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"
35 #include "free-space-cache.h"
38 #undef SCRAMBLE_DELAYED_REFS
41 * control flags for do_chunk_alloc's force field
42 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
43 * if we really need one.
45 * CHUNK_ALLOC_LIMITED means to only try and allocate one
46 * if we have very few chunks already allocated. This is
47 * used as part of the clustering code to help make sure
48 * we have a good pool of storage to cluster in, without
49 * filling the FS with empty chunks
51 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 CHUNK_ALLOC_NO_FORCE = 0,
56 CHUNK_ALLOC_LIMITED = 1,
57 CHUNK_ALLOC_FORCE = 2,
61 * Control how reservations are dealt with.
63 * RESERVE_FREE - freeing a reservation.
64 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
66 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
67 * bytes_may_use as the ENOSPC accounting is done elsewhere
72 RESERVE_ALLOC_NO_ACCOUNT = 2,
75 static int update_block_group(struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, int alloc);
77 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
96 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 flags,
99 static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
101 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
105 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 block_group_cache_done(struct btrfs_block_group_cache *cache)
112 return cache->cached == BTRFS_CACHE_FINISHED;
115 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
117 return (cache->flags & bits) == bits;
120 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
122 atomic_inc(&cache->count);
125 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
127 if (atomic_dec_and_test(&cache->count)) {
128 WARN_ON(cache->pinned > 0);
129 WARN_ON(cache->reserved > 0);
130 kfree(cache->free_space_ctl);
136 * this adds the block group to the fs_info rb tree for the block group
139 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
140 struct btrfs_block_group_cache *block_group)
143 struct rb_node *parent = NULL;
144 struct btrfs_block_group_cache *cache;
146 spin_lock(&info->block_group_cache_lock);
147 p = &info->block_group_cache_tree.rb_node;
151 cache = rb_entry(parent, struct btrfs_block_group_cache,
153 if (block_group->key.objectid < cache->key.objectid) {
155 } else if (block_group->key.objectid > cache->key.objectid) {
158 spin_unlock(&info->block_group_cache_lock);
163 rb_link_node(&block_group->cache_node, parent, p);
164 rb_insert_color(&block_group->cache_node,
165 &info->block_group_cache_tree);
167 if (info->first_logical_byte > block_group->key.objectid)
168 info->first_logical_byte = block_group->key.objectid;
170 spin_unlock(&info->block_group_cache_lock);
176 * This will return the block group at or after bytenr if contains is 0, else
177 * it will return the block group that contains the bytenr
179 static struct btrfs_block_group_cache *
180 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
183 struct btrfs_block_group_cache *cache, *ret = NULL;
187 spin_lock(&info->block_group_cache_lock);
188 n = info->block_group_cache_tree.rb_node;
191 cache = rb_entry(n, struct btrfs_block_group_cache,
193 end = cache->key.objectid + cache->key.offset - 1;
194 start = cache->key.objectid;
196 if (bytenr < start) {
197 if (!contains && (!ret || start < ret->key.objectid))
200 } else if (bytenr > start) {
201 if (contains && bytenr <= end) {
212 btrfs_get_block_group(ret);
213 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
214 info->first_logical_byte = ret->key.objectid;
216 spin_unlock(&info->block_group_cache_lock);
221 static int add_excluded_extent(struct btrfs_root *root,
222 u64 start, u64 num_bytes)
224 u64 end = start + num_bytes - 1;
225 set_extent_bits(&root->fs_info->freed_extents[0],
226 start, end, EXTENT_UPTODATE, GFP_NOFS);
227 set_extent_bits(&root->fs_info->freed_extents[1],
228 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 static void free_excluded_extents(struct btrfs_root *root,
233 struct btrfs_block_group_cache *cache)
237 start = cache->key.objectid;
238 end = start + cache->key.offset - 1;
240 clear_extent_bits(&root->fs_info->freed_extents[0],
241 start, end, EXTENT_UPTODATE, GFP_NOFS);
242 clear_extent_bits(&root->fs_info->freed_extents[1],
243 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 static int exclude_super_stripes(struct btrfs_root *root,
247 struct btrfs_block_group_cache *cache)
254 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
255 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
256 cache->bytes_super += stripe_len;
257 ret = add_excluded_extent(root, cache->key.objectid,
259 BUG_ON(ret); /* -ENOMEM */
262 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
263 bytenr = btrfs_sb_offset(i);
264 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
265 cache->key.objectid, bytenr,
266 0, &logical, &nr, &stripe_len);
267 BUG_ON(ret); /* -ENOMEM */
270 cache->bytes_super += stripe_len;
271 ret = add_excluded_extent(root, logical[nr],
273 BUG_ON(ret); /* -ENOMEM */
281 static struct btrfs_caching_control *
282 get_caching_control(struct btrfs_block_group_cache *cache)
284 struct btrfs_caching_control *ctl;
286 spin_lock(&cache->lock);
287 if (cache->cached != BTRFS_CACHE_STARTED) {
288 spin_unlock(&cache->lock);
292 /* We're loading it the fast way, so we don't have a caching_ctl. */
293 if (!cache->caching_ctl) {
294 spin_unlock(&cache->lock);
298 ctl = cache->caching_ctl;
299 atomic_inc(&ctl->count);
300 spin_unlock(&cache->lock);
304 static void put_caching_control(struct btrfs_caching_control *ctl)
306 if (atomic_dec_and_test(&ctl->count))
311 * this is only called by cache_block_group, since we could have freed extents
312 * we need to check the pinned_extents for any extents that can't be used yet
313 * since their free space will be released as soon as the transaction commits.
315 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
316 struct btrfs_fs_info *info, u64 start, u64 end)
318 u64 extent_start, extent_end, size, total_added = 0;
321 while (start < end) {
322 ret = find_first_extent_bit(info->pinned_extents, start,
323 &extent_start, &extent_end,
324 EXTENT_DIRTY | EXTENT_UPTODATE,
329 if (extent_start <= start) {
330 start = extent_end + 1;
331 } else if (extent_start > start && extent_start < end) {
332 size = extent_start - start;
334 ret = btrfs_add_free_space(block_group, start,
336 BUG_ON(ret); /* -ENOMEM or logic error */
337 start = extent_end + 1;
346 ret = btrfs_add_free_space(block_group, start, size);
347 BUG_ON(ret); /* -ENOMEM or logic error */
353 static noinline void caching_thread(struct btrfs_work *work)
355 struct btrfs_block_group_cache *block_group;
356 struct btrfs_fs_info *fs_info;
357 struct btrfs_caching_control *caching_ctl;
358 struct btrfs_root *extent_root;
359 struct btrfs_path *path;
360 struct extent_buffer *leaf;
361 struct btrfs_key key;
367 caching_ctl = container_of(work, struct btrfs_caching_control, work);
368 block_group = caching_ctl->block_group;
369 fs_info = block_group->fs_info;
370 extent_root = fs_info->extent_root;
372 path = btrfs_alloc_path();
376 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
379 * We don't want to deadlock with somebody trying to allocate a new
380 * extent for the extent root while also trying to search the extent
381 * root to add free space. So we skip locking and search the commit
382 * root, since its read-only
384 path->skip_locking = 1;
385 path->search_commit_root = 1;
390 key.type = BTRFS_EXTENT_ITEM_KEY;
392 mutex_lock(&caching_ctl->mutex);
393 /* need to make sure the commit_root doesn't disappear */
394 down_read(&fs_info->extent_commit_sem);
396 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
400 leaf = path->nodes[0];
401 nritems = btrfs_header_nritems(leaf);
404 if (btrfs_fs_closing(fs_info) > 1) {
409 if (path->slots[0] < nritems) {
410 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
412 ret = find_next_key(path, 0, &key);
416 if (need_resched() ||
417 btrfs_next_leaf(extent_root, path)) {
418 caching_ctl->progress = last;
419 btrfs_release_path(path);
420 up_read(&fs_info->extent_commit_sem);
421 mutex_unlock(&caching_ctl->mutex);
425 leaf = path->nodes[0];
426 nritems = btrfs_header_nritems(leaf);
430 if (key.objectid < block_group->key.objectid) {
435 if (key.objectid >= block_group->key.objectid +
436 block_group->key.offset)
439 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
440 total_found += add_new_free_space(block_group,
443 last = key.objectid + key.offset;
445 if (total_found > (1024 * 1024 * 2)) {
447 wake_up(&caching_ctl->wait);
454 total_found += add_new_free_space(block_group, fs_info, last,
455 block_group->key.objectid +
456 block_group->key.offset);
457 caching_ctl->progress = (u64)-1;
459 spin_lock(&block_group->lock);
460 block_group->caching_ctl = NULL;
461 block_group->cached = BTRFS_CACHE_FINISHED;
462 spin_unlock(&block_group->lock);
465 btrfs_free_path(path);
466 up_read(&fs_info->extent_commit_sem);
468 free_excluded_extents(extent_root, block_group);
470 mutex_unlock(&caching_ctl->mutex);
472 wake_up(&caching_ctl->wait);
474 put_caching_control(caching_ctl);
475 btrfs_put_block_group(block_group);
478 static int cache_block_group(struct btrfs_block_group_cache *cache,
482 struct btrfs_fs_info *fs_info = cache->fs_info;
483 struct btrfs_caching_control *caching_ctl;
486 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
490 INIT_LIST_HEAD(&caching_ctl->list);
491 mutex_init(&caching_ctl->mutex);
492 init_waitqueue_head(&caching_ctl->wait);
493 caching_ctl->block_group = cache;
494 caching_ctl->progress = cache->key.objectid;
495 atomic_set(&caching_ctl->count, 1);
496 caching_ctl->work.func = caching_thread;
498 spin_lock(&cache->lock);
500 * This should be a rare occasion, but this could happen I think in the
501 * case where one thread starts to load the space cache info, and then
502 * some other thread starts a transaction commit which tries to do an
503 * allocation while the other thread is still loading the space cache
504 * info. The previous loop should have kept us from choosing this block
505 * group, but if we've moved to the state where we will wait on caching
506 * block groups we need to first check if we're doing a fast load here,
507 * so we can wait for it to finish, otherwise we could end up allocating
508 * from a block group who's cache gets evicted for one reason or
511 while (cache->cached == BTRFS_CACHE_FAST) {
512 struct btrfs_caching_control *ctl;
514 ctl = cache->caching_ctl;
515 atomic_inc(&ctl->count);
516 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
517 spin_unlock(&cache->lock);
521 finish_wait(&ctl->wait, &wait);
522 put_caching_control(ctl);
523 spin_lock(&cache->lock);
526 if (cache->cached != BTRFS_CACHE_NO) {
527 spin_unlock(&cache->lock);
531 WARN_ON(cache->caching_ctl);
532 cache->caching_ctl = caching_ctl;
533 cache->cached = BTRFS_CACHE_FAST;
534 spin_unlock(&cache->lock);
536 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
537 ret = load_free_space_cache(fs_info, cache);
539 spin_lock(&cache->lock);
541 cache->caching_ctl = NULL;
542 cache->cached = BTRFS_CACHE_FINISHED;
543 cache->last_byte_to_unpin = (u64)-1;
545 if (load_cache_only) {
546 cache->caching_ctl = NULL;
547 cache->cached = BTRFS_CACHE_NO;
549 cache->cached = BTRFS_CACHE_STARTED;
552 spin_unlock(&cache->lock);
553 wake_up(&caching_ctl->wait);
555 put_caching_control(caching_ctl);
556 free_excluded_extents(fs_info->extent_root, cache);
561 * We are not going to do the fast caching, set cached to the
562 * appropriate value and wakeup any waiters.
564 spin_lock(&cache->lock);
565 if (load_cache_only) {
566 cache->caching_ctl = NULL;
567 cache->cached = BTRFS_CACHE_NO;
569 cache->cached = BTRFS_CACHE_STARTED;
571 spin_unlock(&cache->lock);
572 wake_up(&caching_ctl->wait);
575 if (load_cache_only) {
576 put_caching_control(caching_ctl);
580 down_write(&fs_info->extent_commit_sem);
581 atomic_inc(&caching_ctl->count);
582 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
583 up_write(&fs_info->extent_commit_sem);
585 btrfs_get_block_group(cache);
587 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
593 * return the block group that starts at or after bytenr
595 static struct btrfs_block_group_cache *
596 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
598 struct btrfs_block_group_cache *cache;
600 cache = block_group_cache_tree_search(info, bytenr, 0);
606 * return the block group that contains the given bytenr
608 struct btrfs_block_group_cache *btrfs_lookup_block_group(
609 struct btrfs_fs_info *info,
612 struct btrfs_block_group_cache *cache;
614 cache = block_group_cache_tree_search(info, bytenr, 1);
619 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
622 struct list_head *head = &info->space_info;
623 struct btrfs_space_info *found;
625 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
628 list_for_each_entry_rcu(found, head, list) {
629 if (found->flags & flags) {
639 * after adding space to the filesystem, we need to clear the full flags
640 * on all the space infos.
642 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
644 struct list_head *head = &info->space_info;
645 struct btrfs_space_info *found;
648 list_for_each_entry_rcu(found, head, list)
653 u64 btrfs_find_block_group(struct btrfs_root *root,
654 u64 search_start, u64 search_hint, int owner)
656 struct btrfs_block_group_cache *cache;
658 u64 last = max(search_hint, search_start);
665 cache = btrfs_lookup_first_block_group(root->fs_info, last);
669 spin_lock(&cache->lock);
670 last = cache->key.objectid + cache->key.offset;
671 used = btrfs_block_group_used(&cache->item);
673 if ((full_search || !cache->ro) &&
674 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
675 if (used + cache->pinned + cache->reserved <
676 div_factor(cache->key.offset, factor)) {
677 group_start = cache->key.objectid;
678 spin_unlock(&cache->lock);
679 btrfs_put_block_group(cache);
683 spin_unlock(&cache->lock);
684 btrfs_put_block_group(cache);
692 if (!full_search && factor < 10) {
702 /* simple helper to search for an existing extent at a given offset */
703 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
706 struct btrfs_key key;
707 struct btrfs_path *path;
709 path = btrfs_alloc_path();
713 key.objectid = start;
715 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
716 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
718 btrfs_free_path(path);
723 * helper function to lookup reference count and flags of extent.
725 * the head node for delayed ref is used to store the sum of all the
726 * reference count modifications queued up in the rbtree. the head
727 * node may also store the extent flags to set. This way you can check
728 * to see what the reference count and extent flags would be if all of
729 * the delayed refs are not processed.
731 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
732 struct btrfs_root *root, u64 bytenr,
733 u64 num_bytes, u64 *refs, u64 *flags)
735 struct btrfs_delayed_ref_head *head;
736 struct btrfs_delayed_ref_root *delayed_refs;
737 struct btrfs_path *path;
738 struct btrfs_extent_item *ei;
739 struct extent_buffer *leaf;
740 struct btrfs_key key;
746 path = btrfs_alloc_path();
750 key.objectid = bytenr;
751 key.type = BTRFS_EXTENT_ITEM_KEY;
752 key.offset = num_bytes;
754 path->skip_locking = 1;
755 path->search_commit_root = 1;
758 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
764 leaf = path->nodes[0];
765 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
766 if (item_size >= sizeof(*ei)) {
767 ei = btrfs_item_ptr(leaf, path->slots[0],
768 struct btrfs_extent_item);
769 num_refs = btrfs_extent_refs(leaf, ei);
770 extent_flags = btrfs_extent_flags(leaf, ei);
772 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
773 struct btrfs_extent_item_v0 *ei0;
774 BUG_ON(item_size != sizeof(*ei0));
775 ei0 = btrfs_item_ptr(leaf, path->slots[0],
776 struct btrfs_extent_item_v0);
777 num_refs = btrfs_extent_refs_v0(leaf, ei0);
778 /* FIXME: this isn't correct for data */
779 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
784 BUG_ON(num_refs == 0);
794 delayed_refs = &trans->transaction->delayed_refs;
795 spin_lock(&delayed_refs->lock);
796 head = btrfs_find_delayed_ref_head(trans, bytenr);
798 if (!mutex_trylock(&head->mutex)) {
799 atomic_inc(&head->node.refs);
800 spin_unlock(&delayed_refs->lock);
802 btrfs_release_path(path);
805 * Mutex was contended, block until it's released and try
808 mutex_lock(&head->mutex);
809 mutex_unlock(&head->mutex);
810 btrfs_put_delayed_ref(&head->node);
813 if (head->extent_op && head->extent_op->update_flags)
814 extent_flags |= head->extent_op->flags_to_set;
816 BUG_ON(num_refs == 0);
818 num_refs += head->node.ref_mod;
819 mutex_unlock(&head->mutex);
821 spin_unlock(&delayed_refs->lock);
823 WARN_ON(num_refs == 0);
827 *flags = extent_flags;
829 btrfs_free_path(path);
834 * Back reference rules. Back refs have three main goals:
836 * 1) differentiate between all holders of references to an extent so that
837 * when a reference is dropped we can make sure it was a valid reference
838 * before freeing the extent.
840 * 2) Provide enough information to quickly find the holders of an extent
841 * if we notice a given block is corrupted or bad.
843 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
844 * maintenance. This is actually the same as #2, but with a slightly
845 * different use case.
847 * There are two kinds of back refs. The implicit back refs is optimized
848 * for pointers in non-shared tree blocks. For a given pointer in a block,
849 * back refs of this kind provide information about the block's owner tree
850 * and the pointer's key. These information allow us to find the block by
851 * b-tree searching. The full back refs is for pointers in tree blocks not
852 * referenced by their owner trees. The location of tree block is recorded
853 * in the back refs. Actually the full back refs is generic, and can be
854 * used in all cases the implicit back refs is used. The major shortcoming
855 * of the full back refs is its overhead. Every time a tree block gets
856 * COWed, we have to update back refs entry for all pointers in it.
858 * For a newly allocated tree block, we use implicit back refs for
859 * pointers in it. This means most tree related operations only involve
860 * implicit back refs. For a tree block created in old transaction, the
861 * only way to drop a reference to it is COW it. So we can detect the
862 * event that tree block loses its owner tree's reference and do the
863 * back refs conversion.
865 * When a tree block is COW'd through a tree, there are four cases:
867 * The reference count of the block is one and the tree is the block's
868 * owner tree. Nothing to do in this case.
870 * The reference count of the block is one and the tree is not the
871 * block's owner tree. In this case, full back refs is used for pointers
872 * in the block. Remove these full back refs, add implicit back refs for
873 * every pointers in the new block.
875 * The reference count of the block is greater than one and the tree is
876 * the block's owner tree. In this case, implicit back refs is used for
877 * pointers in the block. Add full back refs for every pointers in the
878 * block, increase lower level extents' reference counts. The original
879 * implicit back refs are entailed to the new block.
881 * The reference count of the block is greater than one and the tree is
882 * not the block's owner tree. Add implicit back refs for every pointer in
883 * the new block, increase lower level extents' reference count.
885 * Back Reference Key composing:
887 * The key objectid corresponds to the first byte in the extent,
888 * The key type is used to differentiate between types of back refs.
889 * There are different meanings of the key offset for different types
892 * File extents can be referenced by:
894 * - multiple snapshots, subvolumes, or different generations in one subvol
895 * - different files inside a single subvolume
896 * - different offsets inside a file (bookend extents in file.c)
898 * The extent ref structure for the implicit back refs has fields for:
900 * - Objectid of the subvolume root
901 * - objectid of the file holding the reference
902 * - original offset in the file
903 * - how many bookend extents
905 * The key offset for the implicit back refs is hash of the first
908 * The extent ref structure for the full back refs has field for:
910 * - number of pointers in the tree leaf
912 * The key offset for the implicit back refs is the first byte of
915 * When a file extent is allocated, The implicit back refs is used.
916 * the fields are filled in:
918 * (root_key.objectid, inode objectid, offset in file, 1)
920 * When a file extent is removed file truncation, we find the
921 * corresponding implicit back refs and check the following fields:
923 * (btrfs_header_owner(leaf), inode objectid, offset in file)
925 * Btree extents can be referenced by:
927 * - Different subvolumes
929 * Both the implicit back refs and the full back refs for tree blocks
930 * only consist of key. The key offset for the implicit back refs is
931 * objectid of block's owner tree. The key offset for the full back refs
932 * is the first byte of parent block.
934 * When implicit back refs is used, information about the lowest key and
935 * level of the tree block are required. These information are stored in
936 * tree block info structure.
939 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
940 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
941 struct btrfs_root *root,
942 struct btrfs_path *path,
943 u64 owner, u32 extra_size)
945 struct btrfs_extent_item *item;
946 struct btrfs_extent_item_v0 *ei0;
947 struct btrfs_extent_ref_v0 *ref0;
948 struct btrfs_tree_block_info *bi;
949 struct extent_buffer *leaf;
950 struct btrfs_key key;
951 struct btrfs_key found_key;
952 u32 new_size = sizeof(*item);
956 leaf = path->nodes[0];
957 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
959 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
960 ei0 = btrfs_item_ptr(leaf, path->slots[0],
961 struct btrfs_extent_item_v0);
962 refs = btrfs_extent_refs_v0(leaf, ei0);
964 if (owner == (u64)-1) {
966 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
967 ret = btrfs_next_leaf(root, path);
970 BUG_ON(ret > 0); /* Corruption */
971 leaf = path->nodes[0];
973 btrfs_item_key_to_cpu(leaf, &found_key,
975 BUG_ON(key.objectid != found_key.objectid);
976 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
980 ref0 = btrfs_item_ptr(leaf, path->slots[0],
981 struct btrfs_extent_ref_v0);
982 owner = btrfs_ref_objectid_v0(leaf, ref0);
986 btrfs_release_path(path);
988 if (owner < BTRFS_FIRST_FREE_OBJECTID)
989 new_size += sizeof(*bi);
991 new_size -= sizeof(*ei0);
992 ret = btrfs_search_slot(trans, root, &key, path,
993 new_size + extra_size, 1);
996 BUG_ON(ret); /* Corruption */
998 btrfs_extend_item(trans, root, path, new_size);
1000 leaf = path->nodes[0];
1001 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1002 btrfs_set_extent_refs(leaf, item, refs);
1003 /* FIXME: get real generation */
1004 btrfs_set_extent_generation(leaf, item, 0);
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1006 btrfs_set_extent_flags(leaf, item,
1007 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1008 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1009 bi = (struct btrfs_tree_block_info *)(item + 1);
1010 /* FIXME: get first key of the block */
1011 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1012 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1014 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1016 btrfs_mark_buffer_dirty(leaf);
1021 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1023 u32 high_crc = ~(u32)0;
1024 u32 low_crc = ~(u32)0;
1027 lenum = cpu_to_le64(root_objectid);
1028 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1029 lenum = cpu_to_le64(owner);
1030 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1031 lenum = cpu_to_le64(offset);
1032 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1034 return ((u64)high_crc << 31) ^ (u64)low_crc;
1037 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1038 struct btrfs_extent_data_ref *ref)
1040 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1041 btrfs_extent_data_ref_objectid(leaf, ref),
1042 btrfs_extent_data_ref_offset(leaf, ref));
1045 static int match_extent_data_ref(struct extent_buffer *leaf,
1046 struct btrfs_extent_data_ref *ref,
1047 u64 root_objectid, u64 owner, u64 offset)
1049 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1050 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1051 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1056 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path,
1059 u64 bytenr, u64 parent,
1061 u64 owner, u64 offset)
1063 struct btrfs_key key;
1064 struct btrfs_extent_data_ref *ref;
1065 struct extent_buffer *leaf;
1071 key.objectid = bytenr;
1073 key.type = BTRFS_SHARED_DATA_REF_KEY;
1074 key.offset = parent;
1076 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1077 key.offset = hash_extent_data_ref(root_objectid,
1082 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1092 key.type = BTRFS_EXTENT_REF_V0_KEY;
1093 btrfs_release_path(path);
1094 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 leaf = path->nodes[0];
1106 nritems = btrfs_header_nritems(leaf);
1108 if (path->slots[0] >= nritems) {
1109 ret = btrfs_next_leaf(root, path);
1115 leaf = path->nodes[0];
1116 nritems = btrfs_header_nritems(leaf);
1120 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1121 if (key.objectid != bytenr ||
1122 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1125 ref = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1128 if (match_extent_data_ref(leaf, ref, root_objectid,
1131 btrfs_release_path(path);
1143 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1144 struct btrfs_root *root,
1145 struct btrfs_path *path,
1146 u64 bytenr, u64 parent,
1147 u64 root_objectid, u64 owner,
1148 u64 offset, int refs_to_add)
1150 struct btrfs_key key;
1151 struct extent_buffer *leaf;
1156 key.objectid = bytenr;
1158 key.type = BTRFS_SHARED_DATA_REF_KEY;
1159 key.offset = parent;
1160 size = sizeof(struct btrfs_shared_data_ref);
1162 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1163 key.offset = hash_extent_data_ref(root_objectid,
1165 size = sizeof(struct btrfs_extent_data_ref);
1168 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1169 if (ret && ret != -EEXIST)
1172 leaf = path->nodes[0];
1174 struct btrfs_shared_data_ref *ref;
1175 ref = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_shared_data_ref);
1178 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1180 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1181 num_refs += refs_to_add;
1182 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1185 struct btrfs_extent_data_ref *ref;
1186 while (ret == -EEXIST) {
1187 ref = btrfs_item_ptr(leaf, path->slots[0],
1188 struct btrfs_extent_data_ref);
1189 if (match_extent_data_ref(leaf, ref, root_objectid,
1192 btrfs_release_path(path);
1194 ret = btrfs_insert_empty_item(trans, root, path, &key,
1196 if (ret && ret != -EEXIST)
1199 leaf = path->nodes[0];
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1204 btrfs_set_extent_data_ref_root(leaf, ref,
1206 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1207 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1208 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1210 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1211 num_refs += refs_to_add;
1212 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1215 btrfs_mark_buffer_dirty(leaf);
1218 btrfs_release_path(path);
1222 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_path *path,
1227 struct btrfs_key key;
1228 struct btrfs_extent_data_ref *ref1 = NULL;
1229 struct btrfs_shared_data_ref *ref2 = NULL;
1230 struct extent_buffer *leaf;
1234 leaf = path->nodes[0];
1235 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1237 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1238 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_extent_data_ref);
1240 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1241 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1242 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_shared_data_ref);
1244 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1245 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1246 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1247 struct btrfs_extent_ref_v0 *ref0;
1248 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_ref_v0);
1250 num_refs = btrfs_ref_count_v0(leaf, ref0);
1256 BUG_ON(num_refs < refs_to_drop);
1257 num_refs -= refs_to_drop;
1259 if (num_refs == 0) {
1260 ret = btrfs_del_item(trans, root, path);
1262 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1263 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1264 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1265 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1268 struct btrfs_extent_ref_v0 *ref0;
1269 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_ref_v0);
1271 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1274 btrfs_mark_buffer_dirty(leaf);
1279 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1280 struct btrfs_path *path,
1281 struct btrfs_extent_inline_ref *iref)
1283 struct btrfs_key key;
1284 struct extent_buffer *leaf;
1285 struct btrfs_extent_data_ref *ref1;
1286 struct btrfs_shared_data_ref *ref2;
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1292 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1293 BTRFS_EXTENT_DATA_REF_KEY) {
1294 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1295 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1297 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1300 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1301 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_data_ref);
1303 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1304 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1305 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_shared_data_ref);
1307 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1308 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1310 struct btrfs_extent_ref_v0 *ref0;
1311 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1312 struct btrfs_extent_ref_v0);
1313 num_refs = btrfs_ref_count_v0(leaf, ref0);
1321 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1322 struct btrfs_root *root,
1323 struct btrfs_path *path,
1324 u64 bytenr, u64 parent,
1327 struct btrfs_key key;
1330 key.objectid = bytenr;
1332 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1333 key.offset = parent;
1335 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1336 key.offset = root_objectid;
1339 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1342 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1343 if (ret == -ENOENT && parent) {
1344 btrfs_release_path(path);
1345 key.type = BTRFS_EXTENT_REF_V0_KEY;
1346 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1354 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root,
1356 struct btrfs_path *path,
1357 u64 bytenr, u64 parent,
1360 struct btrfs_key key;
1363 key.objectid = bytenr;
1365 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1366 key.offset = parent;
1368 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1369 key.offset = root_objectid;
1372 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1373 btrfs_release_path(path);
1377 static inline int extent_ref_type(u64 parent, u64 owner)
1380 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1382 type = BTRFS_SHARED_BLOCK_REF_KEY;
1384 type = BTRFS_TREE_BLOCK_REF_KEY;
1387 type = BTRFS_SHARED_DATA_REF_KEY;
1389 type = BTRFS_EXTENT_DATA_REF_KEY;
1394 static int find_next_key(struct btrfs_path *path, int level,
1395 struct btrfs_key *key)
1398 for (; level < BTRFS_MAX_LEVEL; level++) {
1399 if (!path->nodes[level])
1401 if (path->slots[level] + 1 >=
1402 btrfs_header_nritems(path->nodes[level]))
1405 btrfs_item_key_to_cpu(path->nodes[level], key,
1406 path->slots[level] + 1);
1408 btrfs_node_key_to_cpu(path->nodes[level], key,
1409 path->slots[level] + 1);
1416 * look for inline back ref. if back ref is found, *ref_ret is set
1417 * to the address of inline back ref, and 0 is returned.
1419 * if back ref isn't found, *ref_ret is set to the address where it
1420 * should be inserted, and -ENOENT is returned.
1422 * if insert is true and there are too many inline back refs, the path
1423 * points to the extent item, and -EAGAIN is returned.
1425 * NOTE: inline back refs are ordered in the same way that back ref
1426 * items in the tree are ordered.
1428 static noinline_for_stack
1429 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1430 struct btrfs_root *root,
1431 struct btrfs_path *path,
1432 struct btrfs_extent_inline_ref **ref_ret,
1433 u64 bytenr, u64 num_bytes,
1434 u64 parent, u64 root_objectid,
1435 u64 owner, u64 offset, int insert)
1437 struct btrfs_key key;
1438 struct extent_buffer *leaf;
1439 struct btrfs_extent_item *ei;
1440 struct btrfs_extent_inline_ref *iref;
1451 key.objectid = bytenr;
1452 key.type = BTRFS_EXTENT_ITEM_KEY;
1453 key.offset = num_bytes;
1455 want = extent_ref_type(parent, owner);
1457 extra_size = btrfs_extent_inline_ref_size(want);
1458 path->keep_locks = 1;
1461 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1466 if (ret && !insert) {
1470 BUG_ON(ret); /* Corruption */
1472 leaf = path->nodes[0];
1473 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1474 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1475 if (item_size < sizeof(*ei)) {
1480 ret = convert_extent_item_v0(trans, root, path, owner,
1486 leaf = path->nodes[0];
1487 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 BUG_ON(item_size < sizeof(*ei));
1492 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1493 flags = btrfs_extent_flags(leaf, ei);
1495 ptr = (unsigned long)(ei + 1);
1496 end = (unsigned long)ei + item_size;
1498 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1499 ptr += sizeof(struct btrfs_tree_block_info);
1502 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1511 iref = (struct btrfs_extent_inline_ref *)ptr;
1512 type = btrfs_extent_inline_ref_type(leaf, iref);
1516 ptr += btrfs_extent_inline_ref_size(type);
1520 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1521 struct btrfs_extent_data_ref *dref;
1522 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1523 if (match_extent_data_ref(leaf, dref, root_objectid,
1528 if (hash_extent_data_ref_item(leaf, dref) <
1529 hash_extent_data_ref(root_objectid, owner, offset))
1533 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1535 if (parent == ref_offset) {
1539 if (ref_offset < parent)
1542 if (root_objectid == ref_offset) {
1546 if (ref_offset < root_objectid)
1550 ptr += btrfs_extent_inline_ref_size(type);
1552 if (err == -ENOENT && insert) {
1553 if (item_size + extra_size >=
1554 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1559 * To add new inline back ref, we have to make sure
1560 * there is no corresponding back ref item.
1561 * For simplicity, we just do not add new inline back
1562 * ref if there is any kind of item for this block
1564 if (find_next_key(path, 0, &key) == 0 &&
1565 key.objectid == bytenr &&
1566 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1571 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1574 path->keep_locks = 0;
1575 btrfs_unlock_up_safe(path, 1);
1581 * helper to add new inline back ref
1583 static noinline_for_stack
1584 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct btrfs_path *path,
1587 struct btrfs_extent_inline_ref *iref,
1588 u64 parent, u64 root_objectid,
1589 u64 owner, u64 offset, int refs_to_add,
1590 struct btrfs_delayed_extent_op *extent_op)
1592 struct extent_buffer *leaf;
1593 struct btrfs_extent_item *ei;
1596 unsigned long item_offset;
1601 leaf = path->nodes[0];
1602 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1603 item_offset = (unsigned long)iref - (unsigned long)ei;
1605 type = extent_ref_type(parent, owner);
1606 size = btrfs_extent_inline_ref_size(type);
1608 btrfs_extend_item(trans, root, path, size);
1610 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1611 refs = btrfs_extent_refs(leaf, ei);
1612 refs += refs_to_add;
1613 btrfs_set_extent_refs(leaf, ei, refs);
1615 __run_delayed_extent_op(extent_op, leaf, ei);
1617 ptr = (unsigned long)ei + item_offset;
1618 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1619 if (ptr < end - size)
1620 memmove_extent_buffer(leaf, ptr + size, ptr,
1623 iref = (struct btrfs_extent_inline_ref *)ptr;
1624 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1625 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1626 struct btrfs_extent_data_ref *dref;
1627 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1628 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1629 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1630 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1631 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1632 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1633 struct btrfs_shared_data_ref *sref;
1634 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1635 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1636 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1637 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1638 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1640 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1642 btrfs_mark_buffer_dirty(leaf);
1645 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1646 struct btrfs_root *root,
1647 struct btrfs_path *path,
1648 struct btrfs_extent_inline_ref **ref_ret,
1649 u64 bytenr, u64 num_bytes, u64 parent,
1650 u64 root_objectid, u64 owner, u64 offset)
1654 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1655 bytenr, num_bytes, parent,
1656 root_objectid, owner, offset, 0);
1660 btrfs_release_path(path);
1663 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1667 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1668 root_objectid, owner, offset);
1674 * helper to update/remove inline back ref
1676 static noinline_for_stack
1677 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1678 struct btrfs_root *root,
1679 struct btrfs_path *path,
1680 struct btrfs_extent_inline_ref *iref,
1682 struct btrfs_delayed_extent_op *extent_op)
1684 struct extent_buffer *leaf;
1685 struct btrfs_extent_item *ei;
1686 struct btrfs_extent_data_ref *dref = NULL;
1687 struct btrfs_shared_data_ref *sref = NULL;
1695 leaf = path->nodes[0];
1696 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1697 refs = btrfs_extent_refs(leaf, ei);
1698 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1699 refs += refs_to_mod;
1700 btrfs_set_extent_refs(leaf, ei, refs);
1702 __run_delayed_extent_op(extent_op, leaf, ei);
1704 type = btrfs_extent_inline_ref_type(leaf, iref);
1706 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1707 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1708 refs = btrfs_extent_data_ref_count(leaf, dref);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 refs = btrfs_shared_data_ref_count(leaf, sref);
1714 BUG_ON(refs_to_mod != -1);
1717 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1718 refs += refs_to_mod;
1721 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1722 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1724 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1726 size = btrfs_extent_inline_ref_size(type);
1727 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1728 ptr = (unsigned long)iref;
1729 end = (unsigned long)ei + item_size;
1730 if (ptr + size < end)
1731 memmove_extent_buffer(leaf, ptr, ptr + size,
1734 btrfs_truncate_item(trans, root, path, item_size, 1);
1736 btrfs_mark_buffer_dirty(leaf);
1739 static noinline_for_stack
1740 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 u64 bytenr, u64 num_bytes, u64 parent,
1744 u64 root_objectid, u64 owner,
1745 u64 offset, int refs_to_add,
1746 struct btrfs_delayed_extent_op *extent_op)
1748 struct btrfs_extent_inline_ref *iref;
1751 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1752 bytenr, num_bytes, parent,
1753 root_objectid, owner, offset, 1);
1755 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1756 update_inline_extent_backref(trans, root, path, iref,
1757 refs_to_add, extent_op);
1758 } else if (ret == -ENOENT) {
1759 setup_inline_extent_backref(trans, root, path, iref, parent,
1760 root_objectid, owner, offset,
1761 refs_to_add, extent_op);
1767 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1768 struct btrfs_root *root,
1769 struct btrfs_path *path,
1770 u64 bytenr, u64 parent, u64 root_objectid,
1771 u64 owner, u64 offset, int refs_to_add)
1774 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1775 BUG_ON(refs_to_add != 1);
1776 ret = insert_tree_block_ref(trans, root, path, bytenr,
1777 parent, root_objectid);
1779 ret = insert_extent_data_ref(trans, root, path, bytenr,
1780 parent, root_objectid,
1781 owner, offset, refs_to_add);
1786 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1787 struct btrfs_root *root,
1788 struct btrfs_path *path,
1789 struct btrfs_extent_inline_ref *iref,
1790 int refs_to_drop, int is_data)
1794 BUG_ON(!is_data && refs_to_drop != 1);
1796 update_inline_extent_backref(trans, root, path, iref,
1797 -refs_to_drop, NULL);
1798 } else if (is_data) {
1799 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1801 ret = btrfs_del_item(trans, root, path);
1806 static int btrfs_issue_discard(struct block_device *bdev,
1809 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1812 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1813 u64 num_bytes, u64 *actual_bytes)
1816 u64 discarded_bytes = 0;
1817 struct btrfs_bio *bbio = NULL;
1820 /* Tell the block device(s) that the sectors can be discarded */
1821 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1822 bytenr, &num_bytes, &bbio, 0);
1823 /* Error condition is -ENOMEM */
1825 struct btrfs_bio_stripe *stripe = bbio->stripes;
1829 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1830 if (!stripe->dev->can_discard)
1833 ret = btrfs_issue_discard(stripe->dev->bdev,
1837 discarded_bytes += stripe->length;
1838 else if (ret != -EOPNOTSUPP)
1839 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1842 * Just in case we get back EOPNOTSUPP for some reason,
1843 * just ignore the return value so we don't screw up
1844 * people calling discard_extent.
1852 *actual_bytes = discarded_bytes;
1858 /* Can return -ENOMEM */
1859 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 u64 bytenr, u64 num_bytes, u64 parent,
1862 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1865 struct btrfs_fs_info *fs_info = root->fs_info;
1867 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1868 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1870 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1871 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1873 parent, root_objectid, (int)owner,
1874 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1876 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1878 parent, root_objectid, owner, offset,
1879 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1884 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 u64 bytenr, u64 num_bytes,
1887 u64 parent, u64 root_objectid,
1888 u64 owner, u64 offset, int refs_to_add,
1889 struct btrfs_delayed_extent_op *extent_op)
1891 struct btrfs_path *path;
1892 struct extent_buffer *leaf;
1893 struct btrfs_extent_item *item;
1898 path = btrfs_alloc_path();
1903 path->leave_spinning = 1;
1904 /* this will setup the path even if it fails to insert the back ref */
1905 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1906 path, bytenr, num_bytes, parent,
1907 root_objectid, owner, offset,
1908 refs_to_add, extent_op);
1912 if (ret != -EAGAIN) {
1917 leaf = path->nodes[0];
1918 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1919 refs = btrfs_extent_refs(leaf, item);
1920 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1922 __run_delayed_extent_op(extent_op, leaf, item);
1924 btrfs_mark_buffer_dirty(leaf);
1925 btrfs_release_path(path);
1928 path->leave_spinning = 1;
1930 /* now insert the actual backref */
1931 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1932 path, bytenr, parent, root_objectid,
1933 owner, offset, refs_to_add);
1935 btrfs_abort_transaction(trans, root, ret);
1937 btrfs_free_path(path);
1941 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 struct btrfs_delayed_ref_node *node,
1944 struct btrfs_delayed_extent_op *extent_op,
1945 int insert_reserved)
1948 struct btrfs_delayed_data_ref *ref;
1949 struct btrfs_key ins;
1954 ins.objectid = node->bytenr;
1955 ins.offset = node->num_bytes;
1956 ins.type = BTRFS_EXTENT_ITEM_KEY;
1958 ref = btrfs_delayed_node_to_data_ref(node);
1959 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1960 parent = ref->parent;
1962 ref_root = ref->root;
1964 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1966 BUG_ON(extent_op->update_key);
1967 flags |= extent_op->flags_to_set;
1969 ret = alloc_reserved_file_extent(trans, root,
1970 parent, ref_root, flags,
1971 ref->objectid, ref->offset,
1972 &ins, node->ref_mod);
1973 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1974 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1975 node->num_bytes, parent,
1976 ref_root, ref->objectid,
1977 ref->offset, node->ref_mod,
1979 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1980 ret = __btrfs_free_extent(trans, root, node->bytenr,
1981 node->num_bytes, parent,
1982 ref_root, ref->objectid,
1983 ref->offset, node->ref_mod,
1991 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1992 struct extent_buffer *leaf,
1993 struct btrfs_extent_item *ei)
1995 u64 flags = btrfs_extent_flags(leaf, ei);
1996 if (extent_op->update_flags) {
1997 flags |= extent_op->flags_to_set;
1998 btrfs_set_extent_flags(leaf, ei, flags);
2001 if (extent_op->update_key) {
2002 struct btrfs_tree_block_info *bi;
2003 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2004 bi = (struct btrfs_tree_block_info *)(ei + 1);
2005 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2009 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 struct btrfs_delayed_ref_node *node,
2012 struct btrfs_delayed_extent_op *extent_op)
2014 struct btrfs_key key;
2015 struct btrfs_path *path;
2016 struct btrfs_extent_item *ei;
2017 struct extent_buffer *leaf;
2025 path = btrfs_alloc_path();
2029 key.objectid = node->bytenr;
2030 key.type = BTRFS_EXTENT_ITEM_KEY;
2031 key.offset = node->num_bytes;
2034 path->leave_spinning = 1;
2035 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2046 leaf = path->nodes[0];
2047 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2048 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2049 if (item_size < sizeof(*ei)) {
2050 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2056 leaf = path->nodes[0];
2057 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2060 BUG_ON(item_size < sizeof(*ei));
2061 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2062 __run_delayed_extent_op(extent_op, leaf, ei);
2064 btrfs_mark_buffer_dirty(leaf);
2066 btrfs_free_path(path);
2070 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2071 struct btrfs_root *root,
2072 struct btrfs_delayed_ref_node *node,
2073 struct btrfs_delayed_extent_op *extent_op,
2074 int insert_reserved)
2077 struct btrfs_delayed_tree_ref *ref;
2078 struct btrfs_key ins;
2082 ins.objectid = node->bytenr;
2083 ins.offset = node->num_bytes;
2084 ins.type = BTRFS_EXTENT_ITEM_KEY;
2086 ref = btrfs_delayed_node_to_tree_ref(node);
2087 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2088 parent = ref->parent;
2090 ref_root = ref->root;
2092 BUG_ON(node->ref_mod != 1);
2093 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2094 BUG_ON(!extent_op || !extent_op->update_flags ||
2095 !extent_op->update_key);
2096 ret = alloc_reserved_tree_block(trans, root,
2098 extent_op->flags_to_set,
2101 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2102 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2103 node->num_bytes, parent, ref_root,
2104 ref->level, 0, 1, extent_op);
2105 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2106 ret = __btrfs_free_extent(trans, root, node->bytenr,
2107 node->num_bytes, parent, ref_root,
2108 ref->level, 0, 1, extent_op);
2115 /* helper function to actually process a single delayed ref entry */
2116 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2117 struct btrfs_root *root,
2118 struct btrfs_delayed_ref_node *node,
2119 struct btrfs_delayed_extent_op *extent_op,
2120 int insert_reserved)
2127 if (btrfs_delayed_ref_is_head(node)) {
2128 struct btrfs_delayed_ref_head *head;
2130 * we've hit the end of the chain and we were supposed
2131 * to insert this extent into the tree. But, it got
2132 * deleted before we ever needed to insert it, so all
2133 * we have to do is clean up the accounting
2136 head = btrfs_delayed_node_to_head(node);
2137 if (insert_reserved) {
2138 btrfs_pin_extent(root, node->bytenr,
2139 node->num_bytes, 1);
2140 if (head->is_data) {
2141 ret = btrfs_del_csums(trans, root,
2149 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2150 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2151 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2153 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2154 node->type == BTRFS_SHARED_DATA_REF_KEY)
2155 ret = run_delayed_data_ref(trans, root, node, extent_op,
2162 static noinline struct btrfs_delayed_ref_node *
2163 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2165 struct rb_node *node;
2166 struct btrfs_delayed_ref_node *ref;
2167 int action = BTRFS_ADD_DELAYED_REF;
2170 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2171 * this prevents ref count from going down to zero when
2172 * there still are pending delayed ref.
2174 node = rb_prev(&head->node.rb_node);
2178 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2180 if (ref->bytenr != head->node.bytenr)
2182 if (ref->action == action)
2184 node = rb_prev(node);
2186 if (action == BTRFS_ADD_DELAYED_REF) {
2187 action = BTRFS_DROP_DELAYED_REF;
2194 * Returns 0 on success or if called with an already aborted transaction.
2195 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2197 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2198 struct btrfs_root *root,
2199 struct list_head *cluster)
2201 struct btrfs_delayed_ref_root *delayed_refs;
2202 struct btrfs_delayed_ref_node *ref;
2203 struct btrfs_delayed_ref_head *locked_ref = NULL;
2204 struct btrfs_delayed_extent_op *extent_op;
2205 struct btrfs_fs_info *fs_info = root->fs_info;
2208 int must_insert_reserved = 0;
2210 delayed_refs = &trans->transaction->delayed_refs;
2213 /* pick a new head ref from the cluster list */
2214 if (list_empty(cluster))
2217 locked_ref = list_entry(cluster->next,
2218 struct btrfs_delayed_ref_head, cluster);
2220 /* grab the lock that says we are going to process
2221 * all the refs for this head */
2222 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2225 * we may have dropped the spin lock to get the head
2226 * mutex lock, and that might have given someone else
2227 * time to free the head. If that's true, it has been
2228 * removed from our list and we can move on.
2230 if (ret == -EAGAIN) {
2238 * We need to try and merge add/drops of the same ref since we
2239 * can run into issues with relocate dropping the implicit ref
2240 * and then it being added back again before the drop can
2241 * finish. If we merged anything we need to re-loop so we can
2244 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2248 * locked_ref is the head node, so we have to go one
2249 * node back for any delayed ref updates
2251 ref = select_delayed_ref(locked_ref);
2253 if (ref && ref->seq &&
2254 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2256 * there are still refs with lower seq numbers in the
2257 * process of being added. Don't run this ref yet.
2259 list_del_init(&locked_ref->cluster);
2260 btrfs_delayed_ref_unlock(locked_ref);
2262 delayed_refs->num_heads_ready++;
2263 spin_unlock(&delayed_refs->lock);
2265 spin_lock(&delayed_refs->lock);
2270 * record the must insert reserved flag before we
2271 * drop the spin lock.
2273 must_insert_reserved = locked_ref->must_insert_reserved;
2274 locked_ref->must_insert_reserved = 0;
2276 extent_op = locked_ref->extent_op;
2277 locked_ref->extent_op = NULL;
2280 /* All delayed refs have been processed, Go ahead
2281 * and send the head node to run_one_delayed_ref,
2282 * so that any accounting fixes can happen
2284 ref = &locked_ref->node;
2286 if (extent_op && must_insert_reserved) {
2287 btrfs_free_delayed_extent_op(extent_op);
2292 spin_unlock(&delayed_refs->lock);
2294 ret = run_delayed_extent_op(trans, root,
2296 btrfs_free_delayed_extent_op(extent_op);
2300 "btrfs: run_delayed_extent_op "
2301 "returned %d\n", ret);
2302 spin_lock(&delayed_refs->lock);
2303 btrfs_delayed_ref_unlock(locked_ref);
2312 rb_erase(&ref->rb_node, &delayed_refs->root);
2313 delayed_refs->num_entries--;
2314 if (!btrfs_delayed_ref_is_head(ref)) {
2316 * when we play the delayed ref, also correct the
2319 switch (ref->action) {
2320 case BTRFS_ADD_DELAYED_REF:
2321 case BTRFS_ADD_DELAYED_EXTENT:
2322 locked_ref->node.ref_mod -= ref->ref_mod;
2324 case BTRFS_DROP_DELAYED_REF:
2325 locked_ref->node.ref_mod += ref->ref_mod;
2331 spin_unlock(&delayed_refs->lock);
2333 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2334 must_insert_reserved);
2336 btrfs_free_delayed_extent_op(extent_op);
2338 btrfs_delayed_ref_unlock(locked_ref);
2339 btrfs_put_delayed_ref(ref);
2341 "btrfs: run_one_delayed_ref returned %d\n", ret);
2342 spin_lock(&delayed_refs->lock);
2347 * If this node is a head, that means all the refs in this head
2348 * have been dealt with, and we will pick the next head to deal
2349 * with, so we must unlock the head and drop it from the cluster
2350 * list before we release it.
2352 if (btrfs_delayed_ref_is_head(ref)) {
2353 list_del_init(&locked_ref->cluster);
2354 btrfs_delayed_ref_unlock(locked_ref);
2357 btrfs_put_delayed_ref(ref);
2361 spin_lock(&delayed_refs->lock);
2366 #ifdef SCRAMBLE_DELAYED_REFS
2368 * Normally delayed refs get processed in ascending bytenr order. This
2369 * correlates in most cases to the order added. To expose dependencies on this
2370 * order, we start to process the tree in the middle instead of the beginning
2372 static u64 find_middle(struct rb_root *root)
2374 struct rb_node *n = root->rb_node;
2375 struct btrfs_delayed_ref_node *entry;
2378 u64 first = 0, last = 0;
2382 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2383 first = entry->bytenr;
2387 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2388 last = entry->bytenr;
2393 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2394 WARN_ON(!entry->in_tree);
2396 middle = entry->bytenr;
2409 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2410 struct btrfs_fs_info *fs_info)
2412 struct qgroup_update *qgroup_update;
2415 if (list_empty(&trans->qgroup_ref_list) !=
2416 !trans->delayed_ref_elem.seq) {
2417 /* list without seq or seq without list */
2418 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2419 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2420 trans->delayed_ref_elem.seq);
2424 if (!trans->delayed_ref_elem.seq)
2427 while (!list_empty(&trans->qgroup_ref_list)) {
2428 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2429 struct qgroup_update, list);
2430 list_del(&qgroup_update->list);
2432 ret = btrfs_qgroup_account_ref(
2433 trans, fs_info, qgroup_update->node,
2434 qgroup_update->extent_op);
2435 kfree(qgroup_update);
2438 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2444 * this starts processing the delayed reference count updates and
2445 * extent insertions we have queued up so far. count can be
2446 * 0, which means to process everything in the tree at the start
2447 * of the run (but not newly added entries), or it can be some target
2448 * number you'd like to process.
2450 * Returns 0 on success or if called with an aborted transaction
2451 * Returns <0 on error and aborts the transaction
2453 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *root, unsigned long count)
2456 struct rb_node *node;
2457 struct btrfs_delayed_ref_root *delayed_refs;
2458 struct btrfs_delayed_ref_node *ref;
2459 struct list_head cluster;
2462 int run_all = count == (unsigned long)-1;
2466 /* We'll clean this up in btrfs_cleanup_transaction */
2470 if (root == root->fs_info->extent_root)
2471 root = root->fs_info->tree_root;
2473 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2475 delayed_refs = &trans->transaction->delayed_refs;
2476 INIT_LIST_HEAD(&cluster);
2479 spin_lock(&delayed_refs->lock);
2481 #ifdef SCRAMBLE_DELAYED_REFS
2482 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2486 count = delayed_refs->num_entries * 2;
2490 if (!(run_all || run_most) &&
2491 delayed_refs->num_heads_ready < 64)
2495 * go find something we can process in the rbtree. We start at
2496 * the beginning of the tree, and then build a cluster
2497 * of refs to process starting at the first one we are able to
2500 delayed_start = delayed_refs->run_delayed_start;
2501 ret = btrfs_find_ref_cluster(trans, &cluster,
2502 delayed_refs->run_delayed_start);
2506 ret = run_clustered_refs(trans, root, &cluster);
2508 btrfs_release_ref_cluster(&cluster);
2509 spin_unlock(&delayed_refs->lock);
2510 btrfs_abort_transaction(trans, root, ret);
2514 count -= min_t(unsigned long, ret, count);
2519 if (delayed_start >= delayed_refs->run_delayed_start) {
2522 * btrfs_find_ref_cluster looped. let's do one
2523 * more cycle. if we don't run any delayed ref
2524 * during that cycle (because we can't because
2525 * all of them are blocked), bail out.
2530 * no runnable refs left, stop trying
2537 /* refs were run, let's reset staleness detection */
2543 if (!list_empty(&trans->new_bgs)) {
2544 spin_unlock(&delayed_refs->lock);
2545 btrfs_create_pending_block_groups(trans, root);
2546 spin_lock(&delayed_refs->lock);
2549 node = rb_first(&delayed_refs->root);
2552 count = (unsigned long)-1;
2555 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2557 if (btrfs_delayed_ref_is_head(ref)) {
2558 struct btrfs_delayed_ref_head *head;
2560 head = btrfs_delayed_node_to_head(ref);
2561 atomic_inc(&ref->refs);
2563 spin_unlock(&delayed_refs->lock);
2565 * Mutex was contended, block until it's
2566 * released and try again
2568 mutex_lock(&head->mutex);
2569 mutex_unlock(&head->mutex);
2571 btrfs_put_delayed_ref(ref);
2575 node = rb_next(node);
2577 spin_unlock(&delayed_refs->lock);
2578 schedule_timeout(1);
2582 spin_unlock(&delayed_refs->lock);
2583 assert_qgroups_uptodate(trans);
2587 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2588 struct btrfs_root *root,
2589 u64 bytenr, u64 num_bytes, u64 flags,
2592 struct btrfs_delayed_extent_op *extent_op;
2595 extent_op = btrfs_alloc_delayed_extent_op();
2599 extent_op->flags_to_set = flags;
2600 extent_op->update_flags = 1;
2601 extent_op->update_key = 0;
2602 extent_op->is_data = is_data ? 1 : 0;
2604 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2605 num_bytes, extent_op);
2607 btrfs_free_delayed_extent_op(extent_op);
2611 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2612 struct btrfs_root *root,
2613 struct btrfs_path *path,
2614 u64 objectid, u64 offset, u64 bytenr)
2616 struct btrfs_delayed_ref_head *head;
2617 struct btrfs_delayed_ref_node *ref;
2618 struct btrfs_delayed_data_ref *data_ref;
2619 struct btrfs_delayed_ref_root *delayed_refs;
2620 struct rb_node *node;
2624 delayed_refs = &trans->transaction->delayed_refs;
2625 spin_lock(&delayed_refs->lock);
2626 head = btrfs_find_delayed_ref_head(trans, bytenr);
2630 if (!mutex_trylock(&head->mutex)) {
2631 atomic_inc(&head->node.refs);
2632 spin_unlock(&delayed_refs->lock);
2634 btrfs_release_path(path);
2637 * Mutex was contended, block until it's released and let
2640 mutex_lock(&head->mutex);
2641 mutex_unlock(&head->mutex);
2642 btrfs_put_delayed_ref(&head->node);
2646 node = rb_prev(&head->node.rb_node);
2650 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2652 if (ref->bytenr != bytenr)
2656 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2659 data_ref = btrfs_delayed_node_to_data_ref(ref);
2661 node = rb_prev(node);
2665 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2666 if (ref->bytenr == bytenr && ref->seq == seq)
2670 if (data_ref->root != root->root_key.objectid ||
2671 data_ref->objectid != objectid || data_ref->offset != offset)
2676 mutex_unlock(&head->mutex);
2678 spin_unlock(&delayed_refs->lock);
2682 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root,
2684 struct btrfs_path *path,
2685 u64 objectid, u64 offset, u64 bytenr)
2687 struct btrfs_root *extent_root = root->fs_info->extent_root;
2688 struct extent_buffer *leaf;
2689 struct btrfs_extent_data_ref *ref;
2690 struct btrfs_extent_inline_ref *iref;
2691 struct btrfs_extent_item *ei;
2692 struct btrfs_key key;
2696 key.objectid = bytenr;
2697 key.offset = (u64)-1;
2698 key.type = BTRFS_EXTENT_ITEM_KEY;
2700 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2703 BUG_ON(ret == 0); /* Corruption */
2706 if (path->slots[0] == 0)
2710 leaf = path->nodes[0];
2711 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2713 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2717 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2719 if (item_size < sizeof(*ei)) {
2720 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2724 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2726 if (item_size != sizeof(*ei) +
2727 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2730 if (btrfs_extent_generation(leaf, ei) <=
2731 btrfs_root_last_snapshot(&root->root_item))
2734 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2735 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2736 BTRFS_EXTENT_DATA_REF_KEY)
2739 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2740 if (btrfs_extent_refs(leaf, ei) !=
2741 btrfs_extent_data_ref_count(leaf, ref) ||
2742 btrfs_extent_data_ref_root(leaf, ref) !=
2743 root->root_key.objectid ||
2744 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2745 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2753 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2754 struct btrfs_root *root,
2755 u64 objectid, u64 offset, u64 bytenr)
2757 struct btrfs_path *path;
2761 path = btrfs_alloc_path();
2766 ret = check_committed_ref(trans, root, path, objectid,
2768 if (ret && ret != -ENOENT)
2771 ret2 = check_delayed_ref(trans, root, path, objectid,
2773 } while (ret2 == -EAGAIN);
2775 if (ret2 && ret2 != -ENOENT) {
2780 if (ret != -ENOENT || ret2 != -ENOENT)
2783 btrfs_free_path(path);
2784 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2789 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2790 struct btrfs_root *root,
2791 struct extent_buffer *buf,
2792 int full_backref, int inc, int for_cow)
2799 struct btrfs_key key;
2800 struct btrfs_file_extent_item *fi;
2804 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2805 u64, u64, u64, u64, u64, u64, int);
2807 ref_root = btrfs_header_owner(buf);
2808 nritems = btrfs_header_nritems(buf);
2809 level = btrfs_header_level(buf);
2811 if (!root->ref_cows && level == 0)
2815 process_func = btrfs_inc_extent_ref;
2817 process_func = btrfs_free_extent;
2820 parent = buf->start;
2824 for (i = 0; i < nritems; i++) {
2826 btrfs_item_key_to_cpu(buf, &key, i);
2827 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2829 fi = btrfs_item_ptr(buf, i,
2830 struct btrfs_file_extent_item);
2831 if (btrfs_file_extent_type(buf, fi) ==
2832 BTRFS_FILE_EXTENT_INLINE)
2834 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2838 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2839 key.offset -= btrfs_file_extent_offset(buf, fi);
2840 ret = process_func(trans, root, bytenr, num_bytes,
2841 parent, ref_root, key.objectid,
2842 key.offset, for_cow);
2846 bytenr = btrfs_node_blockptr(buf, i);
2847 num_bytes = btrfs_level_size(root, level - 1);
2848 ret = process_func(trans, root, bytenr, num_bytes,
2849 parent, ref_root, level - 1, 0,
2860 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2861 struct extent_buffer *buf, int full_backref, int for_cow)
2863 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2866 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2867 struct extent_buffer *buf, int full_backref, int for_cow)
2869 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2872 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2873 struct btrfs_root *root,
2874 struct btrfs_path *path,
2875 struct btrfs_block_group_cache *cache)
2878 struct btrfs_root *extent_root = root->fs_info->extent_root;
2880 struct extent_buffer *leaf;
2882 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2885 BUG_ON(ret); /* Corruption */
2887 leaf = path->nodes[0];
2888 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2889 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2890 btrfs_mark_buffer_dirty(leaf);
2891 btrfs_release_path(path);
2894 btrfs_abort_transaction(trans, root, ret);
2901 static struct btrfs_block_group_cache *
2902 next_block_group(struct btrfs_root *root,
2903 struct btrfs_block_group_cache *cache)
2905 struct rb_node *node;
2906 spin_lock(&root->fs_info->block_group_cache_lock);
2907 node = rb_next(&cache->cache_node);
2908 btrfs_put_block_group(cache);
2910 cache = rb_entry(node, struct btrfs_block_group_cache,
2912 btrfs_get_block_group(cache);
2915 spin_unlock(&root->fs_info->block_group_cache_lock);
2919 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2920 struct btrfs_trans_handle *trans,
2921 struct btrfs_path *path)
2923 struct btrfs_root *root = block_group->fs_info->tree_root;
2924 struct inode *inode = NULL;
2926 int dcs = BTRFS_DC_ERROR;
2932 * If this block group is smaller than 100 megs don't bother caching the
2935 if (block_group->key.offset < (100 * 1024 * 1024)) {
2936 spin_lock(&block_group->lock);
2937 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2938 spin_unlock(&block_group->lock);
2943 inode = lookup_free_space_inode(root, block_group, path);
2944 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2945 ret = PTR_ERR(inode);
2946 btrfs_release_path(path);
2950 if (IS_ERR(inode)) {
2954 if (block_group->ro)
2957 ret = create_free_space_inode(root, trans, block_group, path);
2963 /* We've already setup this transaction, go ahead and exit */
2964 if (block_group->cache_generation == trans->transid &&
2965 i_size_read(inode)) {
2966 dcs = BTRFS_DC_SETUP;
2971 * We want to set the generation to 0, that way if anything goes wrong
2972 * from here on out we know not to trust this cache when we load up next
2975 BTRFS_I(inode)->generation = 0;
2976 ret = btrfs_update_inode(trans, root, inode);
2979 if (i_size_read(inode) > 0) {
2980 ret = btrfs_truncate_free_space_cache(root, trans, path,
2986 spin_lock(&block_group->lock);
2987 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2988 !btrfs_test_opt(root, SPACE_CACHE)) {
2990 * don't bother trying to write stuff out _if_
2991 * a) we're not cached,
2992 * b) we're with nospace_cache mount option.
2994 dcs = BTRFS_DC_WRITTEN;
2995 spin_unlock(&block_group->lock);
2998 spin_unlock(&block_group->lock);
3001 * Try to preallocate enough space based on how big the block group is.
3002 * Keep in mind this has to include any pinned space which could end up
3003 * taking up quite a bit since it's not folded into the other space
3006 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3011 num_pages *= PAGE_CACHE_SIZE;
3013 ret = btrfs_check_data_free_space(inode, num_pages);
3017 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3018 num_pages, num_pages,
3021 dcs = BTRFS_DC_SETUP;
3022 btrfs_free_reserved_data_space(inode, num_pages);
3027 btrfs_release_path(path);
3029 spin_lock(&block_group->lock);
3030 if (!ret && dcs == BTRFS_DC_SETUP)
3031 block_group->cache_generation = trans->transid;
3032 block_group->disk_cache_state = dcs;
3033 spin_unlock(&block_group->lock);
3038 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3039 struct btrfs_root *root)
3041 struct btrfs_block_group_cache *cache;
3043 struct btrfs_path *path;
3046 path = btrfs_alloc_path();
3052 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3054 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3056 cache = next_block_group(root, cache);
3064 err = cache_save_setup(cache, trans, path);
3065 last = cache->key.objectid + cache->key.offset;
3066 btrfs_put_block_group(cache);
3071 err = btrfs_run_delayed_refs(trans, root,
3073 if (err) /* File system offline */
3077 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3079 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3080 btrfs_put_block_group(cache);
3086 cache = next_block_group(root, cache);
3095 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3096 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3098 last = cache->key.objectid + cache->key.offset;
3100 err = write_one_cache_group(trans, root, path, cache);
3101 if (err) /* File system offline */
3104 btrfs_put_block_group(cache);
3109 * I don't think this is needed since we're just marking our
3110 * preallocated extent as written, but just in case it can't
3114 err = btrfs_run_delayed_refs(trans, root,
3116 if (err) /* File system offline */
3120 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3123 * Really this shouldn't happen, but it could if we
3124 * couldn't write the entire preallocated extent and
3125 * splitting the extent resulted in a new block.
3128 btrfs_put_block_group(cache);
3131 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3133 cache = next_block_group(root, cache);
3142 err = btrfs_write_out_cache(root, trans, cache, path);
3145 * If we didn't have an error then the cache state is still
3146 * NEED_WRITE, so we can set it to WRITTEN.
3148 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3149 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3150 last = cache->key.objectid + cache->key.offset;
3151 btrfs_put_block_group(cache);
3155 btrfs_free_path(path);
3159 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3161 struct btrfs_block_group_cache *block_group;
3164 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3165 if (!block_group || block_group->ro)
3168 btrfs_put_block_group(block_group);
3172 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3173 u64 total_bytes, u64 bytes_used,
3174 struct btrfs_space_info **space_info)
3176 struct btrfs_space_info *found;
3180 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3181 BTRFS_BLOCK_GROUP_RAID10))
3186 found = __find_space_info(info, flags);
3188 spin_lock(&found->lock);
3189 found->total_bytes += total_bytes;
3190 found->disk_total += total_bytes * factor;
3191 found->bytes_used += bytes_used;
3192 found->disk_used += bytes_used * factor;
3194 spin_unlock(&found->lock);
3195 *space_info = found;
3198 found = kzalloc(sizeof(*found), GFP_NOFS);
3202 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3203 INIT_LIST_HEAD(&found->block_groups[i]);
3204 init_rwsem(&found->groups_sem);
3205 spin_lock_init(&found->lock);
3206 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3207 found->total_bytes = total_bytes;
3208 found->disk_total = total_bytes * factor;
3209 found->bytes_used = bytes_used;
3210 found->disk_used = bytes_used * factor;
3211 found->bytes_pinned = 0;
3212 found->bytes_reserved = 0;
3213 found->bytes_readonly = 0;
3214 found->bytes_may_use = 0;
3216 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3217 found->chunk_alloc = 0;
3219 init_waitqueue_head(&found->wait);
3220 *space_info = found;
3221 list_add_rcu(&found->list, &info->space_info);
3222 if (flags & BTRFS_BLOCK_GROUP_DATA)
3223 info->data_sinfo = found;
3227 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3229 u64 extra_flags = chunk_to_extended(flags) &
3230 BTRFS_EXTENDED_PROFILE_MASK;
3232 write_seqlock(&fs_info->profiles_lock);
3233 if (flags & BTRFS_BLOCK_GROUP_DATA)
3234 fs_info->avail_data_alloc_bits |= extra_flags;
3235 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3236 fs_info->avail_metadata_alloc_bits |= extra_flags;
3237 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3238 fs_info->avail_system_alloc_bits |= extra_flags;
3239 write_sequnlock(&fs_info->profiles_lock);
3243 * returns target flags in extended format or 0 if restripe for this
3244 * chunk_type is not in progress
3246 * should be called with either volume_mutex or balance_lock held
3248 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3250 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3256 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3257 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3258 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3259 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3260 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3261 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3262 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3263 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3264 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3271 * @flags: available profiles in extended format (see ctree.h)
3273 * Returns reduced profile in chunk format. If profile changing is in
3274 * progress (either running or paused) picks the target profile (if it's
3275 * already available), otherwise falls back to plain reducing.
3277 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3280 * we add in the count of missing devices because we want
3281 * to make sure that any RAID levels on a degraded FS
3282 * continue to be honored.
3284 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3285 root->fs_info->fs_devices->missing_devices;
3289 * see if restripe for this chunk_type is in progress, if so
3290 * try to reduce to the target profile
3292 spin_lock(&root->fs_info->balance_lock);
3293 target = get_restripe_target(root->fs_info, flags);
3295 /* pick target profile only if it's already available */
3296 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3297 spin_unlock(&root->fs_info->balance_lock);
3298 return extended_to_chunk(target);
3301 spin_unlock(&root->fs_info->balance_lock);
3303 if (num_devices == 1)
3304 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3305 if (num_devices < 4)
3306 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3308 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3309 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3310 BTRFS_BLOCK_GROUP_RAID10))) {
3311 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3314 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3315 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3316 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3319 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3320 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3321 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3322 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3323 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3326 return extended_to_chunk(flags);
3329 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3334 seq = read_seqbegin(&root->fs_info->profiles_lock);
3336 if (flags & BTRFS_BLOCK_GROUP_DATA)
3337 flags |= root->fs_info->avail_data_alloc_bits;
3338 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3339 flags |= root->fs_info->avail_system_alloc_bits;
3340 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3341 flags |= root->fs_info->avail_metadata_alloc_bits;
3342 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3344 return btrfs_reduce_alloc_profile(root, flags);
3347 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3352 flags = BTRFS_BLOCK_GROUP_DATA;
3353 else if (root == root->fs_info->chunk_root)
3354 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3356 flags = BTRFS_BLOCK_GROUP_METADATA;
3358 return get_alloc_profile(root, flags);
3362 * This will check the space that the inode allocates from to make sure we have
3363 * enough space for bytes.
3365 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3367 struct btrfs_space_info *data_sinfo;
3368 struct btrfs_root *root = BTRFS_I(inode)->root;
3369 struct btrfs_fs_info *fs_info = root->fs_info;
3371 int ret = 0, committed = 0, alloc_chunk = 1;
3373 /* make sure bytes are sectorsize aligned */
3374 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3376 if (root == root->fs_info->tree_root ||
3377 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3382 data_sinfo = fs_info->data_sinfo;
3387 /* make sure we have enough space to handle the data first */
3388 spin_lock(&data_sinfo->lock);
3389 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3390 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3391 data_sinfo->bytes_may_use;
3393 if (used + bytes > data_sinfo->total_bytes) {
3394 struct btrfs_trans_handle *trans;
3397 * if we don't have enough free bytes in this space then we need
3398 * to alloc a new chunk.
3400 if (!data_sinfo->full && alloc_chunk) {
3403 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3404 spin_unlock(&data_sinfo->lock);
3406 alloc_target = btrfs_get_alloc_profile(root, 1);
3407 trans = btrfs_join_transaction(root);
3409 return PTR_ERR(trans);
3411 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3413 CHUNK_ALLOC_NO_FORCE);
3414 btrfs_end_transaction(trans, root);
3423 data_sinfo = fs_info->data_sinfo;
3429 * If we have less pinned bytes than we want to allocate then
3430 * don't bother committing the transaction, it won't help us.
3432 if (data_sinfo->bytes_pinned < bytes)
3434 spin_unlock(&data_sinfo->lock);
3436 /* commit the current transaction and try again */
3439 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3441 trans = btrfs_join_transaction(root);
3443 return PTR_ERR(trans);
3444 ret = btrfs_commit_transaction(trans, root);
3452 data_sinfo->bytes_may_use += bytes;
3453 trace_btrfs_space_reservation(root->fs_info, "space_info",
3454 data_sinfo->flags, bytes, 1);
3455 spin_unlock(&data_sinfo->lock);
3461 * Called if we need to clear a data reservation for this inode.
3463 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3465 struct btrfs_root *root = BTRFS_I(inode)->root;
3466 struct btrfs_space_info *data_sinfo;
3468 /* make sure bytes are sectorsize aligned */
3469 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3471 data_sinfo = root->fs_info->data_sinfo;
3472 spin_lock(&data_sinfo->lock);
3473 data_sinfo->bytes_may_use -= bytes;
3474 trace_btrfs_space_reservation(root->fs_info, "space_info",
3475 data_sinfo->flags, bytes, 0);
3476 spin_unlock(&data_sinfo->lock);
3479 static void force_metadata_allocation(struct btrfs_fs_info *info)
3481 struct list_head *head = &info->space_info;
3482 struct btrfs_space_info *found;
3485 list_for_each_entry_rcu(found, head, list) {
3486 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3487 found->force_alloc = CHUNK_ALLOC_FORCE;
3492 static int should_alloc_chunk(struct btrfs_root *root,
3493 struct btrfs_space_info *sinfo, int force)
3495 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3496 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3497 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3500 if (force == CHUNK_ALLOC_FORCE)
3504 * We need to take into account the global rsv because for all intents
3505 * and purposes it's used space. Don't worry about locking the
3506 * global_rsv, it doesn't change except when the transaction commits.
3508 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3509 num_allocated += global_rsv->size;
3512 * in limited mode, we want to have some free space up to
3513 * about 1% of the FS size.
3515 if (force == CHUNK_ALLOC_LIMITED) {
3516 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3517 thresh = max_t(u64, 64 * 1024 * 1024,
3518 div_factor_fine(thresh, 1));
3520 if (num_bytes - num_allocated < thresh)
3524 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3529 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3533 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3534 type & BTRFS_BLOCK_GROUP_RAID0)
3535 num_dev = root->fs_info->fs_devices->rw_devices;
3536 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3539 num_dev = 1; /* DUP or single */
3541 /* metadata for updaing devices and chunk tree */
3542 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3545 static void check_system_chunk(struct btrfs_trans_handle *trans,
3546 struct btrfs_root *root, u64 type)
3548 struct btrfs_space_info *info;
3552 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3553 spin_lock(&info->lock);
3554 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3555 info->bytes_reserved - info->bytes_readonly;
3556 spin_unlock(&info->lock);
3558 thresh = get_system_chunk_thresh(root, type);
3559 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3560 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3561 left, thresh, type);
3562 dump_space_info(info, 0, 0);
3565 if (left < thresh) {
3568 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3569 btrfs_alloc_chunk(trans, root, flags);
3573 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3574 struct btrfs_root *extent_root, u64 flags, int force)
3576 struct btrfs_space_info *space_info;
3577 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3578 int wait_for_alloc = 0;
3581 /* Don't re-enter if we're already allocating a chunk */
3582 if (trans->allocating_chunk)
3585 space_info = __find_space_info(extent_root->fs_info, flags);
3587 ret = update_space_info(extent_root->fs_info, flags,
3589 BUG_ON(ret); /* -ENOMEM */
3591 BUG_ON(!space_info); /* Logic error */
3594 spin_lock(&space_info->lock);
3595 if (force < space_info->force_alloc)
3596 force = space_info->force_alloc;
3597 if (space_info->full) {
3598 spin_unlock(&space_info->lock);
3602 if (!should_alloc_chunk(extent_root, space_info, force)) {
3603 spin_unlock(&space_info->lock);
3605 } else if (space_info->chunk_alloc) {
3608 space_info->chunk_alloc = 1;
3611 spin_unlock(&space_info->lock);
3613 mutex_lock(&fs_info->chunk_mutex);
3616 * The chunk_mutex is held throughout the entirety of a chunk
3617 * allocation, so once we've acquired the chunk_mutex we know that the
3618 * other guy is done and we need to recheck and see if we should
3621 if (wait_for_alloc) {
3622 mutex_unlock(&fs_info->chunk_mutex);
3627 trans->allocating_chunk = true;
3630 * If we have mixed data/metadata chunks we want to make sure we keep
3631 * allocating mixed chunks instead of individual chunks.
3633 if (btrfs_mixed_space_info(space_info))
3634 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3637 * if we're doing a data chunk, go ahead and make sure that
3638 * we keep a reasonable number of metadata chunks allocated in the
3641 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3642 fs_info->data_chunk_allocations++;
3643 if (!(fs_info->data_chunk_allocations %
3644 fs_info->metadata_ratio))
3645 force_metadata_allocation(fs_info);
3649 * Check if we have enough space in SYSTEM chunk because we may need
3650 * to update devices.
3652 check_system_chunk(trans, extent_root, flags);
3654 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3655 trans->allocating_chunk = false;
3656 if (ret < 0 && ret != -ENOSPC)
3659 spin_lock(&space_info->lock);
3661 space_info->full = 1;
3665 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3666 space_info->chunk_alloc = 0;
3667 spin_unlock(&space_info->lock);
3669 mutex_unlock(&fs_info->chunk_mutex);
3673 static int can_overcommit(struct btrfs_root *root,
3674 struct btrfs_space_info *space_info, u64 bytes,
3675 enum btrfs_reserve_flush_enum flush)
3677 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3678 u64 profile = btrfs_get_alloc_profile(root, 0);
3684 used = space_info->bytes_used + space_info->bytes_reserved +
3685 space_info->bytes_pinned + space_info->bytes_readonly;
3687 spin_lock(&global_rsv->lock);
3688 rsv_size = global_rsv->size;
3689 spin_unlock(&global_rsv->lock);
3692 * We only want to allow over committing if we have lots of actual space
3693 * free, but if we don't have enough space to handle the global reserve
3694 * space then we could end up having a real enospc problem when trying
3695 * to allocate a chunk or some other such important allocation.
3698 if (used + rsv_size >= space_info->total_bytes)
3701 used += space_info->bytes_may_use;
3703 spin_lock(&root->fs_info->free_chunk_lock);
3704 avail = root->fs_info->free_chunk_space;
3705 spin_unlock(&root->fs_info->free_chunk_lock);
3708 * If we have dup, raid1 or raid10 then only half of the free
3709 * space is actually useable.
3711 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3712 BTRFS_BLOCK_GROUP_RAID1 |
3713 BTRFS_BLOCK_GROUP_RAID10))
3716 to_add = space_info->total_bytes;
3719 * If we aren't flushing all things, let us overcommit up to
3720 * 1/2th of the space. If we can flush, don't let us overcommit
3721 * too much, let it overcommit up to 1/8 of the space.
3723 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3729 * Limit the overcommit to the amount of free space we could possibly
3730 * allocate for chunks.
3732 to_add = min(avail, to_add);
3734 if (used + bytes < space_info->total_bytes + to_add)
3739 static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3740 unsigned long nr_pages,
3741 enum wb_reason reason)
3743 /* the flusher is dealing with the dirty inodes now. */
3744 if (writeback_in_progress(sb->s_bdi))
3747 if (down_read_trylock(&sb->s_umount)) {
3748 writeback_inodes_sb_nr(sb, nr_pages, reason);
3749 up_read(&sb->s_umount);
3756 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3757 unsigned long nr_pages)
3759 struct super_block *sb = root->fs_info->sb;
3762 /* If we can not start writeback, just sync all the delalloc file. */
3763 started = writeback_inodes_sb_nr_if_idle_safe(sb, nr_pages,
3764 WB_REASON_FS_FREE_SPACE);
3767 * We needn't worry the filesystem going from r/w to r/o though
3768 * we don't acquire ->s_umount mutex, because the filesystem
3769 * should guarantee the delalloc inodes list be empty after
3770 * the filesystem is readonly(all dirty pages are written to
3773 btrfs_start_delalloc_inodes(root, 0);
3774 btrfs_wait_ordered_extents(root, 0);
3779 * shrink metadata reservation for delalloc
3781 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3784 struct btrfs_block_rsv *block_rsv;
3785 struct btrfs_space_info *space_info;
3786 struct btrfs_trans_handle *trans;
3790 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3792 enum btrfs_reserve_flush_enum flush;
3794 trans = (struct btrfs_trans_handle *)current->journal_info;
3795 block_rsv = &root->fs_info->delalloc_block_rsv;
3796 space_info = block_rsv->space_info;
3799 delalloc_bytes = percpu_counter_sum_positive(
3800 &root->fs_info->delalloc_bytes);
3801 if (delalloc_bytes == 0) {
3804 btrfs_wait_ordered_extents(root, 0);
3808 while (delalloc_bytes && loops < 3) {
3809 max_reclaim = min(delalloc_bytes, to_reclaim);
3810 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3811 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3813 * We need to wait for the async pages to actually start before
3816 wait_event(root->fs_info->async_submit_wait,
3817 !atomic_read(&root->fs_info->async_delalloc_pages));
3820 flush = BTRFS_RESERVE_FLUSH_ALL;
3822 flush = BTRFS_RESERVE_NO_FLUSH;
3823 spin_lock(&space_info->lock);
3824 if (can_overcommit(root, space_info, orig, flush)) {
3825 spin_unlock(&space_info->lock);
3828 spin_unlock(&space_info->lock);
3831 if (wait_ordered && !trans) {
3832 btrfs_wait_ordered_extents(root, 0);
3834 time_left = schedule_timeout_killable(1);
3839 delalloc_bytes = percpu_counter_sum_positive(
3840 &root->fs_info->delalloc_bytes);
3845 * maybe_commit_transaction - possibly commit the transaction if its ok to
3846 * @root - the root we're allocating for
3847 * @bytes - the number of bytes we want to reserve
3848 * @force - force the commit
3850 * This will check to make sure that committing the transaction will actually
3851 * get us somewhere and then commit the transaction if it does. Otherwise it
3852 * will return -ENOSPC.
3854 static int may_commit_transaction(struct btrfs_root *root,
3855 struct btrfs_space_info *space_info,
3856 u64 bytes, int force)
3858 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3859 struct btrfs_trans_handle *trans;
3861 trans = (struct btrfs_trans_handle *)current->journal_info;
3868 /* See if there is enough pinned space to make this reservation */
3869 spin_lock(&space_info->lock);
3870 if (space_info->bytes_pinned >= bytes) {
3871 spin_unlock(&space_info->lock);
3874 spin_unlock(&space_info->lock);
3877 * See if there is some space in the delayed insertion reservation for
3880 if (space_info != delayed_rsv->space_info)
3883 spin_lock(&space_info->lock);
3884 spin_lock(&delayed_rsv->lock);
3885 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3886 spin_unlock(&delayed_rsv->lock);
3887 spin_unlock(&space_info->lock);
3890 spin_unlock(&delayed_rsv->lock);
3891 spin_unlock(&space_info->lock);
3894 trans = btrfs_join_transaction(root);
3898 return btrfs_commit_transaction(trans, root);
3902 FLUSH_DELAYED_ITEMS_NR = 1,
3903 FLUSH_DELAYED_ITEMS = 2,
3905 FLUSH_DELALLOC_WAIT = 4,
3910 static int flush_space(struct btrfs_root *root,
3911 struct btrfs_space_info *space_info, u64 num_bytes,
3912 u64 orig_bytes, int state)
3914 struct btrfs_trans_handle *trans;
3919 case FLUSH_DELAYED_ITEMS_NR:
3920 case FLUSH_DELAYED_ITEMS:
3921 if (state == FLUSH_DELAYED_ITEMS_NR) {
3922 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3924 nr = (int)div64_u64(num_bytes, bytes);
3931 trans = btrfs_join_transaction(root);
3932 if (IS_ERR(trans)) {
3933 ret = PTR_ERR(trans);
3936 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3937 btrfs_end_transaction(trans, root);
3939 case FLUSH_DELALLOC:
3940 case FLUSH_DELALLOC_WAIT:
3941 shrink_delalloc(root, num_bytes, orig_bytes,
3942 state == FLUSH_DELALLOC_WAIT);
3945 trans = btrfs_join_transaction(root);
3946 if (IS_ERR(trans)) {
3947 ret = PTR_ERR(trans);
3950 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3951 btrfs_get_alloc_profile(root, 0),
3952 CHUNK_ALLOC_NO_FORCE);
3953 btrfs_end_transaction(trans, root);
3958 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3968 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3969 * @root - the root we're allocating for
3970 * @block_rsv - the block_rsv we're allocating for
3971 * @orig_bytes - the number of bytes we want
3972 * @flush - whether or not we can flush to make our reservation
3974 * This will reserve orgi_bytes number of bytes from the space info associated
3975 * with the block_rsv. If there is not enough space it will make an attempt to
3976 * flush out space to make room. It will do this by flushing delalloc if
3977 * possible or committing the transaction. If flush is 0 then no attempts to
3978 * regain reservations will be made and this will fail if there is not enough
3981 static int reserve_metadata_bytes(struct btrfs_root *root,
3982 struct btrfs_block_rsv *block_rsv,
3984 enum btrfs_reserve_flush_enum flush)
3986 struct btrfs_space_info *space_info = block_rsv->space_info;
3988 u64 num_bytes = orig_bytes;
3989 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3991 bool flushing = false;
3995 spin_lock(&space_info->lock);
3997 * We only want to wait if somebody other than us is flushing and we
3998 * are actually allowed to flush all things.
4000 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4001 space_info->flush) {
4002 spin_unlock(&space_info->lock);
4004 * If we have a trans handle we can't wait because the flusher
4005 * may have to commit the transaction, which would mean we would
4006 * deadlock since we are waiting for the flusher to finish, but
4007 * hold the current transaction open.
4009 if (current->journal_info)
4011 ret = wait_event_killable(space_info->wait, !space_info->flush);
4012 /* Must have been killed, return */
4016 spin_lock(&space_info->lock);
4020 used = space_info->bytes_used + space_info->bytes_reserved +
4021 space_info->bytes_pinned + space_info->bytes_readonly +
4022 space_info->bytes_may_use;
4025 * The idea here is that we've not already over-reserved the block group
4026 * then we can go ahead and save our reservation first and then start
4027 * flushing if we need to. Otherwise if we've already overcommitted
4028 * lets start flushing stuff first and then come back and try to make
4031 if (used <= space_info->total_bytes) {
4032 if (used + orig_bytes <= space_info->total_bytes) {
4033 space_info->bytes_may_use += orig_bytes;
4034 trace_btrfs_space_reservation(root->fs_info,
4035 "space_info", space_info->flags, orig_bytes, 1);
4039 * Ok set num_bytes to orig_bytes since we aren't
4040 * overocmmitted, this way we only try and reclaim what
4043 num_bytes = orig_bytes;
4047 * Ok we're over committed, set num_bytes to the overcommitted
4048 * amount plus the amount of bytes that we need for this
4051 num_bytes = used - space_info->total_bytes +
4055 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4056 space_info->bytes_may_use += orig_bytes;
4057 trace_btrfs_space_reservation(root->fs_info, "space_info",
4058 space_info->flags, orig_bytes,
4064 * Couldn't make our reservation, save our place so while we're trying
4065 * to reclaim space we can actually use it instead of somebody else
4066 * stealing it from us.
4068 * We make the other tasks wait for the flush only when we can flush
4071 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4073 space_info->flush = 1;
4076 spin_unlock(&space_info->lock);
4078 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4081 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4086 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4087 * would happen. So skip delalloc flush.
4089 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4090 (flush_state == FLUSH_DELALLOC ||
4091 flush_state == FLUSH_DELALLOC_WAIT))
4092 flush_state = ALLOC_CHUNK;
4096 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4097 flush_state < COMMIT_TRANS)
4099 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4100 flush_state <= COMMIT_TRANS)
4104 if (ret == -ENOSPC &&
4105 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4106 struct btrfs_block_rsv *global_rsv =
4107 &root->fs_info->global_block_rsv;
4109 if (block_rsv != global_rsv &&
4110 !block_rsv_use_bytes(global_rsv, orig_bytes))
4114 spin_lock(&space_info->lock);
4115 space_info->flush = 0;
4116 wake_up_all(&space_info->wait);
4117 spin_unlock(&space_info->lock);
4122 static struct btrfs_block_rsv *get_block_rsv(
4123 const struct btrfs_trans_handle *trans,
4124 const struct btrfs_root *root)
4126 struct btrfs_block_rsv *block_rsv = NULL;
4129 block_rsv = trans->block_rsv;
4131 if (root == root->fs_info->csum_root && trans->adding_csums)
4132 block_rsv = trans->block_rsv;
4135 block_rsv = root->block_rsv;
4138 block_rsv = &root->fs_info->empty_block_rsv;
4143 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4147 spin_lock(&block_rsv->lock);
4148 if (block_rsv->reserved >= num_bytes) {
4149 block_rsv->reserved -= num_bytes;
4150 if (block_rsv->reserved < block_rsv->size)
4151 block_rsv->full = 0;
4154 spin_unlock(&block_rsv->lock);
4158 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4159 u64 num_bytes, int update_size)
4161 spin_lock(&block_rsv->lock);
4162 block_rsv->reserved += num_bytes;
4164 block_rsv->size += num_bytes;
4165 else if (block_rsv->reserved >= block_rsv->size)
4166 block_rsv->full = 1;
4167 spin_unlock(&block_rsv->lock);
4170 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4171 struct btrfs_block_rsv *block_rsv,
4172 struct btrfs_block_rsv *dest, u64 num_bytes)
4174 struct btrfs_space_info *space_info = block_rsv->space_info;
4176 spin_lock(&block_rsv->lock);
4177 if (num_bytes == (u64)-1)
4178 num_bytes = block_rsv->size;
4179 block_rsv->size -= num_bytes;
4180 if (block_rsv->reserved >= block_rsv->size) {
4181 num_bytes = block_rsv->reserved - block_rsv->size;
4182 block_rsv->reserved = block_rsv->size;
4183 block_rsv->full = 1;
4187 spin_unlock(&block_rsv->lock);
4189 if (num_bytes > 0) {
4191 spin_lock(&dest->lock);
4195 bytes_to_add = dest->size - dest->reserved;
4196 bytes_to_add = min(num_bytes, bytes_to_add);
4197 dest->reserved += bytes_to_add;
4198 if (dest->reserved >= dest->size)
4200 num_bytes -= bytes_to_add;
4202 spin_unlock(&dest->lock);
4205 spin_lock(&space_info->lock);
4206 space_info->bytes_may_use -= num_bytes;
4207 trace_btrfs_space_reservation(fs_info, "space_info",
4208 space_info->flags, num_bytes, 0);
4209 space_info->reservation_progress++;
4210 spin_unlock(&space_info->lock);
4215 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4216 struct btrfs_block_rsv *dst, u64 num_bytes)
4220 ret = block_rsv_use_bytes(src, num_bytes);
4224 block_rsv_add_bytes(dst, num_bytes, 1);
4228 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4230 memset(rsv, 0, sizeof(*rsv));
4231 spin_lock_init(&rsv->lock);
4235 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4236 unsigned short type)
4238 struct btrfs_block_rsv *block_rsv;
4239 struct btrfs_fs_info *fs_info = root->fs_info;
4241 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4245 btrfs_init_block_rsv(block_rsv, type);
4246 block_rsv->space_info = __find_space_info(fs_info,
4247 BTRFS_BLOCK_GROUP_METADATA);
4251 void btrfs_free_block_rsv(struct btrfs_root *root,
4252 struct btrfs_block_rsv *rsv)
4256 btrfs_block_rsv_release(root, rsv, (u64)-1);
4260 int btrfs_block_rsv_add(struct btrfs_root *root,
4261 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4262 enum btrfs_reserve_flush_enum flush)
4269 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4271 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4278 int btrfs_block_rsv_check(struct btrfs_root *root,
4279 struct btrfs_block_rsv *block_rsv, int min_factor)
4287 spin_lock(&block_rsv->lock);
4288 num_bytes = div_factor(block_rsv->size, min_factor);
4289 if (block_rsv->reserved >= num_bytes)
4291 spin_unlock(&block_rsv->lock);
4296 int btrfs_block_rsv_refill(struct btrfs_root *root,
4297 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4298 enum btrfs_reserve_flush_enum flush)
4306 spin_lock(&block_rsv->lock);
4307 num_bytes = min_reserved;
4308 if (block_rsv->reserved >= num_bytes)
4311 num_bytes -= block_rsv->reserved;
4312 spin_unlock(&block_rsv->lock);
4317 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4319 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4326 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4327 struct btrfs_block_rsv *dst_rsv,
4330 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4333 void btrfs_block_rsv_release(struct btrfs_root *root,
4334 struct btrfs_block_rsv *block_rsv,
4337 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4338 if (global_rsv->full || global_rsv == block_rsv ||
4339 block_rsv->space_info != global_rsv->space_info)
4341 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4346 * helper to calculate size of global block reservation.
4347 * the desired value is sum of space used by extent tree,
4348 * checksum tree and root tree
4350 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4352 struct btrfs_space_info *sinfo;
4356 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4358 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4359 spin_lock(&sinfo->lock);
4360 data_used = sinfo->bytes_used;
4361 spin_unlock(&sinfo->lock);
4363 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4364 spin_lock(&sinfo->lock);
4365 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4367 meta_used = sinfo->bytes_used;
4368 spin_unlock(&sinfo->lock);
4370 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4372 num_bytes += div64_u64(data_used + meta_used, 50);
4374 if (num_bytes * 3 > meta_used)
4375 num_bytes = div64_u64(meta_used, 3);
4377 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4380 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4382 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4383 struct btrfs_space_info *sinfo = block_rsv->space_info;
4386 num_bytes = calc_global_metadata_size(fs_info);
4388 spin_lock(&sinfo->lock);
4389 spin_lock(&block_rsv->lock);
4391 block_rsv->size = num_bytes;
4393 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4394 sinfo->bytes_reserved + sinfo->bytes_readonly +
4395 sinfo->bytes_may_use;
4397 if (sinfo->total_bytes > num_bytes) {
4398 num_bytes = sinfo->total_bytes - num_bytes;
4399 block_rsv->reserved += num_bytes;
4400 sinfo->bytes_may_use += num_bytes;
4401 trace_btrfs_space_reservation(fs_info, "space_info",
4402 sinfo->flags, num_bytes, 1);
4405 if (block_rsv->reserved >= block_rsv->size) {
4406 num_bytes = block_rsv->reserved - block_rsv->size;
4407 sinfo->bytes_may_use -= num_bytes;
4408 trace_btrfs_space_reservation(fs_info, "space_info",
4409 sinfo->flags, num_bytes, 0);
4410 sinfo->reservation_progress++;
4411 block_rsv->reserved = block_rsv->size;
4412 block_rsv->full = 1;
4415 spin_unlock(&block_rsv->lock);
4416 spin_unlock(&sinfo->lock);
4419 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4421 struct btrfs_space_info *space_info;
4423 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4424 fs_info->chunk_block_rsv.space_info = space_info;
4426 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4427 fs_info->global_block_rsv.space_info = space_info;
4428 fs_info->delalloc_block_rsv.space_info = space_info;
4429 fs_info->trans_block_rsv.space_info = space_info;
4430 fs_info->empty_block_rsv.space_info = space_info;
4431 fs_info->delayed_block_rsv.space_info = space_info;
4433 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4434 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4435 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4436 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4437 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4439 update_global_block_rsv(fs_info);
4442 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4444 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4446 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4447 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4448 WARN_ON(fs_info->trans_block_rsv.size > 0);
4449 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4450 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4451 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4452 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4453 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4456 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4457 struct btrfs_root *root)
4459 if (!trans->block_rsv)
4462 if (!trans->bytes_reserved)
4465 trace_btrfs_space_reservation(root->fs_info, "transaction",
4466 trans->transid, trans->bytes_reserved, 0);
4467 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4468 trans->bytes_reserved = 0;
4471 /* Can only return 0 or -ENOSPC */
4472 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4473 struct inode *inode)
4475 struct btrfs_root *root = BTRFS_I(inode)->root;
4476 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4477 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4480 * We need to hold space in order to delete our orphan item once we've
4481 * added it, so this takes the reservation so we can release it later
4482 * when we are truly done with the orphan item.
4484 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4485 trace_btrfs_space_reservation(root->fs_info, "orphan",
4486 btrfs_ino(inode), num_bytes, 1);
4487 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4490 void btrfs_orphan_release_metadata(struct inode *inode)
4492 struct btrfs_root *root = BTRFS_I(inode)->root;
4493 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4494 trace_btrfs_space_reservation(root->fs_info, "orphan",
4495 btrfs_ino(inode), num_bytes, 0);
4496 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4499 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4500 struct btrfs_pending_snapshot *pending)
4502 struct btrfs_root *root = pending->root;
4503 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4504 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4506 * two for root back/forward refs, two for directory entries,
4507 * one for root of the snapshot and one for parent inode.
4509 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4510 dst_rsv->space_info = src_rsv->space_info;
4511 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4515 * drop_outstanding_extent - drop an outstanding extent
4516 * @inode: the inode we're dropping the extent for
4518 * This is called when we are freeing up an outstanding extent, either called
4519 * after an error or after an extent is written. This will return the number of
4520 * reserved extents that need to be freed. This must be called with
4521 * BTRFS_I(inode)->lock held.
4523 static unsigned drop_outstanding_extent(struct inode *inode)
4525 unsigned drop_inode_space = 0;
4526 unsigned dropped_extents = 0;
4528 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4529 BTRFS_I(inode)->outstanding_extents--;
4531 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4532 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4533 &BTRFS_I(inode)->runtime_flags))
4534 drop_inode_space = 1;
4537 * If we have more or the same amount of outsanding extents than we have
4538 * reserved then we need to leave the reserved extents count alone.
4540 if (BTRFS_I(inode)->outstanding_extents >=
4541 BTRFS_I(inode)->reserved_extents)
4542 return drop_inode_space;
4544 dropped_extents = BTRFS_I(inode)->reserved_extents -
4545 BTRFS_I(inode)->outstanding_extents;
4546 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4547 return dropped_extents + drop_inode_space;
4551 * calc_csum_metadata_size - return the amount of metada space that must be
4552 * reserved/free'd for the given bytes.
4553 * @inode: the inode we're manipulating
4554 * @num_bytes: the number of bytes in question
4555 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4557 * This adjusts the number of csum_bytes in the inode and then returns the
4558 * correct amount of metadata that must either be reserved or freed. We
4559 * calculate how many checksums we can fit into one leaf and then divide the
4560 * number of bytes that will need to be checksumed by this value to figure out
4561 * how many checksums will be required. If we are adding bytes then the number
4562 * may go up and we will return the number of additional bytes that must be
4563 * reserved. If it is going down we will return the number of bytes that must
4566 * This must be called with BTRFS_I(inode)->lock held.
4568 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4571 struct btrfs_root *root = BTRFS_I(inode)->root;
4573 int num_csums_per_leaf;
4577 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4578 BTRFS_I(inode)->csum_bytes == 0)
4581 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4583 BTRFS_I(inode)->csum_bytes += num_bytes;
4585 BTRFS_I(inode)->csum_bytes -= num_bytes;
4586 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4587 num_csums_per_leaf = (int)div64_u64(csum_size,
4588 sizeof(struct btrfs_csum_item) +
4589 sizeof(struct btrfs_disk_key));
4590 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4591 num_csums = num_csums + num_csums_per_leaf - 1;
4592 num_csums = num_csums / num_csums_per_leaf;
4594 old_csums = old_csums + num_csums_per_leaf - 1;
4595 old_csums = old_csums / num_csums_per_leaf;
4597 /* No change, no need to reserve more */
4598 if (old_csums == num_csums)
4602 return btrfs_calc_trans_metadata_size(root,
4603 num_csums - old_csums);
4605 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4608 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4610 struct btrfs_root *root = BTRFS_I(inode)->root;
4611 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4614 unsigned nr_extents = 0;
4615 int extra_reserve = 0;
4616 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4618 bool delalloc_lock = true;
4620 /* If we are a free space inode we need to not flush since we will be in
4621 * the middle of a transaction commit. We also don't need the delalloc
4622 * mutex since we won't race with anybody. We need this mostly to make
4623 * lockdep shut its filthy mouth.
4625 if (btrfs_is_free_space_inode(inode)) {
4626 flush = BTRFS_RESERVE_NO_FLUSH;
4627 delalloc_lock = false;
4630 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4631 btrfs_transaction_in_commit(root->fs_info))
4632 schedule_timeout(1);
4635 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4637 num_bytes = ALIGN(num_bytes, root->sectorsize);
4639 spin_lock(&BTRFS_I(inode)->lock);
4640 BTRFS_I(inode)->outstanding_extents++;
4642 if (BTRFS_I(inode)->outstanding_extents >
4643 BTRFS_I(inode)->reserved_extents)
4644 nr_extents = BTRFS_I(inode)->outstanding_extents -
4645 BTRFS_I(inode)->reserved_extents;
4648 * Add an item to reserve for updating the inode when we complete the
4651 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4652 &BTRFS_I(inode)->runtime_flags)) {
4657 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4658 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4659 csum_bytes = BTRFS_I(inode)->csum_bytes;
4660 spin_unlock(&BTRFS_I(inode)->lock);
4662 if (root->fs_info->quota_enabled)
4663 ret = btrfs_qgroup_reserve(root, num_bytes +
4664 nr_extents * root->leafsize);
4667 * ret != 0 here means the qgroup reservation failed, we go straight to
4668 * the shared error handling then.
4671 ret = reserve_metadata_bytes(root, block_rsv,
4678 spin_lock(&BTRFS_I(inode)->lock);
4679 dropped = drop_outstanding_extent(inode);
4681 * If the inodes csum_bytes is the same as the original
4682 * csum_bytes then we know we haven't raced with any free()ers
4683 * so we can just reduce our inodes csum bytes and carry on.
4684 * Otherwise we have to do the normal free thing to account for
4685 * the case that the free side didn't free up its reserve
4686 * because of this outstanding reservation.
4688 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4689 calc_csum_metadata_size(inode, num_bytes, 0);
4691 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4692 spin_unlock(&BTRFS_I(inode)->lock);
4694 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4697 btrfs_block_rsv_release(root, block_rsv, to_free);
4698 trace_btrfs_space_reservation(root->fs_info,
4703 if (root->fs_info->quota_enabled) {
4704 btrfs_qgroup_free(root, num_bytes +
4705 nr_extents * root->leafsize);
4708 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4712 spin_lock(&BTRFS_I(inode)->lock);
4713 if (extra_reserve) {
4714 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4715 &BTRFS_I(inode)->runtime_flags);
4718 BTRFS_I(inode)->reserved_extents += nr_extents;
4719 spin_unlock(&BTRFS_I(inode)->lock);
4722 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4725 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4726 btrfs_ino(inode), to_reserve, 1);
4727 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4733 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4734 * @inode: the inode to release the reservation for
4735 * @num_bytes: the number of bytes we're releasing
4737 * This will release the metadata reservation for an inode. This can be called
4738 * once we complete IO for a given set of bytes to release their metadata
4741 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4743 struct btrfs_root *root = BTRFS_I(inode)->root;
4747 num_bytes = ALIGN(num_bytes, root->sectorsize);
4748 spin_lock(&BTRFS_I(inode)->lock);
4749 dropped = drop_outstanding_extent(inode);
4752 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4753 spin_unlock(&BTRFS_I(inode)->lock);
4755 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4757 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4758 btrfs_ino(inode), to_free, 0);
4759 if (root->fs_info->quota_enabled) {
4760 btrfs_qgroup_free(root, num_bytes +
4761 dropped * root->leafsize);
4764 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4769 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4770 * @inode: inode we're writing to
4771 * @num_bytes: the number of bytes we want to allocate
4773 * This will do the following things
4775 * o reserve space in the data space info for num_bytes
4776 * o reserve space in the metadata space info based on number of outstanding
4777 * extents and how much csums will be needed
4778 * o add to the inodes ->delalloc_bytes
4779 * o add it to the fs_info's delalloc inodes list.
4781 * This will return 0 for success and -ENOSPC if there is no space left.
4783 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4787 ret = btrfs_check_data_free_space(inode, num_bytes);
4791 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4793 btrfs_free_reserved_data_space(inode, num_bytes);
4801 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4802 * @inode: inode we're releasing space for
4803 * @num_bytes: the number of bytes we want to free up
4805 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4806 * called in the case that we don't need the metadata AND data reservations
4807 * anymore. So if there is an error or we insert an inline extent.
4809 * This function will release the metadata space that was not used and will
4810 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4811 * list if there are no delalloc bytes left.
4813 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4815 btrfs_delalloc_release_metadata(inode, num_bytes);
4816 btrfs_free_reserved_data_space(inode, num_bytes);
4819 static int update_block_group(struct btrfs_root *root,
4820 u64 bytenr, u64 num_bytes, int alloc)
4822 struct btrfs_block_group_cache *cache = NULL;
4823 struct btrfs_fs_info *info = root->fs_info;
4824 u64 total = num_bytes;
4829 /* block accounting for super block */
4830 spin_lock(&info->delalloc_lock);
4831 old_val = btrfs_super_bytes_used(info->super_copy);
4833 old_val += num_bytes;
4835 old_val -= num_bytes;
4836 btrfs_set_super_bytes_used(info->super_copy, old_val);
4837 spin_unlock(&info->delalloc_lock);
4840 cache = btrfs_lookup_block_group(info, bytenr);
4843 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4844 BTRFS_BLOCK_GROUP_RAID1 |
4845 BTRFS_BLOCK_GROUP_RAID10))
4850 * If this block group has free space cache written out, we
4851 * need to make sure to load it if we are removing space. This
4852 * is because we need the unpinning stage to actually add the
4853 * space back to the block group, otherwise we will leak space.
4855 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4856 cache_block_group(cache, 1);
4858 byte_in_group = bytenr - cache->key.objectid;
4859 WARN_ON(byte_in_group > cache->key.offset);
4861 spin_lock(&cache->space_info->lock);
4862 spin_lock(&cache->lock);
4864 if (btrfs_test_opt(root, SPACE_CACHE) &&
4865 cache->disk_cache_state < BTRFS_DC_CLEAR)
4866 cache->disk_cache_state = BTRFS_DC_CLEAR;
4869 old_val = btrfs_block_group_used(&cache->item);
4870 num_bytes = min(total, cache->key.offset - byte_in_group);
4872 old_val += num_bytes;
4873 btrfs_set_block_group_used(&cache->item, old_val);
4874 cache->reserved -= num_bytes;
4875 cache->space_info->bytes_reserved -= num_bytes;
4876 cache->space_info->bytes_used += num_bytes;
4877 cache->space_info->disk_used += num_bytes * factor;
4878 spin_unlock(&cache->lock);
4879 spin_unlock(&cache->space_info->lock);
4881 old_val -= num_bytes;
4882 btrfs_set_block_group_used(&cache->item, old_val);
4883 cache->pinned += num_bytes;
4884 cache->space_info->bytes_pinned += num_bytes;
4885 cache->space_info->bytes_used -= num_bytes;
4886 cache->space_info->disk_used -= num_bytes * factor;
4887 spin_unlock(&cache->lock);
4888 spin_unlock(&cache->space_info->lock);
4890 set_extent_dirty(info->pinned_extents,
4891 bytenr, bytenr + num_bytes - 1,
4892 GFP_NOFS | __GFP_NOFAIL);
4894 btrfs_put_block_group(cache);
4896 bytenr += num_bytes;
4901 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4903 struct btrfs_block_group_cache *cache;
4906 spin_lock(&root->fs_info->block_group_cache_lock);
4907 bytenr = root->fs_info->first_logical_byte;
4908 spin_unlock(&root->fs_info->block_group_cache_lock);
4910 if (bytenr < (u64)-1)
4913 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4917 bytenr = cache->key.objectid;
4918 btrfs_put_block_group(cache);
4923 static int pin_down_extent(struct btrfs_root *root,
4924 struct btrfs_block_group_cache *cache,
4925 u64 bytenr, u64 num_bytes, int reserved)
4927 spin_lock(&cache->space_info->lock);
4928 spin_lock(&cache->lock);
4929 cache->pinned += num_bytes;
4930 cache->space_info->bytes_pinned += num_bytes;
4932 cache->reserved -= num_bytes;
4933 cache->space_info->bytes_reserved -= num_bytes;
4935 spin_unlock(&cache->lock);
4936 spin_unlock(&cache->space_info->lock);
4938 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4939 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4944 * this function must be called within transaction
4946 int btrfs_pin_extent(struct btrfs_root *root,
4947 u64 bytenr, u64 num_bytes, int reserved)
4949 struct btrfs_block_group_cache *cache;
4951 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4952 BUG_ON(!cache); /* Logic error */
4954 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4956 btrfs_put_block_group(cache);
4961 * this function must be called within transaction
4963 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
4964 u64 bytenr, u64 num_bytes)
4966 struct btrfs_block_group_cache *cache;
4968 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4969 BUG_ON(!cache); /* Logic error */
4972 * pull in the free space cache (if any) so that our pin
4973 * removes the free space from the cache. We have load_only set
4974 * to one because the slow code to read in the free extents does check
4975 * the pinned extents.
4977 cache_block_group(cache, 1);
4979 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4981 /* remove us from the free space cache (if we're there at all) */
4982 btrfs_remove_free_space(cache, bytenr, num_bytes);
4983 btrfs_put_block_group(cache);
4988 * btrfs_update_reserved_bytes - update the block_group and space info counters
4989 * @cache: The cache we are manipulating
4990 * @num_bytes: The number of bytes in question
4991 * @reserve: One of the reservation enums
4993 * This is called by the allocator when it reserves space, or by somebody who is
4994 * freeing space that was never actually used on disk. For example if you
4995 * reserve some space for a new leaf in transaction A and before transaction A
4996 * commits you free that leaf, you call this with reserve set to 0 in order to
4997 * clear the reservation.
4999 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5000 * ENOSPC accounting. For data we handle the reservation through clearing the
5001 * delalloc bits in the io_tree. We have to do this since we could end up
5002 * allocating less disk space for the amount of data we have reserved in the
5003 * case of compression.
5005 * If this is a reservation and the block group has become read only we cannot
5006 * make the reservation and return -EAGAIN, otherwise this function always
5009 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5010 u64 num_bytes, int reserve)
5012 struct btrfs_space_info *space_info = cache->space_info;
5015 spin_lock(&space_info->lock);
5016 spin_lock(&cache->lock);
5017 if (reserve != RESERVE_FREE) {
5021 cache->reserved += num_bytes;
5022 space_info->bytes_reserved += num_bytes;
5023 if (reserve == RESERVE_ALLOC) {
5024 trace_btrfs_space_reservation(cache->fs_info,
5025 "space_info", space_info->flags,
5027 space_info->bytes_may_use -= num_bytes;
5032 space_info->bytes_readonly += num_bytes;
5033 cache->reserved -= num_bytes;
5034 space_info->bytes_reserved -= num_bytes;
5035 space_info->reservation_progress++;
5037 spin_unlock(&cache->lock);
5038 spin_unlock(&space_info->lock);
5042 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5043 struct btrfs_root *root)
5045 struct btrfs_fs_info *fs_info = root->fs_info;
5046 struct btrfs_caching_control *next;
5047 struct btrfs_caching_control *caching_ctl;
5048 struct btrfs_block_group_cache *cache;
5050 down_write(&fs_info->extent_commit_sem);
5052 list_for_each_entry_safe(caching_ctl, next,
5053 &fs_info->caching_block_groups, list) {
5054 cache = caching_ctl->block_group;
5055 if (block_group_cache_done(cache)) {
5056 cache->last_byte_to_unpin = (u64)-1;
5057 list_del_init(&caching_ctl->list);
5058 put_caching_control(caching_ctl);
5060 cache->last_byte_to_unpin = caching_ctl->progress;
5064 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5065 fs_info->pinned_extents = &fs_info->freed_extents[1];
5067 fs_info->pinned_extents = &fs_info->freed_extents[0];
5069 up_write(&fs_info->extent_commit_sem);
5071 update_global_block_rsv(fs_info);
5074 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5076 struct btrfs_fs_info *fs_info = root->fs_info;
5077 struct btrfs_block_group_cache *cache = NULL;
5078 struct btrfs_space_info *space_info;
5079 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5083 while (start <= end) {
5086 start >= cache->key.objectid + cache->key.offset) {
5088 btrfs_put_block_group(cache);
5089 cache = btrfs_lookup_block_group(fs_info, start);
5090 BUG_ON(!cache); /* Logic error */
5093 len = cache->key.objectid + cache->key.offset - start;
5094 len = min(len, end + 1 - start);
5096 if (start < cache->last_byte_to_unpin) {
5097 len = min(len, cache->last_byte_to_unpin - start);
5098 btrfs_add_free_space(cache, start, len);
5102 space_info = cache->space_info;
5104 spin_lock(&space_info->lock);
5105 spin_lock(&cache->lock);
5106 cache->pinned -= len;
5107 space_info->bytes_pinned -= len;
5109 space_info->bytes_readonly += len;
5112 spin_unlock(&cache->lock);
5113 if (!readonly && global_rsv->space_info == space_info) {
5114 spin_lock(&global_rsv->lock);
5115 if (!global_rsv->full) {
5116 len = min(len, global_rsv->size -
5117 global_rsv->reserved);
5118 global_rsv->reserved += len;
5119 space_info->bytes_may_use += len;
5120 if (global_rsv->reserved >= global_rsv->size)
5121 global_rsv->full = 1;
5123 spin_unlock(&global_rsv->lock);
5125 spin_unlock(&space_info->lock);
5129 btrfs_put_block_group(cache);
5133 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5134 struct btrfs_root *root)
5136 struct btrfs_fs_info *fs_info = root->fs_info;
5137 struct extent_io_tree *unpin;
5145 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5146 unpin = &fs_info->freed_extents[1];
5148 unpin = &fs_info->freed_extents[0];
5151 ret = find_first_extent_bit(unpin, 0, &start, &end,
5152 EXTENT_DIRTY, NULL);
5156 if (btrfs_test_opt(root, DISCARD))
5157 ret = btrfs_discard_extent(root, start,
5158 end + 1 - start, NULL);
5160 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5161 unpin_extent_range(root, start, end);
5168 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5169 struct btrfs_root *root,
5170 u64 bytenr, u64 num_bytes, u64 parent,
5171 u64 root_objectid, u64 owner_objectid,
5172 u64 owner_offset, int refs_to_drop,
5173 struct btrfs_delayed_extent_op *extent_op)
5175 struct btrfs_key key;
5176 struct btrfs_path *path;
5177 struct btrfs_fs_info *info = root->fs_info;
5178 struct btrfs_root *extent_root = info->extent_root;
5179 struct extent_buffer *leaf;
5180 struct btrfs_extent_item *ei;
5181 struct btrfs_extent_inline_ref *iref;
5184 int extent_slot = 0;
5185 int found_extent = 0;
5190 path = btrfs_alloc_path();
5195 path->leave_spinning = 1;
5197 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5198 BUG_ON(!is_data && refs_to_drop != 1);
5200 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5201 bytenr, num_bytes, parent,
5202 root_objectid, owner_objectid,
5205 extent_slot = path->slots[0];
5206 while (extent_slot >= 0) {
5207 btrfs_item_key_to_cpu(path->nodes[0], &key,
5209 if (key.objectid != bytenr)
5211 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5212 key.offset == num_bytes) {
5216 if (path->slots[0] - extent_slot > 5)
5220 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5221 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5222 if (found_extent && item_size < sizeof(*ei))
5225 if (!found_extent) {
5227 ret = remove_extent_backref(trans, extent_root, path,
5231 btrfs_abort_transaction(trans, extent_root, ret);
5234 btrfs_release_path(path);
5235 path->leave_spinning = 1;
5237 key.objectid = bytenr;
5238 key.type = BTRFS_EXTENT_ITEM_KEY;
5239 key.offset = num_bytes;
5241 ret = btrfs_search_slot(trans, extent_root,
5244 printk(KERN_ERR "umm, got %d back from search"
5245 ", was looking for %llu\n", ret,
5246 (unsigned long long)bytenr);
5248 btrfs_print_leaf(extent_root,
5252 btrfs_abort_transaction(trans, extent_root, ret);
5255 extent_slot = path->slots[0];
5257 } else if (ret == -ENOENT) {
5258 btrfs_print_leaf(extent_root, path->nodes[0]);
5260 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5261 "parent %llu root %llu owner %llu offset %llu\n",
5262 (unsigned long long)bytenr,
5263 (unsigned long long)parent,
5264 (unsigned long long)root_objectid,
5265 (unsigned long long)owner_objectid,
5266 (unsigned long long)owner_offset);
5268 btrfs_abort_transaction(trans, extent_root, ret);
5272 leaf = path->nodes[0];
5273 item_size = btrfs_item_size_nr(leaf, extent_slot);
5274 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5275 if (item_size < sizeof(*ei)) {
5276 BUG_ON(found_extent || extent_slot != path->slots[0]);
5277 ret = convert_extent_item_v0(trans, extent_root, path,
5280 btrfs_abort_transaction(trans, extent_root, ret);
5284 btrfs_release_path(path);
5285 path->leave_spinning = 1;
5287 key.objectid = bytenr;
5288 key.type = BTRFS_EXTENT_ITEM_KEY;
5289 key.offset = num_bytes;
5291 ret = btrfs_search_slot(trans, extent_root, &key, path,
5294 printk(KERN_ERR "umm, got %d back from search"
5295 ", was looking for %llu\n", ret,
5296 (unsigned long long)bytenr);
5297 btrfs_print_leaf(extent_root, path->nodes[0]);
5300 btrfs_abort_transaction(trans, extent_root, ret);
5304 extent_slot = path->slots[0];
5305 leaf = path->nodes[0];
5306 item_size = btrfs_item_size_nr(leaf, extent_slot);
5309 BUG_ON(item_size < sizeof(*ei));
5310 ei = btrfs_item_ptr(leaf, extent_slot,
5311 struct btrfs_extent_item);
5312 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5313 struct btrfs_tree_block_info *bi;
5314 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5315 bi = (struct btrfs_tree_block_info *)(ei + 1);
5316 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5319 refs = btrfs_extent_refs(leaf, ei);
5320 BUG_ON(refs < refs_to_drop);
5321 refs -= refs_to_drop;
5325 __run_delayed_extent_op(extent_op, leaf, ei);
5327 * In the case of inline back ref, reference count will
5328 * be updated by remove_extent_backref
5331 BUG_ON(!found_extent);
5333 btrfs_set_extent_refs(leaf, ei, refs);
5334 btrfs_mark_buffer_dirty(leaf);
5337 ret = remove_extent_backref(trans, extent_root, path,
5341 btrfs_abort_transaction(trans, extent_root, ret);
5347 BUG_ON(is_data && refs_to_drop !=
5348 extent_data_ref_count(root, path, iref));
5350 BUG_ON(path->slots[0] != extent_slot);
5352 BUG_ON(path->slots[0] != extent_slot + 1);
5353 path->slots[0] = extent_slot;
5358 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5361 btrfs_abort_transaction(trans, extent_root, ret);
5364 btrfs_release_path(path);
5367 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5369 btrfs_abort_transaction(trans, extent_root, ret);
5374 ret = update_block_group(root, bytenr, num_bytes, 0);
5376 btrfs_abort_transaction(trans, extent_root, ret);
5381 btrfs_free_path(path);
5386 * when we free an block, it is possible (and likely) that we free the last
5387 * delayed ref for that extent as well. This searches the delayed ref tree for
5388 * a given extent, and if there are no other delayed refs to be processed, it
5389 * removes it from the tree.
5391 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5392 struct btrfs_root *root, u64 bytenr)
5394 struct btrfs_delayed_ref_head *head;
5395 struct btrfs_delayed_ref_root *delayed_refs;
5396 struct btrfs_delayed_ref_node *ref;
5397 struct rb_node *node;
5400 delayed_refs = &trans->transaction->delayed_refs;
5401 spin_lock(&delayed_refs->lock);
5402 head = btrfs_find_delayed_ref_head(trans, bytenr);
5406 node = rb_prev(&head->node.rb_node);
5410 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5412 /* there are still entries for this ref, we can't drop it */
5413 if (ref->bytenr == bytenr)
5416 if (head->extent_op) {
5417 if (!head->must_insert_reserved)
5419 btrfs_free_delayed_extent_op(head->extent_op);
5420 head->extent_op = NULL;
5424 * waiting for the lock here would deadlock. If someone else has it
5425 * locked they are already in the process of dropping it anyway
5427 if (!mutex_trylock(&head->mutex))
5431 * at this point we have a head with no other entries. Go
5432 * ahead and process it.
5434 head->node.in_tree = 0;
5435 rb_erase(&head->node.rb_node, &delayed_refs->root);
5437 delayed_refs->num_entries--;
5440 * we don't take a ref on the node because we're removing it from the
5441 * tree, so we just steal the ref the tree was holding.
5443 delayed_refs->num_heads--;
5444 if (list_empty(&head->cluster))
5445 delayed_refs->num_heads_ready--;
5447 list_del_init(&head->cluster);
5448 spin_unlock(&delayed_refs->lock);
5450 BUG_ON(head->extent_op);
5451 if (head->must_insert_reserved)
5454 mutex_unlock(&head->mutex);
5455 btrfs_put_delayed_ref(&head->node);
5458 spin_unlock(&delayed_refs->lock);
5462 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5463 struct btrfs_root *root,
5464 struct extent_buffer *buf,
5465 u64 parent, int last_ref)
5467 struct btrfs_block_group_cache *cache = NULL;
5470 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5471 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5472 buf->start, buf->len,
5473 parent, root->root_key.objectid,
5474 btrfs_header_level(buf),
5475 BTRFS_DROP_DELAYED_REF, NULL, 0);
5476 BUG_ON(ret); /* -ENOMEM */
5482 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5484 if (btrfs_header_generation(buf) == trans->transid) {
5485 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5486 ret = check_ref_cleanup(trans, root, buf->start);
5491 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5492 pin_down_extent(root, cache, buf->start, buf->len, 1);
5496 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5498 btrfs_add_free_space(cache, buf->start, buf->len);
5499 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5503 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5506 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5507 btrfs_put_block_group(cache);
5510 /* Can return -ENOMEM */
5511 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5512 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5513 u64 owner, u64 offset, int for_cow)
5516 struct btrfs_fs_info *fs_info = root->fs_info;
5519 * tree log blocks never actually go into the extent allocation
5520 * tree, just update pinning info and exit early.
5522 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5523 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5524 /* unlocks the pinned mutex */
5525 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5527 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5528 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5530 parent, root_objectid, (int)owner,
5531 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5533 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5535 parent, root_objectid, owner,
5536 offset, BTRFS_DROP_DELAYED_REF,
5542 static u64 stripe_align(struct btrfs_root *root, u64 val)
5544 u64 mask = ((u64)root->stripesize - 1);
5545 u64 ret = (val + mask) & ~mask;
5550 * when we wait for progress in the block group caching, its because
5551 * our allocation attempt failed at least once. So, we must sleep
5552 * and let some progress happen before we try again.
5554 * This function will sleep at least once waiting for new free space to
5555 * show up, and then it will check the block group free space numbers
5556 * for our min num_bytes. Another option is to have it go ahead
5557 * and look in the rbtree for a free extent of a given size, but this
5561 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5564 struct btrfs_caching_control *caching_ctl;
5566 caching_ctl = get_caching_control(cache);
5570 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5571 (cache->free_space_ctl->free_space >= num_bytes));
5573 put_caching_control(caching_ctl);
5578 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5580 struct btrfs_caching_control *caching_ctl;
5582 caching_ctl = get_caching_control(cache);
5586 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5588 put_caching_control(caching_ctl);
5592 int __get_raid_index(u64 flags)
5594 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5595 return BTRFS_RAID_RAID10;
5596 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5597 return BTRFS_RAID_RAID1;
5598 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5599 return BTRFS_RAID_DUP;
5600 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5601 return BTRFS_RAID_RAID0;
5603 return BTRFS_RAID_SINGLE;
5606 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5608 return __get_raid_index(cache->flags);
5611 enum btrfs_loop_type {
5612 LOOP_CACHING_NOWAIT = 0,
5613 LOOP_CACHING_WAIT = 1,
5614 LOOP_ALLOC_CHUNK = 2,
5615 LOOP_NO_EMPTY_SIZE = 3,
5619 * walks the btree of allocated extents and find a hole of a given size.
5620 * The key ins is changed to record the hole:
5621 * ins->objectid == block start
5622 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5623 * ins->offset == number of blocks
5624 * Any available blocks before search_start are skipped.
5626 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5627 struct btrfs_root *orig_root,
5628 u64 num_bytes, u64 empty_size,
5629 u64 hint_byte, struct btrfs_key *ins,
5633 struct btrfs_root *root = orig_root->fs_info->extent_root;
5634 struct btrfs_free_cluster *last_ptr = NULL;
5635 struct btrfs_block_group_cache *block_group = NULL;
5636 struct btrfs_block_group_cache *used_block_group;
5637 u64 search_start = 0;
5638 int empty_cluster = 2 * 1024 * 1024;
5639 struct btrfs_space_info *space_info;
5641 int index = __get_raid_index(data);
5642 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5643 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5644 bool found_uncached_bg = false;
5645 bool failed_cluster_refill = false;
5646 bool failed_alloc = false;
5647 bool use_cluster = true;
5648 bool have_caching_bg = false;
5650 WARN_ON(num_bytes < root->sectorsize);
5651 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5655 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5657 space_info = __find_space_info(root->fs_info, data);
5659 printk(KERN_ERR "No space info for %llu\n", data);
5664 * If the space info is for both data and metadata it means we have a
5665 * small filesystem and we can't use the clustering stuff.
5667 if (btrfs_mixed_space_info(space_info))
5668 use_cluster = false;
5670 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5671 last_ptr = &root->fs_info->meta_alloc_cluster;
5672 if (!btrfs_test_opt(root, SSD))
5673 empty_cluster = 64 * 1024;
5676 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5677 btrfs_test_opt(root, SSD)) {
5678 last_ptr = &root->fs_info->data_alloc_cluster;
5682 spin_lock(&last_ptr->lock);
5683 if (last_ptr->block_group)
5684 hint_byte = last_ptr->window_start;
5685 spin_unlock(&last_ptr->lock);
5688 search_start = max(search_start, first_logical_byte(root, 0));
5689 search_start = max(search_start, hint_byte);
5694 if (search_start == hint_byte) {
5695 block_group = btrfs_lookup_block_group(root->fs_info,
5697 used_block_group = block_group;
5699 * we don't want to use the block group if it doesn't match our
5700 * allocation bits, or if its not cached.
5702 * However if we are re-searching with an ideal block group
5703 * picked out then we don't care that the block group is cached.
5705 if (block_group && block_group_bits(block_group, data) &&
5706 block_group->cached != BTRFS_CACHE_NO) {
5707 down_read(&space_info->groups_sem);
5708 if (list_empty(&block_group->list) ||
5711 * someone is removing this block group,
5712 * we can't jump into the have_block_group
5713 * target because our list pointers are not
5716 btrfs_put_block_group(block_group);
5717 up_read(&space_info->groups_sem);
5719 index = get_block_group_index(block_group);
5720 goto have_block_group;
5722 } else if (block_group) {
5723 btrfs_put_block_group(block_group);
5727 have_caching_bg = false;
5728 down_read(&space_info->groups_sem);
5729 list_for_each_entry(block_group, &space_info->block_groups[index],
5734 used_block_group = block_group;
5735 btrfs_get_block_group(block_group);
5736 search_start = block_group->key.objectid;
5739 * this can happen if we end up cycling through all the
5740 * raid types, but we want to make sure we only allocate
5741 * for the proper type.
5743 if (!block_group_bits(block_group, data)) {
5744 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5745 BTRFS_BLOCK_GROUP_RAID1 |
5746 BTRFS_BLOCK_GROUP_RAID10;
5749 * if they asked for extra copies and this block group
5750 * doesn't provide them, bail. This does allow us to
5751 * fill raid0 from raid1.
5753 if ((data & extra) && !(block_group->flags & extra))
5758 cached = block_group_cache_done(block_group);
5759 if (unlikely(!cached)) {
5760 found_uncached_bg = true;
5761 ret = cache_block_group(block_group, 0);
5766 if (unlikely(block_group->ro))
5770 * Ok we want to try and use the cluster allocator, so
5775 * the refill lock keeps out other
5776 * people trying to start a new cluster
5778 spin_lock(&last_ptr->refill_lock);
5779 used_block_group = last_ptr->block_group;
5780 if (used_block_group != block_group &&
5781 (!used_block_group ||
5782 used_block_group->ro ||
5783 !block_group_bits(used_block_group, data))) {
5784 used_block_group = block_group;
5785 goto refill_cluster;
5788 if (used_block_group != block_group)
5789 btrfs_get_block_group(used_block_group);
5791 offset = btrfs_alloc_from_cluster(used_block_group,
5792 last_ptr, num_bytes, used_block_group->key.objectid);
5794 /* we have a block, we're done */
5795 spin_unlock(&last_ptr->refill_lock);
5796 trace_btrfs_reserve_extent_cluster(root,
5797 block_group, search_start, num_bytes);
5801 WARN_ON(last_ptr->block_group != used_block_group);
5802 if (used_block_group != block_group) {
5803 btrfs_put_block_group(used_block_group);
5804 used_block_group = block_group;
5807 BUG_ON(used_block_group != block_group);
5808 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5809 * set up a new clusters, so lets just skip it
5810 * and let the allocator find whatever block
5811 * it can find. If we reach this point, we
5812 * will have tried the cluster allocator
5813 * plenty of times and not have found
5814 * anything, so we are likely way too
5815 * fragmented for the clustering stuff to find
5818 * However, if the cluster is taken from the
5819 * current block group, release the cluster
5820 * first, so that we stand a better chance of
5821 * succeeding in the unclustered
5823 if (loop >= LOOP_NO_EMPTY_SIZE &&
5824 last_ptr->block_group != block_group) {
5825 spin_unlock(&last_ptr->refill_lock);
5826 goto unclustered_alloc;
5830 * this cluster didn't work out, free it and
5833 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5835 if (loop >= LOOP_NO_EMPTY_SIZE) {
5836 spin_unlock(&last_ptr->refill_lock);
5837 goto unclustered_alloc;
5840 /* allocate a cluster in this block group */
5841 ret = btrfs_find_space_cluster(trans, root,
5842 block_group, last_ptr,
5843 search_start, num_bytes,
5844 empty_cluster + empty_size);
5847 * now pull our allocation out of this
5850 offset = btrfs_alloc_from_cluster(block_group,
5851 last_ptr, num_bytes,
5854 /* we found one, proceed */
5855 spin_unlock(&last_ptr->refill_lock);
5856 trace_btrfs_reserve_extent_cluster(root,
5857 block_group, search_start,
5861 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5862 && !failed_cluster_refill) {
5863 spin_unlock(&last_ptr->refill_lock);
5865 failed_cluster_refill = true;
5866 wait_block_group_cache_progress(block_group,
5867 num_bytes + empty_cluster + empty_size);
5868 goto have_block_group;
5872 * at this point we either didn't find a cluster
5873 * or we weren't able to allocate a block from our
5874 * cluster. Free the cluster we've been trying
5875 * to use, and go to the next block group
5877 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5878 spin_unlock(&last_ptr->refill_lock);
5883 spin_lock(&block_group->free_space_ctl->tree_lock);
5885 block_group->free_space_ctl->free_space <
5886 num_bytes + empty_cluster + empty_size) {
5887 spin_unlock(&block_group->free_space_ctl->tree_lock);
5890 spin_unlock(&block_group->free_space_ctl->tree_lock);
5892 offset = btrfs_find_space_for_alloc(block_group, search_start,
5893 num_bytes, empty_size);
5895 * If we didn't find a chunk, and we haven't failed on this
5896 * block group before, and this block group is in the middle of
5897 * caching and we are ok with waiting, then go ahead and wait
5898 * for progress to be made, and set failed_alloc to true.
5900 * If failed_alloc is true then we've already waited on this
5901 * block group once and should move on to the next block group.
5903 if (!offset && !failed_alloc && !cached &&
5904 loop > LOOP_CACHING_NOWAIT) {
5905 wait_block_group_cache_progress(block_group,
5906 num_bytes + empty_size);
5907 failed_alloc = true;
5908 goto have_block_group;
5909 } else if (!offset) {
5911 have_caching_bg = true;
5915 search_start = stripe_align(root, offset);
5917 /* move on to the next group */
5918 if (search_start + num_bytes >
5919 used_block_group->key.objectid + used_block_group->key.offset) {
5920 btrfs_add_free_space(used_block_group, offset, num_bytes);
5924 if (offset < search_start)
5925 btrfs_add_free_space(used_block_group, offset,
5926 search_start - offset);
5927 BUG_ON(offset > search_start);
5929 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5931 if (ret == -EAGAIN) {
5932 btrfs_add_free_space(used_block_group, offset, num_bytes);
5936 /* we are all good, lets return */
5937 ins->objectid = search_start;
5938 ins->offset = num_bytes;
5940 trace_btrfs_reserve_extent(orig_root, block_group,
5941 search_start, num_bytes);
5942 if (used_block_group != block_group)
5943 btrfs_put_block_group(used_block_group);
5944 btrfs_put_block_group(block_group);
5947 failed_cluster_refill = false;
5948 failed_alloc = false;
5949 BUG_ON(index != get_block_group_index(block_group));
5950 if (used_block_group != block_group)
5951 btrfs_put_block_group(used_block_group);
5952 btrfs_put_block_group(block_group);
5954 up_read(&space_info->groups_sem);
5956 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5959 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5963 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5964 * caching kthreads as we move along
5965 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5966 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5967 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5970 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5973 if (loop == LOOP_ALLOC_CHUNK) {
5974 ret = do_chunk_alloc(trans, root, data,
5977 * Do not bail out on ENOSPC since we
5978 * can do more things.
5980 if (ret < 0 && ret != -ENOSPC) {
5981 btrfs_abort_transaction(trans,
5987 if (loop == LOOP_NO_EMPTY_SIZE) {
5993 } else if (!ins->objectid) {
5995 } else if (ins->objectid) {
6003 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6004 int dump_block_groups)
6006 struct btrfs_block_group_cache *cache;
6009 spin_lock(&info->lock);
6010 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6011 (unsigned long long)info->flags,
6012 (unsigned long long)(info->total_bytes - info->bytes_used -
6013 info->bytes_pinned - info->bytes_reserved -
6014 info->bytes_readonly),
6015 (info->full) ? "" : "not ");
6016 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6017 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6018 (unsigned long long)info->total_bytes,
6019 (unsigned long long)info->bytes_used,
6020 (unsigned long long)info->bytes_pinned,
6021 (unsigned long long)info->bytes_reserved,
6022 (unsigned long long)info->bytes_may_use,
6023 (unsigned long long)info->bytes_readonly);
6024 spin_unlock(&info->lock);
6026 if (!dump_block_groups)
6029 down_read(&info->groups_sem);
6031 list_for_each_entry(cache, &info->block_groups[index], list) {
6032 spin_lock(&cache->lock);
6033 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6034 (unsigned long long)cache->key.objectid,
6035 (unsigned long long)cache->key.offset,
6036 (unsigned long long)btrfs_block_group_used(&cache->item),
6037 (unsigned long long)cache->pinned,
6038 (unsigned long long)cache->reserved,
6039 cache->ro ? "[readonly]" : "");
6040 btrfs_dump_free_space(cache, bytes);
6041 spin_unlock(&cache->lock);
6043 if (++index < BTRFS_NR_RAID_TYPES)
6045 up_read(&info->groups_sem);
6048 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6049 struct btrfs_root *root,
6050 u64 num_bytes, u64 min_alloc_size,
6051 u64 empty_size, u64 hint_byte,
6052 struct btrfs_key *ins, u64 data)
6054 bool final_tried = false;
6057 data = btrfs_get_alloc_profile(root, data);
6059 WARN_ON(num_bytes < root->sectorsize);
6060 ret = find_free_extent(trans, root, num_bytes, empty_size,
6061 hint_byte, ins, data);
6063 if (ret == -ENOSPC) {
6065 num_bytes = num_bytes >> 1;
6066 num_bytes = num_bytes & ~(root->sectorsize - 1);
6067 num_bytes = max(num_bytes, min_alloc_size);
6068 if (num_bytes == min_alloc_size)
6071 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6072 struct btrfs_space_info *sinfo;
6074 sinfo = __find_space_info(root->fs_info, data);
6075 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6076 "wanted %llu\n", (unsigned long long)data,
6077 (unsigned long long)num_bytes);
6079 dump_space_info(sinfo, num_bytes, 1);
6083 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6088 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6089 u64 start, u64 len, int pin)
6091 struct btrfs_block_group_cache *cache;
6094 cache = btrfs_lookup_block_group(root->fs_info, start);
6096 printk(KERN_ERR "Unable to find block group for %llu\n",
6097 (unsigned long long)start);
6101 if (btrfs_test_opt(root, DISCARD))
6102 ret = btrfs_discard_extent(root, start, len, NULL);
6105 pin_down_extent(root, cache, start, len, 1);
6107 btrfs_add_free_space(cache, start, len);
6108 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6110 btrfs_put_block_group(cache);
6112 trace_btrfs_reserved_extent_free(root, start, len);
6117 int btrfs_free_reserved_extent(struct btrfs_root *root,
6120 return __btrfs_free_reserved_extent(root, start, len, 0);
6123 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6126 return __btrfs_free_reserved_extent(root, start, len, 1);
6129 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6130 struct btrfs_root *root,
6131 u64 parent, u64 root_objectid,
6132 u64 flags, u64 owner, u64 offset,
6133 struct btrfs_key *ins, int ref_mod)
6136 struct btrfs_fs_info *fs_info = root->fs_info;
6137 struct btrfs_extent_item *extent_item;
6138 struct btrfs_extent_inline_ref *iref;
6139 struct btrfs_path *path;
6140 struct extent_buffer *leaf;
6145 type = BTRFS_SHARED_DATA_REF_KEY;
6147 type = BTRFS_EXTENT_DATA_REF_KEY;
6149 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6151 path = btrfs_alloc_path();
6155 path->leave_spinning = 1;
6156 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6159 btrfs_free_path(path);
6163 leaf = path->nodes[0];
6164 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6165 struct btrfs_extent_item);
6166 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6167 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6168 btrfs_set_extent_flags(leaf, extent_item,
6169 flags | BTRFS_EXTENT_FLAG_DATA);
6171 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6172 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6174 struct btrfs_shared_data_ref *ref;
6175 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6176 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6177 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6179 struct btrfs_extent_data_ref *ref;
6180 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6181 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6182 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6183 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6184 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6187 btrfs_mark_buffer_dirty(path->nodes[0]);
6188 btrfs_free_path(path);
6190 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6191 if (ret) { /* -ENOENT, logic error */
6192 printk(KERN_ERR "btrfs update block group failed for %llu "
6193 "%llu\n", (unsigned long long)ins->objectid,
6194 (unsigned long long)ins->offset);
6200 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6201 struct btrfs_root *root,
6202 u64 parent, u64 root_objectid,
6203 u64 flags, struct btrfs_disk_key *key,
6204 int level, struct btrfs_key *ins)
6207 struct btrfs_fs_info *fs_info = root->fs_info;
6208 struct btrfs_extent_item *extent_item;
6209 struct btrfs_tree_block_info *block_info;
6210 struct btrfs_extent_inline_ref *iref;
6211 struct btrfs_path *path;
6212 struct extent_buffer *leaf;
6213 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6215 path = btrfs_alloc_path();
6219 path->leave_spinning = 1;
6220 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6223 btrfs_free_path(path);
6227 leaf = path->nodes[0];
6228 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6229 struct btrfs_extent_item);
6230 btrfs_set_extent_refs(leaf, extent_item, 1);
6231 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6232 btrfs_set_extent_flags(leaf, extent_item,
6233 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6234 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6236 btrfs_set_tree_block_key(leaf, block_info, key);
6237 btrfs_set_tree_block_level(leaf, block_info, level);
6239 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6241 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6242 btrfs_set_extent_inline_ref_type(leaf, iref,
6243 BTRFS_SHARED_BLOCK_REF_KEY);
6244 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6246 btrfs_set_extent_inline_ref_type(leaf, iref,
6247 BTRFS_TREE_BLOCK_REF_KEY);
6248 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6251 btrfs_mark_buffer_dirty(leaf);
6252 btrfs_free_path(path);
6254 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6255 if (ret) { /* -ENOENT, logic error */
6256 printk(KERN_ERR "btrfs update block group failed for %llu "
6257 "%llu\n", (unsigned long long)ins->objectid,
6258 (unsigned long long)ins->offset);
6264 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6265 struct btrfs_root *root,
6266 u64 root_objectid, u64 owner,
6267 u64 offset, struct btrfs_key *ins)
6271 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6273 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6275 root_objectid, owner, offset,
6276 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6281 * this is used by the tree logging recovery code. It records that
6282 * an extent has been allocated and makes sure to clear the free
6283 * space cache bits as well
6285 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6286 struct btrfs_root *root,
6287 u64 root_objectid, u64 owner, u64 offset,
6288 struct btrfs_key *ins)
6291 struct btrfs_block_group_cache *block_group;
6292 struct btrfs_caching_control *caching_ctl;
6293 u64 start = ins->objectid;
6294 u64 num_bytes = ins->offset;
6296 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6297 cache_block_group(block_group, 0);
6298 caching_ctl = get_caching_control(block_group);
6301 BUG_ON(!block_group_cache_done(block_group));
6302 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6303 BUG_ON(ret); /* -ENOMEM */
6305 mutex_lock(&caching_ctl->mutex);
6307 if (start >= caching_ctl->progress) {
6308 ret = add_excluded_extent(root, start, num_bytes);
6309 BUG_ON(ret); /* -ENOMEM */
6310 } else if (start + num_bytes <= caching_ctl->progress) {
6311 ret = btrfs_remove_free_space(block_group,
6313 BUG_ON(ret); /* -ENOMEM */
6315 num_bytes = caching_ctl->progress - start;
6316 ret = btrfs_remove_free_space(block_group,
6318 BUG_ON(ret); /* -ENOMEM */
6320 start = caching_ctl->progress;
6321 num_bytes = ins->objectid + ins->offset -
6322 caching_ctl->progress;
6323 ret = add_excluded_extent(root, start, num_bytes);
6324 BUG_ON(ret); /* -ENOMEM */
6327 mutex_unlock(&caching_ctl->mutex);
6328 put_caching_control(caching_ctl);
6331 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6332 RESERVE_ALLOC_NO_ACCOUNT);
6333 BUG_ON(ret); /* logic error */
6334 btrfs_put_block_group(block_group);
6335 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6336 0, owner, offset, ins, 1);
6340 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6341 struct btrfs_root *root,
6342 u64 bytenr, u32 blocksize,
6345 struct extent_buffer *buf;
6347 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6349 return ERR_PTR(-ENOMEM);
6350 btrfs_set_header_generation(buf, trans->transid);
6351 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6352 btrfs_tree_lock(buf);
6353 clean_tree_block(trans, root, buf);
6354 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6356 btrfs_set_lock_blocking(buf);
6357 btrfs_set_buffer_uptodate(buf);
6359 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6361 * we allow two log transactions at a time, use different
6362 * EXENT bit to differentiate dirty pages.
6364 if (root->log_transid % 2 == 0)
6365 set_extent_dirty(&root->dirty_log_pages, buf->start,
6366 buf->start + buf->len - 1, GFP_NOFS);
6368 set_extent_new(&root->dirty_log_pages, buf->start,
6369 buf->start + buf->len - 1, GFP_NOFS);
6371 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6372 buf->start + buf->len - 1, GFP_NOFS);
6374 trans->blocks_used++;
6375 /* this returns a buffer locked for blocking */
6379 static struct btrfs_block_rsv *
6380 use_block_rsv(struct btrfs_trans_handle *trans,
6381 struct btrfs_root *root, u32 blocksize)
6383 struct btrfs_block_rsv *block_rsv;
6384 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6387 block_rsv = get_block_rsv(trans, root);
6389 if (block_rsv->size == 0) {
6390 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6391 BTRFS_RESERVE_NO_FLUSH);
6393 * If we couldn't reserve metadata bytes try and use some from
6394 * the global reserve.
6396 if (ret && block_rsv != global_rsv) {
6397 ret = block_rsv_use_bytes(global_rsv, blocksize);
6400 return ERR_PTR(ret);
6402 return ERR_PTR(ret);
6407 ret = block_rsv_use_bytes(block_rsv, blocksize);
6410 if (ret && !block_rsv->failfast) {
6411 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6412 static DEFINE_RATELIMIT_STATE(_rs,
6413 DEFAULT_RATELIMIT_INTERVAL * 10,
6414 /*DEFAULT_RATELIMIT_BURST*/ 1);
6415 if (__ratelimit(&_rs))
6417 "btrfs: block rsv returned %d\n", ret);
6419 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6420 BTRFS_RESERVE_NO_FLUSH);
6423 } else if (ret && block_rsv != global_rsv) {
6424 ret = block_rsv_use_bytes(global_rsv, blocksize);
6430 return ERR_PTR(-ENOSPC);
6433 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6434 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6436 block_rsv_add_bytes(block_rsv, blocksize, 0);
6437 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6441 * finds a free extent and does all the dirty work required for allocation
6442 * returns the key for the extent through ins, and a tree buffer for
6443 * the first block of the extent through buf.
6445 * returns the tree buffer or NULL.
6447 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6448 struct btrfs_root *root, u32 blocksize,
6449 u64 parent, u64 root_objectid,
6450 struct btrfs_disk_key *key, int level,
6451 u64 hint, u64 empty_size)
6453 struct btrfs_key ins;
6454 struct btrfs_block_rsv *block_rsv;
6455 struct extent_buffer *buf;
6460 block_rsv = use_block_rsv(trans, root, blocksize);
6461 if (IS_ERR(block_rsv))
6462 return ERR_CAST(block_rsv);
6464 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6465 empty_size, hint, &ins, 0);
6467 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6468 return ERR_PTR(ret);
6471 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6473 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6475 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6477 parent = ins.objectid;
6478 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6482 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6483 struct btrfs_delayed_extent_op *extent_op;
6484 extent_op = btrfs_alloc_delayed_extent_op();
6485 BUG_ON(!extent_op); /* -ENOMEM */
6487 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6489 memset(&extent_op->key, 0, sizeof(extent_op->key));
6490 extent_op->flags_to_set = flags;
6491 extent_op->update_key = 1;
6492 extent_op->update_flags = 1;
6493 extent_op->is_data = 0;
6495 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6497 ins.offset, parent, root_objectid,
6498 level, BTRFS_ADD_DELAYED_EXTENT,
6500 BUG_ON(ret); /* -ENOMEM */
6505 struct walk_control {
6506 u64 refs[BTRFS_MAX_LEVEL];
6507 u64 flags[BTRFS_MAX_LEVEL];
6508 struct btrfs_key update_progress;
6519 #define DROP_REFERENCE 1
6520 #define UPDATE_BACKREF 2
6522 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6523 struct btrfs_root *root,
6524 struct walk_control *wc,
6525 struct btrfs_path *path)
6533 struct btrfs_key key;
6534 struct extent_buffer *eb;
6539 if (path->slots[wc->level] < wc->reada_slot) {
6540 wc->reada_count = wc->reada_count * 2 / 3;
6541 wc->reada_count = max(wc->reada_count, 2);
6543 wc->reada_count = wc->reada_count * 3 / 2;
6544 wc->reada_count = min_t(int, wc->reada_count,
6545 BTRFS_NODEPTRS_PER_BLOCK(root));
6548 eb = path->nodes[wc->level];
6549 nritems = btrfs_header_nritems(eb);
6550 blocksize = btrfs_level_size(root, wc->level - 1);
6552 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6553 if (nread >= wc->reada_count)
6557 bytenr = btrfs_node_blockptr(eb, slot);
6558 generation = btrfs_node_ptr_generation(eb, slot);
6560 if (slot == path->slots[wc->level])
6563 if (wc->stage == UPDATE_BACKREF &&
6564 generation <= root->root_key.offset)
6567 /* We don't lock the tree block, it's OK to be racy here */
6568 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6570 /* We don't care about errors in readahead. */
6575 if (wc->stage == DROP_REFERENCE) {
6579 if (wc->level == 1 &&
6580 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6582 if (!wc->update_ref ||
6583 generation <= root->root_key.offset)
6585 btrfs_node_key_to_cpu(eb, &key, slot);
6586 ret = btrfs_comp_cpu_keys(&key,
6587 &wc->update_progress);
6591 if (wc->level == 1 &&
6592 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6596 ret = readahead_tree_block(root, bytenr, blocksize,
6602 wc->reada_slot = slot;
6606 * hepler to process tree block while walking down the tree.
6608 * when wc->stage == UPDATE_BACKREF, this function updates
6609 * back refs for pointers in the block.
6611 * NOTE: return value 1 means we should stop walking down.
6613 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6614 struct btrfs_root *root,
6615 struct btrfs_path *path,
6616 struct walk_control *wc, int lookup_info)
6618 int level = wc->level;
6619 struct extent_buffer *eb = path->nodes[level];
6620 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6623 if (wc->stage == UPDATE_BACKREF &&
6624 btrfs_header_owner(eb) != root->root_key.objectid)
6628 * when reference count of tree block is 1, it won't increase
6629 * again. once full backref flag is set, we never clear it.
6632 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6633 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6634 BUG_ON(!path->locks[level]);
6635 ret = btrfs_lookup_extent_info(trans, root,
6639 BUG_ON(ret == -ENOMEM);
6642 BUG_ON(wc->refs[level] == 0);
6645 if (wc->stage == DROP_REFERENCE) {
6646 if (wc->refs[level] > 1)
6649 if (path->locks[level] && !wc->keep_locks) {
6650 btrfs_tree_unlock_rw(eb, path->locks[level]);
6651 path->locks[level] = 0;
6656 /* wc->stage == UPDATE_BACKREF */
6657 if (!(wc->flags[level] & flag)) {
6658 BUG_ON(!path->locks[level]);
6659 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6660 BUG_ON(ret); /* -ENOMEM */
6661 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6662 BUG_ON(ret); /* -ENOMEM */
6663 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6665 BUG_ON(ret); /* -ENOMEM */
6666 wc->flags[level] |= flag;
6670 * the block is shared by multiple trees, so it's not good to
6671 * keep the tree lock
6673 if (path->locks[level] && level > 0) {
6674 btrfs_tree_unlock_rw(eb, path->locks[level]);
6675 path->locks[level] = 0;
6681 * hepler to process tree block pointer.
6683 * when wc->stage == DROP_REFERENCE, this function checks
6684 * reference count of the block pointed to. if the block
6685 * is shared and we need update back refs for the subtree
6686 * rooted at the block, this function changes wc->stage to
6687 * UPDATE_BACKREF. if the block is shared and there is no
6688 * need to update back, this function drops the reference
6691 * NOTE: return value 1 means we should stop walking down.
6693 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6694 struct btrfs_root *root,
6695 struct btrfs_path *path,
6696 struct walk_control *wc, int *lookup_info)
6702 struct btrfs_key key;
6703 struct extent_buffer *next;
6704 int level = wc->level;
6708 generation = btrfs_node_ptr_generation(path->nodes[level],
6709 path->slots[level]);
6711 * if the lower level block was created before the snapshot
6712 * was created, we know there is no need to update back refs
6715 if (wc->stage == UPDATE_BACKREF &&
6716 generation <= root->root_key.offset) {
6721 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6722 blocksize = btrfs_level_size(root, level - 1);
6724 next = btrfs_find_tree_block(root, bytenr, blocksize);
6726 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6731 btrfs_tree_lock(next);
6732 btrfs_set_lock_blocking(next);
6734 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6735 &wc->refs[level - 1],
6736 &wc->flags[level - 1]);
6738 btrfs_tree_unlock(next);
6742 BUG_ON(wc->refs[level - 1] == 0);
6745 if (wc->stage == DROP_REFERENCE) {
6746 if (wc->refs[level - 1] > 1) {
6748 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6751 if (!wc->update_ref ||
6752 generation <= root->root_key.offset)
6755 btrfs_node_key_to_cpu(path->nodes[level], &key,
6756 path->slots[level]);
6757 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6761 wc->stage = UPDATE_BACKREF;
6762 wc->shared_level = level - 1;
6766 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6770 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6771 btrfs_tree_unlock(next);
6772 free_extent_buffer(next);
6778 if (reada && level == 1)
6779 reada_walk_down(trans, root, wc, path);
6780 next = read_tree_block(root, bytenr, blocksize, generation);
6783 btrfs_tree_lock(next);
6784 btrfs_set_lock_blocking(next);
6788 BUG_ON(level != btrfs_header_level(next));
6789 path->nodes[level] = next;
6790 path->slots[level] = 0;
6791 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6797 wc->refs[level - 1] = 0;
6798 wc->flags[level - 1] = 0;
6799 if (wc->stage == DROP_REFERENCE) {
6800 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6801 parent = path->nodes[level]->start;
6803 BUG_ON(root->root_key.objectid !=
6804 btrfs_header_owner(path->nodes[level]));
6808 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6809 root->root_key.objectid, level - 1, 0, 0);
6810 BUG_ON(ret); /* -ENOMEM */
6812 btrfs_tree_unlock(next);
6813 free_extent_buffer(next);
6819 * hepler to process tree block while walking up the tree.
6821 * when wc->stage == DROP_REFERENCE, this function drops
6822 * reference count on the block.
6824 * when wc->stage == UPDATE_BACKREF, this function changes
6825 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6826 * to UPDATE_BACKREF previously while processing the block.
6828 * NOTE: return value 1 means we should stop walking up.
6830 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6831 struct btrfs_root *root,
6832 struct btrfs_path *path,
6833 struct walk_control *wc)
6836 int level = wc->level;
6837 struct extent_buffer *eb = path->nodes[level];
6840 if (wc->stage == UPDATE_BACKREF) {
6841 BUG_ON(wc->shared_level < level);
6842 if (level < wc->shared_level)
6845 ret = find_next_key(path, level + 1, &wc->update_progress);
6849 wc->stage = DROP_REFERENCE;
6850 wc->shared_level = -1;
6851 path->slots[level] = 0;
6854 * check reference count again if the block isn't locked.
6855 * we should start walking down the tree again if reference
6858 if (!path->locks[level]) {
6860 btrfs_tree_lock(eb);
6861 btrfs_set_lock_blocking(eb);
6862 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6864 ret = btrfs_lookup_extent_info(trans, root,
6869 btrfs_tree_unlock_rw(eb, path->locks[level]);
6870 path->locks[level] = 0;
6873 BUG_ON(wc->refs[level] == 0);
6874 if (wc->refs[level] == 1) {
6875 btrfs_tree_unlock_rw(eb, path->locks[level]);
6876 path->locks[level] = 0;
6882 /* wc->stage == DROP_REFERENCE */
6883 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6885 if (wc->refs[level] == 1) {
6887 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6888 ret = btrfs_dec_ref(trans, root, eb, 1,
6891 ret = btrfs_dec_ref(trans, root, eb, 0,
6893 BUG_ON(ret); /* -ENOMEM */
6895 /* make block locked assertion in clean_tree_block happy */
6896 if (!path->locks[level] &&
6897 btrfs_header_generation(eb) == trans->transid) {
6898 btrfs_tree_lock(eb);
6899 btrfs_set_lock_blocking(eb);
6900 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6902 clean_tree_block(trans, root, eb);
6905 if (eb == root->node) {
6906 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6909 BUG_ON(root->root_key.objectid !=
6910 btrfs_header_owner(eb));
6912 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6913 parent = path->nodes[level + 1]->start;
6915 BUG_ON(root->root_key.objectid !=
6916 btrfs_header_owner(path->nodes[level + 1]));
6919 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6921 wc->refs[level] = 0;
6922 wc->flags[level] = 0;
6926 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6927 struct btrfs_root *root,
6928 struct btrfs_path *path,
6929 struct walk_control *wc)
6931 int level = wc->level;
6932 int lookup_info = 1;
6935 while (level >= 0) {
6936 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6943 if (path->slots[level] >=
6944 btrfs_header_nritems(path->nodes[level]))
6947 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6949 path->slots[level]++;
6958 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6959 struct btrfs_root *root,
6960 struct btrfs_path *path,
6961 struct walk_control *wc, int max_level)
6963 int level = wc->level;
6966 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6967 while (level < max_level && path->nodes[level]) {
6969 if (path->slots[level] + 1 <
6970 btrfs_header_nritems(path->nodes[level])) {
6971 path->slots[level]++;
6974 ret = walk_up_proc(trans, root, path, wc);
6978 if (path->locks[level]) {
6979 btrfs_tree_unlock_rw(path->nodes[level],
6980 path->locks[level]);
6981 path->locks[level] = 0;
6983 free_extent_buffer(path->nodes[level]);
6984 path->nodes[level] = NULL;
6992 * drop a subvolume tree.
6994 * this function traverses the tree freeing any blocks that only
6995 * referenced by the tree.
6997 * when a shared tree block is found. this function decreases its
6998 * reference count by one. if update_ref is true, this function
6999 * also make sure backrefs for the shared block and all lower level
7000 * blocks are properly updated.
7002 int btrfs_drop_snapshot(struct btrfs_root *root,
7003 struct btrfs_block_rsv *block_rsv, int update_ref,
7006 struct btrfs_path *path;
7007 struct btrfs_trans_handle *trans;
7008 struct btrfs_root *tree_root = root->fs_info->tree_root;
7009 struct btrfs_root_item *root_item = &root->root_item;
7010 struct walk_control *wc;
7011 struct btrfs_key key;
7016 path = btrfs_alloc_path();
7022 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7024 btrfs_free_path(path);
7029 trans = btrfs_start_transaction(tree_root, 0);
7030 if (IS_ERR(trans)) {
7031 err = PTR_ERR(trans);
7036 trans->block_rsv = block_rsv;
7038 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7039 level = btrfs_header_level(root->node);
7040 path->nodes[level] = btrfs_lock_root_node(root);
7041 btrfs_set_lock_blocking(path->nodes[level]);
7042 path->slots[level] = 0;
7043 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7044 memset(&wc->update_progress, 0,
7045 sizeof(wc->update_progress));
7047 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7048 memcpy(&wc->update_progress, &key,
7049 sizeof(wc->update_progress));
7051 level = root_item->drop_level;
7053 path->lowest_level = level;
7054 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7055 path->lowest_level = 0;
7063 * unlock our path, this is safe because only this
7064 * function is allowed to delete this snapshot
7066 btrfs_unlock_up_safe(path, 0);
7068 level = btrfs_header_level(root->node);
7070 btrfs_tree_lock(path->nodes[level]);
7071 btrfs_set_lock_blocking(path->nodes[level]);
7073 ret = btrfs_lookup_extent_info(trans, root,
7074 path->nodes[level]->start,
7075 path->nodes[level]->len,
7082 BUG_ON(wc->refs[level] == 0);
7084 if (level == root_item->drop_level)
7087 btrfs_tree_unlock(path->nodes[level]);
7088 WARN_ON(wc->refs[level] != 1);
7094 wc->shared_level = -1;
7095 wc->stage = DROP_REFERENCE;
7096 wc->update_ref = update_ref;
7098 wc->for_reloc = for_reloc;
7099 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7102 ret = walk_down_tree(trans, root, path, wc);
7108 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7115 BUG_ON(wc->stage != DROP_REFERENCE);
7119 if (wc->stage == DROP_REFERENCE) {
7121 btrfs_node_key(path->nodes[level],
7122 &root_item->drop_progress,
7123 path->slots[level]);
7124 root_item->drop_level = level;
7127 BUG_ON(wc->level == 0);
7128 if (btrfs_should_end_transaction(trans, tree_root)) {
7129 ret = btrfs_update_root(trans, tree_root,
7133 btrfs_abort_transaction(trans, tree_root, ret);
7138 btrfs_end_transaction_throttle(trans, tree_root);
7139 trans = btrfs_start_transaction(tree_root, 0);
7140 if (IS_ERR(trans)) {
7141 err = PTR_ERR(trans);
7145 trans->block_rsv = block_rsv;
7148 btrfs_release_path(path);
7152 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7154 btrfs_abort_transaction(trans, tree_root, ret);
7158 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7159 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7162 btrfs_abort_transaction(trans, tree_root, ret);
7165 } else if (ret > 0) {
7166 /* if we fail to delete the orphan item this time
7167 * around, it'll get picked up the next time.
7169 * The most common failure here is just -ENOENT.
7171 btrfs_del_orphan_item(trans, tree_root,
7172 root->root_key.objectid);
7176 if (root->in_radix) {
7177 btrfs_free_fs_root(tree_root->fs_info, root);
7179 free_extent_buffer(root->node);
7180 free_extent_buffer(root->commit_root);
7184 btrfs_end_transaction_throttle(trans, tree_root);
7187 btrfs_free_path(path);
7190 btrfs_std_error(root->fs_info, err);
7195 * drop subtree rooted at tree block 'node'.
7197 * NOTE: this function will unlock and release tree block 'node'
7198 * only used by relocation code
7200 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7201 struct btrfs_root *root,
7202 struct extent_buffer *node,
7203 struct extent_buffer *parent)
7205 struct btrfs_path *path;
7206 struct walk_control *wc;
7212 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7214 path = btrfs_alloc_path();
7218 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7220 btrfs_free_path(path);
7224 btrfs_assert_tree_locked(parent);
7225 parent_level = btrfs_header_level(parent);
7226 extent_buffer_get(parent);
7227 path->nodes[parent_level] = parent;
7228 path->slots[parent_level] = btrfs_header_nritems(parent);
7230 btrfs_assert_tree_locked(node);
7231 level = btrfs_header_level(node);
7232 path->nodes[level] = node;
7233 path->slots[level] = 0;
7234 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7236 wc->refs[parent_level] = 1;
7237 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7239 wc->shared_level = -1;
7240 wc->stage = DROP_REFERENCE;
7244 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7247 wret = walk_down_tree(trans, root, path, wc);
7253 wret = walk_up_tree(trans, root, path, wc, parent_level);
7261 btrfs_free_path(path);
7265 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7271 * if restripe for this chunk_type is on pick target profile and
7272 * return, otherwise do the usual balance
7274 stripped = get_restripe_target(root->fs_info, flags);
7276 return extended_to_chunk(stripped);
7279 * we add in the count of missing devices because we want
7280 * to make sure that any RAID levels on a degraded FS
7281 * continue to be honored.
7283 num_devices = root->fs_info->fs_devices->rw_devices +
7284 root->fs_info->fs_devices->missing_devices;
7286 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7287 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7289 if (num_devices == 1) {
7290 stripped |= BTRFS_BLOCK_GROUP_DUP;
7291 stripped = flags & ~stripped;
7293 /* turn raid0 into single device chunks */
7294 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7297 /* turn mirroring into duplication */
7298 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7299 BTRFS_BLOCK_GROUP_RAID10))
7300 return stripped | BTRFS_BLOCK_GROUP_DUP;
7302 /* they already had raid on here, just return */
7303 if (flags & stripped)
7306 stripped |= BTRFS_BLOCK_GROUP_DUP;
7307 stripped = flags & ~stripped;
7309 /* switch duplicated blocks with raid1 */
7310 if (flags & BTRFS_BLOCK_GROUP_DUP)
7311 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7313 /* this is drive concat, leave it alone */
7319 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7321 struct btrfs_space_info *sinfo = cache->space_info;
7323 u64 min_allocable_bytes;
7328 * We need some metadata space and system metadata space for
7329 * allocating chunks in some corner cases until we force to set
7330 * it to be readonly.
7333 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7335 min_allocable_bytes = 1 * 1024 * 1024;
7337 min_allocable_bytes = 0;
7339 spin_lock(&sinfo->lock);
7340 spin_lock(&cache->lock);
7347 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7348 cache->bytes_super - btrfs_block_group_used(&cache->item);
7350 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7351 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7352 min_allocable_bytes <= sinfo->total_bytes) {
7353 sinfo->bytes_readonly += num_bytes;
7358 spin_unlock(&cache->lock);
7359 spin_unlock(&sinfo->lock);
7363 int btrfs_set_block_group_ro(struct btrfs_root *root,
7364 struct btrfs_block_group_cache *cache)
7367 struct btrfs_trans_handle *trans;
7373 trans = btrfs_join_transaction(root);
7375 return PTR_ERR(trans);
7377 alloc_flags = update_block_group_flags(root, cache->flags);
7378 if (alloc_flags != cache->flags) {
7379 ret = do_chunk_alloc(trans, root, alloc_flags,
7385 ret = set_block_group_ro(cache, 0);
7388 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7389 ret = do_chunk_alloc(trans, root, alloc_flags,
7393 ret = set_block_group_ro(cache, 0);
7395 btrfs_end_transaction(trans, root);
7399 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7400 struct btrfs_root *root, u64 type)
7402 u64 alloc_flags = get_alloc_profile(root, type);
7403 return do_chunk_alloc(trans, root, alloc_flags,
7408 * helper to account the unused space of all the readonly block group in the
7409 * list. takes mirrors into account.
7411 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7413 struct btrfs_block_group_cache *block_group;
7417 list_for_each_entry(block_group, groups_list, list) {
7418 spin_lock(&block_group->lock);
7420 if (!block_group->ro) {
7421 spin_unlock(&block_group->lock);
7425 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7426 BTRFS_BLOCK_GROUP_RAID10 |
7427 BTRFS_BLOCK_GROUP_DUP))
7432 free_bytes += (block_group->key.offset -
7433 btrfs_block_group_used(&block_group->item)) *
7436 spin_unlock(&block_group->lock);
7443 * helper to account the unused space of all the readonly block group in the
7444 * space_info. takes mirrors into account.
7446 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7451 spin_lock(&sinfo->lock);
7453 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7454 if (!list_empty(&sinfo->block_groups[i]))
7455 free_bytes += __btrfs_get_ro_block_group_free_space(
7456 &sinfo->block_groups[i]);
7458 spin_unlock(&sinfo->lock);
7463 void btrfs_set_block_group_rw(struct btrfs_root *root,
7464 struct btrfs_block_group_cache *cache)
7466 struct btrfs_space_info *sinfo = cache->space_info;
7471 spin_lock(&sinfo->lock);
7472 spin_lock(&cache->lock);
7473 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7474 cache->bytes_super - btrfs_block_group_used(&cache->item);
7475 sinfo->bytes_readonly -= num_bytes;
7477 spin_unlock(&cache->lock);
7478 spin_unlock(&sinfo->lock);
7482 * checks to see if its even possible to relocate this block group.
7484 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7485 * ok to go ahead and try.
7487 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7489 struct btrfs_block_group_cache *block_group;
7490 struct btrfs_space_info *space_info;
7491 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7492 struct btrfs_device *device;
7501 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7503 /* odd, couldn't find the block group, leave it alone */
7507 min_free = btrfs_block_group_used(&block_group->item);
7509 /* no bytes used, we're good */
7513 space_info = block_group->space_info;
7514 spin_lock(&space_info->lock);
7516 full = space_info->full;
7519 * if this is the last block group we have in this space, we can't
7520 * relocate it unless we're able to allocate a new chunk below.
7522 * Otherwise, we need to make sure we have room in the space to handle
7523 * all of the extents from this block group. If we can, we're good
7525 if ((space_info->total_bytes != block_group->key.offset) &&
7526 (space_info->bytes_used + space_info->bytes_reserved +
7527 space_info->bytes_pinned + space_info->bytes_readonly +
7528 min_free < space_info->total_bytes)) {
7529 spin_unlock(&space_info->lock);
7532 spin_unlock(&space_info->lock);
7535 * ok we don't have enough space, but maybe we have free space on our
7536 * devices to allocate new chunks for relocation, so loop through our
7537 * alloc devices and guess if we have enough space. if this block
7538 * group is going to be restriped, run checks against the target
7539 * profile instead of the current one.
7551 target = get_restripe_target(root->fs_info, block_group->flags);
7553 index = __get_raid_index(extended_to_chunk(target));
7556 * this is just a balance, so if we were marked as full
7557 * we know there is no space for a new chunk
7562 index = get_block_group_index(block_group);
7565 if (index == BTRFS_RAID_RAID10) {
7569 } else if (index == BTRFS_RAID_RAID1) {
7571 } else if (index == BTRFS_RAID_DUP) {
7574 } else if (index == BTRFS_RAID_RAID0) {
7575 dev_min = fs_devices->rw_devices;
7576 do_div(min_free, dev_min);
7579 mutex_lock(&root->fs_info->chunk_mutex);
7580 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7584 * check to make sure we can actually find a chunk with enough
7585 * space to fit our block group in.
7587 if (device->total_bytes > device->bytes_used + min_free &&
7588 !device->is_tgtdev_for_dev_replace) {
7589 ret = find_free_dev_extent(device, min_free,
7594 if (dev_nr >= dev_min)
7600 mutex_unlock(&root->fs_info->chunk_mutex);
7602 btrfs_put_block_group(block_group);
7606 static int find_first_block_group(struct btrfs_root *root,
7607 struct btrfs_path *path, struct btrfs_key *key)
7610 struct btrfs_key found_key;
7611 struct extent_buffer *leaf;
7614 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7619 slot = path->slots[0];
7620 leaf = path->nodes[0];
7621 if (slot >= btrfs_header_nritems(leaf)) {
7622 ret = btrfs_next_leaf(root, path);
7629 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7631 if (found_key.objectid >= key->objectid &&
7632 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7642 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7644 struct btrfs_block_group_cache *block_group;
7648 struct inode *inode;
7650 block_group = btrfs_lookup_first_block_group(info, last);
7651 while (block_group) {
7652 spin_lock(&block_group->lock);
7653 if (block_group->iref)
7655 spin_unlock(&block_group->lock);
7656 block_group = next_block_group(info->tree_root,
7666 inode = block_group->inode;
7667 block_group->iref = 0;
7668 block_group->inode = NULL;
7669 spin_unlock(&block_group->lock);
7671 last = block_group->key.objectid + block_group->key.offset;
7672 btrfs_put_block_group(block_group);
7676 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7678 struct btrfs_block_group_cache *block_group;
7679 struct btrfs_space_info *space_info;
7680 struct btrfs_caching_control *caching_ctl;
7683 down_write(&info->extent_commit_sem);
7684 while (!list_empty(&info->caching_block_groups)) {
7685 caching_ctl = list_entry(info->caching_block_groups.next,
7686 struct btrfs_caching_control, list);
7687 list_del(&caching_ctl->list);
7688 put_caching_control(caching_ctl);
7690 up_write(&info->extent_commit_sem);
7692 spin_lock(&info->block_group_cache_lock);
7693 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7694 block_group = rb_entry(n, struct btrfs_block_group_cache,
7696 rb_erase(&block_group->cache_node,
7697 &info->block_group_cache_tree);
7698 spin_unlock(&info->block_group_cache_lock);
7700 down_write(&block_group->space_info->groups_sem);
7701 list_del(&block_group->list);
7702 up_write(&block_group->space_info->groups_sem);
7704 if (block_group->cached == BTRFS_CACHE_STARTED)
7705 wait_block_group_cache_done(block_group);
7708 * We haven't cached this block group, which means we could
7709 * possibly have excluded extents on this block group.
7711 if (block_group->cached == BTRFS_CACHE_NO)
7712 free_excluded_extents(info->extent_root, block_group);
7714 btrfs_remove_free_space_cache(block_group);
7715 btrfs_put_block_group(block_group);
7717 spin_lock(&info->block_group_cache_lock);
7719 spin_unlock(&info->block_group_cache_lock);
7721 /* now that all the block groups are freed, go through and
7722 * free all the space_info structs. This is only called during
7723 * the final stages of unmount, and so we know nobody is
7724 * using them. We call synchronize_rcu() once before we start,
7725 * just to be on the safe side.
7729 release_global_block_rsv(info);
7731 while(!list_empty(&info->space_info)) {
7732 space_info = list_entry(info->space_info.next,
7733 struct btrfs_space_info,
7735 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
7736 if (space_info->bytes_pinned > 0 ||
7737 space_info->bytes_reserved > 0 ||
7738 space_info->bytes_may_use > 0) {
7740 dump_space_info(space_info, 0, 0);
7743 list_del(&space_info->list);
7749 static void __link_block_group(struct btrfs_space_info *space_info,
7750 struct btrfs_block_group_cache *cache)
7752 int index = get_block_group_index(cache);
7754 down_write(&space_info->groups_sem);
7755 list_add_tail(&cache->list, &space_info->block_groups[index]);
7756 up_write(&space_info->groups_sem);
7759 int btrfs_read_block_groups(struct btrfs_root *root)
7761 struct btrfs_path *path;
7763 struct btrfs_block_group_cache *cache;
7764 struct btrfs_fs_info *info = root->fs_info;
7765 struct btrfs_space_info *space_info;
7766 struct btrfs_key key;
7767 struct btrfs_key found_key;
7768 struct extent_buffer *leaf;
7772 root = info->extent_root;
7775 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7776 path = btrfs_alloc_path();
7781 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7782 if (btrfs_test_opt(root, SPACE_CACHE) &&
7783 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7785 if (btrfs_test_opt(root, CLEAR_CACHE))
7789 ret = find_first_block_group(root, path, &key);
7794 leaf = path->nodes[0];
7795 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7796 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7801 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7803 if (!cache->free_space_ctl) {
7809 atomic_set(&cache->count, 1);
7810 spin_lock_init(&cache->lock);
7811 cache->fs_info = info;
7812 INIT_LIST_HEAD(&cache->list);
7813 INIT_LIST_HEAD(&cache->cluster_list);
7817 * When we mount with old space cache, we need to
7818 * set BTRFS_DC_CLEAR and set dirty flag.
7820 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7821 * truncate the old free space cache inode and
7823 * b) Setting 'dirty flag' makes sure that we flush
7824 * the new space cache info onto disk.
7826 cache->disk_cache_state = BTRFS_DC_CLEAR;
7827 if (btrfs_test_opt(root, SPACE_CACHE))
7831 read_extent_buffer(leaf, &cache->item,
7832 btrfs_item_ptr_offset(leaf, path->slots[0]),
7833 sizeof(cache->item));
7834 memcpy(&cache->key, &found_key, sizeof(found_key));
7836 key.objectid = found_key.objectid + found_key.offset;
7837 btrfs_release_path(path);
7838 cache->flags = btrfs_block_group_flags(&cache->item);
7839 cache->sectorsize = root->sectorsize;
7841 btrfs_init_free_space_ctl(cache);
7844 * We need to exclude the super stripes now so that the space
7845 * info has super bytes accounted for, otherwise we'll think
7846 * we have more space than we actually do.
7848 exclude_super_stripes(root, cache);
7851 * check for two cases, either we are full, and therefore
7852 * don't need to bother with the caching work since we won't
7853 * find any space, or we are empty, and we can just add all
7854 * the space in and be done with it. This saves us _alot_ of
7855 * time, particularly in the full case.
7857 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7858 cache->last_byte_to_unpin = (u64)-1;
7859 cache->cached = BTRFS_CACHE_FINISHED;
7860 free_excluded_extents(root, cache);
7861 } else if (btrfs_block_group_used(&cache->item) == 0) {
7862 cache->last_byte_to_unpin = (u64)-1;
7863 cache->cached = BTRFS_CACHE_FINISHED;
7864 add_new_free_space(cache, root->fs_info,
7866 found_key.objectid +
7868 free_excluded_extents(root, cache);
7871 ret = update_space_info(info, cache->flags, found_key.offset,
7872 btrfs_block_group_used(&cache->item),
7874 BUG_ON(ret); /* -ENOMEM */
7875 cache->space_info = space_info;
7876 spin_lock(&cache->space_info->lock);
7877 cache->space_info->bytes_readonly += cache->bytes_super;
7878 spin_unlock(&cache->space_info->lock);
7880 __link_block_group(space_info, cache);
7882 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7883 BUG_ON(ret); /* Logic error */
7885 set_avail_alloc_bits(root->fs_info, cache->flags);
7886 if (btrfs_chunk_readonly(root, cache->key.objectid))
7887 set_block_group_ro(cache, 1);
7890 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7891 if (!(get_alloc_profile(root, space_info->flags) &
7892 (BTRFS_BLOCK_GROUP_RAID10 |
7893 BTRFS_BLOCK_GROUP_RAID1 |
7894 BTRFS_BLOCK_GROUP_DUP)))
7897 * avoid allocating from un-mirrored block group if there are
7898 * mirrored block groups.
7900 list_for_each_entry(cache, &space_info->block_groups[3], list)
7901 set_block_group_ro(cache, 1);
7902 list_for_each_entry(cache, &space_info->block_groups[4], list)
7903 set_block_group_ro(cache, 1);
7906 init_global_block_rsv(info);
7909 btrfs_free_path(path);
7913 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7914 struct btrfs_root *root)
7916 struct btrfs_block_group_cache *block_group, *tmp;
7917 struct btrfs_root *extent_root = root->fs_info->extent_root;
7918 struct btrfs_block_group_item item;
7919 struct btrfs_key key;
7922 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7924 list_del_init(&block_group->new_bg_list);
7929 spin_lock(&block_group->lock);
7930 memcpy(&item, &block_group->item, sizeof(item));
7931 memcpy(&key, &block_group->key, sizeof(key));
7932 spin_unlock(&block_group->lock);
7934 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7937 btrfs_abort_transaction(trans, extent_root, ret);
7941 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7942 struct btrfs_root *root, u64 bytes_used,
7943 u64 type, u64 chunk_objectid, u64 chunk_offset,
7947 struct btrfs_root *extent_root;
7948 struct btrfs_block_group_cache *cache;
7950 extent_root = root->fs_info->extent_root;
7952 root->fs_info->last_trans_log_full_commit = trans->transid;
7954 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7957 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7959 if (!cache->free_space_ctl) {
7964 cache->key.objectid = chunk_offset;
7965 cache->key.offset = size;
7966 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7967 cache->sectorsize = root->sectorsize;
7968 cache->fs_info = root->fs_info;
7970 atomic_set(&cache->count, 1);
7971 spin_lock_init(&cache->lock);
7972 INIT_LIST_HEAD(&cache->list);
7973 INIT_LIST_HEAD(&cache->cluster_list);
7974 INIT_LIST_HEAD(&cache->new_bg_list);
7976 btrfs_init_free_space_ctl(cache);
7978 btrfs_set_block_group_used(&cache->item, bytes_used);
7979 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7980 cache->flags = type;
7981 btrfs_set_block_group_flags(&cache->item, type);
7983 cache->last_byte_to_unpin = (u64)-1;
7984 cache->cached = BTRFS_CACHE_FINISHED;
7985 exclude_super_stripes(root, cache);
7987 add_new_free_space(cache, root->fs_info, chunk_offset,
7988 chunk_offset + size);
7990 free_excluded_extents(root, cache);
7992 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7993 &cache->space_info);
7994 BUG_ON(ret); /* -ENOMEM */
7995 update_global_block_rsv(root->fs_info);
7997 spin_lock(&cache->space_info->lock);
7998 cache->space_info->bytes_readonly += cache->bytes_super;
7999 spin_unlock(&cache->space_info->lock);
8001 __link_block_group(cache->space_info, cache);
8003 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8004 BUG_ON(ret); /* Logic error */
8006 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8008 set_avail_alloc_bits(extent_root->fs_info, type);
8013 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8015 u64 extra_flags = chunk_to_extended(flags) &
8016 BTRFS_EXTENDED_PROFILE_MASK;
8018 write_seqlock(&fs_info->profiles_lock);
8019 if (flags & BTRFS_BLOCK_GROUP_DATA)
8020 fs_info->avail_data_alloc_bits &= ~extra_flags;
8021 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8022 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8023 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8024 fs_info->avail_system_alloc_bits &= ~extra_flags;
8025 write_sequnlock(&fs_info->profiles_lock);
8028 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8029 struct btrfs_root *root, u64 group_start)
8031 struct btrfs_path *path;
8032 struct btrfs_block_group_cache *block_group;
8033 struct btrfs_free_cluster *cluster;
8034 struct btrfs_root *tree_root = root->fs_info->tree_root;
8035 struct btrfs_key key;
8036 struct inode *inode;
8041 root = root->fs_info->extent_root;
8043 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8044 BUG_ON(!block_group);
8045 BUG_ON(!block_group->ro);
8048 * Free the reserved super bytes from this block group before
8051 free_excluded_extents(root, block_group);
8053 memcpy(&key, &block_group->key, sizeof(key));
8054 index = get_block_group_index(block_group);
8055 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8056 BTRFS_BLOCK_GROUP_RAID1 |
8057 BTRFS_BLOCK_GROUP_RAID10))
8062 /* make sure this block group isn't part of an allocation cluster */
8063 cluster = &root->fs_info->data_alloc_cluster;
8064 spin_lock(&cluster->refill_lock);
8065 btrfs_return_cluster_to_free_space(block_group, cluster);
8066 spin_unlock(&cluster->refill_lock);
8069 * make sure this block group isn't part of a metadata
8070 * allocation cluster
8072 cluster = &root->fs_info->meta_alloc_cluster;
8073 spin_lock(&cluster->refill_lock);
8074 btrfs_return_cluster_to_free_space(block_group, cluster);
8075 spin_unlock(&cluster->refill_lock);
8077 path = btrfs_alloc_path();
8083 inode = lookup_free_space_inode(tree_root, block_group, path);
8084 if (!IS_ERR(inode)) {
8085 ret = btrfs_orphan_add(trans, inode);
8087 btrfs_add_delayed_iput(inode);
8091 /* One for the block groups ref */
8092 spin_lock(&block_group->lock);
8093 if (block_group->iref) {
8094 block_group->iref = 0;
8095 block_group->inode = NULL;
8096 spin_unlock(&block_group->lock);
8099 spin_unlock(&block_group->lock);
8101 /* One for our lookup ref */
8102 btrfs_add_delayed_iput(inode);
8105 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8106 key.offset = block_group->key.objectid;
8109 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8113 btrfs_release_path(path);
8115 ret = btrfs_del_item(trans, tree_root, path);
8118 btrfs_release_path(path);
8121 spin_lock(&root->fs_info->block_group_cache_lock);
8122 rb_erase(&block_group->cache_node,
8123 &root->fs_info->block_group_cache_tree);
8125 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8126 root->fs_info->first_logical_byte = (u64)-1;
8127 spin_unlock(&root->fs_info->block_group_cache_lock);
8129 down_write(&block_group->space_info->groups_sem);
8131 * we must use list_del_init so people can check to see if they
8132 * are still on the list after taking the semaphore
8134 list_del_init(&block_group->list);
8135 if (list_empty(&block_group->space_info->block_groups[index]))
8136 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8137 up_write(&block_group->space_info->groups_sem);
8139 if (block_group->cached == BTRFS_CACHE_STARTED)
8140 wait_block_group_cache_done(block_group);
8142 btrfs_remove_free_space_cache(block_group);
8144 spin_lock(&block_group->space_info->lock);
8145 block_group->space_info->total_bytes -= block_group->key.offset;
8146 block_group->space_info->bytes_readonly -= block_group->key.offset;
8147 block_group->space_info->disk_total -= block_group->key.offset * factor;
8148 spin_unlock(&block_group->space_info->lock);
8150 memcpy(&key, &block_group->key, sizeof(key));
8152 btrfs_clear_space_info_full(root->fs_info);
8154 btrfs_put_block_group(block_group);
8155 btrfs_put_block_group(block_group);
8157 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8163 ret = btrfs_del_item(trans, root, path);
8165 btrfs_free_path(path);
8169 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8171 struct btrfs_space_info *space_info;
8172 struct btrfs_super_block *disk_super;
8178 disk_super = fs_info->super_copy;
8179 if (!btrfs_super_root(disk_super))
8182 features = btrfs_super_incompat_flags(disk_super);
8183 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8186 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8187 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8192 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8193 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8195 flags = BTRFS_BLOCK_GROUP_METADATA;
8196 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8200 flags = BTRFS_BLOCK_GROUP_DATA;
8201 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8207 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8209 return unpin_extent_range(root, start, end);
8212 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8213 u64 num_bytes, u64 *actual_bytes)
8215 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8218 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8220 struct btrfs_fs_info *fs_info = root->fs_info;
8221 struct btrfs_block_group_cache *cache = NULL;
8226 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8230 * try to trim all FS space, our block group may start from non-zero.
8232 if (range->len == total_bytes)
8233 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8235 cache = btrfs_lookup_block_group(fs_info, range->start);
8238 if (cache->key.objectid >= (range->start + range->len)) {
8239 btrfs_put_block_group(cache);
8243 start = max(range->start, cache->key.objectid);
8244 end = min(range->start + range->len,
8245 cache->key.objectid + cache->key.offset);
8247 if (end - start >= range->minlen) {
8248 if (!block_group_cache_done(cache)) {
8249 ret = cache_block_group(cache, 0);
8251 wait_block_group_cache_done(cache);
8253 ret = btrfs_trim_block_group(cache,
8259 trimmed += group_trimmed;
8261 btrfs_put_block_group(cache);
8266 cache = next_block_group(fs_info->tree_root, cache);
8269 range->len = trimmed;