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_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
108 block_group_cache_done(struct btrfs_block_group_cache *cache)
111 return cache->cached == BTRFS_CACHE_FINISHED;
114 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
116 return (cache->flags & bits) == bits;
119 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
121 atomic_inc(&cache->count);
124 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
126 if (atomic_dec_and_test(&cache->count)) {
127 WARN_ON(cache->pinned > 0);
128 WARN_ON(cache->reserved > 0);
129 kfree(cache->free_space_ctl);
135 * this adds the block group to the fs_info rb tree for the block group
138 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
139 struct btrfs_block_group_cache *block_group)
142 struct rb_node *parent = NULL;
143 struct btrfs_block_group_cache *cache;
145 spin_lock(&info->block_group_cache_lock);
146 p = &info->block_group_cache_tree.rb_node;
150 cache = rb_entry(parent, struct btrfs_block_group_cache,
152 if (block_group->key.objectid < cache->key.objectid) {
154 } else if (block_group->key.objectid > cache->key.objectid) {
157 spin_unlock(&info->block_group_cache_lock);
162 rb_link_node(&block_group->cache_node, parent, p);
163 rb_insert_color(&block_group->cache_node,
164 &info->block_group_cache_tree);
165 spin_unlock(&info->block_group_cache_lock);
171 * This will return the block group at or after bytenr if contains is 0, else
172 * it will return the block group that contains the bytenr
174 static struct btrfs_block_group_cache *
175 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
178 struct btrfs_block_group_cache *cache, *ret = NULL;
182 spin_lock(&info->block_group_cache_lock);
183 n = info->block_group_cache_tree.rb_node;
186 cache = rb_entry(n, struct btrfs_block_group_cache,
188 end = cache->key.objectid + cache->key.offset - 1;
189 start = cache->key.objectid;
191 if (bytenr < start) {
192 if (!contains && (!ret || start < ret->key.objectid))
195 } else if (bytenr > start) {
196 if (contains && bytenr <= end) {
207 btrfs_get_block_group(ret);
208 spin_unlock(&info->block_group_cache_lock);
213 static int add_excluded_extent(struct btrfs_root *root,
214 u64 start, u64 num_bytes)
216 u64 end = start + num_bytes - 1;
217 set_extent_bits(&root->fs_info->freed_extents[0],
218 start, end, EXTENT_UPTODATE, GFP_NOFS);
219 set_extent_bits(&root->fs_info->freed_extents[1],
220 start, end, EXTENT_UPTODATE, GFP_NOFS);
224 static void free_excluded_extents(struct btrfs_root *root,
225 struct btrfs_block_group_cache *cache)
229 start = cache->key.objectid;
230 end = start + cache->key.offset - 1;
232 clear_extent_bits(&root->fs_info->freed_extents[0],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 clear_extent_bits(&root->fs_info->freed_extents[1],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
238 static int exclude_super_stripes(struct btrfs_root *root,
239 struct btrfs_block_group_cache *cache)
246 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
247 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
248 cache->bytes_super += stripe_len;
249 ret = add_excluded_extent(root, cache->key.objectid,
251 BUG_ON(ret); /* -ENOMEM */
254 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
255 bytenr = btrfs_sb_offset(i);
256 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
257 cache->key.objectid, bytenr,
258 0, &logical, &nr, &stripe_len);
259 BUG_ON(ret); /* -ENOMEM */
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(root, logical[nr],
265 BUG_ON(ret); /* -ENOMEM */
273 static struct btrfs_caching_control *
274 get_caching_control(struct btrfs_block_group_cache *cache)
276 struct btrfs_caching_control *ctl;
278 spin_lock(&cache->lock);
279 if (cache->cached != BTRFS_CACHE_STARTED) {
280 spin_unlock(&cache->lock);
284 /* We're loading it the fast way, so we don't have a caching_ctl. */
285 if (!cache->caching_ctl) {
286 spin_unlock(&cache->lock);
290 ctl = cache->caching_ctl;
291 atomic_inc(&ctl->count);
292 spin_unlock(&cache->lock);
296 static void put_caching_control(struct btrfs_caching_control *ctl)
298 if (atomic_dec_and_test(&ctl->count))
303 * this is only called by cache_block_group, since we could have freed extents
304 * we need to check the pinned_extents for any extents that can't be used yet
305 * since their free space will be released as soon as the transaction commits.
307 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
308 struct btrfs_fs_info *info, u64 start, u64 end)
310 u64 extent_start, extent_end, size, total_added = 0;
313 while (start < end) {
314 ret = find_first_extent_bit(info->pinned_extents, start,
315 &extent_start, &extent_end,
316 EXTENT_DIRTY | EXTENT_UPTODATE,
321 if (extent_start <= start) {
322 start = extent_end + 1;
323 } else if (extent_start > start && extent_start < end) {
324 size = extent_start - start;
326 ret = btrfs_add_free_space(block_group, start,
328 BUG_ON(ret); /* -ENOMEM or logic error */
329 start = extent_end + 1;
338 ret = btrfs_add_free_space(block_group, start, size);
339 BUG_ON(ret); /* -ENOMEM or logic error */
345 static noinline void caching_thread(struct btrfs_work *work)
347 struct btrfs_block_group_cache *block_group;
348 struct btrfs_fs_info *fs_info;
349 struct btrfs_caching_control *caching_ctl;
350 struct btrfs_root *extent_root;
351 struct btrfs_path *path;
352 struct extent_buffer *leaf;
353 struct btrfs_key key;
359 caching_ctl = container_of(work, struct btrfs_caching_control, work);
360 block_group = caching_ctl->block_group;
361 fs_info = block_group->fs_info;
362 extent_root = fs_info->extent_root;
364 path = btrfs_alloc_path();
368 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
371 * We don't want to deadlock with somebody trying to allocate a new
372 * extent for the extent root while also trying to search the extent
373 * root to add free space. So we skip locking and search the commit
374 * root, since its read-only
376 path->skip_locking = 1;
377 path->search_commit_root = 1;
382 key.type = BTRFS_EXTENT_ITEM_KEY;
384 mutex_lock(&caching_ctl->mutex);
385 /* need to make sure the commit_root doesn't disappear */
386 down_read(&fs_info->extent_commit_sem);
388 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
392 leaf = path->nodes[0];
393 nritems = btrfs_header_nritems(leaf);
396 if (btrfs_fs_closing(fs_info) > 1) {
401 if (path->slots[0] < nritems) {
402 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
404 ret = find_next_key(path, 0, &key);
408 if (need_resched() ||
409 btrfs_next_leaf(extent_root, path)) {
410 caching_ctl->progress = last;
411 btrfs_release_path(path);
412 up_read(&fs_info->extent_commit_sem);
413 mutex_unlock(&caching_ctl->mutex);
417 leaf = path->nodes[0];
418 nritems = btrfs_header_nritems(leaf);
422 if (key.objectid < block_group->key.objectid) {
427 if (key.objectid >= block_group->key.objectid +
428 block_group->key.offset)
431 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
432 total_found += add_new_free_space(block_group,
435 last = key.objectid + key.offset;
437 if (total_found > (1024 * 1024 * 2)) {
439 wake_up(&caching_ctl->wait);
446 total_found += add_new_free_space(block_group, fs_info, last,
447 block_group->key.objectid +
448 block_group->key.offset);
449 caching_ctl->progress = (u64)-1;
451 spin_lock(&block_group->lock);
452 block_group->caching_ctl = NULL;
453 block_group->cached = BTRFS_CACHE_FINISHED;
454 spin_unlock(&block_group->lock);
457 btrfs_free_path(path);
458 up_read(&fs_info->extent_commit_sem);
460 free_excluded_extents(extent_root, block_group);
462 mutex_unlock(&caching_ctl->mutex);
464 wake_up(&caching_ctl->wait);
466 put_caching_control(caching_ctl);
467 btrfs_put_block_group(block_group);
470 static int cache_block_group(struct btrfs_block_group_cache *cache,
471 struct btrfs_trans_handle *trans,
472 struct btrfs_root *root,
476 struct btrfs_fs_info *fs_info = cache->fs_info;
477 struct btrfs_caching_control *caching_ctl;
480 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
484 INIT_LIST_HEAD(&caching_ctl->list);
485 mutex_init(&caching_ctl->mutex);
486 init_waitqueue_head(&caching_ctl->wait);
487 caching_ctl->block_group = cache;
488 caching_ctl->progress = cache->key.objectid;
489 atomic_set(&caching_ctl->count, 1);
490 caching_ctl->work.func = caching_thread;
492 spin_lock(&cache->lock);
494 * This should be a rare occasion, but this could happen I think in the
495 * case where one thread starts to load the space cache info, and then
496 * some other thread starts a transaction commit which tries to do an
497 * allocation while the other thread is still loading the space cache
498 * info. The previous loop should have kept us from choosing this block
499 * group, but if we've moved to the state where we will wait on caching
500 * block groups we need to first check if we're doing a fast load here,
501 * so we can wait for it to finish, otherwise we could end up allocating
502 * from a block group who's cache gets evicted for one reason or
505 while (cache->cached == BTRFS_CACHE_FAST) {
506 struct btrfs_caching_control *ctl;
508 ctl = cache->caching_ctl;
509 atomic_inc(&ctl->count);
510 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
511 spin_unlock(&cache->lock);
515 finish_wait(&ctl->wait, &wait);
516 put_caching_control(ctl);
517 spin_lock(&cache->lock);
520 if (cache->cached != BTRFS_CACHE_NO) {
521 spin_unlock(&cache->lock);
525 WARN_ON(cache->caching_ctl);
526 cache->caching_ctl = caching_ctl;
527 cache->cached = BTRFS_CACHE_FAST;
528 spin_unlock(&cache->lock);
531 * We can't do the read from on-disk cache during a commit since we need
532 * to have the normal tree locking. Also if we are currently trying to
533 * allocate blocks for the tree root we can't do the fast caching since
534 * we likely hold important locks.
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,
2146 mutex_unlock(&head->mutex);
2150 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2151 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2152 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2154 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2155 node->type == BTRFS_SHARED_DATA_REF_KEY)
2156 ret = run_delayed_data_ref(trans, root, node, extent_op,
2163 static noinline struct btrfs_delayed_ref_node *
2164 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2166 struct rb_node *node;
2167 struct btrfs_delayed_ref_node *ref;
2168 int action = BTRFS_ADD_DELAYED_REF;
2171 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2172 * this prevents ref count from going down to zero when
2173 * there still are pending delayed ref.
2175 node = rb_prev(&head->node.rb_node);
2179 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2181 if (ref->bytenr != head->node.bytenr)
2183 if (ref->action == action)
2185 node = rb_prev(node);
2187 if (action == BTRFS_ADD_DELAYED_REF) {
2188 action = BTRFS_DROP_DELAYED_REF;
2195 * Returns 0 on success or if called with an already aborted transaction.
2196 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2198 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *root,
2200 struct list_head *cluster)
2202 struct btrfs_delayed_ref_root *delayed_refs;
2203 struct btrfs_delayed_ref_node *ref;
2204 struct btrfs_delayed_ref_head *locked_ref = NULL;
2205 struct btrfs_delayed_extent_op *extent_op;
2206 struct btrfs_fs_info *fs_info = root->fs_info;
2209 int must_insert_reserved = 0;
2211 delayed_refs = &trans->transaction->delayed_refs;
2214 /* pick a new head ref from the cluster list */
2215 if (list_empty(cluster))
2218 locked_ref = list_entry(cluster->next,
2219 struct btrfs_delayed_ref_head, cluster);
2221 /* grab the lock that says we are going to process
2222 * all the refs for this head */
2223 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2226 * we may have dropped the spin lock to get the head
2227 * mutex lock, and that might have given someone else
2228 * time to free the head. If that's true, it has been
2229 * removed from our list and we can move on.
2231 if (ret == -EAGAIN) {
2239 * We need to try and merge add/drops of the same ref since we
2240 * can run into issues with relocate dropping the implicit ref
2241 * and then it being added back again before the drop can
2242 * finish. If we merged anything we need to re-loop so we can
2245 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2249 * locked_ref is the head node, so we have to go one
2250 * node back for any delayed ref updates
2252 ref = select_delayed_ref(locked_ref);
2254 if (ref && ref->seq &&
2255 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2257 * there are still refs with lower seq numbers in the
2258 * process of being added. Don't run this ref yet.
2260 list_del_init(&locked_ref->cluster);
2261 mutex_unlock(&locked_ref->mutex);
2263 delayed_refs->num_heads_ready++;
2264 spin_unlock(&delayed_refs->lock);
2266 spin_lock(&delayed_refs->lock);
2271 * record the must insert reserved flag before we
2272 * drop the spin lock.
2274 must_insert_reserved = locked_ref->must_insert_reserved;
2275 locked_ref->must_insert_reserved = 0;
2277 extent_op = locked_ref->extent_op;
2278 locked_ref->extent_op = NULL;
2281 /* All delayed refs have been processed, Go ahead
2282 * and send the head node to run_one_delayed_ref,
2283 * so that any accounting fixes can happen
2285 ref = &locked_ref->node;
2287 if (extent_op && must_insert_reserved) {
2293 spin_unlock(&delayed_refs->lock);
2295 ret = run_delayed_extent_op(trans, root,
2300 list_del_init(&locked_ref->cluster);
2301 mutex_unlock(&locked_ref->mutex);
2303 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2304 spin_lock(&delayed_refs->lock);
2311 list_del_init(&locked_ref->cluster);
2316 rb_erase(&ref->rb_node, &delayed_refs->root);
2317 delayed_refs->num_entries--;
2320 * when we play the delayed ref, also correct the
2323 switch (ref->action) {
2324 case BTRFS_ADD_DELAYED_REF:
2325 case BTRFS_ADD_DELAYED_EXTENT:
2326 locked_ref->node.ref_mod -= ref->ref_mod;
2328 case BTRFS_DROP_DELAYED_REF:
2329 locked_ref->node.ref_mod += ref->ref_mod;
2335 spin_unlock(&delayed_refs->lock);
2337 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2338 must_insert_reserved);
2340 btrfs_put_delayed_ref(ref);
2346 list_del_init(&locked_ref->cluster);
2347 mutex_unlock(&locked_ref->mutex);
2349 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2350 spin_lock(&delayed_refs->lock);
2356 spin_lock(&delayed_refs->lock);
2361 #ifdef SCRAMBLE_DELAYED_REFS
2363 * Normally delayed refs get processed in ascending bytenr order. This
2364 * correlates in most cases to the order added. To expose dependencies on this
2365 * order, we start to process the tree in the middle instead of the beginning
2367 static u64 find_middle(struct rb_root *root)
2369 struct rb_node *n = root->rb_node;
2370 struct btrfs_delayed_ref_node *entry;
2373 u64 first = 0, last = 0;
2377 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2378 first = entry->bytenr;
2382 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2383 last = entry->bytenr;
2388 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2389 WARN_ON(!entry->in_tree);
2391 middle = entry->bytenr;
2404 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2405 struct btrfs_fs_info *fs_info)
2407 struct qgroup_update *qgroup_update;
2410 if (list_empty(&trans->qgroup_ref_list) !=
2411 !trans->delayed_ref_elem.seq) {
2412 /* list without seq or seq without list */
2413 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2414 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2415 trans->delayed_ref_elem.seq);
2419 if (!trans->delayed_ref_elem.seq)
2422 while (!list_empty(&trans->qgroup_ref_list)) {
2423 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2424 struct qgroup_update, list);
2425 list_del(&qgroup_update->list);
2427 ret = btrfs_qgroup_account_ref(
2428 trans, fs_info, qgroup_update->node,
2429 qgroup_update->extent_op);
2430 kfree(qgroup_update);
2433 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2439 * this starts processing the delayed reference count updates and
2440 * extent insertions we have queued up so far. count can be
2441 * 0, which means to process everything in the tree at the start
2442 * of the run (but not newly added entries), or it can be some target
2443 * number you'd like to process.
2445 * Returns 0 on success or if called with an aborted transaction
2446 * Returns <0 on error and aborts the transaction
2448 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2449 struct btrfs_root *root, unsigned long count)
2451 struct rb_node *node;
2452 struct btrfs_delayed_ref_root *delayed_refs;
2453 struct btrfs_delayed_ref_node *ref;
2454 struct list_head cluster;
2457 int run_all = count == (unsigned long)-1;
2461 /* We'll clean this up in btrfs_cleanup_transaction */
2465 if (root == root->fs_info->extent_root)
2466 root = root->fs_info->tree_root;
2468 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2470 delayed_refs = &trans->transaction->delayed_refs;
2471 INIT_LIST_HEAD(&cluster);
2474 spin_lock(&delayed_refs->lock);
2476 #ifdef SCRAMBLE_DELAYED_REFS
2477 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2481 count = delayed_refs->num_entries * 2;
2485 if (!(run_all || run_most) &&
2486 delayed_refs->num_heads_ready < 64)
2490 * go find something we can process in the rbtree. We start at
2491 * the beginning of the tree, and then build a cluster
2492 * of refs to process starting at the first one we are able to
2495 delayed_start = delayed_refs->run_delayed_start;
2496 ret = btrfs_find_ref_cluster(trans, &cluster,
2497 delayed_refs->run_delayed_start);
2501 ret = run_clustered_refs(trans, root, &cluster);
2503 spin_unlock(&delayed_refs->lock);
2504 btrfs_abort_transaction(trans, root, ret);
2508 count -= min_t(unsigned long, ret, count);
2513 if (delayed_start >= delayed_refs->run_delayed_start) {
2516 * btrfs_find_ref_cluster looped. let's do one
2517 * more cycle. if we don't run any delayed ref
2518 * during that cycle (because we can't because
2519 * all of them are blocked), bail out.
2524 * no runnable refs left, stop trying
2531 /* refs were run, let's reset staleness detection */
2537 if (!list_empty(&trans->new_bgs)) {
2538 spin_unlock(&delayed_refs->lock);
2539 btrfs_create_pending_block_groups(trans, root);
2540 spin_lock(&delayed_refs->lock);
2543 node = rb_first(&delayed_refs->root);
2546 count = (unsigned long)-1;
2549 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2551 if (btrfs_delayed_ref_is_head(ref)) {
2552 struct btrfs_delayed_ref_head *head;
2554 head = btrfs_delayed_node_to_head(ref);
2555 atomic_inc(&ref->refs);
2557 spin_unlock(&delayed_refs->lock);
2559 * Mutex was contended, block until it's
2560 * released and try again
2562 mutex_lock(&head->mutex);
2563 mutex_unlock(&head->mutex);
2565 btrfs_put_delayed_ref(ref);
2569 node = rb_next(node);
2571 spin_unlock(&delayed_refs->lock);
2572 schedule_timeout(1);
2576 spin_unlock(&delayed_refs->lock);
2577 assert_qgroups_uptodate(trans);
2581 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2582 struct btrfs_root *root,
2583 u64 bytenr, u64 num_bytes, u64 flags,
2586 struct btrfs_delayed_extent_op *extent_op;
2589 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2593 extent_op->flags_to_set = flags;
2594 extent_op->update_flags = 1;
2595 extent_op->update_key = 0;
2596 extent_op->is_data = is_data ? 1 : 0;
2598 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2599 num_bytes, extent_op);
2605 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root,
2607 struct btrfs_path *path,
2608 u64 objectid, u64 offset, u64 bytenr)
2610 struct btrfs_delayed_ref_head *head;
2611 struct btrfs_delayed_ref_node *ref;
2612 struct btrfs_delayed_data_ref *data_ref;
2613 struct btrfs_delayed_ref_root *delayed_refs;
2614 struct rb_node *node;
2618 delayed_refs = &trans->transaction->delayed_refs;
2619 spin_lock(&delayed_refs->lock);
2620 head = btrfs_find_delayed_ref_head(trans, bytenr);
2624 if (!mutex_trylock(&head->mutex)) {
2625 atomic_inc(&head->node.refs);
2626 spin_unlock(&delayed_refs->lock);
2628 btrfs_release_path(path);
2631 * Mutex was contended, block until it's released and let
2634 mutex_lock(&head->mutex);
2635 mutex_unlock(&head->mutex);
2636 btrfs_put_delayed_ref(&head->node);
2640 node = rb_prev(&head->node.rb_node);
2644 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2646 if (ref->bytenr != bytenr)
2650 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2653 data_ref = btrfs_delayed_node_to_data_ref(ref);
2655 node = rb_prev(node);
2659 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2660 if (ref->bytenr == bytenr && ref->seq == seq)
2664 if (data_ref->root != root->root_key.objectid ||
2665 data_ref->objectid != objectid || data_ref->offset != offset)
2670 mutex_unlock(&head->mutex);
2672 spin_unlock(&delayed_refs->lock);
2676 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2677 struct btrfs_root *root,
2678 struct btrfs_path *path,
2679 u64 objectid, u64 offset, u64 bytenr)
2681 struct btrfs_root *extent_root = root->fs_info->extent_root;
2682 struct extent_buffer *leaf;
2683 struct btrfs_extent_data_ref *ref;
2684 struct btrfs_extent_inline_ref *iref;
2685 struct btrfs_extent_item *ei;
2686 struct btrfs_key key;
2690 key.objectid = bytenr;
2691 key.offset = (u64)-1;
2692 key.type = BTRFS_EXTENT_ITEM_KEY;
2694 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2697 BUG_ON(ret == 0); /* Corruption */
2700 if (path->slots[0] == 0)
2704 leaf = path->nodes[0];
2705 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2707 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2711 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2712 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2713 if (item_size < sizeof(*ei)) {
2714 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2718 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2720 if (item_size != sizeof(*ei) +
2721 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2724 if (btrfs_extent_generation(leaf, ei) <=
2725 btrfs_root_last_snapshot(&root->root_item))
2728 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2729 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2730 BTRFS_EXTENT_DATA_REF_KEY)
2733 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2734 if (btrfs_extent_refs(leaf, ei) !=
2735 btrfs_extent_data_ref_count(leaf, ref) ||
2736 btrfs_extent_data_ref_root(leaf, ref) !=
2737 root->root_key.objectid ||
2738 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2739 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2747 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 u64 objectid, u64 offset, u64 bytenr)
2751 struct btrfs_path *path;
2755 path = btrfs_alloc_path();
2760 ret = check_committed_ref(trans, root, path, objectid,
2762 if (ret && ret != -ENOENT)
2765 ret2 = check_delayed_ref(trans, root, path, objectid,
2767 } while (ret2 == -EAGAIN);
2769 if (ret2 && ret2 != -ENOENT) {
2774 if (ret != -ENOENT || ret2 != -ENOENT)
2777 btrfs_free_path(path);
2778 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2783 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct extent_buffer *buf,
2786 int full_backref, int inc, int for_cow)
2793 struct btrfs_key key;
2794 struct btrfs_file_extent_item *fi;
2798 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2799 u64, u64, u64, u64, u64, u64, int);
2801 ref_root = btrfs_header_owner(buf);
2802 nritems = btrfs_header_nritems(buf);
2803 level = btrfs_header_level(buf);
2805 if (!root->ref_cows && level == 0)
2809 process_func = btrfs_inc_extent_ref;
2811 process_func = btrfs_free_extent;
2814 parent = buf->start;
2818 for (i = 0; i < nritems; i++) {
2820 btrfs_item_key_to_cpu(buf, &key, i);
2821 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2823 fi = btrfs_item_ptr(buf, i,
2824 struct btrfs_file_extent_item);
2825 if (btrfs_file_extent_type(buf, fi) ==
2826 BTRFS_FILE_EXTENT_INLINE)
2828 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2832 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2833 key.offset -= btrfs_file_extent_offset(buf, fi);
2834 ret = process_func(trans, root, bytenr, num_bytes,
2835 parent, ref_root, key.objectid,
2836 key.offset, for_cow);
2840 bytenr = btrfs_node_blockptr(buf, i);
2841 num_bytes = btrfs_level_size(root, level - 1);
2842 ret = process_func(trans, root, bytenr, num_bytes,
2843 parent, ref_root, level - 1, 0,
2854 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2855 struct extent_buffer *buf, int full_backref, int for_cow)
2857 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2860 int btrfs_dec_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, 0, for_cow);
2866 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2867 struct btrfs_root *root,
2868 struct btrfs_path *path,
2869 struct btrfs_block_group_cache *cache)
2872 struct btrfs_root *extent_root = root->fs_info->extent_root;
2874 struct extent_buffer *leaf;
2876 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2879 BUG_ON(ret); /* Corruption */
2881 leaf = path->nodes[0];
2882 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2883 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2884 btrfs_mark_buffer_dirty(leaf);
2885 btrfs_release_path(path);
2888 btrfs_abort_transaction(trans, root, ret);
2895 static struct btrfs_block_group_cache *
2896 next_block_group(struct btrfs_root *root,
2897 struct btrfs_block_group_cache *cache)
2899 struct rb_node *node;
2900 spin_lock(&root->fs_info->block_group_cache_lock);
2901 node = rb_next(&cache->cache_node);
2902 btrfs_put_block_group(cache);
2904 cache = rb_entry(node, struct btrfs_block_group_cache,
2906 btrfs_get_block_group(cache);
2909 spin_unlock(&root->fs_info->block_group_cache_lock);
2913 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2914 struct btrfs_trans_handle *trans,
2915 struct btrfs_path *path)
2917 struct btrfs_root *root = block_group->fs_info->tree_root;
2918 struct inode *inode = NULL;
2920 int dcs = BTRFS_DC_ERROR;
2926 * If this block group is smaller than 100 megs don't bother caching the
2929 if (block_group->key.offset < (100 * 1024 * 1024)) {
2930 spin_lock(&block_group->lock);
2931 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2932 spin_unlock(&block_group->lock);
2937 inode = lookup_free_space_inode(root, block_group, path);
2938 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2939 ret = PTR_ERR(inode);
2940 btrfs_release_path(path);
2944 if (IS_ERR(inode)) {
2948 if (block_group->ro)
2951 ret = create_free_space_inode(root, trans, block_group, path);
2957 /* We've already setup this transaction, go ahead and exit */
2958 if (block_group->cache_generation == trans->transid &&
2959 i_size_read(inode)) {
2960 dcs = BTRFS_DC_SETUP;
2965 * We want to set the generation to 0, that way if anything goes wrong
2966 * from here on out we know not to trust this cache when we load up next
2969 BTRFS_I(inode)->generation = 0;
2970 ret = btrfs_update_inode(trans, root, inode);
2973 if (i_size_read(inode) > 0) {
2974 ret = btrfs_truncate_free_space_cache(root, trans, path,
2980 spin_lock(&block_group->lock);
2981 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2982 !btrfs_test_opt(root, SPACE_CACHE)) {
2984 * don't bother trying to write stuff out _if_
2985 * a) we're not cached,
2986 * b) we're with nospace_cache mount option.
2988 dcs = BTRFS_DC_WRITTEN;
2989 spin_unlock(&block_group->lock);
2992 spin_unlock(&block_group->lock);
2995 * Try to preallocate enough space based on how big the block group is.
2996 * Keep in mind this has to include any pinned space which could end up
2997 * taking up quite a bit since it's not folded into the other space
3000 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3005 num_pages *= PAGE_CACHE_SIZE;
3007 ret = btrfs_check_data_free_space(inode, num_pages);
3011 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3012 num_pages, num_pages,
3015 dcs = BTRFS_DC_SETUP;
3016 btrfs_free_reserved_data_space(inode, num_pages);
3021 btrfs_release_path(path);
3023 spin_lock(&block_group->lock);
3024 if (!ret && dcs == BTRFS_DC_SETUP)
3025 block_group->cache_generation = trans->transid;
3026 block_group->disk_cache_state = dcs;
3027 spin_unlock(&block_group->lock);
3032 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root)
3035 struct btrfs_block_group_cache *cache;
3037 struct btrfs_path *path;
3040 path = btrfs_alloc_path();
3046 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3048 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3050 cache = next_block_group(root, cache);
3058 err = cache_save_setup(cache, trans, path);
3059 last = cache->key.objectid + cache->key.offset;
3060 btrfs_put_block_group(cache);
3065 err = btrfs_run_delayed_refs(trans, root,
3067 if (err) /* File system offline */
3071 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3073 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3074 btrfs_put_block_group(cache);
3080 cache = next_block_group(root, cache);
3089 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3090 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3092 last = cache->key.objectid + cache->key.offset;
3094 err = write_one_cache_group(trans, root, path, cache);
3095 if (err) /* File system offline */
3098 btrfs_put_block_group(cache);
3103 * I don't think this is needed since we're just marking our
3104 * preallocated extent as written, but just in case it can't
3108 err = btrfs_run_delayed_refs(trans, root,
3110 if (err) /* File system offline */
3114 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3117 * Really this shouldn't happen, but it could if we
3118 * couldn't write the entire preallocated extent and
3119 * splitting the extent resulted in a new block.
3122 btrfs_put_block_group(cache);
3125 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3127 cache = next_block_group(root, cache);
3136 err = btrfs_write_out_cache(root, trans, cache, path);
3139 * If we didn't have an error then the cache state is still
3140 * NEED_WRITE, so we can set it to WRITTEN.
3142 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3143 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3144 last = cache->key.objectid + cache->key.offset;
3145 btrfs_put_block_group(cache);
3149 btrfs_free_path(path);
3153 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3155 struct btrfs_block_group_cache *block_group;
3158 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3159 if (!block_group || block_group->ro)
3162 btrfs_put_block_group(block_group);
3166 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3167 u64 total_bytes, u64 bytes_used,
3168 struct btrfs_space_info **space_info)
3170 struct btrfs_space_info *found;
3174 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3175 BTRFS_BLOCK_GROUP_RAID10))
3180 found = __find_space_info(info, flags);
3182 spin_lock(&found->lock);
3183 found->total_bytes += total_bytes;
3184 found->disk_total += total_bytes * factor;
3185 found->bytes_used += bytes_used;
3186 found->disk_used += bytes_used * factor;
3188 spin_unlock(&found->lock);
3189 *space_info = found;
3192 found = kzalloc(sizeof(*found), GFP_NOFS);
3196 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3197 INIT_LIST_HEAD(&found->block_groups[i]);
3198 init_rwsem(&found->groups_sem);
3199 spin_lock_init(&found->lock);
3200 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3201 found->total_bytes = total_bytes;
3202 found->disk_total = total_bytes * factor;
3203 found->bytes_used = bytes_used;
3204 found->disk_used = bytes_used * factor;
3205 found->bytes_pinned = 0;
3206 found->bytes_reserved = 0;
3207 found->bytes_readonly = 0;
3208 found->bytes_may_use = 0;
3210 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3211 found->chunk_alloc = 0;
3213 init_waitqueue_head(&found->wait);
3214 *space_info = found;
3215 list_add_rcu(&found->list, &info->space_info);
3216 if (flags & BTRFS_BLOCK_GROUP_DATA)
3217 info->data_sinfo = found;
3221 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3223 u64 extra_flags = chunk_to_extended(flags) &
3224 BTRFS_EXTENDED_PROFILE_MASK;
3226 if (flags & BTRFS_BLOCK_GROUP_DATA)
3227 fs_info->avail_data_alloc_bits |= extra_flags;
3228 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3229 fs_info->avail_metadata_alloc_bits |= extra_flags;
3230 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3231 fs_info->avail_system_alloc_bits |= extra_flags;
3235 * returns target flags in extended format or 0 if restripe for this
3236 * chunk_type is not in progress
3238 * should be called with either volume_mutex or balance_lock held
3240 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3242 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3248 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3249 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3250 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3251 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3252 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3253 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3254 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3255 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3256 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3263 * @flags: available profiles in extended format (see ctree.h)
3265 * Returns reduced profile in chunk format. If profile changing is in
3266 * progress (either running or paused) picks the target profile (if it's
3267 * already available), otherwise falls back to plain reducing.
3269 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3272 * we add in the count of missing devices because we want
3273 * to make sure that any RAID levels on a degraded FS
3274 * continue to be honored.
3276 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3277 root->fs_info->fs_devices->missing_devices;
3281 * see if restripe for this chunk_type is in progress, if so
3282 * try to reduce to the target profile
3284 spin_lock(&root->fs_info->balance_lock);
3285 target = get_restripe_target(root->fs_info, flags);
3287 /* pick target profile only if it's already available */
3288 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3289 spin_unlock(&root->fs_info->balance_lock);
3290 return extended_to_chunk(target);
3293 spin_unlock(&root->fs_info->balance_lock);
3295 if (num_devices == 1)
3296 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3297 if (num_devices < 4)
3298 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3300 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3301 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3302 BTRFS_BLOCK_GROUP_RAID10))) {
3303 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3306 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3307 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3308 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3311 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3312 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3313 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3314 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3315 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3318 return extended_to_chunk(flags);
3321 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3323 if (flags & BTRFS_BLOCK_GROUP_DATA)
3324 flags |= root->fs_info->avail_data_alloc_bits;
3325 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3326 flags |= root->fs_info->avail_system_alloc_bits;
3327 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3328 flags |= root->fs_info->avail_metadata_alloc_bits;
3330 return btrfs_reduce_alloc_profile(root, flags);
3333 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3338 flags = BTRFS_BLOCK_GROUP_DATA;
3339 else if (root == root->fs_info->chunk_root)
3340 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3342 flags = BTRFS_BLOCK_GROUP_METADATA;
3344 return get_alloc_profile(root, flags);
3348 * This will check the space that the inode allocates from to make sure we have
3349 * enough space for bytes.
3351 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3353 struct btrfs_space_info *data_sinfo;
3354 struct btrfs_root *root = BTRFS_I(inode)->root;
3355 struct btrfs_fs_info *fs_info = root->fs_info;
3357 int ret = 0, committed = 0, alloc_chunk = 1;
3359 /* make sure bytes are sectorsize aligned */
3360 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3362 if (root == root->fs_info->tree_root ||
3363 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3368 data_sinfo = fs_info->data_sinfo;
3373 /* make sure we have enough space to handle the data first */
3374 spin_lock(&data_sinfo->lock);
3375 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3376 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3377 data_sinfo->bytes_may_use;
3379 if (used + bytes > data_sinfo->total_bytes) {
3380 struct btrfs_trans_handle *trans;
3383 * if we don't have enough free bytes in this space then we need
3384 * to alloc a new chunk.
3386 if (!data_sinfo->full && alloc_chunk) {
3389 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3390 spin_unlock(&data_sinfo->lock);
3392 alloc_target = btrfs_get_alloc_profile(root, 1);
3393 trans = btrfs_join_transaction(root);
3395 return PTR_ERR(trans);
3397 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3399 CHUNK_ALLOC_NO_FORCE);
3400 btrfs_end_transaction(trans, root);
3409 data_sinfo = fs_info->data_sinfo;
3415 * If we have less pinned bytes than we want to allocate then
3416 * don't bother committing the transaction, it won't help us.
3418 if (data_sinfo->bytes_pinned < bytes)
3420 spin_unlock(&data_sinfo->lock);
3422 /* commit the current transaction and try again */
3425 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3427 trans = btrfs_join_transaction(root);
3429 return PTR_ERR(trans);
3430 ret = btrfs_commit_transaction(trans, root);
3438 data_sinfo->bytes_may_use += bytes;
3439 trace_btrfs_space_reservation(root->fs_info, "space_info",
3440 data_sinfo->flags, bytes, 1);
3441 spin_unlock(&data_sinfo->lock);
3447 * Called if we need to clear a data reservation for this inode.
3449 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3451 struct btrfs_root *root = BTRFS_I(inode)->root;
3452 struct btrfs_space_info *data_sinfo;
3454 /* make sure bytes are sectorsize aligned */
3455 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3457 data_sinfo = root->fs_info->data_sinfo;
3458 spin_lock(&data_sinfo->lock);
3459 data_sinfo->bytes_may_use -= bytes;
3460 trace_btrfs_space_reservation(root->fs_info, "space_info",
3461 data_sinfo->flags, bytes, 0);
3462 spin_unlock(&data_sinfo->lock);
3465 static void force_metadata_allocation(struct btrfs_fs_info *info)
3467 struct list_head *head = &info->space_info;
3468 struct btrfs_space_info *found;
3471 list_for_each_entry_rcu(found, head, list) {
3472 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3473 found->force_alloc = CHUNK_ALLOC_FORCE;
3478 static int should_alloc_chunk(struct btrfs_root *root,
3479 struct btrfs_space_info *sinfo, int force)
3481 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3482 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3483 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3486 if (force == CHUNK_ALLOC_FORCE)
3490 * We need to take into account the global rsv because for all intents
3491 * and purposes it's used space. Don't worry about locking the
3492 * global_rsv, it doesn't change except when the transaction commits.
3494 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3495 num_allocated += global_rsv->size;
3498 * in limited mode, we want to have some free space up to
3499 * about 1% of the FS size.
3501 if (force == CHUNK_ALLOC_LIMITED) {
3502 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3503 thresh = max_t(u64, 64 * 1024 * 1024,
3504 div_factor_fine(thresh, 1));
3506 if (num_bytes - num_allocated < thresh)
3510 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3515 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3519 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3520 type & BTRFS_BLOCK_GROUP_RAID0)
3521 num_dev = root->fs_info->fs_devices->rw_devices;
3522 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3525 num_dev = 1; /* DUP or single */
3527 /* metadata for updaing devices and chunk tree */
3528 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3531 static void check_system_chunk(struct btrfs_trans_handle *trans,
3532 struct btrfs_root *root, u64 type)
3534 struct btrfs_space_info *info;
3538 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3539 spin_lock(&info->lock);
3540 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3541 info->bytes_reserved - info->bytes_readonly;
3542 spin_unlock(&info->lock);
3544 thresh = get_system_chunk_thresh(root, type);
3545 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3546 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3547 left, thresh, type);
3548 dump_space_info(info, 0, 0);
3551 if (left < thresh) {
3554 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3555 btrfs_alloc_chunk(trans, root, flags);
3559 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3560 struct btrfs_root *extent_root, u64 flags, int force)
3562 struct btrfs_space_info *space_info;
3563 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3564 int wait_for_alloc = 0;
3567 space_info = __find_space_info(extent_root->fs_info, flags);
3569 ret = update_space_info(extent_root->fs_info, flags,
3571 BUG_ON(ret); /* -ENOMEM */
3573 BUG_ON(!space_info); /* Logic error */
3576 spin_lock(&space_info->lock);
3577 if (force < space_info->force_alloc)
3578 force = space_info->force_alloc;
3579 if (space_info->full) {
3580 spin_unlock(&space_info->lock);
3584 if (!should_alloc_chunk(extent_root, space_info, force)) {
3585 spin_unlock(&space_info->lock);
3587 } else if (space_info->chunk_alloc) {
3590 space_info->chunk_alloc = 1;
3593 spin_unlock(&space_info->lock);
3595 mutex_lock(&fs_info->chunk_mutex);
3598 * The chunk_mutex is held throughout the entirety of a chunk
3599 * allocation, so once we've acquired the chunk_mutex we know that the
3600 * other guy is done and we need to recheck and see if we should
3603 if (wait_for_alloc) {
3604 mutex_unlock(&fs_info->chunk_mutex);
3610 * If we have mixed data/metadata chunks we want to make sure we keep
3611 * allocating mixed chunks instead of individual chunks.
3613 if (btrfs_mixed_space_info(space_info))
3614 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3617 * if we're doing a data chunk, go ahead and make sure that
3618 * we keep a reasonable number of metadata chunks allocated in the
3621 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3622 fs_info->data_chunk_allocations++;
3623 if (!(fs_info->data_chunk_allocations %
3624 fs_info->metadata_ratio))
3625 force_metadata_allocation(fs_info);
3629 * Check if we have enough space in SYSTEM chunk because we may need
3630 * to update devices.
3632 check_system_chunk(trans, extent_root, flags);
3634 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3635 if (ret < 0 && ret != -ENOSPC)
3638 spin_lock(&space_info->lock);
3640 space_info->full = 1;
3644 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3645 space_info->chunk_alloc = 0;
3646 spin_unlock(&space_info->lock);
3648 mutex_unlock(&fs_info->chunk_mutex);
3652 static int can_overcommit(struct btrfs_root *root,
3653 struct btrfs_space_info *space_info, u64 bytes,
3654 enum btrfs_reserve_flush_enum flush)
3656 u64 profile = btrfs_get_alloc_profile(root, 0);
3660 used = space_info->bytes_used + space_info->bytes_reserved +
3661 space_info->bytes_pinned + space_info->bytes_readonly +
3662 space_info->bytes_may_use;
3664 spin_lock(&root->fs_info->free_chunk_lock);
3665 avail = root->fs_info->free_chunk_space;
3666 spin_unlock(&root->fs_info->free_chunk_lock);
3669 * If we have dup, raid1 or raid10 then only half of the free
3670 * space is actually useable.
3672 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3673 BTRFS_BLOCK_GROUP_RAID1 |
3674 BTRFS_BLOCK_GROUP_RAID10))
3678 * If we aren't flushing all things, let us overcommit up to
3679 * 1/2th of the space. If we can flush, don't let us overcommit
3680 * too much, let it overcommit up to 1/8 of the space.
3682 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3687 if (used + bytes < space_info->total_bytes + avail)
3692 static int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3693 unsigned long nr_pages,
3694 enum wb_reason reason)
3696 if (!writeback_in_progress(sb->s_bdi) &&
3697 down_read_trylock(&sb->s_umount)) {
3698 writeback_inodes_sb_nr(sb, nr_pages, reason);
3699 up_read(&sb->s_umount);
3707 * shrink metadata reservation for delalloc
3709 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3712 struct btrfs_block_rsv *block_rsv;
3713 struct btrfs_space_info *space_info;
3714 struct btrfs_trans_handle *trans;
3718 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3720 enum btrfs_reserve_flush_enum flush;
3722 trans = (struct btrfs_trans_handle *)current->journal_info;
3723 block_rsv = &root->fs_info->delalloc_block_rsv;
3724 space_info = block_rsv->space_info;
3727 delalloc_bytes = root->fs_info->delalloc_bytes;
3728 if (delalloc_bytes == 0) {
3731 btrfs_wait_ordered_extents(root, 0);
3735 while (delalloc_bytes && loops < 3) {
3736 max_reclaim = min(delalloc_bytes, to_reclaim);
3737 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3738 writeback_inodes_sb_nr_if_idle_safe(root->fs_info->sb,
3740 WB_REASON_FS_FREE_SPACE);
3743 * We need to wait for the async pages to actually start before
3746 wait_event(root->fs_info->async_submit_wait,
3747 !atomic_read(&root->fs_info->async_delalloc_pages));
3750 flush = BTRFS_RESERVE_FLUSH_ALL;
3752 flush = BTRFS_RESERVE_NO_FLUSH;
3753 spin_lock(&space_info->lock);
3754 if (can_overcommit(root, space_info, orig, flush)) {
3755 spin_unlock(&space_info->lock);
3758 spin_unlock(&space_info->lock);
3761 if (wait_ordered && !trans) {
3762 btrfs_wait_ordered_extents(root, 0);
3764 time_left = schedule_timeout_killable(1);
3769 delalloc_bytes = root->fs_info->delalloc_bytes;
3774 * maybe_commit_transaction - possibly commit the transaction if its ok to
3775 * @root - the root we're allocating for
3776 * @bytes - the number of bytes we want to reserve
3777 * @force - force the commit
3779 * This will check to make sure that committing the transaction will actually
3780 * get us somewhere and then commit the transaction if it does. Otherwise it
3781 * will return -ENOSPC.
3783 static int may_commit_transaction(struct btrfs_root *root,
3784 struct btrfs_space_info *space_info,
3785 u64 bytes, int force)
3787 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3788 struct btrfs_trans_handle *trans;
3790 trans = (struct btrfs_trans_handle *)current->journal_info;
3797 /* See if there is enough pinned space to make this reservation */
3798 spin_lock(&space_info->lock);
3799 if (space_info->bytes_pinned >= bytes) {
3800 spin_unlock(&space_info->lock);
3803 spin_unlock(&space_info->lock);
3806 * See if there is some space in the delayed insertion reservation for
3809 if (space_info != delayed_rsv->space_info)
3812 spin_lock(&space_info->lock);
3813 spin_lock(&delayed_rsv->lock);
3814 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3815 spin_unlock(&delayed_rsv->lock);
3816 spin_unlock(&space_info->lock);
3819 spin_unlock(&delayed_rsv->lock);
3820 spin_unlock(&space_info->lock);
3823 trans = btrfs_join_transaction(root);
3827 return btrfs_commit_transaction(trans, root);
3831 FLUSH_DELAYED_ITEMS_NR = 1,
3832 FLUSH_DELAYED_ITEMS = 2,
3834 FLUSH_DELALLOC_WAIT = 4,
3839 static int flush_space(struct btrfs_root *root,
3840 struct btrfs_space_info *space_info, u64 num_bytes,
3841 u64 orig_bytes, int state)
3843 struct btrfs_trans_handle *trans;
3848 case FLUSH_DELAYED_ITEMS_NR:
3849 case FLUSH_DELAYED_ITEMS:
3850 if (state == FLUSH_DELAYED_ITEMS_NR) {
3851 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3853 nr = (int)div64_u64(num_bytes, bytes);
3860 trans = btrfs_join_transaction(root);
3861 if (IS_ERR(trans)) {
3862 ret = PTR_ERR(trans);
3865 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3866 btrfs_end_transaction(trans, root);
3868 case FLUSH_DELALLOC:
3869 case FLUSH_DELALLOC_WAIT:
3870 shrink_delalloc(root, num_bytes, orig_bytes,
3871 state == FLUSH_DELALLOC_WAIT);
3874 trans = btrfs_join_transaction(root);
3875 if (IS_ERR(trans)) {
3876 ret = PTR_ERR(trans);
3879 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3880 btrfs_get_alloc_profile(root, 0),
3881 CHUNK_ALLOC_NO_FORCE);
3882 btrfs_end_transaction(trans, root);
3887 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3897 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3898 * @root - the root we're allocating for
3899 * @block_rsv - the block_rsv we're allocating for
3900 * @orig_bytes - the number of bytes we want
3901 * @flush - wether or not we can flush to make our reservation
3903 * This will reserve orgi_bytes number of bytes from the space info associated
3904 * with the block_rsv. If there is not enough space it will make an attempt to
3905 * flush out space to make room. It will do this by flushing delalloc if
3906 * possible or committing the transaction. If flush is 0 then no attempts to
3907 * regain reservations will be made and this will fail if there is not enough
3910 static int reserve_metadata_bytes(struct btrfs_root *root,
3911 struct btrfs_block_rsv *block_rsv,
3913 enum btrfs_reserve_flush_enum flush)
3915 struct btrfs_space_info *space_info = block_rsv->space_info;
3917 u64 num_bytes = orig_bytes;
3918 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3920 bool flushing = false;
3924 spin_lock(&space_info->lock);
3926 * We only want to wait if somebody other than us is flushing and we
3927 * are actually allowed to flush all things.
3929 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3930 space_info->flush) {
3931 spin_unlock(&space_info->lock);
3933 * If we have a trans handle we can't wait because the flusher
3934 * may have to commit the transaction, which would mean we would
3935 * deadlock since we are waiting for the flusher to finish, but
3936 * hold the current transaction open.
3938 if (current->journal_info)
3940 ret = wait_event_killable(space_info->wait, !space_info->flush);
3941 /* Must have been killed, return */
3945 spin_lock(&space_info->lock);
3949 used = space_info->bytes_used + space_info->bytes_reserved +
3950 space_info->bytes_pinned + space_info->bytes_readonly +
3951 space_info->bytes_may_use;
3954 * The idea here is that we've not already over-reserved the block group
3955 * then we can go ahead and save our reservation first and then start
3956 * flushing if we need to. Otherwise if we've already overcommitted
3957 * lets start flushing stuff first and then come back and try to make
3960 if (used <= space_info->total_bytes) {
3961 if (used + orig_bytes <= space_info->total_bytes) {
3962 space_info->bytes_may_use += orig_bytes;
3963 trace_btrfs_space_reservation(root->fs_info,
3964 "space_info", space_info->flags, orig_bytes, 1);
3968 * Ok set num_bytes to orig_bytes since we aren't
3969 * overocmmitted, this way we only try and reclaim what
3972 num_bytes = orig_bytes;
3976 * Ok we're over committed, set num_bytes to the overcommitted
3977 * amount plus the amount of bytes that we need for this
3980 num_bytes = used - space_info->total_bytes +
3984 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
3985 space_info->bytes_may_use += orig_bytes;
3986 trace_btrfs_space_reservation(root->fs_info, "space_info",
3987 space_info->flags, orig_bytes,
3993 * Couldn't make our reservation, save our place so while we're trying
3994 * to reclaim space we can actually use it instead of somebody else
3995 * stealing it from us.
3997 * We make the other tasks wait for the flush only when we can flush
4000 if (ret && flush == BTRFS_RESERVE_FLUSH_ALL) {
4002 space_info->flush = 1;
4005 spin_unlock(&space_info->lock);
4007 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4010 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4015 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4016 * would happen. So skip delalloc flush.
4018 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4019 (flush_state == FLUSH_DELALLOC ||
4020 flush_state == FLUSH_DELALLOC_WAIT))
4021 flush_state = ALLOC_CHUNK;
4025 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4026 flush_state < COMMIT_TRANS)
4028 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4029 flush_state <= COMMIT_TRANS)
4034 spin_lock(&space_info->lock);
4035 space_info->flush = 0;
4036 wake_up_all(&space_info->wait);
4037 spin_unlock(&space_info->lock);
4042 static struct btrfs_block_rsv *get_block_rsv(
4043 const struct btrfs_trans_handle *trans,
4044 const struct btrfs_root *root)
4046 struct btrfs_block_rsv *block_rsv = NULL;
4049 block_rsv = trans->block_rsv;
4051 if (root == root->fs_info->csum_root && trans->adding_csums)
4052 block_rsv = trans->block_rsv;
4055 block_rsv = root->block_rsv;
4058 block_rsv = &root->fs_info->empty_block_rsv;
4063 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4067 spin_lock(&block_rsv->lock);
4068 if (block_rsv->reserved >= num_bytes) {
4069 block_rsv->reserved -= num_bytes;
4070 if (block_rsv->reserved < block_rsv->size)
4071 block_rsv->full = 0;
4074 spin_unlock(&block_rsv->lock);
4078 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4079 u64 num_bytes, int update_size)
4081 spin_lock(&block_rsv->lock);
4082 block_rsv->reserved += num_bytes;
4084 block_rsv->size += num_bytes;
4085 else if (block_rsv->reserved >= block_rsv->size)
4086 block_rsv->full = 1;
4087 spin_unlock(&block_rsv->lock);
4090 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4091 struct btrfs_block_rsv *block_rsv,
4092 struct btrfs_block_rsv *dest, u64 num_bytes)
4094 struct btrfs_space_info *space_info = block_rsv->space_info;
4096 spin_lock(&block_rsv->lock);
4097 if (num_bytes == (u64)-1)
4098 num_bytes = block_rsv->size;
4099 block_rsv->size -= num_bytes;
4100 if (block_rsv->reserved >= block_rsv->size) {
4101 num_bytes = block_rsv->reserved - block_rsv->size;
4102 block_rsv->reserved = block_rsv->size;
4103 block_rsv->full = 1;
4107 spin_unlock(&block_rsv->lock);
4109 if (num_bytes > 0) {
4111 spin_lock(&dest->lock);
4115 bytes_to_add = dest->size - dest->reserved;
4116 bytes_to_add = min(num_bytes, bytes_to_add);
4117 dest->reserved += bytes_to_add;
4118 if (dest->reserved >= dest->size)
4120 num_bytes -= bytes_to_add;
4122 spin_unlock(&dest->lock);
4125 spin_lock(&space_info->lock);
4126 space_info->bytes_may_use -= num_bytes;
4127 trace_btrfs_space_reservation(fs_info, "space_info",
4128 space_info->flags, num_bytes, 0);
4129 space_info->reservation_progress++;
4130 spin_unlock(&space_info->lock);
4135 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4136 struct btrfs_block_rsv *dst, u64 num_bytes)
4140 ret = block_rsv_use_bytes(src, num_bytes);
4144 block_rsv_add_bytes(dst, num_bytes, 1);
4148 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4150 memset(rsv, 0, sizeof(*rsv));
4151 spin_lock_init(&rsv->lock);
4155 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4156 unsigned short type)
4158 struct btrfs_block_rsv *block_rsv;
4159 struct btrfs_fs_info *fs_info = root->fs_info;
4161 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4165 btrfs_init_block_rsv(block_rsv, type);
4166 block_rsv->space_info = __find_space_info(fs_info,
4167 BTRFS_BLOCK_GROUP_METADATA);
4171 void btrfs_free_block_rsv(struct btrfs_root *root,
4172 struct btrfs_block_rsv *rsv)
4176 btrfs_block_rsv_release(root, rsv, (u64)-1);
4180 int btrfs_block_rsv_add(struct btrfs_root *root,
4181 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4182 enum btrfs_reserve_flush_enum flush)
4189 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4191 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4198 int btrfs_block_rsv_check(struct btrfs_root *root,
4199 struct btrfs_block_rsv *block_rsv, int min_factor)
4207 spin_lock(&block_rsv->lock);
4208 num_bytes = div_factor(block_rsv->size, min_factor);
4209 if (block_rsv->reserved >= num_bytes)
4211 spin_unlock(&block_rsv->lock);
4216 int btrfs_block_rsv_refill(struct btrfs_root *root,
4217 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4218 enum btrfs_reserve_flush_enum flush)
4226 spin_lock(&block_rsv->lock);
4227 num_bytes = min_reserved;
4228 if (block_rsv->reserved >= num_bytes)
4231 num_bytes -= block_rsv->reserved;
4232 spin_unlock(&block_rsv->lock);
4237 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4239 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4246 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4247 struct btrfs_block_rsv *dst_rsv,
4250 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4253 void btrfs_block_rsv_release(struct btrfs_root *root,
4254 struct btrfs_block_rsv *block_rsv,
4257 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4258 if (global_rsv->full || global_rsv == block_rsv ||
4259 block_rsv->space_info != global_rsv->space_info)
4261 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4266 * helper to calculate size of global block reservation.
4267 * the desired value is sum of space used by extent tree,
4268 * checksum tree and root tree
4270 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4272 struct btrfs_space_info *sinfo;
4276 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4278 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4279 spin_lock(&sinfo->lock);
4280 data_used = sinfo->bytes_used;
4281 spin_unlock(&sinfo->lock);
4283 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4284 spin_lock(&sinfo->lock);
4285 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4287 meta_used = sinfo->bytes_used;
4288 spin_unlock(&sinfo->lock);
4290 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4292 num_bytes += div64_u64(data_used + meta_used, 50);
4294 if (num_bytes * 3 > meta_used)
4295 num_bytes = div64_u64(meta_used, 3);
4297 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4300 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4302 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4303 struct btrfs_space_info *sinfo = block_rsv->space_info;
4306 num_bytes = calc_global_metadata_size(fs_info);
4308 spin_lock(&sinfo->lock);
4309 spin_lock(&block_rsv->lock);
4311 block_rsv->size = num_bytes;
4313 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4314 sinfo->bytes_reserved + sinfo->bytes_readonly +
4315 sinfo->bytes_may_use;
4317 if (sinfo->total_bytes > num_bytes) {
4318 num_bytes = sinfo->total_bytes - num_bytes;
4319 block_rsv->reserved += num_bytes;
4320 sinfo->bytes_may_use += num_bytes;
4321 trace_btrfs_space_reservation(fs_info, "space_info",
4322 sinfo->flags, num_bytes, 1);
4325 if (block_rsv->reserved >= block_rsv->size) {
4326 num_bytes = block_rsv->reserved - block_rsv->size;
4327 sinfo->bytes_may_use -= num_bytes;
4328 trace_btrfs_space_reservation(fs_info, "space_info",
4329 sinfo->flags, num_bytes, 0);
4330 sinfo->reservation_progress++;
4331 block_rsv->reserved = block_rsv->size;
4332 block_rsv->full = 1;
4335 spin_unlock(&block_rsv->lock);
4336 spin_unlock(&sinfo->lock);
4339 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4341 struct btrfs_space_info *space_info;
4343 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4344 fs_info->chunk_block_rsv.space_info = space_info;
4346 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4347 fs_info->global_block_rsv.space_info = space_info;
4348 fs_info->delalloc_block_rsv.space_info = space_info;
4349 fs_info->trans_block_rsv.space_info = space_info;
4350 fs_info->empty_block_rsv.space_info = space_info;
4351 fs_info->delayed_block_rsv.space_info = space_info;
4353 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4354 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4355 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4356 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4357 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4359 update_global_block_rsv(fs_info);
4362 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4364 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4366 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4367 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4368 WARN_ON(fs_info->trans_block_rsv.size > 0);
4369 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4370 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4371 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4372 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4373 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4376 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4377 struct btrfs_root *root)
4379 if (!trans->block_rsv)
4382 if (!trans->bytes_reserved)
4385 trace_btrfs_space_reservation(root->fs_info, "transaction",
4386 trans->transid, trans->bytes_reserved, 0);
4387 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4388 trans->bytes_reserved = 0;
4391 /* Can only return 0 or -ENOSPC */
4392 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4393 struct inode *inode)
4395 struct btrfs_root *root = BTRFS_I(inode)->root;
4396 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4397 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4400 * We need to hold space in order to delete our orphan item once we've
4401 * added it, so this takes the reservation so we can release it later
4402 * when we are truly done with the orphan item.
4404 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4405 trace_btrfs_space_reservation(root->fs_info, "orphan",
4406 btrfs_ino(inode), num_bytes, 1);
4407 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4410 void btrfs_orphan_release_metadata(struct inode *inode)
4412 struct btrfs_root *root = BTRFS_I(inode)->root;
4413 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4414 trace_btrfs_space_reservation(root->fs_info, "orphan",
4415 btrfs_ino(inode), num_bytes, 0);
4416 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4419 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4420 struct btrfs_pending_snapshot *pending)
4422 struct btrfs_root *root = pending->root;
4423 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4424 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4426 * two for root back/forward refs, two for directory entries,
4427 * one for root of the snapshot and one for parent inode.
4429 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4430 dst_rsv->space_info = src_rsv->space_info;
4431 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4435 * drop_outstanding_extent - drop an outstanding extent
4436 * @inode: the inode we're dropping the extent for
4438 * This is called when we are freeing up an outstanding extent, either called
4439 * after an error or after an extent is written. This will return the number of
4440 * reserved extents that need to be freed. This must be called with
4441 * BTRFS_I(inode)->lock held.
4443 static unsigned drop_outstanding_extent(struct inode *inode)
4445 unsigned drop_inode_space = 0;
4446 unsigned dropped_extents = 0;
4448 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4449 BTRFS_I(inode)->outstanding_extents--;
4451 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4452 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4453 &BTRFS_I(inode)->runtime_flags))
4454 drop_inode_space = 1;
4457 * If we have more or the same amount of outsanding extents than we have
4458 * reserved then we need to leave the reserved extents count alone.
4460 if (BTRFS_I(inode)->outstanding_extents >=
4461 BTRFS_I(inode)->reserved_extents)
4462 return drop_inode_space;
4464 dropped_extents = BTRFS_I(inode)->reserved_extents -
4465 BTRFS_I(inode)->outstanding_extents;
4466 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4467 return dropped_extents + drop_inode_space;
4471 * calc_csum_metadata_size - return the amount of metada space that must be
4472 * reserved/free'd for the given bytes.
4473 * @inode: the inode we're manipulating
4474 * @num_bytes: the number of bytes in question
4475 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4477 * This adjusts the number of csum_bytes in the inode and then returns the
4478 * correct amount of metadata that must either be reserved or freed. We
4479 * calculate how many checksums we can fit into one leaf and then divide the
4480 * number of bytes that will need to be checksumed by this value to figure out
4481 * how many checksums will be required. If we are adding bytes then the number
4482 * may go up and we will return the number of additional bytes that must be
4483 * reserved. If it is going down we will return the number of bytes that must
4486 * This must be called with BTRFS_I(inode)->lock held.
4488 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4491 struct btrfs_root *root = BTRFS_I(inode)->root;
4493 int num_csums_per_leaf;
4497 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4498 BTRFS_I(inode)->csum_bytes == 0)
4501 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4503 BTRFS_I(inode)->csum_bytes += num_bytes;
4505 BTRFS_I(inode)->csum_bytes -= num_bytes;
4506 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4507 num_csums_per_leaf = (int)div64_u64(csum_size,
4508 sizeof(struct btrfs_csum_item) +
4509 sizeof(struct btrfs_disk_key));
4510 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4511 num_csums = num_csums + num_csums_per_leaf - 1;
4512 num_csums = num_csums / num_csums_per_leaf;
4514 old_csums = old_csums + num_csums_per_leaf - 1;
4515 old_csums = old_csums / num_csums_per_leaf;
4517 /* No change, no need to reserve more */
4518 if (old_csums == num_csums)
4522 return btrfs_calc_trans_metadata_size(root,
4523 num_csums - old_csums);
4525 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4528 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4530 struct btrfs_root *root = BTRFS_I(inode)->root;
4531 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4534 unsigned nr_extents = 0;
4535 int extra_reserve = 0;
4536 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4539 /* Need to be holding the i_mutex here if we aren't free space cache */
4540 if (btrfs_is_free_space_inode(inode))
4541 flush = BTRFS_RESERVE_NO_FLUSH;
4543 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4544 btrfs_transaction_in_commit(root->fs_info))
4545 schedule_timeout(1);
4547 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4548 num_bytes = ALIGN(num_bytes, root->sectorsize);
4550 spin_lock(&BTRFS_I(inode)->lock);
4551 BTRFS_I(inode)->outstanding_extents++;
4553 if (BTRFS_I(inode)->outstanding_extents >
4554 BTRFS_I(inode)->reserved_extents)
4555 nr_extents = BTRFS_I(inode)->outstanding_extents -
4556 BTRFS_I(inode)->reserved_extents;
4559 * Add an item to reserve for updating the inode when we complete the
4562 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4563 &BTRFS_I(inode)->runtime_flags)) {
4568 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4569 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4570 csum_bytes = BTRFS_I(inode)->csum_bytes;
4571 spin_unlock(&BTRFS_I(inode)->lock);
4573 if (root->fs_info->quota_enabled) {
4574 ret = btrfs_qgroup_reserve(root, num_bytes +
4575 nr_extents * root->leafsize);
4577 spin_lock(&BTRFS_I(inode)->lock);
4578 calc_csum_metadata_size(inode, num_bytes, 0);
4579 spin_unlock(&BTRFS_I(inode)->lock);
4580 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4585 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4590 spin_lock(&BTRFS_I(inode)->lock);
4591 dropped = drop_outstanding_extent(inode);
4593 * If the inodes csum_bytes is the same as the original
4594 * csum_bytes then we know we haven't raced with any free()ers
4595 * so we can just reduce our inodes csum bytes and carry on.
4596 * Otherwise we have to do the normal free thing to account for
4597 * the case that the free side didn't free up its reserve
4598 * because of this outstanding reservation.
4600 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4601 calc_csum_metadata_size(inode, num_bytes, 0);
4603 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4604 spin_unlock(&BTRFS_I(inode)->lock);
4606 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4609 btrfs_block_rsv_release(root, block_rsv, to_free);
4610 trace_btrfs_space_reservation(root->fs_info,
4615 if (root->fs_info->quota_enabled) {
4616 btrfs_qgroup_free(root, num_bytes +
4617 nr_extents * root->leafsize);
4619 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4623 spin_lock(&BTRFS_I(inode)->lock);
4624 if (extra_reserve) {
4625 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4626 &BTRFS_I(inode)->runtime_flags);
4629 BTRFS_I(inode)->reserved_extents += nr_extents;
4630 spin_unlock(&BTRFS_I(inode)->lock);
4631 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4634 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4635 btrfs_ino(inode), to_reserve, 1);
4636 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4642 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4643 * @inode: the inode to release the reservation for
4644 * @num_bytes: the number of bytes we're releasing
4646 * This will release the metadata reservation for an inode. This can be called
4647 * once we complete IO for a given set of bytes to release their metadata
4650 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4652 struct btrfs_root *root = BTRFS_I(inode)->root;
4656 num_bytes = ALIGN(num_bytes, root->sectorsize);
4657 spin_lock(&BTRFS_I(inode)->lock);
4658 dropped = drop_outstanding_extent(inode);
4660 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4661 spin_unlock(&BTRFS_I(inode)->lock);
4663 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4665 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4666 btrfs_ino(inode), to_free, 0);
4667 if (root->fs_info->quota_enabled) {
4668 btrfs_qgroup_free(root, num_bytes +
4669 dropped * root->leafsize);
4672 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4677 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4678 * @inode: inode we're writing to
4679 * @num_bytes: the number of bytes we want to allocate
4681 * This will do the following things
4683 * o reserve space in the data space info for num_bytes
4684 * o reserve space in the metadata space info based on number of outstanding
4685 * extents and how much csums will be needed
4686 * o add to the inodes ->delalloc_bytes
4687 * o add it to the fs_info's delalloc inodes list.
4689 * This will return 0 for success and -ENOSPC if there is no space left.
4691 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4695 ret = btrfs_check_data_free_space(inode, num_bytes);
4699 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4701 btrfs_free_reserved_data_space(inode, num_bytes);
4709 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4710 * @inode: inode we're releasing space for
4711 * @num_bytes: the number of bytes we want to free up
4713 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4714 * called in the case that we don't need the metadata AND data reservations
4715 * anymore. So if there is an error or we insert an inline extent.
4717 * This function will release the metadata space that was not used and will
4718 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4719 * list if there are no delalloc bytes left.
4721 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4723 btrfs_delalloc_release_metadata(inode, num_bytes);
4724 btrfs_free_reserved_data_space(inode, num_bytes);
4727 static int update_block_group(struct btrfs_trans_handle *trans,
4728 struct btrfs_root *root,
4729 u64 bytenr, u64 num_bytes, int alloc)
4731 struct btrfs_block_group_cache *cache = NULL;
4732 struct btrfs_fs_info *info = root->fs_info;
4733 u64 total = num_bytes;
4738 /* block accounting for super block */
4739 spin_lock(&info->delalloc_lock);
4740 old_val = btrfs_super_bytes_used(info->super_copy);
4742 old_val += num_bytes;
4744 old_val -= num_bytes;
4745 btrfs_set_super_bytes_used(info->super_copy, old_val);
4746 spin_unlock(&info->delalloc_lock);
4749 cache = btrfs_lookup_block_group(info, bytenr);
4752 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4753 BTRFS_BLOCK_GROUP_RAID1 |
4754 BTRFS_BLOCK_GROUP_RAID10))
4759 * If this block group has free space cache written out, we
4760 * need to make sure to load it if we are removing space. This
4761 * is because we need the unpinning stage to actually add the
4762 * space back to the block group, otherwise we will leak space.
4764 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4765 cache_block_group(cache, trans, NULL, 1);
4767 byte_in_group = bytenr - cache->key.objectid;
4768 WARN_ON(byte_in_group > cache->key.offset);
4770 spin_lock(&cache->space_info->lock);
4771 spin_lock(&cache->lock);
4773 if (btrfs_test_opt(root, SPACE_CACHE) &&
4774 cache->disk_cache_state < BTRFS_DC_CLEAR)
4775 cache->disk_cache_state = BTRFS_DC_CLEAR;
4778 old_val = btrfs_block_group_used(&cache->item);
4779 num_bytes = min(total, cache->key.offset - byte_in_group);
4781 old_val += num_bytes;
4782 btrfs_set_block_group_used(&cache->item, old_val);
4783 cache->reserved -= num_bytes;
4784 cache->space_info->bytes_reserved -= num_bytes;
4785 cache->space_info->bytes_used += num_bytes;
4786 cache->space_info->disk_used += num_bytes * factor;
4787 spin_unlock(&cache->lock);
4788 spin_unlock(&cache->space_info->lock);
4790 old_val -= num_bytes;
4791 btrfs_set_block_group_used(&cache->item, old_val);
4792 cache->pinned += num_bytes;
4793 cache->space_info->bytes_pinned += num_bytes;
4794 cache->space_info->bytes_used -= num_bytes;
4795 cache->space_info->disk_used -= num_bytes * factor;
4796 spin_unlock(&cache->lock);
4797 spin_unlock(&cache->space_info->lock);
4799 set_extent_dirty(info->pinned_extents,
4800 bytenr, bytenr + num_bytes - 1,
4801 GFP_NOFS | __GFP_NOFAIL);
4803 btrfs_put_block_group(cache);
4805 bytenr += num_bytes;
4810 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4812 struct btrfs_block_group_cache *cache;
4815 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4819 bytenr = cache->key.objectid;
4820 btrfs_put_block_group(cache);
4825 static int pin_down_extent(struct btrfs_root *root,
4826 struct btrfs_block_group_cache *cache,
4827 u64 bytenr, u64 num_bytes, int reserved)
4829 spin_lock(&cache->space_info->lock);
4830 spin_lock(&cache->lock);
4831 cache->pinned += num_bytes;
4832 cache->space_info->bytes_pinned += num_bytes;
4834 cache->reserved -= num_bytes;
4835 cache->space_info->bytes_reserved -= num_bytes;
4837 spin_unlock(&cache->lock);
4838 spin_unlock(&cache->space_info->lock);
4840 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4841 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4846 * this function must be called within transaction
4848 int btrfs_pin_extent(struct btrfs_root *root,
4849 u64 bytenr, u64 num_bytes, int reserved)
4851 struct btrfs_block_group_cache *cache;
4853 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4854 BUG_ON(!cache); /* Logic error */
4856 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4858 btrfs_put_block_group(cache);
4863 * this function must be called within transaction
4865 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4866 struct btrfs_root *root,
4867 u64 bytenr, u64 num_bytes)
4869 struct btrfs_block_group_cache *cache;
4871 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4872 BUG_ON(!cache); /* Logic error */
4875 * pull in the free space cache (if any) so that our pin
4876 * removes the free space from the cache. We have load_only set
4877 * to one because the slow code to read in the free extents does check
4878 * the pinned extents.
4880 cache_block_group(cache, trans, root, 1);
4882 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4884 /* remove us from the free space cache (if we're there at all) */
4885 btrfs_remove_free_space(cache, bytenr, num_bytes);
4886 btrfs_put_block_group(cache);
4891 * btrfs_update_reserved_bytes - update the block_group and space info counters
4892 * @cache: The cache we are manipulating
4893 * @num_bytes: The number of bytes in question
4894 * @reserve: One of the reservation enums
4896 * This is called by the allocator when it reserves space, or by somebody who is
4897 * freeing space that was never actually used on disk. For example if you
4898 * reserve some space for a new leaf in transaction A and before transaction A
4899 * commits you free that leaf, you call this with reserve set to 0 in order to
4900 * clear the reservation.
4902 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4903 * ENOSPC accounting. For data we handle the reservation through clearing the
4904 * delalloc bits in the io_tree. We have to do this since we could end up
4905 * allocating less disk space for the amount of data we have reserved in the
4906 * case of compression.
4908 * If this is a reservation and the block group has become read only we cannot
4909 * make the reservation and return -EAGAIN, otherwise this function always
4912 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4913 u64 num_bytes, int reserve)
4915 struct btrfs_space_info *space_info = cache->space_info;
4918 spin_lock(&space_info->lock);
4919 spin_lock(&cache->lock);
4920 if (reserve != RESERVE_FREE) {
4924 cache->reserved += num_bytes;
4925 space_info->bytes_reserved += num_bytes;
4926 if (reserve == RESERVE_ALLOC) {
4927 trace_btrfs_space_reservation(cache->fs_info,
4928 "space_info", space_info->flags,
4930 space_info->bytes_may_use -= num_bytes;
4935 space_info->bytes_readonly += num_bytes;
4936 cache->reserved -= num_bytes;
4937 space_info->bytes_reserved -= num_bytes;
4938 space_info->reservation_progress++;
4940 spin_unlock(&cache->lock);
4941 spin_unlock(&space_info->lock);
4945 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4946 struct btrfs_root *root)
4948 struct btrfs_fs_info *fs_info = root->fs_info;
4949 struct btrfs_caching_control *next;
4950 struct btrfs_caching_control *caching_ctl;
4951 struct btrfs_block_group_cache *cache;
4953 down_write(&fs_info->extent_commit_sem);
4955 list_for_each_entry_safe(caching_ctl, next,
4956 &fs_info->caching_block_groups, list) {
4957 cache = caching_ctl->block_group;
4958 if (block_group_cache_done(cache)) {
4959 cache->last_byte_to_unpin = (u64)-1;
4960 list_del_init(&caching_ctl->list);
4961 put_caching_control(caching_ctl);
4963 cache->last_byte_to_unpin = caching_ctl->progress;
4967 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4968 fs_info->pinned_extents = &fs_info->freed_extents[1];
4970 fs_info->pinned_extents = &fs_info->freed_extents[0];
4972 up_write(&fs_info->extent_commit_sem);
4974 update_global_block_rsv(fs_info);
4977 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4979 struct btrfs_fs_info *fs_info = root->fs_info;
4980 struct btrfs_block_group_cache *cache = NULL;
4981 struct btrfs_space_info *space_info;
4982 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4986 while (start <= end) {
4989 start >= cache->key.objectid + cache->key.offset) {
4991 btrfs_put_block_group(cache);
4992 cache = btrfs_lookup_block_group(fs_info, start);
4993 BUG_ON(!cache); /* Logic error */
4996 len = cache->key.objectid + cache->key.offset - start;
4997 len = min(len, end + 1 - start);
4999 if (start < cache->last_byte_to_unpin) {
5000 len = min(len, cache->last_byte_to_unpin - start);
5001 btrfs_add_free_space(cache, start, len);
5005 space_info = cache->space_info;
5007 spin_lock(&space_info->lock);
5008 spin_lock(&cache->lock);
5009 cache->pinned -= len;
5010 space_info->bytes_pinned -= len;
5012 space_info->bytes_readonly += len;
5015 spin_unlock(&cache->lock);
5016 if (!readonly && global_rsv->space_info == space_info) {
5017 spin_lock(&global_rsv->lock);
5018 if (!global_rsv->full) {
5019 len = min(len, global_rsv->size -
5020 global_rsv->reserved);
5021 global_rsv->reserved += len;
5022 space_info->bytes_may_use += len;
5023 if (global_rsv->reserved >= global_rsv->size)
5024 global_rsv->full = 1;
5026 spin_unlock(&global_rsv->lock);
5028 spin_unlock(&space_info->lock);
5032 btrfs_put_block_group(cache);
5036 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5037 struct btrfs_root *root)
5039 struct btrfs_fs_info *fs_info = root->fs_info;
5040 struct extent_io_tree *unpin;
5048 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5049 unpin = &fs_info->freed_extents[1];
5051 unpin = &fs_info->freed_extents[0];
5054 ret = find_first_extent_bit(unpin, 0, &start, &end,
5055 EXTENT_DIRTY, NULL);
5059 if (btrfs_test_opt(root, DISCARD))
5060 ret = btrfs_discard_extent(root, start,
5061 end + 1 - start, NULL);
5063 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5064 unpin_extent_range(root, start, end);
5071 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5072 struct btrfs_root *root,
5073 u64 bytenr, u64 num_bytes, u64 parent,
5074 u64 root_objectid, u64 owner_objectid,
5075 u64 owner_offset, int refs_to_drop,
5076 struct btrfs_delayed_extent_op *extent_op)
5078 struct btrfs_key key;
5079 struct btrfs_path *path;
5080 struct btrfs_fs_info *info = root->fs_info;
5081 struct btrfs_root *extent_root = info->extent_root;
5082 struct extent_buffer *leaf;
5083 struct btrfs_extent_item *ei;
5084 struct btrfs_extent_inline_ref *iref;
5087 int extent_slot = 0;
5088 int found_extent = 0;
5093 path = btrfs_alloc_path();
5098 path->leave_spinning = 1;
5100 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5101 BUG_ON(!is_data && refs_to_drop != 1);
5103 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5104 bytenr, num_bytes, parent,
5105 root_objectid, owner_objectid,
5108 extent_slot = path->slots[0];
5109 while (extent_slot >= 0) {
5110 btrfs_item_key_to_cpu(path->nodes[0], &key,
5112 if (key.objectid != bytenr)
5114 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5115 key.offset == num_bytes) {
5119 if (path->slots[0] - extent_slot > 5)
5123 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5124 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5125 if (found_extent && item_size < sizeof(*ei))
5128 if (!found_extent) {
5130 ret = remove_extent_backref(trans, extent_root, path,
5134 btrfs_abort_transaction(trans, extent_root, ret);
5137 btrfs_release_path(path);
5138 path->leave_spinning = 1;
5140 key.objectid = bytenr;
5141 key.type = BTRFS_EXTENT_ITEM_KEY;
5142 key.offset = num_bytes;
5144 ret = btrfs_search_slot(trans, extent_root,
5147 printk(KERN_ERR "umm, got %d back from search"
5148 ", was looking for %llu\n", ret,
5149 (unsigned long long)bytenr);
5151 btrfs_print_leaf(extent_root,
5155 btrfs_abort_transaction(trans, extent_root, ret);
5158 extent_slot = path->slots[0];
5160 } else if (ret == -ENOENT) {
5161 btrfs_print_leaf(extent_root, path->nodes[0]);
5163 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5164 "parent %llu root %llu owner %llu offset %llu\n",
5165 (unsigned long long)bytenr,
5166 (unsigned long long)parent,
5167 (unsigned long long)root_objectid,
5168 (unsigned long long)owner_objectid,
5169 (unsigned long long)owner_offset);
5171 btrfs_abort_transaction(trans, extent_root, ret);
5175 leaf = path->nodes[0];
5176 item_size = btrfs_item_size_nr(leaf, extent_slot);
5177 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5178 if (item_size < sizeof(*ei)) {
5179 BUG_ON(found_extent || extent_slot != path->slots[0]);
5180 ret = convert_extent_item_v0(trans, extent_root, path,
5183 btrfs_abort_transaction(trans, extent_root, ret);
5187 btrfs_release_path(path);
5188 path->leave_spinning = 1;
5190 key.objectid = bytenr;
5191 key.type = BTRFS_EXTENT_ITEM_KEY;
5192 key.offset = num_bytes;
5194 ret = btrfs_search_slot(trans, extent_root, &key, path,
5197 printk(KERN_ERR "umm, got %d back from search"
5198 ", was looking for %llu\n", ret,
5199 (unsigned long long)bytenr);
5200 btrfs_print_leaf(extent_root, path->nodes[0]);
5203 btrfs_abort_transaction(trans, extent_root, ret);
5207 extent_slot = path->slots[0];
5208 leaf = path->nodes[0];
5209 item_size = btrfs_item_size_nr(leaf, extent_slot);
5212 BUG_ON(item_size < sizeof(*ei));
5213 ei = btrfs_item_ptr(leaf, extent_slot,
5214 struct btrfs_extent_item);
5215 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5216 struct btrfs_tree_block_info *bi;
5217 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5218 bi = (struct btrfs_tree_block_info *)(ei + 1);
5219 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5222 refs = btrfs_extent_refs(leaf, ei);
5223 BUG_ON(refs < refs_to_drop);
5224 refs -= refs_to_drop;
5228 __run_delayed_extent_op(extent_op, leaf, ei);
5230 * In the case of inline back ref, reference count will
5231 * be updated by remove_extent_backref
5234 BUG_ON(!found_extent);
5236 btrfs_set_extent_refs(leaf, ei, refs);
5237 btrfs_mark_buffer_dirty(leaf);
5240 ret = remove_extent_backref(trans, extent_root, path,
5244 btrfs_abort_transaction(trans, extent_root, ret);
5250 BUG_ON(is_data && refs_to_drop !=
5251 extent_data_ref_count(root, path, iref));
5253 BUG_ON(path->slots[0] != extent_slot);
5255 BUG_ON(path->slots[0] != extent_slot + 1);
5256 path->slots[0] = extent_slot;
5261 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5264 btrfs_abort_transaction(trans, extent_root, ret);
5267 btrfs_release_path(path);
5270 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5272 btrfs_abort_transaction(trans, extent_root, ret);
5277 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5279 btrfs_abort_transaction(trans, extent_root, ret);
5284 btrfs_free_path(path);
5289 * when we free an block, it is possible (and likely) that we free the last
5290 * delayed ref for that extent as well. This searches the delayed ref tree for
5291 * a given extent, and if there are no other delayed refs to be processed, it
5292 * removes it from the tree.
5294 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5295 struct btrfs_root *root, u64 bytenr)
5297 struct btrfs_delayed_ref_head *head;
5298 struct btrfs_delayed_ref_root *delayed_refs;
5299 struct btrfs_delayed_ref_node *ref;
5300 struct rb_node *node;
5303 delayed_refs = &trans->transaction->delayed_refs;
5304 spin_lock(&delayed_refs->lock);
5305 head = btrfs_find_delayed_ref_head(trans, bytenr);
5309 node = rb_prev(&head->node.rb_node);
5313 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5315 /* there are still entries for this ref, we can't drop it */
5316 if (ref->bytenr == bytenr)
5319 if (head->extent_op) {
5320 if (!head->must_insert_reserved)
5322 kfree(head->extent_op);
5323 head->extent_op = NULL;
5327 * waiting for the lock here would deadlock. If someone else has it
5328 * locked they are already in the process of dropping it anyway
5330 if (!mutex_trylock(&head->mutex))
5334 * at this point we have a head with no other entries. Go
5335 * ahead and process it.
5337 head->node.in_tree = 0;
5338 rb_erase(&head->node.rb_node, &delayed_refs->root);
5340 delayed_refs->num_entries--;
5343 * we don't take a ref on the node because we're removing it from the
5344 * tree, so we just steal the ref the tree was holding.
5346 delayed_refs->num_heads--;
5347 if (list_empty(&head->cluster))
5348 delayed_refs->num_heads_ready--;
5350 list_del_init(&head->cluster);
5351 spin_unlock(&delayed_refs->lock);
5353 BUG_ON(head->extent_op);
5354 if (head->must_insert_reserved)
5357 mutex_unlock(&head->mutex);
5358 btrfs_put_delayed_ref(&head->node);
5361 spin_unlock(&delayed_refs->lock);
5365 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5366 struct btrfs_root *root,
5367 struct extent_buffer *buf,
5368 u64 parent, int last_ref)
5370 struct btrfs_block_group_cache *cache = NULL;
5373 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5374 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5375 buf->start, buf->len,
5376 parent, root->root_key.objectid,
5377 btrfs_header_level(buf),
5378 BTRFS_DROP_DELAYED_REF, NULL, 0);
5379 BUG_ON(ret); /* -ENOMEM */
5385 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5387 if (btrfs_header_generation(buf) == trans->transid) {
5388 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5389 ret = check_ref_cleanup(trans, root, buf->start);
5394 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5395 pin_down_extent(root, cache, buf->start, buf->len, 1);
5399 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5401 btrfs_add_free_space(cache, buf->start, buf->len);
5402 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5406 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5409 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5410 btrfs_put_block_group(cache);
5413 /* Can return -ENOMEM */
5414 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5415 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5416 u64 owner, u64 offset, int for_cow)
5419 struct btrfs_fs_info *fs_info = root->fs_info;
5422 * tree log blocks never actually go into the extent allocation
5423 * tree, just update pinning info and exit early.
5425 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5426 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5427 /* unlocks the pinned mutex */
5428 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5430 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5431 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5433 parent, root_objectid, (int)owner,
5434 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5436 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5438 parent, root_objectid, owner,
5439 offset, BTRFS_DROP_DELAYED_REF,
5445 static u64 stripe_align(struct btrfs_root *root, u64 val)
5447 u64 mask = ((u64)root->stripesize - 1);
5448 u64 ret = (val + mask) & ~mask;
5453 * when we wait for progress in the block group caching, its because
5454 * our allocation attempt failed at least once. So, we must sleep
5455 * and let some progress happen before we try again.
5457 * This function will sleep at least once waiting for new free space to
5458 * show up, and then it will check the block group free space numbers
5459 * for our min num_bytes. Another option is to have it go ahead
5460 * and look in the rbtree for a free extent of a given size, but this
5464 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5467 struct btrfs_caching_control *caching_ctl;
5470 caching_ctl = get_caching_control(cache);
5474 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5475 (cache->free_space_ctl->free_space >= num_bytes));
5477 put_caching_control(caching_ctl);
5482 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5484 struct btrfs_caching_control *caching_ctl;
5487 caching_ctl = get_caching_control(cache);
5491 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5493 put_caching_control(caching_ctl);
5497 int __get_raid_index(u64 flags)
5501 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5503 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5505 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5507 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5515 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5517 return __get_raid_index(cache->flags);
5520 enum btrfs_loop_type {
5521 LOOP_CACHING_NOWAIT = 0,
5522 LOOP_CACHING_WAIT = 1,
5523 LOOP_ALLOC_CHUNK = 2,
5524 LOOP_NO_EMPTY_SIZE = 3,
5528 * walks the btree of allocated extents and find a hole of a given size.
5529 * The key ins is changed to record the hole:
5530 * ins->objectid == block start
5531 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5532 * ins->offset == number of blocks
5533 * Any available blocks before search_start are skipped.
5535 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5536 struct btrfs_root *orig_root,
5537 u64 num_bytes, u64 empty_size,
5538 u64 hint_byte, struct btrfs_key *ins,
5542 struct btrfs_root *root = orig_root->fs_info->extent_root;
5543 struct btrfs_free_cluster *last_ptr = NULL;
5544 struct btrfs_block_group_cache *block_group = NULL;
5545 struct btrfs_block_group_cache *used_block_group;
5546 u64 search_start = 0;
5547 int empty_cluster = 2 * 1024 * 1024;
5548 struct btrfs_space_info *space_info;
5551 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5552 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5553 bool found_uncached_bg = false;
5554 bool failed_cluster_refill = false;
5555 bool failed_alloc = false;
5556 bool use_cluster = true;
5557 bool have_caching_bg = false;
5559 WARN_ON(num_bytes < root->sectorsize);
5560 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5564 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5566 space_info = __find_space_info(root->fs_info, data);
5568 printk(KERN_ERR "No space info for %llu\n", data);
5573 * If the space info is for both data and metadata it means we have a
5574 * small filesystem and we can't use the clustering stuff.
5576 if (btrfs_mixed_space_info(space_info))
5577 use_cluster = false;
5579 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5580 last_ptr = &root->fs_info->meta_alloc_cluster;
5581 if (!btrfs_test_opt(root, SSD))
5582 empty_cluster = 64 * 1024;
5585 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5586 btrfs_test_opt(root, SSD)) {
5587 last_ptr = &root->fs_info->data_alloc_cluster;
5591 spin_lock(&last_ptr->lock);
5592 if (last_ptr->block_group)
5593 hint_byte = last_ptr->window_start;
5594 spin_unlock(&last_ptr->lock);
5597 search_start = max(search_start, first_logical_byte(root, 0));
5598 search_start = max(search_start, hint_byte);
5603 if (search_start == hint_byte) {
5604 block_group = btrfs_lookup_block_group(root->fs_info,
5606 used_block_group = block_group;
5608 * we don't want to use the block group if it doesn't match our
5609 * allocation bits, or if its not cached.
5611 * However if we are re-searching with an ideal block group
5612 * picked out then we don't care that the block group is cached.
5614 if (block_group && block_group_bits(block_group, data) &&
5615 block_group->cached != BTRFS_CACHE_NO) {
5616 down_read(&space_info->groups_sem);
5617 if (list_empty(&block_group->list) ||
5620 * someone is removing this block group,
5621 * we can't jump into the have_block_group
5622 * target because our list pointers are not
5625 btrfs_put_block_group(block_group);
5626 up_read(&space_info->groups_sem);
5628 index = get_block_group_index(block_group);
5629 goto have_block_group;
5631 } else if (block_group) {
5632 btrfs_put_block_group(block_group);
5636 have_caching_bg = false;
5637 down_read(&space_info->groups_sem);
5638 list_for_each_entry(block_group, &space_info->block_groups[index],
5643 used_block_group = block_group;
5644 btrfs_get_block_group(block_group);
5645 search_start = block_group->key.objectid;
5648 * this can happen if we end up cycling through all the
5649 * raid types, but we want to make sure we only allocate
5650 * for the proper type.
5652 if (!block_group_bits(block_group, data)) {
5653 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5654 BTRFS_BLOCK_GROUP_RAID1 |
5655 BTRFS_BLOCK_GROUP_RAID10;
5658 * if they asked for extra copies and this block group
5659 * doesn't provide them, bail. This does allow us to
5660 * fill raid0 from raid1.
5662 if ((data & extra) && !(block_group->flags & extra))
5667 cached = block_group_cache_done(block_group);
5668 if (unlikely(!cached)) {
5669 found_uncached_bg = true;
5670 ret = cache_block_group(block_group, trans,
5676 if (unlikely(block_group->ro))
5680 * Ok we want to try and use the cluster allocator, so
5685 * the refill lock keeps out other
5686 * people trying to start a new cluster
5688 spin_lock(&last_ptr->refill_lock);
5689 used_block_group = last_ptr->block_group;
5690 if (used_block_group != block_group &&
5691 (!used_block_group ||
5692 used_block_group->ro ||
5693 !block_group_bits(used_block_group, data))) {
5694 used_block_group = block_group;
5695 goto refill_cluster;
5698 if (used_block_group != block_group)
5699 btrfs_get_block_group(used_block_group);
5701 offset = btrfs_alloc_from_cluster(used_block_group,
5702 last_ptr, num_bytes, used_block_group->key.objectid);
5704 /* we have a block, we're done */
5705 spin_unlock(&last_ptr->refill_lock);
5706 trace_btrfs_reserve_extent_cluster(root,
5707 block_group, search_start, num_bytes);
5711 WARN_ON(last_ptr->block_group != used_block_group);
5712 if (used_block_group != block_group) {
5713 btrfs_put_block_group(used_block_group);
5714 used_block_group = block_group;
5717 BUG_ON(used_block_group != block_group);
5718 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5719 * set up a new clusters, so lets just skip it
5720 * and let the allocator find whatever block
5721 * it can find. If we reach this point, we
5722 * will have tried the cluster allocator
5723 * plenty of times and not have found
5724 * anything, so we are likely way too
5725 * fragmented for the clustering stuff to find
5728 * However, if the cluster is taken from the
5729 * current block group, release the cluster
5730 * first, so that we stand a better chance of
5731 * succeeding in the unclustered
5733 if (loop >= LOOP_NO_EMPTY_SIZE &&
5734 last_ptr->block_group != block_group) {
5735 spin_unlock(&last_ptr->refill_lock);
5736 goto unclustered_alloc;
5740 * this cluster didn't work out, free it and
5743 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5745 if (loop >= LOOP_NO_EMPTY_SIZE) {
5746 spin_unlock(&last_ptr->refill_lock);
5747 goto unclustered_alloc;
5750 /* allocate a cluster in this block group */
5751 ret = btrfs_find_space_cluster(trans, root,
5752 block_group, last_ptr,
5753 search_start, num_bytes,
5754 empty_cluster + empty_size);
5757 * now pull our allocation out of this
5760 offset = btrfs_alloc_from_cluster(block_group,
5761 last_ptr, num_bytes,
5764 /* we found one, proceed */
5765 spin_unlock(&last_ptr->refill_lock);
5766 trace_btrfs_reserve_extent_cluster(root,
5767 block_group, search_start,
5771 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5772 && !failed_cluster_refill) {
5773 spin_unlock(&last_ptr->refill_lock);
5775 failed_cluster_refill = true;
5776 wait_block_group_cache_progress(block_group,
5777 num_bytes + empty_cluster + empty_size);
5778 goto have_block_group;
5782 * at this point we either didn't find a cluster
5783 * or we weren't able to allocate a block from our
5784 * cluster. Free the cluster we've been trying
5785 * to use, and go to the next block group
5787 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5788 spin_unlock(&last_ptr->refill_lock);
5793 spin_lock(&block_group->free_space_ctl->tree_lock);
5795 block_group->free_space_ctl->free_space <
5796 num_bytes + empty_cluster + empty_size) {
5797 spin_unlock(&block_group->free_space_ctl->tree_lock);
5800 spin_unlock(&block_group->free_space_ctl->tree_lock);
5802 offset = btrfs_find_space_for_alloc(block_group, search_start,
5803 num_bytes, empty_size);
5805 * If we didn't find a chunk, and we haven't failed on this
5806 * block group before, and this block group is in the middle of
5807 * caching and we are ok with waiting, then go ahead and wait
5808 * for progress to be made, and set failed_alloc to true.
5810 * If failed_alloc is true then we've already waited on this
5811 * block group once and should move on to the next block group.
5813 if (!offset && !failed_alloc && !cached &&
5814 loop > LOOP_CACHING_NOWAIT) {
5815 wait_block_group_cache_progress(block_group,
5816 num_bytes + empty_size);
5817 failed_alloc = true;
5818 goto have_block_group;
5819 } else if (!offset) {
5821 have_caching_bg = true;
5825 search_start = stripe_align(root, offset);
5827 /* move on to the next group */
5828 if (search_start + num_bytes >
5829 used_block_group->key.objectid + used_block_group->key.offset) {
5830 btrfs_add_free_space(used_block_group, offset, num_bytes);
5834 if (offset < search_start)
5835 btrfs_add_free_space(used_block_group, offset,
5836 search_start - offset);
5837 BUG_ON(offset > search_start);
5839 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5841 if (ret == -EAGAIN) {
5842 btrfs_add_free_space(used_block_group, offset, num_bytes);
5846 /* we are all good, lets return */
5847 ins->objectid = search_start;
5848 ins->offset = num_bytes;
5850 trace_btrfs_reserve_extent(orig_root, block_group,
5851 search_start, num_bytes);
5852 if (used_block_group != block_group)
5853 btrfs_put_block_group(used_block_group);
5854 btrfs_put_block_group(block_group);
5857 failed_cluster_refill = false;
5858 failed_alloc = false;
5859 BUG_ON(index != get_block_group_index(block_group));
5860 if (used_block_group != block_group)
5861 btrfs_put_block_group(used_block_group);
5862 btrfs_put_block_group(block_group);
5864 up_read(&space_info->groups_sem);
5866 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5869 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5873 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5874 * caching kthreads as we move along
5875 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5876 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5877 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5880 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5883 if (loop == LOOP_ALLOC_CHUNK) {
5884 ret = do_chunk_alloc(trans, root, data,
5887 * Do not bail out on ENOSPC since we
5888 * can do more things.
5890 if (ret < 0 && ret != -ENOSPC) {
5891 btrfs_abort_transaction(trans,
5897 if (loop == LOOP_NO_EMPTY_SIZE) {
5903 } else if (!ins->objectid) {
5905 } else if (ins->objectid) {
5913 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5914 int dump_block_groups)
5916 struct btrfs_block_group_cache *cache;
5919 spin_lock(&info->lock);
5920 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5921 (unsigned long long)info->flags,
5922 (unsigned long long)(info->total_bytes - info->bytes_used -
5923 info->bytes_pinned - info->bytes_reserved -
5924 info->bytes_readonly),
5925 (info->full) ? "" : "not ");
5926 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5927 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5928 (unsigned long long)info->total_bytes,
5929 (unsigned long long)info->bytes_used,
5930 (unsigned long long)info->bytes_pinned,
5931 (unsigned long long)info->bytes_reserved,
5932 (unsigned long long)info->bytes_may_use,
5933 (unsigned long long)info->bytes_readonly);
5934 spin_unlock(&info->lock);
5936 if (!dump_block_groups)
5939 down_read(&info->groups_sem);
5941 list_for_each_entry(cache, &info->block_groups[index], list) {
5942 spin_lock(&cache->lock);
5943 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5944 (unsigned long long)cache->key.objectid,
5945 (unsigned long long)cache->key.offset,
5946 (unsigned long long)btrfs_block_group_used(&cache->item),
5947 (unsigned long long)cache->pinned,
5948 (unsigned long long)cache->reserved,
5949 cache->ro ? "[readonly]" : "");
5950 btrfs_dump_free_space(cache, bytes);
5951 spin_unlock(&cache->lock);
5953 if (++index < BTRFS_NR_RAID_TYPES)
5955 up_read(&info->groups_sem);
5958 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5959 struct btrfs_root *root,
5960 u64 num_bytes, u64 min_alloc_size,
5961 u64 empty_size, u64 hint_byte,
5962 struct btrfs_key *ins, u64 data)
5964 bool final_tried = false;
5967 data = btrfs_get_alloc_profile(root, data);
5969 WARN_ON(num_bytes < root->sectorsize);
5970 ret = find_free_extent(trans, root, num_bytes, empty_size,
5971 hint_byte, ins, data);
5973 if (ret == -ENOSPC) {
5975 num_bytes = num_bytes >> 1;
5976 num_bytes = num_bytes & ~(root->sectorsize - 1);
5977 num_bytes = max(num_bytes, min_alloc_size);
5978 if (num_bytes == min_alloc_size)
5981 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5982 struct btrfs_space_info *sinfo;
5984 sinfo = __find_space_info(root->fs_info, data);
5985 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5986 "wanted %llu\n", (unsigned long long)data,
5987 (unsigned long long)num_bytes);
5989 dump_space_info(sinfo, num_bytes, 1);
5993 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5998 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5999 u64 start, u64 len, int pin)
6001 struct btrfs_block_group_cache *cache;
6004 cache = btrfs_lookup_block_group(root->fs_info, start);
6006 printk(KERN_ERR "Unable to find block group for %llu\n",
6007 (unsigned long long)start);
6011 if (btrfs_test_opt(root, DISCARD))
6012 ret = btrfs_discard_extent(root, start, len, NULL);
6015 pin_down_extent(root, cache, start, len, 1);
6017 btrfs_add_free_space(cache, start, len);
6018 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6020 btrfs_put_block_group(cache);
6022 trace_btrfs_reserved_extent_free(root, start, len);
6027 int btrfs_free_reserved_extent(struct btrfs_root *root,
6030 return __btrfs_free_reserved_extent(root, start, len, 0);
6033 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6036 return __btrfs_free_reserved_extent(root, start, len, 1);
6039 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6040 struct btrfs_root *root,
6041 u64 parent, u64 root_objectid,
6042 u64 flags, u64 owner, u64 offset,
6043 struct btrfs_key *ins, int ref_mod)
6046 struct btrfs_fs_info *fs_info = root->fs_info;
6047 struct btrfs_extent_item *extent_item;
6048 struct btrfs_extent_inline_ref *iref;
6049 struct btrfs_path *path;
6050 struct extent_buffer *leaf;
6055 type = BTRFS_SHARED_DATA_REF_KEY;
6057 type = BTRFS_EXTENT_DATA_REF_KEY;
6059 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6061 path = btrfs_alloc_path();
6065 path->leave_spinning = 1;
6066 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6069 btrfs_free_path(path);
6073 leaf = path->nodes[0];
6074 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6075 struct btrfs_extent_item);
6076 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6077 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6078 btrfs_set_extent_flags(leaf, extent_item,
6079 flags | BTRFS_EXTENT_FLAG_DATA);
6081 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6082 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6084 struct btrfs_shared_data_ref *ref;
6085 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6086 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6087 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6089 struct btrfs_extent_data_ref *ref;
6090 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6091 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6092 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6093 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6094 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6097 btrfs_mark_buffer_dirty(path->nodes[0]);
6098 btrfs_free_path(path);
6100 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6101 if (ret) { /* -ENOENT, logic error */
6102 printk(KERN_ERR "btrfs update block group failed for %llu "
6103 "%llu\n", (unsigned long long)ins->objectid,
6104 (unsigned long long)ins->offset);
6110 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6111 struct btrfs_root *root,
6112 u64 parent, u64 root_objectid,
6113 u64 flags, struct btrfs_disk_key *key,
6114 int level, struct btrfs_key *ins)
6117 struct btrfs_fs_info *fs_info = root->fs_info;
6118 struct btrfs_extent_item *extent_item;
6119 struct btrfs_tree_block_info *block_info;
6120 struct btrfs_extent_inline_ref *iref;
6121 struct btrfs_path *path;
6122 struct extent_buffer *leaf;
6123 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6125 path = btrfs_alloc_path();
6129 path->leave_spinning = 1;
6130 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6133 btrfs_free_path(path);
6137 leaf = path->nodes[0];
6138 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6139 struct btrfs_extent_item);
6140 btrfs_set_extent_refs(leaf, extent_item, 1);
6141 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6142 btrfs_set_extent_flags(leaf, extent_item,
6143 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6144 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6146 btrfs_set_tree_block_key(leaf, block_info, key);
6147 btrfs_set_tree_block_level(leaf, block_info, level);
6149 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6151 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6152 btrfs_set_extent_inline_ref_type(leaf, iref,
6153 BTRFS_SHARED_BLOCK_REF_KEY);
6154 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6156 btrfs_set_extent_inline_ref_type(leaf, iref,
6157 BTRFS_TREE_BLOCK_REF_KEY);
6158 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6161 btrfs_mark_buffer_dirty(leaf);
6162 btrfs_free_path(path);
6164 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6165 if (ret) { /* -ENOENT, logic error */
6166 printk(KERN_ERR "btrfs update block group failed for %llu "
6167 "%llu\n", (unsigned long long)ins->objectid,
6168 (unsigned long long)ins->offset);
6174 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6175 struct btrfs_root *root,
6176 u64 root_objectid, u64 owner,
6177 u64 offset, struct btrfs_key *ins)
6181 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6183 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6185 root_objectid, owner, offset,
6186 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6191 * this is used by the tree logging recovery code. It records that
6192 * an extent has been allocated and makes sure to clear the free
6193 * space cache bits as well
6195 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6196 struct btrfs_root *root,
6197 u64 root_objectid, u64 owner, u64 offset,
6198 struct btrfs_key *ins)
6201 struct btrfs_block_group_cache *block_group;
6202 struct btrfs_caching_control *caching_ctl;
6203 u64 start = ins->objectid;
6204 u64 num_bytes = ins->offset;
6206 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6207 cache_block_group(block_group, trans, NULL, 0);
6208 caching_ctl = get_caching_control(block_group);
6211 BUG_ON(!block_group_cache_done(block_group));
6212 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6213 BUG_ON(ret); /* -ENOMEM */
6215 mutex_lock(&caching_ctl->mutex);
6217 if (start >= caching_ctl->progress) {
6218 ret = add_excluded_extent(root, start, num_bytes);
6219 BUG_ON(ret); /* -ENOMEM */
6220 } else if (start + num_bytes <= caching_ctl->progress) {
6221 ret = btrfs_remove_free_space(block_group,
6223 BUG_ON(ret); /* -ENOMEM */
6225 num_bytes = caching_ctl->progress - start;
6226 ret = btrfs_remove_free_space(block_group,
6228 BUG_ON(ret); /* -ENOMEM */
6230 start = caching_ctl->progress;
6231 num_bytes = ins->objectid + ins->offset -
6232 caching_ctl->progress;
6233 ret = add_excluded_extent(root, start, num_bytes);
6234 BUG_ON(ret); /* -ENOMEM */
6237 mutex_unlock(&caching_ctl->mutex);
6238 put_caching_control(caching_ctl);
6241 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6242 RESERVE_ALLOC_NO_ACCOUNT);
6243 BUG_ON(ret); /* logic error */
6244 btrfs_put_block_group(block_group);
6245 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6246 0, owner, offset, ins, 1);
6250 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6251 struct btrfs_root *root,
6252 u64 bytenr, u32 blocksize,
6255 struct extent_buffer *buf;
6257 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6259 return ERR_PTR(-ENOMEM);
6260 btrfs_set_header_generation(buf, trans->transid);
6261 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6262 btrfs_tree_lock(buf);
6263 clean_tree_block(trans, root, buf);
6264 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6266 btrfs_set_lock_blocking(buf);
6267 btrfs_set_buffer_uptodate(buf);
6269 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6271 * we allow two log transactions at a time, use different
6272 * EXENT bit to differentiate dirty pages.
6274 if (root->log_transid % 2 == 0)
6275 set_extent_dirty(&root->dirty_log_pages, buf->start,
6276 buf->start + buf->len - 1, GFP_NOFS);
6278 set_extent_new(&root->dirty_log_pages, buf->start,
6279 buf->start + buf->len - 1, GFP_NOFS);
6281 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6282 buf->start + buf->len - 1, GFP_NOFS);
6284 trans->blocks_used++;
6285 /* this returns a buffer locked for blocking */
6289 static struct btrfs_block_rsv *
6290 use_block_rsv(struct btrfs_trans_handle *trans,
6291 struct btrfs_root *root, u32 blocksize)
6293 struct btrfs_block_rsv *block_rsv;
6294 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6297 block_rsv = get_block_rsv(trans, root);
6299 if (block_rsv->size == 0) {
6300 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6301 BTRFS_RESERVE_NO_FLUSH);
6303 * If we couldn't reserve metadata bytes try and use some from
6304 * the global reserve.
6306 if (ret && block_rsv != global_rsv) {
6307 ret = block_rsv_use_bytes(global_rsv, blocksize);
6310 return ERR_PTR(ret);
6312 return ERR_PTR(ret);
6317 ret = block_rsv_use_bytes(block_rsv, blocksize);
6320 if (ret && !block_rsv->failfast) {
6321 static DEFINE_RATELIMIT_STATE(_rs,
6322 DEFAULT_RATELIMIT_INTERVAL,
6323 /*DEFAULT_RATELIMIT_BURST*/ 2);
6324 if (__ratelimit(&_rs))
6325 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6327 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6328 BTRFS_RESERVE_NO_FLUSH);
6331 } else if (ret && block_rsv != global_rsv) {
6332 ret = block_rsv_use_bytes(global_rsv, blocksize);
6338 return ERR_PTR(-ENOSPC);
6341 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6342 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6344 block_rsv_add_bytes(block_rsv, blocksize, 0);
6345 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6349 * finds a free extent and does all the dirty work required for allocation
6350 * returns the key for the extent through ins, and a tree buffer for
6351 * the first block of the extent through buf.
6353 * returns the tree buffer or NULL.
6355 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6356 struct btrfs_root *root, u32 blocksize,
6357 u64 parent, u64 root_objectid,
6358 struct btrfs_disk_key *key, int level,
6359 u64 hint, u64 empty_size)
6361 struct btrfs_key ins;
6362 struct btrfs_block_rsv *block_rsv;
6363 struct extent_buffer *buf;
6368 block_rsv = use_block_rsv(trans, root, blocksize);
6369 if (IS_ERR(block_rsv))
6370 return ERR_CAST(block_rsv);
6372 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6373 empty_size, hint, &ins, 0);
6375 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6376 return ERR_PTR(ret);
6379 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6381 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6383 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6385 parent = ins.objectid;
6386 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6390 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6391 struct btrfs_delayed_extent_op *extent_op;
6392 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6393 BUG_ON(!extent_op); /* -ENOMEM */
6395 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6397 memset(&extent_op->key, 0, sizeof(extent_op->key));
6398 extent_op->flags_to_set = flags;
6399 extent_op->update_key = 1;
6400 extent_op->update_flags = 1;
6401 extent_op->is_data = 0;
6403 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6405 ins.offset, parent, root_objectid,
6406 level, BTRFS_ADD_DELAYED_EXTENT,
6408 BUG_ON(ret); /* -ENOMEM */
6413 struct walk_control {
6414 u64 refs[BTRFS_MAX_LEVEL];
6415 u64 flags[BTRFS_MAX_LEVEL];
6416 struct btrfs_key update_progress;
6427 #define DROP_REFERENCE 1
6428 #define UPDATE_BACKREF 2
6430 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6431 struct btrfs_root *root,
6432 struct walk_control *wc,
6433 struct btrfs_path *path)
6441 struct btrfs_key key;
6442 struct extent_buffer *eb;
6447 if (path->slots[wc->level] < wc->reada_slot) {
6448 wc->reada_count = wc->reada_count * 2 / 3;
6449 wc->reada_count = max(wc->reada_count, 2);
6451 wc->reada_count = wc->reada_count * 3 / 2;
6452 wc->reada_count = min_t(int, wc->reada_count,
6453 BTRFS_NODEPTRS_PER_BLOCK(root));
6456 eb = path->nodes[wc->level];
6457 nritems = btrfs_header_nritems(eb);
6458 blocksize = btrfs_level_size(root, wc->level - 1);
6460 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6461 if (nread >= wc->reada_count)
6465 bytenr = btrfs_node_blockptr(eb, slot);
6466 generation = btrfs_node_ptr_generation(eb, slot);
6468 if (slot == path->slots[wc->level])
6471 if (wc->stage == UPDATE_BACKREF &&
6472 generation <= root->root_key.offset)
6475 /* We don't lock the tree block, it's OK to be racy here */
6476 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6478 /* We don't care about errors in readahead. */
6483 if (wc->stage == DROP_REFERENCE) {
6487 if (wc->level == 1 &&
6488 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6490 if (!wc->update_ref ||
6491 generation <= root->root_key.offset)
6493 btrfs_node_key_to_cpu(eb, &key, slot);
6494 ret = btrfs_comp_cpu_keys(&key,
6495 &wc->update_progress);
6499 if (wc->level == 1 &&
6500 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6504 ret = readahead_tree_block(root, bytenr, blocksize,
6510 wc->reada_slot = slot;
6514 * hepler to process tree block while walking down the tree.
6516 * when wc->stage == UPDATE_BACKREF, this function updates
6517 * back refs for pointers in the block.
6519 * NOTE: return value 1 means we should stop walking down.
6521 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6522 struct btrfs_root *root,
6523 struct btrfs_path *path,
6524 struct walk_control *wc, int lookup_info)
6526 int level = wc->level;
6527 struct extent_buffer *eb = path->nodes[level];
6528 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6531 if (wc->stage == UPDATE_BACKREF &&
6532 btrfs_header_owner(eb) != root->root_key.objectid)
6536 * when reference count of tree block is 1, it won't increase
6537 * again. once full backref flag is set, we never clear it.
6540 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6541 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6542 BUG_ON(!path->locks[level]);
6543 ret = btrfs_lookup_extent_info(trans, root,
6547 BUG_ON(ret == -ENOMEM);
6550 BUG_ON(wc->refs[level] == 0);
6553 if (wc->stage == DROP_REFERENCE) {
6554 if (wc->refs[level] > 1)
6557 if (path->locks[level] && !wc->keep_locks) {
6558 btrfs_tree_unlock_rw(eb, path->locks[level]);
6559 path->locks[level] = 0;
6564 /* wc->stage == UPDATE_BACKREF */
6565 if (!(wc->flags[level] & flag)) {
6566 BUG_ON(!path->locks[level]);
6567 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6568 BUG_ON(ret); /* -ENOMEM */
6569 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6570 BUG_ON(ret); /* -ENOMEM */
6571 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6573 BUG_ON(ret); /* -ENOMEM */
6574 wc->flags[level] |= flag;
6578 * the block is shared by multiple trees, so it's not good to
6579 * keep the tree lock
6581 if (path->locks[level] && level > 0) {
6582 btrfs_tree_unlock_rw(eb, path->locks[level]);
6583 path->locks[level] = 0;
6589 * hepler to process tree block pointer.
6591 * when wc->stage == DROP_REFERENCE, this function checks
6592 * reference count of the block pointed to. if the block
6593 * is shared and we need update back refs for the subtree
6594 * rooted at the block, this function changes wc->stage to
6595 * UPDATE_BACKREF. if the block is shared and there is no
6596 * need to update back, this function drops the reference
6599 * NOTE: return value 1 means we should stop walking down.
6601 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6602 struct btrfs_root *root,
6603 struct btrfs_path *path,
6604 struct walk_control *wc, int *lookup_info)
6610 struct btrfs_key key;
6611 struct extent_buffer *next;
6612 int level = wc->level;
6616 generation = btrfs_node_ptr_generation(path->nodes[level],
6617 path->slots[level]);
6619 * if the lower level block was created before the snapshot
6620 * was created, we know there is no need to update back refs
6623 if (wc->stage == UPDATE_BACKREF &&
6624 generation <= root->root_key.offset) {
6629 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6630 blocksize = btrfs_level_size(root, level - 1);
6632 next = btrfs_find_tree_block(root, bytenr, blocksize);
6634 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6639 btrfs_tree_lock(next);
6640 btrfs_set_lock_blocking(next);
6642 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6643 &wc->refs[level - 1],
6644 &wc->flags[level - 1]);
6646 btrfs_tree_unlock(next);
6650 BUG_ON(wc->refs[level - 1] == 0);
6653 if (wc->stage == DROP_REFERENCE) {
6654 if (wc->refs[level - 1] > 1) {
6656 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6659 if (!wc->update_ref ||
6660 generation <= root->root_key.offset)
6663 btrfs_node_key_to_cpu(path->nodes[level], &key,
6664 path->slots[level]);
6665 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6669 wc->stage = UPDATE_BACKREF;
6670 wc->shared_level = level - 1;
6674 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6678 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6679 btrfs_tree_unlock(next);
6680 free_extent_buffer(next);
6686 if (reada && level == 1)
6687 reada_walk_down(trans, root, wc, path);
6688 next = read_tree_block(root, bytenr, blocksize, generation);
6691 btrfs_tree_lock(next);
6692 btrfs_set_lock_blocking(next);
6696 BUG_ON(level != btrfs_header_level(next));
6697 path->nodes[level] = next;
6698 path->slots[level] = 0;
6699 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6705 wc->refs[level - 1] = 0;
6706 wc->flags[level - 1] = 0;
6707 if (wc->stage == DROP_REFERENCE) {
6708 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6709 parent = path->nodes[level]->start;
6711 BUG_ON(root->root_key.objectid !=
6712 btrfs_header_owner(path->nodes[level]));
6716 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6717 root->root_key.objectid, level - 1, 0, 0);
6718 BUG_ON(ret); /* -ENOMEM */
6720 btrfs_tree_unlock(next);
6721 free_extent_buffer(next);
6727 * hepler to process tree block while walking up the tree.
6729 * when wc->stage == DROP_REFERENCE, this function drops
6730 * reference count on the block.
6732 * when wc->stage == UPDATE_BACKREF, this function changes
6733 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6734 * to UPDATE_BACKREF previously while processing the block.
6736 * NOTE: return value 1 means we should stop walking up.
6738 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6739 struct btrfs_root *root,
6740 struct btrfs_path *path,
6741 struct walk_control *wc)
6744 int level = wc->level;
6745 struct extent_buffer *eb = path->nodes[level];
6748 if (wc->stage == UPDATE_BACKREF) {
6749 BUG_ON(wc->shared_level < level);
6750 if (level < wc->shared_level)
6753 ret = find_next_key(path, level + 1, &wc->update_progress);
6757 wc->stage = DROP_REFERENCE;
6758 wc->shared_level = -1;
6759 path->slots[level] = 0;
6762 * check reference count again if the block isn't locked.
6763 * we should start walking down the tree again if reference
6766 if (!path->locks[level]) {
6768 btrfs_tree_lock(eb);
6769 btrfs_set_lock_blocking(eb);
6770 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6772 ret = btrfs_lookup_extent_info(trans, root,
6777 btrfs_tree_unlock_rw(eb, path->locks[level]);
6780 BUG_ON(wc->refs[level] == 0);
6781 if (wc->refs[level] == 1) {
6782 btrfs_tree_unlock_rw(eb, path->locks[level]);
6788 /* wc->stage == DROP_REFERENCE */
6789 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6791 if (wc->refs[level] == 1) {
6793 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6794 ret = btrfs_dec_ref(trans, root, eb, 1,
6797 ret = btrfs_dec_ref(trans, root, eb, 0,
6799 BUG_ON(ret); /* -ENOMEM */
6801 /* make block locked assertion in clean_tree_block happy */
6802 if (!path->locks[level] &&
6803 btrfs_header_generation(eb) == trans->transid) {
6804 btrfs_tree_lock(eb);
6805 btrfs_set_lock_blocking(eb);
6806 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6808 clean_tree_block(trans, root, eb);
6811 if (eb == root->node) {
6812 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6815 BUG_ON(root->root_key.objectid !=
6816 btrfs_header_owner(eb));
6818 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6819 parent = path->nodes[level + 1]->start;
6821 BUG_ON(root->root_key.objectid !=
6822 btrfs_header_owner(path->nodes[level + 1]));
6825 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6827 wc->refs[level] = 0;
6828 wc->flags[level] = 0;
6832 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6833 struct btrfs_root *root,
6834 struct btrfs_path *path,
6835 struct walk_control *wc)
6837 int level = wc->level;
6838 int lookup_info = 1;
6841 while (level >= 0) {
6842 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6849 if (path->slots[level] >=
6850 btrfs_header_nritems(path->nodes[level]))
6853 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6855 path->slots[level]++;
6864 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6865 struct btrfs_root *root,
6866 struct btrfs_path *path,
6867 struct walk_control *wc, int max_level)
6869 int level = wc->level;
6872 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6873 while (level < max_level && path->nodes[level]) {
6875 if (path->slots[level] + 1 <
6876 btrfs_header_nritems(path->nodes[level])) {
6877 path->slots[level]++;
6880 ret = walk_up_proc(trans, root, path, wc);
6884 if (path->locks[level]) {
6885 btrfs_tree_unlock_rw(path->nodes[level],
6886 path->locks[level]);
6887 path->locks[level] = 0;
6889 free_extent_buffer(path->nodes[level]);
6890 path->nodes[level] = NULL;
6898 * drop a subvolume tree.
6900 * this function traverses the tree freeing any blocks that only
6901 * referenced by the tree.
6903 * when a shared tree block is found. this function decreases its
6904 * reference count by one. if update_ref is true, this function
6905 * also make sure backrefs for the shared block and all lower level
6906 * blocks are properly updated.
6908 int btrfs_drop_snapshot(struct btrfs_root *root,
6909 struct btrfs_block_rsv *block_rsv, int update_ref,
6912 struct btrfs_path *path;
6913 struct btrfs_trans_handle *trans;
6914 struct btrfs_root *tree_root = root->fs_info->tree_root;
6915 struct btrfs_root_item *root_item = &root->root_item;
6916 struct walk_control *wc;
6917 struct btrfs_key key;
6922 path = btrfs_alloc_path();
6928 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6930 btrfs_free_path(path);
6935 trans = btrfs_start_transaction(tree_root, 0);
6936 if (IS_ERR(trans)) {
6937 err = PTR_ERR(trans);
6942 trans->block_rsv = block_rsv;
6944 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6945 level = btrfs_header_level(root->node);
6946 path->nodes[level] = btrfs_lock_root_node(root);
6947 btrfs_set_lock_blocking(path->nodes[level]);
6948 path->slots[level] = 0;
6949 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6950 memset(&wc->update_progress, 0,
6951 sizeof(wc->update_progress));
6953 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6954 memcpy(&wc->update_progress, &key,
6955 sizeof(wc->update_progress));
6957 level = root_item->drop_level;
6959 path->lowest_level = level;
6960 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6961 path->lowest_level = 0;
6969 * unlock our path, this is safe because only this
6970 * function is allowed to delete this snapshot
6972 btrfs_unlock_up_safe(path, 0);
6974 level = btrfs_header_level(root->node);
6976 btrfs_tree_lock(path->nodes[level]);
6977 btrfs_set_lock_blocking(path->nodes[level]);
6979 ret = btrfs_lookup_extent_info(trans, root,
6980 path->nodes[level]->start,
6981 path->nodes[level]->len,
6988 BUG_ON(wc->refs[level] == 0);
6990 if (level == root_item->drop_level)
6993 btrfs_tree_unlock(path->nodes[level]);
6994 WARN_ON(wc->refs[level] != 1);
7000 wc->shared_level = -1;
7001 wc->stage = DROP_REFERENCE;
7002 wc->update_ref = update_ref;
7004 wc->for_reloc = for_reloc;
7005 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7008 ret = walk_down_tree(trans, root, path, wc);
7014 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7021 BUG_ON(wc->stage != DROP_REFERENCE);
7025 if (wc->stage == DROP_REFERENCE) {
7027 btrfs_node_key(path->nodes[level],
7028 &root_item->drop_progress,
7029 path->slots[level]);
7030 root_item->drop_level = level;
7033 BUG_ON(wc->level == 0);
7034 if (btrfs_should_end_transaction(trans, tree_root)) {
7035 ret = btrfs_update_root(trans, tree_root,
7039 btrfs_abort_transaction(trans, tree_root, ret);
7044 btrfs_end_transaction_throttle(trans, tree_root);
7045 trans = btrfs_start_transaction(tree_root, 0);
7046 if (IS_ERR(trans)) {
7047 err = PTR_ERR(trans);
7051 trans->block_rsv = block_rsv;
7054 btrfs_release_path(path);
7058 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7060 btrfs_abort_transaction(trans, tree_root, ret);
7064 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7065 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7068 btrfs_abort_transaction(trans, tree_root, ret);
7071 } else if (ret > 0) {
7072 /* if we fail to delete the orphan item this time
7073 * around, it'll get picked up the next time.
7075 * The most common failure here is just -ENOENT.
7077 btrfs_del_orphan_item(trans, tree_root,
7078 root->root_key.objectid);
7082 if (root->in_radix) {
7083 btrfs_free_fs_root(tree_root->fs_info, root);
7085 free_extent_buffer(root->node);
7086 free_extent_buffer(root->commit_root);
7090 btrfs_end_transaction_throttle(trans, tree_root);
7093 btrfs_free_path(path);
7096 btrfs_std_error(root->fs_info, err);
7101 * drop subtree rooted at tree block 'node'.
7103 * NOTE: this function will unlock and release tree block 'node'
7104 * only used by relocation code
7106 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7107 struct btrfs_root *root,
7108 struct extent_buffer *node,
7109 struct extent_buffer *parent)
7111 struct btrfs_path *path;
7112 struct walk_control *wc;
7118 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7120 path = btrfs_alloc_path();
7124 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7126 btrfs_free_path(path);
7130 btrfs_assert_tree_locked(parent);
7131 parent_level = btrfs_header_level(parent);
7132 extent_buffer_get(parent);
7133 path->nodes[parent_level] = parent;
7134 path->slots[parent_level] = btrfs_header_nritems(parent);
7136 btrfs_assert_tree_locked(node);
7137 level = btrfs_header_level(node);
7138 path->nodes[level] = node;
7139 path->slots[level] = 0;
7140 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7142 wc->refs[parent_level] = 1;
7143 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7145 wc->shared_level = -1;
7146 wc->stage = DROP_REFERENCE;
7150 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7153 wret = walk_down_tree(trans, root, path, wc);
7159 wret = walk_up_tree(trans, root, path, wc, parent_level);
7167 btrfs_free_path(path);
7171 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7177 * if restripe for this chunk_type is on pick target profile and
7178 * return, otherwise do the usual balance
7180 stripped = get_restripe_target(root->fs_info, flags);
7182 return extended_to_chunk(stripped);
7185 * we add in the count of missing devices because we want
7186 * to make sure that any RAID levels on a degraded FS
7187 * continue to be honored.
7189 num_devices = root->fs_info->fs_devices->rw_devices +
7190 root->fs_info->fs_devices->missing_devices;
7192 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7193 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7195 if (num_devices == 1) {
7196 stripped |= BTRFS_BLOCK_GROUP_DUP;
7197 stripped = flags & ~stripped;
7199 /* turn raid0 into single device chunks */
7200 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7203 /* turn mirroring into duplication */
7204 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7205 BTRFS_BLOCK_GROUP_RAID10))
7206 return stripped | BTRFS_BLOCK_GROUP_DUP;
7208 /* they already had raid on here, just return */
7209 if (flags & stripped)
7212 stripped |= BTRFS_BLOCK_GROUP_DUP;
7213 stripped = flags & ~stripped;
7215 /* switch duplicated blocks with raid1 */
7216 if (flags & BTRFS_BLOCK_GROUP_DUP)
7217 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7219 /* this is drive concat, leave it alone */
7225 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7227 struct btrfs_space_info *sinfo = cache->space_info;
7229 u64 min_allocable_bytes;
7234 * We need some metadata space and system metadata space for
7235 * allocating chunks in some corner cases until we force to set
7236 * it to be readonly.
7239 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7241 min_allocable_bytes = 1 * 1024 * 1024;
7243 min_allocable_bytes = 0;
7245 spin_lock(&sinfo->lock);
7246 spin_lock(&cache->lock);
7253 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7254 cache->bytes_super - btrfs_block_group_used(&cache->item);
7256 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7257 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7258 min_allocable_bytes <= sinfo->total_bytes) {
7259 sinfo->bytes_readonly += num_bytes;
7264 spin_unlock(&cache->lock);
7265 spin_unlock(&sinfo->lock);
7269 int btrfs_set_block_group_ro(struct btrfs_root *root,
7270 struct btrfs_block_group_cache *cache)
7273 struct btrfs_trans_handle *trans;
7279 trans = btrfs_join_transaction(root);
7281 return PTR_ERR(trans);
7283 alloc_flags = update_block_group_flags(root, cache->flags);
7284 if (alloc_flags != cache->flags) {
7285 ret = do_chunk_alloc(trans, root, alloc_flags,
7291 ret = set_block_group_ro(cache, 0);
7294 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7295 ret = do_chunk_alloc(trans, root, alloc_flags,
7299 ret = set_block_group_ro(cache, 0);
7301 btrfs_end_transaction(trans, root);
7305 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7306 struct btrfs_root *root, u64 type)
7308 u64 alloc_flags = get_alloc_profile(root, type);
7309 return do_chunk_alloc(trans, root, alloc_flags,
7314 * helper to account the unused space of all the readonly block group in the
7315 * list. takes mirrors into account.
7317 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7319 struct btrfs_block_group_cache *block_group;
7323 list_for_each_entry(block_group, groups_list, list) {
7324 spin_lock(&block_group->lock);
7326 if (!block_group->ro) {
7327 spin_unlock(&block_group->lock);
7331 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7332 BTRFS_BLOCK_GROUP_RAID10 |
7333 BTRFS_BLOCK_GROUP_DUP))
7338 free_bytes += (block_group->key.offset -
7339 btrfs_block_group_used(&block_group->item)) *
7342 spin_unlock(&block_group->lock);
7349 * helper to account the unused space of all the readonly block group in the
7350 * space_info. takes mirrors into account.
7352 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7357 spin_lock(&sinfo->lock);
7359 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7360 if (!list_empty(&sinfo->block_groups[i]))
7361 free_bytes += __btrfs_get_ro_block_group_free_space(
7362 &sinfo->block_groups[i]);
7364 spin_unlock(&sinfo->lock);
7369 void btrfs_set_block_group_rw(struct btrfs_root *root,
7370 struct btrfs_block_group_cache *cache)
7372 struct btrfs_space_info *sinfo = cache->space_info;
7377 spin_lock(&sinfo->lock);
7378 spin_lock(&cache->lock);
7379 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7380 cache->bytes_super - btrfs_block_group_used(&cache->item);
7381 sinfo->bytes_readonly -= num_bytes;
7383 spin_unlock(&cache->lock);
7384 spin_unlock(&sinfo->lock);
7388 * checks to see if its even possible to relocate this block group.
7390 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7391 * ok to go ahead and try.
7393 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7395 struct btrfs_block_group_cache *block_group;
7396 struct btrfs_space_info *space_info;
7397 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7398 struct btrfs_device *device;
7407 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7409 /* odd, couldn't find the block group, leave it alone */
7413 min_free = btrfs_block_group_used(&block_group->item);
7415 /* no bytes used, we're good */
7419 space_info = block_group->space_info;
7420 spin_lock(&space_info->lock);
7422 full = space_info->full;
7425 * if this is the last block group we have in this space, we can't
7426 * relocate it unless we're able to allocate a new chunk below.
7428 * Otherwise, we need to make sure we have room in the space to handle
7429 * all of the extents from this block group. If we can, we're good
7431 if ((space_info->total_bytes != block_group->key.offset) &&
7432 (space_info->bytes_used + space_info->bytes_reserved +
7433 space_info->bytes_pinned + space_info->bytes_readonly +
7434 min_free < space_info->total_bytes)) {
7435 spin_unlock(&space_info->lock);
7438 spin_unlock(&space_info->lock);
7441 * ok we don't have enough space, but maybe we have free space on our
7442 * devices to allocate new chunks for relocation, so loop through our
7443 * alloc devices and guess if we have enough space. if this block
7444 * group is going to be restriped, run checks against the target
7445 * profile instead of the current one.
7457 target = get_restripe_target(root->fs_info, block_group->flags);
7459 index = __get_raid_index(extended_to_chunk(target));
7462 * this is just a balance, so if we were marked as full
7463 * we know there is no space for a new chunk
7468 index = get_block_group_index(block_group);
7475 } else if (index == 1) {
7477 } else if (index == 2) {
7480 } else if (index == 3) {
7481 dev_min = fs_devices->rw_devices;
7482 do_div(min_free, dev_min);
7485 mutex_lock(&root->fs_info->chunk_mutex);
7486 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7490 * check to make sure we can actually find a chunk with enough
7491 * space to fit our block group in.
7493 if (device->total_bytes > device->bytes_used + min_free &&
7494 !device->is_tgtdev_for_dev_replace) {
7495 ret = find_free_dev_extent(device, min_free,
7500 if (dev_nr >= dev_min)
7506 mutex_unlock(&root->fs_info->chunk_mutex);
7508 btrfs_put_block_group(block_group);
7512 static int find_first_block_group(struct btrfs_root *root,
7513 struct btrfs_path *path, struct btrfs_key *key)
7516 struct btrfs_key found_key;
7517 struct extent_buffer *leaf;
7520 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7525 slot = path->slots[0];
7526 leaf = path->nodes[0];
7527 if (slot >= btrfs_header_nritems(leaf)) {
7528 ret = btrfs_next_leaf(root, path);
7535 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7537 if (found_key.objectid >= key->objectid &&
7538 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7548 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7550 struct btrfs_block_group_cache *block_group;
7554 struct inode *inode;
7556 block_group = btrfs_lookup_first_block_group(info, last);
7557 while (block_group) {
7558 spin_lock(&block_group->lock);
7559 if (block_group->iref)
7561 spin_unlock(&block_group->lock);
7562 block_group = next_block_group(info->tree_root,
7572 inode = block_group->inode;
7573 block_group->iref = 0;
7574 block_group->inode = NULL;
7575 spin_unlock(&block_group->lock);
7577 last = block_group->key.objectid + block_group->key.offset;
7578 btrfs_put_block_group(block_group);
7582 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7584 struct btrfs_block_group_cache *block_group;
7585 struct btrfs_space_info *space_info;
7586 struct btrfs_caching_control *caching_ctl;
7589 down_write(&info->extent_commit_sem);
7590 while (!list_empty(&info->caching_block_groups)) {
7591 caching_ctl = list_entry(info->caching_block_groups.next,
7592 struct btrfs_caching_control, list);
7593 list_del(&caching_ctl->list);
7594 put_caching_control(caching_ctl);
7596 up_write(&info->extent_commit_sem);
7598 spin_lock(&info->block_group_cache_lock);
7599 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7600 block_group = rb_entry(n, struct btrfs_block_group_cache,
7602 rb_erase(&block_group->cache_node,
7603 &info->block_group_cache_tree);
7604 spin_unlock(&info->block_group_cache_lock);
7606 down_write(&block_group->space_info->groups_sem);
7607 list_del(&block_group->list);
7608 up_write(&block_group->space_info->groups_sem);
7610 if (block_group->cached == BTRFS_CACHE_STARTED)
7611 wait_block_group_cache_done(block_group);
7614 * We haven't cached this block group, which means we could
7615 * possibly have excluded extents on this block group.
7617 if (block_group->cached == BTRFS_CACHE_NO)
7618 free_excluded_extents(info->extent_root, block_group);
7620 btrfs_remove_free_space_cache(block_group);
7621 btrfs_put_block_group(block_group);
7623 spin_lock(&info->block_group_cache_lock);
7625 spin_unlock(&info->block_group_cache_lock);
7627 /* now that all the block groups are freed, go through and
7628 * free all the space_info structs. This is only called during
7629 * the final stages of unmount, and so we know nobody is
7630 * using them. We call synchronize_rcu() once before we start,
7631 * just to be on the safe side.
7635 release_global_block_rsv(info);
7637 while(!list_empty(&info->space_info)) {
7638 space_info = list_entry(info->space_info.next,
7639 struct btrfs_space_info,
7641 if (space_info->bytes_pinned > 0 ||
7642 space_info->bytes_reserved > 0 ||
7643 space_info->bytes_may_use > 0) {
7645 dump_space_info(space_info, 0, 0);
7647 list_del(&space_info->list);
7653 static void __link_block_group(struct btrfs_space_info *space_info,
7654 struct btrfs_block_group_cache *cache)
7656 int index = get_block_group_index(cache);
7658 down_write(&space_info->groups_sem);
7659 list_add_tail(&cache->list, &space_info->block_groups[index]);
7660 up_write(&space_info->groups_sem);
7663 int btrfs_read_block_groups(struct btrfs_root *root)
7665 struct btrfs_path *path;
7667 struct btrfs_block_group_cache *cache;
7668 struct btrfs_fs_info *info = root->fs_info;
7669 struct btrfs_space_info *space_info;
7670 struct btrfs_key key;
7671 struct btrfs_key found_key;
7672 struct extent_buffer *leaf;
7676 root = info->extent_root;
7679 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7680 path = btrfs_alloc_path();
7685 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7686 if (btrfs_test_opt(root, SPACE_CACHE) &&
7687 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7689 if (btrfs_test_opt(root, CLEAR_CACHE))
7693 ret = find_first_block_group(root, path, &key);
7698 leaf = path->nodes[0];
7699 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7700 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7705 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7707 if (!cache->free_space_ctl) {
7713 atomic_set(&cache->count, 1);
7714 spin_lock_init(&cache->lock);
7715 cache->fs_info = info;
7716 INIT_LIST_HEAD(&cache->list);
7717 INIT_LIST_HEAD(&cache->cluster_list);
7721 * When we mount with old space cache, we need to
7722 * set BTRFS_DC_CLEAR and set dirty flag.
7724 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7725 * truncate the old free space cache inode and
7727 * b) Setting 'dirty flag' makes sure that we flush
7728 * the new space cache info onto disk.
7730 cache->disk_cache_state = BTRFS_DC_CLEAR;
7731 if (btrfs_test_opt(root, SPACE_CACHE))
7735 read_extent_buffer(leaf, &cache->item,
7736 btrfs_item_ptr_offset(leaf, path->slots[0]),
7737 sizeof(cache->item));
7738 memcpy(&cache->key, &found_key, sizeof(found_key));
7740 key.objectid = found_key.objectid + found_key.offset;
7741 btrfs_release_path(path);
7742 cache->flags = btrfs_block_group_flags(&cache->item);
7743 cache->sectorsize = root->sectorsize;
7745 btrfs_init_free_space_ctl(cache);
7748 * We need to exclude the super stripes now so that the space
7749 * info has super bytes accounted for, otherwise we'll think
7750 * we have more space than we actually do.
7752 exclude_super_stripes(root, cache);
7755 * check for two cases, either we are full, and therefore
7756 * don't need to bother with the caching work since we won't
7757 * find any space, or we are empty, and we can just add all
7758 * the space in and be done with it. This saves us _alot_ of
7759 * time, particularly in the full case.
7761 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7762 cache->last_byte_to_unpin = (u64)-1;
7763 cache->cached = BTRFS_CACHE_FINISHED;
7764 free_excluded_extents(root, cache);
7765 } else if (btrfs_block_group_used(&cache->item) == 0) {
7766 cache->last_byte_to_unpin = (u64)-1;
7767 cache->cached = BTRFS_CACHE_FINISHED;
7768 add_new_free_space(cache, root->fs_info,
7770 found_key.objectid +
7772 free_excluded_extents(root, cache);
7775 ret = update_space_info(info, cache->flags, found_key.offset,
7776 btrfs_block_group_used(&cache->item),
7778 BUG_ON(ret); /* -ENOMEM */
7779 cache->space_info = space_info;
7780 spin_lock(&cache->space_info->lock);
7781 cache->space_info->bytes_readonly += cache->bytes_super;
7782 spin_unlock(&cache->space_info->lock);
7784 __link_block_group(space_info, cache);
7786 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7787 BUG_ON(ret); /* Logic error */
7789 set_avail_alloc_bits(root->fs_info, cache->flags);
7790 if (btrfs_chunk_readonly(root, cache->key.objectid))
7791 set_block_group_ro(cache, 1);
7794 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7795 if (!(get_alloc_profile(root, space_info->flags) &
7796 (BTRFS_BLOCK_GROUP_RAID10 |
7797 BTRFS_BLOCK_GROUP_RAID1 |
7798 BTRFS_BLOCK_GROUP_DUP)))
7801 * avoid allocating from un-mirrored block group if there are
7802 * mirrored block groups.
7804 list_for_each_entry(cache, &space_info->block_groups[3], list)
7805 set_block_group_ro(cache, 1);
7806 list_for_each_entry(cache, &space_info->block_groups[4], list)
7807 set_block_group_ro(cache, 1);
7810 init_global_block_rsv(info);
7813 btrfs_free_path(path);
7817 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7818 struct btrfs_root *root)
7820 struct btrfs_block_group_cache *block_group, *tmp;
7821 struct btrfs_root *extent_root = root->fs_info->extent_root;
7822 struct btrfs_block_group_item item;
7823 struct btrfs_key key;
7826 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7828 list_del_init(&block_group->new_bg_list);
7833 spin_lock(&block_group->lock);
7834 memcpy(&item, &block_group->item, sizeof(item));
7835 memcpy(&key, &block_group->key, sizeof(key));
7836 spin_unlock(&block_group->lock);
7838 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7841 btrfs_abort_transaction(trans, extent_root, ret);
7845 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7846 struct btrfs_root *root, u64 bytes_used,
7847 u64 type, u64 chunk_objectid, u64 chunk_offset,
7851 struct btrfs_root *extent_root;
7852 struct btrfs_block_group_cache *cache;
7854 extent_root = root->fs_info->extent_root;
7856 root->fs_info->last_trans_log_full_commit = trans->transid;
7858 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7861 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7863 if (!cache->free_space_ctl) {
7868 cache->key.objectid = chunk_offset;
7869 cache->key.offset = size;
7870 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7871 cache->sectorsize = root->sectorsize;
7872 cache->fs_info = root->fs_info;
7874 atomic_set(&cache->count, 1);
7875 spin_lock_init(&cache->lock);
7876 INIT_LIST_HEAD(&cache->list);
7877 INIT_LIST_HEAD(&cache->cluster_list);
7878 INIT_LIST_HEAD(&cache->new_bg_list);
7880 btrfs_init_free_space_ctl(cache);
7882 btrfs_set_block_group_used(&cache->item, bytes_used);
7883 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7884 cache->flags = type;
7885 btrfs_set_block_group_flags(&cache->item, type);
7887 cache->last_byte_to_unpin = (u64)-1;
7888 cache->cached = BTRFS_CACHE_FINISHED;
7889 exclude_super_stripes(root, cache);
7891 add_new_free_space(cache, root->fs_info, chunk_offset,
7892 chunk_offset + size);
7894 free_excluded_extents(root, cache);
7896 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7897 &cache->space_info);
7898 BUG_ON(ret); /* -ENOMEM */
7899 update_global_block_rsv(root->fs_info);
7901 spin_lock(&cache->space_info->lock);
7902 cache->space_info->bytes_readonly += cache->bytes_super;
7903 spin_unlock(&cache->space_info->lock);
7905 __link_block_group(cache->space_info, cache);
7907 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7908 BUG_ON(ret); /* Logic error */
7910 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7912 set_avail_alloc_bits(extent_root->fs_info, type);
7917 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7919 u64 extra_flags = chunk_to_extended(flags) &
7920 BTRFS_EXTENDED_PROFILE_MASK;
7922 if (flags & BTRFS_BLOCK_GROUP_DATA)
7923 fs_info->avail_data_alloc_bits &= ~extra_flags;
7924 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7925 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7926 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7927 fs_info->avail_system_alloc_bits &= ~extra_flags;
7930 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7931 struct btrfs_root *root, u64 group_start)
7933 struct btrfs_path *path;
7934 struct btrfs_block_group_cache *block_group;
7935 struct btrfs_free_cluster *cluster;
7936 struct btrfs_root *tree_root = root->fs_info->tree_root;
7937 struct btrfs_key key;
7938 struct inode *inode;
7943 root = root->fs_info->extent_root;
7945 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7946 BUG_ON(!block_group);
7947 BUG_ON(!block_group->ro);
7950 * Free the reserved super bytes from this block group before
7953 free_excluded_extents(root, block_group);
7955 memcpy(&key, &block_group->key, sizeof(key));
7956 index = get_block_group_index(block_group);
7957 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7958 BTRFS_BLOCK_GROUP_RAID1 |
7959 BTRFS_BLOCK_GROUP_RAID10))
7964 /* make sure this block group isn't part of an allocation cluster */
7965 cluster = &root->fs_info->data_alloc_cluster;
7966 spin_lock(&cluster->refill_lock);
7967 btrfs_return_cluster_to_free_space(block_group, cluster);
7968 spin_unlock(&cluster->refill_lock);
7971 * make sure this block group isn't part of a metadata
7972 * allocation cluster
7974 cluster = &root->fs_info->meta_alloc_cluster;
7975 spin_lock(&cluster->refill_lock);
7976 btrfs_return_cluster_to_free_space(block_group, cluster);
7977 spin_unlock(&cluster->refill_lock);
7979 path = btrfs_alloc_path();
7985 inode = lookup_free_space_inode(tree_root, block_group, path);
7986 if (!IS_ERR(inode)) {
7987 ret = btrfs_orphan_add(trans, inode);
7989 btrfs_add_delayed_iput(inode);
7993 /* One for the block groups ref */
7994 spin_lock(&block_group->lock);
7995 if (block_group->iref) {
7996 block_group->iref = 0;
7997 block_group->inode = NULL;
7998 spin_unlock(&block_group->lock);
8001 spin_unlock(&block_group->lock);
8003 /* One for our lookup ref */
8004 btrfs_add_delayed_iput(inode);
8007 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8008 key.offset = block_group->key.objectid;
8011 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8015 btrfs_release_path(path);
8017 ret = btrfs_del_item(trans, tree_root, path);
8020 btrfs_release_path(path);
8023 spin_lock(&root->fs_info->block_group_cache_lock);
8024 rb_erase(&block_group->cache_node,
8025 &root->fs_info->block_group_cache_tree);
8026 spin_unlock(&root->fs_info->block_group_cache_lock);
8028 down_write(&block_group->space_info->groups_sem);
8030 * we must use list_del_init so people can check to see if they
8031 * are still on the list after taking the semaphore
8033 list_del_init(&block_group->list);
8034 if (list_empty(&block_group->space_info->block_groups[index]))
8035 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8036 up_write(&block_group->space_info->groups_sem);
8038 if (block_group->cached == BTRFS_CACHE_STARTED)
8039 wait_block_group_cache_done(block_group);
8041 btrfs_remove_free_space_cache(block_group);
8043 spin_lock(&block_group->space_info->lock);
8044 block_group->space_info->total_bytes -= block_group->key.offset;
8045 block_group->space_info->bytes_readonly -= block_group->key.offset;
8046 block_group->space_info->disk_total -= block_group->key.offset * factor;
8047 spin_unlock(&block_group->space_info->lock);
8049 memcpy(&key, &block_group->key, sizeof(key));
8051 btrfs_clear_space_info_full(root->fs_info);
8053 btrfs_put_block_group(block_group);
8054 btrfs_put_block_group(block_group);
8056 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8062 ret = btrfs_del_item(trans, root, path);
8064 btrfs_free_path(path);
8068 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8070 struct btrfs_space_info *space_info;
8071 struct btrfs_super_block *disk_super;
8077 disk_super = fs_info->super_copy;
8078 if (!btrfs_super_root(disk_super))
8081 features = btrfs_super_incompat_flags(disk_super);
8082 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8085 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8086 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8091 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8092 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8094 flags = BTRFS_BLOCK_GROUP_METADATA;
8095 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8099 flags = BTRFS_BLOCK_GROUP_DATA;
8100 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8106 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8108 return unpin_extent_range(root, start, end);
8111 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8112 u64 num_bytes, u64 *actual_bytes)
8114 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8117 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8119 struct btrfs_fs_info *fs_info = root->fs_info;
8120 struct btrfs_block_group_cache *cache = NULL;
8125 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8129 * try to trim all FS space, our block group may start from non-zero.
8131 if (range->len == total_bytes)
8132 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8134 cache = btrfs_lookup_block_group(fs_info, range->start);
8137 if (cache->key.objectid >= (range->start + range->len)) {
8138 btrfs_put_block_group(cache);
8142 start = max(range->start, cache->key.objectid);
8143 end = min(range->start + range->len,
8144 cache->key.objectid + cache->key.offset);
8146 if (end - start >= range->minlen) {
8147 if (!block_group_cache_done(cache)) {
8148 ret = cache_block_group(cache, NULL, root, 0);
8150 wait_block_group_cache_done(cache);
8152 ret = btrfs_trim_block_group(cache,
8158 trimmed += group_trimmed;
8160 btrfs_put_block_group(cache);
8165 cache = next_block_group(fs_info->tree_root, cache);
8168 range->len = trimmed;