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->mapping_tree, 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)
3693 * shrink metadata reservation for delalloc
3695 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3698 struct btrfs_block_rsv *block_rsv;
3699 struct btrfs_space_info *space_info;
3700 struct btrfs_trans_handle *trans;
3704 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3706 enum btrfs_reserve_flush_enum flush;
3708 trans = (struct btrfs_trans_handle *)current->journal_info;
3709 block_rsv = &root->fs_info->delalloc_block_rsv;
3710 space_info = block_rsv->space_info;
3713 delalloc_bytes = root->fs_info->delalloc_bytes;
3714 if (delalloc_bytes == 0) {
3717 btrfs_wait_ordered_extents(root, 0);
3721 while (delalloc_bytes && loops < 3) {
3722 max_reclaim = min(delalloc_bytes, to_reclaim);
3723 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3724 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3725 WB_REASON_FS_FREE_SPACE);
3728 * We need to wait for the async pages to actually start before
3731 wait_event(root->fs_info->async_submit_wait,
3732 !atomic_read(&root->fs_info->async_delalloc_pages));
3735 flush = BTRFS_RESERVE_FLUSH_ALL;
3737 flush = BTRFS_RESERVE_NO_FLUSH;
3738 spin_lock(&space_info->lock);
3739 if (can_overcommit(root, space_info, orig, flush)) {
3740 spin_unlock(&space_info->lock);
3743 spin_unlock(&space_info->lock);
3746 if (wait_ordered && !trans) {
3747 btrfs_wait_ordered_extents(root, 0);
3749 time_left = schedule_timeout_killable(1);
3754 delalloc_bytes = root->fs_info->delalloc_bytes;
3759 * maybe_commit_transaction - possibly commit the transaction if its ok to
3760 * @root - the root we're allocating for
3761 * @bytes - the number of bytes we want to reserve
3762 * @force - force the commit
3764 * This will check to make sure that committing the transaction will actually
3765 * get us somewhere and then commit the transaction if it does. Otherwise it
3766 * will return -ENOSPC.
3768 static int may_commit_transaction(struct btrfs_root *root,
3769 struct btrfs_space_info *space_info,
3770 u64 bytes, int force)
3772 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3773 struct btrfs_trans_handle *trans;
3775 trans = (struct btrfs_trans_handle *)current->journal_info;
3782 /* See if there is enough pinned space to make this reservation */
3783 spin_lock(&space_info->lock);
3784 if (space_info->bytes_pinned >= bytes) {
3785 spin_unlock(&space_info->lock);
3788 spin_unlock(&space_info->lock);
3791 * See if there is some space in the delayed insertion reservation for
3794 if (space_info != delayed_rsv->space_info)
3797 spin_lock(&space_info->lock);
3798 spin_lock(&delayed_rsv->lock);
3799 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3800 spin_unlock(&delayed_rsv->lock);
3801 spin_unlock(&space_info->lock);
3804 spin_unlock(&delayed_rsv->lock);
3805 spin_unlock(&space_info->lock);
3808 trans = btrfs_join_transaction(root);
3812 return btrfs_commit_transaction(trans, root);
3816 FLUSH_DELAYED_ITEMS_NR = 1,
3817 FLUSH_DELAYED_ITEMS = 2,
3819 FLUSH_DELALLOC_WAIT = 4,
3824 static int flush_space(struct btrfs_root *root,
3825 struct btrfs_space_info *space_info, u64 num_bytes,
3826 u64 orig_bytes, int state)
3828 struct btrfs_trans_handle *trans;
3833 case FLUSH_DELAYED_ITEMS_NR:
3834 case FLUSH_DELAYED_ITEMS:
3835 if (state == FLUSH_DELAYED_ITEMS_NR) {
3836 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3838 nr = (int)div64_u64(num_bytes, bytes);
3845 trans = btrfs_join_transaction(root);
3846 if (IS_ERR(trans)) {
3847 ret = PTR_ERR(trans);
3850 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3851 btrfs_end_transaction(trans, root);
3853 case FLUSH_DELALLOC:
3854 case FLUSH_DELALLOC_WAIT:
3855 shrink_delalloc(root, num_bytes, orig_bytes,
3856 state == FLUSH_DELALLOC_WAIT);
3859 trans = btrfs_join_transaction(root);
3860 if (IS_ERR(trans)) {
3861 ret = PTR_ERR(trans);
3864 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3865 btrfs_get_alloc_profile(root, 0),
3866 CHUNK_ALLOC_NO_FORCE);
3867 btrfs_end_transaction(trans, root);
3872 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3882 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3883 * @root - the root we're allocating for
3884 * @block_rsv - the block_rsv we're allocating for
3885 * @orig_bytes - the number of bytes we want
3886 * @flush - wether or not we can flush to make our reservation
3888 * This will reserve orgi_bytes number of bytes from the space info associated
3889 * with the block_rsv. If there is not enough space it will make an attempt to
3890 * flush out space to make room. It will do this by flushing delalloc if
3891 * possible or committing the transaction. If flush is 0 then no attempts to
3892 * regain reservations will be made and this will fail if there is not enough
3895 static int reserve_metadata_bytes(struct btrfs_root *root,
3896 struct btrfs_block_rsv *block_rsv,
3898 enum btrfs_reserve_flush_enum flush)
3900 struct btrfs_space_info *space_info = block_rsv->space_info;
3902 u64 num_bytes = orig_bytes;
3903 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3905 bool flushing = false;
3909 spin_lock(&space_info->lock);
3911 * We only want to wait if somebody other than us is flushing and we
3912 * are actually allowed to flush all things.
3914 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3915 space_info->flush) {
3916 spin_unlock(&space_info->lock);
3918 * If we have a trans handle we can't wait because the flusher
3919 * may have to commit the transaction, which would mean we would
3920 * deadlock since we are waiting for the flusher to finish, but
3921 * hold the current transaction open.
3923 if (current->journal_info)
3925 ret = wait_event_killable(space_info->wait, !space_info->flush);
3926 /* Must have been killed, return */
3930 spin_lock(&space_info->lock);
3934 used = space_info->bytes_used + space_info->bytes_reserved +
3935 space_info->bytes_pinned + space_info->bytes_readonly +
3936 space_info->bytes_may_use;
3939 * The idea here is that we've not already over-reserved the block group
3940 * then we can go ahead and save our reservation first and then start
3941 * flushing if we need to. Otherwise if we've already overcommitted
3942 * lets start flushing stuff first and then come back and try to make
3945 if (used <= space_info->total_bytes) {
3946 if (used + orig_bytes <= space_info->total_bytes) {
3947 space_info->bytes_may_use += orig_bytes;
3948 trace_btrfs_space_reservation(root->fs_info,
3949 "space_info", space_info->flags, orig_bytes, 1);
3953 * Ok set num_bytes to orig_bytes since we aren't
3954 * overocmmitted, this way we only try and reclaim what
3957 num_bytes = orig_bytes;
3961 * Ok we're over committed, set num_bytes to the overcommitted
3962 * amount plus the amount of bytes that we need for this
3965 num_bytes = used - space_info->total_bytes +
3969 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
3970 space_info->bytes_may_use += orig_bytes;
3971 trace_btrfs_space_reservation(root->fs_info, "space_info",
3972 space_info->flags, orig_bytes,
3978 * Couldn't make our reservation, save our place so while we're trying
3979 * to reclaim space we can actually use it instead of somebody else
3980 * stealing it from us.
3982 * We make the other tasks wait for the flush only when we can flush
3985 if (ret && flush == BTRFS_RESERVE_FLUSH_ALL) {
3987 space_info->flush = 1;
3990 spin_unlock(&space_info->lock);
3992 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
3995 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4000 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4001 * would happen. So skip delalloc flush.
4003 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4004 (flush_state == FLUSH_DELALLOC ||
4005 flush_state == FLUSH_DELALLOC_WAIT))
4006 flush_state = ALLOC_CHUNK;
4010 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4011 flush_state < COMMIT_TRANS)
4013 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4014 flush_state <= COMMIT_TRANS)
4019 spin_lock(&space_info->lock);
4020 space_info->flush = 0;
4021 wake_up_all(&space_info->wait);
4022 spin_unlock(&space_info->lock);
4027 static struct btrfs_block_rsv *get_block_rsv(
4028 const struct btrfs_trans_handle *trans,
4029 const struct btrfs_root *root)
4031 struct btrfs_block_rsv *block_rsv = NULL;
4034 block_rsv = trans->block_rsv;
4036 if (root == root->fs_info->csum_root && trans->adding_csums)
4037 block_rsv = trans->block_rsv;
4040 block_rsv = root->block_rsv;
4043 block_rsv = &root->fs_info->empty_block_rsv;
4048 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4052 spin_lock(&block_rsv->lock);
4053 if (block_rsv->reserved >= num_bytes) {
4054 block_rsv->reserved -= num_bytes;
4055 if (block_rsv->reserved < block_rsv->size)
4056 block_rsv->full = 0;
4059 spin_unlock(&block_rsv->lock);
4063 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4064 u64 num_bytes, int update_size)
4066 spin_lock(&block_rsv->lock);
4067 block_rsv->reserved += num_bytes;
4069 block_rsv->size += num_bytes;
4070 else if (block_rsv->reserved >= block_rsv->size)
4071 block_rsv->full = 1;
4072 spin_unlock(&block_rsv->lock);
4075 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4076 struct btrfs_block_rsv *block_rsv,
4077 struct btrfs_block_rsv *dest, u64 num_bytes)
4079 struct btrfs_space_info *space_info = block_rsv->space_info;
4081 spin_lock(&block_rsv->lock);
4082 if (num_bytes == (u64)-1)
4083 num_bytes = block_rsv->size;
4084 block_rsv->size -= num_bytes;
4085 if (block_rsv->reserved >= block_rsv->size) {
4086 num_bytes = block_rsv->reserved - block_rsv->size;
4087 block_rsv->reserved = block_rsv->size;
4088 block_rsv->full = 1;
4092 spin_unlock(&block_rsv->lock);
4094 if (num_bytes > 0) {
4096 spin_lock(&dest->lock);
4100 bytes_to_add = dest->size - dest->reserved;
4101 bytes_to_add = min(num_bytes, bytes_to_add);
4102 dest->reserved += bytes_to_add;
4103 if (dest->reserved >= dest->size)
4105 num_bytes -= bytes_to_add;
4107 spin_unlock(&dest->lock);
4110 spin_lock(&space_info->lock);
4111 space_info->bytes_may_use -= num_bytes;
4112 trace_btrfs_space_reservation(fs_info, "space_info",
4113 space_info->flags, num_bytes, 0);
4114 space_info->reservation_progress++;
4115 spin_unlock(&space_info->lock);
4120 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4121 struct btrfs_block_rsv *dst, u64 num_bytes)
4125 ret = block_rsv_use_bytes(src, num_bytes);
4129 block_rsv_add_bytes(dst, num_bytes, 1);
4133 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4135 memset(rsv, 0, sizeof(*rsv));
4136 spin_lock_init(&rsv->lock);
4140 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4141 unsigned short type)
4143 struct btrfs_block_rsv *block_rsv;
4144 struct btrfs_fs_info *fs_info = root->fs_info;
4146 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4150 btrfs_init_block_rsv(block_rsv, type);
4151 block_rsv->space_info = __find_space_info(fs_info,
4152 BTRFS_BLOCK_GROUP_METADATA);
4156 void btrfs_free_block_rsv(struct btrfs_root *root,
4157 struct btrfs_block_rsv *rsv)
4161 btrfs_block_rsv_release(root, rsv, (u64)-1);
4165 int btrfs_block_rsv_add(struct btrfs_root *root,
4166 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4167 enum btrfs_reserve_flush_enum flush)
4174 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4176 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4183 int btrfs_block_rsv_check(struct btrfs_root *root,
4184 struct btrfs_block_rsv *block_rsv, int min_factor)
4192 spin_lock(&block_rsv->lock);
4193 num_bytes = div_factor(block_rsv->size, min_factor);
4194 if (block_rsv->reserved >= num_bytes)
4196 spin_unlock(&block_rsv->lock);
4201 int btrfs_block_rsv_refill(struct btrfs_root *root,
4202 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4203 enum btrfs_reserve_flush_enum flush)
4211 spin_lock(&block_rsv->lock);
4212 num_bytes = min_reserved;
4213 if (block_rsv->reserved >= num_bytes)
4216 num_bytes -= block_rsv->reserved;
4217 spin_unlock(&block_rsv->lock);
4222 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4224 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4231 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4232 struct btrfs_block_rsv *dst_rsv,
4235 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4238 void btrfs_block_rsv_release(struct btrfs_root *root,
4239 struct btrfs_block_rsv *block_rsv,
4242 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4243 if (global_rsv->full || global_rsv == block_rsv ||
4244 block_rsv->space_info != global_rsv->space_info)
4246 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4251 * helper to calculate size of global block reservation.
4252 * the desired value is sum of space used by extent tree,
4253 * checksum tree and root tree
4255 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4257 struct btrfs_space_info *sinfo;
4261 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4263 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4264 spin_lock(&sinfo->lock);
4265 data_used = sinfo->bytes_used;
4266 spin_unlock(&sinfo->lock);
4268 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4269 spin_lock(&sinfo->lock);
4270 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4272 meta_used = sinfo->bytes_used;
4273 spin_unlock(&sinfo->lock);
4275 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4277 num_bytes += div64_u64(data_used + meta_used, 50);
4279 if (num_bytes * 3 > meta_used)
4280 num_bytes = div64_u64(meta_used, 3);
4282 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4285 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4287 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4288 struct btrfs_space_info *sinfo = block_rsv->space_info;
4291 num_bytes = calc_global_metadata_size(fs_info);
4293 spin_lock(&sinfo->lock);
4294 spin_lock(&block_rsv->lock);
4296 block_rsv->size = num_bytes;
4298 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4299 sinfo->bytes_reserved + sinfo->bytes_readonly +
4300 sinfo->bytes_may_use;
4302 if (sinfo->total_bytes > num_bytes) {
4303 num_bytes = sinfo->total_bytes - num_bytes;
4304 block_rsv->reserved += num_bytes;
4305 sinfo->bytes_may_use += num_bytes;
4306 trace_btrfs_space_reservation(fs_info, "space_info",
4307 sinfo->flags, num_bytes, 1);
4310 if (block_rsv->reserved >= block_rsv->size) {
4311 num_bytes = block_rsv->reserved - block_rsv->size;
4312 sinfo->bytes_may_use -= num_bytes;
4313 trace_btrfs_space_reservation(fs_info, "space_info",
4314 sinfo->flags, num_bytes, 0);
4315 sinfo->reservation_progress++;
4316 block_rsv->reserved = block_rsv->size;
4317 block_rsv->full = 1;
4320 spin_unlock(&block_rsv->lock);
4321 spin_unlock(&sinfo->lock);
4324 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4326 struct btrfs_space_info *space_info;
4328 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4329 fs_info->chunk_block_rsv.space_info = space_info;
4331 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4332 fs_info->global_block_rsv.space_info = space_info;
4333 fs_info->delalloc_block_rsv.space_info = space_info;
4334 fs_info->trans_block_rsv.space_info = space_info;
4335 fs_info->empty_block_rsv.space_info = space_info;
4336 fs_info->delayed_block_rsv.space_info = space_info;
4338 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4339 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4340 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4341 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4342 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4344 update_global_block_rsv(fs_info);
4347 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4349 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4351 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4352 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4353 WARN_ON(fs_info->trans_block_rsv.size > 0);
4354 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4355 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4356 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4357 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4358 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4361 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4362 struct btrfs_root *root)
4364 if (!trans->block_rsv)
4367 if (!trans->bytes_reserved)
4370 trace_btrfs_space_reservation(root->fs_info, "transaction",
4371 trans->transid, trans->bytes_reserved, 0);
4372 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4373 trans->bytes_reserved = 0;
4376 /* Can only return 0 or -ENOSPC */
4377 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4378 struct inode *inode)
4380 struct btrfs_root *root = BTRFS_I(inode)->root;
4381 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4382 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4385 * We need to hold space in order to delete our orphan item once we've
4386 * added it, so this takes the reservation so we can release it later
4387 * when we are truly done with the orphan item.
4389 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4390 trace_btrfs_space_reservation(root->fs_info, "orphan",
4391 btrfs_ino(inode), num_bytes, 1);
4392 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4395 void btrfs_orphan_release_metadata(struct inode *inode)
4397 struct btrfs_root *root = BTRFS_I(inode)->root;
4398 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4399 trace_btrfs_space_reservation(root->fs_info, "orphan",
4400 btrfs_ino(inode), num_bytes, 0);
4401 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4404 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4405 struct btrfs_pending_snapshot *pending)
4407 struct btrfs_root *root = pending->root;
4408 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4409 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4411 * two for root back/forward refs, two for directory entries,
4412 * one for root of the snapshot and one for parent inode.
4414 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4415 dst_rsv->space_info = src_rsv->space_info;
4416 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4420 * drop_outstanding_extent - drop an outstanding extent
4421 * @inode: the inode we're dropping the extent for
4423 * This is called when we are freeing up an outstanding extent, either called
4424 * after an error or after an extent is written. This will return the number of
4425 * reserved extents that need to be freed. This must be called with
4426 * BTRFS_I(inode)->lock held.
4428 static unsigned drop_outstanding_extent(struct inode *inode)
4430 unsigned drop_inode_space = 0;
4431 unsigned dropped_extents = 0;
4433 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4434 BTRFS_I(inode)->outstanding_extents--;
4436 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4437 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4438 &BTRFS_I(inode)->runtime_flags))
4439 drop_inode_space = 1;
4442 * If we have more or the same amount of outsanding extents than we have
4443 * reserved then we need to leave the reserved extents count alone.
4445 if (BTRFS_I(inode)->outstanding_extents >=
4446 BTRFS_I(inode)->reserved_extents)
4447 return drop_inode_space;
4449 dropped_extents = BTRFS_I(inode)->reserved_extents -
4450 BTRFS_I(inode)->outstanding_extents;
4451 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4452 return dropped_extents + drop_inode_space;
4456 * calc_csum_metadata_size - return the amount of metada space that must be
4457 * reserved/free'd for the given bytes.
4458 * @inode: the inode we're manipulating
4459 * @num_bytes: the number of bytes in question
4460 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4462 * This adjusts the number of csum_bytes in the inode and then returns the
4463 * correct amount of metadata that must either be reserved or freed. We
4464 * calculate how many checksums we can fit into one leaf and then divide the
4465 * number of bytes that will need to be checksumed by this value to figure out
4466 * how many checksums will be required. If we are adding bytes then the number
4467 * may go up and we will return the number of additional bytes that must be
4468 * reserved. If it is going down we will return the number of bytes that must
4471 * This must be called with BTRFS_I(inode)->lock held.
4473 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4476 struct btrfs_root *root = BTRFS_I(inode)->root;
4478 int num_csums_per_leaf;
4482 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4483 BTRFS_I(inode)->csum_bytes == 0)
4486 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4488 BTRFS_I(inode)->csum_bytes += num_bytes;
4490 BTRFS_I(inode)->csum_bytes -= num_bytes;
4491 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4492 num_csums_per_leaf = (int)div64_u64(csum_size,
4493 sizeof(struct btrfs_csum_item) +
4494 sizeof(struct btrfs_disk_key));
4495 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4496 num_csums = num_csums + num_csums_per_leaf - 1;
4497 num_csums = num_csums / num_csums_per_leaf;
4499 old_csums = old_csums + num_csums_per_leaf - 1;
4500 old_csums = old_csums / num_csums_per_leaf;
4502 /* No change, no need to reserve more */
4503 if (old_csums == num_csums)
4507 return btrfs_calc_trans_metadata_size(root,
4508 num_csums - old_csums);
4510 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4513 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4515 struct btrfs_root *root = BTRFS_I(inode)->root;
4516 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4519 unsigned nr_extents = 0;
4520 int extra_reserve = 0;
4521 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4524 /* Need to be holding the i_mutex here if we aren't free space cache */
4525 if (btrfs_is_free_space_inode(inode))
4526 flush = BTRFS_RESERVE_NO_FLUSH;
4528 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4529 btrfs_transaction_in_commit(root->fs_info))
4530 schedule_timeout(1);
4532 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4533 num_bytes = ALIGN(num_bytes, root->sectorsize);
4535 spin_lock(&BTRFS_I(inode)->lock);
4536 BTRFS_I(inode)->outstanding_extents++;
4538 if (BTRFS_I(inode)->outstanding_extents >
4539 BTRFS_I(inode)->reserved_extents)
4540 nr_extents = BTRFS_I(inode)->outstanding_extents -
4541 BTRFS_I(inode)->reserved_extents;
4544 * Add an item to reserve for updating the inode when we complete the
4547 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4548 &BTRFS_I(inode)->runtime_flags)) {
4553 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4554 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4555 csum_bytes = BTRFS_I(inode)->csum_bytes;
4556 spin_unlock(&BTRFS_I(inode)->lock);
4558 if (root->fs_info->quota_enabled) {
4559 ret = btrfs_qgroup_reserve(root, num_bytes +
4560 nr_extents * root->leafsize);
4562 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4567 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4572 spin_lock(&BTRFS_I(inode)->lock);
4573 dropped = drop_outstanding_extent(inode);
4575 * If the inodes csum_bytes is the same as the original
4576 * csum_bytes then we know we haven't raced with any free()ers
4577 * so we can just reduce our inodes csum bytes and carry on.
4578 * Otherwise we have to do the normal free thing to account for
4579 * the case that the free side didn't free up its reserve
4580 * because of this outstanding reservation.
4582 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4583 calc_csum_metadata_size(inode, num_bytes, 0);
4585 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4586 spin_unlock(&BTRFS_I(inode)->lock);
4588 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4591 btrfs_block_rsv_release(root, block_rsv, to_free);
4592 trace_btrfs_space_reservation(root->fs_info,
4597 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4601 spin_lock(&BTRFS_I(inode)->lock);
4602 if (extra_reserve) {
4603 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4604 &BTRFS_I(inode)->runtime_flags);
4607 BTRFS_I(inode)->reserved_extents += nr_extents;
4608 spin_unlock(&BTRFS_I(inode)->lock);
4609 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4612 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4613 btrfs_ino(inode), to_reserve, 1);
4614 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4620 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4621 * @inode: the inode to release the reservation for
4622 * @num_bytes: the number of bytes we're releasing
4624 * This will release the metadata reservation for an inode. This can be called
4625 * once we complete IO for a given set of bytes to release their metadata
4628 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4630 struct btrfs_root *root = BTRFS_I(inode)->root;
4634 num_bytes = ALIGN(num_bytes, root->sectorsize);
4635 spin_lock(&BTRFS_I(inode)->lock);
4636 dropped = drop_outstanding_extent(inode);
4638 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4639 spin_unlock(&BTRFS_I(inode)->lock);
4641 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4643 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4644 btrfs_ino(inode), to_free, 0);
4645 if (root->fs_info->quota_enabled) {
4646 btrfs_qgroup_free(root, num_bytes +
4647 dropped * root->leafsize);
4650 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4655 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4656 * @inode: inode we're writing to
4657 * @num_bytes: the number of bytes we want to allocate
4659 * This will do the following things
4661 * o reserve space in the data space info for num_bytes
4662 * o reserve space in the metadata space info based on number of outstanding
4663 * extents and how much csums will be needed
4664 * o add to the inodes ->delalloc_bytes
4665 * o add it to the fs_info's delalloc inodes list.
4667 * This will return 0 for success and -ENOSPC if there is no space left.
4669 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4673 ret = btrfs_check_data_free_space(inode, num_bytes);
4677 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4679 btrfs_free_reserved_data_space(inode, num_bytes);
4687 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4688 * @inode: inode we're releasing space for
4689 * @num_bytes: the number of bytes we want to free up
4691 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4692 * called in the case that we don't need the metadata AND data reservations
4693 * anymore. So if there is an error or we insert an inline extent.
4695 * This function will release the metadata space that was not used and will
4696 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4697 * list if there are no delalloc bytes left.
4699 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4701 btrfs_delalloc_release_metadata(inode, num_bytes);
4702 btrfs_free_reserved_data_space(inode, num_bytes);
4705 static int update_block_group(struct btrfs_trans_handle *trans,
4706 struct btrfs_root *root,
4707 u64 bytenr, u64 num_bytes, int alloc)
4709 struct btrfs_block_group_cache *cache = NULL;
4710 struct btrfs_fs_info *info = root->fs_info;
4711 u64 total = num_bytes;
4716 /* block accounting for super block */
4717 spin_lock(&info->delalloc_lock);
4718 old_val = btrfs_super_bytes_used(info->super_copy);
4720 old_val += num_bytes;
4722 old_val -= num_bytes;
4723 btrfs_set_super_bytes_used(info->super_copy, old_val);
4724 spin_unlock(&info->delalloc_lock);
4727 cache = btrfs_lookup_block_group(info, bytenr);
4730 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4731 BTRFS_BLOCK_GROUP_RAID1 |
4732 BTRFS_BLOCK_GROUP_RAID10))
4737 * If this block group has free space cache written out, we
4738 * need to make sure to load it if we are removing space. This
4739 * is because we need the unpinning stage to actually add the
4740 * space back to the block group, otherwise we will leak space.
4742 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4743 cache_block_group(cache, trans, NULL, 1);
4745 byte_in_group = bytenr - cache->key.objectid;
4746 WARN_ON(byte_in_group > cache->key.offset);
4748 spin_lock(&cache->space_info->lock);
4749 spin_lock(&cache->lock);
4751 if (btrfs_test_opt(root, SPACE_CACHE) &&
4752 cache->disk_cache_state < BTRFS_DC_CLEAR)
4753 cache->disk_cache_state = BTRFS_DC_CLEAR;
4756 old_val = btrfs_block_group_used(&cache->item);
4757 num_bytes = min(total, cache->key.offset - byte_in_group);
4759 old_val += num_bytes;
4760 btrfs_set_block_group_used(&cache->item, old_val);
4761 cache->reserved -= num_bytes;
4762 cache->space_info->bytes_reserved -= num_bytes;
4763 cache->space_info->bytes_used += num_bytes;
4764 cache->space_info->disk_used += num_bytes * factor;
4765 spin_unlock(&cache->lock);
4766 spin_unlock(&cache->space_info->lock);
4768 old_val -= num_bytes;
4769 btrfs_set_block_group_used(&cache->item, old_val);
4770 cache->pinned += num_bytes;
4771 cache->space_info->bytes_pinned += num_bytes;
4772 cache->space_info->bytes_used -= num_bytes;
4773 cache->space_info->disk_used -= num_bytes * factor;
4774 spin_unlock(&cache->lock);
4775 spin_unlock(&cache->space_info->lock);
4777 set_extent_dirty(info->pinned_extents,
4778 bytenr, bytenr + num_bytes - 1,
4779 GFP_NOFS | __GFP_NOFAIL);
4781 btrfs_put_block_group(cache);
4783 bytenr += num_bytes;
4788 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4790 struct btrfs_block_group_cache *cache;
4793 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4797 bytenr = cache->key.objectid;
4798 btrfs_put_block_group(cache);
4803 static int pin_down_extent(struct btrfs_root *root,
4804 struct btrfs_block_group_cache *cache,
4805 u64 bytenr, u64 num_bytes, int reserved)
4807 spin_lock(&cache->space_info->lock);
4808 spin_lock(&cache->lock);
4809 cache->pinned += num_bytes;
4810 cache->space_info->bytes_pinned += num_bytes;
4812 cache->reserved -= num_bytes;
4813 cache->space_info->bytes_reserved -= num_bytes;
4815 spin_unlock(&cache->lock);
4816 spin_unlock(&cache->space_info->lock);
4818 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4819 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4824 * this function must be called within transaction
4826 int btrfs_pin_extent(struct btrfs_root *root,
4827 u64 bytenr, u64 num_bytes, int reserved)
4829 struct btrfs_block_group_cache *cache;
4831 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4832 BUG_ON(!cache); /* Logic error */
4834 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4836 btrfs_put_block_group(cache);
4841 * this function must be called within transaction
4843 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4844 struct btrfs_root *root,
4845 u64 bytenr, u64 num_bytes)
4847 struct btrfs_block_group_cache *cache;
4849 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4850 BUG_ON(!cache); /* Logic error */
4853 * pull in the free space cache (if any) so that our pin
4854 * removes the free space from the cache. We have load_only set
4855 * to one because the slow code to read in the free extents does check
4856 * the pinned extents.
4858 cache_block_group(cache, trans, root, 1);
4860 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4862 /* remove us from the free space cache (if we're there at all) */
4863 btrfs_remove_free_space(cache, bytenr, num_bytes);
4864 btrfs_put_block_group(cache);
4869 * btrfs_update_reserved_bytes - update the block_group and space info counters
4870 * @cache: The cache we are manipulating
4871 * @num_bytes: The number of bytes in question
4872 * @reserve: One of the reservation enums
4874 * This is called by the allocator when it reserves space, or by somebody who is
4875 * freeing space that was never actually used on disk. For example if you
4876 * reserve some space for a new leaf in transaction A and before transaction A
4877 * commits you free that leaf, you call this with reserve set to 0 in order to
4878 * clear the reservation.
4880 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4881 * ENOSPC accounting. For data we handle the reservation through clearing the
4882 * delalloc bits in the io_tree. We have to do this since we could end up
4883 * allocating less disk space for the amount of data we have reserved in the
4884 * case of compression.
4886 * If this is a reservation and the block group has become read only we cannot
4887 * make the reservation and return -EAGAIN, otherwise this function always
4890 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4891 u64 num_bytes, int reserve)
4893 struct btrfs_space_info *space_info = cache->space_info;
4896 spin_lock(&space_info->lock);
4897 spin_lock(&cache->lock);
4898 if (reserve != RESERVE_FREE) {
4902 cache->reserved += num_bytes;
4903 space_info->bytes_reserved += num_bytes;
4904 if (reserve == RESERVE_ALLOC) {
4905 trace_btrfs_space_reservation(cache->fs_info,
4906 "space_info", space_info->flags,
4908 space_info->bytes_may_use -= num_bytes;
4913 space_info->bytes_readonly += num_bytes;
4914 cache->reserved -= num_bytes;
4915 space_info->bytes_reserved -= num_bytes;
4916 space_info->reservation_progress++;
4918 spin_unlock(&cache->lock);
4919 spin_unlock(&space_info->lock);
4923 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4924 struct btrfs_root *root)
4926 struct btrfs_fs_info *fs_info = root->fs_info;
4927 struct btrfs_caching_control *next;
4928 struct btrfs_caching_control *caching_ctl;
4929 struct btrfs_block_group_cache *cache;
4931 down_write(&fs_info->extent_commit_sem);
4933 list_for_each_entry_safe(caching_ctl, next,
4934 &fs_info->caching_block_groups, list) {
4935 cache = caching_ctl->block_group;
4936 if (block_group_cache_done(cache)) {
4937 cache->last_byte_to_unpin = (u64)-1;
4938 list_del_init(&caching_ctl->list);
4939 put_caching_control(caching_ctl);
4941 cache->last_byte_to_unpin = caching_ctl->progress;
4945 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4946 fs_info->pinned_extents = &fs_info->freed_extents[1];
4948 fs_info->pinned_extents = &fs_info->freed_extents[0];
4950 up_write(&fs_info->extent_commit_sem);
4952 update_global_block_rsv(fs_info);
4955 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4957 struct btrfs_fs_info *fs_info = root->fs_info;
4958 struct btrfs_block_group_cache *cache = NULL;
4959 struct btrfs_space_info *space_info;
4960 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4964 while (start <= end) {
4967 start >= cache->key.objectid + cache->key.offset) {
4969 btrfs_put_block_group(cache);
4970 cache = btrfs_lookup_block_group(fs_info, start);
4971 BUG_ON(!cache); /* Logic error */
4974 len = cache->key.objectid + cache->key.offset - start;
4975 len = min(len, end + 1 - start);
4977 if (start < cache->last_byte_to_unpin) {
4978 len = min(len, cache->last_byte_to_unpin - start);
4979 btrfs_add_free_space(cache, start, len);
4983 space_info = cache->space_info;
4985 spin_lock(&space_info->lock);
4986 spin_lock(&cache->lock);
4987 cache->pinned -= len;
4988 space_info->bytes_pinned -= len;
4990 space_info->bytes_readonly += len;
4993 spin_unlock(&cache->lock);
4994 if (!readonly && global_rsv->space_info == space_info) {
4995 spin_lock(&global_rsv->lock);
4996 if (!global_rsv->full) {
4997 len = min(len, global_rsv->size -
4998 global_rsv->reserved);
4999 global_rsv->reserved += len;
5000 space_info->bytes_may_use += len;
5001 if (global_rsv->reserved >= global_rsv->size)
5002 global_rsv->full = 1;
5004 spin_unlock(&global_rsv->lock);
5006 spin_unlock(&space_info->lock);
5010 btrfs_put_block_group(cache);
5014 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5015 struct btrfs_root *root)
5017 struct btrfs_fs_info *fs_info = root->fs_info;
5018 struct extent_io_tree *unpin;
5026 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5027 unpin = &fs_info->freed_extents[1];
5029 unpin = &fs_info->freed_extents[0];
5032 ret = find_first_extent_bit(unpin, 0, &start, &end,
5033 EXTENT_DIRTY, NULL);
5037 if (btrfs_test_opt(root, DISCARD))
5038 ret = btrfs_discard_extent(root, start,
5039 end + 1 - start, NULL);
5041 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5042 unpin_extent_range(root, start, end);
5049 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5050 struct btrfs_root *root,
5051 u64 bytenr, u64 num_bytes, u64 parent,
5052 u64 root_objectid, u64 owner_objectid,
5053 u64 owner_offset, int refs_to_drop,
5054 struct btrfs_delayed_extent_op *extent_op)
5056 struct btrfs_key key;
5057 struct btrfs_path *path;
5058 struct btrfs_fs_info *info = root->fs_info;
5059 struct btrfs_root *extent_root = info->extent_root;
5060 struct extent_buffer *leaf;
5061 struct btrfs_extent_item *ei;
5062 struct btrfs_extent_inline_ref *iref;
5065 int extent_slot = 0;
5066 int found_extent = 0;
5071 path = btrfs_alloc_path();
5076 path->leave_spinning = 1;
5078 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5079 BUG_ON(!is_data && refs_to_drop != 1);
5081 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5082 bytenr, num_bytes, parent,
5083 root_objectid, owner_objectid,
5086 extent_slot = path->slots[0];
5087 while (extent_slot >= 0) {
5088 btrfs_item_key_to_cpu(path->nodes[0], &key,
5090 if (key.objectid != bytenr)
5092 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5093 key.offset == num_bytes) {
5097 if (path->slots[0] - extent_slot > 5)
5101 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5102 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5103 if (found_extent && item_size < sizeof(*ei))
5106 if (!found_extent) {
5108 ret = remove_extent_backref(trans, extent_root, path,
5112 btrfs_abort_transaction(trans, extent_root, ret);
5115 btrfs_release_path(path);
5116 path->leave_spinning = 1;
5118 key.objectid = bytenr;
5119 key.type = BTRFS_EXTENT_ITEM_KEY;
5120 key.offset = num_bytes;
5122 ret = btrfs_search_slot(trans, extent_root,
5125 printk(KERN_ERR "umm, got %d back from search"
5126 ", was looking for %llu\n", ret,
5127 (unsigned long long)bytenr);
5129 btrfs_print_leaf(extent_root,
5133 btrfs_abort_transaction(trans, extent_root, ret);
5136 extent_slot = path->slots[0];
5138 } else if (ret == -ENOENT) {
5139 btrfs_print_leaf(extent_root, path->nodes[0]);
5141 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5142 "parent %llu root %llu owner %llu offset %llu\n",
5143 (unsigned long long)bytenr,
5144 (unsigned long long)parent,
5145 (unsigned long long)root_objectid,
5146 (unsigned long long)owner_objectid,
5147 (unsigned long long)owner_offset);
5149 btrfs_abort_transaction(trans, extent_root, ret);
5153 leaf = path->nodes[0];
5154 item_size = btrfs_item_size_nr(leaf, extent_slot);
5155 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5156 if (item_size < sizeof(*ei)) {
5157 BUG_ON(found_extent || extent_slot != path->slots[0]);
5158 ret = convert_extent_item_v0(trans, extent_root, path,
5161 btrfs_abort_transaction(trans, extent_root, ret);
5165 btrfs_release_path(path);
5166 path->leave_spinning = 1;
5168 key.objectid = bytenr;
5169 key.type = BTRFS_EXTENT_ITEM_KEY;
5170 key.offset = num_bytes;
5172 ret = btrfs_search_slot(trans, extent_root, &key, path,
5175 printk(KERN_ERR "umm, got %d back from search"
5176 ", was looking for %llu\n", ret,
5177 (unsigned long long)bytenr);
5178 btrfs_print_leaf(extent_root, path->nodes[0]);
5181 btrfs_abort_transaction(trans, extent_root, ret);
5185 extent_slot = path->slots[0];
5186 leaf = path->nodes[0];
5187 item_size = btrfs_item_size_nr(leaf, extent_slot);
5190 BUG_ON(item_size < sizeof(*ei));
5191 ei = btrfs_item_ptr(leaf, extent_slot,
5192 struct btrfs_extent_item);
5193 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5194 struct btrfs_tree_block_info *bi;
5195 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5196 bi = (struct btrfs_tree_block_info *)(ei + 1);
5197 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5200 refs = btrfs_extent_refs(leaf, ei);
5201 BUG_ON(refs < refs_to_drop);
5202 refs -= refs_to_drop;
5206 __run_delayed_extent_op(extent_op, leaf, ei);
5208 * In the case of inline back ref, reference count will
5209 * be updated by remove_extent_backref
5212 BUG_ON(!found_extent);
5214 btrfs_set_extent_refs(leaf, ei, refs);
5215 btrfs_mark_buffer_dirty(leaf);
5218 ret = remove_extent_backref(trans, extent_root, path,
5222 btrfs_abort_transaction(trans, extent_root, ret);
5228 BUG_ON(is_data && refs_to_drop !=
5229 extent_data_ref_count(root, path, iref));
5231 BUG_ON(path->slots[0] != extent_slot);
5233 BUG_ON(path->slots[0] != extent_slot + 1);
5234 path->slots[0] = extent_slot;
5239 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5242 btrfs_abort_transaction(trans, extent_root, ret);
5245 btrfs_release_path(path);
5248 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5250 btrfs_abort_transaction(trans, extent_root, ret);
5255 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5257 btrfs_abort_transaction(trans, extent_root, ret);
5262 btrfs_free_path(path);
5267 * when we free an block, it is possible (and likely) that we free the last
5268 * delayed ref for that extent as well. This searches the delayed ref tree for
5269 * a given extent, and if there are no other delayed refs to be processed, it
5270 * removes it from the tree.
5272 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5273 struct btrfs_root *root, u64 bytenr)
5275 struct btrfs_delayed_ref_head *head;
5276 struct btrfs_delayed_ref_root *delayed_refs;
5277 struct btrfs_delayed_ref_node *ref;
5278 struct rb_node *node;
5281 delayed_refs = &trans->transaction->delayed_refs;
5282 spin_lock(&delayed_refs->lock);
5283 head = btrfs_find_delayed_ref_head(trans, bytenr);
5287 node = rb_prev(&head->node.rb_node);
5291 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5293 /* there are still entries for this ref, we can't drop it */
5294 if (ref->bytenr == bytenr)
5297 if (head->extent_op) {
5298 if (!head->must_insert_reserved)
5300 kfree(head->extent_op);
5301 head->extent_op = NULL;
5305 * waiting for the lock here would deadlock. If someone else has it
5306 * locked they are already in the process of dropping it anyway
5308 if (!mutex_trylock(&head->mutex))
5312 * at this point we have a head with no other entries. Go
5313 * ahead and process it.
5315 head->node.in_tree = 0;
5316 rb_erase(&head->node.rb_node, &delayed_refs->root);
5318 delayed_refs->num_entries--;
5321 * we don't take a ref on the node because we're removing it from the
5322 * tree, so we just steal the ref the tree was holding.
5324 delayed_refs->num_heads--;
5325 if (list_empty(&head->cluster))
5326 delayed_refs->num_heads_ready--;
5328 list_del_init(&head->cluster);
5329 spin_unlock(&delayed_refs->lock);
5331 BUG_ON(head->extent_op);
5332 if (head->must_insert_reserved)
5335 mutex_unlock(&head->mutex);
5336 btrfs_put_delayed_ref(&head->node);
5339 spin_unlock(&delayed_refs->lock);
5343 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5344 struct btrfs_root *root,
5345 struct extent_buffer *buf,
5346 u64 parent, int last_ref)
5348 struct btrfs_block_group_cache *cache = NULL;
5351 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5352 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5353 buf->start, buf->len,
5354 parent, root->root_key.objectid,
5355 btrfs_header_level(buf),
5356 BTRFS_DROP_DELAYED_REF, NULL, 0);
5357 BUG_ON(ret); /* -ENOMEM */
5363 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5365 if (btrfs_header_generation(buf) == trans->transid) {
5366 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5367 ret = check_ref_cleanup(trans, root, buf->start);
5372 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5373 pin_down_extent(root, cache, buf->start, buf->len, 1);
5377 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5379 btrfs_add_free_space(cache, buf->start, buf->len);
5380 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5384 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5387 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5388 btrfs_put_block_group(cache);
5391 /* Can return -ENOMEM */
5392 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5393 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5394 u64 owner, u64 offset, int for_cow)
5397 struct btrfs_fs_info *fs_info = root->fs_info;
5400 * tree log blocks never actually go into the extent allocation
5401 * tree, just update pinning info and exit early.
5403 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5404 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5405 /* unlocks the pinned mutex */
5406 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5408 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5409 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5411 parent, root_objectid, (int)owner,
5412 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5414 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5416 parent, root_objectid, owner,
5417 offset, BTRFS_DROP_DELAYED_REF,
5423 static u64 stripe_align(struct btrfs_root *root, u64 val)
5425 u64 mask = ((u64)root->stripesize - 1);
5426 u64 ret = (val + mask) & ~mask;
5431 * when we wait for progress in the block group caching, its because
5432 * our allocation attempt failed at least once. So, we must sleep
5433 * and let some progress happen before we try again.
5435 * This function will sleep at least once waiting for new free space to
5436 * show up, and then it will check the block group free space numbers
5437 * for our min num_bytes. Another option is to have it go ahead
5438 * and look in the rbtree for a free extent of a given size, but this
5442 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5445 struct btrfs_caching_control *caching_ctl;
5448 caching_ctl = get_caching_control(cache);
5452 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5453 (cache->free_space_ctl->free_space >= num_bytes));
5455 put_caching_control(caching_ctl);
5460 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5462 struct btrfs_caching_control *caching_ctl;
5465 caching_ctl = get_caching_control(cache);
5469 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5471 put_caching_control(caching_ctl);
5475 static int __get_block_group_index(u64 flags)
5479 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5481 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5483 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5485 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5493 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5495 return __get_block_group_index(cache->flags);
5498 enum btrfs_loop_type {
5499 LOOP_CACHING_NOWAIT = 0,
5500 LOOP_CACHING_WAIT = 1,
5501 LOOP_ALLOC_CHUNK = 2,
5502 LOOP_NO_EMPTY_SIZE = 3,
5506 * walks the btree of allocated extents and find a hole of a given size.
5507 * The key ins is changed to record the hole:
5508 * ins->objectid == block start
5509 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5510 * ins->offset == number of blocks
5511 * Any available blocks before search_start are skipped.
5513 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5514 struct btrfs_root *orig_root,
5515 u64 num_bytes, u64 empty_size,
5516 u64 hint_byte, struct btrfs_key *ins,
5520 struct btrfs_root *root = orig_root->fs_info->extent_root;
5521 struct btrfs_free_cluster *last_ptr = NULL;
5522 struct btrfs_block_group_cache *block_group = NULL;
5523 struct btrfs_block_group_cache *used_block_group;
5524 u64 search_start = 0;
5525 int empty_cluster = 2 * 1024 * 1024;
5526 struct btrfs_space_info *space_info;
5529 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5530 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5531 bool found_uncached_bg = false;
5532 bool failed_cluster_refill = false;
5533 bool failed_alloc = false;
5534 bool use_cluster = true;
5535 bool have_caching_bg = false;
5537 WARN_ON(num_bytes < root->sectorsize);
5538 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5542 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5544 space_info = __find_space_info(root->fs_info, data);
5546 printk(KERN_ERR "No space info for %llu\n", data);
5551 * If the space info is for both data and metadata it means we have a
5552 * small filesystem and we can't use the clustering stuff.
5554 if (btrfs_mixed_space_info(space_info))
5555 use_cluster = false;
5557 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5558 last_ptr = &root->fs_info->meta_alloc_cluster;
5559 if (!btrfs_test_opt(root, SSD))
5560 empty_cluster = 64 * 1024;
5563 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5564 btrfs_test_opt(root, SSD)) {
5565 last_ptr = &root->fs_info->data_alloc_cluster;
5569 spin_lock(&last_ptr->lock);
5570 if (last_ptr->block_group)
5571 hint_byte = last_ptr->window_start;
5572 spin_unlock(&last_ptr->lock);
5575 search_start = max(search_start, first_logical_byte(root, 0));
5576 search_start = max(search_start, hint_byte);
5581 if (search_start == hint_byte) {
5582 block_group = btrfs_lookup_block_group(root->fs_info,
5584 used_block_group = block_group;
5586 * we don't want to use the block group if it doesn't match our
5587 * allocation bits, or if its not cached.
5589 * However if we are re-searching with an ideal block group
5590 * picked out then we don't care that the block group is cached.
5592 if (block_group && block_group_bits(block_group, data) &&
5593 block_group->cached != BTRFS_CACHE_NO) {
5594 down_read(&space_info->groups_sem);
5595 if (list_empty(&block_group->list) ||
5598 * someone is removing this block group,
5599 * we can't jump into the have_block_group
5600 * target because our list pointers are not
5603 btrfs_put_block_group(block_group);
5604 up_read(&space_info->groups_sem);
5606 index = get_block_group_index(block_group);
5607 goto have_block_group;
5609 } else if (block_group) {
5610 btrfs_put_block_group(block_group);
5614 have_caching_bg = false;
5615 down_read(&space_info->groups_sem);
5616 list_for_each_entry(block_group, &space_info->block_groups[index],
5621 used_block_group = block_group;
5622 btrfs_get_block_group(block_group);
5623 search_start = block_group->key.objectid;
5626 * this can happen if we end up cycling through all the
5627 * raid types, but we want to make sure we only allocate
5628 * for the proper type.
5630 if (!block_group_bits(block_group, data)) {
5631 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5632 BTRFS_BLOCK_GROUP_RAID1 |
5633 BTRFS_BLOCK_GROUP_RAID10;
5636 * if they asked for extra copies and this block group
5637 * doesn't provide them, bail. This does allow us to
5638 * fill raid0 from raid1.
5640 if ((data & extra) && !(block_group->flags & extra))
5645 cached = block_group_cache_done(block_group);
5646 if (unlikely(!cached)) {
5647 found_uncached_bg = true;
5648 ret = cache_block_group(block_group, trans,
5654 if (unlikely(block_group->ro))
5658 * Ok we want to try and use the cluster allocator, so
5663 * the refill lock keeps out other
5664 * people trying to start a new cluster
5666 spin_lock(&last_ptr->refill_lock);
5667 used_block_group = last_ptr->block_group;
5668 if (used_block_group != block_group &&
5669 (!used_block_group ||
5670 used_block_group->ro ||
5671 !block_group_bits(used_block_group, data))) {
5672 used_block_group = block_group;
5673 goto refill_cluster;
5676 if (used_block_group != block_group)
5677 btrfs_get_block_group(used_block_group);
5679 offset = btrfs_alloc_from_cluster(used_block_group,
5680 last_ptr, num_bytes, used_block_group->key.objectid);
5682 /* we have a block, we're done */
5683 spin_unlock(&last_ptr->refill_lock);
5684 trace_btrfs_reserve_extent_cluster(root,
5685 block_group, search_start, num_bytes);
5689 WARN_ON(last_ptr->block_group != used_block_group);
5690 if (used_block_group != block_group) {
5691 btrfs_put_block_group(used_block_group);
5692 used_block_group = block_group;
5695 BUG_ON(used_block_group != block_group);
5696 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5697 * set up a new clusters, so lets just skip it
5698 * and let the allocator find whatever block
5699 * it can find. If we reach this point, we
5700 * will have tried the cluster allocator
5701 * plenty of times and not have found
5702 * anything, so we are likely way too
5703 * fragmented for the clustering stuff to find
5706 * However, if the cluster is taken from the
5707 * current block group, release the cluster
5708 * first, so that we stand a better chance of
5709 * succeeding in the unclustered
5711 if (loop >= LOOP_NO_EMPTY_SIZE &&
5712 last_ptr->block_group != block_group) {
5713 spin_unlock(&last_ptr->refill_lock);
5714 goto unclustered_alloc;
5718 * this cluster didn't work out, free it and
5721 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5723 if (loop >= LOOP_NO_EMPTY_SIZE) {
5724 spin_unlock(&last_ptr->refill_lock);
5725 goto unclustered_alloc;
5728 /* allocate a cluster in this block group */
5729 ret = btrfs_find_space_cluster(trans, root,
5730 block_group, last_ptr,
5731 search_start, num_bytes,
5732 empty_cluster + empty_size);
5735 * now pull our allocation out of this
5738 offset = btrfs_alloc_from_cluster(block_group,
5739 last_ptr, num_bytes,
5742 /* we found one, proceed */
5743 spin_unlock(&last_ptr->refill_lock);
5744 trace_btrfs_reserve_extent_cluster(root,
5745 block_group, search_start,
5749 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5750 && !failed_cluster_refill) {
5751 spin_unlock(&last_ptr->refill_lock);
5753 failed_cluster_refill = true;
5754 wait_block_group_cache_progress(block_group,
5755 num_bytes + empty_cluster + empty_size);
5756 goto have_block_group;
5760 * at this point we either didn't find a cluster
5761 * or we weren't able to allocate a block from our
5762 * cluster. Free the cluster we've been trying
5763 * to use, and go to the next block group
5765 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5766 spin_unlock(&last_ptr->refill_lock);
5771 spin_lock(&block_group->free_space_ctl->tree_lock);
5773 block_group->free_space_ctl->free_space <
5774 num_bytes + empty_cluster + empty_size) {
5775 spin_unlock(&block_group->free_space_ctl->tree_lock);
5778 spin_unlock(&block_group->free_space_ctl->tree_lock);
5780 offset = btrfs_find_space_for_alloc(block_group, search_start,
5781 num_bytes, empty_size);
5783 * If we didn't find a chunk, and we haven't failed on this
5784 * block group before, and this block group is in the middle of
5785 * caching and we are ok with waiting, then go ahead and wait
5786 * for progress to be made, and set failed_alloc to true.
5788 * If failed_alloc is true then we've already waited on this
5789 * block group once and should move on to the next block group.
5791 if (!offset && !failed_alloc && !cached &&
5792 loop > LOOP_CACHING_NOWAIT) {
5793 wait_block_group_cache_progress(block_group,
5794 num_bytes + empty_size);
5795 failed_alloc = true;
5796 goto have_block_group;
5797 } else if (!offset) {
5799 have_caching_bg = true;
5803 search_start = stripe_align(root, offset);
5805 /* move on to the next group */
5806 if (search_start + num_bytes >
5807 used_block_group->key.objectid + used_block_group->key.offset) {
5808 btrfs_add_free_space(used_block_group, offset, num_bytes);
5812 if (offset < search_start)
5813 btrfs_add_free_space(used_block_group, offset,
5814 search_start - offset);
5815 BUG_ON(offset > search_start);
5817 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5819 if (ret == -EAGAIN) {
5820 btrfs_add_free_space(used_block_group, offset, num_bytes);
5824 /* we are all good, lets return */
5825 ins->objectid = search_start;
5826 ins->offset = num_bytes;
5828 trace_btrfs_reserve_extent(orig_root, block_group,
5829 search_start, num_bytes);
5830 if (used_block_group != block_group)
5831 btrfs_put_block_group(used_block_group);
5832 btrfs_put_block_group(block_group);
5835 failed_cluster_refill = false;
5836 failed_alloc = false;
5837 BUG_ON(index != get_block_group_index(block_group));
5838 if (used_block_group != block_group)
5839 btrfs_put_block_group(used_block_group);
5840 btrfs_put_block_group(block_group);
5842 up_read(&space_info->groups_sem);
5844 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5847 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5851 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5852 * caching kthreads as we move along
5853 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5854 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5855 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5858 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5861 if (loop == LOOP_ALLOC_CHUNK) {
5862 ret = do_chunk_alloc(trans, root, data,
5865 * Do not bail out on ENOSPC since we
5866 * can do more things.
5868 if (ret < 0 && ret != -ENOSPC) {
5869 btrfs_abort_transaction(trans,
5875 if (loop == LOOP_NO_EMPTY_SIZE) {
5881 } else if (!ins->objectid) {
5883 } else if (ins->objectid) {
5891 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5892 int dump_block_groups)
5894 struct btrfs_block_group_cache *cache;
5897 spin_lock(&info->lock);
5898 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5899 (unsigned long long)info->flags,
5900 (unsigned long long)(info->total_bytes - info->bytes_used -
5901 info->bytes_pinned - info->bytes_reserved -
5902 info->bytes_readonly),
5903 (info->full) ? "" : "not ");
5904 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5905 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5906 (unsigned long long)info->total_bytes,
5907 (unsigned long long)info->bytes_used,
5908 (unsigned long long)info->bytes_pinned,
5909 (unsigned long long)info->bytes_reserved,
5910 (unsigned long long)info->bytes_may_use,
5911 (unsigned long long)info->bytes_readonly);
5912 spin_unlock(&info->lock);
5914 if (!dump_block_groups)
5917 down_read(&info->groups_sem);
5919 list_for_each_entry(cache, &info->block_groups[index], list) {
5920 spin_lock(&cache->lock);
5921 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5922 (unsigned long long)cache->key.objectid,
5923 (unsigned long long)cache->key.offset,
5924 (unsigned long long)btrfs_block_group_used(&cache->item),
5925 (unsigned long long)cache->pinned,
5926 (unsigned long long)cache->reserved,
5927 cache->ro ? "[readonly]" : "");
5928 btrfs_dump_free_space(cache, bytes);
5929 spin_unlock(&cache->lock);
5931 if (++index < BTRFS_NR_RAID_TYPES)
5933 up_read(&info->groups_sem);
5936 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5937 struct btrfs_root *root,
5938 u64 num_bytes, u64 min_alloc_size,
5939 u64 empty_size, u64 hint_byte,
5940 struct btrfs_key *ins, u64 data)
5942 bool final_tried = false;
5945 data = btrfs_get_alloc_profile(root, data);
5947 WARN_ON(num_bytes < root->sectorsize);
5948 ret = find_free_extent(trans, root, num_bytes, empty_size,
5949 hint_byte, ins, data);
5951 if (ret == -ENOSPC) {
5953 num_bytes = num_bytes >> 1;
5954 num_bytes = num_bytes & ~(root->sectorsize - 1);
5955 num_bytes = max(num_bytes, min_alloc_size);
5956 if (num_bytes == min_alloc_size)
5959 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5960 struct btrfs_space_info *sinfo;
5962 sinfo = __find_space_info(root->fs_info, data);
5963 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5964 "wanted %llu\n", (unsigned long long)data,
5965 (unsigned long long)num_bytes);
5967 dump_space_info(sinfo, num_bytes, 1);
5971 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5976 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5977 u64 start, u64 len, int pin)
5979 struct btrfs_block_group_cache *cache;
5982 cache = btrfs_lookup_block_group(root->fs_info, start);
5984 printk(KERN_ERR "Unable to find block group for %llu\n",
5985 (unsigned long long)start);
5989 if (btrfs_test_opt(root, DISCARD))
5990 ret = btrfs_discard_extent(root, start, len, NULL);
5993 pin_down_extent(root, cache, start, len, 1);
5995 btrfs_add_free_space(cache, start, len);
5996 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5998 btrfs_put_block_group(cache);
6000 trace_btrfs_reserved_extent_free(root, start, len);
6005 int btrfs_free_reserved_extent(struct btrfs_root *root,
6008 return __btrfs_free_reserved_extent(root, start, len, 0);
6011 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6014 return __btrfs_free_reserved_extent(root, start, len, 1);
6017 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6018 struct btrfs_root *root,
6019 u64 parent, u64 root_objectid,
6020 u64 flags, u64 owner, u64 offset,
6021 struct btrfs_key *ins, int ref_mod)
6024 struct btrfs_fs_info *fs_info = root->fs_info;
6025 struct btrfs_extent_item *extent_item;
6026 struct btrfs_extent_inline_ref *iref;
6027 struct btrfs_path *path;
6028 struct extent_buffer *leaf;
6033 type = BTRFS_SHARED_DATA_REF_KEY;
6035 type = BTRFS_EXTENT_DATA_REF_KEY;
6037 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6039 path = btrfs_alloc_path();
6043 path->leave_spinning = 1;
6044 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6047 btrfs_free_path(path);
6051 leaf = path->nodes[0];
6052 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6053 struct btrfs_extent_item);
6054 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6055 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6056 btrfs_set_extent_flags(leaf, extent_item,
6057 flags | BTRFS_EXTENT_FLAG_DATA);
6059 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6060 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6062 struct btrfs_shared_data_ref *ref;
6063 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6064 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6065 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6067 struct btrfs_extent_data_ref *ref;
6068 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6069 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6070 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6071 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6072 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6075 btrfs_mark_buffer_dirty(path->nodes[0]);
6076 btrfs_free_path(path);
6078 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6079 if (ret) { /* -ENOENT, logic error */
6080 printk(KERN_ERR "btrfs update block group failed for %llu "
6081 "%llu\n", (unsigned long long)ins->objectid,
6082 (unsigned long long)ins->offset);
6088 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6089 struct btrfs_root *root,
6090 u64 parent, u64 root_objectid,
6091 u64 flags, struct btrfs_disk_key *key,
6092 int level, struct btrfs_key *ins)
6095 struct btrfs_fs_info *fs_info = root->fs_info;
6096 struct btrfs_extent_item *extent_item;
6097 struct btrfs_tree_block_info *block_info;
6098 struct btrfs_extent_inline_ref *iref;
6099 struct btrfs_path *path;
6100 struct extent_buffer *leaf;
6101 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6103 path = btrfs_alloc_path();
6107 path->leave_spinning = 1;
6108 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6111 btrfs_free_path(path);
6115 leaf = path->nodes[0];
6116 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6117 struct btrfs_extent_item);
6118 btrfs_set_extent_refs(leaf, extent_item, 1);
6119 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6120 btrfs_set_extent_flags(leaf, extent_item,
6121 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6122 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6124 btrfs_set_tree_block_key(leaf, block_info, key);
6125 btrfs_set_tree_block_level(leaf, block_info, level);
6127 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6129 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6130 btrfs_set_extent_inline_ref_type(leaf, iref,
6131 BTRFS_SHARED_BLOCK_REF_KEY);
6132 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6134 btrfs_set_extent_inline_ref_type(leaf, iref,
6135 BTRFS_TREE_BLOCK_REF_KEY);
6136 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6139 btrfs_mark_buffer_dirty(leaf);
6140 btrfs_free_path(path);
6142 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6143 if (ret) { /* -ENOENT, logic error */
6144 printk(KERN_ERR "btrfs update block group failed for %llu "
6145 "%llu\n", (unsigned long long)ins->objectid,
6146 (unsigned long long)ins->offset);
6152 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6153 struct btrfs_root *root,
6154 u64 root_objectid, u64 owner,
6155 u64 offset, struct btrfs_key *ins)
6159 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6161 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6163 root_objectid, owner, offset,
6164 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6169 * this is used by the tree logging recovery code. It records that
6170 * an extent has been allocated and makes sure to clear the free
6171 * space cache bits as well
6173 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6174 struct btrfs_root *root,
6175 u64 root_objectid, u64 owner, u64 offset,
6176 struct btrfs_key *ins)
6179 struct btrfs_block_group_cache *block_group;
6180 struct btrfs_caching_control *caching_ctl;
6181 u64 start = ins->objectid;
6182 u64 num_bytes = ins->offset;
6184 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6185 cache_block_group(block_group, trans, NULL, 0);
6186 caching_ctl = get_caching_control(block_group);
6189 BUG_ON(!block_group_cache_done(block_group));
6190 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6191 BUG_ON(ret); /* -ENOMEM */
6193 mutex_lock(&caching_ctl->mutex);
6195 if (start >= caching_ctl->progress) {
6196 ret = add_excluded_extent(root, start, num_bytes);
6197 BUG_ON(ret); /* -ENOMEM */
6198 } else if (start + num_bytes <= caching_ctl->progress) {
6199 ret = btrfs_remove_free_space(block_group,
6201 BUG_ON(ret); /* -ENOMEM */
6203 num_bytes = caching_ctl->progress - start;
6204 ret = btrfs_remove_free_space(block_group,
6206 BUG_ON(ret); /* -ENOMEM */
6208 start = caching_ctl->progress;
6209 num_bytes = ins->objectid + ins->offset -
6210 caching_ctl->progress;
6211 ret = add_excluded_extent(root, start, num_bytes);
6212 BUG_ON(ret); /* -ENOMEM */
6215 mutex_unlock(&caching_ctl->mutex);
6216 put_caching_control(caching_ctl);
6219 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6220 RESERVE_ALLOC_NO_ACCOUNT);
6221 BUG_ON(ret); /* logic error */
6222 btrfs_put_block_group(block_group);
6223 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6224 0, owner, offset, ins, 1);
6228 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6229 struct btrfs_root *root,
6230 u64 bytenr, u32 blocksize,
6233 struct extent_buffer *buf;
6235 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6237 return ERR_PTR(-ENOMEM);
6238 btrfs_set_header_generation(buf, trans->transid);
6239 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6240 btrfs_tree_lock(buf);
6241 clean_tree_block(trans, root, buf);
6242 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6244 btrfs_set_lock_blocking(buf);
6245 btrfs_set_buffer_uptodate(buf);
6247 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6249 * we allow two log transactions at a time, use different
6250 * EXENT bit to differentiate dirty pages.
6252 if (root->log_transid % 2 == 0)
6253 set_extent_dirty(&root->dirty_log_pages, buf->start,
6254 buf->start + buf->len - 1, GFP_NOFS);
6256 set_extent_new(&root->dirty_log_pages, buf->start,
6257 buf->start + buf->len - 1, GFP_NOFS);
6259 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6260 buf->start + buf->len - 1, GFP_NOFS);
6262 trans->blocks_used++;
6263 /* this returns a buffer locked for blocking */
6267 static struct btrfs_block_rsv *
6268 use_block_rsv(struct btrfs_trans_handle *trans,
6269 struct btrfs_root *root, u32 blocksize)
6271 struct btrfs_block_rsv *block_rsv;
6272 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6275 block_rsv = get_block_rsv(trans, root);
6277 if (block_rsv->size == 0) {
6278 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6279 BTRFS_RESERVE_NO_FLUSH);
6281 * If we couldn't reserve metadata bytes try and use some from
6282 * the global reserve.
6284 if (ret && block_rsv != global_rsv) {
6285 ret = block_rsv_use_bytes(global_rsv, blocksize);
6288 return ERR_PTR(ret);
6290 return ERR_PTR(ret);
6295 ret = block_rsv_use_bytes(block_rsv, blocksize);
6298 if (ret && !block_rsv->failfast) {
6299 static DEFINE_RATELIMIT_STATE(_rs,
6300 DEFAULT_RATELIMIT_INTERVAL,
6301 /*DEFAULT_RATELIMIT_BURST*/ 2);
6302 if (__ratelimit(&_rs))
6303 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6305 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6306 BTRFS_RESERVE_NO_FLUSH);
6309 } else if (ret && block_rsv != global_rsv) {
6310 ret = block_rsv_use_bytes(global_rsv, blocksize);
6316 return ERR_PTR(-ENOSPC);
6319 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6320 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6322 block_rsv_add_bytes(block_rsv, blocksize, 0);
6323 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6327 * finds a free extent and does all the dirty work required for allocation
6328 * returns the key for the extent through ins, and a tree buffer for
6329 * the first block of the extent through buf.
6331 * returns the tree buffer or NULL.
6333 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6334 struct btrfs_root *root, u32 blocksize,
6335 u64 parent, u64 root_objectid,
6336 struct btrfs_disk_key *key, int level,
6337 u64 hint, u64 empty_size)
6339 struct btrfs_key ins;
6340 struct btrfs_block_rsv *block_rsv;
6341 struct extent_buffer *buf;
6346 block_rsv = use_block_rsv(trans, root, blocksize);
6347 if (IS_ERR(block_rsv))
6348 return ERR_CAST(block_rsv);
6350 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6351 empty_size, hint, &ins, 0);
6353 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6354 return ERR_PTR(ret);
6357 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6359 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6361 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6363 parent = ins.objectid;
6364 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6368 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6369 struct btrfs_delayed_extent_op *extent_op;
6370 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6371 BUG_ON(!extent_op); /* -ENOMEM */
6373 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6375 memset(&extent_op->key, 0, sizeof(extent_op->key));
6376 extent_op->flags_to_set = flags;
6377 extent_op->update_key = 1;
6378 extent_op->update_flags = 1;
6379 extent_op->is_data = 0;
6381 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6383 ins.offset, parent, root_objectid,
6384 level, BTRFS_ADD_DELAYED_EXTENT,
6386 BUG_ON(ret); /* -ENOMEM */
6391 struct walk_control {
6392 u64 refs[BTRFS_MAX_LEVEL];
6393 u64 flags[BTRFS_MAX_LEVEL];
6394 struct btrfs_key update_progress;
6405 #define DROP_REFERENCE 1
6406 #define UPDATE_BACKREF 2
6408 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6409 struct btrfs_root *root,
6410 struct walk_control *wc,
6411 struct btrfs_path *path)
6419 struct btrfs_key key;
6420 struct extent_buffer *eb;
6425 if (path->slots[wc->level] < wc->reada_slot) {
6426 wc->reada_count = wc->reada_count * 2 / 3;
6427 wc->reada_count = max(wc->reada_count, 2);
6429 wc->reada_count = wc->reada_count * 3 / 2;
6430 wc->reada_count = min_t(int, wc->reada_count,
6431 BTRFS_NODEPTRS_PER_BLOCK(root));
6434 eb = path->nodes[wc->level];
6435 nritems = btrfs_header_nritems(eb);
6436 blocksize = btrfs_level_size(root, wc->level - 1);
6438 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6439 if (nread >= wc->reada_count)
6443 bytenr = btrfs_node_blockptr(eb, slot);
6444 generation = btrfs_node_ptr_generation(eb, slot);
6446 if (slot == path->slots[wc->level])
6449 if (wc->stage == UPDATE_BACKREF &&
6450 generation <= root->root_key.offset)
6453 /* We don't lock the tree block, it's OK to be racy here */
6454 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6456 /* We don't care about errors in readahead. */
6461 if (wc->stage == DROP_REFERENCE) {
6465 if (wc->level == 1 &&
6466 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6468 if (!wc->update_ref ||
6469 generation <= root->root_key.offset)
6471 btrfs_node_key_to_cpu(eb, &key, slot);
6472 ret = btrfs_comp_cpu_keys(&key,
6473 &wc->update_progress);
6477 if (wc->level == 1 &&
6478 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6482 ret = readahead_tree_block(root, bytenr, blocksize,
6488 wc->reada_slot = slot;
6492 * hepler to process tree block while walking down the tree.
6494 * when wc->stage == UPDATE_BACKREF, this function updates
6495 * back refs for pointers in the block.
6497 * NOTE: return value 1 means we should stop walking down.
6499 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6500 struct btrfs_root *root,
6501 struct btrfs_path *path,
6502 struct walk_control *wc, int lookup_info)
6504 int level = wc->level;
6505 struct extent_buffer *eb = path->nodes[level];
6506 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6509 if (wc->stage == UPDATE_BACKREF &&
6510 btrfs_header_owner(eb) != root->root_key.objectid)
6514 * when reference count of tree block is 1, it won't increase
6515 * again. once full backref flag is set, we never clear it.
6518 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6519 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6520 BUG_ON(!path->locks[level]);
6521 ret = btrfs_lookup_extent_info(trans, root,
6525 BUG_ON(ret == -ENOMEM);
6528 BUG_ON(wc->refs[level] == 0);
6531 if (wc->stage == DROP_REFERENCE) {
6532 if (wc->refs[level] > 1)
6535 if (path->locks[level] && !wc->keep_locks) {
6536 btrfs_tree_unlock_rw(eb, path->locks[level]);
6537 path->locks[level] = 0;
6542 /* wc->stage == UPDATE_BACKREF */
6543 if (!(wc->flags[level] & flag)) {
6544 BUG_ON(!path->locks[level]);
6545 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6546 BUG_ON(ret); /* -ENOMEM */
6547 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6548 BUG_ON(ret); /* -ENOMEM */
6549 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6551 BUG_ON(ret); /* -ENOMEM */
6552 wc->flags[level] |= flag;
6556 * the block is shared by multiple trees, so it's not good to
6557 * keep the tree lock
6559 if (path->locks[level] && level > 0) {
6560 btrfs_tree_unlock_rw(eb, path->locks[level]);
6561 path->locks[level] = 0;
6567 * hepler to process tree block pointer.
6569 * when wc->stage == DROP_REFERENCE, this function checks
6570 * reference count of the block pointed to. if the block
6571 * is shared and we need update back refs for the subtree
6572 * rooted at the block, this function changes wc->stage to
6573 * UPDATE_BACKREF. if the block is shared and there is no
6574 * need to update back, this function drops the reference
6577 * NOTE: return value 1 means we should stop walking down.
6579 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6580 struct btrfs_root *root,
6581 struct btrfs_path *path,
6582 struct walk_control *wc, int *lookup_info)
6588 struct btrfs_key key;
6589 struct extent_buffer *next;
6590 int level = wc->level;
6594 generation = btrfs_node_ptr_generation(path->nodes[level],
6595 path->slots[level]);
6597 * if the lower level block was created before the snapshot
6598 * was created, we know there is no need to update back refs
6601 if (wc->stage == UPDATE_BACKREF &&
6602 generation <= root->root_key.offset) {
6607 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6608 blocksize = btrfs_level_size(root, level - 1);
6610 next = btrfs_find_tree_block(root, bytenr, blocksize);
6612 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6617 btrfs_tree_lock(next);
6618 btrfs_set_lock_blocking(next);
6620 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6621 &wc->refs[level - 1],
6622 &wc->flags[level - 1]);
6624 btrfs_tree_unlock(next);
6628 BUG_ON(wc->refs[level - 1] == 0);
6631 if (wc->stage == DROP_REFERENCE) {
6632 if (wc->refs[level - 1] > 1) {
6634 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6637 if (!wc->update_ref ||
6638 generation <= root->root_key.offset)
6641 btrfs_node_key_to_cpu(path->nodes[level], &key,
6642 path->slots[level]);
6643 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6647 wc->stage = UPDATE_BACKREF;
6648 wc->shared_level = level - 1;
6652 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6656 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6657 btrfs_tree_unlock(next);
6658 free_extent_buffer(next);
6664 if (reada && level == 1)
6665 reada_walk_down(trans, root, wc, path);
6666 next = read_tree_block(root, bytenr, blocksize, generation);
6669 btrfs_tree_lock(next);
6670 btrfs_set_lock_blocking(next);
6674 BUG_ON(level != btrfs_header_level(next));
6675 path->nodes[level] = next;
6676 path->slots[level] = 0;
6677 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6683 wc->refs[level - 1] = 0;
6684 wc->flags[level - 1] = 0;
6685 if (wc->stage == DROP_REFERENCE) {
6686 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6687 parent = path->nodes[level]->start;
6689 BUG_ON(root->root_key.objectid !=
6690 btrfs_header_owner(path->nodes[level]));
6694 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6695 root->root_key.objectid, level - 1, 0, 0);
6696 BUG_ON(ret); /* -ENOMEM */
6698 btrfs_tree_unlock(next);
6699 free_extent_buffer(next);
6705 * hepler to process tree block while walking up the tree.
6707 * when wc->stage == DROP_REFERENCE, this function drops
6708 * reference count on the block.
6710 * when wc->stage == UPDATE_BACKREF, this function changes
6711 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6712 * to UPDATE_BACKREF previously while processing the block.
6714 * NOTE: return value 1 means we should stop walking up.
6716 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6717 struct btrfs_root *root,
6718 struct btrfs_path *path,
6719 struct walk_control *wc)
6722 int level = wc->level;
6723 struct extent_buffer *eb = path->nodes[level];
6726 if (wc->stage == UPDATE_BACKREF) {
6727 BUG_ON(wc->shared_level < level);
6728 if (level < wc->shared_level)
6731 ret = find_next_key(path, level + 1, &wc->update_progress);
6735 wc->stage = DROP_REFERENCE;
6736 wc->shared_level = -1;
6737 path->slots[level] = 0;
6740 * check reference count again if the block isn't locked.
6741 * we should start walking down the tree again if reference
6744 if (!path->locks[level]) {
6746 btrfs_tree_lock(eb);
6747 btrfs_set_lock_blocking(eb);
6748 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6750 ret = btrfs_lookup_extent_info(trans, root,
6755 btrfs_tree_unlock_rw(eb, path->locks[level]);
6758 BUG_ON(wc->refs[level] == 0);
6759 if (wc->refs[level] == 1) {
6760 btrfs_tree_unlock_rw(eb, path->locks[level]);
6766 /* wc->stage == DROP_REFERENCE */
6767 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6769 if (wc->refs[level] == 1) {
6771 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6772 ret = btrfs_dec_ref(trans, root, eb, 1,
6775 ret = btrfs_dec_ref(trans, root, eb, 0,
6777 BUG_ON(ret); /* -ENOMEM */
6779 /* make block locked assertion in clean_tree_block happy */
6780 if (!path->locks[level] &&
6781 btrfs_header_generation(eb) == trans->transid) {
6782 btrfs_tree_lock(eb);
6783 btrfs_set_lock_blocking(eb);
6784 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6786 clean_tree_block(trans, root, eb);
6789 if (eb == root->node) {
6790 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6793 BUG_ON(root->root_key.objectid !=
6794 btrfs_header_owner(eb));
6796 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6797 parent = path->nodes[level + 1]->start;
6799 BUG_ON(root->root_key.objectid !=
6800 btrfs_header_owner(path->nodes[level + 1]));
6803 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6805 wc->refs[level] = 0;
6806 wc->flags[level] = 0;
6810 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6811 struct btrfs_root *root,
6812 struct btrfs_path *path,
6813 struct walk_control *wc)
6815 int level = wc->level;
6816 int lookup_info = 1;
6819 while (level >= 0) {
6820 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6827 if (path->slots[level] >=
6828 btrfs_header_nritems(path->nodes[level]))
6831 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6833 path->slots[level]++;
6842 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6843 struct btrfs_root *root,
6844 struct btrfs_path *path,
6845 struct walk_control *wc, int max_level)
6847 int level = wc->level;
6850 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6851 while (level < max_level && path->nodes[level]) {
6853 if (path->slots[level] + 1 <
6854 btrfs_header_nritems(path->nodes[level])) {
6855 path->slots[level]++;
6858 ret = walk_up_proc(trans, root, path, wc);
6862 if (path->locks[level]) {
6863 btrfs_tree_unlock_rw(path->nodes[level],
6864 path->locks[level]);
6865 path->locks[level] = 0;
6867 free_extent_buffer(path->nodes[level]);
6868 path->nodes[level] = NULL;
6876 * drop a subvolume tree.
6878 * this function traverses the tree freeing any blocks that only
6879 * referenced by the tree.
6881 * when a shared tree block is found. this function decreases its
6882 * reference count by one. if update_ref is true, this function
6883 * also make sure backrefs for the shared block and all lower level
6884 * blocks are properly updated.
6886 int btrfs_drop_snapshot(struct btrfs_root *root,
6887 struct btrfs_block_rsv *block_rsv, int update_ref,
6890 struct btrfs_path *path;
6891 struct btrfs_trans_handle *trans;
6892 struct btrfs_root *tree_root = root->fs_info->tree_root;
6893 struct btrfs_root_item *root_item = &root->root_item;
6894 struct walk_control *wc;
6895 struct btrfs_key key;
6900 path = btrfs_alloc_path();
6906 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6908 btrfs_free_path(path);
6913 trans = btrfs_start_transaction(tree_root, 0);
6914 if (IS_ERR(trans)) {
6915 err = PTR_ERR(trans);
6920 trans->block_rsv = block_rsv;
6922 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6923 level = btrfs_header_level(root->node);
6924 path->nodes[level] = btrfs_lock_root_node(root);
6925 btrfs_set_lock_blocking(path->nodes[level]);
6926 path->slots[level] = 0;
6927 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6928 memset(&wc->update_progress, 0,
6929 sizeof(wc->update_progress));
6931 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6932 memcpy(&wc->update_progress, &key,
6933 sizeof(wc->update_progress));
6935 level = root_item->drop_level;
6937 path->lowest_level = level;
6938 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6939 path->lowest_level = 0;
6947 * unlock our path, this is safe because only this
6948 * function is allowed to delete this snapshot
6950 btrfs_unlock_up_safe(path, 0);
6952 level = btrfs_header_level(root->node);
6954 btrfs_tree_lock(path->nodes[level]);
6955 btrfs_set_lock_blocking(path->nodes[level]);
6957 ret = btrfs_lookup_extent_info(trans, root,
6958 path->nodes[level]->start,
6959 path->nodes[level]->len,
6966 BUG_ON(wc->refs[level] == 0);
6968 if (level == root_item->drop_level)
6971 btrfs_tree_unlock(path->nodes[level]);
6972 WARN_ON(wc->refs[level] != 1);
6978 wc->shared_level = -1;
6979 wc->stage = DROP_REFERENCE;
6980 wc->update_ref = update_ref;
6982 wc->for_reloc = for_reloc;
6983 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6986 ret = walk_down_tree(trans, root, path, wc);
6992 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6999 BUG_ON(wc->stage != DROP_REFERENCE);
7003 if (wc->stage == DROP_REFERENCE) {
7005 btrfs_node_key(path->nodes[level],
7006 &root_item->drop_progress,
7007 path->slots[level]);
7008 root_item->drop_level = level;
7011 BUG_ON(wc->level == 0);
7012 if (btrfs_should_end_transaction(trans, tree_root)) {
7013 ret = btrfs_update_root(trans, tree_root,
7017 btrfs_abort_transaction(trans, tree_root, ret);
7022 btrfs_end_transaction_throttle(trans, tree_root);
7023 trans = btrfs_start_transaction(tree_root, 0);
7024 if (IS_ERR(trans)) {
7025 err = PTR_ERR(trans);
7029 trans->block_rsv = block_rsv;
7032 btrfs_release_path(path);
7036 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7038 btrfs_abort_transaction(trans, tree_root, ret);
7042 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7043 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7046 btrfs_abort_transaction(trans, tree_root, ret);
7049 } else if (ret > 0) {
7050 /* if we fail to delete the orphan item this time
7051 * around, it'll get picked up the next time.
7053 * The most common failure here is just -ENOENT.
7055 btrfs_del_orphan_item(trans, tree_root,
7056 root->root_key.objectid);
7060 if (root->in_radix) {
7061 btrfs_free_fs_root(tree_root->fs_info, root);
7063 free_extent_buffer(root->node);
7064 free_extent_buffer(root->commit_root);
7068 btrfs_end_transaction_throttle(trans, tree_root);
7071 btrfs_free_path(path);
7074 btrfs_std_error(root->fs_info, err);
7079 * drop subtree rooted at tree block 'node'.
7081 * NOTE: this function will unlock and release tree block 'node'
7082 * only used by relocation code
7084 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7085 struct btrfs_root *root,
7086 struct extent_buffer *node,
7087 struct extent_buffer *parent)
7089 struct btrfs_path *path;
7090 struct walk_control *wc;
7096 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7098 path = btrfs_alloc_path();
7102 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7104 btrfs_free_path(path);
7108 btrfs_assert_tree_locked(parent);
7109 parent_level = btrfs_header_level(parent);
7110 extent_buffer_get(parent);
7111 path->nodes[parent_level] = parent;
7112 path->slots[parent_level] = btrfs_header_nritems(parent);
7114 btrfs_assert_tree_locked(node);
7115 level = btrfs_header_level(node);
7116 path->nodes[level] = node;
7117 path->slots[level] = 0;
7118 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7120 wc->refs[parent_level] = 1;
7121 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7123 wc->shared_level = -1;
7124 wc->stage = DROP_REFERENCE;
7128 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7131 wret = walk_down_tree(trans, root, path, wc);
7137 wret = walk_up_tree(trans, root, path, wc, parent_level);
7145 btrfs_free_path(path);
7149 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7155 * if restripe for this chunk_type is on pick target profile and
7156 * return, otherwise do the usual balance
7158 stripped = get_restripe_target(root->fs_info, flags);
7160 return extended_to_chunk(stripped);
7163 * we add in the count of missing devices because we want
7164 * to make sure that any RAID levels on a degraded FS
7165 * continue to be honored.
7167 num_devices = root->fs_info->fs_devices->rw_devices +
7168 root->fs_info->fs_devices->missing_devices;
7170 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7171 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7173 if (num_devices == 1) {
7174 stripped |= BTRFS_BLOCK_GROUP_DUP;
7175 stripped = flags & ~stripped;
7177 /* turn raid0 into single device chunks */
7178 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7181 /* turn mirroring into duplication */
7182 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7183 BTRFS_BLOCK_GROUP_RAID10))
7184 return stripped | BTRFS_BLOCK_GROUP_DUP;
7186 /* they already had raid on here, just return */
7187 if (flags & stripped)
7190 stripped |= BTRFS_BLOCK_GROUP_DUP;
7191 stripped = flags & ~stripped;
7193 /* switch duplicated blocks with raid1 */
7194 if (flags & BTRFS_BLOCK_GROUP_DUP)
7195 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7197 /* this is drive concat, leave it alone */
7203 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7205 struct btrfs_space_info *sinfo = cache->space_info;
7207 u64 min_allocable_bytes;
7212 * We need some metadata space and system metadata space for
7213 * allocating chunks in some corner cases until we force to set
7214 * it to be readonly.
7217 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7219 min_allocable_bytes = 1 * 1024 * 1024;
7221 min_allocable_bytes = 0;
7223 spin_lock(&sinfo->lock);
7224 spin_lock(&cache->lock);
7231 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7232 cache->bytes_super - btrfs_block_group_used(&cache->item);
7234 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7235 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7236 min_allocable_bytes <= sinfo->total_bytes) {
7237 sinfo->bytes_readonly += num_bytes;
7242 spin_unlock(&cache->lock);
7243 spin_unlock(&sinfo->lock);
7247 int btrfs_set_block_group_ro(struct btrfs_root *root,
7248 struct btrfs_block_group_cache *cache)
7251 struct btrfs_trans_handle *trans;
7257 trans = btrfs_join_transaction(root);
7259 return PTR_ERR(trans);
7261 alloc_flags = update_block_group_flags(root, cache->flags);
7262 if (alloc_flags != cache->flags) {
7263 ret = do_chunk_alloc(trans, root, alloc_flags,
7269 ret = set_block_group_ro(cache, 0);
7272 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7273 ret = do_chunk_alloc(trans, root, alloc_flags,
7277 ret = set_block_group_ro(cache, 0);
7279 btrfs_end_transaction(trans, root);
7283 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7284 struct btrfs_root *root, u64 type)
7286 u64 alloc_flags = get_alloc_profile(root, type);
7287 return do_chunk_alloc(trans, root, alloc_flags,
7292 * helper to account the unused space of all the readonly block group in the
7293 * list. takes mirrors into account.
7295 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7297 struct btrfs_block_group_cache *block_group;
7301 list_for_each_entry(block_group, groups_list, list) {
7302 spin_lock(&block_group->lock);
7304 if (!block_group->ro) {
7305 spin_unlock(&block_group->lock);
7309 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7310 BTRFS_BLOCK_GROUP_RAID10 |
7311 BTRFS_BLOCK_GROUP_DUP))
7316 free_bytes += (block_group->key.offset -
7317 btrfs_block_group_used(&block_group->item)) *
7320 spin_unlock(&block_group->lock);
7327 * helper to account the unused space of all the readonly block group in the
7328 * space_info. takes mirrors into account.
7330 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7335 spin_lock(&sinfo->lock);
7337 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7338 if (!list_empty(&sinfo->block_groups[i]))
7339 free_bytes += __btrfs_get_ro_block_group_free_space(
7340 &sinfo->block_groups[i]);
7342 spin_unlock(&sinfo->lock);
7347 void btrfs_set_block_group_rw(struct btrfs_root *root,
7348 struct btrfs_block_group_cache *cache)
7350 struct btrfs_space_info *sinfo = cache->space_info;
7355 spin_lock(&sinfo->lock);
7356 spin_lock(&cache->lock);
7357 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7358 cache->bytes_super - btrfs_block_group_used(&cache->item);
7359 sinfo->bytes_readonly -= num_bytes;
7361 spin_unlock(&cache->lock);
7362 spin_unlock(&sinfo->lock);
7366 * checks to see if its even possible to relocate this block group.
7368 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7369 * ok to go ahead and try.
7371 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7373 struct btrfs_block_group_cache *block_group;
7374 struct btrfs_space_info *space_info;
7375 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7376 struct btrfs_device *device;
7385 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7387 /* odd, couldn't find the block group, leave it alone */
7391 min_free = btrfs_block_group_used(&block_group->item);
7393 /* no bytes used, we're good */
7397 space_info = block_group->space_info;
7398 spin_lock(&space_info->lock);
7400 full = space_info->full;
7403 * if this is the last block group we have in this space, we can't
7404 * relocate it unless we're able to allocate a new chunk below.
7406 * Otherwise, we need to make sure we have room in the space to handle
7407 * all of the extents from this block group. If we can, we're good
7409 if ((space_info->total_bytes != block_group->key.offset) &&
7410 (space_info->bytes_used + space_info->bytes_reserved +
7411 space_info->bytes_pinned + space_info->bytes_readonly +
7412 min_free < space_info->total_bytes)) {
7413 spin_unlock(&space_info->lock);
7416 spin_unlock(&space_info->lock);
7419 * ok we don't have enough space, but maybe we have free space on our
7420 * devices to allocate new chunks for relocation, so loop through our
7421 * alloc devices and guess if we have enough space. if this block
7422 * group is going to be restriped, run checks against the target
7423 * profile instead of the current one.
7435 target = get_restripe_target(root->fs_info, block_group->flags);
7437 index = __get_block_group_index(extended_to_chunk(target));
7440 * this is just a balance, so if we were marked as full
7441 * we know there is no space for a new chunk
7446 index = get_block_group_index(block_group);
7453 } else if (index == 1) {
7455 } else if (index == 2) {
7458 } else if (index == 3) {
7459 dev_min = fs_devices->rw_devices;
7460 do_div(min_free, dev_min);
7463 mutex_lock(&root->fs_info->chunk_mutex);
7464 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7468 * check to make sure we can actually find a chunk with enough
7469 * space to fit our block group in.
7471 if (device->total_bytes > device->bytes_used + min_free) {
7472 ret = find_free_dev_extent(device, min_free,
7477 if (dev_nr >= dev_min)
7483 mutex_unlock(&root->fs_info->chunk_mutex);
7485 btrfs_put_block_group(block_group);
7489 static int find_first_block_group(struct btrfs_root *root,
7490 struct btrfs_path *path, struct btrfs_key *key)
7493 struct btrfs_key found_key;
7494 struct extent_buffer *leaf;
7497 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7502 slot = path->slots[0];
7503 leaf = path->nodes[0];
7504 if (slot >= btrfs_header_nritems(leaf)) {
7505 ret = btrfs_next_leaf(root, path);
7512 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7514 if (found_key.objectid >= key->objectid &&
7515 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7525 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7527 struct btrfs_block_group_cache *block_group;
7531 struct inode *inode;
7533 block_group = btrfs_lookup_first_block_group(info, last);
7534 while (block_group) {
7535 spin_lock(&block_group->lock);
7536 if (block_group->iref)
7538 spin_unlock(&block_group->lock);
7539 block_group = next_block_group(info->tree_root,
7549 inode = block_group->inode;
7550 block_group->iref = 0;
7551 block_group->inode = NULL;
7552 spin_unlock(&block_group->lock);
7554 last = block_group->key.objectid + block_group->key.offset;
7555 btrfs_put_block_group(block_group);
7559 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7561 struct btrfs_block_group_cache *block_group;
7562 struct btrfs_space_info *space_info;
7563 struct btrfs_caching_control *caching_ctl;
7566 down_write(&info->extent_commit_sem);
7567 while (!list_empty(&info->caching_block_groups)) {
7568 caching_ctl = list_entry(info->caching_block_groups.next,
7569 struct btrfs_caching_control, list);
7570 list_del(&caching_ctl->list);
7571 put_caching_control(caching_ctl);
7573 up_write(&info->extent_commit_sem);
7575 spin_lock(&info->block_group_cache_lock);
7576 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7577 block_group = rb_entry(n, struct btrfs_block_group_cache,
7579 rb_erase(&block_group->cache_node,
7580 &info->block_group_cache_tree);
7581 spin_unlock(&info->block_group_cache_lock);
7583 down_write(&block_group->space_info->groups_sem);
7584 list_del(&block_group->list);
7585 up_write(&block_group->space_info->groups_sem);
7587 if (block_group->cached == BTRFS_CACHE_STARTED)
7588 wait_block_group_cache_done(block_group);
7591 * We haven't cached this block group, which means we could
7592 * possibly have excluded extents on this block group.
7594 if (block_group->cached == BTRFS_CACHE_NO)
7595 free_excluded_extents(info->extent_root, block_group);
7597 btrfs_remove_free_space_cache(block_group);
7598 btrfs_put_block_group(block_group);
7600 spin_lock(&info->block_group_cache_lock);
7602 spin_unlock(&info->block_group_cache_lock);
7604 /* now that all the block groups are freed, go through and
7605 * free all the space_info structs. This is only called during
7606 * the final stages of unmount, and so we know nobody is
7607 * using them. We call synchronize_rcu() once before we start,
7608 * just to be on the safe side.
7612 release_global_block_rsv(info);
7614 while(!list_empty(&info->space_info)) {
7615 space_info = list_entry(info->space_info.next,
7616 struct btrfs_space_info,
7618 if (space_info->bytes_pinned > 0 ||
7619 space_info->bytes_reserved > 0 ||
7620 space_info->bytes_may_use > 0) {
7622 dump_space_info(space_info, 0, 0);
7624 list_del(&space_info->list);
7630 static void __link_block_group(struct btrfs_space_info *space_info,
7631 struct btrfs_block_group_cache *cache)
7633 int index = get_block_group_index(cache);
7635 down_write(&space_info->groups_sem);
7636 list_add_tail(&cache->list, &space_info->block_groups[index]);
7637 up_write(&space_info->groups_sem);
7640 int btrfs_read_block_groups(struct btrfs_root *root)
7642 struct btrfs_path *path;
7644 struct btrfs_block_group_cache *cache;
7645 struct btrfs_fs_info *info = root->fs_info;
7646 struct btrfs_space_info *space_info;
7647 struct btrfs_key key;
7648 struct btrfs_key found_key;
7649 struct extent_buffer *leaf;
7653 root = info->extent_root;
7656 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7657 path = btrfs_alloc_path();
7662 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7663 if (btrfs_test_opt(root, SPACE_CACHE) &&
7664 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7666 if (btrfs_test_opt(root, CLEAR_CACHE))
7670 ret = find_first_block_group(root, path, &key);
7675 leaf = path->nodes[0];
7676 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7677 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7682 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7684 if (!cache->free_space_ctl) {
7690 atomic_set(&cache->count, 1);
7691 spin_lock_init(&cache->lock);
7692 cache->fs_info = info;
7693 INIT_LIST_HEAD(&cache->list);
7694 INIT_LIST_HEAD(&cache->cluster_list);
7698 * When we mount with old space cache, we need to
7699 * set BTRFS_DC_CLEAR and set dirty flag.
7701 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7702 * truncate the old free space cache inode and
7704 * b) Setting 'dirty flag' makes sure that we flush
7705 * the new space cache info onto disk.
7707 cache->disk_cache_state = BTRFS_DC_CLEAR;
7708 if (btrfs_test_opt(root, SPACE_CACHE))
7712 read_extent_buffer(leaf, &cache->item,
7713 btrfs_item_ptr_offset(leaf, path->slots[0]),
7714 sizeof(cache->item));
7715 memcpy(&cache->key, &found_key, sizeof(found_key));
7717 key.objectid = found_key.objectid + found_key.offset;
7718 btrfs_release_path(path);
7719 cache->flags = btrfs_block_group_flags(&cache->item);
7720 cache->sectorsize = root->sectorsize;
7722 btrfs_init_free_space_ctl(cache);
7725 * We need to exclude the super stripes now so that the space
7726 * info has super bytes accounted for, otherwise we'll think
7727 * we have more space than we actually do.
7729 exclude_super_stripes(root, cache);
7732 * check for two cases, either we are full, and therefore
7733 * don't need to bother with the caching work since we won't
7734 * find any space, or we are empty, and we can just add all
7735 * the space in and be done with it. This saves us _alot_ of
7736 * time, particularly in the full case.
7738 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7739 cache->last_byte_to_unpin = (u64)-1;
7740 cache->cached = BTRFS_CACHE_FINISHED;
7741 free_excluded_extents(root, cache);
7742 } else if (btrfs_block_group_used(&cache->item) == 0) {
7743 cache->last_byte_to_unpin = (u64)-1;
7744 cache->cached = BTRFS_CACHE_FINISHED;
7745 add_new_free_space(cache, root->fs_info,
7747 found_key.objectid +
7749 free_excluded_extents(root, cache);
7752 ret = update_space_info(info, cache->flags, found_key.offset,
7753 btrfs_block_group_used(&cache->item),
7755 BUG_ON(ret); /* -ENOMEM */
7756 cache->space_info = space_info;
7757 spin_lock(&cache->space_info->lock);
7758 cache->space_info->bytes_readonly += cache->bytes_super;
7759 spin_unlock(&cache->space_info->lock);
7761 __link_block_group(space_info, cache);
7763 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7764 BUG_ON(ret); /* Logic error */
7766 set_avail_alloc_bits(root->fs_info, cache->flags);
7767 if (btrfs_chunk_readonly(root, cache->key.objectid))
7768 set_block_group_ro(cache, 1);
7771 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7772 if (!(get_alloc_profile(root, space_info->flags) &
7773 (BTRFS_BLOCK_GROUP_RAID10 |
7774 BTRFS_BLOCK_GROUP_RAID1 |
7775 BTRFS_BLOCK_GROUP_DUP)))
7778 * avoid allocating from un-mirrored block group if there are
7779 * mirrored block groups.
7781 list_for_each_entry(cache, &space_info->block_groups[3], list)
7782 set_block_group_ro(cache, 1);
7783 list_for_each_entry(cache, &space_info->block_groups[4], list)
7784 set_block_group_ro(cache, 1);
7787 init_global_block_rsv(info);
7790 btrfs_free_path(path);
7794 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7795 struct btrfs_root *root)
7797 struct btrfs_block_group_cache *block_group, *tmp;
7798 struct btrfs_root *extent_root = root->fs_info->extent_root;
7799 struct btrfs_block_group_item item;
7800 struct btrfs_key key;
7803 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7805 list_del_init(&block_group->new_bg_list);
7810 spin_lock(&block_group->lock);
7811 memcpy(&item, &block_group->item, sizeof(item));
7812 memcpy(&key, &block_group->key, sizeof(key));
7813 spin_unlock(&block_group->lock);
7815 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7818 btrfs_abort_transaction(trans, extent_root, ret);
7822 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7823 struct btrfs_root *root, u64 bytes_used,
7824 u64 type, u64 chunk_objectid, u64 chunk_offset,
7828 struct btrfs_root *extent_root;
7829 struct btrfs_block_group_cache *cache;
7831 extent_root = root->fs_info->extent_root;
7833 root->fs_info->last_trans_log_full_commit = trans->transid;
7835 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7838 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7840 if (!cache->free_space_ctl) {
7845 cache->key.objectid = chunk_offset;
7846 cache->key.offset = size;
7847 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7848 cache->sectorsize = root->sectorsize;
7849 cache->fs_info = root->fs_info;
7851 atomic_set(&cache->count, 1);
7852 spin_lock_init(&cache->lock);
7853 INIT_LIST_HEAD(&cache->list);
7854 INIT_LIST_HEAD(&cache->cluster_list);
7855 INIT_LIST_HEAD(&cache->new_bg_list);
7857 btrfs_init_free_space_ctl(cache);
7859 btrfs_set_block_group_used(&cache->item, bytes_used);
7860 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7861 cache->flags = type;
7862 btrfs_set_block_group_flags(&cache->item, type);
7864 cache->last_byte_to_unpin = (u64)-1;
7865 cache->cached = BTRFS_CACHE_FINISHED;
7866 exclude_super_stripes(root, cache);
7868 add_new_free_space(cache, root->fs_info, chunk_offset,
7869 chunk_offset + size);
7871 free_excluded_extents(root, cache);
7873 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7874 &cache->space_info);
7875 BUG_ON(ret); /* -ENOMEM */
7876 update_global_block_rsv(root->fs_info);
7878 spin_lock(&cache->space_info->lock);
7879 cache->space_info->bytes_readonly += cache->bytes_super;
7880 spin_unlock(&cache->space_info->lock);
7882 __link_block_group(cache->space_info, cache);
7884 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7885 BUG_ON(ret); /* Logic error */
7887 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7889 set_avail_alloc_bits(extent_root->fs_info, type);
7894 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7896 u64 extra_flags = chunk_to_extended(flags) &
7897 BTRFS_EXTENDED_PROFILE_MASK;
7899 if (flags & BTRFS_BLOCK_GROUP_DATA)
7900 fs_info->avail_data_alloc_bits &= ~extra_flags;
7901 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7902 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7903 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7904 fs_info->avail_system_alloc_bits &= ~extra_flags;
7907 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7908 struct btrfs_root *root, u64 group_start)
7910 struct btrfs_path *path;
7911 struct btrfs_block_group_cache *block_group;
7912 struct btrfs_free_cluster *cluster;
7913 struct btrfs_root *tree_root = root->fs_info->tree_root;
7914 struct btrfs_key key;
7915 struct inode *inode;
7920 root = root->fs_info->extent_root;
7922 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7923 BUG_ON(!block_group);
7924 BUG_ON(!block_group->ro);
7927 * Free the reserved super bytes from this block group before
7930 free_excluded_extents(root, block_group);
7932 memcpy(&key, &block_group->key, sizeof(key));
7933 index = get_block_group_index(block_group);
7934 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7935 BTRFS_BLOCK_GROUP_RAID1 |
7936 BTRFS_BLOCK_GROUP_RAID10))
7941 /* make sure this block group isn't part of an allocation cluster */
7942 cluster = &root->fs_info->data_alloc_cluster;
7943 spin_lock(&cluster->refill_lock);
7944 btrfs_return_cluster_to_free_space(block_group, cluster);
7945 spin_unlock(&cluster->refill_lock);
7948 * make sure this block group isn't part of a metadata
7949 * allocation cluster
7951 cluster = &root->fs_info->meta_alloc_cluster;
7952 spin_lock(&cluster->refill_lock);
7953 btrfs_return_cluster_to_free_space(block_group, cluster);
7954 spin_unlock(&cluster->refill_lock);
7956 path = btrfs_alloc_path();
7962 inode = lookup_free_space_inode(tree_root, block_group, path);
7963 if (!IS_ERR(inode)) {
7964 ret = btrfs_orphan_add(trans, inode);
7966 btrfs_add_delayed_iput(inode);
7970 /* One for the block groups ref */
7971 spin_lock(&block_group->lock);
7972 if (block_group->iref) {
7973 block_group->iref = 0;
7974 block_group->inode = NULL;
7975 spin_unlock(&block_group->lock);
7978 spin_unlock(&block_group->lock);
7980 /* One for our lookup ref */
7981 btrfs_add_delayed_iput(inode);
7984 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7985 key.offset = block_group->key.objectid;
7988 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7992 btrfs_release_path(path);
7994 ret = btrfs_del_item(trans, tree_root, path);
7997 btrfs_release_path(path);
8000 spin_lock(&root->fs_info->block_group_cache_lock);
8001 rb_erase(&block_group->cache_node,
8002 &root->fs_info->block_group_cache_tree);
8003 spin_unlock(&root->fs_info->block_group_cache_lock);
8005 down_write(&block_group->space_info->groups_sem);
8007 * we must use list_del_init so people can check to see if they
8008 * are still on the list after taking the semaphore
8010 list_del_init(&block_group->list);
8011 if (list_empty(&block_group->space_info->block_groups[index]))
8012 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8013 up_write(&block_group->space_info->groups_sem);
8015 if (block_group->cached == BTRFS_CACHE_STARTED)
8016 wait_block_group_cache_done(block_group);
8018 btrfs_remove_free_space_cache(block_group);
8020 spin_lock(&block_group->space_info->lock);
8021 block_group->space_info->total_bytes -= block_group->key.offset;
8022 block_group->space_info->bytes_readonly -= block_group->key.offset;
8023 block_group->space_info->disk_total -= block_group->key.offset * factor;
8024 spin_unlock(&block_group->space_info->lock);
8026 memcpy(&key, &block_group->key, sizeof(key));
8028 btrfs_clear_space_info_full(root->fs_info);
8030 btrfs_put_block_group(block_group);
8031 btrfs_put_block_group(block_group);
8033 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8039 ret = btrfs_del_item(trans, root, path);
8041 btrfs_free_path(path);
8045 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8047 struct btrfs_space_info *space_info;
8048 struct btrfs_super_block *disk_super;
8054 disk_super = fs_info->super_copy;
8055 if (!btrfs_super_root(disk_super))
8058 features = btrfs_super_incompat_flags(disk_super);
8059 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8062 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8063 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8068 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8069 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8071 flags = BTRFS_BLOCK_GROUP_METADATA;
8072 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8076 flags = BTRFS_BLOCK_GROUP_DATA;
8077 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8083 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8085 return unpin_extent_range(root, start, end);
8088 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8089 u64 num_bytes, u64 *actual_bytes)
8091 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8094 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8096 struct btrfs_fs_info *fs_info = root->fs_info;
8097 struct btrfs_block_group_cache *cache = NULL;
8102 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8106 * try to trim all FS space, our block group may start from non-zero.
8108 if (range->len == total_bytes)
8109 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8111 cache = btrfs_lookup_block_group(fs_info, range->start);
8114 if (cache->key.objectid >= (range->start + range->len)) {
8115 btrfs_put_block_group(cache);
8119 start = max(range->start, cache->key.objectid);
8120 end = min(range->start + range->len,
8121 cache->key.objectid + cache->key.offset);
8123 if (end - start >= range->minlen) {
8124 if (!block_group_cache_done(cache)) {
8125 ret = cache_block_group(cache, NULL, root, 0);
8127 wait_block_group_cache_done(cache);
8129 ret = btrfs_trim_block_group(cache,
8135 trimmed += group_trimmed;
8137 btrfs_put_block_group(cache);
8142 cache = next_block_group(fs_info->tree_root, cache);
8145 range->len = trimmed;