2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
38 #undef SCRAMBLE_DELAYED_REFS
41 * control flags for do_chunk_alloc's force field
42 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
43 * if we really need one.
45 * CHUNK_ALLOC_LIMITED means to only try and allocate one
46 * if we have very few chunks already allocated. This is
47 * used as part of the clustering code to help make sure
48 * we have a good pool of storage to cluster in, without
49 * filling the FS with empty chunks
51 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 CHUNK_ALLOC_NO_FORCE = 0,
56 CHUNK_ALLOC_LIMITED = 1,
57 CHUNK_ALLOC_FORCE = 2,
61 * Control how reservations are dealt with.
63 * RESERVE_FREE - freeing a reservation.
64 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
66 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
67 * bytes_may_use as the ENOSPC accounting is done elsewhere
72 RESERVE_ALLOC_NO_ACCOUNT = 2,
75 static int update_block_group(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
108 block_group_cache_done(struct btrfs_block_group_cache *cache)
111 return cache->cached == BTRFS_CACHE_FINISHED;
114 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
116 return (cache->flags & bits) == bits;
119 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
121 atomic_inc(&cache->count);
124 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
126 if (atomic_dec_and_test(&cache->count)) {
127 WARN_ON(cache->pinned > 0);
128 WARN_ON(cache->reserved > 0);
129 kfree(cache->free_space_ctl);
135 * this adds the block group to the fs_info rb tree for the block group
138 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
139 struct btrfs_block_group_cache *block_group)
142 struct rb_node *parent = NULL;
143 struct btrfs_block_group_cache *cache;
145 spin_lock(&info->block_group_cache_lock);
146 p = &info->block_group_cache_tree.rb_node;
150 cache = rb_entry(parent, struct btrfs_block_group_cache,
152 if (block_group->key.objectid < cache->key.objectid) {
154 } else if (block_group->key.objectid > cache->key.objectid) {
157 spin_unlock(&info->block_group_cache_lock);
162 rb_link_node(&block_group->cache_node, parent, p);
163 rb_insert_color(&block_group->cache_node,
164 &info->block_group_cache_tree);
165 spin_unlock(&info->block_group_cache_lock);
171 * This will return the block group at or after bytenr if contains is 0, else
172 * it will return the block group that contains the bytenr
174 static struct btrfs_block_group_cache *
175 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
178 struct btrfs_block_group_cache *cache, *ret = NULL;
182 spin_lock(&info->block_group_cache_lock);
183 n = info->block_group_cache_tree.rb_node;
186 cache = rb_entry(n, struct btrfs_block_group_cache,
188 end = cache->key.objectid + cache->key.offset - 1;
189 start = cache->key.objectid;
191 if (bytenr < start) {
192 if (!contains && (!ret || start < ret->key.objectid))
195 } else if (bytenr > start) {
196 if (contains && bytenr <= end) {
207 btrfs_get_block_group(ret);
208 spin_unlock(&info->block_group_cache_lock);
213 static int add_excluded_extent(struct btrfs_root *root,
214 u64 start, u64 num_bytes)
216 u64 end = start + num_bytes - 1;
217 set_extent_bits(&root->fs_info->freed_extents[0],
218 start, end, EXTENT_UPTODATE, GFP_NOFS);
219 set_extent_bits(&root->fs_info->freed_extents[1],
220 start, end, EXTENT_UPTODATE, GFP_NOFS);
224 static void free_excluded_extents(struct btrfs_root *root,
225 struct btrfs_block_group_cache *cache)
229 start = cache->key.objectid;
230 end = start + cache->key.offset - 1;
232 clear_extent_bits(&root->fs_info->freed_extents[0],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 clear_extent_bits(&root->fs_info->freed_extents[1],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
238 static int exclude_super_stripes(struct btrfs_root *root,
239 struct btrfs_block_group_cache *cache)
246 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
247 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
248 cache->bytes_super += stripe_len;
249 ret = add_excluded_extent(root, cache->key.objectid,
251 BUG_ON(ret); /* -ENOMEM */
254 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
255 bytenr = btrfs_sb_offset(i);
256 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
257 cache->key.objectid, bytenr,
258 0, &logical, &nr, &stripe_len);
259 BUG_ON(ret); /* -ENOMEM */
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(root, logical[nr],
265 BUG_ON(ret); /* -ENOMEM */
273 static struct btrfs_caching_control *
274 get_caching_control(struct btrfs_block_group_cache *cache)
276 struct btrfs_caching_control *ctl;
278 spin_lock(&cache->lock);
279 if (cache->cached != BTRFS_CACHE_STARTED) {
280 spin_unlock(&cache->lock);
284 /* We're loading it the fast way, so we don't have a caching_ctl. */
285 if (!cache->caching_ctl) {
286 spin_unlock(&cache->lock);
290 ctl = cache->caching_ctl;
291 atomic_inc(&ctl->count);
292 spin_unlock(&cache->lock);
296 static void put_caching_control(struct btrfs_caching_control *ctl)
298 if (atomic_dec_and_test(&ctl->count))
303 * this is only called by cache_block_group, since we could have freed extents
304 * we need to check the pinned_extents for any extents that can't be used yet
305 * since their free space will be released as soon as the transaction commits.
307 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
308 struct btrfs_fs_info *info, u64 start, u64 end)
310 u64 extent_start, extent_end, size, total_added = 0;
313 while (start < end) {
314 ret = find_first_extent_bit(info->pinned_extents, start,
315 &extent_start, &extent_end,
316 EXTENT_DIRTY | EXTENT_UPTODATE,
321 if (extent_start <= start) {
322 start = extent_end + 1;
323 } else if (extent_start > start && extent_start < end) {
324 size = extent_start - start;
326 ret = btrfs_add_free_space(block_group, start,
328 BUG_ON(ret); /* -ENOMEM or logic error */
329 start = extent_end + 1;
338 ret = btrfs_add_free_space(block_group, start, size);
339 BUG_ON(ret); /* -ENOMEM or logic error */
345 static noinline void caching_thread(struct btrfs_work *work)
347 struct btrfs_block_group_cache *block_group;
348 struct btrfs_fs_info *fs_info;
349 struct btrfs_caching_control *caching_ctl;
350 struct btrfs_root *extent_root;
351 struct btrfs_path *path;
352 struct extent_buffer *leaf;
353 struct btrfs_key key;
359 caching_ctl = container_of(work, struct btrfs_caching_control, work);
360 block_group = caching_ctl->block_group;
361 fs_info = block_group->fs_info;
362 extent_root = fs_info->extent_root;
364 path = btrfs_alloc_path();
368 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
371 * We don't want to deadlock with somebody trying to allocate a new
372 * extent for the extent root while also trying to search the extent
373 * root to add free space. So we skip locking and search the commit
374 * root, since its read-only
376 path->skip_locking = 1;
377 path->search_commit_root = 1;
382 key.type = BTRFS_EXTENT_ITEM_KEY;
384 mutex_lock(&caching_ctl->mutex);
385 /* need to make sure the commit_root doesn't disappear */
386 down_read(&fs_info->extent_commit_sem);
388 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
392 leaf = path->nodes[0];
393 nritems = btrfs_header_nritems(leaf);
396 if (btrfs_fs_closing(fs_info) > 1) {
401 if (path->slots[0] < nritems) {
402 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
404 ret = find_next_key(path, 0, &key);
408 if (need_resched() ||
409 btrfs_next_leaf(extent_root, path)) {
410 caching_ctl->progress = last;
411 btrfs_release_path(path);
412 up_read(&fs_info->extent_commit_sem);
413 mutex_unlock(&caching_ctl->mutex);
417 leaf = path->nodes[0];
418 nritems = btrfs_header_nritems(leaf);
422 if (key.objectid < block_group->key.objectid) {
427 if (key.objectid >= block_group->key.objectid +
428 block_group->key.offset)
431 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
432 total_found += add_new_free_space(block_group,
435 last = key.objectid + key.offset;
437 if (total_found > (1024 * 1024 * 2)) {
439 wake_up(&caching_ctl->wait);
446 total_found += add_new_free_space(block_group, fs_info, last,
447 block_group->key.objectid +
448 block_group->key.offset);
449 caching_ctl->progress = (u64)-1;
451 spin_lock(&block_group->lock);
452 block_group->caching_ctl = NULL;
453 block_group->cached = BTRFS_CACHE_FINISHED;
454 spin_unlock(&block_group->lock);
457 btrfs_free_path(path);
458 up_read(&fs_info->extent_commit_sem);
460 free_excluded_extents(extent_root, block_group);
462 mutex_unlock(&caching_ctl->mutex);
464 wake_up(&caching_ctl->wait);
466 put_caching_control(caching_ctl);
467 btrfs_put_block_group(block_group);
470 static int cache_block_group(struct btrfs_block_group_cache *cache,
471 struct btrfs_trans_handle *trans,
472 struct btrfs_root *root,
476 struct btrfs_fs_info *fs_info = cache->fs_info;
477 struct btrfs_caching_control *caching_ctl;
480 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
484 INIT_LIST_HEAD(&caching_ctl->list);
485 mutex_init(&caching_ctl->mutex);
486 init_waitqueue_head(&caching_ctl->wait);
487 caching_ctl->block_group = cache;
488 caching_ctl->progress = cache->key.objectid;
489 atomic_set(&caching_ctl->count, 1);
490 caching_ctl->work.func = caching_thread;
492 spin_lock(&cache->lock);
494 * This should be a rare occasion, but this could happen I think in the
495 * case where one thread starts to load the space cache info, and then
496 * some other thread starts a transaction commit which tries to do an
497 * allocation while the other thread is still loading the space cache
498 * info. The previous loop should have kept us from choosing this block
499 * group, but if we've moved to the state where we will wait on caching
500 * block groups we need to first check if we're doing a fast load here,
501 * so we can wait for it to finish, otherwise we could end up allocating
502 * from a block group who's cache gets evicted for one reason or
505 while (cache->cached == BTRFS_CACHE_FAST) {
506 struct btrfs_caching_control *ctl;
508 ctl = cache->caching_ctl;
509 atomic_inc(&ctl->count);
510 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
511 spin_unlock(&cache->lock);
515 finish_wait(&ctl->wait, &wait);
516 put_caching_control(ctl);
517 spin_lock(&cache->lock);
520 if (cache->cached != BTRFS_CACHE_NO) {
521 spin_unlock(&cache->lock);
525 WARN_ON(cache->caching_ctl);
526 cache->caching_ctl = caching_ctl;
527 cache->cached = BTRFS_CACHE_FAST;
528 spin_unlock(&cache->lock);
531 * We can't do the read from on-disk cache during a commit since we need
532 * to have the normal tree locking. Also if we are currently trying to
533 * allocate blocks for the tree root we can't do the fast caching since
534 * we likely hold important locks.
536 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
537 ret = load_free_space_cache(fs_info, cache);
539 spin_lock(&cache->lock);
541 cache->caching_ctl = NULL;
542 cache->cached = BTRFS_CACHE_FINISHED;
543 cache->last_byte_to_unpin = (u64)-1;
545 if (load_cache_only) {
546 cache->caching_ctl = NULL;
547 cache->cached = BTRFS_CACHE_NO;
549 cache->cached = BTRFS_CACHE_STARTED;
552 spin_unlock(&cache->lock);
553 wake_up(&caching_ctl->wait);
555 put_caching_control(caching_ctl);
556 free_excluded_extents(fs_info->extent_root, cache);
561 * We are not going to do the fast caching, set cached to the
562 * appropriate value and wakeup any waiters.
564 spin_lock(&cache->lock);
565 if (load_cache_only) {
566 cache->caching_ctl = NULL;
567 cache->cached = BTRFS_CACHE_NO;
569 cache->cached = BTRFS_CACHE_STARTED;
571 spin_unlock(&cache->lock);
572 wake_up(&caching_ctl->wait);
575 if (load_cache_only) {
576 put_caching_control(caching_ctl);
580 down_write(&fs_info->extent_commit_sem);
581 atomic_inc(&caching_ctl->count);
582 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
583 up_write(&fs_info->extent_commit_sem);
585 btrfs_get_block_group(cache);
587 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
593 * return the block group that starts at or after bytenr
595 static struct btrfs_block_group_cache *
596 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
598 struct btrfs_block_group_cache *cache;
600 cache = block_group_cache_tree_search(info, bytenr, 0);
606 * return the block group that contains the given bytenr
608 struct btrfs_block_group_cache *btrfs_lookup_block_group(
609 struct btrfs_fs_info *info,
612 struct btrfs_block_group_cache *cache;
614 cache = block_group_cache_tree_search(info, bytenr, 1);
619 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
622 struct list_head *head = &info->space_info;
623 struct btrfs_space_info *found;
625 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
628 list_for_each_entry_rcu(found, head, list) {
629 if (found->flags & flags) {
639 * after adding space to the filesystem, we need to clear the full flags
640 * on all the space infos.
642 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
644 struct list_head *head = &info->space_info;
645 struct btrfs_space_info *found;
648 list_for_each_entry_rcu(found, head, list)
653 u64 btrfs_find_block_group(struct btrfs_root *root,
654 u64 search_start, u64 search_hint, int owner)
656 struct btrfs_block_group_cache *cache;
658 u64 last = max(search_hint, search_start);
665 cache = btrfs_lookup_first_block_group(root->fs_info, last);
669 spin_lock(&cache->lock);
670 last = cache->key.objectid + cache->key.offset;
671 used = btrfs_block_group_used(&cache->item);
673 if ((full_search || !cache->ro) &&
674 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
675 if (used + cache->pinned + cache->reserved <
676 div_factor(cache->key.offset, factor)) {
677 group_start = cache->key.objectid;
678 spin_unlock(&cache->lock);
679 btrfs_put_block_group(cache);
683 spin_unlock(&cache->lock);
684 btrfs_put_block_group(cache);
692 if (!full_search && factor < 10) {
702 /* simple helper to search for an existing extent at a given offset */
703 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
706 struct btrfs_key key;
707 struct btrfs_path *path;
709 path = btrfs_alloc_path();
713 key.objectid = start;
715 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
716 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
718 btrfs_free_path(path);
723 * helper function to lookup reference count and flags of extent.
725 * the head node for delayed ref is used to store the sum of all the
726 * reference count modifications queued up in the rbtree. the head
727 * node may also store the extent flags to set. This way you can check
728 * to see what the reference count and extent flags would be if all of
729 * the delayed refs are not processed.
731 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
732 struct btrfs_root *root, u64 bytenr,
733 u64 num_bytes, u64 *refs, u64 *flags)
735 struct btrfs_delayed_ref_head *head;
736 struct btrfs_delayed_ref_root *delayed_refs;
737 struct btrfs_path *path;
738 struct btrfs_extent_item *ei;
739 struct extent_buffer *leaf;
740 struct btrfs_key key;
746 path = btrfs_alloc_path();
750 key.objectid = bytenr;
751 key.type = BTRFS_EXTENT_ITEM_KEY;
752 key.offset = num_bytes;
754 path->skip_locking = 1;
755 path->search_commit_root = 1;
758 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
764 leaf = path->nodes[0];
765 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
766 if (item_size >= sizeof(*ei)) {
767 ei = btrfs_item_ptr(leaf, path->slots[0],
768 struct btrfs_extent_item);
769 num_refs = btrfs_extent_refs(leaf, ei);
770 extent_flags = btrfs_extent_flags(leaf, ei);
772 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
773 struct btrfs_extent_item_v0 *ei0;
774 BUG_ON(item_size != sizeof(*ei0));
775 ei0 = btrfs_item_ptr(leaf, path->slots[0],
776 struct btrfs_extent_item_v0);
777 num_refs = btrfs_extent_refs_v0(leaf, ei0);
778 /* FIXME: this isn't correct for data */
779 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
784 BUG_ON(num_refs == 0);
794 delayed_refs = &trans->transaction->delayed_refs;
795 spin_lock(&delayed_refs->lock);
796 head = btrfs_find_delayed_ref_head(trans, bytenr);
798 if (!mutex_trylock(&head->mutex)) {
799 atomic_inc(&head->node.refs);
800 spin_unlock(&delayed_refs->lock);
802 btrfs_release_path(path);
805 * Mutex was contended, block until it's released and try
808 mutex_lock(&head->mutex);
809 mutex_unlock(&head->mutex);
810 btrfs_put_delayed_ref(&head->node);
813 if (head->extent_op && head->extent_op->update_flags)
814 extent_flags |= head->extent_op->flags_to_set;
816 BUG_ON(num_refs == 0);
818 num_refs += head->node.ref_mod;
819 mutex_unlock(&head->mutex);
821 spin_unlock(&delayed_refs->lock);
823 WARN_ON(num_refs == 0);
827 *flags = extent_flags;
829 btrfs_free_path(path);
834 * Back reference rules. Back refs have three main goals:
836 * 1) differentiate between all holders of references to an extent so that
837 * when a reference is dropped we can make sure it was a valid reference
838 * before freeing the extent.
840 * 2) Provide enough information to quickly find the holders of an extent
841 * if we notice a given block is corrupted or bad.
843 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
844 * maintenance. This is actually the same as #2, but with a slightly
845 * different use case.
847 * There are two kinds of back refs. The implicit back refs is optimized
848 * for pointers in non-shared tree blocks. For a given pointer in a block,
849 * back refs of this kind provide information about the block's owner tree
850 * and the pointer's key. These information allow us to find the block by
851 * b-tree searching. The full back refs is for pointers in tree blocks not
852 * referenced by their owner trees. The location of tree block is recorded
853 * in the back refs. Actually the full back refs is generic, and can be
854 * used in all cases the implicit back refs is used. The major shortcoming
855 * of the full back refs is its overhead. Every time a tree block gets
856 * COWed, we have to update back refs entry for all pointers in it.
858 * For a newly allocated tree block, we use implicit back refs for
859 * pointers in it. This means most tree related operations only involve
860 * implicit back refs. For a tree block created in old transaction, the
861 * only way to drop a reference to it is COW it. So we can detect the
862 * event that tree block loses its owner tree's reference and do the
863 * back refs conversion.
865 * When a tree block is COW'd through a tree, there are four cases:
867 * The reference count of the block is one and the tree is the block's
868 * owner tree. Nothing to do in this case.
870 * The reference count of the block is one and the tree is not the
871 * block's owner tree. In this case, full back refs is used for pointers
872 * in the block. Remove these full back refs, add implicit back refs for
873 * every pointers in the new block.
875 * The reference count of the block is greater than one and the tree is
876 * the block's owner tree. In this case, implicit back refs is used for
877 * pointers in the block. Add full back refs for every pointers in the
878 * block, increase lower level extents' reference counts. The original
879 * implicit back refs are entailed to the new block.
881 * The reference count of the block is greater than one and the tree is
882 * not the block's owner tree. Add implicit back refs for every pointer in
883 * the new block, increase lower level extents' reference count.
885 * Back Reference Key composing:
887 * The key objectid corresponds to the first byte in the extent,
888 * The key type is used to differentiate between types of back refs.
889 * There are different meanings of the key offset for different types
892 * File extents can be referenced by:
894 * - multiple snapshots, subvolumes, or different generations in one subvol
895 * - different files inside a single subvolume
896 * - different offsets inside a file (bookend extents in file.c)
898 * The extent ref structure for the implicit back refs has fields for:
900 * - Objectid of the subvolume root
901 * - objectid of the file holding the reference
902 * - original offset in the file
903 * - how many bookend extents
905 * The key offset for the implicit back refs is hash of the first
908 * The extent ref structure for the full back refs has field for:
910 * - number of pointers in the tree leaf
912 * The key offset for the implicit back refs is the first byte of
915 * When a file extent is allocated, The implicit back refs is used.
916 * the fields are filled in:
918 * (root_key.objectid, inode objectid, offset in file, 1)
920 * When a file extent is removed file truncation, we find the
921 * corresponding implicit back refs and check the following fields:
923 * (btrfs_header_owner(leaf), inode objectid, offset in file)
925 * Btree extents can be referenced by:
927 * - Different subvolumes
929 * Both the implicit back refs and the full back refs for tree blocks
930 * only consist of key. The key offset for the implicit back refs is
931 * objectid of block's owner tree. The key offset for the full back refs
932 * is the first byte of parent block.
934 * When implicit back refs is used, information about the lowest key and
935 * level of the tree block are required. These information are stored in
936 * tree block info structure.
939 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
940 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
941 struct btrfs_root *root,
942 struct btrfs_path *path,
943 u64 owner, u32 extra_size)
945 struct btrfs_extent_item *item;
946 struct btrfs_extent_item_v0 *ei0;
947 struct btrfs_extent_ref_v0 *ref0;
948 struct btrfs_tree_block_info *bi;
949 struct extent_buffer *leaf;
950 struct btrfs_key key;
951 struct btrfs_key found_key;
952 u32 new_size = sizeof(*item);
956 leaf = path->nodes[0];
957 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
959 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
960 ei0 = btrfs_item_ptr(leaf, path->slots[0],
961 struct btrfs_extent_item_v0);
962 refs = btrfs_extent_refs_v0(leaf, ei0);
964 if (owner == (u64)-1) {
966 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
967 ret = btrfs_next_leaf(root, path);
970 BUG_ON(ret > 0); /* Corruption */
971 leaf = path->nodes[0];
973 btrfs_item_key_to_cpu(leaf, &found_key,
975 BUG_ON(key.objectid != found_key.objectid);
976 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
980 ref0 = btrfs_item_ptr(leaf, path->slots[0],
981 struct btrfs_extent_ref_v0);
982 owner = btrfs_ref_objectid_v0(leaf, ref0);
986 btrfs_release_path(path);
988 if (owner < BTRFS_FIRST_FREE_OBJECTID)
989 new_size += sizeof(*bi);
991 new_size -= sizeof(*ei0);
992 ret = btrfs_search_slot(trans, root, &key, path,
993 new_size + extra_size, 1);
996 BUG_ON(ret); /* Corruption */
998 btrfs_extend_item(trans, root, path, new_size);
1000 leaf = path->nodes[0];
1001 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1002 btrfs_set_extent_refs(leaf, item, refs);
1003 /* FIXME: get real generation */
1004 btrfs_set_extent_generation(leaf, item, 0);
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1006 btrfs_set_extent_flags(leaf, item,
1007 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1008 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1009 bi = (struct btrfs_tree_block_info *)(item + 1);
1010 /* FIXME: get first key of the block */
1011 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1012 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1014 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1016 btrfs_mark_buffer_dirty(leaf);
1021 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1023 u32 high_crc = ~(u32)0;
1024 u32 low_crc = ~(u32)0;
1027 lenum = cpu_to_le64(root_objectid);
1028 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1029 lenum = cpu_to_le64(owner);
1030 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1031 lenum = cpu_to_le64(offset);
1032 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1034 return ((u64)high_crc << 31) ^ (u64)low_crc;
1037 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1038 struct btrfs_extent_data_ref *ref)
1040 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1041 btrfs_extent_data_ref_objectid(leaf, ref),
1042 btrfs_extent_data_ref_offset(leaf, ref));
1045 static int match_extent_data_ref(struct extent_buffer *leaf,
1046 struct btrfs_extent_data_ref *ref,
1047 u64 root_objectid, u64 owner, u64 offset)
1049 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1050 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1051 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1056 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path,
1059 u64 bytenr, u64 parent,
1061 u64 owner, u64 offset)
1063 struct btrfs_key key;
1064 struct btrfs_extent_data_ref *ref;
1065 struct extent_buffer *leaf;
1071 key.objectid = bytenr;
1073 key.type = BTRFS_SHARED_DATA_REF_KEY;
1074 key.offset = parent;
1076 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1077 key.offset = hash_extent_data_ref(root_objectid,
1082 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1092 key.type = BTRFS_EXTENT_REF_V0_KEY;
1093 btrfs_release_path(path);
1094 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 leaf = path->nodes[0];
1106 nritems = btrfs_header_nritems(leaf);
1108 if (path->slots[0] >= nritems) {
1109 ret = btrfs_next_leaf(root, path);
1115 leaf = path->nodes[0];
1116 nritems = btrfs_header_nritems(leaf);
1120 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1121 if (key.objectid != bytenr ||
1122 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1125 ref = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1128 if (match_extent_data_ref(leaf, ref, root_objectid,
1131 btrfs_release_path(path);
1143 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1144 struct btrfs_root *root,
1145 struct btrfs_path *path,
1146 u64 bytenr, u64 parent,
1147 u64 root_objectid, u64 owner,
1148 u64 offset, int refs_to_add)
1150 struct btrfs_key key;
1151 struct extent_buffer *leaf;
1156 key.objectid = bytenr;
1158 key.type = BTRFS_SHARED_DATA_REF_KEY;
1159 key.offset = parent;
1160 size = sizeof(struct btrfs_shared_data_ref);
1162 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1163 key.offset = hash_extent_data_ref(root_objectid,
1165 size = sizeof(struct btrfs_extent_data_ref);
1168 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1169 if (ret && ret != -EEXIST)
1172 leaf = path->nodes[0];
1174 struct btrfs_shared_data_ref *ref;
1175 ref = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_shared_data_ref);
1178 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1180 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1181 num_refs += refs_to_add;
1182 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1185 struct btrfs_extent_data_ref *ref;
1186 while (ret == -EEXIST) {
1187 ref = btrfs_item_ptr(leaf, path->slots[0],
1188 struct btrfs_extent_data_ref);
1189 if (match_extent_data_ref(leaf, ref, root_objectid,
1192 btrfs_release_path(path);
1194 ret = btrfs_insert_empty_item(trans, root, path, &key,
1196 if (ret && ret != -EEXIST)
1199 leaf = path->nodes[0];
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1204 btrfs_set_extent_data_ref_root(leaf, ref,
1206 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1207 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1208 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1210 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1211 num_refs += refs_to_add;
1212 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1215 btrfs_mark_buffer_dirty(leaf);
1218 btrfs_release_path(path);
1222 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_path *path,
1227 struct btrfs_key key;
1228 struct btrfs_extent_data_ref *ref1 = NULL;
1229 struct btrfs_shared_data_ref *ref2 = NULL;
1230 struct extent_buffer *leaf;
1234 leaf = path->nodes[0];
1235 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1237 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1238 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_extent_data_ref);
1240 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1241 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1242 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_shared_data_ref);
1244 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1245 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1246 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1247 struct btrfs_extent_ref_v0 *ref0;
1248 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_ref_v0);
1250 num_refs = btrfs_ref_count_v0(leaf, ref0);
1256 BUG_ON(num_refs < refs_to_drop);
1257 num_refs -= refs_to_drop;
1259 if (num_refs == 0) {
1260 ret = btrfs_del_item(trans, root, path);
1262 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1263 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1264 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1265 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1268 struct btrfs_extent_ref_v0 *ref0;
1269 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_ref_v0);
1271 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1274 btrfs_mark_buffer_dirty(leaf);
1279 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1280 struct btrfs_path *path,
1281 struct btrfs_extent_inline_ref *iref)
1283 struct btrfs_key key;
1284 struct extent_buffer *leaf;
1285 struct btrfs_extent_data_ref *ref1;
1286 struct btrfs_shared_data_ref *ref2;
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1292 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1293 BTRFS_EXTENT_DATA_REF_KEY) {
1294 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1295 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1297 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1300 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1301 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_data_ref);
1303 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1304 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1305 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_shared_data_ref);
1307 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1308 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1310 struct btrfs_extent_ref_v0 *ref0;
1311 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1312 struct btrfs_extent_ref_v0);
1313 num_refs = btrfs_ref_count_v0(leaf, ref0);
1321 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1322 struct btrfs_root *root,
1323 struct btrfs_path *path,
1324 u64 bytenr, u64 parent,
1327 struct btrfs_key key;
1330 key.objectid = bytenr;
1332 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1333 key.offset = parent;
1335 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1336 key.offset = root_objectid;
1339 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1342 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1343 if (ret == -ENOENT && parent) {
1344 btrfs_release_path(path);
1345 key.type = BTRFS_EXTENT_REF_V0_KEY;
1346 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1354 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root,
1356 struct btrfs_path *path,
1357 u64 bytenr, u64 parent,
1360 struct btrfs_key key;
1363 key.objectid = bytenr;
1365 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1366 key.offset = parent;
1368 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1369 key.offset = root_objectid;
1372 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1373 btrfs_release_path(path);
1377 static inline int extent_ref_type(u64 parent, u64 owner)
1380 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1382 type = BTRFS_SHARED_BLOCK_REF_KEY;
1384 type = BTRFS_TREE_BLOCK_REF_KEY;
1387 type = BTRFS_SHARED_DATA_REF_KEY;
1389 type = BTRFS_EXTENT_DATA_REF_KEY;
1394 static int find_next_key(struct btrfs_path *path, int level,
1395 struct btrfs_key *key)
1398 for (; level < BTRFS_MAX_LEVEL; level++) {
1399 if (!path->nodes[level])
1401 if (path->slots[level] + 1 >=
1402 btrfs_header_nritems(path->nodes[level]))
1405 btrfs_item_key_to_cpu(path->nodes[level], key,
1406 path->slots[level] + 1);
1408 btrfs_node_key_to_cpu(path->nodes[level], key,
1409 path->slots[level] + 1);
1416 * look for inline back ref. if back ref is found, *ref_ret is set
1417 * to the address of inline back ref, and 0 is returned.
1419 * if back ref isn't found, *ref_ret is set to the address where it
1420 * should be inserted, and -ENOENT is returned.
1422 * if insert is true and there are too many inline back refs, the path
1423 * points to the extent item, and -EAGAIN is returned.
1425 * NOTE: inline back refs are ordered in the same way that back ref
1426 * items in the tree are ordered.
1428 static noinline_for_stack
1429 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1430 struct btrfs_root *root,
1431 struct btrfs_path *path,
1432 struct btrfs_extent_inline_ref **ref_ret,
1433 u64 bytenr, u64 num_bytes,
1434 u64 parent, u64 root_objectid,
1435 u64 owner, u64 offset, int insert)
1437 struct btrfs_key key;
1438 struct extent_buffer *leaf;
1439 struct btrfs_extent_item *ei;
1440 struct btrfs_extent_inline_ref *iref;
1451 key.objectid = bytenr;
1452 key.type = BTRFS_EXTENT_ITEM_KEY;
1453 key.offset = num_bytes;
1455 want = extent_ref_type(parent, owner);
1457 extra_size = btrfs_extent_inline_ref_size(want);
1458 path->keep_locks = 1;
1461 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1466 if (ret && !insert) {
1470 BUG_ON(ret); /* Corruption */
1472 leaf = path->nodes[0];
1473 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1474 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1475 if (item_size < sizeof(*ei)) {
1480 ret = convert_extent_item_v0(trans, root, path, owner,
1486 leaf = path->nodes[0];
1487 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 BUG_ON(item_size < sizeof(*ei));
1492 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1493 flags = btrfs_extent_flags(leaf, ei);
1495 ptr = (unsigned long)(ei + 1);
1496 end = (unsigned long)ei + item_size;
1498 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1499 ptr += sizeof(struct btrfs_tree_block_info);
1502 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1511 iref = (struct btrfs_extent_inline_ref *)ptr;
1512 type = btrfs_extent_inline_ref_type(leaf, iref);
1516 ptr += btrfs_extent_inline_ref_size(type);
1520 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1521 struct btrfs_extent_data_ref *dref;
1522 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1523 if (match_extent_data_ref(leaf, dref, root_objectid,
1528 if (hash_extent_data_ref_item(leaf, dref) <
1529 hash_extent_data_ref(root_objectid, owner, offset))
1533 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1535 if (parent == ref_offset) {
1539 if (ref_offset < parent)
1542 if (root_objectid == ref_offset) {
1546 if (ref_offset < root_objectid)
1550 ptr += btrfs_extent_inline_ref_size(type);
1552 if (err == -ENOENT && insert) {
1553 if (item_size + extra_size >=
1554 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1559 * To add new inline back ref, we have to make sure
1560 * there is no corresponding back ref item.
1561 * For simplicity, we just do not add new inline back
1562 * ref if there is any kind of item for this block
1564 if (find_next_key(path, 0, &key) == 0 &&
1565 key.objectid == bytenr &&
1566 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1571 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1574 path->keep_locks = 0;
1575 btrfs_unlock_up_safe(path, 1);
1581 * helper to add new inline back ref
1583 static noinline_for_stack
1584 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct btrfs_path *path,
1587 struct btrfs_extent_inline_ref *iref,
1588 u64 parent, u64 root_objectid,
1589 u64 owner, u64 offset, int refs_to_add,
1590 struct btrfs_delayed_extent_op *extent_op)
1592 struct extent_buffer *leaf;
1593 struct btrfs_extent_item *ei;
1596 unsigned long item_offset;
1601 leaf = path->nodes[0];
1602 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1603 item_offset = (unsigned long)iref - (unsigned long)ei;
1605 type = extent_ref_type(parent, owner);
1606 size = btrfs_extent_inline_ref_size(type);
1608 btrfs_extend_item(trans, root, path, size);
1610 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1611 refs = btrfs_extent_refs(leaf, ei);
1612 refs += refs_to_add;
1613 btrfs_set_extent_refs(leaf, ei, refs);
1615 __run_delayed_extent_op(extent_op, leaf, ei);
1617 ptr = (unsigned long)ei + item_offset;
1618 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1619 if (ptr < end - size)
1620 memmove_extent_buffer(leaf, ptr + size, ptr,
1623 iref = (struct btrfs_extent_inline_ref *)ptr;
1624 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1625 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1626 struct btrfs_extent_data_ref *dref;
1627 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1628 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1629 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1630 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1631 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1632 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1633 struct btrfs_shared_data_ref *sref;
1634 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1635 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1636 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1637 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1638 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1640 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1642 btrfs_mark_buffer_dirty(leaf);
1645 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1646 struct btrfs_root *root,
1647 struct btrfs_path *path,
1648 struct btrfs_extent_inline_ref **ref_ret,
1649 u64 bytenr, u64 num_bytes, u64 parent,
1650 u64 root_objectid, u64 owner, u64 offset)
1654 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1655 bytenr, num_bytes, parent,
1656 root_objectid, owner, offset, 0);
1660 btrfs_release_path(path);
1663 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1667 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1668 root_objectid, owner, offset);
1674 * helper to update/remove inline back ref
1676 static noinline_for_stack
1677 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1678 struct btrfs_root *root,
1679 struct btrfs_path *path,
1680 struct btrfs_extent_inline_ref *iref,
1682 struct btrfs_delayed_extent_op *extent_op)
1684 struct extent_buffer *leaf;
1685 struct btrfs_extent_item *ei;
1686 struct btrfs_extent_data_ref *dref = NULL;
1687 struct btrfs_shared_data_ref *sref = NULL;
1695 leaf = path->nodes[0];
1696 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1697 refs = btrfs_extent_refs(leaf, ei);
1698 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1699 refs += refs_to_mod;
1700 btrfs_set_extent_refs(leaf, ei, refs);
1702 __run_delayed_extent_op(extent_op, leaf, ei);
1704 type = btrfs_extent_inline_ref_type(leaf, iref);
1706 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1707 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1708 refs = btrfs_extent_data_ref_count(leaf, dref);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 refs = btrfs_shared_data_ref_count(leaf, sref);
1714 BUG_ON(refs_to_mod != -1);
1717 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1718 refs += refs_to_mod;
1721 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1722 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1724 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1726 size = btrfs_extent_inline_ref_size(type);
1727 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1728 ptr = (unsigned long)iref;
1729 end = (unsigned long)ei + item_size;
1730 if (ptr + size < end)
1731 memmove_extent_buffer(leaf, ptr, ptr + size,
1734 btrfs_truncate_item(trans, root, path, item_size, 1);
1736 btrfs_mark_buffer_dirty(leaf);
1739 static noinline_for_stack
1740 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 u64 bytenr, u64 num_bytes, u64 parent,
1744 u64 root_objectid, u64 owner,
1745 u64 offset, int refs_to_add,
1746 struct btrfs_delayed_extent_op *extent_op)
1748 struct btrfs_extent_inline_ref *iref;
1751 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1752 bytenr, num_bytes, parent,
1753 root_objectid, owner, offset, 1);
1755 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1756 update_inline_extent_backref(trans, root, path, iref,
1757 refs_to_add, extent_op);
1758 } else if (ret == -ENOENT) {
1759 setup_inline_extent_backref(trans, root, path, iref, parent,
1760 root_objectid, owner, offset,
1761 refs_to_add, extent_op);
1767 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1768 struct btrfs_root *root,
1769 struct btrfs_path *path,
1770 u64 bytenr, u64 parent, u64 root_objectid,
1771 u64 owner, u64 offset, int refs_to_add)
1774 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1775 BUG_ON(refs_to_add != 1);
1776 ret = insert_tree_block_ref(trans, root, path, bytenr,
1777 parent, root_objectid);
1779 ret = insert_extent_data_ref(trans, root, path, bytenr,
1780 parent, root_objectid,
1781 owner, offset, refs_to_add);
1786 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1787 struct btrfs_root *root,
1788 struct btrfs_path *path,
1789 struct btrfs_extent_inline_ref *iref,
1790 int refs_to_drop, int is_data)
1794 BUG_ON(!is_data && refs_to_drop != 1);
1796 update_inline_extent_backref(trans, root, path, iref,
1797 -refs_to_drop, NULL);
1798 } else if (is_data) {
1799 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1801 ret = btrfs_del_item(trans, root, path);
1806 static int btrfs_issue_discard(struct block_device *bdev,
1809 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1812 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1813 u64 num_bytes, u64 *actual_bytes)
1816 u64 discarded_bytes = 0;
1817 struct btrfs_bio *bbio = NULL;
1820 /* Tell the block device(s) that the sectors can be discarded */
1821 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1822 bytenr, &num_bytes, &bbio, 0);
1823 /* Error condition is -ENOMEM */
1825 struct btrfs_bio_stripe *stripe = bbio->stripes;
1829 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1830 if (!stripe->dev->can_discard)
1833 ret = btrfs_issue_discard(stripe->dev->bdev,
1837 discarded_bytes += stripe->length;
1838 else if (ret != -EOPNOTSUPP)
1839 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1842 * Just in case we get back EOPNOTSUPP for some reason,
1843 * just ignore the return value so we don't screw up
1844 * people calling discard_extent.
1852 *actual_bytes = discarded_bytes;
1858 /* Can return -ENOMEM */
1859 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 u64 bytenr, u64 num_bytes, u64 parent,
1862 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1865 struct btrfs_fs_info *fs_info = root->fs_info;
1867 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1868 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1870 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1871 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1873 parent, root_objectid, (int)owner,
1874 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1876 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1878 parent, root_objectid, owner, offset,
1879 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1884 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 u64 bytenr, u64 num_bytes,
1887 u64 parent, u64 root_objectid,
1888 u64 owner, u64 offset, int refs_to_add,
1889 struct btrfs_delayed_extent_op *extent_op)
1891 struct btrfs_path *path;
1892 struct extent_buffer *leaf;
1893 struct btrfs_extent_item *item;
1898 path = btrfs_alloc_path();
1903 path->leave_spinning = 1;
1904 /* this will setup the path even if it fails to insert the back ref */
1905 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1906 path, bytenr, num_bytes, parent,
1907 root_objectid, owner, offset,
1908 refs_to_add, extent_op);
1912 if (ret != -EAGAIN) {
1917 leaf = path->nodes[0];
1918 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1919 refs = btrfs_extent_refs(leaf, item);
1920 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1922 __run_delayed_extent_op(extent_op, leaf, item);
1924 btrfs_mark_buffer_dirty(leaf);
1925 btrfs_release_path(path);
1928 path->leave_spinning = 1;
1930 /* now insert the actual backref */
1931 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1932 path, bytenr, parent, root_objectid,
1933 owner, offset, refs_to_add);
1935 btrfs_abort_transaction(trans, root, ret);
1937 btrfs_free_path(path);
1941 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 struct btrfs_delayed_ref_node *node,
1944 struct btrfs_delayed_extent_op *extent_op,
1945 int insert_reserved)
1948 struct btrfs_delayed_data_ref *ref;
1949 struct btrfs_key ins;
1954 ins.objectid = node->bytenr;
1955 ins.offset = node->num_bytes;
1956 ins.type = BTRFS_EXTENT_ITEM_KEY;
1958 ref = btrfs_delayed_node_to_data_ref(node);
1959 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1960 parent = ref->parent;
1962 ref_root = ref->root;
1964 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1966 BUG_ON(extent_op->update_key);
1967 flags |= extent_op->flags_to_set;
1969 ret = alloc_reserved_file_extent(trans, root,
1970 parent, ref_root, flags,
1971 ref->objectid, ref->offset,
1972 &ins, node->ref_mod);
1973 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1974 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1975 node->num_bytes, parent,
1976 ref_root, ref->objectid,
1977 ref->offset, node->ref_mod,
1979 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1980 ret = __btrfs_free_extent(trans, root, node->bytenr,
1981 node->num_bytes, parent,
1982 ref_root, ref->objectid,
1983 ref->offset, node->ref_mod,
1991 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1992 struct extent_buffer *leaf,
1993 struct btrfs_extent_item *ei)
1995 u64 flags = btrfs_extent_flags(leaf, ei);
1996 if (extent_op->update_flags) {
1997 flags |= extent_op->flags_to_set;
1998 btrfs_set_extent_flags(leaf, ei, flags);
2001 if (extent_op->update_key) {
2002 struct btrfs_tree_block_info *bi;
2003 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2004 bi = (struct btrfs_tree_block_info *)(ei + 1);
2005 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2009 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 struct btrfs_delayed_ref_node *node,
2012 struct btrfs_delayed_extent_op *extent_op)
2014 struct btrfs_key key;
2015 struct btrfs_path *path;
2016 struct btrfs_extent_item *ei;
2017 struct extent_buffer *leaf;
2025 path = btrfs_alloc_path();
2029 key.objectid = node->bytenr;
2030 key.type = BTRFS_EXTENT_ITEM_KEY;
2031 key.offset = node->num_bytes;
2034 path->leave_spinning = 1;
2035 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2046 leaf = path->nodes[0];
2047 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2048 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2049 if (item_size < sizeof(*ei)) {
2050 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2056 leaf = path->nodes[0];
2057 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2060 BUG_ON(item_size < sizeof(*ei));
2061 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2062 __run_delayed_extent_op(extent_op, leaf, ei);
2064 btrfs_mark_buffer_dirty(leaf);
2066 btrfs_free_path(path);
2070 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2071 struct btrfs_root *root,
2072 struct btrfs_delayed_ref_node *node,
2073 struct btrfs_delayed_extent_op *extent_op,
2074 int insert_reserved)
2077 struct btrfs_delayed_tree_ref *ref;
2078 struct btrfs_key ins;
2082 ins.objectid = node->bytenr;
2083 ins.offset = node->num_bytes;
2084 ins.type = BTRFS_EXTENT_ITEM_KEY;
2086 ref = btrfs_delayed_node_to_tree_ref(node);
2087 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2088 parent = ref->parent;
2090 ref_root = ref->root;
2092 BUG_ON(node->ref_mod != 1);
2093 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2094 BUG_ON(!extent_op || !extent_op->update_flags ||
2095 !extent_op->update_key);
2096 ret = alloc_reserved_tree_block(trans, root,
2098 extent_op->flags_to_set,
2101 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2102 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2103 node->num_bytes, parent, ref_root,
2104 ref->level, 0, 1, extent_op);
2105 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2106 ret = __btrfs_free_extent(trans, root, node->bytenr,
2107 node->num_bytes, parent, ref_root,
2108 ref->level, 0, 1, extent_op);
2115 /* helper function to actually process a single delayed ref entry */
2116 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2117 struct btrfs_root *root,
2118 struct btrfs_delayed_ref_node *node,
2119 struct btrfs_delayed_extent_op *extent_op,
2120 int insert_reserved)
2127 if (btrfs_delayed_ref_is_head(node)) {
2128 struct btrfs_delayed_ref_head *head;
2130 * we've hit the end of the chain and we were supposed
2131 * to insert this extent into the tree. But, it got
2132 * deleted before we ever needed to insert it, so all
2133 * we have to do is clean up the accounting
2136 head = btrfs_delayed_node_to_head(node);
2137 if (insert_reserved) {
2138 btrfs_pin_extent(root, node->bytenr,
2139 node->num_bytes, 1);
2140 if (head->is_data) {
2141 ret = btrfs_del_csums(trans, root,
2146 mutex_unlock(&head->mutex);
2150 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2151 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2152 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2154 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2155 node->type == BTRFS_SHARED_DATA_REF_KEY)
2156 ret = run_delayed_data_ref(trans, root, node, extent_op,
2163 static noinline struct btrfs_delayed_ref_node *
2164 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2166 struct rb_node *node;
2167 struct btrfs_delayed_ref_node *ref;
2168 int action = BTRFS_ADD_DELAYED_REF;
2171 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2172 * this prevents ref count from going down to zero when
2173 * there still are pending delayed ref.
2175 node = rb_prev(&head->node.rb_node);
2179 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2181 if (ref->bytenr != head->node.bytenr)
2183 if (ref->action == action)
2185 node = rb_prev(node);
2187 if (action == BTRFS_ADD_DELAYED_REF) {
2188 action = BTRFS_DROP_DELAYED_REF;
2195 * Returns 0 on success or if called with an already aborted transaction.
2196 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2198 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *root,
2200 struct list_head *cluster)
2202 struct btrfs_delayed_ref_root *delayed_refs;
2203 struct btrfs_delayed_ref_node *ref;
2204 struct btrfs_delayed_ref_head *locked_ref = NULL;
2205 struct btrfs_delayed_extent_op *extent_op;
2206 struct btrfs_fs_info *fs_info = root->fs_info;
2209 int must_insert_reserved = 0;
2211 delayed_refs = &trans->transaction->delayed_refs;
2214 /* pick a new head ref from the cluster list */
2215 if (list_empty(cluster))
2218 locked_ref = list_entry(cluster->next,
2219 struct btrfs_delayed_ref_head, cluster);
2221 /* grab the lock that says we are going to process
2222 * all the refs for this head */
2223 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2226 * we may have dropped the spin lock to get the head
2227 * mutex lock, and that might have given someone else
2228 * time to free the head. If that's true, it has been
2229 * removed from our list and we can move on.
2231 if (ret == -EAGAIN) {
2239 * We need to try and merge add/drops of the same ref since we
2240 * can run into issues with relocate dropping the implicit ref
2241 * and then it being added back again before the drop can
2242 * finish. If we merged anything we need to re-loop so we can
2245 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2249 * locked_ref is the head node, so we have to go one
2250 * node back for any delayed ref updates
2252 ref = select_delayed_ref(locked_ref);
2254 if (ref && ref->seq &&
2255 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2257 * there are still refs with lower seq numbers in the
2258 * process of being added. Don't run this ref yet.
2260 list_del_init(&locked_ref->cluster);
2261 mutex_unlock(&locked_ref->mutex);
2263 delayed_refs->num_heads_ready++;
2264 spin_unlock(&delayed_refs->lock);
2266 spin_lock(&delayed_refs->lock);
2271 * record the must insert reserved flag before we
2272 * drop the spin lock.
2274 must_insert_reserved = locked_ref->must_insert_reserved;
2275 locked_ref->must_insert_reserved = 0;
2277 extent_op = locked_ref->extent_op;
2278 locked_ref->extent_op = NULL;
2281 /* All delayed refs have been processed, Go ahead
2282 * and send the head node to run_one_delayed_ref,
2283 * so that any accounting fixes can happen
2285 ref = &locked_ref->node;
2287 if (extent_op && must_insert_reserved) {
2293 spin_unlock(&delayed_refs->lock);
2295 ret = run_delayed_extent_op(trans, root,
2300 list_del_init(&locked_ref->cluster);
2301 mutex_unlock(&locked_ref->mutex);
2303 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2304 spin_lock(&delayed_refs->lock);
2311 list_del_init(&locked_ref->cluster);
2316 rb_erase(&ref->rb_node, &delayed_refs->root);
2317 delayed_refs->num_entries--;
2320 * when we play the delayed ref, also correct the
2323 switch (ref->action) {
2324 case BTRFS_ADD_DELAYED_REF:
2325 case BTRFS_ADD_DELAYED_EXTENT:
2326 locked_ref->node.ref_mod -= ref->ref_mod;
2328 case BTRFS_DROP_DELAYED_REF:
2329 locked_ref->node.ref_mod += ref->ref_mod;
2335 spin_unlock(&delayed_refs->lock);
2337 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2338 must_insert_reserved);
2340 btrfs_put_delayed_ref(ref);
2346 list_del_init(&locked_ref->cluster);
2347 mutex_unlock(&locked_ref->mutex);
2349 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2350 spin_lock(&delayed_refs->lock);
2356 spin_lock(&delayed_refs->lock);
2361 #ifdef SCRAMBLE_DELAYED_REFS
2363 * Normally delayed refs get processed in ascending bytenr order. This
2364 * correlates in most cases to the order added. To expose dependencies on this
2365 * order, we start to process the tree in the middle instead of the beginning
2367 static u64 find_middle(struct rb_root *root)
2369 struct rb_node *n = root->rb_node;
2370 struct btrfs_delayed_ref_node *entry;
2373 u64 first = 0, last = 0;
2377 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2378 first = entry->bytenr;
2382 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2383 last = entry->bytenr;
2388 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2389 WARN_ON(!entry->in_tree);
2391 middle = entry->bytenr;
2404 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2405 struct btrfs_fs_info *fs_info)
2407 struct qgroup_update *qgroup_update;
2410 if (list_empty(&trans->qgroup_ref_list) !=
2411 !trans->delayed_ref_elem.seq) {
2412 /* list without seq or seq without list */
2413 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2414 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2415 trans->delayed_ref_elem.seq);
2419 if (!trans->delayed_ref_elem.seq)
2422 while (!list_empty(&trans->qgroup_ref_list)) {
2423 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2424 struct qgroup_update, list);
2425 list_del(&qgroup_update->list);
2427 ret = btrfs_qgroup_account_ref(
2428 trans, fs_info, qgroup_update->node,
2429 qgroup_update->extent_op);
2430 kfree(qgroup_update);
2433 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2439 * this starts processing the delayed reference count updates and
2440 * extent insertions we have queued up so far. count can be
2441 * 0, which means to process everything in the tree at the start
2442 * of the run (but not newly added entries), or it can be some target
2443 * number you'd like to process.
2445 * Returns 0 on success or if called with an aborted transaction
2446 * Returns <0 on error and aborts the transaction
2448 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2449 struct btrfs_root *root, unsigned long count)
2451 struct rb_node *node;
2452 struct btrfs_delayed_ref_root *delayed_refs;
2453 struct btrfs_delayed_ref_node *ref;
2454 struct list_head cluster;
2457 int run_all = count == (unsigned long)-1;
2461 /* We'll clean this up in btrfs_cleanup_transaction */
2465 if (root == root->fs_info->extent_root)
2466 root = root->fs_info->tree_root;
2468 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2470 delayed_refs = &trans->transaction->delayed_refs;
2471 INIT_LIST_HEAD(&cluster);
2474 spin_lock(&delayed_refs->lock);
2476 #ifdef SCRAMBLE_DELAYED_REFS
2477 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2481 count = delayed_refs->num_entries * 2;
2485 if (!(run_all || run_most) &&
2486 delayed_refs->num_heads_ready < 64)
2490 * go find something we can process in the rbtree. We start at
2491 * the beginning of the tree, and then build a cluster
2492 * of refs to process starting at the first one we are able to
2495 delayed_start = delayed_refs->run_delayed_start;
2496 ret = btrfs_find_ref_cluster(trans, &cluster,
2497 delayed_refs->run_delayed_start);
2501 ret = run_clustered_refs(trans, root, &cluster);
2503 spin_unlock(&delayed_refs->lock);
2504 btrfs_abort_transaction(trans, root, ret);
2508 count -= min_t(unsigned long, ret, count);
2513 if (delayed_start >= delayed_refs->run_delayed_start) {
2516 * btrfs_find_ref_cluster looped. let's do one
2517 * more cycle. if we don't run any delayed ref
2518 * during that cycle (because we can't because
2519 * all of them are blocked), bail out.
2524 * no runnable refs left, stop trying
2531 /* refs were run, let's reset staleness detection */
2537 if (!list_empty(&trans->new_bgs)) {
2538 spin_unlock(&delayed_refs->lock);
2539 btrfs_create_pending_block_groups(trans, root);
2540 spin_lock(&delayed_refs->lock);
2543 node = rb_first(&delayed_refs->root);
2546 count = (unsigned long)-1;
2549 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2551 if (btrfs_delayed_ref_is_head(ref)) {
2552 struct btrfs_delayed_ref_head *head;
2554 head = btrfs_delayed_node_to_head(ref);
2555 atomic_inc(&ref->refs);
2557 spin_unlock(&delayed_refs->lock);
2559 * Mutex was contended, block until it's
2560 * released and try again
2562 mutex_lock(&head->mutex);
2563 mutex_unlock(&head->mutex);
2565 btrfs_put_delayed_ref(ref);
2569 node = rb_next(node);
2571 spin_unlock(&delayed_refs->lock);
2572 schedule_timeout(1);
2576 spin_unlock(&delayed_refs->lock);
2577 assert_qgroups_uptodate(trans);
2581 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2582 struct btrfs_root *root,
2583 u64 bytenr, u64 num_bytes, u64 flags,
2586 struct btrfs_delayed_extent_op *extent_op;
2589 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2593 extent_op->flags_to_set = flags;
2594 extent_op->update_flags = 1;
2595 extent_op->update_key = 0;
2596 extent_op->is_data = is_data ? 1 : 0;
2598 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2599 num_bytes, extent_op);
2605 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root,
2607 struct btrfs_path *path,
2608 u64 objectid, u64 offset, u64 bytenr)
2610 struct btrfs_delayed_ref_head *head;
2611 struct btrfs_delayed_ref_node *ref;
2612 struct btrfs_delayed_data_ref *data_ref;
2613 struct btrfs_delayed_ref_root *delayed_refs;
2614 struct rb_node *node;
2618 delayed_refs = &trans->transaction->delayed_refs;
2619 spin_lock(&delayed_refs->lock);
2620 head = btrfs_find_delayed_ref_head(trans, bytenr);
2624 if (!mutex_trylock(&head->mutex)) {
2625 atomic_inc(&head->node.refs);
2626 spin_unlock(&delayed_refs->lock);
2628 btrfs_release_path(path);
2631 * Mutex was contended, block until it's released and let
2634 mutex_lock(&head->mutex);
2635 mutex_unlock(&head->mutex);
2636 btrfs_put_delayed_ref(&head->node);
2640 node = rb_prev(&head->node.rb_node);
2644 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2646 if (ref->bytenr != bytenr)
2650 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2653 data_ref = btrfs_delayed_node_to_data_ref(ref);
2655 node = rb_prev(node);
2659 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2660 if (ref->bytenr == bytenr && ref->seq == seq)
2664 if (data_ref->root != root->root_key.objectid ||
2665 data_ref->objectid != objectid || data_ref->offset != offset)
2670 mutex_unlock(&head->mutex);
2672 spin_unlock(&delayed_refs->lock);
2676 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2677 struct btrfs_root *root,
2678 struct btrfs_path *path,
2679 u64 objectid, u64 offset, u64 bytenr)
2681 struct btrfs_root *extent_root = root->fs_info->extent_root;
2682 struct extent_buffer *leaf;
2683 struct btrfs_extent_data_ref *ref;
2684 struct btrfs_extent_inline_ref *iref;
2685 struct btrfs_extent_item *ei;
2686 struct btrfs_key key;
2690 key.objectid = bytenr;
2691 key.offset = (u64)-1;
2692 key.type = BTRFS_EXTENT_ITEM_KEY;
2694 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2697 BUG_ON(ret == 0); /* Corruption */
2700 if (path->slots[0] == 0)
2704 leaf = path->nodes[0];
2705 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2707 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2711 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2712 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2713 if (item_size < sizeof(*ei)) {
2714 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2718 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2720 if (item_size != sizeof(*ei) +
2721 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2724 if (btrfs_extent_generation(leaf, ei) <=
2725 btrfs_root_last_snapshot(&root->root_item))
2728 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2729 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2730 BTRFS_EXTENT_DATA_REF_KEY)
2733 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2734 if (btrfs_extent_refs(leaf, ei) !=
2735 btrfs_extent_data_ref_count(leaf, ref) ||
2736 btrfs_extent_data_ref_root(leaf, ref) !=
2737 root->root_key.objectid ||
2738 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2739 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2747 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 u64 objectid, u64 offset, u64 bytenr)
2751 struct btrfs_path *path;
2755 path = btrfs_alloc_path();
2760 ret = check_committed_ref(trans, root, path, objectid,
2762 if (ret && ret != -ENOENT)
2765 ret2 = check_delayed_ref(trans, root, path, objectid,
2767 } while (ret2 == -EAGAIN);
2769 if (ret2 && ret2 != -ENOENT) {
2774 if (ret != -ENOENT || ret2 != -ENOENT)
2777 btrfs_free_path(path);
2778 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2783 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct extent_buffer *buf,
2786 int full_backref, int inc, int for_cow)
2793 struct btrfs_key key;
2794 struct btrfs_file_extent_item *fi;
2798 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2799 u64, u64, u64, u64, u64, u64, int);
2801 ref_root = btrfs_header_owner(buf);
2802 nritems = btrfs_header_nritems(buf);
2803 level = btrfs_header_level(buf);
2805 if (!root->ref_cows && level == 0)
2809 process_func = btrfs_inc_extent_ref;
2811 process_func = btrfs_free_extent;
2814 parent = buf->start;
2818 for (i = 0; i < nritems; i++) {
2820 btrfs_item_key_to_cpu(buf, &key, i);
2821 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2823 fi = btrfs_item_ptr(buf, i,
2824 struct btrfs_file_extent_item);
2825 if (btrfs_file_extent_type(buf, fi) ==
2826 BTRFS_FILE_EXTENT_INLINE)
2828 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2832 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2833 key.offset -= btrfs_file_extent_offset(buf, fi);
2834 ret = process_func(trans, root, bytenr, num_bytes,
2835 parent, ref_root, key.objectid,
2836 key.offset, for_cow);
2840 bytenr = btrfs_node_blockptr(buf, i);
2841 num_bytes = btrfs_level_size(root, level - 1);
2842 ret = process_func(trans, root, bytenr, num_bytes,
2843 parent, ref_root, level - 1, 0,
2854 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2855 struct extent_buffer *buf, int full_backref, int for_cow)
2857 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2860 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2861 struct extent_buffer *buf, int full_backref, int for_cow)
2863 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2866 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2867 struct btrfs_root *root,
2868 struct btrfs_path *path,
2869 struct btrfs_block_group_cache *cache)
2872 struct btrfs_root *extent_root = root->fs_info->extent_root;
2874 struct extent_buffer *leaf;
2876 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2879 BUG_ON(ret); /* Corruption */
2881 leaf = path->nodes[0];
2882 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2883 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2884 btrfs_mark_buffer_dirty(leaf);
2885 btrfs_release_path(path);
2888 btrfs_abort_transaction(trans, root, ret);
2895 static struct btrfs_block_group_cache *
2896 next_block_group(struct btrfs_root *root,
2897 struct btrfs_block_group_cache *cache)
2899 struct rb_node *node;
2900 spin_lock(&root->fs_info->block_group_cache_lock);
2901 node = rb_next(&cache->cache_node);
2902 btrfs_put_block_group(cache);
2904 cache = rb_entry(node, struct btrfs_block_group_cache,
2906 btrfs_get_block_group(cache);
2909 spin_unlock(&root->fs_info->block_group_cache_lock);
2913 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2914 struct btrfs_trans_handle *trans,
2915 struct btrfs_path *path)
2917 struct btrfs_root *root = block_group->fs_info->tree_root;
2918 struct inode *inode = NULL;
2920 int dcs = BTRFS_DC_ERROR;
2926 * If this block group is smaller than 100 megs don't bother caching the
2929 if (block_group->key.offset < (100 * 1024 * 1024)) {
2930 spin_lock(&block_group->lock);
2931 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2932 spin_unlock(&block_group->lock);
2937 inode = lookup_free_space_inode(root, block_group, path);
2938 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2939 ret = PTR_ERR(inode);
2940 btrfs_release_path(path);
2944 if (IS_ERR(inode)) {
2948 if (block_group->ro)
2951 ret = create_free_space_inode(root, trans, block_group, path);
2957 /* We've already setup this transaction, go ahead and exit */
2958 if (block_group->cache_generation == trans->transid &&
2959 i_size_read(inode)) {
2960 dcs = BTRFS_DC_SETUP;
2965 * We want to set the generation to 0, that way if anything goes wrong
2966 * from here on out we know not to trust this cache when we load up next
2969 BTRFS_I(inode)->generation = 0;
2970 ret = btrfs_update_inode(trans, root, inode);
2973 if (i_size_read(inode) > 0) {
2974 ret = btrfs_truncate_free_space_cache(root, trans, path,
2980 spin_lock(&block_group->lock);
2981 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2982 !btrfs_test_opt(root, SPACE_CACHE)) {
2984 * don't bother trying to write stuff out _if_
2985 * a) we're not cached,
2986 * b) we're with nospace_cache mount option.
2988 dcs = BTRFS_DC_WRITTEN;
2989 spin_unlock(&block_group->lock);
2992 spin_unlock(&block_group->lock);
2995 * Try to preallocate enough space based on how big the block group is.
2996 * Keep in mind this has to include any pinned space which could end up
2997 * taking up quite a bit since it's not folded into the other space
3000 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3005 num_pages *= PAGE_CACHE_SIZE;
3007 ret = btrfs_check_data_free_space(inode, num_pages);
3011 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3012 num_pages, num_pages,
3015 dcs = BTRFS_DC_SETUP;
3016 btrfs_free_reserved_data_space(inode, num_pages);
3021 btrfs_release_path(path);
3023 spin_lock(&block_group->lock);
3024 if (!ret && dcs == BTRFS_DC_SETUP)
3025 block_group->cache_generation = trans->transid;
3026 block_group->disk_cache_state = dcs;
3027 spin_unlock(&block_group->lock);
3032 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root)
3035 struct btrfs_block_group_cache *cache;
3037 struct btrfs_path *path;
3040 path = btrfs_alloc_path();
3046 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3048 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3050 cache = next_block_group(root, cache);
3058 err = cache_save_setup(cache, trans, path);
3059 last = cache->key.objectid + cache->key.offset;
3060 btrfs_put_block_group(cache);
3065 err = btrfs_run_delayed_refs(trans, root,
3067 if (err) /* File system offline */
3071 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3073 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3074 btrfs_put_block_group(cache);
3080 cache = next_block_group(root, cache);
3089 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3090 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3092 last = cache->key.objectid + cache->key.offset;
3094 err = write_one_cache_group(trans, root, path, cache);
3095 if (err) /* File system offline */
3098 btrfs_put_block_group(cache);
3103 * I don't think this is needed since we're just marking our
3104 * preallocated extent as written, but just in case it can't
3108 err = btrfs_run_delayed_refs(trans, root,
3110 if (err) /* File system offline */
3114 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3117 * Really this shouldn't happen, but it could if we
3118 * couldn't write the entire preallocated extent and
3119 * splitting the extent resulted in a new block.
3122 btrfs_put_block_group(cache);
3125 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3127 cache = next_block_group(root, cache);
3136 err = btrfs_write_out_cache(root, trans, cache, path);
3139 * If we didn't have an error then the cache state is still
3140 * NEED_WRITE, so we can set it to WRITTEN.
3142 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3143 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3144 last = cache->key.objectid + cache->key.offset;
3145 btrfs_put_block_group(cache);
3149 btrfs_free_path(path);
3153 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3155 struct btrfs_block_group_cache *block_group;
3158 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3159 if (!block_group || block_group->ro)
3162 btrfs_put_block_group(block_group);
3166 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3167 u64 total_bytes, u64 bytes_used,
3168 struct btrfs_space_info **space_info)
3170 struct btrfs_space_info *found;
3174 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3175 BTRFS_BLOCK_GROUP_RAID10))
3180 found = __find_space_info(info, flags);
3182 spin_lock(&found->lock);
3183 found->total_bytes += total_bytes;
3184 found->disk_total += total_bytes * factor;
3185 found->bytes_used += bytes_used;
3186 found->disk_used += bytes_used * factor;
3188 spin_unlock(&found->lock);
3189 *space_info = found;
3192 found = kzalloc(sizeof(*found), GFP_NOFS);
3196 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3197 INIT_LIST_HEAD(&found->block_groups[i]);
3198 init_rwsem(&found->groups_sem);
3199 spin_lock_init(&found->lock);
3200 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3201 found->total_bytes = total_bytes;
3202 found->disk_total = total_bytes * factor;
3203 found->bytes_used = bytes_used;
3204 found->disk_used = bytes_used * factor;
3205 found->bytes_pinned = 0;
3206 found->bytes_reserved = 0;
3207 found->bytes_readonly = 0;
3208 found->bytes_may_use = 0;
3210 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3211 found->chunk_alloc = 0;
3213 init_waitqueue_head(&found->wait);
3214 *space_info = found;
3215 list_add_rcu(&found->list, &info->space_info);
3216 if (flags & BTRFS_BLOCK_GROUP_DATA)
3217 info->data_sinfo = found;
3221 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3223 u64 extra_flags = chunk_to_extended(flags) &
3224 BTRFS_EXTENDED_PROFILE_MASK;
3226 if (flags & BTRFS_BLOCK_GROUP_DATA)
3227 fs_info->avail_data_alloc_bits |= extra_flags;
3228 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3229 fs_info->avail_metadata_alloc_bits |= extra_flags;
3230 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3231 fs_info->avail_system_alloc_bits |= extra_flags;
3235 * returns target flags in extended format or 0 if restripe for this
3236 * chunk_type is not in progress
3238 * should be called with either volume_mutex or balance_lock held
3240 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3242 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3248 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3249 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3250 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3251 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3252 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3253 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3254 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3255 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3256 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3263 * @flags: available profiles in extended format (see ctree.h)
3265 * Returns reduced profile in chunk format. If profile changing is in
3266 * progress (either running or paused) picks the target profile (if it's
3267 * already available), otherwise falls back to plain reducing.
3269 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3272 * we add in the count of missing devices because we want
3273 * to make sure that any RAID levels on a degraded FS
3274 * continue to be honored.
3276 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3277 root->fs_info->fs_devices->missing_devices;
3281 * see if restripe for this chunk_type is in progress, if so
3282 * try to reduce to the target profile
3284 spin_lock(&root->fs_info->balance_lock);
3285 target = get_restripe_target(root->fs_info, flags);
3287 /* pick target profile only if it's already available */
3288 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3289 spin_unlock(&root->fs_info->balance_lock);
3290 return extended_to_chunk(target);
3293 spin_unlock(&root->fs_info->balance_lock);
3295 if (num_devices == 1)
3296 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3297 if (num_devices < 4)
3298 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3300 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3301 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3302 BTRFS_BLOCK_GROUP_RAID10))) {
3303 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3306 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3307 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3308 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3311 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3312 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3313 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3314 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3315 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3318 return extended_to_chunk(flags);
3321 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3323 if (flags & BTRFS_BLOCK_GROUP_DATA)
3324 flags |= root->fs_info->avail_data_alloc_bits;
3325 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3326 flags |= root->fs_info->avail_system_alloc_bits;
3327 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3328 flags |= root->fs_info->avail_metadata_alloc_bits;
3330 return btrfs_reduce_alloc_profile(root, flags);
3333 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3338 flags = BTRFS_BLOCK_GROUP_DATA;
3339 else if (root == root->fs_info->chunk_root)
3340 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3342 flags = BTRFS_BLOCK_GROUP_METADATA;
3344 return get_alloc_profile(root, flags);
3348 * This will check the space that the inode allocates from to make sure we have
3349 * enough space for bytes.
3351 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3353 struct btrfs_space_info *data_sinfo;
3354 struct btrfs_root *root = BTRFS_I(inode)->root;
3355 struct btrfs_fs_info *fs_info = root->fs_info;
3357 int ret = 0, committed = 0, alloc_chunk = 1;
3359 /* make sure bytes are sectorsize aligned */
3360 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3362 if (root == root->fs_info->tree_root ||
3363 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3368 data_sinfo = fs_info->data_sinfo;
3373 /* make sure we have enough space to handle the data first */
3374 spin_lock(&data_sinfo->lock);
3375 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3376 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3377 data_sinfo->bytes_may_use;
3379 if (used + bytes > data_sinfo->total_bytes) {
3380 struct btrfs_trans_handle *trans;
3383 * if we don't have enough free bytes in this space then we need
3384 * to alloc a new chunk.
3386 if (!data_sinfo->full && alloc_chunk) {
3389 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3390 spin_unlock(&data_sinfo->lock);
3392 alloc_target = btrfs_get_alloc_profile(root, 1);
3393 trans = btrfs_join_transaction(root);
3395 return PTR_ERR(trans);
3397 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3399 CHUNK_ALLOC_NO_FORCE);
3400 btrfs_end_transaction(trans, root);
3409 data_sinfo = fs_info->data_sinfo;
3415 * If we have less pinned bytes than we want to allocate then
3416 * don't bother committing the transaction, it won't help us.
3418 if (data_sinfo->bytes_pinned < bytes)
3420 spin_unlock(&data_sinfo->lock);
3422 /* commit the current transaction and try again */
3425 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3427 trans = btrfs_join_transaction(root);
3429 return PTR_ERR(trans);
3430 ret = btrfs_commit_transaction(trans, root);
3438 data_sinfo->bytes_may_use += bytes;
3439 trace_btrfs_space_reservation(root->fs_info, "space_info",
3440 data_sinfo->flags, bytes, 1);
3441 spin_unlock(&data_sinfo->lock);
3447 * Called if we need to clear a data reservation for this inode.
3449 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3451 struct btrfs_root *root = BTRFS_I(inode)->root;
3452 struct btrfs_space_info *data_sinfo;
3454 /* make sure bytes are sectorsize aligned */
3455 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3457 data_sinfo = root->fs_info->data_sinfo;
3458 spin_lock(&data_sinfo->lock);
3459 data_sinfo->bytes_may_use -= bytes;
3460 trace_btrfs_space_reservation(root->fs_info, "space_info",
3461 data_sinfo->flags, bytes, 0);
3462 spin_unlock(&data_sinfo->lock);
3465 static void force_metadata_allocation(struct btrfs_fs_info *info)
3467 struct list_head *head = &info->space_info;
3468 struct btrfs_space_info *found;
3471 list_for_each_entry_rcu(found, head, list) {
3472 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3473 found->force_alloc = CHUNK_ALLOC_FORCE;
3478 static int should_alloc_chunk(struct btrfs_root *root,
3479 struct btrfs_space_info *sinfo, int force)
3481 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3482 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3483 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3486 if (force == CHUNK_ALLOC_FORCE)
3490 * We need to take into account the global rsv because for all intents
3491 * and purposes it's used space. Don't worry about locking the
3492 * global_rsv, it doesn't change except when the transaction commits.
3494 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3495 num_allocated += global_rsv->size;
3498 * in limited mode, we want to have some free space up to
3499 * about 1% of the FS size.
3501 if (force == CHUNK_ALLOC_LIMITED) {
3502 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3503 thresh = max_t(u64, 64 * 1024 * 1024,
3504 div_factor_fine(thresh, 1));
3506 if (num_bytes - num_allocated < thresh)
3510 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3515 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3519 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3520 type & BTRFS_BLOCK_GROUP_RAID0)
3521 num_dev = root->fs_info->fs_devices->rw_devices;
3522 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3525 num_dev = 1; /* DUP or single */
3527 /* metadata for updaing devices and chunk tree */
3528 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3531 static void check_system_chunk(struct btrfs_trans_handle *trans,
3532 struct btrfs_root *root, u64 type)
3534 struct btrfs_space_info *info;
3538 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3539 spin_lock(&info->lock);
3540 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3541 info->bytes_reserved - info->bytes_readonly;
3542 spin_unlock(&info->lock);
3544 thresh = get_system_chunk_thresh(root, type);
3545 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3546 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3547 left, thresh, type);
3548 dump_space_info(info, 0, 0);
3551 if (left < thresh) {
3554 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3555 btrfs_alloc_chunk(trans, root, flags);
3559 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3560 struct btrfs_root *extent_root, u64 flags, int force)
3562 struct btrfs_space_info *space_info;
3563 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3564 int wait_for_alloc = 0;
3567 space_info = __find_space_info(extent_root->fs_info, flags);
3569 ret = update_space_info(extent_root->fs_info, flags,
3571 BUG_ON(ret); /* -ENOMEM */
3573 BUG_ON(!space_info); /* Logic error */
3576 spin_lock(&space_info->lock);
3577 if (force < space_info->force_alloc)
3578 force = space_info->force_alloc;
3579 if (space_info->full) {
3580 spin_unlock(&space_info->lock);
3584 if (!should_alloc_chunk(extent_root, space_info, force)) {
3585 spin_unlock(&space_info->lock);
3587 } else if (space_info->chunk_alloc) {
3590 space_info->chunk_alloc = 1;
3593 spin_unlock(&space_info->lock);
3595 mutex_lock(&fs_info->chunk_mutex);
3598 * The chunk_mutex is held throughout the entirety of a chunk
3599 * allocation, so once we've acquired the chunk_mutex we know that the
3600 * other guy is done and we need to recheck and see if we should
3603 if (wait_for_alloc) {
3604 mutex_unlock(&fs_info->chunk_mutex);
3610 * If we have mixed data/metadata chunks we want to make sure we keep
3611 * allocating mixed chunks instead of individual chunks.
3613 if (btrfs_mixed_space_info(space_info))
3614 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3617 * if we're doing a data chunk, go ahead and make sure that
3618 * we keep a reasonable number of metadata chunks allocated in the
3621 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3622 fs_info->data_chunk_allocations++;
3623 if (!(fs_info->data_chunk_allocations %
3624 fs_info->metadata_ratio))
3625 force_metadata_allocation(fs_info);
3629 * Check if we have enough space in SYSTEM chunk because we may need
3630 * to update devices.
3632 check_system_chunk(trans, extent_root, flags);
3634 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3635 if (ret < 0 && ret != -ENOSPC)
3638 spin_lock(&space_info->lock);
3640 space_info->full = 1;
3644 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3645 space_info->chunk_alloc = 0;
3646 spin_unlock(&space_info->lock);
3648 mutex_unlock(&fs_info->chunk_mutex);
3652 static int can_overcommit(struct btrfs_root *root,
3653 struct btrfs_space_info *space_info, u64 bytes,
3654 enum btrfs_reserve_flush_enum flush)
3656 u64 profile = btrfs_get_alloc_profile(root, 0);
3660 used = space_info->bytes_used + space_info->bytes_reserved +
3661 space_info->bytes_pinned + space_info->bytes_readonly +
3662 space_info->bytes_may_use;
3664 spin_lock(&root->fs_info->free_chunk_lock);
3665 avail = root->fs_info->free_chunk_space;
3666 spin_unlock(&root->fs_info->free_chunk_lock);
3669 * If we have dup, raid1 or raid10 then only half of the free
3670 * space is actually useable.
3672 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3673 BTRFS_BLOCK_GROUP_RAID1 |
3674 BTRFS_BLOCK_GROUP_RAID10))
3678 * If we aren't flushing all things, let us overcommit up to
3679 * 1/2th of the space. If we can flush, don't let us overcommit
3680 * too much, let it overcommit up to 1/8 of the space.
3682 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3687 if (used + bytes < space_info->total_bytes + avail)
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 try_to_writeback_inodes_sb_nr(root->fs_info->sb,
3726 WB_REASON_FS_FREE_SPACE);
3729 * We need to wait for the async pages to actually start before
3732 wait_event(root->fs_info->async_submit_wait,
3733 !atomic_read(&root->fs_info->async_delalloc_pages));
3736 flush = BTRFS_RESERVE_FLUSH_ALL;
3738 flush = BTRFS_RESERVE_NO_FLUSH;
3739 spin_lock(&space_info->lock);
3740 if (can_overcommit(root, space_info, orig, flush)) {
3741 spin_unlock(&space_info->lock);
3744 spin_unlock(&space_info->lock);
3747 if (wait_ordered && !trans) {
3748 btrfs_wait_ordered_extents(root, 0);
3750 time_left = schedule_timeout_killable(1);
3755 delalloc_bytes = root->fs_info->delalloc_bytes;
3760 * maybe_commit_transaction - possibly commit the transaction if its ok to
3761 * @root - the root we're allocating for
3762 * @bytes - the number of bytes we want to reserve
3763 * @force - force the commit
3765 * This will check to make sure that committing the transaction will actually
3766 * get us somewhere and then commit the transaction if it does. Otherwise it
3767 * will return -ENOSPC.
3769 static int may_commit_transaction(struct btrfs_root *root,
3770 struct btrfs_space_info *space_info,
3771 u64 bytes, int force)
3773 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3774 struct btrfs_trans_handle *trans;
3776 trans = (struct btrfs_trans_handle *)current->journal_info;
3783 /* See if there is enough pinned space to make this reservation */
3784 spin_lock(&space_info->lock);
3785 if (space_info->bytes_pinned >= bytes) {
3786 spin_unlock(&space_info->lock);
3789 spin_unlock(&space_info->lock);
3792 * See if there is some space in the delayed insertion reservation for
3795 if (space_info != delayed_rsv->space_info)
3798 spin_lock(&space_info->lock);
3799 spin_lock(&delayed_rsv->lock);
3800 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3801 spin_unlock(&delayed_rsv->lock);
3802 spin_unlock(&space_info->lock);
3805 spin_unlock(&delayed_rsv->lock);
3806 spin_unlock(&space_info->lock);
3809 trans = btrfs_join_transaction(root);
3813 return btrfs_commit_transaction(trans, root);
3817 FLUSH_DELAYED_ITEMS_NR = 1,
3818 FLUSH_DELAYED_ITEMS = 2,
3820 FLUSH_DELALLOC_WAIT = 4,
3825 static int flush_space(struct btrfs_root *root,
3826 struct btrfs_space_info *space_info, u64 num_bytes,
3827 u64 orig_bytes, int state)
3829 struct btrfs_trans_handle *trans;
3834 case FLUSH_DELAYED_ITEMS_NR:
3835 case FLUSH_DELAYED_ITEMS:
3836 if (state == FLUSH_DELAYED_ITEMS_NR) {
3837 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3839 nr = (int)div64_u64(num_bytes, bytes);
3846 trans = btrfs_join_transaction(root);
3847 if (IS_ERR(trans)) {
3848 ret = PTR_ERR(trans);
3851 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3852 btrfs_end_transaction(trans, root);
3854 case FLUSH_DELALLOC:
3855 case FLUSH_DELALLOC_WAIT:
3856 shrink_delalloc(root, num_bytes, orig_bytes,
3857 state == FLUSH_DELALLOC_WAIT);
3860 trans = btrfs_join_transaction(root);
3861 if (IS_ERR(trans)) {
3862 ret = PTR_ERR(trans);
3865 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3866 btrfs_get_alloc_profile(root, 0),
3867 CHUNK_ALLOC_NO_FORCE);
3868 btrfs_end_transaction(trans, root);
3873 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3883 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3884 * @root - the root we're allocating for
3885 * @block_rsv - the block_rsv we're allocating for
3886 * @orig_bytes - the number of bytes we want
3887 * @flush - whether or not we can flush to make our reservation
3889 * This will reserve orgi_bytes number of bytes from the space info associated
3890 * with the block_rsv. If there is not enough space it will make an attempt to
3891 * flush out space to make room. It will do this by flushing delalloc if
3892 * possible or committing the transaction. If flush is 0 then no attempts to
3893 * regain reservations will be made and this will fail if there is not enough
3896 static int reserve_metadata_bytes(struct btrfs_root *root,
3897 struct btrfs_block_rsv *block_rsv,
3899 enum btrfs_reserve_flush_enum flush)
3901 struct btrfs_space_info *space_info = block_rsv->space_info;
3903 u64 num_bytes = orig_bytes;
3904 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3906 bool flushing = false;
3910 spin_lock(&space_info->lock);
3912 * We only want to wait if somebody other than us is flushing and we
3913 * are actually allowed to flush all things.
3915 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3916 space_info->flush) {
3917 spin_unlock(&space_info->lock);
3919 * If we have a trans handle we can't wait because the flusher
3920 * may have to commit the transaction, which would mean we would
3921 * deadlock since we are waiting for the flusher to finish, but
3922 * hold the current transaction open.
3924 if (current->journal_info)
3926 ret = wait_event_killable(space_info->wait, !space_info->flush);
3927 /* Must have been killed, return */
3931 spin_lock(&space_info->lock);
3935 used = space_info->bytes_used + space_info->bytes_reserved +
3936 space_info->bytes_pinned + space_info->bytes_readonly +
3937 space_info->bytes_may_use;
3940 * The idea here is that we've not already over-reserved the block group
3941 * then we can go ahead and save our reservation first and then start
3942 * flushing if we need to. Otherwise if we've already overcommitted
3943 * lets start flushing stuff first and then come back and try to make
3946 if (used <= space_info->total_bytes) {
3947 if (used + orig_bytes <= space_info->total_bytes) {
3948 space_info->bytes_may_use += orig_bytes;
3949 trace_btrfs_space_reservation(root->fs_info,
3950 "space_info", space_info->flags, orig_bytes, 1);
3954 * Ok set num_bytes to orig_bytes since we aren't
3955 * overocmmitted, this way we only try and reclaim what
3958 num_bytes = orig_bytes;
3962 * Ok we're over committed, set num_bytes to the overcommitted
3963 * amount plus the amount of bytes that we need for this
3966 num_bytes = used - space_info->total_bytes +
3970 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
3971 space_info->bytes_may_use += orig_bytes;
3972 trace_btrfs_space_reservation(root->fs_info, "space_info",
3973 space_info->flags, orig_bytes,
3979 * Couldn't make our reservation, save our place so while we're trying
3980 * to reclaim space we can actually use it instead of somebody else
3981 * stealing it from us.
3983 * We make the other tasks wait for the flush only when we can flush
3986 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
3988 space_info->flush = 1;
3991 spin_unlock(&space_info->lock);
3993 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
3996 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4001 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4002 * would happen. So skip delalloc flush.
4004 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4005 (flush_state == FLUSH_DELALLOC ||
4006 flush_state == FLUSH_DELALLOC_WAIT))
4007 flush_state = ALLOC_CHUNK;
4011 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4012 flush_state < COMMIT_TRANS)
4014 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4015 flush_state <= COMMIT_TRANS)
4020 spin_lock(&space_info->lock);
4021 space_info->flush = 0;
4022 wake_up_all(&space_info->wait);
4023 spin_unlock(&space_info->lock);
4028 static struct btrfs_block_rsv *get_block_rsv(
4029 const struct btrfs_trans_handle *trans,
4030 const struct btrfs_root *root)
4032 struct btrfs_block_rsv *block_rsv = NULL;
4035 block_rsv = trans->block_rsv;
4037 if (root == root->fs_info->csum_root && trans->adding_csums)
4038 block_rsv = trans->block_rsv;
4041 block_rsv = root->block_rsv;
4044 block_rsv = &root->fs_info->empty_block_rsv;
4049 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4053 spin_lock(&block_rsv->lock);
4054 if (block_rsv->reserved >= num_bytes) {
4055 block_rsv->reserved -= num_bytes;
4056 if (block_rsv->reserved < block_rsv->size)
4057 block_rsv->full = 0;
4060 spin_unlock(&block_rsv->lock);
4064 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4065 u64 num_bytes, int update_size)
4067 spin_lock(&block_rsv->lock);
4068 block_rsv->reserved += num_bytes;
4070 block_rsv->size += num_bytes;
4071 else if (block_rsv->reserved >= block_rsv->size)
4072 block_rsv->full = 1;
4073 spin_unlock(&block_rsv->lock);
4076 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4077 struct btrfs_block_rsv *block_rsv,
4078 struct btrfs_block_rsv *dest, u64 num_bytes)
4080 struct btrfs_space_info *space_info = block_rsv->space_info;
4082 spin_lock(&block_rsv->lock);
4083 if (num_bytes == (u64)-1)
4084 num_bytes = block_rsv->size;
4085 block_rsv->size -= num_bytes;
4086 if (block_rsv->reserved >= block_rsv->size) {
4087 num_bytes = block_rsv->reserved - block_rsv->size;
4088 block_rsv->reserved = block_rsv->size;
4089 block_rsv->full = 1;
4093 spin_unlock(&block_rsv->lock);
4095 if (num_bytes > 0) {
4097 spin_lock(&dest->lock);
4101 bytes_to_add = dest->size - dest->reserved;
4102 bytes_to_add = min(num_bytes, bytes_to_add);
4103 dest->reserved += bytes_to_add;
4104 if (dest->reserved >= dest->size)
4106 num_bytes -= bytes_to_add;
4108 spin_unlock(&dest->lock);
4111 spin_lock(&space_info->lock);
4112 space_info->bytes_may_use -= num_bytes;
4113 trace_btrfs_space_reservation(fs_info, "space_info",
4114 space_info->flags, num_bytes, 0);
4115 space_info->reservation_progress++;
4116 spin_unlock(&space_info->lock);
4121 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4122 struct btrfs_block_rsv *dst, u64 num_bytes)
4126 ret = block_rsv_use_bytes(src, num_bytes);
4130 block_rsv_add_bytes(dst, num_bytes, 1);
4134 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4136 memset(rsv, 0, sizeof(*rsv));
4137 spin_lock_init(&rsv->lock);
4141 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4142 unsigned short type)
4144 struct btrfs_block_rsv *block_rsv;
4145 struct btrfs_fs_info *fs_info = root->fs_info;
4147 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4151 btrfs_init_block_rsv(block_rsv, type);
4152 block_rsv->space_info = __find_space_info(fs_info,
4153 BTRFS_BLOCK_GROUP_METADATA);
4157 void btrfs_free_block_rsv(struct btrfs_root *root,
4158 struct btrfs_block_rsv *rsv)
4162 btrfs_block_rsv_release(root, rsv, (u64)-1);
4166 int btrfs_block_rsv_add(struct btrfs_root *root,
4167 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4168 enum btrfs_reserve_flush_enum flush)
4175 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4177 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4184 int btrfs_block_rsv_check(struct btrfs_root *root,
4185 struct btrfs_block_rsv *block_rsv, int min_factor)
4193 spin_lock(&block_rsv->lock);
4194 num_bytes = div_factor(block_rsv->size, min_factor);
4195 if (block_rsv->reserved >= num_bytes)
4197 spin_unlock(&block_rsv->lock);
4202 int btrfs_block_rsv_refill(struct btrfs_root *root,
4203 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4204 enum btrfs_reserve_flush_enum flush)
4212 spin_lock(&block_rsv->lock);
4213 num_bytes = min_reserved;
4214 if (block_rsv->reserved >= num_bytes)
4217 num_bytes -= block_rsv->reserved;
4218 spin_unlock(&block_rsv->lock);
4223 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4225 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4232 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4233 struct btrfs_block_rsv *dst_rsv,
4236 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4239 void btrfs_block_rsv_release(struct btrfs_root *root,
4240 struct btrfs_block_rsv *block_rsv,
4243 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4244 if (global_rsv->full || global_rsv == block_rsv ||
4245 block_rsv->space_info != global_rsv->space_info)
4247 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4252 * helper to calculate size of global block reservation.
4253 * the desired value is sum of space used by extent tree,
4254 * checksum tree and root tree
4256 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4258 struct btrfs_space_info *sinfo;
4262 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4264 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4265 spin_lock(&sinfo->lock);
4266 data_used = sinfo->bytes_used;
4267 spin_unlock(&sinfo->lock);
4269 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4270 spin_lock(&sinfo->lock);
4271 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4273 meta_used = sinfo->bytes_used;
4274 spin_unlock(&sinfo->lock);
4276 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4278 num_bytes += div64_u64(data_used + meta_used, 50);
4280 if (num_bytes * 3 > meta_used)
4281 num_bytes = div64_u64(meta_used, 3);
4283 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4286 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4288 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4289 struct btrfs_space_info *sinfo = block_rsv->space_info;
4292 num_bytes = calc_global_metadata_size(fs_info);
4294 spin_lock(&sinfo->lock);
4295 spin_lock(&block_rsv->lock);
4297 block_rsv->size = num_bytes;
4299 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4300 sinfo->bytes_reserved + sinfo->bytes_readonly +
4301 sinfo->bytes_may_use;
4303 if (sinfo->total_bytes > num_bytes) {
4304 num_bytes = sinfo->total_bytes - num_bytes;
4305 block_rsv->reserved += num_bytes;
4306 sinfo->bytes_may_use += num_bytes;
4307 trace_btrfs_space_reservation(fs_info, "space_info",
4308 sinfo->flags, num_bytes, 1);
4311 if (block_rsv->reserved >= block_rsv->size) {
4312 num_bytes = block_rsv->reserved - block_rsv->size;
4313 sinfo->bytes_may_use -= num_bytes;
4314 trace_btrfs_space_reservation(fs_info, "space_info",
4315 sinfo->flags, num_bytes, 0);
4316 sinfo->reservation_progress++;
4317 block_rsv->reserved = block_rsv->size;
4318 block_rsv->full = 1;
4321 spin_unlock(&block_rsv->lock);
4322 spin_unlock(&sinfo->lock);
4325 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4327 struct btrfs_space_info *space_info;
4329 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4330 fs_info->chunk_block_rsv.space_info = space_info;
4332 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4333 fs_info->global_block_rsv.space_info = space_info;
4334 fs_info->delalloc_block_rsv.space_info = space_info;
4335 fs_info->trans_block_rsv.space_info = space_info;
4336 fs_info->empty_block_rsv.space_info = space_info;
4337 fs_info->delayed_block_rsv.space_info = space_info;
4339 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4340 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4341 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4342 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4343 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4345 update_global_block_rsv(fs_info);
4348 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4350 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4352 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4353 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4354 WARN_ON(fs_info->trans_block_rsv.size > 0);
4355 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4356 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4357 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4358 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4359 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4362 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4363 struct btrfs_root *root)
4365 if (!trans->block_rsv)
4368 if (!trans->bytes_reserved)
4371 trace_btrfs_space_reservation(root->fs_info, "transaction",
4372 trans->transid, trans->bytes_reserved, 0);
4373 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4374 trans->bytes_reserved = 0;
4377 /* Can only return 0 or -ENOSPC */
4378 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4379 struct inode *inode)
4381 struct btrfs_root *root = BTRFS_I(inode)->root;
4382 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4383 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4386 * We need to hold space in order to delete our orphan item once we've
4387 * added it, so this takes the reservation so we can release it later
4388 * when we are truly done with the orphan item.
4390 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4391 trace_btrfs_space_reservation(root->fs_info, "orphan",
4392 btrfs_ino(inode), num_bytes, 1);
4393 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4396 void btrfs_orphan_release_metadata(struct inode *inode)
4398 struct btrfs_root *root = BTRFS_I(inode)->root;
4399 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4400 trace_btrfs_space_reservation(root->fs_info, "orphan",
4401 btrfs_ino(inode), num_bytes, 0);
4402 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4405 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4406 struct btrfs_pending_snapshot *pending)
4408 struct btrfs_root *root = pending->root;
4409 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4410 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4412 * two for root back/forward refs, two for directory entries,
4413 * one for root of the snapshot and one for parent inode.
4415 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4416 dst_rsv->space_info = src_rsv->space_info;
4417 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4421 * drop_outstanding_extent - drop an outstanding extent
4422 * @inode: the inode we're dropping the extent for
4424 * This is called when we are freeing up an outstanding extent, either called
4425 * after an error or after an extent is written. This will return the number of
4426 * reserved extents that need to be freed. This must be called with
4427 * BTRFS_I(inode)->lock held.
4429 static unsigned drop_outstanding_extent(struct inode *inode)
4431 unsigned drop_inode_space = 0;
4432 unsigned dropped_extents = 0;
4434 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4435 BTRFS_I(inode)->outstanding_extents--;
4437 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4438 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4439 &BTRFS_I(inode)->runtime_flags))
4440 drop_inode_space = 1;
4443 * If we have more or the same amount of outsanding extents than we have
4444 * reserved then we need to leave the reserved extents count alone.
4446 if (BTRFS_I(inode)->outstanding_extents >=
4447 BTRFS_I(inode)->reserved_extents)
4448 return drop_inode_space;
4450 dropped_extents = BTRFS_I(inode)->reserved_extents -
4451 BTRFS_I(inode)->outstanding_extents;
4452 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4453 return dropped_extents + drop_inode_space;
4457 * calc_csum_metadata_size - return the amount of metada space that must be
4458 * reserved/free'd for the given bytes.
4459 * @inode: the inode we're manipulating
4460 * @num_bytes: the number of bytes in question
4461 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4463 * This adjusts the number of csum_bytes in the inode and then returns the
4464 * correct amount of metadata that must either be reserved or freed. We
4465 * calculate how many checksums we can fit into one leaf and then divide the
4466 * number of bytes that will need to be checksumed by this value to figure out
4467 * how many checksums will be required. If we are adding bytes then the number
4468 * may go up and we will return the number of additional bytes that must be
4469 * reserved. If it is going down we will return the number of bytes that must
4472 * This must be called with BTRFS_I(inode)->lock held.
4474 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4477 struct btrfs_root *root = BTRFS_I(inode)->root;
4479 int num_csums_per_leaf;
4483 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4484 BTRFS_I(inode)->csum_bytes == 0)
4487 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4489 BTRFS_I(inode)->csum_bytes += num_bytes;
4491 BTRFS_I(inode)->csum_bytes -= num_bytes;
4492 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4493 num_csums_per_leaf = (int)div64_u64(csum_size,
4494 sizeof(struct btrfs_csum_item) +
4495 sizeof(struct btrfs_disk_key));
4496 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4497 num_csums = num_csums + num_csums_per_leaf - 1;
4498 num_csums = num_csums / num_csums_per_leaf;
4500 old_csums = old_csums + num_csums_per_leaf - 1;
4501 old_csums = old_csums / num_csums_per_leaf;
4503 /* No change, no need to reserve more */
4504 if (old_csums == num_csums)
4508 return btrfs_calc_trans_metadata_size(root,
4509 num_csums - old_csums);
4511 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4514 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4516 struct btrfs_root *root = BTRFS_I(inode)->root;
4517 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4520 unsigned nr_extents = 0;
4521 int extra_reserve = 0;
4522 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4524 bool delalloc_lock = true;
4526 /* If we are a free space inode we need to not flush since we will be in
4527 * the middle of a transaction commit. We also don't need the delalloc
4528 * mutex since we won't race with anybody. We need this mostly to make
4529 * lockdep shut its filthy mouth.
4531 if (btrfs_is_free_space_inode(inode)) {
4532 flush = BTRFS_RESERVE_NO_FLUSH;
4533 delalloc_lock = false;
4536 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4537 btrfs_transaction_in_commit(root->fs_info))
4538 schedule_timeout(1);
4541 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4543 num_bytes = ALIGN(num_bytes, root->sectorsize);
4545 spin_lock(&BTRFS_I(inode)->lock);
4546 BTRFS_I(inode)->outstanding_extents++;
4548 if (BTRFS_I(inode)->outstanding_extents >
4549 BTRFS_I(inode)->reserved_extents)
4550 nr_extents = BTRFS_I(inode)->outstanding_extents -
4551 BTRFS_I(inode)->reserved_extents;
4554 * Add an item to reserve for updating the inode when we complete the
4557 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4558 &BTRFS_I(inode)->runtime_flags)) {
4563 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4564 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4565 csum_bytes = BTRFS_I(inode)->csum_bytes;
4566 spin_unlock(&BTRFS_I(inode)->lock);
4568 if (root->fs_info->quota_enabled)
4569 ret = btrfs_qgroup_reserve(root, num_bytes +
4570 nr_extents * root->leafsize);
4573 * ret != 0 here means the qgroup reservation failed, we go straight to
4574 * the shared error handling then.
4577 ret = reserve_metadata_bytes(root, block_rsv,
4584 spin_lock(&BTRFS_I(inode)->lock);
4585 dropped = drop_outstanding_extent(inode);
4587 * If the inodes csum_bytes is the same as the original
4588 * csum_bytes then we know we haven't raced with any free()ers
4589 * so we can just reduce our inodes csum bytes and carry on.
4590 * Otherwise we have to do the normal free thing to account for
4591 * the case that the free side didn't free up its reserve
4592 * because of this outstanding reservation.
4594 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4595 calc_csum_metadata_size(inode, num_bytes, 0);
4597 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4598 spin_unlock(&BTRFS_I(inode)->lock);
4600 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4603 btrfs_block_rsv_release(root, block_rsv, to_free);
4604 trace_btrfs_space_reservation(root->fs_info,
4609 if (root->fs_info->quota_enabled) {
4610 btrfs_qgroup_free(root, num_bytes +
4611 nr_extents * root->leafsize);
4614 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4618 spin_lock(&BTRFS_I(inode)->lock);
4619 if (extra_reserve) {
4620 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4621 &BTRFS_I(inode)->runtime_flags);
4624 BTRFS_I(inode)->reserved_extents += nr_extents;
4625 spin_unlock(&BTRFS_I(inode)->lock);
4628 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4631 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4632 btrfs_ino(inode), to_reserve, 1);
4633 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4639 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4640 * @inode: the inode to release the reservation for
4641 * @num_bytes: the number of bytes we're releasing
4643 * This will release the metadata reservation for an inode. This can be called
4644 * once we complete IO for a given set of bytes to release their metadata
4647 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4649 struct btrfs_root *root = BTRFS_I(inode)->root;
4653 num_bytes = ALIGN(num_bytes, root->sectorsize);
4654 spin_lock(&BTRFS_I(inode)->lock);
4655 dropped = drop_outstanding_extent(inode);
4657 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4658 spin_unlock(&BTRFS_I(inode)->lock);
4660 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4662 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4663 btrfs_ino(inode), to_free, 0);
4664 if (root->fs_info->quota_enabled) {
4665 btrfs_qgroup_free(root, num_bytes +
4666 dropped * root->leafsize);
4669 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4674 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4675 * @inode: inode we're writing to
4676 * @num_bytes: the number of bytes we want to allocate
4678 * This will do the following things
4680 * o reserve space in the data space info for num_bytes
4681 * o reserve space in the metadata space info based on number of outstanding
4682 * extents and how much csums will be needed
4683 * o add to the inodes ->delalloc_bytes
4684 * o add it to the fs_info's delalloc inodes list.
4686 * This will return 0 for success and -ENOSPC if there is no space left.
4688 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4692 ret = btrfs_check_data_free_space(inode, num_bytes);
4696 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4698 btrfs_free_reserved_data_space(inode, num_bytes);
4706 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4707 * @inode: inode we're releasing space for
4708 * @num_bytes: the number of bytes we want to free up
4710 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4711 * called in the case that we don't need the metadata AND data reservations
4712 * anymore. So if there is an error or we insert an inline extent.
4714 * This function will release the metadata space that was not used and will
4715 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4716 * list if there are no delalloc bytes left.
4718 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4720 btrfs_delalloc_release_metadata(inode, num_bytes);
4721 btrfs_free_reserved_data_space(inode, num_bytes);
4724 static int update_block_group(struct btrfs_trans_handle *trans,
4725 struct btrfs_root *root,
4726 u64 bytenr, u64 num_bytes, int alloc)
4728 struct btrfs_block_group_cache *cache = NULL;
4729 struct btrfs_fs_info *info = root->fs_info;
4730 u64 total = num_bytes;
4735 /* block accounting for super block */
4736 spin_lock(&info->delalloc_lock);
4737 old_val = btrfs_super_bytes_used(info->super_copy);
4739 old_val += num_bytes;
4741 old_val -= num_bytes;
4742 btrfs_set_super_bytes_used(info->super_copy, old_val);
4743 spin_unlock(&info->delalloc_lock);
4746 cache = btrfs_lookup_block_group(info, bytenr);
4749 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4750 BTRFS_BLOCK_GROUP_RAID1 |
4751 BTRFS_BLOCK_GROUP_RAID10))
4756 * If this block group has free space cache written out, we
4757 * need to make sure to load it if we are removing space. This
4758 * is because we need the unpinning stage to actually add the
4759 * space back to the block group, otherwise we will leak space.
4761 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4762 cache_block_group(cache, trans, NULL, 1);
4764 byte_in_group = bytenr - cache->key.objectid;
4765 WARN_ON(byte_in_group > cache->key.offset);
4767 spin_lock(&cache->space_info->lock);
4768 spin_lock(&cache->lock);
4770 if (btrfs_test_opt(root, SPACE_CACHE) &&
4771 cache->disk_cache_state < BTRFS_DC_CLEAR)
4772 cache->disk_cache_state = BTRFS_DC_CLEAR;
4775 old_val = btrfs_block_group_used(&cache->item);
4776 num_bytes = min(total, cache->key.offset - byte_in_group);
4778 old_val += num_bytes;
4779 btrfs_set_block_group_used(&cache->item, old_val);
4780 cache->reserved -= num_bytes;
4781 cache->space_info->bytes_reserved -= num_bytes;
4782 cache->space_info->bytes_used += num_bytes;
4783 cache->space_info->disk_used += num_bytes * factor;
4784 spin_unlock(&cache->lock);
4785 spin_unlock(&cache->space_info->lock);
4787 old_val -= num_bytes;
4788 btrfs_set_block_group_used(&cache->item, old_val);
4789 cache->pinned += num_bytes;
4790 cache->space_info->bytes_pinned += num_bytes;
4791 cache->space_info->bytes_used -= num_bytes;
4792 cache->space_info->disk_used -= num_bytes * factor;
4793 spin_unlock(&cache->lock);
4794 spin_unlock(&cache->space_info->lock);
4796 set_extent_dirty(info->pinned_extents,
4797 bytenr, bytenr + num_bytes - 1,
4798 GFP_NOFS | __GFP_NOFAIL);
4800 btrfs_put_block_group(cache);
4802 bytenr += num_bytes;
4807 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4809 struct btrfs_block_group_cache *cache;
4812 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4816 bytenr = cache->key.objectid;
4817 btrfs_put_block_group(cache);
4822 static int pin_down_extent(struct btrfs_root *root,
4823 struct btrfs_block_group_cache *cache,
4824 u64 bytenr, u64 num_bytes, int reserved)
4826 spin_lock(&cache->space_info->lock);
4827 spin_lock(&cache->lock);
4828 cache->pinned += num_bytes;
4829 cache->space_info->bytes_pinned += num_bytes;
4831 cache->reserved -= num_bytes;
4832 cache->space_info->bytes_reserved -= num_bytes;
4834 spin_unlock(&cache->lock);
4835 spin_unlock(&cache->space_info->lock);
4837 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4838 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4843 * this function must be called within transaction
4845 int btrfs_pin_extent(struct btrfs_root *root,
4846 u64 bytenr, u64 num_bytes, int reserved)
4848 struct btrfs_block_group_cache *cache;
4850 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4851 BUG_ON(!cache); /* Logic error */
4853 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4855 btrfs_put_block_group(cache);
4860 * this function must be called within transaction
4862 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4863 struct btrfs_root *root,
4864 u64 bytenr, u64 num_bytes)
4866 struct btrfs_block_group_cache *cache;
4868 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4869 BUG_ON(!cache); /* Logic error */
4872 * pull in the free space cache (if any) so that our pin
4873 * removes the free space from the cache. We have load_only set
4874 * to one because the slow code to read in the free extents does check
4875 * the pinned extents.
4877 cache_block_group(cache, trans, root, 1);
4879 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4881 /* remove us from the free space cache (if we're there at all) */
4882 btrfs_remove_free_space(cache, bytenr, num_bytes);
4883 btrfs_put_block_group(cache);
4888 * btrfs_update_reserved_bytes - update the block_group and space info counters
4889 * @cache: The cache we are manipulating
4890 * @num_bytes: The number of bytes in question
4891 * @reserve: One of the reservation enums
4893 * This is called by the allocator when it reserves space, or by somebody who is
4894 * freeing space that was never actually used on disk. For example if you
4895 * reserve some space for a new leaf in transaction A and before transaction A
4896 * commits you free that leaf, you call this with reserve set to 0 in order to
4897 * clear the reservation.
4899 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4900 * ENOSPC accounting. For data we handle the reservation through clearing the
4901 * delalloc bits in the io_tree. We have to do this since we could end up
4902 * allocating less disk space for the amount of data we have reserved in the
4903 * case of compression.
4905 * If this is a reservation and the block group has become read only we cannot
4906 * make the reservation and return -EAGAIN, otherwise this function always
4909 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4910 u64 num_bytes, int reserve)
4912 struct btrfs_space_info *space_info = cache->space_info;
4915 spin_lock(&space_info->lock);
4916 spin_lock(&cache->lock);
4917 if (reserve != RESERVE_FREE) {
4921 cache->reserved += num_bytes;
4922 space_info->bytes_reserved += num_bytes;
4923 if (reserve == RESERVE_ALLOC) {
4924 trace_btrfs_space_reservation(cache->fs_info,
4925 "space_info", space_info->flags,
4927 space_info->bytes_may_use -= num_bytes;
4932 space_info->bytes_readonly += num_bytes;
4933 cache->reserved -= num_bytes;
4934 space_info->bytes_reserved -= num_bytes;
4935 space_info->reservation_progress++;
4937 spin_unlock(&cache->lock);
4938 spin_unlock(&space_info->lock);
4942 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4943 struct btrfs_root *root)
4945 struct btrfs_fs_info *fs_info = root->fs_info;
4946 struct btrfs_caching_control *next;
4947 struct btrfs_caching_control *caching_ctl;
4948 struct btrfs_block_group_cache *cache;
4950 down_write(&fs_info->extent_commit_sem);
4952 list_for_each_entry_safe(caching_ctl, next,
4953 &fs_info->caching_block_groups, list) {
4954 cache = caching_ctl->block_group;
4955 if (block_group_cache_done(cache)) {
4956 cache->last_byte_to_unpin = (u64)-1;
4957 list_del_init(&caching_ctl->list);
4958 put_caching_control(caching_ctl);
4960 cache->last_byte_to_unpin = caching_ctl->progress;
4964 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4965 fs_info->pinned_extents = &fs_info->freed_extents[1];
4967 fs_info->pinned_extents = &fs_info->freed_extents[0];
4969 up_write(&fs_info->extent_commit_sem);
4971 update_global_block_rsv(fs_info);
4974 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4976 struct btrfs_fs_info *fs_info = root->fs_info;
4977 struct btrfs_block_group_cache *cache = NULL;
4978 struct btrfs_space_info *space_info;
4979 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4983 while (start <= end) {
4986 start >= cache->key.objectid + cache->key.offset) {
4988 btrfs_put_block_group(cache);
4989 cache = btrfs_lookup_block_group(fs_info, start);
4990 BUG_ON(!cache); /* Logic error */
4993 len = cache->key.objectid + cache->key.offset - start;
4994 len = min(len, end + 1 - start);
4996 if (start < cache->last_byte_to_unpin) {
4997 len = min(len, cache->last_byte_to_unpin - start);
4998 btrfs_add_free_space(cache, start, len);
5002 space_info = cache->space_info;
5004 spin_lock(&space_info->lock);
5005 spin_lock(&cache->lock);
5006 cache->pinned -= len;
5007 space_info->bytes_pinned -= len;
5009 space_info->bytes_readonly += len;
5012 spin_unlock(&cache->lock);
5013 if (!readonly && global_rsv->space_info == space_info) {
5014 spin_lock(&global_rsv->lock);
5015 if (!global_rsv->full) {
5016 len = min(len, global_rsv->size -
5017 global_rsv->reserved);
5018 global_rsv->reserved += len;
5019 space_info->bytes_may_use += len;
5020 if (global_rsv->reserved >= global_rsv->size)
5021 global_rsv->full = 1;
5023 spin_unlock(&global_rsv->lock);
5025 spin_unlock(&space_info->lock);
5029 btrfs_put_block_group(cache);
5033 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5034 struct btrfs_root *root)
5036 struct btrfs_fs_info *fs_info = root->fs_info;
5037 struct extent_io_tree *unpin;
5045 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5046 unpin = &fs_info->freed_extents[1];
5048 unpin = &fs_info->freed_extents[0];
5051 ret = find_first_extent_bit(unpin, 0, &start, &end,
5052 EXTENT_DIRTY, NULL);
5056 if (btrfs_test_opt(root, DISCARD))
5057 ret = btrfs_discard_extent(root, start,
5058 end + 1 - start, NULL);
5060 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5061 unpin_extent_range(root, start, end);
5068 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5069 struct btrfs_root *root,
5070 u64 bytenr, u64 num_bytes, u64 parent,
5071 u64 root_objectid, u64 owner_objectid,
5072 u64 owner_offset, int refs_to_drop,
5073 struct btrfs_delayed_extent_op *extent_op)
5075 struct btrfs_key key;
5076 struct btrfs_path *path;
5077 struct btrfs_fs_info *info = root->fs_info;
5078 struct btrfs_root *extent_root = info->extent_root;
5079 struct extent_buffer *leaf;
5080 struct btrfs_extent_item *ei;
5081 struct btrfs_extent_inline_ref *iref;
5084 int extent_slot = 0;
5085 int found_extent = 0;
5090 path = btrfs_alloc_path();
5095 path->leave_spinning = 1;
5097 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5098 BUG_ON(!is_data && refs_to_drop != 1);
5100 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5101 bytenr, num_bytes, parent,
5102 root_objectid, owner_objectid,
5105 extent_slot = path->slots[0];
5106 while (extent_slot >= 0) {
5107 btrfs_item_key_to_cpu(path->nodes[0], &key,
5109 if (key.objectid != bytenr)
5111 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5112 key.offset == num_bytes) {
5116 if (path->slots[0] - extent_slot > 5)
5120 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5121 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5122 if (found_extent && item_size < sizeof(*ei))
5125 if (!found_extent) {
5127 ret = remove_extent_backref(trans, extent_root, path,
5131 btrfs_abort_transaction(trans, extent_root, ret);
5134 btrfs_release_path(path);
5135 path->leave_spinning = 1;
5137 key.objectid = bytenr;
5138 key.type = BTRFS_EXTENT_ITEM_KEY;
5139 key.offset = num_bytes;
5141 ret = btrfs_search_slot(trans, extent_root,
5144 printk(KERN_ERR "umm, got %d back from search"
5145 ", was looking for %llu\n", ret,
5146 (unsigned long long)bytenr);
5148 btrfs_print_leaf(extent_root,
5152 btrfs_abort_transaction(trans, extent_root, ret);
5155 extent_slot = path->slots[0];
5157 } else if (ret == -ENOENT) {
5158 btrfs_print_leaf(extent_root, path->nodes[0]);
5160 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5161 "parent %llu root %llu owner %llu offset %llu\n",
5162 (unsigned long long)bytenr,
5163 (unsigned long long)parent,
5164 (unsigned long long)root_objectid,
5165 (unsigned long long)owner_objectid,
5166 (unsigned long long)owner_offset);
5168 btrfs_abort_transaction(trans, extent_root, ret);
5172 leaf = path->nodes[0];
5173 item_size = btrfs_item_size_nr(leaf, extent_slot);
5174 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5175 if (item_size < sizeof(*ei)) {
5176 BUG_ON(found_extent || extent_slot != path->slots[0]);
5177 ret = convert_extent_item_v0(trans, extent_root, path,
5180 btrfs_abort_transaction(trans, extent_root, ret);
5184 btrfs_release_path(path);
5185 path->leave_spinning = 1;
5187 key.objectid = bytenr;
5188 key.type = BTRFS_EXTENT_ITEM_KEY;
5189 key.offset = num_bytes;
5191 ret = btrfs_search_slot(trans, extent_root, &key, path,
5194 printk(KERN_ERR "umm, got %d back from search"
5195 ", was looking for %llu\n", ret,
5196 (unsigned long long)bytenr);
5197 btrfs_print_leaf(extent_root, path->nodes[0]);
5200 btrfs_abort_transaction(trans, extent_root, ret);
5204 extent_slot = path->slots[0];
5205 leaf = path->nodes[0];
5206 item_size = btrfs_item_size_nr(leaf, extent_slot);
5209 BUG_ON(item_size < sizeof(*ei));
5210 ei = btrfs_item_ptr(leaf, extent_slot,
5211 struct btrfs_extent_item);
5212 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5213 struct btrfs_tree_block_info *bi;
5214 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5215 bi = (struct btrfs_tree_block_info *)(ei + 1);
5216 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5219 refs = btrfs_extent_refs(leaf, ei);
5220 BUG_ON(refs < refs_to_drop);
5221 refs -= refs_to_drop;
5225 __run_delayed_extent_op(extent_op, leaf, ei);
5227 * In the case of inline back ref, reference count will
5228 * be updated by remove_extent_backref
5231 BUG_ON(!found_extent);
5233 btrfs_set_extent_refs(leaf, ei, refs);
5234 btrfs_mark_buffer_dirty(leaf);
5237 ret = remove_extent_backref(trans, extent_root, path,
5241 btrfs_abort_transaction(trans, extent_root, ret);
5247 BUG_ON(is_data && refs_to_drop !=
5248 extent_data_ref_count(root, path, iref));
5250 BUG_ON(path->slots[0] != extent_slot);
5252 BUG_ON(path->slots[0] != extent_slot + 1);
5253 path->slots[0] = extent_slot;
5258 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5261 btrfs_abort_transaction(trans, extent_root, ret);
5264 btrfs_release_path(path);
5267 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5269 btrfs_abort_transaction(trans, extent_root, ret);
5274 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5276 btrfs_abort_transaction(trans, extent_root, ret);
5281 btrfs_free_path(path);
5286 * when we free an block, it is possible (and likely) that we free the last
5287 * delayed ref for that extent as well. This searches the delayed ref tree for
5288 * a given extent, and if there are no other delayed refs to be processed, it
5289 * removes it from the tree.
5291 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5292 struct btrfs_root *root, u64 bytenr)
5294 struct btrfs_delayed_ref_head *head;
5295 struct btrfs_delayed_ref_root *delayed_refs;
5296 struct btrfs_delayed_ref_node *ref;
5297 struct rb_node *node;
5300 delayed_refs = &trans->transaction->delayed_refs;
5301 spin_lock(&delayed_refs->lock);
5302 head = btrfs_find_delayed_ref_head(trans, bytenr);
5306 node = rb_prev(&head->node.rb_node);
5310 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5312 /* there are still entries for this ref, we can't drop it */
5313 if (ref->bytenr == bytenr)
5316 if (head->extent_op) {
5317 if (!head->must_insert_reserved)
5319 kfree(head->extent_op);
5320 head->extent_op = NULL;
5324 * waiting for the lock here would deadlock. If someone else has it
5325 * locked they are already in the process of dropping it anyway
5327 if (!mutex_trylock(&head->mutex))
5331 * at this point we have a head with no other entries. Go
5332 * ahead and process it.
5334 head->node.in_tree = 0;
5335 rb_erase(&head->node.rb_node, &delayed_refs->root);
5337 delayed_refs->num_entries--;
5340 * we don't take a ref on the node because we're removing it from the
5341 * tree, so we just steal the ref the tree was holding.
5343 delayed_refs->num_heads--;
5344 if (list_empty(&head->cluster))
5345 delayed_refs->num_heads_ready--;
5347 list_del_init(&head->cluster);
5348 spin_unlock(&delayed_refs->lock);
5350 BUG_ON(head->extent_op);
5351 if (head->must_insert_reserved)
5354 mutex_unlock(&head->mutex);
5355 btrfs_put_delayed_ref(&head->node);
5358 spin_unlock(&delayed_refs->lock);
5362 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5363 struct btrfs_root *root,
5364 struct extent_buffer *buf,
5365 u64 parent, int last_ref)
5367 struct btrfs_block_group_cache *cache = NULL;
5370 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5371 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5372 buf->start, buf->len,
5373 parent, root->root_key.objectid,
5374 btrfs_header_level(buf),
5375 BTRFS_DROP_DELAYED_REF, NULL, 0);
5376 BUG_ON(ret); /* -ENOMEM */
5382 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5384 if (btrfs_header_generation(buf) == trans->transid) {
5385 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5386 ret = check_ref_cleanup(trans, root, buf->start);
5391 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5392 pin_down_extent(root, cache, buf->start, buf->len, 1);
5396 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5398 btrfs_add_free_space(cache, buf->start, buf->len);
5399 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5403 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5406 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5407 btrfs_put_block_group(cache);
5410 /* Can return -ENOMEM */
5411 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5412 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5413 u64 owner, u64 offset, int for_cow)
5416 struct btrfs_fs_info *fs_info = root->fs_info;
5419 * tree log blocks never actually go into the extent allocation
5420 * tree, just update pinning info and exit early.
5422 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5423 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5424 /* unlocks the pinned mutex */
5425 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5427 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5428 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5430 parent, root_objectid, (int)owner,
5431 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5433 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5435 parent, root_objectid, owner,
5436 offset, BTRFS_DROP_DELAYED_REF,
5442 static u64 stripe_align(struct btrfs_root *root, u64 val)
5444 u64 mask = ((u64)root->stripesize - 1);
5445 u64 ret = (val + mask) & ~mask;
5450 * when we wait for progress in the block group caching, its because
5451 * our allocation attempt failed at least once. So, we must sleep
5452 * and let some progress happen before we try again.
5454 * This function will sleep at least once waiting for new free space to
5455 * show up, and then it will check the block group free space numbers
5456 * for our min num_bytes. Another option is to have it go ahead
5457 * and look in the rbtree for a free extent of a given size, but this
5461 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5464 struct btrfs_caching_control *caching_ctl;
5467 caching_ctl = get_caching_control(cache);
5471 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5472 (cache->free_space_ctl->free_space >= num_bytes));
5474 put_caching_control(caching_ctl);
5479 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5481 struct btrfs_caching_control *caching_ctl;
5484 caching_ctl = get_caching_control(cache);
5488 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5490 put_caching_control(caching_ctl);
5494 int __get_raid_index(u64 flags)
5498 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5500 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5502 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5504 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5512 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5514 return __get_raid_index(cache->flags);
5517 enum btrfs_loop_type {
5518 LOOP_CACHING_NOWAIT = 0,
5519 LOOP_CACHING_WAIT = 1,
5520 LOOP_ALLOC_CHUNK = 2,
5521 LOOP_NO_EMPTY_SIZE = 3,
5525 * walks the btree of allocated extents and find a hole of a given size.
5526 * The key ins is changed to record the hole:
5527 * ins->objectid == block start
5528 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5529 * ins->offset == number of blocks
5530 * Any available blocks before search_start are skipped.
5532 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5533 struct btrfs_root *orig_root,
5534 u64 num_bytes, u64 empty_size,
5535 u64 hint_byte, struct btrfs_key *ins,
5539 struct btrfs_root *root = orig_root->fs_info->extent_root;
5540 struct btrfs_free_cluster *last_ptr = NULL;
5541 struct btrfs_block_group_cache *block_group = NULL;
5542 struct btrfs_block_group_cache *used_block_group;
5543 u64 search_start = 0;
5544 int empty_cluster = 2 * 1024 * 1024;
5545 struct btrfs_space_info *space_info;
5547 int index = __get_raid_index(data);
5548 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5549 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5550 bool found_uncached_bg = false;
5551 bool failed_cluster_refill = false;
5552 bool failed_alloc = false;
5553 bool use_cluster = true;
5554 bool have_caching_bg = false;
5556 WARN_ON(num_bytes < root->sectorsize);
5557 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5561 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5563 space_info = __find_space_info(root->fs_info, data);
5565 printk(KERN_ERR "No space info for %llu\n", data);
5570 * If the space info is for both data and metadata it means we have a
5571 * small filesystem and we can't use the clustering stuff.
5573 if (btrfs_mixed_space_info(space_info))
5574 use_cluster = false;
5576 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5577 last_ptr = &root->fs_info->meta_alloc_cluster;
5578 if (!btrfs_test_opt(root, SSD))
5579 empty_cluster = 64 * 1024;
5582 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5583 btrfs_test_opt(root, SSD)) {
5584 last_ptr = &root->fs_info->data_alloc_cluster;
5588 spin_lock(&last_ptr->lock);
5589 if (last_ptr->block_group)
5590 hint_byte = last_ptr->window_start;
5591 spin_unlock(&last_ptr->lock);
5594 search_start = max(search_start, first_logical_byte(root, 0));
5595 search_start = max(search_start, hint_byte);
5600 if (search_start == hint_byte) {
5601 block_group = btrfs_lookup_block_group(root->fs_info,
5603 used_block_group = block_group;
5605 * we don't want to use the block group if it doesn't match our
5606 * allocation bits, or if its not cached.
5608 * However if we are re-searching with an ideal block group
5609 * picked out then we don't care that the block group is cached.
5611 if (block_group && block_group_bits(block_group, data) &&
5612 block_group->cached != BTRFS_CACHE_NO) {
5613 down_read(&space_info->groups_sem);
5614 if (list_empty(&block_group->list) ||
5617 * someone is removing this block group,
5618 * we can't jump into the have_block_group
5619 * target because our list pointers are not
5622 btrfs_put_block_group(block_group);
5623 up_read(&space_info->groups_sem);
5625 index = get_block_group_index(block_group);
5626 goto have_block_group;
5628 } else if (block_group) {
5629 btrfs_put_block_group(block_group);
5633 have_caching_bg = false;
5634 down_read(&space_info->groups_sem);
5635 list_for_each_entry(block_group, &space_info->block_groups[index],
5640 used_block_group = block_group;
5641 btrfs_get_block_group(block_group);
5642 search_start = block_group->key.objectid;
5645 * this can happen if we end up cycling through all the
5646 * raid types, but we want to make sure we only allocate
5647 * for the proper type.
5649 if (!block_group_bits(block_group, data)) {
5650 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5651 BTRFS_BLOCK_GROUP_RAID1 |
5652 BTRFS_BLOCK_GROUP_RAID10;
5655 * if they asked for extra copies and this block group
5656 * doesn't provide them, bail. This does allow us to
5657 * fill raid0 from raid1.
5659 if ((data & extra) && !(block_group->flags & extra))
5664 cached = block_group_cache_done(block_group);
5665 if (unlikely(!cached)) {
5666 found_uncached_bg = true;
5667 ret = cache_block_group(block_group, trans,
5673 if (unlikely(block_group->ro))
5677 * Ok we want to try and use the cluster allocator, so
5682 * the refill lock keeps out other
5683 * people trying to start a new cluster
5685 spin_lock(&last_ptr->refill_lock);
5686 used_block_group = last_ptr->block_group;
5687 if (used_block_group != block_group &&
5688 (!used_block_group ||
5689 used_block_group->ro ||
5690 !block_group_bits(used_block_group, data))) {
5691 used_block_group = block_group;
5692 goto refill_cluster;
5695 if (used_block_group != block_group)
5696 btrfs_get_block_group(used_block_group);
5698 offset = btrfs_alloc_from_cluster(used_block_group,
5699 last_ptr, num_bytes, used_block_group->key.objectid);
5701 /* we have a block, we're done */
5702 spin_unlock(&last_ptr->refill_lock);
5703 trace_btrfs_reserve_extent_cluster(root,
5704 block_group, search_start, num_bytes);
5708 WARN_ON(last_ptr->block_group != used_block_group);
5709 if (used_block_group != block_group) {
5710 btrfs_put_block_group(used_block_group);
5711 used_block_group = block_group;
5714 BUG_ON(used_block_group != block_group);
5715 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5716 * set up a new clusters, so lets just skip it
5717 * and let the allocator find whatever block
5718 * it can find. If we reach this point, we
5719 * will have tried the cluster allocator
5720 * plenty of times and not have found
5721 * anything, so we are likely way too
5722 * fragmented for the clustering stuff to find
5725 * However, if the cluster is taken from the
5726 * current block group, release the cluster
5727 * first, so that we stand a better chance of
5728 * succeeding in the unclustered
5730 if (loop >= LOOP_NO_EMPTY_SIZE &&
5731 last_ptr->block_group != block_group) {
5732 spin_unlock(&last_ptr->refill_lock);
5733 goto unclustered_alloc;
5737 * this cluster didn't work out, free it and
5740 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5742 if (loop >= LOOP_NO_EMPTY_SIZE) {
5743 spin_unlock(&last_ptr->refill_lock);
5744 goto unclustered_alloc;
5747 /* allocate a cluster in this block group */
5748 ret = btrfs_find_space_cluster(trans, root,
5749 block_group, last_ptr,
5750 search_start, num_bytes,
5751 empty_cluster + empty_size);
5754 * now pull our allocation out of this
5757 offset = btrfs_alloc_from_cluster(block_group,
5758 last_ptr, num_bytes,
5761 /* we found one, proceed */
5762 spin_unlock(&last_ptr->refill_lock);
5763 trace_btrfs_reserve_extent_cluster(root,
5764 block_group, search_start,
5768 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5769 && !failed_cluster_refill) {
5770 spin_unlock(&last_ptr->refill_lock);
5772 failed_cluster_refill = true;
5773 wait_block_group_cache_progress(block_group,
5774 num_bytes + empty_cluster + empty_size);
5775 goto have_block_group;
5779 * at this point we either didn't find a cluster
5780 * or we weren't able to allocate a block from our
5781 * cluster. Free the cluster we've been trying
5782 * to use, and go to the next block group
5784 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5785 spin_unlock(&last_ptr->refill_lock);
5790 spin_lock(&block_group->free_space_ctl->tree_lock);
5792 block_group->free_space_ctl->free_space <
5793 num_bytes + empty_cluster + empty_size) {
5794 spin_unlock(&block_group->free_space_ctl->tree_lock);
5797 spin_unlock(&block_group->free_space_ctl->tree_lock);
5799 offset = btrfs_find_space_for_alloc(block_group, search_start,
5800 num_bytes, empty_size);
5802 * If we didn't find a chunk, and we haven't failed on this
5803 * block group before, and this block group is in the middle of
5804 * caching and we are ok with waiting, then go ahead and wait
5805 * for progress to be made, and set failed_alloc to true.
5807 * If failed_alloc is true then we've already waited on this
5808 * block group once and should move on to the next block group.
5810 if (!offset && !failed_alloc && !cached &&
5811 loop > LOOP_CACHING_NOWAIT) {
5812 wait_block_group_cache_progress(block_group,
5813 num_bytes + empty_size);
5814 failed_alloc = true;
5815 goto have_block_group;
5816 } else if (!offset) {
5818 have_caching_bg = true;
5822 search_start = stripe_align(root, offset);
5824 /* move on to the next group */
5825 if (search_start + num_bytes >
5826 used_block_group->key.objectid + used_block_group->key.offset) {
5827 btrfs_add_free_space(used_block_group, offset, num_bytes);
5831 if (offset < search_start)
5832 btrfs_add_free_space(used_block_group, offset,
5833 search_start - offset);
5834 BUG_ON(offset > search_start);
5836 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5838 if (ret == -EAGAIN) {
5839 btrfs_add_free_space(used_block_group, offset, num_bytes);
5843 /* we are all good, lets return */
5844 ins->objectid = search_start;
5845 ins->offset = num_bytes;
5847 trace_btrfs_reserve_extent(orig_root, block_group,
5848 search_start, num_bytes);
5849 if (used_block_group != block_group)
5850 btrfs_put_block_group(used_block_group);
5851 btrfs_put_block_group(block_group);
5854 failed_cluster_refill = false;
5855 failed_alloc = false;
5856 BUG_ON(index != get_block_group_index(block_group));
5857 if (used_block_group != block_group)
5858 btrfs_put_block_group(used_block_group);
5859 btrfs_put_block_group(block_group);
5861 up_read(&space_info->groups_sem);
5863 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5866 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5870 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5871 * caching kthreads as we move along
5872 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5873 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5874 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5877 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5880 if (loop == LOOP_ALLOC_CHUNK) {
5881 ret = do_chunk_alloc(trans, root, data,
5884 * Do not bail out on ENOSPC since we
5885 * can do more things.
5887 if (ret < 0 && ret != -ENOSPC) {
5888 btrfs_abort_transaction(trans,
5894 if (loop == LOOP_NO_EMPTY_SIZE) {
5900 } else if (!ins->objectid) {
5902 } else if (ins->objectid) {
5910 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5911 int dump_block_groups)
5913 struct btrfs_block_group_cache *cache;
5916 spin_lock(&info->lock);
5917 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5918 (unsigned long long)info->flags,
5919 (unsigned long long)(info->total_bytes - info->bytes_used -
5920 info->bytes_pinned - info->bytes_reserved -
5921 info->bytes_readonly),
5922 (info->full) ? "" : "not ");
5923 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5924 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5925 (unsigned long long)info->total_bytes,
5926 (unsigned long long)info->bytes_used,
5927 (unsigned long long)info->bytes_pinned,
5928 (unsigned long long)info->bytes_reserved,
5929 (unsigned long long)info->bytes_may_use,
5930 (unsigned long long)info->bytes_readonly);
5931 spin_unlock(&info->lock);
5933 if (!dump_block_groups)
5936 down_read(&info->groups_sem);
5938 list_for_each_entry(cache, &info->block_groups[index], list) {
5939 spin_lock(&cache->lock);
5940 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5941 (unsigned long long)cache->key.objectid,
5942 (unsigned long long)cache->key.offset,
5943 (unsigned long long)btrfs_block_group_used(&cache->item),
5944 (unsigned long long)cache->pinned,
5945 (unsigned long long)cache->reserved,
5946 cache->ro ? "[readonly]" : "");
5947 btrfs_dump_free_space(cache, bytes);
5948 spin_unlock(&cache->lock);
5950 if (++index < BTRFS_NR_RAID_TYPES)
5952 up_read(&info->groups_sem);
5955 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5956 struct btrfs_root *root,
5957 u64 num_bytes, u64 min_alloc_size,
5958 u64 empty_size, u64 hint_byte,
5959 struct btrfs_key *ins, u64 data)
5961 bool final_tried = false;
5964 data = btrfs_get_alloc_profile(root, data);
5966 WARN_ON(num_bytes < root->sectorsize);
5967 ret = find_free_extent(trans, root, num_bytes, empty_size,
5968 hint_byte, ins, data);
5970 if (ret == -ENOSPC) {
5972 num_bytes = num_bytes >> 1;
5973 num_bytes = num_bytes & ~(root->sectorsize - 1);
5974 num_bytes = max(num_bytes, min_alloc_size);
5975 if (num_bytes == min_alloc_size)
5978 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5979 struct btrfs_space_info *sinfo;
5981 sinfo = __find_space_info(root->fs_info, data);
5982 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5983 "wanted %llu\n", (unsigned long long)data,
5984 (unsigned long long)num_bytes);
5986 dump_space_info(sinfo, num_bytes, 1);
5990 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5995 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5996 u64 start, u64 len, int pin)
5998 struct btrfs_block_group_cache *cache;
6001 cache = btrfs_lookup_block_group(root->fs_info, start);
6003 printk(KERN_ERR "Unable to find block group for %llu\n",
6004 (unsigned long long)start);
6008 if (btrfs_test_opt(root, DISCARD))
6009 ret = btrfs_discard_extent(root, start, len, NULL);
6012 pin_down_extent(root, cache, start, len, 1);
6014 btrfs_add_free_space(cache, start, len);
6015 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6017 btrfs_put_block_group(cache);
6019 trace_btrfs_reserved_extent_free(root, start, len);
6024 int btrfs_free_reserved_extent(struct btrfs_root *root,
6027 return __btrfs_free_reserved_extent(root, start, len, 0);
6030 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6033 return __btrfs_free_reserved_extent(root, start, len, 1);
6036 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6037 struct btrfs_root *root,
6038 u64 parent, u64 root_objectid,
6039 u64 flags, u64 owner, u64 offset,
6040 struct btrfs_key *ins, int ref_mod)
6043 struct btrfs_fs_info *fs_info = root->fs_info;
6044 struct btrfs_extent_item *extent_item;
6045 struct btrfs_extent_inline_ref *iref;
6046 struct btrfs_path *path;
6047 struct extent_buffer *leaf;
6052 type = BTRFS_SHARED_DATA_REF_KEY;
6054 type = BTRFS_EXTENT_DATA_REF_KEY;
6056 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6058 path = btrfs_alloc_path();
6062 path->leave_spinning = 1;
6063 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6066 btrfs_free_path(path);
6070 leaf = path->nodes[0];
6071 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6072 struct btrfs_extent_item);
6073 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6074 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6075 btrfs_set_extent_flags(leaf, extent_item,
6076 flags | BTRFS_EXTENT_FLAG_DATA);
6078 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6079 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6081 struct btrfs_shared_data_ref *ref;
6082 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6083 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6084 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6086 struct btrfs_extent_data_ref *ref;
6087 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6088 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6089 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6090 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6091 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6094 btrfs_mark_buffer_dirty(path->nodes[0]);
6095 btrfs_free_path(path);
6097 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6098 if (ret) { /* -ENOENT, logic error */
6099 printk(KERN_ERR "btrfs update block group failed for %llu "
6100 "%llu\n", (unsigned long long)ins->objectid,
6101 (unsigned long long)ins->offset);
6107 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6108 struct btrfs_root *root,
6109 u64 parent, u64 root_objectid,
6110 u64 flags, struct btrfs_disk_key *key,
6111 int level, struct btrfs_key *ins)
6114 struct btrfs_fs_info *fs_info = root->fs_info;
6115 struct btrfs_extent_item *extent_item;
6116 struct btrfs_tree_block_info *block_info;
6117 struct btrfs_extent_inline_ref *iref;
6118 struct btrfs_path *path;
6119 struct extent_buffer *leaf;
6120 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6122 path = btrfs_alloc_path();
6126 path->leave_spinning = 1;
6127 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6130 btrfs_free_path(path);
6134 leaf = path->nodes[0];
6135 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6136 struct btrfs_extent_item);
6137 btrfs_set_extent_refs(leaf, extent_item, 1);
6138 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6139 btrfs_set_extent_flags(leaf, extent_item,
6140 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6141 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6143 btrfs_set_tree_block_key(leaf, block_info, key);
6144 btrfs_set_tree_block_level(leaf, block_info, level);
6146 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6148 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6149 btrfs_set_extent_inline_ref_type(leaf, iref,
6150 BTRFS_SHARED_BLOCK_REF_KEY);
6151 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6153 btrfs_set_extent_inline_ref_type(leaf, iref,
6154 BTRFS_TREE_BLOCK_REF_KEY);
6155 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6158 btrfs_mark_buffer_dirty(leaf);
6159 btrfs_free_path(path);
6161 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6162 if (ret) { /* -ENOENT, logic error */
6163 printk(KERN_ERR "btrfs update block group failed for %llu "
6164 "%llu\n", (unsigned long long)ins->objectid,
6165 (unsigned long long)ins->offset);
6171 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6172 struct btrfs_root *root,
6173 u64 root_objectid, u64 owner,
6174 u64 offset, struct btrfs_key *ins)
6178 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6180 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6182 root_objectid, owner, offset,
6183 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6188 * this is used by the tree logging recovery code. It records that
6189 * an extent has been allocated and makes sure to clear the free
6190 * space cache bits as well
6192 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6193 struct btrfs_root *root,
6194 u64 root_objectid, u64 owner, u64 offset,
6195 struct btrfs_key *ins)
6198 struct btrfs_block_group_cache *block_group;
6199 struct btrfs_caching_control *caching_ctl;
6200 u64 start = ins->objectid;
6201 u64 num_bytes = ins->offset;
6203 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6204 cache_block_group(block_group, trans, NULL, 0);
6205 caching_ctl = get_caching_control(block_group);
6208 BUG_ON(!block_group_cache_done(block_group));
6209 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6210 BUG_ON(ret); /* -ENOMEM */
6212 mutex_lock(&caching_ctl->mutex);
6214 if (start >= caching_ctl->progress) {
6215 ret = add_excluded_extent(root, start, num_bytes);
6216 BUG_ON(ret); /* -ENOMEM */
6217 } else if (start + num_bytes <= caching_ctl->progress) {
6218 ret = btrfs_remove_free_space(block_group,
6220 BUG_ON(ret); /* -ENOMEM */
6222 num_bytes = caching_ctl->progress - start;
6223 ret = btrfs_remove_free_space(block_group,
6225 BUG_ON(ret); /* -ENOMEM */
6227 start = caching_ctl->progress;
6228 num_bytes = ins->objectid + ins->offset -
6229 caching_ctl->progress;
6230 ret = add_excluded_extent(root, start, num_bytes);
6231 BUG_ON(ret); /* -ENOMEM */
6234 mutex_unlock(&caching_ctl->mutex);
6235 put_caching_control(caching_ctl);
6238 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6239 RESERVE_ALLOC_NO_ACCOUNT);
6240 BUG_ON(ret); /* logic error */
6241 btrfs_put_block_group(block_group);
6242 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6243 0, owner, offset, ins, 1);
6247 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6248 struct btrfs_root *root,
6249 u64 bytenr, u32 blocksize,
6252 struct extent_buffer *buf;
6254 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6256 return ERR_PTR(-ENOMEM);
6257 btrfs_set_header_generation(buf, trans->transid);
6258 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6259 btrfs_tree_lock(buf);
6260 clean_tree_block(trans, root, buf);
6261 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6263 btrfs_set_lock_blocking(buf);
6264 btrfs_set_buffer_uptodate(buf);
6266 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6268 * we allow two log transactions at a time, use different
6269 * EXENT bit to differentiate dirty pages.
6271 if (root->log_transid % 2 == 0)
6272 set_extent_dirty(&root->dirty_log_pages, buf->start,
6273 buf->start + buf->len - 1, GFP_NOFS);
6275 set_extent_new(&root->dirty_log_pages, buf->start,
6276 buf->start + buf->len - 1, GFP_NOFS);
6278 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6279 buf->start + buf->len - 1, GFP_NOFS);
6281 trans->blocks_used++;
6282 /* this returns a buffer locked for blocking */
6286 static struct btrfs_block_rsv *
6287 use_block_rsv(struct btrfs_trans_handle *trans,
6288 struct btrfs_root *root, u32 blocksize)
6290 struct btrfs_block_rsv *block_rsv;
6291 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6294 block_rsv = get_block_rsv(trans, root);
6296 if (block_rsv->size == 0) {
6297 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6298 BTRFS_RESERVE_NO_FLUSH);
6300 * If we couldn't reserve metadata bytes try and use some from
6301 * the global reserve.
6303 if (ret && block_rsv != global_rsv) {
6304 ret = block_rsv_use_bytes(global_rsv, blocksize);
6307 return ERR_PTR(ret);
6309 return ERR_PTR(ret);
6314 ret = block_rsv_use_bytes(block_rsv, blocksize);
6317 if (ret && !block_rsv->failfast) {
6318 static DEFINE_RATELIMIT_STATE(_rs,
6319 DEFAULT_RATELIMIT_INTERVAL,
6320 /*DEFAULT_RATELIMIT_BURST*/ 2);
6321 if (__ratelimit(&_rs))
6322 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6324 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6325 BTRFS_RESERVE_NO_FLUSH);
6328 } else if (ret && block_rsv != global_rsv) {
6329 ret = block_rsv_use_bytes(global_rsv, blocksize);
6335 return ERR_PTR(-ENOSPC);
6338 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6339 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6341 block_rsv_add_bytes(block_rsv, blocksize, 0);
6342 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6346 * finds a free extent and does all the dirty work required for allocation
6347 * returns the key for the extent through ins, and a tree buffer for
6348 * the first block of the extent through buf.
6350 * returns the tree buffer or NULL.
6352 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6353 struct btrfs_root *root, u32 blocksize,
6354 u64 parent, u64 root_objectid,
6355 struct btrfs_disk_key *key, int level,
6356 u64 hint, u64 empty_size)
6358 struct btrfs_key ins;
6359 struct btrfs_block_rsv *block_rsv;
6360 struct extent_buffer *buf;
6365 block_rsv = use_block_rsv(trans, root, blocksize);
6366 if (IS_ERR(block_rsv))
6367 return ERR_CAST(block_rsv);
6369 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6370 empty_size, hint, &ins, 0);
6372 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6373 return ERR_PTR(ret);
6376 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6378 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6380 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6382 parent = ins.objectid;
6383 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6387 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6388 struct btrfs_delayed_extent_op *extent_op;
6389 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6390 BUG_ON(!extent_op); /* -ENOMEM */
6392 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6394 memset(&extent_op->key, 0, sizeof(extent_op->key));
6395 extent_op->flags_to_set = flags;
6396 extent_op->update_key = 1;
6397 extent_op->update_flags = 1;
6398 extent_op->is_data = 0;
6400 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6402 ins.offset, parent, root_objectid,
6403 level, BTRFS_ADD_DELAYED_EXTENT,
6405 BUG_ON(ret); /* -ENOMEM */
6410 struct walk_control {
6411 u64 refs[BTRFS_MAX_LEVEL];
6412 u64 flags[BTRFS_MAX_LEVEL];
6413 struct btrfs_key update_progress;
6424 #define DROP_REFERENCE 1
6425 #define UPDATE_BACKREF 2
6427 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6428 struct btrfs_root *root,
6429 struct walk_control *wc,
6430 struct btrfs_path *path)
6438 struct btrfs_key key;
6439 struct extent_buffer *eb;
6444 if (path->slots[wc->level] < wc->reada_slot) {
6445 wc->reada_count = wc->reada_count * 2 / 3;
6446 wc->reada_count = max(wc->reada_count, 2);
6448 wc->reada_count = wc->reada_count * 3 / 2;
6449 wc->reada_count = min_t(int, wc->reada_count,
6450 BTRFS_NODEPTRS_PER_BLOCK(root));
6453 eb = path->nodes[wc->level];
6454 nritems = btrfs_header_nritems(eb);
6455 blocksize = btrfs_level_size(root, wc->level - 1);
6457 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6458 if (nread >= wc->reada_count)
6462 bytenr = btrfs_node_blockptr(eb, slot);
6463 generation = btrfs_node_ptr_generation(eb, slot);
6465 if (slot == path->slots[wc->level])
6468 if (wc->stage == UPDATE_BACKREF &&
6469 generation <= root->root_key.offset)
6472 /* We don't lock the tree block, it's OK to be racy here */
6473 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6475 /* We don't care about errors in readahead. */
6480 if (wc->stage == DROP_REFERENCE) {
6484 if (wc->level == 1 &&
6485 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6487 if (!wc->update_ref ||
6488 generation <= root->root_key.offset)
6490 btrfs_node_key_to_cpu(eb, &key, slot);
6491 ret = btrfs_comp_cpu_keys(&key,
6492 &wc->update_progress);
6496 if (wc->level == 1 &&
6497 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6501 ret = readahead_tree_block(root, bytenr, blocksize,
6507 wc->reada_slot = slot;
6511 * helper to process tree block while walking down the tree.
6513 * when wc->stage == UPDATE_BACKREF, this function updates
6514 * back refs for pointers in the block.
6516 * NOTE: return value 1 means we should stop walking down.
6518 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6519 struct btrfs_root *root,
6520 struct btrfs_path *path,
6521 struct walk_control *wc, int lookup_info)
6523 int level = wc->level;
6524 struct extent_buffer *eb = path->nodes[level];
6525 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6528 if (wc->stage == UPDATE_BACKREF &&
6529 btrfs_header_owner(eb) != root->root_key.objectid)
6533 * when reference count of tree block is 1, it won't increase
6534 * again. once full backref flag is set, we never clear it.
6537 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6538 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6539 BUG_ON(!path->locks[level]);
6540 ret = btrfs_lookup_extent_info(trans, root,
6544 BUG_ON(ret == -ENOMEM);
6547 BUG_ON(wc->refs[level] == 0);
6550 if (wc->stage == DROP_REFERENCE) {
6551 if (wc->refs[level] > 1)
6554 if (path->locks[level] && !wc->keep_locks) {
6555 btrfs_tree_unlock_rw(eb, path->locks[level]);
6556 path->locks[level] = 0;
6561 /* wc->stage == UPDATE_BACKREF */
6562 if (!(wc->flags[level] & flag)) {
6563 BUG_ON(!path->locks[level]);
6564 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6565 BUG_ON(ret); /* -ENOMEM */
6566 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6567 BUG_ON(ret); /* -ENOMEM */
6568 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6570 BUG_ON(ret); /* -ENOMEM */
6571 wc->flags[level] |= flag;
6575 * the block is shared by multiple trees, so it's not good to
6576 * keep the tree lock
6578 if (path->locks[level] && level > 0) {
6579 btrfs_tree_unlock_rw(eb, path->locks[level]);
6580 path->locks[level] = 0;
6586 * helper to process tree block pointer.
6588 * when wc->stage == DROP_REFERENCE, this function checks
6589 * reference count of the block pointed to. if the block
6590 * is shared and we need update back refs for the subtree
6591 * rooted at the block, this function changes wc->stage to
6592 * UPDATE_BACKREF. if the block is shared and there is no
6593 * need to update back, this function drops the reference
6596 * NOTE: return value 1 means we should stop walking down.
6598 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6599 struct btrfs_root *root,
6600 struct btrfs_path *path,
6601 struct walk_control *wc, int *lookup_info)
6607 struct btrfs_key key;
6608 struct extent_buffer *next;
6609 int level = wc->level;
6613 generation = btrfs_node_ptr_generation(path->nodes[level],
6614 path->slots[level]);
6616 * if the lower level block was created before the snapshot
6617 * was created, we know there is no need to update back refs
6620 if (wc->stage == UPDATE_BACKREF &&
6621 generation <= root->root_key.offset) {
6626 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6627 blocksize = btrfs_level_size(root, level - 1);
6629 next = btrfs_find_tree_block(root, bytenr, blocksize);
6631 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6636 btrfs_tree_lock(next);
6637 btrfs_set_lock_blocking(next);
6639 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6640 &wc->refs[level - 1],
6641 &wc->flags[level - 1]);
6643 btrfs_tree_unlock(next);
6647 BUG_ON(wc->refs[level - 1] == 0);
6650 if (wc->stage == DROP_REFERENCE) {
6651 if (wc->refs[level - 1] > 1) {
6653 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6656 if (!wc->update_ref ||
6657 generation <= root->root_key.offset)
6660 btrfs_node_key_to_cpu(path->nodes[level], &key,
6661 path->slots[level]);
6662 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6666 wc->stage = UPDATE_BACKREF;
6667 wc->shared_level = level - 1;
6671 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6675 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6676 btrfs_tree_unlock(next);
6677 free_extent_buffer(next);
6683 if (reada && level == 1)
6684 reada_walk_down(trans, root, wc, path);
6685 next = read_tree_block(root, bytenr, blocksize, generation);
6688 btrfs_tree_lock(next);
6689 btrfs_set_lock_blocking(next);
6693 BUG_ON(level != btrfs_header_level(next));
6694 path->nodes[level] = next;
6695 path->slots[level] = 0;
6696 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6702 wc->refs[level - 1] = 0;
6703 wc->flags[level - 1] = 0;
6704 if (wc->stage == DROP_REFERENCE) {
6705 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6706 parent = path->nodes[level]->start;
6708 BUG_ON(root->root_key.objectid !=
6709 btrfs_header_owner(path->nodes[level]));
6713 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6714 root->root_key.objectid, level - 1, 0, 0);
6715 BUG_ON(ret); /* -ENOMEM */
6717 btrfs_tree_unlock(next);
6718 free_extent_buffer(next);
6724 * helper to process tree block while walking up the tree.
6726 * when wc->stage == DROP_REFERENCE, this function drops
6727 * reference count on the block.
6729 * when wc->stage == UPDATE_BACKREF, this function changes
6730 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6731 * to UPDATE_BACKREF previously while processing the block.
6733 * NOTE: return value 1 means we should stop walking up.
6735 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6736 struct btrfs_root *root,
6737 struct btrfs_path *path,
6738 struct walk_control *wc)
6741 int level = wc->level;
6742 struct extent_buffer *eb = path->nodes[level];
6745 if (wc->stage == UPDATE_BACKREF) {
6746 BUG_ON(wc->shared_level < level);
6747 if (level < wc->shared_level)
6750 ret = find_next_key(path, level + 1, &wc->update_progress);
6754 wc->stage = DROP_REFERENCE;
6755 wc->shared_level = -1;
6756 path->slots[level] = 0;
6759 * check reference count again if the block isn't locked.
6760 * we should start walking down the tree again if reference
6763 if (!path->locks[level]) {
6765 btrfs_tree_lock(eb);
6766 btrfs_set_lock_blocking(eb);
6767 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6769 ret = btrfs_lookup_extent_info(trans, root,
6774 btrfs_tree_unlock_rw(eb, path->locks[level]);
6775 path->locks[level] = 0;
6778 BUG_ON(wc->refs[level] == 0);
6779 if (wc->refs[level] == 1) {
6780 btrfs_tree_unlock_rw(eb, path->locks[level]);
6781 path->locks[level] = 0;
6787 /* wc->stage == DROP_REFERENCE */
6788 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6790 if (wc->refs[level] == 1) {
6792 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6793 ret = btrfs_dec_ref(trans, root, eb, 1,
6796 ret = btrfs_dec_ref(trans, root, eb, 0,
6798 BUG_ON(ret); /* -ENOMEM */
6800 /* make block locked assertion in clean_tree_block happy */
6801 if (!path->locks[level] &&
6802 btrfs_header_generation(eb) == trans->transid) {
6803 btrfs_tree_lock(eb);
6804 btrfs_set_lock_blocking(eb);
6805 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6807 clean_tree_block(trans, root, eb);
6810 if (eb == root->node) {
6811 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6814 BUG_ON(root->root_key.objectid !=
6815 btrfs_header_owner(eb));
6817 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6818 parent = path->nodes[level + 1]->start;
6820 BUG_ON(root->root_key.objectid !=
6821 btrfs_header_owner(path->nodes[level + 1]));
6824 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6826 wc->refs[level] = 0;
6827 wc->flags[level] = 0;
6831 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6832 struct btrfs_root *root,
6833 struct btrfs_path *path,
6834 struct walk_control *wc)
6836 int level = wc->level;
6837 int lookup_info = 1;
6840 while (level >= 0) {
6841 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6848 if (path->slots[level] >=
6849 btrfs_header_nritems(path->nodes[level]))
6852 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6854 path->slots[level]++;
6863 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6864 struct btrfs_root *root,
6865 struct btrfs_path *path,
6866 struct walk_control *wc, int max_level)
6868 int level = wc->level;
6871 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6872 while (level < max_level && path->nodes[level]) {
6874 if (path->slots[level] + 1 <
6875 btrfs_header_nritems(path->nodes[level])) {
6876 path->slots[level]++;
6879 ret = walk_up_proc(trans, root, path, wc);
6883 if (path->locks[level]) {
6884 btrfs_tree_unlock_rw(path->nodes[level],
6885 path->locks[level]);
6886 path->locks[level] = 0;
6888 free_extent_buffer(path->nodes[level]);
6889 path->nodes[level] = NULL;
6897 * drop a subvolume tree.
6899 * this function traverses the tree freeing any blocks that only
6900 * referenced by the tree.
6902 * when a shared tree block is found. this function decreases its
6903 * reference count by one. if update_ref is true, this function
6904 * also make sure backrefs for the shared block and all lower level
6905 * blocks are properly updated.
6907 int btrfs_drop_snapshot(struct btrfs_root *root,
6908 struct btrfs_block_rsv *block_rsv, int update_ref,
6911 struct btrfs_path *path;
6912 struct btrfs_trans_handle *trans;
6913 struct btrfs_root *tree_root = root->fs_info->tree_root;
6914 struct btrfs_root_item *root_item = &root->root_item;
6915 struct walk_control *wc;
6916 struct btrfs_key key;
6921 path = btrfs_alloc_path();
6927 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6929 btrfs_free_path(path);
6934 trans = btrfs_start_transaction(tree_root, 0);
6935 if (IS_ERR(trans)) {
6936 err = PTR_ERR(trans);
6941 trans->block_rsv = block_rsv;
6943 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6944 level = btrfs_header_level(root->node);
6945 path->nodes[level] = btrfs_lock_root_node(root);
6946 btrfs_set_lock_blocking(path->nodes[level]);
6947 path->slots[level] = 0;
6948 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6949 memset(&wc->update_progress, 0,
6950 sizeof(wc->update_progress));
6952 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6953 memcpy(&wc->update_progress, &key,
6954 sizeof(wc->update_progress));
6956 level = root_item->drop_level;
6958 path->lowest_level = level;
6959 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6960 path->lowest_level = 0;
6968 * unlock our path, this is safe because only this
6969 * function is allowed to delete this snapshot
6971 btrfs_unlock_up_safe(path, 0);
6973 level = btrfs_header_level(root->node);
6975 btrfs_tree_lock(path->nodes[level]);
6976 btrfs_set_lock_blocking(path->nodes[level]);
6978 ret = btrfs_lookup_extent_info(trans, root,
6979 path->nodes[level]->start,
6980 path->nodes[level]->len,
6987 BUG_ON(wc->refs[level] == 0);
6989 if (level == root_item->drop_level)
6992 btrfs_tree_unlock(path->nodes[level]);
6993 WARN_ON(wc->refs[level] != 1);
6999 wc->shared_level = -1;
7000 wc->stage = DROP_REFERENCE;
7001 wc->update_ref = update_ref;
7003 wc->for_reloc = for_reloc;
7004 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7007 ret = walk_down_tree(trans, root, path, wc);
7013 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7020 BUG_ON(wc->stage != DROP_REFERENCE);
7024 if (wc->stage == DROP_REFERENCE) {
7026 btrfs_node_key(path->nodes[level],
7027 &root_item->drop_progress,
7028 path->slots[level]);
7029 root_item->drop_level = level;
7032 BUG_ON(wc->level == 0);
7033 if (btrfs_should_end_transaction(trans, tree_root)) {
7034 ret = btrfs_update_root(trans, tree_root,
7038 btrfs_abort_transaction(trans, tree_root, ret);
7043 btrfs_end_transaction_throttle(trans, tree_root);
7044 trans = btrfs_start_transaction(tree_root, 0);
7045 if (IS_ERR(trans)) {
7046 err = PTR_ERR(trans);
7050 trans->block_rsv = block_rsv;
7053 btrfs_release_path(path);
7057 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7059 btrfs_abort_transaction(trans, tree_root, ret);
7063 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7064 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7067 btrfs_abort_transaction(trans, tree_root, ret);
7070 } else if (ret > 0) {
7071 /* if we fail to delete the orphan item this time
7072 * around, it'll get picked up the next time.
7074 * The most common failure here is just -ENOENT.
7076 btrfs_del_orphan_item(trans, tree_root,
7077 root->root_key.objectid);
7081 if (root->in_radix) {
7082 btrfs_free_fs_root(tree_root->fs_info, root);
7084 free_extent_buffer(root->node);
7085 free_extent_buffer(root->commit_root);
7089 btrfs_end_transaction_throttle(trans, tree_root);
7092 btrfs_free_path(path);
7095 btrfs_std_error(root->fs_info, err);
7100 * drop subtree rooted at tree block 'node'.
7102 * NOTE: this function will unlock and release tree block 'node'
7103 * only used by relocation code
7105 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7106 struct btrfs_root *root,
7107 struct extent_buffer *node,
7108 struct extent_buffer *parent)
7110 struct btrfs_path *path;
7111 struct walk_control *wc;
7117 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7119 path = btrfs_alloc_path();
7123 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7125 btrfs_free_path(path);
7129 btrfs_assert_tree_locked(parent);
7130 parent_level = btrfs_header_level(parent);
7131 extent_buffer_get(parent);
7132 path->nodes[parent_level] = parent;
7133 path->slots[parent_level] = btrfs_header_nritems(parent);
7135 btrfs_assert_tree_locked(node);
7136 level = btrfs_header_level(node);
7137 path->nodes[level] = node;
7138 path->slots[level] = 0;
7139 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7141 wc->refs[parent_level] = 1;
7142 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7144 wc->shared_level = -1;
7145 wc->stage = DROP_REFERENCE;
7149 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7152 wret = walk_down_tree(trans, root, path, wc);
7158 wret = walk_up_tree(trans, root, path, wc, parent_level);
7166 btrfs_free_path(path);
7170 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7176 * if restripe for this chunk_type is on pick target profile and
7177 * return, otherwise do the usual balance
7179 stripped = get_restripe_target(root->fs_info, flags);
7181 return extended_to_chunk(stripped);
7184 * we add in the count of missing devices because we want
7185 * to make sure that any RAID levels on a degraded FS
7186 * continue to be honored.
7188 num_devices = root->fs_info->fs_devices->rw_devices +
7189 root->fs_info->fs_devices->missing_devices;
7191 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7192 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7194 if (num_devices == 1) {
7195 stripped |= BTRFS_BLOCK_GROUP_DUP;
7196 stripped = flags & ~stripped;
7198 /* turn raid0 into single device chunks */
7199 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7202 /* turn mirroring into duplication */
7203 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7204 BTRFS_BLOCK_GROUP_RAID10))
7205 return stripped | BTRFS_BLOCK_GROUP_DUP;
7207 /* they already had raid on here, just return */
7208 if (flags & stripped)
7211 stripped |= BTRFS_BLOCK_GROUP_DUP;
7212 stripped = flags & ~stripped;
7214 /* switch duplicated blocks with raid1 */
7215 if (flags & BTRFS_BLOCK_GROUP_DUP)
7216 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7218 /* this is drive concat, leave it alone */
7224 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7226 struct btrfs_space_info *sinfo = cache->space_info;
7228 u64 min_allocable_bytes;
7233 * We need some metadata space and system metadata space for
7234 * allocating chunks in some corner cases until we force to set
7235 * it to be readonly.
7238 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7240 min_allocable_bytes = 1 * 1024 * 1024;
7242 min_allocable_bytes = 0;
7244 spin_lock(&sinfo->lock);
7245 spin_lock(&cache->lock);
7252 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7253 cache->bytes_super - btrfs_block_group_used(&cache->item);
7255 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7256 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7257 min_allocable_bytes <= sinfo->total_bytes) {
7258 sinfo->bytes_readonly += num_bytes;
7263 spin_unlock(&cache->lock);
7264 spin_unlock(&sinfo->lock);
7268 int btrfs_set_block_group_ro(struct btrfs_root *root,
7269 struct btrfs_block_group_cache *cache)
7272 struct btrfs_trans_handle *trans;
7278 trans = btrfs_join_transaction(root);
7280 return PTR_ERR(trans);
7282 alloc_flags = update_block_group_flags(root, cache->flags);
7283 if (alloc_flags != cache->flags) {
7284 ret = do_chunk_alloc(trans, root, alloc_flags,
7290 ret = set_block_group_ro(cache, 0);
7293 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7294 ret = do_chunk_alloc(trans, root, alloc_flags,
7298 ret = set_block_group_ro(cache, 0);
7300 btrfs_end_transaction(trans, root);
7304 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7305 struct btrfs_root *root, u64 type)
7307 u64 alloc_flags = get_alloc_profile(root, type);
7308 return do_chunk_alloc(trans, root, alloc_flags,
7313 * helper to account the unused space of all the readonly block group in the
7314 * list. takes mirrors into account.
7316 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7318 struct btrfs_block_group_cache *block_group;
7322 list_for_each_entry(block_group, groups_list, list) {
7323 spin_lock(&block_group->lock);
7325 if (!block_group->ro) {
7326 spin_unlock(&block_group->lock);
7330 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7331 BTRFS_BLOCK_GROUP_RAID10 |
7332 BTRFS_BLOCK_GROUP_DUP))
7337 free_bytes += (block_group->key.offset -
7338 btrfs_block_group_used(&block_group->item)) *
7341 spin_unlock(&block_group->lock);
7348 * helper to account the unused space of all the readonly block group in the
7349 * space_info. takes mirrors into account.
7351 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7356 spin_lock(&sinfo->lock);
7358 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7359 if (!list_empty(&sinfo->block_groups[i]))
7360 free_bytes += __btrfs_get_ro_block_group_free_space(
7361 &sinfo->block_groups[i]);
7363 spin_unlock(&sinfo->lock);
7368 void btrfs_set_block_group_rw(struct btrfs_root *root,
7369 struct btrfs_block_group_cache *cache)
7371 struct btrfs_space_info *sinfo = cache->space_info;
7376 spin_lock(&sinfo->lock);
7377 spin_lock(&cache->lock);
7378 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7379 cache->bytes_super - btrfs_block_group_used(&cache->item);
7380 sinfo->bytes_readonly -= num_bytes;
7382 spin_unlock(&cache->lock);
7383 spin_unlock(&sinfo->lock);
7387 * checks to see if its even possible to relocate this block group.
7389 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7390 * ok to go ahead and try.
7392 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7394 struct btrfs_block_group_cache *block_group;
7395 struct btrfs_space_info *space_info;
7396 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7397 struct btrfs_device *device;
7406 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7408 /* odd, couldn't find the block group, leave it alone */
7412 min_free = btrfs_block_group_used(&block_group->item);
7414 /* no bytes used, we're good */
7418 space_info = block_group->space_info;
7419 spin_lock(&space_info->lock);
7421 full = space_info->full;
7424 * if this is the last block group we have in this space, we can't
7425 * relocate it unless we're able to allocate a new chunk below.
7427 * Otherwise, we need to make sure we have room in the space to handle
7428 * all of the extents from this block group. If we can, we're good
7430 if ((space_info->total_bytes != block_group->key.offset) &&
7431 (space_info->bytes_used + space_info->bytes_reserved +
7432 space_info->bytes_pinned + space_info->bytes_readonly +
7433 min_free < space_info->total_bytes)) {
7434 spin_unlock(&space_info->lock);
7437 spin_unlock(&space_info->lock);
7440 * ok we don't have enough space, but maybe we have free space on our
7441 * devices to allocate new chunks for relocation, so loop through our
7442 * alloc devices and guess if we have enough space. if this block
7443 * group is going to be restriped, run checks against the target
7444 * profile instead of the current one.
7456 target = get_restripe_target(root->fs_info, block_group->flags);
7458 index = __get_raid_index(extended_to_chunk(target));
7461 * this is just a balance, so if we were marked as full
7462 * we know there is no space for a new chunk
7467 index = get_block_group_index(block_group);
7474 } else if (index == 1) {
7476 } else if (index == 2) {
7479 } else if (index == 3) {
7480 dev_min = fs_devices->rw_devices;
7481 do_div(min_free, dev_min);
7484 mutex_lock(&root->fs_info->chunk_mutex);
7485 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7489 * check to make sure we can actually find a chunk with enough
7490 * space to fit our block group in.
7492 if (device->total_bytes > device->bytes_used + min_free &&
7493 !device->is_tgtdev_for_dev_replace) {
7494 ret = find_free_dev_extent(device, min_free,
7499 if (dev_nr >= dev_min)
7505 mutex_unlock(&root->fs_info->chunk_mutex);
7507 btrfs_put_block_group(block_group);
7511 static int find_first_block_group(struct btrfs_root *root,
7512 struct btrfs_path *path, struct btrfs_key *key)
7515 struct btrfs_key found_key;
7516 struct extent_buffer *leaf;
7519 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7524 slot = path->slots[0];
7525 leaf = path->nodes[0];
7526 if (slot >= btrfs_header_nritems(leaf)) {
7527 ret = btrfs_next_leaf(root, path);
7534 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7536 if (found_key.objectid >= key->objectid &&
7537 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7547 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7549 struct btrfs_block_group_cache *block_group;
7553 struct inode *inode;
7555 block_group = btrfs_lookup_first_block_group(info, last);
7556 while (block_group) {
7557 spin_lock(&block_group->lock);
7558 if (block_group->iref)
7560 spin_unlock(&block_group->lock);
7561 block_group = next_block_group(info->tree_root,
7571 inode = block_group->inode;
7572 block_group->iref = 0;
7573 block_group->inode = NULL;
7574 spin_unlock(&block_group->lock);
7576 last = block_group->key.objectid + block_group->key.offset;
7577 btrfs_put_block_group(block_group);
7581 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7583 struct btrfs_block_group_cache *block_group;
7584 struct btrfs_space_info *space_info;
7585 struct btrfs_caching_control *caching_ctl;
7588 down_write(&info->extent_commit_sem);
7589 while (!list_empty(&info->caching_block_groups)) {
7590 caching_ctl = list_entry(info->caching_block_groups.next,
7591 struct btrfs_caching_control, list);
7592 list_del(&caching_ctl->list);
7593 put_caching_control(caching_ctl);
7595 up_write(&info->extent_commit_sem);
7597 spin_lock(&info->block_group_cache_lock);
7598 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7599 block_group = rb_entry(n, struct btrfs_block_group_cache,
7601 rb_erase(&block_group->cache_node,
7602 &info->block_group_cache_tree);
7603 spin_unlock(&info->block_group_cache_lock);
7605 down_write(&block_group->space_info->groups_sem);
7606 list_del(&block_group->list);
7607 up_write(&block_group->space_info->groups_sem);
7609 if (block_group->cached == BTRFS_CACHE_STARTED)
7610 wait_block_group_cache_done(block_group);
7613 * We haven't cached this block group, which means we could
7614 * possibly have excluded extents on this block group.
7616 if (block_group->cached == BTRFS_CACHE_NO)
7617 free_excluded_extents(info->extent_root, block_group);
7619 btrfs_remove_free_space_cache(block_group);
7620 btrfs_put_block_group(block_group);
7622 spin_lock(&info->block_group_cache_lock);
7624 spin_unlock(&info->block_group_cache_lock);
7626 /* now that all the block groups are freed, go through and
7627 * free all the space_info structs. This is only called during
7628 * the final stages of unmount, and so we know nobody is
7629 * using them. We call synchronize_rcu() once before we start,
7630 * just to be on the safe side.
7634 release_global_block_rsv(info);
7636 while(!list_empty(&info->space_info)) {
7637 space_info = list_entry(info->space_info.next,
7638 struct btrfs_space_info,
7640 if (space_info->bytes_pinned > 0 ||
7641 space_info->bytes_reserved > 0 ||
7642 space_info->bytes_may_use > 0) {
7644 dump_space_info(space_info, 0, 0);
7646 list_del(&space_info->list);
7652 static void __link_block_group(struct btrfs_space_info *space_info,
7653 struct btrfs_block_group_cache *cache)
7655 int index = get_block_group_index(cache);
7657 down_write(&space_info->groups_sem);
7658 list_add_tail(&cache->list, &space_info->block_groups[index]);
7659 up_write(&space_info->groups_sem);
7662 int btrfs_read_block_groups(struct btrfs_root *root)
7664 struct btrfs_path *path;
7666 struct btrfs_block_group_cache *cache;
7667 struct btrfs_fs_info *info = root->fs_info;
7668 struct btrfs_space_info *space_info;
7669 struct btrfs_key key;
7670 struct btrfs_key found_key;
7671 struct extent_buffer *leaf;
7675 root = info->extent_root;
7678 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7679 path = btrfs_alloc_path();
7684 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7685 if (btrfs_test_opt(root, SPACE_CACHE) &&
7686 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7688 if (btrfs_test_opt(root, CLEAR_CACHE))
7692 ret = find_first_block_group(root, path, &key);
7697 leaf = path->nodes[0];
7698 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7699 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7704 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7706 if (!cache->free_space_ctl) {
7712 atomic_set(&cache->count, 1);
7713 spin_lock_init(&cache->lock);
7714 cache->fs_info = info;
7715 INIT_LIST_HEAD(&cache->list);
7716 INIT_LIST_HEAD(&cache->cluster_list);
7720 * When we mount with old space cache, we need to
7721 * set BTRFS_DC_CLEAR and set dirty flag.
7723 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7724 * truncate the old free space cache inode and
7726 * b) Setting 'dirty flag' makes sure that we flush
7727 * the new space cache info onto disk.
7729 cache->disk_cache_state = BTRFS_DC_CLEAR;
7730 if (btrfs_test_opt(root, SPACE_CACHE))
7734 read_extent_buffer(leaf, &cache->item,
7735 btrfs_item_ptr_offset(leaf, path->slots[0]),
7736 sizeof(cache->item));
7737 memcpy(&cache->key, &found_key, sizeof(found_key));
7739 key.objectid = found_key.objectid + found_key.offset;
7740 btrfs_release_path(path);
7741 cache->flags = btrfs_block_group_flags(&cache->item);
7742 cache->sectorsize = root->sectorsize;
7744 btrfs_init_free_space_ctl(cache);
7747 * We need to exclude the super stripes now so that the space
7748 * info has super bytes accounted for, otherwise we'll think
7749 * we have more space than we actually do.
7751 exclude_super_stripes(root, cache);
7754 * check for two cases, either we are full, and therefore
7755 * don't need to bother with the caching work since we won't
7756 * find any space, or we are empty, and we can just add all
7757 * the space in and be done with it. This saves us _alot_ of
7758 * time, particularly in the full case.
7760 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7761 cache->last_byte_to_unpin = (u64)-1;
7762 cache->cached = BTRFS_CACHE_FINISHED;
7763 free_excluded_extents(root, cache);
7764 } else if (btrfs_block_group_used(&cache->item) == 0) {
7765 cache->last_byte_to_unpin = (u64)-1;
7766 cache->cached = BTRFS_CACHE_FINISHED;
7767 add_new_free_space(cache, root->fs_info,
7769 found_key.objectid +
7771 free_excluded_extents(root, cache);
7774 ret = update_space_info(info, cache->flags, found_key.offset,
7775 btrfs_block_group_used(&cache->item),
7777 BUG_ON(ret); /* -ENOMEM */
7778 cache->space_info = space_info;
7779 spin_lock(&cache->space_info->lock);
7780 cache->space_info->bytes_readonly += cache->bytes_super;
7781 spin_unlock(&cache->space_info->lock);
7783 __link_block_group(space_info, cache);
7785 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7786 BUG_ON(ret); /* Logic error */
7788 set_avail_alloc_bits(root->fs_info, cache->flags);
7789 if (btrfs_chunk_readonly(root, cache->key.objectid))
7790 set_block_group_ro(cache, 1);
7793 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7794 if (!(get_alloc_profile(root, space_info->flags) &
7795 (BTRFS_BLOCK_GROUP_RAID10 |
7796 BTRFS_BLOCK_GROUP_RAID1 |
7797 BTRFS_BLOCK_GROUP_DUP)))
7800 * avoid allocating from un-mirrored block group if there are
7801 * mirrored block groups.
7803 list_for_each_entry(cache, &space_info->block_groups[3], list)
7804 set_block_group_ro(cache, 1);
7805 list_for_each_entry(cache, &space_info->block_groups[4], list)
7806 set_block_group_ro(cache, 1);
7809 init_global_block_rsv(info);
7812 btrfs_free_path(path);
7816 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7817 struct btrfs_root *root)
7819 struct btrfs_block_group_cache *block_group, *tmp;
7820 struct btrfs_root *extent_root = root->fs_info->extent_root;
7821 struct btrfs_block_group_item item;
7822 struct btrfs_key key;
7825 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7827 list_del_init(&block_group->new_bg_list);
7832 spin_lock(&block_group->lock);
7833 memcpy(&item, &block_group->item, sizeof(item));
7834 memcpy(&key, &block_group->key, sizeof(key));
7835 spin_unlock(&block_group->lock);
7837 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7840 btrfs_abort_transaction(trans, extent_root, ret);
7844 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7845 struct btrfs_root *root, u64 bytes_used,
7846 u64 type, u64 chunk_objectid, u64 chunk_offset,
7850 struct btrfs_root *extent_root;
7851 struct btrfs_block_group_cache *cache;
7853 extent_root = root->fs_info->extent_root;
7855 root->fs_info->last_trans_log_full_commit = trans->transid;
7857 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7860 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7862 if (!cache->free_space_ctl) {
7867 cache->key.objectid = chunk_offset;
7868 cache->key.offset = size;
7869 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7870 cache->sectorsize = root->sectorsize;
7871 cache->fs_info = root->fs_info;
7873 atomic_set(&cache->count, 1);
7874 spin_lock_init(&cache->lock);
7875 INIT_LIST_HEAD(&cache->list);
7876 INIT_LIST_HEAD(&cache->cluster_list);
7877 INIT_LIST_HEAD(&cache->new_bg_list);
7879 btrfs_init_free_space_ctl(cache);
7881 btrfs_set_block_group_used(&cache->item, bytes_used);
7882 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7883 cache->flags = type;
7884 btrfs_set_block_group_flags(&cache->item, type);
7886 cache->last_byte_to_unpin = (u64)-1;
7887 cache->cached = BTRFS_CACHE_FINISHED;
7888 exclude_super_stripes(root, cache);
7890 add_new_free_space(cache, root->fs_info, chunk_offset,
7891 chunk_offset + size);
7893 free_excluded_extents(root, cache);
7895 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7896 &cache->space_info);
7897 BUG_ON(ret); /* -ENOMEM */
7898 update_global_block_rsv(root->fs_info);
7900 spin_lock(&cache->space_info->lock);
7901 cache->space_info->bytes_readonly += cache->bytes_super;
7902 spin_unlock(&cache->space_info->lock);
7904 __link_block_group(cache->space_info, cache);
7906 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7907 BUG_ON(ret); /* Logic error */
7909 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7911 set_avail_alloc_bits(extent_root->fs_info, type);
7916 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7918 u64 extra_flags = chunk_to_extended(flags) &
7919 BTRFS_EXTENDED_PROFILE_MASK;
7921 if (flags & BTRFS_BLOCK_GROUP_DATA)
7922 fs_info->avail_data_alloc_bits &= ~extra_flags;
7923 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7924 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7925 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7926 fs_info->avail_system_alloc_bits &= ~extra_flags;
7929 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7930 struct btrfs_root *root, u64 group_start)
7932 struct btrfs_path *path;
7933 struct btrfs_block_group_cache *block_group;
7934 struct btrfs_free_cluster *cluster;
7935 struct btrfs_root *tree_root = root->fs_info->tree_root;
7936 struct btrfs_key key;
7937 struct inode *inode;
7942 root = root->fs_info->extent_root;
7944 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7945 BUG_ON(!block_group);
7946 BUG_ON(!block_group->ro);
7949 * Free the reserved super bytes from this block group before
7952 free_excluded_extents(root, block_group);
7954 memcpy(&key, &block_group->key, sizeof(key));
7955 index = get_block_group_index(block_group);
7956 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7957 BTRFS_BLOCK_GROUP_RAID1 |
7958 BTRFS_BLOCK_GROUP_RAID10))
7963 /* make sure this block group isn't part of an allocation cluster */
7964 cluster = &root->fs_info->data_alloc_cluster;
7965 spin_lock(&cluster->refill_lock);
7966 btrfs_return_cluster_to_free_space(block_group, cluster);
7967 spin_unlock(&cluster->refill_lock);
7970 * make sure this block group isn't part of a metadata
7971 * allocation cluster
7973 cluster = &root->fs_info->meta_alloc_cluster;
7974 spin_lock(&cluster->refill_lock);
7975 btrfs_return_cluster_to_free_space(block_group, cluster);
7976 spin_unlock(&cluster->refill_lock);
7978 path = btrfs_alloc_path();
7984 inode = lookup_free_space_inode(tree_root, block_group, path);
7985 if (!IS_ERR(inode)) {
7986 ret = btrfs_orphan_add(trans, inode);
7988 btrfs_add_delayed_iput(inode);
7992 /* One for the block groups ref */
7993 spin_lock(&block_group->lock);
7994 if (block_group->iref) {
7995 block_group->iref = 0;
7996 block_group->inode = NULL;
7997 spin_unlock(&block_group->lock);
8000 spin_unlock(&block_group->lock);
8002 /* One for our lookup ref */
8003 btrfs_add_delayed_iput(inode);
8006 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8007 key.offset = block_group->key.objectid;
8010 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8014 btrfs_release_path(path);
8016 ret = btrfs_del_item(trans, tree_root, path);
8019 btrfs_release_path(path);
8022 spin_lock(&root->fs_info->block_group_cache_lock);
8023 rb_erase(&block_group->cache_node,
8024 &root->fs_info->block_group_cache_tree);
8025 spin_unlock(&root->fs_info->block_group_cache_lock);
8027 down_write(&block_group->space_info->groups_sem);
8029 * we must use list_del_init so people can check to see if they
8030 * are still on the list after taking the semaphore
8032 list_del_init(&block_group->list);
8033 if (list_empty(&block_group->space_info->block_groups[index]))
8034 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8035 up_write(&block_group->space_info->groups_sem);
8037 if (block_group->cached == BTRFS_CACHE_STARTED)
8038 wait_block_group_cache_done(block_group);
8040 btrfs_remove_free_space_cache(block_group);
8042 spin_lock(&block_group->space_info->lock);
8043 block_group->space_info->total_bytes -= block_group->key.offset;
8044 block_group->space_info->bytes_readonly -= block_group->key.offset;
8045 block_group->space_info->disk_total -= block_group->key.offset * factor;
8046 spin_unlock(&block_group->space_info->lock);
8048 memcpy(&key, &block_group->key, sizeof(key));
8050 btrfs_clear_space_info_full(root->fs_info);
8052 btrfs_put_block_group(block_group);
8053 btrfs_put_block_group(block_group);
8055 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8061 ret = btrfs_del_item(trans, root, path);
8063 btrfs_free_path(path);
8067 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8069 struct btrfs_space_info *space_info;
8070 struct btrfs_super_block *disk_super;
8076 disk_super = fs_info->super_copy;
8077 if (!btrfs_super_root(disk_super))
8080 features = btrfs_super_incompat_flags(disk_super);
8081 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8084 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8085 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8090 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8091 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8093 flags = BTRFS_BLOCK_GROUP_METADATA;
8094 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8098 flags = BTRFS_BLOCK_GROUP_DATA;
8099 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8105 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8107 return unpin_extent_range(root, start, end);
8110 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8111 u64 num_bytes, u64 *actual_bytes)
8113 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8116 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8118 struct btrfs_fs_info *fs_info = root->fs_info;
8119 struct btrfs_block_group_cache *cache = NULL;
8124 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8128 * try to trim all FS space, our block group may start from non-zero.
8130 if (range->len == total_bytes)
8131 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8133 cache = btrfs_lookup_block_group(fs_info, range->start);
8136 if (cache->key.objectid >= (range->start + range->len)) {
8137 btrfs_put_block_group(cache);
8141 start = max(range->start, cache->key.objectid);
8142 end = min(range->start + range->len,
8143 cache->key.objectid + cache->key.offset);
8145 if (end - start >= range->minlen) {
8146 if (!block_group_cache_done(cache)) {
8147 ret = cache_block_group(cache, NULL, root, 0);
8149 wait_block_group_cache_done(cache);
8151 ret = btrfs_trim_block_group(cache,
8157 trimmed += group_trimmed;
8159 btrfs_put_block_group(cache);
8164 cache = next_block_group(fs_info->tree_root, cache);
8167 range->len = trimmed;