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,
2149 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2150 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2151 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2153 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2154 node->type == BTRFS_SHARED_DATA_REF_KEY)
2155 ret = run_delayed_data_ref(trans, root, node, extent_op,
2162 static noinline struct btrfs_delayed_ref_node *
2163 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2165 struct rb_node *node;
2166 struct btrfs_delayed_ref_node *ref;
2167 int action = BTRFS_ADD_DELAYED_REF;
2170 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2171 * this prevents ref count from going down to zero when
2172 * there still are pending delayed ref.
2174 node = rb_prev(&head->node.rb_node);
2178 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2180 if (ref->bytenr != head->node.bytenr)
2182 if (ref->action == action)
2184 node = rb_prev(node);
2186 if (action == BTRFS_ADD_DELAYED_REF) {
2187 action = BTRFS_DROP_DELAYED_REF;
2194 * Returns 0 on success or if called with an already aborted transaction.
2195 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2197 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2198 struct btrfs_root *root,
2199 struct list_head *cluster)
2201 struct btrfs_delayed_ref_root *delayed_refs;
2202 struct btrfs_delayed_ref_node *ref;
2203 struct btrfs_delayed_ref_head *locked_ref = NULL;
2204 struct btrfs_delayed_extent_op *extent_op;
2205 struct btrfs_fs_info *fs_info = root->fs_info;
2208 int must_insert_reserved = 0;
2210 delayed_refs = &trans->transaction->delayed_refs;
2213 /* pick a new head ref from the cluster list */
2214 if (list_empty(cluster))
2217 locked_ref = list_entry(cluster->next,
2218 struct btrfs_delayed_ref_head, cluster);
2220 /* grab the lock that says we are going to process
2221 * all the refs for this head */
2222 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2225 * we may have dropped the spin lock to get the head
2226 * mutex lock, and that might have given someone else
2227 * time to free the head. If that's true, it has been
2228 * removed from our list and we can move on.
2230 if (ret == -EAGAIN) {
2238 * We need to try and merge add/drops of the same ref since we
2239 * can run into issues with relocate dropping the implicit ref
2240 * and then it being added back again before the drop can
2241 * finish. If we merged anything we need to re-loop so we can
2244 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2248 * locked_ref is the head node, so we have to go one
2249 * node back for any delayed ref updates
2251 ref = select_delayed_ref(locked_ref);
2253 if (ref && ref->seq &&
2254 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2256 * there are still refs with lower seq numbers in the
2257 * process of being added. Don't run this ref yet.
2259 list_del_init(&locked_ref->cluster);
2260 btrfs_delayed_ref_unlock(locked_ref);
2262 delayed_refs->num_heads_ready++;
2263 spin_unlock(&delayed_refs->lock);
2265 spin_lock(&delayed_refs->lock);
2270 * record the must insert reserved flag before we
2271 * drop the spin lock.
2273 must_insert_reserved = locked_ref->must_insert_reserved;
2274 locked_ref->must_insert_reserved = 0;
2276 extent_op = locked_ref->extent_op;
2277 locked_ref->extent_op = NULL;
2280 /* All delayed refs have been processed, Go ahead
2281 * and send the head node to run_one_delayed_ref,
2282 * so that any accounting fixes can happen
2284 ref = &locked_ref->node;
2286 if (extent_op && must_insert_reserved) {
2287 btrfs_free_delayed_extent_op(extent_op);
2292 spin_unlock(&delayed_refs->lock);
2294 ret = run_delayed_extent_op(trans, root,
2296 btrfs_free_delayed_extent_op(extent_op);
2300 "btrfs: run_delayed_extent_op "
2301 "returned %d\n", ret);
2302 spin_lock(&delayed_refs->lock);
2303 btrfs_delayed_ref_unlock(locked_ref);
2312 rb_erase(&ref->rb_node, &delayed_refs->root);
2313 delayed_refs->num_entries--;
2314 if (!btrfs_delayed_ref_is_head(ref)) {
2316 * when we play the delayed ref, also correct the
2319 switch (ref->action) {
2320 case BTRFS_ADD_DELAYED_REF:
2321 case BTRFS_ADD_DELAYED_EXTENT:
2322 locked_ref->node.ref_mod -= ref->ref_mod;
2324 case BTRFS_DROP_DELAYED_REF:
2325 locked_ref->node.ref_mod += ref->ref_mod;
2331 spin_unlock(&delayed_refs->lock);
2333 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2334 must_insert_reserved);
2336 btrfs_free_delayed_extent_op(extent_op);
2338 btrfs_delayed_ref_unlock(locked_ref);
2339 btrfs_put_delayed_ref(ref);
2341 "btrfs: run_one_delayed_ref returned %d\n", ret);
2342 spin_lock(&delayed_refs->lock);
2347 * If this node is a head, that means all the refs in this head
2348 * have been dealt with, and we will pick the next head to deal
2349 * with, so we must unlock the head and drop it from the cluster
2350 * list before we release it.
2352 if (btrfs_delayed_ref_is_head(ref)) {
2353 list_del_init(&locked_ref->cluster);
2354 btrfs_delayed_ref_unlock(locked_ref);
2357 btrfs_put_delayed_ref(ref);
2361 spin_lock(&delayed_refs->lock);
2366 #ifdef SCRAMBLE_DELAYED_REFS
2368 * Normally delayed refs get processed in ascending bytenr order. This
2369 * correlates in most cases to the order added. To expose dependencies on this
2370 * order, we start to process the tree in the middle instead of the beginning
2372 static u64 find_middle(struct rb_root *root)
2374 struct rb_node *n = root->rb_node;
2375 struct btrfs_delayed_ref_node *entry;
2378 u64 first = 0, last = 0;
2382 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2383 first = entry->bytenr;
2387 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2388 last = entry->bytenr;
2393 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2394 WARN_ON(!entry->in_tree);
2396 middle = entry->bytenr;
2409 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2410 struct btrfs_fs_info *fs_info)
2412 struct qgroup_update *qgroup_update;
2415 if (list_empty(&trans->qgroup_ref_list) !=
2416 !trans->delayed_ref_elem.seq) {
2417 /* list without seq or seq without list */
2418 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2419 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2420 trans->delayed_ref_elem.seq);
2424 if (!trans->delayed_ref_elem.seq)
2427 while (!list_empty(&trans->qgroup_ref_list)) {
2428 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2429 struct qgroup_update, list);
2430 list_del(&qgroup_update->list);
2432 ret = btrfs_qgroup_account_ref(
2433 trans, fs_info, qgroup_update->node,
2434 qgroup_update->extent_op);
2435 kfree(qgroup_update);
2438 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2444 * this starts processing the delayed reference count updates and
2445 * extent insertions we have queued up so far. count can be
2446 * 0, which means to process everything in the tree at the start
2447 * of the run (but not newly added entries), or it can be some target
2448 * number you'd like to process.
2450 * Returns 0 on success or if called with an aborted transaction
2451 * Returns <0 on error and aborts the transaction
2453 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *root, unsigned long count)
2456 struct rb_node *node;
2457 struct btrfs_delayed_ref_root *delayed_refs;
2458 struct btrfs_delayed_ref_node *ref;
2459 struct list_head cluster;
2462 int run_all = count == (unsigned long)-1;
2466 /* We'll clean this up in btrfs_cleanup_transaction */
2470 if (root == root->fs_info->extent_root)
2471 root = root->fs_info->tree_root;
2473 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2475 delayed_refs = &trans->transaction->delayed_refs;
2476 INIT_LIST_HEAD(&cluster);
2479 spin_lock(&delayed_refs->lock);
2481 #ifdef SCRAMBLE_DELAYED_REFS
2482 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2486 count = delayed_refs->num_entries * 2;
2490 if (!(run_all || run_most) &&
2491 delayed_refs->num_heads_ready < 64)
2495 * go find something we can process in the rbtree. We start at
2496 * the beginning of the tree, and then build a cluster
2497 * of refs to process starting at the first one we are able to
2500 delayed_start = delayed_refs->run_delayed_start;
2501 ret = btrfs_find_ref_cluster(trans, &cluster,
2502 delayed_refs->run_delayed_start);
2506 ret = run_clustered_refs(trans, root, &cluster);
2508 btrfs_release_ref_cluster(&cluster);
2509 spin_unlock(&delayed_refs->lock);
2510 btrfs_abort_transaction(trans, root, ret);
2514 count -= min_t(unsigned long, ret, count);
2519 if (delayed_start >= delayed_refs->run_delayed_start) {
2522 * btrfs_find_ref_cluster looped. let's do one
2523 * more cycle. if we don't run any delayed ref
2524 * during that cycle (because we can't because
2525 * all of them are blocked), bail out.
2530 * no runnable refs left, stop trying
2537 /* refs were run, let's reset staleness detection */
2543 if (!list_empty(&trans->new_bgs)) {
2544 spin_unlock(&delayed_refs->lock);
2545 btrfs_create_pending_block_groups(trans, root);
2546 spin_lock(&delayed_refs->lock);
2549 node = rb_first(&delayed_refs->root);
2552 count = (unsigned long)-1;
2555 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2557 if (btrfs_delayed_ref_is_head(ref)) {
2558 struct btrfs_delayed_ref_head *head;
2560 head = btrfs_delayed_node_to_head(ref);
2561 atomic_inc(&ref->refs);
2563 spin_unlock(&delayed_refs->lock);
2565 * Mutex was contended, block until it's
2566 * released and try again
2568 mutex_lock(&head->mutex);
2569 mutex_unlock(&head->mutex);
2571 btrfs_put_delayed_ref(ref);
2575 node = rb_next(node);
2577 spin_unlock(&delayed_refs->lock);
2578 schedule_timeout(1);
2582 spin_unlock(&delayed_refs->lock);
2583 assert_qgroups_uptodate(trans);
2587 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2588 struct btrfs_root *root,
2589 u64 bytenr, u64 num_bytes, u64 flags,
2592 struct btrfs_delayed_extent_op *extent_op;
2595 extent_op = btrfs_alloc_delayed_extent_op();
2599 extent_op->flags_to_set = flags;
2600 extent_op->update_flags = 1;
2601 extent_op->update_key = 0;
2602 extent_op->is_data = is_data ? 1 : 0;
2604 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2605 num_bytes, extent_op);
2607 btrfs_free_delayed_extent_op(extent_op);
2611 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2612 struct btrfs_root *root,
2613 struct btrfs_path *path,
2614 u64 objectid, u64 offset, u64 bytenr)
2616 struct btrfs_delayed_ref_head *head;
2617 struct btrfs_delayed_ref_node *ref;
2618 struct btrfs_delayed_data_ref *data_ref;
2619 struct btrfs_delayed_ref_root *delayed_refs;
2620 struct rb_node *node;
2624 delayed_refs = &trans->transaction->delayed_refs;
2625 spin_lock(&delayed_refs->lock);
2626 head = btrfs_find_delayed_ref_head(trans, bytenr);
2630 if (!mutex_trylock(&head->mutex)) {
2631 atomic_inc(&head->node.refs);
2632 spin_unlock(&delayed_refs->lock);
2634 btrfs_release_path(path);
2637 * Mutex was contended, block until it's released and let
2640 mutex_lock(&head->mutex);
2641 mutex_unlock(&head->mutex);
2642 btrfs_put_delayed_ref(&head->node);
2646 node = rb_prev(&head->node.rb_node);
2650 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2652 if (ref->bytenr != bytenr)
2656 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2659 data_ref = btrfs_delayed_node_to_data_ref(ref);
2661 node = rb_prev(node);
2665 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2666 if (ref->bytenr == bytenr && ref->seq == seq)
2670 if (data_ref->root != root->root_key.objectid ||
2671 data_ref->objectid != objectid || data_ref->offset != offset)
2676 mutex_unlock(&head->mutex);
2678 spin_unlock(&delayed_refs->lock);
2682 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root,
2684 struct btrfs_path *path,
2685 u64 objectid, u64 offset, u64 bytenr)
2687 struct btrfs_root *extent_root = root->fs_info->extent_root;
2688 struct extent_buffer *leaf;
2689 struct btrfs_extent_data_ref *ref;
2690 struct btrfs_extent_inline_ref *iref;
2691 struct btrfs_extent_item *ei;
2692 struct btrfs_key key;
2696 key.objectid = bytenr;
2697 key.offset = (u64)-1;
2698 key.type = BTRFS_EXTENT_ITEM_KEY;
2700 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2703 BUG_ON(ret == 0); /* Corruption */
2706 if (path->slots[0] == 0)
2710 leaf = path->nodes[0];
2711 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2713 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2717 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2719 if (item_size < sizeof(*ei)) {
2720 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2724 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2726 if (item_size != sizeof(*ei) +
2727 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2730 if (btrfs_extent_generation(leaf, ei) <=
2731 btrfs_root_last_snapshot(&root->root_item))
2734 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2735 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2736 BTRFS_EXTENT_DATA_REF_KEY)
2739 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2740 if (btrfs_extent_refs(leaf, ei) !=
2741 btrfs_extent_data_ref_count(leaf, ref) ||
2742 btrfs_extent_data_ref_root(leaf, ref) !=
2743 root->root_key.objectid ||
2744 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2745 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2753 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2754 struct btrfs_root *root,
2755 u64 objectid, u64 offset, u64 bytenr)
2757 struct btrfs_path *path;
2761 path = btrfs_alloc_path();
2766 ret = check_committed_ref(trans, root, path, objectid,
2768 if (ret && ret != -ENOENT)
2771 ret2 = check_delayed_ref(trans, root, path, objectid,
2773 } while (ret2 == -EAGAIN);
2775 if (ret2 && ret2 != -ENOENT) {
2780 if (ret != -ENOENT || ret2 != -ENOENT)
2783 btrfs_free_path(path);
2784 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2789 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2790 struct btrfs_root *root,
2791 struct extent_buffer *buf,
2792 int full_backref, int inc, int for_cow)
2799 struct btrfs_key key;
2800 struct btrfs_file_extent_item *fi;
2804 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2805 u64, u64, u64, u64, u64, u64, int);
2807 ref_root = btrfs_header_owner(buf);
2808 nritems = btrfs_header_nritems(buf);
2809 level = btrfs_header_level(buf);
2811 if (!root->ref_cows && level == 0)
2815 process_func = btrfs_inc_extent_ref;
2817 process_func = btrfs_free_extent;
2820 parent = buf->start;
2824 for (i = 0; i < nritems; i++) {
2826 btrfs_item_key_to_cpu(buf, &key, i);
2827 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2829 fi = btrfs_item_ptr(buf, i,
2830 struct btrfs_file_extent_item);
2831 if (btrfs_file_extent_type(buf, fi) ==
2832 BTRFS_FILE_EXTENT_INLINE)
2834 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2838 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2839 key.offset -= btrfs_file_extent_offset(buf, fi);
2840 ret = process_func(trans, root, bytenr, num_bytes,
2841 parent, ref_root, key.objectid,
2842 key.offset, for_cow);
2846 bytenr = btrfs_node_blockptr(buf, i);
2847 num_bytes = btrfs_level_size(root, level - 1);
2848 ret = process_func(trans, root, bytenr, num_bytes,
2849 parent, ref_root, level - 1, 0,
2860 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2861 struct extent_buffer *buf, int full_backref, int for_cow)
2863 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2866 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2867 struct extent_buffer *buf, int full_backref, int for_cow)
2869 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2872 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2873 struct btrfs_root *root,
2874 struct btrfs_path *path,
2875 struct btrfs_block_group_cache *cache)
2878 struct btrfs_root *extent_root = root->fs_info->extent_root;
2880 struct extent_buffer *leaf;
2882 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2885 BUG_ON(ret); /* Corruption */
2887 leaf = path->nodes[0];
2888 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2889 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2890 btrfs_mark_buffer_dirty(leaf);
2891 btrfs_release_path(path);
2894 btrfs_abort_transaction(trans, root, ret);
2901 static struct btrfs_block_group_cache *
2902 next_block_group(struct btrfs_root *root,
2903 struct btrfs_block_group_cache *cache)
2905 struct rb_node *node;
2906 spin_lock(&root->fs_info->block_group_cache_lock);
2907 node = rb_next(&cache->cache_node);
2908 btrfs_put_block_group(cache);
2910 cache = rb_entry(node, struct btrfs_block_group_cache,
2912 btrfs_get_block_group(cache);
2915 spin_unlock(&root->fs_info->block_group_cache_lock);
2919 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2920 struct btrfs_trans_handle *trans,
2921 struct btrfs_path *path)
2923 struct btrfs_root *root = block_group->fs_info->tree_root;
2924 struct inode *inode = NULL;
2926 int dcs = BTRFS_DC_ERROR;
2932 * If this block group is smaller than 100 megs don't bother caching the
2935 if (block_group->key.offset < (100 * 1024 * 1024)) {
2936 spin_lock(&block_group->lock);
2937 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2938 spin_unlock(&block_group->lock);
2943 inode = lookup_free_space_inode(root, block_group, path);
2944 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2945 ret = PTR_ERR(inode);
2946 btrfs_release_path(path);
2950 if (IS_ERR(inode)) {
2954 if (block_group->ro)
2957 ret = create_free_space_inode(root, trans, block_group, path);
2963 /* We've already setup this transaction, go ahead and exit */
2964 if (block_group->cache_generation == trans->transid &&
2965 i_size_read(inode)) {
2966 dcs = BTRFS_DC_SETUP;
2971 * We want to set the generation to 0, that way if anything goes wrong
2972 * from here on out we know not to trust this cache when we load up next
2975 BTRFS_I(inode)->generation = 0;
2976 ret = btrfs_update_inode(trans, root, inode);
2979 if (i_size_read(inode) > 0) {
2980 ret = btrfs_truncate_free_space_cache(root, trans, path,
2986 spin_lock(&block_group->lock);
2987 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2988 !btrfs_test_opt(root, SPACE_CACHE)) {
2990 * don't bother trying to write stuff out _if_
2991 * a) we're not cached,
2992 * b) we're with nospace_cache mount option.
2994 dcs = BTRFS_DC_WRITTEN;
2995 spin_unlock(&block_group->lock);
2998 spin_unlock(&block_group->lock);
3001 * Try to preallocate enough space based on how big the block group is.
3002 * Keep in mind this has to include any pinned space which could end up
3003 * taking up quite a bit since it's not folded into the other space
3006 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3011 num_pages *= PAGE_CACHE_SIZE;
3013 ret = btrfs_check_data_free_space(inode, num_pages);
3017 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3018 num_pages, num_pages,
3021 dcs = BTRFS_DC_SETUP;
3022 btrfs_free_reserved_data_space(inode, num_pages);
3027 btrfs_release_path(path);
3029 spin_lock(&block_group->lock);
3030 if (!ret && dcs == BTRFS_DC_SETUP)
3031 block_group->cache_generation = trans->transid;
3032 block_group->disk_cache_state = dcs;
3033 spin_unlock(&block_group->lock);
3038 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3039 struct btrfs_root *root)
3041 struct btrfs_block_group_cache *cache;
3043 struct btrfs_path *path;
3046 path = btrfs_alloc_path();
3052 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3054 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3056 cache = next_block_group(root, cache);
3064 err = cache_save_setup(cache, trans, path);
3065 last = cache->key.objectid + cache->key.offset;
3066 btrfs_put_block_group(cache);
3071 err = btrfs_run_delayed_refs(trans, root,
3073 if (err) /* File system offline */
3077 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3079 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3080 btrfs_put_block_group(cache);
3086 cache = next_block_group(root, cache);
3095 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3096 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3098 last = cache->key.objectid + cache->key.offset;
3100 err = write_one_cache_group(trans, root, path, cache);
3101 if (err) /* File system offline */
3104 btrfs_put_block_group(cache);
3109 * I don't think this is needed since we're just marking our
3110 * preallocated extent as written, but just in case it can't
3114 err = btrfs_run_delayed_refs(trans, root,
3116 if (err) /* File system offline */
3120 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3123 * Really this shouldn't happen, but it could if we
3124 * couldn't write the entire preallocated extent and
3125 * splitting the extent resulted in a new block.
3128 btrfs_put_block_group(cache);
3131 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3133 cache = next_block_group(root, cache);
3142 err = btrfs_write_out_cache(root, trans, cache, path);
3145 * If we didn't have an error then the cache state is still
3146 * NEED_WRITE, so we can set it to WRITTEN.
3148 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3149 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3150 last = cache->key.objectid + cache->key.offset;
3151 btrfs_put_block_group(cache);
3155 btrfs_free_path(path);
3159 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3161 struct btrfs_block_group_cache *block_group;
3164 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3165 if (!block_group || block_group->ro)
3168 btrfs_put_block_group(block_group);
3172 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3173 u64 total_bytes, u64 bytes_used,
3174 struct btrfs_space_info **space_info)
3176 struct btrfs_space_info *found;
3180 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3181 BTRFS_BLOCK_GROUP_RAID10))
3186 found = __find_space_info(info, flags);
3188 spin_lock(&found->lock);
3189 found->total_bytes += total_bytes;
3190 found->disk_total += total_bytes * factor;
3191 found->bytes_used += bytes_used;
3192 found->disk_used += bytes_used * factor;
3194 spin_unlock(&found->lock);
3195 *space_info = found;
3198 found = kzalloc(sizeof(*found), GFP_NOFS);
3202 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3203 INIT_LIST_HEAD(&found->block_groups[i]);
3204 init_rwsem(&found->groups_sem);
3205 spin_lock_init(&found->lock);
3206 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3207 found->total_bytes = total_bytes;
3208 found->disk_total = total_bytes * factor;
3209 found->bytes_used = bytes_used;
3210 found->disk_used = bytes_used * factor;
3211 found->bytes_pinned = 0;
3212 found->bytes_reserved = 0;
3213 found->bytes_readonly = 0;
3214 found->bytes_may_use = 0;
3216 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3217 found->chunk_alloc = 0;
3219 init_waitqueue_head(&found->wait);
3220 *space_info = found;
3221 list_add_rcu(&found->list, &info->space_info);
3222 if (flags & BTRFS_BLOCK_GROUP_DATA)
3223 info->data_sinfo = found;
3227 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3229 u64 extra_flags = chunk_to_extended(flags) &
3230 BTRFS_EXTENDED_PROFILE_MASK;
3232 if (flags & BTRFS_BLOCK_GROUP_DATA)
3233 fs_info->avail_data_alloc_bits |= extra_flags;
3234 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3235 fs_info->avail_metadata_alloc_bits |= extra_flags;
3236 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3237 fs_info->avail_system_alloc_bits |= extra_flags;
3241 * returns target flags in extended format or 0 if restripe for this
3242 * chunk_type is not in progress
3244 * should be called with either volume_mutex or balance_lock held
3246 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3248 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3254 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3255 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3256 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3257 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3258 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3259 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3260 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3261 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3262 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3269 * @flags: available profiles in extended format (see ctree.h)
3271 * Returns reduced profile in chunk format. If profile changing is in
3272 * progress (either running or paused) picks the target profile (if it's
3273 * already available), otherwise falls back to plain reducing.
3275 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3278 * we add in the count of missing devices because we want
3279 * to make sure that any RAID levels on a degraded FS
3280 * continue to be honored.
3282 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3283 root->fs_info->fs_devices->missing_devices;
3287 * see if restripe for this chunk_type is in progress, if so
3288 * try to reduce to the target profile
3290 spin_lock(&root->fs_info->balance_lock);
3291 target = get_restripe_target(root->fs_info, flags);
3293 /* pick target profile only if it's already available */
3294 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3295 spin_unlock(&root->fs_info->balance_lock);
3296 return extended_to_chunk(target);
3299 spin_unlock(&root->fs_info->balance_lock);
3301 if (num_devices == 1)
3302 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3303 if (num_devices < 4)
3304 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3306 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3307 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3308 BTRFS_BLOCK_GROUP_RAID10))) {
3309 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3312 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3313 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3314 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3317 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3318 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3319 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3320 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3321 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3324 return extended_to_chunk(flags);
3327 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3329 if (flags & BTRFS_BLOCK_GROUP_DATA)
3330 flags |= root->fs_info->avail_data_alloc_bits;
3331 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3332 flags |= root->fs_info->avail_system_alloc_bits;
3333 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3334 flags |= root->fs_info->avail_metadata_alloc_bits;
3336 return btrfs_reduce_alloc_profile(root, flags);
3339 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3344 flags = BTRFS_BLOCK_GROUP_DATA;
3345 else if (root == root->fs_info->chunk_root)
3346 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3348 flags = BTRFS_BLOCK_GROUP_METADATA;
3350 return get_alloc_profile(root, flags);
3354 * This will check the space that the inode allocates from to make sure we have
3355 * enough space for bytes.
3357 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3359 struct btrfs_space_info *data_sinfo;
3360 struct btrfs_root *root = BTRFS_I(inode)->root;
3361 struct btrfs_fs_info *fs_info = root->fs_info;
3363 int ret = 0, committed = 0, alloc_chunk = 1;
3365 /* make sure bytes are sectorsize aligned */
3366 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3368 if (root == root->fs_info->tree_root ||
3369 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3374 data_sinfo = fs_info->data_sinfo;
3379 /* make sure we have enough space to handle the data first */
3380 spin_lock(&data_sinfo->lock);
3381 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3382 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3383 data_sinfo->bytes_may_use;
3385 if (used + bytes > data_sinfo->total_bytes) {
3386 struct btrfs_trans_handle *trans;
3389 * if we don't have enough free bytes in this space then we need
3390 * to alloc a new chunk.
3392 if (!data_sinfo->full && alloc_chunk) {
3395 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3396 spin_unlock(&data_sinfo->lock);
3398 alloc_target = btrfs_get_alloc_profile(root, 1);
3399 trans = btrfs_join_transaction(root);
3401 return PTR_ERR(trans);
3403 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3405 CHUNK_ALLOC_NO_FORCE);
3406 btrfs_end_transaction(trans, root);
3415 data_sinfo = fs_info->data_sinfo;
3421 * If we have less pinned bytes than we want to allocate then
3422 * don't bother committing the transaction, it won't help us.
3424 if (data_sinfo->bytes_pinned < bytes)
3426 spin_unlock(&data_sinfo->lock);
3428 /* commit the current transaction and try again */
3431 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3433 trans = btrfs_join_transaction(root);
3435 return PTR_ERR(trans);
3436 ret = btrfs_commit_transaction(trans, root);
3444 data_sinfo->bytes_may_use += bytes;
3445 trace_btrfs_space_reservation(root->fs_info, "space_info",
3446 data_sinfo->flags, bytes, 1);
3447 spin_unlock(&data_sinfo->lock);
3453 * Called if we need to clear a data reservation for this inode.
3455 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3457 struct btrfs_root *root = BTRFS_I(inode)->root;
3458 struct btrfs_space_info *data_sinfo;
3460 /* make sure bytes are sectorsize aligned */
3461 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3463 data_sinfo = root->fs_info->data_sinfo;
3464 spin_lock(&data_sinfo->lock);
3465 data_sinfo->bytes_may_use -= bytes;
3466 trace_btrfs_space_reservation(root->fs_info, "space_info",
3467 data_sinfo->flags, bytes, 0);
3468 spin_unlock(&data_sinfo->lock);
3471 static void force_metadata_allocation(struct btrfs_fs_info *info)
3473 struct list_head *head = &info->space_info;
3474 struct btrfs_space_info *found;
3477 list_for_each_entry_rcu(found, head, list) {
3478 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3479 found->force_alloc = CHUNK_ALLOC_FORCE;
3484 static int should_alloc_chunk(struct btrfs_root *root,
3485 struct btrfs_space_info *sinfo, int force)
3487 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3488 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3489 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3492 if (force == CHUNK_ALLOC_FORCE)
3496 * We need to take into account the global rsv because for all intents
3497 * and purposes it's used space. Don't worry about locking the
3498 * global_rsv, it doesn't change except when the transaction commits.
3500 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3501 num_allocated += global_rsv->size;
3504 * in limited mode, we want to have some free space up to
3505 * about 1% of the FS size.
3507 if (force == CHUNK_ALLOC_LIMITED) {
3508 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3509 thresh = max_t(u64, 64 * 1024 * 1024,
3510 div_factor_fine(thresh, 1));
3512 if (num_bytes - num_allocated < thresh)
3516 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3521 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3525 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3526 type & BTRFS_BLOCK_GROUP_RAID0)
3527 num_dev = root->fs_info->fs_devices->rw_devices;
3528 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3531 num_dev = 1; /* DUP or single */
3533 /* metadata for updaing devices and chunk tree */
3534 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3537 static void check_system_chunk(struct btrfs_trans_handle *trans,
3538 struct btrfs_root *root, u64 type)
3540 struct btrfs_space_info *info;
3544 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3545 spin_lock(&info->lock);
3546 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3547 info->bytes_reserved - info->bytes_readonly;
3548 spin_unlock(&info->lock);
3550 thresh = get_system_chunk_thresh(root, type);
3551 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3552 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3553 left, thresh, type);
3554 dump_space_info(info, 0, 0);
3557 if (left < thresh) {
3560 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3561 btrfs_alloc_chunk(trans, root, flags);
3565 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3566 struct btrfs_root *extent_root, u64 flags, int force)
3568 struct btrfs_space_info *space_info;
3569 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3570 int wait_for_alloc = 0;
3573 /* Don't re-enter if we're already allocating a chunk */
3574 if (trans->allocating_chunk)
3577 space_info = __find_space_info(extent_root->fs_info, flags);
3579 ret = update_space_info(extent_root->fs_info, flags,
3581 BUG_ON(ret); /* -ENOMEM */
3583 BUG_ON(!space_info); /* Logic error */
3586 spin_lock(&space_info->lock);
3587 if (force < space_info->force_alloc)
3588 force = space_info->force_alloc;
3589 if (space_info->full) {
3590 spin_unlock(&space_info->lock);
3594 if (!should_alloc_chunk(extent_root, space_info, force)) {
3595 spin_unlock(&space_info->lock);
3597 } else if (space_info->chunk_alloc) {
3600 space_info->chunk_alloc = 1;
3603 spin_unlock(&space_info->lock);
3605 mutex_lock(&fs_info->chunk_mutex);
3608 * The chunk_mutex is held throughout the entirety of a chunk
3609 * allocation, so once we've acquired the chunk_mutex we know that the
3610 * other guy is done and we need to recheck and see if we should
3613 if (wait_for_alloc) {
3614 mutex_unlock(&fs_info->chunk_mutex);
3619 trans->allocating_chunk = true;
3622 * If we have mixed data/metadata chunks we want to make sure we keep
3623 * allocating mixed chunks instead of individual chunks.
3625 if (btrfs_mixed_space_info(space_info))
3626 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3629 * if we're doing a data chunk, go ahead and make sure that
3630 * we keep a reasonable number of metadata chunks allocated in the
3633 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3634 fs_info->data_chunk_allocations++;
3635 if (!(fs_info->data_chunk_allocations %
3636 fs_info->metadata_ratio))
3637 force_metadata_allocation(fs_info);
3641 * Check if we have enough space in SYSTEM chunk because we may need
3642 * to update devices.
3644 check_system_chunk(trans, extent_root, flags);
3646 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3647 trans->allocating_chunk = false;
3648 if (ret < 0 && ret != -ENOSPC)
3651 spin_lock(&space_info->lock);
3653 space_info->full = 1;
3657 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3658 space_info->chunk_alloc = 0;
3659 spin_unlock(&space_info->lock);
3661 mutex_unlock(&fs_info->chunk_mutex);
3665 static int can_overcommit(struct btrfs_root *root,
3666 struct btrfs_space_info *space_info, u64 bytes,
3667 enum btrfs_reserve_flush_enum flush)
3669 u64 profile = btrfs_get_alloc_profile(root, 0);
3673 used = space_info->bytes_used + space_info->bytes_reserved +
3674 space_info->bytes_pinned + space_info->bytes_readonly +
3675 space_info->bytes_may_use;
3677 spin_lock(&root->fs_info->free_chunk_lock);
3678 avail = root->fs_info->free_chunk_space;
3679 spin_unlock(&root->fs_info->free_chunk_lock);
3682 * If we have dup, raid1 or raid10 then only half of the free
3683 * space is actually useable.
3685 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3686 BTRFS_BLOCK_GROUP_RAID1 |
3687 BTRFS_BLOCK_GROUP_RAID10))
3691 * If we aren't flushing all things, let us overcommit up to
3692 * 1/2th of the space. If we can flush, don't let us overcommit
3693 * too much, let it overcommit up to 1/8 of the space.
3695 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3700 if (used + bytes < space_info->total_bytes + avail)
3705 static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3706 unsigned long nr_pages,
3707 enum wb_reason reason)
3709 /* the flusher is dealing with the dirty inodes now. */
3710 if (writeback_in_progress(sb->s_bdi))
3713 if (down_read_trylock(&sb->s_umount)) {
3714 writeback_inodes_sb_nr(sb, nr_pages, reason);
3715 up_read(&sb->s_umount);
3722 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3723 unsigned long nr_pages)
3725 struct super_block *sb = root->fs_info->sb;
3728 /* If we can not start writeback, just sync all the delalloc file. */
3729 started = writeback_inodes_sb_nr_if_idle_safe(sb, nr_pages,
3730 WB_REASON_FS_FREE_SPACE);
3733 * We needn't worry the filesystem going from r/w to r/o though
3734 * we don't acquire ->s_umount mutex, because the filesystem
3735 * should guarantee the delalloc inodes list be empty after
3736 * the filesystem is readonly(all dirty pages are written to
3739 btrfs_start_delalloc_inodes(root, 0);
3740 btrfs_wait_ordered_extents(root, 0);
3745 * shrink metadata reservation for delalloc
3747 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3750 struct btrfs_block_rsv *block_rsv;
3751 struct btrfs_space_info *space_info;
3752 struct btrfs_trans_handle *trans;
3756 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3758 enum btrfs_reserve_flush_enum flush;
3760 trans = (struct btrfs_trans_handle *)current->journal_info;
3761 block_rsv = &root->fs_info->delalloc_block_rsv;
3762 space_info = block_rsv->space_info;
3765 delalloc_bytes = root->fs_info->delalloc_bytes;
3766 if (delalloc_bytes == 0) {
3769 btrfs_wait_ordered_extents(root, 0);
3773 while (delalloc_bytes && loops < 3) {
3774 max_reclaim = min(delalloc_bytes, to_reclaim);
3775 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3776 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3778 * We need to wait for the async pages to actually start before
3781 wait_event(root->fs_info->async_submit_wait,
3782 !atomic_read(&root->fs_info->async_delalloc_pages));
3785 flush = BTRFS_RESERVE_FLUSH_ALL;
3787 flush = BTRFS_RESERVE_NO_FLUSH;
3788 spin_lock(&space_info->lock);
3789 if (can_overcommit(root, space_info, orig, flush)) {
3790 spin_unlock(&space_info->lock);
3793 spin_unlock(&space_info->lock);
3796 if (wait_ordered && !trans) {
3797 btrfs_wait_ordered_extents(root, 0);
3799 time_left = schedule_timeout_killable(1);
3804 delalloc_bytes = root->fs_info->delalloc_bytes;
3809 * maybe_commit_transaction - possibly commit the transaction if its ok to
3810 * @root - the root we're allocating for
3811 * @bytes - the number of bytes we want to reserve
3812 * @force - force the commit
3814 * This will check to make sure that committing the transaction will actually
3815 * get us somewhere and then commit the transaction if it does. Otherwise it
3816 * will return -ENOSPC.
3818 static int may_commit_transaction(struct btrfs_root *root,
3819 struct btrfs_space_info *space_info,
3820 u64 bytes, int force)
3822 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3823 struct btrfs_trans_handle *trans;
3825 trans = (struct btrfs_trans_handle *)current->journal_info;
3832 /* See if there is enough pinned space to make this reservation */
3833 spin_lock(&space_info->lock);
3834 if (space_info->bytes_pinned >= bytes) {
3835 spin_unlock(&space_info->lock);
3838 spin_unlock(&space_info->lock);
3841 * See if there is some space in the delayed insertion reservation for
3844 if (space_info != delayed_rsv->space_info)
3847 spin_lock(&space_info->lock);
3848 spin_lock(&delayed_rsv->lock);
3849 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3850 spin_unlock(&delayed_rsv->lock);
3851 spin_unlock(&space_info->lock);
3854 spin_unlock(&delayed_rsv->lock);
3855 spin_unlock(&space_info->lock);
3858 trans = btrfs_join_transaction(root);
3862 return btrfs_commit_transaction(trans, root);
3866 FLUSH_DELAYED_ITEMS_NR = 1,
3867 FLUSH_DELAYED_ITEMS = 2,
3869 FLUSH_DELALLOC_WAIT = 4,
3874 static int flush_space(struct btrfs_root *root,
3875 struct btrfs_space_info *space_info, u64 num_bytes,
3876 u64 orig_bytes, int state)
3878 struct btrfs_trans_handle *trans;
3883 case FLUSH_DELAYED_ITEMS_NR:
3884 case FLUSH_DELAYED_ITEMS:
3885 if (state == FLUSH_DELAYED_ITEMS_NR) {
3886 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3888 nr = (int)div64_u64(num_bytes, bytes);
3895 trans = btrfs_join_transaction(root);
3896 if (IS_ERR(trans)) {
3897 ret = PTR_ERR(trans);
3900 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3901 btrfs_end_transaction(trans, root);
3903 case FLUSH_DELALLOC:
3904 case FLUSH_DELALLOC_WAIT:
3905 shrink_delalloc(root, num_bytes, orig_bytes,
3906 state == FLUSH_DELALLOC_WAIT);
3909 trans = btrfs_join_transaction(root);
3910 if (IS_ERR(trans)) {
3911 ret = PTR_ERR(trans);
3914 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3915 btrfs_get_alloc_profile(root, 0),
3916 CHUNK_ALLOC_NO_FORCE);
3917 btrfs_end_transaction(trans, root);
3922 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3932 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3933 * @root - the root we're allocating for
3934 * @block_rsv - the block_rsv we're allocating for
3935 * @orig_bytes - the number of bytes we want
3936 * @flush - wether or not we can flush to make our reservation
3938 * This will reserve orgi_bytes number of bytes from the space info associated
3939 * with the block_rsv. If there is not enough space it will make an attempt to
3940 * flush out space to make room. It will do this by flushing delalloc if
3941 * possible or committing the transaction. If flush is 0 then no attempts to
3942 * regain reservations will be made and this will fail if there is not enough
3945 static int reserve_metadata_bytes(struct btrfs_root *root,
3946 struct btrfs_block_rsv *block_rsv,
3948 enum btrfs_reserve_flush_enum flush)
3950 struct btrfs_space_info *space_info = block_rsv->space_info;
3952 u64 num_bytes = orig_bytes;
3953 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3955 bool flushing = false;
3959 spin_lock(&space_info->lock);
3961 * We only want to wait if somebody other than us is flushing and we
3962 * are actually allowed to flush all things.
3964 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3965 space_info->flush) {
3966 spin_unlock(&space_info->lock);
3968 * If we have a trans handle we can't wait because the flusher
3969 * may have to commit the transaction, which would mean we would
3970 * deadlock since we are waiting for the flusher to finish, but
3971 * hold the current transaction open.
3973 if (current->journal_info)
3975 ret = wait_event_killable(space_info->wait, !space_info->flush);
3976 /* Must have been killed, return */
3980 spin_lock(&space_info->lock);
3984 used = space_info->bytes_used + space_info->bytes_reserved +
3985 space_info->bytes_pinned + space_info->bytes_readonly +
3986 space_info->bytes_may_use;
3989 * The idea here is that we've not already over-reserved the block group
3990 * then we can go ahead and save our reservation first and then start
3991 * flushing if we need to. Otherwise if we've already overcommitted
3992 * lets start flushing stuff first and then come back and try to make
3995 if (used <= space_info->total_bytes) {
3996 if (used + orig_bytes <= space_info->total_bytes) {
3997 space_info->bytes_may_use += orig_bytes;
3998 trace_btrfs_space_reservation(root->fs_info,
3999 "space_info", space_info->flags, orig_bytes, 1);
4003 * Ok set num_bytes to orig_bytes since we aren't
4004 * overocmmitted, this way we only try and reclaim what
4007 num_bytes = orig_bytes;
4011 * Ok we're over committed, set num_bytes to the overcommitted
4012 * amount plus the amount of bytes that we need for this
4015 num_bytes = used - space_info->total_bytes +
4019 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4020 space_info->bytes_may_use += orig_bytes;
4021 trace_btrfs_space_reservation(root->fs_info, "space_info",
4022 space_info->flags, orig_bytes,
4028 * Couldn't make our reservation, save our place so while we're trying
4029 * to reclaim space we can actually use it instead of somebody else
4030 * stealing it from us.
4032 * We make the other tasks wait for the flush only when we can flush
4035 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4037 space_info->flush = 1;
4040 spin_unlock(&space_info->lock);
4042 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4045 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4050 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4051 * would happen. So skip delalloc flush.
4053 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4054 (flush_state == FLUSH_DELALLOC ||
4055 flush_state == FLUSH_DELALLOC_WAIT))
4056 flush_state = ALLOC_CHUNK;
4060 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4061 flush_state < COMMIT_TRANS)
4063 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4064 flush_state <= COMMIT_TRANS)
4069 spin_lock(&space_info->lock);
4070 space_info->flush = 0;
4071 wake_up_all(&space_info->wait);
4072 spin_unlock(&space_info->lock);
4077 static struct btrfs_block_rsv *get_block_rsv(
4078 const struct btrfs_trans_handle *trans,
4079 const struct btrfs_root *root)
4081 struct btrfs_block_rsv *block_rsv = NULL;
4084 block_rsv = trans->block_rsv;
4086 if (root == root->fs_info->csum_root && trans->adding_csums)
4087 block_rsv = trans->block_rsv;
4090 block_rsv = root->block_rsv;
4093 block_rsv = &root->fs_info->empty_block_rsv;
4098 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4102 spin_lock(&block_rsv->lock);
4103 if (block_rsv->reserved >= num_bytes) {
4104 block_rsv->reserved -= num_bytes;
4105 if (block_rsv->reserved < block_rsv->size)
4106 block_rsv->full = 0;
4109 spin_unlock(&block_rsv->lock);
4113 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4114 u64 num_bytes, int update_size)
4116 spin_lock(&block_rsv->lock);
4117 block_rsv->reserved += num_bytes;
4119 block_rsv->size += num_bytes;
4120 else if (block_rsv->reserved >= block_rsv->size)
4121 block_rsv->full = 1;
4122 spin_unlock(&block_rsv->lock);
4125 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4126 struct btrfs_block_rsv *block_rsv,
4127 struct btrfs_block_rsv *dest, u64 num_bytes)
4129 struct btrfs_space_info *space_info = block_rsv->space_info;
4131 spin_lock(&block_rsv->lock);
4132 if (num_bytes == (u64)-1)
4133 num_bytes = block_rsv->size;
4134 block_rsv->size -= num_bytes;
4135 if (block_rsv->reserved >= block_rsv->size) {
4136 num_bytes = block_rsv->reserved - block_rsv->size;
4137 block_rsv->reserved = block_rsv->size;
4138 block_rsv->full = 1;
4142 spin_unlock(&block_rsv->lock);
4144 if (num_bytes > 0) {
4146 spin_lock(&dest->lock);
4150 bytes_to_add = dest->size - dest->reserved;
4151 bytes_to_add = min(num_bytes, bytes_to_add);
4152 dest->reserved += bytes_to_add;
4153 if (dest->reserved >= dest->size)
4155 num_bytes -= bytes_to_add;
4157 spin_unlock(&dest->lock);
4160 spin_lock(&space_info->lock);
4161 space_info->bytes_may_use -= num_bytes;
4162 trace_btrfs_space_reservation(fs_info, "space_info",
4163 space_info->flags, num_bytes, 0);
4164 space_info->reservation_progress++;
4165 spin_unlock(&space_info->lock);
4170 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4171 struct btrfs_block_rsv *dst, u64 num_bytes)
4175 ret = block_rsv_use_bytes(src, num_bytes);
4179 block_rsv_add_bytes(dst, num_bytes, 1);
4183 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4185 memset(rsv, 0, sizeof(*rsv));
4186 spin_lock_init(&rsv->lock);
4190 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4191 unsigned short type)
4193 struct btrfs_block_rsv *block_rsv;
4194 struct btrfs_fs_info *fs_info = root->fs_info;
4196 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4200 btrfs_init_block_rsv(block_rsv, type);
4201 block_rsv->space_info = __find_space_info(fs_info,
4202 BTRFS_BLOCK_GROUP_METADATA);
4206 void btrfs_free_block_rsv(struct btrfs_root *root,
4207 struct btrfs_block_rsv *rsv)
4211 btrfs_block_rsv_release(root, rsv, (u64)-1);
4215 int btrfs_block_rsv_add(struct btrfs_root *root,
4216 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4217 enum btrfs_reserve_flush_enum flush)
4224 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4226 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4233 int btrfs_block_rsv_check(struct btrfs_root *root,
4234 struct btrfs_block_rsv *block_rsv, int min_factor)
4242 spin_lock(&block_rsv->lock);
4243 num_bytes = div_factor(block_rsv->size, min_factor);
4244 if (block_rsv->reserved >= num_bytes)
4246 spin_unlock(&block_rsv->lock);
4251 int btrfs_block_rsv_refill(struct btrfs_root *root,
4252 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4253 enum btrfs_reserve_flush_enum flush)
4261 spin_lock(&block_rsv->lock);
4262 num_bytes = min_reserved;
4263 if (block_rsv->reserved >= num_bytes)
4266 num_bytes -= block_rsv->reserved;
4267 spin_unlock(&block_rsv->lock);
4272 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4274 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4281 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4282 struct btrfs_block_rsv *dst_rsv,
4285 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4288 void btrfs_block_rsv_release(struct btrfs_root *root,
4289 struct btrfs_block_rsv *block_rsv,
4292 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4293 if (global_rsv->full || global_rsv == block_rsv ||
4294 block_rsv->space_info != global_rsv->space_info)
4296 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4301 * helper to calculate size of global block reservation.
4302 * the desired value is sum of space used by extent tree,
4303 * checksum tree and root tree
4305 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4307 struct btrfs_space_info *sinfo;
4311 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4313 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4314 spin_lock(&sinfo->lock);
4315 data_used = sinfo->bytes_used;
4316 spin_unlock(&sinfo->lock);
4318 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4319 spin_lock(&sinfo->lock);
4320 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4322 meta_used = sinfo->bytes_used;
4323 spin_unlock(&sinfo->lock);
4325 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4327 num_bytes += div64_u64(data_used + meta_used, 50);
4329 if (num_bytes * 3 > meta_used)
4330 num_bytes = div64_u64(meta_used, 3);
4332 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4335 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4337 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4338 struct btrfs_space_info *sinfo = block_rsv->space_info;
4341 num_bytes = calc_global_metadata_size(fs_info);
4343 spin_lock(&sinfo->lock);
4344 spin_lock(&block_rsv->lock);
4346 block_rsv->size = num_bytes;
4348 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4349 sinfo->bytes_reserved + sinfo->bytes_readonly +
4350 sinfo->bytes_may_use;
4352 if (sinfo->total_bytes > num_bytes) {
4353 num_bytes = sinfo->total_bytes - num_bytes;
4354 block_rsv->reserved += num_bytes;
4355 sinfo->bytes_may_use += num_bytes;
4356 trace_btrfs_space_reservation(fs_info, "space_info",
4357 sinfo->flags, num_bytes, 1);
4360 if (block_rsv->reserved >= block_rsv->size) {
4361 num_bytes = block_rsv->reserved - block_rsv->size;
4362 sinfo->bytes_may_use -= num_bytes;
4363 trace_btrfs_space_reservation(fs_info, "space_info",
4364 sinfo->flags, num_bytes, 0);
4365 sinfo->reservation_progress++;
4366 block_rsv->reserved = block_rsv->size;
4367 block_rsv->full = 1;
4370 spin_unlock(&block_rsv->lock);
4371 spin_unlock(&sinfo->lock);
4374 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4376 struct btrfs_space_info *space_info;
4378 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4379 fs_info->chunk_block_rsv.space_info = space_info;
4381 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4382 fs_info->global_block_rsv.space_info = space_info;
4383 fs_info->delalloc_block_rsv.space_info = space_info;
4384 fs_info->trans_block_rsv.space_info = space_info;
4385 fs_info->empty_block_rsv.space_info = space_info;
4386 fs_info->delayed_block_rsv.space_info = space_info;
4388 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4389 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4390 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4391 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4392 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4394 update_global_block_rsv(fs_info);
4397 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4399 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4401 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4402 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4403 WARN_ON(fs_info->trans_block_rsv.size > 0);
4404 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4405 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4406 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4407 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4408 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4411 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4412 struct btrfs_root *root)
4414 if (!trans->block_rsv)
4417 if (!trans->bytes_reserved)
4420 trace_btrfs_space_reservation(root->fs_info, "transaction",
4421 trans->transid, trans->bytes_reserved, 0);
4422 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4423 trans->bytes_reserved = 0;
4426 /* Can only return 0 or -ENOSPC */
4427 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4428 struct inode *inode)
4430 struct btrfs_root *root = BTRFS_I(inode)->root;
4431 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4432 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4435 * We need to hold space in order to delete our orphan item once we've
4436 * added it, so this takes the reservation so we can release it later
4437 * when we are truly done with the orphan item.
4439 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4440 trace_btrfs_space_reservation(root->fs_info, "orphan",
4441 btrfs_ino(inode), num_bytes, 1);
4442 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4445 void btrfs_orphan_release_metadata(struct inode *inode)
4447 struct btrfs_root *root = BTRFS_I(inode)->root;
4448 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4449 trace_btrfs_space_reservation(root->fs_info, "orphan",
4450 btrfs_ino(inode), num_bytes, 0);
4451 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4454 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4455 struct btrfs_pending_snapshot *pending)
4457 struct btrfs_root *root = pending->root;
4458 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4459 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4461 * two for root back/forward refs, two for directory entries,
4462 * one for root of the snapshot and one for parent inode.
4464 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4465 dst_rsv->space_info = src_rsv->space_info;
4466 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4470 * drop_outstanding_extent - drop an outstanding extent
4471 * @inode: the inode we're dropping the extent for
4473 * This is called when we are freeing up an outstanding extent, either called
4474 * after an error or after an extent is written. This will return the number of
4475 * reserved extents that need to be freed. This must be called with
4476 * BTRFS_I(inode)->lock held.
4478 static unsigned drop_outstanding_extent(struct inode *inode)
4480 unsigned drop_inode_space = 0;
4481 unsigned dropped_extents = 0;
4483 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4484 BTRFS_I(inode)->outstanding_extents--;
4486 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4487 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4488 &BTRFS_I(inode)->runtime_flags))
4489 drop_inode_space = 1;
4492 * If we have more or the same amount of outsanding extents than we have
4493 * reserved then we need to leave the reserved extents count alone.
4495 if (BTRFS_I(inode)->outstanding_extents >=
4496 BTRFS_I(inode)->reserved_extents)
4497 return drop_inode_space;
4499 dropped_extents = BTRFS_I(inode)->reserved_extents -
4500 BTRFS_I(inode)->outstanding_extents;
4501 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4502 return dropped_extents + drop_inode_space;
4506 * calc_csum_metadata_size - return the amount of metada space that must be
4507 * reserved/free'd for the given bytes.
4508 * @inode: the inode we're manipulating
4509 * @num_bytes: the number of bytes in question
4510 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4512 * This adjusts the number of csum_bytes in the inode and then returns the
4513 * correct amount of metadata that must either be reserved or freed. We
4514 * calculate how many checksums we can fit into one leaf and then divide the
4515 * number of bytes that will need to be checksumed by this value to figure out
4516 * how many checksums will be required. If we are adding bytes then the number
4517 * may go up and we will return the number of additional bytes that must be
4518 * reserved. If it is going down we will return the number of bytes that must
4521 * This must be called with BTRFS_I(inode)->lock held.
4523 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4526 struct btrfs_root *root = BTRFS_I(inode)->root;
4528 int num_csums_per_leaf;
4532 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4533 BTRFS_I(inode)->csum_bytes == 0)
4536 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4538 BTRFS_I(inode)->csum_bytes += num_bytes;
4540 BTRFS_I(inode)->csum_bytes -= num_bytes;
4541 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4542 num_csums_per_leaf = (int)div64_u64(csum_size,
4543 sizeof(struct btrfs_csum_item) +
4544 sizeof(struct btrfs_disk_key));
4545 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4546 num_csums = num_csums + num_csums_per_leaf - 1;
4547 num_csums = num_csums / num_csums_per_leaf;
4549 old_csums = old_csums + num_csums_per_leaf - 1;
4550 old_csums = old_csums / num_csums_per_leaf;
4552 /* No change, no need to reserve more */
4553 if (old_csums == num_csums)
4557 return btrfs_calc_trans_metadata_size(root,
4558 num_csums - old_csums);
4560 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4563 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4565 struct btrfs_root *root = BTRFS_I(inode)->root;
4566 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4569 unsigned nr_extents = 0;
4570 int extra_reserve = 0;
4571 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4573 bool delalloc_lock = true;
4575 /* If we are a free space inode we need to not flush since we will be in
4576 * the middle of a transaction commit. We also don't need the delalloc
4577 * mutex since we won't race with anybody. We need this mostly to make
4578 * lockdep shut its filthy mouth.
4580 if (btrfs_is_free_space_inode(inode)) {
4581 flush = BTRFS_RESERVE_NO_FLUSH;
4582 delalloc_lock = false;
4585 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4586 btrfs_transaction_in_commit(root->fs_info))
4587 schedule_timeout(1);
4590 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4592 num_bytes = ALIGN(num_bytes, root->sectorsize);
4594 spin_lock(&BTRFS_I(inode)->lock);
4595 BTRFS_I(inode)->outstanding_extents++;
4597 if (BTRFS_I(inode)->outstanding_extents >
4598 BTRFS_I(inode)->reserved_extents)
4599 nr_extents = BTRFS_I(inode)->outstanding_extents -
4600 BTRFS_I(inode)->reserved_extents;
4603 * Add an item to reserve for updating the inode when we complete the
4606 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4607 &BTRFS_I(inode)->runtime_flags)) {
4612 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4613 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4614 csum_bytes = BTRFS_I(inode)->csum_bytes;
4615 spin_unlock(&BTRFS_I(inode)->lock);
4617 if (root->fs_info->quota_enabled)
4618 ret = btrfs_qgroup_reserve(root, num_bytes +
4619 nr_extents * root->leafsize);
4622 * ret != 0 here means the qgroup reservation failed, we go straight to
4623 * the shared error handling then.
4626 ret = reserve_metadata_bytes(root, block_rsv,
4633 spin_lock(&BTRFS_I(inode)->lock);
4634 dropped = drop_outstanding_extent(inode);
4636 * If the inodes csum_bytes is the same as the original
4637 * csum_bytes then we know we haven't raced with any free()ers
4638 * so we can just reduce our inodes csum bytes and carry on.
4639 * Otherwise we have to do the normal free thing to account for
4640 * the case that the free side didn't free up its reserve
4641 * because of this outstanding reservation.
4643 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4644 calc_csum_metadata_size(inode, num_bytes, 0);
4646 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4647 spin_unlock(&BTRFS_I(inode)->lock);
4649 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4652 btrfs_block_rsv_release(root, block_rsv, to_free);
4653 trace_btrfs_space_reservation(root->fs_info,
4658 if (root->fs_info->quota_enabled) {
4659 btrfs_qgroup_free(root, num_bytes +
4660 nr_extents * root->leafsize);
4663 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4667 spin_lock(&BTRFS_I(inode)->lock);
4668 if (extra_reserve) {
4669 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4670 &BTRFS_I(inode)->runtime_flags);
4673 BTRFS_I(inode)->reserved_extents += nr_extents;
4674 spin_unlock(&BTRFS_I(inode)->lock);
4677 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4680 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4681 btrfs_ino(inode), to_reserve, 1);
4682 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4688 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4689 * @inode: the inode to release the reservation for
4690 * @num_bytes: the number of bytes we're releasing
4692 * This will release the metadata reservation for an inode. This can be called
4693 * once we complete IO for a given set of bytes to release their metadata
4696 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4698 struct btrfs_root *root = BTRFS_I(inode)->root;
4702 num_bytes = ALIGN(num_bytes, root->sectorsize);
4703 spin_lock(&BTRFS_I(inode)->lock);
4704 dropped = drop_outstanding_extent(inode);
4706 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4707 spin_unlock(&BTRFS_I(inode)->lock);
4709 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4711 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4712 btrfs_ino(inode), to_free, 0);
4713 if (root->fs_info->quota_enabled) {
4714 btrfs_qgroup_free(root, num_bytes +
4715 dropped * root->leafsize);
4718 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4723 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4724 * @inode: inode we're writing to
4725 * @num_bytes: the number of bytes we want to allocate
4727 * This will do the following things
4729 * o reserve space in the data space info for num_bytes
4730 * o reserve space in the metadata space info based on number of outstanding
4731 * extents and how much csums will be needed
4732 * o add to the inodes ->delalloc_bytes
4733 * o add it to the fs_info's delalloc inodes list.
4735 * This will return 0 for success and -ENOSPC if there is no space left.
4737 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4741 ret = btrfs_check_data_free_space(inode, num_bytes);
4745 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4747 btrfs_free_reserved_data_space(inode, num_bytes);
4755 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4756 * @inode: inode we're releasing space for
4757 * @num_bytes: the number of bytes we want to free up
4759 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4760 * called in the case that we don't need the metadata AND data reservations
4761 * anymore. So if there is an error or we insert an inline extent.
4763 * This function will release the metadata space that was not used and will
4764 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4765 * list if there are no delalloc bytes left.
4767 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4769 btrfs_delalloc_release_metadata(inode, num_bytes);
4770 btrfs_free_reserved_data_space(inode, num_bytes);
4773 static int update_block_group(struct btrfs_trans_handle *trans,
4774 struct btrfs_root *root,
4775 u64 bytenr, u64 num_bytes, int alloc)
4777 struct btrfs_block_group_cache *cache = NULL;
4778 struct btrfs_fs_info *info = root->fs_info;
4779 u64 total = num_bytes;
4784 /* block accounting for super block */
4785 spin_lock(&info->delalloc_lock);
4786 old_val = btrfs_super_bytes_used(info->super_copy);
4788 old_val += num_bytes;
4790 old_val -= num_bytes;
4791 btrfs_set_super_bytes_used(info->super_copy, old_val);
4792 spin_unlock(&info->delalloc_lock);
4795 cache = btrfs_lookup_block_group(info, bytenr);
4798 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4799 BTRFS_BLOCK_GROUP_RAID1 |
4800 BTRFS_BLOCK_GROUP_RAID10))
4805 * If this block group has free space cache written out, we
4806 * need to make sure to load it if we are removing space. This
4807 * is because we need the unpinning stage to actually add the
4808 * space back to the block group, otherwise we will leak space.
4810 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4811 cache_block_group(cache, trans, NULL, 1);
4813 byte_in_group = bytenr - cache->key.objectid;
4814 WARN_ON(byte_in_group > cache->key.offset);
4816 spin_lock(&cache->space_info->lock);
4817 spin_lock(&cache->lock);
4819 if (btrfs_test_opt(root, SPACE_CACHE) &&
4820 cache->disk_cache_state < BTRFS_DC_CLEAR)
4821 cache->disk_cache_state = BTRFS_DC_CLEAR;
4824 old_val = btrfs_block_group_used(&cache->item);
4825 num_bytes = min(total, cache->key.offset - byte_in_group);
4827 old_val += num_bytes;
4828 btrfs_set_block_group_used(&cache->item, old_val);
4829 cache->reserved -= num_bytes;
4830 cache->space_info->bytes_reserved -= num_bytes;
4831 cache->space_info->bytes_used += num_bytes;
4832 cache->space_info->disk_used += num_bytes * factor;
4833 spin_unlock(&cache->lock);
4834 spin_unlock(&cache->space_info->lock);
4836 old_val -= num_bytes;
4837 btrfs_set_block_group_used(&cache->item, old_val);
4838 cache->pinned += num_bytes;
4839 cache->space_info->bytes_pinned += num_bytes;
4840 cache->space_info->bytes_used -= num_bytes;
4841 cache->space_info->disk_used -= num_bytes * factor;
4842 spin_unlock(&cache->lock);
4843 spin_unlock(&cache->space_info->lock);
4845 set_extent_dirty(info->pinned_extents,
4846 bytenr, bytenr + num_bytes - 1,
4847 GFP_NOFS | __GFP_NOFAIL);
4849 btrfs_put_block_group(cache);
4851 bytenr += num_bytes;
4856 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4858 struct btrfs_block_group_cache *cache;
4861 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4865 bytenr = cache->key.objectid;
4866 btrfs_put_block_group(cache);
4871 static int pin_down_extent(struct btrfs_root *root,
4872 struct btrfs_block_group_cache *cache,
4873 u64 bytenr, u64 num_bytes, int reserved)
4875 spin_lock(&cache->space_info->lock);
4876 spin_lock(&cache->lock);
4877 cache->pinned += num_bytes;
4878 cache->space_info->bytes_pinned += num_bytes;
4880 cache->reserved -= num_bytes;
4881 cache->space_info->bytes_reserved -= num_bytes;
4883 spin_unlock(&cache->lock);
4884 spin_unlock(&cache->space_info->lock);
4886 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4887 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4892 * this function must be called within transaction
4894 int btrfs_pin_extent(struct btrfs_root *root,
4895 u64 bytenr, u64 num_bytes, int reserved)
4897 struct btrfs_block_group_cache *cache;
4899 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4900 BUG_ON(!cache); /* Logic error */
4902 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4904 btrfs_put_block_group(cache);
4909 * this function must be called within transaction
4911 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4912 struct btrfs_root *root,
4913 u64 bytenr, u64 num_bytes)
4915 struct btrfs_block_group_cache *cache;
4917 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4918 BUG_ON(!cache); /* Logic error */
4921 * pull in the free space cache (if any) so that our pin
4922 * removes the free space from the cache. We have load_only set
4923 * to one because the slow code to read in the free extents does check
4924 * the pinned extents.
4926 cache_block_group(cache, trans, root, 1);
4928 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4930 /* remove us from the free space cache (if we're there at all) */
4931 btrfs_remove_free_space(cache, bytenr, num_bytes);
4932 btrfs_put_block_group(cache);
4937 * btrfs_update_reserved_bytes - update the block_group and space info counters
4938 * @cache: The cache we are manipulating
4939 * @num_bytes: The number of bytes in question
4940 * @reserve: One of the reservation enums
4942 * This is called by the allocator when it reserves space, or by somebody who is
4943 * freeing space that was never actually used on disk. For example if you
4944 * reserve some space for a new leaf in transaction A and before transaction A
4945 * commits you free that leaf, you call this with reserve set to 0 in order to
4946 * clear the reservation.
4948 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4949 * ENOSPC accounting. For data we handle the reservation through clearing the
4950 * delalloc bits in the io_tree. We have to do this since we could end up
4951 * allocating less disk space for the amount of data we have reserved in the
4952 * case of compression.
4954 * If this is a reservation and the block group has become read only we cannot
4955 * make the reservation and return -EAGAIN, otherwise this function always
4958 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4959 u64 num_bytes, int reserve)
4961 struct btrfs_space_info *space_info = cache->space_info;
4964 spin_lock(&space_info->lock);
4965 spin_lock(&cache->lock);
4966 if (reserve != RESERVE_FREE) {
4970 cache->reserved += num_bytes;
4971 space_info->bytes_reserved += num_bytes;
4972 if (reserve == RESERVE_ALLOC) {
4973 trace_btrfs_space_reservation(cache->fs_info,
4974 "space_info", space_info->flags,
4976 space_info->bytes_may_use -= num_bytes;
4981 space_info->bytes_readonly += num_bytes;
4982 cache->reserved -= num_bytes;
4983 space_info->bytes_reserved -= num_bytes;
4984 space_info->reservation_progress++;
4986 spin_unlock(&cache->lock);
4987 spin_unlock(&space_info->lock);
4991 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4992 struct btrfs_root *root)
4994 struct btrfs_fs_info *fs_info = root->fs_info;
4995 struct btrfs_caching_control *next;
4996 struct btrfs_caching_control *caching_ctl;
4997 struct btrfs_block_group_cache *cache;
4999 down_write(&fs_info->extent_commit_sem);
5001 list_for_each_entry_safe(caching_ctl, next,
5002 &fs_info->caching_block_groups, list) {
5003 cache = caching_ctl->block_group;
5004 if (block_group_cache_done(cache)) {
5005 cache->last_byte_to_unpin = (u64)-1;
5006 list_del_init(&caching_ctl->list);
5007 put_caching_control(caching_ctl);
5009 cache->last_byte_to_unpin = caching_ctl->progress;
5013 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5014 fs_info->pinned_extents = &fs_info->freed_extents[1];
5016 fs_info->pinned_extents = &fs_info->freed_extents[0];
5018 up_write(&fs_info->extent_commit_sem);
5020 update_global_block_rsv(fs_info);
5023 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5025 struct btrfs_fs_info *fs_info = root->fs_info;
5026 struct btrfs_block_group_cache *cache = NULL;
5027 struct btrfs_space_info *space_info;
5028 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5032 while (start <= end) {
5035 start >= cache->key.objectid + cache->key.offset) {
5037 btrfs_put_block_group(cache);
5038 cache = btrfs_lookup_block_group(fs_info, start);
5039 BUG_ON(!cache); /* Logic error */
5042 len = cache->key.objectid + cache->key.offset - start;
5043 len = min(len, end + 1 - start);
5045 if (start < cache->last_byte_to_unpin) {
5046 len = min(len, cache->last_byte_to_unpin - start);
5047 btrfs_add_free_space(cache, start, len);
5051 space_info = cache->space_info;
5053 spin_lock(&space_info->lock);
5054 spin_lock(&cache->lock);
5055 cache->pinned -= len;
5056 space_info->bytes_pinned -= len;
5058 space_info->bytes_readonly += len;
5061 spin_unlock(&cache->lock);
5062 if (!readonly && global_rsv->space_info == space_info) {
5063 spin_lock(&global_rsv->lock);
5064 if (!global_rsv->full) {
5065 len = min(len, global_rsv->size -
5066 global_rsv->reserved);
5067 global_rsv->reserved += len;
5068 space_info->bytes_may_use += len;
5069 if (global_rsv->reserved >= global_rsv->size)
5070 global_rsv->full = 1;
5072 spin_unlock(&global_rsv->lock);
5074 spin_unlock(&space_info->lock);
5078 btrfs_put_block_group(cache);
5082 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5083 struct btrfs_root *root)
5085 struct btrfs_fs_info *fs_info = root->fs_info;
5086 struct extent_io_tree *unpin;
5094 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5095 unpin = &fs_info->freed_extents[1];
5097 unpin = &fs_info->freed_extents[0];
5100 ret = find_first_extent_bit(unpin, 0, &start, &end,
5101 EXTENT_DIRTY, NULL);
5105 if (btrfs_test_opt(root, DISCARD))
5106 ret = btrfs_discard_extent(root, start,
5107 end + 1 - start, NULL);
5109 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5110 unpin_extent_range(root, start, end);
5117 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5118 struct btrfs_root *root,
5119 u64 bytenr, u64 num_bytes, u64 parent,
5120 u64 root_objectid, u64 owner_objectid,
5121 u64 owner_offset, int refs_to_drop,
5122 struct btrfs_delayed_extent_op *extent_op)
5124 struct btrfs_key key;
5125 struct btrfs_path *path;
5126 struct btrfs_fs_info *info = root->fs_info;
5127 struct btrfs_root *extent_root = info->extent_root;
5128 struct extent_buffer *leaf;
5129 struct btrfs_extent_item *ei;
5130 struct btrfs_extent_inline_ref *iref;
5133 int extent_slot = 0;
5134 int found_extent = 0;
5139 path = btrfs_alloc_path();
5144 path->leave_spinning = 1;
5146 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5147 BUG_ON(!is_data && refs_to_drop != 1);
5149 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5150 bytenr, num_bytes, parent,
5151 root_objectid, owner_objectid,
5154 extent_slot = path->slots[0];
5155 while (extent_slot >= 0) {
5156 btrfs_item_key_to_cpu(path->nodes[0], &key,
5158 if (key.objectid != bytenr)
5160 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5161 key.offset == num_bytes) {
5165 if (path->slots[0] - extent_slot > 5)
5169 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5170 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5171 if (found_extent && item_size < sizeof(*ei))
5174 if (!found_extent) {
5176 ret = remove_extent_backref(trans, extent_root, path,
5180 btrfs_abort_transaction(trans, extent_root, ret);
5183 btrfs_release_path(path);
5184 path->leave_spinning = 1;
5186 key.objectid = bytenr;
5187 key.type = BTRFS_EXTENT_ITEM_KEY;
5188 key.offset = num_bytes;
5190 ret = btrfs_search_slot(trans, extent_root,
5193 printk(KERN_ERR "umm, got %d back from search"
5194 ", was looking for %llu\n", ret,
5195 (unsigned long long)bytenr);
5197 btrfs_print_leaf(extent_root,
5201 btrfs_abort_transaction(trans, extent_root, ret);
5204 extent_slot = path->slots[0];
5206 } else if (ret == -ENOENT) {
5207 btrfs_print_leaf(extent_root, path->nodes[0]);
5209 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5210 "parent %llu root %llu owner %llu offset %llu\n",
5211 (unsigned long long)bytenr,
5212 (unsigned long long)parent,
5213 (unsigned long long)root_objectid,
5214 (unsigned long long)owner_objectid,
5215 (unsigned long long)owner_offset);
5217 btrfs_abort_transaction(trans, extent_root, ret);
5221 leaf = path->nodes[0];
5222 item_size = btrfs_item_size_nr(leaf, extent_slot);
5223 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5224 if (item_size < sizeof(*ei)) {
5225 BUG_ON(found_extent || extent_slot != path->slots[0]);
5226 ret = convert_extent_item_v0(trans, extent_root, path,
5229 btrfs_abort_transaction(trans, extent_root, ret);
5233 btrfs_release_path(path);
5234 path->leave_spinning = 1;
5236 key.objectid = bytenr;
5237 key.type = BTRFS_EXTENT_ITEM_KEY;
5238 key.offset = num_bytes;
5240 ret = btrfs_search_slot(trans, extent_root, &key, path,
5243 printk(KERN_ERR "umm, got %d back from search"
5244 ", was looking for %llu\n", ret,
5245 (unsigned long long)bytenr);
5246 btrfs_print_leaf(extent_root, path->nodes[0]);
5249 btrfs_abort_transaction(trans, extent_root, ret);
5253 extent_slot = path->slots[0];
5254 leaf = path->nodes[0];
5255 item_size = btrfs_item_size_nr(leaf, extent_slot);
5258 BUG_ON(item_size < sizeof(*ei));
5259 ei = btrfs_item_ptr(leaf, extent_slot,
5260 struct btrfs_extent_item);
5261 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5262 struct btrfs_tree_block_info *bi;
5263 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5264 bi = (struct btrfs_tree_block_info *)(ei + 1);
5265 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5268 refs = btrfs_extent_refs(leaf, ei);
5269 BUG_ON(refs < refs_to_drop);
5270 refs -= refs_to_drop;
5274 __run_delayed_extent_op(extent_op, leaf, ei);
5276 * In the case of inline back ref, reference count will
5277 * be updated by remove_extent_backref
5280 BUG_ON(!found_extent);
5282 btrfs_set_extent_refs(leaf, ei, refs);
5283 btrfs_mark_buffer_dirty(leaf);
5286 ret = remove_extent_backref(trans, extent_root, path,
5290 btrfs_abort_transaction(trans, extent_root, ret);
5296 BUG_ON(is_data && refs_to_drop !=
5297 extent_data_ref_count(root, path, iref));
5299 BUG_ON(path->slots[0] != extent_slot);
5301 BUG_ON(path->slots[0] != extent_slot + 1);
5302 path->slots[0] = extent_slot;
5307 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5310 btrfs_abort_transaction(trans, extent_root, ret);
5313 btrfs_release_path(path);
5316 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5318 btrfs_abort_transaction(trans, extent_root, ret);
5323 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5325 btrfs_abort_transaction(trans, extent_root, ret);
5330 btrfs_free_path(path);
5335 * when we free an block, it is possible (and likely) that we free the last
5336 * delayed ref for that extent as well. This searches the delayed ref tree for
5337 * a given extent, and if there are no other delayed refs to be processed, it
5338 * removes it from the tree.
5340 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5341 struct btrfs_root *root, u64 bytenr)
5343 struct btrfs_delayed_ref_head *head;
5344 struct btrfs_delayed_ref_root *delayed_refs;
5345 struct btrfs_delayed_ref_node *ref;
5346 struct rb_node *node;
5349 delayed_refs = &trans->transaction->delayed_refs;
5350 spin_lock(&delayed_refs->lock);
5351 head = btrfs_find_delayed_ref_head(trans, bytenr);
5355 node = rb_prev(&head->node.rb_node);
5359 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5361 /* there are still entries for this ref, we can't drop it */
5362 if (ref->bytenr == bytenr)
5365 if (head->extent_op) {
5366 if (!head->must_insert_reserved)
5368 btrfs_free_delayed_extent_op(head->extent_op);
5369 head->extent_op = NULL;
5373 * waiting for the lock here would deadlock. If someone else has it
5374 * locked they are already in the process of dropping it anyway
5376 if (!mutex_trylock(&head->mutex))
5380 * at this point we have a head with no other entries. Go
5381 * ahead and process it.
5383 head->node.in_tree = 0;
5384 rb_erase(&head->node.rb_node, &delayed_refs->root);
5386 delayed_refs->num_entries--;
5389 * we don't take a ref on the node because we're removing it from the
5390 * tree, so we just steal the ref the tree was holding.
5392 delayed_refs->num_heads--;
5393 if (list_empty(&head->cluster))
5394 delayed_refs->num_heads_ready--;
5396 list_del_init(&head->cluster);
5397 spin_unlock(&delayed_refs->lock);
5399 BUG_ON(head->extent_op);
5400 if (head->must_insert_reserved)
5403 mutex_unlock(&head->mutex);
5404 btrfs_put_delayed_ref(&head->node);
5407 spin_unlock(&delayed_refs->lock);
5411 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5412 struct btrfs_root *root,
5413 struct extent_buffer *buf,
5414 u64 parent, int last_ref)
5416 struct btrfs_block_group_cache *cache = NULL;
5419 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5420 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5421 buf->start, buf->len,
5422 parent, root->root_key.objectid,
5423 btrfs_header_level(buf),
5424 BTRFS_DROP_DELAYED_REF, NULL, 0);
5425 BUG_ON(ret); /* -ENOMEM */
5431 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5433 if (btrfs_header_generation(buf) == trans->transid) {
5434 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5435 ret = check_ref_cleanup(trans, root, buf->start);
5440 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5441 pin_down_extent(root, cache, buf->start, buf->len, 1);
5445 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5447 btrfs_add_free_space(cache, buf->start, buf->len);
5448 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5452 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5455 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5456 btrfs_put_block_group(cache);
5459 /* Can return -ENOMEM */
5460 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5461 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5462 u64 owner, u64 offset, int for_cow)
5465 struct btrfs_fs_info *fs_info = root->fs_info;
5468 * tree log blocks never actually go into the extent allocation
5469 * tree, just update pinning info and exit early.
5471 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5472 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5473 /* unlocks the pinned mutex */
5474 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5476 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5477 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5479 parent, root_objectid, (int)owner,
5480 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5482 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5484 parent, root_objectid, owner,
5485 offset, BTRFS_DROP_DELAYED_REF,
5491 static u64 stripe_align(struct btrfs_root *root, u64 val)
5493 u64 mask = ((u64)root->stripesize - 1);
5494 u64 ret = (val + mask) & ~mask;
5499 * when we wait for progress in the block group caching, its because
5500 * our allocation attempt failed at least once. So, we must sleep
5501 * and let some progress happen before we try again.
5503 * This function will sleep at least once waiting for new free space to
5504 * show up, and then it will check the block group free space numbers
5505 * for our min num_bytes. Another option is to have it go ahead
5506 * and look in the rbtree for a free extent of a given size, but this
5510 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5513 struct btrfs_caching_control *caching_ctl;
5516 caching_ctl = get_caching_control(cache);
5520 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5521 (cache->free_space_ctl->free_space >= num_bytes));
5523 put_caching_control(caching_ctl);
5528 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5530 struct btrfs_caching_control *caching_ctl;
5533 caching_ctl = get_caching_control(cache);
5537 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5539 put_caching_control(caching_ctl);
5543 int __get_raid_index(u64 flags)
5547 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5549 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5551 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5553 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5561 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5563 return __get_raid_index(cache->flags);
5566 enum btrfs_loop_type {
5567 LOOP_CACHING_NOWAIT = 0,
5568 LOOP_CACHING_WAIT = 1,
5569 LOOP_ALLOC_CHUNK = 2,
5570 LOOP_NO_EMPTY_SIZE = 3,
5574 * walks the btree of allocated extents and find a hole of a given size.
5575 * The key ins is changed to record the hole:
5576 * ins->objectid == block start
5577 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5578 * ins->offset == number of blocks
5579 * Any available blocks before search_start are skipped.
5581 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5582 struct btrfs_root *orig_root,
5583 u64 num_bytes, u64 empty_size,
5584 u64 hint_byte, struct btrfs_key *ins,
5588 struct btrfs_root *root = orig_root->fs_info->extent_root;
5589 struct btrfs_free_cluster *last_ptr = NULL;
5590 struct btrfs_block_group_cache *block_group = NULL;
5591 struct btrfs_block_group_cache *used_block_group;
5592 u64 search_start = 0;
5593 int empty_cluster = 2 * 1024 * 1024;
5594 struct btrfs_space_info *space_info;
5596 int index = __get_raid_index(data);
5597 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5598 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5599 bool found_uncached_bg = false;
5600 bool failed_cluster_refill = false;
5601 bool failed_alloc = false;
5602 bool use_cluster = true;
5603 bool have_caching_bg = false;
5605 WARN_ON(num_bytes < root->sectorsize);
5606 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5610 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5612 space_info = __find_space_info(root->fs_info, data);
5614 printk(KERN_ERR "No space info for %llu\n", data);
5619 * If the space info is for both data and metadata it means we have a
5620 * small filesystem and we can't use the clustering stuff.
5622 if (btrfs_mixed_space_info(space_info))
5623 use_cluster = false;
5625 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5626 last_ptr = &root->fs_info->meta_alloc_cluster;
5627 if (!btrfs_test_opt(root, SSD))
5628 empty_cluster = 64 * 1024;
5631 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5632 btrfs_test_opt(root, SSD)) {
5633 last_ptr = &root->fs_info->data_alloc_cluster;
5637 spin_lock(&last_ptr->lock);
5638 if (last_ptr->block_group)
5639 hint_byte = last_ptr->window_start;
5640 spin_unlock(&last_ptr->lock);
5643 search_start = max(search_start, first_logical_byte(root, 0));
5644 search_start = max(search_start, hint_byte);
5649 if (search_start == hint_byte) {
5650 block_group = btrfs_lookup_block_group(root->fs_info,
5652 used_block_group = block_group;
5654 * we don't want to use the block group if it doesn't match our
5655 * allocation bits, or if its not cached.
5657 * However if we are re-searching with an ideal block group
5658 * picked out then we don't care that the block group is cached.
5660 if (block_group && block_group_bits(block_group, data) &&
5661 block_group->cached != BTRFS_CACHE_NO) {
5662 down_read(&space_info->groups_sem);
5663 if (list_empty(&block_group->list) ||
5666 * someone is removing this block group,
5667 * we can't jump into the have_block_group
5668 * target because our list pointers are not
5671 btrfs_put_block_group(block_group);
5672 up_read(&space_info->groups_sem);
5674 index = get_block_group_index(block_group);
5675 goto have_block_group;
5677 } else if (block_group) {
5678 btrfs_put_block_group(block_group);
5682 have_caching_bg = false;
5683 down_read(&space_info->groups_sem);
5684 list_for_each_entry(block_group, &space_info->block_groups[index],
5689 used_block_group = block_group;
5690 btrfs_get_block_group(block_group);
5691 search_start = block_group->key.objectid;
5694 * this can happen if we end up cycling through all the
5695 * raid types, but we want to make sure we only allocate
5696 * for the proper type.
5698 if (!block_group_bits(block_group, data)) {
5699 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5700 BTRFS_BLOCK_GROUP_RAID1 |
5701 BTRFS_BLOCK_GROUP_RAID10;
5704 * if they asked for extra copies and this block group
5705 * doesn't provide them, bail. This does allow us to
5706 * fill raid0 from raid1.
5708 if ((data & extra) && !(block_group->flags & extra))
5713 cached = block_group_cache_done(block_group);
5714 if (unlikely(!cached)) {
5715 found_uncached_bg = true;
5716 ret = cache_block_group(block_group, trans,
5722 if (unlikely(block_group->ro))
5726 * Ok we want to try and use the cluster allocator, so
5731 * the refill lock keeps out other
5732 * people trying to start a new cluster
5734 spin_lock(&last_ptr->refill_lock);
5735 used_block_group = last_ptr->block_group;
5736 if (used_block_group != block_group &&
5737 (!used_block_group ||
5738 used_block_group->ro ||
5739 !block_group_bits(used_block_group, data))) {
5740 used_block_group = block_group;
5741 goto refill_cluster;
5744 if (used_block_group != block_group)
5745 btrfs_get_block_group(used_block_group);
5747 offset = btrfs_alloc_from_cluster(used_block_group,
5748 last_ptr, num_bytes, used_block_group->key.objectid);
5750 /* we have a block, we're done */
5751 spin_unlock(&last_ptr->refill_lock);
5752 trace_btrfs_reserve_extent_cluster(root,
5753 block_group, search_start, num_bytes);
5757 WARN_ON(last_ptr->block_group != used_block_group);
5758 if (used_block_group != block_group) {
5759 btrfs_put_block_group(used_block_group);
5760 used_block_group = block_group;
5763 BUG_ON(used_block_group != block_group);
5764 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5765 * set up a new clusters, so lets just skip it
5766 * and let the allocator find whatever block
5767 * it can find. If we reach this point, we
5768 * will have tried the cluster allocator
5769 * plenty of times and not have found
5770 * anything, so we are likely way too
5771 * fragmented for the clustering stuff to find
5774 * However, if the cluster is taken from the
5775 * current block group, release the cluster
5776 * first, so that we stand a better chance of
5777 * succeeding in the unclustered
5779 if (loop >= LOOP_NO_EMPTY_SIZE &&
5780 last_ptr->block_group != block_group) {
5781 spin_unlock(&last_ptr->refill_lock);
5782 goto unclustered_alloc;
5786 * this cluster didn't work out, free it and
5789 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5791 if (loop >= LOOP_NO_EMPTY_SIZE) {
5792 spin_unlock(&last_ptr->refill_lock);
5793 goto unclustered_alloc;
5796 /* allocate a cluster in this block group */
5797 ret = btrfs_find_space_cluster(trans, root,
5798 block_group, last_ptr,
5799 search_start, num_bytes,
5800 empty_cluster + empty_size);
5803 * now pull our allocation out of this
5806 offset = btrfs_alloc_from_cluster(block_group,
5807 last_ptr, num_bytes,
5810 /* we found one, proceed */
5811 spin_unlock(&last_ptr->refill_lock);
5812 trace_btrfs_reserve_extent_cluster(root,
5813 block_group, search_start,
5817 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5818 && !failed_cluster_refill) {
5819 spin_unlock(&last_ptr->refill_lock);
5821 failed_cluster_refill = true;
5822 wait_block_group_cache_progress(block_group,
5823 num_bytes + empty_cluster + empty_size);
5824 goto have_block_group;
5828 * at this point we either didn't find a cluster
5829 * or we weren't able to allocate a block from our
5830 * cluster. Free the cluster we've been trying
5831 * to use, and go to the next block group
5833 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5834 spin_unlock(&last_ptr->refill_lock);
5839 spin_lock(&block_group->free_space_ctl->tree_lock);
5841 block_group->free_space_ctl->free_space <
5842 num_bytes + empty_cluster + empty_size) {
5843 spin_unlock(&block_group->free_space_ctl->tree_lock);
5846 spin_unlock(&block_group->free_space_ctl->tree_lock);
5848 offset = btrfs_find_space_for_alloc(block_group, search_start,
5849 num_bytes, empty_size);
5851 * If we didn't find a chunk, and we haven't failed on this
5852 * block group before, and this block group is in the middle of
5853 * caching and we are ok with waiting, then go ahead and wait
5854 * for progress to be made, and set failed_alloc to true.
5856 * If failed_alloc is true then we've already waited on this
5857 * block group once and should move on to the next block group.
5859 if (!offset && !failed_alloc && !cached &&
5860 loop > LOOP_CACHING_NOWAIT) {
5861 wait_block_group_cache_progress(block_group,
5862 num_bytes + empty_size);
5863 failed_alloc = true;
5864 goto have_block_group;
5865 } else if (!offset) {
5867 have_caching_bg = true;
5871 search_start = stripe_align(root, offset);
5873 /* move on to the next group */
5874 if (search_start + num_bytes >
5875 used_block_group->key.objectid + used_block_group->key.offset) {
5876 btrfs_add_free_space(used_block_group, offset, num_bytes);
5880 if (offset < search_start)
5881 btrfs_add_free_space(used_block_group, offset,
5882 search_start - offset);
5883 BUG_ON(offset > search_start);
5885 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5887 if (ret == -EAGAIN) {
5888 btrfs_add_free_space(used_block_group, offset, num_bytes);
5892 /* we are all good, lets return */
5893 ins->objectid = search_start;
5894 ins->offset = num_bytes;
5896 trace_btrfs_reserve_extent(orig_root, block_group,
5897 search_start, num_bytes);
5898 if (used_block_group != block_group)
5899 btrfs_put_block_group(used_block_group);
5900 btrfs_put_block_group(block_group);
5903 failed_cluster_refill = false;
5904 failed_alloc = false;
5905 BUG_ON(index != get_block_group_index(block_group));
5906 if (used_block_group != block_group)
5907 btrfs_put_block_group(used_block_group);
5908 btrfs_put_block_group(block_group);
5910 up_read(&space_info->groups_sem);
5912 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5915 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5919 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5920 * caching kthreads as we move along
5921 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5922 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5923 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5926 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5929 if (loop == LOOP_ALLOC_CHUNK) {
5930 ret = do_chunk_alloc(trans, root, data,
5933 * Do not bail out on ENOSPC since we
5934 * can do more things.
5936 if (ret < 0 && ret != -ENOSPC) {
5937 btrfs_abort_transaction(trans,
5943 if (loop == LOOP_NO_EMPTY_SIZE) {
5949 } else if (!ins->objectid) {
5951 } else if (ins->objectid) {
5959 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5960 int dump_block_groups)
5962 struct btrfs_block_group_cache *cache;
5965 spin_lock(&info->lock);
5966 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5967 (unsigned long long)info->flags,
5968 (unsigned long long)(info->total_bytes - info->bytes_used -
5969 info->bytes_pinned - info->bytes_reserved -
5970 info->bytes_readonly),
5971 (info->full) ? "" : "not ");
5972 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5973 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5974 (unsigned long long)info->total_bytes,
5975 (unsigned long long)info->bytes_used,
5976 (unsigned long long)info->bytes_pinned,
5977 (unsigned long long)info->bytes_reserved,
5978 (unsigned long long)info->bytes_may_use,
5979 (unsigned long long)info->bytes_readonly);
5980 spin_unlock(&info->lock);
5982 if (!dump_block_groups)
5985 down_read(&info->groups_sem);
5987 list_for_each_entry(cache, &info->block_groups[index], list) {
5988 spin_lock(&cache->lock);
5989 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5990 (unsigned long long)cache->key.objectid,
5991 (unsigned long long)cache->key.offset,
5992 (unsigned long long)btrfs_block_group_used(&cache->item),
5993 (unsigned long long)cache->pinned,
5994 (unsigned long long)cache->reserved,
5995 cache->ro ? "[readonly]" : "");
5996 btrfs_dump_free_space(cache, bytes);
5997 spin_unlock(&cache->lock);
5999 if (++index < BTRFS_NR_RAID_TYPES)
6001 up_read(&info->groups_sem);
6004 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6005 struct btrfs_root *root,
6006 u64 num_bytes, u64 min_alloc_size,
6007 u64 empty_size, u64 hint_byte,
6008 struct btrfs_key *ins, u64 data)
6010 bool final_tried = false;
6013 data = btrfs_get_alloc_profile(root, data);
6015 WARN_ON(num_bytes < root->sectorsize);
6016 ret = find_free_extent(trans, root, num_bytes, empty_size,
6017 hint_byte, ins, data);
6019 if (ret == -ENOSPC) {
6021 num_bytes = num_bytes >> 1;
6022 num_bytes = num_bytes & ~(root->sectorsize - 1);
6023 num_bytes = max(num_bytes, min_alloc_size);
6024 if (num_bytes == min_alloc_size)
6027 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6028 struct btrfs_space_info *sinfo;
6030 sinfo = __find_space_info(root->fs_info, data);
6031 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6032 "wanted %llu\n", (unsigned long long)data,
6033 (unsigned long long)num_bytes);
6035 dump_space_info(sinfo, num_bytes, 1);
6039 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6044 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6045 u64 start, u64 len, int pin)
6047 struct btrfs_block_group_cache *cache;
6050 cache = btrfs_lookup_block_group(root->fs_info, start);
6052 printk(KERN_ERR "Unable to find block group for %llu\n",
6053 (unsigned long long)start);
6057 if (btrfs_test_opt(root, DISCARD))
6058 ret = btrfs_discard_extent(root, start, len, NULL);
6061 pin_down_extent(root, cache, start, len, 1);
6063 btrfs_add_free_space(cache, start, len);
6064 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6066 btrfs_put_block_group(cache);
6068 trace_btrfs_reserved_extent_free(root, start, len);
6073 int btrfs_free_reserved_extent(struct btrfs_root *root,
6076 return __btrfs_free_reserved_extent(root, start, len, 0);
6079 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6082 return __btrfs_free_reserved_extent(root, start, len, 1);
6085 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6086 struct btrfs_root *root,
6087 u64 parent, u64 root_objectid,
6088 u64 flags, u64 owner, u64 offset,
6089 struct btrfs_key *ins, int ref_mod)
6092 struct btrfs_fs_info *fs_info = root->fs_info;
6093 struct btrfs_extent_item *extent_item;
6094 struct btrfs_extent_inline_ref *iref;
6095 struct btrfs_path *path;
6096 struct extent_buffer *leaf;
6101 type = BTRFS_SHARED_DATA_REF_KEY;
6103 type = BTRFS_EXTENT_DATA_REF_KEY;
6105 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6107 path = btrfs_alloc_path();
6111 path->leave_spinning = 1;
6112 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6115 btrfs_free_path(path);
6119 leaf = path->nodes[0];
6120 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6121 struct btrfs_extent_item);
6122 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6123 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6124 btrfs_set_extent_flags(leaf, extent_item,
6125 flags | BTRFS_EXTENT_FLAG_DATA);
6127 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6128 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6130 struct btrfs_shared_data_ref *ref;
6131 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6132 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6133 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6135 struct btrfs_extent_data_ref *ref;
6136 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6137 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6138 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6139 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6140 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6143 btrfs_mark_buffer_dirty(path->nodes[0]);
6144 btrfs_free_path(path);
6146 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6147 if (ret) { /* -ENOENT, logic error */
6148 printk(KERN_ERR "btrfs update block group failed for %llu "
6149 "%llu\n", (unsigned long long)ins->objectid,
6150 (unsigned long long)ins->offset);
6156 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6157 struct btrfs_root *root,
6158 u64 parent, u64 root_objectid,
6159 u64 flags, struct btrfs_disk_key *key,
6160 int level, struct btrfs_key *ins)
6163 struct btrfs_fs_info *fs_info = root->fs_info;
6164 struct btrfs_extent_item *extent_item;
6165 struct btrfs_tree_block_info *block_info;
6166 struct btrfs_extent_inline_ref *iref;
6167 struct btrfs_path *path;
6168 struct extent_buffer *leaf;
6169 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6171 path = btrfs_alloc_path();
6175 path->leave_spinning = 1;
6176 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6179 btrfs_free_path(path);
6183 leaf = path->nodes[0];
6184 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6185 struct btrfs_extent_item);
6186 btrfs_set_extent_refs(leaf, extent_item, 1);
6187 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6188 btrfs_set_extent_flags(leaf, extent_item,
6189 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6190 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6192 btrfs_set_tree_block_key(leaf, block_info, key);
6193 btrfs_set_tree_block_level(leaf, block_info, level);
6195 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6197 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6198 btrfs_set_extent_inline_ref_type(leaf, iref,
6199 BTRFS_SHARED_BLOCK_REF_KEY);
6200 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6202 btrfs_set_extent_inline_ref_type(leaf, iref,
6203 BTRFS_TREE_BLOCK_REF_KEY);
6204 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6207 btrfs_mark_buffer_dirty(leaf);
6208 btrfs_free_path(path);
6210 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6211 if (ret) { /* -ENOENT, logic error */
6212 printk(KERN_ERR "btrfs update block group failed for %llu "
6213 "%llu\n", (unsigned long long)ins->objectid,
6214 (unsigned long long)ins->offset);
6220 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6221 struct btrfs_root *root,
6222 u64 root_objectid, u64 owner,
6223 u64 offset, struct btrfs_key *ins)
6227 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6229 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6231 root_objectid, owner, offset,
6232 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6237 * this is used by the tree logging recovery code. It records that
6238 * an extent has been allocated and makes sure to clear the free
6239 * space cache bits as well
6241 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6242 struct btrfs_root *root,
6243 u64 root_objectid, u64 owner, u64 offset,
6244 struct btrfs_key *ins)
6247 struct btrfs_block_group_cache *block_group;
6248 struct btrfs_caching_control *caching_ctl;
6249 u64 start = ins->objectid;
6250 u64 num_bytes = ins->offset;
6252 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6253 cache_block_group(block_group, trans, NULL, 0);
6254 caching_ctl = get_caching_control(block_group);
6257 BUG_ON(!block_group_cache_done(block_group));
6258 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6259 BUG_ON(ret); /* -ENOMEM */
6261 mutex_lock(&caching_ctl->mutex);
6263 if (start >= caching_ctl->progress) {
6264 ret = add_excluded_extent(root, start, num_bytes);
6265 BUG_ON(ret); /* -ENOMEM */
6266 } else if (start + num_bytes <= caching_ctl->progress) {
6267 ret = btrfs_remove_free_space(block_group,
6269 BUG_ON(ret); /* -ENOMEM */
6271 num_bytes = caching_ctl->progress - start;
6272 ret = btrfs_remove_free_space(block_group,
6274 BUG_ON(ret); /* -ENOMEM */
6276 start = caching_ctl->progress;
6277 num_bytes = ins->objectid + ins->offset -
6278 caching_ctl->progress;
6279 ret = add_excluded_extent(root, start, num_bytes);
6280 BUG_ON(ret); /* -ENOMEM */
6283 mutex_unlock(&caching_ctl->mutex);
6284 put_caching_control(caching_ctl);
6287 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6288 RESERVE_ALLOC_NO_ACCOUNT);
6289 BUG_ON(ret); /* logic error */
6290 btrfs_put_block_group(block_group);
6291 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6292 0, owner, offset, ins, 1);
6296 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6297 struct btrfs_root *root,
6298 u64 bytenr, u32 blocksize,
6301 struct extent_buffer *buf;
6303 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6305 return ERR_PTR(-ENOMEM);
6306 btrfs_set_header_generation(buf, trans->transid);
6307 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6308 btrfs_tree_lock(buf);
6309 clean_tree_block(trans, root, buf);
6310 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6312 btrfs_set_lock_blocking(buf);
6313 btrfs_set_buffer_uptodate(buf);
6315 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6317 * we allow two log transactions at a time, use different
6318 * EXENT bit to differentiate dirty pages.
6320 if (root->log_transid % 2 == 0)
6321 set_extent_dirty(&root->dirty_log_pages, buf->start,
6322 buf->start + buf->len - 1, GFP_NOFS);
6324 set_extent_new(&root->dirty_log_pages, buf->start,
6325 buf->start + buf->len - 1, GFP_NOFS);
6327 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6328 buf->start + buf->len - 1, GFP_NOFS);
6330 trans->blocks_used++;
6331 /* this returns a buffer locked for blocking */
6335 static struct btrfs_block_rsv *
6336 use_block_rsv(struct btrfs_trans_handle *trans,
6337 struct btrfs_root *root, u32 blocksize)
6339 struct btrfs_block_rsv *block_rsv;
6340 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6343 block_rsv = get_block_rsv(trans, root);
6345 if (block_rsv->size == 0) {
6346 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6347 BTRFS_RESERVE_NO_FLUSH);
6349 * If we couldn't reserve metadata bytes try and use some from
6350 * the global reserve.
6352 if (ret && block_rsv != global_rsv) {
6353 ret = block_rsv_use_bytes(global_rsv, blocksize);
6356 return ERR_PTR(ret);
6358 return ERR_PTR(ret);
6363 ret = block_rsv_use_bytes(block_rsv, blocksize);
6366 if (ret && !block_rsv->failfast) {
6367 static DEFINE_RATELIMIT_STATE(_rs,
6368 DEFAULT_RATELIMIT_INTERVAL,
6369 /*DEFAULT_RATELIMIT_BURST*/ 2);
6370 if (__ratelimit(&_rs))
6371 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6373 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6374 BTRFS_RESERVE_NO_FLUSH);
6377 } else if (ret && block_rsv != global_rsv) {
6378 ret = block_rsv_use_bytes(global_rsv, blocksize);
6384 return ERR_PTR(-ENOSPC);
6387 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6388 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6390 block_rsv_add_bytes(block_rsv, blocksize, 0);
6391 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6395 * finds a free extent and does all the dirty work required for allocation
6396 * returns the key for the extent through ins, and a tree buffer for
6397 * the first block of the extent through buf.
6399 * returns the tree buffer or NULL.
6401 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6402 struct btrfs_root *root, u32 blocksize,
6403 u64 parent, u64 root_objectid,
6404 struct btrfs_disk_key *key, int level,
6405 u64 hint, u64 empty_size)
6407 struct btrfs_key ins;
6408 struct btrfs_block_rsv *block_rsv;
6409 struct extent_buffer *buf;
6414 block_rsv = use_block_rsv(trans, root, blocksize);
6415 if (IS_ERR(block_rsv))
6416 return ERR_CAST(block_rsv);
6418 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6419 empty_size, hint, &ins, 0);
6421 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6422 return ERR_PTR(ret);
6425 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6427 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6429 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6431 parent = ins.objectid;
6432 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6436 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6437 struct btrfs_delayed_extent_op *extent_op;
6438 extent_op = btrfs_alloc_delayed_extent_op();
6439 BUG_ON(!extent_op); /* -ENOMEM */
6441 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6443 memset(&extent_op->key, 0, sizeof(extent_op->key));
6444 extent_op->flags_to_set = flags;
6445 extent_op->update_key = 1;
6446 extent_op->update_flags = 1;
6447 extent_op->is_data = 0;
6449 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6451 ins.offset, parent, root_objectid,
6452 level, BTRFS_ADD_DELAYED_EXTENT,
6454 BUG_ON(ret); /* -ENOMEM */
6459 struct walk_control {
6460 u64 refs[BTRFS_MAX_LEVEL];
6461 u64 flags[BTRFS_MAX_LEVEL];
6462 struct btrfs_key update_progress;
6473 #define DROP_REFERENCE 1
6474 #define UPDATE_BACKREF 2
6476 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6477 struct btrfs_root *root,
6478 struct walk_control *wc,
6479 struct btrfs_path *path)
6487 struct btrfs_key key;
6488 struct extent_buffer *eb;
6493 if (path->slots[wc->level] < wc->reada_slot) {
6494 wc->reada_count = wc->reada_count * 2 / 3;
6495 wc->reada_count = max(wc->reada_count, 2);
6497 wc->reada_count = wc->reada_count * 3 / 2;
6498 wc->reada_count = min_t(int, wc->reada_count,
6499 BTRFS_NODEPTRS_PER_BLOCK(root));
6502 eb = path->nodes[wc->level];
6503 nritems = btrfs_header_nritems(eb);
6504 blocksize = btrfs_level_size(root, wc->level - 1);
6506 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6507 if (nread >= wc->reada_count)
6511 bytenr = btrfs_node_blockptr(eb, slot);
6512 generation = btrfs_node_ptr_generation(eb, slot);
6514 if (slot == path->slots[wc->level])
6517 if (wc->stage == UPDATE_BACKREF &&
6518 generation <= root->root_key.offset)
6521 /* We don't lock the tree block, it's OK to be racy here */
6522 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6524 /* We don't care about errors in readahead. */
6529 if (wc->stage == DROP_REFERENCE) {
6533 if (wc->level == 1 &&
6534 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6536 if (!wc->update_ref ||
6537 generation <= root->root_key.offset)
6539 btrfs_node_key_to_cpu(eb, &key, slot);
6540 ret = btrfs_comp_cpu_keys(&key,
6541 &wc->update_progress);
6545 if (wc->level == 1 &&
6546 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6550 ret = readahead_tree_block(root, bytenr, blocksize,
6556 wc->reada_slot = slot;
6560 * hepler to process tree block while walking down the tree.
6562 * when wc->stage == UPDATE_BACKREF, this function updates
6563 * back refs for pointers in the block.
6565 * NOTE: return value 1 means we should stop walking down.
6567 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6568 struct btrfs_root *root,
6569 struct btrfs_path *path,
6570 struct walk_control *wc, int lookup_info)
6572 int level = wc->level;
6573 struct extent_buffer *eb = path->nodes[level];
6574 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6577 if (wc->stage == UPDATE_BACKREF &&
6578 btrfs_header_owner(eb) != root->root_key.objectid)
6582 * when reference count of tree block is 1, it won't increase
6583 * again. once full backref flag is set, we never clear it.
6586 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6587 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6588 BUG_ON(!path->locks[level]);
6589 ret = btrfs_lookup_extent_info(trans, root,
6593 BUG_ON(ret == -ENOMEM);
6596 BUG_ON(wc->refs[level] == 0);
6599 if (wc->stage == DROP_REFERENCE) {
6600 if (wc->refs[level] > 1)
6603 if (path->locks[level] && !wc->keep_locks) {
6604 btrfs_tree_unlock_rw(eb, path->locks[level]);
6605 path->locks[level] = 0;
6610 /* wc->stage == UPDATE_BACKREF */
6611 if (!(wc->flags[level] & flag)) {
6612 BUG_ON(!path->locks[level]);
6613 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6614 BUG_ON(ret); /* -ENOMEM */
6615 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6616 BUG_ON(ret); /* -ENOMEM */
6617 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6619 BUG_ON(ret); /* -ENOMEM */
6620 wc->flags[level] |= flag;
6624 * the block is shared by multiple trees, so it's not good to
6625 * keep the tree lock
6627 if (path->locks[level] && level > 0) {
6628 btrfs_tree_unlock_rw(eb, path->locks[level]);
6629 path->locks[level] = 0;
6635 * hepler to process tree block pointer.
6637 * when wc->stage == DROP_REFERENCE, this function checks
6638 * reference count of the block pointed to. if the block
6639 * is shared and we need update back refs for the subtree
6640 * rooted at the block, this function changes wc->stage to
6641 * UPDATE_BACKREF. if the block is shared and there is no
6642 * need to update back, this function drops the reference
6645 * NOTE: return value 1 means we should stop walking down.
6647 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6648 struct btrfs_root *root,
6649 struct btrfs_path *path,
6650 struct walk_control *wc, int *lookup_info)
6656 struct btrfs_key key;
6657 struct extent_buffer *next;
6658 int level = wc->level;
6662 generation = btrfs_node_ptr_generation(path->nodes[level],
6663 path->slots[level]);
6665 * if the lower level block was created before the snapshot
6666 * was created, we know there is no need to update back refs
6669 if (wc->stage == UPDATE_BACKREF &&
6670 generation <= root->root_key.offset) {
6675 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6676 blocksize = btrfs_level_size(root, level - 1);
6678 next = btrfs_find_tree_block(root, bytenr, blocksize);
6680 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6685 btrfs_tree_lock(next);
6686 btrfs_set_lock_blocking(next);
6688 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6689 &wc->refs[level - 1],
6690 &wc->flags[level - 1]);
6692 btrfs_tree_unlock(next);
6696 BUG_ON(wc->refs[level - 1] == 0);
6699 if (wc->stage == DROP_REFERENCE) {
6700 if (wc->refs[level - 1] > 1) {
6702 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6705 if (!wc->update_ref ||
6706 generation <= root->root_key.offset)
6709 btrfs_node_key_to_cpu(path->nodes[level], &key,
6710 path->slots[level]);
6711 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6715 wc->stage = UPDATE_BACKREF;
6716 wc->shared_level = level - 1;
6720 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6724 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6725 btrfs_tree_unlock(next);
6726 free_extent_buffer(next);
6732 if (reada && level == 1)
6733 reada_walk_down(trans, root, wc, path);
6734 next = read_tree_block(root, bytenr, blocksize, generation);
6737 btrfs_tree_lock(next);
6738 btrfs_set_lock_blocking(next);
6742 BUG_ON(level != btrfs_header_level(next));
6743 path->nodes[level] = next;
6744 path->slots[level] = 0;
6745 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6751 wc->refs[level - 1] = 0;
6752 wc->flags[level - 1] = 0;
6753 if (wc->stage == DROP_REFERENCE) {
6754 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6755 parent = path->nodes[level]->start;
6757 BUG_ON(root->root_key.objectid !=
6758 btrfs_header_owner(path->nodes[level]));
6762 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6763 root->root_key.objectid, level - 1, 0, 0);
6764 BUG_ON(ret); /* -ENOMEM */
6766 btrfs_tree_unlock(next);
6767 free_extent_buffer(next);
6773 * hepler to process tree block while walking up the tree.
6775 * when wc->stage == DROP_REFERENCE, this function drops
6776 * reference count on the block.
6778 * when wc->stage == UPDATE_BACKREF, this function changes
6779 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6780 * to UPDATE_BACKREF previously while processing the block.
6782 * NOTE: return value 1 means we should stop walking up.
6784 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6785 struct btrfs_root *root,
6786 struct btrfs_path *path,
6787 struct walk_control *wc)
6790 int level = wc->level;
6791 struct extent_buffer *eb = path->nodes[level];
6794 if (wc->stage == UPDATE_BACKREF) {
6795 BUG_ON(wc->shared_level < level);
6796 if (level < wc->shared_level)
6799 ret = find_next_key(path, level + 1, &wc->update_progress);
6803 wc->stage = DROP_REFERENCE;
6804 wc->shared_level = -1;
6805 path->slots[level] = 0;
6808 * check reference count again if the block isn't locked.
6809 * we should start walking down the tree again if reference
6812 if (!path->locks[level]) {
6814 btrfs_tree_lock(eb);
6815 btrfs_set_lock_blocking(eb);
6816 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6818 ret = btrfs_lookup_extent_info(trans, root,
6823 btrfs_tree_unlock_rw(eb, path->locks[level]);
6824 path->locks[level] = 0;
6827 BUG_ON(wc->refs[level] == 0);
6828 if (wc->refs[level] == 1) {
6829 btrfs_tree_unlock_rw(eb, path->locks[level]);
6830 path->locks[level] = 0;
6836 /* wc->stage == DROP_REFERENCE */
6837 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6839 if (wc->refs[level] == 1) {
6841 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6842 ret = btrfs_dec_ref(trans, root, eb, 1,
6845 ret = btrfs_dec_ref(trans, root, eb, 0,
6847 BUG_ON(ret); /* -ENOMEM */
6849 /* make block locked assertion in clean_tree_block happy */
6850 if (!path->locks[level] &&
6851 btrfs_header_generation(eb) == trans->transid) {
6852 btrfs_tree_lock(eb);
6853 btrfs_set_lock_blocking(eb);
6854 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6856 clean_tree_block(trans, root, eb);
6859 if (eb == root->node) {
6860 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6863 BUG_ON(root->root_key.objectid !=
6864 btrfs_header_owner(eb));
6866 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6867 parent = path->nodes[level + 1]->start;
6869 BUG_ON(root->root_key.objectid !=
6870 btrfs_header_owner(path->nodes[level + 1]));
6873 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6875 wc->refs[level] = 0;
6876 wc->flags[level] = 0;
6880 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6881 struct btrfs_root *root,
6882 struct btrfs_path *path,
6883 struct walk_control *wc)
6885 int level = wc->level;
6886 int lookup_info = 1;
6889 while (level >= 0) {
6890 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6897 if (path->slots[level] >=
6898 btrfs_header_nritems(path->nodes[level]))
6901 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6903 path->slots[level]++;
6912 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6913 struct btrfs_root *root,
6914 struct btrfs_path *path,
6915 struct walk_control *wc, int max_level)
6917 int level = wc->level;
6920 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6921 while (level < max_level && path->nodes[level]) {
6923 if (path->slots[level] + 1 <
6924 btrfs_header_nritems(path->nodes[level])) {
6925 path->slots[level]++;
6928 ret = walk_up_proc(trans, root, path, wc);
6932 if (path->locks[level]) {
6933 btrfs_tree_unlock_rw(path->nodes[level],
6934 path->locks[level]);
6935 path->locks[level] = 0;
6937 free_extent_buffer(path->nodes[level]);
6938 path->nodes[level] = NULL;
6946 * drop a subvolume tree.
6948 * this function traverses the tree freeing any blocks that only
6949 * referenced by the tree.
6951 * when a shared tree block is found. this function decreases its
6952 * reference count by one. if update_ref is true, this function
6953 * also make sure backrefs for the shared block and all lower level
6954 * blocks are properly updated.
6956 int btrfs_drop_snapshot(struct btrfs_root *root,
6957 struct btrfs_block_rsv *block_rsv, int update_ref,
6960 struct btrfs_path *path;
6961 struct btrfs_trans_handle *trans;
6962 struct btrfs_root *tree_root = root->fs_info->tree_root;
6963 struct btrfs_root_item *root_item = &root->root_item;
6964 struct walk_control *wc;
6965 struct btrfs_key key;
6970 path = btrfs_alloc_path();
6976 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6978 btrfs_free_path(path);
6983 trans = btrfs_start_transaction(tree_root, 0);
6984 if (IS_ERR(trans)) {
6985 err = PTR_ERR(trans);
6990 trans->block_rsv = block_rsv;
6992 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6993 level = btrfs_header_level(root->node);
6994 path->nodes[level] = btrfs_lock_root_node(root);
6995 btrfs_set_lock_blocking(path->nodes[level]);
6996 path->slots[level] = 0;
6997 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6998 memset(&wc->update_progress, 0,
6999 sizeof(wc->update_progress));
7001 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7002 memcpy(&wc->update_progress, &key,
7003 sizeof(wc->update_progress));
7005 level = root_item->drop_level;
7007 path->lowest_level = level;
7008 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7009 path->lowest_level = 0;
7017 * unlock our path, this is safe because only this
7018 * function is allowed to delete this snapshot
7020 btrfs_unlock_up_safe(path, 0);
7022 level = btrfs_header_level(root->node);
7024 btrfs_tree_lock(path->nodes[level]);
7025 btrfs_set_lock_blocking(path->nodes[level]);
7027 ret = btrfs_lookup_extent_info(trans, root,
7028 path->nodes[level]->start,
7029 path->nodes[level]->len,
7036 BUG_ON(wc->refs[level] == 0);
7038 if (level == root_item->drop_level)
7041 btrfs_tree_unlock(path->nodes[level]);
7042 WARN_ON(wc->refs[level] != 1);
7048 wc->shared_level = -1;
7049 wc->stage = DROP_REFERENCE;
7050 wc->update_ref = update_ref;
7052 wc->for_reloc = for_reloc;
7053 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7056 ret = walk_down_tree(trans, root, path, wc);
7062 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7069 BUG_ON(wc->stage != DROP_REFERENCE);
7073 if (wc->stage == DROP_REFERENCE) {
7075 btrfs_node_key(path->nodes[level],
7076 &root_item->drop_progress,
7077 path->slots[level]);
7078 root_item->drop_level = level;
7081 BUG_ON(wc->level == 0);
7082 if (btrfs_should_end_transaction(trans, tree_root)) {
7083 ret = btrfs_update_root(trans, tree_root,
7087 btrfs_abort_transaction(trans, tree_root, ret);
7092 btrfs_end_transaction_throttle(trans, tree_root);
7093 trans = btrfs_start_transaction(tree_root, 0);
7094 if (IS_ERR(trans)) {
7095 err = PTR_ERR(trans);
7099 trans->block_rsv = block_rsv;
7102 btrfs_release_path(path);
7106 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7108 btrfs_abort_transaction(trans, tree_root, ret);
7112 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7113 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7116 btrfs_abort_transaction(trans, tree_root, ret);
7119 } else if (ret > 0) {
7120 /* if we fail to delete the orphan item this time
7121 * around, it'll get picked up the next time.
7123 * The most common failure here is just -ENOENT.
7125 btrfs_del_orphan_item(trans, tree_root,
7126 root->root_key.objectid);
7130 if (root->in_radix) {
7131 btrfs_free_fs_root(tree_root->fs_info, root);
7133 free_extent_buffer(root->node);
7134 free_extent_buffer(root->commit_root);
7138 btrfs_end_transaction_throttle(trans, tree_root);
7141 btrfs_free_path(path);
7144 btrfs_std_error(root->fs_info, err);
7149 * drop subtree rooted at tree block 'node'.
7151 * NOTE: this function will unlock and release tree block 'node'
7152 * only used by relocation code
7154 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7155 struct btrfs_root *root,
7156 struct extent_buffer *node,
7157 struct extent_buffer *parent)
7159 struct btrfs_path *path;
7160 struct walk_control *wc;
7166 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7168 path = btrfs_alloc_path();
7172 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7174 btrfs_free_path(path);
7178 btrfs_assert_tree_locked(parent);
7179 parent_level = btrfs_header_level(parent);
7180 extent_buffer_get(parent);
7181 path->nodes[parent_level] = parent;
7182 path->slots[parent_level] = btrfs_header_nritems(parent);
7184 btrfs_assert_tree_locked(node);
7185 level = btrfs_header_level(node);
7186 path->nodes[level] = node;
7187 path->slots[level] = 0;
7188 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7190 wc->refs[parent_level] = 1;
7191 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7193 wc->shared_level = -1;
7194 wc->stage = DROP_REFERENCE;
7198 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7201 wret = walk_down_tree(trans, root, path, wc);
7207 wret = walk_up_tree(trans, root, path, wc, parent_level);
7215 btrfs_free_path(path);
7219 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7225 * if restripe for this chunk_type is on pick target profile and
7226 * return, otherwise do the usual balance
7228 stripped = get_restripe_target(root->fs_info, flags);
7230 return extended_to_chunk(stripped);
7233 * we add in the count of missing devices because we want
7234 * to make sure that any RAID levels on a degraded FS
7235 * continue to be honored.
7237 num_devices = root->fs_info->fs_devices->rw_devices +
7238 root->fs_info->fs_devices->missing_devices;
7240 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7241 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7243 if (num_devices == 1) {
7244 stripped |= BTRFS_BLOCK_GROUP_DUP;
7245 stripped = flags & ~stripped;
7247 /* turn raid0 into single device chunks */
7248 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7251 /* turn mirroring into duplication */
7252 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7253 BTRFS_BLOCK_GROUP_RAID10))
7254 return stripped | BTRFS_BLOCK_GROUP_DUP;
7256 /* they already had raid on here, just return */
7257 if (flags & stripped)
7260 stripped |= BTRFS_BLOCK_GROUP_DUP;
7261 stripped = flags & ~stripped;
7263 /* switch duplicated blocks with raid1 */
7264 if (flags & BTRFS_BLOCK_GROUP_DUP)
7265 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7267 /* this is drive concat, leave it alone */
7273 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7275 struct btrfs_space_info *sinfo = cache->space_info;
7277 u64 min_allocable_bytes;
7282 * We need some metadata space and system metadata space for
7283 * allocating chunks in some corner cases until we force to set
7284 * it to be readonly.
7287 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7289 min_allocable_bytes = 1 * 1024 * 1024;
7291 min_allocable_bytes = 0;
7293 spin_lock(&sinfo->lock);
7294 spin_lock(&cache->lock);
7301 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7302 cache->bytes_super - btrfs_block_group_used(&cache->item);
7304 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7305 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7306 min_allocable_bytes <= sinfo->total_bytes) {
7307 sinfo->bytes_readonly += num_bytes;
7312 spin_unlock(&cache->lock);
7313 spin_unlock(&sinfo->lock);
7317 int btrfs_set_block_group_ro(struct btrfs_root *root,
7318 struct btrfs_block_group_cache *cache)
7321 struct btrfs_trans_handle *trans;
7327 trans = btrfs_join_transaction(root);
7329 return PTR_ERR(trans);
7331 alloc_flags = update_block_group_flags(root, cache->flags);
7332 if (alloc_flags != cache->flags) {
7333 ret = do_chunk_alloc(trans, root, alloc_flags,
7339 ret = set_block_group_ro(cache, 0);
7342 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7343 ret = do_chunk_alloc(trans, root, alloc_flags,
7347 ret = set_block_group_ro(cache, 0);
7349 btrfs_end_transaction(trans, root);
7353 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7354 struct btrfs_root *root, u64 type)
7356 u64 alloc_flags = get_alloc_profile(root, type);
7357 return do_chunk_alloc(trans, root, alloc_flags,
7362 * helper to account the unused space of all the readonly block group in the
7363 * list. takes mirrors into account.
7365 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7367 struct btrfs_block_group_cache *block_group;
7371 list_for_each_entry(block_group, groups_list, list) {
7372 spin_lock(&block_group->lock);
7374 if (!block_group->ro) {
7375 spin_unlock(&block_group->lock);
7379 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7380 BTRFS_BLOCK_GROUP_RAID10 |
7381 BTRFS_BLOCK_GROUP_DUP))
7386 free_bytes += (block_group->key.offset -
7387 btrfs_block_group_used(&block_group->item)) *
7390 spin_unlock(&block_group->lock);
7397 * helper to account the unused space of all the readonly block group in the
7398 * space_info. takes mirrors into account.
7400 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7405 spin_lock(&sinfo->lock);
7407 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7408 if (!list_empty(&sinfo->block_groups[i]))
7409 free_bytes += __btrfs_get_ro_block_group_free_space(
7410 &sinfo->block_groups[i]);
7412 spin_unlock(&sinfo->lock);
7417 void btrfs_set_block_group_rw(struct btrfs_root *root,
7418 struct btrfs_block_group_cache *cache)
7420 struct btrfs_space_info *sinfo = cache->space_info;
7425 spin_lock(&sinfo->lock);
7426 spin_lock(&cache->lock);
7427 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7428 cache->bytes_super - btrfs_block_group_used(&cache->item);
7429 sinfo->bytes_readonly -= num_bytes;
7431 spin_unlock(&cache->lock);
7432 spin_unlock(&sinfo->lock);
7436 * checks to see if its even possible to relocate this block group.
7438 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7439 * ok to go ahead and try.
7441 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7443 struct btrfs_block_group_cache *block_group;
7444 struct btrfs_space_info *space_info;
7445 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7446 struct btrfs_device *device;
7455 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7457 /* odd, couldn't find the block group, leave it alone */
7461 min_free = btrfs_block_group_used(&block_group->item);
7463 /* no bytes used, we're good */
7467 space_info = block_group->space_info;
7468 spin_lock(&space_info->lock);
7470 full = space_info->full;
7473 * if this is the last block group we have in this space, we can't
7474 * relocate it unless we're able to allocate a new chunk below.
7476 * Otherwise, we need to make sure we have room in the space to handle
7477 * all of the extents from this block group. If we can, we're good
7479 if ((space_info->total_bytes != block_group->key.offset) &&
7480 (space_info->bytes_used + space_info->bytes_reserved +
7481 space_info->bytes_pinned + space_info->bytes_readonly +
7482 min_free < space_info->total_bytes)) {
7483 spin_unlock(&space_info->lock);
7486 spin_unlock(&space_info->lock);
7489 * ok we don't have enough space, but maybe we have free space on our
7490 * devices to allocate new chunks for relocation, so loop through our
7491 * alloc devices and guess if we have enough space. if this block
7492 * group is going to be restriped, run checks against the target
7493 * profile instead of the current one.
7505 target = get_restripe_target(root->fs_info, block_group->flags);
7507 index = __get_raid_index(extended_to_chunk(target));
7510 * this is just a balance, so if we were marked as full
7511 * we know there is no space for a new chunk
7516 index = get_block_group_index(block_group);
7523 } else if (index == 1) {
7525 } else if (index == 2) {
7528 } else if (index == 3) {
7529 dev_min = fs_devices->rw_devices;
7530 do_div(min_free, dev_min);
7533 mutex_lock(&root->fs_info->chunk_mutex);
7534 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7538 * check to make sure we can actually find a chunk with enough
7539 * space to fit our block group in.
7541 if (device->total_bytes > device->bytes_used + min_free &&
7542 !device->is_tgtdev_for_dev_replace) {
7543 ret = find_free_dev_extent(device, min_free,
7548 if (dev_nr >= dev_min)
7554 mutex_unlock(&root->fs_info->chunk_mutex);
7556 btrfs_put_block_group(block_group);
7560 static int find_first_block_group(struct btrfs_root *root,
7561 struct btrfs_path *path, struct btrfs_key *key)
7564 struct btrfs_key found_key;
7565 struct extent_buffer *leaf;
7568 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7573 slot = path->slots[0];
7574 leaf = path->nodes[0];
7575 if (slot >= btrfs_header_nritems(leaf)) {
7576 ret = btrfs_next_leaf(root, path);
7583 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7585 if (found_key.objectid >= key->objectid &&
7586 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7596 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7598 struct btrfs_block_group_cache *block_group;
7602 struct inode *inode;
7604 block_group = btrfs_lookup_first_block_group(info, last);
7605 while (block_group) {
7606 spin_lock(&block_group->lock);
7607 if (block_group->iref)
7609 spin_unlock(&block_group->lock);
7610 block_group = next_block_group(info->tree_root,
7620 inode = block_group->inode;
7621 block_group->iref = 0;
7622 block_group->inode = NULL;
7623 spin_unlock(&block_group->lock);
7625 last = block_group->key.objectid + block_group->key.offset;
7626 btrfs_put_block_group(block_group);
7630 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7632 struct btrfs_block_group_cache *block_group;
7633 struct btrfs_space_info *space_info;
7634 struct btrfs_caching_control *caching_ctl;
7637 down_write(&info->extent_commit_sem);
7638 while (!list_empty(&info->caching_block_groups)) {
7639 caching_ctl = list_entry(info->caching_block_groups.next,
7640 struct btrfs_caching_control, list);
7641 list_del(&caching_ctl->list);
7642 put_caching_control(caching_ctl);
7644 up_write(&info->extent_commit_sem);
7646 spin_lock(&info->block_group_cache_lock);
7647 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7648 block_group = rb_entry(n, struct btrfs_block_group_cache,
7650 rb_erase(&block_group->cache_node,
7651 &info->block_group_cache_tree);
7652 spin_unlock(&info->block_group_cache_lock);
7654 down_write(&block_group->space_info->groups_sem);
7655 list_del(&block_group->list);
7656 up_write(&block_group->space_info->groups_sem);
7658 if (block_group->cached == BTRFS_CACHE_STARTED)
7659 wait_block_group_cache_done(block_group);
7662 * We haven't cached this block group, which means we could
7663 * possibly have excluded extents on this block group.
7665 if (block_group->cached == BTRFS_CACHE_NO)
7666 free_excluded_extents(info->extent_root, block_group);
7668 btrfs_remove_free_space_cache(block_group);
7669 btrfs_put_block_group(block_group);
7671 spin_lock(&info->block_group_cache_lock);
7673 spin_unlock(&info->block_group_cache_lock);
7675 /* now that all the block groups are freed, go through and
7676 * free all the space_info structs. This is only called during
7677 * the final stages of unmount, and so we know nobody is
7678 * using them. We call synchronize_rcu() once before we start,
7679 * just to be on the safe side.
7683 release_global_block_rsv(info);
7685 while(!list_empty(&info->space_info)) {
7686 space_info = list_entry(info->space_info.next,
7687 struct btrfs_space_info,
7689 if (space_info->bytes_pinned > 0 ||
7690 space_info->bytes_reserved > 0 ||
7691 space_info->bytes_may_use > 0) {
7693 dump_space_info(space_info, 0, 0);
7695 list_del(&space_info->list);
7701 static void __link_block_group(struct btrfs_space_info *space_info,
7702 struct btrfs_block_group_cache *cache)
7704 int index = get_block_group_index(cache);
7706 down_write(&space_info->groups_sem);
7707 list_add_tail(&cache->list, &space_info->block_groups[index]);
7708 up_write(&space_info->groups_sem);
7711 int btrfs_read_block_groups(struct btrfs_root *root)
7713 struct btrfs_path *path;
7715 struct btrfs_block_group_cache *cache;
7716 struct btrfs_fs_info *info = root->fs_info;
7717 struct btrfs_space_info *space_info;
7718 struct btrfs_key key;
7719 struct btrfs_key found_key;
7720 struct extent_buffer *leaf;
7724 root = info->extent_root;
7727 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7728 path = btrfs_alloc_path();
7733 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7734 if (btrfs_test_opt(root, SPACE_CACHE) &&
7735 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7737 if (btrfs_test_opt(root, CLEAR_CACHE))
7741 ret = find_first_block_group(root, path, &key);
7746 leaf = path->nodes[0];
7747 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7748 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7753 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7755 if (!cache->free_space_ctl) {
7761 atomic_set(&cache->count, 1);
7762 spin_lock_init(&cache->lock);
7763 cache->fs_info = info;
7764 INIT_LIST_HEAD(&cache->list);
7765 INIT_LIST_HEAD(&cache->cluster_list);
7769 * When we mount with old space cache, we need to
7770 * set BTRFS_DC_CLEAR and set dirty flag.
7772 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7773 * truncate the old free space cache inode and
7775 * b) Setting 'dirty flag' makes sure that we flush
7776 * the new space cache info onto disk.
7778 cache->disk_cache_state = BTRFS_DC_CLEAR;
7779 if (btrfs_test_opt(root, SPACE_CACHE))
7783 read_extent_buffer(leaf, &cache->item,
7784 btrfs_item_ptr_offset(leaf, path->slots[0]),
7785 sizeof(cache->item));
7786 memcpy(&cache->key, &found_key, sizeof(found_key));
7788 key.objectid = found_key.objectid + found_key.offset;
7789 btrfs_release_path(path);
7790 cache->flags = btrfs_block_group_flags(&cache->item);
7791 cache->sectorsize = root->sectorsize;
7793 btrfs_init_free_space_ctl(cache);
7796 * We need to exclude the super stripes now so that the space
7797 * info has super bytes accounted for, otherwise we'll think
7798 * we have more space than we actually do.
7800 exclude_super_stripes(root, cache);
7803 * check for two cases, either we are full, and therefore
7804 * don't need to bother with the caching work since we won't
7805 * find any space, or we are empty, and we can just add all
7806 * the space in and be done with it. This saves us _alot_ of
7807 * time, particularly in the full case.
7809 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7810 cache->last_byte_to_unpin = (u64)-1;
7811 cache->cached = BTRFS_CACHE_FINISHED;
7812 free_excluded_extents(root, cache);
7813 } else if (btrfs_block_group_used(&cache->item) == 0) {
7814 cache->last_byte_to_unpin = (u64)-1;
7815 cache->cached = BTRFS_CACHE_FINISHED;
7816 add_new_free_space(cache, root->fs_info,
7818 found_key.objectid +
7820 free_excluded_extents(root, cache);
7823 ret = update_space_info(info, cache->flags, found_key.offset,
7824 btrfs_block_group_used(&cache->item),
7826 BUG_ON(ret); /* -ENOMEM */
7827 cache->space_info = space_info;
7828 spin_lock(&cache->space_info->lock);
7829 cache->space_info->bytes_readonly += cache->bytes_super;
7830 spin_unlock(&cache->space_info->lock);
7832 __link_block_group(space_info, cache);
7834 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7835 BUG_ON(ret); /* Logic error */
7837 set_avail_alloc_bits(root->fs_info, cache->flags);
7838 if (btrfs_chunk_readonly(root, cache->key.objectid))
7839 set_block_group_ro(cache, 1);
7842 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7843 if (!(get_alloc_profile(root, space_info->flags) &
7844 (BTRFS_BLOCK_GROUP_RAID10 |
7845 BTRFS_BLOCK_GROUP_RAID1 |
7846 BTRFS_BLOCK_GROUP_DUP)))
7849 * avoid allocating from un-mirrored block group if there are
7850 * mirrored block groups.
7852 list_for_each_entry(cache, &space_info->block_groups[3], list)
7853 set_block_group_ro(cache, 1);
7854 list_for_each_entry(cache, &space_info->block_groups[4], list)
7855 set_block_group_ro(cache, 1);
7858 init_global_block_rsv(info);
7861 btrfs_free_path(path);
7865 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7866 struct btrfs_root *root)
7868 struct btrfs_block_group_cache *block_group, *tmp;
7869 struct btrfs_root *extent_root = root->fs_info->extent_root;
7870 struct btrfs_block_group_item item;
7871 struct btrfs_key key;
7874 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7876 list_del_init(&block_group->new_bg_list);
7881 spin_lock(&block_group->lock);
7882 memcpy(&item, &block_group->item, sizeof(item));
7883 memcpy(&key, &block_group->key, sizeof(key));
7884 spin_unlock(&block_group->lock);
7886 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7889 btrfs_abort_transaction(trans, extent_root, ret);
7893 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7894 struct btrfs_root *root, u64 bytes_used,
7895 u64 type, u64 chunk_objectid, u64 chunk_offset,
7899 struct btrfs_root *extent_root;
7900 struct btrfs_block_group_cache *cache;
7902 extent_root = root->fs_info->extent_root;
7904 root->fs_info->last_trans_log_full_commit = trans->transid;
7906 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7909 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7911 if (!cache->free_space_ctl) {
7916 cache->key.objectid = chunk_offset;
7917 cache->key.offset = size;
7918 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7919 cache->sectorsize = root->sectorsize;
7920 cache->fs_info = root->fs_info;
7922 atomic_set(&cache->count, 1);
7923 spin_lock_init(&cache->lock);
7924 INIT_LIST_HEAD(&cache->list);
7925 INIT_LIST_HEAD(&cache->cluster_list);
7926 INIT_LIST_HEAD(&cache->new_bg_list);
7928 btrfs_init_free_space_ctl(cache);
7930 btrfs_set_block_group_used(&cache->item, bytes_used);
7931 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7932 cache->flags = type;
7933 btrfs_set_block_group_flags(&cache->item, type);
7935 cache->last_byte_to_unpin = (u64)-1;
7936 cache->cached = BTRFS_CACHE_FINISHED;
7937 exclude_super_stripes(root, cache);
7939 add_new_free_space(cache, root->fs_info, chunk_offset,
7940 chunk_offset + size);
7942 free_excluded_extents(root, cache);
7944 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7945 &cache->space_info);
7946 BUG_ON(ret); /* -ENOMEM */
7947 update_global_block_rsv(root->fs_info);
7949 spin_lock(&cache->space_info->lock);
7950 cache->space_info->bytes_readonly += cache->bytes_super;
7951 spin_unlock(&cache->space_info->lock);
7953 __link_block_group(cache->space_info, cache);
7955 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7956 BUG_ON(ret); /* Logic error */
7958 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7960 set_avail_alloc_bits(extent_root->fs_info, type);
7965 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7967 u64 extra_flags = chunk_to_extended(flags) &
7968 BTRFS_EXTENDED_PROFILE_MASK;
7970 if (flags & BTRFS_BLOCK_GROUP_DATA)
7971 fs_info->avail_data_alloc_bits &= ~extra_flags;
7972 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7973 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7974 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7975 fs_info->avail_system_alloc_bits &= ~extra_flags;
7978 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7979 struct btrfs_root *root, u64 group_start)
7981 struct btrfs_path *path;
7982 struct btrfs_block_group_cache *block_group;
7983 struct btrfs_free_cluster *cluster;
7984 struct btrfs_root *tree_root = root->fs_info->tree_root;
7985 struct btrfs_key key;
7986 struct inode *inode;
7991 root = root->fs_info->extent_root;
7993 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7994 BUG_ON(!block_group);
7995 BUG_ON(!block_group->ro);
7998 * Free the reserved super bytes from this block group before
8001 free_excluded_extents(root, block_group);
8003 memcpy(&key, &block_group->key, sizeof(key));
8004 index = get_block_group_index(block_group);
8005 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8006 BTRFS_BLOCK_GROUP_RAID1 |
8007 BTRFS_BLOCK_GROUP_RAID10))
8012 /* make sure this block group isn't part of an allocation cluster */
8013 cluster = &root->fs_info->data_alloc_cluster;
8014 spin_lock(&cluster->refill_lock);
8015 btrfs_return_cluster_to_free_space(block_group, cluster);
8016 spin_unlock(&cluster->refill_lock);
8019 * make sure this block group isn't part of a metadata
8020 * allocation cluster
8022 cluster = &root->fs_info->meta_alloc_cluster;
8023 spin_lock(&cluster->refill_lock);
8024 btrfs_return_cluster_to_free_space(block_group, cluster);
8025 spin_unlock(&cluster->refill_lock);
8027 path = btrfs_alloc_path();
8033 inode = lookup_free_space_inode(tree_root, block_group, path);
8034 if (!IS_ERR(inode)) {
8035 ret = btrfs_orphan_add(trans, inode);
8037 btrfs_add_delayed_iput(inode);
8041 /* One for the block groups ref */
8042 spin_lock(&block_group->lock);
8043 if (block_group->iref) {
8044 block_group->iref = 0;
8045 block_group->inode = NULL;
8046 spin_unlock(&block_group->lock);
8049 spin_unlock(&block_group->lock);
8051 /* One for our lookup ref */
8052 btrfs_add_delayed_iput(inode);
8055 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8056 key.offset = block_group->key.objectid;
8059 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8063 btrfs_release_path(path);
8065 ret = btrfs_del_item(trans, tree_root, path);
8068 btrfs_release_path(path);
8071 spin_lock(&root->fs_info->block_group_cache_lock);
8072 rb_erase(&block_group->cache_node,
8073 &root->fs_info->block_group_cache_tree);
8074 spin_unlock(&root->fs_info->block_group_cache_lock);
8076 down_write(&block_group->space_info->groups_sem);
8078 * we must use list_del_init so people can check to see if they
8079 * are still on the list after taking the semaphore
8081 list_del_init(&block_group->list);
8082 if (list_empty(&block_group->space_info->block_groups[index]))
8083 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8084 up_write(&block_group->space_info->groups_sem);
8086 if (block_group->cached == BTRFS_CACHE_STARTED)
8087 wait_block_group_cache_done(block_group);
8089 btrfs_remove_free_space_cache(block_group);
8091 spin_lock(&block_group->space_info->lock);
8092 block_group->space_info->total_bytes -= block_group->key.offset;
8093 block_group->space_info->bytes_readonly -= block_group->key.offset;
8094 block_group->space_info->disk_total -= block_group->key.offset * factor;
8095 spin_unlock(&block_group->space_info->lock);
8097 memcpy(&key, &block_group->key, sizeof(key));
8099 btrfs_clear_space_info_full(root->fs_info);
8101 btrfs_put_block_group(block_group);
8102 btrfs_put_block_group(block_group);
8104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8110 ret = btrfs_del_item(trans, root, path);
8112 btrfs_free_path(path);
8116 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8118 struct btrfs_space_info *space_info;
8119 struct btrfs_super_block *disk_super;
8125 disk_super = fs_info->super_copy;
8126 if (!btrfs_super_root(disk_super))
8129 features = btrfs_super_incompat_flags(disk_super);
8130 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8133 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8134 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8139 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8140 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8142 flags = BTRFS_BLOCK_GROUP_METADATA;
8143 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8147 flags = BTRFS_BLOCK_GROUP_DATA;
8148 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8154 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8156 return unpin_extent_range(root, start, end);
8159 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8160 u64 num_bytes, u64 *actual_bytes)
8162 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8165 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8167 struct btrfs_fs_info *fs_info = root->fs_info;
8168 struct btrfs_block_group_cache *cache = NULL;
8173 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8177 * try to trim all FS space, our block group may start from non-zero.
8179 if (range->len == total_bytes)
8180 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8182 cache = btrfs_lookup_block_group(fs_info, range->start);
8185 if (cache->key.objectid >= (range->start + range->len)) {
8186 btrfs_put_block_group(cache);
8190 start = max(range->start, cache->key.objectid);
8191 end = min(range->start + range->len,
8192 cache->key.objectid + cache->key.offset);
8194 if (end - start >= range->minlen) {
8195 if (!block_group_cache_done(cache)) {
8196 ret = cache_block_group(cache, NULL, root, 0);
8198 wait_block_group_cache_done(cache);
8200 ret = btrfs_trim_block_group(cache,
8206 trimmed += group_trimmed;
8208 btrfs_put_block_group(cache);
8213 cache = next_block_group(fs_info->tree_root, cache);
8216 range->len = trimmed;