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>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
55 static int update_block_group(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int alloc);
58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, u64 parent,
61 u64 root_objectid, u64 owner_objectid,
62 u64 owner_offset, int refs_to_drop,
63 struct btrfs_delayed_extent_op *extra_op);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
65 struct extent_buffer *leaf,
66 struct btrfs_extent_item *ei);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_root *root,
69 u64 parent, u64 root_objectid,
70 u64 flags, u64 owner, u64 offset,
71 struct btrfs_key *ins, int ref_mod);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 parent, u64 root_objectid,
75 u64 flags, struct btrfs_disk_key *key,
76 int level, struct btrfs_key *ins);
77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
78 struct btrfs_root *extent_root, u64 alloc_bytes,
79 u64 flags, int force);
80 static int find_next_key(struct btrfs_path *path, int level,
81 struct btrfs_key *key);
82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
83 int dump_block_groups);
86 block_group_cache_done(struct btrfs_block_group_cache *cache)
89 return cache->cached == BTRFS_CACHE_FINISHED;
92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
94 return (cache->flags & bits) == bits;
97 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
99 atomic_inc(&cache->count);
102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
104 if (atomic_dec_and_test(&cache->count)) {
105 WARN_ON(cache->pinned > 0);
106 WARN_ON(cache->reserved > 0);
107 WARN_ON(cache->reserved_pinned > 0);
108 kfree(cache->free_space_ctl);
114 * this adds the block group to the fs_info rb tree for the block group
117 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
118 struct btrfs_block_group_cache *block_group)
121 struct rb_node *parent = NULL;
122 struct btrfs_block_group_cache *cache;
124 spin_lock(&info->block_group_cache_lock);
125 p = &info->block_group_cache_tree.rb_node;
129 cache = rb_entry(parent, struct btrfs_block_group_cache,
131 if (block_group->key.objectid < cache->key.objectid) {
133 } else if (block_group->key.objectid > cache->key.objectid) {
136 spin_unlock(&info->block_group_cache_lock);
141 rb_link_node(&block_group->cache_node, parent, p);
142 rb_insert_color(&block_group->cache_node,
143 &info->block_group_cache_tree);
144 spin_unlock(&info->block_group_cache_lock);
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
153 static struct btrfs_block_group_cache *
154 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
157 struct btrfs_block_group_cache *cache, *ret = NULL;
161 spin_lock(&info->block_group_cache_lock);
162 n = info->block_group_cache_tree.rb_node;
165 cache = rb_entry(n, struct btrfs_block_group_cache,
167 end = cache->key.objectid + cache->key.offset - 1;
168 start = cache->key.objectid;
170 if (bytenr < start) {
171 if (!contains && (!ret || start < ret->key.objectid))
174 } else if (bytenr > start) {
175 if (contains && bytenr <= end) {
186 btrfs_get_block_group(ret);
187 spin_unlock(&info->block_group_cache_lock);
192 static int add_excluded_extent(struct btrfs_root *root,
193 u64 start, u64 num_bytes)
195 u64 end = start + num_bytes - 1;
196 set_extent_bits(&root->fs_info->freed_extents[0],
197 start, end, EXTENT_UPTODATE, GFP_NOFS);
198 set_extent_bits(&root->fs_info->freed_extents[1],
199 start, end, EXTENT_UPTODATE, GFP_NOFS);
203 static void free_excluded_extents(struct btrfs_root *root,
204 struct btrfs_block_group_cache *cache)
208 start = cache->key.objectid;
209 end = start + cache->key.offset - 1;
211 clear_extent_bits(&root->fs_info->freed_extents[0],
212 start, end, EXTENT_UPTODATE, GFP_NOFS);
213 clear_extent_bits(&root->fs_info->freed_extents[1],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
217 static int exclude_super_stripes(struct btrfs_root *root,
218 struct btrfs_block_group_cache *cache)
225 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
226 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
227 cache->bytes_super += stripe_len;
228 ret = add_excluded_extent(root, cache->key.objectid,
233 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
234 bytenr = btrfs_sb_offset(i);
235 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
236 cache->key.objectid, bytenr,
237 0, &logical, &nr, &stripe_len);
241 cache->bytes_super += stripe_len;
242 ret = add_excluded_extent(root, logical[nr],
252 static struct btrfs_caching_control *
253 get_caching_control(struct btrfs_block_group_cache *cache)
255 struct btrfs_caching_control *ctl;
257 spin_lock(&cache->lock);
258 if (cache->cached != BTRFS_CACHE_STARTED) {
259 spin_unlock(&cache->lock);
263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache->caching_ctl) {
265 spin_unlock(&cache->lock);
269 ctl = cache->caching_ctl;
270 atomic_inc(&ctl->count);
271 spin_unlock(&cache->lock);
275 static void put_caching_control(struct btrfs_caching_control *ctl)
277 if (atomic_dec_and_test(&ctl->count))
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
286 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
287 struct btrfs_fs_info *info, u64 start, u64 end)
289 u64 extent_start, extent_end, size, total_added = 0;
292 while (start < end) {
293 ret = find_first_extent_bit(info->pinned_extents, start,
294 &extent_start, &extent_end,
295 EXTENT_DIRTY | EXTENT_UPTODATE);
299 if (extent_start <= start) {
300 start = extent_end + 1;
301 } else if (extent_start > start && extent_start < end) {
302 size = extent_start - start;
304 ret = btrfs_add_free_space(block_group, start,
307 start = extent_end + 1;
316 ret = btrfs_add_free_space(block_group, start, size);
323 static int caching_kthread(void *data)
325 struct btrfs_block_group_cache *block_group = data;
326 struct btrfs_fs_info *fs_info = block_group->fs_info;
327 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
328 struct btrfs_root *extent_root = fs_info->extent_root;
329 struct btrfs_path *path;
330 struct extent_buffer *leaf;
331 struct btrfs_key key;
337 path = btrfs_alloc_path();
341 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
344 * We don't want to deadlock with somebody trying to allocate a new
345 * extent for the extent root while also trying to search the extent
346 * root to add free space. So we skip locking and search the commit
347 * root, since its read-only
349 path->skip_locking = 1;
350 path->search_commit_root = 1;
355 key.type = BTRFS_EXTENT_ITEM_KEY;
357 mutex_lock(&caching_ctl->mutex);
358 /* need to make sure the commit_root doesn't disappear */
359 down_read(&fs_info->extent_commit_sem);
361 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
365 leaf = path->nodes[0];
366 nritems = btrfs_header_nritems(leaf);
370 if (fs_info->closing > 1) {
375 if (path->slots[0] < nritems) {
376 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
378 ret = find_next_key(path, 0, &key);
382 caching_ctl->progress = last;
383 btrfs_release_path(extent_root, path);
384 up_read(&fs_info->extent_commit_sem);
385 mutex_unlock(&caching_ctl->mutex);
386 if (btrfs_transaction_in_commit(fs_info))
393 if (key.objectid < block_group->key.objectid) {
398 if (key.objectid >= block_group->key.objectid +
399 block_group->key.offset)
402 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
403 total_found += add_new_free_space(block_group,
406 last = key.objectid + key.offset;
408 if (total_found > (1024 * 1024 * 2)) {
410 wake_up(&caching_ctl->wait);
417 total_found += add_new_free_space(block_group, fs_info, last,
418 block_group->key.objectid +
419 block_group->key.offset);
420 caching_ctl->progress = (u64)-1;
422 spin_lock(&block_group->lock);
423 block_group->caching_ctl = NULL;
424 block_group->cached = BTRFS_CACHE_FINISHED;
425 spin_unlock(&block_group->lock);
428 btrfs_free_path(path);
429 up_read(&fs_info->extent_commit_sem);
431 free_excluded_extents(extent_root, block_group);
433 mutex_unlock(&caching_ctl->mutex);
434 wake_up(&caching_ctl->wait);
436 put_caching_control(caching_ctl);
437 atomic_dec(&block_group->space_info->caching_threads);
438 btrfs_put_block_group(block_group);
443 static int cache_block_group(struct btrfs_block_group_cache *cache,
444 struct btrfs_trans_handle *trans,
445 struct btrfs_root *root,
448 struct btrfs_fs_info *fs_info = cache->fs_info;
449 struct btrfs_caching_control *caching_ctl;
450 struct task_struct *tsk;
454 if (cache->cached != BTRFS_CACHE_NO)
458 * We can't do the read from on-disk cache during a commit since we need
459 * to have the normal tree locking. Also if we are currently trying to
460 * allocate blocks for the tree root we can't do the fast caching since
461 * we likely hold important locks.
463 if (trans && (!trans->transaction->in_commit) &&
464 (root && root != root->fs_info->tree_root)) {
465 spin_lock(&cache->lock);
466 if (cache->cached != BTRFS_CACHE_NO) {
467 spin_unlock(&cache->lock);
470 cache->cached = BTRFS_CACHE_STARTED;
471 spin_unlock(&cache->lock);
473 ret = load_free_space_cache(fs_info, cache);
475 spin_lock(&cache->lock);
477 cache->cached = BTRFS_CACHE_FINISHED;
478 cache->last_byte_to_unpin = (u64)-1;
480 cache->cached = BTRFS_CACHE_NO;
482 spin_unlock(&cache->lock);
484 free_excluded_extents(fs_info->extent_root, cache);
492 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
493 BUG_ON(!caching_ctl);
495 INIT_LIST_HEAD(&caching_ctl->list);
496 mutex_init(&caching_ctl->mutex);
497 init_waitqueue_head(&caching_ctl->wait);
498 caching_ctl->block_group = cache;
499 caching_ctl->progress = cache->key.objectid;
500 /* one for caching kthread, one for caching block group list */
501 atomic_set(&caching_ctl->count, 2);
503 spin_lock(&cache->lock);
504 if (cache->cached != BTRFS_CACHE_NO) {
505 spin_unlock(&cache->lock);
509 cache->caching_ctl = caching_ctl;
510 cache->cached = BTRFS_CACHE_STARTED;
511 spin_unlock(&cache->lock);
513 down_write(&fs_info->extent_commit_sem);
514 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
515 up_write(&fs_info->extent_commit_sem);
517 atomic_inc(&cache->space_info->caching_threads);
518 btrfs_get_block_group(cache);
520 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
521 cache->key.objectid);
524 printk(KERN_ERR "error running thread %d\n", ret);
532 * return the block group that starts at or after bytenr
534 static struct btrfs_block_group_cache *
535 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
537 struct btrfs_block_group_cache *cache;
539 cache = block_group_cache_tree_search(info, bytenr, 0);
545 * return the block group that contains the given bytenr
547 struct btrfs_block_group_cache *btrfs_lookup_block_group(
548 struct btrfs_fs_info *info,
551 struct btrfs_block_group_cache *cache;
553 cache = block_group_cache_tree_search(info, bytenr, 1);
558 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
561 struct list_head *head = &info->space_info;
562 struct btrfs_space_info *found;
564 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
565 BTRFS_BLOCK_GROUP_METADATA;
568 list_for_each_entry_rcu(found, head, list) {
569 if (found->flags & flags) {
579 * after adding space to the filesystem, we need to clear the full flags
580 * on all the space infos.
582 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
584 struct list_head *head = &info->space_info;
585 struct btrfs_space_info *found;
588 list_for_each_entry_rcu(found, head, list)
593 static u64 div_factor(u64 num, int factor)
602 static u64 div_factor_fine(u64 num, int factor)
611 u64 btrfs_find_block_group(struct btrfs_root *root,
612 u64 search_start, u64 search_hint, int owner)
614 struct btrfs_block_group_cache *cache;
616 u64 last = max(search_hint, search_start);
623 cache = btrfs_lookup_first_block_group(root->fs_info, last);
627 spin_lock(&cache->lock);
628 last = cache->key.objectid + cache->key.offset;
629 used = btrfs_block_group_used(&cache->item);
631 if ((full_search || !cache->ro) &&
632 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
633 if (used + cache->pinned + cache->reserved <
634 div_factor(cache->key.offset, factor)) {
635 group_start = cache->key.objectid;
636 spin_unlock(&cache->lock);
637 btrfs_put_block_group(cache);
641 spin_unlock(&cache->lock);
642 btrfs_put_block_group(cache);
650 if (!full_search && factor < 10) {
660 /* simple helper to search for an existing extent at a given offset */
661 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
664 struct btrfs_key key;
665 struct btrfs_path *path;
667 path = btrfs_alloc_path();
669 key.objectid = start;
671 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
672 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
674 btrfs_free_path(path);
679 * helper function to lookup reference count and flags of extent.
681 * the head node for delayed ref is used to store the sum of all the
682 * reference count modifications queued up in the rbtree. the head
683 * node may also store the extent flags to set. This way you can check
684 * to see what the reference count and extent flags would be if all of
685 * the delayed refs are not processed.
687 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
688 struct btrfs_root *root, u64 bytenr,
689 u64 num_bytes, u64 *refs, u64 *flags)
691 struct btrfs_delayed_ref_head *head;
692 struct btrfs_delayed_ref_root *delayed_refs;
693 struct btrfs_path *path;
694 struct btrfs_extent_item *ei;
695 struct extent_buffer *leaf;
696 struct btrfs_key key;
702 path = btrfs_alloc_path();
706 key.objectid = bytenr;
707 key.type = BTRFS_EXTENT_ITEM_KEY;
708 key.offset = num_bytes;
710 path->skip_locking = 1;
711 path->search_commit_root = 1;
714 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
720 leaf = path->nodes[0];
721 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
722 if (item_size >= sizeof(*ei)) {
723 ei = btrfs_item_ptr(leaf, path->slots[0],
724 struct btrfs_extent_item);
725 num_refs = btrfs_extent_refs(leaf, ei);
726 extent_flags = btrfs_extent_flags(leaf, ei);
728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
729 struct btrfs_extent_item_v0 *ei0;
730 BUG_ON(item_size != sizeof(*ei0));
731 ei0 = btrfs_item_ptr(leaf, path->slots[0],
732 struct btrfs_extent_item_v0);
733 num_refs = btrfs_extent_refs_v0(leaf, ei0);
734 /* FIXME: this isn't correct for data */
735 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
740 BUG_ON(num_refs == 0);
750 delayed_refs = &trans->transaction->delayed_refs;
751 spin_lock(&delayed_refs->lock);
752 head = btrfs_find_delayed_ref_head(trans, bytenr);
754 if (!mutex_trylock(&head->mutex)) {
755 atomic_inc(&head->node.refs);
756 spin_unlock(&delayed_refs->lock);
758 btrfs_release_path(root->fs_info->extent_root, path);
760 mutex_lock(&head->mutex);
761 mutex_unlock(&head->mutex);
762 btrfs_put_delayed_ref(&head->node);
765 if (head->extent_op && head->extent_op->update_flags)
766 extent_flags |= head->extent_op->flags_to_set;
768 BUG_ON(num_refs == 0);
770 num_refs += head->node.ref_mod;
771 mutex_unlock(&head->mutex);
773 spin_unlock(&delayed_refs->lock);
775 WARN_ON(num_refs == 0);
779 *flags = extent_flags;
781 btrfs_free_path(path);
786 * Back reference rules. Back refs have three main goals:
788 * 1) differentiate between all holders of references to an extent so that
789 * when a reference is dropped we can make sure it was a valid reference
790 * before freeing the extent.
792 * 2) Provide enough information to quickly find the holders of an extent
793 * if we notice a given block is corrupted or bad.
795 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
796 * maintenance. This is actually the same as #2, but with a slightly
797 * different use case.
799 * There are two kinds of back refs. The implicit back refs is optimized
800 * for pointers in non-shared tree blocks. For a given pointer in a block,
801 * back refs of this kind provide information about the block's owner tree
802 * and the pointer's key. These information allow us to find the block by
803 * b-tree searching. The full back refs is for pointers in tree blocks not
804 * referenced by their owner trees. The location of tree block is recorded
805 * in the back refs. Actually the full back refs is generic, and can be
806 * used in all cases the implicit back refs is used. The major shortcoming
807 * of the full back refs is its overhead. Every time a tree block gets
808 * COWed, we have to update back refs entry for all pointers in it.
810 * For a newly allocated tree block, we use implicit back refs for
811 * pointers in it. This means most tree related operations only involve
812 * implicit back refs. For a tree block created in old transaction, the
813 * only way to drop a reference to it is COW it. So we can detect the
814 * event that tree block loses its owner tree's reference and do the
815 * back refs conversion.
817 * When a tree block is COW'd through a tree, there are four cases:
819 * The reference count of the block is one and the tree is the block's
820 * owner tree. Nothing to do in this case.
822 * The reference count of the block is one and the tree is not the
823 * block's owner tree. In this case, full back refs is used for pointers
824 * in the block. Remove these full back refs, add implicit back refs for
825 * every pointers in the new block.
827 * The reference count of the block is greater than one and the tree is
828 * the block's owner tree. In this case, implicit back refs is used for
829 * pointers in the block. Add full back refs for every pointers in the
830 * block, increase lower level extents' reference counts. The original
831 * implicit back refs are entailed to the new block.
833 * The reference count of the block is greater than one and the tree is
834 * not the block's owner tree. Add implicit back refs for every pointer in
835 * the new block, increase lower level extents' reference count.
837 * Back Reference Key composing:
839 * The key objectid corresponds to the first byte in the extent,
840 * The key type is used to differentiate between types of back refs.
841 * There are different meanings of the key offset for different types
844 * File extents can be referenced by:
846 * - multiple snapshots, subvolumes, or different generations in one subvol
847 * - different files inside a single subvolume
848 * - different offsets inside a file (bookend extents in file.c)
850 * The extent ref structure for the implicit back refs has fields for:
852 * - Objectid of the subvolume root
853 * - objectid of the file holding the reference
854 * - original offset in the file
855 * - how many bookend extents
857 * The key offset for the implicit back refs is hash of the first
860 * The extent ref structure for the full back refs has field for:
862 * - number of pointers in the tree leaf
864 * The key offset for the implicit back refs is the first byte of
867 * When a file extent is allocated, The implicit back refs is used.
868 * the fields are filled in:
870 * (root_key.objectid, inode objectid, offset in file, 1)
872 * When a file extent is removed file truncation, we find the
873 * corresponding implicit back refs and check the following fields:
875 * (btrfs_header_owner(leaf), inode objectid, offset in file)
877 * Btree extents can be referenced by:
879 * - Different subvolumes
881 * Both the implicit back refs and the full back refs for tree blocks
882 * only consist of key. The key offset for the implicit back refs is
883 * objectid of block's owner tree. The key offset for the full back refs
884 * is the first byte of parent block.
886 * When implicit back refs is used, information about the lowest key and
887 * level of the tree block are required. These information are stored in
888 * tree block info structure.
891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
892 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
893 struct btrfs_root *root,
894 struct btrfs_path *path,
895 u64 owner, u32 extra_size)
897 struct btrfs_extent_item *item;
898 struct btrfs_extent_item_v0 *ei0;
899 struct btrfs_extent_ref_v0 *ref0;
900 struct btrfs_tree_block_info *bi;
901 struct extent_buffer *leaf;
902 struct btrfs_key key;
903 struct btrfs_key found_key;
904 u32 new_size = sizeof(*item);
908 leaf = path->nodes[0];
909 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
911 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
912 ei0 = btrfs_item_ptr(leaf, path->slots[0],
913 struct btrfs_extent_item_v0);
914 refs = btrfs_extent_refs_v0(leaf, ei0);
916 if (owner == (u64)-1) {
918 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
919 ret = btrfs_next_leaf(root, path);
923 leaf = path->nodes[0];
925 btrfs_item_key_to_cpu(leaf, &found_key,
927 BUG_ON(key.objectid != found_key.objectid);
928 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
932 ref0 = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_extent_ref_v0);
934 owner = btrfs_ref_objectid_v0(leaf, ref0);
938 btrfs_release_path(root, path);
940 if (owner < BTRFS_FIRST_FREE_OBJECTID)
941 new_size += sizeof(*bi);
943 new_size -= sizeof(*ei0);
944 ret = btrfs_search_slot(trans, root, &key, path,
945 new_size + extra_size, 1);
950 ret = btrfs_extend_item(trans, root, path, new_size);
953 leaf = path->nodes[0];
954 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
955 btrfs_set_extent_refs(leaf, item, refs);
956 /* FIXME: get real generation */
957 btrfs_set_extent_generation(leaf, item, 0);
958 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
959 btrfs_set_extent_flags(leaf, item,
960 BTRFS_EXTENT_FLAG_TREE_BLOCK |
961 BTRFS_BLOCK_FLAG_FULL_BACKREF);
962 bi = (struct btrfs_tree_block_info *)(item + 1);
963 /* FIXME: get first key of the block */
964 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
965 btrfs_set_tree_block_level(leaf, bi, (int)owner);
967 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
969 btrfs_mark_buffer_dirty(leaf);
974 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
976 u32 high_crc = ~(u32)0;
977 u32 low_crc = ~(u32)0;
980 lenum = cpu_to_le64(root_objectid);
981 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
982 lenum = cpu_to_le64(owner);
983 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
984 lenum = cpu_to_le64(offset);
985 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
987 return ((u64)high_crc << 31) ^ (u64)low_crc;
990 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
991 struct btrfs_extent_data_ref *ref)
993 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
994 btrfs_extent_data_ref_objectid(leaf, ref),
995 btrfs_extent_data_ref_offset(leaf, ref));
998 static int match_extent_data_ref(struct extent_buffer *leaf,
999 struct btrfs_extent_data_ref *ref,
1000 u64 root_objectid, u64 owner, u64 offset)
1002 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1003 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1004 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1009 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1010 struct btrfs_root *root,
1011 struct btrfs_path *path,
1012 u64 bytenr, u64 parent,
1014 u64 owner, u64 offset)
1016 struct btrfs_key key;
1017 struct btrfs_extent_data_ref *ref;
1018 struct extent_buffer *leaf;
1024 key.objectid = bytenr;
1026 key.type = BTRFS_SHARED_DATA_REF_KEY;
1027 key.offset = parent;
1029 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1030 key.offset = hash_extent_data_ref(root_objectid,
1035 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 key.type = BTRFS_EXTENT_REF_V0_KEY;
1046 btrfs_release_path(root, path);
1047 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1058 leaf = path->nodes[0];
1059 nritems = btrfs_header_nritems(leaf);
1061 if (path->slots[0] >= nritems) {
1062 ret = btrfs_next_leaf(root, path);
1068 leaf = path->nodes[0];
1069 nritems = btrfs_header_nritems(leaf);
1073 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074 if (key.objectid != bytenr ||
1075 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1078 ref = btrfs_item_ptr(leaf, path->slots[0],
1079 struct btrfs_extent_data_ref);
1081 if (match_extent_data_ref(leaf, ref, root_objectid,
1084 btrfs_release_path(root, path);
1096 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1100 u64 root_objectid, u64 owner,
1101 u64 offset, int refs_to_add)
1103 struct btrfs_key key;
1104 struct extent_buffer *leaf;
1109 key.objectid = bytenr;
1111 key.type = BTRFS_SHARED_DATA_REF_KEY;
1112 key.offset = parent;
1113 size = sizeof(struct btrfs_shared_data_ref);
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1118 size = sizeof(struct btrfs_extent_data_ref);
1121 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1122 if (ret && ret != -EEXIST)
1125 leaf = path->nodes[0];
1127 struct btrfs_shared_data_ref *ref;
1128 ref = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_shared_data_ref);
1131 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1133 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1134 num_refs += refs_to_add;
1135 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1138 struct btrfs_extent_data_ref *ref;
1139 while (ret == -EEXIST) {
1140 ref = btrfs_item_ptr(leaf, path->slots[0],
1141 struct btrfs_extent_data_ref);
1142 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 btrfs_release_path(root, path);
1147 ret = btrfs_insert_empty_item(trans, root, path, &key,
1149 if (ret && ret != -EEXIST)
1152 leaf = path->nodes[0];
1154 ref = btrfs_item_ptr(leaf, path->slots[0],
1155 struct btrfs_extent_data_ref);
1157 btrfs_set_extent_data_ref_root(leaf, ref,
1159 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1160 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1161 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1163 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1164 num_refs += refs_to_add;
1165 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1168 btrfs_mark_buffer_dirty(leaf);
1171 btrfs_release_path(root, path);
1175 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1176 struct btrfs_root *root,
1177 struct btrfs_path *path,
1180 struct btrfs_key key;
1181 struct btrfs_extent_data_ref *ref1 = NULL;
1182 struct btrfs_shared_data_ref *ref2 = NULL;
1183 struct extent_buffer *leaf;
1187 leaf = path->nodes[0];
1188 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1190 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1191 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_extent_data_ref);
1193 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1194 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1195 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1196 struct btrfs_shared_data_ref);
1197 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1198 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1199 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1200 struct btrfs_extent_ref_v0 *ref0;
1201 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_ref_v0);
1203 num_refs = btrfs_ref_count_v0(leaf, ref0);
1209 BUG_ON(num_refs < refs_to_drop);
1210 num_refs -= refs_to_drop;
1212 if (num_refs == 0) {
1213 ret = btrfs_del_item(trans, root, path);
1215 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1216 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1217 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1218 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1221 struct btrfs_extent_ref_v0 *ref0;
1222 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_extent_ref_v0);
1224 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1227 btrfs_mark_buffer_dirty(leaf);
1232 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1233 struct btrfs_path *path,
1234 struct btrfs_extent_inline_ref *iref)
1236 struct btrfs_key key;
1237 struct extent_buffer *leaf;
1238 struct btrfs_extent_data_ref *ref1;
1239 struct btrfs_shared_data_ref *ref2;
1242 leaf = path->nodes[0];
1243 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1245 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1246 BTRFS_EXTENT_DATA_REF_KEY) {
1247 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1248 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1250 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1251 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1253 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1274 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1275 struct btrfs_root *root,
1276 struct btrfs_path *path,
1277 u64 bytenr, u64 parent,
1280 struct btrfs_key key;
1283 key.objectid = bytenr;
1285 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1286 key.offset = parent;
1288 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1289 key.offset = root_objectid;
1292 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1295 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1296 if (ret == -ENOENT && parent) {
1297 btrfs_release_path(root, path);
1298 key.type = BTRFS_EXTENT_REF_V0_KEY;
1299 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1307 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1308 struct btrfs_root *root,
1309 struct btrfs_path *path,
1310 u64 bytenr, u64 parent,
1313 struct btrfs_key key;
1316 key.objectid = bytenr;
1318 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1319 key.offset = parent;
1321 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1322 key.offset = root_objectid;
1325 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1326 btrfs_release_path(root, path);
1330 static inline int extent_ref_type(u64 parent, u64 owner)
1333 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1335 type = BTRFS_SHARED_BLOCK_REF_KEY;
1337 type = BTRFS_TREE_BLOCK_REF_KEY;
1340 type = BTRFS_SHARED_DATA_REF_KEY;
1342 type = BTRFS_EXTENT_DATA_REF_KEY;
1347 static int find_next_key(struct btrfs_path *path, int level,
1348 struct btrfs_key *key)
1351 for (; level < BTRFS_MAX_LEVEL; level++) {
1352 if (!path->nodes[level])
1354 if (path->slots[level] + 1 >=
1355 btrfs_header_nritems(path->nodes[level]))
1358 btrfs_item_key_to_cpu(path->nodes[level], key,
1359 path->slots[level] + 1);
1361 btrfs_node_key_to_cpu(path->nodes[level], key,
1362 path->slots[level] + 1);
1369 * look for inline back ref. if back ref is found, *ref_ret is set
1370 * to the address of inline back ref, and 0 is returned.
1372 * if back ref isn't found, *ref_ret is set to the address where it
1373 * should be inserted, and -ENOENT is returned.
1375 * if insert is true and there are too many inline back refs, the path
1376 * points to the extent item, and -EAGAIN is returned.
1378 * NOTE: inline back refs are ordered in the same way that back ref
1379 * items in the tree are ordered.
1381 static noinline_for_stack
1382 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1383 struct btrfs_root *root,
1384 struct btrfs_path *path,
1385 struct btrfs_extent_inline_ref **ref_ret,
1386 u64 bytenr, u64 num_bytes,
1387 u64 parent, u64 root_objectid,
1388 u64 owner, u64 offset, int insert)
1390 struct btrfs_key key;
1391 struct extent_buffer *leaf;
1392 struct btrfs_extent_item *ei;
1393 struct btrfs_extent_inline_ref *iref;
1404 key.objectid = bytenr;
1405 key.type = BTRFS_EXTENT_ITEM_KEY;
1406 key.offset = num_bytes;
1408 want = extent_ref_type(parent, owner);
1410 extra_size = btrfs_extent_inline_ref_size(want);
1411 path->keep_locks = 1;
1414 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1421 leaf = path->nodes[0];
1422 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1424 if (item_size < sizeof(*ei)) {
1429 ret = convert_extent_item_v0(trans, root, path, owner,
1435 leaf = path->nodes[0];
1436 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1439 BUG_ON(item_size < sizeof(*ei));
1441 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1442 flags = btrfs_extent_flags(leaf, ei);
1444 ptr = (unsigned long)(ei + 1);
1445 end = (unsigned long)ei + item_size;
1447 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1448 ptr += sizeof(struct btrfs_tree_block_info);
1451 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1460 iref = (struct btrfs_extent_inline_ref *)ptr;
1461 type = btrfs_extent_inline_ref_type(leaf, iref);
1465 ptr += btrfs_extent_inline_ref_size(type);
1469 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470 struct btrfs_extent_data_ref *dref;
1471 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1472 if (match_extent_data_ref(leaf, dref, root_objectid,
1477 if (hash_extent_data_ref_item(leaf, dref) <
1478 hash_extent_data_ref(root_objectid, owner, offset))
1482 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1484 if (parent == ref_offset) {
1488 if (ref_offset < parent)
1491 if (root_objectid == ref_offset) {
1495 if (ref_offset < root_objectid)
1499 ptr += btrfs_extent_inline_ref_size(type);
1501 if (err == -ENOENT && insert) {
1502 if (item_size + extra_size >=
1503 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1508 * To add new inline back ref, we have to make sure
1509 * there is no corresponding back ref item.
1510 * For simplicity, we just do not add new inline back
1511 * ref if there is any kind of item for this block
1513 if (find_next_key(path, 0, &key) == 0 &&
1514 key.objectid == bytenr &&
1515 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1520 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1523 path->keep_locks = 0;
1524 btrfs_unlock_up_safe(path, 1);
1530 * helper to add new inline back ref
1532 static noinline_for_stack
1533 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1534 struct btrfs_root *root,
1535 struct btrfs_path *path,
1536 struct btrfs_extent_inline_ref *iref,
1537 u64 parent, u64 root_objectid,
1538 u64 owner, u64 offset, int refs_to_add,
1539 struct btrfs_delayed_extent_op *extent_op)
1541 struct extent_buffer *leaf;
1542 struct btrfs_extent_item *ei;
1545 unsigned long item_offset;
1551 leaf = path->nodes[0];
1552 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1553 item_offset = (unsigned long)iref - (unsigned long)ei;
1555 type = extent_ref_type(parent, owner);
1556 size = btrfs_extent_inline_ref_size(type);
1558 ret = btrfs_extend_item(trans, root, path, size);
1561 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1562 refs = btrfs_extent_refs(leaf, ei);
1563 refs += refs_to_add;
1564 btrfs_set_extent_refs(leaf, ei, refs);
1566 __run_delayed_extent_op(extent_op, leaf, ei);
1568 ptr = (unsigned long)ei + item_offset;
1569 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1570 if (ptr < end - size)
1571 memmove_extent_buffer(leaf, ptr + size, ptr,
1574 iref = (struct btrfs_extent_inline_ref *)ptr;
1575 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1576 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1577 struct btrfs_extent_data_ref *dref;
1578 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1579 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1580 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1581 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1582 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1583 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1584 struct btrfs_shared_data_ref *sref;
1585 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1586 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1587 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1588 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1589 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1591 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1593 btrfs_mark_buffer_dirty(leaf);
1597 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct btrfs_path *path,
1600 struct btrfs_extent_inline_ref **ref_ret,
1601 u64 bytenr, u64 num_bytes, u64 parent,
1602 u64 root_objectid, u64 owner, u64 offset)
1606 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1607 bytenr, num_bytes, parent,
1608 root_objectid, owner, offset, 0);
1612 btrfs_release_path(root, path);
1615 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1616 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1619 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1620 root_objectid, owner, offset);
1626 * helper to update/remove inline back ref
1628 static noinline_for_stack
1629 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1630 struct btrfs_root *root,
1631 struct btrfs_path *path,
1632 struct btrfs_extent_inline_ref *iref,
1634 struct btrfs_delayed_extent_op *extent_op)
1636 struct extent_buffer *leaf;
1637 struct btrfs_extent_item *ei;
1638 struct btrfs_extent_data_ref *dref = NULL;
1639 struct btrfs_shared_data_ref *sref = NULL;
1648 leaf = path->nodes[0];
1649 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650 refs = btrfs_extent_refs(leaf, ei);
1651 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1652 refs += refs_to_mod;
1653 btrfs_set_extent_refs(leaf, ei, refs);
1655 __run_delayed_extent_op(extent_op, leaf, ei);
1657 type = btrfs_extent_inline_ref_type(leaf, iref);
1659 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1660 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1661 refs = btrfs_extent_data_ref_count(leaf, dref);
1662 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1663 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1664 refs = btrfs_shared_data_ref_count(leaf, sref);
1667 BUG_ON(refs_to_mod != -1);
1670 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1671 refs += refs_to_mod;
1674 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1675 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1677 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1679 size = btrfs_extent_inline_ref_size(type);
1680 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1681 ptr = (unsigned long)iref;
1682 end = (unsigned long)ei + item_size;
1683 if (ptr + size < end)
1684 memmove_extent_buffer(leaf, ptr, ptr + size,
1687 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1690 btrfs_mark_buffer_dirty(leaf);
1694 static noinline_for_stack
1695 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1696 struct btrfs_root *root,
1697 struct btrfs_path *path,
1698 u64 bytenr, u64 num_bytes, u64 parent,
1699 u64 root_objectid, u64 owner,
1700 u64 offset, int refs_to_add,
1701 struct btrfs_delayed_extent_op *extent_op)
1703 struct btrfs_extent_inline_ref *iref;
1706 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1707 bytenr, num_bytes, parent,
1708 root_objectid, owner, offset, 1);
1710 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1711 ret = update_inline_extent_backref(trans, root, path, iref,
1712 refs_to_add, extent_op);
1713 } else if (ret == -ENOENT) {
1714 ret = setup_inline_extent_backref(trans, root, path, iref,
1715 parent, root_objectid,
1716 owner, offset, refs_to_add,
1722 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 u64 bytenr, u64 parent, u64 root_objectid,
1726 u64 owner, u64 offset, int refs_to_add)
1729 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1730 BUG_ON(refs_to_add != 1);
1731 ret = insert_tree_block_ref(trans, root, path, bytenr,
1732 parent, root_objectid);
1734 ret = insert_extent_data_ref(trans, root, path, bytenr,
1735 parent, root_objectid,
1736 owner, offset, refs_to_add);
1741 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root,
1743 struct btrfs_path *path,
1744 struct btrfs_extent_inline_ref *iref,
1745 int refs_to_drop, int is_data)
1749 BUG_ON(!is_data && refs_to_drop != 1);
1751 ret = update_inline_extent_backref(trans, root, path, iref,
1752 -refs_to_drop, NULL);
1753 } else if (is_data) {
1754 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1756 ret = btrfs_del_item(trans, root, path);
1761 static int btrfs_issue_discard(struct block_device *bdev,
1764 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1767 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1768 u64 num_bytes, u64 *actual_bytes)
1771 u64 discarded_bytes = 0;
1772 struct btrfs_multi_bio *multi = NULL;
1775 /* Tell the block device(s) that the sectors can be discarded */
1776 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1777 bytenr, &num_bytes, &multi, 0);
1779 struct btrfs_bio_stripe *stripe = multi->stripes;
1783 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1784 ret = btrfs_issue_discard(stripe->dev->bdev,
1788 discarded_bytes += stripe->length;
1789 else if (ret != -EOPNOTSUPP)
1794 if (discarded_bytes && ret == -EOPNOTSUPP)
1798 *actual_bytes = discarded_bytes;
1804 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1805 struct btrfs_root *root,
1806 u64 bytenr, u64 num_bytes, u64 parent,
1807 u64 root_objectid, u64 owner, u64 offset)
1810 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1811 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1813 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1814 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1815 parent, root_objectid, (int)owner,
1816 BTRFS_ADD_DELAYED_REF, NULL);
1818 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1819 parent, root_objectid, owner, offset,
1820 BTRFS_ADD_DELAYED_REF, NULL);
1825 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1826 struct btrfs_root *root,
1827 u64 bytenr, u64 num_bytes,
1828 u64 parent, u64 root_objectid,
1829 u64 owner, u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_path *path;
1833 struct extent_buffer *leaf;
1834 struct btrfs_extent_item *item;
1839 path = btrfs_alloc_path();
1844 path->leave_spinning = 1;
1845 /* this will setup the path even if it fails to insert the back ref */
1846 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1847 path, bytenr, num_bytes, parent,
1848 root_objectid, owner, offset,
1849 refs_to_add, extent_op);
1853 if (ret != -EAGAIN) {
1858 leaf = path->nodes[0];
1859 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1860 refs = btrfs_extent_refs(leaf, item);
1861 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1863 __run_delayed_extent_op(extent_op, leaf, item);
1865 btrfs_mark_buffer_dirty(leaf);
1866 btrfs_release_path(root->fs_info->extent_root, path);
1869 path->leave_spinning = 1;
1871 /* now insert the actual backref */
1872 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1873 path, bytenr, parent, root_objectid,
1874 owner, offset, refs_to_add);
1877 btrfs_free_path(path);
1881 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1882 struct btrfs_root *root,
1883 struct btrfs_delayed_ref_node *node,
1884 struct btrfs_delayed_extent_op *extent_op,
1885 int insert_reserved)
1888 struct btrfs_delayed_data_ref *ref;
1889 struct btrfs_key ins;
1894 ins.objectid = node->bytenr;
1895 ins.offset = node->num_bytes;
1896 ins.type = BTRFS_EXTENT_ITEM_KEY;
1898 ref = btrfs_delayed_node_to_data_ref(node);
1899 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1900 parent = ref->parent;
1902 ref_root = ref->root;
1904 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1906 BUG_ON(extent_op->update_key);
1907 flags |= extent_op->flags_to_set;
1909 ret = alloc_reserved_file_extent(trans, root,
1910 parent, ref_root, flags,
1911 ref->objectid, ref->offset,
1912 &ins, node->ref_mod);
1913 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1914 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1915 node->num_bytes, parent,
1916 ref_root, ref->objectid,
1917 ref->offset, node->ref_mod,
1919 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1920 ret = __btrfs_free_extent(trans, root, node->bytenr,
1921 node->num_bytes, parent,
1922 ref_root, ref->objectid,
1923 ref->offset, node->ref_mod,
1931 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1932 struct extent_buffer *leaf,
1933 struct btrfs_extent_item *ei)
1935 u64 flags = btrfs_extent_flags(leaf, ei);
1936 if (extent_op->update_flags) {
1937 flags |= extent_op->flags_to_set;
1938 btrfs_set_extent_flags(leaf, ei, flags);
1941 if (extent_op->update_key) {
1942 struct btrfs_tree_block_info *bi;
1943 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1944 bi = (struct btrfs_tree_block_info *)(ei + 1);
1945 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1949 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1950 struct btrfs_root *root,
1951 struct btrfs_delayed_ref_node *node,
1952 struct btrfs_delayed_extent_op *extent_op)
1954 struct btrfs_key key;
1955 struct btrfs_path *path;
1956 struct btrfs_extent_item *ei;
1957 struct extent_buffer *leaf;
1962 path = btrfs_alloc_path();
1966 key.objectid = node->bytenr;
1967 key.type = BTRFS_EXTENT_ITEM_KEY;
1968 key.offset = node->num_bytes;
1971 path->leave_spinning = 1;
1972 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1983 leaf = path->nodes[0];
1984 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1985 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1986 if (item_size < sizeof(*ei)) {
1987 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1993 leaf = path->nodes[0];
1994 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1997 BUG_ON(item_size < sizeof(*ei));
1998 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 __run_delayed_extent_op(extent_op, leaf, ei);
2001 btrfs_mark_buffer_dirty(leaf);
2003 btrfs_free_path(path);
2007 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2008 struct btrfs_root *root,
2009 struct btrfs_delayed_ref_node *node,
2010 struct btrfs_delayed_extent_op *extent_op,
2011 int insert_reserved)
2014 struct btrfs_delayed_tree_ref *ref;
2015 struct btrfs_key ins;
2019 ins.objectid = node->bytenr;
2020 ins.offset = node->num_bytes;
2021 ins.type = BTRFS_EXTENT_ITEM_KEY;
2023 ref = btrfs_delayed_node_to_tree_ref(node);
2024 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2025 parent = ref->parent;
2027 ref_root = ref->root;
2029 BUG_ON(node->ref_mod != 1);
2030 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2031 BUG_ON(!extent_op || !extent_op->update_flags ||
2032 !extent_op->update_key);
2033 ret = alloc_reserved_tree_block(trans, root,
2035 extent_op->flags_to_set,
2038 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2039 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2040 node->num_bytes, parent, ref_root,
2041 ref->level, 0, 1, extent_op);
2042 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2043 ret = __btrfs_free_extent(trans, root, node->bytenr,
2044 node->num_bytes, parent, ref_root,
2045 ref->level, 0, 1, extent_op);
2052 /* helper function to actually process a single delayed ref entry */
2053 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2054 struct btrfs_root *root,
2055 struct btrfs_delayed_ref_node *node,
2056 struct btrfs_delayed_extent_op *extent_op,
2057 int insert_reserved)
2060 if (btrfs_delayed_ref_is_head(node)) {
2061 struct btrfs_delayed_ref_head *head;
2063 * we've hit the end of the chain and we were supposed
2064 * to insert this extent into the tree. But, it got
2065 * deleted before we ever needed to insert it, so all
2066 * we have to do is clean up the accounting
2069 head = btrfs_delayed_node_to_head(node);
2070 if (insert_reserved) {
2071 btrfs_pin_extent(root, node->bytenr,
2072 node->num_bytes, 1);
2073 if (head->is_data) {
2074 ret = btrfs_del_csums(trans, root,
2080 mutex_unlock(&head->mutex);
2084 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2085 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2086 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2088 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2089 node->type == BTRFS_SHARED_DATA_REF_KEY)
2090 ret = run_delayed_data_ref(trans, root, node, extent_op,
2097 static noinline struct btrfs_delayed_ref_node *
2098 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2100 struct rb_node *node;
2101 struct btrfs_delayed_ref_node *ref;
2102 int action = BTRFS_ADD_DELAYED_REF;
2105 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2106 * this prevents ref count from going down to zero when
2107 * there still are pending delayed ref.
2109 node = rb_prev(&head->node.rb_node);
2113 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2115 if (ref->bytenr != head->node.bytenr)
2117 if (ref->action == action)
2119 node = rb_prev(node);
2121 if (action == BTRFS_ADD_DELAYED_REF) {
2122 action = BTRFS_DROP_DELAYED_REF;
2128 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2129 struct btrfs_root *root,
2130 struct list_head *cluster)
2132 struct btrfs_delayed_ref_root *delayed_refs;
2133 struct btrfs_delayed_ref_node *ref;
2134 struct btrfs_delayed_ref_head *locked_ref = NULL;
2135 struct btrfs_delayed_extent_op *extent_op;
2138 int must_insert_reserved = 0;
2140 delayed_refs = &trans->transaction->delayed_refs;
2143 /* pick a new head ref from the cluster list */
2144 if (list_empty(cluster))
2147 locked_ref = list_entry(cluster->next,
2148 struct btrfs_delayed_ref_head, cluster);
2150 /* grab the lock that says we are going to process
2151 * all the refs for this head */
2152 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2155 * we may have dropped the spin lock to get the head
2156 * mutex lock, and that might have given someone else
2157 * time to free the head. If that's true, it has been
2158 * removed from our list and we can move on.
2160 if (ret == -EAGAIN) {
2168 * record the must insert reserved flag before we
2169 * drop the spin lock.
2171 must_insert_reserved = locked_ref->must_insert_reserved;
2172 locked_ref->must_insert_reserved = 0;
2174 extent_op = locked_ref->extent_op;
2175 locked_ref->extent_op = NULL;
2178 * locked_ref is the head node, so we have to go one
2179 * node back for any delayed ref updates
2181 ref = select_delayed_ref(locked_ref);
2183 /* All delayed refs have been processed, Go ahead
2184 * and send the head node to run_one_delayed_ref,
2185 * so that any accounting fixes can happen
2187 ref = &locked_ref->node;
2189 if (extent_op && must_insert_reserved) {
2195 spin_unlock(&delayed_refs->lock);
2197 ret = run_delayed_extent_op(trans, root,
2203 spin_lock(&delayed_refs->lock);
2207 list_del_init(&locked_ref->cluster);
2212 rb_erase(&ref->rb_node, &delayed_refs->root);
2213 delayed_refs->num_entries--;
2215 spin_unlock(&delayed_refs->lock);
2217 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2218 must_insert_reserved);
2221 btrfs_put_delayed_ref(ref);
2226 spin_lock(&delayed_refs->lock);
2232 * this starts processing the delayed reference count updates and
2233 * extent insertions we have queued up so far. count can be
2234 * 0, which means to process everything in the tree at the start
2235 * of the run (but not newly added entries), or it can be some target
2236 * number you'd like to process.
2238 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2239 struct btrfs_root *root, unsigned long count)
2241 struct rb_node *node;
2242 struct btrfs_delayed_ref_root *delayed_refs;
2243 struct btrfs_delayed_ref_node *ref;
2244 struct list_head cluster;
2246 int run_all = count == (unsigned long)-1;
2249 if (root == root->fs_info->extent_root)
2250 root = root->fs_info->tree_root;
2252 delayed_refs = &trans->transaction->delayed_refs;
2253 INIT_LIST_HEAD(&cluster);
2255 spin_lock(&delayed_refs->lock);
2257 count = delayed_refs->num_entries * 2;
2261 if (!(run_all || run_most) &&
2262 delayed_refs->num_heads_ready < 64)
2266 * go find something we can process in the rbtree. We start at
2267 * the beginning of the tree, and then build a cluster
2268 * of refs to process starting at the first one we are able to
2271 ret = btrfs_find_ref_cluster(trans, &cluster,
2272 delayed_refs->run_delayed_start);
2276 ret = run_clustered_refs(trans, root, &cluster);
2279 count -= min_t(unsigned long, ret, count);
2286 node = rb_first(&delayed_refs->root);
2289 count = (unsigned long)-1;
2292 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2294 if (btrfs_delayed_ref_is_head(ref)) {
2295 struct btrfs_delayed_ref_head *head;
2297 head = btrfs_delayed_node_to_head(ref);
2298 atomic_inc(&ref->refs);
2300 spin_unlock(&delayed_refs->lock);
2301 mutex_lock(&head->mutex);
2302 mutex_unlock(&head->mutex);
2304 btrfs_put_delayed_ref(ref);
2308 node = rb_next(node);
2310 spin_unlock(&delayed_refs->lock);
2311 schedule_timeout(1);
2315 spin_unlock(&delayed_refs->lock);
2319 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2320 struct btrfs_root *root,
2321 u64 bytenr, u64 num_bytes, u64 flags,
2324 struct btrfs_delayed_extent_op *extent_op;
2327 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2331 extent_op->flags_to_set = flags;
2332 extent_op->update_flags = 1;
2333 extent_op->update_key = 0;
2334 extent_op->is_data = is_data ? 1 : 0;
2336 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2342 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2343 struct btrfs_root *root,
2344 struct btrfs_path *path,
2345 u64 objectid, u64 offset, u64 bytenr)
2347 struct btrfs_delayed_ref_head *head;
2348 struct btrfs_delayed_ref_node *ref;
2349 struct btrfs_delayed_data_ref *data_ref;
2350 struct btrfs_delayed_ref_root *delayed_refs;
2351 struct rb_node *node;
2355 delayed_refs = &trans->transaction->delayed_refs;
2356 spin_lock(&delayed_refs->lock);
2357 head = btrfs_find_delayed_ref_head(trans, bytenr);
2361 if (!mutex_trylock(&head->mutex)) {
2362 atomic_inc(&head->node.refs);
2363 spin_unlock(&delayed_refs->lock);
2365 btrfs_release_path(root->fs_info->extent_root, path);
2367 mutex_lock(&head->mutex);
2368 mutex_unlock(&head->mutex);
2369 btrfs_put_delayed_ref(&head->node);
2373 node = rb_prev(&head->node.rb_node);
2377 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2379 if (ref->bytenr != bytenr)
2383 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2386 data_ref = btrfs_delayed_node_to_data_ref(ref);
2388 node = rb_prev(node);
2390 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2391 if (ref->bytenr == bytenr)
2395 if (data_ref->root != root->root_key.objectid ||
2396 data_ref->objectid != objectid || data_ref->offset != offset)
2401 mutex_unlock(&head->mutex);
2403 spin_unlock(&delayed_refs->lock);
2407 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2408 struct btrfs_root *root,
2409 struct btrfs_path *path,
2410 u64 objectid, u64 offset, u64 bytenr)
2412 struct btrfs_root *extent_root = root->fs_info->extent_root;
2413 struct extent_buffer *leaf;
2414 struct btrfs_extent_data_ref *ref;
2415 struct btrfs_extent_inline_ref *iref;
2416 struct btrfs_extent_item *ei;
2417 struct btrfs_key key;
2421 key.objectid = bytenr;
2422 key.offset = (u64)-1;
2423 key.type = BTRFS_EXTENT_ITEM_KEY;
2425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2431 if (path->slots[0] == 0)
2435 leaf = path->nodes[0];
2436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2438 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2442 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2444 if (item_size < sizeof(*ei)) {
2445 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2449 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2451 if (item_size != sizeof(*ei) +
2452 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2455 if (btrfs_extent_generation(leaf, ei) <=
2456 btrfs_root_last_snapshot(&root->root_item))
2459 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2460 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2461 BTRFS_EXTENT_DATA_REF_KEY)
2464 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2465 if (btrfs_extent_refs(leaf, ei) !=
2466 btrfs_extent_data_ref_count(leaf, ref) ||
2467 btrfs_extent_data_ref_root(leaf, ref) !=
2468 root->root_key.objectid ||
2469 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2470 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2478 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2479 struct btrfs_root *root,
2480 u64 objectid, u64 offset, u64 bytenr)
2482 struct btrfs_path *path;
2486 path = btrfs_alloc_path();
2491 ret = check_committed_ref(trans, root, path, objectid,
2493 if (ret && ret != -ENOENT)
2496 ret2 = check_delayed_ref(trans, root, path, objectid,
2498 } while (ret2 == -EAGAIN);
2500 if (ret2 && ret2 != -ENOENT) {
2505 if (ret != -ENOENT || ret2 != -ENOENT)
2508 btrfs_free_path(path);
2509 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2515 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2516 struct extent_buffer *buf, u32 nr_extents)
2518 struct btrfs_key key;
2519 struct btrfs_file_extent_item *fi;
2527 if (!root->ref_cows)
2530 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2532 root_gen = root->root_key.offset;
2535 root_gen = trans->transid - 1;
2538 level = btrfs_header_level(buf);
2539 nritems = btrfs_header_nritems(buf);
2542 struct btrfs_leaf_ref *ref;
2543 struct btrfs_extent_info *info;
2545 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2551 ref->root_gen = root_gen;
2552 ref->bytenr = buf->start;
2553 ref->owner = btrfs_header_owner(buf);
2554 ref->generation = btrfs_header_generation(buf);
2555 ref->nritems = nr_extents;
2556 info = ref->extents;
2558 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2560 btrfs_item_key_to_cpu(buf, &key, i);
2561 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2563 fi = btrfs_item_ptr(buf, i,
2564 struct btrfs_file_extent_item);
2565 if (btrfs_file_extent_type(buf, fi) ==
2566 BTRFS_FILE_EXTENT_INLINE)
2568 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2569 if (disk_bytenr == 0)
2572 info->bytenr = disk_bytenr;
2574 btrfs_file_extent_disk_num_bytes(buf, fi);
2575 info->objectid = key.objectid;
2576 info->offset = key.offset;
2580 ret = btrfs_add_leaf_ref(root, ref, shared);
2581 if (ret == -EEXIST && shared) {
2582 struct btrfs_leaf_ref *old;
2583 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2585 btrfs_remove_leaf_ref(root, old);
2586 btrfs_free_leaf_ref(root, old);
2587 ret = btrfs_add_leaf_ref(root, ref, shared);
2590 btrfs_free_leaf_ref(root, ref);
2596 /* when a block goes through cow, we update the reference counts of
2597 * everything that block points to. The internal pointers of the block
2598 * can be in just about any order, and it is likely to have clusters of
2599 * things that are close together and clusters of things that are not.
2601 * To help reduce the seeks that come with updating all of these reference
2602 * counts, sort them by byte number before actual updates are done.
2604 * struct refsort is used to match byte number to slot in the btree block.
2605 * we sort based on the byte number and then use the slot to actually
2608 * struct refsort is smaller than strcut btrfs_item and smaller than
2609 * struct btrfs_key_ptr. Since we're currently limited to the page size
2610 * for a btree block, there's no way for a kmalloc of refsorts for a
2611 * single node to be bigger than a page.
2619 * for passing into sort()
2621 static int refsort_cmp(const void *a_void, const void *b_void)
2623 const struct refsort *a = a_void;
2624 const struct refsort *b = b_void;
2626 if (a->bytenr < b->bytenr)
2628 if (a->bytenr > b->bytenr)
2634 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2635 struct btrfs_root *root,
2636 struct extent_buffer *buf,
2637 int full_backref, int inc)
2644 struct btrfs_key key;
2645 struct btrfs_file_extent_item *fi;
2649 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2650 u64, u64, u64, u64, u64, u64);
2652 ref_root = btrfs_header_owner(buf);
2653 nritems = btrfs_header_nritems(buf);
2654 level = btrfs_header_level(buf);
2656 if (!root->ref_cows && level == 0)
2660 process_func = btrfs_inc_extent_ref;
2662 process_func = btrfs_free_extent;
2665 parent = buf->start;
2669 for (i = 0; i < nritems; i++) {
2671 btrfs_item_key_to_cpu(buf, &key, i);
2672 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2674 fi = btrfs_item_ptr(buf, i,
2675 struct btrfs_file_extent_item);
2676 if (btrfs_file_extent_type(buf, fi) ==
2677 BTRFS_FILE_EXTENT_INLINE)
2679 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2683 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2684 key.offset -= btrfs_file_extent_offset(buf, fi);
2685 ret = process_func(trans, root, bytenr, num_bytes,
2686 parent, ref_root, key.objectid,
2691 bytenr = btrfs_node_blockptr(buf, i);
2692 num_bytes = btrfs_level_size(root, level - 1);
2693 ret = process_func(trans, root, bytenr, num_bytes,
2694 parent, ref_root, level - 1, 0);
2705 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2706 struct extent_buffer *buf, int full_backref)
2708 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2711 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2712 struct extent_buffer *buf, int full_backref)
2714 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2717 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2718 struct btrfs_root *root,
2719 struct btrfs_path *path,
2720 struct btrfs_block_group_cache *cache)
2723 struct btrfs_root *extent_root = root->fs_info->extent_root;
2725 struct extent_buffer *leaf;
2727 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2732 leaf = path->nodes[0];
2733 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2734 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2735 btrfs_mark_buffer_dirty(leaf);
2736 btrfs_release_path(extent_root, path);
2744 static struct btrfs_block_group_cache *
2745 next_block_group(struct btrfs_root *root,
2746 struct btrfs_block_group_cache *cache)
2748 struct rb_node *node;
2749 spin_lock(&root->fs_info->block_group_cache_lock);
2750 node = rb_next(&cache->cache_node);
2751 btrfs_put_block_group(cache);
2753 cache = rb_entry(node, struct btrfs_block_group_cache,
2755 btrfs_get_block_group(cache);
2758 spin_unlock(&root->fs_info->block_group_cache_lock);
2762 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2763 struct btrfs_trans_handle *trans,
2764 struct btrfs_path *path)
2766 struct btrfs_root *root = block_group->fs_info->tree_root;
2767 struct inode *inode = NULL;
2769 int dcs = BTRFS_DC_ERROR;
2775 * If this block group is smaller than 100 megs don't bother caching the
2778 if (block_group->key.offset < (100 * 1024 * 1024)) {
2779 spin_lock(&block_group->lock);
2780 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2781 spin_unlock(&block_group->lock);
2786 inode = lookup_free_space_inode(root, block_group, path);
2787 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2788 ret = PTR_ERR(inode);
2789 btrfs_release_path(root, path);
2793 if (IS_ERR(inode)) {
2797 if (block_group->ro)
2800 ret = create_free_space_inode(root, trans, block_group, path);
2807 * We want to set the generation to 0, that way if anything goes wrong
2808 * from here on out we know not to trust this cache when we load up next
2811 BTRFS_I(inode)->generation = 0;
2812 ret = btrfs_update_inode(trans, root, inode);
2815 if (i_size_read(inode) > 0) {
2816 ret = btrfs_truncate_free_space_cache(root, trans, path,
2822 spin_lock(&block_group->lock);
2823 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2824 /* We're not cached, don't bother trying to write stuff out */
2825 dcs = BTRFS_DC_WRITTEN;
2826 spin_unlock(&block_group->lock);
2829 spin_unlock(&block_group->lock);
2831 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2836 * Just to make absolutely sure we have enough space, we're going to
2837 * preallocate 12 pages worth of space for each block group. In
2838 * practice we ought to use at most 8, but we need extra space so we can
2839 * add our header and have a terminator between the extents and the
2843 num_pages *= PAGE_CACHE_SIZE;
2845 ret = btrfs_check_data_free_space(inode, num_pages);
2849 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2850 num_pages, num_pages,
2853 dcs = BTRFS_DC_SETUP;
2854 btrfs_free_reserved_data_space(inode, num_pages);
2858 btrfs_release_path(root, path);
2860 spin_lock(&block_group->lock);
2861 block_group->disk_cache_state = dcs;
2862 spin_unlock(&block_group->lock);
2867 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2868 struct btrfs_root *root)
2870 struct btrfs_block_group_cache *cache;
2872 struct btrfs_path *path;
2875 path = btrfs_alloc_path();
2881 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2883 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2885 cache = next_block_group(root, cache);
2893 err = cache_save_setup(cache, trans, path);
2894 last = cache->key.objectid + cache->key.offset;
2895 btrfs_put_block_group(cache);
2900 err = btrfs_run_delayed_refs(trans, root,
2905 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2907 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2908 btrfs_put_block_group(cache);
2914 cache = next_block_group(root, cache);
2923 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2924 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2926 last = cache->key.objectid + cache->key.offset;
2928 err = write_one_cache_group(trans, root, path, cache);
2930 btrfs_put_block_group(cache);
2935 * I don't think this is needed since we're just marking our
2936 * preallocated extent as written, but just in case it can't
2940 err = btrfs_run_delayed_refs(trans, root,
2945 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2948 * Really this shouldn't happen, but it could if we
2949 * couldn't write the entire preallocated extent and
2950 * splitting the extent resulted in a new block.
2953 btrfs_put_block_group(cache);
2956 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2958 cache = next_block_group(root, cache);
2967 btrfs_write_out_cache(root, trans, cache, path);
2970 * If we didn't have an error then the cache state is still
2971 * NEED_WRITE, so we can set it to WRITTEN.
2973 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2974 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2975 last = cache->key.objectid + cache->key.offset;
2976 btrfs_put_block_group(cache);
2979 btrfs_free_path(path);
2983 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2985 struct btrfs_block_group_cache *block_group;
2988 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2989 if (!block_group || block_group->ro)
2992 btrfs_put_block_group(block_group);
2996 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2997 u64 total_bytes, u64 bytes_used,
2998 struct btrfs_space_info **space_info)
3000 struct btrfs_space_info *found;
3004 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3005 BTRFS_BLOCK_GROUP_RAID10))
3010 found = __find_space_info(info, flags);
3012 spin_lock(&found->lock);
3013 found->total_bytes += total_bytes;
3014 found->disk_total += total_bytes * factor;
3015 found->bytes_used += bytes_used;
3016 found->disk_used += bytes_used * factor;
3018 spin_unlock(&found->lock);
3019 *space_info = found;
3022 found = kzalloc(sizeof(*found), GFP_NOFS);
3026 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3027 INIT_LIST_HEAD(&found->block_groups[i]);
3028 init_rwsem(&found->groups_sem);
3029 spin_lock_init(&found->lock);
3030 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3031 BTRFS_BLOCK_GROUP_SYSTEM |
3032 BTRFS_BLOCK_GROUP_METADATA);
3033 found->total_bytes = total_bytes;
3034 found->disk_total = total_bytes * factor;
3035 found->bytes_used = bytes_used;
3036 found->disk_used = bytes_used * factor;
3037 found->bytes_pinned = 0;
3038 found->bytes_reserved = 0;
3039 found->bytes_readonly = 0;
3040 found->bytes_may_use = 0;
3042 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3043 found->chunk_alloc = 0;
3044 *space_info = found;
3045 list_add_rcu(&found->list, &info->space_info);
3046 atomic_set(&found->caching_threads, 0);
3050 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3052 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3053 BTRFS_BLOCK_GROUP_RAID1 |
3054 BTRFS_BLOCK_GROUP_RAID10 |
3055 BTRFS_BLOCK_GROUP_DUP);
3057 if (flags & BTRFS_BLOCK_GROUP_DATA)
3058 fs_info->avail_data_alloc_bits |= extra_flags;
3059 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3060 fs_info->avail_metadata_alloc_bits |= extra_flags;
3061 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3062 fs_info->avail_system_alloc_bits |= extra_flags;
3066 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3069 * we add in the count of missing devices because we want
3070 * to make sure that any RAID levels on a degraded FS
3071 * continue to be honored.
3073 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3074 root->fs_info->fs_devices->missing_devices;
3076 if (num_devices == 1)
3077 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3078 if (num_devices < 4)
3079 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3081 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3082 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3083 BTRFS_BLOCK_GROUP_RAID10))) {
3084 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3087 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3088 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3089 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3092 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3093 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3094 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3095 (flags & BTRFS_BLOCK_GROUP_DUP)))
3096 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3100 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3102 if (flags & BTRFS_BLOCK_GROUP_DATA)
3103 flags |= root->fs_info->avail_data_alloc_bits &
3104 root->fs_info->data_alloc_profile;
3105 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3106 flags |= root->fs_info->avail_system_alloc_bits &
3107 root->fs_info->system_alloc_profile;
3108 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3109 flags |= root->fs_info->avail_metadata_alloc_bits &
3110 root->fs_info->metadata_alloc_profile;
3111 return btrfs_reduce_alloc_profile(root, flags);
3114 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3119 flags = BTRFS_BLOCK_GROUP_DATA;
3120 else if (root == root->fs_info->chunk_root)
3121 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3123 flags = BTRFS_BLOCK_GROUP_METADATA;
3125 return get_alloc_profile(root, flags);
3128 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3130 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3131 BTRFS_BLOCK_GROUP_DATA);
3135 * This will check the space that the inode allocates from to make sure we have
3136 * enough space for bytes.
3138 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3140 struct btrfs_space_info *data_sinfo;
3141 struct btrfs_root *root = BTRFS_I(inode)->root;
3143 int ret = 0, committed = 0, alloc_chunk = 1;
3145 /* make sure bytes are sectorsize aligned */
3146 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3148 if (root == root->fs_info->tree_root) {
3153 data_sinfo = BTRFS_I(inode)->space_info;
3158 /* make sure we have enough space to handle the data first */
3159 spin_lock(&data_sinfo->lock);
3160 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3161 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3162 data_sinfo->bytes_may_use;
3164 if (used + bytes > data_sinfo->total_bytes) {
3165 struct btrfs_trans_handle *trans;
3168 * if we don't have enough free bytes in this space then we need
3169 * to alloc a new chunk.
3171 if (!data_sinfo->full && alloc_chunk) {
3174 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3175 spin_unlock(&data_sinfo->lock);
3177 alloc_target = btrfs_get_alloc_profile(root, 1);
3178 trans = btrfs_join_transaction(root, 1);
3180 return PTR_ERR(trans);
3182 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3183 bytes + 2 * 1024 * 1024,
3185 CHUNK_ALLOC_NO_FORCE);
3186 btrfs_end_transaction(trans, root);
3195 btrfs_set_inode_space_info(root, inode);
3196 data_sinfo = BTRFS_I(inode)->space_info;
3200 spin_unlock(&data_sinfo->lock);
3202 /* commit the current transaction and try again */
3204 if (!committed && !root->fs_info->open_ioctl_trans) {
3206 trans = btrfs_join_transaction(root, 1);
3208 return PTR_ERR(trans);
3209 ret = btrfs_commit_transaction(trans, root);
3215 #if 0 /* I hope we never need this code again, just in case */
3216 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3217 "%llu bytes_reserved, " "%llu bytes_pinned, "
3218 "%llu bytes_readonly, %llu may use %llu total\n",
3219 (unsigned long long)bytes,
3220 (unsigned long long)data_sinfo->bytes_used,
3221 (unsigned long long)data_sinfo->bytes_reserved,
3222 (unsigned long long)data_sinfo->bytes_pinned,
3223 (unsigned long long)data_sinfo->bytes_readonly,
3224 (unsigned long long)data_sinfo->bytes_may_use,
3225 (unsigned long long)data_sinfo->total_bytes);
3229 data_sinfo->bytes_may_use += bytes;
3230 BTRFS_I(inode)->reserved_bytes += bytes;
3231 spin_unlock(&data_sinfo->lock);
3237 * called when we are clearing an delalloc extent from the
3238 * inode's io_tree or there was an error for whatever reason
3239 * after calling btrfs_check_data_free_space
3241 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3243 struct btrfs_root *root = BTRFS_I(inode)->root;
3244 struct btrfs_space_info *data_sinfo;
3246 /* make sure bytes are sectorsize aligned */
3247 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3249 data_sinfo = BTRFS_I(inode)->space_info;
3250 spin_lock(&data_sinfo->lock);
3251 data_sinfo->bytes_may_use -= bytes;
3252 BTRFS_I(inode)->reserved_bytes -= bytes;
3253 spin_unlock(&data_sinfo->lock);
3256 static void force_metadata_allocation(struct btrfs_fs_info *info)
3258 struct list_head *head = &info->space_info;
3259 struct btrfs_space_info *found;
3262 list_for_each_entry_rcu(found, head, list) {
3263 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3264 found->force_alloc = CHUNK_ALLOC_FORCE;
3269 static int should_alloc_chunk(struct btrfs_root *root,
3270 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3273 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3274 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3277 if (force == CHUNK_ALLOC_FORCE)
3281 * in limited mode, we want to have some free space up to
3282 * about 1% of the FS size.
3284 if (force == CHUNK_ALLOC_LIMITED) {
3285 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3286 thresh = max_t(u64, 64 * 1024 * 1024,
3287 div_factor_fine(thresh, 1));
3289 if (num_bytes - num_allocated < thresh)
3294 * we have two similar checks here, one based on percentage
3295 * and once based on a hard number of 256MB. The idea
3296 * is that if we have a good amount of free
3297 * room, don't allocate a chunk. A good mount is
3298 * less than 80% utilized of the chunks we have allocated,
3299 * or more than 256MB free
3301 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3304 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3307 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3309 /* 256MB or 5% of the FS */
3310 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3312 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3317 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3318 struct btrfs_root *extent_root, u64 alloc_bytes,
3319 u64 flags, int force)
3321 struct btrfs_space_info *space_info;
3322 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3323 int wait_for_alloc = 0;
3326 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3328 space_info = __find_space_info(extent_root->fs_info, flags);
3330 ret = update_space_info(extent_root->fs_info, flags,
3334 BUG_ON(!space_info);
3337 spin_lock(&space_info->lock);
3338 if (space_info->force_alloc)
3339 force = space_info->force_alloc;
3340 if (space_info->full) {
3341 spin_unlock(&space_info->lock);
3345 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3346 spin_unlock(&space_info->lock);
3348 } else if (space_info->chunk_alloc) {
3351 space_info->chunk_alloc = 1;
3354 spin_unlock(&space_info->lock);
3356 mutex_lock(&fs_info->chunk_mutex);
3359 * The chunk_mutex is held throughout the entirety of a chunk
3360 * allocation, so once we've acquired the chunk_mutex we know that the
3361 * other guy is done and we need to recheck and see if we should
3364 if (wait_for_alloc) {
3365 mutex_unlock(&fs_info->chunk_mutex);
3371 * If we have mixed data/metadata chunks we want to make sure we keep
3372 * allocating mixed chunks instead of individual chunks.
3374 if (btrfs_mixed_space_info(space_info))
3375 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3378 * if we're doing a data chunk, go ahead and make sure that
3379 * we keep a reasonable number of metadata chunks allocated in the
3382 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3383 fs_info->data_chunk_allocations++;
3384 if (!(fs_info->data_chunk_allocations %
3385 fs_info->metadata_ratio))
3386 force_metadata_allocation(fs_info);
3389 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3390 spin_lock(&space_info->lock);
3392 space_info->full = 1;
3396 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3397 space_info->chunk_alloc = 0;
3398 spin_unlock(&space_info->lock);
3399 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3404 * shrink metadata reservation for delalloc
3406 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3407 struct btrfs_root *root, u64 to_reclaim, int sync)
3409 struct btrfs_block_rsv *block_rsv;
3410 struct btrfs_space_info *space_info;
3415 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3417 unsigned long progress;
3419 block_rsv = &root->fs_info->delalloc_block_rsv;
3420 space_info = block_rsv->space_info;
3423 reserved = space_info->bytes_reserved;
3424 progress = space_info->reservation_progress;
3429 max_reclaim = min(reserved, to_reclaim);
3431 while (loops < 1024) {
3432 /* have the flusher threads jump in and do some IO */
3434 nr_pages = min_t(unsigned long, nr_pages,
3435 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3436 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3438 spin_lock(&space_info->lock);
3439 if (reserved > space_info->bytes_reserved)
3440 reclaimed += reserved - space_info->bytes_reserved;
3441 reserved = space_info->bytes_reserved;
3442 spin_unlock(&space_info->lock);
3446 if (reserved == 0 || reclaimed >= max_reclaim)
3449 if (trans && trans->transaction->blocked)
3452 time_left = schedule_timeout_interruptible(1);
3454 /* We were interrupted, exit */
3458 /* we've kicked the IO a few times, if anything has been freed,
3459 * exit. There is no sense in looping here for a long time
3460 * when we really need to commit the transaction, or there are
3461 * just too many writers without enough free space
3466 if (progress != space_info->reservation_progress)
3471 return reclaimed >= to_reclaim;
3475 * Retries tells us how many times we've called reserve_metadata_bytes. The
3476 * idea is if this is the first call (retries == 0) then we will add to our
3477 * reserved count if we can't make the allocation in order to hold our place
3478 * while we go and try and free up space. That way for retries > 1 we don't try
3479 * and add space, we just check to see if the amount of unused space is >= the
3480 * total space, meaning that our reservation is valid.
3482 * However if we don't intend to retry this reservation, pass -1 as retries so
3483 * that it short circuits this logic.
3485 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3486 struct btrfs_root *root,
3487 struct btrfs_block_rsv *block_rsv,
3488 u64 orig_bytes, int flush)
3490 struct btrfs_space_info *space_info = block_rsv->space_info;
3492 u64 num_bytes = orig_bytes;
3495 bool reserved = false;
3496 bool committed = false;
3503 spin_lock(&space_info->lock);
3504 unused = space_info->bytes_used + space_info->bytes_reserved +
3505 space_info->bytes_pinned + space_info->bytes_readonly +
3506 space_info->bytes_may_use;
3509 * The idea here is that we've not already over-reserved the block group
3510 * then we can go ahead and save our reservation first and then start
3511 * flushing if we need to. Otherwise if we've already overcommitted
3512 * lets start flushing stuff first and then come back and try to make
3515 if (unused <= space_info->total_bytes) {
3516 unused = space_info->total_bytes - unused;
3517 if (unused >= num_bytes) {
3519 space_info->bytes_reserved += orig_bytes;
3523 * Ok set num_bytes to orig_bytes since we aren't
3524 * overocmmitted, this way we only try and reclaim what
3527 num_bytes = orig_bytes;
3531 * Ok we're over committed, set num_bytes to the overcommitted
3532 * amount plus the amount of bytes that we need for this
3535 num_bytes = unused - space_info->total_bytes +
3536 (orig_bytes * (retries + 1));
3540 * Couldn't make our reservation, save our place so while we're trying
3541 * to reclaim space we can actually use it instead of somebody else
3542 * stealing it from us.
3544 if (ret && !reserved) {
3545 space_info->bytes_reserved += orig_bytes;
3549 spin_unlock(&space_info->lock);
3558 * We do synchronous shrinking since we don't actually unreserve
3559 * metadata until after the IO is completed.
3561 ret = shrink_delalloc(trans, root, num_bytes, 1);
3568 * So if we were overcommitted it's possible that somebody else flushed
3569 * out enough space and we simply didn't have enough space to reclaim,
3570 * so go back around and try again.
3577 spin_lock(&space_info->lock);
3579 * Not enough space to be reclaimed, don't bother committing the
3582 if (space_info->bytes_pinned < orig_bytes)
3584 spin_unlock(&space_info->lock);
3589 if (trans || committed)
3593 trans = btrfs_join_transaction(root, 1);
3596 ret = btrfs_commit_transaction(trans, root);
3605 spin_lock(&space_info->lock);
3606 space_info->bytes_reserved -= orig_bytes;
3607 spin_unlock(&space_info->lock);
3613 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3614 struct btrfs_root *root)
3616 struct btrfs_block_rsv *block_rsv;
3618 block_rsv = trans->block_rsv;
3620 block_rsv = root->block_rsv;
3623 block_rsv = &root->fs_info->empty_block_rsv;
3628 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3632 spin_lock(&block_rsv->lock);
3633 if (block_rsv->reserved >= num_bytes) {
3634 block_rsv->reserved -= num_bytes;
3635 if (block_rsv->reserved < block_rsv->size)
3636 block_rsv->full = 0;
3639 spin_unlock(&block_rsv->lock);
3643 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3644 u64 num_bytes, int update_size)
3646 spin_lock(&block_rsv->lock);
3647 block_rsv->reserved += num_bytes;
3649 block_rsv->size += num_bytes;
3650 else if (block_rsv->reserved >= block_rsv->size)
3651 block_rsv->full = 1;
3652 spin_unlock(&block_rsv->lock);
3655 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3656 struct btrfs_block_rsv *dest, u64 num_bytes)
3658 struct btrfs_space_info *space_info = block_rsv->space_info;
3660 spin_lock(&block_rsv->lock);
3661 if (num_bytes == (u64)-1)
3662 num_bytes = block_rsv->size;
3663 block_rsv->size -= num_bytes;
3664 if (block_rsv->reserved >= block_rsv->size) {
3665 num_bytes = block_rsv->reserved - block_rsv->size;
3666 block_rsv->reserved = block_rsv->size;
3667 block_rsv->full = 1;
3671 spin_unlock(&block_rsv->lock);
3673 if (num_bytes > 0) {
3675 spin_lock(&dest->lock);
3679 bytes_to_add = dest->size - dest->reserved;
3680 bytes_to_add = min(num_bytes, bytes_to_add);
3681 dest->reserved += bytes_to_add;
3682 if (dest->reserved >= dest->size)
3684 num_bytes -= bytes_to_add;
3686 spin_unlock(&dest->lock);
3689 spin_lock(&space_info->lock);
3690 space_info->bytes_reserved -= num_bytes;
3691 space_info->reservation_progress++;
3692 spin_unlock(&space_info->lock);
3697 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3698 struct btrfs_block_rsv *dst, u64 num_bytes)
3702 ret = block_rsv_use_bytes(src, num_bytes);
3706 block_rsv_add_bytes(dst, num_bytes, 1);
3710 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3712 memset(rsv, 0, sizeof(*rsv));
3713 spin_lock_init(&rsv->lock);
3714 atomic_set(&rsv->usage, 1);
3716 INIT_LIST_HEAD(&rsv->list);
3719 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3721 struct btrfs_block_rsv *block_rsv;
3722 struct btrfs_fs_info *fs_info = root->fs_info;
3724 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3728 btrfs_init_block_rsv(block_rsv);
3729 block_rsv->space_info = __find_space_info(fs_info,
3730 BTRFS_BLOCK_GROUP_METADATA);
3734 void btrfs_free_block_rsv(struct btrfs_root *root,
3735 struct btrfs_block_rsv *rsv)
3737 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3738 btrfs_block_rsv_release(root, rsv, (u64)-1);
3745 * make the block_rsv struct be able to capture freed space.
3746 * the captured space will re-add to the the block_rsv struct
3747 * after transaction commit
3749 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3750 struct btrfs_block_rsv *block_rsv)
3752 block_rsv->durable = 1;
3753 mutex_lock(&fs_info->durable_block_rsv_mutex);
3754 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3755 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3758 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3759 struct btrfs_root *root,
3760 struct btrfs_block_rsv *block_rsv,
3768 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3770 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3777 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3778 struct btrfs_root *root,
3779 struct btrfs_block_rsv *block_rsv,
3780 u64 min_reserved, int min_factor)
3783 int commit_trans = 0;
3789 spin_lock(&block_rsv->lock);
3791 num_bytes = div_factor(block_rsv->size, min_factor);
3792 if (min_reserved > num_bytes)
3793 num_bytes = min_reserved;
3795 if (block_rsv->reserved >= num_bytes) {
3798 num_bytes -= block_rsv->reserved;
3799 if (block_rsv->durable &&
3800 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3803 spin_unlock(&block_rsv->lock);
3807 if (block_rsv->refill_used) {
3808 ret = reserve_metadata_bytes(trans, root, block_rsv,
3811 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3820 trans = btrfs_join_transaction(root, 1);
3821 BUG_ON(IS_ERR(trans));
3822 ret = btrfs_commit_transaction(trans, root);
3829 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3830 struct btrfs_block_rsv *dst_rsv,
3833 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3836 void btrfs_block_rsv_release(struct btrfs_root *root,
3837 struct btrfs_block_rsv *block_rsv,
3840 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3841 if (global_rsv->full || global_rsv == block_rsv ||
3842 block_rsv->space_info != global_rsv->space_info)
3844 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3848 * helper to calculate size of global block reservation.
3849 * the desired value is sum of space used by extent tree,
3850 * checksum tree and root tree
3852 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3854 struct btrfs_space_info *sinfo;
3858 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3861 * per tree used space accounting can be inaccuracy, so we
3864 spin_lock(&fs_info->extent_root->accounting_lock);
3865 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3866 spin_unlock(&fs_info->extent_root->accounting_lock);
3868 spin_lock(&fs_info->csum_root->accounting_lock);
3869 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3870 spin_unlock(&fs_info->csum_root->accounting_lock);
3872 spin_lock(&fs_info->tree_root->accounting_lock);
3873 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3874 spin_unlock(&fs_info->tree_root->accounting_lock);
3876 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3877 spin_lock(&sinfo->lock);
3878 data_used = sinfo->bytes_used;
3879 spin_unlock(&sinfo->lock);
3881 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3882 spin_lock(&sinfo->lock);
3883 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3885 meta_used = sinfo->bytes_used;
3886 spin_unlock(&sinfo->lock);
3888 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3890 num_bytes += div64_u64(data_used + meta_used, 50);
3892 if (num_bytes * 3 > meta_used)
3893 num_bytes = div64_u64(meta_used, 3);
3895 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3898 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3900 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3901 struct btrfs_space_info *sinfo = block_rsv->space_info;
3904 num_bytes = calc_global_metadata_size(fs_info);
3906 spin_lock(&block_rsv->lock);
3907 spin_lock(&sinfo->lock);
3909 block_rsv->size = num_bytes;
3911 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3912 sinfo->bytes_reserved + sinfo->bytes_readonly +
3913 sinfo->bytes_may_use;
3915 if (sinfo->total_bytes > num_bytes) {
3916 num_bytes = sinfo->total_bytes - num_bytes;
3917 block_rsv->reserved += num_bytes;
3918 sinfo->bytes_reserved += num_bytes;
3921 if (block_rsv->reserved >= block_rsv->size) {
3922 num_bytes = block_rsv->reserved - block_rsv->size;
3923 sinfo->bytes_reserved -= num_bytes;
3924 sinfo->reservation_progress++;
3925 block_rsv->reserved = block_rsv->size;
3926 block_rsv->full = 1;
3929 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3930 block_rsv->size, block_rsv->reserved);
3932 spin_unlock(&sinfo->lock);
3933 spin_unlock(&block_rsv->lock);
3936 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3938 struct btrfs_space_info *space_info;
3940 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3941 fs_info->chunk_block_rsv.space_info = space_info;
3942 fs_info->chunk_block_rsv.priority = 10;
3944 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3945 fs_info->global_block_rsv.space_info = space_info;
3946 fs_info->global_block_rsv.priority = 10;
3947 fs_info->global_block_rsv.refill_used = 1;
3948 fs_info->delalloc_block_rsv.space_info = space_info;
3949 fs_info->trans_block_rsv.space_info = space_info;
3950 fs_info->empty_block_rsv.space_info = space_info;
3951 fs_info->empty_block_rsv.priority = 10;
3953 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3954 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3955 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3956 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3957 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3959 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3961 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3963 update_global_block_rsv(fs_info);
3966 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3968 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3969 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3970 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3971 WARN_ON(fs_info->trans_block_rsv.size > 0);
3972 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3973 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3974 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3977 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3979 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3983 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3984 struct btrfs_root *root,
3990 if (num_items == 0 || root->fs_info->chunk_root == root)
3993 num_bytes = calc_trans_metadata_size(root, num_items);
3994 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3997 trans->bytes_reserved += num_bytes;
3998 trans->block_rsv = &root->fs_info->trans_block_rsv;
4003 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4004 struct btrfs_root *root)
4006 if (!trans->bytes_reserved)
4009 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4010 btrfs_block_rsv_release(root, trans->block_rsv,
4011 trans->bytes_reserved);
4012 trans->bytes_reserved = 0;
4015 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4016 struct inode *inode)
4018 struct btrfs_root *root = BTRFS_I(inode)->root;
4019 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4020 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4023 * one for deleting orphan item, one for updating inode and
4024 * two for calling btrfs_truncate_inode_items.
4026 * btrfs_truncate_inode_items is a delete operation, it frees
4027 * more space than it uses in most cases. So two units of
4028 * metadata space should be enough for calling it many times.
4029 * If all of the metadata space is used, we can commit
4030 * transaction and use space it freed.
4032 u64 num_bytes = calc_trans_metadata_size(root, 4);
4033 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4036 void btrfs_orphan_release_metadata(struct inode *inode)
4038 struct btrfs_root *root = BTRFS_I(inode)->root;
4039 u64 num_bytes = calc_trans_metadata_size(root, 4);
4040 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4043 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4044 struct btrfs_pending_snapshot *pending)
4046 struct btrfs_root *root = pending->root;
4047 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4048 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4050 * two for root back/forward refs, two for directory entries
4051 * and one for root of the snapshot.
4053 u64 num_bytes = calc_trans_metadata_size(root, 5);
4054 dst_rsv->space_info = src_rsv->space_info;
4055 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4058 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4060 return num_bytes >>= 3;
4063 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4065 struct btrfs_root *root = BTRFS_I(inode)->root;
4066 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4069 int reserved_extents;
4072 if (btrfs_transaction_in_commit(root->fs_info))
4073 schedule_timeout(1);
4075 num_bytes = ALIGN(num_bytes, root->sectorsize);
4077 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4078 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4080 if (nr_extents > reserved_extents) {
4081 nr_extents -= reserved_extents;
4082 to_reserve = calc_trans_metadata_size(root, nr_extents);
4088 to_reserve += calc_csum_metadata_size(inode, num_bytes);
4089 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4093 atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4094 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4096 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4098 if (block_rsv->size > 512 * 1024 * 1024)
4099 shrink_delalloc(NULL, root, to_reserve, 0);
4104 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4106 struct btrfs_root *root = BTRFS_I(inode)->root;
4109 int reserved_extents;
4111 num_bytes = ALIGN(num_bytes, root->sectorsize);
4112 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4113 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4115 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4119 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4120 if (nr_extents >= reserved_extents) {
4124 old = reserved_extents;
4125 nr_extents = reserved_extents - nr_extents;
4126 new = reserved_extents - nr_extents;
4127 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4128 reserved_extents, new);
4129 if (likely(old == reserved_extents))
4131 reserved_extents = old;
4134 to_free = calc_csum_metadata_size(inode, num_bytes);
4136 to_free += calc_trans_metadata_size(root, nr_extents);
4138 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4142 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4146 ret = btrfs_check_data_free_space(inode, num_bytes);
4150 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4152 btrfs_free_reserved_data_space(inode, num_bytes);
4159 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4161 btrfs_delalloc_release_metadata(inode, num_bytes);
4162 btrfs_free_reserved_data_space(inode, num_bytes);
4165 static int update_block_group(struct btrfs_trans_handle *trans,
4166 struct btrfs_root *root,
4167 u64 bytenr, u64 num_bytes, int alloc)
4169 struct btrfs_block_group_cache *cache = NULL;
4170 struct btrfs_fs_info *info = root->fs_info;
4171 u64 total = num_bytes;
4176 /* block accounting for super block */
4177 spin_lock(&info->delalloc_lock);
4178 old_val = btrfs_super_bytes_used(&info->super_copy);
4180 old_val += num_bytes;
4182 old_val -= num_bytes;
4183 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4184 spin_unlock(&info->delalloc_lock);
4187 cache = btrfs_lookup_block_group(info, bytenr);
4190 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4191 BTRFS_BLOCK_GROUP_RAID1 |
4192 BTRFS_BLOCK_GROUP_RAID10))
4197 * If this block group has free space cache written out, we
4198 * need to make sure to load it if we are removing space. This
4199 * is because we need the unpinning stage to actually add the
4200 * space back to the block group, otherwise we will leak space.
4202 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4203 cache_block_group(cache, trans, NULL, 1);
4205 byte_in_group = bytenr - cache->key.objectid;
4206 WARN_ON(byte_in_group > cache->key.offset);
4208 spin_lock(&cache->space_info->lock);
4209 spin_lock(&cache->lock);
4211 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4212 cache->disk_cache_state < BTRFS_DC_CLEAR)
4213 cache->disk_cache_state = BTRFS_DC_CLEAR;
4216 old_val = btrfs_block_group_used(&cache->item);
4217 num_bytes = min(total, cache->key.offset - byte_in_group);
4219 old_val += num_bytes;
4220 btrfs_set_block_group_used(&cache->item, old_val);
4221 cache->reserved -= num_bytes;
4222 cache->space_info->bytes_reserved -= num_bytes;
4223 cache->space_info->reservation_progress++;
4224 cache->space_info->bytes_used += num_bytes;
4225 cache->space_info->disk_used += num_bytes * factor;
4226 spin_unlock(&cache->lock);
4227 spin_unlock(&cache->space_info->lock);
4229 old_val -= num_bytes;
4230 btrfs_set_block_group_used(&cache->item, old_val);
4231 cache->pinned += num_bytes;
4232 cache->space_info->bytes_pinned += num_bytes;
4233 cache->space_info->bytes_used -= num_bytes;
4234 cache->space_info->disk_used -= num_bytes * factor;
4235 spin_unlock(&cache->lock);
4236 spin_unlock(&cache->space_info->lock);
4238 set_extent_dirty(info->pinned_extents,
4239 bytenr, bytenr + num_bytes - 1,
4240 GFP_NOFS | __GFP_NOFAIL);
4242 btrfs_put_block_group(cache);
4244 bytenr += num_bytes;
4249 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4251 struct btrfs_block_group_cache *cache;
4254 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4258 bytenr = cache->key.objectid;
4259 btrfs_put_block_group(cache);
4264 static int pin_down_extent(struct btrfs_root *root,
4265 struct btrfs_block_group_cache *cache,
4266 u64 bytenr, u64 num_bytes, int reserved)
4268 spin_lock(&cache->space_info->lock);
4269 spin_lock(&cache->lock);
4270 cache->pinned += num_bytes;
4271 cache->space_info->bytes_pinned += num_bytes;
4273 cache->reserved -= num_bytes;
4274 cache->space_info->bytes_reserved -= num_bytes;
4275 cache->space_info->reservation_progress++;
4277 spin_unlock(&cache->lock);
4278 spin_unlock(&cache->space_info->lock);
4280 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4281 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4286 * this function must be called within transaction
4288 int btrfs_pin_extent(struct btrfs_root *root,
4289 u64 bytenr, u64 num_bytes, int reserved)
4291 struct btrfs_block_group_cache *cache;
4293 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4296 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4298 btrfs_put_block_group(cache);
4303 * update size of reserved extents. this function may return -EAGAIN
4304 * if 'reserve' is true or 'sinfo' is false.
4306 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4307 u64 num_bytes, int reserve, int sinfo)
4311 struct btrfs_space_info *space_info = cache->space_info;
4312 spin_lock(&space_info->lock);
4313 spin_lock(&cache->lock);
4318 cache->reserved += num_bytes;
4319 space_info->bytes_reserved += num_bytes;
4323 space_info->bytes_readonly += num_bytes;
4324 cache->reserved -= num_bytes;
4325 space_info->bytes_reserved -= num_bytes;
4326 space_info->reservation_progress++;
4328 spin_unlock(&cache->lock);
4329 spin_unlock(&space_info->lock);
4331 spin_lock(&cache->lock);
4336 cache->reserved += num_bytes;
4338 cache->reserved -= num_bytes;
4340 spin_unlock(&cache->lock);
4345 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4346 struct btrfs_root *root)
4348 struct btrfs_fs_info *fs_info = root->fs_info;
4349 struct btrfs_caching_control *next;
4350 struct btrfs_caching_control *caching_ctl;
4351 struct btrfs_block_group_cache *cache;
4353 down_write(&fs_info->extent_commit_sem);
4355 list_for_each_entry_safe(caching_ctl, next,
4356 &fs_info->caching_block_groups, list) {
4357 cache = caching_ctl->block_group;
4358 if (block_group_cache_done(cache)) {
4359 cache->last_byte_to_unpin = (u64)-1;
4360 list_del_init(&caching_ctl->list);
4361 put_caching_control(caching_ctl);
4363 cache->last_byte_to_unpin = caching_ctl->progress;
4367 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4368 fs_info->pinned_extents = &fs_info->freed_extents[1];
4370 fs_info->pinned_extents = &fs_info->freed_extents[0];
4372 up_write(&fs_info->extent_commit_sem);
4374 update_global_block_rsv(fs_info);
4378 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4380 struct btrfs_fs_info *fs_info = root->fs_info;
4381 struct btrfs_block_group_cache *cache = NULL;
4384 while (start <= end) {
4386 start >= cache->key.objectid + cache->key.offset) {
4388 btrfs_put_block_group(cache);
4389 cache = btrfs_lookup_block_group(fs_info, start);
4393 len = cache->key.objectid + cache->key.offset - start;
4394 len = min(len, end + 1 - start);
4396 if (start < cache->last_byte_to_unpin) {
4397 len = min(len, cache->last_byte_to_unpin - start);
4398 btrfs_add_free_space(cache, start, len);
4403 spin_lock(&cache->space_info->lock);
4404 spin_lock(&cache->lock);
4405 cache->pinned -= len;
4406 cache->space_info->bytes_pinned -= len;
4408 cache->space_info->bytes_readonly += len;
4409 } else if (cache->reserved_pinned > 0) {
4410 len = min(len, cache->reserved_pinned);
4411 cache->reserved_pinned -= len;
4412 cache->space_info->bytes_reserved += len;
4414 spin_unlock(&cache->lock);
4415 spin_unlock(&cache->space_info->lock);
4419 btrfs_put_block_group(cache);
4423 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root)
4426 struct btrfs_fs_info *fs_info = root->fs_info;
4427 struct extent_io_tree *unpin;
4428 struct btrfs_block_rsv *block_rsv;
4429 struct btrfs_block_rsv *next_rsv;
4435 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4436 unpin = &fs_info->freed_extents[1];
4438 unpin = &fs_info->freed_extents[0];
4441 ret = find_first_extent_bit(unpin, 0, &start, &end,
4446 if (btrfs_test_opt(root, DISCARD))
4447 ret = btrfs_discard_extent(root, start,
4448 end + 1 - start, NULL);
4450 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4451 unpin_extent_range(root, start, end);
4455 mutex_lock(&fs_info->durable_block_rsv_mutex);
4456 list_for_each_entry_safe(block_rsv, next_rsv,
4457 &fs_info->durable_block_rsv_list, list) {
4459 idx = trans->transid & 0x1;
4460 if (block_rsv->freed[idx] > 0) {
4461 block_rsv_add_bytes(block_rsv,
4462 block_rsv->freed[idx], 0);
4463 block_rsv->freed[idx] = 0;
4465 if (atomic_read(&block_rsv->usage) == 0) {
4466 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4468 if (block_rsv->freed[0] == 0 &&
4469 block_rsv->freed[1] == 0) {
4470 list_del_init(&block_rsv->list);
4474 btrfs_block_rsv_release(root, block_rsv, 0);
4477 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4482 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4483 struct btrfs_root *root,
4484 u64 bytenr, u64 num_bytes, u64 parent,
4485 u64 root_objectid, u64 owner_objectid,
4486 u64 owner_offset, int refs_to_drop,
4487 struct btrfs_delayed_extent_op *extent_op)
4489 struct btrfs_key key;
4490 struct btrfs_path *path;
4491 struct btrfs_fs_info *info = root->fs_info;
4492 struct btrfs_root *extent_root = info->extent_root;
4493 struct extent_buffer *leaf;
4494 struct btrfs_extent_item *ei;
4495 struct btrfs_extent_inline_ref *iref;
4498 int extent_slot = 0;
4499 int found_extent = 0;
4504 path = btrfs_alloc_path();
4509 path->leave_spinning = 1;
4511 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4512 BUG_ON(!is_data && refs_to_drop != 1);
4514 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4515 bytenr, num_bytes, parent,
4516 root_objectid, owner_objectid,
4519 extent_slot = path->slots[0];
4520 while (extent_slot >= 0) {
4521 btrfs_item_key_to_cpu(path->nodes[0], &key,
4523 if (key.objectid != bytenr)
4525 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4526 key.offset == num_bytes) {
4530 if (path->slots[0] - extent_slot > 5)
4534 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4535 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4536 if (found_extent && item_size < sizeof(*ei))
4539 if (!found_extent) {
4541 ret = remove_extent_backref(trans, extent_root, path,
4545 btrfs_release_path(extent_root, path);
4546 path->leave_spinning = 1;
4548 key.objectid = bytenr;
4549 key.type = BTRFS_EXTENT_ITEM_KEY;
4550 key.offset = num_bytes;
4552 ret = btrfs_search_slot(trans, extent_root,
4555 printk(KERN_ERR "umm, got %d back from search"
4556 ", was looking for %llu\n", ret,
4557 (unsigned long long)bytenr);
4558 btrfs_print_leaf(extent_root, path->nodes[0]);
4561 extent_slot = path->slots[0];
4564 btrfs_print_leaf(extent_root, path->nodes[0]);
4566 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4567 "parent %llu root %llu owner %llu offset %llu\n",
4568 (unsigned long long)bytenr,
4569 (unsigned long long)parent,
4570 (unsigned long long)root_objectid,
4571 (unsigned long long)owner_objectid,
4572 (unsigned long long)owner_offset);
4575 leaf = path->nodes[0];
4576 item_size = btrfs_item_size_nr(leaf, extent_slot);
4577 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4578 if (item_size < sizeof(*ei)) {
4579 BUG_ON(found_extent || extent_slot != path->slots[0]);
4580 ret = convert_extent_item_v0(trans, extent_root, path,
4584 btrfs_release_path(extent_root, path);
4585 path->leave_spinning = 1;
4587 key.objectid = bytenr;
4588 key.type = BTRFS_EXTENT_ITEM_KEY;
4589 key.offset = num_bytes;
4591 ret = btrfs_search_slot(trans, extent_root, &key, path,
4594 printk(KERN_ERR "umm, got %d back from search"
4595 ", was looking for %llu\n", ret,
4596 (unsigned long long)bytenr);
4597 btrfs_print_leaf(extent_root, path->nodes[0]);
4600 extent_slot = path->slots[0];
4601 leaf = path->nodes[0];
4602 item_size = btrfs_item_size_nr(leaf, extent_slot);
4605 BUG_ON(item_size < sizeof(*ei));
4606 ei = btrfs_item_ptr(leaf, extent_slot,
4607 struct btrfs_extent_item);
4608 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4609 struct btrfs_tree_block_info *bi;
4610 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4611 bi = (struct btrfs_tree_block_info *)(ei + 1);
4612 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4615 refs = btrfs_extent_refs(leaf, ei);
4616 BUG_ON(refs < refs_to_drop);
4617 refs -= refs_to_drop;
4621 __run_delayed_extent_op(extent_op, leaf, ei);
4623 * In the case of inline back ref, reference count will
4624 * be updated by remove_extent_backref
4627 BUG_ON(!found_extent);
4629 btrfs_set_extent_refs(leaf, ei, refs);
4630 btrfs_mark_buffer_dirty(leaf);
4633 ret = remove_extent_backref(trans, extent_root, path,
4640 BUG_ON(is_data && refs_to_drop !=
4641 extent_data_ref_count(root, path, iref));
4643 BUG_ON(path->slots[0] != extent_slot);
4645 BUG_ON(path->slots[0] != extent_slot + 1);
4646 path->slots[0] = extent_slot;
4651 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4654 btrfs_release_path(extent_root, path);
4657 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4660 invalidate_mapping_pages(info->btree_inode->i_mapping,
4661 bytenr >> PAGE_CACHE_SHIFT,
4662 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4665 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4668 btrfs_free_path(path);
4673 * when we free an block, it is possible (and likely) that we free the last
4674 * delayed ref for that extent as well. This searches the delayed ref tree for
4675 * a given extent, and if there are no other delayed refs to be processed, it
4676 * removes it from the tree.
4678 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4679 struct btrfs_root *root, u64 bytenr)
4681 struct btrfs_delayed_ref_head *head;
4682 struct btrfs_delayed_ref_root *delayed_refs;
4683 struct btrfs_delayed_ref_node *ref;
4684 struct rb_node *node;
4687 delayed_refs = &trans->transaction->delayed_refs;
4688 spin_lock(&delayed_refs->lock);
4689 head = btrfs_find_delayed_ref_head(trans, bytenr);
4693 node = rb_prev(&head->node.rb_node);
4697 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4699 /* there are still entries for this ref, we can't drop it */
4700 if (ref->bytenr == bytenr)
4703 if (head->extent_op) {
4704 if (!head->must_insert_reserved)
4706 kfree(head->extent_op);
4707 head->extent_op = NULL;
4711 * waiting for the lock here would deadlock. If someone else has it
4712 * locked they are already in the process of dropping it anyway
4714 if (!mutex_trylock(&head->mutex))
4718 * at this point we have a head with no other entries. Go
4719 * ahead and process it.
4721 head->node.in_tree = 0;
4722 rb_erase(&head->node.rb_node, &delayed_refs->root);
4724 delayed_refs->num_entries--;
4727 * we don't take a ref on the node because we're removing it from the
4728 * tree, so we just steal the ref the tree was holding.
4730 delayed_refs->num_heads--;
4731 if (list_empty(&head->cluster))
4732 delayed_refs->num_heads_ready--;
4734 list_del_init(&head->cluster);
4735 spin_unlock(&delayed_refs->lock);
4737 BUG_ON(head->extent_op);
4738 if (head->must_insert_reserved)
4741 mutex_unlock(&head->mutex);
4742 btrfs_put_delayed_ref(&head->node);
4745 spin_unlock(&delayed_refs->lock);
4749 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4750 struct btrfs_root *root,
4751 struct extent_buffer *buf,
4752 u64 parent, int last_ref)
4754 struct btrfs_block_rsv *block_rsv;
4755 struct btrfs_block_group_cache *cache = NULL;
4758 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4759 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4760 parent, root->root_key.objectid,
4761 btrfs_header_level(buf),
4762 BTRFS_DROP_DELAYED_REF, NULL);
4769 block_rsv = get_block_rsv(trans, root);
4770 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4771 if (block_rsv->space_info != cache->space_info)
4774 if (btrfs_header_generation(buf) == trans->transid) {
4775 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4776 ret = check_ref_cleanup(trans, root, buf->start);
4781 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4782 pin_down_extent(root, cache, buf->start, buf->len, 1);
4786 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4788 btrfs_add_free_space(cache, buf->start, buf->len);
4789 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4790 if (ret == -EAGAIN) {
4791 /* block group became read-only */
4792 btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4797 spin_lock(&block_rsv->lock);
4798 if (block_rsv->reserved < block_rsv->size) {
4799 block_rsv->reserved += buf->len;
4802 spin_unlock(&block_rsv->lock);
4805 spin_lock(&cache->space_info->lock);
4806 cache->space_info->bytes_reserved -= buf->len;
4807 cache->space_info->reservation_progress++;
4808 spin_unlock(&cache->space_info->lock);
4813 if (block_rsv->durable && !cache->ro) {
4815 spin_lock(&cache->lock);
4817 cache->reserved_pinned += buf->len;
4820 spin_unlock(&cache->lock);
4823 spin_lock(&block_rsv->lock);
4824 block_rsv->freed[trans->transid & 0x1] += buf->len;
4825 spin_unlock(&block_rsv->lock);
4830 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4833 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4834 btrfs_put_block_group(cache);
4837 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4838 struct btrfs_root *root,
4839 u64 bytenr, u64 num_bytes, u64 parent,
4840 u64 root_objectid, u64 owner, u64 offset)
4845 * tree log blocks never actually go into the extent allocation
4846 * tree, just update pinning info and exit early.
4848 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4849 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4850 /* unlocks the pinned mutex */
4851 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4853 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4854 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4855 parent, root_objectid, (int)owner,
4856 BTRFS_DROP_DELAYED_REF, NULL);
4859 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4860 parent, root_objectid, owner,
4861 offset, BTRFS_DROP_DELAYED_REF, NULL);
4867 static u64 stripe_align(struct btrfs_root *root, u64 val)
4869 u64 mask = ((u64)root->stripesize - 1);
4870 u64 ret = (val + mask) & ~mask;
4875 * when we wait for progress in the block group caching, its because
4876 * our allocation attempt failed at least once. So, we must sleep
4877 * and let some progress happen before we try again.
4879 * This function will sleep at least once waiting for new free space to
4880 * show up, and then it will check the block group free space numbers
4881 * for our min num_bytes. Another option is to have it go ahead
4882 * and look in the rbtree for a free extent of a given size, but this
4886 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4889 struct btrfs_caching_control *caching_ctl;
4892 caching_ctl = get_caching_control(cache);
4896 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4897 (cache->free_space_ctl->free_space >= num_bytes));
4899 put_caching_control(caching_ctl);
4904 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4906 struct btrfs_caching_control *caching_ctl;
4909 caching_ctl = get_caching_control(cache);
4913 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4915 put_caching_control(caching_ctl);
4919 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4922 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4924 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4926 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4928 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4935 enum btrfs_loop_type {
4936 LOOP_FIND_IDEAL = 0,
4937 LOOP_CACHING_NOWAIT = 1,
4938 LOOP_CACHING_WAIT = 2,
4939 LOOP_ALLOC_CHUNK = 3,
4940 LOOP_NO_EMPTY_SIZE = 4,
4944 * walks the btree of allocated extents and find a hole of a given size.
4945 * The key ins is changed to record the hole:
4946 * ins->objectid == block start
4947 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4948 * ins->offset == number of blocks
4949 * Any available blocks before search_start are skipped.
4951 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4952 struct btrfs_root *orig_root,
4953 u64 num_bytes, u64 empty_size,
4954 u64 search_start, u64 search_end,
4955 u64 hint_byte, struct btrfs_key *ins,
4959 struct btrfs_root *root = orig_root->fs_info->extent_root;
4960 struct btrfs_free_cluster *last_ptr = NULL;
4961 struct btrfs_block_group_cache *block_group = NULL;
4962 int empty_cluster = 2 * 1024 * 1024;
4963 int allowed_chunk_alloc = 0;
4964 int done_chunk_alloc = 0;
4965 struct btrfs_space_info *space_info;
4966 int last_ptr_loop = 0;
4969 bool found_uncached_bg = false;
4970 bool failed_cluster_refill = false;
4971 bool failed_alloc = false;
4972 bool use_cluster = true;
4973 u64 ideal_cache_percent = 0;
4974 u64 ideal_cache_offset = 0;
4976 WARN_ON(num_bytes < root->sectorsize);
4977 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4981 space_info = __find_space_info(root->fs_info, data);
4983 printk(KERN_ERR "No space info for %d\n", data);
4988 * If the space info is for both data and metadata it means we have a
4989 * small filesystem and we can't use the clustering stuff.
4991 if (btrfs_mixed_space_info(space_info))
4992 use_cluster = false;
4994 if (orig_root->ref_cows || empty_size)
4995 allowed_chunk_alloc = 1;
4997 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4998 last_ptr = &root->fs_info->meta_alloc_cluster;
4999 if (!btrfs_test_opt(root, SSD))
5000 empty_cluster = 64 * 1024;
5003 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5004 btrfs_test_opt(root, SSD)) {
5005 last_ptr = &root->fs_info->data_alloc_cluster;
5009 spin_lock(&last_ptr->lock);
5010 if (last_ptr->block_group)
5011 hint_byte = last_ptr->window_start;
5012 spin_unlock(&last_ptr->lock);
5015 search_start = max(search_start, first_logical_byte(root, 0));
5016 search_start = max(search_start, hint_byte);
5021 if (search_start == hint_byte) {
5023 block_group = btrfs_lookup_block_group(root->fs_info,
5026 * we don't want to use the block group if it doesn't match our
5027 * allocation bits, or if its not cached.
5029 * However if we are re-searching with an ideal block group
5030 * picked out then we don't care that the block group is cached.
5032 if (block_group && block_group_bits(block_group, data) &&
5033 (block_group->cached != BTRFS_CACHE_NO ||
5034 search_start == ideal_cache_offset)) {
5035 down_read(&space_info->groups_sem);
5036 if (list_empty(&block_group->list) ||
5039 * someone is removing this block group,
5040 * we can't jump into the have_block_group
5041 * target because our list pointers are not
5044 btrfs_put_block_group(block_group);
5045 up_read(&space_info->groups_sem);
5047 index = get_block_group_index(block_group);
5048 goto have_block_group;
5050 } else if (block_group) {
5051 btrfs_put_block_group(block_group);
5055 down_read(&space_info->groups_sem);
5056 list_for_each_entry(block_group, &space_info->block_groups[index],
5061 btrfs_get_block_group(block_group);
5062 search_start = block_group->key.objectid;
5065 * this can happen if we end up cycling through all the
5066 * raid types, but we want to make sure we only allocate
5067 * for the proper type.
5069 if (!block_group_bits(block_group, data)) {
5070 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5071 BTRFS_BLOCK_GROUP_RAID1 |
5072 BTRFS_BLOCK_GROUP_RAID10;
5075 * if they asked for extra copies and this block group
5076 * doesn't provide them, bail. This does allow us to
5077 * fill raid0 from raid1.
5079 if ((data & extra) && !(block_group->flags & extra))
5084 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5087 ret = cache_block_group(block_group, trans,
5089 if (block_group->cached == BTRFS_CACHE_FINISHED)
5090 goto have_block_group;
5092 free_percent = btrfs_block_group_used(&block_group->item);
5093 free_percent *= 100;
5094 free_percent = div64_u64(free_percent,
5095 block_group->key.offset);
5096 free_percent = 100 - free_percent;
5097 if (free_percent > ideal_cache_percent &&
5098 likely(!block_group->ro)) {
5099 ideal_cache_offset = block_group->key.objectid;
5100 ideal_cache_percent = free_percent;
5104 * We only want to start kthread caching if we are at
5105 * the point where we will wait for caching to make
5106 * progress, or if our ideal search is over and we've
5107 * found somebody to start caching.
5109 if (loop > LOOP_CACHING_NOWAIT ||
5110 (loop > LOOP_FIND_IDEAL &&
5111 atomic_read(&space_info->caching_threads) < 2)) {
5112 ret = cache_block_group(block_group, trans,
5116 found_uncached_bg = true;
5119 * If loop is set for cached only, try the next block
5122 if (loop == LOOP_FIND_IDEAL)
5126 cached = block_group_cache_done(block_group);
5127 if (unlikely(!cached))
5128 found_uncached_bg = true;
5130 if (unlikely(block_group->ro))
5134 * Ok we want to try and use the cluster allocator, so lets look
5135 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5136 * have tried the cluster allocator plenty of times at this
5137 * point and not have found anything, so we are likely way too
5138 * fragmented for the clustering stuff to find anything, so lets
5139 * just skip it and let the allocator find whatever block it can
5142 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5144 * the refill lock keeps out other
5145 * people trying to start a new cluster
5147 spin_lock(&last_ptr->refill_lock);
5148 if (last_ptr->block_group &&
5149 (last_ptr->block_group->ro ||
5150 !block_group_bits(last_ptr->block_group, data))) {
5152 goto refill_cluster;
5155 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5156 num_bytes, search_start);
5158 /* we have a block, we're done */
5159 spin_unlock(&last_ptr->refill_lock);
5163 spin_lock(&last_ptr->lock);
5165 * whoops, this cluster doesn't actually point to
5166 * this block group. Get a ref on the block
5167 * group is does point to and try again
5169 if (!last_ptr_loop && last_ptr->block_group &&
5170 last_ptr->block_group != block_group) {
5172 btrfs_put_block_group(block_group);
5173 block_group = last_ptr->block_group;
5174 btrfs_get_block_group(block_group);
5175 spin_unlock(&last_ptr->lock);
5176 spin_unlock(&last_ptr->refill_lock);
5179 search_start = block_group->key.objectid;
5181 * we know this block group is properly
5182 * in the list because
5183 * btrfs_remove_block_group, drops the
5184 * cluster before it removes the block
5185 * group from the list
5187 goto have_block_group;
5189 spin_unlock(&last_ptr->lock);
5192 * this cluster didn't work out, free it and
5195 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5199 /* allocate a cluster in this block group */
5200 ret = btrfs_find_space_cluster(trans, root,
5201 block_group, last_ptr,
5203 empty_cluster + empty_size);
5206 * now pull our allocation out of this
5209 offset = btrfs_alloc_from_cluster(block_group,
5210 last_ptr, num_bytes,
5213 /* we found one, proceed */
5214 spin_unlock(&last_ptr->refill_lock);
5217 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5218 && !failed_cluster_refill) {
5219 spin_unlock(&last_ptr->refill_lock);
5221 failed_cluster_refill = true;
5222 wait_block_group_cache_progress(block_group,
5223 num_bytes + empty_cluster + empty_size);
5224 goto have_block_group;
5228 * at this point we either didn't find a cluster
5229 * or we weren't able to allocate a block from our
5230 * cluster. Free the cluster we've been trying
5231 * to use, and go to the next block group
5233 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5234 spin_unlock(&last_ptr->refill_lock);
5238 offset = btrfs_find_space_for_alloc(block_group, search_start,
5239 num_bytes, empty_size);
5241 * If we didn't find a chunk, and we haven't failed on this
5242 * block group before, and this block group is in the middle of
5243 * caching and we are ok with waiting, then go ahead and wait
5244 * for progress to be made, and set failed_alloc to true.
5246 * If failed_alloc is true then we've already waited on this
5247 * block group once and should move on to the next block group.
5249 if (!offset && !failed_alloc && !cached &&
5250 loop > LOOP_CACHING_NOWAIT) {
5251 wait_block_group_cache_progress(block_group,
5252 num_bytes + empty_size);
5253 failed_alloc = true;
5254 goto have_block_group;
5255 } else if (!offset) {
5259 search_start = stripe_align(root, offset);
5260 /* move on to the next group */
5261 if (search_start + num_bytes >= search_end) {
5262 btrfs_add_free_space(block_group, offset, num_bytes);
5266 /* move on to the next group */
5267 if (search_start + num_bytes >
5268 block_group->key.objectid + block_group->key.offset) {
5269 btrfs_add_free_space(block_group, offset, num_bytes);
5273 ins->objectid = search_start;
5274 ins->offset = num_bytes;
5276 if (offset < search_start)
5277 btrfs_add_free_space(block_group, offset,
5278 search_start - offset);
5279 BUG_ON(offset > search_start);
5281 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5282 (data & BTRFS_BLOCK_GROUP_DATA));
5283 if (ret == -EAGAIN) {
5284 btrfs_add_free_space(block_group, offset, num_bytes);
5288 /* we are all good, lets return */
5289 ins->objectid = search_start;
5290 ins->offset = num_bytes;
5292 if (offset < search_start)
5293 btrfs_add_free_space(block_group, offset,
5294 search_start - offset);
5295 BUG_ON(offset > search_start);
5298 failed_cluster_refill = false;
5299 failed_alloc = false;
5300 BUG_ON(index != get_block_group_index(block_group));
5301 btrfs_put_block_group(block_group);
5303 up_read(&space_info->groups_sem);
5305 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5308 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5309 * for them to make caching progress. Also
5310 * determine the best possible bg to cache
5311 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5312 * caching kthreads as we move along
5313 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5314 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5315 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5318 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5319 (found_uncached_bg || empty_size || empty_cluster ||
5320 allowed_chunk_alloc)) {
5322 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5323 found_uncached_bg = false;
5325 if (!ideal_cache_percent &&
5326 atomic_read(&space_info->caching_threads))
5330 * 1 of the following 2 things have happened so far
5332 * 1) We found an ideal block group for caching that
5333 * is mostly full and will cache quickly, so we might
5334 * as well wait for it.
5336 * 2) We searched for cached only and we didn't find
5337 * anything, and we didn't start any caching kthreads
5338 * either, so chances are we will loop through and
5339 * start a couple caching kthreads, and then come back
5340 * around and just wait for them. This will be slower
5341 * because we will have 2 caching kthreads reading at
5342 * the same time when we could have just started one
5343 * and waited for it to get far enough to give us an
5344 * allocation, so go ahead and go to the wait caching
5347 loop = LOOP_CACHING_WAIT;
5348 search_start = ideal_cache_offset;
5349 ideal_cache_percent = 0;
5351 } else if (loop == LOOP_FIND_IDEAL) {
5353 * Didn't find a uncached bg, wait on anything we find
5356 loop = LOOP_CACHING_WAIT;
5360 if (loop < LOOP_CACHING_WAIT) {
5365 if (loop == LOOP_ALLOC_CHUNK) {
5370 if (allowed_chunk_alloc) {
5371 ret = do_chunk_alloc(trans, root, num_bytes +
5372 2 * 1024 * 1024, data,
5373 CHUNK_ALLOC_LIMITED);
5374 allowed_chunk_alloc = 0;
5375 done_chunk_alloc = 1;
5376 } else if (!done_chunk_alloc &&
5377 space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5378 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5381 if (loop < LOOP_NO_EMPTY_SIZE) {
5386 } else if (!ins->objectid) {
5390 /* we found what we needed */
5391 if (ins->objectid) {
5392 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5393 trans->block_group = block_group->key.objectid;
5395 btrfs_put_block_group(block_group);
5402 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5403 int dump_block_groups)
5405 struct btrfs_block_group_cache *cache;
5408 spin_lock(&info->lock);
5409 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5410 (unsigned long long)(info->total_bytes - info->bytes_used -
5411 info->bytes_pinned - info->bytes_reserved -
5412 info->bytes_readonly),
5413 (info->full) ? "" : "not ");
5414 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5415 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5416 (unsigned long long)info->total_bytes,
5417 (unsigned long long)info->bytes_used,
5418 (unsigned long long)info->bytes_pinned,
5419 (unsigned long long)info->bytes_reserved,
5420 (unsigned long long)info->bytes_may_use,
5421 (unsigned long long)info->bytes_readonly);
5422 spin_unlock(&info->lock);
5424 if (!dump_block_groups)
5427 down_read(&info->groups_sem);
5429 list_for_each_entry(cache, &info->block_groups[index], list) {
5430 spin_lock(&cache->lock);
5431 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5432 "%llu pinned %llu reserved\n",
5433 (unsigned long long)cache->key.objectid,
5434 (unsigned long long)cache->key.offset,
5435 (unsigned long long)btrfs_block_group_used(&cache->item),
5436 (unsigned long long)cache->pinned,
5437 (unsigned long long)cache->reserved);
5438 btrfs_dump_free_space(cache, bytes);
5439 spin_unlock(&cache->lock);
5441 if (++index < BTRFS_NR_RAID_TYPES)
5443 up_read(&info->groups_sem);
5446 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5447 struct btrfs_root *root,
5448 u64 num_bytes, u64 min_alloc_size,
5449 u64 empty_size, u64 hint_byte,
5450 u64 search_end, struct btrfs_key *ins,
5454 u64 search_start = 0;
5456 data = btrfs_get_alloc_profile(root, data);
5459 * the only place that sets empty_size is btrfs_realloc_node, which
5460 * is not called recursively on allocations
5462 if (empty_size || root->ref_cows)
5463 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5464 num_bytes + 2 * 1024 * 1024, data,
5465 CHUNK_ALLOC_NO_FORCE);
5467 WARN_ON(num_bytes < root->sectorsize);
5468 ret = find_free_extent(trans, root, num_bytes, empty_size,
5469 search_start, search_end, hint_byte,
5472 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5473 num_bytes = num_bytes >> 1;
5474 num_bytes = num_bytes & ~(root->sectorsize - 1);
5475 num_bytes = max(num_bytes, min_alloc_size);
5476 do_chunk_alloc(trans, root->fs_info->extent_root,
5477 num_bytes, data, CHUNK_ALLOC_FORCE);
5480 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5481 struct btrfs_space_info *sinfo;
5483 sinfo = __find_space_info(root->fs_info, data);
5484 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5485 "wanted %llu\n", (unsigned long long)data,
5486 (unsigned long long)num_bytes);
5487 dump_space_info(sinfo, num_bytes, 1);
5490 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5495 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5497 struct btrfs_block_group_cache *cache;
5500 cache = btrfs_lookup_block_group(root->fs_info, start);
5502 printk(KERN_ERR "Unable to find block group for %llu\n",
5503 (unsigned long long)start);
5507 if (btrfs_test_opt(root, DISCARD))
5508 ret = btrfs_discard_extent(root, start, len, NULL);
5510 btrfs_add_free_space(cache, start, len);
5511 btrfs_update_reserved_bytes(cache, len, 0, 1);
5512 btrfs_put_block_group(cache);
5514 trace_btrfs_reserved_extent_free(root, start, len);
5519 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5520 struct btrfs_root *root,
5521 u64 parent, u64 root_objectid,
5522 u64 flags, u64 owner, u64 offset,
5523 struct btrfs_key *ins, int ref_mod)
5526 struct btrfs_fs_info *fs_info = root->fs_info;
5527 struct btrfs_extent_item *extent_item;
5528 struct btrfs_extent_inline_ref *iref;
5529 struct btrfs_path *path;
5530 struct extent_buffer *leaf;
5535 type = BTRFS_SHARED_DATA_REF_KEY;
5537 type = BTRFS_EXTENT_DATA_REF_KEY;
5539 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5541 path = btrfs_alloc_path();
5545 path->leave_spinning = 1;
5546 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5550 leaf = path->nodes[0];
5551 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5552 struct btrfs_extent_item);
5553 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5554 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5555 btrfs_set_extent_flags(leaf, extent_item,
5556 flags | BTRFS_EXTENT_FLAG_DATA);
5558 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5559 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5561 struct btrfs_shared_data_ref *ref;
5562 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5563 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5564 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5566 struct btrfs_extent_data_ref *ref;
5567 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5568 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5569 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5570 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5571 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5574 btrfs_mark_buffer_dirty(path->nodes[0]);
5575 btrfs_free_path(path);
5577 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5579 printk(KERN_ERR "btrfs update block group failed for %llu "
5580 "%llu\n", (unsigned long long)ins->objectid,
5581 (unsigned long long)ins->offset);
5587 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5588 struct btrfs_root *root,
5589 u64 parent, u64 root_objectid,
5590 u64 flags, struct btrfs_disk_key *key,
5591 int level, struct btrfs_key *ins)
5594 struct btrfs_fs_info *fs_info = root->fs_info;
5595 struct btrfs_extent_item *extent_item;
5596 struct btrfs_tree_block_info *block_info;
5597 struct btrfs_extent_inline_ref *iref;
5598 struct btrfs_path *path;
5599 struct extent_buffer *leaf;
5600 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5602 path = btrfs_alloc_path();
5605 path->leave_spinning = 1;
5606 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5610 leaf = path->nodes[0];
5611 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5612 struct btrfs_extent_item);
5613 btrfs_set_extent_refs(leaf, extent_item, 1);
5614 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5615 btrfs_set_extent_flags(leaf, extent_item,
5616 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5617 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5619 btrfs_set_tree_block_key(leaf, block_info, key);
5620 btrfs_set_tree_block_level(leaf, block_info, level);
5622 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5624 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5625 btrfs_set_extent_inline_ref_type(leaf, iref,
5626 BTRFS_SHARED_BLOCK_REF_KEY);
5627 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5629 btrfs_set_extent_inline_ref_type(leaf, iref,
5630 BTRFS_TREE_BLOCK_REF_KEY);
5631 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5634 btrfs_mark_buffer_dirty(leaf);
5635 btrfs_free_path(path);
5637 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5639 printk(KERN_ERR "btrfs update block group failed for %llu "
5640 "%llu\n", (unsigned long long)ins->objectid,
5641 (unsigned long long)ins->offset);
5647 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5648 struct btrfs_root *root,
5649 u64 root_objectid, u64 owner,
5650 u64 offset, struct btrfs_key *ins)
5654 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5656 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5657 0, root_objectid, owner, offset,
5658 BTRFS_ADD_DELAYED_EXTENT, NULL);
5663 * this is used by the tree logging recovery code. It records that
5664 * an extent has been allocated and makes sure to clear the free
5665 * space cache bits as well
5667 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5668 struct btrfs_root *root,
5669 u64 root_objectid, u64 owner, u64 offset,
5670 struct btrfs_key *ins)
5673 struct btrfs_block_group_cache *block_group;
5674 struct btrfs_caching_control *caching_ctl;
5675 u64 start = ins->objectid;
5676 u64 num_bytes = ins->offset;
5678 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5679 cache_block_group(block_group, trans, NULL, 0);
5680 caching_ctl = get_caching_control(block_group);
5683 BUG_ON(!block_group_cache_done(block_group));
5684 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5687 mutex_lock(&caching_ctl->mutex);
5689 if (start >= caching_ctl->progress) {
5690 ret = add_excluded_extent(root, start, num_bytes);
5692 } else if (start + num_bytes <= caching_ctl->progress) {
5693 ret = btrfs_remove_free_space(block_group,
5697 num_bytes = caching_ctl->progress - start;
5698 ret = btrfs_remove_free_space(block_group,
5702 start = caching_ctl->progress;
5703 num_bytes = ins->objectid + ins->offset -
5704 caching_ctl->progress;
5705 ret = add_excluded_extent(root, start, num_bytes);
5709 mutex_unlock(&caching_ctl->mutex);
5710 put_caching_control(caching_ctl);
5713 ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5715 btrfs_put_block_group(block_group);
5716 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5717 0, owner, offset, ins, 1);
5721 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5722 struct btrfs_root *root,
5723 u64 bytenr, u32 blocksize,
5726 struct extent_buffer *buf;
5728 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5730 return ERR_PTR(-ENOMEM);
5731 btrfs_set_header_generation(buf, trans->transid);
5732 btrfs_set_buffer_lockdep_class(buf, level);
5733 btrfs_tree_lock(buf);
5734 clean_tree_block(trans, root, buf);
5736 btrfs_set_lock_blocking(buf);
5737 btrfs_set_buffer_uptodate(buf);
5739 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5741 * we allow two log transactions at a time, use different
5742 * EXENT bit to differentiate dirty pages.
5744 if (root->log_transid % 2 == 0)
5745 set_extent_dirty(&root->dirty_log_pages, buf->start,
5746 buf->start + buf->len - 1, GFP_NOFS);
5748 set_extent_new(&root->dirty_log_pages, buf->start,
5749 buf->start + buf->len - 1, GFP_NOFS);
5751 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5752 buf->start + buf->len - 1, GFP_NOFS);
5754 trans->blocks_used++;
5755 /* this returns a buffer locked for blocking */
5759 static struct btrfs_block_rsv *
5760 use_block_rsv(struct btrfs_trans_handle *trans,
5761 struct btrfs_root *root, u32 blocksize)
5763 struct btrfs_block_rsv *block_rsv;
5764 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5767 block_rsv = get_block_rsv(trans, root);
5769 if (block_rsv->size == 0) {
5770 ret = reserve_metadata_bytes(trans, root, block_rsv,
5773 * If we couldn't reserve metadata bytes try and use some from
5774 * the global reserve.
5776 if (ret && block_rsv != global_rsv) {
5777 ret = block_rsv_use_bytes(global_rsv, blocksize);
5780 return ERR_PTR(ret);
5782 return ERR_PTR(ret);
5787 ret = block_rsv_use_bytes(block_rsv, blocksize);
5792 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5795 spin_lock(&block_rsv->lock);
5796 block_rsv->size += blocksize;
5797 spin_unlock(&block_rsv->lock);
5799 } else if (ret && block_rsv != global_rsv) {
5800 ret = block_rsv_use_bytes(global_rsv, blocksize);
5806 return ERR_PTR(-ENOSPC);
5809 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5811 block_rsv_add_bytes(block_rsv, blocksize, 0);
5812 block_rsv_release_bytes(block_rsv, NULL, 0);
5816 * finds a free extent and does all the dirty work required for allocation
5817 * returns the key for the extent through ins, and a tree buffer for
5818 * the first block of the extent through buf.
5820 * returns the tree buffer or NULL.
5822 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5823 struct btrfs_root *root, u32 blocksize,
5824 u64 parent, u64 root_objectid,
5825 struct btrfs_disk_key *key, int level,
5826 u64 hint, u64 empty_size)
5828 struct btrfs_key ins;
5829 struct btrfs_block_rsv *block_rsv;
5830 struct extent_buffer *buf;
5835 block_rsv = use_block_rsv(trans, root, blocksize);
5836 if (IS_ERR(block_rsv))
5837 return ERR_CAST(block_rsv);
5839 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5840 empty_size, hint, (u64)-1, &ins, 0);
5842 unuse_block_rsv(block_rsv, blocksize);
5843 return ERR_PTR(ret);
5846 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5848 BUG_ON(IS_ERR(buf));
5850 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5852 parent = ins.objectid;
5853 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5857 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5858 struct btrfs_delayed_extent_op *extent_op;
5859 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5862 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5864 memset(&extent_op->key, 0, sizeof(extent_op->key));
5865 extent_op->flags_to_set = flags;
5866 extent_op->update_key = 1;
5867 extent_op->update_flags = 1;
5868 extent_op->is_data = 0;
5870 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5871 ins.offset, parent, root_objectid,
5872 level, BTRFS_ADD_DELAYED_EXTENT,
5879 struct walk_control {
5880 u64 refs[BTRFS_MAX_LEVEL];
5881 u64 flags[BTRFS_MAX_LEVEL];
5882 struct btrfs_key update_progress;
5892 #define DROP_REFERENCE 1
5893 #define UPDATE_BACKREF 2
5895 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5896 struct btrfs_root *root,
5897 struct walk_control *wc,
5898 struct btrfs_path *path)
5906 struct btrfs_key key;
5907 struct extent_buffer *eb;
5912 if (path->slots[wc->level] < wc->reada_slot) {
5913 wc->reada_count = wc->reada_count * 2 / 3;
5914 wc->reada_count = max(wc->reada_count, 2);
5916 wc->reada_count = wc->reada_count * 3 / 2;
5917 wc->reada_count = min_t(int, wc->reada_count,
5918 BTRFS_NODEPTRS_PER_BLOCK(root));
5921 eb = path->nodes[wc->level];
5922 nritems = btrfs_header_nritems(eb);
5923 blocksize = btrfs_level_size(root, wc->level - 1);
5925 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5926 if (nread >= wc->reada_count)
5930 bytenr = btrfs_node_blockptr(eb, slot);
5931 generation = btrfs_node_ptr_generation(eb, slot);
5933 if (slot == path->slots[wc->level])
5936 if (wc->stage == UPDATE_BACKREF &&
5937 generation <= root->root_key.offset)
5940 /* We don't lock the tree block, it's OK to be racy here */
5941 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5946 if (wc->stage == DROP_REFERENCE) {
5950 if (wc->level == 1 &&
5951 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5953 if (!wc->update_ref ||
5954 generation <= root->root_key.offset)
5956 btrfs_node_key_to_cpu(eb, &key, slot);
5957 ret = btrfs_comp_cpu_keys(&key,
5958 &wc->update_progress);
5962 if (wc->level == 1 &&
5963 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5967 ret = readahead_tree_block(root, bytenr, blocksize,
5973 wc->reada_slot = slot;
5977 * hepler to process tree block while walking down the tree.
5979 * when wc->stage == UPDATE_BACKREF, this function updates
5980 * back refs for pointers in the block.
5982 * NOTE: return value 1 means we should stop walking down.
5984 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5985 struct btrfs_root *root,
5986 struct btrfs_path *path,
5987 struct walk_control *wc, int lookup_info)
5989 int level = wc->level;
5990 struct extent_buffer *eb = path->nodes[level];
5991 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5994 if (wc->stage == UPDATE_BACKREF &&
5995 btrfs_header_owner(eb) != root->root_key.objectid)
5999 * when reference count of tree block is 1, it won't increase
6000 * again. once full backref flag is set, we never clear it.
6003 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6004 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6005 BUG_ON(!path->locks[level]);
6006 ret = btrfs_lookup_extent_info(trans, root,
6011 BUG_ON(wc->refs[level] == 0);
6014 if (wc->stage == DROP_REFERENCE) {
6015 if (wc->refs[level] > 1)
6018 if (path->locks[level] && !wc->keep_locks) {
6019 btrfs_tree_unlock(eb);
6020 path->locks[level] = 0;
6025 /* wc->stage == UPDATE_BACKREF */
6026 if (!(wc->flags[level] & flag)) {
6027 BUG_ON(!path->locks[level]);
6028 ret = btrfs_inc_ref(trans, root, eb, 1);
6030 ret = btrfs_dec_ref(trans, root, eb, 0);
6032 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6035 wc->flags[level] |= flag;
6039 * the block is shared by multiple trees, so it's not good to
6040 * keep the tree lock
6042 if (path->locks[level] && level > 0) {
6043 btrfs_tree_unlock(eb);
6044 path->locks[level] = 0;
6050 * hepler to process tree block pointer.
6052 * when wc->stage == DROP_REFERENCE, this function checks
6053 * reference count of the block pointed to. if the block
6054 * is shared and we need update back refs for the subtree
6055 * rooted at the block, this function changes wc->stage to
6056 * UPDATE_BACKREF. if the block is shared and there is no
6057 * need to update back, this function drops the reference
6060 * NOTE: return value 1 means we should stop walking down.
6062 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6063 struct btrfs_root *root,
6064 struct btrfs_path *path,
6065 struct walk_control *wc, int *lookup_info)
6071 struct btrfs_key key;
6072 struct extent_buffer *next;
6073 int level = wc->level;
6077 generation = btrfs_node_ptr_generation(path->nodes[level],
6078 path->slots[level]);
6080 * if the lower level block was created before the snapshot
6081 * was created, we know there is no need to update back refs
6084 if (wc->stage == UPDATE_BACKREF &&
6085 generation <= root->root_key.offset) {
6090 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6091 blocksize = btrfs_level_size(root, level - 1);
6093 next = btrfs_find_tree_block(root, bytenr, blocksize);
6095 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6100 btrfs_tree_lock(next);
6101 btrfs_set_lock_blocking(next);
6103 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6104 &wc->refs[level - 1],
6105 &wc->flags[level - 1]);
6107 BUG_ON(wc->refs[level - 1] == 0);
6110 if (wc->stage == DROP_REFERENCE) {
6111 if (wc->refs[level - 1] > 1) {
6113 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6116 if (!wc->update_ref ||
6117 generation <= root->root_key.offset)
6120 btrfs_node_key_to_cpu(path->nodes[level], &key,
6121 path->slots[level]);
6122 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6126 wc->stage = UPDATE_BACKREF;
6127 wc->shared_level = level - 1;
6131 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6135 if (!btrfs_buffer_uptodate(next, generation)) {
6136 btrfs_tree_unlock(next);
6137 free_extent_buffer(next);
6143 if (reada && level == 1)
6144 reada_walk_down(trans, root, wc, path);
6145 next = read_tree_block(root, bytenr, blocksize, generation);
6148 btrfs_tree_lock(next);
6149 btrfs_set_lock_blocking(next);
6153 BUG_ON(level != btrfs_header_level(next));
6154 path->nodes[level] = next;
6155 path->slots[level] = 0;
6156 path->locks[level] = 1;
6162 wc->refs[level - 1] = 0;
6163 wc->flags[level - 1] = 0;
6164 if (wc->stage == DROP_REFERENCE) {
6165 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6166 parent = path->nodes[level]->start;
6168 BUG_ON(root->root_key.objectid !=
6169 btrfs_header_owner(path->nodes[level]));
6173 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6174 root->root_key.objectid, level - 1, 0);
6177 btrfs_tree_unlock(next);
6178 free_extent_buffer(next);
6184 * hepler to process tree block while walking up the tree.
6186 * when wc->stage == DROP_REFERENCE, this function drops
6187 * reference count on the block.
6189 * when wc->stage == UPDATE_BACKREF, this function changes
6190 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6191 * to UPDATE_BACKREF previously while processing the block.
6193 * NOTE: return value 1 means we should stop walking up.
6195 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6196 struct btrfs_root *root,
6197 struct btrfs_path *path,
6198 struct walk_control *wc)
6201 int level = wc->level;
6202 struct extent_buffer *eb = path->nodes[level];
6205 if (wc->stage == UPDATE_BACKREF) {
6206 BUG_ON(wc->shared_level < level);
6207 if (level < wc->shared_level)
6210 ret = find_next_key(path, level + 1, &wc->update_progress);
6214 wc->stage = DROP_REFERENCE;
6215 wc->shared_level = -1;
6216 path->slots[level] = 0;
6219 * check reference count again if the block isn't locked.
6220 * we should start walking down the tree again if reference
6223 if (!path->locks[level]) {
6225 btrfs_tree_lock(eb);
6226 btrfs_set_lock_blocking(eb);
6227 path->locks[level] = 1;
6229 ret = btrfs_lookup_extent_info(trans, root,
6234 BUG_ON(wc->refs[level] == 0);
6235 if (wc->refs[level] == 1) {
6236 btrfs_tree_unlock(eb);
6237 path->locks[level] = 0;
6243 /* wc->stage == DROP_REFERENCE */
6244 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6246 if (wc->refs[level] == 1) {
6248 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6249 ret = btrfs_dec_ref(trans, root, eb, 1);
6251 ret = btrfs_dec_ref(trans, root, eb, 0);
6254 /* make block locked assertion in clean_tree_block happy */
6255 if (!path->locks[level] &&
6256 btrfs_header_generation(eb) == trans->transid) {
6257 btrfs_tree_lock(eb);
6258 btrfs_set_lock_blocking(eb);
6259 path->locks[level] = 1;
6261 clean_tree_block(trans, root, eb);
6264 if (eb == root->node) {
6265 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6268 BUG_ON(root->root_key.objectid !=
6269 btrfs_header_owner(eb));
6271 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6272 parent = path->nodes[level + 1]->start;
6274 BUG_ON(root->root_key.objectid !=
6275 btrfs_header_owner(path->nodes[level + 1]));
6278 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6280 wc->refs[level] = 0;
6281 wc->flags[level] = 0;
6285 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6286 struct btrfs_root *root,
6287 struct btrfs_path *path,
6288 struct walk_control *wc)
6290 int level = wc->level;
6291 int lookup_info = 1;
6294 while (level >= 0) {
6295 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6302 if (path->slots[level] >=
6303 btrfs_header_nritems(path->nodes[level]))
6306 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6308 path->slots[level]++;
6317 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6318 struct btrfs_root *root,
6319 struct btrfs_path *path,
6320 struct walk_control *wc, int max_level)
6322 int level = wc->level;
6325 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6326 while (level < max_level && path->nodes[level]) {
6328 if (path->slots[level] + 1 <
6329 btrfs_header_nritems(path->nodes[level])) {
6330 path->slots[level]++;
6333 ret = walk_up_proc(trans, root, path, wc);
6337 if (path->locks[level]) {
6338 btrfs_tree_unlock(path->nodes[level]);
6339 path->locks[level] = 0;
6341 free_extent_buffer(path->nodes[level]);
6342 path->nodes[level] = NULL;
6350 * drop a subvolume tree.
6352 * this function traverses the tree freeing any blocks that only
6353 * referenced by the tree.
6355 * when a shared tree block is found. this function decreases its
6356 * reference count by one. if update_ref is true, this function
6357 * also make sure backrefs for the shared block and all lower level
6358 * blocks are properly updated.
6360 int btrfs_drop_snapshot(struct btrfs_root *root,
6361 struct btrfs_block_rsv *block_rsv, int update_ref)
6363 struct btrfs_path *path;
6364 struct btrfs_trans_handle *trans;
6365 struct btrfs_root *tree_root = root->fs_info->tree_root;
6366 struct btrfs_root_item *root_item = &root->root_item;
6367 struct walk_control *wc;
6368 struct btrfs_key key;
6373 path = btrfs_alloc_path();
6376 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6379 trans = btrfs_start_transaction(tree_root, 0);
6380 BUG_ON(IS_ERR(trans));
6383 trans->block_rsv = block_rsv;
6385 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6386 level = btrfs_header_level(root->node);
6387 path->nodes[level] = btrfs_lock_root_node(root);
6388 btrfs_set_lock_blocking(path->nodes[level]);
6389 path->slots[level] = 0;
6390 path->locks[level] = 1;
6391 memset(&wc->update_progress, 0,
6392 sizeof(wc->update_progress));
6394 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6395 memcpy(&wc->update_progress, &key,
6396 sizeof(wc->update_progress));
6398 level = root_item->drop_level;
6400 path->lowest_level = level;
6401 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6402 path->lowest_level = 0;
6410 * unlock our path, this is safe because only this
6411 * function is allowed to delete this snapshot
6413 btrfs_unlock_up_safe(path, 0);
6415 level = btrfs_header_level(root->node);
6417 btrfs_tree_lock(path->nodes[level]);
6418 btrfs_set_lock_blocking(path->nodes[level]);
6420 ret = btrfs_lookup_extent_info(trans, root,
6421 path->nodes[level]->start,
6422 path->nodes[level]->len,
6426 BUG_ON(wc->refs[level] == 0);
6428 if (level == root_item->drop_level)
6431 btrfs_tree_unlock(path->nodes[level]);
6432 WARN_ON(wc->refs[level] != 1);
6438 wc->shared_level = -1;
6439 wc->stage = DROP_REFERENCE;
6440 wc->update_ref = update_ref;
6442 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6445 ret = walk_down_tree(trans, root, path, wc);
6451 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6458 BUG_ON(wc->stage != DROP_REFERENCE);
6462 if (wc->stage == DROP_REFERENCE) {
6464 btrfs_node_key(path->nodes[level],
6465 &root_item->drop_progress,
6466 path->slots[level]);
6467 root_item->drop_level = level;
6470 BUG_ON(wc->level == 0);
6471 if (btrfs_should_end_transaction(trans, tree_root)) {
6472 ret = btrfs_update_root(trans, tree_root,
6477 btrfs_end_transaction_throttle(trans, tree_root);
6478 trans = btrfs_start_transaction(tree_root, 0);
6479 BUG_ON(IS_ERR(trans));
6481 trans->block_rsv = block_rsv;
6484 btrfs_release_path(root, path);
6487 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6490 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6491 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6495 /* if we fail to delete the orphan item this time
6496 * around, it'll get picked up the next time.
6498 * The most common failure here is just -ENOENT.
6500 btrfs_del_orphan_item(trans, tree_root,
6501 root->root_key.objectid);
6505 if (root->in_radix) {
6506 btrfs_free_fs_root(tree_root->fs_info, root);
6508 free_extent_buffer(root->node);
6509 free_extent_buffer(root->commit_root);
6513 btrfs_end_transaction_throttle(trans, tree_root);
6515 btrfs_free_path(path);
6520 * drop subtree rooted at tree block 'node'.
6522 * NOTE: this function will unlock and release tree block 'node'
6524 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6525 struct btrfs_root *root,
6526 struct extent_buffer *node,
6527 struct extent_buffer *parent)
6529 struct btrfs_path *path;
6530 struct walk_control *wc;
6536 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6538 path = btrfs_alloc_path();
6542 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6544 btrfs_free_path(path);
6548 btrfs_assert_tree_locked(parent);
6549 parent_level = btrfs_header_level(parent);
6550 extent_buffer_get(parent);
6551 path->nodes[parent_level] = parent;
6552 path->slots[parent_level] = btrfs_header_nritems(parent);
6554 btrfs_assert_tree_locked(node);
6555 level = btrfs_header_level(node);
6556 path->nodes[level] = node;
6557 path->slots[level] = 0;
6558 path->locks[level] = 1;
6560 wc->refs[parent_level] = 1;
6561 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6563 wc->shared_level = -1;
6564 wc->stage = DROP_REFERENCE;
6567 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6570 wret = walk_down_tree(trans, root, path, wc);
6576 wret = walk_up_tree(trans, root, path, wc, parent_level);
6584 btrfs_free_path(path);
6589 static unsigned long calc_ra(unsigned long start, unsigned long last,
6592 return min(last, start + nr - 1);
6595 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6600 unsigned long first_index;
6601 unsigned long last_index;
6604 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6605 struct file_ra_state *ra;
6606 struct btrfs_ordered_extent *ordered;
6607 unsigned int total_read = 0;
6608 unsigned int total_dirty = 0;
6611 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6615 mutex_lock(&inode->i_mutex);
6616 first_index = start >> PAGE_CACHE_SHIFT;
6617 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6619 /* make sure the dirty trick played by the caller work */
6620 ret = invalidate_inode_pages2_range(inode->i_mapping,
6621 first_index, last_index);
6625 file_ra_state_init(ra, inode->i_mapping);
6627 for (i = first_index ; i <= last_index; i++) {
6628 if (total_read % ra->ra_pages == 0) {
6629 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6630 calc_ra(i, last_index, ra->ra_pages));
6634 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6636 page = grab_cache_page(inode->i_mapping, i);
6641 if (!PageUptodate(page)) {
6642 btrfs_readpage(NULL, page);
6644 if (!PageUptodate(page)) {
6646 page_cache_release(page);
6651 wait_on_page_writeback(page);
6653 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6654 page_end = page_start + PAGE_CACHE_SIZE - 1;
6655 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6657 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6659 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6661 page_cache_release(page);
6662 btrfs_start_ordered_extent(inode, ordered, 1);
6663 btrfs_put_ordered_extent(ordered);
6666 set_page_extent_mapped(page);
6668 if (i == first_index)
6669 set_extent_bits(io_tree, page_start, page_end,
6670 EXTENT_BOUNDARY, GFP_NOFS);
6671 btrfs_set_extent_delalloc(inode, page_start, page_end);
6673 set_page_dirty(page);
6676 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6678 page_cache_release(page);
6683 mutex_unlock(&inode->i_mutex);
6684 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6688 static noinline int relocate_data_extent(struct inode *reloc_inode,
6689 struct btrfs_key *extent_key,
6692 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6693 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6694 struct extent_map *em;
6695 u64 start = extent_key->objectid - offset;
6696 u64 end = start + extent_key->offset - 1;
6698 em = alloc_extent_map(GFP_NOFS);
6702 em->len = extent_key->offset;
6703 em->block_len = extent_key->offset;
6704 em->block_start = extent_key->objectid;
6705 em->bdev = root->fs_info->fs_devices->latest_bdev;
6706 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6708 /* setup extent map to cheat btrfs_readpage */
6709 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6712 write_lock(&em_tree->lock);
6713 ret = add_extent_mapping(em_tree, em);
6714 write_unlock(&em_tree->lock);
6715 if (ret != -EEXIST) {
6716 free_extent_map(em);
6719 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6721 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6723 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6726 struct btrfs_ref_path {
6728 u64 nodes[BTRFS_MAX_LEVEL];
6730 u64 root_generation;
6737 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6738 u64 new_nodes[BTRFS_MAX_LEVEL];
6741 struct disk_extent {
6752 static int is_cowonly_root(u64 root_objectid)
6754 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6755 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6756 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6757 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6758 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6759 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6764 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6765 struct btrfs_root *extent_root,
6766 struct btrfs_ref_path *ref_path,
6769 struct extent_buffer *leaf;
6770 struct btrfs_path *path;
6771 struct btrfs_extent_ref *ref;
6772 struct btrfs_key key;
6773 struct btrfs_key found_key;
6779 path = btrfs_alloc_path();
6784 ref_path->lowest_level = -1;
6785 ref_path->current_level = -1;
6786 ref_path->shared_level = -1;
6790 level = ref_path->current_level - 1;
6791 while (level >= -1) {
6793 if (level < ref_path->lowest_level)
6797 bytenr = ref_path->nodes[level];
6799 bytenr = ref_path->extent_start;
6800 BUG_ON(bytenr == 0);
6802 parent = ref_path->nodes[level + 1];
6803 ref_path->nodes[level + 1] = 0;
6804 ref_path->current_level = level;
6805 BUG_ON(parent == 0);
6807 key.objectid = bytenr;
6808 key.offset = parent + 1;
6809 key.type = BTRFS_EXTENT_REF_KEY;
6811 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6816 leaf = path->nodes[0];
6817 nritems = btrfs_header_nritems(leaf);
6818 if (path->slots[0] >= nritems) {
6819 ret = btrfs_next_leaf(extent_root, path);
6824 leaf = path->nodes[0];
6827 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6828 if (found_key.objectid == bytenr &&
6829 found_key.type == BTRFS_EXTENT_REF_KEY) {
6830 if (level < ref_path->shared_level)
6831 ref_path->shared_level = level;
6836 btrfs_release_path(extent_root, path);
6839 /* reached lowest level */
6843 level = ref_path->current_level;
6844 while (level < BTRFS_MAX_LEVEL - 1) {
6848 bytenr = ref_path->nodes[level];
6850 bytenr = ref_path->extent_start;
6852 BUG_ON(bytenr == 0);
6854 key.objectid = bytenr;
6856 key.type = BTRFS_EXTENT_REF_KEY;
6858 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6862 leaf = path->nodes[0];
6863 nritems = btrfs_header_nritems(leaf);
6864 if (path->slots[0] >= nritems) {
6865 ret = btrfs_next_leaf(extent_root, path);
6869 /* the extent was freed by someone */
6870 if (ref_path->lowest_level == level)
6872 btrfs_release_path(extent_root, path);
6875 leaf = path->nodes[0];
6878 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6879 if (found_key.objectid != bytenr ||
6880 found_key.type != BTRFS_EXTENT_REF_KEY) {
6881 /* the extent was freed by someone */
6882 if (ref_path->lowest_level == level) {
6886 btrfs_release_path(extent_root, path);
6890 ref = btrfs_item_ptr(leaf, path->slots[0],
6891 struct btrfs_extent_ref);
6892 ref_objectid = btrfs_ref_objectid(leaf, ref);
6893 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6895 level = (int)ref_objectid;
6896 BUG_ON(level >= BTRFS_MAX_LEVEL);
6897 ref_path->lowest_level = level;
6898 ref_path->current_level = level;
6899 ref_path->nodes[level] = bytenr;
6901 WARN_ON(ref_objectid != level);
6904 WARN_ON(level != -1);
6908 if (ref_path->lowest_level == level) {
6909 ref_path->owner_objectid = ref_objectid;
6910 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6914 * the block is tree root or the block isn't in reference
6917 if (found_key.objectid == found_key.offset ||
6918 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6919 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6920 ref_path->root_generation =
6921 btrfs_ref_generation(leaf, ref);
6923 /* special reference from the tree log */
6924 ref_path->nodes[0] = found_key.offset;
6925 ref_path->current_level = 0;
6932 BUG_ON(ref_path->nodes[level] != 0);
6933 ref_path->nodes[level] = found_key.offset;
6934 ref_path->current_level = level;
6937 * the reference was created in the running transaction,
6938 * no need to continue walking up.
6940 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6941 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6942 ref_path->root_generation =
6943 btrfs_ref_generation(leaf, ref);
6948 btrfs_release_path(extent_root, path);
6951 /* reached max tree level, but no tree root found. */
6954 btrfs_free_path(path);
6958 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6959 struct btrfs_root *extent_root,
6960 struct btrfs_ref_path *ref_path,
6963 memset(ref_path, 0, sizeof(*ref_path));
6964 ref_path->extent_start = extent_start;
6966 return __next_ref_path(trans, extent_root, ref_path, 1);
6969 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6970 struct btrfs_root *extent_root,
6971 struct btrfs_ref_path *ref_path)
6973 return __next_ref_path(trans, extent_root, ref_path, 0);
6976 static noinline int get_new_locations(struct inode *reloc_inode,
6977 struct btrfs_key *extent_key,
6978 u64 offset, int no_fragment,
6979 struct disk_extent **extents,
6982 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6983 struct btrfs_path *path;
6984 struct btrfs_file_extent_item *fi;
6985 struct extent_buffer *leaf;
6986 struct disk_extent *exts = *extents;
6987 struct btrfs_key found_key;
6992 int max = *nr_extents;
6995 WARN_ON(!no_fragment && *extents);
6998 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
7003 path = btrfs_alloc_path();
7005 if (exts != *extents)
7010 cur_pos = extent_key->objectid - offset;
7011 last_byte = extent_key->objectid + extent_key->offset;
7012 ret = btrfs_lookup_file_extent(NULL, root, path,
7013 btrfs_ino(reloc_inode), cur_pos, 0);
7022 leaf = path->nodes[0];
7023 nritems = btrfs_header_nritems(leaf);
7024 if (path->slots[0] >= nritems) {
7025 ret = btrfs_next_leaf(root, path);
7030 leaf = path->nodes[0];
7033 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7034 if (found_key.offset != cur_pos ||
7035 found_key.type != BTRFS_EXTENT_DATA_KEY ||
7036 found_key.objectid != btrfs_ino(reloc_inode))
7039 fi = btrfs_item_ptr(leaf, path->slots[0],
7040 struct btrfs_file_extent_item);
7041 if (btrfs_file_extent_type(leaf, fi) !=
7042 BTRFS_FILE_EXTENT_REG ||
7043 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7047 struct disk_extent *old = exts;
7049 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7054 memcpy(exts, old, sizeof(*exts) * nr);
7055 if (old != *extents)
7059 exts[nr].disk_bytenr =
7060 btrfs_file_extent_disk_bytenr(leaf, fi);
7061 exts[nr].disk_num_bytes =
7062 btrfs_file_extent_disk_num_bytes(leaf, fi);
7063 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7064 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7065 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7066 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7067 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7068 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7070 BUG_ON(exts[nr].offset > 0);
7071 BUG_ON(exts[nr].compression || exts[nr].encryption);
7072 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7074 cur_pos += exts[nr].num_bytes;
7077 if (cur_pos + offset >= last_byte)
7087 BUG_ON(cur_pos + offset > last_byte);
7088 if (cur_pos + offset < last_byte) {
7094 btrfs_free_path(path);
7096 if (exts != *extents)
7105 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7106 struct btrfs_root *root,
7107 struct btrfs_path *path,
7108 struct btrfs_key *extent_key,
7109 struct btrfs_key *leaf_key,
7110 struct btrfs_ref_path *ref_path,
7111 struct disk_extent *new_extents,
7114 struct extent_buffer *leaf;
7115 struct btrfs_file_extent_item *fi;
7116 struct inode *inode = NULL;
7117 struct btrfs_key key;
7122 u64 search_end = (u64)-1;
7125 int extent_locked = 0;
7129 memcpy(&key, leaf_key, sizeof(key));
7130 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7131 if (key.objectid < ref_path->owner_objectid ||
7132 (key.objectid == ref_path->owner_objectid &&
7133 key.type < BTRFS_EXTENT_DATA_KEY)) {
7134 key.objectid = ref_path->owner_objectid;
7135 key.type = BTRFS_EXTENT_DATA_KEY;
7141 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7145 leaf = path->nodes[0];
7146 nritems = btrfs_header_nritems(leaf);
7148 if (extent_locked && ret > 0) {
7150 * the file extent item was modified by someone
7151 * before the extent got locked.
7153 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7154 lock_end, GFP_NOFS);
7158 if (path->slots[0] >= nritems) {
7159 if (++nr_scaned > 2)
7162 BUG_ON(extent_locked);
7163 ret = btrfs_next_leaf(root, path);
7168 leaf = path->nodes[0];
7169 nritems = btrfs_header_nritems(leaf);
7172 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7174 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7175 if ((key.objectid > ref_path->owner_objectid) ||
7176 (key.objectid == ref_path->owner_objectid &&
7177 key.type > BTRFS_EXTENT_DATA_KEY) ||
7178 key.offset >= search_end)
7182 if (inode && key.objectid != btrfs_ino(inode)) {
7183 BUG_ON(extent_locked);
7184 btrfs_release_path(root, path);
7185 mutex_unlock(&inode->i_mutex);
7191 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7196 fi = btrfs_item_ptr(leaf, path->slots[0],
7197 struct btrfs_file_extent_item);
7198 extent_type = btrfs_file_extent_type(leaf, fi);
7199 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7200 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7201 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7202 extent_key->objectid)) {
7208 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7209 ext_offset = btrfs_file_extent_offset(leaf, fi);
7211 if (search_end == (u64)-1) {
7212 search_end = key.offset - ext_offset +
7213 btrfs_file_extent_ram_bytes(leaf, fi);
7216 if (!extent_locked) {
7217 lock_start = key.offset;
7218 lock_end = lock_start + num_bytes - 1;
7220 if (lock_start > key.offset ||
7221 lock_end + 1 < key.offset + num_bytes) {
7222 unlock_extent(&BTRFS_I(inode)->io_tree,
7223 lock_start, lock_end, GFP_NOFS);
7229 btrfs_release_path(root, path);
7231 inode = btrfs_iget_locked(root->fs_info->sb,
7232 key.objectid, root);
7233 if (inode->i_state & I_NEW) {
7234 BTRFS_I(inode)->root = root;
7235 BTRFS_I(inode)->location.objectid =
7237 BTRFS_I(inode)->location.type =
7238 BTRFS_INODE_ITEM_KEY;
7239 BTRFS_I(inode)->location.offset = 0;
7240 btrfs_read_locked_inode(inode);
7241 unlock_new_inode(inode);
7244 * some code call btrfs_commit_transaction while
7245 * holding the i_mutex, so we can't use mutex_lock
7248 if (is_bad_inode(inode) ||
7249 !mutex_trylock(&inode->i_mutex)) {
7252 key.offset = (u64)-1;
7257 if (!extent_locked) {
7258 struct btrfs_ordered_extent *ordered;
7260 btrfs_release_path(root, path);
7262 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7263 lock_end, GFP_NOFS);
7264 ordered = btrfs_lookup_first_ordered_extent(inode,
7267 ordered->file_offset <= lock_end &&
7268 ordered->file_offset + ordered->len > lock_start) {
7269 unlock_extent(&BTRFS_I(inode)->io_tree,
7270 lock_start, lock_end, GFP_NOFS);
7271 btrfs_start_ordered_extent(inode, ordered, 1);
7272 btrfs_put_ordered_extent(ordered);
7273 key.offset += num_bytes;
7277 btrfs_put_ordered_extent(ordered);
7283 if (nr_extents == 1) {
7284 /* update extent pointer in place */
7285 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7286 new_extents[0].disk_bytenr);
7287 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7288 new_extents[0].disk_num_bytes);
7289 btrfs_mark_buffer_dirty(leaf);
7291 btrfs_drop_extent_cache(inode, key.offset,
7292 key.offset + num_bytes - 1, 0);
7294 ret = btrfs_inc_extent_ref(trans, root,
7295 new_extents[0].disk_bytenr,
7296 new_extents[0].disk_num_bytes,
7298 root->root_key.objectid,
7303 ret = btrfs_free_extent(trans, root,
7304 extent_key->objectid,
7307 btrfs_header_owner(leaf),
7308 btrfs_header_generation(leaf),
7312 btrfs_release_path(root, path);
7313 key.offset += num_bytes;
7321 * drop old extent pointer at first, then insert the
7322 * new pointers one bye one
7324 btrfs_release_path(root, path);
7325 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7326 key.offset + num_bytes,
7327 key.offset, &alloc_hint);
7330 for (i = 0; i < nr_extents; i++) {
7331 if (ext_offset >= new_extents[i].num_bytes) {
7332 ext_offset -= new_extents[i].num_bytes;
7335 extent_len = min(new_extents[i].num_bytes -
7336 ext_offset, num_bytes);
7338 ret = btrfs_insert_empty_item(trans, root,
7343 leaf = path->nodes[0];
7344 fi = btrfs_item_ptr(leaf, path->slots[0],
7345 struct btrfs_file_extent_item);
7346 btrfs_set_file_extent_generation(leaf, fi,
7348 btrfs_set_file_extent_type(leaf, fi,
7349 BTRFS_FILE_EXTENT_REG);
7350 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7351 new_extents[i].disk_bytenr);
7352 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7353 new_extents[i].disk_num_bytes);
7354 btrfs_set_file_extent_ram_bytes(leaf, fi,
7355 new_extents[i].ram_bytes);
7357 btrfs_set_file_extent_compression(leaf, fi,
7358 new_extents[i].compression);
7359 btrfs_set_file_extent_encryption(leaf, fi,
7360 new_extents[i].encryption);
7361 btrfs_set_file_extent_other_encoding(leaf, fi,
7362 new_extents[i].other_encoding);
7364 btrfs_set_file_extent_num_bytes(leaf, fi,
7366 ext_offset += new_extents[i].offset;
7367 btrfs_set_file_extent_offset(leaf, fi,
7369 btrfs_mark_buffer_dirty(leaf);
7371 btrfs_drop_extent_cache(inode, key.offset,
7372 key.offset + extent_len - 1, 0);
7374 ret = btrfs_inc_extent_ref(trans, root,
7375 new_extents[i].disk_bytenr,
7376 new_extents[i].disk_num_bytes,
7378 root->root_key.objectid,
7379 trans->transid, key.objectid);
7381 btrfs_release_path(root, path);
7383 inode_add_bytes(inode, extent_len);
7386 num_bytes -= extent_len;
7387 key.offset += extent_len;
7392 BUG_ON(i >= nr_extents);
7396 if (extent_locked) {
7397 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7398 lock_end, GFP_NOFS);
7402 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7403 key.offset >= search_end)
7410 btrfs_release_path(root, path);
7412 mutex_unlock(&inode->i_mutex);
7413 if (extent_locked) {
7414 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7415 lock_end, GFP_NOFS);
7422 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7423 struct btrfs_root *root,
7424 struct extent_buffer *buf, u64 orig_start)
7429 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7430 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7432 level = btrfs_header_level(buf);
7434 struct btrfs_leaf_ref *ref;
7435 struct btrfs_leaf_ref *orig_ref;
7437 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7441 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7443 btrfs_free_leaf_ref(root, orig_ref);
7447 ref->nritems = orig_ref->nritems;
7448 memcpy(ref->extents, orig_ref->extents,
7449 sizeof(ref->extents[0]) * ref->nritems);
7451 btrfs_free_leaf_ref(root, orig_ref);
7453 ref->root_gen = trans->transid;
7454 ref->bytenr = buf->start;
7455 ref->owner = btrfs_header_owner(buf);
7456 ref->generation = btrfs_header_generation(buf);
7458 ret = btrfs_add_leaf_ref(root, ref, 0);
7460 btrfs_free_leaf_ref(root, ref);
7465 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7466 struct extent_buffer *leaf,
7467 struct btrfs_block_group_cache *group,
7468 struct btrfs_root *target_root)
7470 struct btrfs_key key;
7471 struct inode *inode = NULL;
7472 struct btrfs_file_extent_item *fi;
7473 struct extent_state *cached_state = NULL;
7475 u64 skip_objectid = 0;
7479 nritems = btrfs_header_nritems(leaf);
7480 for (i = 0; i < nritems; i++) {
7481 btrfs_item_key_to_cpu(leaf, &key, i);
7482 if (key.objectid == skip_objectid ||
7483 key.type != BTRFS_EXTENT_DATA_KEY)
7485 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7486 if (btrfs_file_extent_type(leaf, fi) ==
7487 BTRFS_FILE_EXTENT_INLINE)
7489 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7491 if (!inode || btrfs_ino(inode) != key.objectid) {
7493 inode = btrfs_ilookup(target_root->fs_info->sb,
7494 key.objectid, target_root, 1);
7497 skip_objectid = key.objectid;
7500 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7502 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7503 key.offset + num_bytes - 1, 0, &cached_state,
7505 btrfs_drop_extent_cache(inode, key.offset,
7506 key.offset + num_bytes - 1, 1);
7507 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7508 key.offset + num_bytes - 1, &cached_state,
7516 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7517 struct btrfs_root *root,
7518 struct extent_buffer *leaf,
7519 struct btrfs_block_group_cache *group,
7520 struct inode *reloc_inode)
7522 struct btrfs_key key;
7523 struct btrfs_key extent_key;
7524 struct btrfs_file_extent_item *fi;
7525 struct btrfs_leaf_ref *ref;
7526 struct disk_extent *new_extent;
7535 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7539 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7543 nritems = btrfs_header_nritems(leaf);
7544 for (i = 0; i < nritems; i++) {
7545 btrfs_item_key_to_cpu(leaf, &key, i);
7546 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7548 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7549 if (btrfs_file_extent_type(leaf, fi) ==
7550 BTRFS_FILE_EXTENT_INLINE)
7552 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7553 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7558 if (bytenr >= group->key.objectid + group->key.offset ||
7559 bytenr + num_bytes <= group->key.objectid)
7562 extent_key.objectid = bytenr;
7563 extent_key.offset = num_bytes;
7564 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7566 ret = get_new_locations(reloc_inode, &extent_key,
7567 group->key.objectid, 1,
7568 &new_extent, &nr_extent);
7573 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7574 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7575 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7576 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7578 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7579 new_extent->disk_bytenr);
7580 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7581 new_extent->disk_num_bytes);
7582 btrfs_mark_buffer_dirty(leaf);
7584 ret = btrfs_inc_extent_ref(trans, root,
7585 new_extent->disk_bytenr,
7586 new_extent->disk_num_bytes,
7588 root->root_key.objectid,
7589 trans->transid, key.objectid);
7592 ret = btrfs_free_extent(trans, root,
7593 bytenr, num_bytes, leaf->start,
7594 btrfs_header_owner(leaf),
7595 btrfs_header_generation(leaf),
7601 BUG_ON(ext_index + 1 != ref->nritems);
7602 btrfs_free_leaf_ref(root, ref);
7606 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7607 struct btrfs_root *root)
7609 struct btrfs_root *reloc_root;
7612 if (root->reloc_root) {
7613 reloc_root = root->reloc_root;
7614 root->reloc_root = NULL;
7615 list_add(&reloc_root->dead_list,
7616 &root->fs_info->dead_reloc_roots);
7618 btrfs_set_root_bytenr(&reloc_root->root_item,
7619 reloc_root->node->start);
7620 btrfs_set_root_level(&root->root_item,
7621 btrfs_header_level(reloc_root->node));
7622 memset(&reloc_root->root_item.drop_progress, 0,
7623 sizeof(struct btrfs_disk_key));
7624 reloc_root->root_item.drop_level = 0;
7626 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7627 &reloc_root->root_key,
7628 &reloc_root->root_item);
7634 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7636 struct btrfs_trans_handle *trans;
7637 struct btrfs_root *reloc_root;
7638 struct btrfs_root *prev_root = NULL;
7639 struct list_head dead_roots;
7643 INIT_LIST_HEAD(&dead_roots);
7644 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7646 while (!list_empty(&dead_roots)) {
7647 reloc_root = list_entry(dead_roots.prev,
7648 struct btrfs_root, dead_list);
7649 list_del_init(&reloc_root->dead_list);
7651 BUG_ON(reloc_root->commit_root != NULL);
7653 trans = btrfs_join_transaction(root, 1);
7654 BUG_ON(IS_ERR(trans));
7656 mutex_lock(&root->fs_info->drop_mutex);
7657 ret = btrfs_drop_snapshot(trans, reloc_root);
7660 mutex_unlock(&root->fs_info->drop_mutex);
7662 nr = trans->blocks_used;
7663 ret = btrfs_end_transaction(trans, root);
7665 btrfs_btree_balance_dirty(root, nr);
7668 free_extent_buffer(reloc_root->node);
7670 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7671 &reloc_root->root_key);
7673 mutex_unlock(&root->fs_info->drop_mutex);
7675 nr = trans->blocks_used;
7676 ret = btrfs_end_transaction(trans, root);
7678 btrfs_btree_balance_dirty(root, nr);
7681 prev_root = reloc_root;
7684 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7690 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7692 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7696 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7698 struct btrfs_root *reloc_root;
7699 struct btrfs_trans_handle *trans;
7700 struct btrfs_key location;
7704 mutex_lock(&root->fs_info->tree_reloc_mutex);
7705 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7707 found = !list_empty(&root->fs_info->dead_reloc_roots);
7708 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7711 trans = btrfs_start_transaction(root, 1);
7712 BUG_ON(IS_ERR(trans));
7713 ret = btrfs_commit_transaction(trans, root);
7717 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7718 location.offset = (u64)-1;
7719 location.type = BTRFS_ROOT_ITEM_KEY;
7721 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7722 BUG_ON(!reloc_root);
7723 ret = btrfs_orphan_cleanup(reloc_root);
7728 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7729 struct btrfs_root *root)
7731 struct btrfs_root *reloc_root;
7732 struct extent_buffer *eb;
7733 struct btrfs_root_item *root_item;
7734 struct btrfs_key root_key;
7737 BUG_ON(!root->ref_cows);
7738 if (root->reloc_root)
7741 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7745 ret = btrfs_copy_root(trans, root, root->commit_root,
7746 &eb, BTRFS_TREE_RELOC_OBJECTID);
7749 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7750 root_key.offset = root->root_key.objectid;
7751 root_key.type = BTRFS_ROOT_ITEM_KEY;
7753 memcpy(root_item, &root->root_item, sizeof(root_item));
7754 btrfs_set_root_refs(root_item, 0);
7755 btrfs_set_root_bytenr(root_item, eb->start);
7756 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7757 btrfs_set_root_generation(root_item, trans->transid);
7759 btrfs_tree_unlock(eb);
7760 free_extent_buffer(eb);
7762 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7763 &root_key, root_item);
7767 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7769 BUG_ON(IS_ERR(reloc_root));
7770 reloc_root->last_trans = trans->transid;
7771 reloc_root->commit_root = NULL;
7772 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7774 root->reloc_root = reloc_root;
7779 * Core function of space balance.
7781 * The idea is using reloc trees to relocate tree blocks in reference
7782 * counted roots. There is one reloc tree for each subvol, and all
7783 * reloc trees share same root key objectid. Reloc trees are snapshots
7784 * of the latest committed roots of subvols (root->commit_root).
7786 * To relocate a tree block referenced by a subvol, there are two steps.
7787 * COW the block through subvol's reloc tree, then update block pointer
7788 * in the subvol to point to the new block. Since all reloc trees share
7789 * same root key objectid, doing special handing for tree blocks owned
7790 * by them is easy. Once a tree block has been COWed in one reloc tree,
7791 * we can use the resulting new block directly when the same block is
7792 * required to COW again through other reloc trees. By this way, relocated
7793 * tree blocks are shared between reloc trees, so they are also shared
7796 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7797 struct btrfs_root *root,
7798 struct btrfs_path *path,
7799 struct btrfs_key *first_key,
7800 struct btrfs_ref_path *ref_path,
7801 struct btrfs_block_group_cache *group,
7802 struct inode *reloc_inode)
7804 struct btrfs_root *reloc_root;
7805 struct extent_buffer *eb = NULL;
7806 struct btrfs_key *keys;
7810 int lowest_level = 0;
7813 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7814 lowest_level = ref_path->owner_objectid;
7816 if (!root->ref_cows) {
7817 path->lowest_level = lowest_level;
7818 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7820 path->lowest_level = 0;
7821 btrfs_release_path(root, path);
7825 mutex_lock(&root->fs_info->tree_reloc_mutex);
7826 ret = init_reloc_tree(trans, root);
7828 reloc_root = root->reloc_root;
7830 shared_level = ref_path->shared_level;
7831 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7833 keys = ref_path->node_keys;
7834 nodes = ref_path->new_nodes;
7835 memset(&keys[shared_level + 1], 0,
7836 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7837 memset(&nodes[shared_level + 1], 0,
7838 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7840 if (nodes[lowest_level] == 0) {
7841 path->lowest_level = lowest_level;
7842 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7845 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7846 eb = path->nodes[level];
7847 if (!eb || eb == reloc_root->node)
7849 nodes[level] = eb->start;
7851 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7853 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7856 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7857 eb = path->nodes[0];
7858 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7859 group, reloc_inode);
7862 btrfs_release_path(reloc_root, path);
7864 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7870 * replace tree blocks in the fs tree with tree blocks in
7873 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7876 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7877 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7880 extent_buffer_get(path->nodes[0]);
7881 eb = path->nodes[0];
7882 btrfs_release_path(reloc_root, path);
7883 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7885 free_extent_buffer(eb);
7888 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7889 path->lowest_level = 0;
7893 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7894 struct btrfs_root *root,
7895 struct btrfs_path *path,
7896 struct btrfs_key *first_key,
7897 struct btrfs_ref_path *ref_path)
7901 ret = relocate_one_path(trans, root, path, first_key,
7902 ref_path, NULL, NULL);
7908 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7909 struct btrfs_root *extent_root,
7910 struct btrfs_path *path,
7911 struct btrfs_key *extent_key)
7915 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7918 ret = btrfs_del_item(trans, extent_root, path);
7920 btrfs_release_path(extent_root, path);
7924 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7925 struct btrfs_ref_path *ref_path)
7927 struct btrfs_key root_key;
7929 root_key.objectid = ref_path->root_objectid;
7930 root_key.type = BTRFS_ROOT_ITEM_KEY;
7931 if (is_cowonly_root(ref_path->root_objectid))
7932 root_key.offset = 0;
7934 root_key.offset = (u64)-1;
7936 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7939 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7940 struct btrfs_path *path,
7941 struct btrfs_key *extent_key,
7942 struct btrfs_block_group_cache *group,
7943 struct inode *reloc_inode, int pass)
7945 struct btrfs_trans_handle *trans;
7946 struct btrfs_root *found_root;
7947 struct btrfs_ref_path *ref_path = NULL;
7948 struct disk_extent *new_extents = NULL;
7953 struct btrfs_key first_key;
7957 trans = btrfs_start_transaction(extent_root, 1);
7958 BUG_ON(IS_ERR(trans));
7960 if (extent_key->objectid == 0) {
7961 ret = del_extent_zero(trans, extent_root, path, extent_key);
7965 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7971 for (loops = 0; ; loops++) {
7973 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7974 extent_key->objectid);
7976 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7983 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7984 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7987 found_root = read_ref_root(extent_root->fs_info, ref_path);
7988 BUG_ON(!found_root);
7990 * for reference counted tree, only process reference paths
7991 * rooted at the latest committed root.
7993 if (found_root->ref_cows &&
7994 ref_path->root_generation != found_root->root_key.offset)
7997 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8000 * copy data extents to new locations
8002 u64 group_start = group->key.objectid;
8003 ret = relocate_data_extent(reloc_inode,
8012 level = ref_path->owner_objectid;
8015 if (prev_block != ref_path->nodes[level]) {
8016 struct extent_buffer *eb;
8017 u64 block_start = ref_path->nodes[level];
8018 u64 block_size = btrfs_level_size(found_root, level);
8020 eb = read_tree_block(found_root, block_start,
8026 btrfs_tree_lock(eb);
8027 BUG_ON(level != btrfs_header_level(eb));
8030 btrfs_item_key_to_cpu(eb, &first_key, 0);
8032 btrfs_node_key_to_cpu(eb, &first_key, 0);
8034 btrfs_tree_unlock(eb);
8035 free_extent_buffer(eb);
8036 prev_block = block_start;
8039 mutex_lock(&extent_root->fs_info->trans_mutex);
8040 btrfs_record_root_in_trans(found_root);
8041 mutex_unlock(&extent_root->fs_info->trans_mutex);
8042 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8044 * try to update data extent references while
8045 * keeping metadata shared between snapshots.
8048 ret = relocate_one_path(trans, found_root,
8049 path, &first_key, ref_path,
8050 group, reloc_inode);
8056 * use fallback method to process the remaining
8060 u64 group_start = group->key.objectid;
8061 new_extents = kmalloc(sizeof(*new_extents),
8064 ret = get_new_locations(reloc_inode,
8072 ret = replace_one_extent(trans, found_root,
8074 &first_key, ref_path,
8075 new_extents, nr_extents);
8077 ret = relocate_tree_block(trans, found_root, path,
8078 &first_key, ref_path);
8085 btrfs_end_transaction(trans, extent_root);
8092 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8095 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8096 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8099 * we add in the count of missing devices because we want
8100 * to make sure that any RAID levels on a degraded FS
8101 * continue to be honored.
8103 num_devices = root->fs_info->fs_devices->rw_devices +
8104 root->fs_info->fs_devices->missing_devices;
8106 if (num_devices == 1) {
8107 stripped |= BTRFS_BLOCK_GROUP_DUP;
8108 stripped = flags & ~stripped;
8110 /* turn raid0 into single device chunks */
8111 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8114 /* turn mirroring into duplication */
8115 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8116 BTRFS_BLOCK_GROUP_RAID10))
8117 return stripped | BTRFS_BLOCK_GROUP_DUP;
8120 /* they already had raid on here, just return */
8121 if (flags & stripped)
8124 stripped |= BTRFS_BLOCK_GROUP_DUP;
8125 stripped = flags & ~stripped;
8127 /* switch duplicated blocks with raid1 */
8128 if (flags & BTRFS_BLOCK_GROUP_DUP)
8129 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8131 /* turn single device chunks into raid0 */
8132 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8137 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8139 struct btrfs_space_info *sinfo = cache->space_info;
8146 spin_lock(&sinfo->lock);
8147 spin_lock(&cache->lock);
8148 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8149 cache->bytes_super - btrfs_block_group_used(&cache->item);
8151 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8152 sinfo->bytes_may_use + sinfo->bytes_readonly +
8153 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8154 sinfo->bytes_readonly += num_bytes;
8155 sinfo->bytes_reserved += cache->reserved_pinned;
8156 cache->reserved_pinned = 0;
8161 spin_unlock(&cache->lock);
8162 spin_unlock(&sinfo->lock);
8166 int btrfs_set_block_group_ro(struct btrfs_root *root,
8167 struct btrfs_block_group_cache *cache)
8170 struct btrfs_trans_handle *trans;
8176 trans = btrfs_join_transaction(root, 1);
8177 BUG_ON(IS_ERR(trans));
8179 alloc_flags = update_block_group_flags(root, cache->flags);
8180 if (alloc_flags != cache->flags)
8181 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8184 ret = set_block_group_ro(cache);
8187 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8188 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8192 ret = set_block_group_ro(cache);
8194 btrfs_end_transaction(trans, root);
8198 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8199 struct btrfs_root *root, u64 type)
8201 u64 alloc_flags = get_alloc_profile(root, type);
8202 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8207 * helper to account the unused space of all the readonly block group in the
8208 * list. takes mirrors into account.
8210 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8212 struct btrfs_block_group_cache *block_group;
8216 list_for_each_entry(block_group, groups_list, list) {
8217 spin_lock(&block_group->lock);
8219 if (!block_group->ro) {
8220 spin_unlock(&block_group->lock);
8224 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8225 BTRFS_BLOCK_GROUP_RAID10 |
8226 BTRFS_BLOCK_GROUP_DUP))
8231 free_bytes += (block_group->key.offset -
8232 btrfs_block_group_used(&block_group->item)) *
8235 spin_unlock(&block_group->lock);
8242 * helper to account the unused space of all the readonly block group in the
8243 * space_info. takes mirrors into account.
8245 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8250 spin_lock(&sinfo->lock);
8252 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8253 if (!list_empty(&sinfo->block_groups[i]))
8254 free_bytes += __btrfs_get_ro_block_group_free_space(
8255 &sinfo->block_groups[i]);
8257 spin_unlock(&sinfo->lock);
8262 int btrfs_set_block_group_rw(struct btrfs_root *root,
8263 struct btrfs_block_group_cache *cache)
8265 struct btrfs_space_info *sinfo = cache->space_info;
8270 spin_lock(&sinfo->lock);
8271 spin_lock(&cache->lock);
8272 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8273 cache->bytes_super - btrfs_block_group_used(&cache->item);
8274 sinfo->bytes_readonly -= num_bytes;
8276 spin_unlock(&cache->lock);
8277 spin_unlock(&sinfo->lock);
8282 * checks to see if its even possible to relocate this block group.
8284 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8285 * ok to go ahead and try.
8287 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8289 struct btrfs_block_group_cache *block_group;
8290 struct btrfs_space_info *space_info;
8291 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8292 struct btrfs_device *device;
8296 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8298 /* odd, couldn't find the block group, leave it alone */
8302 /* no bytes used, we're good */
8303 if (!btrfs_block_group_used(&block_group->item))
8306 space_info = block_group->space_info;
8307 spin_lock(&space_info->lock);
8309 full = space_info->full;
8312 * if this is the last block group we have in this space, we can't
8313 * relocate it unless we're able to allocate a new chunk below.
8315 * Otherwise, we need to make sure we have room in the space to handle
8316 * all of the extents from this block group. If we can, we're good
8318 if ((space_info->total_bytes != block_group->key.offset) &&
8319 (space_info->bytes_used + space_info->bytes_reserved +
8320 space_info->bytes_pinned + space_info->bytes_readonly +
8321 btrfs_block_group_used(&block_group->item) <
8322 space_info->total_bytes)) {
8323 spin_unlock(&space_info->lock);
8326 spin_unlock(&space_info->lock);
8329 * ok we don't have enough space, but maybe we have free space on our
8330 * devices to allocate new chunks for relocation, so loop through our
8331 * alloc devices and guess if we have enough space. However, if we
8332 * were marked as full, then we know there aren't enough chunks, and we
8339 mutex_lock(&root->fs_info->chunk_mutex);
8340 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8341 u64 min_free = btrfs_block_group_used(&block_group->item);
8345 * check to make sure we can actually find a chunk with enough
8346 * space to fit our block group in.
8348 if (device->total_bytes > device->bytes_used + min_free) {
8349 ret = find_free_dev_extent(NULL, device, min_free,
8356 mutex_unlock(&root->fs_info->chunk_mutex);
8358 btrfs_put_block_group(block_group);
8362 static int find_first_block_group(struct btrfs_root *root,
8363 struct btrfs_path *path, struct btrfs_key *key)
8366 struct btrfs_key found_key;
8367 struct extent_buffer *leaf;
8370 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8375 slot = path->slots[0];
8376 leaf = path->nodes[0];
8377 if (slot >= btrfs_header_nritems(leaf)) {
8378 ret = btrfs_next_leaf(root, path);
8385 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8387 if (found_key.objectid >= key->objectid &&
8388 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8398 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8400 struct btrfs_block_group_cache *block_group;
8404 struct inode *inode;
8406 block_group = btrfs_lookup_first_block_group(info, last);
8407 while (block_group) {
8408 spin_lock(&block_group->lock);
8409 if (block_group->iref)
8411 spin_unlock(&block_group->lock);
8412 block_group = next_block_group(info->tree_root,
8422 inode = block_group->inode;
8423 block_group->iref = 0;
8424 block_group->inode = NULL;
8425 spin_unlock(&block_group->lock);
8427 last = block_group->key.objectid + block_group->key.offset;
8428 btrfs_put_block_group(block_group);
8432 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8434 struct btrfs_block_group_cache *block_group;
8435 struct btrfs_space_info *space_info;
8436 struct btrfs_caching_control *caching_ctl;
8439 down_write(&info->extent_commit_sem);
8440 while (!list_empty(&info->caching_block_groups)) {
8441 caching_ctl = list_entry(info->caching_block_groups.next,
8442 struct btrfs_caching_control, list);
8443 list_del(&caching_ctl->list);
8444 put_caching_control(caching_ctl);
8446 up_write(&info->extent_commit_sem);
8448 spin_lock(&info->block_group_cache_lock);
8449 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8450 block_group = rb_entry(n, struct btrfs_block_group_cache,
8452 rb_erase(&block_group->cache_node,
8453 &info->block_group_cache_tree);
8454 spin_unlock(&info->block_group_cache_lock);
8456 down_write(&block_group->space_info->groups_sem);
8457 list_del(&block_group->list);
8458 up_write(&block_group->space_info->groups_sem);
8460 if (block_group->cached == BTRFS_CACHE_STARTED)
8461 wait_block_group_cache_done(block_group);
8464 * We haven't cached this block group, which means we could
8465 * possibly have excluded extents on this block group.
8467 if (block_group->cached == BTRFS_CACHE_NO)
8468 free_excluded_extents(info->extent_root, block_group);
8470 btrfs_remove_free_space_cache(block_group);
8471 btrfs_put_block_group(block_group);
8473 spin_lock(&info->block_group_cache_lock);
8475 spin_unlock(&info->block_group_cache_lock);
8477 /* now that all the block groups are freed, go through and
8478 * free all the space_info structs. This is only called during
8479 * the final stages of unmount, and so we know nobody is
8480 * using them. We call synchronize_rcu() once before we start,
8481 * just to be on the safe side.
8485 release_global_block_rsv(info);
8487 while(!list_empty(&info->space_info)) {
8488 space_info = list_entry(info->space_info.next,
8489 struct btrfs_space_info,
8491 if (space_info->bytes_pinned > 0 ||
8492 space_info->bytes_reserved > 0) {
8494 dump_space_info(space_info, 0, 0);
8496 list_del(&space_info->list);
8502 static void __link_block_group(struct btrfs_space_info *space_info,
8503 struct btrfs_block_group_cache *cache)
8505 int index = get_block_group_index(cache);
8507 down_write(&space_info->groups_sem);
8508 list_add_tail(&cache->list, &space_info->block_groups[index]);
8509 up_write(&space_info->groups_sem);
8512 int btrfs_read_block_groups(struct btrfs_root *root)
8514 struct btrfs_path *path;
8516 struct btrfs_block_group_cache *cache;
8517 struct btrfs_fs_info *info = root->fs_info;
8518 struct btrfs_space_info *space_info;
8519 struct btrfs_key key;
8520 struct btrfs_key found_key;
8521 struct extent_buffer *leaf;
8525 root = info->extent_root;
8528 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8529 path = btrfs_alloc_path();
8533 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8534 if (cache_gen != 0 &&
8535 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8537 if (btrfs_test_opt(root, CLEAR_CACHE))
8539 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8540 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8543 ret = find_first_block_group(root, path, &key);
8548 leaf = path->nodes[0];
8549 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8550 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8555 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8557 if (!cache->free_space_ctl) {
8563 atomic_set(&cache->count, 1);
8564 spin_lock_init(&cache->lock);
8565 cache->fs_info = info;
8566 INIT_LIST_HEAD(&cache->list);
8567 INIT_LIST_HEAD(&cache->cluster_list);
8570 cache->disk_cache_state = BTRFS_DC_CLEAR;
8572 read_extent_buffer(leaf, &cache->item,
8573 btrfs_item_ptr_offset(leaf, path->slots[0]),
8574 sizeof(cache->item));
8575 memcpy(&cache->key, &found_key, sizeof(found_key));
8577 key.objectid = found_key.objectid + found_key.offset;
8578 btrfs_release_path(root, path);
8579 cache->flags = btrfs_block_group_flags(&cache->item);
8580 cache->sectorsize = root->sectorsize;
8582 btrfs_init_free_space_ctl(cache);
8585 * We need to exclude the super stripes now so that the space
8586 * info has super bytes accounted for, otherwise we'll think
8587 * we have more space than we actually do.
8589 exclude_super_stripes(root, cache);
8592 * check for two cases, either we are full, and therefore
8593 * don't need to bother with the caching work since we won't
8594 * find any space, or we are empty, and we can just add all
8595 * the space in and be done with it. This saves us _alot_ of
8596 * time, particularly in the full case.
8598 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8599 cache->last_byte_to_unpin = (u64)-1;
8600 cache->cached = BTRFS_CACHE_FINISHED;
8601 free_excluded_extents(root, cache);
8602 } else if (btrfs_block_group_used(&cache->item) == 0) {
8603 cache->last_byte_to_unpin = (u64)-1;
8604 cache->cached = BTRFS_CACHE_FINISHED;
8605 add_new_free_space(cache, root->fs_info,
8607 found_key.objectid +
8609 free_excluded_extents(root, cache);
8612 ret = update_space_info(info, cache->flags, found_key.offset,
8613 btrfs_block_group_used(&cache->item),
8616 cache->space_info = space_info;
8617 spin_lock(&cache->space_info->lock);
8618 cache->space_info->bytes_readonly += cache->bytes_super;
8619 spin_unlock(&cache->space_info->lock);
8621 __link_block_group(space_info, cache);
8623 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8626 set_avail_alloc_bits(root->fs_info, cache->flags);
8627 if (btrfs_chunk_readonly(root, cache->key.objectid))
8628 set_block_group_ro(cache);
8631 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8632 if (!(get_alloc_profile(root, space_info->flags) &
8633 (BTRFS_BLOCK_GROUP_RAID10 |
8634 BTRFS_BLOCK_GROUP_RAID1 |
8635 BTRFS_BLOCK_GROUP_DUP)))
8638 * avoid allocating from un-mirrored block group if there are
8639 * mirrored block groups.
8641 list_for_each_entry(cache, &space_info->block_groups[3], list)
8642 set_block_group_ro(cache);
8643 list_for_each_entry(cache, &space_info->block_groups[4], list)
8644 set_block_group_ro(cache);
8647 init_global_block_rsv(info);
8650 btrfs_free_path(path);
8654 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8655 struct btrfs_root *root, u64 bytes_used,
8656 u64 type, u64 chunk_objectid, u64 chunk_offset,
8660 struct btrfs_root *extent_root;
8661 struct btrfs_block_group_cache *cache;
8663 extent_root = root->fs_info->extent_root;
8665 root->fs_info->last_trans_log_full_commit = trans->transid;
8667 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8670 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8672 if (!cache->free_space_ctl) {
8677 cache->key.objectid = chunk_offset;
8678 cache->key.offset = size;
8679 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8680 cache->sectorsize = root->sectorsize;
8681 cache->fs_info = root->fs_info;
8683 atomic_set(&cache->count, 1);
8684 spin_lock_init(&cache->lock);
8685 INIT_LIST_HEAD(&cache->list);
8686 INIT_LIST_HEAD(&cache->cluster_list);
8688 btrfs_init_free_space_ctl(cache);
8690 btrfs_set_block_group_used(&cache->item, bytes_used);
8691 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8692 cache->flags = type;
8693 btrfs_set_block_group_flags(&cache->item, type);
8695 cache->last_byte_to_unpin = (u64)-1;
8696 cache->cached = BTRFS_CACHE_FINISHED;
8697 exclude_super_stripes(root, cache);
8699 add_new_free_space(cache, root->fs_info, chunk_offset,
8700 chunk_offset + size);
8702 free_excluded_extents(root, cache);
8704 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8705 &cache->space_info);
8708 spin_lock(&cache->space_info->lock);
8709 cache->space_info->bytes_readonly += cache->bytes_super;
8710 spin_unlock(&cache->space_info->lock);
8712 __link_block_group(cache->space_info, cache);
8714 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8717 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8718 sizeof(cache->item));
8721 set_avail_alloc_bits(extent_root->fs_info, type);
8726 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8727 struct btrfs_root *root, u64 group_start)
8729 struct btrfs_path *path;
8730 struct btrfs_block_group_cache *block_group;
8731 struct btrfs_free_cluster *cluster;
8732 struct btrfs_root *tree_root = root->fs_info->tree_root;
8733 struct btrfs_key key;
8734 struct inode *inode;
8738 root = root->fs_info->extent_root;
8740 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8741 BUG_ON(!block_group);
8742 BUG_ON(!block_group->ro);
8745 * Free the reserved super bytes from this block group before
8748 free_excluded_extents(root, block_group);
8750 memcpy(&key, &block_group->key, sizeof(key));
8751 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8752 BTRFS_BLOCK_GROUP_RAID1 |
8753 BTRFS_BLOCK_GROUP_RAID10))
8758 /* make sure this block group isn't part of an allocation cluster */
8759 cluster = &root->fs_info->data_alloc_cluster;
8760 spin_lock(&cluster->refill_lock);
8761 btrfs_return_cluster_to_free_space(block_group, cluster);
8762 spin_unlock(&cluster->refill_lock);
8765 * make sure this block group isn't part of a metadata
8766 * allocation cluster
8768 cluster = &root->fs_info->meta_alloc_cluster;
8769 spin_lock(&cluster->refill_lock);
8770 btrfs_return_cluster_to_free_space(block_group, cluster);
8771 spin_unlock(&cluster->refill_lock);
8773 path = btrfs_alloc_path();
8776 inode = lookup_free_space_inode(root, block_group, path);
8777 if (!IS_ERR(inode)) {
8778 btrfs_orphan_add(trans, inode);
8780 /* One for the block groups ref */
8781 spin_lock(&block_group->lock);
8782 if (block_group->iref) {
8783 block_group->iref = 0;
8784 block_group->inode = NULL;
8785 spin_unlock(&block_group->lock);
8788 spin_unlock(&block_group->lock);
8790 /* One for our lookup ref */
8794 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8795 key.offset = block_group->key.objectid;
8798 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8802 btrfs_release_path(tree_root, path);
8804 ret = btrfs_del_item(trans, tree_root, path);
8807 btrfs_release_path(tree_root, path);
8810 spin_lock(&root->fs_info->block_group_cache_lock);
8811 rb_erase(&block_group->cache_node,
8812 &root->fs_info->block_group_cache_tree);
8813 spin_unlock(&root->fs_info->block_group_cache_lock);
8815 down_write(&block_group->space_info->groups_sem);
8817 * we must use list_del_init so people can check to see if they
8818 * are still on the list after taking the semaphore
8820 list_del_init(&block_group->list);
8821 up_write(&block_group->space_info->groups_sem);
8823 if (block_group->cached == BTRFS_CACHE_STARTED)
8824 wait_block_group_cache_done(block_group);
8826 btrfs_remove_free_space_cache(block_group);
8828 spin_lock(&block_group->space_info->lock);
8829 block_group->space_info->total_bytes -= block_group->key.offset;
8830 block_group->space_info->bytes_readonly -= block_group->key.offset;
8831 block_group->space_info->disk_total -= block_group->key.offset * factor;
8832 spin_unlock(&block_group->space_info->lock);
8834 memcpy(&key, &block_group->key, sizeof(key));
8836 btrfs_clear_space_info_full(root->fs_info);
8838 btrfs_put_block_group(block_group);
8839 btrfs_put_block_group(block_group);
8841 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8847 ret = btrfs_del_item(trans, root, path);
8849 btrfs_free_path(path);
8853 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8855 struct btrfs_space_info *space_info;
8858 ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM, 0, 0,
8863 ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA, 0, 0,
8868 ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA, 0, 0,
8876 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8878 return unpin_extent_range(root, start, end);
8881 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8882 u64 num_bytes, u64 *actual_bytes)
8884 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8887 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8889 struct btrfs_fs_info *fs_info = root->fs_info;
8890 struct btrfs_block_group_cache *cache = NULL;
8897 cache = btrfs_lookup_block_group(fs_info, range->start);
8900 if (cache->key.objectid >= (range->start + range->len)) {
8901 btrfs_put_block_group(cache);
8905 start = max(range->start, cache->key.objectid);
8906 end = min(range->start + range->len,
8907 cache->key.objectid + cache->key.offset);
8909 if (end - start >= range->minlen) {
8910 if (!block_group_cache_done(cache)) {
8911 ret = cache_block_group(cache, NULL, root, 0);
8913 wait_block_group_cache_done(cache);
8915 ret = btrfs_trim_block_group(cache,
8921 trimmed += group_trimmed;
8923 btrfs_put_block_group(cache);
8928 cache = next_block_group(fs_info->tree_root, cache);
8931 range->len = trimmed;