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 static int update_block_group(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 u64 bytenr, u64 num_bytes, int alloc);
39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve, int sinfo);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int find_next_key(struct btrfs_path *path, int level,
64 struct btrfs_key *key);
65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
66 int dump_block_groups);
69 block_group_cache_done(struct btrfs_block_group_cache *cache)
72 return cache->cached == BTRFS_CACHE_FINISHED;
75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
77 return (cache->flags & bits) == bits;
80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
82 atomic_inc(&cache->count);
85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
87 if (atomic_dec_and_test(&cache->count)) {
88 WARN_ON(cache->pinned > 0);
89 WARN_ON(cache->reserved > 0);
90 WARN_ON(cache->reserved_pinned > 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
100 struct btrfs_block_group_cache *block_group)
103 struct rb_node *parent = NULL;
104 struct btrfs_block_group_cache *cache;
106 spin_lock(&info->block_group_cache_lock);
107 p = &info->block_group_cache_tree.rb_node;
111 cache = rb_entry(parent, struct btrfs_block_group_cache,
113 if (block_group->key.objectid < cache->key.objectid) {
115 } else if (block_group->key.objectid > cache->key.objectid) {
118 spin_unlock(&info->block_group_cache_lock);
123 rb_link_node(&block_group->cache_node, parent, p);
124 rb_insert_color(&block_group->cache_node,
125 &info->block_group_cache_tree);
126 spin_unlock(&info->block_group_cache_lock);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache *
136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
139 struct btrfs_block_group_cache *cache, *ret = NULL;
143 spin_lock(&info->block_group_cache_lock);
144 n = info->block_group_cache_tree.rb_node;
147 cache = rb_entry(n, struct btrfs_block_group_cache,
149 end = cache->key.objectid + cache->key.offset - 1;
150 start = cache->key.objectid;
152 if (bytenr < start) {
153 if (!contains && (!ret || start < ret->key.objectid))
156 } else if (bytenr > start) {
157 if (contains && bytenr <= end) {
168 btrfs_get_block_group(ret);
169 spin_unlock(&info->block_group_cache_lock);
174 static int add_excluded_extent(struct btrfs_root *root,
175 u64 start, u64 num_bytes)
177 u64 end = start + num_bytes - 1;
178 set_extent_bits(&root->fs_info->freed_extents[0],
179 start, end, EXTENT_UPTODATE, GFP_NOFS);
180 set_extent_bits(&root->fs_info->freed_extents[1],
181 start, end, EXTENT_UPTODATE, GFP_NOFS);
185 static void free_excluded_extents(struct btrfs_root *root,
186 struct btrfs_block_group_cache *cache)
190 start = cache->key.objectid;
191 end = start + cache->key.offset - 1;
193 clear_extent_bits(&root->fs_info->freed_extents[0],
194 start, end, EXTENT_UPTODATE, GFP_NOFS);
195 clear_extent_bits(&root->fs_info->freed_extents[1],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
199 static int exclude_super_stripes(struct btrfs_root *root,
200 struct btrfs_block_group_cache *cache)
207 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
208 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
209 cache->bytes_super += stripe_len;
210 ret = add_excluded_extent(root, cache->key.objectid,
215 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
216 bytenr = btrfs_sb_offset(i);
217 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
218 cache->key.objectid, bytenr,
219 0, &logical, &nr, &stripe_len);
223 cache->bytes_super += stripe_len;
224 ret = add_excluded_extent(root, logical[nr],
234 static struct btrfs_caching_control *
235 get_caching_control(struct btrfs_block_group_cache *cache)
237 struct btrfs_caching_control *ctl;
239 spin_lock(&cache->lock);
240 if (cache->cached != BTRFS_CACHE_STARTED) {
241 spin_unlock(&cache->lock);
245 ctl = cache->caching_ctl;
246 atomic_inc(&ctl->count);
247 spin_unlock(&cache->lock);
251 static void put_caching_control(struct btrfs_caching_control *ctl)
253 if (atomic_dec_and_test(&ctl->count))
258 * this is only called by cache_block_group, since we could have freed extents
259 * we need to check the pinned_extents for any extents that can't be used yet
260 * since their free space will be released as soon as the transaction commits.
262 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
263 struct btrfs_fs_info *info, u64 start, u64 end)
265 u64 extent_start, extent_end, size, total_added = 0;
268 while (start < end) {
269 ret = find_first_extent_bit(info->pinned_extents, start,
270 &extent_start, &extent_end,
271 EXTENT_DIRTY | EXTENT_UPTODATE);
275 if (extent_start <= start) {
276 start = extent_end + 1;
277 } else if (extent_start > start && extent_start < end) {
278 size = extent_start - start;
280 ret = btrfs_add_free_space(block_group, start,
283 start = extent_end + 1;
292 ret = btrfs_add_free_space(block_group, start, size);
299 static int caching_kthread(void *data)
301 struct btrfs_block_group_cache *block_group = data;
302 struct btrfs_fs_info *fs_info = block_group->fs_info;
303 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
304 struct btrfs_root *extent_root = fs_info->extent_root;
305 struct btrfs_path *path;
306 struct extent_buffer *leaf;
307 struct btrfs_key key;
313 path = btrfs_alloc_path();
317 exclude_super_stripes(extent_root, block_group);
318 spin_lock(&block_group->space_info->lock);
319 block_group->space_info->bytes_readonly += block_group->bytes_super;
320 spin_unlock(&block_group->space_info->lock);
322 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
325 * We don't want to deadlock with somebody trying to allocate a new
326 * extent for the extent root while also trying to search the extent
327 * root to add free space. So we skip locking and search the commit
328 * root, since its read-only
330 path->skip_locking = 1;
331 path->search_commit_root = 1;
336 key.type = BTRFS_EXTENT_ITEM_KEY;
338 mutex_lock(&caching_ctl->mutex);
339 /* need to make sure the commit_root doesn't disappear */
340 down_read(&fs_info->extent_commit_sem);
342 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
346 leaf = path->nodes[0];
347 nritems = btrfs_header_nritems(leaf);
351 if (fs_info->closing > 1) {
356 if (path->slots[0] < nritems) {
357 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
359 ret = find_next_key(path, 0, &key);
363 caching_ctl->progress = last;
364 btrfs_release_path(extent_root, path);
365 up_read(&fs_info->extent_commit_sem);
366 mutex_unlock(&caching_ctl->mutex);
367 if (btrfs_transaction_in_commit(fs_info))
374 if (key.objectid < block_group->key.objectid) {
379 if (key.objectid >= block_group->key.objectid +
380 block_group->key.offset)
383 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
384 total_found += add_new_free_space(block_group,
387 last = key.objectid + key.offset;
389 if (total_found > (1024 * 1024 * 2)) {
391 wake_up(&caching_ctl->wait);
398 total_found += add_new_free_space(block_group, fs_info, last,
399 block_group->key.objectid +
400 block_group->key.offset);
401 caching_ctl->progress = (u64)-1;
403 spin_lock(&block_group->lock);
404 block_group->caching_ctl = NULL;
405 block_group->cached = BTRFS_CACHE_FINISHED;
406 spin_unlock(&block_group->lock);
409 btrfs_free_path(path);
410 up_read(&fs_info->extent_commit_sem);
412 free_excluded_extents(extent_root, block_group);
414 mutex_unlock(&caching_ctl->mutex);
415 wake_up(&caching_ctl->wait);
417 put_caching_control(caching_ctl);
418 atomic_dec(&block_group->space_info->caching_threads);
419 btrfs_put_block_group(block_group);
424 static int cache_block_group(struct btrfs_block_group_cache *cache)
426 struct btrfs_fs_info *fs_info = cache->fs_info;
427 struct btrfs_caching_control *caching_ctl;
428 struct task_struct *tsk;
432 if (cache->cached != BTRFS_CACHE_NO)
435 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
436 BUG_ON(!caching_ctl);
438 INIT_LIST_HEAD(&caching_ctl->list);
439 mutex_init(&caching_ctl->mutex);
440 init_waitqueue_head(&caching_ctl->wait);
441 caching_ctl->block_group = cache;
442 caching_ctl->progress = cache->key.objectid;
443 /* one for caching kthread, one for caching block group list */
444 atomic_set(&caching_ctl->count, 2);
446 spin_lock(&cache->lock);
447 if (cache->cached != BTRFS_CACHE_NO) {
448 spin_unlock(&cache->lock);
452 cache->caching_ctl = caching_ctl;
453 cache->cached = BTRFS_CACHE_STARTED;
454 spin_unlock(&cache->lock);
456 down_write(&fs_info->extent_commit_sem);
457 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
458 up_write(&fs_info->extent_commit_sem);
460 atomic_inc(&cache->space_info->caching_threads);
461 btrfs_get_block_group(cache);
463 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
464 cache->key.objectid);
467 printk(KERN_ERR "error running thread %d\n", ret);
475 * return the block group that starts at or after bytenr
477 static struct btrfs_block_group_cache *
478 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
480 struct btrfs_block_group_cache *cache;
482 cache = block_group_cache_tree_search(info, bytenr, 0);
488 * return the block group that contains the given bytenr
490 struct btrfs_block_group_cache *btrfs_lookup_block_group(
491 struct btrfs_fs_info *info,
494 struct btrfs_block_group_cache *cache;
496 cache = block_group_cache_tree_search(info, bytenr, 1);
501 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
504 struct list_head *head = &info->space_info;
505 struct btrfs_space_info *found;
507 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
508 BTRFS_BLOCK_GROUP_METADATA;
511 list_for_each_entry_rcu(found, head, list) {
512 if (found->flags == flags) {
522 * after adding space to the filesystem, we need to clear the full flags
523 * on all the space infos.
525 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
527 struct list_head *head = &info->space_info;
528 struct btrfs_space_info *found;
531 list_for_each_entry_rcu(found, head, list)
536 static u64 div_factor(u64 num, int factor)
545 u64 btrfs_find_block_group(struct btrfs_root *root,
546 u64 search_start, u64 search_hint, int owner)
548 struct btrfs_block_group_cache *cache;
550 u64 last = max(search_hint, search_start);
557 cache = btrfs_lookup_first_block_group(root->fs_info, last);
561 spin_lock(&cache->lock);
562 last = cache->key.objectid + cache->key.offset;
563 used = btrfs_block_group_used(&cache->item);
565 if ((full_search || !cache->ro) &&
566 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
567 if (used + cache->pinned + cache->reserved <
568 div_factor(cache->key.offset, factor)) {
569 group_start = cache->key.objectid;
570 spin_unlock(&cache->lock);
571 btrfs_put_block_group(cache);
575 spin_unlock(&cache->lock);
576 btrfs_put_block_group(cache);
584 if (!full_search && factor < 10) {
594 /* simple helper to search for an existing extent at a given offset */
595 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
598 struct btrfs_key key;
599 struct btrfs_path *path;
601 path = btrfs_alloc_path();
603 key.objectid = start;
605 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
606 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
608 btrfs_free_path(path);
613 * helper function to lookup reference count and flags of extent.
615 * the head node for delayed ref is used to store the sum of all the
616 * reference count modifications queued up in the rbtree. the head
617 * node may also store the extent flags to set. This way you can check
618 * to see what the reference count and extent flags would be if all of
619 * the delayed refs are not processed.
621 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root, u64 bytenr,
623 u64 num_bytes, u64 *refs, u64 *flags)
625 struct btrfs_delayed_ref_head *head;
626 struct btrfs_delayed_ref_root *delayed_refs;
627 struct btrfs_path *path;
628 struct btrfs_extent_item *ei;
629 struct extent_buffer *leaf;
630 struct btrfs_key key;
636 path = btrfs_alloc_path();
640 key.objectid = bytenr;
641 key.type = BTRFS_EXTENT_ITEM_KEY;
642 key.offset = num_bytes;
644 path->skip_locking = 1;
645 path->search_commit_root = 1;
648 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
654 leaf = path->nodes[0];
655 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
656 if (item_size >= sizeof(*ei)) {
657 ei = btrfs_item_ptr(leaf, path->slots[0],
658 struct btrfs_extent_item);
659 num_refs = btrfs_extent_refs(leaf, ei);
660 extent_flags = btrfs_extent_flags(leaf, ei);
662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
663 struct btrfs_extent_item_v0 *ei0;
664 BUG_ON(item_size != sizeof(*ei0));
665 ei0 = btrfs_item_ptr(leaf, path->slots[0],
666 struct btrfs_extent_item_v0);
667 num_refs = btrfs_extent_refs_v0(leaf, ei0);
668 /* FIXME: this isn't correct for data */
669 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
674 BUG_ON(num_refs == 0);
684 delayed_refs = &trans->transaction->delayed_refs;
685 spin_lock(&delayed_refs->lock);
686 head = btrfs_find_delayed_ref_head(trans, bytenr);
688 if (!mutex_trylock(&head->mutex)) {
689 atomic_inc(&head->node.refs);
690 spin_unlock(&delayed_refs->lock);
692 btrfs_release_path(root->fs_info->extent_root, path);
694 mutex_lock(&head->mutex);
695 mutex_unlock(&head->mutex);
696 btrfs_put_delayed_ref(&head->node);
699 if (head->extent_op && head->extent_op->update_flags)
700 extent_flags |= head->extent_op->flags_to_set;
702 BUG_ON(num_refs == 0);
704 num_refs += head->node.ref_mod;
705 mutex_unlock(&head->mutex);
707 spin_unlock(&delayed_refs->lock);
709 WARN_ON(num_refs == 0);
713 *flags = extent_flags;
715 btrfs_free_path(path);
720 * Back reference rules. Back refs have three main goals:
722 * 1) differentiate between all holders of references to an extent so that
723 * when a reference is dropped we can make sure it was a valid reference
724 * before freeing the extent.
726 * 2) Provide enough information to quickly find the holders of an extent
727 * if we notice a given block is corrupted or bad.
729 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
730 * maintenance. This is actually the same as #2, but with a slightly
731 * different use case.
733 * There are two kinds of back refs. The implicit back refs is optimized
734 * for pointers in non-shared tree blocks. For a given pointer in a block,
735 * back refs of this kind provide information about the block's owner tree
736 * and the pointer's key. These information allow us to find the block by
737 * b-tree searching. The full back refs is for pointers in tree blocks not
738 * referenced by their owner trees. The location of tree block is recorded
739 * in the back refs. Actually the full back refs is generic, and can be
740 * used in all cases the implicit back refs is used. The major shortcoming
741 * of the full back refs is its overhead. Every time a tree block gets
742 * COWed, we have to update back refs entry for all pointers in it.
744 * For a newly allocated tree block, we use implicit back refs for
745 * pointers in it. This means most tree related operations only involve
746 * implicit back refs. For a tree block created in old transaction, the
747 * only way to drop a reference to it is COW it. So we can detect the
748 * event that tree block loses its owner tree's reference and do the
749 * back refs conversion.
751 * When a tree block is COW'd through a tree, there are four cases:
753 * The reference count of the block is one and the tree is the block's
754 * owner tree. Nothing to do in this case.
756 * The reference count of the block is one and the tree is not the
757 * block's owner tree. In this case, full back refs is used for pointers
758 * in the block. Remove these full back refs, add implicit back refs for
759 * every pointers in the new block.
761 * The reference count of the block is greater than one and the tree is
762 * the block's owner tree. In this case, implicit back refs is used for
763 * pointers in the block. Add full back refs for every pointers in the
764 * block, increase lower level extents' reference counts. The original
765 * implicit back refs are entailed to the new block.
767 * The reference count of the block is greater than one and the tree is
768 * not the block's owner tree. Add implicit back refs for every pointer in
769 * the new block, increase lower level extents' reference count.
771 * Back Reference Key composing:
773 * The key objectid corresponds to the first byte in the extent,
774 * The key type is used to differentiate between types of back refs.
775 * There are different meanings of the key offset for different types
778 * File extents can be referenced by:
780 * - multiple snapshots, subvolumes, or different generations in one subvol
781 * - different files inside a single subvolume
782 * - different offsets inside a file (bookend extents in file.c)
784 * The extent ref structure for the implicit back refs has fields for:
786 * - Objectid of the subvolume root
787 * - objectid of the file holding the reference
788 * - original offset in the file
789 * - how many bookend extents
791 * The key offset for the implicit back refs is hash of the first
794 * The extent ref structure for the full back refs has field for:
796 * - number of pointers in the tree leaf
798 * The key offset for the implicit back refs is the first byte of
801 * When a file extent is allocated, The implicit back refs is used.
802 * the fields are filled in:
804 * (root_key.objectid, inode objectid, offset in file, 1)
806 * When a file extent is removed file truncation, we find the
807 * corresponding implicit back refs and check the following fields:
809 * (btrfs_header_owner(leaf), inode objectid, offset in file)
811 * Btree extents can be referenced by:
813 * - Different subvolumes
815 * Both the implicit back refs and the full back refs for tree blocks
816 * only consist of key. The key offset for the implicit back refs is
817 * objectid of block's owner tree. The key offset for the full back refs
818 * is the first byte of parent block.
820 * When implicit back refs is used, information about the lowest key and
821 * level of the tree block are required. These information are stored in
822 * tree block info structure.
825 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
826 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_path *path,
829 u64 owner, u32 extra_size)
831 struct btrfs_extent_item *item;
832 struct btrfs_extent_item_v0 *ei0;
833 struct btrfs_extent_ref_v0 *ref0;
834 struct btrfs_tree_block_info *bi;
835 struct extent_buffer *leaf;
836 struct btrfs_key key;
837 struct btrfs_key found_key;
838 u32 new_size = sizeof(*item);
842 leaf = path->nodes[0];
843 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
845 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
846 ei0 = btrfs_item_ptr(leaf, path->slots[0],
847 struct btrfs_extent_item_v0);
848 refs = btrfs_extent_refs_v0(leaf, ei0);
850 if (owner == (u64)-1) {
852 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
853 ret = btrfs_next_leaf(root, path);
857 leaf = path->nodes[0];
859 btrfs_item_key_to_cpu(leaf, &found_key,
861 BUG_ON(key.objectid != found_key.objectid);
862 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
866 ref0 = btrfs_item_ptr(leaf, path->slots[0],
867 struct btrfs_extent_ref_v0);
868 owner = btrfs_ref_objectid_v0(leaf, ref0);
872 btrfs_release_path(root, path);
874 if (owner < BTRFS_FIRST_FREE_OBJECTID)
875 new_size += sizeof(*bi);
877 new_size -= sizeof(*ei0);
878 ret = btrfs_search_slot(trans, root, &key, path,
879 new_size + extra_size, 1);
884 ret = btrfs_extend_item(trans, root, path, new_size);
887 leaf = path->nodes[0];
888 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
889 btrfs_set_extent_refs(leaf, item, refs);
890 /* FIXME: get real generation */
891 btrfs_set_extent_generation(leaf, item, 0);
892 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
893 btrfs_set_extent_flags(leaf, item,
894 BTRFS_EXTENT_FLAG_TREE_BLOCK |
895 BTRFS_BLOCK_FLAG_FULL_BACKREF);
896 bi = (struct btrfs_tree_block_info *)(item + 1);
897 /* FIXME: get first key of the block */
898 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
899 btrfs_set_tree_block_level(leaf, bi, (int)owner);
901 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
903 btrfs_mark_buffer_dirty(leaf);
908 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
910 u32 high_crc = ~(u32)0;
911 u32 low_crc = ~(u32)0;
914 lenum = cpu_to_le64(root_objectid);
915 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
916 lenum = cpu_to_le64(owner);
917 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
918 lenum = cpu_to_le64(offset);
919 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
921 return ((u64)high_crc << 31) ^ (u64)low_crc;
924 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
925 struct btrfs_extent_data_ref *ref)
927 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
928 btrfs_extent_data_ref_objectid(leaf, ref),
929 btrfs_extent_data_ref_offset(leaf, ref));
932 static int match_extent_data_ref(struct extent_buffer *leaf,
933 struct btrfs_extent_data_ref *ref,
934 u64 root_objectid, u64 owner, u64 offset)
936 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
937 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
938 btrfs_extent_data_ref_offset(leaf, ref) != offset)
943 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
944 struct btrfs_root *root,
945 struct btrfs_path *path,
946 u64 bytenr, u64 parent,
948 u64 owner, u64 offset)
950 struct btrfs_key key;
951 struct btrfs_extent_data_ref *ref;
952 struct extent_buffer *leaf;
958 key.objectid = bytenr;
960 key.type = BTRFS_SHARED_DATA_REF_KEY;
963 key.type = BTRFS_EXTENT_DATA_REF_KEY;
964 key.offset = hash_extent_data_ref(root_objectid,
969 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 key.type = BTRFS_EXTENT_REF_V0_KEY;
980 btrfs_release_path(root, path);
981 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
992 leaf = path->nodes[0];
993 nritems = btrfs_header_nritems(leaf);
995 if (path->slots[0] >= nritems) {
996 ret = btrfs_next_leaf(root, path);
1002 leaf = path->nodes[0];
1003 nritems = btrfs_header_nritems(leaf);
1007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008 if (key.objectid != bytenr ||
1009 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1012 ref = btrfs_item_ptr(leaf, path->slots[0],
1013 struct btrfs_extent_data_ref);
1015 if (match_extent_data_ref(leaf, ref, root_objectid,
1018 btrfs_release_path(root, path);
1030 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_path *path,
1033 u64 bytenr, u64 parent,
1034 u64 root_objectid, u64 owner,
1035 u64 offset, int refs_to_add)
1037 struct btrfs_key key;
1038 struct extent_buffer *leaf;
1043 key.objectid = bytenr;
1045 key.type = BTRFS_SHARED_DATA_REF_KEY;
1046 key.offset = parent;
1047 size = sizeof(struct btrfs_shared_data_ref);
1049 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1050 key.offset = hash_extent_data_ref(root_objectid,
1052 size = sizeof(struct btrfs_extent_data_ref);
1055 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1056 if (ret && ret != -EEXIST)
1059 leaf = path->nodes[0];
1061 struct btrfs_shared_data_ref *ref;
1062 ref = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_shared_data_ref);
1065 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1067 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1068 num_refs += refs_to_add;
1069 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1072 struct btrfs_extent_data_ref *ref;
1073 while (ret == -EEXIST) {
1074 ref = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_extent_data_ref);
1076 if (match_extent_data_ref(leaf, ref, root_objectid,
1079 btrfs_release_path(root, path);
1081 ret = btrfs_insert_empty_item(trans, root, path, &key,
1083 if (ret && ret != -EEXIST)
1086 leaf = path->nodes[0];
1088 ref = btrfs_item_ptr(leaf, path->slots[0],
1089 struct btrfs_extent_data_ref);
1091 btrfs_set_extent_data_ref_root(leaf, ref,
1093 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1094 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1095 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1097 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1098 num_refs += refs_to_add;
1099 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1102 btrfs_mark_buffer_dirty(leaf);
1105 btrfs_release_path(root, path);
1109 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 struct btrfs_path *path,
1114 struct btrfs_key key;
1115 struct btrfs_extent_data_ref *ref1 = NULL;
1116 struct btrfs_shared_data_ref *ref2 = NULL;
1117 struct extent_buffer *leaf;
1121 leaf = path->nodes[0];
1122 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1124 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1125 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1127 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1128 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1129 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1130 struct btrfs_shared_data_ref);
1131 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1132 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1133 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1134 struct btrfs_extent_ref_v0 *ref0;
1135 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1136 struct btrfs_extent_ref_v0);
1137 num_refs = btrfs_ref_count_v0(leaf, ref0);
1143 BUG_ON(num_refs < refs_to_drop);
1144 num_refs -= refs_to_drop;
1146 if (num_refs == 0) {
1147 ret = btrfs_del_item(trans, root, path);
1149 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1150 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1151 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1152 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1155 struct btrfs_extent_ref_v0 *ref0;
1156 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1157 struct btrfs_extent_ref_v0);
1158 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1161 btrfs_mark_buffer_dirty(leaf);
1166 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1167 struct btrfs_path *path,
1168 struct btrfs_extent_inline_ref *iref)
1170 struct btrfs_key key;
1171 struct extent_buffer *leaf;
1172 struct btrfs_extent_data_ref *ref1;
1173 struct btrfs_shared_data_ref *ref2;
1176 leaf = path->nodes[0];
1177 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1179 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1180 BTRFS_EXTENT_DATA_REF_KEY) {
1181 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1182 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1184 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1185 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1187 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1188 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1189 struct btrfs_extent_data_ref);
1190 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1191 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1192 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_shared_data_ref);
1194 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1196 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1197 struct btrfs_extent_ref_v0 *ref0;
1198 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1199 struct btrfs_extent_ref_v0);
1200 num_refs = btrfs_ref_count_v0(leaf, ref0);
1208 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root,
1210 struct btrfs_path *path,
1211 u64 bytenr, u64 parent,
1214 struct btrfs_key key;
1217 key.objectid = bytenr;
1219 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1220 key.offset = parent;
1222 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1223 key.offset = root_objectid;
1226 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1229 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1230 if (ret == -ENOENT && parent) {
1231 btrfs_release_path(root, path);
1232 key.type = BTRFS_EXTENT_REF_V0_KEY;
1233 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1241 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1242 struct btrfs_root *root,
1243 struct btrfs_path *path,
1244 u64 bytenr, u64 parent,
1247 struct btrfs_key key;
1250 key.objectid = bytenr;
1252 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1253 key.offset = parent;
1255 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1256 key.offset = root_objectid;
1259 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1260 btrfs_release_path(root, path);
1264 static inline int extent_ref_type(u64 parent, u64 owner)
1267 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1269 type = BTRFS_SHARED_BLOCK_REF_KEY;
1271 type = BTRFS_TREE_BLOCK_REF_KEY;
1274 type = BTRFS_SHARED_DATA_REF_KEY;
1276 type = BTRFS_EXTENT_DATA_REF_KEY;
1281 static int find_next_key(struct btrfs_path *path, int level,
1282 struct btrfs_key *key)
1285 for (; level < BTRFS_MAX_LEVEL; level++) {
1286 if (!path->nodes[level])
1288 if (path->slots[level] + 1 >=
1289 btrfs_header_nritems(path->nodes[level]))
1292 btrfs_item_key_to_cpu(path->nodes[level], key,
1293 path->slots[level] + 1);
1295 btrfs_node_key_to_cpu(path->nodes[level], key,
1296 path->slots[level] + 1);
1303 * look for inline back ref. if back ref is found, *ref_ret is set
1304 * to the address of inline back ref, and 0 is returned.
1306 * if back ref isn't found, *ref_ret is set to the address where it
1307 * should be inserted, and -ENOENT is returned.
1309 * if insert is true and there are too many inline back refs, the path
1310 * points to the extent item, and -EAGAIN is returned.
1312 * NOTE: inline back refs are ordered in the same way that back ref
1313 * items in the tree are ordered.
1315 static noinline_for_stack
1316 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root,
1318 struct btrfs_path *path,
1319 struct btrfs_extent_inline_ref **ref_ret,
1320 u64 bytenr, u64 num_bytes,
1321 u64 parent, u64 root_objectid,
1322 u64 owner, u64 offset, int insert)
1324 struct btrfs_key key;
1325 struct extent_buffer *leaf;
1326 struct btrfs_extent_item *ei;
1327 struct btrfs_extent_inline_ref *iref;
1338 key.objectid = bytenr;
1339 key.type = BTRFS_EXTENT_ITEM_KEY;
1340 key.offset = num_bytes;
1342 want = extent_ref_type(parent, owner);
1344 extra_size = btrfs_extent_inline_ref_size(want);
1345 path->keep_locks = 1;
1348 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1355 leaf = path->nodes[0];
1356 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (item_size < sizeof(*ei)) {
1363 ret = convert_extent_item_v0(trans, root, path, owner,
1369 leaf = path->nodes[0];
1370 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1373 BUG_ON(item_size < sizeof(*ei));
1375 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1376 flags = btrfs_extent_flags(leaf, ei);
1378 ptr = (unsigned long)(ei + 1);
1379 end = (unsigned long)ei + item_size;
1381 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1382 ptr += sizeof(struct btrfs_tree_block_info);
1385 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1394 iref = (struct btrfs_extent_inline_ref *)ptr;
1395 type = btrfs_extent_inline_ref_type(leaf, iref);
1399 ptr += btrfs_extent_inline_ref_size(type);
1403 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 struct btrfs_extent_data_ref *dref;
1405 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1406 if (match_extent_data_ref(leaf, dref, root_objectid,
1411 if (hash_extent_data_ref_item(leaf, dref) <
1412 hash_extent_data_ref(root_objectid, owner, offset))
1416 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1418 if (parent == ref_offset) {
1422 if (ref_offset < parent)
1425 if (root_objectid == ref_offset) {
1429 if (ref_offset < root_objectid)
1433 ptr += btrfs_extent_inline_ref_size(type);
1435 if (err == -ENOENT && insert) {
1436 if (item_size + extra_size >=
1437 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1442 * To add new inline back ref, we have to make sure
1443 * there is no corresponding back ref item.
1444 * For simplicity, we just do not add new inline back
1445 * ref if there is any kind of item for this block
1447 if (find_next_key(path, 0, &key) == 0 &&
1448 key.objectid == bytenr &&
1449 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1454 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1457 path->keep_locks = 0;
1458 btrfs_unlock_up_safe(path, 1);
1464 * helper to add new inline back ref
1466 static noinline_for_stack
1467 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root,
1469 struct btrfs_path *path,
1470 struct btrfs_extent_inline_ref *iref,
1471 u64 parent, u64 root_objectid,
1472 u64 owner, u64 offset, int refs_to_add,
1473 struct btrfs_delayed_extent_op *extent_op)
1475 struct extent_buffer *leaf;
1476 struct btrfs_extent_item *ei;
1479 unsigned long item_offset;
1485 leaf = path->nodes[0];
1486 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1487 item_offset = (unsigned long)iref - (unsigned long)ei;
1489 type = extent_ref_type(parent, owner);
1490 size = btrfs_extent_inline_ref_size(type);
1492 ret = btrfs_extend_item(trans, root, path, size);
1495 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1496 refs = btrfs_extent_refs(leaf, ei);
1497 refs += refs_to_add;
1498 btrfs_set_extent_refs(leaf, ei, refs);
1500 __run_delayed_extent_op(extent_op, leaf, ei);
1502 ptr = (unsigned long)ei + item_offset;
1503 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1504 if (ptr < end - size)
1505 memmove_extent_buffer(leaf, ptr + size, ptr,
1508 iref = (struct btrfs_extent_inline_ref *)ptr;
1509 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1510 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1511 struct btrfs_extent_data_ref *dref;
1512 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1513 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1514 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1515 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1516 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1517 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1518 struct btrfs_shared_data_ref *sref;
1519 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1520 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1521 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1522 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1523 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1525 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1527 btrfs_mark_buffer_dirty(leaf);
1531 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1532 struct btrfs_root *root,
1533 struct btrfs_path *path,
1534 struct btrfs_extent_inline_ref **ref_ret,
1535 u64 bytenr, u64 num_bytes, u64 parent,
1536 u64 root_objectid, u64 owner, u64 offset)
1540 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1541 bytenr, num_bytes, parent,
1542 root_objectid, owner, offset, 0);
1546 btrfs_release_path(root, path);
1549 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1550 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1553 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1554 root_objectid, owner, offset);
1560 * helper to update/remove inline back ref
1562 static noinline_for_stack
1563 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1564 struct btrfs_root *root,
1565 struct btrfs_path *path,
1566 struct btrfs_extent_inline_ref *iref,
1568 struct btrfs_delayed_extent_op *extent_op)
1570 struct extent_buffer *leaf;
1571 struct btrfs_extent_item *ei;
1572 struct btrfs_extent_data_ref *dref = NULL;
1573 struct btrfs_shared_data_ref *sref = NULL;
1582 leaf = path->nodes[0];
1583 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1584 refs = btrfs_extent_refs(leaf, ei);
1585 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1586 refs += refs_to_mod;
1587 btrfs_set_extent_refs(leaf, ei, refs);
1589 __run_delayed_extent_op(extent_op, leaf, ei);
1591 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1594 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1595 refs = btrfs_extent_data_ref_count(leaf, dref);
1596 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1597 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1598 refs = btrfs_shared_data_ref_count(leaf, sref);
1601 BUG_ON(refs_to_mod != -1);
1604 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1605 refs += refs_to_mod;
1608 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1609 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1611 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1613 size = btrfs_extent_inline_ref_size(type);
1614 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1615 ptr = (unsigned long)iref;
1616 end = (unsigned long)ei + item_size;
1617 if (ptr + size < end)
1618 memmove_extent_buffer(leaf, ptr, ptr + size,
1621 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1624 btrfs_mark_buffer_dirty(leaf);
1628 static noinline_for_stack
1629 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1630 struct btrfs_root *root,
1631 struct btrfs_path *path,
1632 u64 bytenr, u64 num_bytes, u64 parent,
1633 u64 root_objectid, u64 owner,
1634 u64 offset, int refs_to_add,
1635 struct btrfs_delayed_extent_op *extent_op)
1637 struct btrfs_extent_inline_ref *iref;
1640 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1641 bytenr, num_bytes, parent,
1642 root_objectid, owner, offset, 1);
1644 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1645 ret = update_inline_extent_backref(trans, root, path, iref,
1646 refs_to_add, extent_op);
1647 } else if (ret == -ENOENT) {
1648 ret = setup_inline_extent_backref(trans, root, path, iref,
1649 parent, root_objectid,
1650 owner, offset, refs_to_add,
1656 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1657 struct btrfs_root *root,
1658 struct btrfs_path *path,
1659 u64 bytenr, u64 parent, u64 root_objectid,
1660 u64 owner, u64 offset, int refs_to_add)
1663 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664 BUG_ON(refs_to_add != 1);
1665 ret = insert_tree_block_ref(trans, root, path, bytenr,
1666 parent, root_objectid);
1668 ret = insert_extent_data_ref(trans, root, path, bytenr,
1669 parent, root_objectid,
1670 owner, offset, refs_to_add);
1675 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1676 struct btrfs_root *root,
1677 struct btrfs_path *path,
1678 struct btrfs_extent_inline_ref *iref,
1679 int refs_to_drop, int is_data)
1683 BUG_ON(!is_data && refs_to_drop != 1);
1685 ret = update_inline_extent_backref(trans, root, path, iref,
1686 -refs_to_drop, NULL);
1687 } else if (is_data) {
1688 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1690 ret = btrfs_del_item(trans, root, path);
1695 static void btrfs_issue_discard(struct block_device *bdev,
1698 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1699 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1702 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1706 u64 map_length = num_bytes;
1707 struct btrfs_multi_bio *multi = NULL;
1709 if (!btrfs_test_opt(root, DISCARD))
1712 /* Tell the block device(s) that the sectors can be discarded */
1713 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1714 bytenr, &map_length, &multi, 0);
1716 struct btrfs_bio_stripe *stripe = multi->stripes;
1719 if (map_length > num_bytes)
1720 map_length = num_bytes;
1722 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1723 btrfs_issue_discard(stripe->dev->bdev,
1733 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1734 struct btrfs_root *root,
1735 u64 bytenr, u64 num_bytes, u64 parent,
1736 u64 root_objectid, u64 owner, u64 offset)
1739 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1740 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1742 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1743 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1744 parent, root_objectid, (int)owner,
1745 BTRFS_ADD_DELAYED_REF, NULL);
1747 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1748 parent, root_objectid, owner, offset,
1749 BTRFS_ADD_DELAYED_REF, NULL);
1754 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 u64 bytenr, u64 num_bytes,
1757 u64 parent, u64 root_objectid,
1758 u64 owner, u64 offset, int refs_to_add,
1759 struct btrfs_delayed_extent_op *extent_op)
1761 struct btrfs_path *path;
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *item;
1768 path = btrfs_alloc_path();
1773 path->leave_spinning = 1;
1774 /* this will setup the path even if it fails to insert the back ref */
1775 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1776 path, bytenr, num_bytes, parent,
1777 root_objectid, owner, offset,
1778 refs_to_add, extent_op);
1782 if (ret != -EAGAIN) {
1787 leaf = path->nodes[0];
1788 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1789 refs = btrfs_extent_refs(leaf, item);
1790 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1792 __run_delayed_extent_op(extent_op, leaf, item);
1794 btrfs_mark_buffer_dirty(leaf);
1795 btrfs_release_path(root->fs_info->extent_root, path);
1798 path->leave_spinning = 1;
1800 /* now insert the actual backref */
1801 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1802 path, bytenr, parent, root_objectid,
1803 owner, offset, refs_to_add);
1806 btrfs_free_path(path);
1810 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1811 struct btrfs_root *root,
1812 struct btrfs_delayed_ref_node *node,
1813 struct btrfs_delayed_extent_op *extent_op,
1814 int insert_reserved)
1817 struct btrfs_delayed_data_ref *ref;
1818 struct btrfs_key ins;
1823 ins.objectid = node->bytenr;
1824 ins.offset = node->num_bytes;
1825 ins.type = BTRFS_EXTENT_ITEM_KEY;
1827 ref = btrfs_delayed_node_to_data_ref(node);
1828 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1829 parent = ref->parent;
1831 ref_root = ref->root;
1833 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1835 BUG_ON(extent_op->update_key);
1836 flags |= extent_op->flags_to_set;
1838 ret = alloc_reserved_file_extent(trans, root,
1839 parent, ref_root, flags,
1840 ref->objectid, ref->offset,
1841 &ins, node->ref_mod);
1842 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844 node->num_bytes, parent,
1845 ref_root, ref->objectid,
1846 ref->offset, node->ref_mod,
1848 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1849 ret = __btrfs_free_extent(trans, root, node->bytenr,
1850 node->num_bytes, parent,
1851 ref_root, ref->objectid,
1852 ref->offset, node->ref_mod,
1860 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1861 struct extent_buffer *leaf,
1862 struct btrfs_extent_item *ei)
1864 u64 flags = btrfs_extent_flags(leaf, ei);
1865 if (extent_op->update_flags) {
1866 flags |= extent_op->flags_to_set;
1867 btrfs_set_extent_flags(leaf, ei, flags);
1870 if (extent_op->update_key) {
1871 struct btrfs_tree_block_info *bi;
1872 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1873 bi = (struct btrfs_tree_block_info *)(ei + 1);
1874 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1878 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1879 struct btrfs_root *root,
1880 struct btrfs_delayed_ref_node *node,
1881 struct btrfs_delayed_extent_op *extent_op)
1883 struct btrfs_key key;
1884 struct btrfs_path *path;
1885 struct btrfs_extent_item *ei;
1886 struct extent_buffer *leaf;
1891 path = btrfs_alloc_path();
1895 key.objectid = node->bytenr;
1896 key.type = BTRFS_EXTENT_ITEM_KEY;
1897 key.offset = node->num_bytes;
1900 path->leave_spinning = 1;
1901 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1912 leaf = path->nodes[0];
1913 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1914 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1915 if (item_size < sizeof(*ei)) {
1916 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1922 leaf = path->nodes[0];
1923 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1926 BUG_ON(item_size < sizeof(*ei));
1927 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928 __run_delayed_extent_op(extent_op, leaf, ei);
1930 btrfs_mark_buffer_dirty(leaf);
1932 btrfs_free_path(path);
1936 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1937 struct btrfs_root *root,
1938 struct btrfs_delayed_ref_node *node,
1939 struct btrfs_delayed_extent_op *extent_op,
1940 int insert_reserved)
1943 struct btrfs_delayed_tree_ref *ref;
1944 struct btrfs_key ins;
1948 ins.objectid = node->bytenr;
1949 ins.offset = node->num_bytes;
1950 ins.type = BTRFS_EXTENT_ITEM_KEY;
1952 ref = btrfs_delayed_node_to_tree_ref(node);
1953 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1954 parent = ref->parent;
1956 ref_root = ref->root;
1958 BUG_ON(node->ref_mod != 1);
1959 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1960 BUG_ON(!extent_op || !extent_op->update_flags ||
1961 !extent_op->update_key);
1962 ret = alloc_reserved_tree_block(trans, root,
1964 extent_op->flags_to_set,
1967 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1968 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1969 node->num_bytes, parent, ref_root,
1970 ref->level, 0, 1, extent_op);
1971 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1972 ret = __btrfs_free_extent(trans, root, node->bytenr,
1973 node->num_bytes, parent, ref_root,
1974 ref->level, 0, 1, extent_op);
1981 /* helper function to actually process a single delayed ref entry */
1982 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1983 struct btrfs_root *root,
1984 struct btrfs_delayed_ref_node *node,
1985 struct btrfs_delayed_extent_op *extent_op,
1986 int insert_reserved)
1989 if (btrfs_delayed_ref_is_head(node)) {
1990 struct btrfs_delayed_ref_head *head;
1992 * we've hit the end of the chain and we were supposed
1993 * to insert this extent into the tree. But, it got
1994 * deleted before we ever needed to insert it, so all
1995 * we have to do is clean up the accounting
1998 head = btrfs_delayed_node_to_head(node);
1999 if (insert_reserved) {
2000 btrfs_pin_extent(root, node->bytenr,
2001 node->num_bytes, 1);
2002 if (head->is_data) {
2003 ret = btrfs_del_csums(trans, root,
2009 mutex_unlock(&head->mutex);
2013 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2014 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2015 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2017 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2018 node->type == BTRFS_SHARED_DATA_REF_KEY)
2019 ret = run_delayed_data_ref(trans, root, node, extent_op,
2026 static noinline struct btrfs_delayed_ref_node *
2027 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2029 struct rb_node *node;
2030 struct btrfs_delayed_ref_node *ref;
2031 int action = BTRFS_ADD_DELAYED_REF;
2034 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2035 * this prevents ref count from going down to zero when
2036 * there still are pending delayed ref.
2038 node = rb_prev(&head->node.rb_node);
2042 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2044 if (ref->bytenr != head->node.bytenr)
2046 if (ref->action == action)
2048 node = rb_prev(node);
2050 if (action == BTRFS_ADD_DELAYED_REF) {
2051 action = BTRFS_DROP_DELAYED_REF;
2057 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2058 struct btrfs_root *root,
2059 struct list_head *cluster)
2061 struct btrfs_delayed_ref_root *delayed_refs;
2062 struct btrfs_delayed_ref_node *ref;
2063 struct btrfs_delayed_ref_head *locked_ref = NULL;
2064 struct btrfs_delayed_extent_op *extent_op;
2067 int must_insert_reserved = 0;
2069 delayed_refs = &trans->transaction->delayed_refs;
2072 /* pick a new head ref from the cluster list */
2073 if (list_empty(cluster))
2076 locked_ref = list_entry(cluster->next,
2077 struct btrfs_delayed_ref_head, cluster);
2079 /* grab the lock that says we are going to process
2080 * all the refs for this head */
2081 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2084 * we may have dropped the spin lock to get the head
2085 * mutex lock, and that might have given someone else
2086 * time to free the head. If that's true, it has been
2087 * removed from our list and we can move on.
2089 if (ret == -EAGAIN) {
2097 * record the must insert reserved flag before we
2098 * drop the spin lock.
2100 must_insert_reserved = locked_ref->must_insert_reserved;
2101 locked_ref->must_insert_reserved = 0;
2103 extent_op = locked_ref->extent_op;
2104 locked_ref->extent_op = NULL;
2107 * locked_ref is the head node, so we have to go one
2108 * node back for any delayed ref updates
2110 ref = select_delayed_ref(locked_ref);
2112 /* All delayed refs have been processed, Go ahead
2113 * and send the head node to run_one_delayed_ref,
2114 * so that any accounting fixes can happen
2116 ref = &locked_ref->node;
2118 if (extent_op && must_insert_reserved) {
2124 spin_unlock(&delayed_refs->lock);
2126 ret = run_delayed_extent_op(trans, root,
2132 spin_lock(&delayed_refs->lock);
2136 list_del_init(&locked_ref->cluster);
2141 rb_erase(&ref->rb_node, &delayed_refs->root);
2142 delayed_refs->num_entries--;
2144 spin_unlock(&delayed_refs->lock);
2146 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2147 must_insert_reserved);
2150 btrfs_put_delayed_ref(ref);
2155 spin_lock(&delayed_refs->lock);
2161 * this starts processing the delayed reference count updates and
2162 * extent insertions we have queued up so far. count can be
2163 * 0, which means to process everything in the tree at the start
2164 * of the run (but not newly added entries), or it can be some target
2165 * number you'd like to process.
2167 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root, unsigned long count)
2170 struct rb_node *node;
2171 struct btrfs_delayed_ref_root *delayed_refs;
2172 struct btrfs_delayed_ref_node *ref;
2173 struct list_head cluster;
2175 int run_all = count == (unsigned long)-1;
2178 if (root == root->fs_info->extent_root)
2179 root = root->fs_info->tree_root;
2181 delayed_refs = &trans->transaction->delayed_refs;
2182 INIT_LIST_HEAD(&cluster);
2184 spin_lock(&delayed_refs->lock);
2186 count = delayed_refs->num_entries * 2;
2190 if (!(run_all || run_most) &&
2191 delayed_refs->num_heads_ready < 64)
2195 * go find something we can process in the rbtree. We start at
2196 * the beginning of the tree, and then build a cluster
2197 * of refs to process starting at the first one we are able to
2200 ret = btrfs_find_ref_cluster(trans, &cluster,
2201 delayed_refs->run_delayed_start);
2205 ret = run_clustered_refs(trans, root, &cluster);
2208 count -= min_t(unsigned long, ret, count);
2215 node = rb_first(&delayed_refs->root);
2218 count = (unsigned long)-1;
2221 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2223 if (btrfs_delayed_ref_is_head(ref)) {
2224 struct btrfs_delayed_ref_head *head;
2226 head = btrfs_delayed_node_to_head(ref);
2227 atomic_inc(&ref->refs);
2229 spin_unlock(&delayed_refs->lock);
2230 mutex_lock(&head->mutex);
2231 mutex_unlock(&head->mutex);
2233 btrfs_put_delayed_ref(ref);
2237 node = rb_next(node);
2239 spin_unlock(&delayed_refs->lock);
2240 schedule_timeout(1);
2244 spin_unlock(&delayed_refs->lock);
2248 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2249 struct btrfs_root *root,
2250 u64 bytenr, u64 num_bytes, u64 flags,
2253 struct btrfs_delayed_extent_op *extent_op;
2256 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2260 extent_op->flags_to_set = flags;
2261 extent_op->update_flags = 1;
2262 extent_op->update_key = 0;
2263 extent_op->is_data = is_data ? 1 : 0;
2265 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2271 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2272 struct btrfs_root *root,
2273 struct btrfs_path *path,
2274 u64 objectid, u64 offset, u64 bytenr)
2276 struct btrfs_delayed_ref_head *head;
2277 struct btrfs_delayed_ref_node *ref;
2278 struct btrfs_delayed_data_ref *data_ref;
2279 struct btrfs_delayed_ref_root *delayed_refs;
2280 struct rb_node *node;
2284 delayed_refs = &trans->transaction->delayed_refs;
2285 spin_lock(&delayed_refs->lock);
2286 head = btrfs_find_delayed_ref_head(trans, bytenr);
2290 if (!mutex_trylock(&head->mutex)) {
2291 atomic_inc(&head->node.refs);
2292 spin_unlock(&delayed_refs->lock);
2294 btrfs_release_path(root->fs_info->extent_root, path);
2296 mutex_lock(&head->mutex);
2297 mutex_unlock(&head->mutex);
2298 btrfs_put_delayed_ref(&head->node);
2302 node = rb_prev(&head->node.rb_node);
2306 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2308 if (ref->bytenr != bytenr)
2312 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2315 data_ref = btrfs_delayed_node_to_data_ref(ref);
2317 node = rb_prev(node);
2319 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2320 if (ref->bytenr == bytenr)
2324 if (data_ref->root != root->root_key.objectid ||
2325 data_ref->objectid != objectid || data_ref->offset != offset)
2330 mutex_unlock(&head->mutex);
2332 spin_unlock(&delayed_refs->lock);
2336 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2337 struct btrfs_root *root,
2338 struct btrfs_path *path,
2339 u64 objectid, u64 offset, u64 bytenr)
2341 struct btrfs_root *extent_root = root->fs_info->extent_root;
2342 struct extent_buffer *leaf;
2343 struct btrfs_extent_data_ref *ref;
2344 struct btrfs_extent_inline_ref *iref;
2345 struct btrfs_extent_item *ei;
2346 struct btrfs_key key;
2350 key.objectid = bytenr;
2351 key.offset = (u64)-1;
2352 key.type = BTRFS_EXTENT_ITEM_KEY;
2354 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2360 if (path->slots[0] == 0)
2364 leaf = path->nodes[0];
2365 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2367 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2371 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2372 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2373 if (item_size < sizeof(*ei)) {
2374 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2378 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2380 if (item_size != sizeof(*ei) +
2381 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2384 if (btrfs_extent_generation(leaf, ei) <=
2385 btrfs_root_last_snapshot(&root->root_item))
2388 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2389 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2390 BTRFS_EXTENT_DATA_REF_KEY)
2393 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2394 if (btrfs_extent_refs(leaf, ei) !=
2395 btrfs_extent_data_ref_count(leaf, ref) ||
2396 btrfs_extent_data_ref_root(leaf, ref) !=
2397 root->root_key.objectid ||
2398 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2399 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2407 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2408 struct btrfs_root *root,
2409 u64 objectid, u64 offset, u64 bytenr)
2411 struct btrfs_path *path;
2415 path = btrfs_alloc_path();
2420 ret = check_committed_ref(trans, root, path, objectid,
2422 if (ret && ret != -ENOENT)
2425 ret2 = check_delayed_ref(trans, root, path, objectid,
2427 } while (ret2 == -EAGAIN);
2429 if (ret2 && ret2 != -ENOENT) {
2434 if (ret != -ENOENT || ret2 != -ENOENT)
2437 btrfs_free_path(path);
2438 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2445 struct extent_buffer *buf, u32 nr_extents)
2447 struct btrfs_key key;
2448 struct btrfs_file_extent_item *fi;
2456 if (!root->ref_cows)
2459 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2461 root_gen = root->root_key.offset;
2464 root_gen = trans->transid - 1;
2467 level = btrfs_header_level(buf);
2468 nritems = btrfs_header_nritems(buf);
2471 struct btrfs_leaf_ref *ref;
2472 struct btrfs_extent_info *info;
2474 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2480 ref->root_gen = root_gen;
2481 ref->bytenr = buf->start;
2482 ref->owner = btrfs_header_owner(buf);
2483 ref->generation = btrfs_header_generation(buf);
2484 ref->nritems = nr_extents;
2485 info = ref->extents;
2487 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2489 btrfs_item_key_to_cpu(buf, &key, i);
2490 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2492 fi = btrfs_item_ptr(buf, i,
2493 struct btrfs_file_extent_item);
2494 if (btrfs_file_extent_type(buf, fi) ==
2495 BTRFS_FILE_EXTENT_INLINE)
2497 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2498 if (disk_bytenr == 0)
2501 info->bytenr = disk_bytenr;
2503 btrfs_file_extent_disk_num_bytes(buf, fi);
2504 info->objectid = key.objectid;
2505 info->offset = key.offset;
2509 ret = btrfs_add_leaf_ref(root, ref, shared);
2510 if (ret == -EEXIST && shared) {
2511 struct btrfs_leaf_ref *old;
2512 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2514 btrfs_remove_leaf_ref(root, old);
2515 btrfs_free_leaf_ref(root, old);
2516 ret = btrfs_add_leaf_ref(root, ref, shared);
2519 btrfs_free_leaf_ref(root, ref);
2525 /* when a block goes through cow, we update the reference counts of
2526 * everything that block points to. The internal pointers of the block
2527 * can be in just about any order, and it is likely to have clusters of
2528 * things that are close together and clusters of things that are not.
2530 * To help reduce the seeks that come with updating all of these reference
2531 * counts, sort them by byte number before actual updates are done.
2533 * struct refsort is used to match byte number to slot in the btree block.
2534 * we sort based on the byte number and then use the slot to actually
2537 * struct refsort is smaller than strcut btrfs_item and smaller than
2538 * struct btrfs_key_ptr. Since we're currently limited to the page size
2539 * for a btree block, there's no way for a kmalloc of refsorts for a
2540 * single node to be bigger than a page.
2548 * for passing into sort()
2550 static int refsort_cmp(const void *a_void, const void *b_void)
2552 const struct refsort *a = a_void;
2553 const struct refsort *b = b_void;
2555 if (a->bytenr < b->bytenr)
2557 if (a->bytenr > b->bytenr)
2563 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2564 struct btrfs_root *root,
2565 struct extent_buffer *buf,
2566 int full_backref, int inc)
2573 struct btrfs_key key;
2574 struct btrfs_file_extent_item *fi;
2578 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2579 u64, u64, u64, u64, u64, u64);
2581 ref_root = btrfs_header_owner(buf);
2582 nritems = btrfs_header_nritems(buf);
2583 level = btrfs_header_level(buf);
2585 if (!root->ref_cows && level == 0)
2589 process_func = btrfs_inc_extent_ref;
2591 process_func = btrfs_free_extent;
2594 parent = buf->start;
2598 for (i = 0; i < nritems; i++) {
2600 btrfs_item_key_to_cpu(buf, &key, i);
2601 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2603 fi = btrfs_item_ptr(buf, i,
2604 struct btrfs_file_extent_item);
2605 if (btrfs_file_extent_type(buf, fi) ==
2606 BTRFS_FILE_EXTENT_INLINE)
2608 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2612 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2613 key.offset -= btrfs_file_extent_offset(buf, fi);
2614 ret = process_func(trans, root, bytenr, num_bytes,
2615 parent, ref_root, key.objectid,
2620 bytenr = btrfs_node_blockptr(buf, i);
2621 num_bytes = btrfs_level_size(root, level - 1);
2622 ret = process_func(trans, root, bytenr, num_bytes,
2623 parent, ref_root, level - 1, 0);
2634 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2635 struct extent_buffer *buf, int full_backref)
2637 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2640 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2641 struct extent_buffer *buf, int full_backref)
2643 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2646 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2647 struct btrfs_root *root,
2648 struct btrfs_path *path,
2649 struct btrfs_block_group_cache *cache)
2652 struct btrfs_root *extent_root = root->fs_info->extent_root;
2654 struct extent_buffer *leaf;
2656 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2661 leaf = path->nodes[0];
2662 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2663 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2664 btrfs_mark_buffer_dirty(leaf);
2665 btrfs_release_path(extent_root, path);
2673 static struct btrfs_block_group_cache *
2674 next_block_group(struct btrfs_root *root,
2675 struct btrfs_block_group_cache *cache)
2677 struct rb_node *node;
2678 spin_lock(&root->fs_info->block_group_cache_lock);
2679 node = rb_next(&cache->cache_node);
2680 btrfs_put_block_group(cache);
2682 cache = rb_entry(node, struct btrfs_block_group_cache,
2684 btrfs_get_block_group(cache);
2687 spin_unlock(&root->fs_info->block_group_cache_lock);
2691 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2692 struct btrfs_root *root)
2694 struct btrfs_block_group_cache *cache;
2696 struct btrfs_path *path;
2699 path = btrfs_alloc_path();
2705 err = btrfs_run_delayed_refs(trans, root,
2710 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2714 cache = next_block_group(root, cache);
2724 last = cache->key.objectid + cache->key.offset;
2726 err = write_one_cache_group(trans, root, path, cache);
2728 btrfs_put_block_group(cache);
2731 btrfs_free_path(path);
2735 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2737 struct btrfs_block_group_cache *block_group;
2740 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2741 if (!block_group || block_group->ro)
2744 btrfs_put_block_group(block_group);
2748 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2749 u64 total_bytes, u64 bytes_used,
2750 struct btrfs_space_info **space_info)
2752 struct btrfs_space_info *found;
2756 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2757 BTRFS_BLOCK_GROUP_RAID10))
2762 found = __find_space_info(info, flags);
2764 spin_lock(&found->lock);
2765 found->total_bytes += total_bytes;
2766 found->bytes_used += bytes_used;
2767 found->disk_used += bytes_used * factor;
2769 spin_unlock(&found->lock);
2770 *space_info = found;
2773 found = kzalloc(sizeof(*found), GFP_NOFS);
2777 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2778 INIT_LIST_HEAD(&found->block_groups[i]);
2779 init_rwsem(&found->groups_sem);
2780 spin_lock_init(&found->lock);
2781 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2782 BTRFS_BLOCK_GROUP_SYSTEM |
2783 BTRFS_BLOCK_GROUP_METADATA);
2784 found->total_bytes = total_bytes;
2785 found->bytes_used = bytes_used;
2786 found->disk_used = bytes_used * factor;
2787 found->bytes_pinned = 0;
2788 found->bytes_reserved = 0;
2789 found->bytes_readonly = 0;
2790 found->bytes_may_use = 0;
2792 found->force_alloc = 0;
2793 *space_info = found;
2794 list_add_rcu(&found->list, &info->space_info);
2795 atomic_set(&found->caching_threads, 0);
2799 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2801 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2802 BTRFS_BLOCK_GROUP_RAID1 |
2803 BTRFS_BLOCK_GROUP_RAID10 |
2804 BTRFS_BLOCK_GROUP_DUP);
2806 if (flags & BTRFS_BLOCK_GROUP_DATA)
2807 fs_info->avail_data_alloc_bits |= extra_flags;
2808 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2809 fs_info->avail_metadata_alloc_bits |= extra_flags;
2810 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2811 fs_info->avail_system_alloc_bits |= extra_flags;
2815 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2817 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2819 if (num_devices == 1)
2820 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2821 if (num_devices < 4)
2822 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2824 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2825 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2826 BTRFS_BLOCK_GROUP_RAID10))) {
2827 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2830 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2831 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2832 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2835 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2836 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2837 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2838 (flags & BTRFS_BLOCK_GROUP_DUP)))
2839 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2843 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2845 if (flags & BTRFS_BLOCK_GROUP_DATA)
2846 flags |= root->fs_info->avail_data_alloc_bits &
2847 root->fs_info->data_alloc_profile;
2848 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2849 flags |= root->fs_info->avail_system_alloc_bits &
2850 root->fs_info->system_alloc_profile;
2851 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2852 flags |= root->fs_info->avail_metadata_alloc_bits &
2853 root->fs_info->metadata_alloc_profile;
2854 return btrfs_reduce_alloc_profile(root, flags);
2857 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2862 flags = BTRFS_BLOCK_GROUP_DATA;
2863 else if (root == root->fs_info->chunk_root)
2864 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2866 flags = BTRFS_BLOCK_GROUP_METADATA;
2868 return get_alloc_profile(root, flags);
2871 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2873 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2874 BTRFS_BLOCK_GROUP_DATA);
2878 * This will check the space that the inode allocates from to make sure we have
2879 * enough space for bytes.
2881 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
2883 struct btrfs_space_info *data_sinfo;
2884 struct btrfs_root *root = BTRFS_I(inode)->root;
2886 int ret = 0, committed = 0;
2888 /* make sure bytes are sectorsize aligned */
2889 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2891 data_sinfo = BTRFS_I(inode)->space_info;
2896 /* make sure we have enough space to handle the data first */
2897 spin_lock(&data_sinfo->lock);
2898 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
2899 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
2900 data_sinfo->bytes_may_use;
2902 if (used + bytes > data_sinfo->total_bytes) {
2903 struct btrfs_trans_handle *trans;
2906 * if we don't have enough free bytes in this space then we need
2907 * to alloc a new chunk.
2909 if (!data_sinfo->full) {
2912 data_sinfo->force_alloc = 1;
2913 spin_unlock(&data_sinfo->lock);
2915 alloc_target = btrfs_get_alloc_profile(root, 1);
2916 trans = btrfs_join_transaction(root, 1);
2918 return PTR_ERR(trans);
2920 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2921 bytes + 2 * 1024 * 1024,
2923 btrfs_end_transaction(trans, root);
2928 btrfs_set_inode_space_info(root, inode);
2929 data_sinfo = BTRFS_I(inode)->space_info;
2933 spin_unlock(&data_sinfo->lock);
2935 /* commit the current transaction and try again */
2936 if (!committed && !root->fs_info->open_ioctl_trans) {
2938 trans = btrfs_join_transaction(root, 1);
2940 return PTR_ERR(trans);
2941 ret = btrfs_commit_transaction(trans, root);
2947 #if 0 /* I hope we never need this code again, just in case */
2948 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
2949 "%llu bytes_reserved, " "%llu bytes_pinned, "
2950 "%llu bytes_readonly, %llu may use %llu total\n",
2951 (unsigned long long)bytes,
2952 (unsigned long long)data_sinfo->bytes_used,
2953 (unsigned long long)data_sinfo->bytes_reserved,
2954 (unsigned long long)data_sinfo->bytes_pinned,
2955 (unsigned long long)data_sinfo->bytes_readonly,
2956 (unsigned long long)data_sinfo->bytes_may_use,
2957 (unsigned long long)data_sinfo->total_bytes);
2961 data_sinfo->bytes_may_use += bytes;
2962 BTRFS_I(inode)->reserved_bytes += bytes;
2963 spin_unlock(&data_sinfo->lock);
2969 * called when we are clearing an delalloc extent from the
2970 * inode's io_tree or there was an error for whatever reason
2971 * after calling btrfs_check_data_free_space
2973 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
2975 struct btrfs_root *root = BTRFS_I(inode)->root;
2976 struct btrfs_space_info *data_sinfo;
2978 /* make sure bytes are sectorsize aligned */
2979 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2981 data_sinfo = BTRFS_I(inode)->space_info;
2982 spin_lock(&data_sinfo->lock);
2983 data_sinfo->bytes_may_use -= bytes;
2984 BTRFS_I(inode)->reserved_bytes -= bytes;
2985 spin_unlock(&data_sinfo->lock);
2988 static void force_metadata_allocation(struct btrfs_fs_info *info)
2990 struct list_head *head = &info->space_info;
2991 struct btrfs_space_info *found;
2994 list_for_each_entry_rcu(found, head, list) {
2995 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2996 found->force_alloc = 1;
3001 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3004 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3006 if (sinfo->bytes_used + sinfo->bytes_reserved +
3007 alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3010 if (sinfo->bytes_used + sinfo->bytes_reserved +
3011 alloc_bytes < div_factor(num_bytes, 8))
3017 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3018 struct btrfs_root *extent_root, u64 alloc_bytes,
3019 u64 flags, int force)
3021 struct btrfs_space_info *space_info;
3022 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3025 mutex_lock(&fs_info->chunk_mutex);
3027 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3029 space_info = __find_space_info(extent_root->fs_info, flags);
3031 ret = update_space_info(extent_root->fs_info, flags,
3035 BUG_ON(!space_info);
3037 spin_lock(&space_info->lock);
3038 if (space_info->force_alloc)
3040 if (space_info->full) {
3041 spin_unlock(&space_info->lock);
3045 if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3046 spin_unlock(&space_info->lock);
3049 spin_unlock(&space_info->lock);
3052 * if we're doing a data chunk, go ahead and make sure that
3053 * we keep a reasonable number of metadata chunks allocated in the
3056 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3057 fs_info->data_chunk_allocations++;
3058 if (!(fs_info->data_chunk_allocations %
3059 fs_info->metadata_ratio))
3060 force_metadata_allocation(fs_info);
3063 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3064 spin_lock(&space_info->lock);
3066 space_info->full = 1;
3069 space_info->force_alloc = 0;
3070 spin_unlock(&space_info->lock);
3072 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3076 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3077 struct btrfs_root *root,
3078 struct btrfs_space_info *sinfo, u64 num_bytes)
3086 spin_lock(&sinfo->lock);
3087 ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3088 spin_unlock(&sinfo->lock);
3093 trans = btrfs_join_transaction(root, 1);
3094 BUG_ON(IS_ERR(trans));
3098 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3099 num_bytes + 2 * 1024 * 1024,
3100 get_alloc_profile(root, sinfo->flags), 0);
3103 btrfs_end_transaction(trans, root);
3105 return ret == 1 ? 1 : 0;
3109 * shrink metadata reservation for delalloc
3111 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3112 struct btrfs_root *root, u64 to_reclaim)
3114 struct btrfs_block_rsv *block_rsv;
3121 block_rsv = &root->fs_info->delalloc_block_rsv;
3122 spin_lock(&block_rsv->lock);
3123 reserved = block_rsv->reserved;
3124 spin_unlock(&block_rsv->lock);
3129 max_reclaim = min(reserved, to_reclaim);
3132 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3134 __set_current_state(TASK_INTERRUPTIBLE);
3135 schedule_timeout(pause);
3137 if (pause > HZ / 10)
3143 spin_lock(&block_rsv->lock);
3144 if (reserved > block_rsv->reserved)
3145 reclaimed = reserved - block_rsv->reserved;
3146 reserved = block_rsv->reserved;
3147 spin_unlock(&block_rsv->lock);
3149 if (reserved == 0 || reclaimed >= max_reclaim)
3152 if (trans && trans->transaction->blocked)
3155 return reclaimed >= to_reclaim;
3158 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3159 struct btrfs_root *root,
3160 struct btrfs_block_rsv *block_rsv,
3161 u64 num_bytes, int *retries)
3163 struct btrfs_space_info *space_info = block_rsv->space_info;
3169 ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3173 if (trans && trans->transaction->in_commit)
3176 ret = shrink_delalloc(trans, root, num_bytes);
3180 spin_lock(&space_info->lock);
3181 if (space_info->bytes_pinned < num_bytes)
3183 spin_unlock(&space_info->lock);
3192 trans = btrfs_join_transaction(root, 1);
3193 BUG_ON(IS_ERR(trans));
3194 ret = btrfs_commit_transaction(trans, root);
3200 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3203 struct btrfs_space_info *space_info = block_rsv->space_info;
3207 spin_lock(&space_info->lock);
3208 unused = space_info->bytes_used + space_info->bytes_reserved +
3209 space_info->bytes_pinned + space_info->bytes_readonly;
3211 if (unused < space_info->total_bytes)
3212 unused = space_info->total_bytes - unused;
3216 if (unused >= num_bytes) {
3217 if (block_rsv->priority >= 10) {
3218 space_info->bytes_reserved += num_bytes;
3221 if ((unused + block_rsv->reserved) *
3222 block_rsv->priority >=
3223 (num_bytes + block_rsv->reserved) * 10) {
3224 space_info->bytes_reserved += num_bytes;
3229 spin_unlock(&space_info->lock);
3234 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3235 struct btrfs_root *root)
3237 struct btrfs_block_rsv *block_rsv;
3239 block_rsv = trans->block_rsv;
3241 block_rsv = root->block_rsv;
3244 block_rsv = &root->fs_info->empty_block_rsv;
3249 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3253 spin_lock(&block_rsv->lock);
3254 if (block_rsv->reserved >= num_bytes) {
3255 block_rsv->reserved -= num_bytes;
3256 if (block_rsv->reserved < block_rsv->size)
3257 block_rsv->full = 0;
3260 spin_unlock(&block_rsv->lock);
3264 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3265 u64 num_bytes, int update_size)
3267 spin_lock(&block_rsv->lock);
3268 block_rsv->reserved += num_bytes;
3270 block_rsv->size += num_bytes;
3271 else if (block_rsv->reserved >= block_rsv->size)
3272 block_rsv->full = 1;
3273 spin_unlock(&block_rsv->lock);
3276 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3277 struct btrfs_block_rsv *dest, u64 num_bytes)
3279 struct btrfs_space_info *space_info = block_rsv->space_info;
3281 spin_lock(&block_rsv->lock);
3282 if (num_bytes == (u64)-1)
3283 num_bytes = block_rsv->size;
3284 block_rsv->size -= num_bytes;
3285 if (block_rsv->reserved >= block_rsv->size) {
3286 num_bytes = block_rsv->reserved - block_rsv->size;
3287 block_rsv->reserved = block_rsv->size;
3288 block_rsv->full = 1;
3292 spin_unlock(&block_rsv->lock);
3294 if (num_bytes > 0) {
3296 block_rsv_add_bytes(dest, num_bytes, 0);
3298 spin_lock(&space_info->lock);
3299 space_info->bytes_reserved -= num_bytes;
3300 spin_unlock(&space_info->lock);
3305 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3306 struct btrfs_block_rsv *dst, u64 num_bytes)
3310 ret = block_rsv_use_bytes(src, num_bytes);
3314 block_rsv_add_bytes(dst, num_bytes, 1);
3318 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3320 memset(rsv, 0, sizeof(*rsv));
3321 spin_lock_init(&rsv->lock);
3322 atomic_set(&rsv->usage, 1);
3324 INIT_LIST_HEAD(&rsv->list);
3327 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3329 struct btrfs_block_rsv *block_rsv;
3330 struct btrfs_fs_info *fs_info = root->fs_info;
3333 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3337 btrfs_init_block_rsv(block_rsv);
3339 alloc_target = btrfs_get_alloc_profile(root, 0);
3340 block_rsv->space_info = __find_space_info(fs_info,
3341 BTRFS_BLOCK_GROUP_METADATA);
3346 void btrfs_free_block_rsv(struct btrfs_root *root,
3347 struct btrfs_block_rsv *rsv)
3349 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3350 btrfs_block_rsv_release(root, rsv, (u64)-1);
3357 * make the block_rsv struct be able to capture freed space.
3358 * the captured space will re-add to the the block_rsv struct
3359 * after transaction commit
3361 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3362 struct btrfs_block_rsv *block_rsv)
3364 block_rsv->durable = 1;
3365 mutex_lock(&fs_info->durable_block_rsv_mutex);
3366 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3367 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3370 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root,
3372 struct btrfs_block_rsv *block_rsv,
3373 u64 num_bytes, int *retries)
3380 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3382 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3386 ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3393 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3394 struct btrfs_root *root,
3395 struct btrfs_block_rsv *block_rsv,
3396 u64 min_reserved, int min_factor)
3399 int commit_trans = 0;
3405 spin_lock(&block_rsv->lock);
3407 num_bytes = div_factor(block_rsv->size, min_factor);
3408 if (min_reserved > num_bytes)
3409 num_bytes = min_reserved;
3411 if (block_rsv->reserved >= num_bytes) {
3414 num_bytes -= block_rsv->reserved;
3415 if (block_rsv->durable &&
3416 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3419 spin_unlock(&block_rsv->lock);
3423 if (block_rsv->refill_used) {
3424 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3426 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3435 trans = btrfs_join_transaction(root, 1);
3436 BUG_ON(IS_ERR(trans));
3437 ret = btrfs_commit_transaction(trans, root);
3442 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3443 block_rsv->size, block_rsv->reserved,
3444 block_rsv->freed[0], block_rsv->freed[1]);
3449 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3450 struct btrfs_block_rsv *dst_rsv,
3453 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3456 void btrfs_block_rsv_release(struct btrfs_root *root,
3457 struct btrfs_block_rsv *block_rsv,
3460 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3461 if (global_rsv->full || global_rsv == block_rsv ||
3462 block_rsv->space_info != global_rsv->space_info)
3464 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3468 * helper to calculate size of global block reservation.
3469 * the desired value is sum of space used by extent tree,
3470 * checksum tree and root tree
3472 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3474 struct btrfs_space_info *sinfo;
3478 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3481 * per tree used space accounting can be inaccuracy, so we
3484 spin_lock(&fs_info->extent_root->accounting_lock);
3485 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3486 spin_unlock(&fs_info->extent_root->accounting_lock);
3488 spin_lock(&fs_info->csum_root->accounting_lock);
3489 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3490 spin_unlock(&fs_info->csum_root->accounting_lock);
3492 spin_lock(&fs_info->tree_root->accounting_lock);
3493 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3494 spin_unlock(&fs_info->tree_root->accounting_lock);
3496 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3497 spin_lock(&sinfo->lock);
3498 data_used = sinfo->bytes_used;
3499 spin_unlock(&sinfo->lock);
3501 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3502 spin_lock(&sinfo->lock);
3503 meta_used = sinfo->bytes_used;
3504 spin_unlock(&sinfo->lock);
3506 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3508 num_bytes += div64_u64(data_used + meta_used, 50);
3510 if (num_bytes * 3 > meta_used)
3511 num_bytes = div64_u64(meta_used, 3);
3513 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3516 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3518 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3519 struct btrfs_space_info *sinfo = block_rsv->space_info;
3522 num_bytes = calc_global_metadata_size(fs_info);
3524 spin_lock(&block_rsv->lock);
3525 spin_lock(&sinfo->lock);
3527 block_rsv->size = num_bytes;
3529 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3530 sinfo->bytes_reserved + sinfo->bytes_readonly;
3532 if (sinfo->total_bytes > num_bytes) {
3533 num_bytes = sinfo->total_bytes - num_bytes;
3534 block_rsv->reserved += num_bytes;
3535 sinfo->bytes_reserved += num_bytes;
3538 if (block_rsv->reserved >= block_rsv->size) {
3539 num_bytes = block_rsv->reserved - block_rsv->size;
3540 sinfo->bytes_reserved -= num_bytes;
3541 block_rsv->reserved = block_rsv->size;
3542 block_rsv->full = 1;
3545 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3546 block_rsv->size, block_rsv->reserved);
3548 spin_unlock(&sinfo->lock);
3549 spin_unlock(&block_rsv->lock);
3552 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3554 struct btrfs_space_info *space_info;
3556 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3557 fs_info->chunk_block_rsv.space_info = space_info;
3558 fs_info->chunk_block_rsv.priority = 10;
3560 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3561 fs_info->global_block_rsv.space_info = space_info;
3562 fs_info->global_block_rsv.priority = 10;
3563 fs_info->global_block_rsv.refill_used = 1;
3564 fs_info->delalloc_block_rsv.space_info = space_info;
3565 fs_info->trans_block_rsv.space_info = space_info;
3566 fs_info->empty_block_rsv.space_info = space_info;
3567 fs_info->empty_block_rsv.priority = 10;
3569 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3570 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3571 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3572 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3573 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3575 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3577 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3579 update_global_block_rsv(fs_info);
3582 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3584 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3585 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3586 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3587 WARN_ON(fs_info->trans_block_rsv.size > 0);
3588 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3589 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3590 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3593 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3595 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3599 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3600 struct btrfs_root *root,
3601 int num_items, int *retries)
3606 if (num_items == 0 || root->fs_info->chunk_root == root)
3609 num_bytes = calc_trans_metadata_size(root, num_items);
3610 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3611 num_bytes, retries);
3613 trans->bytes_reserved += num_bytes;
3614 trans->block_rsv = &root->fs_info->trans_block_rsv;
3619 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3620 struct btrfs_root *root)
3622 if (!trans->bytes_reserved)
3625 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3626 btrfs_block_rsv_release(root, trans->block_rsv,
3627 trans->bytes_reserved);
3628 trans->bytes_reserved = 0;
3631 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3632 struct inode *inode)
3634 struct btrfs_root *root = BTRFS_I(inode)->root;
3635 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3636 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3639 * one for deleting orphan item, one for updating inode and
3640 * two for calling btrfs_truncate_inode_items.
3642 * btrfs_truncate_inode_items is a delete operation, it frees
3643 * more space than it uses in most cases. So two units of
3644 * metadata space should be enough for calling it many times.
3645 * If all of the metadata space is used, we can commit
3646 * transaction and use space it freed.
3648 u64 num_bytes = calc_trans_metadata_size(root, 4);
3649 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3652 void btrfs_orphan_release_metadata(struct inode *inode)
3654 struct btrfs_root *root = BTRFS_I(inode)->root;
3655 u64 num_bytes = calc_trans_metadata_size(root, 4);
3656 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3659 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3660 struct btrfs_pending_snapshot *pending)
3662 struct btrfs_root *root = pending->root;
3663 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3664 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3666 * two for root back/forward refs, two for directory entries
3667 * and one for root of the snapshot.
3669 u64 num_bytes = calc_trans_metadata_size(root, 5);
3670 dst_rsv->space_info = src_rsv->space_info;
3671 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3674 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3676 return num_bytes >>= 3;
3679 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3681 struct btrfs_root *root = BTRFS_I(inode)->root;
3682 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3688 if (btrfs_transaction_in_commit(root->fs_info))
3689 schedule_timeout(1);
3691 num_bytes = ALIGN(num_bytes, root->sectorsize);
3693 spin_lock(&BTRFS_I(inode)->accounting_lock);
3694 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3695 if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3696 nr_extents -= BTRFS_I(inode)->reserved_extents;
3697 to_reserve = calc_trans_metadata_size(root, nr_extents);
3703 to_reserve += calc_csum_metadata_size(inode, num_bytes);
3704 ret = reserve_metadata_bytes(block_rsv, to_reserve);
3706 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3707 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3714 BTRFS_I(inode)->reserved_extents += nr_extents;
3715 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3716 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3718 block_rsv_add_bytes(block_rsv, to_reserve, 1);
3720 if (block_rsv->size > 512 * 1024 * 1024)
3721 shrink_delalloc(NULL, root, to_reserve);
3726 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3728 struct btrfs_root *root = BTRFS_I(inode)->root;
3732 num_bytes = ALIGN(num_bytes, root->sectorsize);
3733 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3735 spin_lock(&BTRFS_I(inode)->accounting_lock);
3736 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3737 if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3738 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3739 BTRFS_I(inode)->reserved_extents -= nr_extents;
3743 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3745 to_free = calc_csum_metadata_size(inode, num_bytes);
3747 to_free += calc_trans_metadata_size(root, nr_extents);
3749 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3753 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3757 ret = btrfs_check_data_free_space(inode, num_bytes);
3761 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3763 btrfs_free_reserved_data_space(inode, num_bytes);
3770 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3772 btrfs_delalloc_release_metadata(inode, num_bytes);
3773 btrfs_free_reserved_data_space(inode, num_bytes);
3776 static int update_block_group(struct btrfs_trans_handle *trans,
3777 struct btrfs_root *root,
3778 u64 bytenr, u64 num_bytes, int alloc)
3780 struct btrfs_block_group_cache *cache;
3781 struct btrfs_fs_info *info = root->fs_info;
3783 u64 total = num_bytes;
3787 /* block accounting for super block */
3788 spin_lock(&info->delalloc_lock);
3789 old_val = btrfs_super_bytes_used(&info->super_copy);
3791 old_val += num_bytes;
3793 old_val -= num_bytes;
3794 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3795 spin_unlock(&info->delalloc_lock);
3798 cache = btrfs_lookup_block_group(info, bytenr);
3801 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3802 BTRFS_BLOCK_GROUP_RAID1 |
3803 BTRFS_BLOCK_GROUP_RAID10))
3807 byte_in_group = bytenr - cache->key.objectid;
3808 WARN_ON(byte_in_group > cache->key.offset);
3810 spin_lock(&cache->space_info->lock);
3811 spin_lock(&cache->lock);
3813 old_val = btrfs_block_group_used(&cache->item);
3814 num_bytes = min(total, cache->key.offset - byte_in_group);
3816 old_val += num_bytes;
3817 btrfs_set_block_group_used(&cache->item, old_val);
3818 cache->reserved -= num_bytes;
3819 cache->space_info->bytes_reserved -= num_bytes;
3820 cache->space_info->bytes_used += num_bytes;
3821 cache->space_info->disk_used += num_bytes * factor;
3822 spin_unlock(&cache->lock);
3823 spin_unlock(&cache->space_info->lock);
3825 old_val -= num_bytes;
3826 btrfs_set_block_group_used(&cache->item, old_val);
3827 cache->pinned += num_bytes;
3828 cache->space_info->bytes_pinned += num_bytes;
3829 cache->space_info->bytes_used -= num_bytes;
3830 cache->space_info->disk_used -= num_bytes * factor;
3831 spin_unlock(&cache->lock);
3832 spin_unlock(&cache->space_info->lock);
3834 set_extent_dirty(info->pinned_extents,
3835 bytenr, bytenr + num_bytes - 1,
3836 GFP_NOFS | __GFP_NOFAIL);
3838 btrfs_put_block_group(cache);
3840 bytenr += num_bytes;
3845 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3847 struct btrfs_block_group_cache *cache;
3850 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3854 bytenr = cache->key.objectid;
3855 btrfs_put_block_group(cache);
3860 static int pin_down_extent(struct btrfs_root *root,
3861 struct btrfs_block_group_cache *cache,
3862 u64 bytenr, u64 num_bytes, int reserved)
3864 spin_lock(&cache->space_info->lock);
3865 spin_lock(&cache->lock);
3866 cache->pinned += num_bytes;
3867 cache->space_info->bytes_pinned += num_bytes;
3869 cache->reserved -= num_bytes;
3870 cache->space_info->bytes_reserved -= num_bytes;
3872 spin_unlock(&cache->lock);
3873 spin_unlock(&cache->space_info->lock);
3875 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
3876 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
3881 * this function must be called within transaction
3883 int btrfs_pin_extent(struct btrfs_root *root,
3884 u64 bytenr, u64 num_bytes, int reserved)
3886 struct btrfs_block_group_cache *cache;
3888 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
3891 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
3893 btrfs_put_block_group(cache);
3898 * update size of reserved extents. this function may return -EAGAIN
3899 * if 'reserve' is true or 'sinfo' is false.
3901 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
3902 u64 num_bytes, int reserve, int sinfo)
3906 struct btrfs_space_info *space_info = cache->space_info;
3907 spin_lock(&space_info->lock);
3908 spin_lock(&cache->lock);
3913 cache->reserved += num_bytes;
3914 space_info->bytes_reserved += num_bytes;
3918 space_info->bytes_readonly += num_bytes;
3919 cache->reserved -= num_bytes;
3920 space_info->bytes_reserved -= num_bytes;
3922 spin_unlock(&cache->lock);
3923 spin_unlock(&space_info->lock);
3925 spin_lock(&cache->lock);
3930 cache->reserved += num_bytes;
3932 cache->reserved -= num_bytes;
3934 spin_unlock(&cache->lock);
3939 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3940 struct btrfs_root *root)
3942 struct btrfs_fs_info *fs_info = root->fs_info;
3943 struct btrfs_caching_control *next;
3944 struct btrfs_caching_control *caching_ctl;
3945 struct btrfs_block_group_cache *cache;
3947 down_write(&fs_info->extent_commit_sem);
3949 list_for_each_entry_safe(caching_ctl, next,
3950 &fs_info->caching_block_groups, list) {
3951 cache = caching_ctl->block_group;
3952 if (block_group_cache_done(cache)) {
3953 cache->last_byte_to_unpin = (u64)-1;
3954 list_del_init(&caching_ctl->list);
3955 put_caching_control(caching_ctl);
3957 cache->last_byte_to_unpin = caching_ctl->progress;
3961 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3962 fs_info->pinned_extents = &fs_info->freed_extents[1];
3964 fs_info->pinned_extents = &fs_info->freed_extents[0];
3966 up_write(&fs_info->extent_commit_sem);
3968 update_global_block_rsv(fs_info);
3972 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3974 struct btrfs_fs_info *fs_info = root->fs_info;
3975 struct btrfs_block_group_cache *cache = NULL;
3978 while (start <= end) {
3980 start >= cache->key.objectid + cache->key.offset) {
3982 btrfs_put_block_group(cache);
3983 cache = btrfs_lookup_block_group(fs_info, start);
3987 len = cache->key.objectid + cache->key.offset - start;
3988 len = min(len, end + 1 - start);
3990 if (start < cache->last_byte_to_unpin) {
3991 len = min(len, cache->last_byte_to_unpin - start);
3992 btrfs_add_free_space(cache, start, len);
3997 spin_lock(&cache->space_info->lock);
3998 spin_lock(&cache->lock);
3999 cache->pinned -= len;
4000 cache->space_info->bytes_pinned -= len;
4002 cache->space_info->bytes_readonly += len;
4003 } else if (cache->reserved_pinned > 0) {
4004 len = min(len, cache->reserved_pinned);
4005 cache->reserved_pinned -= len;
4006 cache->space_info->bytes_reserved += len;
4008 spin_unlock(&cache->lock);
4009 spin_unlock(&cache->space_info->lock);
4013 btrfs_put_block_group(cache);
4017 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4018 struct btrfs_root *root)
4020 struct btrfs_fs_info *fs_info = root->fs_info;
4021 struct extent_io_tree *unpin;
4022 struct btrfs_block_rsv *block_rsv;
4023 struct btrfs_block_rsv *next_rsv;
4029 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4030 unpin = &fs_info->freed_extents[1];
4032 unpin = &fs_info->freed_extents[0];
4035 ret = find_first_extent_bit(unpin, 0, &start, &end,
4040 ret = btrfs_discard_extent(root, start, end + 1 - start);
4042 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4043 unpin_extent_range(root, start, end);
4047 mutex_lock(&fs_info->durable_block_rsv_mutex);
4048 list_for_each_entry_safe(block_rsv, next_rsv,
4049 &fs_info->durable_block_rsv_list, list) {
4051 idx = trans->transid & 0x1;
4052 if (block_rsv->freed[idx] > 0) {
4053 block_rsv_add_bytes(block_rsv,
4054 block_rsv->freed[idx], 0);
4055 block_rsv->freed[idx] = 0;
4057 if (atomic_read(&block_rsv->usage) == 0) {
4058 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4060 if (block_rsv->freed[0] == 0 &&
4061 block_rsv->freed[1] == 0) {
4062 list_del_init(&block_rsv->list);
4066 btrfs_block_rsv_release(root, block_rsv, 0);
4069 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4074 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4075 struct btrfs_root *root,
4076 u64 bytenr, u64 num_bytes, u64 parent,
4077 u64 root_objectid, u64 owner_objectid,
4078 u64 owner_offset, int refs_to_drop,
4079 struct btrfs_delayed_extent_op *extent_op)
4081 struct btrfs_key key;
4082 struct btrfs_path *path;
4083 struct btrfs_fs_info *info = root->fs_info;
4084 struct btrfs_root *extent_root = info->extent_root;
4085 struct extent_buffer *leaf;
4086 struct btrfs_extent_item *ei;
4087 struct btrfs_extent_inline_ref *iref;
4090 int extent_slot = 0;
4091 int found_extent = 0;
4096 path = btrfs_alloc_path();
4101 path->leave_spinning = 1;
4103 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4104 BUG_ON(!is_data && refs_to_drop != 1);
4106 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4107 bytenr, num_bytes, parent,
4108 root_objectid, owner_objectid,
4111 extent_slot = path->slots[0];
4112 while (extent_slot >= 0) {
4113 btrfs_item_key_to_cpu(path->nodes[0], &key,
4115 if (key.objectid != bytenr)
4117 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4118 key.offset == num_bytes) {
4122 if (path->slots[0] - extent_slot > 5)
4126 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4127 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4128 if (found_extent && item_size < sizeof(*ei))
4131 if (!found_extent) {
4133 ret = remove_extent_backref(trans, extent_root, path,
4137 btrfs_release_path(extent_root, path);
4138 path->leave_spinning = 1;
4140 key.objectid = bytenr;
4141 key.type = BTRFS_EXTENT_ITEM_KEY;
4142 key.offset = num_bytes;
4144 ret = btrfs_search_slot(trans, extent_root,
4147 printk(KERN_ERR "umm, got %d back from search"
4148 ", was looking for %llu\n", ret,
4149 (unsigned long long)bytenr);
4150 btrfs_print_leaf(extent_root, path->nodes[0]);
4153 extent_slot = path->slots[0];
4156 btrfs_print_leaf(extent_root, path->nodes[0]);
4158 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4159 "parent %llu root %llu owner %llu offset %llu\n",
4160 (unsigned long long)bytenr,
4161 (unsigned long long)parent,
4162 (unsigned long long)root_objectid,
4163 (unsigned long long)owner_objectid,
4164 (unsigned long long)owner_offset);
4167 leaf = path->nodes[0];
4168 item_size = btrfs_item_size_nr(leaf, extent_slot);
4169 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4170 if (item_size < sizeof(*ei)) {
4171 BUG_ON(found_extent || extent_slot != path->slots[0]);
4172 ret = convert_extent_item_v0(trans, extent_root, path,
4176 btrfs_release_path(extent_root, path);
4177 path->leave_spinning = 1;
4179 key.objectid = bytenr;
4180 key.type = BTRFS_EXTENT_ITEM_KEY;
4181 key.offset = num_bytes;
4183 ret = btrfs_search_slot(trans, extent_root, &key, path,
4186 printk(KERN_ERR "umm, got %d back from search"
4187 ", was looking for %llu\n", ret,
4188 (unsigned long long)bytenr);
4189 btrfs_print_leaf(extent_root, path->nodes[0]);
4192 extent_slot = path->slots[0];
4193 leaf = path->nodes[0];
4194 item_size = btrfs_item_size_nr(leaf, extent_slot);
4197 BUG_ON(item_size < sizeof(*ei));
4198 ei = btrfs_item_ptr(leaf, extent_slot,
4199 struct btrfs_extent_item);
4200 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4201 struct btrfs_tree_block_info *bi;
4202 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4203 bi = (struct btrfs_tree_block_info *)(ei + 1);
4204 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4207 refs = btrfs_extent_refs(leaf, ei);
4208 BUG_ON(refs < refs_to_drop);
4209 refs -= refs_to_drop;
4213 __run_delayed_extent_op(extent_op, leaf, ei);
4215 * In the case of inline back ref, reference count will
4216 * be updated by remove_extent_backref
4219 BUG_ON(!found_extent);
4221 btrfs_set_extent_refs(leaf, ei, refs);
4222 btrfs_mark_buffer_dirty(leaf);
4225 ret = remove_extent_backref(trans, extent_root, path,
4232 BUG_ON(is_data && refs_to_drop !=
4233 extent_data_ref_count(root, path, iref));
4235 BUG_ON(path->slots[0] != extent_slot);
4237 BUG_ON(path->slots[0] != extent_slot + 1);
4238 path->slots[0] = extent_slot;
4243 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4246 btrfs_release_path(extent_root, path);
4249 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4252 invalidate_mapping_pages(info->btree_inode->i_mapping,
4253 bytenr >> PAGE_CACHE_SHIFT,
4254 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4257 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4260 btrfs_free_path(path);
4265 * when we free an block, it is possible (and likely) that we free the last
4266 * delayed ref for that extent as well. This searches the delayed ref tree for
4267 * a given extent, and if there are no other delayed refs to be processed, it
4268 * removes it from the tree.
4270 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4271 struct btrfs_root *root, u64 bytenr)
4273 struct btrfs_delayed_ref_head *head;
4274 struct btrfs_delayed_ref_root *delayed_refs;
4275 struct btrfs_delayed_ref_node *ref;
4276 struct rb_node *node;
4279 delayed_refs = &trans->transaction->delayed_refs;
4280 spin_lock(&delayed_refs->lock);
4281 head = btrfs_find_delayed_ref_head(trans, bytenr);
4285 node = rb_prev(&head->node.rb_node);
4289 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4291 /* there are still entries for this ref, we can't drop it */
4292 if (ref->bytenr == bytenr)
4295 if (head->extent_op) {
4296 if (!head->must_insert_reserved)
4298 kfree(head->extent_op);
4299 head->extent_op = NULL;
4303 * waiting for the lock here would deadlock. If someone else has it
4304 * locked they are already in the process of dropping it anyway
4306 if (!mutex_trylock(&head->mutex))
4310 * at this point we have a head with no other entries. Go
4311 * ahead and process it.
4313 head->node.in_tree = 0;
4314 rb_erase(&head->node.rb_node, &delayed_refs->root);
4316 delayed_refs->num_entries--;
4319 * we don't take a ref on the node because we're removing it from the
4320 * tree, so we just steal the ref the tree was holding.
4322 delayed_refs->num_heads--;
4323 if (list_empty(&head->cluster))
4324 delayed_refs->num_heads_ready--;
4326 list_del_init(&head->cluster);
4327 spin_unlock(&delayed_refs->lock);
4329 BUG_ON(head->extent_op);
4330 if (head->must_insert_reserved)
4333 mutex_unlock(&head->mutex);
4334 btrfs_put_delayed_ref(&head->node);
4337 spin_unlock(&delayed_refs->lock);
4341 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4342 struct btrfs_root *root,
4343 struct extent_buffer *buf,
4344 u64 parent, int last_ref)
4346 struct btrfs_block_rsv *block_rsv;
4347 struct btrfs_block_group_cache *cache = NULL;
4350 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4351 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4352 parent, root->root_key.objectid,
4353 btrfs_header_level(buf),
4354 BTRFS_DROP_DELAYED_REF, NULL);
4361 block_rsv = get_block_rsv(trans, root);
4362 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4363 if (block_rsv->space_info != cache->space_info)
4366 if (btrfs_header_generation(buf) == trans->transid) {
4367 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4368 ret = check_ref_cleanup(trans, root, buf->start);
4373 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4374 pin_down_extent(root, cache, buf->start, buf->len, 1);
4378 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4380 btrfs_add_free_space(cache, buf->start, buf->len);
4381 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4382 if (ret == -EAGAIN) {
4383 /* block group became read-only */
4384 update_reserved_bytes(cache, buf->len, 0, 1);
4389 spin_lock(&block_rsv->lock);
4390 if (block_rsv->reserved < block_rsv->size) {
4391 block_rsv->reserved += buf->len;
4394 spin_unlock(&block_rsv->lock);
4397 spin_lock(&cache->space_info->lock);
4398 cache->space_info->bytes_reserved -= buf->len;
4399 spin_unlock(&cache->space_info->lock);
4404 if (block_rsv->durable && !cache->ro) {
4406 spin_lock(&cache->lock);
4408 cache->reserved_pinned += buf->len;
4411 spin_unlock(&cache->lock);
4414 spin_lock(&block_rsv->lock);
4415 block_rsv->freed[trans->transid & 0x1] += buf->len;
4416 spin_unlock(&block_rsv->lock);
4420 btrfs_put_block_group(cache);
4423 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root,
4425 u64 bytenr, u64 num_bytes, u64 parent,
4426 u64 root_objectid, u64 owner, u64 offset)
4431 * tree log blocks never actually go into the extent allocation
4432 * tree, just update pinning info and exit early.
4434 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4435 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4436 /* unlocks the pinned mutex */
4437 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4439 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4440 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4441 parent, root_objectid, (int)owner,
4442 BTRFS_DROP_DELAYED_REF, NULL);
4445 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4446 parent, root_objectid, owner,
4447 offset, BTRFS_DROP_DELAYED_REF, NULL);
4453 static u64 stripe_align(struct btrfs_root *root, u64 val)
4455 u64 mask = ((u64)root->stripesize - 1);
4456 u64 ret = (val + mask) & ~mask;
4461 * when we wait for progress in the block group caching, its because
4462 * our allocation attempt failed at least once. So, we must sleep
4463 * and let some progress happen before we try again.
4465 * This function will sleep at least once waiting for new free space to
4466 * show up, and then it will check the block group free space numbers
4467 * for our min num_bytes. Another option is to have it go ahead
4468 * and look in the rbtree for a free extent of a given size, but this
4472 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4475 struct btrfs_caching_control *caching_ctl;
4478 caching_ctl = get_caching_control(cache);
4482 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4483 (cache->free_space >= num_bytes));
4485 put_caching_control(caching_ctl);
4490 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4492 struct btrfs_caching_control *caching_ctl;
4495 caching_ctl = get_caching_control(cache);
4499 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4501 put_caching_control(caching_ctl);
4505 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4508 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4510 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4512 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4514 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4521 enum btrfs_loop_type {
4522 LOOP_FIND_IDEAL = 0,
4523 LOOP_CACHING_NOWAIT = 1,
4524 LOOP_CACHING_WAIT = 2,
4525 LOOP_ALLOC_CHUNK = 3,
4526 LOOP_NO_EMPTY_SIZE = 4,
4530 * walks the btree of allocated extents and find a hole of a given size.
4531 * The key ins is changed to record the hole:
4532 * ins->objectid == block start
4533 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4534 * ins->offset == number of blocks
4535 * Any available blocks before search_start are skipped.
4537 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4538 struct btrfs_root *orig_root,
4539 u64 num_bytes, u64 empty_size,
4540 u64 search_start, u64 search_end,
4541 u64 hint_byte, struct btrfs_key *ins,
4545 struct btrfs_root *root = orig_root->fs_info->extent_root;
4546 struct btrfs_free_cluster *last_ptr = NULL;
4547 struct btrfs_block_group_cache *block_group = NULL;
4548 int empty_cluster = 2 * 1024 * 1024;
4549 int allowed_chunk_alloc = 0;
4550 int done_chunk_alloc = 0;
4551 struct btrfs_space_info *space_info;
4552 int last_ptr_loop = 0;
4555 bool found_uncached_bg = false;
4556 bool failed_cluster_refill = false;
4557 bool failed_alloc = false;
4558 u64 ideal_cache_percent = 0;
4559 u64 ideal_cache_offset = 0;
4561 WARN_ON(num_bytes < root->sectorsize);
4562 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4566 space_info = __find_space_info(root->fs_info, data);
4568 printk(KERN_ERR "No space info for %d\n", data);
4572 if (orig_root->ref_cows || empty_size)
4573 allowed_chunk_alloc = 1;
4575 if (data & BTRFS_BLOCK_GROUP_METADATA) {
4576 last_ptr = &root->fs_info->meta_alloc_cluster;
4577 if (!btrfs_test_opt(root, SSD))
4578 empty_cluster = 64 * 1024;
4581 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4582 last_ptr = &root->fs_info->data_alloc_cluster;
4586 spin_lock(&last_ptr->lock);
4587 if (last_ptr->block_group)
4588 hint_byte = last_ptr->window_start;
4589 spin_unlock(&last_ptr->lock);
4592 search_start = max(search_start, first_logical_byte(root, 0));
4593 search_start = max(search_start, hint_byte);
4598 if (search_start == hint_byte) {
4600 block_group = btrfs_lookup_block_group(root->fs_info,
4603 * we don't want to use the block group if it doesn't match our
4604 * allocation bits, or if its not cached.
4606 * However if we are re-searching with an ideal block group
4607 * picked out then we don't care that the block group is cached.
4609 if (block_group && block_group_bits(block_group, data) &&
4610 (block_group->cached != BTRFS_CACHE_NO ||
4611 search_start == ideal_cache_offset)) {
4612 down_read(&space_info->groups_sem);
4613 if (list_empty(&block_group->list) ||
4616 * someone is removing this block group,
4617 * we can't jump into the have_block_group
4618 * target because our list pointers are not
4621 btrfs_put_block_group(block_group);
4622 up_read(&space_info->groups_sem);
4624 index = get_block_group_index(block_group);
4625 goto have_block_group;
4627 } else if (block_group) {
4628 btrfs_put_block_group(block_group);
4632 down_read(&space_info->groups_sem);
4633 list_for_each_entry(block_group, &space_info->block_groups[index],
4638 btrfs_get_block_group(block_group);
4639 search_start = block_group->key.objectid;
4642 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4645 free_percent = btrfs_block_group_used(&block_group->item);
4646 free_percent *= 100;
4647 free_percent = div64_u64(free_percent,
4648 block_group->key.offset);
4649 free_percent = 100 - free_percent;
4650 if (free_percent > ideal_cache_percent &&
4651 likely(!block_group->ro)) {
4652 ideal_cache_offset = block_group->key.objectid;
4653 ideal_cache_percent = free_percent;
4657 * We only want to start kthread caching if we are at
4658 * the point where we will wait for caching to make
4659 * progress, or if our ideal search is over and we've
4660 * found somebody to start caching.
4662 if (loop > LOOP_CACHING_NOWAIT ||
4663 (loop > LOOP_FIND_IDEAL &&
4664 atomic_read(&space_info->caching_threads) < 2)) {
4665 ret = cache_block_group(block_group);
4668 found_uncached_bg = true;
4671 * If loop is set for cached only, try the next block
4674 if (loop == LOOP_FIND_IDEAL)
4678 cached = block_group_cache_done(block_group);
4679 if (unlikely(!cached))
4680 found_uncached_bg = true;
4682 if (unlikely(block_group->ro))
4686 * Ok we want to try and use the cluster allocator, so lets look
4687 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4688 * have tried the cluster allocator plenty of times at this
4689 * point and not have found anything, so we are likely way too
4690 * fragmented for the clustering stuff to find anything, so lets
4691 * just skip it and let the allocator find whatever block it can
4694 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4696 * the refill lock keeps out other
4697 * people trying to start a new cluster
4699 spin_lock(&last_ptr->refill_lock);
4700 if (last_ptr->block_group &&
4701 (last_ptr->block_group->ro ||
4702 !block_group_bits(last_ptr->block_group, data))) {
4704 goto refill_cluster;
4707 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4708 num_bytes, search_start);
4710 /* we have a block, we're done */
4711 spin_unlock(&last_ptr->refill_lock);
4715 spin_lock(&last_ptr->lock);
4717 * whoops, this cluster doesn't actually point to
4718 * this block group. Get a ref on the block
4719 * group is does point to and try again
4721 if (!last_ptr_loop && last_ptr->block_group &&
4722 last_ptr->block_group != block_group) {
4724 btrfs_put_block_group(block_group);
4725 block_group = last_ptr->block_group;
4726 btrfs_get_block_group(block_group);
4727 spin_unlock(&last_ptr->lock);
4728 spin_unlock(&last_ptr->refill_lock);
4731 search_start = block_group->key.objectid;
4733 * we know this block group is properly
4734 * in the list because
4735 * btrfs_remove_block_group, drops the
4736 * cluster before it removes the block
4737 * group from the list
4739 goto have_block_group;
4741 spin_unlock(&last_ptr->lock);
4744 * this cluster didn't work out, free it and
4747 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4751 /* allocate a cluster in this block group */
4752 ret = btrfs_find_space_cluster(trans, root,
4753 block_group, last_ptr,
4755 empty_cluster + empty_size);
4758 * now pull our allocation out of this
4761 offset = btrfs_alloc_from_cluster(block_group,
4762 last_ptr, num_bytes,
4765 /* we found one, proceed */
4766 spin_unlock(&last_ptr->refill_lock);
4769 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4770 && !failed_cluster_refill) {
4771 spin_unlock(&last_ptr->refill_lock);
4773 failed_cluster_refill = true;
4774 wait_block_group_cache_progress(block_group,
4775 num_bytes + empty_cluster + empty_size);
4776 goto have_block_group;
4780 * at this point we either didn't find a cluster
4781 * or we weren't able to allocate a block from our
4782 * cluster. Free the cluster we've been trying
4783 * to use, and go to the next block group
4785 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4786 spin_unlock(&last_ptr->refill_lock);
4790 offset = btrfs_find_space_for_alloc(block_group, search_start,
4791 num_bytes, empty_size);
4793 * If we didn't find a chunk, and we haven't failed on this
4794 * block group before, and this block group is in the middle of
4795 * caching and we are ok with waiting, then go ahead and wait
4796 * for progress to be made, and set failed_alloc to true.
4798 * If failed_alloc is true then we've already waited on this
4799 * block group once and should move on to the next block group.
4801 if (!offset && !failed_alloc && !cached &&
4802 loop > LOOP_CACHING_NOWAIT) {
4803 wait_block_group_cache_progress(block_group,
4804 num_bytes + empty_size);
4805 failed_alloc = true;
4806 goto have_block_group;
4807 } else if (!offset) {
4811 search_start = stripe_align(root, offset);
4812 /* move on to the next group */
4813 if (search_start + num_bytes >= search_end) {
4814 btrfs_add_free_space(block_group, offset, num_bytes);
4818 /* move on to the next group */
4819 if (search_start + num_bytes >
4820 block_group->key.objectid + block_group->key.offset) {
4821 btrfs_add_free_space(block_group, offset, num_bytes);
4825 ins->objectid = search_start;
4826 ins->offset = num_bytes;
4828 if (offset < search_start)
4829 btrfs_add_free_space(block_group, offset,
4830 search_start - offset);
4831 BUG_ON(offset > search_start);
4833 ret = update_reserved_bytes(block_group, num_bytes, 1,
4834 (data & BTRFS_BLOCK_GROUP_DATA));
4835 if (ret == -EAGAIN) {
4836 btrfs_add_free_space(block_group, offset, num_bytes);
4840 /* we are all good, lets return */
4841 ins->objectid = search_start;
4842 ins->offset = num_bytes;
4844 if (offset < search_start)
4845 btrfs_add_free_space(block_group, offset,
4846 search_start - offset);
4847 BUG_ON(offset > search_start);
4850 failed_cluster_refill = false;
4851 failed_alloc = false;
4852 BUG_ON(index != get_block_group_index(block_group));
4853 btrfs_put_block_group(block_group);
4855 up_read(&space_info->groups_sem);
4857 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4860 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4861 * for them to make caching progress. Also
4862 * determine the best possible bg to cache
4863 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4864 * caching kthreads as we move along
4865 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4866 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4867 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4870 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4871 (found_uncached_bg || empty_size || empty_cluster ||
4872 allowed_chunk_alloc)) {
4874 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4875 found_uncached_bg = false;
4877 if (!ideal_cache_percent &&
4878 atomic_read(&space_info->caching_threads))
4882 * 1 of the following 2 things have happened so far
4884 * 1) We found an ideal block group for caching that
4885 * is mostly full and will cache quickly, so we might
4886 * as well wait for it.
4888 * 2) We searched for cached only and we didn't find
4889 * anything, and we didn't start any caching kthreads
4890 * either, so chances are we will loop through and
4891 * start a couple caching kthreads, and then come back
4892 * around and just wait for them. This will be slower
4893 * because we will have 2 caching kthreads reading at
4894 * the same time when we could have just started one
4895 * and waited for it to get far enough to give us an
4896 * allocation, so go ahead and go to the wait caching
4899 loop = LOOP_CACHING_WAIT;
4900 search_start = ideal_cache_offset;
4901 ideal_cache_percent = 0;
4903 } else if (loop == LOOP_FIND_IDEAL) {
4905 * Didn't find a uncached bg, wait on anything we find
4908 loop = LOOP_CACHING_WAIT;
4912 if (loop < LOOP_CACHING_WAIT) {
4917 if (loop == LOOP_ALLOC_CHUNK) {
4922 if (allowed_chunk_alloc) {
4923 ret = do_chunk_alloc(trans, root, num_bytes +
4924 2 * 1024 * 1024, data, 1);
4925 allowed_chunk_alloc = 0;
4926 done_chunk_alloc = 1;
4927 } else if (!done_chunk_alloc) {
4928 space_info->force_alloc = 1;
4931 if (loop < LOOP_NO_EMPTY_SIZE) {
4936 } else if (!ins->objectid) {
4940 /* we found what we needed */
4941 if (ins->objectid) {
4942 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4943 trans->block_group = block_group->key.objectid;
4945 btrfs_put_block_group(block_group);
4952 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4953 int dump_block_groups)
4955 struct btrfs_block_group_cache *cache;
4958 spin_lock(&info->lock);
4959 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4960 (unsigned long long)(info->total_bytes - info->bytes_used -
4961 info->bytes_pinned - info->bytes_reserved -
4962 info->bytes_readonly),
4963 (info->full) ? "" : "not ");
4964 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
4965 "reserved=%llu, may_use=%llu, readonly=%llu\n",
4966 (unsigned long long)info->total_bytes,
4967 (unsigned long long)info->bytes_used,
4968 (unsigned long long)info->bytes_pinned,
4969 (unsigned long long)info->bytes_reserved,
4970 (unsigned long long)info->bytes_may_use,
4971 (unsigned long long)info->bytes_readonly);
4972 spin_unlock(&info->lock);
4974 if (!dump_block_groups)
4977 down_read(&info->groups_sem);
4979 list_for_each_entry(cache, &info->block_groups[index], list) {
4980 spin_lock(&cache->lock);
4981 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4982 "%llu pinned %llu reserved\n",
4983 (unsigned long long)cache->key.objectid,
4984 (unsigned long long)cache->key.offset,
4985 (unsigned long long)btrfs_block_group_used(&cache->item),
4986 (unsigned long long)cache->pinned,
4987 (unsigned long long)cache->reserved);
4988 btrfs_dump_free_space(cache, bytes);
4989 spin_unlock(&cache->lock);
4991 if (++index < BTRFS_NR_RAID_TYPES)
4993 up_read(&info->groups_sem);
4996 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4997 struct btrfs_root *root,
4998 u64 num_bytes, u64 min_alloc_size,
4999 u64 empty_size, u64 hint_byte,
5000 u64 search_end, struct btrfs_key *ins,
5004 u64 search_start = 0;
5006 data = btrfs_get_alloc_profile(root, data);
5009 * the only place that sets empty_size is btrfs_realloc_node, which
5010 * is not called recursively on allocations
5012 if (empty_size || root->ref_cows)
5013 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5014 num_bytes + 2 * 1024 * 1024, data, 0);
5016 WARN_ON(num_bytes < root->sectorsize);
5017 ret = find_free_extent(trans, root, num_bytes, empty_size,
5018 search_start, search_end, hint_byte,
5021 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5022 num_bytes = num_bytes >> 1;
5023 num_bytes = num_bytes & ~(root->sectorsize - 1);
5024 num_bytes = max(num_bytes, min_alloc_size);
5025 do_chunk_alloc(trans, root->fs_info->extent_root,
5026 num_bytes, data, 1);
5029 if (ret == -ENOSPC) {
5030 struct btrfs_space_info *sinfo;
5032 sinfo = __find_space_info(root->fs_info, data);
5033 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5034 "wanted %llu\n", (unsigned long long)data,
5035 (unsigned long long)num_bytes);
5036 dump_space_info(sinfo, num_bytes, 1);
5042 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5044 struct btrfs_block_group_cache *cache;
5047 cache = btrfs_lookup_block_group(root->fs_info, start);
5049 printk(KERN_ERR "Unable to find block group for %llu\n",
5050 (unsigned long long)start);
5054 ret = btrfs_discard_extent(root, start, len);
5056 btrfs_add_free_space(cache, start, len);
5057 update_reserved_bytes(cache, len, 0, 1);
5058 btrfs_put_block_group(cache);
5063 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5064 struct btrfs_root *root,
5065 u64 parent, u64 root_objectid,
5066 u64 flags, u64 owner, u64 offset,
5067 struct btrfs_key *ins, int ref_mod)
5070 struct btrfs_fs_info *fs_info = root->fs_info;
5071 struct btrfs_extent_item *extent_item;
5072 struct btrfs_extent_inline_ref *iref;
5073 struct btrfs_path *path;
5074 struct extent_buffer *leaf;
5079 type = BTRFS_SHARED_DATA_REF_KEY;
5081 type = BTRFS_EXTENT_DATA_REF_KEY;
5083 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5085 path = btrfs_alloc_path();
5088 path->leave_spinning = 1;
5089 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5093 leaf = path->nodes[0];
5094 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5095 struct btrfs_extent_item);
5096 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5097 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5098 btrfs_set_extent_flags(leaf, extent_item,
5099 flags | BTRFS_EXTENT_FLAG_DATA);
5101 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5102 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5104 struct btrfs_shared_data_ref *ref;
5105 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5106 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5107 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5109 struct btrfs_extent_data_ref *ref;
5110 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5111 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5112 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5113 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5114 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5117 btrfs_mark_buffer_dirty(path->nodes[0]);
5118 btrfs_free_path(path);
5120 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5122 printk(KERN_ERR "btrfs update block group failed for %llu "
5123 "%llu\n", (unsigned long long)ins->objectid,
5124 (unsigned long long)ins->offset);
5130 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5131 struct btrfs_root *root,
5132 u64 parent, u64 root_objectid,
5133 u64 flags, struct btrfs_disk_key *key,
5134 int level, struct btrfs_key *ins)
5137 struct btrfs_fs_info *fs_info = root->fs_info;
5138 struct btrfs_extent_item *extent_item;
5139 struct btrfs_tree_block_info *block_info;
5140 struct btrfs_extent_inline_ref *iref;
5141 struct btrfs_path *path;
5142 struct extent_buffer *leaf;
5143 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5145 path = btrfs_alloc_path();
5148 path->leave_spinning = 1;
5149 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5153 leaf = path->nodes[0];
5154 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5155 struct btrfs_extent_item);
5156 btrfs_set_extent_refs(leaf, extent_item, 1);
5157 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5158 btrfs_set_extent_flags(leaf, extent_item,
5159 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5160 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5162 btrfs_set_tree_block_key(leaf, block_info, key);
5163 btrfs_set_tree_block_level(leaf, block_info, level);
5165 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5167 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5168 btrfs_set_extent_inline_ref_type(leaf, iref,
5169 BTRFS_SHARED_BLOCK_REF_KEY);
5170 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5172 btrfs_set_extent_inline_ref_type(leaf, iref,
5173 BTRFS_TREE_BLOCK_REF_KEY);
5174 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5177 btrfs_mark_buffer_dirty(leaf);
5178 btrfs_free_path(path);
5180 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5182 printk(KERN_ERR "btrfs update block group failed for %llu "
5183 "%llu\n", (unsigned long long)ins->objectid,
5184 (unsigned long long)ins->offset);
5190 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5191 struct btrfs_root *root,
5192 u64 root_objectid, u64 owner,
5193 u64 offset, struct btrfs_key *ins)
5197 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5199 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5200 0, root_objectid, owner, offset,
5201 BTRFS_ADD_DELAYED_EXTENT, NULL);
5206 * this is used by the tree logging recovery code. It records that
5207 * an extent has been allocated and makes sure to clear the free
5208 * space cache bits as well
5210 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5211 struct btrfs_root *root,
5212 u64 root_objectid, u64 owner, u64 offset,
5213 struct btrfs_key *ins)
5216 struct btrfs_block_group_cache *block_group;
5217 struct btrfs_caching_control *caching_ctl;
5218 u64 start = ins->objectid;
5219 u64 num_bytes = ins->offset;
5221 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5222 cache_block_group(block_group);
5223 caching_ctl = get_caching_control(block_group);
5226 BUG_ON(!block_group_cache_done(block_group));
5227 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5230 mutex_lock(&caching_ctl->mutex);
5232 if (start >= caching_ctl->progress) {
5233 ret = add_excluded_extent(root, start, num_bytes);
5235 } else if (start + num_bytes <= caching_ctl->progress) {
5236 ret = btrfs_remove_free_space(block_group,
5240 num_bytes = caching_ctl->progress - start;
5241 ret = btrfs_remove_free_space(block_group,
5245 start = caching_ctl->progress;
5246 num_bytes = ins->objectid + ins->offset -
5247 caching_ctl->progress;
5248 ret = add_excluded_extent(root, start, num_bytes);
5252 mutex_unlock(&caching_ctl->mutex);
5253 put_caching_control(caching_ctl);
5256 ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5258 btrfs_put_block_group(block_group);
5259 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5260 0, owner, offset, ins, 1);
5264 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5265 struct btrfs_root *root,
5266 u64 bytenr, u32 blocksize,
5269 struct extent_buffer *buf;
5271 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5273 return ERR_PTR(-ENOMEM);
5274 btrfs_set_header_generation(buf, trans->transid);
5275 btrfs_set_buffer_lockdep_class(buf, level);
5276 btrfs_tree_lock(buf);
5277 clean_tree_block(trans, root, buf);
5279 btrfs_set_lock_blocking(buf);
5280 btrfs_set_buffer_uptodate(buf);
5282 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5284 * we allow two log transactions at a time, use different
5285 * EXENT bit to differentiate dirty pages.
5287 if (root->log_transid % 2 == 0)
5288 set_extent_dirty(&root->dirty_log_pages, buf->start,
5289 buf->start + buf->len - 1, GFP_NOFS);
5291 set_extent_new(&root->dirty_log_pages, buf->start,
5292 buf->start + buf->len - 1, GFP_NOFS);
5294 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5295 buf->start + buf->len - 1, GFP_NOFS);
5297 trans->blocks_used++;
5298 /* this returns a buffer locked for blocking */
5302 static struct btrfs_block_rsv *
5303 use_block_rsv(struct btrfs_trans_handle *trans,
5304 struct btrfs_root *root, u32 blocksize)
5306 struct btrfs_block_rsv *block_rsv;
5309 block_rsv = get_block_rsv(trans, root);
5311 if (block_rsv->size == 0) {
5312 ret = reserve_metadata_bytes(block_rsv, blocksize);
5314 return ERR_PTR(ret);
5318 ret = block_rsv_use_bytes(block_rsv, blocksize);
5323 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5324 block_rsv->size, block_rsv->reserved,
5325 block_rsv->freed[0], block_rsv->freed[1]);
5327 return ERR_PTR(-ENOSPC);
5330 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5332 block_rsv_add_bytes(block_rsv, blocksize, 0);
5333 block_rsv_release_bytes(block_rsv, NULL, 0);
5337 * finds a free extent and does all the dirty work required for allocation
5338 * returns the key for the extent through ins, and a tree buffer for
5339 * the first block of the extent through buf.
5341 * returns the tree buffer or NULL.
5343 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5344 struct btrfs_root *root, u32 blocksize,
5345 u64 parent, u64 root_objectid,
5346 struct btrfs_disk_key *key, int level,
5347 u64 hint, u64 empty_size)
5349 struct btrfs_key ins;
5350 struct btrfs_block_rsv *block_rsv;
5351 struct extent_buffer *buf;
5356 block_rsv = use_block_rsv(trans, root, blocksize);
5357 if (IS_ERR(block_rsv))
5358 return ERR_CAST(block_rsv);
5360 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5361 empty_size, hint, (u64)-1, &ins, 0);
5363 unuse_block_rsv(block_rsv, blocksize);
5364 return ERR_PTR(ret);
5367 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5369 BUG_ON(IS_ERR(buf));
5371 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5373 parent = ins.objectid;
5374 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5378 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5379 struct btrfs_delayed_extent_op *extent_op;
5380 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5383 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5385 memset(&extent_op->key, 0, sizeof(extent_op->key));
5386 extent_op->flags_to_set = flags;
5387 extent_op->update_key = 1;
5388 extent_op->update_flags = 1;
5389 extent_op->is_data = 0;
5391 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5392 ins.offset, parent, root_objectid,
5393 level, BTRFS_ADD_DELAYED_EXTENT,
5400 struct walk_control {
5401 u64 refs[BTRFS_MAX_LEVEL];
5402 u64 flags[BTRFS_MAX_LEVEL];
5403 struct btrfs_key update_progress;
5413 #define DROP_REFERENCE 1
5414 #define UPDATE_BACKREF 2
5416 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5417 struct btrfs_root *root,
5418 struct walk_control *wc,
5419 struct btrfs_path *path)
5428 struct btrfs_key key;
5429 struct extent_buffer *eb;
5434 if (path->slots[wc->level] < wc->reada_slot) {
5435 wc->reada_count = wc->reada_count * 2 / 3;
5436 wc->reada_count = max(wc->reada_count, 2);
5438 wc->reada_count = wc->reada_count * 3 / 2;
5439 wc->reada_count = min_t(int, wc->reada_count,
5440 BTRFS_NODEPTRS_PER_BLOCK(root));
5443 eb = path->nodes[wc->level];
5444 nritems = btrfs_header_nritems(eb);
5445 blocksize = btrfs_level_size(root, wc->level - 1);
5447 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5448 if (nread >= wc->reada_count)
5452 bytenr = btrfs_node_blockptr(eb, slot);
5453 generation = btrfs_node_ptr_generation(eb, slot);
5455 if (slot == path->slots[wc->level])
5458 if (wc->stage == UPDATE_BACKREF &&
5459 generation <= root->root_key.offset)
5462 /* We don't lock the tree block, it's OK to be racy here */
5463 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5468 if (wc->stage == DROP_REFERENCE) {
5472 if (wc->level == 1 &&
5473 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5475 if (!wc->update_ref ||
5476 generation <= root->root_key.offset)
5478 btrfs_node_key_to_cpu(eb, &key, slot);
5479 ret = btrfs_comp_cpu_keys(&key,
5480 &wc->update_progress);
5484 if (wc->level == 1 &&
5485 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5489 ret = readahead_tree_block(root, bytenr, blocksize,
5493 last = bytenr + blocksize;
5496 wc->reada_slot = slot;
5500 * hepler to process tree block while walking down the tree.
5502 * when wc->stage == UPDATE_BACKREF, this function updates
5503 * back refs for pointers in the block.
5505 * NOTE: return value 1 means we should stop walking down.
5507 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5508 struct btrfs_root *root,
5509 struct btrfs_path *path,
5510 struct walk_control *wc, int lookup_info)
5512 int level = wc->level;
5513 struct extent_buffer *eb = path->nodes[level];
5514 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5517 if (wc->stage == UPDATE_BACKREF &&
5518 btrfs_header_owner(eb) != root->root_key.objectid)
5522 * when reference count of tree block is 1, it won't increase
5523 * again. once full backref flag is set, we never clear it.
5526 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5527 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5528 BUG_ON(!path->locks[level]);
5529 ret = btrfs_lookup_extent_info(trans, root,
5534 BUG_ON(wc->refs[level] == 0);
5537 if (wc->stage == DROP_REFERENCE) {
5538 if (wc->refs[level] > 1)
5541 if (path->locks[level] && !wc->keep_locks) {
5542 btrfs_tree_unlock(eb);
5543 path->locks[level] = 0;
5548 /* wc->stage == UPDATE_BACKREF */
5549 if (!(wc->flags[level] & flag)) {
5550 BUG_ON(!path->locks[level]);
5551 ret = btrfs_inc_ref(trans, root, eb, 1);
5553 ret = btrfs_dec_ref(trans, root, eb, 0);
5555 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5558 wc->flags[level] |= flag;
5562 * the block is shared by multiple trees, so it's not good to
5563 * keep the tree lock
5565 if (path->locks[level] && level > 0) {
5566 btrfs_tree_unlock(eb);
5567 path->locks[level] = 0;
5573 * hepler to process tree block pointer.
5575 * when wc->stage == DROP_REFERENCE, this function checks
5576 * reference count of the block pointed to. if the block
5577 * is shared and we need update back refs for the subtree
5578 * rooted at the block, this function changes wc->stage to
5579 * UPDATE_BACKREF. if the block is shared and there is no
5580 * need to update back, this function drops the reference
5583 * NOTE: return value 1 means we should stop walking down.
5585 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5586 struct btrfs_root *root,
5587 struct btrfs_path *path,
5588 struct walk_control *wc, int *lookup_info)
5594 struct btrfs_key key;
5595 struct extent_buffer *next;
5596 int level = wc->level;
5600 generation = btrfs_node_ptr_generation(path->nodes[level],
5601 path->slots[level]);
5603 * if the lower level block was created before the snapshot
5604 * was created, we know there is no need to update back refs
5607 if (wc->stage == UPDATE_BACKREF &&
5608 generation <= root->root_key.offset) {
5613 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5614 blocksize = btrfs_level_size(root, level - 1);
5616 next = btrfs_find_tree_block(root, bytenr, blocksize);
5618 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5623 btrfs_tree_lock(next);
5624 btrfs_set_lock_blocking(next);
5626 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5627 &wc->refs[level - 1],
5628 &wc->flags[level - 1]);
5630 BUG_ON(wc->refs[level - 1] == 0);
5633 if (wc->stage == DROP_REFERENCE) {
5634 if (wc->refs[level - 1] > 1) {
5636 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5639 if (!wc->update_ref ||
5640 generation <= root->root_key.offset)
5643 btrfs_node_key_to_cpu(path->nodes[level], &key,
5644 path->slots[level]);
5645 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5649 wc->stage = UPDATE_BACKREF;
5650 wc->shared_level = level - 1;
5654 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5658 if (!btrfs_buffer_uptodate(next, generation)) {
5659 btrfs_tree_unlock(next);
5660 free_extent_buffer(next);
5666 if (reada && level == 1)
5667 reada_walk_down(trans, root, wc, path);
5668 next = read_tree_block(root, bytenr, blocksize, generation);
5669 btrfs_tree_lock(next);
5670 btrfs_set_lock_blocking(next);
5674 BUG_ON(level != btrfs_header_level(next));
5675 path->nodes[level] = next;
5676 path->slots[level] = 0;
5677 path->locks[level] = 1;
5683 wc->refs[level - 1] = 0;
5684 wc->flags[level - 1] = 0;
5685 if (wc->stage == DROP_REFERENCE) {
5686 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5687 parent = path->nodes[level]->start;
5689 BUG_ON(root->root_key.objectid !=
5690 btrfs_header_owner(path->nodes[level]));
5694 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5695 root->root_key.objectid, level - 1, 0);
5698 btrfs_tree_unlock(next);
5699 free_extent_buffer(next);
5705 * hepler to process tree block while walking up the tree.
5707 * when wc->stage == DROP_REFERENCE, this function drops
5708 * reference count on the block.
5710 * when wc->stage == UPDATE_BACKREF, this function changes
5711 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5712 * to UPDATE_BACKREF previously while processing the block.
5714 * NOTE: return value 1 means we should stop walking up.
5716 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5717 struct btrfs_root *root,
5718 struct btrfs_path *path,
5719 struct walk_control *wc)
5722 int level = wc->level;
5723 struct extent_buffer *eb = path->nodes[level];
5726 if (wc->stage == UPDATE_BACKREF) {
5727 BUG_ON(wc->shared_level < level);
5728 if (level < wc->shared_level)
5731 ret = find_next_key(path, level + 1, &wc->update_progress);
5735 wc->stage = DROP_REFERENCE;
5736 wc->shared_level = -1;
5737 path->slots[level] = 0;
5740 * check reference count again if the block isn't locked.
5741 * we should start walking down the tree again if reference
5744 if (!path->locks[level]) {
5746 btrfs_tree_lock(eb);
5747 btrfs_set_lock_blocking(eb);
5748 path->locks[level] = 1;
5750 ret = btrfs_lookup_extent_info(trans, root,
5755 BUG_ON(wc->refs[level] == 0);
5756 if (wc->refs[level] == 1) {
5757 btrfs_tree_unlock(eb);
5758 path->locks[level] = 0;
5764 /* wc->stage == DROP_REFERENCE */
5765 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5767 if (wc->refs[level] == 1) {
5769 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5770 ret = btrfs_dec_ref(trans, root, eb, 1);
5772 ret = btrfs_dec_ref(trans, root, eb, 0);
5775 /* make block locked assertion in clean_tree_block happy */
5776 if (!path->locks[level] &&
5777 btrfs_header_generation(eb) == trans->transid) {
5778 btrfs_tree_lock(eb);
5779 btrfs_set_lock_blocking(eb);
5780 path->locks[level] = 1;
5782 clean_tree_block(trans, root, eb);
5785 if (eb == root->node) {
5786 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5789 BUG_ON(root->root_key.objectid !=
5790 btrfs_header_owner(eb));
5792 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5793 parent = path->nodes[level + 1]->start;
5795 BUG_ON(root->root_key.objectid !=
5796 btrfs_header_owner(path->nodes[level + 1]));
5799 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5801 wc->refs[level] = 0;
5802 wc->flags[level] = 0;
5806 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5807 struct btrfs_root *root,
5808 struct btrfs_path *path,
5809 struct walk_control *wc)
5811 int level = wc->level;
5812 int lookup_info = 1;
5815 while (level >= 0) {
5816 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5823 if (path->slots[level] >=
5824 btrfs_header_nritems(path->nodes[level]))
5827 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5829 path->slots[level]++;
5838 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5839 struct btrfs_root *root,
5840 struct btrfs_path *path,
5841 struct walk_control *wc, int max_level)
5843 int level = wc->level;
5846 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5847 while (level < max_level && path->nodes[level]) {
5849 if (path->slots[level] + 1 <
5850 btrfs_header_nritems(path->nodes[level])) {
5851 path->slots[level]++;
5854 ret = walk_up_proc(trans, root, path, wc);
5858 if (path->locks[level]) {
5859 btrfs_tree_unlock(path->nodes[level]);
5860 path->locks[level] = 0;
5862 free_extent_buffer(path->nodes[level]);
5863 path->nodes[level] = NULL;
5871 * drop a subvolume tree.
5873 * this function traverses the tree freeing any blocks that only
5874 * referenced by the tree.
5876 * when a shared tree block is found. this function decreases its
5877 * reference count by one. if update_ref is true, this function
5878 * also make sure backrefs for the shared block and all lower level
5879 * blocks are properly updated.
5881 int btrfs_drop_snapshot(struct btrfs_root *root,
5882 struct btrfs_block_rsv *block_rsv, int update_ref)
5884 struct btrfs_path *path;
5885 struct btrfs_trans_handle *trans;
5886 struct btrfs_root *tree_root = root->fs_info->tree_root;
5887 struct btrfs_root_item *root_item = &root->root_item;
5888 struct walk_control *wc;
5889 struct btrfs_key key;
5894 path = btrfs_alloc_path();
5897 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5900 trans = btrfs_start_transaction(tree_root, 0);
5902 trans->block_rsv = block_rsv;
5904 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5905 level = btrfs_header_level(root->node);
5906 path->nodes[level] = btrfs_lock_root_node(root);
5907 btrfs_set_lock_blocking(path->nodes[level]);
5908 path->slots[level] = 0;
5909 path->locks[level] = 1;
5910 memset(&wc->update_progress, 0,
5911 sizeof(wc->update_progress));
5913 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5914 memcpy(&wc->update_progress, &key,
5915 sizeof(wc->update_progress));
5917 level = root_item->drop_level;
5919 path->lowest_level = level;
5920 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5921 path->lowest_level = 0;
5929 * unlock our path, this is safe because only this
5930 * function is allowed to delete this snapshot
5932 btrfs_unlock_up_safe(path, 0);
5934 level = btrfs_header_level(root->node);
5936 btrfs_tree_lock(path->nodes[level]);
5937 btrfs_set_lock_blocking(path->nodes[level]);
5939 ret = btrfs_lookup_extent_info(trans, root,
5940 path->nodes[level]->start,
5941 path->nodes[level]->len,
5945 BUG_ON(wc->refs[level] == 0);
5947 if (level == root_item->drop_level)
5950 btrfs_tree_unlock(path->nodes[level]);
5951 WARN_ON(wc->refs[level] != 1);
5957 wc->shared_level = -1;
5958 wc->stage = DROP_REFERENCE;
5959 wc->update_ref = update_ref;
5961 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5964 ret = walk_down_tree(trans, root, path, wc);
5970 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5977 BUG_ON(wc->stage != DROP_REFERENCE);
5981 if (wc->stage == DROP_REFERENCE) {
5983 btrfs_node_key(path->nodes[level],
5984 &root_item->drop_progress,
5985 path->slots[level]);
5986 root_item->drop_level = level;
5989 BUG_ON(wc->level == 0);
5990 if (btrfs_should_end_transaction(trans, tree_root)) {
5991 ret = btrfs_update_root(trans, tree_root,
5996 btrfs_end_transaction_throttle(trans, tree_root);
5997 trans = btrfs_start_transaction(tree_root, 0);
5999 trans->block_rsv = block_rsv;
6002 btrfs_release_path(root, path);
6005 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6008 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6009 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6013 ret = btrfs_del_orphan_item(trans, tree_root,
6014 root->root_key.objectid);
6019 if (root->in_radix) {
6020 btrfs_free_fs_root(tree_root->fs_info, root);
6022 free_extent_buffer(root->node);
6023 free_extent_buffer(root->commit_root);
6027 btrfs_end_transaction_throttle(trans, tree_root);
6029 btrfs_free_path(path);
6034 * drop subtree rooted at tree block 'node'.
6036 * NOTE: this function will unlock and release tree block 'node'
6038 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6039 struct btrfs_root *root,
6040 struct extent_buffer *node,
6041 struct extent_buffer *parent)
6043 struct btrfs_path *path;
6044 struct walk_control *wc;
6050 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6052 path = btrfs_alloc_path();
6055 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6058 btrfs_assert_tree_locked(parent);
6059 parent_level = btrfs_header_level(parent);
6060 extent_buffer_get(parent);
6061 path->nodes[parent_level] = parent;
6062 path->slots[parent_level] = btrfs_header_nritems(parent);
6064 btrfs_assert_tree_locked(node);
6065 level = btrfs_header_level(node);
6066 path->nodes[level] = node;
6067 path->slots[level] = 0;
6068 path->locks[level] = 1;
6070 wc->refs[parent_level] = 1;
6071 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6073 wc->shared_level = -1;
6074 wc->stage = DROP_REFERENCE;
6077 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6080 wret = walk_down_tree(trans, root, path, wc);
6086 wret = walk_up_tree(trans, root, path, wc, parent_level);
6094 btrfs_free_path(path);
6099 static unsigned long calc_ra(unsigned long start, unsigned long last,
6102 return min(last, start + nr - 1);
6105 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6110 unsigned long first_index;
6111 unsigned long last_index;
6114 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6115 struct file_ra_state *ra;
6116 struct btrfs_ordered_extent *ordered;
6117 unsigned int total_read = 0;
6118 unsigned int total_dirty = 0;
6121 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6123 mutex_lock(&inode->i_mutex);
6124 first_index = start >> PAGE_CACHE_SHIFT;
6125 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6127 /* make sure the dirty trick played by the caller work */
6128 ret = invalidate_inode_pages2_range(inode->i_mapping,
6129 first_index, last_index);
6133 file_ra_state_init(ra, inode->i_mapping);
6135 for (i = first_index ; i <= last_index; i++) {
6136 if (total_read % ra->ra_pages == 0) {
6137 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6138 calc_ra(i, last_index, ra->ra_pages));
6142 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6144 page = grab_cache_page(inode->i_mapping, i);
6149 if (!PageUptodate(page)) {
6150 btrfs_readpage(NULL, page);
6152 if (!PageUptodate(page)) {
6154 page_cache_release(page);
6159 wait_on_page_writeback(page);
6161 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6162 page_end = page_start + PAGE_CACHE_SIZE - 1;
6163 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6165 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6167 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6169 page_cache_release(page);
6170 btrfs_start_ordered_extent(inode, ordered, 1);
6171 btrfs_put_ordered_extent(ordered);
6174 set_page_extent_mapped(page);
6176 if (i == first_index)
6177 set_extent_bits(io_tree, page_start, page_end,
6178 EXTENT_BOUNDARY, GFP_NOFS);
6179 btrfs_set_extent_delalloc(inode, page_start, page_end);
6181 set_page_dirty(page);
6184 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6186 page_cache_release(page);
6191 mutex_unlock(&inode->i_mutex);
6192 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6196 static noinline int relocate_data_extent(struct inode *reloc_inode,
6197 struct btrfs_key *extent_key,
6200 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6201 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6202 struct extent_map *em;
6203 u64 start = extent_key->objectid - offset;
6204 u64 end = start + extent_key->offset - 1;
6206 em = alloc_extent_map(GFP_NOFS);
6207 BUG_ON(!em || IS_ERR(em));
6210 em->len = extent_key->offset;
6211 em->block_len = extent_key->offset;
6212 em->block_start = extent_key->objectid;
6213 em->bdev = root->fs_info->fs_devices->latest_bdev;
6214 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6216 /* setup extent map to cheat btrfs_readpage */
6217 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6220 write_lock(&em_tree->lock);
6221 ret = add_extent_mapping(em_tree, em);
6222 write_unlock(&em_tree->lock);
6223 if (ret != -EEXIST) {
6224 free_extent_map(em);
6227 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6229 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6231 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6234 struct btrfs_ref_path {
6236 u64 nodes[BTRFS_MAX_LEVEL];
6238 u64 root_generation;
6245 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6246 u64 new_nodes[BTRFS_MAX_LEVEL];
6249 struct disk_extent {
6260 static int is_cowonly_root(u64 root_objectid)
6262 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6263 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6264 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6265 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6266 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6267 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6272 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6273 struct btrfs_root *extent_root,
6274 struct btrfs_ref_path *ref_path,
6277 struct extent_buffer *leaf;
6278 struct btrfs_path *path;
6279 struct btrfs_extent_ref *ref;
6280 struct btrfs_key key;
6281 struct btrfs_key found_key;
6287 path = btrfs_alloc_path();
6292 ref_path->lowest_level = -1;
6293 ref_path->current_level = -1;
6294 ref_path->shared_level = -1;
6298 level = ref_path->current_level - 1;
6299 while (level >= -1) {
6301 if (level < ref_path->lowest_level)
6305 bytenr = ref_path->nodes[level];
6307 bytenr = ref_path->extent_start;
6308 BUG_ON(bytenr == 0);
6310 parent = ref_path->nodes[level + 1];
6311 ref_path->nodes[level + 1] = 0;
6312 ref_path->current_level = level;
6313 BUG_ON(parent == 0);
6315 key.objectid = bytenr;
6316 key.offset = parent + 1;
6317 key.type = BTRFS_EXTENT_REF_KEY;
6319 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6324 leaf = path->nodes[0];
6325 nritems = btrfs_header_nritems(leaf);
6326 if (path->slots[0] >= nritems) {
6327 ret = btrfs_next_leaf(extent_root, path);
6332 leaf = path->nodes[0];
6335 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6336 if (found_key.objectid == bytenr &&
6337 found_key.type == BTRFS_EXTENT_REF_KEY) {
6338 if (level < ref_path->shared_level)
6339 ref_path->shared_level = level;
6344 btrfs_release_path(extent_root, path);
6347 /* reached lowest level */
6351 level = ref_path->current_level;
6352 while (level < BTRFS_MAX_LEVEL - 1) {
6356 bytenr = ref_path->nodes[level];
6358 bytenr = ref_path->extent_start;
6360 BUG_ON(bytenr == 0);
6362 key.objectid = bytenr;
6364 key.type = BTRFS_EXTENT_REF_KEY;
6366 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6370 leaf = path->nodes[0];
6371 nritems = btrfs_header_nritems(leaf);
6372 if (path->slots[0] >= nritems) {
6373 ret = btrfs_next_leaf(extent_root, path);
6377 /* the extent was freed by someone */
6378 if (ref_path->lowest_level == level)
6380 btrfs_release_path(extent_root, path);
6383 leaf = path->nodes[0];
6386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6387 if (found_key.objectid != bytenr ||
6388 found_key.type != BTRFS_EXTENT_REF_KEY) {
6389 /* the extent was freed by someone */
6390 if (ref_path->lowest_level == level) {
6394 btrfs_release_path(extent_root, path);
6398 ref = btrfs_item_ptr(leaf, path->slots[0],
6399 struct btrfs_extent_ref);
6400 ref_objectid = btrfs_ref_objectid(leaf, ref);
6401 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6403 level = (int)ref_objectid;
6404 BUG_ON(level >= BTRFS_MAX_LEVEL);
6405 ref_path->lowest_level = level;
6406 ref_path->current_level = level;
6407 ref_path->nodes[level] = bytenr;
6409 WARN_ON(ref_objectid != level);
6412 WARN_ON(level != -1);
6416 if (ref_path->lowest_level == level) {
6417 ref_path->owner_objectid = ref_objectid;
6418 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6422 * the block is tree root or the block isn't in reference
6425 if (found_key.objectid == found_key.offset ||
6426 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6427 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6428 ref_path->root_generation =
6429 btrfs_ref_generation(leaf, ref);
6431 /* special reference from the tree log */
6432 ref_path->nodes[0] = found_key.offset;
6433 ref_path->current_level = 0;
6440 BUG_ON(ref_path->nodes[level] != 0);
6441 ref_path->nodes[level] = found_key.offset;
6442 ref_path->current_level = level;
6445 * the reference was created in the running transaction,
6446 * no need to continue walking up.
6448 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6449 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6450 ref_path->root_generation =
6451 btrfs_ref_generation(leaf, ref);
6456 btrfs_release_path(extent_root, path);
6459 /* reached max tree level, but no tree root found. */
6462 btrfs_free_path(path);
6466 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6467 struct btrfs_root *extent_root,
6468 struct btrfs_ref_path *ref_path,
6471 memset(ref_path, 0, sizeof(*ref_path));
6472 ref_path->extent_start = extent_start;
6474 return __next_ref_path(trans, extent_root, ref_path, 1);
6477 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6478 struct btrfs_root *extent_root,
6479 struct btrfs_ref_path *ref_path)
6481 return __next_ref_path(trans, extent_root, ref_path, 0);
6484 static noinline int get_new_locations(struct inode *reloc_inode,
6485 struct btrfs_key *extent_key,
6486 u64 offset, int no_fragment,
6487 struct disk_extent **extents,
6490 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6491 struct btrfs_path *path;
6492 struct btrfs_file_extent_item *fi;
6493 struct extent_buffer *leaf;
6494 struct disk_extent *exts = *extents;
6495 struct btrfs_key found_key;
6500 int max = *nr_extents;
6503 WARN_ON(!no_fragment && *extents);
6506 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6511 path = btrfs_alloc_path();
6514 cur_pos = extent_key->objectid - offset;
6515 last_byte = extent_key->objectid + extent_key->offset;
6516 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6526 leaf = path->nodes[0];
6527 nritems = btrfs_header_nritems(leaf);
6528 if (path->slots[0] >= nritems) {
6529 ret = btrfs_next_leaf(root, path);
6534 leaf = path->nodes[0];
6537 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6538 if (found_key.offset != cur_pos ||
6539 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6540 found_key.objectid != reloc_inode->i_ino)
6543 fi = btrfs_item_ptr(leaf, path->slots[0],
6544 struct btrfs_file_extent_item);
6545 if (btrfs_file_extent_type(leaf, fi) !=
6546 BTRFS_FILE_EXTENT_REG ||
6547 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6551 struct disk_extent *old = exts;
6553 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6554 memcpy(exts, old, sizeof(*exts) * nr);
6555 if (old != *extents)
6559 exts[nr].disk_bytenr =
6560 btrfs_file_extent_disk_bytenr(leaf, fi);
6561 exts[nr].disk_num_bytes =
6562 btrfs_file_extent_disk_num_bytes(leaf, fi);
6563 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6564 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6565 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6566 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6567 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6568 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6570 BUG_ON(exts[nr].offset > 0);
6571 BUG_ON(exts[nr].compression || exts[nr].encryption);
6572 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6574 cur_pos += exts[nr].num_bytes;
6577 if (cur_pos + offset >= last_byte)
6587 BUG_ON(cur_pos + offset > last_byte);
6588 if (cur_pos + offset < last_byte) {
6594 btrfs_free_path(path);
6596 if (exts != *extents)
6605 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6606 struct btrfs_root *root,
6607 struct btrfs_path *path,
6608 struct btrfs_key *extent_key,
6609 struct btrfs_key *leaf_key,
6610 struct btrfs_ref_path *ref_path,
6611 struct disk_extent *new_extents,
6614 struct extent_buffer *leaf;
6615 struct btrfs_file_extent_item *fi;
6616 struct inode *inode = NULL;
6617 struct btrfs_key key;
6622 u64 search_end = (u64)-1;
6625 int extent_locked = 0;
6629 memcpy(&key, leaf_key, sizeof(key));
6630 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6631 if (key.objectid < ref_path->owner_objectid ||
6632 (key.objectid == ref_path->owner_objectid &&
6633 key.type < BTRFS_EXTENT_DATA_KEY)) {
6634 key.objectid = ref_path->owner_objectid;
6635 key.type = BTRFS_EXTENT_DATA_KEY;
6641 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6645 leaf = path->nodes[0];
6646 nritems = btrfs_header_nritems(leaf);
6648 if (extent_locked && ret > 0) {
6650 * the file extent item was modified by someone
6651 * before the extent got locked.
6653 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6654 lock_end, GFP_NOFS);
6658 if (path->slots[0] >= nritems) {
6659 if (++nr_scaned > 2)
6662 BUG_ON(extent_locked);
6663 ret = btrfs_next_leaf(root, path);
6668 leaf = path->nodes[0];
6669 nritems = btrfs_header_nritems(leaf);
6672 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6674 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6675 if ((key.objectid > ref_path->owner_objectid) ||
6676 (key.objectid == ref_path->owner_objectid &&
6677 key.type > BTRFS_EXTENT_DATA_KEY) ||
6678 key.offset >= search_end)
6682 if (inode && key.objectid != inode->i_ino) {
6683 BUG_ON(extent_locked);
6684 btrfs_release_path(root, path);
6685 mutex_unlock(&inode->i_mutex);
6691 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6696 fi = btrfs_item_ptr(leaf, path->slots[0],
6697 struct btrfs_file_extent_item);
6698 extent_type = btrfs_file_extent_type(leaf, fi);
6699 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6700 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6701 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6702 extent_key->objectid)) {
6708 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6709 ext_offset = btrfs_file_extent_offset(leaf, fi);
6711 if (search_end == (u64)-1) {
6712 search_end = key.offset - ext_offset +
6713 btrfs_file_extent_ram_bytes(leaf, fi);
6716 if (!extent_locked) {
6717 lock_start = key.offset;
6718 lock_end = lock_start + num_bytes - 1;
6720 if (lock_start > key.offset ||
6721 lock_end + 1 < key.offset + num_bytes) {
6722 unlock_extent(&BTRFS_I(inode)->io_tree,
6723 lock_start, lock_end, GFP_NOFS);
6729 btrfs_release_path(root, path);
6731 inode = btrfs_iget_locked(root->fs_info->sb,
6732 key.objectid, root);
6733 if (inode->i_state & I_NEW) {
6734 BTRFS_I(inode)->root = root;
6735 BTRFS_I(inode)->location.objectid =
6737 BTRFS_I(inode)->location.type =
6738 BTRFS_INODE_ITEM_KEY;
6739 BTRFS_I(inode)->location.offset = 0;
6740 btrfs_read_locked_inode(inode);
6741 unlock_new_inode(inode);
6744 * some code call btrfs_commit_transaction while
6745 * holding the i_mutex, so we can't use mutex_lock
6748 if (is_bad_inode(inode) ||
6749 !mutex_trylock(&inode->i_mutex)) {
6752 key.offset = (u64)-1;
6757 if (!extent_locked) {
6758 struct btrfs_ordered_extent *ordered;
6760 btrfs_release_path(root, path);
6762 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6763 lock_end, GFP_NOFS);
6764 ordered = btrfs_lookup_first_ordered_extent(inode,
6767 ordered->file_offset <= lock_end &&
6768 ordered->file_offset + ordered->len > lock_start) {
6769 unlock_extent(&BTRFS_I(inode)->io_tree,
6770 lock_start, lock_end, GFP_NOFS);
6771 btrfs_start_ordered_extent(inode, ordered, 1);
6772 btrfs_put_ordered_extent(ordered);
6773 key.offset += num_bytes;
6777 btrfs_put_ordered_extent(ordered);
6783 if (nr_extents == 1) {
6784 /* update extent pointer in place */
6785 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6786 new_extents[0].disk_bytenr);
6787 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6788 new_extents[0].disk_num_bytes);
6789 btrfs_mark_buffer_dirty(leaf);
6791 btrfs_drop_extent_cache(inode, key.offset,
6792 key.offset + num_bytes - 1, 0);
6794 ret = btrfs_inc_extent_ref(trans, root,
6795 new_extents[0].disk_bytenr,
6796 new_extents[0].disk_num_bytes,
6798 root->root_key.objectid,
6803 ret = btrfs_free_extent(trans, root,
6804 extent_key->objectid,
6807 btrfs_header_owner(leaf),
6808 btrfs_header_generation(leaf),
6812 btrfs_release_path(root, path);
6813 key.offset += num_bytes;
6821 * drop old extent pointer at first, then insert the
6822 * new pointers one bye one
6824 btrfs_release_path(root, path);
6825 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6826 key.offset + num_bytes,
6827 key.offset, &alloc_hint);
6830 for (i = 0; i < nr_extents; i++) {
6831 if (ext_offset >= new_extents[i].num_bytes) {
6832 ext_offset -= new_extents[i].num_bytes;
6835 extent_len = min(new_extents[i].num_bytes -
6836 ext_offset, num_bytes);
6838 ret = btrfs_insert_empty_item(trans, root,
6843 leaf = path->nodes[0];
6844 fi = btrfs_item_ptr(leaf, path->slots[0],
6845 struct btrfs_file_extent_item);
6846 btrfs_set_file_extent_generation(leaf, fi,
6848 btrfs_set_file_extent_type(leaf, fi,
6849 BTRFS_FILE_EXTENT_REG);
6850 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6851 new_extents[i].disk_bytenr);
6852 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6853 new_extents[i].disk_num_bytes);
6854 btrfs_set_file_extent_ram_bytes(leaf, fi,
6855 new_extents[i].ram_bytes);
6857 btrfs_set_file_extent_compression(leaf, fi,
6858 new_extents[i].compression);
6859 btrfs_set_file_extent_encryption(leaf, fi,
6860 new_extents[i].encryption);
6861 btrfs_set_file_extent_other_encoding(leaf, fi,
6862 new_extents[i].other_encoding);
6864 btrfs_set_file_extent_num_bytes(leaf, fi,
6866 ext_offset += new_extents[i].offset;
6867 btrfs_set_file_extent_offset(leaf, fi,
6869 btrfs_mark_buffer_dirty(leaf);
6871 btrfs_drop_extent_cache(inode, key.offset,
6872 key.offset + extent_len - 1, 0);
6874 ret = btrfs_inc_extent_ref(trans, root,
6875 new_extents[i].disk_bytenr,
6876 new_extents[i].disk_num_bytes,
6878 root->root_key.objectid,
6879 trans->transid, key.objectid);
6881 btrfs_release_path(root, path);
6883 inode_add_bytes(inode, extent_len);
6886 num_bytes -= extent_len;
6887 key.offset += extent_len;
6892 BUG_ON(i >= nr_extents);
6896 if (extent_locked) {
6897 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6898 lock_end, GFP_NOFS);
6902 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6903 key.offset >= search_end)
6910 btrfs_release_path(root, path);
6912 mutex_unlock(&inode->i_mutex);
6913 if (extent_locked) {
6914 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6915 lock_end, GFP_NOFS);
6922 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6923 struct btrfs_root *root,
6924 struct extent_buffer *buf, u64 orig_start)
6929 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6930 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6932 level = btrfs_header_level(buf);
6934 struct btrfs_leaf_ref *ref;
6935 struct btrfs_leaf_ref *orig_ref;
6937 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6941 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6943 btrfs_free_leaf_ref(root, orig_ref);
6947 ref->nritems = orig_ref->nritems;
6948 memcpy(ref->extents, orig_ref->extents,
6949 sizeof(ref->extents[0]) * ref->nritems);
6951 btrfs_free_leaf_ref(root, orig_ref);
6953 ref->root_gen = trans->transid;
6954 ref->bytenr = buf->start;
6955 ref->owner = btrfs_header_owner(buf);
6956 ref->generation = btrfs_header_generation(buf);
6958 ret = btrfs_add_leaf_ref(root, ref, 0);
6960 btrfs_free_leaf_ref(root, ref);
6965 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6966 struct extent_buffer *leaf,
6967 struct btrfs_block_group_cache *group,
6968 struct btrfs_root *target_root)
6970 struct btrfs_key key;
6971 struct inode *inode = NULL;
6972 struct btrfs_file_extent_item *fi;
6973 struct extent_state *cached_state = NULL;
6975 u64 skip_objectid = 0;
6979 nritems = btrfs_header_nritems(leaf);
6980 for (i = 0; i < nritems; i++) {
6981 btrfs_item_key_to_cpu(leaf, &key, i);
6982 if (key.objectid == skip_objectid ||
6983 key.type != BTRFS_EXTENT_DATA_KEY)
6985 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6986 if (btrfs_file_extent_type(leaf, fi) ==
6987 BTRFS_FILE_EXTENT_INLINE)
6989 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6991 if (!inode || inode->i_ino != key.objectid) {
6993 inode = btrfs_ilookup(target_root->fs_info->sb,
6994 key.objectid, target_root, 1);
6997 skip_objectid = key.objectid;
7000 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7002 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7003 key.offset + num_bytes - 1, 0, &cached_state,
7005 btrfs_drop_extent_cache(inode, key.offset,
7006 key.offset + num_bytes - 1, 1);
7007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7008 key.offset + num_bytes - 1, &cached_state,
7016 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7017 struct btrfs_root *root,
7018 struct extent_buffer *leaf,
7019 struct btrfs_block_group_cache *group,
7020 struct inode *reloc_inode)
7022 struct btrfs_key key;
7023 struct btrfs_key extent_key;
7024 struct btrfs_file_extent_item *fi;
7025 struct btrfs_leaf_ref *ref;
7026 struct disk_extent *new_extent;
7035 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7036 BUG_ON(!new_extent);
7038 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7042 nritems = btrfs_header_nritems(leaf);
7043 for (i = 0; i < nritems; i++) {
7044 btrfs_item_key_to_cpu(leaf, &key, i);
7045 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7047 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7048 if (btrfs_file_extent_type(leaf, fi) ==
7049 BTRFS_FILE_EXTENT_INLINE)
7051 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7052 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7057 if (bytenr >= group->key.objectid + group->key.offset ||
7058 bytenr + num_bytes <= group->key.objectid)
7061 extent_key.objectid = bytenr;
7062 extent_key.offset = num_bytes;
7063 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7065 ret = get_new_locations(reloc_inode, &extent_key,
7066 group->key.objectid, 1,
7067 &new_extent, &nr_extent);
7072 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7073 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7074 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7075 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7077 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7078 new_extent->disk_bytenr);
7079 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7080 new_extent->disk_num_bytes);
7081 btrfs_mark_buffer_dirty(leaf);
7083 ret = btrfs_inc_extent_ref(trans, root,
7084 new_extent->disk_bytenr,
7085 new_extent->disk_num_bytes,
7087 root->root_key.objectid,
7088 trans->transid, key.objectid);
7091 ret = btrfs_free_extent(trans, root,
7092 bytenr, num_bytes, leaf->start,
7093 btrfs_header_owner(leaf),
7094 btrfs_header_generation(leaf),
7100 BUG_ON(ext_index + 1 != ref->nritems);
7101 btrfs_free_leaf_ref(root, ref);
7105 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7106 struct btrfs_root *root)
7108 struct btrfs_root *reloc_root;
7111 if (root->reloc_root) {
7112 reloc_root = root->reloc_root;
7113 root->reloc_root = NULL;
7114 list_add(&reloc_root->dead_list,
7115 &root->fs_info->dead_reloc_roots);
7117 btrfs_set_root_bytenr(&reloc_root->root_item,
7118 reloc_root->node->start);
7119 btrfs_set_root_level(&root->root_item,
7120 btrfs_header_level(reloc_root->node));
7121 memset(&reloc_root->root_item.drop_progress, 0,
7122 sizeof(struct btrfs_disk_key));
7123 reloc_root->root_item.drop_level = 0;
7125 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7126 &reloc_root->root_key,
7127 &reloc_root->root_item);
7133 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7135 struct btrfs_trans_handle *trans;
7136 struct btrfs_root *reloc_root;
7137 struct btrfs_root *prev_root = NULL;
7138 struct list_head dead_roots;
7142 INIT_LIST_HEAD(&dead_roots);
7143 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7145 while (!list_empty(&dead_roots)) {
7146 reloc_root = list_entry(dead_roots.prev,
7147 struct btrfs_root, dead_list);
7148 list_del_init(&reloc_root->dead_list);
7150 BUG_ON(reloc_root->commit_root != NULL);
7152 trans = btrfs_join_transaction(root, 1);
7155 mutex_lock(&root->fs_info->drop_mutex);
7156 ret = btrfs_drop_snapshot(trans, reloc_root);
7159 mutex_unlock(&root->fs_info->drop_mutex);
7161 nr = trans->blocks_used;
7162 ret = btrfs_end_transaction(trans, root);
7164 btrfs_btree_balance_dirty(root, nr);
7167 free_extent_buffer(reloc_root->node);
7169 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7170 &reloc_root->root_key);
7172 mutex_unlock(&root->fs_info->drop_mutex);
7174 nr = trans->blocks_used;
7175 ret = btrfs_end_transaction(trans, root);
7177 btrfs_btree_balance_dirty(root, nr);
7180 prev_root = reloc_root;
7183 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7189 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7191 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7195 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7197 struct btrfs_root *reloc_root;
7198 struct btrfs_trans_handle *trans;
7199 struct btrfs_key location;
7203 mutex_lock(&root->fs_info->tree_reloc_mutex);
7204 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7206 found = !list_empty(&root->fs_info->dead_reloc_roots);
7207 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7210 trans = btrfs_start_transaction(root, 1);
7212 ret = btrfs_commit_transaction(trans, root);
7216 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7217 location.offset = (u64)-1;
7218 location.type = BTRFS_ROOT_ITEM_KEY;
7220 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7221 BUG_ON(!reloc_root);
7222 btrfs_orphan_cleanup(reloc_root);
7226 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7227 struct btrfs_root *root)
7229 struct btrfs_root *reloc_root;
7230 struct extent_buffer *eb;
7231 struct btrfs_root_item *root_item;
7232 struct btrfs_key root_key;
7235 BUG_ON(!root->ref_cows);
7236 if (root->reloc_root)
7239 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7242 ret = btrfs_copy_root(trans, root, root->commit_root,
7243 &eb, BTRFS_TREE_RELOC_OBJECTID);
7246 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7247 root_key.offset = root->root_key.objectid;
7248 root_key.type = BTRFS_ROOT_ITEM_KEY;
7250 memcpy(root_item, &root->root_item, sizeof(root_item));
7251 btrfs_set_root_refs(root_item, 0);
7252 btrfs_set_root_bytenr(root_item, eb->start);
7253 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7254 btrfs_set_root_generation(root_item, trans->transid);
7256 btrfs_tree_unlock(eb);
7257 free_extent_buffer(eb);
7259 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7260 &root_key, root_item);
7264 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7266 BUG_ON(!reloc_root);
7267 reloc_root->last_trans = trans->transid;
7268 reloc_root->commit_root = NULL;
7269 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7271 root->reloc_root = reloc_root;
7276 * Core function of space balance.
7278 * The idea is using reloc trees to relocate tree blocks in reference
7279 * counted roots. There is one reloc tree for each subvol, and all
7280 * reloc trees share same root key objectid. Reloc trees are snapshots
7281 * of the latest committed roots of subvols (root->commit_root).
7283 * To relocate a tree block referenced by a subvol, there are two steps.
7284 * COW the block through subvol's reloc tree, then update block pointer
7285 * in the subvol to point to the new block. Since all reloc trees share
7286 * same root key objectid, doing special handing for tree blocks owned
7287 * by them is easy. Once a tree block has been COWed in one reloc tree,
7288 * we can use the resulting new block directly when the same block is
7289 * required to COW again through other reloc trees. By this way, relocated
7290 * tree blocks are shared between reloc trees, so they are also shared
7293 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7294 struct btrfs_root *root,
7295 struct btrfs_path *path,
7296 struct btrfs_key *first_key,
7297 struct btrfs_ref_path *ref_path,
7298 struct btrfs_block_group_cache *group,
7299 struct inode *reloc_inode)
7301 struct btrfs_root *reloc_root;
7302 struct extent_buffer *eb = NULL;
7303 struct btrfs_key *keys;
7307 int lowest_level = 0;
7310 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7311 lowest_level = ref_path->owner_objectid;
7313 if (!root->ref_cows) {
7314 path->lowest_level = lowest_level;
7315 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7317 path->lowest_level = 0;
7318 btrfs_release_path(root, path);
7322 mutex_lock(&root->fs_info->tree_reloc_mutex);
7323 ret = init_reloc_tree(trans, root);
7325 reloc_root = root->reloc_root;
7327 shared_level = ref_path->shared_level;
7328 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7330 keys = ref_path->node_keys;
7331 nodes = ref_path->new_nodes;
7332 memset(&keys[shared_level + 1], 0,
7333 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7334 memset(&nodes[shared_level + 1], 0,
7335 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7337 if (nodes[lowest_level] == 0) {
7338 path->lowest_level = lowest_level;
7339 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7342 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7343 eb = path->nodes[level];
7344 if (!eb || eb == reloc_root->node)
7346 nodes[level] = eb->start;
7348 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7350 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7353 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7354 eb = path->nodes[0];
7355 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7356 group, reloc_inode);
7359 btrfs_release_path(reloc_root, path);
7361 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7367 * replace tree blocks in the fs tree with tree blocks in
7370 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7373 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7374 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7377 extent_buffer_get(path->nodes[0]);
7378 eb = path->nodes[0];
7379 btrfs_release_path(reloc_root, path);
7380 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7382 free_extent_buffer(eb);
7385 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7386 path->lowest_level = 0;
7390 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7391 struct btrfs_root *root,
7392 struct btrfs_path *path,
7393 struct btrfs_key *first_key,
7394 struct btrfs_ref_path *ref_path)
7398 ret = relocate_one_path(trans, root, path, first_key,
7399 ref_path, NULL, NULL);
7405 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7406 struct btrfs_root *extent_root,
7407 struct btrfs_path *path,
7408 struct btrfs_key *extent_key)
7412 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7415 ret = btrfs_del_item(trans, extent_root, path);
7417 btrfs_release_path(extent_root, path);
7421 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7422 struct btrfs_ref_path *ref_path)
7424 struct btrfs_key root_key;
7426 root_key.objectid = ref_path->root_objectid;
7427 root_key.type = BTRFS_ROOT_ITEM_KEY;
7428 if (is_cowonly_root(ref_path->root_objectid))
7429 root_key.offset = 0;
7431 root_key.offset = (u64)-1;
7433 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7436 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7437 struct btrfs_path *path,
7438 struct btrfs_key *extent_key,
7439 struct btrfs_block_group_cache *group,
7440 struct inode *reloc_inode, int pass)
7442 struct btrfs_trans_handle *trans;
7443 struct btrfs_root *found_root;
7444 struct btrfs_ref_path *ref_path = NULL;
7445 struct disk_extent *new_extents = NULL;
7450 struct btrfs_key first_key;
7454 trans = btrfs_start_transaction(extent_root, 1);
7457 if (extent_key->objectid == 0) {
7458 ret = del_extent_zero(trans, extent_root, path, extent_key);
7462 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7468 for (loops = 0; ; loops++) {
7470 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7471 extent_key->objectid);
7473 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7480 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7481 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7484 found_root = read_ref_root(extent_root->fs_info, ref_path);
7485 BUG_ON(!found_root);
7487 * for reference counted tree, only process reference paths
7488 * rooted at the latest committed root.
7490 if (found_root->ref_cows &&
7491 ref_path->root_generation != found_root->root_key.offset)
7494 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7497 * copy data extents to new locations
7499 u64 group_start = group->key.objectid;
7500 ret = relocate_data_extent(reloc_inode,
7509 level = ref_path->owner_objectid;
7512 if (prev_block != ref_path->nodes[level]) {
7513 struct extent_buffer *eb;
7514 u64 block_start = ref_path->nodes[level];
7515 u64 block_size = btrfs_level_size(found_root, level);
7517 eb = read_tree_block(found_root, block_start,
7519 btrfs_tree_lock(eb);
7520 BUG_ON(level != btrfs_header_level(eb));
7523 btrfs_item_key_to_cpu(eb, &first_key, 0);
7525 btrfs_node_key_to_cpu(eb, &first_key, 0);
7527 btrfs_tree_unlock(eb);
7528 free_extent_buffer(eb);
7529 prev_block = block_start;
7532 mutex_lock(&extent_root->fs_info->trans_mutex);
7533 btrfs_record_root_in_trans(found_root);
7534 mutex_unlock(&extent_root->fs_info->trans_mutex);
7535 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7537 * try to update data extent references while
7538 * keeping metadata shared between snapshots.
7541 ret = relocate_one_path(trans, found_root,
7542 path, &first_key, ref_path,
7543 group, reloc_inode);
7549 * use fallback method to process the remaining
7553 u64 group_start = group->key.objectid;
7554 new_extents = kmalloc(sizeof(*new_extents),
7557 ret = get_new_locations(reloc_inode,
7565 ret = replace_one_extent(trans, found_root,
7567 &first_key, ref_path,
7568 new_extents, nr_extents);
7570 ret = relocate_tree_block(trans, found_root, path,
7571 &first_key, ref_path);
7578 btrfs_end_transaction(trans, extent_root);
7585 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7588 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7589 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7591 num_devices = root->fs_info->fs_devices->rw_devices;
7592 if (num_devices == 1) {
7593 stripped |= BTRFS_BLOCK_GROUP_DUP;
7594 stripped = flags & ~stripped;
7596 /* turn raid0 into single device chunks */
7597 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7600 /* turn mirroring into duplication */
7601 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7602 BTRFS_BLOCK_GROUP_RAID10))
7603 return stripped | BTRFS_BLOCK_GROUP_DUP;
7606 /* they already had raid on here, just return */
7607 if (flags & stripped)
7610 stripped |= BTRFS_BLOCK_GROUP_DUP;
7611 stripped = flags & ~stripped;
7613 /* switch duplicated blocks with raid1 */
7614 if (flags & BTRFS_BLOCK_GROUP_DUP)
7615 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7617 /* turn single device chunks into raid0 */
7618 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7623 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7625 struct btrfs_space_info *sinfo = cache->space_info;
7632 spin_lock(&sinfo->lock);
7633 spin_lock(&cache->lock);
7634 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7635 cache->bytes_super - btrfs_block_group_used(&cache->item);
7637 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7638 sinfo->bytes_may_use + sinfo->bytes_readonly +
7639 cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7640 sinfo->bytes_readonly += num_bytes;
7641 sinfo->bytes_reserved += cache->reserved_pinned;
7642 cache->reserved_pinned = 0;
7646 spin_unlock(&cache->lock);
7647 spin_unlock(&sinfo->lock);
7651 int btrfs_set_block_group_ro(struct btrfs_root *root,
7652 struct btrfs_block_group_cache *cache)
7655 struct btrfs_trans_handle *trans;
7661 trans = btrfs_join_transaction(root, 1);
7662 BUG_ON(IS_ERR(trans));
7664 alloc_flags = update_block_group_flags(root, cache->flags);
7665 if (alloc_flags != cache->flags)
7666 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7668 ret = set_block_group_ro(cache);
7671 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7672 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7675 ret = set_block_group_ro(cache);
7677 btrfs_end_transaction(trans, root);
7681 int btrfs_set_block_group_rw(struct btrfs_root *root,
7682 struct btrfs_block_group_cache *cache)
7684 struct btrfs_space_info *sinfo = cache->space_info;
7689 spin_lock(&sinfo->lock);
7690 spin_lock(&cache->lock);
7691 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7692 cache->bytes_super - btrfs_block_group_used(&cache->item);
7693 sinfo->bytes_readonly -= num_bytes;
7695 spin_unlock(&cache->lock);
7696 spin_unlock(&sinfo->lock);
7701 * checks to see if its even possible to relocate this block group.
7703 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7704 * ok to go ahead and try.
7706 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7708 struct btrfs_block_group_cache *block_group;
7709 struct btrfs_space_info *space_info;
7710 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7711 struct btrfs_device *device;
7715 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7717 /* odd, couldn't find the block group, leave it alone */
7721 /* no bytes used, we're good */
7722 if (!btrfs_block_group_used(&block_group->item))
7725 space_info = block_group->space_info;
7726 spin_lock(&space_info->lock);
7728 full = space_info->full;
7731 * if this is the last block group we have in this space, we can't
7732 * relocate it unless we're able to allocate a new chunk below.
7734 * Otherwise, we need to make sure we have room in the space to handle
7735 * all of the extents from this block group. If we can, we're good
7737 if ((space_info->total_bytes != block_group->key.offset) &&
7738 (space_info->bytes_used + space_info->bytes_reserved +
7739 space_info->bytes_pinned + space_info->bytes_readonly +
7740 btrfs_block_group_used(&block_group->item) <
7741 space_info->total_bytes)) {
7742 spin_unlock(&space_info->lock);
7745 spin_unlock(&space_info->lock);
7748 * ok we don't have enough space, but maybe we have free space on our
7749 * devices to allocate new chunks for relocation, so loop through our
7750 * alloc devices and guess if we have enough space. However, if we
7751 * were marked as full, then we know there aren't enough chunks, and we
7758 mutex_lock(&root->fs_info->chunk_mutex);
7759 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7760 u64 min_free = btrfs_block_group_used(&block_group->item);
7761 u64 dev_offset, max_avail;
7764 * check to make sure we can actually find a chunk with enough
7765 * space to fit our block group in.
7767 if (device->total_bytes > device->bytes_used + min_free) {
7768 ret = find_free_dev_extent(NULL, device, min_free,
7769 &dev_offset, &max_avail);
7775 mutex_unlock(&root->fs_info->chunk_mutex);
7777 btrfs_put_block_group(block_group);
7781 static int find_first_block_group(struct btrfs_root *root,
7782 struct btrfs_path *path, struct btrfs_key *key)
7785 struct btrfs_key found_key;
7786 struct extent_buffer *leaf;
7789 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7794 slot = path->slots[0];
7795 leaf = path->nodes[0];
7796 if (slot >= btrfs_header_nritems(leaf)) {
7797 ret = btrfs_next_leaf(root, path);
7804 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7806 if (found_key.objectid >= key->objectid &&
7807 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7817 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7819 struct btrfs_block_group_cache *block_group;
7820 struct btrfs_space_info *space_info;
7821 struct btrfs_caching_control *caching_ctl;
7824 down_write(&info->extent_commit_sem);
7825 while (!list_empty(&info->caching_block_groups)) {
7826 caching_ctl = list_entry(info->caching_block_groups.next,
7827 struct btrfs_caching_control, list);
7828 list_del(&caching_ctl->list);
7829 put_caching_control(caching_ctl);
7831 up_write(&info->extent_commit_sem);
7833 spin_lock(&info->block_group_cache_lock);
7834 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7835 block_group = rb_entry(n, struct btrfs_block_group_cache,
7837 rb_erase(&block_group->cache_node,
7838 &info->block_group_cache_tree);
7839 spin_unlock(&info->block_group_cache_lock);
7841 down_write(&block_group->space_info->groups_sem);
7842 list_del(&block_group->list);
7843 up_write(&block_group->space_info->groups_sem);
7845 if (block_group->cached == BTRFS_CACHE_STARTED)
7846 wait_block_group_cache_done(block_group);
7848 btrfs_remove_free_space_cache(block_group);
7849 btrfs_put_block_group(block_group);
7851 spin_lock(&info->block_group_cache_lock);
7853 spin_unlock(&info->block_group_cache_lock);
7855 /* now that all the block groups are freed, go through and
7856 * free all the space_info structs. This is only called during
7857 * the final stages of unmount, and so we know nobody is
7858 * using them. We call synchronize_rcu() once before we start,
7859 * just to be on the safe side.
7863 release_global_block_rsv(info);
7865 while(!list_empty(&info->space_info)) {
7866 space_info = list_entry(info->space_info.next,
7867 struct btrfs_space_info,
7869 if (space_info->bytes_pinned > 0 ||
7870 space_info->bytes_reserved > 0) {
7872 dump_space_info(space_info, 0, 0);
7874 list_del(&space_info->list);
7880 static void __link_block_group(struct btrfs_space_info *space_info,
7881 struct btrfs_block_group_cache *cache)
7883 int index = get_block_group_index(cache);
7885 down_write(&space_info->groups_sem);
7886 list_add_tail(&cache->list, &space_info->block_groups[index]);
7887 up_write(&space_info->groups_sem);
7890 int btrfs_read_block_groups(struct btrfs_root *root)
7892 struct btrfs_path *path;
7894 struct btrfs_block_group_cache *cache;
7895 struct btrfs_fs_info *info = root->fs_info;
7896 struct btrfs_space_info *space_info;
7897 struct btrfs_key key;
7898 struct btrfs_key found_key;
7899 struct extent_buffer *leaf;
7901 root = info->extent_root;
7904 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7905 path = btrfs_alloc_path();
7910 ret = find_first_block_group(root, path, &key);
7916 leaf = path->nodes[0];
7917 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7918 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7924 atomic_set(&cache->count, 1);
7925 spin_lock_init(&cache->lock);
7926 spin_lock_init(&cache->tree_lock);
7927 cache->fs_info = info;
7928 INIT_LIST_HEAD(&cache->list);
7929 INIT_LIST_HEAD(&cache->cluster_list);
7932 * we only want to have 32k of ram per block group for keeping
7933 * track of free space, and if we pass 1/2 of that we want to
7934 * start converting things over to using bitmaps
7936 cache->extents_thresh = ((1024 * 32) / 2) /
7937 sizeof(struct btrfs_free_space);
7939 read_extent_buffer(leaf, &cache->item,
7940 btrfs_item_ptr_offset(leaf, path->slots[0]),
7941 sizeof(cache->item));
7942 memcpy(&cache->key, &found_key, sizeof(found_key));
7944 key.objectid = found_key.objectid + found_key.offset;
7945 btrfs_release_path(root, path);
7946 cache->flags = btrfs_block_group_flags(&cache->item);
7947 cache->sectorsize = root->sectorsize;
7950 * check for two cases, either we are full, and therefore
7951 * don't need to bother with the caching work since we won't
7952 * find any space, or we are empty, and we can just add all
7953 * the space in and be done with it. This saves us _alot_ of
7954 * time, particularly in the full case.
7956 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7957 exclude_super_stripes(root, cache);
7958 cache->last_byte_to_unpin = (u64)-1;
7959 cache->cached = BTRFS_CACHE_FINISHED;
7960 free_excluded_extents(root, cache);
7961 } else if (btrfs_block_group_used(&cache->item) == 0) {
7962 exclude_super_stripes(root, cache);
7963 cache->last_byte_to_unpin = (u64)-1;
7964 cache->cached = BTRFS_CACHE_FINISHED;
7965 add_new_free_space(cache, root->fs_info,
7967 found_key.objectid +
7969 free_excluded_extents(root, cache);
7972 ret = update_space_info(info, cache->flags, found_key.offset,
7973 btrfs_block_group_used(&cache->item),
7976 cache->space_info = space_info;
7977 spin_lock(&cache->space_info->lock);
7978 cache->space_info->bytes_readonly += cache->bytes_super;
7979 spin_unlock(&cache->space_info->lock);
7981 __link_block_group(space_info, cache);
7983 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7986 set_avail_alloc_bits(root->fs_info, cache->flags);
7987 if (btrfs_chunk_readonly(root, cache->key.objectid))
7988 set_block_group_ro(cache);
7991 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7992 if (!(get_alloc_profile(root, space_info->flags) &
7993 (BTRFS_BLOCK_GROUP_RAID10 |
7994 BTRFS_BLOCK_GROUP_RAID1 |
7995 BTRFS_BLOCK_GROUP_DUP)))
7998 * avoid allocating from un-mirrored block group if there are
7999 * mirrored block groups.
8001 list_for_each_entry(cache, &space_info->block_groups[3], list)
8002 set_block_group_ro(cache);
8003 list_for_each_entry(cache, &space_info->block_groups[4], list)
8004 set_block_group_ro(cache);
8007 init_global_block_rsv(info);
8010 btrfs_free_path(path);
8014 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8015 struct btrfs_root *root, u64 bytes_used,
8016 u64 type, u64 chunk_objectid, u64 chunk_offset,
8020 struct btrfs_root *extent_root;
8021 struct btrfs_block_group_cache *cache;
8023 extent_root = root->fs_info->extent_root;
8025 root->fs_info->last_trans_log_full_commit = trans->transid;
8027 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8031 cache->key.objectid = chunk_offset;
8032 cache->key.offset = size;
8033 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8034 cache->sectorsize = root->sectorsize;
8037 * we only want to have 32k of ram per block group for keeping track
8038 * of free space, and if we pass 1/2 of that we want to start
8039 * converting things over to using bitmaps
8041 cache->extents_thresh = ((1024 * 32) / 2) /
8042 sizeof(struct btrfs_free_space);
8043 atomic_set(&cache->count, 1);
8044 spin_lock_init(&cache->lock);
8045 spin_lock_init(&cache->tree_lock);
8046 INIT_LIST_HEAD(&cache->list);
8047 INIT_LIST_HEAD(&cache->cluster_list);
8049 btrfs_set_block_group_used(&cache->item, bytes_used);
8050 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8051 cache->flags = type;
8052 btrfs_set_block_group_flags(&cache->item, type);
8054 cache->last_byte_to_unpin = (u64)-1;
8055 cache->cached = BTRFS_CACHE_FINISHED;
8056 exclude_super_stripes(root, cache);
8058 add_new_free_space(cache, root->fs_info, chunk_offset,
8059 chunk_offset + size);
8061 free_excluded_extents(root, cache);
8063 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8064 &cache->space_info);
8067 spin_lock(&cache->space_info->lock);
8068 cache->space_info->bytes_readonly += cache->bytes_super;
8069 spin_unlock(&cache->space_info->lock);
8071 __link_block_group(cache->space_info, cache);
8073 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8076 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8077 sizeof(cache->item));
8080 set_avail_alloc_bits(extent_root->fs_info, type);
8085 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8086 struct btrfs_root *root, u64 group_start)
8088 struct btrfs_path *path;
8089 struct btrfs_block_group_cache *block_group;
8090 struct btrfs_free_cluster *cluster;
8091 struct btrfs_key key;
8094 root = root->fs_info->extent_root;
8096 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8097 BUG_ON(!block_group);
8098 BUG_ON(!block_group->ro);
8100 memcpy(&key, &block_group->key, sizeof(key));
8102 /* make sure this block group isn't part of an allocation cluster */
8103 cluster = &root->fs_info->data_alloc_cluster;
8104 spin_lock(&cluster->refill_lock);
8105 btrfs_return_cluster_to_free_space(block_group, cluster);
8106 spin_unlock(&cluster->refill_lock);
8109 * make sure this block group isn't part of a metadata
8110 * allocation cluster
8112 cluster = &root->fs_info->meta_alloc_cluster;
8113 spin_lock(&cluster->refill_lock);
8114 btrfs_return_cluster_to_free_space(block_group, cluster);
8115 spin_unlock(&cluster->refill_lock);
8117 path = btrfs_alloc_path();
8120 spin_lock(&root->fs_info->block_group_cache_lock);
8121 rb_erase(&block_group->cache_node,
8122 &root->fs_info->block_group_cache_tree);
8123 spin_unlock(&root->fs_info->block_group_cache_lock);
8125 down_write(&block_group->space_info->groups_sem);
8127 * we must use list_del_init so people can check to see if they
8128 * are still on the list after taking the semaphore
8130 list_del_init(&block_group->list);
8131 up_write(&block_group->space_info->groups_sem);
8133 if (block_group->cached == BTRFS_CACHE_STARTED)
8134 wait_block_group_cache_done(block_group);
8136 btrfs_remove_free_space_cache(block_group);
8138 spin_lock(&block_group->space_info->lock);
8139 block_group->space_info->total_bytes -= block_group->key.offset;
8140 block_group->space_info->bytes_readonly -= block_group->key.offset;
8141 spin_unlock(&block_group->space_info->lock);
8143 btrfs_clear_space_info_full(root->fs_info);
8145 btrfs_put_block_group(block_group);
8146 btrfs_put_block_group(block_group);
8148 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8154 ret = btrfs_del_item(trans, root, path);
8156 btrfs_free_path(path);