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>
28 #include "print-tree.h"
29 #include "transaction.h"
32 #include "ref-cache.h"
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
38 struct pending_extent_op {
47 struct list_head list;
51 static int finish_current_insert(struct btrfs_trans_handle *trans,
52 struct btrfs_root *extent_root, int all);
53 static int del_pending_extents(struct btrfs_trans_handle *trans,
54 struct btrfs_root *extent_root, int all);
55 static int pin_down_bytes(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int is_data);
58 static int update_block_group(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, int alloc,
63 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
65 return (cache->flags & bits) == bits;
69 * this adds the block group to the fs_info rb tree for the block group
72 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
73 struct btrfs_block_group_cache *block_group)
76 struct rb_node *parent = NULL;
77 struct btrfs_block_group_cache *cache;
79 spin_lock(&info->block_group_cache_lock);
80 p = &info->block_group_cache_tree.rb_node;
84 cache = rb_entry(parent, struct btrfs_block_group_cache,
86 if (block_group->key.objectid < cache->key.objectid) {
88 } else if (block_group->key.objectid > cache->key.objectid) {
91 spin_unlock(&info->block_group_cache_lock);
96 rb_link_node(&block_group->cache_node, parent, p);
97 rb_insert_color(&block_group->cache_node,
98 &info->block_group_cache_tree);
99 spin_unlock(&info->block_group_cache_lock);
105 * This will return the block group at or after bytenr if contains is 0, else
106 * it will return the block group that contains the bytenr
108 static struct btrfs_block_group_cache *
109 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
112 struct btrfs_block_group_cache *cache, *ret = NULL;
116 spin_lock(&info->block_group_cache_lock);
117 n = info->block_group_cache_tree.rb_node;
120 cache = rb_entry(n, struct btrfs_block_group_cache,
122 end = cache->key.objectid + cache->key.offset - 1;
123 start = cache->key.objectid;
125 if (bytenr < start) {
126 if (!contains && (!ret || start < ret->key.objectid))
129 } else if (bytenr > start) {
130 if (contains && bytenr <= end) {
141 atomic_inc(&ret->count);
142 spin_unlock(&info->block_group_cache_lock);
148 * this is only called by cache_block_group, since we could have freed extents
149 * we need to check the pinned_extents for any extents that can't be used yet
150 * since their free space will be released as soon as the transaction commits.
152 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
153 struct btrfs_fs_info *info, u64 start, u64 end)
155 u64 extent_start, extent_end, size;
158 mutex_lock(&info->pinned_mutex);
159 while (start < end) {
160 ret = find_first_extent_bit(&info->pinned_extents, start,
161 &extent_start, &extent_end,
166 if (extent_start == start) {
167 start = extent_end + 1;
168 } else if (extent_start > start && extent_start < end) {
169 size = extent_start - start;
170 ret = btrfs_add_free_space(block_group, start,
173 start = extent_end + 1;
181 ret = btrfs_add_free_space(block_group, start, size);
184 mutex_unlock(&info->pinned_mutex);
189 static int remove_sb_from_cache(struct btrfs_root *root,
190 struct btrfs_block_group_cache *cache)
197 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
198 bytenr = btrfs_sb_offset(i);
199 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
200 cache->key.objectid, bytenr, 0,
201 &logical, &nr, &stripe_len);
204 btrfs_remove_free_space(cache, logical[nr],
212 static int cache_block_group(struct btrfs_root *root,
213 struct btrfs_block_group_cache *block_group)
215 struct btrfs_path *path;
217 struct btrfs_key key;
218 struct extent_buffer *leaf;
225 root = root->fs_info->extent_root;
227 if (block_group->cached)
230 path = btrfs_alloc_path();
236 * we get into deadlocks with paths held by callers of this function.
237 * since the alloc_mutex is protecting things right now, just
238 * skip the locking here
240 path->skip_locking = 1;
241 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
244 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
245 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
250 leaf = path->nodes[0];
251 slot = path->slots[0];
252 if (slot >= btrfs_header_nritems(leaf)) {
253 ret = btrfs_next_leaf(root, path);
261 btrfs_item_key_to_cpu(leaf, &key, slot);
262 if (key.objectid < block_group->key.objectid)
265 if (key.objectid >= block_group->key.objectid +
266 block_group->key.offset)
269 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
270 add_new_free_space(block_group, root->fs_info, last,
273 last = key.objectid + key.offset;
279 add_new_free_space(block_group, root->fs_info, last,
280 block_group->key.objectid +
281 block_group->key.offset);
283 remove_sb_from_cache(root, block_group);
284 block_group->cached = 1;
287 btrfs_free_path(path);
292 * return the block group that starts at or after bytenr
294 static struct btrfs_block_group_cache *
295 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
297 struct btrfs_block_group_cache *cache;
299 cache = block_group_cache_tree_search(info, bytenr, 0);
305 * return the block group that contains teh given bytenr
307 struct btrfs_block_group_cache *btrfs_lookup_block_group(
308 struct btrfs_fs_info *info,
311 struct btrfs_block_group_cache *cache;
313 cache = block_group_cache_tree_search(info, bytenr, 1);
318 static inline void put_block_group(struct btrfs_block_group_cache *cache)
320 if (atomic_dec_and_test(&cache->count))
324 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
327 struct list_head *head = &info->space_info;
328 struct btrfs_space_info *found;
329 list_for_each_entry(found, head, list) {
330 if (found->flags == flags)
336 static u64 div_factor(u64 num, int factor)
345 u64 btrfs_find_block_group(struct btrfs_root *root,
346 u64 search_start, u64 search_hint, int owner)
348 struct btrfs_block_group_cache *cache;
350 u64 last = max(search_hint, search_start);
357 cache = btrfs_lookup_first_block_group(root->fs_info, last);
361 spin_lock(&cache->lock);
362 last = cache->key.objectid + cache->key.offset;
363 used = btrfs_block_group_used(&cache->item);
365 if ((full_search || !cache->ro) &&
366 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
367 if (used + cache->pinned + cache->reserved <
368 div_factor(cache->key.offset, factor)) {
369 group_start = cache->key.objectid;
370 spin_unlock(&cache->lock);
371 put_block_group(cache);
375 spin_unlock(&cache->lock);
376 put_block_group(cache);
384 if (!full_search && factor < 10) {
394 /* simple helper to search for an existing extent at a given offset */
395 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
398 struct btrfs_key key;
399 struct btrfs_path *path;
401 path = btrfs_alloc_path();
403 key.objectid = start;
405 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
406 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
408 btrfs_free_path(path);
413 * Back reference rules. Back refs have three main goals:
415 * 1) differentiate between all holders of references to an extent so that
416 * when a reference is dropped we can make sure it was a valid reference
417 * before freeing the extent.
419 * 2) Provide enough information to quickly find the holders of an extent
420 * if we notice a given block is corrupted or bad.
422 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
423 * maintenance. This is actually the same as #2, but with a slightly
424 * different use case.
426 * File extents can be referenced by:
428 * - multiple snapshots, subvolumes, or different generations in one subvol
429 * - different files inside a single subvolume
430 * - different offsets inside a file (bookend extents in file.c)
432 * The extent ref structure has fields for:
434 * - Objectid of the subvolume root
435 * - Generation number of the tree holding the reference
436 * - objectid of the file holding the reference
437 * - number of references holding by parent node (alway 1 for tree blocks)
439 * Btree leaf may hold multiple references to a file extent. In most cases,
440 * these references are from same file and the corresponding offsets inside
441 * the file are close together.
443 * When a file extent is allocated the fields are filled in:
444 * (root_key.objectid, trans->transid, inode objectid, 1)
446 * When a leaf is cow'd new references are added for every file extent found
447 * in the leaf. It looks similar to the create case, but trans->transid will
448 * be different when the block is cow'd.
450 * (root_key.objectid, trans->transid, inode objectid,
451 * number of references in the leaf)
453 * When a file extent is removed either during snapshot deletion or
454 * file truncation, we find the corresponding back reference and check
455 * the following fields:
457 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
460 * Btree extents can be referenced by:
462 * - Different subvolumes
463 * - Different generations of the same subvolume
465 * When a tree block is created, back references are inserted:
467 * (root->root_key.objectid, trans->transid, level, 1)
469 * When a tree block is cow'd, new back references are added for all the
470 * blocks it points to. If the tree block isn't in reference counted root,
471 * the old back references are removed. These new back references are of
472 * the form (trans->transid will have increased since creation):
474 * (root->root_key.objectid, trans->transid, level, 1)
476 * When a backref is in deleting, the following fields are checked:
478 * if backref was for a tree root:
479 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
481 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
483 * Back Reference Key composing:
485 * The key objectid corresponds to the first byte in the extent, the key
486 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
487 * byte of parent extent. If a extent is tree root, the key offset is set
488 * to the key objectid.
491 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
492 struct btrfs_root *root,
493 struct btrfs_path *path,
494 u64 bytenr, u64 parent,
495 u64 ref_root, u64 ref_generation,
496 u64 owner_objectid, int del)
498 struct btrfs_key key;
499 struct btrfs_extent_ref *ref;
500 struct extent_buffer *leaf;
504 key.objectid = bytenr;
505 key.type = BTRFS_EXTENT_REF_KEY;
508 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
516 leaf = path->nodes[0];
517 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
518 ref_objectid = btrfs_ref_objectid(leaf, ref);
519 if (btrfs_ref_root(leaf, ref) != ref_root ||
520 btrfs_ref_generation(leaf, ref) != ref_generation ||
521 (ref_objectid != owner_objectid &&
522 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
533 * updates all the backrefs that are pending on update_list for the
536 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
537 struct btrfs_root *extent_root,
538 struct btrfs_path *path,
539 struct list_head *update_list)
541 struct btrfs_key key;
542 struct btrfs_extent_ref *ref;
543 struct btrfs_fs_info *info = extent_root->fs_info;
544 struct pending_extent_op *op;
545 struct extent_buffer *leaf;
547 struct list_head *cur = update_list->next;
549 u64 ref_root = extent_root->root_key.objectid;
551 op = list_entry(cur, struct pending_extent_op, list);
554 key.objectid = op->bytenr;
555 key.type = BTRFS_EXTENT_REF_KEY;
556 key.offset = op->orig_parent;
558 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
561 leaf = path->nodes[0];
564 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
566 ref_objectid = btrfs_ref_objectid(leaf, ref);
568 if (btrfs_ref_root(leaf, ref) != ref_root ||
569 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
570 (ref_objectid != op->level &&
571 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
572 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
573 "root %llu, owner %u\n",
574 (unsigned long long)op->bytenr,
575 (unsigned long long)op->orig_parent,
576 (unsigned long long)ref_root, op->level);
577 btrfs_print_leaf(extent_root, leaf);
581 key.objectid = op->bytenr;
582 key.offset = op->parent;
583 key.type = BTRFS_EXTENT_REF_KEY;
584 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
586 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
587 btrfs_set_ref_generation(leaf, ref, op->generation);
591 list_del_init(&op->list);
592 unlock_extent(&info->extent_ins, op->bytenr,
593 op->bytenr + op->num_bytes - 1, GFP_NOFS);
596 if (cur == update_list) {
597 btrfs_mark_buffer_dirty(path->nodes[0]);
598 btrfs_release_path(extent_root, path);
602 op = list_entry(cur, struct pending_extent_op, list);
605 while (path->slots[0] < btrfs_header_nritems(leaf)) {
606 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
607 if (key.objectid == op->bytenr &&
608 key.type == BTRFS_EXTENT_REF_KEY)
613 btrfs_mark_buffer_dirty(path->nodes[0]);
614 btrfs_release_path(extent_root, path);
621 static noinline int insert_extents(struct btrfs_trans_handle *trans,
622 struct btrfs_root *extent_root,
623 struct btrfs_path *path,
624 struct list_head *insert_list, int nr)
626 struct btrfs_key *keys;
628 struct pending_extent_op *op;
629 struct extent_buffer *leaf;
630 struct list_head *cur = insert_list->next;
631 struct btrfs_fs_info *info = extent_root->fs_info;
632 u64 ref_root = extent_root->root_key.objectid;
633 int i = 0, last = 0, ret;
639 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
643 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
649 list_for_each_entry(op, insert_list, list) {
650 keys[i].objectid = op->bytenr;
651 keys[i].offset = op->num_bytes;
652 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
653 data_size[i] = sizeof(struct btrfs_extent_item);
656 keys[i].objectid = op->bytenr;
657 keys[i].offset = op->parent;
658 keys[i].type = BTRFS_EXTENT_REF_KEY;
659 data_size[i] = sizeof(struct btrfs_extent_ref);
663 op = list_entry(cur, struct pending_extent_op, list);
667 ret = btrfs_insert_some_items(trans, extent_root, path,
668 keys+i, data_size+i, total-i);
674 leaf = path->nodes[0];
675 for (c = 0; c < ret; c++) {
676 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
679 * if the first item we inserted was a backref, then
680 * the EXTENT_ITEM will be the odd c's, else it will
683 if ((ref_first && (c % 2)) ||
684 (!ref_first && !(c % 2))) {
685 struct btrfs_extent_item *itm;
687 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
688 struct btrfs_extent_item);
689 btrfs_set_extent_refs(path->nodes[0], itm, 1);
692 struct btrfs_extent_ref *ref;
694 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
695 struct btrfs_extent_ref);
696 btrfs_set_ref_root(leaf, ref, ref_root);
697 btrfs_set_ref_generation(leaf, ref,
699 btrfs_set_ref_objectid(leaf, ref, op->level);
700 btrfs_set_ref_num_refs(leaf, ref, 1);
705 * using del to see when its ok to free up the
706 * pending_extent_op. In the case where we insert the
707 * last item on the list in order to help do batching
708 * we need to not free the extent op until we actually
709 * insert the extent_item
712 unlock_extent(&info->extent_ins, op->bytenr,
713 op->bytenr + op->num_bytes - 1,
716 list_del_init(&op->list);
718 if (cur != insert_list)
720 struct pending_extent_op,
724 btrfs_mark_buffer_dirty(leaf);
725 btrfs_release_path(extent_root, path);
728 * Ok backref's and items usually go right next to eachother,
729 * but if we could only insert 1 item that means that we
730 * inserted on the end of a leaf, and we have no idea what may
731 * be on the next leaf so we just play it safe. In order to
732 * try and help this case we insert the last thing on our
733 * insert list so hopefully it will end up being the last
734 * thing on the leaf and everything else will be before it,
735 * which will let us insert a whole bunch of items at the same
738 if (ret == 1 && !last && (i + ret < total)) {
740 * last: where we will pick up the next time around
741 * i: our current key to insert, will be total - 1
742 * cur: the current op we are screwing with
747 cur = insert_list->prev;
748 op = list_entry(cur, struct pending_extent_op, list);
751 * ok we successfully inserted the last item on the
752 * list, lets reset everything
754 * i: our current key to insert, so where we left off
756 * last: done with this
757 * cur: the op we are messing with
759 * total: since we inserted the last key, we need to
760 * decrement total so we dont overflow
766 cur = insert_list->next;
767 op = list_entry(cur, struct pending_extent_op,
782 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
783 struct btrfs_root *root,
784 struct btrfs_path *path,
785 u64 bytenr, u64 parent,
786 u64 ref_root, u64 ref_generation,
789 struct btrfs_key key;
790 struct extent_buffer *leaf;
791 struct btrfs_extent_ref *ref;
795 key.objectid = bytenr;
796 key.type = BTRFS_EXTENT_REF_KEY;
799 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
801 leaf = path->nodes[0];
802 ref = btrfs_item_ptr(leaf, path->slots[0],
803 struct btrfs_extent_ref);
804 btrfs_set_ref_root(leaf, ref, ref_root);
805 btrfs_set_ref_generation(leaf, ref, ref_generation);
806 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
807 btrfs_set_ref_num_refs(leaf, ref, 1);
808 } else if (ret == -EEXIST) {
810 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
811 leaf = path->nodes[0];
812 ref = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_ref);
814 if (btrfs_ref_root(leaf, ref) != ref_root ||
815 btrfs_ref_generation(leaf, ref) != ref_generation) {
821 num_refs = btrfs_ref_num_refs(leaf, ref);
822 BUG_ON(num_refs == 0);
823 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
825 existing_owner = btrfs_ref_objectid(leaf, ref);
826 if (existing_owner != owner_objectid &&
827 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
828 btrfs_set_ref_objectid(leaf, ref,
829 BTRFS_MULTIPLE_OBJECTIDS);
835 btrfs_mark_buffer_dirty(path->nodes[0]);
837 btrfs_release_path(root, path);
841 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
842 struct btrfs_root *root,
843 struct btrfs_path *path)
845 struct extent_buffer *leaf;
846 struct btrfs_extent_ref *ref;
850 leaf = path->nodes[0];
851 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
852 num_refs = btrfs_ref_num_refs(leaf, ref);
853 BUG_ON(num_refs == 0);
856 ret = btrfs_del_item(trans, root, path);
858 btrfs_set_ref_num_refs(leaf, ref, num_refs);
859 btrfs_mark_buffer_dirty(leaf);
861 btrfs_release_path(root, path);
865 #ifdef BIO_RW_DISCARD
866 static void btrfs_issue_discard(struct block_device *bdev,
869 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
873 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
876 #ifdef BIO_RW_DISCARD
878 u64 map_length = num_bytes;
879 struct btrfs_multi_bio *multi = NULL;
881 /* Tell the block device(s) that the sectors can be discarded */
882 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
883 bytenr, &map_length, &multi, 0);
885 struct btrfs_bio_stripe *stripe = multi->stripes;
888 if (map_length > num_bytes)
889 map_length = num_bytes;
891 for (i = 0; i < multi->num_stripes; i++, stripe++) {
892 btrfs_issue_discard(stripe->dev->bdev,
905 static noinline int free_extents(struct btrfs_trans_handle *trans,
906 struct btrfs_root *extent_root,
907 struct list_head *del_list)
909 struct btrfs_fs_info *info = extent_root->fs_info;
910 struct btrfs_path *path;
911 struct btrfs_key key, found_key;
912 struct extent_buffer *leaf;
913 struct list_head *cur;
914 struct pending_extent_op *op;
915 struct btrfs_extent_item *ei;
916 int ret, num_to_del, extent_slot = 0, found_extent = 0;
920 path = btrfs_alloc_path();
926 /* search for the backref for the current ref we want to delete */
927 cur = del_list->next;
928 op = list_entry(cur, struct pending_extent_op, list);
929 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
931 extent_root->root_key.objectid,
932 op->orig_generation, op->level, 1);
934 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
935 "root %llu gen %llu owner %u\n",
936 (unsigned long long)op->bytenr,
937 (unsigned long long)extent_root->root_key.objectid,
938 (unsigned long long)op->orig_generation, op->level);
939 btrfs_print_leaf(extent_root, path->nodes[0]);
944 extent_slot = path->slots[0];
949 * if we aren't the first item on the leaf we can move back one and see
950 * if our ref is right next to our extent item
952 if (likely(extent_slot)) {
954 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
956 if (found_key.objectid == op->bytenr &&
957 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
958 found_key.offset == op->num_bytes) {
965 * if we didn't find the extent we need to delete the backref and then
966 * search for the extent item key so we can update its ref count
969 key.objectid = op->bytenr;
970 key.type = BTRFS_EXTENT_ITEM_KEY;
971 key.offset = op->num_bytes;
973 ret = remove_extent_backref(trans, extent_root, path);
975 btrfs_release_path(extent_root, path);
976 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
978 extent_slot = path->slots[0];
981 /* this is where we update the ref count for the extent */
982 leaf = path->nodes[0];
983 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
984 refs = btrfs_extent_refs(leaf, ei);
987 btrfs_set_extent_refs(leaf, ei, refs);
989 btrfs_mark_buffer_dirty(leaf);
992 * This extent needs deleting. The reason cur_slot is extent_slot +
993 * num_to_del is because extent_slot points to the slot where the extent
994 * is, and if the backref was not right next to the extent we will be
995 * deleting at least 1 item, and will want to start searching at the
996 * slot directly next to extent_slot. However if we did find the
997 * backref next to the extent item them we will be deleting at least 2
998 * items and will want to start searching directly after the ref slot
1001 struct list_head *pos, *n, *end;
1002 int cur_slot = extent_slot+num_to_del;
1006 path->slots[0] = extent_slot;
1007 bytes_freed = op->num_bytes;
1009 mutex_lock(&info->pinned_mutex);
1010 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1011 op->num_bytes, op->level >=
1012 BTRFS_FIRST_FREE_OBJECTID);
1013 mutex_unlock(&info->pinned_mutex);
1018 * we need to see if we can delete multiple things at once, so
1019 * start looping through the list of extents we are wanting to
1020 * delete and see if their extent/backref's are right next to
1021 * eachother and the extents only have 1 ref
1023 for (pos = cur->next; pos != del_list; pos = pos->next) {
1024 struct pending_extent_op *tmp;
1026 tmp = list_entry(pos, struct pending_extent_op, list);
1028 /* we only want to delete extent+ref at this stage */
1029 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1032 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1033 if (found_key.objectid != tmp->bytenr ||
1034 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1035 found_key.offset != tmp->num_bytes)
1038 /* check to make sure this extent only has one ref */
1039 ei = btrfs_item_ptr(leaf, cur_slot,
1040 struct btrfs_extent_item);
1041 if (btrfs_extent_refs(leaf, ei) != 1)
1044 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1045 if (found_key.objectid != tmp->bytenr ||
1046 found_key.type != BTRFS_EXTENT_REF_KEY ||
1047 found_key.offset != tmp->orig_parent)
1051 * the ref is right next to the extent, we can set the
1052 * ref count to 0 since we will delete them both now
1054 btrfs_set_extent_refs(leaf, ei, 0);
1056 /* pin down the bytes for this extent */
1057 mutex_lock(&info->pinned_mutex);
1058 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1059 tmp->num_bytes, tmp->level >=
1060 BTRFS_FIRST_FREE_OBJECTID);
1061 mutex_unlock(&info->pinned_mutex);
1065 * use the del field to tell if we need to go ahead and
1066 * free up the extent when we delete the item or not.
1069 bytes_freed += tmp->num_bytes;
1076 /* update the free space counters */
1077 spin_lock(&info->delalloc_lock);
1078 super_used = btrfs_super_bytes_used(&info->super_copy);
1079 btrfs_set_super_bytes_used(&info->super_copy,
1080 super_used - bytes_freed);
1082 root_used = btrfs_root_used(&extent_root->root_item);
1083 btrfs_set_root_used(&extent_root->root_item,
1084 root_used - bytes_freed);
1085 spin_unlock(&info->delalloc_lock);
1087 /* delete the items */
1088 ret = btrfs_del_items(trans, extent_root, path,
1089 path->slots[0], num_to_del);
1093 * loop through the extents we deleted and do the cleanup work
1096 for (pos = cur, n = pos->next; pos != end;
1097 pos = n, n = pos->next) {
1098 struct pending_extent_op *tmp;
1099 tmp = list_entry(pos, struct pending_extent_op, list);
1102 * remember tmp->del tells us wether or not we pinned
1105 ret = update_block_group(trans, extent_root,
1106 tmp->bytenr, tmp->num_bytes, 0,
1110 list_del_init(&tmp->list);
1111 unlock_extent(&info->extent_ins, tmp->bytenr,
1112 tmp->bytenr + tmp->num_bytes - 1,
1116 } else if (refs && found_extent) {
1118 * the ref and extent were right next to eachother, but the
1119 * extent still has a ref, so just free the backref and keep
1122 ret = remove_extent_backref(trans, extent_root, path);
1125 list_del_init(&op->list);
1126 unlock_extent(&info->extent_ins, op->bytenr,
1127 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1131 * the extent has multiple refs and the backref we were looking
1132 * for was not right next to it, so just unlock and go next,
1135 list_del_init(&op->list);
1136 unlock_extent(&info->extent_ins, op->bytenr,
1137 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1141 btrfs_release_path(extent_root, path);
1142 if (!list_empty(del_list))
1146 btrfs_free_path(path);
1150 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1151 struct btrfs_root *root, u64 bytenr,
1152 u64 orig_parent, u64 parent,
1153 u64 orig_root, u64 ref_root,
1154 u64 orig_generation, u64 ref_generation,
1158 struct btrfs_root *extent_root = root->fs_info->extent_root;
1159 struct btrfs_path *path;
1161 if (root == root->fs_info->extent_root) {
1162 struct pending_extent_op *extent_op;
1165 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1166 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1167 mutex_lock(&root->fs_info->extent_ins_mutex);
1168 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1169 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1171 ret = get_state_private(&root->fs_info->extent_ins,
1174 extent_op = (struct pending_extent_op *)
1175 (unsigned long)priv;
1176 BUG_ON(extent_op->parent != orig_parent);
1177 BUG_ON(extent_op->generation != orig_generation);
1179 extent_op->parent = parent;
1180 extent_op->generation = ref_generation;
1182 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1185 extent_op->type = PENDING_BACKREF_UPDATE;
1186 extent_op->bytenr = bytenr;
1187 extent_op->num_bytes = num_bytes;
1188 extent_op->parent = parent;
1189 extent_op->orig_parent = orig_parent;
1190 extent_op->generation = ref_generation;
1191 extent_op->orig_generation = orig_generation;
1192 extent_op->level = (int)owner_objectid;
1193 INIT_LIST_HEAD(&extent_op->list);
1196 set_extent_bits(&root->fs_info->extent_ins,
1197 bytenr, bytenr + num_bytes - 1,
1198 EXTENT_WRITEBACK, GFP_NOFS);
1199 set_state_private(&root->fs_info->extent_ins,
1200 bytenr, (unsigned long)extent_op);
1202 mutex_unlock(&root->fs_info->extent_ins_mutex);
1206 path = btrfs_alloc_path();
1209 ret = lookup_extent_backref(trans, extent_root, path,
1210 bytenr, orig_parent, orig_root,
1211 orig_generation, owner_objectid, 1);
1214 ret = remove_extent_backref(trans, extent_root, path);
1217 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1218 parent, ref_root, ref_generation,
1221 finish_current_insert(trans, extent_root, 0);
1222 del_pending_extents(trans, extent_root, 0);
1224 btrfs_free_path(path);
1228 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1229 struct btrfs_root *root, u64 bytenr,
1230 u64 orig_parent, u64 parent,
1231 u64 ref_root, u64 ref_generation,
1235 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1236 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1238 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1239 parent, ref_root, ref_root,
1240 ref_generation, ref_generation,
1245 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1246 struct btrfs_root *root, u64 bytenr,
1247 u64 orig_parent, u64 parent,
1248 u64 orig_root, u64 ref_root,
1249 u64 orig_generation, u64 ref_generation,
1252 struct btrfs_path *path;
1254 struct btrfs_key key;
1255 struct extent_buffer *l;
1256 struct btrfs_extent_item *item;
1259 path = btrfs_alloc_path();
1264 key.objectid = bytenr;
1265 key.type = BTRFS_EXTENT_ITEM_KEY;
1266 key.offset = (u64)-1;
1268 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1272 BUG_ON(ret == 0 || path->slots[0] == 0);
1277 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1278 if (key.objectid != bytenr) {
1279 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1280 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1281 (unsigned long long)bytenr,
1282 (unsigned long long)key.objectid);
1285 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1287 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1288 refs = btrfs_extent_refs(l, item);
1289 btrfs_set_extent_refs(l, item, refs + 1);
1290 btrfs_mark_buffer_dirty(path->nodes[0]);
1292 btrfs_release_path(root->fs_info->extent_root, path);
1295 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1296 path, bytenr, parent,
1297 ref_root, ref_generation,
1300 finish_current_insert(trans, root->fs_info->extent_root, 0);
1301 del_pending_extents(trans, root->fs_info->extent_root, 0);
1303 btrfs_free_path(path);
1307 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1308 struct btrfs_root *root,
1309 u64 bytenr, u64 num_bytes, u64 parent,
1310 u64 ref_root, u64 ref_generation,
1314 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1315 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1317 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1318 0, ref_root, 0, ref_generation,
1323 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1324 struct btrfs_root *root)
1326 finish_current_insert(trans, root->fs_info->extent_root, 1);
1327 del_pending_extents(trans, root->fs_info->extent_root, 1);
1331 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1332 struct btrfs_root *root, u64 bytenr,
1333 u64 num_bytes, u32 *refs)
1335 struct btrfs_path *path;
1337 struct btrfs_key key;
1338 struct extent_buffer *l;
1339 struct btrfs_extent_item *item;
1341 WARN_ON(num_bytes < root->sectorsize);
1342 path = btrfs_alloc_path();
1344 key.objectid = bytenr;
1345 key.offset = num_bytes;
1346 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1347 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1352 btrfs_print_leaf(root, path->nodes[0]);
1353 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1354 (unsigned long long)bytenr);
1358 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1359 *refs = btrfs_extent_refs(l, item);
1361 btrfs_free_path(path);
1365 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1366 struct btrfs_root *root, u64 objectid, u64 bytenr)
1368 struct btrfs_root *extent_root = root->fs_info->extent_root;
1369 struct btrfs_path *path;
1370 struct extent_buffer *leaf;
1371 struct btrfs_extent_ref *ref_item;
1372 struct btrfs_key key;
1373 struct btrfs_key found_key;
1379 key.objectid = bytenr;
1380 key.offset = (u64)-1;
1381 key.type = BTRFS_EXTENT_ITEM_KEY;
1383 path = btrfs_alloc_path();
1384 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1390 if (path->slots[0] == 0)
1394 leaf = path->nodes[0];
1395 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1397 if (found_key.objectid != bytenr ||
1398 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1401 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1403 leaf = path->nodes[0];
1404 nritems = btrfs_header_nritems(leaf);
1405 if (path->slots[0] >= nritems) {
1406 ret = btrfs_next_leaf(extent_root, path);
1413 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1414 if (found_key.objectid != bytenr)
1417 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1422 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1423 struct btrfs_extent_ref);
1424 ref_root = btrfs_ref_root(leaf, ref_item);
1425 if ((ref_root != root->root_key.objectid &&
1426 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1427 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1431 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1440 btrfs_free_path(path);
1444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1445 struct extent_buffer *buf, u32 nr_extents)
1447 struct btrfs_key key;
1448 struct btrfs_file_extent_item *fi;
1456 if (!root->ref_cows)
1459 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1461 root_gen = root->root_key.offset;
1464 root_gen = trans->transid - 1;
1467 level = btrfs_header_level(buf);
1468 nritems = btrfs_header_nritems(buf);
1471 struct btrfs_leaf_ref *ref;
1472 struct btrfs_extent_info *info;
1474 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1480 ref->root_gen = root_gen;
1481 ref->bytenr = buf->start;
1482 ref->owner = btrfs_header_owner(buf);
1483 ref->generation = btrfs_header_generation(buf);
1484 ref->nritems = nr_extents;
1485 info = ref->extents;
1487 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1489 btrfs_item_key_to_cpu(buf, &key, i);
1490 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1492 fi = btrfs_item_ptr(buf, i,
1493 struct btrfs_file_extent_item);
1494 if (btrfs_file_extent_type(buf, fi) ==
1495 BTRFS_FILE_EXTENT_INLINE)
1497 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1498 if (disk_bytenr == 0)
1501 info->bytenr = disk_bytenr;
1503 btrfs_file_extent_disk_num_bytes(buf, fi);
1504 info->objectid = key.objectid;
1505 info->offset = key.offset;
1509 ret = btrfs_add_leaf_ref(root, ref, shared);
1510 if (ret == -EEXIST && shared) {
1511 struct btrfs_leaf_ref *old;
1512 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1514 btrfs_remove_leaf_ref(root, old);
1515 btrfs_free_leaf_ref(root, old);
1516 ret = btrfs_add_leaf_ref(root, ref, shared);
1519 btrfs_free_leaf_ref(root, ref);
1525 /* when a block goes through cow, we update the reference counts of
1526 * everything that block points to. The internal pointers of the block
1527 * can be in just about any order, and it is likely to have clusters of
1528 * things that are close together and clusters of things that are not.
1530 * To help reduce the seeks that come with updating all of these reference
1531 * counts, sort them by byte number before actual updates are done.
1533 * struct refsort is used to match byte number to slot in the btree block.
1534 * we sort based on the byte number and then use the slot to actually
1537 * struct refsort is smaller than strcut btrfs_item and smaller than
1538 * struct btrfs_key_ptr. Since we're currently limited to the page size
1539 * for a btree block, there's no way for a kmalloc of refsorts for a
1540 * single node to be bigger than a page.
1548 * for passing into sort()
1550 static int refsort_cmp(const void *a_void, const void *b_void)
1552 const struct refsort *a = a_void;
1553 const struct refsort *b = b_void;
1555 if (a->bytenr < b->bytenr)
1557 if (a->bytenr > b->bytenr)
1563 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1564 struct btrfs_root *root,
1565 struct extent_buffer *orig_buf,
1566 struct extent_buffer *buf, u32 *nr_extents)
1572 u64 orig_generation;
1573 struct refsort *sorted;
1575 u32 nr_file_extents = 0;
1576 struct btrfs_key key;
1577 struct btrfs_file_extent_item *fi;
1584 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1585 u64, u64, u64, u64, u64, u64, u64, u64);
1587 ref_root = btrfs_header_owner(buf);
1588 ref_generation = btrfs_header_generation(buf);
1589 orig_root = btrfs_header_owner(orig_buf);
1590 orig_generation = btrfs_header_generation(orig_buf);
1592 nritems = btrfs_header_nritems(buf);
1593 level = btrfs_header_level(buf);
1595 sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1598 if (root->ref_cows) {
1599 process_func = __btrfs_inc_extent_ref;
1602 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1605 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1607 process_func = __btrfs_update_extent_ref;
1611 * we make two passes through the items. In the first pass we
1612 * only record the byte number and slot. Then we sort based on
1613 * byte number and do the actual work based on the sorted results
1615 for (i = 0; i < nritems; i++) {
1618 btrfs_item_key_to_cpu(buf, &key, i);
1619 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1621 fi = btrfs_item_ptr(buf, i,
1622 struct btrfs_file_extent_item);
1623 if (btrfs_file_extent_type(buf, fi) ==
1624 BTRFS_FILE_EXTENT_INLINE)
1626 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1631 sorted[refi].bytenr = bytenr;
1632 sorted[refi].slot = i;
1635 bytenr = btrfs_node_blockptr(buf, i);
1636 sorted[refi].bytenr = bytenr;
1637 sorted[refi].slot = i;
1642 * if refi == 0, we didn't actually put anything into the sorted
1643 * array and we're done
1648 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1650 for (i = 0; i < refi; i++) {
1652 slot = sorted[i].slot;
1653 bytenr = sorted[i].bytenr;
1656 btrfs_item_key_to_cpu(buf, &key, slot);
1658 ret = process_func(trans, root, bytenr,
1659 orig_buf->start, buf->start,
1660 orig_root, ref_root,
1661 orig_generation, ref_generation,
1670 ret = process_func(trans, root, bytenr,
1671 orig_buf->start, buf->start,
1672 orig_root, ref_root,
1673 orig_generation, ref_generation,
1686 *nr_extents = nr_file_extents;
1688 *nr_extents = nritems;
1697 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1698 struct btrfs_root *root, struct extent_buffer *orig_buf,
1699 struct extent_buffer *buf, int start_slot, int nr)
1706 u64 orig_generation;
1707 struct btrfs_key key;
1708 struct btrfs_file_extent_item *fi;
1714 BUG_ON(start_slot < 0);
1715 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1717 ref_root = btrfs_header_owner(buf);
1718 ref_generation = btrfs_header_generation(buf);
1719 orig_root = btrfs_header_owner(orig_buf);
1720 orig_generation = btrfs_header_generation(orig_buf);
1721 level = btrfs_header_level(buf);
1723 if (!root->ref_cows) {
1725 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1728 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1732 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1735 btrfs_item_key_to_cpu(buf, &key, slot);
1736 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1738 fi = btrfs_item_ptr(buf, slot,
1739 struct btrfs_file_extent_item);
1740 if (btrfs_file_extent_type(buf, fi) ==
1741 BTRFS_FILE_EXTENT_INLINE)
1743 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1746 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1747 orig_buf->start, buf->start,
1748 orig_root, ref_root,
1749 orig_generation, ref_generation,
1754 bytenr = btrfs_node_blockptr(buf, slot);
1755 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1756 orig_buf->start, buf->start,
1757 orig_root, ref_root,
1758 orig_generation, ref_generation,
1770 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1771 struct btrfs_root *root,
1772 struct btrfs_path *path,
1773 struct btrfs_block_group_cache *cache)
1777 struct btrfs_root *extent_root = root->fs_info->extent_root;
1779 struct extent_buffer *leaf;
1781 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1786 leaf = path->nodes[0];
1787 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1788 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1789 btrfs_mark_buffer_dirty(leaf);
1790 btrfs_release_path(extent_root, path);
1792 finish_current_insert(trans, extent_root, 0);
1793 pending_ret = del_pending_extents(trans, extent_root, 0);
1802 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root)
1805 struct btrfs_block_group_cache *cache, *entry;
1809 struct btrfs_path *path;
1812 path = btrfs_alloc_path();
1818 spin_lock(&root->fs_info->block_group_cache_lock);
1819 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1820 n; n = rb_next(n)) {
1821 entry = rb_entry(n, struct btrfs_block_group_cache,
1828 spin_unlock(&root->fs_info->block_group_cache_lock);
1834 last += cache->key.offset;
1836 err = write_one_cache_group(trans, root,
1839 * if we fail to write the cache group, we want
1840 * to keep it marked dirty in hopes that a later
1848 btrfs_free_path(path);
1852 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1854 struct btrfs_block_group_cache *block_group;
1857 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1858 if (!block_group || block_group->ro)
1861 put_block_group(block_group);
1865 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1866 u64 total_bytes, u64 bytes_used,
1867 struct btrfs_space_info **space_info)
1869 struct btrfs_space_info *found;
1871 found = __find_space_info(info, flags);
1873 spin_lock(&found->lock);
1874 found->total_bytes += total_bytes;
1875 found->bytes_used += bytes_used;
1877 spin_unlock(&found->lock);
1878 *space_info = found;
1881 found = kzalloc(sizeof(*found), GFP_NOFS);
1885 list_add(&found->list, &info->space_info);
1886 INIT_LIST_HEAD(&found->block_groups);
1887 init_rwsem(&found->groups_sem);
1888 spin_lock_init(&found->lock);
1889 found->flags = flags;
1890 found->total_bytes = total_bytes;
1891 found->bytes_used = bytes_used;
1892 found->bytes_pinned = 0;
1893 found->bytes_reserved = 0;
1894 found->bytes_readonly = 0;
1896 found->force_alloc = 0;
1897 *space_info = found;
1901 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1903 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1904 BTRFS_BLOCK_GROUP_RAID1 |
1905 BTRFS_BLOCK_GROUP_RAID10 |
1906 BTRFS_BLOCK_GROUP_DUP);
1908 if (flags & BTRFS_BLOCK_GROUP_DATA)
1909 fs_info->avail_data_alloc_bits |= extra_flags;
1910 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1911 fs_info->avail_metadata_alloc_bits |= extra_flags;
1912 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1913 fs_info->avail_system_alloc_bits |= extra_flags;
1917 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1919 spin_lock(&cache->space_info->lock);
1920 spin_lock(&cache->lock);
1922 cache->space_info->bytes_readonly += cache->key.offset -
1923 btrfs_block_group_used(&cache->item);
1926 spin_unlock(&cache->lock);
1927 spin_unlock(&cache->space_info->lock);
1930 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1932 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1934 if (num_devices == 1)
1935 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1936 if (num_devices < 4)
1937 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1939 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1940 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1941 BTRFS_BLOCK_GROUP_RAID10))) {
1942 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1945 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1946 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1947 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1950 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1951 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1952 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1953 (flags & BTRFS_BLOCK_GROUP_DUP)))
1954 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1958 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1959 struct btrfs_root *extent_root, u64 alloc_bytes,
1960 u64 flags, int force)
1962 struct btrfs_space_info *space_info;
1966 mutex_lock(&extent_root->fs_info->chunk_mutex);
1968 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1970 space_info = __find_space_info(extent_root->fs_info, flags);
1972 ret = update_space_info(extent_root->fs_info, flags,
1976 BUG_ON(!space_info);
1978 spin_lock(&space_info->lock);
1979 if (space_info->force_alloc) {
1981 space_info->force_alloc = 0;
1983 if (space_info->full) {
1984 spin_unlock(&space_info->lock);
1988 thresh = space_info->total_bytes - space_info->bytes_readonly;
1989 thresh = div_factor(thresh, 6);
1991 (space_info->bytes_used + space_info->bytes_pinned +
1992 space_info->bytes_reserved + alloc_bytes) < thresh) {
1993 spin_unlock(&space_info->lock);
1996 spin_unlock(&space_info->lock);
1998 ret = btrfs_alloc_chunk(trans, extent_root, flags);
2000 space_info->full = 1;
2002 mutex_unlock(&extent_root->fs_info->chunk_mutex);
2006 static int update_block_group(struct btrfs_trans_handle *trans,
2007 struct btrfs_root *root,
2008 u64 bytenr, u64 num_bytes, int alloc,
2011 struct btrfs_block_group_cache *cache;
2012 struct btrfs_fs_info *info = root->fs_info;
2013 u64 total = num_bytes;
2018 cache = btrfs_lookup_block_group(info, bytenr);
2021 byte_in_group = bytenr - cache->key.objectid;
2022 WARN_ON(byte_in_group > cache->key.offset);
2024 spin_lock(&cache->space_info->lock);
2025 spin_lock(&cache->lock);
2027 old_val = btrfs_block_group_used(&cache->item);
2028 num_bytes = min(total, cache->key.offset - byte_in_group);
2030 old_val += num_bytes;
2031 cache->space_info->bytes_used += num_bytes;
2033 cache->space_info->bytes_readonly -= num_bytes;
2034 btrfs_set_block_group_used(&cache->item, old_val);
2035 spin_unlock(&cache->lock);
2036 spin_unlock(&cache->space_info->lock);
2038 old_val -= num_bytes;
2039 cache->space_info->bytes_used -= num_bytes;
2041 cache->space_info->bytes_readonly += num_bytes;
2042 btrfs_set_block_group_used(&cache->item, old_val);
2043 spin_unlock(&cache->lock);
2044 spin_unlock(&cache->space_info->lock);
2048 ret = btrfs_discard_extent(root, bytenr,
2052 ret = btrfs_add_free_space(cache, bytenr,
2057 put_block_group(cache);
2059 bytenr += num_bytes;
2064 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2066 struct btrfs_block_group_cache *cache;
2069 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2073 bytenr = cache->key.objectid;
2074 put_block_group(cache);
2079 int btrfs_update_pinned_extents(struct btrfs_root *root,
2080 u64 bytenr, u64 num, int pin)
2083 struct btrfs_block_group_cache *cache;
2084 struct btrfs_fs_info *fs_info = root->fs_info;
2086 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2088 set_extent_dirty(&fs_info->pinned_extents,
2089 bytenr, bytenr + num - 1, GFP_NOFS);
2091 clear_extent_dirty(&fs_info->pinned_extents,
2092 bytenr, bytenr + num - 1, GFP_NOFS);
2095 cache = btrfs_lookup_block_group(fs_info, bytenr);
2097 len = min(num, cache->key.offset -
2098 (bytenr - cache->key.objectid));
2100 spin_lock(&cache->space_info->lock);
2101 spin_lock(&cache->lock);
2102 cache->pinned += len;
2103 cache->space_info->bytes_pinned += len;
2104 spin_unlock(&cache->lock);
2105 spin_unlock(&cache->space_info->lock);
2106 fs_info->total_pinned += len;
2108 spin_lock(&cache->space_info->lock);
2109 spin_lock(&cache->lock);
2110 cache->pinned -= len;
2111 cache->space_info->bytes_pinned -= len;
2112 spin_unlock(&cache->lock);
2113 spin_unlock(&cache->space_info->lock);
2114 fs_info->total_pinned -= len;
2116 btrfs_add_free_space(cache, bytenr, len);
2118 put_block_group(cache);
2125 static int update_reserved_extents(struct btrfs_root *root,
2126 u64 bytenr, u64 num, int reserve)
2129 struct btrfs_block_group_cache *cache;
2130 struct btrfs_fs_info *fs_info = root->fs_info;
2133 cache = btrfs_lookup_block_group(fs_info, bytenr);
2135 len = min(num, cache->key.offset -
2136 (bytenr - cache->key.objectid));
2138 spin_lock(&cache->space_info->lock);
2139 spin_lock(&cache->lock);
2141 cache->reserved += len;
2142 cache->space_info->bytes_reserved += len;
2144 cache->reserved -= len;
2145 cache->space_info->bytes_reserved -= len;
2147 spin_unlock(&cache->lock);
2148 spin_unlock(&cache->space_info->lock);
2149 put_block_group(cache);
2156 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2161 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2164 mutex_lock(&root->fs_info->pinned_mutex);
2166 ret = find_first_extent_bit(pinned_extents, last,
2167 &start, &end, EXTENT_DIRTY);
2170 set_extent_dirty(copy, start, end, GFP_NOFS);
2173 mutex_unlock(&root->fs_info->pinned_mutex);
2177 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2178 struct btrfs_root *root,
2179 struct extent_io_tree *unpin)
2185 mutex_lock(&root->fs_info->pinned_mutex);
2187 ret = find_first_extent_bit(unpin, 0, &start, &end,
2192 ret = btrfs_discard_extent(root, start, end + 1 - start);
2194 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2195 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2197 if (need_resched()) {
2198 mutex_unlock(&root->fs_info->pinned_mutex);
2200 mutex_lock(&root->fs_info->pinned_mutex);
2203 mutex_unlock(&root->fs_info->pinned_mutex);
2207 static int finish_current_insert(struct btrfs_trans_handle *trans,
2208 struct btrfs_root *extent_root, int all)
2215 struct btrfs_fs_info *info = extent_root->fs_info;
2216 struct btrfs_path *path;
2217 struct pending_extent_op *extent_op, *tmp;
2218 struct list_head insert_list, update_list;
2220 int num_inserts = 0, max_inserts;
2222 path = btrfs_alloc_path();
2223 INIT_LIST_HEAD(&insert_list);
2224 INIT_LIST_HEAD(&update_list);
2226 max_inserts = extent_root->leafsize /
2227 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2228 sizeof(struct btrfs_extent_ref) +
2229 sizeof(struct btrfs_extent_item));
2231 mutex_lock(&info->extent_ins_mutex);
2233 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2234 &end, EXTENT_WRITEBACK);
2236 if (skipped && all && !num_inserts &&
2237 list_empty(&update_list)) {
2242 mutex_unlock(&info->extent_ins_mutex);
2246 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2250 if (need_resched()) {
2251 mutex_unlock(&info->extent_ins_mutex);
2253 mutex_lock(&info->extent_ins_mutex);
2258 ret = get_state_private(&info->extent_ins, start, &priv);
2260 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2262 if (extent_op->type == PENDING_EXTENT_INSERT) {
2264 list_add_tail(&extent_op->list, &insert_list);
2266 if (num_inserts == max_inserts) {
2267 mutex_unlock(&info->extent_ins_mutex);
2270 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2271 list_add_tail(&extent_op->list, &update_list);
2279 * process the update list, clear the writeback bit for it, and if
2280 * somebody marked this thing for deletion then just unlock it and be
2281 * done, the free_extents will handle it
2283 mutex_lock(&info->extent_ins_mutex);
2284 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2285 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2286 extent_op->bytenr + extent_op->num_bytes - 1,
2287 EXTENT_WRITEBACK, GFP_NOFS);
2288 if (extent_op->del) {
2289 list_del_init(&extent_op->list);
2290 unlock_extent(&info->extent_ins, extent_op->bytenr,
2291 extent_op->bytenr + extent_op->num_bytes
2296 mutex_unlock(&info->extent_ins_mutex);
2299 * still have things left on the update list, go ahead an update
2302 if (!list_empty(&update_list)) {
2303 ret = update_backrefs(trans, extent_root, path, &update_list);
2308 * if no inserts need to be done, but we skipped some extents and we
2309 * need to make sure everything is cleaned then reset everything and
2310 * go back to the beginning
2312 if (!num_inserts && all && skipped) {
2315 INIT_LIST_HEAD(&update_list);
2316 INIT_LIST_HEAD(&insert_list);
2318 } else if (!num_inserts) {
2323 * process the insert extents list. Again if we are deleting this
2324 * extent, then just unlock it, pin down the bytes if need be, and be
2325 * done with it. Saves us from having to actually insert the extent
2326 * into the tree and then subsequently come along and delete it
2328 mutex_lock(&info->extent_ins_mutex);
2329 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2330 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2331 extent_op->bytenr + extent_op->num_bytes - 1,
2332 EXTENT_WRITEBACK, GFP_NOFS);
2333 if (extent_op->del) {
2335 list_del_init(&extent_op->list);
2336 unlock_extent(&info->extent_ins, extent_op->bytenr,
2337 extent_op->bytenr + extent_op->num_bytes
2340 mutex_lock(&extent_root->fs_info->pinned_mutex);
2341 ret = pin_down_bytes(trans, extent_root,
2343 extent_op->num_bytes, 0);
2344 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2346 spin_lock(&info->delalloc_lock);
2347 used = btrfs_super_bytes_used(&info->super_copy);
2348 btrfs_set_super_bytes_used(&info->super_copy,
2349 used - extent_op->num_bytes);
2350 used = btrfs_root_used(&extent_root->root_item);
2351 btrfs_set_root_used(&extent_root->root_item,
2352 used - extent_op->num_bytes);
2353 spin_unlock(&info->delalloc_lock);
2355 ret = update_block_group(trans, extent_root,
2357 extent_op->num_bytes,
2364 mutex_unlock(&info->extent_ins_mutex);
2366 ret = insert_extents(trans, extent_root, path, &insert_list,
2371 * if we broke out of the loop in order to insert stuff because we hit
2372 * the maximum number of inserts at a time we can handle, then loop
2373 * back and pick up where we left off
2375 if (num_inserts == max_inserts) {
2376 INIT_LIST_HEAD(&insert_list);
2377 INIT_LIST_HEAD(&update_list);
2383 * again, if we need to make absolutely sure there are no more pending
2384 * extent operations left and we know that we skipped some, go back to
2385 * the beginning and do it all again
2387 if (all && skipped) {
2388 INIT_LIST_HEAD(&insert_list);
2389 INIT_LIST_HEAD(&update_list);
2396 btrfs_free_path(path);
2400 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2401 struct btrfs_root *root,
2402 u64 bytenr, u64 num_bytes, int is_data)
2405 struct extent_buffer *buf;
2410 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2414 /* we can reuse a block if it hasn't been written
2415 * and it is from this transaction. We can't
2416 * reuse anything from the tree log root because
2417 * it has tiny sub-transactions.
2419 if (btrfs_buffer_uptodate(buf, 0) &&
2420 btrfs_try_tree_lock(buf)) {
2421 u64 header_owner = btrfs_header_owner(buf);
2422 u64 header_transid = btrfs_header_generation(buf);
2423 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2424 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2425 header_transid == trans->transid &&
2426 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2427 clean_tree_block(NULL, root, buf);
2428 btrfs_tree_unlock(buf);
2429 free_extent_buffer(buf);
2432 btrfs_tree_unlock(buf);
2434 free_extent_buffer(buf);
2436 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2443 * remove an extent from the root, returns 0 on success
2445 static int __free_extent(struct btrfs_trans_handle *trans,
2446 struct btrfs_root *root,
2447 u64 bytenr, u64 num_bytes, u64 parent,
2448 u64 root_objectid, u64 ref_generation,
2449 u64 owner_objectid, int pin, int mark_free)
2451 struct btrfs_path *path;
2452 struct btrfs_key key;
2453 struct btrfs_fs_info *info = root->fs_info;
2454 struct btrfs_root *extent_root = info->extent_root;
2455 struct extent_buffer *leaf;
2457 int extent_slot = 0;
2458 int found_extent = 0;
2460 struct btrfs_extent_item *ei;
2463 key.objectid = bytenr;
2464 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2465 key.offset = num_bytes;
2466 path = btrfs_alloc_path();
2471 ret = lookup_extent_backref(trans, extent_root, path,
2472 bytenr, parent, root_objectid,
2473 ref_generation, owner_objectid, 1);
2475 struct btrfs_key found_key;
2476 extent_slot = path->slots[0];
2477 while (extent_slot > 0) {
2479 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2481 if (found_key.objectid != bytenr)
2483 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2484 found_key.offset == num_bytes) {
2488 if (path->slots[0] - extent_slot > 5)
2491 if (!found_extent) {
2492 ret = remove_extent_backref(trans, extent_root, path);
2494 btrfs_release_path(extent_root, path);
2495 ret = btrfs_search_slot(trans, extent_root,
2498 printk(KERN_ERR "umm, got %d back from search"
2499 ", was looking for %llu\n", ret,
2500 (unsigned long long)bytenr);
2501 btrfs_print_leaf(extent_root, path->nodes[0]);
2504 extent_slot = path->slots[0];
2507 btrfs_print_leaf(extent_root, path->nodes[0]);
2509 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2510 "root %llu gen %llu owner %llu\n",
2511 (unsigned long long)bytenr,
2512 (unsigned long long)root_objectid,
2513 (unsigned long long)ref_generation,
2514 (unsigned long long)owner_objectid);
2517 leaf = path->nodes[0];
2518 ei = btrfs_item_ptr(leaf, extent_slot,
2519 struct btrfs_extent_item);
2520 refs = btrfs_extent_refs(leaf, ei);
2523 btrfs_set_extent_refs(leaf, ei, refs);
2525 btrfs_mark_buffer_dirty(leaf);
2527 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2528 struct btrfs_extent_ref *ref;
2529 ref = btrfs_item_ptr(leaf, path->slots[0],
2530 struct btrfs_extent_ref);
2531 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2532 /* if the back ref and the extent are next to each other
2533 * they get deleted below in one shot
2535 path->slots[0] = extent_slot;
2537 } else if (found_extent) {
2538 /* otherwise delete the extent back ref */
2539 ret = remove_extent_backref(trans, extent_root, path);
2541 /* if refs are 0, we need to setup the path for deletion */
2543 btrfs_release_path(extent_root, path);
2544 ret = btrfs_search_slot(trans, extent_root, &key, path,
2555 mutex_lock(&root->fs_info->pinned_mutex);
2556 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2557 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2558 mutex_unlock(&root->fs_info->pinned_mutex);
2563 /* block accounting for super block */
2564 spin_lock(&info->delalloc_lock);
2565 super_used = btrfs_super_bytes_used(&info->super_copy);
2566 btrfs_set_super_bytes_used(&info->super_copy,
2567 super_used - num_bytes);
2569 /* block accounting for root item */
2570 root_used = btrfs_root_used(&root->root_item);
2571 btrfs_set_root_used(&root->root_item,
2572 root_used - num_bytes);
2573 spin_unlock(&info->delalloc_lock);
2574 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2577 btrfs_release_path(extent_root, path);
2579 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2580 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2584 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2588 btrfs_free_path(path);
2589 finish_current_insert(trans, extent_root, 0);
2594 * find all the blocks marked as pending in the radix tree and remove
2595 * them from the extent map
2597 static int del_pending_extents(struct btrfs_trans_handle *trans,
2598 struct btrfs_root *extent_root, int all)
2606 int nr = 0, skipped = 0;
2607 struct extent_io_tree *pending_del;
2608 struct extent_io_tree *extent_ins;
2609 struct pending_extent_op *extent_op;
2610 struct btrfs_fs_info *info = extent_root->fs_info;
2611 struct list_head delete_list;
2613 INIT_LIST_HEAD(&delete_list);
2614 extent_ins = &extent_root->fs_info->extent_ins;
2615 pending_del = &extent_root->fs_info->pending_del;
2618 mutex_lock(&info->extent_ins_mutex);
2620 ret = find_first_extent_bit(pending_del, search, &start, &end,
2623 if (all && skipped && !nr) {
2628 mutex_unlock(&info->extent_ins_mutex);
2632 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2637 if (need_resched()) {
2638 mutex_unlock(&info->extent_ins_mutex);
2640 mutex_lock(&info->extent_ins_mutex);
2647 ret = get_state_private(pending_del, start, &priv);
2649 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2651 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2653 if (!test_range_bit(extent_ins, start, end,
2654 EXTENT_WRITEBACK, 0)) {
2655 list_add_tail(&extent_op->list, &delete_list);
2660 ret = get_state_private(&info->extent_ins, start,
2663 extent_op = (struct pending_extent_op *)
2664 (unsigned long)priv;
2666 clear_extent_bits(&info->extent_ins, start, end,
2667 EXTENT_WRITEBACK, GFP_NOFS);
2669 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2670 list_add_tail(&extent_op->list, &delete_list);
2676 mutex_lock(&extent_root->fs_info->pinned_mutex);
2677 ret = pin_down_bytes(trans, extent_root, start,
2678 end + 1 - start, 0);
2679 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2681 ret = update_block_group(trans, extent_root, start,
2682 end + 1 - start, 0, ret > 0);
2684 unlock_extent(extent_ins, start, end, GFP_NOFS);
2693 if (need_resched()) {
2694 mutex_unlock(&info->extent_ins_mutex);
2696 mutex_lock(&info->extent_ins_mutex);
2701 ret = free_extents(trans, extent_root, &delete_list);
2705 if (all && skipped) {
2706 INIT_LIST_HEAD(&delete_list);
2716 * remove an extent from the root, returns 0 on success
2718 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root,
2720 u64 bytenr, u64 num_bytes, u64 parent,
2721 u64 root_objectid, u64 ref_generation,
2722 u64 owner_objectid, int pin)
2724 struct btrfs_root *extent_root = root->fs_info->extent_root;
2728 WARN_ON(num_bytes < root->sectorsize);
2729 if (root == extent_root) {
2730 struct pending_extent_op *extent_op = NULL;
2732 mutex_lock(&root->fs_info->extent_ins_mutex);
2733 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2734 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2736 ret = get_state_private(&root->fs_info->extent_ins,
2739 extent_op = (struct pending_extent_op *)
2740 (unsigned long)priv;
2743 if (extent_op->type == PENDING_EXTENT_INSERT) {
2744 mutex_unlock(&root->fs_info->extent_ins_mutex);
2750 ref_generation = extent_op->orig_generation;
2751 parent = extent_op->orig_parent;
2754 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2757 extent_op->type = PENDING_EXTENT_DELETE;
2758 extent_op->bytenr = bytenr;
2759 extent_op->num_bytes = num_bytes;
2760 extent_op->parent = parent;
2761 extent_op->orig_parent = parent;
2762 extent_op->generation = ref_generation;
2763 extent_op->orig_generation = ref_generation;
2764 extent_op->level = (int)owner_objectid;
2765 INIT_LIST_HEAD(&extent_op->list);
2768 set_extent_bits(&root->fs_info->pending_del,
2769 bytenr, bytenr + num_bytes - 1,
2770 EXTENT_WRITEBACK, GFP_NOFS);
2771 set_state_private(&root->fs_info->pending_del,
2772 bytenr, (unsigned long)extent_op);
2773 mutex_unlock(&root->fs_info->extent_ins_mutex);
2776 /* if metadata always pin */
2777 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2778 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2779 mutex_lock(&root->fs_info->pinned_mutex);
2780 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2781 mutex_unlock(&root->fs_info->pinned_mutex);
2782 update_reserved_extents(root, bytenr, num_bytes, 0);
2788 /* if data pin when any transaction has committed this */
2789 if (ref_generation != trans->transid)
2792 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2793 root_objectid, ref_generation,
2794 owner_objectid, pin, pin == 0);
2796 finish_current_insert(trans, root->fs_info->extent_root, 0);
2797 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2798 return ret ? ret : pending_ret;
2801 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2802 struct btrfs_root *root,
2803 u64 bytenr, u64 num_bytes, u64 parent,
2804 u64 root_objectid, u64 ref_generation,
2805 u64 owner_objectid, int pin)
2809 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2810 root_objectid, ref_generation,
2811 owner_objectid, pin);
2815 static u64 stripe_align(struct btrfs_root *root, u64 val)
2817 u64 mask = ((u64)root->stripesize - 1);
2818 u64 ret = (val + mask) & ~mask;
2823 * walks the btree of allocated extents and find a hole of a given size.
2824 * The key ins is changed to record the hole:
2825 * ins->objectid == block start
2826 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2827 * ins->offset == number of blocks
2828 * Any available blocks before search_start are skipped.
2830 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2831 struct btrfs_root *orig_root,
2832 u64 num_bytes, u64 empty_size,
2833 u64 search_start, u64 search_end,
2834 u64 hint_byte, struct btrfs_key *ins,
2835 u64 exclude_start, u64 exclude_nr,
2839 struct btrfs_root *root = orig_root->fs_info->extent_root;
2840 u64 total_needed = num_bytes;
2841 u64 *last_ptr = NULL;
2842 u64 last_wanted = 0;
2843 struct btrfs_block_group_cache *block_group = NULL;
2844 int chunk_alloc_done = 0;
2845 int empty_cluster = 2 * 1024 * 1024;
2846 int allowed_chunk_alloc = 0;
2847 struct list_head *head = NULL, *cur = NULL;
2850 struct btrfs_space_info *space_info;
2852 WARN_ON(num_bytes < root->sectorsize);
2853 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2857 if (orig_root->ref_cows || empty_size)
2858 allowed_chunk_alloc = 1;
2860 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2861 last_ptr = &root->fs_info->last_alloc;
2862 empty_cluster = 64 * 1024;
2865 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2866 last_ptr = &root->fs_info->last_data_alloc;
2870 hint_byte = *last_ptr;
2871 last_wanted = *last_ptr;
2873 empty_size += empty_cluster;
2877 search_start = max(search_start, first_logical_byte(root, 0));
2878 search_start = max(search_start, hint_byte);
2880 if (last_wanted && search_start != last_wanted) {
2882 empty_size += empty_cluster;
2885 total_needed += empty_size;
2886 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2888 block_group = btrfs_lookup_first_block_group(root->fs_info,
2890 space_info = __find_space_info(root->fs_info, data);
2892 down_read(&space_info->groups_sem);
2894 struct btrfs_free_space *free_space;
2896 * the only way this happens if our hint points to a block
2897 * group thats not of the proper type, while looping this
2898 * should never happen
2904 goto new_group_no_lock;
2906 if (unlikely(!block_group->cached)) {
2907 mutex_lock(&block_group->cache_mutex);
2908 ret = cache_block_group(root, block_group);
2909 mutex_unlock(&block_group->cache_mutex);
2914 mutex_lock(&block_group->alloc_mutex);
2915 if (unlikely(!block_group_bits(block_group, data)))
2918 if (unlikely(block_group->ro))
2921 free_space = btrfs_find_free_space(block_group, search_start,
2924 u64 start = block_group->key.objectid;
2925 u64 end = block_group->key.objectid +
2926 block_group->key.offset;
2928 search_start = stripe_align(root, free_space->offset);
2930 /* move on to the next group */
2931 if (search_start + num_bytes >= search_end)
2934 /* move on to the next group */
2935 if (search_start + num_bytes > end)
2938 if (last_wanted && search_start != last_wanted) {
2939 total_needed += empty_cluster;
2940 empty_size += empty_cluster;
2943 * if search_start is still in this block group
2944 * then we just re-search this block group
2946 if (search_start >= start &&
2947 search_start < end) {
2948 mutex_unlock(&block_group->alloc_mutex);
2952 /* else we go to the next block group */
2956 if (exclude_nr > 0 &&
2957 (search_start + num_bytes > exclude_start &&
2958 search_start < exclude_start + exclude_nr)) {
2959 search_start = exclude_start + exclude_nr;
2961 * if search_start is still in this block group
2962 * then we just re-search this block group
2964 if (search_start >= start &&
2965 search_start < end) {
2966 mutex_unlock(&block_group->alloc_mutex);
2971 /* else we go to the next block group */
2975 ins->objectid = search_start;
2976 ins->offset = num_bytes;
2978 btrfs_remove_free_space_lock(block_group, search_start,
2980 /* we are all good, lets return */
2981 mutex_unlock(&block_group->alloc_mutex);
2985 mutex_unlock(&block_group->alloc_mutex);
2986 put_block_group(block_group);
2989 /* don't try to compare new allocations against the
2990 * last allocation any more
2995 * Here's how this works.
2996 * loop == 0: we were searching a block group via a hint
2997 * and didn't find anything, so we start at
2998 * the head of the block groups and keep searching
2999 * loop == 1: we're searching through all of the block groups
3000 * if we hit the head again we have searched
3001 * all of the block groups for this space and we
3002 * need to try and allocate, if we cant error out.
3003 * loop == 2: we allocated more space and are looping through
3004 * all of the block groups again.
3007 head = &space_info->block_groups;
3010 } else if (loop == 1 && cur == head) {
3013 /* at this point we give up on the empty_size
3014 * allocations and just try to allocate the min
3017 * The extra_loop field was set if an empty_size
3018 * allocation was attempted above, and if this
3019 * is try we need to try the loop again without
3020 * the additional empty_size.
3022 total_needed -= empty_size;
3024 keep_going = extra_loop;
3027 if (allowed_chunk_alloc && !chunk_alloc_done) {
3028 up_read(&space_info->groups_sem);
3029 ret = do_chunk_alloc(trans, root, num_bytes +
3030 2 * 1024 * 1024, data, 1);
3031 down_read(&space_info->groups_sem);
3034 head = &space_info->block_groups;
3036 * we've allocated a new chunk, keep
3040 chunk_alloc_done = 1;
3041 } else if (!allowed_chunk_alloc) {
3042 space_info->force_alloc = 1;
3051 } else if (cur == head) {
3055 block_group = list_entry(cur, struct btrfs_block_group_cache,
3057 atomic_inc(&block_group->count);
3059 search_start = block_group->key.objectid;
3063 /* we found what we needed */
3064 if (ins->objectid) {
3065 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3066 trans->block_group = block_group->key.objectid;
3069 *last_ptr = ins->objectid + ins->offset;
3072 printk(KERN_ERR "btrfs searching for %llu bytes, "
3073 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3074 (unsigned long long)total_needed,
3075 (unsigned long long)num_bytes,
3076 loop, allowed_chunk_alloc);
3080 put_block_group(block_group);
3082 up_read(&space_info->groups_sem);
3086 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3088 struct btrfs_block_group_cache *cache;
3090 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3091 (unsigned long long)(info->total_bytes - info->bytes_used -
3092 info->bytes_pinned - info->bytes_reserved),
3093 (info->full) ? "" : "not ");
3095 down_read(&info->groups_sem);
3096 list_for_each_entry(cache, &info->block_groups, list) {
3097 spin_lock(&cache->lock);
3098 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3099 "%llu pinned %llu reserved\n",
3100 (unsigned long long)cache->key.objectid,
3101 (unsigned long long)cache->key.offset,
3102 (unsigned long long)btrfs_block_group_used(&cache->item),
3103 (unsigned long long)cache->pinned,
3104 (unsigned long long)cache->reserved);
3105 btrfs_dump_free_space(cache, bytes);
3106 spin_unlock(&cache->lock);
3108 up_read(&info->groups_sem);
3111 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3112 struct btrfs_root *root,
3113 u64 num_bytes, u64 min_alloc_size,
3114 u64 empty_size, u64 hint_byte,
3115 u64 search_end, struct btrfs_key *ins,
3119 u64 search_start = 0;
3121 struct btrfs_fs_info *info = root->fs_info;
3124 alloc_profile = info->avail_data_alloc_bits &
3125 info->data_alloc_profile;
3126 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3127 } else if (root == root->fs_info->chunk_root) {
3128 alloc_profile = info->avail_system_alloc_bits &
3129 info->system_alloc_profile;
3130 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3132 alloc_profile = info->avail_metadata_alloc_bits &
3133 info->metadata_alloc_profile;
3134 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3137 data = btrfs_reduce_alloc_profile(root, data);
3139 * the only place that sets empty_size is btrfs_realloc_node, which
3140 * is not called recursively on allocations
3142 if (empty_size || root->ref_cows) {
3143 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3144 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3146 BTRFS_BLOCK_GROUP_METADATA |
3147 (info->metadata_alloc_profile &
3148 info->avail_metadata_alloc_bits), 0);
3150 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3151 num_bytes + 2 * 1024 * 1024, data, 0);
3154 WARN_ON(num_bytes < root->sectorsize);
3155 ret = find_free_extent(trans, root, num_bytes, empty_size,
3156 search_start, search_end, hint_byte, ins,
3157 trans->alloc_exclude_start,
3158 trans->alloc_exclude_nr, data);
3160 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3161 num_bytes = num_bytes >> 1;
3162 num_bytes = num_bytes & ~(root->sectorsize - 1);
3163 num_bytes = max(num_bytes, min_alloc_size);
3164 do_chunk_alloc(trans, root->fs_info->extent_root,
3165 num_bytes, data, 1);
3169 struct btrfs_space_info *sinfo;
3171 sinfo = __find_space_info(root->fs_info, data);
3172 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3173 "wanted %llu\n", (unsigned long long)data,
3174 (unsigned long long)num_bytes);
3175 dump_space_info(sinfo, num_bytes);
3182 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3184 struct btrfs_block_group_cache *cache;
3187 cache = btrfs_lookup_block_group(root->fs_info, start);
3189 printk(KERN_ERR "Unable to find block group for %llu\n",
3190 (unsigned long long)start);
3194 ret = btrfs_discard_extent(root, start, len);
3196 btrfs_add_free_space(cache, start, len);
3197 put_block_group(cache);
3198 update_reserved_extents(root, start, len, 0);
3203 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3204 struct btrfs_root *root,
3205 u64 num_bytes, u64 min_alloc_size,
3206 u64 empty_size, u64 hint_byte,
3207 u64 search_end, struct btrfs_key *ins,
3211 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3212 empty_size, hint_byte, search_end, ins,
3214 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3218 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3219 struct btrfs_root *root, u64 parent,
3220 u64 root_objectid, u64 ref_generation,
3221 u64 owner, struct btrfs_key *ins)
3227 u64 num_bytes = ins->offset;
3229 struct btrfs_fs_info *info = root->fs_info;
3230 struct btrfs_root *extent_root = info->extent_root;
3231 struct btrfs_extent_item *extent_item;
3232 struct btrfs_extent_ref *ref;
3233 struct btrfs_path *path;
3234 struct btrfs_key keys[2];
3237 parent = ins->objectid;
3239 /* block accounting for super block */
3240 spin_lock(&info->delalloc_lock);
3241 super_used = btrfs_super_bytes_used(&info->super_copy);
3242 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3244 /* block accounting for root item */
3245 root_used = btrfs_root_used(&root->root_item);
3246 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3247 spin_unlock(&info->delalloc_lock);
3249 if (root == extent_root) {
3250 struct pending_extent_op *extent_op;
3252 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3255 extent_op->type = PENDING_EXTENT_INSERT;
3256 extent_op->bytenr = ins->objectid;
3257 extent_op->num_bytes = ins->offset;
3258 extent_op->parent = parent;
3259 extent_op->orig_parent = 0;
3260 extent_op->generation = ref_generation;
3261 extent_op->orig_generation = 0;
3262 extent_op->level = (int)owner;
3263 INIT_LIST_HEAD(&extent_op->list);
3266 mutex_lock(&root->fs_info->extent_ins_mutex);
3267 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3268 ins->objectid + ins->offset - 1,
3269 EXTENT_WRITEBACK, GFP_NOFS);
3270 set_state_private(&root->fs_info->extent_ins,
3271 ins->objectid, (unsigned long)extent_op);
3272 mutex_unlock(&root->fs_info->extent_ins_mutex);
3276 memcpy(&keys[0], ins, sizeof(*ins));
3277 keys[1].objectid = ins->objectid;
3278 keys[1].type = BTRFS_EXTENT_REF_KEY;
3279 keys[1].offset = parent;
3280 sizes[0] = sizeof(*extent_item);
3281 sizes[1] = sizeof(*ref);
3283 path = btrfs_alloc_path();
3286 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3290 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3291 struct btrfs_extent_item);
3292 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3293 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3294 struct btrfs_extent_ref);
3296 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3297 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3298 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3299 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3301 btrfs_mark_buffer_dirty(path->nodes[0]);
3303 trans->alloc_exclude_start = 0;
3304 trans->alloc_exclude_nr = 0;
3305 btrfs_free_path(path);
3306 finish_current_insert(trans, extent_root, 0);
3307 pending_ret = del_pending_extents(trans, extent_root, 0);
3317 ret = update_block_group(trans, root, ins->objectid,
3320 printk(KERN_ERR "btrfs update block group failed for %llu "
3321 "%llu\n", (unsigned long long)ins->objectid,
3322 (unsigned long long)ins->offset);
3329 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root, u64 parent,
3331 u64 root_objectid, u64 ref_generation,
3332 u64 owner, struct btrfs_key *ins)
3336 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3338 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3339 ref_generation, owner, ins);
3340 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3345 * this is used by the tree logging recovery code. It records that
3346 * an extent has been allocated and makes sure to clear the free
3347 * space cache bits as well
3349 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3350 struct btrfs_root *root, u64 parent,
3351 u64 root_objectid, u64 ref_generation,
3352 u64 owner, struct btrfs_key *ins)
3355 struct btrfs_block_group_cache *block_group;
3357 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3358 mutex_lock(&block_group->cache_mutex);
3359 cache_block_group(root, block_group);
3360 mutex_unlock(&block_group->cache_mutex);
3362 ret = btrfs_remove_free_space(block_group, ins->objectid,
3365 put_block_group(block_group);
3366 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3367 ref_generation, owner, ins);
3372 * finds a free extent and does all the dirty work required for allocation
3373 * returns the key for the extent through ins, and a tree buffer for
3374 * the first block of the extent through buf.
3376 * returns 0 if everything worked, non-zero otherwise.
3378 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *root,
3380 u64 num_bytes, u64 parent, u64 min_alloc_size,
3381 u64 root_objectid, u64 ref_generation,
3382 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3383 u64 search_end, struct btrfs_key *ins, u64 data)
3387 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3388 min_alloc_size, empty_size, hint_byte,
3389 search_end, ins, data);
3391 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3392 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3393 root_objectid, ref_generation,
3394 owner_objectid, ins);
3398 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3403 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3404 struct btrfs_root *root,
3405 u64 bytenr, u32 blocksize)
3407 struct extent_buffer *buf;
3409 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3411 return ERR_PTR(-ENOMEM);
3412 btrfs_set_header_generation(buf, trans->transid);
3413 btrfs_tree_lock(buf);
3414 clean_tree_block(trans, root, buf);
3416 btrfs_set_lock_blocking(buf);
3417 btrfs_set_buffer_uptodate(buf);
3419 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3420 set_extent_dirty(&root->dirty_log_pages, buf->start,
3421 buf->start + buf->len - 1, GFP_NOFS);
3423 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3424 buf->start + buf->len - 1, GFP_NOFS);
3426 trans->blocks_used++;
3427 /* this returns a buffer locked for blocking */
3432 * helper function to allocate a block for a given tree
3433 * returns the tree buffer or NULL.
3435 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3436 struct btrfs_root *root,
3437 u32 blocksize, u64 parent,
3444 struct btrfs_key ins;
3446 struct extent_buffer *buf;
3448 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3449 root_objectid, ref_generation, level,
3450 empty_size, hint, (u64)-1, &ins, 0);
3453 return ERR_PTR(ret);
3456 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3460 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3461 struct btrfs_root *root, struct extent_buffer *leaf)
3464 u64 leaf_generation;
3465 struct refsort *sorted;
3466 struct btrfs_key key;
3467 struct btrfs_file_extent_item *fi;
3474 BUG_ON(!btrfs_is_leaf(leaf));
3475 nritems = btrfs_header_nritems(leaf);
3476 leaf_owner = btrfs_header_owner(leaf);
3477 leaf_generation = btrfs_header_generation(leaf);
3479 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3480 /* we do this loop twice. The first time we build a list
3481 * of the extents we have a reference on, then we sort the list
3482 * by bytenr. The second time around we actually do the
3485 for (i = 0; i < nritems; i++) {
3489 btrfs_item_key_to_cpu(leaf, &key, i);
3491 /* only extents have references, skip everything else */
3492 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3495 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3497 /* inline extents live in the btree, they don't have refs */
3498 if (btrfs_file_extent_type(leaf, fi) ==
3499 BTRFS_FILE_EXTENT_INLINE)
3502 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3504 /* holes don't have refs */
3505 if (disk_bytenr == 0)
3508 sorted[refi].bytenr = disk_bytenr;
3509 sorted[refi].slot = i;
3516 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3518 for (i = 0; i < refi; i++) {
3521 disk_bytenr = sorted[i].bytenr;
3522 slot = sorted[i].slot;
3526 btrfs_item_key_to_cpu(leaf, &key, slot);
3527 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3530 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3532 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3533 btrfs_file_extent_disk_num_bytes(leaf, fi),
3534 leaf->start, leaf_owner, leaf_generation,
3538 atomic_inc(&root->fs_info->throttle_gen);
3539 wake_up(&root->fs_info->transaction_throttle);
3547 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3548 struct btrfs_root *root,
3549 struct btrfs_leaf_ref *ref)
3553 struct btrfs_extent_info *info;
3554 struct refsort *sorted;
3556 if (ref->nritems == 0)
3559 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3560 for (i = 0; i < ref->nritems; i++) {
3561 sorted[i].bytenr = ref->extents[i].bytenr;
3564 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3567 * the items in the ref were sorted when the ref was inserted
3568 * into the ref cache, so this is already in order
3570 for (i = 0; i < ref->nritems; i++) {
3571 info = ref->extents + sorted[i].slot;
3572 ret = __btrfs_free_extent(trans, root, info->bytenr,
3573 info->num_bytes, ref->bytenr,
3574 ref->owner, ref->generation,
3577 atomic_inc(&root->fs_info->throttle_gen);
3578 wake_up(&root->fs_info->transaction_throttle);
3588 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3593 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3596 #if 0 /* some debugging code in case we see problems here */
3597 /* if the refs count is one, it won't get increased again. But
3598 * if the ref count is > 1, someone may be decreasing it at
3599 * the same time we are.
3602 struct extent_buffer *eb = NULL;
3603 eb = btrfs_find_create_tree_block(root, start, len);
3605 btrfs_tree_lock(eb);
3607 mutex_lock(&root->fs_info->alloc_mutex);
3608 ret = lookup_extent_ref(NULL, root, start, len, refs);
3610 mutex_unlock(&root->fs_info->alloc_mutex);
3613 btrfs_tree_unlock(eb);
3614 free_extent_buffer(eb);
3617 printk(KERN_ERR "btrfs block %llu went down to one "
3618 "during drop_snap\n", (unsigned long long)start);
3629 * this is used while deleting old snapshots, and it drops the refs
3630 * on a whole subtree starting from a level 1 node.
3632 * The idea is to sort all the leaf pointers, and then drop the
3633 * ref on all the leaves in order. Most of the time the leaves
3634 * will have ref cache entries, so no leaf IOs will be required to
3635 * find the extents they have references on.
3637 * For each leaf, any references it has are also dropped in order
3639 * This ends up dropping the references in something close to optimal
3640 * order for reading and modifying the extent allocation tree.
3642 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3643 struct btrfs_root *root,
3644 struct btrfs_path *path)
3649 struct extent_buffer *eb = path->nodes[1];
3650 struct extent_buffer *leaf;
3651 struct btrfs_leaf_ref *ref;
3652 struct refsort *sorted = NULL;
3653 int nritems = btrfs_header_nritems(eb);
3657 int slot = path->slots[1];
3658 u32 blocksize = btrfs_level_size(root, 0);
3664 root_owner = btrfs_header_owner(eb);
3665 root_gen = btrfs_header_generation(eb);
3666 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3669 * step one, sort all the leaf pointers so we don't scribble
3670 * randomly into the extent allocation tree
3672 for (i = slot; i < nritems; i++) {
3673 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3674 sorted[refi].slot = i;
3679 * nritems won't be zero, but if we're picking up drop_snapshot
3680 * after a crash, slot might be > 0, so double check things
3686 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3689 * the first loop frees everything the leaves point to
3691 for (i = 0; i < refi; i++) {
3694 bytenr = sorted[i].bytenr;
3697 * check the reference count on this leaf. If it is > 1
3698 * we just decrement it below and don't update any
3699 * of the refs the leaf points to.
3701 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3706 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3709 * the leaf only had one reference, which means the
3710 * only thing pointing to this leaf is the snapshot
3711 * we're deleting. It isn't possible for the reference
3712 * count to increase again later
3714 * The reference cache is checked for the leaf,
3715 * and if found we'll be able to drop any refs held by
3716 * the leaf without needing to read it in.
3718 ref = btrfs_lookup_leaf_ref(root, bytenr);
3719 if (ref && ref->generation != ptr_gen) {
3720 btrfs_free_leaf_ref(root, ref);
3724 ret = cache_drop_leaf_ref(trans, root, ref);
3726 btrfs_remove_leaf_ref(root, ref);
3727 btrfs_free_leaf_ref(root, ref);
3730 * the leaf wasn't in the reference cache, so
3731 * we have to read it.
3733 leaf = read_tree_block(root, bytenr, blocksize,
3735 ret = btrfs_drop_leaf_ref(trans, root, leaf);
3737 free_extent_buffer(leaf);
3739 atomic_inc(&root->fs_info->throttle_gen);
3740 wake_up(&root->fs_info->transaction_throttle);
3745 * run through the loop again to free the refs on the leaves.
3746 * This is faster than doing it in the loop above because
3747 * the leaves are likely to be clustered together. We end up
3748 * working in nice chunks on the extent allocation tree.
3750 for (i = 0; i < refi; i++) {
3751 bytenr = sorted[i].bytenr;
3752 ret = __btrfs_free_extent(trans, root, bytenr,
3753 blocksize, eb->start,
3754 root_owner, root_gen, 0, 1);
3757 atomic_inc(&root->fs_info->throttle_gen);
3758 wake_up(&root->fs_info->transaction_throttle);
3765 * update the path to show we've processed the entire level 1
3766 * node. This will get saved into the root's drop_snapshot_progress
3767 * field so these drops are not repeated again if this transaction
3770 path->slots[1] = nritems;
3775 * helper function for drop_snapshot, this walks down the tree dropping ref
3776 * counts as it goes.
3778 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3779 struct btrfs_root *root,
3780 struct btrfs_path *path, int *level)
3786 struct extent_buffer *next;
3787 struct extent_buffer *cur;
3788 struct extent_buffer *parent;
3793 WARN_ON(*level < 0);
3794 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3795 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3796 path->nodes[*level]->len, &refs);
3802 * walk down to the last node level and free all the leaves
3804 while (*level >= 0) {
3805 WARN_ON(*level < 0);
3806 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3807 cur = path->nodes[*level];
3809 if (btrfs_header_level(cur) != *level)
3812 if (path->slots[*level] >=
3813 btrfs_header_nritems(cur))
3816 /* the new code goes down to level 1 and does all the
3817 * leaves pointed to that node in bulk. So, this check
3818 * for level 0 will always be false.
3820 * But, the disk format allows the drop_snapshot_progress
3821 * field in the root to leave things in a state where
3822 * a leaf will need cleaning up here. If someone crashes
3823 * with the old code and then boots with the new code,
3824 * we might find a leaf here.
3827 ret = btrfs_drop_leaf_ref(trans, root, cur);
3833 * once we get to level one, process the whole node
3834 * at once, including everything below it.
3837 ret = drop_level_one_refs(trans, root, path);
3842 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3843 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3844 blocksize = btrfs_level_size(root, *level - 1);
3846 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3850 * if there is more than one reference, we don't need
3851 * to read that node to drop any references it has. We
3852 * just drop the ref we hold on that node and move on to the
3853 * next slot in this level.
3856 parent = path->nodes[*level];
3857 root_owner = btrfs_header_owner(parent);
3858 root_gen = btrfs_header_generation(parent);
3859 path->slots[*level]++;
3861 ret = __btrfs_free_extent(trans, root, bytenr,
3862 blocksize, parent->start,
3863 root_owner, root_gen,
3867 atomic_inc(&root->fs_info->throttle_gen);
3868 wake_up(&root->fs_info->transaction_throttle);
3875 * we need to keep freeing things in the next level down.
3876 * read the block and loop around to process it
3878 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3879 WARN_ON(*level <= 0);
3880 if (path->nodes[*level-1])
3881 free_extent_buffer(path->nodes[*level-1]);
3882 path->nodes[*level-1] = next;
3883 *level = btrfs_header_level(next);
3884 path->slots[*level] = 0;
3888 WARN_ON(*level < 0);
3889 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3891 if (path->nodes[*level] == root->node) {
3892 parent = path->nodes[*level];
3893 bytenr = path->nodes[*level]->start;
3895 parent = path->nodes[*level + 1];
3896 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3899 blocksize = btrfs_level_size(root, *level);
3900 root_owner = btrfs_header_owner(parent);
3901 root_gen = btrfs_header_generation(parent);
3904 * cleanup and free the reference on the last node
3907 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3908 parent->start, root_owner, root_gen,
3910 free_extent_buffer(path->nodes[*level]);
3911 path->nodes[*level] = NULL;
3921 * helper function for drop_subtree, this function is similar to
3922 * walk_down_tree. The main difference is that it checks reference
3923 * counts while tree blocks are locked.
3925 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3926 struct btrfs_root *root,
3927 struct btrfs_path *path, int *level)
3929 struct extent_buffer *next;
3930 struct extent_buffer *cur;
3931 struct extent_buffer *parent;
3938 cur = path->nodes[*level];
3939 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3945 while (*level >= 0) {
3946 cur = path->nodes[*level];
3948 ret = btrfs_drop_leaf_ref(trans, root, cur);
3950 clean_tree_block(trans, root, cur);
3953 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3954 clean_tree_block(trans, root, cur);
3958 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3959 blocksize = btrfs_level_size(root, *level - 1);
3960 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3962 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3963 btrfs_tree_lock(next);
3964 btrfs_set_lock_blocking(next);
3966 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3970 parent = path->nodes[*level];
3971 ret = btrfs_free_extent(trans, root, bytenr,
3972 blocksize, parent->start,
3973 btrfs_header_owner(parent),
3974 btrfs_header_generation(parent),
3977 path->slots[*level]++;
3978 btrfs_tree_unlock(next);
3979 free_extent_buffer(next);
3983 *level = btrfs_header_level(next);
3984 path->nodes[*level] = next;
3985 path->slots[*level] = 0;
3986 path->locks[*level] = 1;
3990 parent = path->nodes[*level + 1];
3991 bytenr = path->nodes[*level]->start;
3992 blocksize = path->nodes[*level]->len;
3994 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3995 parent->start, btrfs_header_owner(parent),
3996 btrfs_header_generation(parent), *level, 1);
3999 if (path->locks[*level]) {
4000 btrfs_tree_unlock(path->nodes[*level]);
4001 path->locks[*level] = 0;
4003 free_extent_buffer(path->nodes[*level]);
4004 path->nodes[*level] = NULL;
4011 * helper for dropping snapshots. This walks back up the tree in the path
4012 * to find the first node higher up where we haven't yet gone through
4015 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4016 struct btrfs_root *root,
4017 struct btrfs_path *path,
4018 int *level, int max_level)
4022 struct btrfs_root_item *root_item = &root->root_item;
4027 for (i = *level; i < max_level && path->nodes[i]; i++) {
4028 slot = path->slots[i];
4029 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
4030 struct extent_buffer *node;
4031 struct btrfs_disk_key disk_key;
4034 * there is more work to do in this level.
4035 * Update the drop_progress marker to reflect
4036 * the work we've done so far, and then bump
4039 node = path->nodes[i];
4042 WARN_ON(*level == 0);
4043 btrfs_node_key(node, &disk_key, path->slots[i]);
4044 memcpy(&root_item->drop_progress,
4045 &disk_key, sizeof(disk_key));
4046 root_item->drop_level = i;
4049 struct extent_buffer *parent;
4052 * this whole node is done, free our reference
4053 * on it and go up one level
4055 if (path->nodes[*level] == root->node)
4056 parent = path->nodes[*level];
4058 parent = path->nodes[*level + 1];
4060 root_owner = btrfs_header_owner(parent);
4061 root_gen = btrfs_header_generation(parent);
4063 clean_tree_block(trans, root, path->nodes[*level]);
4064 ret = btrfs_free_extent(trans, root,
4065 path->nodes[*level]->start,
4066 path->nodes[*level]->len,
4067 parent->start, root_owner,
4068 root_gen, *level, 1);
4070 if (path->locks[*level]) {
4071 btrfs_tree_unlock(path->nodes[*level]);
4072 path->locks[*level] = 0;
4074 free_extent_buffer(path->nodes[*level]);
4075 path->nodes[*level] = NULL;
4083 * drop the reference count on the tree rooted at 'snap'. This traverses
4084 * the tree freeing any blocks that have a ref count of zero after being
4087 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4093 struct btrfs_path *path;
4096 struct btrfs_root_item *root_item = &root->root_item;
4098 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
4099 path = btrfs_alloc_path();
4102 level = btrfs_header_level(root->node);
4104 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4105 path->nodes[level] = root->node;
4106 extent_buffer_get(root->node);
4107 path->slots[level] = 0;
4109 struct btrfs_key key;
4110 struct btrfs_disk_key found_key;
4111 struct extent_buffer *node;
4113 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4114 level = root_item->drop_level;
4115 path->lowest_level = level;
4116 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4121 node = path->nodes[level];
4122 btrfs_node_key(node, &found_key, path->slots[level]);
4123 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4124 sizeof(found_key)));
4126 * unlock our path, this is safe because only this
4127 * function is allowed to delete this snapshot
4129 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4130 if (path->nodes[i] && path->locks[i]) {
4132 btrfs_tree_unlock(path->nodes[i]);
4137 wret = walk_down_tree(trans, root, path, &level);
4143 wret = walk_up_tree(trans, root, path, &level,
4149 if (trans->transaction->in_commit) {
4153 atomic_inc(&root->fs_info->throttle_gen);
4154 wake_up(&root->fs_info->transaction_throttle);
4156 for (i = 0; i <= orig_level; i++) {
4157 if (path->nodes[i]) {
4158 free_extent_buffer(path->nodes[i]);
4159 path->nodes[i] = NULL;
4163 btrfs_free_path(path);
4167 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
4168 struct btrfs_root *root,
4169 struct extent_buffer *node,
4170 struct extent_buffer *parent)
4172 struct btrfs_path *path;
4178 path = btrfs_alloc_path();
4181 BUG_ON(!btrfs_tree_locked(parent));
4182 parent_level = btrfs_header_level(parent);
4183 extent_buffer_get(parent);
4184 path->nodes[parent_level] = parent;
4185 path->slots[parent_level] = btrfs_header_nritems(parent);
4187 BUG_ON(!btrfs_tree_locked(node));
4188 level = btrfs_header_level(node);
4189 extent_buffer_get(node);
4190 path->nodes[level] = node;
4191 path->slots[level] = 0;
4194 wret = walk_down_subtree(trans, root, path, &level);
4200 wret = walk_up_tree(trans, root, path, &level, parent_level);
4207 btrfs_free_path(path);
4211 static unsigned long calc_ra(unsigned long start, unsigned long last,
4214 return min(last, start + nr - 1);
4217 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
4222 unsigned long first_index;
4223 unsigned long last_index;
4226 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4227 struct file_ra_state *ra;
4228 struct btrfs_ordered_extent *ordered;
4229 unsigned int total_read = 0;
4230 unsigned int total_dirty = 0;
4233 ra = kzalloc(sizeof(*ra), GFP_NOFS);
4235 mutex_lock(&inode->i_mutex);
4236 first_index = start >> PAGE_CACHE_SHIFT;
4237 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4239 /* make sure the dirty trick played by the caller work */
4240 ret = invalidate_inode_pages2_range(inode->i_mapping,
4241 first_index, last_index);
4245 file_ra_state_init(ra, inode->i_mapping);
4247 for (i = first_index ; i <= last_index; i++) {
4248 if (total_read % ra->ra_pages == 0) {
4249 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
4250 calc_ra(i, last_index, ra->ra_pages));
4254 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
4256 page = grab_cache_page(inode->i_mapping, i);
4261 if (!PageUptodate(page)) {
4262 btrfs_readpage(NULL, page);
4264 if (!PageUptodate(page)) {
4266 page_cache_release(page);
4271 wait_on_page_writeback(page);
4273 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4274 page_end = page_start + PAGE_CACHE_SIZE - 1;
4275 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4277 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4279 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4281 page_cache_release(page);
4282 btrfs_start_ordered_extent(inode, ordered, 1);
4283 btrfs_put_ordered_extent(ordered);
4286 set_page_extent_mapped(page);
4288 if (i == first_index)
4289 set_extent_bits(io_tree, page_start, page_end,
4290 EXTENT_BOUNDARY, GFP_NOFS);
4291 btrfs_set_extent_delalloc(inode, page_start, page_end);
4293 set_page_dirty(page);
4296 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4298 page_cache_release(page);
4303 mutex_unlock(&inode->i_mutex);
4304 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4308 static noinline int relocate_data_extent(struct inode *reloc_inode,
4309 struct btrfs_key *extent_key,
4312 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4313 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4314 struct extent_map *em;
4315 u64 start = extent_key->objectid - offset;
4316 u64 end = start + extent_key->offset - 1;
4318 em = alloc_extent_map(GFP_NOFS);
4319 BUG_ON(!em || IS_ERR(em));
4322 em->len = extent_key->offset;
4323 em->block_len = extent_key->offset;
4324 em->block_start = extent_key->objectid;
4325 em->bdev = root->fs_info->fs_devices->latest_bdev;
4326 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4328 /* setup extent map to cheat btrfs_readpage */
4329 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4332 spin_lock(&em_tree->lock);
4333 ret = add_extent_mapping(em_tree, em);
4334 spin_unlock(&em_tree->lock);
4335 if (ret != -EEXIST) {
4336 free_extent_map(em);
4339 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4341 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4343 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4346 struct btrfs_ref_path {
4348 u64 nodes[BTRFS_MAX_LEVEL];
4350 u64 root_generation;
4357 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4358 u64 new_nodes[BTRFS_MAX_LEVEL];
4361 struct disk_extent {
4372 static int is_cowonly_root(u64 root_objectid)
4374 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4375 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4376 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4377 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4378 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4379 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4384 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4385 struct btrfs_root *extent_root,
4386 struct btrfs_ref_path *ref_path,
4389 struct extent_buffer *leaf;
4390 struct btrfs_path *path;
4391 struct btrfs_extent_ref *ref;
4392 struct btrfs_key key;
4393 struct btrfs_key found_key;
4399 path = btrfs_alloc_path();
4404 ref_path->lowest_level = -1;
4405 ref_path->current_level = -1;
4406 ref_path->shared_level = -1;
4410 level = ref_path->current_level - 1;
4411 while (level >= -1) {
4413 if (level < ref_path->lowest_level)
4417 bytenr = ref_path->nodes[level];
4419 bytenr = ref_path->extent_start;
4420 BUG_ON(bytenr == 0);
4422 parent = ref_path->nodes[level + 1];
4423 ref_path->nodes[level + 1] = 0;
4424 ref_path->current_level = level;
4425 BUG_ON(parent == 0);
4427 key.objectid = bytenr;
4428 key.offset = parent + 1;
4429 key.type = BTRFS_EXTENT_REF_KEY;
4431 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4436 leaf = path->nodes[0];
4437 nritems = btrfs_header_nritems(leaf);
4438 if (path->slots[0] >= nritems) {
4439 ret = btrfs_next_leaf(extent_root, path);
4444 leaf = path->nodes[0];
4447 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4448 if (found_key.objectid == bytenr &&
4449 found_key.type == BTRFS_EXTENT_REF_KEY) {
4450 if (level < ref_path->shared_level)
4451 ref_path->shared_level = level;
4456 btrfs_release_path(extent_root, path);
4459 /* reached lowest level */
4463 level = ref_path->current_level;
4464 while (level < BTRFS_MAX_LEVEL - 1) {
4468 bytenr = ref_path->nodes[level];
4470 bytenr = ref_path->extent_start;
4472 BUG_ON(bytenr == 0);
4474 key.objectid = bytenr;
4476 key.type = BTRFS_EXTENT_REF_KEY;
4478 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4482 leaf = path->nodes[0];
4483 nritems = btrfs_header_nritems(leaf);
4484 if (path->slots[0] >= nritems) {
4485 ret = btrfs_next_leaf(extent_root, path);
4489 /* the extent was freed by someone */
4490 if (ref_path->lowest_level == level)
4492 btrfs_release_path(extent_root, path);
4495 leaf = path->nodes[0];
4498 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4499 if (found_key.objectid != bytenr ||
4500 found_key.type != BTRFS_EXTENT_REF_KEY) {
4501 /* the extent was freed by someone */
4502 if (ref_path->lowest_level == level) {
4506 btrfs_release_path(extent_root, path);
4510 ref = btrfs_item_ptr(leaf, path->slots[0],
4511 struct btrfs_extent_ref);
4512 ref_objectid = btrfs_ref_objectid(leaf, ref);
4513 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4515 level = (int)ref_objectid;
4516 BUG_ON(level >= BTRFS_MAX_LEVEL);
4517 ref_path->lowest_level = level;
4518 ref_path->current_level = level;
4519 ref_path->nodes[level] = bytenr;
4521 WARN_ON(ref_objectid != level);
4524 WARN_ON(level != -1);
4528 if (ref_path->lowest_level == level) {
4529 ref_path->owner_objectid = ref_objectid;
4530 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4534 * the block is tree root or the block isn't in reference
4537 if (found_key.objectid == found_key.offset ||
4538 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4539 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4540 ref_path->root_generation =
4541 btrfs_ref_generation(leaf, ref);
4543 /* special reference from the tree log */
4544 ref_path->nodes[0] = found_key.offset;
4545 ref_path->current_level = 0;
4552 BUG_ON(ref_path->nodes[level] != 0);
4553 ref_path->nodes[level] = found_key.offset;
4554 ref_path->current_level = level;
4557 * the reference was created in the running transaction,
4558 * no need to continue walking up.
4560 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4561 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4562 ref_path->root_generation =
4563 btrfs_ref_generation(leaf, ref);
4568 btrfs_release_path(extent_root, path);
4571 /* reached max tree level, but no tree root found. */
4574 btrfs_free_path(path);
4578 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4579 struct btrfs_root *extent_root,
4580 struct btrfs_ref_path *ref_path,
4583 memset(ref_path, 0, sizeof(*ref_path));
4584 ref_path->extent_start = extent_start;
4586 return __next_ref_path(trans, extent_root, ref_path, 1);
4589 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4590 struct btrfs_root *extent_root,
4591 struct btrfs_ref_path *ref_path)
4593 return __next_ref_path(trans, extent_root, ref_path, 0);
4596 static noinline int get_new_locations(struct inode *reloc_inode,
4597 struct btrfs_key *extent_key,
4598 u64 offset, int no_fragment,
4599 struct disk_extent **extents,
4602 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4603 struct btrfs_path *path;
4604 struct btrfs_file_extent_item *fi;
4605 struct extent_buffer *leaf;
4606 struct disk_extent *exts = *extents;
4607 struct btrfs_key found_key;
4612 int max = *nr_extents;
4615 WARN_ON(!no_fragment && *extents);
4618 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4623 path = btrfs_alloc_path();
4626 cur_pos = extent_key->objectid - offset;
4627 last_byte = extent_key->objectid + extent_key->offset;
4628 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4638 leaf = path->nodes[0];
4639 nritems = btrfs_header_nritems(leaf);
4640 if (path->slots[0] >= nritems) {
4641 ret = btrfs_next_leaf(root, path);
4646 leaf = path->nodes[0];
4649 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4650 if (found_key.offset != cur_pos ||
4651 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4652 found_key.objectid != reloc_inode->i_ino)
4655 fi = btrfs_item_ptr(leaf, path->slots[0],
4656 struct btrfs_file_extent_item);
4657 if (btrfs_file_extent_type(leaf, fi) !=
4658 BTRFS_FILE_EXTENT_REG ||
4659 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4663 struct disk_extent *old = exts;
4665 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4666 memcpy(exts, old, sizeof(*exts) * nr);
4667 if (old != *extents)
4671 exts[nr].disk_bytenr =
4672 btrfs_file_extent_disk_bytenr(leaf, fi);
4673 exts[nr].disk_num_bytes =
4674 btrfs_file_extent_disk_num_bytes(leaf, fi);
4675 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4676 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4677 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4678 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4679 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4680 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4682 BUG_ON(exts[nr].offset > 0);
4683 BUG_ON(exts[nr].compression || exts[nr].encryption);
4684 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4686 cur_pos += exts[nr].num_bytes;
4689 if (cur_pos + offset >= last_byte)
4699 BUG_ON(cur_pos + offset > last_byte);
4700 if (cur_pos + offset < last_byte) {
4706 btrfs_free_path(path);
4708 if (exts != *extents)
4717 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4718 struct btrfs_root *root,
4719 struct btrfs_path *path,
4720 struct btrfs_key *extent_key,
4721 struct btrfs_key *leaf_key,
4722 struct btrfs_ref_path *ref_path,
4723 struct disk_extent *new_extents,
4726 struct extent_buffer *leaf;
4727 struct btrfs_file_extent_item *fi;
4728 struct inode *inode = NULL;
4729 struct btrfs_key key;
4734 u64 search_end = (u64)-1;
4737 int extent_locked = 0;
4741 memcpy(&key, leaf_key, sizeof(key));
4742 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4743 if (key.objectid < ref_path->owner_objectid ||
4744 (key.objectid == ref_path->owner_objectid &&
4745 key.type < BTRFS_EXTENT_DATA_KEY)) {
4746 key.objectid = ref_path->owner_objectid;
4747 key.type = BTRFS_EXTENT_DATA_KEY;
4753 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4757 leaf = path->nodes[0];
4758 nritems = btrfs_header_nritems(leaf);
4760 if (extent_locked && ret > 0) {
4762 * the file extent item was modified by someone
4763 * before the extent got locked.
4765 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4766 lock_end, GFP_NOFS);
4770 if (path->slots[0] >= nritems) {
4771 if (++nr_scaned > 2)
4774 BUG_ON(extent_locked);
4775 ret = btrfs_next_leaf(root, path);
4780 leaf = path->nodes[0];
4781 nritems = btrfs_header_nritems(leaf);
4784 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4786 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4787 if ((key.objectid > ref_path->owner_objectid) ||
4788 (key.objectid == ref_path->owner_objectid &&
4789 key.type > BTRFS_EXTENT_DATA_KEY) ||
4790 key.offset >= search_end)
4794 if (inode && key.objectid != inode->i_ino) {
4795 BUG_ON(extent_locked);
4796 btrfs_release_path(root, path);
4797 mutex_unlock(&inode->i_mutex);
4803 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4808 fi = btrfs_item_ptr(leaf, path->slots[0],
4809 struct btrfs_file_extent_item);
4810 extent_type = btrfs_file_extent_type(leaf, fi);
4811 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4812 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4813 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4814 extent_key->objectid)) {
4820 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4821 ext_offset = btrfs_file_extent_offset(leaf, fi);
4823 if (search_end == (u64)-1) {
4824 search_end = key.offset - ext_offset +
4825 btrfs_file_extent_ram_bytes(leaf, fi);
4828 if (!extent_locked) {
4829 lock_start = key.offset;
4830 lock_end = lock_start + num_bytes - 1;
4832 if (lock_start > key.offset ||
4833 lock_end + 1 < key.offset + num_bytes) {
4834 unlock_extent(&BTRFS_I(inode)->io_tree,
4835 lock_start, lock_end, GFP_NOFS);
4841 btrfs_release_path(root, path);
4843 inode = btrfs_iget_locked(root->fs_info->sb,
4844 key.objectid, root);
4845 if (inode->i_state & I_NEW) {
4846 BTRFS_I(inode)->root = root;
4847 BTRFS_I(inode)->location.objectid =
4849 BTRFS_I(inode)->location.type =
4850 BTRFS_INODE_ITEM_KEY;
4851 BTRFS_I(inode)->location.offset = 0;
4852 btrfs_read_locked_inode(inode);
4853 unlock_new_inode(inode);
4856 * some code call btrfs_commit_transaction while
4857 * holding the i_mutex, so we can't use mutex_lock
4860 if (is_bad_inode(inode) ||
4861 !mutex_trylock(&inode->i_mutex)) {
4864 key.offset = (u64)-1;
4869 if (!extent_locked) {
4870 struct btrfs_ordered_extent *ordered;
4872 btrfs_release_path(root, path);
4874 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4875 lock_end, GFP_NOFS);
4876 ordered = btrfs_lookup_first_ordered_extent(inode,
4879 ordered->file_offset <= lock_end &&
4880 ordered->file_offset + ordered->len > lock_start) {
4881 unlock_extent(&BTRFS_I(inode)->io_tree,
4882 lock_start, lock_end, GFP_NOFS);
4883 btrfs_start_ordered_extent(inode, ordered, 1);
4884 btrfs_put_ordered_extent(ordered);
4885 key.offset += num_bytes;
4889 btrfs_put_ordered_extent(ordered);
4895 if (nr_extents == 1) {
4896 /* update extent pointer in place */
4897 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4898 new_extents[0].disk_bytenr);
4899 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4900 new_extents[0].disk_num_bytes);
4901 btrfs_mark_buffer_dirty(leaf);
4903 btrfs_drop_extent_cache(inode, key.offset,
4904 key.offset + num_bytes - 1, 0);
4906 ret = btrfs_inc_extent_ref(trans, root,
4907 new_extents[0].disk_bytenr,
4908 new_extents[0].disk_num_bytes,
4910 root->root_key.objectid,
4915 ret = btrfs_free_extent(trans, root,
4916 extent_key->objectid,
4919 btrfs_header_owner(leaf),
4920 btrfs_header_generation(leaf),
4924 btrfs_release_path(root, path);
4925 key.offset += num_bytes;
4933 * drop old extent pointer at first, then insert the
4934 * new pointers one bye one
4936 btrfs_release_path(root, path);
4937 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4938 key.offset + num_bytes,
4939 key.offset, &alloc_hint);
4942 for (i = 0; i < nr_extents; i++) {
4943 if (ext_offset >= new_extents[i].num_bytes) {
4944 ext_offset -= new_extents[i].num_bytes;
4947 extent_len = min(new_extents[i].num_bytes -
4948 ext_offset, num_bytes);
4950 ret = btrfs_insert_empty_item(trans, root,
4955 leaf = path->nodes[0];
4956 fi = btrfs_item_ptr(leaf, path->slots[0],
4957 struct btrfs_file_extent_item);
4958 btrfs_set_file_extent_generation(leaf, fi,
4960 btrfs_set_file_extent_type(leaf, fi,
4961 BTRFS_FILE_EXTENT_REG);
4962 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4963 new_extents[i].disk_bytenr);
4964 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4965 new_extents[i].disk_num_bytes);
4966 btrfs_set_file_extent_ram_bytes(leaf, fi,
4967 new_extents[i].ram_bytes);
4969 btrfs_set_file_extent_compression(leaf, fi,
4970 new_extents[i].compression);
4971 btrfs_set_file_extent_encryption(leaf, fi,
4972 new_extents[i].encryption);
4973 btrfs_set_file_extent_other_encoding(leaf, fi,
4974 new_extents[i].other_encoding);
4976 btrfs_set_file_extent_num_bytes(leaf, fi,
4978 ext_offset += new_extents[i].offset;
4979 btrfs_set_file_extent_offset(leaf, fi,
4981 btrfs_mark_buffer_dirty(leaf);
4983 btrfs_drop_extent_cache(inode, key.offset,
4984 key.offset + extent_len - 1, 0);
4986 ret = btrfs_inc_extent_ref(trans, root,
4987 new_extents[i].disk_bytenr,
4988 new_extents[i].disk_num_bytes,
4990 root->root_key.objectid,
4991 trans->transid, key.objectid);
4993 btrfs_release_path(root, path);
4995 inode_add_bytes(inode, extent_len);
4998 num_bytes -= extent_len;
4999 key.offset += extent_len;
5004 BUG_ON(i >= nr_extents);
5008 if (extent_locked) {
5009 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5010 lock_end, GFP_NOFS);
5014 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5015 key.offset >= search_end)
5022 btrfs_release_path(root, path);
5024 mutex_unlock(&inode->i_mutex);
5025 if (extent_locked) {
5026 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5027 lock_end, GFP_NOFS);
5034 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5035 struct btrfs_root *root,
5036 struct extent_buffer *buf, u64 orig_start)
5041 BUG_ON(btrfs_header_generation(buf) != trans->transid);
5042 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5044 level = btrfs_header_level(buf);
5046 struct btrfs_leaf_ref *ref;
5047 struct btrfs_leaf_ref *orig_ref;
5049 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5053 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5055 btrfs_free_leaf_ref(root, orig_ref);
5059 ref->nritems = orig_ref->nritems;
5060 memcpy(ref->extents, orig_ref->extents,
5061 sizeof(ref->extents[0]) * ref->nritems);
5063 btrfs_free_leaf_ref(root, orig_ref);
5065 ref->root_gen = trans->transid;
5066 ref->bytenr = buf->start;
5067 ref->owner = btrfs_header_owner(buf);
5068 ref->generation = btrfs_header_generation(buf);
5070 ret = btrfs_add_leaf_ref(root, ref, 0);
5072 btrfs_free_leaf_ref(root, ref);
5077 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5078 struct extent_buffer *leaf,
5079 struct btrfs_block_group_cache *group,
5080 struct btrfs_root *target_root)
5082 struct btrfs_key key;
5083 struct inode *inode = NULL;
5084 struct btrfs_file_extent_item *fi;
5086 u64 skip_objectid = 0;
5090 nritems = btrfs_header_nritems(leaf);
5091 for (i = 0; i < nritems; i++) {
5092 btrfs_item_key_to_cpu(leaf, &key, i);
5093 if (key.objectid == skip_objectid ||
5094 key.type != BTRFS_EXTENT_DATA_KEY)
5096 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5097 if (btrfs_file_extent_type(leaf, fi) ==
5098 BTRFS_FILE_EXTENT_INLINE)
5100 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5102 if (!inode || inode->i_ino != key.objectid) {
5104 inode = btrfs_ilookup(target_root->fs_info->sb,
5105 key.objectid, target_root, 1);
5108 skip_objectid = key.objectid;
5111 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5113 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5114 key.offset + num_bytes - 1, GFP_NOFS);
5115 btrfs_drop_extent_cache(inode, key.offset,
5116 key.offset + num_bytes - 1, 1);
5117 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5118 key.offset + num_bytes - 1, GFP_NOFS);
5125 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5126 struct btrfs_root *root,
5127 struct extent_buffer *leaf,
5128 struct btrfs_block_group_cache *group,
5129 struct inode *reloc_inode)
5131 struct btrfs_key key;
5132 struct btrfs_key extent_key;
5133 struct btrfs_file_extent_item *fi;
5134 struct btrfs_leaf_ref *ref;
5135 struct disk_extent *new_extent;
5144 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
5145 BUG_ON(!new_extent);
5147 ref = btrfs_lookup_leaf_ref(root, leaf->start);
5151 nritems = btrfs_header_nritems(leaf);
5152 for (i = 0; i < nritems; i++) {
5153 btrfs_item_key_to_cpu(leaf, &key, i);
5154 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
5156 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5157 if (btrfs_file_extent_type(leaf, fi) ==
5158 BTRFS_FILE_EXTENT_INLINE)
5160 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5161 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5166 if (bytenr >= group->key.objectid + group->key.offset ||
5167 bytenr + num_bytes <= group->key.objectid)
5170 extent_key.objectid = bytenr;
5171 extent_key.offset = num_bytes;
5172 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
5174 ret = get_new_locations(reloc_inode, &extent_key,
5175 group->key.objectid, 1,
5176 &new_extent, &nr_extent);
5181 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
5182 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
5183 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
5184 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
5186 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5187 new_extent->disk_bytenr);
5188 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5189 new_extent->disk_num_bytes);
5190 btrfs_mark_buffer_dirty(leaf);
5192 ret = btrfs_inc_extent_ref(trans, root,
5193 new_extent->disk_bytenr,
5194 new_extent->disk_num_bytes,
5196 root->root_key.objectid,
5197 trans->transid, key.objectid);
5199 ret = btrfs_free_extent(trans, root,
5200 bytenr, num_bytes, leaf->start,
5201 btrfs_header_owner(leaf),
5202 btrfs_header_generation(leaf),
5208 BUG_ON(ext_index + 1 != ref->nritems);
5209 btrfs_free_leaf_ref(root, ref);
5213 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
5214 struct btrfs_root *root)
5216 struct btrfs_root *reloc_root;
5219 if (root->reloc_root) {
5220 reloc_root = root->reloc_root;
5221 root->reloc_root = NULL;
5222 list_add(&reloc_root->dead_list,
5223 &root->fs_info->dead_reloc_roots);
5225 btrfs_set_root_bytenr(&reloc_root->root_item,
5226 reloc_root->node->start);
5227 btrfs_set_root_level(&root->root_item,
5228 btrfs_header_level(reloc_root->node));
5229 memset(&reloc_root->root_item.drop_progress, 0,
5230 sizeof(struct btrfs_disk_key));
5231 reloc_root->root_item.drop_level = 0;
5233 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5234 &reloc_root->root_key,
5235 &reloc_root->root_item);
5241 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5243 struct btrfs_trans_handle *trans;
5244 struct btrfs_root *reloc_root;
5245 struct btrfs_root *prev_root = NULL;
5246 struct list_head dead_roots;
5250 INIT_LIST_HEAD(&dead_roots);
5251 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5253 while (!list_empty(&dead_roots)) {
5254 reloc_root = list_entry(dead_roots.prev,
5255 struct btrfs_root, dead_list);
5256 list_del_init(&reloc_root->dead_list);
5258 BUG_ON(reloc_root->commit_root != NULL);
5260 trans = btrfs_join_transaction(root, 1);
5263 mutex_lock(&root->fs_info->drop_mutex);
5264 ret = btrfs_drop_snapshot(trans, reloc_root);
5267 mutex_unlock(&root->fs_info->drop_mutex);
5269 nr = trans->blocks_used;
5270 ret = btrfs_end_transaction(trans, root);
5272 btrfs_btree_balance_dirty(root, nr);
5275 free_extent_buffer(reloc_root->node);
5277 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5278 &reloc_root->root_key);
5280 mutex_unlock(&root->fs_info->drop_mutex);
5282 nr = trans->blocks_used;
5283 ret = btrfs_end_transaction(trans, root);
5285 btrfs_btree_balance_dirty(root, nr);
5288 prev_root = reloc_root;
5291 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5297 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5299 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5303 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5305 struct btrfs_root *reloc_root;
5306 struct btrfs_trans_handle *trans;
5307 struct btrfs_key location;
5311 mutex_lock(&root->fs_info->tree_reloc_mutex);
5312 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5314 found = !list_empty(&root->fs_info->dead_reloc_roots);
5315 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5318 trans = btrfs_start_transaction(root, 1);
5320 ret = btrfs_commit_transaction(trans, root);
5324 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5325 location.offset = (u64)-1;
5326 location.type = BTRFS_ROOT_ITEM_KEY;
5328 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5329 BUG_ON(!reloc_root);
5330 btrfs_orphan_cleanup(reloc_root);
5334 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5335 struct btrfs_root *root)
5337 struct btrfs_root *reloc_root;
5338 struct extent_buffer *eb;
5339 struct btrfs_root_item *root_item;
5340 struct btrfs_key root_key;
5343 BUG_ON(!root->ref_cows);
5344 if (root->reloc_root)
5347 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5350 ret = btrfs_copy_root(trans, root, root->commit_root,
5351 &eb, BTRFS_TREE_RELOC_OBJECTID);
5354 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5355 root_key.offset = root->root_key.objectid;
5356 root_key.type = BTRFS_ROOT_ITEM_KEY;
5358 memcpy(root_item, &root->root_item, sizeof(root_item));
5359 btrfs_set_root_refs(root_item, 0);
5360 btrfs_set_root_bytenr(root_item, eb->start);
5361 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5362 btrfs_set_root_generation(root_item, trans->transid);
5364 btrfs_tree_unlock(eb);
5365 free_extent_buffer(eb);
5367 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5368 &root_key, root_item);
5372 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5374 BUG_ON(!reloc_root);
5375 reloc_root->last_trans = trans->transid;
5376 reloc_root->commit_root = NULL;
5377 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5379 root->reloc_root = reloc_root;
5384 * Core function of space balance.
5386 * The idea is using reloc trees to relocate tree blocks in reference
5387 * counted roots. There is one reloc tree for each subvol, and all
5388 * reloc trees share same root key objectid. Reloc trees are snapshots
5389 * of the latest committed roots of subvols (root->commit_root).
5391 * To relocate a tree block referenced by a subvol, there are two steps.
5392 * COW the block through subvol's reloc tree, then update block pointer
5393 * in the subvol to point to the new block. Since all reloc trees share
5394 * same root key objectid, doing special handing for tree blocks owned
5395 * by them is easy. Once a tree block has been COWed in one reloc tree,
5396 * we can use the resulting new block directly when the same block is
5397 * required to COW again through other reloc trees. By this way, relocated
5398 * tree blocks are shared between reloc trees, so they are also shared
5401 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5402 struct btrfs_root *root,
5403 struct btrfs_path *path,
5404 struct btrfs_key *first_key,
5405 struct btrfs_ref_path *ref_path,
5406 struct btrfs_block_group_cache *group,
5407 struct inode *reloc_inode)
5409 struct btrfs_root *reloc_root;
5410 struct extent_buffer *eb = NULL;
5411 struct btrfs_key *keys;
5415 int lowest_level = 0;
5418 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5419 lowest_level = ref_path->owner_objectid;
5421 if (!root->ref_cows) {
5422 path->lowest_level = lowest_level;
5423 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5425 path->lowest_level = 0;
5426 btrfs_release_path(root, path);
5430 mutex_lock(&root->fs_info->tree_reloc_mutex);
5431 ret = init_reloc_tree(trans, root);
5433 reloc_root = root->reloc_root;
5435 shared_level = ref_path->shared_level;
5436 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5438 keys = ref_path->node_keys;
5439 nodes = ref_path->new_nodes;
5440 memset(&keys[shared_level + 1], 0,
5441 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5442 memset(&nodes[shared_level + 1], 0,
5443 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5445 if (nodes[lowest_level] == 0) {
5446 path->lowest_level = lowest_level;
5447 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5450 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5451 eb = path->nodes[level];
5452 if (!eb || eb == reloc_root->node)
5454 nodes[level] = eb->start;
5456 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5458 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5461 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5462 eb = path->nodes[0];
5463 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5464 group, reloc_inode);
5467 btrfs_release_path(reloc_root, path);
5469 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5475 * replace tree blocks in the fs tree with tree blocks in
5478 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5481 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5482 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5485 extent_buffer_get(path->nodes[0]);
5486 eb = path->nodes[0];
5487 btrfs_release_path(reloc_root, path);
5488 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5490 free_extent_buffer(eb);
5493 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5494 path->lowest_level = 0;
5498 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5499 struct btrfs_root *root,
5500 struct btrfs_path *path,
5501 struct btrfs_key *first_key,
5502 struct btrfs_ref_path *ref_path)
5506 ret = relocate_one_path(trans, root, path, first_key,
5507 ref_path, NULL, NULL);
5510 if (root == root->fs_info->extent_root)
5511 btrfs_extent_post_op(trans, root);
5516 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5517 struct btrfs_root *extent_root,
5518 struct btrfs_path *path,
5519 struct btrfs_key *extent_key)
5523 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5526 ret = btrfs_del_item(trans, extent_root, path);
5528 btrfs_release_path(extent_root, path);
5532 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5533 struct btrfs_ref_path *ref_path)
5535 struct btrfs_key root_key;
5537 root_key.objectid = ref_path->root_objectid;
5538 root_key.type = BTRFS_ROOT_ITEM_KEY;
5539 if (is_cowonly_root(ref_path->root_objectid))
5540 root_key.offset = 0;
5542 root_key.offset = (u64)-1;
5544 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5547 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5548 struct btrfs_path *path,
5549 struct btrfs_key *extent_key,
5550 struct btrfs_block_group_cache *group,
5551 struct inode *reloc_inode, int pass)
5553 struct btrfs_trans_handle *trans;
5554 struct btrfs_root *found_root;
5555 struct btrfs_ref_path *ref_path = NULL;
5556 struct disk_extent *new_extents = NULL;
5561 struct btrfs_key first_key;
5565 trans = btrfs_start_transaction(extent_root, 1);
5568 if (extent_key->objectid == 0) {
5569 ret = del_extent_zero(trans, extent_root, path, extent_key);
5573 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5579 for (loops = 0; ; loops++) {
5581 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5582 extent_key->objectid);
5584 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5591 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5592 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5595 found_root = read_ref_root(extent_root->fs_info, ref_path);
5596 BUG_ON(!found_root);
5598 * for reference counted tree, only process reference paths
5599 * rooted at the latest committed root.
5601 if (found_root->ref_cows &&
5602 ref_path->root_generation != found_root->root_key.offset)
5605 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5608 * copy data extents to new locations
5610 u64 group_start = group->key.objectid;
5611 ret = relocate_data_extent(reloc_inode,
5620 level = ref_path->owner_objectid;
5623 if (prev_block != ref_path->nodes[level]) {
5624 struct extent_buffer *eb;
5625 u64 block_start = ref_path->nodes[level];
5626 u64 block_size = btrfs_level_size(found_root, level);
5628 eb = read_tree_block(found_root, block_start,
5630 btrfs_tree_lock(eb);
5631 BUG_ON(level != btrfs_header_level(eb));
5634 btrfs_item_key_to_cpu(eb, &first_key, 0);
5636 btrfs_node_key_to_cpu(eb, &first_key, 0);
5638 btrfs_tree_unlock(eb);
5639 free_extent_buffer(eb);
5640 prev_block = block_start;
5643 btrfs_record_root_in_trans(found_root);
5644 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5646 * try to update data extent references while
5647 * keeping metadata shared between snapshots.
5650 ret = relocate_one_path(trans, found_root,
5651 path, &first_key, ref_path,
5652 group, reloc_inode);
5658 * use fallback method to process the remaining
5662 u64 group_start = group->key.objectid;
5663 new_extents = kmalloc(sizeof(*new_extents),
5666 ret = get_new_locations(reloc_inode,
5674 ret = replace_one_extent(trans, found_root,
5676 &first_key, ref_path,
5677 new_extents, nr_extents);
5679 ret = relocate_tree_block(trans, found_root, path,
5680 &first_key, ref_path);
5687 btrfs_end_transaction(trans, extent_root);
5693 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5696 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5697 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5699 num_devices = root->fs_info->fs_devices->rw_devices;
5700 if (num_devices == 1) {
5701 stripped |= BTRFS_BLOCK_GROUP_DUP;
5702 stripped = flags & ~stripped;
5704 /* turn raid0 into single device chunks */
5705 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5708 /* turn mirroring into duplication */
5709 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5710 BTRFS_BLOCK_GROUP_RAID10))
5711 return stripped | BTRFS_BLOCK_GROUP_DUP;
5714 /* they already had raid on here, just return */
5715 if (flags & stripped)
5718 stripped |= BTRFS_BLOCK_GROUP_DUP;
5719 stripped = flags & ~stripped;
5721 /* switch duplicated blocks with raid1 */
5722 if (flags & BTRFS_BLOCK_GROUP_DUP)
5723 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5725 /* turn single device chunks into raid0 */
5726 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5731 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5732 struct btrfs_block_group_cache *shrink_block_group,
5735 struct btrfs_trans_handle *trans;
5736 u64 new_alloc_flags;
5739 spin_lock(&shrink_block_group->lock);
5740 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5741 spin_unlock(&shrink_block_group->lock);
5743 trans = btrfs_start_transaction(root, 1);
5744 spin_lock(&shrink_block_group->lock);
5746 new_alloc_flags = update_block_group_flags(root,
5747 shrink_block_group->flags);
5748 if (new_alloc_flags != shrink_block_group->flags) {
5750 btrfs_block_group_used(&shrink_block_group->item);
5752 calc = shrink_block_group->key.offset;
5754 spin_unlock(&shrink_block_group->lock);
5756 do_chunk_alloc(trans, root->fs_info->extent_root,
5757 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5759 btrfs_end_transaction(trans, root);
5761 spin_unlock(&shrink_block_group->lock);
5765 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5766 struct btrfs_root *root,
5767 u64 objectid, u64 size)
5769 struct btrfs_path *path;
5770 struct btrfs_inode_item *item;
5771 struct extent_buffer *leaf;
5774 path = btrfs_alloc_path();
5778 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5782 leaf = path->nodes[0];
5783 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5784 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5785 btrfs_set_inode_generation(leaf, item, 1);
5786 btrfs_set_inode_size(leaf, item, size);
5787 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5788 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5789 btrfs_mark_buffer_dirty(leaf);
5790 btrfs_release_path(root, path);
5792 btrfs_free_path(path);
5796 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5797 struct btrfs_block_group_cache *group)
5799 struct inode *inode = NULL;
5800 struct btrfs_trans_handle *trans;
5801 struct btrfs_root *root;
5802 struct btrfs_key root_key;
5803 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5806 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5807 root_key.type = BTRFS_ROOT_ITEM_KEY;
5808 root_key.offset = (u64)-1;
5809 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5811 return ERR_CAST(root);
5813 trans = btrfs_start_transaction(root, 1);
5816 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5820 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5823 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5824 group->key.offset, 0, group->key.offset,
5828 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5829 if (inode->i_state & I_NEW) {
5830 BTRFS_I(inode)->root = root;
5831 BTRFS_I(inode)->location.objectid = objectid;
5832 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5833 BTRFS_I(inode)->location.offset = 0;
5834 btrfs_read_locked_inode(inode);
5835 unlock_new_inode(inode);
5836 BUG_ON(is_bad_inode(inode));
5840 BTRFS_I(inode)->index_cnt = group->key.objectid;
5842 err = btrfs_orphan_add(trans, inode);
5844 btrfs_end_transaction(trans, root);
5848 inode = ERR_PTR(err);
5853 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5856 struct btrfs_ordered_sum *sums;
5857 struct btrfs_sector_sum *sector_sum;
5858 struct btrfs_ordered_extent *ordered;
5859 struct btrfs_root *root = BTRFS_I(inode)->root;
5860 struct list_head list;
5865 INIT_LIST_HEAD(&list);
5867 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5868 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5870 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5871 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5872 disk_bytenr + len - 1, &list);
5874 while (!list_empty(&list)) {
5875 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5876 list_del_init(&sums->list);
5878 sector_sum = sums->sums;
5879 sums->bytenr = ordered->start;
5882 while (offset < sums->len) {
5883 sector_sum->bytenr += ordered->start - disk_bytenr;
5885 offset += root->sectorsize;
5888 btrfs_add_ordered_sum(inode, ordered, sums);
5890 btrfs_put_ordered_extent(ordered);
5894 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5896 struct btrfs_trans_handle *trans;
5897 struct btrfs_path *path;
5898 struct btrfs_fs_info *info = root->fs_info;
5899 struct extent_buffer *leaf;
5900 struct inode *reloc_inode;
5901 struct btrfs_block_group_cache *block_group;
5902 struct btrfs_key key;
5911 root = root->fs_info->extent_root;
5913 block_group = btrfs_lookup_block_group(info, group_start);
5914 BUG_ON(!block_group);
5916 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5917 (unsigned long long)block_group->key.objectid,
5918 (unsigned long long)block_group->flags);
5920 path = btrfs_alloc_path();
5923 reloc_inode = create_reloc_inode(info, block_group);
5924 BUG_ON(IS_ERR(reloc_inode));
5926 __alloc_chunk_for_shrink(root, block_group, 1);
5927 set_block_group_readonly(block_group);
5929 btrfs_start_delalloc_inodes(info->tree_root);
5930 btrfs_wait_ordered_extents(info->tree_root, 0);
5935 key.objectid = block_group->key.objectid;
5938 cur_byte = key.objectid;
5940 trans = btrfs_start_transaction(info->tree_root, 1);
5941 btrfs_commit_transaction(trans, info->tree_root);
5943 mutex_lock(&root->fs_info->cleaner_mutex);
5944 btrfs_clean_old_snapshots(info->tree_root);
5945 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5946 mutex_unlock(&root->fs_info->cleaner_mutex);
5949 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5953 leaf = path->nodes[0];
5954 nritems = btrfs_header_nritems(leaf);
5955 if (path->slots[0] >= nritems) {
5956 ret = btrfs_next_leaf(root, path);
5963 leaf = path->nodes[0];
5964 nritems = btrfs_header_nritems(leaf);
5967 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5969 if (key.objectid >= block_group->key.objectid +
5970 block_group->key.offset)
5973 if (progress && need_resched()) {
5974 btrfs_release_path(root, path);
5981 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5982 key.objectid + key.offset <= cur_byte) {
5988 cur_byte = key.objectid + key.offset;
5989 btrfs_release_path(root, path);
5991 __alloc_chunk_for_shrink(root, block_group, 0);
5992 ret = relocate_one_extent(root, path, &key, block_group,
5998 key.objectid = cur_byte;
6003 btrfs_release_path(root, path);
6006 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6007 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6010 if (total_found > 0) {
6011 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6012 (unsigned long long)total_found, pass);
6014 if (total_found == skipped && pass > 2) {
6016 reloc_inode = create_reloc_inode(info, block_group);
6022 /* delete reloc_inode */
6025 /* unpin extents in this range */
6026 trans = btrfs_start_transaction(info->tree_root, 1);
6027 btrfs_commit_transaction(trans, info->tree_root);
6029 spin_lock(&block_group->lock);
6030 WARN_ON(block_group->pinned > 0);
6031 WARN_ON(block_group->reserved > 0);
6032 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6033 spin_unlock(&block_group->lock);
6034 put_block_group(block_group);
6037 btrfs_free_path(path);
6041 static int find_first_block_group(struct btrfs_root *root,
6042 struct btrfs_path *path, struct btrfs_key *key)
6045 struct btrfs_key found_key;
6046 struct extent_buffer *leaf;
6049 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6054 slot = path->slots[0];
6055 leaf = path->nodes[0];
6056 if (slot >= btrfs_header_nritems(leaf)) {
6057 ret = btrfs_next_leaf(root, path);
6064 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6066 if (found_key.objectid >= key->objectid &&
6067 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6078 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6080 struct btrfs_block_group_cache *block_group;
6083 spin_lock(&info->block_group_cache_lock);
6084 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6085 block_group = rb_entry(n, struct btrfs_block_group_cache,
6087 rb_erase(&block_group->cache_node,
6088 &info->block_group_cache_tree);
6089 spin_unlock(&info->block_group_cache_lock);
6091 btrfs_remove_free_space_cache(block_group);
6092 down_write(&block_group->space_info->groups_sem);
6093 list_del(&block_group->list);
6094 up_write(&block_group->space_info->groups_sem);
6096 WARN_ON(atomic_read(&block_group->count) != 1);
6099 spin_lock(&info->block_group_cache_lock);
6101 spin_unlock(&info->block_group_cache_lock);
6105 int btrfs_read_block_groups(struct btrfs_root *root)
6107 struct btrfs_path *path;
6109 struct btrfs_block_group_cache *cache;
6110 struct btrfs_fs_info *info = root->fs_info;
6111 struct btrfs_space_info *space_info;
6112 struct btrfs_key key;
6113 struct btrfs_key found_key;
6114 struct extent_buffer *leaf;
6116 root = info->extent_root;
6119 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6120 path = btrfs_alloc_path();
6125 ret = find_first_block_group(root, path, &key);
6133 leaf = path->nodes[0];
6134 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6135 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6141 atomic_set(&cache->count, 1);
6142 spin_lock_init(&cache->lock);
6143 mutex_init(&cache->alloc_mutex);
6144 mutex_init(&cache->cache_mutex);
6145 INIT_LIST_HEAD(&cache->list);
6146 read_extent_buffer(leaf, &cache->item,
6147 btrfs_item_ptr_offset(leaf, path->slots[0]),
6148 sizeof(cache->item));
6149 memcpy(&cache->key, &found_key, sizeof(found_key));
6151 key.objectid = found_key.objectid + found_key.offset;
6152 btrfs_release_path(root, path);
6153 cache->flags = btrfs_block_group_flags(&cache->item);
6155 ret = update_space_info(info, cache->flags, found_key.offset,
6156 btrfs_block_group_used(&cache->item),
6159 cache->space_info = space_info;
6160 down_write(&space_info->groups_sem);
6161 list_add_tail(&cache->list, &space_info->block_groups);
6162 up_write(&space_info->groups_sem);
6164 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6167 set_avail_alloc_bits(root->fs_info, cache->flags);
6168 if (btrfs_chunk_readonly(root, cache->key.objectid))
6169 set_block_group_readonly(cache);
6173 btrfs_free_path(path);
6177 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6178 struct btrfs_root *root, u64 bytes_used,
6179 u64 type, u64 chunk_objectid, u64 chunk_offset,
6183 struct btrfs_root *extent_root;
6184 struct btrfs_block_group_cache *cache;
6186 extent_root = root->fs_info->extent_root;
6188 root->fs_info->last_trans_new_blockgroup = trans->transid;
6190 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6194 cache->key.objectid = chunk_offset;
6195 cache->key.offset = size;
6196 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
6197 atomic_set(&cache->count, 1);
6198 spin_lock_init(&cache->lock);
6199 mutex_init(&cache->alloc_mutex);
6200 mutex_init(&cache->cache_mutex);
6201 INIT_LIST_HEAD(&cache->list);
6203 btrfs_set_block_group_used(&cache->item, bytes_used);
6204 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
6205 cache->flags = type;
6206 btrfs_set_block_group_flags(&cache->item, type);
6208 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
6209 &cache->space_info);
6211 down_write(&cache->space_info->groups_sem);
6212 list_add_tail(&cache->list, &cache->space_info->block_groups);
6213 up_write(&cache->space_info->groups_sem);
6215 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6218 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
6219 sizeof(cache->item));
6222 finish_current_insert(trans, extent_root, 0);
6223 ret = del_pending_extents(trans, extent_root, 0);
6225 set_avail_alloc_bits(extent_root->fs_info, type);
6230 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6231 struct btrfs_root *root, u64 group_start)
6233 struct btrfs_path *path;
6234 struct btrfs_block_group_cache *block_group;
6235 struct btrfs_key key;
6238 root = root->fs_info->extent_root;
6240 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6241 BUG_ON(!block_group);
6242 BUG_ON(!block_group->ro);
6244 memcpy(&key, &block_group->key, sizeof(key));
6246 path = btrfs_alloc_path();
6249 spin_lock(&root->fs_info->block_group_cache_lock);
6250 rb_erase(&block_group->cache_node,
6251 &root->fs_info->block_group_cache_tree);
6252 spin_unlock(&root->fs_info->block_group_cache_lock);
6253 btrfs_remove_free_space_cache(block_group);
6254 down_write(&block_group->space_info->groups_sem);
6255 list_del(&block_group->list);
6256 up_write(&block_group->space_info->groups_sem);
6258 spin_lock(&block_group->space_info->lock);
6259 block_group->space_info->total_bytes -= block_group->key.offset;
6260 block_group->space_info->bytes_readonly -= block_group->key.offset;
6261 spin_unlock(&block_group->space_info->lock);
6262 block_group->space_info->full = 0;
6264 put_block_group(block_group);
6265 put_block_group(block_group);
6267 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6273 ret = btrfs_del_item(trans, root, path);
6275 btrfs_free_path(path);