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>
26 #include "print-tree.h"
27 #include "transaction.h"
30 #include "ref-cache.h"
32 #define PENDING_EXTENT_INSERT 0
33 #define PENDING_EXTENT_DELETE 1
34 #define PENDING_BACKREF_UPDATE 2
36 struct pending_extent_op {
45 struct list_head list;
49 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
50 btrfs_root *extent_root, int all);
51 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
52 btrfs_root *extent_root, int all);
53 static struct btrfs_block_group_cache *
54 __btrfs_find_block_group(struct btrfs_root *root,
55 struct btrfs_block_group_cache *hint,
56 u64 search_start, int data, int owner);
57 static int pin_down_bytes(struct btrfs_trans_handle *trans,
58 struct btrfs_root *root,
59 u64 bytenr, u64 num_bytes, int is_data);
60 static int update_block_group(struct btrfs_trans_handle *trans,
61 struct btrfs_root *root,
62 u64 bytenr, u64 num_bytes, int alloc,
65 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
67 return (cache->flags & bits) == bits;
71 * this adds the block group to the fs_info rb tree for the block group
74 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
75 struct btrfs_block_group_cache *block_group)
78 struct rb_node *parent = NULL;
79 struct btrfs_block_group_cache *cache;
81 spin_lock(&info->block_group_cache_lock);
82 p = &info->block_group_cache_tree.rb_node;
86 cache = rb_entry(parent, struct btrfs_block_group_cache,
88 if (block_group->key.objectid < cache->key.objectid) {
90 } else if (block_group->key.objectid > cache->key.objectid) {
93 spin_unlock(&info->block_group_cache_lock);
98 rb_link_node(&block_group->cache_node, parent, p);
99 rb_insert_color(&block_group->cache_node,
100 &info->block_group_cache_tree);
101 spin_unlock(&info->block_group_cache_lock);
107 * This will return the block group at or after bytenr if contains is 0, else
108 * it will return the block group that contains the bytenr
110 static struct btrfs_block_group_cache *
111 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
114 struct btrfs_block_group_cache *cache, *ret = NULL;
118 spin_lock(&info->block_group_cache_lock);
119 n = info->block_group_cache_tree.rb_node;
122 cache = rb_entry(n, struct btrfs_block_group_cache,
124 end = cache->key.objectid + cache->key.offset - 1;
125 start = cache->key.objectid;
127 if (bytenr < start) {
128 if (!contains && (!ret || start < ret->key.objectid))
131 } else if (bytenr > start) {
132 if (contains && bytenr <= end) {
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_lock(block_group, start,
173 start = extent_end + 1;
181 ret = btrfs_add_free_space_lock(block_group, start, size);
184 mutex_unlock(&info->pinned_mutex);
189 static int cache_block_group(struct btrfs_root *root,
190 struct btrfs_block_group_cache *block_group)
192 struct btrfs_path *path;
194 struct btrfs_key key;
195 struct extent_buffer *leaf;
204 root = root->fs_info->extent_root;
206 if (block_group->cached)
209 path = btrfs_alloc_path();
215 * we get into deadlocks with paths held by callers of this function.
216 * since the alloc_mutex is protecting things right now, just
217 * skip the locking here
219 path->skip_locking = 1;
220 first_free = max_t(u64, block_group->key.objectid,
221 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
222 key.objectid = block_group->key.objectid;
224 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
225 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
228 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
232 leaf = path->nodes[0];
233 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
234 if (key.objectid + key.offset > first_free)
235 first_free = key.objectid + key.offset;
238 leaf = path->nodes[0];
239 slot = path->slots[0];
240 if (slot >= btrfs_header_nritems(leaf)) {
241 ret = btrfs_next_leaf(root, path);
249 btrfs_item_key_to_cpu(leaf, &key, slot);
250 if (key.objectid < block_group->key.objectid)
253 if (key.objectid >= block_group->key.objectid +
254 block_group->key.offset)
257 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
263 add_new_free_space(block_group, root->fs_info, last,
266 last = key.objectid + key.offset;
275 add_new_free_space(block_group, root->fs_info, last,
276 block_group->key.objectid +
277 block_group->key.offset);
279 block_group->cached = 1;
282 btrfs_free_path(path);
287 * return the block group that starts at or after bytenr
289 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
293 struct btrfs_block_group_cache *cache;
295 cache = block_group_cache_tree_search(info, bytenr, 0);
301 * return the block group that contains teh given bytenr
303 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
307 struct btrfs_block_group_cache *cache;
309 cache = block_group_cache_tree_search(info, bytenr, 1);
314 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
317 struct list_head *head = &info->space_info;
318 struct list_head *cur;
319 struct btrfs_space_info *found;
320 list_for_each(cur, head) {
321 found = list_entry(cur, struct btrfs_space_info, list);
322 if (found->flags == flags)
328 static u64 div_factor(u64 num, int factor)
337 static struct btrfs_block_group_cache *
338 __btrfs_find_block_group(struct btrfs_root *root,
339 struct btrfs_block_group_cache *hint,
340 u64 search_start, int data, int owner)
342 struct btrfs_block_group_cache *cache;
343 struct btrfs_block_group_cache *found_group = NULL;
344 struct btrfs_fs_info *info = root->fs_info;
352 if (data & BTRFS_BLOCK_GROUP_METADATA)
356 struct btrfs_block_group_cache *shint;
357 shint = btrfs_lookup_first_block_group(info, search_start);
358 if (shint && block_group_bits(shint, data) && !shint->ro) {
359 spin_lock(&shint->lock);
360 used = btrfs_block_group_used(&shint->item);
361 if (used + shint->pinned + shint->reserved <
362 div_factor(shint->key.offset, factor)) {
363 spin_unlock(&shint->lock);
366 spin_unlock(&shint->lock);
369 if (hint && !hint->ro && block_group_bits(hint, data)) {
370 spin_lock(&hint->lock);
371 used = btrfs_block_group_used(&hint->item);
372 if (used + hint->pinned + hint->reserved <
373 div_factor(hint->key.offset, factor)) {
374 spin_unlock(&hint->lock);
377 spin_unlock(&hint->lock);
378 last = hint->key.objectid + hint->key.offset;
381 last = max(hint->key.objectid, search_start);
387 cache = btrfs_lookup_first_block_group(root->fs_info, last);
391 spin_lock(&cache->lock);
392 last = cache->key.objectid + cache->key.offset;
393 used = btrfs_block_group_used(&cache->item);
395 if (!cache->ro && block_group_bits(cache, data)) {
396 free_check = div_factor(cache->key.offset, factor);
397 if (used + cache->pinned + cache->reserved <
400 spin_unlock(&cache->lock);
404 spin_unlock(&cache->lock);
412 if (!full_search && factor < 10) {
422 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
423 struct btrfs_block_group_cache
424 *hint, u64 search_start,
428 struct btrfs_block_group_cache *ret;
429 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
433 /* simple helper to search for an existing extent at a given offset */
434 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
437 struct btrfs_key key;
438 struct btrfs_path *path;
440 path = btrfs_alloc_path();
442 key.objectid = start;
444 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
445 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
447 btrfs_free_path(path);
452 * Back reference rules. Back refs have three main goals:
454 * 1) differentiate between all holders of references to an extent so that
455 * when a reference is dropped we can make sure it was a valid reference
456 * before freeing the extent.
458 * 2) Provide enough information to quickly find the holders of an extent
459 * if we notice a given block is corrupted or bad.
461 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
462 * maintenance. This is actually the same as #2, but with a slightly
463 * different use case.
465 * File extents can be referenced by:
467 * - multiple snapshots, subvolumes, or different generations in one subvol
468 * - different files inside a single subvolume
469 * - different offsets inside a file (bookend extents in file.c)
471 * The extent ref structure has fields for:
473 * - Objectid of the subvolume root
474 * - Generation number of the tree holding the reference
475 * - objectid of the file holding the reference
476 * - number of references holding by parent node (alway 1 for tree blocks)
478 * Btree leaf may hold multiple references to a file extent. In most cases,
479 * these references are from same file and the corresponding offsets inside
480 * the file are close together.
482 * When a file extent is allocated the fields are filled in:
483 * (root_key.objectid, trans->transid, inode objectid, 1)
485 * When a leaf is cow'd new references are added for every file extent found
486 * in the leaf. It looks similar to the create case, but trans->transid will
487 * be different when the block is cow'd.
489 * (root_key.objectid, trans->transid, inode objectid,
490 * number of references in the leaf)
492 * When a file extent is removed either during snapshot deletion or
493 * file truncation, we find the corresponding back reference and check
494 * the following fields:
496 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
499 * Btree extents can be referenced by:
501 * - Different subvolumes
502 * - Different generations of the same subvolume
504 * When a tree block is created, back references are inserted:
506 * (root->root_key.objectid, trans->transid, level, 1)
508 * When a tree block is cow'd, new back references are added for all the
509 * blocks it points to. If the tree block isn't in reference counted root,
510 * the old back references are removed. These new back references are of
511 * the form (trans->transid will have increased since creation):
513 * (root->root_key.objectid, trans->transid, level, 1)
515 * When a backref is in deleting, the following fields are checked:
517 * if backref was for a tree root:
518 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
520 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
522 * Back Reference Key composing:
524 * The key objectid corresponds to the first byte in the extent, the key
525 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
526 * byte of parent extent. If a extent is tree root, the key offset is set
527 * to the key objectid.
530 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
531 struct btrfs_root *root,
532 struct btrfs_path *path,
533 u64 bytenr, u64 parent,
534 u64 ref_root, u64 ref_generation,
535 u64 owner_objectid, int del)
537 struct btrfs_key key;
538 struct btrfs_extent_ref *ref;
539 struct extent_buffer *leaf;
543 key.objectid = bytenr;
544 key.type = BTRFS_EXTENT_REF_KEY;
547 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
555 leaf = path->nodes[0];
556 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
557 ref_objectid = btrfs_ref_objectid(leaf, ref);
558 if (btrfs_ref_root(leaf, ref) != ref_root ||
559 btrfs_ref_generation(leaf, ref) != ref_generation ||
560 (ref_objectid != owner_objectid &&
561 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
572 * updates all the backrefs that are pending on update_list for the
575 static int noinline update_backrefs(struct btrfs_trans_handle *trans,
576 struct btrfs_root *extent_root,
577 struct btrfs_path *path,
578 struct list_head *update_list)
580 struct btrfs_key key;
581 struct btrfs_extent_ref *ref;
582 struct btrfs_fs_info *info = extent_root->fs_info;
583 struct pending_extent_op *op;
584 struct extent_buffer *leaf;
586 struct list_head *cur = update_list->next;
588 u64 ref_root = extent_root->root_key.objectid;
590 op = list_entry(cur, struct pending_extent_op, list);
593 key.objectid = op->bytenr;
594 key.type = BTRFS_EXTENT_REF_KEY;
595 key.offset = op->orig_parent;
597 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
600 leaf = path->nodes[0];
603 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
605 ref_objectid = btrfs_ref_objectid(leaf, ref);
607 if (btrfs_ref_root(leaf, ref) != ref_root ||
608 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
609 (ref_objectid != op->level &&
610 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
611 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
612 "owner %u\n", op->bytenr, op->orig_parent,
613 ref_root, op->level);
614 btrfs_print_leaf(extent_root, leaf);
618 key.objectid = op->bytenr;
619 key.offset = op->parent;
620 key.type = BTRFS_EXTENT_REF_KEY;
621 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
623 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
624 btrfs_set_ref_generation(leaf, ref, op->generation);
628 list_del_init(&op->list);
629 unlock_extent(&info->extent_ins, op->bytenr,
630 op->bytenr + op->num_bytes - 1, GFP_NOFS);
633 if (cur == update_list) {
634 btrfs_mark_buffer_dirty(path->nodes[0]);
635 btrfs_release_path(extent_root, path);
639 op = list_entry(cur, struct pending_extent_op, list);
642 while (path->slots[0] < btrfs_header_nritems(leaf)) {
643 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
644 if (key.objectid == op->bytenr &&
645 key.type == BTRFS_EXTENT_REF_KEY)
650 btrfs_mark_buffer_dirty(path->nodes[0]);
651 btrfs_release_path(extent_root, path);
658 static int noinline insert_extents(struct btrfs_trans_handle *trans,
659 struct btrfs_root *extent_root,
660 struct btrfs_path *path,
661 struct list_head *insert_list, int nr)
663 struct btrfs_key *keys;
665 struct pending_extent_op *op;
666 struct extent_buffer *leaf;
667 struct list_head *cur = insert_list->next;
668 struct btrfs_fs_info *info = extent_root->fs_info;
669 u64 ref_root = extent_root->root_key.objectid;
670 int i = 0, last = 0, ret;
676 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
680 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
686 list_for_each_entry(op, insert_list, list) {
687 keys[i].objectid = op->bytenr;
688 keys[i].offset = op->num_bytes;
689 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
690 data_size[i] = sizeof(struct btrfs_extent_item);
693 keys[i].objectid = op->bytenr;
694 keys[i].offset = op->parent;
695 keys[i].type = BTRFS_EXTENT_REF_KEY;
696 data_size[i] = sizeof(struct btrfs_extent_ref);
700 op = list_entry(cur, struct pending_extent_op, list);
704 ret = btrfs_insert_some_items(trans, extent_root, path,
705 keys+i, data_size+i, total-i);
711 leaf = path->nodes[0];
712 for (c = 0; c < ret; c++) {
713 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
716 * if the first item we inserted was a backref, then
717 * the EXTENT_ITEM will be the odd c's, else it will
720 if ((ref_first && (c % 2)) ||
721 (!ref_first && !(c % 2))) {
722 struct btrfs_extent_item *itm;
724 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
725 struct btrfs_extent_item);
726 btrfs_set_extent_refs(path->nodes[0], itm, 1);
729 struct btrfs_extent_ref *ref;
731 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
732 struct btrfs_extent_ref);
733 btrfs_set_ref_root(leaf, ref, ref_root);
734 btrfs_set_ref_generation(leaf, ref,
736 btrfs_set_ref_objectid(leaf, ref, op->level);
737 btrfs_set_ref_num_refs(leaf, ref, 1);
742 * using del to see when its ok to free up the
743 * pending_extent_op. In the case where we insert the
744 * last item on the list in order to help do batching
745 * we need to not free the extent op until we actually
746 * insert the extent_item
749 unlock_extent(&info->extent_ins, op->bytenr,
750 op->bytenr + op->num_bytes - 1,
753 list_del_init(&op->list);
755 if (cur != insert_list)
757 struct pending_extent_op,
761 btrfs_mark_buffer_dirty(leaf);
762 btrfs_release_path(extent_root, path);
765 * Ok backref's and items usually go right next to eachother,
766 * but if we could only insert 1 item that means that we
767 * inserted on the end of a leaf, and we have no idea what may
768 * be on the next leaf so we just play it safe. In order to
769 * try and help this case we insert the last thing on our
770 * insert list so hopefully it will end up being the last
771 * thing on the leaf and everything else will be before it,
772 * which will let us insert a whole bunch of items at the same
775 if (ret == 1 && !last && (i + ret < total)) {
777 * last: where we will pick up the next time around
778 * i: our current key to insert, will be total - 1
779 * cur: the current op we are screwing with
784 cur = insert_list->prev;
785 op = list_entry(cur, struct pending_extent_op, list);
788 * ok we successfully inserted the last item on the
789 * list, lets reset everything
791 * i: our current key to insert, so where we left off
793 * last: done with this
794 * cur: the op we are messing with
796 * total: since we inserted the last key, we need to
797 * decrement total so we dont overflow
801 cur = insert_list->next;
802 op = list_entry(cur, struct pending_extent_op, list);
816 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
817 struct btrfs_root *root,
818 struct btrfs_path *path,
819 u64 bytenr, u64 parent,
820 u64 ref_root, u64 ref_generation,
823 struct btrfs_key key;
824 struct extent_buffer *leaf;
825 struct btrfs_extent_ref *ref;
829 key.objectid = bytenr;
830 key.type = BTRFS_EXTENT_REF_KEY;
833 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
835 leaf = path->nodes[0];
836 ref = btrfs_item_ptr(leaf, path->slots[0],
837 struct btrfs_extent_ref);
838 btrfs_set_ref_root(leaf, ref, ref_root);
839 btrfs_set_ref_generation(leaf, ref, ref_generation);
840 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
841 btrfs_set_ref_num_refs(leaf, ref, 1);
842 } else if (ret == -EEXIST) {
844 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
845 leaf = path->nodes[0];
846 ref = btrfs_item_ptr(leaf, path->slots[0],
847 struct btrfs_extent_ref);
848 if (btrfs_ref_root(leaf, ref) != ref_root ||
849 btrfs_ref_generation(leaf, ref) != ref_generation) {
855 num_refs = btrfs_ref_num_refs(leaf, ref);
856 BUG_ON(num_refs == 0);
857 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
859 existing_owner = btrfs_ref_objectid(leaf, ref);
860 if (existing_owner != owner_objectid &&
861 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
862 btrfs_set_ref_objectid(leaf, ref,
863 BTRFS_MULTIPLE_OBJECTIDS);
869 btrfs_mark_buffer_dirty(path->nodes[0]);
871 btrfs_release_path(root, path);
875 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
876 struct btrfs_root *root,
877 struct btrfs_path *path)
879 struct extent_buffer *leaf;
880 struct btrfs_extent_ref *ref;
884 leaf = path->nodes[0];
885 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
886 num_refs = btrfs_ref_num_refs(leaf, ref);
887 BUG_ON(num_refs == 0);
890 ret = btrfs_del_item(trans, root, path);
892 btrfs_set_ref_num_refs(leaf, ref, num_refs);
893 btrfs_mark_buffer_dirty(leaf);
895 btrfs_release_path(root, path);
899 static int noinline free_extents(struct btrfs_trans_handle *trans,
900 struct btrfs_root *extent_root,
901 struct list_head *del_list)
903 struct btrfs_fs_info *info = extent_root->fs_info;
904 struct btrfs_path *path;
905 struct btrfs_key key, found_key;
906 struct extent_buffer *leaf;
907 struct list_head *cur;
908 struct pending_extent_op *op;
909 struct btrfs_extent_item *ei;
910 int ret, num_to_del, extent_slot = 0, found_extent = 0;
914 path = btrfs_alloc_path();
920 /* search for the backref for the current ref we want to delete */
921 cur = del_list->next;
922 op = list_entry(cur, struct pending_extent_op, list);
923 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
925 extent_root->root_key.objectid,
926 op->orig_generation, op->level, 1);
928 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
929 "owner %u\n", op->bytenr,
930 extent_root->root_key.objectid, op->orig_generation,
932 btrfs_print_leaf(extent_root, path->nodes[0]);
937 extent_slot = path->slots[0];
942 * if we aren't the first item on the leaf we can move back one and see
943 * if our ref is right next to our extent item
945 if (likely(extent_slot)) {
947 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
949 if (found_key.objectid == op->bytenr &&
950 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
951 found_key.offset == op->num_bytes) {
958 * if we didn't find the extent we need to delete the backref and then
959 * search for the extent item key so we can update its ref count
962 key.objectid = op->bytenr;
963 key.type = BTRFS_EXTENT_ITEM_KEY;
964 key.offset = op->num_bytes;
966 ret = remove_extent_backref(trans, extent_root, path);
968 btrfs_release_path(extent_root, path);
969 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
971 extent_slot = path->slots[0];
974 /* this is where we update the ref count for the extent */
975 leaf = path->nodes[0];
976 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
977 refs = btrfs_extent_refs(leaf, ei);
980 btrfs_set_extent_refs(leaf, ei, refs);
982 btrfs_mark_buffer_dirty(leaf);
985 * This extent needs deleting. The reason cur_slot is extent_slot +
986 * num_to_del is because extent_slot points to the slot where the extent
987 * is, and if the backref was not right next to the extent we will be
988 * deleting at least 1 item, and will want to start searching at the
989 * slot directly next to extent_slot. However if we did find the
990 * backref next to the extent item them we will be deleting at least 2
991 * items and will want to start searching directly after the ref slot
994 struct list_head *pos, *n, *end;
995 int cur_slot = extent_slot+num_to_del;
999 path->slots[0] = extent_slot;
1000 bytes_freed = op->num_bytes;
1003 * we need to see if we can delete multiple things at once, so
1004 * start looping through the list of extents we are wanting to
1005 * delete and see if their extent/backref's are right next to
1006 * eachother and the extents only have 1 ref
1008 for (pos = cur->next; pos != del_list; pos = pos->next) {
1009 struct pending_extent_op *tmp;
1011 tmp = list_entry(pos, struct pending_extent_op, list);
1013 /* we only want to delete extent+ref at this stage */
1014 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1017 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1018 if (found_key.objectid != tmp->bytenr ||
1019 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1020 found_key.offset != tmp->num_bytes)
1023 /* check to make sure this extent only has one ref */
1024 ei = btrfs_item_ptr(leaf, cur_slot,
1025 struct btrfs_extent_item);
1026 if (btrfs_extent_refs(leaf, ei) != 1)
1029 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1030 if (found_key.objectid != tmp->bytenr ||
1031 found_key.type != BTRFS_EXTENT_REF_KEY ||
1032 found_key.offset != tmp->orig_parent)
1036 * the ref is right next to the extent, we can set the
1037 * ref count to 0 since we will delete them both now
1039 btrfs_set_extent_refs(leaf, ei, 0);
1041 /* pin down the bytes for this extent */
1042 mutex_lock(&info->pinned_mutex);
1043 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1044 tmp->num_bytes, tmp->level >=
1045 BTRFS_FIRST_FREE_OBJECTID);
1046 mutex_unlock(&info->pinned_mutex);
1050 * use the del field to tell if we need to go ahead and
1051 * free up the extent when we delete the item or not.
1054 bytes_freed += tmp->num_bytes;
1061 /* update the free space counters */
1062 spin_lock_irq(&info->delalloc_lock);
1063 super_used = btrfs_super_bytes_used(&info->super_copy);
1064 btrfs_set_super_bytes_used(&info->super_copy,
1065 super_used - bytes_freed);
1066 spin_unlock_irq(&info->delalloc_lock);
1068 root_used = btrfs_root_used(&extent_root->root_item);
1069 btrfs_set_root_used(&extent_root->root_item,
1070 root_used - bytes_freed);
1072 /* delete the items */
1073 ret = btrfs_del_items(trans, extent_root, path,
1074 path->slots[0], num_to_del);
1078 * loop through the extents we deleted and do the cleanup work
1081 for (pos = cur, n = pos->next; pos != end;
1082 pos = n, n = pos->next) {
1083 struct pending_extent_op *tmp;
1084 #ifdef BIO_RW_DISCARD
1086 struct btrfs_multi_bio *multi = NULL;
1088 tmp = list_entry(pos, struct pending_extent_op, list);
1091 * remember tmp->del tells us wether or not we pinned
1094 ret = update_block_group(trans, extent_root,
1095 tmp->bytenr, tmp->num_bytes, 0,
1099 #ifdef BIO_RW_DISCARD
1100 ret = btrfs_map_block(&info->mapping_tree, READ,
1101 tmp->bytenr, &map_length, &multi,
1104 struct btrfs_bio_stripe *stripe;
1107 stripe = multi->stripe;
1109 if (map_length > tmp->num_bytes)
1110 map_length = tmp->num_bytes;
1112 for (i = 0; i < multi->num_stripes;
1114 blkdev_issue_discard(stripe->dev->bdev,
1115 stripe->physical >> 9,
1120 list_del_init(&tmp->list);
1121 unlock_extent(&info->extent_ins, tmp->bytenr,
1122 tmp->bytenr + tmp->num_bytes - 1,
1126 } else if (refs && found_extent) {
1128 * the ref and extent were right next to eachother, but the
1129 * extent still has a ref, so just free the backref and keep
1132 ret = remove_extent_backref(trans, extent_root, path);
1135 list_del_init(&op->list);
1136 unlock_extent(&info->extent_ins, op->bytenr,
1137 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1141 * the extent has multiple refs and the backref we were looking
1142 * for was not right next to it, so just unlock and go next,
1145 list_del_init(&op->list);
1146 unlock_extent(&info->extent_ins, op->bytenr,
1147 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1151 btrfs_release_path(extent_root, path);
1152 if (!list_empty(del_list))
1156 btrfs_free_path(path);
1160 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root, u64 bytenr,
1162 u64 orig_parent, u64 parent,
1163 u64 orig_root, u64 ref_root,
1164 u64 orig_generation, u64 ref_generation,
1168 struct btrfs_root *extent_root = root->fs_info->extent_root;
1169 struct btrfs_path *path;
1171 if (root == root->fs_info->extent_root) {
1172 struct pending_extent_op *extent_op;
1175 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1176 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1177 mutex_lock(&root->fs_info->extent_ins_mutex);
1178 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1179 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1181 ret = get_state_private(&root->fs_info->extent_ins,
1184 extent_op = (struct pending_extent_op *)
1185 (unsigned long)priv;
1186 BUG_ON(extent_op->parent != orig_parent);
1187 BUG_ON(extent_op->generation != orig_generation);
1189 extent_op->parent = parent;
1190 extent_op->generation = ref_generation;
1192 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1195 extent_op->type = PENDING_BACKREF_UPDATE;
1196 extent_op->bytenr = bytenr;
1197 extent_op->num_bytes = num_bytes;
1198 extent_op->parent = parent;
1199 extent_op->orig_parent = orig_parent;
1200 extent_op->generation = ref_generation;
1201 extent_op->orig_generation = orig_generation;
1202 extent_op->level = (int)owner_objectid;
1203 INIT_LIST_HEAD(&extent_op->list);
1206 set_extent_bits(&root->fs_info->extent_ins,
1207 bytenr, bytenr + num_bytes - 1,
1208 EXTENT_WRITEBACK, GFP_NOFS);
1209 set_state_private(&root->fs_info->extent_ins,
1210 bytenr, (unsigned long)extent_op);
1212 mutex_unlock(&root->fs_info->extent_ins_mutex);
1216 path = btrfs_alloc_path();
1219 ret = lookup_extent_backref(trans, extent_root, path,
1220 bytenr, orig_parent, orig_root,
1221 orig_generation, owner_objectid, 1);
1224 ret = remove_extent_backref(trans, extent_root, path);
1227 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1228 parent, ref_root, ref_generation,
1231 finish_current_insert(trans, extent_root, 0);
1232 del_pending_extents(trans, extent_root, 0);
1234 btrfs_free_path(path);
1238 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root, u64 bytenr,
1240 u64 orig_parent, u64 parent,
1241 u64 ref_root, u64 ref_generation,
1245 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1246 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1248 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1249 parent, ref_root, ref_root,
1250 ref_generation, ref_generation,
1255 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1256 struct btrfs_root *root, u64 bytenr,
1257 u64 orig_parent, u64 parent,
1258 u64 orig_root, u64 ref_root,
1259 u64 orig_generation, u64 ref_generation,
1262 struct btrfs_path *path;
1264 struct btrfs_key key;
1265 struct extent_buffer *l;
1266 struct btrfs_extent_item *item;
1269 path = btrfs_alloc_path();
1274 key.objectid = bytenr;
1275 key.type = BTRFS_EXTENT_ITEM_KEY;
1276 key.offset = (u64)-1;
1278 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1282 BUG_ON(ret == 0 || path->slots[0] == 0);
1287 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1288 if (key.objectid != bytenr) {
1289 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1290 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1293 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1295 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1296 refs = btrfs_extent_refs(l, item);
1297 btrfs_set_extent_refs(l, item, refs + 1);
1298 btrfs_mark_buffer_dirty(path->nodes[0]);
1300 btrfs_release_path(root->fs_info->extent_root, path);
1303 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1304 path, bytenr, parent,
1305 ref_root, ref_generation,
1308 finish_current_insert(trans, root->fs_info->extent_root, 0);
1309 del_pending_extents(trans, root->fs_info->extent_root, 0);
1311 btrfs_free_path(path);
1315 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1316 struct btrfs_root *root,
1317 u64 bytenr, u64 num_bytes, u64 parent,
1318 u64 ref_root, u64 ref_generation,
1322 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1323 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1325 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1326 0, ref_root, 0, ref_generation,
1331 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1332 struct btrfs_root *root)
1334 finish_current_insert(trans, root->fs_info->extent_root, 1);
1335 del_pending_extents(trans, root->fs_info->extent_root, 1);
1339 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1340 struct btrfs_root *root, u64 bytenr,
1341 u64 num_bytes, u32 *refs)
1343 struct btrfs_path *path;
1345 struct btrfs_key key;
1346 struct extent_buffer *l;
1347 struct btrfs_extent_item *item;
1349 WARN_ON(num_bytes < root->sectorsize);
1350 path = btrfs_alloc_path();
1352 key.objectid = bytenr;
1353 key.offset = num_bytes;
1354 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1355 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1360 btrfs_print_leaf(root, path->nodes[0]);
1361 printk("failed to find block number %Lu\n", bytenr);
1365 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1366 *refs = btrfs_extent_refs(l, item);
1368 btrfs_free_path(path);
1372 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1373 struct btrfs_root *root, u64 bytenr)
1375 struct btrfs_root *extent_root = root->fs_info->extent_root;
1376 struct btrfs_path *path;
1377 struct extent_buffer *leaf;
1378 struct btrfs_extent_ref *ref_item;
1379 struct btrfs_key key;
1380 struct btrfs_key found_key;
1386 key.objectid = bytenr;
1387 key.offset = (u64)-1;
1388 key.type = BTRFS_EXTENT_ITEM_KEY;
1390 path = btrfs_alloc_path();
1391 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1397 if (path->slots[0] == 0)
1401 leaf = path->nodes[0];
1402 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1404 if (found_key.objectid != bytenr ||
1405 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1408 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1410 leaf = path->nodes[0];
1411 nritems = btrfs_header_nritems(leaf);
1412 if (path->slots[0] >= nritems) {
1413 ret = btrfs_next_leaf(extent_root, path);
1420 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1421 if (found_key.objectid != bytenr)
1424 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1429 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_ref);
1431 ref_root = btrfs_ref_root(leaf, ref_item);
1432 if (ref_root != root->root_key.objectid &&
1433 ref_root != BTRFS_TREE_LOG_OBJECTID) {
1437 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1446 btrfs_free_path(path);
1450 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1451 struct extent_buffer *buf, u32 nr_extents)
1453 struct btrfs_key key;
1454 struct btrfs_file_extent_item *fi;
1462 if (!root->ref_cows)
1465 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1467 root_gen = root->root_key.offset;
1470 root_gen = trans->transid - 1;
1473 level = btrfs_header_level(buf);
1474 nritems = btrfs_header_nritems(buf);
1477 struct btrfs_leaf_ref *ref;
1478 struct btrfs_extent_info *info;
1480 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1486 ref->root_gen = root_gen;
1487 ref->bytenr = buf->start;
1488 ref->owner = btrfs_header_owner(buf);
1489 ref->generation = btrfs_header_generation(buf);
1490 ref->nritems = nr_extents;
1491 info = ref->extents;
1493 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1495 btrfs_item_key_to_cpu(buf, &key, i);
1496 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1498 fi = btrfs_item_ptr(buf, i,
1499 struct btrfs_file_extent_item);
1500 if (btrfs_file_extent_type(buf, fi) ==
1501 BTRFS_FILE_EXTENT_INLINE)
1503 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1504 if (disk_bytenr == 0)
1507 info->bytenr = disk_bytenr;
1509 btrfs_file_extent_disk_num_bytes(buf, fi);
1510 info->objectid = key.objectid;
1511 info->offset = key.offset;
1515 ret = btrfs_add_leaf_ref(root, ref, shared);
1516 if (ret == -EEXIST && shared) {
1517 struct btrfs_leaf_ref *old;
1518 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1520 btrfs_remove_leaf_ref(root, old);
1521 btrfs_free_leaf_ref(root, old);
1522 ret = btrfs_add_leaf_ref(root, ref, shared);
1525 btrfs_free_leaf_ref(root, ref);
1531 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1532 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1539 u64 orig_generation;
1541 u32 nr_file_extents = 0;
1542 struct btrfs_key key;
1543 struct btrfs_file_extent_item *fi;
1548 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1549 u64, u64, u64, u64, u64, u64, u64, u64);
1551 ref_root = btrfs_header_owner(buf);
1552 ref_generation = btrfs_header_generation(buf);
1553 orig_root = btrfs_header_owner(orig_buf);
1554 orig_generation = btrfs_header_generation(orig_buf);
1556 nritems = btrfs_header_nritems(buf);
1557 level = btrfs_header_level(buf);
1559 if (root->ref_cows) {
1560 process_func = __btrfs_inc_extent_ref;
1563 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1566 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1568 process_func = __btrfs_update_extent_ref;
1571 for (i = 0; i < nritems; i++) {
1574 btrfs_item_key_to_cpu(buf, &key, i);
1575 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1577 fi = btrfs_item_ptr(buf, i,
1578 struct btrfs_file_extent_item);
1579 if (btrfs_file_extent_type(buf, fi) ==
1580 BTRFS_FILE_EXTENT_INLINE)
1582 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1588 ret = process_func(trans, root, bytenr,
1589 orig_buf->start, buf->start,
1590 orig_root, ref_root,
1591 orig_generation, ref_generation,
1600 bytenr = btrfs_node_blockptr(buf, i);
1601 ret = process_func(trans, root, bytenr,
1602 orig_buf->start, buf->start,
1603 orig_root, ref_root,
1604 orig_generation, ref_generation,
1616 *nr_extents = nr_file_extents;
1618 *nr_extents = nritems;
1626 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root, struct extent_buffer *orig_buf,
1628 struct extent_buffer *buf, int start_slot, int nr)
1635 u64 orig_generation;
1636 struct btrfs_key key;
1637 struct btrfs_file_extent_item *fi;
1643 BUG_ON(start_slot < 0);
1644 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1646 ref_root = btrfs_header_owner(buf);
1647 ref_generation = btrfs_header_generation(buf);
1648 orig_root = btrfs_header_owner(orig_buf);
1649 orig_generation = btrfs_header_generation(orig_buf);
1650 level = btrfs_header_level(buf);
1652 if (!root->ref_cows) {
1654 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1657 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1661 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1664 btrfs_item_key_to_cpu(buf, &key, slot);
1665 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1667 fi = btrfs_item_ptr(buf, slot,
1668 struct btrfs_file_extent_item);
1669 if (btrfs_file_extent_type(buf, fi) ==
1670 BTRFS_FILE_EXTENT_INLINE)
1672 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1675 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1676 orig_buf->start, buf->start,
1677 orig_root, ref_root,
1678 orig_generation, ref_generation,
1683 bytenr = btrfs_node_blockptr(buf, slot);
1684 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1685 orig_buf->start, buf->start,
1686 orig_root, ref_root,
1687 orig_generation, ref_generation,
1699 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1700 struct btrfs_root *root,
1701 struct btrfs_path *path,
1702 struct btrfs_block_group_cache *cache)
1706 struct btrfs_root *extent_root = root->fs_info->extent_root;
1708 struct extent_buffer *leaf;
1710 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1715 leaf = path->nodes[0];
1716 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1717 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1718 btrfs_mark_buffer_dirty(leaf);
1719 btrfs_release_path(extent_root, path);
1721 finish_current_insert(trans, extent_root, 0);
1722 pending_ret = del_pending_extents(trans, extent_root, 0);
1731 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1732 struct btrfs_root *root)
1734 struct btrfs_block_group_cache *cache, *entry;
1738 struct btrfs_path *path;
1741 path = btrfs_alloc_path();
1747 spin_lock(&root->fs_info->block_group_cache_lock);
1748 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1749 n; n = rb_next(n)) {
1750 entry = rb_entry(n, struct btrfs_block_group_cache,
1757 spin_unlock(&root->fs_info->block_group_cache_lock);
1763 last += cache->key.offset;
1765 err = write_one_cache_group(trans, root,
1768 * if we fail to write the cache group, we want
1769 * to keep it marked dirty in hopes that a later
1777 btrfs_free_path(path);
1781 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1782 u64 total_bytes, u64 bytes_used,
1783 struct btrfs_space_info **space_info)
1785 struct btrfs_space_info *found;
1787 found = __find_space_info(info, flags);
1789 spin_lock(&found->lock);
1790 found->total_bytes += total_bytes;
1791 found->bytes_used += bytes_used;
1793 spin_unlock(&found->lock);
1794 *space_info = found;
1797 found = kmalloc(sizeof(*found), GFP_NOFS);
1801 list_add(&found->list, &info->space_info);
1802 INIT_LIST_HEAD(&found->block_groups);
1803 init_rwsem(&found->groups_sem);
1804 spin_lock_init(&found->lock);
1805 found->flags = flags;
1806 found->total_bytes = total_bytes;
1807 found->bytes_used = bytes_used;
1808 found->bytes_pinned = 0;
1809 found->bytes_reserved = 0;
1811 found->force_alloc = 0;
1812 *space_info = found;
1816 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1818 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1819 BTRFS_BLOCK_GROUP_RAID1 |
1820 BTRFS_BLOCK_GROUP_RAID10 |
1821 BTRFS_BLOCK_GROUP_DUP);
1823 if (flags & BTRFS_BLOCK_GROUP_DATA)
1824 fs_info->avail_data_alloc_bits |= extra_flags;
1825 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1826 fs_info->avail_metadata_alloc_bits |= extra_flags;
1827 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1828 fs_info->avail_system_alloc_bits |= extra_flags;
1832 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1834 u64 num_devices = root->fs_info->fs_devices->num_devices;
1836 if (num_devices == 1)
1837 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1838 if (num_devices < 4)
1839 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1841 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1842 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1843 BTRFS_BLOCK_GROUP_RAID10))) {
1844 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1847 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1848 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1849 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1852 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1853 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1854 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1855 (flags & BTRFS_BLOCK_GROUP_DUP)))
1856 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1860 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1861 struct btrfs_root *extent_root, u64 alloc_bytes,
1862 u64 flags, int force)
1864 struct btrfs_space_info *space_info;
1868 int ret = 0, waited = 0;
1870 flags = reduce_alloc_profile(extent_root, flags);
1872 space_info = __find_space_info(extent_root->fs_info, flags);
1874 ret = update_space_info(extent_root->fs_info, flags,
1878 BUG_ON(!space_info);
1880 spin_lock(&space_info->lock);
1881 if (space_info->force_alloc) {
1883 space_info->force_alloc = 0;
1885 if (space_info->full) {
1886 spin_unlock(&space_info->lock);
1890 thresh = div_factor(space_info->total_bytes, 6);
1892 (space_info->bytes_used + space_info->bytes_pinned +
1893 space_info->bytes_reserved + alloc_bytes) < thresh) {
1894 spin_unlock(&space_info->lock);
1898 spin_unlock(&space_info->lock);
1900 ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1901 if (!ret && !force) {
1904 mutex_lock(&extent_root->fs_info->chunk_mutex);
1909 spin_lock(&space_info->lock);
1910 if (space_info->full) {
1911 spin_unlock(&space_info->lock);
1914 spin_unlock(&space_info->lock);
1917 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1919 printk("space info full %Lu\n", flags);
1920 space_info->full = 1;
1924 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1925 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1928 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1933 static int update_block_group(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1935 u64 bytenr, u64 num_bytes, int alloc,
1938 struct btrfs_block_group_cache *cache;
1939 struct btrfs_fs_info *info = root->fs_info;
1940 u64 total = num_bytes;
1945 cache = btrfs_lookup_block_group(info, bytenr);
1948 byte_in_group = bytenr - cache->key.objectid;
1949 WARN_ON(byte_in_group > cache->key.offset);
1951 spin_lock(&cache->space_info->lock);
1952 spin_lock(&cache->lock);
1954 old_val = btrfs_block_group_used(&cache->item);
1955 num_bytes = min(total, cache->key.offset - byte_in_group);
1957 old_val += num_bytes;
1958 cache->space_info->bytes_used += num_bytes;
1959 btrfs_set_block_group_used(&cache->item, old_val);
1960 spin_unlock(&cache->lock);
1961 spin_unlock(&cache->space_info->lock);
1963 old_val -= num_bytes;
1964 cache->space_info->bytes_used -= num_bytes;
1965 btrfs_set_block_group_used(&cache->item, old_val);
1966 spin_unlock(&cache->lock);
1967 spin_unlock(&cache->space_info->lock);
1970 ret = btrfs_add_free_space(cache, bytenr,
1977 bytenr += num_bytes;
1982 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1984 struct btrfs_block_group_cache *cache;
1986 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1990 return cache->key.objectid;
1993 int btrfs_update_pinned_extents(struct btrfs_root *root,
1994 u64 bytenr, u64 num, int pin)
1997 struct btrfs_block_group_cache *cache;
1998 struct btrfs_fs_info *fs_info = root->fs_info;
2000 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2002 set_extent_dirty(&fs_info->pinned_extents,
2003 bytenr, bytenr + num - 1, GFP_NOFS);
2005 clear_extent_dirty(&fs_info->pinned_extents,
2006 bytenr, bytenr + num - 1, GFP_NOFS);
2009 cache = btrfs_lookup_block_group(fs_info, bytenr);
2011 len = min(num, cache->key.offset -
2012 (bytenr - cache->key.objectid));
2014 spin_lock(&cache->space_info->lock);
2015 spin_lock(&cache->lock);
2016 cache->pinned += len;
2017 cache->space_info->bytes_pinned += len;
2018 spin_unlock(&cache->lock);
2019 spin_unlock(&cache->space_info->lock);
2020 fs_info->total_pinned += len;
2022 spin_lock(&cache->space_info->lock);
2023 spin_lock(&cache->lock);
2024 cache->pinned -= len;
2025 cache->space_info->bytes_pinned -= len;
2026 spin_unlock(&cache->lock);
2027 spin_unlock(&cache->space_info->lock);
2028 fs_info->total_pinned -= len;
2036 static int update_reserved_extents(struct btrfs_root *root,
2037 u64 bytenr, u64 num, int reserve)
2040 struct btrfs_block_group_cache *cache;
2041 struct btrfs_fs_info *fs_info = root->fs_info;
2044 cache = btrfs_lookup_block_group(fs_info, bytenr);
2046 len = min(num, cache->key.offset -
2047 (bytenr - cache->key.objectid));
2049 spin_lock(&cache->space_info->lock);
2050 spin_lock(&cache->lock);
2052 cache->reserved += len;
2053 cache->space_info->bytes_reserved += len;
2055 cache->reserved -= len;
2056 cache->space_info->bytes_reserved -= len;
2058 spin_unlock(&cache->lock);
2059 spin_unlock(&cache->space_info->lock);
2066 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2071 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2074 mutex_lock(&root->fs_info->pinned_mutex);
2076 ret = find_first_extent_bit(pinned_extents, last,
2077 &start, &end, EXTENT_DIRTY);
2080 set_extent_dirty(copy, start, end, GFP_NOFS);
2083 mutex_unlock(&root->fs_info->pinned_mutex);
2087 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct extent_io_tree *unpin)
2094 struct btrfs_block_group_cache *cache;
2096 mutex_lock(&root->fs_info->pinned_mutex);
2098 ret = find_first_extent_bit(unpin, 0, &start, &end,
2102 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2103 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2104 cache = btrfs_lookup_block_group(root->fs_info, start);
2106 btrfs_add_free_space(cache, start, end - start + 1);
2107 if (need_resched()) {
2108 mutex_unlock(&root->fs_info->pinned_mutex);
2110 mutex_lock(&root->fs_info->pinned_mutex);
2113 mutex_unlock(&root->fs_info->pinned_mutex);
2117 static int finish_current_insert(struct btrfs_trans_handle *trans,
2118 struct btrfs_root *extent_root, int all)
2125 struct btrfs_fs_info *info = extent_root->fs_info;
2126 struct btrfs_path *path;
2127 struct pending_extent_op *extent_op, *tmp;
2128 struct list_head insert_list, update_list;
2130 int num_inserts = 0, max_inserts;
2132 path = btrfs_alloc_path();
2133 INIT_LIST_HEAD(&insert_list);
2134 INIT_LIST_HEAD(&update_list);
2136 max_inserts = extent_root->leafsize /
2137 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2138 sizeof(struct btrfs_extent_ref) +
2139 sizeof(struct btrfs_extent_item));
2141 mutex_lock(&info->extent_ins_mutex);
2143 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2144 &end, EXTENT_WRITEBACK);
2146 if (skipped && all && !num_inserts) {
2150 mutex_unlock(&info->extent_ins_mutex);
2154 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2158 if (need_resched()) {
2159 mutex_unlock(&info->extent_ins_mutex);
2161 mutex_lock(&info->extent_ins_mutex);
2166 ret = get_state_private(&info->extent_ins, start, &priv);
2168 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2170 if (extent_op->type == PENDING_EXTENT_INSERT) {
2172 list_add_tail(&extent_op->list, &insert_list);
2174 if (num_inserts == max_inserts) {
2175 mutex_unlock(&info->extent_ins_mutex);
2178 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2179 list_add_tail(&extent_op->list, &update_list);
2187 * process teh update list, clear the writeback bit for it, and if
2188 * somebody marked this thing for deletion then just unlock it and be
2189 * done, the free_extents will handle it
2191 mutex_lock(&info->extent_ins_mutex);
2192 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2193 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2194 extent_op->bytenr + extent_op->num_bytes - 1,
2195 EXTENT_WRITEBACK, GFP_NOFS);
2196 if (extent_op->del) {
2197 list_del_init(&extent_op->list);
2198 unlock_extent(&info->extent_ins, extent_op->bytenr,
2199 extent_op->bytenr + extent_op->num_bytes
2204 mutex_unlock(&info->extent_ins_mutex);
2207 * still have things left on the update list, go ahead an update
2210 if (!list_empty(&update_list)) {
2211 ret = update_backrefs(trans, extent_root, path, &update_list);
2216 * if no inserts need to be done, but we skipped some extents and we
2217 * need to make sure everything is cleaned then reset everything and
2218 * go back to the beginning
2220 if (!num_inserts && all && skipped) {
2223 INIT_LIST_HEAD(&update_list);
2224 INIT_LIST_HEAD(&insert_list);
2226 } else if (!num_inserts) {
2231 * process the insert extents list. Again if we are deleting this
2232 * extent, then just unlock it, pin down the bytes if need be, and be
2233 * done with it. Saves us from having to actually insert the extent
2234 * into the tree and then subsequently come along and delete it
2236 mutex_lock(&info->extent_ins_mutex);
2237 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2238 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2239 extent_op->bytenr + extent_op->num_bytes - 1,
2240 EXTENT_WRITEBACK, GFP_NOFS);
2241 if (extent_op->del) {
2242 list_del_init(&extent_op->list);
2243 unlock_extent(&info->extent_ins, extent_op->bytenr,
2244 extent_op->bytenr + extent_op->num_bytes
2247 mutex_lock(&extent_root->fs_info->pinned_mutex);
2248 ret = pin_down_bytes(trans, extent_root,
2250 extent_op->num_bytes, 0);
2251 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2253 ret = update_block_group(trans, extent_root,
2255 extent_op->num_bytes,
2262 mutex_unlock(&info->extent_ins_mutex);
2264 ret = insert_extents(trans, extent_root, path, &insert_list,
2269 * if we broke out of the loop in order to insert stuff because we hit
2270 * the maximum number of inserts at a time we can handle, then loop
2271 * back and pick up where we left off
2273 if (num_inserts == max_inserts) {
2274 INIT_LIST_HEAD(&insert_list);
2275 INIT_LIST_HEAD(&update_list);
2281 * again, if we need to make absolutely sure there are no more pending
2282 * extent operations left and we know that we skipped some, go back to
2283 * the beginning and do it all again
2285 if (all && skipped) {
2286 INIT_LIST_HEAD(&insert_list);
2287 INIT_LIST_HEAD(&update_list);
2294 btrfs_free_path(path);
2298 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 u64 bytenr, u64 num_bytes, int is_data)
2303 struct extent_buffer *buf;
2308 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2312 /* we can reuse a block if it hasn't been written
2313 * and it is from this transaction. We can't
2314 * reuse anything from the tree log root because
2315 * it has tiny sub-transactions.
2317 if (btrfs_buffer_uptodate(buf, 0) &&
2318 btrfs_try_tree_lock(buf)) {
2319 u64 header_owner = btrfs_header_owner(buf);
2320 u64 header_transid = btrfs_header_generation(buf);
2321 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2322 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2323 header_transid == trans->transid &&
2324 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2325 clean_tree_block(NULL, root, buf);
2326 btrfs_tree_unlock(buf);
2327 free_extent_buffer(buf);
2330 btrfs_tree_unlock(buf);
2332 free_extent_buffer(buf);
2334 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2341 * remove an extent from the root, returns 0 on success
2343 static int __free_extent(struct btrfs_trans_handle *trans,
2344 struct btrfs_root *root,
2345 u64 bytenr, u64 num_bytes, u64 parent,
2346 u64 root_objectid, u64 ref_generation,
2347 u64 owner_objectid, int pin, int mark_free)
2349 struct btrfs_path *path;
2350 struct btrfs_key key;
2351 struct btrfs_fs_info *info = root->fs_info;
2352 struct btrfs_root *extent_root = info->extent_root;
2353 struct extent_buffer *leaf;
2355 int extent_slot = 0;
2356 int found_extent = 0;
2358 struct btrfs_extent_item *ei;
2361 key.objectid = bytenr;
2362 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2363 key.offset = num_bytes;
2364 path = btrfs_alloc_path();
2369 ret = lookup_extent_backref(trans, extent_root, path,
2370 bytenr, parent, root_objectid,
2371 ref_generation, owner_objectid, 1);
2373 struct btrfs_key found_key;
2374 extent_slot = path->slots[0];
2375 while(extent_slot > 0) {
2377 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2379 if (found_key.objectid != bytenr)
2381 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2382 found_key.offset == num_bytes) {
2386 if (path->slots[0] - extent_slot > 5)
2389 if (!found_extent) {
2390 ret = remove_extent_backref(trans, extent_root, path);
2392 btrfs_release_path(extent_root, path);
2393 ret = btrfs_search_slot(trans, extent_root,
2396 printk(KERN_ERR "umm, got %d back from search"
2397 ", was looking for %Lu\n", ret,
2399 btrfs_print_leaf(extent_root, path->nodes[0]);
2402 extent_slot = path->slots[0];
2405 btrfs_print_leaf(extent_root, path->nodes[0]);
2407 printk("Unable to find ref byte nr %Lu root %Lu "
2408 "gen %Lu owner %Lu\n", bytenr,
2409 root_objectid, ref_generation, owner_objectid);
2412 leaf = path->nodes[0];
2413 ei = btrfs_item_ptr(leaf, extent_slot,
2414 struct btrfs_extent_item);
2415 refs = btrfs_extent_refs(leaf, ei);
2418 btrfs_set_extent_refs(leaf, ei, refs);
2420 btrfs_mark_buffer_dirty(leaf);
2422 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2423 struct btrfs_extent_ref *ref;
2424 ref = btrfs_item_ptr(leaf, path->slots[0],
2425 struct btrfs_extent_ref);
2426 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2427 /* if the back ref and the extent are next to each other
2428 * they get deleted below in one shot
2430 path->slots[0] = extent_slot;
2432 } else if (found_extent) {
2433 /* otherwise delete the extent back ref */
2434 ret = remove_extent_backref(trans, extent_root, path);
2436 /* if refs are 0, we need to setup the path for deletion */
2438 btrfs_release_path(extent_root, path);
2439 ret = btrfs_search_slot(trans, extent_root, &key, path,
2448 #ifdef BIO_RW_DISCARD
2449 u64 map_length = num_bytes;
2450 struct btrfs_multi_bio *multi = NULL;
2454 mutex_lock(&root->fs_info->pinned_mutex);
2455 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2456 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2457 mutex_unlock(&root->fs_info->pinned_mutex);
2463 /* block accounting for super block */
2464 spin_lock_irq(&info->delalloc_lock);
2465 super_used = btrfs_super_bytes_used(&info->super_copy);
2466 btrfs_set_super_bytes_used(&info->super_copy,
2467 super_used - num_bytes);
2468 spin_unlock_irq(&info->delalloc_lock);
2470 /* block accounting for root item */
2471 root_used = btrfs_root_used(&root->root_item);
2472 btrfs_set_root_used(&root->root_item,
2473 root_used - num_bytes);
2474 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2477 btrfs_release_path(extent_root, path);
2478 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2482 #ifdef BIO_RW_DISCARD
2483 /* Tell the block device(s) that the sectors can be discarded */
2484 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2485 bytenr, &map_length, &multi, 0);
2487 struct btrfs_bio_stripe *stripe = multi->stripes;
2490 if (map_length > num_bytes)
2491 map_length = num_bytes;
2493 for (i = 0; i < multi->num_stripes; i++, stripe++) {
2494 blkdev_issue_discard(stripe->dev->bdev,
2495 stripe->physical >> 9,
2502 btrfs_free_path(path);
2503 finish_current_insert(trans, extent_root, 0);
2508 * find all the blocks marked as pending in the radix tree and remove
2509 * them from the extent map
2511 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2512 btrfs_root *extent_root, int all)
2520 int nr = 0, skipped = 0;
2521 struct extent_io_tree *pending_del;
2522 struct extent_io_tree *extent_ins;
2523 struct pending_extent_op *extent_op;
2524 struct btrfs_fs_info *info = extent_root->fs_info;
2525 struct list_head delete_list;
2527 INIT_LIST_HEAD(&delete_list);
2528 extent_ins = &extent_root->fs_info->extent_ins;
2529 pending_del = &extent_root->fs_info->pending_del;
2532 mutex_lock(&info->extent_ins_mutex);
2534 ret = find_first_extent_bit(pending_del, search, &start, &end,
2537 if (all && skipped && !nr) {
2541 mutex_unlock(&info->extent_ins_mutex);
2545 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2550 if (need_resched()) {
2551 mutex_unlock(&info->extent_ins_mutex);
2553 mutex_lock(&info->extent_ins_mutex);
2560 ret = get_state_private(pending_del, start, &priv);
2562 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2564 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2566 if (!test_range_bit(extent_ins, start, end,
2567 EXTENT_WRITEBACK, 0)) {
2568 list_add_tail(&extent_op->list, &delete_list);
2573 ret = get_state_private(&info->extent_ins, start,
2576 extent_op = (struct pending_extent_op *)
2577 (unsigned long)priv;
2579 clear_extent_bits(&info->extent_ins, start, end,
2580 EXTENT_WRITEBACK, GFP_NOFS);
2582 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2583 list_add_tail(&extent_op->list, &delete_list);
2589 mutex_lock(&extent_root->fs_info->pinned_mutex);
2590 ret = pin_down_bytes(trans, extent_root, start,
2591 end + 1 - start, 0);
2592 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2594 ret = update_block_group(trans, extent_root, start,
2595 end + 1 - start, 0, ret > 0);
2597 unlock_extent(extent_ins, start, end, GFP_NOFS);
2606 if (need_resched()) {
2607 mutex_unlock(&info->extent_ins_mutex);
2609 mutex_lock(&info->extent_ins_mutex);
2614 ret = free_extents(trans, extent_root, &delete_list);
2618 if (all && skipped) {
2619 INIT_LIST_HEAD(&delete_list);
2629 * remove an extent from the root, returns 0 on success
2631 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root,
2633 u64 bytenr, u64 num_bytes, u64 parent,
2634 u64 root_objectid, u64 ref_generation,
2635 u64 owner_objectid, int pin)
2637 struct btrfs_root *extent_root = root->fs_info->extent_root;
2641 WARN_ON(num_bytes < root->sectorsize);
2642 if (root == extent_root) {
2643 struct pending_extent_op *extent_op = NULL;
2645 mutex_lock(&root->fs_info->extent_ins_mutex);
2646 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2647 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2649 ret = get_state_private(&root->fs_info->extent_ins,
2652 extent_op = (struct pending_extent_op *)
2653 (unsigned long)priv;
2656 if (extent_op->type == PENDING_EXTENT_INSERT) {
2657 mutex_unlock(&root->fs_info->extent_ins_mutex);
2663 ref_generation = extent_op->orig_generation;
2664 parent = extent_op->orig_parent;
2667 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2670 extent_op->type = PENDING_EXTENT_DELETE;
2671 extent_op->bytenr = bytenr;
2672 extent_op->num_bytes = num_bytes;
2673 extent_op->parent = parent;
2674 extent_op->orig_parent = parent;
2675 extent_op->generation = ref_generation;
2676 extent_op->orig_generation = ref_generation;
2677 extent_op->level = (int)owner_objectid;
2678 INIT_LIST_HEAD(&extent_op->list);
2681 set_extent_bits(&root->fs_info->pending_del,
2682 bytenr, bytenr + num_bytes - 1,
2683 EXTENT_WRITEBACK, GFP_NOFS);
2684 set_state_private(&root->fs_info->pending_del,
2685 bytenr, (unsigned long)extent_op);
2686 mutex_unlock(&root->fs_info->extent_ins_mutex);
2689 /* if metadata always pin */
2690 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2691 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2692 struct btrfs_block_group_cache *cache;
2694 /* btrfs_free_reserved_extent */
2695 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2697 btrfs_add_free_space(cache, bytenr, num_bytes);
2698 update_reserved_extents(root, bytenr, num_bytes, 0);
2704 /* if data pin when any transaction has committed this */
2705 if (ref_generation != trans->transid)
2708 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2709 root_objectid, ref_generation,
2710 owner_objectid, pin, pin == 0);
2712 finish_current_insert(trans, root->fs_info->extent_root, 0);
2713 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2714 return ret ? ret : pending_ret;
2717 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2718 struct btrfs_root *root,
2719 u64 bytenr, u64 num_bytes, u64 parent,
2720 u64 root_objectid, u64 ref_generation,
2721 u64 owner_objectid, int pin)
2725 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2726 root_objectid, ref_generation,
2727 owner_objectid, pin);
2731 static u64 stripe_align(struct btrfs_root *root, u64 val)
2733 u64 mask = ((u64)root->stripesize - 1);
2734 u64 ret = (val + mask) & ~mask;
2739 * walks the btree of allocated extents and find a hole of a given size.
2740 * The key ins is changed to record the hole:
2741 * ins->objectid == block start
2742 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2743 * ins->offset == number of blocks
2744 * Any available blocks before search_start are skipped.
2746 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *orig_root,
2748 u64 num_bytes, u64 empty_size,
2749 u64 search_start, u64 search_end,
2750 u64 hint_byte, struct btrfs_key *ins,
2751 u64 exclude_start, u64 exclude_nr,
2755 struct btrfs_root * root = orig_root->fs_info->extent_root;
2756 u64 total_needed = num_bytes;
2757 u64 *last_ptr = NULL;
2758 u64 last_wanted = 0;
2759 struct btrfs_block_group_cache *block_group = NULL;
2760 int chunk_alloc_done = 0;
2761 int empty_cluster = 2 * 1024 * 1024;
2762 int allowed_chunk_alloc = 0;
2763 struct list_head *head = NULL, *cur = NULL;
2766 struct btrfs_space_info *space_info;
2768 WARN_ON(num_bytes < root->sectorsize);
2769 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2773 if (orig_root->ref_cows || empty_size)
2774 allowed_chunk_alloc = 1;
2776 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2777 last_ptr = &root->fs_info->last_alloc;
2778 empty_cluster = 64 * 1024;
2781 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2782 last_ptr = &root->fs_info->last_data_alloc;
2786 hint_byte = *last_ptr;
2787 last_wanted = *last_ptr;
2789 empty_size += empty_cluster;
2793 search_start = max(search_start, first_logical_byte(root, 0));
2794 search_start = max(search_start, hint_byte);
2796 if (last_wanted && search_start != last_wanted) {
2798 empty_size += empty_cluster;
2801 total_needed += empty_size;
2802 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2804 block_group = btrfs_lookup_first_block_group(root->fs_info,
2806 space_info = __find_space_info(root->fs_info, data);
2808 down_read(&space_info->groups_sem);
2810 struct btrfs_free_space *free_space;
2812 * the only way this happens if our hint points to a block
2813 * group thats not of the proper type, while looping this
2814 * should never happen
2820 goto new_group_no_lock;
2822 mutex_lock(&block_group->alloc_mutex);
2823 if (unlikely(!block_group_bits(block_group, data)))
2826 ret = cache_block_group(root, block_group);
2828 mutex_unlock(&block_group->alloc_mutex);
2832 if (block_group->ro)
2835 free_space = btrfs_find_free_space(block_group, search_start,
2838 u64 start = block_group->key.objectid;
2839 u64 end = block_group->key.objectid +
2840 block_group->key.offset;
2842 search_start = stripe_align(root, free_space->offset);
2844 /* move on to the next group */
2845 if (search_start + num_bytes >= search_end)
2848 /* move on to the next group */
2849 if (search_start + num_bytes > end)
2852 if (last_wanted && search_start != last_wanted) {
2853 total_needed += empty_cluster;
2854 empty_size += empty_cluster;
2857 * if search_start is still in this block group
2858 * then we just re-search this block group
2860 if (search_start >= start &&
2861 search_start < end) {
2862 mutex_unlock(&block_group->alloc_mutex);
2866 /* else we go to the next block group */
2870 if (exclude_nr > 0 &&
2871 (search_start + num_bytes > exclude_start &&
2872 search_start < exclude_start + exclude_nr)) {
2873 search_start = exclude_start + exclude_nr;
2875 * if search_start is still in this block group
2876 * then we just re-search this block group
2878 if (search_start >= start &&
2879 search_start < end) {
2880 mutex_unlock(&block_group->alloc_mutex);
2885 /* else we go to the next block group */
2889 ins->objectid = search_start;
2890 ins->offset = num_bytes;
2892 btrfs_remove_free_space_lock(block_group, search_start,
2894 /* we are all good, lets return */
2895 mutex_unlock(&block_group->alloc_mutex);
2899 mutex_unlock(&block_group->alloc_mutex);
2901 /* don't try to compare new allocations against the
2902 * last allocation any more
2907 * Here's how this works.
2908 * loop == 0: we were searching a block group via a hint
2909 * and didn't find anything, so we start at
2910 * the head of the block groups and keep searching
2911 * loop == 1: we're searching through all of the block groups
2912 * if we hit the head again we have searched
2913 * all of the block groups for this space and we
2914 * need to try and allocate, if we cant error out.
2915 * loop == 2: we allocated more space and are looping through
2916 * all of the block groups again.
2919 head = &space_info->block_groups;
2922 } else if (loop == 1 && cur == head) {
2925 /* at this point we give up on the empty_size
2926 * allocations and just try to allocate the min
2929 * The extra_loop field was set if an empty_size
2930 * allocation was attempted above, and if this
2931 * is try we need to try the loop again without
2932 * the additional empty_size.
2934 total_needed -= empty_size;
2936 keep_going = extra_loop;
2939 if (allowed_chunk_alloc && !chunk_alloc_done) {
2940 up_read(&space_info->groups_sem);
2941 ret = do_chunk_alloc(trans, root, num_bytes +
2942 2 * 1024 * 1024, data, 1);
2943 down_read(&space_info->groups_sem);
2946 head = &space_info->block_groups;
2948 * we've allocated a new chunk, keep
2952 chunk_alloc_done = 1;
2953 } else if (!allowed_chunk_alloc) {
2954 space_info->force_alloc = 1;
2963 } else if (cur == head) {
2967 block_group = list_entry(cur, struct btrfs_block_group_cache,
2969 search_start = block_group->key.objectid;
2973 /* we found what we needed */
2974 if (ins->objectid) {
2975 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2976 trans->block_group = block_group;
2979 *last_ptr = ins->objectid + ins->offset;
2985 up_read(&space_info->groups_sem);
2989 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2991 struct btrfs_block_group_cache *cache;
2992 struct list_head *l;
2994 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2995 info->total_bytes - info->bytes_used - info->bytes_pinned -
2996 info->bytes_reserved, (info->full) ? "" : "not ");
2998 down_read(&info->groups_sem);
2999 list_for_each(l, &info->block_groups) {
3000 cache = list_entry(l, struct btrfs_block_group_cache, list);
3001 spin_lock(&cache->lock);
3002 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3003 "%Lu pinned %Lu reserved\n",
3004 cache->key.objectid, cache->key.offset,
3005 btrfs_block_group_used(&cache->item),
3006 cache->pinned, cache->reserved);
3007 btrfs_dump_free_space(cache, bytes);
3008 spin_unlock(&cache->lock);
3010 up_read(&info->groups_sem);
3013 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3014 struct btrfs_root *root,
3015 u64 num_bytes, u64 min_alloc_size,
3016 u64 empty_size, u64 hint_byte,
3017 u64 search_end, struct btrfs_key *ins,
3021 u64 search_start = 0;
3023 struct btrfs_fs_info *info = root->fs_info;
3026 alloc_profile = info->avail_data_alloc_bits &
3027 info->data_alloc_profile;
3028 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3029 } else if (root == root->fs_info->chunk_root) {
3030 alloc_profile = info->avail_system_alloc_bits &
3031 info->system_alloc_profile;
3032 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3034 alloc_profile = info->avail_metadata_alloc_bits &
3035 info->metadata_alloc_profile;
3036 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3039 data = reduce_alloc_profile(root, data);
3041 * the only place that sets empty_size is btrfs_realloc_node, which
3042 * is not called recursively on allocations
3044 if (empty_size || root->ref_cows) {
3045 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3046 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3048 BTRFS_BLOCK_GROUP_METADATA |
3049 (info->metadata_alloc_profile &
3050 info->avail_metadata_alloc_bits), 0);
3052 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3053 num_bytes + 2 * 1024 * 1024, data, 0);
3056 WARN_ON(num_bytes < root->sectorsize);
3057 ret = find_free_extent(trans, root, num_bytes, empty_size,
3058 search_start, search_end, hint_byte, ins,
3059 trans->alloc_exclude_start,
3060 trans->alloc_exclude_nr, data);
3062 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3063 num_bytes = num_bytes >> 1;
3064 num_bytes = num_bytes & ~(root->sectorsize - 1);
3065 num_bytes = max(num_bytes, min_alloc_size);
3066 do_chunk_alloc(trans, root->fs_info->extent_root,
3067 num_bytes, data, 1);
3071 struct btrfs_space_info *sinfo;
3073 sinfo = __find_space_info(root->fs_info, data);
3074 printk("allocation failed flags %Lu, wanted %Lu\n",
3076 dump_space_info(sinfo, num_bytes);
3083 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3085 struct btrfs_block_group_cache *cache;
3087 cache = btrfs_lookup_block_group(root->fs_info, start);
3089 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3092 btrfs_add_free_space(cache, start, len);
3093 update_reserved_extents(root, start, len, 0);
3097 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3098 struct btrfs_root *root,
3099 u64 num_bytes, u64 min_alloc_size,
3100 u64 empty_size, u64 hint_byte,
3101 u64 search_end, struct btrfs_key *ins,
3105 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3106 empty_size, hint_byte, search_end, ins,
3108 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3112 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3113 struct btrfs_root *root, u64 parent,
3114 u64 root_objectid, u64 ref_generation,
3115 u64 owner, struct btrfs_key *ins)
3121 u64 num_bytes = ins->offset;
3123 struct btrfs_fs_info *info = root->fs_info;
3124 struct btrfs_root *extent_root = info->extent_root;
3125 struct btrfs_extent_item *extent_item;
3126 struct btrfs_extent_ref *ref;
3127 struct btrfs_path *path;
3128 struct btrfs_key keys[2];
3131 parent = ins->objectid;
3133 /* block accounting for super block */
3134 spin_lock_irq(&info->delalloc_lock);
3135 super_used = btrfs_super_bytes_used(&info->super_copy);
3136 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3137 spin_unlock_irq(&info->delalloc_lock);
3139 /* block accounting for root item */
3140 root_used = btrfs_root_used(&root->root_item);
3141 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3143 if (root == extent_root) {
3144 struct pending_extent_op *extent_op;
3146 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3149 extent_op->type = PENDING_EXTENT_INSERT;
3150 extent_op->bytenr = ins->objectid;
3151 extent_op->num_bytes = ins->offset;
3152 extent_op->parent = parent;
3153 extent_op->orig_parent = 0;
3154 extent_op->generation = ref_generation;
3155 extent_op->orig_generation = 0;
3156 extent_op->level = (int)owner;
3157 INIT_LIST_HEAD(&extent_op->list);
3160 mutex_lock(&root->fs_info->extent_ins_mutex);
3161 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3162 ins->objectid + ins->offset - 1,
3163 EXTENT_WRITEBACK, GFP_NOFS);
3164 set_state_private(&root->fs_info->extent_ins,
3165 ins->objectid, (unsigned long)extent_op);
3166 mutex_unlock(&root->fs_info->extent_ins_mutex);
3170 memcpy(&keys[0], ins, sizeof(*ins));
3171 keys[1].objectid = ins->objectid;
3172 keys[1].type = BTRFS_EXTENT_REF_KEY;
3173 keys[1].offset = parent;
3174 sizes[0] = sizeof(*extent_item);
3175 sizes[1] = sizeof(*ref);
3177 path = btrfs_alloc_path();
3180 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3184 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3185 struct btrfs_extent_item);
3186 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3187 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3188 struct btrfs_extent_ref);
3190 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3191 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3192 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3193 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3195 btrfs_mark_buffer_dirty(path->nodes[0]);
3197 trans->alloc_exclude_start = 0;
3198 trans->alloc_exclude_nr = 0;
3199 btrfs_free_path(path);
3200 finish_current_insert(trans, extent_root, 0);
3201 pending_ret = del_pending_extents(trans, extent_root, 0);
3211 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3213 printk("update block group failed for %Lu %Lu\n",
3214 ins->objectid, ins->offset);
3221 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3222 struct btrfs_root *root, u64 parent,
3223 u64 root_objectid, u64 ref_generation,
3224 u64 owner, struct btrfs_key *ins)
3228 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3230 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3231 ref_generation, owner, ins);
3232 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3237 * this is used by the tree logging recovery code. It records that
3238 * an extent has been allocated and makes sure to clear the free
3239 * space cache bits as well
3241 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3242 struct btrfs_root *root, u64 parent,
3243 u64 root_objectid, u64 ref_generation,
3244 u64 owner, struct btrfs_key *ins)
3247 struct btrfs_block_group_cache *block_group;
3249 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3250 mutex_lock(&block_group->alloc_mutex);
3251 cache_block_group(root, block_group);
3253 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
3255 mutex_unlock(&block_group->alloc_mutex);
3257 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3258 ref_generation, owner, ins);
3263 * finds a free extent and does all the dirty work required for allocation
3264 * returns the key for the extent through ins, and a tree buffer for
3265 * the first block of the extent through buf.
3267 * returns 0 if everything worked, non-zero otherwise.
3269 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3270 struct btrfs_root *root,
3271 u64 num_bytes, u64 parent, u64 min_alloc_size,
3272 u64 root_objectid, u64 ref_generation,
3273 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3274 u64 search_end, struct btrfs_key *ins, u64 data)
3278 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3279 min_alloc_size, empty_size, hint_byte,
3280 search_end, ins, data);
3282 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3283 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3284 root_objectid, ref_generation,
3285 owner_objectid, ins);
3289 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3294 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3295 struct btrfs_root *root,
3296 u64 bytenr, u32 blocksize)
3298 struct extent_buffer *buf;
3300 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3302 return ERR_PTR(-ENOMEM);
3303 btrfs_set_header_generation(buf, trans->transid);
3304 btrfs_tree_lock(buf);
3305 clean_tree_block(trans, root, buf);
3306 btrfs_set_buffer_uptodate(buf);
3307 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3308 set_extent_dirty(&root->dirty_log_pages, buf->start,
3309 buf->start + buf->len - 1, GFP_NOFS);
3311 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3312 buf->start + buf->len - 1, GFP_NOFS);
3314 trans->blocks_used++;
3319 * helper function to allocate a block for a given tree
3320 * returns the tree buffer or NULL.
3322 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3323 struct btrfs_root *root,
3324 u32 blocksize, u64 parent,
3331 struct btrfs_key ins;
3333 struct extent_buffer *buf;
3335 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3336 root_objectid, ref_generation, level,
3337 empty_size, hint, (u64)-1, &ins, 0);
3340 return ERR_PTR(ret);
3343 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3347 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3348 struct btrfs_root *root, struct extent_buffer *leaf)
3351 u64 leaf_generation;
3352 struct btrfs_key key;
3353 struct btrfs_file_extent_item *fi;
3358 BUG_ON(!btrfs_is_leaf(leaf));
3359 nritems = btrfs_header_nritems(leaf);
3360 leaf_owner = btrfs_header_owner(leaf);
3361 leaf_generation = btrfs_header_generation(leaf);
3363 for (i = 0; i < nritems; i++) {
3367 btrfs_item_key_to_cpu(leaf, &key, i);
3368 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3370 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3371 if (btrfs_file_extent_type(leaf, fi) ==
3372 BTRFS_FILE_EXTENT_INLINE)
3375 * FIXME make sure to insert a trans record that
3376 * repeats the snapshot del on crash
3378 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3379 if (disk_bytenr == 0)
3382 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3383 btrfs_file_extent_disk_num_bytes(leaf, fi),
3384 leaf->start, leaf_owner, leaf_generation,
3388 atomic_inc(&root->fs_info->throttle_gen);
3389 wake_up(&root->fs_info->transaction_throttle);
3395 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3396 struct btrfs_root *root,
3397 struct btrfs_leaf_ref *ref)
3401 struct btrfs_extent_info *info = ref->extents;
3403 for (i = 0; i < ref->nritems; i++) {
3404 ret = __btrfs_free_extent(trans, root, info->bytenr,
3405 info->num_bytes, ref->bytenr,
3406 ref->owner, ref->generation,
3409 atomic_inc(&root->fs_info->throttle_gen);
3410 wake_up(&root->fs_info->transaction_throttle);
3420 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3425 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3428 #if 0 // some debugging code in case we see problems here
3429 /* if the refs count is one, it won't get increased again. But
3430 * if the ref count is > 1, someone may be decreasing it at
3431 * the same time we are.
3434 struct extent_buffer *eb = NULL;
3435 eb = btrfs_find_create_tree_block(root, start, len);
3437 btrfs_tree_lock(eb);
3439 mutex_lock(&root->fs_info->alloc_mutex);
3440 ret = lookup_extent_ref(NULL, root, start, len, refs);
3442 mutex_unlock(&root->fs_info->alloc_mutex);
3445 btrfs_tree_unlock(eb);
3446 free_extent_buffer(eb);
3449 printk("block %llu went down to one during drop_snap\n",
3450 (unsigned long long)start);
3461 * helper function for drop_snapshot, this walks down the tree dropping ref
3462 * counts as it goes.
3464 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3465 struct btrfs_root *root,
3466 struct btrfs_path *path, int *level)
3472 struct extent_buffer *next;
3473 struct extent_buffer *cur;
3474 struct extent_buffer *parent;
3475 struct btrfs_leaf_ref *ref;
3480 WARN_ON(*level < 0);
3481 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3482 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3483 path->nodes[*level]->len, &refs);
3489 * walk down to the last node level and free all the leaves
3491 while(*level >= 0) {
3492 WARN_ON(*level < 0);
3493 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3494 cur = path->nodes[*level];
3496 if (btrfs_header_level(cur) != *level)
3499 if (path->slots[*level] >=
3500 btrfs_header_nritems(cur))
3503 ret = btrfs_drop_leaf_ref(trans, root, cur);
3507 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3508 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3509 blocksize = btrfs_level_size(root, *level - 1);
3511 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3514 parent = path->nodes[*level];
3515 root_owner = btrfs_header_owner(parent);
3516 root_gen = btrfs_header_generation(parent);
3517 path->slots[*level]++;
3519 ret = __btrfs_free_extent(trans, root, bytenr,
3520 blocksize, parent->start,
3521 root_owner, root_gen,
3525 atomic_inc(&root->fs_info->throttle_gen);
3526 wake_up(&root->fs_info->transaction_throttle);
3532 * at this point, we have a single ref, and since the
3533 * only place referencing this extent is a dead root
3534 * the reference count should never go higher.
3535 * So, we don't need to check it again
3538 ref = btrfs_lookup_leaf_ref(root, bytenr);
3539 if (ref && ref->generation != ptr_gen) {
3540 btrfs_free_leaf_ref(root, ref);
3544 ret = cache_drop_leaf_ref(trans, root, ref);
3546 btrfs_remove_leaf_ref(root, ref);
3547 btrfs_free_leaf_ref(root, ref);
3551 if (printk_ratelimit()) {
3552 printk("leaf ref miss for bytenr %llu\n",
3553 (unsigned long long)bytenr);
3556 next = btrfs_find_tree_block(root, bytenr, blocksize);
3557 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3558 free_extent_buffer(next);
3560 next = read_tree_block(root, bytenr, blocksize,
3565 * this is a debugging check and can go away
3566 * the ref should never go all the way down to 1
3569 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3575 WARN_ON(*level <= 0);
3576 if (path->nodes[*level-1])
3577 free_extent_buffer(path->nodes[*level-1]);
3578 path->nodes[*level-1] = next;
3579 *level = btrfs_header_level(next);
3580 path->slots[*level] = 0;
3584 WARN_ON(*level < 0);
3585 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3587 if (path->nodes[*level] == root->node) {
3588 parent = path->nodes[*level];
3589 bytenr = path->nodes[*level]->start;
3591 parent = path->nodes[*level + 1];
3592 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3595 blocksize = btrfs_level_size(root, *level);
3596 root_owner = btrfs_header_owner(parent);
3597 root_gen = btrfs_header_generation(parent);
3599 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3600 parent->start, root_owner, root_gen,
3602 free_extent_buffer(path->nodes[*level]);
3603 path->nodes[*level] = NULL;
3612 * helper function for drop_subtree, this function is similar to
3613 * walk_down_tree. The main difference is that it checks reference
3614 * counts while tree blocks are locked.
3616 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *root,
3618 struct btrfs_path *path, int *level)
3620 struct extent_buffer *next;
3621 struct extent_buffer *cur;
3622 struct extent_buffer *parent;
3629 cur = path->nodes[*level];
3630 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3636 while (*level >= 0) {
3637 cur = path->nodes[*level];
3639 ret = btrfs_drop_leaf_ref(trans, root, cur);
3641 clean_tree_block(trans, root, cur);
3644 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3645 clean_tree_block(trans, root, cur);
3649 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3650 blocksize = btrfs_level_size(root, *level - 1);
3651 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3653 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3654 btrfs_tree_lock(next);
3656 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3660 parent = path->nodes[*level];
3661 ret = btrfs_free_extent(trans, root, bytenr,
3662 blocksize, parent->start,
3663 btrfs_header_owner(parent),
3664 btrfs_header_generation(parent),
3667 path->slots[*level]++;
3668 btrfs_tree_unlock(next);
3669 free_extent_buffer(next);
3673 *level = btrfs_header_level(next);
3674 path->nodes[*level] = next;
3675 path->slots[*level] = 0;
3676 path->locks[*level] = 1;
3680 parent = path->nodes[*level + 1];
3681 bytenr = path->nodes[*level]->start;
3682 blocksize = path->nodes[*level]->len;
3684 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3685 parent->start, btrfs_header_owner(parent),
3686 btrfs_header_generation(parent), *level, 1);
3689 if (path->locks[*level]) {
3690 btrfs_tree_unlock(path->nodes[*level]);
3691 path->locks[*level] = 0;
3693 free_extent_buffer(path->nodes[*level]);
3694 path->nodes[*level] = NULL;
3701 * helper for dropping snapshots. This walks back up the tree in the path
3702 * to find the first node higher up where we haven't yet gone through
3705 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3706 struct btrfs_root *root,
3707 struct btrfs_path *path,
3708 int *level, int max_level)
3712 struct btrfs_root_item *root_item = &root->root_item;
3717 for (i = *level; i < max_level && path->nodes[i]; i++) {
3718 slot = path->slots[i];
3719 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3720 struct extent_buffer *node;
3721 struct btrfs_disk_key disk_key;
3722 node = path->nodes[i];
3725 WARN_ON(*level == 0);
3726 btrfs_node_key(node, &disk_key, path->slots[i]);
3727 memcpy(&root_item->drop_progress,
3728 &disk_key, sizeof(disk_key));
3729 root_item->drop_level = i;
3732 struct extent_buffer *parent;
3733 if (path->nodes[*level] == root->node)
3734 parent = path->nodes[*level];
3736 parent = path->nodes[*level + 1];
3738 root_owner = btrfs_header_owner(parent);
3739 root_gen = btrfs_header_generation(parent);
3741 clean_tree_block(trans, root, path->nodes[*level]);
3742 ret = btrfs_free_extent(trans, root,
3743 path->nodes[*level]->start,
3744 path->nodes[*level]->len,
3745 parent->start, root_owner,
3746 root_gen, *level, 1);
3748 if (path->locks[*level]) {
3749 btrfs_tree_unlock(path->nodes[*level]);
3750 path->locks[*level] = 0;
3752 free_extent_buffer(path->nodes[*level]);
3753 path->nodes[*level] = NULL;
3761 * drop the reference count on the tree rooted at 'snap'. This traverses
3762 * the tree freeing any blocks that have a ref count of zero after being
3765 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3771 struct btrfs_path *path;
3774 struct btrfs_root_item *root_item = &root->root_item;
3776 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3777 path = btrfs_alloc_path();
3780 level = btrfs_header_level(root->node);
3782 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3783 path->nodes[level] = root->node;
3784 extent_buffer_get(root->node);
3785 path->slots[level] = 0;
3787 struct btrfs_key key;
3788 struct btrfs_disk_key found_key;
3789 struct extent_buffer *node;
3791 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3792 level = root_item->drop_level;
3793 path->lowest_level = level;
3794 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3799 node = path->nodes[level];
3800 btrfs_node_key(node, &found_key, path->slots[level]);
3801 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3802 sizeof(found_key)));
3804 * unlock our path, this is safe because only this
3805 * function is allowed to delete this snapshot
3807 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3808 if (path->nodes[i] && path->locks[i]) {
3810 btrfs_tree_unlock(path->nodes[i]);
3815 wret = walk_down_tree(trans, root, path, &level);
3821 wret = walk_up_tree(trans, root, path, &level,
3827 if (trans->transaction->in_commit) {
3831 atomic_inc(&root->fs_info->throttle_gen);
3832 wake_up(&root->fs_info->transaction_throttle);
3834 for (i = 0; i <= orig_level; i++) {
3835 if (path->nodes[i]) {
3836 free_extent_buffer(path->nodes[i]);
3837 path->nodes[i] = NULL;
3841 btrfs_free_path(path);
3845 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3846 struct btrfs_root *root,
3847 struct extent_buffer *node,
3848 struct extent_buffer *parent)
3850 struct btrfs_path *path;
3856 path = btrfs_alloc_path();
3859 BUG_ON(!btrfs_tree_locked(parent));
3860 parent_level = btrfs_header_level(parent);
3861 extent_buffer_get(parent);
3862 path->nodes[parent_level] = parent;
3863 path->slots[parent_level] = btrfs_header_nritems(parent);
3865 BUG_ON(!btrfs_tree_locked(node));
3866 level = btrfs_header_level(node);
3867 extent_buffer_get(node);
3868 path->nodes[level] = node;
3869 path->slots[level] = 0;
3872 wret = walk_down_subtree(trans, root, path, &level);
3878 wret = walk_up_tree(trans, root, path, &level, parent_level);
3885 btrfs_free_path(path);
3889 static unsigned long calc_ra(unsigned long start, unsigned long last,
3892 return min(last, start + nr - 1);
3895 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3900 unsigned long first_index;
3901 unsigned long last_index;
3904 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3905 struct file_ra_state *ra;
3906 struct btrfs_ordered_extent *ordered;
3907 unsigned int total_read = 0;
3908 unsigned int total_dirty = 0;
3911 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3913 mutex_lock(&inode->i_mutex);
3914 first_index = start >> PAGE_CACHE_SHIFT;
3915 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3917 /* make sure the dirty trick played by the caller work */
3918 ret = invalidate_inode_pages2_range(inode->i_mapping,
3919 first_index, last_index);
3923 file_ra_state_init(ra, inode->i_mapping);
3925 for (i = first_index ; i <= last_index; i++) {
3926 if (total_read % ra->ra_pages == 0) {
3927 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3928 calc_ra(i, last_index, ra->ra_pages));
3932 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3934 page = grab_cache_page(inode->i_mapping, i);
3939 if (!PageUptodate(page)) {
3940 btrfs_readpage(NULL, page);
3942 if (!PageUptodate(page)) {
3944 page_cache_release(page);
3949 wait_on_page_writeback(page);
3951 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3952 page_end = page_start + PAGE_CACHE_SIZE - 1;
3953 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3955 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3957 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3959 page_cache_release(page);
3960 btrfs_start_ordered_extent(inode, ordered, 1);
3961 btrfs_put_ordered_extent(ordered);
3964 set_page_extent_mapped(page);
3966 btrfs_set_extent_delalloc(inode, page_start, page_end);
3967 if (i == first_index)
3968 set_extent_bits(io_tree, page_start, page_end,
3969 EXTENT_BOUNDARY, GFP_NOFS);
3971 set_page_dirty(page);
3974 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3976 page_cache_release(page);
3981 mutex_unlock(&inode->i_mutex);
3982 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3986 static int noinline relocate_data_extent(struct inode *reloc_inode,
3987 struct btrfs_key *extent_key,
3990 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3991 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3992 struct extent_map *em;
3993 u64 start = extent_key->objectid - offset;
3994 u64 end = start + extent_key->offset - 1;
3996 em = alloc_extent_map(GFP_NOFS);
3997 BUG_ON(!em || IS_ERR(em));
4000 em->len = extent_key->offset;
4001 em->block_len = extent_key->offset;
4002 em->block_start = extent_key->objectid;
4003 em->bdev = root->fs_info->fs_devices->latest_bdev;
4004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4006 /* setup extent map to cheat btrfs_readpage */
4007 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4010 spin_lock(&em_tree->lock);
4011 ret = add_extent_mapping(em_tree, em);
4012 spin_unlock(&em_tree->lock);
4013 if (ret != -EEXIST) {
4014 free_extent_map(em);
4017 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4019 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4021 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4024 struct btrfs_ref_path {
4026 u64 nodes[BTRFS_MAX_LEVEL];
4028 u64 root_generation;
4035 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4036 u64 new_nodes[BTRFS_MAX_LEVEL];
4039 struct disk_extent {
4050 static int is_cowonly_root(u64 root_objectid)
4052 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4053 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4054 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4055 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4056 root_objectid == BTRFS_TREE_LOG_OBJECTID)
4061 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4062 struct btrfs_root *extent_root,
4063 struct btrfs_ref_path *ref_path,
4066 struct extent_buffer *leaf;
4067 struct btrfs_path *path;
4068 struct btrfs_extent_ref *ref;
4069 struct btrfs_key key;
4070 struct btrfs_key found_key;
4076 path = btrfs_alloc_path();
4081 ref_path->lowest_level = -1;
4082 ref_path->current_level = -1;
4083 ref_path->shared_level = -1;
4087 level = ref_path->current_level - 1;
4088 while (level >= -1) {
4090 if (level < ref_path->lowest_level)
4094 bytenr = ref_path->nodes[level];
4096 bytenr = ref_path->extent_start;
4098 BUG_ON(bytenr == 0);
4100 parent = ref_path->nodes[level + 1];
4101 ref_path->nodes[level + 1] = 0;
4102 ref_path->current_level = level;
4103 BUG_ON(parent == 0);
4105 key.objectid = bytenr;
4106 key.offset = parent + 1;
4107 key.type = BTRFS_EXTENT_REF_KEY;
4109 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4114 leaf = path->nodes[0];
4115 nritems = btrfs_header_nritems(leaf);
4116 if (path->slots[0] >= nritems) {
4117 ret = btrfs_next_leaf(extent_root, path);
4122 leaf = path->nodes[0];
4125 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4126 if (found_key.objectid == bytenr &&
4127 found_key.type == BTRFS_EXTENT_REF_KEY) {
4128 if (level < ref_path->shared_level)
4129 ref_path->shared_level = level;
4134 btrfs_release_path(extent_root, path);
4137 /* reached lowest level */
4141 level = ref_path->current_level;
4142 while (level < BTRFS_MAX_LEVEL - 1) {
4145 bytenr = ref_path->nodes[level];
4147 bytenr = ref_path->extent_start;
4149 BUG_ON(bytenr == 0);
4151 key.objectid = bytenr;
4153 key.type = BTRFS_EXTENT_REF_KEY;
4155 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4159 leaf = path->nodes[0];
4160 nritems = btrfs_header_nritems(leaf);
4161 if (path->slots[0] >= nritems) {
4162 ret = btrfs_next_leaf(extent_root, path);
4166 /* the extent was freed by someone */
4167 if (ref_path->lowest_level == level)
4169 btrfs_release_path(extent_root, path);
4172 leaf = path->nodes[0];
4175 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4176 if (found_key.objectid != bytenr ||
4177 found_key.type != BTRFS_EXTENT_REF_KEY) {
4178 /* the extent was freed by someone */
4179 if (ref_path->lowest_level == level) {
4183 btrfs_release_path(extent_root, path);
4187 ref = btrfs_item_ptr(leaf, path->slots[0],
4188 struct btrfs_extent_ref);
4189 ref_objectid = btrfs_ref_objectid(leaf, ref);
4190 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4192 level = (int)ref_objectid;
4193 BUG_ON(level >= BTRFS_MAX_LEVEL);
4194 ref_path->lowest_level = level;
4195 ref_path->current_level = level;
4196 ref_path->nodes[level] = bytenr;
4198 WARN_ON(ref_objectid != level);
4201 WARN_ON(level != -1);
4205 if (ref_path->lowest_level == level) {
4206 ref_path->owner_objectid = ref_objectid;
4207 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4211 * the block is tree root or the block isn't in reference
4214 if (found_key.objectid == found_key.offset ||
4215 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4216 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4217 ref_path->root_generation =
4218 btrfs_ref_generation(leaf, ref);
4220 /* special reference from the tree log */
4221 ref_path->nodes[0] = found_key.offset;
4222 ref_path->current_level = 0;
4229 BUG_ON(ref_path->nodes[level] != 0);
4230 ref_path->nodes[level] = found_key.offset;
4231 ref_path->current_level = level;
4234 * the reference was created in the running transaction,
4235 * no need to continue walking up.
4237 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4238 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4239 ref_path->root_generation =
4240 btrfs_ref_generation(leaf, ref);
4245 btrfs_release_path(extent_root, path);
4248 /* reached max tree level, but no tree root found. */
4251 btrfs_free_path(path);
4255 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4256 struct btrfs_root *extent_root,
4257 struct btrfs_ref_path *ref_path,
4260 memset(ref_path, 0, sizeof(*ref_path));
4261 ref_path->extent_start = extent_start;
4263 return __next_ref_path(trans, extent_root, ref_path, 1);
4266 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4267 struct btrfs_root *extent_root,
4268 struct btrfs_ref_path *ref_path)
4270 return __next_ref_path(trans, extent_root, ref_path, 0);
4273 static int noinline get_new_locations(struct inode *reloc_inode,
4274 struct btrfs_key *extent_key,
4275 u64 offset, int no_fragment,
4276 struct disk_extent **extents,
4279 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4280 struct btrfs_path *path;
4281 struct btrfs_file_extent_item *fi;
4282 struct extent_buffer *leaf;
4283 struct disk_extent *exts = *extents;
4284 struct btrfs_key found_key;
4289 int max = *nr_extents;
4292 WARN_ON(!no_fragment && *extents);
4295 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4300 path = btrfs_alloc_path();
4303 cur_pos = extent_key->objectid - offset;
4304 last_byte = extent_key->objectid + extent_key->offset;
4305 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4315 leaf = path->nodes[0];
4316 nritems = btrfs_header_nritems(leaf);
4317 if (path->slots[0] >= nritems) {
4318 ret = btrfs_next_leaf(root, path);
4323 leaf = path->nodes[0];
4326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4327 if (found_key.offset != cur_pos ||
4328 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4329 found_key.objectid != reloc_inode->i_ino)
4332 fi = btrfs_item_ptr(leaf, path->slots[0],
4333 struct btrfs_file_extent_item);
4334 if (btrfs_file_extent_type(leaf, fi) !=
4335 BTRFS_FILE_EXTENT_REG ||
4336 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4340 struct disk_extent *old = exts;
4342 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4343 memcpy(exts, old, sizeof(*exts) * nr);
4344 if (old != *extents)
4348 exts[nr].disk_bytenr =
4349 btrfs_file_extent_disk_bytenr(leaf, fi);
4350 exts[nr].disk_num_bytes =
4351 btrfs_file_extent_disk_num_bytes(leaf, fi);
4352 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4353 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4354 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4355 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4356 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4357 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4359 BUG_ON(exts[nr].offset > 0);
4360 BUG_ON(exts[nr].compression || exts[nr].encryption);
4361 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4363 cur_pos += exts[nr].num_bytes;
4366 if (cur_pos + offset >= last_byte)
4376 WARN_ON(cur_pos + offset > last_byte);
4377 if (cur_pos + offset < last_byte) {
4383 btrfs_free_path(path);
4385 if (exts != *extents)
4394 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4395 struct btrfs_root *root,
4396 struct btrfs_path *path,
4397 struct btrfs_key *extent_key,
4398 struct btrfs_key *leaf_key,
4399 struct btrfs_ref_path *ref_path,
4400 struct disk_extent *new_extents,
4403 struct extent_buffer *leaf;
4404 struct btrfs_file_extent_item *fi;
4405 struct inode *inode = NULL;
4406 struct btrfs_key key;
4414 int extent_locked = 0;
4418 memcpy(&key, leaf_key, sizeof(key));
4419 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4420 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4421 if (key.objectid < ref_path->owner_objectid ||
4422 (key.objectid == ref_path->owner_objectid &&
4423 key.type < BTRFS_EXTENT_DATA_KEY)) {
4424 key.objectid = ref_path->owner_objectid;
4425 key.type = BTRFS_EXTENT_DATA_KEY;
4431 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4435 leaf = path->nodes[0];
4436 nritems = btrfs_header_nritems(leaf);
4438 if (extent_locked && ret > 0) {
4440 * the file extent item was modified by someone
4441 * before the extent got locked.
4443 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4444 lock_end, GFP_NOFS);
4448 if (path->slots[0] >= nritems) {
4449 if (++nr_scaned > 2)
4452 BUG_ON(extent_locked);
4453 ret = btrfs_next_leaf(root, path);
4458 leaf = path->nodes[0];
4459 nritems = btrfs_header_nritems(leaf);
4462 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4464 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4465 if ((key.objectid > ref_path->owner_objectid) ||
4466 (key.objectid == ref_path->owner_objectid &&
4467 key.type > BTRFS_EXTENT_DATA_KEY) ||
4468 (key.offset >= first_pos + extent_key->offset))
4472 if (inode && key.objectid != inode->i_ino) {
4473 BUG_ON(extent_locked);
4474 btrfs_release_path(root, path);
4475 mutex_unlock(&inode->i_mutex);
4481 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4486 fi = btrfs_item_ptr(leaf, path->slots[0],
4487 struct btrfs_file_extent_item);
4488 extent_type = btrfs_file_extent_type(leaf, fi);
4489 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4490 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4491 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4492 extent_key->objectid)) {
4498 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4499 ext_offset = btrfs_file_extent_offset(leaf, fi);
4501 if (first_pos > key.offset - ext_offset)
4502 first_pos = key.offset - ext_offset;
4504 if (!extent_locked) {
4505 lock_start = key.offset;
4506 lock_end = lock_start + num_bytes - 1;
4508 if (lock_start > key.offset ||
4509 lock_end + 1 < key.offset + num_bytes) {
4510 unlock_extent(&BTRFS_I(inode)->io_tree,
4511 lock_start, lock_end, GFP_NOFS);
4517 btrfs_release_path(root, path);
4519 inode = btrfs_iget_locked(root->fs_info->sb,
4520 key.objectid, root);
4521 if (inode->i_state & I_NEW) {
4522 BTRFS_I(inode)->root = root;
4523 BTRFS_I(inode)->location.objectid =
4525 BTRFS_I(inode)->location.type =
4526 BTRFS_INODE_ITEM_KEY;
4527 BTRFS_I(inode)->location.offset = 0;
4528 btrfs_read_locked_inode(inode);
4529 unlock_new_inode(inode);
4532 * some code call btrfs_commit_transaction while
4533 * holding the i_mutex, so we can't use mutex_lock
4536 if (is_bad_inode(inode) ||
4537 !mutex_trylock(&inode->i_mutex)) {
4540 key.offset = (u64)-1;
4545 if (!extent_locked) {
4546 struct btrfs_ordered_extent *ordered;
4548 btrfs_release_path(root, path);
4550 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4551 lock_end, GFP_NOFS);
4552 ordered = btrfs_lookup_first_ordered_extent(inode,
4555 ordered->file_offset <= lock_end &&
4556 ordered->file_offset + ordered->len > lock_start) {
4557 unlock_extent(&BTRFS_I(inode)->io_tree,
4558 lock_start, lock_end, GFP_NOFS);
4559 btrfs_start_ordered_extent(inode, ordered, 1);
4560 btrfs_put_ordered_extent(ordered);
4561 key.offset += num_bytes;
4565 btrfs_put_ordered_extent(ordered);
4571 if (nr_extents == 1) {
4572 /* update extent pointer in place */
4573 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4574 new_extents[0].disk_bytenr);
4575 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4576 new_extents[0].disk_num_bytes);
4577 btrfs_mark_buffer_dirty(leaf);
4579 btrfs_drop_extent_cache(inode, key.offset,
4580 key.offset + num_bytes - 1, 0);
4582 ret = btrfs_inc_extent_ref(trans, root,
4583 new_extents[0].disk_bytenr,
4584 new_extents[0].disk_num_bytes,
4586 root->root_key.objectid,
4591 ret = btrfs_free_extent(trans, root,
4592 extent_key->objectid,
4595 btrfs_header_owner(leaf),
4596 btrfs_header_generation(leaf),
4600 btrfs_release_path(root, path);
4601 key.offset += num_bytes;
4609 * drop old extent pointer at first, then insert the
4610 * new pointers one bye one
4612 btrfs_release_path(root, path);
4613 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4614 key.offset + num_bytes,
4615 key.offset, &alloc_hint);
4618 for (i = 0; i < nr_extents; i++) {
4619 if (ext_offset >= new_extents[i].num_bytes) {
4620 ext_offset -= new_extents[i].num_bytes;
4623 extent_len = min(new_extents[i].num_bytes -
4624 ext_offset, num_bytes);
4626 ret = btrfs_insert_empty_item(trans, root,
4631 leaf = path->nodes[0];
4632 fi = btrfs_item_ptr(leaf, path->slots[0],
4633 struct btrfs_file_extent_item);
4634 btrfs_set_file_extent_generation(leaf, fi,
4636 btrfs_set_file_extent_type(leaf, fi,
4637 BTRFS_FILE_EXTENT_REG);
4638 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4639 new_extents[i].disk_bytenr);
4640 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4641 new_extents[i].disk_num_bytes);
4642 btrfs_set_file_extent_ram_bytes(leaf, fi,
4643 new_extents[i].ram_bytes);
4645 btrfs_set_file_extent_compression(leaf, fi,
4646 new_extents[i].compression);
4647 btrfs_set_file_extent_encryption(leaf, fi,
4648 new_extents[i].encryption);
4649 btrfs_set_file_extent_other_encoding(leaf, fi,
4650 new_extents[i].other_encoding);
4652 btrfs_set_file_extent_num_bytes(leaf, fi,
4654 ext_offset += new_extents[i].offset;
4655 btrfs_set_file_extent_offset(leaf, fi,
4657 btrfs_mark_buffer_dirty(leaf);
4659 btrfs_drop_extent_cache(inode, key.offset,
4660 key.offset + extent_len - 1, 0);
4662 ret = btrfs_inc_extent_ref(trans, root,
4663 new_extents[i].disk_bytenr,
4664 new_extents[i].disk_num_bytes,
4666 root->root_key.objectid,
4667 trans->transid, key.objectid);
4669 btrfs_release_path(root, path);
4671 inode_add_bytes(inode, extent_len);
4674 num_bytes -= extent_len;
4675 key.offset += extent_len;
4680 BUG_ON(i >= nr_extents);
4684 if (extent_locked) {
4685 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4686 lock_end, GFP_NOFS);
4690 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4691 key.offset >= first_pos + extent_key->offset)
4698 btrfs_release_path(root, path);
4700 mutex_unlock(&inode->i_mutex);
4701 if (extent_locked) {
4702 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4703 lock_end, GFP_NOFS);
4710 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4711 struct btrfs_root *root,
4712 struct extent_buffer *buf, u64 orig_start)
4717 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4718 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4720 level = btrfs_header_level(buf);
4722 struct btrfs_leaf_ref *ref;
4723 struct btrfs_leaf_ref *orig_ref;
4725 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4729 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4731 btrfs_free_leaf_ref(root, orig_ref);
4735 ref->nritems = orig_ref->nritems;
4736 memcpy(ref->extents, orig_ref->extents,
4737 sizeof(ref->extents[0]) * ref->nritems);
4739 btrfs_free_leaf_ref(root, orig_ref);
4741 ref->root_gen = trans->transid;
4742 ref->bytenr = buf->start;
4743 ref->owner = btrfs_header_owner(buf);
4744 ref->generation = btrfs_header_generation(buf);
4745 ret = btrfs_add_leaf_ref(root, ref, 0);
4747 btrfs_free_leaf_ref(root, ref);
4752 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4753 struct extent_buffer *leaf,
4754 struct btrfs_block_group_cache *group,
4755 struct btrfs_root *target_root)
4757 struct btrfs_key key;
4758 struct inode *inode = NULL;
4759 struct btrfs_file_extent_item *fi;
4761 u64 skip_objectid = 0;
4765 nritems = btrfs_header_nritems(leaf);
4766 for (i = 0; i < nritems; i++) {
4767 btrfs_item_key_to_cpu(leaf, &key, i);
4768 if (key.objectid == skip_objectid ||
4769 key.type != BTRFS_EXTENT_DATA_KEY)
4771 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4772 if (btrfs_file_extent_type(leaf, fi) ==
4773 BTRFS_FILE_EXTENT_INLINE)
4775 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4777 if (!inode || inode->i_ino != key.objectid) {
4779 inode = btrfs_ilookup(target_root->fs_info->sb,
4780 key.objectid, target_root, 1);
4783 skip_objectid = key.objectid;
4786 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4788 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4789 key.offset + num_bytes - 1, GFP_NOFS);
4790 btrfs_drop_extent_cache(inode, key.offset,
4791 key.offset + num_bytes - 1, 1);
4792 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4793 key.offset + num_bytes - 1, GFP_NOFS);
4800 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4801 struct btrfs_root *root,
4802 struct extent_buffer *leaf,
4803 struct btrfs_block_group_cache *group,
4804 struct inode *reloc_inode)
4806 struct btrfs_key key;
4807 struct btrfs_key extent_key;
4808 struct btrfs_file_extent_item *fi;
4809 struct btrfs_leaf_ref *ref;
4810 struct disk_extent *new_extent;
4819 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4820 BUG_ON(!new_extent);
4822 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4826 nritems = btrfs_header_nritems(leaf);
4827 for (i = 0; i < nritems; i++) {
4828 btrfs_item_key_to_cpu(leaf, &key, i);
4829 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4831 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4832 if (btrfs_file_extent_type(leaf, fi) ==
4833 BTRFS_FILE_EXTENT_INLINE)
4835 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4836 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4841 if (bytenr >= group->key.objectid + group->key.offset ||
4842 bytenr + num_bytes <= group->key.objectid)
4845 extent_key.objectid = bytenr;
4846 extent_key.offset = num_bytes;
4847 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4849 ret = get_new_locations(reloc_inode, &extent_key,
4850 group->key.objectid, 1,
4851 &new_extent, &nr_extent);
4856 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4857 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4858 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4859 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4861 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4862 new_extent->disk_bytenr);
4863 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4864 new_extent->disk_num_bytes);
4865 btrfs_mark_buffer_dirty(leaf);
4867 ret = btrfs_inc_extent_ref(trans, root,
4868 new_extent->disk_bytenr,
4869 new_extent->disk_num_bytes,
4871 root->root_key.objectid,
4872 trans->transid, key.objectid);
4874 ret = btrfs_free_extent(trans, root,
4875 bytenr, num_bytes, leaf->start,
4876 btrfs_header_owner(leaf),
4877 btrfs_header_generation(leaf),
4883 BUG_ON(ext_index + 1 != ref->nritems);
4884 btrfs_free_leaf_ref(root, ref);
4888 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4889 struct btrfs_root *root)
4891 struct btrfs_root *reloc_root;
4894 if (root->reloc_root) {
4895 reloc_root = root->reloc_root;
4896 root->reloc_root = NULL;
4897 list_add(&reloc_root->dead_list,
4898 &root->fs_info->dead_reloc_roots);
4900 btrfs_set_root_bytenr(&reloc_root->root_item,
4901 reloc_root->node->start);
4902 btrfs_set_root_level(&root->root_item,
4903 btrfs_header_level(reloc_root->node));
4904 memset(&reloc_root->root_item.drop_progress, 0,
4905 sizeof(struct btrfs_disk_key));
4906 reloc_root->root_item.drop_level = 0;
4908 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4909 &reloc_root->root_key,
4910 &reloc_root->root_item);
4916 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4918 struct btrfs_trans_handle *trans;
4919 struct btrfs_root *reloc_root;
4920 struct btrfs_root *prev_root = NULL;
4921 struct list_head dead_roots;
4925 INIT_LIST_HEAD(&dead_roots);
4926 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4928 while (!list_empty(&dead_roots)) {
4929 reloc_root = list_entry(dead_roots.prev,
4930 struct btrfs_root, dead_list);
4931 list_del_init(&reloc_root->dead_list);
4933 BUG_ON(reloc_root->commit_root != NULL);
4935 trans = btrfs_join_transaction(root, 1);
4938 mutex_lock(&root->fs_info->drop_mutex);
4939 ret = btrfs_drop_snapshot(trans, reloc_root);
4942 mutex_unlock(&root->fs_info->drop_mutex);
4944 nr = trans->blocks_used;
4945 ret = btrfs_end_transaction(trans, root);
4947 btrfs_btree_balance_dirty(root, nr);
4950 free_extent_buffer(reloc_root->node);
4952 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4953 &reloc_root->root_key);
4955 mutex_unlock(&root->fs_info->drop_mutex);
4957 nr = trans->blocks_used;
4958 ret = btrfs_end_transaction(trans, root);
4960 btrfs_btree_balance_dirty(root, nr);
4963 prev_root = reloc_root;
4966 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4972 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4974 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4978 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4980 struct btrfs_root *reloc_root;
4981 struct btrfs_trans_handle *trans;
4982 struct btrfs_key location;
4986 mutex_lock(&root->fs_info->tree_reloc_mutex);
4987 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4989 found = !list_empty(&root->fs_info->dead_reloc_roots);
4990 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4993 trans = btrfs_start_transaction(root, 1);
4995 ret = btrfs_commit_transaction(trans, root);
4999 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5000 location.offset = (u64)-1;
5001 location.type = BTRFS_ROOT_ITEM_KEY;
5003 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5004 BUG_ON(!reloc_root);
5005 btrfs_orphan_cleanup(reloc_root);
5009 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5010 struct btrfs_root *root)
5012 struct btrfs_root *reloc_root;
5013 struct extent_buffer *eb;
5014 struct btrfs_root_item *root_item;
5015 struct btrfs_key root_key;
5018 BUG_ON(!root->ref_cows);
5019 if (root->reloc_root)
5022 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5025 ret = btrfs_copy_root(trans, root, root->commit_root,
5026 &eb, BTRFS_TREE_RELOC_OBJECTID);
5029 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5030 root_key.offset = root->root_key.objectid;
5031 root_key.type = BTRFS_ROOT_ITEM_KEY;
5033 memcpy(root_item, &root->root_item, sizeof(root_item));
5034 btrfs_set_root_refs(root_item, 0);
5035 btrfs_set_root_bytenr(root_item, eb->start);
5036 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5037 btrfs_set_root_generation(root_item, trans->transid);
5039 btrfs_tree_unlock(eb);
5040 free_extent_buffer(eb);
5042 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5043 &root_key, root_item);
5047 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5049 BUG_ON(!reloc_root);
5050 reloc_root->last_trans = trans->transid;
5051 reloc_root->commit_root = NULL;
5052 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5054 root->reloc_root = reloc_root;
5059 * Core function of space balance.
5061 * The idea is using reloc trees to relocate tree blocks in reference
5062 * counted roots. There is one reloc tree for each subvol, and all
5063 * reloc trees share same root key objectid. Reloc trees are snapshots
5064 * of the latest committed roots of subvols (root->commit_root).
5066 * To relocate a tree block referenced by a subvol, there are two steps.
5067 * COW the block through subvol's reloc tree, then update block pointer
5068 * in the subvol to point to the new block. Since all reloc trees share
5069 * same root key objectid, doing special handing for tree blocks owned
5070 * by them is easy. Once a tree block has been COWed in one reloc tree,
5071 * we can use the resulting new block directly when the same block is
5072 * required to COW again through other reloc trees. By this way, relocated
5073 * tree blocks are shared between reloc trees, so they are also shared
5076 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5077 struct btrfs_root *root,
5078 struct btrfs_path *path,
5079 struct btrfs_key *first_key,
5080 struct btrfs_ref_path *ref_path,
5081 struct btrfs_block_group_cache *group,
5082 struct inode *reloc_inode)
5084 struct btrfs_root *reloc_root;
5085 struct extent_buffer *eb = NULL;
5086 struct btrfs_key *keys;
5090 int lowest_level = 0;
5093 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5094 lowest_level = ref_path->owner_objectid;
5096 if (!root->ref_cows) {
5097 path->lowest_level = lowest_level;
5098 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5100 path->lowest_level = 0;
5101 btrfs_release_path(root, path);
5105 mutex_lock(&root->fs_info->tree_reloc_mutex);
5106 ret = init_reloc_tree(trans, root);
5108 reloc_root = root->reloc_root;
5110 shared_level = ref_path->shared_level;
5111 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5113 keys = ref_path->node_keys;
5114 nodes = ref_path->new_nodes;
5115 memset(&keys[shared_level + 1], 0,
5116 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5117 memset(&nodes[shared_level + 1], 0,
5118 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5120 if (nodes[lowest_level] == 0) {
5121 path->lowest_level = lowest_level;
5122 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5125 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5126 eb = path->nodes[level];
5127 if (!eb || eb == reloc_root->node)
5129 nodes[level] = eb->start;
5131 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5133 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5135 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5136 eb = path->nodes[0];
5137 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5138 group, reloc_inode);
5141 btrfs_release_path(reloc_root, path);
5143 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5149 * replace tree blocks in the fs tree with tree blocks in
5152 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5155 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5156 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5159 extent_buffer_get(path->nodes[0]);
5160 eb = path->nodes[0];
5161 btrfs_release_path(reloc_root, path);
5162 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5164 free_extent_buffer(eb);
5167 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5168 path->lowest_level = 0;
5172 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5173 struct btrfs_root *root,
5174 struct btrfs_path *path,
5175 struct btrfs_key *first_key,
5176 struct btrfs_ref_path *ref_path)
5180 ret = relocate_one_path(trans, root, path, first_key,
5181 ref_path, NULL, NULL);
5184 if (root == root->fs_info->extent_root)
5185 btrfs_extent_post_op(trans, root);
5190 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5191 struct btrfs_root *extent_root,
5192 struct btrfs_path *path,
5193 struct btrfs_key *extent_key)
5197 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5200 ret = btrfs_del_item(trans, extent_root, path);
5202 btrfs_release_path(extent_root, path);
5206 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5207 struct btrfs_ref_path *ref_path)
5209 struct btrfs_key root_key;
5211 root_key.objectid = ref_path->root_objectid;
5212 root_key.type = BTRFS_ROOT_ITEM_KEY;
5213 if (is_cowonly_root(ref_path->root_objectid))
5214 root_key.offset = 0;
5216 root_key.offset = (u64)-1;
5218 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5221 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5222 struct btrfs_path *path,
5223 struct btrfs_key *extent_key,
5224 struct btrfs_block_group_cache *group,
5225 struct inode *reloc_inode, int pass)
5227 struct btrfs_trans_handle *trans;
5228 struct btrfs_root *found_root;
5229 struct btrfs_ref_path *ref_path = NULL;
5230 struct disk_extent *new_extents = NULL;
5235 struct btrfs_key first_key;
5239 trans = btrfs_start_transaction(extent_root, 1);
5242 if (extent_key->objectid == 0) {
5243 ret = del_extent_zero(trans, extent_root, path, extent_key);
5247 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5253 for (loops = 0; ; loops++) {
5255 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5256 extent_key->objectid);
5258 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5265 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5266 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5269 found_root = read_ref_root(extent_root->fs_info, ref_path);
5270 BUG_ON(!found_root);
5272 * for reference counted tree, only process reference paths
5273 * rooted at the latest committed root.
5275 if (found_root->ref_cows &&
5276 ref_path->root_generation != found_root->root_key.offset)
5279 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5282 * copy data extents to new locations
5284 u64 group_start = group->key.objectid;
5285 ret = relocate_data_extent(reloc_inode,
5294 level = ref_path->owner_objectid;
5297 if (prev_block != ref_path->nodes[level]) {
5298 struct extent_buffer *eb;
5299 u64 block_start = ref_path->nodes[level];
5300 u64 block_size = btrfs_level_size(found_root, level);
5302 eb = read_tree_block(found_root, block_start,
5304 btrfs_tree_lock(eb);
5305 BUG_ON(level != btrfs_header_level(eb));
5308 btrfs_item_key_to_cpu(eb, &first_key, 0);
5310 btrfs_node_key_to_cpu(eb, &first_key, 0);
5312 btrfs_tree_unlock(eb);
5313 free_extent_buffer(eb);
5314 prev_block = block_start;
5317 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5320 * use fallback method to process the remaining
5324 u64 group_start = group->key.objectid;
5325 new_extents = kmalloc(sizeof(*new_extents),
5328 ret = get_new_locations(reloc_inode,
5336 btrfs_record_root_in_trans(found_root);
5337 ret = replace_one_extent(trans, found_root,
5339 &first_key, ref_path,
5340 new_extents, nr_extents);
5346 btrfs_record_root_in_trans(found_root);
5347 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5348 ret = relocate_tree_block(trans, found_root, path,
5349 &first_key, ref_path);
5352 * try to update data extent references while
5353 * keeping metadata shared between snapshots.
5355 ret = relocate_one_path(trans, found_root, path,
5356 &first_key, ref_path,
5357 group, reloc_inode);
5364 btrfs_end_transaction(trans, extent_root);
5370 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5373 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5374 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5376 num_devices = root->fs_info->fs_devices->num_devices;
5377 if (num_devices == 1) {
5378 stripped |= BTRFS_BLOCK_GROUP_DUP;
5379 stripped = flags & ~stripped;
5381 /* turn raid0 into single device chunks */
5382 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5385 /* turn mirroring into duplication */
5386 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5387 BTRFS_BLOCK_GROUP_RAID10))
5388 return stripped | BTRFS_BLOCK_GROUP_DUP;
5391 /* they already had raid on here, just return */
5392 if (flags & stripped)
5395 stripped |= BTRFS_BLOCK_GROUP_DUP;
5396 stripped = flags & ~stripped;
5398 /* switch duplicated blocks with raid1 */
5399 if (flags & BTRFS_BLOCK_GROUP_DUP)
5400 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5402 /* turn single device chunks into raid0 */
5403 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5408 int __alloc_chunk_for_shrink(struct btrfs_root *root,
5409 struct btrfs_block_group_cache *shrink_block_group,
5412 struct btrfs_trans_handle *trans;
5413 u64 new_alloc_flags;
5416 spin_lock(&shrink_block_group->lock);
5417 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5418 spin_unlock(&shrink_block_group->lock);
5420 trans = btrfs_start_transaction(root, 1);
5421 spin_lock(&shrink_block_group->lock);
5423 new_alloc_flags = update_block_group_flags(root,
5424 shrink_block_group->flags);
5425 if (new_alloc_flags != shrink_block_group->flags) {
5427 btrfs_block_group_used(&shrink_block_group->item);
5429 calc = shrink_block_group->key.offset;
5431 spin_unlock(&shrink_block_group->lock);
5433 do_chunk_alloc(trans, root->fs_info->extent_root,
5434 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5436 btrfs_end_transaction(trans, root);
5438 spin_unlock(&shrink_block_group->lock);
5442 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5443 struct btrfs_root *root,
5444 u64 objectid, u64 size)
5446 struct btrfs_path *path;
5447 struct btrfs_inode_item *item;
5448 struct extent_buffer *leaf;
5451 path = btrfs_alloc_path();
5455 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5459 leaf = path->nodes[0];
5460 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5461 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5462 btrfs_set_inode_generation(leaf, item, 1);
5463 btrfs_set_inode_size(leaf, item, size);
5464 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5465 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
5466 BTRFS_INODE_NOCOMPRESS);
5467 btrfs_mark_buffer_dirty(leaf);
5468 btrfs_release_path(root, path);
5470 btrfs_free_path(path);
5474 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5475 struct btrfs_block_group_cache *group)
5477 struct inode *inode = NULL;
5478 struct btrfs_trans_handle *trans;
5479 struct btrfs_root *root;
5480 struct btrfs_key root_key;
5481 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5484 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5485 root_key.type = BTRFS_ROOT_ITEM_KEY;
5486 root_key.offset = (u64)-1;
5487 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5489 return ERR_CAST(root);
5491 trans = btrfs_start_transaction(root, 1);
5494 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5498 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5501 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5502 group->key.offset, 0, group->key.offset,
5506 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5507 if (inode->i_state & I_NEW) {
5508 BTRFS_I(inode)->root = root;
5509 BTRFS_I(inode)->location.objectid = objectid;
5510 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5511 BTRFS_I(inode)->location.offset = 0;
5512 btrfs_read_locked_inode(inode);
5513 unlock_new_inode(inode);
5514 BUG_ON(is_bad_inode(inode));
5519 err = btrfs_orphan_add(trans, inode);
5521 btrfs_end_transaction(trans, root);
5525 inode = ERR_PTR(err);
5530 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5532 struct btrfs_trans_handle *trans;
5533 struct btrfs_path *path;
5534 struct btrfs_fs_info *info = root->fs_info;
5535 struct extent_buffer *leaf;
5536 struct inode *reloc_inode;
5537 struct btrfs_block_group_cache *block_group;
5538 struct btrfs_key key;
5547 root = root->fs_info->extent_root;
5549 block_group = btrfs_lookup_block_group(info, group_start);
5550 BUG_ON(!block_group);
5552 printk("btrfs relocating block group %llu flags %llu\n",
5553 (unsigned long long)block_group->key.objectid,
5554 (unsigned long long)block_group->flags);
5556 path = btrfs_alloc_path();
5559 reloc_inode = create_reloc_inode(info, block_group);
5560 BUG_ON(IS_ERR(reloc_inode));
5562 __alloc_chunk_for_shrink(root, block_group, 1);
5563 block_group->ro = 1;
5564 block_group->space_info->total_bytes -= block_group->key.offset;
5566 btrfs_start_delalloc_inodes(info->tree_root);
5567 btrfs_wait_ordered_extents(info->tree_root, 0);
5572 key.objectid = block_group->key.objectid;
5575 cur_byte = key.objectid;
5577 trans = btrfs_start_transaction(info->tree_root, 1);
5578 btrfs_commit_transaction(trans, info->tree_root);
5580 mutex_lock(&root->fs_info->cleaner_mutex);
5581 btrfs_clean_old_snapshots(info->tree_root);
5582 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5583 mutex_unlock(&root->fs_info->cleaner_mutex);
5586 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5590 leaf = path->nodes[0];
5591 nritems = btrfs_header_nritems(leaf);
5592 if (path->slots[0] >= nritems) {
5593 ret = btrfs_next_leaf(root, path);
5600 leaf = path->nodes[0];
5601 nritems = btrfs_header_nritems(leaf);
5604 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5606 if (key.objectid >= block_group->key.objectid +
5607 block_group->key.offset)
5610 if (progress && need_resched()) {
5611 btrfs_release_path(root, path);
5618 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5619 key.objectid + key.offset <= cur_byte) {
5625 cur_byte = key.objectid + key.offset;
5626 btrfs_release_path(root, path);
5628 __alloc_chunk_for_shrink(root, block_group, 0);
5629 ret = relocate_one_extent(root, path, &key, block_group,
5635 key.objectid = cur_byte;
5640 btrfs_release_path(root, path);
5643 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5644 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5645 WARN_ON(reloc_inode->i_mapping->nrpages);
5648 if (total_found > 0) {
5649 printk("btrfs found %llu extents in pass %d\n",
5650 (unsigned long long)total_found, pass);
5652 if (total_found == skipped && pass > 2) {
5654 reloc_inode = create_reloc_inode(info, block_group);
5660 /* delete reloc_inode */
5663 /* unpin extents in this range */
5664 trans = btrfs_start_transaction(info->tree_root, 1);
5665 btrfs_commit_transaction(trans, info->tree_root);
5667 spin_lock(&block_group->lock);
5668 WARN_ON(block_group->pinned > 0);
5669 WARN_ON(block_group->reserved > 0);
5670 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5671 spin_unlock(&block_group->lock);
5674 btrfs_free_path(path);
5678 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5679 struct btrfs_key *key)
5682 struct btrfs_key found_key;
5683 struct extent_buffer *leaf;
5686 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5691 slot = path->slots[0];
5692 leaf = path->nodes[0];
5693 if (slot >= btrfs_header_nritems(leaf)) {
5694 ret = btrfs_next_leaf(root, path);
5701 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5703 if (found_key.objectid >= key->objectid &&
5704 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5715 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5717 struct btrfs_block_group_cache *block_group;
5720 spin_lock(&info->block_group_cache_lock);
5721 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5722 block_group = rb_entry(n, struct btrfs_block_group_cache,
5724 rb_erase(&block_group->cache_node,
5725 &info->block_group_cache_tree);
5726 spin_unlock(&info->block_group_cache_lock);
5728 btrfs_remove_free_space_cache(block_group);
5729 down_write(&block_group->space_info->groups_sem);
5730 list_del(&block_group->list);
5731 up_write(&block_group->space_info->groups_sem);
5734 spin_lock(&info->block_group_cache_lock);
5736 spin_unlock(&info->block_group_cache_lock);
5740 int btrfs_read_block_groups(struct btrfs_root *root)
5742 struct btrfs_path *path;
5744 struct btrfs_block_group_cache *cache;
5745 struct btrfs_fs_info *info = root->fs_info;
5746 struct btrfs_space_info *space_info;
5747 struct btrfs_key key;
5748 struct btrfs_key found_key;
5749 struct extent_buffer *leaf;
5751 root = info->extent_root;
5754 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5755 path = btrfs_alloc_path();
5760 ret = find_first_block_group(root, path, &key);
5768 leaf = path->nodes[0];
5769 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5770 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5776 spin_lock_init(&cache->lock);
5777 mutex_init(&cache->alloc_mutex);
5778 INIT_LIST_HEAD(&cache->list);
5779 read_extent_buffer(leaf, &cache->item,
5780 btrfs_item_ptr_offset(leaf, path->slots[0]),
5781 sizeof(cache->item));
5782 memcpy(&cache->key, &found_key, sizeof(found_key));
5784 key.objectid = found_key.objectid + found_key.offset;
5785 btrfs_release_path(root, path);
5786 cache->flags = btrfs_block_group_flags(&cache->item);
5788 ret = update_space_info(info, cache->flags, found_key.offset,
5789 btrfs_block_group_used(&cache->item),
5792 cache->space_info = space_info;
5793 down_write(&space_info->groups_sem);
5794 list_add_tail(&cache->list, &space_info->block_groups);
5795 up_write(&space_info->groups_sem);
5797 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5800 set_avail_alloc_bits(root->fs_info, cache->flags);
5804 btrfs_free_path(path);
5808 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5809 struct btrfs_root *root, u64 bytes_used,
5810 u64 type, u64 chunk_objectid, u64 chunk_offset,
5814 struct btrfs_root *extent_root;
5815 struct btrfs_block_group_cache *cache;
5817 extent_root = root->fs_info->extent_root;
5819 root->fs_info->last_trans_new_blockgroup = trans->transid;
5821 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5825 cache->key.objectid = chunk_offset;
5826 cache->key.offset = size;
5827 spin_lock_init(&cache->lock);
5828 mutex_init(&cache->alloc_mutex);
5829 INIT_LIST_HEAD(&cache->list);
5830 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5832 btrfs_set_block_group_used(&cache->item, bytes_used);
5833 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5834 cache->flags = type;
5835 btrfs_set_block_group_flags(&cache->item, type);
5837 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5838 &cache->space_info);
5840 down_write(&cache->space_info->groups_sem);
5841 list_add_tail(&cache->list, &cache->space_info->block_groups);
5842 up_write(&cache->space_info->groups_sem);
5844 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5847 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5848 sizeof(cache->item));
5851 finish_current_insert(trans, extent_root, 0);
5852 ret = del_pending_extents(trans, extent_root, 0);
5854 set_avail_alloc_bits(extent_root->fs_info, type);
5859 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5860 struct btrfs_root *root, u64 group_start)
5862 struct btrfs_path *path;
5863 struct btrfs_block_group_cache *block_group;
5864 struct btrfs_key key;
5867 root = root->fs_info->extent_root;
5869 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5870 BUG_ON(!block_group);
5872 memcpy(&key, &block_group->key, sizeof(key));
5874 path = btrfs_alloc_path();
5877 btrfs_remove_free_space_cache(block_group);
5878 rb_erase(&block_group->cache_node,
5879 &root->fs_info->block_group_cache_tree);
5880 down_write(&block_group->space_info->groups_sem);
5881 list_del(&block_group->list);
5882 up_write(&block_group->space_info->groups_sem);
5885 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5886 kfree(shrink_block_group);
5889 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5895 ret = btrfs_del_item(trans, root, path);
5897 btrfs_free_path(path);