1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2009 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/error-injection.h>
15 #include "transaction.h"
18 #include "btrfs_inode.h"
19 #include "async-thread.h"
20 #include "free-space-cache.h"
21 #include "inode-map.h"
23 #include "print-tree.h"
24 #include "delalloc-space.h"
25 #include "block-group.h"
32 * [What does relocation do]
34 * The objective of relocation is to relocate all extents of the target block
35 * group to other block groups.
36 * This is utilized by resize (shrink only), profile converting, compacting
37 * space, or balance routine to spread chunks over devices.
40 * ------------------------------------------------------------------
41 * BG A: 10 data extents | BG A: deleted
42 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
43 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
45 * [How does relocation work]
47 * 1. Mark the target block group read-only
48 * New extents won't be allocated from the target block group.
50 * 2.1 Record each extent in the target block group
51 * To build a proper map of extents to be relocated.
53 * 2.2 Build data reloc tree and reloc trees
54 * Data reloc tree will contain an inode, recording all newly relocated
56 * There will be only one data reloc tree for one data block group.
58 * Reloc tree will be a special snapshot of its source tree, containing
59 * relocated tree blocks.
60 * Each tree referring to a tree block in target block group will get its
63 * 2.3 Swap source tree with its corresponding reloc tree
64 * Each involved tree only refers to new extents after swap.
66 * 3. Cleanup reloc trees and data reloc tree.
67 * As old extents in the target block group are still referenced by reloc
68 * trees, we need to clean them up before really freeing the target block
71 * The main complexity is in steps 2.2 and 2.3.
73 * The entry point of relocation is relocate_block_group() function.
76 #define RELOCATION_RESERVED_NODES 256
78 * map address of tree root to tree
82 struct rb_node rb_node;
84 }; /* Use rb_simle_node for search/insert */
89 struct rb_root rb_root;
94 * present a tree block to process
98 struct rb_node rb_node;
100 }; /* Use rb_simple_node for search/insert */
101 struct btrfs_key key;
102 unsigned int level:8;
103 unsigned int key_ready:1;
106 #define MAX_EXTENTS 128
108 struct file_extent_cluster {
111 u64 boundary[MAX_EXTENTS];
115 struct reloc_control {
116 /* block group to relocate */
117 struct btrfs_block_group *block_group;
119 struct btrfs_root *extent_root;
120 /* inode for moving data */
121 struct inode *data_inode;
123 struct btrfs_block_rsv *block_rsv;
125 struct btrfs_backref_cache backref_cache;
127 struct file_extent_cluster cluster;
128 /* tree blocks have been processed */
129 struct extent_io_tree processed_blocks;
130 /* map start of tree root to corresponding reloc tree */
131 struct mapping_tree reloc_root_tree;
132 /* list of reloc trees */
133 struct list_head reloc_roots;
134 /* list of subvolume trees that get relocated */
135 struct list_head dirty_subvol_roots;
136 /* size of metadata reservation for merging reloc trees */
137 u64 merging_rsv_size;
138 /* size of relocated tree nodes */
140 /* reserved size for block group relocation*/
146 unsigned int stage:8;
147 unsigned int create_reloc_tree:1;
148 unsigned int merge_reloc_tree:1;
149 unsigned int found_file_extent:1;
152 /* stages of data relocation */
153 #define MOVE_DATA_EXTENTS 0
154 #define UPDATE_DATA_PTRS 1
156 static void mark_block_processed(struct reloc_control *rc,
157 struct btrfs_backref_node *node)
161 if (node->level == 0 ||
162 in_range(node->bytenr, rc->block_group->start,
163 rc->block_group->length)) {
164 blocksize = rc->extent_root->fs_info->nodesize;
165 set_extent_bits(&rc->processed_blocks, node->bytenr,
166 node->bytenr + blocksize - 1, EXTENT_DIRTY);
172 static void mapping_tree_init(struct mapping_tree *tree)
174 tree->rb_root = RB_ROOT;
175 spin_lock_init(&tree->lock);
179 * walk up backref nodes until reach node presents tree root
181 static struct btrfs_backref_node *walk_up_backref(
182 struct btrfs_backref_node *node,
183 struct btrfs_backref_edge *edges[], int *index)
185 struct btrfs_backref_edge *edge;
188 while (!list_empty(&node->upper)) {
189 edge = list_entry(node->upper.next,
190 struct btrfs_backref_edge, list[LOWER]);
192 node = edge->node[UPPER];
194 BUG_ON(node->detached);
200 * walk down backref nodes to find start of next reference path
202 static struct btrfs_backref_node *walk_down_backref(
203 struct btrfs_backref_edge *edges[], int *index)
205 struct btrfs_backref_edge *edge;
206 struct btrfs_backref_node *lower;
210 edge = edges[idx - 1];
211 lower = edge->node[LOWER];
212 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
216 edge = list_entry(edge->list[LOWER].next,
217 struct btrfs_backref_edge, list[LOWER]);
218 edges[idx - 1] = edge;
220 return edge->node[UPPER];
226 static void update_backref_node(struct btrfs_backref_cache *cache,
227 struct btrfs_backref_node *node, u64 bytenr)
229 struct rb_node *rb_node;
230 rb_erase(&node->rb_node, &cache->rb_root);
231 node->bytenr = bytenr;
232 rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node);
234 btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
238 * update backref cache after a transaction commit
240 static int update_backref_cache(struct btrfs_trans_handle *trans,
241 struct btrfs_backref_cache *cache)
243 struct btrfs_backref_node *node;
246 if (cache->last_trans == 0) {
247 cache->last_trans = trans->transid;
251 if (cache->last_trans == trans->transid)
255 * detached nodes are used to avoid unnecessary backref
256 * lookup. transaction commit changes the extent tree.
257 * so the detached nodes are no longer useful.
259 while (!list_empty(&cache->detached)) {
260 node = list_entry(cache->detached.next,
261 struct btrfs_backref_node, list);
262 btrfs_backref_cleanup_node(cache, node);
265 while (!list_empty(&cache->changed)) {
266 node = list_entry(cache->changed.next,
267 struct btrfs_backref_node, list);
268 list_del_init(&node->list);
269 BUG_ON(node->pending);
270 update_backref_node(cache, node, node->new_bytenr);
274 * some nodes can be left in the pending list if there were
275 * errors during processing the pending nodes.
277 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
278 list_for_each_entry(node, &cache->pending[level], list) {
279 BUG_ON(!node->pending);
280 if (node->bytenr == node->new_bytenr)
282 update_backref_node(cache, node, node->new_bytenr);
286 cache->last_trans = 0;
290 static bool reloc_root_is_dead(struct btrfs_root *root)
293 * Pair with set_bit/clear_bit in clean_dirty_subvols and
294 * btrfs_update_reloc_root. We need to see the updated bit before
295 * trying to access reloc_root
298 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
304 * Check if this subvolume tree has valid reloc tree.
306 * Reloc tree after swap is considered dead, thus not considered as valid.
307 * This is enough for most callers, as they don't distinguish dead reloc root
308 * from no reloc root. But btrfs_should_ignore_reloc_root() below is a
311 static bool have_reloc_root(struct btrfs_root *root)
313 if (reloc_root_is_dead(root))
315 if (!root->reloc_root)
320 int btrfs_should_ignore_reloc_root(struct btrfs_root *root)
322 struct btrfs_root *reloc_root;
324 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
327 /* This root has been merged with its reloc tree, we can ignore it */
328 if (reloc_root_is_dead(root))
331 reloc_root = root->reloc_root;
335 if (btrfs_header_generation(reloc_root->commit_root) ==
336 root->fs_info->running_transaction->transid)
339 * if there is reloc tree and it was created in previous
340 * transaction backref lookup can find the reloc tree,
341 * so backref node for the fs tree root is useless for
348 * find reloc tree by address of tree root
350 struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
352 struct reloc_control *rc = fs_info->reloc_ctl;
353 struct rb_node *rb_node;
354 struct mapping_node *node;
355 struct btrfs_root *root = NULL;
358 spin_lock(&rc->reloc_root_tree.lock);
359 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
361 node = rb_entry(rb_node, struct mapping_node, rb_node);
362 root = (struct btrfs_root *)node->data;
364 spin_unlock(&rc->reloc_root_tree.lock);
365 return btrfs_grab_root(root);
368 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
371 struct btrfs_key key;
373 key.objectid = root_objectid;
374 key.type = BTRFS_ROOT_ITEM_KEY;
375 key.offset = (u64)-1;
377 return btrfs_get_fs_root(fs_info, &key, false);
381 * For useless nodes, do two major clean ups:
383 * - Cleanup the children edges and nodes
384 * If child node is also orphan (no parent) during cleanup, then the child
385 * node will also be cleaned up.
387 * - Freeing up leaves (level 0), keeps nodes detached
388 * For nodes, the node is still cached as "detached"
390 * Return false if @node is not in the @useless_nodes list.
391 * Return true if @node is in the @useless_nodes list.
393 static bool handle_useless_nodes(struct reloc_control *rc,
394 struct btrfs_backref_node *node)
396 struct btrfs_backref_cache *cache = &rc->backref_cache;
397 struct list_head *useless_node = &cache->useless_node;
400 while (!list_empty(useless_node)) {
401 struct btrfs_backref_node *cur;
403 cur = list_first_entry(useless_node, struct btrfs_backref_node,
405 list_del_init(&cur->list);
407 /* Only tree root nodes can be added to @useless_nodes */
408 ASSERT(list_empty(&cur->upper));
413 /* The node is the lowest node */
415 list_del_init(&cur->lower);
419 /* Cleanup the lower edges */
420 while (!list_empty(&cur->lower)) {
421 struct btrfs_backref_edge *edge;
422 struct btrfs_backref_node *lower;
424 edge = list_entry(cur->lower.next,
425 struct btrfs_backref_edge, list[UPPER]);
426 list_del(&edge->list[UPPER]);
427 list_del(&edge->list[LOWER]);
428 lower = edge->node[LOWER];
429 btrfs_backref_free_edge(cache, edge);
431 /* Child node is also orphan, queue for cleanup */
432 if (list_empty(&lower->upper))
433 list_add(&lower->list, useless_node);
435 /* Mark this block processed for relocation */
436 mark_block_processed(rc, cur);
439 * Backref nodes for tree leaves are deleted from the cache.
440 * Backref nodes for upper level tree blocks are left in the
441 * cache to avoid unnecessary backref lookup.
443 if (cur->level > 0) {
444 list_add(&cur->list, &cache->detached);
447 rb_erase(&cur->rb_node, &cache->rb_root);
448 btrfs_backref_free_node(cache, cur);
455 * Build backref tree for a given tree block. Root of the backref tree
456 * corresponds the tree block, leaves of the backref tree correspond roots of
457 * b-trees that reference the tree block.
459 * The basic idea of this function is check backrefs of a given block to find
460 * upper level blocks that reference the block, and then check backrefs of
461 * these upper level blocks recursively. The recursion stops when tree root is
462 * reached or backrefs for the block is cached.
464 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
465 * all upper level blocks that directly/indirectly reference the block are also
468 static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
469 struct reloc_control *rc, struct btrfs_key *node_key,
470 int level, u64 bytenr)
472 struct btrfs_backref_iter *iter;
473 struct btrfs_backref_cache *cache = &rc->backref_cache;
474 /* For searching parent of TREE_BLOCK_REF */
475 struct btrfs_path *path;
476 struct btrfs_backref_node *cur;
477 struct btrfs_backref_node *node = NULL;
478 struct btrfs_backref_edge *edge;
482 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS);
484 return ERR_PTR(-ENOMEM);
485 path = btrfs_alloc_path();
491 node = btrfs_backref_alloc_node(cache, bytenr, level);
500 /* Breadth-first search to build backref cache */
502 ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
508 edge = list_first_entry_or_null(&cache->pending_edge,
509 struct btrfs_backref_edge, list[UPPER]);
511 * The pending list isn't empty, take the first block to
515 list_del_init(&edge->list[UPPER]);
516 cur = edge->node[UPPER];
520 /* Finish the upper linkage of newly added edges/nodes */
521 ret = btrfs_backref_finish_upper_links(cache, node);
527 if (handle_useless_nodes(rc, node))
530 btrfs_backref_iter_free(iter);
531 btrfs_free_path(path);
533 btrfs_backref_error_cleanup(cache, node);
536 ASSERT(!node || !node->detached);
537 ASSERT(list_empty(&cache->useless_node) &&
538 list_empty(&cache->pending_edge));
543 * helper to add backref node for the newly created snapshot.
544 * the backref node is created by cloning backref node that
545 * corresponds to root of source tree
547 static int clone_backref_node(struct btrfs_trans_handle *trans,
548 struct reloc_control *rc,
549 struct btrfs_root *src,
550 struct btrfs_root *dest)
552 struct btrfs_root *reloc_root = src->reloc_root;
553 struct btrfs_backref_cache *cache = &rc->backref_cache;
554 struct btrfs_backref_node *node = NULL;
555 struct btrfs_backref_node *new_node;
556 struct btrfs_backref_edge *edge;
557 struct btrfs_backref_edge *new_edge;
558 struct rb_node *rb_node;
560 if (cache->last_trans > 0)
561 update_backref_cache(trans, cache);
563 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
565 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
569 BUG_ON(node->new_bytenr != reloc_root->node->start);
573 rb_node = rb_simple_search(&cache->rb_root,
574 reloc_root->commit_root->start);
576 node = rb_entry(rb_node, struct btrfs_backref_node,
578 BUG_ON(node->detached);
585 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
590 new_node->lowest = node->lowest;
591 new_node->checked = 1;
592 new_node->root = btrfs_grab_root(dest);
593 ASSERT(new_node->root);
596 list_for_each_entry(edge, &node->lower, list[UPPER]) {
597 new_edge = btrfs_backref_alloc_edge(cache);
601 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
602 new_node, LINK_UPPER);
605 list_add_tail(&new_node->lower, &cache->leaves);
608 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
611 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
613 if (!new_node->lowest) {
614 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
615 list_add_tail(&new_edge->list[LOWER],
616 &new_edge->node[LOWER]->upper);
621 while (!list_empty(&new_node->lower)) {
622 new_edge = list_entry(new_node->lower.next,
623 struct btrfs_backref_edge, list[UPPER]);
624 list_del(&new_edge->list[UPPER]);
625 btrfs_backref_free_edge(cache, new_edge);
627 btrfs_backref_free_node(cache, new_node);
632 * helper to add 'address of tree root -> reloc tree' mapping
634 static int __must_check __add_reloc_root(struct btrfs_root *root)
636 struct btrfs_fs_info *fs_info = root->fs_info;
637 struct rb_node *rb_node;
638 struct mapping_node *node;
639 struct reloc_control *rc = fs_info->reloc_ctl;
641 node = kmalloc(sizeof(*node), GFP_NOFS);
645 node->bytenr = root->commit_root->start;
648 spin_lock(&rc->reloc_root_tree.lock);
649 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
650 node->bytenr, &node->rb_node);
651 spin_unlock(&rc->reloc_root_tree.lock);
653 btrfs_panic(fs_info, -EEXIST,
654 "Duplicate root found for start=%llu while inserting into relocation tree",
658 list_add_tail(&root->root_list, &rc->reloc_roots);
663 * helper to delete the 'address of tree root -> reloc tree'
666 static void __del_reloc_root(struct btrfs_root *root)
668 struct btrfs_fs_info *fs_info = root->fs_info;
669 struct rb_node *rb_node;
670 struct mapping_node *node = NULL;
671 struct reloc_control *rc = fs_info->reloc_ctl;
672 bool put_ref = false;
674 if (rc && root->node) {
675 spin_lock(&rc->reloc_root_tree.lock);
676 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
677 root->commit_root->start);
679 node = rb_entry(rb_node, struct mapping_node, rb_node);
680 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
681 RB_CLEAR_NODE(&node->rb_node);
683 spin_unlock(&rc->reloc_root_tree.lock);
686 BUG_ON((struct btrfs_root *)node->data != root);
690 * We only put the reloc root here if it's on the list. There's a lot
691 * of places where the pattern is to splice the rc->reloc_roots, process
692 * the reloc roots, and then add the reloc root back onto
693 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
694 * list we don't want the reference being dropped, because the guy
695 * messing with the list is in charge of the reference.
697 spin_lock(&fs_info->trans_lock);
698 if (!list_empty(&root->root_list)) {
700 list_del_init(&root->root_list);
702 spin_unlock(&fs_info->trans_lock);
704 btrfs_put_root(root);
709 * helper to update the 'address of tree root -> reloc tree'
712 static int __update_reloc_root(struct btrfs_root *root)
714 struct btrfs_fs_info *fs_info = root->fs_info;
715 struct rb_node *rb_node;
716 struct mapping_node *node = NULL;
717 struct reloc_control *rc = fs_info->reloc_ctl;
719 spin_lock(&rc->reloc_root_tree.lock);
720 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
721 root->commit_root->start);
723 node = rb_entry(rb_node, struct mapping_node, rb_node);
724 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
726 spin_unlock(&rc->reloc_root_tree.lock);
730 BUG_ON((struct btrfs_root *)node->data != root);
732 spin_lock(&rc->reloc_root_tree.lock);
733 node->bytenr = root->node->start;
734 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
735 node->bytenr, &node->rb_node);
736 spin_unlock(&rc->reloc_root_tree.lock);
738 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
742 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 objectid)
745 struct btrfs_fs_info *fs_info = root->fs_info;
746 struct btrfs_root *reloc_root;
747 struct extent_buffer *eb;
748 struct btrfs_root_item *root_item;
749 struct btrfs_key root_key;
752 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
755 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
756 root_key.type = BTRFS_ROOT_ITEM_KEY;
757 root_key.offset = objectid;
759 if (root->root_key.objectid == objectid) {
762 /* called by btrfs_init_reloc_root */
763 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
764 BTRFS_TREE_RELOC_OBJECTID);
767 * Set the last_snapshot field to the generation of the commit
768 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
769 * correctly (returns true) when the relocation root is created
770 * either inside the critical section of a transaction commit
771 * (through transaction.c:qgroup_account_snapshot()) and when
772 * it's created before the transaction commit is started.
774 commit_root_gen = btrfs_header_generation(root->commit_root);
775 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
778 * called by btrfs_reloc_post_snapshot_hook.
779 * the source tree is a reloc tree, all tree blocks
780 * modified after it was created have RELOC flag
781 * set in their headers. so it's OK to not update
782 * the 'last_snapshot'.
784 ret = btrfs_copy_root(trans, root, root->node, &eb,
785 BTRFS_TREE_RELOC_OBJECTID);
789 memcpy(root_item, &root->root_item, sizeof(*root_item));
790 btrfs_set_root_bytenr(root_item, eb->start);
791 btrfs_set_root_level(root_item, btrfs_header_level(eb));
792 btrfs_set_root_generation(root_item, trans->transid);
794 if (root->root_key.objectid == objectid) {
795 btrfs_set_root_refs(root_item, 0);
796 memset(&root_item->drop_progress, 0,
797 sizeof(struct btrfs_disk_key));
798 root_item->drop_level = 0;
801 btrfs_tree_unlock(eb);
802 free_extent_buffer(eb);
804 ret = btrfs_insert_root(trans, fs_info->tree_root,
805 &root_key, root_item);
809 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
810 BUG_ON(IS_ERR(reloc_root));
811 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
812 reloc_root->last_trans = trans->transid;
817 * create reloc tree for a given fs tree. reloc tree is just a
818 * snapshot of the fs tree with special root objectid.
820 * The reloc_root comes out of here with two references, one for
821 * root->reloc_root, and another for being on the rc->reloc_roots list.
823 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
824 struct btrfs_root *root)
826 struct btrfs_fs_info *fs_info = root->fs_info;
827 struct btrfs_root *reloc_root;
828 struct reloc_control *rc = fs_info->reloc_ctl;
829 struct btrfs_block_rsv *rsv;
837 * The subvolume has reloc tree but the swap is finished, no need to
838 * create/update the dead reloc tree
840 if (reloc_root_is_dead(root))
844 * This is subtle but important. We do not do
845 * record_root_in_transaction for reloc roots, instead we record their
846 * corresponding fs root, and then here we update the last trans for the
847 * reloc root. This means that we have to do this for the entire life
848 * of the reloc root, regardless of which stage of the relocation we are
851 if (root->reloc_root) {
852 reloc_root = root->reloc_root;
853 reloc_root->last_trans = trans->transid;
858 * We are merging reloc roots, we do not need new reloc trees. Also
859 * reloc trees never need their own reloc tree.
861 if (!rc->create_reloc_tree ||
862 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
865 if (!trans->reloc_reserved) {
866 rsv = trans->block_rsv;
867 trans->block_rsv = rc->block_rsv;
870 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
872 trans->block_rsv = rsv;
874 ret = __add_reloc_root(reloc_root);
876 root->reloc_root = btrfs_grab_root(reloc_root);
881 * update root item of reloc tree
883 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
884 struct btrfs_root *root)
886 struct btrfs_fs_info *fs_info = root->fs_info;
887 struct btrfs_root *reloc_root;
888 struct btrfs_root_item *root_item;
891 if (!have_reloc_root(root))
894 reloc_root = root->reloc_root;
895 root_item = &reloc_root->root_item;
898 * We are probably ok here, but __del_reloc_root() will drop its ref of
899 * the root. We have the ref for root->reloc_root, but just in case
900 * hold it while we update the reloc root.
902 btrfs_grab_root(reloc_root);
904 /* root->reloc_root will stay until current relocation finished */
905 if (fs_info->reloc_ctl->merge_reloc_tree &&
906 btrfs_root_refs(root_item) == 0) {
907 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
909 * Mark the tree as dead before we change reloc_root so
910 * have_reloc_root will not touch it from now on.
913 __del_reloc_root(reloc_root);
916 if (reloc_root->commit_root != reloc_root->node) {
917 __update_reloc_root(reloc_root);
918 btrfs_set_root_node(root_item, reloc_root->node);
919 free_extent_buffer(reloc_root->commit_root);
920 reloc_root->commit_root = btrfs_root_node(reloc_root);
923 ret = btrfs_update_root(trans, fs_info->tree_root,
924 &reloc_root->root_key, root_item);
926 btrfs_put_root(reloc_root);
932 * helper to find first cached inode with inode number >= objectid
935 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
937 struct rb_node *node;
938 struct rb_node *prev;
939 struct btrfs_inode *entry;
942 spin_lock(&root->inode_lock);
944 node = root->inode_tree.rb_node;
948 entry = rb_entry(node, struct btrfs_inode, rb_node);
950 if (objectid < btrfs_ino(entry))
951 node = node->rb_left;
952 else if (objectid > btrfs_ino(entry))
953 node = node->rb_right;
959 entry = rb_entry(prev, struct btrfs_inode, rb_node);
960 if (objectid <= btrfs_ino(entry)) {
964 prev = rb_next(prev);
968 entry = rb_entry(node, struct btrfs_inode, rb_node);
969 inode = igrab(&entry->vfs_inode);
971 spin_unlock(&root->inode_lock);
975 objectid = btrfs_ino(entry) + 1;
976 if (cond_resched_lock(&root->inode_lock))
979 node = rb_next(node);
981 spin_unlock(&root->inode_lock);
986 * get new location of data
988 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
989 u64 bytenr, u64 num_bytes)
991 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
992 struct btrfs_path *path;
993 struct btrfs_file_extent_item *fi;
994 struct extent_buffer *leaf;
997 path = btrfs_alloc_path();
1001 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1002 ret = btrfs_lookup_file_extent(NULL, root, path,
1003 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1011 leaf = path->nodes[0];
1012 fi = btrfs_item_ptr(leaf, path->slots[0],
1013 struct btrfs_file_extent_item);
1015 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1016 btrfs_file_extent_compression(leaf, fi) ||
1017 btrfs_file_extent_encryption(leaf, fi) ||
1018 btrfs_file_extent_other_encoding(leaf, fi));
1020 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1025 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1028 btrfs_free_path(path);
1033 * update file extent items in the tree leaf to point to
1034 * the new locations.
1036 static noinline_for_stack
1037 int replace_file_extents(struct btrfs_trans_handle *trans,
1038 struct reloc_control *rc,
1039 struct btrfs_root *root,
1040 struct extent_buffer *leaf)
1042 struct btrfs_fs_info *fs_info = root->fs_info;
1043 struct btrfs_key key;
1044 struct btrfs_file_extent_item *fi;
1045 struct inode *inode = NULL;
1057 if (rc->stage != UPDATE_DATA_PTRS)
1060 /* reloc trees always use full backref */
1061 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1062 parent = leaf->start;
1066 nritems = btrfs_header_nritems(leaf);
1067 for (i = 0; i < nritems; i++) {
1068 struct btrfs_ref ref = { 0 };
1071 btrfs_item_key_to_cpu(leaf, &key, i);
1072 if (key.type != BTRFS_EXTENT_DATA_KEY)
1074 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1075 if (btrfs_file_extent_type(leaf, fi) ==
1076 BTRFS_FILE_EXTENT_INLINE)
1078 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1079 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1082 if (!in_range(bytenr, rc->block_group->start,
1083 rc->block_group->length))
1087 * if we are modifying block in fs tree, wait for readpage
1088 * to complete and drop the extent cache
1090 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1092 inode = find_next_inode(root, key.objectid);
1094 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1095 btrfs_add_delayed_iput(inode);
1096 inode = find_next_inode(root, key.objectid);
1098 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1100 btrfs_file_extent_num_bytes(leaf, fi);
1101 WARN_ON(!IS_ALIGNED(key.offset,
1102 fs_info->sectorsize));
1103 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1105 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1110 btrfs_drop_extent_cache(BTRFS_I(inode),
1111 key.offset, end, 1);
1112 unlock_extent(&BTRFS_I(inode)->io_tree,
1117 ret = get_new_location(rc->data_inode, &new_bytenr,
1121 * Don't have to abort since we've not changed anything
1122 * in the file extent yet.
1127 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1130 key.offset -= btrfs_file_extent_offset(leaf, fi);
1131 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1133 ref.real_root = root->root_key.objectid;
1134 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1135 key.objectid, key.offset);
1136 ret = btrfs_inc_extent_ref(trans, &ref);
1138 btrfs_abort_transaction(trans, ret);
1142 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1144 ref.real_root = root->root_key.objectid;
1145 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1146 key.objectid, key.offset);
1147 ret = btrfs_free_extent(trans, &ref);
1149 btrfs_abort_transaction(trans, ret);
1154 btrfs_mark_buffer_dirty(leaf);
1156 btrfs_add_delayed_iput(inode);
1160 static noinline_for_stack
1161 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1162 struct btrfs_path *path, int level)
1164 struct btrfs_disk_key key1;
1165 struct btrfs_disk_key key2;
1166 btrfs_node_key(eb, &key1, slot);
1167 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1168 return memcmp(&key1, &key2, sizeof(key1));
1172 * try to replace tree blocks in fs tree with the new blocks
1173 * in reloc tree. tree blocks haven't been modified since the
1174 * reloc tree was create can be replaced.
1176 * if a block was replaced, level of the block + 1 is returned.
1177 * if no block got replaced, 0 is returned. if there are other
1178 * errors, a negative error number is returned.
1180 static noinline_for_stack
1181 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1182 struct btrfs_root *dest, struct btrfs_root *src,
1183 struct btrfs_path *path, struct btrfs_key *next_key,
1184 int lowest_level, int max_level)
1186 struct btrfs_fs_info *fs_info = dest->fs_info;
1187 struct extent_buffer *eb;
1188 struct extent_buffer *parent;
1189 struct btrfs_ref ref = { 0 };
1190 struct btrfs_key key;
1202 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1203 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1205 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1207 slot = path->slots[lowest_level];
1208 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1210 eb = btrfs_lock_root_node(dest);
1211 btrfs_set_lock_blocking_write(eb);
1212 level = btrfs_header_level(eb);
1214 if (level < lowest_level) {
1215 btrfs_tree_unlock(eb);
1216 free_extent_buffer(eb);
1221 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1224 btrfs_set_lock_blocking_write(eb);
1227 next_key->objectid = (u64)-1;
1228 next_key->type = (u8)-1;
1229 next_key->offset = (u64)-1;
1234 struct btrfs_key first_key;
1236 level = btrfs_header_level(parent);
1237 BUG_ON(level < lowest_level);
1239 ret = btrfs_bin_search(parent, &key, &slot);
1242 if (ret && slot > 0)
1245 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1246 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1248 old_bytenr = btrfs_node_blockptr(parent, slot);
1249 blocksize = fs_info->nodesize;
1250 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1251 btrfs_node_key_to_cpu(parent, &first_key, slot);
1253 if (level <= max_level) {
1254 eb = path->nodes[level];
1255 new_bytenr = btrfs_node_blockptr(eb,
1256 path->slots[level]);
1257 new_ptr_gen = btrfs_node_ptr_generation(eb,
1258 path->slots[level]);
1264 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1269 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1270 memcmp_node_keys(parent, slot, path, level)) {
1271 if (level <= lowest_level) {
1276 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1277 level - 1, &first_key);
1281 } else if (!extent_buffer_uptodate(eb)) {
1283 free_extent_buffer(eb);
1286 btrfs_tree_lock(eb);
1288 ret = btrfs_cow_block(trans, dest, eb, parent,
1292 btrfs_set_lock_blocking_write(eb);
1294 btrfs_tree_unlock(parent);
1295 free_extent_buffer(parent);
1302 btrfs_tree_unlock(parent);
1303 free_extent_buffer(parent);
1308 btrfs_node_key_to_cpu(path->nodes[level], &key,
1309 path->slots[level]);
1310 btrfs_release_path(path);
1312 path->lowest_level = level;
1313 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1314 path->lowest_level = 0;
1318 * Info qgroup to trace both subtrees.
1320 * We must trace both trees.
1321 * 1) Tree reloc subtree
1322 * If not traced, we will leak data numbers
1324 * If not traced, we will double count old data
1326 * We don't scan the subtree right now, but only record
1327 * the swapped tree blocks.
1328 * The real subtree rescan is delayed until we have new
1329 * CoW on the subtree root node before transaction commit.
1331 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1332 rc->block_group, parent, slot,
1333 path->nodes[level], path->slots[level],
1338 * swap blocks in fs tree and reloc tree.
1340 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1341 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1342 btrfs_mark_buffer_dirty(parent);
1344 btrfs_set_node_blockptr(path->nodes[level],
1345 path->slots[level], old_bytenr);
1346 btrfs_set_node_ptr_generation(path->nodes[level],
1347 path->slots[level], old_ptr_gen);
1348 btrfs_mark_buffer_dirty(path->nodes[level]);
1350 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1351 blocksize, path->nodes[level]->start);
1352 ref.skip_qgroup = true;
1353 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1354 ret = btrfs_inc_extent_ref(trans, &ref);
1356 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1358 ref.skip_qgroup = true;
1359 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1360 ret = btrfs_inc_extent_ref(trans, &ref);
1363 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1364 blocksize, path->nodes[level]->start);
1365 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1366 ref.skip_qgroup = true;
1367 ret = btrfs_free_extent(trans, &ref);
1370 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1372 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1373 ref.skip_qgroup = true;
1374 ret = btrfs_free_extent(trans, &ref);
1377 btrfs_unlock_up_safe(path, 0);
1382 btrfs_tree_unlock(parent);
1383 free_extent_buffer(parent);
1388 * helper to find next relocated block in reloc tree
1390 static noinline_for_stack
1391 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1394 struct extent_buffer *eb;
1399 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1401 for (i = 0; i < *level; i++) {
1402 free_extent_buffer(path->nodes[i]);
1403 path->nodes[i] = NULL;
1406 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1407 eb = path->nodes[i];
1408 nritems = btrfs_header_nritems(eb);
1409 while (path->slots[i] + 1 < nritems) {
1411 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1418 free_extent_buffer(path->nodes[i]);
1419 path->nodes[i] = NULL;
1425 * walk down reloc tree to find relocated block of lowest level
1427 static noinline_for_stack
1428 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1431 struct btrfs_fs_info *fs_info = root->fs_info;
1432 struct extent_buffer *eb = NULL;
1439 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1441 for (i = *level; i > 0; i--) {
1442 struct btrfs_key first_key;
1444 eb = path->nodes[i];
1445 nritems = btrfs_header_nritems(eb);
1446 while (path->slots[i] < nritems) {
1447 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1448 if (ptr_gen > last_snapshot)
1452 if (path->slots[i] >= nritems) {
1463 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1464 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
1465 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
1469 } else if (!extent_buffer_uptodate(eb)) {
1470 free_extent_buffer(eb);
1473 BUG_ON(btrfs_header_level(eb) != i - 1);
1474 path->nodes[i - 1] = eb;
1475 path->slots[i - 1] = 0;
1481 * invalidate extent cache for file extents whose key in range of
1482 * [min_key, max_key)
1484 static int invalidate_extent_cache(struct btrfs_root *root,
1485 struct btrfs_key *min_key,
1486 struct btrfs_key *max_key)
1488 struct btrfs_fs_info *fs_info = root->fs_info;
1489 struct inode *inode = NULL;
1494 objectid = min_key->objectid;
1499 if (objectid > max_key->objectid)
1502 inode = find_next_inode(root, objectid);
1505 ino = btrfs_ino(BTRFS_I(inode));
1507 if (ino > max_key->objectid) {
1513 if (!S_ISREG(inode->i_mode))
1516 if (unlikely(min_key->objectid == ino)) {
1517 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1519 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1522 start = min_key->offset;
1523 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1529 if (unlikely(max_key->objectid == ino)) {
1530 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1532 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1535 if (max_key->offset == 0)
1537 end = max_key->offset;
1538 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1545 /* the lock_extent waits for readpage to complete */
1546 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1547 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1548 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1553 static int find_next_key(struct btrfs_path *path, int level,
1554 struct btrfs_key *key)
1557 while (level < BTRFS_MAX_LEVEL) {
1558 if (!path->nodes[level])
1560 if (path->slots[level] + 1 <
1561 btrfs_header_nritems(path->nodes[level])) {
1562 btrfs_node_key_to_cpu(path->nodes[level], key,
1563 path->slots[level] + 1);
1572 * Insert current subvolume into reloc_control::dirty_subvol_roots
1574 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1575 struct reloc_control *rc,
1576 struct btrfs_root *root)
1578 struct btrfs_root *reloc_root = root->reloc_root;
1579 struct btrfs_root_item *reloc_root_item;
1581 /* @root must be a subvolume tree root with a valid reloc tree */
1582 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1585 reloc_root_item = &reloc_root->root_item;
1586 memset(&reloc_root_item->drop_progress, 0,
1587 sizeof(reloc_root_item->drop_progress));
1588 reloc_root_item->drop_level = 0;
1589 btrfs_set_root_refs(reloc_root_item, 0);
1590 btrfs_update_reloc_root(trans, root);
1592 if (list_empty(&root->reloc_dirty_list)) {
1593 btrfs_grab_root(root);
1594 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1598 static int clean_dirty_subvols(struct reloc_control *rc)
1600 struct btrfs_root *root;
1601 struct btrfs_root *next;
1605 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1607 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1608 /* Merged subvolume, cleanup its reloc root */
1609 struct btrfs_root *reloc_root = root->reloc_root;
1611 list_del_init(&root->reloc_dirty_list);
1612 root->reloc_root = NULL;
1614 * Need barrier to ensure clear_bit() only happens after
1615 * root->reloc_root = NULL. Pairs with have_reloc_root.
1618 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1621 * btrfs_drop_snapshot drops our ref we hold for
1622 * ->reloc_root. If it fails however we must
1623 * drop the ref ourselves.
1625 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1627 btrfs_put_root(reloc_root);
1632 btrfs_put_root(root);
1634 /* Orphan reloc tree, just clean it up */
1635 ret2 = btrfs_drop_snapshot(root, 0, 1);
1637 btrfs_put_root(root);
1647 * merge the relocated tree blocks in reloc tree with corresponding
1650 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1651 struct btrfs_root *root)
1653 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1654 struct btrfs_key key;
1655 struct btrfs_key next_key;
1656 struct btrfs_trans_handle *trans = NULL;
1657 struct btrfs_root *reloc_root;
1658 struct btrfs_root_item *root_item;
1659 struct btrfs_path *path;
1660 struct extent_buffer *leaf;
1668 path = btrfs_alloc_path();
1671 path->reada = READA_FORWARD;
1673 reloc_root = root->reloc_root;
1674 root_item = &reloc_root->root_item;
1676 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1677 level = btrfs_root_level(root_item);
1678 atomic_inc(&reloc_root->node->refs);
1679 path->nodes[level] = reloc_root->node;
1680 path->slots[level] = 0;
1682 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1684 level = root_item->drop_level;
1686 path->lowest_level = level;
1687 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1688 path->lowest_level = 0;
1690 btrfs_free_path(path);
1694 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1695 path->slots[level]);
1696 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1698 btrfs_unlock_up_safe(path, 0);
1701 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1702 memset(&next_key, 0, sizeof(next_key));
1705 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1706 BTRFS_RESERVE_FLUSH_ALL);
1711 trans = btrfs_start_transaction(root, 0);
1712 if (IS_ERR(trans)) {
1713 err = PTR_ERR(trans);
1719 * At this point we no longer have a reloc_control, so we can't
1720 * depend on btrfs_init_reloc_root to update our last_trans.
1722 * But that's ok, we started the trans handle on our
1723 * corresponding fs_root, which means it's been added to the
1724 * dirty list. At commit time we'll still call
1725 * btrfs_update_reloc_root() and update our root item
1728 reloc_root->last_trans = trans->transid;
1729 trans->block_rsv = rc->block_rsv;
1734 ret = walk_down_reloc_tree(reloc_root, path, &level);
1742 if (!find_next_key(path, level, &key) &&
1743 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1746 ret = replace_path(trans, rc, root, reloc_root, path,
1747 &next_key, level, max_level);
1756 btrfs_node_key_to_cpu(path->nodes[level], &key,
1757 path->slots[level]);
1761 ret = walk_up_reloc_tree(reloc_root, path, &level);
1767 * save the merging progress in the drop_progress.
1768 * this is OK since root refs == 1 in this case.
1770 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1771 path->slots[level]);
1772 root_item->drop_level = level;
1774 btrfs_end_transaction_throttle(trans);
1777 btrfs_btree_balance_dirty(fs_info);
1779 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1780 invalidate_extent_cache(root, &key, &next_key);
1784 * handle the case only one block in the fs tree need to be
1785 * relocated and the block is tree root.
1787 leaf = btrfs_lock_root_node(root);
1788 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
1789 btrfs_tree_unlock(leaf);
1790 free_extent_buffer(leaf);
1794 btrfs_free_path(path);
1797 insert_dirty_subvol(trans, rc, root);
1800 btrfs_end_transaction_throttle(trans);
1802 btrfs_btree_balance_dirty(fs_info);
1804 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1805 invalidate_extent_cache(root, &key, &next_key);
1810 static noinline_for_stack
1811 int prepare_to_merge(struct reloc_control *rc, int err)
1813 struct btrfs_root *root = rc->extent_root;
1814 struct btrfs_fs_info *fs_info = root->fs_info;
1815 struct btrfs_root *reloc_root;
1816 struct btrfs_trans_handle *trans;
1817 LIST_HEAD(reloc_roots);
1821 mutex_lock(&fs_info->reloc_mutex);
1822 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1823 rc->merging_rsv_size += rc->nodes_relocated * 2;
1824 mutex_unlock(&fs_info->reloc_mutex);
1828 num_bytes = rc->merging_rsv_size;
1829 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1830 BTRFS_RESERVE_FLUSH_ALL);
1835 trans = btrfs_join_transaction(rc->extent_root);
1836 if (IS_ERR(trans)) {
1838 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1840 return PTR_ERR(trans);
1844 if (num_bytes != rc->merging_rsv_size) {
1845 btrfs_end_transaction(trans);
1846 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1852 rc->merge_reloc_tree = 1;
1854 while (!list_empty(&rc->reloc_roots)) {
1855 reloc_root = list_entry(rc->reloc_roots.next,
1856 struct btrfs_root, root_list);
1857 list_del_init(&reloc_root->root_list);
1859 root = read_fs_root(fs_info, reloc_root->root_key.offset);
1860 BUG_ON(IS_ERR(root));
1861 BUG_ON(root->reloc_root != reloc_root);
1864 * set reference count to 1, so btrfs_recover_relocation
1865 * knows it should resumes merging
1868 btrfs_set_root_refs(&reloc_root->root_item, 1);
1869 btrfs_update_reloc_root(trans, root);
1871 list_add(&reloc_root->root_list, &reloc_roots);
1872 btrfs_put_root(root);
1875 list_splice(&reloc_roots, &rc->reloc_roots);
1878 btrfs_commit_transaction(trans);
1880 btrfs_end_transaction(trans);
1884 static noinline_for_stack
1885 void free_reloc_roots(struct list_head *list)
1887 struct btrfs_root *reloc_root, *tmp;
1889 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1890 __del_reloc_root(reloc_root);
1893 static noinline_for_stack
1894 void merge_reloc_roots(struct reloc_control *rc)
1896 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1897 struct btrfs_root *root;
1898 struct btrfs_root *reloc_root;
1899 LIST_HEAD(reloc_roots);
1903 root = rc->extent_root;
1906 * this serializes us with btrfs_record_root_in_transaction,
1907 * we have to make sure nobody is in the middle of
1908 * adding their roots to the list while we are
1911 mutex_lock(&fs_info->reloc_mutex);
1912 list_splice_init(&rc->reloc_roots, &reloc_roots);
1913 mutex_unlock(&fs_info->reloc_mutex);
1915 while (!list_empty(&reloc_roots)) {
1917 reloc_root = list_entry(reloc_roots.next,
1918 struct btrfs_root, root_list);
1920 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1921 root = read_fs_root(fs_info,
1922 reloc_root->root_key.offset);
1923 BUG_ON(IS_ERR(root));
1924 BUG_ON(root->reloc_root != reloc_root);
1926 ret = merge_reloc_root(rc, root);
1927 btrfs_put_root(root);
1929 if (list_empty(&reloc_root->root_list))
1930 list_add_tail(&reloc_root->root_list,
1935 list_del_init(&reloc_root->root_list);
1936 /* Don't forget to queue this reloc root for cleanup */
1937 list_add_tail(&reloc_root->reloc_dirty_list,
1938 &rc->dirty_subvol_roots);
1948 btrfs_handle_fs_error(fs_info, ret, NULL);
1949 free_reloc_roots(&reloc_roots);
1951 /* new reloc root may be added */
1952 mutex_lock(&fs_info->reloc_mutex);
1953 list_splice_init(&rc->reloc_roots, &reloc_roots);
1954 mutex_unlock(&fs_info->reloc_mutex);
1955 free_reloc_roots(&reloc_roots);
1961 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1963 * here, but it's wrong. If we fail to start the transaction in
1964 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1965 * have actually been removed from the reloc_root_tree rb tree. This is
1966 * fine because we're bailing here, and we hold a reference on the root
1967 * for the list that holds it, so these roots will be cleaned up when we
1968 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1969 * will be cleaned up on unmount.
1971 * The remaining nodes will be cleaned up by free_reloc_control.
1975 static void free_block_list(struct rb_root *blocks)
1977 struct tree_block *block;
1978 struct rb_node *rb_node;
1979 while ((rb_node = rb_first(blocks))) {
1980 block = rb_entry(rb_node, struct tree_block, rb_node);
1981 rb_erase(rb_node, blocks);
1986 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1987 struct btrfs_root *reloc_root)
1989 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
1990 struct btrfs_root *root;
1993 if (reloc_root->last_trans == trans->transid)
1996 root = read_fs_root(fs_info, reloc_root->root_key.offset);
1997 BUG_ON(IS_ERR(root));
1998 BUG_ON(root->reloc_root != reloc_root);
1999 ret = btrfs_record_root_in_trans(trans, root);
2000 btrfs_put_root(root);
2005 static noinline_for_stack
2006 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2007 struct reloc_control *rc,
2008 struct btrfs_backref_node *node,
2009 struct btrfs_backref_edge *edges[])
2011 struct btrfs_backref_node *next;
2012 struct btrfs_root *root;
2018 next = walk_up_backref(next, edges, &index);
2021 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2023 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2024 record_reloc_root_in_trans(trans, root);
2028 btrfs_record_root_in_trans(trans, root);
2029 root = root->reloc_root;
2031 if (next->new_bytenr != root->node->start) {
2032 BUG_ON(next->new_bytenr);
2033 BUG_ON(!list_empty(&next->list));
2034 next->new_bytenr = root->node->start;
2035 btrfs_put_root(next->root);
2036 next->root = btrfs_grab_root(root);
2038 list_add_tail(&next->list,
2039 &rc->backref_cache.changed);
2040 mark_block_processed(rc, next);
2046 next = walk_down_backref(edges, &index);
2047 if (!next || next->level <= node->level)
2054 /* setup backref node path for btrfs_reloc_cow_block */
2056 rc->backref_cache.path[next->level] = next;
2059 next = edges[index]->node[UPPER];
2065 * Select a tree root for relocation.
2067 * Return NULL if the block is not shareable. We should use do_relocation() in
2070 * Return a tree root pointer if the block is shareable.
2071 * Return -ENOENT if the block is root of reloc tree.
2073 static noinline_for_stack
2074 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2076 struct btrfs_backref_node *next;
2077 struct btrfs_root *root;
2078 struct btrfs_root *fs_root = NULL;
2079 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2085 next = walk_up_backref(next, edges, &index);
2089 /* No other choice for non-shareable tree */
2090 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2093 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2099 next = walk_down_backref(edges, &index);
2100 if (!next || next->level <= node->level)
2105 return ERR_PTR(-ENOENT);
2109 static noinline_for_stack
2110 u64 calcu_metadata_size(struct reloc_control *rc,
2111 struct btrfs_backref_node *node, int reserve)
2113 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2114 struct btrfs_backref_node *next = node;
2115 struct btrfs_backref_edge *edge;
2116 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2120 BUG_ON(reserve && node->processed);
2125 if (next->processed && (reserve || next != node))
2128 num_bytes += fs_info->nodesize;
2130 if (list_empty(&next->upper))
2133 edge = list_entry(next->upper.next,
2134 struct btrfs_backref_edge, list[LOWER]);
2135 edges[index++] = edge;
2136 next = edge->node[UPPER];
2138 next = walk_down_backref(edges, &index);
2143 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2144 struct reloc_control *rc,
2145 struct btrfs_backref_node *node)
2147 struct btrfs_root *root = rc->extent_root;
2148 struct btrfs_fs_info *fs_info = root->fs_info;
2153 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2155 trans->block_rsv = rc->block_rsv;
2156 rc->reserved_bytes += num_bytes;
2159 * We are under a transaction here so we can only do limited flushing.
2160 * If we get an enospc just kick back -EAGAIN so we know to drop the
2161 * transaction and try to refill when we can flush all the things.
2163 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2164 BTRFS_RESERVE_FLUSH_LIMIT);
2166 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2167 while (tmp <= rc->reserved_bytes)
2170 * only one thread can access block_rsv at this point,
2171 * so we don't need hold lock to protect block_rsv.
2172 * we expand more reservation size here to allow enough
2173 * space for relocation and we will return earlier in
2176 rc->block_rsv->size = tmp + fs_info->nodesize *
2177 RELOCATION_RESERVED_NODES;
2185 * relocate a block tree, and then update pointers in upper level
2186 * blocks that reference the block to point to the new location.
2188 * if called by link_to_upper, the block has already been relocated.
2189 * in that case this function just updates pointers.
2191 static int do_relocation(struct btrfs_trans_handle *trans,
2192 struct reloc_control *rc,
2193 struct btrfs_backref_node *node,
2194 struct btrfs_key *key,
2195 struct btrfs_path *path, int lowest)
2197 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2198 struct btrfs_backref_node *upper;
2199 struct btrfs_backref_edge *edge;
2200 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2201 struct btrfs_root *root;
2202 struct extent_buffer *eb;
2210 BUG_ON(lowest && node->eb);
2212 path->lowest_level = node->level + 1;
2213 rc->backref_cache.path[node->level] = node;
2214 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2215 struct btrfs_key first_key;
2216 struct btrfs_ref ref = { 0 };
2220 upper = edge->node[UPPER];
2221 root = select_reloc_root(trans, rc, upper, edges);
2224 if (upper->eb && !upper->locked) {
2226 ret = btrfs_bin_search(upper->eb, key, &slot);
2232 bytenr = btrfs_node_blockptr(upper->eb, slot);
2233 if (node->eb->start == bytenr)
2236 btrfs_backref_drop_node_buffer(upper);
2240 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2247 btrfs_release_path(path);
2252 upper->eb = path->nodes[upper->level];
2253 path->nodes[upper->level] = NULL;
2255 BUG_ON(upper->eb != path->nodes[upper->level]);
2259 path->locks[upper->level] = 0;
2261 slot = path->slots[upper->level];
2262 btrfs_release_path(path);
2264 ret = btrfs_bin_search(upper->eb, key, &slot);
2272 bytenr = btrfs_node_blockptr(upper->eb, slot);
2274 if (bytenr != node->bytenr) {
2275 btrfs_err(root->fs_info,
2276 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2277 bytenr, node->bytenr, slot,
2283 if (node->eb->start == bytenr)
2287 blocksize = root->fs_info->nodesize;
2288 generation = btrfs_node_ptr_generation(upper->eb, slot);
2289 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2290 eb = read_tree_block(fs_info, bytenr, generation,
2291 upper->level - 1, &first_key);
2295 } else if (!extent_buffer_uptodate(eb)) {
2296 free_extent_buffer(eb);
2300 btrfs_tree_lock(eb);
2301 btrfs_set_lock_blocking_write(eb);
2304 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2306 btrfs_tree_unlock(eb);
2307 free_extent_buffer(eb);
2312 BUG_ON(node->eb != eb);
2314 btrfs_set_node_blockptr(upper->eb, slot,
2316 btrfs_set_node_ptr_generation(upper->eb, slot,
2318 btrfs_mark_buffer_dirty(upper->eb);
2320 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2321 node->eb->start, blocksize,
2323 ref.real_root = root->root_key.objectid;
2324 btrfs_init_tree_ref(&ref, node->level,
2325 btrfs_header_owner(upper->eb));
2326 ret = btrfs_inc_extent_ref(trans, &ref);
2329 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2333 if (!upper->pending)
2334 btrfs_backref_drop_node_buffer(upper);
2336 btrfs_backref_unlock_node_buffer(upper);
2341 if (!err && node->pending) {
2342 btrfs_backref_drop_node_buffer(node);
2343 list_move_tail(&node->list, &rc->backref_cache.changed);
2347 path->lowest_level = 0;
2348 BUG_ON(err == -ENOSPC);
2352 static int link_to_upper(struct btrfs_trans_handle *trans,
2353 struct reloc_control *rc,
2354 struct btrfs_backref_node *node,
2355 struct btrfs_path *path)
2357 struct btrfs_key key;
2359 btrfs_node_key_to_cpu(node->eb, &key, 0);
2360 return do_relocation(trans, rc, node, &key, path, 0);
2363 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2364 struct reloc_control *rc,
2365 struct btrfs_path *path, int err)
2368 struct btrfs_backref_cache *cache = &rc->backref_cache;
2369 struct btrfs_backref_node *node;
2373 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2374 while (!list_empty(&cache->pending[level])) {
2375 node = list_entry(cache->pending[level].next,
2376 struct btrfs_backref_node, list);
2377 list_move_tail(&node->list, &list);
2378 BUG_ON(!node->pending);
2381 ret = link_to_upper(trans, rc, node, path);
2386 list_splice_init(&list, &cache->pending[level]);
2392 * mark a block and all blocks directly/indirectly reference the block
2395 static void update_processed_blocks(struct reloc_control *rc,
2396 struct btrfs_backref_node *node)
2398 struct btrfs_backref_node *next = node;
2399 struct btrfs_backref_edge *edge;
2400 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2406 if (next->processed)
2409 mark_block_processed(rc, next);
2411 if (list_empty(&next->upper))
2414 edge = list_entry(next->upper.next,
2415 struct btrfs_backref_edge, list[LOWER]);
2416 edges[index++] = edge;
2417 next = edge->node[UPPER];
2419 next = walk_down_backref(edges, &index);
2423 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2425 u32 blocksize = rc->extent_root->fs_info->nodesize;
2427 if (test_range_bit(&rc->processed_blocks, bytenr,
2428 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2433 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2434 struct tree_block *block)
2436 struct extent_buffer *eb;
2438 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2439 block->level, NULL);
2442 } else if (!extent_buffer_uptodate(eb)) {
2443 free_extent_buffer(eb);
2446 if (block->level == 0)
2447 btrfs_item_key_to_cpu(eb, &block->key, 0);
2449 btrfs_node_key_to_cpu(eb, &block->key, 0);
2450 free_extent_buffer(eb);
2451 block->key_ready = 1;
2456 * helper function to relocate a tree block
2458 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2459 struct reloc_control *rc,
2460 struct btrfs_backref_node *node,
2461 struct btrfs_key *key,
2462 struct btrfs_path *path)
2464 struct btrfs_root *root;
2471 * If we fail here we want to drop our backref_node because we are going
2472 * to start over and regenerate the tree for it.
2474 ret = reserve_metadata_space(trans, rc, node);
2478 BUG_ON(node->processed);
2479 root = select_one_root(node);
2480 if (root == ERR_PTR(-ENOENT)) {
2481 update_processed_blocks(rc, node);
2486 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2487 BUG_ON(node->new_bytenr);
2488 BUG_ON(!list_empty(&node->list));
2489 btrfs_record_root_in_trans(trans, root);
2490 root = root->reloc_root;
2491 node->new_bytenr = root->node->start;
2492 btrfs_put_root(node->root);
2493 node->root = btrfs_grab_root(root);
2495 list_add_tail(&node->list, &rc->backref_cache.changed);
2497 path->lowest_level = node->level;
2498 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2499 btrfs_release_path(path);
2504 update_processed_blocks(rc, node);
2506 ret = do_relocation(trans, rc, node, key, path, 1);
2509 if (ret || node->level == 0 || node->cowonly)
2510 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2515 * relocate a list of blocks
2517 static noinline_for_stack
2518 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2519 struct reloc_control *rc, struct rb_root *blocks)
2521 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2522 struct btrfs_backref_node *node;
2523 struct btrfs_path *path;
2524 struct tree_block *block;
2525 struct tree_block *next;
2529 path = btrfs_alloc_path();
2532 goto out_free_blocks;
2535 /* Kick in readahead for tree blocks with missing keys */
2536 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2537 if (!block->key_ready)
2538 readahead_tree_block(fs_info, block->bytenr);
2541 /* Get first keys */
2542 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2543 if (!block->key_ready) {
2544 err = get_tree_block_key(fs_info, block);
2550 /* Do tree relocation */
2551 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2552 node = build_backref_tree(rc, &block->key,
2553 block->level, block->bytenr);
2555 err = PTR_ERR(node);
2559 ret = relocate_tree_block(trans, rc, node, &block->key,
2567 err = finish_pending_nodes(trans, rc, path, err);
2570 btrfs_free_path(path);
2572 free_block_list(blocks);
2576 static noinline_for_stack
2577 int prealloc_file_extent_cluster(struct inode *inode,
2578 struct file_extent_cluster *cluster)
2583 u64 offset = BTRFS_I(inode)->index_cnt;
2587 u64 prealloc_start = cluster->start - offset;
2588 u64 prealloc_end = cluster->end - offset;
2590 struct extent_changeset *data_reserved = NULL;
2592 BUG_ON(cluster->start != cluster->boundary[0]);
2595 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
2596 prealloc_end + 1 - prealloc_start);
2600 cur_offset = prealloc_start;
2601 while (nr < cluster->nr) {
2602 start = cluster->boundary[nr] - offset;
2603 if (nr + 1 < cluster->nr)
2604 end = cluster->boundary[nr + 1] - 1 - offset;
2606 end = cluster->end - offset;
2608 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2609 num_bytes = end + 1 - start;
2610 if (cur_offset < start)
2611 btrfs_free_reserved_data_space(inode, data_reserved,
2612 cur_offset, start - cur_offset);
2613 ret = btrfs_prealloc_file_range(inode, 0, start,
2614 num_bytes, num_bytes,
2615 end + 1, &alloc_hint);
2616 cur_offset = end + 1;
2617 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2622 if (cur_offset < prealloc_end)
2623 btrfs_free_reserved_data_space(inode, data_reserved,
2624 cur_offset, prealloc_end + 1 - cur_offset);
2626 inode_unlock(inode);
2627 extent_changeset_free(data_reserved);
2631 static noinline_for_stack
2632 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2635 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2636 struct extent_map *em;
2639 em = alloc_extent_map();
2644 em->len = end + 1 - start;
2645 em->block_len = em->len;
2646 em->block_start = block_start;
2647 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2649 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2651 write_lock(&em_tree->lock);
2652 ret = add_extent_mapping(em_tree, em, 0);
2653 write_unlock(&em_tree->lock);
2654 if (ret != -EEXIST) {
2655 free_extent_map(em);
2658 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2660 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2665 * Allow error injection to test balance cancellation
2667 int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2669 return atomic_read(&fs_info->balance_cancel_req);
2671 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2673 static int relocate_file_extent_cluster(struct inode *inode,
2674 struct file_extent_cluster *cluster)
2676 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2679 u64 offset = BTRFS_I(inode)->index_cnt;
2680 unsigned long index;
2681 unsigned long last_index;
2683 struct file_ra_state *ra;
2684 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2691 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2695 ret = prealloc_file_extent_cluster(inode, cluster);
2699 file_ra_state_init(ra, inode->i_mapping);
2701 ret = setup_extent_mapping(inode, cluster->start - offset,
2702 cluster->end - offset, cluster->start);
2706 index = (cluster->start - offset) >> PAGE_SHIFT;
2707 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2708 while (index <= last_index) {
2709 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2714 page = find_lock_page(inode->i_mapping, index);
2716 page_cache_sync_readahead(inode->i_mapping,
2718 last_index + 1 - index);
2719 page = find_or_create_page(inode->i_mapping, index,
2722 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2724 btrfs_delalloc_release_extents(BTRFS_I(inode),
2731 if (PageReadahead(page)) {
2732 page_cache_async_readahead(inode->i_mapping,
2733 ra, NULL, page, index,
2734 last_index + 1 - index);
2737 if (!PageUptodate(page)) {
2738 btrfs_readpage(NULL, page);
2740 if (!PageUptodate(page)) {
2743 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2745 btrfs_delalloc_release_extents(BTRFS_I(inode),
2752 page_start = page_offset(page);
2753 page_end = page_start + PAGE_SIZE - 1;
2755 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2757 set_page_extent_mapped(page);
2759 if (nr < cluster->nr &&
2760 page_start + offset == cluster->boundary[nr]) {
2761 set_extent_bits(&BTRFS_I(inode)->io_tree,
2762 page_start, page_end,
2767 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2772 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2774 btrfs_delalloc_release_extents(BTRFS_I(inode),
2777 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2778 page_start, page_end,
2779 EXTENT_LOCKED | EXTENT_BOUNDARY);
2783 set_page_dirty(page);
2785 unlock_extent(&BTRFS_I(inode)->io_tree,
2786 page_start, page_end);
2791 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2792 balance_dirty_pages_ratelimited(inode->i_mapping);
2793 btrfs_throttle(fs_info);
2794 if (btrfs_should_cancel_balance(fs_info)) {
2799 WARN_ON(nr != cluster->nr);
2805 static noinline_for_stack
2806 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2807 struct file_extent_cluster *cluster)
2811 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2812 ret = relocate_file_extent_cluster(inode, cluster);
2819 cluster->start = extent_key->objectid;
2821 BUG_ON(cluster->nr >= MAX_EXTENTS);
2822 cluster->end = extent_key->objectid + extent_key->offset - 1;
2823 cluster->boundary[cluster->nr] = extent_key->objectid;
2826 if (cluster->nr >= MAX_EXTENTS) {
2827 ret = relocate_file_extent_cluster(inode, cluster);
2836 * helper to add a tree block to the list.
2837 * the major work is getting the generation and level of the block
2839 static int add_tree_block(struct reloc_control *rc,
2840 struct btrfs_key *extent_key,
2841 struct btrfs_path *path,
2842 struct rb_root *blocks)
2844 struct extent_buffer *eb;
2845 struct btrfs_extent_item *ei;
2846 struct btrfs_tree_block_info *bi;
2847 struct tree_block *block;
2848 struct rb_node *rb_node;
2853 eb = path->nodes[0];
2854 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2856 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2857 item_size >= sizeof(*ei) + sizeof(*bi)) {
2858 ei = btrfs_item_ptr(eb, path->slots[0],
2859 struct btrfs_extent_item);
2860 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2861 bi = (struct btrfs_tree_block_info *)(ei + 1);
2862 level = btrfs_tree_block_level(eb, bi);
2864 level = (int)extent_key->offset;
2866 generation = btrfs_extent_generation(eb, ei);
2867 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2868 btrfs_print_v0_err(eb->fs_info);
2869 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2875 btrfs_release_path(path);
2877 BUG_ON(level == -1);
2879 block = kmalloc(sizeof(*block), GFP_NOFS);
2883 block->bytenr = extent_key->objectid;
2884 block->key.objectid = rc->extent_root->fs_info->nodesize;
2885 block->key.offset = generation;
2886 block->level = level;
2887 block->key_ready = 0;
2889 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2891 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2898 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2900 static int __add_tree_block(struct reloc_control *rc,
2901 u64 bytenr, u32 blocksize,
2902 struct rb_root *blocks)
2904 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2905 struct btrfs_path *path;
2906 struct btrfs_key key;
2908 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2910 if (tree_block_processed(bytenr, rc))
2913 if (rb_simple_search(blocks, bytenr))
2916 path = btrfs_alloc_path();
2920 key.objectid = bytenr;
2922 key.type = BTRFS_METADATA_ITEM_KEY;
2923 key.offset = (u64)-1;
2925 key.type = BTRFS_EXTENT_ITEM_KEY;
2926 key.offset = blocksize;
2929 path->search_commit_root = 1;
2930 path->skip_locking = 1;
2931 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2935 if (ret > 0 && skinny) {
2936 if (path->slots[0]) {
2938 btrfs_item_key_to_cpu(path->nodes[0], &key,
2940 if (key.objectid == bytenr &&
2941 (key.type == BTRFS_METADATA_ITEM_KEY ||
2942 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2943 key.offset == blocksize)))
2949 btrfs_release_path(path);
2955 btrfs_print_leaf(path->nodes[0]);
2957 "tree block extent item (%llu) is not found in extent tree",
2964 ret = add_tree_block(rc, &key, path, blocks);
2966 btrfs_free_path(path);
2970 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2971 struct btrfs_block_group *block_group,
2972 struct inode *inode,
2975 struct btrfs_key key;
2976 struct btrfs_root *root = fs_info->tree_root;
2977 struct btrfs_trans_handle *trans;
2984 key.type = BTRFS_INODE_ITEM_KEY;
2987 inode = btrfs_iget(fs_info->sb, &key, root);
2992 ret = btrfs_check_trunc_cache_free_space(fs_info,
2993 &fs_info->global_block_rsv);
2997 trans = btrfs_join_transaction(root);
2998 if (IS_ERR(trans)) {
2999 ret = PTR_ERR(trans);
3003 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3005 btrfs_end_transaction(trans);
3006 btrfs_btree_balance_dirty(fs_info);
3013 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3014 * cache inode, to avoid free space cache data extent blocking data relocation.
3016 static int delete_v1_space_cache(struct extent_buffer *leaf,
3017 struct btrfs_block_group *block_group,
3020 u64 space_cache_ino;
3021 struct btrfs_file_extent_item *ei;
3022 struct btrfs_key key;
3027 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3030 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3031 btrfs_item_key_to_cpu(leaf, &key, i);
3032 if (key.type != BTRFS_EXTENT_DATA_KEY)
3034 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3035 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG &&
3036 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3038 space_cache_ino = key.objectid;
3044 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3050 * helper to find all tree blocks that reference a given data extent
3052 static noinline_for_stack
3053 int add_data_references(struct reloc_control *rc,
3054 struct btrfs_key *extent_key,
3055 struct btrfs_path *path,
3056 struct rb_root *blocks)
3058 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3059 struct ulist *leaves = NULL;
3060 struct ulist_iterator leaf_uiter;
3061 struct ulist_node *ref_node = NULL;
3062 const u32 blocksize = fs_info->nodesize;
3065 btrfs_release_path(path);
3066 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3067 0, &leaves, NULL, true);
3071 ULIST_ITER_INIT(&leaf_uiter);
3072 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3073 struct extent_buffer *eb;
3075 eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL);
3080 ret = delete_v1_space_cache(eb, rc->block_group,
3081 extent_key->objectid);
3082 free_extent_buffer(eb);
3085 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3090 free_block_list(blocks);
3096 * helper to find next unprocessed extent
3098 static noinline_for_stack
3099 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3100 struct btrfs_key *extent_key)
3102 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3103 struct btrfs_key key;
3104 struct extent_buffer *leaf;
3105 u64 start, end, last;
3108 last = rc->block_group->start + rc->block_group->length;
3111 if (rc->search_start >= last) {
3116 key.objectid = rc->search_start;
3117 key.type = BTRFS_EXTENT_ITEM_KEY;
3120 path->search_commit_root = 1;
3121 path->skip_locking = 1;
3122 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3127 leaf = path->nodes[0];
3128 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3129 ret = btrfs_next_leaf(rc->extent_root, path);
3132 leaf = path->nodes[0];
3135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3136 if (key.objectid >= last) {
3141 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3142 key.type != BTRFS_METADATA_ITEM_KEY) {
3147 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3148 key.objectid + key.offset <= rc->search_start) {
3153 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3154 key.objectid + fs_info->nodesize <=
3160 ret = find_first_extent_bit(&rc->processed_blocks,
3161 key.objectid, &start, &end,
3162 EXTENT_DIRTY, NULL);
3164 if (ret == 0 && start <= key.objectid) {
3165 btrfs_release_path(path);
3166 rc->search_start = end + 1;
3168 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3169 rc->search_start = key.objectid + key.offset;
3171 rc->search_start = key.objectid +
3173 memcpy(extent_key, &key, sizeof(key));
3177 btrfs_release_path(path);
3181 static void set_reloc_control(struct reloc_control *rc)
3183 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3185 mutex_lock(&fs_info->reloc_mutex);
3186 fs_info->reloc_ctl = rc;
3187 mutex_unlock(&fs_info->reloc_mutex);
3190 static void unset_reloc_control(struct reloc_control *rc)
3192 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3194 mutex_lock(&fs_info->reloc_mutex);
3195 fs_info->reloc_ctl = NULL;
3196 mutex_unlock(&fs_info->reloc_mutex);
3199 static int check_extent_flags(u64 flags)
3201 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3202 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3204 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3205 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3207 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3208 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3213 static noinline_for_stack
3214 int prepare_to_relocate(struct reloc_control *rc)
3216 struct btrfs_trans_handle *trans;
3219 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3220 BTRFS_BLOCK_RSV_TEMP);
3224 memset(&rc->cluster, 0, sizeof(rc->cluster));
3225 rc->search_start = rc->block_group->start;
3226 rc->extents_found = 0;
3227 rc->nodes_relocated = 0;
3228 rc->merging_rsv_size = 0;
3229 rc->reserved_bytes = 0;
3230 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3231 RELOCATION_RESERVED_NODES;
3232 ret = btrfs_block_rsv_refill(rc->extent_root,
3233 rc->block_rsv, rc->block_rsv->size,
3234 BTRFS_RESERVE_FLUSH_ALL);
3238 rc->create_reloc_tree = 1;
3239 set_reloc_control(rc);
3241 trans = btrfs_join_transaction(rc->extent_root);
3242 if (IS_ERR(trans)) {
3243 unset_reloc_control(rc);
3245 * extent tree is not a ref_cow tree and has no reloc_root to
3246 * cleanup. And callers are responsible to free the above
3249 return PTR_ERR(trans);
3251 btrfs_commit_transaction(trans);
3255 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3257 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3258 struct rb_root blocks = RB_ROOT;
3259 struct btrfs_key key;
3260 struct btrfs_trans_handle *trans = NULL;
3261 struct btrfs_path *path;
3262 struct btrfs_extent_item *ei;
3269 path = btrfs_alloc_path();
3272 path->reada = READA_FORWARD;
3274 ret = prepare_to_relocate(rc);
3281 rc->reserved_bytes = 0;
3282 ret = btrfs_block_rsv_refill(rc->extent_root,
3283 rc->block_rsv, rc->block_rsv->size,
3284 BTRFS_RESERVE_FLUSH_ALL);
3290 trans = btrfs_start_transaction(rc->extent_root, 0);
3291 if (IS_ERR(trans)) {
3292 err = PTR_ERR(trans);
3297 if (update_backref_cache(trans, &rc->backref_cache)) {
3298 btrfs_end_transaction(trans);
3303 ret = find_next_extent(rc, path, &key);
3309 rc->extents_found++;
3311 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3312 struct btrfs_extent_item);
3313 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3314 if (item_size >= sizeof(*ei)) {
3315 flags = btrfs_extent_flags(path->nodes[0], ei);
3316 ret = check_extent_flags(flags);
3318 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3320 btrfs_print_v0_err(trans->fs_info);
3321 btrfs_abort_transaction(trans, err);
3327 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3328 ret = add_tree_block(rc, &key, path, &blocks);
3329 } else if (rc->stage == UPDATE_DATA_PTRS &&
3330 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3331 ret = add_data_references(rc, &key, path, &blocks);
3333 btrfs_release_path(path);
3341 if (!RB_EMPTY_ROOT(&blocks)) {
3342 ret = relocate_tree_blocks(trans, rc, &blocks);
3344 if (ret != -EAGAIN) {
3348 rc->extents_found--;
3349 rc->search_start = key.objectid;
3353 btrfs_end_transaction_throttle(trans);
3354 btrfs_btree_balance_dirty(fs_info);
3357 if (rc->stage == MOVE_DATA_EXTENTS &&
3358 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3359 rc->found_file_extent = 1;
3360 ret = relocate_data_extent(rc->data_inode,
3361 &key, &rc->cluster);
3367 if (btrfs_should_cancel_balance(fs_info)) {
3372 if (trans && progress && err == -ENOSPC) {
3373 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3381 btrfs_release_path(path);
3382 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3385 btrfs_end_transaction_throttle(trans);
3386 btrfs_btree_balance_dirty(fs_info);
3390 ret = relocate_file_extent_cluster(rc->data_inode,
3396 rc->create_reloc_tree = 0;
3397 set_reloc_control(rc);
3399 btrfs_backref_release_cache(&rc->backref_cache);
3400 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3403 * Even in the case when the relocation is cancelled, we should all go
3404 * through prepare_to_merge() and merge_reloc_roots().
3406 * For error (including cancelled balance), prepare_to_merge() will
3407 * mark all reloc trees orphan, then queue them for cleanup in
3408 * merge_reloc_roots()
3410 err = prepare_to_merge(rc, err);
3412 merge_reloc_roots(rc);
3414 rc->merge_reloc_tree = 0;
3415 unset_reloc_control(rc);
3416 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3418 /* get rid of pinned extents */
3419 trans = btrfs_join_transaction(rc->extent_root);
3420 if (IS_ERR(trans)) {
3421 err = PTR_ERR(trans);
3424 btrfs_commit_transaction(trans);
3426 ret = clean_dirty_subvols(rc);
3427 if (ret < 0 && !err)
3429 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3430 btrfs_free_path(path);
3434 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3435 struct btrfs_root *root, u64 objectid)
3437 struct btrfs_path *path;
3438 struct btrfs_inode_item *item;
3439 struct extent_buffer *leaf;
3442 path = btrfs_alloc_path();
3446 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3450 leaf = path->nodes[0];
3451 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3452 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3453 btrfs_set_inode_generation(leaf, item, 1);
3454 btrfs_set_inode_size(leaf, item, 0);
3455 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3456 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3457 BTRFS_INODE_PREALLOC);
3458 btrfs_mark_buffer_dirty(leaf);
3460 btrfs_free_path(path);
3465 * helper to create inode for data relocation.
3466 * the inode is in data relocation tree and its link count is 0
3468 static noinline_for_stack
3469 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3470 struct btrfs_block_group *group)
3472 struct inode *inode = NULL;
3473 struct btrfs_trans_handle *trans;
3474 struct btrfs_root *root;
3475 struct btrfs_key key;
3479 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3481 return ERR_CAST(root);
3483 trans = btrfs_start_transaction(root, 6);
3484 if (IS_ERR(trans)) {
3485 btrfs_put_root(root);
3486 return ERR_CAST(trans);
3489 err = btrfs_find_free_objectid(root, &objectid);
3493 err = __insert_orphan_inode(trans, root, objectid);
3496 key.objectid = objectid;
3497 key.type = BTRFS_INODE_ITEM_KEY;
3499 inode = btrfs_iget(fs_info->sb, &key, root);
3500 BUG_ON(IS_ERR(inode));
3501 BTRFS_I(inode)->index_cnt = group->start;
3503 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3505 btrfs_put_root(root);
3506 btrfs_end_transaction(trans);
3507 btrfs_btree_balance_dirty(fs_info);
3511 inode = ERR_PTR(err);
3516 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3518 struct reloc_control *rc;
3520 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3524 INIT_LIST_HEAD(&rc->reloc_roots);
3525 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3526 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3527 mapping_tree_init(&rc->reloc_root_tree);
3528 extent_io_tree_init(fs_info, &rc->processed_blocks,
3529 IO_TREE_RELOC_BLOCKS, NULL);
3533 static void free_reloc_control(struct reloc_control *rc)
3535 struct mapping_node *node, *tmp;
3537 free_reloc_roots(&rc->reloc_roots);
3538 rbtree_postorder_for_each_entry_safe(node, tmp,
3539 &rc->reloc_root_tree.rb_root, rb_node)
3546 * Print the block group being relocated
3548 static void describe_relocation(struct btrfs_fs_info *fs_info,
3549 struct btrfs_block_group *block_group)
3551 char buf[128] = {'\0'};
3553 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3556 "relocating block group %llu flags %s",
3557 block_group->start, buf);
3560 static const char *stage_to_string(int stage)
3562 if (stage == MOVE_DATA_EXTENTS)
3563 return "move data extents";
3564 if (stage == UPDATE_DATA_PTRS)
3565 return "update data pointers";
3570 * function to relocate all extents in a block group.
3572 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3574 struct btrfs_block_group *bg;
3575 struct btrfs_root *extent_root = fs_info->extent_root;
3576 struct reloc_control *rc;
3577 struct inode *inode;
3578 struct btrfs_path *path;
3583 bg = btrfs_lookup_block_group(fs_info, group_start);
3587 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3588 btrfs_put_block_group(bg);
3592 rc = alloc_reloc_control(fs_info);
3594 btrfs_put_block_group(bg);
3598 rc->extent_root = extent_root;
3599 rc->block_group = bg;
3601 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3608 path = btrfs_alloc_path();
3614 inode = lookup_free_space_inode(rc->block_group, path);
3615 btrfs_free_path(path);
3618 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3620 ret = PTR_ERR(inode);
3622 if (ret && ret != -ENOENT) {
3627 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3628 if (IS_ERR(rc->data_inode)) {
3629 err = PTR_ERR(rc->data_inode);
3630 rc->data_inode = NULL;
3634 describe_relocation(fs_info, rc->block_group);
3636 btrfs_wait_block_group_reservations(rc->block_group);
3637 btrfs_wait_nocow_writers(rc->block_group);
3638 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3639 rc->block_group->start,
3640 rc->block_group->length);
3645 mutex_lock(&fs_info->cleaner_mutex);
3646 ret = relocate_block_group(rc);
3647 mutex_unlock(&fs_info->cleaner_mutex);
3651 finishes_stage = rc->stage;
3653 * We may have gotten ENOSPC after we already dirtied some
3654 * extents. If writeout happens while we're relocating a
3655 * different block group we could end up hitting the
3656 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3657 * btrfs_reloc_cow_block. Make sure we write everything out
3658 * properly so we don't trip over this problem, and then break
3659 * out of the loop if we hit an error.
3661 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3662 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3666 invalidate_mapping_pages(rc->data_inode->i_mapping,
3668 rc->stage = UPDATE_DATA_PTRS;
3674 if (rc->extents_found == 0)
3677 btrfs_info(fs_info, "found %llu extents, stage: %s",
3678 rc->extents_found, stage_to_string(finishes_stage));
3681 WARN_ON(rc->block_group->pinned > 0);
3682 WARN_ON(rc->block_group->reserved > 0);
3683 WARN_ON(rc->block_group->used > 0);
3686 btrfs_dec_block_group_ro(rc->block_group);
3687 iput(rc->data_inode);
3688 btrfs_put_block_group(rc->block_group);
3689 free_reloc_control(rc);
3693 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3695 struct btrfs_fs_info *fs_info = root->fs_info;
3696 struct btrfs_trans_handle *trans;
3699 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3701 return PTR_ERR(trans);
3703 memset(&root->root_item.drop_progress, 0,
3704 sizeof(root->root_item.drop_progress));
3705 root->root_item.drop_level = 0;
3706 btrfs_set_root_refs(&root->root_item, 0);
3707 ret = btrfs_update_root(trans, fs_info->tree_root,
3708 &root->root_key, &root->root_item);
3710 err = btrfs_end_transaction(trans);
3717 * recover relocation interrupted by system crash.
3719 * this function resumes merging reloc trees with corresponding fs trees.
3720 * this is important for keeping the sharing of tree blocks
3722 int btrfs_recover_relocation(struct btrfs_root *root)
3724 struct btrfs_fs_info *fs_info = root->fs_info;
3725 LIST_HEAD(reloc_roots);
3726 struct btrfs_key key;
3727 struct btrfs_root *fs_root;
3728 struct btrfs_root *reloc_root;
3729 struct btrfs_path *path;
3730 struct extent_buffer *leaf;
3731 struct reloc_control *rc = NULL;
3732 struct btrfs_trans_handle *trans;
3736 path = btrfs_alloc_path();
3739 path->reada = READA_BACK;
3741 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3742 key.type = BTRFS_ROOT_ITEM_KEY;
3743 key.offset = (u64)-1;
3746 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3753 if (path->slots[0] == 0)
3757 leaf = path->nodes[0];
3758 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3759 btrfs_release_path(path);
3761 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3762 key.type != BTRFS_ROOT_ITEM_KEY)
3765 reloc_root = btrfs_read_tree_root(root, &key);
3766 if (IS_ERR(reloc_root)) {
3767 err = PTR_ERR(reloc_root);
3771 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3772 list_add(&reloc_root->root_list, &reloc_roots);
3774 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3775 fs_root = read_fs_root(fs_info,
3776 reloc_root->root_key.offset);
3777 if (IS_ERR(fs_root)) {
3778 ret = PTR_ERR(fs_root);
3779 if (ret != -ENOENT) {
3783 ret = mark_garbage_root(reloc_root);
3789 btrfs_put_root(fs_root);
3793 if (key.offset == 0)
3798 btrfs_release_path(path);
3800 if (list_empty(&reloc_roots))
3803 rc = alloc_reloc_control(fs_info);
3809 rc->extent_root = fs_info->extent_root;
3811 set_reloc_control(rc);
3813 trans = btrfs_join_transaction(rc->extent_root);
3814 if (IS_ERR(trans)) {
3815 err = PTR_ERR(trans);
3819 rc->merge_reloc_tree = 1;
3821 while (!list_empty(&reloc_roots)) {
3822 reloc_root = list_entry(reloc_roots.next,
3823 struct btrfs_root, root_list);
3824 list_del(&reloc_root->root_list);
3826 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3827 list_add_tail(&reloc_root->root_list,
3832 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
3833 if (IS_ERR(fs_root)) {
3834 err = PTR_ERR(fs_root);
3835 list_add_tail(&reloc_root->root_list, &reloc_roots);
3836 btrfs_end_transaction(trans);
3840 err = __add_reloc_root(reloc_root);
3841 BUG_ON(err < 0); /* -ENOMEM or logic error */
3842 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3843 btrfs_put_root(fs_root);
3846 err = btrfs_commit_transaction(trans);
3850 merge_reloc_roots(rc);
3852 unset_reloc_control(rc);
3854 trans = btrfs_join_transaction(rc->extent_root);
3855 if (IS_ERR(trans)) {
3856 err = PTR_ERR(trans);
3859 err = btrfs_commit_transaction(trans);
3861 ret = clean_dirty_subvols(rc);
3862 if (ret < 0 && !err)
3865 unset_reloc_control(rc);
3866 free_reloc_control(rc);
3868 free_reloc_roots(&reloc_roots);
3870 btrfs_free_path(path);
3873 /* cleanup orphan inode in data relocation tree */
3874 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3875 if (IS_ERR(fs_root)) {
3876 err = PTR_ERR(fs_root);
3878 err = btrfs_orphan_cleanup(fs_root);
3879 btrfs_put_root(fs_root);
3886 * helper to add ordered checksum for data relocation.
3888 * cloning checksum properly handles the nodatasum extents.
3889 * it also saves CPU time to re-calculate the checksum.
3891 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
3893 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3894 struct btrfs_ordered_sum *sums;
3895 struct btrfs_ordered_extent *ordered;
3901 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3902 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3904 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
3905 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3906 disk_bytenr + len - 1, &list, 0);
3910 while (!list_empty(&list)) {
3911 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3912 list_del_init(&sums->list);
3915 * We need to offset the new_bytenr based on where the csum is.
3916 * We need to do this because we will read in entire prealloc
3917 * extents but we may have written to say the middle of the
3918 * prealloc extent, so we need to make sure the csum goes with
3919 * the right disk offset.
3921 * We can do this because the data reloc inode refers strictly
3922 * to the on disk bytes, so we don't have to worry about
3923 * disk_len vs real len like with real inodes since it's all
3926 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3927 sums->bytenr = new_bytenr;
3929 btrfs_add_ordered_sum(ordered, sums);
3932 btrfs_put_ordered_extent(ordered);
3936 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3937 struct btrfs_root *root, struct extent_buffer *buf,
3938 struct extent_buffer *cow)
3940 struct btrfs_fs_info *fs_info = root->fs_info;
3941 struct reloc_control *rc;
3942 struct btrfs_backref_node *node;
3947 rc = fs_info->reloc_ctl;
3951 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3952 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3954 level = btrfs_header_level(buf);
3955 if (btrfs_header_generation(buf) <=
3956 btrfs_root_last_snapshot(&root->root_item))
3959 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3960 rc->create_reloc_tree) {
3961 WARN_ON(!first_cow && level == 0);
3963 node = rc->backref_cache.path[level];
3964 BUG_ON(node->bytenr != buf->start &&
3965 node->new_bytenr != buf->start);
3967 btrfs_backref_drop_node_buffer(node);
3968 atomic_inc(&cow->refs);
3970 node->new_bytenr = cow->start;
3972 if (!node->pending) {
3973 list_move_tail(&node->list,
3974 &rc->backref_cache.pending[level]);
3979 mark_block_processed(rc, node);
3981 if (first_cow && level > 0)
3982 rc->nodes_relocated += buf->len;
3985 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
3986 ret = replace_file_extents(trans, rc, root, cow);
3991 * called before creating snapshot. it calculates metadata reservation
3992 * required for relocating tree blocks in the snapshot
3994 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
3995 u64 *bytes_to_reserve)
3997 struct btrfs_root *root = pending->root;
3998 struct reloc_control *rc = root->fs_info->reloc_ctl;
4000 if (!rc || !have_reloc_root(root))
4003 if (!rc->merge_reloc_tree)
4006 root = root->reloc_root;
4007 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4009 * relocation is in the stage of merging trees. the space
4010 * used by merging a reloc tree is twice the size of
4011 * relocated tree nodes in the worst case. half for cowing
4012 * the reloc tree, half for cowing the fs tree. the space
4013 * used by cowing the reloc tree will be freed after the
4014 * tree is dropped. if we create snapshot, cowing the fs
4015 * tree may use more space than it frees. so we need
4016 * reserve extra space.
4018 *bytes_to_reserve += rc->nodes_relocated;
4022 * called after snapshot is created. migrate block reservation
4023 * and create reloc root for the newly created snapshot
4025 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4026 * references held on the reloc_root, one for root->reloc_root and one for
4029 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4030 struct btrfs_pending_snapshot *pending)
4032 struct btrfs_root *root = pending->root;
4033 struct btrfs_root *reloc_root;
4034 struct btrfs_root *new_root;
4035 struct reloc_control *rc = root->fs_info->reloc_ctl;
4038 if (!rc || !have_reloc_root(root))
4041 rc = root->fs_info->reloc_ctl;
4042 rc->merging_rsv_size += rc->nodes_relocated;
4044 if (rc->merge_reloc_tree) {
4045 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4047 rc->nodes_relocated, true);
4052 new_root = pending->snap;
4053 reloc_root = create_reloc_root(trans, root->reloc_root,
4054 new_root->root_key.objectid);
4055 if (IS_ERR(reloc_root))
4056 return PTR_ERR(reloc_root);
4058 ret = __add_reloc_root(reloc_root);
4060 new_root->reloc_root = btrfs_grab_root(reloc_root);
4062 if (rc->create_reloc_tree)
4063 ret = clone_backref_node(trans, rc, root, reloc_root);