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);
369 * For useless nodes, do two major clean ups:
371 * - Cleanup the children edges and nodes
372 * If child node is also orphan (no parent) during cleanup, then the child
373 * node will also be cleaned up.
375 * - Freeing up leaves (level 0), keeps nodes detached
376 * For nodes, the node is still cached as "detached"
378 * Return false if @node is not in the @useless_nodes list.
379 * Return true if @node is in the @useless_nodes list.
381 static bool handle_useless_nodes(struct reloc_control *rc,
382 struct btrfs_backref_node *node)
384 struct btrfs_backref_cache *cache = &rc->backref_cache;
385 struct list_head *useless_node = &cache->useless_node;
388 while (!list_empty(useless_node)) {
389 struct btrfs_backref_node *cur;
391 cur = list_first_entry(useless_node, struct btrfs_backref_node,
393 list_del_init(&cur->list);
395 /* Only tree root nodes can be added to @useless_nodes */
396 ASSERT(list_empty(&cur->upper));
401 /* The node is the lowest node */
403 list_del_init(&cur->lower);
407 /* Cleanup the lower edges */
408 while (!list_empty(&cur->lower)) {
409 struct btrfs_backref_edge *edge;
410 struct btrfs_backref_node *lower;
412 edge = list_entry(cur->lower.next,
413 struct btrfs_backref_edge, list[UPPER]);
414 list_del(&edge->list[UPPER]);
415 list_del(&edge->list[LOWER]);
416 lower = edge->node[LOWER];
417 btrfs_backref_free_edge(cache, edge);
419 /* Child node is also orphan, queue for cleanup */
420 if (list_empty(&lower->upper))
421 list_add(&lower->list, useless_node);
423 /* Mark this block processed for relocation */
424 mark_block_processed(rc, cur);
427 * Backref nodes for tree leaves are deleted from the cache.
428 * Backref nodes for upper level tree blocks are left in the
429 * cache to avoid unnecessary backref lookup.
431 if (cur->level > 0) {
432 list_add(&cur->list, &cache->detached);
435 rb_erase(&cur->rb_node, &cache->rb_root);
436 btrfs_backref_free_node(cache, cur);
443 * Build backref tree for a given tree block. Root of the backref tree
444 * corresponds the tree block, leaves of the backref tree correspond roots of
445 * b-trees that reference the tree block.
447 * The basic idea of this function is check backrefs of a given block to find
448 * upper level blocks that reference the block, and then check backrefs of
449 * these upper level blocks recursively. The recursion stops when tree root is
450 * reached or backrefs for the block is cached.
452 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
453 * all upper level blocks that directly/indirectly reference the block are also
456 static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
457 struct reloc_control *rc, struct btrfs_key *node_key,
458 int level, u64 bytenr)
460 struct btrfs_backref_iter *iter;
461 struct btrfs_backref_cache *cache = &rc->backref_cache;
462 /* For searching parent of TREE_BLOCK_REF */
463 struct btrfs_path *path;
464 struct btrfs_backref_node *cur;
465 struct btrfs_backref_node *node = NULL;
466 struct btrfs_backref_edge *edge;
470 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS);
472 return ERR_PTR(-ENOMEM);
473 path = btrfs_alloc_path();
479 node = btrfs_backref_alloc_node(cache, bytenr, level);
488 /* Breadth-first search to build backref cache */
490 ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
496 edge = list_first_entry_or_null(&cache->pending_edge,
497 struct btrfs_backref_edge, list[UPPER]);
499 * The pending list isn't empty, take the first block to
503 list_del_init(&edge->list[UPPER]);
504 cur = edge->node[UPPER];
508 /* Finish the upper linkage of newly added edges/nodes */
509 ret = btrfs_backref_finish_upper_links(cache, node);
515 if (handle_useless_nodes(rc, node))
518 btrfs_backref_iter_free(iter);
519 btrfs_free_path(path);
521 btrfs_backref_error_cleanup(cache, node);
524 ASSERT(!node || !node->detached);
525 ASSERT(list_empty(&cache->useless_node) &&
526 list_empty(&cache->pending_edge));
531 * helper to add backref node for the newly created snapshot.
532 * the backref node is created by cloning backref node that
533 * corresponds to root of source tree
535 static int clone_backref_node(struct btrfs_trans_handle *trans,
536 struct reloc_control *rc,
537 struct btrfs_root *src,
538 struct btrfs_root *dest)
540 struct btrfs_root *reloc_root = src->reloc_root;
541 struct btrfs_backref_cache *cache = &rc->backref_cache;
542 struct btrfs_backref_node *node = NULL;
543 struct btrfs_backref_node *new_node;
544 struct btrfs_backref_edge *edge;
545 struct btrfs_backref_edge *new_edge;
546 struct rb_node *rb_node;
548 if (cache->last_trans > 0)
549 update_backref_cache(trans, cache);
551 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
553 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
557 BUG_ON(node->new_bytenr != reloc_root->node->start);
561 rb_node = rb_simple_search(&cache->rb_root,
562 reloc_root->commit_root->start);
564 node = rb_entry(rb_node, struct btrfs_backref_node,
566 BUG_ON(node->detached);
573 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
578 new_node->lowest = node->lowest;
579 new_node->checked = 1;
580 new_node->root = btrfs_grab_root(dest);
581 ASSERT(new_node->root);
584 list_for_each_entry(edge, &node->lower, list[UPPER]) {
585 new_edge = btrfs_backref_alloc_edge(cache);
589 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
590 new_node, LINK_UPPER);
593 list_add_tail(&new_node->lower, &cache->leaves);
596 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
599 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
601 if (!new_node->lowest) {
602 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
603 list_add_tail(&new_edge->list[LOWER],
604 &new_edge->node[LOWER]->upper);
609 while (!list_empty(&new_node->lower)) {
610 new_edge = list_entry(new_node->lower.next,
611 struct btrfs_backref_edge, list[UPPER]);
612 list_del(&new_edge->list[UPPER]);
613 btrfs_backref_free_edge(cache, new_edge);
615 btrfs_backref_free_node(cache, new_node);
620 * helper to add 'address of tree root -> reloc tree' mapping
622 static int __must_check __add_reloc_root(struct btrfs_root *root)
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 struct rb_node *rb_node;
626 struct mapping_node *node;
627 struct reloc_control *rc = fs_info->reloc_ctl;
629 node = kmalloc(sizeof(*node), GFP_NOFS);
633 node->bytenr = root->commit_root->start;
636 spin_lock(&rc->reloc_root_tree.lock);
637 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
638 node->bytenr, &node->rb_node);
639 spin_unlock(&rc->reloc_root_tree.lock);
641 btrfs_panic(fs_info, -EEXIST,
642 "Duplicate root found for start=%llu while inserting into relocation tree",
646 list_add_tail(&root->root_list, &rc->reloc_roots);
651 * helper to delete the 'address of tree root -> reloc tree'
654 static void __del_reloc_root(struct btrfs_root *root)
656 struct btrfs_fs_info *fs_info = root->fs_info;
657 struct rb_node *rb_node;
658 struct mapping_node *node = NULL;
659 struct reloc_control *rc = fs_info->reloc_ctl;
660 bool put_ref = false;
662 if (rc && root->node) {
663 spin_lock(&rc->reloc_root_tree.lock);
664 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
665 root->commit_root->start);
667 node = rb_entry(rb_node, struct mapping_node, rb_node);
668 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
669 RB_CLEAR_NODE(&node->rb_node);
671 spin_unlock(&rc->reloc_root_tree.lock);
672 ASSERT(!node || (struct btrfs_root *)node->data == root);
676 * We only put the reloc root here if it's on the list. There's a lot
677 * of places where the pattern is to splice the rc->reloc_roots, process
678 * the reloc roots, and then add the reloc root back onto
679 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
680 * list we don't want the reference being dropped, because the guy
681 * messing with the list is in charge of the reference.
683 spin_lock(&fs_info->trans_lock);
684 if (!list_empty(&root->root_list)) {
686 list_del_init(&root->root_list);
688 spin_unlock(&fs_info->trans_lock);
690 btrfs_put_root(root);
695 * helper to update the 'address of tree root -> reloc tree'
698 static int __update_reloc_root(struct btrfs_root *root)
700 struct btrfs_fs_info *fs_info = root->fs_info;
701 struct rb_node *rb_node;
702 struct mapping_node *node = NULL;
703 struct reloc_control *rc = fs_info->reloc_ctl;
705 spin_lock(&rc->reloc_root_tree.lock);
706 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
707 root->commit_root->start);
709 node = rb_entry(rb_node, struct mapping_node, rb_node);
710 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
712 spin_unlock(&rc->reloc_root_tree.lock);
716 BUG_ON((struct btrfs_root *)node->data != root);
718 spin_lock(&rc->reloc_root_tree.lock);
719 node->bytenr = root->node->start;
720 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
721 node->bytenr, &node->rb_node);
722 spin_unlock(&rc->reloc_root_tree.lock);
724 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
728 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
729 struct btrfs_root *root, u64 objectid)
731 struct btrfs_fs_info *fs_info = root->fs_info;
732 struct btrfs_root *reloc_root;
733 struct extent_buffer *eb;
734 struct btrfs_root_item *root_item;
735 struct btrfs_key root_key;
737 bool must_abort = false;
739 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
741 return ERR_PTR(-ENOMEM);
743 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
744 root_key.type = BTRFS_ROOT_ITEM_KEY;
745 root_key.offset = objectid;
747 if (root->root_key.objectid == objectid) {
750 /* called by btrfs_init_reloc_root */
751 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
752 BTRFS_TREE_RELOC_OBJECTID);
757 * Set the last_snapshot field to the generation of the commit
758 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
759 * correctly (returns true) when the relocation root is created
760 * either inside the critical section of a transaction commit
761 * (through transaction.c:qgroup_account_snapshot()) and when
762 * it's created before the transaction commit is started.
764 commit_root_gen = btrfs_header_generation(root->commit_root);
765 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
768 * called by btrfs_reloc_post_snapshot_hook.
769 * the source tree is a reloc tree, all tree blocks
770 * modified after it was created have RELOC flag
771 * set in their headers. so it's OK to not update
772 * the 'last_snapshot'.
774 ret = btrfs_copy_root(trans, root, root->node, &eb,
775 BTRFS_TREE_RELOC_OBJECTID);
781 * We have changed references at this point, we must abort the
782 * transaction if anything fails.
786 memcpy(root_item, &root->root_item, sizeof(*root_item));
787 btrfs_set_root_bytenr(root_item, eb->start);
788 btrfs_set_root_level(root_item, btrfs_header_level(eb));
789 btrfs_set_root_generation(root_item, trans->transid);
791 if (root->root_key.objectid == objectid) {
792 btrfs_set_root_refs(root_item, 0);
793 memset(&root_item->drop_progress, 0,
794 sizeof(struct btrfs_disk_key));
795 root_item->drop_level = 0;
798 btrfs_tree_unlock(eb);
799 free_extent_buffer(eb);
801 ret = btrfs_insert_root(trans, fs_info->tree_root,
802 &root_key, root_item);
808 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
809 if (IS_ERR(reloc_root)) {
810 ret = PTR_ERR(reloc_root);
813 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
814 reloc_root->last_trans = trans->transid;
820 btrfs_abort_transaction(trans, ret);
825 * create reloc tree for a given fs tree. reloc tree is just a
826 * snapshot of the fs tree with special root objectid.
828 * The reloc_root comes out of here with two references, one for
829 * root->reloc_root, and another for being on the rc->reloc_roots list.
831 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
832 struct btrfs_root *root)
834 struct btrfs_fs_info *fs_info = root->fs_info;
835 struct btrfs_root *reloc_root;
836 struct reloc_control *rc = fs_info->reloc_ctl;
837 struct btrfs_block_rsv *rsv;
845 * The subvolume has reloc tree but the swap is finished, no need to
846 * create/update the dead reloc tree
848 if (reloc_root_is_dead(root))
852 * This is subtle but important. We do not do
853 * record_root_in_transaction for reloc roots, instead we record their
854 * corresponding fs root, and then here we update the last trans for the
855 * reloc root. This means that we have to do this for the entire life
856 * of the reloc root, regardless of which stage of the relocation we are
859 if (root->reloc_root) {
860 reloc_root = root->reloc_root;
861 reloc_root->last_trans = trans->transid;
866 * We are merging reloc roots, we do not need new reloc trees. Also
867 * reloc trees never need their own reloc tree.
869 if (!rc->create_reloc_tree ||
870 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
873 if (!trans->reloc_reserved) {
874 rsv = trans->block_rsv;
875 trans->block_rsv = rc->block_rsv;
878 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
880 trans->block_rsv = rsv;
882 ret = __add_reloc_root(reloc_root);
884 root->reloc_root = btrfs_grab_root(reloc_root);
889 * update root item of reloc tree
891 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
892 struct btrfs_root *root)
894 struct btrfs_fs_info *fs_info = root->fs_info;
895 struct btrfs_root *reloc_root;
896 struct btrfs_root_item *root_item;
899 if (!have_reloc_root(root))
902 reloc_root = root->reloc_root;
903 root_item = &reloc_root->root_item;
906 * We are probably ok here, but __del_reloc_root() will drop its ref of
907 * the root. We have the ref for root->reloc_root, but just in case
908 * hold it while we update the reloc root.
910 btrfs_grab_root(reloc_root);
912 /* root->reloc_root will stay until current relocation finished */
913 if (fs_info->reloc_ctl->merge_reloc_tree &&
914 btrfs_root_refs(root_item) == 0) {
915 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
917 * Mark the tree as dead before we change reloc_root so
918 * have_reloc_root will not touch it from now on.
921 __del_reloc_root(reloc_root);
924 if (reloc_root->commit_root != reloc_root->node) {
925 __update_reloc_root(reloc_root);
926 btrfs_set_root_node(root_item, reloc_root->node);
927 free_extent_buffer(reloc_root->commit_root);
928 reloc_root->commit_root = btrfs_root_node(reloc_root);
931 ret = btrfs_update_root(trans, fs_info->tree_root,
932 &reloc_root->root_key, root_item);
933 btrfs_put_root(reloc_root);
938 * helper to find first cached inode with inode number >= objectid
941 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
943 struct rb_node *node;
944 struct rb_node *prev;
945 struct btrfs_inode *entry;
948 spin_lock(&root->inode_lock);
950 node = root->inode_tree.rb_node;
954 entry = rb_entry(node, struct btrfs_inode, rb_node);
956 if (objectid < btrfs_ino(entry))
957 node = node->rb_left;
958 else if (objectid > btrfs_ino(entry))
959 node = node->rb_right;
965 entry = rb_entry(prev, struct btrfs_inode, rb_node);
966 if (objectid <= btrfs_ino(entry)) {
970 prev = rb_next(prev);
974 entry = rb_entry(node, struct btrfs_inode, rb_node);
975 inode = igrab(&entry->vfs_inode);
977 spin_unlock(&root->inode_lock);
981 objectid = btrfs_ino(entry) + 1;
982 if (cond_resched_lock(&root->inode_lock))
985 node = rb_next(node);
987 spin_unlock(&root->inode_lock);
992 * get new location of data
994 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
995 u64 bytenr, u64 num_bytes)
997 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
998 struct btrfs_path *path;
999 struct btrfs_file_extent_item *fi;
1000 struct extent_buffer *leaf;
1003 path = btrfs_alloc_path();
1007 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1008 ret = btrfs_lookup_file_extent(NULL, root, path,
1009 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1017 leaf = path->nodes[0];
1018 fi = btrfs_item_ptr(leaf, path->slots[0],
1019 struct btrfs_file_extent_item);
1021 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1022 btrfs_file_extent_compression(leaf, fi) ||
1023 btrfs_file_extent_encryption(leaf, fi) ||
1024 btrfs_file_extent_other_encoding(leaf, fi));
1026 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1031 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1034 btrfs_free_path(path);
1039 * update file extent items in the tree leaf to point to
1040 * the new locations.
1042 static noinline_for_stack
1043 int replace_file_extents(struct btrfs_trans_handle *trans,
1044 struct reloc_control *rc,
1045 struct btrfs_root *root,
1046 struct extent_buffer *leaf)
1048 struct btrfs_fs_info *fs_info = root->fs_info;
1049 struct btrfs_key key;
1050 struct btrfs_file_extent_item *fi;
1051 struct inode *inode = NULL;
1063 if (rc->stage != UPDATE_DATA_PTRS)
1066 /* reloc trees always use full backref */
1067 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1068 parent = leaf->start;
1072 nritems = btrfs_header_nritems(leaf);
1073 for (i = 0; i < nritems; i++) {
1074 struct btrfs_ref ref = { 0 };
1077 btrfs_item_key_to_cpu(leaf, &key, i);
1078 if (key.type != BTRFS_EXTENT_DATA_KEY)
1080 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1081 if (btrfs_file_extent_type(leaf, fi) ==
1082 BTRFS_FILE_EXTENT_INLINE)
1084 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1085 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1088 if (!in_range(bytenr, rc->block_group->start,
1089 rc->block_group->length))
1093 * if we are modifying block in fs tree, wait for readpage
1094 * to complete and drop the extent cache
1096 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1098 inode = find_next_inode(root, key.objectid);
1100 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1101 btrfs_add_delayed_iput(inode);
1102 inode = find_next_inode(root, key.objectid);
1104 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1106 btrfs_file_extent_num_bytes(leaf, fi);
1107 WARN_ON(!IS_ALIGNED(key.offset,
1108 fs_info->sectorsize));
1109 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1111 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1116 btrfs_drop_extent_cache(BTRFS_I(inode),
1117 key.offset, end, 1);
1118 unlock_extent(&BTRFS_I(inode)->io_tree,
1123 ret = get_new_location(rc->data_inode, &new_bytenr,
1127 * Don't have to abort since we've not changed anything
1128 * in the file extent yet.
1133 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1136 key.offset -= btrfs_file_extent_offset(leaf, fi);
1137 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1139 ref.real_root = root->root_key.objectid;
1140 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1141 key.objectid, key.offset);
1142 ret = btrfs_inc_extent_ref(trans, &ref);
1144 btrfs_abort_transaction(trans, ret);
1148 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1150 ref.real_root = root->root_key.objectid;
1151 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1152 key.objectid, key.offset);
1153 ret = btrfs_free_extent(trans, &ref);
1155 btrfs_abort_transaction(trans, ret);
1160 btrfs_mark_buffer_dirty(leaf);
1162 btrfs_add_delayed_iput(inode);
1166 static noinline_for_stack
1167 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1168 struct btrfs_path *path, int level)
1170 struct btrfs_disk_key key1;
1171 struct btrfs_disk_key key2;
1172 btrfs_node_key(eb, &key1, slot);
1173 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1174 return memcmp(&key1, &key2, sizeof(key1));
1178 * try to replace tree blocks in fs tree with the new blocks
1179 * in reloc tree. tree blocks haven't been modified since the
1180 * reloc tree was create can be replaced.
1182 * if a block was replaced, level of the block + 1 is returned.
1183 * if no block got replaced, 0 is returned. if there are other
1184 * errors, a negative error number is returned.
1186 static noinline_for_stack
1187 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1188 struct btrfs_root *dest, struct btrfs_root *src,
1189 struct btrfs_path *path, struct btrfs_key *next_key,
1190 int lowest_level, int max_level)
1192 struct btrfs_fs_info *fs_info = dest->fs_info;
1193 struct extent_buffer *eb;
1194 struct extent_buffer *parent;
1195 struct btrfs_ref ref = { 0 };
1196 struct btrfs_key key;
1208 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1209 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1211 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1213 slot = path->slots[lowest_level];
1214 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1216 eb = btrfs_lock_root_node(dest);
1217 btrfs_set_lock_blocking_write(eb);
1218 level = btrfs_header_level(eb);
1220 if (level < lowest_level) {
1221 btrfs_tree_unlock(eb);
1222 free_extent_buffer(eb);
1227 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1231 btrfs_set_lock_blocking_write(eb);
1234 next_key->objectid = (u64)-1;
1235 next_key->type = (u8)-1;
1236 next_key->offset = (u64)-1;
1241 struct btrfs_key first_key;
1243 level = btrfs_header_level(parent);
1244 ASSERT(level >= lowest_level);
1246 ret = btrfs_bin_search(parent, &key, &slot);
1249 if (ret && slot > 0)
1252 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1253 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1255 old_bytenr = btrfs_node_blockptr(parent, slot);
1256 blocksize = fs_info->nodesize;
1257 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1258 btrfs_node_key_to_cpu(parent, &first_key, slot);
1260 if (level <= max_level) {
1261 eb = path->nodes[level];
1262 new_bytenr = btrfs_node_blockptr(eb,
1263 path->slots[level]);
1264 new_ptr_gen = btrfs_node_ptr_generation(eb,
1265 path->slots[level]);
1271 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1276 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1277 memcmp_node_keys(parent, slot, path, level)) {
1278 if (level <= lowest_level) {
1283 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1284 level - 1, &first_key);
1288 } else if (!extent_buffer_uptodate(eb)) {
1290 free_extent_buffer(eb);
1293 btrfs_tree_lock(eb);
1295 ret = btrfs_cow_block(trans, dest, eb, parent,
1300 btrfs_set_lock_blocking_write(eb);
1302 btrfs_tree_unlock(parent);
1303 free_extent_buffer(parent);
1310 btrfs_tree_unlock(parent);
1311 free_extent_buffer(parent);
1316 btrfs_node_key_to_cpu(path->nodes[level], &key,
1317 path->slots[level]);
1318 btrfs_release_path(path);
1320 path->lowest_level = level;
1321 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1322 path->lowest_level = 0;
1326 * Info qgroup to trace both subtrees.
1328 * We must trace both trees.
1329 * 1) Tree reloc subtree
1330 * If not traced, we will leak data numbers
1332 * If not traced, we will double count old data
1334 * We don't scan the subtree right now, but only record
1335 * the swapped tree blocks.
1336 * The real subtree rescan is delayed until we have new
1337 * CoW on the subtree root node before transaction commit.
1339 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1340 rc->block_group, parent, slot,
1341 path->nodes[level], path->slots[level],
1346 * swap blocks in fs tree and reloc tree.
1348 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1349 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1350 btrfs_mark_buffer_dirty(parent);
1352 btrfs_set_node_blockptr(path->nodes[level],
1353 path->slots[level], old_bytenr);
1354 btrfs_set_node_ptr_generation(path->nodes[level],
1355 path->slots[level], old_ptr_gen);
1356 btrfs_mark_buffer_dirty(path->nodes[level]);
1358 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1359 blocksize, path->nodes[level]->start);
1360 ref.skip_qgroup = true;
1361 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1362 ret = btrfs_inc_extent_ref(trans, &ref);
1364 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1366 ref.skip_qgroup = true;
1367 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1368 ret = btrfs_inc_extent_ref(trans, &ref);
1371 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1372 blocksize, path->nodes[level]->start);
1373 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1374 ref.skip_qgroup = true;
1375 ret = btrfs_free_extent(trans, &ref);
1378 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1380 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1381 ref.skip_qgroup = true;
1382 ret = btrfs_free_extent(trans, &ref);
1385 btrfs_unlock_up_safe(path, 0);
1390 btrfs_tree_unlock(parent);
1391 free_extent_buffer(parent);
1396 * helper to find next relocated block in reloc tree
1398 static noinline_for_stack
1399 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1402 struct extent_buffer *eb;
1407 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1409 for (i = 0; i < *level; i++) {
1410 free_extent_buffer(path->nodes[i]);
1411 path->nodes[i] = NULL;
1414 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1415 eb = path->nodes[i];
1416 nritems = btrfs_header_nritems(eb);
1417 while (path->slots[i] + 1 < nritems) {
1419 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1426 free_extent_buffer(path->nodes[i]);
1427 path->nodes[i] = NULL;
1433 * walk down reloc tree to find relocated block of lowest level
1435 static noinline_for_stack
1436 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1439 struct btrfs_fs_info *fs_info = root->fs_info;
1440 struct extent_buffer *eb = NULL;
1447 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1449 for (i = *level; i > 0; i--) {
1450 struct btrfs_key first_key;
1452 eb = path->nodes[i];
1453 nritems = btrfs_header_nritems(eb);
1454 while (path->slots[i] < nritems) {
1455 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1456 if (ptr_gen > last_snapshot)
1460 if (path->slots[i] >= nritems) {
1471 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1472 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
1473 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
1477 } else if (!extent_buffer_uptodate(eb)) {
1478 free_extent_buffer(eb);
1481 BUG_ON(btrfs_header_level(eb) != i - 1);
1482 path->nodes[i - 1] = eb;
1483 path->slots[i - 1] = 0;
1489 * invalidate extent cache for file extents whose key in range of
1490 * [min_key, max_key)
1492 static int invalidate_extent_cache(struct btrfs_root *root,
1493 struct btrfs_key *min_key,
1494 struct btrfs_key *max_key)
1496 struct btrfs_fs_info *fs_info = root->fs_info;
1497 struct inode *inode = NULL;
1502 objectid = min_key->objectid;
1507 if (objectid > max_key->objectid)
1510 inode = find_next_inode(root, objectid);
1513 ino = btrfs_ino(BTRFS_I(inode));
1515 if (ino > max_key->objectid) {
1521 if (!S_ISREG(inode->i_mode))
1524 if (unlikely(min_key->objectid == ino)) {
1525 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1527 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1530 start = min_key->offset;
1531 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1537 if (unlikely(max_key->objectid == ino)) {
1538 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1540 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1543 if (max_key->offset == 0)
1545 end = max_key->offset;
1546 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1553 /* the lock_extent waits for readpage to complete */
1554 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1555 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1556 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1561 static int find_next_key(struct btrfs_path *path, int level,
1562 struct btrfs_key *key)
1565 while (level < BTRFS_MAX_LEVEL) {
1566 if (!path->nodes[level])
1568 if (path->slots[level] + 1 <
1569 btrfs_header_nritems(path->nodes[level])) {
1570 btrfs_node_key_to_cpu(path->nodes[level], key,
1571 path->slots[level] + 1);
1580 * Insert current subvolume into reloc_control::dirty_subvol_roots
1582 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1583 struct reloc_control *rc,
1584 struct btrfs_root *root)
1586 struct btrfs_root *reloc_root = root->reloc_root;
1587 struct btrfs_root_item *reloc_root_item;
1589 /* @root must be a subvolume tree root with a valid reloc tree */
1590 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1593 reloc_root_item = &reloc_root->root_item;
1594 memset(&reloc_root_item->drop_progress, 0,
1595 sizeof(reloc_root_item->drop_progress));
1596 reloc_root_item->drop_level = 0;
1597 btrfs_set_root_refs(reloc_root_item, 0);
1598 btrfs_update_reloc_root(trans, root);
1600 if (list_empty(&root->reloc_dirty_list)) {
1601 btrfs_grab_root(root);
1602 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1606 static int clean_dirty_subvols(struct reloc_control *rc)
1608 struct btrfs_root *root;
1609 struct btrfs_root *next;
1613 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1615 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1616 /* Merged subvolume, cleanup its reloc root */
1617 struct btrfs_root *reloc_root = root->reloc_root;
1619 list_del_init(&root->reloc_dirty_list);
1620 root->reloc_root = NULL;
1622 * Need barrier to ensure clear_bit() only happens after
1623 * root->reloc_root = NULL. Pairs with have_reloc_root.
1626 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1629 * btrfs_drop_snapshot drops our ref we hold for
1630 * ->reloc_root. If it fails however we must
1631 * drop the ref ourselves.
1633 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1635 btrfs_put_root(reloc_root);
1640 btrfs_put_root(root);
1642 /* Orphan reloc tree, just clean it up */
1643 ret2 = btrfs_drop_snapshot(root, 0, 1);
1645 btrfs_put_root(root);
1655 * merge the relocated tree blocks in reloc tree with corresponding
1658 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1659 struct btrfs_root *root)
1661 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1662 struct btrfs_key key;
1663 struct btrfs_key next_key;
1664 struct btrfs_trans_handle *trans = NULL;
1665 struct btrfs_root *reloc_root;
1666 struct btrfs_root_item *root_item;
1667 struct btrfs_path *path;
1668 struct extent_buffer *leaf;
1677 path = btrfs_alloc_path();
1680 path->reada = READA_FORWARD;
1682 reloc_root = root->reloc_root;
1683 root_item = &reloc_root->root_item;
1685 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1686 level = btrfs_root_level(root_item);
1687 atomic_inc(&reloc_root->node->refs);
1688 path->nodes[level] = reloc_root->node;
1689 path->slots[level] = 0;
1691 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1693 level = root_item->drop_level;
1695 path->lowest_level = level;
1696 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1697 path->lowest_level = 0;
1699 btrfs_free_path(path);
1703 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1704 path->slots[level]);
1705 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1707 btrfs_unlock_up_safe(path, 0);
1711 * In merge_reloc_root(), we modify the upper level pointer to swap the
1712 * tree blocks between reloc tree and subvolume tree. Thus for tree
1713 * block COW, we COW at most from level 1 to root level for each tree.
1715 * Thus the needed metadata size is at most root_level * nodesize,
1716 * and * 2 since we have two trees to COW.
1718 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1719 min_reserved = fs_info->nodesize * reserve_level * 2;
1720 memset(&next_key, 0, sizeof(next_key));
1723 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1724 BTRFS_RESERVE_FLUSH_LIMIT);
1729 trans = btrfs_start_transaction(root, 0);
1730 if (IS_ERR(trans)) {
1731 err = PTR_ERR(trans);
1737 * At this point we no longer have a reloc_control, so we can't
1738 * depend on btrfs_init_reloc_root to update our last_trans.
1740 * But that's ok, we started the trans handle on our
1741 * corresponding fs_root, which means it's been added to the
1742 * dirty list. At commit time we'll still call
1743 * btrfs_update_reloc_root() and update our root item
1746 reloc_root->last_trans = trans->transid;
1747 trans->block_rsv = rc->block_rsv;
1752 ret = walk_down_reloc_tree(reloc_root, path, &level);
1760 if (!find_next_key(path, level, &key) &&
1761 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1764 ret = replace_path(trans, rc, root, reloc_root, path,
1765 &next_key, level, max_level);
1774 btrfs_node_key_to_cpu(path->nodes[level], &key,
1775 path->slots[level]);
1779 ret = walk_up_reloc_tree(reloc_root, path, &level);
1785 * save the merging progress in the drop_progress.
1786 * this is OK since root refs == 1 in this case.
1788 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1789 path->slots[level]);
1790 root_item->drop_level = level;
1792 btrfs_end_transaction_throttle(trans);
1795 btrfs_btree_balance_dirty(fs_info);
1797 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1798 invalidate_extent_cache(root, &key, &next_key);
1802 * handle the case only one block in the fs tree need to be
1803 * relocated and the block is tree root.
1805 leaf = btrfs_lock_root_node(root);
1806 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1808 btrfs_tree_unlock(leaf);
1809 free_extent_buffer(leaf);
1813 btrfs_free_path(path);
1816 insert_dirty_subvol(trans, rc, root);
1819 btrfs_end_transaction_throttle(trans);
1821 btrfs_btree_balance_dirty(fs_info);
1823 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1824 invalidate_extent_cache(root, &key, &next_key);
1829 static noinline_for_stack
1830 int prepare_to_merge(struct reloc_control *rc, int err)
1832 struct btrfs_root *root = rc->extent_root;
1833 struct btrfs_fs_info *fs_info = root->fs_info;
1834 struct btrfs_root *reloc_root;
1835 struct btrfs_trans_handle *trans;
1836 LIST_HEAD(reloc_roots);
1840 mutex_lock(&fs_info->reloc_mutex);
1841 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1842 rc->merging_rsv_size += rc->nodes_relocated * 2;
1843 mutex_unlock(&fs_info->reloc_mutex);
1847 num_bytes = rc->merging_rsv_size;
1848 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1849 BTRFS_RESERVE_FLUSH_ALL);
1854 trans = btrfs_join_transaction(rc->extent_root);
1855 if (IS_ERR(trans)) {
1857 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1859 return PTR_ERR(trans);
1863 if (num_bytes != rc->merging_rsv_size) {
1864 btrfs_end_transaction(trans);
1865 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1871 rc->merge_reloc_tree = 1;
1873 while (!list_empty(&rc->reloc_roots)) {
1874 reloc_root = list_entry(rc->reloc_roots.next,
1875 struct btrfs_root, root_list);
1876 list_del_init(&reloc_root->root_list);
1878 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1880 BUG_ON(IS_ERR(root));
1881 BUG_ON(root->reloc_root != reloc_root);
1884 * set reference count to 1, so btrfs_recover_relocation
1885 * knows it should resumes merging
1888 btrfs_set_root_refs(&reloc_root->root_item, 1);
1889 btrfs_update_reloc_root(trans, root);
1891 list_add(&reloc_root->root_list, &reloc_roots);
1892 btrfs_put_root(root);
1895 list_splice(&reloc_roots, &rc->reloc_roots);
1898 err = btrfs_commit_transaction(trans);
1900 btrfs_end_transaction(trans);
1904 static noinline_for_stack
1905 void free_reloc_roots(struct list_head *list)
1907 struct btrfs_root *reloc_root, *tmp;
1909 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1910 __del_reloc_root(reloc_root);
1913 static noinline_for_stack
1914 void merge_reloc_roots(struct reloc_control *rc)
1916 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1917 struct btrfs_root *root;
1918 struct btrfs_root *reloc_root;
1919 LIST_HEAD(reloc_roots);
1923 root = rc->extent_root;
1926 * this serializes us with btrfs_record_root_in_transaction,
1927 * we have to make sure nobody is in the middle of
1928 * adding their roots to the list while we are
1931 mutex_lock(&fs_info->reloc_mutex);
1932 list_splice_init(&rc->reloc_roots, &reloc_roots);
1933 mutex_unlock(&fs_info->reloc_mutex);
1935 while (!list_empty(&reloc_roots)) {
1937 reloc_root = list_entry(reloc_roots.next,
1938 struct btrfs_root, root_list);
1940 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1942 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1943 BUG_ON(IS_ERR(root));
1944 BUG_ON(root->reloc_root != reloc_root);
1945 ret = merge_reloc_root(rc, root);
1946 btrfs_put_root(root);
1948 if (list_empty(&reloc_root->root_list))
1949 list_add_tail(&reloc_root->root_list,
1954 if (!IS_ERR(root)) {
1955 if (root->reloc_root == reloc_root) {
1956 root->reloc_root = NULL;
1957 btrfs_put_root(reloc_root);
1959 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1961 btrfs_put_root(root);
1964 list_del_init(&reloc_root->root_list);
1965 /* Don't forget to queue this reloc root for cleanup */
1966 list_add_tail(&reloc_root->reloc_dirty_list,
1967 &rc->dirty_subvol_roots);
1977 btrfs_handle_fs_error(fs_info, ret, NULL);
1978 free_reloc_roots(&reloc_roots);
1980 /* new reloc root may be added */
1981 mutex_lock(&fs_info->reloc_mutex);
1982 list_splice_init(&rc->reloc_roots, &reloc_roots);
1983 mutex_unlock(&fs_info->reloc_mutex);
1984 free_reloc_roots(&reloc_roots);
1990 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1992 * here, but it's wrong. If we fail to start the transaction in
1993 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1994 * have actually been removed from the reloc_root_tree rb tree. This is
1995 * fine because we're bailing here, and we hold a reference on the root
1996 * for the list that holds it, so these roots will be cleaned up when we
1997 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1998 * will be cleaned up on unmount.
2000 * The remaining nodes will be cleaned up by free_reloc_control.
2004 static void free_block_list(struct rb_root *blocks)
2006 struct tree_block *block;
2007 struct rb_node *rb_node;
2008 while ((rb_node = rb_first(blocks))) {
2009 block = rb_entry(rb_node, struct tree_block, rb_node);
2010 rb_erase(rb_node, blocks);
2015 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2016 struct btrfs_root *reloc_root)
2018 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2019 struct btrfs_root *root;
2022 if (reloc_root->last_trans == trans->transid)
2025 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2026 BUG_ON(IS_ERR(root));
2027 BUG_ON(root->reloc_root != reloc_root);
2028 ret = btrfs_record_root_in_trans(trans, root);
2029 btrfs_put_root(root);
2034 static noinline_for_stack
2035 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2036 struct reloc_control *rc,
2037 struct btrfs_backref_node *node,
2038 struct btrfs_backref_edge *edges[])
2040 struct btrfs_backref_node *next;
2041 struct btrfs_root *root;
2047 next = walk_up_backref(next, edges, &index);
2050 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2052 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2053 record_reloc_root_in_trans(trans, root);
2057 btrfs_record_root_in_trans(trans, root);
2058 root = root->reloc_root;
2060 if (next->new_bytenr != root->node->start) {
2061 BUG_ON(next->new_bytenr);
2062 BUG_ON(!list_empty(&next->list));
2063 next->new_bytenr = root->node->start;
2064 btrfs_put_root(next->root);
2065 next->root = btrfs_grab_root(root);
2067 list_add_tail(&next->list,
2068 &rc->backref_cache.changed);
2069 mark_block_processed(rc, next);
2075 next = walk_down_backref(edges, &index);
2076 if (!next || next->level <= node->level)
2083 /* setup backref node path for btrfs_reloc_cow_block */
2085 rc->backref_cache.path[next->level] = next;
2088 next = edges[index]->node[UPPER];
2094 * Select a tree root for relocation.
2096 * Return NULL if the block is not shareable. We should use do_relocation() in
2099 * Return a tree root pointer if the block is shareable.
2100 * Return -ENOENT if the block is root of reloc tree.
2102 static noinline_for_stack
2103 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2105 struct btrfs_backref_node *next;
2106 struct btrfs_root *root;
2107 struct btrfs_root *fs_root = NULL;
2108 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2114 next = walk_up_backref(next, edges, &index);
2118 /* No other choice for non-shareable tree */
2119 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2122 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2128 next = walk_down_backref(edges, &index);
2129 if (!next || next->level <= node->level)
2134 return ERR_PTR(-ENOENT);
2138 static noinline_for_stack
2139 u64 calcu_metadata_size(struct reloc_control *rc,
2140 struct btrfs_backref_node *node, int reserve)
2142 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2143 struct btrfs_backref_node *next = node;
2144 struct btrfs_backref_edge *edge;
2145 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2149 BUG_ON(reserve && node->processed);
2154 if (next->processed && (reserve || next != node))
2157 num_bytes += fs_info->nodesize;
2159 if (list_empty(&next->upper))
2162 edge = list_entry(next->upper.next,
2163 struct btrfs_backref_edge, list[LOWER]);
2164 edges[index++] = edge;
2165 next = edge->node[UPPER];
2167 next = walk_down_backref(edges, &index);
2172 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2173 struct reloc_control *rc,
2174 struct btrfs_backref_node *node)
2176 struct btrfs_root *root = rc->extent_root;
2177 struct btrfs_fs_info *fs_info = root->fs_info;
2182 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2184 trans->block_rsv = rc->block_rsv;
2185 rc->reserved_bytes += num_bytes;
2188 * We are under a transaction here so we can only do limited flushing.
2189 * If we get an enospc just kick back -EAGAIN so we know to drop the
2190 * transaction and try to refill when we can flush all the things.
2192 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2193 BTRFS_RESERVE_FLUSH_LIMIT);
2195 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2196 while (tmp <= rc->reserved_bytes)
2199 * only one thread can access block_rsv at this point,
2200 * so we don't need hold lock to protect block_rsv.
2201 * we expand more reservation size here to allow enough
2202 * space for relocation and we will return earlier in
2205 rc->block_rsv->size = tmp + fs_info->nodesize *
2206 RELOCATION_RESERVED_NODES;
2214 * relocate a block tree, and then update pointers in upper level
2215 * blocks that reference the block to point to the new location.
2217 * if called by link_to_upper, the block has already been relocated.
2218 * in that case this function just updates pointers.
2220 static int do_relocation(struct btrfs_trans_handle *trans,
2221 struct reloc_control *rc,
2222 struct btrfs_backref_node *node,
2223 struct btrfs_key *key,
2224 struct btrfs_path *path, int lowest)
2226 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2227 struct btrfs_backref_node *upper;
2228 struct btrfs_backref_edge *edge;
2229 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2230 struct btrfs_root *root;
2231 struct extent_buffer *eb;
2239 BUG_ON(lowest && node->eb);
2241 path->lowest_level = node->level + 1;
2242 rc->backref_cache.path[node->level] = node;
2243 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2244 struct btrfs_key first_key;
2245 struct btrfs_ref ref = { 0 };
2249 upper = edge->node[UPPER];
2250 root = select_reloc_root(trans, rc, upper, edges);
2253 if (upper->eb && !upper->locked) {
2255 ret = btrfs_bin_search(upper->eb, key, &slot);
2261 bytenr = btrfs_node_blockptr(upper->eb, slot);
2262 if (node->eb->start == bytenr)
2265 btrfs_backref_drop_node_buffer(upper);
2269 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2276 btrfs_release_path(path);
2281 upper->eb = path->nodes[upper->level];
2282 path->nodes[upper->level] = NULL;
2284 BUG_ON(upper->eb != path->nodes[upper->level]);
2288 path->locks[upper->level] = 0;
2290 slot = path->slots[upper->level];
2291 btrfs_release_path(path);
2293 ret = btrfs_bin_search(upper->eb, key, &slot);
2301 bytenr = btrfs_node_blockptr(upper->eb, slot);
2303 if (bytenr != node->bytenr) {
2304 btrfs_err(root->fs_info,
2305 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2306 bytenr, node->bytenr, slot,
2312 if (node->eb->start == bytenr)
2316 blocksize = root->fs_info->nodesize;
2317 generation = btrfs_node_ptr_generation(upper->eb, slot);
2318 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2319 eb = read_tree_block(fs_info, bytenr, generation,
2320 upper->level - 1, &first_key);
2324 } else if (!extent_buffer_uptodate(eb)) {
2325 free_extent_buffer(eb);
2329 btrfs_tree_lock(eb);
2330 btrfs_set_lock_blocking_write(eb);
2333 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2334 slot, &eb, BTRFS_NESTING_COW);
2335 btrfs_tree_unlock(eb);
2336 free_extent_buffer(eb);
2341 BUG_ON(node->eb != eb);
2343 btrfs_set_node_blockptr(upper->eb, slot,
2345 btrfs_set_node_ptr_generation(upper->eb, slot,
2347 btrfs_mark_buffer_dirty(upper->eb);
2349 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2350 node->eb->start, blocksize,
2352 ref.real_root = root->root_key.objectid;
2353 btrfs_init_tree_ref(&ref, node->level,
2354 btrfs_header_owner(upper->eb));
2355 ret = btrfs_inc_extent_ref(trans, &ref);
2358 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2362 if (!upper->pending)
2363 btrfs_backref_drop_node_buffer(upper);
2365 btrfs_backref_unlock_node_buffer(upper);
2370 if (!err && node->pending) {
2371 btrfs_backref_drop_node_buffer(node);
2372 list_move_tail(&node->list, &rc->backref_cache.changed);
2376 path->lowest_level = 0;
2377 BUG_ON(err == -ENOSPC);
2381 static int link_to_upper(struct btrfs_trans_handle *trans,
2382 struct reloc_control *rc,
2383 struct btrfs_backref_node *node,
2384 struct btrfs_path *path)
2386 struct btrfs_key key;
2388 btrfs_node_key_to_cpu(node->eb, &key, 0);
2389 return do_relocation(trans, rc, node, &key, path, 0);
2392 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2393 struct reloc_control *rc,
2394 struct btrfs_path *path, int err)
2397 struct btrfs_backref_cache *cache = &rc->backref_cache;
2398 struct btrfs_backref_node *node;
2402 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2403 while (!list_empty(&cache->pending[level])) {
2404 node = list_entry(cache->pending[level].next,
2405 struct btrfs_backref_node, list);
2406 list_move_tail(&node->list, &list);
2407 BUG_ON(!node->pending);
2410 ret = link_to_upper(trans, rc, node, path);
2415 list_splice_init(&list, &cache->pending[level]);
2421 * mark a block and all blocks directly/indirectly reference the block
2424 static void update_processed_blocks(struct reloc_control *rc,
2425 struct btrfs_backref_node *node)
2427 struct btrfs_backref_node *next = node;
2428 struct btrfs_backref_edge *edge;
2429 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2435 if (next->processed)
2438 mark_block_processed(rc, next);
2440 if (list_empty(&next->upper))
2443 edge = list_entry(next->upper.next,
2444 struct btrfs_backref_edge, list[LOWER]);
2445 edges[index++] = edge;
2446 next = edge->node[UPPER];
2448 next = walk_down_backref(edges, &index);
2452 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2454 u32 blocksize = rc->extent_root->fs_info->nodesize;
2456 if (test_range_bit(&rc->processed_blocks, bytenr,
2457 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2462 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2463 struct tree_block *block)
2465 struct extent_buffer *eb;
2467 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2468 block->level, NULL);
2471 } else if (!extent_buffer_uptodate(eb)) {
2472 free_extent_buffer(eb);
2475 if (block->level == 0)
2476 btrfs_item_key_to_cpu(eb, &block->key, 0);
2478 btrfs_node_key_to_cpu(eb, &block->key, 0);
2479 free_extent_buffer(eb);
2480 block->key_ready = 1;
2485 * helper function to relocate a tree block
2487 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2488 struct reloc_control *rc,
2489 struct btrfs_backref_node *node,
2490 struct btrfs_key *key,
2491 struct btrfs_path *path)
2493 struct btrfs_root *root;
2500 * If we fail here we want to drop our backref_node because we are going
2501 * to start over and regenerate the tree for it.
2503 ret = reserve_metadata_space(trans, rc, node);
2507 BUG_ON(node->processed);
2508 root = select_one_root(node);
2509 if (root == ERR_PTR(-ENOENT)) {
2510 update_processed_blocks(rc, node);
2515 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2516 BUG_ON(node->new_bytenr);
2517 BUG_ON(!list_empty(&node->list));
2518 btrfs_record_root_in_trans(trans, root);
2519 root = root->reloc_root;
2520 node->new_bytenr = root->node->start;
2521 btrfs_put_root(node->root);
2522 node->root = btrfs_grab_root(root);
2524 list_add_tail(&node->list, &rc->backref_cache.changed);
2526 path->lowest_level = node->level;
2527 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2528 btrfs_release_path(path);
2533 update_processed_blocks(rc, node);
2535 ret = do_relocation(trans, rc, node, key, path, 1);
2538 if (ret || node->level == 0 || node->cowonly)
2539 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2544 * relocate a list of blocks
2546 static noinline_for_stack
2547 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2548 struct reloc_control *rc, struct rb_root *blocks)
2550 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2551 struct btrfs_backref_node *node;
2552 struct btrfs_path *path;
2553 struct tree_block *block;
2554 struct tree_block *next;
2558 path = btrfs_alloc_path();
2561 goto out_free_blocks;
2564 /* Kick in readahead for tree blocks with missing keys */
2565 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2566 if (!block->key_ready)
2567 readahead_tree_block(fs_info, block->bytenr);
2570 /* Get first keys */
2571 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2572 if (!block->key_ready) {
2573 err = get_tree_block_key(fs_info, block);
2579 /* Do tree relocation */
2580 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2581 node = build_backref_tree(rc, &block->key,
2582 block->level, block->bytenr);
2584 err = PTR_ERR(node);
2588 ret = relocate_tree_block(trans, rc, node, &block->key,
2596 err = finish_pending_nodes(trans, rc, path, err);
2599 btrfs_free_path(path);
2601 free_block_list(blocks);
2605 static noinline_for_stack int prealloc_file_extent_cluster(
2606 struct btrfs_inode *inode,
2607 struct file_extent_cluster *cluster)
2612 u64 offset = inode->index_cnt;
2616 u64 prealloc_start = cluster->start - offset;
2617 u64 prealloc_end = cluster->end - offset;
2618 u64 cur_offset = prealloc_start;
2620 BUG_ON(cluster->start != cluster->boundary[0]);
2621 ret = btrfs_alloc_data_chunk_ondemand(inode,
2622 prealloc_end + 1 - prealloc_start);
2626 inode_lock(&inode->vfs_inode);
2627 for (nr = 0; nr < cluster->nr; nr++) {
2628 start = cluster->boundary[nr] - offset;
2629 if (nr + 1 < cluster->nr)
2630 end = cluster->boundary[nr + 1] - 1 - offset;
2632 end = cluster->end - offset;
2634 lock_extent(&inode->io_tree, start, end);
2635 num_bytes = end + 1 - start;
2636 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2637 num_bytes, num_bytes,
2638 end + 1, &alloc_hint);
2639 cur_offset = end + 1;
2640 unlock_extent(&inode->io_tree, start, end);
2644 inode_unlock(&inode->vfs_inode);
2646 if (cur_offset < prealloc_end)
2647 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2648 prealloc_end + 1 - cur_offset);
2652 static noinline_for_stack
2653 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2656 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2657 struct extent_map *em;
2660 em = alloc_extent_map();
2665 em->len = end + 1 - start;
2666 em->block_len = em->len;
2667 em->block_start = block_start;
2668 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2670 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2672 write_lock(&em_tree->lock);
2673 ret = add_extent_mapping(em_tree, em, 0);
2674 write_unlock(&em_tree->lock);
2675 if (ret != -EEXIST) {
2676 free_extent_map(em);
2679 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2681 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2686 * Allow error injection to test balance cancellation
2688 int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2690 return atomic_read(&fs_info->balance_cancel_req) ||
2691 fatal_signal_pending(current);
2693 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2695 static int relocate_file_extent_cluster(struct inode *inode,
2696 struct file_extent_cluster *cluster)
2698 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2701 u64 offset = BTRFS_I(inode)->index_cnt;
2702 unsigned long index;
2703 unsigned long last_index;
2705 struct file_ra_state *ra;
2706 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2713 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2717 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2721 file_ra_state_init(ra, inode->i_mapping);
2723 ret = setup_extent_mapping(inode, cluster->start - offset,
2724 cluster->end - offset, cluster->start);
2728 index = (cluster->start - offset) >> PAGE_SHIFT;
2729 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2730 while (index <= last_index) {
2731 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2736 page = find_lock_page(inode->i_mapping, index);
2738 page_cache_sync_readahead(inode->i_mapping,
2740 last_index + 1 - index);
2741 page = find_or_create_page(inode->i_mapping, index,
2744 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2746 btrfs_delalloc_release_extents(BTRFS_I(inode),
2753 if (PageReadahead(page)) {
2754 page_cache_async_readahead(inode->i_mapping,
2755 ra, NULL, page, index,
2756 last_index + 1 - index);
2759 if (!PageUptodate(page)) {
2760 btrfs_readpage(NULL, page);
2762 if (!PageUptodate(page)) {
2765 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2767 btrfs_delalloc_release_extents(BTRFS_I(inode),
2774 page_start = page_offset(page);
2775 page_end = page_start + PAGE_SIZE - 1;
2777 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2779 set_page_extent_mapped(page);
2781 if (nr < cluster->nr &&
2782 page_start + offset == cluster->boundary[nr]) {
2783 set_extent_bits(&BTRFS_I(inode)->io_tree,
2784 page_start, page_end,
2789 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2794 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2796 btrfs_delalloc_release_extents(BTRFS_I(inode),
2799 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2800 page_start, page_end,
2801 EXTENT_LOCKED | EXTENT_BOUNDARY);
2805 set_page_dirty(page);
2807 unlock_extent(&BTRFS_I(inode)->io_tree,
2808 page_start, page_end);
2813 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2814 balance_dirty_pages_ratelimited(inode->i_mapping);
2815 btrfs_throttle(fs_info);
2816 if (btrfs_should_cancel_balance(fs_info)) {
2821 WARN_ON(nr != cluster->nr);
2827 static noinline_for_stack
2828 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2829 struct file_extent_cluster *cluster)
2833 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2834 ret = relocate_file_extent_cluster(inode, cluster);
2841 cluster->start = extent_key->objectid;
2843 BUG_ON(cluster->nr >= MAX_EXTENTS);
2844 cluster->end = extent_key->objectid + extent_key->offset - 1;
2845 cluster->boundary[cluster->nr] = extent_key->objectid;
2848 if (cluster->nr >= MAX_EXTENTS) {
2849 ret = relocate_file_extent_cluster(inode, cluster);
2858 * helper to add a tree block to the list.
2859 * the major work is getting the generation and level of the block
2861 static int add_tree_block(struct reloc_control *rc,
2862 struct btrfs_key *extent_key,
2863 struct btrfs_path *path,
2864 struct rb_root *blocks)
2866 struct extent_buffer *eb;
2867 struct btrfs_extent_item *ei;
2868 struct btrfs_tree_block_info *bi;
2869 struct tree_block *block;
2870 struct rb_node *rb_node;
2875 eb = path->nodes[0];
2876 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2878 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2879 item_size >= sizeof(*ei) + sizeof(*bi)) {
2880 ei = btrfs_item_ptr(eb, path->slots[0],
2881 struct btrfs_extent_item);
2882 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2883 bi = (struct btrfs_tree_block_info *)(ei + 1);
2884 level = btrfs_tree_block_level(eb, bi);
2886 level = (int)extent_key->offset;
2888 generation = btrfs_extent_generation(eb, ei);
2889 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2890 btrfs_print_v0_err(eb->fs_info);
2891 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2897 btrfs_release_path(path);
2899 BUG_ON(level == -1);
2901 block = kmalloc(sizeof(*block), GFP_NOFS);
2905 block->bytenr = extent_key->objectid;
2906 block->key.objectid = rc->extent_root->fs_info->nodesize;
2907 block->key.offset = generation;
2908 block->level = level;
2909 block->key_ready = 0;
2911 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2913 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2920 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2922 static int __add_tree_block(struct reloc_control *rc,
2923 u64 bytenr, u32 blocksize,
2924 struct rb_root *blocks)
2926 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2927 struct btrfs_path *path;
2928 struct btrfs_key key;
2930 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2932 if (tree_block_processed(bytenr, rc))
2935 if (rb_simple_search(blocks, bytenr))
2938 path = btrfs_alloc_path();
2942 key.objectid = bytenr;
2944 key.type = BTRFS_METADATA_ITEM_KEY;
2945 key.offset = (u64)-1;
2947 key.type = BTRFS_EXTENT_ITEM_KEY;
2948 key.offset = blocksize;
2951 path->search_commit_root = 1;
2952 path->skip_locking = 1;
2953 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2957 if (ret > 0 && skinny) {
2958 if (path->slots[0]) {
2960 btrfs_item_key_to_cpu(path->nodes[0], &key,
2962 if (key.objectid == bytenr &&
2963 (key.type == BTRFS_METADATA_ITEM_KEY ||
2964 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2965 key.offset == blocksize)))
2971 btrfs_release_path(path);
2977 btrfs_print_leaf(path->nodes[0]);
2979 "tree block extent item (%llu) is not found in extent tree",
2986 ret = add_tree_block(rc, &key, path, blocks);
2988 btrfs_free_path(path);
2992 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2993 struct btrfs_block_group *block_group,
2994 struct inode *inode,
2997 struct btrfs_root *root = fs_info->tree_root;
2998 struct btrfs_trans_handle *trans;
3004 inode = btrfs_iget(fs_info->sb, ino, root);
3009 ret = btrfs_check_trunc_cache_free_space(fs_info,
3010 &fs_info->global_block_rsv);
3014 trans = btrfs_join_transaction(root);
3015 if (IS_ERR(trans)) {
3016 ret = PTR_ERR(trans);
3020 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3022 btrfs_end_transaction(trans);
3023 btrfs_btree_balance_dirty(fs_info);
3030 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3031 * cache inode, to avoid free space cache data extent blocking data relocation.
3033 static int delete_v1_space_cache(struct extent_buffer *leaf,
3034 struct btrfs_block_group *block_group,
3037 u64 space_cache_ino;
3038 struct btrfs_file_extent_item *ei;
3039 struct btrfs_key key;
3044 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3047 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3050 btrfs_item_key_to_cpu(leaf, &key, i);
3051 if (key.type != BTRFS_EXTENT_DATA_KEY)
3053 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3054 type = btrfs_file_extent_type(leaf, ei);
3056 if ((type == BTRFS_FILE_EXTENT_REG ||
3057 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3058 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3060 space_cache_ino = key.objectid;
3066 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3072 * helper to find all tree blocks that reference a given data extent
3074 static noinline_for_stack
3075 int add_data_references(struct reloc_control *rc,
3076 struct btrfs_key *extent_key,
3077 struct btrfs_path *path,
3078 struct rb_root *blocks)
3080 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3081 struct ulist *leaves = NULL;
3082 struct ulist_iterator leaf_uiter;
3083 struct ulist_node *ref_node = NULL;
3084 const u32 blocksize = fs_info->nodesize;
3087 btrfs_release_path(path);
3088 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3089 0, &leaves, NULL, true);
3093 ULIST_ITER_INIT(&leaf_uiter);
3094 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3095 struct extent_buffer *eb;
3097 eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL);
3102 ret = delete_v1_space_cache(eb, rc->block_group,
3103 extent_key->objectid);
3104 free_extent_buffer(eb);
3107 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3112 free_block_list(blocks);
3118 * helper to find next unprocessed extent
3120 static noinline_for_stack
3121 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3122 struct btrfs_key *extent_key)
3124 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3125 struct btrfs_key key;
3126 struct extent_buffer *leaf;
3127 u64 start, end, last;
3130 last = rc->block_group->start + rc->block_group->length;
3133 if (rc->search_start >= last) {
3138 key.objectid = rc->search_start;
3139 key.type = BTRFS_EXTENT_ITEM_KEY;
3142 path->search_commit_root = 1;
3143 path->skip_locking = 1;
3144 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3149 leaf = path->nodes[0];
3150 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3151 ret = btrfs_next_leaf(rc->extent_root, path);
3154 leaf = path->nodes[0];
3157 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3158 if (key.objectid >= last) {
3163 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3164 key.type != BTRFS_METADATA_ITEM_KEY) {
3169 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3170 key.objectid + key.offset <= rc->search_start) {
3175 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3176 key.objectid + fs_info->nodesize <=
3182 ret = find_first_extent_bit(&rc->processed_blocks,
3183 key.objectid, &start, &end,
3184 EXTENT_DIRTY, NULL);
3186 if (ret == 0 && start <= key.objectid) {
3187 btrfs_release_path(path);
3188 rc->search_start = end + 1;
3190 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3191 rc->search_start = key.objectid + key.offset;
3193 rc->search_start = key.objectid +
3195 memcpy(extent_key, &key, sizeof(key));
3199 btrfs_release_path(path);
3203 static void set_reloc_control(struct reloc_control *rc)
3205 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3207 mutex_lock(&fs_info->reloc_mutex);
3208 fs_info->reloc_ctl = rc;
3209 mutex_unlock(&fs_info->reloc_mutex);
3212 static void unset_reloc_control(struct reloc_control *rc)
3214 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3216 mutex_lock(&fs_info->reloc_mutex);
3217 fs_info->reloc_ctl = NULL;
3218 mutex_unlock(&fs_info->reloc_mutex);
3221 static int check_extent_flags(u64 flags)
3223 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3224 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3226 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3227 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3229 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3230 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3235 static noinline_for_stack
3236 int prepare_to_relocate(struct reloc_control *rc)
3238 struct btrfs_trans_handle *trans;
3241 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3242 BTRFS_BLOCK_RSV_TEMP);
3246 memset(&rc->cluster, 0, sizeof(rc->cluster));
3247 rc->search_start = rc->block_group->start;
3248 rc->extents_found = 0;
3249 rc->nodes_relocated = 0;
3250 rc->merging_rsv_size = 0;
3251 rc->reserved_bytes = 0;
3252 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3253 RELOCATION_RESERVED_NODES;
3254 ret = btrfs_block_rsv_refill(rc->extent_root,
3255 rc->block_rsv, rc->block_rsv->size,
3256 BTRFS_RESERVE_FLUSH_ALL);
3260 rc->create_reloc_tree = 1;
3261 set_reloc_control(rc);
3263 trans = btrfs_join_transaction(rc->extent_root);
3264 if (IS_ERR(trans)) {
3265 unset_reloc_control(rc);
3267 * extent tree is not a ref_cow tree and has no reloc_root to
3268 * cleanup. And callers are responsible to free the above
3271 return PTR_ERR(trans);
3274 ret = btrfs_commit_transaction(trans);
3276 unset_reloc_control(rc);
3281 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3283 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3284 struct rb_root blocks = RB_ROOT;
3285 struct btrfs_key key;
3286 struct btrfs_trans_handle *trans = NULL;
3287 struct btrfs_path *path;
3288 struct btrfs_extent_item *ei;
3295 path = btrfs_alloc_path();
3298 path->reada = READA_FORWARD;
3300 ret = prepare_to_relocate(rc);
3307 rc->reserved_bytes = 0;
3308 ret = btrfs_block_rsv_refill(rc->extent_root,
3309 rc->block_rsv, rc->block_rsv->size,
3310 BTRFS_RESERVE_FLUSH_ALL);
3316 trans = btrfs_start_transaction(rc->extent_root, 0);
3317 if (IS_ERR(trans)) {
3318 err = PTR_ERR(trans);
3323 if (update_backref_cache(trans, &rc->backref_cache)) {
3324 btrfs_end_transaction(trans);
3329 ret = find_next_extent(rc, path, &key);
3335 rc->extents_found++;
3337 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3338 struct btrfs_extent_item);
3339 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3340 if (item_size >= sizeof(*ei)) {
3341 flags = btrfs_extent_flags(path->nodes[0], ei);
3342 ret = check_extent_flags(flags);
3344 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3346 btrfs_print_v0_err(trans->fs_info);
3347 btrfs_abort_transaction(trans, err);
3353 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3354 ret = add_tree_block(rc, &key, path, &blocks);
3355 } else if (rc->stage == UPDATE_DATA_PTRS &&
3356 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3357 ret = add_data_references(rc, &key, path, &blocks);
3359 btrfs_release_path(path);
3367 if (!RB_EMPTY_ROOT(&blocks)) {
3368 ret = relocate_tree_blocks(trans, rc, &blocks);
3370 if (ret != -EAGAIN) {
3374 rc->extents_found--;
3375 rc->search_start = key.objectid;
3379 btrfs_end_transaction_throttle(trans);
3380 btrfs_btree_balance_dirty(fs_info);
3383 if (rc->stage == MOVE_DATA_EXTENTS &&
3384 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3385 rc->found_file_extent = 1;
3386 ret = relocate_data_extent(rc->data_inode,
3387 &key, &rc->cluster);
3393 if (btrfs_should_cancel_balance(fs_info)) {
3398 if (trans && progress && err == -ENOSPC) {
3399 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3407 btrfs_release_path(path);
3408 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3411 btrfs_end_transaction_throttle(trans);
3412 btrfs_btree_balance_dirty(fs_info);
3416 ret = relocate_file_extent_cluster(rc->data_inode,
3422 rc->create_reloc_tree = 0;
3423 set_reloc_control(rc);
3425 btrfs_backref_release_cache(&rc->backref_cache);
3426 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3429 * Even in the case when the relocation is cancelled, we should all go
3430 * through prepare_to_merge() and merge_reloc_roots().
3432 * For error (including cancelled balance), prepare_to_merge() will
3433 * mark all reloc trees orphan, then queue them for cleanup in
3434 * merge_reloc_roots()
3436 err = prepare_to_merge(rc, err);
3438 merge_reloc_roots(rc);
3440 rc->merge_reloc_tree = 0;
3441 unset_reloc_control(rc);
3442 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3444 /* get rid of pinned extents */
3445 trans = btrfs_join_transaction(rc->extent_root);
3446 if (IS_ERR(trans)) {
3447 err = PTR_ERR(trans);
3450 ret = btrfs_commit_transaction(trans);
3454 ret = clean_dirty_subvols(rc);
3455 if (ret < 0 && !err)
3457 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3458 btrfs_free_path(path);
3462 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3463 struct btrfs_root *root, u64 objectid)
3465 struct btrfs_path *path;
3466 struct btrfs_inode_item *item;
3467 struct extent_buffer *leaf;
3470 path = btrfs_alloc_path();
3474 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3478 leaf = path->nodes[0];
3479 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3480 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3481 btrfs_set_inode_generation(leaf, item, 1);
3482 btrfs_set_inode_size(leaf, item, 0);
3483 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3484 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3485 BTRFS_INODE_PREALLOC);
3486 btrfs_mark_buffer_dirty(leaf);
3488 btrfs_free_path(path);
3493 * helper to create inode for data relocation.
3494 * the inode is in data relocation tree and its link count is 0
3496 static noinline_for_stack
3497 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3498 struct btrfs_block_group *group)
3500 struct inode *inode = NULL;
3501 struct btrfs_trans_handle *trans;
3502 struct btrfs_root *root;
3506 root = btrfs_grab_root(fs_info->data_reloc_root);
3507 trans = btrfs_start_transaction(root, 6);
3508 if (IS_ERR(trans)) {
3509 btrfs_put_root(root);
3510 return ERR_CAST(trans);
3513 err = btrfs_find_free_objectid(root, &objectid);
3517 err = __insert_orphan_inode(trans, root, objectid);
3520 inode = btrfs_iget(fs_info->sb, objectid, root);
3521 BUG_ON(IS_ERR(inode));
3522 BTRFS_I(inode)->index_cnt = group->start;
3524 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3526 btrfs_put_root(root);
3527 btrfs_end_transaction(trans);
3528 btrfs_btree_balance_dirty(fs_info);
3532 inode = ERR_PTR(err);
3537 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3539 struct reloc_control *rc;
3541 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3545 INIT_LIST_HEAD(&rc->reloc_roots);
3546 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3547 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3548 mapping_tree_init(&rc->reloc_root_tree);
3549 extent_io_tree_init(fs_info, &rc->processed_blocks,
3550 IO_TREE_RELOC_BLOCKS, NULL);
3554 static void free_reloc_control(struct reloc_control *rc)
3556 struct mapping_node *node, *tmp;
3558 free_reloc_roots(&rc->reloc_roots);
3559 rbtree_postorder_for_each_entry_safe(node, tmp,
3560 &rc->reloc_root_tree.rb_root, rb_node)
3567 * Print the block group being relocated
3569 static void describe_relocation(struct btrfs_fs_info *fs_info,
3570 struct btrfs_block_group *block_group)
3572 char buf[128] = {'\0'};
3574 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3577 "relocating block group %llu flags %s",
3578 block_group->start, buf);
3581 static const char *stage_to_string(int stage)
3583 if (stage == MOVE_DATA_EXTENTS)
3584 return "move data extents";
3585 if (stage == UPDATE_DATA_PTRS)
3586 return "update data pointers";
3591 * function to relocate all extents in a block group.
3593 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3595 struct btrfs_block_group *bg;
3596 struct btrfs_root *extent_root = fs_info->extent_root;
3597 struct reloc_control *rc;
3598 struct inode *inode;
3599 struct btrfs_path *path;
3604 bg = btrfs_lookup_block_group(fs_info, group_start);
3608 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3609 btrfs_put_block_group(bg);
3613 rc = alloc_reloc_control(fs_info);
3615 btrfs_put_block_group(bg);
3619 rc->extent_root = extent_root;
3620 rc->block_group = bg;
3622 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3629 path = btrfs_alloc_path();
3635 inode = lookup_free_space_inode(rc->block_group, path);
3636 btrfs_free_path(path);
3639 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3641 ret = PTR_ERR(inode);
3643 if (ret && ret != -ENOENT) {
3648 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3649 if (IS_ERR(rc->data_inode)) {
3650 err = PTR_ERR(rc->data_inode);
3651 rc->data_inode = NULL;
3655 describe_relocation(fs_info, rc->block_group);
3657 btrfs_wait_block_group_reservations(rc->block_group);
3658 btrfs_wait_nocow_writers(rc->block_group);
3659 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3660 rc->block_group->start,
3661 rc->block_group->length);
3666 mutex_lock(&fs_info->cleaner_mutex);
3667 ret = relocate_block_group(rc);
3668 mutex_unlock(&fs_info->cleaner_mutex);
3672 finishes_stage = rc->stage;
3674 * We may have gotten ENOSPC after we already dirtied some
3675 * extents. If writeout happens while we're relocating a
3676 * different block group we could end up hitting the
3677 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3678 * btrfs_reloc_cow_block. Make sure we write everything out
3679 * properly so we don't trip over this problem, and then break
3680 * out of the loop if we hit an error.
3682 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3683 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3687 invalidate_mapping_pages(rc->data_inode->i_mapping,
3689 rc->stage = UPDATE_DATA_PTRS;
3695 if (rc->extents_found == 0)
3698 btrfs_info(fs_info, "found %llu extents, stage: %s",
3699 rc->extents_found, stage_to_string(finishes_stage));
3702 WARN_ON(rc->block_group->pinned > 0);
3703 WARN_ON(rc->block_group->reserved > 0);
3704 WARN_ON(rc->block_group->used > 0);
3707 btrfs_dec_block_group_ro(rc->block_group);
3708 iput(rc->data_inode);
3709 btrfs_put_block_group(rc->block_group);
3710 free_reloc_control(rc);
3714 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3716 struct btrfs_fs_info *fs_info = root->fs_info;
3717 struct btrfs_trans_handle *trans;
3720 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3722 return PTR_ERR(trans);
3724 memset(&root->root_item.drop_progress, 0,
3725 sizeof(root->root_item.drop_progress));
3726 root->root_item.drop_level = 0;
3727 btrfs_set_root_refs(&root->root_item, 0);
3728 ret = btrfs_update_root(trans, fs_info->tree_root,
3729 &root->root_key, &root->root_item);
3731 err = btrfs_end_transaction(trans);
3738 * recover relocation interrupted by system crash.
3740 * this function resumes merging reloc trees with corresponding fs trees.
3741 * this is important for keeping the sharing of tree blocks
3743 int btrfs_recover_relocation(struct btrfs_root *root)
3745 struct btrfs_fs_info *fs_info = root->fs_info;
3746 LIST_HEAD(reloc_roots);
3747 struct btrfs_key key;
3748 struct btrfs_root *fs_root;
3749 struct btrfs_root *reloc_root;
3750 struct btrfs_path *path;
3751 struct extent_buffer *leaf;
3752 struct reloc_control *rc = NULL;
3753 struct btrfs_trans_handle *trans;
3757 path = btrfs_alloc_path();
3760 path->reada = READA_BACK;
3762 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3763 key.type = BTRFS_ROOT_ITEM_KEY;
3764 key.offset = (u64)-1;
3767 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3774 if (path->slots[0] == 0)
3778 leaf = path->nodes[0];
3779 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3780 btrfs_release_path(path);
3782 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3783 key.type != BTRFS_ROOT_ITEM_KEY)
3786 reloc_root = btrfs_read_tree_root(root, &key);
3787 if (IS_ERR(reloc_root)) {
3788 err = PTR_ERR(reloc_root);
3792 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3793 list_add(&reloc_root->root_list, &reloc_roots);
3795 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3796 fs_root = btrfs_get_fs_root(fs_info,
3797 reloc_root->root_key.offset, false);
3798 if (IS_ERR(fs_root)) {
3799 ret = PTR_ERR(fs_root);
3800 if (ret != -ENOENT) {
3804 ret = mark_garbage_root(reloc_root);
3810 btrfs_put_root(fs_root);
3814 if (key.offset == 0)
3819 btrfs_release_path(path);
3821 if (list_empty(&reloc_roots))
3824 rc = alloc_reloc_control(fs_info);
3830 rc->extent_root = fs_info->extent_root;
3832 set_reloc_control(rc);
3834 trans = btrfs_join_transaction(rc->extent_root);
3835 if (IS_ERR(trans)) {
3836 err = PTR_ERR(trans);
3840 rc->merge_reloc_tree = 1;
3842 while (!list_empty(&reloc_roots)) {
3843 reloc_root = list_entry(reloc_roots.next,
3844 struct btrfs_root, root_list);
3845 list_del(&reloc_root->root_list);
3847 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3848 list_add_tail(&reloc_root->root_list,
3853 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
3855 if (IS_ERR(fs_root)) {
3856 err = PTR_ERR(fs_root);
3857 list_add_tail(&reloc_root->root_list, &reloc_roots);
3858 btrfs_end_transaction(trans);
3862 err = __add_reloc_root(reloc_root);
3863 BUG_ON(err < 0); /* -ENOMEM or logic error */
3864 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3865 btrfs_put_root(fs_root);
3868 err = btrfs_commit_transaction(trans);
3872 merge_reloc_roots(rc);
3874 unset_reloc_control(rc);
3876 trans = btrfs_join_transaction(rc->extent_root);
3877 if (IS_ERR(trans)) {
3878 err = PTR_ERR(trans);
3881 err = btrfs_commit_transaction(trans);
3883 ret = clean_dirty_subvols(rc);
3884 if (ret < 0 && !err)
3887 unset_reloc_control(rc);
3888 free_reloc_control(rc);
3890 free_reloc_roots(&reloc_roots);
3892 btrfs_free_path(path);
3895 /* cleanup orphan inode in data relocation tree */
3896 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
3898 err = btrfs_orphan_cleanup(fs_root);
3899 btrfs_put_root(fs_root);
3905 * helper to add ordered checksum for data relocation.
3907 * cloning checksum properly handles the nodatasum extents.
3908 * it also saves CPU time to re-calculate the checksum.
3910 int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
3912 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3913 struct btrfs_ordered_sum *sums;
3914 struct btrfs_ordered_extent *ordered;
3920 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3921 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3923 disk_bytenr = file_pos + inode->index_cnt;
3924 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3925 disk_bytenr + len - 1, &list, 0);
3929 while (!list_empty(&list)) {
3930 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3931 list_del_init(&sums->list);
3934 * We need to offset the new_bytenr based on where the csum is.
3935 * We need to do this because we will read in entire prealloc
3936 * extents but we may have written to say the middle of the
3937 * prealloc extent, so we need to make sure the csum goes with
3938 * the right disk offset.
3940 * We can do this because the data reloc inode refers strictly
3941 * to the on disk bytes, so we don't have to worry about
3942 * disk_len vs real len like with real inodes since it's all
3945 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3946 sums->bytenr = new_bytenr;
3948 btrfs_add_ordered_sum(ordered, sums);
3951 btrfs_put_ordered_extent(ordered);
3955 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3956 struct btrfs_root *root, struct extent_buffer *buf,
3957 struct extent_buffer *cow)
3959 struct btrfs_fs_info *fs_info = root->fs_info;
3960 struct reloc_control *rc;
3961 struct btrfs_backref_node *node;
3966 rc = fs_info->reloc_ctl;
3970 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3971 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3973 level = btrfs_header_level(buf);
3974 if (btrfs_header_generation(buf) <=
3975 btrfs_root_last_snapshot(&root->root_item))
3978 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3979 rc->create_reloc_tree) {
3980 WARN_ON(!first_cow && level == 0);
3982 node = rc->backref_cache.path[level];
3983 BUG_ON(node->bytenr != buf->start &&
3984 node->new_bytenr != buf->start);
3986 btrfs_backref_drop_node_buffer(node);
3987 atomic_inc(&cow->refs);
3989 node->new_bytenr = cow->start;
3991 if (!node->pending) {
3992 list_move_tail(&node->list,
3993 &rc->backref_cache.pending[level]);
3998 mark_block_processed(rc, node);
4000 if (first_cow && level > 0)
4001 rc->nodes_relocated += buf->len;
4004 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4005 ret = replace_file_extents(trans, rc, root, cow);
4010 * called before creating snapshot. it calculates metadata reservation
4011 * required for relocating tree blocks in the snapshot
4013 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4014 u64 *bytes_to_reserve)
4016 struct btrfs_root *root = pending->root;
4017 struct reloc_control *rc = root->fs_info->reloc_ctl;
4019 if (!rc || !have_reloc_root(root))
4022 if (!rc->merge_reloc_tree)
4025 root = root->reloc_root;
4026 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4028 * relocation is in the stage of merging trees. the space
4029 * used by merging a reloc tree is twice the size of
4030 * relocated tree nodes in the worst case. half for cowing
4031 * the reloc tree, half for cowing the fs tree. the space
4032 * used by cowing the reloc tree will be freed after the
4033 * tree is dropped. if we create snapshot, cowing the fs
4034 * tree may use more space than it frees. so we need
4035 * reserve extra space.
4037 *bytes_to_reserve += rc->nodes_relocated;
4041 * called after snapshot is created. migrate block reservation
4042 * and create reloc root for the newly created snapshot
4044 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4045 * references held on the reloc_root, one for root->reloc_root and one for
4048 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4049 struct btrfs_pending_snapshot *pending)
4051 struct btrfs_root *root = pending->root;
4052 struct btrfs_root *reloc_root;
4053 struct btrfs_root *new_root;
4054 struct reloc_control *rc = root->fs_info->reloc_ctl;
4057 if (!rc || !have_reloc_root(root))
4060 rc = root->fs_info->reloc_ctl;
4061 rc->merging_rsv_size += rc->nodes_relocated;
4063 if (rc->merge_reloc_tree) {
4064 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4066 rc->nodes_relocated, true);
4071 new_root = pending->snap;
4072 reloc_root = create_reloc_root(trans, root->reloc_root,
4073 new_root->root_key.objectid);
4074 if (IS_ERR(reloc_root))
4075 return PTR_ERR(reloc_root);
4077 ret = __add_reloc_root(reloc_root);
4079 new_root->reloc_root = btrfs_grab_root(reloc_root);
4081 if (rc->create_reloc_tree)
4082 ret = clone_backref_node(trans, rc, root, reloc_root);