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"
22 #include "print-tree.h"
23 #include "delalloc-space.h"
24 #include "block-group.h"
31 * [What does relocation do]
33 * The objective of relocation is to relocate all extents of the target block
34 * group to other block groups.
35 * This is utilized by resize (shrink only), profile converting, compacting
36 * space, or balance routine to spread chunks over devices.
39 * ------------------------------------------------------------------
40 * BG A: 10 data extents | BG A: deleted
41 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
42 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
44 * [How does relocation work]
46 * 1. Mark the target block group read-only
47 * New extents won't be allocated from the target block group.
49 * 2.1 Record each extent in the target block group
50 * To build a proper map of extents to be relocated.
52 * 2.2 Build data reloc tree and reloc trees
53 * Data reloc tree will contain an inode, recording all newly relocated
55 * There will be only one data reloc tree for one data block group.
57 * Reloc tree will be a special snapshot of its source tree, containing
58 * relocated tree blocks.
59 * Each tree referring to a tree block in target block group will get its
62 * 2.3 Swap source tree with its corresponding reloc tree
63 * Each involved tree only refers to new extents after swap.
65 * 3. Cleanup reloc trees and data reloc tree.
66 * As old extents in the target block group are still referenced by reloc
67 * trees, we need to clean them up before really freeing the target block
70 * The main complexity is in steps 2.2 and 2.3.
72 * The entry point of relocation is relocate_block_group() function.
75 #define RELOCATION_RESERVED_NODES 256
77 * map address of tree root to tree
81 struct rb_node rb_node;
83 }; /* Use rb_simle_node for search/insert */
88 struct rb_root rb_root;
93 * present a tree block to process
97 struct rb_node rb_node;
99 }; /* 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;
738 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
741 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
742 root_key.type = BTRFS_ROOT_ITEM_KEY;
743 root_key.offset = objectid;
745 if (root->root_key.objectid == objectid) {
748 /* called by btrfs_init_reloc_root */
749 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
750 BTRFS_TREE_RELOC_OBJECTID);
753 * Set the last_snapshot field to the generation of the commit
754 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
755 * correctly (returns true) when the relocation root is created
756 * either inside the critical section of a transaction commit
757 * (through transaction.c:qgroup_account_snapshot()) and when
758 * it's created before the transaction commit is started.
760 commit_root_gen = btrfs_header_generation(root->commit_root);
761 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
764 * called by btrfs_reloc_post_snapshot_hook.
765 * the source tree is a reloc tree, all tree blocks
766 * modified after it was created have RELOC flag
767 * set in their headers. so it's OK to not update
768 * the 'last_snapshot'.
770 ret = btrfs_copy_root(trans, root, root->node, &eb,
771 BTRFS_TREE_RELOC_OBJECTID);
775 memcpy(root_item, &root->root_item, sizeof(*root_item));
776 btrfs_set_root_bytenr(root_item, eb->start);
777 btrfs_set_root_level(root_item, btrfs_header_level(eb));
778 btrfs_set_root_generation(root_item, trans->transid);
780 if (root->root_key.objectid == objectid) {
781 btrfs_set_root_refs(root_item, 0);
782 memset(&root_item->drop_progress, 0,
783 sizeof(struct btrfs_disk_key));
784 btrfs_set_root_drop_level(root_item, 0);
787 btrfs_tree_unlock(eb);
788 free_extent_buffer(eb);
790 ret = btrfs_insert_root(trans, fs_info->tree_root,
791 &root_key, root_item);
795 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
796 BUG_ON(IS_ERR(reloc_root));
797 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
798 reloc_root->last_trans = trans->transid;
803 * create reloc tree for a given fs tree. reloc tree is just a
804 * snapshot of the fs tree with special root objectid.
806 * The reloc_root comes out of here with two references, one for
807 * root->reloc_root, and another for being on the rc->reloc_roots list.
809 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root)
812 struct btrfs_fs_info *fs_info = root->fs_info;
813 struct btrfs_root *reloc_root;
814 struct reloc_control *rc = fs_info->reloc_ctl;
815 struct btrfs_block_rsv *rsv;
823 * The subvolume has reloc tree but the swap is finished, no need to
824 * create/update the dead reloc tree
826 if (reloc_root_is_dead(root))
830 * This is subtle but important. We do not do
831 * record_root_in_transaction for reloc roots, instead we record their
832 * corresponding fs root, and then here we update the last trans for the
833 * reloc root. This means that we have to do this for the entire life
834 * of the reloc root, regardless of which stage of the relocation we are
837 if (root->reloc_root) {
838 reloc_root = root->reloc_root;
839 reloc_root->last_trans = trans->transid;
844 * We are merging reloc roots, we do not need new reloc trees. Also
845 * reloc trees never need their own reloc tree.
847 if (!rc->create_reloc_tree ||
848 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
851 if (!trans->reloc_reserved) {
852 rsv = trans->block_rsv;
853 trans->block_rsv = rc->block_rsv;
856 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
858 trans->block_rsv = rsv;
860 ret = __add_reloc_root(reloc_root);
862 root->reloc_root = btrfs_grab_root(reloc_root);
867 * update root item of reloc tree
869 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
870 struct btrfs_root *root)
872 struct btrfs_fs_info *fs_info = root->fs_info;
873 struct btrfs_root *reloc_root;
874 struct btrfs_root_item *root_item;
877 if (!have_reloc_root(root))
880 reloc_root = root->reloc_root;
881 root_item = &reloc_root->root_item;
884 * We are probably ok here, but __del_reloc_root() will drop its ref of
885 * the root. We have the ref for root->reloc_root, but just in case
886 * hold it while we update the reloc root.
888 btrfs_grab_root(reloc_root);
890 /* root->reloc_root will stay until current relocation finished */
891 if (fs_info->reloc_ctl->merge_reloc_tree &&
892 btrfs_root_refs(root_item) == 0) {
893 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
895 * Mark the tree as dead before we change reloc_root so
896 * have_reloc_root will not touch it from now on.
899 __del_reloc_root(reloc_root);
902 if (reloc_root->commit_root != reloc_root->node) {
903 __update_reloc_root(reloc_root);
904 btrfs_set_root_node(root_item, reloc_root->node);
905 free_extent_buffer(reloc_root->commit_root);
906 reloc_root->commit_root = btrfs_root_node(reloc_root);
909 ret = btrfs_update_root(trans, fs_info->tree_root,
910 &reloc_root->root_key, root_item);
912 btrfs_put_root(reloc_root);
918 * helper to find first cached inode with inode number >= objectid
921 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
923 struct rb_node *node;
924 struct rb_node *prev;
925 struct btrfs_inode *entry;
928 spin_lock(&root->inode_lock);
930 node = root->inode_tree.rb_node;
934 entry = rb_entry(node, struct btrfs_inode, rb_node);
936 if (objectid < btrfs_ino(entry))
937 node = node->rb_left;
938 else if (objectid > btrfs_ino(entry))
939 node = node->rb_right;
945 entry = rb_entry(prev, struct btrfs_inode, rb_node);
946 if (objectid <= btrfs_ino(entry)) {
950 prev = rb_next(prev);
954 entry = rb_entry(node, struct btrfs_inode, rb_node);
955 inode = igrab(&entry->vfs_inode);
957 spin_unlock(&root->inode_lock);
961 objectid = btrfs_ino(entry) + 1;
962 if (cond_resched_lock(&root->inode_lock))
965 node = rb_next(node);
967 spin_unlock(&root->inode_lock);
972 * get new location of data
974 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
975 u64 bytenr, u64 num_bytes)
977 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
978 struct btrfs_path *path;
979 struct btrfs_file_extent_item *fi;
980 struct extent_buffer *leaf;
983 path = btrfs_alloc_path();
987 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
988 ret = btrfs_lookup_file_extent(NULL, root, path,
989 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
997 leaf = path->nodes[0];
998 fi = btrfs_item_ptr(leaf, path->slots[0],
999 struct btrfs_file_extent_item);
1001 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1002 btrfs_file_extent_compression(leaf, fi) ||
1003 btrfs_file_extent_encryption(leaf, fi) ||
1004 btrfs_file_extent_other_encoding(leaf, fi));
1006 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1011 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1014 btrfs_free_path(path);
1019 * update file extent items in the tree leaf to point to
1020 * the new locations.
1022 static noinline_for_stack
1023 int replace_file_extents(struct btrfs_trans_handle *trans,
1024 struct reloc_control *rc,
1025 struct btrfs_root *root,
1026 struct extent_buffer *leaf)
1028 struct btrfs_fs_info *fs_info = root->fs_info;
1029 struct btrfs_key key;
1030 struct btrfs_file_extent_item *fi;
1031 struct inode *inode = NULL;
1043 if (rc->stage != UPDATE_DATA_PTRS)
1046 /* reloc trees always use full backref */
1047 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1048 parent = leaf->start;
1052 nritems = btrfs_header_nritems(leaf);
1053 for (i = 0; i < nritems; i++) {
1054 struct btrfs_ref ref = { 0 };
1057 btrfs_item_key_to_cpu(leaf, &key, i);
1058 if (key.type != BTRFS_EXTENT_DATA_KEY)
1060 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1061 if (btrfs_file_extent_type(leaf, fi) ==
1062 BTRFS_FILE_EXTENT_INLINE)
1064 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1065 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1068 if (!in_range(bytenr, rc->block_group->start,
1069 rc->block_group->length))
1073 * if we are modifying block in fs tree, wait for readpage
1074 * to complete and drop the extent cache
1076 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1078 inode = find_next_inode(root, key.objectid);
1080 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1081 btrfs_add_delayed_iput(inode);
1082 inode = find_next_inode(root, key.objectid);
1084 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1086 btrfs_file_extent_num_bytes(leaf, fi);
1087 WARN_ON(!IS_ALIGNED(key.offset,
1088 fs_info->sectorsize));
1089 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1091 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1096 btrfs_drop_extent_cache(BTRFS_I(inode),
1097 key.offset, end, 1);
1098 unlock_extent(&BTRFS_I(inode)->io_tree,
1103 ret = get_new_location(rc->data_inode, &new_bytenr,
1107 * Don't have to abort since we've not changed anything
1108 * in the file extent yet.
1113 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1116 key.offset -= btrfs_file_extent_offset(leaf, fi);
1117 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1119 ref.real_root = root->root_key.objectid;
1120 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1121 key.objectid, key.offset);
1122 ret = btrfs_inc_extent_ref(trans, &ref);
1124 btrfs_abort_transaction(trans, ret);
1128 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1130 ref.real_root = root->root_key.objectid;
1131 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1132 key.objectid, key.offset);
1133 ret = btrfs_free_extent(trans, &ref);
1135 btrfs_abort_transaction(trans, ret);
1140 btrfs_mark_buffer_dirty(leaf);
1142 btrfs_add_delayed_iput(inode);
1146 static noinline_for_stack
1147 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1148 struct btrfs_path *path, int level)
1150 struct btrfs_disk_key key1;
1151 struct btrfs_disk_key key2;
1152 btrfs_node_key(eb, &key1, slot);
1153 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1154 return memcmp(&key1, &key2, sizeof(key1));
1158 * try to replace tree blocks in fs tree with the new blocks
1159 * in reloc tree. tree blocks haven't been modified since the
1160 * reloc tree was create can be replaced.
1162 * if a block was replaced, level of the block + 1 is returned.
1163 * if no block got replaced, 0 is returned. if there are other
1164 * errors, a negative error number is returned.
1166 static noinline_for_stack
1167 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1168 struct btrfs_root *dest, struct btrfs_root *src,
1169 struct btrfs_path *path, struct btrfs_key *next_key,
1170 int lowest_level, int max_level)
1172 struct btrfs_fs_info *fs_info = dest->fs_info;
1173 struct extent_buffer *eb;
1174 struct extent_buffer *parent;
1175 struct btrfs_ref ref = { 0 };
1176 struct btrfs_key key;
1188 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1189 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1191 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1193 slot = path->slots[lowest_level];
1194 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1196 eb = btrfs_lock_root_node(dest);
1197 level = btrfs_header_level(eb);
1199 if (level < lowest_level) {
1200 btrfs_tree_unlock(eb);
1201 free_extent_buffer(eb);
1206 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1212 next_key->objectid = (u64)-1;
1213 next_key->type = (u8)-1;
1214 next_key->offset = (u64)-1;
1219 level = btrfs_header_level(parent);
1220 BUG_ON(level < lowest_level);
1222 ret = btrfs_bin_search(parent, &key, &slot);
1225 if (ret && slot > 0)
1228 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1229 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1231 old_bytenr = btrfs_node_blockptr(parent, slot);
1232 blocksize = fs_info->nodesize;
1233 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1235 if (level <= max_level) {
1236 eb = path->nodes[level];
1237 new_bytenr = btrfs_node_blockptr(eb,
1238 path->slots[level]);
1239 new_ptr_gen = btrfs_node_ptr_generation(eb,
1240 path->slots[level]);
1246 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1251 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1252 memcmp_node_keys(parent, slot, path, level)) {
1253 if (level <= lowest_level) {
1258 eb = btrfs_read_node_slot(parent, slot);
1263 btrfs_tree_lock(eb);
1265 ret = btrfs_cow_block(trans, dest, eb, parent,
1271 btrfs_tree_unlock(parent);
1272 free_extent_buffer(parent);
1279 btrfs_tree_unlock(parent);
1280 free_extent_buffer(parent);
1285 btrfs_node_key_to_cpu(path->nodes[level], &key,
1286 path->slots[level]);
1287 btrfs_release_path(path);
1289 path->lowest_level = level;
1290 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1291 path->lowest_level = 0;
1295 * Info qgroup to trace both subtrees.
1297 * We must trace both trees.
1298 * 1) Tree reloc subtree
1299 * If not traced, we will leak data numbers
1301 * If not traced, we will double count old data
1303 * We don't scan the subtree right now, but only record
1304 * the swapped tree blocks.
1305 * The real subtree rescan is delayed until we have new
1306 * CoW on the subtree root node before transaction commit.
1308 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1309 rc->block_group, parent, slot,
1310 path->nodes[level], path->slots[level],
1315 * swap blocks in fs tree and reloc tree.
1317 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1318 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1319 btrfs_mark_buffer_dirty(parent);
1321 btrfs_set_node_blockptr(path->nodes[level],
1322 path->slots[level], old_bytenr);
1323 btrfs_set_node_ptr_generation(path->nodes[level],
1324 path->slots[level], old_ptr_gen);
1325 btrfs_mark_buffer_dirty(path->nodes[level]);
1327 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1328 blocksize, path->nodes[level]->start);
1329 ref.skip_qgroup = true;
1330 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1331 ret = btrfs_inc_extent_ref(trans, &ref);
1333 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1335 ref.skip_qgroup = true;
1336 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1337 ret = btrfs_inc_extent_ref(trans, &ref);
1340 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1341 blocksize, path->nodes[level]->start);
1342 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1343 ref.skip_qgroup = true;
1344 ret = btrfs_free_extent(trans, &ref);
1347 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1349 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1350 ref.skip_qgroup = true;
1351 ret = btrfs_free_extent(trans, &ref);
1354 btrfs_unlock_up_safe(path, 0);
1359 btrfs_tree_unlock(parent);
1360 free_extent_buffer(parent);
1365 * helper to find next relocated block in reloc tree
1367 static noinline_for_stack
1368 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1371 struct extent_buffer *eb;
1376 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1378 for (i = 0; i < *level; i++) {
1379 free_extent_buffer(path->nodes[i]);
1380 path->nodes[i] = NULL;
1383 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1384 eb = path->nodes[i];
1385 nritems = btrfs_header_nritems(eb);
1386 while (path->slots[i] + 1 < nritems) {
1388 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1395 free_extent_buffer(path->nodes[i]);
1396 path->nodes[i] = NULL;
1402 * walk down reloc tree to find relocated block of lowest level
1404 static noinline_for_stack
1405 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1408 struct extent_buffer *eb = NULL;
1414 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1416 for (i = *level; i > 0; i--) {
1417 eb = path->nodes[i];
1418 nritems = btrfs_header_nritems(eb);
1419 while (path->slots[i] < nritems) {
1420 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1421 if (ptr_gen > last_snapshot)
1425 if (path->slots[i] >= nritems) {
1436 eb = btrfs_read_node_slot(eb, path->slots[i]);
1439 BUG_ON(btrfs_header_level(eb) != i - 1);
1440 path->nodes[i - 1] = eb;
1441 path->slots[i - 1] = 0;
1447 * invalidate extent cache for file extents whose key in range of
1448 * [min_key, max_key)
1450 static int invalidate_extent_cache(struct btrfs_root *root,
1451 struct btrfs_key *min_key,
1452 struct btrfs_key *max_key)
1454 struct btrfs_fs_info *fs_info = root->fs_info;
1455 struct inode *inode = NULL;
1460 objectid = min_key->objectid;
1465 if (objectid > max_key->objectid)
1468 inode = find_next_inode(root, objectid);
1471 ino = btrfs_ino(BTRFS_I(inode));
1473 if (ino > max_key->objectid) {
1479 if (!S_ISREG(inode->i_mode))
1482 if (unlikely(min_key->objectid == ino)) {
1483 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1485 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1488 start = min_key->offset;
1489 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1495 if (unlikely(max_key->objectid == ino)) {
1496 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1498 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1501 if (max_key->offset == 0)
1503 end = max_key->offset;
1504 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1511 /* the lock_extent waits for readpage to complete */
1512 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1513 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1514 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1519 static int find_next_key(struct btrfs_path *path, int level,
1520 struct btrfs_key *key)
1523 while (level < BTRFS_MAX_LEVEL) {
1524 if (!path->nodes[level])
1526 if (path->slots[level] + 1 <
1527 btrfs_header_nritems(path->nodes[level])) {
1528 btrfs_node_key_to_cpu(path->nodes[level], key,
1529 path->slots[level] + 1);
1538 * Insert current subvolume into reloc_control::dirty_subvol_roots
1540 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1541 struct reloc_control *rc,
1542 struct btrfs_root *root)
1544 struct btrfs_root *reloc_root = root->reloc_root;
1545 struct btrfs_root_item *reloc_root_item;
1547 /* @root must be a subvolume tree root with a valid reloc tree */
1548 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1551 reloc_root_item = &reloc_root->root_item;
1552 memset(&reloc_root_item->drop_progress, 0,
1553 sizeof(reloc_root_item->drop_progress));
1554 btrfs_set_root_drop_level(reloc_root_item, 0);
1555 btrfs_set_root_refs(reloc_root_item, 0);
1556 btrfs_update_reloc_root(trans, root);
1558 if (list_empty(&root->reloc_dirty_list)) {
1559 btrfs_grab_root(root);
1560 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1564 static int clean_dirty_subvols(struct reloc_control *rc)
1566 struct btrfs_root *root;
1567 struct btrfs_root *next;
1571 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1573 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1574 /* Merged subvolume, cleanup its reloc root */
1575 struct btrfs_root *reloc_root = root->reloc_root;
1577 list_del_init(&root->reloc_dirty_list);
1578 root->reloc_root = NULL;
1580 * Need barrier to ensure clear_bit() only happens after
1581 * root->reloc_root = NULL. Pairs with have_reloc_root.
1584 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1587 * btrfs_drop_snapshot drops our ref we hold for
1588 * ->reloc_root. If it fails however we must
1589 * drop the ref ourselves.
1591 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1593 btrfs_put_root(reloc_root);
1598 btrfs_put_root(root);
1600 /* Orphan reloc tree, just clean it up */
1601 ret2 = btrfs_drop_snapshot(root, 0, 1);
1603 btrfs_put_root(root);
1613 * merge the relocated tree blocks in reloc tree with corresponding
1616 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1617 struct btrfs_root *root)
1619 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1620 struct btrfs_key key;
1621 struct btrfs_key next_key;
1622 struct btrfs_trans_handle *trans = NULL;
1623 struct btrfs_root *reloc_root;
1624 struct btrfs_root_item *root_item;
1625 struct btrfs_path *path;
1626 struct extent_buffer *leaf;
1634 path = btrfs_alloc_path();
1637 path->reada = READA_FORWARD;
1639 reloc_root = root->reloc_root;
1640 root_item = &reloc_root->root_item;
1642 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1643 level = btrfs_root_level(root_item);
1644 atomic_inc(&reloc_root->node->refs);
1645 path->nodes[level] = reloc_root->node;
1646 path->slots[level] = 0;
1648 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1650 level = btrfs_root_drop_level(root_item);
1652 path->lowest_level = level;
1653 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1654 path->lowest_level = 0;
1656 btrfs_free_path(path);
1660 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1661 path->slots[level]);
1662 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1664 btrfs_unlock_up_safe(path, 0);
1668 * In merge_reloc_root(), we modify the upper level pointer to swap the
1669 * tree blocks between reloc tree and subvolume tree. Thus for tree
1670 * block COW, we COW at most from level 1 to root level for each tree.
1672 * Thus the needed metadata size is at most root_level * nodesize,
1673 * and * 2 since we have two trees to COW.
1675 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1676 min_reserved = fs_info->nodesize * reserve_level * 2;
1677 memset(&next_key, 0, sizeof(next_key));
1680 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1681 BTRFS_RESERVE_FLUSH_LIMIT);
1684 trans = btrfs_start_transaction(root, 0);
1685 if (IS_ERR(trans)) {
1686 ret = PTR_ERR(trans);
1692 * At this point we no longer have a reloc_control, so we can't
1693 * depend on btrfs_init_reloc_root to update our last_trans.
1695 * But that's ok, we started the trans handle on our
1696 * corresponding fs_root, which means it's been added to the
1697 * dirty list. At commit time we'll still call
1698 * btrfs_update_reloc_root() and update our root item
1701 reloc_root->last_trans = trans->transid;
1702 trans->block_rsv = rc->block_rsv;
1707 ret = walk_down_reloc_tree(reloc_root, path, &level);
1713 if (!find_next_key(path, level, &key) &&
1714 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1717 ret = replace_path(trans, rc, root, reloc_root, path,
1718 &next_key, level, max_level);
1724 btrfs_node_key_to_cpu(path->nodes[level], &key,
1725 path->slots[level]);
1729 ret = walk_up_reloc_tree(reloc_root, path, &level);
1735 * save the merging progress in the drop_progress.
1736 * this is OK since root refs == 1 in this case.
1738 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1739 path->slots[level]);
1740 btrfs_set_root_drop_level(root_item, level);
1742 btrfs_end_transaction_throttle(trans);
1745 btrfs_btree_balance_dirty(fs_info);
1747 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1748 invalidate_extent_cache(root, &key, &next_key);
1752 * handle the case only one block in the fs tree need to be
1753 * relocated and the block is tree root.
1755 leaf = btrfs_lock_root_node(root);
1756 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1758 btrfs_tree_unlock(leaf);
1759 free_extent_buffer(leaf);
1761 btrfs_free_path(path);
1764 insert_dirty_subvol(trans, rc, root);
1767 btrfs_end_transaction_throttle(trans);
1769 btrfs_btree_balance_dirty(fs_info);
1771 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1772 invalidate_extent_cache(root, &key, &next_key);
1777 static noinline_for_stack
1778 int prepare_to_merge(struct reloc_control *rc, int err)
1780 struct btrfs_root *root = rc->extent_root;
1781 struct btrfs_fs_info *fs_info = root->fs_info;
1782 struct btrfs_root *reloc_root;
1783 struct btrfs_trans_handle *trans;
1784 LIST_HEAD(reloc_roots);
1788 mutex_lock(&fs_info->reloc_mutex);
1789 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1790 rc->merging_rsv_size += rc->nodes_relocated * 2;
1791 mutex_unlock(&fs_info->reloc_mutex);
1795 num_bytes = rc->merging_rsv_size;
1796 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1797 BTRFS_RESERVE_FLUSH_ALL);
1802 trans = btrfs_join_transaction(rc->extent_root);
1803 if (IS_ERR(trans)) {
1805 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1807 return PTR_ERR(trans);
1811 if (num_bytes != rc->merging_rsv_size) {
1812 btrfs_end_transaction(trans);
1813 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1819 rc->merge_reloc_tree = 1;
1821 while (!list_empty(&rc->reloc_roots)) {
1822 reloc_root = list_entry(rc->reloc_roots.next,
1823 struct btrfs_root, root_list);
1824 list_del_init(&reloc_root->root_list);
1826 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1828 BUG_ON(IS_ERR(root));
1829 BUG_ON(root->reloc_root != reloc_root);
1832 * set reference count to 1, so btrfs_recover_relocation
1833 * knows it should resumes merging
1836 btrfs_set_root_refs(&reloc_root->root_item, 1);
1837 btrfs_update_reloc_root(trans, root);
1839 list_add(&reloc_root->root_list, &reloc_roots);
1840 btrfs_put_root(root);
1843 list_splice(&reloc_roots, &rc->reloc_roots);
1846 btrfs_commit_transaction(trans);
1848 btrfs_end_transaction(trans);
1852 static noinline_for_stack
1853 void free_reloc_roots(struct list_head *list)
1855 struct btrfs_root *reloc_root, *tmp;
1857 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1858 __del_reloc_root(reloc_root);
1861 static noinline_for_stack
1862 void merge_reloc_roots(struct reloc_control *rc)
1864 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1865 struct btrfs_root *root;
1866 struct btrfs_root *reloc_root;
1867 LIST_HEAD(reloc_roots);
1871 root = rc->extent_root;
1874 * this serializes us with btrfs_record_root_in_transaction,
1875 * we have to make sure nobody is in the middle of
1876 * adding their roots to the list while we are
1879 mutex_lock(&fs_info->reloc_mutex);
1880 list_splice_init(&rc->reloc_roots, &reloc_roots);
1881 mutex_unlock(&fs_info->reloc_mutex);
1883 while (!list_empty(&reloc_roots)) {
1885 reloc_root = list_entry(reloc_roots.next,
1886 struct btrfs_root, root_list);
1888 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1890 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1891 BUG_ON(IS_ERR(root));
1892 BUG_ON(root->reloc_root != reloc_root);
1893 ret = merge_reloc_root(rc, root);
1894 btrfs_put_root(root);
1896 if (list_empty(&reloc_root->root_list))
1897 list_add_tail(&reloc_root->root_list,
1902 if (!IS_ERR(root)) {
1903 if (root->reloc_root == reloc_root) {
1904 root->reloc_root = NULL;
1905 btrfs_put_root(reloc_root);
1907 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1909 btrfs_put_root(root);
1912 list_del_init(&reloc_root->root_list);
1913 /* Don't forget to queue this reloc root for cleanup */
1914 list_add_tail(&reloc_root->reloc_dirty_list,
1915 &rc->dirty_subvol_roots);
1925 btrfs_handle_fs_error(fs_info, ret, NULL);
1926 free_reloc_roots(&reloc_roots);
1928 /* new reloc root may be added */
1929 mutex_lock(&fs_info->reloc_mutex);
1930 list_splice_init(&rc->reloc_roots, &reloc_roots);
1931 mutex_unlock(&fs_info->reloc_mutex);
1932 free_reloc_roots(&reloc_roots);
1938 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1940 * here, but it's wrong. If we fail to start the transaction in
1941 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1942 * have actually been removed from the reloc_root_tree rb tree. This is
1943 * fine because we're bailing here, and we hold a reference on the root
1944 * for the list that holds it, so these roots will be cleaned up when we
1945 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1946 * will be cleaned up on unmount.
1948 * The remaining nodes will be cleaned up by free_reloc_control.
1952 static void free_block_list(struct rb_root *blocks)
1954 struct tree_block *block;
1955 struct rb_node *rb_node;
1956 while ((rb_node = rb_first(blocks))) {
1957 block = rb_entry(rb_node, struct tree_block, rb_node);
1958 rb_erase(rb_node, blocks);
1963 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1964 struct btrfs_root *reloc_root)
1966 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
1967 struct btrfs_root *root;
1970 if (reloc_root->last_trans == trans->transid)
1973 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
1974 BUG_ON(IS_ERR(root));
1975 BUG_ON(root->reloc_root != reloc_root);
1976 ret = btrfs_record_root_in_trans(trans, root);
1977 btrfs_put_root(root);
1982 static noinline_for_stack
1983 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
1984 struct reloc_control *rc,
1985 struct btrfs_backref_node *node,
1986 struct btrfs_backref_edge *edges[])
1988 struct btrfs_backref_node *next;
1989 struct btrfs_root *root;
1995 next = walk_up_backref(next, edges, &index);
1998 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2000 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2001 record_reloc_root_in_trans(trans, root);
2005 btrfs_record_root_in_trans(trans, root);
2006 root = root->reloc_root;
2008 if (next->new_bytenr != root->node->start) {
2009 BUG_ON(next->new_bytenr);
2010 BUG_ON(!list_empty(&next->list));
2011 next->new_bytenr = root->node->start;
2012 btrfs_put_root(next->root);
2013 next->root = btrfs_grab_root(root);
2015 list_add_tail(&next->list,
2016 &rc->backref_cache.changed);
2017 mark_block_processed(rc, next);
2023 next = walk_down_backref(edges, &index);
2024 if (!next || next->level <= node->level)
2031 /* setup backref node path for btrfs_reloc_cow_block */
2033 rc->backref_cache.path[next->level] = next;
2036 next = edges[index]->node[UPPER];
2042 * Select a tree root for relocation.
2044 * Return NULL if the block is not shareable. We should use do_relocation() in
2047 * Return a tree root pointer if the block is shareable.
2048 * Return -ENOENT if the block is root of reloc tree.
2050 static noinline_for_stack
2051 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2053 struct btrfs_backref_node *next;
2054 struct btrfs_root *root;
2055 struct btrfs_root *fs_root = NULL;
2056 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2062 next = walk_up_backref(next, edges, &index);
2066 /* No other choice for non-shareable tree */
2067 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2070 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2076 next = walk_down_backref(edges, &index);
2077 if (!next || next->level <= node->level)
2082 return ERR_PTR(-ENOENT);
2086 static noinline_for_stack
2087 u64 calcu_metadata_size(struct reloc_control *rc,
2088 struct btrfs_backref_node *node, int reserve)
2090 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2091 struct btrfs_backref_node *next = node;
2092 struct btrfs_backref_edge *edge;
2093 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2097 BUG_ON(reserve && node->processed);
2102 if (next->processed && (reserve || next != node))
2105 num_bytes += fs_info->nodesize;
2107 if (list_empty(&next->upper))
2110 edge = list_entry(next->upper.next,
2111 struct btrfs_backref_edge, list[LOWER]);
2112 edges[index++] = edge;
2113 next = edge->node[UPPER];
2115 next = walk_down_backref(edges, &index);
2120 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2121 struct reloc_control *rc,
2122 struct btrfs_backref_node *node)
2124 struct btrfs_root *root = rc->extent_root;
2125 struct btrfs_fs_info *fs_info = root->fs_info;
2130 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2132 trans->block_rsv = rc->block_rsv;
2133 rc->reserved_bytes += num_bytes;
2136 * We are under a transaction here so we can only do limited flushing.
2137 * If we get an enospc just kick back -EAGAIN so we know to drop the
2138 * transaction and try to refill when we can flush all the things.
2140 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2141 BTRFS_RESERVE_FLUSH_LIMIT);
2143 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2144 while (tmp <= rc->reserved_bytes)
2147 * only one thread can access block_rsv at this point,
2148 * so we don't need hold lock to protect block_rsv.
2149 * we expand more reservation size here to allow enough
2150 * space for relocation and we will return earlier in
2153 rc->block_rsv->size = tmp + fs_info->nodesize *
2154 RELOCATION_RESERVED_NODES;
2162 * relocate a block tree, and then update pointers in upper level
2163 * blocks that reference the block to point to the new location.
2165 * if called by link_to_upper, the block has already been relocated.
2166 * in that case this function just updates pointers.
2168 static int do_relocation(struct btrfs_trans_handle *trans,
2169 struct reloc_control *rc,
2170 struct btrfs_backref_node *node,
2171 struct btrfs_key *key,
2172 struct btrfs_path *path, int lowest)
2174 struct btrfs_backref_node *upper;
2175 struct btrfs_backref_edge *edge;
2176 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2177 struct btrfs_root *root;
2178 struct extent_buffer *eb;
2184 BUG_ON(lowest && node->eb);
2186 path->lowest_level = node->level + 1;
2187 rc->backref_cache.path[node->level] = node;
2188 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2189 struct btrfs_ref ref = { 0 };
2193 upper = edge->node[UPPER];
2194 root = select_reloc_root(trans, rc, upper, edges);
2197 if (upper->eb && !upper->locked) {
2199 ret = btrfs_bin_search(upper->eb, key, &slot);
2203 bytenr = btrfs_node_blockptr(upper->eb, slot);
2204 if (node->eb->start == bytenr)
2207 btrfs_backref_drop_node_buffer(upper);
2211 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2216 btrfs_release_path(path);
2221 upper->eb = path->nodes[upper->level];
2222 path->nodes[upper->level] = NULL;
2224 BUG_ON(upper->eb != path->nodes[upper->level]);
2228 path->locks[upper->level] = 0;
2230 slot = path->slots[upper->level];
2231 btrfs_release_path(path);
2233 ret = btrfs_bin_search(upper->eb, key, &slot);
2239 bytenr = btrfs_node_blockptr(upper->eb, slot);
2241 if (bytenr != node->bytenr) {
2242 btrfs_err(root->fs_info,
2243 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2244 bytenr, node->bytenr, slot,
2250 if (node->eb->start == bytenr)
2254 blocksize = root->fs_info->nodesize;
2255 eb = btrfs_read_node_slot(upper->eb, slot);
2260 btrfs_tree_lock(eb);
2263 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2264 slot, &eb, BTRFS_NESTING_COW);
2265 btrfs_tree_unlock(eb);
2266 free_extent_buffer(eb);
2269 BUG_ON(node->eb != eb);
2271 btrfs_set_node_blockptr(upper->eb, slot,
2273 btrfs_set_node_ptr_generation(upper->eb, slot,
2275 btrfs_mark_buffer_dirty(upper->eb);
2277 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2278 node->eb->start, blocksize,
2280 ref.real_root = root->root_key.objectid;
2281 btrfs_init_tree_ref(&ref, node->level,
2282 btrfs_header_owner(upper->eb));
2283 ret = btrfs_inc_extent_ref(trans, &ref);
2286 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2290 if (!upper->pending)
2291 btrfs_backref_drop_node_buffer(upper);
2293 btrfs_backref_unlock_node_buffer(upper);
2298 if (!ret && node->pending) {
2299 btrfs_backref_drop_node_buffer(node);
2300 list_move_tail(&node->list, &rc->backref_cache.changed);
2304 path->lowest_level = 0;
2305 BUG_ON(ret == -ENOSPC);
2309 static int link_to_upper(struct btrfs_trans_handle *trans,
2310 struct reloc_control *rc,
2311 struct btrfs_backref_node *node,
2312 struct btrfs_path *path)
2314 struct btrfs_key key;
2316 btrfs_node_key_to_cpu(node->eb, &key, 0);
2317 return do_relocation(trans, rc, node, &key, path, 0);
2320 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2321 struct reloc_control *rc,
2322 struct btrfs_path *path, int err)
2325 struct btrfs_backref_cache *cache = &rc->backref_cache;
2326 struct btrfs_backref_node *node;
2330 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2331 while (!list_empty(&cache->pending[level])) {
2332 node = list_entry(cache->pending[level].next,
2333 struct btrfs_backref_node, list);
2334 list_move_tail(&node->list, &list);
2335 BUG_ON(!node->pending);
2338 ret = link_to_upper(trans, rc, node, path);
2343 list_splice_init(&list, &cache->pending[level]);
2349 * mark a block and all blocks directly/indirectly reference the block
2352 static void update_processed_blocks(struct reloc_control *rc,
2353 struct btrfs_backref_node *node)
2355 struct btrfs_backref_node *next = node;
2356 struct btrfs_backref_edge *edge;
2357 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2363 if (next->processed)
2366 mark_block_processed(rc, next);
2368 if (list_empty(&next->upper))
2371 edge = list_entry(next->upper.next,
2372 struct btrfs_backref_edge, list[LOWER]);
2373 edges[index++] = edge;
2374 next = edge->node[UPPER];
2376 next = walk_down_backref(edges, &index);
2380 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2382 u32 blocksize = rc->extent_root->fs_info->nodesize;
2384 if (test_range_bit(&rc->processed_blocks, bytenr,
2385 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2390 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2391 struct tree_block *block)
2393 struct extent_buffer *eb;
2395 eb = read_tree_block(fs_info, block->bytenr, block->owner,
2396 block->key.offset, block->level, NULL);
2399 } else if (!extent_buffer_uptodate(eb)) {
2400 free_extent_buffer(eb);
2403 if (block->level == 0)
2404 btrfs_item_key_to_cpu(eb, &block->key, 0);
2406 btrfs_node_key_to_cpu(eb, &block->key, 0);
2407 free_extent_buffer(eb);
2408 block->key_ready = 1;
2413 * helper function to relocate a tree block
2415 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2416 struct reloc_control *rc,
2417 struct btrfs_backref_node *node,
2418 struct btrfs_key *key,
2419 struct btrfs_path *path)
2421 struct btrfs_root *root;
2428 * If we fail here we want to drop our backref_node because we are going
2429 * to start over and regenerate the tree for it.
2431 ret = reserve_metadata_space(trans, rc, node);
2435 BUG_ON(node->processed);
2436 root = select_one_root(node);
2437 if (root == ERR_PTR(-ENOENT)) {
2438 update_processed_blocks(rc, node);
2443 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2444 BUG_ON(node->new_bytenr);
2445 BUG_ON(!list_empty(&node->list));
2446 btrfs_record_root_in_trans(trans, root);
2447 root = root->reloc_root;
2448 node->new_bytenr = root->node->start;
2449 btrfs_put_root(node->root);
2450 node->root = btrfs_grab_root(root);
2452 list_add_tail(&node->list, &rc->backref_cache.changed);
2454 path->lowest_level = node->level;
2455 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2456 btrfs_release_path(path);
2461 update_processed_blocks(rc, node);
2463 ret = do_relocation(trans, rc, node, key, path, 1);
2466 if (ret || node->level == 0 || node->cowonly)
2467 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2472 * relocate a list of blocks
2474 static noinline_for_stack
2475 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2476 struct reloc_control *rc, struct rb_root *blocks)
2478 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2479 struct btrfs_backref_node *node;
2480 struct btrfs_path *path;
2481 struct tree_block *block;
2482 struct tree_block *next;
2486 path = btrfs_alloc_path();
2489 goto out_free_blocks;
2492 /* Kick in readahead for tree blocks with missing keys */
2493 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2494 if (!block->key_ready)
2495 btrfs_readahead_tree_block(fs_info, block->bytenr,
2500 /* Get first keys */
2501 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2502 if (!block->key_ready) {
2503 err = get_tree_block_key(fs_info, block);
2509 /* Do tree relocation */
2510 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2511 node = build_backref_tree(rc, &block->key,
2512 block->level, block->bytenr);
2514 err = PTR_ERR(node);
2518 ret = relocate_tree_block(trans, rc, node, &block->key,
2526 err = finish_pending_nodes(trans, rc, path, err);
2529 btrfs_free_path(path);
2531 free_block_list(blocks);
2535 static noinline_for_stack int prealloc_file_extent_cluster(
2536 struct btrfs_inode *inode,
2537 struct file_extent_cluster *cluster)
2542 u64 offset = inode->index_cnt;
2546 u64 prealloc_start = cluster->start - offset;
2547 u64 prealloc_end = cluster->end - offset;
2548 u64 cur_offset = prealloc_start;
2550 BUG_ON(cluster->start != cluster->boundary[0]);
2551 ret = btrfs_alloc_data_chunk_ondemand(inode,
2552 prealloc_end + 1 - prealloc_start);
2557 * On a zoned filesystem, we cannot preallocate the file region.
2558 * Instead, we dirty and fiemap_write the region.
2560 if (btrfs_is_zoned(inode->root->fs_info)) {
2561 struct btrfs_root *root = inode->root;
2562 struct btrfs_trans_handle *trans;
2564 end = cluster->end - offset + 1;
2565 trans = btrfs_start_transaction(root, 1);
2567 return PTR_ERR(trans);
2569 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
2570 i_size_write(&inode->vfs_inode, end);
2571 ret = btrfs_update_inode(trans, root, inode);
2573 btrfs_abort_transaction(trans, ret);
2574 btrfs_end_transaction(trans);
2578 return btrfs_end_transaction(trans);
2581 inode_lock(&inode->vfs_inode);
2582 for (nr = 0; nr < cluster->nr; nr++) {
2583 start = cluster->boundary[nr] - offset;
2584 if (nr + 1 < cluster->nr)
2585 end = cluster->boundary[nr + 1] - 1 - offset;
2587 end = cluster->end - offset;
2589 lock_extent(&inode->io_tree, start, end);
2590 num_bytes = end + 1 - start;
2591 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2592 num_bytes, num_bytes,
2593 end + 1, &alloc_hint);
2594 cur_offset = end + 1;
2595 unlock_extent(&inode->io_tree, start, end);
2599 inode_unlock(&inode->vfs_inode);
2601 if (cur_offset < prealloc_end)
2602 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2603 prealloc_end + 1 - cur_offset);
2607 static noinline_for_stack
2608 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2611 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2612 struct extent_map *em;
2615 em = alloc_extent_map();
2620 em->len = end + 1 - start;
2621 em->block_len = em->len;
2622 em->block_start = block_start;
2623 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2625 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2627 write_lock(&em_tree->lock);
2628 ret = add_extent_mapping(em_tree, em, 0);
2629 write_unlock(&em_tree->lock);
2630 if (ret != -EEXIST) {
2631 free_extent_map(em);
2634 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2636 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2641 * Allow error injection to test balance cancellation
2643 noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2645 return atomic_read(&fs_info->balance_cancel_req) ||
2646 fatal_signal_pending(current);
2648 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2650 static int relocate_file_extent_cluster(struct inode *inode,
2651 struct file_extent_cluster *cluster)
2653 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2656 u64 offset = BTRFS_I(inode)->index_cnt;
2657 unsigned long index;
2658 unsigned long last_index;
2660 struct file_ra_state *ra;
2661 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2668 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2672 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2676 file_ra_state_init(ra, inode->i_mapping);
2678 ret = setup_extent_mapping(inode, cluster->start - offset,
2679 cluster->end - offset, cluster->start);
2683 index = (cluster->start - offset) >> PAGE_SHIFT;
2684 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2685 while (index <= last_index) {
2686 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2691 page = find_lock_page(inode->i_mapping, index);
2693 page_cache_sync_readahead(inode->i_mapping,
2695 last_index + 1 - index);
2696 page = find_or_create_page(inode->i_mapping, index,
2699 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2701 btrfs_delalloc_release_extents(BTRFS_I(inode),
2707 ret = set_page_extent_mapped(page);
2709 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2711 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2717 if (PageReadahead(page)) {
2718 page_cache_async_readahead(inode->i_mapping,
2719 ra, NULL, page, index,
2720 last_index + 1 - index);
2723 if (!PageUptodate(page)) {
2724 btrfs_readpage(NULL, page);
2726 if (!PageUptodate(page)) {
2729 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2731 btrfs_delalloc_release_extents(BTRFS_I(inode),
2738 page_start = page_offset(page);
2739 page_end = page_start + PAGE_SIZE - 1;
2741 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2743 if (nr < cluster->nr &&
2744 page_start + offset == cluster->boundary[nr]) {
2745 set_extent_bits(&BTRFS_I(inode)->io_tree,
2746 page_start, page_end,
2751 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2756 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2758 btrfs_delalloc_release_extents(BTRFS_I(inode),
2761 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2762 page_start, page_end,
2763 EXTENT_LOCKED | EXTENT_BOUNDARY);
2767 set_page_dirty(page);
2769 unlock_extent(&BTRFS_I(inode)->io_tree,
2770 page_start, page_end);
2775 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2776 balance_dirty_pages_ratelimited(inode->i_mapping);
2777 btrfs_throttle(fs_info);
2778 if (btrfs_should_cancel_balance(fs_info)) {
2783 WARN_ON(nr != cluster->nr);
2784 if (btrfs_is_zoned(fs_info) && !ret)
2785 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
2791 static noinline_for_stack
2792 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2793 struct file_extent_cluster *cluster)
2797 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2798 ret = relocate_file_extent_cluster(inode, cluster);
2805 cluster->start = extent_key->objectid;
2807 BUG_ON(cluster->nr >= MAX_EXTENTS);
2808 cluster->end = extent_key->objectid + extent_key->offset - 1;
2809 cluster->boundary[cluster->nr] = extent_key->objectid;
2812 if (cluster->nr >= MAX_EXTENTS) {
2813 ret = relocate_file_extent_cluster(inode, cluster);
2822 * helper to add a tree block to the list.
2823 * the major work is getting the generation and level of the block
2825 static int add_tree_block(struct reloc_control *rc,
2826 struct btrfs_key *extent_key,
2827 struct btrfs_path *path,
2828 struct rb_root *blocks)
2830 struct extent_buffer *eb;
2831 struct btrfs_extent_item *ei;
2832 struct btrfs_tree_block_info *bi;
2833 struct tree_block *block;
2834 struct rb_node *rb_node;
2840 eb = path->nodes[0];
2841 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2843 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2844 item_size >= sizeof(*ei) + sizeof(*bi)) {
2845 unsigned long ptr = 0, end;
2847 ei = btrfs_item_ptr(eb, path->slots[0],
2848 struct btrfs_extent_item);
2849 end = (unsigned long)ei + item_size;
2850 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2851 bi = (struct btrfs_tree_block_info *)(ei + 1);
2852 level = btrfs_tree_block_level(eb, bi);
2853 ptr = (unsigned long)(bi + 1);
2855 level = (int)extent_key->offset;
2856 ptr = (unsigned long)(ei + 1);
2858 generation = btrfs_extent_generation(eb, ei);
2861 * We're reading random blocks without knowing their owner ahead
2862 * of time. This is ok most of the time, as all reloc roots and
2863 * fs roots have the same lock type. However normal trees do
2864 * not, and the only way to know ahead of time is to read the
2865 * inline ref offset. We know it's an fs root if
2867 * 1. There's more than one ref.
2868 * 2. There's a SHARED_DATA_REF_KEY set.
2869 * 3. FULL_BACKREF is set on the flags.
2871 * Otherwise it's safe to assume that the ref offset == the
2872 * owner of this block, so we can use that when calling
2875 if (btrfs_extent_refs(eb, ei) == 1 &&
2876 !(btrfs_extent_flags(eb, ei) &
2877 BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
2879 struct btrfs_extent_inline_ref *iref;
2882 iref = (struct btrfs_extent_inline_ref *)ptr;
2883 type = btrfs_get_extent_inline_ref_type(eb, iref,
2884 BTRFS_REF_TYPE_BLOCK);
2885 if (type == BTRFS_REF_TYPE_INVALID)
2887 if (type == BTRFS_TREE_BLOCK_REF_KEY)
2888 owner = btrfs_extent_inline_ref_offset(eb, iref);
2890 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2891 btrfs_print_v0_err(eb->fs_info);
2892 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2898 btrfs_release_path(path);
2900 BUG_ON(level == -1);
2902 block = kmalloc(sizeof(*block), GFP_NOFS);
2906 block->bytenr = extent_key->objectid;
2907 block->key.objectid = rc->extent_root->fs_info->nodesize;
2908 block->key.offset = generation;
2909 block->level = level;
2910 block->key_ready = 0;
2911 block->owner = owner;
2913 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2915 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2922 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2924 static int __add_tree_block(struct reloc_control *rc,
2925 u64 bytenr, u32 blocksize,
2926 struct rb_root *blocks)
2928 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2929 struct btrfs_path *path;
2930 struct btrfs_key key;
2932 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2934 if (tree_block_processed(bytenr, rc))
2937 if (rb_simple_search(blocks, bytenr))
2940 path = btrfs_alloc_path();
2944 key.objectid = bytenr;
2946 key.type = BTRFS_METADATA_ITEM_KEY;
2947 key.offset = (u64)-1;
2949 key.type = BTRFS_EXTENT_ITEM_KEY;
2950 key.offset = blocksize;
2953 path->search_commit_root = 1;
2954 path->skip_locking = 1;
2955 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2959 if (ret > 0 && skinny) {
2960 if (path->slots[0]) {
2962 btrfs_item_key_to_cpu(path->nodes[0], &key,
2964 if (key.objectid == bytenr &&
2965 (key.type == BTRFS_METADATA_ITEM_KEY ||
2966 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2967 key.offset == blocksize)))
2973 btrfs_release_path(path);
2979 btrfs_print_leaf(path->nodes[0]);
2981 "tree block extent item (%llu) is not found in extent tree",
2988 ret = add_tree_block(rc, &key, path, blocks);
2990 btrfs_free_path(path);
2994 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2995 struct btrfs_block_group *block_group,
2996 struct inode *inode,
2999 struct btrfs_root *root = fs_info->tree_root;
3000 struct btrfs_trans_handle *trans;
3006 inode = btrfs_iget(fs_info->sb, ino, root);
3011 ret = btrfs_check_trunc_cache_free_space(fs_info,
3012 &fs_info->global_block_rsv);
3016 trans = btrfs_join_transaction(root);
3017 if (IS_ERR(trans)) {
3018 ret = PTR_ERR(trans);
3022 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3024 btrfs_end_transaction(trans);
3025 btrfs_btree_balance_dirty(fs_info);
3032 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3033 * cache inode, to avoid free space cache data extent blocking data relocation.
3035 static int delete_v1_space_cache(struct extent_buffer *leaf,
3036 struct btrfs_block_group *block_group,
3039 u64 space_cache_ino;
3040 struct btrfs_file_extent_item *ei;
3041 struct btrfs_key key;
3046 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3049 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3052 btrfs_item_key_to_cpu(leaf, &key, i);
3053 if (key.type != BTRFS_EXTENT_DATA_KEY)
3055 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3056 type = btrfs_file_extent_type(leaf, ei);
3058 if ((type == BTRFS_FILE_EXTENT_REG ||
3059 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3060 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3062 space_cache_ino = key.objectid;
3068 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3074 * helper to find all tree blocks that reference a given data extent
3076 static noinline_for_stack
3077 int add_data_references(struct reloc_control *rc,
3078 struct btrfs_key *extent_key,
3079 struct btrfs_path *path,
3080 struct rb_root *blocks)
3082 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3083 struct ulist *leaves = NULL;
3084 struct ulist_iterator leaf_uiter;
3085 struct ulist_node *ref_node = NULL;
3086 const u32 blocksize = fs_info->nodesize;
3089 btrfs_release_path(path);
3090 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3091 0, &leaves, NULL, true);
3095 ULIST_ITER_INIT(&leaf_uiter);
3096 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3097 struct extent_buffer *eb;
3099 eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL);
3104 ret = delete_v1_space_cache(eb, rc->block_group,
3105 extent_key->objectid);
3106 free_extent_buffer(eb);
3109 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3114 free_block_list(blocks);
3120 * helper to find next unprocessed extent
3122 static noinline_for_stack
3123 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3124 struct btrfs_key *extent_key)
3126 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3127 struct btrfs_key key;
3128 struct extent_buffer *leaf;
3129 u64 start, end, last;
3132 last = rc->block_group->start + rc->block_group->length;
3135 if (rc->search_start >= last) {
3140 key.objectid = rc->search_start;
3141 key.type = BTRFS_EXTENT_ITEM_KEY;
3144 path->search_commit_root = 1;
3145 path->skip_locking = 1;
3146 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3151 leaf = path->nodes[0];
3152 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3153 ret = btrfs_next_leaf(rc->extent_root, path);
3156 leaf = path->nodes[0];
3159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3160 if (key.objectid >= last) {
3165 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3166 key.type != BTRFS_METADATA_ITEM_KEY) {
3171 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3172 key.objectid + key.offset <= rc->search_start) {
3177 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3178 key.objectid + fs_info->nodesize <=
3184 ret = find_first_extent_bit(&rc->processed_blocks,
3185 key.objectid, &start, &end,
3186 EXTENT_DIRTY, NULL);
3188 if (ret == 0 && start <= key.objectid) {
3189 btrfs_release_path(path);
3190 rc->search_start = end + 1;
3192 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3193 rc->search_start = key.objectid + key.offset;
3195 rc->search_start = key.objectid +
3197 memcpy(extent_key, &key, sizeof(key));
3201 btrfs_release_path(path);
3205 static void set_reloc_control(struct reloc_control *rc)
3207 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3209 mutex_lock(&fs_info->reloc_mutex);
3210 fs_info->reloc_ctl = rc;
3211 mutex_unlock(&fs_info->reloc_mutex);
3214 static void unset_reloc_control(struct reloc_control *rc)
3216 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3218 mutex_lock(&fs_info->reloc_mutex);
3219 fs_info->reloc_ctl = NULL;
3220 mutex_unlock(&fs_info->reloc_mutex);
3223 static int check_extent_flags(u64 flags)
3225 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3226 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3228 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3229 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3231 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3232 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3237 static noinline_for_stack
3238 int prepare_to_relocate(struct reloc_control *rc)
3240 struct btrfs_trans_handle *trans;
3243 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3244 BTRFS_BLOCK_RSV_TEMP);
3248 memset(&rc->cluster, 0, sizeof(rc->cluster));
3249 rc->search_start = rc->block_group->start;
3250 rc->extents_found = 0;
3251 rc->nodes_relocated = 0;
3252 rc->merging_rsv_size = 0;
3253 rc->reserved_bytes = 0;
3254 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3255 RELOCATION_RESERVED_NODES;
3256 ret = btrfs_block_rsv_refill(rc->extent_root,
3257 rc->block_rsv, rc->block_rsv->size,
3258 BTRFS_RESERVE_FLUSH_ALL);
3262 rc->create_reloc_tree = 1;
3263 set_reloc_control(rc);
3265 trans = btrfs_join_transaction(rc->extent_root);
3266 if (IS_ERR(trans)) {
3267 unset_reloc_control(rc);
3269 * extent tree is not a ref_cow tree and has no reloc_root to
3270 * cleanup. And callers are responsible to free the above
3273 return PTR_ERR(trans);
3275 btrfs_commit_transaction(trans);
3279 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3281 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3282 struct rb_root blocks = RB_ROOT;
3283 struct btrfs_key key;
3284 struct btrfs_trans_handle *trans = NULL;
3285 struct btrfs_path *path;
3286 struct btrfs_extent_item *ei;
3293 path = btrfs_alloc_path();
3296 path->reada = READA_FORWARD;
3298 ret = prepare_to_relocate(rc);
3305 rc->reserved_bytes = 0;
3306 ret = btrfs_block_rsv_refill(rc->extent_root,
3307 rc->block_rsv, rc->block_rsv->size,
3308 BTRFS_RESERVE_FLUSH_ALL);
3314 trans = btrfs_start_transaction(rc->extent_root, 0);
3315 if (IS_ERR(trans)) {
3316 err = PTR_ERR(trans);
3321 if (update_backref_cache(trans, &rc->backref_cache)) {
3322 btrfs_end_transaction(trans);
3327 ret = find_next_extent(rc, path, &key);
3333 rc->extents_found++;
3335 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3336 struct btrfs_extent_item);
3337 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3338 if (item_size >= sizeof(*ei)) {
3339 flags = btrfs_extent_flags(path->nodes[0], ei);
3340 ret = check_extent_flags(flags);
3342 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3344 btrfs_print_v0_err(trans->fs_info);
3345 btrfs_abort_transaction(trans, err);
3351 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3352 ret = add_tree_block(rc, &key, path, &blocks);
3353 } else if (rc->stage == UPDATE_DATA_PTRS &&
3354 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3355 ret = add_data_references(rc, &key, path, &blocks);
3357 btrfs_release_path(path);
3365 if (!RB_EMPTY_ROOT(&blocks)) {
3366 ret = relocate_tree_blocks(trans, rc, &blocks);
3368 if (ret != -EAGAIN) {
3372 rc->extents_found--;
3373 rc->search_start = key.objectid;
3377 btrfs_end_transaction_throttle(trans);
3378 btrfs_btree_balance_dirty(fs_info);
3381 if (rc->stage == MOVE_DATA_EXTENTS &&
3382 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3383 rc->found_file_extent = 1;
3384 ret = relocate_data_extent(rc->data_inode,
3385 &key, &rc->cluster);
3391 if (btrfs_should_cancel_balance(fs_info)) {
3396 if (trans && progress && err == -ENOSPC) {
3397 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3405 btrfs_release_path(path);
3406 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3409 btrfs_end_transaction_throttle(trans);
3410 btrfs_btree_balance_dirty(fs_info);
3414 ret = relocate_file_extent_cluster(rc->data_inode,
3420 rc->create_reloc_tree = 0;
3421 set_reloc_control(rc);
3423 btrfs_backref_release_cache(&rc->backref_cache);
3424 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3427 * Even in the case when the relocation is cancelled, we should all go
3428 * through prepare_to_merge() and merge_reloc_roots().
3430 * For error (including cancelled balance), prepare_to_merge() will
3431 * mark all reloc trees orphan, then queue them for cleanup in
3432 * merge_reloc_roots()
3434 err = prepare_to_merge(rc, err);
3436 merge_reloc_roots(rc);
3438 rc->merge_reloc_tree = 0;
3439 unset_reloc_control(rc);
3440 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3442 /* get rid of pinned extents */
3443 trans = btrfs_join_transaction(rc->extent_root);
3444 if (IS_ERR(trans)) {
3445 err = PTR_ERR(trans);
3448 btrfs_commit_transaction(trans);
3450 ret = clean_dirty_subvols(rc);
3451 if (ret < 0 && !err)
3453 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3454 btrfs_free_path(path);
3458 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3459 struct btrfs_root *root, u64 objectid)
3461 struct btrfs_path *path;
3462 struct btrfs_inode_item *item;
3463 struct extent_buffer *leaf;
3464 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
3467 if (btrfs_is_zoned(trans->fs_info))
3468 flags &= ~BTRFS_INODE_PREALLOC;
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, flags);
3485 btrfs_mark_buffer_dirty(leaf);
3487 btrfs_free_path(path);
3492 * helper to create inode for data relocation.
3493 * the inode is in data relocation tree and its link count is 0
3495 static noinline_for_stack
3496 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3497 struct btrfs_block_group *group)
3499 struct inode *inode = NULL;
3500 struct btrfs_trans_handle *trans;
3501 struct btrfs_root *root;
3505 root = btrfs_grab_root(fs_info->data_reloc_root);
3506 trans = btrfs_start_transaction(root, 6);
3507 if (IS_ERR(trans)) {
3508 btrfs_put_root(root);
3509 return ERR_CAST(trans);
3512 err = btrfs_get_free_objectid(root, &objectid);
3516 err = __insert_orphan_inode(trans, root, objectid);
3519 inode = btrfs_iget(fs_info->sb, objectid, root);
3520 BUG_ON(IS_ERR(inode));
3521 BTRFS_I(inode)->index_cnt = group->start;
3523 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3525 btrfs_put_root(root);
3526 btrfs_end_transaction(trans);
3527 btrfs_btree_balance_dirty(fs_info);
3531 inode = ERR_PTR(err);
3536 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3538 struct reloc_control *rc;
3540 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3544 INIT_LIST_HEAD(&rc->reloc_roots);
3545 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3546 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3547 mapping_tree_init(&rc->reloc_root_tree);
3548 extent_io_tree_init(fs_info, &rc->processed_blocks,
3549 IO_TREE_RELOC_BLOCKS, NULL);
3553 static void free_reloc_control(struct reloc_control *rc)
3555 struct mapping_node *node, *tmp;
3557 free_reloc_roots(&rc->reloc_roots);
3558 rbtree_postorder_for_each_entry_safe(node, tmp,
3559 &rc->reloc_root_tree.rb_root, rb_node)
3566 * Print the block group being relocated
3568 static void describe_relocation(struct btrfs_fs_info *fs_info,
3569 struct btrfs_block_group *block_group)
3571 char buf[128] = {'\0'};
3573 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3576 "relocating block group %llu flags %s",
3577 block_group->start, buf);
3580 static const char *stage_to_string(int stage)
3582 if (stage == MOVE_DATA_EXTENTS)
3583 return "move data extents";
3584 if (stage == UPDATE_DATA_PTRS)
3585 return "update data pointers";
3590 * function to relocate all extents in a block group.
3592 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3594 struct btrfs_block_group *bg;
3595 struct btrfs_root *extent_root = fs_info->extent_root;
3596 struct reloc_control *rc;
3597 struct inode *inode;
3598 struct btrfs_path *path;
3603 bg = btrfs_lookup_block_group(fs_info, group_start);
3607 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3608 btrfs_put_block_group(bg);
3612 rc = alloc_reloc_control(fs_info);
3614 btrfs_put_block_group(bg);
3618 rc->extent_root = extent_root;
3619 rc->block_group = bg;
3621 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3628 path = btrfs_alloc_path();
3634 inode = lookup_free_space_inode(rc->block_group, path);
3635 btrfs_free_path(path);
3638 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3640 ret = PTR_ERR(inode);
3642 if (ret && ret != -ENOENT) {
3647 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3648 if (IS_ERR(rc->data_inode)) {
3649 err = PTR_ERR(rc->data_inode);
3650 rc->data_inode = NULL;
3654 describe_relocation(fs_info, rc->block_group);
3656 btrfs_wait_block_group_reservations(rc->block_group);
3657 btrfs_wait_nocow_writers(rc->block_group);
3658 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3659 rc->block_group->start,
3660 rc->block_group->length);
3665 mutex_lock(&fs_info->cleaner_mutex);
3666 ret = relocate_block_group(rc);
3667 mutex_unlock(&fs_info->cleaner_mutex);
3671 finishes_stage = rc->stage;
3673 * We may have gotten ENOSPC after we already dirtied some
3674 * extents. If writeout happens while we're relocating a
3675 * different block group we could end up hitting the
3676 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3677 * btrfs_reloc_cow_block. Make sure we write everything out
3678 * properly so we don't trip over this problem, and then break
3679 * out of the loop if we hit an error.
3681 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3682 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3686 invalidate_mapping_pages(rc->data_inode->i_mapping,
3688 rc->stage = UPDATE_DATA_PTRS;
3694 if (rc->extents_found == 0)
3697 btrfs_info(fs_info, "found %llu extents, stage: %s",
3698 rc->extents_found, stage_to_string(finishes_stage));
3701 WARN_ON(rc->block_group->pinned > 0);
3702 WARN_ON(rc->block_group->reserved > 0);
3703 WARN_ON(rc->block_group->used > 0);
3706 btrfs_dec_block_group_ro(rc->block_group);
3707 iput(rc->data_inode);
3708 btrfs_put_block_group(rc->block_group);
3709 free_reloc_control(rc);
3713 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3715 struct btrfs_fs_info *fs_info = root->fs_info;
3716 struct btrfs_trans_handle *trans;
3719 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3721 return PTR_ERR(trans);
3723 memset(&root->root_item.drop_progress, 0,
3724 sizeof(root->root_item.drop_progress));
3725 btrfs_set_root_drop_level(&root->root_item, 0);
3726 btrfs_set_root_refs(&root->root_item, 0);
3727 ret = btrfs_update_root(trans, fs_info->tree_root,
3728 &root->root_key, &root->root_item);
3730 err = btrfs_end_transaction(trans);
3737 * recover relocation interrupted by system crash.
3739 * this function resumes merging reloc trees with corresponding fs trees.
3740 * this is important for keeping the sharing of tree blocks
3742 int btrfs_recover_relocation(struct btrfs_root *root)
3744 struct btrfs_fs_info *fs_info = root->fs_info;
3745 LIST_HEAD(reloc_roots);
3746 struct btrfs_key key;
3747 struct btrfs_root *fs_root;
3748 struct btrfs_root *reloc_root;
3749 struct btrfs_path *path;
3750 struct extent_buffer *leaf;
3751 struct reloc_control *rc = NULL;
3752 struct btrfs_trans_handle *trans;
3756 path = btrfs_alloc_path();
3759 path->reada = READA_BACK;
3761 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3762 key.type = BTRFS_ROOT_ITEM_KEY;
3763 key.offset = (u64)-1;
3766 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3773 if (path->slots[0] == 0)
3777 leaf = path->nodes[0];
3778 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3779 btrfs_release_path(path);
3781 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3782 key.type != BTRFS_ROOT_ITEM_KEY)
3785 reloc_root = btrfs_read_tree_root(root, &key);
3786 if (IS_ERR(reloc_root)) {
3787 err = PTR_ERR(reloc_root);
3791 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3792 list_add(&reloc_root->root_list, &reloc_roots);
3794 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3795 fs_root = btrfs_get_fs_root(fs_info,
3796 reloc_root->root_key.offset, false);
3797 if (IS_ERR(fs_root)) {
3798 ret = PTR_ERR(fs_root);
3799 if (ret != -ENOENT) {
3803 ret = mark_garbage_root(reloc_root);
3809 btrfs_put_root(fs_root);
3813 if (key.offset == 0)
3818 btrfs_release_path(path);
3820 if (list_empty(&reloc_roots))
3823 rc = alloc_reloc_control(fs_info);
3829 rc->extent_root = fs_info->extent_root;
3831 set_reloc_control(rc);
3833 trans = btrfs_join_transaction(rc->extent_root);
3834 if (IS_ERR(trans)) {
3835 err = PTR_ERR(trans);
3839 rc->merge_reloc_tree = 1;
3841 while (!list_empty(&reloc_roots)) {
3842 reloc_root = list_entry(reloc_roots.next,
3843 struct btrfs_root, root_list);
3844 list_del(&reloc_root->root_list);
3846 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3847 list_add_tail(&reloc_root->root_list,
3852 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
3854 if (IS_ERR(fs_root)) {
3855 err = PTR_ERR(fs_root);
3856 list_add_tail(&reloc_root->root_list, &reloc_roots);
3857 btrfs_end_transaction(trans);
3861 err = __add_reloc_root(reloc_root);
3862 BUG_ON(err < 0); /* -ENOMEM or logic error */
3863 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3864 btrfs_put_root(fs_root);
3867 err = btrfs_commit_transaction(trans);
3871 merge_reloc_roots(rc);
3873 unset_reloc_control(rc);
3875 trans = btrfs_join_transaction(rc->extent_root);
3876 if (IS_ERR(trans)) {
3877 err = PTR_ERR(trans);
3880 err = btrfs_commit_transaction(trans);
3882 ret = clean_dirty_subvols(rc);
3883 if (ret < 0 && !err)
3886 unset_reloc_control(rc);
3887 free_reloc_control(rc);
3889 free_reloc_roots(&reloc_roots);
3891 btrfs_free_path(path);
3894 /* cleanup orphan inode in data relocation tree */
3895 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
3897 err = btrfs_orphan_cleanup(fs_root);
3898 btrfs_put_root(fs_root);
3904 * helper to add ordered checksum for data relocation.
3906 * cloning checksum properly handles the nodatasum extents.
3907 * it also saves CPU time to re-calculate the checksum.
3909 int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
3911 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3912 struct btrfs_ordered_sum *sums;
3913 struct btrfs_ordered_extent *ordered;
3919 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3920 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3922 disk_bytenr = file_pos + inode->index_cnt;
3923 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3924 disk_bytenr + len - 1, &list, 0);
3928 while (!list_empty(&list)) {
3929 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3930 list_del_init(&sums->list);
3933 * We need to offset the new_bytenr based on where the csum is.
3934 * We need to do this because we will read in entire prealloc
3935 * extents but we may have written to say the middle of the
3936 * prealloc extent, so we need to make sure the csum goes with
3937 * the right disk offset.
3939 * We can do this because the data reloc inode refers strictly
3940 * to the on disk bytes, so we don't have to worry about
3941 * disk_len vs real len like with real inodes since it's all
3944 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3945 sums->bytenr = new_bytenr;
3947 btrfs_add_ordered_sum(ordered, sums);
3950 btrfs_put_ordered_extent(ordered);
3954 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3955 struct btrfs_root *root, struct extent_buffer *buf,
3956 struct extent_buffer *cow)
3958 struct btrfs_fs_info *fs_info = root->fs_info;
3959 struct reloc_control *rc;
3960 struct btrfs_backref_node *node;
3965 rc = fs_info->reloc_ctl;
3969 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3970 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3972 level = btrfs_header_level(buf);
3973 if (btrfs_header_generation(buf) <=
3974 btrfs_root_last_snapshot(&root->root_item))
3977 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3978 rc->create_reloc_tree) {
3979 WARN_ON(!first_cow && level == 0);
3981 node = rc->backref_cache.path[level];
3982 BUG_ON(node->bytenr != buf->start &&
3983 node->new_bytenr != buf->start);
3985 btrfs_backref_drop_node_buffer(node);
3986 atomic_inc(&cow->refs);
3988 node->new_bytenr = cow->start;
3990 if (!node->pending) {
3991 list_move_tail(&node->list,
3992 &rc->backref_cache.pending[level]);
3997 mark_block_processed(rc, node);
3999 if (first_cow && level > 0)
4000 rc->nodes_relocated += buf->len;
4003 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4004 ret = replace_file_extents(trans, rc, root, cow);
4009 * called before creating snapshot. it calculates metadata reservation
4010 * required for relocating tree blocks in the snapshot
4012 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4013 u64 *bytes_to_reserve)
4015 struct btrfs_root *root = pending->root;
4016 struct reloc_control *rc = root->fs_info->reloc_ctl;
4018 if (!rc || !have_reloc_root(root))
4021 if (!rc->merge_reloc_tree)
4024 root = root->reloc_root;
4025 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4027 * relocation is in the stage of merging trees. the space
4028 * used by merging a reloc tree is twice the size of
4029 * relocated tree nodes in the worst case. half for cowing
4030 * the reloc tree, half for cowing the fs tree. the space
4031 * used by cowing the reloc tree will be freed after the
4032 * tree is dropped. if we create snapshot, cowing the fs
4033 * tree may use more space than it frees. so we need
4034 * reserve extra space.
4036 *bytes_to_reserve += rc->nodes_relocated;
4040 * called after snapshot is created. migrate block reservation
4041 * and create reloc root for the newly created snapshot
4043 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4044 * references held on the reloc_root, one for root->reloc_root and one for
4047 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4048 struct btrfs_pending_snapshot *pending)
4050 struct btrfs_root *root = pending->root;
4051 struct btrfs_root *reloc_root;
4052 struct btrfs_root *new_root;
4053 struct reloc_control *rc = root->fs_info->reloc_ctl;
4056 if (!rc || !have_reloc_root(root))
4059 rc = root->fs_info->reloc_ctl;
4060 rc->merging_rsv_size += rc->nodes_relocated;
4062 if (rc->merge_reloc_tree) {
4063 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4065 rc->nodes_relocated, true);
4070 new_root = pending->snap;
4071 reloc_root = create_reloc_root(trans, root->reloc_root,
4072 new_root->root_key.objectid);
4073 if (IS_ERR(reloc_root))
4074 return PTR_ERR(reloc_root);
4076 ret = __add_reloc_root(reloc_root);
4078 new_root->reloc_root = btrfs_grab_root(reloc_root);
4080 if (rc->create_reloc_tree)
4081 ret = clone_backref_node(trans, rc, root, reloc_root);