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);
674 BUG_ON((struct btrfs_root *)node->data != root);
678 * We only put the reloc root here if it's on the list. There's a lot
679 * of places where the pattern is to splice the rc->reloc_roots, process
680 * the reloc roots, and then add the reloc root back onto
681 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
682 * list we don't want the reference being dropped, because the guy
683 * messing with the list is in charge of the reference.
685 spin_lock(&fs_info->trans_lock);
686 if (!list_empty(&root->root_list)) {
688 list_del_init(&root->root_list);
690 spin_unlock(&fs_info->trans_lock);
692 btrfs_put_root(root);
697 * helper to update the 'address of tree root -> reloc tree'
700 static int __update_reloc_root(struct btrfs_root *root)
702 struct btrfs_fs_info *fs_info = root->fs_info;
703 struct rb_node *rb_node;
704 struct mapping_node *node = NULL;
705 struct reloc_control *rc = fs_info->reloc_ctl;
707 spin_lock(&rc->reloc_root_tree.lock);
708 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
709 root->commit_root->start);
711 node = rb_entry(rb_node, struct mapping_node, rb_node);
712 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
714 spin_unlock(&rc->reloc_root_tree.lock);
718 BUG_ON((struct btrfs_root *)node->data != root);
720 spin_lock(&rc->reloc_root_tree.lock);
721 node->bytenr = root->node->start;
722 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
723 node->bytenr, &node->rb_node);
724 spin_unlock(&rc->reloc_root_tree.lock);
726 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
730 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
731 struct btrfs_root *root, u64 objectid)
733 struct btrfs_fs_info *fs_info = root->fs_info;
734 struct btrfs_root *reloc_root;
735 struct extent_buffer *eb;
736 struct btrfs_root_item *root_item;
737 struct btrfs_key root_key;
740 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
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);
755 * Set the last_snapshot field to the generation of the commit
756 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
757 * correctly (returns true) when the relocation root is created
758 * either inside the critical section of a transaction commit
759 * (through transaction.c:qgroup_account_snapshot()) and when
760 * it's created before the transaction commit is started.
762 commit_root_gen = btrfs_header_generation(root->commit_root);
763 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
766 * called by btrfs_reloc_post_snapshot_hook.
767 * the source tree is a reloc tree, all tree blocks
768 * modified after it was created have RELOC flag
769 * set in their headers. so it's OK to not update
770 * the 'last_snapshot'.
772 ret = btrfs_copy_root(trans, root, root->node, &eb,
773 BTRFS_TREE_RELOC_OBJECTID);
777 memcpy(root_item, &root->root_item, sizeof(*root_item));
778 btrfs_set_root_bytenr(root_item, eb->start);
779 btrfs_set_root_level(root_item, btrfs_header_level(eb));
780 btrfs_set_root_generation(root_item, trans->transid);
782 if (root->root_key.objectid == objectid) {
783 btrfs_set_root_refs(root_item, 0);
784 memset(&root_item->drop_progress, 0,
785 sizeof(struct btrfs_disk_key));
786 root_item->drop_level = 0;
789 btrfs_tree_unlock(eb);
790 free_extent_buffer(eb);
792 ret = btrfs_insert_root(trans, fs_info->tree_root,
793 &root_key, root_item);
797 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
798 BUG_ON(IS_ERR(reloc_root));
799 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
800 reloc_root->last_trans = trans->transid;
805 * create reloc tree for a given fs tree. reloc tree is just a
806 * snapshot of the fs tree with special root objectid.
808 * The reloc_root comes out of here with two references, one for
809 * root->reloc_root, and another for being on the rc->reloc_roots list.
811 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
812 struct btrfs_root *root)
814 struct btrfs_fs_info *fs_info = root->fs_info;
815 struct btrfs_root *reloc_root;
816 struct reloc_control *rc = fs_info->reloc_ctl;
817 struct btrfs_block_rsv *rsv;
825 * The subvolume has reloc tree but the swap is finished, no need to
826 * create/update the dead reloc tree
828 if (reloc_root_is_dead(root))
832 * This is subtle but important. We do not do
833 * record_root_in_transaction for reloc roots, instead we record their
834 * corresponding fs root, and then here we update the last trans for the
835 * reloc root. This means that we have to do this for the entire life
836 * of the reloc root, regardless of which stage of the relocation we are
839 if (root->reloc_root) {
840 reloc_root = root->reloc_root;
841 reloc_root->last_trans = trans->transid;
846 * We are merging reloc roots, we do not need new reloc trees. Also
847 * reloc trees never need their own reloc tree.
849 if (!rc->create_reloc_tree ||
850 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
853 if (!trans->reloc_reserved) {
854 rsv = trans->block_rsv;
855 trans->block_rsv = rc->block_rsv;
858 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
860 trans->block_rsv = rsv;
862 ret = __add_reloc_root(reloc_root);
864 root->reloc_root = btrfs_grab_root(reloc_root);
869 * update root item of reloc tree
871 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
872 struct btrfs_root *root)
874 struct btrfs_fs_info *fs_info = root->fs_info;
875 struct btrfs_root *reloc_root;
876 struct btrfs_root_item *root_item;
879 if (!have_reloc_root(root))
882 reloc_root = root->reloc_root;
883 root_item = &reloc_root->root_item;
886 * We are probably ok here, but __del_reloc_root() will drop its ref of
887 * the root. We have the ref for root->reloc_root, but just in case
888 * hold it while we update the reloc root.
890 btrfs_grab_root(reloc_root);
892 /* root->reloc_root will stay until current relocation finished */
893 if (fs_info->reloc_ctl->merge_reloc_tree &&
894 btrfs_root_refs(root_item) == 0) {
895 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
897 * Mark the tree as dead before we change reloc_root so
898 * have_reloc_root will not touch it from now on.
901 __del_reloc_root(reloc_root);
904 if (reloc_root->commit_root != reloc_root->node) {
905 __update_reloc_root(reloc_root);
906 btrfs_set_root_node(root_item, reloc_root->node);
907 free_extent_buffer(reloc_root->commit_root);
908 reloc_root->commit_root = btrfs_root_node(reloc_root);
911 ret = btrfs_update_root(trans, fs_info->tree_root,
912 &reloc_root->root_key, root_item);
914 btrfs_put_root(reloc_root);
920 * helper to find first cached inode with inode number >= objectid
923 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
925 struct rb_node *node;
926 struct rb_node *prev;
927 struct btrfs_inode *entry;
930 spin_lock(&root->inode_lock);
932 node = root->inode_tree.rb_node;
936 entry = rb_entry(node, struct btrfs_inode, rb_node);
938 if (objectid < btrfs_ino(entry))
939 node = node->rb_left;
940 else if (objectid > btrfs_ino(entry))
941 node = node->rb_right;
947 entry = rb_entry(prev, struct btrfs_inode, rb_node);
948 if (objectid <= btrfs_ino(entry)) {
952 prev = rb_next(prev);
956 entry = rb_entry(node, struct btrfs_inode, rb_node);
957 inode = igrab(&entry->vfs_inode);
959 spin_unlock(&root->inode_lock);
963 objectid = btrfs_ino(entry) + 1;
964 if (cond_resched_lock(&root->inode_lock))
967 node = rb_next(node);
969 spin_unlock(&root->inode_lock);
974 * get new location of data
976 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
977 u64 bytenr, u64 num_bytes)
979 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
980 struct btrfs_path *path;
981 struct btrfs_file_extent_item *fi;
982 struct extent_buffer *leaf;
985 path = btrfs_alloc_path();
989 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
990 ret = btrfs_lookup_file_extent(NULL, root, path,
991 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
999 leaf = path->nodes[0];
1000 fi = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_file_extent_item);
1003 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1004 btrfs_file_extent_compression(leaf, fi) ||
1005 btrfs_file_extent_encryption(leaf, fi) ||
1006 btrfs_file_extent_other_encoding(leaf, fi));
1008 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1013 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1016 btrfs_free_path(path);
1021 * update file extent items in the tree leaf to point to
1022 * the new locations.
1024 static noinline_for_stack
1025 int replace_file_extents(struct btrfs_trans_handle *trans,
1026 struct reloc_control *rc,
1027 struct btrfs_root *root,
1028 struct extent_buffer *leaf)
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1031 struct btrfs_key key;
1032 struct btrfs_file_extent_item *fi;
1033 struct inode *inode = NULL;
1045 if (rc->stage != UPDATE_DATA_PTRS)
1048 /* reloc trees always use full backref */
1049 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1050 parent = leaf->start;
1054 nritems = btrfs_header_nritems(leaf);
1055 for (i = 0; i < nritems; i++) {
1056 struct btrfs_ref ref = { 0 };
1059 btrfs_item_key_to_cpu(leaf, &key, i);
1060 if (key.type != BTRFS_EXTENT_DATA_KEY)
1062 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1063 if (btrfs_file_extent_type(leaf, fi) ==
1064 BTRFS_FILE_EXTENT_INLINE)
1066 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1067 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1070 if (!in_range(bytenr, rc->block_group->start,
1071 rc->block_group->length))
1075 * if we are modifying block in fs tree, wait for readpage
1076 * to complete and drop the extent cache
1078 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1080 inode = find_next_inode(root, key.objectid);
1082 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1083 btrfs_add_delayed_iput(inode);
1084 inode = find_next_inode(root, key.objectid);
1086 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1088 btrfs_file_extent_num_bytes(leaf, fi);
1089 WARN_ON(!IS_ALIGNED(key.offset,
1090 fs_info->sectorsize));
1091 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1093 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1098 btrfs_drop_extent_cache(BTRFS_I(inode),
1099 key.offset, end, 1);
1100 unlock_extent(&BTRFS_I(inode)->io_tree,
1105 ret = get_new_location(rc->data_inode, &new_bytenr,
1109 * Don't have to abort since we've not changed anything
1110 * in the file extent yet.
1115 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1118 key.offset -= btrfs_file_extent_offset(leaf, fi);
1119 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1121 ref.real_root = root->root_key.objectid;
1122 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1123 key.objectid, key.offset);
1124 ret = btrfs_inc_extent_ref(trans, &ref);
1126 btrfs_abort_transaction(trans, ret);
1130 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1132 ref.real_root = root->root_key.objectid;
1133 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1134 key.objectid, key.offset);
1135 ret = btrfs_free_extent(trans, &ref);
1137 btrfs_abort_transaction(trans, ret);
1142 btrfs_mark_buffer_dirty(leaf);
1144 btrfs_add_delayed_iput(inode);
1148 static noinline_for_stack
1149 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1150 struct btrfs_path *path, int level)
1152 struct btrfs_disk_key key1;
1153 struct btrfs_disk_key key2;
1154 btrfs_node_key(eb, &key1, slot);
1155 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1156 return memcmp(&key1, &key2, sizeof(key1));
1160 * try to replace tree blocks in fs tree with the new blocks
1161 * in reloc tree. tree blocks haven't been modified since the
1162 * reloc tree was create can be replaced.
1164 * if a block was replaced, level of the block + 1 is returned.
1165 * if no block got replaced, 0 is returned. if there are other
1166 * errors, a negative error number is returned.
1168 static noinline_for_stack
1169 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1170 struct btrfs_root *dest, struct btrfs_root *src,
1171 struct btrfs_path *path, struct btrfs_key *next_key,
1172 int lowest_level, int max_level)
1174 struct btrfs_fs_info *fs_info = dest->fs_info;
1175 struct extent_buffer *eb;
1176 struct extent_buffer *parent;
1177 struct btrfs_ref ref = { 0 };
1178 struct btrfs_key key;
1190 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1191 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1193 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1195 slot = path->slots[lowest_level];
1196 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1198 eb = btrfs_lock_root_node(dest);
1199 btrfs_set_lock_blocking_write(eb);
1200 level = btrfs_header_level(eb);
1202 if (level < lowest_level) {
1203 btrfs_tree_unlock(eb);
1204 free_extent_buffer(eb);
1209 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1212 btrfs_set_lock_blocking_write(eb);
1215 next_key->objectid = (u64)-1;
1216 next_key->type = (u8)-1;
1217 next_key->offset = (u64)-1;
1222 struct btrfs_key first_key;
1224 level = btrfs_header_level(parent);
1225 BUG_ON(level < lowest_level);
1227 ret = btrfs_bin_search(parent, &key, &slot);
1230 if (ret && slot > 0)
1233 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1234 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1236 old_bytenr = btrfs_node_blockptr(parent, slot);
1237 blocksize = fs_info->nodesize;
1238 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1239 btrfs_node_key_to_cpu(parent, &first_key, slot);
1241 if (level <= max_level) {
1242 eb = path->nodes[level];
1243 new_bytenr = btrfs_node_blockptr(eb,
1244 path->slots[level]);
1245 new_ptr_gen = btrfs_node_ptr_generation(eb,
1246 path->slots[level]);
1252 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1257 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1258 memcmp_node_keys(parent, slot, path, level)) {
1259 if (level <= lowest_level) {
1264 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1265 level - 1, &first_key);
1269 } else if (!extent_buffer_uptodate(eb)) {
1271 free_extent_buffer(eb);
1274 btrfs_tree_lock(eb);
1276 ret = btrfs_cow_block(trans, dest, eb, parent,
1280 btrfs_set_lock_blocking_write(eb);
1282 btrfs_tree_unlock(parent);
1283 free_extent_buffer(parent);
1290 btrfs_tree_unlock(parent);
1291 free_extent_buffer(parent);
1296 btrfs_node_key_to_cpu(path->nodes[level], &key,
1297 path->slots[level]);
1298 btrfs_release_path(path);
1300 path->lowest_level = level;
1301 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1302 path->lowest_level = 0;
1306 * Info qgroup to trace both subtrees.
1308 * We must trace both trees.
1309 * 1) Tree reloc subtree
1310 * If not traced, we will leak data numbers
1312 * If not traced, we will double count old data
1314 * We don't scan the subtree right now, but only record
1315 * the swapped tree blocks.
1316 * The real subtree rescan is delayed until we have new
1317 * CoW on the subtree root node before transaction commit.
1319 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1320 rc->block_group, parent, slot,
1321 path->nodes[level], path->slots[level],
1326 * swap blocks in fs tree and reloc tree.
1328 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1329 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1330 btrfs_mark_buffer_dirty(parent);
1332 btrfs_set_node_blockptr(path->nodes[level],
1333 path->slots[level], old_bytenr);
1334 btrfs_set_node_ptr_generation(path->nodes[level],
1335 path->slots[level], old_ptr_gen);
1336 btrfs_mark_buffer_dirty(path->nodes[level]);
1338 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1339 blocksize, path->nodes[level]->start);
1340 ref.skip_qgroup = true;
1341 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1342 ret = btrfs_inc_extent_ref(trans, &ref);
1344 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1346 ref.skip_qgroup = true;
1347 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1348 ret = btrfs_inc_extent_ref(trans, &ref);
1351 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1352 blocksize, path->nodes[level]->start);
1353 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1354 ref.skip_qgroup = true;
1355 ret = btrfs_free_extent(trans, &ref);
1358 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1360 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1361 ref.skip_qgroup = true;
1362 ret = btrfs_free_extent(trans, &ref);
1365 btrfs_unlock_up_safe(path, 0);
1370 btrfs_tree_unlock(parent);
1371 free_extent_buffer(parent);
1376 * helper to find next relocated block in reloc tree
1378 static noinline_for_stack
1379 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1382 struct extent_buffer *eb;
1387 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1389 for (i = 0; i < *level; i++) {
1390 free_extent_buffer(path->nodes[i]);
1391 path->nodes[i] = NULL;
1394 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1395 eb = path->nodes[i];
1396 nritems = btrfs_header_nritems(eb);
1397 while (path->slots[i] + 1 < nritems) {
1399 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1406 free_extent_buffer(path->nodes[i]);
1407 path->nodes[i] = NULL;
1413 * walk down reloc tree to find relocated block of lowest level
1415 static noinline_for_stack
1416 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1419 struct btrfs_fs_info *fs_info = root->fs_info;
1420 struct extent_buffer *eb = NULL;
1427 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1429 for (i = *level; i > 0; i--) {
1430 struct btrfs_key first_key;
1432 eb = path->nodes[i];
1433 nritems = btrfs_header_nritems(eb);
1434 while (path->slots[i] < nritems) {
1435 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1436 if (ptr_gen > last_snapshot)
1440 if (path->slots[i] >= nritems) {
1451 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1452 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
1453 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
1457 } else if (!extent_buffer_uptodate(eb)) {
1458 free_extent_buffer(eb);
1461 BUG_ON(btrfs_header_level(eb) != i - 1);
1462 path->nodes[i - 1] = eb;
1463 path->slots[i - 1] = 0;
1469 * invalidate extent cache for file extents whose key in range of
1470 * [min_key, max_key)
1472 static int invalidate_extent_cache(struct btrfs_root *root,
1473 struct btrfs_key *min_key,
1474 struct btrfs_key *max_key)
1476 struct btrfs_fs_info *fs_info = root->fs_info;
1477 struct inode *inode = NULL;
1482 objectid = min_key->objectid;
1487 if (objectid > max_key->objectid)
1490 inode = find_next_inode(root, objectid);
1493 ino = btrfs_ino(BTRFS_I(inode));
1495 if (ino > max_key->objectid) {
1501 if (!S_ISREG(inode->i_mode))
1504 if (unlikely(min_key->objectid == ino)) {
1505 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1507 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1510 start = min_key->offset;
1511 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1517 if (unlikely(max_key->objectid == ino)) {
1518 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1520 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1523 if (max_key->offset == 0)
1525 end = max_key->offset;
1526 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1533 /* the lock_extent waits for readpage to complete */
1534 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1535 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1536 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1541 static int find_next_key(struct btrfs_path *path, int level,
1542 struct btrfs_key *key)
1545 while (level < BTRFS_MAX_LEVEL) {
1546 if (!path->nodes[level])
1548 if (path->slots[level] + 1 <
1549 btrfs_header_nritems(path->nodes[level])) {
1550 btrfs_node_key_to_cpu(path->nodes[level], key,
1551 path->slots[level] + 1);
1560 * Insert current subvolume into reloc_control::dirty_subvol_roots
1562 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1563 struct reloc_control *rc,
1564 struct btrfs_root *root)
1566 struct btrfs_root *reloc_root = root->reloc_root;
1567 struct btrfs_root_item *reloc_root_item;
1569 /* @root must be a subvolume tree root with a valid reloc tree */
1570 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1573 reloc_root_item = &reloc_root->root_item;
1574 memset(&reloc_root_item->drop_progress, 0,
1575 sizeof(reloc_root_item->drop_progress));
1576 reloc_root_item->drop_level = 0;
1577 btrfs_set_root_refs(reloc_root_item, 0);
1578 btrfs_update_reloc_root(trans, root);
1580 if (list_empty(&root->reloc_dirty_list)) {
1581 btrfs_grab_root(root);
1582 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1586 static int clean_dirty_subvols(struct reloc_control *rc)
1588 struct btrfs_root *root;
1589 struct btrfs_root *next;
1593 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1595 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1596 /* Merged subvolume, cleanup its reloc root */
1597 struct btrfs_root *reloc_root = root->reloc_root;
1599 list_del_init(&root->reloc_dirty_list);
1600 root->reloc_root = NULL;
1602 * Need barrier to ensure clear_bit() only happens after
1603 * root->reloc_root = NULL. Pairs with have_reloc_root.
1606 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1609 * btrfs_drop_snapshot drops our ref we hold for
1610 * ->reloc_root. If it fails however we must
1611 * drop the ref ourselves.
1613 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1615 btrfs_put_root(reloc_root);
1620 btrfs_put_root(root);
1622 /* Orphan reloc tree, just clean it up */
1623 ret2 = btrfs_drop_snapshot(root, 0, 1);
1625 btrfs_put_root(root);
1635 * merge the relocated tree blocks in reloc tree with corresponding
1638 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1639 struct btrfs_root *root)
1641 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1642 struct btrfs_key key;
1643 struct btrfs_key next_key;
1644 struct btrfs_trans_handle *trans = NULL;
1645 struct btrfs_root *reloc_root;
1646 struct btrfs_root_item *root_item;
1647 struct btrfs_path *path;
1648 struct extent_buffer *leaf;
1656 path = btrfs_alloc_path();
1659 path->reada = READA_FORWARD;
1661 reloc_root = root->reloc_root;
1662 root_item = &reloc_root->root_item;
1664 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1665 level = btrfs_root_level(root_item);
1666 atomic_inc(&reloc_root->node->refs);
1667 path->nodes[level] = reloc_root->node;
1668 path->slots[level] = 0;
1670 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1672 level = root_item->drop_level;
1674 path->lowest_level = level;
1675 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1676 path->lowest_level = 0;
1678 btrfs_free_path(path);
1682 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1683 path->slots[level]);
1684 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1686 btrfs_unlock_up_safe(path, 0);
1690 * In merge_reloc_root(), we modify the upper level pointer to swap the
1691 * tree blocks between reloc tree and subvolume tree. Thus for tree
1692 * block COW, we COW at most from level 1 to root level for each tree.
1694 * Thus the needed metadata size is at most root_level * nodesize,
1695 * and * 2 since we have two trees to COW.
1697 min_reserved = fs_info->nodesize * btrfs_root_level(root_item) * 2;
1698 memset(&next_key, 0, sizeof(next_key));
1701 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1702 BTRFS_RESERVE_FLUSH_LIMIT);
1707 trans = btrfs_start_transaction(root, 0);
1708 if (IS_ERR(trans)) {
1709 err = PTR_ERR(trans);
1715 * At this point we no longer have a reloc_control, so we can't
1716 * depend on btrfs_init_reloc_root to update our last_trans.
1718 * But that's ok, we started the trans handle on our
1719 * corresponding fs_root, which means it's been added to the
1720 * dirty list. At commit time we'll still call
1721 * btrfs_update_reloc_root() and update our root item
1724 reloc_root->last_trans = trans->transid;
1725 trans->block_rsv = rc->block_rsv;
1730 ret = walk_down_reloc_tree(reloc_root, path, &level);
1738 if (!find_next_key(path, level, &key) &&
1739 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1742 ret = replace_path(trans, rc, root, reloc_root, path,
1743 &next_key, level, max_level);
1752 btrfs_node_key_to_cpu(path->nodes[level], &key,
1753 path->slots[level]);
1757 ret = walk_up_reloc_tree(reloc_root, path, &level);
1763 * save the merging progress in the drop_progress.
1764 * this is OK since root refs == 1 in this case.
1766 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1767 path->slots[level]);
1768 root_item->drop_level = level;
1770 btrfs_end_transaction_throttle(trans);
1773 btrfs_btree_balance_dirty(fs_info);
1775 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1776 invalidate_extent_cache(root, &key, &next_key);
1780 * handle the case only one block in the fs tree need to be
1781 * relocated and the block is tree root.
1783 leaf = btrfs_lock_root_node(root);
1784 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
1785 btrfs_tree_unlock(leaf);
1786 free_extent_buffer(leaf);
1790 btrfs_free_path(path);
1793 insert_dirty_subvol(trans, rc, root);
1796 btrfs_end_transaction_throttle(trans);
1798 btrfs_btree_balance_dirty(fs_info);
1800 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1801 invalidate_extent_cache(root, &key, &next_key);
1806 static noinline_for_stack
1807 int prepare_to_merge(struct reloc_control *rc, int err)
1809 struct btrfs_root *root = rc->extent_root;
1810 struct btrfs_fs_info *fs_info = root->fs_info;
1811 struct btrfs_root *reloc_root;
1812 struct btrfs_trans_handle *trans;
1813 LIST_HEAD(reloc_roots);
1817 mutex_lock(&fs_info->reloc_mutex);
1818 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1819 rc->merging_rsv_size += rc->nodes_relocated * 2;
1820 mutex_unlock(&fs_info->reloc_mutex);
1824 num_bytes = rc->merging_rsv_size;
1825 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1826 BTRFS_RESERVE_FLUSH_ALL);
1831 trans = btrfs_join_transaction(rc->extent_root);
1832 if (IS_ERR(trans)) {
1834 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1836 return PTR_ERR(trans);
1840 if (num_bytes != rc->merging_rsv_size) {
1841 btrfs_end_transaction(trans);
1842 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1848 rc->merge_reloc_tree = 1;
1850 while (!list_empty(&rc->reloc_roots)) {
1851 reloc_root = list_entry(rc->reloc_roots.next,
1852 struct btrfs_root, root_list);
1853 list_del_init(&reloc_root->root_list);
1855 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1857 BUG_ON(IS_ERR(root));
1858 BUG_ON(root->reloc_root != reloc_root);
1861 * set reference count to 1, so btrfs_recover_relocation
1862 * knows it should resumes merging
1865 btrfs_set_root_refs(&reloc_root->root_item, 1);
1866 btrfs_update_reloc_root(trans, root);
1868 list_add(&reloc_root->root_list, &reloc_roots);
1869 btrfs_put_root(root);
1872 list_splice(&reloc_roots, &rc->reloc_roots);
1875 btrfs_commit_transaction(trans);
1877 btrfs_end_transaction(trans);
1881 static noinline_for_stack
1882 void free_reloc_roots(struct list_head *list)
1884 struct btrfs_root *reloc_root, *tmp;
1886 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1887 __del_reloc_root(reloc_root);
1890 static noinline_for_stack
1891 void merge_reloc_roots(struct reloc_control *rc)
1893 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1894 struct btrfs_root *root;
1895 struct btrfs_root *reloc_root;
1896 LIST_HEAD(reloc_roots);
1900 root = rc->extent_root;
1903 * this serializes us with btrfs_record_root_in_transaction,
1904 * we have to make sure nobody is in the middle of
1905 * adding their roots to the list while we are
1908 mutex_lock(&fs_info->reloc_mutex);
1909 list_splice_init(&rc->reloc_roots, &reloc_roots);
1910 mutex_unlock(&fs_info->reloc_mutex);
1912 while (!list_empty(&reloc_roots)) {
1914 reloc_root = list_entry(reloc_roots.next,
1915 struct btrfs_root, root_list);
1917 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1919 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1920 BUG_ON(IS_ERR(root));
1921 BUG_ON(root->reloc_root != reloc_root);
1922 ret = merge_reloc_root(rc, root);
1923 btrfs_put_root(root);
1925 if (list_empty(&reloc_root->root_list))
1926 list_add_tail(&reloc_root->root_list,
1931 if (!IS_ERR(root)) {
1932 if (root->reloc_root == reloc_root) {
1933 root->reloc_root = NULL;
1934 btrfs_put_root(reloc_root);
1936 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1938 btrfs_put_root(root);
1941 list_del_init(&reloc_root->root_list);
1942 /* Don't forget to queue this reloc root for cleanup */
1943 list_add_tail(&reloc_root->reloc_dirty_list,
1944 &rc->dirty_subvol_roots);
1954 btrfs_handle_fs_error(fs_info, ret, NULL);
1955 free_reloc_roots(&reloc_roots);
1957 /* new reloc root may be added */
1958 mutex_lock(&fs_info->reloc_mutex);
1959 list_splice_init(&rc->reloc_roots, &reloc_roots);
1960 mutex_unlock(&fs_info->reloc_mutex);
1961 free_reloc_roots(&reloc_roots);
1967 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1969 * here, but it's wrong. If we fail to start the transaction in
1970 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1971 * have actually been removed from the reloc_root_tree rb tree. This is
1972 * fine because we're bailing here, and we hold a reference on the root
1973 * for the list that holds it, so these roots will be cleaned up when we
1974 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1975 * will be cleaned up on unmount.
1977 * The remaining nodes will be cleaned up by free_reloc_control.
1981 static void free_block_list(struct rb_root *blocks)
1983 struct tree_block *block;
1984 struct rb_node *rb_node;
1985 while ((rb_node = rb_first(blocks))) {
1986 block = rb_entry(rb_node, struct tree_block, rb_node);
1987 rb_erase(rb_node, blocks);
1992 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1993 struct btrfs_root *reloc_root)
1995 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
1996 struct btrfs_root *root;
1999 if (reloc_root->last_trans == trans->transid)
2002 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2003 BUG_ON(IS_ERR(root));
2004 BUG_ON(root->reloc_root != reloc_root);
2005 ret = btrfs_record_root_in_trans(trans, root);
2006 btrfs_put_root(root);
2011 static noinline_for_stack
2012 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2013 struct reloc_control *rc,
2014 struct btrfs_backref_node *node,
2015 struct btrfs_backref_edge *edges[])
2017 struct btrfs_backref_node *next;
2018 struct btrfs_root *root;
2024 next = walk_up_backref(next, edges, &index);
2027 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2029 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2030 record_reloc_root_in_trans(trans, root);
2034 btrfs_record_root_in_trans(trans, root);
2035 root = root->reloc_root;
2037 if (next->new_bytenr != root->node->start) {
2038 BUG_ON(next->new_bytenr);
2039 BUG_ON(!list_empty(&next->list));
2040 next->new_bytenr = root->node->start;
2041 btrfs_put_root(next->root);
2042 next->root = btrfs_grab_root(root);
2044 list_add_tail(&next->list,
2045 &rc->backref_cache.changed);
2046 mark_block_processed(rc, next);
2052 next = walk_down_backref(edges, &index);
2053 if (!next || next->level <= node->level)
2060 /* setup backref node path for btrfs_reloc_cow_block */
2062 rc->backref_cache.path[next->level] = next;
2065 next = edges[index]->node[UPPER];
2071 * Select a tree root for relocation.
2073 * Return NULL if the block is not shareable. We should use do_relocation() in
2076 * Return a tree root pointer if the block is shareable.
2077 * Return -ENOENT if the block is root of reloc tree.
2079 static noinline_for_stack
2080 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2082 struct btrfs_backref_node *next;
2083 struct btrfs_root *root;
2084 struct btrfs_root *fs_root = NULL;
2085 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2091 next = walk_up_backref(next, edges, &index);
2095 /* No other choice for non-shareable tree */
2096 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2099 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2105 next = walk_down_backref(edges, &index);
2106 if (!next || next->level <= node->level)
2111 return ERR_PTR(-ENOENT);
2115 static noinline_for_stack
2116 u64 calcu_metadata_size(struct reloc_control *rc,
2117 struct btrfs_backref_node *node, int reserve)
2119 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2120 struct btrfs_backref_node *next = node;
2121 struct btrfs_backref_edge *edge;
2122 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2126 BUG_ON(reserve && node->processed);
2131 if (next->processed && (reserve || next != node))
2134 num_bytes += fs_info->nodesize;
2136 if (list_empty(&next->upper))
2139 edge = list_entry(next->upper.next,
2140 struct btrfs_backref_edge, list[LOWER]);
2141 edges[index++] = edge;
2142 next = edge->node[UPPER];
2144 next = walk_down_backref(edges, &index);
2149 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2150 struct reloc_control *rc,
2151 struct btrfs_backref_node *node)
2153 struct btrfs_root *root = rc->extent_root;
2154 struct btrfs_fs_info *fs_info = root->fs_info;
2159 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2161 trans->block_rsv = rc->block_rsv;
2162 rc->reserved_bytes += num_bytes;
2165 * We are under a transaction here so we can only do limited flushing.
2166 * If we get an enospc just kick back -EAGAIN so we know to drop the
2167 * transaction and try to refill when we can flush all the things.
2169 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2170 BTRFS_RESERVE_FLUSH_LIMIT);
2172 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2173 while (tmp <= rc->reserved_bytes)
2176 * only one thread can access block_rsv at this point,
2177 * so we don't need hold lock to protect block_rsv.
2178 * we expand more reservation size here to allow enough
2179 * space for relocation and we will return earlier in
2182 rc->block_rsv->size = tmp + fs_info->nodesize *
2183 RELOCATION_RESERVED_NODES;
2191 * relocate a block tree, and then update pointers in upper level
2192 * blocks that reference the block to point to the new location.
2194 * if called by link_to_upper, the block has already been relocated.
2195 * in that case this function just updates pointers.
2197 static int do_relocation(struct btrfs_trans_handle *trans,
2198 struct reloc_control *rc,
2199 struct btrfs_backref_node *node,
2200 struct btrfs_key *key,
2201 struct btrfs_path *path, int lowest)
2203 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2204 struct btrfs_backref_node *upper;
2205 struct btrfs_backref_edge *edge;
2206 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2207 struct btrfs_root *root;
2208 struct extent_buffer *eb;
2216 BUG_ON(lowest && node->eb);
2218 path->lowest_level = node->level + 1;
2219 rc->backref_cache.path[node->level] = node;
2220 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2221 struct btrfs_key first_key;
2222 struct btrfs_ref ref = { 0 };
2226 upper = edge->node[UPPER];
2227 root = select_reloc_root(trans, rc, upper, edges);
2230 if (upper->eb && !upper->locked) {
2232 ret = btrfs_bin_search(upper->eb, key, &slot);
2238 bytenr = btrfs_node_blockptr(upper->eb, slot);
2239 if (node->eb->start == bytenr)
2242 btrfs_backref_drop_node_buffer(upper);
2246 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2253 btrfs_release_path(path);
2258 upper->eb = path->nodes[upper->level];
2259 path->nodes[upper->level] = NULL;
2261 BUG_ON(upper->eb != path->nodes[upper->level]);
2265 path->locks[upper->level] = 0;
2267 slot = path->slots[upper->level];
2268 btrfs_release_path(path);
2270 ret = btrfs_bin_search(upper->eb, key, &slot);
2278 bytenr = btrfs_node_blockptr(upper->eb, slot);
2280 if (bytenr != node->bytenr) {
2281 btrfs_err(root->fs_info,
2282 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2283 bytenr, node->bytenr, slot,
2289 if (node->eb->start == bytenr)
2293 blocksize = root->fs_info->nodesize;
2294 generation = btrfs_node_ptr_generation(upper->eb, slot);
2295 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2296 eb = read_tree_block(fs_info, bytenr, generation,
2297 upper->level - 1, &first_key);
2301 } else if (!extent_buffer_uptodate(eb)) {
2302 free_extent_buffer(eb);
2306 btrfs_tree_lock(eb);
2307 btrfs_set_lock_blocking_write(eb);
2310 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2312 btrfs_tree_unlock(eb);
2313 free_extent_buffer(eb);
2318 BUG_ON(node->eb != eb);
2320 btrfs_set_node_blockptr(upper->eb, slot,
2322 btrfs_set_node_ptr_generation(upper->eb, slot,
2324 btrfs_mark_buffer_dirty(upper->eb);
2326 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2327 node->eb->start, blocksize,
2329 ref.real_root = root->root_key.objectid;
2330 btrfs_init_tree_ref(&ref, node->level,
2331 btrfs_header_owner(upper->eb));
2332 ret = btrfs_inc_extent_ref(trans, &ref);
2335 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2339 if (!upper->pending)
2340 btrfs_backref_drop_node_buffer(upper);
2342 btrfs_backref_unlock_node_buffer(upper);
2347 if (!err && node->pending) {
2348 btrfs_backref_drop_node_buffer(node);
2349 list_move_tail(&node->list, &rc->backref_cache.changed);
2353 path->lowest_level = 0;
2354 BUG_ON(err == -ENOSPC);
2358 static int link_to_upper(struct btrfs_trans_handle *trans,
2359 struct reloc_control *rc,
2360 struct btrfs_backref_node *node,
2361 struct btrfs_path *path)
2363 struct btrfs_key key;
2365 btrfs_node_key_to_cpu(node->eb, &key, 0);
2366 return do_relocation(trans, rc, node, &key, path, 0);
2369 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2370 struct reloc_control *rc,
2371 struct btrfs_path *path, int err)
2374 struct btrfs_backref_cache *cache = &rc->backref_cache;
2375 struct btrfs_backref_node *node;
2379 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2380 while (!list_empty(&cache->pending[level])) {
2381 node = list_entry(cache->pending[level].next,
2382 struct btrfs_backref_node, list);
2383 list_move_tail(&node->list, &list);
2384 BUG_ON(!node->pending);
2387 ret = link_to_upper(trans, rc, node, path);
2392 list_splice_init(&list, &cache->pending[level]);
2398 * mark a block and all blocks directly/indirectly reference the block
2401 static void update_processed_blocks(struct reloc_control *rc,
2402 struct btrfs_backref_node *node)
2404 struct btrfs_backref_node *next = node;
2405 struct btrfs_backref_edge *edge;
2406 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2412 if (next->processed)
2415 mark_block_processed(rc, next);
2417 if (list_empty(&next->upper))
2420 edge = list_entry(next->upper.next,
2421 struct btrfs_backref_edge, list[LOWER]);
2422 edges[index++] = edge;
2423 next = edge->node[UPPER];
2425 next = walk_down_backref(edges, &index);
2429 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2431 u32 blocksize = rc->extent_root->fs_info->nodesize;
2433 if (test_range_bit(&rc->processed_blocks, bytenr,
2434 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2439 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2440 struct tree_block *block)
2442 struct extent_buffer *eb;
2444 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2445 block->level, NULL);
2448 } else if (!extent_buffer_uptodate(eb)) {
2449 free_extent_buffer(eb);
2452 if (block->level == 0)
2453 btrfs_item_key_to_cpu(eb, &block->key, 0);
2455 btrfs_node_key_to_cpu(eb, &block->key, 0);
2456 free_extent_buffer(eb);
2457 block->key_ready = 1;
2462 * helper function to relocate a tree block
2464 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2465 struct reloc_control *rc,
2466 struct btrfs_backref_node *node,
2467 struct btrfs_key *key,
2468 struct btrfs_path *path)
2470 struct btrfs_root *root;
2477 * If we fail here we want to drop our backref_node because we are going
2478 * to start over and regenerate the tree for it.
2480 ret = reserve_metadata_space(trans, rc, node);
2484 BUG_ON(node->processed);
2485 root = select_one_root(node);
2486 if (root == ERR_PTR(-ENOENT)) {
2487 update_processed_blocks(rc, node);
2492 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2493 BUG_ON(node->new_bytenr);
2494 BUG_ON(!list_empty(&node->list));
2495 btrfs_record_root_in_trans(trans, root);
2496 root = root->reloc_root;
2497 node->new_bytenr = root->node->start;
2498 btrfs_put_root(node->root);
2499 node->root = btrfs_grab_root(root);
2501 list_add_tail(&node->list, &rc->backref_cache.changed);
2503 path->lowest_level = node->level;
2504 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2505 btrfs_release_path(path);
2510 update_processed_blocks(rc, node);
2512 ret = do_relocation(trans, rc, node, key, path, 1);
2515 if (ret || node->level == 0 || node->cowonly)
2516 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2521 * relocate a list of blocks
2523 static noinline_for_stack
2524 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2525 struct reloc_control *rc, struct rb_root *blocks)
2527 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2528 struct btrfs_backref_node *node;
2529 struct btrfs_path *path;
2530 struct tree_block *block;
2531 struct tree_block *next;
2535 path = btrfs_alloc_path();
2538 goto out_free_blocks;
2541 /* Kick in readahead for tree blocks with missing keys */
2542 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2543 if (!block->key_ready)
2544 readahead_tree_block(fs_info, block->bytenr);
2547 /* Get first keys */
2548 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2549 if (!block->key_ready) {
2550 err = get_tree_block_key(fs_info, block);
2556 /* Do tree relocation */
2557 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2558 node = build_backref_tree(rc, &block->key,
2559 block->level, block->bytenr);
2561 err = PTR_ERR(node);
2565 ret = relocate_tree_block(trans, rc, node, &block->key,
2573 err = finish_pending_nodes(trans, rc, path, err);
2576 btrfs_free_path(path);
2578 free_block_list(blocks);
2582 static noinline_for_stack int prealloc_file_extent_cluster(
2583 struct btrfs_inode *inode,
2584 struct file_extent_cluster *cluster)
2589 u64 offset = inode->index_cnt;
2593 u64 prealloc_start = cluster->start - offset;
2594 u64 prealloc_end = cluster->end - offset;
2595 u64 cur_offset = prealloc_start;
2597 BUG_ON(cluster->start != cluster->boundary[0]);
2598 ret = btrfs_alloc_data_chunk_ondemand(inode,
2599 prealloc_end + 1 - prealloc_start);
2603 inode_lock(&inode->vfs_inode);
2604 for (nr = 0; nr < cluster->nr; nr++) {
2605 start = cluster->boundary[nr] - offset;
2606 if (nr + 1 < cluster->nr)
2607 end = cluster->boundary[nr + 1] - 1 - offset;
2609 end = cluster->end - offset;
2611 lock_extent(&inode->io_tree, start, end);
2612 num_bytes = end + 1 - start;
2613 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2614 num_bytes, num_bytes,
2615 end + 1, &alloc_hint);
2616 cur_offset = end + 1;
2617 unlock_extent(&inode->io_tree, start, end);
2621 inode_unlock(&inode->vfs_inode);
2623 if (cur_offset < prealloc_end)
2624 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2625 prealloc_end + 1 - cur_offset);
2629 static noinline_for_stack
2630 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2633 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2634 struct extent_map *em;
2637 em = alloc_extent_map();
2642 em->len = end + 1 - start;
2643 em->block_len = em->len;
2644 em->block_start = block_start;
2645 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2647 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2649 write_lock(&em_tree->lock);
2650 ret = add_extent_mapping(em_tree, em, 0);
2651 write_unlock(&em_tree->lock);
2652 if (ret != -EEXIST) {
2653 free_extent_map(em);
2656 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2658 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2663 * Allow error injection to test balance cancellation
2665 int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2667 return atomic_read(&fs_info->balance_cancel_req) ||
2668 fatal_signal_pending(current);
2670 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2672 static int relocate_file_extent_cluster(struct inode *inode,
2673 struct file_extent_cluster *cluster)
2675 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2678 u64 offset = BTRFS_I(inode)->index_cnt;
2679 unsigned long index;
2680 unsigned long last_index;
2682 struct file_ra_state *ra;
2683 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2690 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2694 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2698 file_ra_state_init(ra, inode->i_mapping);
2700 ret = setup_extent_mapping(inode, cluster->start - offset,
2701 cluster->end - offset, cluster->start);
2705 index = (cluster->start - offset) >> PAGE_SHIFT;
2706 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2707 while (index <= last_index) {
2708 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2713 page = find_lock_page(inode->i_mapping, index);
2715 page_cache_sync_readahead(inode->i_mapping,
2717 last_index + 1 - index);
2718 page = find_or_create_page(inode->i_mapping, index,
2721 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2723 btrfs_delalloc_release_extents(BTRFS_I(inode),
2730 if (PageReadahead(page)) {
2731 page_cache_async_readahead(inode->i_mapping,
2732 ra, NULL, page, index,
2733 last_index + 1 - index);
2736 if (!PageUptodate(page)) {
2737 btrfs_readpage(NULL, page);
2739 if (!PageUptodate(page)) {
2742 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2744 btrfs_delalloc_release_extents(BTRFS_I(inode),
2751 page_start = page_offset(page);
2752 page_end = page_start + PAGE_SIZE - 1;
2754 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2756 set_page_extent_mapped(page);
2758 if (nr < cluster->nr &&
2759 page_start + offset == cluster->boundary[nr]) {
2760 set_extent_bits(&BTRFS_I(inode)->io_tree,
2761 page_start, page_end,
2766 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2771 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2773 btrfs_delalloc_release_extents(BTRFS_I(inode),
2776 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2777 page_start, page_end,
2778 EXTENT_LOCKED | EXTENT_BOUNDARY);
2782 set_page_dirty(page);
2784 unlock_extent(&BTRFS_I(inode)->io_tree,
2785 page_start, page_end);
2790 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2791 balance_dirty_pages_ratelimited(inode->i_mapping);
2792 btrfs_throttle(fs_info);
2793 if (btrfs_should_cancel_balance(fs_info)) {
2798 WARN_ON(nr != cluster->nr);
2804 static noinline_for_stack
2805 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2806 struct file_extent_cluster *cluster)
2810 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2811 ret = relocate_file_extent_cluster(inode, cluster);
2818 cluster->start = extent_key->objectid;
2820 BUG_ON(cluster->nr >= MAX_EXTENTS);
2821 cluster->end = extent_key->objectid + extent_key->offset - 1;
2822 cluster->boundary[cluster->nr] = extent_key->objectid;
2825 if (cluster->nr >= MAX_EXTENTS) {
2826 ret = relocate_file_extent_cluster(inode, cluster);
2835 * helper to add a tree block to the list.
2836 * the major work is getting the generation and level of the block
2838 static int add_tree_block(struct reloc_control *rc,
2839 struct btrfs_key *extent_key,
2840 struct btrfs_path *path,
2841 struct rb_root *blocks)
2843 struct extent_buffer *eb;
2844 struct btrfs_extent_item *ei;
2845 struct btrfs_tree_block_info *bi;
2846 struct tree_block *block;
2847 struct rb_node *rb_node;
2852 eb = path->nodes[0];
2853 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2855 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2856 item_size >= sizeof(*ei) + sizeof(*bi)) {
2857 ei = btrfs_item_ptr(eb, path->slots[0],
2858 struct btrfs_extent_item);
2859 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2860 bi = (struct btrfs_tree_block_info *)(ei + 1);
2861 level = btrfs_tree_block_level(eb, bi);
2863 level = (int)extent_key->offset;
2865 generation = btrfs_extent_generation(eb, ei);
2866 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2867 btrfs_print_v0_err(eb->fs_info);
2868 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2874 btrfs_release_path(path);
2876 BUG_ON(level == -1);
2878 block = kmalloc(sizeof(*block), GFP_NOFS);
2882 block->bytenr = extent_key->objectid;
2883 block->key.objectid = rc->extent_root->fs_info->nodesize;
2884 block->key.offset = generation;
2885 block->level = level;
2886 block->key_ready = 0;
2888 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2890 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2897 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2899 static int __add_tree_block(struct reloc_control *rc,
2900 u64 bytenr, u32 blocksize,
2901 struct rb_root *blocks)
2903 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2904 struct btrfs_path *path;
2905 struct btrfs_key key;
2907 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2909 if (tree_block_processed(bytenr, rc))
2912 if (rb_simple_search(blocks, bytenr))
2915 path = btrfs_alloc_path();
2919 key.objectid = bytenr;
2921 key.type = BTRFS_METADATA_ITEM_KEY;
2922 key.offset = (u64)-1;
2924 key.type = BTRFS_EXTENT_ITEM_KEY;
2925 key.offset = blocksize;
2928 path->search_commit_root = 1;
2929 path->skip_locking = 1;
2930 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2934 if (ret > 0 && skinny) {
2935 if (path->slots[0]) {
2937 btrfs_item_key_to_cpu(path->nodes[0], &key,
2939 if (key.objectid == bytenr &&
2940 (key.type == BTRFS_METADATA_ITEM_KEY ||
2941 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2942 key.offset == blocksize)))
2948 btrfs_release_path(path);
2954 btrfs_print_leaf(path->nodes[0]);
2956 "tree block extent item (%llu) is not found in extent tree",
2963 ret = add_tree_block(rc, &key, path, blocks);
2965 btrfs_free_path(path);
2969 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2970 struct btrfs_block_group *block_group,
2971 struct inode *inode,
2974 struct btrfs_root *root = fs_info->tree_root;
2975 struct btrfs_trans_handle *trans;
2981 inode = btrfs_iget(fs_info->sb, ino, root);
2986 ret = btrfs_check_trunc_cache_free_space(fs_info,
2987 &fs_info->global_block_rsv);
2991 trans = btrfs_join_transaction(root);
2992 if (IS_ERR(trans)) {
2993 ret = PTR_ERR(trans);
2997 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
2999 btrfs_end_transaction(trans);
3000 btrfs_btree_balance_dirty(fs_info);
3007 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3008 * cache inode, to avoid free space cache data extent blocking data relocation.
3010 static int delete_v1_space_cache(struct extent_buffer *leaf,
3011 struct btrfs_block_group *block_group,
3014 u64 space_cache_ino;
3015 struct btrfs_file_extent_item *ei;
3016 struct btrfs_key key;
3021 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3024 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3025 btrfs_item_key_to_cpu(leaf, &key, i);
3026 if (key.type != BTRFS_EXTENT_DATA_KEY)
3028 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3029 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG &&
3030 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3032 space_cache_ino = key.objectid;
3038 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3044 * helper to find all tree blocks that reference a given data extent
3046 static noinline_for_stack
3047 int add_data_references(struct reloc_control *rc,
3048 struct btrfs_key *extent_key,
3049 struct btrfs_path *path,
3050 struct rb_root *blocks)
3052 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3053 struct ulist *leaves = NULL;
3054 struct ulist_iterator leaf_uiter;
3055 struct ulist_node *ref_node = NULL;
3056 const u32 blocksize = fs_info->nodesize;
3059 btrfs_release_path(path);
3060 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3061 0, &leaves, NULL, true);
3065 ULIST_ITER_INIT(&leaf_uiter);
3066 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3067 struct extent_buffer *eb;
3069 eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL);
3074 ret = delete_v1_space_cache(eb, rc->block_group,
3075 extent_key->objectid);
3076 free_extent_buffer(eb);
3079 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3084 free_block_list(blocks);
3090 * helper to find next unprocessed extent
3092 static noinline_for_stack
3093 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3094 struct btrfs_key *extent_key)
3096 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3097 struct btrfs_key key;
3098 struct extent_buffer *leaf;
3099 u64 start, end, last;
3102 last = rc->block_group->start + rc->block_group->length;
3105 if (rc->search_start >= last) {
3110 key.objectid = rc->search_start;
3111 key.type = BTRFS_EXTENT_ITEM_KEY;
3114 path->search_commit_root = 1;
3115 path->skip_locking = 1;
3116 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3121 leaf = path->nodes[0];
3122 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3123 ret = btrfs_next_leaf(rc->extent_root, path);
3126 leaf = path->nodes[0];
3129 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3130 if (key.objectid >= last) {
3135 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3136 key.type != BTRFS_METADATA_ITEM_KEY) {
3141 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3142 key.objectid + key.offset <= rc->search_start) {
3147 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3148 key.objectid + fs_info->nodesize <=
3154 ret = find_first_extent_bit(&rc->processed_blocks,
3155 key.objectid, &start, &end,
3156 EXTENT_DIRTY, NULL);
3158 if (ret == 0 && start <= key.objectid) {
3159 btrfs_release_path(path);
3160 rc->search_start = end + 1;
3162 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3163 rc->search_start = key.objectid + key.offset;
3165 rc->search_start = key.objectid +
3167 memcpy(extent_key, &key, sizeof(key));
3171 btrfs_release_path(path);
3175 static void set_reloc_control(struct reloc_control *rc)
3177 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3179 mutex_lock(&fs_info->reloc_mutex);
3180 fs_info->reloc_ctl = rc;
3181 mutex_unlock(&fs_info->reloc_mutex);
3184 static void unset_reloc_control(struct reloc_control *rc)
3186 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3188 mutex_lock(&fs_info->reloc_mutex);
3189 fs_info->reloc_ctl = NULL;
3190 mutex_unlock(&fs_info->reloc_mutex);
3193 static int check_extent_flags(u64 flags)
3195 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3196 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3198 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3199 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3201 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3202 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3207 static noinline_for_stack
3208 int prepare_to_relocate(struct reloc_control *rc)
3210 struct btrfs_trans_handle *trans;
3213 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3214 BTRFS_BLOCK_RSV_TEMP);
3218 memset(&rc->cluster, 0, sizeof(rc->cluster));
3219 rc->search_start = rc->block_group->start;
3220 rc->extents_found = 0;
3221 rc->nodes_relocated = 0;
3222 rc->merging_rsv_size = 0;
3223 rc->reserved_bytes = 0;
3224 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3225 RELOCATION_RESERVED_NODES;
3226 ret = btrfs_block_rsv_refill(rc->extent_root,
3227 rc->block_rsv, rc->block_rsv->size,
3228 BTRFS_RESERVE_FLUSH_ALL);
3232 rc->create_reloc_tree = 1;
3233 set_reloc_control(rc);
3235 trans = btrfs_join_transaction(rc->extent_root);
3236 if (IS_ERR(trans)) {
3237 unset_reloc_control(rc);
3239 * extent tree is not a ref_cow tree and has no reloc_root to
3240 * cleanup. And callers are responsible to free the above
3243 return PTR_ERR(trans);
3245 btrfs_commit_transaction(trans);
3249 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3251 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3252 struct rb_root blocks = RB_ROOT;
3253 struct btrfs_key key;
3254 struct btrfs_trans_handle *trans = NULL;
3255 struct btrfs_path *path;
3256 struct btrfs_extent_item *ei;
3263 path = btrfs_alloc_path();
3266 path->reada = READA_FORWARD;
3268 ret = prepare_to_relocate(rc);
3275 rc->reserved_bytes = 0;
3276 ret = btrfs_block_rsv_refill(rc->extent_root,
3277 rc->block_rsv, rc->block_rsv->size,
3278 BTRFS_RESERVE_FLUSH_ALL);
3284 trans = btrfs_start_transaction(rc->extent_root, 0);
3285 if (IS_ERR(trans)) {
3286 err = PTR_ERR(trans);
3291 if (update_backref_cache(trans, &rc->backref_cache)) {
3292 btrfs_end_transaction(trans);
3297 ret = find_next_extent(rc, path, &key);
3303 rc->extents_found++;
3305 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3306 struct btrfs_extent_item);
3307 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3308 if (item_size >= sizeof(*ei)) {
3309 flags = btrfs_extent_flags(path->nodes[0], ei);
3310 ret = check_extent_flags(flags);
3312 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3314 btrfs_print_v0_err(trans->fs_info);
3315 btrfs_abort_transaction(trans, err);
3321 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3322 ret = add_tree_block(rc, &key, path, &blocks);
3323 } else if (rc->stage == UPDATE_DATA_PTRS &&
3324 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3325 ret = add_data_references(rc, &key, path, &blocks);
3327 btrfs_release_path(path);
3335 if (!RB_EMPTY_ROOT(&blocks)) {
3336 ret = relocate_tree_blocks(trans, rc, &blocks);
3338 if (ret != -EAGAIN) {
3342 rc->extents_found--;
3343 rc->search_start = key.objectid;
3347 btrfs_end_transaction_throttle(trans);
3348 btrfs_btree_balance_dirty(fs_info);
3351 if (rc->stage == MOVE_DATA_EXTENTS &&
3352 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3353 rc->found_file_extent = 1;
3354 ret = relocate_data_extent(rc->data_inode,
3355 &key, &rc->cluster);
3361 if (btrfs_should_cancel_balance(fs_info)) {
3366 if (trans && progress && err == -ENOSPC) {
3367 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3375 btrfs_release_path(path);
3376 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3379 btrfs_end_transaction_throttle(trans);
3380 btrfs_btree_balance_dirty(fs_info);
3384 ret = relocate_file_extent_cluster(rc->data_inode,
3390 rc->create_reloc_tree = 0;
3391 set_reloc_control(rc);
3393 btrfs_backref_release_cache(&rc->backref_cache);
3394 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3397 * Even in the case when the relocation is cancelled, we should all go
3398 * through prepare_to_merge() and merge_reloc_roots().
3400 * For error (including cancelled balance), prepare_to_merge() will
3401 * mark all reloc trees orphan, then queue them for cleanup in
3402 * merge_reloc_roots()
3404 err = prepare_to_merge(rc, err);
3406 merge_reloc_roots(rc);
3408 rc->merge_reloc_tree = 0;
3409 unset_reloc_control(rc);
3410 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3412 /* get rid of pinned extents */
3413 trans = btrfs_join_transaction(rc->extent_root);
3414 if (IS_ERR(trans)) {
3415 err = PTR_ERR(trans);
3418 btrfs_commit_transaction(trans);
3420 ret = clean_dirty_subvols(rc);
3421 if (ret < 0 && !err)
3423 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3424 btrfs_free_path(path);
3428 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3429 struct btrfs_root *root, u64 objectid)
3431 struct btrfs_path *path;
3432 struct btrfs_inode_item *item;
3433 struct extent_buffer *leaf;
3436 path = btrfs_alloc_path();
3440 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3444 leaf = path->nodes[0];
3445 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3446 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3447 btrfs_set_inode_generation(leaf, item, 1);
3448 btrfs_set_inode_size(leaf, item, 0);
3449 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3450 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3451 BTRFS_INODE_PREALLOC);
3452 btrfs_mark_buffer_dirty(leaf);
3454 btrfs_free_path(path);
3459 * helper to create inode for data relocation.
3460 * the inode is in data relocation tree and its link count is 0
3462 static noinline_for_stack
3463 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3464 struct btrfs_block_group *group)
3466 struct inode *inode = NULL;
3467 struct btrfs_trans_handle *trans;
3468 struct btrfs_root *root;
3472 root = btrfs_grab_root(fs_info->data_reloc_root);
3473 trans = btrfs_start_transaction(root, 6);
3474 if (IS_ERR(trans)) {
3475 btrfs_put_root(root);
3476 return ERR_CAST(trans);
3479 err = btrfs_find_free_objectid(root, &objectid);
3483 err = __insert_orphan_inode(trans, root, objectid);
3486 inode = btrfs_iget(fs_info->sb, objectid, root);
3487 BUG_ON(IS_ERR(inode));
3488 BTRFS_I(inode)->index_cnt = group->start;
3490 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3492 btrfs_put_root(root);
3493 btrfs_end_transaction(trans);
3494 btrfs_btree_balance_dirty(fs_info);
3498 inode = ERR_PTR(err);
3503 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3505 struct reloc_control *rc;
3507 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3511 INIT_LIST_HEAD(&rc->reloc_roots);
3512 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3513 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3514 mapping_tree_init(&rc->reloc_root_tree);
3515 extent_io_tree_init(fs_info, &rc->processed_blocks,
3516 IO_TREE_RELOC_BLOCKS, NULL);
3520 static void free_reloc_control(struct reloc_control *rc)
3522 struct mapping_node *node, *tmp;
3524 free_reloc_roots(&rc->reloc_roots);
3525 rbtree_postorder_for_each_entry_safe(node, tmp,
3526 &rc->reloc_root_tree.rb_root, rb_node)
3533 * Print the block group being relocated
3535 static void describe_relocation(struct btrfs_fs_info *fs_info,
3536 struct btrfs_block_group *block_group)
3538 char buf[128] = {'\0'};
3540 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3543 "relocating block group %llu flags %s",
3544 block_group->start, buf);
3547 static const char *stage_to_string(int stage)
3549 if (stage == MOVE_DATA_EXTENTS)
3550 return "move data extents";
3551 if (stage == UPDATE_DATA_PTRS)
3552 return "update data pointers";
3557 * function to relocate all extents in a block group.
3559 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3561 struct btrfs_block_group *bg;
3562 struct btrfs_root *extent_root = fs_info->extent_root;
3563 struct reloc_control *rc;
3564 struct inode *inode;
3565 struct btrfs_path *path;
3570 bg = btrfs_lookup_block_group(fs_info, group_start);
3574 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3575 btrfs_put_block_group(bg);
3579 rc = alloc_reloc_control(fs_info);
3581 btrfs_put_block_group(bg);
3585 rc->extent_root = extent_root;
3586 rc->block_group = bg;
3588 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3595 path = btrfs_alloc_path();
3601 inode = lookup_free_space_inode(rc->block_group, path);
3602 btrfs_free_path(path);
3605 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3607 ret = PTR_ERR(inode);
3609 if (ret && ret != -ENOENT) {
3614 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3615 if (IS_ERR(rc->data_inode)) {
3616 err = PTR_ERR(rc->data_inode);
3617 rc->data_inode = NULL;
3621 describe_relocation(fs_info, rc->block_group);
3623 btrfs_wait_block_group_reservations(rc->block_group);
3624 btrfs_wait_nocow_writers(rc->block_group);
3625 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3626 rc->block_group->start,
3627 rc->block_group->length);
3632 mutex_lock(&fs_info->cleaner_mutex);
3633 ret = relocate_block_group(rc);
3634 mutex_unlock(&fs_info->cleaner_mutex);
3638 finishes_stage = rc->stage;
3640 * We may have gotten ENOSPC after we already dirtied some
3641 * extents. If writeout happens while we're relocating a
3642 * different block group we could end up hitting the
3643 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3644 * btrfs_reloc_cow_block. Make sure we write everything out
3645 * properly so we don't trip over this problem, and then break
3646 * out of the loop if we hit an error.
3648 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3649 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3653 invalidate_mapping_pages(rc->data_inode->i_mapping,
3655 rc->stage = UPDATE_DATA_PTRS;
3661 if (rc->extents_found == 0)
3664 btrfs_info(fs_info, "found %llu extents, stage: %s",
3665 rc->extents_found, stage_to_string(finishes_stage));
3668 WARN_ON(rc->block_group->pinned > 0);
3669 WARN_ON(rc->block_group->reserved > 0);
3670 WARN_ON(rc->block_group->used > 0);
3673 btrfs_dec_block_group_ro(rc->block_group);
3674 iput(rc->data_inode);
3675 btrfs_put_block_group(rc->block_group);
3676 free_reloc_control(rc);
3680 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3682 struct btrfs_fs_info *fs_info = root->fs_info;
3683 struct btrfs_trans_handle *trans;
3686 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3688 return PTR_ERR(trans);
3690 memset(&root->root_item.drop_progress, 0,
3691 sizeof(root->root_item.drop_progress));
3692 root->root_item.drop_level = 0;
3693 btrfs_set_root_refs(&root->root_item, 0);
3694 ret = btrfs_update_root(trans, fs_info->tree_root,
3695 &root->root_key, &root->root_item);
3697 err = btrfs_end_transaction(trans);
3704 * recover relocation interrupted by system crash.
3706 * this function resumes merging reloc trees with corresponding fs trees.
3707 * this is important for keeping the sharing of tree blocks
3709 int btrfs_recover_relocation(struct btrfs_root *root)
3711 struct btrfs_fs_info *fs_info = root->fs_info;
3712 LIST_HEAD(reloc_roots);
3713 struct btrfs_key key;
3714 struct btrfs_root *fs_root;
3715 struct btrfs_root *reloc_root;
3716 struct btrfs_path *path;
3717 struct extent_buffer *leaf;
3718 struct reloc_control *rc = NULL;
3719 struct btrfs_trans_handle *trans;
3723 path = btrfs_alloc_path();
3726 path->reada = READA_BACK;
3728 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3729 key.type = BTRFS_ROOT_ITEM_KEY;
3730 key.offset = (u64)-1;
3733 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3740 if (path->slots[0] == 0)
3744 leaf = path->nodes[0];
3745 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3746 btrfs_release_path(path);
3748 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3749 key.type != BTRFS_ROOT_ITEM_KEY)
3752 reloc_root = btrfs_read_tree_root(root, &key);
3753 if (IS_ERR(reloc_root)) {
3754 err = PTR_ERR(reloc_root);
3758 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3759 list_add(&reloc_root->root_list, &reloc_roots);
3761 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3762 fs_root = btrfs_get_fs_root(fs_info,
3763 reloc_root->root_key.offset, false);
3764 if (IS_ERR(fs_root)) {
3765 ret = PTR_ERR(fs_root);
3766 if (ret != -ENOENT) {
3770 ret = mark_garbage_root(reloc_root);
3776 btrfs_put_root(fs_root);
3780 if (key.offset == 0)
3785 btrfs_release_path(path);
3787 if (list_empty(&reloc_roots))
3790 rc = alloc_reloc_control(fs_info);
3796 rc->extent_root = fs_info->extent_root;
3798 set_reloc_control(rc);
3800 trans = btrfs_join_transaction(rc->extent_root);
3801 if (IS_ERR(trans)) {
3802 err = PTR_ERR(trans);
3806 rc->merge_reloc_tree = 1;
3808 while (!list_empty(&reloc_roots)) {
3809 reloc_root = list_entry(reloc_roots.next,
3810 struct btrfs_root, root_list);
3811 list_del(&reloc_root->root_list);
3813 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3814 list_add_tail(&reloc_root->root_list,
3819 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
3821 if (IS_ERR(fs_root)) {
3822 err = PTR_ERR(fs_root);
3823 list_add_tail(&reloc_root->root_list, &reloc_roots);
3824 btrfs_end_transaction(trans);
3828 err = __add_reloc_root(reloc_root);
3829 BUG_ON(err < 0); /* -ENOMEM or logic error */
3830 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3831 btrfs_put_root(fs_root);
3834 err = btrfs_commit_transaction(trans);
3838 merge_reloc_roots(rc);
3840 unset_reloc_control(rc);
3842 trans = btrfs_join_transaction(rc->extent_root);
3843 if (IS_ERR(trans)) {
3844 err = PTR_ERR(trans);
3847 err = btrfs_commit_transaction(trans);
3849 ret = clean_dirty_subvols(rc);
3850 if (ret < 0 && !err)
3853 unset_reloc_control(rc);
3854 free_reloc_control(rc);
3856 free_reloc_roots(&reloc_roots);
3858 btrfs_free_path(path);
3861 /* cleanup orphan inode in data relocation tree */
3862 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
3864 err = btrfs_orphan_cleanup(fs_root);
3865 btrfs_put_root(fs_root);
3871 * helper to add ordered checksum for data relocation.
3873 * cloning checksum properly handles the nodatasum extents.
3874 * it also saves CPU time to re-calculate the checksum.
3876 int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
3878 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3879 struct btrfs_ordered_sum *sums;
3880 struct btrfs_ordered_extent *ordered;
3886 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3887 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3889 disk_bytenr = file_pos + inode->index_cnt;
3890 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3891 disk_bytenr + len - 1, &list, 0);
3895 while (!list_empty(&list)) {
3896 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3897 list_del_init(&sums->list);
3900 * We need to offset the new_bytenr based on where the csum is.
3901 * We need to do this because we will read in entire prealloc
3902 * extents but we may have written to say the middle of the
3903 * prealloc extent, so we need to make sure the csum goes with
3904 * the right disk offset.
3906 * We can do this because the data reloc inode refers strictly
3907 * to the on disk bytes, so we don't have to worry about
3908 * disk_len vs real len like with real inodes since it's all
3911 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3912 sums->bytenr = new_bytenr;
3914 btrfs_add_ordered_sum(ordered, sums);
3917 btrfs_put_ordered_extent(ordered);
3921 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3922 struct btrfs_root *root, struct extent_buffer *buf,
3923 struct extent_buffer *cow)
3925 struct btrfs_fs_info *fs_info = root->fs_info;
3926 struct reloc_control *rc;
3927 struct btrfs_backref_node *node;
3932 rc = fs_info->reloc_ctl;
3936 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3937 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3939 level = btrfs_header_level(buf);
3940 if (btrfs_header_generation(buf) <=
3941 btrfs_root_last_snapshot(&root->root_item))
3944 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3945 rc->create_reloc_tree) {
3946 WARN_ON(!first_cow && level == 0);
3948 node = rc->backref_cache.path[level];
3949 BUG_ON(node->bytenr != buf->start &&
3950 node->new_bytenr != buf->start);
3952 btrfs_backref_drop_node_buffer(node);
3953 atomic_inc(&cow->refs);
3955 node->new_bytenr = cow->start;
3957 if (!node->pending) {
3958 list_move_tail(&node->list,
3959 &rc->backref_cache.pending[level]);
3964 mark_block_processed(rc, node);
3966 if (first_cow && level > 0)
3967 rc->nodes_relocated += buf->len;
3970 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
3971 ret = replace_file_extents(trans, rc, root, cow);
3976 * called before creating snapshot. it calculates metadata reservation
3977 * required for relocating tree blocks in the snapshot
3979 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
3980 u64 *bytes_to_reserve)
3982 struct btrfs_root *root = pending->root;
3983 struct reloc_control *rc = root->fs_info->reloc_ctl;
3985 if (!rc || !have_reloc_root(root))
3988 if (!rc->merge_reloc_tree)
3991 root = root->reloc_root;
3992 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
3994 * relocation is in the stage of merging trees. the space
3995 * used by merging a reloc tree is twice the size of
3996 * relocated tree nodes in the worst case. half for cowing
3997 * the reloc tree, half for cowing the fs tree. the space
3998 * used by cowing the reloc tree will be freed after the
3999 * tree is dropped. if we create snapshot, cowing the fs
4000 * tree may use more space than it frees. so we need
4001 * reserve extra space.
4003 *bytes_to_reserve += rc->nodes_relocated;
4007 * called after snapshot is created. migrate block reservation
4008 * and create reloc root for the newly created snapshot
4010 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4011 * references held on the reloc_root, one for root->reloc_root and one for
4014 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4015 struct btrfs_pending_snapshot *pending)
4017 struct btrfs_root *root = pending->root;
4018 struct btrfs_root *reloc_root;
4019 struct btrfs_root *new_root;
4020 struct reloc_control *rc = root->fs_info->reloc_ctl;
4023 if (!rc || !have_reloc_root(root))
4026 rc = root->fs_info->reloc_ctl;
4027 rc->merging_rsv_size += rc->nodes_relocated;
4029 if (rc->merge_reloc_tree) {
4030 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4032 rc->nodes_relocated, true);
4037 new_root = pending->snap;
4038 reloc_root = create_reloc_root(trans, root->reloc_root,
4039 new_root->root_key.objectid);
4040 if (IS_ERR(reloc_root))
4041 return PTR_ERR(reloc_root);
4043 ret = __add_reloc_root(reloc_root);
4045 new_root->reloc_root = btrfs_grab_root(reloc_root);
4047 if (rc->create_reloc_tree)
4048 ret = clone_backref_node(trans, rc, root, reloc_root);