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);
642 "Duplicate root found for start=%llu while inserting into relocation tree",
647 list_add_tail(&root->root_list, &rc->reloc_roots);
652 * helper to delete the 'address of tree root -> reloc tree'
655 static void __del_reloc_root(struct btrfs_root *root)
657 struct btrfs_fs_info *fs_info = root->fs_info;
658 struct rb_node *rb_node;
659 struct mapping_node *node = NULL;
660 struct reloc_control *rc = fs_info->reloc_ctl;
661 bool put_ref = false;
663 if (rc && root->node) {
664 spin_lock(&rc->reloc_root_tree.lock);
665 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
666 root->commit_root->start);
668 node = rb_entry(rb_node, struct mapping_node, rb_node);
669 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
670 RB_CLEAR_NODE(&node->rb_node);
672 spin_unlock(&rc->reloc_root_tree.lock);
673 ASSERT(!node || (struct btrfs_root *)node->data == root);
677 * We only put the reloc root here if it's on the list. There's a lot
678 * of places where the pattern is to splice the rc->reloc_roots, process
679 * the reloc roots, and then add the reloc root back onto
680 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
681 * list we don't want the reference being dropped, because the guy
682 * messing with the list is in charge of the reference.
684 spin_lock(&fs_info->trans_lock);
685 if (!list_empty(&root->root_list)) {
687 list_del_init(&root->root_list);
689 spin_unlock(&fs_info->trans_lock);
691 btrfs_put_root(root);
696 * helper to update the 'address of tree root -> reloc tree'
699 static int __update_reloc_root(struct btrfs_root *root)
701 struct btrfs_fs_info *fs_info = root->fs_info;
702 struct rb_node *rb_node;
703 struct mapping_node *node = NULL;
704 struct reloc_control *rc = fs_info->reloc_ctl;
706 spin_lock(&rc->reloc_root_tree.lock);
707 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
708 root->commit_root->start);
710 node = rb_entry(rb_node, struct mapping_node, rb_node);
711 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
713 spin_unlock(&rc->reloc_root_tree.lock);
717 BUG_ON((struct btrfs_root *)node->data != root);
719 spin_lock(&rc->reloc_root_tree.lock);
720 node->bytenr = root->node->start;
721 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
722 node->bytenr, &node->rb_node);
723 spin_unlock(&rc->reloc_root_tree.lock);
725 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
729 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
730 struct btrfs_root *root, u64 objectid)
732 struct btrfs_fs_info *fs_info = root->fs_info;
733 struct btrfs_root *reloc_root;
734 struct extent_buffer *eb;
735 struct btrfs_root_item *root_item;
736 struct btrfs_key root_key;
738 bool must_abort = false;
740 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
742 return ERR_PTR(-ENOMEM);
744 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
745 root_key.type = BTRFS_ROOT_ITEM_KEY;
746 root_key.offset = objectid;
748 if (root->root_key.objectid == objectid) {
751 /* called by btrfs_init_reloc_root */
752 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
753 BTRFS_TREE_RELOC_OBJECTID);
758 * Set the last_snapshot field to the generation of the commit
759 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
760 * correctly (returns true) when the relocation root is created
761 * either inside the critical section of a transaction commit
762 * (through transaction.c:qgroup_account_snapshot()) and when
763 * it's created before the transaction commit is started.
765 commit_root_gen = btrfs_header_generation(root->commit_root);
766 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
769 * called by btrfs_reloc_post_snapshot_hook.
770 * the source tree is a reloc tree, all tree blocks
771 * modified after it was created have RELOC flag
772 * set in their headers. so it's OK to not update
773 * the 'last_snapshot'.
775 ret = btrfs_copy_root(trans, root, root->node, &eb,
776 BTRFS_TREE_RELOC_OBJECTID);
782 * We have changed references at this point, we must abort the
783 * transaction if anything fails.
787 memcpy(root_item, &root->root_item, sizeof(*root_item));
788 btrfs_set_root_bytenr(root_item, eb->start);
789 btrfs_set_root_level(root_item, btrfs_header_level(eb));
790 btrfs_set_root_generation(root_item, trans->transid);
792 if (root->root_key.objectid == objectid) {
793 btrfs_set_root_refs(root_item, 0);
794 memset(&root_item->drop_progress, 0,
795 sizeof(struct btrfs_disk_key));
796 btrfs_set_root_drop_level(root_item, 0);
799 btrfs_tree_unlock(eb);
800 free_extent_buffer(eb);
802 ret = btrfs_insert_root(trans, fs_info->tree_root,
803 &root_key, root_item);
809 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
810 if (IS_ERR(reloc_root)) {
811 ret = PTR_ERR(reloc_root);
814 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
815 reloc_root->last_trans = trans->transid;
821 btrfs_abort_transaction(trans, ret);
826 * create reloc tree for a given fs tree. reloc tree is just a
827 * snapshot of the fs tree with special root objectid.
829 * The reloc_root comes out of here with two references, one for
830 * root->reloc_root, and another for being on the rc->reloc_roots list.
832 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
833 struct btrfs_root *root)
835 struct btrfs_fs_info *fs_info = root->fs_info;
836 struct btrfs_root *reloc_root;
837 struct reloc_control *rc = fs_info->reloc_ctl;
838 struct btrfs_block_rsv *rsv;
846 * The subvolume has reloc tree but the swap is finished, no need to
847 * create/update the dead reloc tree
849 if (reloc_root_is_dead(root))
853 * This is subtle but important. We do not do
854 * record_root_in_transaction for reloc roots, instead we record their
855 * corresponding fs root, and then here we update the last trans for the
856 * reloc root. This means that we have to do this for the entire life
857 * of the reloc root, regardless of which stage of the relocation we are
860 if (root->reloc_root) {
861 reloc_root = root->reloc_root;
862 reloc_root->last_trans = trans->transid;
867 * We are merging reloc roots, we do not need new reloc trees. Also
868 * reloc trees never need their own reloc tree.
870 if (!rc->create_reloc_tree ||
871 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
874 if (!trans->reloc_reserved) {
875 rsv = trans->block_rsv;
876 trans->block_rsv = rc->block_rsv;
879 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
881 trans->block_rsv = rsv;
882 if (IS_ERR(reloc_root))
883 return PTR_ERR(reloc_root);
885 ret = __add_reloc_root(reloc_root);
886 ASSERT(ret != -EEXIST);
888 /* Pairs with create_reloc_root */
889 btrfs_put_root(reloc_root);
892 root->reloc_root = btrfs_grab_root(reloc_root);
897 * update root item of reloc tree
899 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
900 struct btrfs_root *root)
902 struct btrfs_fs_info *fs_info = root->fs_info;
903 struct btrfs_root *reloc_root;
904 struct btrfs_root_item *root_item;
907 if (!have_reloc_root(root))
910 reloc_root = root->reloc_root;
911 root_item = &reloc_root->root_item;
914 * We are probably ok here, but __del_reloc_root() will drop its ref of
915 * the root. We have the ref for root->reloc_root, but just in case
916 * hold it while we update the reloc root.
918 btrfs_grab_root(reloc_root);
920 /* root->reloc_root will stay until current relocation finished */
921 if (fs_info->reloc_ctl->merge_reloc_tree &&
922 btrfs_root_refs(root_item) == 0) {
923 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
925 * Mark the tree as dead before we change reloc_root so
926 * have_reloc_root will not touch it from now on.
929 __del_reloc_root(reloc_root);
932 if (reloc_root->commit_root != reloc_root->node) {
933 __update_reloc_root(reloc_root);
934 btrfs_set_root_node(root_item, reloc_root->node);
935 free_extent_buffer(reloc_root->commit_root);
936 reloc_root->commit_root = btrfs_root_node(reloc_root);
939 ret = btrfs_update_root(trans, fs_info->tree_root,
940 &reloc_root->root_key, root_item);
941 btrfs_put_root(reloc_root);
946 * helper to find first cached inode with inode number >= objectid
949 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
951 struct rb_node *node;
952 struct rb_node *prev;
953 struct btrfs_inode *entry;
956 spin_lock(&root->inode_lock);
958 node = root->inode_tree.rb_node;
962 entry = rb_entry(node, struct btrfs_inode, rb_node);
964 if (objectid < btrfs_ino(entry))
965 node = node->rb_left;
966 else if (objectid > btrfs_ino(entry))
967 node = node->rb_right;
973 entry = rb_entry(prev, struct btrfs_inode, rb_node);
974 if (objectid <= btrfs_ino(entry)) {
978 prev = rb_next(prev);
982 entry = rb_entry(node, struct btrfs_inode, rb_node);
983 inode = igrab(&entry->vfs_inode);
985 spin_unlock(&root->inode_lock);
989 objectid = btrfs_ino(entry) + 1;
990 if (cond_resched_lock(&root->inode_lock))
993 node = rb_next(node);
995 spin_unlock(&root->inode_lock);
1000 * get new location of data
1002 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1003 u64 bytenr, u64 num_bytes)
1005 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1006 struct btrfs_path *path;
1007 struct btrfs_file_extent_item *fi;
1008 struct extent_buffer *leaf;
1011 path = btrfs_alloc_path();
1015 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1016 ret = btrfs_lookup_file_extent(NULL, root, path,
1017 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1025 leaf = path->nodes[0];
1026 fi = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_file_extent_item);
1029 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1030 btrfs_file_extent_compression(leaf, fi) ||
1031 btrfs_file_extent_encryption(leaf, fi) ||
1032 btrfs_file_extent_other_encoding(leaf, fi));
1034 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1039 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1042 btrfs_free_path(path);
1047 * update file extent items in the tree leaf to point to
1048 * the new locations.
1050 static noinline_for_stack
1051 int replace_file_extents(struct btrfs_trans_handle *trans,
1052 struct reloc_control *rc,
1053 struct btrfs_root *root,
1054 struct extent_buffer *leaf)
1056 struct btrfs_fs_info *fs_info = root->fs_info;
1057 struct btrfs_key key;
1058 struct btrfs_file_extent_item *fi;
1059 struct inode *inode = NULL;
1071 if (rc->stage != UPDATE_DATA_PTRS)
1074 /* reloc trees always use full backref */
1075 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1076 parent = leaf->start;
1080 nritems = btrfs_header_nritems(leaf);
1081 for (i = 0; i < nritems; i++) {
1082 struct btrfs_ref ref = { 0 };
1085 btrfs_item_key_to_cpu(leaf, &key, i);
1086 if (key.type != BTRFS_EXTENT_DATA_KEY)
1088 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1089 if (btrfs_file_extent_type(leaf, fi) ==
1090 BTRFS_FILE_EXTENT_INLINE)
1092 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1093 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1096 if (!in_range(bytenr, rc->block_group->start,
1097 rc->block_group->length))
1101 * if we are modifying block in fs tree, wait for readpage
1102 * to complete and drop the extent cache
1104 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1106 inode = find_next_inode(root, key.objectid);
1108 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1109 btrfs_add_delayed_iput(inode);
1110 inode = find_next_inode(root, key.objectid);
1112 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1114 btrfs_file_extent_num_bytes(leaf, fi);
1115 WARN_ON(!IS_ALIGNED(key.offset,
1116 fs_info->sectorsize));
1117 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1119 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1124 btrfs_drop_extent_cache(BTRFS_I(inode),
1125 key.offset, end, 1);
1126 unlock_extent(&BTRFS_I(inode)->io_tree,
1131 ret = get_new_location(rc->data_inode, &new_bytenr,
1135 * Don't have to abort since we've not changed anything
1136 * in the file extent yet.
1141 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1144 key.offset -= btrfs_file_extent_offset(leaf, fi);
1145 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1147 ref.real_root = root->root_key.objectid;
1148 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1149 key.objectid, key.offset);
1150 ret = btrfs_inc_extent_ref(trans, &ref);
1152 btrfs_abort_transaction(trans, ret);
1156 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1158 ref.real_root = root->root_key.objectid;
1159 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1160 key.objectid, key.offset);
1161 ret = btrfs_free_extent(trans, &ref);
1163 btrfs_abort_transaction(trans, ret);
1168 btrfs_mark_buffer_dirty(leaf);
1170 btrfs_add_delayed_iput(inode);
1174 static noinline_for_stack
1175 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1176 struct btrfs_path *path, int level)
1178 struct btrfs_disk_key key1;
1179 struct btrfs_disk_key key2;
1180 btrfs_node_key(eb, &key1, slot);
1181 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1182 return memcmp(&key1, &key2, sizeof(key1));
1186 * try to replace tree blocks in fs tree with the new blocks
1187 * in reloc tree. tree blocks haven't been modified since the
1188 * reloc tree was create can be replaced.
1190 * if a block was replaced, level of the block + 1 is returned.
1191 * if no block got replaced, 0 is returned. if there are other
1192 * errors, a negative error number is returned.
1194 static noinline_for_stack
1195 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1196 struct btrfs_root *dest, struct btrfs_root *src,
1197 struct btrfs_path *path, struct btrfs_key *next_key,
1198 int lowest_level, int max_level)
1200 struct btrfs_fs_info *fs_info = dest->fs_info;
1201 struct extent_buffer *eb;
1202 struct extent_buffer *parent;
1203 struct btrfs_ref ref = { 0 };
1204 struct btrfs_key key;
1216 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1217 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1219 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1221 slot = path->slots[lowest_level];
1222 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1224 eb = btrfs_lock_root_node(dest);
1225 level = btrfs_header_level(eb);
1227 if (level < lowest_level) {
1228 btrfs_tree_unlock(eb);
1229 free_extent_buffer(eb);
1234 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1237 btrfs_tree_unlock(eb);
1238 free_extent_buffer(eb);
1244 next_key->objectid = (u64)-1;
1245 next_key->type = (u8)-1;
1246 next_key->offset = (u64)-1;
1251 level = btrfs_header_level(parent);
1252 ASSERT(level >= lowest_level);
1254 ret = btrfs_bin_search(parent, &key, &slot);
1257 if (ret && slot > 0)
1260 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1261 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1263 old_bytenr = btrfs_node_blockptr(parent, slot);
1264 blocksize = fs_info->nodesize;
1265 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1267 if (level <= max_level) {
1268 eb = path->nodes[level];
1269 new_bytenr = btrfs_node_blockptr(eb,
1270 path->slots[level]);
1271 new_ptr_gen = btrfs_node_ptr_generation(eb,
1272 path->slots[level]);
1278 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1283 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1284 memcmp_node_keys(parent, slot, path, level)) {
1285 if (level <= lowest_level) {
1290 eb = btrfs_read_node_slot(parent, slot);
1295 btrfs_tree_lock(eb);
1297 ret = btrfs_cow_block(trans, dest, eb, parent,
1301 btrfs_tree_unlock(eb);
1302 free_extent_buffer(eb);
1307 btrfs_tree_unlock(parent);
1308 free_extent_buffer(parent);
1315 btrfs_tree_unlock(parent);
1316 free_extent_buffer(parent);
1321 btrfs_node_key_to_cpu(path->nodes[level], &key,
1322 path->slots[level]);
1323 btrfs_release_path(path);
1325 path->lowest_level = level;
1326 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1327 path->lowest_level = 0;
1335 * Info qgroup to trace both subtrees.
1337 * We must trace both trees.
1338 * 1) Tree reloc subtree
1339 * If not traced, we will leak data numbers
1341 * If not traced, we will double count old data
1343 * We don't scan the subtree right now, but only record
1344 * the swapped tree blocks.
1345 * The real subtree rescan is delayed until we have new
1346 * CoW on the subtree root node before transaction commit.
1348 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1349 rc->block_group, parent, slot,
1350 path->nodes[level], path->slots[level],
1355 * swap blocks in fs tree and reloc tree.
1357 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1358 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1359 btrfs_mark_buffer_dirty(parent);
1361 btrfs_set_node_blockptr(path->nodes[level],
1362 path->slots[level], old_bytenr);
1363 btrfs_set_node_ptr_generation(path->nodes[level],
1364 path->slots[level], old_ptr_gen);
1365 btrfs_mark_buffer_dirty(path->nodes[level]);
1367 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1368 blocksize, path->nodes[level]->start);
1369 ref.skip_qgroup = true;
1370 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1371 ret = btrfs_inc_extent_ref(trans, &ref);
1373 btrfs_abort_transaction(trans, ret);
1376 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1378 ref.skip_qgroup = true;
1379 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1380 ret = btrfs_inc_extent_ref(trans, &ref);
1382 btrfs_abort_transaction(trans, ret);
1386 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1387 blocksize, path->nodes[level]->start);
1388 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1389 ref.skip_qgroup = true;
1390 ret = btrfs_free_extent(trans, &ref);
1392 btrfs_abort_transaction(trans, ret);
1396 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1398 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1399 ref.skip_qgroup = true;
1400 ret = btrfs_free_extent(trans, &ref);
1402 btrfs_abort_transaction(trans, ret);
1406 btrfs_unlock_up_safe(path, 0);
1411 btrfs_tree_unlock(parent);
1412 free_extent_buffer(parent);
1417 * helper to find next relocated block in reloc tree
1419 static noinline_for_stack
1420 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1423 struct extent_buffer *eb;
1428 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1430 for (i = 0; i < *level; i++) {
1431 free_extent_buffer(path->nodes[i]);
1432 path->nodes[i] = NULL;
1435 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1436 eb = path->nodes[i];
1437 nritems = btrfs_header_nritems(eb);
1438 while (path->slots[i] + 1 < nritems) {
1440 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1447 free_extent_buffer(path->nodes[i]);
1448 path->nodes[i] = NULL;
1454 * walk down reloc tree to find relocated block of lowest level
1456 static noinline_for_stack
1457 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1460 struct extent_buffer *eb = NULL;
1466 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1468 for (i = *level; i > 0; i--) {
1469 eb = path->nodes[i];
1470 nritems = btrfs_header_nritems(eb);
1471 while (path->slots[i] < nritems) {
1472 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1473 if (ptr_gen > last_snapshot)
1477 if (path->slots[i] >= nritems) {
1488 eb = btrfs_read_node_slot(eb, path->slots[i]);
1491 BUG_ON(btrfs_header_level(eb) != i - 1);
1492 path->nodes[i - 1] = eb;
1493 path->slots[i - 1] = 0;
1499 * invalidate extent cache for file extents whose key in range of
1500 * [min_key, max_key)
1502 static int invalidate_extent_cache(struct btrfs_root *root,
1503 struct btrfs_key *min_key,
1504 struct btrfs_key *max_key)
1506 struct btrfs_fs_info *fs_info = root->fs_info;
1507 struct inode *inode = NULL;
1512 objectid = min_key->objectid;
1517 if (objectid > max_key->objectid)
1520 inode = find_next_inode(root, objectid);
1523 ino = btrfs_ino(BTRFS_I(inode));
1525 if (ino > max_key->objectid) {
1531 if (!S_ISREG(inode->i_mode))
1534 if (unlikely(min_key->objectid == ino)) {
1535 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1537 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1540 start = min_key->offset;
1541 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1547 if (unlikely(max_key->objectid == ino)) {
1548 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1550 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1553 if (max_key->offset == 0)
1555 end = max_key->offset;
1556 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1563 /* the lock_extent waits for readpage to complete */
1564 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1565 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1566 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1571 static int find_next_key(struct btrfs_path *path, int level,
1572 struct btrfs_key *key)
1575 while (level < BTRFS_MAX_LEVEL) {
1576 if (!path->nodes[level])
1578 if (path->slots[level] + 1 <
1579 btrfs_header_nritems(path->nodes[level])) {
1580 btrfs_node_key_to_cpu(path->nodes[level], key,
1581 path->slots[level] + 1);
1590 * Insert current subvolume into reloc_control::dirty_subvol_roots
1592 static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1593 struct reloc_control *rc,
1594 struct btrfs_root *root)
1596 struct btrfs_root *reloc_root = root->reloc_root;
1597 struct btrfs_root_item *reloc_root_item;
1600 /* @root must be a subvolume tree root with a valid reloc tree */
1601 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1604 reloc_root_item = &reloc_root->root_item;
1605 memset(&reloc_root_item->drop_progress, 0,
1606 sizeof(reloc_root_item->drop_progress));
1607 btrfs_set_root_drop_level(reloc_root_item, 0);
1608 btrfs_set_root_refs(reloc_root_item, 0);
1609 ret = btrfs_update_reloc_root(trans, root);
1613 if (list_empty(&root->reloc_dirty_list)) {
1614 btrfs_grab_root(root);
1615 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1621 static int clean_dirty_subvols(struct reloc_control *rc)
1623 struct btrfs_root *root;
1624 struct btrfs_root *next;
1628 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1630 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1631 /* Merged subvolume, cleanup its reloc root */
1632 struct btrfs_root *reloc_root = root->reloc_root;
1634 list_del_init(&root->reloc_dirty_list);
1635 root->reloc_root = NULL;
1637 * Need barrier to ensure clear_bit() only happens after
1638 * root->reloc_root = NULL. Pairs with have_reloc_root.
1641 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1644 * btrfs_drop_snapshot drops our ref we hold for
1645 * ->reloc_root. If it fails however we must
1646 * drop the ref ourselves.
1648 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1650 btrfs_put_root(reloc_root);
1655 btrfs_put_root(root);
1657 /* Orphan reloc tree, just clean it up */
1658 ret2 = btrfs_drop_snapshot(root, 0, 1);
1660 btrfs_put_root(root);
1670 * merge the relocated tree blocks in reloc tree with corresponding
1673 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1674 struct btrfs_root *root)
1676 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1677 struct btrfs_key key;
1678 struct btrfs_key next_key;
1679 struct btrfs_trans_handle *trans = NULL;
1680 struct btrfs_root *reloc_root;
1681 struct btrfs_root_item *root_item;
1682 struct btrfs_path *path;
1683 struct extent_buffer *leaf;
1691 path = btrfs_alloc_path();
1694 path->reada = READA_FORWARD;
1696 reloc_root = root->reloc_root;
1697 root_item = &reloc_root->root_item;
1699 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1700 level = btrfs_root_level(root_item);
1701 atomic_inc(&reloc_root->node->refs);
1702 path->nodes[level] = reloc_root->node;
1703 path->slots[level] = 0;
1705 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1707 level = btrfs_root_drop_level(root_item);
1709 path->lowest_level = level;
1710 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1711 path->lowest_level = 0;
1713 btrfs_free_path(path);
1717 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1718 path->slots[level]);
1719 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1721 btrfs_unlock_up_safe(path, 0);
1725 * In merge_reloc_root(), we modify the upper level pointer to swap the
1726 * tree blocks between reloc tree and subvolume tree. Thus for tree
1727 * block COW, we COW at most from level 1 to root level for each tree.
1729 * Thus the needed metadata size is at most root_level * nodesize,
1730 * and * 2 since we have two trees to COW.
1732 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1733 min_reserved = fs_info->nodesize * reserve_level * 2;
1734 memset(&next_key, 0, sizeof(next_key));
1737 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1738 BTRFS_RESERVE_FLUSH_LIMIT);
1741 trans = btrfs_start_transaction(root, 0);
1742 if (IS_ERR(trans)) {
1743 ret = PTR_ERR(trans);
1749 * At this point we no longer have a reloc_control, so we can't
1750 * depend on btrfs_init_reloc_root to update our last_trans.
1752 * But that's ok, we started the trans handle on our
1753 * corresponding fs_root, which means it's been added to the
1754 * dirty list. At commit time we'll still call
1755 * btrfs_update_reloc_root() and update our root item
1758 reloc_root->last_trans = trans->transid;
1759 trans->block_rsv = rc->block_rsv;
1764 ret = walk_down_reloc_tree(reloc_root, path, &level);
1770 if (!find_next_key(path, level, &key) &&
1771 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1774 ret = replace_path(trans, rc, root, reloc_root, path,
1775 &next_key, level, max_level);
1781 btrfs_node_key_to_cpu(path->nodes[level], &key,
1782 path->slots[level]);
1786 ret = walk_up_reloc_tree(reloc_root, path, &level);
1792 * save the merging progress in the drop_progress.
1793 * this is OK since root refs == 1 in this case.
1795 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1796 path->slots[level]);
1797 btrfs_set_root_drop_level(root_item, level);
1799 btrfs_end_transaction_throttle(trans);
1802 btrfs_btree_balance_dirty(fs_info);
1804 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1805 invalidate_extent_cache(root, &key, &next_key);
1809 * handle the case only one block in the fs tree need to be
1810 * relocated and the block is tree root.
1812 leaf = btrfs_lock_root_node(root);
1813 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1815 btrfs_tree_unlock(leaf);
1816 free_extent_buffer(leaf);
1818 btrfs_free_path(path);
1821 ret = insert_dirty_subvol(trans, rc, root);
1823 btrfs_abort_transaction(trans, ret);
1827 btrfs_end_transaction_throttle(trans);
1829 btrfs_btree_balance_dirty(fs_info);
1831 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1832 invalidate_extent_cache(root, &key, &next_key);
1837 static noinline_for_stack
1838 int prepare_to_merge(struct reloc_control *rc, int err)
1840 struct btrfs_root *root = rc->extent_root;
1841 struct btrfs_fs_info *fs_info = root->fs_info;
1842 struct btrfs_root *reloc_root;
1843 struct btrfs_trans_handle *trans;
1844 LIST_HEAD(reloc_roots);
1848 mutex_lock(&fs_info->reloc_mutex);
1849 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1850 rc->merging_rsv_size += rc->nodes_relocated * 2;
1851 mutex_unlock(&fs_info->reloc_mutex);
1855 num_bytes = rc->merging_rsv_size;
1856 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1857 BTRFS_RESERVE_FLUSH_ALL);
1862 trans = btrfs_join_transaction(rc->extent_root);
1863 if (IS_ERR(trans)) {
1865 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1867 return PTR_ERR(trans);
1871 if (num_bytes != rc->merging_rsv_size) {
1872 btrfs_end_transaction(trans);
1873 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1879 rc->merge_reloc_tree = 1;
1881 while (!list_empty(&rc->reloc_roots)) {
1882 reloc_root = list_entry(rc->reloc_roots.next,
1883 struct btrfs_root, root_list);
1884 list_del_init(&reloc_root->root_list);
1886 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1890 * Even if we have an error we need this reloc root
1891 * back on our list so we can clean up properly.
1893 list_add(&reloc_root->root_list, &reloc_roots);
1894 btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1896 err = PTR_ERR(root);
1899 ASSERT(root->reloc_root == reloc_root);
1902 * set reference count to 1, so btrfs_recover_relocation
1903 * knows it should resumes merging
1906 btrfs_set_root_refs(&reloc_root->root_item, 1);
1907 ret = btrfs_update_reloc_root(trans, root);
1910 * Even if we have an error we need this reloc root back on our
1911 * list so we can clean up properly.
1913 list_add(&reloc_root->root_list, &reloc_roots);
1914 btrfs_put_root(root);
1917 btrfs_abort_transaction(trans, ret);
1924 list_splice(&reloc_roots, &rc->reloc_roots);
1927 err = btrfs_commit_transaction(trans);
1929 btrfs_end_transaction(trans);
1933 static noinline_for_stack
1934 void free_reloc_roots(struct list_head *list)
1936 struct btrfs_root *reloc_root, *tmp;
1938 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1939 __del_reloc_root(reloc_root);
1942 static noinline_for_stack
1943 void merge_reloc_roots(struct reloc_control *rc)
1945 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1946 struct btrfs_root *root;
1947 struct btrfs_root *reloc_root;
1948 LIST_HEAD(reloc_roots);
1952 root = rc->extent_root;
1955 * this serializes us with btrfs_record_root_in_transaction,
1956 * we have to make sure nobody is in the middle of
1957 * adding their roots to the list while we are
1960 mutex_lock(&fs_info->reloc_mutex);
1961 list_splice_init(&rc->reloc_roots, &reloc_roots);
1962 mutex_unlock(&fs_info->reloc_mutex);
1964 while (!list_empty(&reloc_roots)) {
1966 reloc_root = list_entry(reloc_roots.next,
1967 struct btrfs_root, root_list);
1969 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1971 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1974 * For recovery we read the fs roots on mount,
1975 * and if we didn't find the root then we marked
1976 * the reloc root as a garbage root. For normal
1977 * relocation obviously the root should exist in
1978 * memory. However there's no reason we can't
1979 * handle the error properly here just in case.
1982 ret = PTR_ERR(root);
1985 if (root->reloc_root != reloc_root) {
1987 * This is actually impossible without something
1988 * going really wrong (like weird race condition
1995 ret = merge_reloc_root(rc, root);
1996 btrfs_put_root(root);
1998 if (list_empty(&reloc_root->root_list))
1999 list_add_tail(&reloc_root->root_list,
2004 if (!IS_ERR(root)) {
2005 if (root->reloc_root == reloc_root) {
2006 root->reloc_root = NULL;
2007 btrfs_put_root(reloc_root);
2009 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
2011 btrfs_put_root(root);
2014 list_del_init(&reloc_root->root_list);
2015 /* Don't forget to queue this reloc root for cleanup */
2016 list_add_tail(&reloc_root->reloc_dirty_list,
2017 &rc->dirty_subvol_roots);
2027 btrfs_handle_fs_error(fs_info, ret, NULL);
2028 free_reloc_roots(&reloc_roots);
2030 /* new reloc root may be added */
2031 mutex_lock(&fs_info->reloc_mutex);
2032 list_splice_init(&rc->reloc_roots, &reloc_roots);
2033 mutex_unlock(&fs_info->reloc_mutex);
2034 free_reloc_roots(&reloc_roots);
2040 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2042 * here, but it's wrong. If we fail to start the transaction in
2043 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2044 * have actually been removed from the reloc_root_tree rb tree. This is
2045 * fine because we're bailing here, and we hold a reference on the root
2046 * for the list that holds it, so these roots will be cleaned up when we
2047 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
2048 * will be cleaned up on unmount.
2050 * The remaining nodes will be cleaned up by free_reloc_control.
2054 static void free_block_list(struct rb_root *blocks)
2056 struct tree_block *block;
2057 struct rb_node *rb_node;
2058 while ((rb_node = rb_first(blocks))) {
2059 block = rb_entry(rb_node, struct tree_block, rb_node);
2060 rb_erase(rb_node, blocks);
2065 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *reloc_root)
2068 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2069 struct btrfs_root *root;
2072 if (reloc_root->last_trans == trans->transid)
2075 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2078 * This should succeed, since we can't have a reloc root without having
2079 * already looked up the actual root and created the reloc root for this
2082 * However if there's some sort of corruption where we have a ref to a
2083 * reloc root without a corresponding root this could return ENOENT.
2087 return PTR_ERR(root);
2089 if (root->reloc_root != reloc_root) {
2092 "root %llu has two reloc roots associated with it",
2093 reloc_root->root_key.offset);
2094 btrfs_put_root(root);
2097 ret = btrfs_record_root_in_trans(trans, root);
2098 btrfs_put_root(root);
2103 static noinline_for_stack
2104 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2105 struct reloc_control *rc,
2106 struct btrfs_backref_node *node,
2107 struct btrfs_backref_edge *edges[])
2109 struct btrfs_backref_node *next;
2110 struct btrfs_root *root;
2117 next = walk_up_backref(next, edges, &index);
2121 * If there is no root, then our references for this block are
2122 * incomplete, as we should be able to walk all the way up to a
2123 * block that is owned by a root.
2125 * This path is only for SHAREABLE roots, so if we come upon a
2126 * non-SHAREABLE root then we have backrefs that resolve
2129 * Both of these cases indicate file system corruption, or a bug
2130 * in the backref walking code.
2134 btrfs_err(trans->fs_info,
2135 "bytenr %llu doesn't have a backref path ending in a root",
2137 return ERR_PTR(-EUCLEAN);
2139 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2141 btrfs_err(trans->fs_info,
2142 "bytenr %llu has multiple refs with one ending in a non-shareable root",
2144 return ERR_PTR(-EUCLEAN);
2147 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2148 ret = record_reloc_root_in_trans(trans, root);
2150 return ERR_PTR(ret);
2154 ret = btrfs_record_root_in_trans(trans, root);
2156 return ERR_PTR(ret);
2157 root = root->reloc_root;
2160 * We could have raced with another thread which failed, so
2161 * root->reloc_root may not be set, return ENOENT in this case.
2164 return ERR_PTR(-ENOENT);
2166 if (next->new_bytenr != root->node->start) {
2168 * We just created the reloc root, so we shouldn't have
2169 * ->new_bytenr set and this shouldn't be in the changed
2170 * list. If it is then we have multiple roots pointing
2171 * at the same bytenr which indicates corruption, or
2172 * we've made a mistake in the backref walking code.
2174 ASSERT(next->new_bytenr == 0);
2175 ASSERT(list_empty(&next->list));
2176 if (next->new_bytenr || !list_empty(&next->list)) {
2177 btrfs_err(trans->fs_info,
2178 "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2179 node->bytenr, next->bytenr);
2180 return ERR_PTR(-EUCLEAN);
2183 next->new_bytenr = root->node->start;
2184 btrfs_put_root(next->root);
2185 next->root = btrfs_grab_root(root);
2187 list_add_tail(&next->list,
2188 &rc->backref_cache.changed);
2189 mark_block_processed(rc, next);
2195 next = walk_down_backref(edges, &index);
2196 if (!next || next->level <= node->level)
2201 * This can happen if there's fs corruption or if there's a bug
2202 * in the backref lookup code.
2205 return ERR_PTR(-ENOENT);
2209 /* setup backref node path for btrfs_reloc_cow_block */
2211 rc->backref_cache.path[next->level] = next;
2214 next = edges[index]->node[UPPER];
2220 * Select a tree root for relocation.
2222 * Return NULL if the block is not shareable. We should use do_relocation() in
2225 * Return a tree root pointer if the block is shareable.
2226 * Return -ENOENT if the block is root of reloc tree.
2228 static noinline_for_stack
2229 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2231 struct btrfs_backref_node *next;
2232 struct btrfs_root *root;
2233 struct btrfs_root *fs_root = NULL;
2234 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2240 next = walk_up_backref(next, edges, &index);
2244 * This can occur if we have incomplete extent refs leading all
2245 * the way up a particular path, in this case return -EUCLEAN.
2248 return ERR_PTR(-EUCLEAN);
2250 /* No other choice for non-shareable tree */
2251 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2254 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2260 next = walk_down_backref(edges, &index);
2261 if (!next || next->level <= node->level)
2266 return ERR_PTR(-ENOENT);
2270 static noinline_for_stack
2271 u64 calcu_metadata_size(struct reloc_control *rc,
2272 struct btrfs_backref_node *node, int reserve)
2274 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2275 struct btrfs_backref_node *next = node;
2276 struct btrfs_backref_edge *edge;
2277 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2281 BUG_ON(reserve && node->processed);
2286 if (next->processed && (reserve || next != node))
2289 num_bytes += fs_info->nodesize;
2291 if (list_empty(&next->upper))
2294 edge = list_entry(next->upper.next,
2295 struct btrfs_backref_edge, list[LOWER]);
2296 edges[index++] = edge;
2297 next = edge->node[UPPER];
2299 next = walk_down_backref(edges, &index);
2304 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2305 struct reloc_control *rc,
2306 struct btrfs_backref_node *node)
2308 struct btrfs_root *root = rc->extent_root;
2309 struct btrfs_fs_info *fs_info = root->fs_info;
2314 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2316 trans->block_rsv = rc->block_rsv;
2317 rc->reserved_bytes += num_bytes;
2320 * We are under a transaction here so we can only do limited flushing.
2321 * If we get an enospc just kick back -EAGAIN so we know to drop the
2322 * transaction and try to refill when we can flush all the things.
2324 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2325 BTRFS_RESERVE_FLUSH_LIMIT);
2327 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2328 while (tmp <= rc->reserved_bytes)
2331 * only one thread can access block_rsv at this point,
2332 * so we don't need hold lock to protect block_rsv.
2333 * we expand more reservation size here to allow enough
2334 * space for relocation and we will return earlier in
2337 rc->block_rsv->size = tmp + fs_info->nodesize *
2338 RELOCATION_RESERVED_NODES;
2346 * relocate a block tree, and then update pointers in upper level
2347 * blocks that reference the block to point to the new location.
2349 * if called by link_to_upper, the block has already been relocated.
2350 * in that case this function just updates pointers.
2352 static int do_relocation(struct btrfs_trans_handle *trans,
2353 struct reloc_control *rc,
2354 struct btrfs_backref_node *node,
2355 struct btrfs_key *key,
2356 struct btrfs_path *path, int lowest)
2358 struct btrfs_backref_node *upper;
2359 struct btrfs_backref_edge *edge;
2360 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2361 struct btrfs_root *root;
2362 struct extent_buffer *eb;
2369 * If we are lowest then this is the first time we're processing this
2370 * block, and thus shouldn't have an eb associated with it yet.
2372 ASSERT(!lowest || !node->eb);
2374 path->lowest_level = node->level + 1;
2375 rc->backref_cache.path[node->level] = node;
2376 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2377 struct btrfs_ref ref = { 0 };
2381 upper = edge->node[UPPER];
2382 root = select_reloc_root(trans, rc, upper, edges);
2384 ret = PTR_ERR(root);
2388 if (upper->eb && !upper->locked) {
2390 ret = btrfs_bin_search(upper->eb, key, &slot);
2394 bytenr = btrfs_node_blockptr(upper->eb, slot);
2395 if (node->eb->start == bytenr)
2398 btrfs_backref_drop_node_buffer(upper);
2402 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2407 btrfs_release_path(path);
2412 upper->eb = path->nodes[upper->level];
2413 path->nodes[upper->level] = NULL;
2415 BUG_ON(upper->eb != path->nodes[upper->level]);
2419 path->locks[upper->level] = 0;
2421 slot = path->slots[upper->level];
2422 btrfs_release_path(path);
2424 ret = btrfs_bin_search(upper->eb, key, &slot);
2430 bytenr = btrfs_node_blockptr(upper->eb, slot);
2432 if (bytenr != node->bytenr) {
2433 btrfs_err(root->fs_info,
2434 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2435 bytenr, node->bytenr, slot,
2441 if (node->eb->start == bytenr)
2445 blocksize = root->fs_info->nodesize;
2446 eb = btrfs_read_node_slot(upper->eb, slot);
2451 btrfs_tree_lock(eb);
2454 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2455 slot, &eb, BTRFS_NESTING_COW);
2456 btrfs_tree_unlock(eb);
2457 free_extent_buffer(eb);
2461 * We've just COWed this block, it should have updated
2462 * the correct backref node entry.
2464 ASSERT(node->eb == eb);
2466 btrfs_set_node_blockptr(upper->eb, slot,
2468 btrfs_set_node_ptr_generation(upper->eb, slot,
2470 btrfs_mark_buffer_dirty(upper->eb);
2472 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2473 node->eb->start, blocksize,
2475 ref.real_root = root->root_key.objectid;
2476 btrfs_init_tree_ref(&ref, node->level,
2477 btrfs_header_owner(upper->eb));
2478 ret = btrfs_inc_extent_ref(trans, &ref);
2480 ret = btrfs_drop_subtree(trans, root, eb,
2483 btrfs_abort_transaction(trans, ret);
2486 if (!upper->pending)
2487 btrfs_backref_drop_node_buffer(upper);
2489 btrfs_backref_unlock_node_buffer(upper);
2494 if (!ret && node->pending) {
2495 btrfs_backref_drop_node_buffer(node);
2496 list_move_tail(&node->list, &rc->backref_cache.changed);
2500 path->lowest_level = 0;
2503 * We should have allocated all of our space in the block rsv and thus
2506 ASSERT(ret != -ENOSPC);
2510 static int link_to_upper(struct btrfs_trans_handle *trans,
2511 struct reloc_control *rc,
2512 struct btrfs_backref_node *node,
2513 struct btrfs_path *path)
2515 struct btrfs_key key;
2517 btrfs_node_key_to_cpu(node->eb, &key, 0);
2518 return do_relocation(trans, rc, node, &key, path, 0);
2521 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2522 struct reloc_control *rc,
2523 struct btrfs_path *path, int err)
2526 struct btrfs_backref_cache *cache = &rc->backref_cache;
2527 struct btrfs_backref_node *node;
2531 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2532 while (!list_empty(&cache->pending[level])) {
2533 node = list_entry(cache->pending[level].next,
2534 struct btrfs_backref_node, list);
2535 list_move_tail(&node->list, &list);
2536 BUG_ON(!node->pending);
2539 ret = link_to_upper(trans, rc, node, path);
2544 list_splice_init(&list, &cache->pending[level]);
2550 * mark a block and all blocks directly/indirectly reference the block
2553 static void update_processed_blocks(struct reloc_control *rc,
2554 struct btrfs_backref_node *node)
2556 struct btrfs_backref_node *next = node;
2557 struct btrfs_backref_edge *edge;
2558 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2564 if (next->processed)
2567 mark_block_processed(rc, next);
2569 if (list_empty(&next->upper))
2572 edge = list_entry(next->upper.next,
2573 struct btrfs_backref_edge, list[LOWER]);
2574 edges[index++] = edge;
2575 next = edge->node[UPPER];
2577 next = walk_down_backref(edges, &index);
2581 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2583 u32 blocksize = rc->extent_root->fs_info->nodesize;
2585 if (test_range_bit(&rc->processed_blocks, bytenr,
2586 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2591 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2592 struct tree_block *block)
2594 struct extent_buffer *eb;
2596 eb = read_tree_block(fs_info, block->bytenr, block->owner,
2597 block->key.offset, block->level, NULL);
2600 } else if (!extent_buffer_uptodate(eb)) {
2601 free_extent_buffer(eb);
2604 if (block->level == 0)
2605 btrfs_item_key_to_cpu(eb, &block->key, 0);
2607 btrfs_node_key_to_cpu(eb, &block->key, 0);
2608 free_extent_buffer(eb);
2609 block->key_ready = 1;
2614 * helper function to relocate a tree block
2616 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2617 struct reloc_control *rc,
2618 struct btrfs_backref_node *node,
2619 struct btrfs_key *key,
2620 struct btrfs_path *path)
2622 struct btrfs_root *root;
2629 * If we fail here we want to drop our backref_node because we are going
2630 * to start over and regenerate the tree for it.
2632 ret = reserve_metadata_space(trans, rc, node);
2636 BUG_ON(node->processed);
2637 root = select_one_root(node);
2639 ret = PTR_ERR(root);
2641 /* See explanation in select_one_root for the -EUCLEAN case. */
2642 ASSERT(ret == -ENOENT);
2643 if (ret == -ENOENT) {
2645 update_processed_blocks(rc, node);
2651 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2653 * This block was the root block of a root, and this is
2654 * the first time we're processing the block and thus it
2655 * should not have had the ->new_bytenr modified and
2656 * should have not been included on the changed list.
2658 * However in the case of corruption we could have
2659 * multiple refs pointing to the same block improperly,
2660 * and thus we would trip over these checks. ASSERT()
2661 * for the developer case, because it could indicate a
2662 * bug in the backref code, however error out for a
2663 * normal user in the case of corruption.
2665 ASSERT(node->new_bytenr == 0);
2666 ASSERT(list_empty(&node->list));
2667 if (node->new_bytenr || !list_empty(&node->list)) {
2668 btrfs_err(root->fs_info,
2669 "bytenr %llu has improper references to it",
2674 ret = btrfs_record_root_in_trans(trans, root);
2678 * Another thread could have failed, need to check if we
2679 * have reloc_root actually set.
2681 if (!root->reloc_root) {
2685 root = root->reloc_root;
2686 node->new_bytenr = root->node->start;
2687 btrfs_put_root(node->root);
2688 node->root = btrfs_grab_root(root);
2690 list_add_tail(&node->list, &rc->backref_cache.changed);
2692 path->lowest_level = node->level;
2693 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2694 btrfs_release_path(path);
2699 update_processed_blocks(rc, node);
2701 ret = do_relocation(trans, rc, node, key, path, 1);
2704 if (ret || node->level == 0 || node->cowonly)
2705 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2710 * relocate a list of blocks
2712 static noinline_for_stack
2713 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2714 struct reloc_control *rc, struct rb_root *blocks)
2716 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2717 struct btrfs_backref_node *node;
2718 struct btrfs_path *path;
2719 struct tree_block *block;
2720 struct tree_block *next;
2724 path = btrfs_alloc_path();
2727 goto out_free_blocks;
2730 /* Kick in readahead for tree blocks with missing keys */
2731 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2732 if (!block->key_ready)
2733 btrfs_readahead_tree_block(fs_info, block->bytenr,
2738 /* Get first keys */
2739 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2740 if (!block->key_ready) {
2741 err = get_tree_block_key(fs_info, block);
2747 /* Do tree relocation */
2748 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2749 node = build_backref_tree(rc, &block->key,
2750 block->level, block->bytenr);
2752 err = PTR_ERR(node);
2756 ret = relocate_tree_block(trans, rc, node, &block->key,
2764 err = finish_pending_nodes(trans, rc, path, err);
2767 btrfs_free_path(path);
2769 free_block_list(blocks);
2773 static noinline_for_stack int prealloc_file_extent_cluster(
2774 struct btrfs_inode *inode,
2775 struct file_extent_cluster *cluster)
2780 u64 offset = inode->index_cnt;
2784 u64 prealloc_start = cluster->start - offset;
2785 u64 prealloc_end = cluster->end - offset;
2786 u64 cur_offset = prealloc_start;
2788 BUG_ON(cluster->start != cluster->boundary[0]);
2789 ret = btrfs_alloc_data_chunk_ondemand(inode,
2790 prealloc_end + 1 - prealloc_start);
2795 * On a zoned filesystem, we cannot preallocate the file region.
2796 * Instead, we dirty and fiemap_write the region.
2798 if (btrfs_is_zoned(inode->root->fs_info)) {
2799 struct btrfs_root *root = inode->root;
2800 struct btrfs_trans_handle *trans;
2802 end = cluster->end - offset + 1;
2803 trans = btrfs_start_transaction(root, 1);
2805 return PTR_ERR(trans);
2807 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
2808 i_size_write(&inode->vfs_inode, end);
2809 ret = btrfs_update_inode(trans, root, inode);
2811 btrfs_abort_transaction(trans, ret);
2812 btrfs_end_transaction(trans);
2816 return btrfs_end_transaction(trans);
2819 btrfs_inode_lock(&inode->vfs_inode, 0);
2820 for (nr = 0; nr < cluster->nr; nr++) {
2821 start = cluster->boundary[nr] - offset;
2822 if (nr + 1 < cluster->nr)
2823 end = cluster->boundary[nr + 1] - 1 - offset;
2825 end = cluster->end - offset;
2827 lock_extent(&inode->io_tree, start, end);
2828 num_bytes = end + 1 - start;
2829 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2830 num_bytes, num_bytes,
2831 end + 1, &alloc_hint);
2832 cur_offset = end + 1;
2833 unlock_extent(&inode->io_tree, start, end);
2837 btrfs_inode_unlock(&inode->vfs_inode, 0);
2839 if (cur_offset < prealloc_end)
2840 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2841 prealloc_end + 1 - cur_offset);
2845 static noinline_for_stack
2846 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2849 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2850 struct extent_map *em;
2853 em = alloc_extent_map();
2858 em->len = end + 1 - start;
2859 em->block_len = em->len;
2860 em->block_start = block_start;
2861 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2863 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2865 write_lock(&em_tree->lock);
2866 ret = add_extent_mapping(em_tree, em, 0);
2867 write_unlock(&em_tree->lock);
2868 if (ret != -EEXIST) {
2869 free_extent_map(em);
2872 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2874 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2879 * Allow error injection to test balance cancellation
2881 noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2883 return atomic_read(&fs_info->balance_cancel_req) ||
2884 fatal_signal_pending(current);
2886 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2888 static int relocate_file_extent_cluster(struct inode *inode,
2889 struct file_extent_cluster *cluster)
2891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2894 u64 offset = BTRFS_I(inode)->index_cnt;
2895 unsigned long index;
2896 unsigned long last_index;
2898 struct file_ra_state *ra;
2899 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2906 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2910 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2914 file_ra_state_init(ra, inode->i_mapping);
2916 ret = setup_extent_mapping(inode, cluster->start - offset,
2917 cluster->end - offset, cluster->start);
2921 index = (cluster->start - offset) >> PAGE_SHIFT;
2922 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2923 while (index <= last_index) {
2924 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2929 page = find_lock_page(inode->i_mapping, index);
2931 page_cache_sync_readahead(inode->i_mapping,
2933 last_index + 1 - index);
2934 page = find_or_create_page(inode->i_mapping, index,
2937 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2939 btrfs_delalloc_release_extents(BTRFS_I(inode),
2945 ret = set_page_extent_mapped(page);
2947 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2949 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2955 if (PageReadahead(page)) {
2956 page_cache_async_readahead(inode->i_mapping,
2957 ra, NULL, page, index,
2958 last_index + 1 - index);
2961 if (!PageUptodate(page)) {
2962 btrfs_readpage(NULL, page);
2964 if (!PageUptodate(page)) {
2967 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2969 btrfs_delalloc_release_extents(BTRFS_I(inode),
2976 page_start = page_offset(page);
2977 page_end = page_start + PAGE_SIZE - 1;
2979 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2981 if (nr < cluster->nr &&
2982 page_start + offset == cluster->boundary[nr]) {
2983 set_extent_bits(&BTRFS_I(inode)->io_tree,
2984 page_start, page_end,
2989 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2994 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2996 btrfs_delalloc_release_extents(BTRFS_I(inode),
2999 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3000 page_start, page_end,
3001 EXTENT_LOCKED | EXTENT_BOUNDARY);
3005 set_page_dirty(page);
3007 unlock_extent(&BTRFS_I(inode)->io_tree,
3008 page_start, page_end);
3013 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3014 balance_dirty_pages_ratelimited(inode->i_mapping);
3015 btrfs_throttle(fs_info);
3016 if (btrfs_should_cancel_balance(fs_info)) {
3021 WARN_ON(nr != cluster->nr);
3022 if (btrfs_is_zoned(fs_info) && !ret)
3023 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
3029 static noinline_for_stack
3030 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3031 struct file_extent_cluster *cluster)
3035 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3036 ret = relocate_file_extent_cluster(inode, cluster);
3043 cluster->start = extent_key->objectid;
3045 BUG_ON(cluster->nr >= MAX_EXTENTS);
3046 cluster->end = extent_key->objectid + extent_key->offset - 1;
3047 cluster->boundary[cluster->nr] = extent_key->objectid;
3050 if (cluster->nr >= MAX_EXTENTS) {
3051 ret = relocate_file_extent_cluster(inode, cluster);
3060 * helper to add a tree block to the list.
3061 * the major work is getting the generation and level of the block
3063 static int add_tree_block(struct reloc_control *rc,
3064 struct btrfs_key *extent_key,
3065 struct btrfs_path *path,
3066 struct rb_root *blocks)
3068 struct extent_buffer *eb;
3069 struct btrfs_extent_item *ei;
3070 struct btrfs_tree_block_info *bi;
3071 struct tree_block *block;
3072 struct rb_node *rb_node;
3078 eb = path->nodes[0];
3079 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3081 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3082 item_size >= sizeof(*ei) + sizeof(*bi)) {
3083 unsigned long ptr = 0, end;
3085 ei = btrfs_item_ptr(eb, path->slots[0],
3086 struct btrfs_extent_item);
3087 end = (unsigned long)ei + item_size;
3088 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3089 bi = (struct btrfs_tree_block_info *)(ei + 1);
3090 level = btrfs_tree_block_level(eb, bi);
3091 ptr = (unsigned long)(bi + 1);
3093 level = (int)extent_key->offset;
3094 ptr = (unsigned long)(ei + 1);
3096 generation = btrfs_extent_generation(eb, ei);
3099 * We're reading random blocks without knowing their owner ahead
3100 * of time. This is ok most of the time, as all reloc roots and
3101 * fs roots have the same lock type. However normal trees do
3102 * not, and the only way to know ahead of time is to read the
3103 * inline ref offset. We know it's an fs root if
3105 * 1. There's more than one ref.
3106 * 2. There's a SHARED_DATA_REF_KEY set.
3107 * 3. FULL_BACKREF is set on the flags.
3109 * Otherwise it's safe to assume that the ref offset == the
3110 * owner of this block, so we can use that when calling
3113 if (btrfs_extent_refs(eb, ei) == 1 &&
3114 !(btrfs_extent_flags(eb, ei) &
3115 BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3117 struct btrfs_extent_inline_ref *iref;
3120 iref = (struct btrfs_extent_inline_ref *)ptr;
3121 type = btrfs_get_extent_inline_ref_type(eb, iref,
3122 BTRFS_REF_TYPE_BLOCK);
3123 if (type == BTRFS_REF_TYPE_INVALID)
3125 if (type == BTRFS_TREE_BLOCK_REF_KEY)
3126 owner = btrfs_extent_inline_ref_offset(eb, iref);
3128 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3129 btrfs_print_v0_err(eb->fs_info);
3130 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3136 btrfs_release_path(path);
3138 BUG_ON(level == -1);
3140 block = kmalloc(sizeof(*block), GFP_NOFS);
3144 block->bytenr = extent_key->objectid;
3145 block->key.objectid = rc->extent_root->fs_info->nodesize;
3146 block->key.offset = generation;
3147 block->level = level;
3148 block->key_ready = 0;
3149 block->owner = owner;
3151 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3153 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3160 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3162 static int __add_tree_block(struct reloc_control *rc,
3163 u64 bytenr, u32 blocksize,
3164 struct rb_root *blocks)
3166 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3167 struct btrfs_path *path;
3168 struct btrfs_key key;
3170 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3172 if (tree_block_processed(bytenr, rc))
3175 if (rb_simple_search(blocks, bytenr))
3178 path = btrfs_alloc_path();
3182 key.objectid = bytenr;
3184 key.type = BTRFS_METADATA_ITEM_KEY;
3185 key.offset = (u64)-1;
3187 key.type = BTRFS_EXTENT_ITEM_KEY;
3188 key.offset = blocksize;
3191 path->search_commit_root = 1;
3192 path->skip_locking = 1;
3193 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3197 if (ret > 0 && skinny) {
3198 if (path->slots[0]) {
3200 btrfs_item_key_to_cpu(path->nodes[0], &key,
3202 if (key.objectid == bytenr &&
3203 (key.type == BTRFS_METADATA_ITEM_KEY ||
3204 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3205 key.offset == blocksize)))
3211 btrfs_release_path(path);
3217 btrfs_print_leaf(path->nodes[0]);
3219 "tree block extent item (%llu) is not found in extent tree",
3226 ret = add_tree_block(rc, &key, path, blocks);
3228 btrfs_free_path(path);
3232 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3233 struct btrfs_block_group *block_group,
3234 struct inode *inode,
3237 struct btrfs_root *root = fs_info->tree_root;
3238 struct btrfs_trans_handle *trans;
3244 inode = btrfs_iget(fs_info->sb, ino, root);
3249 ret = btrfs_check_trunc_cache_free_space(fs_info,
3250 &fs_info->global_block_rsv);
3254 trans = btrfs_join_transaction(root);
3255 if (IS_ERR(trans)) {
3256 ret = PTR_ERR(trans);
3260 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3262 btrfs_end_transaction(trans);
3263 btrfs_btree_balance_dirty(fs_info);
3270 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3271 * cache inode, to avoid free space cache data extent blocking data relocation.
3273 static int delete_v1_space_cache(struct extent_buffer *leaf,
3274 struct btrfs_block_group *block_group,
3277 u64 space_cache_ino;
3278 struct btrfs_file_extent_item *ei;
3279 struct btrfs_key key;
3284 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3287 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3290 btrfs_item_key_to_cpu(leaf, &key, i);
3291 if (key.type != BTRFS_EXTENT_DATA_KEY)
3293 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3294 type = btrfs_file_extent_type(leaf, ei);
3296 if ((type == BTRFS_FILE_EXTENT_REG ||
3297 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3298 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3300 space_cache_ino = key.objectid;
3306 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3312 * helper to find all tree blocks that reference a given data extent
3314 static noinline_for_stack
3315 int add_data_references(struct reloc_control *rc,
3316 struct btrfs_key *extent_key,
3317 struct btrfs_path *path,
3318 struct rb_root *blocks)
3320 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3321 struct ulist *leaves = NULL;
3322 struct ulist_iterator leaf_uiter;
3323 struct ulist_node *ref_node = NULL;
3324 const u32 blocksize = fs_info->nodesize;
3327 btrfs_release_path(path);
3328 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3329 0, &leaves, NULL, true);
3333 ULIST_ITER_INIT(&leaf_uiter);
3334 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3335 struct extent_buffer *eb;
3337 eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL);
3342 ret = delete_v1_space_cache(eb, rc->block_group,
3343 extent_key->objectid);
3344 free_extent_buffer(eb);
3347 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3352 free_block_list(blocks);
3358 * helper to find next unprocessed extent
3360 static noinline_for_stack
3361 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3362 struct btrfs_key *extent_key)
3364 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3365 struct btrfs_key key;
3366 struct extent_buffer *leaf;
3367 u64 start, end, last;
3370 last = rc->block_group->start + rc->block_group->length;
3373 if (rc->search_start >= last) {
3378 key.objectid = rc->search_start;
3379 key.type = BTRFS_EXTENT_ITEM_KEY;
3382 path->search_commit_root = 1;
3383 path->skip_locking = 1;
3384 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3389 leaf = path->nodes[0];
3390 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3391 ret = btrfs_next_leaf(rc->extent_root, path);
3394 leaf = path->nodes[0];
3397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3398 if (key.objectid >= last) {
3403 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3404 key.type != BTRFS_METADATA_ITEM_KEY) {
3409 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3410 key.objectid + key.offset <= rc->search_start) {
3415 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3416 key.objectid + fs_info->nodesize <=
3422 ret = find_first_extent_bit(&rc->processed_blocks,
3423 key.objectid, &start, &end,
3424 EXTENT_DIRTY, NULL);
3426 if (ret == 0 && start <= key.objectid) {
3427 btrfs_release_path(path);
3428 rc->search_start = end + 1;
3430 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3431 rc->search_start = key.objectid + key.offset;
3433 rc->search_start = key.objectid +
3435 memcpy(extent_key, &key, sizeof(key));
3439 btrfs_release_path(path);
3443 static void set_reloc_control(struct reloc_control *rc)
3445 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3447 mutex_lock(&fs_info->reloc_mutex);
3448 fs_info->reloc_ctl = rc;
3449 mutex_unlock(&fs_info->reloc_mutex);
3452 static void unset_reloc_control(struct reloc_control *rc)
3454 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3456 mutex_lock(&fs_info->reloc_mutex);
3457 fs_info->reloc_ctl = NULL;
3458 mutex_unlock(&fs_info->reloc_mutex);
3461 static noinline_for_stack
3462 int prepare_to_relocate(struct reloc_control *rc)
3464 struct btrfs_trans_handle *trans;
3467 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3468 BTRFS_BLOCK_RSV_TEMP);
3472 memset(&rc->cluster, 0, sizeof(rc->cluster));
3473 rc->search_start = rc->block_group->start;
3474 rc->extents_found = 0;
3475 rc->nodes_relocated = 0;
3476 rc->merging_rsv_size = 0;
3477 rc->reserved_bytes = 0;
3478 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3479 RELOCATION_RESERVED_NODES;
3480 ret = btrfs_block_rsv_refill(rc->extent_root,
3481 rc->block_rsv, rc->block_rsv->size,
3482 BTRFS_RESERVE_FLUSH_ALL);
3486 rc->create_reloc_tree = 1;
3487 set_reloc_control(rc);
3489 trans = btrfs_join_transaction(rc->extent_root);
3490 if (IS_ERR(trans)) {
3491 unset_reloc_control(rc);
3493 * extent tree is not a ref_cow tree and has no reloc_root to
3494 * cleanup. And callers are responsible to free the above
3497 return PTR_ERR(trans);
3499 return btrfs_commit_transaction(trans);
3502 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3504 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3505 struct rb_root blocks = RB_ROOT;
3506 struct btrfs_key key;
3507 struct btrfs_trans_handle *trans = NULL;
3508 struct btrfs_path *path;
3509 struct btrfs_extent_item *ei;
3515 path = btrfs_alloc_path();
3518 path->reada = READA_FORWARD;
3520 ret = prepare_to_relocate(rc);
3527 rc->reserved_bytes = 0;
3528 ret = btrfs_block_rsv_refill(rc->extent_root,
3529 rc->block_rsv, rc->block_rsv->size,
3530 BTRFS_RESERVE_FLUSH_ALL);
3536 trans = btrfs_start_transaction(rc->extent_root, 0);
3537 if (IS_ERR(trans)) {
3538 err = PTR_ERR(trans);
3543 if (update_backref_cache(trans, &rc->backref_cache)) {
3544 btrfs_end_transaction(trans);
3549 ret = find_next_extent(rc, path, &key);
3555 rc->extents_found++;
3557 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3558 struct btrfs_extent_item);
3559 flags = btrfs_extent_flags(path->nodes[0], ei);
3561 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3562 ret = add_tree_block(rc, &key, path, &blocks);
3563 } else if (rc->stage == UPDATE_DATA_PTRS &&
3564 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3565 ret = add_data_references(rc, &key, path, &blocks);
3567 btrfs_release_path(path);
3575 if (!RB_EMPTY_ROOT(&blocks)) {
3576 ret = relocate_tree_blocks(trans, rc, &blocks);
3578 if (ret != -EAGAIN) {
3582 rc->extents_found--;
3583 rc->search_start = key.objectid;
3587 btrfs_end_transaction_throttle(trans);
3588 btrfs_btree_balance_dirty(fs_info);
3591 if (rc->stage == MOVE_DATA_EXTENTS &&
3592 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3593 rc->found_file_extent = 1;
3594 ret = relocate_data_extent(rc->data_inode,
3595 &key, &rc->cluster);
3601 if (btrfs_should_cancel_balance(fs_info)) {
3606 if (trans && progress && err == -ENOSPC) {
3607 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3615 btrfs_release_path(path);
3616 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3619 btrfs_end_transaction_throttle(trans);
3620 btrfs_btree_balance_dirty(fs_info);
3624 ret = relocate_file_extent_cluster(rc->data_inode,
3630 rc->create_reloc_tree = 0;
3631 set_reloc_control(rc);
3633 btrfs_backref_release_cache(&rc->backref_cache);
3634 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3637 * Even in the case when the relocation is cancelled, we should all go
3638 * through prepare_to_merge() and merge_reloc_roots().
3640 * For error (including cancelled balance), prepare_to_merge() will
3641 * mark all reloc trees orphan, then queue them for cleanup in
3642 * merge_reloc_roots()
3644 err = prepare_to_merge(rc, err);
3646 merge_reloc_roots(rc);
3648 rc->merge_reloc_tree = 0;
3649 unset_reloc_control(rc);
3650 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3652 /* get rid of pinned extents */
3653 trans = btrfs_join_transaction(rc->extent_root);
3654 if (IS_ERR(trans)) {
3655 err = PTR_ERR(trans);
3658 ret = btrfs_commit_transaction(trans);
3662 ret = clean_dirty_subvols(rc);
3663 if (ret < 0 && !err)
3665 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3666 btrfs_free_path(path);
3670 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3671 struct btrfs_root *root, u64 objectid)
3673 struct btrfs_path *path;
3674 struct btrfs_inode_item *item;
3675 struct extent_buffer *leaf;
3676 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
3679 if (btrfs_is_zoned(trans->fs_info))
3680 flags &= ~BTRFS_INODE_PREALLOC;
3682 path = btrfs_alloc_path();
3686 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3690 leaf = path->nodes[0];
3691 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3692 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3693 btrfs_set_inode_generation(leaf, item, 1);
3694 btrfs_set_inode_size(leaf, item, 0);
3695 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3696 btrfs_set_inode_flags(leaf, item, flags);
3697 btrfs_mark_buffer_dirty(leaf);
3699 btrfs_free_path(path);
3703 static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3704 struct btrfs_root *root, u64 objectid)
3706 struct btrfs_path *path;
3707 struct btrfs_key key;
3710 path = btrfs_alloc_path();
3716 key.objectid = objectid;
3717 key.type = BTRFS_INODE_ITEM_KEY;
3719 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3725 ret = btrfs_del_item(trans, root, path);
3728 btrfs_abort_transaction(trans, ret);
3729 btrfs_free_path(path);
3733 * helper to create inode for data relocation.
3734 * the inode is in data relocation tree and its link count is 0
3736 static noinline_for_stack
3737 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3738 struct btrfs_block_group *group)
3740 struct inode *inode = NULL;
3741 struct btrfs_trans_handle *trans;
3742 struct btrfs_root *root;
3746 root = btrfs_grab_root(fs_info->data_reloc_root);
3747 trans = btrfs_start_transaction(root, 6);
3748 if (IS_ERR(trans)) {
3749 btrfs_put_root(root);
3750 return ERR_CAST(trans);
3753 err = btrfs_get_free_objectid(root, &objectid);
3757 err = __insert_orphan_inode(trans, root, objectid);
3761 inode = btrfs_iget(fs_info->sb, objectid, root);
3762 if (IS_ERR(inode)) {
3763 delete_orphan_inode(trans, root, objectid);
3764 err = PTR_ERR(inode);
3768 BTRFS_I(inode)->index_cnt = group->start;
3770 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3772 btrfs_put_root(root);
3773 btrfs_end_transaction(trans);
3774 btrfs_btree_balance_dirty(fs_info);
3778 inode = ERR_PTR(err);
3783 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3785 struct reloc_control *rc;
3787 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3791 INIT_LIST_HEAD(&rc->reloc_roots);
3792 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3793 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3794 mapping_tree_init(&rc->reloc_root_tree);
3795 extent_io_tree_init(fs_info, &rc->processed_blocks,
3796 IO_TREE_RELOC_BLOCKS, NULL);
3800 static void free_reloc_control(struct reloc_control *rc)
3802 struct mapping_node *node, *tmp;
3804 free_reloc_roots(&rc->reloc_roots);
3805 rbtree_postorder_for_each_entry_safe(node, tmp,
3806 &rc->reloc_root_tree.rb_root, rb_node)
3813 * Print the block group being relocated
3815 static void describe_relocation(struct btrfs_fs_info *fs_info,
3816 struct btrfs_block_group *block_group)
3818 char buf[128] = {'\0'};
3820 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3823 "relocating block group %llu flags %s",
3824 block_group->start, buf);
3827 static const char *stage_to_string(int stage)
3829 if (stage == MOVE_DATA_EXTENTS)
3830 return "move data extents";
3831 if (stage == UPDATE_DATA_PTRS)
3832 return "update data pointers";
3837 * function to relocate all extents in a block group.
3839 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3841 struct btrfs_block_group *bg;
3842 struct btrfs_root *extent_root = fs_info->extent_root;
3843 struct reloc_control *rc;
3844 struct inode *inode;
3845 struct btrfs_path *path;
3850 bg = btrfs_lookup_block_group(fs_info, group_start);
3854 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3855 btrfs_put_block_group(bg);
3859 rc = alloc_reloc_control(fs_info);
3861 btrfs_put_block_group(bg);
3865 rc->extent_root = extent_root;
3866 rc->block_group = bg;
3868 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3875 path = btrfs_alloc_path();
3881 inode = lookup_free_space_inode(rc->block_group, path);
3882 btrfs_free_path(path);
3885 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3887 ret = PTR_ERR(inode);
3889 if (ret && ret != -ENOENT) {
3894 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3895 if (IS_ERR(rc->data_inode)) {
3896 err = PTR_ERR(rc->data_inode);
3897 rc->data_inode = NULL;
3901 describe_relocation(fs_info, rc->block_group);
3903 btrfs_wait_block_group_reservations(rc->block_group);
3904 btrfs_wait_nocow_writers(rc->block_group);
3905 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3906 rc->block_group->start,
3907 rc->block_group->length);
3912 mutex_lock(&fs_info->cleaner_mutex);
3913 ret = relocate_block_group(rc);
3914 mutex_unlock(&fs_info->cleaner_mutex);
3918 finishes_stage = rc->stage;
3920 * We may have gotten ENOSPC after we already dirtied some
3921 * extents. If writeout happens while we're relocating a
3922 * different block group we could end up hitting the
3923 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3924 * btrfs_reloc_cow_block. Make sure we write everything out
3925 * properly so we don't trip over this problem, and then break
3926 * out of the loop if we hit an error.
3928 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3929 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3933 invalidate_mapping_pages(rc->data_inode->i_mapping,
3935 rc->stage = UPDATE_DATA_PTRS;
3941 if (rc->extents_found == 0)
3944 btrfs_info(fs_info, "found %llu extents, stage: %s",
3945 rc->extents_found, stage_to_string(finishes_stage));
3948 WARN_ON(rc->block_group->pinned > 0);
3949 WARN_ON(rc->block_group->reserved > 0);
3950 WARN_ON(rc->block_group->used > 0);
3953 btrfs_dec_block_group_ro(rc->block_group);
3954 iput(rc->data_inode);
3955 btrfs_put_block_group(rc->block_group);
3956 free_reloc_control(rc);
3960 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3962 struct btrfs_fs_info *fs_info = root->fs_info;
3963 struct btrfs_trans_handle *trans;
3966 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3968 return PTR_ERR(trans);
3970 memset(&root->root_item.drop_progress, 0,
3971 sizeof(root->root_item.drop_progress));
3972 btrfs_set_root_drop_level(&root->root_item, 0);
3973 btrfs_set_root_refs(&root->root_item, 0);
3974 ret = btrfs_update_root(trans, fs_info->tree_root,
3975 &root->root_key, &root->root_item);
3977 err = btrfs_end_transaction(trans);
3984 * recover relocation interrupted by system crash.
3986 * this function resumes merging reloc trees with corresponding fs trees.
3987 * this is important for keeping the sharing of tree blocks
3989 int btrfs_recover_relocation(struct btrfs_root *root)
3991 struct btrfs_fs_info *fs_info = root->fs_info;
3992 LIST_HEAD(reloc_roots);
3993 struct btrfs_key key;
3994 struct btrfs_root *fs_root;
3995 struct btrfs_root *reloc_root;
3996 struct btrfs_path *path;
3997 struct extent_buffer *leaf;
3998 struct reloc_control *rc = NULL;
3999 struct btrfs_trans_handle *trans;
4003 path = btrfs_alloc_path();
4006 path->reada = READA_BACK;
4008 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4009 key.type = BTRFS_ROOT_ITEM_KEY;
4010 key.offset = (u64)-1;
4013 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4020 if (path->slots[0] == 0)
4024 leaf = path->nodes[0];
4025 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4026 btrfs_release_path(path);
4028 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4029 key.type != BTRFS_ROOT_ITEM_KEY)
4032 reloc_root = btrfs_read_tree_root(root, &key);
4033 if (IS_ERR(reloc_root)) {
4034 err = PTR_ERR(reloc_root);
4038 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4039 list_add(&reloc_root->root_list, &reloc_roots);
4041 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4042 fs_root = btrfs_get_fs_root(fs_info,
4043 reloc_root->root_key.offset, false);
4044 if (IS_ERR(fs_root)) {
4045 ret = PTR_ERR(fs_root);
4046 if (ret != -ENOENT) {
4050 ret = mark_garbage_root(reloc_root);
4056 btrfs_put_root(fs_root);
4060 if (key.offset == 0)
4065 btrfs_release_path(path);
4067 if (list_empty(&reloc_roots))
4070 rc = alloc_reloc_control(fs_info);
4076 rc->extent_root = fs_info->extent_root;
4078 set_reloc_control(rc);
4080 trans = btrfs_join_transaction(rc->extent_root);
4081 if (IS_ERR(trans)) {
4082 err = PTR_ERR(trans);
4086 rc->merge_reloc_tree = 1;
4088 while (!list_empty(&reloc_roots)) {
4089 reloc_root = list_entry(reloc_roots.next,
4090 struct btrfs_root, root_list);
4091 list_del(&reloc_root->root_list);
4093 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4094 list_add_tail(&reloc_root->root_list,
4099 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4101 if (IS_ERR(fs_root)) {
4102 err = PTR_ERR(fs_root);
4103 list_add_tail(&reloc_root->root_list, &reloc_roots);
4104 btrfs_end_transaction(trans);
4108 err = __add_reloc_root(reloc_root);
4109 ASSERT(err != -EEXIST);
4111 list_add_tail(&reloc_root->root_list, &reloc_roots);
4112 btrfs_put_root(fs_root);
4113 btrfs_end_transaction(trans);
4116 fs_root->reloc_root = btrfs_grab_root(reloc_root);
4117 btrfs_put_root(fs_root);
4120 err = btrfs_commit_transaction(trans);
4124 merge_reloc_roots(rc);
4126 unset_reloc_control(rc);
4128 trans = btrfs_join_transaction(rc->extent_root);
4129 if (IS_ERR(trans)) {
4130 err = PTR_ERR(trans);
4133 err = btrfs_commit_transaction(trans);
4135 ret = clean_dirty_subvols(rc);
4136 if (ret < 0 && !err)
4139 unset_reloc_control(rc);
4140 free_reloc_control(rc);
4142 free_reloc_roots(&reloc_roots);
4144 btrfs_free_path(path);
4147 /* cleanup orphan inode in data relocation tree */
4148 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4150 err = btrfs_orphan_cleanup(fs_root);
4151 btrfs_put_root(fs_root);
4157 * helper to add ordered checksum for data relocation.
4159 * cloning checksum properly handles the nodatasum extents.
4160 * it also saves CPU time to re-calculate the checksum.
4162 int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
4164 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4165 struct btrfs_ordered_sum *sums;
4166 struct btrfs_ordered_extent *ordered;
4172 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4173 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
4175 disk_bytenr = file_pos + inode->index_cnt;
4176 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4177 disk_bytenr + len - 1, &list, 0);
4181 while (!list_empty(&list)) {
4182 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4183 list_del_init(&sums->list);
4186 * We need to offset the new_bytenr based on where the csum is.
4187 * We need to do this because we will read in entire prealloc
4188 * extents but we may have written to say the middle of the
4189 * prealloc extent, so we need to make sure the csum goes with
4190 * the right disk offset.
4192 * We can do this because the data reloc inode refers strictly
4193 * to the on disk bytes, so we don't have to worry about
4194 * disk_len vs real len like with real inodes since it's all
4197 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
4198 sums->bytenr = new_bytenr;
4200 btrfs_add_ordered_sum(ordered, sums);
4203 btrfs_put_ordered_extent(ordered);
4207 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root, struct extent_buffer *buf,
4209 struct extent_buffer *cow)
4211 struct btrfs_fs_info *fs_info = root->fs_info;
4212 struct reloc_control *rc;
4213 struct btrfs_backref_node *node;
4218 rc = fs_info->reloc_ctl;
4222 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4223 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4225 level = btrfs_header_level(buf);
4226 if (btrfs_header_generation(buf) <=
4227 btrfs_root_last_snapshot(&root->root_item))
4230 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4231 rc->create_reloc_tree) {
4232 WARN_ON(!first_cow && level == 0);
4234 node = rc->backref_cache.path[level];
4235 BUG_ON(node->bytenr != buf->start &&
4236 node->new_bytenr != buf->start);
4238 btrfs_backref_drop_node_buffer(node);
4239 atomic_inc(&cow->refs);
4241 node->new_bytenr = cow->start;
4243 if (!node->pending) {
4244 list_move_tail(&node->list,
4245 &rc->backref_cache.pending[level]);
4250 mark_block_processed(rc, node);
4252 if (first_cow && level > 0)
4253 rc->nodes_relocated += buf->len;
4256 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4257 ret = replace_file_extents(trans, rc, root, cow);
4262 * called before creating snapshot. it calculates metadata reservation
4263 * required for relocating tree blocks in the snapshot
4265 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4266 u64 *bytes_to_reserve)
4268 struct btrfs_root *root = pending->root;
4269 struct reloc_control *rc = root->fs_info->reloc_ctl;
4271 if (!rc || !have_reloc_root(root))
4274 if (!rc->merge_reloc_tree)
4277 root = root->reloc_root;
4278 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4280 * relocation is in the stage of merging trees. the space
4281 * used by merging a reloc tree is twice the size of
4282 * relocated tree nodes in the worst case. half for cowing
4283 * the reloc tree, half for cowing the fs tree. the space
4284 * used by cowing the reloc tree will be freed after the
4285 * tree is dropped. if we create snapshot, cowing the fs
4286 * tree may use more space than it frees. so we need
4287 * reserve extra space.
4289 *bytes_to_reserve += rc->nodes_relocated;
4293 * called after snapshot is created. migrate block reservation
4294 * and create reloc root for the newly created snapshot
4296 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4297 * references held on the reloc_root, one for root->reloc_root and one for
4300 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4301 struct btrfs_pending_snapshot *pending)
4303 struct btrfs_root *root = pending->root;
4304 struct btrfs_root *reloc_root;
4305 struct btrfs_root *new_root;
4306 struct reloc_control *rc = root->fs_info->reloc_ctl;
4309 if (!rc || !have_reloc_root(root))
4312 rc = root->fs_info->reloc_ctl;
4313 rc->merging_rsv_size += rc->nodes_relocated;
4315 if (rc->merge_reloc_tree) {
4316 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4318 rc->nodes_relocated, true);
4323 new_root = pending->snap;
4324 reloc_root = create_reloc_root(trans, root->reloc_root,
4325 new_root->root_key.objectid);
4326 if (IS_ERR(reloc_root))
4327 return PTR_ERR(reloc_root);
4329 ret = __add_reloc_root(reloc_root);
4330 ASSERT(ret != -EEXIST);
4332 /* Pairs with create_reloc_root */
4333 btrfs_put_root(reloc_root);
4336 new_root->reloc_root = btrfs_grab_root(reloc_root);
4338 if (rc->create_reloc_tree)
4339 ret = clone_backref_node(trans, rc, root, reloc_root);