1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2009 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/error-injection.h>
15 #include "transaction.h"
18 #include "btrfs_inode.h"
19 #include "async-thread.h"
20 #include "free-space-cache.h"
21 #include "inode-map.h"
23 #include "print-tree.h"
24 #include "delalloc-space.h"
25 #include "block-group.h"
32 * [What does relocation do]
34 * The objective of relocation is to relocate all extents of the target block
35 * group to other block groups.
36 * This is utilized by resize (shrink only), profile converting, compacting
37 * space, or balance routine to spread chunks over devices.
40 * ------------------------------------------------------------------
41 * BG A: 10 data extents | BG A: deleted
42 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
43 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
45 * [How does relocation work]
47 * 1. Mark the target block group read-only
48 * New extents won't be allocated from the target block group.
50 * 2.1 Record each extent in the target block group
51 * To build a proper map of extents to be relocated.
53 * 2.2 Build data reloc tree and reloc trees
54 * Data reloc tree will contain an inode, recording all newly relocated
56 * There will be only one data reloc tree for one data block group.
58 * Reloc tree will be a special snapshot of its source tree, containing
59 * relocated tree blocks.
60 * Each tree referring to a tree block in target block group will get its
63 * 2.3 Swap source tree with its corresponding reloc tree
64 * Each involved tree only refers to new extents after swap.
66 * 3. Cleanup reloc trees and data reloc tree.
67 * As old extents in the target block group are still referenced by reloc
68 * trees, we need to clean them up before really freeing the target block
71 * The main complexity is in steps 2.2 and 2.3.
73 * The entry point of relocation is relocate_block_group() function.
76 #define RELOCATION_RESERVED_NODES 256
78 * map address of tree root to tree
82 struct rb_node rb_node;
84 }; /* Use rb_simle_node for search/insert */
89 struct rb_root rb_root;
94 * present a tree block to process
98 struct rb_node rb_node;
100 }; /* Use rb_simple_node for search/insert */
101 struct btrfs_key key;
102 unsigned int level:8;
103 unsigned int key_ready:1;
106 #define MAX_EXTENTS 128
108 struct file_extent_cluster {
111 u64 boundary[MAX_EXTENTS];
115 struct reloc_control {
116 /* block group to relocate */
117 struct btrfs_block_group *block_group;
119 struct btrfs_root *extent_root;
120 /* inode for moving data */
121 struct inode *data_inode;
123 struct btrfs_block_rsv *block_rsv;
125 struct btrfs_backref_cache backref_cache;
127 struct file_extent_cluster cluster;
128 /* tree blocks have been processed */
129 struct extent_io_tree processed_blocks;
130 /* map start of tree root to corresponding reloc tree */
131 struct mapping_tree reloc_root_tree;
132 /* list of reloc trees */
133 struct list_head reloc_roots;
134 /* list of subvolume trees that get relocated */
135 struct list_head dirty_subvol_roots;
136 /* size of metadata reservation for merging reloc trees */
137 u64 merging_rsv_size;
138 /* size of relocated tree nodes */
140 /* reserved size for block group relocation*/
146 unsigned int stage:8;
147 unsigned int create_reloc_tree:1;
148 unsigned int merge_reloc_tree:1;
149 unsigned int found_file_extent:1;
152 /* stages of data relocation */
153 #define MOVE_DATA_EXTENTS 0
154 #define UPDATE_DATA_PTRS 1
156 static void mark_block_processed(struct reloc_control *rc,
157 struct btrfs_backref_node *node)
161 if (node->level == 0 ||
162 in_range(node->bytenr, rc->block_group->start,
163 rc->block_group->length)) {
164 blocksize = rc->extent_root->fs_info->nodesize;
165 set_extent_bits(&rc->processed_blocks, node->bytenr,
166 node->bytenr + blocksize - 1, EXTENT_DIRTY);
172 static void mapping_tree_init(struct mapping_tree *tree)
174 tree->rb_root = RB_ROOT;
175 spin_lock_init(&tree->lock);
179 * walk up backref nodes until reach node presents tree root
181 static struct btrfs_backref_node *walk_up_backref(
182 struct btrfs_backref_node *node,
183 struct btrfs_backref_edge *edges[], int *index)
185 struct btrfs_backref_edge *edge;
188 while (!list_empty(&node->upper)) {
189 edge = list_entry(node->upper.next,
190 struct btrfs_backref_edge, list[LOWER]);
192 node = edge->node[UPPER];
194 BUG_ON(node->detached);
200 * walk down backref nodes to find start of next reference path
202 static struct btrfs_backref_node *walk_down_backref(
203 struct btrfs_backref_edge *edges[], int *index)
205 struct btrfs_backref_edge *edge;
206 struct btrfs_backref_node *lower;
210 edge = edges[idx - 1];
211 lower = edge->node[LOWER];
212 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
216 edge = list_entry(edge->list[LOWER].next,
217 struct btrfs_backref_edge, list[LOWER]);
218 edges[idx - 1] = edge;
220 return edge->node[UPPER];
226 static void update_backref_node(struct btrfs_backref_cache *cache,
227 struct btrfs_backref_node *node, u64 bytenr)
229 struct rb_node *rb_node;
230 rb_erase(&node->rb_node, &cache->rb_root);
231 node->bytenr = bytenr;
232 rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node);
234 btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
238 * update backref cache after a transaction commit
240 static int update_backref_cache(struct btrfs_trans_handle *trans,
241 struct btrfs_backref_cache *cache)
243 struct btrfs_backref_node *node;
246 if (cache->last_trans == 0) {
247 cache->last_trans = trans->transid;
251 if (cache->last_trans == trans->transid)
255 * detached nodes are used to avoid unnecessary backref
256 * lookup. transaction commit changes the extent tree.
257 * so the detached nodes are no longer useful.
259 while (!list_empty(&cache->detached)) {
260 node = list_entry(cache->detached.next,
261 struct btrfs_backref_node, list);
262 btrfs_backref_cleanup_node(cache, node);
265 while (!list_empty(&cache->changed)) {
266 node = list_entry(cache->changed.next,
267 struct btrfs_backref_node, list);
268 list_del_init(&node->list);
269 BUG_ON(node->pending);
270 update_backref_node(cache, node, node->new_bytenr);
274 * some nodes can be left in the pending list if there were
275 * errors during processing the pending nodes.
277 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
278 list_for_each_entry(node, &cache->pending[level], list) {
279 BUG_ON(!node->pending);
280 if (node->bytenr == node->new_bytenr)
282 update_backref_node(cache, node, node->new_bytenr);
286 cache->last_trans = 0;
290 static bool reloc_root_is_dead(struct btrfs_root *root)
293 * Pair with set_bit/clear_bit in clean_dirty_subvols and
294 * btrfs_update_reloc_root. We need to see the updated bit before
295 * trying to access reloc_root
298 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
304 * Check if this subvolume tree has valid reloc tree.
306 * Reloc tree after swap is considered dead, thus not considered as valid.
307 * This is enough for most callers, as they don't distinguish dead reloc root
308 * from no reloc root. But btrfs_should_ignore_reloc_root() below is a
311 static bool have_reloc_root(struct btrfs_root *root)
313 if (reloc_root_is_dead(root))
315 if (!root->reloc_root)
320 int btrfs_should_ignore_reloc_root(struct btrfs_root *root)
322 struct btrfs_root *reloc_root;
324 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
327 /* This root has been merged with its reloc tree, we can ignore it */
328 if (reloc_root_is_dead(root))
331 reloc_root = root->reloc_root;
335 if (btrfs_header_generation(reloc_root->commit_root) ==
336 root->fs_info->running_transaction->transid)
339 * if there is reloc tree and it was created in previous
340 * transaction backref lookup can find the reloc tree,
341 * so backref node for the fs tree root is useless for
348 * find reloc tree by address of tree root
350 struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
352 struct reloc_control *rc = fs_info->reloc_ctl;
353 struct rb_node *rb_node;
354 struct mapping_node *node;
355 struct btrfs_root *root = NULL;
358 spin_lock(&rc->reloc_root_tree.lock);
359 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
361 node = rb_entry(rb_node, struct mapping_node, rb_node);
362 root = (struct btrfs_root *)node->data;
364 spin_unlock(&rc->reloc_root_tree.lock);
365 return btrfs_grab_root(root);
369 * For useless nodes, do two major clean ups:
371 * - Cleanup the children edges and nodes
372 * If child node is also orphan (no parent) during cleanup, then the child
373 * node will also be cleaned up.
375 * - Freeing up leaves (level 0), keeps nodes detached
376 * For nodes, the node is still cached as "detached"
378 * Return false if @node is not in the @useless_nodes list.
379 * Return true if @node is in the @useless_nodes list.
381 static bool handle_useless_nodes(struct reloc_control *rc,
382 struct btrfs_backref_node *node)
384 struct btrfs_backref_cache *cache = &rc->backref_cache;
385 struct list_head *useless_node = &cache->useless_node;
388 while (!list_empty(useless_node)) {
389 struct btrfs_backref_node *cur;
391 cur = list_first_entry(useless_node, struct btrfs_backref_node,
393 list_del_init(&cur->list);
395 /* Only tree root nodes can be added to @useless_nodes */
396 ASSERT(list_empty(&cur->upper));
401 /* The node is the lowest node */
403 list_del_init(&cur->lower);
407 /* Cleanup the lower edges */
408 while (!list_empty(&cur->lower)) {
409 struct btrfs_backref_edge *edge;
410 struct btrfs_backref_node *lower;
412 edge = list_entry(cur->lower.next,
413 struct btrfs_backref_edge, list[UPPER]);
414 list_del(&edge->list[UPPER]);
415 list_del(&edge->list[LOWER]);
416 lower = edge->node[LOWER];
417 btrfs_backref_free_edge(cache, edge);
419 /* Child node is also orphan, queue for cleanup */
420 if (list_empty(&lower->upper))
421 list_add(&lower->list, useless_node);
423 /* Mark this block processed for relocation */
424 mark_block_processed(rc, cur);
427 * Backref nodes for tree leaves are deleted from the cache.
428 * Backref nodes for upper level tree blocks are left in the
429 * cache to avoid unnecessary backref lookup.
431 if (cur->level > 0) {
432 list_add(&cur->list, &cache->detached);
435 rb_erase(&cur->rb_node, &cache->rb_root);
436 btrfs_backref_free_node(cache, cur);
443 * Build backref tree for a given tree block. Root of the backref tree
444 * corresponds the tree block, leaves of the backref tree correspond roots of
445 * b-trees that reference the tree block.
447 * The basic idea of this function is check backrefs of a given block to find
448 * upper level blocks that reference the block, and then check backrefs of
449 * these upper level blocks recursively. The recursion stops when tree root is
450 * reached or backrefs for the block is cached.
452 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
453 * all upper level blocks that directly/indirectly reference the block are also
456 static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
457 struct reloc_control *rc, struct btrfs_key *node_key,
458 int level, u64 bytenr)
460 struct btrfs_backref_iter *iter;
461 struct btrfs_backref_cache *cache = &rc->backref_cache;
462 /* For searching parent of TREE_BLOCK_REF */
463 struct btrfs_path *path;
464 struct btrfs_backref_node *cur;
465 struct btrfs_backref_node *node = NULL;
466 struct btrfs_backref_edge *edge;
470 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS);
472 return ERR_PTR(-ENOMEM);
473 path = btrfs_alloc_path();
479 node = btrfs_backref_alloc_node(cache, bytenr, level);
488 /* Breadth-first search to build backref cache */
490 ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
496 edge = list_first_entry_or_null(&cache->pending_edge,
497 struct btrfs_backref_edge, list[UPPER]);
499 * The pending list isn't empty, take the first block to
503 list_del_init(&edge->list[UPPER]);
504 cur = edge->node[UPPER];
508 /* Finish the upper linkage of newly added edges/nodes */
509 ret = btrfs_backref_finish_upper_links(cache, node);
515 if (handle_useless_nodes(rc, node))
518 btrfs_backref_iter_free(iter);
519 btrfs_free_path(path);
521 btrfs_backref_error_cleanup(cache, node);
524 ASSERT(!node || !node->detached);
525 ASSERT(list_empty(&cache->useless_node) &&
526 list_empty(&cache->pending_edge));
531 * helper to add backref node for the newly created snapshot.
532 * the backref node is created by cloning backref node that
533 * corresponds to root of source tree
535 static int clone_backref_node(struct btrfs_trans_handle *trans,
536 struct reloc_control *rc,
537 struct btrfs_root *src,
538 struct btrfs_root *dest)
540 struct btrfs_root *reloc_root = src->reloc_root;
541 struct btrfs_backref_cache *cache = &rc->backref_cache;
542 struct btrfs_backref_node *node = NULL;
543 struct btrfs_backref_node *new_node;
544 struct btrfs_backref_edge *edge;
545 struct btrfs_backref_edge *new_edge;
546 struct rb_node *rb_node;
548 if (cache->last_trans > 0)
549 update_backref_cache(trans, cache);
551 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
553 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
557 BUG_ON(node->new_bytenr != reloc_root->node->start);
561 rb_node = rb_simple_search(&cache->rb_root,
562 reloc_root->commit_root->start);
564 node = rb_entry(rb_node, struct btrfs_backref_node,
566 BUG_ON(node->detached);
573 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
578 new_node->lowest = node->lowest;
579 new_node->checked = 1;
580 new_node->root = btrfs_grab_root(dest);
581 ASSERT(new_node->root);
584 list_for_each_entry(edge, &node->lower, list[UPPER]) {
585 new_edge = btrfs_backref_alloc_edge(cache);
589 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
590 new_node, LINK_UPPER);
593 list_add_tail(&new_node->lower, &cache->leaves);
596 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
599 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
601 if (!new_node->lowest) {
602 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
603 list_add_tail(&new_edge->list[LOWER],
604 &new_edge->node[LOWER]->upper);
609 while (!list_empty(&new_node->lower)) {
610 new_edge = list_entry(new_node->lower.next,
611 struct btrfs_backref_edge, list[UPPER]);
612 list_del(&new_edge->list[UPPER]);
613 btrfs_backref_free_edge(cache, new_edge);
615 btrfs_backref_free_node(cache, new_node);
620 * helper to add 'address of tree root -> reloc tree' mapping
622 static int __must_check __add_reloc_root(struct btrfs_root *root)
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 struct rb_node *rb_node;
626 struct mapping_node *node;
627 struct reloc_control *rc = fs_info->reloc_ctl;
629 node = kmalloc(sizeof(*node), GFP_NOFS);
633 node->bytenr = root->commit_root->start;
636 spin_lock(&rc->reloc_root_tree.lock);
637 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
638 node->bytenr, &node->rb_node);
639 spin_unlock(&rc->reloc_root_tree.lock);
641 btrfs_panic(fs_info, -EEXIST,
642 "Duplicate root found for start=%llu while inserting into relocation tree",
646 list_add_tail(&root->root_list, &rc->reloc_roots);
651 * helper to delete the 'address of tree root -> reloc tree'
654 static void __del_reloc_root(struct btrfs_root *root)
656 struct btrfs_fs_info *fs_info = root->fs_info;
657 struct rb_node *rb_node;
658 struct mapping_node *node = NULL;
659 struct reloc_control *rc = fs_info->reloc_ctl;
660 bool put_ref = false;
662 if (rc && root->node) {
663 spin_lock(&rc->reloc_root_tree.lock);
664 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
665 root->commit_root->start);
667 node = rb_entry(rb_node, struct mapping_node, rb_node);
668 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
669 RB_CLEAR_NODE(&node->rb_node);
671 spin_unlock(&rc->reloc_root_tree.lock);
674 BUG_ON((struct btrfs_root *)node->data != root);
678 * We only put the reloc root here if it's on the list. There's a lot
679 * of places where the pattern is to splice the rc->reloc_roots, process
680 * the reloc roots, and then add the reloc root back onto
681 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
682 * list we don't want the reference being dropped, because the guy
683 * messing with the list is in charge of the reference.
685 spin_lock(&fs_info->trans_lock);
686 if (!list_empty(&root->root_list)) {
688 list_del_init(&root->root_list);
690 spin_unlock(&fs_info->trans_lock);
692 btrfs_put_root(root);
697 * helper to update the 'address of tree root -> reloc tree'
700 static int __update_reloc_root(struct btrfs_root *root)
702 struct btrfs_fs_info *fs_info = root->fs_info;
703 struct rb_node *rb_node;
704 struct mapping_node *node = NULL;
705 struct reloc_control *rc = fs_info->reloc_ctl;
707 spin_lock(&rc->reloc_root_tree.lock);
708 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
709 root->commit_root->start);
711 node = rb_entry(rb_node, struct mapping_node, rb_node);
712 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
714 spin_unlock(&rc->reloc_root_tree.lock);
718 BUG_ON((struct btrfs_root *)node->data != root);
720 spin_lock(&rc->reloc_root_tree.lock);
721 node->bytenr = root->node->start;
722 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
723 node->bytenr, &node->rb_node);
724 spin_unlock(&rc->reloc_root_tree.lock);
726 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
730 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
731 struct btrfs_root *root, u64 objectid)
733 struct btrfs_fs_info *fs_info = root->fs_info;
734 struct btrfs_root *reloc_root;
735 struct extent_buffer *eb;
736 struct btrfs_root_item *root_item;
737 struct btrfs_key root_key;
740 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
743 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
744 root_key.type = BTRFS_ROOT_ITEM_KEY;
745 root_key.offset = objectid;
747 if (root->root_key.objectid == objectid) {
750 /* called by btrfs_init_reloc_root */
751 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
752 BTRFS_TREE_RELOC_OBJECTID);
755 * Set the last_snapshot field to the generation of the commit
756 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
757 * correctly (returns true) when the relocation root is created
758 * either inside the critical section of a transaction commit
759 * (through transaction.c:qgroup_account_snapshot()) and when
760 * it's created before the transaction commit is started.
762 commit_root_gen = btrfs_header_generation(root->commit_root);
763 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
766 * called by btrfs_reloc_post_snapshot_hook.
767 * the source tree is a reloc tree, all tree blocks
768 * modified after it was created have RELOC flag
769 * set in their headers. so it's OK to not update
770 * the 'last_snapshot'.
772 ret = btrfs_copy_root(trans, root, root->node, &eb,
773 BTRFS_TREE_RELOC_OBJECTID);
777 memcpy(root_item, &root->root_item, sizeof(*root_item));
778 btrfs_set_root_bytenr(root_item, eb->start);
779 btrfs_set_root_level(root_item, btrfs_header_level(eb));
780 btrfs_set_root_generation(root_item, trans->transid);
782 if (root->root_key.objectid == objectid) {
783 btrfs_set_root_refs(root_item, 0);
784 memset(&root_item->drop_progress, 0,
785 sizeof(struct btrfs_disk_key));
786 root_item->drop_level = 0;
789 btrfs_tree_unlock(eb);
790 free_extent_buffer(eb);
792 ret = btrfs_insert_root(trans, fs_info->tree_root,
793 &root_key, root_item);
797 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
798 BUG_ON(IS_ERR(reloc_root));
799 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
800 reloc_root->last_trans = trans->transid;
805 * create reloc tree for a given fs tree. reloc tree is just a
806 * snapshot of the fs tree with special root objectid.
808 * The reloc_root comes out of here with two references, one for
809 * root->reloc_root, and another for being on the rc->reloc_roots list.
811 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
812 struct btrfs_root *root)
814 struct btrfs_fs_info *fs_info = root->fs_info;
815 struct btrfs_root *reloc_root;
816 struct reloc_control *rc = fs_info->reloc_ctl;
817 struct btrfs_block_rsv *rsv;
825 * The subvolume has reloc tree but the swap is finished, no need to
826 * create/update the dead reloc tree
828 if (reloc_root_is_dead(root))
832 * This is subtle but important. We do not do
833 * record_root_in_transaction for reloc roots, instead we record their
834 * corresponding fs root, and then here we update the last trans for the
835 * reloc root. This means that we have to do this for the entire life
836 * of the reloc root, regardless of which stage of the relocation we are
839 if (root->reloc_root) {
840 reloc_root = root->reloc_root;
841 reloc_root->last_trans = trans->transid;
846 * We are merging reloc roots, we do not need new reloc trees. Also
847 * reloc trees never need their own reloc tree.
849 if (!rc->create_reloc_tree ||
850 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
853 if (!trans->reloc_reserved) {
854 rsv = trans->block_rsv;
855 trans->block_rsv = rc->block_rsv;
858 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
860 trans->block_rsv = rsv;
862 ret = __add_reloc_root(reloc_root);
864 root->reloc_root = btrfs_grab_root(reloc_root);
869 * update root item of reloc tree
871 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
872 struct btrfs_root *root)
874 struct btrfs_fs_info *fs_info = root->fs_info;
875 struct btrfs_root *reloc_root;
876 struct btrfs_root_item *root_item;
879 if (!have_reloc_root(root))
882 reloc_root = root->reloc_root;
883 root_item = &reloc_root->root_item;
886 * We are probably ok here, but __del_reloc_root() will drop its ref of
887 * the root. We have the ref for root->reloc_root, but just in case
888 * hold it while we update the reloc root.
890 btrfs_grab_root(reloc_root);
892 /* root->reloc_root will stay until current relocation finished */
893 if (fs_info->reloc_ctl->merge_reloc_tree &&
894 btrfs_root_refs(root_item) == 0) {
895 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
897 * Mark the tree as dead before we change reloc_root so
898 * have_reloc_root will not touch it from now on.
901 __del_reloc_root(reloc_root);
904 if (reloc_root->commit_root != reloc_root->node) {
905 __update_reloc_root(reloc_root);
906 btrfs_set_root_node(root_item, reloc_root->node);
907 free_extent_buffer(reloc_root->commit_root);
908 reloc_root->commit_root = btrfs_root_node(reloc_root);
911 ret = btrfs_update_root(trans, fs_info->tree_root,
912 &reloc_root->root_key, root_item);
914 btrfs_put_root(reloc_root);
920 * helper to find first cached inode with inode number >= objectid
923 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
925 struct rb_node *node;
926 struct rb_node *prev;
927 struct btrfs_inode *entry;
930 spin_lock(&root->inode_lock);
932 node = root->inode_tree.rb_node;
936 entry = rb_entry(node, struct btrfs_inode, rb_node);
938 if (objectid < btrfs_ino(entry))
939 node = node->rb_left;
940 else if (objectid > btrfs_ino(entry))
941 node = node->rb_right;
947 entry = rb_entry(prev, struct btrfs_inode, rb_node);
948 if (objectid <= btrfs_ino(entry)) {
952 prev = rb_next(prev);
956 entry = rb_entry(node, struct btrfs_inode, rb_node);
957 inode = igrab(&entry->vfs_inode);
959 spin_unlock(&root->inode_lock);
963 objectid = btrfs_ino(entry) + 1;
964 if (cond_resched_lock(&root->inode_lock))
967 node = rb_next(node);
969 spin_unlock(&root->inode_lock);
974 * get new location of data
976 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
977 u64 bytenr, u64 num_bytes)
979 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
980 struct btrfs_path *path;
981 struct btrfs_file_extent_item *fi;
982 struct extent_buffer *leaf;
985 path = btrfs_alloc_path();
989 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
990 ret = btrfs_lookup_file_extent(NULL, root, path,
991 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
999 leaf = path->nodes[0];
1000 fi = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_file_extent_item);
1003 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1004 btrfs_file_extent_compression(leaf, fi) ||
1005 btrfs_file_extent_encryption(leaf, fi) ||
1006 btrfs_file_extent_other_encoding(leaf, fi));
1008 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1013 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1016 btrfs_free_path(path);
1021 * update file extent items in the tree leaf to point to
1022 * the new locations.
1024 static noinline_for_stack
1025 int replace_file_extents(struct btrfs_trans_handle *trans,
1026 struct reloc_control *rc,
1027 struct btrfs_root *root,
1028 struct extent_buffer *leaf)
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1031 struct btrfs_key key;
1032 struct btrfs_file_extent_item *fi;
1033 struct inode *inode = NULL;
1045 if (rc->stage != UPDATE_DATA_PTRS)
1048 /* reloc trees always use full backref */
1049 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1050 parent = leaf->start;
1054 nritems = btrfs_header_nritems(leaf);
1055 for (i = 0; i < nritems; i++) {
1056 struct btrfs_ref ref = { 0 };
1059 btrfs_item_key_to_cpu(leaf, &key, i);
1060 if (key.type != BTRFS_EXTENT_DATA_KEY)
1062 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1063 if (btrfs_file_extent_type(leaf, fi) ==
1064 BTRFS_FILE_EXTENT_INLINE)
1066 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1067 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1070 if (!in_range(bytenr, rc->block_group->start,
1071 rc->block_group->length))
1075 * if we are modifying block in fs tree, wait for readpage
1076 * to complete and drop the extent cache
1078 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1080 inode = find_next_inode(root, key.objectid);
1082 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1083 btrfs_add_delayed_iput(inode);
1084 inode = find_next_inode(root, key.objectid);
1086 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1088 btrfs_file_extent_num_bytes(leaf, fi);
1089 WARN_ON(!IS_ALIGNED(key.offset,
1090 fs_info->sectorsize));
1091 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1093 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1098 btrfs_drop_extent_cache(BTRFS_I(inode),
1099 key.offset, end, 1);
1100 unlock_extent(&BTRFS_I(inode)->io_tree,
1105 ret = get_new_location(rc->data_inode, &new_bytenr,
1109 * Don't have to abort since we've not changed anything
1110 * in the file extent yet.
1115 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1118 key.offset -= btrfs_file_extent_offset(leaf, fi);
1119 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1121 ref.real_root = root->root_key.objectid;
1122 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1123 key.objectid, key.offset);
1124 ret = btrfs_inc_extent_ref(trans, &ref);
1126 btrfs_abort_transaction(trans, ret);
1130 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1132 ref.real_root = root->root_key.objectid;
1133 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1134 key.objectid, key.offset);
1135 ret = btrfs_free_extent(trans, &ref);
1137 btrfs_abort_transaction(trans, ret);
1142 btrfs_mark_buffer_dirty(leaf);
1144 btrfs_add_delayed_iput(inode);
1148 static noinline_for_stack
1149 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1150 struct btrfs_path *path, int level)
1152 struct btrfs_disk_key key1;
1153 struct btrfs_disk_key key2;
1154 btrfs_node_key(eb, &key1, slot);
1155 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1156 return memcmp(&key1, &key2, sizeof(key1));
1160 * try to replace tree blocks in fs tree with the new blocks
1161 * in reloc tree. tree blocks haven't been modified since the
1162 * reloc tree was create can be replaced.
1164 * if a block was replaced, level of the block + 1 is returned.
1165 * if no block got replaced, 0 is returned. if there are other
1166 * errors, a negative error number is returned.
1168 static noinline_for_stack
1169 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1170 struct btrfs_root *dest, struct btrfs_root *src,
1171 struct btrfs_path *path, struct btrfs_key *next_key,
1172 int lowest_level, int max_level)
1174 struct btrfs_fs_info *fs_info = dest->fs_info;
1175 struct extent_buffer *eb;
1176 struct extent_buffer *parent;
1177 struct btrfs_ref ref = { 0 };
1178 struct btrfs_key key;
1190 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1191 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1193 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1195 slot = path->slots[lowest_level];
1196 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1198 eb = btrfs_lock_root_node(dest);
1199 btrfs_set_lock_blocking_write(eb);
1200 level = btrfs_header_level(eb);
1202 if (level < lowest_level) {
1203 btrfs_tree_unlock(eb);
1204 free_extent_buffer(eb);
1209 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1213 btrfs_set_lock_blocking_write(eb);
1216 next_key->objectid = (u64)-1;
1217 next_key->type = (u8)-1;
1218 next_key->offset = (u64)-1;
1223 struct btrfs_key first_key;
1225 level = btrfs_header_level(parent);
1226 BUG_ON(level < lowest_level);
1228 ret = btrfs_bin_search(parent, &key, &slot);
1231 if (ret && slot > 0)
1234 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1235 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1237 old_bytenr = btrfs_node_blockptr(parent, slot);
1238 blocksize = fs_info->nodesize;
1239 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1240 btrfs_node_key_to_cpu(parent, &first_key, slot);
1242 if (level <= max_level) {
1243 eb = path->nodes[level];
1244 new_bytenr = btrfs_node_blockptr(eb,
1245 path->slots[level]);
1246 new_ptr_gen = btrfs_node_ptr_generation(eb,
1247 path->slots[level]);
1253 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1258 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1259 memcmp_node_keys(parent, slot, path, level)) {
1260 if (level <= lowest_level) {
1265 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1266 level - 1, &first_key);
1270 } else if (!extent_buffer_uptodate(eb)) {
1272 free_extent_buffer(eb);
1275 btrfs_tree_lock(eb);
1277 ret = btrfs_cow_block(trans, dest, eb, parent,
1282 btrfs_set_lock_blocking_write(eb);
1284 btrfs_tree_unlock(parent);
1285 free_extent_buffer(parent);
1292 btrfs_tree_unlock(parent);
1293 free_extent_buffer(parent);
1298 btrfs_node_key_to_cpu(path->nodes[level], &key,
1299 path->slots[level]);
1300 btrfs_release_path(path);
1302 path->lowest_level = level;
1303 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1304 path->lowest_level = 0;
1308 * Info qgroup to trace both subtrees.
1310 * We must trace both trees.
1311 * 1) Tree reloc subtree
1312 * If not traced, we will leak data numbers
1314 * If not traced, we will double count old data
1316 * We don't scan the subtree right now, but only record
1317 * the swapped tree blocks.
1318 * The real subtree rescan is delayed until we have new
1319 * CoW on the subtree root node before transaction commit.
1321 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1322 rc->block_group, parent, slot,
1323 path->nodes[level], path->slots[level],
1328 * swap blocks in fs tree and reloc tree.
1330 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1331 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1332 btrfs_mark_buffer_dirty(parent);
1334 btrfs_set_node_blockptr(path->nodes[level],
1335 path->slots[level], old_bytenr);
1336 btrfs_set_node_ptr_generation(path->nodes[level],
1337 path->slots[level], old_ptr_gen);
1338 btrfs_mark_buffer_dirty(path->nodes[level]);
1340 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1341 blocksize, path->nodes[level]->start);
1342 ref.skip_qgroup = true;
1343 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1344 ret = btrfs_inc_extent_ref(trans, &ref);
1346 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1348 ref.skip_qgroup = true;
1349 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1350 ret = btrfs_inc_extent_ref(trans, &ref);
1353 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1354 blocksize, path->nodes[level]->start);
1355 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1356 ref.skip_qgroup = true;
1357 ret = btrfs_free_extent(trans, &ref);
1360 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1362 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1363 ref.skip_qgroup = true;
1364 ret = btrfs_free_extent(trans, &ref);
1367 btrfs_unlock_up_safe(path, 0);
1372 btrfs_tree_unlock(parent);
1373 free_extent_buffer(parent);
1378 * helper to find next relocated block in reloc tree
1380 static noinline_for_stack
1381 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1384 struct extent_buffer *eb;
1389 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1391 for (i = 0; i < *level; i++) {
1392 free_extent_buffer(path->nodes[i]);
1393 path->nodes[i] = NULL;
1396 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1397 eb = path->nodes[i];
1398 nritems = btrfs_header_nritems(eb);
1399 while (path->slots[i] + 1 < nritems) {
1401 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1408 free_extent_buffer(path->nodes[i]);
1409 path->nodes[i] = NULL;
1415 * walk down reloc tree to find relocated block of lowest level
1417 static noinline_for_stack
1418 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1421 struct btrfs_fs_info *fs_info = root->fs_info;
1422 struct extent_buffer *eb = NULL;
1429 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1431 for (i = *level; i > 0; i--) {
1432 struct btrfs_key first_key;
1434 eb = path->nodes[i];
1435 nritems = btrfs_header_nritems(eb);
1436 while (path->slots[i] < nritems) {
1437 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1438 if (ptr_gen > last_snapshot)
1442 if (path->slots[i] >= nritems) {
1453 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1454 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
1455 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
1459 } else if (!extent_buffer_uptodate(eb)) {
1460 free_extent_buffer(eb);
1463 BUG_ON(btrfs_header_level(eb) != i - 1);
1464 path->nodes[i - 1] = eb;
1465 path->slots[i - 1] = 0;
1471 * invalidate extent cache for file extents whose key in range of
1472 * [min_key, max_key)
1474 static int invalidate_extent_cache(struct btrfs_root *root,
1475 struct btrfs_key *min_key,
1476 struct btrfs_key *max_key)
1478 struct btrfs_fs_info *fs_info = root->fs_info;
1479 struct inode *inode = NULL;
1484 objectid = min_key->objectid;
1489 if (objectid > max_key->objectid)
1492 inode = find_next_inode(root, objectid);
1495 ino = btrfs_ino(BTRFS_I(inode));
1497 if (ino > max_key->objectid) {
1503 if (!S_ISREG(inode->i_mode))
1506 if (unlikely(min_key->objectid == ino)) {
1507 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1509 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1512 start = min_key->offset;
1513 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1519 if (unlikely(max_key->objectid == ino)) {
1520 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1522 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1525 if (max_key->offset == 0)
1527 end = max_key->offset;
1528 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1535 /* the lock_extent waits for readpage to complete */
1536 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1537 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1538 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1543 static int find_next_key(struct btrfs_path *path, int level,
1544 struct btrfs_key *key)
1547 while (level < BTRFS_MAX_LEVEL) {
1548 if (!path->nodes[level])
1550 if (path->slots[level] + 1 <
1551 btrfs_header_nritems(path->nodes[level])) {
1552 btrfs_node_key_to_cpu(path->nodes[level], key,
1553 path->slots[level] + 1);
1562 * Insert current subvolume into reloc_control::dirty_subvol_roots
1564 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1565 struct reloc_control *rc,
1566 struct btrfs_root *root)
1568 struct btrfs_root *reloc_root = root->reloc_root;
1569 struct btrfs_root_item *reloc_root_item;
1571 /* @root must be a subvolume tree root with a valid reloc tree */
1572 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1575 reloc_root_item = &reloc_root->root_item;
1576 memset(&reloc_root_item->drop_progress, 0,
1577 sizeof(reloc_root_item->drop_progress));
1578 reloc_root_item->drop_level = 0;
1579 btrfs_set_root_refs(reloc_root_item, 0);
1580 btrfs_update_reloc_root(trans, root);
1582 if (list_empty(&root->reloc_dirty_list)) {
1583 btrfs_grab_root(root);
1584 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1588 static int clean_dirty_subvols(struct reloc_control *rc)
1590 struct btrfs_root *root;
1591 struct btrfs_root *next;
1595 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1597 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1598 /* Merged subvolume, cleanup its reloc root */
1599 struct btrfs_root *reloc_root = root->reloc_root;
1601 list_del_init(&root->reloc_dirty_list);
1602 root->reloc_root = NULL;
1604 * Need barrier to ensure clear_bit() only happens after
1605 * root->reloc_root = NULL. Pairs with have_reloc_root.
1608 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1611 * btrfs_drop_snapshot drops our ref we hold for
1612 * ->reloc_root. If it fails however we must
1613 * drop the ref ourselves.
1615 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1617 btrfs_put_root(reloc_root);
1622 btrfs_put_root(root);
1624 /* Orphan reloc tree, just clean it up */
1625 ret2 = btrfs_drop_snapshot(root, 0, 1);
1627 btrfs_put_root(root);
1637 * merge the relocated tree blocks in reloc tree with corresponding
1640 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1641 struct btrfs_root *root)
1643 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1644 struct btrfs_key key;
1645 struct btrfs_key next_key;
1646 struct btrfs_trans_handle *trans = NULL;
1647 struct btrfs_root *reloc_root;
1648 struct btrfs_root_item *root_item;
1649 struct btrfs_path *path;
1650 struct extent_buffer *leaf;
1659 path = btrfs_alloc_path();
1662 path->reada = READA_FORWARD;
1664 reloc_root = root->reloc_root;
1665 root_item = &reloc_root->root_item;
1667 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1668 level = btrfs_root_level(root_item);
1669 atomic_inc(&reloc_root->node->refs);
1670 path->nodes[level] = reloc_root->node;
1671 path->slots[level] = 0;
1673 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1675 level = root_item->drop_level;
1677 path->lowest_level = level;
1678 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1679 path->lowest_level = 0;
1681 btrfs_free_path(path);
1685 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1686 path->slots[level]);
1687 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1689 btrfs_unlock_up_safe(path, 0);
1693 * In merge_reloc_root(), we modify the upper level pointer to swap the
1694 * tree blocks between reloc tree and subvolume tree. Thus for tree
1695 * block COW, we COW at most from level 1 to root level for each tree.
1697 * Thus the needed metadata size is at most root_level * nodesize,
1698 * and * 2 since we have two trees to COW.
1700 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1701 min_reserved = fs_info->nodesize * reserve_level * 2;
1702 memset(&next_key, 0, sizeof(next_key));
1705 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1706 BTRFS_RESERVE_FLUSH_LIMIT);
1711 trans = btrfs_start_transaction(root, 0);
1712 if (IS_ERR(trans)) {
1713 err = PTR_ERR(trans);
1719 * At this point we no longer have a reloc_control, so we can't
1720 * depend on btrfs_init_reloc_root to update our last_trans.
1722 * But that's ok, we started the trans handle on our
1723 * corresponding fs_root, which means it's been added to the
1724 * dirty list. At commit time we'll still call
1725 * btrfs_update_reloc_root() and update our root item
1728 reloc_root->last_trans = trans->transid;
1729 trans->block_rsv = rc->block_rsv;
1734 ret = walk_down_reloc_tree(reloc_root, path, &level);
1742 if (!find_next_key(path, level, &key) &&
1743 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1746 ret = replace_path(trans, rc, root, reloc_root, path,
1747 &next_key, level, max_level);
1756 btrfs_node_key_to_cpu(path->nodes[level], &key,
1757 path->slots[level]);
1761 ret = walk_up_reloc_tree(reloc_root, path, &level);
1767 * save the merging progress in the drop_progress.
1768 * this is OK since root refs == 1 in this case.
1770 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1771 path->slots[level]);
1772 root_item->drop_level = level;
1774 btrfs_end_transaction_throttle(trans);
1777 btrfs_btree_balance_dirty(fs_info);
1779 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1780 invalidate_extent_cache(root, &key, &next_key);
1784 * handle the case only one block in the fs tree need to be
1785 * relocated and the block is tree root.
1787 leaf = btrfs_lock_root_node(root);
1788 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1790 btrfs_tree_unlock(leaf);
1791 free_extent_buffer(leaf);
1795 btrfs_free_path(path);
1798 insert_dirty_subvol(trans, rc, root);
1801 btrfs_end_transaction_throttle(trans);
1803 btrfs_btree_balance_dirty(fs_info);
1805 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1806 invalidate_extent_cache(root, &key, &next_key);
1811 static noinline_for_stack
1812 int prepare_to_merge(struct reloc_control *rc, int err)
1814 struct btrfs_root *root = rc->extent_root;
1815 struct btrfs_fs_info *fs_info = root->fs_info;
1816 struct btrfs_root *reloc_root;
1817 struct btrfs_trans_handle *trans;
1818 LIST_HEAD(reloc_roots);
1822 mutex_lock(&fs_info->reloc_mutex);
1823 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1824 rc->merging_rsv_size += rc->nodes_relocated * 2;
1825 mutex_unlock(&fs_info->reloc_mutex);
1829 num_bytes = rc->merging_rsv_size;
1830 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1831 BTRFS_RESERVE_FLUSH_ALL);
1836 trans = btrfs_join_transaction(rc->extent_root);
1837 if (IS_ERR(trans)) {
1839 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1841 return PTR_ERR(trans);
1845 if (num_bytes != rc->merging_rsv_size) {
1846 btrfs_end_transaction(trans);
1847 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1853 rc->merge_reloc_tree = 1;
1855 while (!list_empty(&rc->reloc_roots)) {
1856 reloc_root = list_entry(rc->reloc_roots.next,
1857 struct btrfs_root, root_list);
1858 list_del_init(&reloc_root->root_list);
1860 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1862 BUG_ON(IS_ERR(root));
1863 BUG_ON(root->reloc_root != reloc_root);
1866 * set reference count to 1, so btrfs_recover_relocation
1867 * knows it should resumes merging
1870 btrfs_set_root_refs(&reloc_root->root_item, 1);
1871 btrfs_update_reloc_root(trans, root);
1873 list_add(&reloc_root->root_list, &reloc_roots);
1874 btrfs_put_root(root);
1877 list_splice(&reloc_roots, &rc->reloc_roots);
1880 btrfs_commit_transaction(trans);
1882 btrfs_end_transaction(trans);
1886 static noinline_for_stack
1887 void free_reloc_roots(struct list_head *list)
1889 struct btrfs_root *reloc_root, *tmp;
1891 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1892 __del_reloc_root(reloc_root);
1895 static noinline_for_stack
1896 void merge_reloc_roots(struct reloc_control *rc)
1898 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1899 struct btrfs_root *root;
1900 struct btrfs_root *reloc_root;
1901 LIST_HEAD(reloc_roots);
1905 root = rc->extent_root;
1908 * this serializes us with btrfs_record_root_in_transaction,
1909 * we have to make sure nobody is in the middle of
1910 * adding their roots to the list while we are
1913 mutex_lock(&fs_info->reloc_mutex);
1914 list_splice_init(&rc->reloc_roots, &reloc_roots);
1915 mutex_unlock(&fs_info->reloc_mutex);
1917 while (!list_empty(&reloc_roots)) {
1919 reloc_root = list_entry(reloc_roots.next,
1920 struct btrfs_root, root_list);
1922 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1924 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1925 BUG_ON(IS_ERR(root));
1926 BUG_ON(root->reloc_root != reloc_root);
1927 ret = merge_reloc_root(rc, root);
1928 btrfs_put_root(root);
1930 if (list_empty(&reloc_root->root_list))
1931 list_add_tail(&reloc_root->root_list,
1936 if (!IS_ERR(root)) {
1937 if (root->reloc_root == reloc_root) {
1938 root->reloc_root = NULL;
1939 btrfs_put_root(reloc_root);
1941 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1943 btrfs_put_root(root);
1946 list_del_init(&reloc_root->root_list);
1947 /* Don't forget to queue this reloc root for cleanup */
1948 list_add_tail(&reloc_root->reloc_dirty_list,
1949 &rc->dirty_subvol_roots);
1959 btrfs_handle_fs_error(fs_info, ret, NULL);
1960 free_reloc_roots(&reloc_roots);
1962 /* new reloc root may be added */
1963 mutex_lock(&fs_info->reloc_mutex);
1964 list_splice_init(&rc->reloc_roots, &reloc_roots);
1965 mutex_unlock(&fs_info->reloc_mutex);
1966 free_reloc_roots(&reloc_roots);
1972 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1974 * here, but it's wrong. If we fail to start the transaction in
1975 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1976 * have actually been removed from the reloc_root_tree rb tree. This is
1977 * fine because we're bailing here, and we hold a reference on the root
1978 * for the list that holds it, so these roots will be cleaned up when we
1979 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1980 * will be cleaned up on unmount.
1982 * The remaining nodes will be cleaned up by free_reloc_control.
1986 static void free_block_list(struct rb_root *blocks)
1988 struct tree_block *block;
1989 struct rb_node *rb_node;
1990 while ((rb_node = rb_first(blocks))) {
1991 block = rb_entry(rb_node, struct tree_block, rb_node);
1992 rb_erase(rb_node, blocks);
1997 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1998 struct btrfs_root *reloc_root)
2000 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2001 struct btrfs_root *root;
2004 if (reloc_root->last_trans == trans->transid)
2007 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2008 BUG_ON(IS_ERR(root));
2009 BUG_ON(root->reloc_root != reloc_root);
2010 ret = btrfs_record_root_in_trans(trans, root);
2011 btrfs_put_root(root);
2016 static noinline_for_stack
2017 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2018 struct reloc_control *rc,
2019 struct btrfs_backref_node *node,
2020 struct btrfs_backref_edge *edges[])
2022 struct btrfs_backref_node *next;
2023 struct btrfs_root *root;
2029 next = walk_up_backref(next, edges, &index);
2032 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2034 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2035 record_reloc_root_in_trans(trans, root);
2039 btrfs_record_root_in_trans(trans, root);
2040 root = root->reloc_root;
2042 if (next->new_bytenr != root->node->start) {
2043 BUG_ON(next->new_bytenr);
2044 BUG_ON(!list_empty(&next->list));
2045 next->new_bytenr = root->node->start;
2046 btrfs_put_root(next->root);
2047 next->root = btrfs_grab_root(root);
2049 list_add_tail(&next->list,
2050 &rc->backref_cache.changed);
2051 mark_block_processed(rc, next);
2057 next = walk_down_backref(edges, &index);
2058 if (!next || next->level <= node->level)
2065 /* setup backref node path for btrfs_reloc_cow_block */
2067 rc->backref_cache.path[next->level] = next;
2070 next = edges[index]->node[UPPER];
2076 * Select a tree root for relocation.
2078 * Return NULL if the block is not shareable. We should use do_relocation() in
2081 * Return a tree root pointer if the block is shareable.
2082 * Return -ENOENT if the block is root of reloc tree.
2084 static noinline_for_stack
2085 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2087 struct btrfs_backref_node *next;
2088 struct btrfs_root *root;
2089 struct btrfs_root *fs_root = NULL;
2090 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2096 next = walk_up_backref(next, edges, &index);
2100 /* No other choice for non-shareable tree */
2101 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2104 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2110 next = walk_down_backref(edges, &index);
2111 if (!next || next->level <= node->level)
2116 return ERR_PTR(-ENOENT);
2120 static noinline_for_stack
2121 u64 calcu_metadata_size(struct reloc_control *rc,
2122 struct btrfs_backref_node *node, int reserve)
2124 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2125 struct btrfs_backref_node *next = node;
2126 struct btrfs_backref_edge *edge;
2127 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2131 BUG_ON(reserve && node->processed);
2136 if (next->processed && (reserve || next != node))
2139 num_bytes += fs_info->nodesize;
2141 if (list_empty(&next->upper))
2144 edge = list_entry(next->upper.next,
2145 struct btrfs_backref_edge, list[LOWER]);
2146 edges[index++] = edge;
2147 next = edge->node[UPPER];
2149 next = walk_down_backref(edges, &index);
2154 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2155 struct reloc_control *rc,
2156 struct btrfs_backref_node *node)
2158 struct btrfs_root *root = rc->extent_root;
2159 struct btrfs_fs_info *fs_info = root->fs_info;
2164 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2166 trans->block_rsv = rc->block_rsv;
2167 rc->reserved_bytes += num_bytes;
2170 * We are under a transaction here so we can only do limited flushing.
2171 * If we get an enospc just kick back -EAGAIN so we know to drop the
2172 * transaction and try to refill when we can flush all the things.
2174 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2175 BTRFS_RESERVE_FLUSH_LIMIT);
2177 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2178 while (tmp <= rc->reserved_bytes)
2181 * only one thread can access block_rsv at this point,
2182 * so we don't need hold lock to protect block_rsv.
2183 * we expand more reservation size here to allow enough
2184 * space for relocation and we will return earlier in
2187 rc->block_rsv->size = tmp + fs_info->nodesize *
2188 RELOCATION_RESERVED_NODES;
2196 * relocate a block tree, and then update pointers in upper level
2197 * blocks that reference the block to point to the new location.
2199 * if called by link_to_upper, the block has already been relocated.
2200 * in that case this function just updates pointers.
2202 static int do_relocation(struct btrfs_trans_handle *trans,
2203 struct reloc_control *rc,
2204 struct btrfs_backref_node *node,
2205 struct btrfs_key *key,
2206 struct btrfs_path *path, int lowest)
2208 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2209 struct btrfs_backref_node *upper;
2210 struct btrfs_backref_edge *edge;
2211 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2212 struct btrfs_root *root;
2213 struct extent_buffer *eb;
2221 BUG_ON(lowest && node->eb);
2223 path->lowest_level = node->level + 1;
2224 rc->backref_cache.path[node->level] = node;
2225 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2226 struct btrfs_key first_key;
2227 struct btrfs_ref ref = { 0 };
2231 upper = edge->node[UPPER];
2232 root = select_reloc_root(trans, rc, upper, edges);
2235 if (upper->eb && !upper->locked) {
2237 ret = btrfs_bin_search(upper->eb, key, &slot);
2243 bytenr = btrfs_node_blockptr(upper->eb, slot);
2244 if (node->eb->start == bytenr)
2247 btrfs_backref_drop_node_buffer(upper);
2251 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2258 btrfs_release_path(path);
2263 upper->eb = path->nodes[upper->level];
2264 path->nodes[upper->level] = NULL;
2266 BUG_ON(upper->eb != path->nodes[upper->level]);
2270 path->locks[upper->level] = 0;
2272 slot = path->slots[upper->level];
2273 btrfs_release_path(path);
2275 ret = btrfs_bin_search(upper->eb, key, &slot);
2283 bytenr = btrfs_node_blockptr(upper->eb, slot);
2285 if (bytenr != node->bytenr) {
2286 btrfs_err(root->fs_info,
2287 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2288 bytenr, node->bytenr, slot,
2294 if (node->eb->start == bytenr)
2298 blocksize = root->fs_info->nodesize;
2299 generation = btrfs_node_ptr_generation(upper->eb, slot);
2300 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2301 eb = read_tree_block(fs_info, bytenr, generation,
2302 upper->level - 1, &first_key);
2306 } else if (!extent_buffer_uptodate(eb)) {
2307 free_extent_buffer(eb);
2311 btrfs_tree_lock(eb);
2312 btrfs_set_lock_blocking_write(eb);
2315 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2316 slot, &eb, BTRFS_NESTING_COW);
2317 btrfs_tree_unlock(eb);
2318 free_extent_buffer(eb);
2323 BUG_ON(node->eb != eb);
2325 btrfs_set_node_blockptr(upper->eb, slot,
2327 btrfs_set_node_ptr_generation(upper->eb, slot,
2329 btrfs_mark_buffer_dirty(upper->eb);
2331 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2332 node->eb->start, blocksize,
2334 ref.real_root = root->root_key.objectid;
2335 btrfs_init_tree_ref(&ref, node->level,
2336 btrfs_header_owner(upper->eb));
2337 ret = btrfs_inc_extent_ref(trans, &ref);
2340 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2344 if (!upper->pending)
2345 btrfs_backref_drop_node_buffer(upper);
2347 btrfs_backref_unlock_node_buffer(upper);
2352 if (!err && node->pending) {
2353 btrfs_backref_drop_node_buffer(node);
2354 list_move_tail(&node->list, &rc->backref_cache.changed);
2358 path->lowest_level = 0;
2359 BUG_ON(err == -ENOSPC);
2363 static int link_to_upper(struct btrfs_trans_handle *trans,
2364 struct reloc_control *rc,
2365 struct btrfs_backref_node *node,
2366 struct btrfs_path *path)
2368 struct btrfs_key key;
2370 btrfs_node_key_to_cpu(node->eb, &key, 0);
2371 return do_relocation(trans, rc, node, &key, path, 0);
2374 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2375 struct reloc_control *rc,
2376 struct btrfs_path *path, int err)
2379 struct btrfs_backref_cache *cache = &rc->backref_cache;
2380 struct btrfs_backref_node *node;
2384 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2385 while (!list_empty(&cache->pending[level])) {
2386 node = list_entry(cache->pending[level].next,
2387 struct btrfs_backref_node, list);
2388 list_move_tail(&node->list, &list);
2389 BUG_ON(!node->pending);
2392 ret = link_to_upper(trans, rc, node, path);
2397 list_splice_init(&list, &cache->pending[level]);
2403 * mark a block and all blocks directly/indirectly reference the block
2406 static void update_processed_blocks(struct reloc_control *rc,
2407 struct btrfs_backref_node *node)
2409 struct btrfs_backref_node *next = node;
2410 struct btrfs_backref_edge *edge;
2411 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2417 if (next->processed)
2420 mark_block_processed(rc, next);
2422 if (list_empty(&next->upper))
2425 edge = list_entry(next->upper.next,
2426 struct btrfs_backref_edge, list[LOWER]);
2427 edges[index++] = edge;
2428 next = edge->node[UPPER];
2430 next = walk_down_backref(edges, &index);
2434 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2436 u32 blocksize = rc->extent_root->fs_info->nodesize;
2438 if (test_range_bit(&rc->processed_blocks, bytenr,
2439 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2444 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2445 struct tree_block *block)
2447 struct extent_buffer *eb;
2449 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2450 block->level, NULL);
2453 } else if (!extent_buffer_uptodate(eb)) {
2454 free_extent_buffer(eb);
2457 if (block->level == 0)
2458 btrfs_item_key_to_cpu(eb, &block->key, 0);
2460 btrfs_node_key_to_cpu(eb, &block->key, 0);
2461 free_extent_buffer(eb);
2462 block->key_ready = 1;
2467 * helper function to relocate a tree block
2469 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2470 struct reloc_control *rc,
2471 struct btrfs_backref_node *node,
2472 struct btrfs_key *key,
2473 struct btrfs_path *path)
2475 struct btrfs_root *root;
2482 * If we fail here we want to drop our backref_node because we are going
2483 * to start over and regenerate the tree for it.
2485 ret = reserve_metadata_space(trans, rc, node);
2489 BUG_ON(node->processed);
2490 root = select_one_root(node);
2491 if (root == ERR_PTR(-ENOENT)) {
2492 update_processed_blocks(rc, node);
2497 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2498 BUG_ON(node->new_bytenr);
2499 BUG_ON(!list_empty(&node->list));
2500 btrfs_record_root_in_trans(trans, root);
2501 root = root->reloc_root;
2502 node->new_bytenr = root->node->start;
2503 btrfs_put_root(node->root);
2504 node->root = btrfs_grab_root(root);
2506 list_add_tail(&node->list, &rc->backref_cache.changed);
2508 path->lowest_level = node->level;
2509 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2510 btrfs_release_path(path);
2515 update_processed_blocks(rc, node);
2517 ret = do_relocation(trans, rc, node, key, path, 1);
2520 if (ret || node->level == 0 || node->cowonly)
2521 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2526 * relocate a list of blocks
2528 static noinline_for_stack
2529 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2530 struct reloc_control *rc, struct rb_root *blocks)
2532 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2533 struct btrfs_backref_node *node;
2534 struct btrfs_path *path;
2535 struct tree_block *block;
2536 struct tree_block *next;
2540 path = btrfs_alloc_path();
2543 goto out_free_blocks;
2546 /* Kick in readahead for tree blocks with missing keys */
2547 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2548 if (!block->key_ready)
2549 readahead_tree_block(fs_info, block->bytenr);
2552 /* Get first keys */
2553 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2554 if (!block->key_ready) {
2555 err = get_tree_block_key(fs_info, block);
2561 /* Do tree relocation */
2562 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2563 node = build_backref_tree(rc, &block->key,
2564 block->level, block->bytenr);
2566 err = PTR_ERR(node);
2570 ret = relocate_tree_block(trans, rc, node, &block->key,
2578 err = finish_pending_nodes(trans, rc, path, err);
2581 btrfs_free_path(path);
2583 free_block_list(blocks);
2587 static noinline_for_stack int prealloc_file_extent_cluster(
2588 struct btrfs_inode *inode,
2589 struct file_extent_cluster *cluster)
2594 u64 offset = inode->index_cnt;
2598 u64 prealloc_start = cluster->start - offset;
2599 u64 prealloc_end = cluster->end - offset;
2600 u64 cur_offset = prealloc_start;
2602 BUG_ON(cluster->start != cluster->boundary[0]);
2603 ret = btrfs_alloc_data_chunk_ondemand(inode,
2604 prealloc_end + 1 - prealloc_start);
2608 inode_lock(&inode->vfs_inode);
2609 for (nr = 0; nr < cluster->nr; nr++) {
2610 start = cluster->boundary[nr] - offset;
2611 if (nr + 1 < cluster->nr)
2612 end = cluster->boundary[nr + 1] - 1 - offset;
2614 end = cluster->end - offset;
2616 lock_extent(&inode->io_tree, start, end);
2617 num_bytes = end + 1 - start;
2618 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2619 num_bytes, num_bytes,
2620 end + 1, &alloc_hint);
2621 cur_offset = end + 1;
2622 unlock_extent(&inode->io_tree, start, end);
2626 inode_unlock(&inode->vfs_inode);
2628 if (cur_offset < prealloc_end)
2629 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2630 prealloc_end + 1 - cur_offset);
2634 static noinline_for_stack
2635 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2638 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2639 struct extent_map *em;
2642 em = alloc_extent_map();
2647 em->len = end + 1 - start;
2648 em->block_len = em->len;
2649 em->block_start = block_start;
2650 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2652 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2654 write_lock(&em_tree->lock);
2655 ret = add_extent_mapping(em_tree, em, 0);
2656 write_unlock(&em_tree->lock);
2657 if (ret != -EEXIST) {
2658 free_extent_map(em);
2661 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2663 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2668 * Allow error injection to test balance cancellation
2670 int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2672 return atomic_read(&fs_info->balance_cancel_req) ||
2673 fatal_signal_pending(current);
2675 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2677 static int relocate_file_extent_cluster(struct inode *inode,
2678 struct file_extent_cluster *cluster)
2680 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2683 u64 offset = BTRFS_I(inode)->index_cnt;
2684 unsigned long index;
2685 unsigned long last_index;
2687 struct file_ra_state *ra;
2688 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2695 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2699 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2703 file_ra_state_init(ra, inode->i_mapping);
2705 ret = setup_extent_mapping(inode, cluster->start - offset,
2706 cluster->end - offset, cluster->start);
2710 index = (cluster->start - offset) >> PAGE_SHIFT;
2711 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2712 while (index <= last_index) {
2713 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2718 page = find_lock_page(inode->i_mapping, index);
2720 page_cache_sync_readahead(inode->i_mapping,
2722 last_index + 1 - index);
2723 page = find_or_create_page(inode->i_mapping, index,
2726 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2728 btrfs_delalloc_release_extents(BTRFS_I(inode),
2735 if (PageReadahead(page)) {
2736 page_cache_async_readahead(inode->i_mapping,
2737 ra, NULL, page, index,
2738 last_index + 1 - index);
2741 if (!PageUptodate(page)) {
2742 btrfs_readpage(NULL, page);
2744 if (!PageUptodate(page)) {
2747 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2749 btrfs_delalloc_release_extents(BTRFS_I(inode),
2756 page_start = page_offset(page);
2757 page_end = page_start + PAGE_SIZE - 1;
2759 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2761 set_page_extent_mapped(page);
2763 if (nr < cluster->nr &&
2764 page_start + offset == cluster->boundary[nr]) {
2765 set_extent_bits(&BTRFS_I(inode)->io_tree,
2766 page_start, page_end,
2771 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2776 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2778 btrfs_delalloc_release_extents(BTRFS_I(inode),
2781 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2782 page_start, page_end,
2783 EXTENT_LOCKED | EXTENT_BOUNDARY);
2787 set_page_dirty(page);
2789 unlock_extent(&BTRFS_I(inode)->io_tree,
2790 page_start, page_end);
2795 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2796 balance_dirty_pages_ratelimited(inode->i_mapping);
2797 btrfs_throttle(fs_info);
2798 if (btrfs_should_cancel_balance(fs_info)) {
2803 WARN_ON(nr != cluster->nr);
2809 static noinline_for_stack
2810 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2811 struct file_extent_cluster *cluster)
2815 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2816 ret = relocate_file_extent_cluster(inode, cluster);
2823 cluster->start = extent_key->objectid;
2825 BUG_ON(cluster->nr >= MAX_EXTENTS);
2826 cluster->end = extent_key->objectid + extent_key->offset - 1;
2827 cluster->boundary[cluster->nr] = extent_key->objectid;
2830 if (cluster->nr >= MAX_EXTENTS) {
2831 ret = relocate_file_extent_cluster(inode, cluster);
2840 * helper to add a tree block to the list.
2841 * the major work is getting the generation and level of the block
2843 static int add_tree_block(struct reloc_control *rc,
2844 struct btrfs_key *extent_key,
2845 struct btrfs_path *path,
2846 struct rb_root *blocks)
2848 struct extent_buffer *eb;
2849 struct btrfs_extent_item *ei;
2850 struct btrfs_tree_block_info *bi;
2851 struct tree_block *block;
2852 struct rb_node *rb_node;
2857 eb = path->nodes[0];
2858 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2860 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2861 item_size >= sizeof(*ei) + sizeof(*bi)) {
2862 ei = btrfs_item_ptr(eb, path->slots[0],
2863 struct btrfs_extent_item);
2864 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2865 bi = (struct btrfs_tree_block_info *)(ei + 1);
2866 level = btrfs_tree_block_level(eb, bi);
2868 level = (int)extent_key->offset;
2870 generation = btrfs_extent_generation(eb, ei);
2871 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2872 btrfs_print_v0_err(eb->fs_info);
2873 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2879 btrfs_release_path(path);
2881 BUG_ON(level == -1);
2883 block = kmalloc(sizeof(*block), GFP_NOFS);
2887 block->bytenr = extent_key->objectid;
2888 block->key.objectid = rc->extent_root->fs_info->nodesize;
2889 block->key.offset = generation;
2890 block->level = level;
2891 block->key_ready = 0;
2893 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2895 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2902 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2904 static int __add_tree_block(struct reloc_control *rc,
2905 u64 bytenr, u32 blocksize,
2906 struct rb_root *blocks)
2908 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2909 struct btrfs_path *path;
2910 struct btrfs_key key;
2912 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2914 if (tree_block_processed(bytenr, rc))
2917 if (rb_simple_search(blocks, bytenr))
2920 path = btrfs_alloc_path();
2924 key.objectid = bytenr;
2926 key.type = BTRFS_METADATA_ITEM_KEY;
2927 key.offset = (u64)-1;
2929 key.type = BTRFS_EXTENT_ITEM_KEY;
2930 key.offset = blocksize;
2933 path->search_commit_root = 1;
2934 path->skip_locking = 1;
2935 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2939 if (ret > 0 && skinny) {
2940 if (path->slots[0]) {
2942 btrfs_item_key_to_cpu(path->nodes[0], &key,
2944 if (key.objectid == bytenr &&
2945 (key.type == BTRFS_METADATA_ITEM_KEY ||
2946 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2947 key.offset == blocksize)))
2953 btrfs_release_path(path);
2959 btrfs_print_leaf(path->nodes[0]);
2961 "tree block extent item (%llu) is not found in extent tree",
2968 ret = add_tree_block(rc, &key, path, blocks);
2970 btrfs_free_path(path);
2974 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2975 struct btrfs_block_group *block_group,
2976 struct inode *inode,
2979 struct btrfs_root *root = fs_info->tree_root;
2980 struct btrfs_trans_handle *trans;
2986 inode = btrfs_iget(fs_info->sb, ino, root);
2991 ret = btrfs_check_trunc_cache_free_space(fs_info,
2992 &fs_info->global_block_rsv);
2996 trans = btrfs_join_transaction(root);
2997 if (IS_ERR(trans)) {
2998 ret = PTR_ERR(trans);
3002 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3004 btrfs_end_transaction(trans);
3005 btrfs_btree_balance_dirty(fs_info);
3012 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3013 * cache inode, to avoid free space cache data extent blocking data relocation.
3015 static int delete_v1_space_cache(struct extent_buffer *leaf,
3016 struct btrfs_block_group *block_group,
3019 u64 space_cache_ino;
3020 struct btrfs_file_extent_item *ei;
3021 struct btrfs_key key;
3026 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3029 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3030 btrfs_item_key_to_cpu(leaf, &key, i);
3031 if (key.type != BTRFS_EXTENT_DATA_KEY)
3033 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3034 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG &&
3035 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3037 space_cache_ino = key.objectid;
3043 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3049 * helper to find all tree blocks that reference a given data extent
3051 static noinline_for_stack
3052 int add_data_references(struct reloc_control *rc,
3053 struct btrfs_key *extent_key,
3054 struct btrfs_path *path,
3055 struct rb_root *blocks)
3057 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3058 struct ulist *leaves = NULL;
3059 struct ulist_iterator leaf_uiter;
3060 struct ulist_node *ref_node = NULL;
3061 const u32 blocksize = fs_info->nodesize;
3064 btrfs_release_path(path);
3065 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3066 0, &leaves, NULL, true);
3070 ULIST_ITER_INIT(&leaf_uiter);
3071 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3072 struct extent_buffer *eb;
3074 eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL);
3079 ret = delete_v1_space_cache(eb, rc->block_group,
3080 extent_key->objectid);
3081 free_extent_buffer(eb);
3084 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3089 free_block_list(blocks);
3095 * helper to find next unprocessed extent
3097 static noinline_for_stack
3098 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3099 struct btrfs_key *extent_key)
3101 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3102 struct btrfs_key key;
3103 struct extent_buffer *leaf;
3104 u64 start, end, last;
3107 last = rc->block_group->start + rc->block_group->length;
3110 if (rc->search_start >= last) {
3115 key.objectid = rc->search_start;
3116 key.type = BTRFS_EXTENT_ITEM_KEY;
3119 path->search_commit_root = 1;
3120 path->skip_locking = 1;
3121 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3126 leaf = path->nodes[0];
3127 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3128 ret = btrfs_next_leaf(rc->extent_root, path);
3131 leaf = path->nodes[0];
3134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3135 if (key.objectid >= last) {
3140 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3141 key.type != BTRFS_METADATA_ITEM_KEY) {
3146 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3147 key.objectid + key.offset <= rc->search_start) {
3152 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3153 key.objectid + fs_info->nodesize <=
3159 ret = find_first_extent_bit(&rc->processed_blocks,
3160 key.objectid, &start, &end,
3161 EXTENT_DIRTY, NULL);
3163 if (ret == 0 && start <= key.objectid) {
3164 btrfs_release_path(path);
3165 rc->search_start = end + 1;
3167 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3168 rc->search_start = key.objectid + key.offset;
3170 rc->search_start = key.objectid +
3172 memcpy(extent_key, &key, sizeof(key));
3176 btrfs_release_path(path);
3180 static void set_reloc_control(struct reloc_control *rc)
3182 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3184 mutex_lock(&fs_info->reloc_mutex);
3185 fs_info->reloc_ctl = rc;
3186 mutex_unlock(&fs_info->reloc_mutex);
3189 static void unset_reloc_control(struct reloc_control *rc)
3191 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3193 mutex_lock(&fs_info->reloc_mutex);
3194 fs_info->reloc_ctl = NULL;
3195 mutex_unlock(&fs_info->reloc_mutex);
3198 static int check_extent_flags(u64 flags)
3200 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3201 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3203 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3204 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3206 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3207 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3212 static noinline_for_stack
3213 int prepare_to_relocate(struct reloc_control *rc)
3215 struct btrfs_trans_handle *trans;
3218 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3219 BTRFS_BLOCK_RSV_TEMP);
3223 memset(&rc->cluster, 0, sizeof(rc->cluster));
3224 rc->search_start = rc->block_group->start;
3225 rc->extents_found = 0;
3226 rc->nodes_relocated = 0;
3227 rc->merging_rsv_size = 0;
3228 rc->reserved_bytes = 0;
3229 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3230 RELOCATION_RESERVED_NODES;
3231 ret = btrfs_block_rsv_refill(rc->extent_root,
3232 rc->block_rsv, rc->block_rsv->size,
3233 BTRFS_RESERVE_FLUSH_ALL);
3237 rc->create_reloc_tree = 1;
3238 set_reloc_control(rc);
3240 trans = btrfs_join_transaction(rc->extent_root);
3241 if (IS_ERR(trans)) {
3242 unset_reloc_control(rc);
3244 * extent tree is not a ref_cow tree and has no reloc_root to
3245 * cleanup. And callers are responsible to free the above
3248 return PTR_ERR(trans);
3250 btrfs_commit_transaction(trans);
3254 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3256 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3257 struct rb_root blocks = RB_ROOT;
3258 struct btrfs_key key;
3259 struct btrfs_trans_handle *trans = NULL;
3260 struct btrfs_path *path;
3261 struct btrfs_extent_item *ei;
3268 path = btrfs_alloc_path();
3271 path->reada = READA_FORWARD;
3273 ret = prepare_to_relocate(rc);
3280 rc->reserved_bytes = 0;
3281 ret = btrfs_block_rsv_refill(rc->extent_root,
3282 rc->block_rsv, rc->block_rsv->size,
3283 BTRFS_RESERVE_FLUSH_ALL);
3289 trans = btrfs_start_transaction(rc->extent_root, 0);
3290 if (IS_ERR(trans)) {
3291 err = PTR_ERR(trans);
3296 if (update_backref_cache(trans, &rc->backref_cache)) {
3297 btrfs_end_transaction(trans);
3302 ret = find_next_extent(rc, path, &key);
3308 rc->extents_found++;
3310 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3311 struct btrfs_extent_item);
3312 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3313 if (item_size >= sizeof(*ei)) {
3314 flags = btrfs_extent_flags(path->nodes[0], ei);
3315 ret = check_extent_flags(flags);
3317 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3319 btrfs_print_v0_err(trans->fs_info);
3320 btrfs_abort_transaction(trans, err);
3326 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3327 ret = add_tree_block(rc, &key, path, &blocks);
3328 } else if (rc->stage == UPDATE_DATA_PTRS &&
3329 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3330 ret = add_data_references(rc, &key, path, &blocks);
3332 btrfs_release_path(path);
3340 if (!RB_EMPTY_ROOT(&blocks)) {
3341 ret = relocate_tree_blocks(trans, rc, &blocks);
3343 if (ret != -EAGAIN) {
3347 rc->extents_found--;
3348 rc->search_start = key.objectid;
3352 btrfs_end_transaction_throttle(trans);
3353 btrfs_btree_balance_dirty(fs_info);
3356 if (rc->stage == MOVE_DATA_EXTENTS &&
3357 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3358 rc->found_file_extent = 1;
3359 ret = relocate_data_extent(rc->data_inode,
3360 &key, &rc->cluster);
3366 if (btrfs_should_cancel_balance(fs_info)) {
3371 if (trans && progress && err == -ENOSPC) {
3372 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3380 btrfs_release_path(path);
3381 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3384 btrfs_end_transaction_throttle(trans);
3385 btrfs_btree_balance_dirty(fs_info);
3389 ret = relocate_file_extent_cluster(rc->data_inode,
3395 rc->create_reloc_tree = 0;
3396 set_reloc_control(rc);
3398 btrfs_backref_release_cache(&rc->backref_cache);
3399 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3402 * Even in the case when the relocation is cancelled, we should all go
3403 * through prepare_to_merge() and merge_reloc_roots().
3405 * For error (including cancelled balance), prepare_to_merge() will
3406 * mark all reloc trees orphan, then queue them for cleanup in
3407 * merge_reloc_roots()
3409 err = prepare_to_merge(rc, err);
3411 merge_reloc_roots(rc);
3413 rc->merge_reloc_tree = 0;
3414 unset_reloc_control(rc);
3415 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3417 /* get rid of pinned extents */
3418 trans = btrfs_join_transaction(rc->extent_root);
3419 if (IS_ERR(trans)) {
3420 err = PTR_ERR(trans);
3423 btrfs_commit_transaction(trans);
3425 ret = clean_dirty_subvols(rc);
3426 if (ret < 0 && !err)
3428 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3429 btrfs_free_path(path);
3433 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3434 struct btrfs_root *root, u64 objectid)
3436 struct btrfs_path *path;
3437 struct btrfs_inode_item *item;
3438 struct extent_buffer *leaf;
3441 path = btrfs_alloc_path();
3445 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3449 leaf = path->nodes[0];
3450 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3451 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3452 btrfs_set_inode_generation(leaf, item, 1);
3453 btrfs_set_inode_size(leaf, item, 0);
3454 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3455 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3456 BTRFS_INODE_PREALLOC);
3457 btrfs_mark_buffer_dirty(leaf);
3459 btrfs_free_path(path);
3464 * helper to create inode for data relocation.
3465 * the inode is in data relocation tree and its link count is 0
3467 static noinline_for_stack
3468 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3469 struct btrfs_block_group *group)
3471 struct inode *inode = NULL;
3472 struct btrfs_trans_handle *trans;
3473 struct btrfs_root *root;
3477 root = btrfs_grab_root(fs_info->data_reloc_root);
3478 trans = btrfs_start_transaction(root, 6);
3479 if (IS_ERR(trans)) {
3480 btrfs_put_root(root);
3481 return ERR_CAST(trans);
3484 err = btrfs_find_free_objectid(root, &objectid);
3488 err = __insert_orphan_inode(trans, root, objectid);
3491 inode = btrfs_iget(fs_info->sb, objectid, root);
3492 BUG_ON(IS_ERR(inode));
3493 BTRFS_I(inode)->index_cnt = group->start;
3495 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3497 btrfs_put_root(root);
3498 btrfs_end_transaction(trans);
3499 btrfs_btree_balance_dirty(fs_info);
3503 inode = ERR_PTR(err);
3508 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3510 struct reloc_control *rc;
3512 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3516 INIT_LIST_HEAD(&rc->reloc_roots);
3517 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3518 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3519 mapping_tree_init(&rc->reloc_root_tree);
3520 extent_io_tree_init(fs_info, &rc->processed_blocks,
3521 IO_TREE_RELOC_BLOCKS, NULL);
3525 static void free_reloc_control(struct reloc_control *rc)
3527 struct mapping_node *node, *tmp;
3529 free_reloc_roots(&rc->reloc_roots);
3530 rbtree_postorder_for_each_entry_safe(node, tmp,
3531 &rc->reloc_root_tree.rb_root, rb_node)
3538 * Print the block group being relocated
3540 static void describe_relocation(struct btrfs_fs_info *fs_info,
3541 struct btrfs_block_group *block_group)
3543 char buf[128] = {'\0'};
3545 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3548 "relocating block group %llu flags %s",
3549 block_group->start, buf);
3552 static const char *stage_to_string(int stage)
3554 if (stage == MOVE_DATA_EXTENTS)
3555 return "move data extents";
3556 if (stage == UPDATE_DATA_PTRS)
3557 return "update data pointers";
3562 * function to relocate all extents in a block group.
3564 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3566 struct btrfs_block_group *bg;
3567 struct btrfs_root *extent_root = fs_info->extent_root;
3568 struct reloc_control *rc;
3569 struct inode *inode;
3570 struct btrfs_path *path;
3575 bg = btrfs_lookup_block_group(fs_info, group_start);
3579 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3580 btrfs_put_block_group(bg);
3584 rc = alloc_reloc_control(fs_info);
3586 btrfs_put_block_group(bg);
3590 rc->extent_root = extent_root;
3591 rc->block_group = bg;
3593 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3600 path = btrfs_alloc_path();
3606 inode = lookup_free_space_inode(rc->block_group, path);
3607 btrfs_free_path(path);
3610 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3612 ret = PTR_ERR(inode);
3614 if (ret && ret != -ENOENT) {
3619 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3620 if (IS_ERR(rc->data_inode)) {
3621 err = PTR_ERR(rc->data_inode);
3622 rc->data_inode = NULL;
3626 describe_relocation(fs_info, rc->block_group);
3628 btrfs_wait_block_group_reservations(rc->block_group);
3629 btrfs_wait_nocow_writers(rc->block_group);
3630 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3631 rc->block_group->start,
3632 rc->block_group->length);
3637 mutex_lock(&fs_info->cleaner_mutex);
3638 ret = relocate_block_group(rc);
3639 mutex_unlock(&fs_info->cleaner_mutex);
3643 finishes_stage = rc->stage;
3645 * We may have gotten ENOSPC after we already dirtied some
3646 * extents. If writeout happens while we're relocating a
3647 * different block group we could end up hitting the
3648 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3649 * btrfs_reloc_cow_block. Make sure we write everything out
3650 * properly so we don't trip over this problem, and then break
3651 * out of the loop if we hit an error.
3653 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3654 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3658 invalidate_mapping_pages(rc->data_inode->i_mapping,
3660 rc->stage = UPDATE_DATA_PTRS;
3666 if (rc->extents_found == 0)
3669 btrfs_info(fs_info, "found %llu extents, stage: %s",
3670 rc->extents_found, stage_to_string(finishes_stage));
3673 WARN_ON(rc->block_group->pinned > 0);
3674 WARN_ON(rc->block_group->reserved > 0);
3675 WARN_ON(rc->block_group->used > 0);
3678 btrfs_dec_block_group_ro(rc->block_group);
3679 iput(rc->data_inode);
3680 btrfs_put_block_group(rc->block_group);
3681 free_reloc_control(rc);
3685 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3687 struct btrfs_fs_info *fs_info = root->fs_info;
3688 struct btrfs_trans_handle *trans;
3691 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3693 return PTR_ERR(trans);
3695 memset(&root->root_item.drop_progress, 0,
3696 sizeof(root->root_item.drop_progress));
3697 root->root_item.drop_level = 0;
3698 btrfs_set_root_refs(&root->root_item, 0);
3699 ret = btrfs_update_root(trans, fs_info->tree_root,
3700 &root->root_key, &root->root_item);
3702 err = btrfs_end_transaction(trans);
3709 * recover relocation interrupted by system crash.
3711 * this function resumes merging reloc trees with corresponding fs trees.
3712 * this is important for keeping the sharing of tree blocks
3714 int btrfs_recover_relocation(struct btrfs_root *root)
3716 struct btrfs_fs_info *fs_info = root->fs_info;
3717 LIST_HEAD(reloc_roots);
3718 struct btrfs_key key;
3719 struct btrfs_root *fs_root;
3720 struct btrfs_root *reloc_root;
3721 struct btrfs_path *path;
3722 struct extent_buffer *leaf;
3723 struct reloc_control *rc = NULL;
3724 struct btrfs_trans_handle *trans;
3728 path = btrfs_alloc_path();
3731 path->reada = READA_BACK;
3733 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3734 key.type = BTRFS_ROOT_ITEM_KEY;
3735 key.offset = (u64)-1;
3738 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3745 if (path->slots[0] == 0)
3749 leaf = path->nodes[0];
3750 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3751 btrfs_release_path(path);
3753 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3754 key.type != BTRFS_ROOT_ITEM_KEY)
3757 reloc_root = btrfs_read_tree_root(root, &key);
3758 if (IS_ERR(reloc_root)) {
3759 err = PTR_ERR(reloc_root);
3763 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3764 list_add(&reloc_root->root_list, &reloc_roots);
3766 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3767 fs_root = btrfs_get_fs_root(fs_info,
3768 reloc_root->root_key.offset, false);
3769 if (IS_ERR(fs_root)) {
3770 ret = PTR_ERR(fs_root);
3771 if (ret != -ENOENT) {
3775 ret = mark_garbage_root(reloc_root);
3781 btrfs_put_root(fs_root);
3785 if (key.offset == 0)
3790 btrfs_release_path(path);
3792 if (list_empty(&reloc_roots))
3795 rc = alloc_reloc_control(fs_info);
3801 rc->extent_root = fs_info->extent_root;
3803 set_reloc_control(rc);
3805 trans = btrfs_join_transaction(rc->extent_root);
3806 if (IS_ERR(trans)) {
3807 err = PTR_ERR(trans);
3811 rc->merge_reloc_tree = 1;
3813 while (!list_empty(&reloc_roots)) {
3814 reloc_root = list_entry(reloc_roots.next,
3815 struct btrfs_root, root_list);
3816 list_del(&reloc_root->root_list);
3818 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3819 list_add_tail(&reloc_root->root_list,
3824 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
3826 if (IS_ERR(fs_root)) {
3827 err = PTR_ERR(fs_root);
3828 list_add_tail(&reloc_root->root_list, &reloc_roots);
3829 btrfs_end_transaction(trans);
3833 err = __add_reloc_root(reloc_root);
3834 BUG_ON(err < 0); /* -ENOMEM or logic error */
3835 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3836 btrfs_put_root(fs_root);
3839 err = btrfs_commit_transaction(trans);
3843 merge_reloc_roots(rc);
3845 unset_reloc_control(rc);
3847 trans = btrfs_join_transaction(rc->extent_root);
3848 if (IS_ERR(trans)) {
3849 err = PTR_ERR(trans);
3852 err = btrfs_commit_transaction(trans);
3854 ret = clean_dirty_subvols(rc);
3855 if (ret < 0 && !err)
3858 unset_reloc_control(rc);
3859 free_reloc_control(rc);
3861 free_reloc_roots(&reloc_roots);
3863 btrfs_free_path(path);
3866 /* cleanup orphan inode in data relocation tree */
3867 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
3869 err = btrfs_orphan_cleanup(fs_root);
3870 btrfs_put_root(fs_root);
3876 * helper to add ordered checksum for data relocation.
3878 * cloning checksum properly handles the nodatasum extents.
3879 * it also saves CPU time to re-calculate the checksum.
3881 int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
3883 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3884 struct btrfs_ordered_sum *sums;
3885 struct btrfs_ordered_extent *ordered;
3891 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3892 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3894 disk_bytenr = file_pos + inode->index_cnt;
3895 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3896 disk_bytenr + len - 1, &list, 0);
3900 while (!list_empty(&list)) {
3901 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3902 list_del_init(&sums->list);
3905 * We need to offset the new_bytenr based on where the csum is.
3906 * We need to do this because we will read in entire prealloc
3907 * extents but we may have written to say the middle of the
3908 * prealloc extent, so we need to make sure the csum goes with
3909 * the right disk offset.
3911 * We can do this because the data reloc inode refers strictly
3912 * to the on disk bytes, so we don't have to worry about
3913 * disk_len vs real len like with real inodes since it's all
3916 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3917 sums->bytenr = new_bytenr;
3919 btrfs_add_ordered_sum(ordered, sums);
3922 btrfs_put_ordered_extent(ordered);
3926 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3927 struct btrfs_root *root, struct extent_buffer *buf,
3928 struct extent_buffer *cow)
3930 struct btrfs_fs_info *fs_info = root->fs_info;
3931 struct reloc_control *rc;
3932 struct btrfs_backref_node *node;
3937 rc = fs_info->reloc_ctl;
3941 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3942 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3944 level = btrfs_header_level(buf);
3945 if (btrfs_header_generation(buf) <=
3946 btrfs_root_last_snapshot(&root->root_item))
3949 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3950 rc->create_reloc_tree) {
3951 WARN_ON(!first_cow && level == 0);
3953 node = rc->backref_cache.path[level];
3954 BUG_ON(node->bytenr != buf->start &&
3955 node->new_bytenr != buf->start);
3957 btrfs_backref_drop_node_buffer(node);
3958 atomic_inc(&cow->refs);
3960 node->new_bytenr = cow->start;
3962 if (!node->pending) {
3963 list_move_tail(&node->list,
3964 &rc->backref_cache.pending[level]);
3969 mark_block_processed(rc, node);
3971 if (first_cow && level > 0)
3972 rc->nodes_relocated += buf->len;
3975 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
3976 ret = replace_file_extents(trans, rc, root, cow);
3981 * called before creating snapshot. it calculates metadata reservation
3982 * required for relocating tree blocks in the snapshot
3984 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
3985 u64 *bytes_to_reserve)
3987 struct btrfs_root *root = pending->root;
3988 struct reloc_control *rc = root->fs_info->reloc_ctl;
3990 if (!rc || !have_reloc_root(root))
3993 if (!rc->merge_reloc_tree)
3996 root = root->reloc_root;
3997 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
3999 * relocation is in the stage of merging trees. the space
4000 * used by merging a reloc tree is twice the size of
4001 * relocated tree nodes in the worst case. half for cowing
4002 * the reloc tree, half for cowing the fs tree. the space
4003 * used by cowing the reloc tree will be freed after the
4004 * tree is dropped. if we create snapshot, cowing the fs
4005 * tree may use more space than it frees. so we need
4006 * reserve extra space.
4008 *bytes_to_reserve += rc->nodes_relocated;
4012 * called after snapshot is created. migrate block reservation
4013 * and create reloc root for the newly created snapshot
4015 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4016 * references held on the reloc_root, one for root->reloc_root and one for
4019 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4020 struct btrfs_pending_snapshot *pending)
4022 struct btrfs_root *root = pending->root;
4023 struct btrfs_root *reloc_root;
4024 struct btrfs_root *new_root;
4025 struct reloc_control *rc = root->fs_info->reloc_ctl;
4028 if (!rc || !have_reloc_root(root))
4031 rc = root->fs_info->reloc_ctl;
4032 rc->merging_rsv_size += rc->nodes_relocated;
4034 if (rc->merge_reloc_tree) {
4035 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4037 rc->nodes_relocated, true);
4042 new_root = pending->snap;
4043 reloc_root = create_reloc_root(trans, root->reloc_root,
4044 new_root->root_key.objectid);
4045 if (IS_ERR(reloc_root))
4046 return PTR_ERR(reloc_root);
4048 ret = __add_reloc_root(reloc_root);
4050 new_root->reloc_root = btrfs_grab_root(reloc_root);
4052 if (rc->create_reloc_tree)
4053 ret = clone_backref_node(trans, rc, root, reloc_root);