Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
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 "ctree.h"
13#include "disk-io.h"
14#include "transaction.h"
15#include "volumes.h"
16#include "locking.h"
17#include "btrfs_inode.h"
18#include "async-thread.h"
19#include "free-space-cache.h"
20#include "inode-map.h"
21#include "qgroup.h"
22#include "print-tree.h"
23#include "delalloc-space.h"
24#include "block-group.h"
25
26/*
27 * backref_node, mapping_node and tree_block start with this
28 */
29struct tree_entry {
30 struct rb_node rb_node;
31 u64 bytenr;
32};
33
34/*
35 * present a tree block in the backref cache
36 */
37struct backref_node {
38 struct rb_node rb_node;
39 u64 bytenr;
40
41 u64 new_bytenr;
42 /* objectid of tree block owner, can be not uptodate */
43 u64 owner;
44 /* link to pending, changed or detached list */
45 struct list_head list;
46 /* list of upper level blocks reference this block */
47 struct list_head upper;
48 /* list of child blocks in the cache */
49 struct list_head lower;
50 /* NULL if this node is not tree root */
51 struct btrfs_root *root;
52 /* extent buffer got by COW the block */
53 struct extent_buffer *eb;
54 /* level of tree block */
55 unsigned int level:8;
56 /* is the block in non-reference counted tree */
57 unsigned int cowonly:1;
58 /* 1 if no child node in the cache */
59 unsigned int lowest:1;
60 /* is the extent buffer locked */
61 unsigned int locked:1;
62 /* has the block been processed */
63 unsigned int processed:1;
64 /* have backrefs of this block been checked */
65 unsigned int checked:1;
66 /*
67 * 1 if corresponding block has been cowed but some upper
68 * level block pointers may not point to the new location
69 */
70 unsigned int pending:1;
71 /*
72 * 1 if the backref node isn't connected to any other
73 * backref node.
74 */
75 unsigned int detached:1;
76};
77
78/*
79 * present a block pointer in the backref cache
80 */
81struct backref_edge {
82 struct list_head list[2];
83 struct backref_node *node[2];
84};
85
86#define LOWER 0
87#define UPPER 1
88#define RELOCATION_RESERVED_NODES 256
89
90struct backref_cache {
91 /* red black tree of all backref nodes in the cache */
92 struct rb_root rb_root;
93 /* for passing backref nodes to btrfs_reloc_cow_block */
94 struct backref_node *path[BTRFS_MAX_LEVEL];
95 /*
96 * list of blocks that have been cowed but some block
97 * pointers in upper level blocks may not reflect the
98 * new location
99 */
100 struct list_head pending[BTRFS_MAX_LEVEL];
101 /* list of backref nodes with no child node */
102 struct list_head leaves;
103 /* list of blocks that have been cowed in current transaction */
104 struct list_head changed;
105 /* list of detached backref node. */
106 struct list_head detached;
107
108 u64 last_trans;
109
110 int nr_nodes;
111 int nr_edges;
112};
113
114/*
115 * map address of tree root to tree
116 */
117struct mapping_node {
118 struct rb_node rb_node;
119 u64 bytenr;
120 void *data;
121};
122
123struct mapping_tree {
124 struct rb_root rb_root;
125 spinlock_t lock;
126};
127
128/*
129 * present a tree block to process
130 */
131struct tree_block {
132 struct rb_node rb_node;
133 u64 bytenr;
134 struct btrfs_key key;
135 unsigned int level:8;
136 unsigned int key_ready:1;
137};
138
139#define MAX_EXTENTS 128
140
141struct file_extent_cluster {
142 u64 start;
143 u64 end;
144 u64 boundary[MAX_EXTENTS];
145 unsigned int nr;
146};
147
148struct reloc_control {
149 /* block group to relocate */
150 struct btrfs_block_group_cache *block_group;
151 /* extent tree */
152 struct btrfs_root *extent_root;
153 /* inode for moving data */
154 struct inode *data_inode;
155
156 struct btrfs_block_rsv *block_rsv;
157
158 struct backref_cache backref_cache;
159
160 struct file_extent_cluster cluster;
161 /* tree blocks have been processed */
162 struct extent_io_tree processed_blocks;
163 /* map start of tree root to corresponding reloc tree */
164 struct mapping_tree reloc_root_tree;
165 /* list of reloc trees */
166 struct list_head reloc_roots;
167 /* list of subvolume trees that get relocated */
168 struct list_head dirty_subvol_roots;
169 /* size of metadata reservation for merging reloc trees */
170 u64 merging_rsv_size;
171 /* size of relocated tree nodes */
172 u64 nodes_relocated;
173 /* reserved size for block group relocation*/
174 u64 reserved_bytes;
175
176 u64 search_start;
177 u64 extents_found;
178
179 unsigned int stage:8;
180 unsigned int create_reloc_tree:1;
181 unsigned int merge_reloc_tree:1;
182 unsigned int found_file_extent:1;
183};
184
185/* stages of data relocation */
186#define MOVE_DATA_EXTENTS 0
187#define UPDATE_DATA_PTRS 1
188
189static void remove_backref_node(struct backref_cache *cache,
190 struct backref_node *node);
191static void __mark_block_processed(struct reloc_control *rc,
192 struct backref_node *node);
193
194static void mapping_tree_init(struct mapping_tree *tree)
195{
196 tree->rb_root = RB_ROOT;
197 spin_lock_init(&tree->lock);
198}
199
200static void backref_cache_init(struct backref_cache *cache)
201{
202 int i;
203 cache->rb_root = RB_ROOT;
204 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
205 INIT_LIST_HEAD(&cache->pending[i]);
206 INIT_LIST_HEAD(&cache->changed);
207 INIT_LIST_HEAD(&cache->detached);
208 INIT_LIST_HEAD(&cache->leaves);
209}
210
211static void backref_cache_cleanup(struct backref_cache *cache)
212{
213 struct backref_node *node;
214 int i;
215
216 while (!list_empty(&cache->detached)) {
217 node = list_entry(cache->detached.next,
218 struct backref_node, list);
219 remove_backref_node(cache, node);
220 }
221
222 while (!list_empty(&cache->leaves)) {
223 node = list_entry(cache->leaves.next,
224 struct backref_node, lower);
225 remove_backref_node(cache, node);
226 }
227
228 cache->last_trans = 0;
229
230 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
231 ASSERT(list_empty(&cache->pending[i]));
232 ASSERT(list_empty(&cache->changed));
233 ASSERT(list_empty(&cache->detached));
234 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
235 ASSERT(!cache->nr_nodes);
236 ASSERT(!cache->nr_edges);
237}
238
239static struct backref_node *alloc_backref_node(struct backref_cache *cache)
240{
241 struct backref_node *node;
242
243 node = kzalloc(sizeof(*node), GFP_NOFS);
244 if (node) {
245 INIT_LIST_HEAD(&node->list);
246 INIT_LIST_HEAD(&node->upper);
247 INIT_LIST_HEAD(&node->lower);
248 RB_CLEAR_NODE(&node->rb_node);
249 cache->nr_nodes++;
250 }
251 return node;
252}
253
254static void free_backref_node(struct backref_cache *cache,
255 struct backref_node *node)
256{
257 if (node) {
258 cache->nr_nodes--;
259 kfree(node);
260 }
261}
262
263static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
264{
265 struct backref_edge *edge;
266
267 edge = kzalloc(sizeof(*edge), GFP_NOFS);
268 if (edge)
269 cache->nr_edges++;
270 return edge;
271}
272
273static void free_backref_edge(struct backref_cache *cache,
274 struct backref_edge *edge)
275{
276 if (edge) {
277 cache->nr_edges--;
278 kfree(edge);
279 }
280}
281
282static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
283 struct rb_node *node)
284{
285 struct rb_node **p = &root->rb_node;
286 struct rb_node *parent = NULL;
287 struct tree_entry *entry;
288
289 while (*p) {
290 parent = *p;
291 entry = rb_entry(parent, struct tree_entry, rb_node);
292
293 if (bytenr < entry->bytenr)
294 p = &(*p)->rb_left;
295 else if (bytenr > entry->bytenr)
296 p = &(*p)->rb_right;
297 else
298 return parent;
299 }
300
301 rb_link_node(node, parent, p);
302 rb_insert_color(node, root);
303 return NULL;
304}
305
306static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
307{
308 struct rb_node *n = root->rb_node;
309 struct tree_entry *entry;
310
311 while (n) {
312 entry = rb_entry(n, struct tree_entry, rb_node);
313
314 if (bytenr < entry->bytenr)
315 n = n->rb_left;
316 else if (bytenr > entry->bytenr)
317 n = n->rb_right;
318 else
319 return n;
320 }
321 return NULL;
322}
323
324static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
325{
326
327 struct btrfs_fs_info *fs_info = NULL;
328 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
329 rb_node);
330 if (bnode->root)
331 fs_info = bnode->root->fs_info;
332 btrfs_panic(fs_info, errno,
333 "Inconsistency in backref cache found at offset %llu",
334 bytenr);
335}
336
337/*
338 * walk up backref nodes until reach node presents tree root
339 */
340static struct backref_node *walk_up_backref(struct backref_node *node,
341 struct backref_edge *edges[],
342 int *index)
343{
344 struct backref_edge *edge;
345 int idx = *index;
346
347 while (!list_empty(&node->upper)) {
348 edge = list_entry(node->upper.next,
349 struct backref_edge, list[LOWER]);
350 edges[idx++] = edge;
351 node = edge->node[UPPER];
352 }
353 BUG_ON(node->detached);
354 *index = idx;
355 return node;
356}
357
358/*
359 * walk down backref nodes to find start of next reference path
360 */
361static struct backref_node *walk_down_backref(struct backref_edge *edges[],
362 int *index)
363{
364 struct backref_edge *edge;
365 struct backref_node *lower;
366 int idx = *index;
367
368 while (idx > 0) {
369 edge = edges[idx - 1];
370 lower = edge->node[LOWER];
371 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
372 idx--;
373 continue;
374 }
375 edge = list_entry(edge->list[LOWER].next,
376 struct backref_edge, list[LOWER]);
377 edges[idx - 1] = edge;
378 *index = idx;
379 return edge->node[UPPER];
380 }
381 *index = 0;
382 return NULL;
383}
384
385static void unlock_node_buffer(struct backref_node *node)
386{
387 if (node->locked) {
388 btrfs_tree_unlock(node->eb);
389 node->locked = 0;
390 }
391}
392
393static void drop_node_buffer(struct backref_node *node)
394{
395 if (node->eb) {
396 unlock_node_buffer(node);
397 free_extent_buffer(node->eb);
398 node->eb = NULL;
399 }
400}
401
402static void drop_backref_node(struct backref_cache *tree,
403 struct backref_node *node)
404{
405 BUG_ON(!list_empty(&node->upper));
406
407 drop_node_buffer(node);
408 list_del(&node->list);
409 list_del(&node->lower);
410 if (!RB_EMPTY_NODE(&node->rb_node))
411 rb_erase(&node->rb_node, &tree->rb_root);
412 free_backref_node(tree, node);
413}
414
415/*
416 * remove a backref node from the backref cache
417 */
418static void remove_backref_node(struct backref_cache *cache,
419 struct backref_node *node)
420{
421 struct backref_node *upper;
422 struct backref_edge *edge;
423
424 if (!node)
425 return;
426
427 BUG_ON(!node->lowest && !node->detached);
428 while (!list_empty(&node->upper)) {
429 edge = list_entry(node->upper.next, struct backref_edge,
430 list[LOWER]);
431 upper = edge->node[UPPER];
432 list_del(&edge->list[LOWER]);
433 list_del(&edge->list[UPPER]);
434 free_backref_edge(cache, edge);
435
436 if (RB_EMPTY_NODE(&upper->rb_node)) {
437 BUG_ON(!list_empty(&node->upper));
438 drop_backref_node(cache, node);
439 node = upper;
440 node->lowest = 1;
441 continue;
442 }
443 /*
444 * add the node to leaf node list if no other
445 * child block cached.
446 */
447 if (list_empty(&upper->lower)) {
448 list_add_tail(&upper->lower, &cache->leaves);
449 upper->lowest = 1;
450 }
451 }
452
453 drop_backref_node(cache, node);
454}
455
456static void update_backref_node(struct backref_cache *cache,
457 struct backref_node *node, u64 bytenr)
458{
459 struct rb_node *rb_node;
460 rb_erase(&node->rb_node, &cache->rb_root);
461 node->bytenr = bytenr;
462 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
463 if (rb_node)
464 backref_tree_panic(rb_node, -EEXIST, bytenr);
465}
466
467/*
468 * update backref cache after a transaction commit
469 */
470static int update_backref_cache(struct btrfs_trans_handle *trans,
471 struct backref_cache *cache)
472{
473 struct backref_node *node;
474 int level = 0;
475
476 if (cache->last_trans == 0) {
477 cache->last_trans = trans->transid;
478 return 0;
479 }
480
481 if (cache->last_trans == trans->transid)
482 return 0;
483
484 /*
485 * detached nodes are used to avoid unnecessary backref
486 * lookup. transaction commit changes the extent tree.
487 * so the detached nodes are no longer useful.
488 */
489 while (!list_empty(&cache->detached)) {
490 node = list_entry(cache->detached.next,
491 struct backref_node, list);
492 remove_backref_node(cache, node);
493 }
494
495 while (!list_empty(&cache->changed)) {
496 node = list_entry(cache->changed.next,
497 struct backref_node, list);
498 list_del_init(&node->list);
499 BUG_ON(node->pending);
500 update_backref_node(cache, node, node->new_bytenr);
501 }
502
503 /*
504 * some nodes can be left in the pending list if there were
505 * errors during processing the pending nodes.
506 */
507 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
508 list_for_each_entry(node, &cache->pending[level], list) {
509 BUG_ON(!node->pending);
510 if (node->bytenr == node->new_bytenr)
511 continue;
512 update_backref_node(cache, node, node->new_bytenr);
513 }
514 }
515
516 cache->last_trans = 0;
517 return 1;
518}
519
520
521static int should_ignore_root(struct btrfs_root *root)
522{
523 struct btrfs_root *reloc_root;
524
525 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
526 return 0;
527
528 reloc_root = root->reloc_root;
529 if (!reloc_root)
530 return 0;
531
532 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
533 root->fs_info->running_transaction->transid - 1)
534 return 0;
535 /*
536 * if there is reloc tree and it was created in previous
537 * transaction backref lookup can find the reloc tree,
538 * so backref node for the fs tree root is useless for
539 * relocation.
540 */
541 return 1;
542}
543/*
544 * find reloc tree by address of tree root
545 */
546static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
547 u64 bytenr)
548{
549 struct rb_node *rb_node;
550 struct mapping_node *node;
551 struct btrfs_root *root = NULL;
552
553 spin_lock(&rc->reloc_root_tree.lock);
554 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
555 if (rb_node) {
556 node = rb_entry(rb_node, struct mapping_node, rb_node);
557 root = (struct btrfs_root *)node->data;
558 }
559 spin_unlock(&rc->reloc_root_tree.lock);
560 return root;
561}
562
563static int is_cowonly_root(u64 root_objectid)
564{
565 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
566 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
567 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
568 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
569 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
570 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
571 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
572 root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
573 root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
574 return 1;
575 return 0;
576}
577
578static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
579 u64 root_objectid)
580{
581 struct btrfs_key key;
582
583 key.objectid = root_objectid;
584 key.type = BTRFS_ROOT_ITEM_KEY;
585 if (is_cowonly_root(root_objectid))
586 key.offset = 0;
587 else
588 key.offset = (u64)-1;
589
590 return btrfs_get_fs_root(fs_info, &key, false);
591}
592
593static noinline_for_stack
594int find_inline_backref(struct extent_buffer *leaf, int slot,
595 unsigned long *ptr, unsigned long *end)
596{
597 struct btrfs_key key;
598 struct btrfs_extent_item *ei;
599 struct btrfs_tree_block_info *bi;
600 u32 item_size;
601
602 btrfs_item_key_to_cpu(leaf, &key, slot);
603
604 item_size = btrfs_item_size_nr(leaf, slot);
605 if (item_size < sizeof(*ei)) {
606 btrfs_print_v0_err(leaf->fs_info);
607 btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
608 return 1;
609 }
610 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
611 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
612 BTRFS_EXTENT_FLAG_TREE_BLOCK));
613
614 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
615 item_size <= sizeof(*ei) + sizeof(*bi)) {
616 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
617 return 1;
618 }
619 if (key.type == BTRFS_METADATA_ITEM_KEY &&
620 item_size <= sizeof(*ei)) {
621 WARN_ON(item_size < sizeof(*ei));
622 return 1;
623 }
624
625 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
626 bi = (struct btrfs_tree_block_info *)(ei + 1);
627 *ptr = (unsigned long)(bi + 1);
628 } else {
629 *ptr = (unsigned long)(ei + 1);
630 }
631 *end = (unsigned long)ei + item_size;
632 return 0;
633}
634
635/*
636 * build backref tree for a given tree block. root of the backref tree
637 * corresponds the tree block, leaves of the backref tree correspond
638 * roots of b-trees that reference the tree block.
639 *
640 * the basic idea of this function is check backrefs of a given block
641 * to find upper level blocks that reference the block, and then check
642 * backrefs of these upper level blocks recursively. the recursion stop
643 * when tree root is reached or backrefs for the block is cached.
644 *
645 * NOTE: if we find backrefs for a block are cached, we know backrefs
646 * for all upper level blocks that directly/indirectly reference the
647 * block are also cached.
648 */
649static noinline_for_stack
650struct backref_node *build_backref_tree(struct reloc_control *rc,
651 struct btrfs_key *node_key,
652 int level, u64 bytenr)
653{
654 struct backref_cache *cache = &rc->backref_cache;
655 struct btrfs_path *path1; /* For searching extent root */
656 struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */
657 struct extent_buffer *eb;
658 struct btrfs_root *root;
659 struct backref_node *cur;
660 struct backref_node *upper;
661 struct backref_node *lower;
662 struct backref_node *node = NULL;
663 struct backref_node *exist = NULL;
664 struct backref_edge *edge;
665 struct rb_node *rb_node;
666 struct btrfs_key key;
667 unsigned long end;
668 unsigned long ptr;
669 LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */
670 LIST_HEAD(useless);
671 int cowonly;
672 int ret;
673 int err = 0;
674 bool need_check = true;
675
676 path1 = btrfs_alloc_path();
677 path2 = btrfs_alloc_path();
678 if (!path1 || !path2) {
679 err = -ENOMEM;
680 goto out;
681 }
682 path1->reada = READA_FORWARD;
683 path2->reada = READA_FORWARD;
684
685 node = alloc_backref_node(cache);
686 if (!node) {
687 err = -ENOMEM;
688 goto out;
689 }
690
691 node->bytenr = bytenr;
692 node->level = level;
693 node->lowest = 1;
694 cur = node;
695again:
696 end = 0;
697 ptr = 0;
698 key.objectid = cur->bytenr;
699 key.type = BTRFS_METADATA_ITEM_KEY;
700 key.offset = (u64)-1;
701
702 path1->search_commit_root = 1;
703 path1->skip_locking = 1;
704 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
705 0, 0);
706 if (ret < 0) {
707 err = ret;
708 goto out;
709 }
710 ASSERT(ret);
711 ASSERT(path1->slots[0]);
712
713 path1->slots[0]--;
714
715 WARN_ON(cur->checked);
716 if (!list_empty(&cur->upper)) {
717 /*
718 * the backref was added previously when processing
719 * backref of type BTRFS_TREE_BLOCK_REF_KEY
720 */
721 ASSERT(list_is_singular(&cur->upper));
722 edge = list_entry(cur->upper.next, struct backref_edge,
723 list[LOWER]);
724 ASSERT(list_empty(&edge->list[UPPER]));
725 exist = edge->node[UPPER];
726 /*
727 * add the upper level block to pending list if we need
728 * check its backrefs
729 */
730 if (!exist->checked)
731 list_add_tail(&edge->list[UPPER], &list);
732 } else {
733 exist = NULL;
734 }
735
736 while (1) {
737 cond_resched();
738 eb = path1->nodes[0];
739
740 if (ptr >= end) {
741 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
742 ret = btrfs_next_leaf(rc->extent_root, path1);
743 if (ret < 0) {
744 err = ret;
745 goto out;
746 }
747 if (ret > 0)
748 break;
749 eb = path1->nodes[0];
750 }
751
752 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
753 if (key.objectid != cur->bytenr) {
754 WARN_ON(exist);
755 break;
756 }
757
758 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
759 key.type == BTRFS_METADATA_ITEM_KEY) {
760 ret = find_inline_backref(eb, path1->slots[0],
761 &ptr, &end);
762 if (ret)
763 goto next;
764 }
765 }
766
767 if (ptr < end) {
768 /* update key for inline back ref */
769 struct btrfs_extent_inline_ref *iref;
770 int type;
771 iref = (struct btrfs_extent_inline_ref *)ptr;
772 type = btrfs_get_extent_inline_ref_type(eb, iref,
773 BTRFS_REF_TYPE_BLOCK);
774 if (type == BTRFS_REF_TYPE_INVALID) {
775 err = -EUCLEAN;
776 goto out;
777 }
778 key.type = type;
779 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
780
781 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
782 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
783 }
784
785 /*
786 * Parent node found and matches current inline ref, no need to
787 * rebuild this node for this inline ref.
788 */
789 if (exist &&
790 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
791 exist->owner == key.offset) ||
792 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
793 exist->bytenr == key.offset))) {
794 exist = NULL;
795 goto next;
796 }
797
798 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
799 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
800 if (key.objectid == key.offset) {
801 /*
802 * Only root blocks of reloc trees use backref
803 * pointing to itself.
804 */
805 root = find_reloc_root(rc, cur->bytenr);
806 ASSERT(root);
807 cur->root = root;
808 break;
809 }
810
811 edge = alloc_backref_edge(cache);
812 if (!edge) {
813 err = -ENOMEM;
814 goto out;
815 }
816 rb_node = tree_search(&cache->rb_root, key.offset);
817 if (!rb_node) {
818 upper = alloc_backref_node(cache);
819 if (!upper) {
820 free_backref_edge(cache, edge);
821 err = -ENOMEM;
822 goto out;
823 }
824 upper->bytenr = key.offset;
825 upper->level = cur->level + 1;
826 /*
827 * backrefs for the upper level block isn't
828 * cached, add the block to pending list
829 */
830 list_add_tail(&edge->list[UPPER], &list);
831 } else {
832 upper = rb_entry(rb_node, struct backref_node,
833 rb_node);
834 ASSERT(upper->checked);
835 INIT_LIST_HEAD(&edge->list[UPPER]);
836 }
837 list_add_tail(&edge->list[LOWER], &cur->upper);
838 edge->node[LOWER] = cur;
839 edge->node[UPPER] = upper;
840
841 goto next;
842 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
843 err = -EINVAL;
844 btrfs_print_v0_err(rc->extent_root->fs_info);
845 btrfs_handle_fs_error(rc->extent_root->fs_info, err,
846 NULL);
847 goto out;
848 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
849 goto next;
850 }
851
852 /*
853 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
854 * means the root objectid. We need to search the tree to get
855 * its parent bytenr.
856 */
857 root = read_fs_root(rc->extent_root->fs_info, key.offset);
858 if (IS_ERR(root)) {
859 err = PTR_ERR(root);
860 goto out;
861 }
862
863 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
864 cur->cowonly = 1;
865
866 if (btrfs_root_level(&root->root_item) == cur->level) {
867 /* tree root */
868 ASSERT(btrfs_root_bytenr(&root->root_item) ==
869 cur->bytenr);
870 if (should_ignore_root(root))
871 list_add(&cur->list, &useless);
872 else
873 cur->root = root;
874 break;
875 }
876
877 level = cur->level + 1;
878
879 /* Search the tree to find parent blocks referring the block. */
880 path2->search_commit_root = 1;
881 path2->skip_locking = 1;
882 path2->lowest_level = level;
883 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
884 path2->lowest_level = 0;
885 if (ret < 0) {
886 err = ret;
887 goto out;
888 }
889 if (ret > 0 && path2->slots[level] > 0)
890 path2->slots[level]--;
891
892 eb = path2->nodes[level];
893 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
894 cur->bytenr) {
895 btrfs_err(root->fs_info,
896 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
897 cur->bytenr, level - 1,
898 root->root_key.objectid,
899 node_key->objectid, node_key->type,
900 node_key->offset);
901 err = -ENOENT;
902 goto out;
903 }
904 lower = cur;
905 need_check = true;
906
907 /* Add all nodes and edges in the path */
908 for (; level < BTRFS_MAX_LEVEL; level++) {
909 if (!path2->nodes[level]) {
910 ASSERT(btrfs_root_bytenr(&root->root_item) ==
911 lower->bytenr);
912 if (should_ignore_root(root))
913 list_add(&lower->list, &useless);
914 else
915 lower->root = root;
916 break;
917 }
918
919 edge = alloc_backref_edge(cache);
920 if (!edge) {
921 err = -ENOMEM;
922 goto out;
923 }
924
925 eb = path2->nodes[level];
926 rb_node = tree_search(&cache->rb_root, eb->start);
927 if (!rb_node) {
928 upper = alloc_backref_node(cache);
929 if (!upper) {
930 free_backref_edge(cache, edge);
931 err = -ENOMEM;
932 goto out;
933 }
934 upper->bytenr = eb->start;
935 upper->owner = btrfs_header_owner(eb);
936 upper->level = lower->level + 1;
937 if (!test_bit(BTRFS_ROOT_REF_COWS,
938 &root->state))
939 upper->cowonly = 1;
940
941 /*
942 * if we know the block isn't shared
943 * we can void checking its backrefs.
944 */
945 if (btrfs_block_can_be_shared(root, eb))
946 upper->checked = 0;
947 else
948 upper->checked = 1;
949
950 /*
951 * add the block to pending list if we
952 * need check its backrefs, we only do this once
953 * while walking up a tree as we will catch
954 * anything else later on.
955 */
956 if (!upper->checked && need_check) {
957 need_check = false;
958 list_add_tail(&edge->list[UPPER],
959 &list);
960 } else {
961 if (upper->checked)
962 need_check = true;
963 INIT_LIST_HEAD(&edge->list[UPPER]);
964 }
965 } else {
966 upper = rb_entry(rb_node, struct backref_node,
967 rb_node);
968 ASSERT(upper->checked);
969 INIT_LIST_HEAD(&edge->list[UPPER]);
970 if (!upper->owner)
971 upper->owner = btrfs_header_owner(eb);
972 }
973 list_add_tail(&edge->list[LOWER], &lower->upper);
974 edge->node[LOWER] = lower;
975 edge->node[UPPER] = upper;
976
977 if (rb_node)
978 break;
979 lower = upper;
980 upper = NULL;
981 }
982 btrfs_release_path(path2);
983next:
984 if (ptr < end) {
985 ptr += btrfs_extent_inline_ref_size(key.type);
986 if (ptr >= end) {
987 WARN_ON(ptr > end);
988 ptr = 0;
989 end = 0;
990 }
991 }
992 if (ptr >= end)
993 path1->slots[0]++;
994 }
995 btrfs_release_path(path1);
996
997 cur->checked = 1;
998 WARN_ON(exist);
999
1000 /* the pending list isn't empty, take the first block to process */
1001 if (!list_empty(&list)) {
1002 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1003 list_del_init(&edge->list[UPPER]);
1004 cur = edge->node[UPPER];
1005 goto again;
1006 }
1007
1008 /*
1009 * everything goes well, connect backref nodes and insert backref nodes
1010 * into the cache.
1011 */
1012 ASSERT(node->checked);
1013 cowonly = node->cowonly;
1014 if (!cowonly) {
1015 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1016 &node->rb_node);
1017 if (rb_node)
1018 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1019 list_add_tail(&node->lower, &cache->leaves);
1020 }
1021
1022 list_for_each_entry(edge, &node->upper, list[LOWER])
1023 list_add_tail(&edge->list[UPPER], &list);
1024
1025 while (!list_empty(&list)) {
1026 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1027 list_del_init(&edge->list[UPPER]);
1028 upper = edge->node[UPPER];
1029 if (upper->detached) {
1030 list_del(&edge->list[LOWER]);
1031 lower = edge->node[LOWER];
1032 free_backref_edge(cache, edge);
1033 if (list_empty(&lower->upper))
1034 list_add(&lower->list, &useless);
1035 continue;
1036 }
1037
1038 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1039 if (upper->lowest) {
1040 list_del_init(&upper->lower);
1041 upper->lowest = 0;
1042 }
1043
1044 list_add_tail(&edge->list[UPPER], &upper->lower);
1045 continue;
1046 }
1047
1048 if (!upper->checked) {
1049 /*
1050 * Still want to blow up for developers since this is a
1051 * logic bug.
1052 */
1053 ASSERT(0);
1054 err = -EINVAL;
1055 goto out;
1056 }
1057 if (cowonly != upper->cowonly) {
1058 ASSERT(0);
1059 err = -EINVAL;
1060 goto out;
1061 }
1062
1063 if (!cowonly) {
1064 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1065 &upper->rb_node);
1066 if (rb_node)
1067 backref_tree_panic(rb_node, -EEXIST,
1068 upper->bytenr);
1069 }
1070
1071 list_add_tail(&edge->list[UPPER], &upper->lower);
1072
1073 list_for_each_entry(edge, &upper->upper, list[LOWER])
1074 list_add_tail(&edge->list[UPPER], &list);
1075 }
1076 /*
1077 * process useless backref nodes. backref nodes for tree leaves
1078 * are deleted from the cache. backref nodes for upper level
1079 * tree blocks are left in the cache to avoid unnecessary backref
1080 * lookup.
1081 */
1082 while (!list_empty(&useless)) {
1083 upper = list_entry(useless.next, struct backref_node, list);
1084 list_del_init(&upper->list);
1085 ASSERT(list_empty(&upper->upper));
1086 if (upper == node)
1087 node = NULL;
1088 if (upper->lowest) {
1089 list_del_init(&upper->lower);
1090 upper->lowest = 0;
1091 }
1092 while (!list_empty(&upper->lower)) {
1093 edge = list_entry(upper->lower.next,
1094 struct backref_edge, list[UPPER]);
1095 list_del(&edge->list[UPPER]);
1096 list_del(&edge->list[LOWER]);
1097 lower = edge->node[LOWER];
1098 free_backref_edge(cache, edge);
1099
1100 if (list_empty(&lower->upper))
1101 list_add(&lower->list, &useless);
1102 }
1103 __mark_block_processed(rc, upper);
1104 if (upper->level > 0) {
1105 list_add(&upper->list, &cache->detached);
1106 upper->detached = 1;
1107 } else {
1108 rb_erase(&upper->rb_node, &cache->rb_root);
1109 free_backref_node(cache, upper);
1110 }
1111 }
1112out:
1113 btrfs_free_path(path1);
1114 btrfs_free_path(path2);
1115 if (err) {
1116 while (!list_empty(&useless)) {
1117 lower = list_entry(useless.next,
1118 struct backref_node, list);
1119 list_del_init(&lower->list);
1120 }
1121 while (!list_empty(&list)) {
1122 edge = list_first_entry(&list, struct backref_edge,
1123 list[UPPER]);
1124 list_del(&edge->list[UPPER]);
1125 list_del(&edge->list[LOWER]);
1126 lower = edge->node[LOWER];
1127 upper = edge->node[UPPER];
1128 free_backref_edge(cache, edge);
1129
1130 /*
1131 * Lower is no longer linked to any upper backref nodes
1132 * and isn't in the cache, we can free it ourselves.
1133 */
1134 if (list_empty(&lower->upper) &&
1135 RB_EMPTY_NODE(&lower->rb_node))
1136 list_add(&lower->list, &useless);
1137
1138 if (!RB_EMPTY_NODE(&upper->rb_node))
1139 continue;
1140
1141 /* Add this guy's upper edges to the list to process */
1142 list_for_each_entry(edge, &upper->upper, list[LOWER])
1143 list_add_tail(&edge->list[UPPER], &list);
1144 if (list_empty(&upper->upper))
1145 list_add(&upper->list, &useless);
1146 }
1147
1148 while (!list_empty(&useless)) {
1149 lower = list_entry(useless.next,
1150 struct backref_node, list);
1151 list_del_init(&lower->list);
1152 if (lower == node)
1153 node = NULL;
1154 free_backref_node(cache, lower);
1155 }
1156
1157 free_backref_node(cache, node);
1158 return ERR_PTR(err);
1159 }
1160 ASSERT(!node || !node->detached);
1161 return node;
1162}
1163
1164/*
1165 * helper to add backref node for the newly created snapshot.
1166 * the backref node is created by cloning backref node that
1167 * corresponds to root of source tree
1168 */
1169static int clone_backref_node(struct btrfs_trans_handle *trans,
1170 struct reloc_control *rc,
1171 struct btrfs_root *src,
1172 struct btrfs_root *dest)
1173{
1174 struct btrfs_root *reloc_root = src->reloc_root;
1175 struct backref_cache *cache = &rc->backref_cache;
1176 struct backref_node *node = NULL;
1177 struct backref_node *new_node;
1178 struct backref_edge *edge;
1179 struct backref_edge *new_edge;
1180 struct rb_node *rb_node;
1181
1182 if (cache->last_trans > 0)
1183 update_backref_cache(trans, cache);
1184
1185 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1186 if (rb_node) {
1187 node = rb_entry(rb_node, struct backref_node, rb_node);
1188 if (node->detached)
1189 node = NULL;
1190 else
1191 BUG_ON(node->new_bytenr != reloc_root->node->start);
1192 }
1193
1194 if (!node) {
1195 rb_node = tree_search(&cache->rb_root,
1196 reloc_root->commit_root->start);
1197 if (rb_node) {
1198 node = rb_entry(rb_node, struct backref_node,
1199 rb_node);
1200 BUG_ON(node->detached);
1201 }
1202 }
1203
1204 if (!node)
1205 return 0;
1206
1207 new_node = alloc_backref_node(cache);
1208 if (!new_node)
1209 return -ENOMEM;
1210
1211 new_node->bytenr = dest->node->start;
1212 new_node->level = node->level;
1213 new_node->lowest = node->lowest;
1214 new_node->checked = 1;
1215 new_node->root = dest;
1216
1217 if (!node->lowest) {
1218 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1219 new_edge = alloc_backref_edge(cache);
1220 if (!new_edge)
1221 goto fail;
1222
1223 new_edge->node[UPPER] = new_node;
1224 new_edge->node[LOWER] = edge->node[LOWER];
1225 list_add_tail(&new_edge->list[UPPER],
1226 &new_node->lower);
1227 }
1228 } else {
1229 list_add_tail(&new_node->lower, &cache->leaves);
1230 }
1231
1232 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1233 &new_node->rb_node);
1234 if (rb_node)
1235 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1236
1237 if (!new_node->lowest) {
1238 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1239 list_add_tail(&new_edge->list[LOWER],
1240 &new_edge->node[LOWER]->upper);
1241 }
1242 }
1243 return 0;
1244fail:
1245 while (!list_empty(&new_node->lower)) {
1246 new_edge = list_entry(new_node->lower.next,
1247 struct backref_edge, list[UPPER]);
1248 list_del(&new_edge->list[UPPER]);
1249 free_backref_edge(cache, new_edge);
1250 }
1251 free_backref_node(cache, new_node);
1252 return -ENOMEM;
1253}
1254
1255/*
1256 * helper to add 'address of tree root -> reloc tree' mapping
1257 */
1258static int __must_check __add_reloc_root(struct btrfs_root *root)
1259{
1260 struct btrfs_fs_info *fs_info = root->fs_info;
1261 struct rb_node *rb_node;
1262 struct mapping_node *node;
1263 struct reloc_control *rc = fs_info->reloc_ctl;
1264
1265 node = kmalloc(sizeof(*node), GFP_NOFS);
1266 if (!node)
1267 return -ENOMEM;
1268
1269 node->bytenr = root->node->start;
1270 node->data = root;
1271
1272 spin_lock(&rc->reloc_root_tree.lock);
1273 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1274 node->bytenr, &node->rb_node);
1275 spin_unlock(&rc->reloc_root_tree.lock);
1276 if (rb_node) {
1277 btrfs_panic(fs_info, -EEXIST,
1278 "Duplicate root found for start=%llu while inserting into relocation tree",
1279 node->bytenr);
1280 }
1281
1282 list_add_tail(&root->root_list, &rc->reloc_roots);
1283 return 0;
1284}
1285
1286/*
1287 * helper to delete the 'address of tree root -> reloc tree'
1288 * mapping
1289 */
1290static void __del_reloc_root(struct btrfs_root *root)
1291{
1292 struct btrfs_fs_info *fs_info = root->fs_info;
1293 struct rb_node *rb_node;
1294 struct mapping_node *node = NULL;
1295 struct reloc_control *rc = fs_info->reloc_ctl;
1296
1297 if (rc && root->node) {
1298 spin_lock(&rc->reloc_root_tree.lock);
1299 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1300 root->node->start);
1301 if (rb_node) {
1302 node = rb_entry(rb_node, struct mapping_node, rb_node);
1303 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1304 }
1305 spin_unlock(&rc->reloc_root_tree.lock);
1306 if (!node)
1307 return;
1308 BUG_ON((struct btrfs_root *)node->data != root);
1309 }
1310
1311 spin_lock(&fs_info->trans_lock);
1312 list_del_init(&root->root_list);
1313 spin_unlock(&fs_info->trans_lock);
1314 kfree(node);
1315}
1316
1317/*
1318 * helper to update the 'address of tree root -> reloc tree'
1319 * mapping
1320 */
1321static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1322{
1323 struct btrfs_fs_info *fs_info = root->fs_info;
1324 struct rb_node *rb_node;
1325 struct mapping_node *node = NULL;
1326 struct reloc_control *rc = fs_info->reloc_ctl;
1327
1328 spin_lock(&rc->reloc_root_tree.lock);
1329 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1330 root->node->start);
1331 if (rb_node) {
1332 node = rb_entry(rb_node, struct mapping_node, rb_node);
1333 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1334 }
1335 spin_unlock(&rc->reloc_root_tree.lock);
1336
1337 if (!node)
1338 return 0;
1339 BUG_ON((struct btrfs_root *)node->data != root);
1340
1341 spin_lock(&rc->reloc_root_tree.lock);
1342 node->bytenr = new_bytenr;
1343 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1344 node->bytenr, &node->rb_node);
1345 spin_unlock(&rc->reloc_root_tree.lock);
1346 if (rb_node)
1347 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1348 return 0;
1349}
1350
1351static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1352 struct btrfs_root *root, u64 objectid)
1353{
1354 struct btrfs_fs_info *fs_info = root->fs_info;
1355 struct btrfs_root *reloc_root;
1356 struct extent_buffer *eb;
1357 struct btrfs_root_item *root_item;
1358 struct btrfs_key root_key;
1359 int ret;
1360
1361 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1362 BUG_ON(!root_item);
1363
1364 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1365 root_key.type = BTRFS_ROOT_ITEM_KEY;
1366 root_key.offset = objectid;
1367
1368 if (root->root_key.objectid == objectid) {
1369 u64 commit_root_gen;
1370
1371 /* called by btrfs_init_reloc_root */
1372 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1373 BTRFS_TREE_RELOC_OBJECTID);
1374 BUG_ON(ret);
1375 /*
1376 * Set the last_snapshot field to the generation of the commit
1377 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1378 * correctly (returns true) when the relocation root is created
1379 * either inside the critical section of a transaction commit
1380 * (through transaction.c:qgroup_account_snapshot()) and when
1381 * it's created before the transaction commit is started.
1382 */
1383 commit_root_gen = btrfs_header_generation(root->commit_root);
1384 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1385 } else {
1386 /*
1387 * called by btrfs_reloc_post_snapshot_hook.
1388 * the source tree is a reloc tree, all tree blocks
1389 * modified after it was created have RELOC flag
1390 * set in their headers. so it's OK to not update
1391 * the 'last_snapshot'.
1392 */
1393 ret = btrfs_copy_root(trans, root, root->node, &eb,
1394 BTRFS_TREE_RELOC_OBJECTID);
1395 BUG_ON(ret);
1396 }
1397
1398 memcpy(root_item, &root->root_item, sizeof(*root_item));
1399 btrfs_set_root_bytenr(root_item, eb->start);
1400 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1401 btrfs_set_root_generation(root_item, trans->transid);
1402
1403 if (root->root_key.objectid == objectid) {
1404 btrfs_set_root_refs(root_item, 0);
1405 memset(&root_item->drop_progress, 0,
1406 sizeof(struct btrfs_disk_key));
1407 root_item->drop_level = 0;
1408 }
1409
1410 btrfs_tree_unlock(eb);
1411 free_extent_buffer(eb);
1412
1413 ret = btrfs_insert_root(trans, fs_info->tree_root,
1414 &root_key, root_item);
1415 BUG_ON(ret);
1416 kfree(root_item);
1417
1418 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1419 BUG_ON(IS_ERR(reloc_root));
1420 reloc_root->last_trans = trans->transid;
1421 return reloc_root;
1422}
1423
1424/*
1425 * create reloc tree for a given fs tree. reloc tree is just a
1426 * snapshot of the fs tree with special root objectid.
1427 */
1428int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1429 struct btrfs_root *root)
1430{
1431 struct btrfs_fs_info *fs_info = root->fs_info;
1432 struct btrfs_root *reloc_root;
1433 struct reloc_control *rc = fs_info->reloc_ctl;
1434 struct btrfs_block_rsv *rsv;
1435 int clear_rsv = 0;
1436 int ret;
1437
1438 /*
1439 * The subvolume has reloc tree but the swap is finished, no need to
1440 * create/update the dead reloc tree
1441 */
1442 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
1443 return 0;
1444
1445 if (root->reloc_root) {
1446 reloc_root = root->reloc_root;
1447 reloc_root->last_trans = trans->transid;
1448 return 0;
1449 }
1450
1451 if (!rc || !rc->create_reloc_tree ||
1452 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1453 return 0;
1454
1455 if (!trans->reloc_reserved) {
1456 rsv = trans->block_rsv;
1457 trans->block_rsv = rc->block_rsv;
1458 clear_rsv = 1;
1459 }
1460 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1461 if (clear_rsv)
1462 trans->block_rsv = rsv;
1463
1464 ret = __add_reloc_root(reloc_root);
1465 BUG_ON(ret < 0);
1466 root->reloc_root = reloc_root;
1467 return 0;
1468}
1469
1470/*
1471 * update root item of reloc tree
1472 */
1473int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1474 struct btrfs_root *root)
1475{
1476 struct btrfs_fs_info *fs_info = root->fs_info;
1477 struct btrfs_root *reloc_root;
1478 struct btrfs_root_item *root_item;
1479 int ret;
1480
1481 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) ||
1482 !root->reloc_root)
1483 goto out;
1484
1485 reloc_root = root->reloc_root;
1486 root_item = &reloc_root->root_item;
1487
1488 /* root->reloc_root will stay until current relocation finished */
1489 if (fs_info->reloc_ctl->merge_reloc_tree &&
1490 btrfs_root_refs(root_item) == 0) {
1491 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1492 __del_reloc_root(reloc_root);
1493 }
1494
1495 if (reloc_root->commit_root != reloc_root->node) {
1496 btrfs_set_root_node(root_item, reloc_root->node);
1497 free_extent_buffer(reloc_root->commit_root);
1498 reloc_root->commit_root = btrfs_root_node(reloc_root);
1499 }
1500
1501 ret = btrfs_update_root(trans, fs_info->tree_root,
1502 &reloc_root->root_key, root_item);
1503 BUG_ON(ret);
1504
1505out:
1506 return 0;
1507}
1508
1509/*
1510 * helper to find first cached inode with inode number >= objectid
1511 * in a subvolume
1512 */
1513static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1514{
1515 struct rb_node *node;
1516 struct rb_node *prev;
1517 struct btrfs_inode *entry;
1518 struct inode *inode;
1519
1520 spin_lock(&root->inode_lock);
1521again:
1522 node = root->inode_tree.rb_node;
1523 prev = NULL;
1524 while (node) {
1525 prev = node;
1526 entry = rb_entry(node, struct btrfs_inode, rb_node);
1527
1528 if (objectid < btrfs_ino(entry))
1529 node = node->rb_left;
1530 else if (objectid > btrfs_ino(entry))
1531 node = node->rb_right;
1532 else
1533 break;
1534 }
1535 if (!node) {
1536 while (prev) {
1537 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1538 if (objectid <= btrfs_ino(entry)) {
1539 node = prev;
1540 break;
1541 }
1542 prev = rb_next(prev);
1543 }
1544 }
1545 while (node) {
1546 entry = rb_entry(node, struct btrfs_inode, rb_node);
1547 inode = igrab(&entry->vfs_inode);
1548 if (inode) {
1549 spin_unlock(&root->inode_lock);
1550 return inode;
1551 }
1552
1553 objectid = btrfs_ino(entry) + 1;
1554 if (cond_resched_lock(&root->inode_lock))
1555 goto again;
1556
1557 node = rb_next(node);
1558 }
1559 spin_unlock(&root->inode_lock);
1560 return NULL;
1561}
1562
1563static int in_block_group(u64 bytenr,
1564 struct btrfs_block_group_cache *block_group)
1565{
1566 if (bytenr >= block_group->key.objectid &&
1567 bytenr < block_group->key.objectid + block_group->key.offset)
1568 return 1;
1569 return 0;
1570}
1571
1572/*
1573 * get new location of data
1574 */
1575static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1576 u64 bytenr, u64 num_bytes)
1577{
1578 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1579 struct btrfs_path *path;
1580 struct btrfs_file_extent_item *fi;
1581 struct extent_buffer *leaf;
1582 int ret;
1583
1584 path = btrfs_alloc_path();
1585 if (!path)
1586 return -ENOMEM;
1587
1588 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1589 ret = btrfs_lookup_file_extent(NULL, root, path,
1590 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1591 if (ret < 0)
1592 goto out;
1593 if (ret > 0) {
1594 ret = -ENOENT;
1595 goto out;
1596 }
1597
1598 leaf = path->nodes[0];
1599 fi = btrfs_item_ptr(leaf, path->slots[0],
1600 struct btrfs_file_extent_item);
1601
1602 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1603 btrfs_file_extent_compression(leaf, fi) ||
1604 btrfs_file_extent_encryption(leaf, fi) ||
1605 btrfs_file_extent_other_encoding(leaf, fi));
1606
1607 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1608 ret = -EINVAL;
1609 goto out;
1610 }
1611
1612 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1613 ret = 0;
1614out:
1615 btrfs_free_path(path);
1616 return ret;
1617}
1618
1619/*
1620 * update file extent items in the tree leaf to point to
1621 * the new locations.
1622 */
1623static noinline_for_stack
1624int replace_file_extents(struct btrfs_trans_handle *trans,
1625 struct reloc_control *rc,
1626 struct btrfs_root *root,
1627 struct extent_buffer *leaf)
1628{
1629 struct btrfs_fs_info *fs_info = root->fs_info;
1630 struct btrfs_key key;
1631 struct btrfs_file_extent_item *fi;
1632 struct inode *inode = NULL;
1633 u64 parent;
1634 u64 bytenr;
1635 u64 new_bytenr = 0;
1636 u64 num_bytes;
1637 u64 end;
1638 u32 nritems;
1639 u32 i;
1640 int ret = 0;
1641 int first = 1;
1642 int dirty = 0;
1643
1644 if (rc->stage != UPDATE_DATA_PTRS)
1645 return 0;
1646
1647 /* reloc trees always use full backref */
1648 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1649 parent = leaf->start;
1650 else
1651 parent = 0;
1652
1653 nritems = btrfs_header_nritems(leaf);
1654 for (i = 0; i < nritems; i++) {
1655 struct btrfs_ref ref = { 0 };
1656
1657 cond_resched();
1658 btrfs_item_key_to_cpu(leaf, &key, i);
1659 if (key.type != BTRFS_EXTENT_DATA_KEY)
1660 continue;
1661 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1662 if (btrfs_file_extent_type(leaf, fi) ==
1663 BTRFS_FILE_EXTENT_INLINE)
1664 continue;
1665 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1666 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1667 if (bytenr == 0)
1668 continue;
1669 if (!in_block_group(bytenr, rc->block_group))
1670 continue;
1671
1672 /*
1673 * if we are modifying block in fs tree, wait for readpage
1674 * to complete and drop the extent cache
1675 */
1676 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1677 if (first) {
1678 inode = find_next_inode(root, key.objectid);
1679 first = 0;
1680 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1681 btrfs_add_delayed_iput(inode);
1682 inode = find_next_inode(root, key.objectid);
1683 }
1684 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1685 end = key.offset +
1686 btrfs_file_extent_num_bytes(leaf, fi);
1687 WARN_ON(!IS_ALIGNED(key.offset,
1688 fs_info->sectorsize));
1689 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1690 end--;
1691 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1692 key.offset, end);
1693 if (!ret)
1694 continue;
1695
1696 btrfs_drop_extent_cache(BTRFS_I(inode),
1697 key.offset, end, 1);
1698 unlock_extent(&BTRFS_I(inode)->io_tree,
1699 key.offset, end);
1700 }
1701 }
1702
1703 ret = get_new_location(rc->data_inode, &new_bytenr,
1704 bytenr, num_bytes);
1705 if (ret) {
1706 /*
1707 * Don't have to abort since we've not changed anything
1708 * in the file extent yet.
1709 */
1710 break;
1711 }
1712
1713 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1714 dirty = 1;
1715
1716 key.offset -= btrfs_file_extent_offset(leaf, fi);
1717 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1718 num_bytes, parent);
1719 ref.real_root = root->root_key.objectid;
1720 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1721 key.objectid, key.offset);
1722 ret = btrfs_inc_extent_ref(trans, &ref);
1723 if (ret) {
1724 btrfs_abort_transaction(trans, ret);
1725 break;
1726 }
1727
1728 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1729 num_bytes, parent);
1730 ref.real_root = root->root_key.objectid;
1731 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1732 key.objectid, key.offset);
1733 ret = btrfs_free_extent(trans, &ref);
1734 if (ret) {
1735 btrfs_abort_transaction(trans, ret);
1736 break;
1737 }
1738 }
1739 if (dirty)
1740 btrfs_mark_buffer_dirty(leaf);
1741 if (inode)
1742 btrfs_add_delayed_iput(inode);
1743 return ret;
1744}
1745
1746static noinline_for_stack
1747int memcmp_node_keys(struct extent_buffer *eb, int slot,
1748 struct btrfs_path *path, int level)
1749{
1750 struct btrfs_disk_key key1;
1751 struct btrfs_disk_key key2;
1752 btrfs_node_key(eb, &key1, slot);
1753 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1754 return memcmp(&key1, &key2, sizeof(key1));
1755}
1756
1757/*
1758 * try to replace tree blocks in fs tree with the new blocks
1759 * in reloc tree. tree blocks haven't been modified since the
1760 * reloc tree was create can be replaced.
1761 *
1762 * if a block was replaced, level of the block + 1 is returned.
1763 * if no block got replaced, 0 is returned. if there are other
1764 * errors, a negative error number is returned.
1765 */
1766static noinline_for_stack
1767int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1768 struct btrfs_root *dest, struct btrfs_root *src,
1769 struct btrfs_path *path, struct btrfs_key *next_key,
1770 int lowest_level, int max_level)
1771{
1772 struct btrfs_fs_info *fs_info = dest->fs_info;
1773 struct extent_buffer *eb;
1774 struct extent_buffer *parent;
1775 struct btrfs_ref ref = { 0 };
1776 struct btrfs_key key;
1777 u64 old_bytenr;
1778 u64 new_bytenr;
1779 u64 old_ptr_gen;
1780 u64 new_ptr_gen;
1781 u64 last_snapshot;
1782 u32 blocksize;
1783 int cow = 0;
1784 int level;
1785 int ret;
1786 int slot;
1787
1788 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1789 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1790
1791 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1792again:
1793 slot = path->slots[lowest_level];
1794 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1795
1796 eb = btrfs_lock_root_node(dest);
1797 btrfs_set_lock_blocking_write(eb);
1798 level = btrfs_header_level(eb);
1799
1800 if (level < lowest_level) {
1801 btrfs_tree_unlock(eb);
1802 free_extent_buffer(eb);
1803 return 0;
1804 }
1805
1806 if (cow) {
1807 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1808 BUG_ON(ret);
1809 }
1810 btrfs_set_lock_blocking_write(eb);
1811
1812 if (next_key) {
1813 next_key->objectid = (u64)-1;
1814 next_key->type = (u8)-1;
1815 next_key->offset = (u64)-1;
1816 }
1817
1818 parent = eb;
1819 while (1) {
1820 struct btrfs_key first_key;
1821
1822 level = btrfs_header_level(parent);
1823 BUG_ON(level < lowest_level);
1824
1825 ret = btrfs_bin_search(parent, &key, level, &slot);
1826 if (ret < 0)
1827 break;
1828 if (ret && slot > 0)
1829 slot--;
1830
1831 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1832 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1833
1834 old_bytenr = btrfs_node_blockptr(parent, slot);
1835 blocksize = fs_info->nodesize;
1836 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1837 btrfs_node_key_to_cpu(parent, &first_key, slot);
1838
1839 if (level <= max_level) {
1840 eb = path->nodes[level];
1841 new_bytenr = btrfs_node_blockptr(eb,
1842 path->slots[level]);
1843 new_ptr_gen = btrfs_node_ptr_generation(eb,
1844 path->slots[level]);
1845 } else {
1846 new_bytenr = 0;
1847 new_ptr_gen = 0;
1848 }
1849
1850 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1851 ret = level;
1852 break;
1853 }
1854
1855 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1856 memcmp_node_keys(parent, slot, path, level)) {
1857 if (level <= lowest_level) {
1858 ret = 0;
1859 break;
1860 }
1861
1862 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1863 level - 1, &first_key);
1864 if (IS_ERR(eb)) {
1865 ret = PTR_ERR(eb);
1866 break;
1867 } else if (!extent_buffer_uptodate(eb)) {
1868 ret = -EIO;
1869 free_extent_buffer(eb);
1870 break;
1871 }
1872 btrfs_tree_lock(eb);
1873 if (cow) {
1874 ret = btrfs_cow_block(trans, dest, eb, parent,
1875 slot, &eb);
1876 BUG_ON(ret);
1877 }
1878 btrfs_set_lock_blocking_write(eb);
1879
1880 btrfs_tree_unlock(parent);
1881 free_extent_buffer(parent);
1882
1883 parent = eb;
1884 continue;
1885 }
1886
1887 if (!cow) {
1888 btrfs_tree_unlock(parent);
1889 free_extent_buffer(parent);
1890 cow = 1;
1891 goto again;
1892 }
1893
1894 btrfs_node_key_to_cpu(path->nodes[level], &key,
1895 path->slots[level]);
1896 btrfs_release_path(path);
1897
1898 path->lowest_level = level;
1899 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1900 path->lowest_level = 0;
1901 BUG_ON(ret);
1902
1903 /*
1904 * Info qgroup to trace both subtrees.
1905 *
1906 * We must trace both trees.
1907 * 1) Tree reloc subtree
1908 * If not traced, we will leak data numbers
1909 * 2) Fs subtree
1910 * If not traced, we will double count old data
1911 *
1912 * We don't scan the subtree right now, but only record
1913 * the swapped tree blocks.
1914 * The real subtree rescan is delayed until we have new
1915 * CoW on the subtree root node before transaction commit.
1916 */
1917 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1918 rc->block_group, parent, slot,
1919 path->nodes[level], path->slots[level],
1920 last_snapshot);
1921 if (ret < 0)
1922 break;
1923 /*
1924 * swap blocks in fs tree and reloc tree.
1925 */
1926 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1927 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1928 btrfs_mark_buffer_dirty(parent);
1929
1930 btrfs_set_node_blockptr(path->nodes[level],
1931 path->slots[level], old_bytenr);
1932 btrfs_set_node_ptr_generation(path->nodes[level],
1933 path->slots[level], old_ptr_gen);
1934 btrfs_mark_buffer_dirty(path->nodes[level]);
1935
1936 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1937 blocksize, path->nodes[level]->start);
1938 ref.skip_qgroup = true;
1939 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1940 ret = btrfs_inc_extent_ref(trans, &ref);
1941 BUG_ON(ret);
1942 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1943 blocksize, 0);
1944 ref.skip_qgroup = true;
1945 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1946 ret = btrfs_inc_extent_ref(trans, &ref);
1947 BUG_ON(ret);
1948
1949 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1950 blocksize, path->nodes[level]->start);
1951 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1952 ref.skip_qgroup = true;
1953 ret = btrfs_free_extent(trans, &ref);
1954 BUG_ON(ret);
1955
1956 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1957 blocksize, 0);
1958 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1959 ref.skip_qgroup = true;
1960 ret = btrfs_free_extent(trans, &ref);
1961 BUG_ON(ret);
1962
1963 btrfs_unlock_up_safe(path, 0);
1964
1965 ret = level;
1966 break;
1967 }
1968 btrfs_tree_unlock(parent);
1969 free_extent_buffer(parent);
1970 return ret;
1971}
1972
1973/*
1974 * helper to find next relocated block in reloc tree
1975 */
1976static noinline_for_stack
1977int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1978 int *level)
1979{
1980 struct extent_buffer *eb;
1981 int i;
1982 u64 last_snapshot;
1983 u32 nritems;
1984
1985 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1986
1987 for (i = 0; i < *level; i++) {
1988 free_extent_buffer(path->nodes[i]);
1989 path->nodes[i] = NULL;
1990 }
1991
1992 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1993 eb = path->nodes[i];
1994 nritems = btrfs_header_nritems(eb);
1995 while (path->slots[i] + 1 < nritems) {
1996 path->slots[i]++;
1997 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1998 last_snapshot)
1999 continue;
2000
2001 *level = i;
2002 return 0;
2003 }
2004 free_extent_buffer(path->nodes[i]);
2005 path->nodes[i] = NULL;
2006 }
2007 return 1;
2008}
2009
2010/*
2011 * walk down reloc tree to find relocated block of lowest level
2012 */
2013static noinline_for_stack
2014int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2015 int *level)
2016{
2017 struct btrfs_fs_info *fs_info = root->fs_info;
2018 struct extent_buffer *eb = NULL;
2019 int i;
2020 u64 bytenr;
2021 u64 ptr_gen = 0;
2022 u64 last_snapshot;
2023 u32 nritems;
2024
2025 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2026
2027 for (i = *level; i > 0; i--) {
2028 struct btrfs_key first_key;
2029
2030 eb = path->nodes[i];
2031 nritems = btrfs_header_nritems(eb);
2032 while (path->slots[i] < nritems) {
2033 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2034 if (ptr_gen > last_snapshot)
2035 break;
2036 path->slots[i]++;
2037 }
2038 if (path->slots[i] >= nritems) {
2039 if (i == *level)
2040 break;
2041 *level = i + 1;
2042 return 0;
2043 }
2044 if (i == 1) {
2045 *level = i;
2046 return 0;
2047 }
2048
2049 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2050 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2051 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2052 &first_key);
2053 if (IS_ERR(eb)) {
2054 return PTR_ERR(eb);
2055 } else if (!extent_buffer_uptodate(eb)) {
2056 free_extent_buffer(eb);
2057 return -EIO;
2058 }
2059 BUG_ON(btrfs_header_level(eb) != i - 1);
2060 path->nodes[i - 1] = eb;
2061 path->slots[i - 1] = 0;
2062 }
2063 return 1;
2064}
2065
2066/*
2067 * invalidate extent cache for file extents whose key in range of
2068 * [min_key, max_key)
2069 */
2070static int invalidate_extent_cache(struct btrfs_root *root,
2071 struct btrfs_key *min_key,
2072 struct btrfs_key *max_key)
2073{
2074 struct btrfs_fs_info *fs_info = root->fs_info;
2075 struct inode *inode = NULL;
2076 u64 objectid;
2077 u64 start, end;
2078 u64 ino;
2079
2080 objectid = min_key->objectid;
2081 while (1) {
2082 cond_resched();
2083 iput(inode);
2084
2085 if (objectid > max_key->objectid)
2086 break;
2087
2088 inode = find_next_inode(root, objectid);
2089 if (!inode)
2090 break;
2091 ino = btrfs_ino(BTRFS_I(inode));
2092
2093 if (ino > max_key->objectid) {
2094 iput(inode);
2095 break;
2096 }
2097
2098 objectid = ino + 1;
2099 if (!S_ISREG(inode->i_mode))
2100 continue;
2101
2102 if (unlikely(min_key->objectid == ino)) {
2103 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2104 continue;
2105 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2106 start = 0;
2107 else {
2108 start = min_key->offset;
2109 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2110 }
2111 } else {
2112 start = 0;
2113 }
2114
2115 if (unlikely(max_key->objectid == ino)) {
2116 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2117 continue;
2118 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2119 end = (u64)-1;
2120 } else {
2121 if (max_key->offset == 0)
2122 continue;
2123 end = max_key->offset;
2124 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2125 end--;
2126 }
2127 } else {
2128 end = (u64)-1;
2129 }
2130
2131 /* the lock_extent waits for readpage to complete */
2132 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2133 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2134 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2135 }
2136 return 0;
2137}
2138
2139static int find_next_key(struct btrfs_path *path, int level,
2140 struct btrfs_key *key)
2141
2142{
2143 while (level < BTRFS_MAX_LEVEL) {
2144 if (!path->nodes[level])
2145 break;
2146 if (path->slots[level] + 1 <
2147 btrfs_header_nritems(path->nodes[level])) {
2148 btrfs_node_key_to_cpu(path->nodes[level], key,
2149 path->slots[level] + 1);
2150 return 0;
2151 }
2152 level++;
2153 }
2154 return 1;
2155}
2156
2157/*
2158 * Insert current subvolume into reloc_control::dirty_subvol_roots
2159 */
2160static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
2161 struct reloc_control *rc,
2162 struct btrfs_root *root)
2163{
2164 struct btrfs_root *reloc_root = root->reloc_root;
2165 struct btrfs_root_item *reloc_root_item;
2166
2167 /* @root must be a subvolume tree root with a valid reloc tree */
2168 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
2169 ASSERT(reloc_root);
2170
2171 reloc_root_item = &reloc_root->root_item;
2172 memset(&reloc_root_item->drop_progress, 0,
2173 sizeof(reloc_root_item->drop_progress));
2174 reloc_root_item->drop_level = 0;
2175 btrfs_set_root_refs(reloc_root_item, 0);
2176 btrfs_update_reloc_root(trans, root);
2177
2178 if (list_empty(&root->reloc_dirty_list)) {
2179 btrfs_grab_fs_root(root);
2180 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
2181 }
2182}
2183
2184static int clean_dirty_subvols(struct reloc_control *rc)
2185{
2186 struct btrfs_root *root;
2187 struct btrfs_root *next;
2188 int ret = 0;
2189 int ret2;
2190
2191 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
2192 reloc_dirty_list) {
2193 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2194 /* Merged subvolume, cleanup its reloc root */
2195 struct btrfs_root *reloc_root = root->reloc_root;
2196
2197 list_del_init(&root->reloc_dirty_list);
2198 root->reloc_root = NULL;
2199 if (reloc_root) {
2200
2201 ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
2202 if (ret2 < 0 && !ret)
2203 ret = ret2;
2204 }
2205 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
2206 btrfs_put_fs_root(root);
2207 } else {
2208 /* Orphan reloc tree, just clean it up */
2209 ret2 = btrfs_drop_snapshot(root, NULL, 0, 1);
2210 if (ret2 < 0 && !ret)
2211 ret = ret2;
2212 }
2213 }
2214 return ret;
2215}
2216
2217/*
2218 * merge the relocated tree blocks in reloc tree with corresponding
2219 * fs tree.
2220 */
2221static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2222 struct btrfs_root *root)
2223{
2224 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2225 struct btrfs_key key;
2226 struct btrfs_key next_key;
2227 struct btrfs_trans_handle *trans = NULL;
2228 struct btrfs_root *reloc_root;
2229 struct btrfs_root_item *root_item;
2230 struct btrfs_path *path;
2231 struct extent_buffer *leaf;
2232 int level;
2233 int max_level;
2234 int replaced = 0;
2235 int ret;
2236 int err = 0;
2237 u32 min_reserved;
2238
2239 path = btrfs_alloc_path();
2240 if (!path)
2241 return -ENOMEM;
2242 path->reada = READA_FORWARD;
2243
2244 reloc_root = root->reloc_root;
2245 root_item = &reloc_root->root_item;
2246
2247 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2248 level = btrfs_root_level(root_item);
2249 extent_buffer_get(reloc_root->node);
2250 path->nodes[level] = reloc_root->node;
2251 path->slots[level] = 0;
2252 } else {
2253 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2254
2255 level = root_item->drop_level;
2256 BUG_ON(level == 0);
2257 path->lowest_level = level;
2258 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2259 path->lowest_level = 0;
2260 if (ret < 0) {
2261 btrfs_free_path(path);
2262 return ret;
2263 }
2264
2265 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2266 path->slots[level]);
2267 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2268
2269 btrfs_unlock_up_safe(path, 0);
2270 }
2271
2272 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2273 memset(&next_key, 0, sizeof(next_key));
2274
2275 while (1) {
2276 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2277 BTRFS_RESERVE_FLUSH_ALL);
2278 if (ret) {
2279 err = ret;
2280 goto out;
2281 }
2282 trans = btrfs_start_transaction(root, 0);
2283 if (IS_ERR(trans)) {
2284 err = PTR_ERR(trans);
2285 trans = NULL;
2286 goto out;
2287 }
2288 trans->block_rsv = rc->block_rsv;
2289
2290 replaced = 0;
2291 max_level = level;
2292
2293 ret = walk_down_reloc_tree(reloc_root, path, &level);
2294 if (ret < 0) {
2295 err = ret;
2296 goto out;
2297 }
2298 if (ret > 0)
2299 break;
2300
2301 if (!find_next_key(path, level, &key) &&
2302 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2303 ret = 0;
2304 } else {
2305 ret = replace_path(trans, rc, root, reloc_root, path,
2306 &next_key, level, max_level);
2307 }
2308 if (ret < 0) {
2309 err = ret;
2310 goto out;
2311 }
2312
2313 if (ret > 0) {
2314 level = ret;
2315 btrfs_node_key_to_cpu(path->nodes[level], &key,
2316 path->slots[level]);
2317 replaced = 1;
2318 }
2319
2320 ret = walk_up_reloc_tree(reloc_root, path, &level);
2321 if (ret > 0)
2322 break;
2323
2324 BUG_ON(level == 0);
2325 /*
2326 * save the merging progress in the drop_progress.
2327 * this is OK since root refs == 1 in this case.
2328 */
2329 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2330 path->slots[level]);
2331 root_item->drop_level = level;
2332
2333 btrfs_end_transaction_throttle(trans);
2334 trans = NULL;
2335
2336 btrfs_btree_balance_dirty(fs_info);
2337
2338 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2339 invalidate_extent_cache(root, &key, &next_key);
2340 }
2341
2342 /*
2343 * handle the case only one block in the fs tree need to be
2344 * relocated and the block is tree root.
2345 */
2346 leaf = btrfs_lock_root_node(root);
2347 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2348 btrfs_tree_unlock(leaf);
2349 free_extent_buffer(leaf);
2350 if (ret < 0)
2351 err = ret;
2352out:
2353 btrfs_free_path(path);
2354
2355 if (err == 0)
2356 insert_dirty_subvol(trans, rc, root);
2357
2358 if (trans)
2359 btrfs_end_transaction_throttle(trans);
2360
2361 btrfs_btree_balance_dirty(fs_info);
2362
2363 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2364 invalidate_extent_cache(root, &key, &next_key);
2365
2366 return err;
2367}
2368
2369static noinline_for_stack
2370int prepare_to_merge(struct reloc_control *rc, int err)
2371{
2372 struct btrfs_root *root = rc->extent_root;
2373 struct btrfs_fs_info *fs_info = root->fs_info;
2374 struct btrfs_root *reloc_root;
2375 struct btrfs_trans_handle *trans;
2376 LIST_HEAD(reloc_roots);
2377 u64 num_bytes = 0;
2378 int ret;
2379
2380 mutex_lock(&fs_info->reloc_mutex);
2381 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2382 rc->merging_rsv_size += rc->nodes_relocated * 2;
2383 mutex_unlock(&fs_info->reloc_mutex);
2384
2385again:
2386 if (!err) {
2387 num_bytes = rc->merging_rsv_size;
2388 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2389 BTRFS_RESERVE_FLUSH_ALL);
2390 if (ret)
2391 err = ret;
2392 }
2393
2394 trans = btrfs_join_transaction(rc->extent_root);
2395 if (IS_ERR(trans)) {
2396 if (!err)
2397 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2398 num_bytes);
2399 return PTR_ERR(trans);
2400 }
2401
2402 if (!err) {
2403 if (num_bytes != rc->merging_rsv_size) {
2404 btrfs_end_transaction(trans);
2405 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2406 num_bytes);
2407 goto again;
2408 }
2409 }
2410
2411 rc->merge_reloc_tree = 1;
2412
2413 while (!list_empty(&rc->reloc_roots)) {
2414 reloc_root = list_entry(rc->reloc_roots.next,
2415 struct btrfs_root, root_list);
2416 list_del_init(&reloc_root->root_list);
2417
2418 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2419 BUG_ON(IS_ERR(root));
2420 BUG_ON(root->reloc_root != reloc_root);
2421
2422 /*
2423 * set reference count to 1, so btrfs_recover_relocation
2424 * knows it should resumes merging
2425 */
2426 if (!err)
2427 btrfs_set_root_refs(&reloc_root->root_item, 1);
2428 btrfs_update_reloc_root(trans, root);
2429
2430 list_add(&reloc_root->root_list, &reloc_roots);
2431 }
2432
2433 list_splice(&reloc_roots, &rc->reloc_roots);
2434
2435 if (!err)
2436 btrfs_commit_transaction(trans);
2437 else
2438 btrfs_end_transaction(trans);
2439 return err;
2440}
2441
2442static noinline_for_stack
2443void free_reloc_roots(struct list_head *list)
2444{
2445 struct btrfs_root *reloc_root;
2446
2447 while (!list_empty(list)) {
2448 reloc_root = list_entry(list->next, struct btrfs_root,
2449 root_list);
2450 __del_reloc_root(reloc_root);
2451 free_extent_buffer(reloc_root->node);
2452 free_extent_buffer(reloc_root->commit_root);
2453 reloc_root->node = NULL;
2454 reloc_root->commit_root = NULL;
2455 }
2456}
2457
2458static noinline_for_stack
2459void merge_reloc_roots(struct reloc_control *rc)
2460{
2461 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2462 struct btrfs_root *root;
2463 struct btrfs_root *reloc_root;
2464 LIST_HEAD(reloc_roots);
2465 int found = 0;
2466 int ret = 0;
2467again:
2468 root = rc->extent_root;
2469
2470 /*
2471 * this serializes us with btrfs_record_root_in_transaction,
2472 * we have to make sure nobody is in the middle of
2473 * adding their roots to the list while we are
2474 * doing this splice
2475 */
2476 mutex_lock(&fs_info->reloc_mutex);
2477 list_splice_init(&rc->reloc_roots, &reloc_roots);
2478 mutex_unlock(&fs_info->reloc_mutex);
2479
2480 while (!list_empty(&reloc_roots)) {
2481 found = 1;
2482 reloc_root = list_entry(reloc_roots.next,
2483 struct btrfs_root, root_list);
2484
2485 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2486 root = read_fs_root(fs_info,
2487 reloc_root->root_key.offset);
2488 BUG_ON(IS_ERR(root));
2489 BUG_ON(root->reloc_root != reloc_root);
2490
2491 ret = merge_reloc_root(rc, root);
2492 if (ret) {
2493 if (list_empty(&reloc_root->root_list))
2494 list_add_tail(&reloc_root->root_list,
2495 &reloc_roots);
2496 goto out;
2497 }
2498 } else {
2499 list_del_init(&reloc_root->root_list);
2500 /* Don't forget to queue this reloc root for cleanup */
2501 list_add_tail(&reloc_root->reloc_dirty_list,
2502 &rc->dirty_subvol_roots);
2503 }
2504 }
2505
2506 if (found) {
2507 found = 0;
2508 goto again;
2509 }
2510out:
2511 if (ret) {
2512 btrfs_handle_fs_error(fs_info, ret, NULL);
2513 if (!list_empty(&reloc_roots))
2514 free_reloc_roots(&reloc_roots);
2515
2516 /* new reloc root may be added */
2517 mutex_lock(&fs_info->reloc_mutex);
2518 list_splice_init(&rc->reloc_roots, &reloc_roots);
2519 mutex_unlock(&fs_info->reloc_mutex);
2520 if (!list_empty(&reloc_roots))
2521 free_reloc_roots(&reloc_roots);
2522 }
2523
2524 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2525}
2526
2527static void free_block_list(struct rb_root *blocks)
2528{
2529 struct tree_block *block;
2530 struct rb_node *rb_node;
2531 while ((rb_node = rb_first(blocks))) {
2532 block = rb_entry(rb_node, struct tree_block, rb_node);
2533 rb_erase(rb_node, blocks);
2534 kfree(block);
2535 }
2536}
2537
2538static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2539 struct btrfs_root *reloc_root)
2540{
2541 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2542 struct btrfs_root *root;
2543
2544 if (reloc_root->last_trans == trans->transid)
2545 return 0;
2546
2547 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2548 BUG_ON(IS_ERR(root));
2549 BUG_ON(root->reloc_root != reloc_root);
2550
2551 return btrfs_record_root_in_trans(trans, root);
2552}
2553
2554static noinline_for_stack
2555struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2556 struct reloc_control *rc,
2557 struct backref_node *node,
2558 struct backref_edge *edges[])
2559{
2560 struct backref_node *next;
2561 struct btrfs_root *root;
2562 int index = 0;
2563
2564 next = node;
2565 while (1) {
2566 cond_resched();
2567 next = walk_up_backref(next, edges, &index);
2568 root = next->root;
2569 BUG_ON(!root);
2570 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2571
2572 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2573 record_reloc_root_in_trans(trans, root);
2574 break;
2575 }
2576
2577 btrfs_record_root_in_trans(trans, root);
2578 root = root->reloc_root;
2579
2580 if (next->new_bytenr != root->node->start) {
2581 BUG_ON(next->new_bytenr);
2582 BUG_ON(!list_empty(&next->list));
2583 next->new_bytenr = root->node->start;
2584 next->root = root;
2585 list_add_tail(&next->list,
2586 &rc->backref_cache.changed);
2587 __mark_block_processed(rc, next);
2588 break;
2589 }
2590
2591 WARN_ON(1);
2592 root = NULL;
2593 next = walk_down_backref(edges, &index);
2594 if (!next || next->level <= node->level)
2595 break;
2596 }
2597 if (!root)
2598 return NULL;
2599
2600 next = node;
2601 /* setup backref node path for btrfs_reloc_cow_block */
2602 while (1) {
2603 rc->backref_cache.path[next->level] = next;
2604 if (--index < 0)
2605 break;
2606 next = edges[index]->node[UPPER];
2607 }
2608 return root;
2609}
2610
2611/*
2612 * select a tree root for relocation. return NULL if the block
2613 * is reference counted. we should use do_relocation() in this
2614 * case. return a tree root pointer if the block isn't reference
2615 * counted. return -ENOENT if the block is root of reloc tree.
2616 */
2617static noinline_for_stack
2618struct btrfs_root *select_one_root(struct backref_node *node)
2619{
2620 struct backref_node *next;
2621 struct btrfs_root *root;
2622 struct btrfs_root *fs_root = NULL;
2623 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2624 int index = 0;
2625
2626 next = node;
2627 while (1) {
2628 cond_resched();
2629 next = walk_up_backref(next, edges, &index);
2630 root = next->root;
2631 BUG_ON(!root);
2632
2633 /* no other choice for non-references counted tree */
2634 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2635 return root;
2636
2637 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2638 fs_root = root;
2639
2640 if (next != node)
2641 return NULL;
2642
2643 next = walk_down_backref(edges, &index);
2644 if (!next || next->level <= node->level)
2645 break;
2646 }
2647
2648 if (!fs_root)
2649 return ERR_PTR(-ENOENT);
2650 return fs_root;
2651}
2652
2653static noinline_for_stack
2654u64 calcu_metadata_size(struct reloc_control *rc,
2655 struct backref_node *node, int reserve)
2656{
2657 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2658 struct backref_node *next = node;
2659 struct backref_edge *edge;
2660 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2661 u64 num_bytes = 0;
2662 int index = 0;
2663
2664 BUG_ON(reserve && node->processed);
2665
2666 while (next) {
2667 cond_resched();
2668 while (1) {
2669 if (next->processed && (reserve || next != node))
2670 break;
2671
2672 num_bytes += fs_info->nodesize;
2673
2674 if (list_empty(&next->upper))
2675 break;
2676
2677 edge = list_entry(next->upper.next,
2678 struct backref_edge, list[LOWER]);
2679 edges[index++] = edge;
2680 next = edge->node[UPPER];
2681 }
2682 next = walk_down_backref(edges, &index);
2683 }
2684 return num_bytes;
2685}
2686
2687static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2688 struct reloc_control *rc,
2689 struct backref_node *node)
2690{
2691 struct btrfs_root *root = rc->extent_root;
2692 struct btrfs_fs_info *fs_info = root->fs_info;
2693 u64 num_bytes;
2694 int ret;
2695 u64 tmp;
2696
2697 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2698
2699 trans->block_rsv = rc->block_rsv;
2700 rc->reserved_bytes += num_bytes;
2701
2702 /*
2703 * We are under a transaction here so we can only do limited flushing.
2704 * If we get an enospc just kick back -EAGAIN so we know to drop the
2705 * transaction and try to refill when we can flush all the things.
2706 */
2707 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2708 BTRFS_RESERVE_FLUSH_LIMIT);
2709 if (ret) {
2710 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2711 while (tmp <= rc->reserved_bytes)
2712 tmp <<= 1;
2713 /*
2714 * only one thread can access block_rsv at this point,
2715 * so we don't need hold lock to protect block_rsv.
2716 * we expand more reservation size here to allow enough
2717 * space for relocation and we will return earlier in
2718 * enospc case.
2719 */
2720 rc->block_rsv->size = tmp + fs_info->nodesize *
2721 RELOCATION_RESERVED_NODES;
2722 return -EAGAIN;
2723 }
2724
2725 return 0;
2726}
2727
2728/*
2729 * relocate a block tree, and then update pointers in upper level
2730 * blocks that reference the block to point to the new location.
2731 *
2732 * if called by link_to_upper, the block has already been relocated.
2733 * in that case this function just updates pointers.
2734 */
2735static int do_relocation(struct btrfs_trans_handle *trans,
2736 struct reloc_control *rc,
2737 struct backref_node *node,
2738 struct btrfs_key *key,
2739 struct btrfs_path *path, int lowest)
2740{
2741 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2742 struct backref_node *upper;
2743 struct backref_edge *edge;
2744 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2745 struct btrfs_root *root;
2746 struct extent_buffer *eb;
2747 u32 blocksize;
2748 u64 bytenr;
2749 u64 generation;
2750 int slot;
2751 int ret;
2752 int err = 0;
2753
2754 BUG_ON(lowest && node->eb);
2755
2756 path->lowest_level = node->level + 1;
2757 rc->backref_cache.path[node->level] = node;
2758 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2759 struct btrfs_key first_key;
2760 struct btrfs_ref ref = { 0 };
2761
2762 cond_resched();
2763
2764 upper = edge->node[UPPER];
2765 root = select_reloc_root(trans, rc, upper, edges);
2766 BUG_ON(!root);
2767
2768 if (upper->eb && !upper->locked) {
2769 if (!lowest) {
2770 ret = btrfs_bin_search(upper->eb, key,
2771 upper->level, &slot);
2772 if (ret < 0) {
2773 err = ret;
2774 goto next;
2775 }
2776 BUG_ON(ret);
2777 bytenr = btrfs_node_blockptr(upper->eb, slot);
2778 if (node->eb->start == bytenr)
2779 goto next;
2780 }
2781 drop_node_buffer(upper);
2782 }
2783
2784 if (!upper->eb) {
2785 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2786 if (ret) {
2787 if (ret < 0)
2788 err = ret;
2789 else
2790 err = -ENOENT;
2791
2792 btrfs_release_path(path);
2793 break;
2794 }
2795
2796 if (!upper->eb) {
2797 upper->eb = path->nodes[upper->level];
2798 path->nodes[upper->level] = NULL;
2799 } else {
2800 BUG_ON(upper->eb != path->nodes[upper->level]);
2801 }
2802
2803 upper->locked = 1;
2804 path->locks[upper->level] = 0;
2805
2806 slot = path->slots[upper->level];
2807 btrfs_release_path(path);
2808 } else {
2809 ret = btrfs_bin_search(upper->eb, key, upper->level,
2810 &slot);
2811 if (ret < 0) {
2812 err = ret;
2813 goto next;
2814 }
2815 BUG_ON(ret);
2816 }
2817
2818 bytenr = btrfs_node_blockptr(upper->eb, slot);
2819 if (lowest) {
2820 if (bytenr != node->bytenr) {
2821 btrfs_err(root->fs_info,
2822 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2823 bytenr, node->bytenr, slot,
2824 upper->eb->start);
2825 err = -EIO;
2826 goto next;
2827 }
2828 } else {
2829 if (node->eb->start == bytenr)
2830 goto next;
2831 }
2832
2833 blocksize = root->fs_info->nodesize;
2834 generation = btrfs_node_ptr_generation(upper->eb, slot);
2835 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2836 eb = read_tree_block(fs_info, bytenr, generation,
2837 upper->level - 1, &first_key);
2838 if (IS_ERR(eb)) {
2839 err = PTR_ERR(eb);
2840 goto next;
2841 } else if (!extent_buffer_uptodate(eb)) {
2842 free_extent_buffer(eb);
2843 err = -EIO;
2844 goto next;
2845 }
2846 btrfs_tree_lock(eb);
2847 btrfs_set_lock_blocking_write(eb);
2848
2849 if (!node->eb) {
2850 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2851 slot, &eb);
2852 btrfs_tree_unlock(eb);
2853 free_extent_buffer(eb);
2854 if (ret < 0) {
2855 err = ret;
2856 goto next;
2857 }
2858 BUG_ON(node->eb != eb);
2859 } else {
2860 btrfs_set_node_blockptr(upper->eb, slot,
2861 node->eb->start);
2862 btrfs_set_node_ptr_generation(upper->eb, slot,
2863 trans->transid);
2864 btrfs_mark_buffer_dirty(upper->eb);
2865
2866 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2867 node->eb->start, blocksize,
2868 upper->eb->start);
2869 ref.real_root = root->root_key.objectid;
2870 btrfs_init_tree_ref(&ref, node->level,
2871 btrfs_header_owner(upper->eb));
2872 ret = btrfs_inc_extent_ref(trans, &ref);
2873 BUG_ON(ret);
2874
2875 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2876 BUG_ON(ret);
2877 }
2878next:
2879 if (!upper->pending)
2880 drop_node_buffer(upper);
2881 else
2882 unlock_node_buffer(upper);
2883 if (err)
2884 break;
2885 }
2886
2887 if (!err && node->pending) {
2888 drop_node_buffer(node);
2889 list_move_tail(&node->list, &rc->backref_cache.changed);
2890 node->pending = 0;
2891 }
2892
2893 path->lowest_level = 0;
2894 BUG_ON(err == -ENOSPC);
2895 return err;
2896}
2897
2898static int link_to_upper(struct btrfs_trans_handle *trans,
2899 struct reloc_control *rc,
2900 struct backref_node *node,
2901 struct btrfs_path *path)
2902{
2903 struct btrfs_key key;
2904
2905 btrfs_node_key_to_cpu(node->eb, &key, 0);
2906 return do_relocation(trans, rc, node, &key, path, 0);
2907}
2908
2909static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2910 struct reloc_control *rc,
2911 struct btrfs_path *path, int err)
2912{
2913 LIST_HEAD(list);
2914 struct backref_cache *cache = &rc->backref_cache;
2915 struct backref_node *node;
2916 int level;
2917 int ret;
2918
2919 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2920 while (!list_empty(&cache->pending[level])) {
2921 node = list_entry(cache->pending[level].next,
2922 struct backref_node, list);
2923 list_move_tail(&node->list, &list);
2924 BUG_ON(!node->pending);
2925
2926 if (!err) {
2927 ret = link_to_upper(trans, rc, node, path);
2928 if (ret < 0)
2929 err = ret;
2930 }
2931 }
2932 list_splice_init(&list, &cache->pending[level]);
2933 }
2934 return err;
2935}
2936
2937static void mark_block_processed(struct reloc_control *rc,
2938 u64 bytenr, u32 blocksize)
2939{
2940 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2941 EXTENT_DIRTY);
2942}
2943
2944static void __mark_block_processed(struct reloc_control *rc,
2945 struct backref_node *node)
2946{
2947 u32 blocksize;
2948 if (node->level == 0 ||
2949 in_block_group(node->bytenr, rc->block_group)) {
2950 blocksize = rc->extent_root->fs_info->nodesize;
2951 mark_block_processed(rc, node->bytenr, blocksize);
2952 }
2953 node->processed = 1;
2954}
2955
2956/*
2957 * mark a block and all blocks directly/indirectly reference the block
2958 * as processed.
2959 */
2960static void update_processed_blocks(struct reloc_control *rc,
2961 struct backref_node *node)
2962{
2963 struct backref_node *next = node;
2964 struct backref_edge *edge;
2965 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2966 int index = 0;
2967
2968 while (next) {
2969 cond_resched();
2970 while (1) {
2971 if (next->processed)
2972 break;
2973
2974 __mark_block_processed(rc, next);
2975
2976 if (list_empty(&next->upper))
2977 break;
2978
2979 edge = list_entry(next->upper.next,
2980 struct backref_edge, list[LOWER]);
2981 edges[index++] = edge;
2982 next = edge->node[UPPER];
2983 }
2984 next = walk_down_backref(edges, &index);
2985 }
2986}
2987
2988static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2989{
2990 u32 blocksize = rc->extent_root->fs_info->nodesize;
2991
2992 if (test_range_bit(&rc->processed_blocks, bytenr,
2993 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2994 return 1;
2995 return 0;
2996}
2997
2998static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2999 struct tree_block *block)
3000{
3001 struct extent_buffer *eb;
3002
3003 BUG_ON(block->key_ready);
3004 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
3005 block->level, NULL);
3006 if (IS_ERR(eb)) {
3007 return PTR_ERR(eb);
3008 } else if (!extent_buffer_uptodate(eb)) {
3009 free_extent_buffer(eb);
3010 return -EIO;
3011 }
3012 if (block->level == 0)
3013 btrfs_item_key_to_cpu(eb, &block->key, 0);
3014 else
3015 btrfs_node_key_to_cpu(eb, &block->key, 0);
3016 free_extent_buffer(eb);
3017 block->key_ready = 1;
3018 return 0;
3019}
3020
3021/*
3022 * helper function to relocate a tree block
3023 */
3024static int relocate_tree_block(struct btrfs_trans_handle *trans,
3025 struct reloc_control *rc,
3026 struct backref_node *node,
3027 struct btrfs_key *key,
3028 struct btrfs_path *path)
3029{
3030 struct btrfs_root *root;
3031 int ret = 0;
3032
3033 if (!node)
3034 return 0;
3035
3036 BUG_ON(node->processed);
3037 root = select_one_root(node);
3038 if (root == ERR_PTR(-ENOENT)) {
3039 update_processed_blocks(rc, node);
3040 goto out;
3041 }
3042
3043 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3044 ret = reserve_metadata_space(trans, rc, node);
3045 if (ret)
3046 goto out;
3047 }
3048
3049 if (root) {
3050 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3051 BUG_ON(node->new_bytenr);
3052 BUG_ON(!list_empty(&node->list));
3053 btrfs_record_root_in_trans(trans, root);
3054 root = root->reloc_root;
3055 node->new_bytenr = root->node->start;
3056 node->root = root;
3057 list_add_tail(&node->list, &rc->backref_cache.changed);
3058 } else {
3059 path->lowest_level = node->level;
3060 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3061 btrfs_release_path(path);
3062 if (ret > 0)
3063 ret = 0;
3064 }
3065 if (!ret)
3066 update_processed_blocks(rc, node);
3067 } else {
3068 ret = do_relocation(trans, rc, node, key, path, 1);
3069 }
3070out:
3071 if (ret || node->level == 0 || node->cowonly)
3072 remove_backref_node(&rc->backref_cache, node);
3073 return ret;
3074}
3075
3076/*
3077 * relocate a list of blocks
3078 */
3079static noinline_for_stack
3080int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3081 struct reloc_control *rc, struct rb_root *blocks)
3082{
3083 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3084 struct backref_node *node;
3085 struct btrfs_path *path;
3086 struct tree_block *block;
3087 struct tree_block *next;
3088 int ret;
3089 int err = 0;
3090
3091 path = btrfs_alloc_path();
3092 if (!path) {
3093 err = -ENOMEM;
3094 goto out_free_blocks;
3095 }
3096
3097 /* Kick in readahead for tree blocks with missing keys */
3098 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3099 if (!block->key_ready)
3100 readahead_tree_block(fs_info, block->bytenr);
3101 }
3102
3103 /* Get first keys */
3104 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3105 if (!block->key_ready) {
3106 err = get_tree_block_key(fs_info, block);
3107 if (err)
3108 goto out_free_path;
3109 }
3110 }
3111
3112 /* Do tree relocation */
3113 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3114 node = build_backref_tree(rc, &block->key,
3115 block->level, block->bytenr);
3116 if (IS_ERR(node)) {
3117 err = PTR_ERR(node);
3118 goto out;
3119 }
3120
3121 ret = relocate_tree_block(trans, rc, node, &block->key,
3122 path);
3123 if (ret < 0) {
3124 if (ret != -EAGAIN || &block->rb_node == rb_first(blocks))
3125 err = ret;
3126 goto out;
3127 }
3128 }
3129out:
3130 err = finish_pending_nodes(trans, rc, path, err);
3131
3132out_free_path:
3133 btrfs_free_path(path);
3134out_free_blocks:
3135 free_block_list(blocks);
3136 return err;
3137}
3138
3139static noinline_for_stack
3140int prealloc_file_extent_cluster(struct inode *inode,
3141 struct file_extent_cluster *cluster)
3142{
3143 u64 alloc_hint = 0;
3144 u64 start;
3145 u64 end;
3146 u64 offset = BTRFS_I(inode)->index_cnt;
3147 u64 num_bytes;
3148 int nr = 0;
3149 int ret = 0;
3150 u64 prealloc_start = cluster->start - offset;
3151 u64 prealloc_end = cluster->end - offset;
3152 u64 cur_offset;
3153 struct extent_changeset *data_reserved = NULL;
3154
3155 BUG_ON(cluster->start != cluster->boundary[0]);
3156 inode_lock(inode);
3157
3158 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3159 prealloc_end + 1 - prealloc_start);
3160 if (ret)
3161 goto out;
3162
3163 cur_offset = prealloc_start;
3164 while (nr < cluster->nr) {
3165 start = cluster->boundary[nr] - offset;
3166 if (nr + 1 < cluster->nr)
3167 end = cluster->boundary[nr + 1] - 1 - offset;
3168 else
3169 end = cluster->end - offset;
3170
3171 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3172 num_bytes = end + 1 - start;
3173 if (cur_offset < start)
3174 btrfs_free_reserved_data_space(inode, data_reserved,
3175 cur_offset, start - cur_offset);
3176 ret = btrfs_prealloc_file_range(inode, 0, start,
3177 num_bytes, num_bytes,
3178 end + 1, &alloc_hint);
3179 cur_offset = end + 1;
3180 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3181 if (ret)
3182 break;
3183 nr++;
3184 }
3185 if (cur_offset < prealloc_end)
3186 btrfs_free_reserved_data_space(inode, data_reserved,
3187 cur_offset, prealloc_end + 1 - cur_offset);
3188out:
3189 inode_unlock(inode);
3190 extent_changeset_free(data_reserved);
3191 return ret;
3192}
3193
3194static noinline_for_stack
3195int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3196 u64 block_start)
3197{
3198 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3199 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3200 struct extent_map *em;
3201 int ret = 0;
3202
3203 em = alloc_extent_map();
3204 if (!em)
3205 return -ENOMEM;
3206
3207 em->start = start;
3208 em->len = end + 1 - start;
3209 em->block_len = em->len;
3210 em->block_start = block_start;
3211 em->bdev = fs_info->fs_devices->latest_bdev;
3212 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3213
3214 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3215 while (1) {
3216 write_lock(&em_tree->lock);
3217 ret = add_extent_mapping(em_tree, em, 0);
3218 write_unlock(&em_tree->lock);
3219 if (ret != -EEXIST) {
3220 free_extent_map(em);
3221 break;
3222 }
3223 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3224 }
3225 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3226 return ret;
3227}
3228
3229static int relocate_file_extent_cluster(struct inode *inode,
3230 struct file_extent_cluster *cluster)
3231{
3232 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3233 u64 page_start;
3234 u64 page_end;
3235 u64 offset = BTRFS_I(inode)->index_cnt;
3236 unsigned long index;
3237 unsigned long last_index;
3238 struct page *page;
3239 struct file_ra_state *ra;
3240 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3241 int nr = 0;
3242 int ret = 0;
3243
3244 if (!cluster->nr)
3245 return 0;
3246
3247 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3248 if (!ra)
3249 return -ENOMEM;
3250
3251 ret = prealloc_file_extent_cluster(inode, cluster);
3252 if (ret)
3253 goto out;
3254
3255 file_ra_state_init(ra, inode->i_mapping);
3256
3257 ret = setup_extent_mapping(inode, cluster->start - offset,
3258 cluster->end - offset, cluster->start);
3259 if (ret)
3260 goto out;
3261
3262 index = (cluster->start - offset) >> PAGE_SHIFT;
3263 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3264 while (index <= last_index) {
3265 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3266 PAGE_SIZE);
3267 if (ret)
3268 goto out;
3269
3270 page = find_lock_page(inode->i_mapping, index);
3271 if (!page) {
3272 page_cache_sync_readahead(inode->i_mapping,
3273 ra, NULL, index,
3274 last_index + 1 - index);
3275 page = find_or_create_page(inode->i_mapping, index,
3276 mask);
3277 if (!page) {
3278 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3279 PAGE_SIZE, true);
3280 btrfs_delalloc_release_extents(BTRFS_I(inode),
3281 PAGE_SIZE);
3282 ret = -ENOMEM;
3283 goto out;
3284 }
3285 }
3286
3287 if (PageReadahead(page)) {
3288 page_cache_async_readahead(inode->i_mapping,
3289 ra, NULL, page, index,
3290 last_index + 1 - index);
3291 }
3292
3293 if (!PageUptodate(page)) {
3294 btrfs_readpage(NULL, page);
3295 lock_page(page);
3296 if (!PageUptodate(page)) {
3297 unlock_page(page);
3298 put_page(page);
3299 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3300 PAGE_SIZE, true);
3301 btrfs_delalloc_release_extents(BTRFS_I(inode),
3302 PAGE_SIZE);
3303 ret = -EIO;
3304 goto out;
3305 }
3306 }
3307
3308 page_start = page_offset(page);
3309 page_end = page_start + PAGE_SIZE - 1;
3310
3311 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3312
3313 set_page_extent_mapped(page);
3314
3315 if (nr < cluster->nr &&
3316 page_start + offset == cluster->boundary[nr]) {
3317 set_extent_bits(&BTRFS_I(inode)->io_tree,
3318 page_start, page_end,
3319 EXTENT_BOUNDARY);
3320 nr++;
3321 }
3322
3323 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3324 NULL);
3325 if (ret) {
3326 unlock_page(page);
3327 put_page(page);
3328 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3329 PAGE_SIZE, true);
3330 btrfs_delalloc_release_extents(BTRFS_I(inode),
3331 PAGE_SIZE);
3332
3333 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3334 page_start, page_end,
3335 EXTENT_LOCKED | EXTENT_BOUNDARY);
3336 goto out;
3337
3338 }
3339 set_page_dirty(page);
3340
3341 unlock_extent(&BTRFS_I(inode)->io_tree,
3342 page_start, page_end);
3343 unlock_page(page);
3344 put_page(page);
3345
3346 index++;
3347 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3348 balance_dirty_pages_ratelimited(inode->i_mapping);
3349 btrfs_throttle(fs_info);
3350 }
3351 WARN_ON(nr != cluster->nr);
3352out:
3353 kfree(ra);
3354 return ret;
3355}
3356
3357static noinline_for_stack
3358int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3359 struct file_extent_cluster *cluster)
3360{
3361 int ret;
3362
3363 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3364 ret = relocate_file_extent_cluster(inode, cluster);
3365 if (ret)
3366 return ret;
3367 cluster->nr = 0;
3368 }
3369
3370 if (!cluster->nr)
3371 cluster->start = extent_key->objectid;
3372 else
3373 BUG_ON(cluster->nr >= MAX_EXTENTS);
3374 cluster->end = extent_key->objectid + extent_key->offset - 1;
3375 cluster->boundary[cluster->nr] = extent_key->objectid;
3376 cluster->nr++;
3377
3378 if (cluster->nr >= MAX_EXTENTS) {
3379 ret = relocate_file_extent_cluster(inode, cluster);
3380 if (ret)
3381 return ret;
3382 cluster->nr = 0;
3383 }
3384 return 0;
3385}
3386
3387/*
3388 * helper to add a tree block to the list.
3389 * the major work is getting the generation and level of the block
3390 */
3391static int add_tree_block(struct reloc_control *rc,
3392 struct btrfs_key *extent_key,
3393 struct btrfs_path *path,
3394 struct rb_root *blocks)
3395{
3396 struct extent_buffer *eb;
3397 struct btrfs_extent_item *ei;
3398 struct btrfs_tree_block_info *bi;
3399 struct tree_block *block;
3400 struct rb_node *rb_node;
3401 u32 item_size;
3402 int level = -1;
3403 u64 generation;
3404
3405 eb = path->nodes[0];
3406 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3407
3408 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3409 item_size >= sizeof(*ei) + sizeof(*bi)) {
3410 ei = btrfs_item_ptr(eb, path->slots[0],
3411 struct btrfs_extent_item);
3412 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3413 bi = (struct btrfs_tree_block_info *)(ei + 1);
3414 level = btrfs_tree_block_level(eb, bi);
3415 } else {
3416 level = (int)extent_key->offset;
3417 }
3418 generation = btrfs_extent_generation(eb, ei);
3419 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3420 btrfs_print_v0_err(eb->fs_info);
3421 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3422 return -EINVAL;
3423 } else {
3424 BUG();
3425 }
3426
3427 btrfs_release_path(path);
3428
3429 BUG_ON(level == -1);
3430
3431 block = kmalloc(sizeof(*block), GFP_NOFS);
3432 if (!block)
3433 return -ENOMEM;
3434
3435 block->bytenr = extent_key->objectid;
3436 block->key.objectid = rc->extent_root->fs_info->nodesize;
3437 block->key.offset = generation;
3438 block->level = level;
3439 block->key_ready = 0;
3440
3441 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3442 if (rb_node)
3443 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3444
3445 return 0;
3446}
3447
3448/*
3449 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3450 */
3451static int __add_tree_block(struct reloc_control *rc,
3452 u64 bytenr, u32 blocksize,
3453 struct rb_root *blocks)
3454{
3455 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3456 struct btrfs_path *path;
3457 struct btrfs_key key;
3458 int ret;
3459 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3460
3461 if (tree_block_processed(bytenr, rc))
3462 return 0;
3463
3464 if (tree_search(blocks, bytenr))
3465 return 0;
3466
3467 path = btrfs_alloc_path();
3468 if (!path)
3469 return -ENOMEM;
3470again:
3471 key.objectid = bytenr;
3472 if (skinny) {
3473 key.type = BTRFS_METADATA_ITEM_KEY;
3474 key.offset = (u64)-1;
3475 } else {
3476 key.type = BTRFS_EXTENT_ITEM_KEY;
3477 key.offset = blocksize;
3478 }
3479
3480 path->search_commit_root = 1;
3481 path->skip_locking = 1;
3482 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3483 if (ret < 0)
3484 goto out;
3485
3486 if (ret > 0 && skinny) {
3487 if (path->slots[0]) {
3488 path->slots[0]--;
3489 btrfs_item_key_to_cpu(path->nodes[0], &key,
3490 path->slots[0]);
3491 if (key.objectid == bytenr &&
3492 (key.type == BTRFS_METADATA_ITEM_KEY ||
3493 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3494 key.offset == blocksize)))
3495 ret = 0;
3496 }
3497
3498 if (ret) {
3499 skinny = false;
3500 btrfs_release_path(path);
3501 goto again;
3502 }
3503 }
3504 if (ret) {
3505 ASSERT(ret == 1);
3506 btrfs_print_leaf(path->nodes[0]);
3507 btrfs_err(fs_info,
3508 "tree block extent item (%llu) is not found in extent tree",
3509 bytenr);
3510 WARN_ON(1);
3511 ret = -EINVAL;
3512 goto out;
3513 }
3514
3515 ret = add_tree_block(rc, &key, path, blocks);
3516out:
3517 btrfs_free_path(path);
3518 return ret;
3519}
3520
3521/*
3522 * helper to check if the block use full backrefs for pointers in it
3523 */
3524static int block_use_full_backref(struct reloc_control *rc,
3525 struct extent_buffer *eb)
3526{
3527 u64 flags;
3528 int ret;
3529
3530 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3531 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3532 return 1;
3533
3534 ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3535 eb->start, btrfs_header_level(eb), 1,
3536 NULL, &flags);
3537 BUG_ON(ret);
3538
3539 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3540 ret = 1;
3541 else
3542 ret = 0;
3543 return ret;
3544}
3545
3546static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3547 struct btrfs_block_group_cache *block_group,
3548 struct inode *inode,
3549 u64 ino)
3550{
3551 struct btrfs_key key;
3552 struct btrfs_root *root = fs_info->tree_root;
3553 struct btrfs_trans_handle *trans;
3554 int ret = 0;
3555
3556 if (inode)
3557 goto truncate;
3558
3559 key.objectid = ino;
3560 key.type = BTRFS_INODE_ITEM_KEY;
3561 key.offset = 0;
3562
3563 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3564 if (IS_ERR(inode))
3565 return -ENOENT;
3566
3567truncate:
3568 ret = btrfs_check_trunc_cache_free_space(fs_info,
3569 &fs_info->global_block_rsv);
3570 if (ret)
3571 goto out;
3572
3573 trans = btrfs_join_transaction(root);
3574 if (IS_ERR(trans)) {
3575 ret = PTR_ERR(trans);
3576 goto out;
3577 }
3578
3579 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3580
3581 btrfs_end_transaction(trans);
3582 btrfs_btree_balance_dirty(fs_info);
3583out:
3584 iput(inode);
3585 return ret;
3586}
3587
3588/*
3589 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3590 * this function scans fs tree to find blocks reference the data extent
3591 */
3592static int find_data_references(struct reloc_control *rc,
3593 struct btrfs_key *extent_key,
3594 struct extent_buffer *leaf,
3595 struct btrfs_extent_data_ref *ref,
3596 struct rb_root *blocks)
3597{
3598 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3599 struct btrfs_path *path;
3600 struct tree_block *block;
3601 struct btrfs_root *root;
3602 struct btrfs_file_extent_item *fi;
3603 struct rb_node *rb_node;
3604 struct btrfs_key key;
3605 u64 ref_root;
3606 u64 ref_objectid;
3607 u64 ref_offset;
3608 u32 ref_count;
3609 u32 nritems;
3610 int err = 0;
3611 int added = 0;
3612 int counted;
3613 int ret;
3614
3615 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3616 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3617 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3618 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3619
3620 /*
3621 * This is an extent belonging to the free space cache, lets just delete
3622 * it and redo the search.
3623 */
3624 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3625 ret = delete_block_group_cache(fs_info, rc->block_group,
3626 NULL, ref_objectid);
3627 if (ret != -ENOENT)
3628 return ret;
3629 ret = 0;
3630 }
3631
3632 path = btrfs_alloc_path();
3633 if (!path)
3634 return -ENOMEM;
3635 path->reada = READA_FORWARD;
3636
3637 root = read_fs_root(fs_info, ref_root);
3638 if (IS_ERR(root)) {
3639 err = PTR_ERR(root);
3640 goto out;
3641 }
3642
3643 key.objectid = ref_objectid;
3644 key.type = BTRFS_EXTENT_DATA_KEY;
3645 if (ref_offset > ((u64)-1 << 32))
3646 key.offset = 0;
3647 else
3648 key.offset = ref_offset;
3649
3650 path->search_commit_root = 1;
3651 path->skip_locking = 1;
3652 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3653 if (ret < 0) {
3654 err = ret;
3655 goto out;
3656 }
3657
3658 leaf = path->nodes[0];
3659 nritems = btrfs_header_nritems(leaf);
3660 /*
3661 * the references in tree blocks that use full backrefs
3662 * are not counted in
3663 */
3664 if (block_use_full_backref(rc, leaf))
3665 counted = 0;
3666 else
3667 counted = 1;
3668 rb_node = tree_search(blocks, leaf->start);
3669 if (rb_node) {
3670 if (counted)
3671 added = 1;
3672 else
3673 path->slots[0] = nritems;
3674 }
3675
3676 while (ref_count > 0) {
3677 while (path->slots[0] >= nritems) {
3678 ret = btrfs_next_leaf(root, path);
3679 if (ret < 0) {
3680 err = ret;
3681 goto out;
3682 }
3683 if (WARN_ON(ret > 0))
3684 goto out;
3685
3686 leaf = path->nodes[0];
3687 nritems = btrfs_header_nritems(leaf);
3688 added = 0;
3689
3690 if (block_use_full_backref(rc, leaf))
3691 counted = 0;
3692 else
3693 counted = 1;
3694 rb_node = tree_search(blocks, leaf->start);
3695 if (rb_node) {
3696 if (counted)
3697 added = 1;
3698 else
3699 path->slots[0] = nritems;
3700 }
3701 }
3702
3703 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3704 if (WARN_ON(key.objectid != ref_objectid ||
3705 key.type != BTRFS_EXTENT_DATA_KEY))
3706 break;
3707
3708 fi = btrfs_item_ptr(leaf, path->slots[0],
3709 struct btrfs_file_extent_item);
3710
3711 if (btrfs_file_extent_type(leaf, fi) ==
3712 BTRFS_FILE_EXTENT_INLINE)
3713 goto next;
3714
3715 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3716 extent_key->objectid)
3717 goto next;
3718
3719 key.offset -= btrfs_file_extent_offset(leaf, fi);
3720 if (key.offset != ref_offset)
3721 goto next;
3722
3723 if (counted)
3724 ref_count--;
3725 if (added)
3726 goto next;
3727
3728 if (!tree_block_processed(leaf->start, rc)) {
3729 block = kmalloc(sizeof(*block), GFP_NOFS);
3730 if (!block) {
3731 err = -ENOMEM;
3732 break;
3733 }
3734 block->bytenr = leaf->start;
3735 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3736 block->level = 0;
3737 block->key_ready = 1;
3738 rb_node = tree_insert(blocks, block->bytenr,
3739 &block->rb_node);
3740 if (rb_node)
3741 backref_tree_panic(rb_node, -EEXIST,
3742 block->bytenr);
3743 }
3744 if (counted)
3745 added = 1;
3746 else
3747 path->slots[0] = nritems;
3748next:
3749 path->slots[0]++;
3750
3751 }
3752out:
3753 btrfs_free_path(path);
3754 return err;
3755}
3756
3757/*
3758 * helper to find all tree blocks that reference a given data extent
3759 */
3760static noinline_for_stack
3761int add_data_references(struct reloc_control *rc,
3762 struct btrfs_key *extent_key,
3763 struct btrfs_path *path,
3764 struct rb_root *blocks)
3765{
3766 struct btrfs_key key;
3767 struct extent_buffer *eb;
3768 struct btrfs_extent_data_ref *dref;
3769 struct btrfs_extent_inline_ref *iref;
3770 unsigned long ptr;
3771 unsigned long end;
3772 u32 blocksize = rc->extent_root->fs_info->nodesize;
3773 int ret = 0;
3774 int err = 0;
3775
3776 eb = path->nodes[0];
3777 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3778 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3779 ptr += sizeof(struct btrfs_extent_item);
3780
3781 while (ptr < end) {
3782 iref = (struct btrfs_extent_inline_ref *)ptr;
3783 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3784 BTRFS_REF_TYPE_DATA);
3785 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3786 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3787 ret = __add_tree_block(rc, key.offset, blocksize,
3788 blocks);
3789 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3790 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3791 ret = find_data_references(rc, extent_key,
3792 eb, dref, blocks);
3793 } else {
3794 ret = -EUCLEAN;
3795 btrfs_err(rc->extent_root->fs_info,
3796 "extent %llu slot %d has an invalid inline ref type",
3797 eb->start, path->slots[0]);
3798 }
3799 if (ret) {
3800 err = ret;
3801 goto out;
3802 }
3803 ptr += btrfs_extent_inline_ref_size(key.type);
3804 }
3805 WARN_ON(ptr > end);
3806
3807 while (1) {
3808 cond_resched();
3809 eb = path->nodes[0];
3810 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3811 ret = btrfs_next_leaf(rc->extent_root, path);
3812 if (ret < 0) {
3813 err = ret;
3814 break;
3815 }
3816 if (ret > 0)
3817 break;
3818 eb = path->nodes[0];
3819 }
3820
3821 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3822 if (key.objectid != extent_key->objectid)
3823 break;
3824
3825 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3826 ret = __add_tree_block(rc, key.offset, blocksize,
3827 blocks);
3828 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3829 dref = btrfs_item_ptr(eb, path->slots[0],
3830 struct btrfs_extent_data_ref);
3831 ret = find_data_references(rc, extent_key,
3832 eb, dref, blocks);
3833 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3834 btrfs_print_v0_err(eb->fs_info);
3835 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3836 ret = -EINVAL;
3837 } else {
3838 ret = 0;
3839 }
3840 if (ret) {
3841 err = ret;
3842 break;
3843 }
3844 path->slots[0]++;
3845 }
3846out:
3847 btrfs_release_path(path);
3848 if (err)
3849 free_block_list(blocks);
3850 return err;
3851}
3852
3853/*
3854 * helper to find next unprocessed extent
3855 */
3856static noinline_for_stack
3857int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3858 struct btrfs_key *extent_key)
3859{
3860 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3861 struct btrfs_key key;
3862 struct extent_buffer *leaf;
3863 u64 start, end, last;
3864 int ret;
3865
3866 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3867 while (1) {
3868 cond_resched();
3869 if (rc->search_start >= last) {
3870 ret = 1;
3871 break;
3872 }
3873
3874 key.objectid = rc->search_start;
3875 key.type = BTRFS_EXTENT_ITEM_KEY;
3876 key.offset = 0;
3877
3878 path->search_commit_root = 1;
3879 path->skip_locking = 1;
3880 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3881 0, 0);
3882 if (ret < 0)
3883 break;
3884next:
3885 leaf = path->nodes[0];
3886 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3887 ret = btrfs_next_leaf(rc->extent_root, path);
3888 if (ret != 0)
3889 break;
3890 leaf = path->nodes[0];
3891 }
3892
3893 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3894 if (key.objectid >= last) {
3895 ret = 1;
3896 break;
3897 }
3898
3899 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3900 key.type != BTRFS_METADATA_ITEM_KEY) {
3901 path->slots[0]++;
3902 goto next;
3903 }
3904
3905 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3906 key.objectid + key.offset <= rc->search_start) {
3907 path->slots[0]++;
3908 goto next;
3909 }
3910
3911 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3912 key.objectid + fs_info->nodesize <=
3913 rc->search_start) {
3914 path->slots[0]++;
3915 goto next;
3916 }
3917
3918 ret = find_first_extent_bit(&rc->processed_blocks,
3919 key.objectid, &start, &end,
3920 EXTENT_DIRTY, NULL);
3921
3922 if (ret == 0 && start <= key.objectid) {
3923 btrfs_release_path(path);
3924 rc->search_start = end + 1;
3925 } else {
3926 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3927 rc->search_start = key.objectid + key.offset;
3928 else
3929 rc->search_start = key.objectid +
3930 fs_info->nodesize;
3931 memcpy(extent_key, &key, sizeof(key));
3932 return 0;
3933 }
3934 }
3935 btrfs_release_path(path);
3936 return ret;
3937}
3938
3939static void set_reloc_control(struct reloc_control *rc)
3940{
3941 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3942
3943 mutex_lock(&fs_info->reloc_mutex);
3944 fs_info->reloc_ctl = rc;
3945 mutex_unlock(&fs_info->reloc_mutex);
3946}
3947
3948static void unset_reloc_control(struct reloc_control *rc)
3949{
3950 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3951
3952 mutex_lock(&fs_info->reloc_mutex);
3953 fs_info->reloc_ctl = NULL;
3954 mutex_unlock(&fs_info->reloc_mutex);
3955}
3956
3957static int check_extent_flags(u64 flags)
3958{
3959 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3960 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3961 return 1;
3962 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3963 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3964 return 1;
3965 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3966 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3967 return 1;
3968 return 0;
3969}
3970
3971static noinline_for_stack
3972int prepare_to_relocate(struct reloc_control *rc)
3973{
3974 struct btrfs_trans_handle *trans;
3975 int ret;
3976
3977 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3978 BTRFS_BLOCK_RSV_TEMP);
3979 if (!rc->block_rsv)
3980 return -ENOMEM;
3981
3982 memset(&rc->cluster, 0, sizeof(rc->cluster));
3983 rc->search_start = rc->block_group->key.objectid;
3984 rc->extents_found = 0;
3985 rc->nodes_relocated = 0;
3986 rc->merging_rsv_size = 0;
3987 rc->reserved_bytes = 0;
3988 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3989 RELOCATION_RESERVED_NODES;
3990 ret = btrfs_block_rsv_refill(rc->extent_root,
3991 rc->block_rsv, rc->block_rsv->size,
3992 BTRFS_RESERVE_FLUSH_ALL);
3993 if (ret)
3994 return ret;
3995
3996 rc->create_reloc_tree = 1;
3997 set_reloc_control(rc);
3998
3999 trans = btrfs_join_transaction(rc->extent_root);
4000 if (IS_ERR(trans)) {
4001 unset_reloc_control(rc);
4002 /*
4003 * extent tree is not a ref_cow tree and has no reloc_root to
4004 * cleanup. And callers are responsible to free the above
4005 * block rsv.
4006 */
4007 return PTR_ERR(trans);
4008 }
4009 btrfs_commit_transaction(trans);
4010 return 0;
4011}
4012
4013static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
4014{
4015 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
4016 struct rb_root blocks = RB_ROOT;
4017 struct btrfs_key key;
4018 struct btrfs_trans_handle *trans = NULL;
4019 struct btrfs_path *path;
4020 struct btrfs_extent_item *ei;
4021 u64 flags;
4022 u32 item_size;
4023 int ret;
4024 int err = 0;
4025 int progress = 0;
4026
4027 path = btrfs_alloc_path();
4028 if (!path)
4029 return -ENOMEM;
4030 path->reada = READA_FORWARD;
4031
4032 ret = prepare_to_relocate(rc);
4033 if (ret) {
4034 err = ret;
4035 goto out_free;
4036 }
4037
4038 while (1) {
4039 rc->reserved_bytes = 0;
4040 ret = btrfs_block_rsv_refill(rc->extent_root,
4041 rc->block_rsv, rc->block_rsv->size,
4042 BTRFS_RESERVE_FLUSH_ALL);
4043 if (ret) {
4044 err = ret;
4045 break;
4046 }
4047 progress++;
4048 trans = btrfs_start_transaction(rc->extent_root, 0);
4049 if (IS_ERR(trans)) {
4050 err = PTR_ERR(trans);
4051 trans = NULL;
4052 break;
4053 }
4054restart:
4055 if (update_backref_cache(trans, &rc->backref_cache)) {
4056 btrfs_end_transaction(trans);
4057 trans = NULL;
4058 continue;
4059 }
4060
4061 ret = find_next_extent(rc, path, &key);
4062 if (ret < 0)
4063 err = ret;
4064 if (ret != 0)
4065 break;
4066
4067 rc->extents_found++;
4068
4069 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4070 struct btrfs_extent_item);
4071 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4072 if (item_size >= sizeof(*ei)) {
4073 flags = btrfs_extent_flags(path->nodes[0], ei);
4074 ret = check_extent_flags(flags);
4075 BUG_ON(ret);
4076 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
4077 err = -EINVAL;
4078 btrfs_print_v0_err(trans->fs_info);
4079 btrfs_abort_transaction(trans, err);
4080 break;
4081 } else {
4082 BUG();
4083 }
4084
4085 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4086 ret = add_tree_block(rc, &key, path, &blocks);
4087 } else if (rc->stage == UPDATE_DATA_PTRS &&
4088 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4089 ret = add_data_references(rc, &key, path, &blocks);
4090 } else {
4091 btrfs_release_path(path);
4092 ret = 0;
4093 }
4094 if (ret < 0) {
4095 err = ret;
4096 break;
4097 }
4098
4099 if (!RB_EMPTY_ROOT(&blocks)) {
4100 ret = relocate_tree_blocks(trans, rc, &blocks);
4101 if (ret < 0) {
4102 /*
4103 * if we fail to relocate tree blocks, force to update
4104 * backref cache when committing transaction.
4105 */
4106 rc->backref_cache.last_trans = trans->transid - 1;
4107
4108 if (ret != -EAGAIN) {
4109 err = ret;
4110 break;
4111 }
4112 rc->extents_found--;
4113 rc->search_start = key.objectid;
4114 }
4115 }
4116
4117 btrfs_end_transaction_throttle(trans);
4118 btrfs_btree_balance_dirty(fs_info);
4119 trans = NULL;
4120
4121 if (rc->stage == MOVE_DATA_EXTENTS &&
4122 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4123 rc->found_file_extent = 1;
4124 ret = relocate_data_extent(rc->data_inode,
4125 &key, &rc->cluster);
4126 if (ret < 0) {
4127 err = ret;
4128 break;
4129 }
4130 }
4131 }
4132 if (trans && progress && err == -ENOSPC) {
4133 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4134 if (ret == 1) {
4135 err = 0;
4136 progress = 0;
4137 goto restart;
4138 }
4139 }
4140
4141 btrfs_release_path(path);
4142 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4143
4144 if (trans) {
4145 btrfs_end_transaction_throttle(trans);
4146 btrfs_btree_balance_dirty(fs_info);
4147 }
4148
4149 if (!err) {
4150 ret = relocate_file_extent_cluster(rc->data_inode,
4151 &rc->cluster);
4152 if (ret < 0)
4153 err = ret;
4154 }
4155
4156 rc->create_reloc_tree = 0;
4157 set_reloc_control(rc);
4158
4159 backref_cache_cleanup(&rc->backref_cache);
4160 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4161
4162 err = prepare_to_merge(rc, err);
4163
4164 merge_reloc_roots(rc);
4165
4166 rc->merge_reloc_tree = 0;
4167 unset_reloc_control(rc);
4168 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4169
4170 /* get rid of pinned extents */
4171 trans = btrfs_join_transaction(rc->extent_root);
4172 if (IS_ERR(trans)) {
4173 err = PTR_ERR(trans);
4174 goto out_free;
4175 }
4176 btrfs_commit_transaction(trans);
4177 ret = clean_dirty_subvols(rc);
4178 if (ret < 0 && !err)
4179 err = ret;
4180out_free:
4181 btrfs_free_block_rsv(fs_info, rc->block_rsv);
4182 btrfs_free_path(path);
4183 return err;
4184}
4185
4186static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4187 struct btrfs_root *root, u64 objectid)
4188{
4189 struct btrfs_path *path;
4190 struct btrfs_inode_item *item;
4191 struct extent_buffer *leaf;
4192 int ret;
4193
4194 path = btrfs_alloc_path();
4195 if (!path)
4196 return -ENOMEM;
4197
4198 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4199 if (ret)
4200 goto out;
4201
4202 leaf = path->nodes[0];
4203 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4204 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4205 btrfs_set_inode_generation(leaf, item, 1);
4206 btrfs_set_inode_size(leaf, item, 0);
4207 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4208 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4209 BTRFS_INODE_PREALLOC);
4210 btrfs_mark_buffer_dirty(leaf);
4211out:
4212 btrfs_free_path(path);
4213 return ret;
4214}
4215
4216/*
4217 * helper to create inode for data relocation.
4218 * the inode is in data relocation tree and its link count is 0
4219 */
4220static noinline_for_stack
4221struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4222 struct btrfs_block_group_cache *group)
4223{
4224 struct inode *inode = NULL;
4225 struct btrfs_trans_handle *trans;
4226 struct btrfs_root *root;
4227 struct btrfs_key key;
4228 u64 objectid;
4229 int err = 0;
4230
4231 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4232 if (IS_ERR(root))
4233 return ERR_CAST(root);
4234
4235 trans = btrfs_start_transaction(root, 6);
4236 if (IS_ERR(trans))
4237 return ERR_CAST(trans);
4238
4239 err = btrfs_find_free_objectid(root, &objectid);
4240 if (err)
4241 goto out;
4242
4243 err = __insert_orphan_inode(trans, root, objectid);
4244 BUG_ON(err);
4245
4246 key.objectid = objectid;
4247 key.type = BTRFS_INODE_ITEM_KEY;
4248 key.offset = 0;
4249 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4250 BUG_ON(IS_ERR(inode));
4251 BTRFS_I(inode)->index_cnt = group->key.objectid;
4252
4253 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4254out:
4255 btrfs_end_transaction(trans);
4256 btrfs_btree_balance_dirty(fs_info);
4257 if (err) {
4258 if (inode)
4259 iput(inode);
4260 inode = ERR_PTR(err);
4261 }
4262 return inode;
4263}
4264
4265static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4266{
4267 struct reloc_control *rc;
4268
4269 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4270 if (!rc)
4271 return NULL;
4272
4273 INIT_LIST_HEAD(&rc->reloc_roots);
4274 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
4275 backref_cache_init(&rc->backref_cache);
4276 mapping_tree_init(&rc->reloc_root_tree);
4277 extent_io_tree_init(fs_info, &rc->processed_blocks,
4278 IO_TREE_RELOC_BLOCKS, NULL);
4279 return rc;
4280}
4281
4282/*
4283 * Print the block group being relocated
4284 */
4285static void describe_relocation(struct btrfs_fs_info *fs_info,
4286 struct btrfs_block_group_cache *block_group)
4287{
4288 char buf[128] = {'\0'};
4289
4290 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
4291
4292 btrfs_info(fs_info,
4293 "relocating block group %llu flags %s",
4294 block_group->key.objectid, buf);
4295}
4296
4297/*
4298 * function to relocate all extents in a block group.
4299 */
4300int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4301{
4302 struct btrfs_block_group_cache *bg;
4303 struct btrfs_root *extent_root = fs_info->extent_root;
4304 struct reloc_control *rc;
4305 struct inode *inode;
4306 struct btrfs_path *path;
4307 int ret;
4308 int rw = 0;
4309 int err = 0;
4310
4311 bg = btrfs_lookup_block_group(fs_info, group_start);
4312 if (!bg)
4313 return -ENOENT;
4314
4315 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4316 btrfs_put_block_group(bg);
4317 return -ETXTBSY;
4318 }
4319
4320 rc = alloc_reloc_control(fs_info);
4321 if (!rc) {
4322 btrfs_put_block_group(bg);
4323 return -ENOMEM;
4324 }
4325
4326 rc->extent_root = extent_root;
4327 rc->block_group = bg;
4328
4329 ret = btrfs_inc_block_group_ro(rc->block_group);
4330 if (ret) {
4331 err = ret;
4332 goto out;
4333 }
4334 rw = 1;
4335
4336 path = btrfs_alloc_path();
4337 if (!path) {
4338 err = -ENOMEM;
4339 goto out;
4340 }
4341
4342 inode = lookup_free_space_inode(rc->block_group, path);
4343 btrfs_free_path(path);
4344
4345 if (!IS_ERR(inode))
4346 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4347 else
4348 ret = PTR_ERR(inode);
4349
4350 if (ret && ret != -ENOENT) {
4351 err = ret;
4352 goto out;
4353 }
4354
4355 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4356 if (IS_ERR(rc->data_inode)) {
4357 err = PTR_ERR(rc->data_inode);
4358 rc->data_inode = NULL;
4359 goto out;
4360 }
4361
4362 describe_relocation(fs_info, rc->block_group);
4363
4364 btrfs_wait_block_group_reservations(rc->block_group);
4365 btrfs_wait_nocow_writers(rc->block_group);
4366 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4367 rc->block_group->key.objectid,
4368 rc->block_group->key.offset);
4369
4370 while (1) {
4371 mutex_lock(&fs_info->cleaner_mutex);
4372 ret = relocate_block_group(rc);
4373 mutex_unlock(&fs_info->cleaner_mutex);
4374 if (ret < 0)
4375 err = ret;
4376
4377 /*
4378 * We may have gotten ENOSPC after we already dirtied some
4379 * extents. If writeout happens while we're relocating a
4380 * different block group we could end up hitting the
4381 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4382 * btrfs_reloc_cow_block. Make sure we write everything out
4383 * properly so we don't trip over this problem, and then break
4384 * out of the loop if we hit an error.
4385 */
4386 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4387 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4388 (u64)-1);
4389 if (ret)
4390 err = ret;
4391 invalidate_mapping_pages(rc->data_inode->i_mapping,
4392 0, -1);
4393 rc->stage = UPDATE_DATA_PTRS;
4394 }
4395
4396 if (err < 0)
4397 goto out;
4398
4399 if (rc->extents_found == 0)
4400 break;
4401
4402 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4403
4404 }
4405
4406 WARN_ON(rc->block_group->pinned > 0);
4407 WARN_ON(rc->block_group->reserved > 0);
4408 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4409out:
4410 if (err && rw)
4411 btrfs_dec_block_group_ro(rc->block_group);
4412 iput(rc->data_inode);
4413 btrfs_put_block_group(rc->block_group);
4414 kfree(rc);
4415 return err;
4416}
4417
4418static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4419{
4420 struct btrfs_fs_info *fs_info = root->fs_info;
4421 struct btrfs_trans_handle *trans;
4422 int ret, err;
4423
4424 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4425 if (IS_ERR(trans))
4426 return PTR_ERR(trans);
4427
4428 memset(&root->root_item.drop_progress, 0,
4429 sizeof(root->root_item.drop_progress));
4430 root->root_item.drop_level = 0;
4431 btrfs_set_root_refs(&root->root_item, 0);
4432 ret = btrfs_update_root(trans, fs_info->tree_root,
4433 &root->root_key, &root->root_item);
4434
4435 err = btrfs_end_transaction(trans);
4436 if (err)
4437 return err;
4438 return ret;
4439}
4440
4441/*
4442 * recover relocation interrupted by system crash.
4443 *
4444 * this function resumes merging reloc trees with corresponding fs trees.
4445 * this is important for keeping the sharing of tree blocks
4446 */
4447int btrfs_recover_relocation(struct btrfs_root *root)
4448{
4449 struct btrfs_fs_info *fs_info = root->fs_info;
4450 LIST_HEAD(reloc_roots);
4451 struct btrfs_key key;
4452 struct btrfs_root *fs_root;
4453 struct btrfs_root *reloc_root;
4454 struct btrfs_path *path;
4455 struct extent_buffer *leaf;
4456 struct reloc_control *rc = NULL;
4457 struct btrfs_trans_handle *trans;
4458 int ret;
4459 int err = 0;
4460
4461 path = btrfs_alloc_path();
4462 if (!path)
4463 return -ENOMEM;
4464 path->reada = READA_BACK;
4465
4466 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4467 key.type = BTRFS_ROOT_ITEM_KEY;
4468 key.offset = (u64)-1;
4469
4470 while (1) {
4471 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4472 path, 0, 0);
4473 if (ret < 0) {
4474 err = ret;
4475 goto out;
4476 }
4477 if (ret > 0) {
4478 if (path->slots[0] == 0)
4479 break;
4480 path->slots[0]--;
4481 }
4482 leaf = path->nodes[0];
4483 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4484 btrfs_release_path(path);
4485
4486 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4487 key.type != BTRFS_ROOT_ITEM_KEY)
4488 break;
4489
4490 reloc_root = btrfs_read_fs_root(root, &key);
4491 if (IS_ERR(reloc_root)) {
4492 err = PTR_ERR(reloc_root);
4493 goto out;
4494 }
4495
4496 list_add(&reloc_root->root_list, &reloc_roots);
4497
4498 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4499 fs_root = read_fs_root(fs_info,
4500 reloc_root->root_key.offset);
4501 if (IS_ERR(fs_root)) {
4502 ret = PTR_ERR(fs_root);
4503 if (ret != -ENOENT) {
4504 err = ret;
4505 goto out;
4506 }
4507 ret = mark_garbage_root(reloc_root);
4508 if (ret < 0) {
4509 err = ret;
4510 goto out;
4511 }
4512 }
4513 }
4514
4515 if (key.offset == 0)
4516 break;
4517
4518 key.offset--;
4519 }
4520 btrfs_release_path(path);
4521
4522 if (list_empty(&reloc_roots))
4523 goto out;
4524
4525 rc = alloc_reloc_control(fs_info);
4526 if (!rc) {
4527 err = -ENOMEM;
4528 goto out;
4529 }
4530
4531 rc->extent_root = fs_info->extent_root;
4532
4533 set_reloc_control(rc);
4534
4535 trans = btrfs_join_transaction(rc->extent_root);
4536 if (IS_ERR(trans)) {
4537 unset_reloc_control(rc);
4538 err = PTR_ERR(trans);
4539 goto out_free;
4540 }
4541
4542 rc->merge_reloc_tree = 1;
4543
4544 while (!list_empty(&reloc_roots)) {
4545 reloc_root = list_entry(reloc_roots.next,
4546 struct btrfs_root, root_list);
4547 list_del(&reloc_root->root_list);
4548
4549 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4550 list_add_tail(&reloc_root->root_list,
4551 &rc->reloc_roots);
4552 continue;
4553 }
4554
4555 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4556 if (IS_ERR(fs_root)) {
4557 err = PTR_ERR(fs_root);
4558 goto out_free;
4559 }
4560
4561 err = __add_reloc_root(reloc_root);
4562 BUG_ON(err < 0); /* -ENOMEM or logic error */
4563 fs_root->reloc_root = reloc_root;
4564 }
4565
4566 err = btrfs_commit_transaction(trans);
4567 if (err)
4568 goto out_free;
4569
4570 merge_reloc_roots(rc);
4571
4572 unset_reloc_control(rc);
4573
4574 trans = btrfs_join_transaction(rc->extent_root);
4575 if (IS_ERR(trans)) {
4576 err = PTR_ERR(trans);
4577 goto out_free;
4578 }
4579 err = btrfs_commit_transaction(trans);
4580
4581 ret = clean_dirty_subvols(rc);
4582 if (ret < 0 && !err)
4583 err = ret;
4584out_free:
4585 kfree(rc);
4586out:
4587 if (!list_empty(&reloc_roots))
4588 free_reloc_roots(&reloc_roots);
4589
4590 btrfs_free_path(path);
4591
4592 if (err == 0) {
4593 /* cleanup orphan inode in data relocation tree */
4594 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4595 if (IS_ERR(fs_root))
4596 err = PTR_ERR(fs_root);
4597 else
4598 err = btrfs_orphan_cleanup(fs_root);
4599 }
4600 return err;
4601}
4602
4603/*
4604 * helper to add ordered checksum for data relocation.
4605 *
4606 * cloning checksum properly handles the nodatasum extents.
4607 * it also saves CPU time to re-calculate the checksum.
4608 */
4609int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4610{
4611 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4612 struct btrfs_ordered_sum *sums;
4613 struct btrfs_ordered_extent *ordered;
4614 int ret;
4615 u64 disk_bytenr;
4616 u64 new_bytenr;
4617 LIST_HEAD(list);
4618
4619 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4620 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4621
4622 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4623 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4624 disk_bytenr + len - 1, &list, 0);
4625 if (ret)
4626 goto out;
4627
4628 while (!list_empty(&list)) {
4629 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4630 list_del_init(&sums->list);
4631
4632 /*
4633 * We need to offset the new_bytenr based on where the csum is.
4634 * We need to do this because we will read in entire prealloc
4635 * extents but we may have written to say the middle of the
4636 * prealloc extent, so we need to make sure the csum goes with
4637 * the right disk offset.
4638 *
4639 * We can do this because the data reloc inode refers strictly
4640 * to the on disk bytes, so we don't have to worry about
4641 * disk_len vs real len like with real inodes since it's all
4642 * disk length.
4643 */
4644 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4645 sums->bytenr = new_bytenr;
4646
4647 btrfs_add_ordered_sum(ordered, sums);
4648 }
4649out:
4650 btrfs_put_ordered_extent(ordered);
4651 return ret;
4652}
4653
4654int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4655 struct btrfs_root *root, struct extent_buffer *buf,
4656 struct extent_buffer *cow)
4657{
4658 struct btrfs_fs_info *fs_info = root->fs_info;
4659 struct reloc_control *rc;
4660 struct backref_node *node;
4661 int first_cow = 0;
4662 int level;
4663 int ret = 0;
4664
4665 rc = fs_info->reloc_ctl;
4666 if (!rc)
4667 return 0;
4668
4669 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4670 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4671
4672 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4673 if (buf == root->node)
4674 __update_reloc_root(root, cow->start);
4675 }
4676
4677 level = btrfs_header_level(buf);
4678 if (btrfs_header_generation(buf) <=
4679 btrfs_root_last_snapshot(&root->root_item))
4680 first_cow = 1;
4681
4682 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4683 rc->create_reloc_tree) {
4684 WARN_ON(!first_cow && level == 0);
4685
4686 node = rc->backref_cache.path[level];
4687 BUG_ON(node->bytenr != buf->start &&
4688 node->new_bytenr != buf->start);
4689
4690 drop_node_buffer(node);
4691 extent_buffer_get(cow);
4692 node->eb = cow;
4693 node->new_bytenr = cow->start;
4694
4695 if (!node->pending) {
4696 list_move_tail(&node->list,
4697 &rc->backref_cache.pending[level]);
4698 node->pending = 1;
4699 }
4700
4701 if (first_cow)
4702 __mark_block_processed(rc, node);
4703
4704 if (first_cow && level > 0)
4705 rc->nodes_relocated += buf->len;
4706 }
4707
4708 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4709 ret = replace_file_extents(trans, rc, root, cow);
4710 return ret;
4711}
4712
4713/*
4714 * called before creating snapshot. it calculates metadata reservation
4715 * required for relocating tree blocks in the snapshot
4716 */
4717void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4718 u64 *bytes_to_reserve)
4719{
4720 struct btrfs_root *root = pending->root;
4721 struct reloc_control *rc = root->fs_info->reloc_ctl;
4722
4723 if (!root->reloc_root || !rc)
4724 return;
4725
4726 if (!rc->merge_reloc_tree)
4727 return;
4728
4729 root = root->reloc_root;
4730 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4731 /*
4732 * relocation is in the stage of merging trees. the space
4733 * used by merging a reloc tree is twice the size of
4734 * relocated tree nodes in the worst case. half for cowing
4735 * the reloc tree, half for cowing the fs tree. the space
4736 * used by cowing the reloc tree will be freed after the
4737 * tree is dropped. if we create snapshot, cowing the fs
4738 * tree may use more space than it frees. so we need
4739 * reserve extra space.
4740 */
4741 *bytes_to_reserve += rc->nodes_relocated;
4742}
4743
4744/*
4745 * called after snapshot is created. migrate block reservation
4746 * and create reloc root for the newly created snapshot
4747 */
4748int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4749 struct btrfs_pending_snapshot *pending)
4750{
4751 struct btrfs_root *root = pending->root;
4752 struct btrfs_root *reloc_root;
4753 struct btrfs_root *new_root;
4754 struct reloc_control *rc = root->fs_info->reloc_ctl;
4755 int ret;
4756
4757 if (!root->reloc_root || !rc)
4758 return 0;
4759
4760 rc = root->fs_info->reloc_ctl;
4761 rc->merging_rsv_size += rc->nodes_relocated;
4762
4763 if (rc->merge_reloc_tree) {
4764 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4765 rc->block_rsv,
4766 rc->nodes_relocated, true);
4767 if (ret)
4768 return ret;
4769 }
4770
4771 new_root = pending->snap;
4772 reloc_root = create_reloc_root(trans, root->reloc_root,
4773 new_root->root_key.objectid);
4774 if (IS_ERR(reloc_root))
4775 return PTR_ERR(reloc_root);
4776
4777 ret = __add_reloc_root(reloc_root);
4778 BUG_ON(ret < 0);
4779 new_root->reloc_root = reloc_root;
4780
4781 if (rc->create_reloc_tree)
4782 ret = clone_backref_node(trans, rc, root, reloc_root);
4783 return ret;
4784}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
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>
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "volumes.h"
17#include "locking.h"
18#include "btrfs_inode.h"
19#include "async-thread.h"
20#include "free-space-cache.h"
21#include "qgroup.h"
22#include "print-tree.h"
23#include "delalloc-space.h"
24#include "block-group.h"
25#include "backref.h"
26#include "misc.h"
27
28/*
29 * Relocation overview
30 *
31 * [What does relocation do]
32 *
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.
37 *
38 * Before | After
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)
43 *
44 * [How does relocation work]
45 *
46 * 1. Mark the target block group read-only
47 * New extents won't be allocated from the target block group.
48 *
49 * 2.1 Record each extent in the target block group
50 * To build a proper map of extents to be relocated.
51 *
52 * 2.2 Build data reloc tree and reloc trees
53 * Data reloc tree will contain an inode, recording all newly relocated
54 * data extents.
55 * There will be only one data reloc tree for one data block group.
56 *
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
60 * reloc tree built.
61 *
62 * 2.3 Swap source tree with its corresponding reloc tree
63 * Each involved tree only refers to new extents after swap.
64 *
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
68 * group.
69 *
70 * The main complexity is in steps 2.2 and 2.3.
71 *
72 * The entry point of relocation is relocate_block_group() function.
73 */
74
75#define RELOCATION_RESERVED_NODES 256
76/*
77 * map address of tree root to tree
78 */
79struct mapping_node {
80 struct {
81 struct rb_node rb_node;
82 u64 bytenr;
83 }; /* Use rb_simle_node for search/insert */
84 void *data;
85};
86
87struct mapping_tree {
88 struct rb_root rb_root;
89 spinlock_t lock;
90};
91
92/*
93 * present a tree block to process
94 */
95struct tree_block {
96 struct {
97 struct rb_node rb_node;
98 u64 bytenr;
99 }; /* Use rb_simple_node for search/insert */
100 u64 owner;
101 struct btrfs_key key;
102 unsigned int level:8;
103 unsigned int key_ready:1;
104};
105
106#define MAX_EXTENTS 128
107
108struct file_extent_cluster {
109 u64 start;
110 u64 end;
111 u64 boundary[MAX_EXTENTS];
112 unsigned int nr;
113};
114
115struct reloc_control {
116 /* block group to relocate */
117 struct btrfs_block_group *block_group;
118 /* extent tree */
119 struct btrfs_root *extent_root;
120 /* inode for moving data */
121 struct inode *data_inode;
122
123 struct btrfs_block_rsv *block_rsv;
124
125 struct btrfs_backref_cache backref_cache;
126
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 */
139 u64 nodes_relocated;
140 /* reserved size for block group relocation*/
141 u64 reserved_bytes;
142
143 u64 search_start;
144 u64 extents_found;
145
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;
150};
151
152/* stages of data relocation */
153#define MOVE_DATA_EXTENTS 0
154#define UPDATE_DATA_PTRS 1
155
156static void mark_block_processed(struct reloc_control *rc,
157 struct btrfs_backref_node *node)
158{
159 u32 blocksize;
160
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);
167 }
168 node->processed = 1;
169}
170
171
172static void mapping_tree_init(struct mapping_tree *tree)
173{
174 tree->rb_root = RB_ROOT;
175 spin_lock_init(&tree->lock);
176}
177
178/*
179 * walk up backref nodes until reach node presents tree root
180 */
181static struct btrfs_backref_node *walk_up_backref(
182 struct btrfs_backref_node *node,
183 struct btrfs_backref_edge *edges[], int *index)
184{
185 struct btrfs_backref_edge *edge;
186 int idx = *index;
187
188 while (!list_empty(&node->upper)) {
189 edge = list_entry(node->upper.next,
190 struct btrfs_backref_edge, list[LOWER]);
191 edges[idx++] = edge;
192 node = edge->node[UPPER];
193 }
194 BUG_ON(node->detached);
195 *index = idx;
196 return node;
197}
198
199/*
200 * walk down backref nodes to find start of next reference path
201 */
202static struct btrfs_backref_node *walk_down_backref(
203 struct btrfs_backref_edge *edges[], int *index)
204{
205 struct btrfs_backref_edge *edge;
206 struct btrfs_backref_node *lower;
207 int idx = *index;
208
209 while (idx > 0) {
210 edge = edges[idx - 1];
211 lower = edge->node[LOWER];
212 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
213 idx--;
214 continue;
215 }
216 edge = list_entry(edge->list[LOWER].next,
217 struct btrfs_backref_edge, list[LOWER]);
218 edges[idx - 1] = edge;
219 *index = idx;
220 return edge->node[UPPER];
221 }
222 *index = 0;
223 return NULL;
224}
225
226static void update_backref_node(struct btrfs_backref_cache *cache,
227 struct btrfs_backref_node *node, u64 bytenr)
228{
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);
233 if (rb_node)
234 btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
235}
236
237/*
238 * update backref cache after a transaction commit
239 */
240static int update_backref_cache(struct btrfs_trans_handle *trans,
241 struct btrfs_backref_cache *cache)
242{
243 struct btrfs_backref_node *node;
244 int level = 0;
245
246 if (cache->last_trans == 0) {
247 cache->last_trans = trans->transid;
248 return 0;
249 }
250
251 if (cache->last_trans == trans->transid)
252 return 0;
253
254 /*
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.
258 */
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);
263 }
264
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);
271 }
272
273 /*
274 * some nodes can be left in the pending list if there were
275 * errors during processing the pending nodes.
276 */
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)
281 continue;
282 update_backref_node(cache, node, node->new_bytenr);
283 }
284 }
285
286 cache->last_trans = 0;
287 return 1;
288}
289
290static bool reloc_root_is_dead(struct btrfs_root *root)
291{
292 /*
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
296 */
297 smp_rmb();
298 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
299 return true;
300 return false;
301}
302
303/*
304 * Check if this subvolume tree has valid reloc tree.
305 *
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
309 * special case.
310 */
311static bool have_reloc_root(struct btrfs_root *root)
312{
313 if (reloc_root_is_dead(root))
314 return false;
315 if (!root->reloc_root)
316 return false;
317 return true;
318}
319
320int btrfs_should_ignore_reloc_root(struct btrfs_root *root)
321{
322 struct btrfs_root *reloc_root;
323
324 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
325 return 0;
326
327 /* This root has been merged with its reloc tree, we can ignore it */
328 if (reloc_root_is_dead(root))
329 return 1;
330
331 reloc_root = root->reloc_root;
332 if (!reloc_root)
333 return 0;
334
335 if (btrfs_header_generation(reloc_root->commit_root) ==
336 root->fs_info->running_transaction->transid)
337 return 0;
338 /*
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
342 * relocation.
343 */
344 return 1;
345}
346
347/*
348 * find reloc tree by address of tree root
349 */
350struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
351{
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;
356
357 ASSERT(rc);
358 spin_lock(&rc->reloc_root_tree.lock);
359 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
360 if (rb_node) {
361 node = rb_entry(rb_node, struct mapping_node, rb_node);
362 root = (struct btrfs_root *)node->data;
363 }
364 spin_unlock(&rc->reloc_root_tree.lock);
365 return btrfs_grab_root(root);
366}
367
368/*
369 * For useless nodes, do two major clean ups:
370 *
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.
374 *
375 * - Freeing up leaves (level 0), keeps nodes detached
376 * For nodes, the node is still cached as "detached"
377 *
378 * Return false if @node is not in the @useless_nodes list.
379 * Return true if @node is in the @useless_nodes list.
380 */
381static bool handle_useless_nodes(struct reloc_control *rc,
382 struct btrfs_backref_node *node)
383{
384 struct btrfs_backref_cache *cache = &rc->backref_cache;
385 struct list_head *useless_node = &cache->useless_node;
386 bool ret = false;
387
388 while (!list_empty(useless_node)) {
389 struct btrfs_backref_node *cur;
390
391 cur = list_first_entry(useless_node, struct btrfs_backref_node,
392 list);
393 list_del_init(&cur->list);
394
395 /* Only tree root nodes can be added to @useless_nodes */
396 ASSERT(list_empty(&cur->upper));
397
398 if (cur == node)
399 ret = true;
400
401 /* The node is the lowest node */
402 if (cur->lowest) {
403 list_del_init(&cur->lower);
404 cur->lowest = 0;
405 }
406
407 /* Cleanup the lower edges */
408 while (!list_empty(&cur->lower)) {
409 struct btrfs_backref_edge *edge;
410 struct btrfs_backref_node *lower;
411
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);
418
419 /* Child node is also orphan, queue for cleanup */
420 if (list_empty(&lower->upper))
421 list_add(&lower->list, useless_node);
422 }
423 /* Mark this block processed for relocation */
424 mark_block_processed(rc, cur);
425
426 /*
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.
430 */
431 if (cur->level > 0) {
432 list_add(&cur->list, &cache->detached);
433 cur->detached = 1;
434 } else {
435 rb_erase(&cur->rb_node, &cache->rb_root);
436 btrfs_backref_free_node(cache, cur);
437 }
438 }
439 return ret;
440}
441
442/*
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.
446 *
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.
451 *
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
454 * cached.
455 */
456static 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)
459{
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;
467 int ret;
468 int err = 0;
469
470 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS);
471 if (!iter)
472 return ERR_PTR(-ENOMEM);
473 path = btrfs_alloc_path();
474 if (!path) {
475 err = -ENOMEM;
476 goto out;
477 }
478
479 node = btrfs_backref_alloc_node(cache, bytenr, level);
480 if (!node) {
481 err = -ENOMEM;
482 goto out;
483 }
484
485 node->lowest = 1;
486 cur = node;
487
488 /* Breadth-first search to build backref cache */
489 do {
490 ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
491 cur);
492 if (ret < 0) {
493 err = ret;
494 goto out;
495 }
496 edge = list_first_entry_or_null(&cache->pending_edge,
497 struct btrfs_backref_edge, list[UPPER]);
498 /*
499 * The pending list isn't empty, take the first block to
500 * process
501 */
502 if (edge) {
503 list_del_init(&edge->list[UPPER]);
504 cur = edge->node[UPPER];
505 }
506 } while (edge);
507
508 /* Finish the upper linkage of newly added edges/nodes */
509 ret = btrfs_backref_finish_upper_links(cache, node);
510 if (ret < 0) {
511 err = ret;
512 goto out;
513 }
514
515 if (handle_useless_nodes(rc, node))
516 node = NULL;
517out:
518 btrfs_backref_iter_free(iter);
519 btrfs_free_path(path);
520 if (err) {
521 btrfs_backref_error_cleanup(cache, node);
522 return ERR_PTR(err);
523 }
524 ASSERT(!node || !node->detached);
525 ASSERT(list_empty(&cache->useless_node) &&
526 list_empty(&cache->pending_edge));
527 return node;
528}
529
530/*
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
534 */
535static int clone_backref_node(struct btrfs_trans_handle *trans,
536 struct reloc_control *rc,
537 struct btrfs_root *src,
538 struct btrfs_root *dest)
539{
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;
547
548 if (cache->last_trans > 0)
549 update_backref_cache(trans, cache);
550
551 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
552 if (rb_node) {
553 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
554 if (node->detached)
555 node = NULL;
556 else
557 BUG_ON(node->new_bytenr != reloc_root->node->start);
558 }
559
560 if (!node) {
561 rb_node = rb_simple_search(&cache->rb_root,
562 reloc_root->commit_root->start);
563 if (rb_node) {
564 node = rb_entry(rb_node, struct btrfs_backref_node,
565 rb_node);
566 BUG_ON(node->detached);
567 }
568 }
569
570 if (!node)
571 return 0;
572
573 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
574 node->level);
575 if (!new_node)
576 return -ENOMEM;
577
578 new_node->lowest = node->lowest;
579 new_node->checked = 1;
580 new_node->root = btrfs_grab_root(dest);
581 ASSERT(new_node->root);
582
583 if (!node->lowest) {
584 list_for_each_entry(edge, &node->lower, list[UPPER]) {
585 new_edge = btrfs_backref_alloc_edge(cache);
586 if (!new_edge)
587 goto fail;
588
589 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
590 new_node, LINK_UPPER);
591 }
592 } else {
593 list_add_tail(&new_node->lower, &cache->leaves);
594 }
595
596 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
597 &new_node->rb_node);
598 if (rb_node)
599 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
600
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);
605 }
606 }
607 return 0;
608fail:
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);
614 }
615 btrfs_backref_free_node(cache, new_node);
616 return -ENOMEM;
617}
618
619/*
620 * helper to add 'address of tree root -> reloc tree' mapping
621 */
622static int __must_check __add_reloc_root(struct btrfs_root *root)
623{
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;
628
629 node = kmalloc(sizeof(*node), GFP_NOFS);
630 if (!node)
631 return -ENOMEM;
632
633 node->bytenr = root->commit_root->start;
634 node->data = root;
635
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);
640 if (rb_node) {
641 btrfs_err(fs_info,
642 "Duplicate root found for start=%llu while inserting into relocation tree",
643 node->bytenr);
644 return -EEXIST;
645 }
646
647 list_add_tail(&root->root_list, &rc->reloc_roots);
648 return 0;
649}
650
651/*
652 * helper to delete the 'address of tree root -> reloc tree'
653 * mapping
654 */
655static void __del_reloc_root(struct btrfs_root *root)
656{
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;
662
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);
667 if (rb_node) {
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);
671 }
672 spin_unlock(&rc->reloc_root_tree.lock);
673 ASSERT(!node || (struct btrfs_root *)node->data == root);
674 }
675
676 /*
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.
683 */
684 spin_lock(&fs_info->trans_lock);
685 if (!list_empty(&root->root_list)) {
686 put_ref = true;
687 list_del_init(&root->root_list);
688 }
689 spin_unlock(&fs_info->trans_lock);
690 if (put_ref)
691 btrfs_put_root(root);
692 kfree(node);
693}
694
695/*
696 * helper to update the 'address of tree root -> reloc tree'
697 * mapping
698 */
699static int __update_reloc_root(struct btrfs_root *root)
700{
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;
705
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);
709 if (rb_node) {
710 node = rb_entry(rb_node, struct mapping_node, rb_node);
711 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
712 }
713 spin_unlock(&rc->reloc_root_tree.lock);
714
715 if (!node)
716 return 0;
717 BUG_ON((struct btrfs_root *)node->data != root);
718
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);
724 if (rb_node)
725 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
726 return 0;
727}
728
729static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
730 struct btrfs_root *root, u64 objectid)
731{
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;
737 int ret = 0;
738 bool must_abort = false;
739
740 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
741 if (!root_item)
742 return ERR_PTR(-ENOMEM);
743
744 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
745 root_key.type = BTRFS_ROOT_ITEM_KEY;
746 root_key.offset = objectid;
747
748 if (root->root_key.objectid == objectid) {
749 u64 commit_root_gen;
750
751 /* called by btrfs_init_reloc_root */
752 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
753 BTRFS_TREE_RELOC_OBJECTID);
754 if (ret)
755 goto fail;
756
757 /*
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.
764 */
765 commit_root_gen = btrfs_header_generation(root->commit_root);
766 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
767 } else {
768 /*
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'.
774 */
775 ret = btrfs_copy_root(trans, root, root->node, &eb,
776 BTRFS_TREE_RELOC_OBJECTID);
777 if (ret)
778 goto fail;
779 }
780
781 /*
782 * We have changed references at this point, we must abort the
783 * transaction if anything fails.
784 */
785 must_abort = true;
786
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);
791
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);
797 }
798
799 btrfs_tree_unlock(eb);
800 free_extent_buffer(eb);
801
802 ret = btrfs_insert_root(trans, fs_info->tree_root,
803 &root_key, root_item);
804 if (ret)
805 goto fail;
806
807 kfree(root_item);
808
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);
812 goto abort;
813 }
814 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
815 reloc_root->last_trans = trans->transid;
816 return reloc_root;
817fail:
818 kfree(root_item);
819abort:
820 if (must_abort)
821 btrfs_abort_transaction(trans, ret);
822 return ERR_PTR(ret);
823}
824
825/*
826 * create reloc tree for a given fs tree. reloc tree is just a
827 * snapshot of the fs tree with special root objectid.
828 *
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.
831 */
832int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
833 struct btrfs_root *root)
834{
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;
839 int clear_rsv = 0;
840 int ret;
841
842 if (!rc)
843 return 0;
844
845 /*
846 * The subvolume has reloc tree but the swap is finished, no need to
847 * create/update the dead reloc tree
848 */
849 if (reloc_root_is_dead(root))
850 return 0;
851
852 /*
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
858 * in.
859 */
860 if (root->reloc_root) {
861 reloc_root = root->reloc_root;
862 reloc_root->last_trans = trans->transid;
863 return 0;
864 }
865
866 /*
867 * We are merging reloc roots, we do not need new reloc trees. Also
868 * reloc trees never need their own reloc tree.
869 */
870 if (!rc->create_reloc_tree ||
871 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
872 return 0;
873
874 if (!trans->reloc_reserved) {
875 rsv = trans->block_rsv;
876 trans->block_rsv = rc->block_rsv;
877 clear_rsv = 1;
878 }
879 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
880 if (clear_rsv)
881 trans->block_rsv = rsv;
882 if (IS_ERR(reloc_root))
883 return PTR_ERR(reloc_root);
884
885 ret = __add_reloc_root(reloc_root);
886 ASSERT(ret != -EEXIST);
887 if (ret) {
888 /* Pairs with create_reloc_root */
889 btrfs_put_root(reloc_root);
890 return ret;
891 }
892 root->reloc_root = btrfs_grab_root(reloc_root);
893 return 0;
894}
895
896/*
897 * update root item of reloc tree
898 */
899int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
900 struct btrfs_root *root)
901{
902 struct btrfs_fs_info *fs_info = root->fs_info;
903 struct btrfs_root *reloc_root;
904 struct btrfs_root_item *root_item;
905 int ret;
906
907 if (!have_reloc_root(root))
908 return 0;
909
910 reloc_root = root->reloc_root;
911 root_item = &reloc_root->root_item;
912
913 /*
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.
917 */
918 btrfs_grab_root(reloc_root);
919
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);
924 /*
925 * Mark the tree as dead before we change reloc_root so
926 * have_reloc_root will not touch it from now on.
927 */
928 smp_wmb();
929 __del_reloc_root(reloc_root);
930 }
931
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);
937 }
938
939 ret = btrfs_update_root(trans, fs_info->tree_root,
940 &reloc_root->root_key, root_item);
941 btrfs_put_root(reloc_root);
942 return ret;
943}
944
945/*
946 * helper to find first cached inode with inode number >= objectid
947 * in a subvolume
948 */
949static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
950{
951 struct rb_node *node;
952 struct rb_node *prev;
953 struct btrfs_inode *entry;
954 struct inode *inode;
955
956 spin_lock(&root->inode_lock);
957again:
958 node = root->inode_tree.rb_node;
959 prev = NULL;
960 while (node) {
961 prev = node;
962 entry = rb_entry(node, struct btrfs_inode, rb_node);
963
964 if (objectid < btrfs_ino(entry))
965 node = node->rb_left;
966 else if (objectid > btrfs_ino(entry))
967 node = node->rb_right;
968 else
969 break;
970 }
971 if (!node) {
972 while (prev) {
973 entry = rb_entry(prev, struct btrfs_inode, rb_node);
974 if (objectid <= btrfs_ino(entry)) {
975 node = prev;
976 break;
977 }
978 prev = rb_next(prev);
979 }
980 }
981 while (node) {
982 entry = rb_entry(node, struct btrfs_inode, rb_node);
983 inode = igrab(&entry->vfs_inode);
984 if (inode) {
985 spin_unlock(&root->inode_lock);
986 return inode;
987 }
988
989 objectid = btrfs_ino(entry) + 1;
990 if (cond_resched_lock(&root->inode_lock))
991 goto again;
992
993 node = rb_next(node);
994 }
995 spin_unlock(&root->inode_lock);
996 return NULL;
997}
998
999/*
1000 * get new location of data
1001 */
1002static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1003 u64 bytenr, u64 num_bytes)
1004{
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;
1009 int ret;
1010
1011 path = btrfs_alloc_path();
1012 if (!path)
1013 return -ENOMEM;
1014
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);
1018 if (ret < 0)
1019 goto out;
1020 if (ret > 0) {
1021 ret = -ENOENT;
1022 goto out;
1023 }
1024
1025 leaf = path->nodes[0];
1026 fi = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_file_extent_item);
1028
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));
1033
1034 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1035 ret = -EINVAL;
1036 goto out;
1037 }
1038
1039 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1040 ret = 0;
1041out:
1042 btrfs_free_path(path);
1043 return ret;
1044}
1045
1046/*
1047 * update file extent items in the tree leaf to point to
1048 * the new locations.
1049 */
1050static noinline_for_stack
1051int replace_file_extents(struct btrfs_trans_handle *trans,
1052 struct reloc_control *rc,
1053 struct btrfs_root *root,
1054 struct extent_buffer *leaf)
1055{
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;
1060 u64 parent;
1061 u64 bytenr;
1062 u64 new_bytenr = 0;
1063 u64 num_bytes;
1064 u64 end;
1065 u32 nritems;
1066 u32 i;
1067 int ret = 0;
1068 int first = 1;
1069 int dirty = 0;
1070
1071 if (rc->stage != UPDATE_DATA_PTRS)
1072 return 0;
1073
1074 /* reloc trees always use full backref */
1075 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1076 parent = leaf->start;
1077 else
1078 parent = 0;
1079
1080 nritems = btrfs_header_nritems(leaf);
1081 for (i = 0; i < nritems; i++) {
1082 struct btrfs_ref ref = { 0 };
1083
1084 cond_resched();
1085 btrfs_item_key_to_cpu(leaf, &key, i);
1086 if (key.type != BTRFS_EXTENT_DATA_KEY)
1087 continue;
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)
1091 continue;
1092 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1093 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1094 if (bytenr == 0)
1095 continue;
1096 if (!in_range(bytenr, rc->block_group->start,
1097 rc->block_group->length))
1098 continue;
1099
1100 /*
1101 * if we are modifying block in fs tree, wait for readpage
1102 * to complete and drop the extent cache
1103 */
1104 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1105 if (first) {
1106 inode = find_next_inode(root, key.objectid);
1107 first = 0;
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);
1111 }
1112 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1113 end = key.offset +
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));
1118 end--;
1119 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1120 key.offset, end);
1121 if (!ret)
1122 continue;
1123
1124 btrfs_drop_extent_cache(BTRFS_I(inode),
1125 key.offset, end, 1);
1126 unlock_extent(&BTRFS_I(inode)->io_tree,
1127 key.offset, end);
1128 }
1129 }
1130
1131 ret = get_new_location(rc->data_inode, &new_bytenr,
1132 bytenr, num_bytes);
1133 if (ret) {
1134 /*
1135 * Don't have to abort since we've not changed anything
1136 * in the file extent yet.
1137 */
1138 break;
1139 }
1140
1141 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1142 dirty = 1;
1143
1144 key.offset -= btrfs_file_extent_offset(leaf, fi);
1145 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1146 num_bytes, parent);
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);
1151 if (ret) {
1152 btrfs_abort_transaction(trans, ret);
1153 break;
1154 }
1155
1156 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1157 num_bytes, parent);
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);
1162 if (ret) {
1163 btrfs_abort_transaction(trans, ret);
1164 break;
1165 }
1166 }
1167 if (dirty)
1168 btrfs_mark_buffer_dirty(leaf);
1169 if (inode)
1170 btrfs_add_delayed_iput(inode);
1171 return ret;
1172}
1173
1174static noinline_for_stack
1175int memcmp_node_keys(struct extent_buffer *eb, int slot,
1176 struct btrfs_path *path, int level)
1177{
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));
1183}
1184
1185/*
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.
1189 *
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.
1193 */
1194static noinline_for_stack
1195int 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)
1199{
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;
1205 u64 old_bytenr;
1206 u64 new_bytenr;
1207 u64 old_ptr_gen;
1208 u64 new_ptr_gen;
1209 u64 last_snapshot;
1210 u32 blocksize;
1211 int cow = 0;
1212 int level;
1213 int ret;
1214 int slot;
1215
1216 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1217 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1218
1219 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1220again:
1221 slot = path->slots[lowest_level];
1222 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1223
1224 eb = btrfs_lock_root_node(dest);
1225 level = btrfs_header_level(eb);
1226
1227 if (level < lowest_level) {
1228 btrfs_tree_unlock(eb);
1229 free_extent_buffer(eb);
1230 return 0;
1231 }
1232
1233 if (cow) {
1234 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1235 BTRFS_NESTING_COW);
1236 if (ret) {
1237 btrfs_tree_unlock(eb);
1238 free_extent_buffer(eb);
1239 return ret;
1240 }
1241 }
1242
1243 if (next_key) {
1244 next_key->objectid = (u64)-1;
1245 next_key->type = (u8)-1;
1246 next_key->offset = (u64)-1;
1247 }
1248
1249 parent = eb;
1250 while (1) {
1251 level = btrfs_header_level(parent);
1252 ASSERT(level >= lowest_level);
1253
1254 ret = btrfs_bin_search(parent, &key, &slot);
1255 if (ret < 0)
1256 break;
1257 if (ret && slot > 0)
1258 slot--;
1259
1260 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1261 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1262
1263 old_bytenr = btrfs_node_blockptr(parent, slot);
1264 blocksize = fs_info->nodesize;
1265 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1266
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]);
1273 } else {
1274 new_bytenr = 0;
1275 new_ptr_gen = 0;
1276 }
1277
1278 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1279 ret = level;
1280 break;
1281 }
1282
1283 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1284 memcmp_node_keys(parent, slot, path, level)) {
1285 if (level <= lowest_level) {
1286 ret = 0;
1287 break;
1288 }
1289
1290 eb = btrfs_read_node_slot(parent, slot);
1291 if (IS_ERR(eb)) {
1292 ret = PTR_ERR(eb);
1293 break;
1294 }
1295 btrfs_tree_lock(eb);
1296 if (cow) {
1297 ret = btrfs_cow_block(trans, dest, eb, parent,
1298 slot, &eb,
1299 BTRFS_NESTING_COW);
1300 if (ret) {
1301 btrfs_tree_unlock(eb);
1302 free_extent_buffer(eb);
1303 break;
1304 }
1305 }
1306
1307 btrfs_tree_unlock(parent);
1308 free_extent_buffer(parent);
1309
1310 parent = eb;
1311 continue;
1312 }
1313
1314 if (!cow) {
1315 btrfs_tree_unlock(parent);
1316 free_extent_buffer(parent);
1317 cow = 1;
1318 goto again;
1319 }
1320
1321 btrfs_node_key_to_cpu(path->nodes[level], &key,
1322 path->slots[level]);
1323 btrfs_release_path(path);
1324
1325 path->lowest_level = level;
1326 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1327 path->lowest_level = 0;
1328 if (ret) {
1329 if (ret > 0)
1330 ret = -ENOENT;
1331 break;
1332 }
1333
1334 /*
1335 * Info qgroup to trace both subtrees.
1336 *
1337 * We must trace both trees.
1338 * 1) Tree reloc subtree
1339 * If not traced, we will leak data numbers
1340 * 2) Fs subtree
1341 * If not traced, we will double count old data
1342 *
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.
1347 */
1348 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1349 rc->block_group, parent, slot,
1350 path->nodes[level], path->slots[level],
1351 last_snapshot);
1352 if (ret < 0)
1353 break;
1354 /*
1355 * swap blocks in fs tree and reloc tree.
1356 */
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);
1360
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]);
1366
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);
1372 if (ret) {
1373 btrfs_abort_transaction(trans, ret);
1374 break;
1375 }
1376 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1377 blocksize, 0);
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);
1381 if (ret) {
1382 btrfs_abort_transaction(trans, ret);
1383 break;
1384 }
1385
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);
1391 if (ret) {
1392 btrfs_abort_transaction(trans, ret);
1393 break;
1394 }
1395
1396 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1397 blocksize, 0);
1398 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1399 ref.skip_qgroup = true;
1400 ret = btrfs_free_extent(trans, &ref);
1401 if (ret) {
1402 btrfs_abort_transaction(trans, ret);
1403 break;
1404 }
1405
1406 btrfs_unlock_up_safe(path, 0);
1407
1408 ret = level;
1409 break;
1410 }
1411 btrfs_tree_unlock(parent);
1412 free_extent_buffer(parent);
1413 return ret;
1414}
1415
1416/*
1417 * helper to find next relocated block in reloc tree
1418 */
1419static noinline_for_stack
1420int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1421 int *level)
1422{
1423 struct extent_buffer *eb;
1424 int i;
1425 u64 last_snapshot;
1426 u32 nritems;
1427
1428 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1429
1430 for (i = 0; i < *level; i++) {
1431 free_extent_buffer(path->nodes[i]);
1432 path->nodes[i] = NULL;
1433 }
1434
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) {
1439 path->slots[i]++;
1440 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1441 last_snapshot)
1442 continue;
1443
1444 *level = i;
1445 return 0;
1446 }
1447 free_extent_buffer(path->nodes[i]);
1448 path->nodes[i] = NULL;
1449 }
1450 return 1;
1451}
1452
1453/*
1454 * walk down reloc tree to find relocated block of lowest level
1455 */
1456static noinline_for_stack
1457int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1458 int *level)
1459{
1460 struct extent_buffer *eb = NULL;
1461 int i;
1462 u64 ptr_gen = 0;
1463 u64 last_snapshot;
1464 u32 nritems;
1465
1466 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1467
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)
1474 break;
1475 path->slots[i]++;
1476 }
1477 if (path->slots[i] >= nritems) {
1478 if (i == *level)
1479 break;
1480 *level = i + 1;
1481 return 0;
1482 }
1483 if (i == 1) {
1484 *level = i;
1485 return 0;
1486 }
1487
1488 eb = btrfs_read_node_slot(eb, path->slots[i]);
1489 if (IS_ERR(eb))
1490 return PTR_ERR(eb);
1491 BUG_ON(btrfs_header_level(eb) != i - 1);
1492 path->nodes[i - 1] = eb;
1493 path->slots[i - 1] = 0;
1494 }
1495 return 1;
1496}
1497
1498/*
1499 * invalidate extent cache for file extents whose key in range of
1500 * [min_key, max_key)
1501 */
1502static int invalidate_extent_cache(struct btrfs_root *root,
1503 struct btrfs_key *min_key,
1504 struct btrfs_key *max_key)
1505{
1506 struct btrfs_fs_info *fs_info = root->fs_info;
1507 struct inode *inode = NULL;
1508 u64 objectid;
1509 u64 start, end;
1510 u64 ino;
1511
1512 objectid = min_key->objectid;
1513 while (1) {
1514 cond_resched();
1515 iput(inode);
1516
1517 if (objectid > max_key->objectid)
1518 break;
1519
1520 inode = find_next_inode(root, objectid);
1521 if (!inode)
1522 break;
1523 ino = btrfs_ino(BTRFS_I(inode));
1524
1525 if (ino > max_key->objectid) {
1526 iput(inode);
1527 break;
1528 }
1529
1530 objectid = ino + 1;
1531 if (!S_ISREG(inode->i_mode))
1532 continue;
1533
1534 if (unlikely(min_key->objectid == ino)) {
1535 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1536 continue;
1537 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1538 start = 0;
1539 else {
1540 start = min_key->offset;
1541 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1542 }
1543 } else {
1544 start = 0;
1545 }
1546
1547 if (unlikely(max_key->objectid == ino)) {
1548 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1549 continue;
1550 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1551 end = (u64)-1;
1552 } else {
1553 if (max_key->offset == 0)
1554 continue;
1555 end = max_key->offset;
1556 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1557 end--;
1558 }
1559 } else {
1560 end = (u64)-1;
1561 }
1562
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);
1567 }
1568 return 0;
1569}
1570
1571static int find_next_key(struct btrfs_path *path, int level,
1572 struct btrfs_key *key)
1573
1574{
1575 while (level < BTRFS_MAX_LEVEL) {
1576 if (!path->nodes[level])
1577 break;
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);
1582 return 0;
1583 }
1584 level++;
1585 }
1586 return 1;
1587}
1588
1589/*
1590 * Insert current subvolume into reloc_control::dirty_subvol_roots
1591 */
1592static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1593 struct reloc_control *rc,
1594 struct btrfs_root *root)
1595{
1596 struct btrfs_root *reloc_root = root->reloc_root;
1597 struct btrfs_root_item *reloc_root_item;
1598 int ret;
1599
1600 /* @root must be a subvolume tree root with a valid reloc tree */
1601 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1602 ASSERT(reloc_root);
1603
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);
1610 if (ret)
1611 return ret;
1612
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);
1616 }
1617
1618 return 0;
1619}
1620
1621static int clean_dirty_subvols(struct reloc_control *rc)
1622{
1623 struct btrfs_root *root;
1624 struct btrfs_root *next;
1625 int ret = 0;
1626 int ret2;
1627
1628 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1629 reloc_dirty_list) {
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;
1633
1634 list_del_init(&root->reloc_dirty_list);
1635 root->reloc_root = NULL;
1636 /*
1637 * Need barrier to ensure clear_bit() only happens after
1638 * root->reloc_root = NULL. Pairs with have_reloc_root.
1639 */
1640 smp_wmb();
1641 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1642 if (reloc_root) {
1643 /*
1644 * btrfs_drop_snapshot drops our ref we hold for
1645 * ->reloc_root. If it fails however we must
1646 * drop the ref ourselves.
1647 */
1648 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1649 if (ret2 < 0) {
1650 btrfs_put_root(reloc_root);
1651 if (!ret)
1652 ret = ret2;
1653 }
1654 }
1655 btrfs_put_root(root);
1656 } else {
1657 /* Orphan reloc tree, just clean it up */
1658 ret2 = btrfs_drop_snapshot(root, 0, 1);
1659 if (ret2 < 0) {
1660 btrfs_put_root(root);
1661 if (!ret)
1662 ret = ret2;
1663 }
1664 }
1665 }
1666 return ret;
1667}
1668
1669/*
1670 * merge the relocated tree blocks in reloc tree with corresponding
1671 * fs tree.
1672 */
1673static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1674 struct btrfs_root *root)
1675{
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;
1684 int reserve_level;
1685 int level;
1686 int max_level;
1687 int replaced = 0;
1688 int ret = 0;
1689 u32 min_reserved;
1690
1691 path = btrfs_alloc_path();
1692 if (!path)
1693 return -ENOMEM;
1694 path->reada = READA_FORWARD;
1695
1696 reloc_root = root->reloc_root;
1697 root_item = &reloc_root->root_item;
1698
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;
1704 } else {
1705 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1706
1707 level = btrfs_root_drop_level(root_item);
1708 BUG_ON(level == 0);
1709 path->lowest_level = level;
1710 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1711 path->lowest_level = 0;
1712 if (ret < 0) {
1713 btrfs_free_path(path);
1714 return ret;
1715 }
1716
1717 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1718 path->slots[level]);
1719 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1720
1721 btrfs_unlock_up_safe(path, 0);
1722 }
1723
1724 /*
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.
1728 *
1729 * Thus the needed metadata size is at most root_level * nodesize,
1730 * and * 2 since we have two trees to COW.
1731 */
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));
1735
1736 while (1) {
1737 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1738 BTRFS_RESERVE_FLUSH_LIMIT);
1739 if (ret)
1740 goto out;
1741 trans = btrfs_start_transaction(root, 0);
1742 if (IS_ERR(trans)) {
1743 ret = PTR_ERR(trans);
1744 trans = NULL;
1745 goto out;
1746 }
1747
1748 /*
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.
1751 *
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
1756 * appropriately.
1757 */
1758 reloc_root->last_trans = trans->transid;
1759 trans->block_rsv = rc->block_rsv;
1760
1761 replaced = 0;
1762 max_level = level;
1763
1764 ret = walk_down_reloc_tree(reloc_root, path, &level);
1765 if (ret < 0)
1766 goto out;
1767 if (ret > 0)
1768 break;
1769
1770 if (!find_next_key(path, level, &key) &&
1771 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1772 ret = 0;
1773 } else {
1774 ret = replace_path(trans, rc, root, reloc_root, path,
1775 &next_key, level, max_level);
1776 }
1777 if (ret < 0)
1778 goto out;
1779 if (ret > 0) {
1780 level = ret;
1781 btrfs_node_key_to_cpu(path->nodes[level], &key,
1782 path->slots[level]);
1783 replaced = 1;
1784 }
1785
1786 ret = walk_up_reloc_tree(reloc_root, path, &level);
1787 if (ret > 0)
1788 break;
1789
1790 BUG_ON(level == 0);
1791 /*
1792 * save the merging progress in the drop_progress.
1793 * this is OK since root refs == 1 in this case.
1794 */
1795 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1796 path->slots[level]);
1797 btrfs_set_root_drop_level(root_item, level);
1798
1799 btrfs_end_transaction_throttle(trans);
1800 trans = NULL;
1801
1802 btrfs_btree_balance_dirty(fs_info);
1803
1804 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1805 invalidate_extent_cache(root, &key, &next_key);
1806 }
1807
1808 /*
1809 * handle the case only one block in the fs tree need to be
1810 * relocated and the block is tree root.
1811 */
1812 leaf = btrfs_lock_root_node(root);
1813 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1814 BTRFS_NESTING_COW);
1815 btrfs_tree_unlock(leaf);
1816 free_extent_buffer(leaf);
1817out:
1818 btrfs_free_path(path);
1819
1820 if (ret == 0) {
1821 ret = insert_dirty_subvol(trans, rc, root);
1822 if (ret)
1823 btrfs_abort_transaction(trans, ret);
1824 }
1825
1826 if (trans)
1827 btrfs_end_transaction_throttle(trans);
1828
1829 btrfs_btree_balance_dirty(fs_info);
1830
1831 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1832 invalidate_extent_cache(root, &key, &next_key);
1833
1834 return ret;
1835}
1836
1837static noinline_for_stack
1838int prepare_to_merge(struct reloc_control *rc, int err)
1839{
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);
1845 u64 num_bytes = 0;
1846 int ret;
1847
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);
1852
1853again:
1854 if (!err) {
1855 num_bytes = rc->merging_rsv_size;
1856 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1857 BTRFS_RESERVE_FLUSH_ALL);
1858 if (ret)
1859 err = ret;
1860 }
1861
1862 trans = btrfs_join_transaction(rc->extent_root);
1863 if (IS_ERR(trans)) {
1864 if (!err)
1865 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1866 num_bytes, NULL);
1867 return PTR_ERR(trans);
1868 }
1869
1870 if (!err) {
1871 if (num_bytes != rc->merging_rsv_size) {
1872 btrfs_end_transaction(trans);
1873 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1874 num_bytes, NULL);
1875 goto again;
1876 }
1877 }
1878
1879 rc->merge_reloc_tree = 1;
1880
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);
1885
1886 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1887 false);
1888 if (IS_ERR(root)) {
1889 /*
1890 * Even if we have an error we need this reloc root
1891 * back on our list so we can clean up properly.
1892 */
1893 list_add(&reloc_root->root_list, &reloc_roots);
1894 btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1895 if (!err)
1896 err = PTR_ERR(root);
1897 break;
1898 }
1899 ASSERT(root->reloc_root == reloc_root);
1900
1901 /*
1902 * set reference count to 1, so btrfs_recover_relocation
1903 * knows it should resumes merging
1904 */
1905 if (!err)
1906 btrfs_set_root_refs(&reloc_root->root_item, 1);
1907 ret = btrfs_update_reloc_root(trans, root);
1908
1909 /*
1910 * Even if we have an error we need this reloc root back on our
1911 * list so we can clean up properly.
1912 */
1913 list_add(&reloc_root->root_list, &reloc_roots);
1914 btrfs_put_root(root);
1915
1916 if (ret) {
1917 btrfs_abort_transaction(trans, ret);
1918 if (!err)
1919 err = ret;
1920 break;
1921 }
1922 }
1923
1924 list_splice(&reloc_roots, &rc->reloc_roots);
1925
1926 if (!err)
1927 err = btrfs_commit_transaction(trans);
1928 else
1929 btrfs_end_transaction(trans);
1930 return err;
1931}
1932
1933static noinline_for_stack
1934void free_reloc_roots(struct list_head *list)
1935{
1936 struct btrfs_root *reloc_root, *tmp;
1937
1938 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1939 __del_reloc_root(reloc_root);
1940}
1941
1942static noinline_for_stack
1943void merge_reloc_roots(struct reloc_control *rc)
1944{
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);
1949 int found = 0;
1950 int ret = 0;
1951again:
1952 root = rc->extent_root;
1953
1954 /*
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
1958 * doing this splice
1959 */
1960 mutex_lock(&fs_info->reloc_mutex);
1961 list_splice_init(&rc->reloc_roots, &reloc_roots);
1962 mutex_unlock(&fs_info->reloc_mutex);
1963
1964 while (!list_empty(&reloc_roots)) {
1965 found = 1;
1966 reloc_root = list_entry(reloc_roots.next,
1967 struct btrfs_root, root_list);
1968
1969 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1970 false);
1971 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1972 if (IS_ERR(root)) {
1973 /*
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.
1980 */
1981 ASSERT(0);
1982 ret = PTR_ERR(root);
1983 goto out;
1984 }
1985 if (root->reloc_root != reloc_root) {
1986 /*
1987 * This is actually impossible without something
1988 * going really wrong (like weird race condition
1989 * or cosmic rays).
1990 */
1991 ASSERT(0);
1992 ret = -EINVAL;
1993 goto out;
1994 }
1995 ret = merge_reloc_root(rc, root);
1996 btrfs_put_root(root);
1997 if (ret) {
1998 if (list_empty(&reloc_root->root_list))
1999 list_add_tail(&reloc_root->root_list,
2000 &reloc_roots);
2001 goto out;
2002 }
2003 } else {
2004 if (!IS_ERR(root)) {
2005 if (root->reloc_root == reloc_root) {
2006 root->reloc_root = NULL;
2007 btrfs_put_root(reloc_root);
2008 }
2009 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
2010 &root->state);
2011 btrfs_put_root(root);
2012 }
2013
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);
2018 }
2019 }
2020
2021 if (found) {
2022 found = 0;
2023 goto again;
2024 }
2025out:
2026 if (ret) {
2027 btrfs_handle_fs_error(fs_info, ret, NULL);
2028 free_reloc_roots(&reloc_roots);
2029
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);
2035 }
2036
2037 /*
2038 * We used to have
2039 *
2040 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2041 *
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.
2049 *
2050 * The remaining nodes will be cleaned up by free_reloc_control.
2051 */
2052}
2053
2054static void free_block_list(struct rb_root *blocks)
2055{
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);
2061 kfree(block);
2062 }
2063}
2064
2065static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *reloc_root)
2067{
2068 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2069 struct btrfs_root *root;
2070 int ret;
2071
2072 if (reloc_root->last_trans == trans->transid)
2073 return 0;
2074
2075 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2076
2077 /*
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
2080 * root.
2081 *
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.
2084 */
2085 if (IS_ERR(root)) {
2086 ASSERT(0);
2087 return PTR_ERR(root);
2088 }
2089 if (root->reloc_root != reloc_root) {
2090 ASSERT(0);
2091 btrfs_err(fs_info,
2092 "root %llu has two reloc roots associated with it",
2093 reloc_root->root_key.offset);
2094 btrfs_put_root(root);
2095 return -EUCLEAN;
2096 }
2097 ret = btrfs_record_root_in_trans(trans, root);
2098 btrfs_put_root(root);
2099
2100 return ret;
2101}
2102
2103static noinline_for_stack
2104struct 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[])
2108{
2109 struct btrfs_backref_node *next;
2110 struct btrfs_root *root;
2111 int index = 0;
2112 int ret;
2113
2114 next = node;
2115 while (1) {
2116 cond_resched();
2117 next = walk_up_backref(next, edges, &index);
2118 root = next->root;
2119
2120 /*
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.
2124 *
2125 * This path is only for SHAREABLE roots, so if we come upon a
2126 * non-SHAREABLE root then we have backrefs that resolve
2127 * improperly.
2128 *
2129 * Both of these cases indicate file system corruption, or a bug
2130 * in the backref walking code.
2131 */
2132 if (!root) {
2133 ASSERT(0);
2134 btrfs_err(trans->fs_info,
2135 "bytenr %llu doesn't have a backref path ending in a root",
2136 node->bytenr);
2137 return ERR_PTR(-EUCLEAN);
2138 }
2139 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2140 ASSERT(0);
2141 btrfs_err(trans->fs_info,
2142 "bytenr %llu has multiple refs with one ending in a non-shareable root",
2143 node->bytenr);
2144 return ERR_PTR(-EUCLEAN);
2145 }
2146
2147 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2148 ret = record_reloc_root_in_trans(trans, root);
2149 if (ret)
2150 return ERR_PTR(ret);
2151 break;
2152 }
2153
2154 ret = btrfs_record_root_in_trans(trans, root);
2155 if (ret)
2156 return ERR_PTR(ret);
2157 root = root->reloc_root;
2158
2159 /*
2160 * We could have raced with another thread which failed, so
2161 * root->reloc_root may not be set, return ENOENT in this case.
2162 */
2163 if (!root)
2164 return ERR_PTR(-ENOENT);
2165
2166 if (next->new_bytenr != root->node->start) {
2167 /*
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.
2173 */
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);
2181 }
2182
2183 next->new_bytenr = root->node->start;
2184 btrfs_put_root(next->root);
2185 next->root = btrfs_grab_root(root);
2186 ASSERT(next->root);
2187 list_add_tail(&next->list,
2188 &rc->backref_cache.changed);
2189 mark_block_processed(rc, next);
2190 break;
2191 }
2192
2193 WARN_ON(1);
2194 root = NULL;
2195 next = walk_down_backref(edges, &index);
2196 if (!next || next->level <= node->level)
2197 break;
2198 }
2199 if (!root) {
2200 /*
2201 * This can happen if there's fs corruption or if there's a bug
2202 * in the backref lookup code.
2203 */
2204 ASSERT(0);
2205 return ERR_PTR(-ENOENT);
2206 }
2207
2208 next = node;
2209 /* setup backref node path for btrfs_reloc_cow_block */
2210 while (1) {
2211 rc->backref_cache.path[next->level] = next;
2212 if (--index < 0)
2213 break;
2214 next = edges[index]->node[UPPER];
2215 }
2216 return root;
2217}
2218
2219/*
2220 * Select a tree root for relocation.
2221 *
2222 * Return NULL if the block is not shareable. We should use do_relocation() in
2223 * this case.
2224 *
2225 * Return a tree root pointer if the block is shareable.
2226 * Return -ENOENT if the block is root of reloc tree.
2227 */
2228static noinline_for_stack
2229struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2230{
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];
2235 int index = 0;
2236
2237 next = node;
2238 while (1) {
2239 cond_resched();
2240 next = walk_up_backref(next, edges, &index);
2241 root = next->root;
2242
2243 /*
2244 * This can occur if we have incomplete extent refs leading all
2245 * the way up a particular path, in this case return -EUCLEAN.
2246 */
2247 if (!root)
2248 return ERR_PTR(-EUCLEAN);
2249
2250 /* No other choice for non-shareable tree */
2251 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2252 return root;
2253
2254 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2255 fs_root = root;
2256
2257 if (next != node)
2258 return NULL;
2259
2260 next = walk_down_backref(edges, &index);
2261 if (!next || next->level <= node->level)
2262 break;
2263 }
2264
2265 if (!fs_root)
2266 return ERR_PTR(-ENOENT);
2267 return fs_root;
2268}
2269
2270static noinline_for_stack
2271u64 calcu_metadata_size(struct reloc_control *rc,
2272 struct btrfs_backref_node *node, int reserve)
2273{
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];
2278 u64 num_bytes = 0;
2279 int index = 0;
2280
2281 BUG_ON(reserve && node->processed);
2282
2283 while (next) {
2284 cond_resched();
2285 while (1) {
2286 if (next->processed && (reserve || next != node))
2287 break;
2288
2289 num_bytes += fs_info->nodesize;
2290
2291 if (list_empty(&next->upper))
2292 break;
2293
2294 edge = list_entry(next->upper.next,
2295 struct btrfs_backref_edge, list[LOWER]);
2296 edges[index++] = edge;
2297 next = edge->node[UPPER];
2298 }
2299 next = walk_down_backref(edges, &index);
2300 }
2301 return num_bytes;
2302}
2303
2304static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2305 struct reloc_control *rc,
2306 struct btrfs_backref_node *node)
2307{
2308 struct btrfs_root *root = rc->extent_root;
2309 struct btrfs_fs_info *fs_info = root->fs_info;
2310 u64 num_bytes;
2311 int ret;
2312 u64 tmp;
2313
2314 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2315
2316 trans->block_rsv = rc->block_rsv;
2317 rc->reserved_bytes += num_bytes;
2318
2319 /*
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.
2323 */
2324 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2325 BTRFS_RESERVE_FLUSH_LIMIT);
2326 if (ret) {
2327 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2328 while (tmp <= rc->reserved_bytes)
2329 tmp <<= 1;
2330 /*
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
2335 * enospc case.
2336 */
2337 rc->block_rsv->size = tmp + fs_info->nodesize *
2338 RELOCATION_RESERVED_NODES;
2339 return -EAGAIN;
2340 }
2341
2342 return 0;
2343}
2344
2345/*
2346 * relocate a block tree, and then update pointers in upper level
2347 * blocks that reference the block to point to the new location.
2348 *
2349 * if called by link_to_upper, the block has already been relocated.
2350 * in that case this function just updates pointers.
2351 */
2352static 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)
2357{
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;
2363 u32 blocksize;
2364 u64 bytenr;
2365 int slot;
2366 int ret = 0;
2367
2368 /*
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.
2371 */
2372 ASSERT(!lowest || !node->eb);
2373
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 };
2378
2379 cond_resched();
2380
2381 upper = edge->node[UPPER];
2382 root = select_reloc_root(trans, rc, upper, edges);
2383 if (IS_ERR(root)) {
2384 ret = PTR_ERR(root);
2385 goto next;
2386 }
2387
2388 if (upper->eb && !upper->locked) {
2389 if (!lowest) {
2390 ret = btrfs_bin_search(upper->eb, key, &slot);
2391 if (ret < 0)
2392 goto next;
2393 BUG_ON(ret);
2394 bytenr = btrfs_node_blockptr(upper->eb, slot);
2395 if (node->eb->start == bytenr)
2396 goto next;
2397 }
2398 btrfs_backref_drop_node_buffer(upper);
2399 }
2400
2401 if (!upper->eb) {
2402 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2403 if (ret) {
2404 if (ret > 0)
2405 ret = -ENOENT;
2406
2407 btrfs_release_path(path);
2408 break;
2409 }
2410
2411 if (!upper->eb) {
2412 upper->eb = path->nodes[upper->level];
2413 path->nodes[upper->level] = NULL;
2414 } else {
2415 BUG_ON(upper->eb != path->nodes[upper->level]);
2416 }
2417
2418 upper->locked = 1;
2419 path->locks[upper->level] = 0;
2420
2421 slot = path->slots[upper->level];
2422 btrfs_release_path(path);
2423 } else {
2424 ret = btrfs_bin_search(upper->eb, key, &slot);
2425 if (ret < 0)
2426 goto next;
2427 BUG_ON(ret);
2428 }
2429
2430 bytenr = btrfs_node_blockptr(upper->eb, slot);
2431 if (lowest) {
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,
2436 upper->eb->start);
2437 ret = -EIO;
2438 goto next;
2439 }
2440 } else {
2441 if (node->eb->start == bytenr)
2442 goto next;
2443 }
2444
2445 blocksize = root->fs_info->nodesize;
2446 eb = btrfs_read_node_slot(upper->eb, slot);
2447 if (IS_ERR(eb)) {
2448 ret = PTR_ERR(eb);
2449 goto next;
2450 }
2451 btrfs_tree_lock(eb);
2452
2453 if (!node->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);
2458 if (ret < 0)
2459 goto next;
2460 /*
2461 * We've just COWed this block, it should have updated
2462 * the correct backref node entry.
2463 */
2464 ASSERT(node->eb == eb);
2465 } else {
2466 btrfs_set_node_blockptr(upper->eb, slot,
2467 node->eb->start);
2468 btrfs_set_node_ptr_generation(upper->eb, slot,
2469 trans->transid);
2470 btrfs_mark_buffer_dirty(upper->eb);
2471
2472 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2473 node->eb->start, blocksize,
2474 upper->eb->start);
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);
2479 if (!ret)
2480 ret = btrfs_drop_subtree(trans, root, eb,
2481 upper->eb);
2482 if (ret)
2483 btrfs_abort_transaction(trans, ret);
2484 }
2485next:
2486 if (!upper->pending)
2487 btrfs_backref_drop_node_buffer(upper);
2488 else
2489 btrfs_backref_unlock_node_buffer(upper);
2490 if (ret)
2491 break;
2492 }
2493
2494 if (!ret && node->pending) {
2495 btrfs_backref_drop_node_buffer(node);
2496 list_move_tail(&node->list, &rc->backref_cache.changed);
2497 node->pending = 0;
2498 }
2499
2500 path->lowest_level = 0;
2501
2502 /*
2503 * We should have allocated all of our space in the block rsv and thus
2504 * shouldn't ENOSPC.
2505 */
2506 ASSERT(ret != -ENOSPC);
2507 return ret;
2508}
2509
2510static 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)
2514{
2515 struct btrfs_key key;
2516
2517 btrfs_node_key_to_cpu(node->eb, &key, 0);
2518 return do_relocation(trans, rc, node, &key, path, 0);
2519}
2520
2521static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2522 struct reloc_control *rc,
2523 struct btrfs_path *path, int err)
2524{
2525 LIST_HEAD(list);
2526 struct btrfs_backref_cache *cache = &rc->backref_cache;
2527 struct btrfs_backref_node *node;
2528 int level;
2529 int ret;
2530
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);
2537
2538 if (!err) {
2539 ret = link_to_upper(trans, rc, node, path);
2540 if (ret < 0)
2541 err = ret;
2542 }
2543 }
2544 list_splice_init(&list, &cache->pending[level]);
2545 }
2546 return err;
2547}
2548
2549/*
2550 * mark a block and all blocks directly/indirectly reference the block
2551 * as processed.
2552 */
2553static void update_processed_blocks(struct reloc_control *rc,
2554 struct btrfs_backref_node *node)
2555{
2556 struct btrfs_backref_node *next = node;
2557 struct btrfs_backref_edge *edge;
2558 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2559 int index = 0;
2560
2561 while (next) {
2562 cond_resched();
2563 while (1) {
2564 if (next->processed)
2565 break;
2566
2567 mark_block_processed(rc, next);
2568
2569 if (list_empty(&next->upper))
2570 break;
2571
2572 edge = list_entry(next->upper.next,
2573 struct btrfs_backref_edge, list[LOWER]);
2574 edges[index++] = edge;
2575 next = edge->node[UPPER];
2576 }
2577 next = walk_down_backref(edges, &index);
2578 }
2579}
2580
2581static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2582{
2583 u32 blocksize = rc->extent_root->fs_info->nodesize;
2584
2585 if (test_range_bit(&rc->processed_blocks, bytenr,
2586 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2587 return 1;
2588 return 0;
2589}
2590
2591static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2592 struct tree_block *block)
2593{
2594 struct extent_buffer *eb;
2595
2596 eb = read_tree_block(fs_info, block->bytenr, block->owner,
2597 block->key.offset, block->level, NULL);
2598 if (IS_ERR(eb)) {
2599 return PTR_ERR(eb);
2600 } else if (!extent_buffer_uptodate(eb)) {
2601 free_extent_buffer(eb);
2602 return -EIO;
2603 }
2604 if (block->level == 0)
2605 btrfs_item_key_to_cpu(eb, &block->key, 0);
2606 else
2607 btrfs_node_key_to_cpu(eb, &block->key, 0);
2608 free_extent_buffer(eb);
2609 block->key_ready = 1;
2610 return 0;
2611}
2612
2613/*
2614 * helper function to relocate a tree block
2615 */
2616static 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)
2621{
2622 struct btrfs_root *root;
2623 int ret = 0;
2624
2625 if (!node)
2626 return 0;
2627
2628 /*
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.
2631 */
2632 ret = reserve_metadata_space(trans, rc, node);
2633 if (ret)
2634 goto out;
2635
2636 BUG_ON(node->processed);
2637 root = select_one_root(node);
2638 if (IS_ERR(root)) {
2639 ret = PTR_ERR(root);
2640
2641 /* See explanation in select_one_root for the -EUCLEAN case. */
2642 ASSERT(ret == -ENOENT);
2643 if (ret == -ENOENT) {
2644 ret = 0;
2645 update_processed_blocks(rc, node);
2646 }
2647 goto out;
2648 }
2649
2650 if (root) {
2651 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2652 /*
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.
2657 *
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.
2664 */
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",
2670 node->bytenr);
2671 ret = -EUCLEAN;
2672 goto out;
2673 }
2674 ret = btrfs_record_root_in_trans(trans, root);
2675 if (ret)
2676 goto out;
2677 /*
2678 * Another thread could have failed, need to check if we
2679 * have reloc_root actually set.
2680 */
2681 if (!root->reloc_root) {
2682 ret = -ENOENT;
2683 goto out;
2684 }
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);
2689 ASSERT(node->root);
2690 list_add_tail(&node->list, &rc->backref_cache.changed);
2691 } else {
2692 path->lowest_level = node->level;
2693 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2694 btrfs_release_path(path);
2695 if (ret > 0)
2696 ret = 0;
2697 }
2698 if (!ret)
2699 update_processed_blocks(rc, node);
2700 } else {
2701 ret = do_relocation(trans, rc, node, key, path, 1);
2702 }
2703out:
2704 if (ret || node->level == 0 || node->cowonly)
2705 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2706 return ret;
2707}
2708
2709/*
2710 * relocate a list of blocks
2711 */
2712static noinline_for_stack
2713int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2714 struct reloc_control *rc, struct rb_root *blocks)
2715{
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;
2721 int ret;
2722 int err = 0;
2723
2724 path = btrfs_alloc_path();
2725 if (!path) {
2726 err = -ENOMEM;
2727 goto out_free_blocks;
2728 }
2729
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,
2734 block->owner, 0,
2735 block->level);
2736 }
2737
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);
2742 if (err)
2743 goto out_free_path;
2744 }
2745 }
2746
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);
2751 if (IS_ERR(node)) {
2752 err = PTR_ERR(node);
2753 goto out;
2754 }
2755
2756 ret = relocate_tree_block(trans, rc, node, &block->key,
2757 path);
2758 if (ret < 0) {
2759 err = ret;
2760 break;
2761 }
2762 }
2763out:
2764 err = finish_pending_nodes(trans, rc, path, err);
2765
2766out_free_path:
2767 btrfs_free_path(path);
2768out_free_blocks:
2769 free_block_list(blocks);
2770 return err;
2771}
2772
2773static noinline_for_stack int prealloc_file_extent_cluster(
2774 struct btrfs_inode *inode,
2775 struct file_extent_cluster *cluster)
2776{
2777 u64 alloc_hint = 0;
2778 u64 start;
2779 u64 end;
2780 u64 offset = inode->index_cnt;
2781 u64 num_bytes;
2782 int nr;
2783 int ret = 0;
2784 u64 prealloc_start = cluster->start - offset;
2785 u64 prealloc_end = cluster->end - offset;
2786 u64 cur_offset = prealloc_start;
2787
2788 BUG_ON(cluster->start != cluster->boundary[0]);
2789 ret = btrfs_alloc_data_chunk_ondemand(inode,
2790 prealloc_end + 1 - prealloc_start);
2791 if (ret)
2792 return ret;
2793
2794 /*
2795 * On a zoned filesystem, we cannot preallocate the file region.
2796 * Instead, we dirty and fiemap_write the region.
2797 */
2798 if (btrfs_is_zoned(inode->root->fs_info)) {
2799 struct btrfs_root *root = inode->root;
2800 struct btrfs_trans_handle *trans;
2801
2802 end = cluster->end - offset + 1;
2803 trans = btrfs_start_transaction(root, 1);
2804 if (IS_ERR(trans))
2805 return PTR_ERR(trans);
2806
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);
2810 if (ret) {
2811 btrfs_abort_transaction(trans, ret);
2812 btrfs_end_transaction(trans);
2813 return ret;
2814 }
2815
2816 return btrfs_end_transaction(trans);
2817 }
2818
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;
2824 else
2825 end = cluster->end - offset;
2826
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);
2834 if (ret)
2835 break;
2836 }
2837 btrfs_inode_unlock(&inode->vfs_inode, 0);
2838
2839 if (cur_offset < prealloc_end)
2840 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2841 prealloc_end + 1 - cur_offset);
2842 return ret;
2843}
2844
2845static noinline_for_stack
2846int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2847 u64 block_start)
2848{
2849 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2850 struct extent_map *em;
2851 int ret = 0;
2852
2853 em = alloc_extent_map();
2854 if (!em)
2855 return -ENOMEM;
2856
2857 em->start = start;
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);
2862
2863 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2864 while (1) {
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);
2870 break;
2871 }
2872 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2873 }
2874 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2875 return ret;
2876}
2877
2878/*
2879 * Allow error injection to test balance/relocation cancellation
2880 */
2881noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2882{
2883 return atomic_read(&fs_info->balance_cancel_req) ||
2884 atomic_read(&fs_info->reloc_cancel_req) ||
2885 fatal_signal_pending(current);
2886}
2887ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2888
2889static int relocate_file_extent_cluster(struct inode *inode,
2890 struct file_extent_cluster *cluster)
2891{
2892 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2893 u64 page_start;
2894 u64 page_end;
2895 u64 offset = BTRFS_I(inode)->index_cnt;
2896 unsigned long index;
2897 unsigned long last_index;
2898 struct page *page;
2899 struct file_ra_state *ra;
2900 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2901 int nr = 0;
2902 int ret = 0;
2903
2904 if (!cluster->nr)
2905 return 0;
2906
2907 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2908 if (!ra)
2909 return -ENOMEM;
2910
2911 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2912 if (ret)
2913 goto out;
2914
2915 file_ra_state_init(ra, inode->i_mapping);
2916
2917 ret = setup_extent_mapping(inode, cluster->start - offset,
2918 cluster->end - offset, cluster->start);
2919 if (ret)
2920 goto out;
2921
2922 index = (cluster->start - offset) >> PAGE_SHIFT;
2923 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2924 while (index <= last_index) {
2925 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2926 PAGE_SIZE);
2927 if (ret)
2928 goto out;
2929
2930 page = find_lock_page(inode->i_mapping, index);
2931 if (!page) {
2932 page_cache_sync_readahead(inode->i_mapping,
2933 ra, NULL, index,
2934 last_index + 1 - index);
2935 page = find_or_create_page(inode->i_mapping, index,
2936 mask);
2937 if (!page) {
2938 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2939 PAGE_SIZE, true);
2940 btrfs_delalloc_release_extents(BTRFS_I(inode),
2941 PAGE_SIZE);
2942 ret = -ENOMEM;
2943 goto out;
2944 }
2945 }
2946 ret = set_page_extent_mapped(page);
2947 if (ret < 0) {
2948 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2949 PAGE_SIZE, true);
2950 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2951 unlock_page(page);
2952 put_page(page);
2953 goto out;
2954 }
2955
2956 if (PageReadahead(page)) {
2957 page_cache_async_readahead(inode->i_mapping,
2958 ra, NULL, page, index,
2959 last_index + 1 - index);
2960 }
2961
2962 if (!PageUptodate(page)) {
2963 btrfs_readpage(NULL, page);
2964 lock_page(page);
2965 if (!PageUptodate(page)) {
2966 unlock_page(page);
2967 put_page(page);
2968 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2969 PAGE_SIZE, true);
2970 btrfs_delalloc_release_extents(BTRFS_I(inode),
2971 PAGE_SIZE);
2972 ret = -EIO;
2973 goto out;
2974 }
2975 }
2976
2977 page_start = page_offset(page);
2978 page_end = page_start + PAGE_SIZE - 1;
2979
2980 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2981
2982 if (nr < cluster->nr &&
2983 page_start + offset == cluster->boundary[nr]) {
2984 set_extent_bits(&BTRFS_I(inode)->io_tree,
2985 page_start, page_end,
2986 EXTENT_BOUNDARY);
2987 nr++;
2988 }
2989
2990 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2991 page_end, 0, NULL);
2992 if (ret) {
2993 unlock_page(page);
2994 put_page(page);
2995 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2996 PAGE_SIZE, true);
2997 btrfs_delalloc_release_extents(BTRFS_I(inode),
2998 PAGE_SIZE);
2999
3000 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3001 page_start, page_end,
3002 EXTENT_LOCKED | EXTENT_BOUNDARY);
3003 goto out;
3004
3005 }
3006 set_page_dirty(page);
3007
3008 unlock_extent(&BTRFS_I(inode)->io_tree,
3009 page_start, page_end);
3010 unlock_page(page);
3011 put_page(page);
3012
3013 index++;
3014 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3015 balance_dirty_pages_ratelimited(inode->i_mapping);
3016 btrfs_throttle(fs_info);
3017 if (btrfs_should_cancel_balance(fs_info)) {
3018 ret = -ECANCELED;
3019 goto out;
3020 }
3021 }
3022 WARN_ON(nr != cluster->nr);
3023 if (btrfs_is_zoned(fs_info) && !ret)
3024 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
3025out:
3026 kfree(ra);
3027 return ret;
3028}
3029
3030static noinline_for_stack
3031int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3032 struct file_extent_cluster *cluster)
3033{
3034 int ret;
3035
3036 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3037 ret = relocate_file_extent_cluster(inode, cluster);
3038 if (ret)
3039 return ret;
3040 cluster->nr = 0;
3041 }
3042
3043 if (!cluster->nr)
3044 cluster->start = extent_key->objectid;
3045 else
3046 BUG_ON(cluster->nr >= MAX_EXTENTS);
3047 cluster->end = extent_key->objectid + extent_key->offset - 1;
3048 cluster->boundary[cluster->nr] = extent_key->objectid;
3049 cluster->nr++;
3050
3051 if (cluster->nr >= MAX_EXTENTS) {
3052 ret = relocate_file_extent_cluster(inode, cluster);
3053 if (ret)
3054 return ret;
3055 cluster->nr = 0;
3056 }
3057 return 0;
3058}
3059
3060/*
3061 * helper to add a tree block to the list.
3062 * the major work is getting the generation and level of the block
3063 */
3064static int add_tree_block(struct reloc_control *rc,
3065 struct btrfs_key *extent_key,
3066 struct btrfs_path *path,
3067 struct rb_root *blocks)
3068{
3069 struct extent_buffer *eb;
3070 struct btrfs_extent_item *ei;
3071 struct btrfs_tree_block_info *bi;
3072 struct tree_block *block;
3073 struct rb_node *rb_node;
3074 u32 item_size;
3075 int level = -1;
3076 u64 generation;
3077 u64 owner = 0;
3078
3079 eb = path->nodes[0];
3080 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3081
3082 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3083 item_size >= sizeof(*ei) + sizeof(*bi)) {
3084 unsigned long ptr = 0, end;
3085
3086 ei = btrfs_item_ptr(eb, path->slots[0],
3087 struct btrfs_extent_item);
3088 end = (unsigned long)ei + item_size;
3089 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3090 bi = (struct btrfs_tree_block_info *)(ei + 1);
3091 level = btrfs_tree_block_level(eb, bi);
3092 ptr = (unsigned long)(bi + 1);
3093 } else {
3094 level = (int)extent_key->offset;
3095 ptr = (unsigned long)(ei + 1);
3096 }
3097 generation = btrfs_extent_generation(eb, ei);
3098
3099 /*
3100 * We're reading random blocks without knowing their owner ahead
3101 * of time. This is ok most of the time, as all reloc roots and
3102 * fs roots have the same lock type. However normal trees do
3103 * not, and the only way to know ahead of time is to read the
3104 * inline ref offset. We know it's an fs root if
3105 *
3106 * 1. There's more than one ref.
3107 * 2. There's a SHARED_DATA_REF_KEY set.
3108 * 3. FULL_BACKREF is set on the flags.
3109 *
3110 * Otherwise it's safe to assume that the ref offset == the
3111 * owner of this block, so we can use that when calling
3112 * read_tree_block.
3113 */
3114 if (btrfs_extent_refs(eb, ei) == 1 &&
3115 !(btrfs_extent_flags(eb, ei) &
3116 BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3117 ptr < end) {
3118 struct btrfs_extent_inline_ref *iref;
3119 int type;
3120
3121 iref = (struct btrfs_extent_inline_ref *)ptr;
3122 type = btrfs_get_extent_inline_ref_type(eb, iref,
3123 BTRFS_REF_TYPE_BLOCK);
3124 if (type == BTRFS_REF_TYPE_INVALID)
3125 return -EINVAL;
3126 if (type == BTRFS_TREE_BLOCK_REF_KEY)
3127 owner = btrfs_extent_inline_ref_offset(eb, iref);
3128 }
3129 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3130 btrfs_print_v0_err(eb->fs_info);
3131 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3132 return -EINVAL;
3133 } else {
3134 BUG();
3135 }
3136
3137 btrfs_release_path(path);
3138
3139 BUG_ON(level == -1);
3140
3141 block = kmalloc(sizeof(*block), GFP_NOFS);
3142 if (!block)
3143 return -ENOMEM;
3144
3145 block->bytenr = extent_key->objectid;
3146 block->key.objectid = rc->extent_root->fs_info->nodesize;
3147 block->key.offset = generation;
3148 block->level = level;
3149 block->key_ready = 0;
3150 block->owner = owner;
3151
3152 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3153 if (rb_node)
3154 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3155 -EEXIST);
3156
3157 return 0;
3158}
3159
3160/*
3161 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3162 */
3163static int __add_tree_block(struct reloc_control *rc,
3164 u64 bytenr, u32 blocksize,
3165 struct rb_root *blocks)
3166{
3167 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3168 struct btrfs_path *path;
3169 struct btrfs_key key;
3170 int ret;
3171 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3172
3173 if (tree_block_processed(bytenr, rc))
3174 return 0;
3175
3176 if (rb_simple_search(blocks, bytenr))
3177 return 0;
3178
3179 path = btrfs_alloc_path();
3180 if (!path)
3181 return -ENOMEM;
3182again:
3183 key.objectid = bytenr;
3184 if (skinny) {
3185 key.type = BTRFS_METADATA_ITEM_KEY;
3186 key.offset = (u64)-1;
3187 } else {
3188 key.type = BTRFS_EXTENT_ITEM_KEY;
3189 key.offset = blocksize;
3190 }
3191
3192 path->search_commit_root = 1;
3193 path->skip_locking = 1;
3194 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3195 if (ret < 0)
3196 goto out;
3197
3198 if (ret > 0 && skinny) {
3199 if (path->slots[0]) {
3200 path->slots[0]--;
3201 btrfs_item_key_to_cpu(path->nodes[0], &key,
3202 path->slots[0]);
3203 if (key.objectid == bytenr &&
3204 (key.type == BTRFS_METADATA_ITEM_KEY ||
3205 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3206 key.offset == blocksize)))
3207 ret = 0;
3208 }
3209
3210 if (ret) {
3211 skinny = false;
3212 btrfs_release_path(path);
3213 goto again;
3214 }
3215 }
3216 if (ret) {
3217 ASSERT(ret == 1);
3218 btrfs_print_leaf(path->nodes[0]);
3219 btrfs_err(fs_info,
3220 "tree block extent item (%llu) is not found in extent tree",
3221 bytenr);
3222 WARN_ON(1);
3223 ret = -EINVAL;
3224 goto out;
3225 }
3226
3227 ret = add_tree_block(rc, &key, path, blocks);
3228out:
3229 btrfs_free_path(path);
3230 return ret;
3231}
3232
3233static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3234 struct btrfs_block_group *block_group,
3235 struct inode *inode,
3236 u64 ino)
3237{
3238 struct btrfs_root *root = fs_info->tree_root;
3239 struct btrfs_trans_handle *trans;
3240 int ret = 0;
3241
3242 if (inode)
3243 goto truncate;
3244
3245 inode = btrfs_iget(fs_info->sb, ino, root);
3246 if (IS_ERR(inode))
3247 return -ENOENT;
3248
3249truncate:
3250 ret = btrfs_check_trunc_cache_free_space(fs_info,
3251 &fs_info->global_block_rsv);
3252 if (ret)
3253 goto out;
3254
3255 trans = btrfs_join_transaction(root);
3256 if (IS_ERR(trans)) {
3257 ret = PTR_ERR(trans);
3258 goto out;
3259 }
3260
3261 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3262
3263 btrfs_end_transaction(trans);
3264 btrfs_btree_balance_dirty(fs_info);
3265out:
3266 iput(inode);
3267 return ret;
3268}
3269
3270/*
3271 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3272 * cache inode, to avoid free space cache data extent blocking data relocation.
3273 */
3274static int delete_v1_space_cache(struct extent_buffer *leaf,
3275 struct btrfs_block_group *block_group,
3276 u64 data_bytenr)
3277{
3278 u64 space_cache_ino;
3279 struct btrfs_file_extent_item *ei;
3280 struct btrfs_key key;
3281 bool found = false;
3282 int i;
3283 int ret;
3284
3285 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3286 return 0;
3287
3288 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3289 u8 type;
3290
3291 btrfs_item_key_to_cpu(leaf, &key, i);
3292 if (key.type != BTRFS_EXTENT_DATA_KEY)
3293 continue;
3294 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3295 type = btrfs_file_extent_type(leaf, ei);
3296
3297 if ((type == BTRFS_FILE_EXTENT_REG ||
3298 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3299 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3300 found = true;
3301 space_cache_ino = key.objectid;
3302 break;
3303 }
3304 }
3305 if (!found)
3306 return -ENOENT;
3307 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3308 space_cache_ino);
3309 return ret;
3310}
3311
3312/*
3313 * helper to find all tree blocks that reference a given data extent
3314 */
3315static noinline_for_stack
3316int add_data_references(struct reloc_control *rc,
3317 struct btrfs_key *extent_key,
3318 struct btrfs_path *path,
3319 struct rb_root *blocks)
3320{
3321 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3322 struct ulist *leaves = NULL;
3323 struct ulist_iterator leaf_uiter;
3324 struct ulist_node *ref_node = NULL;
3325 const u32 blocksize = fs_info->nodesize;
3326 int ret = 0;
3327
3328 btrfs_release_path(path);
3329 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3330 0, &leaves, NULL, true);
3331 if (ret < 0)
3332 return ret;
3333
3334 ULIST_ITER_INIT(&leaf_uiter);
3335 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3336 struct extent_buffer *eb;
3337
3338 eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL);
3339 if (IS_ERR(eb)) {
3340 ret = PTR_ERR(eb);
3341 break;
3342 }
3343 ret = delete_v1_space_cache(eb, rc->block_group,
3344 extent_key->objectid);
3345 free_extent_buffer(eb);
3346 if (ret < 0)
3347 break;
3348 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3349 if (ret < 0)
3350 break;
3351 }
3352 if (ret < 0)
3353 free_block_list(blocks);
3354 ulist_free(leaves);
3355 return ret;
3356}
3357
3358/*
3359 * helper to find next unprocessed extent
3360 */
3361static noinline_for_stack
3362int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3363 struct btrfs_key *extent_key)
3364{
3365 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3366 struct btrfs_key key;
3367 struct extent_buffer *leaf;
3368 u64 start, end, last;
3369 int ret;
3370
3371 last = rc->block_group->start + rc->block_group->length;
3372 while (1) {
3373 cond_resched();
3374 if (rc->search_start >= last) {
3375 ret = 1;
3376 break;
3377 }
3378
3379 key.objectid = rc->search_start;
3380 key.type = BTRFS_EXTENT_ITEM_KEY;
3381 key.offset = 0;
3382
3383 path->search_commit_root = 1;
3384 path->skip_locking = 1;
3385 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3386 0, 0);
3387 if (ret < 0)
3388 break;
3389next:
3390 leaf = path->nodes[0];
3391 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3392 ret = btrfs_next_leaf(rc->extent_root, path);
3393 if (ret != 0)
3394 break;
3395 leaf = path->nodes[0];
3396 }
3397
3398 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3399 if (key.objectid >= last) {
3400 ret = 1;
3401 break;
3402 }
3403
3404 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3405 key.type != BTRFS_METADATA_ITEM_KEY) {
3406 path->slots[0]++;
3407 goto next;
3408 }
3409
3410 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3411 key.objectid + key.offset <= rc->search_start) {
3412 path->slots[0]++;
3413 goto next;
3414 }
3415
3416 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3417 key.objectid + fs_info->nodesize <=
3418 rc->search_start) {
3419 path->slots[0]++;
3420 goto next;
3421 }
3422
3423 ret = find_first_extent_bit(&rc->processed_blocks,
3424 key.objectid, &start, &end,
3425 EXTENT_DIRTY, NULL);
3426
3427 if (ret == 0 && start <= key.objectid) {
3428 btrfs_release_path(path);
3429 rc->search_start = end + 1;
3430 } else {
3431 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3432 rc->search_start = key.objectid + key.offset;
3433 else
3434 rc->search_start = key.objectid +
3435 fs_info->nodesize;
3436 memcpy(extent_key, &key, sizeof(key));
3437 return 0;
3438 }
3439 }
3440 btrfs_release_path(path);
3441 return ret;
3442}
3443
3444static void set_reloc_control(struct reloc_control *rc)
3445{
3446 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3447
3448 mutex_lock(&fs_info->reloc_mutex);
3449 fs_info->reloc_ctl = rc;
3450 mutex_unlock(&fs_info->reloc_mutex);
3451}
3452
3453static void unset_reloc_control(struct reloc_control *rc)
3454{
3455 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3456
3457 mutex_lock(&fs_info->reloc_mutex);
3458 fs_info->reloc_ctl = NULL;
3459 mutex_unlock(&fs_info->reloc_mutex);
3460}
3461
3462static noinline_for_stack
3463int prepare_to_relocate(struct reloc_control *rc)
3464{
3465 struct btrfs_trans_handle *trans;
3466 int ret;
3467
3468 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3469 BTRFS_BLOCK_RSV_TEMP);
3470 if (!rc->block_rsv)
3471 return -ENOMEM;
3472
3473 memset(&rc->cluster, 0, sizeof(rc->cluster));
3474 rc->search_start = rc->block_group->start;
3475 rc->extents_found = 0;
3476 rc->nodes_relocated = 0;
3477 rc->merging_rsv_size = 0;
3478 rc->reserved_bytes = 0;
3479 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3480 RELOCATION_RESERVED_NODES;
3481 ret = btrfs_block_rsv_refill(rc->extent_root,
3482 rc->block_rsv, rc->block_rsv->size,
3483 BTRFS_RESERVE_FLUSH_ALL);
3484 if (ret)
3485 return ret;
3486
3487 rc->create_reloc_tree = 1;
3488 set_reloc_control(rc);
3489
3490 trans = btrfs_join_transaction(rc->extent_root);
3491 if (IS_ERR(trans)) {
3492 unset_reloc_control(rc);
3493 /*
3494 * extent tree is not a ref_cow tree and has no reloc_root to
3495 * cleanup. And callers are responsible to free the above
3496 * block rsv.
3497 */
3498 return PTR_ERR(trans);
3499 }
3500 return btrfs_commit_transaction(trans);
3501}
3502
3503static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3504{
3505 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3506 struct rb_root blocks = RB_ROOT;
3507 struct btrfs_key key;
3508 struct btrfs_trans_handle *trans = NULL;
3509 struct btrfs_path *path;
3510 struct btrfs_extent_item *ei;
3511 u64 flags;
3512 int ret;
3513 int err = 0;
3514 int progress = 0;
3515
3516 path = btrfs_alloc_path();
3517 if (!path)
3518 return -ENOMEM;
3519 path->reada = READA_FORWARD;
3520
3521 ret = prepare_to_relocate(rc);
3522 if (ret) {
3523 err = ret;
3524 goto out_free;
3525 }
3526
3527 while (1) {
3528 rc->reserved_bytes = 0;
3529 ret = btrfs_block_rsv_refill(rc->extent_root,
3530 rc->block_rsv, rc->block_rsv->size,
3531 BTRFS_RESERVE_FLUSH_ALL);
3532 if (ret) {
3533 err = ret;
3534 break;
3535 }
3536 progress++;
3537 trans = btrfs_start_transaction(rc->extent_root, 0);
3538 if (IS_ERR(trans)) {
3539 err = PTR_ERR(trans);
3540 trans = NULL;
3541 break;
3542 }
3543restart:
3544 if (update_backref_cache(trans, &rc->backref_cache)) {
3545 btrfs_end_transaction(trans);
3546 trans = NULL;
3547 continue;
3548 }
3549
3550 ret = find_next_extent(rc, path, &key);
3551 if (ret < 0)
3552 err = ret;
3553 if (ret != 0)
3554 break;
3555
3556 rc->extents_found++;
3557
3558 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3559 struct btrfs_extent_item);
3560 flags = btrfs_extent_flags(path->nodes[0], ei);
3561
3562 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3563 ret = add_tree_block(rc, &key, path, &blocks);
3564 } else if (rc->stage == UPDATE_DATA_PTRS &&
3565 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3566 ret = add_data_references(rc, &key, path, &blocks);
3567 } else {
3568 btrfs_release_path(path);
3569 ret = 0;
3570 }
3571 if (ret < 0) {
3572 err = ret;
3573 break;
3574 }
3575
3576 if (!RB_EMPTY_ROOT(&blocks)) {
3577 ret = relocate_tree_blocks(trans, rc, &blocks);
3578 if (ret < 0) {
3579 if (ret != -EAGAIN) {
3580 err = ret;
3581 break;
3582 }
3583 rc->extents_found--;
3584 rc->search_start = key.objectid;
3585 }
3586 }
3587
3588 btrfs_end_transaction_throttle(trans);
3589 btrfs_btree_balance_dirty(fs_info);
3590 trans = NULL;
3591
3592 if (rc->stage == MOVE_DATA_EXTENTS &&
3593 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3594 rc->found_file_extent = 1;
3595 ret = relocate_data_extent(rc->data_inode,
3596 &key, &rc->cluster);
3597 if (ret < 0) {
3598 err = ret;
3599 break;
3600 }
3601 }
3602 if (btrfs_should_cancel_balance(fs_info)) {
3603 err = -ECANCELED;
3604 break;
3605 }
3606 }
3607 if (trans && progress && err == -ENOSPC) {
3608 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3609 if (ret == 1) {
3610 err = 0;
3611 progress = 0;
3612 goto restart;
3613 }
3614 }
3615
3616 btrfs_release_path(path);
3617 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3618
3619 if (trans) {
3620 btrfs_end_transaction_throttle(trans);
3621 btrfs_btree_balance_dirty(fs_info);
3622 }
3623
3624 if (!err) {
3625 ret = relocate_file_extent_cluster(rc->data_inode,
3626 &rc->cluster);
3627 if (ret < 0)
3628 err = ret;
3629 }
3630
3631 rc->create_reloc_tree = 0;
3632 set_reloc_control(rc);
3633
3634 btrfs_backref_release_cache(&rc->backref_cache);
3635 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3636
3637 /*
3638 * Even in the case when the relocation is cancelled, we should all go
3639 * through prepare_to_merge() and merge_reloc_roots().
3640 *
3641 * For error (including cancelled balance), prepare_to_merge() will
3642 * mark all reloc trees orphan, then queue them for cleanup in
3643 * merge_reloc_roots()
3644 */
3645 err = prepare_to_merge(rc, err);
3646
3647 merge_reloc_roots(rc);
3648
3649 rc->merge_reloc_tree = 0;
3650 unset_reloc_control(rc);
3651 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3652
3653 /* get rid of pinned extents */
3654 trans = btrfs_join_transaction(rc->extent_root);
3655 if (IS_ERR(trans)) {
3656 err = PTR_ERR(trans);
3657 goto out_free;
3658 }
3659 ret = btrfs_commit_transaction(trans);
3660 if (ret && !err)
3661 err = ret;
3662out_free:
3663 ret = clean_dirty_subvols(rc);
3664 if (ret < 0 && !err)
3665 err = ret;
3666 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3667 btrfs_free_path(path);
3668 return err;
3669}
3670
3671static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3672 struct btrfs_root *root, u64 objectid)
3673{
3674 struct btrfs_path *path;
3675 struct btrfs_inode_item *item;
3676 struct extent_buffer *leaf;
3677 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
3678 int ret;
3679
3680 if (btrfs_is_zoned(trans->fs_info))
3681 flags &= ~BTRFS_INODE_PREALLOC;
3682
3683 path = btrfs_alloc_path();
3684 if (!path)
3685 return -ENOMEM;
3686
3687 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3688 if (ret)
3689 goto out;
3690
3691 leaf = path->nodes[0];
3692 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3693 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3694 btrfs_set_inode_generation(leaf, item, 1);
3695 btrfs_set_inode_size(leaf, item, 0);
3696 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3697 btrfs_set_inode_flags(leaf, item, flags);
3698 btrfs_mark_buffer_dirty(leaf);
3699out:
3700 btrfs_free_path(path);
3701 return ret;
3702}
3703
3704static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3705 struct btrfs_root *root, u64 objectid)
3706{
3707 struct btrfs_path *path;
3708 struct btrfs_key key;
3709 int ret = 0;
3710
3711 path = btrfs_alloc_path();
3712 if (!path) {
3713 ret = -ENOMEM;
3714 goto out;
3715 }
3716
3717 key.objectid = objectid;
3718 key.type = BTRFS_INODE_ITEM_KEY;
3719 key.offset = 0;
3720 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3721 if (ret) {
3722 if (ret > 0)
3723 ret = -ENOENT;
3724 goto out;
3725 }
3726 ret = btrfs_del_item(trans, root, path);
3727out:
3728 if (ret)
3729 btrfs_abort_transaction(trans, ret);
3730 btrfs_free_path(path);
3731}
3732
3733/*
3734 * helper to create inode for data relocation.
3735 * the inode is in data relocation tree and its link count is 0
3736 */
3737static noinline_for_stack
3738struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3739 struct btrfs_block_group *group)
3740{
3741 struct inode *inode = NULL;
3742 struct btrfs_trans_handle *trans;
3743 struct btrfs_root *root;
3744 u64 objectid;
3745 int err = 0;
3746
3747 root = btrfs_grab_root(fs_info->data_reloc_root);
3748 trans = btrfs_start_transaction(root, 6);
3749 if (IS_ERR(trans)) {
3750 btrfs_put_root(root);
3751 return ERR_CAST(trans);
3752 }
3753
3754 err = btrfs_get_free_objectid(root, &objectid);
3755 if (err)
3756 goto out;
3757
3758 err = __insert_orphan_inode(trans, root, objectid);
3759 if (err)
3760 goto out;
3761
3762 inode = btrfs_iget(fs_info->sb, objectid, root);
3763 if (IS_ERR(inode)) {
3764 delete_orphan_inode(trans, root, objectid);
3765 err = PTR_ERR(inode);
3766 inode = NULL;
3767 goto out;
3768 }
3769 BTRFS_I(inode)->index_cnt = group->start;
3770
3771 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3772out:
3773 btrfs_put_root(root);
3774 btrfs_end_transaction(trans);
3775 btrfs_btree_balance_dirty(fs_info);
3776 if (err) {
3777 if (inode)
3778 iput(inode);
3779 inode = ERR_PTR(err);
3780 }
3781 return inode;
3782}
3783
3784/*
3785 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3786 * has been requested meanwhile and don't start in that case.
3787 *
3788 * Return:
3789 * 0 success
3790 * -EINPROGRESS operation is already in progress, that's probably a bug
3791 * -ECANCELED cancellation request was set before the operation started
3792 * -EAGAIN can not start because there are ongoing send operations
3793 */
3794static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3795{
3796 spin_lock(&fs_info->send_reloc_lock);
3797 if (fs_info->send_in_progress) {
3798 btrfs_warn_rl(fs_info,
3799"cannot run relocation while send operations are in progress (%d in progress)",
3800 fs_info->send_in_progress);
3801 spin_unlock(&fs_info->send_reloc_lock);
3802 return -EAGAIN;
3803 }
3804 if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3805 /* This should not happen */
3806 spin_unlock(&fs_info->send_reloc_lock);
3807 btrfs_err(fs_info, "reloc already running, cannot start");
3808 return -EINPROGRESS;
3809 }
3810 spin_unlock(&fs_info->send_reloc_lock);
3811
3812 if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3813 btrfs_info(fs_info, "chunk relocation canceled on start");
3814 /*
3815 * On cancel, clear all requests but let the caller mark
3816 * the end after cleanup operations.
3817 */
3818 atomic_set(&fs_info->reloc_cancel_req, 0);
3819 return -ECANCELED;
3820 }
3821 return 0;
3822}
3823
3824/*
3825 * Mark end of chunk relocation that is cancellable and wake any waiters.
3826 */
3827static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
3828{
3829 /* Requested after start, clear bit first so any waiters can continue */
3830 if (atomic_read(&fs_info->reloc_cancel_req) > 0)
3831 btrfs_info(fs_info, "chunk relocation canceled during operation");
3832 spin_lock(&fs_info->send_reloc_lock);
3833 clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
3834 spin_unlock(&fs_info->send_reloc_lock);
3835 atomic_set(&fs_info->reloc_cancel_req, 0);
3836}
3837
3838static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3839{
3840 struct reloc_control *rc;
3841
3842 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3843 if (!rc)
3844 return NULL;
3845
3846 INIT_LIST_HEAD(&rc->reloc_roots);
3847 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3848 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3849 mapping_tree_init(&rc->reloc_root_tree);
3850 extent_io_tree_init(fs_info, &rc->processed_blocks,
3851 IO_TREE_RELOC_BLOCKS, NULL);
3852 return rc;
3853}
3854
3855static void free_reloc_control(struct reloc_control *rc)
3856{
3857 struct mapping_node *node, *tmp;
3858
3859 free_reloc_roots(&rc->reloc_roots);
3860 rbtree_postorder_for_each_entry_safe(node, tmp,
3861 &rc->reloc_root_tree.rb_root, rb_node)
3862 kfree(node);
3863
3864 kfree(rc);
3865}
3866
3867/*
3868 * Print the block group being relocated
3869 */
3870static void describe_relocation(struct btrfs_fs_info *fs_info,
3871 struct btrfs_block_group *block_group)
3872{
3873 char buf[128] = {'\0'};
3874
3875 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3876
3877 btrfs_info(fs_info,
3878 "relocating block group %llu flags %s",
3879 block_group->start, buf);
3880}
3881
3882static const char *stage_to_string(int stage)
3883{
3884 if (stage == MOVE_DATA_EXTENTS)
3885 return "move data extents";
3886 if (stage == UPDATE_DATA_PTRS)
3887 return "update data pointers";
3888 return "unknown";
3889}
3890
3891/*
3892 * function to relocate all extents in a block group.
3893 */
3894int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3895{
3896 struct btrfs_block_group *bg;
3897 struct btrfs_root *extent_root = fs_info->extent_root;
3898 struct reloc_control *rc;
3899 struct inode *inode;
3900 struct btrfs_path *path;
3901 int ret;
3902 int rw = 0;
3903 int err = 0;
3904
3905 bg = btrfs_lookup_block_group(fs_info, group_start);
3906 if (!bg)
3907 return -ENOENT;
3908
3909 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3910 btrfs_put_block_group(bg);
3911 return -ETXTBSY;
3912 }
3913
3914 rc = alloc_reloc_control(fs_info);
3915 if (!rc) {
3916 btrfs_put_block_group(bg);
3917 return -ENOMEM;
3918 }
3919
3920 ret = reloc_chunk_start(fs_info);
3921 if (ret < 0) {
3922 err = ret;
3923 goto out_put_bg;
3924 }
3925
3926 rc->extent_root = extent_root;
3927 rc->block_group = bg;
3928
3929 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3930 if (ret) {
3931 err = ret;
3932 goto out;
3933 }
3934 rw = 1;
3935
3936 path = btrfs_alloc_path();
3937 if (!path) {
3938 err = -ENOMEM;
3939 goto out;
3940 }
3941
3942 inode = lookup_free_space_inode(rc->block_group, path);
3943 btrfs_free_path(path);
3944
3945 if (!IS_ERR(inode))
3946 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3947 else
3948 ret = PTR_ERR(inode);
3949
3950 if (ret && ret != -ENOENT) {
3951 err = ret;
3952 goto out;
3953 }
3954
3955 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3956 if (IS_ERR(rc->data_inode)) {
3957 err = PTR_ERR(rc->data_inode);
3958 rc->data_inode = NULL;
3959 goto out;
3960 }
3961
3962 describe_relocation(fs_info, rc->block_group);
3963
3964 btrfs_wait_block_group_reservations(rc->block_group);
3965 btrfs_wait_nocow_writers(rc->block_group);
3966 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3967 rc->block_group->start,
3968 rc->block_group->length);
3969
3970 while (1) {
3971 int finishes_stage;
3972
3973 mutex_lock(&fs_info->cleaner_mutex);
3974 ret = relocate_block_group(rc);
3975 mutex_unlock(&fs_info->cleaner_mutex);
3976 if (ret < 0)
3977 err = ret;
3978
3979 finishes_stage = rc->stage;
3980 /*
3981 * We may have gotten ENOSPC after we already dirtied some
3982 * extents. If writeout happens while we're relocating a
3983 * different block group we could end up hitting the
3984 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3985 * btrfs_reloc_cow_block. Make sure we write everything out
3986 * properly so we don't trip over this problem, and then break
3987 * out of the loop if we hit an error.
3988 */
3989 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3990 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3991 (u64)-1);
3992 if (ret)
3993 err = ret;
3994 invalidate_mapping_pages(rc->data_inode->i_mapping,
3995 0, -1);
3996 rc->stage = UPDATE_DATA_PTRS;
3997 }
3998
3999 if (err < 0)
4000 goto out;
4001
4002 if (rc->extents_found == 0)
4003 break;
4004
4005 btrfs_info(fs_info, "found %llu extents, stage: %s",
4006 rc->extents_found, stage_to_string(finishes_stage));
4007 }
4008
4009 WARN_ON(rc->block_group->pinned > 0);
4010 WARN_ON(rc->block_group->reserved > 0);
4011 WARN_ON(rc->block_group->used > 0);
4012out:
4013 if (err && rw)
4014 btrfs_dec_block_group_ro(rc->block_group);
4015 iput(rc->data_inode);
4016out_put_bg:
4017 btrfs_put_block_group(bg);
4018 reloc_chunk_end(fs_info);
4019 free_reloc_control(rc);
4020 return err;
4021}
4022
4023static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4024{
4025 struct btrfs_fs_info *fs_info = root->fs_info;
4026 struct btrfs_trans_handle *trans;
4027 int ret, err;
4028
4029 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4030 if (IS_ERR(trans))
4031 return PTR_ERR(trans);
4032
4033 memset(&root->root_item.drop_progress, 0,
4034 sizeof(root->root_item.drop_progress));
4035 btrfs_set_root_drop_level(&root->root_item, 0);
4036 btrfs_set_root_refs(&root->root_item, 0);
4037 ret = btrfs_update_root(trans, fs_info->tree_root,
4038 &root->root_key, &root->root_item);
4039
4040 err = btrfs_end_transaction(trans);
4041 if (err)
4042 return err;
4043 return ret;
4044}
4045
4046/*
4047 * recover relocation interrupted by system crash.
4048 *
4049 * this function resumes merging reloc trees with corresponding fs trees.
4050 * this is important for keeping the sharing of tree blocks
4051 */
4052int btrfs_recover_relocation(struct btrfs_root *root)
4053{
4054 struct btrfs_fs_info *fs_info = root->fs_info;
4055 LIST_HEAD(reloc_roots);
4056 struct btrfs_key key;
4057 struct btrfs_root *fs_root;
4058 struct btrfs_root *reloc_root;
4059 struct btrfs_path *path;
4060 struct extent_buffer *leaf;
4061 struct reloc_control *rc = NULL;
4062 struct btrfs_trans_handle *trans;
4063 int ret;
4064 int err = 0;
4065
4066 path = btrfs_alloc_path();
4067 if (!path)
4068 return -ENOMEM;
4069 path->reada = READA_BACK;
4070
4071 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4072 key.type = BTRFS_ROOT_ITEM_KEY;
4073 key.offset = (u64)-1;
4074
4075 while (1) {
4076 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4077 path, 0, 0);
4078 if (ret < 0) {
4079 err = ret;
4080 goto out;
4081 }
4082 if (ret > 0) {
4083 if (path->slots[0] == 0)
4084 break;
4085 path->slots[0]--;
4086 }
4087 leaf = path->nodes[0];
4088 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4089 btrfs_release_path(path);
4090
4091 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4092 key.type != BTRFS_ROOT_ITEM_KEY)
4093 break;
4094
4095 reloc_root = btrfs_read_tree_root(root, &key);
4096 if (IS_ERR(reloc_root)) {
4097 err = PTR_ERR(reloc_root);
4098 goto out;
4099 }
4100
4101 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4102 list_add(&reloc_root->root_list, &reloc_roots);
4103
4104 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4105 fs_root = btrfs_get_fs_root(fs_info,
4106 reloc_root->root_key.offset, false);
4107 if (IS_ERR(fs_root)) {
4108 ret = PTR_ERR(fs_root);
4109 if (ret != -ENOENT) {
4110 err = ret;
4111 goto out;
4112 }
4113 ret = mark_garbage_root(reloc_root);
4114 if (ret < 0) {
4115 err = ret;
4116 goto out;
4117 }
4118 } else {
4119 btrfs_put_root(fs_root);
4120 }
4121 }
4122
4123 if (key.offset == 0)
4124 break;
4125
4126 key.offset--;
4127 }
4128 btrfs_release_path(path);
4129
4130 if (list_empty(&reloc_roots))
4131 goto out;
4132
4133 rc = alloc_reloc_control(fs_info);
4134 if (!rc) {
4135 err = -ENOMEM;
4136 goto out;
4137 }
4138
4139 ret = reloc_chunk_start(fs_info);
4140 if (ret < 0) {
4141 err = ret;
4142 goto out_end;
4143 }
4144
4145 rc->extent_root = fs_info->extent_root;
4146
4147 set_reloc_control(rc);
4148
4149 trans = btrfs_join_transaction(rc->extent_root);
4150 if (IS_ERR(trans)) {
4151 err = PTR_ERR(trans);
4152 goto out_unset;
4153 }
4154
4155 rc->merge_reloc_tree = 1;
4156
4157 while (!list_empty(&reloc_roots)) {
4158 reloc_root = list_entry(reloc_roots.next,
4159 struct btrfs_root, root_list);
4160 list_del(&reloc_root->root_list);
4161
4162 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4163 list_add_tail(&reloc_root->root_list,
4164 &rc->reloc_roots);
4165 continue;
4166 }
4167
4168 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4169 false);
4170 if (IS_ERR(fs_root)) {
4171 err = PTR_ERR(fs_root);
4172 list_add_tail(&reloc_root->root_list, &reloc_roots);
4173 btrfs_end_transaction(trans);
4174 goto out_unset;
4175 }
4176
4177 err = __add_reloc_root(reloc_root);
4178 ASSERT(err != -EEXIST);
4179 if (err) {
4180 list_add_tail(&reloc_root->root_list, &reloc_roots);
4181 btrfs_put_root(fs_root);
4182 btrfs_end_transaction(trans);
4183 goto out_unset;
4184 }
4185 fs_root->reloc_root = btrfs_grab_root(reloc_root);
4186 btrfs_put_root(fs_root);
4187 }
4188
4189 err = btrfs_commit_transaction(trans);
4190 if (err)
4191 goto out_unset;
4192
4193 merge_reloc_roots(rc);
4194
4195 unset_reloc_control(rc);
4196
4197 trans = btrfs_join_transaction(rc->extent_root);
4198 if (IS_ERR(trans)) {
4199 err = PTR_ERR(trans);
4200 goto out_clean;
4201 }
4202 err = btrfs_commit_transaction(trans);
4203out_clean:
4204 ret = clean_dirty_subvols(rc);
4205 if (ret < 0 && !err)
4206 err = ret;
4207out_unset:
4208 unset_reloc_control(rc);
4209out_end:
4210 reloc_chunk_end(fs_info);
4211 free_reloc_control(rc);
4212out:
4213 free_reloc_roots(&reloc_roots);
4214
4215 btrfs_free_path(path);
4216
4217 if (err == 0) {
4218 /* cleanup orphan inode in data relocation tree */
4219 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4220 ASSERT(fs_root);
4221 err = btrfs_orphan_cleanup(fs_root);
4222 btrfs_put_root(fs_root);
4223 }
4224 return err;
4225}
4226
4227/*
4228 * helper to add ordered checksum for data relocation.
4229 *
4230 * cloning checksum properly handles the nodatasum extents.
4231 * it also saves CPU time to re-calculate the checksum.
4232 */
4233int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
4234{
4235 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4236 struct btrfs_ordered_sum *sums;
4237 struct btrfs_ordered_extent *ordered;
4238 int ret;
4239 u64 disk_bytenr;
4240 u64 new_bytenr;
4241 LIST_HEAD(list);
4242
4243 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4244 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
4245
4246 disk_bytenr = file_pos + inode->index_cnt;
4247 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4248 disk_bytenr + len - 1, &list, 0);
4249 if (ret)
4250 goto out;
4251
4252 while (!list_empty(&list)) {
4253 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4254 list_del_init(&sums->list);
4255
4256 /*
4257 * We need to offset the new_bytenr based on where the csum is.
4258 * We need to do this because we will read in entire prealloc
4259 * extents but we may have written to say the middle of the
4260 * prealloc extent, so we need to make sure the csum goes with
4261 * the right disk offset.
4262 *
4263 * We can do this because the data reloc inode refers strictly
4264 * to the on disk bytes, so we don't have to worry about
4265 * disk_len vs real len like with real inodes since it's all
4266 * disk length.
4267 */
4268 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
4269 sums->bytenr = new_bytenr;
4270
4271 btrfs_add_ordered_sum(ordered, sums);
4272 }
4273out:
4274 btrfs_put_ordered_extent(ordered);
4275 return ret;
4276}
4277
4278int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *root, struct extent_buffer *buf,
4280 struct extent_buffer *cow)
4281{
4282 struct btrfs_fs_info *fs_info = root->fs_info;
4283 struct reloc_control *rc;
4284 struct btrfs_backref_node *node;
4285 int first_cow = 0;
4286 int level;
4287 int ret = 0;
4288
4289 rc = fs_info->reloc_ctl;
4290 if (!rc)
4291 return 0;
4292
4293 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4294 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4295
4296 level = btrfs_header_level(buf);
4297 if (btrfs_header_generation(buf) <=
4298 btrfs_root_last_snapshot(&root->root_item))
4299 first_cow = 1;
4300
4301 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4302 rc->create_reloc_tree) {
4303 WARN_ON(!first_cow && level == 0);
4304
4305 node = rc->backref_cache.path[level];
4306 BUG_ON(node->bytenr != buf->start &&
4307 node->new_bytenr != buf->start);
4308
4309 btrfs_backref_drop_node_buffer(node);
4310 atomic_inc(&cow->refs);
4311 node->eb = cow;
4312 node->new_bytenr = cow->start;
4313
4314 if (!node->pending) {
4315 list_move_tail(&node->list,
4316 &rc->backref_cache.pending[level]);
4317 node->pending = 1;
4318 }
4319
4320 if (first_cow)
4321 mark_block_processed(rc, node);
4322
4323 if (first_cow && level > 0)
4324 rc->nodes_relocated += buf->len;
4325 }
4326
4327 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4328 ret = replace_file_extents(trans, rc, root, cow);
4329 return ret;
4330}
4331
4332/*
4333 * called before creating snapshot. it calculates metadata reservation
4334 * required for relocating tree blocks in the snapshot
4335 */
4336void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4337 u64 *bytes_to_reserve)
4338{
4339 struct btrfs_root *root = pending->root;
4340 struct reloc_control *rc = root->fs_info->reloc_ctl;
4341
4342 if (!rc || !have_reloc_root(root))
4343 return;
4344
4345 if (!rc->merge_reloc_tree)
4346 return;
4347
4348 root = root->reloc_root;
4349 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4350 /*
4351 * relocation is in the stage of merging trees. the space
4352 * used by merging a reloc tree is twice the size of
4353 * relocated tree nodes in the worst case. half for cowing
4354 * the reloc tree, half for cowing the fs tree. the space
4355 * used by cowing the reloc tree will be freed after the
4356 * tree is dropped. if we create snapshot, cowing the fs
4357 * tree may use more space than it frees. so we need
4358 * reserve extra space.
4359 */
4360 *bytes_to_reserve += rc->nodes_relocated;
4361}
4362
4363/*
4364 * called after snapshot is created. migrate block reservation
4365 * and create reloc root for the newly created snapshot
4366 *
4367 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4368 * references held on the reloc_root, one for root->reloc_root and one for
4369 * rc->reloc_roots.
4370 */
4371int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4372 struct btrfs_pending_snapshot *pending)
4373{
4374 struct btrfs_root *root = pending->root;
4375 struct btrfs_root *reloc_root;
4376 struct btrfs_root *new_root;
4377 struct reloc_control *rc = root->fs_info->reloc_ctl;
4378 int ret;
4379
4380 if (!rc || !have_reloc_root(root))
4381 return 0;
4382
4383 rc = root->fs_info->reloc_ctl;
4384 rc->merging_rsv_size += rc->nodes_relocated;
4385
4386 if (rc->merge_reloc_tree) {
4387 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4388 rc->block_rsv,
4389 rc->nodes_relocated, true);
4390 if (ret)
4391 return ret;
4392 }
4393
4394 new_root = pending->snap;
4395 reloc_root = create_reloc_root(trans, root->reloc_root,
4396 new_root->root_key.objectid);
4397 if (IS_ERR(reloc_root))
4398 return PTR_ERR(reloc_root);
4399
4400 ret = __add_reloc_root(reloc_root);
4401 ASSERT(ret != -EEXIST);
4402 if (ret) {
4403 /* Pairs with create_reloc_root */
4404 btrfs_put_root(reloc_root);
4405 return ret;
4406 }
4407 new_root->reloc_root = btrfs_grab_root(reloc_root);
4408
4409 if (rc->create_reloc_tree)
4410 ret = clone_backref_node(trans, rc, root, reloc_root);
4411 return ret;
4412}