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 #include "subpage.h"
28 #include "zoned.h"
29 #include "inode-item.h"
30 #include "space-info.h"
31 #include "fs.h"
32 #include "accessors.h"
33 #include "extent-tree.h"
34 #include "root-tree.h"
35 #include "file-item.h"
36 #include "relocation.h"
37 #include "super.h"
38 #include "tree-checker.h"
39 #include "raid-stripe-tree.h"
40
41 /*
42 * Relocation overview
43 *
44 * [What does relocation do]
45 *
46 * The objective of relocation is to relocate all extents of the target block
47 * group to other block groups.
48 * This is utilized by resize (shrink only), profile converting, compacting
49 * space, or balance routine to spread chunks over devices.
50 *
51 * Before | After
52 * ------------------------------------------------------------------
53 * BG A: 10 data extents | BG A: deleted
54 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
55 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
56 *
57 * [How does relocation work]
58 *
59 * 1. Mark the target block group read-only
60 * New extents won't be allocated from the target block group.
61 *
62 * 2.1 Record each extent in the target block group
63 * To build a proper map of extents to be relocated.
64 *
65 * 2.2 Build data reloc tree and reloc trees
66 * Data reloc tree will contain an inode, recording all newly relocated
67 * data extents.
68 * There will be only one data reloc tree for one data block group.
69 *
70 * Reloc tree will be a special snapshot of its source tree, containing
71 * relocated tree blocks.
72 * Each tree referring to a tree block in target block group will get its
73 * reloc tree built.
74 *
75 * 2.3 Swap source tree with its corresponding reloc tree
76 * Each involved tree only refers to new extents after swap.
77 *
78 * 3. Cleanup reloc trees and data reloc tree.
79 * As old extents in the target block group are still referenced by reloc
80 * trees, we need to clean them up before really freeing the target block
81 * group.
82 *
83 * The main complexity is in steps 2.2 and 2.3.
84 *
85 * The entry point of relocation is relocate_block_group() function.
86 */
87
88 #define RELOCATION_RESERVED_NODES 256
89 /*
90 * map address of tree root to tree
91 */
92 struct mapping_node {
93 union {
94 /* Use rb_simple_node for search/insert */
95 struct {
96 struct rb_node rb_node;
97 u64 bytenr;
98 };
99
100 struct rb_simple_node simple_node;
101 };
102 void *data;
103 };
104
105 struct mapping_tree {
106 struct rb_root rb_root;
107 spinlock_t lock;
108 };
109
110 /*
111 * present a tree block to process
112 */
113 struct tree_block {
114 union {
115 /* Use rb_simple_node for search/insert */
116 struct {
117 struct rb_node rb_node;
118 u64 bytenr;
119 };
120
121 struct rb_simple_node simple_node;
122 };
123 u64 owner;
124 struct btrfs_key key;
125 u8 level;
126 bool key_ready;
127 };
128
129 #define MAX_EXTENTS 128
130
131 struct file_extent_cluster {
132 u64 start;
133 u64 end;
134 u64 boundary[MAX_EXTENTS];
135 unsigned int nr;
136 u64 owning_root;
137 };
138
139 /* Stages of data relocation. */
140 enum reloc_stage {
141 MOVE_DATA_EXTENTS,
142 UPDATE_DATA_PTRS
143 };
144
145 struct reloc_control {
146 /* block group to relocate */
147 struct btrfs_block_group *block_group;
148 /* extent tree */
149 struct btrfs_root *extent_root;
150 /* inode for moving data */
151 struct inode *data_inode;
152
153 struct btrfs_block_rsv *block_rsv;
154
155 struct btrfs_backref_cache backref_cache;
156
157 struct file_extent_cluster cluster;
158 /* tree blocks have been processed */
159 struct extent_io_tree processed_blocks;
160 /* map start of tree root to corresponding reloc tree */
161 struct mapping_tree reloc_root_tree;
162 /* list of reloc trees */
163 struct list_head reloc_roots;
164 /* list of subvolume trees that get relocated */
165 struct list_head dirty_subvol_roots;
166 /* size of metadata reservation for merging reloc trees */
167 u64 merging_rsv_size;
168 /* size of relocated tree nodes */
169 u64 nodes_relocated;
170 /* reserved size for block group relocation*/
171 u64 reserved_bytes;
172
173 u64 search_start;
174 u64 extents_found;
175
176 enum reloc_stage stage;
177 bool create_reloc_tree;
178 bool merge_reloc_tree;
179 bool found_file_extent;
180 };
181
mark_block_processed(struct reloc_control * rc,struct btrfs_backref_node * node)182 static void mark_block_processed(struct reloc_control *rc,
183 struct btrfs_backref_node *node)
184 {
185 u32 blocksize;
186
187 if (node->level == 0 ||
188 in_range(node->bytenr, rc->block_group->start,
189 rc->block_group->length)) {
190 blocksize = rc->extent_root->fs_info->nodesize;
191 btrfs_set_extent_bit(&rc->processed_blocks, node->bytenr,
192 node->bytenr + blocksize - 1, EXTENT_DIRTY,
193 NULL);
194 }
195 node->processed = 1;
196 }
197
198 /*
199 * walk up backref nodes until reach node presents tree root
200 */
walk_up_backref(struct btrfs_backref_node * node,struct btrfs_backref_edge * edges[],int * index)201 static struct btrfs_backref_node *walk_up_backref(
202 struct btrfs_backref_node *node,
203 struct btrfs_backref_edge *edges[], int *index)
204 {
205 struct btrfs_backref_edge *edge;
206 int idx = *index;
207
208 while (!list_empty(&node->upper)) {
209 edge = list_first_entry(&node->upper, struct btrfs_backref_edge,
210 list[LOWER]);
211 edges[idx++] = edge;
212 node = edge->node[UPPER];
213 }
214 BUG_ON(node->detached);
215 *index = idx;
216 return node;
217 }
218
219 /*
220 * walk down backref nodes to find start of next reference path
221 */
walk_down_backref(struct btrfs_backref_edge * edges[],int * index)222 static struct btrfs_backref_node *walk_down_backref(
223 struct btrfs_backref_edge *edges[], int *index)
224 {
225 struct btrfs_backref_edge *edge;
226 struct btrfs_backref_node *lower;
227 int idx = *index;
228
229 while (idx > 0) {
230 edge = edges[idx - 1];
231 lower = edge->node[LOWER];
232 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
233 idx--;
234 continue;
235 }
236 edge = list_first_entry(&edge->list[LOWER], struct btrfs_backref_edge,
237 list[LOWER]);
238 edges[idx - 1] = edge;
239 *index = idx;
240 return edge->node[UPPER];
241 }
242 *index = 0;
243 return NULL;
244 }
245
reloc_root_is_dead(const struct btrfs_root * root)246 static bool reloc_root_is_dead(const struct btrfs_root *root)
247 {
248 /*
249 * Pair with set_bit/clear_bit in clean_dirty_subvols and
250 * btrfs_update_reloc_root. We need to see the updated bit before
251 * trying to access reloc_root
252 */
253 smp_rmb();
254 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
255 return true;
256 return false;
257 }
258
259 /*
260 * Check if this subvolume tree has valid reloc tree.
261 *
262 * Reloc tree after swap is considered dead, thus not considered as valid.
263 * This is enough for most callers, as they don't distinguish dead reloc root
264 * from no reloc root. But btrfs_should_ignore_reloc_root() below is a
265 * special case.
266 */
have_reloc_root(const struct btrfs_root * root)267 static bool have_reloc_root(const struct btrfs_root *root)
268 {
269 if (reloc_root_is_dead(root))
270 return false;
271 if (!root->reloc_root)
272 return false;
273 return true;
274 }
275
btrfs_should_ignore_reloc_root(const struct btrfs_root * root)276 bool btrfs_should_ignore_reloc_root(const struct btrfs_root *root)
277 {
278 struct btrfs_root *reloc_root;
279
280 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
281 return false;
282
283 /* This root has been merged with its reloc tree, we can ignore it */
284 if (reloc_root_is_dead(root))
285 return true;
286
287 reloc_root = root->reloc_root;
288 if (!reloc_root)
289 return false;
290
291 if (btrfs_header_generation(reloc_root->commit_root) ==
292 root->fs_info->running_transaction->transid)
293 return false;
294 /*
295 * If there is reloc tree and it was created in previous transaction
296 * backref lookup can find the reloc tree, so backref node for the fs
297 * tree root is useless for relocation.
298 */
299 return true;
300 }
301
302 /*
303 * find reloc tree by address of tree root
304 */
find_reloc_root(struct btrfs_fs_info * fs_info,u64 bytenr)305 struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
306 {
307 struct reloc_control *rc = fs_info->reloc_ctl;
308 struct rb_node *rb_node;
309 struct mapping_node *node;
310 struct btrfs_root *root = NULL;
311
312 ASSERT(rc);
313 spin_lock(&rc->reloc_root_tree.lock);
314 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
315 if (rb_node) {
316 node = rb_entry(rb_node, struct mapping_node, rb_node);
317 root = node->data;
318 }
319 spin_unlock(&rc->reloc_root_tree.lock);
320 return btrfs_grab_root(root);
321 }
322
323 /*
324 * For useless nodes, do two major clean ups:
325 *
326 * - Cleanup the children edges and nodes
327 * If child node is also orphan (no parent) during cleanup, then the child
328 * node will also be cleaned up.
329 *
330 * - Freeing up leaves (level 0), keeps nodes detached
331 * For nodes, the node is still cached as "detached"
332 *
333 * Return false if @node is not in the @useless_nodes list.
334 * Return true if @node is in the @useless_nodes list.
335 */
handle_useless_nodes(struct reloc_control * rc,struct btrfs_backref_node * node)336 static bool handle_useless_nodes(struct reloc_control *rc,
337 struct btrfs_backref_node *node)
338 {
339 struct btrfs_backref_cache *cache = &rc->backref_cache;
340 struct list_head *useless_node = &cache->useless_node;
341 bool ret = false;
342
343 while (!list_empty(useless_node)) {
344 struct btrfs_backref_node *cur;
345
346 cur = list_first_entry(useless_node, struct btrfs_backref_node,
347 list);
348 list_del_init(&cur->list);
349
350 /* Only tree root nodes can be added to @useless_nodes */
351 ASSERT(list_empty(&cur->upper));
352
353 if (cur == node)
354 ret = true;
355
356 /* Cleanup the lower edges */
357 while (!list_empty(&cur->lower)) {
358 struct btrfs_backref_edge *edge;
359 struct btrfs_backref_node *lower;
360
361 edge = list_first_entry(&cur->lower, struct btrfs_backref_edge,
362 list[UPPER]);
363 list_del(&edge->list[UPPER]);
364 list_del(&edge->list[LOWER]);
365 lower = edge->node[LOWER];
366 btrfs_backref_free_edge(cache, edge);
367
368 /* Child node is also orphan, queue for cleanup */
369 if (list_empty(&lower->upper))
370 list_add(&lower->list, useless_node);
371 }
372 /* Mark this block processed for relocation */
373 mark_block_processed(rc, cur);
374
375 /*
376 * Backref nodes for tree leaves are deleted from the cache.
377 * Backref nodes for upper level tree blocks are left in the
378 * cache to avoid unnecessary backref lookup.
379 */
380 if (cur->level > 0) {
381 cur->detached = 1;
382 } else {
383 rb_erase(&cur->rb_node, &cache->rb_root);
384 btrfs_backref_free_node(cache, cur);
385 }
386 }
387 return ret;
388 }
389
390 /*
391 * Build backref tree for a given tree block. Root of the backref tree
392 * corresponds the tree block, leaves of the backref tree correspond roots of
393 * b-trees that reference the tree block.
394 *
395 * The basic idea of this function is check backrefs of a given block to find
396 * upper level blocks that reference the block, and then check backrefs of
397 * these upper level blocks recursively. The recursion stops when tree root is
398 * reached or backrefs for the block is cached.
399 *
400 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
401 * all upper level blocks that directly/indirectly reference the block are also
402 * cached.
403 */
build_backref_tree(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_key * node_key,int level,u64 bytenr)404 static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
405 struct btrfs_trans_handle *trans,
406 struct reloc_control *rc, struct btrfs_key *node_key,
407 int level, u64 bytenr)
408 {
409 struct btrfs_backref_iter *iter;
410 struct btrfs_backref_cache *cache = &rc->backref_cache;
411 /* For searching parent of TREE_BLOCK_REF */
412 struct btrfs_path *path;
413 struct btrfs_backref_node *cur;
414 struct btrfs_backref_node *node = NULL;
415 struct btrfs_backref_edge *edge;
416 int ret;
417
418 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info);
419 if (!iter)
420 return ERR_PTR(-ENOMEM);
421 path = btrfs_alloc_path();
422 if (!path) {
423 ret = -ENOMEM;
424 goto out;
425 }
426
427 node = btrfs_backref_alloc_node(cache, bytenr, level);
428 if (!node) {
429 ret = -ENOMEM;
430 goto out;
431 }
432
433 cur = node;
434
435 /* Breadth-first search to build backref cache */
436 do {
437 ret = btrfs_backref_add_tree_node(trans, cache, path, iter,
438 node_key, cur);
439 if (ret < 0)
440 goto out;
441
442 edge = list_first_entry_or_null(&cache->pending_edge,
443 struct btrfs_backref_edge, list[UPPER]);
444 /*
445 * The pending list isn't empty, take the first block to
446 * process
447 */
448 if (edge) {
449 list_del_init(&edge->list[UPPER]);
450 cur = edge->node[UPPER];
451 }
452 } while (edge);
453
454 /* Finish the upper linkage of newly added edges/nodes */
455 ret = btrfs_backref_finish_upper_links(cache, node);
456 if (ret < 0)
457 goto out;
458
459 if (handle_useless_nodes(rc, node))
460 node = NULL;
461 out:
462 btrfs_free_path(iter->path);
463 kfree(iter);
464 btrfs_free_path(path);
465 if (ret) {
466 btrfs_backref_error_cleanup(cache, node);
467 return ERR_PTR(ret);
468 }
469 ASSERT(!node || !node->detached);
470 ASSERT(list_empty(&cache->useless_node) &&
471 list_empty(&cache->pending_edge));
472 return node;
473 }
474
475 /*
476 * helper to add 'address of tree root -> reloc tree' mapping
477 */
__add_reloc_root(struct btrfs_root * root)478 static int __add_reloc_root(struct btrfs_root *root)
479 {
480 struct btrfs_fs_info *fs_info = root->fs_info;
481 struct rb_node *rb_node;
482 struct mapping_node *node;
483 struct reloc_control *rc = fs_info->reloc_ctl;
484
485 node = kmalloc(sizeof(*node), GFP_NOFS);
486 if (!node)
487 return -ENOMEM;
488
489 node->bytenr = root->commit_root->start;
490 node->data = root;
491
492 spin_lock(&rc->reloc_root_tree.lock);
493 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, &node->simple_node);
494 spin_unlock(&rc->reloc_root_tree.lock);
495 if (rb_node) {
496 btrfs_err(fs_info,
497 "Duplicate root found for start=%llu while inserting into relocation tree",
498 node->bytenr);
499 return -EEXIST;
500 }
501
502 list_add_tail(&root->root_list, &rc->reloc_roots);
503 return 0;
504 }
505
506 /*
507 * helper to delete the 'address of tree root -> reloc tree'
508 * mapping
509 */
__del_reloc_root(struct btrfs_root * root)510 static void __del_reloc_root(struct btrfs_root *root)
511 {
512 struct btrfs_fs_info *fs_info = root->fs_info;
513 struct rb_node *rb_node;
514 struct mapping_node *node = NULL;
515 struct reloc_control *rc = fs_info->reloc_ctl;
516 bool put_ref = false;
517
518 if (rc && root->node) {
519 spin_lock(&rc->reloc_root_tree.lock);
520 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
521 root->commit_root->start);
522 if (rb_node) {
523 node = rb_entry(rb_node, struct mapping_node, rb_node);
524 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
525 RB_CLEAR_NODE(&node->rb_node);
526 }
527 spin_unlock(&rc->reloc_root_tree.lock);
528 ASSERT(!node || (struct btrfs_root *)node->data == root);
529 }
530
531 /*
532 * We only put the reloc root here if it's on the list. There's a lot
533 * of places where the pattern is to splice the rc->reloc_roots, process
534 * the reloc roots, and then add the reloc root back onto
535 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
536 * list we don't want the reference being dropped, because the guy
537 * messing with the list is in charge of the reference.
538 */
539 spin_lock(&fs_info->trans_lock);
540 if (!list_empty(&root->root_list)) {
541 put_ref = true;
542 list_del_init(&root->root_list);
543 }
544 spin_unlock(&fs_info->trans_lock);
545 if (put_ref)
546 btrfs_put_root(root);
547 kfree(node);
548 }
549
550 /*
551 * helper to update the 'address of tree root -> reloc tree'
552 * mapping
553 */
__update_reloc_root(struct btrfs_root * root)554 static int __update_reloc_root(struct btrfs_root *root)
555 {
556 struct btrfs_fs_info *fs_info = root->fs_info;
557 struct rb_node *rb_node;
558 struct mapping_node *node = NULL;
559 struct reloc_control *rc = fs_info->reloc_ctl;
560
561 spin_lock(&rc->reloc_root_tree.lock);
562 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
563 root->commit_root->start);
564 if (rb_node) {
565 node = rb_entry(rb_node, struct mapping_node, rb_node);
566 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
567 }
568 spin_unlock(&rc->reloc_root_tree.lock);
569
570 if (!node)
571 return 0;
572 BUG_ON((struct btrfs_root *)node->data != root);
573
574 spin_lock(&rc->reloc_root_tree.lock);
575 node->bytenr = root->node->start;
576 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, &node->simple_node);
577 spin_unlock(&rc->reloc_root_tree.lock);
578 if (rb_node)
579 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
580 return 0;
581 }
582
create_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)583 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root, u64 objectid)
585 {
586 struct btrfs_fs_info *fs_info = root->fs_info;
587 struct btrfs_root *reloc_root;
588 struct extent_buffer *eb;
589 struct btrfs_root_item *root_item;
590 struct btrfs_key root_key;
591 int ret = 0;
592 bool must_abort = false;
593
594 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
595 if (!root_item)
596 return ERR_PTR(-ENOMEM);
597
598 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
599 root_key.type = BTRFS_ROOT_ITEM_KEY;
600 root_key.offset = objectid;
601
602 if (btrfs_root_id(root) == objectid) {
603 u64 commit_root_gen;
604
605 /*
606 * Relocation will wait for cleaner thread, and any half-dropped
607 * subvolume will be fully cleaned up at mount time.
608 * So here we shouldn't hit a subvolume with non-zero drop_progress.
609 *
610 * If this isn't the case, error out since it can make us attempt to
611 * drop references for extents that were already dropped before.
612 */
613 if (unlikely(btrfs_disk_key_objectid(&root->root_item.drop_progress))) {
614 struct btrfs_key cpu_key;
615
616 btrfs_disk_key_to_cpu(&cpu_key, &root->root_item.drop_progress);
617 btrfs_err(fs_info,
618 "cannot relocate partially dropped subvolume %llu, drop progress key (%llu %u %llu)",
619 objectid, cpu_key.objectid, cpu_key.type, cpu_key.offset);
620 ret = -EUCLEAN;
621 goto fail;
622 }
623
624 /* called by btrfs_init_reloc_root */
625 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
626 BTRFS_TREE_RELOC_OBJECTID);
627 if (ret)
628 goto fail;
629
630 /*
631 * Set the last_snapshot field to the generation of the commit
632 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
633 * correctly (returns true) when the relocation root is created
634 * either inside the critical section of a transaction commit
635 * (through transaction.c:qgroup_account_snapshot()) and when
636 * it's created before the transaction commit is started.
637 */
638 commit_root_gen = btrfs_header_generation(root->commit_root);
639 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
640 } else {
641 /*
642 * called by btrfs_reloc_post_snapshot_hook.
643 * the source tree is a reloc tree, all tree blocks
644 * modified after it was created have RELOC flag
645 * set in their headers. so it's OK to not update
646 * the 'last_snapshot'.
647 */
648 ret = btrfs_copy_root(trans, root, root->node, &eb,
649 BTRFS_TREE_RELOC_OBJECTID);
650 if (ret)
651 goto fail;
652 }
653
654 /*
655 * We have changed references at this point, we must abort the
656 * transaction if anything fails.
657 */
658 must_abort = true;
659
660 memcpy(root_item, &root->root_item, sizeof(*root_item));
661 btrfs_set_root_bytenr(root_item, eb->start);
662 btrfs_set_root_level(root_item, btrfs_header_level(eb));
663 btrfs_set_root_generation(root_item, trans->transid);
664
665 if (btrfs_root_id(root) == objectid) {
666 btrfs_set_root_refs(root_item, 0);
667 memset(&root_item->drop_progress, 0,
668 sizeof(struct btrfs_disk_key));
669 btrfs_set_root_drop_level(root_item, 0);
670 }
671
672 btrfs_tree_unlock(eb);
673 free_extent_buffer(eb);
674
675 ret = btrfs_insert_root(trans, fs_info->tree_root,
676 &root_key, root_item);
677 if (ret)
678 goto fail;
679
680 kfree(root_item);
681
682 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
683 if (IS_ERR(reloc_root)) {
684 ret = PTR_ERR(reloc_root);
685 goto abort;
686 }
687 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
688 btrfs_set_root_last_trans(reloc_root, trans->transid);
689 return reloc_root;
690 fail:
691 kfree(root_item);
692 abort:
693 if (must_abort)
694 btrfs_abort_transaction(trans, ret);
695 return ERR_PTR(ret);
696 }
697
698 /*
699 * create reloc tree for a given fs tree. reloc tree is just a
700 * snapshot of the fs tree with special root objectid.
701 *
702 * The reloc_root comes out of here with two references, one for
703 * root->reloc_root, and another for being on the rc->reloc_roots list.
704 */
btrfs_init_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)705 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
706 struct btrfs_root *root)
707 {
708 struct btrfs_fs_info *fs_info = root->fs_info;
709 struct btrfs_root *reloc_root;
710 struct reloc_control *rc = fs_info->reloc_ctl;
711 struct btrfs_block_rsv *rsv;
712 int clear_rsv = 0;
713 int ret;
714
715 if (!rc)
716 return 0;
717
718 /*
719 * The subvolume has reloc tree but the swap is finished, no need to
720 * create/update the dead reloc tree
721 */
722 if (reloc_root_is_dead(root))
723 return 0;
724
725 /*
726 * This is subtle but important. We do not do
727 * record_root_in_transaction for reloc roots, instead we record their
728 * corresponding fs root, and then here we update the last trans for the
729 * reloc root. This means that we have to do this for the entire life
730 * of the reloc root, regardless of which stage of the relocation we are
731 * in.
732 */
733 if (root->reloc_root) {
734 reloc_root = root->reloc_root;
735 btrfs_set_root_last_trans(reloc_root, trans->transid);
736 return 0;
737 }
738
739 /*
740 * We are merging reloc roots, we do not need new reloc trees. Also
741 * reloc trees never need their own reloc tree.
742 */
743 if (!rc->create_reloc_tree || btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID)
744 return 0;
745
746 if (!trans->reloc_reserved) {
747 rsv = trans->block_rsv;
748 trans->block_rsv = rc->block_rsv;
749 clear_rsv = 1;
750 }
751 reloc_root = create_reloc_root(trans, root, btrfs_root_id(root));
752 if (clear_rsv)
753 trans->block_rsv = rsv;
754 if (IS_ERR(reloc_root))
755 return PTR_ERR(reloc_root);
756
757 ret = __add_reloc_root(reloc_root);
758 ASSERT(ret != -EEXIST);
759 if (ret) {
760 /* Pairs with create_reloc_root */
761 btrfs_put_root(reloc_root);
762 return ret;
763 }
764 root->reloc_root = btrfs_grab_root(reloc_root);
765 return 0;
766 }
767
768 /*
769 * update root item of reloc tree
770 */
btrfs_update_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)771 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
772 struct btrfs_root *root)
773 {
774 struct btrfs_fs_info *fs_info = root->fs_info;
775 struct btrfs_root *reloc_root;
776 struct btrfs_root_item *root_item;
777 int ret;
778
779 if (!have_reloc_root(root))
780 return 0;
781
782 reloc_root = root->reloc_root;
783 root_item = &reloc_root->root_item;
784
785 /*
786 * We are probably ok here, but __del_reloc_root() will drop its ref of
787 * the root. We have the ref for root->reloc_root, but just in case
788 * hold it while we update the reloc root.
789 */
790 btrfs_grab_root(reloc_root);
791
792 /* root->reloc_root will stay until current relocation finished */
793 if (fs_info->reloc_ctl && fs_info->reloc_ctl->merge_reloc_tree &&
794 btrfs_root_refs(root_item) == 0) {
795 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
796 /*
797 * Mark the tree as dead before we change reloc_root so
798 * have_reloc_root will not touch it from now on.
799 */
800 smp_wmb();
801 __del_reloc_root(reloc_root);
802 }
803
804 if (reloc_root->commit_root != reloc_root->node) {
805 __update_reloc_root(reloc_root);
806 btrfs_set_root_node(root_item, reloc_root->node);
807 free_extent_buffer(reloc_root->commit_root);
808 reloc_root->commit_root = btrfs_root_node(reloc_root);
809 }
810
811 ret = btrfs_update_root(trans, fs_info->tree_root,
812 &reloc_root->root_key, root_item);
813 btrfs_put_root(reloc_root);
814 return ret;
815 }
816
817 /*
818 * get new location of data
819 */
get_new_location(struct inode * reloc_inode,u64 * new_bytenr,u64 bytenr,u64 num_bytes)820 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
821 u64 bytenr, u64 num_bytes)
822 {
823 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
824 BTRFS_PATH_AUTO_FREE(path);
825 struct btrfs_file_extent_item *fi;
826 struct extent_buffer *leaf;
827 int ret;
828
829 path = btrfs_alloc_path();
830 if (!path)
831 return -ENOMEM;
832
833 bytenr -= BTRFS_I(reloc_inode)->reloc_block_group_start;
834 ret = btrfs_lookup_file_extent(NULL, root, path,
835 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
836 if (ret < 0)
837 return ret;
838 if (ret > 0)
839 return -ENOENT;
840
841 leaf = path->nodes[0];
842 fi = btrfs_item_ptr(leaf, path->slots[0],
843 struct btrfs_file_extent_item);
844
845 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
846 btrfs_file_extent_compression(leaf, fi) ||
847 btrfs_file_extent_encryption(leaf, fi) ||
848 btrfs_file_extent_other_encoding(leaf, fi));
849
850 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi))
851 return -EINVAL;
852
853 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
854 return 0;
855 }
856
857 /*
858 * update file extent items in the tree leaf to point to
859 * the new locations.
860 */
861 static noinline_for_stack
replace_file_extents(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * root,struct extent_buffer * leaf)862 int replace_file_extents(struct btrfs_trans_handle *trans,
863 struct reloc_control *rc,
864 struct btrfs_root *root,
865 struct extent_buffer *leaf)
866 {
867 struct btrfs_fs_info *fs_info = root->fs_info;
868 struct btrfs_key key;
869 struct btrfs_file_extent_item *fi;
870 struct btrfs_inode *inode = NULL;
871 u64 parent;
872 u64 bytenr;
873 u64 new_bytenr = 0;
874 u64 num_bytes;
875 u64 end;
876 u32 nritems;
877 u32 i;
878 int ret = 0;
879 int first = 1;
880
881 if (rc->stage != UPDATE_DATA_PTRS)
882 return 0;
883
884 /* reloc trees always use full backref */
885 if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID)
886 parent = leaf->start;
887 else
888 parent = 0;
889
890 nritems = btrfs_header_nritems(leaf);
891 for (i = 0; i < nritems; i++) {
892 struct btrfs_ref ref = { 0 };
893
894 cond_resched();
895 btrfs_item_key_to_cpu(leaf, &key, i);
896 if (key.type != BTRFS_EXTENT_DATA_KEY)
897 continue;
898 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
899 if (btrfs_file_extent_type(leaf, fi) ==
900 BTRFS_FILE_EXTENT_INLINE)
901 continue;
902 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
903 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
904 if (bytenr == 0)
905 continue;
906 if (!in_range(bytenr, rc->block_group->start,
907 rc->block_group->length))
908 continue;
909
910 /*
911 * if we are modifying block in fs tree, wait for read_folio
912 * to complete and drop the extent cache
913 */
914 if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) {
915 if (first) {
916 inode = btrfs_find_first_inode(root, key.objectid);
917 first = 0;
918 } else if (inode && btrfs_ino(inode) < key.objectid) {
919 btrfs_add_delayed_iput(inode);
920 inode = btrfs_find_first_inode(root, key.objectid);
921 }
922 if (inode && btrfs_ino(inode) == key.objectid) {
923 struct extent_state *cached_state = NULL;
924
925 end = key.offset +
926 btrfs_file_extent_num_bytes(leaf, fi);
927 WARN_ON(!IS_ALIGNED(key.offset,
928 fs_info->sectorsize));
929 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
930 end--;
931 /* Take mmap lock to serialize with reflinks. */
932 if (!down_read_trylock(&inode->i_mmap_lock))
933 continue;
934 ret = btrfs_try_lock_extent(&inode->io_tree, key.offset,
935 end, &cached_state);
936 if (!ret) {
937 up_read(&inode->i_mmap_lock);
938 continue;
939 }
940
941 btrfs_drop_extent_map_range(inode, key.offset, end, true);
942 btrfs_unlock_extent(&inode->io_tree, key.offset, end,
943 &cached_state);
944 up_read(&inode->i_mmap_lock);
945 }
946 }
947
948 ret = get_new_location(rc->data_inode, &new_bytenr,
949 bytenr, num_bytes);
950 if (ret) {
951 /*
952 * Don't have to abort since we've not changed anything
953 * in the file extent yet.
954 */
955 break;
956 }
957
958 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
959
960 key.offset -= btrfs_file_extent_offset(leaf, fi);
961 ref.action = BTRFS_ADD_DELAYED_REF;
962 ref.bytenr = new_bytenr;
963 ref.num_bytes = num_bytes;
964 ref.parent = parent;
965 ref.owning_root = btrfs_root_id(root);
966 ref.ref_root = btrfs_header_owner(leaf);
967 btrfs_init_data_ref(&ref, key.objectid, key.offset,
968 btrfs_root_id(root), false);
969 ret = btrfs_inc_extent_ref(trans, &ref);
970 if (unlikely(ret)) {
971 btrfs_abort_transaction(trans, ret);
972 break;
973 }
974
975 ref.action = BTRFS_DROP_DELAYED_REF;
976 ref.bytenr = bytenr;
977 ref.num_bytes = num_bytes;
978 ref.parent = parent;
979 ref.owning_root = btrfs_root_id(root);
980 ref.ref_root = btrfs_header_owner(leaf);
981 btrfs_init_data_ref(&ref, key.objectid, key.offset,
982 btrfs_root_id(root), false);
983 ret = btrfs_free_extent(trans, &ref);
984 if (unlikely(ret)) {
985 btrfs_abort_transaction(trans, ret);
986 break;
987 }
988 }
989 if (inode)
990 btrfs_add_delayed_iput(inode);
991 return ret;
992 }
993
memcmp_node_keys(const struct extent_buffer * eb,int slot,const struct btrfs_path * path,int level)994 static noinline_for_stack int memcmp_node_keys(const struct extent_buffer *eb,
995 int slot, const struct btrfs_path *path,
996 int level)
997 {
998 struct btrfs_disk_key key1;
999 struct btrfs_disk_key key2;
1000 btrfs_node_key(eb, &key1, slot);
1001 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1002 return memcmp(&key1, &key2, sizeof(key1));
1003 }
1004
1005 /*
1006 * try to replace tree blocks in fs tree with the new blocks
1007 * in reloc tree. tree blocks haven't been modified since the
1008 * reloc tree was create can be replaced.
1009 *
1010 * if a block was replaced, level of the block + 1 is returned.
1011 * if no block got replaced, 0 is returned. if there are other
1012 * errors, a negative error number is returned.
1013 */
1014 static noinline_for_stack
replace_path(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * dest,struct btrfs_root * src,struct btrfs_path * path,struct btrfs_key * next_key,int lowest_level,int max_level)1015 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1016 struct btrfs_root *dest, struct btrfs_root *src,
1017 struct btrfs_path *path, struct btrfs_key *next_key,
1018 int lowest_level, int max_level)
1019 {
1020 struct btrfs_fs_info *fs_info = dest->fs_info;
1021 struct extent_buffer *eb;
1022 struct extent_buffer *parent;
1023 struct btrfs_ref ref = { 0 };
1024 struct btrfs_key key;
1025 u64 old_bytenr;
1026 u64 new_bytenr;
1027 u64 old_ptr_gen;
1028 u64 new_ptr_gen;
1029 u64 last_snapshot;
1030 u32 blocksize;
1031 int cow = 0;
1032 int level;
1033 int ret;
1034 int slot;
1035
1036 ASSERT(btrfs_root_id(src) == BTRFS_TREE_RELOC_OBJECTID);
1037 ASSERT(btrfs_root_id(dest) != BTRFS_TREE_RELOC_OBJECTID);
1038
1039 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1040 again:
1041 slot = path->slots[lowest_level];
1042 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1043
1044 eb = btrfs_lock_root_node(dest);
1045 level = btrfs_header_level(eb);
1046
1047 if (level < lowest_level) {
1048 btrfs_tree_unlock(eb);
1049 free_extent_buffer(eb);
1050 return 0;
1051 }
1052
1053 if (cow) {
1054 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1055 BTRFS_NESTING_COW);
1056 if (ret) {
1057 btrfs_tree_unlock(eb);
1058 free_extent_buffer(eb);
1059 return ret;
1060 }
1061 }
1062
1063 if (next_key) {
1064 next_key->objectid = (u64)-1;
1065 next_key->type = (u8)-1;
1066 next_key->offset = (u64)-1;
1067 }
1068
1069 parent = eb;
1070 while (1) {
1071 level = btrfs_header_level(parent);
1072 ASSERT(level >= lowest_level);
1073
1074 ret = btrfs_bin_search(parent, 0, &key, &slot);
1075 if (ret < 0)
1076 break;
1077 if (ret && slot > 0)
1078 slot--;
1079
1080 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1081 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1082
1083 old_bytenr = btrfs_node_blockptr(parent, slot);
1084 blocksize = fs_info->nodesize;
1085 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1086
1087 if (level <= max_level) {
1088 eb = path->nodes[level];
1089 new_bytenr = btrfs_node_blockptr(eb,
1090 path->slots[level]);
1091 new_ptr_gen = btrfs_node_ptr_generation(eb,
1092 path->slots[level]);
1093 } else {
1094 new_bytenr = 0;
1095 new_ptr_gen = 0;
1096 }
1097
1098 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1099 ret = level;
1100 break;
1101 }
1102
1103 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1104 memcmp_node_keys(parent, slot, path, level)) {
1105 if (level <= lowest_level) {
1106 ret = 0;
1107 break;
1108 }
1109
1110 eb = btrfs_read_node_slot(parent, slot);
1111 if (IS_ERR(eb)) {
1112 ret = PTR_ERR(eb);
1113 break;
1114 }
1115 btrfs_tree_lock(eb);
1116 if (cow) {
1117 ret = btrfs_cow_block(trans, dest, eb, parent,
1118 slot, &eb,
1119 BTRFS_NESTING_COW);
1120 if (ret) {
1121 btrfs_tree_unlock(eb);
1122 free_extent_buffer(eb);
1123 break;
1124 }
1125 }
1126
1127 btrfs_tree_unlock(parent);
1128 free_extent_buffer(parent);
1129
1130 parent = eb;
1131 continue;
1132 }
1133
1134 if (!cow) {
1135 btrfs_tree_unlock(parent);
1136 free_extent_buffer(parent);
1137 cow = 1;
1138 goto again;
1139 }
1140
1141 btrfs_node_key_to_cpu(path->nodes[level], &key,
1142 path->slots[level]);
1143 btrfs_release_path(path);
1144
1145 path->lowest_level = level;
1146 set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1147 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1148 clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1149 path->lowest_level = 0;
1150 if (ret) {
1151 if (ret > 0)
1152 ret = -ENOENT;
1153 break;
1154 }
1155
1156 /*
1157 * Info qgroup to trace both subtrees.
1158 *
1159 * We must trace both trees.
1160 * 1) Tree reloc subtree
1161 * If not traced, we will leak data numbers
1162 * 2) Fs subtree
1163 * If not traced, we will double count old data
1164 *
1165 * We don't scan the subtree right now, but only record
1166 * the swapped tree blocks.
1167 * The real subtree rescan is delayed until we have new
1168 * CoW on the subtree root node before transaction commit.
1169 */
1170 ret = btrfs_qgroup_add_swapped_blocks(dest,
1171 rc->block_group, parent, slot,
1172 path->nodes[level], path->slots[level],
1173 last_snapshot);
1174 if (ret < 0)
1175 break;
1176 /*
1177 * swap blocks in fs tree and reloc tree.
1178 */
1179 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1180 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1181
1182 btrfs_set_node_blockptr(path->nodes[level],
1183 path->slots[level], old_bytenr);
1184 btrfs_set_node_ptr_generation(path->nodes[level],
1185 path->slots[level], old_ptr_gen);
1186
1187 ref.action = BTRFS_ADD_DELAYED_REF;
1188 ref.bytenr = old_bytenr;
1189 ref.num_bytes = blocksize;
1190 ref.parent = path->nodes[level]->start;
1191 ref.owning_root = btrfs_root_id(src);
1192 ref.ref_root = btrfs_root_id(src);
1193 btrfs_init_tree_ref(&ref, level - 1, 0, true);
1194 ret = btrfs_inc_extent_ref(trans, &ref);
1195 if (unlikely(ret)) {
1196 btrfs_abort_transaction(trans, ret);
1197 break;
1198 }
1199
1200 ref.action = BTRFS_ADD_DELAYED_REF;
1201 ref.bytenr = new_bytenr;
1202 ref.num_bytes = blocksize;
1203 ref.parent = 0;
1204 ref.owning_root = btrfs_root_id(dest);
1205 ref.ref_root = btrfs_root_id(dest);
1206 btrfs_init_tree_ref(&ref, level - 1, 0, true);
1207 ret = btrfs_inc_extent_ref(trans, &ref);
1208 if (unlikely(ret)) {
1209 btrfs_abort_transaction(trans, ret);
1210 break;
1211 }
1212
1213 /* We don't know the real owning_root, use 0. */
1214 ref.action = BTRFS_DROP_DELAYED_REF;
1215 ref.bytenr = new_bytenr;
1216 ref.num_bytes = blocksize;
1217 ref.parent = path->nodes[level]->start;
1218 ref.owning_root = 0;
1219 ref.ref_root = btrfs_root_id(src);
1220 btrfs_init_tree_ref(&ref, level - 1, 0, true);
1221 ret = btrfs_free_extent(trans, &ref);
1222 if (unlikely(ret)) {
1223 btrfs_abort_transaction(trans, ret);
1224 break;
1225 }
1226
1227 /* We don't know the real owning_root, use 0. */
1228 ref.action = BTRFS_DROP_DELAYED_REF;
1229 ref.bytenr = old_bytenr;
1230 ref.num_bytes = blocksize;
1231 ref.parent = 0;
1232 ref.owning_root = 0;
1233 ref.ref_root = btrfs_root_id(dest);
1234 btrfs_init_tree_ref(&ref, level - 1, 0, true);
1235 ret = btrfs_free_extent(trans, &ref);
1236 if (unlikely(ret)) {
1237 btrfs_abort_transaction(trans, ret);
1238 break;
1239 }
1240
1241 btrfs_unlock_up_safe(path, 0);
1242
1243 ret = level;
1244 break;
1245 }
1246 btrfs_tree_unlock(parent);
1247 free_extent_buffer(parent);
1248 return ret;
1249 }
1250
1251 /*
1252 * helper to find next relocated block in reloc tree
1253 */
1254 static noinline_for_stack
walk_up_reloc_tree(struct btrfs_root * root,struct btrfs_path * path,int * level)1255 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1256 int *level)
1257 {
1258 struct extent_buffer *eb;
1259 int i;
1260 u64 last_snapshot;
1261 u32 nritems;
1262
1263 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1264
1265 for (i = 0; i < *level; i++) {
1266 free_extent_buffer(path->nodes[i]);
1267 path->nodes[i] = NULL;
1268 }
1269
1270 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1271 eb = path->nodes[i];
1272 nritems = btrfs_header_nritems(eb);
1273 while (path->slots[i] + 1 < nritems) {
1274 path->slots[i]++;
1275 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1276 last_snapshot)
1277 continue;
1278
1279 *level = i;
1280 return 0;
1281 }
1282 free_extent_buffer(path->nodes[i]);
1283 path->nodes[i] = NULL;
1284 }
1285 return 1;
1286 }
1287
1288 /*
1289 * walk down reloc tree to find relocated block of lowest level
1290 */
1291 static noinline_for_stack
walk_down_reloc_tree(struct btrfs_root * root,struct btrfs_path * path,int * level)1292 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1293 int *level)
1294 {
1295 struct extent_buffer *eb = NULL;
1296 int i;
1297 u64 ptr_gen = 0;
1298 u64 last_snapshot;
1299 u32 nritems;
1300
1301 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1302
1303 for (i = *level; i > 0; i--) {
1304 eb = path->nodes[i];
1305 nritems = btrfs_header_nritems(eb);
1306 while (path->slots[i] < nritems) {
1307 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1308 if (ptr_gen > last_snapshot)
1309 break;
1310 path->slots[i]++;
1311 }
1312 if (path->slots[i] >= nritems) {
1313 if (i == *level)
1314 break;
1315 *level = i + 1;
1316 return 0;
1317 }
1318 if (i == 1) {
1319 *level = i;
1320 return 0;
1321 }
1322
1323 eb = btrfs_read_node_slot(eb, path->slots[i]);
1324 if (IS_ERR(eb))
1325 return PTR_ERR(eb);
1326 BUG_ON(btrfs_header_level(eb) != i - 1);
1327 path->nodes[i - 1] = eb;
1328 path->slots[i - 1] = 0;
1329 }
1330 return 1;
1331 }
1332
1333 /*
1334 * invalidate extent cache for file extents whose key in range of
1335 * [min_key, max_key)
1336 */
invalidate_extent_cache(struct btrfs_root * root,const struct btrfs_key * min_key,const struct btrfs_key * max_key)1337 static int invalidate_extent_cache(struct btrfs_root *root,
1338 const struct btrfs_key *min_key,
1339 const struct btrfs_key *max_key)
1340 {
1341 struct btrfs_fs_info *fs_info = root->fs_info;
1342 struct btrfs_inode *inode = NULL;
1343 u64 objectid;
1344 u64 start, end;
1345 u64 ino;
1346
1347 objectid = min_key->objectid;
1348 while (1) {
1349 struct extent_state *cached_state = NULL;
1350
1351 cond_resched();
1352 if (inode)
1353 iput(&inode->vfs_inode);
1354
1355 if (objectid > max_key->objectid)
1356 break;
1357
1358 inode = btrfs_find_first_inode(root, objectid);
1359 if (!inode)
1360 break;
1361 ino = btrfs_ino(inode);
1362
1363 if (ino > max_key->objectid) {
1364 iput(&inode->vfs_inode);
1365 break;
1366 }
1367
1368 objectid = ino + 1;
1369 if (!S_ISREG(inode->vfs_inode.i_mode))
1370 continue;
1371
1372 if (unlikely(min_key->objectid == ino)) {
1373 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1374 continue;
1375 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1376 start = 0;
1377 else {
1378 start = min_key->offset;
1379 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1380 }
1381 } else {
1382 start = 0;
1383 }
1384
1385 if (unlikely(max_key->objectid == ino)) {
1386 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1387 continue;
1388 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1389 end = (u64)-1;
1390 } else {
1391 if (max_key->offset == 0)
1392 continue;
1393 end = max_key->offset;
1394 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1395 end--;
1396 }
1397 } else {
1398 end = (u64)-1;
1399 }
1400
1401 /* the lock_extent waits for read_folio to complete */
1402 btrfs_lock_extent(&inode->io_tree, start, end, &cached_state);
1403 btrfs_drop_extent_map_range(inode, start, end, true);
1404 btrfs_unlock_extent(&inode->io_tree, start, end, &cached_state);
1405 }
1406 return 0;
1407 }
1408
find_next_key(struct btrfs_path * path,int level,struct btrfs_key * key)1409 static int find_next_key(struct btrfs_path *path, int level,
1410 struct btrfs_key *key)
1411
1412 {
1413 while (level < BTRFS_MAX_LEVEL) {
1414 if (!path->nodes[level])
1415 break;
1416 if (path->slots[level] + 1 <
1417 btrfs_header_nritems(path->nodes[level])) {
1418 btrfs_node_key_to_cpu(path->nodes[level], key,
1419 path->slots[level] + 1);
1420 return 0;
1421 }
1422 level++;
1423 }
1424 return 1;
1425 }
1426
1427 /*
1428 * Insert current subvolume into reloc_control::dirty_subvol_roots
1429 */
insert_dirty_subvol(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * root)1430 static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1431 struct reloc_control *rc,
1432 struct btrfs_root *root)
1433 {
1434 struct btrfs_root *reloc_root = root->reloc_root;
1435 struct btrfs_root_item *reloc_root_item;
1436 int ret;
1437
1438 /* @root must be a subvolume tree root with a valid reloc tree */
1439 ASSERT(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID);
1440 ASSERT(reloc_root);
1441
1442 reloc_root_item = &reloc_root->root_item;
1443 memset(&reloc_root_item->drop_progress, 0,
1444 sizeof(reloc_root_item->drop_progress));
1445 btrfs_set_root_drop_level(reloc_root_item, 0);
1446 btrfs_set_root_refs(reloc_root_item, 0);
1447 ret = btrfs_update_reloc_root(trans, root);
1448 if (ret)
1449 return ret;
1450
1451 if (list_empty(&root->reloc_dirty_list)) {
1452 btrfs_grab_root(root);
1453 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1454 }
1455
1456 return 0;
1457 }
1458
clean_dirty_subvols(struct reloc_control * rc)1459 static int clean_dirty_subvols(struct reloc_control *rc)
1460 {
1461 struct btrfs_root *root;
1462 struct btrfs_root *next;
1463 int ret = 0;
1464 int ret2;
1465
1466 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1467 reloc_dirty_list) {
1468 if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) {
1469 /* Merged subvolume, cleanup its reloc root */
1470 struct btrfs_root *reloc_root = root->reloc_root;
1471
1472 list_del_init(&root->reloc_dirty_list);
1473 root->reloc_root = NULL;
1474 /*
1475 * Need barrier to ensure clear_bit() only happens after
1476 * root->reloc_root = NULL. Pairs with have_reloc_root.
1477 */
1478 smp_wmb();
1479 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1480 if (reloc_root) {
1481 /*
1482 * btrfs_drop_snapshot drops our ref we hold for
1483 * ->reloc_root. If it fails however we must
1484 * drop the ref ourselves.
1485 */
1486 ret2 = btrfs_drop_snapshot(reloc_root, false, true);
1487 if (ret2 < 0) {
1488 btrfs_put_root(reloc_root);
1489 if (!ret)
1490 ret = ret2;
1491 }
1492 }
1493 btrfs_put_root(root);
1494 } else {
1495 /* Orphan reloc tree, just clean it up */
1496 ret2 = btrfs_drop_snapshot(root, false, true);
1497 if (ret2 < 0) {
1498 btrfs_put_root(root);
1499 if (!ret)
1500 ret = ret2;
1501 }
1502 }
1503 }
1504 return ret;
1505 }
1506
1507 /*
1508 * merge the relocated tree blocks in reloc tree with corresponding
1509 * fs tree.
1510 */
merge_reloc_root(struct reloc_control * rc,struct btrfs_root * root)1511 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1512 struct btrfs_root *root)
1513 {
1514 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1515 struct btrfs_key key;
1516 struct btrfs_key next_key;
1517 struct btrfs_trans_handle *trans = NULL;
1518 struct btrfs_root *reloc_root;
1519 struct btrfs_root_item *root_item;
1520 struct btrfs_path *path;
1521 struct extent_buffer *leaf;
1522 int reserve_level;
1523 int level;
1524 int max_level;
1525 int replaced = 0;
1526 int ret = 0;
1527 u32 min_reserved;
1528
1529 path = btrfs_alloc_path();
1530 if (!path)
1531 return -ENOMEM;
1532 path->reada = READA_FORWARD;
1533
1534 reloc_root = root->reloc_root;
1535 root_item = &reloc_root->root_item;
1536
1537 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1538 level = btrfs_root_level(root_item);
1539 refcount_inc(&reloc_root->node->refs);
1540 path->nodes[level] = reloc_root->node;
1541 path->slots[level] = 0;
1542 } else {
1543 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1544
1545 level = btrfs_root_drop_level(root_item);
1546 BUG_ON(level == 0);
1547 path->lowest_level = level;
1548 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1549 path->lowest_level = 0;
1550 if (ret < 0) {
1551 btrfs_free_path(path);
1552 return ret;
1553 }
1554
1555 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1556 path->slots[level]);
1557 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1558
1559 btrfs_unlock_up_safe(path, 0);
1560 }
1561
1562 /*
1563 * In merge_reloc_root(), we modify the upper level pointer to swap the
1564 * tree blocks between reloc tree and subvolume tree. Thus for tree
1565 * block COW, we COW at most from level 1 to root level for each tree.
1566 *
1567 * Thus the needed metadata size is at most root_level * nodesize,
1568 * and * 2 since we have two trees to COW.
1569 */
1570 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1571 min_reserved = fs_info->nodesize * reserve_level * 2;
1572 memset(&next_key, 0, sizeof(next_key));
1573
1574 while (1) {
1575 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
1576 min_reserved,
1577 BTRFS_RESERVE_FLUSH_LIMIT);
1578 if (ret)
1579 goto out;
1580 trans = btrfs_start_transaction(root, 0);
1581 if (IS_ERR(trans)) {
1582 ret = PTR_ERR(trans);
1583 trans = NULL;
1584 goto out;
1585 }
1586
1587 /*
1588 * At this point we no longer have a reloc_control, so we can't
1589 * depend on btrfs_init_reloc_root to update our last_trans.
1590 *
1591 * But that's ok, we started the trans handle on our
1592 * corresponding fs_root, which means it's been added to the
1593 * dirty list. At commit time we'll still call
1594 * btrfs_update_reloc_root() and update our root item
1595 * appropriately.
1596 */
1597 btrfs_set_root_last_trans(reloc_root, trans->transid);
1598 trans->block_rsv = rc->block_rsv;
1599
1600 replaced = 0;
1601 max_level = level;
1602
1603 ret = walk_down_reloc_tree(reloc_root, path, &level);
1604 if (ret < 0)
1605 goto out;
1606 if (ret > 0)
1607 break;
1608
1609 if (!find_next_key(path, level, &key) &&
1610 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1611 ret = 0;
1612 } else {
1613 ret = replace_path(trans, rc, root, reloc_root, path,
1614 &next_key, level, max_level);
1615 }
1616 if (ret < 0)
1617 goto out;
1618 if (ret > 0) {
1619 level = ret;
1620 btrfs_node_key_to_cpu(path->nodes[level], &key,
1621 path->slots[level]);
1622 replaced = 1;
1623 }
1624
1625 ret = walk_up_reloc_tree(reloc_root, path, &level);
1626 if (ret > 0)
1627 break;
1628
1629 BUG_ON(level == 0);
1630 /*
1631 * save the merging progress in the drop_progress.
1632 * this is OK since root refs == 1 in this case.
1633 */
1634 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1635 path->slots[level]);
1636 btrfs_set_root_drop_level(root_item, level);
1637
1638 btrfs_end_transaction_throttle(trans);
1639 trans = NULL;
1640
1641 btrfs_btree_balance_dirty(fs_info);
1642
1643 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1644 invalidate_extent_cache(root, &key, &next_key);
1645 }
1646
1647 /*
1648 * handle the case only one block in the fs tree need to be
1649 * relocated and the block is tree root.
1650 */
1651 leaf = btrfs_lock_root_node(root);
1652 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1653 BTRFS_NESTING_COW);
1654 btrfs_tree_unlock(leaf);
1655 free_extent_buffer(leaf);
1656 out:
1657 btrfs_free_path(path);
1658
1659 if (ret == 0) {
1660 ret = insert_dirty_subvol(trans, rc, root);
1661 if (ret)
1662 btrfs_abort_transaction(trans, ret);
1663 }
1664
1665 if (trans)
1666 btrfs_end_transaction_throttle(trans);
1667
1668 btrfs_btree_balance_dirty(fs_info);
1669
1670 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1671 invalidate_extent_cache(root, &key, &next_key);
1672
1673 return ret;
1674 }
1675
1676 static noinline_for_stack
prepare_to_merge(struct reloc_control * rc,int err)1677 int prepare_to_merge(struct reloc_control *rc, int err)
1678 {
1679 struct btrfs_root *root = rc->extent_root;
1680 struct btrfs_fs_info *fs_info = root->fs_info;
1681 struct btrfs_root *reloc_root;
1682 struct btrfs_trans_handle *trans;
1683 LIST_HEAD(reloc_roots);
1684 u64 num_bytes = 0;
1685 int ret;
1686
1687 mutex_lock(&fs_info->reloc_mutex);
1688 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1689 rc->merging_rsv_size += rc->nodes_relocated * 2;
1690 mutex_unlock(&fs_info->reloc_mutex);
1691
1692 again:
1693 if (!err) {
1694 num_bytes = rc->merging_rsv_size;
1695 ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
1696 BTRFS_RESERVE_FLUSH_ALL);
1697 if (ret)
1698 err = ret;
1699 }
1700
1701 trans = btrfs_join_transaction(rc->extent_root);
1702 if (IS_ERR(trans)) {
1703 if (!err)
1704 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1705 num_bytes, NULL);
1706 return PTR_ERR(trans);
1707 }
1708
1709 if (!err) {
1710 if (num_bytes != rc->merging_rsv_size) {
1711 btrfs_end_transaction(trans);
1712 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1713 num_bytes, NULL);
1714 goto again;
1715 }
1716 }
1717
1718 rc->merge_reloc_tree = true;
1719
1720 while (!list_empty(&rc->reloc_roots)) {
1721 reloc_root = list_first_entry(&rc->reloc_roots,
1722 struct btrfs_root, root_list);
1723 list_del_init(&reloc_root->root_list);
1724
1725 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1726 false);
1727 if (IS_ERR(root)) {
1728 /*
1729 * Even if we have an error we need this reloc root
1730 * back on our list so we can clean up properly.
1731 */
1732 list_add(&reloc_root->root_list, &reloc_roots);
1733 btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1734 if (!err)
1735 err = PTR_ERR(root);
1736 break;
1737 }
1738
1739 if (unlikely(root->reloc_root != reloc_root)) {
1740 if (root->reloc_root) {
1741 btrfs_err(fs_info,
1742 "reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu",
1743 btrfs_root_id(root),
1744 btrfs_root_id(root->reloc_root),
1745 root->reloc_root->root_key.type,
1746 root->reloc_root->root_key.offset,
1747 btrfs_root_generation(
1748 &root->reloc_root->root_item),
1749 btrfs_root_id(reloc_root),
1750 reloc_root->root_key.type,
1751 reloc_root->root_key.offset,
1752 btrfs_root_generation(
1753 &reloc_root->root_item));
1754 } else {
1755 btrfs_err(fs_info,
1756 "reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu",
1757 btrfs_root_id(root),
1758 btrfs_root_id(reloc_root),
1759 reloc_root->root_key.type,
1760 reloc_root->root_key.offset,
1761 btrfs_root_generation(
1762 &reloc_root->root_item));
1763 }
1764 list_add(&reloc_root->root_list, &reloc_roots);
1765 btrfs_put_root(root);
1766 btrfs_abort_transaction(trans, -EUCLEAN);
1767 if (!err)
1768 err = -EUCLEAN;
1769 break;
1770 }
1771
1772 /*
1773 * set reference count to 1, so btrfs_recover_relocation
1774 * knows it should resumes merging
1775 */
1776 if (!err)
1777 btrfs_set_root_refs(&reloc_root->root_item, 1);
1778 ret = btrfs_update_reloc_root(trans, root);
1779
1780 /*
1781 * Even if we have an error we need this reloc root back on our
1782 * list so we can clean up properly.
1783 */
1784 list_add(&reloc_root->root_list, &reloc_roots);
1785 btrfs_put_root(root);
1786
1787 if (unlikely(ret)) {
1788 btrfs_abort_transaction(trans, ret);
1789 if (!err)
1790 err = ret;
1791 break;
1792 }
1793 }
1794
1795 list_splice(&reloc_roots, &rc->reloc_roots);
1796
1797 if (!err)
1798 err = btrfs_commit_transaction(trans);
1799 else
1800 btrfs_end_transaction(trans);
1801 return err;
1802 }
1803
1804 static noinline_for_stack
free_reloc_roots(struct list_head * list)1805 void free_reloc_roots(struct list_head *list)
1806 {
1807 struct btrfs_root *reloc_root, *tmp;
1808
1809 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1810 __del_reloc_root(reloc_root);
1811 }
1812
1813 static noinline_for_stack
merge_reloc_roots(struct reloc_control * rc)1814 void merge_reloc_roots(struct reloc_control *rc)
1815 {
1816 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1817 struct btrfs_root *root;
1818 struct btrfs_root *reloc_root;
1819 LIST_HEAD(reloc_roots);
1820 int found = 0;
1821 int ret = 0;
1822 again:
1823 root = rc->extent_root;
1824
1825 /*
1826 * this serializes us with btrfs_record_root_in_transaction,
1827 * we have to make sure nobody is in the middle of
1828 * adding their roots to the list while we are
1829 * doing this splice
1830 */
1831 mutex_lock(&fs_info->reloc_mutex);
1832 list_splice_init(&rc->reloc_roots, &reloc_roots);
1833 mutex_unlock(&fs_info->reloc_mutex);
1834
1835 while (!list_empty(&reloc_roots)) {
1836 found = 1;
1837 reloc_root = list_first_entry(&reloc_roots, struct btrfs_root, root_list);
1838
1839 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1840 false);
1841 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1842 if (WARN_ON(IS_ERR(root))) {
1843 /*
1844 * For recovery we read the fs roots on mount,
1845 * and if we didn't find the root then we marked
1846 * the reloc root as a garbage root. For normal
1847 * relocation obviously the root should exist in
1848 * memory. However there's no reason we can't
1849 * handle the error properly here just in case.
1850 */
1851 ret = PTR_ERR(root);
1852 goto out;
1853 }
1854 if (WARN_ON(root->reloc_root != reloc_root)) {
1855 /*
1856 * This can happen if on-disk metadata has some
1857 * corruption, e.g. bad reloc tree key offset.
1858 */
1859 ret = -EINVAL;
1860 goto out;
1861 }
1862 ret = merge_reloc_root(rc, root);
1863 btrfs_put_root(root);
1864 if (ret) {
1865 if (list_empty(&reloc_root->root_list))
1866 list_add_tail(&reloc_root->root_list,
1867 &reloc_roots);
1868 goto out;
1869 }
1870 } else {
1871 if (!IS_ERR(root)) {
1872 if (root->reloc_root == reloc_root) {
1873 root->reloc_root = NULL;
1874 btrfs_put_root(reloc_root);
1875 }
1876 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1877 &root->state);
1878 btrfs_put_root(root);
1879 }
1880
1881 list_del_init(&reloc_root->root_list);
1882 /* Don't forget to queue this reloc root for cleanup */
1883 list_add_tail(&reloc_root->reloc_dirty_list,
1884 &rc->dirty_subvol_roots);
1885 }
1886 }
1887
1888 if (found) {
1889 found = 0;
1890 goto again;
1891 }
1892 out:
1893 if (ret) {
1894 btrfs_handle_fs_error(fs_info, ret, NULL);
1895 free_reloc_roots(&reloc_roots);
1896
1897 /* new reloc root may be added */
1898 mutex_lock(&fs_info->reloc_mutex);
1899 list_splice_init(&rc->reloc_roots, &reloc_roots);
1900 mutex_unlock(&fs_info->reloc_mutex);
1901 free_reloc_roots(&reloc_roots);
1902 }
1903
1904 /*
1905 * We used to have
1906 *
1907 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1908 *
1909 * here, but it's wrong. If we fail to start the transaction in
1910 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1911 * have actually been removed from the reloc_root_tree rb tree. This is
1912 * fine because we're bailing here, and we hold a reference on the root
1913 * for the list that holds it, so these roots will be cleaned up when we
1914 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1915 * will be cleaned up on unmount.
1916 *
1917 * The remaining nodes will be cleaned up by free_reloc_control.
1918 */
1919 }
1920
free_block_list(struct rb_root * blocks)1921 static void free_block_list(struct rb_root *blocks)
1922 {
1923 struct tree_block *block;
1924 struct rb_node *rb_node;
1925 while ((rb_node = rb_first(blocks))) {
1926 block = rb_entry(rb_node, struct tree_block, rb_node);
1927 rb_erase(rb_node, blocks);
1928 kfree(block);
1929 }
1930 }
1931
record_reloc_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * reloc_root)1932 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1933 struct btrfs_root *reloc_root)
1934 {
1935 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
1936 struct btrfs_root *root;
1937 int ret;
1938
1939 if (btrfs_get_root_last_trans(reloc_root) == trans->transid)
1940 return 0;
1941
1942 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
1943
1944 /*
1945 * This should succeed, since we can't have a reloc root without having
1946 * already looked up the actual root and created the reloc root for this
1947 * root.
1948 *
1949 * However if there's some sort of corruption where we have a ref to a
1950 * reloc root without a corresponding root this could return ENOENT.
1951 */
1952 if (IS_ERR(root)) {
1953 DEBUG_WARN("error %ld reading root for reloc root", PTR_ERR(root));
1954 return PTR_ERR(root);
1955 }
1956 if (unlikely(root->reloc_root != reloc_root)) {
1957 DEBUG_WARN("unexpected reloc root found");
1958 btrfs_err(fs_info,
1959 "root %llu has two reloc roots associated with it",
1960 reloc_root->root_key.offset);
1961 btrfs_put_root(root);
1962 return -EUCLEAN;
1963 }
1964 ret = btrfs_record_root_in_trans(trans, root);
1965 btrfs_put_root(root);
1966
1967 return ret;
1968 }
1969
1970 static noinline_for_stack
select_reloc_root(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_backref_edge * edges[])1971 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
1972 struct reloc_control *rc,
1973 struct btrfs_backref_node *node,
1974 struct btrfs_backref_edge *edges[])
1975 {
1976 struct btrfs_backref_node *next;
1977 struct btrfs_root *root;
1978 int index = 0;
1979 int ret;
1980
1981 next = walk_up_backref(node, edges, &index);
1982 root = next->root;
1983
1984 /*
1985 * If there is no root, then our references for this block are
1986 * incomplete, as we should be able to walk all the way up to a block
1987 * that is owned by a root.
1988 *
1989 * This path is only for SHAREABLE roots, so if we come upon a
1990 * non-SHAREABLE root then we have backrefs that resolve improperly.
1991 *
1992 * Both of these cases indicate file system corruption, or a bug in the
1993 * backref walking code.
1994 */
1995 if (unlikely(!root)) {
1996 btrfs_err(trans->fs_info,
1997 "bytenr %llu doesn't have a backref path ending in a root",
1998 node->bytenr);
1999 return ERR_PTR(-EUCLEAN);
2000 }
2001 if (unlikely(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))) {
2002 btrfs_err(trans->fs_info,
2003 "bytenr %llu has multiple refs with one ending in a non-shareable root",
2004 node->bytenr);
2005 return ERR_PTR(-EUCLEAN);
2006 }
2007
2008 if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) {
2009 ret = record_reloc_root_in_trans(trans, root);
2010 if (ret)
2011 return ERR_PTR(ret);
2012 goto found;
2013 }
2014
2015 ret = btrfs_record_root_in_trans(trans, root);
2016 if (ret)
2017 return ERR_PTR(ret);
2018 root = root->reloc_root;
2019
2020 /*
2021 * We could have raced with another thread which failed, so
2022 * root->reloc_root may not be set, return ENOENT in this case.
2023 */
2024 if (!root)
2025 return ERR_PTR(-ENOENT);
2026
2027 if (unlikely(next->new_bytenr)) {
2028 /*
2029 * We just created the reloc root, so we shouldn't have
2030 * ->new_bytenr set yet. If it is then we have multiple roots
2031 * pointing at the same bytenr which indicates corruption, or
2032 * we've made a mistake in the backref walking code.
2033 */
2034 ASSERT(next->new_bytenr == 0);
2035 btrfs_err(trans->fs_info,
2036 "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2037 node->bytenr, next->bytenr);
2038 return ERR_PTR(-EUCLEAN);
2039 }
2040
2041 next->new_bytenr = root->node->start;
2042 btrfs_put_root(next->root);
2043 next->root = btrfs_grab_root(root);
2044 ASSERT(next->root);
2045 mark_block_processed(rc, next);
2046 found:
2047 next = node;
2048 /* setup backref node path for btrfs_reloc_cow_block */
2049 while (1) {
2050 rc->backref_cache.path[next->level] = next;
2051 if (--index < 0)
2052 break;
2053 next = edges[index]->node[UPPER];
2054 }
2055 return root;
2056 }
2057
2058 /*
2059 * Select a tree root for relocation.
2060 *
2061 * Return NULL if the block is not shareable. We should use do_relocation() in
2062 * this case.
2063 *
2064 * Return a tree root pointer if the block is shareable.
2065 * Return -ENOENT if the block is root of reloc tree.
2066 */
2067 static noinline_for_stack
select_one_root(struct btrfs_backref_node * node)2068 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2069 {
2070 struct btrfs_backref_node *next;
2071 struct btrfs_root *root;
2072 struct btrfs_root *fs_root = NULL;
2073 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2074 int index = 0;
2075
2076 next = node;
2077 while (1) {
2078 cond_resched();
2079 next = walk_up_backref(next, edges, &index);
2080 root = next->root;
2081
2082 /*
2083 * This can occur if we have incomplete extent refs leading all
2084 * the way up a particular path, in this case return -EUCLEAN.
2085 */
2086 if (unlikely(!root))
2087 return ERR_PTR(-EUCLEAN);
2088
2089 /* No other choice for non-shareable tree */
2090 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2091 return root;
2092
2093 if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID)
2094 fs_root = root;
2095
2096 if (next != node)
2097 return NULL;
2098
2099 next = walk_down_backref(edges, &index);
2100 if (!next || next->level <= node->level)
2101 break;
2102 }
2103
2104 if (!fs_root)
2105 return ERR_PTR(-ENOENT);
2106 return fs_root;
2107 }
2108
calcu_metadata_size(struct reloc_control * rc,struct btrfs_backref_node * node)2109 static noinline_for_stack u64 calcu_metadata_size(struct reloc_control *rc,
2110 struct btrfs_backref_node *node)
2111 {
2112 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2113 struct btrfs_backref_node *next = node;
2114 struct btrfs_backref_edge *edge;
2115 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2116 u64 num_bytes = 0;
2117 int index = 0;
2118
2119 BUG_ON(node->processed);
2120
2121 while (next) {
2122 cond_resched();
2123 while (1) {
2124 if (next->processed)
2125 break;
2126
2127 num_bytes += fs_info->nodesize;
2128
2129 if (list_empty(&next->upper))
2130 break;
2131
2132 edge = list_first_entry(&next->upper, struct btrfs_backref_edge,
2133 list[LOWER]);
2134 edges[index++] = edge;
2135 next = edge->node[UPPER];
2136 }
2137 next = walk_down_backref(edges, &index);
2138 }
2139 return num_bytes;
2140 }
2141
refill_metadata_space(struct btrfs_trans_handle * trans,struct reloc_control * rc,u64 num_bytes)2142 static int refill_metadata_space(struct btrfs_trans_handle *trans,
2143 struct reloc_control *rc, u64 num_bytes)
2144 {
2145 struct btrfs_fs_info *fs_info = trans->fs_info;
2146 int ret;
2147
2148 trans->block_rsv = rc->block_rsv;
2149 rc->reserved_bytes += num_bytes;
2150
2151 /*
2152 * We are under a transaction here so we can only do limited flushing.
2153 * If we get an enospc just kick back -EAGAIN so we know to drop the
2154 * transaction and try to refill when we can flush all the things.
2155 */
2156 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
2157 BTRFS_RESERVE_FLUSH_LIMIT);
2158 if (ret) {
2159 u64 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2160
2161 while (tmp <= rc->reserved_bytes)
2162 tmp <<= 1;
2163 /*
2164 * only one thread can access block_rsv at this point,
2165 * so we don't need hold lock to protect block_rsv.
2166 * we expand more reservation size here to allow enough
2167 * space for relocation and we will return earlier in
2168 * enospc case.
2169 */
2170 rc->block_rsv->size = tmp + fs_info->nodesize *
2171 RELOCATION_RESERVED_NODES;
2172 return -EAGAIN;
2173 }
2174
2175 return 0;
2176 }
2177
reserve_metadata_space(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node)2178 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2179 struct reloc_control *rc,
2180 struct btrfs_backref_node *node)
2181 {
2182 u64 num_bytes;
2183
2184 num_bytes = calcu_metadata_size(rc, node) * 2;
2185 return refill_metadata_space(trans, rc, num_bytes);
2186 }
2187
2188 /*
2189 * relocate a block tree, and then update pointers in upper level
2190 * blocks that reference the block to point to the new location.
2191 *
2192 * if called by link_to_upper, the block has already been relocated.
2193 * in that case this function just updates pointers.
2194 */
do_relocation(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_key * key,struct btrfs_path * path,int lowest)2195 static int do_relocation(struct btrfs_trans_handle *trans,
2196 struct reloc_control *rc,
2197 struct btrfs_backref_node *node,
2198 struct btrfs_key *key,
2199 struct btrfs_path *path, int lowest)
2200 {
2201 struct btrfs_backref_node *upper;
2202 struct btrfs_backref_edge *edge;
2203 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2204 struct btrfs_root *root;
2205 struct extent_buffer *eb;
2206 u32 blocksize;
2207 u64 bytenr;
2208 int slot;
2209 int ret = 0;
2210
2211 /*
2212 * If we are lowest then this is the first time we're processing this
2213 * block, and thus shouldn't have an eb associated with it yet.
2214 */
2215 ASSERT(!lowest || !node->eb);
2216
2217 path->lowest_level = node->level + 1;
2218 rc->backref_cache.path[node->level] = node;
2219 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2220 cond_resched();
2221
2222 upper = edge->node[UPPER];
2223 root = select_reloc_root(trans, rc, upper, edges);
2224 if (IS_ERR(root)) {
2225 ret = PTR_ERR(root);
2226 goto next;
2227 }
2228
2229 if (upper->eb && !upper->locked) {
2230 if (!lowest) {
2231 ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2232 if (ret < 0)
2233 goto next;
2234 BUG_ON(ret);
2235 bytenr = btrfs_node_blockptr(upper->eb, slot);
2236 if (node->eb->start == bytenr)
2237 goto next;
2238 }
2239 btrfs_backref_drop_node_buffer(upper);
2240 }
2241
2242 if (!upper->eb) {
2243 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2244 if (ret) {
2245 if (ret > 0)
2246 ret = -ENOENT;
2247
2248 btrfs_release_path(path);
2249 break;
2250 }
2251
2252 if (!upper->eb) {
2253 upper->eb = path->nodes[upper->level];
2254 path->nodes[upper->level] = NULL;
2255 } else {
2256 BUG_ON(upper->eb != path->nodes[upper->level]);
2257 }
2258
2259 upper->locked = 1;
2260 path->locks[upper->level] = 0;
2261
2262 slot = path->slots[upper->level];
2263 btrfs_release_path(path);
2264 } else {
2265 ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2266 if (ret < 0)
2267 goto next;
2268 BUG_ON(ret);
2269 }
2270
2271 bytenr = btrfs_node_blockptr(upper->eb, slot);
2272 if (lowest) {
2273 if (unlikely(bytenr != node->bytenr)) {
2274 btrfs_err(root->fs_info,
2275 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2276 bytenr, node->bytenr, slot,
2277 upper->eb->start);
2278 ret = -EIO;
2279 goto next;
2280 }
2281 } else {
2282 if (node->eb->start == bytenr)
2283 goto next;
2284 }
2285
2286 blocksize = root->fs_info->nodesize;
2287 eb = btrfs_read_node_slot(upper->eb, slot);
2288 if (IS_ERR(eb)) {
2289 ret = PTR_ERR(eb);
2290 goto next;
2291 }
2292 btrfs_tree_lock(eb);
2293
2294 if (!node->eb) {
2295 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2296 slot, &eb, BTRFS_NESTING_COW);
2297 btrfs_tree_unlock(eb);
2298 free_extent_buffer(eb);
2299 if (ret < 0)
2300 goto next;
2301 /*
2302 * We've just COWed this block, it should have updated
2303 * the correct backref node entry.
2304 */
2305 ASSERT(node->eb == eb);
2306 } else {
2307 struct btrfs_ref ref = {
2308 .action = BTRFS_ADD_DELAYED_REF,
2309 .bytenr = node->eb->start,
2310 .num_bytes = blocksize,
2311 .parent = upper->eb->start,
2312 .owning_root = btrfs_header_owner(upper->eb),
2313 .ref_root = btrfs_header_owner(upper->eb),
2314 };
2315
2316 btrfs_set_node_blockptr(upper->eb, slot,
2317 node->eb->start);
2318 btrfs_set_node_ptr_generation(upper->eb, slot,
2319 trans->transid);
2320 btrfs_mark_buffer_dirty(trans, upper->eb);
2321
2322 btrfs_init_tree_ref(&ref, node->level,
2323 btrfs_root_id(root), false);
2324 ret = btrfs_inc_extent_ref(trans, &ref);
2325 if (!ret)
2326 ret = btrfs_drop_subtree(trans, root, eb,
2327 upper->eb);
2328 if (unlikely(ret))
2329 btrfs_abort_transaction(trans, ret);
2330 }
2331 next:
2332 if (!upper->pending)
2333 btrfs_backref_drop_node_buffer(upper);
2334 else
2335 btrfs_backref_unlock_node_buffer(upper);
2336 if (ret)
2337 break;
2338 }
2339
2340 if (!ret && node->pending) {
2341 btrfs_backref_drop_node_buffer(node);
2342 list_del_init(&node->list);
2343 node->pending = 0;
2344 }
2345
2346 path->lowest_level = 0;
2347
2348 /*
2349 * We should have allocated all of our space in the block rsv and thus
2350 * shouldn't ENOSPC.
2351 */
2352 ASSERT(ret != -ENOSPC);
2353 return ret;
2354 }
2355
link_to_upper(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_path * path)2356 static int link_to_upper(struct btrfs_trans_handle *trans,
2357 struct reloc_control *rc,
2358 struct btrfs_backref_node *node,
2359 struct btrfs_path *path)
2360 {
2361 struct btrfs_key key;
2362
2363 btrfs_node_key_to_cpu(node->eb, &key, 0);
2364 return do_relocation(trans, rc, node, &key, path, 0);
2365 }
2366
finish_pending_nodes(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_path * path,int err)2367 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2368 struct reloc_control *rc,
2369 struct btrfs_path *path, int err)
2370 {
2371 LIST_HEAD(list);
2372 struct btrfs_backref_cache *cache = &rc->backref_cache;
2373 struct btrfs_backref_node *node;
2374 int level;
2375 int ret;
2376
2377 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2378 while (!list_empty(&cache->pending[level])) {
2379 node = list_first_entry(&cache->pending[level],
2380 struct btrfs_backref_node, list);
2381 list_move_tail(&node->list, &list);
2382 BUG_ON(!node->pending);
2383
2384 if (!err) {
2385 ret = link_to_upper(trans, rc, node, path);
2386 if (ret < 0)
2387 err = ret;
2388 }
2389 }
2390 list_splice_init(&list, &cache->pending[level]);
2391 }
2392 return err;
2393 }
2394
2395 /*
2396 * mark a block and all blocks directly/indirectly reference the block
2397 * as processed.
2398 */
update_processed_blocks(struct reloc_control * rc,struct btrfs_backref_node * node)2399 static void update_processed_blocks(struct reloc_control *rc,
2400 struct btrfs_backref_node *node)
2401 {
2402 struct btrfs_backref_node *next = node;
2403 struct btrfs_backref_edge *edge;
2404 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2405 int index = 0;
2406
2407 while (next) {
2408 cond_resched();
2409 while (1) {
2410 if (next->processed)
2411 break;
2412
2413 mark_block_processed(rc, next);
2414
2415 if (list_empty(&next->upper))
2416 break;
2417
2418 edge = list_first_entry(&next->upper, struct btrfs_backref_edge,
2419 list[LOWER]);
2420 edges[index++] = edge;
2421 next = edge->node[UPPER];
2422 }
2423 next = walk_down_backref(edges, &index);
2424 }
2425 }
2426
tree_block_processed(u64 bytenr,struct reloc_control * rc)2427 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2428 {
2429 u32 blocksize = rc->extent_root->fs_info->nodesize;
2430
2431 if (btrfs_test_range_bit(&rc->processed_blocks, bytenr,
2432 bytenr + blocksize - 1, EXTENT_DIRTY, NULL))
2433 return 1;
2434 return 0;
2435 }
2436
get_tree_block_key(struct btrfs_fs_info * fs_info,struct tree_block * block)2437 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2438 struct tree_block *block)
2439 {
2440 struct btrfs_tree_parent_check check = {
2441 .level = block->level,
2442 .owner_root = block->owner,
2443 .transid = block->key.offset
2444 };
2445 struct extent_buffer *eb;
2446
2447 eb = read_tree_block(fs_info, block->bytenr, &check);
2448 if (IS_ERR(eb))
2449 return PTR_ERR(eb);
2450 if (unlikely(!extent_buffer_uptodate(eb))) {
2451 free_extent_buffer(eb);
2452 return -EIO;
2453 }
2454 if (block->level == 0)
2455 btrfs_item_key_to_cpu(eb, &block->key, 0);
2456 else
2457 btrfs_node_key_to_cpu(eb, &block->key, 0);
2458 free_extent_buffer(eb);
2459 block->key_ready = true;
2460 return 0;
2461 }
2462
2463 /*
2464 * helper function to relocate a tree block
2465 */
relocate_tree_block(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_key * key,struct btrfs_path * path)2466 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2467 struct reloc_control *rc,
2468 struct btrfs_backref_node *node,
2469 struct btrfs_key *key,
2470 struct btrfs_path *path)
2471 {
2472 struct btrfs_root *root;
2473 int ret = 0;
2474
2475 if (!node)
2476 return 0;
2477
2478 /*
2479 * If we fail here we want to drop our backref_node because we are going
2480 * to start over and regenerate the tree for it.
2481 */
2482 ret = reserve_metadata_space(trans, rc, node);
2483 if (ret)
2484 goto out;
2485
2486 BUG_ON(node->processed);
2487 root = select_one_root(node);
2488 if (IS_ERR(root)) {
2489 ret = PTR_ERR(root);
2490
2491 /* See explanation in select_one_root for the -EUCLEAN case. */
2492 ASSERT(ret == -ENOENT);
2493 if (ret == -ENOENT) {
2494 ret = 0;
2495 update_processed_blocks(rc, node);
2496 }
2497 goto out;
2498 }
2499
2500 if (root) {
2501 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2502 /*
2503 * This block was the root block of a root, and this is
2504 * the first time we're processing the block and thus it
2505 * should not have had the ->new_bytenr modified.
2506 *
2507 * However in the case of corruption we could have
2508 * multiple refs pointing to the same block improperly,
2509 * and thus we would trip over these checks. ASSERT()
2510 * for the developer case, because it could indicate a
2511 * bug in the backref code, however error out for a
2512 * normal user in the case of corruption.
2513 */
2514 ASSERT(node->new_bytenr == 0);
2515 if (unlikely(node->new_bytenr)) {
2516 btrfs_err(root->fs_info,
2517 "bytenr %llu has improper references to it",
2518 node->bytenr);
2519 ret = -EUCLEAN;
2520 goto out;
2521 }
2522 ret = btrfs_record_root_in_trans(trans, root);
2523 if (ret)
2524 goto out;
2525 /*
2526 * Another thread could have failed, need to check if we
2527 * have reloc_root actually set.
2528 */
2529 if (!root->reloc_root) {
2530 ret = -ENOENT;
2531 goto out;
2532 }
2533 root = root->reloc_root;
2534 node->new_bytenr = root->node->start;
2535 btrfs_put_root(node->root);
2536 node->root = btrfs_grab_root(root);
2537 ASSERT(node->root);
2538 } else {
2539 btrfs_err(root->fs_info,
2540 "bytenr %llu resolved to a non-shareable root",
2541 node->bytenr);
2542 ret = -EUCLEAN;
2543 goto out;
2544 }
2545 if (!ret)
2546 update_processed_blocks(rc, node);
2547 } else {
2548 ret = do_relocation(trans, rc, node, key, path, 1);
2549 }
2550 out:
2551 if (ret || node->level == 0)
2552 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2553 return ret;
2554 }
2555
relocate_cowonly_block(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct tree_block * block,struct btrfs_path * path)2556 static int relocate_cowonly_block(struct btrfs_trans_handle *trans,
2557 struct reloc_control *rc, struct tree_block *block,
2558 struct btrfs_path *path)
2559 {
2560 struct btrfs_fs_info *fs_info = trans->fs_info;
2561 struct btrfs_root *root;
2562 u64 num_bytes;
2563 int nr_levels;
2564 int ret;
2565
2566 root = btrfs_get_fs_root(fs_info, block->owner, true);
2567 if (IS_ERR(root))
2568 return PTR_ERR(root);
2569
2570 nr_levels = max(btrfs_header_level(root->node) - block->level, 0) + 1;
2571
2572 num_bytes = fs_info->nodesize * nr_levels;
2573 ret = refill_metadata_space(trans, rc, num_bytes);
2574 if (ret) {
2575 btrfs_put_root(root);
2576 return ret;
2577 }
2578 path->lowest_level = block->level;
2579 if (root == root->fs_info->chunk_root)
2580 btrfs_reserve_chunk_metadata(trans, false);
2581
2582 ret = btrfs_search_slot(trans, root, &block->key, path, 0, 1);
2583 path->lowest_level = 0;
2584 btrfs_release_path(path);
2585
2586 if (root == root->fs_info->chunk_root)
2587 btrfs_trans_release_chunk_metadata(trans);
2588 if (ret > 0)
2589 ret = 0;
2590 btrfs_put_root(root);
2591
2592 return ret;
2593 }
2594
2595 /*
2596 * relocate a list of blocks
2597 */
2598 static noinline_for_stack
relocate_tree_blocks(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct rb_root * blocks)2599 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2600 struct reloc_control *rc, struct rb_root *blocks)
2601 {
2602 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2603 struct btrfs_backref_node *node;
2604 struct btrfs_path *path;
2605 struct tree_block *block;
2606 struct tree_block *next;
2607 int ret = 0;
2608
2609 path = btrfs_alloc_path();
2610 if (!path) {
2611 ret = -ENOMEM;
2612 goto out_free_blocks;
2613 }
2614
2615 /* Kick in readahead for tree blocks with missing keys */
2616 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2617 if (!block->key_ready)
2618 btrfs_readahead_tree_block(fs_info, block->bytenr,
2619 block->owner, 0,
2620 block->level);
2621 }
2622
2623 /* Get first keys */
2624 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2625 if (!block->key_ready) {
2626 ret = get_tree_block_key(fs_info, block);
2627 if (ret)
2628 goto out_free_path;
2629 }
2630 }
2631
2632 /* Do tree relocation */
2633 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2634 /*
2635 * For COWonly blocks, or the data reloc tree, we only need to
2636 * COW down to the block, there's no need to generate a backref
2637 * tree.
2638 */
2639 if (block->owner &&
2640 (!btrfs_is_fstree(block->owner) ||
2641 block->owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
2642 ret = relocate_cowonly_block(trans, rc, block, path);
2643 if (ret)
2644 break;
2645 continue;
2646 }
2647
2648 node = build_backref_tree(trans, rc, &block->key,
2649 block->level, block->bytenr);
2650 if (IS_ERR(node)) {
2651 ret = PTR_ERR(node);
2652 goto out;
2653 }
2654
2655 ret = relocate_tree_block(trans, rc, node, &block->key,
2656 path);
2657 if (ret < 0)
2658 break;
2659 }
2660 out:
2661 ret = finish_pending_nodes(trans, rc, path, ret);
2662
2663 out_free_path:
2664 btrfs_free_path(path);
2665 out_free_blocks:
2666 free_block_list(blocks);
2667 return ret;
2668 }
2669
prealloc_file_extent_cluster(struct reloc_control * rc)2670 static noinline_for_stack int prealloc_file_extent_cluster(struct reloc_control *rc)
2671 {
2672 const struct file_extent_cluster *cluster = &rc->cluster;
2673 struct btrfs_inode *inode = BTRFS_I(rc->data_inode);
2674 u64 alloc_hint = 0;
2675 u64 start;
2676 u64 end;
2677 u64 offset = inode->reloc_block_group_start;
2678 u64 num_bytes;
2679 int nr;
2680 int ret = 0;
2681 u64 prealloc_start = cluster->start - offset;
2682 u64 prealloc_end = cluster->end - offset;
2683 u64 cur_offset = prealloc_start;
2684
2685 /*
2686 * For blocksize < folio size case (either bs < page size or large folios),
2687 * beyond i_size, all blocks are filled with zero.
2688 *
2689 * If the current cluster covers the above range, btrfs_do_readpage()
2690 * will skip the read, and relocate_one_folio() will later writeback
2691 * the padding zeros as new data, causing data corruption.
2692 *
2693 * Here we have to invalidate the cache covering our cluster.
2694 */
2695 ret = filemap_invalidate_inode(&inode->vfs_inode, true, prealloc_start,
2696 prealloc_end);
2697 if (ret < 0)
2698 return ret;
2699
2700 BUG_ON(cluster->start != cluster->boundary[0]);
2701 ret = btrfs_alloc_data_chunk_ondemand(inode,
2702 prealloc_end + 1 - prealloc_start);
2703 if (ret)
2704 return ret;
2705
2706 btrfs_inode_lock(inode, 0);
2707 for (nr = 0; nr < cluster->nr; nr++) {
2708 struct extent_state *cached_state = NULL;
2709
2710 start = cluster->boundary[nr] - offset;
2711 if (nr + 1 < cluster->nr)
2712 end = cluster->boundary[nr + 1] - 1 - offset;
2713 else
2714 end = cluster->end - offset;
2715
2716 btrfs_lock_extent(&inode->io_tree, start, end, &cached_state);
2717 num_bytes = end + 1 - start;
2718 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2719 num_bytes, num_bytes,
2720 end + 1, &alloc_hint);
2721 cur_offset = end + 1;
2722 btrfs_unlock_extent(&inode->io_tree, start, end, &cached_state);
2723 if (ret)
2724 break;
2725 }
2726 btrfs_inode_unlock(inode, 0);
2727
2728 if (cur_offset < prealloc_end)
2729 btrfs_free_reserved_data_space_noquota(inode,
2730 prealloc_end + 1 - cur_offset);
2731 return ret;
2732 }
2733
setup_relocation_extent_mapping(struct reloc_control * rc)2734 static noinline_for_stack int setup_relocation_extent_mapping(struct reloc_control *rc)
2735 {
2736 struct btrfs_inode *inode = BTRFS_I(rc->data_inode);
2737 struct extent_map *em;
2738 struct extent_state *cached_state = NULL;
2739 u64 offset = inode->reloc_block_group_start;
2740 u64 start = rc->cluster.start - offset;
2741 u64 end = rc->cluster.end - offset;
2742 int ret = 0;
2743
2744 em = btrfs_alloc_extent_map();
2745 if (!em)
2746 return -ENOMEM;
2747
2748 em->start = start;
2749 em->len = end + 1 - start;
2750 em->disk_bytenr = rc->cluster.start;
2751 em->disk_num_bytes = em->len;
2752 em->ram_bytes = em->len;
2753 em->flags |= EXTENT_FLAG_PINNED;
2754
2755 btrfs_lock_extent(&inode->io_tree, start, end, &cached_state);
2756 ret = btrfs_replace_extent_map_range(inode, em, false);
2757 btrfs_unlock_extent(&inode->io_tree, start, end, &cached_state);
2758 btrfs_free_extent_map(em);
2759
2760 return ret;
2761 }
2762
2763 /*
2764 * Allow error injection to test balance/relocation cancellation
2765 */
btrfs_should_cancel_balance(const struct btrfs_fs_info * fs_info)2766 noinline int btrfs_should_cancel_balance(const struct btrfs_fs_info *fs_info)
2767 {
2768 return atomic_read(&fs_info->balance_cancel_req) ||
2769 atomic_read(&fs_info->reloc_cancel_req) ||
2770 fatal_signal_pending(current);
2771 }
2772 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2773
get_cluster_boundary_end(const struct file_extent_cluster * cluster,int cluster_nr)2774 static u64 get_cluster_boundary_end(const struct file_extent_cluster *cluster,
2775 int cluster_nr)
2776 {
2777 /* Last extent, use cluster end directly */
2778 if (cluster_nr >= cluster->nr - 1)
2779 return cluster->end;
2780
2781 /* Use next boundary start*/
2782 return cluster->boundary[cluster_nr + 1] - 1;
2783 }
2784
relocate_one_folio(struct reloc_control * rc,struct file_ra_state * ra,int * cluster_nr,u64 * file_offset_ret)2785 static int relocate_one_folio(struct reloc_control *rc,
2786 struct file_ra_state *ra,
2787 int *cluster_nr, u64 *file_offset_ret)
2788 {
2789 const struct file_extent_cluster *cluster = &rc->cluster;
2790 struct inode *inode = rc->data_inode;
2791 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2792 const u64 orig_file_offset = *file_offset_ret;
2793 u64 offset = BTRFS_I(inode)->reloc_block_group_start;
2794 const pgoff_t last_index = (cluster->end - offset) >> PAGE_SHIFT;
2795 const pgoff_t index = orig_file_offset >> PAGE_SHIFT;
2796 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2797 struct folio *folio;
2798 u64 folio_start;
2799 u64 folio_end;
2800 u64 cur;
2801 int ret;
2802 const bool use_rst = btrfs_need_stripe_tree_update(fs_info, rc->block_group->flags);
2803
2804 ASSERT(index <= last_index);
2805 again:
2806 folio = filemap_lock_folio(inode->i_mapping, index);
2807 if (IS_ERR(folio)) {
2808
2809 /*
2810 * On relocation we're doing readahead on the relocation inode,
2811 * but if the filesystem is backed by a RAID stripe tree we can
2812 * get ENOENT (e.g. due to preallocated extents not being
2813 * mapped in the RST) from the lookup.
2814 *
2815 * But readahead doesn't handle the error and submits invalid
2816 * reads to the device, causing a assertion failures.
2817 */
2818 if (!use_rst)
2819 page_cache_sync_readahead(inode->i_mapping, ra, NULL,
2820 index, last_index + 1 - index);
2821 folio = __filemap_get_folio(inode->i_mapping, index,
2822 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
2823 mask);
2824 if (IS_ERR(folio))
2825 return PTR_ERR(folio);
2826 }
2827
2828 if (folio_test_readahead(folio) && !use_rst)
2829 page_cache_async_readahead(inode->i_mapping, ra, NULL,
2830 folio, last_index + 1 - index);
2831
2832 if (!folio_test_uptodate(folio)) {
2833 btrfs_read_folio(NULL, folio);
2834 folio_lock(folio);
2835 if (unlikely(!folio_test_uptodate(folio))) {
2836 ret = -EIO;
2837 goto release_folio;
2838 }
2839 if (folio->mapping != inode->i_mapping) {
2840 folio_unlock(folio);
2841 folio_put(folio);
2842 goto again;
2843 }
2844 }
2845
2846 /*
2847 * We could have lost folio private when we dropped the lock to read the
2848 * folio above, make sure we set_folio_extent_mapped() here so we have any
2849 * of the subpage blocksize stuff we need in place.
2850 */
2851 ret = set_folio_extent_mapped(folio);
2852 if (ret < 0)
2853 goto release_folio;
2854
2855 folio_start = folio_pos(folio);
2856 folio_end = folio_start + folio_size(folio) - 1;
2857
2858 /*
2859 * Start from the cluster, as for subpage case, the cluster can start
2860 * inside the folio.
2861 */
2862 cur = max(folio_start, cluster->boundary[*cluster_nr] - offset);
2863 while (cur <= folio_end) {
2864 struct extent_state *cached_state = NULL;
2865 u64 extent_start = cluster->boundary[*cluster_nr] - offset;
2866 u64 extent_end = get_cluster_boundary_end(cluster,
2867 *cluster_nr) - offset;
2868 u64 clamped_start = max(folio_start, extent_start);
2869 u64 clamped_end = min(folio_end, extent_end);
2870 u32 clamped_len = clamped_end + 1 - clamped_start;
2871
2872 /* Reserve metadata for this range */
2873 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2874 clamped_len, clamped_len,
2875 false);
2876 if (ret)
2877 goto release_folio;
2878
2879 /* Mark the range delalloc and dirty for later writeback */
2880 btrfs_lock_extent(&BTRFS_I(inode)->io_tree, clamped_start,
2881 clamped_end, &cached_state);
2882 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start,
2883 clamped_end, 0, &cached_state);
2884 if (ret) {
2885 btrfs_clear_extent_bit(&BTRFS_I(inode)->io_tree,
2886 clamped_start, clamped_end,
2887 EXTENT_LOCKED | EXTENT_BOUNDARY,
2888 &cached_state);
2889 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2890 clamped_len, true);
2891 btrfs_delalloc_release_extents(BTRFS_I(inode),
2892 clamped_len);
2893 goto release_folio;
2894 }
2895 btrfs_folio_set_dirty(fs_info, folio, clamped_start, clamped_len);
2896
2897 /*
2898 * Set the boundary if it's inside the folio.
2899 * Data relocation requires the destination extents to have the
2900 * same size as the source.
2901 * EXTENT_BOUNDARY bit prevents current extent from being merged
2902 * with previous extent.
2903 */
2904 if (in_range(cluster->boundary[*cluster_nr] - offset,
2905 folio_start, folio_size(folio))) {
2906 u64 boundary_start = cluster->boundary[*cluster_nr] -
2907 offset;
2908 u64 boundary_end = boundary_start +
2909 fs_info->sectorsize - 1;
2910
2911 btrfs_set_extent_bit(&BTRFS_I(inode)->io_tree,
2912 boundary_start, boundary_end,
2913 EXTENT_BOUNDARY, NULL);
2914 }
2915 btrfs_unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
2916 &cached_state);
2917 btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len);
2918 cur += clamped_len;
2919
2920 /* Crossed extent end, go to next extent */
2921 if (cur >= extent_end) {
2922 (*cluster_nr)++;
2923 /* Just finished the last extent of the cluster, exit. */
2924 if (*cluster_nr >= cluster->nr)
2925 break;
2926 }
2927 }
2928 folio_unlock(folio);
2929 folio_put(folio);
2930
2931 balance_dirty_pages_ratelimited(inode->i_mapping);
2932 btrfs_throttle(fs_info);
2933 if (btrfs_should_cancel_balance(fs_info))
2934 ret = -ECANCELED;
2935 *file_offset_ret = folio_end + 1;
2936 return ret;
2937
2938 release_folio:
2939 folio_unlock(folio);
2940 folio_put(folio);
2941 return ret;
2942 }
2943
relocate_file_extent_cluster(struct reloc_control * rc)2944 static int relocate_file_extent_cluster(struct reloc_control *rc)
2945 {
2946 struct inode *inode = rc->data_inode;
2947 const struct file_extent_cluster *cluster = &rc->cluster;
2948 u64 offset = BTRFS_I(inode)->reloc_block_group_start;
2949 u64 cur_file_offset = cluster->start - offset;
2950 struct file_ra_state *ra;
2951 int cluster_nr = 0;
2952 int ret = 0;
2953
2954 if (!cluster->nr)
2955 return 0;
2956
2957 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2958 if (!ra)
2959 return -ENOMEM;
2960
2961 ret = prealloc_file_extent_cluster(rc);
2962 if (ret)
2963 goto out;
2964
2965 file_ra_state_init(ra, inode->i_mapping);
2966
2967 ret = setup_relocation_extent_mapping(rc);
2968 if (ret)
2969 goto out;
2970
2971 while (cur_file_offset < cluster->end - offset) {
2972 ret = relocate_one_folio(rc, ra, &cluster_nr, &cur_file_offset);
2973 if (ret)
2974 break;
2975 }
2976 if (ret == 0)
2977 WARN_ON(cluster_nr != cluster->nr);
2978 out:
2979 kfree(ra);
2980 return ret;
2981 }
2982
relocate_data_extent(struct reloc_control * rc,const struct btrfs_key * extent_key)2983 static noinline_for_stack int relocate_data_extent(struct reloc_control *rc,
2984 const struct btrfs_key *extent_key)
2985 {
2986 struct inode *inode = rc->data_inode;
2987 struct file_extent_cluster *cluster = &rc->cluster;
2988 int ret;
2989 struct btrfs_root *root = BTRFS_I(inode)->root;
2990
2991 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2992 ret = relocate_file_extent_cluster(rc);
2993 if (ret)
2994 return ret;
2995 cluster->nr = 0;
2996 }
2997
2998 /*
2999 * Under simple quotas, we set root->relocation_src_root when we find
3000 * the extent. If adjacent extents have different owners, we can't merge
3001 * them while relocating. Handle this by storing the owning root that
3002 * started a cluster and if we see an extent from a different root break
3003 * cluster formation (just like the above case of non-adjacent extents).
3004 *
3005 * Without simple quotas, relocation_src_root is always 0, so we should
3006 * never see a mismatch, and it should have no effect on relocation
3007 * clusters.
3008 */
3009 if (cluster->nr > 0 && cluster->owning_root != root->relocation_src_root) {
3010 u64 tmp = root->relocation_src_root;
3011
3012 /*
3013 * root->relocation_src_root is the state that actually affects
3014 * the preallocation we do here, so set it to the root owning
3015 * the cluster we need to relocate.
3016 */
3017 root->relocation_src_root = cluster->owning_root;
3018 ret = relocate_file_extent_cluster(rc);
3019 if (ret)
3020 return ret;
3021 cluster->nr = 0;
3022 /* And reset it back for the current extent's owning root. */
3023 root->relocation_src_root = tmp;
3024 }
3025
3026 if (!cluster->nr) {
3027 cluster->start = extent_key->objectid;
3028 cluster->owning_root = root->relocation_src_root;
3029 }
3030 else
3031 BUG_ON(cluster->nr >= MAX_EXTENTS);
3032 cluster->end = extent_key->objectid + extent_key->offset - 1;
3033 cluster->boundary[cluster->nr] = extent_key->objectid;
3034 cluster->nr++;
3035
3036 if (cluster->nr >= MAX_EXTENTS) {
3037 ret = relocate_file_extent_cluster(rc);
3038 if (ret)
3039 return ret;
3040 cluster->nr = 0;
3041 }
3042 return 0;
3043 }
3044
3045 /*
3046 * helper to add a tree block to the list.
3047 * the major work is getting the generation and level of the block
3048 */
add_tree_block(struct reloc_control * rc,const struct btrfs_key * extent_key,struct btrfs_path * path,struct rb_root * blocks)3049 static int add_tree_block(struct reloc_control *rc,
3050 const struct btrfs_key *extent_key,
3051 struct btrfs_path *path,
3052 struct rb_root *blocks)
3053 {
3054 struct extent_buffer *eb;
3055 struct btrfs_extent_item *ei;
3056 struct btrfs_tree_block_info *bi;
3057 struct tree_block *block;
3058 struct rb_node *rb_node;
3059 u32 item_size;
3060 int level = -1;
3061 u64 generation;
3062 u64 owner = 0;
3063
3064 eb = path->nodes[0];
3065 item_size = btrfs_item_size(eb, path->slots[0]);
3066
3067 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3068 item_size >= sizeof(*ei) + sizeof(*bi)) {
3069 unsigned long ptr = 0, end;
3070
3071 ei = btrfs_item_ptr(eb, path->slots[0],
3072 struct btrfs_extent_item);
3073 end = (unsigned long)ei + item_size;
3074 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3075 bi = (struct btrfs_tree_block_info *)(ei + 1);
3076 level = btrfs_tree_block_level(eb, bi);
3077 ptr = (unsigned long)(bi + 1);
3078 } else {
3079 level = (int)extent_key->offset;
3080 ptr = (unsigned long)(ei + 1);
3081 }
3082 generation = btrfs_extent_generation(eb, ei);
3083
3084 /*
3085 * We're reading random blocks without knowing their owner ahead
3086 * of time. This is ok most of the time, as all reloc roots and
3087 * fs roots have the same lock type. However normal trees do
3088 * not, and the only way to know ahead of time is to read the
3089 * inline ref offset. We know it's an fs root if
3090 *
3091 * 1. There's more than one ref.
3092 * 2. There's a SHARED_DATA_REF_KEY set.
3093 * 3. FULL_BACKREF is set on the flags.
3094 *
3095 * Otherwise it's safe to assume that the ref offset == the
3096 * owner of this block, so we can use that when calling
3097 * read_tree_block.
3098 */
3099 if (btrfs_extent_refs(eb, ei) == 1 &&
3100 !(btrfs_extent_flags(eb, ei) &
3101 BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3102 ptr < end) {
3103 struct btrfs_extent_inline_ref *iref;
3104 int type;
3105
3106 iref = (struct btrfs_extent_inline_ref *)ptr;
3107 type = btrfs_get_extent_inline_ref_type(eb, iref,
3108 BTRFS_REF_TYPE_BLOCK);
3109 if (type == BTRFS_REF_TYPE_INVALID)
3110 return -EINVAL;
3111 if (type == BTRFS_TREE_BLOCK_REF_KEY)
3112 owner = btrfs_extent_inline_ref_offset(eb, iref);
3113 }
3114 } else {
3115 btrfs_print_leaf(eb);
3116 btrfs_err(rc->block_group->fs_info,
3117 "unrecognized tree backref at tree block %llu slot %u",
3118 eb->start, path->slots[0]);
3119 btrfs_release_path(path);
3120 return -EUCLEAN;
3121 }
3122
3123 btrfs_release_path(path);
3124
3125 BUG_ON(level == -1);
3126
3127 block = kmalloc(sizeof(*block), GFP_NOFS);
3128 if (!block)
3129 return -ENOMEM;
3130
3131 block->bytenr = extent_key->objectid;
3132 block->key.objectid = rc->extent_root->fs_info->nodesize;
3133 block->key.offset = generation;
3134 block->level = level;
3135 block->key_ready = false;
3136 block->owner = owner;
3137
3138 rb_node = rb_simple_insert(blocks, &block->simple_node);
3139 if (rb_node)
3140 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3141 -EEXIST);
3142
3143 return 0;
3144 }
3145
3146 /*
3147 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3148 */
__add_tree_block(struct reloc_control * rc,u64 bytenr,u32 blocksize,struct rb_root * blocks)3149 static int __add_tree_block(struct reloc_control *rc,
3150 u64 bytenr, u32 blocksize,
3151 struct rb_root *blocks)
3152 {
3153 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3154 BTRFS_PATH_AUTO_FREE(path);
3155 struct btrfs_key key;
3156 int ret;
3157 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3158
3159 if (tree_block_processed(bytenr, rc))
3160 return 0;
3161
3162 if (rb_simple_search(blocks, bytenr))
3163 return 0;
3164
3165 path = btrfs_alloc_path();
3166 if (!path)
3167 return -ENOMEM;
3168 again:
3169 key.objectid = bytenr;
3170 if (skinny) {
3171 key.type = BTRFS_METADATA_ITEM_KEY;
3172 key.offset = (u64)-1;
3173 } else {
3174 key.type = BTRFS_EXTENT_ITEM_KEY;
3175 key.offset = blocksize;
3176 }
3177
3178 path->search_commit_root = 1;
3179 path->skip_locking = 1;
3180 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3181 if (ret < 0)
3182 return ret;
3183
3184 if (ret > 0 && skinny) {
3185 if (path->slots[0]) {
3186 path->slots[0]--;
3187 btrfs_item_key_to_cpu(path->nodes[0], &key,
3188 path->slots[0]);
3189 if (key.objectid == bytenr &&
3190 (key.type == BTRFS_METADATA_ITEM_KEY ||
3191 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3192 key.offset == blocksize)))
3193 ret = 0;
3194 }
3195
3196 if (ret) {
3197 skinny = false;
3198 btrfs_release_path(path);
3199 goto again;
3200 }
3201 }
3202 if (ret) {
3203 ASSERT(ret == 1);
3204 btrfs_print_leaf(path->nodes[0]);
3205 btrfs_err(fs_info,
3206 "tree block extent item (%llu) is not found in extent tree",
3207 bytenr);
3208 WARN_ON(1);
3209 return -EINVAL;
3210 }
3211
3212 return add_tree_block(rc, &key, path, blocks);
3213 }
3214
delete_block_group_cache(struct btrfs_block_group * block_group,struct inode * inode,u64 ino)3215 static int delete_block_group_cache(struct btrfs_block_group *block_group,
3216 struct inode *inode,
3217 u64 ino)
3218 {
3219 struct btrfs_fs_info *fs_info = block_group->fs_info;
3220 struct btrfs_root *root = fs_info->tree_root;
3221 struct btrfs_trans_handle *trans;
3222 struct btrfs_inode *btrfs_inode;
3223 int ret = 0;
3224
3225 if (inode)
3226 goto truncate;
3227
3228 btrfs_inode = btrfs_iget(ino, root);
3229 if (IS_ERR(btrfs_inode))
3230 return -ENOENT;
3231 inode = &btrfs_inode->vfs_inode;
3232
3233 truncate:
3234 ret = btrfs_check_trunc_cache_free_space(fs_info,
3235 &fs_info->global_block_rsv);
3236 if (ret)
3237 goto out;
3238
3239 trans = btrfs_join_transaction(root);
3240 if (IS_ERR(trans)) {
3241 ret = PTR_ERR(trans);
3242 goto out;
3243 }
3244
3245 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3246
3247 btrfs_end_transaction(trans);
3248 btrfs_btree_balance_dirty(fs_info);
3249 out:
3250 iput(inode);
3251 return ret;
3252 }
3253
3254 /*
3255 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3256 * cache inode, to avoid free space cache data extent blocking data relocation.
3257 */
delete_v1_space_cache(struct extent_buffer * leaf,struct btrfs_block_group * block_group,u64 data_bytenr)3258 static int delete_v1_space_cache(struct extent_buffer *leaf,
3259 struct btrfs_block_group *block_group,
3260 u64 data_bytenr)
3261 {
3262 u64 space_cache_ino;
3263 struct btrfs_file_extent_item *ei;
3264 struct btrfs_key key;
3265 bool found = false;
3266 int i;
3267 int ret;
3268
3269 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3270 return 0;
3271
3272 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3273 u8 type;
3274
3275 btrfs_item_key_to_cpu(leaf, &key, i);
3276 if (key.type != BTRFS_EXTENT_DATA_KEY)
3277 continue;
3278 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3279 type = btrfs_file_extent_type(leaf, ei);
3280
3281 if ((type == BTRFS_FILE_EXTENT_REG ||
3282 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3283 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3284 found = true;
3285 space_cache_ino = key.objectid;
3286 break;
3287 }
3288 }
3289 if (!found)
3290 return -ENOENT;
3291 ret = delete_block_group_cache(block_group, NULL, space_cache_ino);
3292 return ret;
3293 }
3294
3295 /*
3296 * helper to find all tree blocks that reference a given data extent
3297 */
add_data_references(struct reloc_control * rc,const struct btrfs_key * extent_key,struct btrfs_path * path,struct rb_root * blocks)3298 static noinline_for_stack int add_data_references(struct reloc_control *rc,
3299 const struct btrfs_key *extent_key,
3300 struct btrfs_path *path,
3301 struct rb_root *blocks)
3302 {
3303 struct btrfs_backref_walk_ctx ctx = { 0 };
3304 struct ulist_iterator leaf_uiter;
3305 struct ulist_node *ref_node = NULL;
3306 const u32 blocksize = rc->extent_root->fs_info->nodesize;
3307 int ret = 0;
3308
3309 btrfs_release_path(path);
3310
3311 ctx.bytenr = extent_key->objectid;
3312 ctx.skip_inode_ref_list = true;
3313 ctx.fs_info = rc->extent_root->fs_info;
3314
3315 ret = btrfs_find_all_leafs(&ctx);
3316 if (ret < 0)
3317 return ret;
3318
3319 ULIST_ITER_INIT(&leaf_uiter);
3320 while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) {
3321 struct btrfs_tree_parent_check check = { 0 };
3322 struct extent_buffer *eb;
3323
3324 eb = read_tree_block(ctx.fs_info, ref_node->val, &check);
3325 if (IS_ERR(eb)) {
3326 ret = PTR_ERR(eb);
3327 break;
3328 }
3329 ret = delete_v1_space_cache(eb, rc->block_group,
3330 extent_key->objectid);
3331 free_extent_buffer(eb);
3332 if (ret < 0)
3333 break;
3334 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3335 if (ret < 0)
3336 break;
3337 }
3338 if (ret < 0)
3339 free_block_list(blocks);
3340 ulist_free(ctx.refs);
3341 return ret;
3342 }
3343
3344 /*
3345 * helper to find next unprocessed extent
3346 */
3347 static noinline_for_stack
find_next_extent(struct reloc_control * rc,struct btrfs_path * path,struct btrfs_key * extent_key)3348 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3349 struct btrfs_key *extent_key)
3350 {
3351 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3352 struct btrfs_key key;
3353 struct extent_buffer *leaf;
3354 u64 start, end, last;
3355 int ret;
3356
3357 last = rc->block_group->start + rc->block_group->length;
3358 while (1) {
3359 bool block_found;
3360
3361 cond_resched();
3362 if (rc->search_start >= last) {
3363 ret = 1;
3364 break;
3365 }
3366
3367 key.objectid = rc->search_start;
3368 key.type = BTRFS_EXTENT_ITEM_KEY;
3369 key.offset = 0;
3370
3371 path->search_commit_root = 1;
3372 path->skip_locking = 1;
3373 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3374 0, 0);
3375 if (ret < 0)
3376 break;
3377 next:
3378 leaf = path->nodes[0];
3379 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3380 ret = btrfs_next_leaf(rc->extent_root, path);
3381 if (ret != 0)
3382 break;
3383 leaf = path->nodes[0];
3384 }
3385
3386 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3387 if (key.objectid >= last) {
3388 ret = 1;
3389 break;
3390 }
3391
3392 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3393 key.type != BTRFS_METADATA_ITEM_KEY) {
3394 path->slots[0]++;
3395 goto next;
3396 }
3397
3398 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3399 key.objectid + key.offset <= rc->search_start) {
3400 path->slots[0]++;
3401 goto next;
3402 }
3403
3404 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3405 key.objectid + fs_info->nodesize <=
3406 rc->search_start) {
3407 path->slots[0]++;
3408 goto next;
3409 }
3410
3411 block_found = btrfs_find_first_extent_bit(&rc->processed_blocks,
3412 key.objectid, &start, &end,
3413 EXTENT_DIRTY, NULL);
3414
3415 if (block_found && start <= key.objectid) {
3416 btrfs_release_path(path);
3417 rc->search_start = end + 1;
3418 } else {
3419 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3420 rc->search_start = key.objectid + key.offset;
3421 else
3422 rc->search_start = key.objectid +
3423 fs_info->nodesize;
3424 memcpy(extent_key, &key, sizeof(key));
3425 return 0;
3426 }
3427 }
3428 btrfs_release_path(path);
3429 return ret;
3430 }
3431
set_reloc_control(struct reloc_control * rc)3432 static void set_reloc_control(struct reloc_control *rc)
3433 {
3434 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3435
3436 mutex_lock(&fs_info->reloc_mutex);
3437 fs_info->reloc_ctl = rc;
3438 mutex_unlock(&fs_info->reloc_mutex);
3439 }
3440
unset_reloc_control(struct reloc_control * rc)3441 static void unset_reloc_control(struct reloc_control *rc)
3442 {
3443 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3444
3445 mutex_lock(&fs_info->reloc_mutex);
3446 fs_info->reloc_ctl = NULL;
3447 mutex_unlock(&fs_info->reloc_mutex);
3448 }
3449
3450 static noinline_for_stack
prepare_to_relocate(struct reloc_control * rc)3451 int prepare_to_relocate(struct reloc_control *rc)
3452 {
3453 struct btrfs_trans_handle *trans;
3454 int ret;
3455
3456 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3457 BTRFS_BLOCK_RSV_TEMP);
3458 if (!rc->block_rsv)
3459 return -ENOMEM;
3460
3461 memset(&rc->cluster, 0, sizeof(rc->cluster));
3462 rc->search_start = rc->block_group->start;
3463 rc->extents_found = 0;
3464 rc->nodes_relocated = 0;
3465 rc->merging_rsv_size = 0;
3466 rc->reserved_bytes = 0;
3467 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3468 RELOCATION_RESERVED_NODES;
3469 ret = btrfs_block_rsv_refill(rc->extent_root->fs_info,
3470 rc->block_rsv, rc->block_rsv->size,
3471 BTRFS_RESERVE_FLUSH_ALL);
3472 if (ret)
3473 return ret;
3474
3475 rc->create_reloc_tree = true;
3476 set_reloc_control(rc);
3477
3478 trans = btrfs_join_transaction(rc->extent_root);
3479 if (IS_ERR(trans)) {
3480 unset_reloc_control(rc);
3481 /*
3482 * extent tree is not a ref_cow tree and has no reloc_root to
3483 * cleanup. And callers are responsible to free the above
3484 * block rsv.
3485 */
3486 return PTR_ERR(trans);
3487 }
3488
3489 ret = btrfs_commit_transaction(trans);
3490 if (ret)
3491 unset_reloc_control(rc);
3492
3493 return ret;
3494 }
3495
relocate_block_group(struct reloc_control * rc)3496 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3497 {
3498 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3499 struct rb_root blocks = RB_ROOT;
3500 struct btrfs_key key;
3501 struct btrfs_trans_handle *trans = NULL;
3502 BTRFS_PATH_AUTO_FREE(path);
3503 struct btrfs_extent_item *ei;
3504 u64 flags;
3505 int ret;
3506 int err = 0;
3507 int progress = 0;
3508
3509 path = btrfs_alloc_path();
3510 if (!path)
3511 return -ENOMEM;
3512 path->reada = READA_FORWARD;
3513
3514 ret = prepare_to_relocate(rc);
3515 if (ret) {
3516 err = ret;
3517 goto out_free;
3518 }
3519
3520 while (1) {
3521 rc->reserved_bytes = 0;
3522 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
3523 rc->block_rsv->size,
3524 BTRFS_RESERVE_FLUSH_ALL);
3525 if (ret) {
3526 err = ret;
3527 break;
3528 }
3529 progress++;
3530 trans = btrfs_start_transaction(rc->extent_root, 0);
3531 if (IS_ERR(trans)) {
3532 err = PTR_ERR(trans);
3533 trans = NULL;
3534 break;
3535 }
3536 restart:
3537 if (rc->backref_cache.last_trans != trans->transid)
3538 btrfs_backref_release_cache(&rc->backref_cache);
3539 rc->backref_cache.last_trans = trans->transid;
3540
3541 ret = find_next_extent(rc, path, &key);
3542 if (ret < 0)
3543 err = ret;
3544 if (ret != 0)
3545 break;
3546
3547 rc->extents_found++;
3548
3549 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3550 struct btrfs_extent_item);
3551 flags = btrfs_extent_flags(path->nodes[0], ei);
3552
3553 /*
3554 * If we are relocating a simple quota owned extent item, we
3555 * need to note the owner on the reloc data root so that when
3556 * we allocate the replacement item, we can attribute it to the
3557 * correct eventual owner (rather than the reloc data root).
3558 */
3559 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
3560 struct btrfs_root *root = BTRFS_I(rc->data_inode)->root;
3561 u64 owning_root_id = btrfs_get_extent_owner_root(fs_info,
3562 path->nodes[0],
3563 path->slots[0]);
3564
3565 root->relocation_src_root = owning_root_id;
3566 }
3567
3568 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3569 ret = add_tree_block(rc, &key, path, &blocks);
3570 } else if (rc->stage == UPDATE_DATA_PTRS &&
3571 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3572 ret = add_data_references(rc, &key, path, &blocks);
3573 } else {
3574 btrfs_release_path(path);
3575 ret = 0;
3576 }
3577 if (ret < 0) {
3578 err = ret;
3579 break;
3580 }
3581
3582 if (!RB_EMPTY_ROOT(&blocks)) {
3583 ret = relocate_tree_blocks(trans, rc, &blocks);
3584 if (ret < 0) {
3585 if (ret != -EAGAIN) {
3586 err = ret;
3587 break;
3588 }
3589 rc->extents_found--;
3590 rc->search_start = key.objectid;
3591 }
3592 }
3593
3594 btrfs_end_transaction_throttle(trans);
3595 btrfs_btree_balance_dirty(fs_info);
3596 trans = NULL;
3597
3598 if (rc->stage == MOVE_DATA_EXTENTS &&
3599 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3600 rc->found_file_extent = true;
3601 ret = relocate_data_extent(rc, &key);
3602 if (ret < 0) {
3603 err = ret;
3604 break;
3605 }
3606 }
3607 if (btrfs_should_cancel_balance(fs_info)) {
3608 err = -ECANCELED;
3609 break;
3610 }
3611 }
3612 if (trans && progress && err == -ENOSPC) {
3613 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3614 if (ret == 1) {
3615 err = 0;
3616 progress = 0;
3617 goto restart;
3618 }
3619 }
3620
3621 btrfs_release_path(path);
3622 btrfs_clear_extent_bit(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY, NULL);
3623
3624 if (trans) {
3625 btrfs_end_transaction_throttle(trans);
3626 btrfs_btree_balance_dirty(fs_info);
3627 }
3628
3629 if (!err) {
3630 ret = relocate_file_extent_cluster(rc);
3631 if (ret < 0)
3632 err = ret;
3633 }
3634
3635 rc->create_reloc_tree = false;
3636 set_reloc_control(rc);
3637
3638 btrfs_backref_release_cache(&rc->backref_cache);
3639 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3640
3641 /*
3642 * Even in the case when the relocation is cancelled, we should all go
3643 * through prepare_to_merge() and merge_reloc_roots().
3644 *
3645 * For error (including cancelled balance), prepare_to_merge() will
3646 * mark all reloc trees orphan, then queue them for cleanup in
3647 * merge_reloc_roots()
3648 */
3649 err = prepare_to_merge(rc, err);
3650
3651 merge_reloc_roots(rc);
3652
3653 rc->merge_reloc_tree = false;
3654 unset_reloc_control(rc);
3655 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3656
3657 /* get rid of pinned extents */
3658 trans = btrfs_join_transaction(rc->extent_root);
3659 if (IS_ERR(trans)) {
3660 err = PTR_ERR(trans);
3661 goto out_free;
3662 }
3663 ret = btrfs_commit_transaction(trans);
3664 if (ret && !err)
3665 err = ret;
3666 out_free:
3667 ret = clean_dirty_subvols(rc);
3668 if (ret < 0 && !err)
3669 err = ret;
3670 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3671 return err;
3672 }
3673
__insert_orphan_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)3674 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3675 struct btrfs_root *root, u64 objectid)
3676 {
3677 BTRFS_PATH_AUTO_FREE(path);
3678 struct btrfs_inode_item *item;
3679 struct extent_buffer *leaf;
3680 int ret;
3681
3682 path = btrfs_alloc_path();
3683 if (!path)
3684 return -ENOMEM;
3685
3686 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3687 if (ret)
3688 return ret;
3689
3690 leaf = path->nodes[0];
3691 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3692 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3693 btrfs_set_inode_generation(leaf, item, 1);
3694 btrfs_set_inode_size(leaf, item, 0);
3695 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3696 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3697 BTRFS_INODE_PREALLOC);
3698 return 0;
3699 }
3700
delete_orphan_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)3701 static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3702 struct btrfs_root *root, u64 objectid)
3703 {
3704 BTRFS_PATH_AUTO_FREE(path);
3705 struct btrfs_key key;
3706 int ret = 0;
3707
3708 path = btrfs_alloc_path();
3709 if (!path) {
3710 ret = -ENOMEM;
3711 goto out;
3712 }
3713
3714 key.objectid = objectid;
3715 key.type = BTRFS_INODE_ITEM_KEY;
3716 key.offset = 0;
3717 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3718 if (ret) {
3719 if (ret > 0)
3720 ret = -ENOENT;
3721 goto out;
3722 }
3723 ret = btrfs_del_item(trans, root, path);
3724 out:
3725 if (ret)
3726 btrfs_abort_transaction(trans, ret);
3727 }
3728
3729 /*
3730 * helper to create inode for data relocation.
3731 * the inode is in data relocation tree and its link count is 0
3732 */
create_reloc_inode(const struct btrfs_block_group * group)3733 static noinline_for_stack struct inode *create_reloc_inode(
3734 const struct btrfs_block_group *group)
3735 {
3736 struct btrfs_fs_info *fs_info = group->fs_info;
3737 struct btrfs_inode *inode = NULL;
3738 struct btrfs_trans_handle *trans;
3739 struct btrfs_root *root;
3740 u64 objectid;
3741 int ret = 0;
3742
3743 root = btrfs_grab_root(fs_info->data_reloc_root);
3744 trans = btrfs_start_transaction(root, 6);
3745 if (IS_ERR(trans)) {
3746 btrfs_put_root(root);
3747 return ERR_CAST(trans);
3748 }
3749
3750 ret = btrfs_get_free_objectid(root, &objectid);
3751 if (ret)
3752 goto out;
3753
3754 ret = __insert_orphan_inode(trans, root, objectid);
3755 if (ret)
3756 goto out;
3757
3758 inode = btrfs_iget(objectid, root);
3759 if (IS_ERR(inode)) {
3760 delete_orphan_inode(trans, root, objectid);
3761 ret = PTR_ERR(inode);
3762 inode = NULL;
3763 goto out;
3764 }
3765 inode->reloc_block_group_start = group->start;
3766
3767 ret = btrfs_orphan_add(trans, inode);
3768 out:
3769 btrfs_put_root(root);
3770 btrfs_end_transaction(trans);
3771 btrfs_btree_balance_dirty(fs_info);
3772 if (ret) {
3773 if (inode)
3774 iput(&inode->vfs_inode);
3775 return ERR_PTR(ret);
3776 }
3777 return &inode->vfs_inode;
3778 }
3779
3780 /*
3781 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3782 * has been requested meanwhile and don't start in that case.
3783 *
3784 * Return:
3785 * 0 success
3786 * -EINPROGRESS operation is already in progress, that's probably a bug
3787 * -ECANCELED cancellation request was set before the operation started
3788 */
reloc_chunk_start(struct btrfs_fs_info * fs_info)3789 static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3790 {
3791 if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3792 /* This should not happen */
3793 btrfs_err(fs_info, "reloc already running, cannot start");
3794 return -EINPROGRESS;
3795 }
3796
3797 if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3798 btrfs_info(fs_info, "chunk relocation canceled on start");
3799 /*
3800 * On cancel, clear all requests but let the caller mark
3801 * the end after cleanup operations.
3802 */
3803 atomic_set(&fs_info->reloc_cancel_req, 0);
3804 return -ECANCELED;
3805 }
3806 return 0;
3807 }
3808
3809 /*
3810 * Mark end of chunk relocation that is cancellable and wake any waiters.
3811 */
reloc_chunk_end(struct btrfs_fs_info * fs_info)3812 static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
3813 {
3814 /* Requested after start, clear bit first so any waiters can continue */
3815 if (atomic_read(&fs_info->reloc_cancel_req) > 0)
3816 btrfs_info(fs_info, "chunk relocation canceled during operation");
3817 clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
3818 atomic_set(&fs_info->reloc_cancel_req, 0);
3819 }
3820
alloc_reloc_control(struct btrfs_fs_info * fs_info)3821 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3822 {
3823 struct reloc_control *rc;
3824
3825 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3826 if (!rc)
3827 return NULL;
3828
3829 INIT_LIST_HEAD(&rc->reloc_roots);
3830 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3831 btrfs_backref_init_cache(fs_info, &rc->backref_cache, true);
3832 rc->reloc_root_tree.rb_root = RB_ROOT;
3833 spin_lock_init(&rc->reloc_root_tree.lock);
3834 btrfs_extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS);
3835 return rc;
3836 }
3837
free_reloc_control(struct reloc_control * rc)3838 static void free_reloc_control(struct reloc_control *rc)
3839 {
3840 struct mapping_node *node, *tmp;
3841
3842 free_reloc_roots(&rc->reloc_roots);
3843 rbtree_postorder_for_each_entry_safe(node, tmp,
3844 &rc->reloc_root_tree.rb_root, rb_node)
3845 kfree(node);
3846
3847 kfree(rc);
3848 }
3849
3850 /*
3851 * Print the block group being relocated
3852 */
describe_relocation(struct btrfs_block_group * block_group)3853 static void describe_relocation(struct btrfs_block_group *block_group)
3854 {
3855 char buf[128] = "NONE";
3856
3857 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3858
3859 btrfs_info(block_group->fs_info, "relocating block group %llu flags %s",
3860 block_group->start, buf);
3861 }
3862
stage_to_string(enum reloc_stage stage)3863 static const char *stage_to_string(enum reloc_stage stage)
3864 {
3865 if (stage == MOVE_DATA_EXTENTS)
3866 return "move data extents";
3867 if (stage == UPDATE_DATA_PTRS)
3868 return "update data pointers";
3869 return "unknown";
3870 }
3871
3872 /*
3873 * function to relocate all extents in a block group.
3874 */
btrfs_relocate_block_group(struct btrfs_fs_info * fs_info,u64 group_start,bool verbose)3875 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start,
3876 bool verbose)
3877 {
3878 struct btrfs_block_group *bg;
3879 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start);
3880 struct reloc_control *rc;
3881 struct inode *inode;
3882 struct btrfs_path *path;
3883 int ret;
3884 int rw = 0;
3885 int err = 0;
3886
3887 /*
3888 * This only gets set if we had a half-deleted snapshot on mount. We
3889 * cannot allow relocation to start while we're still trying to clean up
3890 * these pending deletions.
3891 */
3892 ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
3893 if (ret)
3894 return ret;
3895
3896 /* We may have been woken up by close_ctree, so bail if we're closing. */
3897 if (btrfs_fs_closing(fs_info))
3898 return -EINTR;
3899
3900 bg = btrfs_lookup_block_group(fs_info, group_start);
3901 if (!bg)
3902 return -ENOENT;
3903
3904 /*
3905 * Relocation of a data block group creates ordered extents. Without
3906 * sb_start_write(), we can freeze the filesystem while unfinished
3907 * ordered extents are left. Such ordered extents can cause a deadlock
3908 * e.g. when syncfs() is waiting for their completion but they can't
3909 * finish because they block when joining a transaction, due to the
3910 * fact that the freeze locks are being held in write mode.
3911 */
3912 if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3913 ASSERT(sb_write_started(fs_info->sb));
3914
3915 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3916 btrfs_put_block_group(bg);
3917 return -ETXTBSY;
3918 }
3919
3920 rc = alloc_reloc_control(fs_info);
3921 if (!rc) {
3922 btrfs_put_block_group(bg);
3923 return -ENOMEM;
3924 }
3925
3926 ret = reloc_chunk_start(fs_info);
3927 if (ret < 0) {
3928 err = ret;
3929 goto out_put_bg;
3930 }
3931
3932 rc->extent_root = extent_root;
3933 rc->block_group = bg;
3934
3935 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3936 if (ret) {
3937 err = ret;
3938 goto out;
3939 }
3940 rw = 1;
3941
3942 path = btrfs_alloc_path();
3943 if (!path) {
3944 err = -ENOMEM;
3945 goto out;
3946 }
3947
3948 inode = lookup_free_space_inode(rc->block_group, path);
3949 btrfs_free_path(path);
3950
3951 if (!IS_ERR(inode))
3952 ret = delete_block_group_cache(rc->block_group, inode, 0);
3953 else
3954 ret = PTR_ERR(inode);
3955
3956 if (ret && ret != -ENOENT) {
3957 err = ret;
3958 goto out;
3959 }
3960
3961 rc->data_inode = create_reloc_inode(rc->block_group);
3962 if (IS_ERR(rc->data_inode)) {
3963 err = PTR_ERR(rc->data_inode);
3964 rc->data_inode = NULL;
3965 goto out;
3966 }
3967
3968 if (verbose)
3969 describe_relocation(rc->block_group);
3970
3971 btrfs_wait_block_group_reservations(rc->block_group);
3972 btrfs_wait_nocow_writers(rc->block_group);
3973 btrfs_wait_ordered_roots(fs_info, U64_MAX, rc->block_group);
3974
3975 ret = btrfs_zone_finish(rc->block_group);
3976 WARN_ON(ret && ret != -EAGAIN);
3977
3978 while (1) {
3979 enum reloc_stage finishes_stage;
3980
3981 mutex_lock(&fs_info->cleaner_mutex);
3982 ret = relocate_block_group(rc);
3983 mutex_unlock(&fs_info->cleaner_mutex);
3984 if (ret < 0)
3985 err = ret;
3986
3987 finishes_stage = rc->stage;
3988 /*
3989 * We may have gotten ENOSPC after we already dirtied some
3990 * extents. If writeout happens while we're relocating a
3991 * different block group we could end up hitting the
3992 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3993 * btrfs_reloc_cow_block. Make sure we write everything out
3994 * properly so we don't trip over this problem, and then break
3995 * out of the loop if we hit an error.
3996 */
3997 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3998 ret = btrfs_wait_ordered_range(BTRFS_I(rc->data_inode), 0,
3999 (u64)-1);
4000 if (ret)
4001 err = ret;
4002 invalidate_mapping_pages(rc->data_inode->i_mapping,
4003 0, -1);
4004 rc->stage = UPDATE_DATA_PTRS;
4005 }
4006
4007 if (err < 0)
4008 goto out;
4009
4010 if (rc->extents_found == 0)
4011 break;
4012
4013 if (verbose)
4014 btrfs_info(fs_info, "found %llu extents, stage: %s",
4015 rc->extents_found,
4016 stage_to_string(finishes_stage));
4017 }
4018
4019 WARN_ON(rc->block_group->pinned > 0);
4020 WARN_ON(rc->block_group->reserved > 0);
4021 WARN_ON(rc->block_group->used > 0);
4022 out:
4023 if (err && rw)
4024 btrfs_dec_block_group_ro(rc->block_group);
4025 iput(rc->data_inode);
4026 out_put_bg:
4027 btrfs_put_block_group(bg);
4028 reloc_chunk_end(fs_info);
4029 free_reloc_control(rc);
4030 return err;
4031 }
4032
mark_garbage_root(struct btrfs_root * root)4033 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4034 {
4035 struct btrfs_fs_info *fs_info = root->fs_info;
4036 struct btrfs_trans_handle *trans;
4037 int ret, err;
4038
4039 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4040 if (IS_ERR(trans))
4041 return PTR_ERR(trans);
4042
4043 memset(&root->root_item.drop_progress, 0,
4044 sizeof(root->root_item.drop_progress));
4045 btrfs_set_root_drop_level(&root->root_item, 0);
4046 btrfs_set_root_refs(&root->root_item, 0);
4047 ret = btrfs_update_root(trans, fs_info->tree_root,
4048 &root->root_key, &root->root_item);
4049
4050 err = btrfs_end_transaction(trans);
4051 if (err)
4052 return err;
4053 return ret;
4054 }
4055
4056 /*
4057 * recover relocation interrupted by system crash.
4058 *
4059 * this function resumes merging reloc trees with corresponding fs trees.
4060 * this is important for keeping the sharing of tree blocks
4061 */
btrfs_recover_relocation(struct btrfs_fs_info * fs_info)4062 int btrfs_recover_relocation(struct btrfs_fs_info *fs_info)
4063 {
4064 LIST_HEAD(reloc_roots);
4065 struct btrfs_key key;
4066 struct btrfs_root *fs_root;
4067 struct btrfs_root *reloc_root;
4068 struct btrfs_path *path;
4069 struct extent_buffer *leaf;
4070 struct reloc_control *rc = NULL;
4071 struct btrfs_trans_handle *trans;
4072 int ret2;
4073 int ret = 0;
4074
4075 path = btrfs_alloc_path();
4076 if (!path)
4077 return -ENOMEM;
4078 path->reada = READA_BACK;
4079
4080 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4081 key.type = BTRFS_ROOT_ITEM_KEY;
4082 key.offset = (u64)-1;
4083
4084 while (1) {
4085 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4086 path, 0, 0);
4087 if (ret < 0)
4088 goto out;
4089 if (ret > 0) {
4090 if (path->slots[0] == 0)
4091 break;
4092 path->slots[0]--;
4093 }
4094 ret = 0;
4095 leaf = path->nodes[0];
4096 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4097 btrfs_release_path(path);
4098
4099 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4100 key.type != BTRFS_ROOT_ITEM_KEY)
4101 break;
4102
4103 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key);
4104 if (IS_ERR(reloc_root)) {
4105 ret = PTR_ERR(reloc_root);
4106 goto out;
4107 }
4108
4109 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4110 list_add(&reloc_root->root_list, &reloc_roots);
4111
4112 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4113 fs_root = btrfs_get_fs_root(fs_info,
4114 reloc_root->root_key.offset, false);
4115 if (IS_ERR(fs_root)) {
4116 ret = PTR_ERR(fs_root);
4117 if (ret != -ENOENT)
4118 goto out;
4119 ret = mark_garbage_root(reloc_root);
4120 if (ret < 0)
4121 goto out;
4122 ret = 0;
4123 } else {
4124 btrfs_put_root(fs_root);
4125 }
4126 }
4127
4128 if (key.offset == 0)
4129 break;
4130
4131 key.offset--;
4132 }
4133 btrfs_release_path(path);
4134
4135 if (list_empty(&reloc_roots))
4136 goto out;
4137
4138 rc = alloc_reloc_control(fs_info);
4139 if (!rc) {
4140 ret = -ENOMEM;
4141 goto out;
4142 }
4143
4144 ret = reloc_chunk_start(fs_info);
4145 if (ret < 0)
4146 goto out_end;
4147
4148 rc->extent_root = btrfs_extent_root(fs_info, 0);
4149
4150 set_reloc_control(rc);
4151
4152 trans = btrfs_join_transaction(rc->extent_root);
4153 if (IS_ERR(trans)) {
4154 ret = PTR_ERR(trans);
4155 goto out_unset;
4156 }
4157
4158 rc->merge_reloc_tree = true;
4159
4160 while (!list_empty(&reloc_roots)) {
4161 reloc_root = list_first_entry(&reloc_roots, struct btrfs_root, root_list);
4162 list_del(&reloc_root->root_list);
4163
4164 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4165 list_add_tail(&reloc_root->root_list,
4166 &rc->reloc_roots);
4167 continue;
4168 }
4169
4170 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4171 false);
4172 if (IS_ERR(fs_root)) {
4173 ret = PTR_ERR(fs_root);
4174 list_add_tail(&reloc_root->root_list, &reloc_roots);
4175 btrfs_end_transaction(trans);
4176 goto out_unset;
4177 }
4178
4179 ret = __add_reloc_root(reloc_root);
4180 ASSERT(ret != -EEXIST);
4181 if (ret) {
4182 list_add_tail(&reloc_root->root_list, &reloc_roots);
4183 btrfs_put_root(fs_root);
4184 btrfs_end_transaction(trans);
4185 goto out_unset;
4186 }
4187 fs_root->reloc_root = btrfs_grab_root(reloc_root);
4188 btrfs_put_root(fs_root);
4189 }
4190
4191 ret = btrfs_commit_transaction(trans);
4192 if (ret)
4193 goto out_unset;
4194
4195 merge_reloc_roots(rc);
4196
4197 unset_reloc_control(rc);
4198
4199 trans = btrfs_join_transaction(rc->extent_root);
4200 if (IS_ERR(trans)) {
4201 ret = PTR_ERR(trans);
4202 goto out_clean;
4203 }
4204 ret = btrfs_commit_transaction(trans);
4205 out_clean:
4206 ret2 = clean_dirty_subvols(rc);
4207 if (ret2 < 0 && !ret)
4208 ret = ret2;
4209 out_unset:
4210 unset_reloc_control(rc);
4211 out_end:
4212 reloc_chunk_end(fs_info);
4213 free_reloc_control(rc);
4214 out:
4215 free_reloc_roots(&reloc_roots);
4216
4217 btrfs_free_path(path);
4218
4219 if (ret == 0) {
4220 /* cleanup orphan inode in data relocation tree */
4221 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4222 ASSERT(fs_root);
4223 ret = btrfs_orphan_cleanup(fs_root);
4224 btrfs_put_root(fs_root);
4225 }
4226 return ret;
4227 }
4228
4229 /*
4230 * helper to add ordered checksum for data relocation.
4231 *
4232 * cloning checksum properly handles the nodatasum extents.
4233 * it also saves CPU time to re-calculate the checksum.
4234 */
btrfs_reloc_clone_csums(struct btrfs_ordered_extent * ordered)4235 int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered)
4236 {
4237 struct btrfs_inode *inode = ordered->inode;
4238 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4239 u64 disk_bytenr = ordered->file_offset + inode->reloc_block_group_start;
4240 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr);
4241 LIST_HEAD(list);
4242 int ret;
4243
4244 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4245 disk_bytenr + ordered->num_bytes - 1,
4246 &list, false);
4247 if (ret < 0) {
4248 btrfs_mark_ordered_extent_error(ordered);
4249 return ret;
4250 }
4251
4252 while (!list_empty(&list)) {
4253 struct btrfs_ordered_sum *sums =
4254 list_first_entry(&list, struct btrfs_ordered_sum, list);
4255
4256 list_del_init(&sums->list);
4257
4258 /*
4259 * We need to offset the new_bytenr based on where the csum is.
4260 * We need to do this because we will read in entire prealloc
4261 * extents but we may have written to say the middle of the
4262 * prealloc extent, so we need to make sure the csum goes with
4263 * the right disk offset.
4264 *
4265 * We can do this because the data reloc inode refers strictly
4266 * to the on disk bytes, so we don't have to worry about
4267 * disk_len vs real len like with real inodes since it's all
4268 * disk length.
4269 */
4270 sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr;
4271 btrfs_add_ordered_sum(ordered, sums);
4272 }
4273
4274 return 0;
4275 }
4276
btrfs_reloc_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct extent_buffer * buf,struct extent_buffer * cow)4277 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4278 struct btrfs_root *root,
4279 const 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 && btrfs_is_data_reloc_root(root));
4294
4295 level = btrfs_header_level(buf);
4296 if (btrfs_header_generation(buf) <=
4297 btrfs_root_last_snapshot(&root->root_item))
4298 first_cow = 1;
4299
4300 if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID && rc->create_reloc_tree) {
4301 WARN_ON(!first_cow && level == 0);
4302
4303 node = rc->backref_cache.path[level];
4304
4305 /*
4306 * If node->bytenr != buf->start and node->new_bytenr !=
4307 * buf->start then we've got the wrong backref node for what we
4308 * expected to see here and the cache is incorrect.
4309 */
4310 if (unlikely(node->bytenr != buf->start && node->new_bytenr != buf->start)) {
4311 btrfs_err(fs_info,
4312 "bytenr %llu was found but our backref cache was expecting %llu or %llu",
4313 buf->start, node->bytenr, node->new_bytenr);
4314 return -EUCLEAN;
4315 }
4316
4317 btrfs_backref_drop_node_buffer(node);
4318 refcount_inc(&cow->refs);
4319 node->eb = cow;
4320 node->new_bytenr = cow->start;
4321
4322 if (!node->pending) {
4323 list_move_tail(&node->list,
4324 &rc->backref_cache.pending[level]);
4325 node->pending = 1;
4326 }
4327
4328 if (first_cow)
4329 mark_block_processed(rc, node);
4330
4331 if (first_cow && level > 0)
4332 rc->nodes_relocated += buf->len;
4333 }
4334
4335 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4336 ret = replace_file_extents(trans, rc, root, cow);
4337 return ret;
4338 }
4339
4340 /*
4341 * called before creating snapshot. it calculates metadata reservation
4342 * required for relocating tree blocks in the snapshot
4343 */
btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot * pending,u64 * bytes_to_reserve)4344 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4345 u64 *bytes_to_reserve)
4346 {
4347 struct btrfs_root *root = pending->root;
4348 struct reloc_control *rc = root->fs_info->reloc_ctl;
4349
4350 if (!rc || !have_reloc_root(root))
4351 return;
4352
4353 if (!rc->merge_reloc_tree)
4354 return;
4355
4356 root = root->reloc_root;
4357 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4358 /*
4359 * relocation is in the stage of merging trees. the space
4360 * used by merging a reloc tree is twice the size of
4361 * relocated tree nodes in the worst case. half for cowing
4362 * the reloc tree, half for cowing the fs tree. the space
4363 * used by cowing the reloc tree will be freed after the
4364 * tree is dropped. if we create snapshot, cowing the fs
4365 * tree may use more space than it frees. so we need
4366 * reserve extra space.
4367 */
4368 *bytes_to_reserve += rc->nodes_relocated;
4369 }
4370
4371 /*
4372 * called after snapshot is created. migrate block reservation
4373 * and create reloc root for the newly created snapshot
4374 *
4375 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4376 * references held on the reloc_root, one for root->reloc_root and one for
4377 * rc->reloc_roots.
4378 */
btrfs_reloc_post_snapshot(struct btrfs_trans_handle * trans,struct btrfs_pending_snapshot * pending)4379 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4380 struct btrfs_pending_snapshot *pending)
4381 {
4382 struct btrfs_root *root = pending->root;
4383 struct btrfs_root *reloc_root;
4384 struct btrfs_root *new_root;
4385 struct reloc_control *rc = root->fs_info->reloc_ctl;
4386 int ret;
4387
4388 if (!rc || !have_reloc_root(root))
4389 return 0;
4390
4391 rc = root->fs_info->reloc_ctl;
4392 rc->merging_rsv_size += rc->nodes_relocated;
4393
4394 if (rc->merge_reloc_tree) {
4395 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4396 rc->block_rsv,
4397 rc->nodes_relocated, true);
4398 if (ret)
4399 return ret;
4400 }
4401
4402 new_root = pending->snap;
4403 reloc_root = create_reloc_root(trans, root->reloc_root, btrfs_root_id(new_root));
4404 if (IS_ERR(reloc_root))
4405 return PTR_ERR(reloc_root);
4406
4407 ret = __add_reloc_root(reloc_root);
4408 ASSERT(ret != -EEXIST);
4409 if (ret) {
4410 /* Pairs with create_reloc_root */
4411 btrfs_put_root(reloc_root);
4412 return ret;
4413 }
4414 new_root->reloc_root = btrfs_grab_root(reloc_root);
4415 return 0;
4416 }
4417
4418 /*
4419 * Get the current bytenr for the block group which is being relocated.
4420 *
4421 * Return U64_MAX if no running relocation.
4422 */
btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info * fs_info)4423 u64 btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info *fs_info)
4424 {
4425 u64 logical = U64_MAX;
4426
4427 lockdep_assert_held(&fs_info->reloc_mutex);
4428
4429 if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group)
4430 logical = fs_info->reloc_ctl->block_group->start;
4431 return logical;
4432 }
4433