blob: b67171e6d688362167fed9db03cdc659b88061ea [file] [log] [blame]
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001/*
2 * Copyright (C) 2009 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include <linux/pagemap.h>
21#include <linux/writeback.h>
22#include <linux/blkdev.h>
23#include <linux/rbtree.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/slab.h>
Yan Zheng5d4f98a2009-06-10 10:45:14 -040025#include "ctree.h"
26#include "disk-io.h"
27#include "transaction.h"
28#include "volumes.h"
29#include "locking.h"
30#include "btrfs_inode.h"
31#include "async-thread.h"
Josef Bacik0af3d002010-06-21 14:48:16 -040032#include "free-space-cache.h"
Li Zefan581bb052011-04-20 10:06:11 +080033#include "inode-map.h"
Yan Zheng5d4f98a2009-06-10 10:45:14 -040034
35/*
36 * backref_node, mapping_node and tree_block start with this
37 */
38struct tree_entry {
39 struct rb_node rb_node;
40 u64 bytenr;
41};
42
43/*
44 * present a tree block in the backref cache
45 */
46struct backref_node {
47 struct rb_node rb_node;
48 u64 bytenr;
Yan, Zheng3fd0a552010-05-16 10:49:59 -040049
50 u64 new_bytenr;
51 /* objectid of tree block owner, can be not uptodate */
Yan Zheng5d4f98a2009-06-10 10:45:14 -040052 u64 owner;
Yan, Zheng3fd0a552010-05-16 10:49:59 -040053 /* link to pending, changed or detached list */
54 struct list_head list;
Yan Zheng5d4f98a2009-06-10 10:45:14 -040055 /* list of upper level blocks reference this block */
56 struct list_head upper;
57 /* list of child blocks in the cache */
58 struct list_head lower;
59 /* NULL if this node is not tree root */
60 struct btrfs_root *root;
61 /* extent buffer got by COW the block */
62 struct extent_buffer *eb;
63 /* level of tree block */
64 unsigned int level:8;
Yan, Zheng3fd0a552010-05-16 10:49:59 -040065 /* is the block in non-reference counted tree */
66 unsigned int cowonly:1;
67 /* 1 if no child node in the cache */
Yan Zheng5d4f98a2009-06-10 10:45:14 -040068 unsigned int lowest:1;
69 /* is the extent buffer locked */
70 unsigned int locked:1;
71 /* has the block been processed */
72 unsigned int processed:1;
73 /* have backrefs of this block been checked */
74 unsigned int checked:1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -040075 /*
76 * 1 if corresponding block has been cowed but some upper
77 * level block pointers may not point to the new location
78 */
79 unsigned int pending:1;
80 /*
81 * 1 if the backref node isn't connected to any other
82 * backref node.
83 */
84 unsigned int detached:1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -040085};
86
87/*
88 * present a block pointer in the backref cache
89 */
90struct backref_edge {
91 struct list_head list[2];
92 struct backref_node *node[2];
Yan Zheng5d4f98a2009-06-10 10:45:14 -040093};
94
95#define LOWER 0
96#define UPPER 1
97
98struct backref_cache {
99 /* red black tree of all backref nodes in the cache */
100 struct rb_root rb_root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400101 /* for passing backref nodes to btrfs_reloc_cow_block */
102 struct backref_node *path[BTRFS_MAX_LEVEL];
103 /*
104 * list of blocks that have been cowed but some block
105 * pointers in upper level blocks may not reflect the
106 * new location
107 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400108 struct list_head pending[BTRFS_MAX_LEVEL];
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400109 /* list of backref nodes with no child node */
110 struct list_head leaves;
111 /* list of blocks that have been cowed in current transaction */
112 struct list_head changed;
113 /* list of detached backref node. */
114 struct list_head detached;
115
116 u64 last_trans;
117
118 int nr_nodes;
119 int nr_edges;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400120};
121
122/*
123 * map address of tree root to tree
124 */
125struct mapping_node {
126 struct rb_node rb_node;
127 u64 bytenr;
128 void *data;
129};
130
131struct mapping_tree {
132 struct rb_root rb_root;
133 spinlock_t lock;
134};
135
136/*
137 * present a tree block to process
138 */
139struct tree_block {
140 struct rb_node rb_node;
141 u64 bytenr;
142 struct btrfs_key key;
143 unsigned int level:8;
144 unsigned int key_ready:1;
145};
146
Yan, Zheng0257bb82009-09-24 09:17:31 -0400147#define MAX_EXTENTS 128
148
149struct file_extent_cluster {
150 u64 start;
151 u64 end;
152 u64 boundary[MAX_EXTENTS];
153 unsigned int nr;
154};
155
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400156struct reloc_control {
157 /* block group to relocate */
158 struct btrfs_block_group_cache *block_group;
159 /* extent tree */
160 struct btrfs_root *extent_root;
161 /* inode for moving data */
162 struct inode *data_inode;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400163
164 struct btrfs_block_rsv *block_rsv;
165
166 struct backref_cache backref_cache;
167
168 struct file_extent_cluster cluster;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400169 /* tree blocks have been processed */
170 struct extent_io_tree processed_blocks;
171 /* map start of tree root to corresponding reloc tree */
172 struct mapping_tree reloc_root_tree;
173 /* list of reloc trees */
174 struct list_head reloc_roots;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400175 /* size of metadata reservation for merging reloc trees */
176 u64 merging_rsv_size;
177 /* size of relocated tree nodes */
178 u64 nodes_relocated;
179
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400180 u64 search_start;
181 u64 extents_found;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400182
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400183 unsigned int stage:8;
184 unsigned int create_reloc_tree:1;
185 unsigned int merge_reloc_tree:1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400186 unsigned int found_file_extent:1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400187 unsigned int commit_transaction:1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400188};
189
190/* stages of data relocation */
191#define MOVE_DATA_EXTENTS 0
192#define UPDATE_DATA_PTRS 1
193
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400194static void remove_backref_node(struct backref_cache *cache,
195 struct backref_node *node);
196static void __mark_block_processed(struct reloc_control *rc,
197 struct backref_node *node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400198
199static void mapping_tree_init(struct mapping_tree *tree)
200{
Eric Paris6bef4d32010-02-23 19:43:04 +0000201 tree->rb_root = RB_ROOT;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400202 spin_lock_init(&tree->lock);
203}
204
205static void backref_cache_init(struct backref_cache *cache)
206{
207 int i;
Eric Paris6bef4d32010-02-23 19:43:04 +0000208 cache->rb_root = RB_ROOT;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400209 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
210 INIT_LIST_HEAD(&cache->pending[i]);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400211 INIT_LIST_HEAD(&cache->changed);
212 INIT_LIST_HEAD(&cache->detached);
213 INIT_LIST_HEAD(&cache->leaves);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400214}
215
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400216static void backref_cache_cleanup(struct backref_cache *cache)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400217{
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400218 struct backref_node *node;
219 int i;
220
221 while (!list_empty(&cache->detached)) {
222 node = list_entry(cache->detached.next,
223 struct backref_node, list);
224 remove_backref_node(cache, node);
225 }
226
227 while (!list_empty(&cache->leaves)) {
228 node = list_entry(cache->leaves.next,
229 struct backref_node, lower);
230 remove_backref_node(cache, node);
231 }
232
233 cache->last_trans = 0;
234
235 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
236 BUG_ON(!list_empty(&cache->pending[i]));
237 BUG_ON(!list_empty(&cache->changed));
238 BUG_ON(!list_empty(&cache->detached));
239 BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
240 BUG_ON(cache->nr_nodes);
241 BUG_ON(cache->nr_edges);
242}
243
244static struct backref_node *alloc_backref_node(struct backref_cache *cache)
245{
246 struct backref_node *node;
247
248 node = kzalloc(sizeof(*node), GFP_NOFS);
249 if (node) {
250 INIT_LIST_HEAD(&node->list);
251 INIT_LIST_HEAD(&node->upper);
252 INIT_LIST_HEAD(&node->lower);
253 RB_CLEAR_NODE(&node->rb_node);
254 cache->nr_nodes++;
255 }
256 return node;
257}
258
259static void free_backref_node(struct backref_cache *cache,
260 struct backref_node *node)
261{
262 if (node) {
263 cache->nr_nodes--;
264 kfree(node);
265 }
266}
267
268static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
269{
270 struct backref_edge *edge;
271
272 edge = kzalloc(sizeof(*edge), GFP_NOFS);
273 if (edge)
274 cache->nr_edges++;
275 return edge;
276}
277
278static void free_backref_edge(struct backref_cache *cache,
279 struct backref_edge *edge)
280{
281 if (edge) {
282 cache->nr_edges--;
283 kfree(edge);
284 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400285}
286
287static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
288 struct rb_node *node)
289{
290 struct rb_node **p = &root->rb_node;
291 struct rb_node *parent = NULL;
292 struct tree_entry *entry;
293
294 while (*p) {
295 parent = *p;
296 entry = rb_entry(parent, struct tree_entry, rb_node);
297
298 if (bytenr < entry->bytenr)
299 p = &(*p)->rb_left;
300 else if (bytenr > entry->bytenr)
301 p = &(*p)->rb_right;
302 else
303 return parent;
304 }
305
306 rb_link_node(node, parent, p);
307 rb_insert_color(node, root);
308 return NULL;
309}
310
311static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
312{
313 struct rb_node *n = root->rb_node;
314 struct tree_entry *entry;
315
316 while (n) {
317 entry = rb_entry(n, struct tree_entry, rb_node);
318
319 if (bytenr < entry->bytenr)
320 n = n->rb_left;
321 else if (bytenr > entry->bytenr)
322 n = n->rb_right;
323 else
324 return n;
325 }
326 return NULL;
327}
328
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400329void backref_tree_panic(struct rb_node *rb_node, int errno,
330 u64 bytenr)
331{
332
333 struct btrfs_fs_info *fs_info = NULL;
334 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
335 rb_node);
336 if (bnode->root)
337 fs_info = bnode->root->fs_info;
338 btrfs_panic(fs_info, errno, "Inconsistency in backref cache "
339 "found at offset %llu\n", (unsigned long long)bytenr);
340}
341
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400342/*
343 * walk up backref nodes until reach node presents tree root
344 */
345static struct backref_node *walk_up_backref(struct backref_node *node,
346 struct backref_edge *edges[],
347 int *index)
348{
349 struct backref_edge *edge;
350 int idx = *index;
351
352 while (!list_empty(&node->upper)) {
353 edge = list_entry(node->upper.next,
354 struct backref_edge, list[LOWER]);
355 edges[idx++] = edge;
356 node = edge->node[UPPER];
357 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400358 BUG_ON(node->detached);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400359 *index = idx;
360 return node;
361}
362
363/*
364 * walk down backref nodes to find start of next reference path
365 */
366static struct backref_node *walk_down_backref(struct backref_edge *edges[],
367 int *index)
368{
369 struct backref_edge *edge;
370 struct backref_node *lower;
371 int idx = *index;
372
373 while (idx > 0) {
374 edge = edges[idx - 1];
375 lower = edge->node[LOWER];
376 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
377 idx--;
378 continue;
379 }
380 edge = list_entry(edge->list[LOWER].next,
381 struct backref_edge, list[LOWER]);
382 edges[idx - 1] = edge;
383 *index = idx;
384 return edge->node[UPPER];
385 }
386 *index = 0;
387 return NULL;
388}
389
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400390static void unlock_node_buffer(struct backref_node *node)
391{
392 if (node->locked) {
393 btrfs_tree_unlock(node->eb);
394 node->locked = 0;
395 }
396}
397
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400398static void drop_node_buffer(struct backref_node *node)
399{
400 if (node->eb) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400401 unlock_node_buffer(node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400402 free_extent_buffer(node->eb);
403 node->eb = NULL;
404 }
405}
406
407static void drop_backref_node(struct backref_cache *tree,
408 struct backref_node *node)
409{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400410 BUG_ON(!list_empty(&node->upper));
411
412 drop_node_buffer(node);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400413 list_del(&node->list);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400414 list_del(&node->lower);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400415 if (!RB_EMPTY_NODE(&node->rb_node))
416 rb_erase(&node->rb_node, &tree->rb_root);
417 free_backref_node(tree, node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400418}
419
420/*
421 * remove a backref node from the backref cache
422 */
423static void remove_backref_node(struct backref_cache *cache,
424 struct backref_node *node)
425{
426 struct backref_node *upper;
427 struct backref_edge *edge;
428
429 if (!node)
430 return;
431
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400432 BUG_ON(!node->lowest && !node->detached);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400433 while (!list_empty(&node->upper)) {
434 edge = list_entry(node->upper.next, struct backref_edge,
435 list[LOWER]);
436 upper = edge->node[UPPER];
437 list_del(&edge->list[LOWER]);
438 list_del(&edge->list[UPPER]);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400439 free_backref_edge(cache, edge);
440
441 if (RB_EMPTY_NODE(&upper->rb_node)) {
442 BUG_ON(!list_empty(&node->upper));
443 drop_backref_node(cache, node);
444 node = upper;
445 node->lowest = 1;
446 continue;
447 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400448 /*
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400449 * add the node to leaf node list if no other
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400450 * child block cached.
451 */
452 if (list_empty(&upper->lower)) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400453 list_add_tail(&upper->lower, &cache->leaves);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400454 upper->lowest = 1;
455 }
456 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400457
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400458 drop_backref_node(cache, node);
459}
460
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400461static void update_backref_node(struct backref_cache *cache,
462 struct backref_node *node, u64 bytenr)
463{
464 struct rb_node *rb_node;
465 rb_erase(&node->rb_node, &cache->rb_root);
466 node->bytenr = bytenr;
467 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400468 if (rb_node)
469 backref_tree_panic(rb_node, -EEXIST, bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400470}
471
472/*
473 * update backref cache after a transaction commit
474 */
475static int update_backref_cache(struct btrfs_trans_handle *trans,
476 struct backref_cache *cache)
477{
478 struct backref_node *node;
479 int level = 0;
480
481 if (cache->last_trans == 0) {
482 cache->last_trans = trans->transid;
483 return 0;
484 }
485
486 if (cache->last_trans == trans->transid)
487 return 0;
488
489 /*
490 * detached nodes are used to avoid unnecessary backref
491 * lookup. transaction commit changes the extent tree.
492 * so the detached nodes are no longer useful.
493 */
494 while (!list_empty(&cache->detached)) {
495 node = list_entry(cache->detached.next,
496 struct backref_node, list);
497 remove_backref_node(cache, node);
498 }
499
500 while (!list_empty(&cache->changed)) {
501 node = list_entry(cache->changed.next,
502 struct backref_node, list);
503 list_del_init(&node->list);
504 BUG_ON(node->pending);
505 update_backref_node(cache, node, node->new_bytenr);
506 }
507
508 /*
509 * some nodes can be left in the pending list if there were
510 * errors during processing the pending nodes.
511 */
512 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
513 list_for_each_entry(node, &cache->pending[level], list) {
514 BUG_ON(!node->pending);
515 if (node->bytenr == node->new_bytenr)
516 continue;
517 update_backref_node(cache, node, node->new_bytenr);
518 }
519 }
520
521 cache->last_trans = 0;
522 return 1;
523}
524
David Sterbaf2a97a92011-05-05 12:44:41 +0200525
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400526static int should_ignore_root(struct btrfs_root *root)
527{
528 struct btrfs_root *reloc_root;
529
530 if (!root->ref_cows)
531 return 0;
532
533 reloc_root = root->reloc_root;
534 if (!reloc_root)
535 return 0;
536
537 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
538 root->fs_info->running_transaction->transid - 1)
539 return 0;
540 /*
541 * if there is reloc tree and it was created in previous
542 * transaction backref lookup can find the reloc tree,
543 * so backref node for the fs tree root is useless for
544 * relocation.
545 */
546 return 1;
547}
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400548/*
549 * find reloc tree by address of tree root
550 */
551static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
552 u64 bytenr)
553{
554 struct rb_node *rb_node;
555 struct mapping_node *node;
556 struct btrfs_root *root = NULL;
557
558 spin_lock(&rc->reloc_root_tree.lock);
559 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
560 if (rb_node) {
561 node = rb_entry(rb_node, struct mapping_node, rb_node);
562 root = (struct btrfs_root *)node->data;
563 }
564 spin_unlock(&rc->reloc_root_tree.lock);
565 return root;
566}
567
568static int is_cowonly_root(u64 root_objectid)
569{
570 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
571 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
572 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
573 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
574 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
575 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
576 return 1;
577 return 0;
578}
579
580static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
581 u64 root_objectid)
582{
583 struct btrfs_key key;
584
585 key.objectid = root_objectid;
586 key.type = BTRFS_ROOT_ITEM_KEY;
587 if (is_cowonly_root(root_objectid))
588 key.offset = 0;
589 else
590 key.offset = (u64)-1;
591
592 return btrfs_read_fs_root_no_name(fs_info, &key);
593}
594
595#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
596static noinline_for_stack
597struct btrfs_root *find_tree_root(struct reloc_control *rc,
598 struct extent_buffer *leaf,
599 struct btrfs_extent_ref_v0 *ref0)
600{
601 struct btrfs_root *root;
602 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
603 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
604
605 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
606
607 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
608 BUG_ON(IS_ERR(root));
609
610 if (root->ref_cows &&
611 generation != btrfs_root_generation(&root->root_item))
612 return NULL;
613
614 return root;
615}
616#endif
617
618static noinline_for_stack
619int find_inline_backref(struct extent_buffer *leaf, int slot,
620 unsigned long *ptr, unsigned long *end)
621{
622 struct btrfs_extent_item *ei;
623 struct btrfs_tree_block_info *bi;
624 u32 item_size;
625
626 item_size = btrfs_item_size_nr(leaf, slot);
627#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
628 if (item_size < sizeof(*ei)) {
629 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
630 return 1;
631 }
632#endif
633 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
634 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
635 BTRFS_EXTENT_FLAG_TREE_BLOCK));
636
637 if (item_size <= sizeof(*ei) + sizeof(*bi)) {
638 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
639 return 1;
640 }
641
642 bi = (struct btrfs_tree_block_info *)(ei + 1);
643 *ptr = (unsigned long)(bi + 1);
644 *end = (unsigned long)ei + item_size;
645 return 0;
646}
647
648/*
649 * build backref tree for a given tree block. root of the backref tree
650 * corresponds the tree block, leaves of the backref tree correspond
651 * roots of b-trees that reference the tree block.
652 *
653 * the basic idea of this function is check backrefs of a given block
654 * to find upper level blocks that refernece the block, and then check
655 * bakcrefs of these upper level blocks recursively. the recursion stop
656 * when tree root is reached or backrefs for the block is cached.
657 *
658 * NOTE: if we find backrefs for a block are cached, we know backrefs
659 * for all upper level blocks that directly/indirectly reference the
660 * block are also cached.
661 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400662static noinline_for_stack
663struct backref_node *build_backref_tree(struct reloc_control *rc,
664 struct btrfs_key *node_key,
665 int level, u64 bytenr)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400666{
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400667 struct backref_cache *cache = &rc->backref_cache;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400668 struct btrfs_path *path1;
669 struct btrfs_path *path2;
670 struct extent_buffer *eb;
671 struct btrfs_root *root;
672 struct backref_node *cur;
673 struct backref_node *upper;
674 struct backref_node *lower;
675 struct backref_node *node = NULL;
676 struct backref_node *exist = NULL;
677 struct backref_edge *edge;
678 struct rb_node *rb_node;
679 struct btrfs_key key;
680 unsigned long end;
681 unsigned long ptr;
682 LIST_HEAD(list);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400683 LIST_HEAD(useless);
684 int cowonly;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400685 int ret;
686 int err = 0;
687
688 path1 = btrfs_alloc_path();
689 path2 = btrfs_alloc_path();
690 if (!path1 || !path2) {
691 err = -ENOMEM;
692 goto out;
693 }
Josef Bacik026fd312011-05-13 10:32:11 -0400694 path1->reada = 1;
695 path2->reada = 2;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400696
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400697 node = alloc_backref_node(cache);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400698 if (!node) {
699 err = -ENOMEM;
700 goto out;
701 }
702
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400703 node->bytenr = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400704 node->level = level;
705 node->lowest = 1;
706 cur = node;
707again:
708 end = 0;
709 ptr = 0;
710 key.objectid = cur->bytenr;
711 key.type = BTRFS_EXTENT_ITEM_KEY;
712 key.offset = (u64)-1;
713
714 path1->search_commit_root = 1;
715 path1->skip_locking = 1;
716 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
717 0, 0);
718 if (ret < 0) {
719 err = ret;
720 goto out;
721 }
722 BUG_ON(!ret || !path1->slots[0]);
723
724 path1->slots[0]--;
725
726 WARN_ON(cur->checked);
727 if (!list_empty(&cur->upper)) {
728 /*
Justin P. Mattock70f23fd2011-05-10 10:16:21 +0200729 * the backref was added previously when processing
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400730 * backref of type BTRFS_TREE_BLOCK_REF_KEY
731 */
732 BUG_ON(!list_is_singular(&cur->upper));
733 edge = list_entry(cur->upper.next, struct backref_edge,
734 list[LOWER]);
735 BUG_ON(!list_empty(&edge->list[UPPER]));
736 exist = edge->node[UPPER];
737 /*
738 * add the upper level block to pending list if we need
739 * check its backrefs
740 */
741 if (!exist->checked)
742 list_add_tail(&edge->list[UPPER], &list);
743 } else {
744 exist = NULL;
745 }
746
747 while (1) {
748 cond_resched();
749 eb = path1->nodes[0];
750
751 if (ptr >= end) {
752 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
753 ret = btrfs_next_leaf(rc->extent_root, path1);
754 if (ret < 0) {
755 err = ret;
756 goto out;
757 }
758 if (ret > 0)
759 break;
760 eb = path1->nodes[0];
761 }
762
763 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
764 if (key.objectid != cur->bytenr) {
765 WARN_ON(exist);
766 break;
767 }
768
769 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
770 ret = find_inline_backref(eb, path1->slots[0],
771 &ptr, &end);
772 if (ret)
773 goto next;
774 }
775 }
776
777 if (ptr < end) {
778 /* update key for inline back ref */
779 struct btrfs_extent_inline_ref *iref;
780 iref = (struct btrfs_extent_inline_ref *)ptr;
781 key.type = btrfs_extent_inline_ref_type(eb, iref);
782 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
783 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
784 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
785 }
786
787 if (exist &&
788 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
789 exist->owner == key.offset) ||
790 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
791 exist->bytenr == key.offset))) {
792 exist = NULL;
793 goto next;
794 }
795
796#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
797 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
798 key.type == BTRFS_EXTENT_REF_V0_KEY) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400799 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400800 struct btrfs_extent_ref_v0 *ref0;
801 ref0 = btrfs_item_ptr(eb, path1->slots[0],
802 struct btrfs_extent_ref_v0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400803 if (key.objectid == key.offset) {
Yan, Zheng046f2642010-05-31 08:58:47 +0000804 root = find_tree_root(rc, eb, ref0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400805 if (root && !should_ignore_root(root))
806 cur->root = root;
807 else
808 list_add(&cur->list, &useless);
809 break;
810 }
Yan, Zheng046f2642010-05-31 08:58:47 +0000811 if (is_cowonly_root(btrfs_ref_root_v0(eb,
812 ref0)))
813 cur->cowonly = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400814 }
815#else
816 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
817 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
818#endif
819 if (key.objectid == key.offset) {
820 /*
821 * only root blocks of reloc trees use
822 * backref of this type.
823 */
824 root = find_reloc_root(rc, cur->bytenr);
825 BUG_ON(!root);
826 cur->root = root;
827 break;
828 }
829
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400830 edge = alloc_backref_edge(cache);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400831 if (!edge) {
832 err = -ENOMEM;
833 goto out;
834 }
835 rb_node = tree_search(&cache->rb_root, key.offset);
836 if (!rb_node) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400837 upper = alloc_backref_node(cache);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400838 if (!upper) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400839 free_backref_edge(cache, edge);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400840 err = -ENOMEM;
841 goto out;
842 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400843 upper->bytenr = key.offset;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400844 upper->level = cur->level + 1;
845 /*
846 * backrefs for the upper level block isn't
847 * cached, add the block to pending list
848 */
849 list_add_tail(&edge->list[UPPER], &list);
850 } else {
851 upper = rb_entry(rb_node, struct backref_node,
852 rb_node);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400853 BUG_ON(!upper->checked);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400854 INIT_LIST_HEAD(&edge->list[UPPER]);
855 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400856 list_add_tail(&edge->list[LOWER], &cur->upper);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400857 edge->node[LOWER] = cur;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400858 edge->node[UPPER] = upper;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400859
860 goto next;
861 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
862 goto next;
863 }
864
865 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
866 root = read_fs_root(rc->extent_root->fs_info, key.offset);
867 if (IS_ERR(root)) {
868 err = PTR_ERR(root);
869 goto out;
870 }
871
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400872 if (!root->ref_cows)
873 cur->cowonly = 1;
874
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400875 if (btrfs_root_level(&root->root_item) == cur->level) {
876 /* tree root */
877 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
878 cur->bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400879 if (should_ignore_root(root))
880 list_add(&cur->list, &useless);
881 else
882 cur->root = root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400883 break;
884 }
885
886 level = cur->level + 1;
887
888 /*
889 * searching the tree to find upper level blocks
890 * reference the block.
891 */
892 path2->search_commit_root = 1;
893 path2->skip_locking = 1;
894 path2->lowest_level = level;
895 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
896 path2->lowest_level = 0;
897 if (ret < 0) {
898 err = ret;
899 goto out;
900 }
Yan Zheng33c66f42009-07-22 09:59:00 -0400901 if (ret > 0 && path2->slots[level] > 0)
902 path2->slots[level]--;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400903
904 eb = path2->nodes[level];
905 WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
906 cur->bytenr);
907
908 lower = cur;
909 for (; level < BTRFS_MAX_LEVEL; level++) {
910 if (!path2->nodes[level]) {
911 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
912 lower->bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400913 if (should_ignore_root(root))
914 list_add(&lower->list, &useless);
915 else
916 lower->root = root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400917 break;
918 }
919
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400920 edge = alloc_backref_edge(cache);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400921 if (!edge) {
922 err = -ENOMEM;
923 goto out;
924 }
925
926 eb = path2->nodes[level];
927 rb_node = tree_search(&cache->rb_root, eb->start);
928 if (!rb_node) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400929 upper = alloc_backref_node(cache);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400930 if (!upper) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400931 free_backref_edge(cache, edge);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400932 err = -ENOMEM;
933 goto out;
934 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400935 upper->bytenr = eb->start;
936 upper->owner = btrfs_header_owner(eb);
937 upper->level = lower->level + 1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400938 if (!root->ref_cows)
939 upper->cowonly = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400940
941 /*
942 * if we know the block isn't shared
943 * we can void checking its backrefs.
944 */
945 if (btrfs_block_can_be_shared(root, eb))
946 upper->checked = 0;
947 else
948 upper->checked = 1;
949
950 /*
951 * add the block to pending list if we
952 * need check its backrefs. only block
953 * at 'cur->level + 1' is added to the
954 * tail of pending list. this guarantees
955 * we check backrefs from lower level
956 * blocks to upper level blocks.
957 */
958 if (!upper->checked &&
959 level == cur->level + 1) {
960 list_add_tail(&edge->list[UPPER],
961 &list);
962 } else
963 INIT_LIST_HEAD(&edge->list[UPPER]);
964 } else {
965 upper = rb_entry(rb_node, struct backref_node,
966 rb_node);
967 BUG_ON(!upper->checked);
968 INIT_LIST_HEAD(&edge->list[UPPER]);
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400969 if (!upper->owner)
970 upper->owner = btrfs_header_owner(eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400971 }
972 list_add_tail(&edge->list[LOWER], &lower->upper);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400973 edge->node[LOWER] = lower;
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400974 edge->node[UPPER] = upper;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400975
976 if (rb_node)
977 break;
978 lower = upper;
979 upper = NULL;
980 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200981 btrfs_release_path(path2);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400982next:
983 if (ptr < end) {
984 ptr += btrfs_extent_inline_ref_size(key.type);
985 if (ptr >= end) {
986 WARN_ON(ptr > end);
987 ptr = 0;
988 end = 0;
989 }
990 }
991 if (ptr >= end)
992 path1->slots[0]++;
993 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200994 btrfs_release_path(path1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400995
996 cur->checked = 1;
997 WARN_ON(exist);
998
999 /* the pending list isn't empty, take the first block to process */
1000 if (!list_empty(&list)) {
1001 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1002 list_del_init(&edge->list[UPPER]);
1003 cur = edge->node[UPPER];
1004 goto again;
1005 }
1006
1007 /*
1008 * everything goes well, connect backref nodes and insert backref nodes
1009 * into the cache.
1010 */
1011 BUG_ON(!node->checked);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001012 cowonly = node->cowonly;
1013 if (!cowonly) {
1014 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1015 &node->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04001016 if (rb_node)
1017 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001018 list_add_tail(&node->lower, &cache->leaves);
1019 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001020
1021 list_for_each_entry(edge, &node->upper, list[LOWER])
1022 list_add_tail(&edge->list[UPPER], &list);
1023
1024 while (!list_empty(&list)) {
1025 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1026 list_del_init(&edge->list[UPPER]);
1027 upper = edge->node[UPPER];
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001028 if (upper->detached) {
1029 list_del(&edge->list[LOWER]);
1030 lower = edge->node[LOWER];
1031 free_backref_edge(cache, edge);
1032 if (list_empty(&lower->upper))
1033 list_add(&lower->list, &useless);
1034 continue;
1035 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001036
1037 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1038 if (upper->lowest) {
1039 list_del_init(&upper->lower);
1040 upper->lowest = 0;
1041 }
1042
1043 list_add_tail(&edge->list[UPPER], &upper->lower);
1044 continue;
1045 }
1046
1047 BUG_ON(!upper->checked);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001048 BUG_ON(cowonly != upper->cowonly);
1049 if (!cowonly) {
1050 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1051 &upper->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04001052 if (rb_node)
1053 backref_tree_panic(rb_node, -EEXIST,
1054 upper->bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001055 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001056
1057 list_add_tail(&edge->list[UPPER], &upper->lower);
1058
1059 list_for_each_entry(edge, &upper->upper, list[LOWER])
1060 list_add_tail(&edge->list[UPPER], &list);
1061 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001062 /*
1063 * process useless backref nodes. backref nodes for tree leaves
1064 * are deleted from the cache. backref nodes for upper level
1065 * tree blocks are left in the cache to avoid unnecessary backref
1066 * lookup.
1067 */
1068 while (!list_empty(&useless)) {
1069 upper = list_entry(useless.next, struct backref_node, list);
1070 list_del_init(&upper->list);
1071 BUG_ON(!list_empty(&upper->upper));
1072 if (upper == node)
1073 node = NULL;
1074 if (upper->lowest) {
1075 list_del_init(&upper->lower);
1076 upper->lowest = 0;
1077 }
1078 while (!list_empty(&upper->lower)) {
1079 edge = list_entry(upper->lower.next,
1080 struct backref_edge, list[UPPER]);
1081 list_del(&edge->list[UPPER]);
1082 list_del(&edge->list[LOWER]);
1083 lower = edge->node[LOWER];
1084 free_backref_edge(cache, edge);
1085
1086 if (list_empty(&lower->upper))
1087 list_add(&lower->list, &useless);
1088 }
1089 __mark_block_processed(rc, upper);
1090 if (upper->level > 0) {
1091 list_add(&upper->list, &cache->detached);
1092 upper->detached = 1;
1093 } else {
1094 rb_erase(&upper->rb_node, &cache->rb_root);
1095 free_backref_node(cache, upper);
1096 }
1097 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001098out:
1099 btrfs_free_path(path1);
1100 btrfs_free_path(path2);
1101 if (err) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001102 while (!list_empty(&useless)) {
1103 lower = list_entry(useless.next,
1104 struct backref_node, upper);
1105 list_del_init(&lower->upper);
1106 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001107 upper = node;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001108 INIT_LIST_HEAD(&list);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001109 while (upper) {
1110 if (RB_EMPTY_NODE(&upper->rb_node)) {
1111 list_splice_tail(&upper->upper, &list);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001112 free_backref_node(cache, upper);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001113 }
1114
1115 if (list_empty(&list))
1116 break;
1117
1118 edge = list_entry(list.next, struct backref_edge,
1119 list[LOWER]);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001120 list_del(&edge->list[LOWER]);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001121 upper = edge->node[UPPER];
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001122 free_backref_edge(cache, edge);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001123 }
1124 return ERR_PTR(err);
1125 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001126 BUG_ON(node && node->detached);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001127 return node;
1128}
1129
1130/*
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001131 * helper to add backref node for the newly created snapshot.
1132 * the backref node is created by cloning backref node that
1133 * corresponds to root of source tree
1134 */
1135static int clone_backref_node(struct btrfs_trans_handle *trans,
1136 struct reloc_control *rc,
1137 struct btrfs_root *src,
1138 struct btrfs_root *dest)
1139{
1140 struct btrfs_root *reloc_root = src->reloc_root;
1141 struct backref_cache *cache = &rc->backref_cache;
1142 struct backref_node *node = NULL;
1143 struct backref_node *new_node;
1144 struct backref_edge *edge;
1145 struct backref_edge *new_edge;
1146 struct rb_node *rb_node;
1147
1148 if (cache->last_trans > 0)
1149 update_backref_cache(trans, cache);
1150
1151 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1152 if (rb_node) {
1153 node = rb_entry(rb_node, struct backref_node, rb_node);
1154 if (node->detached)
1155 node = NULL;
1156 else
1157 BUG_ON(node->new_bytenr != reloc_root->node->start);
1158 }
1159
1160 if (!node) {
1161 rb_node = tree_search(&cache->rb_root,
1162 reloc_root->commit_root->start);
1163 if (rb_node) {
1164 node = rb_entry(rb_node, struct backref_node,
1165 rb_node);
1166 BUG_ON(node->detached);
1167 }
1168 }
1169
1170 if (!node)
1171 return 0;
1172
1173 new_node = alloc_backref_node(cache);
1174 if (!new_node)
1175 return -ENOMEM;
1176
1177 new_node->bytenr = dest->node->start;
1178 new_node->level = node->level;
1179 new_node->lowest = node->lowest;
Yan, Zheng6848ad62011-02-14 16:00:03 -05001180 new_node->checked = 1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001181 new_node->root = dest;
1182
1183 if (!node->lowest) {
1184 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1185 new_edge = alloc_backref_edge(cache);
1186 if (!new_edge)
1187 goto fail;
1188
1189 new_edge->node[UPPER] = new_node;
1190 new_edge->node[LOWER] = edge->node[LOWER];
1191 list_add_tail(&new_edge->list[UPPER],
1192 &new_node->lower);
1193 }
Miao Xie76b9e232011-11-10 20:45:05 -05001194 } else {
1195 list_add_tail(&new_node->lower, &cache->leaves);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001196 }
1197
1198 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1199 &new_node->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04001200 if (rb_node)
1201 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001202
1203 if (!new_node->lowest) {
1204 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1205 list_add_tail(&new_edge->list[LOWER],
1206 &new_edge->node[LOWER]->upper);
1207 }
1208 }
1209 return 0;
1210fail:
1211 while (!list_empty(&new_node->lower)) {
1212 new_edge = list_entry(new_node->lower.next,
1213 struct backref_edge, list[UPPER]);
1214 list_del(&new_edge->list[UPPER]);
1215 free_backref_edge(cache, new_edge);
1216 }
1217 free_backref_node(cache, new_node);
1218 return -ENOMEM;
1219}
1220
1221/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001222 * helper to add 'address of tree root -> reloc tree' mapping
1223 */
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001224static int __must_check __add_reloc_root(struct btrfs_root *root)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001225{
1226 struct rb_node *rb_node;
1227 struct mapping_node *node;
1228 struct reloc_control *rc = root->fs_info->reloc_ctl;
1229
1230 node = kmalloc(sizeof(*node), GFP_NOFS);
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001231 if (!node)
1232 return -ENOMEM;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001233
1234 node->bytenr = root->node->start;
1235 node->data = root;
1236
1237 spin_lock(&rc->reloc_root_tree.lock);
1238 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1239 node->bytenr, &node->rb_node);
1240 spin_unlock(&rc->reloc_root_tree.lock);
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001241 if (rb_node) {
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001242 btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found "
1243 "for start=%llu while inserting into relocation "
Joe Perches533574c2012-07-30 14:40:13 -07001244 "tree\n", node->bytenr);
Dan Carpenter23291a02012-06-25 05:15:23 -06001245 kfree(node);
1246 return -EEXIST;
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001247 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001248
1249 list_add_tail(&root->root_list, &rc->reloc_roots);
1250 return 0;
1251}
1252
1253/*
1254 * helper to update/delete the 'address of tree root -> reloc tree'
1255 * mapping
1256 */
1257static int __update_reloc_root(struct btrfs_root *root, int del)
1258{
1259 struct rb_node *rb_node;
1260 struct mapping_node *node = NULL;
1261 struct reloc_control *rc = root->fs_info->reloc_ctl;
1262
1263 spin_lock(&rc->reloc_root_tree.lock);
1264 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1265 root->commit_root->start);
1266 if (rb_node) {
1267 node = rb_entry(rb_node, struct mapping_node, rb_node);
1268 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1269 }
1270 spin_unlock(&rc->reloc_root_tree.lock);
1271
Liu Bo8f71f3e2013-03-04 16:25:36 +00001272 if (!node)
1273 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001274 BUG_ON((struct btrfs_root *)node->data != root);
1275
1276 if (!del) {
1277 spin_lock(&rc->reloc_root_tree.lock);
1278 node->bytenr = root->node->start;
1279 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1280 node->bytenr, &node->rb_node);
1281 spin_unlock(&rc->reloc_root_tree.lock);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04001282 if (rb_node)
1283 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001284 } else {
Daniel J Blueman1daf3542012-04-27 12:41:46 -04001285 spin_lock(&root->fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001286 list_del_init(&root->root_list);
Daniel J Blueman1daf3542012-04-27 12:41:46 -04001287 spin_unlock(&root->fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001288 kfree(node);
1289 }
1290 return 0;
1291}
1292
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001293static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1294 struct btrfs_root *root, u64 objectid)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001295{
1296 struct btrfs_root *reloc_root;
1297 struct extent_buffer *eb;
1298 struct btrfs_root_item *root_item;
1299 struct btrfs_key root_key;
1300 int ret;
1301
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001302 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1303 BUG_ON(!root_item);
1304
1305 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1306 root_key.type = BTRFS_ROOT_ITEM_KEY;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001307 root_key.offset = objectid;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001308
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001309 if (root->root_key.objectid == objectid) {
1310 /* called by btrfs_init_reloc_root */
1311 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1312 BTRFS_TREE_RELOC_OBJECTID);
1313 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001314
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001315 btrfs_set_root_last_snapshot(&root->root_item,
1316 trans->transid - 1);
1317 } else {
1318 /*
1319 * called by btrfs_reloc_post_snapshot_hook.
1320 * the source tree is a reloc tree, all tree blocks
1321 * modified after it was created have RELOC flag
1322 * set in their headers. so it's OK to not update
1323 * the 'last_snapshot'.
1324 */
1325 ret = btrfs_copy_root(trans, root, root->node, &eb,
1326 BTRFS_TREE_RELOC_OBJECTID);
1327 BUG_ON(ret);
1328 }
1329
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001330 memcpy(root_item, &root->root_item, sizeof(*root_item));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001331 btrfs_set_root_bytenr(root_item, eb->start);
1332 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1333 btrfs_set_root_generation(root_item, trans->transid);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001334
1335 if (root->root_key.objectid == objectid) {
1336 btrfs_set_root_refs(root_item, 0);
1337 memset(&root_item->drop_progress, 0,
1338 sizeof(struct btrfs_disk_key));
1339 root_item->drop_level = 0;
1340 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001341
1342 btrfs_tree_unlock(eb);
1343 free_extent_buffer(eb);
1344
1345 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1346 &root_key, root_item);
1347 BUG_ON(ret);
1348 kfree(root_item);
1349
1350 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
1351 &root_key);
1352 BUG_ON(IS_ERR(reloc_root));
1353 reloc_root->last_trans = trans->transid;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001354 return reloc_root;
1355}
1356
1357/*
1358 * create reloc tree for a given fs tree. reloc tree is just a
1359 * snapshot of the fs tree with special root objectid.
1360 */
1361int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root)
1363{
1364 struct btrfs_root *reloc_root;
1365 struct reloc_control *rc = root->fs_info->reloc_ctl;
1366 int clear_rsv = 0;
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001367 int ret;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001368
1369 if (root->reloc_root) {
1370 reloc_root = root->reloc_root;
1371 reloc_root->last_trans = trans->transid;
1372 return 0;
1373 }
1374
1375 if (!rc || !rc->create_reloc_tree ||
1376 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1377 return 0;
1378
1379 if (!trans->block_rsv) {
1380 trans->block_rsv = rc->block_rsv;
1381 clear_rsv = 1;
1382 }
1383 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1384 if (clear_rsv)
1385 trans->block_rsv = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001386
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04001387 ret = __add_reloc_root(reloc_root);
1388 BUG_ON(ret < 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001389 root->reloc_root = reloc_root;
1390 return 0;
1391}
1392
1393/*
1394 * update root item of reloc tree
1395 */
1396int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1397 struct btrfs_root *root)
1398{
1399 struct btrfs_root *reloc_root;
1400 struct btrfs_root_item *root_item;
1401 int del = 0;
1402 int ret;
1403
1404 if (!root->reloc_root)
Chris Mason75857172011-06-13 20:00:16 -04001405 goto out;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001406
1407 reloc_root = root->reloc_root;
1408 root_item = &reloc_root->root_item;
1409
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001410 if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1411 btrfs_root_refs(root_item) == 0) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001412 root->reloc_root = NULL;
1413 del = 1;
1414 }
1415
1416 __update_reloc_root(reloc_root, del);
1417
1418 if (reloc_root->commit_root != reloc_root->node) {
1419 btrfs_set_root_node(root_item, reloc_root->node);
1420 free_extent_buffer(reloc_root->commit_root);
1421 reloc_root->commit_root = btrfs_root_node(reloc_root);
1422 }
1423
1424 ret = btrfs_update_root(trans, root->fs_info->tree_root,
1425 &reloc_root->root_key, root_item);
1426 BUG_ON(ret);
Chris Mason75857172011-06-13 20:00:16 -04001427
1428out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001429 return 0;
1430}
1431
1432/*
1433 * helper to find first cached inode with inode number >= objectid
1434 * in a subvolume
1435 */
1436static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1437{
1438 struct rb_node *node;
1439 struct rb_node *prev;
1440 struct btrfs_inode *entry;
1441 struct inode *inode;
1442
1443 spin_lock(&root->inode_lock);
1444again:
1445 node = root->inode_tree.rb_node;
1446 prev = NULL;
1447 while (node) {
1448 prev = node;
1449 entry = rb_entry(node, struct btrfs_inode, rb_node);
1450
Li Zefan33345d012011-04-20 10:31:50 +08001451 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001452 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08001453 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001454 node = node->rb_right;
1455 else
1456 break;
1457 }
1458 if (!node) {
1459 while (prev) {
1460 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08001461 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001462 node = prev;
1463 break;
1464 }
1465 prev = rb_next(prev);
1466 }
1467 }
1468 while (node) {
1469 entry = rb_entry(node, struct btrfs_inode, rb_node);
1470 inode = igrab(&entry->vfs_inode);
1471 if (inode) {
1472 spin_unlock(&root->inode_lock);
1473 return inode;
1474 }
1475
Li Zefan33345d012011-04-20 10:31:50 +08001476 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001477 if (cond_resched_lock(&root->inode_lock))
1478 goto again;
1479
1480 node = rb_next(node);
1481 }
1482 spin_unlock(&root->inode_lock);
1483 return NULL;
1484}
1485
1486static int in_block_group(u64 bytenr,
1487 struct btrfs_block_group_cache *block_group)
1488{
1489 if (bytenr >= block_group->key.objectid &&
1490 bytenr < block_group->key.objectid + block_group->key.offset)
1491 return 1;
1492 return 0;
1493}
1494
1495/*
1496 * get new location of data
1497 */
1498static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1499 u64 bytenr, u64 num_bytes)
1500{
1501 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1502 struct btrfs_path *path;
1503 struct btrfs_file_extent_item *fi;
1504 struct extent_buffer *leaf;
1505 int ret;
1506
1507 path = btrfs_alloc_path();
1508 if (!path)
1509 return -ENOMEM;
1510
1511 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
Li Zefan33345d012011-04-20 10:31:50 +08001512 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001513 bytenr, 0);
1514 if (ret < 0)
1515 goto out;
1516 if (ret > 0) {
1517 ret = -ENOENT;
1518 goto out;
1519 }
1520
1521 leaf = path->nodes[0];
1522 fi = btrfs_item_ptr(leaf, path->slots[0],
1523 struct btrfs_file_extent_item);
1524
1525 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1526 btrfs_file_extent_compression(leaf, fi) ||
1527 btrfs_file_extent_encryption(leaf, fi) ||
1528 btrfs_file_extent_other_encoding(leaf, fi));
1529
1530 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1531 ret = 1;
1532 goto out;
1533 }
1534
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001535 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001536 ret = 0;
1537out:
1538 btrfs_free_path(path);
1539 return ret;
1540}
1541
1542/*
1543 * update file extent items in the tree leaf to point to
1544 * the new locations.
1545 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001546static noinline_for_stack
1547int replace_file_extents(struct btrfs_trans_handle *trans,
1548 struct reloc_control *rc,
1549 struct btrfs_root *root,
1550 struct extent_buffer *leaf)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001551{
1552 struct btrfs_key key;
1553 struct btrfs_file_extent_item *fi;
1554 struct inode *inode = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001555 u64 parent;
1556 u64 bytenr;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001557 u64 new_bytenr = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001558 u64 num_bytes;
1559 u64 end;
1560 u32 nritems;
1561 u32 i;
1562 int ret;
1563 int first = 1;
1564 int dirty = 0;
1565
1566 if (rc->stage != UPDATE_DATA_PTRS)
1567 return 0;
1568
1569 /* reloc trees always use full backref */
1570 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1571 parent = leaf->start;
1572 else
1573 parent = 0;
1574
1575 nritems = btrfs_header_nritems(leaf);
1576 for (i = 0; i < nritems; i++) {
1577 cond_resched();
1578 btrfs_item_key_to_cpu(leaf, &key, i);
1579 if (key.type != BTRFS_EXTENT_DATA_KEY)
1580 continue;
1581 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1582 if (btrfs_file_extent_type(leaf, fi) ==
1583 BTRFS_FILE_EXTENT_INLINE)
1584 continue;
1585 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1586 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1587 if (bytenr == 0)
1588 continue;
1589 if (!in_block_group(bytenr, rc->block_group))
1590 continue;
1591
1592 /*
1593 * if we are modifying block in fs tree, wait for readpage
1594 * to complete and drop the extent cache
1595 */
1596 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001597 if (first) {
1598 inode = find_next_inode(root, key.objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001599 first = 0;
Li Zefan33345d012011-04-20 10:31:50 +08001600 } else if (inode && btrfs_ino(inode) < key.objectid) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001601 btrfs_add_delayed_iput(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001602 inode = find_next_inode(root, key.objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001603 }
Li Zefan33345d012011-04-20 10:31:50 +08001604 if (inode && btrfs_ino(inode) == key.objectid) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001605 end = key.offset +
1606 btrfs_file_extent_num_bytes(leaf, fi);
1607 WARN_ON(!IS_ALIGNED(key.offset,
1608 root->sectorsize));
1609 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1610 end--;
1611 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001612 key.offset, end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001613 if (!ret)
1614 continue;
1615
1616 btrfs_drop_extent_cache(inode, key.offset, end,
1617 1);
1618 unlock_extent(&BTRFS_I(inode)->io_tree,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001619 key.offset, end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001620 }
1621 }
1622
1623 ret = get_new_location(rc->data_inode, &new_bytenr,
1624 bytenr, num_bytes);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001625 if (ret > 0) {
1626 WARN_ON(1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001627 continue;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001628 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001629 BUG_ON(ret < 0);
1630
1631 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1632 dirty = 1;
1633
1634 key.offset -= btrfs_file_extent_offset(leaf, fi);
1635 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1636 num_bytes, parent,
1637 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001638 key.objectid, key.offset, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001639 BUG_ON(ret);
1640
1641 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1642 parent, btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001643 key.objectid, key.offset, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001644 BUG_ON(ret);
1645 }
1646 if (dirty)
1647 btrfs_mark_buffer_dirty(leaf);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001648 if (inode)
1649 btrfs_add_delayed_iput(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001650 return 0;
1651}
1652
1653static noinline_for_stack
1654int memcmp_node_keys(struct extent_buffer *eb, int slot,
1655 struct btrfs_path *path, int level)
1656{
1657 struct btrfs_disk_key key1;
1658 struct btrfs_disk_key key2;
1659 btrfs_node_key(eb, &key1, slot);
1660 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1661 return memcmp(&key1, &key2, sizeof(key1));
1662}
1663
1664/*
1665 * try to replace tree blocks in fs tree with the new blocks
1666 * in reloc tree. tree blocks haven't been modified since the
1667 * reloc tree was create can be replaced.
1668 *
1669 * if a block was replaced, level of the block + 1 is returned.
1670 * if no block got replaced, 0 is returned. if there are other
1671 * errors, a negative error number is returned.
1672 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001673static noinline_for_stack
1674int replace_path(struct btrfs_trans_handle *trans,
1675 struct btrfs_root *dest, struct btrfs_root *src,
1676 struct btrfs_path *path, struct btrfs_key *next_key,
1677 int lowest_level, int max_level)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001678{
1679 struct extent_buffer *eb;
1680 struct extent_buffer *parent;
1681 struct btrfs_key key;
1682 u64 old_bytenr;
1683 u64 new_bytenr;
1684 u64 old_ptr_gen;
1685 u64 new_ptr_gen;
1686 u64 last_snapshot;
1687 u32 blocksize;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001688 int cow = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001689 int level;
1690 int ret;
1691 int slot;
1692
1693 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1694 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001695
1696 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001697again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001698 slot = path->slots[lowest_level];
1699 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1700
1701 eb = btrfs_lock_root_node(dest);
1702 btrfs_set_lock_blocking(eb);
1703 level = btrfs_header_level(eb);
1704
1705 if (level < lowest_level) {
1706 btrfs_tree_unlock(eb);
1707 free_extent_buffer(eb);
1708 return 0;
1709 }
1710
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001711 if (cow) {
1712 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1713 BUG_ON(ret);
1714 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001715 btrfs_set_lock_blocking(eb);
1716
1717 if (next_key) {
1718 next_key->objectid = (u64)-1;
1719 next_key->type = (u8)-1;
1720 next_key->offset = (u64)-1;
1721 }
1722
1723 parent = eb;
1724 while (1) {
1725 level = btrfs_header_level(parent);
1726 BUG_ON(level < lowest_level);
1727
1728 ret = btrfs_bin_search(parent, &key, level, &slot);
1729 if (ret && slot > 0)
1730 slot--;
1731
1732 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1733 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1734
1735 old_bytenr = btrfs_node_blockptr(parent, slot);
1736 blocksize = btrfs_level_size(dest, level - 1);
1737 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1738
1739 if (level <= max_level) {
1740 eb = path->nodes[level];
1741 new_bytenr = btrfs_node_blockptr(eb,
1742 path->slots[level]);
1743 new_ptr_gen = btrfs_node_ptr_generation(eb,
1744 path->slots[level]);
1745 } else {
1746 new_bytenr = 0;
1747 new_ptr_gen = 0;
1748 }
1749
1750 if (new_bytenr > 0 && new_bytenr == old_bytenr) {
1751 WARN_ON(1);
1752 ret = level;
1753 break;
1754 }
1755
1756 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1757 memcmp_node_keys(parent, slot, path, level)) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001758 if (level <= lowest_level) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001759 ret = 0;
1760 break;
1761 }
1762
1763 eb = read_tree_block(dest, old_bytenr, blocksize,
1764 old_ptr_gen);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00001765 BUG_ON(!eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001766 btrfs_tree_lock(eb);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001767 if (cow) {
1768 ret = btrfs_cow_block(trans, dest, eb, parent,
1769 slot, &eb);
1770 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001771 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001772 btrfs_set_lock_blocking(eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001773
1774 btrfs_tree_unlock(parent);
1775 free_extent_buffer(parent);
1776
1777 parent = eb;
1778 continue;
1779 }
1780
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001781 if (!cow) {
1782 btrfs_tree_unlock(parent);
1783 free_extent_buffer(parent);
1784 cow = 1;
1785 goto again;
1786 }
1787
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001788 btrfs_node_key_to_cpu(path->nodes[level], &key,
1789 path->slots[level]);
David Sterbab3b4aa72011-04-21 01:20:15 +02001790 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001791
1792 path->lowest_level = level;
1793 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1794 path->lowest_level = 0;
1795 BUG_ON(ret);
1796
1797 /*
1798 * swap blocks in fs tree and reloc tree.
1799 */
1800 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1801 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1802 btrfs_mark_buffer_dirty(parent);
1803
1804 btrfs_set_node_blockptr(path->nodes[level],
1805 path->slots[level], old_bytenr);
1806 btrfs_set_node_ptr_generation(path->nodes[level],
1807 path->slots[level], old_ptr_gen);
1808 btrfs_mark_buffer_dirty(path->nodes[level]);
1809
1810 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1811 path->nodes[level]->start,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001812 src->root_key.objectid, level - 1, 0,
1813 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001814 BUG_ON(ret);
1815 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1816 0, dest->root_key.objectid, level - 1,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001817 0, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001818 BUG_ON(ret);
1819
1820 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1821 path->nodes[level]->start,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001822 src->root_key.objectid, level - 1, 0,
1823 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001824 BUG_ON(ret);
1825
1826 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1827 0, dest->root_key.objectid, level - 1,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001828 0, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001829 BUG_ON(ret);
1830
1831 btrfs_unlock_up_safe(path, 0);
1832
1833 ret = level;
1834 break;
1835 }
1836 btrfs_tree_unlock(parent);
1837 free_extent_buffer(parent);
1838 return ret;
1839}
1840
1841/*
1842 * helper to find next relocated block in reloc tree
1843 */
1844static noinline_for_stack
1845int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1846 int *level)
1847{
1848 struct extent_buffer *eb;
1849 int i;
1850 u64 last_snapshot;
1851 u32 nritems;
1852
1853 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1854
1855 for (i = 0; i < *level; i++) {
1856 free_extent_buffer(path->nodes[i]);
1857 path->nodes[i] = NULL;
1858 }
1859
1860 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1861 eb = path->nodes[i];
1862 nritems = btrfs_header_nritems(eb);
1863 while (path->slots[i] + 1 < nritems) {
1864 path->slots[i]++;
1865 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1866 last_snapshot)
1867 continue;
1868
1869 *level = i;
1870 return 0;
1871 }
1872 free_extent_buffer(path->nodes[i]);
1873 path->nodes[i] = NULL;
1874 }
1875 return 1;
1876}
1877
1878/*
1879 * walk down reloc tree to find relocated block of lowest level
1880 */
1881static noinline_for_stack
1882int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1883 int *level)
1884{
1885 struct extent_buffer *eb = NULL;
1886 int i;
1887 u64 bytenr;
1888 u64 ptr_gen = 0;
1889 u64 last_snapshot;
1890 u32 blocksize;
1891 u32 nritems;
1892
1893 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1894
1895 for (i = *level; i > 0; i--) {
1896 eb = path->nodes[i];
1897 nritems = btrfs_header_nritems(eb);
1898 while (path->slots[i] < nritems) {
1899 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1900 if (ptr_gen > last_snapshot)
1901 break;
1902 path->slots[i]++;
1903 }
1904 if (path->slots[i] >= nritems) {
1905 if (i == *level)
1906 break;
1907 *level = i + 1;
1908 return 0;
1909 }
1910 if (i == 1) {
1911 *level = i;
1912 return 0;
1913 }
1914
1915 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1916 blocksize = btrfs_level_size(root, i - 1);
1917 eb = read_tree_block(root, bytenr, blocksize, ptr_gen);
1918 BUG_ON(btrfs_header_level(eb) != i - 1);
1919 path->nodes[i - 1] = eb;
1920 path->slots[i - 1] = 0;
1921 }
1922 return 1;
1923}
1924
1925/*
1926 * invalidate extent cache for file extents whose key in range of
1927 * [min_key, max_key)
1928 */
1929static int invalidate_extent_cache(struct btrfs_root *root,
1930 struct btrfs_key *min_key,
1931 struct btrfs_key *max_key)
1932{
1933 struct inode *inode = NULL;
1934 u64 objectid;
1935 u64 start, end;
Li Zefan33345d012011-04-20 10:31:50 +08001936 u64 ino;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001937
1938 objectid = min_key->objectid;
1939 while (1) {
1940 cond_resched();
1941 iput(inode);
1942
1943 if (objectid > max_key->objectid)
1944 break;
1945
1946 inode = find_next_inode(root, objectid);
1947 if (!inode)
1948 break;
Li Zefan33345d012011-04-20 10:31:50 +08001949 ino = btrfs_ino(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001950
Li Zefan33345d012011-04-20 10:31:50 +08001951 if (ino > max_key->objectid) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001952 iput(inode);
1953 break;
1954 }
1955
Li Zefan33345d012011-04-20 10:31:50 +08001956 objectid = ino + 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001957 if (!S_ISREG(inode->i_mode))
1958 continue;
1959
Li Zefan33345d012011-04-20 10:31:50 +08001960 if (unlikely(min_key->objectid == ino)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001961 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1962 continue;
1963 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1964 start = 0;
1965 else {
1966 start = min_key->offset;
1967 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
1968 }
1969 } else {
1970 start = 0;
1971 }
1972
Li Zefan33345d012011-04-20 10:31:50 +08001973 if (unlikely(max_key->objectid == ino)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001974 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1975 continue;
1976 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1977 end = (u64)-1;
1978 } else {
1979 if (max_key->offset == 0)
1980 continue;
1981 end = max_key->offset;
1982 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1983 end--;
1984 }
1985 } else {
1986 end = (u64)-1;
1987 }
1988
1989 /* the lock_extent waits for readpage to complete */
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001990 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001991 btrfs_drop_extent_cache(inode, start, end, 1);
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001992 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001993 }
1994 return 0;
1995}
1996
1997static int find_next_key(struct btrfs_path *path, int level,
1998 struct btrfs_key *key)
1999
2000{
2001 while (level < BTRFS_MAX_LEVEL) {
2002 if (!path->nodes[level])
2003 break;
2004 if (path->slots[level] + 1 <
2005 btrfs_header_nritems(path->nodes[level])) {
2006 btrfs_node_key_to_cpu(path->nodes[level], key,
2007 path->slots[level] + 1);
2008 return 0;
2009 }
2010 level++;
2011 }
2012 return 1;
2013}
2014
2015/*
2016 * merge the relocated tree blocks in reloc tree with corresponding
2017 * fs tree.
2018 */
2019static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2020 struct btrfs_root *root)
2021{
2022 LIST_HEAD(inode_list);
2023 struct btrfs_key key;
2024 struct btrfs_key next_key;
2025 struct btrfs_trans_handle *trans;
2026 struct btrfs_root *reloc_root;
2027 struct btrfs_root_item *root_item;
2028 struct btrfs_path *path;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002029 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002030 int level;
2031 int max_level;
2032 int replaced = 0;
2033 int ret;
2034 int err = 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002035 u32 min_reserved;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002036
2037 path = btrfs_alloc_path();
2038 if (!path)
2039 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04002040 path->reada = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002041
2042 reloc_root = root->reloc_root;
2043 root_item = &reloc_root->root_item;
2044
2045 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2046 level = btrfs_root_level(root_item);
2047 extent_buffer_get(reloc_root->node);
2048 path->nodes[level] = reloc_root->node;
2049 path->slots[level] = 0;
2050 } else {
2051 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2052
2053 level = root_item->drop_level;
2054 BUG_ON(level == 0);
2055 path->lowest_level = level;
2056 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
Yan Zheng33c66f42009-07-22 09:59:00 -04002057 path->lowest_level = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002058 if (ret < 0) {
2059 btrfs_free_path(path);
2060 return ret;
2061 }
2062
2063 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2064 path->slots[level]);
2065 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2066
2067 btrfs_unlock_up_safe(path, 0);
2068 }
2069
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002070 min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002071 memset(&next_key, 0, sizeof(next_key));
2072
2073 while (1) {
Yan, Zhenga22285a2010-05-16 10:48:46 -04002074 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002075 BUG_ON(IS_ERR(trans));
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002076 trans->block_rsv = rc->block_rsv;
2077
Miao Xie08e007d2012-10-16 11:33:38 +00002078 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2079 BTRFS_RESERVE_FLUSH_ALL);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002080 if (ret) {
2081 BUG_ON(ret != -EAGAIN);
2082 ret = btrfs_commit_transaction(trans, root);
2083 BUG_ON(ret);
2084 continue;
2085 }
2086
2087 replaced = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002088 max_level = level;
2089
2090 ret = walk_down_reloc_tree(reloc_root, path, &level);
2091 if (ret < 0) {
2092 err = ret;
2093 goto out;
2094 }
2095 if (ret > 0)
2096 break;
2097
2098 if (!find_next_key(path, level, &key) &&
2099 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2100 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002101 } else {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002102 ret = replace_path(trans, root, reloc_root, path,
2103 &next_key, level, max_level);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002104 }
2105 if (ret < 0) {
2106 err = ret;
2107 goto out;
2108 }
2109
2110 if (ret > 0) {
2111 level = ret;
2112 btrfs_node_key_to_cpu(path->nodes[level], &key,
2113 path->slots[level]);
2114 replaced = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002115 }
2116
2117 ret = walk_up_reloc_tree(reloc_root, path, &level);
2118 if (ret > 0)
2119 break;
2120
2121 BUG_ON(level == 0);
2122 /*
2123 * save the merging progress in the drop_progress.
2124 * this is OK since root refs == 1 in this case.
2125 */
2126 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2127 path->slots[level]);
2128 root_item->drop_level = level;
2129
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002130 btrfs_end_transaction_throttle(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002131
Liu Bob53d3f52012-11-14 14:34:34 +00002132 btrfs_btree_balance_dirty(root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002133
2134 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2135 invalidate_extent_cache(root, &key, &next_key);
2136 }
2137
2138 /*
2139 * handle the case only one block in the fs tree need to be
2140 * relocated and the block is tree root.
2141 */
2142 leaf = btrfs_lock_root_node(root);
2143 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2144 btrfs_tree_unlock(leaf);
2145 free_extent_buffer(leaf);
2146 if (ret < 0)
2147 err = ret;
2148out:
2149 btrfs_free_path(path);
2150
2151 if (err == 0) {
2152 memset(&root_item->drop_progress, 0,
2153 sizeof(root_item->drop_progress));
2154 root_item->drop_level = 0;
2155 btrfs_set_root_refs(root_item, 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002156 btrfs_update_reloc_root(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002157 }
2158
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002159 btrfs_end_transaction_throttle(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002160
Liu Bob53d3f52012-11-14 14:34:34 +00002161 btrfs_btree_balance_dirty(root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002162
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002163 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2164 invalidate_extent_cache(root, &key, &next_key);
2165
2166 return err;
2167}
2168
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002169static noinline_for_stack
2170int prepare_to_merge(struct reloc_control *rc, int err)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002171{
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002172 struct btrfs_root *root = rc->extent_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002173 struct btrfs_root *reloc_root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002174 struct btrfs_trans_handle *trans;
2175 LIST_HEAD(reloc_roots);
2176 u64 num_bytes = 0;
2177 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002178
Chris Mason75857172011-06-13 20:00:16 -04002179 mutex_lock(&root->fs_info->reloc_mutex);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002180 rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2181 rc->merging_rsv_size += rc->nodes_relocated * 2;
Chris Mason75857172011-06-13 20:00:16 -04002182 mutex_unlock(&root->fs_info->reloc_mutex);
2183
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002184again:
2185 if (!err) {
2186 num_bytes = rc->merging_rsv_size;
Miao Xie08e007d2012-10-16 11:33:38 +00002187 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2188 BTRFS_RESERVE_FLUSH_ALL);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002189 if (ret)
2190 err = ret;
2191 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002192
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002193 trans = btrfs_join_transaction(rc->extent_root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00002194 if (IS_ERR(trans)) {
2195 if (!err)
2196 btrfs_block_rsv_release(rc->extent_root,
2197 rc->block_rsv, num_bytes);
2198 return PTR_ERR(trans);
2199 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002200
2201 if (!err) {
2202 if (num_bytes != rc->merging_rsv_size) {
2203 btrfs_end_transaction(trans, rc->extent_root);
2204 btrfs_block_rsv_release(rc->extent_root,
2205 rc->block_rsv, num_bytes);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002206 goto again;
2207 }
2208 }
2209
2210 rc->merge_reloc_tree = 1;
2211
2212 while (!list_empty(&rc->reloc_roots)) {
2213 reloc_root = list_entry(rc->reloc_roots.next,
2214 struct btrfs_root, root_list);
2215 list_del_init(&reloc_root->root_list);
2216
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002217 root = read_fs_root(reloc_root->fs_info,
2218 reloc_root->root_key.offset);
2219 BUG_ON(IS_ERR(root));
2220 BUG_ON(root->reloc_root != reloc_root);
2221
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002222 /*
2223 * set reference count to 1, so btrfs_recover_relocation
2224 * knows it should resumes merging
2225 */
2226 if (!err)
2227 btrfs_set_root_refs(&reloc_root->root_item, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002228 btrfs_update_reloc_root(trans, root);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002229
2230 list_add(&reloc_root->root_list, &reloc_roots);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002231 }
2232
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002233 list_splice(&reloc_roots, &rc->reloc_roots);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002234
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002235 if (!err)
2236 btrfs_commit_transaction(trans, rc->extent_root);
2237 else
2238 btrfs_end_transaction(trans, rc->extent_root);
2239 return err;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002240}
2241
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002242static noinline_for_stack
Liu Boaca1bba2013-03-04 16:25:37 +00002243void free_reloc_roots(struct list_head *list)
2244{
2245 struct btrfs_root *reloc_root;
2246
2247 while (!list_empty(list)) {
2248 reloc_root = list_entry(list->next, struct btrfs_root,
2249 root_list);
2250 __update_reloc_root(reloc_root, 1);
2251 free_extent_buffer(reloc_root->node);
2252 free_extent_buffer(reloc_root->commit_root);
2253 kfree(reloc_root);
2254 }
2255}
2256
2257static noinline_for_stack
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002258int merge_reloc_roots(struct reloc_control *rc)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002259{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002260 struct btrfs_root *root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002261 struct btrfs_root *reloc_root;
2262 LIST_HEAD(reloc_roots);
2263 int found = 0;
Liu Boaca1bba2013-03-04 16:25:37 +00002264 int ret = 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002265again:
2266 root = rc->extent_root;
Chris Mason75857172011-06-13 20:00:16 -04002267
2268 /*
2269 * this serializes us with btrfs_record_root_in_transaction,
2270 * we have to make sure nobody is in the middle of
2271 * adding their roots to the list while we are
2272 * doing this splice
2273 */
2274 mutex_lock(&root->fs_info->reloc_mutex);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002275 list_splice_init(&rc->reloc_roots, &reloc_roots);
Chris Mason75857172011-06-13 20:00:16 -04002276 mutex_unlock(&root->fs_info->reloc_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002277
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002278 while (!list_empty(&reloc_roots)) {
2279 found = 1;
2280 reloc_root = list_entry(reloc_roots.next,
2281 struct btrfs_root, root_list);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002282
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002283 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2284 root = read_fs_root(reloc_root->fs_info,
2285 reloc_root->root_key.offset);
2286 BUG_ON(IS_ERR(root));
2287 BUG_ON(root->reloc_root != reloc_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002288
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002289 ret = merge_reloc_root(rc, root);
Liu Boaca1bba2013-03-04 16:25:37 +00002290 if (ret)
2291 goto out;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002292 } else {
2293 list_del_init(&reloc_root->root_list);
2294 }
Jeff Mahoney2c536792011-10-03 23:22:41 -04002295 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
Liu Boaca1bba2013-03-04 16:25:37 +00002296 if (ret < 0) {
2297 if (list_empty(&reloc_root->root_list))
2298 list_add_tail(&reloc_root->root_list,
2299 &reloc_roots);
2300 goto out;
2301 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002302 }
2303
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002304 if (found) {
2305 found = 0;
2306 goto again;
2307 }
Liu Boaca1bba2013-03-04 16:25:37 +00002308out:
2309 if (ret) {
2310 btrfs_std_error(root->fs_info, ret);
2311 if (!list_empty(&reloc_roots))
2312 free_reloc_roots(&reloc_roots);
2313 }
2314
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002315 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
Liu Boaca1bba2013-03-04 16:25:37 +00002316 return ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002317}
2318
2319static void free_block_list(struct rb_root *blocks)
2320{
2321 struct tree_block *block;
2322 struct rb_node *rb_node;
2323 while ((rb_node = rb_first(blocks))) {
2324 block = rb_entry(rb_node, struct tree_block, rb_node);
2325 rb_erase(rb_node, blocks);
2326 kfree(block);
2327 }
2328}
2329
2330static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2331 struct btrfs_root *reloc_root)
2332{
2333 struct btrfs_root *root;
2334
2335 if (reloc_root->last_trans == trans->transid)
2336 return 0;
2337
2338 root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2339 BUG_ON(IS_ERR(root));
2340 BUG_ON(root->reloc_root != reloc_root);
2341
2342 return btrfs_record_root_in_trans(trans, root);
2343}
2344
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002345static noinline_for_stack
2346struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2347 struct reloc_control *rc,
2348 struct backref_node *node,
2349 struct backref_edge *edges[], int *nr)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002350{
2351 struct backref_node *next;
2352 struct btrfs_root *root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002353 int index = 0;
2354
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002355 next = node;
2356 while (1) {
2357 cond_resched();
2358 next = walk_up_backref(next, edges, &index);
2359 root = next->root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002360 BUG_ON(!root);
2361 BUG_ON(!root->ref_cows);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002362
2363 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2364 record_reloc_root_in_trans(trans, root);
2365 break;
2366 }
2367
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002368 btrfs_record_root_in_trans(trans, root);
2369 root = root->reloc_root;
2370
2371 if (next->new_bytenr != root->node->start) {
2372 BUG_ON(next->new_bytenr);
2373 BUG_ON(!list_empty(&next->list));
2374 next->new_bytenr = root->node->start;
2375 next->root = root;
2376 list_add_tail(&next->list,
2377 &rc->backref_cache.changed);
2378 __mark_block_processed(rc, next);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002379 break;
2380 }
2381
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002382 WARN_ON(1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002383 root = NULL;
2384 next = walk_down_backref(edges, &index);
2385 if (!next || next->level <= node->level)
2386 break;
2387 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002388 if (!root)
2389 return NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002390
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002391 *nr = index;
2392 next = node;
2393 /* setup backref node path for btrfs_reloc_cow_block */
2394 while (1) {
2395 rc->backref_cache.path[next->level] = next;
2396 if (--index < 0)
2397 break;
2398 next = edges[index]->node[UPPER];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002399 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002400 return root;
2401}
2402
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002403/*
2404 * select a tree root for relocation. return NULL if the block
2405 * is reference counted. we should use do_relocation() in this
2406 * case. return a tree root pointer if the block isn't reference
2407 * counted. return -ENOENT if the block is root of reloc tree.
2408 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002409static noinline_for_stack
2410struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
2411 struct backref_node *node)
2412{
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002413 struct backref_node *next;
2414 struct btrfs_root *root;
2415 struct btrfs_root *fs_root = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002416 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002417 int index = 0;
2418
2419 next = node;
2420 while (1) {
2421 cond_resched();
2422 next = walk_up_backref(next, edges, &index);
2423 root = next->root;
2424 BUG_ON(!root);
2425
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002426 /* no other choice for non-references counted tree */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002427 if (!root->ref_cows)
2428 return root;
2429
2430 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2431 fs_root = root;
2432
2433 if (next != node)
2434 return NULL;
2435
2436 next = walk_down_backref(edges, &index);
2437 if (!next || next->level <= node->level)
2438 break;
2439 }
2440
2441 if (!fs_root)
2442 return ERR_PTR(-ENOENT);
2443 return fs_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002444}
2445
2446static noinline_for_stack
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002447u64 calcu_metadata_size(struct reloc_control *rc,
2448 struct backref_node *node, int reserve)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002449{
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002450 struct backref_node *next = node;
2451 struct backref_edge *edge;
2452 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2453 u64 num_bytes = 0;
2454 int index = 0;
2455
2456 BUG_ON(reserve && node->processed);
2457
2458 while (next) {
2459 cond_resched();
2460 while (1) {
2461 if (next->processed && (reserve || next != node))
2462 break;
2463
2464 num_bytes += btrfs_level_size(rc->extent_root,
2465 next->level);
2466
2467 if (list_empty(&next->upper))
2468 break;
2469
2470 edge = list_entry(next->upper.next,
2471 struct backref_edge, list[LOWER]);
2472 edges[index++] = edge;
2473 next = edge->node[UPPER];
2474 }
2475 next = walk_down_backref(edges, &index);
2476 }
2477 return num_bytes;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002478}
2479
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002480static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2481 struct reloc_control *rc,
2482 struct backref_node *node)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002483{
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002484 struct btrfs_root *root = rc->extent_root;
2485 u64 num_bytes;
2486 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002487
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002488 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002489
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002490 trans->block_rsv = rc->block_rsv;
Miao Xie08e007d2012-10-16 11:33:38 +00002491 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2492 BTRFS_RESERVE_FLUSH_ALL);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002493 if (ret) {
2494 if (ret == -EAGAIN)
2495 rc->commit_transaction = 1;
2496 return ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002497 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002498
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002499 return 0;
2500}
2501
2502static void release_metadata_space(struct reloc_control *rc,
2503 struct backref_node *node)
2504{
2505 u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
2506 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002507}
2508
2509/*
2510 * relocate a block tree, and then update pointers in upper level
2511 * blocks that reference the block to point to the new location.
2512 *
2513 * if called by link_to_upper, the block has already been relocated.
2514 * in that case this function just updates pointers.
2515 */
2516static int do_relocation(struct btrfs_trans_handle *trans,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002517 struct reloc_control *rc,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002518 struct backref_node *node,
2519 struct btrfs_key *key,
2520 struct btrfs_path *path, int lowest)
2521{
2522 struct backref_node *upper;
2523 struct backref_edge *edge;
2524 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2525 struct btrfs_root *root;
2526 struct extent_buffer *eb;
2527 u32 blocksize;
2528 u64 bytenr;
2529 u64 generation;
2530 int nr;
2531 int slot;
2532 int ret;
2533 int err = 0;
2534
2535 BUG_ON(lowest && node->eb);
2536
2537 path->lowest_level = node->level + 1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002538 rc->backref_cache.path[node->level] = node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002539 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2540 cond_resched();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002541
2542 upper = edge->node[UPPER];
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002543 root = select_reloc_root(trans, rc, upper, edges, &nr);
2544 BUG_ON(!root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002545
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002546 if (upper->eb && !upper->locked) {
2547 if (!lowest) {
2548 ret = btrfs_bin_search(upper->eb, key,
2549 upper->level, &slot);
2550 BUG_ON(ret);
2551 bytenr = btrfs_node_blockptr(upper->eb, slot);
2552 if (node->eb->start == bytenr)
2553 goto next;
2554 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002555 drop_node_buffer(upper);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002556 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002557
2558 if (!upper->eb) {
2559 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2560 if (ret < 0) {
2561 err = ret;
2562 break;
2563 }
2564 BUG_ON(ret > 0);
2565
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002566 if (!upper->eb) {
2567 upper->eb = path->nodes[upper->level];
2568 path->nodes[upper->level] = NULL;
2569 } else {
2570 BUG_ON(upper->eb != path->nodes[upper->level]);
2571 }
2572
2573 upper->locked = 1;
2574 path->locks[upper->level] = 0;
2575
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002576 slot = path->slots[upper->level];
David Sterbab3b4aa72011-04-21 01:20:15 +02002577 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002578 } else {
2579 ret = btrfs_bin_search(upper->eb, key, upper->level,
2580 &slot);
2581 BUG_ON(ret);
2582 }
2583
2584 bytenr = btrfs_node_blockptr(upper->eb, slot);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002585 if (lowest) {
2586 BUG_ON(bytenr != node->bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002587 } else {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002588 if (node->eb->start == bytenr)
2589 goto next;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002590 }
2591
2592 blocksize = btrfs_level_size(root, node->level);
2593 generation = btrfs_node_ptr_generation(upper->eb, slot);
2594 eb = read_tree_block(root, bytenr, blocksize, generation);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00002595 if (!eb) {
2596 err = -EIO;
2597 goto next;
2598 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002599 btrfs_tree_lock(eb);
2600 btrfs_set_lock_blocking(eb);
2601
2602 if (!node->eb) {
2603 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2604 slot, &eb);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002605 btrfs_tree_unlock(eb);
2606 free_extent_buffer(eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002607 if (ret < 0) {
2608 err = ret;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002609 goto next;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002610 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002611 BUG_ON(node->eb != eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002612 } else {
2613 btrfs_set_node_blockptr(upper->eb, slot,
2614 node->eb->start);
2615 btrfs_set_node_ptr_generation(upper->eb, slot,
2616 trans->transid);
2617 btrfs_mark_buffer_dirty(upper->eb);
2618
2619 ret = btrfs_inc_extent_ref(trans, root,
2620 node->eb->start, blocksize,
2621 upper->eb->start,
2622 btrfs_header_owner(upper->eb),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002623 node->level, 0, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002624 BUG_ON(ret);
2625
2626 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2627 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002628 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002629next:
2630 if (!upper->pending)
2631 drop_node_buffer(upper);
2632 else
2633 unlock_node_buffer(upper);
2634 if (err)
2635 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002636 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002637
2638 if (!err && node->pending) {
2639 drop_node_buffer(node);
2640 list_move_tail(&node->list, &rc->backref_cache.changed);
2641 node->pending = 0;
2642 }
2643
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002644 path->lowest_level = 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002645 BUG_ON(err == -ENOSPC);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002646 return err;
2647}
2648
2649static int link_to_upper(struct btrfs_trans_handle *trans,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002650 struct reloc_control *rc,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002651 struct backref_node *node,
2652 struct btrfs_path *path)
2653{
2654 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002655
2656 btrfs_node_key_to_cpu(node->eb, &key, 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002657 return do_relocation(trans, rc, node, &key, path, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002658}
2659
2660static int finish_pending_nodes(struct btrfs_trans_handle *trans,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002661 struct reloc_control *rc,
2662 struct btrfs_path *path, int err)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002663{
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002664 LIST_HEAD(list);
2665 struct backref_cache *cache = &rc->backref_cache;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002666 struct backref_node *node;
2667 int level;
2668 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002669
2670 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2671 while (!list_empty(&cache->pending[level])) {
2672 node = list_entry(cache->pending[level].next,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002673 struct backref_node, list);
2674 list_move_tail(&node->list, &list);
2675 BUG_ON(!node->pending);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002676
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002677 if (!err) {
2678 ret = link_to_upper(trans, rc, node, path);
2679 if (ret < 0)
2680 err = ret;
2681 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002682 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002683 list_splice_init(&list, &cache->pending[level]);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002684 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002685 return err;
2686}
2687
2688static void mark_block_processed(struct reloc_control *rc,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002689 u64 bytenr, u32 blocksize)
2690{
2691 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2692 EXTENT_DIRTY, GFP_NOFS);
2693}
2694
2695static void __mark_block_processed(struct reloc_control *rc,
2696 struct backref_node *node)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002697{
2698 u32 blocksize;
2699 if (node->level == 0 ||
2700 in_block_group(node->bytenr, rc->block_group)) {
2701 blocksize = btrfs_level_size(rc->extent_root, node->level);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002702 mark_block_processed(rc, node->bytenr, blocksize);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002703 }
2704 node->processed = 1;
2705}
2706
2707/*
2708 * mark a block and all blocks directly/indirectly reference the block
2709 * as processed.
2710 */
2711static void update_processed_blocks(struct reloc_control *rc,
2712 struct backref_node *node)
2713{
2714 struct backref_node *next = node;
2715 struct backref_edge *edge;
2716 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2717 int index = 0;
2718
2719 while (next) {
2720 cond_resched();
2721 while (1) {
2722 if (next->processed)
2723 break;
2724
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002725 __mark_block_processed(rc, next);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002726
2727 if (list_empty(&next->upper))
2728 break;
2729
2730 edge = list_entry(next->upper.next,
2731 struct backref_edge, list[LOWER]);
2732 edges[index++] = edge;
2733 next = edge->node[UPPER];
2734 }
2735 next = walk_down_backref(edges, &index);
2736 }
2737}
2738
2739static int tree_block_processed(u64 bytenr, u32 blocksize,
2740 struct reloc_control *rc)
2741{
2742 if (test_range_bit(&rc->processed_blocks, bytenr,
Chris Mason9655d292009-09-02 15:22:30 -04002743 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002744 return 1;
2745 return 0;
2746}
2747
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002748static int get_tree_block_key(struct reloc_control *rc,
2749 struct tree_block *block)
2750{
2751 struct extent_buffer *eb;
2752
2753 BUG_ON(block->key_ready);
2754 eb = read_tree_block(rc->extent_root, block->bytenr,
2755 block->key.objectid, block->key.offset);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00002756 BUG_ON(!eb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002757 WARN_ON(btrfs_header_level(eb) != block->level);
2758 if (block->level == 0)
2759 btrfs_item_key_to_cpu(eb, &block->key, 0);
2760 else
2761 btrfs_node_key_to_cpu(eb, &block->key, 0);
2762 free_extent_buffer(eb);
2763 block->key_ready = 1;
2764 return 0;
2765}
2766
2767static int reada_tree_block(struct reloc_control *rc,
2768 struct tree_block *block)
2769{
2770 BUG_ON(block->key_ready);
2771 readahead_tree_block(rc->extent_root, block->bytenr,
2772 block->key.objectid, block->key.offset);
2773 return 0;
2774}
2775
2776/*
2777 * helper function to relocate a tree block
2778 */
2779static int relocate_tree_block(struct btrfs_trans_handle *trans,
2780 struct reloc_control *rc,
2781 struct backref_node *node,
2782 struct btrfs_key *key,
2783 struct btrfs_path *path)
2784{
2785 struct btrfs_root *root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002786 int release = 0;
2787 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002788
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002789 if (!node)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002790 return 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002791
2792 BUG_ON(node->processed);
2793 root = select_one_root(trans, node);
2794 if (root == ERR_PTR(-ENOENT)) {
2795 update_processed_blocks(rc, node);
2796 goto out;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002797 }
2798
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002799 if (!root || root->ref_cows) {
2800 ret = reserve_metadata_space(trans, rc, node);
2801 if (ret)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002802 goto out;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002803 release = 1;
2804 }
2805
2806 if (root) {
2807 if (root->ref_cows) {
2808 BUG_ON(node->new_bytenr);
2809 BUG_ON(!list_empty(&node->list));
2810 btrfs_record_root_in_trans(trans, root);
2811 root = root->reloc_root;
2812 node->new_bytenr = root->node->start;
2813 node->root = root;
2814 list_add_tail(&node->list, &rc->backref_cache.changed);
2815 } else {
2816 path->lowest_level = node->level;
2817 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
David Sterbab3b4aa72011-04-21 01:20:15 +02002818 btrfs_release_path(path);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002819 if (ret > 0)
2820 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002821 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002822 if (!ret)
2823 update_processed_blocks(rc, node);
2824 } else {
2825 ret = do_relocation(trans, rc, node, key, path, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002826 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002827out:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002828 if (ret || node->level == 0 || node->cowonly) {
2829 if (release)
2830 release_metadata_space(rc, node);
2831 remove_backref_node(&rc->backref_cache, node);
2832 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002833 return ret;
2834}
2835
2836/*
2837 * relocate a list of blocks
2838 */
2839static noinline_for_stack
2840int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2841 struct reloc_control *rc, struct rb_root *blocks)
2842{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002843 struct backref_node *node;
2844 struct btrfs_path *path;
2845 struct tree_block *block;
2846 struct rb_node *rb_node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002847 int ret;
2848 int err = 0;
2849
2850 path = btrfs_alloc_path();
Liu Boe1a12672013-03-04 16:25:38 +00002851 if (!path) {
2852 err = -ENOMEM;
2853 goto out_path;
2854 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002855
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002856 rb_node = rb_first(blocks);
2857 while (rb_node) {
2858 block = rb_entry(rb_node, struct tree_block, rb_node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002859 if (!block->key_ready)
2860 reada_tree_block(rc, block);
2861 rb_node = rb_next(rb_node);
2862 }
2863
2864 rb_node = rb_first(blocks);
2865 while (rb_node) {
2866 block = rb_entry(rb_node, struct tree_block, rb_node);
2867 if (!block->key_ready)
2868 get_tree_block_key(rc, block);
2869 rb_node = rb_next(rb_node);
2870 }
2871
2872 rb_node = rb_first(blocks);
2873 while (rb_node) {
2874 block = rb_entry(rb_node, struct tree_block, rb_node);
2875
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002876 node = build_backref_tree(rc, &block->key,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002877 block->level, block->bytenr);
2878 if (IS_ERR(node)) {
2879 err = PTR_ERR(node);
2880 goto out;
2881 }
2882
2883 ret = relocate_tree_block(trans, rc, node, &block->key,
2884 path);
2885 if (ret < 0) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002886 if (ret != -EAGAIN || rb_node == rb_first(blocks))
2887 err = ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002888 goto out;
2889 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002890 rb_node = rb_next(rb_node);
2891 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002892out:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04002893 err = finish_pending_nodes(trans, rc, path, err);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002894
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002895 btrfs_free_path(path);
Liu Boe1a12672013-03-04 16:25:38 +00002896out_path:
2897 free_block_list(blocks);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002898 return err;
2899}
2900
2901static noinline_for_stack
Yan, Zhengefa56462010-05-16 10:49:59 -04002902int prealloc_file_extent_cluster(struct inode *inode,
2903 struct file_extent_cluster *cluster)
2904{
2905 u64 alloc_hint = 0;
2906 u64 start;
2907 u64 end;
2908 u64 offset = BTRFS_I(inode)->index_cnt;
2909 u64 num_bytes;
2910 int nr = 0;
2911 int ret = 0;
2912
2913 BUG_ON(cluster->start != cluster->boundary[0]);
2914 mutex_lock(&inode->i_mutex);
2915
2916 ret = btrfs_check_data_free_space(inode, cluster->end +
2917 1 - cluster->start);
2918 if (ret)
2919 goto out;
2920
2921 while (nr < cluster->nr) {
2922 start = cluster->boundary[nr] - offset;
2923 if (nr + 1 < cluster->nr)
2924 end = cluster->boundary[nr + 1] - 1 - offset;
2925 else
2926 end = cluster->end - offset;
2927
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002928 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan, Zhengefa56462010-05-16 10:49:59 -04002929 num_bytes = end + 1 - start;
2930 ret = btrfs_prealloc_file_range(inode, 0, start,
2931 num_bytes, num_bytes,
2932 end + 1, &alloc_hint);
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002933 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan, Zhengefa56462010-05-16 10:49:59 -04002934 if (ret)
2935 break;
2936 nr++;
2937 }
2938 btrfs_free_reserved_data_space(inode, cluster->end +
2939 1 - cluster->start);
2940out:
2941 mutex_unlock(&inode->i_mutex);
2942 return ret;
2943}
2944
2945static noinline_for_stack
Yan, Zheng0257bb82009-09-24 09:17:31 -04002946int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2947 u64 block_start)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002948{
2949 struct btrfs_root *root = BTRFS_I(inode)->root;
2950 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2951 struct extent_map *em;
Yan, Zheng0257bb82009-09-24 09:17:31 -04002952 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002953
David Sterba172ddd62011-04-21 00:48:27 +02002954 em = alloc_extent_map();
Yan, Zheng0257bb82009-09-24 09:17:31 -04002955 if (!em)
2956 return -ENOMEM;
2957
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002958 em->start = start;
Yan, Zheng0257bb82009-09-24 09:17:31 -04002959 em->len = end + 1 - start;
2960 em->block_len = em->len;
2961 em->block_start = block_start;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002962 em->bdev = root->fs_info->fs_devices->latest_bdev;
2963 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2964
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002965 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002966 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04002967 write_lock(&em_tree->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002968 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04002969 write_unlock(&em_tree->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002970 if (ret != -EEXIST) {
2971 free_extent_map(em);
2972 break;
2973 }
2974 btrfs_drop_extent_cache(inode, start, end, 0);
2975 }
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002976 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
Yan, Zheng0257bb82009-09-24 09:17:31 -04002977 return ret;
2978}
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002979
Yan, Zheng0257bb82009-09-24 09:17:31 -04002980static int relocate_file_extent_cluster(struct inode *inode,
2981 struct file_extent_cluster *cluster)
2982{
2983 u64 page_start;
2984 u64 page_end;
2985 u64 offset = BTRFS_I(inode)->index_cnt;
2986 unsigned long index;
2987 unsigned long last_index;
Yan, Zheng0257bb82009-09-24 09:17:31 -04002988 struct page *page;
2989 struct file_ra_state *ra;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04002990 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
Yan, Zheng0257bb82009-09-24 09:17:31 -04002991 int nr = 0;
2992 int ret = 0;
2993
2994 if (!cluster->nr)
2995 return 0;
2996
2997 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2998 if (!ra)
2999 return -ENOMEM;
3000
Yan, Zhengefa56462010-05-16 10:49:59 -04003001 ret = prealloc_file_extent_cluster(inode, cluster);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003002 if (ret)
Yan, Zhengefa56462010-05-16 10:49:59 -04003003 goto out;
Yan, Zheng0257bb82009-09-24 09:17:31 -04003004
3005 file_ra_state_init(ra, inode->i_mapping);
3006
Yan, Zhengefa56462010-05-16 10:49:59 -04003007 ret = setup_extent_mapping(inode, cluster->start - offset,
3008 cluster->end - offset, cluster->start);
3009 if (ret)
3010 goto out;
3011
3012 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
3013 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
Yan, Zheng0257bb82009-09-24 09:17:31 -04003014 while (index <= last_index) {
Yan, Zhengefa56462010-05-16 10:49:59 -04003015 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
3016 if (ret)
3017 goto out;
3018
Yan, Zheng0257bb82009-09-24 09:17:31 -04003019 page = find_lock_page(inode->i_mapping, index);
3020 if (!page) {
3021 page_cache_sync_readahead(inode->i_mapping,
3022 ra, NULL, index,
3023 last_index + 1 - index);
Josef Bacika94733d2011-07-11 10:47:06 -04003024 page = find_or_create_page(inode->i_mapping, index,
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003025 mask);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003026 if (!page) {
Yan, Zhengefa56462010-05-16 10:49:59 -04003027 btrfs_delalloc_release_metadata(inode,
3028 PAGE_CACHE_SIZE);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003029 ret = -ENOMEM;
Yan, Zhengefa56462010-05-16 10:49:59 -04003030 goto out;
Yan, Zheng0257bb82009-09-24 09:17:31 -04003031 }
3032 }
3033
3034 if (PageReadahead(page)) {
3035 page_cache_async_readahead(inode->i_mapping,
3036 ra, NULL, page, index,
3037 last_index + 1 - index);
3038 }
3039
3040 if (!PageUptodate(page)) {
3041 btrfs_readpage(NULL, page);
3042 lock_page(page);
3043 if (!PageUptodate(page)) {
3044 unlock_page(page);
3045 page_cache_release(page);
Yan, Zhengefa56462010-05-16 10:49:59 -04003046 btrfs_delalloc_release_metadata(inode,
3047 PAGE_CACHE_SIZE);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003048 ret = -EIO;
Yan, Zhengefa56462010-05-16 10:49:59 -04003049 goto out;
Yan, Zheng0257bb82009-09-24 09:17:31 -04003050 }
3051 }
3052
Miao Xie4eee4fa2012-12-21 09:17:45 +00003053 page_start = page_offset(page);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003054 page_end = page_start + PAGE_CACHE_SIZE - 1;
3055
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003056 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003057
3058 set_page_extent_mapped(page);
3059
3060 if (nr < cluster->nr &&
3061 page_start + offset == cluster->boundary[nr]) {
3062 set_extent_bits(&BTRFS_I(inode)->io_tree,
3063 page_start, page_end,
3064 EXTENT_BOUNDARY, GFP_NOFS);
3065 nr++;
3066 }
Yan, Zheng0257bb82009-09-24 09:17:31 -04003067
Yan, Zhengefa56462010-05-16 10:49:59 -04003068 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003069 set_page_dirty(page);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003070
3071 unlock_extent(&BTRFS_I(inode)->io_tree,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003072 page_start, page_end);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003073 unlock_page(page);
3074 page_cache_release(page);
3075
3076 index++;
Yan, Zhengefa56462010-05-16 10:49:59 -04003077 balance_dirty_pages_ratelimited(inode->i_mapping);
3078 btrfs_throttle(BTRFS_I(inode)->root);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003079 }
3080 WARN_ON(nr != cluster->nr);
Yan, Zhengefa56462010-05-16 10:49:59 -04003081out:
Yan, Zheng0257bb82009-09-24 09:17:31 -04003082 kfree(ra);
3083 return ret;
3084}
3085
3086static noinline_for_stack
3087int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3088 struct file_extent_cluster *cluster)
3089{
3090 int ret;
3091
3092 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3093 ret = relocate_file_extent_cluster(inode, cluster);
3094 if (ret)
3095 return ret;
3096 cluster->nr = 0;
3097 }
3098
3099 if (!cluster->nr)
3100 cluster->start = extent_key->objectid;
3101 else
3102 BUG_ON(cluster->nr >= MAX_EXTENTS);
3103 cluster->end = extent_key->objectid + extent_key->offset - 1;
3104 cluster->boundary[cluster->nr] = extent_key->objectid;
3105 cluster->nr++;
3106
3107 if (cluster->nr >= MAX_EXTENTS) {
3108 ret = relocate_file_extent_cluster(inode, cluster);
3109 if (ret)
3110 return ret;
3111 cluster->nr = 0;
3112 }
3113 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003114}
3115
3116#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3117static int get_ref_objectid_v0(struct reloc_control *rc,
3118 struct btrfs_path *path,
3119 struct btrfs_key *extent_key,
3120 u64 *ref_objectid, int *path_change)
3121{
3122 struct btrfs_key key;
3123 struct extent_buffer *leaf;
3124 struct btrfs_extent_ref_v0 *ref0;
3125 int ret;
3126 int slot;
3127
3128 leaf = path->nodes[0];
3129 slot = path->slots[0];
3130 while (1) {
3131 if (slot >= btrfs_header_nritems(leaf)) {
3132 ret = btrfs_next_leaf(rc->extent_root, path);
3133 if (ret < 0)
3134 return ret;
3135 BUG_ON(ret > 0);
3136 leaf = path->nodes[0];
3137 slot = path->slots[0];
3138 if (path_change)
3139 *path_change = 1;
3140 }
3141 btrfs_item_key_to_cpu(leaf, &key, slot);
3142 if (key.objectid != extent_key->objectid)
3143 return -ENOENT;
3144
3145 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3146 slot++;
3147 continue;
3148 }
3149 ref0 = btrfs_item_ptr(leaf, slot,
3150 struct btrfs_extent_ref_v0);
3151 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3152 break;
3153 }
3154 return 0;
3155}
3156#endif
3157
3158/*
3159 * helper to add a tree block to the list.
3160 * the major work is getting the generation and level of the block
3161 */
3162static int add_tree_block(struct reloc_control *rc,
3163 struct btrfs_key *extent_key,
3164 struct btrfs_path *path,
3165 struct rb_root *blocks)
3166{
3167 struct extent_buffer *eb;
3168 struct btrfs_extent_item *ei;
3169 struct btrfs_tree_block_info *bi;
3170 struct tree_block *block;
3171 struct rb_node *rb_node;
3172 u32 item_size;
3173 int level = -1;
3174 int generation;
3175
3176 eb = path->nodes[0];
3177 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3178
3179 if (item_size >= sizeof(*ei) + sizeof(*bi)) {
3180 ei = btrfs_item_ptr(eb, path->slots[0],
3181 struct btrfs_extent_item);
3182 bi = (struct btrfs_tree_block_info *)(ei + 1);
3183 generation = btrfs_extent_generation(eb, ei);
3184 level = btrfs_tree_block_level(eb, bi);
3185 } else {
3186#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3187 u64 ref_owner;
3188 int ret;
3189
3190 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3191 ret = get_ref_objectid_v0(rc, path, extent_key,
3192 &ref_owner, NULL);
Andi Kleen411fc6b2010-10-29 15:14:31 -04003193 if (ret < 0)
3194 return ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003195 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3196 level = (int)ref_owner;
3197 /* FIXME: get real generation */
3198 generation = 0;
3199#else
3200 BUG();
3201#endif
3202 }
3203
David Sterbab3b4aa72011-04-21 01:20:15 +02003204 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003205
3206 BUG_ON(level == -1);
3207
3208 block = kmalloc(sizeof(*block), GFP_NOFS);
3209 if (!block)
3210 return -ENOMEM;
3211
3212 block->bytenr = extent_key->objectid;
3213 block->key.objectid = extent_key->offset;
3214 block->key.offset = generation;
3215 block->level = level;
3216 block->key_ready = 0;
3217
3218 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04003219 if (rb_node)
3220 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003221
3222 return 0;
3223}
3224
3225/*
3226 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3227 */
3228static int __add_tree_block(struct reloc_control *rc,
3229 u64 bytenr, u32 blocksize,
3230 struct rb_root *blocks)
3231{
3232 struct btrfs_path *path;
3233 struct btrfs_key key;
3234 int ret;
3235
3236 if (tree_block_processed(bytenr, blocksize, rc))
3237 return 0;
3238
3239 if (tree_search(blocks, bytenr))
3240 return 0;
3241
3242 path = btrfs_alloc_path();
3243 if (!path)
3244 return -ENOMEM;
3245
3246 key.objectid = bytenr;
3247 key.type = BTRFS_EXTENT_ITEM_KEY;
3248 key.offset = blocksize;
3249
3250 path->search_commit_root = 1;
3251 path->skip_locking = 1;
3252 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3253 if (ret < 0)
3254 goto out;
3255 BUG_ON(ret);
3256
3257 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3258 ret = add_tree_block(rc, &key, path, blocks);
3259out:
3260 btrfs_free_path(path);
3261 return ret;
3262}
3263
3264/*
3265 * helper to check if the block use full backrefs for pointers in it
3266 */
3267static int block_use_full_backref(struct reloc_control *rc,
3268 struct extent_buffer *eb)
3269{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003270 u64 flags;
3271 int ret;
3272
3273 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3274 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3275 return 1;
3276
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003277 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3278 eb->start, eb->len, NULL, &flags);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003279 BUG_ON(ret);
3280
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003281 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3282 ret = 1;
3283 else
3284 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003285 return ret;
3286}
3287
Josef Bacik0af3d002010-06-21 14:48:16 -04003288static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3289 struct inode *inode, u64 ino)
3290{
3291 struct btrfs_key key;
3292 struct btrfs_path *path;
3293 struct btrfs_root *root = fs_info->tree_root;
3294 struct btrfs_trans_handle *trans;
Josef Bacik0af3d002010-06-21 14:48:16 -04003295 int ret = 0;
3296
3297 if (inode)
3298 goto truncate;
3299
3300 key.objectid = ino;
3301 key.type = BTRFS_INODE_ITEM_KEY;
3302 key.offset = 0;
3303
3304 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
Tsutomu Itohf54fb852012-09-06 03:08:59 -06003305 if (IS_ERR(inode) || is_bad_inode(inode)) {
3306 if (!IS_ERR(inode))
Josef Bacik0af3d002010-06-21 14:48:16 -04003307 iput(inode);
3308 return -ENOENT;
3309 }
3310
3311truncate:
3312 path = btrfs_alloc_path();
3313 if (!path) {
3314 ret = -ENOMEM;
3315 goto out;
3316 }
3317
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003318 trans = btrfs_join_transaction(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04003319 if (IS_ERR(trans)) {
3320 btrfs_free_path(path);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00003321 ret = PTR_ERR(trans);
Josef Bacik0af3d002010-06-21 14:48:16 -04003322 goto out;
3323 }
3324
3325 ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
3326
3327 btrfs_free_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04003328 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003329 btrfs_btree_balance_dirty(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04003330out:
3331 iput(inode);
3332 return ret;
3333}
3334
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003335/*
3336 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3337 * this function scans fs tree to find blocks reference the data extent
3338 */
3339static int find_data_references(struct reloc_control *rc,
3340 struct btrfs_key *extent_key,
3341 struct extent_buffer *leaf,
3342 struct btrfs_extent_data_ref *ref,
3343 struct rb_root *blocks)
3344{
3345 struct btrfs_path *path;
3346 struct tree_block *block;
3347 struct btrfs_root *root;
3348 struct btrfs_file_extent_item *fi;
3349 struct rb_node *rb_node;
3350 struct btrfs_key key;
3351 u64 ref_root;
3352 u64 ref_objectid;
3353 u64 ref_offset;
3354 u32 ref_count;
3355 u32 nritems;
3356 int err = 0;
3357 int added = 0;
3358 int counted;
3359 int ret;
3360
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003361 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3362 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3363 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3364 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3365
Josef Bacik0af3d002010-06-21 14:48:16 -04003366 /*
3367 * This is an extent belonging to the free space cache, lets just delete
3368 * it and redo the search.
3369 */
3370 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3371 ret = delete_block_group_cache(rc->extent_root->fs_info,
3372 NULL, ref_objectid);
3373 if (ret != -ENOENT)
3374 return ret;
3375 ret = 0;
3376 }
3377
3378 path = btrfs_alloc_path();
3379 if (!path)
3380 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04003381 path->reada = 1;
Josef Bacik0af3d002010-06-21 14:48:16 -04003382
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003383 root = read_fs_root(rc->extent_root->fs_info, ref_root);
3384 if (IS_ERR(root)) {
3385 err = PTR_ERR(root);
3386 goto out;
3387 }
3388
3389 key.objectid = ref_objectid;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003390 key.type = BTRFS_EXTENT_DATA_KEY;
Yan, Zheng84850e82011-08-29 09:25:53 +08003391 if (ref_offset > ((u64)-1 << 32))
3392 key.offset = 0;
3393 else
3394 key.offset = ref_offset;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003395
3396 path->search_commit_root = 1;
3397 path->skip_locking = 1;
3398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3399 if (ret < 0) {
3400 err = ret;
3401 goto out;
3402 }
3403
3404 leaf = path->nodes[0];
3405 nritems = btrfs_header_nritems(leaf);
3406 /*
3407 * the references in tree blocks that use full backrefs
3408 * are not counted in
3409 */
3410 if (block_use_full_backref(rc, leaf))
3411 counted = 0;
3412 else
3413 counted = 1;
3414 rb_node = tree_search(blocks, leaf->start);
3415 if (rb_node) {
3416 if (counted)
3417 added = 1;
3418 else
3419 path->slots[0] = nritems;
3420 }
3421
3422 while (ref_count > 0) {
3423 while (path->slots[0] >= nritems) {
3424 ret = btrfs_next_leaf(root, path);
3425 if (ret < 0) {
3426 err = ret;
3427 goto out;
3428 }
3429 if (ret > 0) {
3430 WARN_ON(1);
3431 goto out;
3432 }
3433
3434 leaf = path->nodes[0];
3435 nritems = btrfs_header_nritems(leaf);
3436 added = 0;
3437
3438 if (block_use_full_backref(rc, leaf))
3439 counted = 0;
3440 else
3441 counted = 1;
3442 rb_node = tree_search(blocks, leaf->start);
3443 if (rb_node) {
3444 if (counted)
3445 added = 1;
3446 else
3447 path->slots[0] = nritems;
3448 }
3449 }
3450
3451 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3452 if (key.objectid != ref_objectid ||
3453 key.type != BTRFS_EXTENT_DATA_KEY) {
3454 WARN_ON(1);
3455 break;
3456 }
3457
3458 fi = btrfs_item_ptr(leaf, path->slots[0],
3459 struct btrfs_file_extent_item);
3460
3461 if (btrfs_file_extent_type(leaf, fi) ==
3462 BTRFS_FILE_EXTENT_INLINE)
3463 goto next;
3464
3465 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3466 extent_key->objectid)
3467 goto next;
3468
3469 key.offset -= btrfs_file_extent_offset(leaf, fi);
3470 if (key.offset != ref_offset)
3471 goto next;
3472
3473 if (counted)
3474 ref_count--;
3475 if (added)
3476 goto next;
3477
3478 if (!tree_block_processed(leaf->start, leaf->len, rc)) {
3479 block = kmalloc(sizeof(*block), GFP_NOFS);
3480 if (!block) {
3481 err = -ENOMEM;
3482 break;
3483 }
3484 block->bytenr = leaf->start;
3485 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3486 block->level = 0;
3487 block->key_ready = 1;
3488 rb_node = tree_insert(blocks, block->bytenr,
3489 &block->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -04003490 if (rb_node)
3491 backref_tree_panic(rb_node, -EEXIST,
3492 block->bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003493 }
3494 if (counted)
3495 added = 1;
3496 else
3497 path->slots[0] = nritems;
3498next:
3499 path->slots[0]++;
3500
3501 }
3502out:
3503 btrfs_free_path(path);
3504 return err;
3505}
3506
3507/*
Liu Bo2c016dc2012-12-26 15:32:17 +08003508 * helper to find all tree blocks that reference a given data extent
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003509 */
3510static noinline_for_stack
3511int add_data_references(struct reloc_control *rc,
3512 struct btrfs_key *extent_key,
3513 struct btrfs_path *path,
3514 struct rb_root *blocks)
3515{
3516 struct btrfs_key key;
3517 struct extent_buffer *eb;
3518 struct btrfs_extent_data_ref *dref;
3519 struct btrfs_extent_inline_ref *iref;
3520 unsigned long ptr;
3521 unsigned long end;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003522 u32 blocksize = btrfs_level_size(rc->extent_root, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003523 int ret;
3524 int err = 0;
3525
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003526 eb = path->nodes[0];
3527 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3528 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3529#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3530 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3531 ptr = end;
3532 else
3533#endif
3534 ptr += sizeof(struct btrfs_extent_item);
3535
3536 while (ptr < end) {
3537 iref = (struct btrfs_extent_inline_ref *)ptr;
3538 key.type = btrfs_extent_inline_ref_type(eb, iref);
3539 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3540 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3541 ret = __add_tree_block(rc, key.offset, blocksize,
3542 blocks);
3543 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3544 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3545 ret = find_data_references(rc, extent_key,
3546 eb, dref, blocks);
3547 } else {
3548 BUG();
3549 }
3550 ptr += btrfs_extent_inline_ref_size(key.type);
3551 }
3552 WARN_ON(ptr > end);
3553
3554 while (1) {
3555 cond_resched();
3556 eb = path->nodes[0];
3557 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3558 ret = btrfs_next_leaf(rc->extent_root, path);
3559 if (ret < 0) {
3560 err = ret;
3561 break;
3562 }
3563 if (ret > 0)
3564 break;
3565 eb = path->nodes[0];
3566 }
3567
3568 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3569 if (key.objectid != extent_key->objectid)
3570 break;
3571
3572#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3573 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3574 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3575#else
3576 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3577 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3578#endif
3579 ret = __add_tree_block(rc, key.offset, blocksize,
3580 blocks);
3581 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3582 dref = btrfs_item_ptr(eb, path->slots[0],
3583 struct btrfs_extent_data_ref);
3584 ret = find_data_references(rc, extent_key,
3585 eb, dref, blocks);
3586 } else {
3587 ret = 0;
3588 }
3589 if (ret) {
3590 err = ret;
3591 break;
3592 }
3593 path->slots[0]++;
3594 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003595 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003596 if (err)
3597 free_block_list(blocks);
3598 return err;
3599}
3600
3601/*
Liu Bo2c016dc2012-12-26 15:32:17 +08003602 * helper to find next unprocessed extent
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003603 */
3604static noinline_for_stack
3605int find_next_extent(struct btrfs_trans_handle *trans,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003606 struct reloc_control *rc, struct btrfs_path *path,
3607 struct btrfs_key *extent_key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003608{
3609 struct btrfs_key key;
3610 struct extent_buffer *leaf;
3611 u64 start, end, last;
3612 int ret;
3613
3614 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3615 while (1) {
3616 cond_resched();
3617 if (rc->search_start >= last) {
3618 ret = 1;
3619 break;
3620 }
3621
3622 key.objectid = rc->search_start;
3623 key.type = BTRFS_EXTENT_ITEM_KEY;
3624 key.offset = 0;
3625
3626 path->search_commit_root = 1;
3627 path->skip_locking = 1;
3628 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3629 0, 0);
3630 if (ret < 0)
3631 break;
3632next:
3633 leaf = path->nodes[0];
3634 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3635 ret = btrfs_next_leaf(rc->extent_root, path);
3636 if (ret != 0)
3637 break;
3638 leaf = path->nodes[0];
3639 }
3640
3641 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3642 if (key.objectid >= last) {
3643 ret = 1;
3644 break;
3645 }
3646
3647 if (key.type != BTRFS_EXTENT_ITEM_KEY ||
3648 key.objectid + key.offset <= rc->search_start) {
3649 path->slots[0]++;
3650 goto next;
3651 }
3652
3653 ret = find_first_extent_bit(&rc->processed_blocks,
3654 key.objectid, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -04003655 EXTENT_DIRTY, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003656
3657 if (ret == 0 && start <= key.objectid) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003658 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003659 rc->search_start = end + 1;
3660 } else {
3661 rc->search_start = key.objectid + key.offset;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003662 memcpy(extent_key, &key, sizeof(key));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003663 return 0;
3664 }
3665 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003666 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003667 return ret;
3668}
3669
3670static void set_reloc_control(struct reloc_control *rc)
3671{
3672 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
Chris Mason75857172011-06-13 20:00:16 -04003673
3674 mutex_lock(&fs_info->reloc_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003675 fs_info->reloc_ctl = rc;
Chris Mason75857172011-06-13 20:00:16 -04003676 mutex_unlock(&fs_info->reloc_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003677}
3678
3679static void unset_reloc_control(struct reloc_control *rc)
3680{
3681 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
Chris Mason75857172011-06-13 20:00:16 -04003682
3683 mutex_lock(&fs_info->reloc_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003684 fs_info->reloc_ctl = NULL;
Chris Mason75857172011-06-13 20:00:16 -04003685 mutex_unlock(&fs_info->reloc_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003686}
3687
3688static int check_extent_flags(u64 flags)
3689{
3690 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3691 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3692 return 1;
3693 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3694 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3695 return 1;
3696 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3697 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3698 return 1;
3699 return 0;
3700}
3701
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003702static noinline_for_stack
3703int prepare_to_relocate(struct reloc_control *rc)
3704{
3705 struct btrfs_trans_handle *trans;
3706 int ret;
3707
Miao Xie66d8f3d2012-09-06 04:02:28 -06003708 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root,
3709 BTRFS_BLOCK_RSV_TEMP);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003710 if (!rc->block_rsv)
3711 return -ENOMEM;
3712
3713 /*
3714 * reserve some space for creating reloc trees.
3715 * btrfs_init_reloc_root will use them when there
3716 * is no reservation in transaction handle.
3717 */
Josef Bacik4a92b1b2011-08-30 12:34:28 -04003718 ret = btrfs_block_rsv_add(rc->extent_root, rc->block_rsv,
Miao Xie08e007d2012-10-16 11:33:38 +00003719 rc->extent_root->nodesize * 256,
3720 BTRFS_RESERVE_FLUSH_ALL);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003721 if (ret)
3722 return ret;
3723
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003724 memset(&rc->cluster, 0, sizeof(rc->cluster));
3725 rc->search_start = rc->block_group->key.objectid;
3726 rc->extents_found = 0;
3727 rc->nodes_relocated = 0;
3728 rc->merging_rsv_size = 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003729
3730 rc->create_reloc_tree = 1;
3731 set_reloc_control(rc);
3732
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003733 trans = btrfs_join_transaction(rc->extent_root);
Liu Bo28818942013-03-04 16:25:39 +00003734 if (IS_ERR(trans)) {
3735 unset_reloc_control(rc);
3736 /*
3737 * extent tree is not a ref_cow tree and has no reloc_root to
3738 * cleanup. And callers are responsible to free the above
3739 * block rsv.
3740 */
3741 return PTR_ERR(trans);
3742 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003743 btrfs_commit_transaction(trans, rc->extent_root);
3744 return 0;
3745}
Yan, Zheng76dda932009-09-21 16:00:26 -04003746
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003747static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3748{
3749 struct rb_root blocks = RB_ROOT;
3750 struct btrfs_key key;
3751 struct btrfs_trans_handle *trans = NULL;
3752 struct btrfs_path *path;
3753 struct btrfs_extent_item *ei;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003754 u64 flags;
3755 u32 item_size;
3756 int ret;
3757 int err = 0;
Chris Masonc87f08c2011-02-16 13:57:04 -05003758 int progress = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003759
3760 path = btrfs_alloc_path();
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003761 if (!path)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003762 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04003763 path->reada = 1;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003764
3765 ret = prepare_to_relocate(rc);
3766 if (ret) {
3767 err = ret;
3768 goto out_free;
Jiri Slaby2423fdf2010-01-06 16:57:22 +00003769 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003770
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003771 while (1) {
Chris Masonc87f08c2011-02-16 13:57:04 -05003772 progress++;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003773 trans = btrfs_start_transaction(rc->extent_root, 0);
Liu Bo0f788c52013-03-04 16:25:40 +00003774 if (IS_ERR(trans)) {
3775 err = PTR_ERR(trans);
3776 trans = NULL;
3777 break;
3778 }
Chris Masonc87f08c2011-02-16 13:57:04 -05003779restart:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003780 if (update_backref_cache(trans, &rc->backref_cache)) {
3781 btrfs_end_transaction(trans, rc->extent_root);
3782 continue;
3783 }
3784
3785 ret = find_next_extent(trans, rc, path, &key);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003786 if (ret < 0)
3787 err = ret;
3788 if (ret != 0)
3789 break;
3790
3791 rc->extents_found++;
3792
3793 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3794 struct btrfs_extent_item);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003795 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003796 if (item_size >= sizeof(*ei)) {
3797 flags = btrfs_extent_flags(path->nodes[0], ei);
3798 ret = check_extent_flags(flags);
3799 BUG_ON(ret);
3800
3801 } else {
3802#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3803 u64 ref_owner;
3804 int path_change = 0;
3805
3806 BUG_ON(item_size !=
3807 sizeof(struct btrfs_extent_item_v0));
3808 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3809 &path_change);
3810 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3811 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3812 else
3813 flags = BTRFS_EXTENT_FLAG_DATA;
3814
3815 if (path_change) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003816 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003817
3818 path->search_commit_root = 1;
3819 path->skip_locking = 1;
3820 ret = btrfs_search_slot(NULL, rc->extent_root,
3821 &key, path, 0, 0);
3822 if (ret < 0) {
3823 err = ret;
3824 break;
3825 }
3826 BUG_ON(ret > 0);
3827 }
3828#else
3829 BUG();
3830#endif
3831 }
3832
3833 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3834 ret = add_tree_block(rc, &key, path, &blocks);
3835 } else if (rc->stage == UPDATE_DATA_PTRS &&
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003836 (flags & BTRFS_EXTENT_FLAG_DATA)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003837 ret = add_data_references(rc, &key, path, &blocks);
3838 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +02003839 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003840 ret = 0;
3841 }
3842 if (ret < 0) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003843 err = ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003844 break;
3845 }
3846
3847 if (!RB_EMPTY_ROOT(&blocks)) {
3848 ret = relocate_tree_blocks(trans, rc, &blocks);
3849 if (ret < 0) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003850 if (ret != -EAGAIN) {
3851 err = ret;
3852 break;
3853 }
3854 rc->extents_found--;
3855 rc->search_start = key.objectid;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003856 }
3857 }
3858
Josef Bacik36ba0222011-10-18 12:15:48 -04003859 ret = btrfs_block_rsv_check(rc->extent_root, rc->block_rsv, 5);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003860 if (ret < 0) {
Daniel J Blueman7654b722012-04-27 12:41:46 -04003861 if (ret != -ENOSPC) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003862 err = ret;
3863 WARN_ON(1);
3864 break;
3865 }
3866 rc->commit_transaction = 1;
3867 }
3868
3869 if (rc->commit_transaction) {
3870 rc->commit_transaction = 0;
3871 ret = btrfs_commit_transaction(trans, rc->extent_root);
3872 BUG_ON(ret);
3873 } else {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003874 btrfs_end_transaction_throttle(trans, rc->extent_root);
Liu Bob53d3f52012-11-14 14:34:34 +00003875 btrfs_btree_balance_dirty(rc->extent_root);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003876 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003877 trans = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003878
3879 if (rc->stage == MOVE_DATA_EXTENTS &&
3880 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3881 rc->found_file_extent = 1;
Yan, Zheng0257bb82009-09-24 09:17:31 -04003882 ret = relocate_data_extent(rc->data_inode,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003883 &key, &rc->cluster);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003884 if (ret < 0) {
3885 err = ret;
3886 break;
3887 }
3888 }
3889 }
Chris Masonc87f08c2011-02-16 13:57:04 -05003890 if (trans && progress && err == -ENOSPC) {
3891 ret = btrfs_force_chunk_alloc(trans, rc->extent_root,
3892 rc->block_group->flags);
3893 if (ret == 0) {
3894 err = 0;
3895 progress = 0;
3896 goto restart;
3897 }
3898 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003899
David Sterbab3b4aa72011-04-21 01:20:15 +02003900 btrfs_release_path(path);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003901 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3902 GFP_NOFS);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003903
3904 if (trans) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003905 btrfs_end_transaction_throttle(trans, rc->extent_root);
Liu Bob53d3f52012-11-14 14:34:34 +00003906 btrfs_btree_balance_dirty(rc->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003907 }
3908
Yan, Zheng0257bb82009-09-24 09:17:31 -04003909 if (!err) {
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003910 ret = relocate_file_extent_cluster(rc->data_inode,
3911 &rc->cluster);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003912 if (ret < 0)
3913 err = ret;
3914 }
3915
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003916 rc->create_reloc_tree = 0;
3917 set_reloc_control(rc);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003918
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003919 backref_cache_cleanup(&rc->backref_cache);
3920 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003921
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003922 err = prepare_to_merge(rc, err);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003923
3924 merge_reloc_roots(rc);
3925
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003926 rc->merge_reloc_tree = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003927 unset_reloc_control(rc);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003928 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003929
3930 /* get rid of pinned extents */
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003931 trans = btrfs_join_transaction(rc->extent_root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00003932 if (IS_ERR(trans))
3933 err = PTR_ERR(trans);
3934 else
3935 btrfs_commit_transaction(trans, rc->extent_root);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003936out_free:
3937 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
3938 btrfs_free_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003939 return err;
3940}
3941
3942static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
Yan, Zheng0257bb82009-09-24 09:17:31 -04003943 struct btrfs_root *root, u64 objectid)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003944{
3945 struct btrfs_path *path;
3946 struct btrfs_inode_item *item;
3947 struct extent_buffer *leaf;
3948 int ret;
3949
3950 path = btrfs_alloc_path();
3951 if (!path)
3952 return -ENOMEM;
3953
3954 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3955 if (ret)
3956 goto out;
3957
3958 leaf = path->nodes[0];
3959 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3960 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3961 btrfs_set_inode_generation(leaf, item, 1);
Yan, Zheng0257bb82009-09-24 09:17:31 -04003962 btrfs_set_inode_size(leaf, item, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003963 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003964 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3965 BTRFS_INODE_PREALLOC);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003966 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02003967 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003968out:
3969 btrfs_free_path(path);
3970 return ret;
3971}
3972
3973/*
3974 * helper to create inode for data relocation.
3975 * the inode is in data relocation tree and its link count is 0
3976 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003977static noinline_for_stack
3978struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3979 struct btrfs_block_group_cache *group)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003980{
3981 struct inode *inode = NULL;
3982 struct btrfs_trans_handle *trans;
3983 struct btrfs_root *root;
3984 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003985 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
3986 int err = 0;
3987
3988 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3989 if (IS_ERR(root))
3990 return ERR_CAST(root);
3991
Yan, Zhenga22285a2010-05-16 10:48:46 -04003992 trans = btrfs_start_transaction(root, 6);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04003993 if (IS_ERR(trans))
3994 return ERR_CAST(trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003995
Li Zefan581bb052011-04-20 10:06:11 +08003996 err = btrfs_find_free_objectid(root, &objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003997 if (err)
3998 goto out;
3999
Yan, Zheng0257bb82009-09-24 09:17:31 -04004000 err = __insert_orphan_inode(trans, root, objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004001 BUG_ON(err);
4002
4003 key.objectid = objectid;
4004 key.type = BTRFS_INODE_ITEM_KEY;
4005 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00004006 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004007 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4008 BTRFS_I(inode)->index_cnt = group->key.objectid;
4009
4010 err = btrfs_orphan_add(trans, inode);
4011out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004012 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004013 btrfs_btree_balance_dirty(root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004014 if (err) {
4015 if (inode)
4016 iput(inode);
4017 inode = ERR_PTR(err);
4018 }
4019 return inode;
4020}
4021
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004022static struct reloc_control *alloc_reloc_control(void)
4023{
4024 struct reloc_control *rc;
4025
4026 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4027 if (!rc)
4028 return NULL;
4029
4030 INIT_LIST_HEAD(&rc->reloc_roots);
4031 backref_cache_init(&rc->backref_cache);
4032 mapping_tree_init(&rc->reloc_root_tree);
David Sterbaf993c882011-04-20 23:35:57 +02004033 extent_io_tree_init(&rc->processed_blocks, NULL);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004034 return rc;
4035}
4036
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004037/*
4038 * function to relocate all extents in a block group.
4039 */
4040int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
4041{
4042 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4043 struct reloc_control *rc;
Josef Bacik0af3d002010-06-21 14:48:16 -04004044 struct inode *inode;
4045 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004046 int ret;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004047 int rw = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004048 int err = 0;
4049
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004050 rc = alloc_reloc_control();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004051 if (!rc)
4052 return -ENOMEM;
4053
Yan, Zhengf0486c62010-05-16 10:46:25 -04004054 rc->extent_root = extent_root;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004055
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004056 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4057 BUG_ON(!rc->block_group);
4058
Yan, Zhengf0486c62010-05-16 10:46:25 -04004059 if (!rc->block_group->ro) {
4060 ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
4061 if (ret) {
4062 err = ret;
4063 goto out;
4064 }
4065 rw = 1;
4066 }
4067
Josef Bacik0af3d002010-06-21 14:48:16 -04004068 path = btrfs_alloc_path();
4069 if (!path) {
4070 err = -ENOMEM;
4071 goto out;
4072 }
4073
4074 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4075 path);
4076 btrfs_free_path(path);
4077
4078 if (!IS_ERR(inode))
4079 ret = delete_block_group_cache(fs_info, inode, 0);
4080 else
4081 ret = PTR_ERR(inode);
4082
4083 if (ret && ret != -ENOENT) {
4084 err = ret;
4085 goto out;
4086 }
4087
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004088 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4089 if (IS_ERR(rc->data_inode)) {
4090 err = PTR_ERR(rc->data_inode);
4091 rc->data_inode = NULL;
4092 goto out;
4093 }
4094
4095 printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
4096 (unsigned long long)rc->block_group->key.objectid,
4097 (unsigned long long)rc->block_group->flags);
4098
Miao Xie8ccf6f192012-10-25 09:28:04 +00004099 ret = btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
4100 if (ret < 0) {
4101 err = ret;
4102 goto out;
4103 }
Liu Bo6bbe3a92012-09-14 02:58:07 -06004104 btrfs_wait_ordered_extents(fs_info->tree_root, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004105
4106 while (1) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004107 mutex_lock(&fs_info->cleaner_mutex);
4108
4109 btrfs_clean_old_snapshots(fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004110 ret = relocate_block_group(rc);
Yan, Zheng76dda932009-09-21 16:00:26 -04004111
4112 mutex_unlock(&fs_info->cleaner_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004113 if (ret < 0) {
4114 err = ret;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004115 goto out;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004116 }
4117
4118 if (rc->extents_found == 0)
4119 break;
4120
4121 printk(KERN_INFO "btrfs: found %llu extents\n",
4122 (unsigned long long)rc->extents_found);
4123
4124 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4125 btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
4126 invalidate_mapping_pages(rc->data_inode->i_mapping,
4127 0, -1);
4128 rc->stage = UPDATE_DATA_PTRS;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004129 }
4130 }
4131
Yan, Zheng0257bb82009-09-24 09:17:31 -04004132 filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
4133 rc->block_group->key.objectid,
4134 rc->block_group->key.objectid +
4135 rc->block_group->key.offset - 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004136
4137 WARN_ON(rc->block_group->pinned > 0);
4138 WARN_ON(rc->block_group->reserved > 0);
4139 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4140out:
Yan, Zhengf0486c62010-05-16 10:46:25 -04004141 if (err && rw)
4142 btrfs_set_block_group_rw(extent_root, rc->block_group);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004143 iput(rc->data_inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004144 btrfs_put_block_group(rc->block_group);
4145 kfree(rc);
4146 return err;
4147}
4148
Yan, Zheng76dda932009-09-21 16:00:26 -04004149static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4150{
4151 struct btrfs_trans_handle *trans;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004152 int ret, err;
Yan, Zheng76dda932009-09-21 16:00:26 -04004153
Yan, Zhenga22285a2010-05-16 10:48:46 -04004154 trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004155 if (IS_ERR(trans))
4156 return PTR_ERR(trans);
Yan, Zheng76dda932009-09-21 16:00:26 -04004157
4158 memset(&root->root_item.drop_progress, 0,
4159 sizeof(root->root_item.drop_progress));
4160 root->root_item.drop_level = 0;
4161 btrfs_set_root_refs(&root->root_item, 0);
4162 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4163 &root->root_key, &root->root_item);
Yan, Zheng76dda932009-09-21 16:00:26 -04004164
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004165 err = btrfs_end_transaction(trans, root->fs_info->tree_root);
4166 if (err)
4167 return err;
4168 return ret;
Yan, Zheng76dda932009-09-21 16:00:26 -04004169}
4170
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004171/*
4172 * recover relocation interrupted by system crash.
4173 *
4174 * this function resumes merging reloc trees with corresponding fs trees.
4175 * this is important for keeping the sharing of tree blocks
4176 */
4177int btrfs_recover_relocation(struct btrfs_root *root)
4178{
4179 LIST_HEAD(reloc_roots);
4180 struct btrfs_key key;
4181 struct btrfs_root *fs_root;
4182 struct btrfs_root *reloc_root;
4183 struct btrfs_path *path;
4184 struct extent_buffer *leaf;
4185 struct reloc_control *rc = NULL;
4186 struct btrfs_trans_handle *trans;
4187 int ret;
4188 int err = 0;
4189
4190 path = btrfs_alloc_path();
4191 if (!path)
4192 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04004193 path->reada = -1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004194
4195 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4196 key.type = BTRFS_ROOT_ITEM_KEY;
4197 key.offset = (u64)-1;
4198
4199 while (1) {
4200 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
4201 path, 0, 0);
4202 if (ret < 0) {
4203 err = ret;
4204 goto out;
4205 }
4206 if (ret > 0) {
4207 if (path->slots[0] == 0)
4208 break;
4209 path->slots[0]--;
4210 }
4211 leaf = path->nodes[0];
4212 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004213 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004214
4215 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4216 key.type != BTRFS_ROOT_ITEM_KEY)
4217 break;
4218
4219 reloc_root = btrfs_read_fs_root_no_radix(root, &key);
4220 if (IS_ERR(reloc_root)) {
4221 err = PTR_ERR(reloc_root);
4222 goto out;
4223 }
4224
4225 list_add(&reloc_root->root_list, &reloc_roots);
4226
4227 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4228 fs_root = read_fs_root(root->fs_info,
4229 reloc_root->root_key.offset);
4230 if (IS_ERR(fs_root)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004231 ret = PTR_ERR(fs_root);
4232 if (ret != -ENOENT) {
4233 err = ret;
4234 goto out;
4235 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004236 ret = mark_garbage_root(reloc_root);
4237 if (ret < 0) {
4238 err = ret;
4239 goto out;
4240 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004241 }
4242 }
4243
4244 if (key.offset == 0)
4245 break;
4246
4247 key.offset--;
4248 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004249 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004250
4251 if (list_empty(&reloc_roots))
4252 goto out;
4253
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004254 rc = alloc_reloc_control();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004255 if (!rc) {
4256 err = -ENOMEM;
4257 goto out;
4258 }
4259
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004260 rc->extent_root = root->fs_info->extent_root;
4261
4262 set_reloc_control(rc);
4263
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004264 trans = btrfs_join_transaction(rc->extent_root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004265 if (IS_ERR(trans)) {
4266 unset_reloc_control(rc);
4267 err = PTR_ERR(trans);
4268 goto out_free;
4269 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004270
4271 rc->merge_reloc_tree = 1;
4272
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004273 while (!list_empty(&reloc_roots)) {
4274 reloc_root = list_entry(reloc_roots.next,
4275 struct btrfs_root, root_list);
4276 list_del(&reloc_root->root_list);
4277
4278 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4279 list_add_tail(&reloc_root->root_list,
4280 &rc->reloc_roots);
4281 continue;
4282 }
4283
4284 fs_root = read_fs_root(root->fs_info,
4285 reloc_root->root_key.offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004286 if (IS_ERR(fs_root)) {
4287 err = PTR_ERR(fs_root);
4288 goto out_free;
4289 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004290
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04004291 err = __add_reloc_root(reloc_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004292 BUG_ON(err < 0); /* -ENOMEM or logic error */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004293 fs_root->reloc_root = reloc_root;
4294 }
4295
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004296 err = btrfs_commit_transaction(trans, rc->extent_root);
4297 if (err)
4298 goto out_free;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004299
4300 merge_reloc_roots(rc);
4301
4302 unset_reloc_control(rc);
4303
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004304 trans = btrfs_join_transaction(rc->extent_root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004305 if (IS_ERR(trans))
4306 err = PTR_ERR(trans);
4307 else
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004308 err = btrfs_commit_transaction(trans, rc->extent_root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004309out_free:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004310 kfree(rc);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004311out:
Liu Boaca1bba2013-03-04 16:25:37 +00004312 if (!list_empty(&reloc_roots))
4313 free_reloc_roots(&reloc_roots);
4314
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004315 btrfs_free_path(path);
4316
4317 if (err == 0) {
4318 /* cleanup orphan inode in data relocation tree */
4319 fs_root = read_fs_root(root->fs_info,
4320 BTRFS_DATA_RELOC_TREE_OBJECTID);
4321 if (IS_ERR(fs_root))
4322 err = PTR_ERR(fs_root);
Miao Xied7ce5842010-02-02 08:46:44 +00004323 else
Josef Bacik66b4ffd2011-01-31 16:22:42 -05004324 err = btrfs_orphan_cleanup(fs_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004325 }
4326 return err;
4327}
4328
4329/*
4330 * helper to add ordered checksum for data relocation.
4331 *
4332 * cloning checksum properly handles the nodatasum extents.
4333 * it also saves CPU time to re-calculate the checksum.
4334 */
4335int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4336{
4337 struct btrfs_ordered_sum *sums;
4338 struct btrfs_sector_sum *sector_sum;
4339 struct btrfs_ordered_extent *ordered;
4340 struct btrfs_root *root = BTRFS_I(inode)->root;
4341 size_t offset;
4342 int ret;
4343 u64 disk_bytenr;
4344 LIST_HEAD(list);
4345
4346 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4347 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4348
4349 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4350 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01004351 disk_bytenr + len - 1, &list, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004352 if (ret)
4353 goto out;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004354
4355 while (!list_empty(&list)) {
4356 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4357 list_del_init(&sums->list);
4358
4359 sector_sum = sums->sums;
4360 sums->bytenr = ordered->start;
4361
4362 offset = 0;
4363 while (offset < sums->len) {
4364 sector_sum->bytenr += ordered->start - disk_bytenr;
4365 sector_sum++;
4366 offset += root->sectorsize;
4367 }
4368
4369 btrfs_add_ordered_sum(inode, ordered, sums);
4370 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004371out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004372 btrfs_put_ordered_extent(ordered);
Andi Kleen411fc6b2010-10-29 15:14:31 -04004373 return ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004374}
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004375
4376void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4377 struct btrfs_root *root, struct extent_buffer *buf,
4378 struct extent_buffer *cow)
4379{
4380 struct reloc_control *rc;
4381 struct backref_node *node;
4382 int first_cow = 0;
4383 int level;
4384 int ret;
4385
4386 rc = root->fs_info->reloc_ctl;
4387 if (!rc)
4388 return;
4389
4390 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4391 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4392
4393 level = btrfs_header_level(buf);
4394 if (btrfs_header_generation(buf) <=
4395 btrfs_root_last_snapshot(&root->root_item))
4396 first_cow = 1;
4397
4398 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4399 rc->create_reloc_tree) {
4400 WARN_ON(!first_cow && level == 0);
4401
4402 node = rc->backref_cache.path[level];
4403 BUG_ON(node->bytenr != buf->start &&
4404 node->new_bytenr != buf->start);
4405
4406 drop_node_buffer(node);
4407 extent_buffer_get(cow);
4408 node->eb = cow;
4409 node->new_bytenr = cow->start;
4410
4411 if (!node->pending) {
4412 list_move_tail(&node->list,
4413 &rc->backref_cache.pending[level]);
4414 node->pending = 1;
4415 }
4416
4417 if (first_cow)
4418 __mark_block_processed(rc, node);
4419
4420 if (first_cow && level > 0)
4421 rc->nodes_relocated += buf->len;
4422 }
4423
4424 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
4425 ret = replace_file_extents(trans, rc, root, cow);
4426 BUG_ON(ret);
4427 }
4428}
4429
4430/*
4431 * called before creating snapshot. it calculates metadata reservation
4432 * requried for relocating tree blocks in the snapshot
4433 */
4434void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
4435 struct btrfs_pending_snapshot *pending,
4436 u64 *bytes_to_reserve)
4437{
4438 struct btrfs_root *root;
4439 struct reloc_control *rc;
4440
4441 root = pending->root;
4442 if (!root->reloc_root)
4443 return;
4444
4445 rc = root->fs_info->reloc_ctl;
4446 if (!rc->merge_reloc_tree)
4447 return;
4448
4449 root = root->reloc_root;
4450 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4451 /*
4452 * relocation is in the stage of merging trees. the space
4453 * used by merging a reloc tree is twice the size of
4454 * relocated tree nodes in the worst case. half for cowing
4455 * the reloc tree, half for cowing the fs tree. the space
4456 * used by cowing the reloc tree will be freed after the
4457 * tree is dropped. if we create snapshot, cowing the fs
4458 * tree may use more space than it frees. so we need
4459 * reserve extra space.
4460 */
4461 *bytes_to_reserve += rc->nodes_relocated;
4462}
4463
4464/*
4465 * called after snapshot is created. migrate block reservation
4466 * and create reloc root for the newly created snapshot
4467 */
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004468int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004469 struct btrfs_pending_snapshot *pending)
4470{
4471 struct btrfs_root *root = pending->root;
4472 struct btrfs_root *reloc_root;
4473 struct btrfs_root *new_root;
4474 struct reloc_control *rc;
4475 int ret;
4476
4477 if (!root->reloc_root)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004478 return 0;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004479
4480 rc = root->fs_info->reloc_ctl;
4481 rc->merging_rsv_size += rc->nodes_relocated;
4482
4483 if (rc->merge_reloc_tree) {
4484 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4485 rc->block_rsv,
4486 rc->nodes_relocated);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004487 if (ret)
4488 return ret;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004489 }
4490
4491 new_root = pending->snap;
4492 reloc_root = create_reloc_root(trans, root->reloc_root,
4493 new_root->root_key.objectid);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004494 if (IS_ERR(reloc_root))
4495 return PTR_ERR(reloc_root);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004496
Jeff Mahoneyffd7b332011-10-03 23:23:15 -04004497 ret = __add_reloc_root(reloc_root);
4498 BUG_ON(ret < 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004499 new_root->reloc_root = reloc_root;
4500
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004501 if (rc->create_reloc_tree)
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004502 ret = clone_backref_node(trans, rc, root, reloc_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01004503 return ret;
Yan, Zheng3fd0a552010-05-16 10:49:59 -04004504}