]>
Commit | Line | Data |
---|---|---|
5d4f98a2 YZ |
1 | /* |
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> | |
5a0e3ad6 | 24 | #include <linux/slab.h> |
5d4f98a2 YZ |
25 | #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" | |
0af3d00b | 32 | #include "free-space-cache.h" |
581bb050 | 33 | #include "inode-map.h" |
62b99540 | 34 | #include "qgroup.h" |
5d4f98a2 YZ |
35 | |
36 | /* | |
37 | * backref_node, mapping_node and tree_block start with this | |
38 | */ | |
39 | struct tree_entry { | |
40 | struct rb_node rb_node; | |
41 | u64 bytenr; | |
42 | }; | |
43 | ||
44 | /* | |
45 | * present a tree block in the backref cache | |
46 | */ | |
47 | struct backref_node { | |
48 | struct rb_node rb_node; | |
49 | u64 bytenr; | |
3fd0a558 YZ |
50 | |
51 | u64 new_bytenr; | |
52 | /* objectid of tree block owner, can be not uptodate */ | |
5d4f98a2 | 53 | u64 owner; |
3fd0a558 YZ |
54 | /* link to pending, changed or detached list */ |
55 | struct list_head list; | |
5d4f98a2 YZ |
56 | /* list of upper level blocks reference this block */ |
57 | struct list_head upper; | |
58 | /* list of child blocks in the cache */ | |
59 | struct list_head lower; | |
60 | /* NULL if this node is not tree root */ | |
61 | struct btrfs_root *root; | |
62 | /* extent buffer got by COW the block */ | |
63 | struct extent_buffer *eb; | |
64 | /* level of tree block */ | |
65 | unsigned int level:8; | |
3fd0a558 YZ |
66 | /* is the block in non-reference counted tree */ |
67 | unsigned int cowonly:1; | |
68 | /* 1 if no child node in the cache */ | |
5d4f98a2 YZ |
69 | unsigned int lowest:1; |
70 | /* is the extent buffer locked */ | |
71 | unsigned int locked:1; | |
72 | /* has the block been processed */ | |
73 | unsigned int processed:1; | |
74 | /* have backrefs of this block been checked */ | |
75 | unsigned int checked:1; | |
3fd0a558 YZ |
76 | /* |
77 | * 1 if corresponding block has been cowed but some upper | |
78 | * level block pointers may not point to the new location | |
79 | */ | |
80 | unsigned int pending:1; | |
81 | /* | |
82 | * 1 if the backref node isn't connected to any other | |
83 | * backref node. | |
84 | */ | |
85 | unsigned int detached:1; | |
5d4f98a2 YZ |
86 | }; |
87 | ||
88 | /* | |
89 | * present a block pointer in the backref cache | |
90 | */ | |
91 | struct backref_edge { | |
92 | struct list_head list[2]; | |
93 | struct backref_node *node[2]; | |
5d4f98a2 YZ |
94 | }; |
95 | ||
96 | #define LOWER 0 | |
97 | #define UPPER 1 | |
0647bf56 | 98 | #define RELOCATION_RESERVED_NODES 256 |
5d4f98a2 YZ |
99 | |
100 | struct backref_cache { | |
101 | /* red black tree of all backref nodes in the cache */ | |
102 | struct rb_root rb_root; | |
3fd0a558 YZ |
103 | /* for passing backref nodes to btrfs_reloc_cow_block */ |
104 | struct backref_node *path[BTRFS_MAX_LEVEL]; | |
105 | /* | |
106 | * list of blocks that have been cowed but some block | |
107 | * pointers in upper level blocks may not reflect the | |
108 | * new location | |
109 | */ | |
5d4f98a2 | 110 | struct list_head pending[BTRFS_MAX_LEVEL]; |
3fd0a558 YZ |
111 | /* list of backref nodes with no child node */ |
112 | struct list_head leaves; | |
113 | /* list of blocks that have been cowed in current transaction */ | |
114 | struct list_head changed; | |
115 | /* list of detached backref node. */ | |
116 | struct list_head detached; | |
117 | ||
118 | u64 last_trans; | |
119 | ||
120 | int nr_nodes; | |
121 | int nr_edges; | |
5d4f98a2 YZ |
122 | }; |
123 | ||
124 | /* | |
125 | * map address of tree root to tree | |
126 | */ | |
127 | struct mapping_node { | |
128 | struct rb_node rb_node; | |
129 | u64 bytenr; | |
130 | void *data; | |
131 | }; | |
132 | ||
133 | struct mapping_tree { | |
134 | struct rb_root rb_root; | |
135 | spinlock_t lock; | |
136 | }; | |
137 | ||
138 | /* | |
139 | * present a tree block to process | |
140 | */ | |
141 | struct tree_block { | |
142 | struct rb_node rb_node; | |
143 | u64 bytenr; | |
144 | struct btrfs_key key; | |
145 | unsigned int level:8; | |
146 | unsigned int key_ready:1; | |
147 | }; | |
148 | ||
0257bb82 YZ |
149 | #define MAX_EXTENTS 128 |
150 | ||
151 | struct file_extent_cluster { | |
152 | u64 start; | |
153 | u64 end; | |
154 | u64 boundary[MAX_EXTENTS]; | |
155 | unsigned int nr; | |
156 | }; | |
157 | ||
5d4f98a2 YZ |
158 | struct reloc_control { |
159 | /* block group to relocate */ | |
160 | struct btrfs_block_group_cache *block_group; | |
161 | /* extent tree */ | |
162 | struct btrfs_root *extent_root; | |
163 | /* inode for moving data */ | |
164 | struct inode *data_inode; | |
3fd0a558 YZ |
165 | |
166 | struct btrfs_block_rsv *block_rsv; | |
167 | ||
168 | struct backref_cache backref_cache; | |
169 | ||
170 | struct file_extent_cluster cluster; | |
5d4f98a2 YZ |
171 | /* tree blocks have been processed */ |
172 | struct extent_io_tree processed_blocks; | |
173 | /* map start of tree root to corresponding reloc tree */ | |
174 | struct mapping_tree reloc_root_tree; | |
175 | /* list of reloc trees */ | |
176 | struct list_head reloc_roots; | |
3fd0a558 YZ |
177 | /* size of metadata reservation for merging reloc trees */ |
178 | u64 merging_rsv_size; | |
179 | /* size of relocated tree nodes */ | |
180 | u64 nodes_relocated; | |
0647bf56 WS |
181 | /* reserved size for block group relocation*/ |
182 | u64 reserved_bytes; | |
3fd0a558 | 183 | |
5d4f98a2 YZ |
184 | u64 search_start; |
185 | u64 extents_found; | |
3fd0a558 | 186 | |
3fd0a558 YZ |
187 | unsigned int stage:8; |
188 | unsigned int create_reloc_tree:1; | |
189 | unsigned int merge_reloc_tree:1; | |
5d4f98a2 | 190 | unsigned int found_file_extent:1; |
5d4f98a2 YZ |
191 | }; |
192 | ||
193 | /* stages of data relocation */ | |
194 | #define MOVE_DATA_EXTENTS 0 | |
195 | #define UPDATE_DATA_PTRS 1 | |
196 | ||
3fd0a558 YZ |
197 | static void remove_backref_node(struct backref_cache *cache, |
198 | struct backref_node *node); | |
199 | static void __mark_block_processed(struct reloc_control *rc, | |
200 | struct backref_node *node); | |
5d4f98a2 YZ |
201 | |
202 | static void mapping_tree_init(struct mapping_tree *tree) | |
203 | { | |
6bef4d31 | 204 | tree->rb_root = RB_ROOT; |
5d4f98a2 YZ |
205 | spin_lock_init(&tree->lock); |
206 | } | |
207 | ||
208 | static void backref_cache_init(struct backref_cache *cache) | |
209 | { | |
210 | int i; | |
6bef4d31 | 211 | cache->rb_root = RB_ROOT; |
5d4f98a2 YZ |
212 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) |
213 | INIT_LIST_HEAD(&cache->pending[i]); | |
3fd0a558 YZ |
214 | INIT_LIST_HEAD(&cache->changed); |
215 | INIT_LIST_HEAD(&cache->detached); | |
216 | INIT_LIST_HEAD(&cache->leaves); | |
217 | } | |
218 | ||
219 | static void backref_cache_cleanup(struct backref_cache *cache) | |
220 | { | |
221 | struct backref_node *node; | |
222 | int i; | |
223 | ||
224 | while (!list_empty(&cache->detached)) { | |
225 | node = list_entry(cache->detached.next, | |
226 | struct backref_node, list); | |
227 | remove_backref_node(cache, node); | |
228 | } | |
229 | ||
230 | while (!list_empty(&cache->leaves)) { | |
231 | node = list_entry(cache->leaves.next, | |
232 | struct backref_node, lower); | |
233 | remove_backref_node(cache, node); | |
234 | } | |
235 | ||
236 | cache->last_trans = 0; | |
237 | ||
238 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
f4907095 LB |
239 | ASSERT(list_empty(&cache->pending[i])); |
240 | ASSERT(list_empty(&cache->changed)); | |
241 | ASSERT(list_empty(&cache->detached)); | |
242 | ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); | |
243 | ASSERT(!cache->nr_nodes); | |
244 | ASSERT(!cache->nr_edges); | |
3fd0a558 YZ |
245 | } |
246 | ||
247 | static struct backref_node *alloc_backref_node(struct backref_cache *cache) | |
248 | { | |
249 | struct backref_node *node; | |
250 | ||
251 | node = kzalloc(sizeof(*node), GFP_NOFS); | |
252 | if (node) { | |
253 | INIT_LIST_HEAD(&node->list); | |
254 | INIT_LIST_HEAD(&node->upper); | |
255 | INIT_LIST_HEAD(&node->lower); | |
256 | RB_CLEAR_NODE(&node->rb_node); | |
257 | cache->nr_nodes++; | |
258 | } | |
259 | return node; | |
260 | } | |
261 | ||
262 | static void free_backref_node(struct backref_cache *cache, | |
263 | struct backref_node *node) | |
264 | { | |
265 | if (node) { | |
266 | cache->nr_nodes--; | |
267 | kfree(node); | |
268 | } | |
269 | } | |
270 | ||
271 | static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) | |
272 | { | |
273 | struct backref_edge *edge; | |
274 | ||
275 | edge = kzalloc(sizeof(*edge), GFP_NOFS); | |
276 | if (edge) | |
277 | cache->nr_edges++; | |
278 | return edge; | |
5d4f98a2 YZ |
279 | } |
280 | ||
3fd0a558 YZ |
281 | static void free_backref_edge(struct backref_cache *cache, |
282 | struct backref_edge *edge) | |
5d4f98a2 | 283 | { |
3fd0a558 YZ |
284 | if (edge) { |
285 | cache->nr_edges--; | |
286 | kfree(edge); | |
287 | } | |
5d4f98a2 YZ |
288 | } |
289 | ||
290 | static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, | |
291 | struct rb_node *node) | |
292 | { | |
293 | struct rb_node **p = &root->rb_node; | |
294 | struct rb_node *parent = NULL; | |
295 | struct tree_entry *entry; | |
296 | ||
297 | while (*p) { | |
298 | parent = *p; | |
299 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
300 | ||
301 | if (bytenr < entry->bytenr) | |
302 | p = &(*p)->rb_left; | |
303 | else if (bytenr > entry->bytenr) | |
304 | p = &(*p)->rb_right; | |
305 | else | |
306 | return parent; | |
307 | } | |
308 | ||
309 | rb_link_node(node, parent, p); | |
310 | rb_insert_color(node, root); | |
311 | return NULL; | |
312 | } | |
313 | ||
314 | static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) | |
315 | { | |
316 | struct rb_node *n = root->rb_node; | |
317 | struct tree_entry *entry; | |
318 | ||
319 | while (n) { | |
320 | entry = rb_entry(n, struct tree_entry, rb_node); | |
321 | ||
322 | if (bytenr < entry->bytenr) | |
323 | n = n->rb_left; | |
324 | else if (bytenr > entry->bytenr) | |
325 | n = n->rb_right; | |
326 | else | |
327 | return n; | |
328 | } | |
329 | return NULL; | |
330 | } | |
331 | ||
48a3b636 | 332 | static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) |
43c04fb1 JM |
333 | { |
334 | ||
335 | struct btrfs_fs_info *fs_info = NULL; | |
336 | struct backref_node *bnode = rb_entry(rb_node, struct backref_node, | |
337 | rb_node); | |
338 | if (bnode->root) | |
339 | fs_info = bnode->root->fs_info; | |
5d163e0e JM |
340 | btrfs_panic(fs_info, errno, |
341 | "Inconsistency in backref cache found at offset %llu", | |
342 | bytenr); | |
43c04fb1 JM |
343 | } |
344 | ||
5d4f98a2 YZ |
345 | /* |
346 | * walk up backref nodes until reach node presents tree root | |
347 | */ | |
348 | static struct backref_node *walk_up_backref(struct backref_node *node, | |
349 | struct backref_edge *edges[], | |
350 | int *index) | |
351 | { | |
352 | struct backref_edge *edge; | |
353 | int idx = *index; | |
354 | ||
355 | while (!list_empty(&node->upper)) { | |
356 | edge = list_entry(node->upper.next, | |
357 | struct backref_edge, list[LOWER]); | |
358 | edges[idx++] = edge; | |
359 | node = edge->node[UPPER]; | |
360 | } | |
3fd0a558 | 361 | BUG_ON(node->detached); |
5d4f98a2 YZ |
362 | *index = idx; |
363 | return node; | |
364 | } | |
365 | ||
366 | /* | |
367 | * walk down backref nodes to find start of next reference path | |
368 | */ | |
369 | static struct backref_node *walk_down_backref(struct backref_edge *edges[], | |
370 | int *index) | |
371 | { | |
372 | struct backref_edge *edge; | |
373 | struct backref_node *lower; | |
374 | int idx = *index; | |
375 | ||
376 | while (idx > 0) { | |
377 | edge = edges[idx - 1]; | |
378 | lower = edge->node[LOWER]; | |
379 | if (list_is_last(&edge->list[LOWER], &lower->upper)) { | |
380 | idx--; | |
381 | continue; | |
382 | } | |
383 | edge = list_entry(edge->list[LOWER].next, | |
384 | struct backref_edge, list[LOWER]); | |
385 | edges[idx - 1] = edge; | |
386 | *index = idx; | |
387 | return edge->node[UPPER]; | |
388 | } | |
389 | *index = 0; | |
390 | return NULL; | |
391 | } | |
392 | ||
3fd0a558 YZ |
393 | static void unlock_node_buffer(struct backref_node *node) |
394 | { | |
395 | if (node->locked) { | |
396 | btrfs_tree_unlock(node->eb); | |
397 | node->locked = 0; | |
398 | } | |
399 | } | |
400 | ||
5d4f98a2 YZ |
401 | static void drop_node_buffer(struct backref_node *node) |
402 | { | |
403 | if (node->eb) { | |
3fd0a558 | 404 | unlock_node_buffer(node); |
5d4f98a2 YZ |
405 | free_extent_buffer(node->eb); |
406 | node->eb = NULL; | |
407 | } | |
408 | } | |
409 | ||
410 | static void drop_backref_node(struct backref_cache *tree, | |
411 | struct backref_node *node) | |
412 | { | |
5d4f98a2 YZ |
413 | BUG_ON(!list_empty(&node->upper)); |
414 | ||
415 | drop_node_buffer(node); | |
3fd0a558 | 416 | list_del(&node->list); |
5d4f98a2 | 417 | list_del(&node->lower); |
3fd0a558 YZ |
418 | if (!RB_EMPTY_NODE(&node->rb_node)) |
419 | rb_erase(&node->rb_node, &tree->rb_root); | |
420 | free_backref_node(tree, node); | |
5d4f98a2 YZ |
421 | } |
422 | ||
423 | /* | |
424 | * remove a backref node from the backref cache | |
425 | */ | |
426 | static void remove_backref_node(struct backref_cache *cache, | |
427 | struct backref_node *node) | |
428 | { | |
429 | struct backref_node *upper; | |
430 | struct backref_edge *edge; | |
431 | ||
432 | if (!node) | |
433 | return; | |
434 | ||
3fd0a558 | 435 | BUG_ON(!node->lowest && !node->detached); |
5d4f98a2 YZ |
436 | while (!list_empty(&node->upper)) { |
437 | edge = list_entry(node->upper.next, struct backref_edge, | |
438 | list[LOWER]); | |
439 | upper = edge->node[UPPER]; | |
440 | list_del(&edge->list[LOWER]); | |
441 | list_del(&edge->list[UPPER]); | |
3fd0a558 YZ |
442 | free_backref_edge(cache, edge); |
443 | ||
444 | if (RB_EMPTY_NODE(&upper->rb_node)) { | |
445 | BUG_ON(!list_empty(&node->upper)); | |
446 | drop_backref_node(cache, node); | |
447 | node = upper; | |
448 | node->lowest = 1; | |
449 | continue; | |
450 | } | |
5d4f98a2 | 451 | /* |
3fd0a558 | 452 | * add the node to leaf node list if no other |
5d4f98a2 YZ |
453 | * child block cached. |
454 | */ | |
455 | if (list_empty(&upper->lower)) { | |
3fd0a558 | 456 | list_add_tail(&upper->lower, &cache->leaves); |
5d4f98a2 YZ |
457 | upper->lowest = 1; |
458 | } | |
459 | } | |
3fd0a558 | 460 | |
5d4f98a2 YZ |
461 | drop_backref_node(cache, node); |
462 | } | |
463 | ||
3fd0a558 YZ |
464 | static void update_backref_node(struct backref_cache *cache, |
465 | struct backref_node *node, u64 bytenr) | |
466 | { | |
467 | struct rb_node *rb_node; | |
468 | rb_erase(&node->rb_node, &cache->rb_root); | |
469 | node->bytenr = bytenr; | |
470 | rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); | |
43c04fb1 JM |
471 | if (rb_node) |
472 | backref_tree_panic(rb_node, -EEXIST, bytenr); | |
3fd0a558 YZ |
473 | } |
474 | ||
475 | /* | |
476 | * update backref cache after a transaction commit | |
477 | */ | |
478 | static int update_backref_cache(struct btrfs_trans_handle *trans, | |
479 | struct backref_cache *cache) | |
480 | { | |
481 | struct backref_node *node; | |
482 | int level = 0; | |
483 | ||
484 | if (cache->last_trans == 0) { | |
485 | cache->last_trans = trans->transid; | |
486 | return 0; | |
487 | } | |
488 | ||
489 | if (cache->last_trans == trans->transid) | |
490 | return 0; | |
491 | ||
492 | /* | |
493 | * detached nodes are used to avoid unnecessary backref | |
494 | * lookup. transaction commit changes the extent tree. | |
495 | * so the detached nodes are no longer useful. | |
496 | */ | |
497 | while (!list_empty(&cache->detached)) { | |
498 | node = list_entry(cache->detached.next, | |
499 | struct backref_node, list); | |
500 | remove_backref_node(cache, node); | |
501 | } | |
502 | ||
503 | while (!list_empty(&cache->changed)) { | |
504 | node = list_entry(cache->changed.next, | |
505 | struct backref_node, list); | |
506 | list_del_init(&node->list); | |
507 | BUG_ON(node->pending); | |
508 | update_backref_node(cache, node, node->new_bytenr); | |
509 | } | |
510 | ||
511 | /* | |
512 | * some nodes can be left in the pending list if there were | |
513 | * errors during processing the pending nodes. | |
514 | */ | |
515 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
516 | list_for_each_entry(node, &cache->pending[level], list) { | |
517 | BUG_ON(!node->pending); | |
518 | if (node->bytenr == node->new_bytenr) | |
519 | continue; | |
520 | update_backref_node(cache, node, node->new_bytenr); | |
521 | } | |
522 | } | |
523 | ||
524 | cache->last_trans = 0; | |
525 | return 1; | |
526 | } | |
527 | ||
f2a97a9d | 528 | |
3fd0a558 YZ |
529 | static int should_ignore_root(struct btrfs_root *root) |
530 | { | |
531 | struct btrfs_root *reloc_root; | |
532 | ||
27cdeb70 | 533 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
3fd0a558 YZ |
534 | return 0; |
535 | ||
536 | reloc_root = root->reloc_root; | |
537 | if (!reloc_root) | |
538 | return 0; | |
539 | ||
540 | if (btrfs_root_last_snapshot(&reloc_root->root_item) == | |
541 | root->fs_info->running_transaction->transid - 1) | |
542 | return 0; | |
543 | /* | |
544 | * if there is reloc tree and it was created in previous | |
545 | * transaction backref lookup can find the reloc tree, | |
546 | * so backref node for the fs tree root is useless for | |
547 | * relocation. | |
548 | */ | |
549 | return 1; | |
550 | } | |
5d4f98a2 YZ |
551 | /* |
552 | * find reloc tree by address of tree root | |
553 | */ | |
554 | static struct btrfs_root *find_reloc_root(struct reloc_control *rc, | |
555 | u64 bytenr) | |
556 | { | |
557 | struct rb_node *rb_node; | |
558 | struct mapping_node *node; | |
559 | struct btrfs_root *root = NULL; | |
560 | ||
561 | spin_lock(&rc->reloc_root_tree.lock); | |
562 | rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); | |
563 | if (rb_node) { | |
564 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
565 | root = (struct btrfs_root *)node->data; | |
566 | } | |
567 | spin_unlock(&rc->reloc_root_tree.lock); | |
568 | return root; | |
569 | } | |
570 | ||
571 | static int is_cowonly_root(u64 root_objectid) | |
572 | { | |
573 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
574 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
575 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
576 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
577 | root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
66463748 WS |
578 | root_objectid == BTRFS_CSUM_TREE_OBJECTID || |
579 | root_objectid == BTRFS_UUID_TREE_OBJECTID || | |
3e4c5efb DS |
580 | root_objectid == BTRFS_QUOTA_TREE_OBJECTID || |
581 | root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) | |
5d4f98a2 YZ |
582 | return 1; |
583 | return 0; | |
584 | } | |
585 | ||
586 | static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, | |
587 | u64 root_objectid) | |
588 | { | |
589 | struct btrfs_key key; | |
590 | ||
591 | key.objectid = root_objectid; | |
592 | key.type = BTRFS_ROOT_ITEM_KEY; | |
593 | if (is_cowonly_root(root_objectid)) | |
594 | key.offset = 0; | |
595 | else | |
596 | key.offset = (u64)-1; | |
597 | ||
c00869f1 | 598 | return btrfs_get_fs_root(fs_info, &key, false); |
5d4f98a2 YZ |
599 | } |
600 | ||
601 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
602 | static noinline_for_stack | |
603 | struct btrfs_root *find_tree_root(struct reloc_control *rc, | |
604 | struct extent_buffer *leaf, | |
605 | struct btrfs_extent_ref_v0 *ref0) | |
606 | { | |
607 | struct btrfs_root *root; | |
608 | u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); | |
609 | u64 generation = btrfs_ref_generation_v0(leaf, ref0); | |
610 | ||
611 | BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); | |
612 | ||
613 | root = read_fs_root(rc->extent_root->fs_info, root_objectid); | |
614 | BUG_ON(IS_ERR(root)); | |
615 | ||
27cdeb70 | 616 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
617 | generation != btrfs_root_generation(&root->root_item)) |
618 | return NULL; | |
619 | ||
620 | return root; | |
621 | } | |
622 | #endif | |
623 | ||
624 | static noinline_for_stack | |
625 | int find_inline_backref(struct extent_buffer *leaf, int slot, | |
626 | unsigned long *ptr, unsigned long *end) | |
627 | { | |
3173a18f | 628 | struct btrfs_key key; |
5d4f98a2 YZ |
629 | struct btrfs_extent_item *ei; |
630 | struct btrfs_tree_block_info *bi; | |
631 | u32 item_size; | |
632 | ||
3173a18f JB |
633 | btrfs_item_key_to_cpu(leaf, &key, slot); |
634 | ||
5d4f98a2 YZ |
635 | item_size = btrfs_item_size_nr(leaf, slot); |
636 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
637 | if (item_size < sizeof(*ei)) { | |
638 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
639 | return 1; | |
640 | } | |
641 | #endif | |
642 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
643 | WARN_ON(!(btrfs_extent_flags(leaf, ei) & | |
644 | BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
645 | ||
3173a18f JB |
646 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
647 | item_size <= sizeof(*ei) + sizeof(*bi)) { | |
5d4f98a2 YZ |
648 | WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); |
649 | return 1; | |
650 | } | |
d062d13c JB |
651 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
652 | item_size <= sizeof(*ei)) { | |
653 | WARN_ON(item_size < sizeof(*ei)); | |
654 | return 1; | |
655 | } | |
5d4f98a2 | 656 | |
3173a18f JB |
657 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
658 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
659 | *ptr = (unsigned long)(bi + 1); | |
660 | } else { | |
661 | *ptr = (unsigned long)(ei + 1); | |
662 | } | |
5d4f98a2 YZ |
663 | *end = (unsigned long)ei + item_size; |
664 | return 0; | |
665 | } | |
666 | ||
667 | /* | |
668 | * build backref tree for a given tree block. root of the backref tree | |
669 | * corresponds the tree block, leaves of the backref tree correspond | |
670 | * roots of b-trees that reference the tree block. | |
671 | * | |
672 | * the basic idea of this function is check backrefs of a given block | |
01327610 NS |
673 | * to find upper level blocks that reference the block, and then check |
674 | * backrefs of these upper level blocks recursively. the recursion stop | |
5d4f98a2 YZ |
675 | * when tree root is reached or backrefs for the block is cached. |
676 | * | |
677 | * NOTE: if we find backrefs for a block are cached, we know backrefs | |
678 | * for all upper level blocks that directly/indirectly reference the | |
679 | * block are also cached. | |
680 | */ | |
3fd0a558 YZ |
681 | static noinline_for_stack |
682 | struct backref_node *build_backref_tree(struct reloc_control *rc, | |
683 | struct btrfs_key *node_key, | |
684 | int level, u64 bytenr) | |
5d4f98a2 | 685 | { |
3fd0a558 | 686 | struct backref_cache *cache = &rc->backref_cache; |
5d4f98a2 YZ |
687 | struct btrfs_path *path1; |
688 | struct btrfs_path *path2; | |
689 | struct extent_buffer *eb; | |
690 | struct btrfs_root *root; | |
691 | struct backref_node *cur; | |
692 | struct backref_node *upper; | |
693 | struct backref_node *lower; | |
694 | struct backref_node *node = NULL; | |
695 | struct backref_node *exist = NULL; | |
696 | struct backref_edge *edge; | |
697 | struct rb_node *rb_node; | |
698 | struct btrfs_key key; | |
699 | unsigned long end; | |
700 | unsigned long ptr; | |
701 | LIST_HEAD(list); | |
3fd0a558 YZ |
702 | LIST_HEAD(useless); |
703 | int cowonly; | |
5d4f98a2 YZ |
704 | int ret; |
705 | int err = 0; | |
b6c60c80 | 706 | bool need_check = true; |
5d4f98a2 YZ |
707 | |
708 | path1 = btrfs_alloc_path(); | |
709 | path2 = btrfs_alloc_path(); | |
710 | if (!path1 || !path2) { | |
711 | err = -ENOMEM; | |
712 | goto out; | |
713 | } | |
e4058b54 DS |
714 | path1->reada = READA_FORWARD; |
715 | path2->reada = READA_FORWARD; | |
5d4f98a2 | 716 | |
3fd0a558 | 717 | node = alloc_backref_node(cache); |
5d4f98a2 YZ |
718 | if (!node) { |
719 | err = -ENOMEM; | |
720 | goto out; | |
721 | } | |
722 | ||
5d4f98a2 | 723 | node->bytenr = bytenr; |
5d4f98a2 YZ |
724 | node->level = level; |
725 | node->lowest = 1; | |
726 | cur = node; | |
727 | again: | |
728 | end = 0; | |
729 | ptr = 0; | |
730 | key.objectid = cur->bytenr; | |
3173a18f | 731 | key.type = BTRFS_METADATA_ITEM_KEY; |
5d4f98a2 YZ |
732 | key.offset = (u64)-1; |
733 | ||
734 | path1->search_commit_root = 1; | |
735 | path1->skip_locking = 1; | |
736 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, | |
737 | 0, 0); | |
738 | if (ret < 0) { | |
739 | err = ret; | |
740 | goto out; | |
741 | } | |
75bfb9af JB |
742 | ASSERT(ret); |
743 | ASSERT(path1->slots[0]); | |
5d4f98a2 YZ |
744 | |
745 | path1->slots[0]--; | |
746 | ||
747 | WARN_ON(cur->checked); | |
748 | if (!list_empty(&cur->upper)) { | |
749 | /* | |
70f23fd6 | 750 | * the backref was added previously when processing |
5d4f98a2 YZ |
751 | * backref of type BTRFS_TREE_BLOCK_REF_KEY |
752 | */ | |
75bfb9af | 753 | ASSERT(list_is_singular(&cur->upper)); |
5d4f98a2 YZ |
754 | edge = list_entry(cur->upper.next, struct backref_edge, |
755 | list[LOWER]); | |
75bfb9af | 756 | ASSERT(list_empty(&edge->list[UPPER])); |
5d4f98a2 YZ |
757 | exist = edge->node[UPPER]; |
758 | /* | |
759 | * add the upper level block to pending list if we need | |
760 | * check its backrefs | |
761 | */ | |
762 | if (!exist->checked) | |
763 | list_add_tail(&edge->list[UPPER], &list); | |
764 | } else { | |
765 | exist = NULL; | |
766 | } | |
767 | ||
768 | while (1) { | |
769 | cond_resched(); | |
770 | eb = path1->nodes[0]; | |
771 | ||
772 | if (ptr >= end) { | |
773 | if (path1->slots[0] >= btrfs_header_nritems(eb)) { | |
774 | ret = btrfs_next_leaf(rc->extent_root, path1); | |
775 | if (ret < 0) { | |
776 | err = ret; | |
777 | goto out; | |
778 | } | |
779 | if (ret > 0) | |
780 | break; | |
781 | eb = path1->nodes[0]; | |
782 | } | |
783 | ||
784 | btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); | |
785 | if (key.objectid != cur->bytenr) { | |
786 | WARN_ON(exist); | |
787 | break; | |
788 | } | |
789 | ||
3173a18f JB |
790 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
791 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
5d4f98a2 YZ |
792 | ret = find_inline_backref(eb, path1->slots[0], |
793 | &ptr, &end); | |
794 | if (ret) | |
795 | goto next; | |
796 | } | |
797 | } | |
798 | ||
799 | if (ptr < end) { | |
800 | /* update key for inline back ref */ | |
801 | struct btrfs_extent_inline_ref *iref; | |
3de28d57 | 802 | int type; |
5d4f98a2 | 803 | iref = (struct btrfs_extent_inline_ref *)ptr; |
3de28d57 LB |
804 | type = btrfs_get_extent_inline_ref_type(eb, iref, |
805 | BTRFS_REF_TYPE_BLOCK); | |
806 | if (type == BTRFS_REF_TYPE_INVALID) { | |
807 | err = -EINVAL; | |
808 | goto out; | |
809 | } | |
810 | key.type = type; | |
5d4f98a2 | 811 | key.offset = btrfs_extent_inline_ref_offset(eb, iref); |
3de28d57 | 812 | |
5d4f98a2 YZ |
813 | WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && |
814 | key.type != BTRFS_SHARED_BLOCK_REF_KEY); | |
815 | } | |
816 | ||
817 | if (exist && | |
818 | ((key.type == BTRFS_TREE_BLOCK_REF_KEY && | |
819 | exist->owner == key.offset) || | |
820 | (key.type == BTRFS_SHARED_BLOCK_REF_KEY && | |
821 | exist->bytenr == key.offset))) { | |
822 | exist = NULL; | |
823 | goto next; | |
824 | } | |
825 | ||
826 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
827 | if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || | |
828 | key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
3fd0a558 | 829 | if (key.type == BTRFS_EXTENT_REF_V0_KEY) { |
5d4f98a2 YZ |
830 | struct btrfs_extent_ref_v0 *ref0; |
831 | ref0 = btrfs_item_ptr(eb, path1->slots[0], | |
832 | struct btrfs_extent_ref_v0); | |
3fd0a558 | 833 | if (key.objectid == key.offset) { |
046f264f | 834 | root = find_tree_root(rc, eb, ref0); |
3fd0a558 YZ |
835 | if (root && !should_ignore_root(root)) |
836 | cur->root = root; | |
837 | else | |
838 | list_add(&cur->list, &useless); | |
839 | break; | |
840 | } | |
046f264f YZ |
841 | if (is_cowonly_root(btrfs_ref_root_v0(eb, |
842 | ref0))) | |
843 | cur->cowonly = 1; | |
5d4f98a2 YZ |
844 | } |
845 | #else | |
75bfb9af | 846 | ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY); |
5d4f98a2 YZ |
847 | if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { |
848 | #endif | |
849 | if (key.objectid == key.offset) { | |
850 | /* | |
851 | * only root blocks of reloc trees use | |
852 | * backref of this type. | |
853 | */ | |
854 | root = find_reloc_root(rc, cur->bytenr); | |
75bfb9af | 855 | ASSERT(root); |
5d4f98a2 YZ |
856 | cur->root = root; |
857 | break; | |
858 | } | |
859 | ||
3fd0a558 | 860 | edge = alloc_backref_edge(cache); |
5d4f98a2 YZ |
861 | if (!edge) { |
862 | err = -ENOMEM; | |
863 | goto out; | |
864 | } | |
865 | rb_node = tree_search(&cache->rb_root, key.offset); | |
866 | if (!rb_node) { | |
3fd0a558 | 867 | upper = alloc_backref_node(cache); |
5d4f98a2 | 868 | if (!upper) { |
3fd0a558 | 869 | free_backref_edge(cache, edge); |
5d4f98a2 YZ |
870 | err = -ENOMEM; |
871 | goto out; | |
872 | } | |
5d4f98a2 | 873 | upper->bytenr = key.offset; |
5d4f98a2 YZ |
874 | upper->level = cur->level + 1; |
875 | /* | |
876 | * backrefs for the upper level block isn't | |
877 | * cached, add the block to pending list | |
878 | */ | |
879 | list_add_tail(&edge->list[UPPER], &list); | |
880 | } else { | |
881 | upper = rb_entry(rb_node, struct backref_node, | |
882 | rb_node); | |
75bfb9af | 883 | ASSERT(upper->checked); |
5d4f98a2 YZ |
884 | INIT_LIST_HEAD(&edge->list[UPPER]); |
885 | } | |
3fd0a558 | 886 | list_add_tail(&edge->list[LOWER], &cur->upper); |
5d4f98a2 | 887 | edge->node[LOWER] = cur; |
3fd0a558 | 888 | edge->node[UPPER] = upper; |
5d4f98a2 YZ |
889 | |
890 | goto next; | |
891 | } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { | |
892 | goto next; | |
893 | } | |
894 | ||
895 | /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ | |
896 | root = read_fs_root(rc->extent_root->fs_info, key.offset); | |
897 | if (IS_ERR(root)) { | |
898 | err = PTR_ERR(root); | |
899 | goto out; | |
900 | } | |
901 | ||
27cdeb70 | 902 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
3fd0a558 YZ |
903 | cur->cowonly = 1; |
904 | ||
5d4f98a2 YZ |
905 | if (btrfs_root_level(&root->root_item) == cur->level) { |
906 | /* tree root */ | |
75bfb9af | 907 | ASSERT(btrfs_root_bytenr(&root->root_item) == |
5d4f98a2 | 908 | cur->bytenr); |
3fd0a558 YZ |
909 | if (should_ignore_root(root)) |
910 | list_add(&cur->list, &useless); | |
911 | else | |
912 | cur->root = root; | |
5d4f98a2 YZ |
913 | break; |
914 | } | |
915 | ||
916 | level = cur->level + 1; | |
917 | ||
918 | /* | |
919 | * searching the tree to find upper level blocks | |
920 | * reference the block. | |
921 | */ | |
922 | path2->search_commit_root = 1; | |
923 | path2->skip_locking = 1; | |
924 | path2->lowest_level = level; | |
925 | ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); | |
926 | path2->lowest_level = 0; | |
927 | if (ret < 0) { | |
928 | err = ret; | |
929 | goto out; | |
930 | } | |
33c66f43 YZ |
931 | if (ret > 0 && path2->slots[level] > 0) |
932 | path2->slots[level]--; | |
5d4f98a2 YZ |
933 | |
934 | eb = path2->nodes[level]; | |
3561b9db LB |
935 | if (btrfs_node_blockptr(eb, path2->slots[level]) != |
936 | cur->bytenr) { | |
937 | btrfs_err(root->fs_info, | |
938 | "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", | |
939 | cur->bytenr, level - 1, root->objectid, | |
940 | node_key->objectid, node_key->type, | |
941 | node_key->offset); | |
942 | err = -ENOENT; | |
943 | goto out; | |
944 | } | |
5d4f98a2 | 945 | lower = cur; |
b6c60c80 | 946 | need_check = true; |
5d4f98a2 YZ |
947 | for (; level < BTRFS_MAX_LEVEL; level++) { |
948 | if (!path2->nodes[level]) { | |
75bfb9af | 949 | ASSERT(btrfs_root_bytenr(&root->root_item) == |
5d4f98a2 | 950 | lower->bytenr); |
3fd0a558 YZ |
951 | if (should_ignore_root(root)) |
952 | list_add(&lower->list, &useless); | |
953 | else | |
954 | lower->root = root; | |
5d4f98a2 YZ |
955 | break; |
956 | } | |
957 | ||
3fd0a558 | 958 | edge = alloc_backref_edge(cache); |
5d4f98a2 YZ |
959 | if (!edge) { |
960 | err = -ENOMEM; | |
961 | goto out; | |
962 | } | |
963 | ||
964 | eb = path2->nodes[level]; | |
965 | rb_node = tree_search(&cache->rb_root, eb->start); | |
966 | if (!rb_node) { | |
3fd0a558 | 967 | upper = alloc_backref_node(cache); |
5d4f98a2 | 968 | if (!upper) { |
3fd0a558 | 969 | free_backref_edge(cache, edge); |
5d4f98a2 YZ |
970 | err = -ENOMEM; |
971 | goto out; | |
972 | } | |
5d4f98a2 YZ |
973 | upper->bytenr = eb->start; |
974 | upper->owner = btrfs_header_owner(eb); | |
975 | upper->level = lower->level + 1; | |
27cdeb70 MX |
976 | if (!test_bit(BTRFS_ROOT_REF_COWS, |
977 | &root->state)) | |
3fd0a558 | 978 | upper->cowonly = 1; |
5d4f98a2 YZ |
979 | |
980 | /* | |
981 | * if we know the block isn't shared | |
982 | * we can void checking its backrefs. | |
983 | */ | |
984 | if (btrfs_block_can_be_shared(root, eb)) | |
985 | upper->checked = 0; | |
986 | else | |
987 | upper->checked = 1; | |
988 | ||
989 | /* | |
990 | * add the block to pending list if we | |
b6c60c80 JB |
991 | * need check its backrefs, we only do this once |
992 | * while walking up a tree as we will catch | |
993 | * anything else later on. | |
5d4f98a2 | 994 | */ |
b6c60c80 JB |
995 | if (!upper->checked && need_check) { |
996 | need_check = false; | |
5d4f98a2 YZ |
997 | list_add_tail(&edge->list[UPPER], |
998 | &list); | |
bbe90514 JB |
999 | } else { |
1000 | if (upper->checked) | |
1001 | need_check = true; | |
5d4f98a2 | 1002 | INIT_LIST_HEAD(&edge->list[UPPER]); |
bbe90514 | 1003 | } |
5d4f98a2 YZ |
1004 | } else { |
1005 | upper = rb_entry(rb_node, struct backref_node, | |
1006 | rb_node); | |
75bfb9af | 1007 | ASSERT(upper->checked); |
5d4f98a2 | 1008 | INIT_LIST_HEAD(&edge->list[UPPER]); |
3fd0a558 YZ |
1009 | if (!upper->owner) |
1010 | upper->owner = btrfs_header_owner(eb); | |
5d4f98a2 YZ |
1011 | } |
1012 | list_add_tail(&edge->list[LOWER], &lower->upper); | |
5d4f98a2 | 1013 | edge->node[LOWER] = lower; |
3fd0a558 | 1014 | edge->node[UPPER] = upper; |
5d4f98a2 YZ |
1015 | |
1016 | if (rb_node) | |
1017 | break; | |
1018 | lower = upper; | |
1019 | upper = NULL; | |
1020 | } | |
b3b4aa74 | 1021 | btrfs_release_path(path2); |
5d4f98a2 YZ |
1022 | next: |
1023 | if (ptr < end) { | |
1024 | ptr += btrfs_extent_inline_ref_size(key.type); | |
1025 | if (ptr >= end) { | |
1026 | WARN_ON(ptr > end); | |
1027 | ptr = 0; | |
1028 | end = 0; | |
1029 | } | |
1030 | } | |
1031 | if (ptr >= end) | |
1032 | path1->slots[0]++; | |
1033 | } | |
b3b4aa74 | 1034 | btrfs_release_path(path1); |
5d4f98a2 YZ |
1035 | |
1036 | cur->checked = 1; | |
1037 | WARN_ON(exist); | |
1038 | ||
1039 | /* the pending list isn't empty, take the first block to process */ | |
1040 | if (!list_empty(&list)) { | |
1041 | edge = list_entry(list.next, struct backref_edge, list[UPPER]); | |
1042 | list_del_init(&edge->list[UPPER]); | |
1043 | cur = edge->node[UPPER]; | |
1044 | goto again; | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * everything goes well, connect backref nodes and insert backref nodes | |
1049 | * into the cache. | |
1050 | */ | |
75bfb9af | 1051 | ASSERT(node->checked); |
3fd0a558 YZ |
1052 | cowonly = node->cowonly; |
1053 | if (!cowonly) { | |
1054 | rb_node = tree_insert(&cache->rb_root, node->bytenr, | |
1055 | &node->rb_node); | |
43c04fb1 JM |
1056 | if (rb_node) |
1057 | backref_tree_panic(rb_node, -EEXIST, node->bytenr); | |
3fd0a558 YZ |
1058 | list_add_tail(&node->lower, &cache->leaves); |
1059 | } | |
5d4f98a2 YZ |
1060 | |
1061 | list_for_each_entry(edge, &node->upper, list[LOWER]) | |
1062 | list_add_tail(&edge->list[UPPER], &list); | |
1063 | ||
1064 | while (!list_empty(&list)) { | |
1065 | edge = list_entry(list.next, struct backref_edge, list[UPPER]); | |
1066 | list_del_init(&edge->list[UPPER]); | |
1067 | upper = edge->node[UPPER]; | |
3fd0a558 YZ |
1068 | if (upper->detached) { |
1069 | list_del(&edge->list[LOWER]); | |
1070 | lower = edge->node[LOWER]; | |
1071 | free_backref_edge(cache, edge); | |
1072 | if (list_empty(&lower->upper)) | |
1073 | list_add(&lower->list, &useless); | |
1074 | continue; | |
1075 | } | |
5d4f98a2 YZ |
1076 | |
1077 | if (!RB_EMPTY_NODE(&upper->rb_node)) { | |
1078 | if (upper->lowest) { | |
1079 | list_del_init(&upper->lower); | |
1080 | upper->lowest = 0; | |
1081 | } | |
1082 | ||
1083 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
1084 | continue; | |
1085 | } | |
1086 | ||
75bfb9af JB |
1087 | if (!upper->checked) { |
1088 | /* | |
1089 | * Still want to blow up for developers since this is a | |
1090 | * logic bug. | |
1091 | */ | |
1092 | ASSERT(0); | |
1093 | err = -EINVAL; | |
1094 | goto out; | |
1095 | } | |
1096 | if (cowonly != upper->cowonly) { | |
1097 | ASSERT(0); | |
1098 | err = -EINVAL; | |
1099 | goto out; | |
1100 | } | |
1101 | ||
3fd0a558 YZ |
1102 | if (!cowonly) { |
1103 | rb_node = tree_insert(&cache->rb_root, upper->bytenr, | |
1104 | &upper->rb_node); | |
43c04fb1 JM |
1105 | if (rb_node) |
1106 | backref_tree_panic(rb_node, -EEXIST, | |
1107 | upper->bytenr); | |
3fd0a558 | 1108 | } |
5d4f98a2 YZ |
1109 | |
1110 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
1111 | ||
1112 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
1113 | list_add_tail(&edge->list[UPPER], &list); | |
1114 | } | |
3fd0a558 YZ |
1115 | /* |
1116 | * process useless backref nodes. backref nodes for tree leaves | |
1117 | * are deleted from the cache. backref nodes for upper level | |
1118 | * tree blocks are left in the cache to avoid unnecessary backref | |
1119 | * lookup. | |
1120 | */ | |
1121 | while (!list_empty(&useless)) { | |
1122 | upper = list_entry(useless.next, struct backref_node, list); | |
1123 | list_del_init(&upper->list); | |
75bfb9af | 1124 | ASSERT(list_empty(&upper->upper)); |
3fd0a558 YZ |
1125 | if (upper == node) |
1126 | node = NULL; | |
1127 | if (upper->lowest) { | |
1128 | list_del_init(&upper->lower); | |
1129 | upper->lowest = 0; | |
1130 | } | |
1131 | while (!list_empty(&upper->lower)) { | |
1132 | edge = list_entry(upper->lower.next, | |
1133 | struct backref_edge, list[UPPER]); | |
1134 | list_del(&edge->list[UPPER]); | |
1135 | list_del(&edge->list[LOWER]); | |
1136 | lower = edge->node[LOWER]; | |
1137 | free_backref_edge(cache, edge); | |
1138 | ||
1139 | if (list_empty(&lower->upper)) | |
1140 | list_add(&lower->list, &useless); | |
1141 | } | |
1142 | __mark_block_processed(rc, upper); | |
1143 | if (upper->level > 0) { | |
1144 | list_add(&upper->list, &cache->detached); | |
1145 | upper->detached = 1; | |
1146 | } else { | |
1147 | rb_erase(&upper->rb_node, &cache->rb_root); | |
1148 | free_backref_node(cache, upper); | |
1149 | } | |
1150 | } | |
5d4f98a2 YZ |
1151 | out: |
1152 | btrfs_free_path(path1); | |
1153 | btrfs_free_path(path2); | |
1154 | if (err) { | |
3fd0a558 YZ |
1155 | while (!list_empty(&useless)) { |
1156 | lower = list_entry(useless.next, | |
75bfb9af JB |
1157 | struct backref_node, list); |
1158 | list_del_init(&lower->list); | |
3fd0a558 | 1159 | } |
75bfb9af JB |
1160 | while (!list_empty(&list)) { |
1161 | edge = list_first_entry(&list, struct backref_edge, | |
1162 | list[UPPER]); | |
1163 | list_del(&edge->list[UPPER]); | |
3fd0a558 | 1164 | list_del(&edge->list[LOWER]); |
75bfb9af | 1165 | lower = edge->node[LOWER]; |
5d4f98a2 | 1166 | upper = edge->node[UPPER]; |
3fd0a558 | 1167 | free_backref_edge(cache, edge); |
75bfb9af JB |
1168 | |
1169 | /* | |
1170 | * Lower is no longer linked to any upper backref nodes | |
1171 | * and isn't in the cache, we can free it ourselves. | |
1172 | */ | |
1173 | if (list_empty(&lower->upper) && | |
1174 | RB_EMPTY_NODE(&lower->rb_node)) | |
1175 | list_add(&lower->list, &useless); | |
1176 | ||
1177 | if (!RB_EMPTY_NODE(&upper->rb_node)) | |
1178 | continue; | |
1179 | ||
01327610 | 1180 | /* Add this guy's upper edges to the list to process */ |
75bfb9af JB |
1181 | list_for_each_entry(edge, &upper->upper, list[LOWER]) |
1182 | list_add_tail(&edge->list[UPPER], &list); | |
1183 | if (list_empty(&upper->upper)) | |
1184 | list_add(&upper->list, &useless); | |
1185 | } | |
1186 | ||
1187 | while (!list_empty(&useless)) { | |
1188 | lower = list_entry(useless.next, | |
1189 | struct backref_node, list); | |
1190 | list_del_init(&lower->list); | |
0fd8c3da LB |
1191 | if (lower == node) |
1192 | node = NULL; | |
75bfb9af | 1193 | free_backref_node(cache, lower); |
5d4f98a2 | 1194 | } |
0fd8c3da LB |
1195 | |
1196 | free_backref_node(cache, node); | |
5d4f98a2 YZ |
1197 | return ERR_PTR(err); |
1198 | } | |
75bfb9af | 1199 | ASSERT(!node || !node->detached); |
5d4f98a2 YZ |
1200 | return node; |
1201 | } | |
1202 | ||
3fd0a558 YZ |
1203 | /* |
1204 | * helper to add backref node for the newly created snapshot. | |
1205 | * the backref node is created by cloning backref node that | |
1206 | * corresponds to root of source tree | |
1207 | */ | |
1208 | static int clone_backref_node(struct btrfs_trans_handle *trans, | |
1209 | struct reloc_control *rc, | |
1210 | struct btrfs_root *src, | |
1211 | struct btrfs_root *dest) | |
1212 | { | |
1213 | struct btrfs_root *reloc_root = src->reloc_root; | |
1214 | struct backref_cache *cache = &rc->backref_cache; | |
1215 | struct backref_node *node = NULL; | |
1216 | struct backref_node *new_node; | |
1217 | struct backref_edge *edge; | |
1218 | struct backref_edge *new_edge; | |
1219 | struct rb_node *rb_node; | |
1220 | ||
1221 | if (cache->last_trans > 0) | |
1222 | update_backref_cache(trans, cache); | |
1223 | ||
1224 | rb_node = tree_search(&cache->rb_root, src->commit_root->start); | |
1225 | if (rb_node) { | |
1226 | node = rb_entry(rb_node, struct backref_node, rb_node); | |
1227 | if (node->detached) | |
1228 | node = NULL; | |
1229 | else | |
1230 | BUG_ON(node->new_bytenr != reloc_root->node->start); | |
1231 | } | |
1232 | ||
1233 | if (!node) { | |
1234 | rb_node = tree_search(&cache->rb_root, | |
1235 | reloc_root->commit_root->start); | |
1236 | if (rb_node) { | |
1237 | node = rb_entry(rb_node, struct backref_node, | |
1238 | rb_node); | |
1239 | BUG_ON(node->detached); | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | if (!node) | |
1244 | return 0; | |
1245 | ||
1246 | new_node = alloc_backref_node(cache); | |
1247 | if (!new_node) | |
1248 | return -ENOMEM; | |
1249 | ||
1250 | new_node->bytenr = dest->node->start; | |
1251 | new_node->level = node->level; | |
1252 | new_node->lowest = node->lowest; | |
6848ad64 | 1253 | new_node->checked = 1; |
3fd0a558 YZ |
1254 | new_node->root = dest; |
1255 | ||
1256 | if (!node->lowest) { | |
1257 | list_for_each_entry(edge, &node->lower, list[UPPER]) { | |
1258 | new_edge = alloc_backref_edge(cache); | |
1259 | if (!new_edge) | |
1260 | goto fail; | |
1261 | ||
1262 | new_edge->node[UPPER] = new_node; | |
1263 | new_edge->node[LOWER] = edge->node[LOWER]; | |
1264 | list_add_tail(&new_edge->list[UPPER], | |
1265 | &new_node->lower); | |
1266 | } | |
76b9e23d MX |
1267 | } else { |
1268 | list_add_tail(&new_node->lower, &cache->leaves); | |
3fd0a558 YZ |
1269 | } |
1270 | ||
1271 | rb_node = tree_insert(&cache->rb_root, new_node->bytenr, | |
1272 | &new_node->rb_node); | |
43c04fb1 JM |
1273 | if (rb_node) |
1274 | backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); | |
3fd0a558 YZ |
1275 | |
1276 | if (!new_node->lowest) { | |
1277 | list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { | |
1278 | list_add_tail(&new_edge->list[LOWER], | |
1279 | &new_edge->node[LOWER]->upper); | |
1280 | } | |
1281 | } | |
1282 | return 0; | |
1283 | fail: | |
1284 | while (!list_empty(&new_node->lower)) { | |
1285 | new_edge = list_entry(new_node->lower.next, | |
1286 | struct backref_edge, list[UPPER]); | |
1287 | list_del(&new_edge->list[UPPER]); | |
1288 | free_backref_edge(cache, new_edge); | |
1289 | } | |
1290 | free_backref_node(cache, new_node); | |
1291 | return -ENOMEM; | |
1292 | } | |
1293 | ||
5d4f98a2 YZ |
1294 | /* |
1295 | * helper to add 'address of tree root -> reloc tree' mapping | |
1296 | */ | |
ffd7b339 | 1297 | static int __must_check __add_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 1298 | { |
0b246afa | 1299 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1300 | struct rb_node *rb_node; |
1301 | struct mapping_node *node; | |
0b246afa | 1302 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
1303 | |
1304 | node = kmalloc(sizeof(*node), GFP_NOFS); | |
ffd7b339 JM |
1305 | if (!node) |
1306 | return -ENOMEM; | |
5d4f98a2 YZ |
1307 | |
1308 | node->bytenr = root->node->start; | |
1309 | node->data = root; | |
1310 | ||
1311 | spin_lock(&rc->reloc_root_tree.lock); | |
1312 | rb_node = tree_insert(&rc->reloc_root_tree.rb_root, | |
1313 | node->bytenr, &node->rb_node); | |
1314 | spin_unlock(&rc->reloc_root_tree.lock); | |
ffd7b339 | 1315 | if (rb_node) { |
0b246afa | 1316 | btrfs_panic(fs_info, -EEXIST, |
5d163e0e JM |
1317 | "Duplicate root found for start=%llu while inserting into relocation tree", |
1318 | node->bytenr); | |
ffd7b339 | 1319 | } |
5d4f98a2 YZ |
1320 | |
1321 | list_add_tail(&root->root_list, &rc->reloc_roots); | |
1322 | return 0; | |
1323 | } | |
1324 | ||
1325 | /* | |
c974c464 | 1326 | * helper to delete the 'address of tree root -> reloc tree' |
5d4f98a2 YZ |
1327 | * mapping |
1328 | */ | |
c974c464 | 1329 | static void __del_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 1330 | { |
0b246afa | 1331 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1332 | struct rb_node *rb_node; |
1333 | struct mapping_node *node = NULL; | |
0b246afa | 1334 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
1335 | |
1336 | spin_lock(&rc->reloc_root_tree.lock); | |
1337 | rb_node = tree_search(&rc->reloc_root_tree.rb_root, | |
c974c464 | 1338 | root->node->start); |
5d4f98a2 YZ |
1339 | if (rb_node) { |
1340 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
1341 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
1342 | } | |
1343 | spin_unlock(&rc->reloc_root_tree.lock); | |
1344 | ||
8f71f3e0 | 1345 | if (!node) |
c974c464 | 1346 | return; |
5d4f98a2 YZ |
1347 | BUG_ON((struct btrfs_root *)node->data != root); |
1348 | ||
0b246afa | 1349 | spin_lock(&fs_info->trans_lock); |
c974c464 | 1350 | list_del_init(&root->root_list); |
0b246afa | 1351 | spin_unlock(&fs_info->trans_lock); |
c974c464 WS |
1352 | kfree(node); |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * helper to update the 'address of tree root -> reloc tree' | |
1357 | * mapping | |
1358 | */ | |
1359 | static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) | |
1360 | { | |
0b246afa | 1361 | struct btrfs_fs_info *fs_info = root->fs_info; |
c974c464 WS |
1362 | struct rb_node *rb_node; |
1363 | struct mapping_node *node = NULL; | |
0b246afa | 1364 | struct reloc_control *rc = fs_info->reloc_ctl; |
c974c464 WS |
1365 | |
1366 | spin_lock(&rc->reloc_root_tree.lock); | |
1367 | rb_node = tree_search(&rc->reloc_root_tree.rb_root, | |
1368 | root->node->start); | |
1369 | if (rb_node) { | |
1370 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
1371 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
5d4f98a2 | 1372 | } |
c974c464 WS |
1373 | spin_unlock(&rc->reloc_root_tree.lock); |
1374 | ||
1375 | if (!node) | |
1376 | return 0; | |
1377 | BUG_ON((struct btrfs_root *)node->data != root); | |
1378 | ||
1379 | spin_lock(&rc->reloc_root_tree.lock); | |
1380 | node->bytenr = new_bytenr; | |
1381 | rb_node = tree_insert(&rc->reloc_root_tree.rb_root, | |
1382 | node->bytenr, &node->rb_node); | |
1383 | spin_unlock(&rc->reloc_root_tree.lock); | |
1384 | if (rb_node) | |
1385 | backref_tree_panic(rb_node, -EEXIST, node->bytenr); | |
5d4f98a2 YZ |
1386 | return 0; |
1387 | } | |
1388 | ||
3fd0a558 YZ |
1389 | static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, |
1390 | struct btrfs_root *root, u64 objectid) | |
5d4f98a2 | 1391 | { |
0b246afa | 1392 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1393 | struct btrfs_root *reloc_root; |
1394 | struct extent_buffer *eb; | |
1395 | struct btrfs_root_item *root_item; | |
1396 | struct btrfs_key root_key; | |
1397 | int ret; | |
1398 | ||
5d4f98a2 YZ |
1399 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); |
1400 | BUG_ON(!root_item); | |
1401 | ||
1402 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
1403 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
3fd0a558 | 1404 | root_key.offset = objectid; |
5d4f98a2 | 1405 | |
3fd0a558 | 1406 | if (root->root_key.objectid == objectid) { |
054570a1 FM |
1407 | u64 commit_root_gen; |
1408 | ||
3fd0a558 YZ |
1409 | /* called by btrfs_init_reloc_root */ |
1410 | ret = btrfs_copy_root(trans, root, root->commit_root, &eb, | |
1411 | BTRFS_TREE_RELOC_OBJECTID); | |
1412 | BUG_ON(ret); | |
054570a1 FM |
1413 | /* |
1414 | * Set the last_snapshot field to the generation of the commit | |
1415 | * root - like this ctree.c:btrfs_block_can_be_shared() behaves | |
1416 | * correctly (returns true) when the relocation root is created | |
1417 | * either inside the critical section of a transaction commit | |
1418 | * (through transaction.c:qgroup_account_snapshot()) and when | |
1419 | * it's created before the transaction commit is started. | |
1420 | */ | |
1421 | commit_root_gen = btrfs_header_generation(root->commit_root); | |
1422 | btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); | |
3fd0a558 YZ |
1423 | } else { |
1424 | /* | |
1425 | * called by btrfs_reloc_post_snapshot_hook. | |
1426 | * the source tree is a reloc tree, all tree blocks | |
1427 | * modified after it was created have RELOC flag | |
1428 | * set in their headers. so it's OK to not update | |
1429 | * the 'last_snapshot'. | |
1430 | */ | |
1431 | ret = btrfs_copy_root(trans, root, root->node, &eb, | |
1432 | BTRFS_TREE_RELOC_OBJECTID); | |
1433 | BUG_ON(ret); | |
1434 | } | |
5d4f98a2 | 1435 | |
5d4f98a2 | 1436 | memcpy(root_item, &root->root_item, sizeof(*root_item)); |
5d4f98a2 YZ |
1437 | btrfs_set_root_bytenr(root_item, eb->start); |
1438 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
1439 | btrfs_set_root_generation(root_item, trans->transid); | |
3fd0a558 YZ |
1440 | |
1441 | if (root->root_key.objectid == objectid) { | |
1442 | btrfs_set_root_refs(root_item, 0); | |
1443 | memset(&root_item->drop_progress, 0, | |
1444 | sizeof(struct btrfs_disk_key)); | |
1445 | root_item->drop_level = 0; | |
1446 | } | |
5d4f98a2 YZ |
1447 | |
1448 | btrfs_tree_unlock(eb); | |
1449 | free_extent_buffer(eb); | |
1450 | ||
0b246afa | 1451 | ret = btrfs_insert_root(trans, fs_info->tree_root, |
5d4f98a2 YZ |
1452 | &root_key, root_item); |
1453 | BUG_ON(ret); | |
1454 | kfree(root_item); | |
1455 | ||
0b246afa | 1456 | reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); |
5d4f98a2 YZ |
1457 | BUG_ON(IS_ERR(reloc_root)); |
1458 | reloc_root->last_trans = trans->transid; | |
3fd0a558 YZ |
1459 | return reloc_root; |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * create reloc tree for a given fs tree. reloc tree is just a | |
1464 | * snapshot of the fs tree with special root objectid. | |
1465 | */ | |
1466 | int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, | |
1467 | struct btrfs_root *root) | |
1468 | { | |
0b246afa | 1469 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 | 1470 | struct btrfs_root *reloc_root; |
0b246afa | 1471 | struct reloc_control *rc = fs_info->reloc_ctl; |
20dd2cbf | 1472 | struct btrfs_block_rsv *rsv; |
3fd0a558 | 1473 | int clear_rsv = 0; |
ffd7b339 | 1474 | int ret; |
3fd0a558 YZ |
1475 | |
1476 | if (root->reloc_root) { | |
1477 | reloc_root = root->reloc_root; | |
1478 | reloc_root->last_trans = trans->transid; | |
1479 | return 0; | |
1480 | } | |
1481 | ||
1482 | if (!rc || !rc->create_reloc_tree || | |
1483 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1484 | return 0; | |
1485 | ||
20dd2cbf MX |
1486 | if (!trans->reloc_reserved) { |
1487 | rsv = trans->block_rsv; | |
3fd0a558 YZ |
1488 | trans->block_rsv = rc->block_rsv; |
1489 | clear_rsv = 1; | |
1490 | } | |
1491 | reloc_root = create_reloc_root(trans, root, root->root_key.objectid); | |
1492 | if (clear_rsv) | |
20dd2cbf | 1493 | trans->block_rsv = rsv; |
5d4f98a2 | 1494 | |
ffd7b339 JM |
1495 | ret = __add_reloc_root(reloc_root); |
1496 | BUG_ON(ret < 0); | |
5d4f98a2 YZ |
1497 | root->reloc_root = reloc_root; |
1498 | return 0; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * update root item of reloc tree | |
1503 | */ | |
1504 | int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, | |
1505 | struct btrfs_root *root) | |
1506 | { | |
0b246afa | 1507 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1508 | struct btrfs_root *reloc_root; |
1509 | struct btrfs_root_item *root_item; | |
5d4f98a2 YZ |
1510 | int ret; |
1511 | ||
1512 | if (!root->reloc_root) | |
7585717f | 1513 | goto out; |
5d4f98a2 YZ |
1514 | |
1515 | reloc_root = root->reloc_root; | |
1516 | root_item = &reloc_root->root_item; | |
1517 | ||
0b246afa | 1518 | if (fs_info->reloc_ctl->merge_reloc_tree && |
3fd0a558 | 1519 | btrfs_root_refs(root_item) == 0) { |
5d4f98a2 | 1520 | root->reloc_root = NULL; |
c974c464 | 1521 | __del_reloc_root(reloc_root); |
5d4f98a2 YZ |
1522 | } |
1523 | ||
5d4f98a2 YZ |
1524 | if (reloc_root->commit_root != reloc_root->node) { |
1525 | btrfs_set_root_node(root_item, reloc_root->node); | |
1526 | free_extent_buffer(reloc_root->commit_root); | |
1527 | reloc_root->commit_root = btrfs_root_node(reloc_root); | |
1528 | } | |
1529 | ||
0b246afa | 1530 | ret = btrfs_update_root(trans, fs_info->tree_root, |
5d4f98a2 YZ |
1531 | &reloc_root->root_key, root_item); |
1532 | BUG_ON(ret); | |
7585717f CM |
1533 | |
1534 | out: | |
5d4f98a2 YZ |
1535 | return 0; |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * helper to find first cached inode with inode number >= objectid | |
1540 | * in a subvolume | |
1541 | */ | |
1542 | static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) | |
1543 | { | |
1544 | struct rb_node *node; | |
1545 | struct rb_node *prev; | |
1546 | struct btrfs_inode *entry; | |
1547 | struct inode *inode; | |
1548 | ||
1549 | spin_lock(&root->inode_lock); | |
1550 | again: | |
1551 | node = root->inode_tree.rb_node; | |
1552 | prev = NULL; | |
1553 | while (node) { | |
1554 | prev = node; | |
1555 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
1556 | ||
4a0cc7ca | 1557 | if (objectid < btrfs_ino(entry)) |
5d4f98a2 | 1558 | node = node->rb_left; |
4a0cc7ca | 1559 | else if (objectid > btrfs_ino(entry)) |
5d4f98a2 YZ |
1560 | node = node->rb_right; |
1561 | else | |
1562 | break; | |
1563 | } | |
1564 | if (!node) { | |
1565 | while (prev) { | |
1566 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 1567 | if (objectid <= btrfs_ino(entry)) { |
5d4f98a2 YZ |
1568 | node = prev; |
1569 | break; | |
1570 | } | |
1571 | prev = rb_next(prev); | |
1572 | } | |
1573 | } | |
1574 | while (node) { | |
1575 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
1576 | inode = igrab(&entry->vfs_inode); | |
1577 | if (inode) { | |
1578 | spin_unlock(&root->inode_lock); | |
1579 | return inode; | |
1580 | } | |
1581 | ||
4a0cc7ca | 1582 | objectid = btrfs_ino(entry) + 1; |
5d4f98a2 YZ |
1583 | if (cond_resched_lock(&root->inode_lock)) |
1584 | goto again; | |
1585 | ||
1586 | node = rb_next(node); | |
1587 | } | |
1588 | spin_unlock(&root->inode_lock); | |
1589 | return NULL; | |
1590 | } | |
1591 | ||
1592 | static int in_block_group(u64 bytenr, | |
1593 | struct btrfs_block_group_cache *block_group) | |
1594 | { | |
1595 | if (bytenr >= block_group->key.objectid && | |
1596 | bytenr < block_group->key.objectid + block_group->key.offset) | |
1597 | return 1; | |
1598 | return 0; | |
1599 | } | |
1600 | ||
1601 | /* | |
1602 | * get new location of data | |
1603 | */ | |
1604 | static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, | |
1605 | u64 bytenr, u64 num_bytes) | |
1606 | { | |
1607 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
1608 | struct btrfs_path *path; | |
1609 | struct btrfs_file_extent_item *fi; | |
1610 | struct extent_buffer *leaf; | |
1611 | int ret; | |
1612 | ||
1613 | path = btrfs_alloc_path(); | |
1614 | if (!path) | |
1615 | return -ENOMEM; | |
1616 | ||
1617 | bytenr -= BTRFS_I(reloc_inode)->index_cnt; | |
f85b7379 DS |
1618 | ret = btrfs_lookup_file_extent(NULL, root, path, |
1619 | btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); | |
5d4f98a2 YZ |
1620 | if (ret < 0) |
1621 | goto out; | |
1622 | if (ret > 0) { | |
1623 | ret = -ENOENT; | |
1624 | goto out; | |
1625 | } | |
1626 | ||
1627 | leaf = path->nodes[0]; | |
1628 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1629 | struct btrfs_file_extent_item); | |
1630 | ||
1631 | BUG_ON(btrfs_file_extent_offset(leaf, fi) || | |
1632 | btrfs_file_extent_compression(leaf, fi) || | |
1633 | btrfs_file_extent_encryption(leaf, fi) || | |
1634 | btrfs_file_extent_other_encoding(leaf, fi)); | |
1635 | ||
1636 | if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { | |
83d4cfd4 | 1637 | ret = -EINVAL; |
5d4f98a2 YZ |
1638 | goto out; |
1639 | } | |
1640 | ||
3fd0a558 | 1641 | *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 YZ |
1642 | ret = 0; |
1643 | out: | |
1644 | btrfs_free_path(path); | |
1645 | return ret; | |
1646 | } | |
1647 | ||
1648 | /* | |
1649 | * update file extent items in the tree leaf to point to | |
1650 | * the new locations. | |
1651 | */ | |
3fd0a558 YZ |
1652 | static noinline_for_stack |
1653 | int replace_file_extents(struct btrfs_trans_handle *trans, | |
1654 | struct reloc_control *rc, | |
1655 | struct btrfs_root *root, | |
1656 | struct extent_buffer *leaf) | |
5d4f98a2 | 1657 | { |
0b246afa | 1658 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1659 | struct btrfs_key key; |
1660 | struct btrfs_file_extent_item *fi; | |
1661 | struct inode *inode = NULL; | |
5d4f98a2 YZ |
1662 | u64 parent; |
1663 | u64 bytenr; | |
3fd0a558 | 1664 | u64 new_bytenr = 0; |
5d4f98a2 YZ |
1665 | u64 num_bytes; |
1666 | u64 end; | |
1667 | u32 nritems; | |
1668 | u32 i; | |
83d4cfd4 | 1669 | int ret = 0; |
5d4f98a2 YZ |
1670 | int first = 1; |
1671 | int dirty = 0; | |
1672 | ||
1673 | if (rc->stage != UPDATE_DATA_PTRS) | |
1674 | return 0; | |
1675 | ||
1676 | /* reloc trees always use full backref */ | |
1677 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1678 | parent = leaf->start; | |
1679 | else | |
1680 | parent = 0; | |
1681 | ||
1682 | nritems = btrfs_header_nritems(leaf); | |
1683 | for (i = 0; i < nritems; i++) { | |
1684 | cond_resched(); | |
1685 | btrfs_item_key_to_cpu(leaf, &key, i); | |
1686 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
1687 | continue; | |
1688 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
1689 | if (btrfs_file_extent_type(leaf, fi) == | |
1690 | BTRFS_FILE_EXTENT_INLINE) | |
1691 | continue; | |
1692 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1693 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
1694 | if (bytenr == 0) | |
1695 | continue; | |
1696 | if (!in_block_group(bytenr, rc->block_group)) | |
1697 | continue; | |
1698 | ||
1699 | /* | |
1700 | * if we are modifying block in fs tree, wait for readpage | |
1701 | * to complete and drop the extent cache | |
1702 | */ | |
1703 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { | |
5d4f98a2 YZ |
1704 | if (first) { |
1705 | inode = find_next_inode(root, key.objectid); | |
5d4f98a2 | 1706 | first = 0; |
4a0cc7ca | 1707 | } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { |
3fd0a558 | 1708 | btrfs_add_delayed_iput(inode); |
5d4f98a2 | 1709 | inode = find_next_inode(root, key.objectid); |
5d4f98a2 | 1710 | } |
4a0cc7ca | 1711 | if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { |
5d4f98a2 YZ |
1712 | end = key.offset + |
1713 | btrfs_file_extent_num_bytes(leaf, fi); | |
1714 | WARN_ON(!IS_ALIGNED(key.offset, | |
0b246afa JM |
1715 | fs_info->sectorsize)); |
1716 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); | |
5d4f98a2 YZ |
1717 | end--; |
1718 | ret = try_lock_extent(&BTRFS_I(inode)->io_tree, | |
d0082371 | 1719 | key.offset, end); |
5d4f98a2 YZ |
1720 | if (!ret) |
1721 | continue; | |
1722 | ||
dcdbc059 NB |
1723 | btrfs_drop_extent_cache(BTRFS_I(inode), |
1724 | key.offset, end, 1); | |
5d4f98a2 | 1725 | unlock_extent(&BTRFS_I(inode)->io_tree, |
d0082371 | 1726 | key.offset, end); |
5d4f98a2 YZ |
1727 | } |
1728 | } | |
1729 | ||
1730 | ret = get_new_location(rc->data_inode, &new_bytenr, | |
1731 | bytenr, num_bytes); | |
83d4cfd4 JB |
1732 | if (ret) { |
1733 | /* | |
1734 | * Don't have to abort since we've not changed anything | |
1735 | * in the file extent yet. | |
1736 | */ | |
1737 | break; | |
3fd0a558 | 1738 | } |
5d4f98a2 YZ |
1739 | |
1740 | btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); | |
1741 | dirty = 1; | |
1742 | ||
1743 | key.offset -= btrfs_file_extent_offset(leaf, fi); | |
2ff7e61e | 1744 | ret = btrfs_inc_extent_ref(trans, fs_info, new_bytenr, |
5d4f98a2 YZ |
1745 | num_bytes, parent, |
1746 | btrfs_header_owner(leaf), | |
b06c4bf5 | 1747 | key.objectid, key.offset); |
83d4cfd4 | 1748 | if (ret) { |
66642832 | 1749 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1750 | break; |
1751 | } | |
5d4f98a2 | 1752 | |
2ff7e61e | 1753 | ret = btrfs_free_extent(trans, fs_info, bytenr, num_bytes, |
5d4f98a2 | 1754 | parent, btrfs_header_owner(leaf), |
b06c4bf5 | 1755 | key.objectid, key.offset); |
83d4cfd4 | 1756 | if (ret) { |
66642832 | 1757 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1758 | break; |
1759 | } | |
5d4f98a2 YZ |
1760 | } |
1761 | if (dirty) | |
1762 | btrfs_mark_buffer_dirty(leaf); | |
3fd0a558 YZ |
1763 | if (inode) |
1764 | btrfs_add_delayed_iput(inode); | |
83d4cfd4 | 1765 | return ret; |
5d4f98a2 YZ |
1766 | } |
1767 | ||
1768 | static noinline_for_stack | |
1769 | int memcmp_node_keys(struct extent_buffer *eb, int slot, | |
1770 | struct btrfs_path *path, int level) | |
1771 | { | |
1772 | struct btrfs_disk_key key1; | |
1773 | struct btrfs_disk_key key2; | |
1774 | btrfs_node_key(eb, &key1, slot); | |
1775 | btrfs_node_key(path->nodes[level], &key2, path->slots[level]); | |
1776 | return memcmp(&key1, &key2, sizeof(key1)); | |
1777 | } | |
1778 | ||
1779 | /* | |
1780 | * try to replace tree blocks in fs tree with the new blocks | |
1781 | * in reloc tree. tree blocks haven't been modified since the | |
1782 | * reloc tree was create can be replaced. | |
1783 | * | |
1784 | * if a block was replaced, level of the block + 1 is returned. | |
1785 | * if no block got replaced, 0 is returned. if there are other | |
1786 | * errors, a negative error number is returned. | |
1787 | */ | |
3fd0a558 YZ |
1788 | static noinline_for_stack |
1789 | int replace_path(struct btrfs_trans_handle *trans, | |
1790 | struct btrfs_root *dest, struct btrfs_root *src, | |
1791 | struct btrfs_path *path, struct btrfs_key *next_key, | |
1792 | int lowest_level, int max_level) | |
5d4f98a2 | 1793 | { |
0b246afa | 1794 | struct btrfs_fs_info *fs_info = dest->fs_info; |
5d4f98a2 YZ |
1795 | struct extent_buffer *eb; |
1796 | struct extent_buffer *parent; | |
1797 | struct btrfs_key key; | |
1798 | u64 old_bytenr; | |
1799 | u64 new_bytenr; | |
1800 | u64 old_ptr_gen; | |
1801 | u64 new_ptr_gen; | |
1802 | u64 last_snapshot; | |
1803 | u32 blocksize; | |
3fd0a558 | 1804 | int cow = 0; |
5d4f98a2 YZ |
1805 | int level; |
1806 | int ret; | |
1807 | int slot; | |
1808 | ||
1809 | BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
1810 | BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); | |
5d4f98a2 YZ |
1811 | |
1812 | last_snapshot = btrfs_root_last_snapshot(&src->root_item); | |
3fd0a558 | 1813 | again: |
5d4f98a2 YZ |
1814 | slot = path->slots[lowest_level]; |
1815 | btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); | |
1816 | ||
1817 | eb = btrfs_lock_root_node(dest); | |
1818 | btrfs_set_lock_blocking(eb); | |
1819 | level = btrfs_header_level(eb); | |
1820 | ||
1821 | if (level < lowest_level) { | |
1822 | btrfs_tree_unlock(eb); | |
1823 | free_extent_buffer(eb); | |
1824 | return 0; | |
1825 | } | |
1826 | ||
3fd0a558 YZ |
1827 | if (cow) { |
1828 | ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); | |
1829 | BUG_ON(ret); | |
1830 | } | |
5d4f98a2 YZ |
1831 | btrfs_set_lock_blocking(eb); |
1832 | ||
1833 | if (next_key) { | |
1834 | next_key->objectid = (u64)-1; | |
1835 | next_key->type = (u8)-1; | |
1836 | next_key->offset = (u64)-1; | |
1837 | } | |
1838 | ||
1839 | parent = eb; | |
1840 | while (1) { | |
1841 | level = btrfs_header_level(parent); | |
1842 | BUG_ON(level < lowest_level); | |
1843 | ||
1844 | ret = btrfs_bin_search(parent, &key, level, &slot); | |
1845 | if (ret && slot > 0) | |
1846 | slot--; | |
1847 | ||
1848 | if (next_key && slot + 1 < btrfs_header_nritems(parent)) | |
1849 | btrfs_node_key_to_cpu(parent, next_key, slot + 1); | |
1850 | ||
1851 | old_bytenr = btrfs_node_blockptr(parent, slot); | |
0b246afa | 1852 | blocksize = fs_info->nodesize; |
5d4f98a2 YZ |
1853 | old_ptr_gen = btrfs_node_ptr_generation(parent, slot); |
1854 | ||
1855 | if (level <= max_level) { | |
1856 | eb = path->nodes[level]; | |
1857 | new_bytenr = btrfs_node_blockptr(eb, | |
1858 | path->slots[level]); | |
1859 | new_ptr_gen = btrfs_node_ptr_generation(eb, | |
1860 | path->slots[level]); | |
1861 | } else { | |
1862 | new_bytenr = 0; | |
1863 | new_ptr_gen = 0; | |
1864 | } | |
1865 | ||
fae7f21c | 1866 | if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { |
5d4f98a2 YZ |
1867 | ret = level; |
1868 | break; | |
1869 | } | |
1870 | ||
1871 | if (new_bytenr == 0 || old_ptr_gen > last_snapshot || | |
1872 | memcmp_node_keys(parent, slot, path, level)) { | |
3fd0a558 | 1873 | if (level <= lowest_level) { |
5d4f98a2 YZ |
1874 | ret = 0; |
1875 | break; | |
1876 | } | |
1877 | ||
2ff7e61e | 1878 | eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen); |
64c043de LB |
1879 | if (IS_ERR(eb)) { |
1880 | ret = PTR_ERR(eb); | |
264813ac | 1881 | break; |
64c043de LB |
1882 | } else if (!extent_buffer_uptodate(eb)) { |
1883 | ret = -EIO; | |
416bc658 | 1884 | free_extent_buffer(eb); |
379cde74 | 1885 | break; |
416bc658 | 1886 | } |
5d4f98a2 | 1887 | btrfs_tree_lock(eb); |
3fd0a558 YZ |
1888 | if (cow) { |
1889 | ret = btrfs_cow_block(trans, dest, eb, parent, | |
1890 | slot, &eb); | |
1891 | BUG_ON(ret); | |
5d4f98a2 | 1892 | } |
3fd0a558 | 1893 | btrfs_set_lock_blocking(eb); |
5d4f98a2 YZ |
1894 | |
1895 | btrfs_tree_unlock(parent); | |
1896 | free_extent_buffer(parent); | |
1897 | ||
1898 | parent = eb; | |
1899 | continue; | |
1900 | } | |
1901 | ||
3fd0a558 YZ |
1902 | if (!cow) { |
1903 | btrfs_tree_unlock(parent); | |
1904 | free_extent_buffer(parent); | |
1905 | cow = 1; | |
1906 | goto again; | |
1907 | } | |
1908 | ||
5d4f98a2 YZ |
1909 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
1910 | path->slots[level]); | |
b3b4aa74 | 1911 | btrfs_release_path(path); |
5d4f98a2 YZ |
1912 | |
1913 | path->lowest_level = level; | |
1914 | ret = btrfs_search_slot(trans, src, &key, path, 0, 1); | |
1915 | path->lowest_level = 0; | |
1916 | BUG_ON(ret); | |
1917 | ||
824d8dff QW |
1918 | /* |
1919 | * Info qgroup to trace both subtrees. | |
1920 | * | |
1921 | * We must trace both trees. | |
1922 | * 1) Tree reloc subtree | |
1923 | * If not traced, we will leak data numbers | |
1924 | * 2) Fs subtree | |
1925 | * If not traced, we will double count old data | |
1926 | * and tree block numbers, if current trans doesn't free | |
1927 | * data reloc tree inode. | |
1928 | */ | |
1929 | ret = btrfs_qgroup_trace_subtree(trans, src, parent, | |
1930 | btrfs_header_generation(parent), | |
1931 | btrfs_header_level(parent)); | |
1932 | if (ret < 0) | |
1933 | break; | |
1934 | ret = btrfs_qgroup_trace_subtree(trans, dest, | |
1935 | path->nodes[level], | |
1936 | btrfs_header_generation(path->nodes[level]), | |
1937 | btrfs_header_level(path->nodes[level])); | |
1938 | if (ret < 0) | |
1939 | break; | |
1940 | ||
5d4f98a2 YZ |
1941 | /* |
1942 | * swap blocks in fs tree and reloc tree. | |
1943 | */ | |
1944 | btrfs_set_node_blockptr(parent, slot, new_bytenr); | |
1945 | btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); | |
1946 | btrfs_mark_buffer_dirty(parent); | |
1947 | ||
1948 | btrfs_set_node_blockptr(path->nodes[level], | |
1949 | path->slots[level], old_bytenr); | |
1950 | btrfs_set_node_ptr_generation(path->nodes[level], | |
1951 | path->slots[level], old_ptr_gen); | |
1952 | btrfs_mark_buffer_dirty(path->nodes[level]); | |
1953 | ||
2ff7e61e JM |
1954 | ret = btrfs_inc_extent_ref(trans, fs_info, old_bytenr, |
1955 | blocksize, path->nodes[level]->start, | |
b06c4bf5 | 1956 | src->root_key.objectid, level - 1, 0); |
5d4f98a2 | 1957 | BUG_ON(ret); |
2ff7e61e JM |
1958 | ret = btrfs_inc_extent_ref(trans, fs_info, new_bytenr, |
1959 | blocksize, 0, dest->root_key.objectid, | |
1960 | level - 1, 0); | |
5d4f98a2 YZ |
1961 | BUG_ON(ret); |
1962 | ||
2ff7e61e | 1963 | ret = btrfs_free_extent(trans, fs_info, new_bytenr, blocksize, |
5d4f98a2 | 1964 | path->nodes[level]->start, |
b06c4bf5 | 1965 | src->root_key.objectid, level - 1, 0); |
5d4f98a2 YZ |
1966 | BUG_ON(ret); |
1967 | ||
2ff7e61e | 1968 | ret = btrfs_free_extent(trans, fs_info, old_bytenr, blocksize, |
5d4f98a2 | 1969 | 0, dest->root_key.objectid, level - 1, |
b06c4bf5 | 1970 | 0); |
5d4f98a2 YZ |
1971 | BUG_ON(ret); |
1972 | ||
1973 | btrfs_unlock_up_safe(path, 0); | |
1974 | ||
1975 | ret = level; | |
1976 | break; | |
1977 | } | |
1978 | btrfs_tree_unlock(parent); | |
1979 | free_extent_buffer(parent); | |
1980 | return ret; | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * helper to find next relocated block in reloc tree | |
1985 | */ | |
1986 | static noinline_for_stack | |
1987 | int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
1988 | int *level) | |
1989 | { | |
1990 | struct extent_buffer *eb; | |
1991 | int i; | |
1992 | u64 last_snapshot; | |
1993 | u32 nritems; | |
1994 | ||
1995 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
1996 | ||
1997 | for (i = 0; i < *level; i++) { | |
1998 | free_extent_buffer(path->nodes[i]); | |
1999 | path->nodes[i] = NULL; | |
2000 | } | |
2001 | ||
2002 | for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { | |
2003 | eb = path->nodes[i]; | |
2004 | nritems = btrfs_header_nritems(eb); | |
2005 | while (path->slots[i] + 1 < nritems) { | |
2006 | path->slots[i]++; | |
2007 | if (btrfs_node_ptr_generation(eb, path->slots[i]) <= | |
2008 | last_snapshot) | |
2009 | continue; | |
2010 | ||
2011 | *level = i; | |
2012 | return 0; | |
2013 | } | |
2014 | free_extent_buffer(path->nodes[i]); | |
2015 | path->nodes[i] = NULL; | |
2016 | } | |
2017 | return 1; | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * walk down reloc tree to find relocated block of lowest level | |
2022 | */ | |
2023 | static noinline_for_stack | |
2024 | int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
2025 | int *level) | |
2026 | { | |
2ff7e61e | 2027 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2028 | struct extent_buffer *eb = NULL; |
2029 | int i; | |
2030 | u64 bytenr; | |
2031 | u64 ptr_gen = 0; | |
2032 | u64 last_snapshot; | |
5d4f98a2 YZ |
2033 | u32 nritems; |
2034 | ||
2035 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
2036 | ||
2037 | for (i = *level; i > 0; i--) { | |
2038 | eb = path->nodes[i]; | |
2039 | nritems = btrfs_header_nritems(eb); | |
2040 | while (path->slots[i] < nritems) { | |
2041 | ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); | |
2042 | if (ptr_gen > last_snapshot) | |
2043 | break; | |
2044 | path->slots[i]++; | |
2045 | } | |
2046 | if (path->slots[i] >= nritems) { | |
2047 | if (i == *level) | |
2048 | break; | |
2049 | *level = i + 1; | |
2050 | return 0; | |
2051 | } | |
2052 | if (i == 1) { | |
2053 | *level = i; | |
2054 | return 0; | |
2055 | } | |
2056 | ||
2057 | bytenr = btrfs_node_blockptr(eb, path->slots[i]); | |
2ff7e61e | 2058 | eb = read_tree_block(fs_info, bytenr, ptr_gen); |
64c043de LB |
2059 | if (IS_ERR(eb)) { |
2060 | return PTR_ERR(eb); | |
2061 | } else if (!extent_buffer_uptodate(eb)) { | |
416bc658 JB |
2062 | free_extent_buffer(eb); |
2063 | return -EIO; | |
2064 | } | |
5d4f98a2 YZ |
2065 | BUG_ON(btrfs_header_level(eb) != i - 1); |
2066 | path->nodes[i - 1] = eb; | |
2067 | path->slots[i - 1] = 0; | |
2068 | } | |
2069 | return 1; | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * invalidate extent cache for file extents whose key in range of | |
2074 | * [min_key, max_key) | |
2075 | */ | |
2076 | static int invalidate_extent_cache(struct btrfs_root *root, | |
2077 | struct btrfs_key *min_key, | |
2078 | struct btrfs_key *max_key) | |
2079 | { | |
0b246afa | 2080 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2081 | struct inode *inode = NULL; |
2082 | u64 objectid; | |
2083 | u64 start, end; | |
33345d01 | 2084 | u64 ino; |
5d4f98a2 YZ |
2085 | |
2086 | objectid = min_key->objectid; | |
2087 | while (1) { | |
2088 | cond_resched(); | |
2089 | iput(inode); | |
2090 | ||
2091 | if (objectid > max_key->objectid) | |
2092 | break; | |
2093 | ||
2094 | inode = find_next_inode(root, objectid); | |
2095 | if (!inode) | |
2096 | break; | |
4a0cc7ca | 2097 | ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 2098 | |
33345d01 | 2099 | if (ino > max_key->objectid) { |
5d4f98a2 YZ |
2100 | iput(inode); |
2101 | break; | |
2102 | } | |
2103 | ||
33345d01 | 2104 | objectid = ino + 1; |
5d4f98a2 YZ |
2105 | if (!S_ISREG(inode->i_mode)) |
2106 | continue; | |
2107 | ||
33345d01 | 2108 | if (unlikely(min_key->objectid == ino)) { |
5d4f98a2 YZ |
2109 | if (min_key->type > BTRFS_EXTENT_DATA_KEY) |
2110 | continue; | |
2111 | if (min_key->type < BTRFS_EXTENT_DATA_KEY) | |
2112 | start = 0; | |
2113 | else { | |
2114 | start = min_key->offset; | |
0b246afa | 2115 | WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); |
5d4f98a2 YZ |
2116 | } |
2117 | } else { | |
2118 | start = 0; | |
2119 | } | |
2120 | ||
33345d01 | 2121 | if (unlikely(max_key->objectid == ino)) { |
5d4f98a2 YZ |
2122 | if (max_key->type < BTRFS_EXTENT_DATA_KEY) |
2123 | continue; | |
2124 | if (max_key->type > BTRFS_EXTENT_DATA_KEY) { | |
2125 | end = (u64)-1; | |
2126 | } else { | |
2127 | if (max_key->offset == 0) | |
2128 | continue; | |
2129 | end = max_key->offset; | |
0b246afa | 2130 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); |
5d4f98a2 YZ |
2131 | end--; |
2132 | } | |
2133 | } else { | |
2134 | end = (u64)-1; | |
2135 | } | |
2136 | ||
2137 | /* the lock_extent waits for readpage to complete */ | |
d0082371 | 2138 | lock_extent(&BTRFS_I(inode)->io_tree, start, end); |
dcdbc059 | 2139 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); |
d0082371 | 2140 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end); |
5d4f98a2 YZ |
2141 | } |
2142 | return 0; | |
2143 | } | |
2144 | ||
2145 | static int find_next_key(struct btrfs_path *path, int level, | |
2146 | struct btrfs_key *key) | |
2147 | ||
2148 | { | |
2149 | while (level < BTRFS_MAX_LEVEL) { | |
2150 | if (!path->nodes[level]) | |
2151 | break; | |
2152 | if (path->slots[level] + 1 < | |
2153 | btrfs_header_nritems(path->nodes[level])) { | |
2154 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
2155 | path->slots[level] + 1); | |
2156 | return 0; | |
2157 | } | |
2158 | level++; | |
2159 | } | |
2160 | return 1; | |
2161 | } | |
2162 | ||
2163 | /* | |
2164 | * merge the relocated tree blocks in reloc tree with corresponding | |
2165 | * fs tree. | |
2166 | */ | |
2167 | static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, | |
2168 | struct btrfs_root *root) | |
2169 | { | |
0b246afa | 2170 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
2171 | LIST_HEAD(inode_list); |
2172 | struct btrfs_key key; | |
2173 | struct btrfs_key next_key; | |
9e6a0c52 | 2174 | struct btrfs_trans_handle *trans = NULL; |
5d4f98a2 YZ |
2175 | struct btrfs_root *reloc_root; |
2176 | struct btrfs_root_item *root_item; | |
2177 | struct btrfs_path *path; | |
3fd0a558 | 2178 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2179 | int level; |
2180 | int max_level; | |
2181 | int replaced = 0; | |
2182 | int ret; | |
2183 | int err = 0; | |
3fd0a558 | 2184 | u32 min_reserved; |
5d4f98a2 YZ |
2185 | |
2186 | path = btrfs_alloc_path(); | |
2187 | if (!path) | |
2188 | return -ENOMEM; | |
e4058b54 | 2189 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2190 | |
2191 | reloc_root = root->reloc_root; | |
2192 | root_item = &reloc_root->root_item; | |
2193 | ||
2194 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { | |
2195 | level = btrfs_root_level(root_item); | |
2196 | extent_buffer_get(reloc_root->node); | |
2197 | path->nodes[level] = reloc_root->node; | |
2198 | path->slots[level] = 0; | |
2199 | } else { | |
2200 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); | |
2201 | ||
2202 | level = root_item->drop_level; | |
2203 | BUG_ON(level == 0); | |
2204 | path->lowest_level = level; | |
2205 | ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); | |
33c66f43 | 2206 | path->lowest_level = 0; |
5d4f98a2 YZ |
2207 | if (ret < 0) { |
2208 | btrfs_free_path(path); | |
2209 | return ret; | |
2210 | } | |
2211 | ||
2212 | btrfs_node_key_to_cpu(path->nodes[level], &next_key, | |
2213 | path->slots[level]); | |
2214 | WARN_ON(memcmp(&key, &next_key, sizeof(key))); | |
2215 | ||
2216 | btrfs_unlock_up_safe(path, 0); | |
2217 | } | |
2218 | ||
0b246afa | 2219 | min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; |
3fd0a558 | 2220 | memset(&next_key, 0, sizeof(next_key)); |
5d4f98a2 | 2221 | |
3fd0a558 | 2222 | while (1) { |
08e007d2 MX |
2223 | ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, |
2224 | BTRFS_RESERVE_FLUSH_ALL); | |
3fd0a558 | 2225 | if (ret) { |
9e6a0c52 JB |
2226 | err = ret; |
2227 | goto out; | |
5d4f98a2 | 2228 | } |
9e6a0c52 JB |
2229 | trans = btrfs_start_transaction(root, 0); |
2230 | if (IS_ERR(trans)) { | |
2231 | err = PTR_ERR(trans); | |
2232 | trans = NULL; | |
2233 | goto out; | |
2234 | } | |
2235 | trans->block_rsv = rc->block_rsv; | |
5d4f98a2 | 2236 | |
5d4f98a2 | 2237 | replaced = 0; |
5d4f98a2 YZ |
2238 | max_level = level; |
2239 | ||
2240 | ret = walk_down_reloc_tree(reloc_root, path, &level); | |
2241 | if (ret < 0) { | |
2242 | err = ret; | |
2243 | goto out; | |
2244 | } | |
2245 | if (ret > 0) | |
2246 | break; | |
2247 | ||
2248 | if (!find_next_key(path, level, &key) && | |
2249 | btrfs_comp_cpu_keys(&next_key, &key) >= 0) { | |
2250 | ret = 0; | |
5d4f98a2 | 2251 | } else { |
3fd0a558 YZ |
2252 | ret = replace_path(trans, root, reloc_root, path, |
2253 | &next_key, level, max_level); | |
5d4f98a2 YZ |
2254 | } |
2255 | if (ret < 0) { | |
2256 | err = ret; | |
2257 | goto out; | |
2258 | } | |
2259 | ||
2260 | if (ret > 0) { | |
2261 | level = ret; | |
2262 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
2263 | path->slots[level]); | |
2264 | replaced = 1; | |
5d4f98a2 YZ |
2265 | } |
2266 | ||
2267 | ret = walk_up_reloc_tree(reloc_root, path, &level); | |
2268 | if (ret > 0) | |
2269 | break; | |
2270 | ||
2271 | BUG_ON(level == 0); | |
2272 | /* | |
2273 | * save the merging progress in the drop_progress. | |
2274 | * this is OK since root refs == 1 in this case. | |
2275 | */ | |
2276 | btrfs_node_key(path->nodes[level], &root_item->drop_progress, | |
2277 | path->slots[level]); | |
2278 | root_item->drop_level = level; | |
2279 | ||
3a45bb20 | 2280 | btrfs_end_transaction_throttle(trans); |
9e6a0c52 | 2281 | trans = NULL; |
5d4f98a2 | 2282 | |
2ff7e61e | 2283 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
2284 | |
2285 | if (replaced && rc->stage == UPDATE_DATA_PTRS) | |
2286 | invalidate_extent_cache(root, &key, &next_key); | |
2287 | } | |
2288 | ||
2289 | /* | |
2290 | * handle the case only one block in the fs tree need to be | |
2291 | * relocated and the block is tree root. | |
2292 | */ | |
2293 | leaf = btrfs_lock_root_node(root); | |
2294 | ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); | |
2295 | btrfs_tree_unlock(leaf); | |
2296 | free_extent_buffer(leaf); | |
2297 | if (ret < 0) | |
2298 | err = ret; | |
2299 | out: | |
2300 | btrfs_free_path(path); | |
2301 | ||
2302 | if (err == 0) { | |
2303 | memset(&root_item->drop_progress, 0, | |
2304 | sizeof(root_item->drop_progress)); | |
2305 | root_item->drop_level = 0; | |
2306 | btrfs_set_root_refs(root_item, 0); | |
3fd0a558 | 2307 | btrfs_update_reloc_root(trans, root); |
5d4f98a2 YZ |
2308 | } |
2309 | ||
9e6a0c52 | 2310 | if (trans) |
3a45bb20 | 2311 | btrfs_end_transaction_throttle(trans); |
5d4f98a2 | 2312 | |
2ff7e61e | 2313 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 2314 | |
5d4f98a2 YZ |
2315 | if (replaced && rc->stage == UPDATE_DATA_PTRS) |
2316 | invalidate_extent_cache(root, &key, &next_key); | |
2317 | ||
2318 | return err; | |
2319 | } | |
2320 | ||
3fd0a558 YZ |
2321 | static noinline_for_stack |
2322 | int prepare_to_merge(struct reloc_control *rc, int err) | |
5d4f98a2 | 2323 | { |
3fd0a558 | 2324 | struct btrfs_root *root = rc->extent_root; |
0b246afa | 2325 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 2326 | struct btrfs_root *reloc_root; |
3fd0a558 YZ |
2327 | struct btrfs_trans_handle *trans; |
2328 | LIST_HEAD(reloc_roots); | |
2329 | u64 num_bytes = 0; | |
2330 | int ret; | |
3fd0a558 | 2331 | |
0b246afa JM |
2332 | mutex_lock(&fs_info->reloc_mutex); |
2333 | rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; | |
3fd0a558 | 2334 | rc->merging_rsv_size += rc->nodes_relocated * 2; |
0b246afa | 2335 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 2336 | |
3fd0a558 YZ |
2337 | again: |
2338 | if (!err) { | |
2339 | num_bytes = rc->merging_rsv_size; | |
08e007d2 MX |
2340 | ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, |
2341 | BTRFS_RESERVE_FLUSH_ALL); | |
3fd0a558 YZ |
2342 | if (ret) |
2343 | err = ret; | |
2344 | } | |
2345 | ||
7a7eaa40 | 2346 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 TI |
2347 | if (IS_ERR(trans)) { |
2348 | if (!err) | |
2ff7e61e JM |
2349 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
2350 | num_bytes); | |
3612b495 TI |
2351 | return PTR_ERR(trans); |
2352 | } | |
3fd0a558 YZ |
2353 | |
2354 | if (!err) { | |
2355 | if (num_bytes != rc->merging_rsv_size) { | |
3a45bb20 | 2356 | btrfs_end_transaction(trans); |
2ff7e61e JM |
2357 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
2358 | num_bytes); | |
3fd0a558 YZ |
2359 | goto again; |
2360 | } | |
2361 | } | |
5d4f98a2 | 2362 | |
3fd0a558 YZ |
2363 | rc->merge_reloc_tree = 1; |
2364 | ||
2365 | while (!list_empty(&rc->reloc_roots)) { | |
2366 | reloc_root = list_entry(rc->reloc_roots.next, | |
2367 | struct btrfs_root, root_list); | |
2368 | list_del_init(&reloc_root->root_list); | |
5d4f98a2 | 2369 | |
0b246afa | 2370 | root = read_fs_root(fs_info, reloc_root->root_key.offset); |
5d4f98a2 YZ |
2371 | BUG_ON(IS_ERR(root)); |
2372 | BUG_ON(root->reloc_root != reloc_root); | |
2373 | ||
3fd0a558 YZ |
2374 | /* |
2375 | * set reference count to 1, so btrfs_recover_relocation | |
2376 | * knows it should resumes merging | |
2377 | */ | |
2378 | if (!err) | |
2379 | btrfs_set_root_refs(&reloc_root->root_item, 1); | |
5d4f98a2 | 2380 | btrfs_update_reloc_root(trans, root); |
5d4f98a2 | 2381 | |
3fd0a558 YZ |
2382 | list_add(&reloc_root->root_list, &reloc_roots); |
2383 | } | |
5d4f98a2 | 2384 | |
3fd0a558 | 2385 | list_splice(&reloc_roots, &rc->reloc_roots); |
5d4f98a2 | 2386 | |
3fd0a558 | 2387 | if (!err) |
3a45bb20 | 2388 | btrfs_commit_transaction(trans); |
3fd0a558 | 2389 | else |
3a45bb20 | 2390 | btrfs_end_transaction(trans); |
3fd0a558 | 2391 | return err; |
5d4f98a2 YZ |
2392 | } |
2393 | ||
aca1bba6 LB |
2394 | static noinline_for_stack |
2395 | void free_reloc_roots(struct list_head *list) | |
2396 | { | |
2397 | struct btrfs_root *reloc_root; | |
2398 | ||
2399 | while (!list_empty(list)) { | |
2400 | reloc_root = list_entry(list->next, struct btrfs_root, | |
2401 | root_list); | |
6bdf131f JB |
2402 | free_extent_buffer(reloc_root->node); |
2403 | free_extent_buffer(reloc_root->commit_root); | |
2404 | reloc_root->node = NULL; | |
2405 | reloc_root->commit_root = NULL; | |
c974c464 | 2406 | __del_reloc_root(reloc_root); |
aca1bba6 LB |
2407 | } |
2408 | } | |
2409 | ||
3fd0a558 | 2410 | static noinline_for_stack |
94404e82 | 2411 | void merge_reloc_roots(struct reloc_control *rc) |
5d4f98a2 | 2412 | { |
0b246afa | 2413 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 | 2414 | struct btrfs_root *root; |
3fd0a558 YZ |
2415 | struct btrfs_root *reloc_root; |
2416 | LIST_HEAD(reloc_roots); | |
2417 | int found = 0; | |
aca1bba6 | 2418 | int ret = 0; |
3fd0a558 YZ |
2419 | again: |
2420 | root = rc->extent_root; | |
7585717f CM |
2421 | |
2422 | /* | |
2423 | * this serializes us with btrfs_record_root_in_transaction, | |
2424 | * we have to make sure nobody is in the middle of | |
2425 | * adding their roots to the list while we are | |
2426 | * doing this splice | |
2427 | */ | |
0b246afa | 2428 | mutex_lock(&fs_info->reloc_mutex); |
3fd0a558 | 2429 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 2430 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 | 2431 | |
3fd0a558 YZ |
2432 | while (!list_empty(&reloc_roots)) { |
2433 | found = 1; | |
2434 | reloc_root = list_entry(reloc_roots.next, | |
2435 | struct btrfs_root, root_list); | |
5d4f98a2 | 2436 | |
3fd0a558 | 2437 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { |
0b246afa | 2438 | root = read_fs_root(fs_info, |
3fd0a558 YZ |
2439 | reloc_root->root_key.offset); |
2440 | BUG_ON(IS_ERR(root)); | |
2441 | BUG_ON(root->reloc_root != reloc_root); | |
5d4f98a2 | 2442 | |
3fd0a558 | 2443 | ret = merge_reloc_root(rc, root); |
b37b39cd | 2444 | if (ret) { |
25e293c2 WS |
2445 | if (list_empty(&reloc_root->root_list)) |
2446 | list_add_tail(&reloc_root->root_list, | |
2447 | &reloc_roots); | |
aca1bba6 | 2448 | goto out; |
b37b39cd | 2449 | } |
3fd0a558 YZ |
2450 | } else { |
2451 | list_del_init(&reloc_root->root_list); | |
2452 | } | |
5bc7247a | 2453 | |
2c536799 | 2454 | ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); |
aca1bba6 LB |
2455 | if (ret < 0) { |
2456 | if (list_empty(&reloc_root->root_list)) | |
2457 | list_add_tail(&reloc_root->root_list, | |
2458 | &reloc_roots); | |
2459 | goto out; | |
2460 | } | |
5d4f98a2 YZ |
2461 | } |
2462 | ||
3fd0a558 YZ |
2463 | if (found) { |
2464 | found = 0; | |
2465 | goto again; | |
2466 | } | |
aca1bba6 LB |
2467 | out: |
2468 | if (ret) { | |
0b246afa | 2469 | btrfs_handle_fs_error(fs_info, ret, NULL); |
aca1bba6 LB |
2470 | if (!list_empty(&reloc_roots)) |
2471 | free_reloc_roots(&reloc_roots); | |
467bb1d2 WS |
2472 | |
2473 | /* new reloc root may be added */ | |
0b246afa | 2474 | mutex_lock(&fs_info->reloc_mutex); |
467bb1d2 | 2475 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 2476 | mutex_unlock(&fs_info->reloc_mutex); |
467bb1d2 WS |
2477 | if (!list_empty(&reloc_roots)) |
2478 | free_reloc_roots(&reloc_roots); | |
aca1bba6 LB |
2479 | } |
2480 | ||
5d4f98a2 | 2481 | BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); |
5d4f98a2 YZ |
2482 | } |
2483 | ||
2484 | static void free_block_list(struct rb_root *blocks) | |
2485 | { | |
2486 | struct tree_block *block; | |
2487 | struct rb_node *rb_node; | |
2488 | while ((rb_node = rb_first(blocks))) { | |
2489 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
2490 | rb_erase(rb_node, blocks); | |
2491 | kfree(block); | |
2492 | } | |
2493 | } | |
2494 | ||
2495 | static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, | |
2496 | struct btrfs_root *reloc_root) | |
2497 | { | |
0b246afa | 2498 | struct btrfs_fs_info *fs_info = reloc_root->fs_info; |
5d4f98a2 YZ |
2499 | struct btrfs_root *root; |
2500 | ||
2501 | if (reloc_root->last_trans == trans->transid) | |
2502 | return 0; | |
2503 | ||
0b246afa | 2504 | root = read_fs_root(fs_info, reloc_root->root_key.offset); |
5d4f98a2 YZ |
2505 | BUG_ON(IS_ERR(root)); |
2506 | BUG_ON(root->reloc_root != reloc_root); | |
2507 | ||
2508 | return btrfs_record_root_in_trans(trans, root); | |
2509 | } | |
2510 | ||
3fd0a558 YZ |
2511 | static noinline_for_stack |
2512 | struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, | |
2513 | struct reloc_control *rc, | |
2514 | struct backref_node *node, | |
dc4103f9 | 2515 | struct backref_edge *edges[]) |
5d4f98a2 YZ |
2516 | { |
2517 | struct backref_node *next; | |
2518 | struct btrfs_root *root; | |
3fd0a558 YZ |
2519 | int index = 0; |
2520 | ||
5d4f98a2 YZ |
2521 | next = node; |
2522 | while (1) { | |
2523 | cond_resched(); | |
2524 | next = walk_up_backref(next, edges, &index); | |
2525 | root = next->root; | |
3fd0a558 | 2526 | BUG_ON(!root); |
27cdeb70 | 2527 | BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); |
5d4f98a2 YZ |
2528 | |
2529 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
2530 | record_reloc_root_in_trans(trans, root); | |
2531 | break; | |
2532 | } | |
2533 | ||
3fd0a558 YZ |
2534 | btrfs_record_root_in_trans(trans, root); |
2535 | root = root->reloc_root; | |
2536 | ||
2537 | if (next->new_bytenr != root->node->start) { | |
2538 | BUG_ON(next->new_bytenr); | |
2539 | BUG_ON(!list_empty(&next->list)); | |
2540 | next->new_bytenr = root->node->start; | |
2541 | next->root = root; | |
2542 | list_add_tail(&next->list, | |
2543 | &rc->backref_cache.changed); | |
2544 | __mark_block_processed(rc, next); | |
5d4f98a2 YZ |
2545 | break; |
2546 | } | |
2547 | ||
3fd0a558 | 2548 | WARN_ON(1); |
5d4f98a2 YZ |
2549 | root = NULL; |
2550 | next = walk_down_backref(edges, &index); | |
2551 | if (!next || next->level <= node->level) | |
2552 | break; | |
2553 | } | |
3fd0a558 YZ |
2554 | if (!root) |
2555 | return NULL; | |
5d4f98a2 | 2556 | |
3fd0a558 YZ |
2557 | next = node; |
2558 | /* setup backref node path for btrfs_reloc_cow_block */ | |
2559 | while (1) { | |
2560 | rc->backref_cache.path[next->level] = next; | |
2561 | if (--index < 0) | |
2562 | break; | |
2563 | next = edges[index]->node[UPPER]; | |
5d4f98a2 | 2564 | } |
5d4f98a2 YZ |
2565 | return root; |
2566 | } | |
2567 | ||
3fd0a558 YZ |
2568 | /* |
2569 | * select a tree root for relocation. return NULL if the block | |
2570 | * is reference counted. we should use do_relocation() in this | |
2571 | * case. return a tree root pointer if the block isn't reference | |
2572 | * counted. return -ENOENT if the block is root of reloc tree. | |
2573 | */ | |
5d4f98a2 | 2574 | static noinline_for_stack |
147d256e | 2575 | struct btrfs_root *select_one_root(struct backref_node *node) |
5d4f98a2 | 2576 | { |
3fd0a558 YZ |
2577 | struct backref_node *next; |
2578 | struct btrfs_root *root; | |
2579 | struct btrfs_root *fs_root = NULL; | |
5d4f98a2 | 2580 | struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; |
3fd0a558 YZ |
2581 | int index = 0; |
2582 | ||
2583 | next = node; | |
2584 | while (1) { | |
2585 | cond_resched(); | |
2586 | next = walk_up_backref(next, edges, &index); | |
2587 | root = next->root; | |
2588 | BUG_ON(!root); | |
2589 | ||
25985edc | 2590 | /* no other choice for non-references counted tree */ |
27cdeb70 | 2591 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
3fd0a558 YZ |
2592 | return root; |
2593 | ||
2594 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) | |
2595 | fs_root = root; | |
2596 | ||
2597 | if (next != node) | |
2598 | return NULL; | |
2599 | ||
2600 | next = walk_down_backref(edges, &index); | |
2601 | if (!next || next->level <= node->level) | |
2602 | break; | |
2603 | } | |
2604 | ||
2605 | if (!fs_root) | |
2606 | return ERR_PTR(-ENOENT); | |
2607 | return fs_root; | |
5d4f98a2 YZ |
2608 | } |
2609 | ||
2610 | static noinline_for_stack | |
3fd0a558 YZ |
2611 | u64 calcu_metadata_size(struct reloc_control *rc, |
2612 | struct backref_node *node, int reserve) | |
5d4f98a2 | 2613 | { |
0b246afa | 2614 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
3fd0a558 YZ |
2615 | struct backref_node *next = node; |
2616 | struct backref_edge *edge; | |
2617 | struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
2618 | u64 num_bytes = 0; | |
2619 | int index = 0; | |
2620 | ||
2621 | BUG_ON(reserve && node->processed); | |
2622 | ||
2623 | while (next) { | |
2624 | cond_resched(); | |
2625 | while (1) { | |
2626 | if (next->processed && (reserve || next != node)) | |
2627 | break; | |
2628 | ||
0b246afa | 2629 | num_bytes += fs_info->nodesize; |
3fd0a558 YZ |
2630 | |
2631 | if (list_empty(&next->upper)) | |
2632 | break; | |
2633 | ||
2634 | edge = list_entry(next->upper.next, | |
2635 | struct backref_edge, list[LOWER]); | |
2636 | edges[index++] = edge; | |
2637 | next = edge->node[UPPER]; | |
2638 | } | |
2639 | next = walk_down_backref(edges, &index); | |
2640 | } | |
2641 | return num_bytes; | |
5d4f98a2 YZ |
2642 | } |
2643 | ||
3fd0a558 YZ |
2644 | static int reserve_metadata_space(struct btrfs_trans_handle *trans, |
2645 | struct reloc_control *rc, | |
2646 | struct backref_node *node) | |
5d4f98a2 | 2647 | { |
3fd0a558 | 2648 | struct btrfs_root *root = rc->extent_root; |
da17066c | 2649 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 YZ |
2650 | u64 num_bytes; |
2651 | int ret; | |
0647bf56 | 2652 | u64 tmp; |
3fd0a558 YZ |
2653 | |
2654 | num_bytes = calcu_metadata_size(rc, node, 1) * 2; | |
5d4f98a2 | 2655 | |
3fd0a558 | 2656 | trans->block_rsv = rc->block_rsv; |
0647bf56 | 2657 | rc->reserved_bytes += num_bytes; |
8ca17f0f JB |
2658 | |
2659 | /* | |
2660 | * We are under a transaction here so we can only do limited flushing. | |
2661 | * If we get an enospc just kick back -EAGAIN so we know to drop the | |
2662 | * transaction and try to refill when we can flush all the things. | |
2663 | */ | |
0647bf56 | 2664 | ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, |
8ca17f0f | 2665 | BTRFS_RESERVE_FLUSH_LIMIT); |
3fd0a558 | 2666 | if (ret) { |
da17066c | 2667 | tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; |
8ca17f0f JB |
2668 | while (tmp <= rc->reserved_bytes) |
2669 | tmp <<= 1; | |
2670 | /* | |
2671 | * only one thread can access block_rsv at this point, | |
2672 | * so we don't need hold lock to protect block_rsv. | |
2673 | * we expand more reservation size here to allow enough | |
2674 | * space for relocation and we will return eailer in | |
2675 | * enospc case. | |
2676 | */ | |
da17066c JM |
2677 | rc->block_rsv->size = tmp + fs_info->nodesize * |
2678 | RELOCATION_RESERVED_NODES; | |
8ca17f0f | 2679 | return -EAGAIN; |
5d4f98a2 | 2680 | } |
3fd0a558 | 2681 | |
3fd0a558 YZ |
2682 | return 0; |
2683 | } | |
2684 | ||
5d4f98a2 YZ |
2685 | /* |
2686 | * relocate a block tree, and then update pointers in upper level | |
2687 | * blocks that reference the block to point to the new location. | |
2688 | * | |
2689 | * if called by link_to_upper, the block has already been relocated. | |
2690 | * in that case this function just updates pointers. | |
2691 | */ | |
2692 | static int do_relocation(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2693 | struct reloc_control *rc, |
5d4f98a2 YZ |
2694 | struct backref_node *node, |
2695 | struct btrfs_key *key, | |
2696 | struct btrfs_path *path, int lowest) | |
2697 | { | |
2ff7e61e | 2698 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
2699 | struct backref_node *upper; |
2700 | struct backref_edge *edge; | |
2701 | struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
2702 | struct btrfs_root *root; | |
2703 | struct extent_buffer *eb; | |
2704 | u32 blocksize; | |
2705 | u64 bytenr; | |
2706 | u64 generation; | |
5d4f98a2 YZ |
2707 | int slot; |
2708 | int ret; | |
2709 | int err = 0; | |
2710 | ||
2711 | BUG_ON(lowest && node->eb); | |
2712 | ||
2713 | path->lowest_level = node->level + 1; | |
3fd0a558 | 2714 | rc->backref_cache.path[node->level] = node; |
5d4f98a2 YZ |
2715 | list_for_each_entry(edge, &node->upper, list[LOWER]) { |
2716 | cond_resched(); | |
5d4f98a2 YZ |
2717 | |
2718 | upper = edge->node[UPPER]; | |
dc4103f9 | 2719 | root = select_reloc_root(trans, rc, upper, edges); |
3fd0a558 YZ |
2720 | BUG_ON(!root); |
2721 | ||
2722 | if (upper->eb && !upper->locked) { | |
2723 | if (!lowest) { | |
2724 | ret = btrfs_bin_search(upper->eb, key, | |
2725 | upper->level, &slot); | |
2726 | BUG_ON(ret); | |
2727 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
2728 | if (node->eb->start == bytenr) | |
2729 | goto next; | |
2730 | } | |
5d4f98a2 | 2731 | drop_node_buffer(upper); |
3fd0a558 | 2732 | } |
5d4f98a2 YZ |
2733 | |
2734 | if (!upper->eb) { | |
2735 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); | |
3561b9db LB |
2736 | if (ret) { |
2737 | if (ret < 0) | |
2738 | err = ret; | |
2739 | else | |
2740 | err = -ENOENT; | |
2741 | ||
2742 | btrfs_release_path(path); | |
5d4f98a2 YZ |
2743 | break; |
2744 | } | |
5d4f98a2 | 2745 | |
3fd0a558 YZ |
2746 | if (!upper->eb) { |
2747 | upper->eb = path->nodes[upper->level]; | |
2748 | path->nodes[upper->level] = NULL; | |
2749 | } else { | |
2750 | BUG_ON(upper->eb != path->nodes[upper->level]); | |
2751 | } | |
5d4f98a2 | 2752 | |
3fd0a558 YZ |
2753 | upper->locked = 1; |
2754 | path->locks[upper->level] = 0; | |
5d4f98a2 | 2755 | |
3fd0a558 | 2756 | slot = path->slots[upper->level]; |
b3b4aa74 | 2757 | btrfs_release_path(path); |
5d4f98a2 YZ |
2758 | } else { |
2759 | ret = btrfs_bin_search(upper->eb, key, upper->level, | |
2760 | &slot); | |
2761 | BUG_ON(ret); | |
2762 | } | |
2763 | ||
2764 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
3fd0a558 | 2765 | if (lowest) { |
4547f4d8 LB |
2766 | if (bytenr != node->bytenr) { |
2767 | btrfs_err(root->fs_info, | |
2768 | "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", | |
2769 | bytenr, node->bytenr, slot, | |
2770 | upper->eb->start); | |
2771 | err = -EIO; | |
2772 | goto next; | |
2773 | } | |
5d4f98a2 | 2774 | } else { |
3fd0a558 YZ |
2775 | if (node->eb->start == bytenr) |
2776 | goto next; | |
5d4f98a2 YZ |
2777 | } |
2778 | ||
da17066c | 2779 | blocksize = root->fs_info->nodesize; |
5d4f98a2 | 2780 | generation = btrfs_node_ptr_generation(upper->eb, slot); |
2ff7e61e | 2781 | eb = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
2782 | if (IS_ERR(eb)) { |
2783 | err = PTR_ERR(eb); | |
2784 | goto next; | |
2785 | } else if (!extent_buffer_uptodate(eb)) { | |
416bc658 | 2786 | free_extent_buffer(eb); |
97d9a8a4 TI |
2787 | err = -EIO; |
2788 | goto next; | |
2789 | } | |
5d4f98a2 YZ |
2790 | btrfs_tree_lock(eb); |
2791 | btrfs_set_lock_blocking(eb); | |
2792 | ||
2793 | if (!node->eb) { | |
2794 | ret = btrfs_cow_block(trans, root, eb, upper->eb, | |
2795 | slot, &eb); | |
3fd0a558 YZ |
2796 | btrfs_tree_unlock(eb); |
2797 | free_extent_buffer(eb); | |
5d4f98a2 YZ |
2798 | if (ret < 0) { |
2799 | err = ret; | |
3fd0a558 | 2800 | goto next; |
5d4f98a2 | 2801 | } |
3fd0a558 | 2802 | BUG_ON(node->eb != eb); |
5d4f98a2 YZ |
2803 | } else { |
2804 | btrfs_set_node_blockptr(upper->eb, slot, | |
2805 | node->eb->start); | |
2806 | btrfs_set_node_ptr_generation(upper->eb, slot, | |
2807 | trans->transid); | |
2808 | btrfs_mark_buffer_dirty(upper->eb); | |
2809 | ||
2ff7e61e | 2810 | ret = btrfs_inc_extent_ref(trans, root->fs_info, |
5d4f98a2 YZ |
2811 | node->eb->start, blocksize, |
2812 | upper->eb->start, | |
2813 | btrfs_header_owner(upper->eb), | |
b06c4bf5 | 2814 | node->level, 0); |
5d4f98a2 YZ |
2815 | BUG_ON(ret); |
2816 | ||
2817 | ret = btrfs_drop_subtree(trans, root, eb, upper->eb); | |
2818 | BUG_ON(ret); | |
5d4f98a2 | 2819 | } |
3fd0a558 YZ |
2820 | next: |
2821 | if (!upper->pending) | |
2822 | drop_node_buffer(upper); | |
2823 | else | |
2824 | unlock_node_buffer(upper); | |
2825 | if (err) | |
2826 | break; | |
5d4f98a2 | 2827 | } |
3fd0a558 YZ |
2828 | |
2829 | if (!err && node->pending) { | |
2830 | drop_node_buffer(node); | |
2831 | list_move_tail(&node->list, &rc->backref_cache.changed); | |
2832 | node->pending = 0; | |
2833 | } | |
2834 | ||
5d4f98a2 | 2835 | path->lowest_level = 0; |
3fd0a558 | 2836 | BUG_ON(err == -ENOSPC); |
5d4f98a2 YZ |
2837 | return err; |
2838 | } | |
2839 | ||
2840 | static int link_to_upper(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2841 | struct reloc_control *rc, |
5d4f98a2 YZ |
2842 | struct backref_node *node, |
2843 | struct btrfs_path *path) | |
2844 | { | |
2845 | struct btrfs_key key; | |
5d4f98a2 YZ |
2846 | |
2847 | btrfs_node_key_to_cpu(node->eb, &key, 0); | |
3fd0a558 | 2848 | return do_relocation(trans, rc, node, &key, path, 0); |
5d4f98a2 YZ |
2849 | } |
2850 | ||
2851 | static int finish_pending_nodes(struct btrfs_trans_handle *trans, | |
3fd0a558 YZ |
2852 | struct reloc_control *rc, |
2853 | struct btrfs_path *path, int err) | |
5d4f98a2 | 2854 | { |
3fd0a558 YZ |
2855 | LIST_HEAD(list); |
2856 | struct backref_cache *cache = &rc->backref_cache; | |
5d4f98a2 YZ |
2857 | struct backref_node *node; |
2858 | int level; | |
2859 | int ret; | |
5d4f98a2 YZ |
2860 | |
2861 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
2862 | while (!list_empty(&cache->pending[level])) { | |
2863 | node = list_entry(cache->pending[level].next, | |
3fd0a558 YZ |
2864 | struct backref_node, list); |
2865 | list_move_tail(&node->list, &list); | |
2866 | BUG_ON(!node->pending); | |
5d4f98a2 | 2867 | |
3fd0a558 YZ |
2868 | if (!err) { |
2869 | ret = link_to_upper(trans, rc, node, path); | |
2870 | if (ret < 0) | |
2871 | err = ret; | |
2872 | } | |
5d4f98a2 | 2873 | } |
3fd0a558 | 2874 | list_splice_init(&list, &cache->pending[level]); |
5d4f98a2 | 2875 | } |
5d4f98a2 YZ |
2876 | return err; |
2877 | } | |
2878 | ||
2879 | static void mark_block_processed(struct reloc_control *rc, | |
3fd0a558 YZ |
2880 | u64 bytenr, u32 blocksize) |
2881 | { | |
2882 | set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, | |
ceeb0ae7 | 2883 | EXTENT_DIRTY); |
3fd0a558 YZ |
2884 | } |
2885 | ||
2886 | static void __mark_block_processed(struct reloc_control *rc, | |
2887 | struct backref_node *node) | |
5d4f98a2 YZ |
2888 | { |
2889 | u32 blocksize; | |
2890 | if (node->level == 0 || | |
2891 | in_block_group(node->bytenr, rc->block_group)) { | |
da17066c | 2892 | blocksize = rc->extent_root->fs_info->nodesize; |
3fd0a558 | 2893 | mark_block_processed(rc, node->bytenr, blocksize); |
5d4f98a2 YZ |
2894 | } |
2895 | node->processed = 1; | |
2896 | } | |
2897 | ||
2898 | /* | |
2899 | * mark a block and all blocks directly/indirectly reference the block | |
2900 | * as processed. | |
2901 | */ | |
2902 | static void update_processed_blocks(struct reloc_control *rc, | |
2903 | struct backref_node *node) | |
2904 | { | |
2905 | struct backref_node *next = node; | |
2906 | struct backref_edge *edge; | |
2907 | struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
2908 | int index = 0; | |
2909 | ||
2910 | while (next) { | |
2911 | cond_resched(); | |
2912 | while (1) { | |
2913 | if (next->processed) | |
2914 | break; | |
2915 | ||
3fd0a558 | 2916 | __mark_block_processed(rc, next); |
5d4f98a2 YZ |
2917 | |
2918 | if (list_empty(&next->upper)) | |
2919 | break; | |
2920 | ||
2921 | edge = list_entry(next->upper.next, | |
2922 | struct backref_edge, list[LOWER]); | |
2923 | edges[index++] = edge; | |
2924 | next = edge->node[UPPER]; | |
2925 | } | |
2926 | next = walk_down_backref(edges, &index); | |
2927 | } | |
2928 | } | |
2929 | ||
7476dfda | 2930 | static int tree_block_processed(u64 bytenr, struct reloc_control *rc) |
3fd0a558 | 2931 | { |
da17066c | 2932 | u32 blocksize = rc->extent_root->fs_info->nodesize; |
7476dfda | 2933 | |
3fd0a558 YZ |
2934 | if (test_range_bit(&rc->processed_blocks, bytenr, |
2935 | bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) | |
2936 | return 1; | |
2937 | return 0; | |
5d4f98a2 YZ |
2938 | } |
2939 | ||
2ff7e61e | 2940 | static int get_tree_block_key(struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2941 | struct tree_block *block) |
2942 | { | |
2943 | struct extent_buffer *eb; | |
2944 | ||
2945 | BUG_ON(block->key_ready); | |
2ff7e61e | 2946 | eb = read_tree_block(fs_info, block->bytenr, block->key.offset); |
64c043de LB |
2947 | if (IS_ERR(eb)) { |
2948 | return PTR_ERR(eb); | |
2949 | } else if (!extent_buffer_uptodate(eb)) { | |
416bc658 JB |
2950 | free_extent_buffer(eb); |
2951 | return -EIO; | |
2952 | } | |
5d4f98a2 YZ |
2953 | WARN_ON(btrfs_header_level(eb) != block->level); |
2954 | if (block->level == 0) | |
2955 | btrfs_item_key_to_cpu(eb, &block->key, 0); | |
2956 | else | |
2957 | btrfs_node_key_to_cpu(eb, &block->key, 0); | |
2958 | free_extent_buffer(eb); | |
2959 | block->key_ready = 1; | |
2960 | return 0; | |
2961 | } | |
2962 | ||
5d4f98a2 YZ |
2963 | /* |
2964 | * helper function to relocate a tree block | |
2965 | */ | |
2966 | static int relocate_tree_block(struct btrfs_trans_handle *trans, | |
2967 | struct reloc_control *rc, | |
2968 | struct backref_node *node, | |
2969 | struct btrfs_key *key, | |
2970 | struct btrfs_path *path) | |
2971 | { | |
2972 | struct btrfs_root *root; | |
3fd0a558 YZ |
2973 | int ret = 0; |
2974 | ||
2975 | if (!node) | |
2976 | return 0; | |
5d4f98a2 | 2977 | |
3fd0a558 | 2978 | BUG_ON(node->processed); |
147d256e | 2979 | root = select_one_root(node); |
3fd0a558 | 2980 | if (root == ERR_PTR(-ENOENT)) { |
5d4f98a2 | 2981 | update_processed_blocks(rc, node); |
3fd0a558 | 2982 | goto out; |
5d4f98a2 YZ |
2983 | } |
2984 | ||
27cdeb70 | 2985 | if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
3fd0a558 YZ |
2986 | ret = reserve_metadata_space(trans, rc, node); |
2987 | if (ret) | |
5d4f98a2 | 2988 | goto out; |
5d4f98a2 YZ |
2989 | } |
2990 | ||
3fd0a558 | 2991 | if (root) { |
27cdeb70 | 2992 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
3fd0a558 YZ |
2993 | BUG_ON(node->new_bytenr); |
2994 | BUG_ON(!list_empty(&node->list)); | |
2995 | btrfs_record_root_in_trans(trans, root); | |
2996 | root = root->reloc_root; | |
2997 | node->new_bytenr = root->node->start; | |
2998 | node->root = root; | |
2999 | list_add_tail(&node->list, &rc->backref_cache.changed); | |
3000 | } else { | |
3001 | path->lowest_level = node->level; | |
3002 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); | |
b3b4aa74 | 3003 | btrfs_release_path(path); |
3fd0a558 YZ |
3004 | if (ret > 0) |
3005 | ret = 0; | |
3006 | } | |
3007 | if (!ret) | |
3008 | update_processed_blocks(rc, node); | |
3009 | } else { | |
3010 | ret = do_relocation(trans, rc, node, key, path, 1); | |
3011 | } | |
5d4f98a2 | 3012 | out: |
0647bf56 | 3013 | if (ret || node->level == 0 || node->cowonly) |
3fd0a558 | 3014 | remove_backref_node(&rc->backref_cache, node); |
5d4f98a2 YZ |
3015 | return ret; |
3016 | } | |
3017 | ||
3018 | /* | |
3019 | * relocate a list of blocks | |
3020 | */ | |
3021 | static noinline_for_stack | |
3022 | int relocate_tree_blocks(struct btrfs_trans_handle *trans, | |
3023 | struct reloc_control *rc, struct rb_root *blocks) | |
3024 | { | |
2ff7e61e | 3025 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3026 | struct backref_node *node; |
3027 | struct btrfs_path *path; | |
3028 | struct tree_block *block; | |
3029 | struct rb_node *rb_node; | |
5d4f98a2 YZ |
3030 | int ret; |
3031 | int err = 0; | |
3032 | ||
3033 | path = btrfs_alloc_path(); | |
e1a12670 LB |
3034 | if (!path) { |
3035 | err = -ENOMEM; | |
34c2b290 | 3036 | goto out_free_blocks; |
e1a12670 | 3037 | } |
5d4f98a2 | 3038 | |
5d4f98a2 YZ |
3039 | rb_node = rb_first(blocks); |
3040 | while (rb_node) { | |
3041 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
5d4f98a2 | 3042 | if (!block->key_ready) |
2ff7e61e | 3043 | readahead_tree_block(fs_info, block->bytenr); |
5d4f98a2 YZ |
3044 | rb_node = rb_next(rb_node); |
3045 | } | |
3046 | ||
3047 | rb_node = rb_first(blocks); | |
3048 | while (rb_node) { | |
3049 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
34c2b290 | 3050 | if (!block->key_ready) { |
2ff7e61e | 3051 | err = get_tree_block_key(fs_info, block); |
34c2b290 DS |
3052 | if (err) |
3053 | goto out_free_path; | |
3054 | } | |
5d4f98a2 YZ |
3055 | rb_node = rb_next(rb_node); |
3056 | } | |
3057 | ||
3058 | rb_node = rb_first(blocks); | |
3059 | while (rb_node) { | |
3060 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
3061 | ||
3fd0a558 | 3062 | node = build_backref_tree(rc, &block->key, |
5d4f98a2 YZ |
3063 | block->level, block->bytenr); |
3064 | if (IS_ERR(node)) { | |
3065 | err = PTR_ERR(node); | |
3066 | goto out; | |
3067 | } | |
3068 | ||
3069 | ret = relocate_tree_block(trans, rc, node, &block->key, | |
3070 | path); | |
3071 | if (ret < 0) { | |
3fd0a558 YZ |
3072 | if (ret != -EAGAIN || rb_node == rb_first(blocks)) |
3073 | err = ret; | |
5d4f98a2 YZ |
3074 | goto out; |
3075 | } | |
5d4f98a2 YZ |
3076 | rb_node = rb_next(rb_node); |
3077 | } | |
5d4f98a2 | 3078 | out: |
3fd0a558 | 3079 | err = finish_pending_nodes(trans, rc, path, err); |
5d4f98a2 | 3080 | |
34c2b290 | 3081 | out_free_path: |
5d4f98a2 | 3082 | btrfs_free_path(path); |
34c2b290 | 3083 | out_free_blocks: |
e1a12670 | 3084 | free_block_list(blocks); |
5d4f98a2 YZ |
3085 | return err; |
3086 | } | |
3087 | ||
efa56464 YZ |
3088 | static noinline_for_stack |
3089 | int prealloc_file_extent_cluster(struct inode *inode, | |
3090 | struct file_extent_cluster *cluster) | |
3091 | { | |
3092 | u64 alloc_hint = 0; | |
3093 | u64 start; | |
3094 | u64 end; | |
3095 | u64 offset = BTRFS_I(inode)->index_cnt; | |
3096 | u64 num_bytes; | |
3097 | int nr = 0; | |
3098 | int ret = 0; | |
dcb40c19 WX |
3099 | u64 prealloc_start = cluster->start - offset; |
3100 | u64 prealloc_end = cluster->end - offset; | |
18513091 | 3101 | u64 cur_offset; |
364ecf36 | 3102 | struct extent_changeset *data_reserved = NULL; |
efa56464 YZ |
3103 | |
3104 | BUG_ON(cluster->start != cluster->boundary[0]); | |
5955102c | 3105 | inode_lock(inode); |
efa56464 | 3106 | |
364ecf36 | 3107 | ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, |
dcb40c19 | 3108 | prealloc_end + 1 - prealloc_start); |
efa56464 YZ |
3109 | if (ret) |
3110 | goto out; | |
3111 | ||
18513091 | 3112 | cur_offset = prealloc_start; |
efa56464 YZ |
3113 | while (nr < cluster->nr) { |
3114 | start = cluster->boundary[nr] - offset; | |
3115 | if (nr + 1 < cluster->nr) | |
3116 | end = cluster->boundary[nr + 1] - 1 - offset; | |
3117 | else | |
3118 | end = cluster->end - offset; | |
3119 | ||
d0082371 | 3120 | lock_extent(&BTRFS_I(inode)->io_tree, start, end); |
efa56464 | 3121 | num_bytes = end + 1 - start; |
18513091 | 3122 | if (cur_offset < start) |
bc42bda2 QW |
3123 | btrfs_free_reserved_data_space(inode, data_reserved, |
3124 | cur_offset, start - cur_offset); | |
efa56464 YZ |
3125 | ret = btrfs_prealloc_file_range(inode, 0, start, |
3126 | num_bytes, num_bytes, | |
3127 | end + 1, &alloc_hint); | |
18513091 | 3128 | cur_offset = end + 1; |
d0082371 | 3129 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end); |
efa56464 YZ |
3130 | if (ret) |
3131 | break; | |
3132 | nr++; | |
3133 | } | |
18513091 | 3134 | if (cur_offset < prealloc_end) |
bc42bda2 QW |
3135 | btrfs_free_reserved_data_space(inode, data_reserved, |
3136 | cur_offset, prealloc_end + 1 - cur_offset); | |
efa56464 | 3137 | out: |
5955102c | 3138 | inode_unlock(inode); |
364ecf36 | 3139 | extent_changeset_free(data_reserved); |
efa56464 YZ |
3140 | return ret; |
3141 | } | |
3142 | ||
5d4f98a2 | 3143 | static noinline_for_stack |
0257bb82 YZ |
3144 | int setup_extent_mapping(struct inode *inode, u64 start, u64 end, |
3145 | u64 block_start) | |
3146 | { | |
0b246afa | 3147 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0257bb82 YZ |
3148 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
3149 | struct extent_map *em; | |
3150 | int ret = 0; | |
3151 | ||
172ddd60 | 3152 | em = alloc_extent_map(); |
0257bb82 YZ |
3153 | if (!em) |
3154 | return -ENOMEM; | |
3155 | ||
3156 | em->start = start; | |
3157 | em->len = end + 1 - start; | |
3158 | em->block_len = em->len; | |
3159 | em->block_start = block_start; | |
0b246afa | 3160 | em->bdev = fs_info->fs_devices->latest_bdev; |
0257bb82 YZ |
3161 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
3162 | ||
d0082371 | 3163 | lock_extent(&BTRFS_I(inode)->io_tree, start, end); |
0257bb82 YZ |
3164 | while (1) { |
3165 | write_lock(&em_tree->lock); | |
09a2a8f9 | 3166 | ret = add_extent_mapping(em_tree, em, 0); |
0257bb82 YZ |
3167 | write_unlock(&em_tree->lock); |
3168 | if (ret != -EEXIST) { | |
3169 | free_extent_map(em); | |
3170 | break; | |
3171 | } | |
dcdbc059 | 3172 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
0257bb82 | 3173 | } |
d0082371 | 3174 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end); |
0257bb82 YZ |
3175 | return ret; |
3176 | } | |
3177 | ||
3178 | static int relocate_file_extent_cluster(struct inode *inode, | |
3179 | struct file_extent_cluster *cluster) | |
5d4f98a2 | 3180 | { |
2ff7e61e | 3181 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 YZ |
3182 | u64 page_start; |
3183 | u64 page_end; | |
0257bb82 YZ |
3184 | u64 offset = BTRFS_I(inode)->index_cnt; |
3185 | unsigned long index; | |
5d4f98a2 | 3186 | unsigned long last_index; |
5d4f98a2 YZ |
3187 | struct page *page; |
3188 | struct file_ra_state *ra; | |
3b16a4e3 | 3189 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
0257bb82 | 3190 | int nr = 0; |
5d4f98a2 YZ |
3191 | int ret = 0; |
3192 | ||
0257bb82 YZ |
3193 | if (!cluster->nr) |
3194 | return 0; | |
3195 | ||
5d4f98a2 YZ |
3196 | ra = kzalloc(sizeof(*ra), GFP_NOFS); |
3197 | if (!ra) | |
3198 | return -ENOMEM; | |
3199 | ||
efa56464 YZ |
3200 | ret = prealloc_file_extent_cluster(inode, cluster); |
3201 | if (ret) | |
3202 | goto out; | |
0257bb82 | 3203 | |
efa56464 | 3204 | file_ra_state_init(ra, inode->i_mapping); |
5d4f98a2 | 3205 | |
0257bb82 YZ |
3206 | ret = setup_extent_mapping(inode, cluster->start - offset, |
3207 | cluster->end - offset, cluster->start); | |
5d4f98a2 | 3208 | if (ret) |
efa56464 | 3209 | goto out; |
5d4f98a2 | 3210 | |
09cbfeaf KS |
3211 | index = (cluster->start - offset) >> PAGE_SHIFT; |
3212 | last_index = (cluster->end - offset) >> PAGE_SHIFT; | |
0257bb82 | 3213 | while (index <= last_index) { |
9f3db423 NB |
3214 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
3215 | PAGE_SIZE); | |
efa56464 YZ |
3216 | if (ret) |
3217 | goto out; | |
3218 | ||
0257bb82 | 3219 | page = find_lock_page(inode->i_mapping, index); |
5d4f98a2 | 3220 | if (!page) { |
0257bb82 YZ |
3221 | page_cache_sync_readahead(inode->i_mapping, |
3222 | ra, NULL, index, | |
3223 | last_index + 1 - index); | |
a94733d0 | 3224 | page = find_or_create_page(inode->i_mapping, index, |
3b16a4e3 | 3225 | mask); |
0257bb82 | 3226 | if (!page) { |
691fa059 | 3227 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
09cbfeaf | 3228 | PAGE_SIZE); |
0257bb82 | 3229 | ret = -ENOMEM; |
efa56464 | 3230 | goto out; |
0257bb82 | 3231 | } |
5d4f98a2 | 3232 | } |
0257bb82 YZ |
3233 | |
3234 | if (PageReadahead(page)) { | |
3235 | page_cache_async_readahead(inode->i_mapping, | |
3236 | ra, NULL, page, index, | |
3237 | last_index + 1 - index); | |
3238 | } | |
3239 | ||
5d4f98a2 YZ |
3240 | if (!PageUptodate(page)) { |
3241 | btrfs_readpage(NULL, page); | |
3242 | lock_page(page); | |
3243 | if (!PageUptodate(page)) { | |
3244 | unlock_page(page); | |
09cbfeaf | 3245 | put_page(page); |
691fa059 | 3246 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
09cbfeaf | 3247 | PAGE_SIZE); |
5d4f98a2 | 3248 | ret = -EIO; |
efa56464 | 3249 | goto out; |
5d4f98a2 YZ |
3250 | } |
3251 | } | |
5d4f98a2 | 3252 | |
4eee4fa4 | 3253 | page_start = page_offset(page); |
09cbfeaf | 3254 | page_end = page_start + PAGE_SIZE - 1; |
0257bb82 | 3255 | |
d0082371 | 3256 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); |
0257bb82 | 3257 | |
5d4f98a2 YZ |
3258 | set_page_extent_mapped(page); |
3259 | ||
0257bb82 YZ |
3260 | if (nr < cluster->nr && |
3261 | page_start + offset == cluster->boundary[nr]) { | |
3262 | set_extent_bits(&BTRFS_I(inode)->io_tree, | |
3263 | page_start, page_end, | |
ceeb0ae7 | 3264 | EXTENT_BOUNDARY); |
0257bb82 YZ |
3265 | nr++; |
3266 | } | |
5d4f98a2 | 3267 | |
ba8b04c1 | 3268 | btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0); |
5d4f98a2 | 3269 | set_page_dirty(page); |
5d4f98a2 | 3270 | |
0257bb82 | 3271 | unlock_extent(&BTRFS_I(inode)->io_tree, |
d0082371 | 3272 | page_start, page_end); |
5d4f98a2 | 3273 | unlock_page(page); |
09cbfeaf | 3274 | put_page(page); |
0257bb82 YZ |
3275 | |
3276 | index++; | |
efa56464 | 3277 | balance_dirty_pages_ratelimited(inode->i_mapping); |
2ff7e61e | 3278 | btrfs_throttle(fs_info); |
5d4f98a2 | 3279 | } |
0257bb82 | 3280 | WARN_ON(nr != cluster->nr); |
efa56464 | 3281 | out: |
5d4f98a2 | 3282 | kfree(ra); |
5d4f98a2 YZ |
3283 | return ret; |
3284 | } | |
3285 | ||
3286 | static noinline_for_stack | |
0257bb82 YZ |
3287 | int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, |
3288 | struct file_extent_cluster *cluster) | |
5d4f98a2 | 3289 | { |
0257bb82 | 3290 | int ret; |
5d4f98a2 | 3291 | |
0257bb82 YZ |
3292 | if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { |
3293 | ret = relocate_file_extent_cluster(inode, cluster); | |
3294 | if (ret) | |
3295 | return ret; | |
3296 | cluster->nr = 0; | |
5d4f98a2 | 3297 | } |
5d4f98a2 | 3298 | |
0257bb82 YZ |
3299 | if (!cluster->nr) |
3300 | cluster->start = extent_key->objectid; | |
3301 | else | |
3302 | BUG_ON(cluster->nr >= MAX_EXTENTS); | |
3303 | cluster->end = extent_key->objectid + extent_key->offset - 1; | |
3304 | cluster->boundary[cluster->nr] = extent_key->objectid; | |
3305 | cluster->nr++; | |
3306 | ||
3307 | if (cluster->nr >= MAX_EXTENTS) { | |
3308 | ret = relocate_file_extent_cluster(inode, cluster); | |
3309 | if (ret) | |
3310 | return ret; | |
3311 | cluster->nr = 0; | |
3312 | } | |
3313 | return 0; | |
5d4f98a2 YZ |
3314 | } |
3315 | ||
3316 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3317 | static int get_ref_objectid_v0(struct reloc_control *rc, | |
3318 | struct btrfs_path *path, | |
3319 | struct btrfs_key *extent_key, | |
3320 | u64 *ref_objectid, int *path_change) | |
3321 | { | |
3322 | struct btrfs_key key; | |
3323 | struct extent_buffer *leaf; | |
3324 | struct btrfs_extent_ref_v0 *ref0; | |
3325 | int ret; | |
3326 | int slot; | |
3327 | ||
3328 | leaf = path->nodes[0]; | |
3329 | slot = path->slots[0]; | |
3330 | while (1) { | |
3331 | if (slot >= btrfs_header_nritems(leaf)) { | |
3332 | ret = btrfs_next_leaf(rc->extent_root, path); | |
3333 | if (ret < 0) | |
3334 | return ret; | |
3335 | BUG_ON(ret > 0); | |
3336 | leaf = path->nodes[0]; | |
3337 | slot = path->slots[0]; | |
3338 | if (path_change) | |
3339 | *path_change = 1; | |
3340 | } | |
3341 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3342 | if (key.objectid != extent_key->objectid) | |
3343 | return -ENOENT; | |
3344 | ||
3345 | if (key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
3346 | slot++; | |
3347 | continue; | |
3348 | } | |
3349 | ref0 = btrfs_item_ptr(leaf, slot, | |
3350 | struct btrfs_extent_ref_v0); | |
3351 | *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); | |
3352 | break; | |
3353 | } | |
3354 | return 0; | |
3355 | } | |
3356 | #endif | |
3357 | ||
3358 | /* | |
3359 | * helper to add a tree block to the list. | |
3360 | * the major work is getting the generation and level of the block | |
3361 | */ | |
3362 | static int add_tree_block(struct reloc_control *rc, | |
3363 | struct btrfs_key *extent_key, | |
3364 | struct btrfs_path *path, | |
3365 | struct rb_root *blocks) | |
3366 | { | |
3367 | struct extent_buffer *eb; | |
3368 | struct btrfs_extent_item *ei; | |
3369 | struct btrfs_tree_block_info *bi; | |
3370 | struct tree_block *block; | |
3371 | struct rb_node *rb_node; | |
3372 | u32 item_size; | |
3373 | int level = -1; | |
7fdf4b60 | 3374 | u64 generation; |
5d4f98a2 YZ |
3375 | |
3376 | eb = path->nodes[0]; | |
3377 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
3378 | ||
3173a18f JB |
3379 | if (extent_key->type == BTRFS_METADATA_ITEM_KEY || |
3380 | item_size >= sizeof(*ei) + sizeof(*bi)) { | |
5d4f98a2 YZ |
3381 | ei = btrfs_item_ptr(eb, path->slots[0], |
3382 | struct btrfs_extent_item); | |
3173a18f JB |
3383 | if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { |
3384 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3385 | level = btrfs_tree_block_level(eb, bi); | |
3386 | } else { | |
3387 | level = (int)extent_key->offset; | |
3388 | } | |
5d4f98a2 | 3389 | generation = btrfs_extent_generation(eb, ei); |
5d4f98a2 YZ |
3390 | } else { |
3391 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3392 | u64 ref_owner; | |
3393 | int ret; | |
3394 | ||
3395 | BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3396 | ret = get_ref_objectid_v0(rc, path, extent_key, | |
3397 | &ref_owner, NULL); | |
411fc6bc AK |
3398 | if (ret < 0) |
3399 | return ret; | |
5d4f98a2 YZ |
3400 | BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); |
3401 | level = (int)ref_owner; | |
3402 | /* FIXME: get real generation */ | |
3403 | generation = 0; | |
3404 | #else | |
3405 | BUG(); | |
3406 | #endif | |
3407 | } | |
3408 | ||
b3b4aa74 | 3409 | btrfs_release_path(path); |
5d4f98a2 YZ |
3410 | |
3411 | BUG_ON(level == -1); | |
3412 | ||
3413 | block = kmalloc(sizeof(*block), GFP_NOFS); | |
3414 | if (!block) | |
3415 | return -ENOMEM; | |
3416 | ||
3417 | block->bytenr = extent_key->objectid; | |
da17066c | 3418 | block->key.objectid = rc->extent_root->fs_info->nodesize; |
5d4f98a2 YZ |
3419 | block->key.offset = generation; |
3420 | block->level = level; | |
3421 | block->key_ready = 0; | |
3422 | ||
3423 | rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); | |
43c04fb1 JM |
3424 | if (rb_node) |
3425 | backref_tree_panic(rb_node, -EEXIST, block->bytenr); | |
5d4f98a2 YZ |
3426 | |
3427 | return 0; | |
3428 | } | |
3429 | ||
3430 | /* | |
3431 | * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY | |
3432 | */ | |
3433 | static int __add_tree_block(struct reloc_control *rc, | |
3434 | u64 bytenr, u32 blocksize, | |
3435 | struct rb_root *blocks) | |
3436 | { | |
0b246afa | 3437 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3438 | struct btrfs_path *path; |
3439 | struct btrfs_key key; | |
3440 | int ret; | |
0b246afa | 3441 | bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
5d4f98a2 | 3442 | |
7476dfda | 3443 | if (tree_block_processed(bytenr, rc)) |
5d4f98a2 YZ |
3444 | return 0; |
3445 | ||
3446 | if (tree_search(blocks, bytenr)) | |
3447 | return 0; | |
3448 | ||
3449 | path = btrfs_alloc_path(); | |
3450 | if (!path) | |
3451 | return -ENOMEM; | |
aee68ee5 | 3452 | again: |
5d4f98a2 | 3453 | key.objectid = bytenr; |
aee68ee5 JB |
3454 | if (skinny) { |
3455 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3456 | key.offset = (u64)-1; | |
3457 | } else { | |
3458 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3459 | key.offset = blocksize; | |
3460 | } | |
5d4f98a2 YZ |
3461 | |
3462 | path->search_commit_root = 1; | |
3463 | path->skip_locking = 1; | |
3464 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); | |
3465 | if (ret < 0) | |
3466 | goto out; | |
5d4f98a2 | 3467 | |
aee68ee5 JB |
3468 | if (ret > 0 && skinny) { |
3469 | if (path->slots[0]) { | |
3470 | path->slots[0]--; | |
3471 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3472 | path->slots[0]); | |
3473 | if (key.objectid == bytenr && | |
3474 | (key.type == BTRFS_METADATA_ITEM_KEY || | |
3475 | (key.type == BTRFS_EXTENT_ITEM_KEY && | |
3476 | key.offset == blocksize))) | |
3477 | ret = 0; | |
3478 | } | |
3479 | ||
3480 | if (ret) { | |
3481 | skinny = false; | |
3482 | btrfs_release_path(path); | |
3483 | goto again; | |
3484 | } | |
3173a18f JB |
3485 | } |
3486 | BUG_ON(ret); | |
3487 | ||
5d4f98a2 YZ |
3488 | ret = add_tree_block(rc, &key, path, blocks); |
3489 | out: | |
3490 | btrfs_free_path(path); | |
3491 | return ret; | |
3492 | } | |
3493 | ||
3494 | /* | |
3495 | * helper to check if the block use full backrefs for pointers in it | |
3496 | */ | |
3497 | static int block_use_full_backref(struct reloc_control *rc, | |
3498 | struct extent_buffer *eb) | |
3499 | { | |
5d4f98a2 YZ |
3500 | u64 flags; |
3501 | int ret; | |
3502 | ||
3503 | if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || | |
3504 | btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) | |
3505 | return 1; | |
3506 | ||
2ff7e61e | 3507 | ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, |
3173a18f JB |
3508 | eb->start, btrfs_header_level(eb), 1, |
3509 | NULL, &flags); | |
5d4f98a2 YZ |
3510 | BUG_ON(ret); |
3511 | ||
5d4f98a2 YZ |
3512 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
3513 | ret = 1; | |
3514 | else | |
3515 | ret = 0; | |
5d4f98a2 YZ |
3516 | return ret; |
3517 | } | |
3518 | ||
0af3d00b | 3519 | static int delete_block_group_cache(struct btrfs_fs_info *fs_info, |
1bbc621e CM |
3520 | struct btrfs_block_group_cache *block_group, |
3521 | struct inode *inode, | |
3522 | u64 ino) | |
0af3d00b JB |
3523 | { |
3524 | struct btrfs_key key; | |
0af3d00b JB |
3525 | struct btrfs_root *root = fs_info->tree_root; |
3526 | struct btrfs_trans_handle *trans; | |
0af3d00b JB |
3527 | int ret = 0; |
3528 | ||
3529 | if (inode) | |
3530 | goto truncate; | |
3531 | ||
3532 | key.objectid = ino; | |
3533 | key.type = BTRFS_INODE_ITEM_KEY; | |
3534 | key.offset = 0; | |
3535 | ||
3536 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
f54fb859 TI |
3537 | if (IS_ERR(inode) || is_bad_inode(inode)) { |
3538 | if (!IS_ERR(inode)) | |
0af3d00b JB |
3539 | iput(inode); |
3540 | return -ENOENT; | |
3541 | } | |
3542 | ||
3543 | truncate: | |
2ff7e61e | 3544 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
7b61cd92 MX |
3545 | &fs_info->global_block_rsv); |
3546 | if (ret) | |
3547 | goto out; | |
3548 | ||
7a7eaa40 | 3549 | trans = btrfs_join_transaction(root); |
0af3d00b | 3550 | if (IS_ERR(trans)) { |
3612b495 | 3551 | ret = PTR_ERR(trans); |
0af3d00b JB |
3552 | goto out; |
3553 | } | |
3554 | ||
77ab86bf | 3555 | ret = btrfs_truncate_free_space_cache(trans, block_group, inode); |
0af3d00b | 3556 | |
3a45bb20 | 3557 | btrfs_end_transaction(trans); |
2ff7e61e | 3558 | btrfs_btree_balance_dirty(fs_info); |
0af3d00b JB |
3559 | out: |
3560 | iput(inode); | |
3561 | return ret; | |
3562 | } | |
3563 | ||
5d4f98a2 YZ |
3564 | /* |
3565 | * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY | |
3566 | * this function scans fs tree to find blocks reference the data extent | |
3567 | */ | |
3568 | static int find_data_references(struct reloc_control *rc, | |
3569 | struct btrfs_key *extent_key, | |
3570 | struct extent_buffer *leaf, | |
3571 | struct btrfs_extent_data_ref *ref, | |
3572 | struct rb_root *blocks) | |
3573 | { | |
0b246afa | 3574 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3575 | struct btrfs_path *path; |
3576 | struct tree_block *block; | |
3577 | struct btrfs_root *root; | |
3578 | struct btrfs_file_extent_item *fi; | |
3579 | struct rb_node *rb_node; | |
3580 | struct btrfs_key key; | |
3581 | u64 ref_root; | |
3582 | u64 ref_objectid; | |
3583 | u64 ref_offset; | |
3584 | u32 ref_count; | |
3585 | u32 nritems; | |
3586 | int err = 0; | |
3587 | int added = 0; | |
3588 | int counted; | |
3589 | int ret; | |
3590 | ||
5d4f98a2 YZ |
3591 | ref_root = btrfs_extent_data_ref_root(leaf, ref); |
3592 | ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); | |
3593 | ref_offset = btrfs_extent_data_ref_offset(leaf, ref); | |
3594 | ref_count = btrfs_extent_data_ref_count(leaf, ref); | |
3595 | ||
0af3d00b JB |
3596 | /* |
3597 | * This is an extent belonging to the free space cache, lets just delete | |
3598 | * it and redo the search. | |
3599 | */ | |
3600 | if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { | |
0b246afa | 3601 | ret = delete_block_group_cache(fs_info, rc->block_group, |
0af3d00b JB |
3602 | NULL, ref_objectid); |
3603 | if (ret != -ENOENT) | |
3604 | return ret; | |
3605 | ret = 0; | |
3606 | } | |
3607 | ||
3608 | path = btrfs_alloc_path(); | |
3609 | if (!path) | |
3610 | return -ENOMEM; | |
e4058b54 | 3611 | path->reada = READA_FORWARD; |
0af3d00b | 3612 | |
0b246afa | 3613 | root = read_fs_root(fs_info, ref_root); |
5d4f98a2 YZ |
3614 | if (IS_ERR(root)) { |
3615 | err = PTR_ERR(root); | |
3616 | goto out; | |
3617 | } | |
3618 | ||
3619 | key.objectid = ref_objectid; | |
5d4f98a2 | 3620 | key.type = BTRFS_EXTENT_DATA_KEY; |
84850e8d YZ |
3621 | if (ref_offset > ((u64)-1 << 32)) |
3622 | key.offset = 0; | |
3623 | else | |
3624 | key.offset = ref_offset; | |
5d4f98a2 YZ |
3625 | |
3626 | path->search_commit_root = 1; | |
3627 | path->skip_locking = 1; | |
3628 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3629 | if (ret < 0) { | |
3630 | err = ret; | |
3631 | goto out; | |
3632 | } | |
3633 | ||
3634 | leaf = path->nodes[0]; | |
3635 | nritems = btrfs_header_nritems(leaf); | |
3636 | /* | |
3637 | * the references in tree blocks that use full backrefs | |
3638 | * are not counted in | |
3639 | */ | |
3640 | if (block_use_full_backref(rc, leaf)) | |
3641 | counted = 0; | |
3642 | else | |
3643 | counted = 1; | |
3644 | rb_node = tree_search(blocks, leaf->start); | |
3645 | if (rb_node) { | |
3646 | if (counted) | |
3647 | added = 1; | |
3648 | else | |
3649 | path->slots[0] = nritems; | |
3650 | } | |
3651 | ||
3652 | while (ref_count > 0) { | |
3653 | while (path->slots[0] >= nritems) { | |
3654 | ret = btrfs_next_leaf(root, path); | |
3655 | if (ret < 0) { | |
3656 | err = ret; | |
3657 | goto out; | |
3658 | } | |
fae7f21c | 3659 | if (WARN_ON(ret > 0)) |
5d4f98a2 | 3660 | goto out; |
5d4f98a2 YZ |
3661 | |
3662 | leaf = path->nodes[0]; | |
3663 | nritems = btrfs_header_nritems(leaf); | |
3664 | added = 0; | |
3665 | ||
3666 | if (block_use_full_backref(rc, leaf)) | |
3667 | counted = 0; | |
3668 | else | |
3669 | counted = 1; | |
3670 | rb_node = tree_search(blocks, leaf->start); | |
3671 | if (rb_node) { | |
3672 | if (counted) | |
3673 | added = 1; | |
3674 | else | |
3675 | path->slots[0] = nritems; | |
3676 | } | |
3677 | } | |
3678 | ||
3679 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
fae7f21c DG |
3680 | if (WARN_ON(key.objectid != ref_objectid || |
3681 | key.type != BTRFS_EXTENT_DATA_KEY)) | |
5d4f98a2 | 3682 | break; |
5d4f98a2 YZ |
3683 | |
3684 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3685 | struct btrfs_file_extent_item); | |
3686 | ||
3687 | if (btrfs_file_extent_type(leaf, fi) == | |
3688 | BTRFS_FILE_EXTENT_INLINE) | |
3689 | goto next; | |
3690 | ||
3691 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != | |
3692 | extent_key->objectid) | |
3693 | goto next; | |
3694 | ||
3695 | key.offset -= btrfs_file_extent_offset(leaf, fi); | |
3696 | if (key.offset != ref_offset) | |
3697 | goto next; | |
3698 | ||
3699 | if (counted) | |
3700 | ref_count--; | |
3701 | if (added) | |
3702 | goto next; | |
3703 | ||
7476dfda | 3704 | if (!tree_block_processed(leaf->start, rc)) { |
5d4f98a2 YZ |
3705 | block = kmalloc(sizeof(*block), GFP_NOFS); |
3706 | if (!block) { | |
3707 | err = -ENOMEM; | |
3708 | break; | |
3709 | } | |
3710 | block->bytenr = leaf->start; | |
3711 | btrfs_item_key_to_cpu(leaf, &block->key, 0); | |
3712 | block->level = 0; | |
3713 | block->key_ready = 1; | |
3714 | rb_node = tree_insert(blocks, block->bytenr, | |
3715 | &block->rb_node); | |
43c04fb1 JM |
3716 | if (rb_node) |
3717 | backref_tree_panic(rb_node, -EEXIST, | |
3718 | block->bytenr); | |
5d4f98a2 YZ |
3719 | } |
3720 | if (counted) | |
3721 | added = 1; | |
3722 | else | |
3723 | path->slots[0] = nritems; | |
3724 | next: | |
3725 | path->slots[0]++; | |
3726 | ||
3727 | } | |
3728 | out: | |
3729 | btrfs_free_path(path); | |
3730 | return err; | |
3731 | } | |
3732 | ||
3733 | /* | |
2c016dc2 | 3734 | * helper to find all tree blocks that reference a given data extent |
5d4f98a2 YZ |
3735 | */ |
3736 | static noinline_for_stack | |
3737 | int add_data_references(struct reloc_control *rc, | |
3738 | struct btrfs_key *extent_key, | |
3739 | struct btrfs_path *path, | |
3740 | struct rb_root *blocks) | |
3741 | { | |
3742 | struct btrfs_key key; | |
3743 | struct extent_buffer *eb; | |
3744 | struct btrfs_extent_data_ref *dref; | |
3745 | struct btrfs_extent_inline_ref *iref; | |
3746 | unsigned long ptr; | |
3747 | unsigned long end; | |
da17066c | 3748 | u32 blocksize = rc->extent_root->fs_info->nodesize; |
647f63bd | 3749 | int ret = 0; |
5d4f98a2 YZ |
3750 | int err = 0; |
3751 | ||
5d4f98a2 YZ |
3752 | eb = path->nodes[0]; |
3753 | ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
3754 | end = ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
3755 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3756 | if (ptr + sizeof(struct btrfs_extent_item_v0) == end) | |
3757 | ptr = end; | |
3758 | else | |
3759 | #endif | |
3760 | ptr += sizeof(struct btrfs_extent_item); | |
3761 | ||
3762 | while (ptr < end) { | |
3763 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
3764 | key.type = btrfs_get_extent_inline_ref_type(eb, iref, |
3765 | BTRFS_REF_TYPE_DATA); | |
5d4f98a2 YZ |
3766 | if (key.type == BTRFS_SHARED_DATA_REF_KEY) { |
3767 | key.offset = btrfs_extent_inline_ref_offset(eb, iref); | |
3768 | ret = __add_tree_block(rc, key.offset, blocksize, | |
3769 | blocks); | |
3770 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
3771 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3772 | ret = find_data_references(rc, extent_key, | |
3773 | eb, dref, blocks); | |
3774 | } else { | |
b14c55a1 LB |
3775 | ret = -EINVAL; |
3776 | btrfs_err(rc->extent_root->fs_info, | |
3777 | "extent %llu slot %d has an invalid inline ref type", | |
3778 | eb->start, path->slots[0]); | |
5d4f98a2 | 3779 | } |
647f63bd FDBM |
3780 | if (ret) { |
3781 | err = ret; | |
3782 | goto out; | |
3783 | } | |
5d4f98a2 YZ |
3784 | ptr += btrfs_extent_inline_ref_size(key.type); |
3785 | } | |
3786 | WARN_ON(ptr > end); | |
3787 | ||
3788 | while (1) { | |
3789 | cond_resched(); | |
3790 | eb = path->nodes[0]; | |
3791 | if (path->slots[0] >= btrfs_header_nritems(eb)) { | |
3792 | ret = btrfs_next_leaf(rc->extent_root, path); | |
3793 | if (ret < 0) { | |
3794 | err = ret; | |
3795 | break; | |
3796 | } | |
3797 | if (ret > 0) | |
3798 | break; | |
3799 | eb = path->nodes[0]; | |
3800 | } | |
3801 | ||
3802 | btrfs_item_key_to_cpu(eb, &key, path->slots[0]); | |
3803 | if (key.objectid != extent_key->objectid) | |
3804 | break; | |
3805 | ||
3806 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3807 | if (key.type == BTRFS_SHARED_DATA_REF_KEY || | |
3808 | key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
3809 | #else | |
3810 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
3811 | if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
3812 | #endif | |
3813 | ret = __add_tree_block(rc, key.offset, blocksize, | |
3814 | blocks); | |
3815 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
3816 | dref = btrfs_item_ptr(eb, path->slots[0], | |
3817 | struct btrfs_extent_data_ref); | |
3818 | ret = find_data_references(rc, extent_key, | |
3819 | eb, dref, blocks); | |
3820 | } else { | |
3821 | ret = 0; | |
3822 | } | |
3823 | if (ret) { | |
3824 | err = ret; | |
3825 | break; | |
3826 | } | |
3827 | path->slots[0]++; | |
3828 | } | |
647f63bd | 3829 | out: |
b3b4aa74 | 3830 | btrfs_release_path(path); |
5d4f98a2 YZ |
3831 | if (err) |
3832 | free_block_list(blocks); | |
3833 | return err; | |
3834 | } | |
3835 | ||
3836 | /* | |
2c016dc2 | 3837 | * helper to find next unprocessed extent |
5d4f98a2 YZ |
3838 | */ |
3839 | static noinline_for_stack | |
147d256e | 3840 | int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, |
3fd0a558 | 3841 | struct btrfs_key *extent_key) |
5d4f98a2 | 3842 | { |
0b246afa | 3843 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3844 | struct btrfs_key key; |
3845 | struct extent_buffer *leaf; | |
3846 | u64 start, end, last; | |
3847 | int ret; | |
3848 | ||
3849 | last = rc->block_group->key.objectid + rc->block_group->key.offset; | |
3850 | while (1) { | |
3851 | cond_resched(); | |
3852 | if (rc->search_start >= last) { | |
3853 | ret = 1; | |
3854 | break; | |
3855 | } | |
3856 | ||
3857 | key.objectid = rc->search_start; | |
3858 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3859 | key.offset = 0; | |
3860 | ||
3861 | path->search_commit_root = 1; | |
3862 | path->skip_locking = 1; | |
3863 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, | |
3864 | 0, 0); | |
3865 | if (ret < 0) | |
3866 | break; | |
3867 | next: | |
3868 | leaf = path->nodes[0]; | |
3869 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3870 | ret = btrfs_next_leaf(rc->extent_root, path); | |
3871 | if (ret != 0) | |
3872 | break; | |
3873 | leaf = path->nodes[0]; | |
3874 | } | |
3875 | ||
3876 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3877 | if (key.objectid >= last) { | |
3878 | ret = 1; | |
3879 | break; | |
3880 | } | |
3881 | ||
3173a18f JB |
3882 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3883 | key.type != BTRFS_METADATA_ITEM_KEY) { | |
3884 | path->slots[0]++; | |
3885 | goto next; | |
3886 | } | |
3887 | ||
3888 | if (key.type == BTRFS_EXTENT_ITEM_KEY && | |
5d4f98a2 YZ |
3889 | key.objectid + key.offset <= rc->search_start) { |
3890 | path->slots[0]++; | |
3891 | goto next; | |
3892 | } | |
3893 | ||
3173a18f | 3894 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
0b246afa | 3895 | key.objectid + fs_info->nodesize <= |
3173a18f JB |
3896 | rc->search_start) { |
3897 | path->slots[0]++; | |
3898 | goto next; | |
3899 | } | |
3900 | ||
5d4f98a2 YZ |
3901 | ret = find_first_extent_bit(&rc->processed_blocks, |
3902 | key.objectid, &start, &end, | |
e6138876 | 3903 | EXTENT_DIRTY, NULL); |
5d4f98a2 YZ |
3904 | |
3905 | if (ret == 0 && start <= key.objectid) { | |
b3b4aa74 | 3906 | btrfs_release_path(path); |
5d4f98a2 YZ |
3907 | rc->search_start = end + 1; |
3908 | } else { | |
3173a18f JB |
3909 | if (key.type == BTRFS_EXTENT_ITEM_KEY) |
3910 | rc->search_start = key.objectid + key.offset; | |
3911 | else | |
3912 | rc->search_start = key.objectid + | |
0b246afa | 3913 | fs_info->nodesize; |
3fd0a558 | 3914 | memcpy(extent_key, &key, sizeof(key)); |
5d4f98a2 YZ |
3915 | return 0; |
3916 | } | |
3917 | } | |
b3b4aa74 | 3918 | btrfs_release_path(path); |
5d4f98a2 YZ |
3919 | return ret; |
3920 | } | |
3921 | ||
3922 | static void set_reloc_control(struct reloc_control *rc) | |
3923 | { | |
3924 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3925 | |
3926 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3927 | fs_info->reloc_ctl = rc; |
7585717f | 3928 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3929 | } |
3930 | ||
3931 | static void unset_reloc_control(struct reloc_control *rc) | |
3932 | { | |
3933 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3934 | |
3935 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3936 | fs_info->reloc_ctl = NULL; |
7585717f | 3937 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3938 | } |
3939 | ||
3940 | static int check_extent_flags(u64 flags) | |
3941 | { | |
3942 | if ((flags & BTRFS_EXTENT_FLAG_DATA) && | |
3943 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) | |
3944 | return 1; | |
3945 | if (!(flags & BTRFS_EXTENT_FLAG_DATA) && | |
3946 | !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) | |
3947 | return 1; | |
3948 | if ((flags & BTRFS_EXTENT_FLAG_DATA) && | |
3949 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
3950 | return 1; | |
3951 | return 0; | |
3952 | } | |
3953 | ||
3fd0a558 YZ |
3954 | static noinline_for_stack |
3955 | int prepare_to_relocate(struct reloc_control *rc) | |
3956 | { | |
3957 | struct btrfs_trans_handle *trans; | |
ac2fabac | 3958 | int ret; |
3fd0a558 | 3959 | |
2ff7e61e | 3960 | rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, |
66d8f3dd | 3961 | BTRFS_BLOCK_RSV_TEMP); |
3fd0a558 YZ |
3962 | if (!rc->block_rsv) |
3963 | return -ENOMEM; | |
3964 | ||
3fd0a558 YZ |
3965 | memset(&rc->cluster, 0, sizeof(rc->cluster)); |
3966 | rc->search_start = rc->block_group->key.objectid; | |
3967 | rc->extents_found = 0; | |
3968 | rc->nodes_relocated = 0; | |
3969 | rc->merging_rsv_size = 0; | |
0647bf56 | 3970 | rc->reserved_bytes = 0; |
da17066c | 3971 | rc->block_rsv->size = rc->extent_root->fs_info->nodesize * |
0647bf56 | 3972 | RELOCATION_RESERVED_NODES; |
ac2fabac JB |
3973 | ret = btrfs_block_rsv_refill(rc->extent_root, |
3974 | rc->block_rsv, rc->block_rsv->size, | |
3975 | BTRFS_RESERVE_FLUSH_ALL); | |
3976 | if (ret) | |
3977 | return ret; | |
3fd0a558 YZ |
3978 | |
3979 | rc->create_reloc_tree = 1; | |
3980 | set_reloc_control(rc); | |
3981 | ||
7a7eaa40 | 3982 | trans = btrfs_join_transaction(rc->extent_root); |
28818947 LB |
3983 | if (IS_ERR(trans)) { |
3984 | unset_reloc_control(rc); | |
3985 | /* | |
3986 | * extent tree is not a ref_cow tree and has no reloc_root to | |
3987 | * cleanup. And callers are responsible to free the above | |
3988 | * block rsv. | |
3989 | */ | |
3990 | return PTR_ERR(trans); | |
3991 | } | |
3a45bb20 | 3992 | btrfs_commit_transaction(trans); |
3fd0a558 YZ |
3993 | return 0; |
3994 | } | |
76dda93c | 3995 | |
5d4f98a2 YZ |
3996 | static noinline_for_stack int relocate_block_group(struct reloc_control *rc) |
3997 | { | |
2ff7e61e | 3998 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3999 | struct rb_root blocks = RB_ROOT; |
4000 | struct btrfs_key key; | |
4001 | struct btrfs_trans_handle *trans = NULL; | |
4002 | struct btrfs_path *path; | |
4003 | struct btrfs_extent_item *ei; | |
5d4f98a2 YZ |
4004 | u64 flags; |
4005 | u32 item_size; | |
4006 | int ret; | |
4007 | int err = 0; | |
c87f08ca | 4008 | int progress = 0; |
5d4f98a2 YZ |
4009 | |
4010 | path = btrfs_alloc_path(); | |
3fd0a558 | 4011 | if (!path) |
5d4f98a2 | 4012 | return -ENOMEM; |
e4058b54 | 4013 | path->reada = READA_FORWARD; |
5d4f98a2 | 4014 | |
3fd0a558 YZ |
4015 | ret = prepare_to_relocate(rc); |
4016 | if (ret) { | |
4017 | err = ret; | |
4018 | goto out_free; | |
4019 | } | |
5d4f98a2 YZ |
4020 | |
4021 | while (1) { | |
0647bf56 WS |
4022 | rc->reserved_bytes = 0; |
4023 | ret = btrfs_block_rsv_refill(rc->extent_root, | |
4024 | rc->block_rsv, rc->block_rsv->size, | |
4025 | BTRFS_RESERVE_FLUSH_ALL); | |
4026 | if (ret) { | |
4027 | err = ret; | |
4028 | break; | |
4029 | } | |
c87f08ca | 4030 | progress++; |
a22285a6 | 4031 | trans = btrfs_start_transaction(rc->extent_root, 0); |
0f788c58 LB |
4032 | if (IS_ERR(trans)) { |
4033 | err = PTR_ERR(trans); | |
4034 | trans = NULL; | |
4035 | break; | |
4036 | } | |
c87f08ca | 4037 | restart: |
3fd0a558 | 4038 | if (update_backref_cache(trans, &rc->backref_cache)) { |
3a45bb20 | 4039 | btrfs_end_transaction(trans); |
3fd0a558 YZ |
4040 | continue; |
4041 | } | |
4042 | ||
147d256e | 4043 | ret = find_next_extent(rc, path, &key); |
5d4f98a2 YZ |
4044 | if (ret < 0) |
4045 | err = ret; | |
4046 | if (ret != 0) | |
4047 | break; | |
4048 | ||
4049 | rc->extents_found++; | |
4050 | ||
4051 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4052 | struct btrfs_extent_item); | |
3fd0a558 | 4053 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); |
5d4f98a2 YZ |
4054 | if (item_size >= sizeof(*ei)) { |
4055 | flags = btrfs_extent_flags(path->nodes[0], ei); | |
4056 | ret = check_extent_flags(flags); | |
4057 | BUG_ON(ret); | |
4058 | ||
4059 | } else { | |
4060 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4061 | u64 ref_owner; | |
4062 | int path_change = 0; | |
4063 | ||
4064 | BUG_ON(item_size != | |
4065 | sizeof(struct btrfs_extent_item_v0)); | |
4066 | ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, | |
4067 | &path_change); | |
4b3576e4 Z |
4068 | if (ret < 0) { |
4069 | err = ret; | |
4070 | break; | |
4071 | } | |
5d4f98a2 YZ |
4072 | if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) |
4073 | flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
4074 | else | |
4075 | flags = BTRFS_EXTENT_FLAG_DATA; | |
4076 | ||
4077 | if (path_change) { | |
b3b4aa74 | 4078 | btrfs_release_path(path); |
5d4f98a2 YZ |
4079 | |
4080 | path->search_commit_root = 1; | |
4081 | path->skip_locking = 1; | |
4082 | ret = btrfs_search_slot(NULL, rc->extent_root, | |
4083 | &key, path, 0, 0); | |
4084 | if (ret < 0) { | |
4085 | err = ret; | |
4086 | break; | |
4087 | } | |
4088 | BUG_ON(ret > 0); | |
4089 | } | |
4090 | #else | |
4091 | BUG(); | |
4092 | #endif | |
4093 | } | |
4094 | ||
4095 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
4096 | ret = add_tree_block(rc, &key, path, &blocks); | |
4097 | } else if (rc->stage == UPDATE_DATA_PTRS && | |
3fd0a558 | 4098 | (flags & BTRFS_EXTENT_FLAG_DATA)) { |
5d4f98a2 YZ |
4099 | ret = add_data_references(rc, &key, path, &blocks); |
4100 | } else { | |
b3b4aa74 | 4101 | btrfs_release_path(path); |
5d4f98a2 YZ |
4102 | ret = 0; |
4103 | } | |
4104 | if (ret < 0) { | |
3fd0a558 | 4105 | err = ret; |
5d4f98a2 YZ |
4106 | break; |
4107 | } | |
4108 | ||
4109 | if (!RB_EMPTY_ROOT(&blocks)) { | |
4110 | ret = relocate_tree_blocks(trans, rc, &blocks); | |
4111 | if (ret < 0) { | |
1708cc57 WS |
4112 | /* |
4113 | * if we fail to relocate tree blocks, force to update | |
4114 | * backref cache when committing transaction. | |
4115 | */ | |
4116 | rc->backref_cache.last_trans = trans->transid - 1; | |
4117 | ||
3fd0a558 YZ |
4118 | if (ret != -EAGAIN) { |
4119 | err = ret; | |
4120 | break; | |
4121 | } | |
4122 | rc->extents_found--; | |
4123 | rc->search_start = key.objectid; | |
4124 | } | |
4125 | } | |
4126 | ||
3a45bb20 | 4127 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 4128 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 4129 | trans = NULL; |
5d4f98a2 YZ |
4130 | |
4131 | if (rc->stage == MOVE_DATA_EXTENTS && | |
4132 | (flags & BTRFS_EXTENT_FLAG_DATA)) { | |
4133 | rc->found_file_extent = 1; | |
0257bb82 | 4134 | ret = relocate_data_extent(rc->data_inode, |
3fd0a558 | 4135 | &key, &rc->cluster); |
5d4f98a2 YZ |
4136 | if (ret < 0) { |
4137 | err = ret; | |
4138 | break; | |
4139 | } | |
4140 | } | |
4141 | } | |
c87f08ca | 4142 | if (trans && progress && err == -ENOSPC) { |
2ff7e61e | 4143 | ret = btrfs_force_chunk_alloc(trans, fs_info, |
c87f08ca | 4144 | rc->block_group->flags); |
9689457b | 4145 | if (ret == 1) { |
c87f08ca CM |
4146 | err = 0; |
4147 | progress = 0; | |
4148 | goto restart; | |
4149 | } | |
4150 | } | |
3fd0a558 | 4151 | |
b3b4aa74 | 4152 | btrfs_release_path(path); |
91166212 | 4153 | clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); |
5d4f98a2 YZ |
4154 | |
4155 | if (trans) { | |
3a45bb20 | 4156 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 4157 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
4158 | } |
4159 | ||
0257bb82 | 4160 | if (!err) { |
3fd0a558 YZ |
4161 | ret = relocate_file_extent_cluster(rc->data_inode, |
4162 | &rc->cluster); | |
0257bb82 YZ |
4163 | if (ret < 0) |
4164 | err = ret; | |
4165 | } | |
4166 | ||
3fd0a558 YZ |
4167 | rc->create_reloc_tree = 0; |
4168 | set_reloc_control(rc); | |
0257bb82 | 4169 | |
3fd0a558 | 4170 | backref_cache_cleanup(&rc->backref_cache); |
2ff7e61e | 4171 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); |
5d4f98a2 | 4172 | |
3fd0a558 | 4173 | err = prepare_to_merge(rc, err); |
5d4f98a2 YZ |
4174 | |
4175 | merge_reloc_roots(rc); | |
4176 | ||
3fd0a558 | 4177 | rc->merge_reloc_tree = 0; |
5d4f98a2 | 4178 | unset_reloc_control(rc); |
2ff7e61e | 4179 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); |
5d4f98a2 YZ |
4180 | |
4181 | /* get rid of pinned extents */ | |
7a7eaa40 | 4182 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 4183 | if (IS_ERR(trans)) { |
3612b495 | 4184 | err = PTR_ERR(trans); |
62b99540 QW |
4185 | goto out_free; |
4186 | } | |
3a45bb20 | 4187 | btrfs_commit_transaction(trans); |
3fd0a558 | 4188 | out_free: |
2ff7e61e | 4189 | btrfs_free_block_rsv(fs_info, rc->block_rsv); |
3fd0a558 | 4190 | btrfs_free_path(path); |
5d4f98a2 YZ |
4191 | return err; |
4192 | } | |
4193 | ||
4194 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, | |
0257bb82 | 4195 | struct btrfs_root *root, u64 objectid) |
5d4f98a2 YZ |
4196 | { |
4197 | struct btrfs_path *path; | |
4198 | struct btrfs_inode_item *item; | |
4199 | struct extent_buffer *leaf; | |
4200 | int ret; | |
4201 | ||
4202 | path = btrfs_alloc_path(); | |
4203 | if (!path) | |
4204 | return -ENOMEM; | |
4205 | ||
4206 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
4207 | if (ret) | |
4208 | goto out; | |
4209 | ||
4210 | leaf = path->nodes[0]; | |
4211 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
b159fa28 | 4212 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
5d4f98a2 | 4213 | btrfs_set_inode_generation(leaf, item, 1); |
0257bb82 | 4214 | btrfs_set_inode_size(leaf, item, 0); |
5d4f98a2 | 4215 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); |
3fd0a558 YZ |
4216 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | |
4217 | BTRFS_INODE_PREALLOC); | |
5d4f98a2 | 4218 | btrfs_mark_buffer_dirty(leaf); |
5d4f98a2 YZ |
4219 | out: |
4220 | btrfs_free_path(path); | |
4221 | return ret; | |
4222 | } | |
4223 | ||
4224 | /* | |
4225 | * helper to create inode for data relocation. | |
4226 | * the inode is in data relocation tree and its link count is 0 | |
4227 | */ | |
3fd0a558 YZ |
4228 | static noinline_for_stack |
4229 | struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, | |
4230 | struct btrfs_block_group_cache *group) | |
5d4f98a2 YZ |
4231 | { |
4232 | struct inode *inode = NULL; | |
4233 | struct btrfs_trans_handle *trans; | |
4234 | struct btrfs_root *root; | |
4235 | struct btrfs_key key; | |
4624900d | 4236 | u64 objectid; |
5d4f98a2 YZ |
4237 | int err = 0; |
4238 | ||
4239 | root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); | |
4240 | if (IS_ERR(root)) | |
4241 | return ERR_CAST(root); | |
4242 | ||
a22285a6 | 4243 | trans = btrfs_start_transaction(root, 6); |
3fd0a558 YZ |
4244 | if (IS_ERR(trans)) |
4245 | return ERR_CAST(trans); | |
5d4f98a2 | 4246 | |
581bb050 | 4247 | err = btrfs_find_free_objectid(root, &objectid); |
5d4f98a2 YZ |
4248 | if (err) |
4249 | goto out; | |
4250 | ||
0257bb82 | 4251 | err = __insert_orphan_inode(trans, root, objectid); |
5d4f98a2 YZ |
4252 | BUG_ON(err); |
4253 | ||
4254 | key.objectid = objectid; | |
4255 | key.type = BTRFS_INODE_ITEM_KEY; | |
4256 | key.offset = 0; | |
0b246afa | 4257 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
4258 | BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); |
4259 | BTRFS_I(inode)->index_cnt = group->key.objectid; | |
4260 | ||
73f2e545 | 4261 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
5d4f98a2 | 4262 | out: |
3a45bb20 | 4263 | btrfs_end_transaction(trans); |
2ff7e61e | 4264 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
4265 | if (err) { |
4266 | if (inode) | |
4267 | iput(inode); | |
4268 | inode = ERR_PTR(err); | |
4269 | } | |
4270 | return inode; | |
4271 | } | |
4272 | ||
a9995eec | 4273 | static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) |
3fd0a558 YZ |
4274 | { |
4275 | struct reloc_control *rc; | |
4276 | ||
4277 | rc = kzalloc(sizeof(*rc), GFP_NOFS); | |
4278 | if (!rc) | |
4279 | return NULL; | |
4280 | ||
4281 | INIT_LIST_HEAD(&rc->reloc_roots); | |
4282 | backref_cache_init(&rc->backref_cache); | |
4283 | mapping_tree_init(&rc->reloc_root_tree); | |
c6100a4b | 4284 | extent_io_tree_init(&rc->processed_blocks, NULL); |
3fd0a558 YZ |
4285 | return rc; |
4286 | } | |
4287 | ||
ebce0e01 AB |
4288 | /* |
4289 | * Print the block group being relocated | |
4290 | */ | |
4291 | static void describe_relocation(struct btrfs_fs_info *fs_info, | |
4292 | struct btrfs_block_group_cache *block_group) | |
4293 | { | |
4294 | char buf[128]; /* prefixed by a '|' that'll be dropped */ | |
4295 | u64 flags = block_group->flags; | |
4296 | ||
4297 | /* Shouldn't happen */ | |
4298 | if (!flags) { | |
4299 | strcpy(buf, "|NONE"); | |
4300 | } else { | |
4301 | char *bp = buf; | |
4302 | ||
4303 | #define DESCRIBE_FLAG(f, d) \ | |
4304 | if (flags & BTRFS_BLOCK_GROUP_##f) { \ | |
4305 | bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \ | |
4306 | flags &= ~BTRFS_BLOCK_GROUP_##f; \ | |
4307 | } | |
4308 | DESCRIBE_FLAG(DATA, "data"); | |
4309 | DESCRIBE_FLAG(SYSTEM, "system"); | |
4310 | DESCRIBE_FLAG(METADATA, "metadata"); | |
4311 | DESCRIBE_FLAG(RAID0, "raid0"); | |
4312 | DESCRIBE_FLAG(RAID1, "raid1"); | |
4313 | DESCRIBE_FLAG(DUP, "dup"); | |
4314 | DESCRIBE_FLAG(RAID10, "raid10"); | |
4315 | DESCRIBE_FLAG(RAID5, "raid5"); | |
4316 | DESCRIBE_FLAG(RAID6, "raid6"); | |
4317 | if (flags) | |
4318 | snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags); | |
4319 | #undef DESCRIBE_FLAG | |
4320 | } | |
4321 | ||
4322 | btrfs_info(fs_info, | |
4323 | "relocating block group %llu flags %s", | |
4324 | block_group->key.objectid, buf + 1); | |
4325 | } | |
4326 | ||
5d4f98a2 YZ |
4327 | /* |
4328 | * function to relocate all extents in a block group. | |
4329 | */ | |
6bccf3ab | 4330 | int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) |
5d4f98a2 | 4331 | { |
6bccf3ab | 4332 | struct btrfs_root *extent_root = fs_info->extent_root; |
5d4f98a2 | 4333 | struct reloc_control *rc; |
0af3d00b JB |
4334 | struct inode *inode; |
4335 | struct btrfs_path *path; | |
5d4f98a2 | 4336 | int ret; |
f0486c68 | 4337 | int rw = 0; |
5d4f98a2 YZ |
4338 | int err = 0; |
4339 | ||
a9995eec | 4340 | rc = alloc_reloc_control(fs_info); |
5d4f98a2 YZ |
4341 | if (!rc) |
4342 | return -ENOMEM; | |
4343 | ||
f0486c68 | 4344 | rc->extent_root = extent_root; |
3fd0a558 | 4345 | |
5d4f98a2 YZ |
4346 | rc->block_group = btrfs_lookup_block_group(fs_info, group_start); |
4347 | BUG_ON(!rc->block_group); | |
4348 | ||
5e00f193 | 4349 | ret = btrfs_inc_block_group_ro(fs_info, rc->block_group); |
868f401a Z |
4350 | if (ret) { |
4351 | err = ret; | |
4352 | goto out; | |
f0486c68 | 4353 | } |
868f401a | 4354 | rw = 1; |
f0486c68 | 4355 | |
0af3d00b JB |
4356 | path = btrfs_alloc_path(); |
4357 | if (!path) { | |
4358 | err = -ENOMEM; | |
4359 | goto out; | |
4360 | } | |
4361 | ||
77ab86bf | 4362 | inode = lookup_free_space_inode(fs_info, rc->block_group, path); |
0af3d00b JB |
4363 | btrfs_free_path(path); |
4364 | ||
4365 | if (!IS_ERR(inode)) | |
1bbc621e | 4366 | ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); |
0af3d00b JB |
4367 | else |
4368 | ret = PTR_ERR(inode); | |
4369 | ||
4370 | if (ret && ret != -ENOENT) { | |
4371 | err = ret; | |
4372 | goto out; | |
4373 | } | |
4374 | ||
5d4f98a2 YZ |
4375 | rc->data_inode = create_reloc_inode(fs_info, rc->block_group); |
4376 | if (IS_ERR(rc->data_inode)) { | |
4377 | err = PTR_ERR(rc->data_inode); | |
4378 | rc->data_inode = NULL; | |
4379 | goto out; | |
4380 | } | |
4381 | ||
0b246afa | 4382 | describe_relocation(fs_info, rc->block_group); |
5d4f98a2 | 4383 | |
9cfa3e34 | 4384 | btrfs_wait_block_group_reservations(rc->block_group); |
f78c436c | 4385 | btrfs_wait_nocow_writers(rc->block_group); |
6374e57a | 4386 | btrfs_wait_ordered_roots(fs_info, U64_MAX, |
578def7c FM |
4387 | rc->block_group->key.objectid, |
4388 | rc->block_group->key.offset); | |
5d4f98a2 YZ |
4389 | |
4390 | while (1) { | |
76dda93c | 4391 | mutex_lock(&fs_info->cleaner_mutex); |
5d4f98a2 | 4392 | ret = relocate_block_group(rc); |
76dda93c | 4393 | mutex_unlock(&fs_info->cleaner_mutex); |
5d4f98a2 YZ |
4394 | if (ret < 0) { |
4395 | err = ret; | |
3fd0a558 | 4396 | goto out; |
5d4f98a2 YZ |
4397 | } |
4398 | ||
4399 | if (rc->extents_found == 0) | |
4400 | break; | |
4401 | ||
0b246afa | 4402 | btrfs_info(fs_info, "found %llu extents", rc->extents_found); |
5d4f98a2 YZ |
4403 | |
4404 | if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { | |
0ef8b726 JB |
4405 | ret = btrfs_wait_ordered_range(rc->data_inode, 0, |
4406 | (u64)-1); | |
4407 | if (ret) { | |
4408 | err = ret; | |
4409 | goto out; | |
4410 | } | |
5d4f98a2 YZ |
4411 | invalidate_mapping_pages(rc->data_inode->i_mapping, |
4412 | 0, -1); | |
4413 | rc->stage = UPDATE_DATA_PTRS; | |
5d4f98a2 YZ |
4414 | } |
4415 | } | |
4416 | ||
5d4f98a2 YZ |
4417 | WARN_ON(rc->block_group->pinned > 0); |
4418 | WARN_ON(rc->block_group->reserved > 0); | |
4419 | WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); | |
4420 | out: | |
f0486c68 | 4421 | if (err && rw) |
2ff7e61e | 4422 | btrfs_dec_block_group_ro(rc->block_group); |
5d4f98a2 | 4423 | iput(rc->data_inode); |
5d4f98a2 YZ |
4424 | btrfs_put_block_group(rc->block_group); |
4425 | kfree(rc); | |
4426 | return err; | |
4427 | } | |
4428 | ||
76dda93c YZ |
4429 | static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) |
4430 | { | |
0b246afa | 4431 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c | 4432 | struct btrfs_trans_handle *trans; |
79787eaa | 4433 | int ret, err; |
76dda93c | 4434 | |
0b246afa | 4435 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
4436 | if (IS_ERR(trans)) |
4437 | return PTR_ERR(trans); | |
76dda93c YZ |
4438 | |
4439 | memset(&root->root_item.drop_progress, 0, | |
4440 | sizeof(root->root_item.drop_progress)); | |
4441 | root->root_item.drop_level = 0; | |
4442 | btrfs_set_root_refs(&root->root_item, 0); | |
0b246afa | 4443 | ret = btrfs_update_root(trans, fs_info->tree_root, |
76dda93c | 4444 | &root->root_key, &root->root_item); |
76dda93c | 4445 | |
3a45bb20 | 4446 | err = btrfs_end_transaction(trans); |
79787eaa JM |
4447 | if (err) |
4448 | return err; | |
4449 | return ret; | |
76dda93c YZ |
4450 | } |
4451 | ||
5d4f98a2 YZ |
4452 | /* |
4453 | * recover relocation interrupted by system crash. | |
4454 | * | |
4455 | * this function resumes merging reloc trees with corresponding fs trees. | |
4456 | * this is important for keeping the sharing of tree blocks | |
4457 | */ | |
4458 | int btrfs_recover_relocation(struct btrfs_root *root) | |
4459 | { | |
0b246afa | 4460 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
4461 | LIST_HEAD(reloc_roots); |
4462 | struct btrfs_key key; | |
4463 | struct btrfs_root *fs_root; | |
4464 | struct btrfs_root *reloc_root; | |
4465 | struct btrfs_path *path; | |
4466 | struct extent_buffer *leaf; | |
4467 | struct reloc_control *rc = NULL; | |
4468 | struct btrfs_trans_handle *trans; | |
4469 | int ret; | |
4470 | int err = 0; | |
4471 | ||
4472 | path = btrfs_alloc_path(); | |
4473 | if (!path) | |
4474 | return -ENOMEM; | |
e4058b54 | 4475 | path->reada = READA_BACK; |
5d4f98a2 YZ |
4476 | |
4477 | key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
4478 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4479 | key.offset = (u64)-1; | |
4480 | ||
4481 | while (1) { | |
0b246afa | 4482 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, |
5d4f98a2 YZ |
4483 | path, 0, 0); |
4484 | if (ret < 0) { | |
4485 | err = ret; | |
4486 | goto out; | |
4487 | } | |
4488 | if (ret > 0) { | |
4489 | if (path->slots[0] == 0) | |
4490 | break; | |
4491 | path->slots[0]--; | |
4492 | } | |
4493 | leaf = path->nodes[0]; | |
4494 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4495 | btrfs_release_path(path); |
5d4f98a2 YZ |
4496 | |
4497 | if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || | |
4498 | key.type != BTRFS_ROOT_ITEM_KEY) | |
4499 | break; | |
4500 | ||
cb517eab | 4501 | reloc_root = btrfs_read_fs_root(root, &key); |
5d4f98a2 YZ |
4502 | if (IS_ERR(reloc_root)) { |
4503 | err = PTR_ERR(reloc_root); | |
4504 | goto out; | |
4505 | } | |
4506 | ||
4507 | list_add(&reloc_root->root_list, &reloc_roots); | |
4508 | ||
4509 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { | |
0b246afa | 4510 | fs_root = read_fs_root(fs_info, |
5d4f98a2 YZ |
4511 | reloc_root->root_key.offset); |
4512 | if (IS_ERR(fs_root)) { | |
76dda93c YZ |
4513 | ret = PTR_ERR(fs_root); |
4514 | if (ret != -ENOENT) { | |
4515 | err = ret; | |
4516 | goto out; | |
4517 | } | |
79787eaa JM |
4518 | ret = mark_garbage_root(reloc_root); |
4519 | if (ret < 0) { | |
4520 | err = ret; | |
4521 | goto out; | |
4522 | } | |
5d4f98a2 YZ |
4523 | } |
4524 | } | |
4525 | ||
4526 | if (key.offset == 0) | |
4527 | break; | |
4528 | ||
4529 | key.offset--; | |
4530 | } | |
b3b4aa74 | 4531 | btrfs_release_path(path); |
5d4f98a2 YZ |
4532 | |
4533 | if (list_empty(&reloc_roots)) | |
4534 | goto out; | |
4535 | ||
0b246afa | 4536 | rc = alloc_reloc_control(fs_info); |
5d4f98a2 YZ |
4537 | if (!rc) { |
4538 | err = -ENOMEM; | |
4539 | goto out; | |
4540 | } | |
4541 | ||
0b246afa | 4542 | rc->extent_root = fs_info->extent_root; |
5d4f98a2 YZ |
4543 | |
4544 | set_reloc_control(rc); | |
4545 | ||
7a7eaa40 | 4546 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 TI |
4547 | if (IS_ERR(trans)) { |
4548 | unset_reloc_control(rc); | |
4549 | err = PTR_ERR(trans); | |
4550 | goto out_free; | |
4551 | } | |
3fd0a558 YZ |
4552 | |
4553 | rc->merge_reloc_tree = 1; | |
4554 | ||
5d4f98a2 YZ |
4555 | while (!list_empty(&reloc_roots)) { |
4556 | reloc_root = list_entry(reloc_roots.next, | |
4557 | struct btrfs_root, root_list); | |
4558 | list_del(&reloc_root->root_list); | |
4559 | ||
4560 | if (btrfs_root_refs(&reloc_root->root_item) == 0) { | |
4561 | list_add_tail(&reloc_root->root_list, | |
4562 | &rc->reloc_roots); | |
4563 | continue; | |
4564 | } | |
4565 | ||
0b246afa | 4566 | fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); |
79787eaa JM |
4567 | if (IS_ERR(fs_root)) { |
4568 | err = PTR_ERR(fs_root); | |
4569 | goto out_free; | |
4570 | } | |
5d4f98a2 | 4571 | |
ffd7b339 | 4572 | err = __add_reloc_root(reloc_root); |
79787eaa | 4573 | BUG_ON(err < 0); /* -ENOMEM or logic error */ |
5d4f98a2 YZ |
4574 | fs_root->reloc_root = reloc_root; |
4575 | } | |
4576 | ||
3a45bb20 | 4577 | err = btrfs_commit_transaction(trans); |
79787eaa JM |
4578 | if (err) |
4579 | goto out_free; | |
5d4f98a2 YZ |
4580 | |
4581 | merge_reloc_roots(rc); | |
4582 | ||
4583 | unset_reloc_control(rc); | |
4584 | ||
7a7eaa40 | 4585 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 4586 | if (IS_ERR(trans)) { |
3612b495 | 4587 | err = PTR_ERR(trans); |
62b99540 QW |
4588 | goto out_free; |
4589 | } | |
3a45bb20 | 4590 | err = btrfs_commit_transaction(trans); |
3612b495 | 4591 | out_free: |
3fd0a558 | 4592 | kfree(rc); |
3612b495 | 4593 | out: |
aca1bba6 LB |
4594 | if (!list_empty(&reloc_roots)) |
4595 | free_reloc_roots(&reloc_roots); | |
4596 | ||
5d4f98a2 YZ |
4597 | btrfs_free_path(path); |
4598 | ||
4599 | if (err == 0) { | |
4600 | /* cleanup orphan inode in data relocation tree */ | |
0b246afa | 4601 | fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); |
5d4f98a2 YZ |
4602 | if (IS_ERR(fs_root)) |
4603 | err = PTR_ERR(fs_root); | |
d7ce5843 | 4604 | else |
66b4ffd1 | 4605 | err = btrfs_orphan_cleanup(fs_root); |
5d4f98a2 YZ |
4606 | } |
4607 | return err; | |
4608 | } | |
4609 | ||
4610 | /* | |
4611 | * helper to add ordered checksum for data relocation. | |
4612 | * | |
4613 | * cloning checksum properly handles the nodatasum extents. | |
4614 | * it also saves CPU time to re-calculate the checksum. | |
4615 | */ | |
4616 | int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) | |
4617 | { | |
0b246afa | 4618 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 4619 | struct btrfs_ordered_sum *sums; |
5d4f98a2 | 4620 | struct btrfs_ordered_extent *ordered; |
5d4f98a2 YZ |
4621 | int ret; |
4622 | u64 disk_bytenr; | |
4577b014 | 4623 | u64 new_bytenr; |
5d4f98a2 YZ |
4624 | LIST_HEAD(list); |
4625 | ||
4626 | ordered = btrfs_lookup_ordered_extent(inode, file_pos); | |
4627 | BUG_ON(ordered->file_offset != file_pos || ordered->len != len); | |
4628 | ||
4629 | disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; | |
0b246afa | 4630 | ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, |
a2de733c | 4631 | disk_bytenr + len - 1, &list, 0); |
79787eaa JM |
4632 | if (ret) |
4633 | goto out; | |
5d4f98a2 YZ |
4634 | |
4635 | while (!list_empty(&list)) { | |
4636 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
4637 | list_del_init(&sums->list); | |
4638 | ||
4577b014 JB |
4639 | /* |
4640 | * We need to offset the new_bytenr based on where the csum is. | |
4641 | * We need to do this because we will read in entire prealloc | |
4642 | * extents but we may have written to say the middle of the | |
4643 | * prealloc extent, so we need to make sure the csum goes with | |
4644 | * the right disk offset. | |
4645 | * | |
4646 | * We can do this because the data reloc inode refers strictly | |
4647 | * to the on disk bytes, so we don't have to worry about | |
4648 | * disk_len vs real len like with real inodes since it's all | |
4649 | * disk length. | |
4650 | */ | |
4651 | new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); | |
4652 | sums->bytenr = new_bytenr; | |
5d4f98a2 YZ |
4653 | |
4654 | btrfs_add_ordered_sum(inode, ordered, sums); | |
4655 | } | |
79787eaa | 4656 | out: |
5d4f98a2 | 4657 | btrfs_put_ordered_extent(ordered); |
411fc6bc | 4658 | return ret; |
5d4f98a2 | 4659 | } |
3fd0a558 | 4660 | |
83d4cfd4 JB |
4661 | int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, |
4662 | struct btrfs_root *root, struct extent_buffer *buf, | |
4663 | struct extent_buffer *cow) | |
3fd0a558 | 4664 | { |
0b246afa | 4665 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 YZ |
4666 | struct reloc_control *rc; |
4667 | struct backref_node *node; | |
4668 | int first_cow = 0; | |
4669 | int level; | |
83d4cfd4 | 4670 | int ret = 0; |
3fd0a558 | 4671 | |
0b246afa | 4672 | rc = fs_info->reloc_ctl; |
3fd0a558 | 4673 | if (!rc) |
83d4cfd4 | 4674 | return 0; |
3fd0a558 YZ |
4675 | |
4676 | BUG_ON(rc->stage == UPDATE_DATA_PTRS && | |
4677 | root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); | |
4678 | ||
c974c464 WS |
4679 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { |
4680 | if (buf == root->node) | |
4681 | __update_reloc_root(root, cow->start); | |
4682 | } | |
4683 | ||
3fd0a558 YZ |
4684 | level = btrfs_header_level(buf); |
4685 | if (btrfs_header_generation(buf) <= | |
4686 | btrfs_root_last_snapshot(&root->root_item)) | |
4687 | first_cow = 1; | |
4688 | ||
4689 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && | |
4690 | rc->create_reloc_tree) { | |
4691 | WARN_ON(!first_cow && level == 0); | |
4692 | ||
4693 | node = rc->backref_cache.path[level]; | |
4694 | BUG_ON(node->bytenr != buf->start && | |
4695 | node->new_bytenr != buf->start); | |
4696 | ||
4697 | drop_node_buffer(node); | |
4698 | extent_buffer_get(cow); | |
4699 | node->eb = cow; | |
4700 | node->new_bytenr = cow->start; | |
4701 | ||
4702 | if (!node->pending) { | |
4703 | list_move_tail(&node->list, | |
4704 | &rc->backref_cache.pending[level]); | |
4705 | node->pending = 1; | |
4706 | } | |
4707 | ||
4708 | if (first_cow) | |
4709 | __mark_block_processed(rc, node); | |
4710 | ||
4711 | if (first_cow && level > 0) | |
4712 | rc->nodes_relocated += buf->len; | |
4713 | } | |
4714 | ||
83d4cfd4 | 4715 | if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) |
3fd0a558 | 4716 | ret = replace_file_extents(trans, rc, root, cow); |
83d4cfd4 | 4717 | return ret; |
3fd0a558 YZ |
4718 | } |
4719 | ||
4720 | /* | |
4721 | * called before creating snapshot. it calculates metadata reservation | |
01327610 | 4722 | * required for relocating tree blocks in the snapshot |
3fd0a558 | 4723 | */ |
147d256e | 4724 | void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, |
3fd0a558 YZ |
4725 | u64 *bytes_to_reserve) |
4726 | { | |
4727 | struct btrfs_root *root; | |
4728 | struct reloc_control *rc; | |
4729 | ||
4730 | root = pending->root; | |
4731 | if (!root->reloc_root) | |
4732 | return; | |
4733 | ||
4734 | rc = root->fs_info->reloc_ctl; | |
4735 | if (!rc->merge_reloc_tree) | |
4736 | return; | |
4737 | ||
4738 | root = root->reloc_root; | |
4739 | BUG_ON(btrfs_root_refs(&root->root_item) == 0); | |
4740 | /* | |
4741 | * relocation is in the stage of merging trees. the space | |
4742 | * used by merging a reloc tree is twice the size of | |
4743 | * relocated tree nodes in the worst case. half for cowing | |
4744 | * the reloc tree, half for cowing the fs tree. the space | |
4745 | * used by cowing the reloc tree will be freed after the | |
4746 | * tree is dropped. if we create snapshot, cowing the fs | |
4747 | * tree may use more space than it frees. so we need | |
4748 | * reserve extra space. | |
4749 | */ | |
4750 | *bytes_to_reserve += rc->nodes_relocated; | |
4751 | } | |
4752 | ||
4753 | /* | |
4754 | * called after snapshot is created. migrate block reservation | |
4755 | * and create reloc root for the newly created snapshot | |
4756 | */ | |
49b25e05 | 4757 | int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, |
3fd0a558 YZ |
4758 | struct btrfs_pending_snapshot *pending) |
4759 | { | |
4760 | struct btrfs_root *root = pending->root; | |
4761 | struct btrfs_root *reloc_root; | |
4762 | struct btrfs_root *new_root; | |
4763 | struct reloc_control *rc; | |
4764 | int ret; | |
4765 | ||
4766 | if (!root->reloc_root) | |
49b25e05 | 4767 | return 0; |
3fd0a558 YZ |
4768 | |
4769 | rc = root->fs_info->reloc_ctl; | |
4770 | rc->merging_rsv_size += rc->nodes_relocated; | |
4771 | ||
4772 | if (rc->merge_reloc_tree) { | |
4773 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
4774 | rc->block_rsv, | |
25d609f8 | 4775 | rc->nodes_relocated, 1); |
49b25e05 JM |
4776 | if (ret) |
4777 | return ret; | |
3fd0a558 YZ |
4778 | } |
4779 | ||
4780 | new_root = pending->snap; | |
4781 | reloc_root = create_reloc_root(trans, root->reloc_root, | |
4782 | new_root->root_key.objectid); | |
49b25e05 JM |
4783 | if (IS_ERR(reloc_root)) |
4784 | return PTR_ERR(reloc_root); | |
3fd0a558 | 4785 | |
ffd7b339 JM |
4786 | ret = __add_reloc_root(reloc_root); |
4787 | BUG_ON(ret < 0); | |
3fd0a558 YZ |
4788 | new_root->reloc_root = reloc_root; |
4789 | ||
49b25e05 | 4790 | if (rc->create_reloc_tree) |
3fd0a558 | 4791 | ret = clone_backref_node(trans, rc, root, reloc_root); |
49b25e05 | 4792 | return ret; |
3fd0a558 | 4793 | } |