Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
fd708b81 JB |
2 | /* |
3 | * Copyright (C) 2014 Facebook. All rights reserved. | |
fd708b81 JB |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
7 | #include <linux/stacktrace.h> | |
9b569ea0 | 8 | #include "messages.h" |
fd708b81 JB |
9 | #include "ctree.h" |
10 | #include "disk-io.h" | |
11 | #include "locking.h" | |
12 | #include "delayed-ref.h" | |
13 | #include "ref-verify.h" | |
fc97a410 | 14 | #include "fs.h" |
fd708b81 JB |
15 | |
16 | /* | |
17 | * Used to keep track the roots and number of refs each root has for a given | |
18 | * bytenr. This just tracks the number of direct references, no shared | |
19 | * references. | |
20 | */ | |
21 | struct root_entry { | |
22 | u64 root_objectid; | |
23 | u64 num_refs; | |
24 | struct rb_node node; | |
25 | }; | |
26 | ||
27 | /* | |
28 | * These are meant to represent what should exist in the extent tree, these can | |
29 | * be used to verify the extent tree is consistent as these should all match | |
30 | * what the extent tree says. | |
31 | */ | |
32 | struct ref_entry { | |
33 | u64 root_objectid; | |
34 | u64 parent; | |
35 | u64 owner; | |
36 | u64 offset; | |
37 | u64 num_refs; | |
38 | struct rb_node node; | |
39 | }; | |
40 | ||
41 | #define MAX_TRACE 16 | |
42 | ||
43 | /* | |
44 | * Whenever we add/remove a reference we record the action. The action maps | |
45 | * back to the delayed ref action. We hold the ref we are changing in the | |
46 | * action so we can account for the history properly, and we record the root we | |
47 | * were called with since it could be different from ref_root. We also store | |
52042d8e | 48 | * stack traces because that's how I roll. |
fd708b81 JB |
49 | */ |
50 | struct ref_action { | |
51 | int action; | |
52 | u64 root; | |
53 | struct ref_entry ref; | |
54 | struct list_head list; | |
55 | unsigned long trace[MAX_TRACE]; | |
56 | unsigned int trace_len; | |
57 | }; | |
58 | ||
59 | /* | |
60 | * One of these for every block we reference, it holds the roots and references | |
52042d8e | 61 | * to it as well as all of the ref actions that have occurred to it. We never |
fd708b81 JB |
62 | * free it until we unmount the file system in order to make sure re-allocations |
63 | * are happening properly. | |
64 | */ | |
65 | struct block_entry { | |
66 | u64 bytenr; | |
67 | u64 len; | |
68 | u64 num_refs; | |
69 | int metadata; | |
70 | int from_disk; | |
71 | struct rb_root roots; | |
72 | struct rb_root refs; | |
73 | struct rb_node node; | |
74 | struct list_head actions; | |
75 | }; | |
76 | ||
77 | static struct block_entry *insert_block_entry(struct rb_root *root, | |
78 | struct block_entry *be) | |
79 | { | |
80 | struct rb_node **p = &root->rb_node; | |
81 | struct rb_node *parent_node = NULL; | |
82 | struct block_entry *entry; | |
83 | ||
84 | while (*p) { | |
85 | parent_node = *p; | |
86 | entry = rb_entry(parent_node, struct block_entry, node); | |
87 | if (entry->bytenr > be->bytenr) | |
88 | p = &(*p)->rb_left; | |
89 | else if (entry->bytenr < be->bytenr) | |
90 | p = &(*p)->rb_right; | |
91 | else | |
92 | return entry; | |
93 | } | |
94 | ||
95 | rb_link_node(&be->node, parent_node, p); | |
96 | rb_insert_color(&be->node, root); | |
97 | return NULL; | |
98 | } | |
99 | ||
100 | static struct block_entry *lookup_block_entry(struct rb_root *root, u64 bytenr) | |
101 | { | |
102 | struct rb_node *n; | |
103 | struct block_entry *entry = NULL; | |
104 | ||
105 | n = root->rb_node; | |
106 | while (n) { | |
107 | entry = rb_entry(n, struct block_entry, node); | |
108 | if (entry->bytenr < bytenr) | |
109 | n = n->rb_right; | |
110 | else if (entry->bytenr > bytenr) | |
111 | n = n->rb_left; | |
112 | else | |
113 | return entry; | |
114 | } | |
115 | return NULL; | |
116 | } | |
117 | ||
118 | static struct root_entry *insert_root_entry(struct rb_root *root, | |
119 | struct root_entry *re) | |
120 | { | |
121 | struct rb_node **p = &root->rb_node; | |
122 | struct rb_node *parent_node = NULL; | |
123 | struct root_entry *entry; | |
124 | ||
125 | while (*p) { | |
126 | parent_node = *p; | |
127 | entry = rb_entry(parent_node, struct root_entry, node); | |
128 | if (entry->root_objectid > re->root_objectid) | |
129 | p = &(*p)->rb_left; | |
130 | else if (entry->root_objectid < re->root_objectid) | |
131 | p = &(*p)->rb_right; | |
132 | else | |
133 | return entry; | |
134 | } | |
135 | ||
136 | rb_link_node(&re->node, parent_node, p); | |
137 | rb_insert_color(&re->node, root); | |
138 | return NULL; | |
139 | ||
140 | } | |
141 | ||
142 | static int comp_refs(struct ref_entry *ref1, struct ref_entry *ref2) | |
143 | { | |
144 | if (ref1->root_objectid < ref2->root_objectid) | |
145 | return -1; | |
146 | if (ref1->root_objectid > ref2->root_objectid) | |
147 | return 1; | |
148 | if (ref1->parent < ref2->parent) | |
149 | return -1; | |
150 | if (ref1->parent > ref2->parent) | |
151 | return 1; | |
152 | if (ref1->owner < ref2->owner) | |
153 | return -1; | |
154 | if (ref1->owner > ref2->owner) | |
155 | return 1; | |
156 | if (ref1->offset < ref2->offset) | |
157 | return -1; | |
158 | if (ref1->offset > ref2->offset) | |
159 | return 1; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static struct ref_entry *insert_ref_entry(struct rb_root *root, | |
164 | struct ref_entry *ref) | |
165 | { | |
166 | struct rb_node **p = &root->rb_node; | |
167 | struct rb_node *parent_node = NULL; | |
168 | struct ref_entry *entry; | |
169 | int cmp; | |
170 | ||
171 | while (*p) { | |
172 | parent_node = *p; | |
173 | entry = rb_entry(parent_node, struct ref_entry, node); | |
174 | cmp = comp_refs(entry, ref); | |
175 | if (cmp > 0) | |
176 | p = &(*p)->rb_left; | |
177 | else if (cmp < 0) | |
178 | p = &(*p)->rb_right; | |
179 | else | |
180 | return entry; | |
181 | } | |
182 | ||
183 | rb_link_node(&ref->node, parent_node, p); | |
184 | rb_insert_color(&ref->node, root); | |
185 | return NULL; | |
186 | ||
187 | } | |
188 | ||
189 | static struct root_entry *lookup_root_entry(struct rb_root *root, u64 objectid) | |
190 | { | |
191 | struct rb_node *n; | |
192 | struct root_entry *entry = NULL; | |
193 | ||
194 | n = root->rb_node; | |
195 | while (n) { | |
196 | entry = rb_entry(n, struct root_entry, node); | |
197 | if (entry->root_objectid < objectid) | |
198 | n = n->rb_right; | |
199 | else if (entry->root_objectid > objectid) | |
200 | n = n->rb_left; | |
201 | else | |
202 | return entry; | |
203 | } | |
204 | return NULL; | |
205 | } | |
206 | ||
207 | #ifdef CONFIG_STACKTRACE | |
208 | static void __save_stack_trace(struct ref_action *ra) | |
209 | { | |
6924f5fe | 210 | ra->trace_len = stack_trace_save(ra->trace, MAX_TRACE, 2); |
fd708b81 JB |
211 | } |
212 | ||
213 | static void __print_stack_trace(struct btrfs_fs_info *fs_info, | |
214 | struct ref_action *ra) | |
215 | { | |
fd708b81 JB |
216 | if (ra->trace_len == 0) { |
217 | btrfs_err(fs_info, " ref-verify: no stacktrace"); | |
218 | return; | |
219 | } | |
6924f5fe | 220 | stack_trace_print(ra->trace, ra->trace_len, 2); |
fd708b81 JB |
221 | } |
222 | #else | |
aedb9d90 | 223 | static inline void __save_stack_trace(struct ref_action *ra) |
fd708b81 JB |
224 | { |
225 | } | |
226 | ||
aedb9d90 | 227 | static inline void __print_stack_trace(struct btrfs_fs_info *fs_info, |
fd708b81 JB |
228 | struct ref_action *ra) |
229 | { | |
230 | btrfs_err(fs_info, " ref-verify: no stacktrace support"); | |
231 | } | |
232 | #endif | |
233 | ||
234 | static void free_block_entry(struct block_entry *be) | |
235 | { | |
236 | struct root_entry *re; | |
237 | struct ref_entry *ref; | |
238 | struct ref_action *ra; | |
239 | struct rb_node *n; | |
240 | ||
241 | while ((n = rb_first(&be->roots))) { | |
242 | re = rb_entry(n, struct root_entry, node); | |
243 | rb_erase(&re->node, &be->roots); | |
244 | kfree(re); | |
245 | } | |
246 | ||
247 | while((n = rb_first(&be->refs))) { | |
248 | ref = rb_entry(n, struct ref_entry, node); | |
249 | rb_erase(&ref->node, &be->refs); | |
250 | kfree(ref); | |
251 | } | |
252 | ||
253 | while (!list_empty(&be->actions)) { | |
254 | ra = list_first_entry(&be->actions, struct ref_action, | |
255 | list); | |
256 | list_del(&ra->list); | |
257 | kfree(ra); | |
258 | } | |
259 | kfree(be); | |
260 | } | |
261 | ||
262 | static struct block_entry *add_block_entry(struct btrfs_fs_info *fs_info, | |
263 | u64 bytenr, u64 len, | |
264 | u64 root_objectid) | |
265 | { | |
266 | struct block_entry *be = NULL, *exist; | |
267 | struct root_entry *re = NULL; | |
268 | ||
5a656c36 FM |
269 | re = kzalloc(sizeof(struct root_entry), GFP_NOFS); |
270 | be = kzalloc(sizeof(struct block_entry), GFP_NOFS); | |
fd708b81 JB |
271 | if (!be || !re) { |
272 | kfree(re); | |
273 | kfree(be); | |
274 | return ERR_PTR(-ENOMEM); | |
275 | } | |
276 | be->bytenr = bytenr; | |
277 | be->len = len; | |
278 | ||
279 | re->root_objectid = root_objectid; | |
280 | re->num_refs = 0; | |
281 | ||
282 | spin_lock(&fs_info->ref_verify_lock); | |
283 | exist = insert_block_entry(&fs_info->block_tree, be); | |
284 | if (exist) { | |
285 | if (root_objectid) { | |
286 | struct root_entry *exist_re; | |
287 | ||
288 | exist_re = insert_root_entry(&exist->roots, re); | |
289 | if (exist_re) | |
290 | kfree(re); | |
d60ba8de TR |
291 | } else { |
292 | kfree(re); | |
fd708b81 JB |
293 | } |
294 | kfree(be); | |
295 | return exist; | |
296 | } | |
297 | ||
298 | be->num_refs = 0; | |
299 | be->metadata = 0; | |
300 | be->from_disk = 0; | |
301 | be->roots = RB_ROOT; | |
302 | be->refs = RB_ROOT; | |
303 | INIT_LIST_HEAD(&be->actions); | |
304 | if (root_objectid) | |
305 | insert_root_entry(&be->roots, re); | |
306 | else | |
307 | kfree(re); | |
308 | return be; | |
309 | } | |
310 | ||
311 | static int add_tree_block(struct btrfs_fs_info *fs_info, u64 ref_root, | |
312 | u64 parent, u64 bytenr, int level) | |
313 | { | |
314 | struct block_entry *be; | |
315 | struct root_entry *re; | |
316 | struct ref_entry *ref = NULL, *exist; | |
317 | ||
5a656c36 | 318 | ref = kmalloc(sizeof(struct ref_entry), GFP_NOFS); |
fd708b81 JB |
319 | if (!ref) |
320 | return -ENOMEM; | |
321 | ||
322 | if (parent) | |
323 | ref->root_objectid = 0; | |
324 | else | |
325 | ref->root_objectid = ref_root; | |
326 | ref->parent = parent; | |
327 | ref->owner = level; | |
328 | ref->offset = 0; | |
329 | ref->num_refs = 1; | |
330 | ||
331 | be = add_block_entry(fs_info, bytenr, fs_info->nodesize, ref_root); | |
332 | if (IS_ERR(be)) { | |
333 | kfree(ref); | |
334 | return PTR_ERR(be); | |
335 | } | |
336 | be->num_refs++; | |
337 | be->from_disk = 1; | |
338 | be->metadata = 1; | |
339 | ||
340 | if (!parent) { | |
341 | ASSERT(ref_root); | |
342 | re = lookup_root_entry(&be->roots, ref_root); | |
343 | ASSERT(re); | |
344 | re->num_refs++; | |
345 | } | |
346 | exist = insert_ref_entry(&be->refs, ref); | |
347 | if (exist) { | |
348 | exist->num_refs++; | |
349 | kfree(ref); | |
350 | } | |
351 | spin_unlock(&fs_info->ref_verify_lock); | |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
356 | static int add_shared_data_ref(struct btrfs_fs_info *fs_info, | |
357 | u64 parent, u32 num_refs, u64 bytenr, | |
358 | u64 num_bytes) | |
359 | { | |
360 | struct block_entry *be; | |
361 | struct ref_entry *ref; | |
362 | ||
5a656c36 | 363 | ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS); |
fd708b81 JB |
364 | if (!ref) |
365 | return -ENOMEM; | |
366 | be = add_block_entry(fs_info, bytenr, num_bytes, 0); | |
367 | if (IS_ERR(be)) { | |
368 | kfree(ref); | |
369 | return PTR_ERR(be); | |
370 | } | |
371 | be->num_refs += num_refs; | |
372 | ||
373 | ref->parent = parent; | |
374 | ref->num_refs = num_refs; | |
375 | if (insert_ref_entry(&be->refs, ref)) { | |
376 | spin_unlock(&fs_info->ref_verify_lock); | |
377 | btrfs_err(fs_info, "existing shared ref when reading from disk?"); | |
378 | kfree(ref); | |
379 | return -EINVAL; | |
380 | } | |
381 | spin_unlock(&fs_info->ref_verify_lock); | |
382 | return 0; | |
383 | } | |
384 | ||
385 | static int add_extent_data_ref(struct btrfs_fs_info *fs_info, | |
386 | struct extent_buffer *leaf, | |
387 | struct btrfs_extent_data_ref *dref, | |
388 | u64 bytenr, u64 num_bytes) | |
389 | { | |
390 | struct block_entry *be; | |
391 | struct ref_entry *ref; | |
392 | struct root_entry *re; | |
393 | u64 ref_root = btrfs_extent_data_ref_root(leaf, dref); | |
394 | u64 owner = btrfs_extent_data_ref_objectid(leaf, dref); | |
395 | u64 offset = btrfs_extent_data_ref_offset(leaf, dref); | |
396 | u32 num_refs = btrfs_extent_data_ref_count(leaf, dref); | |
397 | ||
5a656c36 | 398 | ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS); |
fd708b81 JB |
399 | if (!ref) |
400 | return -ENOMEM; | |
401 | be = add_block_entry(fs_info, bytenr, num_bytes, ref_root); | |
402 | if (IS_ERR(be)) { | |
403 | kfree(ref); | |
404 | return PTR_ERR(be); | |
405 | } | |
406 | be->num_refs += num_refs; | |
407 | ||
408 | ref->parent = 0; | |
409 | ref->owner = owner; | |
410 | ref->root_objectid = ref_root; | |
411 | ref->offset = offset; | |
412 | ref->num_refs = num_refs; | |
413 | if (insert_ref_entry(&be->refs, ref)) { | |
414 | spin_unlock(&fs_info->ref_verify_lock); | |
415 | btrfs_err(fs_info, "existing ref when reading from disk?"); | |
416 | kfree(ref); | |
417 | return -EINVAL; | |
418 | } | |
419 | ||
420 | re = lookup_root_entry(&be->roots, ref_root); | |
421 | if (!re) { | |
422 | spin_unlock(&fs_info->ref_verify_lock); | |
423 | btrfs_err(fs_info, "missing root in new block entry?"); | |
424 | return -EINVAL; | |
425 | } | |
426 | re->num_refs += num_refs; | |
427 | spin_unlock(&fs_info->ref_verify_lock); | |
428 | return 0; | |
429 | } | |
430 | ||
431 | static int process_extent_item(struct btrfs_fs_info *fs_info, | |
432 | struct btrfs_path *path, struct btrfs_key *key, | |
433 | int slot, int *tree_block_level) | |
434 | { | |
435 | struct btrfs_extent_item *ei; | |
436 | struct btrfs_extent_inline_ref *iref; | |
437 | struct btrfs_extent_data_ref *dref; | |
438 | struct btrfs_shared_data_ref *sref; | |
439 | struct extent_buffer *leaf = path->nodes[0]; | |
3212fa14 | 440 | u32 item_size = btrfs_item_size(leaf, slot); |
fd708b81 JB |
441 | unsigned long end, ptr; |
442 | u64 offset, flags, count; | |
443 | int type, ret; | |
444 | ||
445 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
446 | flags = btrfs_extent_flags(leaf, ei); | |
447 | ||
448 | if ((key->type == BTRFS_EXTENT_ITEM_KEY) && | |
449 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
450 | struct btrfs_tree_block_info *info; | |
451 | ||
452 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
453 | *tree_block_level = btrfs_tree_block_level(leaf, info); | |
454 | iref = (struct btrfs_extent_inline_ref *)(info + 1); | |
455 | } else { | |
456 | if (key->type == BTRFS_METADATA_ITEM_KEY) | |
457 | *tree_block_level = key->offset; | |
458 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
459 | } | |
460 | ||
461 | ptr = (unsigned long)iref; | |
462 | end = (unsigned long)ei + item_size; | |
463 | while (ptr < end) { | |
464 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
465 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
466 | offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
467 | switch (type) { | |
468 | case BTRFS_TREE_BLOCK_REF_KEY: | |
469 | ret = add_tree_block(fs_info, offset, 0, key->objectid, | |
470 | *tree_block_level); | |
471 | break; | |
472 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
473 | ret = add_tree_block(fs_info, 0, offset, key->objectid, | |
474 | *tree_block_level); | |
475 | break; | |
476 | case BTRFS_EXTENT_DATA_REF_KEY: | |
477 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
478 | ret = add_extent_data_ref(fs_info, leaf, dref, | |
479 | key->objectid, key->offset); | |
480 | break; | |
481 | case BTRFS_SHARED_DATA_REF_KEY: | |
482 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
483 | count = btrfs_shared_data_ref_count(leaf, sref); | |
484 | ret = add_shared_data_ref(fs_info, offset, count, | |
485 | key->objectid, key->offset); | |
486 | break; | |
487 | default: | |
488 | btrfs_err(fs_info, "invalid key type in iref"); | |
489 | ret = -EINVAL; | |
490 | break; | |
491 | } | |
492 | if (ret) | |
493 | break; | |
494 | ptr += btrfs_extent_inline_ref_size(type); | |
495 | } | |
496 | return ret; | |
497 | } | |
498 | ||
499 | static int process_leaf(struct btrfs_root *root, | |
0d73a11c JB |
500 | struct btrfs_path *path, u64 *bytenr, u64 *num_bytes, |
501 | int *tree_block_level) | |
fd708b81 JB |
502 | { |
503 | struct btrfs_fs_info *fs_info = root->fs_info; | |
504 | struct extent_buffer *leaf = path->nodes[0]; | |
505 | struct btrfs_extent_data_ref *dref; | |
506 | struct btrfs_shared_data_ref *sref; | |
507 | u32 count; | |
0d73a11c | 508 | int i = 0, ret = 0; |
fd708b81 JB |
509 | struct btrfs_key key; |
510 | int nritems = btrfs_header_nritems(leaf); | |
511 | ||
512 | for (i = 0; i < nritems; i++) { | |
513 | btrfs_item_key_to_cpu(leaf, &key, i); | |
514 | switch (key.type) { | |
515 | case BTRFS_EXTENT_ITEM_KEY: | |
516 | *num_bytes = key.offset; | |
c730ae0c | 517 | fallthrough; |
fd708b81 JB |
518 | case BTRFS_METADATA_ITEM_KEY: |
519 | *bytenr = key.objectid; | |
520 | ret = process_extent_item(fs_info, path, &key, i, | |
0d73a11c | 521 | tree_block_level); |
fd708b81 JB |
522 | break; |
523 | case BTRFS_TREE_BLOCK_REF_KEY: | |
524 | ret = add_tree_block(fs_info, key.offset, 0, | |
0d73a11c | 525 | key.objectid, *tree_block_level); |
fd708b81 JB |
526 | break; |
527 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
528 | ret = add_tree_block(fs_info, 0, key.offset, | |
0d73a11c | 529 | key.objectid, *tree_block_level); |
fd708b81 JB |
530 | break; |
531 | case BTRFS_EXTENT_DATA_REF_KEY: | |
532 | dref = btrfs_item_ptr(leaf, i, | |
533 | struct btrfs_extent_data_ref); | |
534 | ret = add_extent_data_ref(fs_info, leaf, dref, *bytenr, | |
535 | *num_bytes); | |
536 | break; | |
537 | case BTRFS_SHARED_DATA_REF_KEY: | |
538 | sref = btrfs_item_ptr(leaf, i, | |
539 | struct btrfs_shared_data_ref); | |
540 | count = btrfs_shared_data_ref_count(leaf, sref); | |
541 | ret = add_shared_data_ref(fs_info, key.offset, count, | |
542 | *bytenr, *num_bytes); | |
543 | break; | |
544 | default: | |
545 | break; | |
546 | } | |
547 | if (ret) | |
548 | break; | |
549 | } | |
550 | return ret; | |
551 | } | |
552 | ||
553 | /* Walk down to the leaf from the given level */ | |
554 | static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path, | |
0d73a11c JB |
555 | int level, u64 *bytenr, u64 *num_bytes, |
556 | int *tree_block_level) | |
fd708b81 | 557 | { |
fd708b81 | 558 | struct extent_buffer *eb; |
fd708b81 JB |
559 | int ret = 0; |
560 | ||
561 | while (level >= 0) { | |
562 | if (level) { | |
c990ada2 JB |
563 | eb = btrfs_read_node_slot(path->nodes[level], |
564 | path->slots[level]); | |
fd708b81 JB |
565 | if (IS_ERR(eb)) |
566 | return PTR_ERR(eb); | |
fd708b81 | 567 | btrfs_tree_read_lock(eb); |
fd708b81 JB |
568 | path->nodes[level-1] = eb; |
569 | path->slots[level-1] = 0; | |
ac5887c8 | 570 | path->locks[level-1] = BTRFS_READ_LOCK; |
fd708b81 | 571 | } else { |
0d73a11c JB |
572 | ret = process_leaf(root, path, bytenr, num_bytes, |
573 | tree_block_level); | |
fd708b81 JB |
574 | if (ret) |
575 | break; | |
576 | } | |
577 | level--; | |
578 | } | |
579 | return ret; | |
580 | } | |
581 | ||
582 | /* Walk up to the next node that needs to be processed */ | |
02cfe779 | 583 | static int walk_up_tree(struct btrfs_path *path, int *level) |
fd708b81 JB |
584 | { |
585 | int l; | |
586 | ||
587 | for (l = 0; l < BTRFS_MAX_LEVEL; l++) { | |
588 | if (!path->nodes[l]) | |
589 | continue; | |
590 | if (l) { | |
591 | path->slots[l]++; | |
592 | if (path->slots[l] < | |
593 | btrfs_header_nritems(path->nodes[l])) { | |
594 | *level = l; | |
595 | return 0; | |
596 | } | |
597 | } | |
598 | btrfs_tree_unlock_rw(path->nodes[l], path->locks[l]); | |
599 | free_extent_buffer(path->nodes[l]); | |
600 | path->nodes[l] = NULL; | |
601 | path->slots[l] = 0; | |
602 | path->locks[l] = 0; | |
603 | } | |
604 | ||
605 | return 1; | |
606 | } | |
607 | ||
608 | static void dump_ref_action(struct btrfs_fs_info *fs_info, | |
609 | struct ref_action *ra) | |
610 | { | |
611 | btrfs_err(fs_info, | |
612 | " Ref action %d, root %llu, ref_root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu", | |
613 | ra->action, ra->root, ra->ref.root_objectid, ra->ref.parent, | |
614 | ra->ref.owner, ra->ref.offset, ra->ref.num_refs); | |
615 | __print_stack_trace(fs_info, ra); | |
616 | } | |
617 | ||
618 | /* | |
619 | * Dumps all the information from the block entry to printk, it's going to be | |
620 | * awesome. | |
621 | */ | |
622 | static void dump_block_entry(struct btrfs_fs_info *fs_info, | |
623 | struct block_entry *be) | |
624 | { | |
625 | struct ref_entry *ref; | |
626 | struct root_entry *re; | |
627 | struct ref_action *ra; | |
628 | struct rb_node *n; | |
629 | ||
630 | btrfs_err(fs_info, | |
631 | "dumping block entry [%llu %llu], num_refs %llu, metadata %d, from disk %d", | |
632 | be->bytenr, be->len, be->num_refs, be->metadata, | |
633 | be->from_disk); | |
634 | ||
635 | for (n = rb_first(&be->refs); n; n = rb_next(n)) { | |
636 | ref = rb_entry(n, struct ref_entry, node); | |
637 | btrfs_err(fs_info, | |
638 | " ref root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu", | |
639 | ref->root_objectid, ref->parent, ref->owner, | |
640 | ref->offset, ref->num_refs); | |
641 | } | |
642 | ||
643 | for (n = rb_first(&be->roots); n; n = rb_next(n)) { | |
644 | re = rb_entry(n, struct root_entry, node); | |
645 | btrfs_err(fs_info, " root entry %llu, num_refs %llu", | |
646 | re->root_objectid, re->num_refs); | |
647 | } | |
648 | ||
649 | list_for_each_entry(ra, &be->actions, list) | |
650 | dump_ref_action(fs_info, ra); | |
651 | } | |
652 | ||
653 | /* | |
654 | * btrfs_ref_tree_mod: called when we modify a ref for a bytenr | |
fd708b81 JB |
655 | * |
656 | * This will add an action item to the given bytenr and do sanity checks to make | |
657 | * sure we haven't messed something up. If we are making a new allocation and | |
658 | * this block entry has history we will delete all previous actions as long as | |
659 | * our sanity checks pass as they are no longer needed. | |
660 | */ | |
8a5040f7 QW |
661 | int btrfs_ref_tree_mod(struct btrfs_fs_info *fs_info, |
662 | struct btrfs_ref *generic_ref) | |
fd708b81 | 663 | { |
fd708b81 JB |
664 | struct ref_entry *ref = NULL, *exist; |
665 | struct ref_action *ra = NULL; | |
666 | struct block_entry *be = NULL; | |
667 | struct root_entry *re = NULL; | |
8a5040f7 | 668 | int action = generic_ref->action; |
fd708b81 | 669 | int ret = 0; |
8a5040f7 QW |
670 | bool metadata; |
671 | u64 bytenr = generic_ref->bytenr; | |
672 | u64 num_bytes = generic_ref->len; | |
673 | u64 parent = generic_ref->parent; | |
1478143a JB |
674 | u64 ref_root = 0; |
675 | u64 owner = 0; | |
676 | u64 offset = 0; | |
fd708b81 | 677 | |
8a5040f7 | 678 | if (!btrfs_test_opt(fs_info, REF_VERIFY)) |
fd708b81 JB |
679 | return 0; |
680 | ||
8a5040f7 | 681 | if (generic_ref->type == BTRFS_REF_METADATA) { |
1478143a | 682 | if (!parent) |
113479d5 | 683 | ref_root = generic_ref->tree_ref.owning_root; |
8a5040f7 | 684 | owner = generic_ref->tree_ref.level; |
1478143a | 685 | } else if (!parent) { |
113479d5 | 686 | ref_root = generic_ref->data_ref.owning_root; |
8a5040f7 QW |
687 | owner = generic_ref->data_ref.ino; |
688 | offset = generic_ref->data_ref.offset; | |
689 | } | |
690 | metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
691 | ||
fd708b81 JB |
692 | ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS); |
693 | ra = kmalloc(sizeof(struct ref_action), GFP_NOFS); | |
694 | if (!ra || !ref) { | |
695 | kfree(ref); | |
696 | kfree(ra); | |
697 | ret = -ENOMEM; | |
698 | goto out; | |
699 | } | |
700 | ||
1478143a JB |
701 | ref->parent = parent; |
702 | ref->owner = owner; | |
703 | ref->root_objectid = ref_root; | |
704 | ref->offset = offset; | |
fd708b81 JB |
705 | ref->num_refs = (action == BTRFS_DROP_DELAYED_REF) ? -1 : 1; |
706 | ||
707 | memcpy(&ra->ref, ref, sizeof(struct ref_entry)); | |
708 | /* | |
709 | * Save the extra info from the delayed ref in the ref action to make it | |
710 | * easier to figure out what is happening. The real ref's we add to the | |
711 | * ref tree need to reflect what we save on disk so it matches any | |
712 | * on-disk refs we pre-loaded. | |
713 | */ | |
714 | ra->ref.owner = owner; | |
715 | ra->ref.offset = offset; | |
716 | ra->ref.root_objectid = ref_root; | |
717 | __save_stack_trace(ra); | |
718 | ||
719 | INIT_LIST_HEAD(&ra->list); | |
720 | ra->action = action; | |
8a5040f7 | 721 | ra->root = generic_ref->real_root; |
fd708b81 JB |
722 | |
723 | /* | |
724 | * This is an allocation, preallocate the block_entry in case we haven't | |
725 | * used it before. | |
726 | */ | |
727 | ret = -EINVAL; | |
728 | if (action == BTRFS_ADD_DELAYED_EXTENT) { | |
729 | /* | |
730 | * For subvol_create we'll just pass in whatever the parent root | |
731 | * is and the new root objectid, so let's not treat the passed | |
732 | * in root as if it really has a ref for this bytenr. | |
733 | */ | |
8a5040f7 | 734 | be = add_block_entry(fs_info, bytenr, num_bytes, ref_root); |
fd708b81 | 735 | if (IS_ERR(be)) { |
f311ade3 | 736 | kfree(ref); |
fd708b81 JB |
737 | kfree(ra); |
738 | ret = PTR_ERR(be); | |
739 | goto out; | |
740 | } | |
741 | be->num_refs++; | |
742 | if (metadata) | |
743 | be->metadata = 1; | |
744 | ||
745 | if (be->num_refs != 1) { | |
746 | btrfs_err(fs_info, | |
747 | "re-allocated a block that still has references to it!"); | |
748 | dump_block_entry(fs_info, be); | |
749 | dump_ref_action(fs_info, ra); | |
f311ade3 WW |
750 | kfree(ref); |
751 | kfree(ra); | |
fd708b81 JB |
752 | goto out_unlock; |
753 | } | |
754 | ||
755 | while (!list_empty(&be->actions)) { | |
756 | struct ref_action *tmp; | |
757 | ||
758 | tmp = list_first_entry(&be->actions, struct ref_action, | |
759 | list); | |
760 | list_del(&tmp->list); | |
761 | kfree(tmp); | |
762 | } | |
763 | } else { | |
764 | struct root_entry *tmp; | |
765 | ||
766 | if (!parent) { | |
767 | re = kmalloc(sizeof(struct root_entry), GFP_NOFS); | |
768 | if (!re) { | |
769 | kfree(ref); | |
770 | kfree(ra); | |
771 | ret = -ENOMEM; | |
772 | goto out; | |
773 | } | |
774 | /* | |
775 | * This is the root that is modifying us, so it's the | |
776 | * one we want to lookup below when we modify the | |
777 | * re->num_refs. | |
778 | */ | |
8a5040f7 QW |
779 | ref_root = generic_ref->real_root; |
780 | re->root_objectid = generic_ref->real_root; | |
fd708b81 JB |
781 | re->num_refs = 0; |
782 | } | |
783 | ||
8a5040f7 QW |
784 | spin_lock(&fs_info->ref_verify_lock); |
785 | be = lookup_block_entry(&fs_info->block_tree, bytenr); | |
fd708b81 JB |
786 | if (!be) { |
787 | btrfs_err(fs_info, | |
788 | "trying to do action %d to bytenr %llu num_bytes %llu but there is no existing entry!", | |
cc7c7714 | 789 | action, bytenr, num_bytes); |
fd708b81 JB |
790 | dump_ref_action(fs_info, ra); |
791 | kfree(ref); | |
792 | kfree(ra); | |
793 | goto out_unlock; | |
b39c8f5a JB |
794 | } else if (be->num_refs == 0) { |
795 | btrfs_err(fs_info, | |
796 | "trying to do action %d for a bytenr that has 0 total references", | |
797 | action); | |
798 | dump_block_entry(fs_info, be); | |
799 | dump_ref_action(fs_info, ra); | |
800 | kfree(ref); | |
801 | kfree(ra); | |
802 | goto out_unlock; | |
fd708b81 JB |
803 | } |
804 | ||
805 | if (!parent) { | |
806 | tmp = insert_root_entry(&be->roots, re); | |
807 | if (tmp) { | |
808 | kfree(re); | |
809 | re = tmp; | |
810 | } | |
811 | } | |
812 | } | |
813 | ||
814 | exist = insert_ref_entry(&be->refs, ref); | |
815 | if (exist) { | |
816 | if (action == BTRFS_DROP_DELAYED_REF) { | |
817 | if (exist->num_refs == 0) { | |
818 | btrfs_err(fs_info, | |
819 | "dropping a ref for a existing root that doesn't have a ref on the block"); | |
820 | dump_block_entry(fs_info, be); | |
821 | dump_ref_action(fs_info, ra); | |
f311ade3 | 822 | kfree(ref); |
fd708b81 JB |
823 | kfree(ra); |
824 | goto out_unlock; | |
825 | } | |
826 | exist->num_refs--; | |
827 | if (exist->num_refs == 0) { | |
828 | rb_erase(&exist->node, &be->refs); | |
829 | kfree(exist); | |
830 | } | |
831 | } else if (!be->metadata) { | |
832 | exist->num_refs++; | |
833 | } else { | |
834 | btrfs_err(fs_info, | |
835 | "attempting to add another ref for an existing ref on a tree block"); | |
836 | dump_block_entry(fs_info, be); | |
837 | dump_ref_action(fs_info, ra); | |
f311ade3 | 838 | kfree(ref); |
fd708b81 JB |
839 | kfree(ra); |
840 | goto out_unlock; | |
841 | } | |
842 | kfree(ref); | |
843 | } else { | |
844 | if (action == BTRFS_DROP_DELAYED_REF) { | |
845 | btrfs_err(fs_info, | |
846 | "dropping a ref for a root that doesn't have a ref on the block"); | |
847 | dump_block_entry(fs_info, be); | |
848 | dump_ref_action(fs_info, ra); | |
468600c6 | 849 | kfree(ref); |
fd708b81 JB |
850 | kfree(ra); |
851 | goto out_unlock; | |
852 | } | |
853 | } | |
854 | ||
855 | if (!parent && !re) { | |
856 | re = lookup_root_entry(&be->roots, ref_root); | |
857 | if (!re) { | |
858 | /* | |
859 | * This shouldn't happen because we will add our re | |
860 | * above when we lookup the be with !parent, but just in | |
861 | * case catch this case so we don't panic because I | |
52042d8e | 862 | * didn't think of some other corner case. |
fd708b81 JB |
863 | */ |
864 | btrfs_err(fs_info, "failed to find root %llu for %llu", | |
8a5040f7 | 865 | generic_ref->real_root, be->bytenr); |
fd708b81 JB |
866 | dump_block_entry(fs_info, be); |
867 | dump_ref_action(fs_info, ra); | |
868 | kfree(ra); | |
869 | goto out_unlock; | |
870 | } | |
871 | } | |
872 | if (action == BTRFS_DROP_DELAYED_REF) { | |
873 | if (re) | |
874 | re->num_refs--; | |
875 | be->num_refs--; | |
876 | } else if (action == BTRFS_ADD_DELAYED_REF) { | |
877 | be->num_refs++; | |
878 | if (re) | |
879 | re->num_refs++; | |
880 | } | |
881 | list_add_tail(&ra->list, &be->actions); | |
882 | ret = 0; | |
883 | out_unlock: | |
8a5040f7 | 884 | spin_unlock(&fs_info->ref_verify_lock); |
fd708b81 JB |
885 | out: |
886 | if (ret) | |
887 | btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY); | |
888 | return ret; | |
889 | } | |
890 | ||
891 | /* Free up the ref cache */ | |
892 | void btrfs_free_ref_cache(struct btrfs_fs_info *fs_info) | |
893 | { | |
894 | struct block_entry *be; | |
895 | struct rb_node *n; | |
896 | ||
897 | if (!btrfs_test_opt(fs_info, REF_VERIFY)) | |
898 | return; | |
899 | ||
900 | spin_lock(&fs_info->ref_verify_lock); | |
901 | while ((n = rb_first(&fs_info->block_tree))) { | |
902 | be = rb_entry(n, struct block_entry, node); | |
903 | rb_erase(&be->node, &fs_info->block_tree); | |
904 | free_block_entry(be); | |
905 | cond_resched_lock(&fs_info->ref_verify_lock); | |
906 | } | |
907 | spin_unlock(&fs_info->ref_verify_lock); | |
908 | } | |
909 | ||
910 | void btrfs_free_ref_tree_range(struct btrfs_fs_info *fs_info, u64 start, | |
911 | u64 len) | |
912 | { | |
913 | struct block_entry *be = NULL, *entry; | |
914 | struct rb_node *n; | |
915 | ||
916 | if (!btrfs_test_opt(fs_info, REF_VERIFY)) | |
917 | return; | |
918 | ||
919 | spin_lock(&fs_info->ref_verify_lock); | |
920 | n = fs_info->block_tree.rb_node; | |
921 | while (n) { | |
922 | entry = rb_entry(n, struct block_entry, node); | |
923 | if (entry->bytenr < start) { | |
924 | n = n->rb_right; | |
925 | } else if (entry->bytenr > start) { | |
926 | n = n->rb_left; | |
927 | } else { | |
928 | be = entry; | |
929 | break; | |
930 | } | |
931 | /* We want to get as close to start as possible */ | |
932 | if (be == NULL || | |
933 | (entry->bytenr < start && be->bytenr > start) || | |
934 | (entry->bytenr < start && entry->bytenr > be->bytenr)) | |
935 | be = entry; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Could have an empty block group, maybe have something to check for | |
940 | * this case to verify we were actually empty? | |
941 | */ | |
942 | if (!be) { | |
943 | spin_unlock(&fs_info->ref_verify_lock); | |
944 | return; | |
945 | } | |
946 | ||
947 | n = &be->node; | |
948 | while (n) { | |
949 | be = rb_entry(n, struct block_entry, node); | |
950 | n = rb_next(n); | |
951 | if (be->bytenr < start && be->bytenr + be->len > start) { | |
952 | btrfs_err(fs_info, | |
953 | "block entry overlaps a block group [%llu,%llu]!", | |
954 | start, len); | |
955 | dump_block_entry(fs_info, be); | |
956 | continue; | |
957 | } | |
958 | if (be->bytenr < start) | |
959 | continue; | |
960 | if (be->bytenr >= start + len) | |
961 | break; | |
962 | if (be->bytenr + be->len > start + len) { | |
963 | btrfs_err(fs_info, | |
964 | "block entry overlaps a block group [%llu,%llu]!", | |
965 | start, len); | |
966 | dump_block_entry(fs_info, be); | |
967 | } | |
968 | rb_erase(&be->node, &fs_info->block_tree); | |
969 | free_block_entry(be); | |
970 | } | |
971 | spin_unlock(&fs_info->ref_verify_lock); | |
972 | } | |
973 | ||
974 | /* Walk down all roots and build the ref tree, meant to be called at mount */ | |
975 | int btrfs_build_ref_tree(struct btrfs_fs_info *fs_info) | |
976 | { | |
29cbcf40 | 977 | struct btrfs_root *extent_root; |
fd708b81 | 978 | struct btrfs_path *path; |
fd708b81 | 979 | struct extent_buffer *eb; |
0d73a11c | 980 | int tree_block_level = 0; |
fd708b81 JB |
981 | u64 bytenr = 0, num_bytes = 0; |
982 | int ret, level; | |
983 | ||
984 | if (!btrfs_test_opt(fs_info, REF_VERIFY)) | |
985 | return 0; | |
986 | ||
987 | path = btrfs_alloc_path(); | |
988 | if (!path) | |
989 | return -ENOMEM; | |
990 | ||
29cbcf40 JB |
991 | extent_root = btrfs_extent_root(fs_info, 0); |
992 | eb = btrfs_read_lock_root_node(extent_root); | |
fd708b81 JB |
993 | level = btrfs_header_level(eb); |
994 | path->nodes[level] = eb; | |
995 | path->slots[level] = 0; | |
ac5887c8 | 996 | path->locks[level] = BTRFS_READ_LOCK; |
fd708b81 JB |
997 | |
998 | while (1) { | |
999 | /* | |
1000 | * We have to keep track of the bytenr/num_bytes we last hit | |
1001 | * because we could have run out of space for an inline ref, and | |
1002 | * would have had to added a ref key item which may appear on a | |
1003 | * different leaf from the original extent item. | |
1004 | */ | |
29cbcf40 | 1005 | ret = walk_down_tree(extent_root, path, level, |
0d73a11c | 1006 | &bytenr, &num_bytes, &tree_block_level); |
fd708b81 JB |
1007 | if (ret) |
1008 | break; | |
02cfe779 | 1009 | ret = walk_up_tree(path, &level); |
fd708b81 JB |
1010 | if (ret < 0) |
1011 | break; | |
1012 | if (ret > 0) { | |
1013 | ret = 0; | |
1014 | break; | |
1015 | } | |
1016 | } | |
1017 | if (ret) { | |
1018 | btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY); | |
1019 | btrfs_free_ref_cache(fs_info); | |
1020 | } | |
1021 | btrfs_free_path(path); | |
1022 | return ret; | |
1023 | } |