Commit | Line | Data |
---|---|---|
a542ad1b JS |
1 | /* |
2 | * Copyright (C) 2011 STRATO. 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 | ||
425d17a2 | 19 | #include <linux/vmalloc.h> |
a542ad1b JS |
20 | #include "ctree.h" |
21 | #include "disk-io.h" | |
22 | #include "backref.h" | |
8da6d581 JS |
23 | #include "ulist.h" |
24 | #include "transaction.h" | |
25 | #include "delayed-ref.h" | |
b916a59a | 26 | #include "locking.h" |
a542ad1b | 27 | |
976b1908 JS |
28 | struct extent_inode_elem { |
29 | u64 inum; | |
30 | u64 offset; | |
31 | struct extent_inode_elem *next; | |
32 | }; | |
33 | ||
34 | static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb, | |
35 | struct btrfs_file_extent_item *fi, | |
36 | u64 extent_item_pos, | |
37 | struct extent_inode_elem **eie) | |
38 | { | |
39 | u64 data_offset; | |
40 | u64 data_len; | |
41 | struct extent_inode_elem *e; | |
42 | ||
43 | data_offset = btrfs_file_extent_offset(eb, fi); | |
44 | data_len = btrfs_file_extent_num_bytes(eb, fi); | |
45 | ||
46 | if (extent_item_pos < data_offset || | |
47 | extent_item_pos >= data_offset + data_len) | |
48 | return 1; | |
49 | ||
50 | e = kmalloc(sizeof(*e), GFP_NOFS); | |
51 | if (!e) | |
52 | return -ENOMEM; | |
53 | ||
54 | e->next = *eie; | |
55 | e->inum = key->objectid; | |
56 | e->offset = key->offset + (extent_item_pos - data_offset); | |
57 | *eie = e; | |
58 | ||
59 | return 0; | |
60 | } | |
61 | ||
62 | static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte, | |
63 | u64 extent_item_pos, | |
64 | struct extent_inode_elem **eie) | |
65 | { | |
66 | u64 disk_byte; | |
67 | struct btrfs_key key; | |
68 | struct btrfs_file_extent_item *fi; | |
69 | int slot; | |
70 | int nritems; | |
71 | int extent_type; | |
72 | int ret; | |
73 | ||
74 | /* | |
75 | * from the shared data ref, we only have the leaf but we need | |
76 | * the key. thus, we must look into all items and see that we | |
77 | * find one (some) with a reference to our extent item. | |
78 | */ | |
79 | nritems = btrfs_header_nritems(eb); | |
80 | for (slot = 0; slot < nritems; ++slot) { | |
81 | btrfs_item_key_to_cpu(eb, &key, slot); | |
82 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
83 | continue; | |
84 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
85 | extent_type = btrfs_file_extent_type(eb, fi); | |
86 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
87 | continue; | |
88 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
89 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
90 | if (disk_byte != wanted_disk_byte) | |
91 | continue; | |
92 | ||
93 | ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie); | |
94 | if (ret < 0) | |
95 | return ret; | |
96 | } | |
97 | ||
98 | return 0; | |
99 | } | |
100 | ||
8da6d581 JS |
101 | /* |
102 | * this structure records all encountered refs on the way up to the root | |
103 | */ | |
104 | struct __prelim_ref { | |
105 | struct list_head list; | |
106 | u64 root_id; | |
d5c88b73 | 107 | struct btrfs_key key_for_search; |
8da6d581 JS |
108 | int level; |
109 | int count; | |
3301958b | 110 | struct extent_inode_elem *inode_list; |
8da6d581 JS |
111 | u64 parent; |
112 | u64 wanted_disk_byte; | |
113 | }; | |
114 | ||
d5c88b73 JS |
115 | /* |
116 | * the rules for all callers of this function are: | |
117 | * - obtaining the parent is the goal | |
118 | * - if you add a key, you must know that it is a correct key | |
119 | * - if you cannot add the parent or a correct key, then we will look into the | |
120 | * block later to set a correct key | |
121 | * | |
122 | * delayed refs | |
123 | * ============ | |
124 | * backref type | shared | indirect | shared | indirect | |
125 | * information | tree | tree | data | data | |
126 | * --------------------+--------+----------+--------+---------- | |
127 | * parent logical | y | - | - | - | |
128 | * key to resolve | - | y | y | y | |
129 | * tree block logical | - | - | - | - | |
130 | * root for resolving | y | y | y | y | |
131 | * | |
132 | * - column 1: we've the parent -> done | |
133 | * - column 2, 3, 4: we use the key to find the parent | |
134 | * | |
135 | * on disk refs (inline or keyed) | |
136 | * ============================== | |
137 | * backref type | shared | indirect | shared | indirect | |
138 | * information | tree | tree | data | data | |
139 | * --------------------+--------+----------+--------+---------- | |
140 | * parent logical | y | - | y | - | |
141 | * key to resolve | - | - | - | y | |
142 | * tree block logical | y | y | y | y | |
143 | * root for resolving | - | y | y | y | |
144 | * | |
145 | * - column 1, 3: we've the parent -> done | |
146 | * - column 2: we take the first key from the block to find the parent | |
147 | * (see __add_missing_keys) | |
148 | * - column 4: we use the key to find the parent | |
149 | * | |
150 | * additional information that's available but not required to find the parent | |
151 | * block might help in merging entries to gain some speed. | |
152 | */ | |
153 | ||
8da6d581 | 154 | static int __add_prelim_ref(struct list_head *head, u64 root_id, |
d5c88b73 JS |
155 | struct btrfs_key *key, int level, |
156 | u64 parent, u64 wanted_disk_byte, int count) | |
8da6d581 JS |
157 | { |
158 | struct __prelim_ref *ref; | |
159 | ||
160 | /* in case we're adding delayed refs, we're holding the refs spinlock */ | |
161 | ref = kmalloc(sizeof(*ref), GFP_ATOMIC); | |
162 | if (!ref) | |
163 | return -ENOMEM; | |
164 | ||
165 | ref->root_id = root_id; | |
166 | if (key) | |
d5c88b73 | 167 | ref->key_for_search = *key; |
8da6d581 | 168 | else |
d5c88b73 | 169 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 170 | |
3301958b | 171 | ref->inode_list = NULL; |
8da6d581 JS |
172 | ref->level = level; |
173 | ref->count = count; | |
174 | ref->parent = parent; | |
175 | ref->wanted_disk_byte = wanted_disk_byte; | |
176 | list_add_tail(&ref->list, head); | |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, | |
976b1908 | 182 | struct ulist *parents, int level, |
69bca40d | 183 | struct btrfs_key *key_for_search, u64 time_seq, |
3d7806ec | 184 | u64 wanted_disk_byte, |
976b1908 | 185 | const u64 *extent_item_pos) |
8da6d581 | 186 | { |
69bca40d AB |
187 | int ret = 0; |
188 | int slot; | |
189 | struct extent_buffer *eb; | |
190 | struct btrfs_key key; | |
8da6d581 | 191 | struct btrfs_file_extent_item *fi; |
3301958b | 192 | struct extent_inode_elem *eie = NULL; |
8da6d581 JS |
193 | u64 disk_byte; |
194 | ||
69bca40d AB |
195 | if (level != 0) { |
196 | eb = path->nodes[level]; | |
197 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
198 | if (ret < 0) |
199 | return ret; | |
8da6d581 | 200 | return 0; |
69bca40d | 201 | } |
8da6d581 JS |
202 | |
203 | /* | |
69bca40d AB |
204 | * We normally enter this function with the path already pointing to |
205 | * the first item to check. But sometimes, we may enter it with | |
206 | * slot==nritems. In that case, go to the next leaf before we continue. | |
8da6d581 | 207 | */ |
69bca40d | 208 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) |
3d7806ec | 209 | ret = btrfs_next_old_leaf(root, path, time_seq); |
8da6d581 | 210 | |
69bca40d | 211 | while (!ret) { |
8da6d581 | 212 | eb = path->nodes[0]; |
69bca40d AB |
213 | slot = path->slots[0]; |
214 | ||
215 | btrfs_item_key_to_cpu(eb, &key, slot); | |
216 | ||
217 | if (key.objectid != key_for_search->objectid || | |
218 | key.type != BTRFS_EXTENT_DATA_KEY) | |
219 | break; | |
220 | ||
221 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
222 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
223 | ||
224 | if (disk_byte == wanted_disk_byte) { | |
225 | eie = NULL; | |
226 | if (extent_item_pos) { | |
227 | ret = check_extent_in_eb(&key, eb, fi, | |
228 | *extent_item_pos, | |
229 | &eie); | |
230 | if (ret < 0) | |
231 | break; | |
232 | } | |
233 | if (!ret) { | |
234 | ret = ulist_add(parents, eb->start, | |
995e01b7 | 235 | (uintptr_t)eie, GFP_NOFS); |
69bca40d AB |
236 | if (ret < 0) |
237 | break; | |
238 | if (!extent_item_pos) { | |
239 | ret = btrfs_next_old_leaf(root, path, | |
240 | time_seq); | |
241 | continue; | |
242 | } | |
243 | } | |
8da6d581 | 244 | } |
69bca40d | 245 | ret = btrfs_next_old_item(root, path, time_seq); |
8da6d581 JS |
246 | } |
247 | ||
69bca40d AB |
248 | if (ret > 0) |
249 | ret = 0; | |
250 | return ret; | |
8da6d581 JS |
251 | } |
252 | ||
253 | /* | |
254 | * resolve an indirect backref in the form (root_id, key, level) | |
255 | * to a logical address | |
256 | */ | |
257 | static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info, | |
7a3ae2f8 | 258 | int search_commit_root, |
8445f61c | 259 | u64 time_seq, |
8da6d581 | 260 | struct __prelim_ref *ref, |
976b1908 JS |
261 | struct ulist *parents, |
262 | const u64 *extent_item_pos) | |
8da6d581 JS |
263 | { |
264 | struct btrfs_path *path; | |
265 | struct btrfs_root *root; | |
266 | struct btrfs_key root_key; | |
8da6d581 JS |
267 | struct extent_buffer *eb; |
268 | int ret = 0; | |
269 | int root_level; | |
270 | int level = ref->level; | |
271 | ||
272 | path = btrfs_alloc_path(); | |
273 | if (!path) | |
274 | return -ENOMEM; | |
7a3ae2f8 | 275 | path->search_commit_root = !!search_commit_root; |
8da6d581 JS |
276 | |
277 | root_key.objectid = ref->root_id; | |
278 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
279 | root_key.offset = (u64)-1; | |
280 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
281 | if (IS_ERR(root)) { | |
282 | ret = PTR_ERR(root); | |
283 | goto out; | |
284 | } | |
285 | ||
5b6602e7 | 286 | root_level = btrfs_old_root_level(root, time_seq); |
8da6d581 JS |
287 | |
288 | if (root_level + 1 == level) | |
289 | goto out; | |
290 | ||
291 | path->lowest_level = level; | |
8445f61c | 292 | ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq); |
8da6d581 JS |
293 | pr_debug("search slot in root %llu (level %d, ref count %d) returned " |
294 | "%d for key (%llu %u %llu)\n", | |
295 | (unsigned long long)ref->root_id, level, ref->count, ret, | |
d5c88b73 JS |
296 | (unsigned long long)ref->key_for_search.objectid, |
297 | ref->key_for_search.type, | |
298 | (unsigned long long)ref->key_for_search.offset); | |
8da6d581 JS |
299 | if (ret < 0) |
300 | goto out; | |
301 | ||
302 | eb = path->nodes[level]; | |
9345457f JS |
303 | while (!eb) { |
304 | if (!level) { | |
305 | WARN_ON(1); | |
306 | ret = 1; | |
307 | goto out; | |
308 | } | |
309 | level--; | |
310 | eb = path->nodes[level]; | |
8da6d581 JS |
311 | } |
312 | ||
69bca40d AB |
313 | ret = add_all_parents(root, path, parents, level, &ref->key_for_search, |
314 | time_seq, ref->wanted_disk_byte, | |
315 | extent_item_pos); | |
8da6d581 JS |
316 | out: |
317 | btrfs_free_path(path); | |
318 | return ret; | |
319 | } | |
320 | ||
321 | /* | |
322 | * resolve all indirect backrefs from the list | |
323 | */ | |
324 | static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info, | |
8445f61c | 325 | int search_commit_root, u64 time_seq, |
976b1908 JS |
326 | struct list_head *head, |
327 | const u64 *extent_item_pos) | |
8da6d581 JS |
328 | { |
329 | int err; | |
330 | int ret = 0; | |
331 | struct __prelim_ref *ref; | |
332 | struct __prelim_ref *ref_safe; | |
333 | struct __prelim_ref *new_ref; | |
334 | struct ulist *parents; | |
335 | struct ulist_node *node; | |
cd1b413c | 336 | struct ulist_iterator uiter; |
8da6d581 JS |
337 | |
338 | parents = ulist_alloc(GFP_NOFS); | |
339 | if (!parents) | |
340 | return -ENOMEM; | |
341 | ||
342 | /* | |
343 | * _safe allows us to insert directly after the current item without | |
344 | * iterating over the newly inserted items. | |
345 | * we're also allowed to re-assign ref during iteration. | |
346 | */ | |
347 | list_for_each_entry_safe(ref, ref_safe, head, list) { | |
348 | if (ref->parent) /* already direct */ | |
349 | continue; | |
350 | if (ref->count == 0) | |
351 | continue; | |
7a3ae2f8 | 352 | err = __resolve_indirect_ref(fs_info, search_commit_root, |
8445f61c JS |
353 | time_seq, ref, parents, |
354 | extent_item_pos); | |
ca60ebfa | 355 | if (err) |
8da6d581 | 356 | continue; |
8da6d581 JS |
357 | |
358 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
359 | ULIST_ITER_INIT(&uiter); |
360 | node = ulist_next(parents, &uiter); | |
8da6d581 | 361 | ref->parent = node ? node->val : 0; |
995e01b7 JS |
362 | ref->inode_list = node ? |
363 | (struct extent_inode_elem *)(uintptr_t)node->aux : 0; | |
8da6d581 JS |
364 | |
365 | /* additional parents require new refs being added here */ | |
cd1b413c | 366 | while ((node = ulist_next(parents, &uiter))) { |
8da6d581 JS |
367 | new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS); |
368 | if (!new_ref) { | |
369 | ret = -ENOMEM; | |
370 | break; | |
371 | } | |
372 | memcpy(new_ref, ref, sizeof(*ref)); | |
373 | new_ref->parent = node->val; | |
995e01b7 JS |
374 | new_ref->inode_list = (struct extent_inode_elem *) |
375 | (uintptr_t)node->aux; | |
8da6d581 JS |
376 | list_add(&new_ref->list, &ref->list); |
377 | } | |
378 | ulist_reinit(parents); | |
379 | } | |
380 | ||
381 | ulist_free(parents); | |
382 | return ret; | |
383 | } | |
384 | ||
d5c88b73 JS |
385 | static inline int ref_for_same_block(struct __prelim_ref *ref1, |
386 | struct __prelim_ref *ref2) | |
387 | { | |
388 | if (ref1->level != ref2->level) | |
389 | return 0; | |
390 | if (ref1->root_id != ref2->root_id) | |
391 | return 0; | |
392 | if (ref1->key_for_search.type != ref2->key_for_search.type) | |
393 | return 0; | |
394 | if (ref1->key_for_search.objectid != ref2->key_for_search.objectid) | |
395 | return 0; | |
396 | if (ref1->key_for_search.offset != ref2->key_for_search.offset) | |
397 | return 0; | |
398 | if (ref1->parent != ref2->parent) | |
399 | return 0; | |
400 | ||
401 | return 1; | |
402 | } | |
403 | ||
404 | /* | |
405 | * read tree blocks and add keys where required. | |
406 | */ | |
407 | static int __add_missing_keys(struct btrfs_fs_info *fs_info, | |
408 | struct list_head *head) | |
409 | { | |
410 | struct list_head *pos; | |
411 | struct extent_buffer *eb; | |
412 | ||
413 | list_for_each(pos, head) { | |
414 | struct __prelim_ref *ref; | |
415 | ref = list_entry(pos, struct __prelim_ref, list); | |
416 | ||
417 | if (ref->parent) | |
418 | continue; | |
419 | if (ref->key_for_search.type) | |
420 | continue; | |
421 | BUG_ON(!ref->wanted_disk_byte); | |
422 | eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte, | |
423 | fs_info->tree_root->leafsize, 0); | |
424 | BUG_ON(!eb); | |
425 | btrfs_tree_read_lock(eb); | |
426 | if (btrfs_header_level(eb) == 0) | |
427 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
428 | else | |
429 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
430 | btrfs_tree_read_unlock(eb); | |
431 | free_extent_buffer(eb); | |
432 | } | |
433 | return 0; | |
434 | } | |
435 | ||
8da6d581 JS |
436 | /* |
437 | * merge two lists of backrefs and adjust counts accordingly | |
438 | * | |
439 | * mode = 1: merge identical keys, if key is set | |
d5c88b73 JS |
440 | * FIXME: if we add more keys in __add_prelim_ref, we can merge more here. |
441 | * additionally, we could even add a key range for the blocks we | |
442 | * looked into to merge even more (-> replace unresolved refs by those | |
443 | * having a parent). | |
8da6d581 JS |
444 | * mode = 2: merge identical parents |
445 | */ | |
692206b1 | 446 | static void __merge_refs(struct list_head *head, int mode) |
8da6d581 JS |
447 | { |
448 | struct list_head *pos1; | |
449 | ||
450 | list_for_each(pos1, head) { | |
451 | struct list_head *n2; | |
452 | struct list_head *pos2; | |
453 | struct __prelim_ref *ref1; | |
454 | ||
455 | ref1 = list_entry(pos1, struct __prelim_ref, list); | |
456 | ||
8da6d581 JS |
457 | for (pos2 = pos1->next, n2 = pos2->next; pos2 != head; |
458 | pos2 = n2, n2 = pos2->next) { | |
459 | struct __prelim_ref *ref2; | |
d5c88b73 | 460 | struct __prelim_ref *xchg; |
3ef5969c | 461 | struct extent_inode_elem *eie; |
8da6d581 JS |
462 | |
463 | ref2 = list_entry(pos2, struct __prelim_ref, list); | |
464 | ||
465 | if (mode == 1) { | |
d5c88b73 | 466 | if (!ref_for_same_block(ref1, ref2)) |
8da6d581 | 467 | continue; |
d5c88b73 JS |
468 | if (!ref1->parent && ref2->parent) { |
469 | xchg = ref1; | |
470 | ref1 = ref2; | |
471 | ref2 = xchg; | |
472 | } | |
8da6d581 JS |
473 | } else { |
474 | if (ref1->parent != ref2->parent) | |
475 | continue; | |
8da6d581 | 476 | } |
3ef5969c AB |
477 | |
478 | eie = ref1->inode_list; | |
479 | while (eie && eie->next) | |
480 | eie = eie->next; | |
481 | if (eie) | |
482 | eie->next = ref2->inode_list; | |
483 | else | |
484 | ref1->inode_list = ref2->inode_list; | |
485 | ref1->count += ref2->count; | |
486 | ||
8da6d581 JS |
487 | list_del(&ref2->list); |
488 | kfree(ref2); | |
489 | } | |
490 | ||
491 | } | |
8da6d581 JS |
492 | } |
493 | ||
494 | /* | |
495 | * add all currently queued delayed refs from this head whose seq nr is | |
496 | * smaller or equal that seq to the list | |
497 | */ | |
498 | static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq, | |
8da6d581 JS |
499 | struct list_head *prefs) |
500 | { | |
501 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
502 | struct rb_node *n = &head->node.rb_node; | |
d5c88b73 JS |
503 | struct btrfs_key key; |
504 | struct btrfs_key op_key = {0}; | |
8da6d581 | 505 | int sgn; |
b1375d64 | 506 | int ret = 0; |
8da6d581 JS |
507 | |
508 | if (extent_op && extent_op->update_key) | |
d5c88b73 | 509 | btrfs_disk_key_to_cpu(&op_key, &extent_op->key); |
8da6d581 JS |
510 | |
511 | while ((n = rb_prev(n))) { | |
512 | struct btrfs_delayed_ref_node *node; | |
513 | node = rb_entry(n, struct btrfs_delayed_ref_node, | |
514 | rb_node); | |
515 | if (node->bytenr != head->node.bytenr) | |
516 | break; | |
517 | WARN_ON(node->is_head); | |
518 | ||
519 | if (node->seq > seq) | |
520 | continue; | |
521 | ||
522 | switch (node->action) { | |
523 | case BTRFS_ADD_DELAYED_EXTENT: | |
524 | case BTRFS_UPDATE_DELAYED_HEAD: | |
525 | WARN_ON(1); | |
526 | continue; | |
527 | case BTRFS_ADD_DELAYED_REF: | |
528 | sgn = 1; | |
529 | break; | |
530 | case BTRFS_DROP_DELAYED_REF: | |
531 | sgn = -1; | |
532 | break; | |
533 | default: | |
534 | BUG_ON(1); | |
535 | } | |
536 | switch (node->type) { | |
537 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
538 | struct btrfs_delayed_tree_ref *ref; | |
539 | ||
540 | ref = btrfs_delayed_node_to_tree_ref(node); | |
d5c88b73 | 541 | ret = __add_prelim_ref(prefs, ref->root, &op_key, |
8da6d581 JS |
542 | ref->level + 1, 0, node->bytenr, |
543 | node->ref_mod * sgn); | |
544 | break; | |
545 | } | |
546 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
547 | struct btrfs_delayed_tree_ref *ref; | |
548 | ||
549 | ref = btrfs_delayed_node_to_tree_ref(node); | |
d5c88b73 | 550 | ret = __add_prelim_ref(prefs, ref->root, NULL, |
8da6d581 JS |
551 | ref->level + 1, ref->parent, |
552 | node->bytenr, | |
553 | node->ref_mod * sgn); | |
554 | break; | |
555 | } | |
556 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
557 | struct btrfs_delayed_data_ref *ref; | |
8da6d581 JS |
558 | ref = btrfs_delayed_node_to_data_ref(node); |
559 | ||
560 | key.objectid = ref->objectid; | |
561 | key.type = BTRFS_EXTENT_DATA_KEY; | |
562 | key.offset = ref->offset; | |
563 | ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0, | |
564 | node->bytenr, | |
565 | node->ref_mod * sgn); | |
566 | break; | |
567 | } | |
568 | case BTRFS_SHARED_DATA_REF_KEY: { | |
569 | struct btrfs_delayed_data_ref *ref; | |
8da6d581 JS |
570 | |
571 | ref = btrfs_delayed_node_to_data_ref(node); | |
572 | ||
573 | key.objectid = ref->objectid; | |
574 | key.type = BTRFS_EXTENT_DATA_KEY; | |
575 | key.offset = ref->offset; | |
576 | ret = __add_prelim_ref(prefs, ref->root, &key, 0, | |
577 | ref->parent, node->bytenr, | |
578 | node->ref_mod * sgn); | |
579 | break; | |
580 | } | |
581 | default: | |
582 | WARN_ON(1); | |
583 | } | |
1149ab6b WS |
584 | if (ret) |
585 | return ret; | |
8da6d581 JS |
586 | } |
587 | ||
588 | return 0; | |
589 | } | |
590 | ||
591 | /* | |
592 | * add all inline backrefs for bytenr to the list | |
593 | */ | |
594 | static int __add_inline_refs(struct btrfs_fs_info *fs_info, | |
595 | struct btrfs_path *path, u64 bytenr, | |
d5c88b73 | 596 | int *info_level, struct list_head *prefs) |
8da6d581 | 597 | { |
b1375d64 | 598 | int ret = 0; |
8da6d581 JS |
599 | int slot; |
600 | struct extent_buffer *leaf; | |
601 | struct btrfs_key key; | |
602 | unsigned long ptr; | |
603 | unsigned long end; | |
604 | struct btrfs_extent_item *ei; | |
605 | u64 flags; | |
606 | u64 item_size; | |
607 | ||
608 | /* | |
609 | * enumerate all inline refs | |
610 | */ | |
611 | leaf = path->nodes[0]; | |
dadcaf78 | 612 | slot = path->slots[0]; |
8da6d581 JS |
613 | |
614 | item_size = btrfs_item_size_nr(leaf, slot); | |
615 | BUG_ON(item_size < sizeof(*ei)); | |
616 | ||
617 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
618 | flags = btrfs_extent_flags(leaf, ei); | |
619 | ||
620 | ptr = (unsigned long)(ei + 1); | |
621 | end = (unsigned long)ei + item_size; | |
622 | ||
623 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
624 | struct btrfs_tree_block_info *info; | |
8da6d581 JS |
625 | |
626 | info = (struct btrfs_tree_block_info *)ptr; | |
627 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
628 | ptr += sizeof(struct btrfs_tree_block_info); |
629 | BUG_ON(ptr > end); | |
630 | } else { | |
631 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
632 | } | |
633 | ||
634 | while (ptr < end) { | |
635 | struct btrfs_extent_inline_ref *iref; | |
636 | u64 offset; | |
637 | int type; | |
638 | ||
639 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
640 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
641 | offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
642 | ||
643 | switch (type) { | |
644 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
d5c88b73 | 645 | ret = __add_prelim_ref(prefs, 0, NULL, |
8da6d581 JS |
646 | *info_level + 1, offset, |
647 | bytenr, 1); | |
648 | break; | |
649 | case BTRFS_SHARED_DATA_REF_KEY: { | |
650 | struct btrfs_shared_data_ref *sdref; | |
651 | int count; | |
652 | ||
653 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
654 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
655 | ret = __add_prelim_ref(prefs, 0, NULL, 0, offset, | |
656 | bytenr, count); | |
657 | break; | |
658 | } | |
659 | case BTRFS_TREE_BLOCK_REF_KEY: | |
d5c88b73 JS |
660 | ret = __add_prelim_ref(prefs, offset, NULL, |
661 | *info_level + 1, 0, | |
662 | bytenr, 1); | |
8da6d581 JS |
663 | break; |
664 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
665 | struct btrfs_extent_data_ref *dref; | |
666 | int count; | |
667 | u64 root; | |
668 | ||
669 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
670 | count = btrfs_extent_data_ref_count(leaf, dref); | |
671 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
672 | dref); | |
673 | key.type = BTRFS_EXTENT_DATA_KEY; | |
674 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
675 | root = btrfs_extent_data_ref_root(leaf, dref); | |
d5c88b73 JS |
676 | ret = __add_prelim_ref(prefs, root, &key, 0, 0, |
677 | bytenr, count); | |
8da6d581 JS |
678 | break; |
679 | } | |
680 | default: | |
681 | WARN_ON(1); | |
682 | } | |
1149ab6b WS |
683 | if (ret) |
684 | return ret; | |
8da6d581 JS |
685 | ptr += btrfs_extent_inline_ref_size(type); |
686 | } | |
687 | ||
688 | return 0; | |
689 | } | |
690 | ||
691 | /* | |
692 | * add all non-inline backrefs for bytenr to the list | |
693 | */ | |
694 | static int __add_keyed_refs(struct btrfs_fs_info *fs_info, | |
695 | struct btrfs_path *path, u64 bytenr, | |
d5c88b73 | 696 | int info_level, struct list_head *prefs) |
8da6d581 JS |
697 | { |
698 | struct btrfs_root *extent_root = fs_info->extent_root; | |
699 | int ret; | |
700 | int slot; | |
701 | struct extent_buffer *leaf; | |
702 | struct btrfs_key key; | |
703 | ||
704 | while (1) { | |
705 | ret = btrfs_next_item(extent_root, path); | |
706 | if (ret < 0) | |
707 | break; | |
708 | if (ret) { | |
709 | ret = 0; | |
710 | break; | |
711 | } | |
712 | ||
713 | slot = path->slots[0]; | |
714 | leaf = path->nodes[0]; | |
715 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
716 | ||
717 | if (key.objectid != bytenr) | |
718 | break; | |
719 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
720 | continue; | |
721 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
722 | break; | |
723 | ||
724 | switch (key.type) { | |
725 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
d5c88b73 | 726 | ret = __add_prelim_ref(prefs, 0, NULL, |
8da6d581 JS |
727 | info_level + 1, key.offset, |
728 | bytenr, 1); | |
729 | break; | |
730 | case BTRFS_SHARED_DATA_REF_KEY: { | |
731 | struct btrfs_shared_data_ref *sdref; | |
732 | int count; | |
733 | ||
734 | sdref = btrfs_item_ptr(leaf, slot, | |
735 | struct btrfs_shared_data_ref); | |
736 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
737 | ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset, | |
738 | bytenr, count); | |
739 | break; | |
740 | } | |
741 | case BTRFS_TREE_BLOCK_REF_KEY: | |
d5c88b73 JS |
742 | ret = __add_prelim_ref(prefs, key.offset, NULL, |
743 | info_level + 1, 0, | |
744 | bytenr, 1); | |
8da6d581 JS |
745 | break; |
746 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
747 | struct btrfs_extent_data_ref *dref; | |
748 | int count; | |
749 | u64 root; | |
750 | ||
751 | dref = btrfs_item_ptr(leaf, slot, | |
752 | struct btrfs_extent_data_ref); | |
753 | count = btrfs_extent_data_ref_count(leaf, dref); | |
754 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
755 | dref); | |
756 | key.type = BTRFS_EXTENT_DATA_KEY; | |
757 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
758 | root = btrfs_extent_data_ref_root(leaf, dref); | |
759 | ret = __add_prelim_ref(prefs, root, &key, 0, 0, | |
d5c88b73 | 760 | bytenr, count); |
8da6d581 JS |
761 | break; |
762 | } | |
763 | default: | |
764 | WARN_ON(1); | |
765 | } | |
1149ab6b WS |
766 | if (ret) |
767 | return ret; | |
768 | ||
8da6d581 JS |
769 | } |
770 | ||
771 | return ret; | |
772 | } | |
773 | ||
774 | /* | |
775 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
776 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
777 | * indirect refs to their parent bytenr. | |
778 | * When roots are found, they're added to the roots list | |
779 | * | |
780 | * FIXME some caching might speed things up | |
781 | */ | |
782 | static int find_parent_nodes(struct btrfs_trans_handle *trans, | |
783 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c JS |
784 | u64 time_seq, struct ulist *refs, |
785 | struct ulist *roots, const u64 *extent_item_pos) | |
8da6d581 JS |
786 | { |
787 | struct btrfs_key key; | |
788 | struct btrfs_path *path; | |
8da6d581 | 789 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 790 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
791 | int info_level = 0; |
792 | int ret; | |
7a3ae2f8 | 793 | int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT); |
8da6d581 JS |
794 | struct list_head prefs_delayed; |
795 | struct list_head prefs; | |
796 | struct __prelim_ref *ref; | |
797 | ||
798 | INIT_LIST_HEAD(&prefs); | |
799 | INIT_LIST_HEAD(&prefs_delayed); | |
800 | ||
801 | key.objectid = bytenr; | |
802 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
803 | key.offset = (u64)-1; | |
804 | ||
805 | path = btrfs_alloc_path(); | |
806 | if (!path) | |
807 | return -ENOMEM; | |
7a3ae2f8 | 808 | path->search_commit_root = !!search_commit_root; |
8da6d581 JS |
809 | |
810 | /* | |
811 | * grab both a lock on the path and a lock on the delayed ref head. | |
812 | * We need both to get a consistent picture of how the refs look | |
813 | * at a specified point in time | |
814 | */ | |
815 | again: | |
d3b01064 LZ |
816 | head = NULL; |
817 | ||
8da6d581 JS |
818 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
819 | if (ret < 0) | |
820 | goto out; | |
821 | BUG_ON(ret == 0); | |
822 | ||
7a3ae2f8 JS |
823 | if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) { |
824 | /* | |
825 | * look if there are updates for this ref queued and lock the | |
826 | * head | |
827 | */ | |
828 | delayed_refs = &trans->transaction->delayed_refs; | |
829 | spin_lock(&delayed_refs->lock); | |
830 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
831 | if (head) { | |
832 | if (!mutex_trylock(&head->mutex)) { | |
833 | atomic_inc(&head->node.refs); | |
834 | spin_unlock(&delayed_refs->lock); | |
835 | ||
836 | btrfs_release_path(path); | |
837 | ||
838 | /* | |
839 | * Mutex was contended, block until it's | |
840 | * released and try again | |
841 | */ | |
842 | mutex_lock(&head->mutex); | |
843 | mutex_unlock(&head->mutex); | |
844 | btrfs_put_delayed_ref(&head->node); | |
845 | goto again; | |
846 | } | |
097b8a7c | 847 | ret = __add_delayed_refs(head, time_seq, |
8445f61c | 848 | &prefs_delayed); |
155725c9 | 849 | mutex_unlock(&head->mutex); |
7a3ae2f8 JS |
850 | if (ret) { |
851 | spin_unlock(&delayed_refs->lock); | |
852 | goto out; | |
853 | } | |
d3b01064 | 854 | } |
7a3ae2f8 | 855 | spin_unlock(&delayed_refs->lock); |
8da6d581 | 856 | } |
8da6d581 JS |
857 | |
858 | if (path->slots[0]) { | |
859 | struct extent_buffer *leaf; | |
860 | int slot; | |
861 | ||
dadcaf78 | 862 | path->slots[0]--; |
8da6d581 | 863 | leaf = path->nodes[0]; |
dadcaf78 | 864 | slot = path->slots[0]; |
8da6d581 JS |
865 | btrfs_item_key_to_cpu(leaf, &key, slot); |
866 | if (key.objectid == bytenr && | |
867 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
868 | ret = __add_inline_refs(fs_info, path, bytenr, | |
d5c88b73 | 869 | &info_level, &prefs); |
8da6d581 JS |
870 | if (ret) |
871 | goto out; | |
d5c88b73 | 872 | ret = __add_keyed_refs(fs_info, path, bytenr, |
8da6d581 JS |
873 | info_level, &prefs); |
874 | if (ret) | |
875 | goto out; | |
876 | } | |
877 | } | |
878 | btrfs_release_path(path); | |
879 | ||
8da6d581 JS |
880 | list_splice_init(&prefs_delayed, &prefs); |
881 | ||
d5c88b73 JS |
882 | ret = __add_missing_keys(fs_info, &prefs); |
883 | if (ret) | |
884 | goto out; | |
885 | ||
692206b1 | 886 | __merge_refs(&prefs, 1); |
8da6d581 | 887 | |
8445f61c JS |
888 | ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq, |
889 | &prefs, extent_item_pos); | |
8da6d581 JS |
890 | if (ret) |
891 | goto out; | |
892 | ||
692206b1 | 893 | __merge_refs(&prefs, 2); |
8da6d581 JS |
894 | |
895 | while (!list_empty(&prefs)) { | |
896 | ref = list_first_entry(&prefs, struct __prelim_ref, list); | |
897 | list_del(&ref->list); | |
6c1500f2 | 898 | WARN_ON(ref->count < 0); |
8da6d581 JS |
899 | if (ref->count && ref->root_id && ref->parent == 0) { |
900 | /* no parent == root of tree */ | |
901 | ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); | |
f1723939 WS |
902 | if (ret < 0) |
903 | goto out; | |
8da6d581 JS |
904 | } |
905 | if (ref->count && ref->parent) { | |
976b1908 | 906 | struct extent_inode_elem *eie = NULL; |
3301958b | 907 | if (extent_item_pos && !ref->inode_list) { |
976b1908 JS |
908 | u32 bsz; |
909 | struct extent_buffer *eb; | |
910 | bsz = btrfs_level_size(fs_info->extent_root, | |
911 | info_level); | |
912 | eb = read_tree_block(fs_info->extent_root, | |
913 | ref->parent, bsz, 0); | |
914 | BUG_ON(!eb); | |
915 | ret = find_extent_in_eb(eb, bytenr, | |
916 | *extent_item_pos, &eie); | |
3301958b | 917 | ref->inode_list = eie; |
976b1908 JS |
918 | free_extent_buffer(eb); |
919 | } | |
3301958b | 920 | ret = ulist_add_merge(refs, ref->parent, |
995e01b7 | 921 | (uintptr_t)ref->inode_list, |
34d73f54 | 922 | (u64 *)&eie, GFP_NOFS); |
f1723939 WS |
923 | if (ret < 0) |
924 | goto out; | |
3301958b JS |
925 | if (!ret && extent_item_pos) { |
926 | /* | |
927 | * we've recorded that parent, so we must extend | |
928 | * its inode list here | |
929 | */ | |
930 | BUG_ON(!eie); | |
931 | while (eie->next) | |
932 | eie = eie->next; | |
933 | eie->next = ref->inode_list; | |
934 | } | |
8da6d581 JS |
935 | } |
936 | kfree(ref); | |
937 | } | |
938 | ||
939 | out: | |
8da6d581 JS |
940 | btrfs_free_path(path); |
941 | while (!list_empty(&prefs)) { | |
942 | ref = list_first_entry(&prefs, struct __prelim_ref, list); | |
943 | list_del(&ref->list); | |
944 | kfree(ref); | |
945 | } | |
946 | while (!list_empty(&prefs_delayed)) { | |
947 | ref = list_first_entry(&prefs_delayed, struct __prelim_ref, | |
948 | list); | |
949 | list_del(&ref->list); | |
950 | kfree(ref); | |
951 | } | |
952 | ||
953 | return ret; | |
954 | } | |
955 | ||
976b1908 JS |
956 | static void free_leaf_list(struct ulist *blocks) |
957 | { | |
958 | struct ulist_node *node = NULL; | |
959 | struct extent_inode_elem *eie; | |
960 | struct extent_inode_elem *eie_next; | |
961 | struct ulist_iterator uiter; | |
962 | ||
963 | ULIST_ITER_INIT(&uiter); | |
964 | while ((node = ulist_next(blocks, &uiter))) { | |
965 | if (!node->aux) | |
966 | continue; | |
995e01b7 | 967 | eie = (struct extent_inode_elem *)(uintptr_t)node->aux; |
976b1908 JS |
968 | for (; eie; eie = eie_next) { |
969 | eie_next = eie->next; | |
970 | kfree(eie); | |
971 | } | |
972 | node->aux = 0; | |
973 | } | |
974 | ||
975 | ulist_free(blocks); | |
976 | } | |
977 | ||
8da6d581 JS |
978 | /* |
979 | * Finds all leafs with a reference to the specified combination of bytenr and | |
980 | * offset. key_list_head will point to a list of corresponding keys (caller must | |
981 | * free each list element). The leafs will be stored in the leafs ulist, which | |
982 | * must be freed with ulist_free. | |
983 | * | |
984 | * returns 0 on success, <0 on error | |
985 | */ | |
986 | static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, | |
987 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 988 | u64 time_seq, struct ulist **leafs, |
976b1908 | 989 | const u64 *extent_item_pos) |
8da6d581 JS |
990 | { |
991 | struct ulist *tmp; | |
992 | int ret; | |
993 | ||
994 | tmp = ulist_alloc(GFP_NOFS); | |
995 | if (!tmp) | |
996 | return -ENOMEM; | |
997 | *leafs = ulist_alloc(GFP_NOFS); | |
998 | if (!*leafs) { | |
999 | ulist_free(tmp); | |
1000 | return -ENOMEM; | |
1001 | } | |
1002 | ||
097b8a7c | 1003 | ret = find_parent_nodes(trans, fs_info, bytenr, |
8445f61c | 1004 | time_seq, *leafs, tmp, extent_item_pos); |
8da6d581 JS |
1005 | ulist_free(tmp); |
1006 | ||
1007 | if (ret < 0 && ret != -ENOENT) { | |
976b1908 | 1008 | free_leaf_list(*leafs); |
8da6d581 JS |
1009 | return ret; |
1010 | } | |
1011 | ||
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * walk all backrefs for a given extent to find all roots that reference this | |
1017 | * extent. Walking a backref means finding all extents that reference this | |
1018 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1019 | * recursive process, but here it is implemented in an iterative fashion: We | |
1020 | * find all referencing extents for the extent in question and put them on a | |
1021 | * list. In turn, we find all referencing extents for those, further appending | |
1022 | * to the list. The way we iterate the list allows adding more elements after | |
1023 | * the current while iterating. The process stops when we reach the end of the | |
1024 | * list. Found roots are added to the roots list. | |
1025 | * | |
1026 | * returns 0 on success, < 0 on error. | |
1027 | */ | |
1028 | int btrfs_find_all_roots(struct btrfs_trans_handle *trans, | |
1029 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1030 | u64 time_seq, struct ulist **roots) |
8da6d581 JS |
1031 | { |
1032 | struct ulist *tmp; | |
1033 | struct ulist_node *node = NULL; | |
cd1b413c | 1034 | struct ulist_iterator uiter; |
8da6d581 JS |
1035 | int ret; |
1036 | ||
1037 | tmp = ulist_alloc(GFP_NOFS); | |
1038 | if (!tmp) | |
1039 | return -ENOMEM; | |
1040 | *roots = ulist_alloc(GFP_NOFS); | |
1041 | if (!*roots) { | |
1042 | ulist_free(tmp); | |
1043 | return -ENOMEM; | |
1044 | } | |
1045 | ||
cd1b413c | 1046 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1047 | while (1) { |
097b8a7c | 1048 | ret = find_parent_nodes(trans, fs_info, bytenr, |
8445f61c | 1049 | time_seq, tmp, *roots, NULL); |
8da6d581 JS |
1050 | if (ret < 0 && ret != -ENOENT) { |
1051 | ulist_free(tmp); | |
1052 | ulist_free(*roots); | |
1053 | return ret; | |
1054 | } | |
cd1b413c | 1055 | node = ulist_next(tmp, &uiter); |
8da6d581 JS |
1056 | if (!node) |
1057 | break; | |
1058 | bytenr = node->val; | |
1059 | } | |
1060 | ||
1061 | ulist_free(tmp); | |
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | ||
a542ad1b JS |
1066 | static int __inode_info(u64 inum, u64 ioff, u8 key_type, |
1067 | struct btrfs_root *fs_root, struct btrfs_path *path, | |
1068 | struct btrfs_key *found_key) | |
1069 | { | |
1070 | int ret; | |
1071 | struct btrfs_key key; | |
1072 | struct extent_buffer *eb; | |
1073 | ||
1074 | key.type = key_type; | |
1075 | key.objectid = inum; | |
1076 | key.offset = ioff; | |
1077 | ||
1078 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1079 | if (ret < 0) | |
1080 | return ret; | |
1081 | ||
1082 | eb = path->nodes[0]; | |
1083 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
1084 | ret = btrfs_next_leaf(fs_root, path); | |
1085 | if (ret) | |
1086 | return ret; | |
1087 | eb = path->nodes[0]; | |
1088 | } | |
1089 | ||
1090 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
1091 | if (found_key->type != key.type || found_key->objectid != key.objectid) | |
1092 | return 1; | |
1093 | ||
1094 | return 0; | |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * this makes the path point to (inum INODE_ITEM ioff) | |
1099 | */ | |
1100 | int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, | |
1101 | struct btrfs_path *path) | |
1102 | { | |
1103 | struct btrfs_key key; | |
1104 | return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path, | |
1105 | &key); | |
1106 | } | |
1107 | ||
1108 | static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, | |
1109 | struct btrfs_path *path, | |
1110 | struct btrfs_key *found_key) | |
1111 | { | |
1112 | return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path, | |
1113 | found_key); | |
1114 | } | |
1115 | ||
f186373f MF |
1116 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
1117 | u64 start_off, struct btrfs_path *path, | |
1118 | struct btrfs_inode_extref **ret_extref, | |
1119 | u64 *found_off) | |
1120 | { | |
1121 | int ret, slot; | |
1122 | struct btrfs_key key; | |
1123 | struct btrfs_key found_key; | |
1124 | struct btrfs_inode_extref *extref; | |
1125 | struct extent_buffer *leaf; | |
1126 | unsigned long ptr; | |
1127 | ||
1128 | key.objectid = inode_objectid; | |
1129 | btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY); | |
1130 | key.offset = start_off; | |
1131 | ||
1132 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1133 | if (ret < 0) | |
1134 | return ret; | |
1135 | ||
1136 | while (1) { | |
1137 | leaf = path->nodes[0]; | |
1138 | slot = path->slots[0]; | |
1139 | if (slot >= btrfs_header_nritems(leaf)) { | |
1140 | /* | |
1141 | * If the item at offset is not found, | |
1142 | * btrfs_search_slot will point us to the slot | |
1143 | * where it should be inserted. In our case | |
1144 | * that will be the slot directly before the | |
1145 | * next INODE_REF_KEY_V2 item. In the case | |
1146 | * that we're pointing to the last slot in a | |
1147 | * leaf, we must move one leaf over. | |
1148 | */ | |
1149 | ret = btrfs_next_leaf(root, path); | |
1150 | if (ret) { | |
1151 | if (ret >= 1) | |
1152 | ret = -ENOENT; | |
1153 | break; | |
1154 | } | |
1155 | continue; | |
1156 | } | |
1157 | ||
1158 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1159 | ||
1160 | /* | |
1161 | * Check that we're still looking at an extended ref key for | |
1162 | * this particular objectid. If we have different | |
1163 | * objectid or type then there are no more to be found | |
1164 | * in the tree and we can exit. | |
1165 | */ | |
1166 | ret = -ENOENT; | |
1167 | if (found_key.objectid != inode_objectid) | |
1168 | break; | |
1169 | if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY) | |
1170 | break; | |
1171 | ||
1172 | ret = 0; | |
1173 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1174 | extref = (struct btrfs_inode_extref *)ptr; | |
1175 | *ret_extref = extref; | |
1176 | if (found_off) | |
1177 | *found_off = found_key.offset; | |
1178 | break; | |
1179 | } | |
1180 | ||
1181 | return ret; | |
1182 | } | |
1183 | ||
96b5bd77 JS |
1184 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
1185 | u32 name_len, unsigned long name_off, | |
1186 | struct extent_buffer *eb_in, u64 parent, | |
1187 | char *dest, u32 size) | |
a542ad1b | 1188 | { |
a542ad1b JS |
1189 | int slot; |
1190 | u64 next_inum; | |
1191 | int ret; | |
661bec6b | 1192 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
1193 | struct extent_buffer *eb = eb_in; |
1194 | struct btrfs_key found_key; | |
b916a59a | 1195 | int leave_spinning = path->leave_spinning; |
d24bec3a | 1196 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
1197 | |
1198 | if (bytes_left >= 0) | |
1199 | dest[bytes_left] = '\0'; | |
1200 | ||
b916a59a | 1201 | path->leave_spinning = 1; |
a542ad1b | 1202 | while (1) { |
d24bec3a | 1203 | bytes_left -= name_len; |
a542ad1b JS |
1204 | if (bytes_left >= 0) |
1205 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 1206 | name_off, name_len); |
b916a59a JS |
1207 | if (eb != eb_in) { |
1208 | btrfs_tree_read_unlock_blocking(eb); | |
a542ad1b | 1209 | free_extent_buffer(eb); |
b916a59a | 1210 | } |
a542ad1b | 1211 | ret = inode_ref_info(parent, 0, fs_root, path, &found_key); |
8f24b496 JS |
1212 | if (ret > 0) |
1213 | ret = -ENOENT; | |
a542ad1b JS |
1214 | if (ret) |
1215 | break; | |
d24bec3a | 1216 | |
a542ad1b JS |
1217 | next_inum = found_key.offset; |
1218 | ||
1219 | /* regular exit ahead */ | |
1220 | if (parent == next_inum) | |
1221 | break; | |
1222 | ||
1223 | slot = path->slots[0]; | |
1224 | eb = path->nodes[0]; | |
1225 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 1226 | if (eb != eb_in) { |
a542ad1b | 1227 | atomic_inc(&eb->refs); |
b916a59a JS |
1228 | btrfs_tree_read_lock(eb); |
1229 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1230 | } | |
a542ad1b | 1231 | btrfs_release_path(path); |
a542ad1b | 1232 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
1233 | |
1234 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1235 | name_off = (unsigned long)(iref + 1); | |
1236 | ||
a542ad1b JS |
1237 | parent = next_inum; |
1238 | --bytes_left; | |
1239 | if (bytes_left >= 0) | |
1240 | dest[bytes_left] = '/'; | |
1241 | } | |
1242 | ||
1243 | btrfs_release_path(path); | |
b916a59a | 1244 | path->leave_spinning = leave_spinning; |
a542ad1b JS |
1245 | |
1246 | if (ret) | |
1247 | return ERR_PTR(ret); | |
1248 | ||
1249 | return dest + bytes_left; | |
1250 | } | |
1251 | ||
d24bec3a MF |
1252 | /* |
1253 | * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements | |
1254 | * of the path are separated by '/' and the path is guaranteed to be | |
1255 | * 0-terminated. the path is only given within the current file system. | |
1256 | * Therefore, it never starts with a '/'. the caller is responsible to provide | |
1257 | * "size" bytes in "dest". the dest buffer will be filled backwards. finally, | |
1258 | * the start point of the resulting string is returned. this pointer is within | |
1259 | * dest, normally. | |
1260 | * in case the path buffer would overflow, the pointer is decremented further | |
1261 | * as if output was written to the buffer, though no more output is actually | |
1262 | * generated. that way, the caller can determine how much space would be | |
1263 | * required for the path to fit into the buffer. in that case, the returned | |
1264 | * value will be smaller than dest. callers must check this! | |
1265 | */ | |
1266 | char *btrfs_iref_to_path(struct btrfs_root *fs_root, | |
1267 | struct btrfs_path *path, | |
1268 | struct btrfs_inode_ref *iref, | |
1269 | struct extent_buffer *eb_in, u64 parent, | |
1270 | char *dest, u32 size) | |
1271 | { | |
96b5bd77 JS |
1272 | return btrfs_ref_to_path(fs_root, path, |
1273 | btrfs_inode_ref_name_len(eb_in, iref), | |
1274 | (unsigned long)(iref + 1), | |
1275 | eb_in, parent, dest, size); | |
d24bec3a MF |
1276 | } |
1277 | ||
a542ad1b JS |
1278 | /* |
1279 | * this makes the path point to (logical EXTENT_ITEM *) | |
1280 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
1281 | * tree blocks and <0 on error. | |
1282 | */ | |
1283 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
1284 | struct btrfs_path *path, struct btrfs_key *found_key, |
1285 | u64 *flags_ret) | |
a542ad1b JS |
1286 | { |
1287 | int ret; | |
1288 | u64 flags; | |
1289 | u32 item_size; | |
1290 | struct extent_buffer *eb; | |
1291 | struct btrfs_extent_item *ei; | |
1292 | struct btrfs_key key; | |
1293 | ||
1294 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1295 | key.objectid = logical; | |
1296 | key.offset = (u64)-1; | |
1297 | ||
1298 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); | |
1299 | if (ret < 0) | |
1300 | return ret; | |
1301 | ret = btrfs_previous_item(fs_info->extent_root, path, | |
1302 | 0, BTRFS_EXTENT_ITEM_KEY); | |
1303 | if (ret < 0) | |
1304 | return ret; | |
1305 | ||
1306 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); | |
1307 | if (found_key->type != BTRFS_EXTENT_ITEM_KEY || | |
1308 | found_key->objectid > logical || | |
4692cf58 JS |
1309 | found_key->objectid + found_key->offset <= logical) { |
1310 | pr_debug("logical %llu is not within any extent\n", | |
1311 | (unsigned long long)logical); | |
a542ad1b | 1312 | return -ENOENT; |
4692cf58 | 1313 | } |
a542ad1b JS |
1314 | |
1315 | eb = path->nodes[0]; | |
1316 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
1317 | BUG_ON(item_size < sizeof(*ei)); | |
1318 | ||
1319 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
1320 | flags = btrfs_extent_flags(eb, ei); | |
1321 | ||
4692cf58 JS |
1322 | pr_debug("logical %llu is at position %llu within the extent (%llu " |
1323 | "EXTENT_ITEM %llu) flags %#llx size %u\n", | |
1324 | (unsigned long long)logical, | |
1325 | (unsigned long long)(logical - found_key->objectid), | |
1326 | (unsigned long long)found_key->objectid, | |
1327 | (unsigned long long)found_key->offset, | |
1328 | (unsigned long long)flags, item_size); | |
69917e43 LB |
1329 | |
1330 | WARN_ON(!flags_ret); | |
1331 | if (flags_ret) { | |
1332 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1333 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
1334 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1335 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
1336 | else | |
1337 | BUG_ON(1); | |
1338 | return 0; | |
1339 | } | |
a542ad1b JS |
1340 | |
1341 | return -EIO; | |
1342 | } | |
1343 | ||
1344 | /* | |
1345 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
1346 | * for the first call and may be modified. it is used to track state. | |
1347 | * if more refs exist, 0 is returned and the next call to | |
1348 | * __get_extent_inline_ref must pass the modified ptr parameter to get the | |
1349 | * next ref. after the last ref was processed, 1 is returned. | |
1350 | * returns <0 on error | |
1351 | */ | |
1352 | static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, | |
1353 | struct btrfs_extent_item *ei, u32 item_size, | |
1354 | struct btrfs_extent_inline_ref **out_eiref, | |
1355 | int *out_type) | |
1356 | { | |
1357 | unsigned long end; | |
1358 | u64 flags; | |
1359 | struct btrfs_tree_block_info *info; | |
1360 | ||
1361 | if (!*ptr) { | |
1362 | /* first call */ | |
1363 | flags = btrfs_extent_flags(eb, ei); | |
1364 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1365 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1366 | *out_eiref = | |
1367 | (struct btrfs_extent_inline_ref *)(info + 1); | |
1368 | } else { | |
1369 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
1370 | } | |
1371 | *ptr = (unsigned long)*out_eiref; | |
1372 | if ((void *)*ptr >= (void *)ei + item_size) | |
1373 | return -ENOENT; | |
1374 | } | |
1375 | ||
1376 | end = (unsigned long)ei + item_size; | |
1377 | *out_eiref = (struct btrfs_extent_inline_ref *)*ptr; | |
1378 | *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref); | |
1379 | ||
1380 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
1381 | WARN_ON(*ptr > end); | |
1382 | if (*ptr == end) | |
1383 | return 1; /* last */ | |
1384 | ||
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * reads the tree block backref for an extent. tree level and root are returned | |
1390 | * through out_level and out_root. ptr must point to a 0 value for the first | |
1391 | * call and may be modified (see __get_extent_inline_ref comment). | |
1392 | * returns 0 if data was provided, 1 if there was no more data to provide or | |
1393 | * <0 on error. | |
1394 | */ | |
1395 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
1396 | struct btrfs_extent_item *ei, u32 item_size, | |
1397 | u64 *out_root, u8 *out_level) | |
1398 | { | |
1399 | int ret; | |
1400 | int type; | |
1401 | struct btrfs_tree_block_info *info; | |
1402 | struct btrfs_extent_inline_ref *eiref; | |
1403 | ||
1404 | if (*ptr == (unsigned long)-1) | |
1405 | return 1; | |
1406 | ||
1407 | while (1) { | |
1408 | ret = __get_extent_inline_ref(ptr, eb, ei, item_size, | |
1409 | &eiref, &type); | |
1410 | if (ret < 0) | |
1411 | return ret; | |
1412 | ||
1413 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1414 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1415 | break; | |
1416 | ||
1417 | if (ret == 1) | |
1418 | return 1; | |
1419 | } | |
1420 | ||
1421 | /* we can treat both ref types equally here */ | |
1422 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1423 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); | |
1424 | *out_level = btrfs_tree_block_level(eb, info); | |
1425 | ||
1426 | if (ret == 1) | |
1427 | *ptr = (unsigned long)-1; | |
1428 | ||
1429 | return 0; | |
1430 | } | |
1431 | ||
976b1908 JS |
1432 | static int iterate_leaf_refs(struct extent_inode_elem *inode_list, |
1433 | u64 root, u64 extent_item_objectid, | |
4692cf58 | 1434 | iterate_extent_inodes_t *iterate, void *ctx) |
a542ad1b | 1435 | { |
976b1908 | 1436 | struct extent_inode_elem *eie; |
4692cf58 | 1437 | int ret = 0; |
4692cf58 | 1438 | |
976b1908 | 1439 | for (eie = inode_list; eie; eie = eie->next) { |
4692cf58 | 1440 | pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), " |
976b1908 JS |
1441 | "root %llu\n", extent_item_objectid, |
1442 | eie->inum, eie->offset, root); | |
1443 | ret = iterate(eie->inum, eie->offset, root, ctx); | |
4692cf58 | 1444 | if (ret) { |
976b1908 JS |
1445 | pr_debug("stopping iteration for %llu due to ret=%d\n", |
1446 | extent_item_objectid, ret); | |
4692cf58 JS |
1447 | break; |
1448 | } | |
a542ad1b JS |
1449 | } |
1450 | ||
a542ad1b JS |
1451 | return ret; |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 1456 | * the given parameters. |
a542ad1b JS |
1457 | * when the iterator function returns a non-zero value, iteration stops. |
1458 | */ | |
1459 | int iterate_extent_inodes(struct btrfs_fs_info *fs_info, | |
4692cf58 | 1460 | u64 extent_item_objectid, u64 extent_item_pos, |
7a3ae2f8 | 1461 | int search_commit_root, |
a542ad1b JS |
1462 | iterate_extent_inodes_t *iterate, void *ctx) |
1463 | { | |
a542ad1b | 1464 | int ret; |
a542ad1b JS |
1465 | struct list_head data_refs = LIST_HEAD_INIT(data_refs); |
1466 | struct list_head shared_refs = LIST_HEAD_INIT(shared_refs); | |
4692cf58 | 1467 | struct btrfs_trans_handle *trans; |
7a3ae2f8 JS |
1468 | struct ulist *refs = NULL; |
1469 | struct ulist *roots = NULL; | |
4692cf58 JS |
1470 | struct ulist_node *ref_node = NULL; |
1471 | struct ulist_node *root_node = NULL; | |
8445f61c | 1472 | struct seq_list tree_mod_seq_elem = {}; |
cd1b413c JS |
1473 | struct ulist_iterator ref_uiter; |
1474 | struct ulist_iterator root_uiter; | |
a542ad1b | 1475 | |
4692cf58 JS |
1476 | pr_debug("resolving all inodes for extent %llu\n", |
1477 | extent_item_objectid); | |
a542ad1b | 1478 | |
7a3ae2f8 JS |
1479 | if (search_commit_root) { |
1480 | trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT; | |
1481 | } else { | |
1482 | trans = btrfs_join_transaction(fs_info->extent_root); | |
1483 | if (IS_ERR(trans)) | |
1484 | return PTR_ERR(trans); | |
8445f61c | 1485 | btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
7a3ae2f8 | 1486 | } |
a542ad1b | 1487 | |
4692cf58 | 1488 | ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, |
097b8a7c | 1489 | tree_mod_seq_elem.seq, &refs, |
8445f61c | 1490 | &extent_item_pos); |
4692cf58 JS |
1491 | if (ret) |
1492 | goto out; | |
a542ad1b | 1493 | |
cd1b413c JS |
1494 | ULIST_ITER_INIT(&ref_uiter); |
1495 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
976b1908 | 1496 | ret = btrfs_find_all_roots(trans, fs_info, ref_node->val, |
097b8a7c | 1497 | tree_mod_seq_elem.seq, &roots); |
4692cf58 JS |
1498 | if (ret) |
1499 | break; | |
cd1b413c JS |
1500 | ULIST_ITER_INIT(&root_uiter); |
1501 | while (!ret && (root_node = ulist_next(roots, &root_uiter))) { | |
976b1908 | 1502 | pr_debug("root %llu references leaf %llu, data list " |
34d73f54 | 1503 | "%#llx\n", root_node->val, ref_node->val, |
995e01b7 JS |
1504 | (long long)ref_node->aux); |
1505 | ret = iterate_leaf_refs((struct extent_inode_elem *) | |
1506 | (uintptr_t)ref_node->aux, | |
1507 | root_node->val, | |
1508 | extent_item_objectid, | |
1509 | iterate, ctx); | |
4692cf58 | 1510 | } |
976b1908 | 1511 | ulist_free(roots); |
a542ad1b JS |
1512 | } |
1513 | ||
976b1908 | 1514 | free_leaf_list(refs); |
4692cf58 | 1515 | out: |
7a3ae2f8 | 1516 | if (!search_commit_root) { |
8445f61c | 1517 | btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
7a3ae2f8 JS |
1518 | btrfs_end_transaction(trans, fs_info->extent_root); |
1519 | } | |
1520 | ||
a542ad1b JS |
1521 | return ret; |
1522 | } | |
1523 | ||
1524 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, | |
1525 | struct btrfs_path *path, | |
1526 | iterate_extent_inodes_t *iterate, void *ctx) | |
1527 | { | |
1528 | int ret; | |
4692cf58 | 1529 | u64 extent_item_pos; |
69917e43 | 1530 | u64 flags = 0; |
a542ad1b | 1531 | struct btrfs_key found_key; |
7a3ae2f8 | 1532 | int search_commit_root = path->search_commit_root; |
a542ad1b | 1533 | |
69917e43 | 1534 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 1535 | btrfs_release_path(path); |
a542ad1b JS |
1536 | if (ret < 0) |
1537 | return ret; | |
69917e43 | 1538 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 1539 | return -EINVAL; |
a542ad1b | 1540 | |
4692cf58 | 1541 | extent_item_pos = logical - found_key.objectid; |
7a3ae2f8 JS |
1542 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1543 | extent_item_pos, search_commit_root, | |
1544 | iterate, ctx); | |
a542ad1b JS |
1545 | |
1546 | return ret; | |
1547 | } | |
1548 | ||
d24bec3a MF |
1549 | typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off, |
1550 | struct extent_buffer *eb, void *ctx); | |
1551 | ||
1552 | static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, | |
1553 | struct btrfs_path *path, | |
1554 | iterate_irefs_t *iterate, void *ctx) | |
a542ad1b | 1555 | { |
aefc1eb1 | 1556 | int ret = 0; |
a542ad1b JS |
1557 | int slot; |
1558 | u32 cur; | |
1559 | u32 len; | |
1560 | u32 name_len; | |
1561 | u64 parent = 0; | |
1562 | int found = 0; | |
1563 | struct extent_buffer *eb; | |
1564 | struct btrfs_item *item; | |
1565 | struct btrfs_inode_ref *iref; | |
1566 | struct btrfs_key found_key; | |
1567 | ||
aefc1eb1 | 1568 | while (!ret) { |
b916a59a | 1569 | path->leave_spinning = 1; |
a542ad1b | 1570 | ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path, |
d24bec3a | 1571 | &found_key); |
a542ad1b JS |
1572 | if (ret < 0) |
1573 | break; | |
1574 | if (ret) { | |
1575 | ret = found ? 0 : -ENOENT; | |
1576 | break; | |
1577 | } | |
1578 | ++found; | |
1579 | ||
1580 | parent = found_key.offset; | |
1581 | slot = path->slots[0]; | |
1582 | eb = path->nodes[0]; | |
1583 | /* make sure we can use eb after releasing the path */ | |
1584 | atomic_inc(&eb->refs); | |
b916a59a JS |
1585 | btrfs_tree_read_lock(eb); |
1586 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
a542ad1b JS |
1587 | btrfs_release_path(path); |
1588 | ||
1589 | item = btrfs_item_nr(eb, slot); | |
1590 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); | |
1591 | ||
1592 | for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) { | |
1593 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1594 | /* path must be released before calling iterate()! */ | |
4692cf58 JS |
1595 | pr_debug("following ref at offset %u for inode %llu in " |
1596 | "tree %llu\n", cur, | |
1597 | (unsigned long long)found_key.objectid, | |
1598 | (unsigned long long)fs_root->objectid); | |
d24bec3a MF |
1599 | ret = iterate(parent, name_len, |
1600 | (unsigned long)(iref + 1), eb, ctx); | |
aefc1eb1 | 1601 | if (ret) |
a542ad1b | 1602 | break; |
a542ad1b JS |
1603 | len = sizeof(*iref) + name_len; |
1604 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
1605 | } | |
b916a59a | 1606 | btrfs_tree_read_unlock_blocking(eb); |
a542ad1b JS |
1607 | free_extent_buffer(eb); |
1608 | } | |
1609 | ||
1610 | btrfs_release_path(path); | |
1611 | ||
1612 | return ret; | |
1613 | } | |
1614 | ||
d24bec3a MF |
1615 | static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, |
1616 | struct btrfs_path *path, | |
1617 | iterate_irefs_t *iterate, void *ctx) | |
1618 | { | |
1619 | int ret; | |
1620 | int slot; | |
1621 | u64 offset = 0; | |
1622 | u64 parent; | |
1623 | int found = 0; | |
1624 | struct extent_buffer *eb; | |
1625 | struct btrfs_inode_extref *extref; | |
1626 | struct extent_buffer *leaf; | |
1627 | u32 item_size; | |
1628 | u32 cur_offset; | |
1629 | unsigned long ptr; | |
1630 | ||
1631 | while (1) { | |
1632 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
1633 | &offset); | |
1634 | if (ret < 0) | |
1635 | break; | |
1636 | if (ret) { | |
1637 | ret = found ? 0 : -ENOENT; | |
1638 | break; | |
1639 | } | |
1640 | ++found; | |
1641 | ||
1642 | slot = path->slots[0]; | |
1643 | eb = path->nodes[0]; | |
1644 | /* make sure we can use eb after releasing the path */ | |
1645 | atomic_inc(&eb->refs); | |
1646 | ||
1647 | btrfs_tree_read_lock(eb); | |
1648 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1649 | btrfs_release_path(path); | |
1650 | ||
1651 | leaf = path->nodes[0]; | |
1652 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1653 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1654 | cur_offset = 0; | |
1655 | ||
1656 | while (cur_offset < item_size) { | |
1657 | u32 name_len; | |
1658 | ||
1659 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
1660 | parent = btrfs_inode_extref_parent(eb, extref); | |
1661 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
1662 | ret = iterate(parent, name_len, | |
1663 | (unsigned long)&extref->name, eb, ctx); | |
1664 | if (ret) | |
1665 | break; | |
1666 | ||
1667 | cur_offset += btrfs_inode_extref_name_len(leaf, extref); | |
1668 | cur_offset += sizeof(*extref); | |
1669 | } | |
1670 | btrfs_tree_read_unlock_blocking(eb); | |
1671 | free_extent_buffer(eb); | |
1672 | ||
1673 | offset++; | |
1674 | } | |
1675 | ||
1676 | btrfs_release_path(path); | |
1677 | ||
1678 | return ret; | |
1679 | } | |
1680 | ||
1681 | static int iterate_irefs(u64 inum, struct btrfs_root *fs_root, | |
1682 | struct btrfs_path *path, iterate_irefs_t *iterate, | |
1683 | void *ctx) | |
1684 | { | |
1685 | int ret; | |
1686 | int found_refs = 0; | |
1687 | ||
1688 | ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx); | |
1689 | if (!ret) | |
1690 | ++found_refs; | |
1691 | else if (ret != -ENOENT) | |
1692 | return ret; | |
1693 | ||
1694 | ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx); | |
1695 | if (ret == -ENOENT && found_refs) | |
1696 | return 0; | |
1697 | ||
1698 | return ret; | |
1699 | } | |
1700 | ||
a542ad1b JS |
1701 | /* |
1702 | * returns 0 if the path could be dumped (probably truncated) | |
1703 | * returns <0 in case of an error | |
1704 | */ | |
d24bec3a MF |
1705 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
1706 | struct extent_buffer *eb, void *ctx) | |
a542ad1b JS |
1707 | { |
1708 | struct inode_fs_paths *ipath = ctx; | |
1709 | char *fspath; | |
1710 | char *fspath_min; | |
1711 | int i = ipath->fspath->elem_cnt; | |
1712 | const int s_ptr = sizeof(char *); | |
1713 | u32 bytes_left; | |
1714 | ||
1715 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
1716 | ipath->fspath->bytes_left - s_ptr : 0; | |
1717 | ||
740c3d22 | 1718 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
1719 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
1720 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
1721 | if (IS_ERR(fspath)) |
1722 | return PTR_ERR(fspath); | |
1723 | ||
1724 | if (fspath > fspath_min) { | |
745c4d8e | 1725 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
1726 | ++ipath->fspath->elem_cnt; |
1727 | ipath->fspath->bytes_left = fspath - fspath_min; | |
1728 | } else { | |
1729 | ++ipath->fspath->elem_missed; | |
1730 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
1731 | ipath->fspath->bytes_left = 0; | |
1732 | } | |
1733 | ||
1734 | return 0; | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * this dumps all file system paths to the inode into the ipath struct, provided | |
1739 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 1740 | * from ipath->fspath->val[i]. |
a542ad1b | 1741 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 1742 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
a542ad1b JS |
1743 | * number of missed paths in recored in ipath->fspath->elem_missed, otherwise, |
1744 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would | |
1745 | * have been needed to return all paths. | |
1746 | */ | |
1747 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
1748 | { | |
1749 | return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path, | |
d24bec3a | 1750 | inode_to_path, ipath); |
a542ad1b JS |
1751 | } |
1752 | ||
a542ad1b JS |
1753 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
1754 | { | |
1755 | struct btrfs_data_container *data; | |
1756 | size_t alloc_bytes; | |
1757 | ||
1758 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
425d17a2 | 1759 | data = vmalloc(alloc_bytes); |
a542ad1b JS |
1760 | if (!data) |
1761 | return ERR_PTR(-ENOMEM); | |
1762 | ||
1763 | if (total_bytes >= sizeof(*data)) { | |
1764 | data->bytes_left = total_bytes - sizeof(*data); | |
1765 | data->bytes_missing = 0; | |
1766 | } else { | |
1767 | data->bytes_missing = sizeof(*data) - total_bytes; | |
1768 | data->bytes_left = 0; | |
1769 | } | |
1770 | ||
1771 | data->elem_cnt = 0; | |
1772 | data->elem_missed = 0; | |
1773 | ||
1774 | return data; | |
1775 | } | |
1776 | ||
1777 | /* | |
1778 | * allocates space to return multiple file system paths for an inode. | |
1779 | * total_bytes to allocate are passed, note that space usable for actual path | |
1780 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
1781 | * the returned pointer must be freed with free_ipath() in the end. | |
1782 | */ | |
1783 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
1784 | struct btrfs_path *path) | |
1785 | { | |
1786 | struct inode_fs_paths *ifp; | |
1787 | struct btrfs_data_container *fspath; | |
1788 | ||
1789 | fspath = init_data_container(total_bytes); | |
1790 | if (IS_ERR(fspath)) | |
1791 | return (void *)fspath; | |
1792 | ||
1793 | ifp = kmalloc(sizeof(*ifp), GFP_NOFS); | |
1794 | if (!ifp) { | |
1795 | kfree(fspath); | |
1796 | return ERR_PTR(-ENOMEM); | |
1797 | } | |
1798 | ||
1799 | ifp->btrfs_path = path; | |
1800 | ifp->fspath = fspath; | |
1801 | ifp->fs_root = fs_root; | |
1802 | ||
1803 | return ifp; | |
1804 | } | |
1805 | ||
1806 | void free_ipath(struct inode_fs_paths *ipath) | |
1807 | { | |
4735fb28 JJ |
1808 | if (!ipath) |
1809 | return; | |
425d17a2 | 1810 | vfree(ipath->fspath); |
a542ad1b JS |
1811 | kfree(ipath); |
1812 | } |