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 | { | |
8ca15e05 | 39 | u64 offset = 0; |
976b1908 JS |
40 | struct extent_inode_elem *e; |
41 | ||
8ca15e05 JB |
42 | if (!btrfs_file_extent_compression(eb, fi) && |
43 | !btrfs_file_extent_encryption(eb, fi) && | |
44 | !btrfs_file_extent_other_encoding(eb, fi)) { | |
45 | u64 data_offset; | |
46 | u64 data_len; | |
976b1908 | 47 | |
8ca15e05 JB |
48 | data_offset = btrfs_file_extent_offset(eb, fi); |
49 | data_len = btrfs_file_extent_num_bytes(eb, fi); | |
50 | ||
51 | if (extent_item_pos < data_offset || | |
52 | extent_item_pos >= data_offset + data_len) | |
53 | return 1; | |
54 | offset = extent_item_pos - data_offset; | |
55 | } | |
976b1908 JS |
56 | |
57 | e = kmalloc(sizeof(*e), GFP_NOFS); | |
58 | if (!e) | |
59 | return -ENOMEM; | |
60 | ||
61 | e->next = *eie; | |
62 | e->inum = key->objectid; | |
8ca15e05 | 63 | e->offset = key->offset + offset; |
976b1908 JS |
64 | *eie = e; |
65 | ||
66 | return 0; | |
67 | } | |
68 | ||
f05c4746 WS |
69 | static void free_inode_elem_list(struct extent_inode_elem *eie) |
70 | { | |
71 | struct extent_inode_elem *eie_next; | |
72 | ||
73 | for (; eie; eie = eie_next) { | |
74 | eie_next = eie->next; | |
75 | kfree(eie); | |
76 | } | |
77 | } | |
78 | ||
976b1908 JS |
79 | static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte, |
80 | u64 extent_item_pos, | |
81 | struct extent_inode_elem **eie) | |
82 | { | |
83 | u64 disk_byte; | |
84 | struct btrfs_key key; | |
85 | struct btrfs_file_extent_item *fi; | |
86 | int slot; | |
87 | int nritems; | |
88 | int extent_type; | |
89 | int ret; | |
90 | ||
91 | /* | |
92 | * from the shared data ref, we only have the leaf but we need | |
93 | * the key. thus, we must look into all items and see that we | |
94 | * find one (some) with a reference to our extent item. | |
95 | */ | |
96 | nritems = btrfs_header_nritems(eb); | |
97 | for (slot = 0; slot < nritems; ++slot) { | |
98 | btrfs_item_key_to_cpu(eb, &key, slot); | |
99 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
100 | continue; | |
101 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
102 | extent_type = btrfs_file_extent_type(eb, fi); | |
103 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
104 | continue; | |
105 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
106 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
107 | if (disk_byte != wanted_disk_byte) | |
108 | continue; | |
109 | ||
110 | ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie); | |
111 | if (ret < 0) | |
112 | return ret; | |
113 | } | |
114 | ||
115 | return 0; | |
116 | } | |
117 | ||
8da6d581 JS |
118 | /* |
119 | * this structure records all encountered refs on the way up to the root | |
120 | */ | |
121 | struct __prelim_ref { | |
122 | struct list_head list; | |
123 | u64 root_id; | |
d5c88b73 | 124 | struct btrfs_key key_for_search; |
8da6d581 JS |
125 | int level; |
126 | int count; | |
3301958b | 127 | struct extent_inode_elem *inode_list; |
8da6d581 JS |
128 | u64 parent; |
129 | u64 wanted_disk_byte; | |
130 | }; | |
131 | ||
b9e9a6cb WS |
132 | static struct kmem_cache *btrfs_prelim_ref_cache; |
133 | ||
134 | int __init btrfs_prelim_ref_init(void) | |
135 | { | |
136 | btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", | |
137 | sizeof(struct __prelim_ref), | |
138 | 0, | |
139 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
140 | NULL); | |
141 | if (!btrfs_prelim_ref_cache) | |
142 | return -ENOMEM; | |
143 | return 0; | |
144 | } | |
145 | ||
146 | void btrfs_prelim_ref_exit(void) | |
147 | { | |
148 | if (btrfs_prelim_ref_cache) | |
149 | kmem_cache_destroy(btrfs_prelim_ref_cache); | |
150 | } | |
151 | ||
d5c88b73 JS |
152 | /* |
153 | * the rules for all callers of this function are: | |
154 | * - obtaining the parent is the goal | |
155 | * - if you add a key, you must know that it is a correct key | |
156 | * - if you cannot add the parent or a correct key, then we will look into the | |
157 | * block later to set a correct key | |
158 | * | |
159 | * delayed refs | |
160 | * ============ | |
161 | * backref type | shared | indirect | shared | indirect | |
162 | * information | tree | tree | data | data | |
163 | * --------------------+--------+----------+--------+---------- | |
164 | * parent logical | y | - | - | - | |
165 | * key to resolve | - | y | y | y | |
166 | * tree block logical | - | - | - | - | |
167 | * root for resolving | y | y | y | y | |
168 | * | |
169 | * - column 1: we've the parent -> done | |
170 | * - column 2, 3, 4: we use the key to find the parent | |
171 | * | |
172 | * on disk refs (inline or keyed) | |
173 | * ============================== | |
174 | * backref type | shared | indirect | shared | indirect | |
175 | * information | tree | tree | data | data | |
176 | * --------------------+--------+----------+--------+---------- | |
177 | * parent logical | y | - | y | - | |
178 | * key to resolve | - | - | - | y | |
179 | * tree block logical | y | y | y | y | |
180 | * root for resolving | - | y | y | y | |
181 | * | |
182 | * - column 1, 3: we've the parent -> done | |
183 | * - column 2: we take the first key from the block to find the parent | |
184 | * (see __add_missing_keys) | |
185 | * - column 4: we use the key to find the parent | |
186 | * | |
187 | * additional information that's available but not required to find the parent | |
188 | * block might help in merging entries to gain some speed. | |
189 | */ | |
190 | ||
8da6d581 | 191 | static int __add_prelim_ref(struct list_head *head, u64 root_id, |
d5c88b73 | 192 | struct btrfs_key *key, int level, |
742916b8 WS |
193 | u64 parent, u64 wanted_disk_byte, int count, |
194 | gfp_t gfp_mask) | |
8da6d581 JS |
195 | { |
196 | struct __prelim_ref *ref; | |
197 | ||
48ec4736 LB |
198 | if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) |
199 | return 0; | |
200 | ||
b9e9a6cb | 201 | ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); |
8da6d581 JS |
202 | if (!ref) |
203 | return -ENOMEM; | |
204 | ||
205 | ref->root_id = root_id; | |
206 | if (key) | |
d5c88b73 | 207 | ref->key_for_search = *key; |
8da6d581 | 208 | else |
d5c88b73 | 209 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 210 | |
3301958b | 211 | ref->inode_list = NULL; |
8da6d581 JS |
212 | ref->level = level; |
213 | ref->count = count; | |
214 | ref->parent = parent; | |
215 | ref->wanted_disk_byte = wanted_disk_byte; | |
216 | list_add_tail(&ref->list, head); | |
217 | ||
218 | return 0; | |
219 | } | |
220 | ||
221 | static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, | |
7ef81ac8 | 222 | struct ulist *parents, struct __prelim_ref *ref, |
44853868 JB |
223 | int level, u64 time_seq, const u64 *extent_item_pos, |
224 | u64 total_refs) | |
8da6d581 | 225 | { |
69bca40d AB |
226 | int ret = 0; |
227 | int slot; | |
228 | struct extent_buffer *eb; | |
229 | struct btrfs_key key; | |
7ef81ac8 | 230 | struct btrfs_key *key_for_search = &ref->key_for_search; |
8da6d581 | 231 | struct btrfs_file_extent_item *fi; |
ed8c4913 | 232 | struct extent_inode_elem *eie = NULL, *old = NULL; |
8da6d581 | 233 | u64 disk_byte; |
7ef81ac8 JB |
234 | u64 wanted_disk_byte = ref->wanted_disk_byte; |
235 | u64 count = 0; | |
8da6d581 | 236 | |
69bca40d AB |
237 | if (level != 0) { |
238 | eb = path->nodes[level]; | |
239 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
240 | if (ret < 0) |
241 | return ret; | |
8da6d581 | 242 | return 0; |
69bca40d | 243 | } |
8da6d581 JS |
244 | |
245 | /* | |
69bca40d AB |
246 | * We normally enter this function with the path already pointing to |
247 | * the first item to check. But sometimes, we may enter it with | |
248 | * slot==nritems. In that case, go to the next leaf before we continue. | |
8da6d581 | 249 | */ |
69bca40d | 250 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) |
3d7806ec | 251 | ret = btrfs_next_old_leaf(root, path, time_seq); |
8da6d581 | 252 | |
44853868 | 253 | while (!ret && count < total_refs) { |
8da6d581 | 254 | eb = path->nodes[0]; |
69bca40d AB |
255 | slot = path->slots[0]; |
256 | ||
257 | btrfs_item_key_to_cpu(eb, &key, slot); | |
258 | ||
259 | if (key.objectid != key_for_search->objectid || | |
260 | key.type != BTRFS_EXTENT_DATA_KEY) | |
261 | break; | |
262 | ||
263 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
264 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
265 | ||
266 | if (disk_byte == wanted_disk_byte) { | |
267 | eie = NULL; | |
ed8c4913 | 268 | old = NULL; |
7ef81ac8 | 269 | count++; |
69bca40d AB |
270 | if (extent_item_pos) { |
271 | ret = check_extent_in_eb(&key, eb, fi, | |
272 | *extent_item_pos, | |
273 | &eie); | |
274 | if (ret < 0) | |
275 | break; | |
276 | } | |
ed8c4913 JB |
277 | if (ret > 0) |
278 | goto next; | |
279 | ret = ulist_add_merge(parents, eb->start, | |
280 | (uintptr_t)eie, | |
281 | (u64 *)&old, GFP_NOFS); | |
282 | if (ret < 0) | |
283 | break; | |
284 | if (!ret && extent_item_pos) { | |
285 | while (old->next) | |
286 | old = old->next; | |
287 | old->next = eie; | |
69bca40d | 288 | } |
f05c4746 | 289 | eie = NULL; |
8da6d581 | 290 | } |
ed8c4913 | 291 | next: |
69bca40d | 292 | ret = btrfs_next_old_item(root, path, time_seq); |
8da6d581 JS |
293 | } |
294 | ||
69bca40d AB |
295 | if (ret > 0) |
296 | ret = 0; | |
f05c4746 WS |
297 | else if (ret < 0) |
298 | free_inode_elem_list(eie); | |
69bca40d | 299 | return ret; |
8da6d581 JS |
300 | } |
301 | ||
302 | /* | |
303 | * resolve an indirect backref in the form (root_id, key, level) | |
304 | * to a logical address | |
305 | */ | |
306 | static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info, | |
da61d31a JB |
307 | struct btrfs_path *path, u64 time_seq, |
308 | struct __prelim_ref *ref, | |
309 | struct ulist *parents, | |
44853868 | 310 | const u64 *extent_item_pos, u64 total_refs) |
8da6d581 | 311 | { |
8da6d581 JS |
312 | struct btrfs_root *root; |
313 | struct btrfs_key root_key; | |
8da6d581 JS |
314 | struct extent_buffer *eb; |
315 | int ret = 0; | |
316 | int root_level; | |
317 | int level = ref->level; | |
538f72cd | 318 | int index; |
8da6d581 | 319 | |
8da6d581 JS |
320 | root_key.objectid = ref->root_id; |
321 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
322 | root_key.offset = (u64)-1; | |
538f72cd WS |
323 | |
324 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
325 | ||
8da6d581 JS |
326 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); |
327 | if (IS_ERR(root)) { | |
538f72cd | 328 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
8da6d581 JS |
329 | ret = PTR_ERR(root); |
330 | goto out; | |
331 | } | |
332 | ||
9e351cc8 JB |
333 | if (path->search_commit_root) |
334 | root_level = btrfs_header_level(root->commit_root); | |
335 | else | |
336 | root_level = btrfs_old_root_level(root, time_seq); | |
8da6d581 | 337 | |
538f72cd WS |
338 | if (root_level + 1 == level) { |
339 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
8da6d581 | 340 | goto out; |
538f72cd | 341 | } |
8da6d581 JS |
342 | |
343 | path->lowest_level = level; | |
8445f61c | 344 | ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq); |
538f72cd WS |
345 | |
346 | /* root node has been locked, we can release @subvol_srcu safely here */ | |
347 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
348 | ||
8da6d581 JS |
349 | pr_debug("search slot in root %llu (level %d, ref count %d) returned " |
350 | "%d for key (%llu %u %llu)\n", | |
c1c9ff7c GU |
351 | ref->root_id, level, ref->count, ret, |
352 | ref->key_for_search.objectid, ref->key_for_search.type, | |
353 | ref->key_for_search.offset); | |
8da6d581 JS |
354 | if (ret < 0) |
355 | goto out; | |
356 | ||
357 | eb = path->nodes[level]; | |
9345457f | 358 | while (!eb) { |
fae7f21c | 359 | if (WARN_ON(!level)) { |
9345457f JS |
360 | ret = 1; |
361 | goto out; | |
362 | } | |
363 | level--; | |
364 | eb = path->nodes[level]; | |
8da6d581 JS |
365 | } |
366 | ||
7ef81ac8 | 367 | ret = add_all_parents(root, path, parents, ref, level, time_seq, |
44853868 | 368 | extent_item_pos, total_refs); |
8da6d581 | 369 | out: |
da61d31a JB |
370 | path->lowest_level = 0; |
371 | btrfs_release_path(path); | |
8da6d581 JS |
372 | return ret; |
373 | } | |
374 | ||
375 | /* | |
376 | * resolve all indirect backrefs from the list | |
377 | */ | |
378 | static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info, | |
da61d31a | 379 | struct btrfs_path *path, u64 time_seq, |
976b1908 | 380 | struct list_head *head, |
44853868 | 381 | const u64 *extent_item_pos, u64 total_refs) |
8da6d581 JS |
382 | { |
383 | int err; | |
384 | int ret = 0; | |
385 | struct __prelim_ref *ref; | |
386 | struct __prelim_ref *ref_safe; | |
387 | struct __prelim_ref *new_ref; | |
388 | struct ulist *parents; | |
389 | struct ulist_node *node; | |
cd1b413c | 390 | struct ulist_iterator uiter; |
8da6d581 JS |
391 | |
392 | parents = ulist_alloc(GFP_NOFS); | |
393 | if (!parents) | |
394 | return -ENOMEM; | |
395 | ||
396 | /* | |
397 | * _safe allows us to insert directly after the current item without | |
398 | * iterating over the newly inserted items. | |
399 | * we're also allowed to re-assign ref during iteration. | |
400 | */ | |
401 | list_for_each_entry_safe(ref, ref_safe, head, list) { | |
402 | if (ref->parent) /* already direct */ | |
403 | continue; | |
404 | if (ref->count == 0) | |
405 | continue; | |
da61d31a | 406 | err = __resolve_indirect_ref(fs_info, path, time_seq, ref, |
44853868 JB |
407 | parents, extent_item_pos, |
408 | total_refs); | |
95def2ed WS |
409 | /* |
410 | * we can only tolerate ENOENT,otherwise,we should catch error | |
411 | * and return directly. | |
412 | */ | |
413 | if (err == -ENOENT) { | |
8da6d581 | 414 | continue; |
95def2ed WS |
415 | } else if (err) { |
416 | ret = err; | |
417 | goto out; | |
418 | } | |
8da6d581 JS |
419 | |
420 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
421 | ULIST_ITER_INIT(&uiter); |
422 | node = ulist_next(parents, &uiter); | |
8da6d581 | 423 | ref->parent = node ? node->val : 0; |
995e01b7 | 424 | ref->inode_list = node ? |
35a3621b | 425 | (struct extent_inode_elem *)(uintptr_t)node->aux : NULL; |
8da6d581 JS |
426 | |
427 | /* additional parents require new refs being added here */ | |
cd1b413c | 428 | while ((node = ulist_next(parents, &uiter))) { |
b9e9a6cb WS |
429 | new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, |
430 | GFP_NOFS); | |
8da6d581 JS |
431 | if (!new_ref) { |
432 | ret = -ENOMEM; | |
e36902d4 | 433 | goto out; |
8da6d581 JS |
434 | } |
435 | memcpy(new_ref, ref, sizeof(*ref)); | |
436 | new_ref->parent = node->val; | |
995e01b7 JS |
437 | new_ref->inode_list = (struct extent_inode_elem *) |
438 | (uintptr_t)node->aux; | |
8da6d581 JS |
439 | list_add(&new_ref->list, &ref->list); |
440 | } | |
441 | ulist_reinit(parents); | |
442 | } | |
e36902d4 | 443 | out: |
8da6d581 JS |
444 | ulist_free(parents); |
445 | return ret; | |
446 | } | |
447 | ||
d5c88b73 JS |
448 | static inline int ref_for_same_block(struct __prelim_ref *ref1, |
449 | struct __prelim_ref *ref2) | |
450 | { | |
451 | if (ref1->level != ref2->level) | |
452 | return 0; | |
453 | if (ref1->root_id != ref2->root_id) | |
454 | return 0; | |
455 | if (ref1->key_for_search.type != ref2->key_for_search.type) | |
456 | return 0; | |
457 | if (ref1->key_for_search.objectid != ref2->key_for_search.objectid) | |
458 | return 0; | |
459 | if (ref1->key_for_search.offset != ref2->key_for_search.offset) | |
460 | return 0; | |
461 | if (ref1->parent != ref2->parent) | |
462 | return 0; | |
463 | ||
464 | return 1; | |
465 | } | |
466 | ||
467 | /* | |
468 | * read tree blocks and add keys where required. | |
469 | */ | |
470 | static int __add_missing_keys(struct btrfs_fs_info *fs_info, | |
471 | struct list_head *head) | |
472 | { | |
473 | struct list_head *pos; | |
474 | struct extent_buffer *eb; | |
475 | ||
476 | list_for_each(pos, head) { | |
477 | struct __prelim_ref *ref; | |
478 | ref = list_entry(pos, struct __prelim_ref, list); | |
479 | ||
480 | if (ref->parent) | |
481 | continue; | |
482 | if (ref->key_for_search.type) | |
483 | continue; | |
484 | BUG_ON(!ref->wanted_disk_byte); | |
485 | eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte, | |
486 | fs_info->tree_root->leafsize, 0); | |
416bc658 JB |
487 | if (!eb || !extent_buffer_uptodate(eb)) { |
488 | free_extent_buffer(eb); | |
489 | return -EIO; | |
490 | } | |
d5c88b73 JS |
491 | btrfs_tree_read_lock(eb); |
492 | if (btrfs_header_level(eb) == 0) | |
493 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
494 | else | |
495 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
496 | btrfs_tree_read_unlock(eb); | |
497 | free_extent_buffer(eb); | |
498 | } | |
499 | return 0; | |
500 | } | |
501 | ||
8da6d581 JS |
502 | /* |
503 | * merge two lists of backrefs and adjust counts accordingly | |
504 | * | |
505 | * mode = 1: merge identical keys, if key is set | |
d5c88b73 JS |
506 | * FIXME: if we add more keys in __add_prelim_ref, we can merge more here. |
507 | * additionally, we could even add a key range for the blocks we | |
508 | * looked into to merge even more (-> replace unresolved refs by those | |
509 | * having a parent). | |
8da6d581 JS |
510 | * mode = 2: merge identical parents |
511 | */ | |
692206b1 | 512 | static void __merge_refs(struct list_head *head, int mode) |
8da6d581 JS |
513 | { |
514 | struct list_head *pos1; | |
515 | ||
516 | list_for_each(pos1, head) { | |
517 | struct list_head *n2; | |
518 | struct list_head *pos2; | |
519 | struct __prelim_ref *ref1; | |
520 | ||
521 | ref1 = list_entry(pos1, struct __prelim_ref, list); | |
522 | ||
8da6d581 JS |
523 | for (pos2 = pos1->next, n2 = pos2->next; pos2 != head; |
524 | pos2 = n2, n2 = pos2->next) { | |
525 | struct __prelim_ref *ref2; | |
d5c88b73 | 526 | struct __prelim_ref *xchg; |
3ef5969c | 527 | struct extent_inode_elem *eie; |
8da6d581 JS |
528 | |
529 | ref2 = list_entry(pos2, struct __prelim_ref, list); | |
530 | ||
531 | if (mode == 1) { | |
d5c88b73 | 532 | if (!ref_for_same_block(ref1, ref2)) |
8da6d581 | 533 | continue; |
d5c88b73 JS |
534 | if (!ref1->parent && ref2->parent) { |
535 | xchg = ref1; | |
536 | ref1 = ref2; | |
537 | ref2 = xchg; | |
538 | } | |
8da6d581 JS |
539 | } else { |
540 | if (ref1->parent != ref2->parent) | |
541 | continue; | |
8da6d581 | 542 | } |
3ef5969c AB |
543 | |
544 | eie = ref1->inode_list; | |
545 | while (eie && eie->next) | |
546 | eie = eie->next; | |
547 | if (eie) | |
548 | eie->next = ref2->inode_list; | |
549 | else | |
550 | ref1->inode_list = ref2->inode_list; | |
551 | ref1->count += ref2->count; | |
552 | ||
8da6d581 | 553 | list_del(&ref2->list); |
b9e9a6cb | 554 | kmem_cache_free(btrfs_prelim_ref_cache, ref2); |
8da6d581 JS |
555 | } |
556 | ||
557 | } | |
8da6d581 JS |
558 | } |
559 | ||
560 | /* | |
561 | * add all currently queued delayed refs from this head whose seq nr is | |
562 | * smaller or equal that seq to the list | |
563 | */ | |
564 | static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq, | |
44853868 | 565 | struct list_head *prefs, u64 *total_refs) |
8da6d581 JS |
566 | { |
567 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
568 | struct rb_node *n = &head->node.rb_node; | |
d5c88b73 JS |
569 | struct btrfs_key key; |
570 | struct btrfs_key op_key = {0}; | |
8da6d581 | 571 | int sgn; |
b1375d64 | 572 | int ret = 0; |
8da6d581 JS |
573 | |
574 | if (extent_op && extent_op->update_key) | |
d5c88b73 | 575 | btrfs_disk_key_to_cpu(&op_key, &extent_op->key); |
8da6d581 | 576 | |
d7df2c79 JB |
577 | spin_lock(&head->lock); |
578 | n = rb_first(&head->ref_root); | |
579 | while (n) { | |
8da6d581 JS |
580 | struct btrfs_delayed_ref_node *node; |
581 | node = rb_entry(n, struct btrfs_delayed_ref_node, | |
582 | rb_node); | |
d7df2c79 | 583 | n = rb_next(n); |
8da6d581 JS |
584 | if (node->seq > seq) |
585 | continue; | |
586 | ||
587 | switch (node->action) { | |
588 | case BTRFS_ADD_DELAYED_EXTENT: | |
589 | case BTRFS_UPDATE_DELAYED_HEAD: | |
590 | WARN_ON(1); | |
591 | continue; | |
592 | case BTRFS_ADD_DELAYED_REF: | |
593 | sgn = 1; | |
594 | break; | |
595 | case BTRFS_DROP_DELAYED_REF: | |
596 | sgn = -1; | |
597 | break; | |
598 | default: | |
599 | BUG_ON(1); | |
600 | } | |
44853868 | 601 | *total_refs += (node->ref_mod * sgn); |
8da6d581 JS |
602 | switch (node->type) { |
603 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
604 | struct btrfs_delayed_tree_ref *ref; | |
605 | ||
606 | ref = btrfs_delayed_node_to_tree_ref(node); | |
d5c88b73 | 607 | ret = __add_prelim_ref(prefs, ref->root, &op_key, |
8da6d581 | 608 | ref->level + 1, 0, node->bytenr, |
742916b8 | 609 | node->ref_mod * sgn, GFP_ATOMIC); |
8da6d581 JS |
610 | break; |
611 | } | |
612 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
613 | struct btrfs_delayed_tree_ref *ref; | |
614 | ||
615 | ref = btrfs_delayed_node_to_tree_ref(node); | |
d5c88b73 | 616 | ret = __add_prelim_ref(prefs, ref->root, NULL, |
8da6d581 JS |
617 | ref->level + 1, ref->parent, |
618 | node->bytenr, | |
742916b8 | 619 | node->ref_mod * sgn, GFP_ATOMIC); |
8da6d581 JS |
620 | break; |
621 | } | |
622 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
623 | struct btrfs_delayed_data_ref *ref; | |
8da6d581 JS |
624 | ref = btrfs_delayed_node_to_data_ref(node); |
625 | ||
626 | key.objectid = ref->objectid; | |
627 | key.type = BTRFS_EXTENT_DATA_KEY; | |
628 | key.offset = ref->offset; | |
629 | ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0, | |
630 | node->bytenr, | |
742916b8 | 631 | node->ref_mod * sgn, GFP_ATOMIC); |
8da6d581 JS |
632 | break; |
633 | } | |
634 | case BTRFS_SHARED_DATA_REF_KEY: { | |
635 | struct btrfs_delayed_data_ref *ref; | |
8da6d581 JS |
636 | |
637 | ref = btrfs_delayed_node_to_data_ref(node); | |
638 | ||
639 | key.objectid = ref->objectid; | |
640 | key.type = BTRFS_EXTENT_DATA_KEY; | |
641 | key.offset = ref->offset; | |
642 | ret = __add_prelim_ref(prefs, ref->root, &key, 0, | |
643 | ref->parent, node->bytenr, | |
742916b8 | 644 | node->ref_mod * sgn, GFP_ATOMIC); |
8da6d581 JS |
645 | break; |
646 | } | |
647 | default: | |
648 | WARN_ON(1); | |
649 | } | |
1149ab6b | 650 | if (ret) |
d7df2c79 | 651 | break; |
8da6d581 | 652 | } |
d7df2c79 JB |
653 | spin_unlock(&head->lock); |
654 | return ret; | |
8da6d581 JS |
655 | } |
656 | ||
657 | /* | |
658 | * add all inline backrefs for bytenr to the list | |
659 | */ | |
660 | static int __add_inline_refs(struct btrfs_fs_info *fs_info, | |
661 | struct btrfs_path *path, u64 bytenr, | |
44853868 JB |
662 | int *info_level, struct list_head *prefs, |
663 | u64 *total_refs) | |
8da6d581 | 664 | { |
b1375d64 | 665 | int ret = 0; |
8da6d581 JS |
666 | int slot; |
667 | struct extent_buffer *leaf; | |
668 | struct btrfs_key key; | |
261c84b6 | 669 | struct btrfs_key found_key; |
8da6d581 JS |
670 | unsigned long ptr; |
671 | unsigned long end; | |
672 | struct btrfs_extent_item *ei; | |
673 | u64 flags; | |
674 | u64 item_size; | |
675 | ||
676 | /* | |
677 | * enumerate all inline refs | |
678 | */ | |
679 | leaf = path->nodes[0]; | |
dadcaf78 | 680 | slot = path->slots[0]; |
8da6d581 JS |
681 | |
682 | item_size = btrfs_item_size_nr(leaf, slot); | |
683 | BUG_ON(item_size < sizeof(*ei)); | |
684 | ||
685 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
686 | flags = btrfs_extent_flags(leaf, ei); | |
44853868 | 687 | *total_refs += btrfs_extent_refs(leaf, ei); |
261c84b6 | 688 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
8da6d581 JS |
689 | |
690 | ptr = (unsigned long)(ei + 1); | |
691 | end = (unsigned long)ei + item_size; | |
692 | ||
261c84b6 JB |
693 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && |
694 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
8da6d581 | 695 | struct btrfs_tree_block_info *info; |
8da6d581 JS |
696 | |
697 | info = (struct btrfs_tree_block_info *)ptr; | |
698 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
699 | ptr += sizeof(struct btrfs_tree_block_info); |
700 | BUG_ON(ptr > end); | |
261c84b6 JB |
701 | } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { |
702 | *info_level = found_key.offset; | |
8da6d581 JS |
703 | } else { |
704 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
705 | } | |
706 | ||
707 | while (ptr < end) { | |
708 | struct btrfs_extent_inline_ref *iref; | |
709 | u64 offset; | |
710 | int type; | |
711 | ||
712 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
713 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
714 | offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
715 | ||
716 | switch (type) { | |
717 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
d5c88b73 | 718 | ret = __add_prelim_ref(prefs, 0, NULL, |
8da6d581 | 719 | *info_level + 1, offset, |
742916b8 | 720 | bytenr, 1, GFP_NOFS); |
8da6d581 JS |
721 | break; |
722 | case BTRFS_SHARED_DATA_REF_KEY: { | |
723 | struct btrfs_shared_data_ref *sdref; | |
724 | int count; | |
725 | ||
726 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
727 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
728 | ret = __add_prelim_ref(prefs, 0, NULL, 0, offset, | |
742916b8 | 729 | bytenr, count, GFP_NOFS); |
8da6d581 JS |
730 | break; |
731 | } | |
732 | case BTRFS_TREE_BLOCK_REF_KEY: | |
d5c88b73 JS |
733 | ret = __add_prelim_ref(prefs, offset, NULL, |
734 | *info_level + 1, 0, | |
742916b8 | 735 | bytenr, 1, GFP_NOFS); |
8da6d581 JS |
736 | break; |
737 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
738 | struct btrfs_extent_data_ref *dref; | |
739 | int count; | |
740 | u64 root; | |
741 | ||
742 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
743 | count = btrfs_extent_data_ref_count(leaf, dref); | |
744 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
745 | dref); | |
746 | key.type = BTRFS_EXTENT_DATA_KEY; | |
747 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
748 | root = btrfs_extent_data_ref_root(leaf, dref); | |
d5c88b73 | 749 | ret = __add_prelim_ref(prefs, root, &key, 0, 0, |
742916b8 | 750 | bytenr, count, GFP_NOFS); |
8da6d581 JS |
751 | break; |
752 | } | |
753 | default: | |
754 | WARN_ON(1); | |
755 | } | |
1149ab6b WS |
756 | if (ret) |
757 | return ret; | |
8da6d581 JS |
758 | ptr += btrfs_extent_inline_ref_size(type); |
759 | } | |
760 | ||
761 | return 0; | |
762 | } | |
763 | ||
764 | /* | |
765 | * add all non-inline backrefs for bytenr to the list | |
766 | */ | |
767 | static int __add_keyed_refs(struct btrfs_fs_info *fs_info, | |
768 | struct btrfs_path *path, u64 bytenr, | |
d5c88b73 | 769 | int info_level, struct list_head *prefs) |
8da6d581 JS |
770 | { |
771 | struct btrfs_root *extent_root = fs_info->extent_root; | |
772 | int ret; | |
773 | int slot; | |
774 | struct extent_buffer *leaf; | |
775 | struct btrfs_key key; | |
776 | ||
777 | while (1) { | |
778 | ret = btrfs_next_item(extent_root, path); | |
779 | if (ret < 0) | |
780 | break; | |
781 | if (ret) { | |
782 | ret = 0; | |
783 | break; | |
784 | } | |
785 | ||
786 | slot = path->slots[0]; | |
787 | leaf = path->nodes[0]; | |
788 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
789 | ||
790 | if (key.objectid != bytenr) | |
791 | break; | |
792 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
793 | continue; | |
794 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
795 | break; | |
796 | ||
797 | switch (key.type) { | |
798 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
d5c88b73 | 799 | ret = __add_prelim_ref(prefs, 0, NULL, |
8da6d581 | 800 | info_level + 1, key.offset, |
742916b8 | 801 | bytenr, 1, GFP_NOFS); |
8da6d581 JS |
802 | break; |
803 | case BTRFS_SHARED_DATA_REF_KEY: { | |
804 | struct btrfs_shared_data_ref *sdref; | |
805 | int count; | |
806 | ||
807 | sdref = btrfs_item_ptr(leaf, slot, | |
808 | struct btrfs_shared_data_ref); | |
809 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
810 | ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset, | |
742916b8 | 811 | bytenr, count, GFP_NOFS); |
8da6d581 JS |
812 | break; |
813 | } | |
814 | case BTRFS_TREE_BLOCK_REF_KEY: | |
d5c88b73 JS |
815 | ret = __add_prelim_ref(prefs, key.offset, NULL, |
816 | info_level + 1, 0, | |
742916b8 | 817 | bytenr, 1, GFP_NOFS); |
8da6d581 JS |
818 | break; |
819 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
820 | struct btrfs_extent_data_ref *dref; | |
821 | int count; | |
822 | u64 root; | |
823 | ||
824 | dref = btrfs_item_ptr(leaf, slot, | |
825 | struct btrfs_extent_data_ref); | |
826 | count = btrfs_extent_data_ref_count(leaf, dref); | |
827 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
828 | dref); | |
829 | key.type = BTRFS_EXTENT_DATA_KEY; | |
830 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
831 | root = btrfs_extent_data_ref_root(leaf, dref); | |
832 | ret = __add_prelim_ref(prefs, root, &key, 0, 0, | |
742916b8 | 833 | bytenr, count, GFP_NOFS); |
8da6d581 JS |
834 | break; |
835 | } | |
836 | default: | |
837 | WARN_ON(1); | |
838 | } | |
1149ab6b WS |
839 | if (ret) |
840 | return ret; | |
841 | ||
8da6d581 JS |
842 | } |
843 | ||
844 | return ret; | |
845 | } | |
846 | ||
847 | /* | |
848 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
849 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
850 | * indirect refs to their parent bytenr. | |
851 | * When roots are found, they're added to the roots list | |
852 | * | |
853 | * FIXME some caching might speed things up | |
854 | */ | |
855 | static int find_parent_nodes(struct btrfs_trans_handle *trans, | |
856 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c JS |
857 | u64 time_seq, struct ulist *refs, |
858 | struct ulist *roots, const u64 *extent_item_pos) | |
8da6d581 JS |
859 | { |
860 | struct btrfs_key key; | |
861 | struct btrfs_path *path; | |
8da6d581 | 862 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 863 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
864 | int info_level = 0; |
865 | int ret; | |
866 | struct list_head prefs_delayed; | |
867 | struct list_head prefs; | |
868 | struct __prelim_ref *ref; | |
f05c4746 | 869 | struct extent_inode_elem *eie = NULL; |
44853868 | 870 | u64 total_refs = 0; |
8da6d581 JS |
871 | |
872 | INIT_LIST_HEAD(&prefs); | |
873 | INIT_LIST_HEAD(&prefs_delayed); | |
874 | ||
875 | key.objectid = bytenr; | |
8da6d581 | 876 | key.offset = (u64)-1; |
261c84b6 JB |
877 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
878 | key.type = BTRFS_METADATA_ITEM_KEY; | |
879 | else | |
880 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
8da6d581 JS |
881 | |
882 | path = btrfs_alloc_path(); | |
883 | if (!path) | |
884 | return -ENOMEM; | |
e84752d4 | 885 | if (!trans) { |
da61d31a | 886 | path->search_commit_root = 1; |
e84752d4 WS |
887 | path->skip_locking = 1; |
888 | } | |
8da6d581 JS |
889 | |
890 | /* | |
891 | * grab both a lock on the path and a lock on the delayed ref head. | |
892 | * We need both to get a consistent picture of how the refs look | |
893 | * at a specified point in time | |
894 | */ | |
895 | again: | |
d3b01064 LZ |
896 | head = NULL; |
897 | ||
8da6d581 JS |
898 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
899 | if (ret < 0) | |
900 | goto out; | |
901 | BUG_ON(ret == 0); | |
902 | ||
faa2dbf0 JB |
903 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
904 | if (trans && likely(trans->type != __TRANS_DUMMY)) { | |
905 | #else | |
da61d31a | 906 | if (trans) { |
faa2dbf0 | 907 | #endif |
7a3ae2f8 JS |
908 | /* |
909 | * look if there are updates for this ref queued and lock the | |
910 | * head | |
911 | */ | |
912 | delayed_refs = &trans->transaction->delayed_refs; | |
913 | spin_lock(&delayed_refs->lock); | |
914 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
915 | if (head) { | |
916 | if (!mutex_trylock(&head->mutex)) { | |
917 | atomic_inc(&head->node.refs); | |
918 | spin_unlock(&delayed_refs->lock); | |
919 | ||
920 | btrfs_release_path(path); | |
921 | ||
922 | /* | |
923 | * Mutex was contended, block until it's | |
924 | * released and try again | |
925 | */ | |
926 | mutex_lock(&head->mutex); | |
927 | mutex_unlock(&head->mutex); | |
928 | btrfs_put_delayed_ref(&head->node); | |
929 | goto again; | |
930 | } | |
d7df2c79 | 931 | spin_unlock(&delayed_refs->lock); |
097b8a7c | 932 | ret = __add_delayed_refs(head, time_seq, |
44853868 | 933 | &prefs_delayed, &total_refs); |
155725c9 | 934 | mutex_unlock(&head->mutex); |
d7df2c79 | 935 | if (ret) |
7a3ae2f8 | 936 | goto out; |
d7df2c79 JB |
937 | } else { |
938 | spin_unlock(&delayed_refs->lock); | |
d3b01064 | 939 | } |
8da6d581 | 940 | } |
8da6d581 JS |
941 | |
942 | if (path->slots[0]) { | |
943 | struct extent_buffer *leaf; | |
944 | int slot; | |
945 | ||
dadcaf78 | 946 | path->slots[0]--; |
8da6d581 | 947 | leaf = path->nodes[0]; |
dadcaf78 | 948 | slot = path->slots[0]; |
8da6d581 JS |
949 | btrfs_item_key_to_cpu(leaf, &key, slot); |
950 | if (key.objectid == bytenr && | |
261c84b6 JB |
951 | (key.type == BTRFS_EXTENT_ITEM_KEY || |
952 | key.type == BTRFS_METADATA_ITEM_KEY)) { | |
8da6d581 | 953 | ret = __add_inline_refs(fs_info, path, bytenr, |
44853868 JB |
954 | &info_level, &prefs, |
955 | &total_refs); | |
8da6d581 JS |
956 | if (ret) |
957 | goto out; | |
d5c88b73 | 958 | ret = __add_keyed_refs(fs_info, path, bytenr, |
8da6d581 JS |
959 | info_level, &prefs); |
960 | if (ret) | |
961 | goto out; | |
962 | } | |
963 | } | |
964 | btrfs_release_path(path); | |
965 | ||
8da6d581 JS |
966 | list_splice_init(&prefs_delayed, &prefs); |
967 | ||
d5c88b73 JS |
968 | ret = __add_missing_keys(fs_info, &prefs); |
969 | if (ret) | |
970 | goto out; | |
971 | ||
692206b1 | 972 | __merge_refs(&prefs, 1); |
8da6d581 | 973 | |
da61d31a | 974 | ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs, |
44853868 | 975 | extent_item_pos, total_refs); |
8da6d581 JS |
976 | if (ret) |
977 | goto out; | |
978 | ||
692206b1 | 979 | __merge_refs(&prefs, 2); |
8da6d581 JS |
980 | |
981 | while (!list_empty(&prefs)) { | |
982 | ref = list_first_entry(&prefs, struct __prelim_ref, list); | |
6c1500f2 | 983 | WARN_ON(ref->count < 0); |
98cfee21 | 984 | if (roots && ref->count && ref->root_id && ref->parent == 0) { |
8da6d581 JS |
985 | /* no parent == root of tree */ |
986 | ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); | |
f1723939 WS |
987 | if (ret < 0) |
988 | goto out; | |
8da6d581 JS |
989 | } |
990 | if (ref->count && ref->parent) { | |
8a56457f JB |
991 | if (extent_item_pos && !ref->inode_list && |
992 | ref->level == 0) { | |
976b1908 JS |
993 | u32 bsz; |
994 | struct extent_buffer *eb; | |
995 | bsz = btrfs_level_size(fs_info->extent_root, | |
8a56457f | 996 | ref->level); |
976b1908 JS |
997 | eb = read_tree_block(fs_info->extent_root, |
998 | ref->parent, bsz, 0); | |
416bc658 JB |
999 | if (!eb || !extent_buffer_uptodate(eb)) { |
1000 | free_extent_buffer(eb); | |
c16c2e2e WS |
1001 | ret = -EIO; |
1002 | goto out; | |
416bc658 | 1003 | } |
6f7ff6d7 FM |
1004 | btrfs_tree_read_lock(eb); |
1005 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
976b1908 JS |
1006 | ret = find_extent_in_eb(eb, bytenr, |
1007 | *extent_item_pos, &eie); | |
6f7ff6d7 | 1008 | btrfs_tree_read_unlock_blocking(eb); |
976b1908 | 1009 | free_extent_buffer(eb); |
f5929cd8 FDBM |
1010 | if (ret < 0) |
1011 | goto out; | |
1012 | ref->inode_list = eie; | |
976b1908 | 1013 | } |
3301958b | 1014 | ret = ulist_add_merge(refs, ref->parent, |
995e01b7 | 1015 | (uintptr_t)ref->inode_list, |
34d73f54 | 1016 | (u64 *)&eie, GFP_NOFS); |
f1723939 WS |
1017 | if (ret < 0) |
1018 | goto out; | |
3301958b JS |
1019 | if (!ret && extent_item_pos) { |
1020 | /* | |
1021 | * we've recorded that parent, so we must extend | |
1022 | * its inode list here | |
1023 | */ | |
1024 | BUG_ON(!eie); | |
1025 | while (eie->next) | |
1026 | eie = eie->next; | |
1027 | eie->next = ref->inode_list; | |
1028 | } | |
f05c4746 | 1029 | eie = NULL; |
8da6d581 | 1030 | } |
a4fdb61e | 1031 | list_del(&ref->list); |
b9e9a6cb | 1032 | kmem_cache_free(btrfs_prelim_ref_cache, ref); |
8da6d581 JS |
1033 | } |
1034 | ||
1035 | out: | |
8da6d581 JS |
1036 | btrfs_free_path(path); |
1037 | while (!list_empty(&prefs)) { | |
1038 | ref = list_first_entry(&prefs, struct __prelim_ref, list); | |
1039 | list_del(&ref->list); | |
b9e9a6cb | 1040 | kmem_cache_free(btrfs_prelim_ref_cache, ref); |
8da6d581 JS |
1041 | } |
1042 | while (!list_empty(&prefs_delayed)) { | |
1043 | ref = list_first_entry(&prefs_delayed, struct __prelim_ref, | |
1044 | list); | |
1045 | list_del(&ref->list); | |
b9e9a6cb | 1046 | kmem_cache_free(btrfs_prelim_ref_cache, ref); |
8da6d581 | 1047 | } |
f05c4746 WS |
1048 | if (ret < 0) |
1049 | free_inode_elem_list(eie); | |
8da6d581 JS |
1050 | return ret; |
1051 | } | |
1052 | ||
976b1908 JS |
1053 | static void free_leaf_list(struct ulist *blocks) |
1054 | { | |
1055 | struct ulist_node *node = NULL; | |
1056 | struct extent_inode_elem *eie; | |
976b1908 JS |
1057 | struct ulist_iterator uiter; |
1058 | ||
1059 | ULIST_ITER_INIT(&uiter); | |
1060 | while ((node = ulist_next(blocks, &uiter))) { | |
1061 | if (!node->aux) | |
1062 | continue; | |
995e01b7 | 1063 | eie = (struct extent_inode_elem *)(uintptr_t)node->aux; |
f05c4746 | 1064 | free_inode_elem_list(eie); |
976b1908 JS |
1065 | node->aux = 0; |
1066 | } | |
1067 | ||
1068 | ulist_free(blocks); | |
1069 | } | |
1070 | ||
8da6d581 JS |
1071 | /* |
1072 | * Finds all leafs with a reference to the specified combination of bytenr and | |
1073 | * offset. key_list_head will point to a list of corresponding keys (caller must | |
1074 | * free each list element). The leafs will be stored in the leafs ulist, which | |
1075 | * must be freed with ulist_free. | |
1076 | * | |
1077 | * returns 0 on success, <0 on error | |
1078 | */ | |
1079 | static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, | |
1080 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1081 | u64 time_seq, struct ulist **leafs, |
976b1908 | 1082 | const u64 *extent_item_pos) |
8da6d581 | 1083 | { |
8da6d581 JS |
1084 | int ret; |
1085 | ||
8da6d581 | 1086 | *leafs = ulist_alloc(GFP_NOFS); |
98cfee21 | 1087 | if (!*leafs) |
8da6d581 | 1088 | return -ENOMEM; |
8da6d581 | 1089 | |
097b8a7c | 1090 | ret = find_parent_nodes(trans, fs_info, bytenr, |
98cfee21 | 1091 | time_seq, *leafs, NULL, extent_item_pos); |
8da6d581 | 1092 | if (ret < 0 && ret != -ENOENT) { |
976b1908 | 1093 | free_leaf_list(*leafs); |
8da6d581 JS |
1094 | return ret; |
1095 | } | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * walk all backrefs for a given extent to find all roots that reference this | |
1102 | * extent. Walking a backref means finding all extents that reference this | |
1103 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1104 | * recursive process, but here it is implemented in an iterative fashion: We | |
1105 | * find all referencing extents for the extent in question and put them on a | |
1106 | * list. In turn, we find all referencing extents for those, further appending | |
1107 | * to the list. The way we iterate the list allows adding more elements after | |
1108 | * the current while iterating. The process stops when we reach the end of the | |
1109 | * list. Found roots are added to the roots list. | |
1110 | * | |
1111 | * returns 0 on success, < 0 on error. | |
1112 | */ | |
9e351cc8 JB |
1113 | static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans, |
1114 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
1115 | u64 time_seq, struct ulist **roots) | |
8da6d581 JS |
1116 | { |
1117 | struct ulist *tmp; | |
1118 | struct ulist_node *node = NULL; | |
cd1b413c | 1119 | struct ulist_iterator uiter; |
8da6d581 JS |
1120 | int ret; |
1121 | ||
1122 | tmp = ulist_alloc(GFP_NOFS); | |
1123 | if (!tmp) | |
1124 | return -ENOMEM; | |
1125 | *roots = ulist_alloc(GFP_NOFS); | |
1126 | if (!*roots) { | |
1127 | ulist_free(tmp); | |
1128 | return -ENOMEM; | |
1129 | } | |
1130 | ||
cd1b413c | 1131 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1132 | while (1) { |
097b8a7c | 1133 | ret = find_parent_nodes(trans, fs_info, bytenr, |
8445f61c | 1134 | time_seq, tmp, *roots, NULL); |
8da6d581 JS |
1135 | if (ret < 0 && ret != -ENOENT) { |
1136 | ulist_free(tmp); | |
1137 | ulist_free(*roots); | |
1138 | return ret; | |
1139 | } | |
cd1b413c | 1140 | node = ulist_next(tmp, &uiter); |
8da6d581 JS |
1141 | if (!node) |
1142 | break; | |
1143 | bytenr = node->val; | |
bca1a290 | 1144 | cond_resched(); |
8da6d581 JS |
1145 | } |
1146 | ||
1147 | ulist_free(tmp); | |
1148 | return 0; | |
1149 | } | |
1150 | ||
9e351cc8 JB |
1151 | int btrfs_find_all_roots(struct btrfs_trans_handle *trans, |
1152 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
1153 | u64 time_seq, struct ulist **roots) | |
1154 | { | |
1155 | int ret; | |
1156 | ||
1157 | if (!trans) | |
1158 | down_read(&fs_info->commit_root_sem); | |
1159 | ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots); | |
1160 | if (!trans) | |
1161 | up_read(&fs_info->commit_root_sem); | |
1162 | return ret; | |
1163 | } | |
1164 | ||
a542ad1b JS |
1165 | /* |
1166 | * this makes the path point to (inum INODE_ITEM ioff) | |
1167 | */ | |
1168 | int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, | |
1169 | struct btrfs_path *path) | |
1170 | { | |
1171 | struct btrfs_key key; | |
e33d5c3d KN |
1172 | return btrfs_find_item(fs_root, path, inum, ioff, |
1173 | BTRFS_INODE_ITEM_KEY, &key); | |
a542ad1b JS |
1174 | } |
1175 | ||
1176 | static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, | |
1177 | struct btrfs_path *path, | |
1178 | struct btrfs_key *found_key) | |
1179 | { | |
e33d5c3d KN |
1180 | return btrfs_find_item(fs_root, path, inum, ioff, |
1181 | BTRFS_INODE_REF_KEY, found_key); | |
a542ad1b JS |
1182 | } |
1183 | ||
f186373f MF |
1184 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
1185 | u64 start_off, struct btrfs_path *path, | |
1186 | struct btrfs_inode_extref **ret_extref, | |
1187 | u64 *found_off) | |
1188 | { | |
1189 | int ret, slot; | |
1190 | struct btrfs_key key; | |
1191 | struct btrfs_key found_key; | |
1192 | struct btrfs_inode_extref *extref; | |
1193 | struct extent_buffer *leaf; | |
1194 | unsigned long ptr; | |
1195 | ||
1196 | key.objectid = inode_objectid; | |
1197 | btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY); | |
1198 | key.offset = start_off; | |
1199 | ||
1200 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1201 | if (ret < 0) | |
1202 | return ret; | |
1203 | ||
1204 | while (1) { | |
1205 | leaf = path->nodes[0]; | |
1206 | slot = path->slots[0]; | |
1207 | if (slot >= btrfs_header_nritems(leaf)) { | |
1208 | /* | |
1209 | * If the item at offset is not found, | |
1210 | * btrfs_search_slot will point us to the slot | |
1211 | * where it should be inserted. In our case | |
1212 | * that will be the slot directly before the | |
1213 | * next INODE_REF_KEY_V2 item. In the case | |
1214 | * that we're pointing to the last slot in a | |
1215 | * leaf, we must move one leaf over. | |
1216 | */ | |
1217 | ret = btrfs_next_leaf(root, path); | |
1218 | if (ret) { | |
1219 | if (ret >= 1) | |
1220 | ret = -ENOENT; | |
1221 | break; | |
1222 | } | |
1223 | continue; | |
1224 | } | |
1225 | ||
1226 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1227 | ||
1228 | /* | |
1229 | * Check that we're still looking at an extended ref key for | |
1230 | * this particular objectid. If we have different | |
1231 | * objectid or type then there are no more to be found | |
1232 | * in the tree and we can exit. | |
1233 | */ | |
1234 | ret = -ENOENT; | |
1235 | if (found_key.objectid != inode_objectid) | |
1236 | break; | |
1237 | if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY) | |
1238 | break; | |
1239 | ||
1240 | ret = 0; | |
1241 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1242 | extref = (struct btrfs_inode_extref *)ptr; | |
1243 | *ret_extref = extref; | |
1244 | if (found_off) | |
1245 | *found_off = found_key.offset; | |
1246 | break; | |
1247 | } | |
1248 | ||
1249 | return ret; | |
1250 | } | |
1251 | ||
48a3b636 ES |
1252 | /* |
1253 | * this iterates to turn a name (from iref/extref) into a full filesystem path. | |
1254 | * Elements 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 | */ | |
96b5bd77 JS |
1266 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
1267 | u32 name_len, unsigned long name_off, | |
1268 | struct extent_buffer *eb_in, u64 parent, | |
1269 | char *dest, u32 size) | |
a542ad1b | 1270 | { |
a542ad1b JS |
1271 | int slot; |
1272 | u64 next_inum; | |
1273 | int ret; | |
661bec6b | 1274 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
1275 | struct extent_buffer *eb = eb_in; |
1276 | struct btrfs_key found_key; | |
b916a59a | 1277 | int leave_spinning = path->leave_spinning; |
d24bec3a | 1278 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
1279 | |
1280 | if (bytes_left >= 0) | |
1281 | dest[bytes_left] = '\0'; | |
1282 | ||
b916a59a | 1283 | path->leave_spinning = 1; |
a542ad1b | 1284 | while (1) { |
d24bec3a | 1285 | bytes_left -= name_len; |
a542ad1b JS |
1286 | if (bytes_left >= 0) |
1287 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 1288 | name_off, name_len); |
b916a59a JS |
1289 | if (eb != eb_in) { |
1290 | btrfs_tree_read_unlock_blocking(eb); | |
a542ad1b | 1291 | free_extent_buffer(eb); |
b916a59a | 1292 | } |
a542ad1b | 1293 | ret = inode_ref_info(parent, 0, fs_root, path, &found_key); |
8f24b496 JS |
1294 | if (ret > 0) |
1295 | ret = -ENOENT; | |
a542ad1b JS |
1296 | if (ret) |
1297 | break; | |
d24bec3a | 1298 | |
a542ad1b JS |
1299 | next_inum = found_key.offset; |
1300 | ||
1301 | /* regular exit ahead */ | |
1302 | if (parent == next_inum) | |
1303 | break; | |
1304 | ||
1305 | slot = path->slots[0]; | |
1306 | eb = path->nodes[0]; | |
1307 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 1308 | if (eb != eb_in) { |
a542ad1b | 1309 | atomic_inc(&eb->refs); |
b916a59a JS |
1310 | btrfs_tree_read_lock(eb); |
1311 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1312 | } | |
a542ad1b | 1313 | btrfs_release_path(path); |
a542ad1b | 1314 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
1315 | |
1316 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1317 | name_off = (unsigned long)(iref + 1); | |
1318 | ||
a542ad1b JS |
1319 | parent = next_inum; |
1320 | --bytes_left; | |
1321 | if (bytes_left >= 0) | |
1322 | dest[bytes_left] = '/'; | |
1323 | } | |
1324 | ||
1325 | btrfs_release_path(path); | |
b916a59a | 1326 | path->leave_spinning = leave_spinning; |
a542ad1b JS |
1327 | |
1328 | if (ret) | |
1329 | return ERR_PTR(ret); | |
1330 | ||
1331 | return dest + bytes_left; | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * this makes the path point to (logical EXTENT_ITEM *) | |
1336 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
1337 | * tree blocks and <0 on error. | |
1338 | */ | |
1339 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
1340 | struct btrfs_path *path, struct btrfs_key *found_key, |
1341 | u64 *flags_ret) | |
a542ad1b JS |
1342 | { |
1343 | int ret; | |
1344 | u64 flags; | |
261c84b6 | 1345 | u64 size = 0; |
a542ad1b JS |
1346 | u32 item_size; |
1347 | struct extent_buffer *eb; | |
1348 | struct btrfs_extent_item *ei; | |
1349 | struct btrfs_key key; | |
1350 | ||
261c84b6 JB |
1351 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1352 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1353 | else | |
1354 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a542ad1b JS |
1355 | key.objectid = logical; |
1356 | key.offset = (u64)-1; | |
1357 | ||
1358 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); | |
1359 | if (ret < 0) | |
1360 | return ret; | |
a542ad1b | 1361 | |
850a8cdf WS |
1362 | ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0); |
1363 | if (ret) { | |
1364 | if (ret > 0) | |
1365 | ret = -ENOENT; | |
1366 | return ret; | |
580f0a67 | 1367 | } |
850a8cdf | 1368 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); |
261c84b6 JB |
1369 | if (found_key->type == BTRFS_METADATA_ITEM_KEY) |
1370 | size = fs_info->extent_root->leafsize; | |
1371 | else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) | |
1372 | size = found_key->offset; | |
1373 | ||
580f0a67 | 1374 | if (found_key->objectid > logical || |
261c84b6 | 1375 | found_key->objectid + size <= logical) { |
c1c9ff7c | 1376 | pr_debug("logical %llu is not within any extent\n", logical); |
a542ad1b | 1377 | return -ENOENT; |
4692cf58 | 1378 | } |
a542ad1b JS |
1379 | |
1380 | eb = path->nodes[0]; | |
1381 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
1382 | BUG_ON(item_size < sizeof(*ei)); | |
1383 | ||
1384 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
1385 | flags = btrfs_extent_flags(eb, ei); | |
1386 | ||
4692cf58 JS |
1387 | pr_debug("logical %llu is at position %llu within the extent (%llu " |
1388 | "EXTENT_ITEM %llu) flags %#llx size %u\n", | |
c1c9ff7c GU |
1389 | logical, logical - found_key->objectid, found_key->objectid, |
1390 | found_key->offset, flags, item_size); | |
69917e43 LB |
1391 | |
1392 | WARN_ON(!flags_ret); | |
1393 | if (flags_ret) { | |
1394 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1395 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
1396 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1397 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
1398 | else | |
1399 | BUG_ON(1); | |
1400 | return 0; | |
1401 | } | |
a542ad1b JS |
1402 | |
1403 | return -EIO; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
1408 | * for the first call and may be modified. it is used to track state. | |
1409 | * if more refs exist, 0 is returned and the next call to | |
1410 | * __get_extent_inline_ref must pass the modified ptr parameter to get the | |
1411 | * next ref. after the last ref was processed, 1 is returned. | |
1412 | * returns <0 on error | |
1413 | */ | |
1414 | static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
1415 | struct btrfs_key *key, |
1416 | struct btrfs_extent_item *ei, u32 item_size, | |
1417 | struct btrfs_extent_inline_ref **out_eiref, | |
1418 | int *out_type) | |
a542ad1b JS |
1419 | { |
1420 | unsigned long end; | |
1421 | u64 flags; | |
1422 | struct btrfs_tree_block_info *info; | |
1423 | ||
1424 | if (!*ptr) { | |
1425 | /* first call */ | |
1426 | flags = btrfs_extent_flags(eb, ei); | |
1427 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
6eda71d0 LB |
1428 | if (key->type == BTRFS_METADATA_ITEM_KEY) { |
1429 | /* a skinny metadata extent */ | |
1430 | *out_eiref = | |
1431 | (struct btrfs_extent_inline_ref *)(ei + 1); | |
1432 | } else { | |
1433 | WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY); | |
1434 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1435 | *out_eiref = | |
1436 | (struct btrfs_extent_inline_ref *)(info + 1); | |
1437 | } | |
a542ad1b JS |
1438 | } else { |
1439 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
1440 | } | |
1441 | *ptr = (unsigned long)*out_eiref; | |
cd857dd6 | 1442 | if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) |
a542ad1b JS |
1443 | return -ENOENT; |
1444 | } | |
1445 | ||
1446 | end = (unsigned long)ei + item_size; | |
6eda71d0 | 1447 | *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); |
a542ad1b JS |
1448 | *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref); |
1449 | ||
1450 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
1451 | WARN_ON(*ptr > end); | |
1452 | if (*ptr == end) | |
1453 | return 1; /* last */ | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | ||
1458 | /* | |
1459 | * reads the tree block backref for an extent. tree level and root are returned | |
1460 | * through out_level and out_root. ptr must point to a 0 value for the first | |
1461 | * call and may be modified (see __get_extent_inline_ref comment). | |
1462 | * returns 0 if data was provided, 1 if there was no more data to provide or | |
1463 | * <0 on error. | |
1464 | */ | |
1465 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
1466 | struct btrfs_key *key, struct btrfs_extent_item *ei, |
1467 | u32 item_size, u64 *out_root, u8 *out_level) | |
a542ad1b JS |
1468 | { |
1469 | int ret; | |
1470 | int type; | |
1471 | struct btrfs_tree_block_info *info; | |
1472 | struct btrfs_extent_inline_ref *eiref; | |
1473 | ||
1474 | if (*ptr == (unsigned long)-1) | |
1475 | return 1; | |
1476 | ||
1477 | while (1) { | |
6eda71d0 LB |
1478 | ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size, |
1479 | &eiref, &type); | |
a542ad1b JS |
1480 | if (ret < 0) |
1481 | return ret; | |
1482 | ||
1483 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1484 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1485 | break; | |
1486 | ||
1487 | if (ret == 1) | |
1488 | return 1; | |
1489 | } | |
1490 | ||
1491 | /* we can treat both ref types equally here */ | |
1492 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1493 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); | |
1494 | *out_level = btrfs_tree_block_level(eb, info); | |
1495 | ||
1496 | if (ret == 1) | |
1497 | *ptr = (unsigned long)-1; | |
1498 | ||
1499 | return 0; | |
1500 | } | |
1501 | ||
976b1908 JS |
1502 | static int iterate_leaf_refs(struct extent_inode_elem *inode_list, |
1503 | u64 root, u64 extent_item_objectid, | |
4692cf58 | 1504 | iterate_extent_inodes_t *iterate, void *ctx) |
a542ad1b | 1505 | { |
976b1908 | 1506 | struct extent_inode_elem *eie; |
4692cf58 | 1507 | int ret = 0; |
4692cf58 | 1508 | |
976b1908 | 1509 | for (eie = inode_list; eie; eie = eie->next) { |
4692cf58 | 1510 | pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), " |
976b1908 JS |
1511 | "root %llu\n", extent_item_objectid, |
1512 | eie->inum, eie->offset, root); | |
1513 | ret = iterate(eie->inum, eie->offset, root, ctx); | |
4692cf58 | 1514 | if (ret) { |
976b1908 JS |
1515 | pr_debug("stopping iteration for %llu due to ret=%d\n", |
1516 | extent_item_objectid, ret); | |
4692cf58 JS |
1517 | break; |
1518 | } | |
a542ad1b JS |
1519 | } |
1520 | ||
a542ad1b JS |
1521 | return ret; |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 1526 | * the given parameters. |
a542ad1b JS |
1527 | * when the iterator function returns a non-zero value, iteration stops. |
1528 | */ | |
1529 | int iterate_extent_inodes(struct btrfs_fs_info *fs_info, | |
4692cf58 | 1530 | u64 extent_item_objectid, u64 extent_item_pos, |
7a3ae2f8 | 1531 | int search_commit_root, |
a542ad1b JS |
1532 | iterate_extent_inodes_t *iterate, void *ctx) |
1533 | { | |
a542ad1b | 1534 | int ret; |
da61d31a | 1535 | struct btrfs_trans_handle *trans = NULL; |
7a3ae2f8 JS |
1536 | struct ulist *refs = NULL; |
1537 | struct ulist *roots = NULL; | |
4692cf58 JS |
1538 | struct ulist_node *ref_node = NULL; |
1539 | struct ulist_node *root_node = NULL; | |
8445f61c | 1540 | struct seq_list tree_mod_seq_elem = {}; |
cd1b413c JS |
1541 | struct ulist_iterator ref_uiter; |
1542 | struct ulist_iterator root_uiter; | |
a542ad1b | 1543 | |
4692cf58 JS |
1544 | pr_debug("resolving all inodes for extent %llu\n", |
1545 | extent_item_objectid); | |
a542ad1b | 1546 | |
da61d31a | 1547 | if (!search_commit_root) { |
7a3ae2f8 JS |
1548 | trans = btrfs_join_transaction(fs_info->extent_root); |
1549 | if (IS_ERR(trans)) | |
1550 | return PTR_ERR(trans); | |
8445f61c | 1551 | btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
9e351cc8 JB |
1552 | } else { |
1553 | down_read(&fs_info->commit_root_sem); | |
7a3ae2f8 | 1554 | } |
a542ad1b | 1555 | |
4692cf58 | 1556 | ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, |
097b8a7c | 1557 | tree_mod_seq_elem.seq, &refs, |
8445f61c | 1558 | &extent_item_pos); |
4692cf58 JS |
1559 | if (ret) |
1560 | goto out; | |
a542ad1b | 1561 | |
cd1b413c JS |
1562 | ULIST_ITER_INIT(&ref_uiter); |
1563 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
9e351cc8 JB |
1564 | ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val, |
1565 | tree_mod_seq_elem.seq, &roots); | |
4692cf58 JS |
1566 | if (ret) |
1567 | break; | |
cd1b413c JS |
1568 | ULIST_ITER_INIT(&root_uiter); |
1569 | while (!ret && (root_node = ulist_next(roots, &root_uiter))) { | |
976b1908 | 1570 | pr_debug("root %llu references leaf %llu, data list " |
34d73f54 | 1571 | "%#llx\n", root_node->val, ref_node->val, |
c1c9ff7c | 1572 | ref_node->aux); |
995e01b7 JS |
1573 | ret = iterate_leaf_refs((struct extent_inode_elem *) |
1574 | (uintptr_t)ref_node->aux, | |
1575 | root_node->val, | |
1576 | extent_item_objectid, | |
1577 | iterate, ctx); | |
4692cf58 | 1578 | } |
976b1908 | 1579 | ulist_free(roots); |
a542ad1b JS |
1580 | } |
1581 | ||
976b1908 | 1582 | free_leaf_list(refs); |
4692cf58 | 1583 | out: |
7a3ae2f8 | 1584 | if (!search_commit_root) { |
8445f61c | 1585 | btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
7a3ae2f8 | 1586 | btrfs_end_transaction(trans, fs_info->extent_root); |
9e351cc8 JB |
1587 | } else { |
1588 | up_read(&fs_info->commit_root_sem); | |
7a3ae2f8 JS |
1589 | } |
1590 | ||
a542ad1b JS |
1591 | return ret; |
1592 | } | |
1593 | ||
1594 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, | |
1595 | struct btrfs_path *path, | |
1596 | iterate_extent_inodes_t *iterate, void *ctx) | |
1597 | { | |
1598 | int ret; | |
4692cf58 | 1599 | u64 extent_item_pos; |
69917e43 | 1600 | u64 flags = 0; |
a542ad1b | 1601 | struct btrfs_key found_key; |
7a3ae2f8 | 1602 | int search_commit_root = path->search_commit_root; |
a542ad1b | 1603 | |
69917e43 | 1604 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 1605 | btrfs_release_path(path); |
a542ad1b JS |
1606 | if (ret < 0) |
1607 | return ret; | |
69917e43 | 1608 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 1609 | return -EINVAL; |
a542ad1b | 1610 | |
4692cf58 | 1611 | extent_item_pos = logical - found_key.objectid; |
7a3ae2f8 JS |
1612 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1613 | extent_item_pos, search_commit_root, | |
1614 | iterate, ctx); | |
a542ad1b JS |
1615 | |
1616 | return ret; | |
1617 | } | |
1618 | ||
d24bec3a MF |
1619 | typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off, |
1620 | struct extent_buffer *eb, void *ctx); | |
1621 | ||
1622 | static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, | |
1623 | struct btrfs_path *path, | |
1624 | iterate_irefs_t *iterate, void *ctx) | |
a542ad1b | 1625 | { |
aefc1eb1 | 1626 | int ret = 0; |
a542ad1b JS |
1627 | int slot; |
1628 | u32 cur; | |
1629 | u32 len; | |
1630 | u32 name_len; | |
1631 | u64 parent = 0; | |
1632 | int found = 0; | |
1633 | struct extent_buffer *eb; | |
1634 | struct btrfs_item *item; | |
1635 | struct btrfs_inode_ref *iref; | |
1636 | struct btrfs_key found_key; | |
1637 | ||
aefc1eb1 | 1638 | while (!ret) { |
a542ad1b | 1639 | ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path, |
d24bec3a | 1640 | &found_key); |
a542ad1b JS |
1641 | if (ret < 0) |
1642 | break; | |
1643 | if (ret) { | |
1644 | ret = found ? 0 : -ENOENT; | |
1645 | break; | |
1646 | } | |
1647 | ++found; | |
1648 | ||
1649 | parent = found_key.offset; | |
1650 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
1651 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
1652 | if (!eb) { | |
1653 | ret = -ENOMEM; | |
1654 | break; | |
1655 | } | |
1656 | extent_buffer_get(eb); | |
b916a59a JS |
1657 | btrfs_tree_read_lock(eb); |
1658 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
a542ad1b JS |
1659 | btrfs_release_path(path); |
1660 | ||
dd3cc16b | 1661 | item = btrfs_item_nr(slot); |
a542ad1b JS |
1662 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
1663 | ||
1664 | for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) { | |
1665 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1666 | /* path must be released before calling iterate()! */ | |
4692cf58 | 1667 | pr_debug("following ref at offset %u for inode %llu in " |
c1c9ff7c GU |
1668 | "tree %llu\n", cur, found_key.objectid, |
1669 | fs_root->objectid); | |
d24bec3a MF |
1670 | ret = iterate(parent, name_len, |
1671 | (unsigned long)(iref + 1), eb, ctx); | |
aefc1eb1 | 1672 | if (ret) |
a542ad1b | 1673 | break; |
a542ad1b JS |
1674 | len = sizeof(*iref) + name_len; |
1675 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
1676 | } | |
b916a59a | 1677 | btrfs_tree_read_unlock_blocking(eb); |
a542ad1b JS |
1678 | free_extent_buffer(eb); |
1679 | } | |
1680 | ||
1681 | btrfs_release_path(path); | |
1682 | ||
1683 | return ret; | |
1684 | } | |
1685 | ||
d24bec3a MF |
1686 | static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, |
1687 | struct btrfs_path *path, | |
1688 | iterate_irefs_t *iterate, void *ctx) | |
1689 | { | |
1690 | int ret; | |
1691 | int slot; | |
1692 | u64 offset = 0; | |
1693 | u64 parent; | |
1694 | int found = 0; | |
1695 | struct extent_buffer *eb; | |
1696 | struct btrfs_inode_extref *extref; | |
1697 | struct extent_buffer *leaf; | |
1698 | u32 item_size; | |
1699 | u32 cur_offset; | |
1700 | unsigned long ptr; | |
1701 | ||
1702 | while (1) { | |
1703 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
1704 | &offset); | |
1705 | if (ret < 0) | |
1706 | break; | |
1707 | if (ret) { | |
1708 | ret = found ? 0 : -ENOENT; | |
1709 | break; | |
1710 | } | |
1711 | ++found; | |
1712 | ||
1713 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
1714 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
1715 | if (!eb) { | |
1716 | ret = -ENOMEM; | |
1717 | break; | |
1718 | } | |
1719 | extent_buffer_get(eb); | |
d24bec3a MF |
1720 | |
1721 | btrfs_tree_read_lock(eb); | |
1722 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1723 | btrfs_release_path(path); | |
1724 | ||
1725 | leaf = path->nodes[0]; | |
e94acd86 VG |
1726 | item_size = btrfs_item_size_nr(leaf, slot); |
1727 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
d24bec3a MF |
1728 | cur_offset = 0; |
1729 | ||
1730 | while (cur_offset < item_size) { | |
1731 | u32 name_len; | |
1732 | ||
1733 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
1734 | parent = btrfs_inode_extref_parent(eb, extref); | |
1735 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
1736 | ret = iterate(parent, name_len, | |
1737 | (unsigned long)&extref->name, eb, ctx); | |
1738 | if (ret) | |
1739 | break; | |
1740 | ||
1741 | cur_offset += btrfs_inode_extref_name_len(leaf, extref); | |
1742 | cur_offset += sizeof(*extref); | |
1743 | } | |
1744 | btrfs_tree_read_unlock_blocking(eb); | |
1745 | free_extent_buffer(eb); | |
1746 | ||
1747 | offset++; | |
1748 | } | |
1749 | ||
1750 | btrfs_release_path(path); | |
1751 | ||
1752 | return ret; | |
1753 | } | |
1754 | ||
1755 | static int iterate_irefs(u64 inum, struct btrfs_root *fs_root, | |
1756 | struct btrfs_path *path, iterate_irefs_t *iterate, | |
1757 | void *ctx) | |
1758 | { | |
1759 | int ret; | |
1760 | int found_refs = 0; | |
1761 | ||
1762 | ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx); | |
1763 | if (!ret) | |
1764 | ++found_refs; | |
1765 | else if (ret != -ENOENT) | |
1766 | return ret; | |
1767 | ||
1768 | ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx); | |
1769 | if (ret == -ENOENT && found_refs) | |
1770 | return 0; | |
1771 | ||
1772 | return ret; | |
1773 | } | |
1774 | ||
a542ad1b JS |
1775 | /* |
1776 | * returns 0 if the path could be dumped (probably truncated) | |
1777 | * returns <0 in case of an error | |
1778 | */ | |
d24bec3a MF |
1779 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
1780 | struct extent_buffer *eb, void *ctx) | |
a542ad1b JS |
1781 | { |
1782 | struct inode_fs_paths *ipath = ctx; | |
1783 | char *fspath; | |
1784 | char *fspath_min; | |
1785 | int i = ipath->fspath->elem_cnt; | |
1786 | const int s_ptr = sizeof(char *); | |
1787 | u32 bytes_left; | |
1788 | ||
1789 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
1790 | ipath->fspath->bytes_left - s_ptr : 0; | |
1791 | ||
740c3d22 | 1792 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
1793 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
1794 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
1795 | if (IS_ERR(fspath)) |
1796 | return PTR_ERR(fspath); | |
1797 | ||
1798 | if (fspath > fspath_min) { | |
745c4d8e | 1799 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
1800 | ++ipath->fspath->elem_cnt; |
1801 | ipath->fspath->bytes_left = fspath - fspath_min; | |
1802 | } else { | |
1803 | ++ipath->fspath->elem_missed; | |
1804 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
1805 | ipath->fspath->bytes_left = 0; | |
1806 | } | |
1807 | ||
1808 | return 0; | |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * this dumps all file system paths to the inode into the ipath struct, provided | |
1813 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 1814 | * from ipath->fspath->val[i]. |
a542ad1b | 1815 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 1816 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
a542ad1b JS |
1817 | * number of missed paths in recored in ipath->fspath->elem_missed, otherwise, |
1818 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would | |
1819 | * have been needed to return all paths. | |
1820 | */ | |
1821 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
1822 | { | |
1823 | return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path, | |
d24bec3a | 1824 | inode_to_path, ipath); |
a542ad1b JS |
1825 | } |
1826 | ||
a542ad1b JS |
1827 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
1828 | { | |
1829 | struct btrfs_data_container *data; | |
1830 | size_t alloc_bytes; | |
1831 | ||
1832 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
425d17a2 | 1833 | data = vmalloc(alloc_bytes); |
a542ad1b JS |
1834 | if (!data) |
1835 | return ERR_PTR(-ENOMEM); | |
1836 | ||
1837 | if (total_bytes >= sizeof(*data)) { | |
1838 | data->bytes_left = total_bytes - sizeof(*data); | |
1839 | data->bytes_missing = 0; | |
1840 | } else { | |
1841 | data->bytes_missing = sizeof(*data) - total_bytes; | |
1842 | data->bytes_left = 0; | |
1843 | } | |
1844 | ||
1845 | data->elem_cnt = 0; | |
1846 | data->elem_missed = 0; | |
1847 | ||
1848 | return data; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * allocates space to return multiple file system paths for an inode. | |
1853 | * total_bytes to allocate are passed, note that space usable for actual path | |
1854 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
1855 | * the returned pointer must be freed with free_ipath() in the end. | |
1856 | */ | |
1857 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
1858 | struct btrfs_path *path) | |
1859 | { | |
1860 | struct inode_fs_paths *ifp; | |
1861 | struct btrfs_data_container *fspath; | |
1862 | ||
1863 | fspath = init_data_container(total_bytes); | |
1864 | if (IS_ERR(fspath)) | |
1865 | return (void *)fspath; | |
1866 | ||
1867 | ifp = kmalloc(sizeof(*ifp), GFP_NOFS); | |
1868 | if (!ifp) { | |
1869 | kfree(fspath); | |
1870 | return ERR_PTR(-ENOMEM); | |
1871 | } | |
1872 | ||
1873 | ifp->btrfs_path = path; | |
1874 | ifp->fspath = fspath; | |
1875 | ifp->fs_root = fs_root; | |
1876 | ||
1877 | return ifp; | |
1878 | } | |
1879 | ||
1880 | void free_ipath(struct inode_fs_paths *ipath) | |
1881 | { | |
4735fb28 JJ |
1882 | if (!ipath) |
1883 | return; | |
425d17a2 | 1884 | vfree(ipath->fspath); |
a542ad1b JS |
1885 | kfree(ipath); |
1886 | } |