Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a542ad1b JS |
2 | /* |
3 | * Copyright (C) 2011 STRATO. All rights reserved. | |
a542ad1b JS |
4 | */ |
5 | ||
f54de068 | 6 | #include <linux/mm.h> |
afce772e | 7 | #include <linux/rbtree.h> |
00142756 | 8 | #include <trace/events/btrfs.h> |
a542ad1b JS |
9 | #include "ctree.h" |
10 | #include "disk-io.h" | |
11 | #include "backref.h" | |
8da6d581 JS |
12 | #include "ulist.h" |
13 | #include "transaction.h" | |
14 | #include "delayed-ref.h" | |
b916a59a | 15 | #include "locking.h" |
1b60d2ec | 16 | #include "misc.h" |
f3a84ccd | 17 | #include "tree-mod-log.h" |
c7f13d42 | 18 | #include "fs.h" |
07e81dc9 | 19 | #include "accessors.h" |
a542ad1b | 20 | |
877c1476 FM |
21 | /* Just arbitrary numbers so we can be sure one of these happened. */ |
22 | #define BACKREF_FOUND_SHARED 6 | |
23 | #define BACKREF_FOUND_NOT_SHARED 7 | |
dc046b10 | 24 | |
976b1908 JS |
25 | struct extent_inode_elem { |
26 | u64 inum; | |
27 | u64 offset; | |
28 | struct extent_inode_elem *next; | |
29 | }; | |
30 | ||
73980bec JM |
31 | static int check_extent_in_eb(const struct btrfs_key *key, |
32 | const struct extent_buffer *eb, | |
33 | const struct btrfs_file_extent_item *fi, | |
34 | u64 extent_item_pos, | |
c995ab3c ZB |
35 | struct extent_inode_elem **eie, |
36 | bool ignore_offset) | |
976b1908 | 37 | { |
8ca15e05 | 38 | u64 offset = 0; |
976b1908 JS |
39 | struct extent_inode_elem *e; |
40 | ||
c995ab3c ZB |
41 | if (!ignore_offset && |
42 | !btrfs_file_extent_compression(eb, fi) && | |
8ca15e05 JB |
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 | ||
73980bec JM |
79 | static int find_extent_in_eb(const struct extent_buffer *eb, |
80 | u64 wanted_disk_byte, u64 extent_item_pos, | |
c995ab3c ZB |
81 | struct extent_inode_elem **eie, |
82 | bool ignore_offset) | |
976b1908 JS |
83 | { |
84 | u64 disk_byte; | |
85 | struct btrfs_key key; | |
86 | struct btrfs_file_extent_item *fi; | |
87 | int slot; | |
88 | int nritems; | |
89 | int extent_type; | |
90 | int ret; | |
91 | ||
92 | /* | |
93 | * from the shared data ref, we only have the leaf but we need | |
94 | * the key. thus, we must look into all items and see that we | |
95 | * find one (some) with a reference to our extent item. | |
96 | */ | |
97 | nritems = btrfs_header_nritems(eb); | |
98 | for (slot = 0; slot < nritems; ++slot) { | |
99 | btrfs_item_key_to_cpu(eb, &key, slot); | |
100 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
101 | continue; | |
102 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
103 | extent_type = btrfs_file_extent_type(eb, fi); | |
104 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
105 | continue; | |
106 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
107 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
108 | if (disk_byte != wanted_disk_byte) | |
109 | continue; | |
110 | ||
c995ab3c | 111 | ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset); |
976b1908 JS |
112 | if (ret < 0) |
113 | return ret; | |
114 | } | |
115 | ||
116 | return 0; | |
117 | } | |
118 | ||
86d5f994 | 119 | struct preftree { |
ecf160b4 | 120 | struct rb_root_cached root; |
6c336b21 | 121 | unsigned int count; |
86d5f994 EN |
122 | }; |
123 | ||
ecf160b4 | 124 | #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 } |
86d5f994 EN |
125 | |
126 | struct preftrees { | |
127 | struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */ | |
128 | struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */ | |
129 | struct preftree indirect_missing_keys; | |
130 | }; | |
131 | ||
3ec4d323 EN |
132 | /* |
133 | * Checks for a shared extent during backref search. | |
134 | * | |
135 | * The share_count tracks prelim_refs (direct and indirect) having a | |
136 | * ref->count >0: | |
137 | * - incremented when a ref->count transitions to >0 | |
138 | * - decremented when a ref->count transitions to <1 | |
139 | */ | |
140 | struct share_check { | |
877c1476 FM |
141 | struct btrfs_backref_share_check_ctx *ctx; |
142 | struct btrfs_root *root; | |
3ec4d323 | 143 | u64 inum; |
73e339e6 | 144 | u64 data_bytenr; |
6976201f | 145 | u64 data_extent_gen; |
73e339e6 FM |
146 | /* |
147 | * Counts number of inodes that refer to an extent (different inodes in | |
148 | * the same root or different roots) that we could find. The sharedness | |
149 | * check typically stops once this counter gets greater than 1, so it | |
150 | * may not reflect the total number of inodes. | |
151 | */ | |
3ec4d323 | 152 | int share_count; |
73e339e6 FM |
153 | /* |
154 | * The number of times we found our inode refers to the data extent we | |
155 | * are determining the sharedness. In other words, how many file extent | |
156 | * items we could find for our inode that point to our target data | |
157 | * extent. The value we get here after finishing the extent sharedness | |
158 | * check may be smaller than reality, but if it ends up being greater | |
159 | * than 1, then we know for sure the inode has multiple file extent | |
160 | * items that point to our inode, and we can safely assume it's useful | |
161 | * to cache the sharedness check result. | |
162 | */ | |
163 | int self_ref_count; | |
4fc7b572 | 164 | bool have_delayed_delete_refs; |
3ec4d323 EN |
165 | }; |
166 | ||
167 | static inline int extent_is_shared(struct share_check *sc) | |
168 | { | |
169 | return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0; | |
170 | } | |
171 | ||
b9e9a6cb WS |
172 | static struct kmem_cache *btrfs_prelim_ref_cache; |
173 | ||
174 | int __init btrfs_prelim_ref_init(void) | |
175 | { | |
176 | btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", | |
e0c476b1 | 177 | sizeof(struct prelim_ref), |
b9e9a6cb | 178 | 0, |
fba4b697 | 179 | SLAB_MEM_SPREAD, |
b9e9a6cb WS |
180 | NULL); |
181 | if (!btrfs_prelim_ref_cache) | |
182 | return -ENOMEM; | |
183 | return 0; | |
184 | } | |
185 | ||
e67c718b | 186 | void __cold btrfs_prelim_ref_exit(void) |
b9e9a6cb | 187 | { |
5598e900 | 188 | kmem_cache_destroy(btrfs_prelim_ref_cache); |
b9e9a6cb WS |
189 | } |
190 | ||
86d5f994 EN |
191 | static void free_pref(struct prelim_ref *ref) |
192 | { | |
193 | kmem_cache_free(btrfs_prelim_ref_cache, ref); | |
194 | } | |
195 | ||
196 | /* | |
197 | * Return 0 when both refs are for the same block (and can be merged). | |
198 | * A -1 return indicates ref1 is a 'lower' block than ref2, while 1 | |
199 | * indicates a 'higher' block. | |
200 | */ | |
201 | static int prelim_ref_compare(struct prelim_ref *ref1, | |
202 | struct prelim_ref *ref2) | |
203 | { | |
204 | if (ref1->level < ref2->level) | |
205 | return -1; | |
206 | if (ref1->level > ref2->level) | |
207 | return 1; | |
208 | if (ref1->root_id < ref2->root_id) | |
209 | return -1; | |
210 | if (ref1->root_id > ref2->root_id) | |
211 | return 1; | |
212 | if (ref1->key_for_search.type < ref2->key_for_search.type) | |
213 | return -1; | |
214 | if (ref1->key_for_search.type > ref2->key_for_search.type) | |
215 | return 1; | |
216 | if (ref1->key_for_search.objectid < ref2->key_for_search.objectid) | |
217 | return -1; | |
218 | if (ref1->key_for_search.objectid > ref2->key_for_search.objectid) | |
219 | return 1; | |
220 | if (ref1->key_for_search.offset < ref2->key_for_search.offset) | |
221 | return -1; | |
222 | if (ref1->key_for_search.offset > ref2->key_for_search.offset) | |
223 | return 1; | |
224 | if (ref1->parent < ref2->parent) | |
225 | return -1; | |
226 | if (ref1->parent > ref2->parent) | |
227 | return 1; | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
ccc8dc75 | 232 | static void update_share_count(struct share_check *sc, int oldcount, |
73e339e6 | 233 | int newcount, struct prelim_ref *newref) |
3ec4d323 EN |
234 | { |
235 | if ((!sc) || (oldcount == 0 && newcount < 1)) | |
236 | return; | |
237 | ||
238 | if (oldcount > 0 && newcount < 1) | |
239 | sc->share_count--; | |
240 | else if (oldcount < 1 && newcount > 0) | |
241 | sc->share_count++; | |
73e339e6 | 242 | |
877c1476 | 243 | if (newref->root_id == sc->root->root_key.objectid && |
73e339e6 FM |
244 | newref->wanted_disk_byte == sc->data_bytenr && |
245 | newref->key_for_search.objectid == sc->inum) | |
246 | sc->self_ref_count += newref->count; | |
3ec4d323 EN |
247 | } |
248 | ||
86d5f994 EN |
249 | /* |
250 | * Add @newref to the @root rbtree, merging identical refs. | |
251 | * | |
3ec4d323 | 252 | * Callers should assume that newref has been freed after calling. |
86d5f994 | 253 | */ |
00142756 JM |
254 | static void prelim_ref_insert(const struct btrfs_fs_info *fs_info, |
255 | struct preftree *preftree, | |
3ec4d323 EN |
256 | struct prelim_ref *newref, |
257 | struct share_check *sc) | |
86d5f994 | 258 | { |
ecf160b4 | 259 | struct rb_root_cached *root; |
86d5f994 EN |
260 | struct rb_node **p; |
261 | struct rb_node *parent = NULL; | |
262 | struct prelim_ref *ref; | |
263 | int result; | |
ecf160b4 | 264 | bool leftmost = true; |
86d5f994 EN |
265 | |
266 | root = &preftree->root; | |
ecf160b4 | 267 | p = &root->rb_root.rb_node; |
86d5f994 EN |
268 | |
269 | while (*p) { | |
270 | parent = *p; | |
271 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
272 | result = prelim_ref_compare(ref, newref); | |
273 | if (result < 0) { | |
274 | p = &(*p)->rb_left; | |
275 | } else if (result > 0) { | |
276 | p = &(*p)->rb_right; | |
ecf160b4 | 277 | leftmost = false; |
86d5f994 EN |
278 | } else { |
279 | /* Identical refs, merge them and free @newref */ | |
280 | struct extent_inode_elem *eie = ref->inode_list; | |
281 | ||
282 | while (eie && eie->next) | |
283 | eie = eie->next; | |
284 | ||
285 | if (!eie) | |
286 | ref->inode_list = newref->inode_list; | |
287 | else | |
288 | eie->next = newref->inode_list; | |
00142756 JM |
289 | trace_btrfs_prelim_ref_merge(fs_info, ref, newref, |
290 | preftree->count); | |
3ec4d323 EN |
291 | /* |
292 | * A delayed ref can have newref->count < 0. | |
293 | * The ref->count is updated to follow any | |
294 | * BTRFS_[ADD|DROP]_DELAYED_REF actions. | |
295 | */ | |
296 | update_share_count(sc, ref->count, | |
73e339e6 | 297 | ref->count + newref->count, newref); |
86d5f994 EN |
298 | ref->count += newref->count; |
299 | free_pref(newref); | |
300 | return; | |
301 | } | |
302 | } | |
303 | ||
73e339e6 | 304 | update_share_count(sc, 0, newref->count, newref); |
6c336b21 | 305 | preftree->count++; |
00142756 | 306 | trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count); |
86d5f994 | 307 | rb_link_node(&newref->rbnode, parent, p); |
ecf160b4 | 308 | rb_insert_color_cached(&newref->rbnode, root, leftmost); |
86d5f994 EN |
309 | } |
310 | ||
311 | /* | |
312 | * Release the entire tree. We don't care about internal consistency so | |
313 | * just free everything and then reset the tree root. | |
314 | */ | |
315 | static void prelim_release(struct preftree *preftree) | |
316 | { | |
317 | struct prelim_ref *ref, *next_ref; | |
318 | ||
ecf160b4 | 319 | rbtree_postorder_for_each_entry_safe(ref, next_ref, |
92876eec FM |
320 | &preftree->root.rb_root, rbnode) { |
321 | free_inode_elem_list(ref->inode_list); | |
86d5f994 | 322 | free_pref(ref); |
92876eec | 323 | } |
86d5f994 | 324 | |
ecf160b4 | 325 | preftree->root = RB_ROOT_CACHED; |
6c336b21 | 326 | preftree->count = 0; |
86d5f994 EN |
327 | } |
328 | ||
d5c88b73 JS |
329 | /* |
330 | * the rules for all callers of this function are: | |
331 | * - obtaining the parent is the goal | |
332 | * - if you add a key, you must know that it is a correct key | |
333 | * - if you cannot add the parent or a correct key, then we will look into the | |
334 | * block later to set a correct key | |
335 | * | |
336 | * delayed refs | |
337 | * ============ | |
338 | * backref type | shared | indirect | shared | indirect | |
339 | * information | tree | tree | data | data | |
340 | * --------------------+--------+----------+--------+---------- | |
341 | * parent logical | y | - | - | - | |
342 | * key to resolve | - | y | y | y | |
343 | * tree block logical | - | - | - | - | |
344 | * root for resolving | y | y | y | y | |
345 | * | |
346 | * - column 1: we've the parent -> done | |
347 | * - column 2, 3, 4: we use the key to find the parent | |
348 | * | |
349 | * on disk refs (inline or keyed) | |
350 | * ============================== | |
351 | * backref type | shared | indirect | shared | indirect | |
352 | * information | tree | tree | data | data | |
353 | * --------------------+--------+----------+--------+---------- | |
354 | * parent logical | y | - | y | - | |
355 | * key to resolve | - | - | - | y | |
356 | * tree block logical | y | y | y | y | |
357 | * root for resolving | - | y | y | y | |
358 | * | |
359 | * - column 1, 3: we've the parent -> done | |
360 | * - column 2: we take the first key from the block to find the parent | |
e0c476b1 | 361 | * (see add_missing_keys) |
d5c88b73 JS |
362 | * - column 4: we use the key to find the parent |
363 | * | |
364 | * additional information that's available but not required to find the parent | |
365 | * block might help in merging entries to gain some speed. | |
366 | */ | |
00142756 JM |
367 | static int add_prelim_ref(const struct btrfs_fs_info *fs_info, |
368 | struct preftree *preftree, u64 root_id, | |
e0c476b1 | 369 | const struct btrfs_key *key, int level, u64 parent, |
3ec4d323 EN |
370 | u64 wanted_disk_byte, int count, |
371 | struct share_check *sc, gfp_t gfp_mask) | |
8da6d581 | 372 | { |
e0c476b1 | 373 | struct prelim_ref *ref; |
8da6d581 | 374 | |
48ec4736 LB |
375 | if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) |
376 | return 0; | |
377 | ||
b9e9a6cb | 378 | ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); |
8da6d581 JS |
379 | if (!ref) |
380 | return -ENOMEM; | |
381 | ||
382 | ref->root_id = root_id; | |
7ac8b88e | 383 | if (key) |
d5c88b73 | 384 | ref->key_for_search = *key; |
7ac8b88e | 385 | else |
d5c88b73 | 386 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 387 | |
3301958b | 388 | ref->inode_list = NULL; |
8da6d581 JS |
389 | ref->level = level; |
390 | ref->count = count; | |
391 | ref->parent = parent; | |
392 | ref->wanted_disk_byte = wanted_disk_byte; | |
3ec4d323 EN |
393 | prelim_ref_insert(fs_info, preftree, ref, sc); |
394 | return extent_is_shared(sc); | |
8da6d581 JS |
395 | } |
396 | ||
86d5f994 | 397 | /* direct refs use root == 0, key == NULL */ |
00142756 JM |
398 | static int add_direct_ref(const struct btrfs_fs_info *fs_info, |
399 | struct preftrees *preftrees, int level, u64 parent, | |
3ec4d323 EN |
400 | u64 wanted_disk_byte, int count, |
401 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 | 402 | { |
00142756 | 403 | return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level, |
3ec4d323 | 404 | parent, wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
405 | } |
406 | ||
407 | /* indirect refs use parent == 0 */ | |
00142756 JM |
408 | static int add_indirect_ref(const struct btrfs_fs_info *fs_info, |
409 | struct preftrees *preftrees, u64 root_id, | |
86d5f994 | 410 | const struct btrfs_key *key, int level, |
3ec4d323 EN |
411 | u64 wanted_disk_byte, int count, |
412 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 EN |
413 | { |
414 | struct preftree *tree = &preftrees->indirect; | |
415 | ||
416 | if (!key) | |
417 | tree = &preftrees->indirect_missing_keys; | |
00142756 | 418 | return add_prelim_ref(fs_info, tree, root_id, key, level, 0, |
3ec4d323 | 419 | wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
420 | } |
421 | ||
ed58f2e6 | 422 | static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr) |
423 | { | |
424 | struct rb_node **p = &preftrees->direct.root.rb_root.rb_node; | |
425 | struct rb_node *parent = NULL; | |
426 | struct prelim_ref *ref = NULL; | |
9c6c723f | 427 | struct prelim_ref target = {}; |
ed58f2e6 | 428 | int result; |
429 | ||
430 | target.parent = bytenr; | |
431 | ||
432 | while (*p) { | |
433 | parent = *p; | |
434 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
435 | result = prelim_ref_compare(ref, &target); | |
436 | ||
437 | if (result < 0) | |
438 | p = &(*p)->rb_left; | |
439 | else if (result > 0) | |
440 | p = &(*p)->rb_right; | |
441 | else | |
442 | return 1; | |
443 | } | |
444 | return 0; | |
445 | } | |
446 | ||
8da6d581 | 447 | static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, |
ed58f2e6 | 448 | struct ulist *parents, |
449 | struct preftrees *preftrees, struct prelim_ref *ref, | |
44853868 | 450 | int level, u64 time_seq, const u64 *extent_item_pos, |
b25b0b87 | 451 | bool ignore_offset) |
8da6d581 | 452 | { |
69bca40d AB |
453 | int ret = 0; |
454 | int slot; | |
455 | struct extent_buffer *eb; | |
456 | struct btrfs_key key; | |
7ef81ac8 | 457 | struct btrfs_key *key_for_search = &ref->key_for_search; |
8da6d581 | 458 | struct btrfs_file_extent_item *fi; |
ed8c4913 | 459 | struct extent_inode_elem *eie = NULL, *old = NULL; |
8da6d581 | 460 | u64 disk_byte; |
7ef81ac8 JB |
461 | u64 wanted_disk_byte = ref->wanted_disk_byte; |
462 | u64 count = 0; | |
7ac8b88e | 463 | u64 data_offset; |
8da6d581 | 464 | |
69bca40d AB |
465 | if (level != 0) { |
466 | eb = path->nodes[level]; | |
467 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
468 | if (ret < 0) |
469 | return ret; | |
8da6d581 | 470 | return 0; |
69bca40d | 471 | } |
8da6d581 JS |
472 | |
473 | /* | |
ed58f2e6 | 474 | * 1. We normally enter this function with the path already pointing to |
475 | * the first item to check. But sometimes, we may enter it with | |
476 | * slot == nritems. | |
477 | * 2. We are searching for normal backref but bytenr of this leaf | |
478 | * matches shared data backref | |
cfc0eed0 | 479 | * 3. The leaf owner is not equal to the root we are searching |
480 | * | |
ed58f2e6 | 481 | * For these cases, go to the next leaf before we continue. |
8da6d581 | 482 | */ |
ed58f2e6 | 483 | eb = path->nodes[0]; |
484 | if (path->slots[0] >= btrfs_header_nritems(eb) || | |
cfc0eed0 | 485 | is_shared_data_backref(preftrees, eb->start) || |
486 | ref->root_id != btrfs_header_owner(eb)) { | |
f3a84ccd | 487 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
488 | ret = btrfs_next_leaf(root, path); |
489 | else | |
490 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
491 | } | |
8da6d581 | 492 | |
b25b0b87 | 493 | while (!ret && count < ref->count) { |
8da6d581 | 494 | eb = path->nodes[0]; |
69bca40d AB |
495 | slot = path->slots[0]; |
496 | ||
497 | btrfs_item_key_to_cpu(eb, &key, slot); | |
498 | ||
499 | if (key.objectid != key_for_search->objectid || | |
500 | key.type != BTRFS_EXTENT_DATA_KEY) | |
501 | break; | |
502 | ||
ed58f2e6 | 503 | /* |
504 | * We are searching for normal backref but bytenr of this leaf | |
cfc0eed0 | 505 | * matches shared data backref, OR |
506 | * the leaf owner is not equal to the root we are searching for | |
ed58f2e6 | 507 | */ |
cfc0eed0 | 508 | if (slot == 0 && |
509 | (is_shared_data_backref(preftrees, eb->start) || | |
510 | ref->root_id != btrfs_header_owner(eb))) { | |
f3a84ccd | 511 | if (time_seq == BTRFS_SEQ_LAST) |
ed58f2e6 | 512 | ret = btrfs_next_leaf(root, path); |
513 | else | |
514 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
515 | continue; | |
516 | } | |
69bca40d AB |
517 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
518 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
7ac8b88e | 519 | data_offset = btrfs_file_extent_offset(eb, fi); |
69bca40d AB |
520 | |
521 | if (disk_byte == wanted_disk_byte) { | |
522 | eie = NULL; | |
ed8c4913 | 523 | old = NULL; |
7ac8b88e | 524 | if (ref->key_for_search.offset == key.offset - data_offset) |
525 | count++; | |
526 | else | |
527 | goto next; | |
69bca40d AB |
528 | if (extent_item_pos) { |
529 | ret = check_extent_in_eb(&key, eb, fi, | |
530 | *extent_item_pos, | |
c995ab3c | 531 | &eie, ignore_offset); |
69bca40d AB |
532 | if (ret < 0) |
533 | break; | |
534 | } | |
ed8c4913 JB |
535 | if (ret > 0) |
536 | goto next; | |
4eb1f66d TI |
537 | ret = ulist_add_merge_ptr(parents, eb->start, |
538 | eie, (void **)&old, GFP_NOFS); | |
ed8c4913 JB |
539 | if (ret < 0) |
540 | break; | |
541 | if (!ret && extent_item_pos) { | |
542 | while (old->next) | |
543 | old = old->next; | |
544 | old->next = eie; | |
69bca40d | 545 | } |
f05c4746 | 546 | eie = NULL; |
8da6d581 | 547 | } |
ed8c4913 | 548 | next: |
f3a84ccd | 549 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
550 | ret = btrfs_next_item(root, path); |
551 | else | |
552 | ret = btrfs_next_old_item(root, path, time_seq); | |
8da6d581 JS |
553 | } |
554 | ||
69bca40d AB |
555 | if (ret > 0) |
556 | ret = 0; | |
f05c4746 WS |
557 | else if (ret < 0) |
558 | free_inode_elem_list(eie); | |
69bca40d | 559 | return ret; |
8da6d581 JS |
560 | } |
561 | ||
562 | /* | |
563 | * resolve an indirect backref in the form (root_id, key, level) | |
564 | * to a logical address | |
565 | */ | |
e0c476b1 JM |
566 | static int resolve_indirect_ref(struct btrfs_fs_info *fs_info, |
567 | struct btrfs_path *path, u64 time_seq, | |
ed58f2e6 | 568 | struct preftrees *preftrees, |
e0c476b1 | 569 | struct prelim_ref *ref, struct ulist *parents, |
b25b0b87 | 570 | const u64 *extent_item_pos, bool ignore_offset) |
8da6d581 | 571 | { |
8da6d581 | 572 | struct btrfs_root *root; |
8da6d581 JS |
573 | struct extent_buffer *eb; |
574 | int ret = 0; | |
575 | int root_level; | |
576 | int level = ref->level; | |
7ac8b88e | 577 | struct btrfs_key search_key = ref->key_for_search; |
8da6d581 | 578 | |
49d11bea JB |
579 | /* |
580 | * If we're search_commit_root we could possibly be holding locks on | |
581 | * other tree nodes. This happens when qgroups does backref walks when | |
582 | * adding new delayed refs. To deal with this we need to look in cache | |
583 | * for the root, and if we don't find it then we need to search the | |
584 | * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage | |
585 | * here. | |
586 | */ | |
587 | if (path->search_commit_root) | |
588 | root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id); | |
589 | else | |
590 | root = btrfs_get_fs_root(fs_info, ref->root_id, false); | |
8da6d581 JS |
591 | if (IS_ERR(root)) { |
592 | ret = PTR_ERR(root); | |
9326f76f JB |
593 | goto out_free; |
594 | } | |
595 | ||
39dba873 JB |
596 | if (!path->search_commit_root && |
597 | test_bit(BTRFS_ROOT_DELETING, &root->state)) { | |
598 | ret = -ENOENT; | |
599 | goto out; | |
600 | } | |
601 | ||
f5ee5c9a | 602 | if (btrfs_is_testing(fs_info)) { |
d9ee522b JB |
603 | ret = -ENOENT; |
604 | goto out; | |
605 | } | |
606 | ||
9e351cc8 JB |
607 | if (path->search_commit_root) |
608 | root_level = btrfs_header_level(root->commit_root); | |
f3a84ccd | 609 | else if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 | 610 | root_level = btrfs_header_level(root->node); |
9e351cc8 JB |
611 | else |
612 | root_level = btrfs_old_root_level(root, time_seq); | |
8da6d581 | 613 | |
c75e8394 | 614 | if (root_level + 1 == level) |
8da6d581 JS |
615 | goto out; |
616 | ||
7ac8b88e | 617 | /* |
618 | * We can often find data backrefs with an offset that is too large | |
619 | * (>= LLONG_MAX, maximum allowed file offset) due to underflows when | |
620 | * subtracting a file's offset with the data offset of its | |
621 | * corresponding extent data item. This can happen for example in the | |
622 | * clone ioctl. | |
623 | * | |
624 | * So if we detect such case we set the search key's offset to zero to | |
625 | * make sure we will find the matching file extent item at | |
626 | * add_all_parents(), otherwise we will miss it because the offset | |
627 | * taken form the backref is much larger then the offset of the file | |
628 | * extent item. This can make us scan a very large number of file | |
629 | * extent items, but at least it will not make us miss any. | |
630 | * | |
631 | * This is an ugly workaround for a behaviour that should have never | |
632 | * existed, but it does and a fix for the clone ioctl would touch a lot | |
633 | * of places, cause backwards incompatibility and would not fix the | |
634 | * problem for extents cloned with older kernels. | |
635 | */ | |
636 | if (search_key.type == BTRFS_EXTENT_DATA_KEY && | |
637 | search_key.offset >= LLONG_MAX) | |
638 | search_key.offset = 0; | |
8da6d581 | 639 | path->lowest_level = level; |
f3a84ccd | 640 | if (time_seq == BTRFS_SEQ_LAST) |
7ac8b88e | 641 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); |
21633fc6 | 642 | else |
7ac8b88e | 643 | ret = btrfs_search_old_slot(root, &search_key, path, time_seq); |
538f72cd | 644 | |
ab8d0fc4 JM |
645 | btrfs_debug(fs_info, |
646 | "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)", | |
c1c9ff7c GU |
647 | ref->root_id, level, ref->count, ret, |
648 | ref->key_for_search.objectid, ref->key_for_search.type, | |
649 | ref->key_for_search.offset); | |
8da6d581 JS |
650 | if (ret < 0) |
651 | goto out; | |
652 | ||
653 | eb = path->nodes[level]; | |
9345457f | 654 | while (!eb) { |
fae7f21c | 655 | if (WARN_ON(!level)) { |
9345457f JS |
656 | ret = 1; |
657 | goto out; | |
658 | } | |
659 | level--; | |
660 | eb = path->nodes[level]; | |
8da6d581 JS |
661 | } |
662 | ||
ed58f2e6 | 663 | ret = add_all_parents(root, path, parents, preftrees, ref, level, |
b25b0b87 | 664 | time_seq, extent_item_pos, ignore_offset); |
8da6d581 | 665 | out: |
00246528 | 666 | btrfs_put_root(root); |
9326f76f | 667 | out_free: |
da61d31a JB |
668 | path->lowest_level = 0; |
669 | btrfs_release_path(path); | |
8da6d581 JS |
670 | return ret; |
671 | } | |
672 | ||
4dae077a JM |
673 | static struct extent_inode_elem * |
674 | unode_aux_to_inode_list(struct ulist_node *node) | |
675 | { | |
676 | if (!node) | |
677 | return NULL; | |
678 | return (struct extent_inode_elem *)(uintptr_t)node->aux; | |
679 | } | |
680 | ||
5614dc3a FM |
681 | static void free_leaf_list(struct ulist *ulist) |
682 | { | |
683 | struct ulist_node *node; | |
684 | struct ulist_iterator uiter; | |
685 | ||
686 | ULIST_ITER_INIT(&uiter); | |
687 | while ((node = ulist_next(ulist, &uiter))) | |
688 | free_inode_elem_list(unode_aux_to_inode_list(node)); | |
689 | ||
690 | ulist_free(ulist); | |
691 | } | |
692 | ||
8da6d581 | 693 | /* |
52042d8e | 694 | * We maintain three separate rbtrees: one for direct refs, one for |
86d5f994 EN |
695 | * indirect refs which have a key, and one for indirect refs which do not |
696 | * have a key. Each tree does merge on insertion. | |
697 | * | |
698 | * Once all of the references are located, we iterate over the tree of | |
699 | * indirect refs with missing keys. An appropriate key is located and | |
700 | * the ref is moved onto the tree for indirect refs. After all missing | |
701 | * keys are thus located, we iterate over the indirect ref tree, resolve | |
702 | * each reference, and then insert the resolved reference onto the | |
703 | * direct tree (merging there too). | |
704 | * | |
705 | * New backrefs (i.e., for parent nodes) are added to the appropriate | |
706 | * rbtree as they are encountered. The new backrefs are subsequently | |
707 | * resolved as above. | |
8da6d581 | 708 | */ |
e0c476b1 JM |
709 | static int resolve_indirect_refs(struct btrfs_fs_info *fs_info, |
710 | struct btrfs_path *path, u64 time_seq, | |
86d5f994 | 711 | struct preftrees *preftrees, |
b25b0b87 | 712 | const u64 *extent_item_pos, |
c995ab3c | 713 | struct share_check *sc, bool ignore_offset) |
8da6d581 JS |
714 | { |
715 | int err; | |
716 | int ret = 0; | |
8da6d581 JS |
717 | struct ulist *parents; |
718 | struct ulist_node *node; | |
cd1b413c | 719 | struct ulist_iterator uiter; |
86d5f994 | 720 | struct rb_node *rnode; |
8da6d581 JS |
721 | |
722 | parents = ulist_alloc(GFP_NOFS); | |
723 | if (!parents) | |
724 | return -ENOMEM; | |
725 | ||
726 | /* | |
86d5f994 EN |
727 | * We could trade memory usage for performance here by iterating |
728 | * the tree, allocating new refs for each insertion, and then | |
729 | * freeing the entire indirect tree when we're done. In some test | |
730 | * cases, the tree can grow quite large (~200k objects). | |
8da6d581 | 731 | */ |
ecf160b4 | 732 | while ((rnode = rb_first_cached(&preftrees->indirect.root))) { |
86d5f994 EN |
733 | struct prelim_ref *ref; |
734 | ||
735 | ref = rb_entry(rnode, struct prelim_ref, rbnode); | |
736 | if (WARN(ref->parent, | |
737 | "BUG: direct ref found in indirect tree")) { | |
738 | ret = -EINVAL; | |
739 | goto out; | |
740 | } | |
741 | ||
ecf160b4 | 742 | rb_erase_cached(&ref->rbnode, &preftrees->indirect.root); |
6c336b21 | 743 | preftrees->indirect.count--; |
86d5f994 EN |
744 | |
745 | if (ref->count == 0) { | |
746 | free_pref(ref); | |
8da6d581 | 747 | continue; |
86d5f994 EN |
748 | } |
749 | ||
877c1476 | 750 | if (sc && ref->root_id != sc->root->root_key.objectid) { |
86d5f994 | 751 | free_pref(ref); |
dc046b10 JB |
752 | ret = BACKREF_FOUND_SHARED; |
753 | goto out; | |
754 | } | |
ed58f2e6 | 755 | err = resolve_indirect_ref(fs_info, path, time_seq, preftrees, |
756 | ref, parents, extent_item_pos, | |
b25b0b87 | 757 | ignore_offset); |
95def2ed WS |
758 | /* |
759 | * we can only tolerate ENOENT,otherwise,we should catch error | |
760 | * and return directly. | |
761 | */ | |
762 | if (err == -ENOENT) { | |
3ec4d323 EN |
763 | prelim_ref_insert(fs_info, &preftrees->direct, ref, |
764 | NULL); | |
8da6d581 | 765 | continue; |
95def2ed | 766 | } else if (err) { |
86d5f994 | 767 | free_pref(ref); |
95def2ed WS |
768 | ret = err; |
769 | goto out; | |
770 | } | |
8da6d581 JS |
771 | |
772 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
773 | ULIST_ITER_INIT(&uiter); |
774 | node = ulist_next(parents, &uiter); | |
8da6d581 | 775 | ref->parent = node ? node->val : 0; |
4dae077a | 776 | ref->inode_list = unode_aux_to_inode_list(node); |
8da6d581 | 777 | |
86d5f994 | 778 | /* Add a prelim_ref(s) for any other parent(s). */ |
cd1b413c | 779 | while ((node = ulist_next(parents, &uiter))) { |
86d5f994 EN |
780 | struct prelim_ref *new_ref; |
781 | ||
b9e9a6cb WS |
782 | new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, |
783 | GFP_NOFS); | |
8da6d581 | 784 | if (!new_ref) { |
86d5f994 | 785 | free_pref(ref); |
8da6d581 | 786 | ret = -ENOMEM; |
e36902d4 | 787 | goto out; |
8da6d581 JS |
788 | } |
789 | memcpy(new_ref, ref, sizeof(*ref)); | |
790 | new_ref->parent = node->val; | |
4dae077a | 791 | new_ref->inode_list = unode_aux_to_inode_list(node); |
3ec4d323 EN |
792 | prelim_ref_insert(fs_info, &preftrees->direct, |
793 | new_ref, NULL); | |
8da6d581 | 794 | } |
86d5f994 | 795 | |
3ec4d323 | 796 | /* |
52042d8e | 797 | * Now it's a direct ref, put it in the direct tree. We must |
3ec4d323 EN |
798 | * do this last because the ref could be merged/freed here. |
799 | */ | |
800 | prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL); | |
86d5f994 | 801 | |
8da6d581 | 802 | ulist_reinit(parents); |
9dd14fd6 | 803 | cond_resched(); |
8da6d581 | 804 | } |
e36902d4 | 805 | out: |
5614dc3a FM |
806 | /* |
807 | * We may have inode lists attached to refs in the parents ulist, so we | |
808 | * must free them before freeing the ulist and its refs. | |
809 | */ | |
810 | free_leaf_list(parents); | |
8da6d581 JS |
811 | return ret; |
812 | } | |
813 | ||
d5c88b73 JS |
814 | /* |
815 | * read tree blocks and add keys where required. | |
816 | */ | |
e0c476b1 | 817 | static int add_missing_keys(struct btrfs_fs_info *fs_info, |
38e3eebf | 818 | struct preftrees *preftrees, bool lock) |
d5c88b73 | 819 | { |
e0c476b1 | 820 | struct prelim_ref *ref; |
d5c88b73 | 821 | struct extent_buffer *eb; |
86d5f994 EN |
822 | struct preftree *tree = &preftrees->indirect_missing_keys; |
823 | struct rb_node *node; | |
d5c88b73 | 824 | |
ecf160b4 | 825 | while ((node = rb_first_cached(&tree->root))) { |
86d5f994 | 826 | ref = rb_entry(node, struct prelim_ref, rbnode); |
ecf160b4 | 827 | rb_erase_cached(node, &tree->root); |
86d5f994 EN |
828 | |
829 | BUG_ON(ref->parent); /* should not be a direct ref */ | |
830 | BUG_ON(ref->key_for_search.type); | |
d5c88b73 | 831 | BUG_ON(!ref->wanted_disk_byte); |
86d5f994 | 832 | |
1b7ec85e JB |
833 | eb = read_tree_block(fs_info, ref->wanted_disk_byte, |
834 | ref->root_id, 0, ref->level - 1, NULL); | |
64c043de | 835 | if (IS_ERR(eb)) { |
86d5f994 | 836 | free_pref(ref); |
64c043de | 837 | return PTR_ERR(eb); |
4eb150d6 QW |
838 | } |
839 | if (!extent_buffer_uptodate(eb)) { | |
86d5f994 | 840 | free_pref(ref); |
416bc658 JB |
841 | free_extent_buffer(eb); |
842 | return -EIO; | |
843 | } | |
4eb150d6 | 844 | |
38e3eebf JB |
845 | if (lock) |
846 | btrfs_tree_read_lock(eb); | |
d5c88b73 JS |
847 | if (btrfs_header_level(eb) == 0) |
848 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
849 | else | |
850 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
38e3eebf JB |
851 | if (lock) |
852 | btrfs_tree_read_unlock(eb); | |
d5c88b73 | 853 | free_extent_buffer(eb); |
3ec4d323 | 854 | prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL); |
9dd14fd6 | 855 | cond_resched(); |
d5c88b73 JS |
856 | } |
857 | return 0; | |
858 | } | |
859 | ||
8da6d581 JS |
860 | /* |
861 | * add all currently queued delayed refs from this head whose seq nr is | |
862 | * smaller or equal that seq to the list | |
863 | */ | |
00142756 JM |
864 | static int add_delayed_refs(const struct btrfs_fs_info *fs_info, |
865 | struct btrfs_delayed_ref_head *head, u64 seq, | |
b25b0b87 | 866 | struct preftrees *preftrees, struct share_check *sc) |
8da6d581 | 867 | { |
c6fc2454 | 868 | struct btrfs_delayed_ref_node *node; |
d5c88b73 | 869 | struct btrfs_key key; |
0e0adbcf | 870 | struct rb_node *n; |
01747e92 | 871 | int count; |
b1375d64 | 872 | int ret = 0; |
8da6d581 | 873 | |
d7df2c79 | 874 | spin_lock(&head->lock); |
e3d03965 | 875 | for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) { |
0e0adbcf JB |
876 | node = rb_entry(n, struct btrfs_delayed_ref_node, |
877 | ref_node); | |
8da6d581 JS |
878 | if (node->seq > seq) |
879 | continue; | |
880 | ||
881 | switch (node->action) { | |
882 | case BTRFS_ADD_DELAYED_EXTENT: | |
883 | case BTRFS_UPDATE_DELAYED_HEAD: | |
884 | WARN_ON(1); | |
885 | continue; | |
886 | case BTRFS_ADD_DELAYED_REF: | |
01747e92 | 887 | count = node->ref_mod; |
8da6d581 JS |
888 | break; |
889 | case BTRFS_DROP_DELAYED_REF: | |
01747e92 | 890 | count = node->ref_mod * -1; |
8da6d581 JS |
891 | break; |
892 | default: | |
290342f6 | 893 | BUG(); |
8da6d581 JS |
894 | } |
895 | switch (node->type) { | |
896 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
86d5f994 | 897 | /* NORMAL INDIRECT METADATA backref */ |
8da6d581 | 898 | struct btrfs_delayed_tree_ref *ref; |
943553ef FM |
899 | struct btrfs_key *key_ptr = NULL; |
900 | ||
901 | if (head->extent_op && head->extent_op->update_key) { | |
902 | btrfs_disk_key_to_cpu(&key, &head->extent_op->key); | |
903 | key_ptr = &key; | |
904 | } | |
8da6d581 JS |
905 | |
906 | ref = btrfs_delayed_node_to_tree_ref(node); | |
00142756 | 907 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
943553ef | 908 | key_ptr, ref->level + 1, |
01747e92 EN |
909 | node->bytenr, count, sc, |
910 | GFP_ATOMIC); | |
8da6d581 JS |
911 | break; |
912 | } | |
913 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
86d5f994 | 914 | /* SHARED DIRECT METADATA backref */ |
8da6d581 JS |
915 | struct btrfs_delayed_tree_ref *ref; |
916 | ||
917 | ref = btrfs_delayed_node_to_tree_ref(node); | |
86d5f994 | 918 | |
01747e92 EN |
919 | ret = add_direct_ref(fs_info, preftrees, ref->level + 1, |
920 | ref->parent, node->bytenr, count, | |
3ec4d323 | 921 | sc, GFP_ATOMIC); |
8da6d581 JS |
922 | break; |
923 | } | |
924 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 925 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 | 926 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
927 | ref = btrfs_delayed_node_to_data_ref(node); |
928 | ||
929 | key.objectid = ref->objectid; | |
930 | key.type = BTRFS_EXTENT_DATA_KEY; | |
931 | key.offset = ref->offset; | |
dc046b10 JB |
932 | |
933 | /* | |
4fc7b572 FM |
934 | * If we have a share check context and a reference for |
935 | * another inode, we can't exit immediately. This is | |
936 | * because even if this is a BTRFS_ADD_DELAYED_REF | |
937 | * reference we may find next a BTRFS_DROP_DELAYED_REF | |
938 | * which cancels out this ADD reference. | |
939 | * | |
940 | * If this is a DROP reference and there was no previous | |
941 | * ADD reference, then we need to signal that when we | |
942 | * process references from the extent tree (through | |
943 | * add_inline_refs() and add_keyed_refs()), we should | |
944 | * not exit early if we find a reference for another | |
945 | * inode, because one of the delayed DROP references | |
946 | * may cancel that reference in the extent tree. | |
dc046b10 | 947 | */ |
4fc7b572 FM |
948 | if (sc && count < 0) |
949 | sc->have_delayed_delete_refs = true; | |
dc046b10 | 950 | |
00142756 | 951 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
01747e92 EN |
952 | &key, 0, node->bytenr, count, sc, |
953 | GFP_ATOMIC); | |
8da6d581 JS |
954 | break; |
955 | } | |
956 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 957 | /* SHARED DIRECT FULL backref */ |
8da6d581 | 958 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
959 | |
960 | ref = btrfs_delayed_node_to_data_ref(node); | |
86d5f994 | 961 | |
01747e92 EN |
962 | ret = add_direct_ref(fs_info, preftrees, 0, ref->parent, |
963 | node->bytenr, count, sc, | |
964 | GFP_ATOMIC); | |
8da6d581 JS |
965 | break; |
966 | } | |
967 | default: | |
968 | WARN_ON(1); | |
969 | } | |
3ec4d323 EN |
970 | /* |
971 | * We must ignore BACKREF_FOUND_SHARED until all delayed | |
972 | * refs have been checked. | |
973 | */ | |
974 | if (ret && (ret != BACKREF_FOUND_SHARED)) | |
d7df2c79 | 975 | break; |
8da6d581 | 976 | } |
3ec4d323 EN |
977 | if (!ret) |
978 | ret = extent_is_shared(sc); | |
4fc7b572 | 979 | |
d7df2c79 JB |
980 | spin_unlock(&head->lock); |
981 | return ret; | |
8da6d581 JS |
982 | } |
983 | ||
984 | /* | |
985 | * add all inline backrefs for bytenr to the list | |
3ec4d323 EN |
986 | * |
987 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 988 | */ |
00142756 JM |
989 | static int add_inline_refs(const struct btrfs_fs_info *fs_info, |
990 | struct btrfs_path *path, u64 bytenr, | |
86d5f994 | 991 | int *info_level, struct preftrees *preftrees, |
b25b0b87 | 992 | struct share_check *sc) |
8da6d581 | 993 | { |
b1375d64 | 994 | int ret = 0; |
8da6d581 JS |
995 | int slot; |
996 | struct extent_buffer *leaf; | |
997 | struct btrfs_key key; | |
261c84b6 | 998 | struct btrfs_key found_key; |
8da6d581 JS |
999 | unsigned long ptr; |
1000 | unsigned long end; | |
1001 | struct btrfs_extent_item *ei; | |
1002 | u64 flags; | |
1003 | u64 item_size; | |
1004 | ||
1005 | /* | |
1006 | * enumerate all inline refs | |
1007 | */ | |
1008 | leaf = path->nodes[0]; | |
dadcaf78 | 1009 | slot = path->slots[0]; |
8da6d581 | 1010 | |
3212fa14 | 1011 | item_size = btrfs_item_size(leaf, slot); |
8da6d581 JS |
1012 | BUG_ON(item_size < sizeof(*ei)); |
1013 | ||
1014 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
1015 | flags = btrfs_extent_flags(leaf, ei); | |
261c84b6 | 1016 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
8da6d581 JS |
1017 | |
1018 | ptr = (unsigned long)(ei + 1); | |
1019 | end = (unsigned long)ei + item_size; | |
1020 | ||
261c84b6 JB |
1021 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && |
1022 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
8da6d581 | 1023 | struct btrfs_tree_block_info *info; |
8da6d581 JS |
1024 | |
1025 | info = (struct btrfs_tree_block_info *)ptr; | |
1026 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
1027 | ptr += sizeof(struct btrfs_tree_block_info); |
1028 | BUG_ON(ptr > end); | |
261c84b6 JB |
1029 | } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { |
1030 | *info_level = found_key.offset; | |
8da6d581 JS |
1031 | } else { |
1032 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1033 | } | |
1034 | ||
1035 | while (ptr < end) { | |
1036 | struct btrfs_extent_inline_ref *iref; | |
1037 | u64 offset; | |
1038 | int type; | |
1039 | ||
1040 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1041 | type = btrfs_get_extent_inline_ref_type(leaf, iref, |
1042 | BTRFS_REF_TYPE_ANY); | |
1043 | if (type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 1044 | return -EUCLEAN; |
3de28d57 | 1045 | |
8da6d581 JS |
1046 | offset = btrfs_extent_inline_ref_offset(leaf, iref); |
1047 | ||
1048 | switch (type) { | |
1049 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
00142756 JM |
1050 | ret = add_direct_ref(fs_info, preftrees, |
1051 | *info_level + 1, offset, | |
3ec4d323 | 1052 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1053 | break; |
1054 | case BTRFS_SHARED_DATA_REF_KEY: { | |
1055 | struct btrfs_shared_data_ref *sdref; | |
1056 | int count; | |
1057 | ||
1058 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1059 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
86d5f994 | 1060 | |
00142756 | 1061 | ret = add_direct_ref(fs_info, preftrees, 0, offset, |
3ec4d323 | 1062 | bytenr, count, sc, GFP_NOFS); |
8da6d581 JS |
1063 | break; |
1064 | } | |
1065 | case BTRFS_TREE_BLOCK_REF_KEY: | |
00142756 JM |
1066 | ret = add_indirect_ref(fs_info, preftrees, offset, |
1067 | NULL, *info_level + 1, | |
3ec4d323 | 1068 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1069 | break; |
1070 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
1071 | struct btrfs_extent_data_ref *dref; | |
1072 | int count; | |
1073 | u64 root; | |
1074 | ||
1075 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1076 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1077 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1078 | dref); | |
1079 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1080 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1081 | |
a0a5472a | 1082 | if (sc && key.objectid != sc->inum && |
4fc7b572 | 1083 | !sc->have_delayed_delete_refs) { |
dc046b10 JB |
1084 | ret = BACKREF_FOUND_SHARED; |
1085 | break; | |
1086 | } | |
1087 | ||
8da6d581 | 1088 | root = btrfs_extent_data_ref_root(leaf, dref); |
86d5f994 | 1089 | |
00142756 JM |
1090 | ret = add_indirect_ref(fs_info, preftrees, root, |
1091 | &key, 0, bytenr, count, | |
3ec4d323 | 1092 | sc, GFP_NOFS); |
4fc7b572 | 1093 | |
8da6d581 JS |
1094 | break; |
1095 | } | |
1096 | default: | |
1097 | WARN_ON(1); | |
1098 | } | |
1149ab6b WS |
1099 | if (ret) |
1100 | return ret; | |
8da6d581 JS |
1101 | ptr += btrfs_extent_inline_ref_size(type); |
1102 | } | |
1103 | ||
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * add all non-inline backrefs for bytenr to the list | |
3ec4d323 EN |
1109 | * |
1110 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 1111 | */ |
98cc4222 | 1112 | static int add_keyed_refs(struct btrfs_root *extent_root, |
e0c476b1 | 1113 | struct btrfs_path *path, u64 bytenr, |
86d5f994 | 1114 | int info_level, struct preftrees *preftrees, |
3ec4d323 | 1115 | struct share_check *sc) |
8da6d581 | 1116 | { |
98cc4222 | 1117 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
8da6d581 JS |
1118 | int ret; |
1119 | int slot; | |
1120 | struct extent_buffer *leaf; | |
1121 | struct btrfs_key key; | |
1122 | ||
1123 | while (1) { | |
1124 | ret = btrfs_next_item(extent_root, path); | |
1125 | if (ret < 0) | |
1126 | break; | |
1127 | if (ret) { | |
1128 | ret = 0; | |
1129 | break; | |
1130 | } | |
1131 | ||
1132 | slot = path->slots[0]; | |
1133 | leaf = path->nodes[0]; | |
1134 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1135 | ||
1136 | if (key.objectid != bytenr) | |
1137 | break; | |
1138 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
1139 | continue; | |
1140 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
1141 | break; | |
1142 | ||
1143 | switch (key.type) { | |
1144 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
86d5f994 | 1145 | /* SHARED DIRECT METADATA backref */ |
00142756 JM |
1146 | ret = add_direct_ref(fs_info, preftrees, |
1147 | info_level + 1, key.offset, | |
3ec4d323 | 1148 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1149 | break; |
1150 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 1151 | /* SHARED DIRECT FULL backref */ |
8da6d581 JS |
1152 | struct btrfs_shared_data_ref *sdref; |
1153 | int count; | |
1154 | ||
1155 | sdref = btrfs_item_ptr(leaf, slot, | |
1156 | struct btrfs_shared_data_ref); | |
1157 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
00142756 JM |
1158 | ret = add_direct_ref(fs_info, preftrees, 0, |
1159 | key.offset, bytenr, count, | |
3ec4d323 | 1160 | sc, GFP_NOFS); |
8da6d581 JS |
1161 | break; |
1162 | } | |
1163 | case BTRFS_TREE_BLOCK_REF_KEY: | |
86d5f994 | 1164 | /* NORMAL INDIRECT METADATA backref */ |
00142756 JM |
1165 | ret = add_indirect_ref(fs_info, preftrees, key.offset, |
1166 | NULL, info_level + 1, bytenr, | |
3ec4d323 | 1167 | 1, NULL, GFP_NOFS); |
8da6d581 JS |
1168 | break; |
1169 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 1170 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 JS |
1171 | struct btrfs_extent_data_ref *dref; |
1172 | int count; | |
1173 | u64 root; | |
1174 | ||
1175 | dref = btrfs_item_ptr(leaf, slot, | |
1176 | struct btrfs_extent_data_ref); | |
1177 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1178 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1179 | dref); | |
1180 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1181 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1182 | |
a0a5472a | 1183 | if (sc && key.objectid != sc->inum && |
4fc7b572 | 1184 | !sc->have_delayed_delete_refs) { |
dc046b10 JB |
1185 | ret = BACKREF_FOUND_SHARED; |
1186 | break; | |
1187 | } | |
1188 | ||
8da6d581 | 1189 | root = btrfs_extent_data_ref_root(leaf, dref); |
00142756 JM |
1190 | ret = add_indirect_ref(fs_info, preftrees, root, |
1191 | &key, 0, bytenr, count, | |
3ec4d323 | 1192 | sc, GFP_NOFS); |
8da6d581 JS |
1193 | break; |
1194 | } | |
1195 | default: | |
1196 | WARN_ON(1); | |
1197 | } | |
1149ab6b WS |
1198 | if (ret) |
1199 | return ret; | |
1200 | ||
8da6d581 JS |
1201 | } |
1202 | ||
1203 | return ret; | |
1204 | } | |
1205 | ||
583f4ac5 FM |
1206 | /* |
1207 | * The caller has joined a transaction or is holding a read lock on the | |
1208 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1209 | * snapshot field changing while updating or checking the cache. | |
1210 | */ | |
1211 | static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, | |
1212 | struct btrfs_root *root, | |
1213 | u64 bytenr, int level, bool *is_shared) | |
1214 | { | |
1215 | struct btrfs_backref_shared_cache_entry *entry; | |
1216 | ||
1217 | if (!ctx->use_path_cache) | |
1218 | return false; | |
1219 | ||
1220 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1221 | return false; | |
1222 | ||
1223 | /* | |
1224 | * Level -1 is used for the data extent, which is not reliable to cache | |
1225 | * because its reference count can increase or decrease without us | |
1226 | * realizing. We cache results only for extent buffers that lead from | |
1227 | * the root node down to the leaf with the file extent item. | |
1228 | */ | |
1229 | ASSERT(level >= 0); | |
1230 | ||
1231 | entry = &ctx->path_cache_entries[level]; | |
1232 | ||
1233 | /* Unused cache entry or being used for some other extent buffer. */ | |
1234 | if (entry->bytenr != bytenr) | |
1235 | return false; | |
1236 | ||
1237 | /* | |
1238 | * We cached a false result, but the last snapshot generation of the | |
1239 | * root changed, so we now have a snapshot. Don't trust the result. | |
1240 | */ | |
1241 | if (!entry->is_shared && | |
1242 | entry->gen != btrfs_root_last_snapshot(&root->root_item)) | |
1243 | return false; | |
1244 | ||
1245 | /* | |
1246 | * If we cached a true result and the last generation used for dropping | |
1247 | * a root changed, we can not trust the result, because the dropped root | |
1248 | * could be a snapshot sharing this extent buffer. | |
1249 | */ | |
1250 | if (entry->is_shared && | |
1251 | entry->gen != btrfs_get_last_root_drop_gen(root->fs_info)) | |
1252 | return false; | |
1253 | ||
1254 | *is_shared = entry->is_shared; | |
1255 | /* | |
1256 | * If the node at this level is shared, than all nodes below are also | |
1257 | * shared. Currently some of the nodes below may be marked as not shared | |
1258 | * because we have just switched from one leaf to another, and switched | |
1259 | * also other nodes above the leaf and below the current level, so mark | |
1260 | * them as shared. | |
1261 | */ | |
1262 | if (*is_shared) { | |
1263 | for (int i = 0; i < level; i++) { | |
1264 | ctx->path_cache_entries[i].is_shared = true; | |
1265 | ctx->path_cache_entries[i].gen = entry->gen; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | return true; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * The caller has joined a transaction or is holding a read lock on the | |
1274 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1275 | * snapshot field changing while updating or checking the cache. | |
1276 | */ | |
1277 | static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, | |
1278 | struct btrfs_root *root, | |
1279 | u64 bytenr, int level, bool is_shared) | |
1280 | { | |
1281 | struct btrfs_backref_shared_cache_entry *entry; | |
1282 | u64 gen; | |
1283 | ||
1284 | if (!ctx->use_path_cache) | |
1285 | return; | |
1286 | ||
1287 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1288 | return; | |
1289 | ||
1290 | /* | |
1291 | * Level -1 is used for the data extent, which is not reliable to cache | |
1292 | * because its reference count can increase or decrease without us | |
1293 | * realizing. We cache results only for extent buffers that lead from | |
1294 | * the root node down to the leaf with the file extent item. | |
1295 | */ | |
1296 | ASSERT(level >= 0); | |
1297 | ||
1298 | if (is_shared) | |
1299 | gen = btrfs_get_last_root_drop_gen(root->fs_info); | |
1300 | else | |
1301 | gen = btrfs_root_last_snapshot(&root->root_item); | |
1302 | ||
1303 | entry = &ctx->path_cache_entries[level]; | |
1304 | entry->bytenr = bytenr; | |
1305 | entry->is_shared = is_shared; | |
1306 | entry->gen = gen; | |
1307 | ||
1308 | /* | |
1309 | * If we found an extent buffer is shared, set the cache result for all | |
1310 | * extent buffers below it to true. As nodes in the path are COWed, | |
1311 | * their sharedness is moved to their children, and if a leaf is COWed, | |
1312 | * then the sharedness of a data extent becomes direct, the refcount of | |
1313 | * data extent is increased in the extent item at the extent tree. | |
1314 | */ | |
1315 | if (is_shared) { | |
1316 | for (int i = 0; i < level; i++) { | |
1317 | entry = &ctx->path_cache_entries[i]; | |
1318 | entry->is_shared = is_shared; | |
1319 | entry->gen = gen; | |
1320 | } | |
1321 | } | |
1322 | } | |
1323 | ||
8da6d581 JS |
1324 | /* |
1325 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
1326 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
1327 | * indirect refs to their parent bytenr. | |
1328 | * When roots are found, they're added to the roots list | |
1329 | * | |
f3a84ccd FM |
1330 | * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and |
1331 | * behave much like trans == NULL case, the difference only lies in it will not | |
21633fc6 QW |
1332 | * commit root. |
1333 | * The special case is for qgroup to search roots in commit_transaction(). | |
1334 | * | |
3ec4d323 EN |
1335 | * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a |
1336 | * shared extent is detected. | |
1337 | * | |
1338 | * Otherwise this returns 0 for success and <0 for an error. | |
1339 | * | |
c995ab3c ZB |
1340 | * If ignore_offset is set to false, only extent refs whose offsets match |
1341 | * extent_item_pos are returned. If true, every extent ref is returned | |
1342 | * and extent_item_pos is ignored. | |
1343 | * | |
8da6d581 JS |
1344 | * FIXME some caching might speed things up |
1345 | */ | |
1346 | static int find_parent_nodes(struct btrfs_trans_handle *trans, | |
1347 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1348 | u64 time_seq, struct ulist *refs, |
dc046b10 | 1349 | struct ulist *roots, const u64 *extent_item_pos, |
c995ab3c | 1350 | struct share_check *sc, bool ignore_offset) |
8da6d581 | 1351 | { |
29cbcf40 | 1352 | struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr); |
8da6d581 JS |
1353 | struct btrfs_key key; |
1354 | struct btrfs_path *path; | |
8da6d581 | 1355 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 1356 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
1357 | int info_level = 0; |
1358 | int ret; | |
e0c476b1 | 1359 | struct prelim_ref *ref; |
86d5f994 | 1360 | struct rb_node *node; |
f05c4746 | 1361 | struct extent_inode_elem *eie = NULL; |
86d5f994 EN |
1362 | struct preftrees preftrees = { |
1363 | .direct = PREFTREE_INIT, | |
1364 | .indirect = PREFTREE_INIT, | |
1365 | .indirect_missing_keys = PREFTREE_INIT | |
1366 | }; | |
8da6d581 | 1367 | |
56f5c199 FM |
1368 | /* Roots ulist is not needed when using a sharedness check context. */ |
1369 | if (sc) | |
1370 | ASSERT(roots == NULL); | |
1371 | ||
8da6d581 | 1372 | key.objectid = bytenr; |
8da6d581 | 1373 | key.offset = (u64)-1; |
261c84b6 JB |
1374 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1375 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1376 | else | |
1377 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
8da6d581 JS |
1378 | |
1379 | path = btrfs_alloc_path(); | |
1380 | if (!path) | |
1381 | return -ENOMEM; | |
e84752d4 | 1382 | if (!trans) { |
da61d31a | 1383 | path->search_commit_root = 1; |
e84752d4 WS |
1384 | path->skip_locking = 1; |
1385 | } | |
8da6d581 | 1386 | |
f3a84ccd | 1387 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
1388 | path->skip_locking = 1; |
1389 | ||
8da6d581 | 1390 | again: |
d3b01064 LZ |
1391 | head = NULL; |
1392 | ||
98cc4222 | 1393 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8da6d581 JS |
1394 | if (ret < 0) |
1395 | goto out; | |
fcba0120 JB |
1396 | if (ret == 0) { |
1397 | /* This shouldn't happen, indicates a bug or fs corruption. */ | |
1398 | ASSERT(ret != 0); | |
1399 | ret = -EUCLEAN; | |
1400 | goto out; | |
1401 | } | |
8da6d581 | 1402 | |
21633fc6 | 1403 | if (trans && likely(trans->type != __TRANS_DUMMY) && |
f3a84ccd | 1404 | time_seq != BTRFS_SEQ_LAST) { |
7a3ae2f8 | 1405 | /* |
9665ebd5 JB |
1406 | * We have a specific time_seq we care about and trans which |
1407 | * means we have the path lock, we need to grab the ref head and | |
1408 | * lock it so we have a consistent view of the refs at the given | |
1409 | * time. | |
7a3ae2f8 JS |
1410 | */ |
1411 | delayed_refs = &trans->transaction->delayed_refs; | |
1412 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 1413 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
7a3ae2f8 JS |
1414 | if (head) { |
1415 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 1416 | refcount_inc(&head->refs); |
7a3ae2f8 JS |
1417 | spin_unlock(&delayed_refs->lock); |
1418 | ||
1419 | btrfs_release_path(path); | |
1420 | ||
1421 | /* | |
1422 | * Mutex was contended, block until it's | |
1423 | * released and try again | |
1424 | */ | |
1425 | mutex_lock(&head->mutex); | |
1426 | mutex_unlock(&head->mutex); | |
d278850e | 1427 | btrfs_put_delayed_ref_head(head); |
7a3ae2f8 JS |
1428 | goto again; |
1429 | } | |
d7df2c79 | 1430 | spin_unlock(&delayed_refs->lock); |
00142756 | 1431 | ret = add_delayed_refs(fs_info, head, time_seq, |
b25b0b87 | 1432 | &preftrees, sc); |
155725c9 | 1433 | mutex_unlock(&head->mutex); |
d7df2c79 | 1434 | if (ret) |
7a3ae2f8 | 1435 | goto out; |
d7df2c79 JB |
1436 | } else { |
1437 | spin_unlock(&delayed_refs->lock); | |
d3b01064 | 1438 | } |
8da6d581 | 1439 | } |
8da6d581 JS |
1440 | |
1441 | if (path->slots[0]) { | |
1442 | struct extent_buffer *leaf; | |
1443 | int slot; | |
1444 | ||
dadcaf78 | 1445 | path->slots[0]--; |
8da6d581 | 1446 | leaf = path->nodes[0]; |
dadcaf78 | 1447 | slot = path->slots[0]; |
8da6d581 JS |
1448 | btrfs_item_key_to_cpu(leaf, &key, slot); |
1449 | if (key.objectid == bytenr && | |
261c84b6 JB |
1450 | (key.type == BTRFS_EXTENT_ITEM_KEY || |
1451 | key.type == BTRFS_METADATA_ITEM_KEY)) { | |
00142756 | 1452 | ret = add_inline_refs(fs_info, path, bytenr, |
b25b0b87 | 1453 | &info_level, &preftrees, sc); |
8da6d581 JS |
1454 | if (ret) |
1455 | goto out; | |
98cc4222 | 1456 | ret = add_keyed_refs(root, path, bytenr, info_level, |
3ec4d323 | 1457 | &preftrees, sc); |
8da6d581 JS |
1458 | if (ret) |
1459 | goto out; | |
1460 | } | |
1461 | } | |
8da6d581 | 1462 | |
56f5c199 FM |
1463 | /* |
1464 | * If we have a share context and we reached here, it means the extent | |
1465 | * is not directly shared (no multiple reference items for it), | |
1466 | * otherwise we would have exited earlier with a return value of | |
1467 | * BACKREF_FOUND_SHARED after processing delayed references or while | |
1468 | * processing inline or keyed references from the extent tree. | |
1469 | * The extent may however be indirectly shared through shared subtrees | |
1470 | * as a result from creating snapshots, so we determine below what is | |
1471 | * its parent node, in case we are dealing with a metadata extent, or | |
1472 | * what's the leaf (or leaves), from a fs tree, that has a file extent | |
1473 | * item pointing to it in case we are dealing with a data extent. | |
1474 | */ | |
1475 | ASSERT(extent_is_shared(sc) == 0); | |
1476 | ||
877c1476 FM |
1477 | /* |
1478 | * If we are here for a data extent and we have a share_check structure | |
1479 | * it means the data extent is not directly shared (does not have | |
1480 | * multiple reference items), so we have to check if a path in the fs | |
1481 | * tree (going from the root node down to the leaf that has the file | |
1482 | * extent item pointing to the data extent) is shared, that is, if any | |
1483 | * of the extent buffers in the path is referenced by other trees. | |
1484 | */ | |
1485 | if (sc && bytenr == sc->data_bytenr) { | |
6976201f FM |
1486 | /* |
1487 | * If our data extent is from a generation more recent than the | |
1488 | * last generation used to snapshot the root, then we know that | |
1489 | * it can not be shared through subtrees, so we can skip | |
1490 | * resolving indirect references, there's no point in | |
1491 | * determining the extent buffers for the path from the fs tree | |
1492 | * root node down to the leaf that has the file extent item that | |
1493 | * points to the data extent. | |
1494 | */ | |
1495 | if (sc->data_extent_gen > | |
1496 | btrfs_root_last_snapshot(&sc->root->root_item)) { | |
1497 | ret = BACKREF_FOUND_NOT_SHARED; | |
1498 | goto out; | |
1499 | } | |
1500 | ||
877c1476 FM |
1501 | /* |
1502 | * If we are only determining if a data extent is shared or not | |
1503 | * and the corresponding file extent item is located in the same | |
1504 | * leaf as the previous file extent item, we can skip resolving | |
1505 | * indirect references for a data extent, since the fs tree path | |
1506 | * is the same (same leaf, so same path). We skip as long as the | |
1507 | * cached result for the leaf is valid and only if there's only | |
1508 | * one file extent item pointing to the data extent, because in | |
1509 | * the case of multiple file extent items, they may be located | |
1510 | * in different leaves and therefore we have multiple paths. | |
1511 | */ | |
1512 | if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr && | |
1513 | sc->self_ref_count == 1) { | |
1514 | bool cached; | |
1515 | bool is_shared; | |
1516 | ||
1517 | cached = lookup_backref_shared_cache(sc->ctx, sc->root, | |
1518 | sc->ctx->curr_leaf_bytenr, | |
1519 | 0, &is_shared); | |
1520 | if (cached) { | |
1521 | if (is_shared) | |
1522 | ret = BACKREF_FOUND_SHARED; | |
1523 | else | |
1524 | ret = BACKREF_FOUND_NOT_SHARED; | |
1525 | goto out; | |
1526 | } | |
1527 | } | |
1528 | } | |
1529 | ||
86d5f994 | 1530 | btrfs_release_path(path); |
8da6d581 | 1531 | |
38e3eebf | 1532 | ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0); |
d5c88b73 JS |
1533 | if (ret) |
1534 | goto out; | |
1535 | ||
ecf160b4 | 1536 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root)); |
8da6d581 | 1537 | |
86d5f994 | 1538 | ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees, |
b25b0b87 | 1539 | extent_item_pos, sc, ignore_offset); |
8da6d581 JS |
1540 | if (ret) |
1541 | goto out; | |
1542 | ||
ecf160b4 | 1543 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root)); |
8da6d581 | 1544 | |
86d5f994 EN |
1545 | /* |
1546 | * This walks the tree of merged and resolved refs. Tree blocks are | |
1547 | * read in as needed. Unique entries are added to the ulist, and | |
1548 | * the list of found roots is updated. | |
1549 | * | |
1550 | * We release the entire tree in one go before returning. | |
1551 | */ | |
ecf160b4 | 1552 | node = rb_first_cached(&preftrees.direct.root); |
86d5f994 EN |
1553 | while (node) { |
1554 | ref = rb_entry(node, struct prelim_ref, rbnode); | |
1555 | node = rb_next(&ref->rbnode); | |
c8195a7b ZB |
1556 | /* |
1557 | * ref->count < 0 can happen here if there are delayed | |
1558 | * refs with a node->action of BTRFS_DROP_DELAYED_REF. | |
1559 | * prelim_ref_insert() relies on this when merging | |
1560 | * identical refs to keep the overall count correct. | |
1561 | * prelim_ref_insert() will merge only those refs | |
1562 | * which compare identically. Any refs having | |
1563 | * e.g. different offsets would not be merged, | |
1564 | * and would retain their original ref->count < 0. | |
1565 | */ | |
98cfee21 | 1566 | if (roots && ref->count && ref->root_id && ref->parent == 0) { |
8da6d581 JS |
1567 | /* no parent == root of tree */ |
1568 | ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); | |
f1723939 WS |
1569 | if (ret < 0) |
1570 | goto out; | |
8da6d581 JS |
1571 | } |
1572 | if (ref->count && ref->parent) { | |
8a56457f JB |
1573 | if (extent_item_pos && !ref->inode_list && |
1574 | ref->level == 0) { | |
976b1908 | 1575 | struct extent_buffer *eb; |
707e8a07 | 1576 | |
581c1760 | 1577 | eb = read_tree_block(fs_info, ref->parent, 0, |
1b7ec85e | 1578 | 0, ref->level, NULL); |
64c043de LB |
1579 | if (IS_ERR(eb)) { |
1580 | ret = PTR_ERR(eb); | |
1581 | goto out; | |
4eb150d6 QW |
1582 | } |
1583 | if (!extent_buffer_uptodate(eb)) { | |
416bc658 | 1584 | free_extent_buffer(eb); |
c16c2e2e WS |
1585 | ret = -EIO; |
1586 | goto out; | |
416bc658 | 1587 | } |
38e3eebf | 1588 | |
ac5887c8 | 1589 | if (!path->skip_locking) |
38e3eebf | 1590 | btrfs_tree_read_lock(eb); |
976b1908 | 1591 | ret = find_extent_in_eb(eb, bytenr, |
c995ab3c | 1592 | *extent_item_pos, &eie, ignore_offset); |
38e3eebf | 1593 | if (!path->skip_locking) |
ac5887c8 | 1594 | btrfs_tree_read_unlock(eb); |
976b1908 | 1595 | free_extent_buffer(eb); |
f5929cd8 FDBM |
1596 | if (ret < 0) |
1597 | goto out; | |
1598 | ref->inode_list = eie; | |
92876eec FM |
1599 | /* |
1600 | * We transferred the list ownership to the ref, | |
1601 | * so set to NULL to avoid a double free in case | |
1602 | * an error happens after this. | |
1603 | */ | |
1604 | eie = NULL; | |
976b1908 | 1605 | } |
4eb1f66d TI |
1606 | ret = ulist_add_merge_ptr(refs, ref->parent, |
1607 | ref->inode_list, | |
1608 | (void **)&eie, GFP_NOFS); | |
f1723939 WS |
1609 | if (ret < 0) |
1610 | goto out; | |
3301958b JS |
1611 | if (!ret && extent_item_pos) { |
1612 | /* | |
9f05c09d JB |
1613 | * We've recorded that parent, so we must extend |
1614 | * its inode list here. | |
1615 | * | |
1616 | * However if there was corruption we may not | |
1617 | * have found an eie, return an error in this | |
1618 | * case. | |
3301958b | 1619 | */ |
9f05c09d JB |
1620 | ASSERT(eie); |
1621 | if (!eie) { | |
1622 | ret = -EUCLEAN; | |
1623 | goto out; | |
1624 | } | |
3301958b JS |
1625 | while (eie->next) |
1626 | eie = eie->next; | |
1627 | eie->next = ref->inode_list; | |
1628 | } | |
f05c4746 | 1629 | eie = NULL; |
92876eec FM |
1630 | /* |
1631 | * We have transferred the inode list ownership from | |
1632 | * this ref to the ref we added to the 'refs' ulist. | |
1633 | * So set this ref's inode list to NULL to avoid | |
1634 | * use-after-free when our caller uses it or double | |
1635 | * frees in case an error happens before we return. | |
1636 | */ | |
1637 | ref->inode_list = NULL; | |
8da6d581 | 1638 | } |
9dd14fd6 | 1639 | cond_resched(); |
8da6d581 JS |
1640 | } |
1641 | ||
1642 | out: | |
8da6d581 | 1643 | btrfs_free_path(path); |
86d5f994 EN |
1644 | |
1645 | prelim_release(&preftrees.direct); | |
1646 | prelim_release(&preftrees.indirect); | |
1647 | prelim_release(&preftrees.indirect_missing_keys); | |
1648 | ||
f05c4746 WS |
1649 | if (ret < 0) |
1650 | free_inode_elem_list(eie); | |
8da6d581 JS |
1651 | return ret; |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * Finds all leafs with a reference to the specified combination of bytenr and | |
1656 | * offset. key_list_head will point to a list of corresponding keys (caller must | |
1657 | * free each list element). The leafs will be stored in the leafs ulist, which | |
1658 | * must be freed with ulist_free. | |
1659 | * | |
1660 | * returns 0 on success, <0 on error | |
1661 | */ | |
19b546d7 QW |
1662 | int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, |
1663 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
1664 | u64 time_seq, struct ulist **leafs, | |
1665 | const u64 *extent_item_pos, bool ignore_offset) | |
8da6d581 | 1666 | { |
8da6d581 JS |
1667 | int ret; |
1668 | ||
8da6d581 | 1669 | *leafs = ulist_alloc(GFP_NOFS); |
98cfee21 | 1670 | if (!*leafs) |
8da6d581 | 1671 | return -ENOMEM; |
8da6d581 | 1672 | |
afce772e | 1673 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1674 | *leafs, NULL, extent_item_pos, NULL, ignore_offset); |
8da6d581 | 1675 | if (ret < 0 && ret != -ENOENT) { |
976b1908 | 1676 | free_leaf_list(*leafs); |
8da6d581 JS |
1677 | return ret; |
1678 | } | |
1679 | ||
1680 | return 0; | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * walk all backrefs for a given extent to find all roots that reference this | |
1685 | * extent. Walking a backref means finding all extents that reference this | |
1686 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1687 | * recursive process, but here it is implemented in an iterative fashion: We | |
1688 | * find all referencing extents for the extent in question and put them on a | |
1689 | * list. In turn, we find all referencing extents for those, further appending | |
1690 | * to the list. The way we iterate the list allows adding more elements after | |
1691 | * the current while iterating. The process stops when we reach the end of the | |
1692 | * list. Found roots are added to the roots list. | |
1693 | * | |
1694 | * returns 0 on success, < 0 on error. | |
1695 | */ | |
e0c476b1 JM |
1696 | static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans, |
1697 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c ZB |
1698 | u64 time_seq, struct ulist **roots, |
1699 | bool ignore_offset) | |
8da6d581 JS |
1700 | { |
1701 | struct ulist *tmp; | |
1702 | struct ulist_node *node = NULL; | |
cd1b413c | 1703 | struct ulist_iterator uiter; |
8da6d581 JS |
1704 | int ret; |
1705 | ||
1706 | tmp = ulist_alloc(GFP_NOFS); | |
1707 | if (!tmp) | |
1708 | return -ENOMEM; | |
1709 | *roots = ulist_alloc(GFP_NOFS); | |
1710 | if (!*roots) { | |
1711 | ulist_free(tmp); | |
1712 | return -ENOMEM; | |
1713 | } | |
1714 | ||
cd1b413c | 1715 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1716 | while (1) { |
afce772e | 1717 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1718 | tmp, *roots, NULL, NULL, ignore_offset); |
8da6d581 JS |
1719 | if (ret < 0 && ret != -ENOENT) { |
1720 | ulist_free(tmp); | |
1721 | ulist_free(*roots); | |
580c079b | 1722 | *roots = NULL; |
8da6d581 JS |
1723 | return ret; |
1724 | } | |
cd1b413c | 1725 | node = ulist_next(tmp, &uiter); |
8da6d581 JS |
1726 | if (!node) |
1727 | break; | |
1728 | bytenr = node->val; | |
bca1a290 | 1729 | cond_resched(); |
8da6d581 JS |
1730 | } |
1731 | ||
1732 | ulist_free(tmp); | |
1733 | return 0; | |
1734 | } | |
1735 | ||
9e351cc8 JB |
1736 | int btrfs_find_all_roots(struct btrfs_trans_handle *trans, |
1737 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c | 1738 | u64 time_seq, struct ulist **roots, |
c7bcbb21 | 1739 | bool skip_commit_root_sem) |
9e351cc8 JB |
1740 | { |
1741 | int ret; | |
1742 | ||
8949b9a1 | 1743 | if (!trans && !skip_commit_root_sem) |
9e351cc8 | 1744 | down_read(&fs_info->commit_root_sem); |
e0c476b1 | 1745 | ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr, |
c7bcbb21 | 1746 | time_seq, roots, false); |
8949b9a1 | 1747 | if (!trans && !skip_commit_root_sem) |
9e351cc8 JB |
1748 | up_read(&fs_info->commit_root_sem); |
1749 | return ret; | |
1750 | } | |
1751 | ||
84a7949d FM |
1752 | struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void) |
1753 | { | |
1754 | struct btrfs_backref_share_check_ctx *ctx; | |
1755 | ||
1756 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1757 | if (!ctx) | |
1758 | return NULL; | |
1759 | ||
1760 | ulist_init(&ctx->refs); | |
84a7949d FM |
1761 | |
1762 | return ctx; | |
1763 | } | |
1764 | ||
1765 | void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx) | |
1766 | { | |
1767 | if (!ctx) | |
1768 | return; | |
1769 | ||
1770 | ulist_release(&ctx->refs); | |
84a7949d FM |
1771 | kfree(ctx); |
1772 | } | |
1773 | ||
8eedadda FM |
1774 | /* |
1775 | * Check if a data extent is shared or not. | |
6e353e3b | 1776 | * |
ceb707da | 1777 | * @inode: The inode whose extent we are checking. |
b8f164e3 FM |
1778 | * @bytenr: Logical bytenr of the extent we are checking. |
1779 | * @extent_gen: Generation of the extent (file extent item) or 0 if it is | |
1780 | * not known. | |
61dbb952 | 1781 | * @ctx: A backref sharedness check context. |
2c2ed5aa | 1782 | * |
8eedadda | 1783 | * btrfs_is_data_extent_shared uses the backref walking code but will short |
2c2ed5aa MF |
1784 | * circuit as soon as it finds a root or inode that doesn't match the |
1785 | * one passed in. This provides a significant performance benefit for | |
1786 | * callers (such as fiemap) which want to know whether the extent is | |
1787 | * shared but do not need a ref count. | |
1788 | * | |
03628cdb FM |
1789 | * This attempts to attach to the running transaction in order to account for |
1790 | * delayed refs, but continues on even when no running transaction exists. | |
bb739cf0 | 1791 | * |
2c2ed5aa MF |
1792 | * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. |
1793 | */ | |
ceb707da | 1794 | int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr, |
b8f164e3 | 1795 | u64 extent_gen, |
61dbb952 | 1796 | struct btrfs_backref_share_check_ctx *ctx) |
dc046b10 | 1797 | { |
ceb707da | 1798 | struct btrfs_root *root = inode->root; |
bb739cf0 EN |
1799 | struct btrfs_fs_info *fs_info = root->fs_info; |
1800 | struct btrfs_trans_handle *trans; | |
dc046b10 JB |
1801 | struct ulist_iterator uiter; |
1802 | struct ulist_node *node; | |
f3a84ccd | 1803 | struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem); |
dc046b10 | 1804 | int ret = 0; |
3ec4d323 | 1805 | struct share_check shared = { |
877c1476 FM |
1806 | .ctx = ctx, |
1807 | .root = root, | |
ceb707da | 1808 | .inum = btrfs_ino(inode), |
73e339e6 | 1809 | .data_bytenr = bytenr, |
6976201f | 1810 | .data_extent_gen = extent_gen, |
3ec4d323 | 1811 | .share_count = 0, |
73e339e6 | 1812 | .self_ref_count = 0, |
4fc7b572 | 1813 | .have_delayed_delete_refs = false, |
3ec4d323 | 1814 | }; |
12a824dc | 1815 | int level; |
dc046b10 | 1816 | |
73e339e6 FM |
1817 | for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) { |
1818 | if (ctx->prev_extents_cache[i].bytenr == bytenr) | |
1819 | return ctx->prev_extents_cache[i].is_shared; | |
1820 | } | |
1821 | ||
84a7949d | 1822 | ulist_init(&ctx->refs); |
dc046b10 | 1823 | |
a6d155d2 | 1824 | trans = btrfs_join_transaction_nostart(root); |
bb739cf0 | 1825 | if (IS_ERR(trans)) { |
03628cdb FM |
1826 | if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) { |
1827 | ret = PTR_ERR(trans); | |
1828 | goto out; | |
1829 | } | |
bb739cf0 | 1830 | trans = NULL; |
dc046b10 | 1831 | down_read(&fs_info->commit_root_sem); |
bb739cf0 EN |
1832 | } else { |
1833 | btrfs_get_tree_mod_seq(fs_info, &elem); | |
1834 | } | |
1835 | ||
12a824dc FM |
1836 | /* -1 means we are in the bytenr of the data extent. */ |
1837 | level = -1; | |
dc046b10 | 1838 | ULIST_ITER_INIT(&uiter); |
61dbb952 | 1839 | ctx->use_path_cache = true; |
dc046b10 | 1840 | while (1) { |
12a824dc FM |
1841 | bool is_shared; |
1842 | bool cached; | |
1843 | ||
84a7949d | 1844 | ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, &ctx->refs, |
b6296858 | 1845 | NULL, NULL, &shared, false); |
877c1476 FM |
1846 | if (ret == BACKREF_FOUND_SHARED || |
1847 | ret == BACKREF_FOUND_NOT_SHARED) { | |
1848 | /* If shared must return 1, otherwise return 0. */ | |
1849 | ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0; | |
12a824dc | 1850 | if (level >= 0) |
61dbb952 | 1851 | store_backref_shared_cache(ctx, root, bytenr, |
877c1476 | 1852 | level, ret == 1); |
dc046b10 JB |
1853 | break; |
1854 | } | |
1855 | if (ret < 0 && ret != -ENOENT) | |
1856 | break; | |
2c2ed5aa | 1857 | ret = 0; |
b8f164e3 | 1858 | |
63c84b46 FM |
1859 | /* |
1860 | * If our data extent was not directly shared (without multiple | |
1861 | * reference items), than it might have a single reference item | |
1862 | * with a count > 1 for the same offset, which means there are 2 | |
1863 | * (or more) file extent items that point to the data extent - | |
1864 | * this happens when a file extent item needs to be split and | |
1865 | * then one item gets moved to another leaf due to a b+tree leaf | |
1866 | * split when inserting some item. In this case the file extent | |
1867 | * items may be located in different leaves and therefore some | |
1868 | * of the leaves may be referenced through shared subtrees while | |
1869 | * others are not. Since our extent buffer cache only works for | |
1870 | * a single path (by far the most common case and simpler to | |
1871 | * deal with), we can not use it if we have multiple leaves | |
1872 | * (which implies multiple paths). | |
1873 | */ | |
84a7949d | 1874 | if (level == -1 && ctx->refs.nnodes > 1) |
61dbb952 | 1875 | ctx->use_path_cache = false; |
63c84b46 | 1876 | |
12a824dc | 1877 | if (level >= 0) |
61dbb952 | 1878 | store_backref_shared_cache(ctx, root, bytenr, |
12a824dc | 1879 | level, false); |
84a7949d | 1880 | node = ulist_next(&ctx->refs, &uiter); |
dc046b10 JB |
1881 | if (!node) |
1882 | break; | |
1883 | bytenr = node->val; | |
12a824dc | 1884 | level++; |
61dbb952 | 1885 | cached = lookup_backref_shared_cache(ctx, root, bytenr, level, |
12a824dc FM |
1886 | &is_shared); |
1887 | if (cached) { | |
1888 | ret = (is_shared ? 1 : 0); | |
1889 | break; | |
1890 | } | |
18bf591b | 1891 | shared.share_count = 0; |
4fc7b572 | 1892 | shared.have_delayed_delete_refs = false; |
dc046b10 JB |
1893 | cond_resched(); |
1894 | } | |
bb739cf0 | 1895 | |
73e339e6 FM |
1896 | /* |
1897 | * Cache the sharedness result for the data extent if we know our inode | |
1898 | * has more than 1 file extent item that refers to the data extent. | |
1899 | */ | |
1900 | if (ret >= 0 && shared.self_ref_count > 1) { | |
1901 | int slot = ctx->prev_extents_cache_slot; | |
1902 | ||
1903 | ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr; | |
1904 | ctx->prev_extents_cache[slot].is_shared = (ret == 1); | |
1905 | ||
1906 | slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; | |
1907 | ctx->prev_extents_cache_slot = slot; | |
1908 | } | |
1909 | ||
bb739cf0 | 1910 | if (trans) { |
dc046b10 | 1911 | btrfs_put_tree_mod_seq(fs_info, &elem); |
bb739cf0 EN |
1912 | btrfs_end_transaction(trans); |
1913 | } else { | |
dc046b10 | 1914 | up_read(&fs_info->commit_root_sem); |
bb739cf0 | 1915 | } |
03628cdb | 1916 | out: |
84a7949d | 1917 | ulist_release(&ctx->refs); |
877c1476 FM |
1918 | ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr; |
1919 | ||
dc046b10 JB |
1920 | return ret; |
1921 | } | |
1922 | ||
f186373f MF |
1923 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
1924 | u64 start_off, struct btrfs_path *path, | |
1925 | struct btrfs_inode_extref **ret_extref, | |
1926 | u64 *found_off) | |
1927 | { | |
1928 | int ret, slot; | |
1929 | struct btrfs_key key; | |
1930 | struct btrfs_key found_key; | |
1931 | struct btrfs_inode_extref *extref; | |
73980bec | 1932 | const struct extent_buffer *leaf; |
f186373f MF |
1933 | unsigned long ptr; |
1934 | ||
1935 | key.objectid = inode_objectid; | |
962a298f | 1936 | key.type = BTRFS_INODE_EXTREF_KEY; |
f186373f MF |
1937 | key.offset = start_off; |
1938 | ||
1939 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1940 | if (ret < 0) | |
1941 | return ret; | |
1942 | ||
1943 | while (1) { | |
1944 | leaf = path->nodes[0]; | |
1945 | slot = path->slots[0]; | |
1946 | if (slot >= btrfs_header_nritems(leaf)) { | |
1947 | /* | |
1948 | * If the item at offset is not found, | |
1949 | * btrfs_search_slot will point us to the slot | |
1950 | * where it should be inserted. In our case | |
1951 | * that will be the slot directly before the | |
1952 | * next INODE_REF_KEY_V2 item. In the case | |
1953 | * that we're pointing to the last slot in a | |
1954 | * leaf, we must move one leaf over. | |
1955 | */ | |
1956 | ret = btrfs_next_leaf(root, path); | |
1957 | if (ret) { | |
1958 | if (ret >= 1) | |
1959 | ret = -ENOENT; | |
1960 | break; | |
1961 | } | |
1962 | continue; | |
1963 | } | |
1964 | ||
1965 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1966 | ||
1967 | /* | |
1968 | * Check that we're still looking at an extended ref key for | |
1969 | * this particular objectid. If we have different | |
1970 | * objectid or type then there are no more to be found | |
1971 | * in the tree and we can exit. | |
1972 | */ | |
1973 | ret = -ENOENT; | |
1974 | if (found_key.objectid != inode_objectid) | |
1975 | break; | |
962a298f | 1976 | if (found_key.type != BTRFS_INODE_EXTREF_KEY) |
f186373f MF |
1977 | break; |
1978 | ||
1979 | ret = 0; | |
1980 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1981 | extref = (struct btrfs_inode_extref *)ptr; | |
1982 | *ret_extref = extref; | |
1983 | if (found_off) | |
1984 | *found_off = found_key.offset; | |
1985 | break; | |
1986 | } | |
1987 | ||
1988 | return ret; | |
1989 | } | |
1990 | ||
48a3b636 ES |
1991 | /* |
1992 | * this iterates to turn a name (from iref/extref) into a full filesystem path. | |
1993 | * Elements of the path are separated by '/' and the path is guaranteed to be | |
1994 | * 0-terminated. the path is only given within the current file system. | |
1995 | * Therefore, it never starts with a '/'. the caller is responsible to provide | |
1996 | * "size" bytes in "dest". the dest buffer will be filled backwards. finally, | |
1997 | * the start point of the resulting string is returned. this pointer is within | |
1998 | * dest, normally. | |
1999 | * in case the path buffer would overflow, the pointer is decremented further | |
2000 | * as if output was written to the buffer, though no more output is actually | |
2001 | * generated. that way, the caller can determine how much space would be | |
2002 | * required for the path to fit into the buffer. in that case, the returned | |
2003 | * value will be smaller than dest. callers must check this! | |
2004 | */ | |
96b5bd77 JS |
2005 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
2006 | u32 name_len, unsigned long name_off, | |
2007 | struct extent_buffer *eb_in, u64 parent, | |
2008 | char *dest, u32 size) | |
a542ad1b | 2009 | { |
a542ad1b JS |
2010 | int slot; |
2011 | u64 next_inum; | |
2012 | int ret; | |
661bec6b | 2013 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
2014 | struct extent_buffer *eb = eb_in; |
2015 | struct btrfs_key found_key; | |
d24bec3a | 2016 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
2017 | |
2018 | if (bytes_left >= 0) | |
2019 | dest[bytes_left] = '\0'; | |
2020 | ||
2021 | while (1) { | |
d24bec3a | 2022 | bytes_left -= name_len; |
a542ad1b JS |
2023 | if (bytes_left >= 0) |
2024 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 2025 | name_off, name_len); |
b916a59a | 2026 | if (eb != eb_in) { |
0c0fe3b0 | 2027 | if (!path->skip_locking) |
ac5887c8 | 2028 | btrfs_tree_read_unlock(eb); |
a542ad1b | 2029 | free_extent_buffer(eb); |
b916a59a | 2030 | } |
c234a24d DS |
2031 | ret = btrfs_find_item(fs_root, path, parent, 0, |
2032 | BTRFS_INODE_REF_KEY, &found_key); | |
8f24b496 JS |
2033 | if (ret > 0) |
2034 | ret = -ENOENT; | |
a542ad1b JS |
2035 | if (ret) |
2036 | break; | |
d24bec3a | 2037 | |
a542ad1b JS |
2038 | next_inum = found_key.offset; |
2039 | ||
2040 | /* regular exit ahead */ | |
2041 | if (parent == next_inum) | |
2042 | break; | |
2043 | ||
2044 | slot = path->slots[0]; | |
2045 | eb = path->nodes[0]; | |
2046 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 2047 | if (eb != eb_in) { |
0c0fe3b0 FM |
2048 | path->nodes[0] = NULL; |
2049 | path->locks[0] = 0; | |
b916a59a | 2050 | } |
a542ad1b | 2051 | btrfs_release_path(path); |
a542ad1b | 2052 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
2053 | |
2054 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
2055 | name_off = (unsigned long)(iref + 1); | |
2056 | ||
a542ad1b JS |
2057 | parent = next_inum; |
2058 | --bytes_left; | |
2059 | if (bytes_left >= 0) | |
2060 | dest[bytes_left] = '/'; | |
2061 | } | |
2062 | ||
2063 | btrfs_release_path(path); | |
2064 | ||
2065 | if (ret) | |
2066 | return ERR_PTR(ret); | |
2067 | ||
2068 | return dest + bytes_left; | |
2069 | } | |
2070 | ||
2071 | /* | |
2072 | * this makes the path point to (logical EXTENT_ITEM *) | |
2073 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
2074 | * tree blocks and <0 on error. | |
2075 | */ | |
2076 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
2077 | struct btrfs_path *path, struct btrfs_key *found_key, |
2078 | u64 *flags_ret) | |
a542ad1b | 2079 | { |
29cbcf40 | 2080 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical); |
a542ad1b JS |
2081 | int ret; |
2082 | u64 flags; | |
261c84b6 | 2083 | u64 size = 0; |
a542ad1b | 2084 | u32 item_size; |
73980bec | 2085 | const struct extent_buffer *eb; |
a542ad1b JS |
2086 | struct btrfs_extent_item *ei; |
2087 | struct btrfs_key key; | |
2088 | ||
261c84b6 JB |
2089 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
2090 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2091 | else | |
2092 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a542ad1b JS |
2093 | key.objectid = logical; |
2094 | key.offset = (u64)-1; | |
2095 | ||
29cbcf40 | 2096 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a542ad1b JS |
2097 | if (ret < 0) |
2098 | return ret; | |
a542ad1b | 2099 | |
29cbcf40 | 2100 | ret = btrfs_previous_extent_item(extent_root, path, 0); |
850a8cdf WS |
2101 | if (ret) { |
2102 | if (ret > 0) | |
2103 | ret = -ENOENT; | |
2104 | return ret; | |
580f0a67 | 2105 | } |
850a8cdf | 2106 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); |
261c84b6 | 2107 | if (found_key->type == BTRFS_METADATA_ITEM_KEY) |
da17066c | 2108 | size = fs_info->nodesize; |
261c84b6 JB |
2109 | else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) |
2110 | size = found_key->offset; | |
2111 | ||
580f0a67 | 2112 | if (found_key->objectid > logical || |
261c84b6 | 2113 | found_key->objectid + size <= logical) { |
ab8d0fc4 JM |
2114 | btrfs_debug(fs_info, |
2115 | "logical %llu is not within any extent", logical); | |
a542ad1b | 2116 | return -ENOENT; |
4692cf58 | 2117 | } |
a542ad1b JS |
2118 | |
2119 | eb = path->nodes[0]; | |
3212fa14 | 2120 | item_size = btrfs_item_size(eb, path->slots[0]); |
a542ad1b JS |
2121 | BUG_ON(item_size < sizeof(*ei)); |
2122 | ||
2123 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
2124 | flags = btrfs_extent_flags(eb, ei); | |
2125 | ||
ab8d0fc4 JM |
2126 | btrfs_debug(fs_info, |
2127 | "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u", | |
c1c9ff7c GU |
2128 | logical, logical - found_key->objectid, found_key->objectid, |
2129 | found_key->offset, flags, item_size); | |
69917e43 LB |
2130 | |
2131 | WARN_ON(!flags_ret); | |
2132 | if (flags_ret) { | |
2133 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
2134 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
2135 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
2136 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
2137 | else | |
290342f6 | 2138 | BUG(); |
69917e43 LB |
2139 | return 0; |
2140 | } | |
a542ad1b JS |
2141 | |
2142 | return -EIO; | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
2147 | * for the first call and may be modified. it is used to track state. | |
2148 | * if more refs exist, 0 is returned and the next call to | |
e0c476b1 | 2149 | * get_extent_inline_ref must pass the modified ptr parameter to get the |
a542ad1b JS |
2150 | * next ref. after the last ref was processed, 1 is returned. |
2151 | * returns <0 on error | |
2152 | */ | |
e0c476b1 JM |
2153 | static int get_extent_inline_ref(unsigned long *ptr, |
2154 | const struct extent_buffer *eb, | |
2155 | const struct btrfs_key *key, | |
2156 | const struct btrfs_extent_item *ei, | |
2157 | u32 item_size, | |
2158 | struct btrfs_extent_inline_ref **out_eiref, | |
2159 | int *out_type) | |
a542ad1b JS |
2160 | { |
2161 | unsigned long end; | |
2162 | u64 flags; | |
2163 | struct btrfs_tree_block_info *info; | |
2164 | ||
2165 | if (!*ptr) { | |
2166 | /* first call */ | |
2167 | flags = btrfs_extent_flags(eb, ei); | |
2168 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
6eda71d0 LB |
2169 | if (key->type == BTRFS_METADATA_ITEM_KEY) { |
2170 | /* a skinny metadata extent */ | |
2171 | *out_eiref = | |
2172 | (struct btrfs_extent_inline_ref *)(ei + 1); | |
2173 | } else { | |
2174 | WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY); | |
2175 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2176 | *out_eiref = | |
2177 | (struct btrfs_extent_inline_ref *)(info + 1); | |
2178 | } | |
a542ad1b JS |
2179 | } else { |
2180 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2181 | } | |
2182 | *ptr = (unsigned long)*out_eiref; | |
cd857dd6 | 2183 | if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) |
a542ad1b JS |
2184 | return -ENOENT; |
2185 | } | |
2186 | ||
2187 | end = (unsigned long)ei + item_size; | |
6eda71d0 | 2188 | *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); |
3de28d57 LB |
2189 | *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref, |
2190 | BTRFS_REF_TYPE_ANY); | |
2191 | if (*out_type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 2192 | return -EUCLEAN; |
a542ad1b JS |
2193 | |
2194 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
2195 | WARN_ON(*ptr > end); | |
2196 | if (*ptr == end) | |
2197 | return 1; /* last */ | |
2198 | ||
2199 | return 0; | |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * reads the tree block backref for an extent. tree level and root are returned | |
2204 | * through out_level and out_root. ptr must point to a 0 value for the first | |
e0c476b1 | 2205 | * call and may be modified (see get_extent_inline_ref comment). |
a542ad1b JS |
2206 | * returns 0 if data was provided, 1 if there was no more data to provide or |
2207 | * <0 on error. | |
2208 | */ | |
2209 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
2210 | struct btrfs_key *key, struct btrfs_extent_item *ei, |
2211 | u32 item_size, u64 *out_root, u8 *out_level) | |
a542ad1b JS |
2212 | { |
2213 | int ret; | |
2214 | int type; | |
a542ad1b JS |
2215 | struct btrfs_extent_inline_ref *eiref; |
2216 | ||
2217 | if (*ptr == (unsigned long)-1) | |
2218 | return 1; | |
2219 | ||
2220 | while (1) { | |
e0c476b1 | 2221 | ret = get_extent_inline_ref(ptr, eb, key, ei, item_size, |
6eda71d0 | 2222 | &eiref, &type); |
a542ad1b JS |
2223 | if (ret < 0) |
2224 | return ret; | |
2225 | ||
2226 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
2227 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2228 | break; | |
2229 | ||
2230 | if (ret == 1) | |
2231 | return 1; | |
2232 | } | |
2233 | ||
2234 | /* we can treat both ref types equally here */ | |
a542ad1b | 2235 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); |
a1317f45 FM |
2236 | |
2237 | if (key->type == BTRFS_EXTENT_ITEM_KEY) { | |
2238 | struct btrfs_tree_block_info *info; | |
2239 | ||
2240 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2241 | *out_level = btrfs_tree_block_level(eb, info); | |
2242 | } else { | |
2243 | ASSERT(key->type == BTRFS_METADATA_ITEM_KEY); | |
2244 | *out_level = (u8)key->offset; | |
2245 | } | |
a542ad1b JS |
2246 | |
2247 | if (ret == 1) | |
2248 | *ptr = (unsigned long)-1; | |
2249 | ||
2250 | return 0; | |
2251 | } | |
2252 | ||
ab8d0fc4 JM |
2253 | static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, |
2254 | struct extent_inode_elem *inode_list, | |
2255 | u64 root, u64 extent_item_objectid, | |
2256 | iterate_extent_inodes_t *iterate, void *ctx) | |
a542ad1b | 2257 | { |
976b1908 | 2258 | struct extent_inode_elem *eie; |
4692cf58 | 2259 | int ret = 0; |
4692cf58 | 2260 | |
976b1908 | 2261 | for (eie = inode_list; eie; eie = eie->next) { |
ab8d0fc4 JM |
2262 | btrfs_debug(fs_info, |
2263 | "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu", | |
2264 | extent_item_objectid, eie->inum, | |
2265 | eie->offset, root); | |
976b1908 | 2266 | ret = iterate(eie->inum, eie->offset, root, ctx); |
4692cf58 | 2267 | if (ret) { |
ab8d0fc4 JM |
2268 | btrfs_debug(fs_info, |
2269 | "stopping iteration for %llu due to ret=%d", | |
2270 | extent_item_objectid, ret); | |
4692cf58 JS |
2271 | break; |
2272 | } | |
a542ad1b JS |
2273 | } |
2274 | ||
a542ad1b JS |
2275 | return ret; |
2276 | } | |
2277 | ||
2278 | /* | |
2279 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 2280 | * the given parameters. |
a542ad1b JS |
2281 | * when the iterator function returns a non-zero value, iteration stops. |
2282 | */ | |
2283 | int iterate_extent_inodes(struct btrfs_fs_info *fs_info, | |
4692cf58 | 2284 | u64 extent_item_objectid, u64 extent_item_pos, |
7a3ae2f8 | 2285 | int search_commit_root, |
c995ab3c ZB |
2286 | iterate_extent_inodes_t *iterate, void *ctx, |
2287 | bool ignore_offset) | |
a542ad1b | 2288 | { |
a542ad1b | 2289 | int ret; |
da61d31a | 2290 | struct btrfs_trans_handle *trans = NULL; |
7a3ae2f8 JS |
2291 | struct ulist *refs = NULL; |
2292 | struct ulist *roots = NULL; | |
4692cf58 JS |
2293 | struct ulist_node *ref_node = NULL; |
2294 | struct ulist_node *root_node = NULL; | |
f3a84ccd | 2295 | struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem); |
cd1b413c JS |
2296 | struct ulist_iterator ref_uiter; |
2297 | struct ulist_iterator root_uiter; | |
a542ad1b | 2298 | |
ab8d0fc4 | 2299 | btrfs_debug(fs_info, "resolving all inodes for extent %llu", |
4692cf58 | 2300 | extent_item_objectid); |
a542ad1b | 2301 | |
da61d31a | 2302 | if (!search_commit_root) { |
30a9da5d | 2303 | trans = btrfs_attach_transaction(fs_info->tree_root); |
bfc61c36 FM |
2304 | if (IS_ERR(trans)) { |
2305 | if (PTR_ERR(trans) != -ENOENT && | |
2306 | PTR_ERR(trans) != -EROFS) | |
2307 | return PTR_ERR(trans); | |
2308 | trans = NULL; | |
2309 | } | |
2310 | } | |
2311 | ||
2312 | if (trans) | |
f3a84ccd | 2313 | btrfs_get_tree_mod_seq(fs_info, &seq_elem); |
bfc61c36 | 2314 | else |
9e351cc8 | 2315 | down_read(&fs_info->commit_root_sem); |
a542ad1b | 2316 | |
4692cf58 | 2317 | ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, |
f3a84ccd | 2318 | seq_elem.seq, &refs, |
c995ab3c | 2319 | &extent_item_pos, ignore_offset); |
4692cf58 JS |
2320 | if (ret) |
2321 | goto out; | |
a542ad1b | 2322 | |
cd1b413c JS |
2323 | ULIST_ITER_INIT(&ref_uiter); |
2324 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
e0c476b1 | 2325 | ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val, |
f3a84ccd | 2326 | seq_elem.seq, &roots, |
c995ab3c | 2327 | ignore_offset); |
4692cf58 JS |
2328 | if (ret) |
2329 | break; | |
cd1b413c JS |
2330 | ULIST_ITER_INIT(&root_uiter); |
2331 | while (!ret && (root_node = ulist_next(roots, &root_uiter))) { | |
ab8d0fc4 JM |
2332 | btrfs_debug(fs_info, |
2333 | "root %llu references leaf %llu, data list %#llx", | |
2334 | root_node->val, ref_node->val, | |
2335 | ref_node->aux); | |
2336 | ret = iterate_leaf_refs(fs_info, | |
2337 | (struct extent_inode_elem *) | |
995e01b7 JS |
2338 | (uintptr_t)ref_node->aux, |
2339 | root_node->val, | |
2340 | extent_item_objectid, | |
2341 | iterate, ctx); | |
4692cf58 | 2342 | } |
976b1908 | 2343 | ulist_free(roots); |
a542ad1b JS |
2344 | } |
2345 | ||
976b1908 | 2346 | free_leaf_list(refs); |
4692cf58 | 2347 | out: |
bfc61c36 | 2348 | if (trans) { |
f3a84ccd | 2349 | btrfs_put_tree_mod_seq(fs_info, &seq_elem); |
3a45bb20 | 2350 | btrfs_end_transaction(trans); |
9e351cc8 JB |
2351 | } else { |
2352 | up_read(&fs_info->commit_root_sem); | |
7a3ae2f8 JS |
2353 | } |
2354 | ||
a542ad1b JS |
2355 | return ret; |
2356 | } | |
2357 | ||
e3059ec0 DS |
2358 | static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) |
2359 | { | |
2360 | struct btrfs_data_container *inodes = ctx; | |
2361 | const size_t c = 3 * sizeof(u64); | |
2362 | ||
2363 | if (inodes->bytes_left >= c) { | |
2364 | inodes->bytes_left -= c; | |
2365 | inodes->val[inodes->elem_cnt] = inum; | |
2366 | inodes->val[inodes->elem_cnt + 1] = offset; | |
2367 | inodes->val[inodes->elem_cnt + 2] = root; | |
2368 | inodes->elem_cnt += 3; | |
2369 | } else { | |
2370 | inodes->bytes_missing += c - inodes->bytes_left; | |
2371 | inodes->bytes_left = 0; | |
2372 | inodes->elem_missed += 3; | |
2373 | } | |
2374 | ||
2375 | return 0; | |
2376 | } | |
2377 | ||
a542ad1b JS |
2378 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, |
2379 | struct btrfs_path *path, | |
e3059ec0 | 2380 | void *ctx, bool ignore_offset) |
a542ad1b JS |
2381 | { |
2382 | int ret; | |
4692cf58 | 2383 | u64 extent_item_pos; |
69917e43 | 2384 | u64 flags = 0; |
a542ad1b | 2385 | struct btrfs_key found_key; |
7a3ae2f8 | 2386 | int search_commit_root = path->search_commit_root; |
a542ad1b | 2387 | |
69917e43 | 2388 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 2389 | btrfs_release_path(path); |
a542ad1b JS |
2390 | if (ret < 0) |
2391 | return ret; | |
69917e43 | 2392 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 2393 | return -EINVAL; |
a542ad1b | 2394 | |
4692cf58 | 2395 | extent_item_pos = logical - found_key.objectid; |
7a3ae2f8 JS |
2396 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
2397 | extent_item_pos, search_commit_root, | |
e3059ec0 | 2398 | build_ino_list, ctx, ignore_offset); |
a542ad1b JS |
2399 | |
2400 | return ret; | |
2401 | } | |
2402 | ||
ad6240f6 | 2403 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2404 | struct extent_buffer *eb, struct inode_fs_paths *ipath); |
d24bec3a | 2405 | |
875d1daa | 2406 | static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath) |
a542ad1b | 2407 | { |
aefc1eb1 | 2408 | int ret = 0; |
a542ad1b JS |
2409 | int slot; |
2410 | u32 cur; | |
2411 | u32 len; | |
2412 | u32 name_len; | |
2413 | u64 parent = 0; | |
2414 | int found = 0; | |
875d1daa DS |
2415 | struct btrfs_root *fs_root = ipath->fs_root; |
2416 | struct btrfs_path *path = ipath->btrfs_path; | |
a542ad1b | 2417 | struct extent_buffer *eb; |
a542ad1b JS |
2418 | struct btrfs_inode_ref *iref; |
2419 | struct btrfs_key found_key; | |
2420 | ||
aefc1eb1 | 2421 | while (!ret) { |
c234a24d DS |
2422 | ret = btrfs_find_item(fs_root, path, inum, |
2423 | parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY, | |
2424 | &found_key); | |
2425 | ||
a542ad1b JS |
2426 | if (ret < 0) |
2427 | break; | |
2428 | if (ret) { | |
2429 | ret = found ? 0 : -ENOENT; | |
2430 | break; | |
2431 | } | |
2432 | ++found; | |
2433 | ||
2434 | parent = found_key.offset; | |
2435 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2436 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2437 | if (!eb) { | |
2438 | ret = -ENOMEM; | |
2439 | break; | |
2440 | } | |
a542ad1b JS |
2441 | btrfs_release_path(path); |
2442 | ||
a542ad1b JS |
2443 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
2444 | ||
3212fa14 | 2445 | for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) { |
a542ad1b JS |
2446 | name_len = btrfs_inode_ref_name_len(eb, iref); |
2447 | /* path must be released before calling iterate()! */ | |
ab8d0fc4 JM |
2448 | btrfs_debug(fs_root->fs_info, |
2449 | "following ref at offset %u for inode %llu in tree %llu", | |
4fd786e6 MT |
2450 | cur, found_key.objectid, |
2451 | fs_root->root_key.objectid); | |
ad6240f6 | 2452 | ret = inode_to_path(parent, name_len, |
875d1daa | 2453 | (unsigned long)(iref + 1), eb, ipath); |
aefc1eb1 | 2454 | if (ret) |
a542ad1b | 2455 | break; |
a542ad1b JS |
2456 | len = sizeof(*iref) + name_len; |
2457 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
2458 | } | |
2459 | free_extent_buffer(eb); | |
2460 | } | |
2461 | ||
2462 | btrfs_release_path(path); | |
2463 | ||
2464 | return ret; | |
2465 | } | |
2466 | ||
875d1daa | 2467 | static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath) |
d24bec3a MF |
2468 | { |
2469 | int ret; | |
2470 | int slot; | |
2471 | u64 offset = 0; | |
2472 | u64 parent; | |
2473 | int found = 0; | |
875d1daa DS |
2474 | struct btrfs_root *fs_root = ipath->fs_root; |
2475 | struct btrfs_path *path = ipath->btrfs_path; | |
d24bec3a MF |
2476 | struct extent_buffer *eb; |
2477 | struct btrfs_inode_extref *extref; | |
d24bec3a MF |
2478 | u32 item_size; |
2479 | u32 cur_offset; | |
2480 | unsigned long ptr; | |
2481 | ||
2482 | while (1) { | |
2483 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
2484 | &offset); | |
2485 | if (ret < 0) | |
2486 | break; | |
2487 | if (ret) { | |
2488 | ret = found ? 0 : -ENOENT; | |
2489 | break; | |
2490 | } | |
2491 | ++found; | |
2492 | ||
2493 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2494 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2495 | if (!eb) { | |
2496 | ret = -ENOMEM; | |
2497 | break; | |
2498 | } | |
d24bec3a MF |
2499 | btrfs_release_path(path); |
2500 | ||
3212fa14 | 2501 | item_size = btrfs_item_size(eb, slot); |
2849a854 | 2502 | ptr = btrfs_item_ptr_offset(eb, slot); |
d24bec3a MF |
2503 | cur_offset = 0; |
2504 | ||
2505 | while (cur_offset < item_size) { | |
2506 | u32 name_len; | |
2507 | ||
2508 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
2509 | parent = btrfs_inode_extref_parent(eb, extref); | |
2510 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
ad6240f6 | 2511 | ret = inode_to_path(parent, name_len, |
875d1daa | 2512 | (unsigned long)&extref->name, eb, ipath); |
d24bec3a MF |
2513 | if (ret) |
2514 | break; | |
2515 | ||
2849a854 | 2516 | cur_offset += btrfs_inode_extref_name_len(eb, extref); |
d24bec3a MF |
2517 | cur_offset += sizeof(*extref); |
2518 | } | |
d24bec3a MF |
2519 | free_extent_buffer(eb); |
2520 | ||
2521 | offset++; | |
2522 | } | |
2523 | ||
2524 | btrfs_release_path(path); | |
2525 | ||
2526 | return ret; | |
2527 | } | |
2528 | ||
a542ad1b JS |
2529 | /* |
2530 | * returns 0 if the path could be dumped (probably truncated) | |
2531 | * returns <0 in case of an error | |
2532 | */ | |
d24bec3a | 2533 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2534 | struct extent_buffer *eb, struct inode_fs_paths *ipath) |
a542ad1b | 2535 | { |
a542ad1b JS |
2536 | char *fspath; |
2537 | char *fspath_min; | |
2538 | int i = ipath->fspath->elem_cnt; | |
2539 | const int s_ptr = sizeof(char *); | |
2540 | u32 bytes_left; | |
2541 | ||
2542 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
2543 | ipath->fspath->bytes_left - s_ptr : 0; | |
2544 | ||
740c3d22 | 2545 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
2546 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
2547 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
2548 | if (IS_ERR(fspath)) |
2549 | return PTR_ERR(fspath); | |
2550 | ||
2551 | if (fspath > fspath_min) { | |
745c4d8e | 2552 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
2553 | ++ipath->fspath->elem_cnt; |
2554 | ipath->fspath->bytes_left = fspath - fspath_min; | |
2555 | } else { | |
2556 | ++ipath->fspath->elem_missed; | |
2557 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
2558 | ipath->fspath->bytes_left = 0; | |
2559 | } | |
2560 | ||
2561 | return 0; | |
2562 | } | |
2563 | ||
2564 | /* | |
2565 | * this dumps all file system paths to the inode into the ipath struct, provided | |
2566 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 2567 | * from ipath->fspath->val[i]. |
a542ad1b | 2568 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 2569 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
01327610 | 2570 | * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise, |
a542ad1b JS |
2571 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would |
2572 | * have been needed to return all paths. | |
2573 | */ | |
2574 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
2575 | { | |
ad6240f6 DS |
2576 | int ret; |
2577 | int found_refs = 0; | |
2578 | ||
875d1daa | 2579 | ret = iterate_inode_refs(inum, ipath); |
ad6240f6 DS |
2580 | if (!ret) |
2581 | ++found_refs; | |
2582 | else if (ret != -ENOENT) | |
2583 | return ret; | |
2584 | ||
875d1daa | 2585 | ret = iterate_inode_extrefs(inum, ipath); |
ad6240f6 DS |
2586 | if (ret == -ENOENT && found_refs) |
2587 | return 0; | |
2588 | ||
2589 | return ret; | |
a542ad1b JS |
2590 | } |
2591 | ||
a542ad1b JS |
2592 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
2593 | { | |
2594 | struct btrfs_data_container *data; | |
2595 | size_t alloc_bytes; | |
2596 | ||
2597 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
f54de068 | 2598 | data = kvmalloc(alloc_bytes, GFP_KERNEL); |
a542ad1b JS |
2599 | if (!data) |
2600 | return ERR_PTR(-ENOMEM); | |
2601 | ||
2602 | if (total_bytes >= sizeof(*data)) { | |
2603 | data->bytes_left = total_bytes - sizeof(*data); | |
2604 | data->bytes_missing = 0; | |
2605 | } else { | |
2606 | data->bytes_missing = sizeof(*data) - total_bytes; | |
2607 | data->bytes_left = 0; | |
2608 | } | |
2609 | ||
2610 | data->elem_cnt = 0; | |
2611 | data->elem_missed = 0; | |
2612 | ||
2613 | return data; | |
2614 | } | |
2615 | ||
2616 | /* | |
2617 | * allocates space to return multiple file system paths for an inode. | |
2618 | * total_bytes to allocate are passed, note that space usable for actual path | |
2619 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
2620 | * the returned pointer must be freed with free_ipath() in the end. | |
2621 | */ | |
2622 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
2623 | struct btrfs_path *path) | |
2624 | { | |
2625 | struct inode_fs_paths *ifp; | |
2626 | struct btrfs_data_container *fspath; | |
2627 | ||
2628 | fspath = init_data_container(total_bytes); | |
2629 | if (IS_ERR(fspath)) | |
afc6961f | 2630 | return ERR_CAST(fspath); |
a542ad1b | 2631 | |
f54de068 | 2632 | ifp = kmalloc(sizeof(*ifp), GFP_KERNEL); |
a542ad1b | 2633 | if (!ifp) { |
f54de068 | 2634 | kvfree(fspath); |
a542ad1b JS |
2635 | return ERR_PTR(-ENOMEM); |
2636 | } | |
2637 | ||
2638 | ifp->btrfs_path = path; | |
2639 | ifp->fspath = fspath; | |
2640 | ifp->fs_root = fs_root; | |
2641 | ||
2642 | return ifp; | |
2643 | } | |
2644 | ||
2645 | void free_ipath(struct inode_fs_paths *ipath) | |
2646 | { | |
4735fb28 JJ |
2647 | if (!ipath) |
2648 | return; | |
f54de068 | 2649 | kvfree(ipath->fspath); |
a542ad1b JS |
2650 | kfree(ipath); |
2651 | } | |
a37f232b | 2652 | |
d68194b2 | 2653 | struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info) |
a37f232b QW |
2654 | { |
2655 | struct btrfs_backref_iter *ret; | |
2656 | ||
d68194b2 | 2657 | ret = kzalloc(sizeof(*ret), GFP_NOFS); |
a37f232b QW |
2658 | if (!ret) |
2659 | return NULL; | |
2660 | ||
2661 | ret->path = btrfs_alloc_path(); | |
c15c2ec0 | 2662 | if (!ret->path) { |
a37f232b QW |
2663 | kfree(ret); |
2664 | return NULL; | |
2665 | } | |
2666 | ||
2667 | /* Current backref iterator only supports iteration in commit root */ | |
2668 | ret->path->search_commit_root = 1; | |
2669 | ret->path->skip_locking = 1; | |
2670 | ret->fs_info = fs_info; | |
2671 | ||
2672 | return ret; | |
2673 | } | |
2674 | ||
2675 | int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr) | |
2676 | { | |
2677 | struct btrfs_fs_info *fs_info = iter->fs_info; | |
29cbcf40 | 2678 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); |
a37f232b QW |
2679 | struct btrfs_path *path = iter->path; |
2680 | struct btrfs_extent_item *ei; | |
2681 | struct btrfs_key key; | |
2682 | int ret; | |
2683 | ||
2684 | key.objectid = bytenr; | |
2685 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2686 | key.offset = (u64)-1; | |
2687 | iter->bytenr = bytenr; | |
2688 | ||
29cbcf40 | 2689 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a37f232b QW |
2690 | if (ret < 0) |
2691 | return ret; | |
2692 | if (ret == 0) { | |
2693 | ret = -EUCLEAN; | |
2694 | goto release; | |
2695 | } | |
2696 | if (path->slots[0] == 0) { | |
2697 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
2698 | ret = -EUCLEAN; | |
2699 | goto release; | |
2700 | } | |
2701 | path->slots[0]--; | |
2702 | ||
2703 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2704 | if ((key.type != BTRFS_EXTENT_ITEM_KEY && | |
2705 | key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) { | |
2706 | ret = -ENOENT; | |
2707 | goto release; | |
2708 | } | |
2709 | memcpy(&iter->cur_key, &key, sizeof(key)); | |
2710 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2711 | path->slots[0]); | |
2712 | iter->end_ptr = (u32)(iter->item_ptr + | |
3212fa14 | 2713 | btrfs_item_size(path->nodes[0], path->slots[0])); |
a37f232b QW |
2714 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2715 | struct btrfs_extent_item); | |
2716 | ||
2717 | /* | |
2718 | * Only support iteration on tree backref yet. | |
2719 | * | |
2720 | * This is an extra precaution for non skinny-metadata, where | |
2721 | * EXTENT_ITEM is also used for tree blocks, that we can only use | |
2722 | * extent flags to determine if it's a tree block. | |
2723 | */ | |
2724 | if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) { | |
2725 | ret = -ENOTSUPP; | |
2726 | goto release; | |
2727 | } | |
2728 | iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei)); | |
2729 | ||
2730 | /* If there is no inline backref, go search for keyed backref */ | |
2731 | if (iter->cur_ptr >= iter->end_ptr) { | |
29cbcf40 | 2732 | ret = btrfs_next_item(extent_root, path); |
a37f232b QW |
2733 | |
2734 | /* No inline nor keyed ref */ | |
2735 | if (ret > 0) { | |
2736 | ret = -ENOENT; | |
2737 | goto release; | |
2738 | } | |
2739 | if (ret < 0) | |
2740 | goto release; | |
2741 | ||
2742 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, | |
2743 | path->slots[0]); | |
2744 | if (iter->cur_key.objectid != bytenr || | |
2745 | (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY && | |
2746 | iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) { | |
2747 | ret = -ENOENT; | |
2748 | goto release; | |
2749 | } | |
2750 | iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2751 | path->slots[0]); | |
2752 | iter->item_ptr = iter->cur_ptr; | |
3212fa14 | 2753 | iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size( |
a37f232b QW |
2754 | path->nodes[0], path->slots[0])); |
2755 | } | |
2756 | ||
2757 | return 0; | |
2758 | release: | |
2759 | btrfs_backref_iter_release(iter); | |
2760 | return ret; | |
2761 | } | |
c39c2ddc QW |
2762 | |
2763 | /* | |
2764 | * Go to the next backref item of current bytenr, can be either inlined or | |
2765 | * keyed. | |
2766 | * | |
2767 | * Caller needs to check whether it's inline ref or not by iter->cur_key. | |
2768 | * | |
2769 | * Return 0 if we get next backref without problem. | |
2770 | * Return >0 if there is no extra backref for this bytenr. | |
2771 | * Return <0 if there is something wrong happened. | |
2772 | */ | |
2773 | int btrfs_backref_iter_next(struct btrfs_backref_iter *iter) | |
2774 | { | |
2775 | struct extent_buffer *eb = btrfs_backref_get_eb(iter); | |
29cbcf40 | 2776 | struct btrfs_root *extent_root; |
c39c2ddc QW |
2777 | struct btrfs_path *path = iter->path; |
2778 | struct btrfs_extent_inline_ref *iref; | |
2779 | int ret; | |
2780 | u32 size; | |
2781 | ||
2782 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
2783 | /* We're still inside the inline refs */ | |
2784 | ASSERT(iter->cur_ptr < iter->end_ptr); | |
2785 | ||
2786 | if (btrfs_backref_has_tree_block_info(iter)) { | |
2787 | /* First tree block info */ | |
2788 | size = sizeof(struct btrfs_tree_block_info); | |
2789 | } else { | |
2790 | /* Use inline ref type to determine the size */ | |
2791 | int type; | |
2792 | ||
2793 | iref = (struct btrfs_extent_inline_ref *) | |
2794 | ((unsigned long)iter->cur_ptr); | |
2795 | type = btrfs_extent_inline_ref_type(eb, iref); | |
2796 | ||
2797 | size = btrfs_extent_inline_ref_size(type); | |
2798 | } | |
2799 | iter->cur_ptr += size; | |
2800 | if (iter->cur_ptr < iter->end_ptr) | |
2801 | return 0; | |
2802 | ||
2803 | /* All inline items iterated, fall through */ | |
2804 | } | |
2805 | ||
2806 | /* We're at keyed items, there is no inline item, go to the next one */ | |
29cbcf40 JB |
2807 | extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr); |
2808 | ret = btrfs_next_item(extent_root, iter->path); | |
c39c2ddc QW |
2809 | if (ret) |
2810 | return ret; | |
2811 | ||
2812 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]); | |
2813 | if (iter->cur_key.objectid != iter->bytenr || | |
2814 | (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY && | |
2815 | iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY)) | |
2816 | return 1; | |
2817 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2818 | path->slots[0]); | |
2819 | iter->cur_ptr = iter->item_ptr; | |
3212fa14 | 2820 | iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0], |
c39c2ddc QW |
2821 | path->slots[0]); |
2822 | return 0; | |
2823 | } | |
584fb121 QW |
2824 | |
2825 | void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info, | |
2826 | struct btrfs_backref_cache *cache, int is_reloc) | |
2827 | { | |
2828 | int i; | |
2829 | ||
2830 | cache->rb_root = RB_ROOT; | |
2831 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
2832 | INIT_LIST_HEAD(&cache->pending[i]); | |
2833 | INIT_LIST_HEAD(&cache->changed); | |
2834 | INIT_LIST_HEAD(&cache->detached); | |
2835 | INIT_LIST_HEAD(&cache->leaves); | |
2836 | INIT_LIST_HEAD(&cache->pending_edge); | |
2837 | INIT_LIST_HEAD(&cache->useless_node); | |
2838 | cache->fs_info = fs_info; | |
2839 | cache->is_reloc = is_reloc; | |
2840 | } | |
b1818dab QW |
2841 | |
2842 | struct btrfs_backref_node *btrfs_backref_alloc_node( | |
2843 | struct btrfs_backref_cache *cache, u64 bytenr, int level) | |
2844 | { | |
2845 | struct btrfs_backref_node *node; | |
2846 | ||
2847 | ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL); | |
2848 | node = kzalloc(sizeof(*node), GFP_NOFS); | |
2849 | if (!node) | |
2850 | return node; | |
2851 | ||
2852 | INIT_LIST_HEAD(&node->list); | |
2853 | INIT_LIST_HEAD(&node->upper); | |
2854 | INIT_LIST_HEAD(&node->lower); | |
2855 | RB_CLEAR_NODE(&node->rb_node); | |
2856 | cache->nr_nodes++; | |
2857 | node->level = level; | |
2858 | node->bytenr = bytenr; | |
2859 | ||
2860 | return node; | |
2861 | } | |
47254d07 QW |
2862 | |
2863 | struct btrfs_backref_edge *btrfs_backref_alloc_edge( | |
2864 | struct btrfs_backref_cache *cache) | |
2865 | { | |
2866 | struct btrfs_backref_edge *edge; | |
2867 | ||
2868 | edge = kzalloc(sizeof(*edge), GFP_NOFS); | |
2869 | if (edge) | |
2870 | cache->nr_edges++; | |
2871 | return edge; | |
2872 | } | |
023acb07 QW |
2873 | |
2874 | /* | |
2875 | * Drop the backref node from cache, also cleaning up all its | |
2876 | * upper edges and any uncached nodes in the path. | |
2877 | * | |
2878 | * This cleanup happens bottom up, thus the node should either | |
2879 | * be the lowest node in the cache or a detached node. | |
2880 | */ | |
2881 | void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache, | |
2882 | struct btrfs_backref_node *node) | |
2883 | { | |
2884 | struct btrfs_backref_node *upper; | |
2885 | struct btrfs_backref_edge *edge; | |
2886 | ||
2887 | if (!node) | |
2888 | return; | |
2889 | ||
2890 | BUG_ON(!node->lowest && !node->detached); | |
2891 | while (!list_empty(&node->upper)) { | |
2892 | edge = list_entry(node->upper.next, struct btrfs_backref_edge, | |
2893 | list[LOWER]); | |
2894 | upper = edge->node[UPPER]; | |
2895 | list_del(&edge->list[LOWER]); | |
2896 | list_del(&edge->list[UPPER]); | |
2897 | btrfs_backref_free_edge(cache, edge); | |
2898 | ||
023acb07 QW |
2899 | /* |
2900 | * Add the node to leaf node list if no other child block | |
2901 | * cached. | |
2902 | */ | |
2903 | if (list_empty(&upper->lower)) { | |
2904 | list_add_tail(&upper->lower, &cache->leaves); | |
2905 | upper->lowest = 1; | |
2906 | } | |
2907 | } | |
2908 | ||
2909 | btrfs_backref_drop_node(cache, node); | |
2910 | } | |
13fe1bdb QW |
2911 | |
2912 | /* | |
2913 | * Release all nodes/edges from current cache | |
2914 | */ | |
2915 | void btrfs_backref_release_cache(struct btrfs_backref_cache *cache) | |
2916 | { | |
2917 | struct btrfs_backref_node *node; | |
2918 | int i; | |
2919 | ||
2920 | while (!list_empty(&cache->detached)) { | |
2921 | node = list_entry(cache->detached.next, | |
2922 | struct btrfs_backref_node, list); | |
2923 | btrfs_backref_cleanup_node(cache, node); | |
2924 | } | |
2925 | ||
2926 | while (!list_empty(&cache->leaves)) { | |
2927 | node = list_entry(cache->leaves.next, | |
2928 | struct btrfs_backref_node, lower); | |
2929 | btrfs_backref_cleanup_node(cache, node); | |
2930 | } | |
2931 | ||
2932 | cache->last_trans = 0; | |
2933 | ||
2934 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
2935 | ASSERT(list_empty(&cache->pending[i])); | |
2936 | ASSERT(list_empty(&cache->pending_edge)); | |
2937 | ASSERT(list_empty(&cache->useless_node)); | |
2938 | ASSERT(list_empty(&cache->changed)); | |
2939 | ASSERT(list_empty(&cache->detached)); | |
2940 | ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); | |
2941 | ASSERT(!cache->nr_nodes); | |
2942 | ASSERT(!cache->nr_edges); | |
2943 | } | |
1b60d2ec QW |
2944 | |
2945 | /* | |
2946 | * Handle direct tree backref | |
2947 | * | |
2948 | * Direct tree backref means, the backref item shows its parent bytenr | |
2949 | * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined). | |
2950 | * | |
2951 | * @ref_key: The converted backref key. | |
2952 | * For keyed backref, it's the item key. | |
2953 | * For inlined backref, objectid is the bytenr, | |
2954 | * type is btrfs_inline_ref_type, offset is | |
2955 | * btrfs_inline_ref_offset. | |
2956 | */ | |
2957 | static int handle_direct_tree_backref(struct btrfs_backref_cache *cache, | |
2958 | struct btrfs_key *ref_key, | |
2959 | struct btrfs_backref_node *cur) | |
2960 | { | |
2961 | struct btrfs_backref_edge *edge; | |
2962 | struct btrfs_backref_node *upper; | |
2963 | struct rb_node *rb_node; | |
2964 | ||
2965 | ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
2966 | ||
2967 | /* Only reloc root uses backref pointing to itself */ | |
2968 | if (ref_key->objectid == ref_key->offset) { | |
2969 | struct btrfs_root *root; | |
2970 | ||
2971 | cur->is_reloc_root = 1; | |
2972 | /* Only reloc backref cache cares about a specific root */ | |
2973 | if (cache->is_reloc) { | |
2974 | root = find_reloc_root(cache->fs_info, cur->bytenr); | |
f78743fb | 2975 | if (!root) |
1b60d2ec QW |
2976 | return -ENOENT; |
2977 | cur->root = root; | |
2978 | } else { | |
2979 | /* | |
2980 | * For generic purpose backref cache, reloc root node | |
2981 | * is useless. | |
2982 | */ | |
2983 | list_add(&cur->list, &cache->useless_node); | |
2984 | } | |
2985 | return 0; | |
2986 | } | |
2987 | ||
2988 | edge = btrfs_backref_alloc_edge(cache); | |
2989 | if (!edge) | |
2990 | return -ENOMEM; | |
2991 | ||
2992 | rb_node = rb_simple_search(&cache->rb_root, ref_key->offset); | |
2993 | if (!rb_node) { | |
2994 | /* Parent node not yet cached */ | |
2995 | upper = btrfs_backref_alloc_node(cache, ref_key->offset, | |
2996 | cur->level + 1); | |
2997 | if (!upper) { | |
2998 | btrfs_backref_free_edge(cache, edge); | |
2999 | return -ENOMEM; | |
3000 | } | |
3001 | ||
3002 | /* | |
3003 | * Backrefs for the upper level block isn't cached, add the | |
3004 | * block to pending list | |
3005 | */ | |
3006 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
3007 | } else { | |
3008 | /* Parent node already cached */ | |
3009 | upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node); | |
3010 | ASSERT(upper->checked); | |
3011 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3012 | } | |
3013 | btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER); | |
3014 | return 0; | |
3015 | } | |
3016 | ||
3017 | /* | |
3018 | * Handle indirect tree backref | |
3019 | * | |
3020 | * Indirect tree backref means, we only know which tree the node belongs to. | |
3021 | * We still need to do a tree search to find out the parents. This is for | |
3022 | * TREE_BLOCK_REF backref (keyed or inlined). | |
3023 | * | |
3024 | * @ref_key: The same as @ref_key in handle_direct_tree_backref() | |
3025 | * @tree_key: The first key of this tree block. | |
1a9fd417 | 3026 | * @path: A clean (released) path, to avoid allocating path every time |
1b60d2ec QW |
3027 | * the function get called. |
3028 | */ | |
3029 | static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache, | |
3030 | struct btrfs_path *path, | |
3031 | struct btrfs_key *ref_key, | |
3032 | struct btrfs_key *tree_key, | |
3033 | struct btrfs_backref_node *cur) | |
3034 | { | |
3035 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3036 | struct btrfs_backref_node *upper; | |
3037 | struct btrfs_backref_node *lower; | |
3038 | struct btrfs_backref_edge *edge; | |
3039 | struct extent_buffer *eb; | |
3040 | struct btrfs_root *root; | |
1b60d2ec QW |
3041 | struct rb_node *rb_node; |
3042 | int level; | |
3043 | bool need_check = true; | |
3044 | int ret; | |
3045 | ||
56e9357a | 3046 | root = btrfs_get_fs_root(fs_info, ref_key->offset, false); |
1b60d2ec QW |
3047 | if (IS_ERR(root)) |
3048 | return PTR_ERR(root); | |
92a7cc42 | 3049 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
3050 | cur->cowonly = 1; |
3051 | ||
3052 | if (btrfs_root_level(&root->root_item) == cur->level) { | |
3053 | /* Tree root */ | |
3054 | ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr); | |
876de781 QW |
3055 | /* |
3056 | * For reloc backref cache, we may ignore reloc root. But for | |
3057 | * general purpose backref cache, we can't rely on | |
3058 | * btrfs_should_ignore_reloc_root() as it may conflict with | |
3059 | * current running relocation and lead to missing root. | |
3060 | * | |
3061 | * For general purpose backref cache, reloc root detection is | |
3062 | * completely relying on direct backref (key->offset is parent | |
3063 | * bytenr), thus only do such check for reloc cache. | |
3064 | */ | |
3065 | if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) { | |
1b60d2ec QW |
3066 | btrfs_put_root(root); |
3067 | list_add(&cur->list, &cache->useless_node); | |
3068 | } else { | |
3069 | cur->root = root; | |
3070 | } | |
3071 | return 0; | |
3072 | } | |
3073 | ||
3074 | level = cur->level + 1; | |
3075 | ||
3076 | /* Search the tree to find parent blocks referring to the block */ | |
3077 | path->search_commit_root = 1; | |
3078 | path->skip_locking = 1; | |
3079 | path->lowest_level = level; | |
3080 | ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0); | |
3081 | path->lowest_level = 0; | |
3082 | if (ret < 0) { | |
3083 | btrfs_put_root(root); | |
3084 | return ret; | |
3085 | } | |
3086 | if (ret > 0 && path->slots[level] > 0) | |
3087 | path->slots[level]--; | |
3088 | ||
3089 | eb = path->nodes[level]; | |
3090 | if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) { | |
3091 | btrfs_err(fs_info, | |
3092 | "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", | |
3093 | cur->bytenr, level - 1, root->root_key.objectid, | |
3094 | tree_key->objectid, tree_key->type, tree_key->offset); | |
3095 | btrfs_put_root(root); | |
3096 | ret = -ENOENT; | |
3097 | goto out; | |
3098 | } | |
3099 | lower = cur; | |
3100 | ||
3101 | /* Add all nodes and edges in the path */ | |
3102 | for (; level < BTRFS_MAX_LEVEL; level++) { | |
3103 | if (!path->nodes[level]) { | |
3104 | ASSERT(btrfs_root_bytenr(&root->root_item) == | |
3105 | lower->bytenr); | |
876de781 QW |
3106 | /* Same as previous should_ignore_reloc_root() call */ |
3107 | if (btrfs_should_ignore_reloc_root(root) && | |
3108 | cache->is_reloc) { | |
1b60d2ec QW |
3109 | btrfs_put_root(root); |
3110 | list_add(&lower->list, &cache->useless_node); | |
3111 | } else { | |
3112 | lower->root = root; | |
3113 | } | |
3114 | break; | |
3115 | } | |
3116 | ||
3117 | edge = btrfs_backref_alloc_edge(cache); | |
3118 | if (!edge) { | |
3119 | btrfs_put_root(root); | |
3120 | ret = -ENOMEM; | |
3121 | goto out; | |
3122 | } | |
3123 | ||
3124 | eb = path->nodes[level]; | |
3125 | rb_node = rb_simple_search(&cache->rb_root, eb->start); | |
3126 | if (!rb_node) { | |
3127 | upper = btrfs_backref_alloc_node(cache, eb->start, | |
3128 | lower->level + 1); | |
3129 | if (!upper) { | |
3130 | btrfs_put_root(root); | |
3131 | btrfs_backref_free_edge(cache, edge); | |
3132 | ret = -ENOMEM; | |
3133 | goto out; | |
3134 | } | |
3135 | upper->owner = btrfs_header_owner(eb); | |
92a7cc42 | 3136 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
3137 | upper->cowonly = 1; |
3138 | ||
3139 | /* | |
3140 | * If we know the block isn't shared we can avoid | |
3141 | * checking its backrefs. | |
3142 | */ | |
3143 | if (btrfs_block_can_be_shared(root, eb)) | |
3144 | upper->checked = 0; | |
3145 | else | |
3146 | upper->checked = 1; | |
3147 | ||
3148 | /* | |
3149 | * Add the block to pending list if we need to check its | |
3150 | * backrefs, we only do this once while walking up a | |
3151 | * tree as we will catch anything else later on. | |
3152 | */ | |
3153 | if (!upper->checked && need_check) { | |
3154 | need_check = false; | |
3155 | list_add_tail(&edge->list[UPPER], | |
3156 | &cache->pending_edge); | |
3157 | } else { | |
3158 | if (upper->checked) | |
3159 | need_check = true; | |
3160 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3161 | } | |
3162 | } else { | |
3163 | upper = rb_entry(rb_node, struct btrfs_backref_node, | |
3164 | rb_node); | |
3165 | ASSERT(upper->checked); | |
3166 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3167 | if (!upper->owner) | |
3168 | upper->owner = btrfs_header_owner(eb); | |
3169 | } | |
3170 | btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER); | |
3171 | ||
3172 | if (rb_node) { | |
3173 | btrfs_put_root(root); | |
3174 | break; | |
3175 | } | |
3176 | lower = upper; | |
3177 | upper = NULL; | |
3178 | } | |
3179 | out: | |
3180 | btrfs_release_path(path); | |
3181 | return ret; | |
3182 | } | |
3183 | ||
3184 | /* | |
3185 | * Add backref node @cur into @cache. | |
3186 | * | |
3187 | * NOTE: Even if the function returned 0, @cur is not yet cached as its upper | |
3188 | * links aren't yet bi-directional. Needs to finish such links. | |
fc997ed0 | 3189 | * Use btrfs_backref_finish_upper_links() to finish such linkage. |
1b60d2ec QW |
3190 | * |
3191 | * @path: Released path for indirect tree backref lookup | |
3192 | * @iter: Released backref iter for extent tree search | |
3193 | * @node_key: The first key of the tree block | |
3194 | */ | |
3195 | int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache, | |
3196 | struct btrfs_path *path, | |
3197 | struct btrfs_backref_iter *iter, | |
3198 | struct btrfs_key *node_key, | |
3199 | struct btrfs_backref_node *cur) | |
3200 | { | |
3201 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3202 | struct btrfs_backref_edge *edge; | |
3203 | struct btrfs_backref_node *exist; | |
3204 | int ret; | |
3205 | ||
3206 | ret = btrfs_backref_iter_start(iter, cur->bytenr); | |
3207 | if (ret < 0) | |
3208 | return ret; | |
3209 | /* | |
3210 | * We skip the first btrfs_tree_block_info, as we don't use the key | |
3211 | * stored in it, but fetch it from the tree block | |
3212 | */ | |
3213 | if (btrfs_backref_has_tree_block_info(iter)) { | |
3214 | ret = btrfs_backref_iter_next(iter); | |
3215 | if (ret < 0) | |
3216 | goto out; | |
3217 | /* No extra backref? This means the tree block is corrupted */ | |
3218 | if (ret > 0) { | |
3219 | ret = -EUCLEAN; | |
3220 | goto out; | |
3221 | } | |
3222 | } | |
3223 | WARN_ON(cur->checked); | |
3224 | if (!list_empty(&cur->upper)) { | |
3225 | /* | |
3226 | * The backref was added previously when processing backref of | |
3227 | * type BTRFS_TREE_BLOCK_REF_KEY | |
3228 | */ | |
3229 | ASSERT(list_is_singular(&cur->upper)); | |
3230 | edge = list_entry(cur->upper.next, struct btrfs_backref_edge, | |
3231 | list[LOWER]); | |
3232 | ASSERT(list_empty(&edge->list[UPPER])); | |
3233 | exist = edge->node[UPPER]; | |
3234 | /* | |
3235 | * Add the upper level block to pending list if we need check | |
3236 | * its backrefs | |
3237 | */ | |
3238 | if (!exist->checked) | |
3239 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
3240 | } else { | |
3241 | exist = NULL; | |
3242 | } | |
3243 | ||
3244 | for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { | |
3245 | struct extent_buffer *eb; | |
3246 | struct btrfs_key key; | |
3247 | int type; | |
3248 | ||
3249 | cond_resched(); | |
3250 | eb = btrfs_backref_get_eb(iter); | |
3251 | ||
3252 | key.objectid = iter->bytenr; | |
3253 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
3254 | struct btrfs_extent_inline_ref *iref; | |
3255 | ||
3256 | /* Update key for inline backref */ | |
3257 | iref = (struct btrfs_extent_inline_ref *) | |
3258 | ((unsigned long)iter->cur_ptr); | |
3259 | type = btrfs_get_extent_inline_ref_type(eb, iref, | |
3260 | BTRFS_REF_TYPE_BLOCK); | |
3261 | if (type == BTRFS_REF_TYPE_INVALID) { | |
3262 | ret = -EUCLEAN; | |
3263 | goto out; | |
3264 | } | |
3265 | key.type = type; | |
3266 | key.offset = btrfs_extent_inline_ref_offset(eb, iref); | |
3267 | } else { | |
3268 | key.type = iter->cur_key.type; | |
3269 | key.offset = iter->cur_key.offset; | |
3270 | } | |
3271 | ||
3272 | /* | |
3273 | * Parent node found and matches current inline ref, no need to | |
3274 | * rebuild this node for this inline ref | |
3275 | */ | |
3276 | if (exist && | |
3277 | ((key.type == BTRFS_TREE_BLOCK_REF_KEY && | |
3278 | exist->owner == key.offset) || | |
3279 | (key.type == BTRFS_SHARED_BLOCK_REF_KEY && | |
3280 | exist->bytenr == key.offset))) { | |
3281 | exist = NULL; | |
3282 | continue; | |
3283 | } | |
3284 | ||
3285 | /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ | |
3286 | if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
3287 | ret = handle_direct_tree_backref(cache, &key, cur); | |
3288 | if (ret < 0) | |
3289 | goto out; | |
3290 | continue; | |
3291 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { | |
3292 | ret = -EINVAL; | |
3293 | btrfs_print_v0_err(fs_info); | |
3294 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
3295 | goto out; | |
3296 | } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { | |
3297 | continue; | |
3298 | } | |
3299 | ||
3300 | /* | |
3301 | * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset | |
3302 | * means the root objectid. We need to search the tree to get | |
3303 | * its parent bytenr. | |
3304 | */ | |
3305 | ret = handle_indirect_tree_backref(cache, path, &key, node_key, | |
3306 | cur); | |
3307 | if (ret < 0) | |
3308 | goto out; | |
3309 | } | |
3310 | ret = 0; | |
3311 | cur->checked = 1; | |
3312 | WARN_ON(exist); | |
3313 | out: | |
3314 | btrfs_backref_iter_release(iter); | |
3315 | return ret; | |
3316 | } | |
fc997ed0 QW |
3317 | |
3318 | /* | |
3319 | * Finish the upwards linkage created by btrfs_backref_add_tree_node() | |
3320 | */ | |
3321 | int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache, | |
3322 | struct btrfs_backref_node *start) | |
3323 | { | |
3324 | struct list_head *useless_node = &cache->useless_node; | |
3325 | struct btrfs_backref_edge *edge; | |
3326 | struct rb_node *rb_node; | |
3327 | LIST_HEAD(pending_edge); | |
3328 | ||
3329 | ASSERT(start->checked); | |
3330 | ||
3331 | /* Insert this node to cache if it's not COW-only */ | |
3332 | if (!start->cowonly) { | |
3333 | rb_node = rb_simple_insert(&cache->rb_root, start->bytenr, | |
3334 | &start->rb_node); | |
3335 | if (rb_node) | |
3336 | btrfs_backref_panic(cache->fs_info, start->bytenr, | |
3337 | -EEXIST); | |
3338 | list_add_tail(&start->lower, &cache->leaves); | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * Use breadth first search to iterate all related edges. | |
3343 | * | |
3344 | * The starting points are all the edges of this node | |
3345 | */ | |
3346 | list_for_each_entry(edge, &start->upper, list[LOWER]) | |
3347 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3348 | ||
3349 | while (!list_empty(&pending_edge)) { | |
3350 | struct btrfs_backref_node *upper; | |
3351 | struct btrfs_backref_node *lower; | |
fc997ed0 QW |
3352 | |
3353 | edge = list_first_entry(&pending_edge, | |
3354 | struct btrfs_backref_edge, list[UPPER]); | |
3355 | list_del_init(&edge->list[UPPER]); | |
3356 | upper = edge->node[UPPER]; | |
3357 | lower = edge->node[LOWER]; | |
3358 | ||
3359 | /* Parent is detached, no need to keep any edges */ | |
3360 | if (upper->detached) { | |
3361 | list_del(&edge->list[LOWER]); | |
3362 | btrfs_backref_free_edge(cache, edge); | |
3363 | ||
3364 | /* Lower node is orphan, queue for cleanup */ | |
3365 | if (list_empty(&lower->upper)) | |
3366 | list_add(&lower->list, useless_node); | |
3367 | continue; | |
3368 | } | |
3369 | ||
3370 | /* | |
3371 | * All new nodes added in current build_backref_tree() haven't | |
3372 | * been linked to the cache rb tree. | |
3373 | * So if we have upper->rb_node populated, this means a cache | |
3374 | * hit. We only need to link the edge, as @upper and all its | |
3375 | * parents have already been linked. | |
3376 | */ | |
3377 | if (!RB_EMPTY_NODE(&upper->rb_node)) { | |
3378 | if (upper->lowest) { | |
3379 | list_del_init(&upper->lower); | |
3380 | upper->lowest = 0; | |
3381 | } | |
3382 | ||
3383 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3384 | continue; | |
3385 | } | |
3386 | ||
3387 | /* Sanity check, we shouldn't have any unchecked nodes */ | |
3388 | if (!upper->checked) { | |
3389 | ASSERT(0); | |
3390 | return -EUCLEAN; | |
3391 | } | |
3392 | ||
3393 | /* Sanity check, COW-only node has non-COW-only parent */ | |
3394 | if (start->cowonly != upper->cowonly) { | |
3395 | ASSERT(0); | |
3396 | return -EUCLEAN; | |
3397 | } | |
3398 | ||
3399 | /* Only cache non-COW-only (subvolume trees) tree blocks */ | |
3400 | if (!upper->cowonly) { | |
3401 | rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr, | |
3402 | &upper->rb_node); | |
3403 | if (rb_node) { | |
3404 | btrfs_backref_panic(cache->fs_info, | |
3405 | upper->bytenr, -EEXIST); | |
3406 | return -EUCLEAN; | |
3407 | } | |
3408 | } | |
3409 | ||
3410 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3411 | ||
3412 | /* | |
3413 | * Also queue all the parent edges of this uncached node | |
3414 | * to finish the upper linkage | |
3415 | */ | |
3416 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3417 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3418 | } | |
3419 | return 0; | |
3420 | } | |
1b23ea18 QW |
3421 | |
3422 | void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache, | |
3423 | struct btrfs_backref_node *node) | |
3424 | { | |
3425 | struct btrfs_backref_node *lower; | |
3426 | struct btrfs_backref_node *upper; | |
3427 | struct btrfs_backref_edge *edge; | |
3428 | ||
3429 | while (!list_empty(&cache->useless_node)) { | |
3430 | lower = list_first_entry(&cache->useless_node, | |
3431 | struct btrfs_backref_node, list); | |
3432 | list_del_init(&lower->list); | |
3433 | } | |
3434 | while (!list_empty(&cache->pending_edge)) { | |
3435 | edge = list_first_entry(&cache->pending_edge, | |
3436 | struct btrfs_backref_edge, list[UPPER]); | |
3437 | list_del(&edge->list[UPPER]); | |
3438 | list_del(&edge->list[LOWER]); | |
3439 | lower = edge->node[LOWER]; | |
3440 | upper = edge->node[UPPER]; | |
3441 | btrfs_backref_free_edge(cache, edge); | |
3442 | ||
3443 | /* | |
3444 | * Lower is no longer linked to any upper backref nodes and | |
3445 | * isn't in the cache, we can free it ourselves. | |
3446 | */ | |
3447 | if (list_empty(&lower->upper) && | |
3448 | RB_EMPTY_NODE(&lower->rb_node)) | |
3449 | list_add(&lower->list, &cache->useless_node); | |
3450 | ||
3451 | if (!RB_EMPTY_NODE(&upper->rb_node)) | |
3452 | continue; | |
3453 | ||
3454 | /* Add this guy's upper edges to the list to process */ | |
3455 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3456 | list_add_tail(&edge->list[UPPER], | |
3457 | &cache->pending_edge); | |
3458 | if (list_empty(&upper->upper)) | |
3459 | list_add(&upper->list, &cache->useless_node); | |
3460 | } | |
3461 | ||
3462 | while (!list_empty(&cache->useless_node)) { | |
3463 | lower = list_first_entry(&cache->useless_node, | |
3464 | struct btrfs_backref_node, list); | |
3465 | list_del_init(&lower->list); | |
3466 | if (lower == node) | |
3467 | node = NULL; | |
49ecc679 | 3468 | btrfs_backref_drop_node(cache, lower); |
1b23ea18 QW |
3469 | } |
3470 | ||
3471 | btrfs_backref_cleanup_node(cache, node); | |
3472 | ASSERT(list_empty(&cache->useless_node) && | |
3473 | list_empty(&cache->pending_edge)); | |
3474 | } |