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" |
a0231804 | 20 | #include "extent-tree.h" |
67707479 | 21 | #include "relocation.h" |
27137fac | 22 | #include "tree-checker.h" |
a542ad1b | 23 | |
877c1476 FM |
24 | /* Just arbitrary numbers so we can be sure one of these happened. */ |
25 | #define BACKREF_FOUND_SHARED 6 | |
26 | #define BACKREF_FOUND_NOT_SHARED 7 | |
dc046b10 | 27 | |
976b1908 JS |
28 | struct extent_inode_elem { |
29 | u64 inum; | |
30 | u64 offset; | |
c7499a64 | 31 | u64 num_bytes; |
976b1908 JS |
32 | struct extent_inode_elem *next; |
33 | }; | |
34 | ||
88ffb665 FM |
35 | static int check_extent_in_eb(struct btrfs_backref_walk_ctx *ctx, |
36 | const struct btrfs_key *key, | |
73980bec JM |
37 | const struct extent_buffer *eb, |
38 | const struct btrfs_file_extent_item *fi, | |
6ce6ba53 | 39 | struct extent_inode_elem **eie) |
976b1908 | 40 | { |
c7499a64 | 41 | const u64 data_len = btrfs_file_extent_num_bytes(eb, fi); |
88ffb665 | 42 | u64 offset = key->offset; |
976b1908 | 43 | struct extent_inode_elem *e; |
88ffb665 FM |
44 | const u64 *root_ids; |
45 | int root_count; | |
46 | bool cached; | |
976b1908 | 47 | |
6ce6ba53 | 48 | if (!btrfs_file_extent_compression(eb, fi) && |
8ca15e05 JB |
49 | !btrfs_file_extent_encryption(eb, fi) && |
50 | !btrfs_file_extent_other_encoding(eb, fi)) { | |
51 | u64 data_offset; | |
976b1908 | 52 | |
8ca15e05 | 53 | data_offset = btrfs_file_extent_offset(eb, fi); |
8ca15e05 | 54 | |
88ffb665 FM |
55 | if (ctx->extent_item_pos < data_offset || |
56 | ctx->extent_item_pos >= data_offset + data_len) | |
8ca15e05 | 57 | return 1; |
88ffb665 | 58 | offset += ctx->extent_item_pos - data_offset; |
8ca15e05 | 59 | } |
976b1908 | 60 | |
88ffb665 FM |
61 | if (!ctx->indirect_ref_iterator || !ctx->cache_lookup) |
62 | goto add_inode_elem; | |
63 | ||
64 | cached = ctx->cache_lookup(eb->start, ctx->user_ctx, &root_ids, | |
65 | &root_count); | |
66 | if (!cached) | |
67 | goto add_inode_elem; | |
68 | ||
69 | for (int i = 0; i < root_count; i++) { | |
70 | int ret; | |
71 | ||
72 | ret = ctx->indirect_ref_iterator(key->objectid, offset, | |
73 | data_len, root_ids[i], | |
74 | ctx->user_ctx); | |
75 | if (ret) | |
76 | return ret; | |
77 | } | |
78 | ||
79 | add_inode_elem: | |
976b1908 JS |
80 | e = kmalloc(sizeof(*e), GFP_NOFS); |
81 | if (!e) | |
82 | return -ENOMEM; | |
83 | ||
84 | e->next = *eie; | |
85 | e->inum = key->objectid; | |
88ffb665 | 86 | e->offset = offset; |
c7499a64 | 87 | e->num_bytes = data_len; |
976b1908 JS |
88 | *eie = e; |
89 | ||
90 | return 0; | |
91 | } | |
92 | ||
f05c4746 WS |
93 | static void free_inode_elem_list(struct extent_inode_elem *eie) |
94 | { | |
95 | struct extent_inode_elem *eie_next; | |
96 | ||
97 | for (; eie; eie = eie_next) { | |
98 | eie_next = eie->next; | |
99 | kfree(eie); | |
100 | } | |
101 | } | |
102 | ||
88ffb665 FM |
103 | static int find_extent_in_eb(struct btrfs_backref_walk_ctx *ctx, |
104 | const struct extent_buffer *eb, | |
6ce6ba53 | 105 | struct extent_inode_elem **eie) |
976b1908 JS |
106 | { |
107 | u64 disk_byte; | |
108 | struct btrfs_key key; | |
109 | struct btrfs_file_extent_item *fi; | |
110 | int slot; | |
111 | int nritems; | |
112 | int extent_type; | |
113 | int ret; | |
114 | ||
115 | /* | |
116 | * from the shared data ref, we only have the leaf but we need | |
117 | * the key. thus, we must look into all items and see that we | |
118 | * find one (some) with a reference to our extent item. | |
119 | */ | |
120 | nritems = btrfs_header_nritems(eb); | |
121 | for (slot = 0; slot < nritems; ++slot) { | |
122 | btrfs_item_key_to_cpu(eb, &key, slot); | |
123 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
124 | continue; | |
125 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
126 | extent_type = btrfs_file_extent_type(eb, fi); | |
127 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
128 | continue; | |
129 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
130 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
88ffb665 | 131 | if (disk_byte != ctx->bytenr) |
976b1908 JS |
132 | continue; |
133 | ||
88ffb665 FM |
134 | ret = check_extent_in_eb(ctx, &key, eb, fi, eie); |
135 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) | |
976b1908 JS |
136 | return ret; |
137 | } | |
138 | ||
139 | return 0; | |
140 | } | |
141 | ||
86d5f994 | 142 | struct preftree { |
ecf160b4 | 143 | struct rb_root_cached root; |
6c336b21 | 144 | unsigned int count; |
86d5f994 EN |
145 | }; |
146 | ||
ecf160b4 | 147 | #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 } |
86d5f994 EN |
148 | |
149 | struct preftrees { | |
150 | struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */ | |
151 | struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */ | |
152 | struct preftree indirect_missing_keys; | |
153 | }; | |
154 | ||
3ec4d323 EN |
155 | /* |
156 | * Checks for a shared extent during backref search. | |
157 | * | |
158 | * The share_count tracks prelim_refs (direct and indirect) having a | |
159 | * ref->count >0: | |
160 | * - incremented when a ref->count transitions to >0 | |
161 | * - decremented when a ref->count transitions to <1 | |
162 | */ | |
163 | struct share_check { | |
877c1476 FM |
164 | struct btrfs_backref_share_check_ctx *ctx; |
165 | struct btrfs_root *root; | |
3ec4d323 | 166 | u64 inum; |
73e339e6 | 167 | u64 data_bytenr; |
6976201f | 168 | u64 data_extent_gen; |
73e339e6 FM |
169 | /* |
170 | * Counts number of inodes that refer to an extent (different inodes in | |
171 | * the same root or different roots) that we could find. The sharedness | |
172 | * check typically stops once this counter gets greater than 1, so it | |
173 | * may not reflect the total number of inodes. | |
174 | */ | |
3ec4d323 | 175 | int share_count; |
73e339e6 FM |
176 | /* |
177 | * The number of times we found our inode refers to the data extent we | |
178 | * are determining the sharedness. In other words, how many file extent | |
179 | * items we could find for our inode that point to our target data | |
180 | * extent. The value we get here after finishing the extent sharedness | |
181 | * check may be smaller than reality, but if it ends up being greater | |
182 | * than 1, then we know for sure the inode has multiple file extent | |
183 | * items that point to our inode, and we can safely assume it's useful | |
184 | * to cache the sharedness check result. | |
185 | */ | |
186 | int self_ref_count; | |
4fc7b572 | 187 | bool have_delayed_delete_refs; |
3ec4d323 EN |
188 | }; |
189 | ||
190 | static inline int extent_is_shared(struct share_check *sc) | |
191 | { | |
192 | return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0; | |
193 | } | |
194 | ||
b9e9a6cb WS |
195 | static struct kmem_cache *btrfs_prelim_ref_cache; |
196 | ||
197 | int __init btrfs_prelim_ref_init(void) | |
198 | { | |
199 | btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", | |
e0c476b1 | 200 | sizeof(struct prelim_ref), |
b9e9a6cb | 201 | 0, |
fba4b697 | 202 | SLAB_MEM_SPREAD, |
b9e9a6cb WS |
203 | NULL); |
204 | if (!btrfs_prelim_ref_cache) | |
205 | return -ENOMEM; | |
206 | return 0; | |
207 | } | |
208 | ||
e67c718b | 209 | void __cold btrfs_prelim_ref_exit(void) |
b9e9a6cb | 210 | { |
5598e900 | 211 | kmem_cache_destroy(btrfs_prelim_ref_cache); |
b9e9a6cb WS |
212 | } |
213 | ||
86d5f994 EN |
214 | static void free_pref(struct prelim_ref *ref) |
215 | { | |
216 | kmem_cache_free(btrfs_prelim_ref_cache, ref); | |
217 | } | |
218 | ||
219 | /* | |
220 | * Return 0 when both refs are for the same block (and can be merged). | |
221 | * A -1 return indicates ref1 is a 'lower' block than ref2, while 1 | |
222 | * indicates a 'higher' block. | |
223 | */ | |
224 | static int prelim_ref_compare(struct prelim_ref *ref1, | |
225 | struct prelim_ref *ref2) | |
226 | { | |
227 | if (ref1->level < ref2->level) | |
228 | return -1; | |
229 | if (ref1->level > ref2->level) | |
230 | return 1; | |
231 | if (ref1->root_id < ref2->root_id) | |
232 | return -1; | |
233 | if (ref1->root_id > ref2->root_id) | |
234 | return 1; | |
235 | if (ref1->key_for_search.type < ref2->key_for_search.type) | |
236 | return -1; | |
237 | if (ref1->key_for_search.type > ref2->key_for_search.type) | |
238 | return 1; | |
239 | if (ref1->key_for_search.objectid < ref2->key_for_search.objectid) | |
240 | return -1; | |
241 | if (ref1->key_for_search.objectid > ref2->key_for_search.objectid) | |
242 | return 1; | |
243 | if (ref1->key_for_search.offset < ref2->key_for_search.offset) | |
244 | return -1; | |
245 | if (ref1->key_for_search.offset > ref2->key_for_search.offset) | |
246 | return 1; | |
247 | if (ref1->parent < ref2->parent) | |
248 | return -1; | |
249 | if (ref1->parent > ref2->parent) | |
250 | return 1; | |
251 | ||
252 | return 0; | |
253 | } | |
254 | ||
ccc8dc75 | 255 | static void update_share_count(struct share_check *sc, int oldcount, |
73e339e6 | 256 | int newcount, struct prelim_ref *newref) |
3ec4d323 EN |
257 | { |
258 | if ((!sc) || (oldcount == 0 && newcount < 1)) | |
259 | return; | |
260 | ||
261 | if (oldcount > 0 && newcount < 1) | |
262 | sc->share_count--; | |
263 | else if (oldcount < 1 && newcount > 0) | |
264 | sc->share_count++; | |
73e339e6 | 265 | |
877c1476 | 266 | if (newref->root_id == sc->root->root_key.objectid && |
73e339e6 FM |
267 | newref->wanted_disk_byte == sc->data_bytenr && |
268 | newref->key_for_search.objectid == sc->inum) | |
269 | sc->self_ref_count += newref->count; | |
3ec4d323 EN |
270 | } |
271 | ||
86d5f994 EN |
272 | /* |
273 | * Add @newref to the @root rbtree, merging identical refs. | |
274 | * | |
3ec4d323 | 275 | * Callers should assume that newref has been freed after calling. |
86d5f994 | 276 | */ |
00142756 JM |
277 | static void prelim_ref_insert(const struct btrfs_fs_info *fs_info, |
278 | struct preftree *preftree, | |
3ec4d323 EN |
279 | struct prelim_ref *newref, |
280 | struct share_check *sc) | |
86d5f994 | 281 | { |
ecf160b4 | 282 | struct rb_root_cached *root; |
86d5f994 EN |
283 | struct rb_node **p; |
284 | struct rb_node *parent = NULL; | |
285 | struct prelim_ref *ref; | |
286 | int result; | |
ecf160b4 | 287 | bool leftmost = true; |
86d5f994 EN |
288 | |
289 | root = &preftree->root; | |
ecf160b4 | 290 | p = &root->rb_root.rb_node; |
86d5f994 EN |
291 | |
292 | while (*p) { | |
293 | parent = *p; | |
294 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
295 | result = prelim_ref_compare(ref, newref); | |
296 | if (result < 0) { | |
297 | p = &(*p)->rb_left; | |
298 | } else if (result > 0) { | |
299 | p = &(*p)->rb_right; | |
ecf160b4 | 300 | leftmost = false; |
86d5f994 EN |
301 | } else { |
302 | /* Identical refs, merge them and free @newref */ | |
303 | struct extent_inode_elem *eie = ref->inode_list; | |
304 | ||
305 | while (eie && eie->next) | |
306 | eie = eie->next; | |
307 | ||
308 | if (!eie) | |
309 | ref->inode_list = newref->inode_list; | |
310 | else | |
311 | eie->next = newref->inode_list; | |
00142756 JM |
312 | trace_btrfs_prelim_ref_merge(fs_info, ref, newref, |
313 | preftree->count); | |
3ec4d323 EN |
314 | /* |
315 | * A delayed ref can have newref->count < 0. | |
316 | * The ref->count is updated to follow any | |
317 | * BTRFS_[ADD|DROP]_DELAYED_REF actions. | |
318 | */ | |
319 | update_share_count(sc, ref->count, | |
73e339e6 | 320 | ref->count + newref->count, newref); |
86d5f994 EN |
321 | ref->count += newref->count; |
322 | free_pref(newref); | |
323 | return; | |
324 | } | |
325 | } | |
326 | ||
73e339e6 | 327 | update_share_count(sc, 0, newref->count, newref); |
6c336b21 | 328 | preftree->count++; |
00142756 | 329 | trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count); |
86d5f994 | 330 | rb_link_node(&newref->rbnode, parent, p); |
ecf160b4 | 331 | rb_insert_color_cached(&newref->rbnode, root, leftmost); |
86d5f994 EN |
332 | } |
333 | ||
334 | /* | |
335 | * Release the entire tree. We don't care about internal consistency so | |
336 | * just free everything and then reset the tree root. | |
337 | */ | |
338 | static void prelim_release(struct preftree *preftree) | |
339 | { | |
340 | struct prelim_ref *ref, *next_ref; | |
341 | ||
ecf160b4 | 342 | rbtree_postorder_for_each_entry_safe(ref, next_ref, |
92876eec FM |
343 | &preftree->root.rb_root, rbnode) { |
344 | free_inode_elem_list(ref->inode_list); | |
86d5f994 | 345 | free_pref(ref); |
92876eec | 346 | } |
86d5f994 | 347 | |
ecf160b4 | 348 | preftree->root = RB_ROOT_CACHED; |
6c336b21 | 349 | preftree->count = 0; |
86d5f994 EN |
350 | } |
351 | ||
d5c88b73 JS |
352 | /* |
353 | * the rules for all callers of this function are: | |
354 | * - obtaining the parent is the goal | |
355 | * - if you add a key, you must know that it is a correct key | |
356 | * - if you cannot add the parent or a correct key, then we will look into the | |
357 | * block later to set a correct key | |
358 | * | |
359 | * delayed refs | |
360 | * ============ | |
361 | * backref type | shared | indirect | shared | indirect | |
362 | * information | tree | tree | data | data | |
363 | * --------------------+--------+----------+--------+---------- | |
364 | * parent logical | y | - | - | - | |
365 | * key to resolve | - | y | y | y | |
366 | * tree block logical | - | - | - | - | |
367 | * root for resolving | y | y | y | y | |
368 | * | |
369 | * - column 1: we've the parent -> done | |
370 | * - column 2, 3, 4: we use the key to find the parent | |
371 | * | |
372 | * on disk refs (inline or keyed) | |
373 | * ============================== | |
374 | * backref type | shared | indirect | shared | indirect | |
375 | * information | tree | tree | data | data | |
376 | * --------------------+--------+----------+--------+---------- | |
377 | * parent logical | y | - | y | - | |
378 | * key to resolve | - | - | - | y | |
379 | * tree block logical | y | y | y | y | |
380 | * root for resolving | - | y | y | y | |
381 | * | |
382 | * - column 1, 3: we've the parent -> done | |
383 | * - column 2: we take the first key from the block to find the parent | |
e0c476b1 | 384 | * (see add_missing_keys) |
d5c88b73 JS |
385 | * - column 4: we use the key to find the parent |
386 | * | |
387 | * additional information that's available but not required to find the parent | |
388 | * block might help in merging entries to gain some speed. | |
389 | */ | |
00142756 JM |
390 | static int add_prelim_ref(const struct btrfs_fs_info *fs_info, |
391 | struct preftree *preftree, u64 root_id, | |
e0c476b1 | 392 | const struct btrfs_key *key, int level, u64 parent, |
3ec4d323 EN |
393 | u64 wanted_disk_byte, int count, |
394 | struct share_check *sc, gfp_t gfp_mask) | |
8da6d581 | 395 | { |
e0c476b1 | 396 | struct prelim_ref *ref; |
8da6d581 | 397 | |
48ec4736 LB |
398 | if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) |
399 | return 0; | |
400 | ||
b9e9a6cb | 401 | ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); |
8da6d581 JS |
402 | if (!ref) |
403 | return -ENOMEM; | |
404 | ||
405 | ref->root_id = root_id; | |
7ac8b88e | 406 | if (key) |
d5c88b73 | 407 | ref->key_for_search = *key; |
7ac8b88e | 408 | else |
d5c88b73 | 409 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 410 | |
3301958b | 411 | ref->inode_list = NULL; |
8da6d581 JS |
412 | ref->level = level; |
413 | ref->count = count; | |
414 | ref->parent = parent; | |
415 | ref->wanted_disk_byte = wanted_disk_byte; | |
3ec4d323 EN |
416 | prelim_ref_insert(fs_info, preftree, ref, sc); |
417 | return extent_is_shared(sc); | |
8da6d581 JS |
418 | } |
419 | ||
86d5f994 | 420 | /* direct refs use root == 0, key == NULL */ |
00142756 JM |
421 | static int add_direct_ref(const struct btrfs_fs_info *fs_info, |
422 | struct preftrees *preftrees, int level, u64 parent, | |
3ec4d323 EN |
423 | u64 wanted_disk_byte, int count, |
424 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 | 425 | { |
00142756 | 426 | return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level, |
3ec4d323 | 427 | parent, wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
428 | } |
429 | ||
430 | /* indirect refs use parent == 0 */ | |
00142756 JM |
431 | static int add_indirect_ref(const struct btrfs_fs_info *fs_info, |
432 | struct preftrees *preftrees, u64 root_id, | |
86d5f994 | 433 | const struct btrfs_key *key, int level, |
3ec4d323 EN |
434 | u64 wanted_disk_byte, int count, |
435 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 EN |
436 | { |
437 | struct preftree *tree = &preftrees->indirect; | |
438 | ||
439 | if (!key) | |
440 | tree = &preftrees->indirect_missing_keys; | |
00142756 | 441 | return add_prelim_ref(fs_info, tree, root_id, key, level, 0, |
3ec4d323 | 442 | wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
443 | } |
444 | ||
ed58f2e6 | 445 | static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr) |
446 | { | |
447 | struct rb_node **p = &preftrees->direct.root.rb_root.rb_node; | |
448 | struct rb_node *parent = NULL; | |
449 | struct prelim_ref *ref = NULL; | |
9c6c723f | 450 | struct prelim_ref target = {}; |
ed58f2e6 | 451 | int result; |
452 | ||
453 | target.parent = bytenr; | |
454 | ||
455 | while (*p) { | |
456 | parent = *p; | |
457 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
458 | result = prelim_ref_compare(ref, &target); | |
459 | ||
460 | if (result < 0) | |
461 | p = &(*p)->rb_left; | |
462 | else if (result > 0) | |
463 | p = &(*p)->rb_right; | |
464 | else | |
465 | return 1; | |
466 | } | |
467 | return 0; | |
468 | } | |
469 | ||
a2c8d27e FM |
470 | static int add_all_parents(struct btrfs_backref_walk_ctx *ctx, |
471 | struct btrfs_root *root, struct btrfs_path *path, | |
ed58f2e6 | 472 | struct ulist *parents, |
473 | struct preftrees *preftrees, struct prelim_ref *ref, | |
a2c8d27e | 474 | int level) |
8da6d581 | 475 | { |
69bca40d AB |
476 | int ret = 0; |
477 | int slot; | |
478 | struct extent_buffer *eb; | |
479 | struct btrfs_key key; | |
7ef81ac8 | 480 | struct btrfs_key *key_for_search = &ref->key_for_search; |
8da6d581 | 481 | struct btrfs_file_extent_item *fi; |
ed8c4913 | 482 | struct extent_inode_elem *eie = NULL, *old = NULL; |
8da6d581 | 483 | u64 disk_byte; |
7ef81ac8 JB |
484 | u64 wanted_disk_byte = ref->wanted_disk_byte; |
485 | u64 count = 0; | |
7ac8b88e | 486 | u64 data_offset; |
560840af | 487 | u8 type; |
8da6d581 | 488 | |
69bca40d AB |
489 | if (level != 0) { |
490 | eb = path->nodes[level]; | |
491 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
492 | if (ret < 0) |
493 | return ret; | |
8da6d581 | 494 | return 0; |
69bca40d | 495 | } |
8da6d581 JS |
496 | |
497 | /* | |
ed58f2e6 | 498 | * 1. We normally enter this function with the path already pointing to |
499 | * the first item to check. But sometimes, we may enter it with | |
500 | * slot == nritems. | |
501 | * 2. We are searching for normal backref but bytenr of this leaf | |
502 | * matches shared data backref | |
cfc0eed0 | 503 | * 3. The leaf owner is not equal to the root we are searching |
504 | * | |
ed58f2e6 | 505 | * For these cases, go to the next leaf before we continue. |
8da6d581 | 506 | */ |
ed58f2e6 | 507 | eb = path->nodes[0]; |
508 | if (path->slots[0] >= btrfs_header_nritems(eb) || | |
cfc0eed0 | 509 | is_shared_data_backref(preftrees, eb->start) || |
510 | ref->root_id != btrfs_header_owner(eb)) { | |
a2c8d27e | 511 | if (ctx->time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
512 | ret = btrfs_next_leaf(root, path); |
513 | else | |
a2c8d27e | 514 | ret = btrfs_next_old_leaf(root, path, ctx->time_seq); |
21633fc6 | 515 | } |
8da6d581 | 516 | |
b25b0b87 | 517 | while (!ret && count < ref->count) { |
8da6d581 | 518 | eb = path->nodes[0]; |
69bca40d AB |
519 | slot = path->slots[0]; |
520 | ||
521 | btrfs_item_key_to_cpu(eb, &key, slot); | |
522 | ||
523 | if (key.objectid != key_for_search->objectid || | |
524 | key.type != BTRFS_EXTENT_DATA_KEY) | |
525 | break; | |
526 | ||
ed58f2e6 | 527 | /* |
528 | * We are searching for normal backref but bytenr of this leaf | |
cfc0eed0 | 529 | * matches shared data backref, OR |
530 | * the leaf owner is not equal to the root we are searching for | |
ed58f2e6 | 531 | */ |
cfc0eed0 | 532 | if (slot == 0 && |
533 | (is_shared_data_backref(preftrees, eb->start) || | |
534 | ref->root_id != btrfs_header_owner(eb))) { | |
a2c8d27e | 535 | if (ctx->time_seq == BTRFS_SEQ_LAST) |
ed58f2e6 | 536 | ret = btrfs_next_leaf(root, path); |
537 | else | |
a2c8d27e | 538 | ret = btrfs_next_old_leaf(root, path, ctx->time_seq); |
ed58f2e6 | 539 | continue; |
540 | } | |
69bca40d | 541 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
560840af BB |
542 | type = btrfs_file_extent_type(eb, fi); |
543 | if (type == BTRFS_FILE_EXTENT_INLINE) | |
544 | goto next; | |
69bca40d | 545 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
7ac8b88e | 546 | data_offset = btrfs_file_extent_offset(eb, fi); |
69bca40d AB |
547 | |
548 | if (disk_byte == wanted_disk_byte) { | |
549 | eie = NULL; | |
ed8c4913 | 550 | old = NULL; |
7ac8b88e | 551 | if (ref->key_for_search.offset == key.offset - data_offset) |
552 | count++; | |
553 | else | |
554 | goto next; | |
a2c8d27e | 555 | if (!ctx->ignore_extent_item_pos) { |
88ffb665 FM |
556 | ret = check_extent_in_eb(ctx, &key, eb, fi, &eie); |
557 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || | |
558 | ret < 0) | |
69bca40d AB |
559 | break; |
560 | } | |
ed8c4913 JB |
561 | if (ret > 0) |
562 | goto next; | |
4eb1f66d TI |
563 | ret = ulist_add_merge_ptr(parents, eb->start, |
564 | eie, (void **)&old, GFP_NOFS); | |
ed8c4913 JB |
565 | if (ret < 0) |
566 | break; | |
a2c8d27e | 567 | if (!ret && !ctx->ignore_extent_item_pos) { |
ed8c4913 JB |
568 | while (old->next) |
569 | old = old->next; | |
570 | old->next = eie; | |
69bca40d | 571 | } |
f05c4746 | 572 | eie = NULL; |
8da6d581 | 573 | } |
ed8c4913 | 574 | next: |
a2c8d27e | 575 | if (ctx->time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
576 | ret = btrfs_next_item(root, path); |
577 | else | |
a2c8d27e | 578 | ret = btrfs_next_old_item(root, path, ctx->time_seq); |
8da6d581 JS |
579 | } |
580 | ||
88ffb665 | 581 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) |
f05c4746 | 582 | free_inode_elem_list(eie); |
88ffb665 FM |
583 | else if (ret > 0) |
584 | ret = 0; | |
585 | ||
69bca40d | 586 | return ret; |
8da6d581 JS |
587 | } |
588 | ||
589 | /* | |
590 | * resolve an indirect backref in the form (root_id, key, level) | |
591 | * to a logical address | |
592 | */ | |
a2c8d27e FM |
593 | static int resolve_indirect_ref(struct btrfs_backref_walk_ctx *ctx, |
594 | struct btrfs_path *path, | |
ed58f2e6 | 595 | struct preftrees *preftrees, |
a2c8d27e | 596 | struct prelim_ref *ref, struct ulist *parents) |
8da6d581 | 597 | { |
8da6d581 | 598 | struct btrfs_root *root; |
8da6d581 JS |
599 | struct extent_buffer *eb; |
600 | int ret = 0; | |
601 | int root_level; | |
602 | int level = ref->level; | |
7ac8b88e | 603 | struct btrfs_key search_key = ref->key_for_search; |
8da6d581 | 604 | |
49d11bea JB |
605 | /* |
606 | * If we're search_commit_root we could possibly be holding locks on | |
607 | * other tree nodes. This happens when qgroups does backref walks when | |
608 | * adding new delayed refs. To deal with this we need to look in cache | |
609 | * for the root, and if we don't find it then we need to search the | |
610 | * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage | |
611 | * here. | |
612 | */ | |
613 | if (path->search_commit_root) | |
a2c8d27e | 614 | root = btrfs_get_fs_root_commit_root(ctx->fs_info, path, ref->root_id); |
49d11bea | 615 | else |
a2c8d27e | 616 | root = btrfs_get_fs_root(ctx->fs_info, ref->root_id, false); |
8da6d581 JS |
617 | if (IS_ERR(root)) { |
618 | ret = PTR_ERR(root); | |
9326f76f JB |
619 | goto out_free; |
620 | } | |
621 | ||
39dba873 JB |
622 | if (!path->search_commit_root && |
623 | test_bit(BTRFS_ROOT_DELETING, &root->state)) { | |
624 | ret = -ENOENT; | |
625 | goto out; | |
626 | } | |
627 | ||
a2c8d27e | 628 | if (btrfs_is_testing(ctx->fs_info)) { |
d9ee522b JB |
629 | ret = -ENOENT; |
630 | goto out; | |
631 | } | |
632 | ||
9e351cc8 JB |
633 | if (path->search_commit_root) |
634 | root_level = btrfs_header_level(root->commit_root); | |
a2c8d27e | 635 | else if (ctx->time_seq == BTRFS_SEQ_LAST) |
21633fc6 | 636 | root_level = btrfs_header_level(root->node); |
9e351cc8 | 637 | else |
a2c8d27e | 638 | root_level = btrfs_old_root_level(root, ctx->time_seq); |
8da6d581 | 639 | |
c75e8394 | 640 | if (root_level + 1 == level) |
8da6d581 JS |
641 | goto out; |
642 | ||
7ac8b88e | 643 | /* |
644 | * We can often find data backrefs with an offset that is too large | |
645 | * (>= LLONG_MAX, maximum allowed file offset) due to underflows when | |
646 | * subtracting a file's offset with the data offset of its | |
647 | * corresponding extent data item. This can happen for example in the | |
648 | * clone ioctl. | |
649 | * | |
650 | * So if we detect such case we set the search key's offset to zero to | |
651 | * make sure we will find the matching file extent item at | |
652 | * add_all_parents(), otherwise we will miss it because the offset | |
653 | * taken form the backref is much larger then the offset of the file | |
654 | * extent item. This can make us scan a very large number of file | |
655 | * extent items, but at least it will not make us miss any. | |
656 | * | |
657 | * This is an ugly workaround for a behaviour that should have never | |
658 | * existed, but it does and a fix for the clone ioctl would touch a lot | |
659 | * of places, cause backwards incompatibility and would not fix the | |
660 | * problem for extents cloned with older kernels. | |
661 | */ | |
662 | if (search_key.type == BTRFS_EXTENT_DATA_KEY && | |
663 | search_key.offset >= LLONG_MAX) | |
664 | search_key.offset = 0; | |
8da6d581 | 665 | path->lowest_level = level; |
a2c8d27e | 666 | if (ctx->time_seq == BTRFS_SEQ_LAST) |
7ac8b88e | 667 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); |
21633fc6 | 668 | else |
a2c8d27e | 669 | ret = btrfs_search_old_slot(root, &search_key, path, ctx->time_seq); |
538f72cd | 670 | |
a2c8d27e | 671 | btrfs_debug(ctx->fs_info, |
ab8d0fc4 | 672 | "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)", |
c1c9ff7c GU |
673 | ref->root_id, level, ref->count, ret, |
674 | ref->key_for_search.objectid, ref->key_for_search.type, | |
675 | ref->key_for_search.offset); | |
8da6d581 JS |
676 | if (ret < 0) |
677 | goto out; | |
678 | ||
679 | eb = path->nodes[level]; | |
9345457f | 680 | while (!eb) { |
fae7f21c | 681 | if (WARN_ON(!level)) { |
9345457f JS |
682 | ret = 1; |
683 | goto out; | |
684 | } | |
685 | level--; | |
686 | eb = path->nodes[level]; | |
8da6d581 JS |
687 | } |
688 | ||
a2c8d27e | 689 | ret = add_all_parents(ctx, root, path, parents, preftrees, ref, level); |
8da6d581 | 690 | out: |
00246528 | 691 | btrfs_put_root(root); |
9326f76f | 692 | out_free: |
da61d31a JB |
693 | path->lowest_level = 0; |
694 | btrfs_release_path(path); | |
8da6d581 JS |
695 | return ret; |
696 | } | |
697 | ||
4dae077a JM |
698 | static struct extent_inode_elem * |
699 | unode_aux_to_inode_list(struct ulist_node *node) | |
700 | { | |
701 | if (!node) | |
702 | return NULL; | |
703 | return (struct extent_inode_elem *)(uintptr_t)node->aux; | |
704 | } | |
705 | ||
5614dc3a FM |
706 | static void free_leaf_list(struct ulist *ulist) |
707 | { | |
708 | struct ulist_node *node; | |
709 | struct ulist_iterator uiter; | |
710 | ||
711 | ULIST_ITER_INIT(&uiter); | |
712 | while ((node = ulist_next(ulist, &uiter))) | |
713 | free_inode_elem_list(unode_aux_to_inode_list(node)); | |
714 | ||
715 | ulist_free(ulist); | |
716 | } | |
717 | ||
8da6d581 | 718 | /* |
52042d8e | 719 | * We maintain three separate rbtrees: one for direct refs, one for |
86d5f994 EN |
720 | * indirect refs which have a key, and one for indirect refs which do not |
721 | * have a key. Each tree does merge on insertion. | |
722 | * | |
723 | * Once all of the references are located, we iterate over the tree of | |
724 | * indirect refs with missing keys. An appropriate key is located and | |
725 | * the ref is moved onto the tree for indirect refs. After all missing | |
726 | * keys are thus located, we iterate over the indirect ref tree, resolve | |
727 | * each reference, and then insert the resolved reference onto the | |
728 | * direct tree (merging there too). | |
729 | * | |
730 | * New backrefs (i.e., for parent nodes) are added to the appropriate | |
731 | * rbtree as they are encountered. The new backrefs are subsequently | |
732 | * resolved as above. | |
8da6d581 | 733 | */ |
a2c8d27e FM |
734 | static int resolve_indirect_refs(struct btrfs_backref_walk_ctx *ctx, |
735 | struct btrfs_path *path, | |
86d5f994 | 736 | struct preftrees *preftrees, |
6ce6ba53 | 737 | struct share_check *sc) |
8da6d581 JS |
738 | { |
739 | int err; | |
740 | int ret = 0; | |
8da6d581 JS |
741 | struct ulist *parents; |
742 | struct ulist_node *node; | |
cd1b413c | 743 | struct ulist_iterator uiter; |
86d5f994 | 744 | struct rb_node *rnode; |
8da6d581 JS |
745 | |
746 | parents = ulist_alloc(GFP_NOFS); | |
747 | if (!parents) | |
748 | return -ENOMEM; | |
749 | ||
750 | /* | |
86d5f994 EN |
751 | * We could trade memory usage for performance here by iterating |
752 | * the tree, allocating new refs for each insertion, and then | |
753 | * freeing the entire indirect tree when we're done. In some test | |
754 | * cases, the tree can grow quite large (~200k objects). | |
8da6d581 | 755 | */ |
ecf160b4 | 756 | while ((rnode = rb_first_cached(&preftrees->indirect.root))) { |
86d5f994 EN |
757 | struct prelim_ref *ref; |
758 | ||
759 | ref = rb_entry(rnode, struct prelim_ref, rbnode); | |
760 | if (WARN(ref->parent, | |
761 | "BUG: direct ref found in indirect tree")) { | |
762 | ret = -EINVAL; | |
763 | goto out; | |
764 | } | |
765 | ||
ecf160b4 | 766 | rb_erase_cached(&ref->rbnode, &preftrees->indirect.root); |
6c336b21 | 767 | preftrees->indirect.count--; |
86d5f994 EN |
768 | |
769 | if (ref->count == 0) { | |
770 | free_pref(ref); | |
8da6d581 | 771 | continue; |
86d5f994 EN |
772 | } |
773 | ||
877c1476 | 774 | if (sc && ref->root_id != sc->root->root_key.objectid) { |
86d5f994 | 775 | free_pref(ref); |
dc046b10 JB |
776 | ret = BACKREF_FOUND_SHARED; |
777 | goto out; | |
778 | } | |
a2c8d27e | 779 | err = resolve_indirect_ref(ctx, path, preftrees, ref, parents); |
95def2ed WS |
780 | /* |
781 | * we can only tolerate ENOENT,otherwise,we should catch error | |
782 | * and return directly. | |
783 | */ | |
784 | if (err == -ENOENT) { | |
a2c8d27e | 785 | prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, |
3ec4d323 | 786 | NULL); |
8da6d581 | 787 | continue; |
95def2ed | 788 | } else if (err) { |
86d5f994 | 789 | free_pref(ref); |
95def2ed WS |
790 | ret = err; |
791 | goto out; | |
792 | } | |
8da6d581 JS |
793 | |
794 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
795 | ULIST_ITER_INIT(&uiter); |
796 | node = ulist_next(parents, &uiter); | |
8da6d581 | 797 | ref->parent = node ? node->val : 0; |
4dae077a | 798 | ref->inode_list = unode_aux_to_inode_list(node); |
8da6d581 | 799 | |
86d5f994 | 800 | /* Add a prelim_ref(s) for any other parent(s). */ |
cd1b413c | 801 | while ((node = ulist_next(parents, &uiter))) { |
86d5f994 EN |
802 | struct prelim_ref *new_ref; |
803 | ||
b9e9a6cb WS |
804 | new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, |
805 | GFP_NOFS); | |
8da6d581 | 806 | if (!new_ref) { |
86d5f994 | 807 | free_pref(ref); |
8da6d581 | 808 | ret = -ENOMEM; |
e36902d4 | 809 | goto out; |
8da6d581 JS |
810 | } |
811 | memcpy(new_ref, ref, sizeof(*ref)); | |
812 | new_ref->parent = node->val; | |
4dae077a | 813 | new_ref->inode_list = unode_aux_to_inode_list(node); |
a2c8d27e | 814 | prelim_ref_insert(ctx->fs_info, &preftrees->direct, |
3ec4d323 | 815 | new_ref, NULL); |
8da6d581 | 816 | } |
86d5f994 | 817 | |
3ec4d323 | 818 | /* |
52042d8e | 819 | * Now it's a direct ref, put it in the direct tree. We must |
3ec4d323 EN |
820 | * do this last because the ref could be merged/freed here. |
821 | */ | |
a2c8d27e | 822 | prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, NULL); |
86d5f994 | 823 | |
8da6d581 | 824 | ulist_reinit(parents); |
9dd14fd6 | 825 | cond_resched(); |
8da6d581 | 826 | } |
e36902d4 | 827 | out: |
5614dc3a FM |
828 | /* |
829 | * We may have inode lists attached to refs in the parents ulist, so we | |
830 | * must free them before freeing the ulist and its refs. | |
831 | */ | |
832 | free_leaf_list(parents); | |
8da6d581 JS |
833 | return ret; |
834 | } | |
835 | ||
d5c88b73 JS |
836 | /* |
837 | * read tree blocks and add keys where required. | |
838 | */ | |
e0c476b1 | 839 | static int add_missing_keys(struct btrfs_fs_info *fs_info, |
38e3eebf | 840 | struct preftrees *preftrees, bool lock) |
d5c88b73 | 841 | { |
e0c476b1 | 842 | struct prelim_ref *ref; |
d5c88b73 | 843 | struct extent_buffer *eb; |
86d5f994 EN |
844 | struct preftree *tree = &preftrees->indirect_missing_keys; |
845 | struct rb_node *node; | |
d5c88b73 | 846 | |
ecf160b4 | 847 | while ((node = rb_first_cached(&tree->root))) { |
789d6a3a QW |
848 | struct btrfs_tree_parent_check check = { 0 }; |
849 | ||
86d5f994 | 850 | ref = rb_entry(node, struct prelim_ref, rbnode); |
ecf160b4 | 851 | rb_erase_cached(node, &tree->root); |
86d5f994 EN |
852 | |
853 | BUG_ON(ref->parent); /* should not be a direct ref */ | |
854 | BUG_ON(ref->key_for_search.type); | |
d5c88b73 | 855 | BUG_ON(!ref->wanted_disk_byte); |
86d5f994 | 856 | |
789d6a3a QW |
857 | check.level = ref->level - 1; |
858 | check.owner_root = ref->root_id; | |
859 | ||
860 | eb = read_tree_block(fs_info, ref->wanted_disk_byte, &check); | |
64c043de | 861 | if (IS_ERR(eb)) { |
86d5f994 | 862 | free_pref(ref); |
64c043de | 863 | return PTR_ERR(eb); |
4eb150d6 QW |
864 | } |
865 | if (!extent_buffer_uptodate(eb)) { | |
86d5f994 | 866 | free_pref(ref); |
416bc658 JB |
867 | free_extent_buffer(eb); |
868 | return -EIO; | |
869 | } | |
4eb150d6 | 870 | |
38e3eebf JB |
871 | if (lock) |
872 | btrfs_tree_read_lock(eb); | |
d5c88b73 JS |
873 | if (btrfs_header_level(eb) == 0) |
874 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
875 | else | |
876 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
38e3eebf JB |
877 | if (lock) |
878 | btrfs_tree_read_unlock(eb); | |
d5c88b73 | 879 | free_extent_buffer(eb); |
3ec4d323 | 880 | prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL); |
9dd14fd6 | 881 | cond_resched(); |
d5c88b73 JS |
882 | } |
883 | return 0; | |
884 | } | |
885 | ||
8da6d581 JS |
886 | /* |
887 | * add all currently queued delayed refs from this head whose seq nr is | |
888 | * smaller or equal that seq to the list | |
889 | */ | |
00142756 JM |
890 | static int add_delayed_refs(const struct btrfs_fs_info *fs_info, |
891 | struct btrfs_delayed_ref_head *head, u64 seq, | |
b25b0b87 | 892 | struct preftrees *preftrees, struct share_check *sc) |
8da6d581 | 893 | { |
c6fc2454 | 894 | struct btrfs_delayed_ref_node *node; |
d5c88b73 | 895 | struct btrfs_key key; |
0e0adbcf | 896 | struct rb_node *n; |
01747e92 | 897 | int count; |
b1375d64 | 898 | int ret = 0; |
8da6d581 | 899 | |
d7df2c79 | 900 | spin_lock(&head->lock); |
e3d03965 | 901 | for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) { |
0e0adbcf JB |
902 | node = rb_entry(n, struct btrfs_delayed_ref_node, |
903 | ref_node); | |
8da6d581 JS |
904 | if (node->seq > seq) |
905 | continue; | |
906 | ||
907 | switch (node->action) { | |
908 | case BTRFS_ADD_DELAYED_EXTENT: | |
909 | case BTRFS_UPDATE_DELAYED_HEAD: | |
910 | WARN_ON(1); | |
911 | continue; | |
912 | case BTRFS_ADD_DELAYED_REF: | |
01747e92 | 913 | count = node->ref_mod; |
8da6d581 JS |
914 | break; |
915 | case BTRFS_DROP_DELAYED_REF: | |
01747e92 | 916 | count = node->ref_mod * -1; |
8da6d581 JS |
917 | break; |
918 | default: | |
290342f6 | 919 | BUG(); |
8da6d581 JS |
920 | } |
921 | switch (node->type) { | |
922 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
86d5f994 | 923 | /* NORMAL INDIRECT METADATA backref */ |
8da6d581 | 924 | struct btrfs_delayed_tree_ref *ref; |
943553ef FM |
925 | struct btrfs_key *key_ptr = NULL; |
926 | ||
927 | if (head->extent_op && head->extent_op->update_key) { | |
928 | btrfs_disk_key_to_cpu(&key, &head->extent_op->key); | |
929 | key_ptr = &key; | |
930 | } | |
8da6d581 JS |
931 | |
932 | ref = btrfs_delayed_node_to_tree_ref(node); | |
00142756 | 933 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
943553ef | 934 | key_ptr, ref->level + 1, |
01747e92 EN |
935 | node->bytenr, count, sc, |
936 | GFP_ATOMIC); | |
8da6d581 JS |
937 | break; |
938 | } | |
939 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
86d5f994 | 940 | /* SHARED DIRECT METADATA backref */ |
8da6d581 JS |
941 | struct btrfs_delayed_tree_ref *ref; |
942 | ||
943 | ref = btrfs_delayed_node_to_tree_ref(node); | |
86d5f994 | 944 | |
01747e92 EN |
945 | ret = add_direct_ref(fs_info, preftrees, ref->level + 1, |
946 | ref->parent, node->bytenr, count, | |
3ec4d323 | 947 | sc, GFP_ATOMIC); |
8da6d581 JS |
948 | break; |
949 | } | |
950 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 951 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 | 952 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
953 | ref = btrfs_delayed_node_to_data_ref(node); |
954 | ||
955 | key.objectid = ref->objectid; | |
956 | key.type = BTRFS_EXTENT_DATA_KEY; | |
957 | key.offset = ref->offset; | |
dc046b10 JB |
958 | |
959 | /* | |
4fc7b572 FM |
960 | * If we have a share check context and a reference for |
961 | * another inode, we can't exit immediately. This is | |
962 | * because even if this is a BTRFS_ADD_DELAYED_REF | |
963 | * reference we may find next a BTRFS_DROP_DELAYED_REF | |
964 | * which cancels out this ADD reference. | |
965 | * | |
966 | * If this is a DROP reference and there was no previous | |
967 | * ADD reference, then we need to signal that when we | |
968 | * process references from the extent tree (through | |
969 | * add_inline_refs() and add_keyed_refs()), we should | |
970 | * not exit early if we find a reference for another | |
971 | * inode, because one of the delayed DROP references | |
972 | * may cancel that reference in the extent tree. | |
dc046b10 | 973 | */ |
4fc7b572 FM |
974 | if (sc && count < 0) |
975 | sc->have_delayed_delete_refs = true; | |
dc046b10 | 976 | |
00142756 | 977 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
01747e92 EN |
978 | &key, 0, node->bytenr, count, sc, |
979 | GFP_ATOMIC); | |
8da6d581 JS |
980 | break; |
981 | } | |
982 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 983 | /* SHARED DIRECT FULL backref */ |
8da6d581 | 984 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
985 | |
986 | ref = btrfs_delayed_node_to_data_ref(node); | |
86d5f994 | 987 | |
01747e92 EN |
988 | ret = add_direct_ref(fs_info, preftrees, 0, ref->parent, |
989 | node->bytenr, count, sc, | |
990 | GFP_ATOMIC); | |
8da6d581 JS |
991 | break; |
992 | } | |
993 | default: | |
994 | WARN_ON(1); | |
995 | } | |
3ec4d323 EN |
996 | /* |
997 | * We must ignore BACKREF_FOUND_SHARED until all delayed | |
998 | * refs have been checked. | |
999 | */ | |
1000 | if (ret && (ret != BACKREF_FOUND_SHARED)) | |
d7df2c79 | 1001 | break; |
8da6d581 | 1002 | } |
3ec4d323 EN |
1003 | if (!ret) |
1004 | ret = extent_is_shared(sc); | |
4fc7b572 | 1005 | |
d7df2c79 JB |
1006 | spin_unlock(&head->lock); |
1007 | return ret; | |
8da6d581 JS |
1008 | } |
1009 | ||
1010 | /* | |
1011 | * add all inline backrefs for bytenr to the list | |
3ec4d323 EN |
1012 | * |
1013 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 1014 | */ |
f73853c7 FM |
1015 | static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx, |
1016 | struct btrfs_path *path, | |
86d5f994 | 1017 | int *info_level, struct preftrees *preftrees, |
b25b0b87 | 1018 | struct share_check *sc) |
8da6d581 | 1019 | { |
b1375d64 | 1020 | int ret = 0; |
8da6d581 JS |
1021 | int slot; |
1022 | struct extent_buffer *leaf; | |
1023 | struct btrfs_key key; | |
261c84b6 | 1024 | struct btrfs_key found_key; |
8da6d581 JS |
1025 | unsigned long ptr; |
1026 | unsigned long end; | |
1027 | struct btrfs_extent_item *ei; | |
1028 | u64 flags; | |
1029 | u64 item_size; | |
1030 | ||
1031 | /* | |
1032 | * enumerate all inline refs | |
1033 | */ | |
1034 | leaf = path->nodes[0]; | |
dadcaf78 | 1035 | slot = path->slots[0]; |
8da6d581 | 1036 | |
3212fa14 | 1037 | item_size = btrfs_item_size(leaf, slot); |
8da6d581 JS |
1038 | BUG_ON(item_size < sizeof(*ei)); |
1039 | ||
1040 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
f73853c7 FM |
1041 | |
1042 | if (ctx->check_extent_item) { | |
1043 | ret = ctx->check_extent_item(ctx->bytenr, ei, leaf, ctx->user_ctx); | |
1044 | if (ret) | |
1045 | return ret; | |
1046 | } | |
1047 | ||
8da6d581 | 1048 | flags = btrfs_extent_flags(leaf, ei); |
261c84b6 | 1049 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
8da6d581 JS |
1050 | |
1051 | ptr = (unsigned long)(ei + 1); | |
1052 | end = (unsigned long)ei + item_size; | |
1053 | ||
261c84b6 JB |
1054 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && |
1055 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
8da6d581 | 1056 | struct btrfs_tree_block_info *info; |
8da6d581 JS |
1057 | |
1058 | info = (struct btrfs_tree_block_info *)ptr; | |
1059 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
1060 | ptr += sizeof(struct btrfs_tree_block_info); |
1061 | BUG_ON(ptr > end); | |
261c84b6 JB |
1062 | } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { |
1063 | *info_level = found_key.offset; | |
8da6d581 JS |
1064 | } else { |
1065 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1066 | } | |
1067 | ||
1068 | while (ptr < end) { | |
1069 | struct btrfs_extent_inline_ref *iref; | |
1070 | u64 offset; | |
1071 | int type; | |
1072 | ||
1073 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1074 | type = btrfs_get_extent_inline_ref_type(leaf, iref, |
1075 | BTRFS_REF_TYPE_ANY); | |
1076 | if (type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 1077 | return -EUCLEAN; |
3de28d57 | 1078 | |
8da6d581 JS |
1079 | offset = btrfs_extent_inline_ref_offset(leaf, iref); |
1080 | ||
1081 | switch (type) { | |
1082 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
f73853c7 | 1083 | ret = add_direct_ref(ctx->fs_info, preftrees, |
00142756 | 1084 | *info_level + 1, offset, |
f73853c7 | 1085 | ctx->bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1086 | break; |
1087 | case BTRFS_SHARED_DATA_REF_KEY: { | |
1088 | struct btrfs_shared_data_ref *sdref; | |
1089 | int count; | |
1090 | ||
1091 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1092 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
86d5f994 | 1093 | |
f73853c7 FM |
1094 | ret = add_direct_ref(ctx->fs_info, preftrees, 0, offset, |
1095 | ctx->bytenr, count, sc, GFP_NOFS); | |
8da6d581 JS |
1096 | break; |
1097 | } | |
1098 | case BTRFS_TREE_BLOCK_REF_KEY: | |
f73853c7 | 1099 | ret = add_indirect_ref(ctx->fs_info, preftrees, offset, |
00142756 | 1100 | NULL, *info_level + 1, |
f73853c7 | 1101 | ctx->bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1102 | break; |
1103 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
1104 | struct btrfs_extent_data_ref *dref; | |
1105 | int count; | |
1106 | u64 root; | |
1107 | ||
1108 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1109 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1110 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1111 | dref); | |
1112 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1113 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1114 | |
a0a5472a | 1115 | if (sc && key.objectid != sc->inum && |
4fc7b572 | 1116 | !sc->have_delayed_delete_refs) { |
dc046b10 JB |
1117 | ret = BACKREF_FOUND_SHARED; |
1118 | break; | |
1119 | } | |
1120 | ||
8da6d581 | 1121 | root = btrfs_extent_data_ref_root(leaf, dref); |
86d5f994 | 1122 | |
adf02418 FM |
1123 | if (!ctx->skip_data_ref || |
1124 | !ctx->skip_data_ref(root, key.objectid, key.offset, | |
1125 | ctx->user_ctx)) | |
1126 | ret = add_indirect_ref(ctx->fs_info, preftrees, | |
1127 | root, &key, 0, ctx->bytenr, | |
1128 | count, sc, GFP_NOFS); | |
8da6d581 JS |
1129 | break; |
1130 | } | |
1131 | default: | |
1132 | WARN_ON(1); | |
1133 | } | |
1149ab6b WS |
1134 | if (ret) |
1135 | return ret; | |
8da6d581 JS |
1136 | ptr += btrfs_extent_inline_ref_size(type); |
1137 | } | |
1138 | ||
1139 | return 0; | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * add all non-inline backrefs for bytenr to the list | |
3ec4d323 EN |
1144 | * |
1145 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 1146 | */ |
adf02418 FM |
1147 | static int add_keyed_refs(struct btrfs_backref_walk_ctx *ctx, |
1148 | struct btrfs_root *extent_root, | |
1149 | struct btrfs_path *path, | |
86d5f994 | 1150 | int info_level, struct preftrees *preftrees, |
3ec4d323 | 1151 | struct share_check *sc) |
8da6d581 | 1152 | { |
98cc4222 | 1153 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
8da6d581 JS |
1154 | int ret; |
1155 | int slot; | |
1156 | struct extent_buffer *leaf; | |
1157 | struct btrfs_key key; | |
1158 | ||
1159 | while (1) { | |
1160 | ret = btrfs_next_item(extent_root, path); | |
1161 | if (ret < 0) | |
1162 | break; | |
1163 | if (ret) { | |
1164 | ret = 0; | |
1165 | break; | |
1166 | } | |
1167 | ||
1168 | slot = path->slots[0]; | |
1169 | leaf = path->nodes[0]; | |
1170 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1171 | ||
adf02418 | 1172 | if (key.objectid != ctx->bytenr) |
8da6d581 JS |
1173 | break; |
1174 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
1175 | continue; | |
1176 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
1177 | break; | |
1178 | ||
1179 | switch (key.type) { | |
1180 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
86d5f994 | 1181 | /* SHARED DIRECT METADATA backref */ |
00142756 JM |
1182 | ret = add_direct_ref(fs_info, preftrees, |
1183 | info_level + 1, key.offset, | |
adf02418 | 1184 | ctx->bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1185 | break; |
1186 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 1187 | /* SHARED DIRECT FULL backref */ |
8da6d581 JS |
1188 | struct btrfs_shared_data_ref *sdref; |
1189 | int count; | |
1190 | ||
1191 | sdref = btrfs_item_ptr(leaf, slot, | |
1192 | struct btrfs_shared_data_ref); | |
1193 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
00142756 | 1194 | ret = add_direct_ref(fs_info, preftrees, 0, |
adf02418 | 1195 | key.offset, ctx->bytenr, count, |
3ec4d323 | 1196 | sc, GFP_NOFS); |
8da6d581 JS |
1197 | break; |
1198 | } | |
1199 | case BTRFS_TREE_BLOCK_REF_KEY: | |
86d5f994 | 1200 | /* NORMAL INDIRECT METADATA backref */ |
00142756 | 1201 | ret = add_indirect_ref(fs_info, preftrees, key.offset, |
adf02418 | 1202 | NULL, info_level + 1, ctx->bytenr, |
3ec4d323 | 1203 | 1, NULL, GFP_NOFS); |
8da6d581 JS |
1204 | break; |
1205 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 1206 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 JS |
1207 | struct btrfs_extent_data_ref *dref; |
1208 | int count; | |
1209 | u64 root; | |
1210 | ||
1211 | dref = btrfs_item_ptr(leaf, slot, | |
1212 | struct btrfs_extent_data_ref); | |
1213 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1214 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1215 | dref); | |
1216 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1217 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1218 | |
a0a5472a | 1219 | if (sc && key.objectid != sc->inum && |
4fc7b572 | 1220 | !sc->have_delayed_delete_refs) { |
dc046b10 JB |
1221 | ret = BACKREF_FOUND_SHARED; |
1222 | break; | |
1223 | } | |
1224 | ||
8da6d581 | 1225 | root = btrfs_extent_data_ref_root(leaf, dref); |
adf02418 FM |
1226 | |
1227 | if (!ctx->skip_data_ref || | |
1228 | !ctx->skip_data_ref(root, key.objectid, key.offset, | |
1229 | ctx->user_ctx)) | |
1230 | ret = add_indirect_ref(fs_info, preftrees, root, | |
1231 | &key, 0, ctx->bytenr, | |
1232 | count, sc, GFP_NOFS); | |
8da6d581 JS |
1233 | break; |
1234 | } | |
1235 | default: | |
1236 | WARN_ON(1); | |
1237 | } | |
1149ab6b WS |
1238 | if (ret) |
1239 | return ret; | |
1240 | ||
8da6d581 JS |
1241 | } |
1242 | ||
1243 | return ret; | |
1244 | } | |
1245 | ||
583f4ac5 FM |
1246 | /* |
1247 | * The caller has joined a transaction or is holding a read lock on the | |
1248 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1249 | * snapshot field changing while updating or checking the cache. | |
1250 | */ | |
1251 | static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, | |
1252 | struct btrfs_root *root, | |
1253 | u64 bytenr, int level, bool *is_shared) | |
1254 | { | |
4e4488d4 | 1255 | const struct btrfs_fs_info *fs_info = root->fs_info; |
583f4ac5 FM |
1256 | struct btrfs_backref_shared_cache_entry *entry; |
1257 | ||
4e4488d4 FM |
1258 | if (!current->journal_info) |
1259 | lockdep_assert_held(&fs_info->commit_root_sem); | |
1260 | ||
583f4ac5 FM |
1261 | if (!ctx->use_path_cache) |
1262 | return false; | |
1263 | ||
1264 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1265 | return false; | |
1266 | ||
1267 | /* | |
1268 | * Level -1 is used for the data extent, which is not reliable to cache | |
1269 | * because its reference count can increase or decrease without us | |
1270 | * realizing. We cache results only for extent buffers that lead from | |
1271 | * the root node down to the leaf with the file extent item. | |
1272 | */ | |
1273 | ASSERT(level >= 0); | |
1274 | ||
1275 | entry = &ctx->path_cache_entries[level]; | |
1276 | ||
1277 | /* Unused cache entry or being used for some other extent buffer. */ | |
1278 | if (entry->bytenr != bytenr) | |
1279 | return false; | |
1280 | ||
1281 | /* | |
1282 | * We cached a false result, but the last snapshot generation of the | |
1283 | * root changed, so we now have a snapshot. Don't trust the result. | |
1284 | */ | |
1285 | if (!entry->is_shared && | |
1286 | entry->gen != btrfs_root_last_snapshot(&root->root_item)) | |
1287 | return false; | |
1288 | ||
1289 | /* | |
1290 | * If we cached a true result and the last generation used for dropping | |
1291 | * a root changed, we can not trust the result, because the dropped root | |
1292 | * could be a snapshot sharing this extent buffer. | |
1293 | */ | |
1294 | if (entry->is_shared && | |
4e4488d4 | 1295 | entry->gen != btrfs_get_last_root_drop_gen(fs_info)) |
583f4ac5 FM |
1296 | return false; |
1297 | ||
1298 | *is_shared = entry->is_shared; | |
1299 | /* | |
1300 | * If the node at this level is shared, than all nodes below are also | |
1301 | * shared. Currently some of the nodes below may be marked as not shared | |
1302 | * because we have just switched from one leaf to another, and switched | |
1303 | * also other nodes above the leaf and below the current level, so mark | |
1304 | * them as shared. | |
1305 | */ | |
1306 | if (*is_shared) { | |
1307 | for (int i = 0; i < level; i++) { | |
1308 | ctx->path_cache_entries[i].is_shared = true; | |
1309 | ctx->path_cache_entries[i].gen = entry->gen; | |
1310 | } | |
1311 | } | |
1312 | ||
1313 | return true; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * The caller has joined a transaction or is holding a read lock on the | |
1318 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1319 | * snapshot field changing while updating or checking the cache. | |
1320 | */ | |
1321 | static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, | |
1322 | struct btrfs_root *root, | |
1323 | u64 bytenr, int level, bool is_shared) | |
1324 | { | |
4e4488d4 | 1325 | const struct btrfs_fs_info *fs_info = root->fs_info; |
583f4ac5 FM |
1326 | struct btrfs_backref_shared_cache_entry *entry; |
1327 | u64 gen; | |
1328 | ||
4e4488d4 FM |
1329 | if (!current->journal_info) |
1330 | lockdep_assert_held(&fs_info->commit_root_sem); | |
1331 | ||
583f4ac5 FM |
1332 | if (!ctx->use_path_cache) |
1333 | return; | |
1334 | ||
1335 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1336 | return; | |
1337 | ||
1338 | /* | |
1339 | * Level -1 is used for the data extent, which is not reliable to cache | |
1340 | * because its reference count can increase or decrease without us | |
1341 | * realizing. We cache results only for extent buffers that lead from | |
1342 | * the root node down to the leaf with the file extent item. | |
1343 | */ | |
1344 | ASSERT(level >= 0); | |
1345 | ||
1346 | if (is_shared) | |
4e4488d4 | 1347 | gen = btrfs_get_last_root_drop_gen(fs_info); |
583f4ac5 FM |
1348 | else |
1349 | gen = btrfs_root_last_snapshot(&root->root_item); | |
1350 | ||
1351 | entry = &ctx->path_cache_entries[level]; | |
1352 | entry->bytenr = bytenr; | |
1353 | entry->is_shared = is_shared; | |
1354 | entry->gen = gen; | |
1355 | ||
1356 | /* | |
1357 | * If we found an extent buffer is shared, set the cache result for all | |
1358 | * extent buffers below it to true. As nodes in the path are COWed, | |
1359 | * their sharedness is moved to their children, and if a leaf is COWed, | |
1360 | * then the sharedness of a data extent becomes direct, the refcount of | |
1361 | * data extent is increased in the extent item at the extent tree. | |
1362 | */ | |
1363 | if (is_shared) { | |
1364 | for (int i = 0; i < level; i++) { | |
1365 | entry = &ctx->path_cache_entries[i]; | |
1366 | entry->is_shared = is_shared; | |
1367 | entry->gen = gen; | |
1368 | } | |
1369 | } | |
1370 | } | |
1371 | ||
8da6d581 JS |
1372 | /* |
1373 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
1374 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
1375 | * indirect refs to their parent bytenr. | |
1376 | * When roots are found, they're added to the roots list | |
1377 | * | |
a2c8d27e FM |
1378 | * @ctx: Backref walking context object, must be not NULL. |
1379 | * @sc: If !NULL, then immediately return BACKREF_FOUND_SHARED when a | |
1380 | * shared extent is detected. | |
3ec4d323 EN |
1381 | * |
1382 | * Otherwise this returns 0 for success and <0 for an error. | |
1383 | * | |
8da6d581 JS |
1384 | * FIXME some caching might speed things up |
1385 | */ | |
a2c8d27e | 1386 | static int find_parent_nodes(struct btrfs_backref_walk_ctx *ctx, |
6ce6ba53 | 1387 | struct share_check *sc) |
8da6d581 | 1388 | { |
a2c8d27e | 1389 | struct btrfs_root *root = btrfs_extent_root(ctx->fs_info, ctx->bytenr); |
8da6d581 JS |
1390 | struct btrfs_key key; |
1391 | struct btrfs_path *path; | |
8da6d581 | 1392 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 1393 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
1394 | int info_level = 0; |
1395 | int ret; | |
e0c476b1 | 1396 | struct prelim_ref *ref; |
86d5f994 | 1397 | struct rb_node *node; |
f05c4746 | 1398 | struct extent_inode_elem *eie = NULL; |
86d5f994 EN |
1399 | struct preftrees preftrees = { |
1400 | .direct = PREFTREE_INIT, | |
1401 | .indirect = PREFTREE_INIT, | |
1402 | .indirect_missing_keys = PREFTREE_INIT | |
1403 | }; | |
8da6d581 | 1404 | |
56f5c199 FM |
1405 | /* Roots ulist is not needed when using a sharedness check context. */ |
1406 | if (sc) | |
a2c8d27e | 1407 | ASSERT(ctx->roots == NULL); |
56f5c199 | 1408 | |
a2c8d27e | 1409 | key.objectid = ctx->bytenr; |
8da6d581 | 1410 | key.offset = (u64)-1; |
a2c8d27e | 1411 | if (btrfs_fs_incompat(ctx->fs_info, SKINNY_METADATA)) |
261c84b6 JB |
1412 | key.type = BTRFS_METADATA_ITEM_KEY; |
1413 | else | |
1414 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
8da6d581 JS |
1415 | |
1416 | path = btrfs_alloc_path(); | |
1417 | if (!path) | |
1418 | return -ENOMEM; | |
a2c8d27e | 1419 | if (!ctx->trans) { |
da61d31a | 1420 | path->search_commit_root = 1; |
e84752d4 WS |
1421 | path->skip_locking = 1; |
1422 | } | |
8da6d581 | 1423 | |
a2c8d27e | 1424 | if (ctx->time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
1425 | path->skip_locking = 1; |
1426 | ||
8da6d581 | 1427 | again: |
d3b01064 LZ |
1428 | head = NULL; |
1429 | ||
98cc4222 | 1430 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8da6d581 JS |
1431 | if (ret < 0) |
1432 | goto out; | |
fcba0120 JB |
1433 | if (ret == 0) { |
1434 | /* This shouldn't happen, indicates a bug or fs corruption. */ | |
1435 | ASSERT(ret != 0); | |
1436 | ret = -EUCLEAN; | |
1437 | goto out; | |
1438 | } | |
8da6d581 | 1439 | |
a2c8d27e FM |
1440 | if (ctx->trans && likely(ctx->trans->type != __TRANS_DUMMY) && |
1441 | ctx->time_seq != BTRFS_SEQ_LAST) { | |
7a3ae2f8 | 1442 | /* |
9665ebd5 JB |
1443 | * We have a specific time_seq we care about and trans which |
1444 | * means we have the path lock, we need to grab the ref head and | |
1445 | * lock it so we have a consistent view of the refs at the given | |
1446 | * time. | |
7a3ae2f8 | 1447 | */ |
a2c8d27e | 1448 | delayed_refs = &ctx->trans->transaction->delayed_refs; |
7a3ae2f8 | 1449 | spin_lock(&delayed_refs->lock); |
a2c8d27e | 1450 | head = btrfs_find_delayed_ref_head(delayed_refs, ctx->bytenr); |
7a3ae2f8 JS |
1451 | if (head) { |
1452 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 1453 | refcount_inc(&head->refs); |
7a3ae2f8 JS |
1454 | spin_unlock(&delayed_refs->lock); |
1455 | ||
1456 | btrfs_release_path(path); | |
1457 | ||
1458 | /* | |
1459 | * Mutex was contended, block until it's | |
1460 | * released and try again | |
1461 | */ | |
1462 | mutex_lock(&head->mutex); | |
1463 | mutex_unlock(&head->mutex); | |
d278850e | 1464 | btrfs_put_delayed_ref_head(head); |
7a3ae2f8 JS |
1465 | goto again; |
1466 | } | |
d7df2c79 | 1467 | spin_unlock(&delayed_refs->lock); |
a2c8d27e | 1468 | ret = add_delayed_refs(ctx->fs_info, head, ctx->time_seq, |
b25b0b87 | 1469 | &preftrees, sc); |
155725c9 | 1470 | mutex_unlock(&head->mutex); |
d7df2c79 | 1471 | if (ret) |
7a3ae2f8 | 1472 | goto out; |
d7df2c79 JB |
1473 | } else { |
1474 | spin_unlock(&delayed_refs->lock); | |
d3b01064 | 1475 | } |
8da6d581 | 1476 | } |
8da6d581 JS |
1477 | |
1478 | if (path->slots[0]) { | |
1479 | struct extent_buffer *leaf; | |
1480 | int slot; | |
1481 | ||
dadcaf78 | 1482 | path->slots[0]--; |
8da6d581 | 1483 | leaf = path->nodes[0]; |
dadcaf78 | 1484 | slot = path->slots[0]; |
8da6d581 | 1485 | btrfs_item_key_to_cpu(leaf, &key, slot); |
a2c8d27e | 1486 | if (key.objectid == ctx->bytenr && |
261c84b6 JB |
1487 | (key.type == BTRFS_EXTENT_ITEM_KEY || |
1488 | key.type == BTRFS_METADATA_ITEM_KEY)) { | |
f73853c7 FM |
1489 | ret = add_inline_refs(ctx, path, &info_level, |
1490 | &preftrees, sc); | |
8da6d581 JS |
1491 | if (ret) |
1492 | goto out; | |
adf02418 | 1493 | ret = add_keyed_refs(ctx, root, path, info_level, |
3ec4d323 | 1494 | &preftrees, sc); |
8da6d581 JS |
1495 | if (ret) |
1496 | goto out; | |
1497 | } | |
1498 | } | |
8da6d581 | 1499 | |
56f5c199 FM |
1500 | /* |
1501 | * If we have a share context and we reached here, it means the extent | |
1502 | * is not directly shared (no multiple reference items for it), | |
1503 | * otherwise we would have exited earlier with a return value of | |
1504 | * BACKREF_FOUND_SHARED after processing delayed references or while | |
1505 | * processing inline or keyed references from the extent tree. | |
1506 | * The extent may however be indirectly shared through shared subtrees | |
1507 | * as a result from creating snapshots, so we determine below what is | |
1508 | * its parent node, in case we are dealing with a metadata extent, or | |
1509 | * what's the leaf (or leaves), from a fs tree, that has a file extent | |
1510 | * item pointing to it in case we are dealing with a data extent. | |
1511 | */ | |
1512 | ASSERT(extent_is_shared(sc) == 0); | |
1513 | ||
877c1476 FM |
1514 | /* |
1515 | * If we are here for a data extent and we have a share_check structure | |
1516 | * it means the data extent is not directly shared (does not have | |
1517 | * multiple reference items), so we have to check if a path in the fs | |
1518 | * tree (going from the root node down to the leaf that has the file | |
1519 | * extent item pointing to the data extent) is shared, that is, if any | |
1520 | * of the extent buffers in the path is referenced by other trees. | |
1521 | */ | |
a2c8d27e | 1522 | if (sc && ctx->bytenr == sc->data_bytenr) { |
6976201f FM |
1523 | /* |
1524 | * If our data extent is from a generation more recent than the | |
1525 | * last generation used to snapshot the root, then we know that | |
1526 | * it can not be shared through subtrees, so we can skip | |
1527 | * resolving indirect references, there's no point in | |
1528 | * determining the extent buffers for the path from the fs tree | |
1529 | * root node down to the leaf that has the file extent item that | |
1530 | * points to the data extent. | |
1531 | */ | |
1532 | if (sc->data_extent_gen > | |
1533 | btrfs_root_last_snapshot(&sc->root->root_item)) { | |
1534 | ret = BACKREF_FOUND_NOT_SHARED; | |
1535 | goto out; | |
1536 | } | |
1537 | ||
877c1476 FM |
1538 | /* |
1539 | * If we are only determining if a data extent is shared or not | |
1540 | * and the corresponding file extent item is located in the same | |
1541 | * leaf as the previous file extent item, we can skip resolving | |
1542 | * indirect references for a data extent, since the fs tree path | |
1543 | * is the same (same leaf, so same path). We skip as long as the | |
1544 | * cached result for the leaf is valid and only if there's only | |
1545 | * one file extent item pointing to the data extent, because in | |
1546 | * the case of multiple file extent items, they may be located | |
1547 | * in different leaves and therefore we have multiple paths. | |
1548 | */ | |
1549 | if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr && | |
1550 | sc->self_ref_count == 1) { | |
1551 | bool cached; | |
1552 | bool is_shared; | |
1553 | ||
1554 | cached = lookup_backref_shared_cache(sc->ctx, sc->root, | |
1555 | sc->ctx->curr_leaf_bytenr, | |
1556 | 0, &is_shared); | |
1557 | if (cached) { | |
1558 | if (is_shared) | |
1559 | ret = BACKREF_FOUND_SHARED; | |
1560 | else | |
1561 | ret = BACKREF_FOUND_NOT_SHARED; | |
1562 | goto out; | |
1563 | } | |
1564 | } | |
1565 | } | |
1566 | ||
86d5f994 | 1567 | btrfs_release_path(path); |
8da6d581 | 1568 | |
a2c8d27e | 1569 | ret = add_missing_keys(ctx->fs_info, &preftrees, path->skip_locking == 0); |
d5c88b73 JS |
1570 | if (ret) |
1571 | goto out; | |
1572 | ||
ecf160b4 | 1573 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root)); |
8da6d581 | 1574 | |
a2c8d27e | 1575 | ret = resolve_indirect_refs(ctx, path, &preftrees, sc); |
8da6d581 JS |
1576 | if (ret) |
1577 | goto out; | |
1578 | ||
ecf160b4 | 1579 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root)); |
8da6d581 | 1580 | |
86d5f994 EN |
1581 | /* |
1582 | * This walks the tree of merged and resolved refs. Tree blocks are | |
1583 | * read in as needed. Unique entries are added to the ulist, and | |
1584 | * the list of found roots is updated. | |
1585 | * | |
1586 | * We release the entire tree in one go before returning. | |
1587 | */ | |
ecf160b4 | 1588 | node = rb_first_cached(&preftrees.direct.root); |
86d5f994 EN |
1589 | while (node) { |
1590 | ref = rb_entry(node, struct prelim_ref, rbnode); | |
1591 | node = rb_next(&ref->rbnode); | |
c8195a7b ZB |
1592 | /* |
1593 | * ref->count < 0 can happen here if there are delayed | |
1594 | * refs with a node->action of BTRFS_DROP_DELAYED_REF. | |
1595 | * prelim_ref_insert() relies on this when merging | |
1596 | * identical refs to keep the overall count correct. | |
1597 | * prelim_ref_insert() will merge only those refs | |
1598 | * which compare identically. Any refs having | |
1599 | * e.g. different offsets would not be merged, | |
1600 | * and would retain their original ref->count < 0. | |
1601 | */ | |
a2c8d27e | 1602 | if (ctx->roots && ref->count && ref->root_id && ref->parent == 0) { |
8da6d581 | 1603 | /* no parent == root of tree */ |
a2c8d27e | 1604 | ret = ulist_add(ctx->roots, ref->root_id, 0, GFP_NOFS); |
f1723939 WS |
1605 | if (ret < 0) |
1606 | goto out; | |
8da6d581 JS |
1607 | } |
1608 | if (ref->count && ref->parent) { | |
a2c8d27e FM |
1609 | if (!ctx->ignore_extent_item_pos && !ref->inode_list && |
1610 | ref->level == 0) { | |
789d6a3a | 1611 | struct btrfs_tree_parent_check check = { 0 }; |
976b1908 | 1612 | struct extent_buffer *eb; |
707e8a07 | 1613 | |
789d6a3a QW |
1614 | check.level = ref->level; |
1615 | ||
1616 | eb = read_tree_block(ctx->fs_info, ref->parent, | |
1617 | &check); | |
64c043de LB |
1618 | if (IS_ERR(eb)) { |
1619 | ret = PTR_ERR(eb); | |
1620 | goto out; | |
4eb150d6 QW |
1621 | } |
1622 | if (!extent_buffer_uptodate(eb)) { | |
416bc658 | 1623 | free_extent_buffer(eb); |
c16c2e2e WS |
1624 | ret = -EIO; |
1625 | goto out; | |
416bc658 | 1626 | } |
38e3eebf | 1627 | |
ac5887c8 | 1628 | if (!path->skip_locking) |
38e3eebf | 1629 | btrfs_tree_read_lock(eb); |
88ffb665 | 1630 | ret = find_extent_in_eb(ctx, eb, &eie); |
38e3eebf | 1631 | if (!path->skip_locking) |
ac5887c8 | 1632 | btrfs_tree_read_unlock(eb); |
976b1908 | 1633 | free_extent_buffer(eb); |
88ffb665 FM |
1634 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || |
1635 | ret < 0) | |
f5929cd8 FDBM |
1636 | goto out; |
1637 | ref->inode_list = eie; | |
92876eec FM |
1638 | /* |
1639 | * We transferred the list ownership to the ref, | |
1640 | * so set to NULL to avoid a double free in case | |
1641 | * an error happens after this. | |
1642 | */ | |
1643 | eie = NULL; | |
976b1908 | 1644 | } |
a2c8d27e | 1645 | ret = ulist_add_merge_ptr(ctx->refs, ref->parent, |
4eb1f66d TI |
1646 | ref->inode_list, |
1647 | (void **)&eie, GFP_NOFS); | |
f1723939 WS |
1648 | if (ret < 0) |
1649 | goto out; | |
a2c8d27e | 1650 | if (!ret && !ctx->ignore_extent_item_pos) { |
3301958b | 1651 | /* |
9f05c09d JB |
1652 | * We've recorded that parent, so we must extend |
1653 | * its inode list here. | |
1654 | * | |
1655 | * However if there was corruption we may not | |
1656 | * have found an eie, return an error in this | |
1657 | * case. | |
3301958b | 1658 | */ |
9f05c09d JB |
1659 | ASSERT(eie); |
1660 | if (!eie) { | |
1661 | ret = -EUCLEAN; | |
1662 | goto out; | |
1663 | } | |
3301958b JS |
1664 | while (eie->next) |
1665 | eie = eie->next; | |
1666 | eie->next = ref->inode_list; | |
1667 | } | |
f05c4746 | 1668 | eie = NULL; |
92876eec FM |
1669 | /* |
1670 | * We have transferred the inode list ownership from | |
1671 | * this ref to the ref we added to the 'refs' ulist. | |
1672 | * So set this ref's inode list to NULL to avoid | |
1673 | * use-after-free when our caller uses it or double | |
1674 | * frees in case an error happens before we return. | |
1675 | */ | |
1676 | ref->inode_list = NULL; | |
8da6d581 | 1677 | } |
9dd14fd6 | 1678 | cond_resched(); |
8da6d581 JS |
1679 | } |
1680 | ||
1681 | out: | |
8da6d581 | 1682 | btrfs_free_path(path); |
86d5f994 EN |
1683 | |
1684 | prelim_release(&preftrees.direct); | |
1685 | prelim_release(&preftrees.indirect); | |
1686 | prelim_release(&preftrees.indirect_missing_keys); | |
1687 | ||
88ffb665 | 1688 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) |
f05c4746 | 1689 | free_inode_elem_list(eie); |
8da6d581 JS |
1690 | return ret; |
1691 | } | |
1692 | ||
1693 | /* | |
a2c8d27e FM |
1694 | * Finds all leaves with a reference to the specified combination of |
1695 | * @ctx->bytenr and @ctx->extent_item_pos. The bytenr of the found leaves are | |
1696 | * added to the ulist at @ctx->refs, and that ulist is allocated by this | |
1697 | * function. The caller should free the ulist with free_leaf_list() if | |
1698 | * @ctx->ignore_extent_item_pos is false, otherwise a fimple ulist_free() is | |
1699 | * enough. | |
8da6d581 | 1700 | * |
a2c8d27e | 1701 | * Returns 0 on success and < 0 on error. On error @ctx->refs is not allocated. |
8da6d581 | 1702 | */ |
a2c8d27e | 1703 | int btrfs_find_all_leafs(struct btrfs_backref_walk_ctx *ctx) |
8da6d581 | 1704 | { |
8da6d581 JS |
1705 | int ret; |
1706 | ||
a2c8d27e FM |
1707 | ASSERT(ctx->refs == NULL); |
1708 | ||
1709 | ctx->refs = ulist_alloc(GFP_NOFS); | |
1710 | if (!ctx->refs) | |
8da6d581 | 1711 | return -ENOMEM; |
8da6d581 | 1712 | |
a2c8d27e | 1713 | ret = find_parent_nodes(ctx, NULL); |
88ffb665 FM |
1714 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || |
1715 | (ret < 0 && ret != -ENOENT)) { | |
a2c8d27e FM |
1716 | free_leaf_list(ctx->refs); |
1717 | ctx->refs = NULL; | |
8da6d581 JS |
1718 | return ret; |
1719 | } | |
1720 | ||
1721 | return 0; | |
1722 | } | |
1723 | ||
1724 | /* | |
a2c8d27e | 1725 | * Walk all backrefs for a given extent to find all roots that reference this |
8da6d581 JS |
1726 | * extent. Walking a backref means finding all extents that reference this |
1727 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1728 | * recursive process, but here it is implemented in an iterative fashion: We | |
1729 | * find all referencing extents for the extent in question and put them on a | |
1730 | * list. In turn, we find all referencing extents for those, further appending | |
1731 | * to the list. The way we iterate the list allows adding more elements after | |
1732 | * the current while iterating. The process stops when we reach the end of the | |
a2c8d27e FM |
1733 | * list. |
1734 | * | |
1baea6f1 FM |
1735 | * Found roots are added to @ctx->roots, which is allocated by this function if |
1736 | * it points to NULL, in which case the caller is responsible for freeing it | |
1737 | * after it's not needed anymore. | |
1738 | * This function requires @ctx->refs to be NULL, as it uses it for allocating a | |
1739 | * ulist to do temporary work, and frees it before returning. | |
8da6d581 | 1740 | * |
1baea6f1 | 1741 | * Returns 0 on success, < 0 on error. |
8da6d581 | 1742 | */ |
a2c8d27e | 1743 | static int btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx *ctx) |
8da6d581 | 1744 | { |
a2c8d27e FM |
1745 | const u64 orig_bytenr = ctx->bytenr; |
1746 | const bool orig_ignore_extent_item_pos = ctx->ignore_extent_item_pos; | |
1baea6f1 | 1747 | bool roots_ulist_allocated = false; |
cd1b413c | 1748 | struct ulist_iterator uiter; |
a2c8d27e FM |
1749 | int ret = 0; |
1750 | ||
1751 | ASSERT(ctx->refs == NULL); | |
8da6d581 | 1752 | |
a2c8d27e FM |
1753 | ctx->refs = ulist_alloc(GFP_NOFS); |
1754 | if (!ctx->refs) | |
8da6d581 | 1755 | return -ENOMEM; |
a2c8d27e | 1756 | |
a2c8d27e | 1757 | if (!ctx->roots) { |
1baea6f1 FM |
1758 | ctx->roots = ulist_alloc(GFP_NOFS); |
1759 | if (!ctx->roots) { | |
1760 | ulist_free(ctx->refs); | |
1761 | ctx->refs = NULL; | |
1762 | return -ENOMEM; | |
1763 | } | |
1764 | roots_ulist_allocated = true; | |
8da6d581 JS |
1765 | } |
1766 | ||
a2c8d27e FM |
1767 | ctx->ignore_extent_item_pos = true; |
1768 | ||
cd1b413c | 1769 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1770 | while (1) { |
a2c8d27e FM |
1771 | struct ulist_node *node; |
1772 | ||
1773 | ret = find_parent_nodes(ctx, NULL); | |
8da6d581 | 1774 | if (ret < 0 && ret != -ENOENT) { |
1baea6f1 FM |
1775 | if (roots_ulist_allocated) { |
1776 | ulist_free(ctx->roots); | |
1777 | ctx->roots = NULL; | |
1778 | } | |
a2c8d27e | 1779 | break; |
8da6d581 | 1780 | } |
a2c8d27e FM |
1781 | ret = 0; |
1782 | node = ulist_next(ctx->refs, &uiter); | |
8da6d581 JS |
1783 | if (!node) |
1784 | break; | |
a2c8d27e | 1785 | ctx->bytenr = node->val; |
bca1a290 | 1786 | cond_resched(); |
8da6d581 JS |
1787 | } |
1788 | ||
a2c8d27e FM |
1789 | ulist_free(ctx->refs); |
1790 | ctx->refs = NULL; | |
1791 | ctx->bytenr = orig_bytenr; | |
1792 | ctx->ignore_extent_item_pos = orig_ignore_extent_item_pos; | |
1793 | ||
1794 | return ret; | |
8da6d581 JS |
1795 | } |
1796 | ||
a2c8d27e | 1797 | int btrfs_find_all_roots(struct btrfs_backref_walk_ctx *ctx, |
c7bcbb21 | 1798 | bool skip_commit_root_sem) |
9e351cc8 JB |
1799 | { |
1800 | int ret; | |
1801 | ||
a2c8d27e FM |
1802 | if (!ctx->trans && !skip_commit_root_sem) |
1803 | down_read(&ctx->fs_info->commit_root_sem); | |
1804 | ret = btrfs_find_all_roots_safe(ctx); | |
1805 | if (!ctx->trans && !skip_commit_root_sem) | |
1806 | up_read(&ctx->fs_info->commit_root_sem); | |
9e351cc8 JB |
1807 | return ret; |
1808 | } | |
1809 | ||
84a7949d FM |
1810 | struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void) |
1811 | { | |
1812 | struct btrfs_backref_share_check_ctx *ctx; | |
1813 | ||
1814 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1815 | if (!ctx) | |
1816 | return NULL; | |
1817 | ||
1818 | ulist_init(&ctx->refs); | |
84a7949d FM |
1819 | |
1820 | return ctx; | |
1821 | } | |
1822 | ||
1823 | void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx) | |
1824 | { | |
1825 | if (!ctx) | |
1826 | return; | |
1827 | ||
1828 | ulist_release(&ctx->refs); | |
84a7949d FM |
1829 | kfree(ctx); |
1830 | } | |
1831 | ||
8eedadda FM |
1832 | /* |
1833 | * Check if a data extent is shared or not. | |
6e353e3b | 1834 | * |
ceb707da | 1835 | * @inode: The inode whose extent we are checking. |
b8f164e3 FM |
1836 | * @bytenr: Logical bytenr of the extent we are checking. |
1837 | * @extent_gen: Generation of the extent (file extent item) or 0 if it is | |
1838 | * not known. | |
61dbb952 | 1839 | * @ctx: A backref sharedness check context. |
2c2ed5aa | 1840 | * |
8eedadda | 1841 | * btrfs_is_data_extent_shared uses the backref walking code but will short |
2c2ed5aa MF |
1842 | * circuit as soon as it finds a root or inode that doesn't match the |
1843 | * one passed in. This provides a significant performance benefit for | |
1844 | * callers (such as fiemap) which want to know whether the extent is | |
1845 | * shared but do not need a ref count. | |
1846 | * | |
03628cdb FM |
1847 | * This attempts to attach to the running transaction in order to account for |
1848 | * delayed refs, but continues on even when no running transaction exists. | |
bb739cf0 | 1849 | * |
2c2ed5aa MF |
1850 | * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. |
1851 | */ | |
ceb707da | 1852 | int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr, |
b8f164e3 | 1853 | u64 extent_gen, |
61dbb952 | 1854 | struct btrfs_backref_share_check_ctx *ctx) |
dc046b10 | 1855 | { |
a2c8d27e | 1856 | struct btrfs_backref_walk_ctx walk_ctx = { 0 }; |
ceb707da | 1857 | struct btrfs_root *root = inode->root; |
bb739cf0 EN |
1858 | struct btrfs_fs_info *fs_info = root->fs_info; |
1859 | struct btrfs_trans_handle *trans; | |
dc046b10 JB |
1860 | struct ulist_iterator uiter; |
1861 | struct ulist_node *node; | |
f3a84ccd | 1862 | struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem); |
dc046b10 | 1863 | int ret = 0; |
3ec4d323 | 1864 | struct share_check shared = { |
877c1476 FM |
1865 | .ctx = ctx, |
1866 | .root = root, | |
ceb707da | 1867 | .inum = btrfs_ino(inode), |
73e339e6 | 1868 | .data_bytenr = bytenr, |
6976201f | 1869 | .data_extent_gen = extent_gen, |
3ec4d323 | 1870 | .share_count = 0, |
73e339e6 | 1871 | .self_ref_count = 0, |
4fc7b572 | 1872 | .have_delayed_delete_refs = false, |
3ec4d323 | 1873 | }; |
12a824dc | 1874 | int level; |
e2fd8306 FM |
1875 | bool leaf_cached; |
1876 | bool leaf_is_shared; | |
dc046b10 | 1877 | |
73e339e6 FM |
1878 | for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) { |
1879 | if (ctx->prev_extents_cache[i].bytenr == bytenr) | |
1880 | return ctx->prev_extents_cache[i].is_shared; | |
1881 | } | |
1882 | ||
84a7949d | 1883 | ulist_init(&ctx->refs); |
dc046b10 | 1884 | |
a6d155d2 | 1885 | trans = btrfs_join_transaction_nostart(root); |
bb739cf0 | 1886 | if (IS_ERR(trans)) { |
03628cdb FM |
1887 | if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) { |
1888 | ret = PTR_ERR(trans); | |
1889 | goto out; | |
1890 | } | |
bb739cf0 | 1891 | trans = NULL; |
dc046b10 | 1892 | down_read(&fs_info->commit_root_sem); |
bb739cf0 EN |
1893 | } else { |
1894 | btrfs_get_tree_mod_seq(fs_info, &elem); | |
a2c8d27e | 1895 | walk_ctx.time_seq = elem.seq; |
bb739cf0 EN |
1896 | } |
1897 | ||
e2fd8306 FM |
1898 | ctx->use_path_cache = true; |
1899 | ||
1900 | /* | |
1901 | * We may have previously determined that the current leaf is shared. | |
1902 | * If it is, then we have a data extent that is shared due to a shared | |
1903 | * subtree (caused by snapshotting) and we don't need to check for data | |
1904 | * backrefs. If the leaf is not shared, then we must do backref walking | |
1905 | * to determine if the data extent is shared through reflinks. | |
1906 | */ | |
1907 | leaf_cached = lookup_backref_shared_cache(ctx, root, | |
1908 | ctx->curr_leaf_bytenr, 0, | |
1909 | &leaf_is_shared); | |
1910 | if (leaf_cached && leaf_is_shared) { | |
1911 | ret = 1; | |
1912 | goto out_trans; | |
1913 | } | |
1914 | ||
a2c8d27e FM |
1915 | walk_ctx.ignore_extent_item_pos = true; |
1916 | walk_ctx.trans = trans; | |
1917 | walk_ctx.fs_info = fs_info; | |
1918 | walk_ctx.refs = &ctx->refs; | |
1919 | ||
12a824dc FM |
1920 | /* -1 means we are in the bytenr of the data extent. */ |
1921 | level = -1; | |
dc046b10 JB |
1922 | ULIST_ITER_INIT(&uiter); |
1923 | while (1) { | |
2280d425 | 1924 | const unsigned long prev_ref_count = ctx->refs.nnodes; |
12a824dc | 1925 | |
a2c8d27e FM |
1926 | walk_ctx.bytenr = bytenr; |
1927 | ret = find_parent_nodes(&walk_ctx, &shared); | |
877c1476 FM |
1928 | if (ret == BACKREF_FOUND_SHARED || |
1929 | ret == BACKREF_FOUND_NOT_SHARED) { | |
1930 | /* If shared must return 1, otherwise return 0. */ | |
1931 | ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0; | |
12a824dc | 1932 | if (level >= 0) |
61dbb952 | 1933 | store_backref_shared_cache(ctx, root, bytenr, |
877c1476 | 1934 | level, ret == 1); |
dc046b10 JB |
1935 | break; |
1936 | } | |
1937 | if (ret < 0 && ret != -ENOENT) | |
1938 | break; | |
2c2ed5aa | 1939 | ret = 0; |
b8f164e3 | 1940 | |
63c84b46 | 1941 | /* |
2280d425 FM |
1942 | * More than one extent buffer (bytenr) may have been added to |
1943 | * the ctx->refs ulist, in which case we have to check multiple | |
1944 | * tree paths in case the first one is not shared, so we can not | |
1945 | * use the path cache which is made for a single path. Multiple | |
1946 | * extent buffers at the current level happen when: | |
1947 | * | |
1948 | * 1) level -1, the data extent: If our data extent was not | |
1949 | * directly shared (without multiple reference items), then | |
1950 | * it might have a single reference item with a count > 1 for | |
1951 | * the same offset, which means there are 2 (or more) file | |
1952 | * extent items that point to the data extent - this happens | |
1953 | * when a file extent item needs to be split and then one | |
1954 | * item gets moved to another leaf due to a b+tree leaf split | |
1955 | * when inserting some item. In this case the file extent | |
1956 | * items may be located in different leaves and therefore | |
1957 | * some of the leaves may be referenced through shared | |
1958 | * subtrees while others are not. Since our extent buffer | |
1959 | * cache only works for a single path (by far the most common | |
1960 | * case and simpler to deal with), we can not use it if we | |
1961 | * have multiple leaves (which implies multiple paths). | |
1962 | * | |
1963 | * 2) level >= 0, a tree node/leaf: We can have a mix of direct | |
1964 | * and indirect references on a b+tree node/leaf, so we have | |
1965 | * to check multiple paths, and the extent buffer (the | |
1966 | * current bytenr) may be shared or not. One example is | |
1967 | * during relocation as we may get a shared tree block ref | |
1968 | * (direct ref) and a non-shared tree block ref (indirect | |
1969 | * ref) for the same node/leaf. | |
63c84b46 | 1970 | */ |
2280d425 | 1971 | if ((ctx->refs.nnodes - prev_ref_count) > 1) |
61dbb952 | 1972 | ctx->use_path_cache = false; |
63c84b46 | 1973 | |
12a824dc | 1974 | if (level >= 0) |
61dbb952 | 1975 | store_backref_shared_cache(ctx, root, bytenr, |
12a824dc | 1976 | level, false); |
84a7949d | 1977 | node = ulist_next(&ctx->refs, &uiter); |
dc046b10 JB |
1978 | if (!node) |
1979 | break; | |
1980 | bytenr = node->val; | |
2280d425 FM |
1981 | if (ctx->use_path_cache) { |
1982 | bool is_shared; | |
1983 | bool cached; | |
1984 | ||
1985 | level++; | |
1986 | cached = lookup_backref_shared_cache(ctx, root, bytenr, | |
1987 | level, &is_shared); | |
1988 | if (cached) { | |
1989 | ret = (is_shared ? 1 : 0); | |
1990 | break; | |
1991 | } | |
12a824dc | 1992 | } |
18bf591b | 1993 | shared.share_count = 0; |
4fc7b572 | 1994 | shared.have_delayed_delete_refs = false; |
dc046b10 JB |
1995 | cond_resched(); |
1996 | } | |
bb739cf0 | 1997 | |
2280d425 FM |
1998 | /* |
1999 | * If the path cache is disabled, then it means at some tree level we | |
2000 | * got multiple parents due to a mix of direct and indirect backrefs or | |
2001 | * multiple leaves with file extent items pointing to the same data | |
2002 | * extent. We have to invalidate the cache and cache only the sharedness | |
2003 | * result for the levels where we got only one node/reference. | |
2004 | */ | |
2005 | if (!ctx->use_path_cache) { | |
2006 | int i = 0; | |
2007 | ||
2008 | level--; | |
2009 | if (ret >= 0 && level >= 0) { | |
2010 | bytenr = ctx->path_cache_entries[level].bytenr; | |
2011 | ctx->use_path_cache = true; | |
2012 | store_backref_shared_cache(ctx, root, bytenr, level, ret); | |
2013 | i = level + 1; | |
2014 | } | |
2015 | ||
2016 | for ( ; i < BTRFS_MAX_LEVEL; i++) | |
2017 | ctx->path_cache_entries[i].bytenr = 0; | |
2018 | } | |
2019 | ||
73e339e6 FM |
2020 | /* |
2021 | * Cache the sharedness result for the data extent if we know our inode | |
2022 | * has more than 1 file extent item that refers to the data extent. | |
2023 | */ | |
2024 | if (ret >= 0 && shared.self_ref_count > 1) { | |
2025 | int slot = ctx->prev_extents_cache_slot; | |
2026 | ||
2027 | ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr; | |
2028 | ctx->prev_extents_cache[slot].is_shared = (ret == 1); | |
2029 | ||
2030 | slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; | |
2031 | ctx->prev_extents_cache_slot = slot; | |
2032 | } | |
2033 | ||
e2fd8306 | 2034 | out_trans: |
bb739cf0 | 2035 | if (trans) { |
dc046b10 | 2036 | btrfs_put_tree_mod_seq(fs_info, &elem); |
bb739cf0 EN |
2037 | btrfs_end_transaction(trans); |
2038 | } else { | |
dc046b10 | 2039 | up_read(&fs_info->commit_root_sem); |
bb739cf0 | 2040 | } |
03628cdb | 2041 | out: |
84a7949d | 2042 | ulist_release(&ctx->refs); |
877c1476 FM |
2043 | ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr; |
2044 | ||
dc046b10 JB |
2045 | return ret; |
2046 | } | |
2047 | ||
f186373f MF |
2048 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
2049 | u64 start_off, struct btrfs_path *path, | |
2050 | struct btrfs_inode_extref **ret_extref, | |
2051 | u64 *found_off) | |
2052 | { | |
2053 | int ret, slot; | |
2054 | struct btrfs_key key; | |
2055 | struct btrfs_key found_key; | |
2056 | struct btrfs_inode_extref *extref; | |
73980bec | 2057 | const struct extent_buffer *leaf; |
f186373f MF |
2058 | unsigned long ptr; |
2059 | ||
2060 | key.objectid = inode_objectid; | |
962a298f | 2061 | key.type = BTRFS_INODE_EXTREF_KEY; |
f186373f MF |
2062 | key.offset = start_off; |
2063 | ||
2064 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2065 | if (ret < 0) | |
2066 | return ret; | |
2067 | ||
2068 | while (1) { | |
2069 | leaf = path->nodes[0]; | |
2070 | slot = path->slots[0]; | |
2071 | if (slot >= btrfs_header_nritems(leaf)) { | |
2072 | /* | |
2073 | * If the item at offset is not found, | |
2074 | * btrfs_search_slot will point us to the slot | |
2075 | * where it should be inserted. In our case | |
2076 | * that will be the slot directly before the | |
2077 | * next INODE_REF_KEY_V2 item. In the case | |
2078 | * that we're pointing to the last slot in a | |
2079 | * leaf, we must move one leaf over. | |
2080 | */ | |
2081 | ret = btrfs_next_leaf(root, path); | |
2082 | if (ret) { | |
2083 | if (ret >= 1) | |
2084 | ret = -ENOENT; | |
2085 | break; | |
2086 | } | |
2087 | continue; | |
2088 | } | |
2089 | ||
2090 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2091 | ||
2092 | /* | |
2093 | * Check that we're still looking at an extended ref key for | |
2094 | * this particular objectid. If we have different | |
2095 | * objectid or type then there are no more to be found | |
2096 | * in the tree and we can exit. | |
2097 | */ | |
2098 | ret = -ENOENT; | |
2099 | if (found_key.objectid != inode_objectid) | |
2100 | break; | |
962a298f | 2101 | if (found_key.type != BTRFS_INODE_EXTREF_KEY) |
f186373f MF |
2102 | break; |
2103 | ||
2104 | ret = 0; | |
2105 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2106 | extref = (struct btrfs_inode_extref *)ptr; | |
2107 | *ret_extref = extref; | |
2108 | if (found_off) | |
2109 | *found_off = found_key.offset; | |
2110 | break; | |
2111 | } | |
2112 | ||
2113 | return ret; | |
2114 | } | |
2115 | ||
48a3b636 ES |
2116 | /* |
2117 | * this iterates to turn a name (from iref/extref) into a full filesystem path. | |
2118 | * Elements of the path are separated by '/' and the path is guaranteed to be | |
2119 | * 0-terminated. the path is only given within the current file system. | |
2120 | * Therefore, it never starts with a '/'. the caller is responsible to provide | |
2121 | * "size" bytes in "dest". the dest buffer will be filled backwards. finally, | |
2122 | * the start point of the resulting string is returned. this pointer is within | |
2123 | * dest, normally. | |
2124 | * in case the path buffer would overflow, the pointer is decremented further | |
2125 | * as if output was written to the buffer, though no more output is actually | |
2126 | * generated. that way, the caller can determine how much space would be | |
2127 | * required for the path to fit into the buffer. in that case, the returned | |
2128 | * value will be smaller than dest. callers must check this! | |
2129 | */ | |
96b5bd77 JS |
2130 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
2131 | u32 name_len, unsigned long name_off, | |
2132 | struct extent_buffer *eb_in, u64 parent, | |
2133 | char *dest, u32 size) | |
a542ad1b | 2134 | { |
a542ad1b JS |
2135 | int slot; |
2136 | u64 next_inum; | |
2137 | int ret; | |
661bec6b | 2138 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
2139 | struct extent_buffer *eb = eb_in; |
2140 | struct btrfs_key found_key; | |
d24bec3a | 2141 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
2142 | |
2143 | if (bytes_left >= 0) | |
2144 | dest[bytes_left] = '\0'; | |
2145 | ||
2146 | while (1) { | |
d24bec3a | 2147 | bytes_left -= name_len; |
a542ad1b JS |
2148 | if (bytes_left >= 0) |
2149 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 2150 | name_off, name_len); |
b916a59a | 2151 | if (eb != eb_in) { |
0c0fe3b0 | 2152 | if (!path->skip_locking) |
ac5887c8 | 2153 | btrfs_tree_read_unlock(eb); |
a542ad1b | 2154 | free_extent_buffer(eb); |
b916a59a | 2155 | } |
c234a24d DS |
2156 | ret = btrfs_find_item(fs_root, path, parent, 0, |
2157 | BTRFS_INODE_REF_KEY, &found_key); | |
8f24b496 JS |
2158 | if (ret > 0) |
2159 | ret = -ENOENT; | |
a542ad1b JS |
2160 | if (ret) |
2161 | break; | |
d24bec3a | 2162 | |
a542ad1b JS |
2163 | next_inum = found_key.offset; |
2164 | ||
2165 | /* regular exit ahead */ | |
2166 | if (parent == next_inum) | |
2167 | break; | |
2168 | ||
2169 | slot = path->slots[0]; | |
2170 | eb = path->nodes[0]; | |
2171 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 2172 | if (eb != eb_in) { |
0c0fe3b0 FM |
2173 | path->nodes[0] = NULL; |
2174 | path->locks[0] = 0; | |
b916a59a | 2175 | } |
a542ad1b | 2176 | btrfs_release_path(path); |
a542ad1b | 2177 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
2178 | |
2179 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
2180 | name_off = (unsigned long)(iref + 1); | |
2181 | ||
a542ad1b JS |
2182 | parent = next_inum; |
2183 | --bytes_left; | |
2184 | if (bytes_left >= 0) | |
2185 | dest[bytes_left] = '/'; | |
2186 | } | |
2187 | ||
2188 | btrfs_release_path(path); | |
2189 | ||
2190 | if (ret) | |
2191 | return ERR_PTR(ret); | |
2192 | ||
2193 | return dest + bytes_left; | |
2194 | } | |
2195 | ||
2196 | /* | |
2197 | * this makes the path point to (logical EXTENT_ITEM *) | |
2198 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
2199 | * tree blocks and <0 on error. | |
2200 | */ | |
2201 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
2202 | struct btrfs_path *path, struct btrfs_key *found_key, |
2203 | u64 *flags_ret) | |
a542ad1b | 2204 | { |
29cbcf40 | 2205 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical); |
a542ad1b JS |
2206 | int ret; |
2207 | u64 flags; | |
261c84b6 | 2208 | u64 size = 0; |
a542ad1b | 2209 | u32 item_size; |
73980bec | 2210 | const struct extent_buffer *eb; |
a542ad1b JS |
2211 | struct btrfs_extent_item *ei; |
2212 | struct btrfs_key key; | |
2213 | ||
261c84b6 JB |
2214 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
2215 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2216 | else | |
2217 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a542ad1b JS |
2218 | key.objectid = logical; |
2219 | key.offset = (u64)-1; | |
2220 | ||
29cbcf40 | 2221 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a542ad1b JS |
2222 | if (ret < 0) |
2223 | return ret; | |
a542ad1b | 2224 | |
29cbcf40 | 2225 | ret = btrfs_previous_extent_item(extent_root, path, 0); |
850a8cdf WS |
2226 | if (ret) { |
2227 | if (ret > 0) | |
2228 | ret = -ENOENT; | |
2229 | return ret; | |
580f0a67 | 2230 | } |
850a8cdf | 2231 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); |
261c84b6 | 2232 | if (found_key->type == BTRFS_METADATA_ITEM_KEY) |
da17066c | 2233 | size = fs_info->nodesize; |
261c84b6 JB |
2234 | else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) |
2235 | size = found_key->offset; | |
2236 | ||
580f0a67 | 2237 | if (found_key->objectid > logical || |
261c84b6 | 2238 | found_key->objectid + size <= logical) { |
ab8d0fc4 JM |
2239 | btrfs_debug(fs_info, |
2240 | "logical %llu is not within any extent", logical); | |
a542ad1b | 2241 | return -ENOENT; |
4692cf58 | 2242 | } |
a542ad1b JS |
2243 | |
2244 | eb = path->nodes[0]; | |
3212fa14 | 2245 | item_size = btrfs_item_size(eb, path->slots[0]); |
a542ad1b JS |
2246 | BUG_ON(item_size < sizeof(*ei)); |
2247 | ||
2248 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
2249 | flags = btrfs_extent_flags(eb, ei); | |
2250 | ||
ab8d0fc4 JM |
2251 | btrfs_debug(fs_info, |
2252 | "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u", | |
c1c9ff7c GU |
2253 | logical, logical - found_key->objectid, found_key->objectid, |
2254 | found_key->offset, flags, item_size); | |
69917e43 LB |
2255 | |
2256 | WARN_ON(!flags_ret); | |
2257 | if (flags_ret) { | |
2258 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
2259 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
2260 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
2261 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
2262 | else | |
290342f6 | 2263 | BUG(); |
69917e43 LB |
2264 | return 0; |
2265 | } | |
a542ad1b JS |
2266 | |
2267 | return -EIO; | |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
2272 | * for the first call and may be modified. it is used to track state. | |
2273 | * if more refs exist, 0 is returned and the next call to | |
e0c476b1 | 2274 | * get_extent_inline_ref must pass the modified ptr parameter to get the |
a542ad1b JS |
2275 | * next ref. after the last ref was processed, 1 is returned. |
2276 | * returns <0 on error | |
2277 | */ | |
e0c476b1 JM |
2278 | static int get_extent_inline_ref(unsigned long *ptr, |
2279 | const struct extent_buffer *eb, | |
2280 | const struct btrfs_key *key, | |
2281 | const struct btrfs_extent_item *ei, | |
2282 | u32 item_size, | |
2283 | struct btrfs_extent_inline_ref **out_eiref, | |
2284 | int *out_type) | |
a542ad1b JS |
2285 | { |
2286 | unsigned long end; | |
2287 | u64 flags; | |
2288 | struct btrfs_tree_block_info *info; | |
2289 | ||
2290 | if (!*ptr) { | |
2291 | /* first call */ | |
2292 | flags = btrfs_extent_flags(eb, ei); | |
2293 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
6eda71d0 LB |
2294 | if (key->type == BTRFS_METADATA_ITEM_KEY) { |
2295 | /* a skinny metadata extent */ | |
2296 | *out_eiref = | |
2297 | (struct btrfs_extent_inline_ref *)(ei + 1); | |
2298 | } else { | |
2299 | WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY); | |
2300 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2301 | *out_eiref = | |
2302 | (struct btrfs_extent_inline_ref *)(info + 1); | |
2303 | } | |
a542ad1b JS |
2304 | } else { |
2305 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2306 | } | |
2307 | *ptr = (unsigned long)*out_eiref; | |
cd857dd6 | 2308 | if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) |
a542ad1b JS |
2309 | return -ENOENT; |
2310 | } | |
2311 | ||
2312 | end = (unsigned long)ei + item_size; | |
6eda71d0 | 2313 | *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); |
3de28d57 LB |
2314 | *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref, |
2315 | BTRFS_REF_TYPE_ANY); | |
2316 | if (*out_type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 2317 | return -EUCLEAN; |
a542ad1b JS |
2318 | |
2319 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
2320 | WARN_ON(*ptr > end); | |
2321 | if (*ptr == end) | |
2322 | return 1; /* last */ | |
2323 | ||
2324 | return 0; | |
2325 | } | |
2326 | ||
2327 | /* | |
2328 | * reads the tree block backref for an extent. tree level and root are returned | |
2329 | * through out_level and out_root. ptr must point to a 0 value for the first | |
e0c476b1 | 2330 | * call and may be modified (see get_extent_inline_ref comment). |
a542ad1b JS |
2331 | * returns 0 if data was provided, 1 if there was no more data to provide or |
2332 | * <0 on error. | |
2333 | */ | |
2334 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
2335 | struct btrfs_key *key, struct btrfs_extent_item *ei, |
2336 | u32 item_size, u64 *out_root, u8 *out_level) | |
a542ad1b JS |
2337 | { |
2338 | int ret; | |
2339 | int type; | |
a542ad1b JS |
2340 | struct btrfs_extent_inline_ref *eiref; |
2341 | ||
2342 | if (*ptr == (unsigned long)-1) | |
2343 | return 1; | |
2344 | ||
2345 | while (1) { | |
e0c476b1 | 2346 | ret = get_extent_inline_ref(ptr, eb, key, ei, item_size, |
6eda71d0 | 2347 | &eiref, &type); |
a542ad1b JS |
2348 | if (ret < 0) |
2349 | return ret; | |
2350 | ||
2351 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
2352 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2353 | break; | |
2354 | ||
2355 | if (ret == 1) | |
2356 | return 1; | |
2357 | } | |
2358 | ||
2359 | /* we can treat both ref types equally here */ | |
a542ad1b | 2360 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); |
a1317f45 FM |
2361 | |
2362 | if (key->type == BTRFS_EXTENT_ITEM_KEY) { | |
2363 | struct btrfs_tree_block_info *info; | |
2364 | ||
2365 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2366 | *out_level = btrfs_tree_block_level(eb, info); | |
2367 | } else { | |
2368 | ASSERT(key->type == BTRFS_METADATA_ITEM_KEY); | |
2369 | *out_level = (u8)key->offset; | |
2370 | } | |
a542ad1b JS |
2371 | |
2372 | if (ret == 1) | |
2373 | *ptr = (unsigned long)-1; | |
2374 | ||
2375 | return 0; | |
2376 | } | |
2377 | ||
ab8d0fc4 JM |
2378 | static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, |
2379 | struct extent_inode_elem *inode_list, | |
2380 | u64 root, u64 extent_item_objectid, | |
2381 | iterate_extent_inodes_t *iterate, void *ctx) | |
a542ad1b | 2382 | { |
976b1908 | 2383 | struct extent_inode_elem *eie; |
4692cf58 | 2384 | int ret = 0; |
4692cf58 | 2385 | |
976b1908 | 2386 | for (eie = inode_list; eie; eie = eie->next) { |
ab8d0fc4 JM |
2387 | btrfs_debug(fs_info, |
2388 | "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu", | |
2389 | extent_item_objectid, eie->inum, | |
2390 | eie->offset, root); | |
c7499a64 | 2391 | ret = iterate(eie->inum, eie->offset, eie->num_bytes, root, ctx); |
4692cf58 | 2392 | if (ret) { |
ab8d0fc4 JM |
2393 | btrfs_debug(fs_info, |
2394 | "stopping iteration for %llu due to ret=%d", | |
2395 | extent_item_objectid, ret); | |
4692cf58 JS |
2396 | break; |
2397 | } | |
a542ad1b JS |
2398 | } |
2399 | ||
a542ad1b JS |
2400 | return ret; |
2401 | } | |
2402 | ||
2403 | /* | |
2404 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 2405 | * the given parameters. |
a542ad1b JS |
2406 | * when the iterator function returns a non-zero value, iteration stops. |
2407 | */ | |
a2c8d27e FM |
2408 | int iterate_extent_inodes(struct btrfs_backref_walk_ctx *ctx, |
2409 | bool search_commit_root, | |
2410 | iterate_extent_inodes_t *iterate, void *user_ctx) | |
a542ad1b | 2411 | { |
a542ad1b | 2412 | int ret; |
a2c8d27e FM |
2413 | struct ulist *refs; |
2414 | struct ulist_node *ref_node; | |
f3a84ccd | 2415 | struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem); |
cd1b413c | 2416 | struct ulist_iterator ref_uiter; |
a542ad1b | 2417 | |
a2c8d27e FM |
2418 | btrfs_debug(ctx->fs_info, "resolving all inodes for extent %llu", |
2419 | ctx->bytenr); | |
2420 | ||
2421 | ASSERT(ctx->trans == NULL); | |
1baea6f1 FM |
2422 | ASSERT(ctx->roots == NULL); |
2423 | ||
da61d31a | 2424 | if (!search_commit_root) { |
a2c8d27e FM |
2425 | struct btrfs_trans_handle *trans; |
2426 | ||
2427 | trans = btrfs_attach_transaction(ctx->fs_info->tree_root); | |
bfc61c36 FM |
2428 | if (IS_ERR(trans)) { |
2429 | if (PTR_ERR(trans) != -ENOENT && | |
66d04209 | 2430 | PTR_ERR(trans) != -EROFS) |
bfc61c36 FM |
2431 | return PTR_ERR(trans); |
2432 | trans = NULL; | |
2433 | } | |
a2c8d27e | 2434 | ctx->trans = trans; |
bfc61c36 FM |
2435 | } |
2436 | ||
a2c8d27e FM |
2437 | if (ctx->trans) { |
2438 | btrfs_get_tree_mod_seq(ctx->fs_info, &seq_elem); | |
2439 | ctx->time_seq = seq_elem.seq; | |
2440 | } else { | |
2441 | down_read(&ctx->fs_info->commit_root_sem); | |
2442 | } | |
a542ad1b | 2443 | |
a2c8d27e | 2444 | ret = btrfs_find_all_leafs(ctx); |
4692cf58 JS |
2445 | if (ret) |
2446 | goto out; | |
a2c8d27e FM |
2447 | refs = ctx->refs; |
2448 | ctx->refs = NULL; | |
a542ad1b | 2449 | |
cd1b413c JS |
2450 | ULIST_ITER_INIT(&ref_uiter); |
2451 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
66d04209 | 2452 | const u64 leaf_bytenr = ref_node->val; |
a2c8d27e FM |
2453 | struct ulist_node *root_node; |
2454 | struct ulist_iterator root_uiter; | |
66d04209 FM |
2455 | struct extent_inode_elem *inode_list; |
2456 | ||
2457 | inode_list = (struct extent_inode_elem *)(uintptr_t)ref_node->aux; | |
2458 | ||
2459 | if (ctx->cache_lookup) { | |
2460 | const u64 *root_ids; | |
2461 | int root_count; | |
2462 | bool cached; | |
2463 | ||
2464 | cached = ctx->cache_lookup(leaf_bytenr, ctx->user_ctx, | |
2465 | &root_ids, &root_count); | |
2466 | if (cached) { | |
2467 | for (int i = 0; i < root_count; i++) { | |
2468 | ret = iterate_leaf_refs(ctx->fs_info, | |
2469 | inode_list, | |
2470 | root_ids[i], | |
2471 | leaf_bytenr, | |
2472 | iterate, | |
2473 | user_ctx); | |
2474 | if (ret) | |
2475 | break; | |
2476 | } | |
2477 | continue; | |
2478 | } | |
2479 | } | |
2480 | ||
2481 | if (!ctx->roots) { | |
2482 | ctx->roots = ulist_alloc(GFP_NOFS); | |
2483 | if (!ctx->roots) { | |
2484 | ret = -ENOMEM; | |
2485 | break; | |
2486 | } | |
2487 | } | |
a2c8d27e | 2488 | |
66d04209 | 2489 | ctx->bytenr = leaf_bytenr; |
a2c8d27e | 2490 | ret = btrfs_find_all_roots_safe(ctx); |
4692cf58 JS |
2491 | if (ret) |
2492 | break; | |
a2c8d27e | 2493 | |
66d04209 FM |
2494 | if (ctx->cache_store) |
2495 | ctx->cache_store(leaf_bytenr, ctx->roots, ctx->user_ctx); | |
2496 | ||
cd1b413c | 2497 | ULIST_ITER_INIT(&root_uiter); |
a2c8d27e FM |
2498 | while (!ret && (root_node = ulist_next(ctx->roots, &root_uiter))) { |
2499 | btrfs_debug(ctx->fs_info, | |
ab8d0fc4 JM |
2500 | "root %llu references leaf %llu, data list %#llx", |
2501 | root_node->val, ref_node->val, | |
2502 | ref_node->aux); | |
66d04209 | 2503 | ret = iterate_leaf_refs(ctx->fs_info, inode_list, |
a2c8d27e FM |
2504 | root_node->val, ctx->bytenr, |
2505 | iterate, user_ctx); | |
4692cf58 | 2506 | } |
1baea6f1 | 2507 | ulist_reinit(ctx->roots); |
a542ad1b JS |
2508 | } |
2509 | ||
976b1908 | 2510 | free_leaf_list(refs); |
4692cf58 | 2511 | out: |
a2c8d27e FM |
2512 | if (ctx->trans) { |
2513 | btrfs_put_tree_mod_seq(ctx->fs_info, &seq_elem); | |
2514 | btrfs_end_transaction(ctx->trans); | |
2515 | ctx->trans = NULL; | |
9e351cc8 | 2516 | } else { |
a2c8d27e | 2517 | up_read(&ctx->fs_info->commit_root_sem); |
7a3ae2f8 JS |
2518 | } |
2519 | ||
1baea6f1 FM |
2520 | ulist_free(ctx->roots); |
2521 | ctx->roots = NULL; | |
2522 | ||
88ffb665 FM |
2523 | if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP) |
2524 | ret = 0; | |
2525 | ||
a542ad1b JS |
2526 | return ret; |
2527 | } | |
2528 | ||
c7499a64 | 2529 | static int build_ino_list(u64 inum, u64 offset, u64 num_bytes, u64 root, void *ctx) |
e3059ec0 DS |
2530 | { |
2531 | struct btrfs_data_container *inodes = ctx; | |
2532 | const size_t c = 3 * sizeof(u64); | |
2533 | ||
2534 | if (inodes->bytes_left >= c) { | |
2535 | inodes->bytes_left -= c; | |
2536 | inodes->val[inodes->elem_cnt] = inum; | |
2537 | inodes->val[inodes->elem_cnt + 1] = offset; | |
2538 | inodes->val[inodes->elem_cnt + 2] = root; | |
2539 | inodes->elem_cnt += 3; | |
2540 | } else { | |
2541 | inodes->bytes_missing += c - inodes->bytes_left; | |
2542 | inodes->bytes_left = 0; | |
2543 | inodes->elem_missed += 3; | |
2544 | } | |
2545 | ||
2546 | return 0; | |
2547 | } | |
2548 | ||
a542ad1b JS |
2549 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, |
2550 | struct btrfs_path *path, | |
e3059ec0 | 2551 | void *ctx, bool ignore_offset) |
a542ad1b | 2552 | { |
a2c8d27e | 2553 | struct btrfs_backref_walk_ctx walk_ctx = { 0 }; |
a542ad1b | 2554 | int ret; |
69917e43 | 2555 | u64 flags = 0; |
a542ad1b | 2556 | struct btrfs_key found_key; |
7a3ae2f8 | 2557 | int search_commit_root = path->search_commit_root; |
a542ad1b | 2558 | |
69917e43 | 2559 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 2560 | btrfs_release_path(path); |
a542ad1b JS |
2561 | if (ret < 0) |
2562 | return ret; | |
69917e43 | 2563 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 2564 | return -EINVAL; |
a542ad1b | 2565 | |
a2c8d27e | 2566 | walk_ctx.bytenr = found_key.objectid; |
6ce6ba53 | 2567 | if (ignore_offset) |
a2c8d27e | 2568 | walk_ctx.ignore_extent_item_pos = true; |
6ce6ba53 | 2569 | else |
a2c8d27e FM |
2570 | walk_ctx.extent_item_pos = logical - found_key.objectid; |
2571 | walk_ctx.fs_info = fs_info; | |
6ce6ba53 | 2572 | |
a2c8d27e FM |
2573 | return iterate_extent_inodes(&walk_ctx, search_commit_root, |
2574 | build_ino_list, ctx); | |
a542ad1b JS |
2575 | } |
2576 | ||
ad6240f6 | 2577 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2578 | struct extent_buffer *eb, struct inode_fs_paths *ipath); |
d24bec3a | 2579 | |
875d1daa | 2580 | static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath) |
a542ad1b | 2581 | { |
aefc1eb1 | 2582 | int ret = 0; |
a542ad1b JS |
2583 | int slot; |
2584 | u32 cur; | |
2585 | u32 len; | |
2586 | u32 name_len; | |
2587 | u64 parent = 0; | |
2588 | int found = 0; | |
875d1daa DS |
2589 | struct btrfs_root *fs_root = ipath->fs_root; |
2590 | struct btrfs_path *path = ipath->btrfs_path; | |
a542ad1b | 2591 | struct extent_buffer *eb; |
a542ad1b JS |
2592 | struct btrfs_inode_ref *iref; |
2593 | struct btrfs_key found_key; | |
2594 | ||
aefc1eb1 | 2595 | while (!ret) { |
c234a24d DS |
2596 | ret = btrfs_find_item(fs_root, path, inum, |
2597 | parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY, | |
2598 | &found_key); | |
2599 | ||
a542ad1b JS |
2600 | if (ret < 0) |
2601 | break; | |
2602 | if (ret) { | |
2603 | ret = found ? 0 : -ENOENT; | |
2604 | break; | |
2605 | } | |
2606 | ++found; | |
2607 | ||
2608 | parent = found_key.offset; | |
2609 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2610 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2611 | if (!eb) { | |
2612 | ret = -ENOMEM; | |
2613 | break; | |
2614 | } | |
a542ad1b JS |
2615 | btrfs_release_path(path); |
2616 | ||
a542ad1b JS |
2617 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
2618 | ||
3212fa14 | 2619 | for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) { |
a542ad1b JS |
2620 | name_len = btrfs_inode_ref_name_len(eb, iref); |
2621 | /* path must be released before calling iterate()! */ | |
ab8d0fc4 JM |
2622 | btrfs_debug(fs_root->fs_info, |
2623 | "following ref at offset %u for inode %llu in tree %llu", | |
4fd786e6 MT |
2624 | cur, found_key.objectid, |
2625 | fs_root->root_key.objectid); | |
ad6240f6 | 2626 | ret = inode_to_path(parent, name_len, |
875d1daa | 2627 | (unsigned long)(iref + 1), eb, ipath); |
aefc1eb1 | 2628 | if (ret) |
a542ad1b | 2629 | break; |
a542ad1b JS |
2630 | len = sizeof(*iref) + name_len; |
2631 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
2632 | } | |
2633 | free_extent_buffer(eb); | |
2634 | } | |
2635 | ||
2636 | btrfs_release_path(path); | |
2637 | ||
2638 | return ret; | |
2639 | } | |
2640 | ||
875d1daa | 2641 | static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath) |
d24bec3a MF |
2642 | { |
2643 | int ret; | |
2644 | int slot; | |
2645 | u64 offset = 0; | |
2646 | u64 parent; | |
2647 | int found = 0; | |
875d1daa DS |
2648 | struct btrfs_root *fs_root = ipath->fs_root; |
2649 | struct btrfs_path *path = ipath->btrfs_path; | |
d24bec3a MF |
2650 | struct extent_buffer *eb; |
2651 | struct btrfs_inode_extref *extref; | |
d24bec3a MF |
2652 | u32 item_size; |
2653 | u32 cur_offset; | |
2654 | unsigned long ptr; | |
2655 | ||
2656 | while (1) { | |
2657 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
2658 | &offset); | |
2659 | if (ret < 0) | |
2660 | break; | |
2661 | if (ret) { | |
2662 | ret = found ? 0 : -ENOENT; | |
2663 | break; | |
2664 | } | |
2665 | ++found; | |
2666 | ||
2667 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2668 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2669 | if (!eb) { | |
2670 | ret = -ENOMEM; | |
2671 | break; | |
2672 | } | |
d24bec3a MF |
2673 | btrfs_release_path(path); |
2674 | ||
3212fa14 | 2675 | item_size = btrfs_item_size(eb, slot); |
2849a854 | 2676 | ptr = btrfs_item_ptr_offset(eb, slot); |
d24bec3a MF |
2677 | cur_offset = 0; |
2678 | ||
2679 | while (cur_offset < item_size) { | |
2680 | u32 name_len; | |
2681 | ||
2682 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
2683 | parent = btrfs_inode_extref_parent(eb, extref); | |
2684 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
ad6240f6 | 2685 | ret = inode_to_path(parent, name_len, |
875d1daa | 2686 | (unsigned long)&extref->name, eb, ipath); |
d24bec3a MF |
2687 | if (ret) |
2688 | break; | |
2689 | ||
2849a854 | 2690 | cur_offset += btrfs_inode_extref_name_len(eb, extref); |
d24bec3a MF |
2691 | cur_offset += sizeof(*extref); |
2692 | } | |
d24bec3a MF |
2693 | free_extent_buffer(eb); |
2694 | ||
2695 | offset++; | |
2696 | } | |
2697 | ||
2698 | btrfs_release_path(path); | |
2699 | ||
2700 | return ret; | |
2701 | } | |
2702 | ||
a542ad1b JS |
2703 | /* |
2704 | * returns 0 if the path could be dumped (probably truncated) | |
2705 | * returns <0 in case of an error | |
2706 | */ | |
d24bec3a | 2707 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2708 | struct extent_buffer *eb, struct inode_fs_paths *ipath) |
a542ad1b | 2709 | { |
a542ad1b JS |
2710 | char *fspath; |
2711 | char *fspath_min; | |
2712 | int i = ipath->fspath->elem_cnt; | |
2713 | const int s_ptr = sizeof(char *); | |
2714 | u32 bytes_left; | |
2715 | ||
2716 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
2717 | ipath->fspath->bytes_left - s_ptr : 0; | |
2718 | ||
740c3d22 | 2719 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
2720 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
2721 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
2722 | if (IS_ERR(fspath)) |
2723 | return PTR_ERR(fspath); | |
2724 | ||
2725 | if (fspath > fspath_min) { | |
745c4d8e | 2726 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
2727 | ++ipath->fspath->elem_cnt; |
2728 | ipath->fspath->bytes_left = fspath - fspath_min; | |
2729 | } else { | |
2730 | ++ipath->fspath->elem_missed; | |
2731 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
2732 | ipath->fspath->bytes_left = 0; | |
2733 | } | |
2734 | ||
2735 | return 0; | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * this dumps all file system paths to the inode into the ipath struct, provided | |
2740 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 2741 | * from ipath->fspath->val[i]. |
a542ad1b | 2742 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 2743 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
01327610 | 2744 | * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise, |
a542ad1b JS |
2745 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would |
2746 | * have been needed to return all paths. | |
2747 | */ | |
2748 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
2749 | { | |
ad6240f6 DS |
2750 | int ret; |
2751 | int found_refs = 0; | |
2752 | ||
875d1daa | 2753 | ret = iterate_inode_refs(inum, ipath); |
ad6240f6 DS |
2754 | if (!ret) |
2755 | ++found_refs; | |
2756 | else if (ret != -ENOENT) | |
2757 | return ret; | |
2758 | ||
875d1daa | 2759 | ret = iterate_inode_extrefs(inum, ipath); |
ad6240f6 DS |
2760 | if (ret == -ENOENT && found_refs) |
2761 | return 0; | |
2762 | ||
2763 | return ret; | |
a542ad1b JS |
2764 | } |
2765 | ||
a542ad1b JS |
2766 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
2767 | { | |
2768 | struct btrfs_data_container *data; | |
2769 | size_t alloc_bytes; | |
2770 | ||
2771 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
f54de068 | 2772 | data = kvmalloc(alloc_bytes, GFP_KERNEL); |
a542ad1b JS |
2773 | if (!data) |
2774 | return ERR_PTR(-ENOMEM); | |
2775 | ||
2776 | if (total_bytes >= sizeof(*data)) { | |
2777 | data->bytes_left = total_bytes - sizeof(*data); | |
2778 | data->bytes_missing = 0; | |
2779 | } else { | |
2780 | data->bytes_missing = sizeof(*data) - total_bytes; | |
2781 | data->bytes_left = 0; | |
2782 | } | |
2783 | ||
2784 | data->elem_cnt = 0; | |
2785 | data->elem_missed = 0; | |
2786 | ||
2787 | return data; | |
2788 | } | |
2789 | ||
2790 | /* | |
2791 | * allocates space to return multiple file system paths for an inode. | |
2792 | * total_bytes to allocate are passed, note that space usable for actual path | |
2793 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
2794 | * the returned pointer must be freed with free_ipath() in the end. | |
2795 | */ | |
2796 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
2797 | struct btrfs_path *path) | |
2798 | { | |
2799 | struct inode_fs_paths *ifp; | |
2800 | struct btrfs_data_container *fspath; | |
2801 | ||
2802 | fspath = init_data_container(total_bytes); | |
2803 | if (IS_ERR(fspath)) | |
afc6961f | 2804 | return ERR_CAST(fspath); |
a542ad1b | 2805 | |
f54de068 | 2806 | ifp = kmalloc(sizeof(*ifp), GFP_KERNEL); |
a542ad1b | 2807 | if (!ifp) { |
f54de068 | 2808 | kvfree(fspath); |
a542ad1b JS |
2809 | return ERR_PTR(-ENOMEM); |
2810 | } | |
2811 | ||
2812 | ifp->btrfs_path = path; | |
2813 | ifp->fspath = fspath; | |
2814 | ifp->fs_root = fs_root; | |
2815 | ||
2816 | return ifp; | |
2817 | } | |
2818 | ||
2819 | void free_ipath(struct inode_fs_paths *ipath) | |
2820 | { | |
4735fb28 JJ |
2821 | if (!ipath) |
2822 | return; | |
f54de068 | 2823 | kvfree(ipath->fspath); |
a542ad1b JS |
2824 | kfree(ipath); |
2825 | } | |
a37f232b | 2826 | |
d68194b2 | 2827 | struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info) |
a37f232b QW |
2828 | { |
2829 | struct btrfs_backref_iter *ret; | |
2830 | ||
d68194b2 | 2831 | ret = kzalloc(sizeof(*ret), GFP_NOFS); |
a37f232b QW |
2832 | if (!ret) |
2833 | return NULL; | |
2834 | ||
2835 | ret->path = btrfs_alloc_path(); | |
c15c2ec0 | 2836 | if (!ret->path) { |
a37f232b QW |
2837 | kfree(ret); |
2838 | return NULL; | |
2839 | } | |
2840 | ||
2841 | /* Current backref iterator only supports iteration in commit root */ | |
2842 | ret->path->search_commit_root = 1; | |
2843 | ret->path->skip_locking = 1; | |
2844 | ret->fs_info = fs_info; | |
2845 | ||
2846 | return ret; | |
2847 | } | |
2848 | ||
2849 | int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr) | |
2850 | { | |
2851 | struct btrfs_fs_info *fs_info = iter->fs_info; | |
29cbcf40 | 2852 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); |
a37f232b QW |
2853 | struct btrfs_path *path = iter->path; |
2854 | struct btrfs_extent_item *ei; | |
2855 | struct btrfs_key key; | |
2856 | int ret; | |
2857 | ||
2858 | key.objectid = bytenr; | |
2859 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2860 | key.offset = (u64)-1; | |
2861 | iter->bytenr = bytenr; | |
2862 | ||
29cbcf40 | 2863 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a37f232b QW |
2864 | if (ret < 0) |
2865 | return ret; | |
2866 | if (ret == 0) { | |
2867 | ret = -EUCLEAN; | |
2868 | goto release; | |
2869 | } | |
2870 | if (path->slots[0] == 0) { | |
2871 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
2872 | ret = -EUCLEAN; | |
2873 | goto release; | |
2874 | } | |
2875 | path->slots[0]--; | |
2876 | ||
2877 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2878 | if ((key.type != BTRFS_EXTENT_ITEM_KEY && | |
2879 | key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) { | |
2880 | ret = -ENOENT; | |
2881 | goto release; | |
2882 | } | |
2883 | memcpy(&iter->cur_key, &key, sizeof(key)); | |
2884 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2885 | path->slots[0]); | |
2886 | iter->end_ptr = (u32)(iter->item_ptr + | |
3212fa14 | 2887 | btrfs_item_size(path->nodes[0], path->slots[0])); |
a37f232b QW |
2888 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2889 | struct btrfs_extent_item); | |
2890 | ||
2891 | /* | |
2892 | * Only support iteration on tree backref yet. | |
2893 | * | |
2894 | * This is an extra precaution for non skinny-metadata, where | |
2895 | * EXTENT_ITEM is also used for tree blocks, that we can only use | |
2896 | * extent flags to determine if it's a tree block. | |
2897 | */ | |
2898 | if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) { | |
2899 | ret = -ENOTSUPP; | |
2900 | goto release; | |
2901 | } | |
2902 | iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei)); | |
2903 | ||
2904 | /* If there is no inline backref, go search for keyed backref */ | |
2905 | if (iter->cur_ptr >= iter->end_ptr) { | |
29cbcf40 | 2906 | ret = btrfs_next_item(extent_root, path); |
a37f232b QW |
2907 | |
2908 | /* No inline nor keyed ref */ | |
2909 | if (ret > 0) { | |
2910 | ret = -ENOENT; | |
2911 | goto release; | |
2912 | } | |
2913 | if (ret < 0) | |
2914 | goto release; | |
2915 | ||
2916 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, | |
2917 | path->slots[0]); | |
2918 | if (iter->cur_key.objectid != bytenr || | |
2919 | (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY && | |
2920 | iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) { | |
2921 | ret = -ENOENT; | |
2922 | goto release; | |
2923 | } | |
2924 | iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2925 | path->slots[0]); | |
2926 | iter->item_ptr = iter->cur_ptr; | |
3212fa14 | 2927 | iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size( |
a37f232b QW |
2928 | path->nodes[0], path->slots[0])); |
2929 | } | |
2930 | ||
2931 | return 0; | |
2932 | release: | |
2933 | btrfs_backref_iter_release(iter); | |
2934 | return ret; | |
2935 | } | |
c39c2ddc QW |
2936 | |
2937 | /* | |
2938 | * Go to the next backref item of current bytenr, can be either inlined or | |
2939 | * keyed. | |
2940 | * | |
2941 | * Caller needs to check whether it's inline ref or not by iter->cur_key. | |
2942 | * | |
2943 | * Return 0 if we get next backref without problem. | |
2944 | * Return >0 if there is no extra backref for this bytenr. | |
2945 | * Return <0 if there is something wrong happened. | |
2946 | */ | |
2947 | int btrfs_backref_iter_next(struct btrfs_backref_iter *iter) | |
2948 | { | |
2949 | struct extent_buffer *eb = btrfs_backref_get_eb(iter); | |
29cbcf40 | 2950 | struct btrfs_root *extent_root; |
c39c2ddc QW |
2951 | struct btrfs_path *path = iter->path; |
2952 | struct btrfs_extent_inline_ref *iref; | |
2953 | int ret; | |
2954 | u32 size; | |
2955 | ||
2956 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
2957 | /* We're still inside the inline refs */ | |
2958 | ASSERT(iter->cur_ptr < iter->end_ptr); | |
2959 | ||
2960 | if (btrfs_backref_has_tree_block_info(iter)) { | |
2961 | /* First tree block info */ | |
2962 | size = sizeof(struct btrfs_tree_block_info); | |
2963 | } else { | |
2964 | /* Use inline ref type to determine the size */ | |
2965 | int type; | |
2966 | ||
2967 | iref = (struct btrfs_extent_inline_ref *) | |
2968 | ((unsigned long)iter->cur_ptr); | |
2969 | type = btrfs_extent_inline_ref_type(eb, iref); | |
2970 | ||
2971 | size = btrfs_extent_inline_ref_size(type); | |
2972 | } | |
2973 | iter->cur_ptr += size; | |
2974 | if (iter->cur_ptr < iter->end_ptr) | |
2975 | return 0; | |
2976 | ||
2977 | /* All inline items iterated, fall through */ | |
2978 | } | |
2979 | ||
2980 | /* We're at keyed items, there is no inline item, go to the next one */ | |
29cbcf40 JB |
2981 | extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr); |
2982 | ret = btrfs_next_item(extent_root, iter->path); | |
c39c2ddc QW |
2983 | if (ret) |
2984 | return ret; | |
2985 | ||
2986 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]); | |
2987 | if (iter->cur_key.objectid != iter->bytenr || | |
2988 | (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY && | |
2989 | iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY)) | |
2990 | return 1; | |
2991 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2992 | path->slots[0]); | |
2993 | iter->cur_ptr = iter->item_ptr; | |
3212fa14 | 2994 | iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0], |
c39c2ddc QW |
2995 | path->slots[0]); |
2996 | return 0; | |
2997 | } | |
584fb121 QW |
2998 | |
2999 | void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info, | |
3000 | struct btrfs_backref_cache *cache, int is_reloc) | |
3001 | { | |
3002 | int i; | |
3003 | ||
3004 | cache->rb_root = RB_ROOT; | |
3005 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
3006 | INIT_LIST_HEAD(&cache->pending[i]); | |
3007 | INIT_LIST_HEAD(&cache->changed); | |
3008 | INIT_LIST_HEAD(&cache->detached); | |
3009 | INIT_LIST_HEAD(&cache->leaves); | |
3010 | INIT_LIST_HEAD(&cache->pending_edge); | |
3011 | INIT_LIST_HEAD(&cache->useless_node); | |
3012 | cache->fs_info = fs_info; | |
3013 | cache->is_reloc = is_reloc; | |
3014 | } | |
b1818dab QW |
3015 | |
3016 | struct btrfs_backref_node *btrfs_backref_alloc_node( | |
3017 | struct btrfs_backref_cache *cache, u64 bytenr, int level) | |
3018 | { | |
3019 | struct btrfs_backref_node *node; | |
3020 | ||
3021 | ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL); | |
3022 | node = kzalloc(sizeof(*node), GFP_NOFS); | |
3023 | if (!node) | |
3024 | return node; | |
3025 | ||
3026 | INIT_LIST_HEAD(&node->list); | |
3027 | INIT_LIST_HEAD(&node->upper); | |
3028 | INIT_LIST_HEAD(&node->lower); | |
3029 | RB_CLEAR_NODE(&node->rb_node); | |
3030 | cache->nr_nodes++; | |
3031 | node->level = level; | |
3032 | node->bytenr = bytenr; | |
3033 | ||
3034 | return node; | |
3035 | } | |
47254d07 QW |
3036 | |
3037 | struct btrfs_backref_edge *btrfs_backref_alloc_edge( | |
3038 | struct btrfs_backref_cache *cache) | |
3039 | { | |
3040 | struct btrfs_backref_edge *edge; | |
3041 | ||
3042 | edge = kzalloc(sizeof(*edge), GFP_NOFS); | |
3043 | if (edge) | |
3044 | cache->nr_edges++; | |
3045 | return edge; | |
3046 | } | |
023acb07 QW |
3047 | |
3048 | /* | |
3049 | * Drop the backref node from cache, also cleaning up all its | |
3050 | * upper edges and any uncached nodes in the path. | |
3051 | * | |
3052 | * This cleanup happens bottom up, thus the node should either | |
3053 | * be the lowest node in the cache or a detached node. | |
3054 | */ | |
3055 | void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache, | |
3056 | struct btrfs_backref_node *node) | |
3057 | { | |
3058 | struct btrfs_backref_node *upper; | |
3059 | struct btrfs_backref_edge *edge; | |
3060 | ||
3061 | if (!node) | |
3062 | return; | |
3063 | ||
3064 | BUG_ON(!node->lowest && !node->detached); | |
3065 | while (!list_empty(&node->upper)) { | |
3066 | edge = list_entry(node->upper.next, struct btrfs_backref_edge, | |
3067 | list[LOWER]); | |
3068 | upper = edge->node[UPPER]; | |
3069 | list_del(&edge->list[LOWER]); | |
3070 | list_del(&edge->list[UPPER]); | |
3071 | btrfs_backref_free_edge(cache, edge); | |
3072 | ||
023acb07 QW |
3073 | /* |
3074 | * Add the node to leaf node list if no other child block | |
3075 | * cached. | |
3076 | */ | |
3077 | if (list_empty(&upper->lower)) { | |
3078 | list_add_tail(&upper->lower, &cache->leaves); | |
3079 | upper->lowest = 1; | |
3080 | } | |
3081 | } | |
3082 | ||
3083 | btrfs_backref_drop_node(cache, node); | |
3084 | } | |
13fe1bdb QW |
3085 | |
3086 | /* | |
3087 | * Release all nodes/edges from current cache | |
3088 | */ | |
3089 | void btrfs_backref_release_cache(struct btrfs_backref_cache *cache) | |
3090 | { | |
3091 | struct btrfs_backref_node *node; | |
3092 | int i; | |
3093 | ||
3094 | while (!list_empty(&cache->detached)) { | |
3095 | node = list_entry(cache->detached.next, | |
3096 | struct btrfs_backref_node, list); | |
3097 | btrfs_backref_cleanup_node(cache, node); | |
3098 | } | |
3099 | ||
3100 | while (!list_empty(&cache->leaves)) { | |
3101 | node = list_entry(cache->leaves.next, | |
3102 | struct btrfs_backref_node, lower); | |
3103 | btrfs_backref_cleanup_node(cache, node); | |
3104 | } | |
3105 | ||
3106 | cache->last_trans = 0; | |
3107 | ||
3108 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
3109 | ASSERT(list_empty(&cache->pending[i])); | |
3110 | ASSERT(list_empty(&cache->pending_edge)); | |
3111 | ASSERT(list_empty(&cache->useless_node)); | |
3112 | ASSERT(list_empty(&cache->changed)); | |
3113 | ASSERT(list_empty(&cache->detached)); | |
3114 | ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); | |
3115 | ASSERT(!cache->nr_nodes); | |
3116 | ASSERT(!cache->nr_edges); | |
3117 | } | |
1b60d2ec QW |
3118 | |
3119 | /* | |
3120 | * Handle direct tree backref | |
3121 | * | |
3122 | * Direct tree backref means, the backref item shows its parent bytenr | |
3123 | * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined). | |
3124 | * | |
3125 | * @ref_key: The converted backref key. | |
3126 | * For keyed backref, it's the item key. | |
3127 | * For inlined backref, objectid is the bytenr, | |
3128 | * type is btrfs_inline_ref_type, offset is | |
3129 | * btrfs_inline_ref_offset. | |
3130 | */ | |
3131 | static int handle_direct_tree_backref(struct btrfs_backref_cache *cache, | |
3132 | struct btrfs_key *ref_key, | |
3133 | struct btrfs_backref_node *cur) | |
3134 | { | |
3135 | struct btrfs_backref_edge *edge; | |
3136 | struct btrfs_backref_node *upper; | |
3137 | struct rb_node *rb_node; | |
3138 | ||
3139 | ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
3140 | ||
3141 | /* Only reloc root uses backref pointing to itself */ | |
3142 | if (ref_key->objectid == ref_key->offset) { | |
3143 | struct btrfs_root *root; | |
3144 | ||
3145 | cur->is_reloc_root = 1; | |
3146 | /* Only reloc backref cache cares about a specific root */ | |
3147 | if (cache->is_reloc) { | |
3148 | root = find_reloc_root(cache->fs_info, cur->bytenr); | |
f78743fb | 3149 | if (!root) |
1b60d2ec QW |
3150 | return -ENOENT; |
3151 | cur->root = root; | |
3152 | } else { | |
3153 | /* | |
3154 | * For generic purpose backref cache, reloc root node | |
3155 | * is useless. | |
3156 | */ | |
3157 | list_add(&cur->list, &cache->useless_node); | |
3158 | } | |
3159 | return 0; | |
3160 | } | |
3161 | ||
3162 | edge = btrfs_backref_alloc_edge(cache); | |
3163 | if (!edge) | |
3164 | return -ENOMEM; | |
3165 | ||
3166 | rb_node = rb_simple_search(&cache->rb_root, ref_key->offset); | |
3167 | if (!rb_node) { | |
3168 | /* Parent node not yet cached */ | |
3169 | upper = btrfs_backref_alloc_node(cache, ref_key->offset, | |
3170 | cur->level + 1); | |
3171 | if (!upper) { | |
3172 | btrfs_backref_free_edge(cache, edge); | |
3173 | return -ENOMEM; | |
3174 | } | |
3175 | ||
3176 | /* | |
3177 | * Backrefs for the upper level block isn't cached, add the | |
3178 | * block to pending list | |
3179 | */ | |
3180 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
3181 | } else { | |
3182 | /* Parent node already cached */ | |
3183 | upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node); | |
3184 | ASSERT(upper->checked); | |
3185 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3186 | } | |
3187 | btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER); | |
3188 | return 0; | |
3189 | } | |
3190 | ||
3191 | /* | |
3192 | * Handle indirect tree backref | |
3193 | * | |
3194 | * Indirect tree backref means, we only know which tree the node belongs to. | |
3195 | * We still need to do a tree search to find out the parents. This is for | |
3196 | * TREE_BLOCK_REF backref (keyed or inlined). | |
3197 | * | |
3198 | * @ref_key: The same as @ref_key in handle_direct_tree_backref() | |
3199 | * @tree_key: The first key of this tree block. | |
1a9fd417 | 3200 | * @path: A clean (released) path, to avoid allocating path every time |
1b60d2ec QW |
3201 | * the function get called. |
3202 | */ | |
3203 | static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache, | |
3204 | struct btrfs_path *path, | |
3205 | struct btrfs_key *ref_key, | |
3206 | struct btrfs_key *tree_key, | |
3207 | struct btrfs_backref_node *cur) | |
3208 | { | |
3209 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3210 | struct btrfs_backref_node *upper; | |
3211 | struct btrfs_backref_node *lower; | |
3212 | struct btrfs_backref_edge *edge; | |
3213 | struct extent_buffer *eb; | |
3214 | struct btrfs_root *root; | |
1b60d2ec QW |
3215 | struct rb_node *rb_node; |
3216 | int level; | |
3217 | bool need_check = true; | |
3218 | int ret; | |
3219 | ||
56e9357a | 3220 | root = btrfs_get_fs_root(fs_info, ref_key->offset, false); |
1b60d2ec QW |
3221 | if (IS_ERR(root)) |
3222 | return PTR_ERR(root); | |
92a7cc42 | 3223 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
3224 | cur->cowonly = 1; |
3225 | ||
3226 | if (btrfs_root_level(&root->root_item) == cur->level) { | |
3227 | /* Tree root */ | |
3228 | ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr); | |
876de781 QW |
3229 | /* |
3230 | * For reloc backref cache, we may ignore reloc root. But for | |
3231 | * general purpose backref cache, we can't rely on | |
3232 | * btrfs_should_ignore_reloc_root() as it may conflict with | |
3233 | * current running relocation and lead to missing root. | |
3234 | * | |
3235 | * For general purpose backref cache, reloc root detection is | |
3236 | * completely relying on direct backref (key->offset is parent | |
3237 | * bytenr), thus only do such check for reloc cache. | |
3238 | */ | |
3239 | if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) { | |
1b60d2ec QW |
3240 | btrfs_put_root(root); |
3241 | list_add(&cur->list, &cache->useless_node); | |
3242 | } else { | |
3243 | cur->root = root; | |
3244 | } | |
3245 | return 0; | |
3246 | } | |
3247 | ||
3248 | level = cur->level + 1; | |
3249 | ||
3250 | /* Search the tree to find parent blocks referring to the block */ | |
3251 | path->search_commit_root = 1; | |
3252 | path->skip_locking = 1; | |
3253 | path->lowest_level = level; | |
3254 | ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0); | |
3255 | path->lowest_level = 0; | |
3256 | if (ret < 0) { | |
3257 | btrfs_put_root(root); | |
3258 | return ret; | |
3259 | } | |
3260 | if (ret > 0 && path->slots[level] > 0) | |
3261 | path->slots[level]--; | |
3262 | ||
3263 | eb = path->nodes[level]; | |
3264 | if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) { | |
3265 | btrfs_err(fs_info, | |
3266 | "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", | |
3267 | cur->bytenr, level - 1, root->root_key.objectid, | |
3268 | tree_key->objectid, tree_key->type, tree_key->offset); | |
3269 | btrfs_put_root(root); | |
3270 | ret = -ENOENT; | |
3271 | goto out; | |
3272 | } | |
3273 | lower = cur; | |
3274 | ||
3275 | /* Add all nodes and edges in the path */ | |
3276 | for (; level < BTRFS_MAX_LEVEL; level++) { | |
3277 | if (!path->nodes[level]) { | |
3278 | ASSERT(btrfs_root_bytenr(&root->root_item) == | |
3279 | lower->bytenr); | |
876de781 QW |
3280 | /* Same as previous should_ignore_reloc_root() call */ |
3281 | if (btrfs_should_ignore_reloc_root(root) && | |
3282 | cache->is_reloc) { | |
1b60d2ec QW |
3283 | btrfs_put_root(root); |
3284 | list_add(&lower->list, &cache->useless_node); | |
3285 | } else { | |
3286 | lower->root = root; | |
3287 | } | |
3288 | break; | |
3289 | } | |
3290 | ||
3291 | edge = btrfs_backref_alloc_edge(cache); | |
3292 | if (!edge) { | |
3293 | btrfs_put_root(root); | |
3294 | ret = -ENOMEM; | |
3295 | goto out; | |
3296 | } | |
3297 | ||
3298 | eb = path->nodes[level]; | |
3299 | rb_node = rb_simple_search(&cache->rb_root, eb->start); | |
3300 | if (!rb_node) { | |
3301 | upper = btrfs_backref_alloc_node(cache, eb->start, | |
3302 | lower->level + 1); | |
3303 | if (!upper) { | |
3304 | btrfs_put_root(root); | |
3305 | btrfs_backref_free_edge(cache, edge); | |
3306 | ret = -ENOMEM; | |
3307 | goto out; | |
3308 | } | |
3309 | upper->owner = btrfs_header_owner(eb); | |
92a7cc42 | 3310 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
3311 | upper->cowonly = 1; |
3312 | ||
3313 | /* | |
3314 | * If we know the block isn't shared we can avoid | |
3315 | * checking its backrefs. | |
3316 | */ | |
3317 | if (btrfs_block_can_be_shared(root, eb)) | |
3318 | upper->checked = 0; | |
3319 | else | |
3320 | upper->checked = 1; | |
3321 | ||
3322 | /* | |
3323 | * Add the block to pending list if we need to check its | |
3324 | * backrefs, we only do this once while walking up a | |
3325 | * tree as we will catch anything else later on. | |
3326 | */ | |
3327 | if (!upper->checked && need_check) { | |
3328 | need_check = false; | |
3329 | list_add_tail(&edge->list[UPPER], | |
3330 | &cache->pending_edge); | |
3331 | } else { | |
3332 | if (upper->checked) | |
3333 | need_check = true; | |
3334 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3335 | } | |
3336 | } else { | |
3337 | upper = rb_entry(rb_node, struct btrfs_backref_node, | |
3338 | rb_node); | |
3339 | ASSERT(upper->checked); | |
3340 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3341 | if (!upper->owner) | |
3342 | upper->owner = btrfs_header_owner(eb); | |
3343 | } | |
3344 | btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER); | |
3345 | ||
3346 | if (rb_node) { | |
3347 | btrfs_put_root(root); | |
3348 | break; | |
3349 | } | |
3350 | lower = upper; | |
3351 | upper = NULL; | |
3352 | } | |
3353 | out: | |
3354 | btrfs_release_path(path); | |
3355 | return ret; | |
3356 | } | |
3357 | ||
3358 | /* | |
3359 | * Add backref node @cur into @cache. | |
3360 | * | |
3361 | * NOTE: Even if the function returned 0, @cur is not yet cached as its upper | |
3362 | * links aren't yet bi-directional. Needs to finish such links. | |
fc997ed0 | 3363 | * Use btrfs_backref_finish_upper_links() to finish such linkage. |
1b60d2ec QW |
3364 | * |
3365 | * @path: Released path for indirect tree backref lookup | |
3366 | * @iter: Released backref iter for extent tree search | |
3367 | * @node_key: The first key of the tree block | |
3368 | */ | |
3369 | int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache, | |
3370 | struct btrfs_path *path, | |
3371 | struct btrfs_backref_iter *iter, | |
3372 | struct btrfs_key *node_key, | |
3373 | struct btrfs_backref_node *cur) | |
3374 | { | |
3375 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3376 | struct btrfs_backref_edge *edge; | |
3377 | struct btrfs_backref_node *exist; | |
3378 | int ret; | |
3379 | ||
3380 | ret = btrfs_backref_iter_start(iter, cur->bytenr); | |
3381 | if (ret < 0) | |
3382 | return ret; | |
3383 | /* | |
3384 | * We skip the first btrfs_tree_block_info, as we don't use the key | |
3385 | * stored in it, but fetch it from the tree block | |
3386 | */ | |
3387 | if (btrfs_backref_has_tree_block_info(iter)) { | |
3388 | ret = btrfs_backref_iter_next(iter); | |
3389 | if (ret < 0) | |
3390 | goto out; | |
3391 | /* No extra backref? This means the tree block is corrupted */ | |
3392 | if (ret > 0) { | |
3393 | ret = -EUCLEAN; | |
3394 | goto out; | |
3395 | } | |
3396 | } | |
3397 | WARN_ON(cur->checked); | |
3398 | if (!list_empty(&cur->upper)) { | |
3399 | /* | |
3400 | * The backref was added previously when processing backref of | |
3401 | * type BTRFS_TREE_BLOCK_REF_KEY | |
3402 | */ | |
3403 | ASSERT(list_is_singular(&cur->upper)); | |
3404 | edge = list_entry(cur->upper.next, struct btrfs_backref_edge, | |
3405 | list[LOWER]); | |
3406 | ASSERT(list_empty(&edge->list[UPPER])); | |
3407 | exist = edge->node[UPPER]; | |
3408 | /* | |
3409 | * Add the upper level block to pending list if we need check | |
3410 | * its backrefs | |
3411 | */ | |
3412 | if (!exist->checked) | |
3413 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
3414 | } else { | |
3415 | exist = NULL; | |
3416 | } | |
3417 | ||
3418 | for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { | |
3419 | struct extent_buffer *eb; | |
3420 | struct btrfs_key key; | |
3421 | int type; | |
3422 | ||
3423 | cond_resched(); | |
3424 | eb = btrfs_backref_get_eb(iter); | |
3425 | ||
3426 | key.objectid = iter->bytenr; | |
3427 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
3428 | struct btrfs_extent_inline_ref *iref; | |
3429 | ||
3430 | /* Update key for inline backref */ | |
3431 | iref = (struct btrfs_extent_inline_ref *) | |
3432 | ((unsigned long)iter->cur_ptr); | |
3433 | type = btrfs_get_extent_inline_ref_type(eb, iref, | |
3434 | BTRFS_REF_TYPE_BLOCK); | |
3435 | if (type == BTRFS_REF_TYPE_INVALID) { | |
3436 | ret = -EUCLEAN; | |
3437 | goto out; | |
3438 | } | |
3439 | key.type = type; | |
3440 | key.offset = btrfs_extent_inline_ref_offset(eb, iref); | |
3441 | } else { | |
3442 | key.type = iter->cur_key.type; | |
3443 | key.offset = iter->cur_key.offset; | |
3444 | } | |
3445 | ||
3446 | /* | |
3447 | * Parent node found and matches current inline ref, no need to | |
3448 | * rebuild this node for this inline ref | |
3449 | */ | |
3450 | if (exist && | |
3451 | ((key.type == BTRFS_TREE_BLOCK_REF_KEY && | |
3452 | exist->owner == key.offset) || | |
3453 | (key.type == BTRFS_SHARED_BLOCK_REF_KEY && | |
3454 | exist->bytenr == key.offset))) { | |
3455 | exist = NULL; | |
3456 | continue; | |
3457 | } | |
3458 | ||
3459 | /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ | |
3460 | if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
3461 | ret = handle_direct_tree_backref(cache, &key, cur); | |
3462 | if (ret < 0) | |
3463 | goto out; | |
3464 | continue; | |
3465 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { | |
3466 | ret = -EINVAL; | |
3467 | btrfs_print_v0_err(fs_info); | |
3468 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
3469 | goto out; | |
3470 | } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { | |
3471 | continue; | |
3472 | } | |
3473 | ||
3474 | /* | |
3475 | * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset | |
3476 | * means the root objectid. We need to search the tree to get | |
3477 | * its parent bytenr. | |
3478 | */ | |
3479 | ret = handle_indirect_tree_backref(cache, path, &key, node_key, | |
3480 | cur); | |
3481 | if (ret < 0) | |
3482 | goto out; | |
3483 | } | |
3484 | ret = 0; | |
3485 | cur->checked = 1; | |
3486 | WARN_ON(exist); | |
3487 | out: | |
3488 | btrfs_backref_iter_release(iter); | |
3489 | return ret; | |
3490 | } | |
fc997ed0 QW |
3491 | |
3492 | /* | |
3493 | * Finish the upwards linkage created by btrfs_backref_add_tree_node() | |
3494 | */ | |
3495 | int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache, | |
3496 | struct btrfs_backref_node *start) | |
3497 | { | |
3498 | struct list_head *useless_node = &cache->useless_node; | |
3499 | struct btrfs_backref_edge *edge; | |
3500 | struct rb_node *rb_node; | |
3501 | LIST_HEAD(pending_edge); | |
3502 | ||
3503 | ASSERT(start->checked); | |
3504 | ||
3505 | /* Insert this node to cache if it's not COW-only */ | |
3506 | if (!start->cowonly) { | |
3507 | rb_node = rb_simple_insert(&cache->rb_root, start->bytenr, | |
3508 | &start->rb_node); | |
3509 | if (rb_node) | |
3510 | btrfs_backref_panic(cache->fs_info, start->bytenr, | |
3511 | -EEXIST); | |
3512 | list_add_tail(&start->lower, &cache->leaves); | |
3513 | } | |
3514 | ||
3515 | /* | |
3516 | * Use breadth first search to iterate all related edges. | |
3517 | * | |
3518 | * The starting points are all the edges of this node | |
3519 | */ | |
3520 | list_for_each_entry(edge, &start->upper, list[LOWER]) | |
3521 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3522 | ||
3523 | while (!list_empty(&pending_edge)) { | |
3524 | struct btrfs_backref_node *upper; | |
3525 | struct btrfs_backref_node *lower; | |
fc997ed0 QW |
3526 | |
3527 | edge = list_first_entry(&pending_edge, | |
3528 | struct btrfs_backref_edge, list[UPPER]); | |
3529 | list_del_init(&edge->list[UPPER]); | |
3530 | upper = edge->node[UPPER]; | |
3531 | lower = edge->node[LOWER]; | |
3532 | ||
3533 | /* Parent is detached, no need to keep any edges */ | |
3534 | if (upper->detached) { | |
3535 | list_del(&edge->list[LOWER]); | |
3536 | btrfs_backref_free_edge(cache, edge); | |
3537 | ||
3538 | /* Lower node is orphan, queue for cleanup */ | |
3539 | if (list_empty(&lower->upper)) | |
3540 | list_add(&lower->list, useless_node); | |
3541 | continue; | |
3542 | } | |
3543 | ||
3544 | /* | |
3545 | * All new nodes added in current build_backref_tree() haven't | |
3546 | * been linked to the cache rb tree. | |
3547 | * So if we have upper->rb_node populated, this means a cache | |
3548 | * hit. We only need to link the edge, as @upper and all its | |
3549 | * parents have already been linked. | |
3550 | */ | |
3551 | if (!RB_EMPTY_NODE(&upper->rb_node)) { | |
3552 | if (upper->lowest) { | |
3553 | list_del_init(&upper->lower); | |
3554 | upper->lowest = 0; | |
3555 | } | |
3556 | ||
3557 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3558 | continue; | |
3559 | } | |
3560 | ||
3561 | /* Sanity check, we shouldn't have any unchecked nodes */ | |
3562 | if (!upper->checked) { | |
3563 | ASSERT(0); | |
3564 | return -EUCLEAN; | |
3565 | } | |
3566 | ||
3567 | /* Sanity check, COW-only node has non-COW-only parent */ | |
3568 | if (start->cowonly != upper->cowonly) { | |
3569 | ASSERT(0); | |
3570 | return -EUCLEAN; | |
3571 | } | |
3572 | ||
3573 | /* Only cache non-COW-only (subvolume trees) tree blocks */ | |
3574 | if (!upper->cowonly) { | |
3575 | rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr, | |
3576 | &upper->rb_node); | |
3577 | if (rb_node) { | |
3578 | btrfs_backref_panic(cache->fs_info, | |
3579 | upper->bytenr, -EEXIST); | |
3580 | return -EUCLEAN; | |
3581 | } | |
3582 | } | |
3583 | ||
3584 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3585 | ||
3586 | /* | |
3587 | * Also queue all the parent edges of this uncached node | |
3588 | * to finish the upper linkage | |
3589 | */ | |
3590 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3591 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3592 | } | |
3593 | return 0; | |
3594 | } | |
1b23ea18 QW |
3595 | |
3596 | void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache, | |
3597 | struct btrfs_backref_node *node) | |
3598 | { | |
3599 | struct btrfs_backref_node *lower; | |
3600 | struct btrfs_backref_node *upper; | |
3601 | struct btrfs_backref_edge *edge; | |
3602 | ||
3603 | while (!list_empty(&cache->useless_node)) { | |
3604 | lower = list_first_entry(&cache->useless_node, | |
3605 | struct btrfs_backref_node, list); | |
3606 | list_del_init(&lower->list); | |
3607 | } | |
3608 | while (!list_empty(&cache->pending_edge)) { | |
3609 | edge = list_first_entry(&cache->pending_edge, | |
3610 | struct btrfs_backref_edge, list[UPPER]); | |
3611 | list_del(&edge->list[UPPER]); | |
3612 | list_del(&edge->list[LOWER]); | |
3613 | lower = edge->node[LOWER]; | |
3614 | upper = edge->node[UPPER]; | |
3615 | btrfs_backref_free_edge(cache, edge); | |
3616 | ||
3617 | /* | |
3618 | * Lower is no longer linked to any upper backref nodes and | |
3619 | * isn't in the cache, we can free it ourselves. | |
3620 | */ | |
3621 | if (list_empty(&lower->upper) && | |
3622 | RB_EMPTY_NODE(&lower->rb_node)) | |
3623 | list_add(&lower->list, &cache->useless_node); | |
3624 | ||
3625 | if (!RB_EMPTY_NODE(&upper->rb_node)) | |
3626 | continue; | |
3627 | ||
3628 | /* Add this guy's upper edges to the list to process */ | |
3629 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3630 | list_add_tail(&edge->list[UPPER], | |
3631 | &cache->pending_edge); | |
3632 | if (list_empty(&upper->upper)) | |
3633 | list_add(&upper->list, &cache->useless_node); | |
3634 | } | |
3635 | ||
3636 | while (!list_empty(&cache->useless_node)) { | |
3637 | lower = list_first_entry(&cache->useless_node, | |
3638 | struct btrfs_backref_node, list); | |
3639 | list_del_init(&lower->list); | |
3640 | if (lower == node) | |
3641 | node = NULL; | |
49ecc679 | 3642 | btrfs_backref_drop_node(cache, lower); |
1b23ea18 QW |
3643 | } |
3644 | ||
3645 | btrfs_backref_cleanup_node(cache, node); | |
3646 | ASSERT(list_empty(&cache->useless_node) && | |
3647 | list_empty(&cache->pending_edge)); | |
3648 | } |