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" |
a542ad1b | 16 | |
dc046b10 JB |
17 | /* Just an arbitrary number so we can be sure this happened */ |
18 | #define BACKREF_FOUND_SHARED 6 | |
19 | ||
976b1908 JS |
20 | struct extent_inode_elem { |
21 | u64 inum; | |
22 | u64 offset; | |
23 | struct extent_inode_elem *next; | |
24 | }; | |
25 | ||
73980bec JM |
26 | static int check_extent_in_eb(const struct btrfs_key *key, |
27 | const struct extent_buffer *eb, | |
28 | const struct btrfs_file_extent_item *fi, | |
29 | u64 extent_item_pos, | |
c995ab3c ZB |
30 | struct extent_inode_elem **eie, |
31 | bool ignore_offset) | |
976b1908 | 32 | { |
8ca15e05 | 33 | u64 offset = 0; |
976b1908 JS |
34 | struct extent_inode_elem *e; |
35 | ||
c995ab3c ZB |
36 | if (!ignore_offset && |
37 | !btrfs_file_extent_compression(eb, fi) && | |
8ca15e05 JB |
38 | !btrfs_file_extent_encryption(eb, fi) && |
39 | !btrfs_file_extent_other_encoding(eb, fi)) { | |
40 | u64 data_offset; | |
41 | u64 data_len; | |
976b1908 | 42 | |
8ca15e05 JB |
43 | data_offset = btrfs_file_extent_offset(eb, fi); |
44 | data_len = btrfs_file_extent_num_bytes(eb, fi); | |
45 | ||
46 | if (extent_item_pos < data_offset || | |
47 | extent_item_pos >= data_offset + data_len) | |
48 | return 1; | |
49 | offset = extent_item_pos - data_offset; | |
50 | } | |
976b1908 JS |
51 | |
52 | e = kmalloc(sizeof(*e), GFP_NOFS); | |
53 | if (!e) | |
54 | return -ENOMEM; | |
55 | ||
56 | e->next = *eie; | |
57 | e->inum = key->objectid; | |
8ca15e05 | 58 | e->offset = key->offset + offset; |
976b1908 JS |
59 | *eie = e; |
60 | ||
61 | return 0; | |
62 | } | |
63 | ||
f05c4746 WS |
64 | static void free_inode_elem_list(struct extent_inode_elem *eie) |
65 | { | |
66 | struct extent_inode_elem *eie_next; | |
67 | ||
68 | for (; eie; eie = eie_next) { | |
69 | eie_next = eie->next; | |
70 | kfree(eie); | |
71 | } | |
72 | } | |
73 | ||
73980bec JM |
74 | static int find_extent_in_eb(const struct extent_buffer *eb, |
75 | u64 wanted_disk_byte, u64 extent_item_pos, | |
c995ab3c ZB |
76 | struct extent_inode_elem **eie, |
77 | bool ignore_offset) | |
976b1908 JS |
78 | { |
79 | u64 disk_byte; | |
80 | struct btrfs_key key; | |
81 | struct btrfs_file_extent_item *fi; | |
82 | int slot; | |
83 | int nritems; | |
84 | int extent_type; | |
85 | int ret; | |
86 | ||
87 | /* | |
88 | * from the shared data ref, we only have the leaf but we need | |
89 | * the key. thus, we must look into all items and see that we | |
90 | * find one (some) with a reference to our extent item. | |
91 | */ | |
92 | nritems = btrfs_header_nritems(eb); | |
93 | for (slot = 0; slot < nritems; ++slot) { | |
94 | btrfs_item_key_to_cpu(eb, &key, slot); | |
95 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
96 | continue; | |
97 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
98 | extent_type = btrfs_file_extent_type(eb, fi); | |
99 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
100 | continue; | |
101 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
102 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
103 | if (disk_byte != wanted_disk_byte) | |
104 | continue; | |
105 | ||
c995ab3c | 106 | ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset); |
976b1908 JS |
107 | if (ret < 0) |
108 | return ret; | |
109 | } | |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
86d5f994 | 114 | struct preftree { |
ecf160b4 | 115 | struct rb_root_cached root; |
6c336b21 | 116 | unsigned int count; |
86d5f994 EN |
117 | }; |
118 | ||
ecf160b4 | 119 | #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 } |
86d5f994 EN |
120 | |
121 | struct preftrees { | |
122 | struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */ | |
123 | struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */ | |
124 | struct preftree indirect_missing_keys; | |
125 | }; | |
126 | ||
3ec4d323 EN |
127 | /* |
128 | * Checks for a shared extent during backref search. | |
129 | * | |
130 | * The share_count tracks prelim_refs (direct and indirect) having a | |
131 | * ref->count >0: | |
132 | * - incremented when a ref->count transitions to >0 | |
133 | * - decremented when a ref->count transitions to <1 | |
134 | */ | |
135 | struct share_check { | |
136 | u64 root_objectid; | |
137 | u64 inum; | |
138 | int share_count; | |
139 | }; | |
140 | ||
141 | static inline int extent_is_shared(struct share_check *sc) | |
142 | { | |
143 | return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0; | |
144 | } | |
145 | ||
b9e9a6cb WS |
146 | static struct kmem_cache *btrfs_prelim_ref_cache; |
147 | ||
148 | int __init btrfs_prelim_ref_init(void) | |
149 | { | |
150 | btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", | |
e0c476b1 | 151 | sizeof(struct prelim_ref), |
b9e9a6cb | 152 | 0, |
fba4b697 | 153 | SLAB_MEM_SPREAD, |
b9e9a6cb WS |
154 | NULL); |
155 | if (!btrfs_prelim_ref_cache) | |
156 | return -ENOMEM; | |
157 | return 0; | |
158 | } | |
159 | ||
e67c718b | 160 | void __cold btrfs_prelim_ref_exit(void) |
b9e9a6cb | 161 | { |
5598e900 | 162 | kmem_cache_destroy(btrfs_prelim_ref_cache); |
b9e9a6cb WS |
163 | } |
164 | ||
86d5f994 EN |
165 | static void free_pref(struct prelim_ref *ref) |
166 | { | |
167 | kmem_cache_free(btrfs_prelim_ref_cache, ref); | |
168 | } | |
169 | ||
170 | /* | |
171 | * Return 0 when both refs are for the same block (and can be merged). | |
172 | * A -1 return indicates ref1 is a 'lower' block than ref2, while 1 | |
173 | * indicates a 'higher' block. | |
174 | */ | |
175 | static int prelim_ref_compare(struct prelim_ref *ref1, | |
176 | struct prelim_ref *ref2) | |
177 | { | |
178 | if (ref1->level < ref2->level) | |
179 | return -1; | |
180 | if (ref1->level > ref2->level) | |
181 | return 1; | |
182 | if (ref1->root_id < ref2->root_id) | |
183 | return -1; | |
184 | if (ref1->root_id > ref2->root_id) | |
185 | return 1; | |
186 | if (ref1->key_for_search.type < ref2->key_for_search.type) | |
187 | return -1; | |
188 | if (ref1->key_for_search.type > ref2->key_for_search.type) | |
189 | return 1; | |
190 | if (ref1->key_for_search.objectid < ref2->key_for_search.objectid) | |
191 | return -1; | |
192 | if (ref1->key_for_search.objectid > ref2->key_for_search.objectid) | |
193 | return 1; | |
194 | if (ref1->key_for_search.offset < ref2->key_for_search.offset) | |
195 | return -1; | |
196 | if (ref1->key_for_search.offset > ref2->key_for_search.offset) | |
197 | return 1; | |
198 | if (ref1->parent < ref2->parent) | |
199 | return -1; | |
200 | if (ref1->parent > ref2->parent) | |
201 | return 1; | |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
ccc8dc75 CIK |
206 | static void update_share_count(struct share_check *sc, int oldcount, |
207 | int newcount) | |
3ec4d323 EN |
208 | { |
209 | if ((!sc) || (oldcount == 0 && newcount < 1)) | |
210 | return; | |
211 | ||
212 | if (oldcount > 0 && newcount < 1) | |
213 | sc->share_count--; | |
214 | else if (oldcount < 1 && newcount > 0) | |
215 | sc->share_count++; | |
216 | } | |
217 | ||
86d5f994 EN |
218 | /* |
219 | * Add @newref to the @root rbtree, merging identical refs. | |
220 | * | |
3ec4d323 | 221 | * Callers should assume that newref has been freed after calling. |
86d5f994 | 222 | */ |
00142756 JM |
223 | static void prelim_ref_insert(const struct btrfs_fs_info *fs_info, |
224 | struct preftree *preftree, | |
3ec4d323 EN |
225 | struct prelim_ref *newref, |
226 | struct share_check *sc) | |
86d5f994 | 227 | { |
ecf160b4 | 228 | struct rb_root_cached *root; |
86d5f994 EN |
229 | struct rb_node **p; |
230 | struct rb_node *parent = NULL; | |
231 | struct prelim_ref *ref; | |
232 | int result; | |
ecf160b4 | 233 | bool leftmost = true; |
86d5f994 EN |
234 | |
235 | root = &preftree->root; | |
ecf160b4 | 236 | p = &root->rb_root.rb_node; |
86d5f994 EN |
237 | |
238 | while (*p) { | |
239 | parent = *p; | |
240 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
241 | result = prelim_ref_compare(ref, newref); | |
242 | if (result < 0) { | |
243 | p = &(*p)->rb_left; | |
244 | } else if (result > 0) { | |
245 | p = &(*p)->rb_right; | |
ecf160b4 | 246 | leftmost = false; |
86d5f994 EN |
247 | } else { |
248 | /* Identical refs, merge them and free @newref */ | |
249 | struct extent_inode_elem *eie = ref->inode_list; | |
250 | ||
251 | while (eie && eie->next) | |
252 | eie = eie->next; | |
253 | ||
254 | if (!eie) | |
255 | ref->inode_list = newref->inode_list; | |
256 | else | |
257 | eie->next = newref->inode_list; | |
00142756 JM |
258 | trace_btrfs_prelim_ref_merge(fs_info, ref, newref, |
259 | preftree->count); | |
3ec4d323 EN |
260 | /* |
261 | * A delayed ref can have newref->count < 0. | |
262 | * The ref->count is updated to follow any | |
263 | * BTRFS_[ADD|DROP]_DELAYED_REF actions. | |
264 | */ | |
265 | update_share_count(sc, ref->count, | |
266 | ref->count + newref->count); | |
86d5f994 EN |
267 | ref->count += newref->count; |
268 | free_pref(newref); | |
269 | return; | |
270 | } | |
271 | } | |
272 | ||
3ec4d323 | 273 | update_share_count(sc, 0, newref->count); |
6c336b21 | 274 | preftree->count++; |
00142756 | 275 | trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count); |
86d5f994 | 276 | rb_link_node(&newref->rbnode, parent, p); |
ecf160b4 | 277 | rb_insert_color_cached(&newref->rbnode, root, leftmost); |
86d5f994 EN |
278 | } |
279 | ||
280 | /* | |
281 | * Release the entire tree. We don't care about internal consistency so | |
282 | * just free everything and then reset the tree root. | |
283 | */ | |
284 | static void prelim_release(struct preftree *preftree) | |
285 | { | |
286 | struct prelim_ref *ref, *next_ref; | |
287 | ||
ecf160b4 LB |
288 | rbtree_postorder_for_each_entry_safe(ref, next_ref, |
289 | &preftree->root.rb_root, rbnode) | |
86d5f994 EN |
290 | free_pref(ref); |
291 | ||
ecf160b4 | 292 | preftree->root = RB_ROOT_CACHED; |
6c336b21 | 293 | preftree->count = 0; |
86d5f994 EN |
294 | } |
295 | ||
d5c88b73 JS |
296 | /* |
297 | * the rules for all callers of this function are: | |
298 | * - obtaining the parent is the goal | |
299 | * - if you add a key, you must know that it is a correct key | |
300 | * - if you cannot add the parent or a correct key, then we will look into the | |
301 | * block later to set a correct key | |
302 | * | |
303 | * delayed refs | |
304 | * ============ | |
305 | * backref type | shared | indirect | shared | indirect | |
306 | * information | tree | tree | data | data | |
307 | * --------------------+--------+----------+--------+---------- | |
308 | * parent logical | y | - | - | - | |
309 | * key to resolve | - | y | y | y | |
310 | * tree block logical | - | - | - | - | |
311 | * root for resolving | y | y | y | y | |
312 | * | |
313 | * - column 1: we've the parent -> done | |
314 | * - column 2, 3, 4: we use the key to find the parent | |
315 | * | |
316 | * on disk refs (inline or keyed) | |
317 | * ============================== | |
318 | * backref type | shared | indirect | shared | indirect | |
319 | * information | tree | tree | data | data | |
320 | * --------------------+--------+----------+--------+---------- | |
321 | * parent logical | y | - | y | - | |
322 | * key to resolve | - | - | - | y | |
323 | * tree block logical | y | y | y | y | |
324 | * root for resolving | - | y | y | y | |
325 | * | |
326 | * - column 1, 3: we've the parent -> done | |
327 | * - column 2: we take the first key from the block to find the parent | |
e0c476b1 | 328 | * (see add_missing_keys) |
d5c88b73 JS |
329 | * - column 4: we use the key to find the parent |
330 | * | |
331 | * additional information that's available but not required to find the parent | |
332 | * block might help in merging entries to gain some speed. | |
333 | */ | |
00142756 JM |
334 | static int add_prelim_ref(const struct btrfs_fs_info *fs_info, |
335 | struct preftree *preftree, u64 root_id, | |
e0c476b1 | 336 | const struct btrfs_key *key, int level, u64 parent, |
3ec4d323 EN |
337 | u64 wanted_disk_byte, int count, |
338 | struct share_check *sc, gfp_t gfp_mask) | |
8da6d581 | 339 | { |
e0c476b1 | 340 | struct prelim_ref *ref; |
8da6d581 | 341 | |
48ec4736 LB |
342 | if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) |
343 | return 0; | |
344 | ||
b9e9a6cb | 345 | ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); |
8da6d581 JS |
346 | if (!ref) |
347 | return -ENOMEM; | |
348 | ||
349 | ref->root_id = root_id; | |
7ac8b88e | 350 | if (key) |
d5c88b73 | 351 | ref->key_for_search = *key; |
7ac8b88e | 352 | else |
d5c88b73 | 353 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 354 | |
3301958b | 355 | ref->inode_list = NULL; |
8da6d581 JS |
356 | ref->level = level; |
357 | ref->count = count; | |
358 | ref->parent = parent; | |
359 | ref->wanted_disk_byte = wanted_disk_byte; | |
3ec4d323 EN |
360 | prelim_ref_insert(fs_info, preftree, ref, sc); |
361 | return extent_is_shared(sc); | |
8da6d581 JS |
362 | } |
363 | ||
86d5f994 | 364 | /* direct refs use root == 0, key == NULL */ |
00142756 JM |
365 | static int add_direct_ref(const struct btrfs_fs_info *fs_info, |
366 | struct preftrees *preftrees, int level, u64 parent, | |
3ec4d323 EN |
367 | u64 wanted_disk_byte, int count, |
368 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 | 369 | { |
00142756 | 370 | return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level, |
3ec4d323 | 371 | parent, wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
372 | } |
373 | ||
374 | /* indirect refs use parent == 0 */ | |
00142756 JM |
375 | static int add_indirect_ref(const struct btrfs_fs_info *fs_info, |
376 | struct preftrees *preftrees, u64 root_id, | |
86d5f994 | 377 | const struct btrfs_key *key, int level, |
3ec4d323 EN |
378 | u64 wanted_disk_byte, int count, |
379 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 EN |
380 | { |
381 | struct preftree *tree = &preftrees->indirect; | |
382 | ||
383 | if (!key) | |
384 | tree = &preftrees->indirect_missing_keys; | |
00142756 | 385 | return add_prelim_ref(fs_info, tree, root_id, key, level, 0, |
3ec4d323 | 386 | wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
387 | } |
388 | ||
ed58f2e6 | 389 | static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr) |
390 | { | |
391 | struct rb_node **p = &preftrees->direct.root.rb_root.rb_node; | |
392 | struct rb_node *parent = NULL; | |
393 | struct prelim_ref *ref = NULL; | |
394 | struct prelim_ref target = {0}; | |
395 | int result; | |
396 | ||
397 | target.parent = bytenr; | |
398 | ||
399 | while (*p) { | |
400 | parent = *p; | |
401 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
402 | result = prelim_ref_compare(ref, &target); | |
403 | ||
404 | if (result < 0) | |
405 | p = &(*p)->rb_left; | |
406 | else if (result > 0) | |
407 | p = &(*p)->rb_right; | |
408 | else | |
409 | return 1; | |
410 | } | |
411 | return 0; | |
412 | } | |
413 | ||
8da6d581 | 414 | static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, |
ed58f2e6 | 415 | struct ulist *parents, |
416 | struct preftrees *preftrees, struct prelim_ref *ref, | |
44853868 | 417 | int level, u64 time_seq, const u64 *extent_item_pos, |
c995ab3c | 418 | u64 total_refs, bool ignore_offset) |
8da6d581 | 419 | { |
69bca40d AB |
420 | int ret = 0; |
421 | int slot; | |
422 | struct extent_buffer *eb; | |
423 | struct btrfs_key key; | |
7ef81ac8 | 424 | struct btrfs_key *key_for_search = &ref->key_for_search; |
8da6d581 | 425 | struct btrfs_file_extent_item *fi; |
ed8c4913 | 426 | struct extent_inode_elem *eie = NULL, *old = NULL; |
8da6d581 | 427 | u64 disk_byte; |
7ef81ac8 JB |
428 | u64 wanted_disk_byte = ref->wanted_disk_byte; |
429 | u64 count = 0; | |
7ac8b88e | 430 | u64 data_offset; |
8da6d581 | 431 | |
69bca40d AB |
432 | if (level != 0) { |
433 | eb = path->nodes[level]; | |
434 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
435 | if (ret < 0) |
436 | return ret; | |
8da6d581 | 437 | return 0; |
69bca40d | 438 | } |
8da6d581 JS |
439 | |
440 | /* | |
ed58f2e6 | 441 | * 1. We normally enter this function with the path already pointing to |
442 | * the first item to check. But sometimes, we may enter it with | |
443 | * slot == nritems. | |
444 | * 2. We are searching for normal backref but bytenr of this leaf | |
445 | * matches shared data backref | |
cfc0eed0 | 446 | * 3. The leaf owner is not equal to the root we are searching |
447 | * | |
ed58f2e6 | 448 | * For these cases, go to the next leaf before we continue. |
8da6d581 | 449 | */ |
ed58f2e6 | 450 | eb = path->nodes[0]; |
451 | if (path->slots[0] >= btrfs_header_nritems(eb) || | |
cfc0eed0 | 452 | is_shared_data_backref(preftrees, eb->start) || |
453 | ref->root_id != btrfs_header_owner(eb)) { | |
de47c9d3 | 454 | if (time_seq == SEQ_LAST) |
21633fc6 QW |
455 | ret = btrfs_next_leaf(root, path); |
456 | else | |
457 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
458 | } | |
8da6d581 | 459 | |
44853868 | 460 | while (!ret && count < total_refs) { |
8da6d581 | 461 | eb = path->nodes[0]; |
69bca40d AB |
462 | slot = path->slots[0]; |
463 | ||
464 | btrfs_item_key_to_cpu(eb, &key, slot); | |
465 | ||
466 | if (key.objectid != key_for_search->objectid || | |
467 | key.type != BTRFS_EXTENT_DATA_KEY) | |
468 | break; | |
469 | ||
ed58f2e6 | 470 | /* |
471 | * We are searching for normal backref but bytenr of this leaf | |
cfc0eed0 | 472 | * matches shared data backref, OR |
473 | * the leaf owner is not equal to the root we are searching for | |
ed58f2e6 | 474 | */ |
cfc0eed0 | 475 | if (slot == 0 && |
476 | (is_shared_data_backref(preftrees, eb->start) || | |
477 | ref->root_id != btrfs_header_owner(eb))) { | |
ed58f2e6 | 478 | if (time_seq == SEQ_LAST) |
479 | ret = btrfs_next_leaf(root, path); | |
480 | else | |
481 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
482 | continue; | |
483 | } | |
69bca40d AB |
484 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
485 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
7ac8b88e | 486 | data_offset = btrfs_file_extent_offset(eb, fi); |
69bca40d AB |
487 | |
488 | if (disk_byte == wanted_disk_byte) { | |
489 | eie = NULL; | |
ed8c4913 | 490 | old = NULL; |
7ac8b88e | 491 | if (ref->key_for_search.offset == key.offset - data_offset) |
492 | count++; | |
493 | else | |
494 | goto next; | |
69bca40d AB |
495 | if (extent_item_pos) { |
496 | ret = check_extent_in_eb(&key, eb, fi, | |
497 | *extent_item_pos, | |
c995ab3c | 498 | &eie, ignore_offset); |
69bca40d AB |
499 | if (ret < 0) |
500 | break; | |
501 | } | |
ed8c4913 JB |
502 | if (ret > 0) |
503 | goto next; | |
4eb1f66d TI |
504 | ret = ulist_add_merge_ptr(parents, eb->start, |
505 | eie, (void **)&old, GFP_NOFS); | |
ed8c4913 JB |
506 | if (ret < 0) |
507 | break; | |
508 | if (!ret && extent_item_pos) { | |
509 | while (old->next) | |
510 | old = old->next; | |
511 | old->next = eie; | |
69bca40d | 512 | } |
f05c4746 | 513 | eie = NULL; |
8da6d581 | 514 | } |
ed8c4913 | 515 | next: |
de47c9d3 | 516 | if (time_seq == SEQ_LAST) |
21633fc6 QW |
517 | ret = btrfs_next_item(root, path); |
518 | else | |
519 | ret = btrfs_next_old_item(root, path, time_seq); | |
8da6d581 JS |
520 | } |
521 | ||
69bca40d AB |
522 | if (ret > 0) |
523 | ret = 0; | |
f05c4746 WS |
524 | else if (ret < 0) |
525 | free_inode_elem_list(eie); | |
69bca40d | 526 | return ret; |
8da6d581 JS |
527 | } |
528 | ||
529 | /* | |
530 | * resolve an indirect backref in the form (root_id, key, level) | |
531 | * to a logical address | |
532 | */ | |
e0c476b1 JM |
533 | static int resolve_indirect_ref(struct btrfs_fs_info *fs_info, |
534 | struct btrfs_path *path, u64 time_seq, | |
ed58f2e6 | 535 | struct preftrees *preftrees, |
e0c476b1 | 536 | struct prelim_ref *ref, struct ulist *parents, |
c995ab3c ZB |
537 | const u64 *extent_item_pos, u64 total_refs, |
538 | bool ignore_offset) | |
8da6d581 | 539 | { |
8da6d581 JS |
540 | struct btrfs_root *root; |
541 | struct btrfs_key root_key; | |
8da6d581 JS |
542 | struct extent_buffer *eb; |
543 | int ret = 0; | |
544 | int root_level; | |
545 | int level = ref->level; | |
538f72cd | 546 | int index; |
7ac8b88e | 547 | struct btrfs_key search_key = ref->key_for_search; |
8da6d581 | 548 | |
8da6d581 JS |
549 | root_key.objectid = ref->root_id; |
550 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
551 | root_key.offset = (u64)-1; | |
538f72cd WS |
552 | |
553 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
554 | ||
2d9e9776 | 555 | root = btrfs_get_fs_root(fs_info, &root_key, false); |
8da6d581 | 556 | if (IS_ERR(root)) { |
538f72cd | 557 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
8da6d581 | 558 | ret = PTR_ERR(root); |
9326f76f JB |
559 | goto out_free; |
560 | } | |
561 | ||
f5ee5c9a | 562 | if (btrfs_is_testing(fs_info)) { |
d9ee522b JB |
563 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
564 | ret = -ENOENT; | |
565 | goto out; | |
566 | } | |
567 | ||
9e351cc8 JB |
568 | if (path->search_commit_root) |
569 | root_level = btrfs_header_level(root->commit_root); | |
de47c9d3 | 570 | else if (time_seq == SEQ_LAST) |
21633fc6 | 571 | root_level = btrfs_header_level(root->node); |
9e351cc8 JB |
572 | else |
573 | root_level = btrfs_old_root_level(root, time_seq); | |
8da6d581 | 574 | |
538f72cd WS |
575 | if (root_level + 1 == level) { |
576 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
8da6d581 | 577 | goto out; |
538f72cd | 578 | } |
8da6d581 | 579 | |
7ac8b88e | 580 | /* |
581 | * We can often find data backrefs with an offset that is too large | |
582 | * (>= LLONG_MAX, maximum allowed file offset) due to underflows when | |
583 | * subtracting a file's offset with the data offset of its | |
584 | * corresponding extent data item. This can happen for example in the | |
585 | * clone ioctl. | |
586 | * | |
587 | * So if we detect such case we set the search key's offset to zero to | |
588 | * make sure we will find the matching file extent item at | |
589 | * add_all_parents(), otherwise we will miss it because the offset | |
590 | * taken form the backref is much larger then the offset of the file | |
591 | * extent item. This can make us scan a very large number of file | |
592 | * extent items, but at least it will not make us miss any. | |
593 | * | |
594 | * This is an ugly workaround for a behaviour that should have never | |
595 | * existed, but it does and a fix for the clone ioctl would touch a lot | |
596 | * of places, cause backwards incompatibility and would not fix the | |
597 | * problem for extents cloned with older kernels. | |
598 | */ | |
599 | if (search_key.type == BTRFS_EXTENT_DATA_KEY && | |
600 | search_key.offset >= LLONG_MAX) | |
601 | search_key.offset = 0; | |
8da6d581 | 602 | path->lowest_level = level; |
de47c9d3 | 603 | if (time_seq == SEQ_LAST) |
7ac8b88e | 604 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); |
21633fc6 | 605 | else |
7ac8b88e | 606 | ret = btrfs_search_old_slot(root, &search_key, path, time_seq); |
538f72cd WS |
607 | |
608 | /* root node has been locked, we can release @subvol_srcu safely here */ | |
609 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
610 | ||
ab8d0fc4 JM |
611 | btrfs_debug(fs_info, |
612 | "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)", | |
c1c9ff7c GU |
613 | ref->root_id, level, ref->count, ret, |
614 | ref->key_for_search.objectid, ref->key_for_search.type, | |
615 | ref->key_for_search.offset); | |
8da6d581 JS |
616 | if (ret < 0) |
617 | goto out; | |
618 | ||
619 | eb = path->nodes[level]; | |
9345457f | 620 | while (!eb) { |
fae7f21c | 621 | if (WARN_ON(!level)) { |
9345457f JS |
622 | ret = 1; |
623 | goto out; | |
624 | } | |
625 | level--; | |
626 | eb = path->nodes[level]; | |
8da6d581 JS |
627 | } |
628 | ||
ed58f2e6 | 629 | ret = add_all_parents(root, path, parents, preftrees, ref, level, |
630 | time_seq, extent_item_pos, total_refs, ignore_offset); | |
8da6d581 | 631 | out: |
00246528 | 632 | btrfs_put_root(root); |
9326f76f | 633 | out_free: |
da61d31a JB |
634 | path->lowest_level = 0; |
635 | btrfs_release_path(path); | |
8da6d581 JS |
636 | return ret; |
637 | } | |
638 | ||
4dae077a JM |
639 | static struct extent_inode_elem * |
640 | unode_aux_to_inode_list(struct ulist_node *node) | |
641 | { | |
642 | if (!node) | |
643 | return NULL; | |
644 | return (struct extent_inode_elem *)(uintptr_t)node->aux; | |
645 | } | |
646 | ||
8da6d581 | 647 | /* |
52042d8e | 648 | * We maintain three separate rbtrees: one for direct refs, one for |
86d5f994 EN |
649 | * indirect refs which have a key, and one for indirect refs which do not |
650 | * have a key. Each tree does merge on insertion. | |
651 | * | |
652 | * Once all of the references are located, we iterate over the tree of | |
653 | * indirect refs with missing keys. An appropriate key is located and | |
654 | * the ref is moved onto the tree for indirect refs. After all missing | |
655 | * keys are thus located, we iterate over the indirect ref tree, resolve | |
656 | * each reference, and then insert the resolved reference onto the | |
657 | * direct tree (merging there too). | |
658 | * | |
659 | * New backrefs (i.e., for parent nodes) are added to the appropriate | |
660 | * rbtree as they are encountered. The new backrefs are subsequently | |
661 | * resolved as above. | |
8da6d581 | 662 | */ |
e0c476b1 JM |
663 | static int resolve_indirect_refs(struct btrfs_fs_info *fs_info, |
664 | struct btrfs_path *path, u64 time_seq, | |
86d5f994 | 665 | struct preftrees *preftrees, |
e0c476b1 | 666 | const u64 *extent_item_pos, u64 total_refs, |
c995ab3c | 667 | struct share_check *sc, bool ignore_offset) |
8da6d581 JS |
668 | { |
669 | int err; | |
670 | int ret = 0; | |
8da6d581 JS |
671 | struct ulist *parents; |
672 | struct ulist_node *node; | |
cd1b413c | 673 | struct ulist_iterator uiter; |
86d5f994 | 674 | struct rb_node *rnode; |
8da6d581 JS |
675 | |
676 | parents = ulist_alloc(GFP_NOFS); | |
677 | if (!parents) | |
678 | return -ENOMEM; | |
679 | ||
680 | /* | |
86d5f994 EN |
681 | * We could trade memory usage for performance here by iterating |
682 | * the tree, allocating new refs for each insertion, and then | |
683 | * freeing the entire indirect tree when we're done. In some test | |
684 | * cases, the tree can grow quite large (~200k objects). | |
8da6d581 | 685 | */ |
ecf160b4 | 686 | while ((rnode = rb_first_cached(&preftrees->indirect.root))) { |
86d5f994 EN |
687 | struct prelim_ref *ref; |
688 | ||
689 | ref = rb_entry(rnode, struct prelim_ref, rbnode); | |
690 | if (WARN(ref->parent, | |
691 | "BUG: direct ref found in indirect tree")) { | |
692 | ret = -EINVAL; | |
693 | goto out; | |
694 | } | |
695 | ||
ecf160b4 | 696 | rb_erase_cached(&ref->rbnode, &preftrees->indirect.root); |
6c336b21 | 697 | preftrees->indirect.count--; |
86d5f994 EN |
698 | |
699 | if (ref->count == 0) { | |
700 | free_pref(ref); | |
8da6d581 | 701 | continue; |
86d5f994 EN |
702 | } |
703 | ||
3ec4d323 EN |
704 | if (sc && sc->root_objectid && |
705 | ref->root_id != sc->root_objectid) { | |
86d5f994 | 706 | free_pref(ref); |
dc046b10 JB |
707 | ret = BACKREF_FOUND_SHARED; |
708 | goto out; | |
709 | } | |
ed58f2e6 | 710 | err = resolve_indirect_ref(fs_info, path, time_seq, preftrees, |
711 | ref, parents, extent_item_pos, | |
c995ab3c | 712 | total_refs, ignore_offset); |
95def2ed WS |
713 | /* |
714 | * we can only tolerate ENOENT,otherwise,we should catch error | |
715 | * and return directly. | |
716 | */ | |
717 | if (err == -ENOENT) { | |
3ec4d323 EN |
718 | prelim_ref_insert(fs_info, &preftrees->direct, ref, |
719 | NULL); | |
8da6d581 | 720 | continue; |
95def2ed | 721 | } else if (err) { |
86d5f994 | 722 | free_pref(ref); |
95def2ed WS |
723 | ret = err; |
724 | goto out; | |
725 | } | |
8da6d581 JS |
726 | |
727 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
728 | ULIST_ITER_INIT(&uiter); |
729 | node = ulist_next(parents, &uiter); | |
8da6d581 | 730 | ref->parent = node ? node->val : 0; |
4dae077a | 731 | ref->inode_list = unode_aux_to_inode_list(node); |
8da6d581 | 732 | |
86d5f994 | 733 | /* Add a prelim_ref(s) for any other parent(s). */ |
cd1b413c | 734 | while ((node = ulist_next(parents, &uiter))) { |
86d5f994 EN |
735 | struct prelim_ref *new_ref; |
736 | ||
b9e9a6cb WS |
737 | new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, |
738 | GFP_NOFS); | |
8da6d581 | 739 | if (!new_ref) { |
86d5f994 | 740 | free_pref(ref); |
8da6d581 | 741 | ret = -ENOMEM; |
e36902d4 | 742 | goto out; |
8da6d581 JS |
743 | } |
744 | memcpy(new_ref, ref, sizeof(*ref)); | |
745 | new_ref->parent = node->val; | |
4dae077a | 746 | new_ref->inode_list = unode_aux_to_inode_list(node); |
3ec4d323 EN |
747 | prelim_ref_insert(fs_info, &preftrees->direct, |
748 | new_ref, NULL); | |
8da6d581 | 749 | } |
86d5f994 | 750 | |
3ec4d323 | 751 | /* |
52042d8e | 752 | * Now it's a direct ref, put it in the direct tree. We must |
3ec4d323 EN |
753 | * do this last because the ref could be merged/freed here. |
754 | */ | |
755 | prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL); | |
86d5f994 | 756 | |
8da6d581 | 757 | ulist_reinit(parents); |
9dd14fd6 | 758 | cond_resched(); |
8da6d581 | 759 | } |
e36902d4 | 760 | out: |
8da6d581 JS |
761 | ulist_free(parents); |
762 | return ret; | |
763 | } | |
764 | ||
d5c88b73 JS |
765 | /* |
766 | * read tree blocks and add keys where required. | |
767 | */ | |
e0c476b1 | 768 | static int add_missing_keys(struct btrfs_fs_info *fs_info, |
38e3eebf | 769 | struct preftrees *preftrees, bool lock) |
d5c88b73 | 770 | { |
e0c476b1 | 771 | struct prelim_ref *ref; |
d5c88b73 | 772 | struct extent_buffer *eb; |
86d5f994 EN |
773 | struct preftree *tree = &preftrees->indirect_missing_keys; |
774 | struct rb_node *node; | |
d5c88b73 | 775 | |
ecf160b4 | 776 | while ((node = rb_first_cached(&tree->root))) { |
86d5f994 | 777 | ref = rb_entry(node, struct prelim_ref, rbnode); |
ecf160b4 | 778 | rb_erase_cached(node, &tree->root); |
86d5f994 EN |
779 | |
780 | BUG_ON(ref->parent); /* should not be a direct ref */ | |
781 | BUG_ON(ref->key_for_search.type); | |
d5c88b73 | 782 | BUG_ON(!ref->wanted_disk_byte); |
86d5f994 | 783 | |
581c1760 QW |
784 | eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0, |
785 | ref->level - 1, NULL); | |
64c043de | 786 | if (IS_ERR(eb)) { |
86d5f994 | 787 | free_pref(ref); |
64c043de LB |
788 | return PTR_ERR(eb); |
789 | } else if (!extent_buffer_uptodate(eb)) { | |
86d5f994 | 790 | free_pref(ref); |
416bc658 JB |
791 | free_extent_buffer(eb); |
792 | return -EIO; | |
793 | } | |
38e3eebf JB |
794 | if (lock) |
795 | btrfs_tree_read_lock(eb); | |
d5c88b73 JS |
796 | if (btrfs_header_level(eb) == 0) |
797 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
798 | else | |
799 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
38e3eebf JB |
800 | if (lock) |
801 | btrfs_tree_read_unlock(eb); | |
d5c88b73 | 802 | free_extent_buffer(eb); |
3ec4d323 | 803 | prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL); |
9dd14fd6 | 804 | cond_resched(); |
d5c88b73 JS |
805 | } |
806 | return 0; | |
807 | } | |
808 | ||
8da6d581 JS |
809 | /* |
810 | * add all currently queued delayed refs from this head whose seq nr is | |
811 | * smaller or equal that seq to the list | |
812 | */ | |
00142756 JM |
813 | static int add_delayed_refs(const struct btrfs_fs_info *fs_info, |
814 | struct btrfs_delayed_ref_head *head, u64 seq, | |
86d5f994 | 815 | struct preftrees *preftrees, u64 *total_refs, |
3ec4d323 | 816 | struct share_check *sc) |
8da6d581 | 817 | { |
c6fc2454 | 818 | struct btrfs_delayed_ref_node *node; |
8da6d581 | 819 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; |
d5c88b73 | 820 | struct btrfs_key key; |
86d5f994 | 821 | struct btrfs_key tmp_op_key; |
0e0adbcf | 822 | struct rb_node *n; |
01747e92 | 823 | int count; |
b1375d64 | 824 | int ret = 0; |
8da6d581 | 825 | |
a6dbceaf | 826 | if (extent_op && extent_op->update_key) |
86d5f994 | 827 | btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key); |
8da6d581 | 828 | |
d7df2c79 | 829 | spin_lock(&head->lock); |
e3d03965 | 830 | for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) { |
0e0adbcf JB |
831 | node = rb_entry(n, struct btrfs_delayed_ref_node, |
832 | ref_node); | |
8da6d581 JS |
833 | if (node->seq > seq) |
834 | continue; | |
835 | ||
836 | switch (node->action) { | |
837 | case BTRFS_ADD_DELAYED_EXTENT: | |
838 | case BTRFS_UPDATE_DELAYED_HEAD: | |
839 | WARN_ON(1); | |
840 | continue; | |
841 | case BTRFS_ADD_DELAYED_REF: | |
01747e92 | 842 | count = node->ref_mod; |
8da6d581 JS |
843 | break; |
844 | case BTRFS_DROP_DELAYED_REF: | |
01747e92 | 845 | count = node->ref_mod * -1; |
8da6d581 JS |
846 | break; |
847 | default: | |
290342f6 | 848 | BUG(); |
8da6d581 | 849 | } |
01747e92 | 850 | *total_refs += count; |
8da6d581 JS |
851 | switch (node->type) { |
852 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
86d5f994 | 853 | /* NORMAL INDIRECT METADATA backref */ |
8da6d581 JS |
854 | struct btrfs_delayed_tree_ref *ref; |
855 | ||
856 | ref = btrfs_delayed_node_to_tree_ref(node); | |
00142756 JM |
857 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
858 | &tmp_op_key, ref->level + 1, | |
01747e92 EN |
859 | node->bytenr, count, sc, |
860 | GFP_ATOMIC); | |
8da6d581 JS |
861 | break; |
862 | } | |
863 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
86d5f994 | 864 | /* SHARED DIRECT METADATA backref */ |
8da6d581 JS |
865 | struct btrfs_delayed_tree_ref *ref; |
866 | ||
867 | ref = btrfs_delayed_node_to_tree_ref(node); | |
86d5f994 | 868 | |
01747e92 EN |
869 | ret = add_direct_ref(fs_info, preftrees, ref->level + 1, |
870 | ref->parent, node->bytenr, count, | |
3ec4d323 | 871 | sc, GFP_ATOMIC); |
8da6d581 JS |
872 | break; |
873 | } | |
874 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 875 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 | 876 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
877 | ref = btrfs_delayed_node_to_data_ref(node); |
878 | ||
879 | key.objectid = ref->objectid; | |
880 | key.type = BTRFS_EXTENT_DATA_KEY; | |
881 | key.offset = ref->offset; | |
dc046b10 JB |
882 | |
883 | /* | |
884 | * Found a inum that doesn't match our known inum, we | |
885 | * know it's shared. | |
886 | */ | |
3ec4d323 | 887 | if (sc && sc->inum && ref->objectid != sc->inum) { |
dc046b10 | 888 | ret = BACKREF_FOUND_SHARED; |
3ec4d323 | 889 | goto out; |
dc046b10 JB |
890 | } |
891 | ||
00142756 | 892 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
01747e92 EN |
893 | &key, 0, node->bytenr, count, sc, |
894 | GFP_ATOMIC); | |
8da6d581 JS |
895 | break; |
896 | } | |
897 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 898 | /* SHARED DIRECT FULL backref */ |
8da6d581 | 899 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
900 | |
901 | ref = btrfs_delayed_node_to_data_ref(node); | |
86d5f994 | 902 | |
01747e92 EN |
903 | ret = add_direct_ref(fs_info, preftrees, 0, ref->parent, |
904 | node->bytenr, count, sc, | |
905 | GFP_ATOMIC); | |
8da6d581 JS |
906 | break; |
907 | } | |
908 | default: | |
909 | WARN_ON(1); | |
910 | } | |
3ec4d323 EN |
911 | /* |
912 | * We must ignore BACKREF_FOUND_SHARED until all delayed | |
913 | * refs have been checked. | |
914 | */ | |
915 | if (ret && (ret != BACKREF_FOUND_SHARED)) | |
d7df2c79 | 916 | break; |
8da6d581 | 917 | } |
3ec4d323 EN |
918 | if (!ret) |
919 | ret = extent_is_shared(sc); | |
920 | out: | |
d7df2c79 JB |
921 | spin_unlock(&head->lock); |
922 | return ret; | |
8da6d581 JS |
923 | } |
924 | ||
925 | /* | |
926 | * add all inline backrefs for bytenr to the list | |
3ec4d323 EN |
927 | * |
928 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 929 | */ |
00142756 JM |
930 | static int add_inline_refs(const struct btrfs_fs_info *fs_info, |
931 | struct btrfs_path *path, u64 bytenr, | |
86d5f994 | 932 | int *info_level, struct preftrees *preftrees, |
3ec4d323 | 933 | u64 *total_refs, struct share_check *sc) |
8da6d581 | 934 | { |
b1375d64 | 935 | int ret = 0; |
8da6d581 JS |
936 | int slot; |
937 | struct extent_buffer *leaf; | |
938 | struct btrfs_key key; | |
261c84b6 | 939 | struct btrfs_key found_key; |
8da6d581 JS |
940 | unsigned long ptr; |
941 | unsigned long end; | |
942 | struct btrfs_extent_item *ei; | |
943 | u64 flags; | |
944 | u64 item_size; | |
945 | ||
946 | /* | |
947 | * enumerate all inline refs | |
948 | */ | |
949 | leaf = path->nodes[0]; | |
dadcaf78 | 950 | slot = path->slots[0]; |
8da6d581 JS |
951 | |
952 | item_size = btrfs_item_size_nr(leaf, slot); | |
953 | BUG_ON(item_size < sizeof(*ei)); | |
954 | ||
955 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
956 | flags = btrfs_extent_flags(leaf, ei); | |
44853868 | 957 | *total_refs += btrfs_extent_refs(leaf, ei); |
261c84b6 | 958 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
8da6d581 JS |
959 | |
960 | ptr = (unsigned long)(ei + 1); | |
961 | end = (unsigned long)ei + item_size; | |
962 | ||
261c84b6 JB |
963 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && |
964 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
8da6d581 | 965 | struct btrfs_tree_block_info *info; |
8da6d581 JS |
966 | |
967 | info = (struct btrfs_tree_block_info *)ptr; | |
968 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
969 | ptr += sizeof(struct btrfs_tree_block_info); |
970 | BUG_ON(ptr > end); | |
261c84b6 JB |
971 | } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { |
972 | *info_level = found_key.offset; | |
8da6d581 JS |
973 | } else { |
974 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
975 | } | |
976 | ||
977 | while (ptr < end) { | |
978 | struct btrfs_extent_inline_ref *iref; | |
979 | u64 offset; | |
980 | int type; | |
981 | ||
982 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
983 | type = btrfs_get_extent_inline_ref_type(leaf, iref, |
984 | BTRFS_REF_TYPE_ANY); | |
985 | if (type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 986 | return -EUCLEAN; |
3de28d57 | 987 | |
8da6d581 JS |
988 | offset = btrfs_extent_inline_ref_offset(leaf, iref); |
989 | ||
990 | switch (type) { | |
991 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
00142756 JM |
992 | ret = add_direct_ref(fs_info, preftrees, |
993 | *info_level + 1, offset, | |
3ec4d323 | 994 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
995 | break; |
996 | case BTRFS_SHARED_DATA_REF_KEY: { | |
997 | struct btrfs_shared_data_ref *sdref; | |
998 | int count; | |
999 | ||
1000 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1001 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
86d5f994 | 1002 | |
00142756 | 1003 | ret = add_direct_ref(fs_info, preftrees, 0, offset, |
3ec4d323 | 1004 | bytenr, count, sc, GFP_NOFS); |
8da6d581 JS |
1005 | break; |
1006 | } | |
1007 | case BTRFS_TREE_BLOCK_REF_KEY: | |
00142756 JM |
1008 | ret = add_indirect_ref(fs_info, preftrees, offset, |
1009 | NULL, *info_level + 1, | |
3ec4d323 | 1010 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1011 | break; |
1012 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
1013 | struct btrfs_extent_data_ref *dref; | |
1014 | int count; | |
1015 | u64 root; | |
1016 | ||
1017 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1018 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1019 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1020 | dref); | |
1021 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1022 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1023 | |
3ec4d323 | 1024 | if (sc && sc->inum && key.objectid != sc->inum) { |
dc046b10 JB |
1025 | ret = BACKREF_FOUND_SHARED; |
1026 | break; | |
1027 | } | |
1028 | ||
8da6d581 | 1029 | root = btrfs_extent_data_ref_root(leaf, dref); |
86d5f994 | 1030 | |
00142756 JM |
1031 | ret = add_indirect_ref(fs_info, preftrees, root, |
1032 | &key, 0, bytenr, count, | |
3ec4d323 | 1033 | sc, GFP_NOFS); |
8da6d581 JS |
1034 | break; |
1035 | } | |
1036 | default: | |
1037 | WARN_ON(1); | |
1038 | } | |
1149ab6b WS |
1039 | if (ret) |
1040 | return ret; | |
8da6d581 JS |
1041 | ptr += btrfs_extent_inline_ref_size(type); |
1042 | } | |
1043 | ||
1044 | return 0; | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * add all non-inline backrefs for bytenr to the list | |
3ec4d323 EN |
1049 | * |
1050 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 1051 | */ |
e0c476b1 JM |
1052 | static int add_keyed_refs(struct btrfs_fs_info *fs_info, |
1053 | struct btrfs_path *path, u64 bytenr, | |
86d5f994 | 1054 | int info_level, struct preftrees *preftrees, |
3ec4d323 | 1055 | struct share_check *sc) |
8da6d581 JS |
1056 | { |
1057 | struct btrfs_root *extent_root = fs_info->extent_root; | |
1058 | int ret; | |
1059 | int slot; | |
1060 | struct extent_buffer *leaf; | |
1061 | struct btrfs_key key; | |
1062 | ||
1063 | while (1) { | |
1064 | ret = btrfs_next_item(extent_root, path); | |
1065 | if (ret < 0) | |
1066 | break; | |
1067 | if (ret) { | |
1068 | ret = 0; | |
1069 | break; | |
1070 | } | |
1071 | ||
1072 | slot = path->slots[0]; | |
1073 | leaf = path->nodes[0]; | |
1074 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1075 | ||
1076 | if (key.objectid != bytenr) | |
1077 | break; | |
1078 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
1079 | continue; | |
1080 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
1081 | break; | |
1082 | ||
1083 | switch (key.type) { | |
1084 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
86d5f994 | 1085 | /* SHARED DIRECT METADATA backref */ |
00142756 JM |
1086 | ret = add_direct_ref(fs_info, preftrees, |
1087 | info_level + 1, key.offset, | |
3ec4d323 | 1088 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1089 | break; |
1090 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 1091 | /* SHARED DIRECT FULL backref */ |
8da6d581 JS |
1092 | struct btrfs_shared_data_ref *sdref; |
1093 | int count; | |
1094 | ||
1095 | sdref = btrfs_item_ptr(leaf, slot, | |
1096 | struct btrfs_shared_data_ref); | |
1097 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
00142756 JM |
1098 | ret = add_direct_ref(fs_info, preftrees, 0, |
1099 | key.offset, bytenr, count, | |
3ec4d323 | 1100 | sc, GFP_NOFS); |
8da6d581 JS |
1101 | break; |
1102 | } | |
1103 | case BTRFS_TREE_BLOCK_REF_KEY: | |
86d5f994 | 1104 | /* NORMAL INDIRECT METADATA backref */ |
00142756 JM |
1105 | ret = add_indirect_ref(fs_info, preftrees, key.offset, |
1106 | NULL, info_level + 1, bytenr, | |
3ec4d323 | 1107 | 1, NULL, GFP_NOFS); |
8da6d581 JS |
1108 | break; |
1109 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 1110 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 JS |
1111 | struct btrfs_extent_data_ref *dref; |
1112 | int count; | |
1113 | u64 root; | |
1114 | ||
1115 | dref = btrfs_item_ptr(leaf, slot, | |
1116 | struct btrfs_extent_data_ref); | |
1117 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1118 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1119 | dref); | |
1120 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1121 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1122 | |
3ec4d323 | 1123 | if (sc && sc->inum && key.objectid != sc->inum) { |
dc046b10 JB |
1124 | ret = BACKREF_FOUND_SHARED; |
1125 | break; | |
1126 | } | |
1127 | ||
8da6d581 | 1128 | root = btrfs_extent_data_ref_root(leaf, dref); |
00142756 JM |
1129 | ret = add_indirect_ref(fs_info, preftrees, root, |
1130 | &key, 0, bytenr, count, | |
3ec4d323 | 1131 | sc, GFP_NOFS); |
8da6d581 JS |
1132 | break; |
1133 | } | |
1134 | default: | |
1135 | WARN_ON(1); | |
1136 | } | |
1149ab6b WS |
1137 | if (ret) |
1138 | return ret; | |
1139 | ||
8da6d581 JS |
1140 | } |
1141 | ||
1142 | return ret; | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
1147 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
1148 | * indirect refs to their parent bytenr. | |
1149 | * When roots are found, they're added to the roots list | |
1150 | * | |
de47c9d3 | 1151 | * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave |
21633fc6 QW |
1152 | * much like trans == NULL case, the difference only lies in it will not |
1153 | * commit root. | |
1154 | * The special case is for qgroup to search roots in commit_transaction(). | |
1155 | * | |
3ec4d323 EN |
1156 | * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a |
1157 | * shared extent is detected. | |
1158 | * | |
1159 | * Otherwise this returns 0 for success and <0 for an error. | |
1160 | * | |
c995ab3c ZB |
1161 | * If ignore_offset is set to false, only extent refs whose offsets match |
1162 | * extent_item_pos are returned. If true, every extent ref is returned | |
1163 | * and extent_item_pos is ignored. | |
1164 | * | |
8da6d581 JS |
1165 | * FIXME some caching might speed things up |
1166 | */ | |
1167 | static int find_parent_nodes(struct btrfs_trans_handle *trans, | |
1168 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1169 | u64 time_seq, struct ulist *refs, |
dc046b10 | 1170 | struct ulist *roots, const u64 *extent_item_pos, |
c995ab3c | 1171 | struct share_check *sc, bool ignore_offset) |
8da6d581 JS |
1172 | { |
1173 | struct btrfs_key key; | |
1174 | struct btrfs_path *path; | |
8da6d581 | 1175 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 1176 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
1177 | int info_level = 0; |
1178 | int ret; | |
e0c476b1 | 1179 | struct prelim_ref *ref; |
86d5f994 | 1180 | struct rb_node *node; |
f05c4746 | 1181 | struct extent_inode_elem *eie = NULL; |
86d5f994 | 1182 | /* total of both direct AND indirect refs! */ |
44853868 | 1183 | u64 total_refs = 0; |
86d5f994 EN |
1184 | struct preftrees preftrees = { |
1185 | .direct = PREFTREE_INIT, | |
1186 | .indirect = PREFTREE_INIT, | |
1187 | .indirect_missing_keys = PREFTREE_INIT | |
1188 | }; | |
8da6d581 JS |
1189 | |
1190 | key.objectid = bytenr; | |
8da6d581 | 1191 | key.offset = (u64)-1; |
261c84b6 JB |
1192 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1193 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1194 | else | |
1195 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
8da6d581 JS |
1196 | |
1197 | path = btrfs_alloc_path(); | |
1198 | if (!path) | |
1199 | return -ENOMEM; | |
e84752d4 | 1200 | if (!trans) { |
da61d31a | 1201 | path->search_commit_root = 1; |
e84752d4 WS |
1202 | path->skip_locking = 1; |
1203 | } | |
8da6d581 | 1204 | |
de47c9d3 | 1205 | if (time_seq == SEQ_LAST) |
21633fc6 QW |
1206 | path->skip_locking = 1; |
1207 | ||
8da6d581 JS |
1208 | /* |
1209 | * grab both a lock on the path and a lock on the delayed ref head. | |
1210 | * We need both to get a consistent picture of how the refs look | |
1211 | * at a specified point in time | |
1212 | */ | |
1213 | again: | |
d3b01064 LZ |
1214 | head = NULL; |
1215 | ||
8da6d581 JS |
1216 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
1217 | if (ret < 0) | |
1218 | goto out; | |
1219 | BUG_ON(ret == 0); | |
1220 | ||
faa2dbf0 | 1221 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
21633fc6 | 1222 | if (trans && likely(trans->type != __TRANS_DUMMY) && |
de47c9d3 | 1223 | time_seq != SEQ_LAST) { |
faa2dbf0 | 1224 | #else |
de47c9d3 | 1225 | if (trans && time_seq != SEQ_LAST) { |
faa2dbf0 | 1226 | #endif |
7a3ae2f8 JS |
1227 | /* |
1228 | * look if there are updates for this ref queued and lock the | |
1229 | * head | |
1230 | */ | |
1231 | delayed_refs = &trans->transaction->delayed_refs; | |
1232 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 1233 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
7a3ae2f8 JS |
1234 | if (head) { |
1235 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 1236 | refcount_inc(&head->refs); |
7a3ae2f8 JS |
1237 | spin_unlock(&delayed_refs->lock); |
1238 | ||
1239 | btrfs_release_path(path); | |
1240 | ||
1241 | /* | |
1242 | * Mutex was contended, block until it's | |
1243 | * released and try again | |
1244 | */ | |
1245 | mutex_lock(&head->mutex); | |
1246 | mutex_unlock(&head->mutex); | |
d278850e | 1247 | btrfs_put_delayed_ref_head(head); |
7a3ae2f8 JS |
1248 | goto again; |
1249 | } | |
d7df2c79 | 1250 | spin_unlock(&delayed_refs->lock); |
00142756 | 1251 | ret = add_delayed_refs(fs_info, head, time_seq, |
3ec4d323 | 1252 | &preftrees, &total_refs, sc); |
155725c9 | 1253 | mutex_unlock(&head->mutex); |
d7df2c79 | 1254 | if (ret) |
7a3ae2f8 | 1255 | goto out; |
d7df2c79 JB |
1256 | } else { |
1257 | spin_unlock(&delayed_refs->lock); | |
d3b01064 | 1258 | } |
8da6d581 | 1259 | } |
8da6d581 JS |
1260 | |
1261 | if (path->slots[0]) { | |
1262 | struct extent_buffer *leaf; | |
1263 | int slot; | |
1264 | ||
dadcaf78 | 1265 | path->slots[0]--; |
8da6d581 | 1266 | leaf = path->nodes[0]; |
dadcaf78 | 1267 | slot = path->slots[0]; |
8da6d581 JS |
1268 | btrfs_item_key_to_cpu(leaf, &key, slot); |
1269 | if (key.objectid == bytenr && | |
261c84b6 JB |
1270 | (key.type == BTRFS_EXTENT_ITEM_KEY || |
1271 | key.type == BTRFS_METADATA_ITEM_KEY)) { | |
00142756 JM |
1272 | ret = add_inline_refs(fs_info, path, bytenr, |
1273 | &info_level, &preftrees, | |
3ec4d323 | 1274 | &total_refs, sc); |
8da6d581 JS |
1275 | if (ret) |
1276 | goto out; | |
e0c476b1 | 1277 | ret = add_keyed_refs(fs_info, path, bytenr, info_level, |
3ec4d323 | 1278 | &preftrees, sc); |
8da6d581 JS |
1279 | if (ret) |
1280 | goto out; | |
1281 | } | |
1282 | } | |
8da6d581 | 1283 | |
86d5f994 | 1284 | btrfs_release_path(path); |
8da6d581 | 1285 | |
38e3eebf | 1286 | ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0); |
d5c88b73 JS |
1287 | if (ret) |
1288 | goto out; | |
1289 | ||
ecf160b4 | 1290 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root)); |
8da6d581 | 1291 | |
86d5f994 | 1292 | ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees, |
c995ab3c | 1293 | extent_item_pos, total_refs, sc, ignore_offset); |
8da6d581 JS |
1294 | if (ret) |
1295 | goto out; | |
1296 | ||
ecf160b4 | 1297 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root)); |
8da6d581 | 1298 | |
86d5f994 EN |
1299 | /* |
1300 | * This walks the tree of merged and resolved refs. Tree blocks are | |
1301 | * read in as needed. Unique entries are added to the ulist, and | |
1302 | * the list of found roots is updated. | |
1303 | * | |
1304 | * We release the entire tree in one go before returning. | |
1305 | */ | |
ecf160b4 | 1306 | node = rb_first_cached(&preftrees.direct.root); |
86d5f994 EN |
1307 | while (node) { |
1308 | ref = rb_entry(node, struct prelim_ref, rbnode); | |
1309 | node = rb_next(&ref->rbnode); | |
c8195a7b ZB |
1310 | /* |
1311 | * ref->count < 0 can happen here if there are delayed | |
1312 | * refs with a node->action of BTRFS_DROP_DELAYED_REF. | |
1313 | * prelim_ref_insert() relies on this when merging | |
1314 | * identical refs to keep the overall count correct. | |
1315 | * prelim_ref_insert() will merge only those refs | |
1316 | * which compare identically. Any refs having | |
1317 | * e.g. different offsets would not be merged, | |
1318 | * and would retain their original ref->count < 0. | |
1319 | */ | |
98cfee21 | 1320 | if (roots && ref->count && ref->root_id && ref->parent == 0) { |
3ec4d323 EN |
1321 | if (sc && sc->root_objectid && |
1322 | ref->root_id != sc->root_objectid) { | |
dc046b10 JB |
1323 | ret = BACKREF_FOUND_SHARED; |
1324 | goto out; | |
1325 | } | |
1326 | ||
8da6d581 JS |
1327 | /* no parent == root of tree */ |
1328 | ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); | |
f1723939 WS |
1329 | if (ret < 0) |
1330 | goto out; | |
8da6d581 JS |
1331 | } |
1332 | if (ref->count && ref->parent) { | |
8a56457f JB |
1333 | if (extent_item_pos && !ref->inode_list && |
1334 | ref->level == 0) { | |
976b1908 | 1335 | struct extent_buffer *eb; |
707e8a07 | 1336 | |
581c1760 QW |
1337 | eb = read_tree_block(fs_info, ref->parent, 0, |
1338 | ref->level, NULL); | |
64c043de LB |
1339 | if (IS_ERR(eb)) { |
1340 | ret = PTR_ERR(eb); | |
1341 | goto out; | |
1342 | } else if (!extent_buffer_uptodate(eb)) { | |
416bc658 | 1343 | free_extent_buffer(eb); |
c16c2e2e WS |
1344 | ret = -EIO; |
1345 | goto out; | |
416bc658 | 1346 | } |
38e3eebf JB |
1347 | |
1348 | if (!path->skip_locking) { | |
1349 | btrfs_tree_read_lock(eb); | |
1350 | btrfs_set_lock_blocking_read(eb); | |
1351 | } | |
976b1908 | 1352 | ret = find_extent_in_eb(eb, bytenr, |
c995ab3c | 1353 | *extent_item_pos, &eie, ignore_offset); |
38e3eebf JB |
1354 | if (!path->skip_locking) |
1355 | btrfs_tree_read_unlock_blocking(eb); | |
976b1908 | 1356 | free_extent_buffer(eb); |
f5929cd8 FDBM |
1357 | if (ret < 0) |
1358 | goto out; | |
1359 | ref->inode_list = eie; | |
976b1908 | 1360 | } |
4eb1f66d TI |
1361 | ret = ulist_add_merge_ptr(refs, ref->parent, |
1362 | ref->inode_list, | |
1363 | (void **)&eie, GFP_NOFS); | |
f1723939 WS |
1364 | if (ret < 0) |
1365 | goto out; | |
3301958b JS |
1366 | if (!ret && extent_item_pos) { |
1367 | /* | |
1368 | * we've recorded that parent, so we must extend | |
1369 | * its inode list here | |
1370 | */ | |
1371 | BUG_ON(!eie); | |
1372 | while (eie->next) | |
1373 | eie = eie->next; | |
1374 | eie->next = ref->inode_list; | |
1375 | } | |
f05c4746 | 1376 | eie = NULL; |
8da6d581 | 1377 | } |
9dd14fd6 | 1378 | cond_resched(); |
8da6d581 JS |
1379 | } |
1380 | ||
1381 | out: | |
8da6d581 | 1382 | btrfs_free_path(path); |
86d5f994 EN |
1383 | |
1384 | prelim_release(&preftrees.direct); | |
1385 | prelim_release(&preftrees.indirect); | |
1386 | prelim_release(&preftrees.indirect_missing_keys); | |
1387 | ||
f05c4746 WS |
1388 | if (ret < 0) |
1389 | free_inode_elem_list(eie); | |
8da6d581 JS |
1390 | return ret; |
1391 | } | |
1392 | ||
976b1908 JS |
1393 | static void free_leaf_list(struct ulist *blocks) |
1394 | { | |
1395 | struct ulist_node *node = NULL; | |
1396 | struct extent_inode_elem *eie; | |
976b1908 JS |
1397 | struct ulist_iterator uiter; |
1398 | ||
1399 | ULIST_ITER_INIT(&uiter); | |
1400 | while ((node = ulist_next(blocks, &uiter))) { | |
1401 | if (!node->aux) | |
1402 | continue; | |
4dae077a | 1403 | eie = unode_aux_to_inode_list(node); |
f05c4746 | 1404 | free_inode_elem_list(eie); |
976b1908 JS |
1405 | node->aux = 0; |
1406 | } | |
1407 | ||
1408 | ulist_free(blocks); | |
1409 | } | |
1410 | ||
8da6d581 JS |
1411 | /* |
1412 | * Finds all leafs with a reference to the specified combination of bytenr and | |
1413 | * offset. key_list_head will point to a list of corresponding keys (caller must | |
1414 | * free each list element). The leafs will be stored in the leafs ulist, which | |
1415 | * must be freed with ulist_free. | |
1416 | * | |
1417 | * returns 0 on success, <0 on error | |
1418 | */ | |
1419 | static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, | |
1420 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1421 | u64 time_seq, struct ulist **leafs, |
c995ab3c | 1422 | const u64 *extent_item_pos, bool ignore_offset) |
8da6d581 | 1423 | { |
8da6d581 JS |
1424 | int ret; |
1425 | ||
8da6d581 | 1426 | *leafs = ulist_alloc(GFP_NOFS); |
98cfee21 | 1427 | if (!*leafs) |
8da6d581 | 1428 | return -ENOMEM; |
8da6d581 | 1429 | |
afce772e | 1430 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1431 | *leafs, NULL, extent_item_pos, NULL, ignore_offset); |
8da6d581 | 1432 | if (ret < 0 && ret != -ENOENT) { |
976b1908 | 1433 | free_leaf_list(*leafs); |
8da6d581 JS |
1434 | return ret; |
1435 | } | |
1436 | ||
1437 | return 0; | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * walk all backrefs for a given extent to find all roots that reference this | |
1442 | * extent. Walking a backref means finding all extents that reference this | |
1443 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1444 | * recursive process, but here it is implemented in an iterative fashion: We | |
1445 | * find all referencing extents for the extent in question and put them on a | |
1446 | * list. In turn, we find all referencing extents for those, further appending | |
1447 | * to the list. The way we iterate the list allows adding more elements after | |
1448 | * the current while iterating. The process stops when we reach the end of the | |
1449 | * list. Found roots are added to the roots list. | |
1450 | * | |
1451 | * returns 0 on success, < 0 on error. | |
1452 | */ | |
e0c476b1 JM |
1453 | static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans, |
1454 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c ZB |
1455 | u64 time_seq, struct ulist **roots, |
1456 | bool ignore_offset) | |
8da6d581 JS |
1457 | { |
1458 | struct ulist *tmp; | |
1459 | struct ulist_node *node = NULL; | |
cd1b413c | 1460 | struct ulist_iterator uiter; |
8da6d581 JS |
1461 | int ret; |
1462 | ||
1463 | tmp = ulist_alloc(GFP_NOFS); | |
1464 | if (!tmp) | |
1465 | return -ENOMEM; | |
1466 | *roots = ulist_alloc(GFP_NOFS); | |
1467 | if (!*roots) { | |
1468 | ulist_free(tmp); | |
1469 | return -ENOMEM; | |
1470 | } | |
1471 | ||
cd1b413c | 1472 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1473 | while (1) { |
afce772e | 1474 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1475 | tmp, *roots, NULL, NULL, ignore_offset); |
8da6d581 JS |
1476 | if (ret < 0 && ret != -ENOENT) { |
1477 | ulist_free(tmp); | |
1478 | ulist_free(*roots); | |
1479 | return ret; | |
1480 | } | |
cd1b413c | 1481 | node = ulist_next(tmp, &uiter); |
8da6d581 JS |
1482 | if (!node) |
1483 | break; | |
1484 | bytenr = node->val; | |
bca1a290 | 1485 | cond_resched(); |
8da6d581 JS |
1486 | } |
1487 | ||
1488 | ulist_free(tmp); | |
1489 | return 0; | |
1490 | } | |
1491 | ||
9e351cc8 JB |
1492 | int btrfs_find_all_roots(struct btrfs_trans_handle *trans, |
1493 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c ZB |
1494 | u64 time_seq, struct ulist **roots, |
1495 | bool ignore_offset) | |
9e351cc8 JB |
1496 | { |
1497 | int ret; | |
1498 | ||
1499 | if (!trans) | |
1500 | down_read(&fs_info->commit_root_sem); | |
e0c476b1 | 1501 | ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr, |
c995ab3c | 1502 | time_seq, roots, ignore_offset); |
9e351cc8 JB |
1503 | if (!trans) |
1504 | up_read(&fs_info->commit_root_sem); | |
1505 | return ret; | |
1506 | } | |
1507 | ||
2c2ed5aa MF |
1508 | /** |
1509 | * btrfs_check_shared - tell us whether an extent is shared | |
1510 | * | |
2c2ed5aa MF |
1511 | * btrfs_check_shared uses the backref walking code but will short |
1512 | * circuit as soon as it finds a root or inode that doesn't match the | |
1513 | * one passed in. This provides a significant performance benefit for | |
1514 | * callers (such as fiemap) which want to know whether the extent is | |
1515 | * shared but do not need a ref count. | |
1516 | * | |
03628cdb FM |
1517 | * This attempts to attach to the running transaction in order to account for |
1518 | * delayed refs, but continues on even when no running transaction exists. | |
bb739cf0 | 1519 | * |
2c2ed5aa MF |
1520 | * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. |
1521 | */ | |
5911c8fe DS |
1522 | int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr, |
1523 | struct ulist *roots, struct ulist *tmp) | |
dc046b10 | 1524 | { |
bb739cf0 EN |
1525 | struct btrfs_fs_info *fs_info = root->fs_info; |
1526 | struct btrfs_trans_handle *trans; | |
dc046b10 JB |
1527 | struct ulist_iterator uiter; |
1528 | struct ulist_node *node; | |
3284da7b | 1529 | struct seq_list elem = SEQ_LIST_INIT(elem); |
dc046b10 | 1530 | int ret = 0; |
3ec4d323 | 1531 | struct share_check shared = { |
4fd786e6 | 1532 | .root_objectid = root->root_key.objectid, |
3ec4d323 EN |
1533 | .inum = inum, |
1534 | .share_count = 0, | |
1535 | }; | |
dc046b10 | 1536 | |
5911c8fe DS |
1537 | ulist_init(roots); |
1538 | ulist_init(tmp); | |
dc046b10 | 1539 | |
a6d155d2 | 1540 | trans = btrfs_join_transaction_nostart(root); |
bb739cf0 | 1541 | if (IS_ERR(trans)) { |
03628cdb FM |
1542 | if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) { |
1543 | ret = PTR_ERR(trans); | |
1544 | goto out; | |
1545 | } | |
bb739cf0 | 1546 | trans = NULL; |
dc046b10 | 1547 | down_read(&fs_info->commit_root_sem); |
bb739cf0 EN |
1548 | } else { |
1549 | btrfs_get_tree_mod_seq(fs_info, &elem); | |
1550 | } | |
1551 | ||
dc046b10 JB |
1552 | ULIST_ITER_INIT(&uiter); |
1553 | while (1) { | |
1554 | ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp, | |
c995ab3c | 1555 | roots, NULL, &shared, false); |
dc046b10 | 1556 | if (ret == BACKREF_FOUND_SHARED) { |
2c2ed5aa | 1557 | /* this is the only condition under which we return 1 */ |
dc046b10 JB |
1558 | ret = 1; |
1559 | break; | |
1560 | } | |
1561 | if (ret < 0 && ret != -ENOENT) | |
1562 | break; | |
2c2ed5aa | 1563 | ret = 0; |
dc046b10 JB |
1564 | node = ulist_next(tmp, &uiter); |
1565 | if (!node) | |
1566 | break; | |
1567 | bytenr = node->val; | |
18bf591b | 1568 | shared.share_count = 0; |
dc046b10 JB |
1569 | cond_resched(); |
1570 | } | |
bb739cf0 EN |
1571 | |
1572 | if (trans) { | |
dc046b10 | 1573 | btrfs_put_tree_mod_seq(fs_info, &elem); |
bb739cf0 EN |
1574 | btrfs_end_transaction(trans); |
1575 | } else { | |
dc046b10 | 1576 | up_read(&fs_info->commit_root_sem); |
bb739cf0 | 1577 | } |
03628cdb | 1578 | out: |
5911c8fe DS |
1579 | ulist_release(roots); |
1580 | ulist_release(tmp); | |
dc046b10 JB |
1581 | return ret; |
1582 | } | |
1583 | ||
f186373f MF |
1584 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
1585 | u64 start_off, struct btrfs_path *path, | |
1586 | struct btrfs_inode_extref **ret_extref, | |
1587 | u64 *found_off) | |
1588 | { | |
1589 | int ret, slot; | |
1590 | struct btrfs_key key; | |
1591 | struct btrfs_key found_key; | |
1592 | struct btrfs_inode_extref *extref; | |
73980bec | 1593 | const struct extent_buffer *leaf; |
f186373f MF |
1594 | unsigned long ptr; |
1595 | ||
1596 | key.objectid = inode_objectid; | |
962a298f | 1597 | key.type = BTRFS_INODE_EXTREF_KEY; |
f186373f MF |
1598 | key.offset = start_off; |
1599 | ||
1600 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1601 | if (ret < 0) | |
1602 | return ret; | |
1603 | ||
1604 | while (1) { | |
1605 | leaf = path->nodes[0]; | |
1606 | slot = path->slots[0]; | |
1607 | if (slot >= btrfs_header_nritems(leaf)) { | |
1608 | /* | |
1609 | * If the item at offset is not found, | |
1610 | * btrfs_search_slot will point us to the slot | |
1611 | * where it should be inserted. In our case | |
1612 | * that will be the slot directly before the | |
1613 | * next INODE_REF_KEY_V2 item. In the case | |
1614 | * that we're pointing to the last slot in a | |
1615 | * leaf, we must move one leaf over. | |
1616 | */ | |
1617 | ret = btrfs_next_leaf(root, path); | |
1618 | if (ret) { | |
1619 | if (ret >= 1) | |
1620 | ret = -ENOENT; | |
1621 | break; | |
1622 | } | |
1623 | continue; | |
1624 | } | |
1625 | ||
1626 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1627 | ||
1628 | /* | |
1629 | * Check that we're still looking at an extended ref key for | |
1630 | * this particular objectid. If we have different | |
1631 | * objectid or type then there are no more to be found | |
1632 | * in the tree and we can exit. | |
1633 | */ | |
1634 | ret = -ENOENT; | |
1635 | if (found_key.objectid != inode_objectid) | |
1636 | break; | |
962a298f | 1637 | if (found_key.type != BTRFS_INODE_EXTREF_KEY) |
f186373f MF |
1638 | break; |
1639 | ||
1640 | ret = 0; | |
1641 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1642 | extref = (struct btrfs_inode_extref *)ptr; | |
1643 | *ret_extref = extref; | |
1644 | if (found_off) | |
1645 | *found_off = found_key.offset; | |
1646 | break; | |
1647 | } | |
1648 | ||
1649 | return ret; | |
1650 | } | |
1651 | ||
48a3b636 ES |
1652 | /* |
1653 | * this iterates to turn a name (from iref/extref) into a full filesystem path. | |
1654 | * Elements of the path are separated by '/' and the path is guaranteed to be | |
1655 | * 0-terminated. the path is only given within the current file system. | |
1656 | * Therefore, it never starts with a '/'. the caller is responsible to provide | |
1657 | * "size" bytes in "dest". the dest buffer will be filled backwards. finally, | |
1658 | * the start point of the resulting string is returned. this pointer is within | |
1659 | * dest, normally. | |
1660 | * in case the path buffer would overflow, the pointer is decremented further | |
1661 | * as if output was written to the buffer, though no more output is actually | |
1662 | * generated. that way, the caller can determine how much space would be | |
1663 | * required for the path to fit into the buffer. in that case, the returned | |
1664 | * value will be smaller than dest. callers must check this! | |
1665 | */ | |
96b5bd77 JS |
1666 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
1667 | u32 name_len, unsigned long name_off, | |
1668 | struct extent_buffer *eb_in, u64 parent, | |
1669 | char *dest, u32 size) | |
a542ad1b | 1670 | { |
a542ad1b JS |
1671 | int slot; |
1672 | u64 next_inum; | |
1673 | int ret; | |
661bec6b | 1674 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
1675 | struct extent_buffer *eb = eb_in; |
1676 | struct btrfs_key found_key; | |
b916a59a | 1677 | int leave_spinning = path->leave_spinning; |
d24bec3a | 1678 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
1679 | |
1680 | if (bytes_left >= 0) | |
1681 | dest[bytes_left] = '\0'; | |
1682 | ||
b916a59a | 1683 | path->leave_spinning = 1; |
a542ad1b | 1684 | while (1) { |
d24bec3a | 1685 | bytes_left -= name_len; |
a542ad1b JS |
1686 | if (bytes_left >= 0) |
1687 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 1688 | name_off, name_len); |
b916a59a | 1689 | if (eb != eb_in) { |
0c0fe3b0 FM |
1690 | if (!path->skip_locking) |
1691 | btrfs_tree_read_unlock_blocking(eb); | |
a542ad1b | 1692 | free_extent_buffer(eb); |
b916a59a | 1693 | } |
c234a24d DS |
1694 | ret = btrfs_find_item(fs_root, path, parent, 0, |
1695 | BTRFS_INODE_REF_KEY, &found_key); | |
8f24b496 JS |
1696 | if (ret > 0) |
1697 | ret = -ENOENT; | |
a542ad1b JS |
1698 | if (ret) |
1699 | break; | |
d24bec3a | 1700 | |
a542ad1b JS |
1701 | next_inum = found_key.offset; |
1702 | ||
1703 | /* regular exit ahead */ | |
1704 | if (parent == next_inum) | |
1705 | break; | |
1706 | ||
1707 | slot = path->slots[0]; | |
1708 | eb = path->nodes[0]; | |
1709 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 1710 | if (eb != eb_in) { |
0c0fe3b0 | 1711 | if (!path->skip_locking) |
300aa896 | 1712 | btrfs_set_lock_blocking_read(eb); |
0c0fe3b0 FM |
1713 | path->nodes[0] = NULL; |
1714 | path->locks[0] = 0; | |
b916a59a | 1715 | } |
a542ad1b | 1716 | btrfs_release_path(path); |
a542ad1b | 1717 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
1718 | |
1719 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1720 | name_off = (unsigned long)(iref + 1); | |
1721 | ||
a542ad1b JS |
1722 | parent = next_inum; |
1723 | --bytes_left; | |
1724 | if (bytes_left >= 0) | |
1725 | dest[bytes_left] = '/'; | |
1726 | } | |
1727 | ||
1728 | btrfs_release_path(path); | |
b916a59a | 1729 | path->leave_spinning = leave_spinning; |
a542ad1b JS |
1730 | |
1731 | if (ret) | |
1732 | return ERR_PTR(ret); | |
1733 | ||
1734 | return dest + bytes_left; | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * this makes the path point to (logical EXTENT_ITEM *) | |
1739 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
1740 | * tree blocks and <0 on error. | |
1741 | */ | |
1742 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
1743 | struct btrfs_path *path, struct btrfs_key *found_key, |
1744 | u64 *flags_ret) | |
a542ad1b JS |
1745 | { |
1746 | int ret; | |
1747 | u64 flags; | |
261c84b6 | 1748 | u64 size = 0; |
a542ad1b | 1749 | u32 item_size; |
73980bec | 1750 | const struct extent_buffer *eb; |
a542ad1b JS |
1751 | struct btrfs_extent_item *ei; |
1752 | struct btrfs_key key; | |
1753 | ||
261c84b6 JB |
1754 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1755 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1756 | else | |
1757 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a542ad1b JS |
1758 | key.objectid = logical; |
1759 | key.offset = (u64)-1; | |
1760 | ||
1761 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); | |
1762 | if (ret < 0) | |
1763 | return ret; | |
a542ad1b | 1764 | |
850a8cdf WS |
1765 | ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0); |
1766 | if (ret) { | |
1767 | if (ret > 0) | |
1768 | ret = -ENOENT; | |
1769 | return ret; | |
580f0a67 | 1770 | } |
850a8cdf | 1771 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); |
261c84b6 | 1772 | if (found_key->type == BTRFS_METADATA_ITEM_KEY) |
da17066c | 1773 | size = fs_info->nodesize; |
261c84b6 JB |
1774 | else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) |
1775 | size = found_key->offset; | |
1776 | ||
580f0a67 | 1777 | if (found_key->objectid > logical || |
261c84b6 | 1778 | found_key->objectid + size <= logical) { |
ab8d0fc4 JM |
1779 | btrfs_debug(fs_info, |
1780 | "logical %llu is not within any extent", logical); | |
a542ad1b | 1781 | return -ENOENT; |
4692cf58 | 1782 | } |
a542ad1b JS |
1783 | |
1784 | eb = path->nodes[0]; | |
1785 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
1786 | BUG_ON(item_size < sizeof(*ei)); | |
1787 | ||
1788 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
1789 | flags = btrfs_extent_flags(eb, ei); | |
1790 | ||
ab8d0fc4 JM |
1791 | btrfs_debug(fs_info, |
1792 | "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u", | |
c1c9ff7c GU |
1793 | logical, logical - found_key->objectid, found_key->objectid, |
1794 | found_key->offset, flags, item_size); | |
69917e43 LB |
1795 | |
1796 | WARN_ON(!flags_ret); | |
1797 | if (flags_ret) { | |
1798 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1799 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
1800 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1801 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
1802 | else | |
290342f6 | 1803 | BUG(); |
69917e43 LB |
1804 | return 0; |
1805 | } | |
a542ad1b JS |
1806 | |
1807 | return -EIO; | |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
1812 | * for the first call and may be modified. it is used to track state. | |
1813 | * if more refs exist, 0 is returned and the next call to | |
e0c476b1 | 1814 | * get_extent_inline_ref must pass the modified ptr parameter to get the |
a542ad1b JS |
1815 | * next ref. after the last ref was processed, 1 is returned. |
1816 | * returns <0 on error | |
1817 | */ | |
e0c476b1 JM |
1818 | static int get_extent_inline_ref(unsigned long *ptr, |
1819 | const struct extent_buffer *eb, | |
1820 | const struct btrfs_key *key, | |
1821 | const struct btrfs_extent_item *ei, | |
1822 | u32 item_size, | |
1823 | struct btrfs_extent_inline_ref **out_eiref, | |
1824 | int *out_type) | |
a542ad1b JS |
1825 | { |
1826 | unsigned long end; | |
1827 | u64 flags; | |
1828 | struct btrfs_tree_block_info *info; | |
1829 | ||
1830 | if (!*ptr) { | |
1831 | /* first call */ | |
1832 | flags = btrfs_extent_flags(eb, ei); | |
1833 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
6eda71d0 LB |
1834 | if (key->type == BTRFS_METADATA_ITEM_KEY) { |
1835 | /* a skinny metadata extent */ | |
1836 | *out_eiref = | |
1837 | (struct btrfs_extent_inline_ref *)(ei + 1); | |
1838 | } else { | |
1839 | WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY); | |
1840 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1841 | *out_eiref = | |
1842 | (struct btrfs_extent_inline_ref *)(info + 1); | |
1843 | } | |
a542ad1b JS |
1844 | } else { |
1845 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
1846 | } | |
1847 | *ptr = (unsigned long)*out_eiref; | |
cd857dd6 | 1848 | if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) |
a542ad1b JS |
1849 | return -ENOENT; |
1850 | } | |
1851 | ||
1852 | end = (unsigned long)ei + item_size; | |
6eda71d0 | 1853 | *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); |
3de28d57 LB |
1854 | *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref, |
1855 | BTRFS_REF_TYPE_ANY); | |
1856 | if (*out_type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 1857 | return -EUCLEAN; |
a542ad1b JS |
1858 | |
1859 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
1860 | WARN_ON(*ptr > end); | |
1861 | if (*ptr == end) | |
1862 | return 1; /* last */ | |
1863 | ||
1864 | return 0; | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * reads the tree block backref for an extent. tree level and root are returned | |
1869 | * through out_level and out_root. ptr must point to a 0 value for the first | |
e0c476b1 | 1870 | * call and may be modified (see get_extent_inline_ref comment). |
a542ad1b JS |
1871 | * returns 0 if data was provided, 1 if there was no more data to provide or |
1872 | * <0 on error. | |
1873 | */ | |
1874 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
1875 | struct btrfs_key *key, struct btrfs_extent_item *ei, |
1876 | u32 item_size, u64 *out_root, u8 *out_level) | |
a542ad1b JS |
1877 | { |
1878 | int ret; | |
1879 | int type; | |
a542ad1b JS |
1880 | struct btrfs_extent_inline_ref *eiref; |
1881 | ||
1882 | if (*ptr == (unsigned long)-1) | |
1883 | return 1; | |
1884 | ||
1885 | while (1) { | |
e0c476b1 | 1886 | ret = get_extent_inline_ref(ptr, eb, key, ei, item_size, |
6eda71d0 | 1887 | &eiref, &type); |
a542ad1b JS |
1888 | if (ret < 0) |
1889 | return ret; | |
1890 | ||
1891 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1892 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1893 | break; | |
1894 | ||
1895 | if (ret == 1) | |
1896 | return 1; | |
1897 | } | |
1898 | ||
1899 | /* we can treat both ref types equally here */ | |
a542ad1b | 1900 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); |
a1317f45 FM |
1901 | |
1902 | if (key->type == BTRFS_EXTENT_ITEM_KEY) { | |
1903 | struct btrfs_tree_block_info *info; | |
1904 | ||
1905 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
1906 | *out_level = btrfs_tree_block_level(eb, info); | |
1907 | } else { | |
1908 | ASSERT(key->type == BTRFS_METADATA_ITEM_KEY); | |
1909 | *out_level = (u8)key->offset; | |
1910 | } | |
a542ad1b JS |
1911 | |
1912 | if (ret == 1) | |
1913 | *ptr = (unsigned long)-1; | |
1914 | ||
1915 | return 0; | |
1916 | } | |
1917 | ||
ab8d0fc4 JM |
1918 | static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, |
1919 | struct extent_inode_elem *inode_list, | |
1920 | u64 root, u64 extent_item_objectid, | |
1921 | iterate_extent_inodes_t *iterate, void *ctx) | |
a542ad1b | 1922 | { |
976b1908 | 1923 | struct extent_inode_elem *eie; |
4692cf58 | 1924 | int ret = 0; |
4692cf58 | 1925 | |
976b1908 | 1926 | for (eie = inode_list; eie; eie = eie->next) { |
ab8d0fc4 JM |
1927 | btrfs_debug(fs_info, |
1928 | "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu", | |
1929 | extent_item_objectid, eie->inum, | |
1930 | eie->offset, root); | |
976b1908 | 1931 | ret = iterate(eie->inum, eie->offset, root, ctx); |
4692cf58 | 1932 | if (ret) { |
ab8d0fc4 JM |
1933 | btrfs_debug(fs_info, |
1934 | "stopping iteration for %llu due to ret=%d", | |
1935 | extent_item_objectid, ret); | |
4692cf58 JS |
1936 | break; |
1937 | } | |
a542ad1b JS |
1938 | } |
1939 | ||
a542ad1b JS |
1940 | return ret; |
1941 | } | |
1942 | ||
1943 | /* | |
1944 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 1945 | * the given parameters. |
a542ad1b JS |
1946 | * when the iterator function returns a non-zero value, iteration stops. |
1947 | */ | |
1948 | int iterate_extent_inodes(struct btrfs_fs_info *fs_info, | |
4692cf58 | 1949 | u64 extent_item_objectid, u64 extent_item_pos, |
7a3ae2f8 | 1950 | int search_commit_root, |
c995ab3c ZB |
1951 | iterate_extent_inodes_t *iterate, void *ctx, |
1952 | bool ignore_offset) | |
a542ad1b | 1953 | { |
a542ad1b | 1954 | int ret; |
da61d31a | 1955 | struct btrfs_trans_handle *trans = NULL; |
7a3ae2f8 JS |
1956 | struct ulist *refs = NULL; |
1957 | struct ulist *roots = NULL; | |
4692cf58 JS |
1958 | struct ulist_node *ref_node = NULL; |
1959 | struct ulist_node *root_node = NULL; | |
3284da7b | 1960 | struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem); |
cd1b413c JS |
1961 | struct ulist_iterator ref_uiter; |
1962 | struct ulist_iterator root_uiter; | |
a542ad1b | 1963 | |
ab8d0fc4 | 1964 | btrfs_debug(fs_info, "resolving all inodes for extent %llu", |
4692cf58 | 1965 | extent_item_objectid); |
a542ad1b | 1966 | |
da61d31a | 1967 | if (!search_commit_root) { |
bfc61c36 FM |
1968 | trans = btrfs_attach_transaction(fs_info->extent_root); |
1969 | if (IS_ERR(trans)) { | |
1970 | if (PTR_ERR(trans) != -ENOENT && | |
1971 | PTR_ERR(trans) != -EROFS) | |
1972 | return PTR_ERR(trans); | |
1973 | trans = NULL; | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | if (trans) | |
8445f61c | 1978 | btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
bfc61c36 | 1979 | else |
9e351cc8 | 1980 | down_read(&fs_info->commit_root_sem); |
a542ad1b | 1981 | |
4692cf58 | 1982 | ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, |
097b8a7c | 1983 | tree_mod_seq_elem.seq, &refs, |
c995ab3c | 1984 | &extent_item_pos, ignore_offset); |
4692cf58 JS |
1985 | if (ret) |
1986 | goto out; | |
a542ad1b | 1987 | |
cd1b413c JS |
1988 | ULIST_ITER_INIT(&ref_uiter); |
1989 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
e0c476b1 | 1990 | ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val, |
c995ab3c ZB |
1991 | tree_mod_seq_elem.seq, &roots, |
1992 | ignore_offset); | |
4692cf58 JS |
1993 | if (ret) |
1994 | break; | |
cd1b413c JS |
1995 | ULIST_ITER_INIT(&root_uiter); |
1996 | while (!ret && (root_node = ulist_next(roots, &root_uiter))) { | |
ab8d0fc4 JM |
1997 | btrfs_debug(fs_info, |
1998 | "root %llu references leaf %llu, data list %#llx", | |
1999 | root_node->val, ref_node->val, | |
2000 | ref_node->aux); | |
2001 | ret = iterate_leaf_refs(fs_info, | |
2002 | (struct extent_inode_elem *) | |
995e01b7 JS |
2003 | (uintptr_t)ref_node->aux, |
2004 | root_node->val, | |
2005 | extent_item_objectid, | |
2006 | iterate, ctx); | |
4692cf58 | 2007 | } |
976b1908 | 2008 | ulist_free(roots); |
a542ad1b JS |
2009 | } |
2010 | ||
976b1908 | 2011 | free_leaf_list(refs); |
4692cf58 | 2012 | out: |
bfc61c36 | 2013 | if (trans) { |
8445f61c | 2014 | btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
3a45bb20 | 2015 | btrfs_end_transaction(trans); |
9e351cc8 JB |
2016 | } else { |
2017 | up_read(&fs_info->commit_root_sem); | |
7a3ae2f8 JS |
2018 | } |
2019 | ||
a542ad1b JS |
2020 | return ret; |
2021 | } | |
2022 | ||
2023 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, | |
2024 | struct btrfs_path *path, | |
c995ab3c ZB |
2025 | iterate_extent_inodes_t *iterate, void *ctx, |
2026 | bool ignore_offset) | |
a542ad1b JS |
2027 | { |
2028 | int ret; | |
4692cf58 | 2029 | u64 extent_item_pos; |
69917e43 | 2030 | u64 flags = 0; |
a542ad1b | 2031 | struct btrfs_key found_key; |
7a3ae2f8 | 2032 | int search_commit_root = path->search_commit_root; |
a542ad1b | 2033 | |
69917e43 | 2034 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 2035 | btrfs_release_path(path); |
a542ad1b JS |
2036 | if (ret < 0) |
2037 | return ret; | |
69917e43 | 2038 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 2039 | return -EINVAL; |
a542ad1b | 2040 | |
4692cf58 | 2041 | extent_item_pos = logical - found_key.objectid; |
7a3ae2f8 JS |
2042 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
2043 | extent_item_pos, search_commit_root, | |
c995ab3c | 2044 | iterate, ctx, ignore_offset); |
a542ad1b JS |
2045 | |
2046 | return ret; | |
2047 | } | |
2048 | ||
d24bec3a MF |
2049 | typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off, |
2050 | struct extent_buffer *eb, void *ctx); | |
2051 | ||
2052 | static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, | |
2053 | struct btrfs_path *path, | |
2054 | iterate_irefs_t *iterate, void *ctx) | |
a542ad1b | 2055 | { |
aefc1eb1 | 2056 | int ret = 0; |
a542ad1b JS |
2057 | int slot; |
2058 | u32 cur; | |
2059 | u32 len; | |
2060 | u32 name_len; | |
2061 | u64 parent = 0; | |
2062 | int found = 0; | |
2063 | struct extent_buffer *eb; | |
2064 | struct btrfs_item *item; | |
2065 | struct btrfs_inode_ref *iref; | |
2066 | struct btrfs_key found_key; | |
2067 | ||
aefc1eb1 | 2068 | while (!ret) { |
c234a24d DS |
2069 | ret = btrfs_find_item(fs_root, path, inum, |
2070 | parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY, | |
2071 | &found_key); | |
2072 | ||
a542ad1b JS |
2073 | if (ret < 0) |
2074 | break; | |
2075 | if (ret) { | |
2076 | ret = found ? 0 : -ENOENT; | |
2077 | break; | |
2078 | } | |
2079 | ++found; | |
2080 | ||
2081 | parent = found_key.offset; | |
2082 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2083 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2084 | if (!eb) { | |
2085 | ret = -ENOMEM; | |
2086 | break; | |
2087 | } | |
a542ad1b JS |
2088 | btrfs_release_path(path); |
2089 | ||
dd3cc16b | 2090 | item = btrfs_item_nr(slot); |
a542ad1b JS |
2091 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
2092 | ||
2093 | for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) { | |
2094 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
2095 | /* path must be released before calling iterate()! */ | |
ab8d0fc4 JM |
2096 | btrfs_debug(fs_root->fs_info, |
2097 | "following ref at offset %u for inode %llu in tree %llu", | |
4fd786e6 MT |
2098 | cur, found_key.objectid, |
2099 | fs_root->root_key.objectid); | |
d24bec3a MF |
2100 | ret = iterate(parent, name_len, |
2101 | (unsigned long)(iref + 1), eb, ctx); | |
aefc1eb1 | 2102 | if (ret) |
a542ad1b | 2103 | break; |
a542ad1b JS |
2104 | len = sizeof(*iref) + name_len; |
2105 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
2106 | } | |
2107 | free_extent_buffer(eb); | |
2108 | } | |
2109 | ||
2110 | btrfs_release_path(path); | |
2111 | ||
2112 | return ret; | |
2113 | } | |
2114 | ||
d24bec3a MF |
2115 | static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, |
2116 | struct btrfs_path *path, | |
2117 | iterate_irefs_t *iterate, void *ctx) | |
2118 | { | |
2119 | int ret; | |
2120 | int slot; | |
2121 | u64 offset = 0; | |
2122 | u64 parent; | |
2123 | int found = 0; | |
2124 | struct extent_buffer *eb; | |
2125 | struct btrfs_inode_extref *extref; | |
d24bec3a MF |
2126 | u32 item_size; |
2127 | u32 cur_offset; | |
2128 | unsigned long ptr; | |
2129 | ||
2130 | while (1) { | |
2131 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
2132 | &offset); | |
2133 | if (ret < 0) | |
2134 | break; | |
2135 | if (ret) { | |
2136 | ret = found ? 0 : -ENOENT; | |
2137 | break; | |
2138 | } | |
2139 | ++found; | |
2140 | ||
2141 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2142 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2143 | if (!eb) { | |
2144 | ret = -ENOMEM; | |
2145 | break; | |
2146 | } | |
d24bec3a MF |
2147 | btrfs_release_path(path); |
2148 | ||
2849a854 CM |
2149 | item_size = btrfs_item_size_nr(eb, slot); |
2150 | ptr = btrfs_item_ptr_offset(eb, slot); | |
d24bec3a MF |
2151 | cur_offset = 0; |
2152 | ||
2153 | while (cur_offset < item_size) { | |
2154 | u32 name_len; | |
2155 | ||
2156 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
2157 | parent = btrfs_inode_extref_parent(eb, extref); | |
2158 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
2159 | ret = iterate(parent, name_len, | |
2160 | (unsigned long)&extref->name, eb, ctx); | |
2161 | if (ret) | |
2162 | break; | |
2163 | ||
2849a854 | 2164 | cur_offset += btrfs_inode_extref_name_len(eb, extref); |
d24bec3a MF |
2165 | cur_offset += sizeof(*extref); |
2166 | } | |
d24bec3a MF |
2167 | free_extent_buffer(eb); |
2168 | ||
2169 | offset++; | |
2170 | } | |
2171 | ||
2172 | btrfs_release_path(path); | |
2173 | ||
2174 | return ret; | |
2175 | } | |
2176 | ||
2177 | static int iterate_irefs(u64 inum, struct btrfs_root *fs_root, | |
2178 | struct btrfs_path *path, iterate_irefs_t *iterate, | |
2179 | void *ctx) | |
2180 | { | |
2181 | int ret; | |
2182 | int found_refs = 0; | |
2183 | ||
2184 | ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx); | |
2185 | if (!ret) | |
2186 | ++found_refs; | |
2187 | else if (ret != -ENOENT) | |
2188 | return ret; | |
2189 | ||
2190 | ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx); | |
2191 | if (ret == -ENOENT && found_refs) | |
2192 | return 0; | |
2193 | ||
2194 | return ret; | |
2195 | } | |
2196 | ||
a542ad1b JS |
2197 | /* |
2198 | * returns 0 if the path could be dumped (probably truncated) | |
2199 | * returns <0 in case of an error | |
2200 | */ | |
d24bec3a MF |
2201 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
2202 | struct extent_buffer *eb, void *ctx) | |
a542ad1b JS |
2203 | { |
2204 | struct inode_fs_paths *ipath = ctx; | |
2205 | char *fspath; | |
2206 | char *fspath_min; | |
2207 | int i = ipath->fspath->elem_cnt; | |
2208 | const int s_ptr = sizeof(char *); | |
2209 | u32 bytes_left; | |
2210 | ||
2211 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
2212 | ipath->fspath->bytes_left - s_ptr : 0; | |
2213 | ||
740c3d22 | 2214 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
2215 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
2216 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
2217 | if (IS_ERR(fspath)) |
2218 | return PTR_ERR(fspath); | |
2219 | ||
2220 | if (fspath > fspath_min) { | |
745c4d8e | 2221 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
2222 | ++ipath->fspath->elem_cnt; |
2223 | ipath->fspath->bytes_left = fspath - fspath_min; | |
2224 | } else { | |
2225 | ++ipath->fspath->elem_missed; | |
2226 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
2227 | ipath->fspath->bytes_left = 0; | |
2228 | } | |
2229 | ||
2230 | return 0; | |
2231 | } | |
2232 | ||
2233 | /* | |
2234 | * this dumps all file system paths to the inode into the ipath struct, provided | |
2235 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 2236 | * from ipath->fspath->val[i]. |
a542ad1b | 2237 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 2238 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
01327610 | 2239 | * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise, |
a542ad1b JS |
2240 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would |
2241 | * have been needed to return all paths. | |
2242 | */ | |
2243 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
2244 | { | |
2245 | return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path, | |
d24bec3a | 2246 | inode_to_path, ipath); |
a542ad1b JS |
2247 | } |
2248 | ||
a542ad1b JS |
2249 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
2250 | { | |
2251 | struct btrfs_data_container *data; | |
2252 | size_t alloc_bytes; | |
2253 | ||
2254 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
f54de068 | 2255 | data = kvmalloc(alloc_bytes, GFP_KERNEL); |
a542ad1b JS |
2256 | if (!data) |
2257 | return ERR_PTR(-ENOMEM); | |
2258 | ||
2259 | if (total_bytes >= sizeof(*data)) { | |
2260 | data->bytes_left = total_bytes - sizeof(*data); | |
2261 | data->bytes_missing = 0; | |
2262 | } else { | |
2263 | data->bytes_missing = sizeof(*data) - total_bytes; | |
2264 | data->bytes_left = 0; | |
2265 | } | |
2266 | ||
2267 | data->elem_cnt = 0; | |
2268 | data->elem_missed = 0; | |
2269 | ||
2270 | return data; | |
2271 | } | |
2272 | ||
2273 | /* | |
2274 | * allocates space to return multiple file system paths for an inode. | |
2275 | * total_bytes to allocate are passed, note that space usable for actual path | |
2276 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
2277 | * the returned pointer must be freed with free_ipath() in the end. | |
2278 | */ | |
2279 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
2280 | struct btrfs_path *path) | |
2281 | { | |
2282 | struct inode_fs_paths *ifp; | |
2283 | struct btrfs_data_container *fspath; | |
2284 | ||
2285 | fspath = init_data_container(total_bytes); | |
2286 | if (IS_ERR(fspath)) | |
afc6961f | 2287 | return ERR_CAST(fspath); |
a542ad1b | 2288 | |
f54de068 | 2289 | ifp = kmalloc(sizeof(*ifp), GFP_KERNEL); |
a542ad1b | 2290 | if (!ifp) { |
f54de068 | 2291 | kvfree(fspath); |
a542ad1b JS |
2292 | return ERR_PTR(-ENOMEM); |
2293 | } | |
2294 | ||
2295 | ifp->btrfs_path = path; | |
2296 | ifp->fspath = fspath; | |
2297 | ifp->fs_root = fs_root; | |
2298 | ||
2299 | return ifp; | |
2300 | } | |
2301 | ||
2302 | void free_ipath(struct inode_fs_paths *ipath) | |
2303 | { | |
4735fb28 JJ |
2304 | if (!ipath) |
2305 | return; | |
f54de068 | 2306 | kvfree(ipath->fspath); |
a542ad1b JS |
2307 | kfree(ipath); |
2308 | } |