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" |
a542ad1b | 18 | |
dc046b10 JB |
19 | /* Just an arbitrary number so we can be sure this happened */ |
20 | #define BACKREF_FOUND_SHARED 6 | |
21 | ||
976b1908 JS |
22 | struct extent_inode_elem { |
23 | u64 inum; | |
24 | u64 offset; | |
25 | struct extent_inode_elem *next; | |
26 | }; | |
27 | ||
73980bec JM |
28 | static int check_extent_in_eb(const struct btrfs_key *key, |
29 | const struct extent_buffer *eb, | |
30 | const struct btrfs_file_extent_item *fi, | |
31 | u64 extent_item_pos, | |
c995ab3c ZB |
32 | struct extent_inode_elem **eie, |
33 | bool ignore_offset) | |
976b1908 | 34 | { |
8ca15e05 | 35 | u64 offset = 0; |
976b1908 JS |
36 | struct extent_inode_elem *e; |
37 | ||
c995ab3c ZB |
38 | if (!ignore_offset && |
39 | !btrfs_file_extent_compression(eb, fi) && | |
8ca15e05 JB |
40 | !btrfs_file_extent_encryption(eb, fi) && |
41 | !btrfs_file_extent_other_encoding(eb, fi)) { | |
42 | u64 data_offset; | |
43 | u64 data_len; | |
976b1908 | 44 | |
8ca15e05 JB |
45 | data_offset = btrfs_file_extent_offset(eb, fi); |
46 | data_len = btrfs_file_extent_num_bytes(eb, fi); | |
47 | ||
48 | if (extent_item_pos < data_offset || | |
49 | extent_item_pos >= data_offset + data_len) | |
50 | return 1; | |
51 | offset = extent_item_pos - data_offset; | |
52 | } | |
976b1908 JS |
53 | |
54 | e = kmalloc(sizeof(*e), GFP_NOFS); | |
55 | if (!e) | |
56 | return -ENOMEM; | |
57 | ||
58 | e->next = *eie; | |
59 | e->inum = key->objectid; | |
8ca15e05 | 60 | e->offset = key->offset + offset; |
976b1908 JS |
61 | *eie = e; |
62 | ||
63 | return 0; | |
64 | } | |
65 | ||
f05c4746 WS |
66 | static void free_inode_elem_list(struct extent_inode_elem *eie) |
67 | { | |
68 | struct extent_inode_elem *eie_next; | |
69 | ||
70 | for (; eie; eie = eie_next) { | |
71 | eie_next = eie->next; | |
72 | kfree(eie); | |
73 | } | |
74 | } | |
75 | ||
73980bec JM |
76 | static int find_extent_in_eb(const struct extent_buffer *eb, |
77 | u64 wanted_disk_byte, u64 extent_item_pos, | |
c995ab3c ZB |
78 | struct extent_inode_elem **eie, |
79 | bool ignore_offset) | |
976b1908 JS |
80 | { |
81 | u64 disk_byte; | |
82 | struct btrfs_key key; | |
83 | struct btrfs_file_extent_item *fi; | |
84 | int slot; | |
85 | int nritems; | |
86 | int extent_type; | |
87 | int ret; | |
88 | ||
89 | /* | |
90 | * from the shared data ref, we only have the leaf but we need | |
91 | * the key. thus, we must look into all items and see that we | |
92 | * find one (some) with a reference to our extent item. | |
93 | */ | |
94 | nritems = btrfs_header_nritems(eb); | |
95 | for (slot = 0; slot < nritems; ++slot) { | |
96 | btrfs_item_key_to_cpu(eb, &key, slot); | |
97 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
98 | continue; | |
99 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
100 | extent_type = btrfs_file_extent_type(eb, fi); | |
101 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
102 | continue; | |
103 | /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ | |
104 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
105 | if (disk_byte != wanted_disk_byte) | |
106 | continue; | |
107 | ||
c995ab3c | 108 | ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset); |
976b1908 JS |
109 | if (ret < 0) |
110 | return ret; | |
111 | } | |
112 | ||
113 | return 0; | |
114 | } | |
115 | ||
86d5f994 | 116 | struct preftree { |
ecf160b4 | 117 | struct rb_root_cached root; |
6c336b21 | 118 | unsigned int count; |
86d5f994 EN |
119 | }; |
120 | ||
ecf160b4 | 121 | #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 } |
86d5f994 EN |
122 | |
123 | struct preftrees { | |
124 | struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */ | |
125 | struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */ | |
126 | struct preftree indirect_missing_keys; | |
127 | }; | |
128 | ||
3ec4d323 EN |
129 | /* |
130 | * Checks for a shared extent during backref search. | |
131 | * | |
132 | * The share_count tracks prelim_refs (direct and indirect) having a | |
133 | * ref->count >0: | |
134 | * - incremented when a ref->count transitions to >0 | |
135 | * - decremented when a ref->count transitions to <1 | |
136 | */ | |
137 | struct share_check { | |
138 | u64 root_objectid; | |
139 | u64 inum; | |
140 | int share_count; | |
4fc7b572 | 141 | bool have_delayed_delete_refs; |
3ec4d323 EN |
142 | }; |
143 | ||
144 | static inline int extent_is_shared(struct share_check *sc) | |
145 | { | |
146 | return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0; | |
147 | } | |
148 | ||
b9e9a6cb WS |
149 | static struct kmem_cache *btrfs_prelim_ref_cache; |
150 | ||
151 | int __init btrfs_prelim_ref_init(void) | |
152 | { | |
153 | btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", | |
e0c476b1 | 154 | sizeof(struct prelim_ref), |
b9e9a6cb | 155 | 0, |
fba4b697 | 156 | SLAB_MEM_SPREAD, |
b9e9a6cb WS |
157 | NULL); |
158 | if (!btrfs_prelim_ref_cache) | |
159 | return -ENOMEM; | |
160 | return 0; | |
161 | } | |
162 | ||
e67c718b | 163 | void __cold btrfs_prelim_ref_exit(void) |
b9e9a6cb | 164 | { |
5598e900 | 165 | kmem_cache_destroy(btrfs_prelim_ref_cache); |
b9e9a6cb WS |
166 | } |
167 | ||
86d5f994 EN |
168 | static void free_pref(struct prelim_ref *ref) |
169 | { | |
170 | kmem_cache_free(btrfs_prelim_ref_cache, ref); | |
171 | } | |
172 | ||
173 | /* | |
174 | * Return 0 when both refs are for the same block (and can be merged). | |
175 | * A -1 return indicates ref1 is a 'lower' block than ref2, while 1 | |
176 | * indicates a 'higher' block. | |
177 | */ | |
178 | static int prelim_ref_compare(struct prelim_ref *ref1, | |
179 | struct prelim_ref *ref2) | |
180 | { | |
181 | if (ref1->level < ref2->level) | |
182 | return -1; | |
183 | if (ref1->level > ref2->level) | |
184 | return 1; | |
185 | if (ref1->root_id < ref2->root_id) | |
186 | return -1; | |
187 | if (ref1->root_id > ref2->root_id) | |
188 | return 1; | |
189 | if (ref1->key_for_search.type < ref2->key_for_search.type) | |
190 | return -1; | |
191 | if (ref1->key_for_search.type > ref2->key_for_search.type) | |
192 | return 1; | |
193 | if (ref1->key_for_search.objectid < ref2->key_for_search.objectid) | |
194 | return -1; | |
195 | if (ref1->key_for_search.objectid > ref2->key_for_search.objectid) | |
196 | return 1; | |
197 | if (ref1->key_for_search.offset < ref2->key_for_search.offset) | |
198 | return -1; | |
199 | if (ref1->key_for_search.offset > ref2->key_for_search.offset) | |
200 | return 1; | |
201 | if (ref1->parent < ref2->parent) | |
202 | return -1; | |
203 | if (ref1->parent > ref2->parent) | |
204 | return 1; | |
205 | ||
206 | return 0; | |
207 | } | |
208 | ||
ccc8dc75 CIK |
209 | static void update_share_count(struct share_check *sc, int oldcount, |
210 | int newcount) | |
3ec4d323 EN |
211 | { |
212 | if ((!sc) || (oldcount == 0 && newcount < 1)) | |
213 | return; | |
214 | ||
215 | if (oldcount > 0 && newcount < 1) | |
216 | sc->share_count--; | |
217 | else if (oldcount < 1 && newcount > 0) | |
218 | sc->share_count++; | |
219 | } | |
220 | ||
86d5f994 EN |
221 | /* |
222 | * Add @newref to the @root rbtree, merging identical refs. | |
223 | * | |
3ec4d323 | 224 | * Callers should assume that newref has been freed after calling. |
86d5f994 | 225 | */ |
00142756 JM |
226 | static void prelim_ref_insert(const struct btrfs_fs_info *fs_info, |
227 | struct preftree *preftree, | |
3ec4d323 EN |
228 | struct prelim_ref *newref, |
229 | struct share_check *sc) | |
86d5f994 | 230 | { |
ecf160b4 | 231 | struct rb_root_cached *root; |
86d5f994 EN |
232 | struct rb_node **p; |
233 | struct rb_node *parent = NULL; | |
234 | struct prelim_ref *ref; | |
235 | int result; | |
ecf160b4 | 236 | bool leftmost = true; |
86d5f994 EN |
237 | |
238 | root = &preftree->root; | |
ecf160b4 | 239 | p = &root->rb_root.rb_node; |
86d5f994 EN |
240 | |
241 | while (*p) { | |
242 | parent = *p; | |
243 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
244 | result = prelim_ref_compare(ref, newref); | |
245 | if (result < 0) { | |
246 | p = &(*p)->rb_left; | |
247 | } else if (result > 0) { | |
248 | p = &(*p)->rb_right; | |
ecf160b4 | 249 | leftmost = false; |
86d5f994 EN |
250 | } else { |
251 | /* Identical refs, merge them and free @newref */ | |
252 | struct extent_inode_elem *eie = ref->inode_list; | |
253 | ||
254 | while (eie && eie->next) | |
255 | eie = eie->next; | |
256 | ||
257 | if (!eie) | |
258 | ref->inode_list = newref->inode_list; | |
259 | else | |
260 | eie->next = newref->inode_list; | |
00142756 JM |
261 | trace_btrfs_prelim_ref_merge(fs_info, ref, newref, |
262 | preftree->count); | |
3ec4d323 EN |
263 | /* |
264 | * A delayed ref can have newref->count < 0. | |
265 | * The ref->count is updated to follow any | |
266 | * BTRFS_[ADD|DROP]_DELAYED_REF actions. | |
267 | */ | |
268 | update_share_count(sc, ref->count, | |
269 | ref->count + newref->count); | |
86d5f994 EN |
270 | ref->count += newref->count; |
271 | free_pref(newref); | |
272 | return; | |
273 | } | |
274 | } | |
275 | ||
3ec4d323 | 276 | update_share_count(sc, 0, newref->count); |
6c336b21 | 277 | preftree->count++; |
00142756 | 278 | trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count); |
86d5f994 | 279 | rb_link_node(&newref->rbnode, parent, p); |
ecf160b4 | 280 | rb_insert_color_cached(&newref->rbnode, root, leftmost); |
86d5f994 EN |
281 | } |
282 | ||
283 | /* | |
284 | * Release the entire tree. We don't care about internal consistency so | |
285 | * just free everything and then reset the tree root. | |
286 | */ | |
287 | static void prelim_release(struct preftree *preftree) | |
288 | { | |
289 | struct prelim_ref *ref, *next_ref; | |
290 | ||
ecf160b4 LB |
291 | rbtree_postorder_for_each_entry_safe(ref, next_ref, |
292 | &preftree->root.rb_root, rbnode) | |
86d5f994 EN |
293 | free_pref(ref); |
294 | ||
ecf160b4 | 295 | preftree->root = RB_ROOT_CACHED; |
6c336b21 | 296 | preftree->count = 0; |
86d5f994 EN |
297 | } |
298 | ||
d5c88b73 JS |
299 | /* |
300 | * the rules for all callers of this function are: | |
301 | * - obtaining the parent is the goal | |
302 | * - if you add a key, you must know that it is a correct key | |
303 | * - if you cannot add the parent or a correct key, then we will look into the | |
304 | * block later to set a correct key | |
305 | * | |
306 | * delayed refs | |
307 | * ============ | |
308 | * backref type | shared | indirect | shared | indirect | |
309 | * information | tree | tree | data | data | |
310 | * --------------------+--------+----------+--------+---------- | |
311 | * parent logical | y | - | - | - | |
312 | * key to resolve | - | y | y | y | |
313 | * tree block logical | - | - | - | - | |
314 | * root for resolving | y | y | y | y | |
315 | * | |
316 | * - column 1: we've the parent -> done | |
317 | * - column 2, 3, 4: we use the key to find the parent | |
318 | * | |
319 | * on disk refs (inline or keyed) | |
320 | * ============================== | |
321 | * backref type | shared | indirect | shared | indirect | |
322 | * information | tree | tree | data | data | |
323 | * --------------------+--------+----------+--------+---------- | |
324 | * parent logical | y | - | y | - | |
325 | * key to resolve | - | - | - | y | |
326 | * tree block logical | y | y | y | y | |
327 | * root for resolving | - | y | y | y | |
328 | * | |
329 | * - column 1, 3: we've the parent -> done | |
330 | * - column 2: we take the first key from the block to find the parent | |
e0c476b1 | 331 | * (see add_missing_keys) |
d5c88b73 JS |
332 | * - column 4: we use the key to find the parent |
333 | * | |
334 | * additional information that's available but not required to find the parent | |
335 | * block might help in merging entries to gain some speed. | |
336 | */ | |
00142756 JM |
337 | static int add_prelim_ref(const struct btrfs_fs_info *fs_info, |
338 | struct preftree *preftree, u64 root_id, | |
e0c476b1 | 339 | const struct btrfs_key *key, int level, u64 parent, |
3ec4d323 EN |
340 | u64 wanted_disk_byte, int count, |
341 | struct share_check *sc, gfp_t gfp_mask) | |
8da6d581 | 342 | { |
e0c476b1 | 343 | struct prelim_ref *ref; |
8da6d581 | 344 | |
48ec4736 LB |
345 | if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) |
346 | return 0; | |
347 | ||
b9e9a6cb | 348 | ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); |
8da6d581 JS |
349 | if (!ref) |
350 | return -ENOMEM; | |
351 | ||
352 | ref->root_id = root_id; | |
7ac8b88e | 353 | if (key) |
d5c88b73 | 354 | ref->key_for_search = *key; |
7ac8b88e | 355 | else |
d5c88b73 | 356 | memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); |
8da6d581 | 357 | |
3301958b | 358 | ref->inode_list = NULL; |
8da6d581 JS |
359 | ref->level = level; |
360 | ref->count = count; | |
361 | ref->parent = parent; | |
362 | ref->wanted_disk_byte = wanted_disk_byte; | |
3ec4d323 EN |
363 | prelim_ref_insert(fs_info, preftree, ref, sc); |
364 | return extent_is_shared(sc); | |
8da6d581 JS |
365 | } |
366 | ||
86d5f994 | 367 | /* direct refs use root == 0, key == NULL */ |
00142756 JM |
368 | static int add_direct_ref(const struct btrfs_fs_info *fs_info, |
369 | struct preftrees *preftrees, int level, u64 parent, | |
3ec4d323 EN |
370 | u64 wanted_disk_byte, int count, |
371 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 | 372 | { |
00142756 | 373 | return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level, |
3ec4d323 | 374 | parent, wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
375 | } |
376 | ||
377 | /* indirect refs use parent == 0 */ | |
00142756 JM |
378 | static int add_indirect_ref(const struct btrfs_fs_info *fs_info, |
379 | struct preftrees *preftrees, u64 root_id, | |
86d5f994 | 380 | const struct btrfs_key *key, int level, |
3ec4d323 EN |
381 | u64 wanted_disk_byte, int count, |
382 | struct share_check *sc, gfp_t gfp_mask) | |
86d5f994 EN |
383 | { |
384 | struct preftree *tree = &preftrees->indirect; | |
385 | ||
386 | if (!key) | |
387 | tree = &preftrees->indirect_missing_keys; | |
00142756 | 388 | return add_prelim_ref(fs_info, tree, root_id, key, level, 0, |
3ec4d323 | 389 | wanted_disk_byte, count, sc, gfp_mask); |
86d5f994 EN |
390 | } |
391 | ||
ed58f2e6 | 392 | static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr) |
393 | { | |
394 | struct rb_node **p = &preftrees->direct.root.rb_root.rb_node; | |
395 | struct rb_node *parent = NULL; | |
396 | struct prelim_ref *ref = NULL; | |
9c6c723f | 397 | struct prelim_ref target = {}; |
ed58f2e6 | 398 | int result; |
399 | ||
400 | target.parent = bytenr; | |
401 | ||
402 | while (*p) { | |
403 | parent = *p; | |
404 | ref = rb_entry(parent, struct prelim_ref, rbnode); | |
405 | result = prelim_ref_compare(ref, &target); | |
406 | ||
407 | if (result < 0) | |
408 | p = &(*p)->rb_left; | |
409 | else if (result > 0) | |
410 | p = &(*p)->rb_right; | |
411 | else | |
412 | return 1; | |
413 | } | |
414 | return 0; | |
415 | } | |
416 | ||
8da6d581 | 417 | static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, |
ed58f2e6 | 418 | struct ulist *parents, |
419 | struct preftrees *preftrees, struct prelim_ref *ref, | |
44853868 | 420 | int level, u64 time_seq, const u64 *extent_item_pos, |
b25b0b87 | 421 | bool ignore_offset) |
8da6d581 | 422 | { |
69bca40d AB |
423 | int ret = 0; |
424 | int slot; | |
425 | struct extent_buffer *eb; | |
426 | struct btrfs_key key; | |
7ef81ac8 | 427 | struct btrfs_key *key_for_search = &ref->key_for_search; |
8da6d581 | 428 | struct btrfs_file_extent_item *fi; |
ed8c4913 | 429 | struct extent_inode_elem *eie = NULL, *old = NULL; |
8da6d581 | 430 | u64 disk_byte; |
7ef81ac8 JB |
431 | u64 wanted_disk_byte = ref->wanted_disk_byte; |
432 | u64 count = 0; | |
7ac8b88e | 433 | u64 data_offset; |
8da6d581 | 434 | |
69bca40d AB |
435 | if (level != 0) { |
436 | eb = path->nodes[level]; | |
437 | ret = ulist_add(parents, eb->start, 0, GFP_NOFS); | |
3301958b JS |
438 | if (ret < 0) |
439 | return ret; | |
8da6d581 | 440 | return 0; |
69bca40d | 441 | } |
8da6d581 JS |
442 | |
443 | /* | |
ed58f2e6 | 444 | * 1. We normally enter this function with the path already pointing to |
445 | * the first item to check. But sometimes, we may enter it with | |
446 | * slot == nritems. | |
447 | * 2. We are searching for normal backref but bytenr of this leaf | |
448 | * matches shared data backref | |
cfc0eed0 | 449 | * 3. The leaf owner is not equal to the root we are searching |
450 | * | |
ed58f2e6 | 451 | * For these cases, go to the next leaf before we continue. |
8da6d581 | 452 | */ |
ed58f2e6 | 453 | eb = path->nodes[0]; |
454 | if (path->slots[0] >= btrfs_header_nritems(eb) || | |
cfc0eed0 | 455 | is_shared_data_backref(preftrees, eb->start) || |
456 | ref->root_id != btrfs_header_owner(eb)) { | |
f3a84ccd | 457 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
458 | ret = btrfs_next_leaf(root, path); |
459 | else | |
460 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
461 | } | |
8da6d581 | 462 | |
b25b0b87 | 463 | while (!ret && count < ref->count) { |
8da6d581 | 464 | eb = path->nodes[0]; |
69bca40d AB |
465 | slot = path->slots[0]; |
466 | ||
467 | btrfs_item_key_to_cpu(eb, &key, slot); | |
468 | ||
469 | if (key.objectid != key_for_search->objectid || | |
470 | key.type != BTRFS_EXTENT_DATA_KEY) | |
471 | break; | |
472 | ||
ed58f2e6 | 473 | /* |
474 | * We are searching for normal backref but bytenr of this leaf | |
cfc0eed0 | 475 | * matches shared data backref, OR |
476 | * the leaf owner is not equal to the root we are searching for | |
ed58f2e6 | 477 | */ |
cfc0eed0 | 478 | if (slot == 0 && |
479 | (is_shared_data_backref(preftrees, eb->start) || | |
480 | ref->root_id != btrfs_header_owner(eb))) { | |
f3a84ccd | 481 | if (time_seq == BTRFS_SEQ_LAST) |
ed58f2e6 | 482 | ret = btrfs_next_leaf(root, path); |
483 | else | |
484 | ret = btrfs_next_old_leaf(root, path, time_seq); | |
485 | continue; | |
486 | } | |
69bca40d AB |
487 | fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
488 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); | |
7ac8b88e | 489 | data_offset = btrfs_file_extent_offset(eb, fi); |
69bca40d AB |
490 | |
491 | if (disk_byte == wanted_disk_byte) { | |
492 | eie = NULL; | |
ed8c4913 | 493 | old = NULL; |
7ac8b88e | 494 | if (ref->key_for_search.offset == key.offset - data_offset) |
495 | count++; | |
496 | else | |
497 | goto next; | |
69bca40d AB |
498 | if (extent_item_pos) { |
499 | ret = check_extent_in_eb(&key, eb, fi, | |
500 | *extent_item_pos, | |
c995ab3c | 501 | &eie, ignore_offset); |
69bca40d AB |
502 | if (ret < 0) |
503 | break; | |
504 | } | |
ed8c4913 JB |
505 | if (ret > 0) |
506 | goto next; | |
4eb1f66d TI |
507 | ret = ulist_add_merge_ptr(parents, eb->start, |
508 | eie, (void **)&old, GFP_NOFS); | |
ed8c4913 JB |
509 | if (ret < 0) |
510 | break; | |
511 | if (!ret && extent_item_pos) { | |
512 | while (old->next) | |
513 | old = old->next; | |
514 | old->next = eie; | |
69bca40d | 515 | } |
f05c4746 | 516 | eie = NULL; |
8da6d581 | 517 | } |
ed8c4913 | 518 | next: |
f3a84ccd | 519 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
520 | ret = btrfs_next_item(root, path); |
521 | else | |
522 | ret = btrfs_next_old_item(root, path, time_seq); | |
8da6d581 JS |
523 | } |
524 | ||
69bca40d AB |
525 | if (ret > 0) |
526 | ret = 0; | |
f05c4746 WS |
527 | else if (ret < 0) |
528 | free_inode_elem_list(eie); | |
69bca40d | 529 | return ret; |
8da6d581 JS |
530 | } |
531 | ||
532 | /* | |
533 | * resolve an indirect backref in the form (root_id, key, level) | |
534 | * to a logical address | |
535 | */ | |
e0c476b1 JM |
536 | static int resolve_indirect_ref(struct btrfs_fs_info *fs_info, |
537 | struct btrfs_path *path, u64 time_seq, | |
ed58f2e6 | 538 | struct preftrees *preftrees, |
e0c476b1 | 539 | struct prelim_ref *ref, struct ulist *parents, |
b25b0b87 | 540 | const u64 *extent_item_pos, bool ignore_offset) |
8da6d581 | 541 | { |
8da6d581 | 542 | struct btrfs_root *root; |
8da6d581 JS |
543 | struct extent_buffer *eb; |
544 | int ret = 0; | |
545 | int root_level; | |
546 | int level = ref->level; | |
7ac8b88e | 547 | struct btrfs_key search_key = ref->key_for_search; |
8da6d581 | 548 | |
49d11bea JB |
549 | /* |
550 | * If we're search_commit_root we could possibly be holding locks on | |
551 | * other tree nodes. This happens when qgroups does backref walks when | |
552 | * adding new delayed refs. To deal with this we need to look in cache | |
553 | * for the root, and if we don't find it then we need to search the | |
554 | * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage | |
555 | * here. | |
556 | */ | |
557 | if (path->search_commit_root) | |
558 | root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id); | |
559 | else | |
560 | root = btrfs_get_fs_root(fs_info, ref->root_id, false); | |
8da6d581 JS |
561 | if (IS_ERR(root)) { |
562 | ret = PTR_ERR(root); | |
9326f76f JB |
563 | goto out_free; |
564 | } | |
565 | ||
39dba873 JB |
566 | if (!path->search_commit_root && |
567 | test_bit(BTRFS_ROOT_DELETING, &root->state)) { | |
568 | ret = -ENOENT; | |
569 | goto out; | |
570 | } | |
571 | ||
f5ee5c9a | 572 | if (btrfs_is_testing(fs_info)) { |
d9ee522b JB |
573 | ret = -ENOENT; |
574 | goto out; | |
575 | } | |
576 | ||
9e351cc8 JB |
577 | if (path->search_commit_root) |
578 | root_level = btrfs_header_level(root->commit_root); | |
f3a84ccd | 579 | else if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 | 580 | root_level = btrfs_header_level(root->node); |
9e351cc8 JB |
581 | else |
582 | root_level = btrfs_old_root_level(root, time_seq); | |
8da6d581 | 583 | |
c75e8394 | 584 | if (root_level + 1 == level) |
8da6d581 JS |
585 | goto out; |
586 | ||
7ac8b88e | 587 | /* |
588 | * We can often find data backrefs with an offset that is too large | |
589 | * (>= LLONG_MAX, maximum allowed file offset) due to underflows when | |
590 | * subtracting a file's offset with the data offset of its | |
591 | * corresponding extent data item. This can happen for example in the | |
592 | * clone ioctl. | |
593 | * | |
594 | * So if we detect such case we set the search key's offset to zero to | |
595 | * make sure we will find the matching file extent item at | |
596 | * add_all_parents(), otherwise we will miss it because the offset | |
597 | * taken form the backref is much larger then the offset of the file | |
598 | * extent item. This can make us scan a very large number of file | |
599 | * extent items, but at least it will not make us miss any. | |
600 | * | |
601 | * This is an ugly workaround for a behaviour that should have never | |
602 | * existed, but it does and a fix for the clone ioctl would touch a lot | |
603 | * of places, cause backwards incompatibility and would not fix the | |
604 | * problem for extents cloned with older kernels. | |
605 | */ | |
606 | if (search_key.type == BTRFS_EXTENT_DATA_KEY && | |
607 | search_key.offset >= LLONG_MAX) | |
608 | search_key.offset = 0; | |
8da6d581 | 609 | path->lowest_level = level; |
f3a84ccd | 610 | if (time_seq == BTRFS_SEQ_LAST) |
7ac8b88e | 611 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); |
21633fc6 | 612 | else |
7ac8b88e | 613 | ret = btrfs_search_old_slot(root, &search_key, path, time_seq); |
538f72cd | 614 | |
ab8d0fc4 JM |
615 | btrfs_debug(fs_info, |
616 | "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)", | |
c1c9ff7c GU |
617 | ref->root_id, level, ref->count, ret, |
618 | ref->key_for_search.objectid, ref->key_for_search.type, | |
619 | ref->key_for_search.offset); | |
8da6d581 JS |
620 | if (ret < 0) |
621 | goto out; | |
622 | ||
623 | eb = path->nodes[level]; | |
9345457f | 624 | while (!eb) { |
fae7f21c | 625 | if (WARN_ON(!level)) { |
9345457f JS |
626 | ret = 1; |
627 | goto out; | |
628 | } | |
629 | level--; | |
630 | eb = path->nodes[level]; | |
8da6d581 JS |
631 | } |
632 | ||
ed58f2e6 | 633 | ret = add_all_parents(root, path, parents, preftrees, ref, level, |
b25b0b87 | 634 | time_seq, extent_item_pos, ignore_offset); |
8da6d581 | 635 | out: |
00246528 | 636 | btrfs_put_root(root); |
9326f76f | 637 | out_free: |
da61d31a JB |
638 | path->lowest_level = 0; |
639 | btrfs_release_path(path); | |
8da6d581 JS |
640 | return ret; |
641 | } | |
642 | ||
4dae077a JM |
643 | static struct extent_inode_elem * |
644 | unode_aux_to_inode_list(struct ulist_node *node) | |
645 | { | |
646 | if (!node) | |
647 | return NULL; | |
648 | return (struct extent_inode_elem *)(uintptr_t)node->aux; | |
649 | } | |
650 | ||
8da6d581 | 651 | /* |
52042d8e | 652 | * We maintain three separate rbtrees: one for direct refs, one for |
86d5f994 EN |
653 | * indirect refs which have a key, and one for indirect refs which do not |
654 | * have a key. Each tree does merge on insertion. | |
655 | * | |
656 | * Once all of the references are located, we iterate over the tree of | |
657 | * indirect refs with missing keys. An appropriate key is located and | |
658 | * the ref is moved onto the tree for indirect refs. After all missing | |
659 | * keys are thus located, we iterate over the indirect ref tree, resolve | |
660 | * each reference, and then insert the resolved reference onto the | |
661 | * direct tree (merging there too). | |
662 | * | |
663 | * New backrefs (i.e., for parent nodes) are added to the appropriate | |
664 | * rbtree as they are encountered. The new backrefs are subsequently | |
665 | * resolved as above. | |
8da6d581 | 666 | */ |
e0c476b1 JM |
667 | static int resolve_indirect_refs(struct btrfs_fs_info *fs_info, |
668 | struct btrfs_path *path, u64 time_seq, | |
86d5f994 | 669 | struct preftrees *preftrees, |
b25b0b87 | 670 | const u64 *extent_item_pos, |
c995ab3c | 671 | struct share_check *sc, bool ignore_offset) |
8da6d581 JS |
672 | { |
673 | int err; | |
674 | int ret = 0; | |
8da6d581 JS |
675 | struct ulist *parents; |
676 | struct ulist_node *node; | |
cd1b413c | 677 | struct ulist_iterator uiter; |
86d5f994 | 678 | struct rb_node *rnode; |
8da6d581 JS |
679 | |
680 | parents = ulist_alloc(GFP_NOFS); | |
681 | if (!parents) | |
682 | return -ENOMEM; | |
683 | ||
684 | /* | |
86d5f994 EN |
685 | * We could trade memory usage for performance here by iterating |
686 | * the tree, allocating new refs for each insertion, and then | |
687 | * freeing the entire indirect tree when we're done. In some test | |
688 | * cases, the tree can grow quite large (~200k objects). | |
8da6d581 | 689 | */ |
ecf160b4 | 690 | while ((rnode = rb_first_cached(&preftrees->indirect.root))) { |
86d5f994 EN |
691 | struct prelim_ref *ref; |
692 | ||
693 | ref = rb_entry(rnode, struct prelim_ref, rbnode); | |
694 | if (WARN(ref->parent, | |
695 | "BUG: direct ref found in indirect tree")) { | |
696 | ret = -EINVAL; | |
697 | goto out; | |
698 | } | |
699 | ||
ecf160b4 | 700 | rb_erase_cached(&ref->rbnode, &preftrees->indirect.root); |
6c336b21 | 701 | preftrees->indirect.count--; |
86d5f994 EN |
702 | |
703 | if (ref->count == 0) { | |
704 | free_pref(ref); | |
8da6d581 | 705 | continue; |
86d5f994 EN |
706 | } |
707 | ||
3ec4d323 EN |
708 | if (sc && sc->root_objectid && |
709 | ref->root_id != sc->root_objectid) { | |
86d5f994 | 710 | free_pref(ref); |
dc046b10 JB |
711 | ret = BACKREF_FOUND_SHARED; |
712 | goto out; | |
713 | } | |
ed58f2e6 | 714 | err = resolve_indirect_ref(fs_info, path, time_seq, preftrees, |
715 | ref, parents, extent_item_pos, | |
b25b0b87 | 716 | ignore_offset); |
95def2ed WS |
717 | /* |
718 | * we can only tolerate ENOENT,otherwise,we should catch error | |
719 | * and return directly. | |
720 | */ | |
721 | if (err == -ENOENT) { | |
3ec4d323 EN |
722 | prelim_ref_insert(fs_info, &preftrees->direct, ref, |
723 | NULL); | |
8da6d581 | 724 | continue; |
95def2ed | 725 | } else if (err) { |
86d5f994 | 726 | free_pref(ref); |
95def2ed WS |
727 | ret = err; |
728 | goto out; | |
729 | } | |
8da6d581 JS |
730 | |
731 | /* we put the first parent into the ref at hand */ | |
cd1b413c JS |
732 | ULIST_ITER_INIT(&uiter); |
733 | node = ulist_next(parents, &uiter); | |
8da6d581 | 734 | ref->parent = node ? node->val : 0; |
4dae077a | 735 | ref->inode_list = unode_aux_to_inode_list(node); |
8da6d581 | 736 | |
86d5f994 | 737 | /* Add a prelim_ref(s) for any other parent(s). */ |
cd1b413c | 738 | while ((node = ulist_next(parents, &uiter))) { |
86d5f994 EN |
739 | struct prelim_ref *new_ref; |
740 | ||
b9e9a6cb WS |
741 | new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, |
742 | GFP_NOFS); | |
8da6d581 | 743 | if (!new_ref) { |
86d5f994 | 744 | free_pref(ref); |
8da6d581 | 745 | ret = -ENOMEM; |
e36902d4 | 746 | goto out; |
8da6d581 JS |
747 | } |
748 | memcpy(new_ref, ref, sizeof(*ref)); | |
749 | new_ref->parent = node->val; | |
4dae077a | 750 | new_ref->inode_list = unode_aux_to_inode_list(node); |
3ec4d323 EN |
751 | prelim_ref_insert(fs_info, &preftrees->direct, |
752 | new_ref, NULL); | |
8da6d581 | 753 | } |
86d5f994 | 754 | |
3ec4d323 | 755 | /* |
52042d8e | 756 | * Now it's a direct ref, put it in the direct tree. We must |
3ec4d323 EN |
757 | * do this last because the ref could be merged/freed here. |
758 | */ | |
759 | prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL); | |
86d5f994 | 760 | |
8da6d581 | 761 | ulist_reinit(parents); |
9dd14fd6 | 762 | cond_resched(); |
8da6d581 | 763 | } |
e36902d4 | 764 | out: |
8da6d581 JS |
765 | ulist_free(parents); |
766 | return ret; | |
767 | } | |
768 | ||
d5c88b73 JS |
769 | /* |
770 | * read tree blocks and add keys where required. | |
771 | */ | |
e0c476b1 | 772 | static int add_missing_keys(struct btrfs_fs_info *fs_info, |
38e3eebf | 773 | struct preftrees *preftrees, bool lock) |
d5c88b73 | 774 | { |
e0c476b1 | 775 | struct prelim_ref *ref; |
d5c88b73 | 776 | struct extent_buffer *eb; |
86d5f994 EN |
777 | struct preftree *tree = &preftrees->indirect_missing_keys; |
778 | struct rb_node *node; | |
d5c88b73 | 779 | |
ecf160b4 | 780 | while ((node = rb_first_cached(&tree->root))) { |
86d5f994 | 781 | ref = rb_entry(node, struct prelim_ref, rbnode); |
ecf160b4 | 782 | rb_erase_cached(node, &tree->root); |
86d5f994 EN |
783 | |
784 | BUG_ON(ref->parent); /* should not be a direct ref */ | |
785 | BUG_ON(ref->key_for_search.type); | |
d5c88b73 | 786 | BUG_ON(!ref->wanted_disk_byte); |
86d5f994 | 787 | |
1b7ec85e JB |
788 | eb = read_tree_block(fs_info, ref->wanted_disk_byte, |
789 | ref->root_id, 0, ref->level - 1, NULL); | |
64c043de | 790 | if (IS_ERR(eb)) { |
86d5f994 | 791 | free_pref(ref); |
64c043de | 792 | return PTR_ERR(eb); |
4eb150d6 QW |
793 | } |
794 | if (!extent_buffer_uptodate(eb)) { | |
86d5f994 | 795 | free_pref(ref); |
416bc658 JB |
796 | free_extent_buffer(eb); |
797 | return -EIO; | |
798 | } | |
4eb150d6 | 799 | |
38e3eebf JB |
800 | if (lock) |
801 | btrfs_tree_read_lock(eb); | |
d5c88b73 JS |
802 | if (btrfs_header_level(eb) == 0) |
803 | btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); | |
804 | else | |
805 | btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); | |
38e3eebf JB |
806 | if (lock) |
807 | btrfs_tree_read_unlock(eb); | |
d5c88b73 | 808 | free_extent_buffer(eb); |
3ec4d323 | 809 | prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL); |
9dd14fd6 | 810 | cond_resched(); |
d5c88b73 JS |
811 | } |
812 | return 0; | |
813 | } | |
814 | ||
8da6d581 JS |
815 | /* |
816 | * add all currently queued delayed refs from this head whose seq nr is | |
817 | * smaller or equal that seq to the list | |
818 | */ | |
00142756 JM |
819 | static int add_delayed_refs(const struct btrfs_fs_info *fs_info, |
820 | struct btrfs_delayed_ref_head *head, u64 seq, | |
b25b0b87 | 821 | struct preftrees *preftrees, struct share_check *sc) |
8da6d581 | 822 | { |
c6fc2454 | 823 | struct btrfs_delayed_ref_node *node; |
8da6d581 | 824 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; |
d5c88b73 | 825 | struct btrfs_key key; |
86d5f994 | 826 | struct btrfs_key tmp_op_key; |
0e0adbcf | 827 | struct rb_node *n; |
01747e92 | 828 | int count; |
b1375d64 | 829 | int ret = 0; |
8da6d581 | 830 | |
a6dbceaf | 831 | if (extent_op && extent_op->update_key) |
86d5f994 | 832 | btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key); |
8da6d581 | 833 | |
d7df2c79 | 834 | spin_lock(&head->lock); |
e3d03965 | 835 | for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) { |
0e0adbcf JB |
836 | node = rb_entry(n, struct btrfs_delayed_ref_node, |
837 | ref_node); | |
8da6d581 JS |
838 | if (node->seq > seq) |
839 | continue; | |
840 | ||
841 | switch (node->action) { | |
842 | case BTRFS_ADD_DELAYED_EXTENT: | |
843 | case BTRFS_UPDATE_DELAYED_HEAD: | |
844 | WARN_ON(1); | |
845 | continue; | |
846 | case BTRFS_ADD_DELAYED_REF: | |
01747e92 | 847 | count = node->ref_mod; |
8da6d581 JS |
848 | break; |
849 | case BTRFS_DROP_DELAYED_REF: | |
01747e92 | 850 | count = node->ref_mod * -1; |
8da6d581 JS |
851 | break; |
852 | default: | |
290342f6 | 853 | BUG(); |
8da6d581 JS |
854 | } |
855 | switch (node->type) { | |
856 | case BTRFS_TREE_BLOCK_REF_KEY: { | |
86d5f994 | 857 | /* NORMAL INDIRECT METADATA backref */ |
8da6d581 JS |
858 | struct btrfs_delayed_tree_ref *ref; |
859 | ||
860 | ref = btrfs_delayed_node_to_tree_ref(node); | |
00142756 JM |
861 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
862 | &tmp_op_key, ref->level + 1, | |
01747e92 EN |
863 | node->bytenr, count, sc, |
864 | GFP_ATOMIC); | |
8da6d581 JS |
865 | break; |
866 | } | |
867 | case BTRFS_SHARED_BLOCK_REF_KEY: { | |
86d5f994 | 868 | /* SHARED DIRECT METADATA backref */ |
8da6d581 JS |
869 | struct btrfs_delayed_tree_ref *ref; |
870 | ||
871 | ref = btrfs_delayed_node_to_tree_ref(node); | |
86d5f994 | 872 | |
01747e92 EN |
873 | ret = add_direct_ref(fs_info, preftrees, ref->level + 1, |
874 | ref->parent, node->bytenr, count, | |
3ec4d323 | 875 | sc, GFP_ATOMIC); |
8da6d581 JS |
876 | break; |
877 | } | |
878 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 879 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 | 880 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
881 | ref = btrfs_delayed_node_to_data_ref(node); |
882 | ||
883 | key.objectid = ref->objectid; | |
884 | key.type = BTRFS_EXTENT_DATA_KEY; | |
885 | key.offset = ref->offset; | |
dc046b10 JB |
886 | |
887 | /* | |
4fc7b572 FM |
888 | * If we have a share check context and a reference for |
889 | * another inode, we can't exit immediately. This is | |
890 | * because even if this is a BTRFS_ADD_DELAYED_REF | |
891 | * reference we may find next a BTRFS_DROP_DELAYED_REF | |
892 | * which cancels out this ADD reference. | |
893 | * | |
894 | * If this is a DROP reference and there was no previous | |
895 | * ADD reference, then we need to signal that when we | |
896 | * process references from the extent tree (through | |
897 | * add_inline_refs() and add_keyed_refs()), we should | |
898 | * not exit early if we find a reference for another | |
899 | * inode, because one of the delayed DROP references | |
900 | * may cancel that reference in the extent tree. | |
dc046b10 | 901 | */ |
4fc7b572 FM |
902 | if (sc && count < 0) |
903 | sc->have_delayed_delete_refs = true; | |
dc046b10 | 904 | |
00142756 | 905 | ret = add_indirect_ref(fs_info, preftrees, ref->root, |
01747e92 EN |
906 | &key, 0, node->bytenr, count, sc, |
907 | GFP_ATOMIC); | |
8da6d581 JS |
908 | break; |
909 | } | |
910 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 911 | /* SHARED DIRECT FULL backref */ |
8da6d581 | 912 | struct btrfs_delayed_data_ref *ref; |
8da6d581 JS |
913 | |
914 | ref = btrfs_delayed_node_to_data_ref(node); | |
86d5f994 | 915 | |
01747e92 EN |
916 | ret = add_direct_ref(fs_info, preftrees, 0, ref->parent, |
917 | node->bytenr, count, sc, | |
918 | GFP_ATOMIC); | |
8da6d581 JS |
919 | break; |
920 | } | |
921 | default: | |
922 | WARN_ON(1); | |
923 | } | |
3ec4d323 EN |
924 | /* |
925 | * We must ignore BACKREF_FOUND_SHARED until all delayed | |
926 | * refs have been checked. | |
927 | */ | |
928 | if (ret && (ret != BACKREF_FOUND_SHARED)) | |
d7df2c79 | 929 | break; |
8da6d581 | 930 | } |
3ec4d323 EN |
931 | if (!ret) |
932 | ret = extent_is_shared(sc); | |
4fc7b572 | 933 | |
d7df2c79 JB |
934 | spin_unlock(&head->lock); |
935 | return ret; | |
8da6d581 JS |
936 | } |
937 | ||
938 | /* | |
939 | * add all inline backrefs for bytenr to the list | |
3ec4d323 EN |
940 | * |
941 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 942 | */ |
00142756 JM |
943 | static int add_inline_refs(const struct btrfs_fs_info *fs_info, |
944 | struct btrfs_path *path, u64 bytenr, | |
86d5f994 | 945 | int *info_level, struct preftrees *preftrees, |
b25b0b87 | 946 | struct share_check *sc) |
8da6d581 | 947 | { |
b1375d64 | 948 | int ret = 0; |
8da6d581 JS |
949 | int slot; |
950 | struct extent_buffer *leaf; | |
951 | struct btrfs_key key; | |
261c84b6 | 952 | struct btrfs_key found_key; |
8da6d581 JS |
953 | unsigned long ptr; |
954 | unsigned long end; | |
955 | struct btrfs_extent_item *ei; | |
956 | u64 flags; | |
957 | u64 item_size; | |
958 | ||
959 | /* | |
960 | * enumerate all inline refs | |
961 | */ | |
962 | leaf = path->nodes[0]; | |
dadcaf78 | 963 | slot = path->slots[0]; |
8da6d581 | 964 | |
3212fa14 | 965 | item_size = btrfs_item_size(leaf, slot); |
8da6d581 JS |
966 | BUG_ON(item_size < sizeof(*ei)); |
967 | ||
968 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
969 | flags = btrfs_extent_flags(leaf, ei); | |
261c84b6 | 970 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
8da6d581 JS |
971 | |
972 | ptr = (unsigned long)(ei + 1); | |
973 | end = (unsigned long)ei + item_size; | |
974 | ||
261c84b6 JB |
975 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && |
976 | flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
8da6d581 | 977 | struct btrfs_tree_block_info *info; |
8da6d581 JS |
978 | |
979 | info = (struct btrfs_tree_block_info *)ptr; | |
980 | *info_level = btrfs_tree_block_level(leaf, info); | |
8da6d581 JS |
981 | ptr += sizeof(struct btrfs_tree_block_info); |
982 | BUG_ON(ptr > end); | |
261c84b6 JB |
983 | } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { |
984 | *info_level = found_key.offset; | |
8da6d581 JS |
985 | } else { |
986 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
987 | } | |
988 | ||
989 | while (ptr < end) { | |
990 | struct btrfs_extent_inline_ref *iref; | |
991 | u64 offset; | |
992 | int type; | |
993 | ||
994 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
995 | type = btrfs_get_extent_inline_ref_type(leaf, iref, |
996 | BTRFS_REF_TYPE_ANY); | |
997 | if (type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 998 | return -EUCLEAN; |
3de28d57 | 999 | |
8da6d581 JS |
1000 | offset = btrfs_extent_inline_ref_offset(leaf, iref); |
1001 | ||
1002 | switch (type) { | |
1003 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
00142756 JM |
1004 | ret = add_direct_ref(fs_info, preftrees, |
1005 | *info_level + 1, offset, | |
3ec4d323 | 1006 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1007 | break; |
1008 | case BTRFS_SHARED_DATA_REF_KEY: { | |
1009 | struct btrfs_shared_data_ref *sdref; | |
1010 | int count; | |
1011 | ||
1012 | sdref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1013 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
86d5f994 | 1014 | |
00142756 | 1015 | ret = add_direct_ref(fs_info, preftrees, 0, offset, |
3ec4d323 | 1016 | bytenr, count, sc, GFP_NOFS); |
8da6d581 JS |
1017 | break; |
1018 | } | |
1019 | case BTRFS_TREE_BLOCK_REF_KEY: | |
00142756 JM |
1020 | ret = add_indirect_ref(fs_info, preftrees, offset, |
1021 | NULL, *info_level + 1, | |
3ec4d323 | 1022 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1023 | break; |
1024 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
1025 | struct btrfs_extent_data_ref *dref; | |
1026 | int count; | |
1027 | u64 root; | |
1028 | ||
1029 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1030 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1031 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1032 | dref); | |
1033 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1034 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1035 | |
4fc7b572 FM |
1036 | if (sc && sc->inum && key.objectid != sc->inum && |
1037 | !sc->have_delayed_delete_refs) { | |
dc046b10 JB |
1038 | ret = BACKREF_FOUND_SHARED; |
1039 | break; | |
1040 | } | |
1041 | ||
8da6d581 | 1042 | root = btrfs_extent_data_ref_root(leaf, dref); |
86d5f994 | 1043 | |
00142756 JM |
1044 | ret = add_indirect_ref(fs_info, preftrees, root, |
1045 | &key, 0, bytenr, count, | |
3ec4d323 | 1046 | sc, GFP_NOFS); |
4fc7b572 | 1047 | |
8da6d581 JS |
1048 | break; |
1049 | } | |
1050 | default: | |
1051 | WARN_ON(1); | |
1052 | } | |
1149ab6b WS |
1053 | if (ret) |
1054 | return ret; | |
8da6d581 JS |
1055 | ptr += btrfs_extent_inline_ref_size(type); |
1056 | } | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * add all non-inline backrefs for bytenr to the list | |
3ec4d323 EN |
1063 | * |
1064 | * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. | |
8da6d581 | 1065 | */ |
98cc4222 | 1066 | static int add_keyed_refs(struct btrfs_root *extent_root, |
e0c476b1 | 1067 | struct btrfs_path *path, u64 bytenr, |
86d5f994 | 1068 | int info_level, struct preftrees *preftrees, |
3ec4d323 | 1069 | struct share_check *sc) |
8da6d581 | 1070 | { |
98cc4222 | 1071 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
8da6d581 JS |
1072 | int ret; |
1073 | int slot; | |
1074 | struct extent_buffer *leaf; | |
1075 | struct btrfs_key key; | |
1076 | ||
1077 | while (1) { | |
1078 | ret = btrfs_next_item(extent_root, path); | |
1079 | if (ret < 0) | |
1080 | break; | |
1081 | if (ret) { | |
1082 | ret = 0; | |
1083 | break; | |
1084 | } | |
1085 | ||
1086 | slot = path->slots[0]; | |
1087 | leaf = path->nodes[0]; | |
1088 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1089 | ||
1090 | if (key.objectid != bytenr) | |
1091 | break; | |
1092 | if (key.type < BTRFS_TREE_BLOCK_REF_KEY) | |
1093 | continue; | |
1094 | if (key.type > BTRFS_SHARED_DATA_REF_KEY) | |
1095 | break; | |
1096 | ||
1097 | switch (key.type) { | |
1098 | case BTRFS_SHARED_BLOCK_REF_KEY: | |
86d5f994 | 1099 | /* SHARED DIRECT METADATA backref */ |
00142756 JM |
1100 | ret = add_direct_ref(fs_info, preftrees, |
1101 | info_level + 1, key.offset, | |
3ec4d323 | 1102 | bytenr, 1, NULL, GFP_NOFS); |
8da6d581 JS |
1103 | break; |
1104 | case BTRFS_SHARED_DATA_REF_KEY: { | |
86d5f994 | 1105 | /* SHARED DIRECT FULL backref */ |
8da6d581 JS |
1106 | struct btrfs_shared_data_ref *sdref; |
1107 | int count; | |
1108 | ||
1109 | sdref = btrfs_item_ptr(leaf, slot, | |
1110 | struct btrfs_shared_data_ref); | |
1111 | count = btrfs_shared_data_ref_count(leaf, sdref); | |
00142756 JM |
1112 | ret = add_direct_ref(fs_info, preftrees, 0, |
1113 | key.offset, bytenr, count, | |
3ec4d323 | 1114 | sc, GFP_NOFS); |
8da6d581 JS |
1115 | break; |
1116 | } | |
1117 | case BTRFS_TREE_BLOCK_REF_KEY: | |
86d5f994 | 1118 | /* NORMAL INDIRECT METADATA backref */ |
00142756 JM |
1119 | ret = add_indirect_ref(fs_info, preftrees, key.offset, |
1120 | NULL, info_level + 1, bytenr, | |
3ec4d323 | 1121 | 1, NULL, GFP_NOFS); |
8da6d581 JS |
1122 | break; |
1123 | case BTRFS_EXTENT_DATA_REF_KEY: { | |
86d5f994 | 1124 | /* NORMAL INDIRECT DATA backref */ |
8da6d581 JS |
1125 | struct btrfs_extent_data_ref *dref; |
1126 | int count; | |
1127 | u64 root; | |
1128 | ||
1129 | dref = btrfs_item_ptr(leaf, slot, | |
1130 | struct btrfs_extent_data_ref); | |
1131 | count = btrfs_extent_data_ref_count(leaf, dref); | |
1132 | key.objectid = btrfs_extent_data_ref_objectid(leaf, | |
1133 | dref); | |
1134 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1135 | key.offset = btrfs_extent_data_ref_offset(leaf, dref); | |
dc046b10 | 1136 | |
4fc7b572 FM |
1137 | if (sc && sc->inum && key.objectid != sc->inum && |
1138 | !sc->have_delayed_delete_refs) { | |
dc046b10 JB |
1139 | ret = BACKREF_FOUND_SHARED; |
1140 | break; | |
1141 | } | |
1142 | ||
8da6d581 | 1143 | root = btrfs_extent_data_ref_root(leaf, dref); |
00142756 JM |
1144 | ret = add_indirect_ref(fs_info, preftrees, root, |
1145 | &key, 0, bytenr, count, | |
3ec4d323 | 1146 | sc, GFP_NOFS); |
8da6d581 JS |
1147 | break; |
1148 | } | |
1149 | default: | |
1150 | WARN_ON(1); | |
1151 | } | |
1149ab6b WS |
1152 | if (ret) |
1153 | return ret; | |
1154 | ||
8da6d581 JS |
1155 | } |
1156 | ||
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * this adds all existing backrefs (inline backrefs, backrefs and delayed | |
1162 | * refs) for the given bytenr to the refs list, merges duplicates and resolves | |
1163 | * indirect refs to their parent bytenr. | |
1164 | * When roots are found, they're added to the roots list | |
1165 | * | |
f3a84ccd FM |
1166 | * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and |
1167 | * behave much like trans == NULL case, the difference only lies in it will not | |
21633fc6 QW |
1168 | * commit root. |
1169 | * The special case is for qgroup to search roots in commit_transaction(). | |
1170 | * | |
3ec4d323 EN |
1171 | * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a |
1172 | * shared extent is detected. | |
1173 | * | |
1174 | * Otherwise this returns 0 for success and <0 for an error. | |
1175 | * | |
c995ab3c ZB |
1176 | * If ignore_offset is set to false, only extent refs whose offsets match |
1177 | * extent_item_pos are returned. If true, every extent ref is returned | |
1178 | * and extent_item_pos is ignored. | |
1179 | * | |
8da6d581 JS |
1180 | * FIXME some caching might speed things up |
1181 | */ | |
1182 | static int find_parent_nodes(struct btrfs_trans_handle *trans, | |
1183 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
097b8a7c | 1184 | u64 time_seq, struct ulist *refs, |
dc046b10 | 1185 | struct ulist *roots, const u64 *extent_item_pos, |
c995ab3c | 1186 | struct share_check *sc, bool ignore_offset) |
8da6d581 | 1187 | { |
29cbcf40 | 1188 | struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr); |
8da6d581 JS |
1189 | struct btrfs_key key; |
1190 | struct btrfs_path *path; | |
8da6d581 | 1191 | struct btrfs_delayed_ref_root *delayed_refs = NULL; |
d3b01064 | 1192 | struct btrfs_delayed_ref_head *head; |
8da6d581 JS |
1193 | int info_level = 0; |
1194 | int ret; | |
e0c476b1 | 1195 | struct prelim_ref *ref; |
86d5f994 | 1196 | struct rb_node *node; |
f05c4746 | 1197 | struct extent_inode_elem *eie = NULL; |
86d5f994 EN |
1198 | struct preftrees preftrees = { |
1199 | .direct = PREFTREE_INIT, | |
1200 | .indirect = PREFTREE_INIT, | |
1201 | .indirect_missing_keys = PREFTREE_INIT | |
1202 | }; | |
8da6d581 JS |
1203 | |
1204 | key.objectid = bytenr; | |
8da6d581 | 1205 | key.offset = (u64)-1; |
261c84b6 JB |
1206 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1207 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1208 | else | |
1209 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
8da6d581 JS |
1210 | |
1211 | path = btrfs_alloc_path(); | |
1212 | if (!path) | |
1213 | return -ENOMEM; | |
e84752d4 | 1214 | if (!trans) { |
da61d31a | 1215 | path->search_commit_root = 1; |
e84752d4 WS |
1216 | path->skip_locking = 1; |
1217 | } | |
8da6d581 | 1218 | |
f3a84ccd | 1219 | if (time_seq == BTRFS_SEQ_LAST) |
21633fc6 QW |
1220 | path->skip_locking = 1; |
1221 | ||
8da6d581 | 1222 | again: |
d3b01064 LZ |
1223 | head = NULL; |
1224 | ||
98cc4222 | 1225 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8da6d581 JS |
1226 | if (ret < 0) |
1227 | goto out; | |
fcba0120 JB |
1228 | if (ret == 0) { |
1229 | /* This shouldn't happen, indicates a bug or fs corruption. */ | |
1230 | ASSERT(ret != 0); | |
1231 | ret = -EUCLEAN; | |
1232 | goto out; | |
1233 | } | |
8da6d581 | 1234 | |
21633fc6 | 1235 | if (trans && likely(trans->type != __TRANS_DUMMY) && |
f3a84ccd | 1236 | time_seq != BTRFS_SEQ_LAST) { |
7a3ae2f8 | 1237 | /* |
9665ebd5 JB |
1238 | * We have a specific time_seq we care about and trans which |
1239 | * means we have the path lock, we need to grab the ref head and | |
1240 | * lock it so we have a consistent view of the refs at the given | |
1241 | * time. | |
7a3ae2f8 JS |
1242 | */ |
1243 | delayed_refs = &trans->transaction->delayed_refs; | |
1244 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 1245 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
7a3ae2f8 JS |
1246 | if (head) { |
1247 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 1248 | refcount_inc(&head->refs); |
7a3ae2f8 JS |
1249 | spin_unlock(&delayed_refs->lock); |
1250 | ||
1251 | btrfs_release_path(path); | |
1252 | ||
1253 | /* | |
1254 | * Mutex was contended, block until it's | |
1255 | * released and try again | |
1256 | */ | |
1257 | mutex_lock(&head->mutex); | |
1258 | mutex_unlock(&head->mutex); | |
d278850e | 1259 | btrfs_put_delayed_ref_head(head); |
7a3ae2f8 JS |
1260 | goto again; |
1261 | } | |
d7df2c79 | 1262 | spin_unlock(&delayed_refs->lock); |
00142756 | 1263 | ret = add_delayed_refs(fs_info, head, time_seq, |
b25b0b87 | 1264 | &preftrees, sc); |
155725c9 | 1265 | mutex_unlock(&head->mutex); |
d7df2c79 | 1266 | if (ret) |
7a3ae2f8 | 1267 | goto out; |
d7df2c79 JB |
1268 | } else { |
1269 | spin_unlock(&delayed_refs->lock); | |
d3b01064 | 1270 | } |
8da6d581 | 1271 | } |
8da6d581 JS |
1272 | |
1273 | if (path->slots[0]) { | |
1274 | struct extent_buffer *leaf; | |
1275 | int slot; | |
1276 | ||
dadcaf78 | 1277 | path->slots[0]--; |
8da6d581 | 1278 | leaf = path->nodes[0]; |
dadcaf78 | 1279 | slot = path->slots[0]; |
8da6d581 JS |
1280 | btrfs_item_key_to_cpu(leaf, &key, slot); |
1281 | if (key.objectid == bytenr && | |
261c84b6 JB |
1282 | (key.type == BTRFS_EXTENT_ITEM_KEY || |
1283 | key.type == BTRFS_METADATA_ITEM_KEY)) { | |
00142756 | 1284 | ret = add_inline_refs(fs_info, path, bytenr, |
b25b0b87 | 1285 | &info_level, &preftrees, sc); |
8da6d581 JS |
1286 | if (ret) |
1287 | goto out; | |
98cc4222 | 1288 | ret = add_keyed_refs(root, path, bytenr, info_level, |
3ec4d323 | 1289 | &preftrees, sc); |
8da6d581 JS |
1290 | if (ret) |
1291 | goto out; | |
1292 | } | |
1293 | } | |
8da6d581 | 1294 | |
86d5f994 | 1295 | btrfs_release_path(path); |
8da6d581 | 1296 | |
38e3eebf | 1297 | ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0); |
d5c88b73 JS |
1298 | if (ret) |
1299 | goto out; | |
1300 | ||
ecf160b4 | 1301 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root)); |
8da6d581 | 1302 | |
86d5f994 | 1303 | ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees, |
b25b0b87 | 1304 | extent_item_pos, sc, ignore_offset); |
8da6d581 JS |
1305 | if (ret) |
1306 | goto out; | |
1307 | ||
ecf160b4 | 1308 | WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root)); |
8da6d581 | 1309 | |
86d5f994 EN |
1310 | /* |
1311 | * This walks the tree of merged and resolved refs. Tree blocks are | |
1312 | * read in as needed. Unique entries are added to the ulist, and | |
1313 | * the list of found roots is updated. | |
1314 | * | |
1315 | * We release the entire tree in one go before returning. | |
1316 | */ | |
ecf160b4 | 1317 | node = rb_first_cached(&preftrees.direct.root); |
86d5f994 EN |
1318 | while (node) { |
1319 | ref = rb_entry(node, struct prelim_ref, rbnode); | |
1320 | node = rb_next(&ref->rbnode); | |
c8195a7b ZB |
1321 | /* |
1322 | * ref->count < 0 can happen here if there are delayed | |
1323 | * refs with a node->action of BTRFS_DROP_DELAYED_REF. | |
1324 | * prelim_ref_insert() relies on this when merging | |
1325 | * identical refs to keep the overall count correct. | |
1326 | * prelim_ref_insert() will merge only those refs | |
1327 | * which compare identically. Any refs having | |
1328 | * e.g. different offsets would not be merged, | |
1329 | * and would retain their original ref->count < 0. | |
1330 | */ | |
98cfee21 | 1331 | if (roots && ref->count && ref->root_id && ref->parent == 0) { |
3ec4d323 EN |
1332 | if (sc && sc->root_objectid && |
1333 | ref->root_id != sc->root_objectid) { | |
dc046b10 JB |
1334 | ret = BACKREF_FOUND_SHARED; |
1335 | goto out; | |
1336 | } | |
1337 | ||
8da6d581 JS |
1338 | /* no parent == root of tree */ |
1339 | ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); | |
f1723939 WS |
1340 | if (ret < 0) |
1341 | goto out; | |
8da6d581 JS |
1342 | } |
1343 | if (ref->count && ref->parent) { | |
8a56457f JB |
1344 | if (extent_item_pos && !ref->inode_list && |
1345 | ref->level == 0) { | |
976b1908 | 1346 | struct extent_buffer *eb; |
707e8a07 | 1347 | |
581c1760 | 1348 | eb = read_tree_block(fs_info, ref->parent, 0, |
1b7ec85e | 1349 | 0, ref->level, NULL); |
64c043de LB |
1350 | if (IS_ERR(eb)) { |
1351 | ret = PTR_ERR(eb); | |
1352 | goto out; | |
4eb150d6 QW |
1353 | } |
1354 | if (!extent_buffer_uptodate(eb)) { | |
416bc658 | 1355 | free_extent_buffer(eb); |
c16c2e2e WS |
1356 | ret = -EIO; |
1357 | goto out; | |
416bc658 | 1358 | } |
38e3eebf | 1359 | |
ac5887c8 | 1360 | if (!path->skip_locking) |
38e3eebf | 1361 | btrfs_tree_read_lock(eb); |
976b1908 | 1362 | ret = find_extent_in_eb(eb, bytenr, |
c995ab3c | 1363 | *extent_item_pos, &eie, ignore_offset); |
38e3eebf | 1364 | if (!path->skip_locking) |
ac5887c8 | 1365 | btrfs_tree_read_unlock(eb); |
976b1908 | 1366 | free_extent_buffer(eb); |
f5929cd8 FDBM |
1367 | if (ret < 0) |
1368 | goto out; | |
1369 | ref->inode_list = eie; | |
976b1908 | 1370 | } |
4eb1f66d TI |
1371 | ret = ulist_add_merge_ptr(refs, ref->parent, |
1372 | ref->inode_list, | |
1373 | (void **)&eie, GFP_NOFS); | |
f1723939 WS |
1374 | if (ret < 0) |
1375 | goto out; | |
3301958b JS |
1376 | if (!ret && extent_item_pos) { |
1377 | /* | |
9f05c09d JB |
1378 | * We've recorded that parent, so we must extend |
1379 | * its inode list here. | |
1380 | * | |
1381 | * However if there was corruption we may not | |
1382 | * have found an eie, return an error in this | |
1383 | * case. | |
3301958b | 1384 | */ |
9f05c09d JB |
1385 | ASSERT(eie); |
1386 | if (!eie) { | |
1387 | ret = -EUCLEAN; | |
1388 | goto out; | |
1389 | } | |
3301958b JS |
1390 | while (eie->next) |
1391 | eie = eie->next; | |
1392 | eie->next = ref->inode_list; | |
1393 | } | |
f05c4746 | 1394 | eie = NULL; |
8da6d581 | 1395 | } |
9dd14fd6 | 1396 | cond_resched(); |
8da6d581 JS |
1397 | } |
1398 | ||
1399 | out: | |
8da6d581 | 1400 | btrfs_free_path(path); |
86d5f994 EN |
1401 | |
1402 | prelim_release(&preftrees.direct); | |
1403 | prelim_release(&preftrees.indirect); | |
1404 | prelim_release(&preftrees.indirect_missing_keys); | |
1405 | ||
f05c4746 WS |
1406 | if (ret < 0) |
1407 | free_inode_elem_list(eie); | |
8da6d581 JS |
1408 | return ret; |
1409 | } | |
1410 | ||
976b1908 JS |
1411 | static void free_leaf_list(struct ulist *blocks) |
1412 | { | |
1413 | struct ulist_node *node = NULL; | |
1414 | struct extent_inode_elem *eie; | |
976b1908 JS |
1415 | struct ulist_iterator uiter; |
1416 | ||
1417 | ULIST_ITER_INIT(&uiter); | |
1418 | while ((node = ulist_next(blocks, &uiter))) { | |
1419 | if (!node->aux) | |
1420 | continue; | |
4dae077a | 1421 | eie = unode_aux_to_inode_list(node); |
f05c4746 | 1422 | free_inode_elem_list(eie); |
976b1908 JS |
1423 | node->aux = 0; |
1424 | } | |
1425 | ||
1426 | ulist_free(blocks); | |
1427 | } | |
1428 | ||
8da6d581 JS |
1429 | /* |
1430 | * Finds all leafs with a reference to the specified combination of bytenr and | |
1431 | * offset. key_list_head will point to a list of corresponding keys (caller must | |
1432 | * free each list element). The leafs will be stored in the leafs ulist, which | |
1433 | * must be freed with ulist_free. | |
1434 | * | |
1435 | * returns 0 on success, <0 on error | |
1436 | */ | |
19b546d7 QW |
1437 | int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, |
1438 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
1439 | u64 time_seq, struct ulist **leafs, | |
1440 | const u64 *extent_item_pos, bool ignore_offset) | |
8da6d581 | 1441 | { |
8da6d581 JS |
1442 | int ret; |
1443 | ||
8da6d581 | 1444 | *leafs = ulist_alloc(GFP_NOFS); |
98cfee21 | 1445 | if (!*leafs) |
8da6d581 | 1446 | return -ENOMEM; |
8da6d581 | 1447 | |
afce772e | 1448 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1449 | *leafs, NULL, extent_item_pos, NULL, ignore_offset); |
8da6d581 | 1450 | if (ret < 0 && ret != -ENOENT) { |
976b1908 | 1451 | free_leaf_list(*leafs); |
8da6d581 JS |
1452 | return ret; |
1453 | } | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | ||
1458 | /* | |
1459 | * walk all backrefs for a given extent to find all roots that reference this | |
1460 | * extent. Walking a backref means finding all extents that reference this | |
1461 | * extent and in turn walk the backrefs of those, too. Naturally this is a | |
1462 | * recursive process, but here it is implemented in an iterative fashion: We | |
1463 | * find all referencing extents for the extent in question and put them on a | |
1464 | * list. In turn, we find all referencing extents for those, further appending | |
1465 | * to the list. The way we iterate the list allows adding more elements after | |
1466 | * the current while iterating. The process stops when we reach the end of the | |
1467 | * list. Found roots are added to the roots list. | |
1468 | * | |
1469 | * returns 0 on success, < 0 on error. | |
1470 | */ | |
e0c476b1 JM |
1471 | static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans, |
1472 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c ZB |
1473 | u64 time_seq, struct ulist **roots, |
1474 | bool ignore_offset) | |
8da6d581 JS |
1475 | { |
1476 | struct ulist *tmp; | |
1477 | struct ulist_node *node = NULL; | |
cd1b413c | 1478 | struct ulist_iterator uiter; |
8da6d581 JS |
1479 | int ret; |
1480 | ||
1481 | tmp = ulist_alloc(GFP_NOFS); | |
1482 | if (!tmp) | |
1483 | return -ENOMEM; | |
1484 | *roots = ulist_alloc(GFP_NOFS); | |
1485 | if (!*roots) { | |
1486 | ulist_free(tmp); | |
1487 | return -ENOMEM; | |
1488 | } | |
1489 | ||
cd1b413c | 1490 | ULIST_ITER_INIT(&uiter); |
8da6d581 | 1491 | while (1) { |
afce772e | 1492 | ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, |
c995ab3c | 1493 | tmp, *roots, NULL, NULL, ignore_offset); |
8da6d581 JS |
1494 | if (ret < 0 && ret != -ENOENT) { |
1495 | ulist_free(tmp); | |
1496 | ulist_free(*roots); | |
580c079b | 1497 | *roots = NULL; |
8da6d581 JS |
1498 | return ret; |
1499 | } | |
cd1b413c | 1500 | node = ulist_next(tmp, &uiter); |
8da6d581 JS |
1501 | if (!node) |
1502 | break; | |
1503 | bytenr = node->val; | |
bca1a290 | 1504 | cond_resched(); |
8da6d581 JS |
1505 | } |
1506 | ||
1507 | ulist_free(tmp); | |
1508 | return 0; | |
1509 | } | |
1510 | ||
9e351cc8 JB |
1511 | int btrfs_find_all_roots(struct btrfs_trans_handle *trans, |
1512 | struct btrfs_fs_info *fs_info, u64 bytenr, | |
c995ab3c | 1513 | u64 time_seq, struct ulist **roots, |
c7bcbb21 | 1514 | bool skip_commit_root_sem) |
9e351cc8 JB |
1515 | { |
1516 | int ret; | |
1517 | ||
8949b9a1 | 1518 | if (!trans && !skip_commit_root_sem) |
9e351cc8 | 1519 | down_read(&fs_info->commit_root_sem); |
e0c476b1 | 1520 | ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr, |
c7bcbb21 | 1521 | time_seq, roots, false); |
8949b9a1 | 1522 | if (!trans && !skip_commit_root_sem) |
9e351cc8 JB |
1523 | up_read(&fs_info->commit_root_sem); |
1524 | return ret; | |
1525 | } | |
1526 | ||
12a824dc FM |
1527 | /* |
1528 | * The caller has joined a transaction or is holding a read lock on the | |
1529 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1530 | * snapshot field changing while updating or checking the cache. | |
1531 | */ | |
1532 | static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache, | |
1533 | struct btrfs_root *root, | |
1534 | u64 bytenr, int level, bool *is_shared) | |
1535 | { | |
1536 | struct btrfs_backref_shared_cache_entry *entry; | |
1537 | ||
1538 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1539 | return false; | |
1540 | ||
1541 | /* | |
1542 | * Level -1 is used for the data extent, which is not reliable to cache | |
1543 | * because its reference count can increase or decrease without us | |
1544 | * realizing. We cache results only for extent buffers that lead from | |
1545 | * the root node down to the leaf with the file extent item. | |
1546 | */ | |
1547 | ASSERT(level >= 0); | |
1548 | ||
1549 | entry = &cache->entries[level]; | |
1550 | ||
1551 | /* Unused cache entry or being used for some other extent buffer. */ | |
1552 | if (entry->bytenr != bytenr) | |
1553 | return false; | |
1554 | ||
1555 | /* | |
1556 | * We cached a false result, but the last snapshot generation of the | |
1557 | * root changed, so we now have a snapshot. Don't trust the result. | |
1558 | */ | |
1559 | if (!entry->is_shared && | |
1560 | entry->gen != btrfs_root_last_snapshot(&root->root_item)) | |
1561 | return false; | |
1562 | ||
1563 | /* | |
1564 | * If we cached a true result and the last generation used for dropping | |
1565 | * a root changed, we can not trust the result, because the dropped root | |
1566 | * could be a snapshot sharing this extent buffer. | |
1567 | */ | |
1568 | if (entry->is_shared && | |
1569 | entry->gen != btrfs_get_last_root_drop_gen(root->fs_info)) | |
1570 | return false; | |
1571 | ||
1572 | *is_shared = entry->is_shared; | |
96dbcc00 FM |
1573 | /* |
1574 | * If the node at this level is shared, than all nodes below are also | |
1575 | * shared. Currently some of the nodes below may be marked as not shared | |
1576 | * because we have just switched from one leaf to another, and switched | |
1577 | * also other nodes above the leaf and below the current level, so mark | |
1578 | * them as shared. | |
1579 | */ | |
1580 | if (*is_shared) { | |
1581 | for (int i = 0; i < level; i++) { | |
1582 | cache->entries[i].is_shared = true; | |
1583 | cache->entries[i].gen = entry->gen; | |
1584 | } | |
1585 | } | |
12a824dc FM |
1586 | |
1587 | return true; | |
1588 | } | |
1589 | ||
1590 | /* | |
1591 | * The caller has joined a transaction or is holding a read lock on the | |
1592 | * fs_info->commit_root_sem semaphore, so no need to worry about the root's last | |
1593 | * snapshot field changing while updating or checking the cache. | |
1594 | */ | |
1595 | static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache, | |
1596 | struct btrfs_root *root, | |
1597 | u64 bytenr, int level, bool is_shared) | |
1598 | { | |
1599 | struct btrfs_backref_shared_cache_entry *entry; | |
1600 | u64 gen; | |
1601 | ||
1602 | if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) | |
1603 | return; | |
1604 | ||
1605 | /* | |
1606 | * Level -1 is used for the data extent, which is not reliable to cache | |
1607 | * because its reference count can increase or decrease without us | |
1608 | * realizing. We cache results only for extent buffers that lead from | |
1609 | * the root node down to the leaf with the file extent item. | |
1610 | */ | |
1611 | ASSERT(level >= 0); | |
1612 | ||
1613 | if (is_shared) | |
1614 | gen = btrfs_get_last_root_drop_gen(root->fs_info); | |
1615 | else | |
1616 | gen = btrfs_root_last_snapshot(&root->root_item); | |
1617 | ||
1618 | entry = &cache->entries[level]; | |
1619 | entry->bytenr = bytenr; | |
1620 | entry->is_shared = is_shared; | |
1621 | entry->gen = gen; | |
1622 | ||
1623 | /* | |
1624 | * If we found an extent buffer is shared, set the cache result for all | |
1625 | * extent buffers below it to true. As nodes in the path are COWed, | |
1626 | * their sharedness is moved to their children, and if a leaf is COWed, | |
1627 | * then the sharedness of a data extent becomes direct, the refcount of | |
1628 | * data extent is increased in the extent item at the extent tree. | |
1629 | */ | |
1630 | if (is_shared) { | |
1631 | for (int i = 0; i < level; i++) { | |
1632 | entry = &cache->entries[i]; | |
1633 | entry->is_shared = is_shared; | |
1634 | entry->gen = gen; | |
1635 | } | |
1636 | } | |
1637 | } | |
1638 | ||
8eedadda FM |
1639 | /* |
1640 | * Check if a data extent is shared or not. | |
6e353e3b | 1641 | * |
b8f164e3 FM |
1642 | * @root: The root the inode belongs to. |
1643 | * @inum: Number of the inode whose extent we are checking. | |
1644 | * @bytenr: Logical bytenr of the extent we are checking. | |
1645 | * @extent_gen: Generation of the extent (file extent item) or 0 if it is | |
1646 | * not known. | |
1647 | * @roots: List of roots this extent is shared among. | |
1648 | * @tmp: Temporary list used for iteration. | |
1649 | * @cache: A backref lookup result cache. | |
2c2ed5aa | 1650 | * |
8eedadda | 1651 | * btrfs_is_data_extent_shared uses the backref walking code but will short |
2c2ed5aa MF |
1652 | * circuit as soon as it finds a root or inode that doesn't match the |
1653 | * one passed in. This provides a significant performance benefit for | |
1654 | * callers (such as fiemap) which want to know whether the extent is | |
1655 | * shared but do not need a ref count. | |
1656 | * | |
03628cdb FM |
1657 | * This attempts to attach to the running transaction in order to account for |
1658 | * delayed refs, but continues on even when no running transaction exists. | |
bb739cf0 | 1659 | * |
2c2ed5aa MF |
1660 | * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. |
1661 | */ | |
8eedadda | 1662 | int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr, |
b8f164e3 | 1663 | u64 extent_gen, |
12a824dc FM |
1664 | struct ulist *roots, struct ulist *tmp, |
1665 | struct btrfs_backref_shared_cache *cache) | |
dc046b10 | 1666 | { |
bb739cf0 EN |
1667 | struct btrfs_fs_info *fs_info = root->fs_info; |
1668 | struct btrfs_trans_handle *trans; | |
dc046b10 JB |
1669 | struct ulist_iterator uiter; |
1670 | struct ulist_node *node; | |
f3a84ccd | 1671 | struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem); |
dc046b10 | 1672 | int ret = 0; |
3ec4d323 | 1673 | struct share_check shared = { |
4fd786e6 | 1674 | .root_objectid = root->root_key.objectid, |
3ec4d323 EN |
1675 | .inum = inum, |
1676 | .share_count = 0, | |
4fc7b572 | 1677 | .have_delayed_delete_refs = false, |
3ec4d323 | 1678 | }; |
12a824dc | 1679 | int level; |
dc046b10 | 1680 | |
5911c8fe DS |
1681 | ulist_init(roots); |
1682 | ulist_init(tmp); | |
dc046b10 | 1683 | |
a6d155d2 | 1684 | trans = btrfs_join_transaction_nostart(root); |
bb739cf0 | 1685 | if (IS_ERR(trans)) { |
03628cdb FM |
1686 | if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) { |
1687 | ret = PTR_ERR(trans); | |
1688 | goto out; | |
1689 | } | |
bb739cf0 | 1690 | trans = NULL; |
dc046b10 | 1691 | down_read(&fs_info->commit_root_sem); |
bb739cf0 EN |
1692 | } else { |
1693 | btrfs_get_tree_mod_seq(fs_info, &elem); | |
1694 | } | |
1695 | ||
12a824dc FM |
1696 | /* -1 means we are in the bytenr of the data extent. */ |
1697 | level = -1; | |
dc046b10 JB |
1698 | ULIST_ITER_INIT(&uiter); |
1699 | while (1) { | |
12a824dc FM |
1700 | bool is_shared; |
1701 | bool cached; | |
1702 | ||
dc046b10 | 1703 | ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp, |
c995ab3c | 1704 | roots, NULL, &shared, false); |
dc046b10 | 1705 | if (ret == BACKREF_FOUND_SHARED) { |
2c2ed5aa | 1706 | /* this is the only condition under which we return 1 */ |
dc046b10 | 1707 | ret = 1; |
12a824dc FM |
1708 | if (level >= 0) |
1709 | store_backref_shared_cache(cache, root, bytenr, | |
1710 | level, true); | |
dc046b10 JB |
1711 | break; |
1712 | } | |
1713 | if (ret < 0 && ret != -ENOENT) | |
1714 | break; | |
2c2ed5aa | 1715 | ret = 0; |
b8f164e3 FM |
1716 | /* |
1717 | * If our data extent is not shared through reflinks and it was | |
1718 | * created in a generation after the last one used to create a | |
1719 | * snapshot of the inode's root, then it can not be shared | |
1720 | * indirectly through subtrees, as that can only happen with | |
1721 | * snapshots. In this case bail out, no need to check for the | |
1722 | * sharedness of extent buffers. | |
1723 | */ | |
1724 | if (level == -1 && | |
1725 | extent_gen > btrfs_root_last_snapshot(&root->root_item)) | |
1726 | break; | |
1727 | ||
12a824dc FM |
1728 | if (level >= 0) |
1729 | store_backref_shared_cache(cache, root, bytenr, | |
1730 | level, false); | |
dc046b10 JB |
1731 | node = ulist_next(tmp, &uiter); |
1732 | if (!node) | |
1733 | break; | |
1734 | bytenr = node->val; | |
12a824dc FM |
1735 | level++; |
1736 | cached = lookup_backref_shared_cache(cache, root, bytenr, level, | |
1737 | &is_shared); | |
1738 | if (cached) { | |
1739 | ret = (is_shared ? 1 : 0); | |
1740 | break; | |
1741 | } | |
18bf591b | 1742 | shared.share_count = 0; |
4fc7b572 | 1743 | shared.have_delayed_delete_refs = false; |
dc046b10 JB |
1744 | cond_resched(); |
1745 | } | |
bb739cf0 EN |
1746 | |
1747 | if (trans) { | |
dc046b10 | 1748 | btrfs_put_tree_mod_seq(fs_info, &elem); |
bb739cf0 EN |
1749 | btrfs_end_transaction(trans); |
1750 | } else { | |
dc046b10 | 1751 | up_read(&fs_info->commit_root_sem); |
bb739cf0 | 1752 | } |
03628cdb | 1753 | out: |
5911c8fe DS |
1754 | ulist_release(roots); |
1755 | ulist_release(tmp); | |
dc046b10 JB |
1756 | return ret; |
1757 | } | |
1758 | ||
f186373f MF |
1759 | int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, |
1760 | u64 start_off, struct btrfs_path *path, | |
1761 | struct btrfs_inode_extref **ret_extref, | |
1762 | u64 *found_off) | |
1763 | { | |
1764 | int ret, slot; | |
1765 | struct btrfs_key key; | |
1766 | struct btrfs_key found_key; | |
1767 | struct btrfs_inode_extref *extref; | |
73980bec | 1768 | const struct extent_buffer *leaf; |
f186373f MF |
1769 | unsigned long ptr; |
1770 | ||
1771 | key.objectid = inode_objectid; | |
962a298f | 1772 | key.type = BTRFS_INODE_EXTREF_KEY; |
f186373f MF |
1773 | key.offset = start_off; |
1774 | ||
1775 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1776 | if (ret < 0) | |
1777 | return ret; | |
1778 | ||
1779 | while (1) { | |
1780 | leaf = path->nodes[0]; | |
1781 | slot = path->slots[0]; | |
1782 | if (slot >= btrfs_header_nritems(leaf)) { | |
1783 | /* | |
1784 | * If the item at offset is not found, | |
1785 | * btrfs_search_slot will point us to the slot | |
1786 | * where it should be inserted. In our case | |
1787 | * that will be the slot directly before the | |
1788 | * next INODE_REF_KEY_V2 item. In the case | |
1789 | * that we're pointing to the last slot in a | |
1790 | * leaf, we must move one leaf over. | |
1791 | */ | |
1792 | ret = btrfs_next_leaf(root, path); | |
1793 | if (ret) { | |
1794 | if (ret >= 1) | |
1795 | ret = -ENOENT; | |
1796 | break; | |
1797 | } | |
1798 | continue; | |
1799 | } | |
1800 | ||
1801 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1802 | ||
1803 | /* | |
1804 | * Check that we're still looking at an extended ref key for | |
1805 | * this particular objectid. If we have different | |
1806 | * objectid or type then there are no more to be found | |
1807 | * in the tree and we can exit. | |
1808 | */ | |
1809 | ret = -ENOENT; | |
1810 | if (found_key.objectid != inode_objectid) | |
1811 | break; | |
962a298f | 1812 | if (found_key.type != BTRFS_INODE_EXTREF_KEY) |
f186373f MF |
1813 | break; |
1814 | ||
1815 | ret = 0; | |
1816 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1817 | extref = (struct btrfs_inode_extref *)ptr; | |
1818 | *ret_extref = extref; | |
1819 | if (found_off) | |
1820 | *found_off = found_key.offset; | |
1821 | break; | |
1822 | } | |
1823 | ||
1824 | return ret; | |
1825 | } | |
1826 | ||
48a3b636 ES |
1827 | /* |
1828 | * this iterates to turn a name (from iref/extref) into a full filesystem path. | |
1829 | * Elements of the path are separated by '/' and the path is guaranteed to be | |
1830 | * 0-terminated. the path is only given within the current file system. | |
1831 | * Therefore, it never starts with a '/'. the caller is responsible to provide | |
1832 | * "size" bytes in "dest". the dest buffer will be filled backwards. finally, | |
1833 | * the start point of the resulting string is returned. this pointer is within | |
1834 | * dest, normally. | |
1835 | * in case the path buffer would overflow, the pointer is decremented further | |
1836 | * as if output was written to the buffer, though no more output is actually | |
1837 | * generated. that way, the caller can determine how much space would be | |
1838 | * required for the path to fit into the buffer. in that case, the returned | |
1839 | * value will be smaller than dest. callers must check this! | |
1840 | */ | |
96b5bd77 JS |
1841 | char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, |
1842 | u32 name_len, unsigned long name_off, | |
1843 | struct extent_buffer *eb_in, u64 parent, | |
1844 | char *dest, u32 size) | |
a542ad1b | 1845 | { |
a542ad1b JS |
1846 | int slot; |
1847 | u64 next_inum; | |
1848 | int ret; | |
661bec6b | 1849 | s64 bytes_left = ((s64)size) - 1; |
a542ad1b JS |
1850 | struct extent_buffer *eb = eb_in; |
1851 | struct btrfs_key found_key; | |
d24bec3a | 1852 | struct btrfs_inode_ref *iref; |
a542ad1b JS |
1853 | |
1854 | if (bytes_left >= 0) | |
1855 | dest[bytes_left] = '\0'; | |
1856 | ||
1857 | while (1) { | |
d24bec3a | 1858 | bytes_left -= name_len; |
a542ad1b JS |
1859 | if (bytes_left >= 0) |
1860 | read_extent_buffer(eb, dest + bytes_left, | |
d24bec3a | 1861 | name_off, name_len); |
b916a59a | 1862 | if (eb != eb_in) { |
0c0fe3b0 | 1863 | if (!path->skip_locking) |
ac5887c8 | 1864 | btrfs_tree_read_unlock(eb); |
a542ad1b | 1865 | free_extent_buffer(eb); |
b916a59a | 1866 | } |
c234a24d DS |
1867 | ret = btrfs_find_item(fs_root, path, parent, 0, |
1868 | BTRFS_INODE_REF_KEY, &found_key); | |
8f24b496 JS |
1869 | if (ret > 0) |
1870 | ret = -ENOENT; | |
a542ad1b JS |
1871 | if (ret) |
1872 | break; | |
d24bec3a | 1873 | |
a542ad1b JS |
1874 | next_inum = found_key.offset; |
1875 | ||
1876 | /* regular exit ahead */ | |
1877 | if (parent == next_inum) | |
1878 | break; | |
1879 | ||
1880 | slot = path->slots[0]; | |
1881 | eb = path->nodes[0]; | |
1882 | /* make sure we can use eb after releasing the path */ | |
b916a59a | 1883 | if (eb != eb_in) { |
0c0fe3b0 FM |
1884 | path->nodes[0] = NULL; |
1885 | path->locks[0] = 0; | |
b916a59a | 1886 | } |
a542ad1b | 1887 | btrfs_release_path(path); |
a542ad1b | 1888 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
d24bec3a MF |
1889 | |
1890 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1891 | name_off = (unsigned long)(iref + 1); | |
1892 | ||
a542ad1b JS |
1893 | parent = next_inum; |
1894 | --bytes_left; | |
1895 | if (bytes_left >= 0) | |
1896 | dest[bytes_left] = '/'; | |
1897 | } | |
1898 | ||
1899 | btrfs_release_path(path); | |
1900 | ||
1901 | if (ret) | |
1902 | return ERR_PTR(ret); | |
1903 | ||
1904 | return dest + bytes_left; | |
1905 | } | |
1906 | ||
1907 | /* | |
1908 | * this makes the path point to (logical EXTENT_ITEM *) | |
1909 | * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for | |
1910 | * tree blocks and <0 on error. | |
1911 | */ | |
1912 | int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, | |
69917e43 LB |
1913 | struct btrfs_path *path, struct btrfs_key *found_key, |
1914 | u64 *flags_ret) | |
a542ad1b | 1915 | { |
29cbcf40 | 1916 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical); |
a542ad1b JS |
1917 | int ret; |
1918 | u64 flags; | |
261c84b6 | 1919 | u64 size = 0; |
a542ad1b | 1920 | u32 item_size; |
73980bec | 1921 | const struct extent_buffer *eb; |
a542ad1b JS |
1922 | struct btrfs_extent_item *ei; |
1923 | struct btrfs_key key; | |
1924 | ||
261c84b6 JB |
1925 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1926 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1927 | else | |
1928 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a542ad1b JS |
1929 | key.objectid = logical; |
1930 | key.offset = (u64)-1; | |
1931 | ||
29cbcf40 | 1932 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a542ad1b JS |
1933 | if (ret < 0) |
1934 | return ret; | |
a542ad1b | 1935 | |
29cbcf40 | 1936 | ret = btrfs_previous_extent_item(extent_root, path, 0); |
850a8cdf WS |
1937 | if (ret) { |
1938 | if (ret > 0) | |
1939 | ret = -ENOENT; | |
1940 | return ret; | |
580f0a67 | 1941 | } |
850a8cdf | 1942 | btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); |
261c84b6 | 1943 | if (found_key->type == BTRFS_METADATA_ITEM_KEY) |
da17066c | 1944 | size = fs_info->nodesize; |
261c84b6 JB |
1945 | else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) |
1946 | size = found_key->offset; | |
1947 | ||
580f0a67 | 1948 | if (found_key->objectid > logical || |
261c84b6 | 1949 | found_key->objectid + size <= logical) { |
ab8d0fc4 JM |
1950 | btrfs_debug(fs_info, |
1951 | "logical %llu is not within any extent", logical); | |
a542ad1b | 1952 | return -ENOENT; |
4692cf58 | 1953 | } |
a542ad1b JS |
1954 | |
1955 | eb = path->nodes[0]; | |
3212fa14 | 1956 | item_size = btrfs_item_size(eb, path->slots[0]); |
a542ad1b JS |
1957 | BUG_ON(item_size < sizeof(*ei)); |
1958 | ||
1959 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
1960 | flags = btrfs_extent_flags(eb, ei); | |
1961 | ||
ab8d0fc4 JM |
1962 | btrfs_debug(fs_info, |
1963 | "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u", | |
c1c9ff7c GU |
1964 | logical, logical - found_key->objectid, found_key->objectid, |
1965 | found_key->offset, flags, item_size); | |
69917e43 LB |
1966 | |
1967 | WARN_ON(!flags_ret); | |
1968 | if (flags_ret) { | |
1969 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1970 | *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; | |
1971 | else if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1972 | *flags_ret = BTRFS_EXTENT_FLAG_DATA; | |
1973 | else | |
290342f6 | 1974 | BUG(); |
69917e43 LB |
1975 | return 0; |
1976 | } | |
a542ad1b JS |
1977 | |
1978 | return -EIO; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * helper function to iterate extent inline refs. ptr must point to a 0 value | |
1983 | * for the first call and may be modified. it is used to track state. | |
1984 | * if more refs exist, 0 is returned and the next call to | |
e0c476b1 | 1985 | * get_extent_inline_ref must pass the modified ptr parameter to get the |
a542ad1b JS |
1986 | * next ref. after the last ref was processed, 1 is returned. |
1987 | * returns <0 on error | |
1988 | */ | |
e0c476b1 JM |
1989 | static int get_extent_inline_ref(unsigned long *ptr, |
1990 | const struct extent_buffer *eb, | |
1991 | const struct btrfs_key *key, | |
1992 | const struct btrfs_extent_item *ei, | |
1993 | u32 item_size, | |
1994 | struct btrfs_extent_inline_ref **out_eiref, | |
1995 | int *out_type) | |
a542ad1b JS |
1996 | { |
1997 | unsigned long end; | |
1998 | u64 flags; | |
1999 | struct btrfs_tree_block_info *info; | |
2000 | ||
2001 | if (!*ptr) { | |
2002 | /* first call */ | |
2003 | flags = btrfs_extent_flags(eb, ei); | |
2004 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
6eda71d0 LB |
2005 | if (key->type == BTRFS_METADATA_ITEM_KEY) { |
2006 | /* a skinny metadata extent */ | |
2007 | *out_eiref = | |
2008 | (struct btrfs_extent_inline_ref *)(ei + 1); | |
2009 | } else { | |
2010 | WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY); | |
2011 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2012 | *out_eiref = | |
2013 | (struct btrfs_extent_inline_ref *)(info + 1); | |
2014 | } | |
a542ad1b JS |
2015 | } else { |
2016 | *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2017 | } | |
2018 | *ptr = (unsigned long)*out_eiref; | |
cd857dd6 | 2019 | if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) |
a542ad1b JS |
2020 | return -ENOENT; |
2021 | } | |
2022 | ||
2023 | end = (unsigned long)ei + item_size; | |
6eda71d0 | 2024 | *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); |
3de28d57 LB |
2025 | *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref, |
2026 | BTRFS_REF_TYPE_ANY); | |
2027 | if (*out_type == BTRFS_REF_TYPE_INVALID) | |
af431dcb | 2028 | return -EUCLEAN; |
a542ad1b JS |
2029 | |
2030 | *ptr += btrfs_extent_inline_ref_size(*out_type); | |
2031 | WARN_ON(*ptr > end); | |
2032 | if (*ptr == end) | |
2033 | return 1; /* last */ | |
2034 | ||
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * reads the tree block backref for an extent. tree level and root are returned | |
2040 | * through out_level and out_root. ptr must point to a 0 value for the first | |
e0c476b1 | 2041 | * call and may be modified (see get_extent_inline_ref comment). |
a542ad1b JS |
2042 | * returns 0 if data was provided, 1 if there was no more data to provide or |
2043 | * <0 on error. | |
2044 | */ | |
2045 | int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, | |
6eda71d0 LB |
2046 | struct btrfs_key *key, struct btrfs_extent_item *ei, |
2047 | u32 item_size, u64 *out_root, u8 *out_level) | |
a542ad1b JS |
2048 | { |
2049 | int ret; | |
2050 | int type; | |
a542ad1b JS |
2051 | struct btrfs_extent_inline_ref *eiref; |
2052 | ||
2053 | if (*ptr == (unsigned long)-1) | |
2054 | return 1; | |
2055 | ||
2056 | while (1) { | |
e0c476b1 | 2057 | ret = get_extent_inline_ref(ptr, eb, key, ei, item_size, |
6eda71d0 | 2058 | &eiref, &type); |
a542ad1b JS |
2059 | if (ret < 0) |
2060 | return ret; | |
2061 | ||
2062 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
2063 | type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2064 | break; | |
2065 | ||
2066 | if (ret == 1) | |
2067 | return 1; | |
2068 | } | |
2069 | ||
2070 | /* we can treat both ref types equally here */ | |
a542ad1b | 2071 | *out_root = btrfs_extent_inline_ref_offset(eb, eiref); |
a1317f45 FM |
2072 | |
2073 | if (key->type == BTRFS_EXTENT_ITEM_KEY) { | |
2074 | struct btrfs_tree_block_info *info; | |
2075 | ||
2076 | info = (struct btrfs_tree_block_info *)(ei + 1); | |
2077 | *out_level = btrfs_tree_block_level(eb, info); | |
2078 | } else { | |
2079 | ASSERT(key->type == BTRFS_METADATA_ITEM_KEY); | |
2080 | *out_level = (u8)key->offset; | |
2081 | } | |
a542ad1b JS |
2082 | |
2083 | if (ret == 1) | |
2084 | *ptr = (unsigned long)-1; | |
2085 | ||
2086 | return 0; | |
2087 | } | |
2088 | ||
ab8d0fc4 JM |
2089 | static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, |
2090 | struct extent_inode_elem *inode_list, | |
2091 | u64 root, u64 extent_item_objectid, | |
2092 | iterate_extent_inodes_t *iterate, void *ctx) | |
a542ad1b | 2093 | { |
976b1908 | 2094 | struct extent_inode_elem *eie; |
4692cf58 | 2095 | int ret = 0; |
4692cf58 | 2096 | |
976b1908 | 2097 | for (eie = inode_list; eie; eie = eie->next) { |
ab8d0fc4 JM |
2098 | btrfs_debug(fs_info, |
2099 | "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu", | |
2100 | extent_item_objectid, eie->inum, | |
2101 | eie->offset, root); | |
976b1908 | 2102 | ret = iterate(eie->inum, eie->offset, root, ctx); |
4692cf58 | 2103 | if (ret) { |
ab8d0fc4 JM |
2104 | btrfs_debug(fs_info, |
2105 | "stopping iteration for %llu due to ret=%d", | |
2106 | extent_item_objectid, ret); | |
4692cf58 JS |
2107 | break; |
2108 | } | |
a542ad1b JS |
2109 | } |
2110 | ||
a542ad1b JS |
2111 | return ret; |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * calls iterate() for every inode that references the extent identified by | |
4692cf58 | 2116 | * the given parameters. |
a542ad1b JS |
2117 | * when the iterator function returns a non-zero value, iteration stops. |
2118 | */ | |
2119 | int iterate_extent_inodes(struct btrfs_fs_info *fs_info, | |
4692cf58 | 2120 | u64 extent_item_objectid, u64 extent_item_pos, |
7a3ae2f8 | 2121 | int search_commit_root, |
c995ab3c ZB |
2122 | iterate_extent_inodes_t *iterate, void *ctx, |
2123 | bool ignore_offset) | |
a542ad1b | 2124 | { |
a542ad1b | 2125 | int ret; |
da61d31a | 2126 | struct btrfs_trans_handle *trans = NULL; |
7a3ae2f8 JS |
2127 | struct ulist *refs = NULL; |
2128 | struct ulist *roots = NULL; | |
4692cf58 JS |
2129 | struct ulist_node *ref_node = NULL; |
2130 | struct ulist_node *root_node = NULL; | |
f3a84ccd | 2131 | struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem); |
cd1b413c JS |
2132 | struct ulist_iterator ref_uiter; |
2133 | struct ulist_iterator root_uiter; | |
a542ad1b | 2134 | |
ab8d0fc4 | 2135 | btrfs_debug(fs_info, "resolving all inodes for extent %llu", |
4692cf58 | 2136 | extent_item_objectid); |
a542ad1b | 2137 | |
da61d31a | 2138 | if (!search_commit_root) { |
30a9da5d | 2139 | trans = btrfs_attach_transaction(fs_info->tree_root); |
bfc61c36 FM |
2140 | if (IS_ERR(trans)) { |
2141 | if (PTR_ERR(trans) != -ENOENT && | |
2142 | PTR_ERR(trans) != -EROFS) | |
2143 | return PTR_ERR(trans); | |
2144 | trans = NULL; | |
2145 | } | |
2146 | } | |
2147 | ||
2148 | if (trans) | |
f3a84ccd | 2149 | btrfs_get_tree_mod_seq(fs_info, &seq_elem); |
bfc61c36 | 2150 | else |
9e351cc8 | 2151 | down_read(&fs_info->commit_root_sem); |
a542ad1b | 2152 | |
4692cf58 | 2153 | ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, |
f3a84ccd | 2154 | seq_elem.seq, &refs, |
c995ab3c | 2155 | &extent_item_pos, ignore_offset); |
4692cf58 JS |
2156 | if (ret) |
2157 | goto out; | |
a542ad1b | 2158 | |
cd1b413c JS |
2159 | ULIST_ITER_INIT(&ref_uiter); |
2160 | while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { | |
e0c476b1 | 2161 | ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val, |
f3a84ccd | 2162 | seq_elem.seq, &roots, |
c995ab3c | 2163 | ignore_offset); |
4692cf58 JS |
2164 | if (ret) |
2165 | break; | |
cd1b413c JS |
2166 | ULIST_ITER_INIT(&root_uiter); |
2167 | while (!ret && (root_node = ulist_next(roots, &root_uiter))) { | |
ab8d0fc4 JM |
2168 | btrfs_debug(fs_info, |
2169 | "root %llu references leaf %llu, data list %#llx", | |
2170 | root_node->val, ref_node->val, | |
2171 | ref_node->aux); | |
2172 | ret = iterate_leaf_refs(fs_info, | |
2173 | (struct extent_inode_elem *) | |
995e01b7 JS |
2174 | (uintptr_t)ref_node->aux, |
2175 | root_node->val, | |
2176 | extent_item_objectid, | |
2177 | iterate, ctx); | |
4692cf58 | 2178 | } |
976b1908 | 2179 | ulist_free(roots); |
a542ad1b JS |
2180 | } |
2181 | ||
976b1908 | 2182 | free_leaf_list(refs); |
4692cf58 | 2183 | out: |
bfc61c36 | 2184 | if (trans) { |
f3a84ccd | 2185 | btrfs_put_tree_mod_seq(fs_info, &seq_elem); |
3a45bb20 | 2186 | btrfs_end_transaction(trans); |
9e351cc8 JB |
2187 | } else { |
2188 | up_read(&fs_info->commit_root_sem); | |
7a3ae2f8 JS |
2189 | } |
2190 | ||
a542ad1b JS |
2191 | return ret; |
2192 | } | |
2193 | ||
e3059ec0 DS |
2194 | static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) |
2195 | { | |
2196 | struct btrfs_data_container *inodes = ctx; | |
2197 | const size_t c = 3 * sizeof(u64); | |
2198 | ||
2199 | if (inodes->bytes_left >= c) { | |
2200 | inodes->bytes_left -= c; | |
2201 | inodes->val[inodes->elem_cnt] = inum; | |
2202 | inodes->val[inodes->elem_cnt + 1] = offset; | |
2203 | inodes->val[inodes->elem_cnt + 2] = root; | |
2204 | inodes->elem_cnt += 3; | |
2205 | } else { | |
2206 | inodes->bytes_missing += c - inodes->bytes_left; | |
2207 | inodes->bytes_left = 0; | |
2208 | inodes->elem_missed += 3; | |
2209 | } | |
2210 | ||
2211 | return 0; | |
2212 | } | |
2213 | ||
a542ad1b JS |
2214 | int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, |
2215 | struct btrfs_path *path, | |
e3059ec0 | 2216 | void *ctx, bool ignore_offset) |
a542ad1b JS |
2217 | { |
2218 | int ret; | |
4692cf58 | 2219 | u64 extent_item_pos; |
69917e43 | 2220 | u64 flags = 0; |
a542ad1b | 2221 | struct btrfs_key found_key; |
7a3ae2f8 | 2222 | int search_commit_root = path->search_commit_root; |
a542ad1b | 2223 | |
69917e43 | 2224 | ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); |
4692cf58 | 2225 | btrfs_release_path(path); |
a542ad1b JS |
2226 | if (ret < 0) |
2227 | return ret; | |
69917e43 | 2228 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) |
3627bf45 | 2229 | return -EINVAL; |
a542ad1b | 2230 | |
4692cf58 | 2231 | extent_item_pos = logical - found_key.objectid; |
7a3ae2f8 JS |
2232 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
2233 | extent_item_pos, search_commit_root, | |
e3059ec0 | 2234 | build_ino_list, ctx, ignore_offset); |
a542ad1b JS |
2235 | |
2236 | return ret; | |
2237 | } | |
2238 | ||
ad6240f6 | 2239 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2240 | struct extent_buffer *eb, struct inode_fs_paths *ipath); |
d24bec3a | 2241 | |
875d1daa | 2242 | static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath) |
a542ad1b | 2243 | { |
aefc1eb1 | 2244 | int ret = 0; |
a542ad1b JS |
2245 | int slot; |
2246 | u32 cur; | |
2247 | u32 len; | |
2248 | u32 name_len; | |
2249 | u64 parent = 0; | |
2250 | int found = 0; | |
875d1daa DS |
2251 | struct btrfs_root *fs_root = ipath->fs_root; |
2252 | struct btrfs_path *path = ipath->btrfs_path; | |
a542ad1b | 2253 | struct extent_buffer *eb; |
a542ad1b JS |
2254 | struct btrfs_inode_ref *iref; |
2255 | struct btrfs_key found_key; | |
2256 | ||
aefc1eb1 | 2257 | while (!ret) { |
c234a24d DS |
2258 | ret = btrfs_find_item(fs_root, path, inum, |
2259 | parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY, | |
2260 | &found_key); | |
2261 | ||
a542ad1b JS |
2262 | if (ret < 0) |
2263 | break; | |
2264 | if (ret) { | |
2265 | ret = found ? 0 : -ENOENT; | |
2266 | break; | |
2267 | } | |
2268 | ++found; | |
2269 | ||
2270 | parent = found_key.offset; | |
2271 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2272 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2273 | if (!eb) { | |
2274 | ret = -ENOMEM; | |
2275 | break; | |
2276 | } | |
a542ad1b JS |
2277 | btrfs_release_path(path); |
2278 | ||
a542ad1b JS |
2279 | iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); |
2280 | ||
3212fa14 | 2281 | for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) { |
a542ad1b JS |
2282 | name_len = btrfs_inode_ref_name_len(eb, iref); |
2283 | /* path must be released before calling iterate()! */ | |
ab8d0fc4 JM |
2284 | btrfs_debug(fs_root->fs_info, |
2285 | "following ref at offset %u for inode %llu in tree %llu", | |
4fd786e6 MT |
2286 | cur, found_key.objectid, |
2287 | fs_root->root_key.objectid); | |
ad6240f6 | 2288 | ret = inode_to_path(parent, name_len, |
875d1daa | 2289 | (unsigned long)(iref + 1), eb, ipath); |
aefc1eb1 | 2290 | if (ret) |
a542ad1b | 2291 | break; |
a542ad1b JS |
2292 | len = sizeof(*iref) + name_len; |
2293 | iref = (struct btrfs_inode_ref *)((char *)iref + len); | |
2294 | } | |
2295 | free_extent_buffer(eb); | |
2296 | } | |
2297 | ||
2298 | btrfs_release_path(path); | |
2299 | ||
2300 | return ret; | |
2301 | } | |
2302 | ||
875d1daa | 2303 | static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath) |
d24bec3a MF |
2304 | { |
2305 | int ret; | |
2306 | int slot; | |
2307 | u64 offset = 0; | |
2308 | u64 parent; | |
2309 | int found = 0; | |
875d1daa DS |
2310 | struct btrfs_root *fs_root = ipath->fs_root; |
2311 | struct btrfs_path *path = ipath->btrfs_path; | |
d24bec3a MF |
2312 | struct extent_buffer *eb; |
2313 | struct btrfs_inode_extref *extref; | |
d24bec3a MF |
2314 | u32 item_size; |
2315 | u32 cur_offset; | |
2316 | unsigned long ptr; | |
2317 | ||
2318 | while (1) { | |
2319 | ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, | |
2320 | &offset); | |
2321 | if (ret < 0) | |
2322 | break; | |
2323 | if (ret) { | |
2324 | ret = found ? 0 : -ENOENT; | |
2325 | break; | |
2326 | } | |
2327 | ++found; | |
2328 | ||
2329 | slot = path->slots[0]; | |
3fe81ce2 FDBM |
2330 | eb = btrfs_clone_extent_buffer(path->nodes[0]); |
2331 | if (!eb) { | |
2332 | ret = -ENOMEM; | |
2333 | break; | |
2334 | } | |
d24bec3a MF |
2335 | btrfs_release_path(path); |
2336 | ||
3212fa14 | 2337 | item_size = btrfs_item_size(eb, slot); |
2849a854 | 2338 | ptr = btrfs_item_ptr_offset(eb, slot); |
d24bec3a MF |
2339 | cur_offset = 0; |
2340 | ||
2341 | while (cur_offset < item_size) { | |
2342 | u32 name_len; | |
2343 | ||
2344 | extref = (struct btrfs_inode_extref *)(ptr + cur_offset); | |
2345 | parent = btrfs_inode_extref_parent(eb, extref); | |
2346 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
ad6240f6 | 2347 | ret = inode_to_path(parent, name_len, |
875d1daa | 2348 | (unsigned long)&extref->name, eb, ipath); |
d24bec3a MF |
2349 | if (ret) |
2350 | break; | |
2351 | ||
2849a854 | 2352 | cur_offset += btrfs_inode_extref_name_len(eb, extref); |
d24bec3a MF |
2353 | cur_offset += sizeof(*extref); |
2354 | } | |
d24bec3a MF |
2355 | free_extent_buffer(eb); |
2356 | ||
2357 | offset++; | |
2358 | } | |
2359 | ||
2360 | btrfs_release_path(path); | |
2361 | ||
2362 | return ret; | |
2363 | } | |
2364 | ||
a542ad1b JS |
2365 | /* |
2366 | * returns 0 if the path could be dumped (probably truncated) | |
2367 | * returns <0 in case of an error | |
2368 | */ | |
d24bec3a | 2369 | static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, |
875d1daa | 2370 | struct extent_buffer *eb, struct inode_fs_paths *ipath) |
a542ad1b | 2371 | { |
a542ad1b JS |
2372 | char *fspath; |
2373 | char *fspath_min; | |
2374 | int i = ipath->fspath->elem_cnt; | |
2375 | const int s_ptr = sizeof(char *); | |
2376 | u32 bytes_left; | |
2377 | ||
2378 | bytes_left = ipath->fspath->bytes_left > s_ptr ? | |
2379 | ipath->fspath->bytes_left - s_ptr : 0; | |
2380 | ||
740c3d22 | 2381 | fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; |
96b5bd77 JS |
2382 | fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, |
2383 | name_off, eb, inum, fspath_min, bytes_left); | |
a542ad1b JS |
2384 | if (IS_ERR(fspath)) |
2385 | return PTR_ERR(fspath); | |
2386 | ||
2387 | if (fspath > fspath_min) { | |
745c4d8e | 2388 | ipath->fspath->val[i] = (u64)(unsigned long)fspath; |
a542ad1b JS |
2389 | ++ipath->fspath->elem_cnt; |
2390 | ipath->fspath->bytes_left = fspath - fspath_min; | |
2391 | } else { | |
2392 | ++ipath->fspath->elem_missed; | |
2393 | ipath->fspath->bytes_missing += fspath_min - fspath; | |
2394 | ipath->fspath->bytes_left = 0; | |
2395 | } | |
2396 | ||
2397 | return 0; | |
2398 | } | |
2399 | ||
2400 | /* | |
2401 | * this dumps all file system paths to the inode into the ipath struct, provided | |
2402 | * is has been created large enough. each path is zero-terminated and accessed | |
740c3d22 | 2403 | * from ipath->fspath->val[i]. |
a542ad1b | 2404 | * when it returns, there are ipath->fspath->elem_cnt number of paths available |
740c3d22 | 2405 | * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the |
01327610 | 2406 | * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise, |
a542ad1b JS |
2407 | * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would |
2408 | * have been needed to return all paths. | |
2409 | */ | |
2410 | int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) | |
2411 | { | |
ad6240f6 DS |
2412 | int ret; |
2413 | int found_refs = 0; | |
2414 | ||
875d1daa | 2415 | ret = iterate_inode_refs(inum, ipath); |
ad6240f6 DS |
2416 | if (!ret) |
2417 | ++found_refs; | |
2418 | else if (ret != -ENOENT) | |
2419 | return ret; | |
2420 | ||
875d1daa | 2421 | ret = iterate_inode_extrefs(inum, ipath); |
ad6240f6 DS |
2422 | if (ret == -ENOENT && found_refs) |
2423 | return 0; | |
2424 | ||
2425 | return ret; | |
a542ad1b JS |
2426 | } |
2427 | ||
a542ad1b JS |
2428 | struct btrfs_data_container *init_data_container(u32 total_bytes) |
2429 | { | |
2430 | struct btrfs_data_container *data; | |
2431 | size_t alloc_bytes; | |
2432 | ||
2433 | alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); | |
f54de068 | 2434 | data = kvmalloc(alloc_bytes, GFP_KERNEL); |
a542ad1b JS |
2435 | if (!data) |
2436 | return ERR_PTR(-ENOMEM); | |
2437 | ||
2438 | if (total_bytes >= sizeof(*data)) { | |
2439 | data->bytes_left = total_bytes - sizeof(*data); | |
2440 | data->bytes_missing = 0; | |
2441 | } else { | |
2442 | data->bytes_missing = sizeof(*data) - total_bytes; | |
2443 | data->bytes_left = 0; | |
2444 | } | |
2445 | ||
2446 | data->elem_cnt = 0; | |
2447 | data->elem_missed = 0; | |
2448 | ||
2449 | return data; | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * allocates space to return multiple file system paths for an inode. | |
2454 | * total_bytes to allocate are passed, note that space usable for actual path | |
2455 | * information will be total_bytes - sizeof(struct inode_fs_paths). | |
2456 | * the returned pointer must be freed with free_ipath() in the end. | |
2457 | */ | |
2458 | struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, | |
2459 | struct btrfs_path *path) | |
2460 | { | |
2461 | struct inode_fs_paths *ifp; | |
2462 | struct btrfs_data_container *fspath; | |
2463 | ||
2464 | fspath = init_data_container(total_bytes); | |
2465 | if (IS_ERR(fspath)) | |
afc6961f | 2466 | return ERR_CAST(fspath); |
a542ad1b | 2467 | |
f54de068 | 2468 | ifp = kmalloc(sizeof(*ifp), GFP_KERNEL); |
a542ad1b | 2469 | if (!ifp) { |
f54de068 | 2470 | kvfree(fspath); |
a542ad1b JS |
2471 | return ERR_PTR(-ENOMEM); |
2472 | } | |
2473 | ||
2474 | ifp->btrfs_path = path; | |
2475 | ifp->fspath = fspath; | |
2476 | ifp->fs_root = fs_root; | |
2477 | ||
2478 | return ifp; | |
2479 | } | |
2480 | ||
2481 | void free_ipath(struct inode_fs_paths *ipath) | |
2482 | { | |
4735fb28 JJ |
2483 | if (!ipath) |
2484 | return; | |
f54de068 | 2485 | kvfree(ipath->fspath); |
a542ad1b JS |
2486 | kfree(ipath); |
2487 | } | |
a37f232b QW |
2488 | |
2489 | struct btrfs_backref_iter *btrfs_backref_iter_alloc( | |
2490 | struct btrfs_fs_info *fs_info, gfp_t gfp_flag) | |
2491 | { | |
2492 | struct btrfs_backref_iter *ret; | |
2493 | ||
2494 | ret = kzalloc(sizeof(*ret), gfp_flag); | |
2495 | if (!ret) | |
2496 | return NULL; | |
2497 | ||
2498 | ret->path = btrfs_alloc_path(); | |
c15c2ec0 | 2499 | if (!ret->path) { |
a37f232b QW |
2500 | kfree(ret); |
2501 | return NULL; | |
2502 | } | |
2503 | ||
2504 | /* Current backref iterator only supports iteration in commit root */ | |
2505 | ret->path->search_commit_root = 1; | |
2506 | ret->path->skip_locking = 1; | |
2507 | ret->fs_info = fs_info; | |
2508 | ||
2509 | return ret; | |
2510 | } | |
2511 | ||
2512 | int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr) | |
2513 | { | |
2514 | struct btrfs_fs_info *fs_info = iter->fs_info; | |
29cbcf40 | 2515 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); |
a37f232b QW |
2516 | struct btrfs_path *path = iter->path; |
2517 | struct btrfs_extent_item *ei; | |
2518 | struct btrfs_key key; | |
2519 | int ret; | |
2520 | ||
2521 | key.objectid = bytenr; | |
2522 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2523 | key.offset = (u64)-1; | |
2524 | iter->bytenr = bytenr; | |
2525 | ||
29cbcf40 | 2526 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
a37f232b QW |
2527 | if (ret < 0) |
2528 | return ret; | |
2529 | if (ret == 0) { | |
2530 | ret = -EUCLEAN; | |
2531 | goto release; | |
2532 | } | |
2533 | if (path->slots[0] == 0) { | |
2534 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
2535 | ret = -EUCLEAN; | |
2536 | goto release; | |
2537 | } | |
2538 | path->slots[0]--; | |
2539 | ||
2540 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2541 | if ((key.type != BTRFS_EXTENT_ITEM_KEY && | |
2542 | key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) { | |
2543 | ret = -ENOENT; | |
2544 | goto release; | |
2545 | } | |
2546 | memcpy(&iter->cur_key, &key, sizeof(key)); | |
2547 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2548 | path->slots[0]); | |
2549 | iter->end_ptr = (u32)(iter->item_ptr + | |
3212fa14 | 2550 | btrfs_item_size(path->nodes[0], path->slots[0])); |
a37f232b QW |
2551 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2552 | struct btrfs_extent_item); | |
2553 | ||
2554 | /* | |
2555 | * Only support iteration on tree backref yet. | |
2556 | * | |
2557 | * This is an extra precaution for non skinny-metadata, where | |
2558 | * EXTENT_ITEM is also used for tree blocks, that we can only use | |
2559 | * extent flags to determine if it's a tree block. | |
2560 | */ | |
2561 | if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) { | |
2562 | ret = -ENOTSUPP; | |
2563 | goto release; | |
2564 | } | |
2565 | iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei)); | |
2566 | ||
2567 | /* If there is no inline backref, go search for keyed backref */ | |
2568 | if (iter->cur_ptr >= iter->end_ptr) { | |
29cbcf40 | 2569 | ret = btrfs_next_item(extent_root, path); |
a37f232b QW |
2570 | |
2571 | /* No inline nor keyed ref */ | |
2572 | if (ret > 0) { | |
2573 | ret = -ENOENT; | |
2574 | goto release; | |
2575 | } | |
2576 | if (ret < 0) | |
2577 | goto release; | |
2578 | ||
2579 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, | |
2580 | path->slots[0]); | |
2581 | if (iter->cur_key.objectid != bytenr || | |
2582 | (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY && | |
2583 | iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) { | |
2584 | ret = -ENOENT; | |
2585 | goto release; | |
2586 | } | |
2587 | iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2588 | path->slots[0]); | |
2589 | iter->item_ptr = iter->cur_ptr; | |
3212fa14 | 2590 | iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size( |
a37f232b QW |
2591 | path->nodes[0], path->slots[0])); |
2592 | } | |
2593 | ||
2594 | return 0; | |
2595 | release: | |
2596 | btrfs_backref_iter_release(iter); | |
2597 | return ret; | |
2598 | } | |
c39c2ddc QW |
2599 | |
2600 | /* | |
2601 | * Go to the next backref item of current bytenr, can be either inlined or | |
2602 | * keyed. | |
2603 | * | |
2604 | * Caller needs to check whether it's inline ref or not by iter->cur_key. | |
2605 | * | |
2606 | * Return 0 if we get next backref without problem. | |
2607 | * Return >0 if there is no extra backref for this bytenr. | |
2608 | * Return <0 if there is something wrong happened. | |
2609 | */ | |
2610 | int btrfs_backref_iter_next(struct btrfs_backref_iter *iter) | |
2611 | { | |
2612 | struct extent_buffer *eb = btrfs_backref_get_eb(iter); | |
29cbcf40 | 2613 | struct btrfs_root *extent_root; |
c39c2ddc QW |
2614 | struct btrfs_path *path = iter->path; |
2615 | struct btrfs_extent_inline_ref *iref; | |
2616 | int ret; | |
2617 | u32 size; | |
2618 | ||
2619 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
2620 | /* We're still inside the inline refs */ | |
2621 | ASSERT(iter->cur_ptr < iter->end_ptr); | |
2622 | ||
2623 | if (btrfs_backref_has_tree_block_info(iter)) { | |
2624 | /* First tree block info */ | |
2625 | size = sizeof(struct btrfs_tree_block_info); | |
2626 | } else { | |
2627 | /* Use inline ref type to determine the size */ | |
2628 | int type; | |
2629 | ||
2630 | iref = (struct btrfs_extent_inline_ref *) | |
2631 | ((unsigned long)iter->cur_ptr); | |
2632 | type = btrfs_extent_inline_ref_type(eb, iref); | |
2633 | ||
2634 | size = btrfs_extent_inline_ref_size(type); | |
2635 | } | |
2636 | iter->cur_ptr += size; | |
2637 | if (iter->cur_ptr < iter->end_ptr) | |
2638 | return 0; | |
2639 | ||
2640 | /* All inline items iterated, fall through */ | |
2641 | } | |
2642 | ||
2643 | /* We're at keyed items, there is no inline item, go to the next one */ | |
29cbcf40 JB |
2644 | extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr); |
2645 | ret = btrfs_next_item(extent_root, iter->path); | |
c39c2ddc QW |
2646 | if (ret) |
2647 | return ret; | |
2648 | ||
2649 | btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]); | |
2650 | if (iter->cur_key.objectid != iter->bytenr || | |
2651 | (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY && | |
2652 | iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY)) | |
2653 | return 1; | |
2654 | iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], | |
2655 | path->slots[0]); | |
2656 | iter->cur_ptr = iter->item_ptr; | |
3212fa14 | 2657 | iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0], |
c39c2ddc QW |
2658 | path->slots[0]); |
2659 | return 0; | |
2660 | } | |
584fb121 QW |
2661 | |
2662 | void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info, | |
2663 | struct btrfs_backref_cache *cache, int is_reloc) | |
2664 | { | |
2665 | int i; | |
2666 | ||
2667 | cache->rb_root = RB_ROOT; | |
2668 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
2669 | INIT_LIST_HEAD(&cache->pending[i]); | |
2670 | INIT_LIST_HEAD(&cache->changed); | |
2671 | INIT_LIST_HEAD(&cache->detached); | |
2672 | INIT_LIST_HEAD(&cache->leaves); | |
2673 | INIT_LIST_HEAD(&cache->pending_edge); | |
2674 | INIT_LIST_HEAD(&cache->useless_node); | |
2675 | cache->fs_info = fs_info; | |
2676 | cache->is_reloc = is_reloc; | |
2677 | } | |
b1818dab QW |
2678 | |
2679 | struct btrfs_backref_node *btrfs_backref_alloc_node( | |
2680 | struct btrfs_backref_cache *cache, u64 bytenr, int level) | |
2681 | { | |
2682 | struct btrfs_backref_node *node; | |
2683 | ||
2684 | ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL); | |
2685 | node = kzalloc(sizeof(*node), GFP_NOFS); | |
2686 | if (!node) | |
2687 | return node; | |
2688 | ||
2689 | INIT_LIST_HEAD(&node->list); | |
2690 | INIT_LIST_HEAD(&node->upper); | |
2691 | INIT_LIST_HEAD(&node->lower); | |
2692 | RB_CLEAR_NODE(&node->rb_node); | |
2693 | cache->nr_nodes++; | |
2694 | node->level = level; | |
2695 | node->bytenr = bytenr; | |
2696 | ||
2697 | return node; | |
2698 | } | |
47254d07 QW |
2699 | |
2700 | struct btrfs_backref_edge *btrfs_backref_alloc_edge( | |
2701 | struct btrfs_backref_cache *cache) | |
2702 | { | |
2703 | struct btrfs_backref_edge *edge; | |
2704 | ||
2705 | edge = kzalloc(sizeof(*edge), GFP_NOFS); | |
2706 | if (edge) | |
2707 | cache->nr_edges++; | |
2708 | return edge; | |
2709 | } | |
023acb07 QW |
2710 | |
2711 | /* | |
2712 | * Drop the backref node from cache, also cleaning up all its | |
2713 | * upper edges and any uncached nodes in the path. | |
2714 | * | |
2715 | * This cleanup happens bottom up, thus the node should either | |
2716 | * be the lowest node in the cache or a detached node. | |
2717 | */ | |
2718 | void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache, | |
2719 | struct btrfs_backref_node *node) | |
2720 | { | |
2721 | struct btrfs_backref_node *upper; | |
2722 | struct btrfs_backref_edge *edge; | |
2723 | ||
2724 | if (!node) | |
2725 | return; | |
2726 | ||
2727 | BUG_ON(!node->lowest && !node->detached); | |
2728 | while (!list_empty(&node->upper)) { | |
2729 | edge = list_entry(node->upper.next, struct btrfs_backref_edge, | |
2730 | list[LOWER]); | |
2731 | upper = edge->node[UPPER]; | |
2732 | list_del(&edge->list[LOWER]); | |
2733 | list_del(&edge->list[UPPER]); | |
2734 | btrfs_backref_free_edge(cache, edge); | |
2735 | ||
023acb07 QW |
2736 | /* |
2737 | * Add the node to leaf node list if no other child block | |
2738 | * cached. | |
2739 | */ | |
2740 | if (list_empty(&upper->lower)) { | |
2741 | list_add_tail(&upper->lower, &cache->leaves); | |
2742 | upper->lowest = 1; | |
2743 | } | |
2744 | } | |
2745 | ||
2746 | btrfs_backref_drop_node(cache, node); | |
2747 | } | |
13fe1bdb QW |
2748 | |
2749 | /* | |
2750 | * Release all nodes/edges from current cache | |
2751 | */ | |
2752 | void btrfs_backref_release_cache(struct btrfs_backref_cache *cache) | |
2753 | { | |
2754 | struct btrfs_backref_node *node; | |
2755 | int i; | |
2756 | ||
2757 | while (!list_empty(&cache->detached)) { | |
2758 | node = list_entry(cache->detached.next, | |
2759 | struct btrfs_backref_node, list); | |
2760 | btrfs_backref_cleanup_node(cache, node); | |
2761 | } | |
2762 | ||
2763 | while (!list_empty(&cache->leaves)) { | |
2764 | node = list_entry(cache->leaves.next, | |
2765 | struct btrfs_backref_node, lower); | |
2766 | btrfs_backref_cleanup_node(cache, node); | |
2767 | } | |
2768 | ||
2769 | cache->last_trans = 0; | |
2770 | ||
2771 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) | |
2772 | ASSERT(list_empty(&cache->pending[i])); | |
2773 | ASSERT(list_empty(&cache->pending_edge)); | |
2774 | ASSERT(list_empty(&cache->useless_node)); | |
2775 | ASSERT(list_empty(&cache->changed)); | |
2776 | ASSERT(list_empty(&cache->detached)); | |
2777 | ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); | |
2778 | ASSERT(!cache->nr_nodes); | |
2779 | ASSERT(!cache->nr_edges); | |
2780 | } | |
1b60d2ec QW |
2781 | |
2782 | /* | |
2783 | * Handle direct tree backref | |
2784 | * | |
2785 | * Direct tree backref means, the backref item shows its parent bytenr | |
2786 | * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined). | |
2787 | * | |
2788 | * @ref_key: The converted backref key. | |
2789 | * For keyed backref, it's the item key. | |
2790 | * For inlined backref, objectid is the bytenr, | |
2791 | * type is btrfs_inline_ref_type, offset is | |
2792 | * btrfs_inline_ref_offset. | |
2793 | */ | |
2794 | static int handle_direct_tree_backref(struct btrfs_backref_cache *cache, | |
2795 | struct btrfs_key *ref_key, | |
2796 | struct btrfs_backref_node *cur) | |
2797 | { | |
2798 | struct btrfs_backref_edge *edge; | |
2799 | struct btrfs_backref_node *upper; | |
2800 | struct rb_node *rb_node; | |
2801 | ||
2802 | ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
2803 | ||
2804 | /* Only reloc root uses backref pointing to itself */ | |
2805 | if (ref_key->objectid == ref_key->offset) { | |
2806 | struct btrfs_root *root; | |
2807 | ||
2808 | cur->is_reloc_root = 1; | |
2809 | /* Only reloc backref cache cares about a specific root */ | |
2810 | if (cache->is_reloc) { | |
2811 | root = find_reloc_root(cache->fs_info, cur->bytenr); | |
f78743fb | 2812 | if (!root) |
1b60d2ec QW |
2813 | return -ENOENT; |
2814 | cur->root = root; | |
2815 | } else { | |
2816 | /* | |
2817 | * For generic purpose backref cache, reloc root node | |
2818 | * is useless. | |
2819 | */ | |
2820 | list_add(&cur->list, &cache->useless_node); | |
2821 | } | |
2822 | return 0; | |
2823 | } | |
2824 | ||
2825 | edge = btrfs_backref_alloc_edge(cache); | |
2826 | if (!edge) | |
2827 | return -ENOMEM; | |
2828 | ||
2829 | rb_node = rb_simple_search(&cache->rb_root, ref_key->offset); | |
2830 | if (!rb_node) { | |
2831 | /* Parent node not yet cached */ | |
2832 | upper = btrfs_backref_alloc_node(cache, ref_key->offset, | |
2833 | cur->level + 1); | |
2834 | if (!upper) { | |
2835 | btrfs_backref_free_edge(cache, edge); | |
2836 | return -ENOMEM; | |
2837 | } | |
2838 | ||
2839 | /* | |
2840 | * Backrefs for the upper level block isn't cached, add the | |
2841 | * block to pending list | |
2842 | */ | |
2843 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
2844 | } else { | |
2845 | /* Parent node already cached */ | |
2846 | upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node); | |
2847 | ASSERT(upper->checked); | |
2848 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
2849 | } | |
2850 | btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER); | |
2851 | return 0; | |
2852 | } | |
2853 | ||
2854 | /* | |
2855 | * Handle indirect tree backref | |
2856 | * | |
2857 | * Indirect tree backref means, we only know which tree the node belongs to. | |
2858 | * We still need to do a tree search to find out the parents. This is for | |
2859 | * TREE_BLOCK_REF backref (keyed or inlined). | |
2860 | * | |
2861 | * @ref_key: The same as @ref_key in handle_direct_tree_backref() | |
2862 | * @tree_key: The first key of this tree block. | |
1a9fd417 | 2863 | * @path: A clean (released) path, to avoid allocating path every time |
1b60d2ec QW |
2864 | * the function get called. |
2865 | */ | |
2866 | static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache, | |
2867 | struct btrfs_path *path, | |
2868 | struct btrfs_key *ref_key, | |
2869 | struct btrfs_key *tree_key, | |
2870 | struct btrfs_backref_node *cur) | |
2871 | { | |
2872 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2873 | struct btrfs_backref_node *upper; | |
2874 | struct btrfs_backref_node *lower; | |
2875 | struct btrfs_backref_edge *edge; | |
2876 | struct extent_buffer *eb; | |
2877 | struct btrfs_root *root; | |
1b60d2ec QW |
2878 | struct rb_node *rb_node; |
2879 | int level; | |
2880 | bool need_check = true; | |
2881 | int ret; | |
2882 | ||
56e9357a | 2883 | root = btrfs_get_fs_root(fs_info, ref_key->offset, false); |
1b60d2ec QW |
2884 | if (IS_ERR(root)) |
2885 | return PTR_ERR(root); | |
92a7cc42 | 2886 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
2887 | cur->cowonly = 1; |
2888 | ||
2889 | if (btrfs_root_level(&root->root_item) == cur->level) { | |
2890 | /* Tree root */ | |
2891 | ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr); | |
876de781 QW |
2892 | /* |
2893 | * For reloc backref cache, we may ignore reloc root. But for | |
2894 | * general purpose backref cache, we can't rely on | |
2895 | * btrfs_should_ignore_reloc_root() as it may conflict with | |
2896 | * current running relocation and lead to missing root. | |
2897 | * | |
2898 | * For general purpose backref cache, reloc root detection is | |
2899 | * completely relying on direct backref (key->offset is parent | |
2900 | * bytenr), thus only do such check for reloc cache. | |
2901 | */ | |
2902 | if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) { | |
1b60d2ec QW |
2903 | btrfs_put_root(root); |
2904 | list_add(&cur->list, &cache->useless_node); | |
2905 | } else { | |
2906 | cur->root = root; | |
2907 | } | |
2908 | return 0; | |
2909 | } | |
2910 | ||
2911 | level = cur->level + 1; | |
2912 | ||
2913 | /* Search the tree to find parent blocks referring to the block */ | |
2914 | path->search_commit_root = 1; | |
2915 | path->skip_locking = 1; | |
2916 | path->lowest_level = level; | |
2917 | ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0); | |
2918 | path->lowest_level = 0; | |
2919 | if (ret < 0) { | |
2920 | btrfs_put_root(root); | |
2921 | return ret; | |
2922 | } | |
2923 | if (ret > 0 && path->slots[level] > 0) | |
2924 | path->slots[level]--; | |
2925 | ||
2926 | eb = path->nodes[level]; | |
2927 | if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) { | |
2928 | btrfs_err(fs_info, | |
2929 | "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", | |
2930 | cur->bytenr, level - 1, root->root_key.objectid, | |
2931 | tree_key->objectid, tree_key->type, tree_key->offset); | |
2932 | btrfs_put_root(root); | |
2933 | ret = -ENOENT; | |
2934 | goto out; | |
2935 | } | |
2936 | lower = cur; | |
2937 | ||
2938 | /* Add all nodes and edges in the path */ | |
2939 | for (; level < BTRFS_MAX_LEVEL; level++) { | |
2940 | if (!path->nodes[level]) { | |
2941 | ASSERT(btrfs_root_bytenr(&root->root_item) == | |
2942 | lower->bytenr); | |
876de781 QW |
2943 | /* Same as previous should_ignore_reloc_root() call */ |
2944 | if (btrfs_should_ignore_reloc_root(root) && | |
2945 | cache->is_reloc) { | |
1b60d2ec QW |
2946 | btrfs_put_root(root); |
2947 | list_add(&lower->list, &cache->useless_node); | |
2948 | } else { | |
2949 | lower->root = root; | |
2950 | } | |
2951 | break; | |
2952 | } | |
2953 | ||
2954 | edge = btrfs_backref_alloc_edge(cache); | |
2955 | if (!edge) { | |
2956 | btrfs_put_root(root); | |
2957 | ret = -ENOMEM; | |
2958 | goto out; | |
2959 | } | |
2960 | ||
2961 | eb = path->nodes[level]; | |
2962 | rb_node = rb_simple_search(&cache->rb_root, eb->start); | |
2963 | if (!rb_node) { | |
2964 | upper = btrfs_backref_alloc_node(cache, eb->start, | |
2965 | lower->level + 1); | |
2966 | if (!upper) { | |
2967 | btrfs_put_root(root); | |
2968 | btrfs_backref_free_edge(cache, edge); | |
2969 | ret = -ENOMEM; | |
2970 | goto out; | |
2971 | } | |
2972 | upper->owner = btrfs_header_owner(eb); | |
92a7cc42 | 2973 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
1b60d2ec QW |
2974 | upper->cowonly = 1; |
2975 | ||
2976 | /* | |
2977 | * If we know the block isn't shared we can avoid | |
2978 | * checking its backrefs. | |
2979 | */ | |
2980 | if (btrfs_block_can_be_shared(root, eb)) | |
2981 | upper->checked = 0; | |
2982 | else | |
2983 | upper->checked = 1; | |
2984 | ||
2985 | /* | |
2986 | * Add the block to pending list if we need to check its | |
2987 | * backrefs, we only do this once while walking up a | |
2988 | * tree as we will catch anything else later on. | |
2989 | */ | |
2990 | if (!upper->checked && need_check) { | |
2991 | need_check = false; | |
2992 | list_add_tail(&edge->list[UPPER], | |
2993 | &cache->pending_edge); | |
2994 | } else { | |
2995 | if (upper->checked) | |
2996 | need_check = true; | |
2997 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
2998 | } | |
2999 | } else { | |
3000 | upper = rb_entry(rb_node, struct btrfs_backref_node, | |
3001 | rb_node); | |
3002 | ASSERT(upper->checked); | |
3003 | INIT_LIST_HEAD(&edge->list[UPPER]); | |
3004 | if (!upper->owner) | |
3005 | upper->owner = btrfs_header_owner(eb); | |
3006 | } | |
3007 | btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER); | |
3008 | ||
3009 | if (rb_node) { | |
3010 | btrfs_put_root(root); | |
3011 | break; | |
3012 | } | |
3013 | lower = upper; | |
3014 | upper = NULL; | |
3015 | } | |
3016 | out: | |
3017 | btrfs_release_path(path); | |
3018 | return ret; | |
3019 | } | |
3020 | ||
3021 | /* | |
3022 | * Add backref node @cur into @cache. | |
3023 | * | |
3024 | * NOTE: Even if the function returned 0, @cur is not yet cached as its upper | |
3025 | * links aren't yet bi-directional. Needs to finish such links. | |
fc997ed0 | 3026 | * Use btrfs_backref_finish_upper_links() to finish such linkage. |
1b60d2ec QW |
3027 | * |
3028 | * @path: Released path for indirect tree backref lookup | |
3029 | * @iter: Released backref iter for extent tree search | |
3030 | * @node_key: The first key of the tree block | |
3031 | */ | |
3032 | int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache, | |
3033 | struct btrfs_path *path, | |
3034 | struct btrfs_backref_iter *iter, | |
3035 | struct btrfs_key *node_key, | |
3036 | struct btrfs_backref_node *cur) | |
3037 | { | |
3038 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3039 | struct btrfs_backref_edge *edge; | |
3040 | struct btrfs_backref_node *exist; | |
3041 | int ret; | |
3042 | ||
3043 | ret = btrfs_backref_iter_start(iter, cur->bytenr); | |
3044 | if (ret < 0) | |
3045 | return ret; | |
3046 | /* | |
3047 | * We skip the first btrfs_tree_block_info, as we don't use the key | |
3048 | * stored in it, but fetch it from the tree block | |
3049 | */ | |
3050 | if (btrfs_backref_has_tree_block_info(iter)) { | |
3051 | ret = btrfs_backref_iter_next(iter); | |
3052 | if (ret < 0) | |
3053 | goto out; | |
3054 | /* No extra backref? This means the tree block is corrupted */ | |
3055 | if (ret > 0) { | |
3056 | ret = -EUCLEAN; | |
3057 | goto out; | |
3058 | } | |
3059 | } | |
3060 | WARN_ON(cur->checked); | |
3061 | if (!list_empty(&cur->upper)) { | |
3062 | /* | |
3063 | * The backref was added previously when processing backref of | |
3064 | * type BTRFS_TREE_BLOCK_REF_KEY | |
3065 | */ | |
3066 | ASSERT(list_is_singular(&cur->upper)); | |
3067 | edge = list_entry(cur->upper.next, struct btrfs_backref_edge, | |
3068 | list[LOWER]); | |
3069 | ASSERT(list_empty(&edge->list[UPPER])); | |
3070 | exist = edge->node[UPPER]; | |
3071 | /* | |
3072 | * Add the upper level block to pending list if we need check | |
3073 | * its backrefs | |
3074 | */ | |
3075 | if (!exist->checked) | |
3076 | list_add_tail(&edge->list[UPPER], &cache->pending_edge); | |
3077 | } else { | |
3078 | exist = NULL; | |
3079 | } | |
3080 | ||
3081 | for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { | |
3082 | struct extent_buffer *eb; | |
3083 | struct btrfs_key key; | |
3084 | int type; | |
3085 | ||
3086 | cond_resched(); | |
3087 | eb = btrfs_backref_get_eb(iter); | |
3088 | ||
3089 | key.objectid = iter->bytenr; | |
3090 | if (btrfs_backref_iter_is_inline_ref(iter)) { | |
3091 | struct btrfs_extent_inline_ref *iref; | |
3092 | ||
3093 | /* Update key for inline backref */ | |
3094 | iref = (struct btrfs_extent_inline_ref *) | |
3095 | ((unsigned long)iter->cur_ptr); | |
3096 | type = btrfs_get_extent_inline_ref_type(eb, iref, | |
3097 | BTRFS_REF_TYPE_BLOCK); | |
3098 | if (type == BTRFS_REF_TYPE_INVALID) { | |
3099 | ret = -EUCLEAN; | |
3100 | goto out; | |
3101 | } | |
3102 | key.type = type; | |
3103 | key.offset = btrfs_extent_inline_ref_offset(eb, iref); | |
3104 | } else { | |
3105 | key.type = iter->cur_key.type; | |
3106 | key.offset = iter->cur_key.offset; | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * Parent node found and matches current inline ref, no need to | |
3111 | * rebuild this node for this inline ref | |
3112 | */ | |
3113 | if (exist && | |
3114 | ((key.type == BTRFS_TREE_BLOCK_REF_KEY && | |
3115 | exist->owner == key.offset) || | |
3116 | (key.type == BTRFS_SHARED_BLOCK_REF_KEY && | |
3117 | exist->bytenr == key.offset))) { | |
3118 | exist = NULL; | |
3119 | continue; | |
3120 | } | |
3121 | ||
3122 | /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ | |
3123 | if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
3124 | ret = handle_direct_tree_backref(cache, &key, cur); | |
3125 | if (ret < 0) | |
3126 | goto out; | |
3127 | continue; | |
3128 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { | |
3129 | ret = -EINVAL; | |
3130 | btrfs_print_v0_err(fs_info); | |
3131 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
3132 | goto out; | |
3133 | } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { | |
3134 | continue; | |
3135 | } | |
3136 | ||
3137 | /* | |
3138 | * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset | |
3139 | * means the root objectid. We need to search the tree to get | |
3140 | * its parent bytenr. | |
3141 | */ | |
3142 | ret = handle_indirect_tree_backref(cache, path, &key, node_key, | |
3143 | cur); | |
3144 | if (ret < 0) | |
3145 | goto out; | |
3146 | } | |
3147 | ret = 0; | |
3148 | cur->checked = 1; | |
3149 | WARN_ON(exist); | |
3150 | out: | |
3151 | btrfs_backref_iter_release(iter); | |
3152 | return ret; | |
3153 | } | |
fc997ed0 QW |
3154 | |
3155 | /* | |
3156 | * Finish the upwards linkage created by btrfs_backref_add_tree_node() | |
3157 | */ | |
3158 | int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache, | |
3159 | struct btrfs_backref_node *start) | |
3160 | { | |
3161 | struct list_head *useless_node = &cache->useless_node; | |
3162 | struct btrfs_backref_edge *edge; | |
3163 | struct rb_node *rb_node; | |
3164 | LIST_HEAD(pending_edge); | |
3165 | ||
3166 | ASSERT(start->checked); | |
3167 | ||
3168 | /* Insert this node to cache if it's not COW-only */ | |
3169 | if (!start->cowonly) { | |
3170 | rb_node = rb_simple_insert(&cache->rb_root, start->bytenr, | |
3171 | &start->rb_node); | |
3172 | if (rb_node) | |
3173 | btrfs_backref_panic(cache->fs_info, start->bytenr, | |
3174 | -EEXIST); | |
3175 | list_add_tail(&start->lower, &cache->leaves); | |
3176 | } | |
3177 | ||
3178 | /* | |
3179 | * Use breadth first search to iterate all related edges. | |
3180 | * | |
3181 | * The starting points are all the edges of this node | |
3182 | */ | |
3183 | list_for_each_entry(edge, &start->upper, list[LOWER]) | |
3184 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3185 | ||
3186 | while (!list_empty(&pending_edge)) { | |
3187 | struct btrfs_backref_node *upper; | |
3188 | struct btrfs_backref_node *lower; | |
fc997ed0 QW |
3189 | |
3190 | edge = list_first_entry(&pending_edge, | |
3191 | struct btrfs_backref_edge, list[UPPER]); | |
3192 | list_del_init(&edge->list[UPPER]); | |
3193 | upper = edge->node[UPPER]; | |
3194 | lower = edge->node[LOWER]; | |
3195 | ||
3196 | /* Parent is detached, no need to keep any edges */ | |
3197 | if (upper->detached) { | |
3198 | list_del(&edge->list[LOWER]); | |
3199 | btrfs_backref_free_edge(cache, edge); | |
3200 | ||
3201 | /* Lower node is orphan, queue for cleanup */ | |
3202 | if (list_empty(&lower->upper)) | |
3203 | list_add(&lower->list, useless_node); | |
3204 | continue; | |
3205 | } | |
3206 | ||
3207 | /* | |
3208 | * All new nodes added in current build_backref_tree() haven't | |
3209 | * been linked to the cache rb tree. | |
3210 | * So if we have upper->rb_node populated, this means a cache | |
3211 | * hit. We only need to link the edge, as @upper and all its | |
3212 | * parents have already been linked. | |
3213 | */ | |
3214 | if (!RB_EMPTY_NODE(&upper->rb_node)) { | |
3215 | if (upper->lowest) { | |
3216 | list_del_init(&upper->lower); | |
3217 | upper->lowest = 0; | |
3218 | } | |
3219 | ||
3220 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3221 | continue; | |
3222 | } | |
3223 | ||
3224 | /* Sanity check, we shouldn't have any unchecked nodes */ | |
3225 | if (!upper->checked) { | |
3226 | ASSERT(0); | |
3227 | return -EUCLEAN; | |
3228 | } | |
3229 | ||
3230 | /* Sanity check, COW-only node has non-COW-only parent */ | |
3231 | if (start->cowonly != upper->cowonly) { | |
3232 | ASSERT(0); | |
3233 | return -EUCLEAN; | |
3234 | } | |
3235 | ||
3236 | /* Only cache non-COW-only (subvolume trees) tree blocks */ | |
3237 | if (!upper->cowonly) { | |
3238 | rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr, | |
3239 | &upper->rb_node); | |
3240 | if (rb_node) { | |
3241 | btrfs_backref_panic(cache->fs_info, | |
3242 | upper->bytenr, -EEXIST); | |
3243 | return -EUCLEAN; | |
3244 | } | |
3245 | } | |
3246 | ||
3247 | list_add_tail(&edge->list[UPPER], &upper->lower); | |
3248 | ||
3249 | /* | |
3250 | * Also queue all the parent edges of this uncached node | |
3251 | * to finish the upper linkage | |
3252 | */ | |
3253 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3254 | list_add_tail(&edge->list[UPPER], &pending_edge); | |
3255 | } | |
3256 | return 0; | |
3257 | } | |
1b23ea18 QW |
3258 | |
3259 | void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache, | |
3260 | struct btrfs_backref_node *node) | |
3261 | { | |
3262 | struct btrfs_backref_node *lower; | |
3263 | struct btrfs_backref_node *upper; | |
3264 | struct btrfs_backref_edge *edge; | |
3265 | ||
3266 | while (!list_empty(&cache->useless_node)) { | |
3267 | lower = list_first_entry(&cache->useless_node, | |
3268 | struct btrfs_backref_node, list); | |
3269 | list_del_init(&lower->list); | |
3270 | } | |
3271 | while (!list_empty(&cache->pending_edge)) { | |
3272 | edge = list_first_entry(&cache->pending_edge, | |
3273 | struct btrfs_backref_edge, list[UPPER]); | |
3274 | list_del(&edge->list[UPPER]); | |
3275 | list_del(&edge->list[LOWER]); | |
3276 | lower = edge->node[LOWER]; | |
3277 | upper = edge->node[UPPER]; | |
3278 | btrfs_backref_free_edge(cache, edge); | |
3279 | ||
3280 | /* | |
3281 | * Lower is no longer linked to any upper backref nodes and | |
3282 | * isn't in the cache, we can free it ourselves. | |
3283 | */ | |
3284 | if (list_empty(&lower->upper) && | |
3285 | RB_EMPTY_NODE(&lower->rb_node)) | |
3286 | list_add(&lower->list, &cache->useless_node); | |
3287 | ||
3288 | if (!RB_EMPTY_NODE(&upper->rb_node)) | |
3289 | continue; | |
3290 | ||
3291 | /* Add this guy's upper edges to the list to process */ | |
3292 | list_for_each_entry(edge, &upper->upper, list[LOWER]) | |
3293 | list_add_tail(&edge->list[UPPER], | |
3294 | &cache->pending_edge); | |
3295 | if (list_empty(&upper->upper)) | |
3296 | list_add(&upper->list, &cache->useless_node); | |
3297 | } | |
3298 | ||
3299 | while (!list_empty(&cache->useless_node)) { | |
3300 | lower = list_first_entry(&cache->useless_node, | |
3301 | struct btrfs_backref_node, list); | |
3302 | list_del_init(&lower->list); | |
3303 | if (lower == node) | |
3304 | node = NULL; | |
49ecc679 | 3305 | btrfs_backref_drop_node(cache, lower); |
1b23ea18 QW |
3306 | } |
3307 | ||
3308 | btrfs_backref_cleanup_node(cache, node); | |
3309 | ASSERT(list_empty(&cache->useless_node) && | |
3310 | list_empty(&cache->pending_edge)); | |
3311 | } |