Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2011 STRATO. All rights reserved.
4 : */
5 :
6 : #include <linux/mm.h>
7 : #include <linux/rbtree.h>
8 : #include <trace/events/btrfs.h>
9 : #include "ctree.h"
10 : #include "disk-io.h"
11 : #include "backref.h"
12 : #include "ulist.h"
13 : #include "transaction.h"
14 : #include "delayed-ref.h"
15 : #include "locking.h"
16 : #include "misc.h"
17 : #include "tree-mod-log.h"
18 : #include "fs.h"
19 : #include "accessors.h"
20 : #include "extent-tree.h"
21 : #include "relocation.h"
22 : #include "tree-checker.h"
23 :
24 : /* Just arbitrary numbers so we can be sure one of these happened. */
25 : #define BACKREF_FOUND_SHARED 6
26 : #define BACKREF_FOUND_NOT_SHARED 7
27 :
28 : struct extent_inode_elem {
29 : u64 inum;
30 : u64 offset;
31 : u64 num_bytes;
32 : struct extent_inode_elem *next;
33 : };
34 :
35 2008740 : static int check_extent_in_eb(struct btrfs_backref_walk_ctx *ctx,
36 : const struct btrfs_key *key,
37 : const struct extent_buffer *eb,
38 : const struct btrfs_file_extent_item *fi,
39 : struct extent_inode_elem **eie)
40 : {
41 2008740 : const u64 data_len = btrfs_file_extent_num_bytes(eb, fi);
42 2008738 : u64 offset = key->offset;
43 2008738 : struct extent_inode_elem *e;
44 2008738 : const u64 *root_ids;
45 2008738 : int root_count;
46 2008738 : bool cached;
47 :
48 4017440 : if (!ctx->ignore_extent_item_pos &&
49 2008653 : !btrfs_file_extent_compression(eb, fi) &&
50 2008647 : !btrfs_file_extent_encryption(eb, fi) &&
51 : !btrfs_file_extent_other_encoding(eb, fi)) {
52 2008647 : u64 data_offset;
53 :
54 2008647 : data_offset = btrfs_file_extent_offset(eb, fi);
55 :
56 2008653 : if (ctx->extent_item_pos < data_offset ||
57 1795297 : ctx->extent_item_pos >= data_offset + data_len)
58 : return 1;
59 1674837 : offset += ctx->extent_item_pos - data_offset;
60 : }
61 :
62 1674924 : if (!ctx->indirect_ref_iterator || !ctx->cache_lookup)
63 172225 : goto add_inode_elem;
64 :
65 1502699 : cached = ctx->cache_lookup(eb->start, ctx->user_ctx, &root_ids,
66 : &root_count);
67 1502699 : if (!cached)
68 534438 : goto add_inode_elem;
69 :
70 1459077 : for (int i = 0; i < root_count; i++) {
71 494050 : int ret;
72 :
73 494050 : ret = ctx->indirect_ref_iterator(key->objectid, offset,
74 494050 : data_len, root_ids[i],
75 : ctx->user_ctx);
76 494052 : if (ret)
77 3236 : return ret;
78 : }
79 :
80 965027 : add_inode_elem:
81 1671690 : e = kmalloc(sizeof(*e), GFP_NOFS);
82 1671686 : if (!e)
83 : return -ENOMEM;
84 :
85 1671686 : e->next = *eie;
86 1671686 : e->inum = key->objectid;
87 1671686 : e->offset = offset;
88 1671686 : e->num_bytes = data_len;
89 1671686 : *eie = e;
90 :
91 1671686 : return 0;
92 : }
93 :
94 : static void free_inode_elem_list(struct extent_inode_elem *eie)
95 : {
96 20345109 : struct extent_inode_elem *eie_next;
97 :
98 23456583 : for (; eie; eie = eie_next) {
99 1671680 : eie_next = eie->next;
100 1671680 : kfree(eie);
101 : }
102 : }
103 :
104 1378410 : static int find_extent_in_eb(struct btrfs_backref_walk_ctx *ctx,
105 : const struct extent_buffer *eb,
106 : struct extent_inode_elem **eie)
107 : {
108 1378410 : u64 disk_byte;
109 1378410 : struct btrfs_key key;
110 1378410 : struct btrfs_file_extent_item *fi;
111 1378410 : int slot;
112 1378410 : int nritems;
113 1378410 : int extent_type;
114 1378410 : int ret;
115 :
116 : /*
117 : * from the shared data ref, we only have the leaf but we need
118 : * the key. thus, we must look into all items and see that we
119 : * find one (some) with a reference to our extent item.
120 : */
121 1378410 : nritems = btrfs_header_nritems(eb);
122 89336525 : for (slot = 0; slot < nritems; ++slot) {
123 87961176 : btrfs_item_key_to_cpu(eb, &key, slot);
124 87961114 : if (key.type != BTRFS_EXTENT_DATA_KEY)
125 50085213 : continue;
126 37875901 : fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
127 37875932 : extent_type = btrfs_file_extent_type(eb, fi);
128 37875931 : if (extent_type == BTRFS_FILE_EXTENT_INLINE)
129 584814 : continue;
130 : /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
131 37291117 : disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
132 37291152 : if (disk_byte != ctx->bytenr)
133 35683847 : continue;
134 :
135 1607305 : ret = check_extent_in_eb(ctx, &key, eb, fi, eie);
136 1607304 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
137 3063 : return ret;
138 : }
139 :
140 : return 0;
141 : }
142 :
143 : struct preftree {
144 : struct rb_root_cached root;
145 : unsigned int count;
146 : };
147 :
148 : #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
149 :
150 : struct preftrees {
151 : struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
152 : struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
153 : struct preftree indirect_missing_keys;
154 : };
155 :
156 : /*
157 : * Checks for a shared extent during backref search.
158 : *
159 : * The share_count tracks prelim_refs (direct and indirect) having a
160 : * ref->count >0:
161 : * - incremented when a ref->count transitions to >0
162 : * - decremented when a ref->count transitions to <1
163 : */
164 : struct share_check {
165 : struct btrfs_backref_share_check_ctx *ctx;
166 : struct btrfs_root *root;
167 : u64 inum;
168 : u64 data_bytenr;
169 : u64 data_extent_gen;
170 : /*
171 : * Counts number of inodes that refer to an extent (different inodes in
172 : * the same root or different roots) that we could find. The sharedness
173 : * check typically stops once this counter gets greater than 1, so it
174 : * may not reflect the total number of inodes.
175 : */
176 : int share_count;
177 : /*
178 : * The number of times we found our inode refers to the data extent we
179 : * are determining the sharedness. In other words, how many file extent
180 : * items we could find for our inode that point to our target data
181 : * extent. The value we get here after finishing the extent sharedness
182 : * check may be smaller than reality, but if it ends up being greater
183 : * than 1, then we know for sure the inode has multiple file extent
184 : * items that point to our inode, and we can safely assume it's useful
185 : * to cache the sharedness check result.
186 : */
187 : int self_ref_count;
188 : bool have_delayed_delete_refs;
189 : };
190 :
191 : static inline int extent_is_shared(struct share_check *sc)
192 : {
193 1521242 : return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
194 : }
195 :
196 : static struct kmem_cache *btrfs_prelim_ref_cache;
197 :
198 11 : int __init btrfs_prelim_ref_init(void)
199 : {
200 11 : btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
201 : sizeof(struct prelim_ref),
202 : 0,
203 : SLAB_MEM_SPREAD,
204 : NULL);
205 11 : if (!btrfs_prelim_ref_cache)
206 0 : return -ENOMEM;
207 : return 0;
208 : }
209 :
210 0 : void __cold btrfs_prelim_ref_exit(void)
211 : {
212 0 : kmem_cache_destroy(btrfs_prelim_ref_cache);
213 0 : }
214 :
215 : static void free_pref(struct prelim_ref *ref)
216 : {
217 20384671 : kmem_cache_free(btrfs_prelim_ref_cache, ref);
218 : }
219 :
220 : /*
221 : * Return 0 when both refs are for the same block (and can be merged).
222 : * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
223 : * indicates a 'higher' block.
224 : */
225 44605651 : static int prelim_ref_compare(struct prelim_ref *ref1,
226 : struct prelim_ref *ref2)
227 : {
228 44605651 : if (ref1->level < ref2->level)
229 : return -1;
230 44605651 : if (ref1->level > ref2->level)
231 : return 1;
232 44605651 : if (ref1->root_id < ref2->root_id)
233 : return -1;
234 42977506 : if (ref1->root_id > ref2->root_id)
235 : return 1;
236 35147886 : if (ref1->key_for_search.type < ref2->key_for_search.type)
237 : return -1;
238 35147886 : if (ref1->key_for_search.type > ref2->key_for_search.type)
239 : return 1;
240 35147886 : if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
241 : return -1;
242 33689043 : if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
243 : return 1;
244 30829340 : if (ref1->key_for_search.offset < ref2->key_for_search.offset)
245 : return -1;
246 27505570 : if (ref1->key_for_search.offset > ref2->key_for_search.offset)
247 : return 1;
248 26575211 : if (ref1->parent < ref2->parent)
249 : return -1;
250 21250965 : if (ref1->parent > ref2->parent)
251 21250502 : return 1;
252 :
253 : return 0;
254 : }
255 :
256 44126114 : static void update_share_count(struct share_check *sc, int oldcount,
257 : int newcount, struct prelim_ref *newref)
258 : {
259 44126114 : if ((!sc) || (oldcount == 0 && newcount < 1))
260 : return;
261 :
262 1143911 : if (oldcount > 0 && newcount < 1)
263 73 : sc->share_count--;
264 1143838 : else if (oldcount < 1 && newcount > 0)
265 1143432 : sc->share_count++;
266 :
267 1143911 : if (newref->root_id == sc->root->root_key.objectid &&
268 1138462 : newref->wanted_disk_byte == sc->data_bytenr &&
269 1138458 : newref->key_for_search.objectid == sc->inum)
270 1000279 : sc->self_ref_count += newref->count;
271 : }
272 :
273 : /*
274 : * Add @newref to the @root rbtree, merging identical refs.
275 : *
276 : * Callers should assume that newref has been freed after calling.
277 : */
278 44126167 : static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
279 : struct preftree *preftree,
280 : struct prelim_ref *newref,
281 : struct share_check *sc)
282 : {
283 44126167 : struct rb_root_cached *root;
284 44126167 : struct rb_node **p;
285 44126167 : struct rb_node *parent = NULL;
286 44126167 : struct prelim_ref *ref;
287 44126167 : int result;
288 44126167 : bool leftmost = true;
289 :
290 44126167 : root = &preftree->root;
291 44126167 : p = &root->rb_root.rb_node;
292 :
293 85272784 : while (*p) {
294 41147066 : parent = *p;
295 41147066 : ref = rb_entry(parent, struct prelim_ref, rbnode);
296 41147066 : result = prelim_ref_compare(ref, newref);
297 41147066 : if (result < 0) {
298 11606442 : p = &(*p)->rb_left;
299 29540624 : } else if (result > 0) {
300 29540175 : p = &(*p)->rb_right;
301 29540175 : leftmost = false;
302 : } else {
303 : /* Identical refs, merge them and free @newref */
304 449 : struct extent_inode_elem *eie = ref->inode_list;
305 :
306 449 : while (eie && eie->next)
307 : eie = eie->next;
308 :
309 449 : if (!eie)
310 449 : ref->inode_list = newref->inode_list;
311 : else
312 0 : eie->next = newref->inode_list;
313 449 : trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
314 449 : preftree->count);
315 : /*
316 : * A delayed ref can have newref->count < 0.
317 : * The ref->count is updated to follow any
318 : * BTRFS_[ADD|DROP]_DELAYED_REF actions.
319 : */
320 449 : update_share_count(sc, ref->count,
321 449 : ref->count + newref->count, newref);
322 449 : ref->count += newref->count;
323 449 : free_pref(newref);
324 449 : return;
325 : }
326 : }
327 :
328 44125718 : update_share_count(sc, 0, newref->count, newref);
329 44124540 : preftree->count++;
330 44124540 : trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
331 44123293 : rb_link_node(&newref->rbnode, parent, p);
332 44123293 : rb_insert_color_cached(&newref->rbnode, root, leftmost);
333 : }
334 :
335 : /*
336 : * Release the entire tree. We don't care about internal consistency so
337 : * just free everything and then reset the tree root.
338 : */
339 40298177 : static void prelim_release(struct preftree *preftree)
340 : {
341 40298177 : struct prelim_ref *ref, *next_ref;
342 :
343 100915379 : rbtree_postorder_for_each_entry_safe(ref, next_ref,
344 : &preftree->root.rb_root, rbnode) {
345 20319622 : free_inode_elem_list(ref->inode_list);
346 20319559 : free_pref(ref);
347 : }
348 :
349 40298013 : preftree->root = RB_ROOT_CACHED;
350 40298013 : preftree->count = 0;
351 40298013 : }
352 :
353 : /*
354 : * the rules for all callers of this function are:
355 : * - obtaining the parent is the goal
356 : * - if you add a key, you must know that it is a correct key
357 : * - if you cannot add the parent or a correct key, then we will look into the
358 : * block later to set a correct key
359 : *
360 : * delayed refs
361 : * ============
362 : * backref type | shared | indirect | shared | indirect
363 : * information | tree | tree | data | data
364 : * --------------------+--------+----------+--------+----------
365 : * parent logical | y | - | - | -
366 : * key to resolve | - | y | y | y
367 : * tree block logical | - | - | - | -
368 : * root for resolving | y | y | y | y
369 : *
370 : * - column 1: we've the parent -> done
371 : * - column 2, 3, 4: we use the key to find the parent
372 : *
373 : * on disk refs (inline or keyed)
374 : * ==============================
375 : * backref type | shared | indirect | shared | indirect
376 : * information | tree | tree | data | data
377 : * --------------------+--------+----------+--------+----------
378 : * parent logical | y | - | y | -
379 : * key to resolve | - | - | - | y
380 : * tree block logical | y | y | y | y
381 : * root for resolving | - | y | y | y
382 : *
383 : * - column 1, 3: we've the parent -> done
384 : * - column 2: we take the first key from the block to find the parent
385 : * (see add_missing_keys)
386 : * - column 4: we use the key to find the parent
387 : *
388 : * additional information that's available but not required to find the parent
389 : * block might help in merging entries to gain some speed.
390 : */
391 20393503 : static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
392 : struct preftree *preftree, u64 root_id,
393 : const struct btrfs_key *key, int level, u64 parent,
394 : u64 wanted_disk_byte, int count,
395 : struct share_check *sc, gfp_t gfp_mask)
396 : {
397 20393503 : struct prelim_ref *ref;
398 :
399 20393503 : if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
400 : return 0;
401 :
402 20380241 : ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
403 20384572 : if (!ref)
404 : return -ENOMEM;
405 :
406 20384572 : ref->root_id = root_id;
407 20384572 : if (key)
408 1628774 : ref->key_for_search = *key;
409 : else
410 37511596 : memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
411 :
412 20384572 : ref->inode_list = NULL;
413 20384572 : ref->level = level;
414 20384572 : ref->count = count;
415 20384572 : ref->parent = parent;
416 20384572 : ref->wanted_disk_byte = wanted_disk_byte;
417 20384572 : prelim_ref_insert(fs_info, preftree, ref, sc);
418 20383307 : return extent_is_shared(sc);
419 : }
420 :
421 : /* direct refs use root == 0, key == NULL */
422 : static int add_direct_ref(const struct btrfs_fs_info *fs_info,
423 : struct preftrees *preftrees, int level, u64 parent,
424 : u64 wanted_disk_byte, int count,
425 : struct share_check *sc, gfp_t gfp_mask)
426 : {
427 7109621 : return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
428 : parent, wanted_disk_byte, count, sc, gfp_mask);
429 : }
430 :
431 : /* indirect refs use parent == 0 */
432 : static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
433 : struct preftrees *preftrees, u64 root_id,
434 : const struct btrfs_key *key, int level,
435 : u64 wanted_disk_byte, int count,
436 : struct share_check *sc, gfp_t gfp_mask)
437 : {
438 13285618 : struct preftree *tree = &preftrees->indirect;
439 :
440 13285618 : if (!key)
441 22721 : tree = &preftrees->indirect_missing_keys;
442 13285618 : return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
443 : wanted_disk_byte, count, sc, gfp_mask);
444 : }
445 :
446 540467 : static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
447 : {
448 540467 : struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
449 540467 : struct rb_node *parent = NULL;
450 540467 : struct prelim_ref *ref = NULL;
451 540467 : struct prelim_ref target = {};
452 540467 : int result;
453 :
454 540467 : target.parent = bytenr;
455 :
456 3998126 : while (*p) {
457 3457659 : parent = *p;
458 3457659 : ref = rb_entry(parent, struct prelim_ref, rbnode);
459 3457659 : result = prelim_ref_compare(ref, &target);
460 :
461 3457659 : if (result < 0)
462 127823 : p = &(*p)->rb_left;
463 3329836 : else if (result > 0)
464 3329836 : p = &(*p)->rb_right;
465 : else
466 : return 1;
467 : }
468 : return 0;
469 : }
470 :
471 6464979 : static int add_all_parents(struct btrfs_backref_walk_ctx *ctx,
472 : struct btrfs_root *root, struct btrfs_path *path,
473 : struct ulist *parents,
474 : struct preftrees *preftrees, struct prelim_ref *ref,
475 : int level)
476 : {
477 6464979 : int ret = 0;
478 6464979 : int slot;
479 6464979 : struct extent_buffer *eb;
480 6464979 : struct btrfs_key key;
481 6464979 : struct btrfs_key *key_for_search = &ref->key_for_search;
482 6464979 : struct btrfs_file_extent_item *fi;
483 6464979 : struct extent_inode_elem *eie = NULL, *old = NULL;
484 6464979 : u64 disk_byte;
485 6464979 : u64 wanted_disk_byte = ref->wanted_disk_byte;
486 6464979 : u64 count = 0;
487 6464979 : u64 data_offset;
488 6464979 : u8 type;
489 :
490 6464979 : if (level != 0) {
491 5936298 : eb = path->nodes[level];
492 5936298 : ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
493 5936242 : if (ret < 0)
494 : return ret;
495 5936242 : return 0;
496 : }
497 :
498 : /*
499 : * 1. We normally enter this function with the path already pointing to
500 : * the first item to check. But sometimes, we may enter it with
501 : * slot == nritems.
502 : * 2. We are searching for normal backref but bytenr of this leaf
503 : * matches shared data backref
504 : * 3. The leaf owner is not equal to the root we are searching
505 : *
506 : * For these cases, go to the next leaf before we continue.
507 : */
508 528681 : eb = path->nodes[0];
509 1053526 : if (path->slots[0] >= btrfs_header_nritems(eb) ||
510 1049689 : is_shared_data_backref(preftrees, eb->start) ||
511 524845 : ref->root_id != btrfs_header_owner(eb)) {
512 4812 : if (ctx->time_seq == BTRFS_SEQ_LAST)
513 17 : ret = btrfs_next_leaf(root, path);
514 : else
515 4795 : ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
516 : }
517 :
518 1554720 : while (!ret && count < ref->count) {
519 1026213 : eb = path->nodes[0];
520 1026213 : slot = path->slots[0];
521 :
522 1026213 : btrfs_item_key_to_cpu(eb, &key, slot);
523 :
524 1026214 : if (key.objectid != key_for_search->objectid ||
525 1026214 : key.type != BTRFS_EXTENT_DATA_KEY)
526 : break;
527 :
528 : /*
529 : * We are searching for normal backref but bytenr of this leaf
530 : * matches shared data backref, OR
531 : * the leaf owner is not equal to the root we are searching for
532 : */
533 1041836 : if (slot == 0 &&
534 31244 : (is_shared_data_backref(preftrees, eb->start) ||
535 15622 : ref->root_id != btrfs_header_owner(eb))) {
536 0 : if (ctx->time_seq == BTRFS_SEQ_LAST)
537 0 : ret = btrfs_next_leaf(root, path);
538 : else
539 0 : ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
540 0 : continue;
541 : }
542 1026214 : fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
543 1026214 : type = btrfs_file_extent_type(eb, fi);
544 1026213 : if (type == BTRFS_FILE_EXTENT_INLINE)
545 1 : goto next;
546 1026212 : disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
547 1026213 : data_offset = btrfs_file_extent_offset(eb, fi);
548 :
549 1026212 : if (disk_byte == wanted_disk_byte) {
550 664275 : eie = NULL;
551 664275 : old = NULL;
552 664275 : if (ref->key_for_search.offset == key.offset - data_offset)
553 614247 : count++;
554 : else
555 50028 : goto next;
556 614247 : if (!ctx->skip_inode_ref_list) {
557 401435 : ret = check_extent_in_eb(ctx, &key, eb, fi, &eie);
558 401435 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
559 401435 : ret < 0)
560 : break;
561 : }
562 614074 : if (ret > 0)
563 178088 : goto next;
564 435986 : ret = ulist_add_merge_ptr(parents, eb->start,
565 : eie, (void **)&old, GFP_NOFS);
566 435986 : if (ret < 0)
567 : break;
568 435986 : if (!ret && !ctx->skip_inode_ref_list) {
569 5 : while (old->next)
570 0 : old = old->next;
571 5 : old->next = eie;
572 : }
573 435986 : eie = NULL;
574 : }
575 361937 : next:
576 1026040 : if (ctx->time_seq == BTRFS_SEQ_LAST)
577 68394 : ret = btrfs_next_item(root, path);
578 : else
579 957646 : ret = btrfs_next_old_item(root, path, ctx->time_seq);
580 : }
581 :
582 528680 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
583 173 : free_inode_elem_list(eie);
584 : else if (ret > 0)
585 : ret = 0;
586 :
587 : return ret;
588 : }
589 :
590 : /*
591 : * resolve an indirect backref in the form (root_id, key, level)
592 : * to a logical address
593 : */
594 12098025 : static int resolve_indirect_ref(struct btrfs_backref_walk_ctx *ctx,
595 : struct btrfs_path *path,
596 : struct preftrees *preftrees,
597 : struct prelim_ref *ref, struct ulist *parents)
598 : {
599 12098025 : struct btrfs_root *root;
600 12098025 : struct extent_buffer *eb;
601 12098025 : int ret = 0;
602 12098025 : int root_level;
603 12098025 : int level = ref->level;
604 12098025 : struct btrfs_key search_key = ref->key_for_search;
605 :
606 : /*
607 : * If we're search_commit_root we could possibly be holding locks on
608 : * other tree nodes. This happens when qgroups does backref walks when
609 : * adding new delayed refs. To deal with this we need to look in cache
610 : * for the root, and if we don't find it then we need to search the
611 : * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
612 : * here.
613 : */
614 12098025 : if (path->search_commit_root)
615 6432870 : root = btrfs_get_fs_root_commit_root(ctx->fs_info, path, ref->root_id);
616 : else
617 5665155 : root = btrfs_get_fs_root(ctx->fs_info, ref->root_id, false);
618 12099203 : if (IS_ERR(root)) {
619 0 : ret = PTR_ERR(root);
620 0 : goto out_free;
621 : }
622 :
623 17764358 : if (!path->search_commit_root &&
624 5665155 : test_bit(BTRFS_ROOT_DELETING, &root->state)) {
625 4 : ret = -ENOENT;
626 4 : goto out;
627 : }
628 :
629 12099199 : if (btrfs_is_testing(ctx->fs_info)) {
630 : ret = -ENOENT;
631 : goto out;
632 : }
633 :
634 12099199 : if (path->search_commit_root)
635 6434048 : root_level = btrfs_header_level(root->commit_root);
636 5665151 : else if (ctx->time_seq == BTRFS_SEQ_LAST)
637 5140122 : root_level = btrfs_header_level(root->node);
638 : else
639 525029 : root_level = btrfs_old_root_level(root, ctx->time_seq);
640 :
641 12099199 : if (root_level + 1 == level)
642 5634196 : goto out;
643 :
644 : /*
645 : * We can often find data backrefs with an offset that is too large
646 : * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
647 : * subtracting a file's offset with the data offset of its
648 : * corresponding extent data item. This can happen for example in the
649 : * clone ioctl.
650 : *
651 : * So if we detect such case we set the search key's offset to zero to
652 : * make sure we will find the matching file extent item at
653 : * add_all_parents(), otherwise we will miss it because the offset
654 : * taken form the backref is much larger then the offset of the file
655 : * extent item. This can make us scan a very large number of file
656 : * extent items, but at least it will not make us miss any.
657 : *
658 : * This is an ugly workaround for a behaviour that should have never
659 : * existed, but it does and a fix for the clone ioctl would touch a lot
660 : * of places, cause backwards incompatibility and would not fix the
661 : * problem for extents cloned with older kernels.
662 : */
663 6465003 : if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
664 3342491 : search_key.offset >= LLONG_MAX)
665 15637 : search_key.offset = 0;
666 6465003 : path->lowest_level = level;
667 6465003 : if (ctx->time_seq == BTRFS_SEQ_LAST)
668 2649157 : ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
669 : else
670 3815846 : ret = btrfs_search_old_slot(root, &search_key, path, ctx->time_seq);
671 :
672 6464973 : btrfs_debug(ctx->fs_info,
673 : "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
674 : ref->root_id, level, ref->count, ret,
675 : ref->key_for_search.objectid, ref->key_for_search.type,
676 : ref->key_for_search.offset);
677 6464973 : if (ret < 0)
678 0 : goto out;
679 :
680 6464973 : eb = path->nodes[level];
681 6464973 : while (!eb) {
682 1 : if (WARN_ON(!level)) {
683 0 : ret = 1;
684 0 : goto out;
685 : }
686 0 : level--;
687 0 : eb = path->nodes[level];
688 : }
689 :
690 6464972 : ret = add_all_parents(ctx, root, path, parents, preftrees, ref, level);
691 12099112 : out:
692 12099112 : btrfs_put_root(root);
693 12099027 : out_free:
694 12099027 : path->lowest_level = 0;
695 12099027 : btrfs_release_path(path);
696 12098900 : return ret;
697 : }
698 :
699 : static struct extent_inode_elem *
700 : unode_aux_to_inode_list(struct ulist_node *node)
701 : {
702 13538810 : if (!node)
703 : return NULL;
704 6366544 : return (struct extent_inode_elem *)(uintptr_t)node->aux;
705 : }
706 :
707 13306066 : static void free_leaf_list(struct ulist *ulist)
708 : {
709 13306066 : struct ulist_node *node;
710 13306066 : struct ulist_iterator uiter;
711 :
712 13306066 : ULIST_ITER_INIT(&uiter);
713 14745923 : while ((node = ulist_next(ulist, &uiter)))
714 1439853 : free_inode_elem_list(unode_aux_to_inode_list(node));
715 :
716 13306005 : ulist_free(ulist);
717 13306395 : }
718 :
719 : /*
720 : * We maintain three separate rbtrees: one for direct refs, one for
721 : * indirect refs which have a key, and one for indirect refs which do not
722 : * have a key. Each tree does merge on insertion.
723 : *
724 : * Once all of the references are located, we iterate over the tree of
725 : * indirect refs with missing keys. An appropriate key is located and
726 : * the ref is moved onto the tree for indirect refs. After all missing
727 : * keys are thus located, we iterate over the indirect ref tree, resolve
728 : * each reference, and then insert the resolved reference onto the
729 : * direct tree (merging there too).
730 : *
731 : * New backrefs (i.e., for parent nodes) are added to the appropriate
732 : * rbtree as they are encountered. The new backrefs are subsequently
733 : * resolved as above.
734 : */
735 12671428 : static int resolve_indirect_refs(struct btrfs_backref_walk_ctx *ctx,
736 : struct btrfs_path *path,
737 : struct preftrees *preftrees,
738 : struct share_check *sc)
739 : {
740 12671428 : int err;
741 12671428 : int ret = 0;
742 12671428 : struct ulist *parents;
743 12671428 : struct ulist_node *node;
744 12671428 : struct ulist_iterator uiter;
745 12671428 : struct rb_node *rnode;
746 :
747 12671428 : parents = ulist_alloc(GFP_NOFS);
748 12671759 : if (!parents)
749 : return -ENOMEM;
750 :
751 : /*
752 : * We could trade memory usage for performance here by iterating
753 : * the tree, allocating new refs for each insertion, and then
754 : * freeing the entire indirect tree when we're done. In some test
755 : * cases, the tree can grow quite large (~200k objects).
756 : */
757 24770104 : while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
758 12162719 : struct prelim_ref *ref;
759 :
760 12162719 : ref = rb_entry(rnode, struct prelim_ref, rbnode);
761 12162719 : if (WARN(ref->parent,
762 : "BUG: direct ref found in indirect tree")) {
763 0 : ret = -EINVAL;
764 0 : goto out;
765 : }
766 :
767 12162719 : rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
768 12162585 : preftrees->indirect.count--;
769 :
770 12162585 : if (ref->count == 0) {
771 8 : free_pref(ref);
772 8 : continue;
773 : }
774 :
775 12162577 : if (sc && ref->root_id != sc->root->root_key.objectid) {
776 64482 : free_pref(ref);
777 64482 : ret = BACKREF_FOUND_SHARED;
778 64482 : goto out;
779 : }
780 12098095 : err = resolve_indirect_ref(ctx, path, preftrees, ref, parents);
781 : /*
782 : * we can only tolerate ENOENT,otherwise,we should catch error
783 : * and return directly.
784 : */
785 12098876 : if (err == -ENOENT) {
786 4 : prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref,
787 : NULL);
788 4 : continue;
789 12098872 : } else if (err) {
790 173 : free_pref(ref);
791 173 : ret = err;
792 173 : goto out;
793 : }
794 :
795 : /* we put the first parent into the ref at hand */
796 12098699 : ULIST_ITER_INIT(&uiter);
797 12098699 : node = ulist_next(parents, &uiter);
798 12098934 : ref->parent = node ? node->val : 0;
799 12098934 : ref->inode_list = unode_aux_to_inode_list(node);
800 :
801 : /* Add a prelim_ref(s) for any other parent(s). */
802 12098957 : while ((node = ulist_next(parents, &uiter))) {
803 23 : struct prelim_ref *new_ref;
804 :
805 23 : new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
806 : GFP_NOFS);
807 23 : if (!new_ref) {
808 0 : free_pref(ref);
809 0 : ret = -ENOMEM;
810 0 : goto out;
811 : }
812 46 : memcpy(new_ref, ref, sizeof(*ref));
813 23 : new_ref->parent = node->val;
814 23 : new_ref->inode_list = unode_aux_to_inode_list(node);
815 23 : prelim_ref_insert(ctx->fs_info, &preftrees->direct,
816 : new_ref, NULL);
817 : }
818 :
819 : /*
820 : * Now it's a direct ref, put it in the direct tree. We must
821 : * do this last because the ref could be merged/freed here.
822 : */
823 12098584 : prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, NULL);
824 :
825 12098414 : ulist_reinit(parents);
826 12098200 : cond_resched();
827 : }
828 12607385 : out:
829 : /*
830 : * We may have inode lists attached to refs in the parents ulist, so we
831 : * must free them before freeing the ulist and its refs.
832 : */
833 12672040 : free_leaf_list(parents);
834 12672040 : return ret;
835 : }
836 :
837 : /*
838 : * read tree blocks and add keys where required.
839 : */
840 12671700 : static int add_missing_keys(struct btrfs_fs_info *fs_info,
841 : struct preftrees *preftrees, bool lock)
842 : {
843 12671700 : struct prelim_ref *ref;
844 12671700 : struct extent_buffer *eb;
845 12671700 : struct preftree *tree = &preftrees->indirect_missing_keys;
846 12671700 : struct rb_node *node;
847 :
848 24317460 : while ((node = rb_first_cached(&tree->root))) {
849 11645819 : struct btrfs_tree_parent_check check = { 0 };
850 :
851 11645819 : ref = rb_entry(node, struct prelim_ref, rbnode);
852 11645819 : rb_erase_cached(node, &tree->root);
853 :
854 11645116 : BUG_ON(ref->parent); /* should not be a direct ref */
855 11645116 : BUG_ON(ref->key_for_search.type);
856 11645116 : BUG_ON(!ref->wanted_disk_byte);
857 :
858 11645116 : check.level = ref->level - 1;
859 11645116 : check.owner_root = ref->root_id;
860 :
861 11645116 : eb = read_tree_block(fs_info, ref->wanted_disk_byte, &check);
862 11645813 : if (IS_ERR(eb)) {
863 0 : free_pref(ref);
864 0 : return PTR_ERR(eb);
865 : }
866 23291626 : if (!extent_buffer_uptodate(eb)) {
867 0 : free_pref(ref);
868 0 : free_extent_buffer(eb);
869 0 : return -EIO;
870 : }
871 :
872 11645813 : if (lock)
873 302009 : btrfs_tree_read_lock(eb);
874 11645813 : if (btrfs_header_level(eb) == 0)
875 895759 : btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
876 : else
877 10750054 : btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
878 11645698 : if (lock)
879 302009 : btrfs_tree_read_unlock(eb);
880 11645698 : free_extent_buffer(eb);
881 11646196 : prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
882 11645665 : cond_resched();
883 : }
884 : return 0;
885 : }
886 :
887 : /*
888 : * add all currently queued delayed refs from this head whose seq nr is
889 : * smaller or equal that seq to the list
890 : */
891 406004 : static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
892 : struct btrfs_delayed_ref_head *head, u64 seq,
893 : struct preftrees *preftrees, struct share_check *sc)
894 : {
895 406004 : struct btrfs_delayed_ref_node *node;
896 406004 : struct btrfs_key key;
897 406004 : struct rb_node *n;
898 406004 : int count;
899 406004 : int ret = 0;
900 :
901 406004 : spin_lock(&head->lock);
902 1151166 : for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
903 745280 : node = rb_entry(n, struct btrfs_delayed_ref_node,
904 : ref_node);
905 745280 : if (node->seq > seq)
906 580 : continue;
907 :
908 744700 : switch (node->action) {
909 : case BTRFS_ADD_DELAYED_EXTENT:
910 : case BTRFS_UPDATE_DELAYED_HEAD:
911 0 : WARN_ON(1);
912 0 : continue;
913 743234 : case BTRFS_ADD_DELAYED_REF:
914 743234 : count = node->ref_mod;
915 743234 : break;
916 1466 : case BTRFS_DROP_DELAYED_REF:
917 1466 : count = node->ref_mod * -1;
918 1466 : break;
919 0 : default:
920 0 : BUG();
921 : }
922 744700 : switch (node->type) {
923 22721 : case BTRFS_TREE_BLOCK_REF_KEY: {
924 : /* NORMAL INDIRECT METADATA backref */
925 22721 : struct btrfs_delayed_tree_ref *ref;
926 22721 : struct btrfs_key *key_ptr = NULL;
927 :
928 22721 : if (head->extent_op && head->extent_op->update_key) {
929 0 : btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
930 0 : key_ptr = &key;
931 : }
932 :
933 22721 : ref = btrfs_delayed_node_to_tree_ref(node);
934 22721 : ret = add_indirect_ref(fs_info, preftrees, ref->root,
935 22721 : key_ptr, ref->level + 1,
936 : node->bytenr, count, sc,
937 : GFP_ATOMIC);
938 22721 : break;
939 : }
940 : case BTRFS_SHARED_BLOCK_REF_KEY: {
941 : /* SHARED DIRECT METADATA backref */
942 0 : struct btrfs_delayed_tree_ref *ref;
943 :
944 0 : ref = btrfs_delayed_node_to_tree_ref(node);
945 :
946 0 : ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
947 : ref->parent, node->bytenr, count,
948 : sc, GFP_ATOMIC);
949 0 : break;
950 : }
951 : case BTRFS_EXTENT_DATA_REF_KEY: {
952 : /* NORMAL INDIRECT DATA backref */
953 718907 : struct btrfs_delayed_data_ref *ref;
954 718907 : ref = btrfs_delayed_node_to_data_ref(node);
955 :
956 718907 : key.objectid = ref->objectid;
957 718907 : key.type = BTRFS_EXTENT_DATA_KEY;
958 718907 : key.offset = ref->offset;
959 :
960 : /*
961 : * If we have a share check context and a reference for
962 : * another inode, we can't exit immediately. This is
963 : * because even if this is a BTRFS_ADD_DELAYED_REF
964 : * reference we may find next a BTRFS_DROP_DELAYED_REF
965 : * which cancels out this ADD reference.
966 : *
967 : * If this is a DROP reference and there was no previous
968 : * ADD reference, then we need to signal that when we
969 : * process references from the extent tree (through
970 : * add_inline_refs() and add_keyed_refs()), we should
971 : * not exit early if we find a reference for another
972 : * inode, because one of the delayed DROP references
973 : * may cancel that reference in the extent tree.
974 : */
975 718907 : if (sc && count < 0)
976 1506 : sc->have_delayed_delete_refs = true;
977 :
978 718907 : ret = add_indirect_ref(fs_info, preftrees, ref->root,
979 : &key, 0, node->bytenr, count, sc,
980 : GFP_ATOMIC);
981 718907 : break;
982 : }
983 : case BTRFS_SHARED_DATA_REF_KEY: {
984 : /* SHARED DIRECT FULL backref */
985 3072 : struct btrfs_delayed_data_ref *ref;
986 :
987 3072 : ref = btrfs_delayed_node_to_data_ref(node);
988 :
989 3072 : ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
990 : node->bytenr, count, sc,
991 : GFP_ATOMIC);
992 3072 : break;
993 : }
994 : default:
995 0 : WARN_ON(1);
996 : }
997 : /*
998 : * We must ignore BACKREF_FOUND_SHARED until all delayed
999 : * refs have been checked.
1000 : */
1001 744584 : if (ret && (ret != BACKREF_FOUND_SHARED))
1002 : break;
1003 : }
1004 405994 : if (!ret)
1005 399317 : ret = extent_is_shared(sc);
1006 :
1007 405994 : spin_unlock(&head->lock);
1008 405999 : return ret;
1009 : }
1010 :
1011 : /*
1012 : * add all inline backrefs for bytenr to the list
1013 : *
1014 : * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1015 : */
1016 12977125 : static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx,
1017 : struct btrfs_path *path,
1018 : int *info_level, struct preftrees *preftrees,
1019 : struct share_check *sc)
1020 : {
1021 12977125 : int ret = 0;
1022 12977125 : int slot;
1023 12977125 : struct extent_buffer *leaf;
1024 12977125 : struct btrfs_key key;
1025 12977125 : struct btrfs_key found_key;
1026 12977125 : unsigned long ptr;
1027 12977125 : unsigned long end;
1028 12977125 : struct btrfs_extent_item *ei;
1029 12977125 : u64 flags;
1030 12977125 : u64 item_size;
1031 :
1032 : /*
1033 : * enumerate all inline refs
1034 : */
1035 12977125 : leaf = path->nodes[0];
1036 12977125 : slot = path->slots[0];
1037 :
1038 12977125 : item_size = btrfs_item_size(leaf, slot);
1039 12976945 : BUG_ON(item_size < sizeof(*ei));
1040 :
1041 12976945 : ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1042 :
1043 12976961 : if (ctx->check_extent_item) {
1044 2236312 : ret = ctx->check_extent_item(ctx->bytenr, ei, leaf, ctx->user_ctx);
1045 2236310 : if (ret)
1046 : return ret;
1047 : }
1048 :
1049 12954881 : flags = btrfs_extent_flags(leaf, ei);
1050 12954799 : btrfs_item_key_to_cpu(leaf, &found_key, slot);
1051 :
1052 12954762 : ptr = (unsigned long)(ei + 1);
1053 12954762 : end = (unsigned long)ei + item_size;
1054 :
1055 12954762 : if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1056 1207538 : flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1057 0 : struct btrfs_tree_block_info *info;
1058 :
1059 0 : info = (struct btrfs_tree_block_info *)ptr;
1060 0 : *info_level = btrfs_tree_block_level(leaf, info);
1061 0 : ptr += sizeof(struct btrfs_tree_block_info);
1062 0 : BUG_ON(ptr > end);
1063 12954762 : } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
1064 11747224 : *info_level = found_key.offset;
1065 : } else {
1066 1207538 : BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1067 : }
1068 :
1069 31825125 : while (ptr < end) {
1070 18899529 : struct btrfs_extent_inline_ref *iref;
1071 18899529 : u64 offset;
1072 18899529 : int type;
1073 :
1074 18899529 : iref = (struct btrfs_extent_inline_ref *)ptr;
1075 18899529 : type = btrfs_get_extent_inline_ref_type(leaf, iref,
1076 : BTRFS_REF_TYPE_ANY);
1077 18896417 : if (type == BTRFS_REF_TYPE_INVALID)
1078 : return -EUCLEAN;
1079 :
1080 18896417 : offset = btrfs_extent_inline_ref_offset(leaf, iref);
1081 :
1082 18897688 : switch (type) {
1083 5344152 : case BTRFS_SHARED_BLOCK_REF_KEY:
1084 5344152 : ret = add_direct_ref(ctx->fs_info, preftrees,
1085 5344152 : *info_level + 1, offset,
1086 : ctx->bytenr, 1, NULL, GFP_NOFS);
1087 5344152 : break;
1088 1281671 : case BTRFS_SHARED_DATA_REF_KEY: {
1089 1281671 : struct btrfs_shared_data_ref *sdref;
1090 1281671 : int count;
1091 :
1092 1281671 : sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1093 1281671 : count = btrfs_shared_data_ref_count(leaf, sdref);
1094 :
1095 1281665 : ret = add_direct_ref(ctx->fs_info, preftrees, 0, offset,
1096 : ctx->bytenr, count, sc, GFP_NOFS);
1097 1281665 : break;
1098 : }
1099 11621328 : case BTRFS_TREE_BLOCK_REF_KEY:
1100 11621328 : ret = add_indirect_ref(ctx->fs_info, preftrees, offset,
1101 11621328 : NULL, *info_level + 1,
1102 : ctx->bytenr, 1, NULL, GFP_NOFS);
1103 11621328 : break;
1104 650537 : case BTRFS_EXTENT_DATA_REF_KEY: {
1105 650537 : struct btrfs_extent_data_ref *dref;
1106 650537 : int count;
1107 650537 : u64 root;
1108 :
1109 650537 : dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1110 650537 : count = btrfs_extent_data_ref_count(leaf, dref);
1111 650539 : key.objectid = btrfs_extent_data_ref_objectid(leaf,
1112 : dref);
1113 650535 : key.type = BTRFS_EXTENT_DATA_KEY;
1114 650535 : key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1115 :
1116 650606 : if (sc && key.objectid != sc->inum &&
1117 24431 : !sc->have_delayed_delete_refs) {
1118 : ret = BACKREF_FOUND_SHARED;
1119 : break;
1120 : }
1121 :
1122 627698 : root = btrfs_extent_data_ref_root(leaf, dref);
1123 :
1124 640573 : if (!ctx->skip_data_ref ||
1125 12883 : !ctx->skip_data_ref(root, key.objectid, key.offset,
1126 : ctx->user_ctx))
1127 623421 : ret = add_indirect_ref(ctx->fs_info, preftrees,
1128 : root, &key, 0, ctx->bytenr,
1129 : count, sc, GFP_NOFS);
1130 : break;
1131 : }
1132 : default:
1133 0 : WARN_ON(1);
1134 : }
1135 18876794 : if (ret)
1136 29339 : return ret;
1137 20773902 : ptr += btrfs_extent_inline_ref_size(type);
1138 : }
1139 :
1140 : return 0;
1141 : }
1142 :
1143 : /*
1144 : * add all non-inline backrefs for bytenr to the list
1145 : *
1146 : * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1147 : */
1148 12925409 : static int add_keyed_refs(struct btrfs_backref_walk_ctx *ctx,
1149 : struct btrfs_root *extent_root,
1150 : struct btrfs_path *path,
1151 : int info_level, struct preftrees *preftrees,
1152 : struct share_check *sc)
1153 : {
1154 12925409 : struct btrfs_fs_info *fs_info = extent_root->fs_info;
1155 13705616 : int ret;
1156 13705616 : int slot;
1157 13705616 : struct extent_buffer *leaf;
1158 13705616 : struct btrfs_key key;
1159 :
1160 13705616 : while (1) {
1161 13705616 : ret = btrfs_next_item(extent_root, path);
1162 13705597 : if (ret < 0)
1163 : break;
1164 13705597 : if (ret) {
1165 : ret = 0;
1166 : break;
1167 : }
1168 :
1169 13702688 : slot = path->slots[0];
1170 13702688 : leaf = path->nodes[0];
1171 13702688 : btrfs_item_key_to_cpu(leaf, &key, slot);
1172 :
1173 13702320 : if (key.objectid != ctx->bytenr)
1174 : break;
1175 782649 : if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1176 0 : continue;
1177 782649 : if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1178 : break;
1179 :
1180 780212 : switch (key.type) {
1181 0 : case BTRFS_SHARED_BLOCK_REF_KEY:
1182 : /* SHARED DIRECT METADATA backref */
1183 0 : ret = add_direct_ref(fs_info, preftrees,
1184 : info_level + 1, key.offset,
1185 : ctx->bytenr, 1, NULL, GFP_NOFS);
1186 0 : break;
1187 : case BTRFS_SHARED_DATA_REF_KEY: {
1188 : /* SHARED DIRECT FULL backref */
1189 480747 : struct btrfs_shared_data_ref *sdref;
1190 480747 : int count;
1191 :
1192 480747 : sdref = btrfs_item_ptr(leaf, slot,
1193 : struct btrfs_shared_data_ref);
1194 480736 : count = btrfs_shared_data_ref_count(leaf, sdref);
1195 480732 : ret = add_direct_ref(fs_info, preftrees, 0,
1196 : key.offset, ctx->bytenr, count,
1197 : sc, GFP_NOFS);
1198 480732 : break;
1199 : }
1200 0 : case BTRFS_TREE_BLOCK_REF_KEY:
1201 : /* NORMAL INDIRECT METADATA backref */
1202 0 : ret = add_indirect_ref(fs_info, preftrees, key.offset,
1203 : NULL, info_level + 1, ctx->bytenr,
1204 : 1, NULL, GFP_NOFS);
1205 0 : break;
1206 : case BTRFS_EXTENT_DATA_REF_KEY: {
1207 : /* NORMAL INDIRECT DATA backref */
1208 299465 : struct btrfs_extent_data_ref *dref;
1209 299465 : int count;
1210 299465 : u64 root;
1211 :
1212 299465 : dref = btrfs_item_ptr(leaf, slot,
1213 : struct btrfs_extent_data_ref);
1214 299465 : count = btrfs_extent_data_ref_count(leaf, dref);
1215 299465 : key.objectid = btrfs_extent_data_ref_objectid(leaf,
1216 : dref);
1217 299465 : key.type = BTRFS_EXTENT_DATA_KEY;
1218 299465 : key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1219 :
1220 299465 : if (sc && key.objectid != sc->inum &&
1221 1 : !sc->have_delayed_delete_refs) {
1222 : ret = BACKREF_FOUND_SHARED;
1223 : break;
1224 : }
1225 :
1226 299464 : root = btrfs_extent_data_ref_root(leaf, dref);
1227 :
1228 356558 : if (!ctx->skip_data_ref ||
1229 57094 : !ctx->skip_data_ref(root, key.objectid, key.offset,
1230 : ctx->user_ctx))
1231 299241 : ret = add_indirect_ref(fs_info, preftrees, root,
1232 : &key, 0, ctx->bytenr,
1233 : count, sc, GFP_NOFS);
1234 : break;
1235 : }
1236 : default:
1237 0 : WARN_ON(1);
1238 : }
1239 780207 : if (ret)
1240 1 : return ret;
1241 :
1242 : }
1243 :
1244 : return ret;
1245 : }
1246 :
1247 : /*
1248 : * The caller has joined a transaction or is holding a read lock on the
1249 : * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1250 : * snapshot field changing while updating or checking the cache.
1251 : */
1252 1412947 : static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1253 : struct btrfs_root *root,
1254 : u64 bytenr, int level, bool *is_shared)
1255 : {
1256 1412947 : const struct btrfs_fs_info *fs_info = root->fs_info;
1257 1412947 : struct btrfs_backref_shared_cache_entry *entry;
1258 :
1259 1412947 : if (!current->journal_info)
1260 1412947 : lockdep_assert_held(&fs_info->commit_root_sem);
1261 :
1262 1412947 : if (!ctx->use_path_cache)
1263 : return false;
1264 :
1265 1412947 : if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1266 : return false;
1267 :
1268 : /*
1269 : * Level -1 is used for the data extent, which is not reliable to cache
1270 : * because its reference count can increase or decrease without us
1271 : * realizing. We cache results only for extent buffers that lead from
1272 : * the root node down to the leaf with the file extent item.
1273 : */
1274 1412947 : ASSERT(level >= 0);
1275 :
1276 1412947 : entry = &ctx->path_cache_entries[level];
1277 :
1278 : /* Unused cache entry or being used for some other extent buffer. */
1279 1412947 : if (entry->bytenr != bytenr)
1280 : return false;
1281 :
1282 : /*
1283 : * We cached a false result, but the last snapshot generation of the
1284 : * root changed, so we now have a snapshot. Don't trust the result.
1285 : */
1286 603985 : if (!entry->is_shared &&
1287 261726 : entry->gen != btrfs_root_last_snapshot(&root->root_item))
1288 : return false;
1289 :
1290 : /*
1291 : * If we cached a true result and the last generation used for dropping
1292 : * a root changed, we can not trust the result, because the dropped root
1293 : * could be a snapshot sharing this extent buffer.
1294 : */
1295 603985 : if (entry->is_shared &&
1296 342259 : entry->gen != btrfs_get_last_root_drop_gen(fs_info))
1297 : return false;
1298 :
1299 603985 : *is_shared = entry->is_shared;
1300 : /*
1301 : * If the node at this level is shared, than all nodes below are also
1302 : * shared. Currently some of the nodes below may be marked as not shared
1303 : * because we have just switched from one leaf to another, and switched
1304 : * also other nodes above the leaf and below the current level, so mark
1305 : * them as shared.
1306 : */
1307 603985 : if (*is_shared) {
1308 342646 : for (int i = 0; i < level; i++) {
1309 387 : ctx->path_cache_entries[i].is_shared = true;
1310 387 : ctx->path_cache_entries[i].gen = entry->gen;
1311 : }
1312 : }
1313 :
1314 : return true;
1315 : }
1316 :
1317 : /*
1318 : * The caller has joined a transaction or is holding a read lock on the
1319 : * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1320 : * snapshot field changing while updating or checking the cache.
1321 : */
1322 184548 : static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1323 : struct btrfs_root *root,
1324 : u64 bytenr, int level, bool is_shared)
1325 : {
1326 184548 : const struct btrfs_fs_info *fs_info = root->fs_info;
1327 184548 : struct btrfs_backref_shared_cache_entry *entry;
1328 184548 : u64 gen;
1329 :
1330 184548 : if (!current->journal_info)
1331 184548 : lockdep_assert_held(&fs_info->commit_root_sem);
1332 :
1333 184548 : if (!ctx->use_path_cache)
1334 : return;
1335 :
1336 27861 : if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1337 : return;
1338 :
1339 : /*
1340 : * Level -1 is used for the data extent, which is not reliable to cache
1341 : * because its reference count can increase or decrease without us
1342 : * realizing. We cache results only for extent buffers that lead from
1343 : * the root node down to the leaf with the file extent item.
1344 : */
1345 27861 : ASSERT(level >= 0);
1346 :
1347 27861 : if (is_shared)
1348 12007 : gen = btrfs_get_last_root_drop_gen(fs_info);
1349 : else
1350 15854 : gen = btrfs_root_last_snapshot(&root->root_item);
1351 :
1352 27861 : entry = &ctx->path_cache_entries[level];
1353 27861 : entry->bytenr = bytenr;
1354 27861 : entry->is_shared = is_shared;
1355 27861 : entry->gen = gen;
1356 :
1357 : /*
1358 : * If we found an extent buffer is shared, set the cache result for all
1359 : * extent buffers below it to true. As nodes in the path are COWed,
1360 : * their sharedness is moved to their children, and if a leaf is COWed,
1361 : * then the sharedness of a data extent becomes direct, the refcount of
1362 : * data extent is increased in the extent item at the extent tree.
1363 : */
1364 27861 : if (is_shared) {
1365 14518 : for (int i = 0; i < level; i++) {
1366 2511 : entry = &ctx->path_cache_entries[i];
1367 2511 : entry->is_shared = is_shared;
1368 2511 : entry->gen = gen;
1369 : }
1370 : }
1371 : }
1372 :
1373 : /*
1374 : * this adds all existing backrefs (inline backrefs, backrefs and delayed
1375 : * refs) for the given bytenr to the refs list, merges duplicates and resolves
1376 : * indirect refs to their parent bytenr.
1377 : * When roots are found, they're added to the roots list
1378 : *
1379 : * @ctx: Backref walking context object, must be not NULL.
1380 : * @sc: If !NULL, then immediately return BACKREF_FOUND_SHARED when a
1381 : * shared extent is detected.
1382 : *
1383 : * Otherwise this returns 0 for success and <0 for an error.
1384 : *
1385 : * FIXME some caching might speed things up
1386 : */
1387 13433997 : static int find_parent_nodes(struct btrfs_backref_walk_ctx *ctx,
1388 : struct share_check *sc)
1389 : {
1390 13433997 : struct btrfs_root *root = btrfs_extent_root(ctx->fs_info, ctx->bytenr);
1391 13432903 : struct btrfs_key key;
1392 13432903 : struct btrfs_path *path;
1393 13432903 : struct btrfs_delayed_ref_root *delayed_refs = NULL;
1394 13432903 : struct btrfs_delayed_ref_head *head;
1395 13432903 : int info_level = 0;
1396 13432903 : int ret;
1397 13432903 : struct prelim_ref *ref;
1398 13432903 : struct rb_node *node;
1399 13432903 : struct extent_inode_elem *eie = NULL;
1400 13432903 : struct preftrees preftrees = {
1401 : .direct = PREFTREE_INIT,
1402 : .indirect = PREFTREE_INIT,
1403 : .indirect_missing_keys = PREFTREE_INIT
1404 : };
1405 :
1406 : /* Roots ulist is not needed when using a sharedness check context. */
1407 13432903 : if (sc)
1408 : ASSERT(ctx->roots == NULL);
1409 :
1410 13432903 : key.objectid = ctx->bytenr;
1411 13432903 : key.offset = (u64)-1;
1412 13432903 : if (btrfs_fs_incompat(ctx->fs_info, SKINNY_METADATA))
1413 13432903 : key.type = BTRFS_METADATA_ITEM_KEY;
1414 : else
1415 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
1416 :
1417 13432903 : path = btrfs_alloc_path();
1418 13434029 : if (!path)
1419 : return -ENOMEM;
1420 13434029 : if (!ctx->trans) {
1421 7390683 : path->search_commit_root = 1;
1422 7390683 : path->skip_locking = 1;
1423 : }
1424 :
1425 13434029 : if (ctx->time_seq == BTRFS_SEQ_LAST)
1426 4948653 : path->skip_locking = 1;
1427 :
1428 8485376 : again:
1429 13434188 : head = NULL;
1430 :
1431 13434188 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1432 13433338 : if (ret < 0)
1433 0 : goto out;
1434 13433338 : if (ret == 0) {
1435 : /* This shouldn't happen, indicates a bug or fs corruption. */
1436 0 : ASSERT(ret != 0);
1437 0 : ret = -EUCLEAN;
1438 0 : goto out;
1439 : }
1440 :
1441 13433338 : if (ctx->trans && likely(ctx->trans->type != __TRANS_DUMMY) &&
1442 6043483 : ctx->time_seq != BTRFS_SEQ_LAST) {
1443 : /*
1444 : * We have a specific time_seq we care about and trans which
1445 : * means we have the path lock, we need to grab the ref head and
1446 : * lock it so we have a consistent view of the refs at the given
1447 : * time.
1448 : */
1449 1094830 : delayed_refs = &ctx->trans->transaction->delayed_refs;
1450 1094830 : spin_lock(&delayed_refs->lock);
1451 1094945 : head = btrfs_find_delayed_ref_head(delayed_refs, ctx->bytenr);
1452 1094945 : if (head) {
1453 406163 : if (!mutex_trylock(&head->mutex)) {
1454 159 : refcount_inc(&head->refs);
1455 159 : spin_unlock(&delayed_refs->lock);
1456 :
1457 159 : btrfs_release_path(path);
1458 :
1459 : /*
1460 : * Mutex was contended, block until it's
1461 : * released and try again
1462 : */
1463 159 : mutex_lock(&head->mutex);
1464 159 : mutex_unlock(&head->mutex);
1465 159 : btrfs_put_delayed_ref_head(head);
1466 159 : goto again;
1467 : }
1468 406004 : spin_unlock(&delayed_refs->lock);
1469 406004 : ret = add_delayed_refs(ctx->fs_info, head, ctx->time_seq,
1470 : &preftrees, sc);
1471 405996 : mutex_unlock(&head->mutex);
1472 405999 : if (ret)
1473 6677 : goto out;
1474 : } else {
1475 688782 : spin_unlock(&delayed_refs->lock);
1476 : }
1477 : }
1478 :
1479 13426612 : if (path->slots[0]) {
1480 13426527 : struct extent_buffer *leaf;
1481 13426527 : int slot;
1482 :
1483 13426527 : path->slots[0]--;
1484 13426527 : leaf = path->nodes[0];
1485 13426527 : slot = path->slots[0];
1486 13426527 : btrfs_item_key_to_cpu(leaf, &key, slot);
1487 13426241 : if (key.objectid == ctx->bytenr &&
1488 12977555 : (key.type == BTRFS_EXTENT_ITEM_KEY ||
1489 : key.type == BTRFS_METADATA_ITEM_KEY)) {
1490 12977441 : ret = add_inline_refs(ctx, path, &info_level,
1491 : &preftrees, sc);
1492 12976917 : if (ret)
1493 51410 : goto out;
1494 12925507 : ret = add_keyed_refs(ctx, root, path, info_level,
1495 : &preftrees, sc);
1496 12924848 : if (ret)
1497 1 : goto out;
1498 : }
1499 : }
1500 :
1501 : /*
1502 : * If we have a share context and we reached here, it means the extent
1503 : * is not directly shared (no multiple reference items for it),
1504 : * otherwise we would have exited earlier with a return value of
1505 : * BACKREF_FOUND_SHARED after processing delayed references or while
1506 : * processing inline or keyed references from the extent tree.
1507 : * The extent may however be indirectly shared through shared subtrees
1508 : * as a result from creating snapshots, so we determine below what is
1509 : * its parent node, in case we are dealing with a metadata extent, or
1510 : * what's the leaf (or leaves), from a fs tree, that has a file extent
1511 : * item pointing to it in case we are dealing with a data extent.
1512 : */
1513 13373732 : ASSERT(extent_is_shared(sc) == 0);
1514 :
1515 : /*
1516 : * If we are here for a data extent and we have a share_check structure
1517 : * it means the data extent is not directly shared (does not have
1518 : * multiple reference items), so we have to check if a path in the fs
1519 : * tree (going from the root node down to the leaf that has the file
1520 : * extent item pointing to the data extent) is shared, that is, if any
1521 : * of the extent buffers in the path is referenced by other trees.
1522 : */
1523 13373732 : if (sc && ctx->bytenr == sc->data_bytenr) {
1524 : /*
1525 : * If our data extent is from a generation more recent than the
1526 : * last generation used to snapshot the root, then we know that
1527 : * it can not be shared through subtrees, so we can skip
1528 : * resolving indirect references, there's no point in
1529 : * determining the extent buffers for the path from the fs tree
1530 : * root node down to the leaf that has the file extent item that
1531 : * points to the data extent.
1532 : */
1533 772973 : if (sc->data_extent_gen >
1534 772973 : btrfs_root_last_snapshot(&sc->root->root_item)) {
1535 572882 : ret = BACKREF_FOUND_NOT_SHARED;
1536 572882 : goto out;
1537 : }
1538 :
1539 : /*
1540 : * If we are only determining if a data extent is shared or not
1541 : * and the corresponding file extent item is located in the same
1542 : * leaf as the previous file extent item, we can skip resolving
1543 : * indirect references for a data extent, since the fs tree path
1544 : * is the same (same leaf, so same path). We skip as long as the
1545 : * cached result for the leaf is valid and only if there's only
1546 : * one file extent item pointing to the data extent, because in
1547 : * the case of multiple file extent items, they may be located
1548 : * in different leaves and therefore we have multiple paths.
1549 : */
1550 200091 : if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr &&
1551 181129 : sc->self_ref_count == 1) {
1552 180894 : bool cached;
1553 180894 : bool is_shared;
1554 :
1555 180894 : cached = lookup_backref_shared_cache(sc->ctx, sc->root,
1556 : sc->ctx->curr_leaf_bytenr,
1557 : 0, &is_shared);
1558 180894 : if (cached) {
1559 130481 : if (is_shared)
1560 : ret = BACKREF_FOUND_SHARED;
1561 : else
1562 130481 : ret = BACKREF_FOUND_NOT_SHARED;
1563 130481 : goto out;
1564 : }
1565 : }
1566 : }
1567 :
1568 12670369 : btrfs_release_path(path);
1569 :
1570 12671702 : ret = add_missing_keys(ctx->fs_info, &preftrees, path->skip_locking == 0);
1571 12671454 : if (ret)
1572 0 : goto out;
1573 :
1574 12671454 : WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1575 :
1576 12671454 : ret = resolve_indirect_refs(ctx, path, &preftrees, sc);
1577 12672435 : if (ret)
1578 64655 : goto out;
1579 :
1580 12607780 : WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1581 :
1582 : /*
1583 : * This walks the tree of merged and resolved refs. Tree blocks are
1584 : * read in as needed. Unique entries are added to the ulist, and
1585 : * the list of found roots is updated.
1586 : *
1587 : * We release the entire tree in one go before returning.
1588 : */
1589 12607780 : node = rb_first_cached(&preftrees.direct.root);
1590 31587714 : while (node) {
1591 18983746 : ref = rb_entry(node, struct prelim_ref, rbnode);
1592 18983746 : node = rb_next(&ref->rbnode);
1593 : /*
1594 : * ref->count < 0 can happen here if there are delayed
1595 : * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1596 : * prelim_ref_insert() relies on this when merging
1597 : * identical refs to keep the overall count correct.
1598 : * prelim_ref_insert() will merge only those refs
1599 : * which compare identically. Any refs having
1600 : * e.g. different offsets would not be merged,
1601 : * and would retain their original ref->count < 0.
1602 : */
1603 18983518 : if (ctx->roots && ref->count && ref->root_id && ref->parent == 0) {
1604 : /* no parent == root of tree */
1605 5630059 : ret = ulist_add(ctx->roots, ref->root_id, 0, GFP_NOFS);
1606 5629928 : if (ret < 0)
1607 0 : goto out;
1608 : }
1609 18983387 : if (ref->count && ref->parent) {
1610 13251777 : if (!ctx->skip_inode_ref_list && !ref->inode_list &&
1611 1378401 : ref->level == 0) {
1612 1378396 : struct btrfs_tree_parent_check check = { 0 };
1613 1378396 : struct extent_buffer *eb;
1614 :
1615 1378396 : check.level = ref->level;
1616 :
1617 1378396 : eb = read_tree_block(ctx->fs_info, ref->parent,
1618 : &check);
1619 1378408 : if (IS_ERR(eb)) {
1620 0 : ret = PTR_ERR(eb);
1621 3063 : goto out;
1622 : }
1623 2756816 : if (!extent_buffer_uptodate(eb)) {
1624 0 : free_extent_buffer(eb);
1625 0 : ret = -EIO;
1626 0 : goto out;
1627 : }
1628 :
1629 1378408 : if (!path->skip_locking)
1630 0 : btrfs_tree_read_lock(eb);
1631 1378408 : ret = find_extent_in_eb(ctx, eb, &eie);
1632 1378410 : if (!path->skip_locking)
1633 0 : btrfs_tree_read_unlock(eb);
1634 1378410 : free_extent_buffer(eb);
1635 1378414 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1636 1378414 : ret < 0)
1637 3063 : goto out;
1638 1375351 : ref->inode_list = eie;
1639 : /*
1640 : * We transferred the list ownership to the ref,
1641 : * so set to NULL to avoid a double free in case
1642 : * an error happens after this.
1643 : */
1644 1375351 : eie = NULL;
1645 : }
1646 13248732 : ret = ulist_add_merge_ptr(ctx->refs, ref->parent,
1647 13248732 : ref->inode_list,
1648 : (void **)&eie, GFP_NOFS);
1649 13248174 : if (ret < 0)
1650 0 : goto out;
1651 13248174 : if (!ret && !ctx->skip_inode_ref_list) {
1652 : /*
1653 : * We've recorded that parent, so we must extend
1654 : * its inode list here.
1655 : *
1656 : * However if there was corruption we may not
1657 : * have found an eie, return an error in this
1658 : * case.
1659 : */
1660 146552 : ASSERT(eie);
1661 146552 : if (!eie) {
1662 0 : ret = -EUCLEAN;
1663 0 : goto out;
1664 : }
1665 2028260 : while (eie->next)
1666 1881708 : eie = eie->next;
1667 146552 : eie->next = ref->inode_list;
1668 : }
1669 13248174 : eie = NULL;
1670 : /*
1671 : * We have transferred the inode list ownership from
1672 : * this ref to the ref we added to the 'refs' ulist.
1673 : * So set this ref's inode list to NULL to avoid
1674 : * use-after-free when our caller uses it or double
1675 : * frees in case an error happens before we return.
1676 : */
1677 13248174 : ref->inode_list = NULL;
1678 : }
1679 18979784 : cond_resched();
1680 : }
1681 :
1682 12603968 : out:
1683 13433137 : btrfs_free_path(path);
1684 :
1685 13433913 : prelim_release(&preftrees.direct);
1686 13433810 : prelim_release(&preftrees.indirect);
1687 13433069 : prelim_release(&preftrees.indirect_missing_keys);
1688 :
1689 13433287 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
1690 25314 : free_inode_elem_list(eie);
1691 : return ret;
1692 : }
1693 :
1694 : /*
1695 : * Finds all leaves with a reference to the specified combination of
1696 : * @ctx->bytenr and @ctx->extent_item_pos. The bytenr of the found leaves are
1697 : * added to the ulist at @ctx->refs, and that ulist is allocated by this
1698 : * function. The caller should free the ulist with free_leaf_list() if
1699 : * @ctx->ignore_extent_item_pos is false, otherwise a fimple ulist_free() is
1700 : * enough.
1701 : *
1702 : * Returns 0 on success and < 0 on error. On error @ctx->refs is not allocated.
1703 : */
1704 656402 : int btrfs_find_all_leafs(struct btrfs_backref_walk_ctx *ctx)
1705 : {
1706 656402 : int ret;
1707 :
1708 656402 : ASSERT(ctx->refs == NULL);
1709 :
1710 656402 : ctx->refs = ulist_alloc(GFP_NOFS);
1711 656403 : if (!ctx->refs)
1712 : return -ENOMEM;
1713 :
1714 656403 : ret = find_parent_nodes(ctx, NULL);
1715 656400 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1716 653164 : (ret < 0 && ret != -ENOENT)) {
1717 3236 : free_leaf_list(ctx->refs);
1718 3236 : ctx->refs = NULL;
1719 3236 : return ret;
1720 : }
1721 :
1722 : return 0;
1723 : }
1724 :
1725 : /*
1726 : * Walk all backrefs for a given extent to find all roots that reference this
1727 : * extent. Walking a backref means finding all extents that reference this
1728 : * extent and in turn walk the backrefs of those, too. Naturally this is a
1729 : * recursive process, but here it is implemented in an iterative fashion: We
1730 : * find all referencing extents for the extent in question and put them on a
1731 : * list. In turn, we find all referencing extents for those, further appending
1732 : * to the list. The way we iterate the list allows adding more elements after
1733 : * the current while iterating. The process stops when we reach the end of the
1734 : * list.
1735 : *
1736 : * Found roots are added to @ctx->roots, which is allocated by this function if
1737 : * it points to NULL, in which case the caller is responsible for freeing it
1738 : * after it's not needed anymore.
1739 : * This function requires @ctx->refs to be NULL, as it uses it for allocating a
1740 : * ulist to do temporary work, and frees it before returning.
1741 : *
1742 : * Returns 0 on success, < 0 on error.
1743 : */
1744 794741 : static int btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx *ctx)
1745 : {
1746 794741 : const u64 orig_bytenr = ctx->bytenr;
1747 794741 : const bool orig_skip_inode_ref_list = ctx->skip_inode_ref_list;
1748 794741 : bool roots_ulist_allocated = false;
1749 794741 : struct ulist_iterator uiter;
1750 794741 : int ret = 0;
1751 :
1752 794741 : ASSERT(ctx->refs == NULL);
1753 :
1754 794741 : ctx->refs = ulist_alloc(GFP_NOFS);
1755 794740 : if (!ctx->refs)
1756 : return -ENOMEM;
1757 :
1758 794740 : if (!ctx->roots) {
1759 407109 : ctx->roots = ulist_alloc(GFP_NOFS);
1760 407108 : if (!ctx->roots) {
1761 0 : ulist_free(ctx->refs);
1762 0 : ctx->refs = NULL;
1763 0 : return -ENOMEM;
1764 : }
1765 : roots_ulist_allocated = true;
1766 : }
1767 :
1768 794739 : ctx->skip_inode_ref_list = true;
1769 :
1770 794739 : ULIST_ITER_INIT(&uiter);
1771 10988773 : while (1) {
1772 11783502 : struct ulist_node *node;
1773 :
1774 11783502 : ret = find_parent_nodes(ctx, NULL);
1775 11783389 : if (ret < 0 && ret != -ENOENT) {
1776 0 : if (roots_ulist_allocated) {
1777 0 : ulist_free(ctx->roots);
1778 0 : ctx->roots = NULL;
1779 : }
1780 : break;
1781 : }
1782 11783389 : ret = 0;
1783 11783389 : node = ulist_next(ctx->refs, &uiter);
1784 11783400 : if (!node)
1785 : break;
1786 10988713 : ctx->bytenr = node->val;
1787 10988713 : cond_resched();
1788 : }
1789 :
1790 794687 : ulist_free(ctx->refs);
1791 794693 : ctx->refs = NULL;
1792 794693 : ctx->bytenr = orig_bytenr;
1793 794693 : ctx->skip_inode_ref_list = orig_skip_inode_ref_list;
1794 :
1795 794693 : return ret;
1796 : }
1797 :
1798 407110 : int btrfs_find_all_roots(struct btrfs_backref_walk_ctx *ctx,
1799 : bool skip_commit_root_sem)
1800 : {
1801 407110 : int ret;
1802 :
1803 407110 : if (!ctx->trans && !skip_commit_root_sem)
1804 22480 : down_read(&ctx->fs_info->commit_root_sem);
1805 407110 : ret = btrfs_find_all_roots_safe(ctx);
1806 407078 : if (!ctx->trans && !skip_commit_root_sem)
1807 22480 : up_read(&ctx->fs_info->commit_root_sem);
1808 407078 : return ret;
1809 : }
1810 :
1811 49029 : struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void)
1812 : {
1813 49029 : struct btrfs_backref_share_check_ctx *ctx;
1814 :
1815 49029 : ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1816 49029 : if (!ctx)
1817 : return NULL;
1818 :
1819 49029 : ulist_init(&ctx->refs);
1820 :
1821 49029 : return ctx;
1822 : }
1823 :
1824 49020 : void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx)
1825 : {
1826 49020 : if (!ctx)
1827 : return;
1828 :
1829 49020 : ulist_release(&ctx->refs);
1830 49023 : kfree(ctx);
1831 : }
1832 :
1833 : /*
1834 : * Check if a data extent is shared or not.
1835 : *
1836 : * @inode: The inode whose extent we are checking.
1837 : * @bytenr: Logical bytenr of the extent we are checking.
1838 : * @extent_gen: Generation of the extent (file extent item) or 0 if it is
1839 : * not known.
1840 : * @ctx: A backref sharedness check context.
1841 : *
1842 : * btrfs_is_data_extent_shared uses the backref walking code but will short
1843 : * circuit as soon as it finds a root or inode that doesn't match the
1844 : * one passed in. This provides a significant performance benefit for
1845 : * callers (such as fiemap) which want to know whether the extent is
1846 : * shared but do not need a ref count.
1847 : *
1848 : * This attempts to attach to the running transaction in order to account for
1849 : * delayed refs, but continues on even when no running transaction exists.
1850 : *
1851 : * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1852 : */
1853 1173496 : int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr,
1854 : u64 extent_gen,
1855 : struct btrfs_backref_share_check_ctx *ctx)
1856 : {
1857 1173496 : struct btrfs_backref_walk_ctx walk_ctx = { 0 };
1858 1173496 : struct btrfs_root *root = inode->root;
1859 1173496 : struct btrfs_fs_info *fs_info = root->fs_info;
1860 1173496 : struct btrfs_trans_handle *trans;
1861 1173496 : struct ulist_iterator uiter;
1862 1173496 : struct ulist_node *node;
1863 1173496 : struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1864 1173496 : int ret = 0;
1865 1173496 : struct share_check shared = {
1866 : .ctx = ctx,
1867 : .root = root,
1868 : .inum = btrfs_ino(inode),
1869 : .data_bytenr = bytenr,
1870 : .data_extent_gen = extent_gen,
1871 : .share_count = 0,
1872 : .self_ref_count = 0,
1873 : .have_delayed_delete_refs = false,
1874 : };
1875 1173496 : int level;
1876 1173496 : bool leaf_cached;
1877 1173496 : bool leaf_is_shared;
1878 :
1879 10398502 : for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) {
1880 9247572 : if (ctx->prev_extents_cache[i].bytenr == bytenr)
1881 22566 : return ctx->prev_extents_cache[i].is_shared;
1882 : }
1883 :
1884 1150930 : ulist_init(&ctx->refs);
1885 :
1886 1150912 : trans = btrfs_join_transaction_nostart(root);
1887 1150932 : if (IS_ERR(trans)) {
1888 17 : if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1889 0 : ret = PTR_ERR(trans);
1890 0 : goto out;
1891 : }
1892 17 : trans = NULL;
1893 17 : down_read(&fs_info->commit_root_sem);
1894 : } else {
1895 1150915 : btrfs_get_tree_mod_seq(fs_info, &elem);
1896 1150930 : walk_ctx.time_seq = elem.seq;
1897 : }
1898 :
1899 1150947 : ctx->use_path_cache = true;
1900 :
1901 : /*
1902 : * We may have previously determined that the current leaf is shared.
1903 : * If it is, then we have a data extent that is shared due to a shared
1904 : * subtree (caused by snapshotting) and we don't need to check for data
1905 : * backrefs. If the leaf is not shared, then we must do backref walking
1906 : * to determine if the data extent is shared through reflinks.
1907 : */
1908 1150947 : leaf_cached = lookup_backref_shared_cache(ctx, root,
1909 : ctx->curr_leaf_bytenr, 0,
1910 : &leaf_is_shared);
1911 1150937 : if (leaf_cached && leaf_is_shared) {
1912 341872 : ret = 1;
1913 341872 : goto out_trans;
1914 : }
1915 :
1916 809065 : walk_ctx.skip_inode_ref_list = true;
1917 809065 : walk_ctx.trans = trans;
1918 809065 : walk_ctx.fs_info = fs_info;
1919 809065 : walk_ctx.refs = &ctx->refs;
1920 :
1921 : /* -1 means we are in the bytenr of the data extent. */
1922 809065 : level = -1;
1923 809065 : ULIST_ITER_INIT(&uiter);
1924 184548 : while (1) {
1925 993613 : const unsigned long prev_ref_count = ctx->refs.nnodes;
1926 :
1927 993613 : walk_ctx.bytenr = bytenr;
1928 993613 : ret = find_parent_nodes(&walk_ctx, &shared);
1929 993488 : if (ret == BACKREF_FOUND_SHARED ||
1930 : ret == BACKREF_FOUND_NOT_SHARED) {
1931 : /* If shared must return 1, otherwise return 0. */
1932 803818 : ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0;
1933 803818 : if (level >= 0)
1934 64236 : store_backref_shared_cache(ctx, root, bytenr,
1935 : level, ret == 1);
1936 : break;
1937 : }
1938 189670 : if (ret < 0 && ret != -ENOENT)
1939 : break;
1940 189676 : ret = 0;
1941 :
1942 : /*
1943 : * More than one extent buffer (bytenr) may have been added to
1944 : * the ctx->refs ulist, in which case we have to check multiple
1945 : * tree paths in case the first one is not shared, so we can not
1946 : * use the path cache which is made for a single path. Multiple
1947 : * extent buffers at the current level happen when:
1948 : *
1949 : * 1) level -1, the data extent: If our data extent was not
1950 : * directly shared (without multiple reference items), then
1951 : * it might have a single reference item with a count > 1 for
1952 : * the same offset, which means there are 2 (or more) file
1953 : * extent items that point to the data extent - this happens
1954 : * when a file extent item needs to be split and then one
1955 : * item gets moved to another leaf due to a b+tree leaf split
1956 : * when inserting some item. In this case the file extent
1957 : * items may be located in different leaves and therefore
1958 : * some of the leaves may be referenced through shared
1959 : * subtrees while others are not. Since our extent buffer
1960 : * cache only works for a single path (by far the most common
1961 : * case and simpler to deal with), we can not use it if we
1962 : * have multiple leaves (which implies multiple paths).
1963 : *
1964 : * 2) level >= 0, a tree node/leaf: We can have a mix of direct
1965 : * and indirect references on a b+tree node/leaf, so we have
1966 : * to check multiple paths, and the extent buffer (the
1967 : * current bytenr) may be shared or not. One example is
1968 : * during relocation as we may get a shared tree block ref
1969 : * (direct ref) and a non-shared tree block ref (indirect
1970 : * ref) for the same node/leaf.
1971 : */
1972 189676 : if ((ctx->refs.nnodes - prev_ref_count) > 1)
1973 52229 : ctx->use_path_cache = false;
1974 :
1975 189676 : if (level >= 0)
1976 120312 : store_backref_shared_cache(ctx, root, bytenr,
1977 : level, false);
1978 189676 : node = ulist_next(&ctx->refs, &uiter);
1979 189676 : if (!node)
1980 : break;
1981 185567 : bytenr = node->val;
1982 185567 : if (ctx->use_path_cache) {
1983 81109 : bool is_shared;
1984 81109 : bool cached;
1985 :
1986 81109 : level++;
1987 81109 : cached = lookup_backref_shared_cache(ctx, root, bytenr,
1988 : level, &is_shared);
1989 81109 : if (cached) {
1990 1019 : ret = (is_shared ? 1 : 0);
1991 1019 : break;
1992 : }
1993 : }
1994 184548 : shared.share_count = 0;
1995 184548 : shared.have_delayed_delete_refs = false;
1996 184548 : cond_resched();
1997 : }
1998 :
1999 : /*
2000 : * If the path cache is disabled, then it means at some tree level we
2001 : * got multiple parents due to a mix of direct and indirect backrefs or
2002 : * multiple leaves with file extent items pointing to the same data
2003 : * extent. We have to invalidate the cache and cache only the sharedness
2004 : * result for the levels where we got only one node/reference.
2005 : */
2006 808940 : if (!ctx->use_path_cache) {
2007 52229 : int i = 0;
2008 :
2009 52229 : level--;
2010 52229 : if (ret >= 0 && level >= 0) {
2011 0 : bytenr = ctx->path_cache_entries[level].bytenr;
2012 0 : ctx->use_path_cache = true;
2013 0 : store_backref_shared_cache(ctx, root, bytenr, level, ret);
2014 0 : i = level + 1;
2015 : }
2016 :
2017 470061 : for ( ; i < BTRFS_MAX_LEVEL; i++)
2018 417832 : ctx->path_cache_entries[i].bytenr = 0;
2019 : }
2020 :
2021 : /*
2022 : * Cache the sharedness result for the data extent if we know our inode
2023 : * has more than 1 file extent item that refers to the data extent.
2024 : */
2025 808940 : if (ret >= 0 && shared.self_ref_count > 1) {
2026 3900 : int slot = ctx->prev_extents_cache_slot;
2027 :
2028 3900 : ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr;
2029 3900 : ctx->prev_extents_cache[slot].is_shared = (ret == 1);
2030 :
2031 3900 : slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE;
2032 3900 : ctx->prev_extents_cache_slot = slot;
2033 : }
2034 :
2035 805040 : out_trans:
2036 1150812 : if (trans) {
2037 1150795 : btrfs_put_tree_mod_seq(fs_info, &elem);
2038 1150927 : btrfs_end_transaction(trans);
2039 : } else {
2040 17 : up_read(&fs_info->commit_root_sem);
2041 : }
2042 1150910 : out:
2043 1150910 : ulist_release(&ctx->refs);
2044 1150882 : ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr;
2045 :
2046 1150882 : return ret;
2047 : }
2048 :
2049 6633 : int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
2050 : u64 start_off, struct btrfs_path *path,
2051 : struct btrfs_inode_extref **ret_extref,
2052 : u64 *found_off)
2053 : {
2054 6633 : int ret, slot;
2055 6633 : struct btrfs_key key;
2056 6633 : struct btrfs_key found_key;
2057 6633 : struct btrfs_inode_extref *extref;
2058 6633 : const struct extent_buffer *leaf;
2059 6633 : unsigned long ptr;
2060 :
2061 6633 : key.objectid = inode_objectid;
2062 6633 : key.type = BTRFS_INODE_EXTREF_KEY;
2063 6633 : key.offset = start_off;
2064 :
2065 6633 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066 6633 : if (ret < 0)
2067 : return ret;
2068 :
2069 6961 : while (1) {
2070 6797 : leaf = path->nodes[0];
2071 6797 : slot = path->slots[0];
2072 6797 : if (slot >= btrfs_header_nritems(leaf)) {
2073 : /*
2074 : * If the item at offset is not found,
2075 : * btrfs_search_slot will point us to the slot
2076 : * where it should be inserted. In our case
2077 : * that will be the slot directly before the
2078 : * next INODE_REF_KEY_V2 item. In the case
2079 : * that we're pointing to the last slot in a
2080 : * leaf, we must move one leaf over.
2081 : */
2082 174 : ret = btrfs_next_leaf(root, path);
2083 174 : if (ret) {
2084 10 : if (ret >= 1)
2085 10 : ret = -ENOENT;
2086 : break;
2087 : }
2088 164 : continue;
2089 : }
2090 :
2091 6623 : btrfs_item_key_to_cpu(leaf, &found_key, slot);
2092 :
2093 : /*
2094 : * Check that we're still looking at an extended ref key for
2095 : * this particular objectid. If we have different
2096 : * objectid or type then there are no more to be found
2097 : * in the tree and we can exit.
2098 : */
2099 6623 : ret = -ENOENT;
2100 6623 : if (found_key.objectid != inode_objectid)
2101 : break;
2102 6554 : if (found_key.type != BTRFS_INODE_EXTREF_KEY)
2103 : break;
2104 :
2105 0 : ret = 0;
2106 0 : ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2107 0 : extref = (struct btrfs_inode_extref *)ptr;
2108 0 : *ret_extref = extref;
2109 0 : if (found_off)
2110 0 : *found_off = found_key.offset;
2111 : break;
2112 : }
2113 :
2114 : return ret;
2115 : }
2116 :
2117 : /*
2118 : * this iterates to turn a name (from iref/extref) into a full filesystem path.
2119 : * Elements of the path are separated by '/' and the path is guaranteed to be
2120 : * 0-terminated. the path is only given within the current file system.
2121 : * Therefore, it never starts with a '/'. the caller is responsible to provide
2122 : * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
2123 : * the start point of the resulting string is returned. this pointer is within
2124 : * dest, normally.
2125 : * in case the path buffer would overflow, the pointer is decremented further
2126 : * as if output was written to the buffer, though no more output is actually
2127 : * generated. that way, the caller can determine how much space would be
2128 : * required for the path to fit into the buffer. in that case, the returned
2129 : * value will be smaller than dest. callers must check this!
2130 : */
2131 4371 : char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
2132 : u32 name_len, unsigned long name_off,
2133 : struct extent_buffer *eb_in, u64 parent,
2134 : char *dest, u32 size)
2135 : {
2136 4371 : int slot;
2137 4371 : u64 next_inum;
2138 4371 : int ret;
2139 4371 : s64 bytes_left = ((s64)size) - 1;
2140 4371 : struct extent_buffer *eb = eb_in;
2141 4371 : struct btrfs_key found_key;
2142 4371 : struct btrfs_inode_ref *iref;
2143 :
2144 4371 : if (bytes_left >= 0)
2145 4371 : dest[bytes_left] = '\0';
2146 :
2147 32780 : while (1) {
2148 32780 : bytes_left -= name_len;
2149 32780 : if (bytes_left >= 0)
2150 32780 : read_extent_buffer(eb, dest + bytes_left,
2151 : name_off, name_len);
2152 32780 : if (eb != eb_in) {
2153 28409 : if (!path->skip_locking)
2154 17887 : btrfs_tree_read_unlock(eb);
2155 28409 : free_extent_buffer(eb);
2156 : }
2157 32780 : ret = btrfs_find_item(fs_root, path, parent, 0,
2158 : BTRFS_INODE_REF_KEY, &found_key);
2159 32780 : if (ret > 0)
2160 : ret = -ENOENT;
2161 32780 : if (ret)
2162 : break;
2163 :
2164 32780 : next_inum = found_key.offset;
2165 :
2166 : /* regular exit ahead */
2167 32780 : if (parent == next_inum)
2168 : break;
2169 :
2170 28409 : slot = path->slots[0];
2171 28409 : eb = path->nodes[0];
2172 : /* make sure we can use eb after releasing the path */
2173 28409 : if (eb != eb_in) {
2174 28409 : path->nodes[0] = NULL;
2175 28409 : path->locks[0] = 0;
2176 : }
2177 28409 : btrfs_release_path(path);
2178 28409 : iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2179 :
2180 28409 : name_len = btrfs_inode_ref_name_len(eb, iref);
2181 28409 : name_off = (unsigned long)(iref + 1);
2182 :
2183 28409 : parent = next_inum;
2184 28409 : --bytes_left;
2185 28409 : if (bytes_left >= 0)
2186 28409 : dest[bytes_left] = '/';
2187 : }
2188 :
2189 4371 : btrfs_release_path(path);
2190 :
2191 4371 : if (ret)
2192 0 : return ERR_PTR(ret);
2193 :
2194 4371 : return dest + bytes_left;
2195 : }
2196 :
2197 : /*
2198 : * this makes the path point to (logical EXTENT_ITEM *)
2199 : * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
2200 : * tree blocks and <0 on error.
2201 : */
2202 2091 : int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
2203 : struct btrfs_path *path, struct btrfs_key *found_key,
2204 : u64 *flags_ret)
2205 : {
2206 2091 : struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
2207 2091 : int ret;
2208 2091 : u64 flags;
2209 2091 : u64 size = 0;
2210 2091 : u32 item_size;
2211 2091 : const struct extent_buffer *eb;
2212 2091 : struct btrfs_extent_item *ei;
2213 2091 : struct btrfs_key key;
2214 :
2215 2091 : if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2216 2091 : key.type = BTRFS_METADATA_ITEM_KEY;
2217 : else
2218 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
2219 2091 : key.objectid = logical;
2220 2091 : key.offset = (u64)-1;
2221 :
2222 2091 : ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2223 2091 : if (ret < 0)
2224 : return ret;
2225 :
2226 2091 : ret = btrfs_previous_extent_item(extent_root, path, 0);
2227 2091 : if (ret) {
2228 0 : if (ret > 0)
2229 0 : ret = -ENOENT;
2230 0 : return ret;
2231 : }
2232 2091 : btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
2233 2091 : if (found_key->type == BTRFS_METADATA_ITEM_KEY)
2234 0 : size = fs_info->nodesize;
2235 2091 : else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
2236 2091 : size = found_key->offset;
2237 :
2238 2091 : if (found_key->objectid > logical ||
2239 2091 : found_key->objectid + size <= logical) {
2240 : btrfs_debug(fs_info,
2241 : "logical %llu is not within any extent", logical);
2242 : return -ENOENT;
2243 : }
2244 :
2245 2091 : eb = path->nodes[0];
2246 2091 : item_size = btrfs_item_size(eb, path->slots[0]);
2247 2091 : BUG_ON(item_size < sizeof(*ei));
2248 :
2249 2091 : ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
2250 2091 : flags = btrfs_extent_flags(eb, ei);
2251 :
2252 2091 : btrfs_debug(fs_info,
2253 : "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
2254 : logical, logical - found_key->objectid, found_key->objectid,
2255 : found_key->offset, flags, item_size);
2256 :
2257 2091 : WARN_ON(!flags_ret);
2258 2091 : if (flags_ret) {
2259 2091 : if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2260 0 : *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2261 2091 : else if (flags & BTRFS_EXTENT_FLAG_DATA)
2262 2091 : *flags_ret = BTRFS_EXTENT_FLAG_DATA;
2263 : else
2264 0 : BUG();
2265 2091 : return 0;
2266 : }
2267 :
2268 : return -EIO;
2269 : }
2270 :
2271 : /*
2272 : * helper function to iterate extent inline refs. ptr must point to a 0 value
2273 : * for the first call and may be modified. it is used to track state.
2274 : * if more refs exist, 0 is returned and the next call to
2275 : * get_extent_inline_ref must pass the modified ptr parameter to get the
2276 : * next ref. after the last ref was processed, 1 is returned.
2277 : * returns <0 on error
2278 : */
2279 0 : static int get_extent_inline_ref(unsigned long *ptr,
2280 : const struct extent_buffer *eb,
2281 : const struct btrfs_key *key,
2282 : const struct btrfs_extent_item *ei,
2283 : u32 item_size,
2284 : struct btrfs_extent_inline_ref **out_eiref,
2285 : int *out_type)
2286 : {
2287 0 : unsigned long end;
2288 0 : u64 flags;
2289 0 : struct btrfs_tree_block_info *info;
2290 :
2291 0 : if (!*ptr) {
2292 : /* first call */
2293 0 : flags = btrfs_extent_flags(eb, ei);
2294 0 : if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2295 0 : if (key->type == BTRFS_METADATA_ITEM_KEY) {
2296 : /* a skinny metadata extent */
2297 0 : *out_eiref =
2298 0 : (struct btrfs_extent_inline_ref *)(ei + 1);
2299 : } else {
2300 0 : WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2301 0 : info = (struct btrfs_tree_block_info *)(ei + 1);
2302 0 : *out_eiref =
2303 0 : (struct btrfs_extent_inline_ref *)(info + 1);
2304 : }
2305 : } else {
2306 0 : *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2307 : }
2308 0 : *ptr = (unsigned long)*out_eiref;
2309 0 : if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2310 : return -ENOENT;
2311 : }
2312 :
2313 0 : end = (unsigned long)ei + item_size;
2314 0 : *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2315 0 : *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2316 : BTRFS_REF_TYPE_ANY);
2317 0 : if (*out_type == BTRFS_REF_TYPE_INVALID)
2318 : return -EUCLEAN;
2319 :
2320 0 : *ptr += btrfs_extent_inline_ref_size(*out_type);
2321 0 : WARN_ON(*ptr > end);
2322 0 : if (*ptr == end)
2323 0 : return 1; /* last */
2324 :
2325 : return 0;
2326 : }
2327 :
2328 : /*
2329 : * reads the tree block backref for an extent. tree level and root are returned
2330 : * through out_level and out_root. ptr must point to a 0 value for the first
2331 : * call and may be modified (see get_extent_inline_ref comment).
2332 : * returns 0 if data was provided, 1 if there was no more data to provide or
2333 : * <0 on error.
2334 : */
2335 0 : int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2336 : struct btrfs_key *key, struct btrfs_extent_item *ei,
2337 : u32 item_size, u64 *out_root, u8 *out_level)
2338 : {
2339 0 : int ret;
2340 0 : int type;
2341 0 : struct btrfs_extent_inline_ref *eiref;
2342 :
2343 0 : if (*ptr == (unsigned long)-1)
2344 : return 1;
2345 :
2346 0 : while (1) {
2347 0 : ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2348 : &eiref, &type);
2349 0 : if (ret < 0)
2350 0 : return ret;
2351 :
2352 0 : if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2353 : type == BTRFS_SHARED_BLOCK_REF_KEY)
2354 : break;
2355 :
2356 0 : if (ret == 1)
2357 : return 1;
2358 : }
2359 :
2360 : /* we can treat both ref types equally here */
2361 0 : *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2362 :
2363 0 : if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2364 0 : struct btrfs_tree_block_info *info;
2365 :
2366 0 : info = (struct btrfs_tree_block_info *)(ei + 1);
2367 0 : *out_level = btrfs_tree_block_level(eb, info);
2368 : } else {
2369 0 : ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2370 0 : *out_level = (u8)key->offset;
2371 : }
2372 :
2373 0 : if (ret == 1)
2374 0 : *ptr = (unsigned long)-1;
2375 :
2376 : return 0;
2377 : }
2378 :
2379 990382 : static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2380 : struct extent_inode_elem *inode_list,
2381 : u64 root, u64 extent_item_objectid,
2382 : iterate_extent_inodes_t *iterate, void *ctx)
2383 : {
2384 990382 : struct extent_inode_elem *eie;
2385 990382 : int ret = 0;
2386 :
2387 2119566 : for (eie = inode_list; eie; eie = eie->next) {
2388 1130050 : btrfs_debug(fs_info,
2389 : "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2390 : extent_item_objectid, eie->inum,
2391 : eie->offset, root);
2392 1130050 : ret = iterate(eie->inum, eie->offset, eie->num_bytes, root, ctx);
2393 1130050 : if (ret) {
2394 : btrfs_debug(fs_info,
2395 : "stopping iteration for %llu due to ret=%d",
2396 : extent_item_objectid, ret);
2397 : break;
2398 : }
2399 : }
2400 :
2401 990382 : return ret;
2402 : }
2403 :
2404 : /*
2405 : * calls iterate() for every inode that references the extent identified by
2406 : * the given parameters.
2407 : * when the iterator function returns a non-zero value, iteration stops.
2408 : */
2409 633890 : int iterate_extent_inodes(struct btrfs_backref_walk_ctx *ctx,
2410 : bool search_commit_root,
2411 : iterate_extent_inodes_t *iterate, void *user_ctx)
2412 : {
2413 633890 : int ret;
2414 633890 : struct ulist *refs;
2415 633890 : struct ulist_node *ref_node;
2416 633890 : struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2417 633890 : struct ulist_iterator ref_uiter;
2418 :
2419 633890 : btrfs_debug(ctx->fs_info, "resolving all inodes for extent %llu",
2420 : ctx->bytenr);
2421 :
2422 633890 : ASSERT(ctx->trans == NULL);
2423 633890 : ASSERT(ctx->roots == NULL);
2424 :
2425 633890 : if (!search_commit_root) {
2426 2091 : struct btrfs_trans_handle *trans;
2427 :
2428 2091 : trans = btrfs_attach_transaction(ctx->fs_info->tree_root);
2429 2091 : if (IS_ERR(trans)) {
2430 249 : if (PTR_ERR(trans) != -ENOENT &&
2431 : PTR_ERR(trans) != -EROFS)
2432 0 : return PTR_ERR(trans);
2433 : trans = NULL;
2434 : }
2435 2091 : ctx->trans = trans;
2436 : }
2437 :
2438 633890 : if (ctx->trans) {
2439 1842 : btrfs_get_tree_mod_seq(ctx->fs_info, &seq_elem);
2440 1842 : ctx->time_seq = seq_elem.seq;
2441 : } else {
2442 632048 : down_read(&ctx->fs_info->commit_root_sem);
2443 : }
2444 :
2445 633889 : ret = btrfs_find_all_leafs(ctx);
2446 633888 : if (ret)
2447 3236 : goto out;
2448 630652 : refs = ctx->refs;
2449 630652 : ctx->refs = NULL;
2450 :
2451 630652 : ULIST_ITER_INIT(&ref_uiter);
2452 1790041 : while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2453 1159385 : const u64 leaf_bytenr = ref_node->val;
2454 1159385 : struct ulist_node *root_node;
2455 1159385 : struct ulist_iterator root_uiter;
2456 1159385 : struct extent_inode_elem *inode_list;
2457 :
2458 1159385 : inode_list = (struct extent_inode_elem *)(uintptr_t)ref_node->aux;
2459 :
2460 1159385 : if (ctx->cache_lookup) {
2461 1106045 : const u64 *root_ids;
2462 1106045 : int root_count;
2463 1106045 : bool cached;
2464 :
2465 1106045 : cached = ctx->cache_lookup(leaf_bytenr, ctx->user_ctx,
2466 : &root_ids, &root_count);
2467 1106052 : if (cached) {
2468 1219943 : for (int i = 0; i < root_count; i++) {
2469 448183 : ret = iterate_leaf_refs(ctx->fs_info,
2470 : inode_list,
2471 448183 : root_ids[i],
2472 : leaf_bytenr,
2473 : iterate,
2474 : user_ctx);
2475 448182 : if (ret)
2476 : break;
2477 : }
2478 771760 : continue;
2479 : }
2480 : }
2481 :
2482 387631 : if (!ctx->roots) {
2483 234194 : ctx->roots = ulist_alloc(GFP_NOFS);
2484 234194 : if (!ctx->roots) {
2485 : ret = -ENOMEM;
2486 0 : break;
2487 : }
2488 : }
2489 :
2490 387631 : ctx->bytenr = leaf_bytenr;
2491 387631 : ret = btrfs_find_all_roots_safe(ctx);
2492 387626 : if (ret)
2493 : break;
2494 :
2495 387626 : if (ctx->cache_store)
2496 334286 : ctx->cache_store(leaf_bytenr, ctx->roots, ctx->user_ctx);
2497 :
2498 387630 : ULIST_ITER_INIT(&root_uiter);
2499 929834 : while (!ret && (root_node = ulist_next(ctx->roots, &root_uiter))) {
2500 542201 : btrfs_debug(ctx->fs_info,
2501 : "root %llu references leaf %llu, data list %#llx",
2502 : root_node->val, ref_node->val,
2503 : ref_node->aux);
2504 542201 : ret = iterate_leaf_refs(ctx->fs_info, inode_list,
2505 : root_node->val, ctx->bytenr,
2506 : iterate, user_ctx);
2507 : }
2508 387629 : ulist_reinit(ctx->roots);
2509 : }
2510 :
2511 630652 : free_leaf_list(refs);
2512 633887 : out:
2513 633887 : if (ctx->trans) {
2514 1842 : btrfs_put_tree_mod_seq(ctx->fs_info, &seq_elem);
2515 1842 : btrfs_end_transaction(ctx->trans);
2516 1842 : ctx->trans = NULL;
2517 : } else {
2518 632045 : up_read(&ctx->fs_info->commit_root_sem);
2519 : }
2520 :
2521 633885 : ulist_free(ctx->roots);
2522 633882 : ctx->roots = NULL;
2523 :
2524 633882 : if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP)
2525 4102 : ret = 0;
2526 :
2527 : return ret;
2528 : }
2529 :
2530 202858 : static int build_ino_list(u64 inum, u64 offset, u64 num_bytes, u64 root, void *ctx)
2531 : {
2532 202858 : struct btrfs_data_container *inodes = ctx;
2533 202858 : const size_t c = 3 * sizeof(u64);
2534 :
2535 202858 : if (inodes->bytes_left >= c) {
2536 202858 : inodes->bytes_left -= c;
2537 202858 : inodes->val[inodes->elem_cnt] = inum;
2538 202858 : inodes->val[inodes->elem_cnt + 1] = offset;
2539 202858 : inodes->val[inodes->elem_cnt + 2] = root;
2540 202858 : inodes->elem_cnt += 3;
2541 : } else {
2542 0 : inodes->bytes_missing += c - inodes->bytes_left;
2543 0 : inodes->bytes_left = 0;
2544 0 : inodes->elem_missed += 3;
2545 : }
2546 :
2547 202858 : return 0;
2548 : }
2549 :
2550 2091 : int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2551 : struct btrfs_path *path,
2552 : void *ctx, bool ignore_offset)
2553 : {
2554 2091 : struct btrfs_backref_walk_ctx walk_ctx = { 0 };
2555 2091 : int ret;
2556 2091 : u64 flags = 0;
2557 2091 : struct btrfs_key found_key;
2558 2091 : int search_commit_root = path->search_commit_root;
2559 :
2560 2091 : ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2561 2091 : btrfs_release_path(path);
2562 2091 : if (ret < 0)
2563 : return ret;
2564 2091 : if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2565 : return -EINVAL;
2566 :
2567 2091 : walk_ctx.bytenr = found_key.objectid;
2568 2091 : if (ignore_offset)
2569 7 : walk_ctx.ignore_extent_item_pos = true;
2570 : else
2571 2084 : walk_ctx.extent_item_pos = logical - found_key.objectid;
2572 2091 : walk_ctx.fs_info = fs_info;
2573 :
2574 2091 : return iterate_extent_inodes(&walk_ctx, search_commit_root,
2575 : build_ino_list, ctx);
2576 : }
2577 :
2578 : static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2579 : struct extent_buffer *eb, struct inode_fs_paths *ipath);
2580 :
2581 2074 : static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2582 : {
2583 2074 : int ret = 0;
2584 2074 : int slot;
2585 2074 : u32 cur;
2586 2074 : u32 len;
2587 2074 : u32 name_len;
2588 2074 : u64 parent = 0;
2589 2074 : int found = 0;
2590 2074 : struct btrfs_root *fs_root = ipath->fs_root;
2591 2074 : struct btrfs_path *path = ipath->btrfs_path;
2592 2074 : struct extent_buffer *eb;
2593 2074 : struct btrfs_inode_ref *iref;
2594 2074 : struct btrfs_key found_key;
2595 :
2596 4670 : while (!ret) {
2597 4670 : ret = btrfs_find_item(fs_root, path, inum,
2598 : parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2599 : &found_key);
2600 :
2601 4670 : if (ret < 0)
2602 : break;
2603 4670 : if (ret) {
2604 2074 : ret = found ? 0 : -ENOENT;
2605 : break;
2606 : }
2607 2596 : ++found;
2608 :
2609 2596 : parent = found_key.offset;
2610 2596 : slot = path->slots[0];
2611 2596 : eb = btrfs_clone_extent_buffer(path->nodes[0]);
2612 2596 : if (!eb) {
2613 : ret = -ENOMEM;
2614 : break;
2615 : }
2616 2596 : btrfs_release_path(path);
2617 :
2618 2596 : iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2619 :
2620 5192 : for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2621 2596 : name_len = btrfs_inode_ref_name_len(eb, iref);
2622 : /* path must be released before calling iterate()! */
2623 2596 : btrfs_debug(fs_root->fs_info,
2624 : "following ref at offset %u for inode %llu in tree %llu",
2625 : cur, found_key.objectid,
2626 : fs_root->root_key.objectid);
2627 2596 : ret = inode_to_path(parent, name_len,
2628 2596 : (unsigned long)(iref + 1), eb, ipath);
2629 2596 : if (ret)
2630 : break;
2631 2596 : len = sizeof(*iref) + name_len;
2632 2596 : iref = (struct btrfs_inode_ref *)((char *)iref + len);
2633 : }
2634 2596 : free_extent_buffer(eb);
2635 : }
2636 :
2637 2074 : btrfs_release_path(path);
2638 :
2639 2074 : return ret;
2640 : }
2641 :
2642 2074 : static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2643 : {
2644 2074 : int ret;
2645 2074 : int slot;
2646 2074 : u64 offset = 0;
2647 2074 : u64 parent;
2648 2074 : int found = 0;
2649 2074 : struct btrfs_root *fs_root = ipath->fs_root;
2650 2074 : struct btrfs_path *path = ipath->btrfs_path;
2651 2074 : struct extent_buffer *eb;
2652 2074 : struct btrfs_inode_extref *extref;
2653 2074 : u32 item_size;
2654 2074 : u32 cur_offset;
2655 2074 : unsigned long ptr;
2656 :
2657 2074 : while (1) {
2658 2074 : ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2659 : &offset);
2660 2074 : if (ret < 0)
2661 : break;
2662 0 : if (ret) {
2663 0 : ret = found ? 0 : -ENOENT;
2664 : break;
2665 : }
2666 0 : ++found;
2667 :
2668 0 : slot = path->slots[0];
2669 0 : eb = btrfs_clone_extent_buffer(path->nodes[0]);
2670 0 : if (!eb) {
2671 : ret = -ENOMEM;
2672 : break;
2673 : }
2674 0 : btrfs_release_path(path);
2675 :
2676 0 : item_size = btrfs_item_size(eb, slot);
2677 0 : ptr = btrfs_item_ptr_offset(eb, slot);
2678 0 : cur_offset = 0;
2679 :
2680 0 : while (cur_offset < item_size) {
2681 0 : u32 name_len;
2682 :
2683 0 : extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2684 0 : parent = btrfs_inode_extref_parent(eb, extref);
2685 0 : name_len = btrfs_inode_extref_name_len(eb, extref);
2686 0 : ret = inode_to_path(parent, name_len,
2687 0 : (unsigned long)&extref->name, eb, ipath);
2688 0 : if (ret)
2689 : break;
2690 :
2691 0 : cur_offset += btrfs_inode_extref_name_len(eb, extref);
2692 0 : cur_offset += sizeof(*extref);
2693 : }
2694 0 : free_extent_buffer(eb);
2695 :
2696 0 : offset++;
2697 : }
2698 :
2699 2074 : btrfs_release_path(path);
2700 :
2701 2074 : return ret;
2702 : }
2703 :
2704 : /*
2705 : * returns 0 if the path could be dumped (probably truncated)
2706 : * returns <0 in case of an error
2707 : */
2708 2596 : static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2709 : struct extent_buffer *eb, struct inode_fs_paths *ipath)
2710 : {
2711 2596 : char *fspath;
2712 2596 : char *fspath_min;
2713 2596 : int i = ipath->fspath->elem_cnt;
2714 2596 : const int s_ptr = sizeof(char *);
2715 2596 : u32 bytes_left;
2716 :
2717 2596 : bytes_left = ipath->fspath->bytes_left > s_ptr ?
2718 2596 : ipath->fspath->bytes_left - s_ptr : 0;
2719 :
2720 2596 : fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2721 2596 : fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2722 : name_off, eb, inum, fspath_min, bytes_left);
2723 2596 : if (IS_ERR(fspath))
2724 0 : return PTR_ERR(fspath);
2725 :
2726 2596 : if (fspath > fspath_min) {
2727 2596 : ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2728 2596 : ++ipath->fspath->elem_cnt;
2729 2596 : ipath->fspath->bytes_left = fspath - fspath_min;
2730 : } else {
2731 0 : ++ipath->fspath->elem_missed;
2732 0 : ipath->fspath->bytes_missing += fspath_min - fspath;
2733 0 : ipath->fspath->bytes_left = 0;
2734 : }
2735 :
2736 : return 0;
2737 : }
2738 :
2739 : /*
2740 : * this dumps all file system paths to the inode into the ipath struct, provided
2741 : * is has been created large enough. each path is zero-terminated and accessed
2742 : * from ipath->fspath->val[i].
2743 : * when it returns, there are ipath->fspath->elem_cnt number of paths available
2744 : * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2745 : * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2746 : * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2747 : * have been needed to return all paths.
2748 : */
2749 2074 : int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2750 : {
2751 2074 : int ret;
2752 2074 : int found_refs = 0;
2753 :
2754 2074 : ret = iterate_inode_refs(inum, ipath);
2755 2074 : if (!ret)
2756 : ++found_refs;
2757 0 : else if (ret != -ENOENT)
2758 : return ret;
2759 :
2760 2074 : ret = iterate_inode_extrefs(inum, ipath);
2761 2074 : if (ret == -ENOENT && found_refs)
2762 2074 : return 0;
2763 :
2764 : return ret;
2765 : }
2766 :
2767 4165 : struct btrfs_data_container *init_data_container(u32 total_bytes)
2768 : {
2769 4165 : struct btrfs_data_container *data;
2770 4165 : size_t alloc_bytes;
2771 :
2772 4165 : alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2773 4165 : data = kvmalloc(alloc_bytes, GFP_KERNEL);
2774 4165 : if (!data)
2775 : return ERR_PTR(-ENOMEM);
2776 :
2777 4165 : if (total_bytes >= sizeof(*data)) {
2778 4165 : data->bytes_left = total_bytes - sizeof(*data);
2779 4165 : data->bytes_missing = 0;
2780 : } else {
2781 0 : data->bytes_missing = sizeof(*data) - total_bytes;
2782 0 : data->bytes_left = 0;
2783 : }
2784 :
2785 4165 : data->elem_cnt = 0;
2786 4165 : data->elem_missed = 0;
2787 :
2788 4165 : return data;
2789 : }
2790 :
2791 : /*
2792 : * allocates space to return multiple file system paths for an inode.
2793 : * total_bytes to allocate are passed, note that space usable for actual path
2794 : * information will be total_bytes - sizeof(struct inode_fs_paths).
2795 : * the returned pointer must be freed with free_ipath() in the end.
2796 : */
2797 2074 : struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2798 : struct btrfs_path *path)
2799 : {
2800 2074 : struct inode_fs_paths *ifp;
2801 2074 : struct btrfs_data_container *fspath;
2802 :
2803 2074 : fspath = init_data_container(total_bytes);
2804 2074 : if (IS_ERR(fspath))
2805 : return ERR_CAST(fspath);
2806 :
2807 2074 : ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2808 2074 : if (!ifp) {
2809 0 : kvfree(fspath);
2810 0 : return ERR_PTR(-ENOMEM);
2811 : }
2812 :
2813 2074 : ifp->btrfs_path = path;
2814 2074 : ifp->fspath = fspath;
2815 2074 : ifp->fs_root = fs_root;
2816 :
2817 2074 : return ifp;
2818 : }
2819 :
2820 2074 : void free_ipath(struct inode_fs_paths *ipath)
2821 : {
2822 2074 : if (!ipath)
2823 : return;
2824 2074 : kvfree(ipath->fspath);
2825 2074 : kfree(ipath);
2826 : }
2827 :
2828 1954101 : struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info)
2829 : {
2830 1954101 : struct btrfs_backref_iter *ret;
2831 :
2832 1954101 : ret = kzalloc(sizeof(*ret), GFP_NOFS);
2833 1954101 : if (!ret)
2834 : return NULL;
2835 :
2836 1954101 : ret->path = btrfs_alloc_path();
2837 1954101 : if (!ret->path) {
2838 0 : kfree(ret);
2839 0 : return NULL;
2840 : }
2841 :
2842 : /* Current backref iterator only supports iteration in commit root */
2843 1954101 : ret->path->search_commit_root = 1;
2844 1954101 : ret->path->skip_locking = 1;
2845 1954101 : ret->fs_info = fs_info;
2846 :
2847 1954101 : return ret;
2848 : }
2849 :
2850 1973913 : int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2851 : {
2852 1973913 : struct btrfs_fs_info *fs_info = iter->fs_info;
2853 1973913 : struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2854 1973913 : struct btrfs_path *path = iter->path;
2855 1973913 : struct btrfs_extent_item *ei;
2856 1973913 : struct btrfs_key key;
2857 1973913 : int ret;
2858 :
2859 1973913 : key.objectid = bytenr;
2860 1973913 : key.type = BTRFS_METADATA_ITEM_KEY;
2861 1973913 : key.offset = (u64)-1;
2862 1973913 : iter->bytenr = bytenr;
2863 :
2864 1973913 : ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2865 1973913 : if (ret < 0)
2866 : return ret;
2867 1973913 : if (ret == 0) {
2868 0 : ret = -EUCLEAN;
2869 0 : goto release;
2870 : }
2871 1973913 : if (path->slots[0] == 0) {
2872 0 : WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2873 0 : ret = -EUCLEAN;
2874 0 : goto release;
2875 : }
2876 1973913 : path->slots[0]--;
2877 :
2878 1973913 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2879 1973913 : if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2880 1973913 : key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2881 0 : ret = -ENOENT;
2882 0 : goto release;
2883 : }
2884 3947826 : memcpy(&iter->cur_key, &key, sizeof(key));
2885 1973913 : iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2886 : path->slots[0]);
2887 3947826 : iter->end_ptr = (u32)(iter->item_ptr +
2888 1973913 : btrfs_item_size(path->nodes[0], path->slots[0]));
2889 1973913 : ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2890 : struct btrfs_extent_item);
2891 :
2892 : /*
2893 : * Only support iteration on tree backref yet.
2894 : *
2895 : * This is an extra precaution for non skinny-metadata, where
2896 : * EXTENT_ITEM is also used for tree blocks, that we can only use
2897 : * extent flags to determine if it's a tree block.
2898 : */
2899 1973913 : if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2900 0 : ret = -ENOTSUPP;
2901 0 : goto release;
2902 : }
2903 1973913 : iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2904 :
2905 : /* If there is no inline backref, go search for keyed backref */
2906 1973913 : if (iter->cur_ptr >= iter->end_ptr) {
2907 0 : ret = btrfs_next_item(extent_root, path);
2908 :
2909 : /* No inline nor keyed ref */
2910 0 : if (ret > 0) {
2911 0 : ret = -ENOENT;
2912 0 : goto release;
2913 : }
2914 0 : if (ret < 0)
2915 0 : goto release;
2916 :
2917 0 : btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2918 : path->slots[0]);
2919 0 : if (iter->cur_key.objectid != bytenr ||
2920 0 : (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2921 : iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2922 0 : ret = -ENOENT;
2923 0 : goto release;
2924 : }
2925 0 : iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2926 : path->slots[0]);
2927 0 : iter->item_ptr = iter->cur_ptr;
2928 0 : iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2929 0 : path->nodes[0], path->slots[0]));
2930 : }
2931 :
2932 : return 0;
2933 0 : release:
2934 0 : btrfs_backref_iter_release(iter);
2935 0 : return ret;
2936 : }
2937 :
2938 : /*
2939 : * Go to the next backref item of current bytenr, can be either inlined or
2940 : * keyed.
2941 : *
2942 : * Caller needs to check whether it's inline ref or not by iter->cur_key.
2943 : *
2944 : * Return 0 if we get next backref without problem.
2945 : * Return >0 if there is no extra backref for this bytenr.
2946 : * Return <0 if there is something wrong happened.
2947 : */
2948 2673621 : int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2949 : {
2950 2673621 : struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2951 2673621 : struct btrfs_root *extent_root;
2952 2673621 : struct btrfs_path *path = iter->path;
2953 2673621 : struct btrfs_extent_inline_ref *iref;
2954 2673621 : int ret;
2955 2673621 : u32 size;
2956 :
2957 2673621 : if (btrfs_backref_iter_is_inline_ref(iter)) {
2958 : /* We're still inside the inline refs */
2959 2673621 : ASSERT(iter->cur_ptr < iter->end_ptr);
2960 :
2961 2673621 : if (btrfs_backref_has_tree_block_info(iter)) {
2962 : /* First tree block info */
2963 : size = sizeof(struct btrfs_tree_block_info);
2964 : } else {
2965 : /* Use inline ref type to determine the size */
2966 2673621 : int type;
2967 :
2968 2673621 : iref = (struct btrfs_extent_inline_ref *)
2969 2673621 : ((unsigned long)iter->cur_ptr);
2970 2673621 : type = btrfs_extent_inline_ref_type(eb, iref);
2971 :
2972 2673621 : size = btrfs_extent_inline_ref_size(type);
2973 : }
2974 2673621 : iter->cur_ptr += size;
2975 2673621 : if (iter->cur_ptr < iter->end_ptr)
2976 : return 0;
2977 :
2978 : /* All inline items iterated, fall through */
2979 : }
2980 :
2981 : /* We're at keyed items, there is no inline item, go to the next one */
2982 1973913 : extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2983 1973913 : ret = btrfs_next_item(extent_root, iter->path);
2984 1973913 : if (ret)
2985 : return ret;
2986 :
2987 1973885 : btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2988 1973885 : if (iter->cur_key.objectid != iter->bytenr ||
2989 169 : (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2990 : iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2991 : return 1;
2992 0 : iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2993 : path->slots[0]);
2994 0 : iter->cur_ptr = iter->item_ptr;
2995 0 : iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
2996 : path->slots[0]);
2997 0 : return 0;
2998 : }
2999 :
3000 521 : void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
3001 : struct btrfs_backref_cache *cache, int is_reloc)
3002 : {
3003 521 : int i;
3004 :
3005 521 : cache->rb_root = RB_ROOT;
3006 4689 : for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3007 4168 : INIT_LIST_HEAD(&cache->pending[i]);
3008 521 : INIT_LIST_HEAD(&cache->changed);
3009 521 : INIT_LIST_HEAD(&cache->detached);
3010 521 : INIT_LIST_HEAD(&cache->leaves);
3011 521 : INIT_LIST_HEAD(&cache->pending_edge);
3012 521 : INIT_LIST_HEAD(&cache->useless_node);
3013 521 : cache->fs_info = fs_info;
3014 521 : cache->is_reloc = is_reloc;
3015 521 : }
3016 :
3017 2068888 : struct btrfs_backref_node *btrfs_backref_alloc_node(
3018 : struct btrfs_backref_cache *cache, u64 bytenr, int level)
3019 : {
3020 2068888 : struct btrfs_backref_node *node;
3021 :
3022 2068888 : ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
3023 2068888 : node = kzalloc(sizeof(*node), GFP_NOFS);
3024 2068888 : if (!node)
3025 : return node;
3026 :
3027 2068888 : INIT_LIST_HEAD(&node->list);
3028 2068888 : INIT_LIST_HEAD(&node->upper);
3029 2068888 : INIT_LIST_HEAD(&node->lower);
3030 2068888 : RB_CLEAR_NODE(&node->rb_node);
3031 2068888 : cache->nr_nodes++;
3032 2068888 : node->level = level;
3033 2068888 : node->bytenr = bytenr;
3034 :
3035 2068888 : return node;
3036 : }
3037 :
3038 2708202 : struct btrfs_backref_edge *btrfs_backref_alloc_edge(
3039 : struct btrfs_backref_cache *cache)
3040 : {
3041 2708202 : struct btrfs_backref_edge *edge;
3042 :
3043 2708202 : edge = kzalloc(sizeof(*edge), GFP_NOFS);
3044 2708202 : if (edge)
3045 2708202 : cache->nr_edges++;
3046 2708202 : return edge;
3047 : }
3048 :
3049 : /*
3050 : * Drop the backref node from cache, also cleaning up all its
3051 : * upper edges and any uncached nodes in the path.
3052 : *
3053 : * This cleanup happens bottom up, thus the node should either
3054 : * be the lowest node in the cache or a detached node.
3055 : */
3056 2068888 : void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
3057 : struct btrfs_backref_node *node)
3058 : {
3059 2068888 : struct btrfs_backref_node *upper;
3060 2068888 : struct btrfs_backref_edge *edge;
3061 :
3062 2068888 : if (!node)
3063 : return;
3064 :
3065 2068888 : BUG_ON(!node->lowest && !node->detached);
3066 4758476 : while (!list_empty(&node->upper)) {
3067 2689588 : edge = list_entry(node->upper.next, struct btrfs_backref_edge,
3068 : list[LOWER]);
3069 2689588 : upper = edge->node[UPPER];
3070 2689588 : list_del(&edge->list[LOWER]);
3071 2689588 : list_del(&edge->list[UPPER]);
3072 2689588 : btrfs_backref_free_edge(cache, edge);
3073 :
3074 : /*
3075 : * Add the node to leaf node list if no other child block
3076 : * cached.
3077 : */
3078 2689588 : if (list_empty(&upper->lower)) {
3079 2660966 : list_add_tail(&upper->lower, &cache->leaves);
3080 2660966 : upper->lowest = 1;
3081 : }
3082 : }
3083 :
3084 2068888 : btrfs_backref_drop_node(cache, node);
3085 : }
3086 :
3087 : /*
3088 : * Release all nodes/edges from current cache
3089 : */
3090 1135 : void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
3091 : {
3092 1135 : struct btrfs_backref_node *node;
3093 1135 : int i;
3094 :
3095 1290 : while (!list_empty(&cache->detached)) {
3096 155 : node = list_entry(cache->detached.next,
3097 : struct btrfs_backref_node, list);
3098 155 : btrfs_backref_cleanup_node(cache, node);
3099 : }
3100 :
3101 116719 : while (!list_empty(&cache->leaves)) {
3102 115584 : node = list_entry(cache->leaves.next,
3103 : struct btrfs_backref_node, lower);
3104 115584 : btrfs_backref_cleanup_node(cache, node);
3105 : }
3106 :
3107 1135 : cache->last_trans = 0;
3108 :
3109 10215 : for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3110 9080 : ASSERT(list_empty(&cache->pending[i]));
3111 1135 : ASSERT(list_empty(&cache->pending_edge));
3112 1135 : ASSERT(list_empty(&cache->useless_node));
3113 1135 : ASSERT(list_empty(&cache->changed));
3114 1135 : ASSERT(list_empty(&cache->detached));
3115 1135 : ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
3116 1135 : ASSERT(!cache->nr_nodes);
3117 1135 : ASSERT(!cache->nr_edges);
3118 1135 : }
3119 :
3120 : /*
3121 : * Handle direct tree backref
3122 : *
3123 : * Direct tree backref means, the backref item shows its parent bytenr
3124 : * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
3125 : *
3126 : * @ref_key: The converted backref key.
3127 : * For keyed backref, it's the item key.
3128 : * For inlined backref, objectid is the bytenr,
3129 : * type is btrfs_inline_ref_type, offset is
3130 : * btrfs_inline_ref_offset.
3131 : */
3132 2544533 : static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
3133 : struct btrfs_key *ref_key,
3134 : struct btrfs_backref_node *cur)
3135 : {
3136 2544533 : struct btrfs_backref_edge *edge;
3137 2544533 : struct btrfs_backref_node *upper;
3138 2544533 : struct rb_node *rb_node;
3139 :
3140 2544533 : ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
3141 :
3142 : /* Only reloc root uses backref pointing to itself */
3143 2544533 : if (ref_key->objectid == ref_key->offset) {
3144 3 : struct btrfs_root *root;
3145 :
3146 3 : cur->is_reloc_root = 1;
3147 : /* Only reloc backref cache cares about a specific root */
3148 3 : if (cache->is_reloc) {
3149 3 : root = find_reloc_root(cache->fs_info, cur->bytenr);
3150 3 : if (!root)
3151 : return -ENOENT;
3152 3 : cur->root = root;
3153 : } else {
3154 : /*
3155 : * For generic purpose backref cache, reloc root node
3156 : * is useless.
3157 : */
3158 0 : list_add(&cur->list, &cache->useless_node);
3159 : }
3160 3 : return 0;
3161 : }
3162 :
3163 2544530 : edge = btrfs_backref_alloc_edge(cache);
3164 2544530 : if (!edge)
3165 : return -ENOMEM;
3166 :
3167 2544530 : rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
3168 2544530 : if (!rb_node) {
3169 : /* Parent node not yet cached */
3170 19684 : upper = btrfs_backref_alloc_node(cache, ref_key->offset,
3171 19684 : cur->level + 1);
3172 19684 : if (!upper) {
3173 0 : btrfs_backref_free_edge(cache, edge);
3174 0 : return -ENOMEM;
3175 : }
3176 :
3177 : /*
3178 : * Backrefs for the upper level block isn't cached, add the
3179 : * block to pending list
3180 : */
3181 19684 : list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3182 : } else {
3183 : /* Parent node already cached */
3184 2524846 : upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
3185 2524846 : ASSERT(upper->checked);
3186 2524846 : INIT_LIST_HEAD(&edge->list[UPPER]);
3187 : }
3188 2544530 : btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
3189 : return 0;
3190 : }
3191 :
3192 : /*
3193 : * Handle indirect tree backref
3194 : *
3195 : * Indirect tree backref means, we only know which tree the node belongs to.
3196 : * We still need to do a tree search to find out the parents. This is for
3197 : * TREE_BLOCK_REF backref (keyed or inlined).
3198 : *
3199 : * @ref_key: The same as @ref_key in handle_direct_tree_backref()
3200 : * @tree_key: The first key of this tree block.
3201 : * @path: A clean (released) path, to avoid allocating path every time
3202 : * the function get called.
3203 : */
3204 128960 : static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
3205 : struct btrfs_path *path,
3206 : struct btrfs_key *ref_key,
3207 : struct btrfs_key *tree_key,
3208 : struct btrfs_backref_node *cur)
3209 : {
3210 128960 : struct btrfs_fs_info *fs_info = cache->fs_info;
3211 128960 : struct btrfs_backref_node *upper;
3212 128960 : struct btrfs_backref_node *lower;
3213 128960 : struct btrfs_backref_edge *edge;
3214 128960 : struct extent_buffer *eb;
3215 128960 : struct btrfs_root *root;
3216 128960 : struct rb_node *rb_node;
3217 128960 : int level;
3218 128960 : bool need_check = true;
3219 128960 : int ret;
3220 :
3221 128960 : root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
3222 128960 : if (IS_ERR(root))
3223 0 : return PTR_ERR(root);
3224 128960 : if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3225 59175 : cur->cowonly = 1;
3226 :
3227 128960 : if (btrfs_root_level(&root->root_item) == cur->level) {
3228 : /* Tree root */
3229 778 : ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
3230 : /*
3231 : * For reloc backref cache, we may ignore reloc root. But for
3232 : * general purpose backref cache, we can't rely on
3233 : * btrfs_should_ignore_reloc_root() as it may conflict with
3234 : * current running relocation and lead to missing root.
3235 : *
3236 : * For general purpose backref cache, reloc root detection is
3237 : * completely relying on direct backref (key->offset is parent
3238 : * bytenr), thus only do such check for reloc cache.
3239 : */
3240 778 : if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
3241 0 : btrfs_put_root(root);
3242 0 : list_add(&cur->list, &cache->useless_node);
3243 : } else {
3244 778 : cur->root = root;
3245 : }
3246 778 : return 0;
3247 : }
3248 :
3249 128182 : level = cur->level + 1;
3250 :
3251 : /* Search the tree to find parent blocks referring to the block */
3252 128182 : path->search_commit_root = 1;
3253 128182 : path->skip_locking = 1;
3254 128182 : path->lowest_level = level;
3255 128182 : ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
3256 128182 : path->lowest_level = 0;
3257 128182 : if (ret < 0) {
3258 0 : btrfs_put_root(root);
3259 0 : return ret;
3260 : }
3261 128182 : if (ret > 0 && path->slots[level] > 0)
3262 27032 : path->slots[level]--;
3263 :
3264 128182 : eb = path->nodes[level];
3265 128182 : if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
3266 0 : btrfs_err(fs_info,
3267 : "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
3268 : cur->bytenr, level - 1, root->root_key.objectid,
3269 : tree_key->objectid, tree_key->type, tree_key->offset);
3270 0 : btrfs_put_root(root);
3271 0 : ret = -ENOENT;
3272 0 : goto out;
3273 : }
3274 : lower = cur;
3275 :
3276 : /* Add all nodes and edges in the path */
3277 223285 : for (; level < BTRFS_MAX_LEVEL; level++) {
3278 223285 : if (!path->nodes[level]) {
3279 59613 : ASSERT(btrfs_root_bytenr(&root->root_item) ==
3280 : lower->bytenr);
3281 : /* Same as previous should_ignore_reloc_root() call */
3282 59613 : if (btrfs_should_ignore_reloc_root(root) &&
3283 4 : cache->is_reloc) {
3284 4 : btrfs_put_root(root);
3285 4 : list_add(&lower->list, &cache->useless_node);
3286 : } else {
3287 59609 : lower->root = root;
3288 : }
3289 : break;
3290 : }
3291 :
3292 163672 : edge = btrfs_backref_alloc_edge(cache);
3293 163672 : if (!edge) {
3294 0 : btrfs_put_root(root);
3295 0 : ret = -ENOMEM;
3296 0 : goto out;
3297 : }
3298 :
3299 163672 : eb = path->nodes[level];
3300 163672 : rb_node = rb_simple_search(&cache->rb_root, eb->start);
3301 163672 : if (!rb_node) {
3302 95103 : upper = btrfs_backref_alloc_node(cache, eb->start,
3303 95103 : lower->level + 1);
3304 95103 : if (!upper) {
3305 0 : btrfs_put_root(root);
3306 0 : btrfs_backref_free_edge(cache, edge);
3307 0 : ret = -ENOMEM;
3308 0 : goto out;
3309 : }
3310 95103 : upper->owner = btrfs_header_owner(eb);
3311 95103 : if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3312 94178 : upper->cowonly = 1;
3313 :
3314 : /*
3315 : * If we know the block isn't shared we can avoid
3316 : * checking its backrefs.
3317 : */
3318 95103 : if (btrfs_block_can_be_shared(root, eb))
3319 128 : upper->checked = 0;
3320 : else
3321 94975 : upper->checked = 1;
3322 :
3323 : /*
3324 : * Add the block to pending list if we need to check its
3325 : * backrefs, we only do this once while walking up a
3326 : * tree as we will catch anything else later on.
3327 : */
3328 95103 : if (!upper->checked && need_check) {
3329 128 : need_check = false;
3330 128 : list_add_tail(&edge->list[UPPER],
3331 : &cache->pending_edge);
3332 : } else {
3333 94975 : if (upper->checked)
3334 94975 : need_check = true;
3335 94975 : INIT_LIST_HEAD(&edge->list[UPPER]);
3336 : }
3337 : } else {
3338 68569 : upper = rb_entry(rb_node, struct btrfs_backref_node,
3339 : rb_node);
3340 68569 : ASSERT(upper->checked);
3341 68569 : INIT_LIST_HEAD(&edge->list[UPPER]);
3342 68569 : if (!upper->owner)
3343 34 : upper->owner = btrfs_header_owner(eb);
3344 : }
3345 163672 : btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3346 :
3347 163672 : if (rb_node) {
3348 68569 : btrfs_put_root(root);
3349 68569 : break;
3350 : }
3351 95103 : lower = upper;
3352 95103 : upper = NULL;
3353 : }
3354 0 : out:
3355 128182 : btrfs_release_path(path);
3356 128182 : return ret;
3357 : }
3358 :
3359 : /*
3360 : * Add backref node @cur into @cache.
3361 : *
3362 : * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3363 : * links aren't yet bi-directional. Needs to finish such links.
3364 : * Use btrfs_backref_finish_upper_links() to finish such linkage.
3365 : *
3366 : * @path: Released path for indirect tree backref lookup
3367 : * @iter: Released backref iter for extent tree search
3368 : * @node_key: The first key of the tree block
3369 : */
3370 1973913 : int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
3371 : struct btrfs_path *path,
3372 : struct btrfs_backref_iter *iter,
3373 : struct btrfs_key *node_key,
3374 : struct btrfs_backref_node *cur)
3375 : {
3376 1973913 : struct btrfs_fs_info *fs_info = cache->fs_info;
3377 1973913 : struct btrfs_backref_edge *edge;
3378 1973913 : struct btrfs_backref_node *exist;
3379 1973913 : int ret;
3380 :
3381 1973913 : ret = btrfs_backref_iter_start(iter, cur->bytenr);
3382 1973913 : if (ret < 0)
3383 : return ret;
3384 : /*
3385 : * We skip the first btrfs_tree_block_info, as we don't use the key
3386 : * stored in it, but fetch it from the tree block
3387 : */
3388 1973913 : if (btrfs_backref_has_tree_block_info(iter)) {
3389 0 : ret = btrfs_backref_iter_next(iter);
3390 0 : if (ret < 0)
3391 0 : goto out;
3392 : /* No extra backref? This means the tree block is corrupted */
3393 0 : if (ret > 0) {
3394 0 : ret = -EUCLEAN;
3395 0 : goto out;
3396 : }
3397 : }
3398 1973913 : WARN_ON(cur->checked);
3399 1973913 : if (!list_empty(&cur->upper)) {
3400 : /*
3401 : * The backref was added previously when processing backref of
3402 : * type BTRFS_TREE_BLOCK_REF_KEY
3403 : */
3404 128 : ASSERT(list_is_singular(&cur->upper));
3405 128 : edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3406 : list[LOWER]);
3407 128 : ASSERT(list_empty(&edge->list[UPPER]));
3408 128 : exist = edge->node[UPPER];
3409 : /*
3410 : * Add the upper level block to pending list if we need check
3411 : * its backrefs
3412 : */
3413 128 : if (!exist->checked)
3414 0 : list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3415 : } else {
3416 : exist = NULL;
3417 : }
3418 :
3419 4647534 : for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3420 2673621 : struct extent_buffer *eb;
3421 2673621 : struct btrfs_key key;
3422 2673621 : int type;
3423 :
3424 2673621 : cond_resched();
3425 2673621 : eb = btrfs_backref_get_eb(iter);
3426 :
3427 2673621 : key.objectid = iter->bytenr;
3428 2673621 : if (btrfs_backref_iter_is_inline_ref(iter)) {
3429 2673621 : struct btrfs_extent_inline_ref *iref;
3430 :
3431 : /* Update key for inline backref */
3432 2673621 : iref = (struct btrfs_extent_inline_ref *)
3433 2673621 : ((unsigned long)iter->cur_ptr);
3434 2673621 : type = btrfs_get_extent_inline_ref_type(eb, iref,
3435 : BTRFS_REF_TYPE_BLOCK);
3436 2673621 : if (type == BTRFS_REF_TYPE_INVALID) {
3437 0 : ret = -EUCLEAN;
3438 0 : goto out;
3439 : }
3440 2673621 : key.type = type;
3441 2673621 : key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3442 : } else {
3443 0 : key.type = iter->cur_key.type;
3444 0 : key.offset = iter->cur_key.offset;
3445 : }
3446 :
3447 : /*
3448 : * Parent node found and matches current inline ref, no need to
3449 : * rebuild this node for this inline ref
3450 : */
3451 2673621 : if (exist &&
3452 188 : ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3453 188 : exist->owner == key.offset) ||
3454 0 : (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3455 0 : exist->bytenr == key.offset))) {
3456 128 : exist = NULL;
3457 2544661 : continue;
3458 : }
3459 :
3460 : /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3461 2673493 : if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3462 2544533 : ret = handle_direct_tree_backref(cache, &key, cur);
3463 2544533 : if (ret < 0)
3464 0 : goto out;
3465 2544533 : continue;
3466 128960 : } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3467 0 : ret = -EINVAL;
3468 0 : btrfs_print_v0_err(fs_info);
3469 0 : btrfs_handle_fs_error(fs_info, ret, NULL);
3470 0 : goto out;
3471 128960 : } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
3472 0 : continue;
3473 : }
3474 :
3475 : /*
3476 : * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
3477 : * means the root objectid. We need to search the tree to get
3478 : * its parent bytenr.
3479 : */
3480 128960 : ret = handle_indirect_tree_backref(cache, path, &key, node_key,
3481 : cur);
3482 128960 : if (ret < 0)
3483 0 : goto out;
3484 : }
3485 1973913 : ret = 0;
3486 1973913 : cur->checked = 1;
3487 1973913 : WARN_ON(exist);
3488 1973913 : out:
3489 1973913 : btrfs_backref_iter_release(iter);
3490 1973913 : return ret;
3491 : }
3492 :
3493 : /*
3494 : * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3495 : */
3496 1954101 : int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3497 : struct btrfs_backref_node *start)
3498 : {
3499 1954101 : struct list_head *useless_node = &cache->useless_node;
3500 1954101 : struct btrfs_backref_edge *edge;
3501 1954101 : struct rb_node *rb_node;
3502 1954101 : LIST_HEAD(pending_edge);
3503 :
3504 1954101 : ASSERT(start->checked);
3505 :
3506 : /* Insert this node to cache if it's not COW-only */
3507 1954101 : if (!start->cowonly) {
3508 1894926 : rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3509 : &start->rb_node);
3510 1894926 : if (rb_node)
3511 0 : btrfs_backref_panic(cache->fs_info, start->bytenr,
3512 : -EEXIST);
3513 1894926 : list_add_tail(&start->lower, &cache->leaves);
3514 : }
3515 :
3516 : /*
3517 : * Use breadth first search to iterate all related edges.
3518 : *
3519 : * The starting points are all the edges of this node
3520 : */
3521 4599350 : list_for_each_entry(edge, &start->upper, list[LOWER])
3522 2645249 : list_add_tail(&edge->list[UPPER], &pending_edge);
3523 :
3524 4662303 : while (!list_empty(&pending_edge)) {
3525 2708202 : struct btrfs_backref_node *upper;
3526 2708202 : struct btrfs_backref_node *lower;
3527 :
3528 2708202 : edge = list_first_entry(&pending_edge,
3529 : struct btrfs_backref_edge, list[UPPER]);
3530 2708202 : list_del_init(&edge->list[UPPER]);
3531 2708202 : upper = edge->node[UPPER];
3532 2708202 : lower = edge->node[LOWER];
3533 :
3534 : /* Parent is detached, no need to keep any edges */
3535 2708202 : if (upper->detached) {
3536 18459 : list_del(&edge->list[LOWER]);
3537 18459 : btrfs_backref_free_edge(cache, edge);
3538 :
3539 : /* Lower node is orphan, queue for cleanup */
3540 18459 : if (list_empty(&lower->upper))
3541 149 : list_add(&lower->list, useless_node);
3542 18459 : continue;
3543 : }
3544 :
3545 : /*
3546 : * All new nodes added in current build_backref_tree() haven't
3547 : * been linked to the cache rb tree.
3548 : * So if we have upper->rb_node populated, this means a cache
3549 : * hit. We only need to link the edge, as @upper and all its
3550 : * parents have already been linked.
3551 : */
3552 2689743 : if (!RB_EMPTY_NODE(&upper->rb_node)) {
3553 2574956 : if (upper->lowest) {
3554 2546334 : list_del_init(&upper->lower);
3555 2546334 : upper->lowest = 0;
3556 : }
3557 :
3558 2574956 : list_add_tail(&edge->list[UPPER], &upper->lower);
3559 2574956 : continue;
3560 : }
3561 :
3562 : /* Sanity check, we shouldn't have any unchecked nodes */
3563 114787 : if (!upper->checked) {
3564 : ASSERT(0);
3565 : return -EUCLEAN;
3566 : }
3567 :
3568 : /* Sanity check, COW-only node has non-COW-only parent */
3569 114787 : if (start->cowonly != upper->cowonly) {
3570 : ASSERT(0);
3571 : return -EUCLEAN;
3572 : }
3573 :
3574 : /* Only cache non-COW-only (subvolume trees) tree blocks */
3575 114787 : if (!upper->cowonly) {
3576 20609 : rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3577 : &upper->rb_node);
3578 20609 : if (rb_node) {
3579 0 : btrfs_backref_panic(cache->fs_info,
3580 : upper->bytenr, -EEXIST);
3581 : return -EUCLEAN;
3582 : }
3583 : }
3584 :
3585 114787 : list_add_tail(&edge->list[UPPER], &upper->lower);
3586 :
3587 : /*
3588 : * Also queue all the parent edges of this uncached node
3589 : * to finish the upper linkage
3590 : */
3591 177740 : list_for_each_entry(edge, &upper->upper, list[LOWER])
3592 62953 : list_add_tail(&edge->list[UPPER], &pending_edge);
3593 : }
3594 : return 0;
3595 : }
3596 :
3597 0 : void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3598 : struct btrfs_backref_node *node)
3599 : {
3600 0 : struct btrfs_backref_node *lower;
3601 0 : struct btrfs_backref_node *upper;
3602 0 : struct btrfs_backref_edge *edge;
3603 :
3604 0 : while (!list_empty(&cache->useless_node)) {
3605 0 : lower = list_first_entry(&cache->useless_node,
3606 : struct btrfs_backref_node, list);
3607 0 : list_del_init(&lower->list);
3608 : }
3609 0 : while (!list_empty(&cache->pending_edge)) {
3610 0 : edge = list_first_entry(&cache->pending_edge,
3611 : struct btrfs_backref_edge, list[UPPER]);
3612 0 : list_del(&edge->list[UPPER]);
3613 0 : list_del(&edge->list[LOWER]);
3614 0 : lower = edge->node[LOWER];
3615 0 : upper = edge->node[UPPER];
3616 0 : btrfs_backref_free_edge(cache, edge);
3617 :
3618 : /*
3619 : * Lower is no longer linked to any upper backref nodes and
3620 : * isn't in the cache, we can free it ourselves.
3621 : */
3622 0 : if (list_empty(&lower->upper) &&
3623 0 : RB_EMPTY_NODE(&lower->rb_node))
3624 0 : list_add(&lower->list, &cache->useless_node);
3625 :
3626 0 : if (!RB_EMPTY_NODE(&upper->rb_node))
3627 0 : continue;
3628 :
3629 : /* Add this guy's upper edges to the list to process */
3630 0 : list_for_each_entry(edge, &upper->upper, list[LOWER])
3631 0 : list_add_tail(&edge->list[UPPER],
3632 : &cache->pending_edge);
3633 0 : if (list_empty(&upper->upper))
3634 0 : list_add(&upper->list, &cache->useless_node);
3635 : }
3636 :
3637 0 : while (!list_empty(&cache->useless_node)) {
3638 0 : lower = list_first_entry(&cache->useless_node,
3639 : struct btrfs_backref_node, list);
3640 0 : list_del_init(&lower->list);
3641 0 : if (lower == node)
3642 0 : node = NULL;
3643 0 : btrfs_backref_drop_node(cache, lower);
3644 : }
3645 :
3646 0 : btrfs_backref_cleanup_node(cache, node);
3647 0 : ASSERT(list_empty(&cache->useless_node) &&
3648 : list_empty(&cache->pending_edge));
3649 0 : }
|
