1// SPDX-License-Identifier: GPL-2.0
2
3#include "bcachefs.h"
4#include "bkey_buf.h"
5#include "bkey_methods.h"
6#include "btree_update.h"
7#include "extents.h"
8#include "dirent.h"
9#include "fs.h"
10#include "keylist.h"
11#include "str_hash.h"
12#include "subvolume.h"
13
14#include <linux/dcache.h>
15
16static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
17{
18 unsigned bkey_u64s = bkey_val_u64s(d.k);
19 unsigned bkey_bytes = bkey_u64s * sizeof(u64);
20 u64 last_u64 = ((u64*)d.v)[bkey_u64s - 1];
21#if CPU_BIG_ENDIAN
22 unsigned trailing_nuls = last_u64 ? __builtin_ctzll(last_u64) / 8 : 64 / 8;
23#else
24 unsigned trailing_nuls = last_u64 ? __builtin_clzll(last_u64) / 8 : 64 / 8;
25#endif
26
27 return bkey_bytes -
28 offsetof(struct bch_dirent, d_name) -
29 trailing_nuls;
30}
31
32struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
33{
34 return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
35}
36
37static u64 bch2_dirent_hash(const struct bch_hash_info *info,
38 const struct qstr *name)
39{
40 struct bch_str_hash_ctx ctx;
41
42 bch2_str_hash_init(ctx: &ctx, info);
43 bch2_str_hash_update(ctx: &ctx, info, data: name->name, len: name->len);
44
45 /* [0,2) reserved for dots */
46 return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
47}
48
49static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
50{
51 return bch2_dirent_hash(info, name: key);
52}
53
54static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
55{
56 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
57 struct qstr name = bch2_dirent_get_name(d);
58
59 return bch2_dirent_hash(info, name: &name);
60}
61
62static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
63{
64 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(k: _l);
65 const struct qstr l_name = bch2_dirent_get_name(d: l);
66 const struct qstr *r_name = _r;
67
68 return !qstr_eq(l: l_name, r: *r_name);
69}
70
71static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
72{
73 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(k: _l);
74 struct bkey_s_c_dirent r = bkey_s_c_to_dirent(k: _r);
75 const struct qstr l_name = bch2_dirent_get_name(d: l);
76 const struct qstr r_name = bch2_dirent_get_name(d: r);
77
78 return !qstr_eq(l: l_name, r: r_name);
79}
80
81static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
82{
83 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
84
85 if (d.v->d_type == DT_SUBVOL)
86 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
87 return true;
88}
89
90const struct bch_hash_desc bch2_dirent_hash_desc = {
91 .btree_id = BTREE_ID_dirents,
92 .key_type = KEY_TYPE_dirent,
93 .hash_key = dirent_hash_key,
94 .hash_bkey = dirent_hash_bkey,
95 .cmp_key = dirent_cmp_key,
96 .cmp_bkey = dirent_cmp_bkey,
97 .is_visible = dirent_is_visible,
98};
99
100int bch2_dirent_invalid(struct bch_fs *c, struct bkey_s_c k,
101 enum bkey_invalid_flags flags,
102 struct printbuf *err)
103{
104 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
105 struct qstr d_name = bch2_dirent_get_name(d);
106 int ret = 0;
107
108 bkey_fsck_err_on(!d_name.len, c, err,
109 dirent_empty_name,
110 "empty name");
111
112 bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len), c, err,
113 dirent_val_too_big,
114 "value too big (%zu > %u)",
115 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
116
117 /*
118 * Check new keys don't exceed the max length
119 * (older keys may be larger.)
120 */
121 bkey_fsck_err_on((flags & BKEY_INVALID_COMMIT) && d_name.len > BCH_NAME_MAX, c, err,
122 dirent_name_too_long,
123 "dirent name too big (%u > %u)",
124 d_name.len, BCH_NAME_MAX);
125
126 bkey_fsck_err_on(d_name.len != strnlen(d_name.name, d_name.len), c, err,
127 dirent_name_embedded_nul,
128 "dirent has stray data after name's NUL");
129
130 bkey_fsck_err_on((d_name.len == 1 && !memcmp(d_name.name, ".", 1)) ||
131 (d_name.len == 2 && !memcmp(d_name.name, "..", 2)), c, err,
132 dirent_name_dot_or_dotdot,
133 "invalid name");
134
135 bkey_fsck_err_on(memchr(d_name.name, '/', d_name.len), c, err,
136 dirent_name_has_slash,
137 "name with /");
138
139 bkey_fsck_err_on(d.v->d_type != DT_SUBVOL &&
140 le64_to_cpu(d.v->d_inum) == d.k->p.inode, c, err,
141 dirent_to_itself,
142 "dirent points to own directory");
143fsck_err:
144 return ret;
145}
146
147void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
148{
149 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
150 struct qstr d_name = bch2_dirent_get_name(d);
151
152 prt_printf(out, "%.*s -> ", d_name.len, d_name.name);
153
154 if (d.v->d_type != DT_SUBVOL)
155 prt_printf(out, "%llu", le64_to_cpu(d.v->d_inum));
156 else
157 prt_printf(out, "%u -> %u",
158 le32_to_cpu(d.v->d_parent_subvol),
159 le32_to_cpu(d.v->d_child_subvol));
160
161 prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
162}
163
164static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
165 subvol_inum dir, u8 type,
166 const struct qstr *name, u64 dst)
167{
168 struct bkey_i_dirent *dirent;
169 unsigned u64s = BKEY_U64s + dirent_val_u64s(len: name->len);
170
171 if (name->len > BCH_NAME_MAX)
172 return ERR_PTR(error: -ENAMETOOLONG);
173
174 BUG_ON(u64s > U8_MAX);
175
176 dirent = bch2_trans_kmalloc(trans, size: u64s * sizeof(u64));
177 if (IS_ERR(ptr: dirent))
178 return dirent;
179
180 bkey_dirent_init(k: &dirent->k_i);
181 dirent->k.u64s = u64s;
182
183 if (type != DT_SUBVOL) {
184 dirent->v.d_inum = cpu_to_le64(dst);
185 } else {
186 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
187 dirent->v.d_child_subvol = cpu_to_le32(dst);
188 }
189
190 dirent->v.d_type = type;
191
192 memcpy(dirent->v.d_name, name->name, name->len);
193 memset(dirent->v.d_name + name->len, 0,
194 bkey_val_bytes(&dirent->k) -
195 offsetof(struct bch_dirent, d_name) -
196 name->len);
197
198 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
199
200 return dirent;
201}
202
203int bch2_dirent_create_snapshot(struct btree_trans *trans,
204 u32 dir_subvol, u64 dir, u32 snapshot,
205 const struct bch_hash_info *hash_info,
206 u8 type, const struct qstr *name, u64 dst_inum,
207 u64 *dir_offset,
208 bch_str_hash_flags_t str_hash_flags)
209{
210 subvol_inum dir_inum = { .subvol = dir_subvol, .inum = dir };
211 struct bkey_i_dirent *dirent;
212 int ret;
213
214 dirent = dirent_create_key(trans, dir: dir_inum, type, name, dst: dst_inum);
215 ret = PTR_ERR_OR_ZERO(ptr: dirent);
216 if (ret)
217 return ret;
218
219 dirent->k.p.inode = dir;
220 dirent->k.p.snapshot = snapshot;
221
222 ret = bch2_hash_set_in_snapshot(trans, desc: bch2_dirent_hash_desc, info: hash_info,
223 inum: dir_inum, snapshot,
224 insert: &dirent->k_i, str_hash_flags,
225 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
226 *dir_offset = dirent->k.p.offset;
227
228 return ret;
229}
230
231int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
232 const struct bch_hash_info *hash_info,
233 u8 type, const struct qstr *name, u64 dst_inum,
234 u64 *dir_offset,
235 bch_str_hash_flags_t str_hash_flags)
236{
237 struct bkey_i_dirent *dirent;
238 int ret;
239
240 dirent = dirent_create_key(trans, dir, type, name, dst: dst_inum);
241 ret = PTR_ERR_OR_ZERO(ptr: dirent);
242 if (ret)
243 return ret;
244
245 ret = bch2_hash_set(trans, desc: bch2_dirent_hash_desc, info: hash_info,
246 inum: dir, insert: &dirent->k_i, str_hash_flags);
247 *dir_offset = dirent->k.p.offset;
248
249 return ret;
250}
251
252static void dirent_copy_target(struct bkey_i_dirent *dst,
253 struct bkey_s_c_dirent src)
254{
255 dst->v.d_inum = src.v->d_inum;
256 dst->v.d_type = src.v->d_type;
257}
258
259int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
260 struct bkey_s_c_dirent d, subvol_inum *target)
261{
262 struct bch_subvolume s;
263 int ret = 0;
264
265 if (d.v->d_type == DT_SUBVOL &&
266 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
267 return 1;
268
269 if (likely(d.v->d_type != DT_SUBVOL)) {
270 target->subvol = dir.subvol;
271 target->inum = le64_to_cpu(d.v->d_inum);
272 } else {
273 target->subvol = le32_to_cpu(d.v->d_child_subvol);
274
275 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
276
277 target->inum = le64_to_cpu(s.inode);
278 }
279
280 return ret;
281}
282
283int bch2_dirent_rename(struct btree_trans *trans,
284 subvol_inum src_dir, struct bch_hash_info *src_hash,
285 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
286 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
287 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
288 enum bch_rename_mode mode)
289{
290 struct btree_iter src_iter = { NULL };
291 struct btree_iter dst_iter = { NULL };
292 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
293 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
294 struct bpos dst_pos =
295 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
296 unsigned src_update_flags = 0;
297 bool delete_src, delete_dst;
298 int ret = 0;
299
300 memset(src_inum, 0, sizeof(*src_inum));
301 memset(dst_inum, 0, sizeof(*dst_inum));
302
303 /* Lookup src: */
304 ret = bch2_hash_lookup(trans, iter: &src_iter, desc: bch2_dirent_hash_desc,
305 info: src_hash, inum: src_dir, key: src_name,
306 flags: BTREE_ITER_INTENT);
307 if (ret)
308 goto out;
309
310 old_src = bch2_btree_iter_peek_slot(&src_iter);
311 ret = bkey_err(old_src);
312 if (ret)
313 goto out;
314
315 ret = bch2_dirent_read_target(trans, dir: src_dir,
316 d: bkey_s_c_to_dirent(k: old_src), target: src_inum);
317 if (ret)
318 goto out;
319
320 /* Lookup dst: */
321 if (mode == BCH_RENAME) {
322 /*
323 * Note that we're _not_ checking if the target already exists -
324 * we're relying on the VFS to do that check for us for
325 * correctness:
326 */
327 ret = bch2_hash_hole(trans, iter: &dst_iter, desc: bch2_dirent_hash_desc,
328 info: dst_hash, inum: dst_dir, key: dst_name);
329 if (ret)
330 goto out;
331 } else {
332 ret = bch2_hash_lookup(trans, iter: &dst_iter, desc: bch2_dirent_hash_desc,
333 info: dst_hash, inum: dst_dir, key: dst_name,
334 flags: BTREE_ITER_INTENT);
335 if (ret)
336 goto out;
337
338 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
339 ret = bkey_err(old_dst);
340 if (ret)
341 goto out;
342
343 ret = bch2_dirent_read_target(trans, dir: dst_dir,
344 d: bkey_s_c_to_dirent(k: old_dst), target: dst_inum);
345 if (ret)
346 goto out;
347 }
348
349 if (mode != BCH_RENAME_EXCHANGE)
350 *src_offset = dst_iter.pos.offset;
351
352 /* Create new dst key: */
353 new_dst = dirent_create_key(trans, dir: dst_dir, type: 0, name: dst_name, dst: 0);
354 ret = PTR_ERR_OR_ZERO(ptr: new_dst);
355 if (ret)
356 goto out;
357
358 dirent_copy_target(dst: new_dst, src: bkey_s_c_to_dirent(k: old_src));
359 new_dst->k.p = dst_iter.pos;
360
361 /* Create new src key: */
362 if (mode == BCH_RENAME_EXCHANGE) {
363 new_src = dirent_create_key(trans, dir: src_dir, type: 0, name: src_name, dst: 0);
364 ret = PTR_ERR_OR_ZERO(ptr: new_src);
365 if (ret)
366 goto out;
367
368 dirent_copy_target(dst: new_src, src: bkey_s_c_to_dirent(k: old_dst));
369 new_src->k.p = src_iter.pos;
370 } else {
371 new_src = bch2_trans_kmalloc(trans, size: sizeof(struct bkey_i));
372 ret = PTR_ERR_OR_ZERO(ptr: new_src);
373 if (ret)
374 goto out;
375
376 bkey_init(k: &new_src->k);
377 new_src->k.p = src_iter.pos;
378
379 if (bkey_le(l: dst_pos, r: src_iter.pos) &&
380 bkey_lt(l: src_iter.pos, r: dst_iter.pos)) {
381 /*
382 * We have a hash collision for the new dst key,
383 * and new_src - the key we're deleting - is between
384 * new_dst's hashed slot and the slot we're going to be
385 * inserting it into - oops. This will break the hash
386 * table if we don't deal with it:
387 */
388 if (mode == BCH_RENAME) {
389 /*
390 * If we're not overwriting, we can just insert
391 * new_dst at the src position:
392 */
393 new_src = new_dst;
394 new_src->k.p = src_iter.pos;
395 goto out_set_src;
396 } else {
397 /* If we're overwriting, we can't insert new_dst
398 * at a different slot because it has to
399 * overwrite old_dst - just make sure to use a
400 * whiteout when deleting src:
401 */
402 new_src->k.type = KEY_TYPE_hash_whiteout;
403 }
404 } else {
405 /* Check if we need a whiteout to delete src: */
406 ret = bch2_hash_needs_whiteout(trans, desc: bch2_dirent_hash_desc,
407 info: src_hash, start: &src_iter);
408 if (ret < 0)
409 goto out;
410
411 if (ret)
412 new_src->k.type = KEY_TYPE_hash_whiteout;
413 }
414 }
415
416 if (new_dst->v.d_type == DT_SUBVOL)
417 new_dst->v.d_parent_subvol = cpu_to_le32(dst_dir.subvol);
418
419 if ((mode == BCH_RENAME_EXCHANGE) &&
420 new_src->v.d_type == DT_SUBVOL)
421 new_src->v.d_parent_subvol = cpu_to_le32(src_dir.subvol);
422
423 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
424 if (ret)
425 goto out;
426out_set_src:
427 /*
428 * If we're deleting a subvolume we need to really delete the dirent,
429 * not just emit a whiteout in the current snapshot - there can only be
430 * single dirent that points to a given subvolume.
431 *
432 * IOW, we don't maintain multiple versions in different snapshots of
433 * dirents that point to subvolumes - dirents that point to subvolumes
434 * are only visible in one particular subvolume so it's not necessary,
435 * and it would be particularly confusing for fsck to have to deal with.
436 */
437 delete_src = bkey_s_c_to_dirent(k: old_src).v->d_type == DT_SUBVOL &&
438 new_src->k.p.snapshot != old_src.k->p.snapshot;
439
440 delete_dst = old_dst.k &&
441 bkey_s_c_to_dirent(k: old_dst).v->d_type == DT_SUBVOL &&
442 new_dst->k.p.snapshot != old_dst.k->p.snapshot;
443
444 if (!delete_src || !bkey_deleted(&new_src->k)) {
445 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
446 if (ret)
447 goto out;
448 }
449
450 if (delete_src) {
451 bch2_btree_iter_set_snapshot(iter: &src_iter, snapshot: old_src.k->p.snapshot);
452 ret = bch2_btree_iter_traverse(&src_iter) ?:
453 bch2_btree_delete_at(trans, &src_iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
454 if (ret)
455 goto out;
456 }
457
458 if (delete_dst) {
459 bch2_btree_iter_set_snapshot(iter: &dst_iter, snapshot: old_dst.k->p.snapshot);
460 ret = bch2_btree_iter_traverse(&dst_iter) ?:
461 bch2_btree_delete_at(trans, &dst_iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
462 if (ret)
463 goto out;
464 }
465
466 if (mode == BCH_RENAME_EXCHANGE)
467 *src_offset = new_src->k.p.offset;
468 *dst_offset = new_dst->k.p.offset;
469out:
470 bch2_trans_iter_exit(trans, &src_iter);
471 bch2_trans_iter_exit(trans, &dst_iter);
472 return ret;
473}
474
475int bch2_dirent_lookup_trans(struct btree_trans *trans,
476 struct btree_iter *iter,
477 subvol_inum dir,
478 const struct bch_hash_info *hash_info,
479 const struct qstr *name, subvol_inum *inum,
480 unsigned flags)
481{
482 int ret = bch2_hash_lookup(trans, iter, desc: bch2_dirent_hash_desc,
483 info: hash_info, inum: dir, key: name, flags);
484 if (ret)
485 return ret;
486
487 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
488 ret = bkey_err(k);
489 if (ret)
490 goto err;
491
492 ret = bch2_dirent_read_target(trans, dir, d: bkey_s_c_to_dirent(k), target: inum);
493 if (ret > 0)
494 ret = -ENOENT;
495err:
496 if (ret)
497 bch2_trans_iter_exit(trans, iter);
498 return ret;
499}
500
501u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
502 const struct bch_hash_info *hash_info,
503 const struct qstr *name, subvol_inum *inum)
504{
505 struct btree_trans *trans = bch2_trans_get(c);
506 struct btree_iter iter = { NULL };
507
508 int ret = lockrestart_do(trans,
509 bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
510 bch2_trans_iter_exit(trans, &iter);
511 bch2_trans_put(trans);
512 return ret;
513}
514
515int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32 snapshot)
516{
517 struct btree_iter iter;
518 struct bkey_s_c k;
519 int ret;
520
521 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
522 SPOS(dir, 0, snapshot),
523 POS(dir, U64_MAX), 0, k, ret)
524 if (k.k->type == KEY_TYPE_dirent) {
525 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
526 if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
527 continue;
528 ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
529 break;
530 }
531 bch2_trans_iter_exit(trans, &iter);
532
533 return ret;
534}
535
536int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
537{
538 u32 snapshot;
539
540 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
541 bch2_empty_dir_snapshot(trans, dir: dir.inum, subvol: dir.subvol, snapshot);
542}
543
544int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
545{
546 struct btree_trans *trans = bch2_trans_get(c);
547 struct btree_iter iter;
548 struct bkey_s_c k;
549 struct bkey_s_c_dirent dirent;
550 subvol_inum target;
551 u32 snapshot;
552 struct bkey_buf sk;
553 struct qstr name;
554 int ret;
555
556 bch2_bkey_buf_init(s: &sk);
557retry:
558 bch2_trans_begin(trans);
559
560 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
561 if (ret)
562 goto err;
563
564 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
565 SPOS(inum.inum, ctx->pos, snapshot),
566 POS(inum.inum, U64_MAX), 0, k, ret) {
567 if (k.k->type != KEY_TYPE_dirent)
568 continue;
569
570 dirent = bkey_s_c_to_dirent(k);
571
572 ret = bch2_dirent_read_target(trans, dir: inum, d: dirent, target: &target);
573 if (ret < 0)
574 break;
575 if (ret)
576 continue;
577
578 /* dir_emit() can fault and block: */
579 bch2_bkey_buf_reassemble(s: &sk, c, k);
580 dirent = bkey_i_to_s_c_dirent(k: sk.k);
581 bch2_trans_unlock(trans);
582
583 name = bch2_dirent_get_name(d: dirent);
584
585 ctx->pos = dirent.k->p.offset;
586 if (!dir_emit(ctx, name: name.name,
587 namelen: name.len,
588 ino: target.inum,
589 type: vfs_d_type(type: dirent.v->d_type)))
590 break;
591 ctx->pos = dirent.k->p.offset + 1;
592
593 /*
594 * read_target looks up subvolumes, we can overflow paths if the
595 * directory has many subvolumes in it
596 */
597 ret = btree_trans_too_many_iters(trans);
598 if (ret)
599 break;
600 }
601 bch2_trans_iter_exit(trans, &iter);
602err:
603 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
604 goto retry;
605
606 bch2_trans_put(trans);
607 bch2_bkey_buf_exit(s: &sk, c);
608
609 return ret;
610}
611

source code of linux/fs/bcachefs/dirent.c