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
16 static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
17 {
18 if (bkey_val_bytes(d.k) < offsetof(struct bch_dirent, d_name))
19 return 0;
20
21 unsigned bkey_u64s = bkey_val_u64s(d.k);
22 unsigned bkey_bytes = bkey_u64s * sizeof(u64);
23 u64 last_u64 = ((u64*)d.v)[bkey_u64s - 1];
24 #if CPU_BIG_ENDIAN
25 unsigned trailing_nuls = last_u64 ? __builtin_ctzll(last_u64) / 8 : 64 / 8;
26 #else
27 unsigned trailing_nuls = last_u64 ? __builtin_clzll(last_u64) / 8 : 64 / 8;
28 #endif
29
30 return bkey_bytes -
31 offsetof(struct bch_dirent, d_name) -
32 trailing_nuls;
33 }
34
35 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
36 {
37 return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
38 }
39
40 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
41 const struct qstr *name)
42 {
43 struct bch_str_hash_ctx ctx;
44
45 bch2_str_hash_init(&ctx, info);
46 bch2_str_hash_update(&ctx, info, name->name, name->len);
47
48 /* [0,2) reserved for dots */
49 return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
50 }
51
52 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
53 {
54 return bch2_dirent_hash(info, key);
55 }
56
57 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
58 {
59 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
60 struct qstr name = bch2_dirent_get_name(d);
61
62 return bch2_dirent_hash(info, &name);
63 }
64
65 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
66 {
67 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
68 const struct qstr l_name = bch2_dirent_get_name(l);
69 const struct qstr *r_name = _r;
70
71 return !qstr_eq(l_name, *r_name);
72 }
73
74 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
75 {
76 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
77 struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
78 const struct qstr l_name = bch2_dirent_get_name(l);
79 const struct qstr r_name = bch2_dirent_get_name(r);
80
81 return !qstr_eq(l_name, r_name);
82 }
83
84 static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
85 {
86 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
87
88 if (d.v->d_type == DT_SUBVOL)
89 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
90 return true;
91 }
92
93 const struct bch_hash_desc bch2_dirent_hash_desc = {
94 .btree_id = BTREE_ID_dirents,
95 .key_type = KEY_TYPE_dirent,
96 .hash_key = dirent_hash_key,
97 .hash_bkey = dirent_hash_bkey,
98 .cmp_key = dirent_cmp_key,
99 .cmp_bkey = dirent_cmp_bkey,
100 .is_visible = dirent_is_visible,
101 };
102
103 int bch2_dirent_validate(struct bch_fs *c, struct bkey_s_c k,
104 enum bch_validate_flags flags)
105 {
106 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
107 struct qstr d_name = bch2_dirent_get_name(d);
108 int ret = 0;
109
110 bkey_fsck_err_on(!d_name.len,
111 c, dirent_empty_name,
112 "empty name");
113
114 bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len),
115 c, dirent_val_too_big,
116 "value too big (%zu > %u)",
117 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
118
119 /*
120 * Check new keys don't exceed the max length
121 * (older keys may be larger.)
122 */
123 bkey_fsck_err_on((flags & BCH_VALIDATE_commit) && d_name.len > BCH_NAME_MAX,
124 c, dirent_name_too_long,
125 "dirent name too big (%u > %u)",
126 d_name.len, BCH_NAME_MAX);
127
128 bkey_fsck_err_on(d_name.len != strnlen(d_name.name, d_name.len),
129 c, dirent_name_embedded_nul,
130 "dirent has stray data after name's NUL");
131
132 bkey_fsck_err_on((d_name.len == 1 && !memcmp(d_name.name, ".", 1)) ||
133 (d_name.len == 2 && !memcmp(d_name.name, "..", 2)),
134 c, dirent_name_dot_or_dotdot,
135 "invalid name");
136
137 bkey_fsck_err_on(memchr(d_name.name, '/', d_name.len),
138 c, dirent_name_has_slash,
139 "name with /");
140
141 bkey_fsck_err_on(d.v->d_type != DT_SUBVOL &&
142 le64_to_cpu(d.v->d_inum) == d.k->p.inode,
143 c, dirent_to_itself,
144 "dirent points to own directory");
145 fsck_err:
146 return ret;
147 }
148
149 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
150 {
151 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
152 struct qstr d_name = bch2_dirent_get_name(d);
153
154 prt_printf(out, "%.*s -> ", d_name.len, d_name.name);
155
156 if (d.v->d_type != DT_SUBVOL)
157 prt_printf(out, "%llu", le64_to_cpu(d.v->d_inum));
158 else
159 prt_printf(out, "%u -> %u",
160 le32_to_cpu(d.v->d_parent_subvol),
161 le32_to_cpu(d.v->d_child_subvol));
162
163 prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
164 }
165
166 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
167 subvol_inum dir, u8 type,
168 const struct qstr *name, u64 dst)
169 {
170 struct bkey_i_dirent *dirent;
171 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
172
173 if (name->len > BCH_NAME_MAX)
174 return ERR_PTR(-ENAMETOOLONG);
175
176 BUG_ON(u64s > U8_MAX);
177
178 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
179 if (IS_ERR(dirent))
180 return dirent;
181
182 bkey_dirent_init(&dirent->k_i);
183 dirent->k.u64s = u64s;
184
185 if (type != DT_SUBVOL) {
186 dirent->v.d_inum = cpu_to_le64(dst);
187 } else {
188 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
189 dirent->v.d_child_subvol = cpu_to_le32(dst);
190 }
191
192 dirent->v.d_type = type;
193
194 memcpy(dirent->v.d_name, name->name, name->len);
195 memset(dirent->v.d_name + name->len, 0,
196 bkey_val_bytes(&dirent->k) -
197 offsetof(struct bch_dirent, d_name) -
198 name->len);
199
200 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
201
202 return dirent;
203 }
204
205 int bch2_dirent_create_snapshot(struct btree_trans *trans,
206 u32 dir_subvol, u64 dir, u32 snapshot,
207 const struct bch_hash_info *hash_info,
208 u8 type, const struct qstr *name, u64 dst_inum,
209 u64 *dir_offset,
210 enum btree_iter_update_trigger_flags flags)
211 {
212 subvol_inum dir_inum = { .subvol = dir_subvol, .inum = dir };
213 struct bkey_i_dirent *dirent;
214 int ret;
215
216 dirent = dirent_create_key(trans, dir_inum, type, name, dst_inum);
217 ret = PTR_ERR_OR_ZERO(dirent);
218 if (ret)
219 return ret;
220
221 dirent->k.p.inode = dir;
222 dirent->k.p.snapshot = snapshot;
223
224 ret = bch2_hash_set_in_snapshot(trans, bch2_dirent_hash_desc, hash_info,
225 dir_inum, snapshot, &dirent->k_i,
226 flags|BTREE_UPDATE_internal_snapshot_node);
227 *dir_offset = dirent->k.p.offset;
228
229 return ret;
230 }
231
232 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
233 const struct bch_hash_info *hash_info,
234 u8 type, const struct qstr *name, u64 dst_inum,
235 u64 *dir_offset,
236 enum btree_iter_update_trigger_flags flags)
237 {
238 struct bkey_i_dirent *dirent;
239 int ret;
240
241 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
242 ret = PTR_ERR_OR_ZERO(dirent);
243 if (ret)
244 return ret;
245
246 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
247 dir, &dirent->k_i, flags);
248 *dir_offset = dirent->k.p.offset;
249
250 return ret;
251 }
252
253 static void dirent_copy_target(struct bkey_i_dirent *dst,
254 struct bkey_s_c_dirent src)
255 {
256 dst->v.d_inum = src.v->d_inum;
257 dst->v.d_type = src.v->d_type;
258 }
259
260 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
261 struct bkey_s_c_dirent d, subvol_inum *target)
262 {
263 struct bch_subvolume s;
264 int ret = 0;
265
266 if (d.v->d_type == DT_SUBVOL &&
267 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
268 return 1;
269
270 if (likely(d.v->d_type != DT_SUBVOL)) {
271 target->subvol = dir.subvol;
272 target->inum = le64_to_cpu(d.v->d_inum);
273 } else {
274 target->subvol = le32_to_cpu(d.v->d_child_subvol);
275
276 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_cached, &s);
277
278 target->inum = le64_to_cpu(s.inode);
279 }
280
281 return ret;
282 }
283
284 int bch2_dirent_rename(struct btree_trans *trans,
285 subvol_inum src_dir, struct bch_hash_info *src_hash,
286 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
287 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
288 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
289 enum bch_rename_mode mode)
290 {
291 struct btree_iter src_iter = { NULL };
292 struct btree_iter dst_iter = { NULL };
293 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
294 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
295 struct bpos dst_pos =
296 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
297 unsigned src_update_flags = 0;
298 bool delete_src, delete_dst;
299 int ret = 0;
300
301 memset(src_inum, 0, sizeof(*src_inum));
302 memset(dst_inum, 0, sizeof(*dst_inum));
303
304 /* Lookup src: */
305 old_src = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
306 src_hash, src_dir, src_name,
307 BTREE_ITER_intent);
308 ret = bkey_err(old_src);
309 if (ret)
310 goto out;
311
312 ret = bch2_dirent_read_target(trans, src_dir,
313 bkey_s_c_to_dirent(old_src), src_inum);
314 if (ret)
315 goto out;
316
317 /* Lookup dst: */
318 if (mode == BCH_RENAME) {
319 /*
320 * Note that we're _not_ checking if the target already exists -
321 * we're relying on the VFS to do that check for us for
322 * correctness:
323 */
324 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
325 dst_hash, dst_dir, dst_name);
326 if (ret)
327 goto out;
328 } else {
329 old_dst = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
330 dst_hash, dst_dir, dst_name,
331 BTREE_ITER_intent);
332 ret = bkey_err(old_dst);
333 if (ret)
334 goto out;
335
336 ret = bch2_dirent_read_target(trans, dst_dir,
337 bkey_s_c_to_dirent(old_dst), dst_inum);
338 if (ret)
339 goto out;
340 }
341
342 if (mode != BCH_RENAME_EXCHANGE)
343 *src_offset = dst_iter.pos.offset;
344
345 /* Create new dst key: */
346 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
347 ret = PTR_ERR_OR_ZERO(new_dst);
348 if (ret)
349 goto out;
350
351 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
352 new_dst->k.p = dst_iter.pos;
353
354 /* Create new src key: */
355 if (mode == BCH_RENAME_EXCHANGE) {
356 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
357 ret = PTR_ERR_OR_ZERO(new_src);
358 if (ret)
359 goto out;
360
361 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
362 new_src->k.p = src_iter.pos;
363 } else {
364 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
365 ret = PTR_ERR_OR_ZERO(new_src);
366 if (ret)
367 goto out;
368
369 bkey_init(&new_src->k);
370 new_src->k.p = src_iter.pos;
371
372 if (bkey_le(dst_pos, src_iter.pos) &&
373 bkey_lt(src_iter.pos, dst_iter.pos)) {
374 /*
375 * We have a hash collision for the new dst key,
376 * and new_src - the key we're deleting - is between
377 * new_dst's hashed slot and the slot we're going to be
378 * inserting it into - oops. This will break the hash
379 * table if we don't deal with it:
380 */
381 if (mode == BCH_RENAME) {
382 /*
383 * If we're not overwriting, we can just insert
384 * new_dst at the src position:
385 */
386 new_src = new_dst;
387 new_src->k.p = src_iter.pos;
388 goto out_set_src;
389 } else {
390 /* If we're overwriting, we can't insert new_dst
391 * at a different slot because it has to
392 * overwrite old_dst - just make sure to use a
393 * whiteout when deleting src:
394 */
395 new_src->k.type = KEY_TYPE_hash_whiteout;
396 }
397 } else {
398 /* Check if we need a whiteout to delete src: */
399 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
400 src_hash, &src_iter);
401 if (ret < 0)
402 goto out;
403
404 if (ret)
405 new_src->k.type = KEY_TYPE_hash_whiteout;
406 }
407 }
408
409 if (new_dst->v.d_type == DT_SUBVOL)
410 new_dst->v.d_parent_subvol = cpu_to_le32(dst_dir.subvol);
411
412 if ((mode == BCH_RENAME_EXCHANGE) &&
413 new_src->v.d_type == DT_SUBVOL)
414 new_src->v.d_parent_subvol = cpu_to_le32(src_dir.subvol);
415
416 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
417 if (ret)
418 goto out;
419 out_set_src:
420 /*
421 * If we're deleting a subvolume we need to really delete the dirent,
422 * not just emit a whiteout in the current snapshot - there can only be
423 * single dirent that points to a given subvolume.
424 *
425 * IOW, we don't maintain multiple versions in different snapshots of
426 * dirents that point to subvolumes - dirents that point to subvolumes
427 * are only visible in one particular subvolume so it's not necessary,
428 * and it would be particularly confusing for fsck to have to deal with.
429 */
430 delete_src = bkey_s_c_to_dirent(old_src).v->d_type == DT_SUBVOL &&
431 new_src->k.p.snapshot != old_src.k->p.snapshot;
432
433 delete_dst = old_dst.k &&
434 bkey_s_c_to_dirent(old_dst).v->d_type == DT_SUBVOL &&
435 new_dst->k.p.snapshot != old_dst.k->p.snapshot;
436
437 if (!delete_src || !bkey_deleted(&new_src->k)) {
438 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
439 if (ret)
440 goto out;
441 }
442
443 if (delete_src) {
444 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
445 ret = bch2_btree_iter_traverse(&src_iter) ?:
446 bch2_btree_delete_at(trans, &src_iter, BTREE_UPDATE_internal_snapshot_node);
447 if (ret)
448 goto out;
449 }
450
451 if (delete_dst) {
452 bch2_btree_iter_set_snapshot(&dst_iter, old_dst.k->p.snapshot);
453 ret = bch2_btree_iter_traverse(&dst_iter) ?:
454 bch2_btree_delete_at(trans, &dst_iter, BTREE_UPDATE_internal_snapshot_node);
455 if (ret)
456 goto out;
457 }
458
459 if (mode == BCH_RENAME_EXCHANGE)
460 *src_offset = new_src->k.p.offset;
461 *dst_offset = new_dst->k.p.offset;
462 out:
463 bch2_trans_iter_exit(trans, &src_iter);
464 bch2_trans_iter_exit(trans, &dst_iter);
465 return ret;
466 }
467
468 int bch2_dirent_lookup_trans(struct btree_trans *trans,
469 struct btree_iter *iter,
470 subvol_inum dir,
471 const struct bch_hash_info *hash_info,
472 const struct qstr *name, subvol_inum *inum,
473 unsigned flags)
474 {
475 struct bkey_s_c k = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
476 hash_info, dir, name, flags);
477 int ret = bkey_err(k);
478 if (ret)
479 goto err;
480
481 ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), inum);
482 if (ret > 0)
483 ret = -ENOENT;
484 err:
485 if (ret)
486 bch2_trans_iter_exit(trans, iter);
487 return ret;
488 }
489
490 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
491 const struct bch_hash_info *hash_info,
492 const struct qstr *name, subvol_inum *inum)
493 {
494 struct btree_trans *trans = bch2_trans_get(c);
495 struct btree_iter iter = { NULL };
496
497 int ret = lockrestart_do(trans,
498 bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
499 bch2_trans_iter_exit(trans, &iter);
500 bch2_trans_put(trans);
501 return ret;
502 }
503
504 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32 snapshot)
505 {
506 struct btree_iter iter;
507 struct bkey_s_c k;
508 int ret;
509
510 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
511 SPOS(dir, 0, snapshot),
512 POS(dir, U64_MAX), 0, k, ret)
513 if (k.k->type == KEY_TYPE_dirent) {
514 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
515 if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
516 continue;
517 ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
518 break;
519 }
520 bch2_trans_iter_exit(trans, &iter);
521
522 return ret;
523 }
524
525 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
526 {
527 u32 snapshot;
528
529 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
530 bch2_empty_dir_snapshot(trans, dir.inum, dir.subvol, snapshot);
531 }
532
533 static int bch2_dir_emit(struct dir_context *ctx, struct bkey_s_c_dirent d, subvol_inum target)
534 {
535 struct qstr name = bch2_dirent_get_name(d);
536 /*
537 * Although not required by the kernel code, updating ctx->pos is needed
538 * for the bcachefs FUSE driver. Without this update, the FUSE
539 * implementation will be stuck in an infinite loop when reading
540 * directories (via the bcachefs_fuse_readdir callback).
541 * In kernel space, ctx->pos is updated by the VFS code.
542 */
543 ctx->pos = d.k->p.offset;
544 bool ret = dir_emit(ctx, name.name,
545 name.len,
546 target.inum,
547 vfs_d_type(d.v->d_type));
548 if (ret)
549 ctx->pos = d.k->p.offset + 1;
550 return ret;
551 }
552
553 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
554 {
555 struct btree_trans *trans = bch2_trans_get(c);
556 struct btree_iter iter;
557 struct bkey_s_c k;
558 subvol_inum target;
559 u32 snapshot;
560 struct bkey_buf sk;
561 int ret;
562
563 bch2_bkey_buf_init(&sk);
564 retry:
565 bch2_trans_begin(trans);
566
567 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
568 if (ret)
569 goto err;
570
571 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
572 SPOS(inum.inum, ctx->pos, snapshot),
573 POS(inum.inum, U64_MAX), 0, k, ret) {
574 if (k.k->type != KEY_TYPE_dirent)
575 continue;
576
577 /* dir_emit() can fault and block: */
578 bch2_bkey_buf_reassemble(&sk, c, k);
579 struct bkey_s_c_dirent dirent = bkey_i_to_s_c_dirent(sk.k);
580
581 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
582 if (ret < 0)
583 break;
584 if (ret)
585 continue;
586
587 /*
588 * read_target looks up subvolumes, we can overflow paths if the
589 * directory has many subvolumes in it
590 *
591 * XXX: btree_trans_too_many_iters() is something we'd like to
592 * get rid of, and there's no good reason to be using it here
593 * except that we don't yet have a for_each_btree_key() helper
594 * that does subvolume_get_snapshot().
595 */
596 ret = drop_locks_do(trans,
597 bch2_dir_emit(ctx, dirent, target)) ?:
598 btree_trans_too_many_iters(trans);
599 if (ret) {
600 ret = ret < 0 ? ret : 0;
601 break;
602 }
603 }
604 bch2_trans_iter_exit(trans, &iter);
605 err:
606 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
607 goto retry;
608
609 bch2_trans_put(trans);
610 bch2_bkey_buf_exit(&sk, c);
611
612 return ret;
613 }
614
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.