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_invalid(struct bch_fs *c, struct bkey_s_c k,
104 enum bch_validate_flags flags,
105 struct printbuf *err)
106 {
107 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
108 struct qstr d_name = bch2_dirent_get_name(d);
109 int ret = 0;
110
111 bkey_fsck_err_on(!d_name.len, c, err,
112 dirent_empty_name,
113 "empty name");
114
115 bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len), c, err,
116 dirent_val_too_big,
117 "value too big (%zu > %u)",
118 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
119
120 /*
121 * Check new keys don't exceed the max length
122 * (older keys may be larger.)
123 */
124 bkey_fsck_err_on((flags & BCH_VALIDATE_commit) && d_name.len > BCH_NAME_MAX, c, err,
125 dirent_name_too_long,
126 "dirent name too big (%u > %u)",
127 d_name.len, BCH_NAME_MAX);
128
129 bkey_fsck_err_on(d_name.len != strnlen(d_name.name, d_name.len), c, err,
130 dirent_name_embedded_nul,
131 "dirent has stray data after name's NUL");
132
133 bkey_fsck_err_on((d_name.len == 1 && !memcmp(d_name.name, ".", 1)) ||
134 (d_name.len == 2 && !memcmp(d_name.name, "..", 2)), c, err,
135 dirent_name_dot_or_dotdot,
136 "invalid name");
137
138 bkey_fsck_err_on(memchr(d_name.name, '/', d_name.len), c, err,
139 dirent_name_has_slash,
140 "name with /");
141
142 bkey_fsck_err_on(d.v->d_type != DT_SUBVOL &&
143 le64_to_cpu(d.v->d_inum) == d.k->p.inode, c, err,
144 dirent_to_itself,
145 "dirent points to own directory");
146 fsck_err:
147 return ret;
148 }
149
150 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
151 {
152 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
153 struct qstr d_name = bch2_dirent_get_name(d);
154
155 prt_printf(out, "%.*s -> ", d_name.len, d_name.name);
156
157 if (d.v->d_type != DT_SUBVOL)
158 prt_printf(out, "%llu", le64_to_cpu(d.v->d_inum));
159 else
160 prt_printf(out, "%u -> %u",
161 le32_to_cpu(d.v->d_parent_subvol),
162 le32_to_cpu(d.v->d_child_subvol));
163
164 prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
165 }
166
167 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
168 subvol_inum dir, u8 type,
169 const struct qstr *name, u64 dst)
170 {
171 struct bkey_i_dirent *dirent;
172 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
173
174 if (name->len > BCH_NAME_MAX)
175 return ERR_PTR(-ENAMETOOLONG);
176
177 BUG_ON(u64s > U8_MAX);
178
179 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
180 if (IS_ERR(dirent))
181 return dirent;
182
183 bkey_dirent_init(&dirent->k_i);
184 dirent->k.u64s = u64s;
185
186 if (type != DT_SUBVOL) {
187 dirent->v.d_inum = cpu_to_le64(dst);
188 } else {
189 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
190 dirent->v.d_child_subvol = cpu_to_le32(dst);
191 }
192
193 dirent->v.d_type = type;
194
195 memcpy(dirent->v.d_name, name->name, name->len);
196 memset(dirent->v.d_name + name->len, 0,
197 bkey_val_bytes(&dirent->k) -
198 offsetof(struct bch_dirent, d_name) -
199 name->len);
200
201 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
202
203 return dirent;
204 }
205
206 int bch2_dirent_create_snapshot(struct btree_trans *trans,
207 u32 dir_subvol, u64 dir, u32 snapshot,
208 const struct bch_hash_info *hash_info,
209 u8 type, const struct qstr *name, u64 dst_inum,
210 u64 *dir_offset,
211 enum btree_iter_update_trigger_flags flags)
212 {
213 subvol_inum dir_inum = { .subvol = dir_subvol, .inum = dir };
214 struct bkey_i_dirent *dirent;
215 int ret;
216
217 dirent = dirent_create_key(trans, dir_inum, type, name, dst_inum);
218 ret = PTR_ERR_OR_ZERO(dirent);
219 if (ret)
220 return ret;
221
222 dirent->k.p.inode = dir;
223 dirent->k.p.snapshot = snapshot;
224
225 ret = bch2_hash_set_in_snapshot(trans, bch2_dirent_hash_desc, hash_info,
226 dir_inum, snapshot, &dirent->k_i,
227 flags|BTREE_UPDATE_internal_snapshot_node);
228 *dir_offset = dirent->k.p.offset;
229
230 return ret;
231 }
232
233 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
234 const struct bch_hash_info *hash_info,
235 u8 type, const struct qstr *name, u64 dst_inum,
236 u64 *dir_offset,
237 enum btree_iter_update_trigger_flags flags)
238 {
239 struct bkey_i_dirent *dirent;
240 int ret;
241
242 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
243 ret = PTR_ERR_OR_ZERO(dirent);
244 if (ret)
245 return ret;
246
247 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
248 dir, &dirent->k_i, flags);
249 *dir_offset = dirent->k.p.offset;
250
251 return ret;
252 }
253
254 static void dirent_copy_target(struct bkey_i_dirent *dst,
255 struct bkey_s_c_dirent src)
256 {
257 dst->v.d_inum = src.v->d_inum;
258 dst->v.d_type = src.v->d_type;
259 }
260
261 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
262 struct bkey_s_c_dirent d, subvol_inum *target)
263 {
264 struct bch_subvolume s;
265 int ret = 0;
266
267 if (d.v->d_type == DT_SUBVOL &&
268 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
269 return 1;
270
271 if (likely(d.v->d_type != DT_SUBVOL)) {
272 target->subvol = dir.subvol;
273 target->inum = le64_to_cpu(d.v->d_inum);
274 } else {
275 target->subvol = le32_to_cpu(d.v->d_child_subvol);
276
277 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_cached, &s);
278
279 target->inum = le64_to_cpu(s.inode);
280 }
281
282 return ret;
283 }
284
285 int bch2_dirent_rename(struct btree_trans *trans,
286 subvol_inum src_dir, struct bch_hash_info *src_hash,
287 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
288 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
289 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
290 enum bch_rename_mode mode)
291 {
292 struct btree_iter src_iter = { NULL };
293 struct btree_iter dst_iter = { NULL };
294 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
295 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
296 struct bpos dst_pos =
297 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
298 unsigned src_update_flags = 0;
299 bool delete_src, delete_dst;
300 int ret = 0;
301
302 memset(src_inum, 0, sizeof(*src_inum));
303 memset(dst_inum, 0, sizeof(*dst_inum));
304
305 /* Lookup src: */
306 old_src = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
307 src_hash, src_dir, src_name,
308 BTREE_ITER_intent);
309 ret = bkey_err(old_src);
310 if (ret)
311 goto out;
312
313 ret = bch2_dirent_read_target(trans, src_dir,
314 bkey_s_c_to_dirent(old_src), src_inum);
315 if (ret)
316 goto out;
317
318 /* Lookup dst: */
319 if (mode == BCH_RENAME) {
320 /*
321 * Note that we're _not_ checking if the target already exists -
322 * we're relying on the VFS to do that check for us for
323 * correctness:
324 */
325 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
326 dst_hash, dst_dir, dst_name);
327 if (ret)
328 goto out;
329 } else {
330 old_dst = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
331 dst_hash, dst_dir, dst_name,
332 BTREE_ITER_intent);
333 ret = bkey_err(old_dst);
334 if (ret)
335 goto out;
336
337 ret = bch2_dirent_read_target(trans, dst_dir,
338 bkey_s_c_to_dirent(old_dst), dst_inum);
339 if (ret)
340 goto out;
341 }
342
343 if (mode != BCH_RENAME_EXCHANGE)
344 *src_offset = dst_iter.pos.offset;
345
346 /* Create new dst key: */
347 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
348 ret = PTR_ERR_OR_ZERO(new_dst);
349 if (ret)
350 goto out;
351
352 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
353 new_dst->k.p = dst_iter.pos;
354
355 /* Create new src key: */
356 if (mode == BCH_RENAME_EXCHANGE) {
357 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
358 ret = PTR_ERR_OR_ZERO(new_src);
359 if (ret)
360 goto out;
361
362 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
363 new_src->k.p = src_iter.pos;
364 } else {
365 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
366 ret = PTR_ERR_OR_ZERO(new_src);
367 if (ret)
368 goto out;
369
370 bkey_init(&new_src->k);
371 new_src->k.p = src_iter.pos;
372
373 if (bkey_le(dst_pos, src_iter.pos) &&
374 bkey_lt(src_iter.pos, dst_iter.pos)) {
375 /*
376 * We have a hash collision for the new dst key,
377 * and new_src - the key we're deleting - is between
378 * new_dst's hashed slot and the slot we're going to be
379 * inserting it into - oops. This will break the hash
380 * table if we don't deal with it:
381 */
382 if (mode == BCH_RENAME) {
383 /*
384 * If we're not overwriting, we can just insert
385 * new_dst at the src position:
386 */
387 new_src = new_dst;
388 new_src->k.p = src_iter.pos;
389 goto out_set_src;
390 } else {
391 /* If we're overwriting, we can't insert new_dst
392 * at a different slot because it has to
393 * overwrite old_dst - just make sure to use a
394 * whiteout when deleting src:
395 */
396 new_src->k.type = KEY_TYPE_hash_whiteout;
397 }
398 } else {
399 /* Check if we need a whiteout to delete src: */
400 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
401 src_hash, &src_iter);
402 if (ret < 0)
403 goto out;
404
405 if (ret)
406 new_src->k.type = KEY_TYPE_hash_whiteout;
407 }
408 }
409
410 if (new_dst->v.d_type == DT_SUBVOL)
411 new_dst->v.d_parent_subvol = cpu_to_le32(dst_dir.subvol);
412
413 if ((mode == BCH_RENAME_EXCHANGE) &&
414 new_src->v.d_type == DT_SUBVOL)
415 new_src->v.d_parent_subvol = cpu_to_le32(src_dir.subvol);
416
417 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
418 if (ret)
419 goto out;
420 out_set_src:
421 /*
422 * If we're deleting a subvolume we need to really delete the dirent,
423 * not just emit a whiteout in the current snapshot - there can only be
424 * single dirent that points to a given subvolume.
425 *
426 * IOW, we don't maintain multiple versions in different snapshots of
427 * dirents that point to subvolumes - dirents that point to subvolumes
428 * are only visible in one particular subvolume so it's not necessary,
429 * and it would be particularly confusing for fsck to have to deal with.
430 */
431 delete_src = bkey_s_c_to_dirent(old_src).v->d_type == DT_SUBVOL &&
432 new_src->k.p.snapshot != old_src.k->p.snapshot;
433
434 delete_dst = old_dst.k &&
435 bkey_s_c_to_dirent(old_dst).v->d_type == DT_SUBVOL &&
436 new_dst->k.p.snapshot != old_dst.k->p.snapshot;
437
438 if (!delete_src || !bkey_deleted(&new_src->k)) {
439 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
440 if (ret)
441 goto out;
442 }
443
444 if (delete_src) {
445 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
446 ret = bch2_btree_iter_traverse(&src_iter) ?:
447 bch2_btree_delete_at(trans, &src_iter, BTREE_UPDATE_internal_snapshot_node);
448 if (ret)
449 goto out;
450 }
451
452 if (delete_dst) {
453 bch2_btree_iter_set_snapshot(&dst_iter, old_dst.k->p.snapshot);
454 ret = bch2_btree_iter_traverse(&dst_iter) ?:
455 bch2_btree_delete_at(trans, &dst_iter, BTREE_UPDATE_internal_snapshot_node);
456 if (ret)
457 goto out;
458 }
459
460 if (mode == BCH_RENAME_EXCHANGE)
461 *src_offset = new_src->k.p.offset;
462 *dst_offset = new_dst->k.p.offset;
463 out:
464 bch2_trans_iter_exit(trans, &src_iter);
465 bch2_trans_iter_exit(trans, &dst_iter);
466 return ret;
467 }
468
469 int bch2_dirent_lookup_trans(struct btree_trans *trans,
470 struct btree_iter *iter,
471 subvol_inum dir,
472 const struct bch_hash_info *hash_info,
473 const struct qstr *name, subvol_inum *inum,
474 unsigned flags)
475 {
476 struct bkey_s_c k = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
477 hash_info, dir, name, flags);
478 int ret = bkey_err(k);
479 if (ret)
480 goto err;
481
482 ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), inum);
483 if (ret > 0)
484 ret = -ENOENT;
485 err:
486 if (ret)
487 bch2_trans_iter_exit(trans, iter);
488 return ret;
489 }
490
491 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
492 const struct bch_hash_info *hash_info,
493 const struct qstr *name, subvol_inum *inum)
494 {
495 struct btree_trans *trans = bch2_trans_get(c);
496 struct btree_iter iter = { NULL };
497
498 int ret = lockrestart_do(trans,
499 bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
500 bch2_trans_iter_exit(trans, &iter);
501 bch2_trans_put(trans);
502 return ret;
503 }
504
505 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32 snapshot)
506 {
507 struct btree_iter iter;
508 struct bkey_s_c k;
509 int ret;
510
511 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
512 SPOS(dir, 0, snapshot),
513 POS(dir, U64_MAX), 0, k, ret)
514 if (k.k->type == KEY_TYPE_dirent) {
515 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
516 if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
517 continue;
518 ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
519 break;
520 }
521 bch2_trans_iter_exit(trans, &iter);
522
523 return ret;
524 }
525
526 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
527 {
528 u32 snapshot;
529
530 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
531 bch2_empty_dir_snapshot(trans, dir.inum, dir.subvol, snapshot);
532 }
533
534 static int bch2_dir_emit(struct dir_context *ctx, struct bkey_s_c_dirent d, subvol_inum target)
535 {
536 struct qstr name = bch2_dirent_get_name(d);
537 /*
538 * Although not required by the kernel code, updating ctx->pos is needed
539 * for the bcachefs FUSE driver. Without this update, the FUSE
540 * implementation will be stuck in an infinite loop when reading
541 * directories (via the bcachefs_fuse_readdir callback).
542 * In kernel space, ctx->pos is updated by the VFS code.
543 */
544 ctx->pos = d.k->p.offset;
545 bool ret = dir_emit(ctx, name.name,
546 name.len,
547 target.inum,
548 vfs_d_type(d.v->d_type));
549 if (ret)
550 ctx->pos = d.k->p.offset + 1;
551 return ret;
552 }
553
554 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
555 {
556 struct btree_trans *trans = bch2_trans_get(c);
557 struct btree_iter iter;
558 struct bkey_s_c k;
559 subvol_inum target;
560 u32 snapshot;
561 struct bkey_buf sk;
562 int ret;
563
564 bch2_bkey_buf_init(&sk);
565 retry:
566 bch2_trans_begin(trans);
567
568 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
569 if (ret)
570 goto err;
571
572 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
573 SPOS(inum.inum, ctx->pos, snapshot),
574 POS(inum.inum, U64_MAX), 0, k, ret) {
575 if (k.k->type != KEY_TYPE_dirent)
576 continue;
577
578 /* dir_emit() can fault and block: */
579 bch2_bkey_buf_reassemble(&sk, c, k);
580 struct bkey_s_c_dirent dirent = bkey_i_to_s_c_dirent(sk.k);
581
582 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
583 if (ret < 0)
584 break;
585 if (ret)
586 continue;
587
588 /*
589 * read_target looks up subvolumes, we can overflow paths if the
590 * directory has many subvolumes in it
591 *
592 * XXX: btree_trans_too_many_iters() is something we'd like to
593 * get rid of, and there's no good reason to be using it here
594 * except that we don't yet have a for_each_btree_key() helper
595 * that does subvolume_get_snapshot().
596 */
597 ret = drop_locks_do(trans,
598 bch2_dir_emit(ctx, dirent, target)) ?:
599 btree_trans_too_many_iters(trans);
600 if (ret) {
601 ret = ret < 0 ? ret : 0;
602 break;
603 }
604 }
605 bch2_trans_iter_exit(trans, &iter);
606 err:
607 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
608 goto retry;
609
610 bch2_trans_put(trans);
611 bch2_bkey_buf_exit(&sk, c);
612
613 return ret;
614 }
615
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.