xref: /linux/fs/bcachefs/dirent.c (revision 0678df8271820bcf8fb4f877129f05d68a237de4)
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 	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 
32 struct 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 
37 static 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, info);
43 	bch2_str_hash_update(&ctx, info, name->name, name->len);
44 
45 	/* [0,2) reserved for dots */
46 	return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
47 }
48 
49 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
50 {
51 	return bch2_dirent_hash(info, key);
52 }
53 
54 static 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);
60 }
61 
62 static 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(_l);
65 	const struct qstr l_name = bch2_dirent_get_name(l);
66 	const struct qstr *r_name = _r;
67 
68 	return l_name.len - r_name->len ?: memcmp(l_name.name, r_name->name, l_name.len);
69 }
70 
71 static 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(_l);
74 	struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
75 	const struct qstr l_name = bch2_dirent_get_name(l);
76 	const struct qstr r_name = bch2_dirent_get_name(r);
77 
78 	return l_name.len - r_name.len ?: memcmp(l_name.name, r_name.name, l_name.len);
79 }
80 
81 static 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 
90 const 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 
100 int 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");
143 fsck_err:
144 	return ret;
145 }
146 
147 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
148 			 struct bkey_s_c k)
149 {
150 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
151 	struct qstr d_name = bch2_dirent_get_name(d);
152 
153 	prt_printf(out, "%.*s -> %llu type %s",
154 	       d_name.len,
155 	       d_name.name,
156 	       d.v->d_type != DT_SUBVOL
157 	       ? le64_to_cpu(d.v->d_inum)
158 	       : le32_to_cpu(d.v->d_child_subvol),
159 	       bch2_d_type_str(d.v->d_type));
160 }
161 
162 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
163 				subvol_inum dir, u8 type,
164 				const struct qstr *name, u64 dst)
165 {
166 	struct bkey_i_dirent *dirent;
167 	unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
168 
169 	if (name->len > BCH_NAME_MAX)
170 		return ERR_PTR(-ENAMETOOLONG);
171 
172 	BUG_ON(u64s > U8_MAX);
173 
174 	dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
175 	if (IS_ERR(dirent))
176 		return dirent;
177 
178 	bkey_dirent_init(&dirent->k_i);
179 	dirent->k.u64s = u64s;
180 
181 	if (type != DT_SUBVOL) {
182 		dirent->v.d_inum = cpu_to_le64(dst);
183 	} else {
184 		dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
185 		dirent->v.d_child_subvol = cpu_to_le32(dst);
186 	}
187 
188 	dirent->v.d_type = type;
189 
190 	memcpy(dirent->v.d_name, name->name, name->len);
191 	memset(dirent->v.d_name + name->len, 0,
192 	       bkey_val_bytes(&dirent->k) -
193 	       offsetof(struct bch_dirent, d_name) -
194 	       name->len);
195 
196 	EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
197 
198 	return dirent;
199 }
200 
201 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
202 		       const struct bch_hash_info *hash_info,
203 		       u8 type, const struct qstr *name, u64 dst_inum,
204 		       u64 *dir_offset, int flags)
205 {
206 	struct bkey_i_dirent *dirent;
207 	int ret;
208 
209 	dirent = dirent_create_key(trans, dir, type, name, dst_inum);
210 	ret = PTR_ERR_OR_ZERO(dirent);
211 	if (ret)
212 		return ret;
213 
214 	ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
215 			    dir, &dirent->k_i, flags);
216 	*dir_offset = dirent->k.p.offset;
217 
218 	return ret;
219 }
220 
221 static void dirent_copy_target(struct bkey_i_dirent *dst,
222 			       struct bkey_s_c_dirent src)
223 {
224 	dst->v.d_inum = src.v->d_inum;
225 	dst->v.d_type = src.v->d_type;
226 }
227 
228 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
229 			    struct bkey_s_c_dirent d, subvol_inum *target)
230 {
231 	struct bch_subvolume s;
232 	int ret = 0;
233 
234 	if (d.v->d_type == DT_SUBVOL &&
235 	    le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
236 		return 1;
237 
238 	if (likely(d.v->d_type != DT_SUBVOL)) {
239 		target->subvol	= dir.subvol;
240 		target->inum	= le64_to_cpu(d.v->d_inum);
241 	} else {
242 		target->subvol	= le32_to_cpu(d.v->d_child_subvol);
243 
244 		ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
245 
246 		target->inum	= le64_to_cpu(s.inode);
247 	}
248 
249 	return ret;
250 }
251 
252 int bch2_dirent_rename(struct btree_trans *trans,
253 		subvol_inum src_dir, struct bch_hash_info *src_hash,
254 		subvol_inum dst_dir, struct bch_hash_info *dst_hash,
255 		const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
256 		const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
257 		enum bch_rename_mode mode)
258 {
259 	struct btree_iter src_iter = { NULL };
260 	struct btree_iter dst_iter = { NULL };
261 	struct bkey_s_c old_src, old_dst = bkey_s_c_null;
262 	struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
263 	struct bpos dst_pos =
264 		POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
265 	unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
266 	int ret = 0;
267 
268 	if (src_dir.subvol != dst_dir.subvol)
269 		return -EXDEV;
270 
271 	memset(src_inum, 0, sizeof(*src_inum));
272 	memset(dst_inum, 0, sizeof(*dst_inum));
273 
274 	/* Lookup src: */
275 	ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
276 			       src_hash, src_dir, src_name,
277 			       BTREE_ITER_INTENT);
278 	if (ret)
279 		goto out;
280 
281 	old_src = bch2_btree_iter_peek_slot(&src_iter);
282 	ret = bkey_err(old_src);
283 	if (ret)
284 		goto out;
285 
286 	ret = bch2_dirent_read_target(trans, src_dir,
287 			bkey_s_c_to_dirent(old_src), src_inum);
288 	if (ret)
289 		goto out;
290 
291 	src_type = bkey_s_c_to_dirent(old_src).v->d_type;
292 
293 	if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
294 		return -EOPNOTSUPP;
295 
296 
297 	/* Lookup dst: */
298 	if (mode == BCH_RENAME) {
299 		/*
300 		 * Note that we're _not_ checking if the target already exists -
301 		 * we're relying on the VFS to do that check for us for
302 		 * correctness:
303 		 */
304 		ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
305 				     dst_hash, dst_dir, dst_name);
306 		if (ret)
307 			goto out;
308 	} else {
309 		ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
310 				       dst_hash, dst_dir, dst_name,
311 				       BTREE_ITER_INTENT);
312 		if (ret)
313 			goto out;
314 
315 		old_dst = bch2_btree_iter_peek_slot(&dst_iter);
316 		ret = bkey_err(old_dst);
317 		if (ret)
318 			goto out;
319 
320 		ret = bch2_dirent_read_target(trans, dst_dir,
321 				bkey_s_c_to_dirent(old_dst), dst_inum);
322 		if (ret)
323 			goto out;
324 
325 		dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
326 
327 		if (dst_type == DT_SUBVOL)
328 			return -EOPNOTSUPP;
329 	}
330 
331 	if (mode != BCH_RENAME_EXCHANGE)
332 		*src_offset = dst_iter.pos.offset;
333 
334 	/* Create new dst key: */
335 	new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
336 	ret = PTR_ERR_OR_ZERO(new_dst);
337 	if (ret)
338 		goto out;
339 
340 	dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
341 	new_dst->k.p = dst_iter.pos;
342 
343 	/* Create new src key: */
344 	if (mode == BCH_RENAME_EXCHANGE) {
345 		new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
346 		ret = PTR_ERR_OR_ZERO(new_src);
347 		if (ret)
348 			goto out;
349 
350 		dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
351 		new_src->k.p = src_iter.pos;
352 	} else {
353 		new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
354 		ret = PTR_ERR_OR_ZERO(new_src);
355 		if (ret)
356 			goto out;
357 
358 		bkey_init(&new_src->k);
359 		new_src->k.p = src_iter.pos;
360 
361 		if (bkey_le(dst_pos, src_iter.pos) &&
362 		    bkey_lt(src_iter.pos, dst_iter.pos)) {
363 			/*
364 			 * We have a hash collision for the new dst key,
365 			 * and new_src - the key we're deleting - is between
366 			 * new_dst's hashed slot and the slot we're going to be
367 			 * inserting it into - oops.  This will break the hash
368 			 * table if we don't deal with it:
369 			 */
370 			if (mode == BCH_RENAME) {
371 				/*
372 				 * If we're not overwriting, we can just insert
373 				 * new_dst at the src position:
374 				 */
375 				new_src = new_dst;
376 				new_src->k.p = src_iter.pos;
377 				goto out_set_src;
378 			} else {
379 				/* If we're overwriting, we can't insert new_dst
380 				 * at a different slot because it has to
381 				 * overwrite old_dst - just make sure to use a
382 				 * whiteout when deleting src:
383 				 */
384 				new_src->k.type = KEY_TYPE_hash_whiteout;
385 			}
386 		} else {
387 			/* Check if we need a whiteout to delete src: */
388 			ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
389 						       src_hash, &src_iter);
390 			if (ret < 0)
391 				goto out;
392 
393 			if (ret)
394 				new_src->k.type = KEY_TYPE_hash_whiteout;
395 		}
396 	}
397 
398 	ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
399 	if (ret)
400 		goto out;
401 out_set_src:
402 
403 	/*
404 	 * If we're deleting a subvolume, we need to really delete the dirent,
405 	 * not just emit a whiteout in the current snapshot:
406 	 */
407 	if (src_type == DT_SUBVOL) {
408 		bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
409 		ret = bch2_btree_iter_traverse(&src_iter);
410 		if (ret)
411 			goto out;
412 
413 		new_src->k.p = src_iter.pos;
414 		src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
415 	}
416 
417 	ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
418 	if (ret)
419 		goto out;
420 
421 	if (mode == BCH_RENAME_EXCHANGE)
422 		*src_offset = new_src->k.p.offset;
423 	*dst_offset = new_dst->k.p.offset;
424 out:
425 	bch2_trans_iter_exit(trans, &src_iter);
426 	bch2_trans_iter_exit(trans, &dst_iter);
427 	return ret;
428 }
429 
430 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
431 			       struct btree_iter *iter,
432 			       subvol_inum dir,
433 			       const struct bch_hash_info *hash_info,
434 			       const struct qstr *name, subvol_inum *inum,
435 			       unsigned flags)
436 {
437 	struct bkey_s_c k;
438 	struct bkey_s_c_dirent d;
439 	u32 snapshot;
440 	int ret;
441 
442 	ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
443 	if (ret)
444 		return ret;
445 
446 	ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
447 			       hash_info, dir, name, flags);
448 	if (ret)
449 		return ret;
450 
451 	k = bch2_btree_iter_peek_slot(iter);
452 	ret = bkey_err(k);
453 	if (ret)
454 		goto err;
455 
456 	d = bkey_s_c_to_dirent(k);
457 
458 	ret = bch2_dirent_read_target(trans, dir, d, inum);
459 	if (ret > 0)
460 		ret = -ENOENT;
461 err:
462 	if (ret)
463 		bch2_trans_iter_exit(trans, iter);
464 
465 	return ret;
466 }
467 
468 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
469 		       const struct bch_hash_info *hash_info,
470 		       const struct qstr *name, subvol_inum *inum)
471 {
472 	struct btree_trans *trans = bch2_trans_get(c);
473 	struct btree_iter iter;
474 	int ret;
475 retry:
476 	bch2_trans_begin(trans);
477 
478 	ret = __bch2_dirent_lookup_trans(trans, &iter, dir, hash_info,
479 					  name, inum, 0);
480 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
481 		goto retry;
482 	if (!ret)
483 		bch2_trans_iter_exit(trans, &iter);
484 	bch2_trans_put(trans);
485 	return ret;
486 }
487 
488 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
489 {
490 	struct btree_iter iter;
491 	struct bkey_s_c k;
492 	u32 snapshot;
493 	int ret;
494 
495 	ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
496 	if (ret)
497 		return ret;
498 
499 	for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
500 			   SPOS(dir.inum, 0, snapshot),
501 			   POS(dir.inum, U64_MAX), 0, k, ret)
502 		if (k.k->type == KEY_TYPE_dirent) {
503 			ret = -ENOTEMPTY;
504 			break;
505 		}
506 	bch2_trans_iter_exit(trans, &iter);
507 
508 	return ret;
509 }
510 
511 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
512 {
513 	struct btree_trans *trans = bch2_trans_get(c);
514 	struct btree_iter iter;
515 	struct bkey_s_c k;
516 	struct bkey_s_c_dirent dirent;
517 	subvol_inum target;
518 	u32 snapshot;
519 	struct bkey_buf sk;
520 	struct qstr name;
521 	int ret;
522 
523 	bch2_bkey_buf_init(&sk);
524 retry:
525 	bch2_trans_begin(trans);
526 
527 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
528 	if (ret)
529 		goto err;
530 
531 	for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
532 			   SPOS(inum.inum, ctx->pos, snapshot),
533 			   POS(inum.inum, U64_MAX), 0, k, ret) {
534 		if (k.k->type != KEY_TYPE_dirent)
535 			continue;
536 
537 		dirent = bkey_s_c_to_dirent(k);
538 
539 		ret = bch2_dirent_read_target(trans, inum, dirent, &target);
540 		if (ret < 0)
541 			break;
542 		if (ret)
543 			continue;
544 
545 		/* dir_emit() can fault and block: */
546 		bch2_bkey_buf_reassemble(&sk, c, k);
547 		dirent = bkey_i_to_s_c_dirent(sk.k);
548 		bch2_trans_unlock(trans);
549 
550 		name = bch2_dirent_get_name(dirent);
551 
552 		ctx->pos = dirent.k->p.offset;
553 		if (!dir_emit(ctx, name.name,
554 			      name.len,
555 			      target.inum,
556 			      vfs_d_type(dirent.v->d_type)))
557 			break;
558 		ctx->pos = dirent.k->p.offset + 1;
559 
560 		/*
561 		 * read_target looks up subvolumes, we can overflow paths if the
562 		 * directory has many subvolumes in it
563 		 */
564 		ret = btree_trans_too_many_iters(trans);
565 		if (ret)
566 			break;
567 	}
568 	bch2_trans_iter_exit(trans, &iter);
569 err:
570 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
571 		goto retry;
572 
573 	bch2_trans_put(trans);
574 	bch2_bkey_buf_exit(&sk, c);
575 
576 	return ret;
577 }
578