xref: /linux/fs/bcachefs/dirent.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
254 			    struct bkey_s_c_dirent d, subvol_inum *target)
255 {
256 	struct bch_subvolume s;
257 	int ret = 0;
258 
259 	if (d.v->d_type == DT_SUBVOL &&
260 	    le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
261 		return 1;
262 
263 	if (likely(d.v->d_type != DT_SUBVOL)) {
264 		target->subvol	= dir.subvol;
265 		target->inum	= le64_to_cpu(d.v->d_inum);
266 	} else {
267 		target->subvol	= le32_to_cpu(d.v->d_child_subvol);
268 
269 		ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_cached, &s);
270 
271 		target->inum	= le64_to_cpu(s.inode);
272 	}
273 
274 	return ret;
275 }
276 
277 int bch2_dirent_rename(struct btree_trans *trans,
278 		subvol_inum src_dir, struct bch_hash_info *src_hash,
279 		subvol_inum dst_dir, struct bch_hash_info *dst_hash,
280 		const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
281 		const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
282 		enum bch_rename_mode mode)
283 {
284 	struct btree_iter src_iter = { NULL };
285 	struct btree_iter dst_iter = { NULL };
286 	struct bkey_s_c old_src, old_dst = bkey_s_c_null;
287 	struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
288 	struct bpos dst_pos =
289 		POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
290 	unsigned src_update_flags = 0;
291 	bool delete_src, delete_dst;
292 	int ret = 0;
293 
294 	memset(src_inum, 0, sizeof(*src_inum));
295 	memset(dst_inum, 0, sizeof(*dst_inum));
296 
297 	/* Lookup src: */
298 	old_src = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
299 				   src_hash, src_dir, src_name,
300 				   BTREE_ITER_intent);
301 	ret = bkey_err(old_src);
302 	if (ret)
303 		goto out;
304 
305 	ret = bch2_dirent_read_target(trans, src_dir,
306 			bkey_s_c_to_dirent(old_src), src_inum);
307 	if (ret)
308 		goto out;
309 
310 	/* Lookup dst: */
311 	if (mode == BCH_RENAME) {
312 		/*
313 		 * Note that we're _not_ checking if the target already exists -
314 		 * we're relying on the VFS to do that check for us for
315 		 * correctness:
316 		 */
317 		ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
318 				     dst_hash, dst_dir, dst_name);
319 		if (ret)
320 			goto out;
321 	} else {
322 		old_dst = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
323 					    dst_hash, dst_dir, dst_name,
324 					    BTREE_ITER_intent);
325 		ret = bkey_err(old_dst);
326 		if (ret)
327 			goto out;
328 
329 		ret = bch2_dirent_read_target(trans, dst_dir,
330 				bkey_s_c_to_dirent(old_dst), dst_inum);
331 		if (ret)
332 			goto out;
333 	}
334 
335 	if (mode != BCH_RENAME_EXCHANGE)
336 		*src_offset = dst_iter.pos.offset;
337 
338 	/* Create new dst key: */
339 	new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
340 	ret = PTR_ERR_OR_ZERO(new_dst);
341 	if (ret)
342 		goto out;
343 
344 	dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
345 	new_dst->k.p = dst_iter.pos;
346 
347 	/* Create new src key: */
348 	if (mode == BCH_RENAME_EXCHANGE) {
349 		new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
350 		ret = PTR_ERR_OR_ZERO(new_src);
351 		if (ret)
352 			goto out;
353 
354 		dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
355 		new_src->k.p = src_iter.pos;
356 	} else {
357 		new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
358 		ret = PTR_ERR_OR_ZERO(new_src);
359 		if (ret)
360 			goto out;
361 
362 		bkey_init(&new_src->k);
363 		new_src->k.p = src_iter.pos;
364 
365 		if (bkey_le(dst_pos, src_iter.pos) &&
366 		    bkey_lt(src_iter.pos, dst_iter.pos)) {
367 			/*
368 			 * We have a hash collision for the new dst key,
369 			 * and new_src - the key we're deleting - is between
370 			 * new_dst's hashed slot and the slot we're going to be
371 			 * inserting it into - oops.  This will break the hash
372 			 * table if we don't deal with it:
373 			 */
374 			if (mode == BCH_RENAME) {
375 				/*
376 				 * If we're not overwriting, we can just insert
377 				 * new_dst at the src position:
378 				 */
379 				new_src = new_dst;
380 				new_src->k.p = src_iter.pos;
381 				goto out_set_src;
382 			} else {
383 				/* If we're overwriting, we can't insert new_dst
384 				 * at a different slot because it has to
385 				 * overwrite old_dst - just make sure to use a
386 				 * whiteout when deleting src:
387 				 */
388 				new_src->k.type = KEY_TYPE_hash_whiteout;
389 			}
390 		} else {
391 			/* Check if we need a whiteout to delete src: */
392 			ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
393 						       src_hash, &src_iter);
394 			if (ret < 0)
395 				goto out;
396 
397 			if (ret)
398 				new_src->k.type = KEY_TYPE_hash_whiteout;
399 		}
400 	}
401 
402 	if (new_dst->v.d_type == DT_SUBVOL)
403 		new_dst->v.d_parent_subvol = cpu_to_le32(dst_dir.subvol);
404 
405 	if ((mode == BCH_RENAME_EXCHANGE) &&
406 	    new_src->v.d_type == DT_SUBVOL)
407 		new_src->v.d_parent_subvol = cpu_to_le32(src_dir.subvol);
408 
409 	ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
410 	if (ret)
411 		goto out;
412 out_set_src:
413 	/*
414 	 * If we're deleting a subvolume we need to really delete the dirent,
415 	 * not just emit a whiteout in the current snapshot - there can only be
416 	 * single dirent that points to a given subvolume.
417 	 *
418 	 * IOW, we don't maintain multiple versions in different snapshots of
419 	 * dirents that point to subvolumes - dirents that point to subvolumes
420 	 * are only visible in one particular subvolume so it's not necessary,
421 	 * and it would be particularly confusing for fsck to have to deal with.
422 	 */
423 	delete_src = bkey_s_c_to_dirent(old_src).v->d_type == DT_SUBVOL &&
424 		new_src->k.p.snapshot != old_src.k->p.snapshot;
425 
426 	delete_dst = old_dst.k &&
427 		bkey_s_c_to_dirent(old_dst).v->d_type == DT_SUBVOL &&
428 		new_dst->k.p.snapshot != old_dst.k->p.snapshot;
429 
430 	if (!delete_src || !bkey_deleted(&new_src->k)) {
431 		ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
432 		if (ret)
433 			goto out;
434 	}
435 
436 	if (delete_src) {
437 		bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
438 		ret =   bch2_btree_iter_traverse(&src_iter) ?:
439 			bch2_btree_delete_at(trans, &src_iter, BTREE_UPDATE_internal_snapshot_node);
440 		if (ret)
441 			goto out;
442 	}
443 
444 	if (delete_dst) {
445 		bch2_btree_iter_set_snapshot(&dst_iter, old_dst.k->p.snapshot);
446 		ret =   bch2_btree_iter_traverse(&dst_iter) ?:
447 			bch2_btree_delete_at(trans, &dst_iter, BTREE_UPDATE_internal_snapshot_node);
448 		if (ret)
449 			goto out;
450 	}
451 
452 	if (mode == BCH_RENAME_EXCHANGE)
453 		*src_offset = new_src->k.p.offset;
454 	*dst_offset = new_dst->k.p.offset;
455 out:
456 	bch2_trans_iter_exit(trans, &src_iter);
457 	bch2_trans_iter_exit(trans, &dst_iter);
458 	return ret;
459 }
460 
461 int bch2_dirent_lookup_trans(struct btree_trans *trans,
462 			     struct btree_iter *iter,
463 			     subvol_inum dir,
464 			     const struct bch_hash_info *hash_info,
465 			     const struct qstr *name, subvol_inum *inum,
466 			     unsigned flags)
467 {
468 	struct bkey_s_c k = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
469 					     hash_info, dir, name, flags);
470 	int ret = bkey_err(k);
471 	if (ret)
472 		goto err;
473 
474 	ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), inum);
475 	if (ret > 0)
476 		ret = -ENOENT;
477 err:
478 	if (ret)
479 		bch2_trans_iter_exit(trans, iter);
480 	return ret;
481 }
482 
483 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
484 		       const struct bch_hash_info *hash_info,
485 		       const struct qstr *name, subvol_inum *inum)
486 {
487 	struct btree_trans *trans = bch2_trans_get(c);
488 	struct btree_iter iter = { NULL };
489 
490 	int ret = lockrestart_do(trans,
491 		bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
492 	bch2_trans_iter_exit(trans, &iter);
493 	bch2_trans_put(trans);
494 	return ret;
495 }
496 
497 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32 snapshot)
498 {
499 	struct btree_iter iter;
500 	struct bkey_s_c k;
501 	int ret;
502 
503 	for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
504 			   SPOS(dir, 0, snapshot),
505 			   POS(dir, U64_MAX), 0, k, ret)
506 		if (k.k->type == KEY_TYPE_dirent) {
507 			struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
508 			if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
509 				continue;
510 			ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
511 			break;
512 		}
513 	bch2_trans_iter_exit(trans, &iter);
514 
515 	return ret;
516 }
517 
518 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
519 {
520 	u32 snapshot;
521 
522 	return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
523 		bch2_empty_dir_snapshot(trans, dir.inum, dir.subvol, snapshot);
524 }
525 
526 static int bch2_dir_emit(struct dir_context *ctx, struct bkey_s_c_dirent d, subvol_inum target)
527 {
528 	struct qstr name = bch2_dirent_get_name(d);
529 	/*
530 	 * Although not required by the kernel code, updating ctx->pos is needed
531 	 * for the bcachefs FUSE driver. Without this update, the FUSE
532 	 * implementation will be stuck in an infinite loop when reading
533 	 * directories (via the bcachefs_fuse_readdir callback).
534 	 * In kernel space, ctx->pos is updated by the VFS code.
535 	 */
536 	ctx->pos = d.k->p.offset;
537 	bool ret = dir_emit(ctx, name.name,
538 		      name.len,
539 		      target.inum,
540 		      vfs_d_type(d.v->d_type));
541 	if (ret)
542 		ctx->pos = d.k->p.offset + 1;
543 	return ret;
544 }
545 
546 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
547 {
548 	struct bkey_buf sk;
549 	bch2_bkey_buf_init(&sk);
550 
551 	int ret = bch2_trans_run(c,
552 		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_dirents,
553 				   POS(inum.inum, ctx->pos),
554 				   POS(inum.inum, U64_MAX),
555 				   inum.subvol, 0, k, ({
556 			if (k.k->type != KEY_TYPE_dirent)
557 				continue;
558 
559 			/* dir_emit() can fault and block: */
560 			bch2_bkey_buf_reassemble(&sk, c, k);
561 			struct bkey_s_c_dirent dirent = bkey_i_to_s_c_dirent(sk.k);
562 
563 			subvol_inum target;
564 			int ret2 = bch2_dirent_read_target(trans, inum, dirent, &target);
565 			if (ret2 > 0)
566 				continue;
567 
568 			ret2 ?: drop_locks_do(trans, bch2_dir_emit(ctx, dirent, target));
569 		})));
570 
571 	bch2_bkey_buf_exit(&sk, c);
572 
573 	return ret < 0 ? ret : 0;
574 }
575