1 // SPDX-License-Identifier: GPL-2.0
5 #include "bkey_methods.h"
6 #include "btree_update.h"
12 #include "subvolume.h"
14 #include <linux/dcache.h>
16 static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
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];
22 unsigned trailing_nuls = last_u64 ? __builtin_ctzll(last_u64) / 8 : 64 / 8;
24 unsigned trailing_nuls = last_u64 ? __builtin_clzll(last_u64) / 8 : 64 / 8;
28 offsetof(struct bch_dirent, d_name) -
32 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
34 return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
37 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
38 const struct qstr *name)
40 struct bch_str_hash_ctx ctx;
42 bch2_str_hash_init(&ctx, info);
43 bch2_str_hash_update(&ctx, info, name->name, name->len);
45 /* [0,2) reserved for dots */
46 return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
49 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
51 return bch2_dirent_hash(info, key);
54 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
56 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
57 struct qstr name = bch2_dirent_get_name(d);
59 return bch2_dirent_hash(info, &name);
62 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
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;
68 return !qstr_eq(l_name, *r_name);
71 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
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);
78 return !qstr_eq(l_name, r_name);
81 static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
83 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
85 if (d.v->d_type == DT_SUBVOL)
86 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
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,
100 int bch2_dirent_invalid(struct bch_fs *c, struct bkey_s_c k,
101 enum bkey_invalid_flags flags,
102 struct printbuf *err)
104 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
105 struct qstr d_name = bch2_dirent_get_name(d);
108 bkey_fsck_err_on(!d_name.len, c, err,
112 bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len), c, err,
114 "value too big (%zu > %u)",
115 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
118 * Check new keys don't exceed the max length
119 * (older keys may be larger.)
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);
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");
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,
135 bkey_fsck_err_on(memchr(d_name.name, '/', d_name.len), c, err,
136 dirent_name_has_slash,
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,
142 "dirent points to own directory");
147 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
149 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
150 struct qstr d_name = bch2_dirent_get_name(d);
152 prt_printf(out, "%.*s -> ", d_name.len, d_name.name);
154 if (d.v->d_type != DT_SUBVOL)
155 prt_printf(out, "%llu", le64_to_cpu(d.v->d_inum));
157 prt_printf(out, "%u -> %u",
158 le32_to_cpu(d.v->d_parent_subvol),
159 le32_to_cpu(d.v->d_child_subvol));
161 prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
164 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
165 subvol_inum dir, u8 type,
166 const struct qstr *name, u64 dst)
168 struct bkey_i_dirent *dirent;
169 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
171 if (name->len > BCH_NAME_MAX)
172 return ERR_PTR(-ENAMETOOLONG);
174 BUG_ON(u64s > U8_MAX);
176 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
180 bkey_dirent_init(&dirent->k_i);
181 dirent->k.u64s = u64s;
183 if (type != DT_SUBVOL) {
184 dirent->v.d_inum = cpu_to_le64(dst);
186 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
187 dirent->v.d_child_subvol = cpu_to_le32(dst);
190 dirent->v.d_type = type;
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) -
198 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
203 int 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,
208 bch_str_hash_flags_t str_hash_flags)
210 subvol_inum dir_inum = { .subvol = dir_subvol, .inum = dir };
211 struct bkey_i_dirent *dirent;
214 dirent = dirent_create_key(trans, dir_inum, type, name, dst_inum);
215 ret = PTR_ERR_OR_ZERO(dirent);
219 dirent->k.p.inode = dir;
220 dirent->k.p.snapshot = snapshot;
222 ret = bch2_hash_set_in_snapshot(trans, bch2_dirent_hash_desc, hash_info,
224 &dirent->k_i, str_hash_flags,
225 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
226 *dir_offset = dirent->k.p.offset;
231 int 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,
235 bch_str_hash_flags_t str_hash_flags)
237 struct bkey_i_dirent *dirent;
240 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
241 ret = PTR_ERR_OR_ZERO(dirent);
245 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
246 dir, &dirent->k_i, str_hash_flags);
247 *dir_offset = dirent->k.p.offset;
252 static void dirent_copy_target(struct bkey_i_dirent *dst,
253 struct bkey_s_c_dirent src)
255 dst->v.d_inum = src.v->d_inum;
256 dst->v.d_type = src.v->d_type;
259 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
260 struct bkey_s_c_dirent d, subvol_inum *target)
262 struct bch_subvolume s;
265 if (d.v->d_type == DT_SUBVOL &&
266 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
269 if (likely(d.v->d_type != DT_SUBVOL)) {
270 target->subvol = dir.subvol;
271 target->inum = le64_to_cpu(d.v->d_inum);
273 target->subvol = le32_to_cpu(d.v->d_child_subvol);
275 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
277 target->inum = le64_to_cpu(s.inode);
283 int 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)
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;
300 memset(src_inum, 0, sizeof(*src_inum));
301 memset(dst_inum, 0, sizeof(*dst_inum));
304 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
305 src_hash, src_dir, src_name,
310 old_src = bch2_btree_iter_peek_slot(&src_iter);
311 ret = bkey_err(old_src);
315 ret = bch2_dirent_read_target(trans, src_dir,
316 bkey_s_c_to_dirent(old_src), src_inum);
321 if (mode == BCH_RENAME) {
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
327 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
328 dst_hash, dst_dir, dst_name);
332 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
333 dst_hash, dst_dir, dst_name,
338 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
339 ret = bkey_err(old_dst);
343 ret = bch2_dirent_read_target(trans, dst_dir,
344 bkey_s_c_to_dirent(old_dst), dst_inum);
349 if (mode != BCH_RENAME_EXCHANGE)
350 *src_offset = dst_iter.pos.offset;
352 /* Create new dst key: */
353 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
354 ret = PTR_ERR_OR_ZERO(new_dst);
358 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
359 new_dst->k.p = dst_iter.pos;
361 /* Create new src key: */
362 if (mode == BCH_RENAME_EXCHANGE) {
363 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
364 ret = PTR_ERR_OR_ZERO(new_src);
368 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
369 new_src->k.p = src_iter.pos;
371 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
372 ret = PTR_ERR_OR_ZERO(new_src);
376 bkey_init(&new_src->k);
377 new_src->k.p = src_iter.pos;
379 if (bkey_le(dst_pos, src_iter.pos) &&
380 bkey_lt(src_iter.pos, dst_iter.pos)) {
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:
388 if (mode == BCH_RENAME) {
390 * If we're not overwriting, we can just insert
391 * new_dst at the src position:
394 new_src->k.p = src_iter.pos;
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:
402 new_src->k.type = KEY_TYPE_hash_whiteout;
405 /* Check if we need a whiteout to delete src: */
406 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
407 src_hash, &src_iter);
412 new_src->k.type = KEY_TYPE_hash_whiteout;
416 if (new_dst->v.d_type == DT_SUBVOL)
417 new_dst->v.d_parent_subvol = cpu_to_le32(dst_dir.subvol);
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);
423 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
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.
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.
437 delete_src = bkey_s_c_to_dirent(old_src).v->d_type == DT_SUBVOL &&
438 new_src->k.p.snapshot != old_src.k->p.snapshot;
440 delete_dst = old_dst.k &&
441 bkey_s_c_to_dirent(old_dst).v->d_type == DT_SUBVOL &&
442 new_dst->k.p.snapshot != old_dst.k->p.snapshot;
444 if (!delete_src || !bkey_deleted(&new_src->k)) {
445 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
451 bch2_btree_iter_set_snapshot(&src_iter, 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);
459 bch2_btree_iter_set_snapshot(&dst_iter, 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);
466 if (mode == BCH_RENAME_EXCHANGE)
467 *src_offset = new_src->k.p.offset;
468 *dst_offset = new_dst->k.p.offset;
470 bch2_trans_iter_exit(trans, &src_iter);
471 bch2_trans_iter_exit(trans, &dst_iter);
475 int bch2_dirent_lookup_trans(struct btree_trans *trans,
476 struct btree_iter *iter,
478 const struct bch_hash_info *hash_info,
479 const struct qstr *name, subvol_inum *inum,
482 int ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
483 hash_info, dir, name, flags);
487 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
492 ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), inum);
497 bch2_trans_iter_exit(trans, iter);
501 u64 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)
505 struct btree_trans *trans = bch2_trans_get(c);
506 struct btree_iter iter = { NULL };
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);
515 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32 snapshot)
517 struct btree_iter iter;
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)
528 ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
531 bch2_trans_iter_exit(trans, &iter);
536 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
540 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
541 bch2_empty_dir_snapshot(trans, dir.inum, dir.subvol, snapshot);
544 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
546 struct btree_trans *trans = bch2_trans_get(c);
547 struct btree_iter iter;
549 struct bkey_s_c_dirent dirent;
556 bch2_bkey_buf_init(&sk);
558 bch2_trans_begin(trans);
560 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
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)
570 dirent = bkey_s_c_to_dirent(k);
572 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
578 /* dir_emit() can fault and block: */
579 bch2_bkey_buf_reassemble(&sk, c, k);
580 dirent = bkey_i_to_s_c_dirent(sk.k);
581 bch2_trans_unlock(trans);
583 name = bch2_dirent_get_name(dirent);
585 ctx->pos = dirent.k->p.offset;
586 if (!dir_emit(ctx, name.name,
589 vfs_d_type(dirent.v->d_type)))
591 ctx->pos = dirent.k->p.offset + 1;
594 * read_target looks up subvolumes, we can overflow paths if the
595 * directory has many subvolumes in it
597 ret = btree_trans_too_many_iters(trans);
601 bch2_trans_iter_exit(trans, &iter);
603 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
606 bch2_trans_put(trans);
607 bch2_bkey_buf_exit(&sk, c);