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,
150 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
151 struct qstr d_name = bch2_dirent_get_name(d);
153 prt_printf(out, "%.*s -> %llu type %s",
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));
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)
166 struct bkey_i_dirent *dirent;
167 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
169 if (name->len > BCH_NAME_MAX)
170 return ERR_PTR(-ENAMETOOLONG);
172 BUG_ON(u64s > U8_MAX);
174 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
178 bkey_dirent_init(&dirent->k_i);
179 dirent->k.u64s = u64s;
181 if (type != DT_SUBVOL) {
182 dirent->v.d_inum = cpu_to_le64(dst);
184 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
185 dirent->v.d_child_subvol = cpu_to_le32(dst);
188 dirent->v.d_type = type;
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) -
196 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
201 int bch2_dirent_create_snapshot(struct btree_trans *trans,
202 u64 dir, u32 snapshot,
203 const struct bch_hash_info *hash_info,
204 u8 type, const struct qstr *name, u64 dst_inum,
206 bch_str_hash_flags_t str_hash_flags)
208 subvol_inum zero_inum = { 0 };
209 struct bkey_i_dirent *dirent;
212 dirent = dirent_create_key(trans, zero_inum, type, name, dst_inum);
213 ret = PTR_ERR_OR_ZERO(dirent);
217 dirent->k.p.inode = dir;
218 dirent->k.p.snapshot = snapshot;
220 ret = bch2_hash_set_snapshot(trans, bch2_dirent_hash_desc, hash_info,
222 &dirent->k_i, str_hash_flags,
223 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
224 *dir_offset = dirent->k.p.offset;
229 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
230 const struct bch_hash_info *hash_info,
231 u8 type, const struct qstr *name, u64 dst_inum,
233 bch_str_hash_flags_t str_hash_flags)
235 struct bkey_i_dirent *dirent;
238 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
239 ret = PTR_ERR_OR_ZERO(dirent);
243 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
244 dir, &dirent->k_i, str_hash_flags);
245 *dir_offset = dirent->k.p.offset;
250 static void dirent_copy_target(struct bkey_i_dirent *dst,
251 struct bkey_s_c_dirent src)
253 dst->v.d_inum = src.v->d_inum;
254 dst->v.d_type = src.v->d_type;
257 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
258 struct bkey_s_c_dirent d, subvol_inum *target)
260 struct bch_subvolume s;
263 if (d.v->d_type == DT_SUBVOL &&
264 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
267 if (likely(d.v->d_type != DT_SUBVOL)) {
268 target->subvol = dir.subvol;
269 target->inum = le64_to_cpu(d.v->d_inum);
271 target->subvol = le32_to_cpu(d.v->d_child_subvol);
273 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
275 target->inum = le64_to_cpu(s.inode);
281 int bch2_dirent_rename(struct btree_trans *trans,
282 subvol_inum src_dir, struct bch_hash_info *src_hash,
283 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
284 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
285 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
286 enum bch_rename_mode mode)
288 struct btree_iter src_iter = { NULL };
289 struct btree_iter dst_iter = { NULL };
290 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
291 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
292 struct bpos dst_pos =
293 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
294 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
297 if (src_dir.subvol != dst_dir.subvol)
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);
320 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
322 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
327 if (mode == BCH_RENAME) {
329 * Note that we're _not_ checking if the target already exists -
330 * we're relying on the VFS to do that check for us for
333 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
334 dst_hash, dst_dir, dst_name);
338 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
339 dst_hash, dst_dir, dst_name,
344 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
345 ret = bkey_err(old_dst);
349 ret = bch2_dirent_read_target(trans, dst_dir,
350 bkey_s_c_to_dirent(old_dst), dst_inum);
354 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
356 if (dst_type == DT_SUBVOL)
360 if (mode != BCH_RENAME_EXCHANGE)
361 *src_offset = dst_iter.pos.offset;
363 /* Create new dst key: */
364 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
365 ret = PTR_ERR_OR_ZERO(new_dst);
369 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
370 new_dst->k.p = dst_iter.pos;
372 /* Create new src key: */
373 if (mode == BCH_RENAME_EXCHANGE) {
374 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
375 ret = PTR_ERR_OR_ZERO(new_src);
379 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
380 new_src->k.p = src_iter.pos;
382 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
383 ret = PTR_ERR_OR_ZERO(new_src);
387 bkey_init(&new_src->k);
388 new_src->k.p = src_iter.pos;
390 if (bkey_le(dst_pos, src_iter.pos) &&
391 bkey_lt(src_iter.pos, dst_iter.pos)) {
393 * We have a hash collision for the new dst key,
394 * and new_src - the key we're deleting - is between
395 * new_dst's hashed slot and the slot we're going to be
396 * inserting it into - oops. This will break the hash
397 * table if we don't deal with it:
399 if (mode == BCH_RENAME) {
401 * If we're not overwriting, we can just insert
402 * new_dst at the src position:
405 new_src->k.p = src_iter.pos;
408 /* If we're overwriting, we can't insert new_dst
409 * at a different slot because it has to
410 * overwrite old_dst - just make sure to use a
411 * whiteout when deleting src:
413 new_src->k.type = KEY_TYPE_hash_whiteout;
416 /* Check if we need a whiteout to delete src: */
417 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
418 src_hash, &src_iter);
423 new_src->k.type = KEY_TYPE_hash_whiteout;
427 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
433 * If we're deleting a subvolume, we need to really delete the dirent,
434 * not just emit a whiteout in the current snapshot:
436 if (src_type == DT_SUBVOL) {
437 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
438 ret = bch2_btree_iter_traverse(&src_iter);
442 new_src->k.p = src_iter.pos;
443 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
446 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
450 if (mode == BCH_RENAME_EXCHANGE)
451 *src_offset = new_src->k.p.offset;
452 *dst_offset = new_dst->k.p.offset;
454 bch2_trans_iter_exit(trans, &src_iter);
455 bch2_trans_iter_exit(trans, &dst_iter);
459 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
460 struct btree_iter *iter,
462 const struct bch_hash_info *hash_info,
463 const struct qstr *name, subvol_inum *inum,
467 struct bkey_s_c_dirent d;
471 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
475 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
476 hash_info, dir, name, flags);
480 k = bch2_btree_iter_peek_slot(iter);
485 d = bkey_s_c_to_dirent(k);
487 ret = bch2_dirent_read_target(trans, dir, d, inum);
492 bch2_trans_iter_exit(trans, iter);
497 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
498 const struct bch_hash_info *hash_info,
499 const struct qstr *name, subvol_inum *inum)
501 struct btree_trans *trans = bch2_trans_get(c);
502 struct btree_iter iter = { NULL };
504 int ret = lockrestart_do(trans,
505 __bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
506 bch2_trans_iter_exit(trans, &iter);
507 bch2_trans_put(trans);
511 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 snapshot)
513 struct btree_iter iter;
517 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
518 SPOS(dir, 0, snapshot),
519 POS(dir, U64_MAX), 0, k, ret)
520 if (k.k->type == KEY_TYPE_dirent) {
524 bch2_trans_iter_exit(trans, &iter);
529 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
533 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
534 bch2_empty_dir_snapshot(trans, dir.inum, snapshot);
537 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
539 struct btree_trans *trans = bch2_trans_get(c);
540 struct btree_iter iter;
542 struct bkey_s_c_dirent dirent;
549 bch2_bkey_buf_init(&sk);
551 bch2_trans_begin(trans);
553 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
557 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
558 SPOS(inum.inum, ctx->pos, snapshot),
559 POS(inum.inum, U64_MAX), 0, k, ret) {
560 if (k.k->type != KEY_TYPE_dirent)
563 dirent = bkey_s_c_to_dirent(k);
565 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
571 /* dir_emit() can fault and block: */
572 bch2_bkey_buf_reassemble(&sk, c, k);
573 dirent = bkey_i_to_s_c_dirent(sk.k);
574 bch2_trans_unlock(trans);
576 name = bch2_dirent_get_name(dirent);
578 ctx->pos = dirent.k->p.offset;
579 if (!dir_emit(ctx, name.name,
582 vfs_d_type(dirent.v->d_type)))
584 ctx->pos = dirent.k->p.offset + 1;
587 * read_target looks up subvolumes, we can overflow paths if the
588 * directory has many subvolumes in it
590 ret = btree_trans_too_many_iters(trans);
594 bch2_trans_iter_exit(trans, &iter);
596 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
599 bch2_trans_put(trans);
600 bch2_bkey_buf_exit(&sk, c);