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 l_name.len - r_name->len ?: memcmp(l_name.name, r_name->name, l_name.len);
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 l_name.len - r_name.len ?: memcmp(l_name.name, r_name.name, l_name.len);
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(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,
205 bch_str_hash_flags_t str_hash_flags)
207 struct bkey_i_dirent *dirent;
210 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
211 ret = PTR_ERR_OR_ZERO(dirent);
215 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
216 dir, &dirent->k_i, str_hash_flags);
217 *dir_offset = dirent->k.p.offset;
222 static void dirent_copy_target(struct bkey_i_dirent *dst,
223 struct bkey_s_c_dirent src)
225 dst->v.d_inum = src.v->d_inum;
226 dst->v.d_type = src.v->d_type;
229 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
230 struct bkey_s_c_dirent d, subvol_inum *target)
232 struct bch_subvolume s;
235 if (d.v->d_type == DT_SUBVOL &&
236 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
239 if (likely(d.v->d_type != DT_SUBVOL)) {
240 target->subvol = dir.subvol;
241 target->inum = le64_to_cpu(d.v->d_inum);
243 target->subvol = le32_to_cpu(d.v->d_child_subvol);
245 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
247 target->inum = le64_to_cpu(s.inode);
253 int bch2_dirent_rename(struct btree_trans *trans,
254 subvol_inum src_dir, struct bch_hash_info *src_hash,
255 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
256 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
257 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
258 enum bch_rename_mode mode)
260 struct btree_iter src_iter = { NULL };
261 struct btree_iter dst_iter = { NULL };
262 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
263 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
264 struct bpos dst_pos =
265 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
266 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
269 if (src_dir.subvol != dst_dir.subvol)
272 memset(src_inum, 0, sizeof(*src_inum));
273 memset(dst_inum, 0, sizeof(*dst_inum));
276 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
277 src_hash, src_dir, src_name,
282 old_src = bch2_btree_iter_peek_slot(&src_iter);
283 ret = bkey_err(old_src);
287 ret = bch2_dirent_read_target(trans, src_dir,
288 bkey_s_c_to_dirent(old_src), src_inum);
292 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
294 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
299 if (mode == BCH_RENAME) {
301 * Note that we're _not_ checking if the target already exists -
302 * we're relying on the VFS to do that check for us for
305 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
306 dst_hash, dst_dir, dst_name);
310 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
311 dst_hash, dst_dir, dst_name,
316 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
317 ret = bkey_err(old_dst);
321 ret = bch2_dirent_read_target(trans, dst_dir,
322 bkey_s_c_to_dirent(old_dst), dst_inum);
326 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
328 if (dst_type == DT_SUBVOL)
332 if (mode != BCH_RENAME_EXCHANGE)
333 *src_offset = dst_iter.pos.offset;
335 /* Create new dst key: */
336 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
337 ret = PTR_ERR_OR_ZERO(new_dst);
341 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
342 new_dst->k.p = dst_iter.pos;
344 /* Create new src key: */
345 if (mode == BCH_RENAME_EXCHANGE) {
346 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
347 ret = PTR_ERR_OR_ZERO(new_src);
351 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
352 new_src->k.p = src_iter.pos;
354 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
355 ret = PTR_ERR_OR_ZERO(new_src);
359 bkey_init(&new_src->k);
360 new_src->k.p = src_iter.pos;
362 if (bkey_le(dst_pos, src_iter.pos) &&
363 bkey_lt(src_iter.pos, dst_iter.pos)) {
365 * We have a hash collision for the new dst key,
366 * and new_src - the key we're deleting - is between
367 * new_dst's hashed slot and the slot we're going to be
368 * inserting it into - oops. This will break the hash
369 * table if we don't deal with it:
371 if (mode == BCH_RENAME) {
373 * If we're not overwriting, we can just insert
374 * new_dst at the src position:
377 new_src->k.p = src_iter.pos;
380 /* If we're overwriting, we can't insert new_dst
381 * at a different slot because it has to
382 * overwrite old_dst - just make sure to use a
383 * whiteout when deleting src:
385 new_src->k.type = KEY_TYPE_hash_whiteout;
388 /* Check if we need a whiteout to delete src: */
389 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
390 src_hash, &src_iter);
395 new_src->k.type = KEY_TYPE_hash_whiteout;
399 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
405 * If we're deleting a subvolume, we need to really delete the dirent,
406 * not just emit a whiteout in the current snapshot:
408 if (src_type == DT_SUBVOL) {
409 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
410 ret = bch2_btree_iter_traverse(&src_iter);
414 new_src->k.p = src_iter.pos;
415 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
418 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
422 if (mode == BCH_RENAME_EXCHANGE)
423 *src_offset = new_src->k.p.offset;
424 *dst_offset = new_dst->k.p.offset;
426 bch2_trans_iter_exit(trans, &src_iter);
427 bch2_trans_iter_exit(trans, &dst_iter);
431 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
432 struct btree_iter *iter,
434 const struct bch_hash_info *hash_info,
435 const struct qstr *name, subvol_inum *inum,
439 struct bkey_s_c_dirent d;
443 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
447 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
448 hash_info, dir, name, flags);
452 k = bch2_btree_iter_peek_slot(iter);
457 d = bkey_s_c_to_dirent(k);
459 ret = bch2_dirent_read_target(trans, dir, d, inum);
464 bch2_trans_iter_exit(trans, iter);
469 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
470 const struct bch_hash_info *hash_info,
471 const struct qstr *name, subvol_inum *inum)
473 struct btree_trans *trans = bch2_trans_get(c);
474 struct btree_iter iter;
477 bch2_trans_begin(trans);
479 ret = __bch2_dirent_lookup_trans(trans, &iter, dir, hash_info,
481 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
484 bch2_trans_iter_exit(trans, &iter);
485 bch2_trans_put(trans);
489 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 snapshot)
491 struct btree_iter iter;
495 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
496 SPOS(dir, 0, snapshot),
497 POS(dir, U64_MAX), 0, k, ret)
498 if (k.k->type == KEY_TYPE_dirent) {
502 bch2_trans_iter_exit(trans, &iter);
507 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
511 return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
512 bch2_empty_dir_snapshot(trans, dir.inum, snapshot);
515 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
517 struct btree_trans *trans = bch2_trans_get(c);
518 struct btree_iter iter;
520 struct bkey_s_c_dirent dirent;
527 bch2_bkey_buf_init(&sk);
529 bch2_trans_begin(trans);
531 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
535 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
536 SPOS(inum.inum, ctx->pos, snapshot),
537 POS(inum.inum, U64_MAX), 0, k, ret) {
538 if (k.k->type != KEY_TYPE_dirent)
541 dirent = bkey_s_c_to_dirent(k);
543 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
549 /* dir_emit() can fault and block: */
550 bch2_bkey_buf_reassemble(&sk, c, k);
551 dirent = bkey_i_to_s_c_dirent(sk.k);
552 bch2_trans_unlock(trans);
554 name = bch2_dirent_get_name(dirent);
556 ctx->pos = dirent.k->p.offset;
557 if (!dir_emit(ctx, name.name,
560 vfs_d_type(dirent.v->d_type)))
562 ctx->pos = dirent.k->p.offset + 1;
565 * read_target looks up subvolumes, we can overflow paths if the
566 * directory has many subvolumes in it
568 ret = btree_trans_too_many_iters(trans);
572 bch2_trans_iter_exit(trans, &iter);
574 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
577 bch2_trans_put(trans);
578 bch2_bkey_buf_exit(&sk, c);