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(const 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 prt_printf(err, "empty name");
109 return -BCH_ERR_invalid_bkey;
112 if (bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len)) {
113 prt_printf(err, "value too big (%zu > %u)",
114 bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
115 return -BCH_ERR_invalid_bkey;
119 * Check new keys don't exceed the max length
120 * (older keys may be larger.)
122 if ((flags & BKEY_INVALID_COMMIT) && d_name.len > BCH_NAME_MAX) {
123 prt_printf(err, "dirent name too big (%u > %u)",
124 d_name.len, BCH_NAME_MAX);
125 return -BCH_ERR_invalid_bkey;
128 if (d_name.len != strnlen(d_name.name, d_name.len)) {
129 prt_printf(err, "dirent has stray data after name's NUL");
130 return -BCH_ERR_invalid_bkey;
133 if (d_name.len == 1 && !memcmp(d_name.name, ".", 1)) {
134 prt_printf(err, "invalid name");
135 return -BCH_ERR_invalid_bkey;
138 if (d_name.len == 2 && !memcmp(d_name.name, "..", 2)) {
139 prt_printf(err, "invalid name");
140 return -BCH_ERR_invalid_bkey;
143 if (memchr(d_name.name, '/', d_name.len)) {
144 prt_printf(err, "invalid name");
145 return -BCH_ERR_invalid_bkey;
148 if (d.v->d_type != DT_SUBVOL &&
149 le64_to_cpu(d.v->d_inum) == d.k->p.inode) {
150 prt_printf(err, "dirent points to own directory");
151 return -BCH_ERR_invalid_bkey;
157 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
160 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
161 struct qstr d_name = bch2_dirent_get_name(d);
163 prt_printf(out, "%.*s -> %llu type %s",
166 d.v->d_type != DT_SUBVOL
167 ? le64_to_cpu(d.v->d_inum)
168 : le32_to_cpu(d.v->d_child_subvol),
169 bch2_d_type_str(d.v->d_type));
172 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
173 subvol_inum dir, u8 type,
174 const struct qstr *name, u64 dst)
176 struct bkey_i_dirent *dirent;
177 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
179 if (name->len > BCH_NAME_MAX)
180 return ERR_PTR(-ENAMETOOLONG);
182 BUG_ON(u64s > U8_MAX);
184 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
188 bkey_dirent_init(&dirent->k_i);
189 dirent->k.u64s = u64s;
191 if (type != DT_SUBVOL) {
192 dirent->v.d_inum = cpu_to_le64(dst);
194 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
195 dirent->v.d_child_subvol = cpu_to_le32(dst);
198 dirent->v.d_type = type;
200 memcpy(dirent->v.d_name, name->name, name->len);
201 memset(dirent->v.d_name + name->len, 0,
202 bkey_val_bytes(&dirent->k) -
203 offsetof(struct bch_dirent, d_name) -
206 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
211 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
212 const struct bch_hash_info *hash_info,
213 u8 type, const struct qstr *name, u64 dst_inum,
214 u64 *dir_offset, int flags)
216 struct bkey_i_dirent *dirent;
219 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
220 ret = PTR_ERR_OR_ZERO(dirent);
224 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
225 dir, &dirent->k_i, flags);
226 *dir_offset = dirent->k.p.offset;
231 static void dirent_copy_target(struct bkey_i_dirent *dst,
232 struct bkey_s_c_dirent src)
234 dst->v.d_inum = src.v->d_inum;
235 dst->v.d_type = src.v->d_type;
238 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
239 struct bkey_s_c_dirent d, subvol_inum *target)
241 struct bch_subvolume s;
244 if (d.v->d_type == DT_SUBVOL &&
245 le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
248 if (likely(d.v->d_type != DT_SUBVOL)) {
249 target->subvol = dir.subvol;
250 target->inum = le64_to_cpu(d.v->d_inum);
252 target->subvol = le32_to_cpu(d.v->d_child_subvol);
254 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
256 target->inum = le64_to_cpu(s.inode);
262 int bch2_dirent_rename(struct btree_trans *trans,
263 subvol_inum src_dir, struct bch_hash_info *src_hash,
264 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
265 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
266 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
267 enum bch_rename_mode mode)
269 struct btree_iter src_iter = { NULL };
270 struct btree_iter dst_iter = { NULL };
271 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
272 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
273 struct bpos dst_pos =
274 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
275 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
278 if (src_dir.subvol != dst_dir.subvol)
281 memset(src_inum, 0, sizeof(*src_inum));
282 memset(dst_inum, 0, sizeof(*dst_inum));
285 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
286 src_hash, src_dir, src_name,
291 old_src = bch2_btree_iter_peek_slot(&src_iter);
292 ret = bkey_err(old_src);
296 ret = bch2_dirent_read_target(trans, src_dir,
297 bkey_s_c_to_dirent(old_src), src_inum);
301 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
303 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
308 if (mode == BCH_RENAME) {
310 * Note that we're _not_ checking if the target already exists -
311 * we're relying on the VFS to do that check for us for
314 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
315 dst_hash, dst_dir, dst_name);
319 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
320 dst_hash, dst_dir, dst_name,
325 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
326 ret = bkey_err(old_dst);
330 ret = bch2_dirent_read_target(trans, dst_dir,
331 bkey_s_c_to_dirent(old_dst), dst_inum);
335 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
337 if (dst_type == DT_SUBVOL)
341 if (mode != BCH_RENAME_EXCHANGE)
342 *src_offset = dst_iter.pos.offset;
344 /* Create new dst key: */
345 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
346 ret = PTR_ERR_OR_ZERO(new_dst);
350 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
351 new_dst->k.p = dst_iter.pos;
353 /* Create new src key: */
354 if (mode == BCH_RENAME_EXCHANGE) {
355 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
356 ret = PTR_ERR_OR_ZERO(new_src);
360 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
361 new_src->k.p = src_iter.pos;
363 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
364 ret = PTR_ERR_OR_ZERO(new_src);
368 bkey_init(&new_src->k);
369 new_src->k.p = src_iter.pos;
371 if (bkey_le(dst_pos, src_iter.pos) &&
372 bkey_lt(src_iter.pos, dst_iter.pos)) {
374 * We have a hash collision for the new dst key,
375 * and new_src - the key we're deleting - is between
376 * new_dst's hashed slot and the slot we're going to be
377 * inserting it into - oops. This will break the hash
378 * table if we don't deal with it:
380 if (mode == BCH_RENAME) {
382 * If we're not overwriting, we can just insert
383 * new_dst at the src position:
386 new_src->k.p = src_iter.pos;
389 /* If we're overwriting, we can't insert new_dst
390 * at a different slot because it has to
391 * overwrite old_dst - just make sure to use a
392 * whiteout when deleting src:
394 new_src->k.type = KEY_TYPE_hash_whiteout;
397 /* Check if we need a whiteout to delete src: */
398 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
399 src_hash, &src_iter);
404 new_src->k.type = KEY_TYPE_hash_whiteout;
408 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
414 * If we're deleting a subvolume, we need to really delete the dirent,
415 * not just emit a whiteout in the current snapshot:
417 if (src_type == DT_SUBVOL) {
418 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
419 ret = bch2_btree_iter_traverse(&src_iter);
423 new_src->k.p = src_iter.pos;
424 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
427 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
431 if (mode == BCH_RENAME_EXCHANGE)
432 *src_offset = new_src->k.p.offset;
433 *dst_offset = new_dst->k.p.offset;
435 bch2_trans_iter_exit(trans, &src_iter);
436 bch2_trans_iter_exit(trans, &dst_iter);
440 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
441 struct btree_iter *iter,
443 const struct bch_hash_info *hash_info,
444 const struct qstr *name, subvol_inum *inum,
448 struct bkey_s_c_dirent d;
452 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
456 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
457 hash_info, dir, name, flags);
461 k = bch2_btree_iter_peek_slot(iter);
466 d = bkey_s_c_to_dirent(k);
468 ret = bch2_dirent_read_target(trans, dir, d, inum);
473 bch2_trans_iter_exit(trans, iter);
478 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
479 const struct bch_hash_info *hash_info,
480 const struct qstr *name, subvol_inum *inum)
482 struct btree_trans trans;
483 struct btree_iter iter;
486 bch2_trans_init(&trans, c, 0, 0);
488 bch2_trans_begin(&trans);
490 ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
492 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
495 bch2_trans_iter_exit(&trans, &iter);
496 bch2_trans_exit(&trans);
500 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
502 struct btree_iter iter;
507 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
511 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
512 SPOS(dir.inum, 0, snapshot),
513 POS(dir.inum, U64_MAX), 0, k, ret)
514 if (k.k->type == KEY_TYPE_dirent) {
518 bch2_trans_iter_exit(trans, &iter);
523 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
525 struct btree_trans trans;
526 struct btree_iter iter;
528 struct bkey_s_c_dirent dirent;
535 bch2_bkey_buf_init(&sk);
536 bch2_trans_init(&trans, c, 0, 0);
538 bch2_trans_begin(&trans);
540 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
544 for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_dirents,
545 SPOS(inum.inum, ctx->pos, snapshot),
546 POS(inum.inum, U64_MAX), 0, k, ret) {
547 if (k.k->type != KEY_TYPE_dirent)
550 dirent = bkey_s_c_to_dirent(k);
552 ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
558 /* dir_emit() can fault and block: */
559 bch2_bkey_buf_reassemble(&sk, c, k);
560 dirent = bkey_i_to_s_c_dirent(sk.k);
561 bch2_trans_unlock(&trans);
563 name = bch2_dirent_get_name(dirent);
565 ctx->pos = dirent.k->p.offset;
566 if (!dir_emit(ctx, name.name,
569 vfs_d_type(dirent.v->d_type)))
571 ctx->pos = dirent.k->p.offset + 1;
574 * read_target looks up subvolumes, we can overflow paths if the
575 * directory has many subvolumes in it
577 ret = btree_trans_too_many_iters(&trans);
581 bch2_trans_iter_exit(&trans, &iter);
583 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
586 bch2_trans_exit(&trans);
587 bch2_bkey_buf_exit(&sk, c);