1 // SPDX-License-Identifier: GPL-2.0
4 #include "bkey_methods.h"
5 #include "btree_update.h"
11 #include "subvolume.h"
13 #include <linux/dcache.h>
15 unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
17 unsigned len = bkey_val_bytes(d.k) -
18 offsetof(struct bch_dirent, d_name);
20 return strnlen(d.v->d_name, len);
23 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
24 const struct qstr *name)
26 struct bch_str_hash_ctx ctx;
28 bch2_str_hash_init(&ctx, info);
29 bch2_str_hash_update(&ctx, info, name->name, name->len);
31 /* [0,2) reserved for dots */
32 return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
35 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
37 return bch2_dirent_hash(info, key);
40 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
42 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
43 struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
45 return bch2_dirent_hash(info, &name);
48 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
50 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
51 int len = bch2_dirent_name_bytes(l);
52 const struct qstr *r = _r;
54 return len - r->len ?: memcmp(l.v->d_name, r->name, len);
57 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
59 struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
60 struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
61 int l_len = bch2_dirent_name_bytes(l);
62 int r_len = bch2_dirent_name_bytes(r);
64 return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
67 static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
69 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
71 if (d.v->d_type == DT_SUBVOL)
72 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
76 const struct bch_hash_desc bch2_dirent_hash_desc = {
77 .btree_id = BTREE_ID_dirents,
78 .key_type = KEY_TYPE_dirent,
79 .hash_key = dirent_hash_key,
80 .hash_bkey = dirent_hash_bkey,
81 .cmp_key = dirent_cmp_key,
82 .cmp_bkey = dirent_cmp_bkey,
83 .is_visible = dirent_is_visible,
86 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
88 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
91 if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
92 return "value too small";
94 len = bch2_dirent_name_bytes(d);
98 if (bkey_val_u64s(k.k) > dirent_val_u64s(len))
99 return "value too big";
101 if (len > BCH_NAME_MAX)
102 return "dirent name too big";
104 if (len == 1 && !memcmp(d.v->d_name, ".", 1))
105 return "invalid name";
107 if (len == 2 && !memcmp(d.v->d_name, "..", 2))
108 return "invalid name";
110 if (memchr(d.v->d_name, '/', len))
111 return "invalid name";
113 if (d.v->d_type != DT_SUBVOL &&
114 le64_to_cpu(d.v->d_inum) == d.k->p.inode)
115 return "dirent points to own directory";
120 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
123 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
125 pr_buf(out, "%.*s -> %llu type %s",
126 bch2_dirent_name_bytes(d),
128 d.v->d_type != DT_SUBVOL
129 ? le64_to_cpu(d.v->d_inum)
130 : le32_to_cpu(d.v->d_child_subvol),
131 bch2_d_type_str(d.v->d_type));
134 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
135 subvol_inum dir, u8 type,
136 const struct qstr *name, u64 dst)
138 struct bkey_i_dirent *dirent;
139 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
141 if (name->len > BCH_NAME_MAX)
142 return ERR_PTR(-ENAMETOOLONG);
144 BUG_ON(u64s > U8_MAX);
146 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
150 bkey_dirent_init(&dirent->k_i);
151 dirent->k.u64s = u64s;
153 if (type != DT_SUBVOL) {
154 dirent->v.d_inum = cpu_to_le64(dst);
156 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
157 dirent->v.d_child_subvol = cpu_to_le32(dst);
160 dirent->v.d_type = type;
162 memcpy(dirent->v.d_name, name->name, name->len);
163 memset(dirent->v.d_name + name->len, 0,
164 bkey_val_bytes(&dirent->k) -
165 offsetof(struct bch_dirent, d_name) -
168 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
173 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
174 const struct bch_hash_info *hash_info,
175 u8 type, const struct qstr *name, u64 dst_inum,
176 u64 *dir_offset, int flags)
178 struct bkey_i_dirent *dirent;
181 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
182 ret = PTR_ERR_OR_ZERO(dirent);
186 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
187 dir, &dirent->k_i, flags);
188 *dir_offset = dirent->k.p.offset;
193 static void dirent_copy_target(struct bkey_i_dirent *dst,
194 struct bkey_s_c_dirent src)
196 dst->v.d_inum = src.v->d_inum;
197 dst->v.d_type = src.v->d_type;
200 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
201 struct bkey_s_c_dirent d, subvol_inum *target)
203 struct bch_subvolume s;
206 if (d.v->d_type == DT_SUBVOL &&
207 d.v->d_parent_subvol != dir.subvol)
210 if (likely(d.v->d_type != DT_SUBVOL)) {
211 target->subvol = dir.subvol;
212 target->inum = le64_to_cpu(d.v->d_inum);
214 target->subvol = le32_to_cpu(d.v->d_child_subvol);
216 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
218 target->inum = le64_to_cpu(s.inode);
224 int bch2_dirent_rename(struct btree_trans *trans,
225 subvol_inum src_dir, struct bch_hash_info *src_hash,
226 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
227 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
228 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
229 enum bch_rename_mode mode)
231 struct btree_iter src_iter = { NULL };
232 struct btree_iter dst_iter = { NULL };
233 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
234 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
235 struct bpos dst_pos =
236 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
237 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
240 if (src_dir.subvol != dst_dir.subvol)
243 memset(src_inum, 0, sizeof(*src_inum));
244 memset(dst_inum, 0, sizeof(*dst_inum));
247 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
248 src_hash, src_dir, src_name,
253 old_src = bch2_btree_iter_peek_slot(&src_iter);
254 ret = bkey_err(old_src);
258 ret = bch2_dirent_read_target(trans, src_dir,
259 bkey_s_c_to_dirent(old_src), src_inum);
263 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
265 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
270 if (mode == BCH_RENAME) {
272 * Note that we're _not_ checking if the target already exists -
273 * we're relying on the VFS to do that check for us for
276 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
277 dst_hash, dst_dir, dst_name);
281 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
282 dst_hash, dst_dir, dst_name,
287 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
288 ret = bkey_err(old_dst);
292 ret = bch2_dirent_read_target(trans, dst_dir,
293 bkey_s_c_to_dirent(old_dst), dst_inum);
297 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
299 if (dst_type == DT_SUBVOL)
303 if (mode != BCH_RENAME_EXCHANGE)
304 *src_offset = dst_iter.pos.offset;
306 /* Create new dst key: */
307 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
308 ret = PTR_ERR_OR_ZERO(new_dst);
312 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
313 new_dst->k.p = dst_iter.pos;
315 /* Create new src key: */
316 if (mode == BCH_RENAME_EXCHANGE) {
317 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
318 ret = PTR_ERR_OR_ZERO(new_src);
322 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
323 new_src->k.p = src_iter.pos;
325 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
326 ret = PTR_ERR_OR_ZERO(new_src);
330 bkey_init(&new_src->k);
331 new_src->k.p = src_iter.pos;
333 if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
334 bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
336 * We have a hash collision for the new dst key,
337 * and new_src - the key we're deleting - is between
338 * new_dst's hashed slot and the slot we're going to be
339 * inserting it into - oops. This will break the hash
340 * table if we don't deal with it:
342 if (mode == BCH_RENAME) {
344 * If we're not overwriting, we can just insert
345 * new_dst at the src position:
348 new_src->k.p = src_iter.pos;
351 /* If we're overwriting, we can't insert new_dst
352 * at a different slot because it has to
353 * overwrite old_dst - just make sure to use a
354 * whiteout when deleting src:
356 new_src->k.type = KEY_TYPE_hash_whiteout;
359 /* Check if we need a whiteout to delete src: */
360 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
361 src_hash, &src_iter);
366 new_src->k.type = KEY_TYPE_hash_whiteout;
370 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
376 * If we're deleting a subvolume, we need to really delete the dirent,
377 * not just emit a whiteout in the current snapshot:
379 if (src_type == DT_SUBVOL) {
380 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
381 ret = bch2_btree_iter_traverse(&src_iter);
385 new_src->k.p = src_iter.pos;
386 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
389 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
393 if (mode == BCH_RENAME_EXCHANGE)
394 *src_offset = new_src->k.p.offset;
395 *dst_offset = new_dst->k.p.offset;
397 bch2_trans_iter_exit(trans, &src_iter);
398 bch2_trans_iter_exit(trans, &dst_iter);
402 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
403 struct btree_iter *iter,
405 const struct bch_hash_info *hash_info,
406 const struct qstr *name, subvol_inum *inum,
410 struct bkey_s_c_dirent d;
414 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
418 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
419 hash_info, dir, name, flags);
423 k = bch2_btree_iter_peek_slot(iter);
428 d = bkey_s_c_to_dirent(k);
430 ret = bch2_dirent_read_target(trans, dir, d, inum);
435 bch2_trans_iter_exit(trans, iter);
440 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
441 const struct bch_hash_info *hash_info,
442 const struct qstr *name, subvol_inum *inum)
444 struct btree_trans trans;
445 struct btree_iter iter;
448 bch2_trans_init(&trans, c, 0, 0);
450 bch2_trans_begin(&trans);
452 ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
457 bch2_trans_iter_exit(&trans, &iter);
458 bch2_trans_exit(&trans);
462 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
464 struct btree_iter iter;
469 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
473 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
474 SPOS(dir.inum, 0, snapshot),
475 POS(dir.inum, U64_MAX), 0, k, ret)
476 if (k.k->type == KEY_TYPE_dirent) {
480 bch2_trans_iter_exit(trans, &iter);
485 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
487 struct btree_trans trans;
488 struct btree_iter iter;
490 struct bkey_s_c_dirent dirent;
495 bch2_trans_init(&trans, c, 0, 0);
497 bch2_trans_begin(&trans);
499 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
503 for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_dirents,
504 SPOS(inum.inum, ctx->pos, snapshot),
505 POS(inum.inum, U64_MAX), 0, k, ret) {
506 if (k.k->type != KEY_TYPE_dirent)
509 dirent = bkey_s_c_to_dirent(k);
511 ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
518 * XXX: dir_emit() can fault and block, while we're holding
521 ctx->pos = dirent.k->p.offset;
522 if (!dir_emit(ctx, dirent.v->d_name,
523 bch2_dirent_name_bytes(dirent),
525 vfs_d_type(dirent.v->d_type)))
527 ctx->pos = dirent.k->p.offset + 1;
530 * read_target looks up subvolumes, we can overflow paths if the
531 * directory has many subvolumes in it
533 ret = btree_trans_too_many_iters(&trans);
537 bch2_trans_iter_exit(&trans, &iter);
542 bch2_trans_exit(&trans);