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 int bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k,
87 unsigned flags, struct printbuf *err)
89 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
92 len = bch2_dirent_name_bytes(d);
94 prt_printf(err, "empty name");
95 return -BCH_ERR_invalid_bkey;
98 if (bkey_val_u64s(k.k) > dirent_val_u64s(len)) {
99 prt_printf(err, "value too big (%zu > %u)",
100 bkey_val_u64s(k.k), dirent_val_u64s(len));
101 return -BCH_ERR_invalid_bkey;
104 if (len > BCH_NAME_MAX) {
105 prt_printf(err, "dirent name too big (%u > %u)",
107 return -BCH_ERR_invalid_bkey;
110 if (len == 1 && !memcmp(d.v->d_name, ".", 1)) {
111 prt_printf(err, "invalid name");
112 return -BCH_ERR_invalid_bkey;
115 if (len == 2 && !memcmp(d.v->d_name, "..", 2)) {
116 prt_printf(err, "invalid name");
117 return -BCH_ERR_invalid_bkey;
120 if (memchr(d.v->d_name, '/', len)) {
121 prt_printf(err, "invalid name");
122 return -BCH_ERR_invalid_bkey;
125 if (d.v->d_type != DT_SUBVOL &&
126 le64_to_cpu(d.v->d_inum) == d.k->p.inode) {
127 prt_printf(err, "dirent points to own directory");
128 return -BCH_ERR_invalid_bkey;
134 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
137 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
139 prt_printf(out, "%.*s -> %llu type %s",
140 bch2_dirent_name_bytes(d),
142 d.v->d_type != DT_SUBVOL
143 ? le64_to_cpu(d.v->d_inum)
144 : le32_to_cpu(d.v->d_child_subvol),
145 bch2_d_type_str(d.v->d_type));
148 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
149 subvol_inum dir, u8 type,
150 const struct qstr *name, u64 dst)
152 struct bkey_i_dirent *dirent;
153 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
155 if (name->len > BCH_NAME_MAX)
156 return ERR_PTR(-ENAMETOOLONG);
158 BUG_ON(u64s > U8_MAX);
160 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
164 bkey_dirent_init(&dirent->k_i);
165 dirent->k.u64s = u64s;
167 if (type != DT_SUBVOL) {
168 dirent->v.d_inum = cpu_to_le64(dst);
170 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
171 dirent->v.d_child_subvol = cpu_to_le32(dst);
174 dirent->v.d_type = type;
176 memcpy(dirent->v.d_name, name->name, name->len);
177 memset(dirent->v.d_name + name->len, 0,
178 bkey_val_bytes(&dirent->k) -
179 offsetof(struct bch_dirent, d_name) -
182 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
187 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
188 const struct bch_hash_info *hash_info,
189 u8 type, const struct qstr *name, u64 dst_inum,
190 u64 *dir_offset, int flags)
192 struct bkey_i_dirent *dirent;
195 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
196 ret = PTR_ERR_OR_ZERO(dirent);
200 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
201 dir, &dirent->k_i, flags);
202 *dir_offset = dirent->k.p.offset;
207 static void dirent_copy_target(struct bkey_i_dirent *dst,
208 struct bkey_s_c_dirent src)
210 dst->v.d_inum = src.v->d_inum;
211 dst->v.d_type = src.v->d_type;
214 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
215 struct bkey_s_c_dirent d, subvol_inum *target)
217 struct bch_subvolume s;
220 if (d.v->d_type == DT_SUBVOL &&
221 d.v->d_parent_subvol != dir.subvol)
224 if (likely(d.v->d_type != DT_SUBVOL)) {
225 target->subvol = dir.subvol;
226 target->inum = le64_to_cpu(d.v->d_inum);
228 target->subvol = le32_to_cpu(d.v->d_child_subvol);
230 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
232 target->inum = le64_to_cpu(s.inode);
238 int bch2_dirent_rename(struct btree_trans *trans,
239 subvol_inum src_dir, struct bch_hash_info *src_hash,
240 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
241 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
242 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
243 enum bch_rename_mode mode)
245 struct btree_iter src_iter = { NULL };
246 struct btree_iter dst_iter = { NULL };
247 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
248 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
249 struct bpos dst_pos =
250 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
251 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
254 if (src_dir.subvol != dst_dir.subvol)
257 memset(src_inum, 0, sizeof(*src_inum));
258 memset(dst_inum, 0, sizeof(*dst_inum));
261 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
262 src_hash, src_dir, src_name,
267 old_src = bch2_btree_iter_peek_slot(&src_iter);
268 ret = bkey_err(old_src);
272 ret = bch2_dirent_read_target(trans, src_dir,
273 bkey_s_c_to_dirent(old_src), src_inum);
277 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
279 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
284 if (mode == BCH_RENAME) {
286 * Note that we're _not_ checking if the target already exists -
287 * we're relying on the VFS to do that check for us for
290 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
291 dst_hash, dst_dir, dst_name);
295 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
296 dst_hash, dst_dir, dst_name,
301 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
302 ret = bkey_err(old_dst);
306 ret = bch2_dirent_read_target(trans, dst_dir,
307 bkey_s_c_to_dirent(old_dst), dst_inum);
311 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
313 if (dst_type == DT_SUBVOL)
317 if (mode != BCH_RENAME_EXCHANGE)
318 *src_offset = dst_iter.pos.offset;
320 /* Create new dst key: */
321 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
322 ret = PTR_ERR_OR_ZERO(new_dst);
326 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
327 new_dst->k.p = dst_iter.pos;
329 /* Create new src key: */
330 if (mode == BCH_RENAME_EXCHANGE) {
331 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
332 ret = PTR_ERR_OR_ZERO(new_src);
336 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
337 new_src->k.p = src_iter.pos;
339 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
340 ret = PTR_ERR_OR_ZERO(new_src);
344 bkey_init(&new_src->k);
345 new_src->k.p = src_iter.pos;
347 if (bkey_le(dst_pos, src_iter.pos) &&
348 bkey_lt(src_iter.pos, dst_iter.pos)) {
350 * We have a hash collision for the new dst key,
351 * and new_src - the key we're deleting - is between
352 * new_dst's hashed slot and the slot we're going to be
353 * inserting it into - oops. This will break the hash
354 * table if we don't deal with it:
356 if (mode == BCH_RENAME) {
358 * If we're not overwriting, we can just insert
359 * new_dst at the src position:
362 new_src->k.p = src_iter.pos;
365 /* If we're overwriting, we can't insert new_dst
366 * at a different slot because it has to
367 * overwrite old_dst - just make sure to use a
368 * whiteout when deleting src:
370 new_src->k.type = KEY_TYPE_hash_whiteout;
373 /* Check if we need a whiteout to delete src: */
374 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
375 src_hash, &src_iter);
380 new_src->k.type = KEY_TYPE_hash_whiteout;
384 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
390 * If we're deleting a subvolume, we need to really delete the dirent,
391 * not just emit a whiteout in the current snapshot:
393 if (src_type == DT_SUBVOL) {
394 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
395 ret = bch2_btree_iter_traverse(&src_iter);
399 new_src->k.p = src_iter.pos;
400 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
403 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
407 if (mode == BCH_RENAME_EXCHANGE)
408 *src_offset = new_src->k.p.offset;
409 *dst_offset = new_dst->k.p.offset;
411 bch2_trans_iter_exit(trans, &src_iter);
412 bch2_trans_iter_exit(trans, &dst_iter);
416 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
417 struct btree_iter *iter,
419 const struct bch_hash_info *hash_info,
420 const struct qstr *name, subvol_inum *inum,
424 struct bkey_s_c_dirent d;
428 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
432 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
433 hash_info, dir, name, flags);
437 k = bch2_btree_iter_peek_slot(iter);
442 d = bkey_s_c_to_dirent(k);
444 ret = bch2_dirent_read_target(trans, dir, d, inum);
449 bch2_trans_iter_exit(trans, iter);
454 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
455 const struct bch_hash_info *hash_info,
456 const struct qstr *name, subvol_inum *inum)
458 struct btree_trans trans;
459 struct btree_iter iter;
462 bch2_trans_init(&trans, c, 0, 0);
464 bch2_trans_begin(&trans);
466 ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
468 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
471 bch2_trans_iter_exit(&trans, &iter);
472 bch2_trans_exit(&trans);
476 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
478 struct btree_iter iter;
483 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
487 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
488 SPOS(dir.inum, 0, snapshot),
489 POS(dir.inum, U64_MAX), 0, k, ret)
490 if (k.k->type == KEY_TYPE_dirent) {
494 bch2_trans_iter_exit(trans, &iter);
499 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
501 struct btree_trans trans;
502 struct btree_iter iter;
504 struct bkey_s_c_dirent dirent;
509 bch2_trans_init(&trans, c, 0, 0);
511 bch2_trans_begin(&trans);
513 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
517 for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_dirents,
518 SPOS(inum.inum, ctx->pos, snapshot),
519 POS(inum.inum, U64_MAX), 0, k, ret) {
520 if (k.k->type != KEY_TYPE_dirent)
523 dirent = bkey_s_c_to_dirent(k);
525 ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
532 * XXX: dir_emit() can fault and block, while we're holding
535 ctx->pos = dirent.k->p.offset;
536 if (!dir_emit(ctx, dirent.v->d_name,
537 bch2_dirent_name_bytes(dirent),
539 vfs_d_type(dirent.v->d_type)))
541 ctx->pos = dirent.k->p.offset + 1;
544 * read_target looks up subvolumes, we can overflow paths if the
545 * directory has many subvolumes in it
547 ret = btree_trans_too_many_iters(&trans);
551 bch2_trans_iter_exit(&trans, &iter);
553 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
556 bch2_trans_exit(&trans);