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 int rw, struct printbuf *err)
89 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
92 if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent)) {
93 prt_printf(err, "incorrect value size (%zu < %zu)",
94 bkey_val_bytes(k.k), sizeof(*d.v));
98 len = bch2_dirent_name_bytes(d);
100 prt_printf(err, "empty name");
104 if (bkey_val_u64s(k.k) > dirent_val_u64s(len)) {
105 prt_printf(err, "value too big (%zu > %u)",
106 bkey_val_u64s(k.k), dirent_val_u64s(len));
110 if (len > BCH_NAME_MAX) {
111 prt_printf(err, "dirent name too big (%u > %u)",
116 if (len == 1 && !memcmp(d.v->d_name, ".", 1)) {
117 prt_printf(err, "invalid name");
121 if (len == 2 && !memcmp(d.v->d_name, "..", 2)) {
122 prt_printf(err, "invalid name");
126 if (memchr(d.v->d_name, '/', len)) {
127 prt_printf(err, "invalid name");
131 if (d.v->d_type != DT_SUBVOL &&
132 le64_to_cpu(d.v->d_inum) == d.k->p.inode) {
133 prt_printf(err, "dirent points to own directory");
140 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
143 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
145 prt_printf(out, "%.*s -> %llu type %s",
146 bch2_dirent_name_bytes(d),
148 d.v->d_type != DT_SUBVOL
149 ? le64_to_cpu(d.v->d_inum)
150 : le32_to_cpu(d.v->d_child_subvol),
151 bch2_d_type_str(d.v->d_type));
154 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
155 subvol_inum dir, u8 type,
156 const struct qstr *name, u64 dst)
158 struct bkey_i_dirent *dirent;
159 unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
161 if (name->len > BCH_NAME_MAX)
162 return ERR_PTR(-ENAMETOOLONG);
164 BUG_ON(u64s > U8_MAX);
166 dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
170 bkey_dirent_init(&dirent->k_i);
171 dirent->k.u64s = u64s;
173 if (type != DT_SUBVOL) {
174 dirent->v.d_inum = cpu_to_le64(dst);
176 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
177 dirent->v.d_child_subvol = cpu_to_le32(dst);
180 dirent->v.d_type = type;
182 memcpy(dirent->v.d_name, name->name, name->len);
183 memset(dirent->v.d_name + name->len, 0,
184 bkey_val_bytes(&dirent->k) -
185 offsetof(struct bch_dirent, d_name) -
188 EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
193 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
194 const struct bch_hash_info *hash_info,
195 u8 type, const struct qstr *name, u64 dst_inum,
196 u64 *dir_offset, int flags)
198 struct bkey_i_dirent *dirent;
201 dirent = dirent_create_key(trans, dir, type, name, dst_inum);
202 ret = PTR_ERR_OR_ZERO(dirent);
206 ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
207 dir, &dirent->k_i, flags);
208 *dir_offset = dirent->k.p.offset;
213 static void dirent_copy_target(struct bkey_i_dirent *dst,
214 struct bkey_s_c_dirent src)
216 dst->v.d_inum = src.v->d_inum;
217 dst->v.d_type = src.v->d_type;
220 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
221 struct bkey_s_c_dirent d, subvol_inum *target)
223 struct bch_subvolume s;
226 if (d.v->d_type == DT_SUBVOL &&
227 d.v->d_parent_subvol != dir.subvol)
230 if (likely(d.v->d_type != DT_SUBVOL)) {
231 target->subvol = dir.subvol;
232 target->inum = le64_to_cpu(d.v->d_inum);
234 target->subvol = le32_to_cpu(d.v->d_child_subvol);
236 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
238 target->inum = le64_to_cpu(s.inode);
244 int bch2_dirent_rename(struct btree_trans *trans,
245 subvol_inum src_dir, struct bch_hash_info *src_hash,
246 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
247 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
248 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
249 enum bch_rename_mode mode)
251 struct btree_iter src_iter = { NULL };
252 struct btree_iter dst_iter = { NULL };
253 struct bkey_s_c old_src, old_dst = bkey_s_c_null;
254 struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
255 struct bpos dst_pos =
256 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
257 unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
260 if (src_dir.subvol != dst_dir.subvol)
263 memset(src_inum, 0, sizeof(*src_inum));
264 memset(dst_inum, 0, sizeof(*dst_inum));
267 ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
268 src_hash, src_dir, src_name,
273 old_src = bch2_btree_iter_peek_slot(&src_iter);
274 ret = bkey_err(old_src);
278 ret = bch2_dirent_read_target(trans, src_dir,
279 bkey_s_c_to_dirent(old_src), src_inum);
283 src_type = bkey_s_c_to_dirent(old_src).v->d_type;
285 if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
290 if (mode == BCH_RENAME) {
292 * Note that we're _not_ checking if the target already exists -
293 * we're relying on the VFS to do that check for us for
296 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
297 dst_hash, dst_dir, dst_name);
301 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
302 dst_hash, dst_dir, dst_name,
307 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
308 ret = bkey_err(old_dst);
312 ret = bch2_dirent_read_target(trans, dst_dir,
313 bkey_s_c_to_dirent(old_dst), dst_inum);
317 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
319 if (dst_type == DT_SUBVOL)
323 if (mode != BCH_RENAME_EXCHANGE)
324 *src_offset = dst_iter.pos.offset;
326 /* Create new dst key: */
327 new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
328 ret = PTR_ERR_OR_ZERO(new_dst);
332 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
333 new_dst->k.p = dst_iter.pos;
335 /* Create new src key: */
336 if (mode == BCH_RENAME_EXCHANGE) {
337 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
338 ret = PTR_ERR_OR_ZERO(new_src);
342 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
343 new_src->k.p = src_iter.pos;
345 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
346 ret = PTR_ERR_OR_ZERO(new_src);
350 bkey_init(&new_src->k);
351 new_src->k.p = src_iter.pos;
353 if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
354 bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
356 * We have a hash collision for the new dst key,
357 * and new_src - the key we're deleting - is between
358 * new_dst's hashed slot and the slot we're going to be
359 * inserting it into - oops. This will break the hash
360 * table if we don't deal with it:
362 if (mode == BCH_RENAME) {
364 * If we're not overwriting, we can just insert
365 * new_dst at the src position:
368 new_src->k.p = src_iter.pos;
371 /* If we're overwriting, we can't insert new_dst
372 * at a different slot because it has to
373 * overwrite old_dst - just make sure to use a
374 * whiteout when deleting src:
376 new_src->k.type = KEY_TYPE_hash_whiteout;
379 /* Check if we need a whiteout to delete src: */
380 ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
381 src_hash, &src_iter);
386 new_src->k.type = KEY_TYPE_hash_whiteout;
390 ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
396 * If we're deleting a subvolume, we need to really delete the dirent,
397 * not just emit a whiteout in the current snapshot:
399 if (src_type == DT_SUBVOL) {
400 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
401 ret = bch2_btree_iter_traverse(&src_iter);
405 new_src->k.p = src_iter.pos;
406 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
409 ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
413 if (mode == BCH_RENAME_EXCHANGE)
414 *src_offset = new_src->k.p.offset;
415 *dst_offset = new_dst->k.p.offset;
417 bch2_trans_iter_exit(trans, &src_iter);
418 bch2_trans_iter_exit(trans, &dst_iter);
422 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
423 struct btree_iter *iter,
425 const struct bch_hash_info *hash_info,
426 const struct qstr *name, subvol_inum *inum,
430 struct bkey_s_c_dirent d;
434 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
438 ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
439 hash_info, dir, name, flags);
443 k = bch2_btree_iter_peek_slot(iter);
448 d = bkey_s_c_to_dirent(k);
450 ret = bch2_dirent_read_target(trans, dir, d, inum);
455 bch2_trans_iter_exit(trans, iter);
460 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
461 const struct bch_hash_info *hash_info,
462 const struct qstr *name, subvol_inum *inum)
464 struct btree_trans trans;
465 struct btree_iter iter;
468 bch2_trans_init(&trans, c, 0, 0);
470 bch2_trans_begin(&trans);
472 ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
474 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
477 bch2_trans_iter_exit(&trans, &iter);
478 bch2_trans_exit(&trans);
482 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
484 struct btree_iter iter;
489 ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
493 for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
494 SPOS(dir.inum, 0, snapshot),
495 POS(dir.inum, U64_MAX), 0, k, ret)
496 if (k.k->type == KEY_TYPE_dirent) {
500 bch2_trans_iter_exit(trans, &iter);
505 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
507 struct btree_trans trans;
508 struct btree_iter iter;
510 struct bkey_s_c_dirent dirent;
515 bch2_trans_init(&trans, c, 0, 0);
517 bch2_trans_begin(&trans);
519 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
523 for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_dirents,
524 SPOS(inum.inum, ctx->pos, snapshot),
525 POS(inum.inum, U64_MAX), 0, k, ret) {
526 if (k.k->type != KEY_TYPE_dirent)
529 dirent = bkey_s_c_to_dirent(k);
531 ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
538 * XXX: dir_emit() can fault and block, while we're holding
541 ctx->pos = dirent.k->p.offset;
542 if (!dir_emit(ctx, dirent.v->d_name,
543 bch2_dirent_name_bytes(dirent),
545 vfs_d_type(dirent.v->d_type)))
547 ctx->pos = dirent.k->p.offset + 1;
550 * read_target looks up subvolumes, we can overflow paths if the
551 * directory has many subvolumes in it
553 ret = btree_trans_too_many_iters(&trans);
557 bch2_trans_iter_exit(&trans, &iter);
559 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
562 bch2_trans_exit(&trans);