+// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
#include "bkey_methods.h"
unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
{
- unsigned len = bkey_val_bytes(d.k) - sizeof(struct bch_dirent);
+ unsigned len = bkey_val_bytes(d.k) -
+ offsetof(struct bch_dirent, d_name);
- while (len && !d.v->d_name[len - 1])
- --len;
-
- return len;
-}
-
-static unsigned dirent_val_u64s(unsigned len)
-{
- return DIV_ROUND_UP(sizeof(struct bch_dirent) + len, sizeof(u64));
+ return strnlen(d.v->d_name, len);
}
static u64 bch2_dirent_hash(const struct bch_hash_info *info,
}
const struct bch_hash_desc bch2_dirent_hash_desc = {
- .btree_id = BTREE_ID_DIRENTS,
- .key_type = BCH_DIRENT,
- .whiteout_type = BCH_DIRENT_WHITEOUT,
+ .btree_id = BTREE_ID_dirents,
+ .key_type = KEY_TYPE_dirent,
.hash_key = dirent_hash_key,
.hash_bkey = dirent_hash_bkey,
.cmp_key = dirent_cmp_key,
.cmp_bkey = dirent_cmp_bkey,
};
-static const char *bch2_dirent_invalid(const struct bch_fs *c,
- struct bkey_s_c k)
+const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
- struct bkey_s_c_dirent d;
+ struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
unsigned len;
- switch (k.k->type) {
- case BCH_DIRENT:
- if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
- return "value too small";
+ if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
+ return "value too small";
- d = bkey_s_c_to_dirent(k);
- len = bch2_dirent_name_bytes(d);
+ len = bch2_dirent_name_bytes(d);
+ if (!len)
+ return "empty name";
- if (!len)
- return "empty name";
+ if (bkey_val_u64s(k.k) > dirent_val_u64s(len))
+ return "value too big";
- if (bkey_val_u64s(k.k) > dirent_val_u64s(len))
- return "value too big";
+ if (len > BCH_NAME_MAX)
+ return "dirent name too big";
- if (len > NAME_MAX)
- return "dirent name too big";
+ if (len == 1 && !memcmp(d.v->d_name, ".", 1))
+ return "invalid name";
- if (memchr(d.v->d_name, '/', len))
- return "dirent name has invalid characters";
+ if (len == 2 && !memcmp(d.v->d_name, "..", 2))
+ return "invalid name";
- return NULL;
- case BCH_DIRENT_WHITEOUT:
- return bkey_val_bytes(k.k) != 0
- ? "value size should be zero"
- : NULL;
+ if (memchr(d.v->d_name, '/', len))
+ return "invalid name";
- default:
- return "invalid type";
- }
+ if (le64_to_cpu(d.v->d_inum) == d.k->p.inode)
+ return "dirent points to own directory";
+
+ return NULL;
}
-static void bch2_dirent_to_text(struct bch_fs *c, char *buf,
- size_t size, struct bkey_s_c k)
+void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
{
- struct bkey_s_c_dirent d;
- size_t n = 0;
-
- switch (k.k->type) {
- case BCH_DIRENT:
- d = bkey_s_c_to_dirent(k);
-
- n += bch_scnmemcpy(buf + n, size - n, d.v->d_name,
- bch2_dirent_name_bytes(d));
- n += scnprintf(buf + n, size - n, " -> %llu", d.v->d_inum);
- break;
- case BCH_DIRENT_WHITEOUT:
- scnprintf(buf, size, "whiteout");
- break;
- }
-}
+ struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
-const struct bkey_ops bch2_bkey_dirent_ops = {
- .key_invalid = bch2_dirent_invalid,
- .val_to_text = bch2_dirent_to_text,
-};
+ bch_scnmemcpy(out, d.v->d_name,
+ bch2_dirent_name_bytes(d));
+ pr_buf(out, " -> %llu type %s", d.v->d_inum,
+ d.v->d_type < DT_MAX
+ ? bch2_d_types[d.v->d_type]
+ : "(bad d_type)");
+}
-static struct bkey_i_dirent *dirent_create_key(u8 type,
- const struct qstr *name, u64 dst)
+static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
+ u8 type, const struct qstr *name, u64 dst)
{
struct bkey_i_dirent *dirent;
unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
- dirent = kmalloc(u64s * sizeof(u64), GFP_NOFS);
- if (!dirent)
- return NULL;
+ if (name->len > BCH_NAME_MAX)
+ return ERR_PTR(-ENAMETOOLONG);
+
+ BUG_ON(u64s > U8_MAX);
+
+ dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
+ if (IS_ERR(dirent))
+ return dirent;
bkey_dirent_init(&dirent->k_i);
dirent->k.u64s = u64s;
memcpy(dirent->v.d_name, name->name, name->len);
memset(dirent->v.d_name + name->len, 0,
bkey_val_bytes(&dirent->k) -
- (sizeof(struct bch_dirent) + name->len));
+ offsetof(struct bch_dirent, d_name) -
+ name->len);
EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
return dirent;
}
-int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
- const struct bch_hash_info *hash_info,
+int bch2_dirent_create(struct btree_trans *trans,
+ u64 dir_inum, const struct bch_hash_info *hash_info,
u8 type, const struct qstr *name, u64 dst_inum,
- u64 *journal_seq, int flags)
+ u64 *dir_offset, int flags)
{
struct bkey_i_dirent *dirent;
int ret;
- dirent = dirent_create_key(type, name, dst_inum);
- if (!dirent)
- return -ENOMEM;
+ dirent = dirent_create_key(trans, type, name, dst_inum);
+ ret = PTR_ERR_OR_ZERO(dirent);
+ if (ret)
+ return ret;
- ret = bch2_hash_set(bch2_dirent_hash_desc, hash_info, c, dir_inum,
- journal_seq, &dirent->k_i, flags);
- kfree(dirent);
+ ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
+ dir_inum, &dirent->k_i, flags);
+ *dir_offset = dirent->k.p.offset;
return ret;
}
dst->v.d_type = src.v->d_type;
}
-static struct bpos bch2_dirent_pos(struct bch_inode_info *ei,
- const struct qstr *name)
+int bch2_dirent_rename(struct btree_trans *trans,
+ u64 src_dir, struct bch_hash_info *src_hash,
+ u64 dst_dir, struct bch_hash_info *dst_hash,
+ const struct qstr *src_name, u64 *src_inum, u64 *src_offset,
+ const struct qstr *dst_name, u64 *dst_inum, u64 *dst_offset,
+ enum bch_rename_mode mode)
{
- return POS(ei->vfs_inode.i_ino, bch2_dirent_hash(&ei->str_hash, name));
-}
-
-int bch2_dirent_rename(struct bch_fs *c,
- struct inode *src_dir, const struct qstr *src_name,
- struct inode *dst_dir, const struct qstr *dst_name,
- u64 *journal_seq, enum bch_rename_mode mode)
-{
- struct bch_inode_info *src_ei = to_bch_ei(src_dir);
- struct bch_inode_info *dst_ei = to_bch_ei(dst_dir);
- struct btree_iter src_iter, dst_iter, whiteout_iter;
+ struct btree_iter src_iter = { NULL };
+ struct btree_iter dst_iter = { NULL };
struct bkey_s_c old_src, old_dst;
- struct bkey delete;
struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
- struct bpos src_pos = bch2_dirent_pos(src_ei, src_name);
- struct bpos dst_pos = bch2_dirent_pos(dst_ei, dst_name);
- bool need_whiteout;
- int ret = -ENOMEM;
-
- bch2_btree_iter_init_intent(&src_iter, c, BTREE_ID_DIRENTS, src_pos);
- bch2_btree_iter_init_intent(&dst_iter, c, BTREE_ID_DIRENTS, dst_pos);
- bch2_btree_iter_link(&src_iter, &dst_iter);
-
- bch2_btree_iter_init(&whiteout_iter, c, BTREE_ID_DIRENTS, src_pos);
- bch2_btree_iter_link(&src_iter, &whiteout_iter);
-
- if (mode == BCH_RENAME_EXCHANGE) {
- new_src = dirent_create_key(0, src_name, 0);
- if (!new_src)
- goto err;
- } else {
- new_src = (void *) &delete;
- }
+ struct bpos dst_pos =
+ POS(dst_dir, bch2_dirent_hash(dst_hash, dst_name));
+ int ret = 0;
- new_dst = dirent_create_key(0, dst_name, 0);
- if (!new_dst)
- goto err;
-retry:
- /*
- * Note that on -EINTR/dropped locks we're not restarting the lookup
- * from the original hashed position (like we do when creating dirents,
- * in bch_hash_set) - we never move existing dirents to different slot:
- */
- old_src = bch2_hash_lookup_at(bch2_dirent_hash_desc,
- &src_ei->str_hash,
- &src_iter, src_name);
- if ((ret = btree_iter_err(old_src)))
- goto err;
-
- ret = bch2_hash_needs_whiteout(bch2_dirent_hash_desc,
- &src_ei->str_hash,
- &whiteout_iter, &src_iter);
- if (ret < 0)
- goto err;
- need_whiteout = ret;
+ *src_inum = *dst_inum = 0;
/*
+ * Lookup dst:
+ *
* Note that in BCH_RENAME mode, we're _not_ checking if
* the target already exists - we're relying on the VFS
* to do that check for us for correctness:
*/
- old_dst = mode == BCH_RENAME
- ? bch2_hash_hole_at(bch2_dirent_hash_desc, &dst_iter)
- : bch2_hash_lookup_at(bch2_dirent_hash_desc,
- &dst_ei->str_hash,
- &dst_iter, dst_name);
- if ((ret = btree_iter_err(old_dst)))
- goto err;
-
- switch (mode) {
- case BCH_RENAME:
- bkey_init(&new_src->k);
- dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
+ ret = mode == BCH_RENAME
+ ? bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
+ dst_hash, dst_dir, dst_name)
+ : bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
+ dst_hash, dst_dir, dst_name,
+ BTREE_ITER_INTENT);
+ if (ret)
+ goto out;
+
+ old_dst = bch2_btree_iter_peek_slot(&dst_iter);
+ ret = bkey_err(old_dst);
+ if (ret)
+ goto out;
+
+ if (mode != BCH_RENAME)
+ *dst_inum = le64_to_cpu(bkey_s_c_to_dirent(old_dst).v->d_inum);
+ if (mode != BCH_RENAME_EXCHANGE)
+ *src_offset = dst_iter.pos.offset;
+
+ /* Lookup src: */
+ ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
+ src_hash, src_dir, src_name,
+ BTREE_ITER_INTENT);
+ if (ret)
+ goto out;
+
+ old_src = bch2_btree_iter_peek_slot(&src_iter);
+ ret = bkey_err(old_src);
+ if (ret)
+ goto out;
+
+ *src_inum = le64_to_cpu(bkey_s_c_to_dirent(old_src).v->d_inum);
+
+ /* Create new dst key: */
+ new_dst = dirent_create_key(trans, 0, dst_name, 0);
+ ret = PTR_ERR_OR_ZERO(new_dst);
+ if (ret)
+ goto out;
+
+ dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
+ new_dst->k.p = dst_iter.pos;
- if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
- bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
- /*
- * If we couldn't insert new_dst at its hashed
- * position (dst_pos) due to a hash collision,
- * and we're going to be deleting in
- * between the hashed position and first empty
- * slot we found - just overwrite the pos we
- * were going to delete:
- *
- * Note: this is a correctness issue, in this
- * situation bch2_hash_needs_whiteout() could
- * return false when the whiteout would have
- * been needed if we inserted at the pos
- * __dirent_find_hole() found
- */
- new_dst->k.p = src_iter.pos;
- ret = bch2_btree_insert_at(c, NULL, NULL,
- journal_seq,
- BTREE_INSERT_ATOMIC,
- BTREE_INSERT_ENTRY(&src_iter,
- &new_dst->k_i));
- goto err;
- }
+ /* Create new src key: */
+ if (mode == BCH_RENAME_EXCHANGE) {
+ new_src = dirent_create_key(trans, 0, src_name, 0);
+ ret = PTR_ERR_OR_ZERO(new_src);
+ if (ret)
+ goto out;
+
+ dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
+ new_src->k.p = src_iter.pos;
+ } else {
+ new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
+ ret = PTR_ERR_OR_ZERO(new_src);
+ if (ret)
+ goto out;
- if (need_whiteout)
- new_src->k.type = BCH_DIRENT_WHITEOUT;
- break;
- case BCH_RENAME_OVERWRITE:
bkey_init(&new_src->k);
- dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
+ new_src->k.p = src_iter.pos;
if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
/*
- * Same case described above -
- * bch_hash_needs_whiteout could spuriously
- * return false, but we have to insert at
- * dst_iter.pos because we're overwriting
- * another dirent:
+ * We have a hash collision for the new dst key,
+ * and new_src - the key we're deleting - is between
+ * new_dst's hashed slot and the slot we're going to be
+ * inserting it into - oops. This will break the hash
+ * table if we don't deal with it:
*/
- new_src->k.type = BCH_DIRENT_WHITEOUT;
- } else if (need_whiteout)
- new_src->k.type = BCH_DIRENT_WHITEOUT;
- break;
- case BCH_RENAME_EXCHANGE:
- dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
- dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
- break;
+ if (mode == BCH_RENAME) {
+ /*
+ * If we're not overwriting, we can just insert
+ * new_dst at the src position:
+ */
+ new_dst->k.p = src_iter.pos;
+ bch2_trans_update(trans, &src_iter,
+ &new_dst->k_i, 0);
+ goto out_set_offset;
+ } else {
+ /* If we're overwriting, we can't insert new_dst
+ * at a different slot because it has to
+ * overwrite old_dst - just make sure to use a
+ * whiteout when deleting src:
+ */
+ new_src->k.type = KEY_TYPE_hash_whiteout;
+ }
+ } else {
+ /* Check if we need a whiteout to delete src: */
+ ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
+ src_hash, &src_iter);
+ if (ret < 0)
+ goto out;
+
+ if (ret)
+ new_src->k.type = KEY_TYPE_hash_whiteout;
+ }
}
- new_src->k.p = src_iter.pos;
- new_dst->k.p = dst_iter.pos;
- ret = bch2_btree_insert_at(c, NULL, NULL, journal_seq,
- BTREE_INSERT_ATOMIC,
- BTREE_INSERT_ENTRY(&src_iter, &new_src->k_i),
- BTREE_INSERT_ENTRY(&dst_iter, &new_dst->k_i));
-err:
- if (ret == -EINTR)
- goto retry;
-
- bch2_btree_iter_unlock(&whiteout_iter);
- bch2_btree_iter_unlock(&dst_iter);
- bch2_btree_iter_unlock(&src_iter);
-
- if (new_src != (void *) &delete)
- kfree(new_src);
- kfree(new_dst);
+ bch2_trans_update(trans, &src_iter, &new_src->k_i, 0);
+ bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
+out_set_offset:
+ if (mode == BCH_RENAME_EXCHANGE)
+ *src_offset = new_src->k.p.offset;
+ *dst_offset = new_dst->k.p.offset;
+out:
+ bch2_trans_iter_exit(trans, &src_iter);
+ bch2_trans_iter_exit(trans, &dst_iter);
return ret;
}
-int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
- const struct bch_hash_info *hash_info,
- const struct qstr *name,
- u64 *journal_seq)
+int bch2_dirent_delete_at(struct btree_trans *trans,
+ const struct bch_hash_info *hash_info,
+ struct btree_iter *iter)
+{
+ return bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
+ hash_info, iter);
+}
+
+int __bch2_dirent_lookup_trans(struct btree_trans *trans,
+ struct btree_iter *iter,
+ u64 dir_inum,
+ const struct bch_hash_info *hash_info,
+ const struct qstr *name, unsigned flags)
{
- return bch2_hash_delete(bch2_dirent_hash_desc, hash_info,
- c, dir_inum, journal_seq, name);
+ return bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
+ hash_info, dir_inum, name, flags);
}
u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
const struct bch_hash_info *hash_info,
const struct qstr *name)
{
+ struct btree_trans trans;
struct btree_iter iter;
struct bkey_s_c k;
- u64 inum;
+ u64 inum = 0;
+ int ret = 0;
- k = bch2_hash_lookup(bch2_dirent_hash_desc, hash_info, c,
- dir_inum, &iter, name);
- if (IS_ERR(k.k)) {
- bch2_btree_iter_unlock(&iter);
- return 0;
- }
+ bch2_trans_init(&trans, c, 0, 0);
- inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
- bch2_btree_iter_unlock(&iter);
+ ret = __bch2_dirent_lookup_trans(&trans, &iter, dir_inum,
+ hash_info, name, 0);
+ if (ret)
+ goto out;
+
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto out;
+ inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
+ bch2_trans_iter_exit(&trans, &iter);
+out:
+ BUG_ON(ret == -EINTR);
+ bch2_trans_exit(&trans);
return inum;
}
-int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
+int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
{
struct btree_iter iter;
struct bkey_s_c k;
- int ret = 0;
+ int ret;
- for_each_btree_key(&iter, c, BTREE_ID_DIRENTS, POS(dir_inum, 0), k) {
+ for_each_btree_key(trans, iter, BTREE_ID_dirents,
+ POS(dir_inum, 0), 0, k, ret) {
if (k.k->p.inode > dir_inum)
break;
- if (k.k->type == BCH_DIRENT) {
+ if (k.k->type == KEY_TYPE_dirent) {
ret = -ENOTEMPTY;
break;
}
}
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
-int bch2_readdir(struct bch_fs *c, struct file *file,
- struct dir_context *ctx)
+int bch2_readdir(struct bch_fs *c, u64 inum, struct dir_context *ctx)
{
- struct inode *inode = file_inode(file);
+ struct btree_trans trans;
struct btree_iter iter;
struct bkey_s_c k;
struct bkey_s_c_dirent dirent;
- unsigned len;
+ int ret;
- if (!dir_emit_dots(file, ctx))
- return 0;
+ bch2_trans_init(&trans, c, 0, 0);
- pr_debug("listing for %lu from %llu", inode->i_ino, ctx->pos);
+ for_each_btree_key(&trans, iter, BTREE_ID_dirents,
+ POS(inum, ctx->pos), 0, k, ret) {
+ if (k.k->p.inode > inum)
+ break;
- for_each_btree_key(&iter, c, BTREE_ID_DIRENTS,
- POS(inode->i_ino, ctx->pos), k) {
- if (k.k->type != BCH_DIRENT)
+ if (k.k->type != KEY_TYPE_dirent)
continue;
dirent = bkey_s_c_to_dirent(k);
- pr_debug("saw %llu:%llu (%s) -> %llu",
- k.k->p.inode, k.k->p.offset,
- dirent.v->d_name, dirent.v->d_inum);
-
- if (bkey_cmp(k.k->p, POS(inode->i_ino, ctx->pos)) < 0)
- continue;
-
- if (k.k->p.inode > inode->i_ino)
- break;
-
- len = bch2_dirent_name_bytes(dirent);
-
- pr_debug("emitting %s", dirent.v->d_name);
-
/*
* XXX: dir_emit() can fault and block, while we're holding
* locks
*/
- if (!dir_emit(ctx, dirent.v->d_name, len,
+ ctx->pos = dirent.k->p.offset;
+ if (!dir_emit(ctx, dirent.v->d_name,
+ bch2_dirent_name_bytes(dirent),
le64_to_cpu(dirent.v->d_inum),
dirent.v->d_type))
break;
-
- ctx->pos = k.k->p.offset + 1;
+ ctx->pos = dirent.k->p.offset + 1;
}
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_iter_exit(&trans, &iter);
- return 0;
+ ret = bch2_trans_exit(&trans) ?: ret;
+
+ return ret;
}