X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fdirent.c;h=a7559ab03802cb3a39a6dc17d563cb9f3dced5a4;hb=54b6beabf05b6cf62092f98f0c06395e4242b064;hp=14a9a2c0ef0243297c432dc1369070008429e3b8;hpb=d13bbb2955f899f10be4ab315ad229d2951fdcda;p=bcachefs-tools-debian

diff --git a/libbcachefs/dirent.c b/libbcachefs/dirent.c
index 14a9a2c..a7559ab 100644
--- a/libbcachefs/dirent.c
+++ b/libbcachefs/dirent.c
@@ -1,5 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "bkey_buf.h"
 #include "bkey_methods.h"
 #include "btree_update.h"
 #include "extents.h"
@@ -7,15 +9,29 @@
 #include "fs.h"
 #include "keylist.h"
 #include "str_hash.h"
+#include "subvolume.h"
 
 #include <linux/dcache.h>
 
-unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
+static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
 {
-	unsigned len = bkey_val_bytes(d.k) -
-		offsetof(struct bch_dirent, d_name);
+	unsigned bkey_u64s = bkey_val_u64s(d.k);
+	unsigned bkey_bytes = bkey_u64s * sizeof(u64);
+	u64 last_u64 = ((u64*)d.v)[bkey_u64s - 1];
+#if CPU_BIG_ENDIAN
+	unsigned trailing_nuls = last_u64 ? __builtin_ctzll(last_u64) / 8 : 64 / 8;
+#else
+	unsigned trailing_nuls = last_u64 ? __builtin_clzll(last_u64) / 8 : 64 / 8;
+#endif
+
+	return bkey_bytes -
+		offsetof(struct bch_dirent, d_name) -
+		trailing_nuls;
+}
 
-	return strnlen(d.v->d_name, len);
+struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
+{
+	return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
 }
 
 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
@@ -38,7 +54,7 @@ static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
-	struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
+	struct qstr name = bch2_dirent_get_name(d);
 
 	return bch2_dirent_hash(info, &name);
 }
@@ -46,68 +62,116 @@ static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
 {
 	struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
-	int len = bch2_dirent_name_bytes(l);
-	const struct qstr *r = _r;
+	const struct qstr l_name = bch2_dirent_get_name(l);
+	const struct qstr *r_name = _r;
 
-	return len - r->len ?: memcmp(l.v->d_name, r->name, len);
+	return l_name.len - r_name->len ?: memcmp(l_name.name, r_name->name, l_name.len);
 }
 
 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
 {
 	struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
 	struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
-	int l_len = bch2_dirent_name_bytes(l);
-	int r_len = bch2_dirent_name_bytes(r);
+	const struct qstr l_name = bch2_dirent_get_name(l);
+	const struct qstr r_name = bch2_dirent_get_name(r);
+
+	return l_name.len - r_name.len ?: memcmp(l_name.name, r_name.name, l_name.len);
+}
 
-	return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
+static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
+{
+	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
+
+	if (d.v->d_type == DT_SUBVOL)
+		return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
+	return true;
 }
 
 const struct bch_hash_desc bch2_dirent_hash_desc = {
-	.btree_id	= BTREE_ID_DIRENTS,
+	.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,
+	.is_visible	= dirent_is_visible,
 };
 
-const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
+int bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			enum bkey_invalid_flags flags,
+			struct printbuf *err)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
-	unsigned len;
+	struct qstr d_name = bch2_dirent_get_name(d);
 
-	if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
-		return "value too small";
+	if (!d_name.len) {
+		prt_printf(err, "empty name");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	len = bch2_dirent_name_bytes(d);
-	if (!len)
-		return "empty name";
+	if (bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len)) {
+		prt_printf(err, "value too big (%zu > %u)",
+		       bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
+		return -BCH_ERR_invalid_bkey;
+	}
 
 	/*
-	 * older versions of bcachefs were buggy and creating dirent
-	 * keys that were bigger than necessary:
+	 * Check new keys don't exceed the max length
+	 * (older keys may be larger.)
 	 */
-	if (bkey_val_u64s(k.k) > dirent_val_u64s(len + 7))
-		return "value too big";
+	if ((flags & BKEY_INVALID_COMMIT) && d_name.len > BCH_NAME_MAX) {
+		prt_printf(err, "dirent name too big (%u > %u)",
+		       d_name.len, BCH_NAME_MAX);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (d_name.len != strnlen(d_name.name, d_name.len)) {
+		prt_printf(err, "dirent has stray data after name's NUL");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (d_name.len == 1 && !memcmp(d_name.name, ".", 1)) {
+		prt_printf(err, "invalid name");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (d_name.len == 2 && !memcmp(d_name.name, "..", 2)) {
+		prt_printf(err, "invalid name");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	if (len > BCH_NAME_MAX)
-		return "dirent name too big";
+	if (memchr(d_name.name, '/', d_name.len)) {
+		prt_printf(err, "invalid name");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (d.v->d_type != DT_SUBVOL &&
+	    le64_to_cpu(d.v->d_inum) == d.k->p.inode) {
+		prt_printf(err, "dirent points to own directory");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	return NULL;
+	return 0;
 }
 
 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
 			 struct bkey_s_c k)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
-
-	bch_scnmemcpy(out, d.v->d_name,
-		      bch2_dirent_name_bytes(d));
-	pr_buf(out, " -> %llu", d.v->d_inum);
+	struct qstr d_name = bch2_dirent_get_name(d);
+
+	prt_printf(out, "%.*s -> %llu type %s",
+	       d_name.len,
+	       d_name.name,
+	       d.v->d_type != DT_SUBVOL
+	       ? le64_to_cpu(d.v->d_inum)
+	       : le32_to_cpu(d.v->d_child_subvol),
+	       bch2_d_type_str(d.v->d_type));
 }
 
 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
-				u8 type, const struct qstr *name, u64 dst)
+				subvol_inum dir, u8 type,
+				const struct qstr *name, u64 dst)
 {
 	struct bkey_i_dirent *dirent;
 	unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
@@ -123,7 +187,14 @@ static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 
 	bkey_dirent_init(&dirent->k_i);
 	dirent->k.u64s = u64s;
-	dirent->v.d_inum = cpu_to_le64(dst);
+
+	if (type != DT_SUBVOL) {
+		dirent->v.d_inum = cpu_to_le64(dst);
+	} else {
+		dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
+		dirent->v.d_child_subvol = cpu_to_le32(dst);
+	}
+
 	dirent->v.d_type = type;
 
 	memcpy(dirent->v.d_name, name->name, name->len);
@@ -137,31 +208,24 @@ static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 	return dirent;
 }
 
-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,
-			 int flags)
+int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
+		       const struct bch_hash_info *hash_info,
+		       u8 type, const struct qstr *name, u64 dst_inum,
+		       u64 *dir_offset, int flags)
 {
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, type, name, dst_inum);
+	dirent = dirent_create_key(trans, dir, type, name, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
 
-	return bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
-			     dir_inum, &dirent->k_i, flags);
-}
+	ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
+			    dir, &dirent->k_i, flags);
+	*dir_offset = dirent->k.p.offset;
 
-int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
-		       const struct bch_hash_info *hash_info,
-		       u8 type, const struct qstr *name, u64 dst_inum,
-		       u64 *journal_seq, int flags)
-{
-	return bch2_trans_do(c, journal_seq, flags,
-		__bch2_dirent_create(&trans, dir_inum, hash_info,
-				     type, name, dst_inum, flags));
+	return ret;
 }
 
 static void dirent_copy_target(struct bkey_i_dirent *dst,
@@ -171,76 +235,141 @@ static void dirent_copy_target(struct bkey_i_dirent *dst,
 	dst->v.d_type = src.v->d_type;
 }
 
-static struct bpos bch2_dirent_pos(struct bch_inode_info *inode,
-				   const struct qstr *name)
+int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
+			    struct bkey_s_c_dirent d, subvol_inum *target)
 {
-	return POS(inode->v.i_ino, bch2_dirent_hash(&inode->ei_str_hash, name));
+	struct bch_subvolume s;
+	int ret = 0;
+
+	if (d.v->d_type == DT_SUBVOL &&
+	    le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
+		return 1;
+
+	if (likely(d.v->d_type != DT_SUBVOL)) {
+		target->subvol	= dir.subvol;
+		target->inum	= le64_to_cpu(d.v->d_inum);
+	} else {
+		target->subvol	= le32_to_cpu(d.v->d_child_subvol);
+
+		ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
+
+		target->inum	= le64_to_cpu(s.inode);
+	}
+
+	return ret;
 }
 
 int bch2_dirent_rename(struct btree_trans *trans,
-		struct bch_inode_info *src_dir, const struct qstr *src_name,
-		struct bch_inode_info *dst_dir, const struct qstr *dst_name,
+		subvol_inum src_dir, struct bch_hash_info *src_hash,
+		subvol_inum dst_dir, struct bch_hash_info *dst_hash,
+		const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
+		const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
 		enum bch_rename_mode mode)
 {
-	struct btree_iter *src_iter, *dst_iter;
-	struct bkey_s_c old_src, old_dst;
+	struct btree_iter src_iter = { NULL };
+	struct btree_iter dst_iter = { NULL };
+	struct bkey_s_c old_src, old_dst = bkey_s_c_null;
 	struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
-	struct bpos dst_pos = bch2_dirent_pos(dst_dir, dst_name);
-	int ret;
+	struct bpos dst_pos =
+		POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
+	unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
+	int ret = 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:
-	 */
-	dst_iter = mode == BCH_RENAME
-		? bch2_hash_hole(trans, bch2_dirent_hash_desc,
-				 &dst_dir->ei_str_hash,
-				 dst_dir->v.i_ino, dst_name)
-		: bch2_hash_lookup(trans, bch2_dirent_hash_desc,
-				   &dst_dir->ei_str_hash,
-				   dst_dir->v.i_ino, dst_name,
-				   BTREE_ITER_INTENT);
-	if (IS_ERR(dst_iter))
-		return PTR_ERR(dst_iter);
-	old_dst = bch2_btree_iter_peek_slot(dst_iter);
+	if (src_dir.subvol != dst_dir.subvol)
+		return -EXDEV;
+
+	memset(src_inum, 0, sizeof(*src_inum));
+	memset(dst_inum, 0, sizeof(*dst_inum));
 
 	/* Lookup src: */
-	src_iter = bch2_hash_lookup(trans, bch2_dirent_hash_desc,
-				    &src_dir->ei_str_hash,
-				    src_dir->v.i_ino, src_name,
-				    BTREE_ITER_INTENT);
-	if (IS_ERR(src_iter))
-		return PTR_ERR(src_iter);
-	old_src = bch2_btree_iter_peek_slot(src_iter);
+	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;
+
+	ret = bch2_dirent_read_target(trans, src_dir,
+			bkey_s_c_to_dirent(old_src), src_inum);
+	if (ret)
+		goto out;
+
+	src_type = bkey_s_c_to_dirent(old_src).v->d_type;
+
+	if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
+		return -EOPNOTSUPP;
+
+
+	/* Lookup dst: */
+	if (mode == BCH_RENAME) {
+		/*
+		 * Note that we're _not_ checking if the target already exists -
+		 * we're relying on the VFS to do that check for us for
+		 * correctness:
+		 */
+		ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
+				     dst_hash, dst_dir, dst_name);
+		if (ret)
+			goto out;
+	} else {
+		ret = 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;
+
+		ret = bch2_dirent_read_target(trans, dst_dir,
+				bkey_s_c_to_dirent(old_dst), dst_inum);
+		if (ret)
+			goto out;
+
+		dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
+
+		if (dst_type == DT_SUBVOL)
+			return -EOPNOTSUPP;
+	}
+
+	if (mode != BCH_RENAME_EXCHANGE)
+		*src_offset = dst_iter.pos.offset;
 
 	/* Create new dst key: */
-	new_dst = dirent_create_key(trans, 0, dst_name, 0);
-	if (IS_ERR(new_dst))
-		return PTR_ERR(new_dst);
+	new_dst = dirent_create_key(trans, dst_dir, 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;
+	new_dst->k.p = dst_iter.pos;
 
 	/* Create new src key: */
 	if (mode == BCH_RENAME_EXCHANGE) {
-		new_src = dirent_create_key(trans, 0, src_name, 0);
-		if (IS_ERR(new_src))
-			return PTR_ERR(new_src);
+		new_src = dirent_create_key(trans, src_dir, 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;
+		new_src->k.p = src_iter.pos;
 	} else {
 		new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
-		if (IS_ERR(new_src))
-			return PTR_ERR(new_src);
+		ret = PTR_ERR_OR_ZERO(new_src);
+		if (ret)
+			goto out;
+
 		bkey_init(&new_src->k);
-		new_src->k.p = src_iter->pos;
+		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) {
+		if (bkey_le(dst_pos, src_iter.pos) &&
+		    bkey_lt(src_iter.pos, dst_iter.pos)) {
 			/*
 			 * We have a hash collision for the new dst key,
 			 * and new_src - the key we're deleting - is between
@@ -253,154 +382,209 @@ int bch2_dirent_rename(struct btree_trans *trans,
 				 * 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,
-					BTREE_INSERT_ENTRY(src_iter,
-							   &new_dst->k_i));
-				return 0;
+				new_src = new_dst;
+				new_src->k.p = src_iter.pos;
+				goto out_set_src;
 			} 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_whiteout;
+				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_dir->ei_str_hash,
-						       src_iter);
+						       src_hash, &src_iter);
 			if (ret < 0)
-				return ret;
+				goto out;
 
 			if (ret)
-				new_src->k.type = KEY_TYPE_whiteout;
+				new_src->k.type = KEY_TYPE_hash_whiteout;
 		}
 	}
 
-	bch2_trans_update(trans, BTREE_INSERT_ENTRY(src_iter, &new_src->k_i));
-	bch2_trans_update(trans, BTREE_INSERT_ENTRY(dst_iter, &new_dst->k_i));
-	return 0;
-}
+	ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
+	if (ret)
+		goto out;
+out_set_src:
 
-int __bch2_dirent_delete(struct btree_trans *trans, u64 dir_inum,
-			 const struct bch_hash_info *hash_info,
-			 const struct qstr *name)
-{
-	return bch2_hash_delete(trans, bch2_dirent_hash_desc, hash_info,
-				dir_inum, name);
+	/*
+	 * If we're deleting a subvolume, we need to really delete the dirent,
+	 * not just emit a whiteout in the current snapshot:
+	 */
+	if (src_type == DT_SUBVOL) {
+		bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
+		ret = bch2_btree_iter_traverse(&src_iter);
+		if (ret)
+			goto out;
+
+		new_src->k.p = src_iter.pos;
+		src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
+	}
+
+	ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
+	if (ret)
+		goto out;
+
+	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_lookup_trans(struct btree_trans *trans,
+			       struct btree_iter *iter,
+			       subvol_inum dir,
+			       const struct bch_hash_info *hash_info,
+			       const struct qstr *name, subvol_inum *inum,
+			       unsigned flags)
 {
-	return bch2_trans_do(c, journal_seq,
-			     BTREE_INSERT_ATOMIC|
-			     BTREE_INSERT_NOFAIL,
-		__bch2_dirent_delete(&trans, dir_inum, hash_info, name));
+	struct bkey_s_c k;
+	struct bkey_s_c_dirent d;
+	u32 snapshot;
+	int ret;
+
+	ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
+	if (ret)
+		return ret;
+
+	ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
+			       hash_info, dir, name, flags);
+	if (ret)
+		return ret;
+
+	k = bch2_btree_iter_peek_slot(iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
+
+	d = bkey_s_c_to_dirent(k);
+
+	ret = bch2_dirent_read_target(trans, dir, d, inum);
+	if (ret > 0)
+		ret = -ENOENT;
+err:
+	if (ret)
+		bch2_trans_iter_exit(trans, iter);
+
+	return ret;
 }
 
-u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
+u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
 		       const struct bch_hash_info *hash_info,
-		       const struct qstr *name)
+		       const struct qstr *name, subvol_inum *inum)
 {
 	struct btree_trans trans;
-	struct btree_iter *iter;
-	struct bkey_s_c k;
-	u64 inum = 0;
-
-	bch2_trans_init(&trans, c);
+	struct btree_iter iter;
+	int ret;
 
-	iter = bch2_hash_lookup(&trans, bch2_dirent_hash_desc,
-				hash_info, dir_inum, name, 0);
-	if (IS_ERR(iter)) {
-		BUG_ON(PTR_ERR(iter) == -EINTR);
-		goto out;
-	}
+	bch2_trans_init(&trans, c, 0, 0);
+retry:
+	bch2_trans_begin(&trans);
 
-	k = bch2_btree_iter_peek_slot(iter);
-	inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
-out:
+	ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
+					  name, inum, 0);
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		goto retry;
+	if (!ret)
+		bch2_trans_iter_exit(&trans, &iter);
 	bch2_trans_exit(&trans);
-	return inum;
+	return ret;
 }
 
-int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
+int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
 {
-	struct btree_iter *iter;
+	struct btree_iter iter;
 	struct bkey_s_c k;
-	int ret = 0;
-
-	iter = bch2_trans_get_iter(trans, BTREE_ID_DIRENTS,
-				   POS(dir_inum, 0), 0);
-	if (IS_ERR(iter))
-		return PTR_ERR(iter);
+	u32 snapshot;
+	int ret;
 
-	for_each_btree_key_continue(iter, 0, k) {
-		if (k.k->p.inode > dir_inum)
-			break;
+	ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
+	if (ret)
+		return ret;
 
+	for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
+			   SPOS(dir.inum, 0, snapshot),
+			   POS(dir.inum, U64_MAX), 0, k, ret)
 		if (k.k->type == KEY_TYPE_dirent) {
 			ret = -ENOTEMPTY;
 			break;
 		}
-	}
-	bch2_trans_iter_put(trans, iter);
+	bch2_trans_iter_exit(trans, &iter);
 
 	return ret;
 }
 
-int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
-{
-	return bch2_trans_do(c, NULL, 0,
-		bch2_empty_dir_trans(&trans, dir_inum));
-}
-
-int bch2_readdir(struct bch_fs *c, struct file *file,
-		 struct dir_context *ctx)
+int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 {
-	struct bch_inode_info *inode = file_bch_inode(file);
 	struct btree_trans trans;
-	struct btree_iter *iter;
+	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct bkey_s_c_dirent dirent;
-	unsigned len;
+	subvol_inum target;
+	u32 snapshot;
+	struct bkey_buf sk;
+	struct qstr name;
+	int ret;
 
-	if (!dir_emit_dots(file, ctx))
-		return 0;
+	bch2_bkey_buf_init(&sk);
+	bch2_trans_init(&trans, c, 0, 0);
+retry:
+	bch2_trans_begin(&trans);
 
-	bch2_trans_init(&trans, c);
+	ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
+	if (ret)
+		goto err;
 
-	for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
-			   POS(inode->v.i_ino, ctx->pos), 0, k) {
+	for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_dirents,
+			   SPOS(inum.inum, ctx->pos, snapshot),
+			   POS(inum.inum, U64_MAX), 0, k, ret) {
 		if (k.k->type != KEY_TYPE_dirent)
 			continue;
 
 		dirent = bkey_s_c_to_dirent(k);
 
-		if (bkey_cmp(k.k->p, POS(inode->v.i_ino, ctx->pos)) < 0)
+		ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
+		if (ret < 0)
+			break;
+		if (ret)
 			continue;
 
-		if (k.k->p.inode > inode->v.i_ino)
-			break;
+		/* dir_emit() can fault and block: */
+		bch2_bkey_buf_reassemble(&sk, c, k);
+		dirent = bkey_i_to_s_c_dirent(sk.k);
+		bch2_trans_unlock(&trans);
+
+		name = bch2_dirent_get_name(dirent);
 
-		len = bch2_dirent_name_bytes(dirent);
+		ctx->pos = dirent.k->p.offset;
+		if (!dir_emit(ctx, name.name,
+			      name.len,
+			      target.inum,
+			      vfs_d_type(dirent.v->d_type)))
+			break;
+		ctx->pos = dirent.k->p.offset + 1;
 
 		/*
-		 * XXX: dir_emit() can fault and block, while we're holding
-		 * locks
+		 * read_target looks up subvolumes, we can overflow paths if the
+		 * directory has many subvolumes in it
 		 */
-		if (!dir_emit(ctx, dirent.v->d_name, len,
-			      le64_to_cpu(dirent.v->d_inum),
-			      dirent.v->d_type))
+		ret = btree_trans_too_many_iters(&trans);
+		if (ret)
 			break;
-
-		ctx->pos = k.k->p.offset + 1;
 	}
+	bch2_trans_iter_exit(&trans, &iter);
+err:
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		goto retry;
+
 	bch2_trans_exit(&trans);
+	bch2_bkey_buf_exit(&sk, c);
 
-	return 0;
+	return ret;
 }