Skip to contentSkip to conversation

fs: allow clone_private_mount() for a path on real rootfs

PATCH Superseded KONDO KAZUMA(近藤 和真) 2 participants 5 messages May 14, 2025 – May 15, 2025 VFS

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fs: allow clone_private_mount() for a path on real rootfs

Mounting overlayfs with a directory on real rootfs (initramfs) as upperdir has failed with following message since commit [1].

  [    4.080134] overlayfs: failed to clone upperpath

Overlayfs mount uses clone_private_mount() to create internal mount for the underlying layers.

The commit [1] made clone_private_mount() reject real rootfs because it does not have a parent mount and is in the initial mount namespace, that is not an anonymous mount namespace.

This issue can be fixed by relaxing the permission check of clone_private_mount(), similar to [2].

[1] commit db04662e2f4f ("fs: allow detached mounts in clone_private_mount()")
[2] commit 46f5ab762d04 ("fs: relax mount_setattr() permission checks")

Fixes db04662e2f4f ("fs: allow detached mounts in clone_private_mount()")
---
 fs/namespace.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)
fs/namespace.c +4 -8
index 1b466c54a357..277dbf18e160 100644
@@ -2482,17 +2482,13 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	if (IS_MNT_UNBINDABLE(old_mnt))
 		return ERR_PTR(-EINVAL);
 
-	if (mnt_has_parent(old_mnt)) {
+	if (!is_mounted(&old_mnt->mnt))
+		return ERR_PTR(-EINVAL);
+
+	if (mnt_has_parent(old_mnt) || !is_anon_ns(old_mnt->mnt_ns)) {
 		if (!check_mnt(old_mnt))
 			return ERR_PTR(-EINVAL);
 	} else {
-		if (!is_mounted(&old_mnt->mnt))
-			return ERR_PTR(-EINVAL);
-
-		/* Make sure this isn't something purely kernel internal. */
-		if (!is_anon_ns(old_mnt->mnt_ns))
-			return ERR_PTR(-EINVAL);
-
 		/* Make sure we don't create mount namespace loops. */
 		if (!check_for_nsfs_mounts(old_mnt))
 			return ERR_PTR(-EINVAL);
Signature · 1 line
2.49.0
maintainer
#1 lore raw
Quoted 12 lines · KONDO KAZUMA(近藤 和真)
fs/namespace.c +4 -1
@@ -2482,17 +2482,13 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	if (IS_MNT_UNBINDABLE(old_mnt))
 		return ERR_PTR(-EINVAL);
 
-	if (mnt_has_parent(old_mnt)) {
+	if (!is_mounted(&old_mnt->mnt))
+		return ERR_PTR(-EINVAL);
+
+	if (mnt_has_parent(old_mnt) || !is_anon_ns(old_mnt->mnt_ns)) {
 		if (!check_mnt(old_mnt))
 			return ERR_PTR(-EINVAL);
 	} else {

fs/namespace.c struct vfsmount *clone_private_mount(const struct path *path) · line 2494

Quoted 10 lines
fs/namespace.c +0 -7
-		if (!is_mounted(&old_mnt->mnt))
-			return ERR_PTR(-EINVAL);
-
-		/* Make sure this isn't something purely kernel internal. */
-		if (!is_anon_ns(old_mnt->mnt_ns))
-			return ERR_PTR(-EINVAL);
-
 		/* Make sure we don't create mount namespace loops. */
 		if (!check_for_nsfs_mounts(old_mnt))
 			return ERR_PTR(-EINVAL);

Not the right way to do that. What we want is

	/* ours are always fine */
	if (!check_mnt(old_mnt)) {
		/* they'd better be mounted _somewhere */
		if (!is_mounted(old_mnt))
			return -EINVAL;
		/* no other real namespaces; only anon */
		if (!is_anon_ns(old_mnt->mnt_ns))
			return -EINVAL;
		/* ... and root of that anon */
		if (mnt_has_parent(old_mnt))
			return -EINVAL;
		/* Make sure we don't create mount namespace loops. */
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}
#2 lore raw
Quoted 27 lines · Al Viro
Quoted 12 lines · KONDO KAZUMA(近藤 和真)
fs/namespace.c +4 -1
@@ -2482,17 +2482,13 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	if (IS_MNT_UNBINDABLE(old_mnt))
 		return ERR_PTR(-EINVAL);
 
-	if (mnt_has_parent(old_mnt)) {
+	if (!is_mounted(&old_mnt->mnt))
+		return ERR_PTR(-EINVAL);
+
+	if (mnt_has_parent(old_mnt) || !is_anon_ns(old_mnt->mnt_ns)) {
 		if (!check_mnt(old_mnt))
 			return ERR_PTR(-EINVAL);
 	} else {

fs/namespace.c struct vfsmount *clone_private_mount(const struct path *path) · line 2494

Quoted 10 lines
fs/namespace.c +0 -7
-		if (!is_mounted(&old_mnt->mnt))
-			return ERR_PTR(-EINVAL);
-
-		/* Make sure this isn't something purely kernel internal. */
-		if (!is_anon_ns(old_mnt->mnt_ns))
-			return ERR_PTR(-EINVAL);
-
 		/* Make sure we don't create mount namespace loops. */
 		if (!check_for_nsfs_mounts(old_mnt))
 			return ERR_PTR(-EINVAL);

Not the right way to do that. What we want is

Quoted 15 lines
	/* ours are always fine */
	if (!check_mnt(old_mnt)) {
		/* they'd better be mounted _somewhere */
		if (!is_mounted(old_mnt))
			return -EINVAL;
		/* no other real namespaces; only anon */
		if (!is_anon_ns(old_mnt->mnt_ns))
			return -EINVAL;
		/* ... and root of that anon */
		if (mnt_has_parent(old_mnt))
			return -EINVAL;
		/* Make sure we don't create mount namespace loops. */
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}

Hello Al Viro,

Thank you for your comment.
That code can solve my problem, and it seems to be better!

So, I will revise my patch and resend it.

Thanks,
Kazuma Kondo

maintainer
#3 lore raw
Quoted 44 lines · KONDO KAZUMA(近藤 和真)
Quoted 27 lines · Al Viro
Quoted 12 lines · KONDO KAZUMA(近藤 和真)
fs/namespace.c +4 -1
@@ -2482,17 +2482,13 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	if (IS_MNT_UNBINDABLE(old_mnt))
 		return ERR_PTR(-EINVAL);
 
-	if (mnt_has_parent(old_mnt)) {
+	if (!is_mounted(&old_mnt->mnt))
+		return ERR_PTR(-EINVAL);
+
+	if (mnt_has_parent(old_mnt) || !is_anon_ns(old_mnt->mnt_ns)) {
 		if (!check_mnt(old_mnt))
 			return ERR_PTR(-EINVAL);
 	} else {

fs/namespace.c struct vfsmount *clone_private_mount(const struct path *path) · line 2494

Quoted 10 lines
fs/namespace.c +0 -7
-		if (!is_mounted(&old_mnt->mnt))
-			return ERR_PTR(-EINVAL);
-
-		/* Make sure this isn't something purely kernel internal. */
-		if (!is_anon_ns(old_mnt->mnt_ns))
-			return ERR_PTR(-EINVAL);
-
 		/* Make sure we don't create mount namespace loops. */
 		if (!check_for_nsfs_mounts(old_mnt))
 			return ERR_PTR(-EINVAL);

Not the right way to do that. What we want is

Quoted 15 lines
	/* ours are always fine */
	if (!check_mnt(old_mnt)) {
		/* they'd better be mounted _somewhere */
		if (!is_mounted(old_mnt))
			return -EINVAL;
		/* no other real namespaces; only anon */
		if (!is_anon_ns(old_mnt->mnt_ns))
			return -EINVAL;
		/* ... and root of that anon */
		if (mnt_has_parent(old_mnt))
			return -EINVAL;
		/* Make sure we don't create mount namespace loops. */
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}
Quoted 4 lines

Hello Al Viro,

Thank you for your comment.
That code can solve my problem, and it seems to be better!

BTW, see https://lore.kernel.org/all/20250506194849.GT2023217@ZenIV/ for discussion about a week ago when that got noticed:

|| In case of clone_private_mount(), though, there's nothing wrong
|| with "clone me a subtree of absolute root", so it has to be
|| done other way round - check if it's ours first, then in "not
|| ours" case check that it's a root of anon namespace.
||
|| Failing btrfs mount has ended up with upper layer pathname
|| pointing to initramfs directory where btrfs would've been
|| mounted, which had walked into that corner case.  In your
|| case the problem has already happened by that point, but on
|| a setup a-la X Terminal it would cause trouble...

Looks like such setups are less theoretical than I thought.

Quoted 1 line

So, I will revise my patch and resend it.

Probably worth gathering the comments in one place. Something like

	/*
	 * Check if the source is acceptable; anything mounted in
	 * our namespace is fine, otherwise it must be the root of
	 * some anon namespace and we need to make sure no namespace
	 * loops get created.
	 */
	if (!check_mnt(old_mnt)) {
		if (!is_mounted(&old_mnt->mnt) ||
		    !is_anon_ns(old_mnt->mnt_ns) ||
		    mnt_has_parent(old_mnt))
			return ERR_PTR(-EINVAL);
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}

might be easier to follow.

#4 lore raw
Quoted 83 lines · Al Viro
Quoted 44 lines · KONDO KAZUMA(近藤 和真)
Quoted 27 lines · Al Viro
Quoted 12 lines · KONDO KAZUMA(近藤 和真)
fs/namespace.c +4 -1
@@ -2482,17 +2482,13 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	if (IS_MNT_UNBINDABLE(old_mnt))
 		return ERR_PTR(-EINVAL);
 
-	if (mnt_has_parent(old_mnt)) {
+	if (!is_mounted(&old_mnt->mnt))
+		return ERR_PTR(-EINVAL);
+
+	if (mnt_has_parent(old_mnt) || !is_anon_ns(old_mnt->mnt_ns)) {
 		if (!check_mnt(old_mnt))
 			return ERR_PTR(-EINVAL);
 	} else {

fs/namespace.c struct vfsmount *clone_private_mount(const struct path *path) · line 2494

Quoted 10 lines
fs/namespace.c +0 -7
-		if (!is_mounted(&old_mnt->mnt))
-			return ERR_PTR(-EINVAL);
-
-		/* Make sure this isn't something purely kernel internal. */
-		if (!is_anon_ns(old_mnt->mnt_ns))
-			return ERR_PTR(-EINVAL);
-
 		/* Make sure we don't create mount namespace loops. */
 		if (!check_for_nsfs_mounts(old_mnt))
 			return ERR_PTR(-EINVAL);

Not the right way to do that. What we want is

Quoted 15 lines
	/* ours are always fine */
	if (!check_mnt(old_mnt)) {
		/* they'd better be mounted _somewhere */
		if (!is_mounted(old_mnt))
			return -EINVAL;
		/* no other real namespaces; only anon */
		if (!is_anon_ns(old_mnt->mnt_ns))
			return -EINVAL;
		/* ... and root of that anon */
		if (mnt_has_parent(old_mnt))
			return -EINVAL;
		/* Make sure we don't create mount namespace loops. */
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}
Quoted 4 lines

Hello Al Viro,

Thank you for your comment.
That code can solve my problem, and it seems to be better!

BTW, see https://lore.kernel.org/all/20250506194849.GT2023217@ZenIV/ for discussion about a week ago when that got noticed:

|| In case of clone_private_mount(), though, there's nothing wrong
|| with "clone me a subtree of absolute root", so it has to be
|| done other way round - check if it's ours first, then in "not
|| ours" case check that it's a root of anon namespace.
||
|| Failing btrfs mount has ended up with upper layer pathname
|| pointing to initramfs directory where btrfs would've been
|| mounted, which had walked into that corner case.  In your
|| case the problem has already happened by that point, but on
|| a setup a-la X Terminal it would cause trouble...

Looks like such setups are less theoretical than I thought.

Quoted 1 line

So, I will revise my patch and resend it.

Probably worth gathering the comments in one place. Something like

	/*
	 * Check if the source is acceptable; anything mounted in
	 * our namespace is fine, otherwise it must be the root of
	 * some anon namespace and we need to make sure no namespace
	 * loops get created.
	 */
	if (!check_mnt(old_mnt)) {
		if (!is_mounted(&old_mnt->mnt) ||
		    !is_anon_ns(old_mnt->mnt_ns) ||
		    mnt_has_parent(old_mnt))
			return ERR_PTR(-EINVAL);
		if (!check_for_nsfs_mounts(old_mnt))
			return ERR_PTR(-EINVAL);
	}
Quoted 1 line

might be easier to follow.

I think it's good idea.
I will send a patch with that comment.

Thanks