[PATCH] userns: allow privileged user to operate locked mount

[PATCH] userns: allow privileged user to operate locked mount

[Gao feng]

Privileged user should have rights to mount/umount/move
these even locked mount.

Signed-off-by: Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>
---
 fs/namespace.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index da5c494..7097fc7 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1297,6 +1297,11 @@ static inline bool may_mount(void)
 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
 }
 
+static inline bool may_mount_lock(struct mount *mnt)
+{
+	return !(mnt->mnt.mnt_flags & MNT_LOCKED) || capable(CAP_SYS_ADMIN);
+}
+
 /*
  * Now umount can handle mount points as well as block devices.
  * This is important for filesystems which use unnamed block devices.
@@ -1330,7 +1335,7 @@ SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
 		goto dput_and_out;
 	if (!check_mnt(mnt))
 		goto dput_and_out;
-	if (mnt->mnt.mnt_flags & MNT_LOCKED)
+	if (!may_mount_lock(mnt))
 		goto dput_and_out;
 
 	retval = do_umount(mnt, flags);
@@ -1768,7 +1773,8 @@ static int do_loopback(struct path *path, const char *old_name,
 	if (!check_mnt(parent) || !check_mnt(old))
 		goto out2;
 
-	if (!recurse && has_locked_children(old, old_path.dentry))
+	if (!recurse && has_locked_children(old, old_path.dentry) &&
+	    !capable(CAP_SYS_ADMIN))
 		goto out2;
 
 	if (recurse)
@@ -1895,7 +1901,7 @@ static int do_move_mount(struct path *path, const char *old_name)
 	if (!check_mnt(p) || !check_mnt(old))
 		goto out1;
 
-	if (old->mnt.mnt_flags & MNT_LOCKED)
+	if (!may_mount_lock(old))
 		goto out1;
 
 	err = -EINVAL;
@@ -2679,7 +2685,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out4;
 	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
 		goto out4;
-	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
+	if (!may_mount_lock(new_mnt))
 		goto out4;
 	error = -ENOENT;
 	if (d_unlinked(new.dentry))
-- 
1.8.3.1

Re: [PATCH] userns: allow privileged user to operate locked mount

[Eric W. Biederman]

Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org> writes:

> Privileged user should have rights to mount/umount/move
> these even locked mount.

Hmm. This is pretty much a can't happen case, as the only exist in mount
namespaces where the global root isn't the root.  How are you getting
into this situation?  Using setns() ?

Why would we even care?

As implemented this patch does not handle nested user namespaces and
that really worries me at a semantic level.

We don't want to design cases where we can create containers in
containers.

Eric


> Signed-off-by: Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>
> ---
>  fs/namespace.c | 14 ++++++++++----
>  1 file changed, 10 insertions(+), 4 deletions(-)
>
> diff --git a/fs/namespace.c b/fs/namespace.c
> index da5c494..7097fc7 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -1297,6 +1297,11 @@ static inline bool may_mount(void)
>  	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
>  }
>  
> +static inline bool may_mount_lock(struct mount *mnt)
> +{
> +	return !(mnt->mnt.mnt_flags & MNT_LOCKED) || capable(CAP_SYS_ADMIN);
> +}
> +
>  /*
>   * Now umount can handle mount points as well as block devices.
>   * This is important for filesystems which use unnamed block devices.
> @@ -1330,7 +1335,7 @@ SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
>  		goto dput_and_out;
>  	if (!check_mnt(mnt))
>  		goto dput_and_out;
> -	if (mnt->mnt.mnt_flags & MNT_LOCKED)
> +	if (!may_mount_lock(mnt))
>  		goto dput_and_out;
>  
>  	retval = do_umount(mnt, flags);
> @@ -1768,7 +1773,8 @@ static int do_loopback(struct path *path, const char *old_name,
>  	if (!check_mnt(parent) || !check_mnt(old))
>  		goto out2;
>  
> -	if (!recurse && has_locked_children(old, old_path.dentry))
> +	if (!recurse && has_locked_children(old, old_path.dentry) &&
> +	    !capable(CAP_SYS_ADMIN))
>  		goto out2;
>  
>  	if (recurse)
> @@ -1895,7 +1901,7 @@ static int do_move_mount(struct path *path, const char *old_name)
>  	if (!check_mnt(p) || !check_mnt(old))
>  		goto out1;
>  
> -	if (old->mnt.mnt_flags & MNT_LOCKED)
> +	if (!may_mount_lock(old))
>  		goto out1;
>  
>  	err = -EINVAL;
> @@ -2679,7 +2685,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>  		goto out4;
>  	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
>  		goto out4;
> -	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
> +	if (!may_mount_lock(new_mnt))
>  		goto out4;
>  	error = -ENOENT;
>  	if (d_unlinked(new.dentry))

Re: [PATCH] userns: allow privileged user to operate locked mount

[Gao feng]

On 11/15/2013 07:50 AM, Eric W. Biederman wrote:
> Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org> writes:
> 
>> Privileged user should have rights to mount/umount/move
>> these even locked mount.
> 
> Hmm. This is pretty much a can't happen case, as the only exist in mount
> namespaces where the global root isn't the root.  How are you getting
> into this situation?  Using setns() ?
> 

Before, priviged user can use setns to set his mount namespace to the
container's mount namespace, and change container's mount directly.
this patch just gives back host the control of container.

> Why would we even care?
> 
> As implemented this patch does not handle nested user namespaces and
> that really worries me at a semantic level.
> 
> We don't want to design cases where we can create containers in
> containers.
> 
> Eric
> 
> 
>> Signed-off-by: Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>
>> ---
>>  fs/namespace.c | 14 ++++++++++----
>>  1 file changed, 10 insertions(+), 4 deletions(-)
>>
>> diff --git a/fs/namespace.c b/fs/namespace.c
>> index da5c494..7097fc7 100644
>> --- a/fs/namespace.c
>> +++ b/fs/namespace.c
>> @@ -1297,6 +1297,11 @@ static inline bool may_mount(void)
>>  	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
>>  }
>>  
>> +static inline bool may_mount_lock(struct mount *mnt)
>> +{
>> +	return !(mnt->mnt.mnt_flags & MNT_LOCKED) || capable(CAP_SYS_ADMIN);
>> +}
>> +
>>  /*
>>   * Now umount can handle mount points as well as block devices.
>>   * This is important for filesystems which use unnamed block devices.
>> @@ -1330,7 +1335,7 @@ SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
>>  		goto dput_and_out;
>>  	if (!check_mnt(mnt))
>>  		goto dput_and_out;
>> -	if (mnt->mnt.mnt_flags & MNT_LOCKED)
>> +	if (!may_mount_lock(mnt))
>>  		goto dput_and_out;
>>  
>>  	retval = do_umount(mnt, flags);
>> @@ -1768,7 +1773,8 @@ static int do_loopback(struct path *path, const char *old_name,
>>  	if (!check_mnt(parent) || !check_mnt(old))
>>  		goto out2;
>>  
>> -	if (!recurse && has_locked_children(old, old_path.dentry))
>> +	if (!recurse && has_locked_children(old, old_path.dentry) &&
>> +	    !capable(CAP_SYS_ADMIN))
>>  		goto out2;
>>  
>>  	if (recurse)
>> @@ -1895,7 +1901,7 @@ static int do_move_mount(struct path *path, const char *old_name)
>>  	if (!check_mnt(p) || !check_mnt(old))
>>  		goto out1;
>>  
>> -	if (old->mnt.mnt_flags & MNT_LOCKED)
>> +	if (!may_mount_lock(old))
>>  		goto out1;
>>  
>>  	err = -EINVAL;
>> @@ -2679,7 +2685,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>>  		goto out4;
>>  	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
>>  		goto out4;
>> -	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
>> +	if (!may_mount_lock(new_mnt))
>>  		goto out4;
>>  	error = -ENOENT;
>>  	if (d_unlinked(new.dentry))
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

Re: [PATCH] userns: allow privileged user to operate locked mount

[Eric W. Biederman]

Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org> writes:

> On 11/15/2013 07:50 AM, Eric W. Biederman wrote:
>> Gao feng <gaofeng-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org> writes:
>> 
>>> Privileged user should have rights to mount/umount/move
>>> these even locked mount.
>> 
>> Hmm. This is pretty much a can't happen case, as the only exist in mount
>> namespaces where the global root isn't the root.  How are you getting
>> into this situation?  Using setns() ?
>> 
>
> Before, priviged user can use setns to set his mount namespace to the
> container's mount namespace, and change container's mount directly.
> this patch just gives back host the control of container.

Having thought about this patch a little more I really don't like it.

There are other ways for a privileged user to get around the limitations
when the mount namespace is being created or the mounts are being
propagated.

This approach would require more then a signgle bit of accounting to
work in the nested user namespace case.

The lock says one or several mounts are mounted as a unit and need to
stay that way.

If there are real advantages to splitting things up I might be persuaded
to change my mind.  But right now it looks like you are introducing
extra complexity for a very corner edge case that we don't want to
encourage people to use.

Eric