[RFC PATCH] vfs: add MNT_NOFOLLOW flag to umount(2)

[RFC PATCH] vfs: add MNT_NOFOLLOW flag to umount(2)

[Miklos Szeredi]

From: Miklos Szeredi <mszeredi@suse.cz>

Add a new MNT_NOFOLLOW flag to umount(2).  This is needed to prevent
symlink attacks in unprivileged unmounts (fuse, samba, ncpfs).

Additionally, return -EINVAL if an unknown flag is encountered.  This
makes it possible for the caller to determine if a flag is supported
or not (at least on kernels with this patch).

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/namespace.c     |    9 ++++++++-
 include/linux/fs.h |    1 +
 2 files changed, 9 insertions(+), 1 deletion(-)

Index: linux-2.6/fs/namespace.c
===================================================================
--- linux-2.6.orig/fs/namespace.c	2010-01-29 09:21:45.000000000 +0100
+++ linux-2.6/fs/namespace.c	2010-02-04 10:03:08.000000000 +0100
@@ -1121,8 +1121,15 @@ SYSCALL_DEFINE2(umount, char __user *, n
 {
 	struct path path;
 	int retval;
+	int lookup_flags = 0;
 
-	retval = user_path(name, &path);
+	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | MNT_NOFOLLOW))
+		return -EINVAL;
+
+	if (!(flags & MNT_NOFOLLOW))
+		lookup_flags |= LOOKUP_FOLLOW;
+
+	retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
 	if (retval)
 		goto out;
 	retval = -EINVAL;
Index: linux-2.6/include/linux/fs.h
===================================================================
--- linux-2.6.orig/include/linux/fs.h	2010-01-22 08:46:57.000000000 +0100
+++ linux-2.6/include/linux/fs.h	2010-02-04 09:57:39.000000000 +0100
@@ -1305,6 +1305,7 @@ extern int send_sigurg(struct fown_struc
 #define MNT_FORCE	0x00000001	/* Attempt to forcibily umount */
 #define MNT_DETACH	0x00000002	/* Just detach from the tree */
 #define MNT_EXPIRE	0x00000004	/* Mark for expiry */
+#define MNT_NOFOLLOW	0x00000008	/* Don't follow symlink on umount */
 
 extern struct list_head super_blocks;
 extern spinlock_t sb_lock;

Re: [RFC PATCH] vfs: add MNT_NOFOLLOW flag to umount(2)

[Bodo Eggert]

Miklos Szeredi <miklos@szeredi.hu> wrote:

> Additionally, return -EINVAL if an unknown flag is encountered.  This
> makes it possible for the caller to determine if a flag is supported
> or not (at least on kernels with this patch).

There should be a guaranteed-to-be-invalid flag or flag-combination in
order to safely detect this feature.

Re: [RFC PATCH] vfs: add MNT_NOFOLLOW flag to umount(2)

[Miklos Szeredi]

On Thu, 04 Feb 2010, Bodo Eggert wrote:
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
> > Additionally, return -EINVAL if an unknown flag is encountered.  This
> > makes it possible for the caller to determine if a flag is supported
> > or not (at least on kernels with this patch).
> 
> There should be a guaranteed-to-be-invalid flag or flag-combination in
> order to safely detect this feature.

It's difficult though, because the app would have to make sure the
detection itself would reliably fail, and with a different error.

Simply checking the kernel version might be easier.

Thanks,
Miklos

Re: [RFC PATCH] vfs: add MNT_NOFOLLOW flag to umount(2)

[Bodo Eggert]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 986 bytes --]

On Fri, 5 Feb 2010, Miklos Szeredi wrote:

> On Thu, 04 Feb 2010, Bodo Eggert wrote:
>> Miklos Szeredi <miklos@szeredi.hu> wrote:
>>
>>> Additionally, return -EINVAL if an unknown flag is encountered.  This
>>> makes it possible for the caller to determine if a flag is supported
>>> or not (at least on kernels with this patch).
>>
>> There should be a guaranteed-to-be-invalid flag or flag-combination in
>> order to safely detect this feature.
>
> It's difficult though, because the app would have to make sure the
> detection itself would reliably fail, and with a different error.
>
> Simply checking the kernel version might be easier.

Yes, that's exactly what I intend to make possible:

| ret = umount("/tmp/mkstmp", UMOUNT_IMPOSSIBLE_FLAG);
| if (ret != -1 || errno != -EINVAL)
|     goto extended_mount_flags_are_not_supported;

Checking the kernel version is a bad idea, because the application might 
be portable, or features might be backported. ¢¢