2022-09-29 11:09:37

by Mickaël Salaün

[permalink] [raw]
Subject: [PATCH v1] ksmbd: Fix user namespace mapping

A kernel daemon should not rely on the current thread, which is unknown
and might be malicious. Before this security fix,
ksmbd_override_fsids() didn't correctly override FS UID/GID which means
that arbitrary user space threads could trick the kernel to impersonate
arbitrary users or groups for file system access checks, leading to
file system access bypass.

This was found while investigating truncate support for Landlock:
https://lore.kernel.org/r/CAKYAXd8fpMJ7guizOjHgxEyyjoUwPsx3jLOPZP=wPYcbhkVXqA@mail.gmail.com

Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
Cc: Hyunchul Lee <[email protected]>
Cc: Namjae Jeon <[email protected]>
Cc: Steve French <[email protected]>
Cc: [email protected]
Signed-off-by: Mickaël Salaün <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
fs/ksmbd/smb_common.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/ksmbd/smb_common.c b/fs/ksmbd/smb_common.c
index 7f8ab14fb8ec..d96da872d70a 100644
--- a/fs/ksmbd/smb_common.c
+++ b/fs/ksmbd/smb_common.c
@@ -4,6 +4,8 @@
* Copyright (C) 2018 Namjae Jeon <[email protected]>
*/

+#include <linux/user_namespace.h>
+
#include "smb_common.h"
#include "server.h"
#include "misc.h"
@@ -625,8 +627,8 @@ int ksmbd_override_fsids(struct ksmbd_work *work)
if (!cred)
return -ENOMEM;

- cred->fsuid = make_kuid(current_user_ns(), uid);
- cred->fsgid = make_kgid(current_user_ns(), gid);
+ cred->fsuid = make_kuid(&init_user_ns, uid);
+ cred->fsgid = make_kgid(&init_user_ns, gid);

gi = groups_alloc(0);
if (!gi) {

base-commit: f76349cf41451c5c42a99f18a9163377e4b364ff
--
2.37.2


2022-09-29 12:13:59

by Christian Brauner

[permalink] [raw]
Subject: Re: [PATCH v1] ksmbd: Fix user namespace mapping

On Thu, Sep 29, 2022 at 12:04:47PM +0200, Mickaël Salaün wrote:
> A kernel daemon should not rely on the current thread, which is unknown
> and might be malicious. Before this security fix,
> ksmbd_override_fsids() didn't correctly override FS UID/GID which means
> that arbitrary user space threads could trick the kernel to impersonate
> arbitrary users or groups for file system access checks, leading to
> file system access bypass.
>
> This was found while investigating truncate support for Landlock:
> https://lore.kernel.org/r/CAKYAXd8fpMJ7guizOjHgxEyyjoUwPsx3jLOPZP=wPYcbhkVXqA@mail.gmail.com
>
> Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
> Cc: Hyunchul Lee <[email protected]>
> Cc: Namjae Jeon <[email protected]>
> Cc: Steve French <[email protected]>
> Cc: [email protected]
> Signed-off-by: Mickaël Salaün <[email protected]>
> Link: https://lore.kernel.org/r/[email protected]
> ---

I think this is ok. The alternative would probably be to somehow use a
relevant userns when struct ksmbd_user is created when the session is
established. But these are deeper ksmbd design questions. The fix
proposed here itself seems good.

2022-09-29 12:53:22

by Namjae Jeon

[permalink] [raw]
Subject: Re: [PATCH v1] ksmbd: Fix user namespace mapping

2022-09-29 19:04 GMT+09:00, Mickaël Salaün <[email protected]>:
> A kernel daemon should not rely on the current thread, which is unknown
> and might be malicious. Before this security fix,
> ksmbd_override_fsids() didn't correctly override FS UID/GID which means
> that arbitrary user space threads could trick the kernel to impersonate
> arbitrary users or groups for file system access checks, leading to
> file system access bypass.
>
> This was found while investigating truncate support for Landlock:
> https://lore.kernel.org/r/CAKYAXd8fpMJ7guizOjHgxEyyjoUwPsx3jLOPZP=wPYcbhkVXqA@mail.gmail.com
>
> Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
> Cc: Hyunchul Lee <[email protected]>
> Cc: Namjae Jeon <[email protected]>
> Cc: Steve French <[email protected]>
> Cc: [email protected]
> Signed-off-by: Mickaël Salaün <[email protected]>
> Link: https://lore.kernel.org/r/[email protected]
Acked-by: Namjae Jeon <[email protected]>

Thanks!

2022-09-29 13:14:57

by Mickaël Salaün

[permalink] [raw]
Subject: Re: [PATCH v1] ksmbd: Fix user namespace mapping


On 29/09/2022 13:37, Christian Brauner wrote:
> On Thu, Sep 29, 2022 at 12:04:47PM +0200, Mickaël Salaün wrote:
>> A kernel daemon should not rely on the current thread, which is unknown
>> and might be malicious. Before this security fix,
>> ksmbd_override_fsids() didn't correctly override FS UID/GID which means
>> that arbitrary user space threads could trick the kernel to impersonate
>> arbitrary users or groups for file system access checks, leading to
>> file system access bypass.
>>
>> This was found while investigating truncate support for Landlock:
>> https://lore.kernel.org/r/CAKYAXd8fpMJ7guizOjHgxEyyjoUwPsx3jLOPZP=wPYcbhkVXqA@mail.gmail.com
>>
>> Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
>> Cc: Hyunchul Lee <[email protected]>
>> Cc: Namjae Jeon <[email protected]>
>> Cc: Steve French <[email protected]>
>> Cc: [email protected]
>> Signed-off-by: Mickaël Salaün <[email protected]>
>> Link: https://lore.kernel.org/r/[email protected]
>> ---
>
> I think this is ok. The alternative would probably be to somehow use a
> relevant userns when struct ksmbd_user is created when the session is
> established. But these are deeper ksmbd design questions. The fix
> proposed here itself seems good.

That would be better indeed. I guess ksmbd works whenever the netlink
peer is not in a user namespace with mapped UID/GID, but it should
result in obvious access bugs otherwise (which is already the case
anyway). It seems that the netlink peer must be trusted because it is
the source of truth for account/user mapping anyway. This change fixes
the more critical side of the issue and it should fit well for backports.

2022-09-29 13:25:15

by Christian Brauner

[permalink] [raw]
Subject: Re: [PATCH v1] ksmbd: Fix user namespace mapping

On Thu, Sep 29, 2022 at 02:18:43PM +0200, Mickaël Salaün wrote:
>
> On 29/09/2022 13:37, Christian Brauner wrote:
> > On Thu, Sep 29, 2022 at 12:04:47PM +0200, Mickaël Salaün wrote:
> > > A kernel daemon should not rely on the current thread, which is unknown
> > > and might be malicious. Before this security fix,
> > > ksmbd_override_fsids() didn't correctly override FS UID/GID which means
> > > that arbitrary user space threads could trick the kernel to impersonate
> > > arbitrary users or groups for file system access checks, leading to
> > > file system access bypass.
> > >
> > > This was found while investigating truncate support for Landlock:
> > > https://lore.kernel.org/r/CAKYAXd8fpMJ7guizOjHgxEyyjoUwPsx3jLOPZP=wPYcbhkVXqA@mail.gmail.com
> > >
> > > Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
> > > Cc: Hyunchul Lee <[email protected]>
> > > Cc: Namjae Jeon <[email protected]>
> > > Cc: Steve French <[email protected]>
> > > Cc: [email protected]
> > > Signed-off-by: Mickaël Salaün <[email protected]>
> > > Link: https://lore.kernel.org/r/[email protected]
> > > ---
> >
> > I think this is ok. The alternative would probably be to somehow use a
> > relevant userns when struct ksmbd_user is created when the session is
> > established. But these are deeper ksmbd design questions. The fix
> > proposed here itself seems good.
>
> That would be better indeed. I guess ksmbd works whenever the netlink peer
> is not in a user namespace with mapped UID/GID, but it should result in
> obvious access bugs otherwise (which is already the case anyway). It seems
> that the netlink peer must be trusted because it is the source of truth for
> account/user mapping anyway. This change fixes the more critical side of the
> issue and it should fit well for backports.

Sorry, I also forgot,
Acked-by: Christian Brauner (Microsoft) <[email protected]>