Re: [PATCH] selinux: read and write sid under lock

From: Paul Moore
Date: Mon Mar 10 2025 - 16:20:37 EST


On Mon, Mar 10, 2025 at 3:53 PM Stephen Smalley
<stephen.smalley.work@xxxxxxxxx> wrote:
> On Sat, Mar 8, 2025 at 11:55 PM Edward Adam Davis <eadavis@xxxxxx> wrote:
> >
> > syzbot reported a data-race in selinux_socket_post_create /
> > selinux_socket_sock_rcv_skb. [1]
> >
> > When creating the socket path and receiving the network data packet path,
> > effective data access protection is not performed when reading and writing
> > the sid, resulting in a race condition.
> >
> > Add a lock to synchronize the two.

...

> > Reported-by: syzbot+00c633585760c05507c3@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Closes: https://syzkaller.appspot.com/bug?extid=00c633585760c05507c3
> > Signed-off-by: Edward Adam Davis <eadavis@xxxxxx>
> > ---
> > security/selinux/hooks.c | 7 ++++++-
> > 1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> > index 7b867dfec88b..ea5d0273f9d5 100644
> > --- a/security/selinux/hooks.c
> > +++ b/security/selinux/hooks.c
> > @@ -4677,8 +4677,10 @@ static int selinux_socket_post_create(struct socket *sock, int family,
> >
> > if (sock->sk) {
> > sksec = selinux_sock(sock->sk);
> > + spin_lock(&sksec->lock);
>
> You didn't include the diff that adds this lock field to
> sk_security_struct, but aside from that, I would suggest something
> lighter-weight like READ_ONCE/WRITE_ONCE if possible.

Yes, please don't add a spinlock to something that is potentially
going to be hit on every packet entering the system.

--
paul-moore.com