Re: tun oops dereferencing garbage nsproxy-> address.

[ebiederm@xmission.com (Eric W. Biederman)]

Dave Jones <davej@redhat.com> writes:

> On Tue, Mar 13, 2012 at 11:19:40AM -0700, Eric W. Biederman wrote:
>
>  > > oops happened here..
>  > >
>  > >         tfile->net = get_net(current->nsproxy->net_ns);
>  > >      548:       48 8b 92 50 05 00 00    mov    0x550(%rdx),%rdx
>  > >      54f:       48 8b 52 28             mov    0x28(%rdx),%rdx
>  > >
>  > > My guess is the fuzzer called some syscall that set current->nsproxy
>  > > to garbage (0x0000000100000001), which later got dereferenced when it
>  > > subsequently randomly did an open() on tun.
>  > >
>  > > Any thoughts ?
>  > 
>  > It smells like a memory stomp.  current->nsproxy is always supposed to
>  > have a valid value, and it never would have an odd value.  The value
>  > should always be at least 8 byte aligned.
>  > 
>  > Since the value is impossible this doesn't feel like a path where the
>  > error handling is wrong.
>
> 0x0000000100000001 looks like one of strange values my fuzzer passes syscalls
> when they ask for an address.
>
> So something managed to get that set as nsproxy.  The fuzzer avoids calling
> clone(), so are there other syscalls that might set this ?

setns and unshare might touch the nsproxy for the same reasons as clone,
but the rules are very similar to clone.

>  > So I am guessing this is a memory stomp.  My guess it would take the
>  > same sequence of system calls on the same kernel build to reproduce
>  > this problem in the same place.
>  > 
>  > Do you have any more information?
>
> I've not managed to reproduce it, and that run sadly had logging turned off,
> so I don't have the exact syscall sequence that caused it.

Grr.  All of the interesting failures seem to happen with logging turned off.

Eric