netdev - Re: how is the bpfilter sockopt processing supposed to work

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAADnVQJoMC=vfS4yb7gYZF4fmwrHd+gdOf9zmPm2XyK1jfosHg@mail.gmail.com>
Date:   Fri, 17 Jul 2020 10:28:47 -0700
From:   Alexei Starovoitov <alexei.starovoitov@...il.com>
To:     Christoph Hellwig <hch@....de>
Cc:     Stanislav Fomichev <sdf@...gle.com>,
        Alexei Starovoitov <ast@...nel.org>,
        "David S. Miller" <davem@...emloft.net>,
        Network Development <netdev@...r.kernel.org>,
        bpf <bpf@...r.kernel.org>
Subject: Re: how is the bpfilter sockopt processing supposed to work

On Fri, Jul 17, 2020 at 9:25 AM Christoph Hellwig <hch@....de> wrote:
>
> On Fri, Jul 17, 2020 at 09:13:07AM -0700, Alexei Starovoitov wrote:
> > On Thu, Jul 16, 2020 at 10:52 PM Christoph Hellwig <hch@....de> wrote:
> > >
> > > Hi Alexei,
> > >
> > > I've just been auditing the sockopt code, and bpfilter looks really
> > > odd.  Both getsockopts and setsockopt eventually end up
> > > in__bpfilter_process_sockopt, which then passes record to the
> > > userspace helper containing the address of the optval buffer.
> > > Which depending on bpf-cgroup might be in user or kernel space.
> > > But even if it is in userspace it would be in a different process
> > > than the bpfiler helper.  What makes all this work?
> >
> > Hmm. Good point. bpfilter assumes user addresses. It will break
> > if bpf cgroup sockopt messes with it.
> > We had a different issue with bpf-cgroup-sockopt and iptables in the past.
> > Probably the easiest way forward is to special case this particular one.
> > With your new series is there a way to tell in bpfilter_ip_get_sockopt()
> > whether addr is kernel or user? And if it's the kernel just return with error.
>
> Yes, I can send a fix.  But how do even the user space addressed work?
> If some random process calls getsockopt or setsockopt, how does the
> bpfilter user mode helper attach to its address space?

The actual bpfilter processing is in two patches that we didn't land:
https://lore.kernel.org/patchwork/patch/902785/
https://lore.kernel.org/patchwork/patch/902783/
UMD is using process_vm_readv().
The target process is waiting for the sockopt syscall to return,
so from the toctou perspective it's the same as the kernel doing copy_from_user.