netdev - Re: [Patch net] ip_gre: set dev->hard_header

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]

Date:   Thu, 8 Oct 2020 15:04:01 -0400
From:   Willem de Bruijn <willemdebruijn.kernel@...il.com>
To:     Cong Wang <xiyou.wangcong@...il.com>
Cc:     Network Development <netdev@...r.kernel.org>,
        syzbot <syzbot+4a2c52677a8a1aa283cb@...kaller.appspotmail.com>,
        William Tu <u9012063@...il.com>, Xie He <xie.he.0141@...il.com>
Subject: Re: [Patch net] ip_gre: set dev->hard_header_len properly

On Thu, Oct 8, 2020 at 1:34 PM Cong Wang <xiyou.wangcong@...il.com> wrote:
>
> On Thu, Oct 8, 2020 at 4:49 AM Willem de Bruijn
> <willemdebruijn.kernel@...il.com> wrote:
> >
> > On Wed, Oct 7, 2020 at 9:22 PM Cong Wang <xiyou.wangcong@...il.com> wrote:
> > >
> > > GRE tunnel has its own header_ops, ipgre_header_ops, and sets it
> > > conditionally. When it is set, it assumes the outer IP header is
> > > already created before ipgre_xmit().
> > >
> > > This is not true when we send packets through a raw packet socket,
> > > where L2 headers are supposed to be constructed by user. Packet
> > > socket calls dev_validate_header() to validate the header. But
> > > GRE tunnel does not set dev->hard_header_len, so that check can
> > > be simply bypassed, therefore uninit memory could be passed down
> > > to ipgre_xmit().
> >
> > If dev->hard_header_len is zero, the packet socket will not reserve
> > room for the link layer header, so skb->data points to network_header.
> > But I don't see any uninitialized packet data?
>
> The uninit data is allocated by packet_alloc_skb(), if dev->hard_header_len
> is 0 and 'len' is anything between [0, tunnel->hlen + sizeof(struct iphdr)),
> dev_validate_header() still returns true obviously but only 'len'
> bytes are copied
> from user-space by skb_copy_datagram_from_iter(). Therefore, those bytes
> within range (len, tunnel->hlen + sizeof(struct iphdr)] are uninitialized.

With dev->hard_header_len of zero, packet_alloc_skb() only allocates len bytes.

With SOCK_RAW, the writer is expected to write the ip and gre header
and include these in the send len argument. The only difference I see
is that with hard_header_len the data starts reserve bytes before
skb_network_header, and an additional tail has been allocated that is
not used.

But this also fixes a potentially more serious bug. With SOCK_DGRAM,
dev_hard_header/ipgre_header just assumes that there is enough room in
the packet to skb_push(skb, t->hlen + sizeof(*iph)). Which may be
false if this header length had not been reserved.

Though I've mainly looked at packet_snd. Perhaps you are referring to
tpacket_snd?