netdev - Re: [PATCH] openvswitch: Trim off padding before L3 conntrack processing

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:   Sun, 17 Dec 2017 11:22:45 -0800
From:   Pravin Shelar <pshelar@....org>
To:     Ed Swierk <eswierk@...portsystems.com>
Cc:     ovs-dev <ovs-dev@...nvswitch.org>,
        Linux Kernel Network Developers <netdev@...r.kernel.org>,
        Lance Richardson <lrichard@...hat.com>,
        Benjamin Warren <ben@...portsystems.com>,
        Keith Holleman <holleman@...portsystems.com>
Subject: Re: [PATCH] openvswitch: Trim off padding before L3 conntrack processing

On Thu, Dec 14, 2017 at 12:05 PM, Ed Swierk <eswierk@...portsystems.com> wrote:
> On Wed, Dec 13, 2017 at 4:58 PM, Pravin Shelar <pshelar@....org> wrote:
>> On Tue, Dec 12, 2017 at 8:17 AM, Ed Swierk <eswierk@...portsystems.com> wrote:
>>> A short IPv4 packet may have up to 6 bytes of padding following the IP
>>> payload when received on an Ethernet device.
>>>
>>> In the normal IPv4 receive path, ip_rcv() trims the packet to
>>> ip_hdr->tot_len before invoking NF_INET_PRE_ROUTING hooks (including
>>> conntrack). Then any subsequent L3+ processing steps, like
>>> nf_checksum(), use skb->len as the length of the packet, rather than
>>> referring back to ip_hdr->tot_len. In the IPv6 receive path, ip6_rcv()
>>> does the same using ipv6_hdr->payload_len.
>>>
>>> In the OVS conntrack receive path, this trimming does not occur, so
>>> the checksum verification in tcp_header() fails, printing "nf_ct_tcp:
>>> bad TCP checksum". Extra zero bytes don't affect the checksum, but the
>>> length in the IP pseudoheader does. That length is based on skb->len,
>>> and without trimming, it doesn't match the length the sender used when
>>> computing the checksum.
>>>
>>> With this change, OVS conntrack trims IPv4 and IPv6 packets prior to
>>> L3 processing.
>>>
>>> Signed-off-by: Ed Swierk <eswierk@...portsystems.com>
>>> ---
>>>  net/openvswitch/conntrack.c | 17 +++++++++++++++++
>>>  1 file changed, 17 insertions(+)
>>>
>>> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
>>> index d558e882ca0c..3a7c9215c431 100644
>>> --- a/net/openvswitch/conntrack.c
>>> +++ b/net/openvswitch/conntrack.c
>>> @@ -1105,12 +1105,29 @@ int ovs_ct_execute(struct net *net, struct sk_buff *skb,
>>>                    const struct ovs_conntrack_info *info)
>>>  {
>>>         int nh_ofs;
>>> +       unsigned int nh_len;
>>>         int err;
>>>
>>>         /* The conntrack module expects to be working at L3. */
>>>         nh_ofs = skb_network_offset(skb);
>>>         skb_pull_rcsum(skb, nh_ofs);
>>>
>>> +       /* Trim to L3 length since nf_checksum() doesn't expect padding. */
>> Can you explore if nf_checksum can be changed to avoid the padding?
>
> The nf_ip_checksum() and nf_ip6_checksum() helper functions can easily
> be changed to avoid the padding.
>
> My worry is that conntrack is just one of many netfilter hooks that
> perform L3+ processing, and may assume that once skb->data points to
> the L3 header, skb->len reflects the length of the L3 header and
> payload. For example, in nf_conntrack_ftp.c, help() uses skb->len to
> determine the length of the FTP payload and the TCP sequence number of
> the next packet; this would be thrown off by lower-layer padding.
>
> br_netfilter, a cousin of OVS, has always preserved this
> assumption--like ip_rcv() and ip6_rcv(), br_validate_ipv4() and
> br_validate_ipv6() trim the skb to the L3 length before they invoke
> NF_INET_PRE_ROUTING hooks. Modifying OVS to fit the mold seems more
> straightforward than changing this assumption.
>
we could avoid extra processing in fast path, thats why I wanted to
explore this, But if it is too complex, I am fine with this patch.

>>> +       switch (skb->protocol) {
>>> +       case htons(ETH_P_IP):
>>> +               nh_len = ntohs(ip_hdr(skb)->tot_len);
>>> +               break;
>>> +       case htons(ETH_P_IPV6):
>>> +               nh_len = ntohs(ipv6_hdr(skb)->payload_len)
>>> +                       + sizeof(struct ipv6hdr);
>>> +               break;
>>> +       default:
>>> +               nh_len = skb->len;
>>> +       }
>>> +       err = pskb_trim_rcsum(skb, nh_len);
>>> +       if (err)
>>> +               return err;
>>> +
>> In case of error skb needs to be freed.
>
> Thanks, I will fix this.
>
> --Ed