From mboxrd@z Thu Jan  1 00:00:00 1970
From: Hyong Youb Kim <hyonkim@cisco.com>
Subject: Re: [dpdk-users] IPV4/IPV6 TCP/UDP Pseudo Header
 Checksum APIs
Date: Thu, 29 Nov 2018 00:07:19 +0900
Message-ID: <20181128150718.GA25832@HYONKIM-7R0DR.cisco.com>
References: <A69E93D1CF6DFA4BA33FB8D8DAFFFD19035A0280@dggemm512-mbx.china.huawei.com>
 <CAGSp03nOTeCO8xdJ8wbhi+Z2_6kMAyhYYNQvV9BaFS=txFn4dg@mail.gmail.com>
 <CAGSp03ncwhn4ZxHtqnMoZEVJYQjc27bcWwVBvhwstpzyJWa67w@mail.gmail.com>
 <A69E93D1CF6DFA4BA33FB8D8DAFFFD19035A0973@dggemm512-mbx.china.huawei.com>
 <CAGSp03mCDamL2a=fSq=1rRd+_tdPLbsnY+2CsA6aYpd5vU3RwQ@mail.gmail.com>
 <bf22fb5e-cefd-c7f1-07aa-a059c98d0692@intel.com>
 <20181023090158.z5w3gtvdzax247w6@platinum>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Ferruh Yigit <ferruh.yigit@intel.com>,
 Shyam Shrivastav <shrivastav.shyam@gmail.com>, lidejun1@huawei.com,
 users <users@dpdk.org>, dev@dpdk.org, lichunhe@huawei.com,
 zhangxufeng4@huawei.com
To: Olivier Matz <olivier.matz@6wind.com>
Return-path: <dev-bounces@dpdk.org>
Content-Disposition: inline
In-Reply-To: <20181023090158.z5w3gtvdzax247w6@platinum>
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <https://mails.dpdk.org/options/dev>,
 <mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://mails.dpdk.org/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <https://mails.dpdk.org/listinfo/dev>,
 <mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>

On Tue, Oct 23, 2018 at 11:01:58AM +0200, Olivier Matz wrote:
> Hi,
> 
> You are right, the current code does not take IP or IPv6 options
> in account. I think this should be considered as a bug.
> 
> The fix for IPv4 is not complicated, I did a quick draft here:
> http://git.droids-corp.org/?p=dpdk.git;a=commitdiff;h=96a6978ef6814e1450e1bd65fbce91c3d85b3121
> 
> For IPv6, it is more complex than expected:
> https://tools.ietf.org/html/rfc2460.html#section-8.1
> 
> - we need to skip extension headers
> - we need to parse routing headers and use the proper destination
>   address in the pseudo header checksum
> 
> This makes me think that the API is not adequate. Asking the user
> to provide the headers in a contiguous memory without specifying
> the length is quite dangerous, especially if the header comes from
> outside, as it can trigger out of bound accesses.
> 
> I wonder if we shouldn't switch to a mbuf based API instead, and
> deprecate the old one.
> 
> Thoughts?
> Olivier
> 

I have been looking into a similar issue because
rte_net_intel_cksum_prepare(), which is used by most tx_pkt_prepare
handlers, does not work when ipv6 extensions are present. That
function is using rte_ipv6_phdr_cksum(). And this makes
rte_eth_tx_prepare() kinda useless for any workloads that encounter
ipv6 extensions.

There are 2 routing header types now (2 and 3).

https://www.iana.org/assignments/ipv6-parameters/ipv6-parameters.xhtml#ipv6-parameters-3

In addition to these routing headers, there is also ipv6
mobility. Pseudo header's source address is supposed to be the address
in the Home Address option.

https://tools.ietf.org/html/rfc6275#page-36

Who knows there may be future extensions that affect pseudo
header.. We can probably make rte_ipv6_phdr_cksum() to understand all
existing headers that affect pseudo header, but it will still not be future
proof. Should at least document the limitations for rte_ipv6_phdr_cksum()..

-Hyong