* Re: R: Why we need add :: at the last when encap seg6 with inline mode?
[not found] ` <AM6PR07MB56691CDB704895F0E13689AA9125A@AM6PR07MB5669.eurprd07.prod.outlook.com>
@ 2023-07-04 2:48 ` Hangbin Liu
0 siblings, 0 replies; 2+ messages in thread
From: Hangbin Liu @ 2023-07-04 2:48 UTC (permalink / raw)
To: paolo lungaroni
Cc: Andrea Mayer, netdev@vger.kernel.org, stefano.salsano,
Ahmed Abdelsalam, David Lebrun
Hi Paolo,
On Thu, Jun 29, 2023 at 10:57:32PM +0000, paolo lungaroni wrote:
> Just for the sake of clarity let's consider Example 1 where the SID List is
> built as follows: <S3,S2,S1>.
>
> Before applying H.Insert
> IPv6 DA=B2
>
> After applying H.Insert:
> IPv6 DA=S1 | SRH <B2,S3,S2,S1>
>
> In this representation, we stick with the draft conventions where the SID List
> is represented in the reversed order. That means that S1 will be the first SID,
> S2 the second one, etc. B2 is the original IPv6 DA that is pushed into the
> SID List and will be restored once the SRH is removed.
>
> Now let's consider the H.Insert implementation in Linux.
>
> User space
> =========
>
> Currently, when a H.Insert tunnel is instantiated the userspace (iproute2, in
> this case) acquires all the info (e.g. inline, segs, etc) from the CLI and
> builds directly the required SRH header [2].
> Considering the H.Insert tunnel (i.e. inline mode), parse_srh() (line 894)
> accommodates for an extra space required to store the original IPv6 DA (i.e.
> B2 considering the Example 1) which is not known at the moment of the tunnel
> creation.
> For this reason, iproute2 fills the reserved space with 16 bytes set to zero
> (and that's why we see ::).
Thanks for the explanation. Compared with dealing each segment route and
adding the original DA in the middle of struct ipv6_sr_hdr, this way looks
more easier. Although the users may feel a little confused :)
Regards
Hangbin
^ permalink raw reply [flat|nested] 2+ messages in thread