Re: kern/53664 (nfs client broken on netbsd-8)

[Maxime Villard <max%m00nbsd.net@localhost>]

Le 13/10/2018 à 13:15, Michael van Elst a écrit :
> The following reply was made to PR kern/53664; it has been noted by GNATS.
>
> From: mlelstv%serpens.de@localhost (Michael van Elst)
> To: gnats-bugs%netbsd.org@localhost
> Cc:
> Subject: Re: kern/53664 (nfs client broken on netbsd-8)
> Date: Sat, 13 Oct 2018 11:14:28 -0000 (UTC)
>
>   martin%duskware.de@localhost (Martin Husemann) writes:
>
>   > Just to make sure: default nfs mounts changed from UDP to TCP and I have
>   > not seen any NFS fallout with any -current version in the meantime,
>   > even with r1.20. Is the fallout specific to UDP maybe, or specific
>   > network devices or server settings?
>
>   It's specific to IP reassembly, and that's very rarely used with TCP
>   but for NFS almost always with UDP.

No, the fallout is when someone reads ip_off on a reassembled packet, which
very few places do in practice. Fragmented UDP worked correctly, that's how
I did basic testing. So did TCP, etc.

Probably htodd's configuration reads ip_off somewhere in a way that matters,
and in all cases, it was a mistake (from the code before me, and also from me)
not to explicitly reset ip_off to zero.