Subject: Re: NFS-mounted (TCP) root partitions and "halt" --> no FIN_ACK
To: Sean Doran <firstname.lastname@example.org>
From: Perry E. Metzger <email@example.com>
Date: 10/28/1998 15:06:11
Sean Doran writes:
> Perry Metzger wrote:
> | That's a rather hard problem. At the time where "doshutdownhooks"
> | gets called in most ports, interrupts have already been stopped and
> | you're past the point where you could reasonably handle that
> | cleanly. Before that, of course, you need to finish your last sync...
> Oki, fair enough. How about a cheat by not having NFS_BOOT_TCP choose
> port 1023 every time, so you aren't penalized by 2*MSL when you reboot?
Do you mean *not* use port 1023? Because if you reuse the same port,
you are supposed to wait 2*MSL to allow all old segements to clear,
but not if you use a different port pair.
> (Well, statistically less likely to be penalized, etc...)
BTW, maybe we can just have the remote host send a keepalive probe if
a second connection from the first host appears, to see if it gets a