Date: Mon, 17 Sep 2007 01:50:16 -0700 From: "Kip Macy" <kip.macy@gmail.com> To: "Andre Oppermann" <andre@freebsd.org> Cc: Perforce Change Reviews <perforce@freebsd.org>, Kip Macy <kmacy@freebsd.org> Subject: Re: PERFORCE change 126488 for review Message-ID: <b1fa29170709170150y3b391a22x1a52f4858b6882ca@mail.gmail.com> In-Reply-To: <46EE30A8.7020809@freebsd.org> References: <200709162157.l8GLvpx2080125@repoman.freebsd.org> <46EE30A8.7020809@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 9/17/07, Andre Oppermann <andre@freebsd.org> wrote: > Kip Macy wrote: > > http://perforce.freebsd.org/chv.cgi?CH=126488 > > > > Change 126488 by kmacy@kmacy_home:ethng on 2007/09/16 21:57:27 > > > > add a route structure to the inpcb > > release the reference to the rt_entry if its set when we free the inpcb > > We already had this once before and it was nasty as you had to > do a full inpcb walk to nuke any references if a route changed. Nope. If you keep a generation counter for the route in the inpcb and check it each time in udp_output against the rtentry / route table you'll know that your cached route is no longer valid. I'm not currently doing that, but it won't be hard to add. In any event, incurring a 20% slowdown (in the uncontended case, its basically crippling in the contended case i.e. 1 flow on the same route) for an event that occurs < .01% of the time on most networks seems questionable. -Kip
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b1fa29170709170150y3b391a22x1a52f4858b6882ca>