Date: Thu, 10 Apr 2008 08:35:06 +0200 From: Vitezslav Novy <vnovy@vnovy.net> To: Jack Vogel <jfvogel@gmail.com> Cc: freebsd-current@freebsd.org, Steven Hartland <killing@multiplay.co.uk> Subject: Re: IP bad-len 0 ( on em0 ) Message-ID: <47FDB51A.1030606@vnovy.net> In-Reply-To: <2a41acea0804091659l7ac2d9adqcbdd0caf900469b@mail.gmail.com> References: <200802042142.38606.qpadla@gmail.com> <200802070018.54429.qpadla@gmail.com> <006801c87f19$a14d8060$b6db87d4@multiplay.co.uk> <200804092043.24500.qpadla@gmail.com> <2a41acea0804091659l7ac2d9adqcbdd0caf900469b@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Jack Vogel wrote: > On Wed, Apr 9, 2008 at 10:43 AM, Nikolay Pavlov <qpadla@gmail.com> wrote: >> On Thursday 06 March 2008 01:35:43 Steven Hartland wrote: >> > Did you ever get anywhere with this? Did Jack respond? >> >> Nope. I've disabled tso. > > > I've looked into this a little, and then got interrupted with other issues. The > reason the thing is zero'ed is because the hardware is going to repacketize > this big wad that its been handled, it should be making new headers that > appear in the packets on the wire. So its not yet clear to me what the > real brokenness is, you are actually SUPPOSED to zero that value and > csum according to documentation, but the rewritten headers should have > correct len's in them, so the question is why in some cases they do not. I think packets on wire have correct IP-len, but after sending packet to card, driver injects original long packet with zeroed IP-len to BPF. So in tcpdump, we see packet with zero IP-len. vita
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?47FDB51A.1030606>