On 27 November 2014 at 03:12, Theo de Raadt <dera...@cvs.openbsd.org> wrote: >> On Tue, Nov 25, 2014 at 18:42 +0100, Mike Belopuhov wrote: >> > On Mon, Nov 24, 2014 at 19:04 +0100, Mike Belopuhov wrote: >> > > Hi, >> > > >> > > IP header is not always aligned since bpf copies out the mbuf >> > > chain into the contigous buffer provided by the userland. I've >> > > seen this with large packet sizes on VLANs. ip_print will then >> > > copy the packet but the Ethernet header into the internal buffer >> > > so that it can cast it to the IP header structure and update >> > > global packetp and snapend pointers hence preventing the -Xx >> > > dumping code from printing out the Ethernet header itself. >> > > >> > > Diff below fixes it. OK? >> > > >> > >> > better diff. the problem is that dissectors use packetp and >> > snapend pointers themselves therefore they should be pointing >> > to the newly allocated structure. we can restore them once >> > we're done with the inner content and go back to the caller >> > to see if we need to hexdump the contents. >> > >> > i'll see if i can cook and test the ipv6 version. >> > >> > OK? >> > >> >> now with an ip6 version and i've made sure that this fixes >> dumping unaligned ipv6 packets as well. in the meantime >> jsg@ has lured me into looking at the afl crash in the same >> code and it looks like the check from ip6_print is useful >> here: if we haven't got enough data for a header, don't >> bother with anything else and just bail. >> >> ok? > > > Did you test on a strict alignment machine?
works fine on sparc64.
