On 6/7/18 5:49 PM, Jakub Kicinski wrote: > On Thu, 7 Jun 2018 17:28:59 -0700, Eric Dumazet wrote: >> On 06/07/2018 05:11 PM, David Miller wrote: >>> From: Jakub Kicinski <jakub.kicin...@netronome.com> >>> Date: Thu, 7 Jun 2018 17:06:23 -0700 >>> >>>> [ 293.213661] ip_send_unicast_reply+0x1b67/0x1d0e >>> >>> This calls ip_setup_cork() which can NULL out the 'rt' route >>> pointer. Hmmm... :-/ >>> >> >> >> UBSAN seems unhappy with dst being NULL in : >> >> dst_release(&rt->dst); >> >> But the code obviously is ready for dst being NULL, it is even documented :) > > Oh, so the code depends on dst being the first member? Would it make > sense to just cast the pointer instead? >
I've been going the other way with 'rt to dst' and 'dst to rt' transformations. Perhaps UBSAN should be updated to understand that NULL + 0 is ok.
