On 23/10/19(Wed) 17:55, Lauri Tirkkonen wrote: > [...] > I did some bisecting today: using the github src mirror, I built > bsd.rd's on a fresh 6.5 install (which I had to manually bootstrap > libc.so.95.1 on to build all the necessary revisions), booted them on > a Hetzner VM, and checked whether 'dhclient vio0; ifconfig vio0 -soii' > lost connectivity and started displaying the "no arp information" kernel > messages. The results: > > a571f6fc23a13c4832f0066f64a7d855785533f4 is the first bad commit > commit a571f6fc23a13c4832f0066f64a7d855785533f4 > Author: krw <[email protected]> > Date: Fri May 10 18:50:11 2019 +0000 > > Restart the protocol and get a new/renewed lease for any relevant > RTM_IFINFO seen. As dhclient no longer commits suicide to restart > the > protocol this should be very low cost. > > This will help some trunk(4) setups after some pending network stack > commits. > > Requested by & ok florian@ > > :040000 040000 038c213a90ba80990a4ab6e74ff3de09227749e8 > fdfc486143c5c06197c590f5f2e59d33c0c7568f M sbin > > This corresponds to dhclient.c r1.634. I have no idea why this diff > would cause '-soii' to make the dhclient-added v4 default route lose > RTF_LLINFO (that seems to be the reason for the error message, from > netinet/if_ether.c), but it does.
Could you repeat the steps to make that issue arise while capturing routing socket messages with "# route monitor" on another terminal? If you do that before and after the commit it might help figuring out what's different.
