On Sun, 26 Apr 2020 at 03:38, Michael Richardson <mcr@xxxxxxxxxxxx> wrote: > > > David Balažic <xerces9@xxxxxxxxx> wrote: > >> OK, more data: > >> I started tcpdump with the -e option as suggested on openwrt forum: > >> tcpdump -e -v -i eth1.3902 pppoed > >> > >> on disconnect, this was logged (times are UTC): > >> > >> > >> tcpdump: listening on eth1.3902, link-type EN10MB (Ethernet), capture > >> size 262144 bytes > >> 18:23:09.204991 a4:7b:2c:9e:c7:44 (oui Unknown) > 44:4e:6d:fd:c7:39 > >> (oui Unknown), ethertype PPPoE D (0x8863), length 97: PPPoE PADO > >> [Service-Name] [AC-Name "SIMB_TABOR_BNG1"] [Host-Uniq > >> 0x44************long_number******************************AA] > >> [AC-Cookie ".5b************v"] > > > Isnt this strange? The dst addr is not my router or any other known. > > Also the Host-Uniq value is different. > > As if is traffic meant for someone else. > > oh, I understand. > thank you for noting this. > So, it's as if you are getting a PPPoE message for someone else, and when > that happens, the interface is dying. > Is it always the same ethernet address? Yes,the same. > Do you have ebtables? Could you arrange to filter out packets like that? > That would point to there being some kernel bug. > > Clearly, your ISP has some other bug that they are sending stuff down the > wrong pipe, but that could just be hash collisions that they assume are > "harmless". Those packets don't seem to be a problem. In the last connection, I received 15 of them and no reconnect. As suggested on OpenWRT, I removed a IPv6 wan device config that was running DHCPv6 on eth1. Then he connection was up for 45 hours before dropping. Could be a coincidence... I posted the logs at https://forum.openwrt.org/t/pppoe-disconnects-every-few-hours/61239/22 (gmail wraps lines...) Regards, David