Hi, On Tue, Sep 25, 2018 at 11:23 AM Jo-Philipp Wich <j...@mein.io> wrote: > Maybe netlink congestion or something related to privilege dropping? Can > you manage to capture an strace log of the running dnsmasq instance > while the network is getting restarted?
After some discussion on the dnsmasq mailing list, the cause has been found and your theory about "resubscribe" is correct. When dnsmasq is configured to listened to one particular interface, dnsmasq binds (SO_BINDTODEVICE) to this interface during startup if the interface is available. On my devices, I had configured dnsmasq to only listen to br-lan. When I restart networking, br-lan is re-created. This event is not detected by dnsmasq and the socket will silently fail. If I configure dnsmasq to listen to two interfaces, SO_BINDTODEVICE is never used and the error disappears. Why this error has appeared now is still unknown, but it could be because no one has gone looking for it. My current work-around, base on the advice of Simon Kelley, is to have the function which looks up the interface to bind to (whichdevice()) return NULL. Filtering of received packets is anyway done not dependent on the bound socket. BR, Kristian _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel