Hello, I'm running a dual socket NUMA CURRENT host (Fujitsu RX host) running several jails. Jails are attached to a bridge device (bridge1), the physical device on that bridge is igb1 (i350 based NIC). The bridge is created via host's rc scripts, adding and/or deleting epair members of the bridge is performed by the jail.conf script.
I do not know how long the setup worked, but out of the blue, last week after a longish poudriere run after updating the host to most recent CURRENT (as of today, latest update kernel and world) and performing "etcupdate" on both the host and all jails, traffic beyond the bridge is not seen on the network! All jails can communicate with each other. Traffic from the host itself is routed via igb0 to network and back via igb1 onto the bridge. I check all setups for net.link.bridge: net.link.bridge.ipfw: 0 net.link.bridge.log_mac_flap: 1 net.link.bridge.allow_llz_overlap: 0 net.link.bridge.inherit_mac: 0 net.link.bridge.log_stp: 0 net.link.bridge.pfil_local_phys: 0 net.link.bridge.pfil_member: 0 net.link.bridge.ipfw_arp: 0 net.link.bridge.pfil_bridge: 0 net.link.bridge.pfil_onlyip: 0 I did not change anything (knowingly). I also have an oldish box running single socket processor, also driven by the very same CURRENT and similar, but not identical setup. The box is running very well and the bridge is working as expected. I was wondering if something in detail has changed in the handling of jails, epair and bridges. I followed the setup "after the book", nothing suspicious. Maybe someone has a clue what might break the bridge. By the way: ifconfig bridge1 looks as always, igb1 as member and it doesn't make any difference whether I force the bridge to inherit igb1's MAC or not. We also checked for the switches whether BPDU Guard may have been triggered, but everything looks good from the outside - execept the fact the brdiged interface seems inactive (but up) from the outside ... Kind regards oh -- O. Hartmann
