Package: libvirt-daemon Version: 7.6.0-1 Since the upgrade of Debian 11 to current testing, libvirt has a regression: It does not recognize veth devices as "nodedevs" (devices available on the host and known by libvirt) any more. This breaks assigning them to VMs for networking or PXE booting.
# ip link add name eth42 type veth peer name v_eth42 # virsh nodedev-list --cap net net_eth0_52_54_00_12_34_56 net_lo_00_00_00_00_00_00 → No trace of eth42 and v_eth42. This still works fine on Debian 11, with libvirt 7.0.0-3, systemd 247.3-6, and kernel 5.10.0-8-cloud-amd64: # ip link add name eth42 type veth peer name v_eth42 # virsh nodedev-list --cap net net_eth0_52_54_00_12_34_56 net_eth42_ba_f8_89_e3_96_41 net_lo_00_00_00_00_00_00 net_v_eth42_c2_28_c1_aa_03_0b I file this against Debian, as none of our other cockpit-tested operating systems show this regression (Ubuntu 20.04 LTS/21.10, Fedora 34/35, RHEL/CentOS 8/9, etc.). In particular current Fedora 35 has the exact same libvirt version 7.6.0, just a newer systemd (249.4 in Fedora, 247.9 in Debian testing). I don't think it's related to udev: In both Debian 11 and testing the device looks exactly alike in `udevadm info --export-db`: P: /devices/virtual/net/eth42 L: 0 E: DEVPATH=/devices/virtual/net/eth42 E: SUBSYSTEM=net E: INTERFACE=eth42 E: IFINDEX=6 E: USEC_INITIALIZED=106226972 E: ID_MM_CANDIDATE=1 E: ID_NET_DRIVER=veth E: ID_NET_LINK_FILE=/usr/lib/systemd/network/99-default.link E: ID_NET_NAME=eth42 E: NM_UNMANAGED=1 E: SYSTEMD_ALIAS=/sys/subsystem/net/devices/eth42 E: TAGS=:systemd: E: CURRENT_TAGS=:systemd: Thanks, Martin
