andrew bezella wrote: > as reported in https://bugs.debian.org/1105223, the systemd predictable > network device names change between bookworm and its default 6.1 > kernel and trixie's 6.12 kernel for interfaces using the i40e driver.
Even if it doesn't go in the Release Notes, it's definitely one for https://wiki.debian.org/NetworkInterfaceNames#Complications_and_corner_cases so thanks for reporting it. > in particular it seems that the newer kernel starts setting the > phys_port_name: > ia2060# uname -vr > 6.1.0-20-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.1.85-1 (2024-04-11) > ia2060# cat > /sys/devices/pci0000:5d/0000:5d:00.0/0000:5e:00.0/net/enp94s0f0/phys_port_name > cat: > '/sys/devices/pci0000:5d/0000:5d:00.0/0000:5e:00.0/net/enp94s0f0/phys_port_name': > Operation not supported > > ia2060# uname -vr > 6.12.22+bpo-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.22-1~bpo12+1 (2025-04-25) > ia2060# cat > /sys/devices/pci0000:5d/0000:5d:00.0/0000:5e:00.0/net/enp94s0f0np0/phys_port_name > p0 > > this and similar bugs have been closed as "This is expected due > to kernel changes." Ah, yes, "expected" in the same messed-up sense that this is the "predictable names" scheme. But as your backports kernel shows, this isn't specifically a dist-upgrade issue; it's just one of those things all GNU/Linux admins ought to be warned about, in between "Backups!" and "Backups!" However, we've already got another "predictable names" bugreport for the Trixie Release Notes involving QEMU (#1092176) so it might make sense to consolidate them into one reminder in the Release Notes. (Were the "similar bugs" similar NICs with similar symptoms?) > however, from a user's point of view losing > networking on reboot was very much unexpected and might benefit from > being documented in the trixie release notes[1]. per the Network > device naming schemes document[2] setting the ID_NET_NAME_ALLOW > udev property might disable this behavior if it can be predicted > (untested locally). Interesting. The trouble is, if you know you're going to need that sort of precisely targeted override then you'd be better off just planning for the change in the first place, or creating a custom .link file and opting out of the whole mess. Anyway, collecting information to go on the Wiki page: I gather this is an Intel Ethernet Connection X722, which is onboard on Lenovo ThinkSystem servers. It does seem especially messed up that they're complicating ID_NET_NAME_ONBOARD, which was hitherto the only kind that users had a good chance of actually predicting... -- JBR with qualifications in linguistics, experience as a Debian sysadmin, and probably no clue about this particular package
