Lennart Poettering

hey, @sean-anderson-seco any chance you can weigh in on this? does this make sense? (you did the original devicetree support for udev interface naming, and you might have a tak...

i don't grok the csv business, can you explain?

Overriding the port number is simply not supported at this time. I guess we could add that, but that'd be a new feature, not a bugfix.

And yes, @DaanDeMeyer is right, my idea was that networkd takes the lock continously for the sysctls it wants to manage (i.e. it would continously keep open an fd to...

> And keep the file descriptor forever in networkd? That still would not prevent manually setting the values with `sysctl` or by writing in `/proc/sys` Not forever, only as long...

so my (evolved) current thinking goes like this: for each network interface we manage, install: 1. one inotify watch on /proc/sys/net/ipv4/conf/$iface/ 2. one inotify watch on /proc/sys/net/ipv6/conf/$iface/ and that's it....

> Another idea: what about publishing on dbus the list of the sysctls used by networkd? D-Bus is not a great option really for networkd: not available in initrd/early boot....

> Does inotify trigger an event when writing a file in the monitored directory? If so, that's OK, but if not, we need to monitor so many files?? Hm... yes...

Hmm, I think adding this would be OK. but I think we should overload the existing systemd.machine_id= setting for this. i.e. if systemd.machine_id=smbios would mean: read the uuid from smbios....

actually, it should be called systemd.machine_id=firmware I guess. After all we already support both smbios and devicetree for this, hence a more generic name sounds better