[Bloat] SQM scripts seems to break dhcpv6 client

[vom513]

Hello all,

This one was (and kind of still is) a head scratcher.

I am running sqm-scripts on my Debian 12 Linux firewall - installed from: https://github.com/tohojo/sqm-scripts

For DHCPv6 client I use wide-dhcpv6-client (dhcp6c).

I run simple.qos, and it’s on a Chelsio 10G adapter.  I have disabled all the fancy offload on my WAN port on this card (seemed to cause kernel issues/oopses with sqm running).

I have two internet connections at home (primary and backup).

My interface config:

UPLINK=380000
DOWNLINK=380000
SCRIPT=simple.qos

Backup and primary config are the same except for UPLINK/DOWNLINK values.

I use nftables, and am very careful about making sure I an letting in ICMPv6 (ND RA etc), as well as DHCPv6.  In fact all my policies are default accept (I use manual cleanup/block rules at the ends of them).

On my primary provider, I don’t seem to have any issues.  Something about the DHCPv6 packets from my backup provider - it seems like the packets are “eaten”.  I can see them come in via tcpdump, they just never make it to my client daemon.

After tons of hair pulling (our primary is having a known outage this morning, so I kicked to backup) - I decided to reboot with no SQM.

After doing this - I get my IPv6 addresses on all interfaces like normal/expected.

Any thoughts on this ?  I really like SQM scripts and want to keep it working, but if I have to choose I will choose having full dual-stack working :(

Thanks in advance.