kern/57972: rge* interface stops communicating after a while

[john%ziaspace.com@localhost]

>Number:         57972
>Category:       kern
>Synopsis:       rge* interface stops communicating after a while
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Feb 29 05:55:00 +0000 2024
>Originator:     John Klos
>Release:        NetBSD 10.0_RC5
>Organization:
	
>Environment:
	
	
System: NetBSD sage.zia.io 10.0_RC5 NetBSD 10.0_RC5 (SAGE) #0: Tue Feb 27 07:17:37 UTC 2024 john%sage.zia.io@localhost:/usr/obj-amd64/sys/arch/amd64/compile/SAGE amd64
Architecture: x86_64
Machine: amd64
>Description:
	
Running a system with options GATEWAY as a NAT router with npf and 
with rge* as the primary public interface occasionally leads to a 
state where traffic on the public rge* interface stops flowing.

In case this was an issue with the specific card, I tried a completely 
different card. No change.

Relevant lines from npf:

$ext_if = rge0
$ext_ip = { inet4($ext_if) }
map $ext_if dynamic $localnet_lan -> $ext_ip
group "external" on $ext_ip {
	pass stateful out final all
	pass stateful in final family inet4 proto tcp to $ext_ip port ssh apply "log"
	block in final from <blocklist>
	block final all apply "log"
}
group default {
	pass final on lo0 all
	pass in final all
	pass out final all
}

When in this state, netstat -m shows:

7360 mbufs in use:
	7302 mbufs allocated to data
	51 mbufs allocated to packet headers
	7 mbufs allocated to socket names and addresses
0 calls to protocol drain routines

Even though no traffic appears to be flowing, mbufs change a bit:

7363 mbufs in use:
	7300 mbufs allocated to data
	52 mbufs allocated to packet headers
	11 mbufs allocated to socket names and addresses
0 calls to protocol drain routines

After running "ifconfig rge0 down" and waiting a few seconds, netstat -m gives:

4109 mbufs in use:
	4102 mbufs allocated to data
	2 mbufs allocated to packet headers
	5 mbufs allocated to socket names and addresses
0 calls to protocol drain routines

Then, running "ifconfig rge0 up" gives working communications again.

Kernel config file is GENERIC plus options GATEWAY. Machine also routes public IPv6 
and runs dhcpcd on public interface.
>How-To-Repeat:
	
>Fix:
	

>Unformatted: