#486 IPv6: mDNSResponder eating up 100% - PowerD fail to throttle
Background:
Running XigmaNAS 14.1.0.5
Host: HP Elitedesk 800G2 SFF,
[...]
em0: <intel(r) i219-lm="" spt-h(2)=""> mem 0xe1000000-0xe101ffff at device 31.6 on pci0
em0: EEPROM V0.8-4
em0: Using 1024 TX descriptors and 1024 RX descriptors
em0: Using an MSI interrupt
em0: Ethernet address: ---------------
em0: netmap queues/slots: TX 1/1024, RX 1/1024</intel(r)>
Attached to em0 (designated LAN): vlan1, vlan2, vlan3.
vlan interfaces are not assigned with any IP (neither static, nor DHCP).
Using Bastille with jails attached to three different vlans: vlan1, vlan2, vlan3.
This scenario works without problems as long as either IPv6 is disabled on the master NIC (em0) and only IPv4 is used, or IPv6 is enabled, and configured statically (I used ULA/ULA prefices). The configuration of the epair pseudo NICs within the jails seems to have NO effect on the phenomenon described below.
For IPv6, via rc.conf settings are performed like:
ifconfig_em0_ipv6="inet6 fd00:47a::11 prefixlen 64 auto_linklocal accept_rtadv -no_radr up"
ipv6_defaultrouter="fd00:47a::1"
I need ULA to address the host within my IPv6 network.
When IPv6 on the master (physical) NIC (em0) is enabled AND Type is set to Auto, NIC em0 obtains an IPv6 address provided by another FreeBSD router advertising instance running rtadvd and advertising both a prefix delegated from ISP's pool AND ULA. In that scenario, the default route is set to an
fe80::/16%em0 address.
I;m quite the novice to IPv6, so this kind of setup works on other FreeBSD hosts so far.
IPFW is enabled. Added a rule to pass FROM fe80::/16 -> ff02::fb via ALL interfaces as this seems not to be met by the default ruleset.
PROBLEM:
With the setup AUTO as described above, mDNSresponder is running at 100% on one Core of the 4 core CPU never letting powerd throtteling the power the clock and therefore the host is consuming always at 100% power although doing nothing - speeding at 3,4 GHz at all times.
The only way to fix this is either disabling IPv6 (not an option) or switching back to configure static IPv6 addresses, which leaves me with ULA - also a bad option.
Checking the mDNS process with truss gives me some strange results which might be usefull:
truss -fade -p 5223
[...]
5223: 1.239141331 __sysctl("net.routetable.0.0.5.0",6,0x0,0x820bee748,0x0,0) = 0 (0x0)
5223: 1.239181244 __sysctl("net.routetable.0.0.5.0",6,0x11ffa8809000,0x820bee748,0x0,0) = 0 (0x0)
5223: 1.239211063 socket(PF_INET,SOCK_DGRAM|SOCK_CLOEXEC,0) = 14 (0xe)
5223: 1.239235239 ioctl(14,SIOCGIFINDEX,0x820bee7c0) ERR#6 'Device not configured'
5223: 1.239258685 close(14) = 0 (0x0)