Re: loopback interface broken on current

Am 01/09/13 09:59, schrieb Gleb Smirnoff:
> On Wed, Jan 09, 2013 at 09:56:25AM +0100, Hartmann, O. wrote:
> H> Same here.
> H> The OS: FreeBSD 10.0-CURRENT/amd64 r245218. Since I have three boxes
> H> running approximately the same configurations (I share my configs
> H> between lab and home), but different hardware, I'm confused.
> H> 
> H> The symptoms in my case are:
> H> 
> H> Booting the box is all right until it comes to the start of nfsuserd.
> H> Prior to that, ntp adjusts the clock properly with an external time
> H> server  - so this implies network connectivity. Start of nfsuserd is
> H> stuck forever.
> H> Interrupting the start of nfsuserd restarts several other services, but
> H> winbindd and slapd (OpenLDAP) get stuck again. In case I also interrupt
> H> them, there are other services which will not start.
> H> 
> H> Trying to login as root on the console fails - I never get a password
> H> tag after having issued the root login name. Since this machine is bound
> H> to a local and remote OpenLDAP backend, I'm used to have an emergency
> H> local user which usually works - but this time, neither root nor this
> H> user can login!
> H> 
> H> Bringing up the box in SINGLEUSER allows me to login. Investigating the
> H> network interfaces with ifconfig reveals, that the loopback did not get
> H> assigned to any inet 127.0.0.1 address. Sometimes there is only inet6
> H> linklocal address, some nd6 options, but sometimes even IPv6 assignments
> H> do not show up.
> H> 
> H> In a desperate move I tried to recompile a kernel. In /etc/src.conf, I
> H> recompile also the kernel module for the most recent virtual-box kernel
> H> module. While the kernel and module (*.ko) get installed properly, the
> H> recompilation of the VirtualBox port gets stuck when the system unfolds
> H> the source tarball. Hitting Ctrl-T say "sbwait" for the process. Other
> H> processes seem to have trouble getting a proper ownership or UID for a
> H> file - this is my naiv interpretation what I see at the surface.
> H> 
> H> The funny thing is, that after several reboots, the box gets up as normal.
> H> 
> H> I revealed this issue approx. two weeks ago when out of the sudden the
> H> amd automounter stopped working and the NFSv4 network drives didn't
> H> attach properly and made the whole box being stuck.
> H> 
> H> Sorry for the more superficial description of the problem ...
> H> 
> H> Has the problem been identified? Is there a solution? Since it affects
> H> only my very modern hardware (i7-3930, 32GB RAM, ASUS P9X79 WS
> H> mainboard), while a very same setup on older hardware (our local server
> H> is Intel Q6600 with 8GB RAM and and oldish Intel P45-chipset based
> H> mainboard), both systems do have Intel NICs, I'm a bit confused.
> 
> This looks unrelated to the problem discussed, because r245218 is
> later than r244989 which backed out my change.
> 
> Can you do a binary search to identify which revision broke things?
> 

Sorry for the delay.

Today I realized that the problems occured and described are due to the
fact that I use jumbo frames (mtu 9100 or mtu 6150) on the em0-device
(em0_at_pci0:0:25:0:        class=0x020000 card=0x844e1043 chip=0x15038086
rev=0x05 hdr=0x00
    vendor     = 'Intel Corporation'
    device     = '82579V Gigabit Network Connection'
    class      = network
    subclass   = ethernet
). Using the default of mtu 1500 does not make the problem occur.

My time constraints disallow to do further or deeper investigations - at
least until the end of next week, I'm sorry. The problems arose around
Christmas, it might be even earlier, since I didn't access the machine
since 10th December.

I have other boxes with Intel NICs - different chiptypes. They do not
show this problem.

Oliver