CURRENT: "broken pipe" / SIGNAL 13 / no X11 / OpenLDAP/nscd weirdness

From: Hartmann, O. <ohartman_at_zedat.fu-berlin.de>
Date: Wed, 06 Mar 2013 11:06:29 +0100
The most recent CURRENT (FreeBSD 10.0-CURRENT #0 r247865: Wed Mar  6
08:52:15 CET 2013/amd64, built with CLANG and
CXXFLAGS+=              -stdlib=libc++
CXXFLAGS+=              -std=c++11
set in /etc/src.conf, for the record) does have some serious issues and
I'm wondering why others do not.

The problems I face affect ALL most recent FreeBSD CURRENT boxes. It
doesn't matter whether I use my customized kernel config or GENERIC.

The symptoms are weird. Using system commands like top(1) give me a
"broken pipe" on the console. Repeating the command gives me then after
a while top screen as expected.

Services, like OpenLDAP (which I use throughout all systems with nscd
and /etc/nsswitch.conf configured), nagios, hald, dbusd or starting X11,
doesn't work properly: starting or not starting seems to be a
statistical game.

Since some kernel modifications needs recompiling ports which provides
kernel modules, like

emulators/virtual-box-ose-kmod
x11/nvidia-driver

fail with

[...]
*** [do-extract] Signal 13

This happens to all ports I try to compile/recompile, but sometimes,
again the gamble, a simple make in the port's directory works and I can
recompile.

Also strange and possibly a sort of indication for the problem is the
behaviour of sudo(1):

hartmann_at_gate: [~] sudo su -
Password:
hartmann_at_gate: [~]

sudo(1) does not switch to the superuser as it is expected, sometimes I
receive a "broken pipe" error, sometimes it works as expected.

In single user mode, I can cmpile ports without problems and I do not
see this "broken pipe" issue. To be able to compile and install a port,
I need to perform*** [do-extract] Signal 13

service ldconfig start

which works, so all "suspicious" libraries should be loaded (as a naiv
guess, I thought a miscompiled shared lib could casue the problem).
Well, in general for the ports problem, the SINGLEUSER mode doesn't show
the problem.

The X11 system isn't starting up anymore. The problem why this doesn't
work seems to "change":

with sources/kernel prior or equal to r247839, dbusd couldn't start
(obviously broken pipe ...).

Sporadically non-working services, like slapd (OpenLDAP) seem to start
after repeatedly issue "service slapd restart", but I can not reproduce
by a number the failings. Sometimes the services start at boot time on a
regular basis, sometimes not.

For further informations of the settings: I use TEMPFS for /var/run and
/tmp. This is common to all systems (but it doesn't affect single user
mode, somehow).

Things which do not work or solve the problem:

a) use a GENERIC kernel
b) avoid either in GENERIC or CUSTOM kernel CXXFLAGS+= in /etc/src.conf
c) disable nscd
d) recompile ports which cause trouble or tend to be highly
inresponsive: openldap24-sasl-client/server, pam_ldap, nss_ldap, sudo,
dbus and hal, xdm, xorg-server, virtualbox-ose-kmod

I tried to recompile for now several times the whole sources with all
standard setting (no custom /etc/src.conf or make.conf, also GENERIC
kernel). No salvation.

I also switched off SMP via setting

#kern.smp.disabled=1

since this caused earlier some trouble - but this is more a desperate
move to narrow down the problem a bit.

I hope someone can reproduce this problem since I think I'm not the only
one who is using CURRENT in a server-like multiuser/workstation
environment (not only Vbox images, jails).

Regards,

oh
Received on Wed Mar 06 2013 - 09:05:36 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:35 UTC