Debian Lenny 5.0.7 / Squeeze
Kernel 2.6.26-2-xen-amd64
ipv6-enabled host

Additional information:

Building with "--disable-ipv6 --enable-agent" changes nothing, same errors.

This error arises at attempt to run Zabbix agent compiled on version 2.6.27 or above on a platform with a kernel 2.6.26 and lower.

Note that static linking will not help in this case because it is the socket() system call that does not support SOCK_CLOEXEC flag on earlier kernels.

Solved, indeed.

Just bootet the squeeze-VM with 2.6.32-5, agent now works as expected:

===
512:20110111:132326.619 Zabbix Agent started. Zabbix 1.8.4 (revision 16604).
520:20110111:132326.630 zabbix_agentd listener started
519:20110111:132326.631 zabbix_agentd listener started
518:20110111:132326.631 zabbix_agentd listener started
517:20110111:132326.631 zabbix_agentd collector started
===

Thanks for checking into this!

Can this ticket be reopened?

I'm having the same issue (same log lines) on a genuine ipv6 enabled machine, having compiled Zabbix 1.8.4 stable on a 2.6.26 kernel running debian squeeze.

Is it just simply that Zabbix 1.8.4 now REQUIRES 2.6.27+ kernel?

Just FYI: I have been through the same troubleshooting steps. a) enabled/disabled ipv6 at configure time b) with ipv4/ipv6 for IP, or blank

The problem is not ZABBIX, it's Debian Squeeze:

You HAVE to run Squeeze with 2.6.32!

Even compiling on Squeeze does not solve this issue, I assume that libc6 is compiled against 2.6.32 and therefor also ZABBIX agent requries 2.6.27+.

If you also have a squeeze-VM running on a lenny XEN Dom0: install linux-image-2.6.32-xen-amd64 in the VM, copy /boot/vmlinuz-2.6.32*, /boot/initrd-2.6.32* and /lib/modules/2.6.32* to the Dom0, change your domU's config to use the new kernel on the dom0 and restart the VM.

This works - resulting in Debian squeeze with 2.6.32 kernel, zabbix works.

The problem is not ZABBIX, it's Debian Squeeze:

You HAVE to run Squeeze with 2.6.32!

Even compiling on Squeeze does not solve this issue, I assume that libc6 is compiled against 2.6.32 and therefor also ZABBIX agent requries 2.6.27+.

If you also have a squeeze-VM running on a lenny XEN Dom0: install linux-image-2.6.32-xen-amd64 in the VM, copy /boot/vmlinuz-2.6.32*, /boot/initrd-2.6.32* and /lib/modules/2.6.32* to the Dom0, change your domU's config to use the new kernel on the dom0 and restart the VM.

This works - resulting in Debian squeeze with 2.6.32 kernel, zabbix works.

It's not always possible to use Debian Squeeze with kernel > 2.6.26.

Many hosting providers use Virtuozzo/OpenVZ. The most of them use a kernel like 2.6.18, 2.6.24 or 2.6.26.
This will not change in the near future.

Other applications are working in this environment. So I think, something must be done.

I maintain an host in a virtuozzo container with 2.6.18-028stab070.14 and Debian Squeeze - the zabbix agent works flawlessly, although I didn't expect that to happen. The reported bug does not show up here.

I have some systems with OpenVZ and Debian Lenny (2.6.26)/Ubuntu Hardy(2.6.24) on my site. The Debian Squeeze containers have the same problem that is described here.

marcel, have you tried compiling the agent on a platform with kernel 2.6.26 or lower (maybe the same you plan to run it on) ?
if the problem description is correct, that should solve your issue

same problem,

compiled on a newer kernal, ran on a 2.6.26 kernel.

problem resolved once I compiled on the 2.6.26 kernel.

time to upgrade my kernels

I compiled it in a OpenVZ container on a Debian Lenny host (2.6.26 x86_64) and Debian Squeeze (x86_64) inside the container.
libc: 2.11.2-10
gcc: 4.4.5-1
kernel: 2.6.26-2-openvz-amd64

Problem still exists.

I compiled Zabbix 1.8.10 in a xen domU with Debian Squeeze amd64 and

libc: 2.11.3-2
gcc: 4.4.5-8
kernel: 2.6.26-2-xen-amd64

and the agentd fails to start in the exact same machine with the same socket error as above.
So the statement "This error arises at attempt to run Zabbix agent compiled on version 2.6.27 or above on a platform with a kernel 2.6.26 and lower." also written in the FAQ cannot be true...

The Zabbix 1.8.2 from the Debian Squeeze repositories run without problems though...

Why is this report set to "wont' fix"? The problem apparently still exists for Zabbix agents of version 1.8.11 at least.
On several machines I randomly get:
27271:20120531:110804.261 Starting Zabbix Agent [svn]. Zabbix 1.8.11 (revision 26205).
27271:20120531:110804.261 listener failed: socket() for [[0.0.0.0]:10050] failed: [22] Invalid argument

This is a very annoying issue and I can't figure out how to solve it other than install an ancient 1.8.2 agent on those systems.

I believe I found something. This problem happened on Debian GNU/Linux systems where I installed security updates. Apparently it involves kernel updates. A reboot fixes the problem. So perhaps some networking APIs are using parts of the wrong kernel.

You can build zabbix with -DOLD_LIBC_MODE

diff -ur zabbix-2.0.2_orig/src/libs/zbxcomms/comms.c zabbix-2.0.2/src/libs/zbxcomms/comms.c
— zabbix-2.0.2_orig/src/libs/zbxcomms/comms.c 2012-07-31 13:16:46.000000000 +0400
+++ zabbix-2.0.2/src/libs/zbxcomms/comms.c 2012-08-30 18:56:36.525523544 +0400
@@ -31,8 +31,12 @@

1. define ZBX_SOCKLEN_T socklen_t
   #endif

-#if !defined(SOCK_CLOEXEC)

• 1. define SOCK_CLOEXEC 0 /* SOCK_CLOEXEC is Linux-specific, available since 2.6.23 */
     +#if !defined(OLD_LIBC_MODE)
     +# if !defined(SOCK_CLOEXEC)
     +# define SOCK_CLOEXEC 0 /* SOCK_CLOEXEC is Linux-specific, available since 2.6.23 */
     +# endif
     +#else
     +# define SOCK_CLOEXEC 0
     #endif

/******************************************************************************

Just applied gescheit's patch. Now Zabbix-server 2.0.4 runs on Linux kernel 2.6.21.