snmpdm coredumps

[David Knickmeyer]

Hi,

I have a VM running 11.23 where the snmp daemons refuse to run. I have other identical virtual machines that have no problem. Okay, obviously they are not identical because it runs there, but they are on the same software versions/patches. Even the Microsoft solution (rebooting) didn't help. Here's what I get if I run the snmpdm command myself:

(root@sapbde)# snmpdm -apall -n
SNMP Research SNMP Agent Resident Module Version 15.3.1.0
Copyright 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001 SNMP Research, Inc.
Successfully opened log file /var/adm/snmpd.log
at line 593 in file ../mastmain.c
init_fnames: searching for configuration files in /etc/srconf/agt (default)
at line 134 in file ../../../snmpd/shared/fnames.c
Binding to port 161
at line 482 in file /view/anu.IA.6.2/DE/snmpr/seclib/t_ip.c
Attempt #5 to bind to socket
at line 484 in file /view/anu.IA.6.2/DE/snmpr/seclib/t_ip.c
GetLocalIPAddress: hostname is sapbde
at line 166 in file /view/anu.IA.6.2/DE/snmpr/seclib/t_ip.c
Memory fault(coredump)

I haven't been able to find any differences in any configuration files I know of (/etc/rc.config.d/SnmpMaster.conf or /etc/snmpd.conf. The only line there is trap-dest: xxxx.xxxx.com)

Any help would be appreciated.

Thanks,

Dave

[Steven E. Protter]

Shalom Dave,

Microsoft solution. Cute phrase but so true.

I would check the patch levels of the working VM machines against this one. Perhaps a critical patch for snmp is missing.

swlist -l product

Maybe use diff to compare the output.

SEP

[David Knickmeyer]

Thanks for the idea. I did just assume that the servers were patched the same but I hadn't actually bothered to check. The servers are only a few months old.

Unfortunately, they are patched identically and on the same software versions.

Thanks,

Dave

[Padma Asrani]

Hi David

Could you write a sample C program on the machine where it is dumping core to evaluate this call

LocalIPAddress = inet_addr(hostname);

Regards
Padma

[David Knickmeyer]

C program, huh? Okay, it's been a little while and I never did much with the network stuff, but I cribbed this together:

#include
#include
#include
#include


int main () {

struct sockaddr_in sa;

sa.sin_addr.s_addr = inet_addr("xxx");
printf ("%s\n",sa.sin_addr.s_addr);
}

The result? "Memory fault(coredump)"

The interesting thing (and which makes my program suspicious) is that I get the same result on the other VM guests and the host as well, even though the snmpdm is running on them.

So, obvious bug in my C program or a bug in my server configurations?

Thanks much,

Dave

[Dennis Handly]

printf("%s\n",sa.sin_addr.s_addr);
>obvious bug in my C program

If you used a real C compiler it would have told you the format was wrong:
warning #2181-D: argument is incompatible with corresponding format string conversion

The right format is %lx in 32 bit mode and %x in 64 bit mode.

[Padma Asrani]

Hi David,

Sorry for coming late on this. Is it okay with you to get me the core file and snmpd.conf file . I can try to get you the c program which you can run on your system.

Padma

[David Knickmeyer]

Yep, that was an obvious bug. Sorry about that.

The revised program returns 'ffffff' with the nodename and 'a5688e6b' with the IP (wasn't positive which one to use).

I've attached the core file. /etc/snmpd.conf file only has one line: trap-dest: xxxx.xxxx.com

[Dennis Handly]

>I've attached the core file.

Corefiles are useless if you are not on the original machine. You must use gdb's packcore so you capture ALL of the loadmodules used by the application.

All I get is:
ELF-32 core file - IA64 from 'snmpdm' - received SIGSEGV

But you can use gdb and get a stack trace:
$ gdb snmpdm core
(gdb) bt
(gdb) info reg
(gdb) disas $pc-16*8 $pc+16*4
(gdb) q

[David Knickmeyer]

Hi,

Sorry to be obtuse, but I can't find gdb for 11.23. I did find a website 'Merjin's HP-UX software for ITRC users' that says gdb is not supported under 11.23. Is this still true? If not, where can I find it (not hpux.cs.utah.edu, apparently) and if so, is there a different debugger available?

Thanks,

Dave

[Dennis Handly]

>but I can't find gdb for 11.23.

You should use HP's gdb:
http://www.hp.com/go/wdb

[David Knickmeyer]

Okay, finally got _that_ installed.

Ran commands listed above and attached the output. I'm sad to say it doesn't mean a whole lot to me...

Thanks again.

[Dennis Handly]

>I'm sad to say it doesn't mean a whole lot to me.

Not much to me either unless I look into it more or I'm able to run it in the debugger.

Perhaps it will be useful to Padma?

Basically what you have is a null pointer dereference in readSnmpdConf:
#0 0x4054b60:0 in readSnmpdConf+0x680
#1 0x4030690:0 in InitMaster+0xac0
#2 0x4029970:0 in main+0x550

: br.call.dptk.few b0=0x4054300;; HpTranslateAddress
: addp4 r8=0,r8
: ld4 r8=[r8]
: addp4 r9=5,r8;;
: st1 [r9]=r57 <<<<

The result from HpTranslateAddress contains a pointer to a pointer, which contains a null pointer, R8. 5 is being added to it, giving R9. And we die on the store.