sshd crashing on IRIX (3.6.1p1)

[Kevin Taylor]

Darren Tucker wrote:
> Kevin Taylor wrote:
> 
>>Occasionally, we're noticing that sshd is core dumping on our IRIX
>>6.5.18f machine.
> 
> [snip]
> 
>>Any helpful thoughts?
> 
> 
> Is there anything (eg a "fatal") in wherever syslog is sending sshd's
> logs (possibly authlog)? 
> 
> A change recently went into sshd so the monitor will pass some signals
> on to its child, but SEGV isn't one of them.  Should the monitor shut
> down the child in this case too?  It probably shouldn't pass the SEGV
> through but maybe it should send a SIGTERM instead?
> 

I didn't see any fatals related to this particular crash, although we're 
only logging at the INFO level. Unfortunately, logging at a debug level 
might be more than we can handle.

I do like the 'feature' that the user is not getting logged out in this 
case, but like you mention, it's probably not what should properly 
happen in the code.

I've opened bugzilla case #574 on this. There's some more notes about 
things we've discovered in looking through the core dump. Apparently the 
system calls are listed in reverse order when looking through dbx, I 
didn't know that, so I was looking in the wrong place to start.

We basically think that the problem might be stemming from the 
verify_reverse_mapping functions. We noticed in the core files that on 
occasion, sshd was garbling the hostname that people were logging in 
from, and after we disabled VerifyReverseMapping in the sshd_config, the 
core dumps seemed to have subsided, however I'd like to be able to turn 
that back on in the future, because it's a useful function at times.