Re: Crashes in ldap_unbind

re Solaris crash i cant tell anything from the stack posted, will need
a full stacktrace / corefile, preferably taken with DBG build. re Doze
and 4.1 try checking NSS and NSPR versions [especially NSPR] you have
there and if you can upgrade them to something a wee bit more recent.
we had many problems with old NSPR and Doze fibers older versions used
by default. obviously i cant bet this is your root cause here but it
does worth a check. if you have problems upgrading NSPR, have a look @
http://www.mozilla.org/projects/security/pki/jss/faq.html#SSL%20Hanging
nevermind JSS and hanging talk there, just note NSPR_NATIVE_THREADS_ONLY
env var and give it a try but make sure it is actually being set because
i personally found it is rather tricky on Doze.

Pradnyesh wrote:
> Hi Anton,
> 
> One more addition, we were encountering network connection problems to
> the directory server during the Solaris - ldap sdk 5.12 crash.
> 
> Thanks & Regards,
> Pradnyesh
> 
> Pradnyesh wrote:
>> Hi Anton,
>>
>> The ldap sdk  4.1 crash environment is happening with SSL. We haven't
>> been able to verify it on non-Windows platforms as of yet. Is there a
>> known problem with ldap sdk 4.1 ldap_unbind with SSL enabled on
>> Windows.
>>
>> On a separate note, we are also observing segamentation faults with
>> ldap sdk 5.12 on Solaris with ldap_unbind(). Has such a crash been
>> encountered before?
>>
>> -----------------  lwp# 8 / thread# 8  --------------------
>>  ff1d2c24 nsldapi_clear_from_cb_pollfds (69bfe40, 0, 1880e08, ff1343b8,
>> 16af468, 0) + 2c
>>  ff1d62ac nsldapi_free_connection (69bfe40, 20e0188, 0, 0, 1, 1) + 28
>>  ff1e0a48 ldap_ld_free (69bfe40, 0, 0, 1, 0, ff132870) + 210
>>  ff1e07e8 ldap_unbind (69bfe40, 0, 2050010, 5420000, 1, 1630468) + 10
>>
>> Thanks & Regards,
>> Pradnyesh
>>
>> Anton Bobrov wrote:
>>> even if say you know you hitting some know issue with 4.x it doesnt
>>> do you any good because you cant get any fixes for 4.x today, back
>>> ported or otherwise. with that said you might wanna check if :
>>>
>>> - the problem happens on Doze only.
>>> - you are doing SSL on that handle.
>>> - you can confirm that problem is
>>>    *not* reproducible with 5.x or 6.
>>>
>>> Pradnyesh wrote:
>>>> Hi All,
>>>>
>>>> We have a multi-threaded server that uses netscape ldap c sdk 4.1 in
>>>> it's older release. We are experiencing some random crashes, while
>>>> freeing memory, in the call to ldap_unbind() in our code. Due to its
>>>> random nature, We haven't been able to create a test case for the
>>>> issue so far.
>>>>
>>>> Our environment is as follows
>>>>
>>>> - The backend directory server is AD.NET 2003 and our server is
>>>> running on a Win 2003 box. AD.NET is by default configured to close
>>>> idle ldap connections every 15 mins.
>>>> - ldap sdk version is 4.1. [We have alread migrated to ldap sdk 5.X in
>>>> the later versions of our product.]
>>>> - Another thing that is hampering us is that there are no Windows
>>>> symbol files[pdb] available for ldap sdk 4.1, which prevents us from
>>>> capturing an exact stacktrace inside ldap_unbind.
>>>> - Our server creates a pool of ldap connections with separate
>>>> dedicated threads to create new connections as well as clean-up old
>>>> connections.
>>>> - multiple worker threads can share the same ldap connection handle at
>>>> the same time. e.g. you can have multiple ldap_simple_bind going
>>>> asynchronously over the same ldap connection handle at the same time.
>>>> - we have registered callback functions for createmutex, lockmutex,
>>>> unlockmutex, freemutex, getthreadid, getldaperror etc.
>>>> - we have confirmed via logging that we are not calling ldap_unbind
>>>> twice on the same ldap handle.
>>>>
>>>> Please let me know if anyone has any suggestions.
>>>>
>>>> I also noticed that there were already a couple of posts regarding
>>>> crashes in ldap_unbind. Did we have any resolutions for these?
>>>>
>>>> Heroux, Bernard R wrote on 3/30/2006
>>>>> Every so often the ldap_unbind_s will cause a crash.  This occurs even
>>>>> when checking the ldapHandle for NULL before calling the unbind.
>>>> Mike wrote on 3/2/2006
>>>>> I was reading through the documentation on how to handle failover or
>>>>> being disconnected from an LDAP server in my apps.  One method is to
>>>>> call ldap_unbind and reconnect and rebind, while the other method is set
>>>>> the LDAP_OPT_RECONNECT option and only rebind (i.e. don't call
>>>>> ldap_unbind).
>>>>>
>>>>> http://www.mozilla.org/directory/csdk-docs/using.htm#handle_failover
>>>>>
>>>>> We had been using the former of the two methods, but some of our apps
>>>>> were coring randomly during the ldap_unbind call
>>>> Thanks & Regards,
>>>> Pradnyesh
>>>> _______________________________________________
>>>> dev-tech-ldap mailing list
>>>> [email protected]
>>>> https://lists.mozilla.org/listinfo/dev-tech-ldap
> 
> _______________________________________________
> dev-tech-ldap mailing list
> [email protected]
> https://lists.mozilla.org/listinfo/dev-tech-ldap