[Orca-users] Segmentation Fault

[Cockcroft, Adrian]

The stdio definition of FILE is a byte for the file number for 32bit
APIs, when designing the 64bit APIs this structure was made opaque and
the file number is at least a four byte value.

On SPARC Solaris 8 or 10, the SE toolkit is running as a 64bit
application, so this problem cannot occur. On x86 Solaris 10 we could
still have this issue.

Adrian

-----Original Message-----
From: orca-users-bounces+acockcroft=ebay.com at orcaware.com
[mailto:orca-users-bounces+acockcroft=ebay.com at orcaware.com] On Behalf
Of David Michaels
Sent: Friday, November 04, 2005 3:48 PM
To: orca at bias.org
Cc: orca-users at orcaware.com
Subject: Re: [Orca-users] Segmentation Fault

I think this might be a 64-bit vs. 32-bit issue.  I seem to recall a 
similar issue with file handle limits when I was trying to get Samba to 
work on my Solaris 8 machine. I recompiled Samba to 64-bit, and the 
problem went away.  As I recall, the small file-handle limit on 32-bit 
Solaris is not something ulimit can exceed, even for root--it requires a

64-bit OS and a 64-bit application.

--Dragon


orca at bias.org wrote:

> Thanks for the suggestion.  I tried increasing ulimit, but that did 
> not solve.
>
> I do a lot of lun adding/removing luns (daily) on Solaris 8 and 
> Solaris 10 boxes.  I have not seen the issue at all on Solaris 8.  On 
> Solaris 10, I have a boxes with over a thousands luns that have never 
> changed since built and they do not have the issue.  I only see the 
> issue when I remove luns.  I can fix it by removing all the devices 
> and then rebuilding (tar) from cdrom or network boot.
>
> Regards,
> Liston
>
>
> On Fri, 4 Nov 2005, Cockcroft, Adrian wrote:
>
>> 250-ish sounds like a file descriptor limit issue
>>
>> Try increasing the limit in the shell and see if it fails at a
different
>> point, you can use ulimit -n to check and set the limit
>>
>> The opendir call uses a fd for each directory it opens, I'm not sure
why
>> this would be an issue for you and not others, perhaps your directory
>> structure is different somehow and the directories aren't being
closed
>> properly in the code, so this function would leak file descriptors.
>>
>> Adrian
>>
>> -----Original Message-----
>> From: orca-users-bounces+acockcroft=ebay.com at orcaware.com
>> [mailto:orca-users-bounces+acockcroft=ebay.com at orcaware.com] On
Behalf
>> Of orca at bias.org
>> Sent: Friday, November 04, 2005 8:23 AM
>> To: orca-users at orcaware.com
>> Subject: RE: [Orca-users] Segmentation Fault
>>
>> On Thu, 29 Sep 2005, Cockcroft, Adrian wrote:
>>
>>> 3)  the problem below seems to be that its reading a directory and
>>> getting a bad directory entry that is crashing it when it
dereferences
>>
>>
>>> the pointer. This is not the same as the common problem of running
out
>>
>>
>>> of disk slots. You san see that the current index is 30, and the
size
>>
>> is
>>
>>> 540 in the debug output below.
>>
>>
>> Found a workaround for Segmentation Fault(coredump) when trying to
start
>>
>> orcallator on system where I have removed luns previously there.
Doing
>> a
>> reconfiguration reboot did not resolve this issues.
>>
>> The issue appears to be with SE when doing:
>>
>>   setup_diskinfo_inst()
>>   in /opt/RICHPse/include/diskinfo.se
>>
>> I can test that the issue exist by running:
>>
>>   # /opt/RICHPse/bin/se disks.se
>>
>> The diskinfo.se creates a listing of diskinfo by searching /dev/dsk.
I
>> checking this directory for invalid entires and I could not find any.
>>
>> I copied all the links in /dev/dsk to /dev2/dsk and made the
following
>> change in /opt/RICHPse/include/diskinfo.se:
>>
>>   string disk_dirs[3] = { "/dev2/dsk", "/dev/osa/dev/dsk", nil };A
>>
>> One by one, I tested every link (~1800) individually, running se, and
>> then
>> moving it out.  All links works fine.  I received no Segmentation
Faults
>>
>> during this test.
>>
>> I then testing moving them in, running se, and NOT moving them out.
>>
>> When I get to about 250 directory entries, it starts to Segmentation
>> Fault.  I changed which 250 exist by manually selecting the order and
>> received the same result.
>>
>> I don't need stats on all these disk so I just picked the one that
are
>> most critical and never change to be in the directory /dev2/dsk and
am
>> now
>> able to start orcallator.
>>
>> If this helps identify a more global solution, please let me know.
>>
>> Regards,
>> Liston
>> _______________________________________________
>> Orca-users mailing list
>> Orca-users at orcaware.com
>> http://www.orcaware.com/mailman/listinfo/orca-users
>>
>
> _______________________________________________
> Orca-users mailing list
> Orca-users at orcaware.com
> http://www.orcaware.com/mailman/listinfo/orca-users


_______________________________________________
Orca-users mailing list
Orca-users at orcaware.com
http://www.orcaware.com/mailman/listinfo/orca-users