Page 1 of 1

stability issues

PostPosted: Wed Nov 06, 2002 5:37 pm
by ryan
Im experiencing issues of multiple servers , with different specs on each (some SMP, some single proc, some scsi, some ide....)...

Anyways im seeing servers crash daily, and was wondering if it is at all possible to be caused from gradm or grsecurity as a whole ?

Also just before when i went to re-enable gradm on a system, it crashed.
No errors of note , the boxes just crash.

Using current release of grsec (1.9.7d ?) and current gradm and such...

PostPosted: Wed Nov 06, 2002 8:11 pm
by spender
On your UP machines, do you have SMP support compiled in anyway? I believe I've seen your problem on my system, and I've heard reports from one other person of the same problem. In their case they were using SMP also. I believe the problem is due to locking, and I have reason to believe PaX is the culprit. If you could tell us what was going on on the system when it crashed, that would be helpful. Was it under heavy load? Can you duplicate the bug?

-Brad

PostPosted: Fri Nov 08, 2002 1:22 am
by ryan
I like clean and simple kernels, My smp systems have SMP and applicable config's and my non-smp are not compiled with SMP (i dont know why people leave SMP enabled for UP :P).

When yous had the 1.9.6 releases out or whatever that had the openwall and pax BOF protection methods clearly defined. I never had stability issues (opted always with openwall). Ever since the sorta merg or whatever, things have acted flacky for me.

My servers on typical before the crashes are doing nothing more than serving webpages or media streams. Iv discounted hardware problems as the culprit.

PostPosted: Fri Nov 08, 2002 5:47 am
by PaX Team
ryan wrote:When yous had the 1.9.6 releases out or whatever that had the openwall and pax BOF protection methods clearly defined. I never had stability issues (opted always with openwall). Ever since the sorta merg or whatever, things have acted flacky for me.

My servers on typical before the crashes are doing nothing more than serving webpages or media streams. Iv discounted hardware problems as the culprit.

There is indeed a bug we have yet to find, most likely somewhere in the vma mirroring code (which is new and the most complex part). It would be helpful if you could tell us which combination of PAGEEXEC/SEGMEXEC/RANDEXEC you use in both the kernel config and on your executables.