Subject: Re: [vserver] Patch for 2.6.38.4 vserver + GR Security
From:Sandino Araico Sánchez <sandino@sandino.net>
Date: Sat, 14 May 2011 03:01:12 -0500
Sat, 14 May 2011 03:01:12 -0500
After the refcount fix (thanks Rik for your advice) I have the new 
kernel running for more than 23 hours with no visible failure.

The vserver patch against a grsec-patched kernel:
http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-against-grsec-2.2.2-201104232142-KB2-unstable.diff

The combined patch against 2.6.38.4 vanilla:
http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-grsec-2.2.2-201104232142-KB2-unstable.diff



On 12/05/11 06:33, Sandino Araico Sánchez wrote:
> On 12/05/11 05:34, Rik Bobbaers wrote:
>> this is the refcount overflow i was talking about.
> I have found your refcount patch.
> Most of the rejects happened because the hunk was already applied.
> The new patch is 
> http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-against-grsec-2.2.2-201104232142-KB2-unstable.diff
> As Herbert asked, the patch is vserver against a grsec-patched kernel.
>
> I have to go to sleep. I will try the new patch tomorrow.
>> I'll try to spin a new kernel this evening...
>>
>> (ps. what system calls are you talking about? i haven't done any 2.6.38
>> kernel yet, so it would be nice to know and not have to look ;)
> Merging parameters in function calls like 'var = 
> func(vx xxx(foo,grsec yyy(bar)));'
> A couple of those; the rest of the rejects are cleaner.
> Not too complicated.
>> KR
>>
>> Rik Bobbaers
>>
>> -- http://harry.enzoverder.be
>>
>>> I checked all the rejects and applied them manually. I had to merge a
>>> couple of function calls manually either because both vserver and grsec
>>> patches modified the same line.
>>>
>>> The kernel compiles and boots without crashing but I have found some
>>> functionality missing like the old token bucket sched, having to remove
>>> the sched directory from /etc/vservers/myvserver/sched
>>>
>>> Running the cherokee webserver inside a vserver with ~40 requests per
>>> second makes the kernel crash but I haven't had the time to try to
>>> reproduce the crash. The same webserver running inside a simple chroot
>>> does not crash the kernel.
>
>
> -- 
> Sandino Araico Sánchez
> http://sandino.net


-- 
Sandino Araico Sánchez
http://sandino.net



After the refcount fix (thanks Rik for your advice) I have the new kernel running for more than 23 hours with no visible failure.

The vserver patch against a grsec-patched kernel:
http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-against-grsec-2.2.2-201104232142-KB2-unstable.diff

The combined patch against 2.6.38.4 vanilla:
http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-grsec-2.2.2-201104232142-KB2-unstable.diff



On 12/05/11 06:33, Sandino Araico Sánchez wrote:
On 12/05/11 05:34, Rik Bobbaers wrote:
this is the refcount overflow i was talking about.
I have found your refcount patch. 
Most of the rejects happened because the hunk was already applied.
The new patch is http://sandino.araico.net/parches/vserver/patch-2.6.38.4-vs2.3.0.37-rc15-against-grsec-2.2.2-201104232142-KB2-unstable.diff
As Herbert asked, the patch is vserver against a grsec-patched kernel.

I have to go to sleep. I will try the new patch tomorrow.
I'll try to spin a new kernel this evening...

(ps. what system calls are you talking about? i haven't done any 2.6.38
kernel yet, so it would be nice to know and not have to look ;)
Merging parameters in function calls like 'var = func(vx_xxx(foo,grsec_yyy(bar)));'
A couple of those; the rest of the rejects are cleaner.
Not too complicated.
KR

Rik Bobbaers

-- http://harry.enzoverder.be

I checked all the rejects and applied them manually. I had to merge a
couple of function calls manually either because both vserver and grsec
patches modified the same line.

The kernel compiles and boots without crashing but I have found some
functionality missing like the old token bucket sched, having to remove
the sched directory from /etc/vservers/myvserver/sched

Running the cherokee webserver inside a vserver with ~40 requests per
second makes the kernel crash but I haven't had the time to try to
reproduce the crash. The same webserver running inside a simple chroot
does not crash the kernel.


-- 
Sandino Araico Sánchez 
http://sandino.net


-- 
Sandino Araico Sánchez 
http://sandino.net