You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@river.apache.org by Peter Firmstone <ji...@zeus.net.au> on 2011/08/19 05:16:05 UTC
monitoring security checks in a distributed environment
Getting security policies right in a distributed environment is
difficult, getting information back when things break in deployment is
also difficult.
What would be nice is a service you can utilise to give you a list of
failed security checks and the failed ProtectionDomain's.
ProtectionDomain isn't Serializable and neither is ClassLoader (although
PrefferedClassLoader overrides toString()).
But we could call ProtectionDomain.toString() and serialize that.
I thought about using RemoteEvent's, however an attacker could use this
for DOS attacks. Instead, each node could register a SecurityAudit
service with the lookup service. It would be up to an administrator
client to contact it.
Then you could use that service to periodically retrieve a list of
failed permission checks.
Or, we could log it? But logging doesn't seem to work too well from the
extension classloader...
But if we did use logging we could create a remote logger service, where
an administrator could inspect logs remotely...
Not sure...
Thoughts?
Regards,
Peter.
Re: monitoring security checks in a distributed environment
Posted by Peter <ji...@zeus.net.au>.
No me either, single point of failure subject to DOS.
I'm thinking each node provides the service, the admin client would need to know the service id.
Having an admin client means the thread in the node runs as the admin Subject.
Cheers,
Peter.
----- Original message -----
> Mmmm, so I'm not really a fan of centralised services to handle such
> stuff, preferring arrangements like this:
>
> (1) Permission issues are logged locally in each JVM.
> (2) We go find those JVM's and gather them up via some suitable
> service interface (found via e.g. LUS).
>
> Of course, if you're having security problems with LUS registration
> this won't be so helpful (although you'll know which service isn't
> registering and be able to tackle it there and then). Equally, if
> security is a concern, you mightn't be able to register with a logman
> either.
>
> On 20 August 2011 06:20, Peter Firmstone <ji...@zeus.net.au> wrote:
> > Gregg Wonderly wrote:
> > >
> > > On 8/18/2011 10:16 PM, Peter Firmstone wrote:
> > > >
> > > > Getting security policies right in a distributed environment is
> > > > difficult,
> > > > getting information back when things break in deployment is also
> > > > difficult.
> > > >
> > > > What would be nice is a service you can utilise to give you a list of
> > > > failed
> > > > security checks and the failed ProtectionDomain's.
> > > >
> > > > ProtectionDomain isn't Serializable and neither is ClassLoader (although
> > > > PrefferedClassLoader overrides toString()).
> > > >
> > > > But we could call ProtectionDomain.toString() and serialize that.
> > > >
> > > > I thought about using RemoteEvent's, however an attacker could use this
> > > > for DOS
> > > > attacks. Instead, each node could register a SecurityAudit service with
> > > > the
> > > > lookup service. It would be up to an administrator client to contact it.
> > > >
> > > > Then you could use that service to periodically retrieve a list of failed
> > > > permission checks.
> > > >
> > > > Or, we could log it? But logging doesn't seem to work too well from the
> > > > extension classloader...
> > > >
> > > > But if we did use logging we could create a remote logger service, where
> > > > an
> > > > administrator could inspect logs remotely...
> > >
> > > My logman.dev.java.net project provides this with command line activation.
> > > Just set the LogManager class, and the configuration file, and you get
> > > remotely accessible logging as well as configuration of log levels.
> > >
> > > Gregg
> > >
> > I like the sound of that Gregg, you've spent plenty of time dealing with
> > these issues.
> >
> > What are the security risks of logging failed permission checks?
> >
> > Does anyone want this feature?
> >
> > The big picture:
> >
> > You're having trouble with some services in your djinn, so you check your
> > client logs remotely, find you're missing some permissions (a list of the
> > permissions the proxy needs and those it has are listed), you update your
> > remote policy because you find your users don't have GrantPermission for a
> > the needed permission and the problem is fixed.
> >
> > Regards,
> >
> > Peter.
> >
Re: monitoring security checks in a distributed environment
Posted by Dan Creswell <da...@gmail.com>.
Mmmm, so I'm not really a fan of centralised services to handle such
stuff, preferring arrangements like this:
(1) Permission issues are logged locally in each JVM.
(2) We go find those JVM's and gather them up via some suitable
service interface (found via e.g. LUS).
Of course, if you're having security problems with LUS registration
this won't be so helpful (although you'll know which service isn't
registering and be able to tackle it there and then). Equally, if
security is a concern, you mightn't be able to register with a logman
either.
On 20 August 2011 06:20, Peter Firmstone <ji...@zeus.net.au> wrote:
> Gregg Wonderly wrote:
>>
>> On 8/18/2011 10:16 PM, Peter Firmstone wrote:
>>>
>>> Getting security policies right in a distributed environment is
>>> difficult,
>>> getting information back when things break in deployment is also
>>> difficult.
>>>
>>> What would be nice is a service you can utilise to give you a list of
>>> failed
>>> security checks and the failed ProtectionDomain's.
>>>
>>> ProtectionDomain isn't Serializable and neither is ClassLoader (although
>>> PrefferedClassLoader overrides toString()).
>>>
>>> But we could call ProtectionDomain.toString() and serialize that.
>>>
>>> I thought about using RemoteEvent's, however an attacker could use this
>>> for DOS
>>> attacks. Instead, each node could register a SecurityAudit service with
>>> the
>>> lookup service. It would be up to an administrator client to contact it.
>>>
>>> Then you could use that service to periodically retrieve a list of failed
>>> permission checks.
>>>
>>> Or, we could log it? But logging doesn't seem to work too well from the
>>> extension classloader...
>>>
>>> But if we did use logging we could create a remote logger service, where
>>> an
>>> administrator could inspect logs remotely...
>>
>> My logman.dev.java.net project provides this with command line activation.
>> Just set the LogManager class, and the configuration file, and you get
>> remotely accessible logging as well as configuration of log levels.
>>
>> Gregg
>>
> I like the sound of that Gregg, you've spent plenty of time dealing with
> these issues.
>
> What are the security risks of logging failed permission checks?
>
> Does anyone want this feature?
>
> The big picture:
>
> You're having trouble with some services in your djinn, so you check your
> client logs remotely, find you're missing some permissions (a list of the
> permissions the proxy needs and those it has are listed), you update your
> remote policy because you find your users don't have GrantPermission for a
> the needed permission and the problem is fixed.
>
> Regards,
>
> Peter.
>
Re: monitoring security checks in a distributed environment
Posted by Peter Firmstone <ji...@zeus.net.au>.
Gregg Wonderly wrote:
> On 8/18/2011 10:16 PM, Peter Firmstone wrote:
>> Getting security policies right in a distributed environment is
>> difficult,
>> getting information back when things break in deployment is also
>> difficult.
>>
>> What would be nice is a service you can utilise to give you a list of
>> failed
>> security checks and the failed ProtectionDomain's.
>>
>> ProtectionDomain isn't Serializable and neither is ClassLoader (although
>> PrefferedClassLoader overrides toString()).
>>
>> But we could call ProtectionDomain.toString() and serialize that.
>>
>> I thought about using RemoteEvent's, however an attacker could use
>> this for DOS
>> attacks. Instead, each node could register a SecurityAudit service
>> with the
>> lookup service. It would be up to an administrator client to contact it.
>>
>> Then you could use that service to periodically retrieve a list of
>> failed
>> permission checks.
>>
>> Or, we could log it? But logging doesn't seem to work too well from the
>> extension classloader...
>>
>> But if we did use logging we could create a remote logger service,
>> where an
>> administrator could inspect logs remotely...
>
> My logman.dev.java.net project provides this with command line
> activation. Just set the LogManager class, and the configuration
> file, and you get remotely accessible logging as well as configuration
> of log levels.
>
> Gregg
>
I like the sound of that Gregg, you've spent plenty of time dealing with
these issues.
What are the security risks of logging failed permission checks?
Does anyone want this feature?
The big picture:
You're having trouble with some services in your djinn, so you check
your client logs remotely, find you're missing some permissions (a list
of the permissions the proxy needs and those it has are listed), you
update your remote policy because you find your users don't have
GrantPermission for a the needed permission and the problem is fixed.
Regards,
Peter.
Re: monitoring security checks in a distributed environment
Posted by Gregg Wonderly <gr...@wonderly.org>.
On 8/18/2011 10:16 PM, Peter Firmstone wrote:
> Getting security policies right in a distributed environment is difficult,
> getting information back when things break in deployment is also difficult.
>
> What would be nice is a service you can utilise to give you a list of failed
> security checks and the failed ProtectionDomain's.
>
> ProtectionDomain isn't Serializable and neither is ClassLoader (although
> PrefferedClassLoader overrides toString()).
>
> But we could call ProtectionDomain.toString() and serialize that.
>
> I thought about using RemoteEvent's, however an attacker could use this for DOS
> attacks. Instead, each node could register a SecurityAudit service with the
> lookup service. It would be up to an administrator client to contact it.
>
> Then you could use that service to periodically retrieve a list of failed
> permission checks.
>
> Or, we could log it? But logging doesn't seem to work too well from the
> extension classloader...
>
> But if we did use logging we could create a remote logger service, where an
> administrator could inspect logs remotely...
My logman.dev.java.net project provides this with command line activation. Just
set the LogManager class, and the configuration file, and you get remotely
accessible logging as well as configuration of log levels.
Gregg