Re: [PATCH] capabilities: audit capability use

[Topi Miettinen <toiwoton-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

On 07/12/16 14:59, Tejun Heo wrote:
> On Mon, Jul 11, 2016 at 07:47:44PM +0000, Topi Miettinen wrote:
>> It's really critical to be able to associate a task in the logs to
>> cgroups which were valid that time. Or can we infer somehow what cgroups
> 
> When is "that time"?  Without logging all operations, this is
> meaningless.
> 
>> a task was taking part, long time after task exit? Perhaps task cgroup
>> membership changes and changes in available cgroups should be logged too?
>>
>> Some kind of cgroup IDs could be logged instead of long paths. Then
>> these IDs should be reliably resolvable to paths offline somehow.
> 
> I don't think that's doable.  That pretty much requires the kernel to
> remember paths of all past cgroups.

That's a show stopper for audit approach for getting helpful information
for configuration. I'll try something different, probably cgroupstats.

-Topi

> 
>> How usual migrations between cgroups are? Why would a task ever move
>> from (say) systemd/system.slice/smartd.service to anywhere else?
> 
> In most cases, they won't move once set up initially but that's not
> the point of audit subsystem.  Logging this once one exit isn't gonna
> help anything for auditing the system.
> 
> Thanks.
>