Re: [BUG] perf record: --uid=x fails

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Mon, Jan 14, 2013 at 03:14:02PM +0100, Stephane Eranian escreveu:
> Hi,
> 
> I was trying to use the --uid option of perf record but it fails for
> me no matter
> what I tried. Looks like the goal of this option is to measure ALL the processes
> owned by the specified uid. Each process is measured in per-thread mode.
> 
> However for me it failed on all my attempts when running with 3.8.0-rc3 on
> Ubuntu Quantal.
> 
> $ perf record --uid=eranian sleep 4
> Error:
> Permission error - are you root?
> Consider tweaking /proc/sys/kernel/perf_event_paranoid:
>  -1 - Not paranoid at all
>   0 - Disallow raw tracepoint access for unpriv
>   1 - Disallow cpu events for unpriv
>   2 - Disallow kernel profiling for unpriv
> sleep: Terminated
> 
> You don't want to be root to run this command.  Should not require

But you have to due to the problem you noticed below.

> to measure the processes I own. So the error message is confusing
> here.

indeed
 
> After some debugging, I came to the conclusion that this command
> fails when it hits the sshd daemon:
> 
> eranian   2439  0.0  0.0 110424  1968 ?        S    15:07   0:00 sshd:
> eranian@pts/2
> root      2301  0.0  0.0 110424  4420 ?        Ss   15:07   0:00 sshd:
> eranian [priv]
> 
> I mean the sshd process owned by me. It is owned by me but I cannot attached
> an event to it. I get EACCES and I suspect it's because of missing
> ptrace privilege.
> The sshd binary is obviously not setuid. So there is something else preventing
> ptrace. In fact, even strace -p 2349 fails. Looked online and there
> were a couple
> of mentions to the Yama security model and the ptrace_scope sysctl
> control. I tried
> that and it did not help.
> 
> So looks to me that something is broken somewhere. If the kernel
> restrictions are
> normal, then I think perf record should warn that it has to skip the
> sshd process and
> continue with the other processes owned by me. That  seems more useful than the
> current situation. But I may be missing something here. If so please
> explain to me.

Right, I need to resume working on properly fixing this, I got
sidetracked when I got the same point as you in the above analysis :-\

Will try to recover the conversations I had about this.

- Arnaldo