[Bug runtime/14223] New: perf probes fail if not root

[Bug runtime/14223] New: perf probes fail if not root

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14223

             Bug #: 14223
           Summary: perf probes fail if not root
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: runtime
        AssignedTo: systemtap@sourceware.org
        ReportedBy: dsmith@redhat.com
    Classification: Unclassified


On rawhide (3.5.0-0.rc1.git0.1.fc18.x86_64).

As root, the following one-liner works fine:

====
# sudo /usr/local/bin/stap -ve 'probe perf.sw.cpu_clock { exit() }'
Pass 1: parsed user script and 85 library script(s) using
202020virt/23848res/2800shr/21660data kb, in 150usr/30sys/178real ms.
Pass 2: analyzed script: 1 probe(s), 1 function(s), 0 embed(s), 0 global(s)
using 202416virt/24640res/2968shr/22056data kb, in 0usr/0sys/6real ms.
Pass 3: using cached
/root/.systemtap/cache/73/stap_7333e571f12104c405ac9f21508f723b_902.c
Pass 4: using cached
/root/.systemtap/cache/73/stap_7333e571f12104c405ac9f21508f723b_902.ko
Pass 5: starting run.
Pass 5: run completed in 20usr/30sys/337real ms.
====

But, as myself, the perf probe registrations fails (even though I'm in
stapusr/stapdev):

====
# groups
dsmith wheel users stapusr stapdev
# /usr/local/bin/stap -ve 'probe perf.sw.cpu_clock { exit() }'
Pass 1: parsed user script and 85 library script(s) using
204136virt/24432res/2884shr/21676data kb, in 150usr/20sys/178real ms.
Pass 2: analyzed script: 1 probe(s), 1 function(s), 0 embed(s), 0 global(s)
using 204532virt/24960res/2956shr/22072data kb, in 10usr/0sys/6real ms.
Pass 3: using cached
/home/dsmith/.systemtap/cache/7d/stap_7d3ce5a812ac75c47b94040fadd2de76_905.c
Pass 4: using cached
/home/dsmith/.systemtap/cache/7d/stap_7d3ce5a812ac75c47b94040fadd2de76_905.ko
Pass 5: starting run.
ERROR: probe perf.type(1).config(0).sample(1000000) registration error (rc -13)
WARNING: /usr/local/bin/staprun exited with status: 1
Pass 5: run completed in 10usr/40sys/60real ms.
Pass 5: run failed.  Try again with another '--vp 00001' option.
====

From looking at the source and doing some debugging, it looks like the
following code in find_get_context() (called by
perf_event_create_kernel_counter()) is failing:

====
find_get_context(struct pmu *pmu, struct task_struct *task, int cpu)
{
    struct perf_event_context *ctx;
    struct perf_cpu_context *cpuctx;
    unsigned long flags;
    int ctxn, err;

    if (!task) {
        /* Must be root to operate on a CPU event: */
        if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
            return ERR_PTR(-EACCES);
====

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/14223] perf probes fail if not root

[fche at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14223

Frank Ch. Eigler <fche at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fche at redhat dot com

--- Comment #1 from Frank Ch. Eigler <fche at redhat dot com> 2012-06-12 15:37:08 UTC ---
Wacky.  It's strange for a deeper-down kernel api to enforce authorization, as
opposed to a close-to-userspace syscall handler, but alas.  We should
investigate whether this part of the module runs during a callback caused by
(setuid-root) staprun vs. (unprivileged) stapio that staprun execs.  If the
latter, we may be able to "fix" it by having some probe registration occur
earlier, perhaps with an additional transport message.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/14223] perf probes fail if not root

[jistone at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=14223

Josh Stone <jistone at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |jistone at redhat dot com
         Resolution|---                         |FIXED

--- Comment #2 from Josh Stone <jistone at redhat dot com> ---
commit ad6c8f90a60ee9c223e31e1d61c066d7c9a6a6b4

Rather than messing with early registration, I chose to use schedule_work() to
run this bit on an events/X kernel thread.

-- 
You are receiving this mail because:
You are the assignee for the bug.