Re: [PATCH v3 3/4] perf: Use CAP_SYSLOG with kptr_restrict checks

[Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>]

Em Mon, Aug 19, 2019 at 10:22:07PM +0000, Lubashev, Igor escreveu:
> On Mon, August 19, 2019 at 12:51 PM Mathieu Poirier <mathieu.poirier@linaro.org> wrote:
> > On Thu, 15 Aug 2019 at 15:42, Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com> wrote:
> > Things are working properly on your perf/cap branch.  I tested with on both
> > x86 and ARM.
 
> Mathieu, you are probably testing with euid==0.  If you were to test
> with euid!=0 but with CAP_SYSLOG and no libcap (and kptr_restrict=0,
> perf_event_paranoid=2), you would likely hit the bug that you
> identified in __perf_event__synthesize_kermel_mmap().
 
> See https://lkml.kernel.org/lkml/930a59730c0d495f8c5acf4f99048e8d@usma1ex-dag1mb6.msg.corp.akamai.com for the fix (Option 1 only or Options 1+2).
> 
> Arnaldo, once we decide what the right fix is, I am happy to post the update (options 1, 1+2) as a patch series.

I think you should get the checks for ref_reloc_sym in place so as to
make the code overall more robust, and also go on continuing to make the
checks in tools/perf/ to match what is checked on the other side of the
mirror, i.e. by the kernel, so from a quick read, please put first the
robustness patches (check ref_reloc_sym) do your other suggestions and
update the warnings, then refresh the two patches that still are not in
my perf/core branch:

[acme@quaco perf]$ git rebase perf/core
First, rewinding head to replay your work on top of it...
Applying: perf tools: Use CAP_SYS_ADMIN with perf_event_paranoid checks
Applying: perf symbols: Use CAP_SYSLOG with kptr_restrict checks
[acme@quaco perf]$ 

I've pushed out perf/cap, so you can go from there as it is rebased on
my current perf/core.

Then test all these cases: with/without libcap, with euid==0 and
different than zero, with capabilities, etc, patch by patch so that we
don't break bisection nor regress,

Thanks and keep up the good work!

- Arnaldo
 
> - Igor
> 
> 
> > > > I am not sure how this can be fixed.  I counted a total of 19
> > > > instances where kmap->ref_reloc_sym->XYZ is called, only 2 of wich
> > > > care to check if kmap->ref_reloc_sym is valid before proceeding.  As
> > > > such I must hope that in the 17 other cases, kmap->ref_reloc_sym is
> > > > guaranteed to be valid.  If I am correct then all we need is to
> > > > check for a valid pointer in _perf_event__synthesize_kernel_mmap().
> > > > Otherwise it will be a little harder.
> > > >
> > > > Mathieu
> 

-- 

- Arnaldo