Re: [PATCH] tracer for sys_open() - sreadahead

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

Em Wed, Jan 28, 2009 at 02:05:20PM -0800, Kok, Auke escreveu:
> Frederic Weisbecker wrote:
> > On Tue, Jan 27, 2009 at 06:51:37PM -0200, Arnaldo Carvalho de Melo wrote:
> >> Em Tue, Jan 27, 2009 at 12:08:04PM -0800, Kok, Auke escreveu:
> >>> This tracer monitors regular file open() syscalls. This is a fast
> >>> and low-overhead alternative to strace, and does not allow or
> >>> require to be attached to every process.
> >>>
> >>> The tracer only logs succesfull calls, as those are the only ones we
> >>> are currently interested in, and we can determine the absolute path
> >>> of these files as we log.
> >>>
> >>> Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
> >> <SNIP>
> >>  
> >>> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> >>> index 4d3d381..24c17d2 100644
> >>> --- a/kernel/trace/trace.h
> >>> +++ b/kernel/trace/trace.h
> >>> @@ -30,6 +30,7 @@ enum trace_type {
> >>>         TRACE_USER_STACK,
> >>>         TRACE_HW_BRANCHES,
> >>>         TRACE_POWER,
> >>> +       TRACE_OPEN,
> >> Why not a TRACE_VFS or TRACE_SYSCALL and then multiplex there open,
> >> close, etc? trace_assign_type will get humongous in no time this way.
> > 
> > 
> > That's what I was about to answer too.
> > That would be sad to find one tracer for open, one for read, one for write...
> > 
> >  
> >> TRACE_BLK does multiplexing, as does TRACE_PROCESS that Frank just
> >> posted too, and I'm working on a plugin for the patch that Neil
> >> submitted for the socket layer that also will need an entry there for
> >> its tracepoints.
> >>
> >> Frederic, it seems that discussion about subtypes has to continue :-)
> > 
> > 
> > Indeed, that becomes serious :-)
> > 
> 
> I totally agree - this tracer was purely made quick-and-dirty to get sreadahead to
> the next level. The in-kernel syscall trace facility is extremely potent and I
> don't doubt that my patch may have been found slightly absurd by some :)

hey, hey, your patch just illustrated that we need a subtype facility.

I'm (as all the other people in some shape or form involved in pushing
some sort of common tracing infrastructure into the kernel, I guess)
excited about more developers using what is being put in place :-)
 
> In all it was not productive to write a giant trace facility just to get
> sreadahead one tracer type. Now that the code works and the benefit is proven, I'm
> happy to see if we can work on making a decent long-term non-specialistic solution.

That is the spirit!

- Arnaldo