perf inject jit: Ignore memfd mmap events if jitdump present

[PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Brian Robbins 3 years, 8 months ago

Some processes store jitted code in memfd mappings to avoid having rwx
mappings.  These processes map the code with a writeable mapping and a
read-execute mapping.  They write the code using the writeable mapping
and then unmap the writeable mapping.  All subsequent execution is
through the read-execute mapping.

perf inject --jit ignores //anon* mappings for each process where a
jitdump is present because it expects to inject mmap events for each
jitted code range, and said jitted code ranges will overlap with the
//anon* mappings.

Ignore /memfd: mappings so that jitted code contained in /memfd:
mappings is treated the same way as jitted code contained in //anon*
mappings.

Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>
---
 tools/perf/util/jitdump.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
index a23255773c60..335a3c61940b 100644
--- a/tools/perf/util/jitdump.c
+++ b/tools/perf/util/jitdump.c
@@ -845,8 +845,12 @@ jit_process(struct perf_session *session,
 	if (jit_detect(filename, pid, nsi)) {
 		nsinfo__put(nsi);
 
-		// Strip //anon* mmaps if we processed a jitdump for this pid
-		if (jit_has_pid(machine, pid) && (strncmp(filename, "//anon", 6) == 0))
+		/*
+		 * Strip //anon* and /memfd:* mmaps if we processed a jitdump for this pid
+		 */
+		if (jit_has_pid(machine, pid) &&
+			((strncmp(filename, "//anon", 6) == 0) ||
+			 (strncmp(filename, "/memfd:", 7) == 0))
 			return 1;
 
 		return 0;
-- 
2.25.1

Re: [PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Ian Rogers 3 years, 8 months ago

On Tue, Aug 2, 2022 at 11:25 AM Brian Robbins
<brianrob@linux.microsoft.com> wrote:
>
> Some processes store jitted code in memfd mappings to avoid having rwx
> mappings.  These processes map the code with a writeable mapping and a
> read-execute mapping.  They write the code using the writeable mapping
> and then unmap the writeable mapping.  All subsequent execution is
> through the read-execute mapping.
>
> perf inject --jit ignores //anon* mappings for each process where a
> jitdump is present because it expects to inject mmap events for each
> jitted code range, and said jitted code ranges will overlap with the
> //anon* mappings.
>
> Ignore /memfd: mappings so that jitted code contained in /memfd:
> mappings is treated the same way as jitted code contained in //anon*
> mappings.
>
> Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>

Acked-by: Ian Rogers <irogers@google.com>

> ---
>  tools/perf/util/jitdump.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
> index a23255773c60..335a3c61940b 100644
> --- a/tools/perf/util/jitdump.c
> +++ b/tools/perf/util/jitdump.c
> @@ -845,8 +845,12 @@ jit_process(struct perf_session *session,
>         if (jit_detect(filename, pid, nsi)) {
>                 nsinfo__put(nsi);
>
> -               // Strip //anon* mmaps if we processed a jitdump for this pid
> -               if (jit_has_pid(machine, pid) && (strncmp(filename, "//anon", 6) == 0))
> +               /*
> +                * Strip //anon* and /memfd:* mmaps if we processed a jitdump for this pid
> +                */
> +               if (jit_has_pid(machine, pid) &&
> +                       ((strncmp(filename, "//anon", 6) == 0) ||
> +                        (strncmp(filename, "/memfd:", 7) == 0))

Related to this there is the prctl PR_SET_VMA_ANON_NAME which will
name mapping to start with "[anon:"
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/filesystems/proc.rst#n434
I wonder also we should be checking the pages are executable.

Thanks,
Ian

>                         return 1;
>
>                 return 0;
> --
> 2.25.1
>

Re: [PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Arnaldo Carvalho de Melo 3 years, 8 months ago

Em Thu, Aug 04, 2022 at 08:22:14AM -0700, Ian Rogers escreveu:
> On Tue, Aug 2, 2022 at 11:25 AM Brian Robbins
> <brianrob@linux.microsoft.com> wrote:
> >
> > Some processes store jitted code in memfd mappings to avoid having rwx
> > mappings.  These processes map the code with a writeable mapping and a
> > read-execute mapping.  They write the code using the writeable mapping
> > and then unmap the writeable mapping.  All subsequent execution is
> > through the read-execute mapping.
> >
> > perf inject --jit ignores //anon* mappings for each process where a
> > jitdump is present because it expects to inject mmap events for each
> > jitted code range, and said jitted code ranges will overlap with the
> > //anon* mappings.
> >
> > Ignore /memfd: mappings so that jitted code contained in /memfd:
> > mappings is treated the same way as jitted code contained in //anon*
> > mappings.
> >
> > Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>
> 
> Acked-by: Ian Rogers <irogers@google.com>

  CC      /tmp/build/perf/util/jitdump.o
  CC      /tmp/build/perf/pmu-events/pmu-events.o
  LD      /tmp/build/perf/pmu-events/pmu-events-in.o
util/jitdump.c: In function ‘jit_process’:
util/jitdump.c:853:65: error: expected ‘)’ before ‘return’
  853 |                          (strncmp(filename, "/memfd:", 7) == 0))
      |                                                                 ^
      |                                                                 )
  854 |                         return 1;
      |                         ~~~~~~
util/jitdump.c:851:20: note: to match this ‘(’
  851 |                 if (jit_has_pid(machine, pid) &&
      |                    ^
util/jitdump.c:857:9: error: expected expression before ‘}’ token
  857 |         }
      |         ^
make[4]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:96: /tmp/build/perf/util/jitdump.o] Error 1
make[4]: *** Waiting for unfinished jobs....
make[3]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:139: util] Error 2


Trying to fix now.

- Arnaldo
 
> > ---
> >  tools/perf/util/jitdump.c | 8 ++++++--
> >  1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
> > index a23255773c60..335a3c61940b 100644
> > --- a/tools/perf/util/jitdump.c
> > +++ b/tools/perf/util/jitdump.c
> > @@ -845,8 +845,12 @@ jit_process(struct perf_session *session,
> >         if (jit_detect(filename, pid, nsi)) {
> >                 nsinfo__put(nsi);
> >
> > -               // Strip //anon* mmaps if we processed a jitdump for this pid
> > -               if (jit_has_pid(machine, pid) && (strncmp(filename, "//anon", 6) == 0))
> > +               /*
> > +                * Strip //anon* and /memfd:* mmaps if we processed a jitdump for this pid
> > +                */
> > +               if (jit_has_pid(machine, pid) &&
> > +                       ((strncmp(filename, "//anon", 6) == 0) ||
> > +                        (strncmp(filename, "/memfd:", 7) == 0))
> 
> Related to this there is the prctl PR_SET_VMA_ANON_NAME which will
> name mapping to start with "[anon:"
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/filesystems/proc.rst#n434
> I wonder also we should be checking the pages are executable.
> 
> Thanks,
> Ian
> 
> >                         return 1;
> >
> >                 return 0;
> > --
> > 2.25.1
> >

-- 

- Arnaldo

Re: [PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Arnaldo Carvalho de Melo 3 years, 8 months ago

Em Wed, Aug 10, 2022 at 10:00:14AM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Thu, Aug 04, 2022 at 08:22:14AM -0700, Ian Rogers escreveu:
> > On Tue, Aug 2, 2022 at 11:25 AM Brian Robbins <brianrob@linux.microsoft.com> wrote:
> > > Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>
> > 
> > Acked-by: Ian Rogers <irogers@google.com>
> 
>   CC      /tmp/build/perf/util/jitdump.o
>   CC      /tmp/build/perf/pmu-events/pmu-events.o
>   LD      /tmp/build/perf/pmu-events/pmu-events-in.o
> util/jitdump.c: In function ‘jit_process’:
> util/jitdump.c:853:65: error: expected ‘)’ before ‘return’
>   853 |                          (strncmp(filename, "/memfd:", 7) == 0))
>       |                                                                 ^
>       |                                                                 )
>   854 |                         return 1;
>       |                         ~~~~~~
> util/jitdump.c:851:20: note: to match this ‘(’
>   851 |                 if (jit_has_pid(machine, pid) &&
>       |                    ^
> util/jitdump.c:857:9: error: expected expression before ‘}’ token
>   857 |         }
>       |         ^
> make[4]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:96: /tmp/build/perf/util/jitdump.o] Error 1
> make[4]: *** Waiting for unfinished jobs....
> make[3]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:139: util] Error 2
> 
> 
> Trying to fix now.

Trivial, please compile test it...

- Arnaldo

diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
index 335a3c61940b7b39..9983d21a5c42d70e 100644
--- a/tools/perf/util/jitdump.c
+++ b/tools/perf/util/jitdump.c
@@ -850,7 +850,7 @@ jit_process(struct perf_session *session,
 		 */
 		if (jit_has_pid(machine, pid) &&
 			((strncmp(filename, "//anon", 6) == 0) ||
-			 (strncmp(filename, "/memfd:", 7) == 0))
+			 (strncmp(filename, "/memfd:", 7) == 0)))
 			return 1;
 
 		return 0;

Re: [PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Brian Robbins 3 years, 8 months ago

On Wed, Aug 10, 2022 at 10:03:17AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, Aug 10, 2022 at 10:00:14AM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Thu, Aug 04, 2022 at 08:22:14AM -0700, Ian Rogers escreveu:
> > > On Tue, Aug 2, 2022 at 11:25 AM Brian Robbins <brianrob@linux.microsoft.com> wrote:
> > > > Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>
> > > 
> > > Acked-by: Ian Rogers <irogers@google.com>
> > 
> >   CC      /tmp/build/perf/util/jitdump.o
> >   CC      /tmp/build/perf/pmu-events/pmu-events.o
> >   LD      /tmp/build/perf/pmu-events/pmu-events-in.o
> > util/jitdump.c: In function ‘jit_process’:
> > util/jitdump.c:853:65: error: expected ‘)’ before ‘return’
> >   853 |                          (strncmp(filename, "/memfd:", 7) == 0))
> >       |                                                                 ^
> >       |                                                                 )
> >   854 |                         return 1;
> >       |                         ~~~~~~
> > util/jitdump.c:851:20: note: to match this ‘(’
> >   851 |                 if (jit_has_pid(machine, pid) &&
> >       |                    ^
> > util/jitdump.c:857:9: error: expected expression before ‘}’ token
> >   857 |         }
> >       |         ^
> > make[4]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:96: /tmp/build/perf/util/jitdump.o] Error 1
> > make[4]: *** Waiting for unfinished jobs....
> > make[3]: *** [/var/home/acme/git/perf/tools/build/Makefile.build:139: util] Error 2
> > 
> > 
> > Trying to fix now.
> 
> Trivial, please compile test it...
> 
> - Arnaldo
> 
> diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
> index 335a3c61940b7b39..9983d21a5c42d70e 100644
> --- a/tools/perf/util/jitdump.c
> +++ b/tools/perf/util/jitdump.c
> @@ -850,7 +850,7 @@ jit_process(struct perf_session *session,
>  		 */
>  		if (jit_has_pid(machine, pid) &&
>  			((strncmp(filename, "//anon", 6) == 0) ||
> -			 (strncmp(filename, "/memfd:", 7) == 0))
> +			 (strncmp(filename, "/memfd:", 7) == 0)))
>  			return 1;
>  
>  		return 0;

Apologies about that.  Fixed this in v2.

--Brian

Re: [PATCH] perf inject jit: Ignore memfd mmap events if jitdump present

Posted by Brian Robbins 3 years, 8 months ago

On Thu, Aug 04, 2022 at 08:22:14AM -0700, Ian Rogers wrote:
> On Tue, Aug 2, 2022 at 11:25 AM Brian Robbins
> <brianrob@linux.microsoft.com> wrote:
> >
> > Some processes store jitted code in memfd mappings to avoid having rwx
> > mappings.  These processes map the code with a writeable mapping and a
> > read-execute mapping.  They write the code using the writeable mapping
> > and then unmap the writeable mapping.  All subsequent execution is
> > through the read-execute mapping.
> >
> > perf inject --jit ignores //anon* mappings for each process where a
> > jitdump is present because it expects to inject mmap events for each
> > jitted code range, and said jitted code ranges will overlap with the
> > //anon* mappings.
> >
> > Ignore /memfd: mappings so that jitted code contained in /memfd:
> > mappings is treated the same way as jitted code contained in //anon*
> > mappings.
> >
> > Signed-off-by: Brian Robbins <brianrob@linux.microsoft.com>
> 
> Acked-by: Ian Rogers <irogers@google.com>
> 
> > ---
> >  tools/perf/util/jitdump.c | 8 ++++++--
> >  1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
> > index a23255773c60..335a3c61940b 100644
> > --- a/tools/perf/util/jitdump.c
> > +++ b/tools/perf/util/jitdump.c
> > @@ -845,8 +845,12 @@ jit_process(struct perf_session *session,
> >         if (jit_detect(filename, pid, nsi)) {
> >                 nsinfo__put(nsi);
> >
> > -               // Strip //anon* mmaps if we processed a jitdump for this pid
> > -               if (jit_has_pid(machine, pid) && (strncmp(filename, "//anon", 6) == 0))
> > +               /*
> > +                * Strip //anon* and /memfd:* mmaps if we processed a jitdump for this pid
> > +                */
> > +               if (jit_has_pid(machine, pid) &&
> > +                       ((strncmp(filename, "//anon", 6) == 0) ||
> > +                        (strncmp(filename, "/memfd:", 7) == 0))
> 
> Related to this there is the prctl PR_SET_VMA_ANON_NAME which will
> name mapping to start with "[anon:"
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/filesystems/proc.rst#n434
> I wonder also we should be checking the pages are executable.
> 
> Thanks,
> Ian
> 
> >                         return 1;
> >
> >                 return 0;
> > --
> > 2.25.1
> >

I have not run into this case yet, but I suspect you are right that this should be handled as well.  I can create a follow-up patch for this.

Thanks.
--Brian