Re: [PATCH] perf parse-events: fix 3 use after frees

[Ian Rogers]

On Sat, Mar 14, 2020 at 6:59 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> On Fri, Mar 13, 2020 at 04:02:49PM -0700, Ian Rogers wrote:
> > Reproducible with a clang asan build and then running perf test in
> > particular 'Parse event definition strings'.
> >
> > Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> > ---
> >  tools/perf/util/parse-events.c | 6 +++---
> >  1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
> > index a14995835d85..593b6b03785d 100644
> > --- a/tools/perf/util/parse-events.c
> > +++ b/tools/perf/util/parse-events.c
> > @@ -1449,7 +1449,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
> >               evsel = __add_event(list, &parse_state->idx, &attr, NULL, pmu, NULL,
> >                                   auto_merge_stats, NULL);
> >               if (evsel) {
> > -                     evsel->pmu_name = name;
> > +                     evsel->pmu_name = name ? strdup(name) : NULL;
> >                       evsel->use_uncore_alias = use_uncore_alias;
> >                       return 0;
> >               } else {
> > @@ -1497,7 +1497,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
> >               evsel->snapshot = info.snapshot;
> >               evsel->metric_expr = info.metric_expr;
> >               evsel->metric_name = info.metric_name;
> > -             evsel->pmu_name = name;
> > +             evsel->pmu_name = name ? strdup(name) : NULL;
>
> so it's pmu->name pointer.. does pmu get destroyed before the evsel?
> also should we free that then like below?
>
> >               evsel->use_uncore_alias = use_uncore_alias;
> >               evsel->percore = config_term_percore(&evsel->config_terms);
> >       }
> > @@ -1547,7 +1547,7 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,
> >                               if (!parse_events_add_pmu(parse_state, list,
> >                                                         pmu->name, head,
> >                                                         true, true)) {
> > -                                     pr_debug("%s -> %s/%s/\n", config,
> > +                                     pr_debug("%s -> %s/%s/\n", str,
>
> nice catch ;-)
>
> >                                                pmu->name, alias->str);
> >                                       ok++;
> >                               }
> > --
> > 2.25.1.481.gfbce0eb801-goog
> >
>
> thanks,
> jirka
>
>
> ---
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 816d930d774e..15ccd193483f 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -1287,6 +1287,7 @@ void perf_evsel__exit(struct evsel *evsel)
>         perf_thread_map__put(evsel->core.threads);
>         zfree(&evsel->group_name);
>         zfree(&evsel->name);
> +       zfree(&evsel->pmu_name);
>         perf_evsel__object.fini(evsel);
>  }

Agreed on the zfree! I'll upload a v2 with it. I've tested with
address sanitizer in case of double frees, etc. I haven't yet fuzz
tested the latest version of parse events, libfuzz detects memory
leaks between iterations so I expect there will be future small
patches like this coming.

Thanks,
Ian