Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230317100514.1373-1-laihangliang1@huawei.com>
In-Reply-To: <20230317100514.1373-1-laihangliang1@huawei.com>
From:   Namhyung Kim <namhyung@kernel.org>
Date:   Fri, 31 Mar 2023 17:11:01 -0700
Message-ID: <CAM9d7cjrkWdg18wTA_tRch8oN0YUEy3ZJ2mhrYtk23iLqd1XPw@mail.gmail.com>
Subject: Re: [PATCH] perf top: expand the range of multithreaded phase
To:     Hangliang Lai <laihangliang1@huawei.com>
Cc:     liuwenyu7@huawei.com, acme@kernel.org, adrian.hunter@intel.com,
        alexander.shishkin@linux.intel.com, brauner@kernel.org,
        hewenliang4@huawei.com, irogers@google.com, jolsa@kernel.org,
        linfeilong@huawei.com, linux-kernel@vger.kernel.org,
        linux-perf-users@vger.kernel.org, mark.rutland@arm.com,
        mingo@redhat.com, yeyunfeng@huawei.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
Precedence: bulk

Hello,

Sorry for the late reply.

On Fri, Mar 17, 2023 at 3:05 AM Hangliang Lai <laihangliang1@huawei.com> wrote:
>
> In __cmd_top, perf_set_multithreaded is used to enable pthread_rwlock, thus
> donw_read and down_write can work to handle concurrency problems. Then top
> use perf_set_singlethreaded and switch to single threaded phase, assuming
> that no thread concurrency will happen later.
>
> However, a UAF problem could occur in perf top in single threaded phase,
> The concurrent procedure is like this:
>
> display_thread                              process_thread
> --------------                              --------------
>
> thread__comm_len
>   -> thread__comm_str
>     -> __thread__comm_str(thread)
>                                             thread__delete
>                                              -> comm__free
>                                               -> comm_str__put
>                                                -> zfree(&cs->str)
>     -> thread->comm_len = strlen(comm);
>
> Since in single thread phase, perf_singlethreaded is true, down_read and
> down_write can not work to avoid concurrency problems.
>
> This patch put perf_set_singlethreaded to the function tail to expand the
> multithreaded phase range, make display_thread and process_thread run
> safe.

I think it should be unconditional as perf top is always multi-threaded.

>
> Signed-off-by: Hangliang Lai  <laihangliang1@huawei.com>
> Reported-by: Wenyu Liu <liuwenyu7@huawei.com>
> Reviewed-by: Yunfeng Ye <yeyunfeng@huawei.com>
> ---
>  tools/perf/builtin-top.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> index 7c6413447..74239940b 100644
> --- a/tools/perf/builtin-top.c
> +++ b/tools/perf/builtin-top.c
> @@ -1280,9 +1280,6 @@ static int __cmd_top(struct perf_top *top)
>                                     top->evlist->core.threads, false,
>                                     top->nr_threads_synthesize);
>
> -       if (top->nr_threads_synthesize > 1)
> -               perf_set_singlethreaded();

Instead, we can simply do

    perf_set_multithreaded();

If top->nr_threads_synthesize > 1, no effect.  If not, it turns
the switch on here.

> -
>         if (perf_hpp_list.socket) {
>                 ret = perf_env__read_cpu_topology_map(&perf_env);
>                 if (ret < 0) {
> @@ -1359,6 +1356,10 @@ out_join:
>  out_join_thread:
>         pthread_cond_signal(&top->qe.cond);
>         pthread_join(thread_process, NULL);
> +
> +       if (top->nr_threads_synthesize > 1)
> +               perf_set_singlethreaded();

And remove the condition here.

Thanks,
Namhyung


> +
>         return ret;
>  }
>
> --
> 2.33.0
>