Re: [RFC][PATCH 0/4] tracing/kprobes/uprobes: Fix race between opening probe event files and deleting probe

From: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
To: Steven Rostedt <rostedt@goodmis.org>
Cc: linux-kernel@vger.kernel.org, Oleg Nesterov <oleg@redhat.com>,
	"zhangwei(Jovi)" <jovi.zhangwei@huawei.com>,
	Jiri Olsa <jolsa@redhat.com>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>,
	Arnaldo Carvalho de Melo <acme@ghostprotocols.net>,
	Srikar Dronamraju <srikar@linux.vnet.ibm.com>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	Ingo Molnar <mingo@redhat.com>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [RFC][PATCH 0/4] tracing/kprobes/uprobes: Fix race between opening probe event files and deleting probe
Date: Fri, 12 Jul 2013 22:09:50 +0900	[thread overview]
Message-ID: <51E0001E.3050108@hitachi.com> (raw)
In-Reply-To: <20130704033347.807661713@goodmis.org>

(2013/07/04 12:33), Steven Rostedt wrote:
> Currently there exists a race with deleting a kprobe or uprobe and
> a user opening the probe event file or using perf events.
> 
> The problem stems from not being able to take the probe_lock from the
> unregister code because we may have the event_mutex at the time, and
> the event mutex may be taken with the probe_lock held.
> 
> To solve this, the events get a ref count (using the flags field), where
> when an event file is opened, the ftrace_event_call ref count increments.
> Then this is checked under event_mutex and if set, the unregistering
> of the probe will fail.
> 
> Here's a test that shows how things break:
> 
>      # cd /sys/kernel/debug/tracing
>      # echo 'p:sigprocmask sigprocmask' > kprobe_events || exit -1
>      # enable_probe() {
>         sleep 10
>         echo 1
>      }
>      # file=events/kprobes/sigprocmask/enable
>      # enable_probe > $file &
>      > kprobe_events
>     
>     The above will corrupt the kprobe system, as the write to the enable
>     file will happen after the kprobe was deleted.
>     
>     Trying to create the probe again fails:
>     
>      # echo 'p:sigprocmask sigprocmask' > kprobe_events
>      # cat kprobe_events
>     p:kprobes/sigprocmask sigprocmask
>      # ls events/kprobes/
>     enable  filter
> 
> After applying these patches, the "> kprobe_events" fails due to the
> event being busy.
> 
> Masami, please review these patches and give your ack.

Steve, Oleg, could you also take a look on my additional 2 patches too?

I think both this series and my patches are required for
fixing most of the problem. (what we need is a patch to ensure
finishing all running uprobe handlers at disabling it *if needed*)

Thank you,

-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com