[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20190319093122.70bfc751@gandalf.local.home>
Date: Tue, 19 Mar 2019 09:31:22 -0400
From: Steven Rostedt <rostedt@...dmis.org>
To: Prateek Sood <prsood@...eaurora.org>
Cc: mingo@...hat.com, linux-kernel@...r.kernel.org,
sramana@...eaurora.org, fweisbec@...il.com, jolsa@...hat.com
Subject: Re: [PATCH] perf: fix use after free of perf_trace_buf
On Tue, 19 Mar 2019 17:51:00 +0530
Prateek Sood <prsood@...eaurora.org> wrote:
> SyS_perf_event_open()
> free_event()
> _free_event()
> tp_perf_event_destroy()
> perf_trace_destroy()
> perf_trace_event_unreg() //free perf_trace_buf
>
> trace_cpu_frequency()
> perf_trace_cpu()
> perf_trace_buf_alloc() //access perf_trace_buf
>
> CPU0 CPU1
> perf_trace_event_unreg() perf_trace_cpu()
> head = (event_call->perf_events)
>
But here there's a:
tracepoint_synchronize_unregister();
Which is suppose to prevent this. Are you saying that the
tracepoint_synchronize_unregister() is broken?
-- Steve
> free_percpu(tp_event->perf_events)
> tp_event->perf_events = NULL
> --total_ref_count
> free_percpu(perf_trace_buf[i])
> perf_trace_buf[i] = NULL
>
> raw_data = perf_trace_buf[rctx]
> memset(raw_data)
>
> A potential race exists between access of perf_trace_buf from
> perf_trace_buf_alloc() and perf_trace_event_unreg(). This can
> result in perf_trace_buf[rctx] being NULL during access from memset()
> in perf_trace_buf_alloc().
>
Powered by blists - more mailing lists