[FFmpeg-devel] [PATCH] avformat/rtsp: fix infinite loop with udp transport

[Andriy Gelman]

On Sun, 30. Aug 11:50, Zhao Zhili wrote:
> 
> 
> > On Aug 30, 2020, at 1:45 AM, Andriy Gelman <andriy.gelman at gmail.com> wrote:
> > 
> > Hi Zhao,
> > 
> > On Tue, 25. Aug 01:17, Zhao Zhili wrote:
> >> Ping again.
> >> 
> >>> On Aug 17, 2020, at 8:05 AM, zhilizhao <quinkblack at foxmail.com> wrote:
> >>> 
> >>> Please help review the patch, thanks!
> >>> 
> >>>> On Aug 6, 2020, at 4:50 PM, quinkblack at foxmail.com wrote:
> >>>> 
> >>>> From: Zhao Zhili <quinkblack at foxmail.com>
> >>>> 
> >>>> User report: http://ffmpeg.org/pipermail/ffmpeg-user/2020-August/049494.html
> >>>> 
> >>>> server:
> >>>> ./ffmpeg -i test.mp4 -c copy -f rtsp -rtsp_transport udp  rtsp://localhost:12345/live.sdp
> >>>> 
> >>>> client:
> >>>> ./ffmpeg_g -y -rtsp_flags listen -timeout 100 -i rtsp://localhost:12345/live.sdp -c copy test.mp4
> >>>> ---
> >>>> libavformat/rtsp.c | 5 +++++
> >>>> 1 file changed, 5 insertions(+)
> >>>> 
> >>>> diff --git a/libavformat/rtsp.c b/libavformat/rtsp.c
> >>>> index 5d8491b74b..0fb9fde6b4 100644
> >>>> --- a/libavformat/rtsp.c
> >>>> +++ b/libavformat/rtsp.c
> >>>> @@ -2051,6 +2051,11 @@ static int udp_read_packet(AVFormatContext *s, RTSPStream **prtsp_st,
> >>>>               if ((ret = parse_rtsp_message(s)) < 0) {
> >>>>                   return ret;
> >>>>               }
> >>>> +                /* Since there is no way to detect eof of udp streams, break
> >>>> +                 * the loop in teardown state to prevent run into infinite.
> >>>> +                 */
> >>>> +                if (rt->state == RTSP_STATE_IDLE)
> >>>> +                    return AVERROR_EOF;
> >>>>           }
> >>>> #endif
> >>>>       } else if (n == 0 && ++timeout_cnt >= MAX_TIMEOUTS) {
> > 
> > Currently when the stream finishes a BYE message is not sent from the server.  
> > 
> > You can ensure that the message is sent by adding the send_bye option with the
> > flag:
> > "-rtpflags send_bye"
> > 
> > This will also fix the problem.
> 
> Thanks for your reply. It's a clean solution from the viewpoint of FFmpeg users.
> However, I think FFmpeg should handle the case without BYE packets, even
> without RTCP. The busy loop leads to high cpu usage. I'm not saying the patch
> is the right solution, I'd like to hear other suggestions to fix the problem on the
> rtsp receiver's side.
> 

The server does send a teardown message when it exits. Currently this sets 
the client state to rt->state = RTSP_STATE_IDLE.

Could adding RTSP_STATE_TEARDOWN to RTSPClientState enum be better? You could
then return AVERROR_EOF if the client is in the teardown state.

-- 
Andriy