FFMPEG-rtsp-h264负载剥离

rtsp剥离负载，发送在连接建立以后，OPTION \DESCRIPT\SETUP\PLAY等方法已相继发送，udp(tcp) 数据传输通道、媒体信息已相应建立，ffmpeg通过注册回调函数read_packet开始读取发来的负载信息，并组装成AVPacket格式返回。

AVInputFormat ff_rtsp_demuxer = {
    .name           = "rtsp",
    .long_name      = NULL_IF_CONFIG_SMALL("RTSP input"),
    .priv_data_size = sizeof(RTSPState),
    .read_probe     = rtsp_probe,
    .read_header    = rtsp_read_header,
    .read_packet    = rtsp_read_packet,
    .read_close     = rtsp_read_close,
    .read_seek      = rtsp_read_seek,
    .flags          = AVFMT_NOFILE,
    .read_play      = rtsp_read_play,
    .read_pause     = rtsp_read_pause,
    .priv_class     = &rtsp_demuxer_class,
};

对应回调函数rtsp_read_packet，主要做了两件事1、通过套接字读取远端数据 2、解析数据剥离负载

一、读取远端数据

static int rtsp_read_packet(AVFormatContext *s, AVPacket *pkt)
{
    RTSPState *rt = s->priv_data;
    int ret;
    RTSPMessageHeader reply1, *reply = &reply1;
    char cmd[1024];

retry:
      ......................
    ret = ff_rtsp_fetch_packet(s, pkt);
    ................
    }

此过程发生再ff_rtsp_fetch_packet中，

int ff_rtsp_fetch_packet(AVFormatContext *s, AVPacket *pkt)
{
    RTSPState *rt = s->priv_data;
     .....
    switch(rt->lower_transport) {
    default:
#if CONFIG_RTSP_DEMUXER
    case RTSP_LOWER_TRANSPORT_TCP:
        len = ff_rtsp_tcp_read_packet(s, &rtsp_st, rt->recvbuf, RECVBUF_SIZE);
        break;
#endif
    case RTSP_LOWER_TRANSPORT_UDP:
    case RTSP_LOWER_TRANSPORT_UDP_MULTICAST:
        len = udp_read_packet(s, &rtsp_st, rt->recvbuf, RECVBUF_SIZE, wait_end);
        if (len > 0 && rtsp_st->transport_priv && rt->transport == RTSP_TRANSPORT_RTP)
            ff_rtp_check_and_send_back_rr(rtsp_st->transport_priv, rtsp_st->rtp_handle, NULL, len);
        break;
   ....
   }
     if (len < 0)
        return len;
    if (len == 0)
        return AVERROR_EOF;
    if (rt->transport == RTSP_TRANSPORT_RDT) {
        ret = ff_rdt_parse_packet(rtsp_st->transport_priv, pkt, &rt->recvbuf, len);
    } else if (rt->transport == RTSP_TRANSPORT_RTP) {
        ret = ff_rtp_parse_packet(rtsp_st->transport_priv, pkt, &rt->recvbuf, len);
         .....
    }

如果是使用UDP协议，那么通过读取数据报，得到远端发送的rtp数据。

二、解析音视频负载

在（一）得到远端数据后，ffmpeg马上判断传输类型，并调用ff_rdt_parse_packet/ff_rtp_parse_packet解析。这里我们使用的是rtp数据，所以只看ff_rtp_parse_packet。下面是该函数的核心代码

static int rtp_parse_one_packet(RTPDemuxContext *s, AVPacket *pkt,
                                uint8_t **bufptr, int len)
{
   .....
    if ((buf[0] & 0xc0) != (RTP_VERSION << 6))
        return -1;
    if (RTP_PT_IS_RTCP(buf[1])) {
        return rtcp_parse_packet(s, buf, len);
    }
....
    if ((s->seq == 0 && !s->queue) || s->queue_size <= 1) {
        /* First packet, or no reordering */
        return rtp_parse_packet_internal(s, pkt, buf, len);
    } else {
        uint16_t seq = AV_RB16(buf + 2);
        int16_t diff = seq - s->seq;
        if (diff < 0) {
            /* Packet older than the previously emitted one, drop */
            av_log(s->ic, AV_LOG_WARNING,
                   "RTP: dropping old packet received too late\n");
            return -1;
        } else if (diff <= 1) {
            /* Correct packet */
            rv = rtp_parse_packet_internal(s, pkt, buf, len);
            return rv;
        } else {
            /* Still missing some packet, enqueue this one. */
            rv = enqueue_packet(s, buf, len);
            if (rv < 0)
                return rv;
            *bufptr = NULL;
            /* Return the first enqueued packet if the queue is full,
             * even if we're missing something */
            if (s->queue_len >= s->queue_size) {
                av_log(s->ic, AV_LOG_WARNING, "jitter buffer full\n");
                return rtp_parse_queued_packet(s, pkt);
            }
            return -1;
        }
    }
}

这部分开头做了一层过滤，通过PT数值过滤掉RTCP数据，随后判断了rtp-seq值，正常来说seq应该是连续增的，否则做相应的异常处理，得到正常的rtp数据后，便开始真正分析rtp包了（rtp_parse_packet_internal）

static int rtp_parse_packet_internal(RTPDemuxContext *s, AVPacket *pkt,
                                     const uint8_t *buf, int len)
{
    unsigned int ssrc;
    int payload_type, seq, flags = 0;
    int ext, csrc;
    AVStream *st;
    uint32_t timestamp;
    int rv = 0;

    csrc         = buf[0] & 0x0f;
    ext          = buf[0] & 0x10;
    payload_type = buf[1] & 0x7f;
    if (buf[1] & 0x80)
        flags |= RTP_FLAG_MARKER;
    seq       = AV_RB16(buf + 2);
    timestamp = AV_RB32(buf + 4);
    ssrc      = AV_RB32(buf + 8);
    /* store the ssrc in the RTPDemuxContext */
    s->ssrc = ssrc;

    /* NOTE: we can handle only one payload type */
    if (s->payload_type != payload_type)
        return -1;

    st = s->st;
    // only do something with this if all the rtp checks pass...
    if (!rtp_valid_packet_in_sequence(&s->statistics, seq)) {
        av_log(s->ic, AV_LOG_ERROR,
               "RTP: PT=%02x: bad cseq %04x expected=%04x\n",
               payload_type, seq, ((s->seq + 1) & 0xffff));
        return -1;
    }

    if (buf[0] & 0x20) {
        int padding = buf[len - 1];
        if (len >= 12 + padding)
            len -= padding;
    }

    s->seq = seq;
    len   -= 12;
    buf   += 12;

    len   -= 4 * csrc;
    buf   += 4 * csrc;
    if (len < 0)
        return AVERROR_INVALIDDATA;

    /* RFC 3550 Section 5.3.1 RTP Header Extension handling */
    if (ext) {
        if (len < 4)
            return -1;
        /* calculate the header extension length (stored as number
         * of 32-bit words) */
        ext = (AV_RB16(buf + 2) + 1) << 2;

        if (len < ext)
            return -1;
        // skip past RTP header extension
        len -= ext;
        buf += ext;
    }

    if (s->handler && s->handler->parse_packet) {
        rv = s->handler->parse_packet(s->ic, s->dynamic_protocol_context,
                                      s->st, pkt, &timestamp, buf, len, seq,
                                      flags);
    } else if (st) {
        if ((rv = av_new_packet(pkt, len)) < 0)
            return rv;
        memcpy(pkt->data, buf, len);
        pkt->stream_index = st->index;
    } else {
        return AVERROR(EINVAL);
    }

    // now perform timestamp things....
    finalize_packet(s, pkt, timestamp);

    return rv;
}

解析函数跳过了前面的head，直接指到rtp音视频数据所在的地方，并调用s->handler->parse_packet，将数据剥离到AVPacket中。现在疑问的是，parse_packet是在哪定义的呢。

 

经过一番寻找，不难发现，定义该结构的地方如下，再通过GDB断点，打印堆栈，发现就是在解析DESCRIPT负载数据的时候。这一部分定义了rtp所有可能的负载类型，并按ffmpeg特有的风格模块化，各自提供解析回调函数供外界调用。

void ff_register_dynamic_payload_handler(RTPDynamicProtocolHandler *handler)
{
    handler->next = rtp_first_dynamic_payload_handler;
    rtp_first_dynamic_payload_handler = handler;
}

void ff_register_rtp_dynamic_payload_handlers(void)
{
    ff_register_dynamic_payload_handler(&ff_ac3_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_amr_nb_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_amr_wb_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_dv_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726_16_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726_24_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726_32_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726_40_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726le_16_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726le_24_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726le_32_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_g726le_40_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_h261_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_h263_1998_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_h263_2000_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_h263_rfc2190_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_h264_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_hevc_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_ilbc_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_jpeg_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mp4a_latm_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mp4v_es_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mpeg_audio_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mpeg_audio_robust_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mpeg_video_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mpeg4_generic_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_mpegts_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_ms_rtp_asf_pfa_handler);
    ff_register_dynamic_payload_handler(&ff_ms_rtp_asf_pfv_handler);
    ff_register_dynamic_payload_handler(&ff_qcelp_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_qdm2_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_qt_rtp_aud_handler);
    ff_register_dynamic_payload_handler(&ff_qt_rtp_vid_handler);
    ff_register_dynamic_payload_handler(&ff_quicktime_rtp_aud_handler);
    ff_register_dynamic_payload_handler(&ff_quicktime_rtp_vid_handler);
    ff_register_dynamic_payload_handler(&ff_svq3_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_theora_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_vc2hq_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_vorbis_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_vp8_dynamic_handler);
    ff_register_dynamic_payload_handler(&ff_vp9_dynamic_handler);
    ff_register_dynamic_payload_handler(&gsm_dynamic_handler);
    ff_register_dynamic_payload_handler(&opus_dynamic_handler);
    ff_register_dynamic_payload_handler(&realmedia_mp3_dynamic_handler);
    ff_register_dynamic_payload_handler(&speex_dynamic_handler);
    ff_register_dynamic_payload_handler(&t140_dynamic_handler);
}

RTPDynamicProtocolHandler *ff_rtp_handler_find_by_name(const char *name,
                                                       enum AVMediaType codec_type)
{
    RTPDynamicProtocolHandler *handler;
    for (handler = rtp_first_dynamic_payload_handler;
         handler; handler = handler->next)
        if (handler->enc_name &&
            !av_strcasecmp(name, handler->enc_name) &&
            codec_type == handler->codec_type)
            return handler;
    return NULL;
}