gstreamer/ext/smoothstreaming/gstmssdemux.c

1320 lines
39 KiB
C
Raw Normal View History

/* GStreamer
* Copyright (C) 2012 Smart TV Alliance
* Author: Thiago Sousa Santos <thiago.sousa.santos@collabora.com>, Collabora Ltd.
*
* gstmssdemux.c:
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 02111-1307, USA.
*/
/**
* SECTION:element-mssdemux
*
* Demuxes a Microsoft's Smooth Streaming manifest into its audio and/or video streams.
*
*
*/
/*
* == Internals
*
* = Smooth streaming in a few lines
* A SS stream is defined by a xml manifest file. This file has a list of
* tracks (StreamIndex), each one can have multiple QualityLevels, that define
* different encoding/bitrates. When playing a track, only one of those
* QualityLevels can be active at a time (per stream).
*
* The StreamIndex defines a URL with {time} and {bitrate} tags that are
* replaced by values indicated by the fragment start times and the selected
* QualityLevel, that generates the fragments URLs.
*
* Another relevant detail is that the Isomedia fragments for smoothstreaming
* won't contains a 'moov' atom, nor a 'stsd', so there is no information
* about the media type/configuration on the fragments, it must be extracted
* from the Manifest and passed downstream. mssdemux does this via GstCaps.
*
* = How mssdemux works
* There is a gstmssmanifest.c utility that holds the manifest and parses
* and has functions to extract information from it. mssdemux received the
* manifest from its sink pad and starts processing it when it gets EOS.
*
* The Manifest is parsed and the streams are exposed, 1 pad for each, with
* a initially selected QualityLevel. Each stream starts its own GstTaks that
* is responsible for downloading fragments and pushing them downstream.
*
* When a new connection-speed is set, mssdemux evaluates the available
* QualityLevels and might decide to switch to another one. In this case it
* pushes a new GstCaps event indicating the new caps on the pads.
*
* All operations that intend to update the GstTasks state should be protected
* with the GST_OBJECT_LOCK.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "gst/gst-i18n-plugin.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "gstmssdemux.h"
GST_DEBUG_CATEGORY (mssdemux_debug);
#define DEFAULT_CONNECTION_SPEED 0
#define DEFAULT_MAX_QUEUE_SIZE_BUFFERS 0
#define DEFAULT_BITRATE_LIMIT 0.8
#define DOWNLOAD_RATE_MAX_HISTORY_LENGTH 5
#define MAX_DOWNLOAD_ERROR_COUNT 3
enum
{
PROP_0,
PROP_CONNECTION_SPEED,
PROP_MAX_QUEUE_SIZE_BUFFERS,
PROP_BITRATE_LIMIT,
PROP_LAST
};
static GstStaticPadTemplate gst_mss_demux_sink_template =
GST_STATIC_PAD_TEMPLATE ("sink",
GST_PAD_SINK,
GST_PAD_ALWAYS,
GST_STATIC_CAPS ("application/vnd.ms-sstr+xml")
);
static GstStaticPadTemplate gst_mss_demux_videosrc_template =
GST_STATIC_PAD_TEMPLATE ("video_%02u",
GST_PAD_SRC,
GST_PAD_SOMETIMES,
GST_STATIC_CAPS_ANY);
static GstStaticPadTemplate gst_mss_demux_audiosrc_template =
GST_STATIC_PAD_TEMPLATE ("audio_%02u",
GST_PAD_SRC,
GST_PAD_SOMETIMES,
GST_STATIC_CAPS_ANY);
2013-04-16 22:35:03 +00:00
#define gst_mss_demux_parent_class parent_class
G_DEFINE_TYPE (GstMssDemux, gst_mss_demux, GST_TYPE_ELEMENT);
static void gst_mss_demux_dispose (GObject * object);
static void gst_mss_demux_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec);
static void gst_mss_demux_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec);
static GstStateChangeReturn gst_mss_demux_change_state (GstElement * element,
GstStateChange transition);
2013-04-16 22:35:03 +00:00
static GstFlowReturn gst_mss_demux_chain (GstPad * pad, GstObject * parent,
GstBuffer * buffer);
static GstFlowReturn gst_mss_demux_event (GstPad * pad, GstObject * parent,
GstEvent * event);
static gboolean gst_mss_demux_src_query (GstPad * pad, GstObject * parent,
GstQuery * query);
static void gst_mss_demux_download_loop (GstMssDemuxStream * stream);
static GstFlowReturn gst_mss_demux_stream_push (GstMssDemuxStream * stream,
GstBuffer * buffer);
static GstFlowReturn gst_mss_demux_stream_push_event (GstMssDemuxStream *
stream, GstEvent * event);
static GstFlowReturn gst_mss_demux_combine_flows (GstMssDemux * mssdemux);
static gboolean gst_mss_demux_process_manifest (GstMssDemux * mssdemux);
static void
gst_mss_demux_stop_tasks (GstMssDemux * mssdemux, gboolean immediate);
static void
gst_mss_demux_class_init (GstMssDemuxClass * klass)
{
GObjectClass *gobject_class;
GstElementClass *gstelement_class;
gobject_class = (GObjectClass *) klass;
gstelement_class = (GstElementClass *) klass;
2013-04-16 22:35:03 +00:00
gst_element_class_add_pad_template (gstelement_class,
gst_static_pad_template_get (&gst_mss_demux_sink_template));
gst_element_class_add_pad_template (gstelement_class,
gst_static_pad_template_get (&gst_mss_demux_videosrc_template));
gst_element_class_add_pad_template (gstelement_class,
gst_static_pad_template_get (&gst_mss_demux_audiosrc_template));
gst_element_class_set_static_metadata (gstelement_class,
"Smooth Streaming demuxer", "Codec/Demuxer/Adaptive",
2013-04-16 22:35:03 +00:00
"Parse and demultiplex a Smooth Streaming manifest into audio and video "
"streams", "Thiago Santos <thiago.sousa.santos@collabora.com>");
gobject_class->dispose = gst_mss_demux_dispose;
gobject_class->set_property = gst_mss_demux_set_property;
gobject_class->get_property = gst_mss_demux_get_property;
g_object_class_install_property (gobject_class, PROP_CONNECTION_SPEED,
g_param_spec_uint ("connection-speed", "Connection Speed",
"Network connection speed in kbps (0 = unknown)",
0, G_MAXUINT / 1000, DEFAULT_CONNECTION_SPEED,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_MAX_QUEUE_SIZE_BUFFERS,
g_param_spec_uint ("max-queue-size-buffers", "Max queue size in buffers",
"Maximum buffers that can be stored in each internal stream queue "
"(0 = infinite)", 0, G_MAXUINT, DEFAULT_MAX_QUEUE_SIZE_BUFFERS,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_BITRATE_LIMIT,
g_param_spec_float ("bitrate-limit",
"Bitrate limit in %",
"Limit of the available bitrate to use when switching to alternates.",
0, 1, DEFAULT_BITRATE_LIMIT,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
gstelement_class->change_state =
GST_DEBUG_FUNCPTR (gst_mss_demux_change_state);
2013-04-16 22:35:03 +00:00
GST_DEBUG_CATEGORY_INIT (mssdemux_debug, "mssdemux", 0, "mssdemux plugin");
}
static void
2013-04-16 22:35:03 +00:00
gst_mss_demux_init (GstMssDemux * mssdemux)
{
mssdemux->sinkpad =
gst_pad_new_from_static_template (&gst_mss_demux_sink_template, "sink");
gst_pad_set_chain_function (mssdemux->sinkpad,
GST_DEBUG_FUNCPTR (gst_mss_demux_chain));
gst_pad_set_event_function (mssdemux->sinkpad,
GST_DEBUG_FUNCPTR (gst_mss_demux_event));
gst_element_add_pad (GST_ELEMENT_CAST (mssdemux), mssdemux->sinkpad);
mssdemux->data_queue_max_size = DEFAULT_MAX_QUEUE_SIZE_BUFFERS;
mssdemux->bitrate_limit = DEFAULT_BITRATE_LIMIT;
mssdemux->have_group_id = FALSE;
mssdemux->group_id = G_MAXUINT;
}
static GstMssDemuxStream *
gst_mss_demux_stream_new (GstMssDemux * mssdemux,
GstMssStream * manifeststream, GstPad * srcpad)
{
GstMssDemuxStream *stream;
stream = g_new0 (GstMssDemuxStream, 1);
stream->downloader = gst_uri_downloader_new ();
/* Downloading task */
2013-04-16 22:35:03 +00:00
g_rec_mutex_init (&stream->download_lock);
stream->download_task =
2013-04-16 22:35:03 +00:00
gst_task_new ((GstTaskFunction) gst_mss_demux_download_loop, stream,
NULL);
gst_task_set_lock (stream->download_task, &stream->download_lock);
stream->pad = srcpad;
stream->manifest_stream = manifeststream;
stream->parent = mssdemux;
gst_download_rate_init (&stream->download_rate);
gst_download_rate_set_max_length (&stream->download_rate,
DOWNLOAD_RATE_MAX_HISTORY_LENGTH);
gst_segment_init (&stream->segment, GST_FORMAT_TIME);
return stream;
}
static void
gst_mss_demux_stream_free (GstMssDemuxStream * stream)
{
if (stream->download_task) {
if (GST_TASK_STATE (stream->download_task) != GST_TASK_STOPPED) {
GST_DEBUG_OBJECT (stream->parent, "Leaving streaming task %s:%s",
GST_DEBUG_PAD_NAME (stream->pad));
gst_uri_downloader_cancel (stream->downloader);
gst_task_stop (stream->download_task);
GST_LOG_OBJECT (stream->parent, "Waiting for task to finish");
gst_task_join (stream->download_task);
GST_LOG_OBJECT (stream->parent, "Finished");
}
gst_object_unref (stream->download_task);
2013-04-16 22:35:03 +00:00
g_rec_mutex_clear (&stream->download_lock);
stream->download_task = NULL;
}
gst_download_rate_deinit (&stream->download_rate);
if (stream->pending_newsegment) {
gst_event_unref (stream->pending_newsegment);
stream->pending_newsegment = NULL;
}
if (stream->downloader != NULL) {
g_object_unref (stream->downloader);
stream->downloader = NULL;
}
if (stream->pad) {
gst_object_unref (stream->pad);
stream->pad = NULL;
}
if (stream->caps)
gst_caps_unref (stream->caps);
g_free (stream);
}
static void
gst_mss_demux_reset (GstMssDemux * mssdemux)
{
GSList *iter;
gst_mss_demux_stop_tasks (mssdemux, TRUE);
if (mssdemux->manifest_buffer) {
gst_buffer_unref (mssdemux->manifest_buffer);
mssdemux->manifest_buffer = NULL;
}
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
if (stream->pad)
gst_element_remove_pad (GST_ELEMENT_CAST (mssdemux), stream->pad);
gst_mss_demux_stream_free (stream);
}
g_slist_free (mssdemux->streams);
mssdemux->streams = NULL;
if (mssdemux->manifest) {
gst_mss_manifest_free (mssdemux->manifest);
mssdemux->manifest = NULL;
}
mssdemux->n_videos = mssdemux->n_audios = 0;
g_free (mssdemux->base_url);
mssdemux->base_url = NULL;
g_free (mssdemux->manifest_uri);
mssdemux->manifest_uri = NULL;
mssdemux->have_group_id = FALSE;
mssdemux->group_id = G_MAXUINT;
}
static void
gst_mss_demux_dispose (GObject * object)
{
GstMssDemux *mssdemux = GST_MSS_DEMUX_CAST (object);
gst_mss_demux_reset (mssdemux);
G_OBJECT_CLASS (parent_class)->dispose (object);
}
static void
gst_mss_demux_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec)
{
GstMssDemux *mssdemux = GST_MSS_DEMUX (object);
switch (prop_id) {
case PROP_CONNECTION_SPEED:
GST_OBJECT_LOCK (mssdemux);
mssdemux->connection_speed = g_value_get_uint (value) * 1000;
mssdemux->update_bitrates = TRUE;
GST_DEBUG_OBJECT (mssdemux, "Connection speed set to %" G_GUINT64_FORMAT,
mssdemux->connection_speed);
GST_OBJECT_UNLOCK (mssdemux);
break;
case PROP_MAX_QUEUE_SIZE_BUFFERS:
mssdemux->data_queue_max_size = g_value_get_uint (value);
break;
case PROP_BITRATE_LIMIT:
mssdemux->bitrate_limit = g_value_get_float (value);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_mss_demux_get_property (GObject * object, guint prop_id, GValue * value,
GParamSpec * pspec)
{
GstMssDemux *mssdemux = GST_MSS_DEMUX (object);
switch (prop_id) {
case PROP_CONNECTION_SPEED:
g_value_set_uint (value, mssdemux->connection_speed / 1000);
break;
case PROP_MAX_QUEUE_SIZE_BUFFERS:
g_value_set_uint (value, mssdemux->data_queue_max_size);
break;
case PROP_BITRATE_LIMIT:
g_value_set_float (value, mssdemux->bitrate_limit);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static GstStateChangeReturn
gst_mss_demux_change_state (GstElement * element, GstStateChange transition)
{
GstMssDemux *mssdemux = GST_MSS_DEMUX_CAST (element);
GstStateChangeReturn result = GST_STATE_CHANGE_FAILURE;
switch (transition) {
case GST_STATE_CHANGE_PAUSED_TO_READY:
gst_mss_demux_reset (mssdemux);
break;
case GST_STATE_CHANGE_READY_TO_NULL:
break;
default:
break;
}
result = GST_ELEMENT_CLASS (parent_class)->change_state (element, transition);
return result;
}
static GstFlowReturn
2013-04-16 22:35:03 +00:00
gst_mss_demux_chain (GstPad * pad, GstObject * parent, GstBuffer * buffer)
{
2013-04-16 22:35:03 +00:00
GstMssDemux *mssdemux = GST_MSS_DEMUX_CAST (parent);
if (mssdemux->manifest_buffer == NULL)
mssdemux->manifest_buffer = buffer;
else
mssdemux->manifest_buffer =
2013-04-16 22:35:03 +00:00
gst_buffer_append (mssdemux->manifest_buffer, buffer);
GST_INFO_OBJECT (mssdemux, "Received manifest buffer, total size is %i bytes",
(gint) gst_buffer_get_size (mssdemux->manifest_buffer));
return GST_FLOW_OK;
}
static void
gst_mss_demux_start (GstMssDemux * mssdemux)
{
GSList *iter;
GST_INFO_OBJECT (mssdemux, "Starting streams' tasks");
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
gst_task_start (stream->download_task);
}
}
static gboolean
gst_mss_demux_push_src_event (GstMssDemux * mssdemux, GstEvent * event)
{
GSList *iter;
gboolean ret = TRUE;
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
gst_event_ref (event);
ret = ret & gst_pad_push_event (stream->pad, event);
}
gst_event_unref (event);
return ret;
}
static gboolean
2013-04-16 22:35:03 +00:00
gst_mss_demux_event (GstPad * pad, GstObject * parent, GstEvent * event)
{
2013-04-16 22:35:03 +00:00
GstMssDemux *mssdemux = GST_MSS_DEMUX_CAST (parent);
gboolean forward = TRUE;
gboolean ret = TRUE;
switch (GST_EVENT_TYPE (event)) {
case GST_EVENT_FLUSH_STOP:
gst_mss_demux_reset (mssdemux);
break;
case GST_EVENT_EOS:
if (mssdemux->manifest_buffer == NULL) {
GST_WARNING_OBJECT (mssdemux, "Received EOS without a manifest.");
break;
}
GST_INFO_OBJECT (mssdemux, "Received EOS");
if (gst_mss_demux_process_manifest (mssdemux))
gst_mss_demux_start (mssdemux);
forward = FALSE;
break;
default:
break;
}
if (forward) {
2013-04-16 22:35:03 +00:00
ret = gst_pad_event_default (pad, parent, event);
} else {
gst_event_unref (event);
}
return ret;
}
static void
gst_mss_demux_stop_tasks (GstMssDemux * mssdemux, gboolean immediate)
{
GSList *iter;
GST_OBJECT_LOCK (mssdemux);
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
gst_task_stop (stream->download_task);
stream->cancelled = TRUE;
if (immediate)
gst_uri_downloader_cancel (stream->downloader);
}
GST_OBJECT_UNLOCK (mssdemux);
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
gst_task_join (stream->download_task);
stream->download_error_count = 0;
}
}
static void
gst_mss_demux_restart_tasks (GstMssDemux * mssdemux)
{
GSList *iter;
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
gst_uri_downloader_reset (stream->downloader);
stream->cancelled = FALSE;
gst_task_start (stream->download_task);
}
}
static gboolean
2013-04-16 22:35:03 +00:00
gst_mss_demux_src_event (GstPad * pad, GstObject * parent, GstEvent * event)
{
GstMssDemux *mssdemux;
2013-04-16 22:35:03 +00:00
mssdemux = GST_MSS_DEMUX_CAST (parent);
switch (event->type) {
case GST_EVENT_SEEK:
{
gdouble rate;
GstFormat format;
GstSeekFlags flags;
GstSeekType start_type, stop_type;
gint64 start, stop;
GstEvent *newsegment;
GSList *iter;
2013-04-16 22:35:03 +00:00
gboolean update;
GST_INFO_OBJECT (mssdemux, "Received GST_EVENT_SEEK");
gst_event_parse_seek (event, &rate, &format, &flags, &start_type, &start,
&stop_type, &stop);
if (format != GST_FORMAT_TIME)
goto not_supported;
GST_DEBUG_OBJECT (mssdemux,
"seek event, rate: %f start: %" GST_TIME_FORMAT " stop: %"
GST_TIME_FORMAT, rate, GST_TIME_ARGS (start), GST_TIME_ARGS (stop));
if (flags & GST_SEEK_FLAG_FLUSH) {
GstEvent *flush = gst_event_new_flush_start ();
GST_DEBUG_OBJECT (mssdemux, "sending flush start");
gst_event_set_seqnum (flush, gst_event_get_seqnum (event));
gst_mss_demux_push_src_event (mssdemux, flush);
}
gst_mss_demux_stop_tasks (mssdemux, TRUE);
if (!gst_mss_manifest_seek (mssdemux->manifest, start)) {;
GST_WARNING_OBJECT (mssdemux, "Could not find seeked fragment");
goto not_supported;
}
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
stream->eos = FALSE;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
if (flags & GST_SEEK_FLAG_FLUSH) {
stream->last_ret = GST_FLOW_OK;
}
gst_segment_do_seek (&stream->segment, rate, format, flags,
start_type, start, stop_type, stop, &update);
newsegment = gst_event_new_segment (&stream->segment);
gst_event_set_seqnum (newsegment, gst_event_get_seqnum (event));
if (stream->pending_newsegment)
gst_event_unref (stream->pending_newsegment);
stream->pending_newsegment = newsegment;
}
if (flags & GST_SEEK_FLAG_FLUSH) {
2013-04-16 22:35:03 +00:00
GstEvent *flush = gst_event_new_flush_stop (TRUE);
GST_DEBUG_OBJECT (mssdemux, "sending flush stop");
gst_event_set_seqnum (flush, gst_event_get_seqnum (event));
gst_mss_demux_push_src_event (mssdemux, flush);
}
gst_mss_demux_restart_tasks (mssdemux);
gst_event_unref (event);
return TRUE;
}
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
case GST_EVENT_RECONFIGURE:{
GSList *iter;
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
if (stream->pad == pad) {
GST_OBJECT_LOCK (mssdemux);
if (stream->last_ret == GST_FLOW_NOT_LINKED) {
GST_DEBUG_OBJECT (stream->pad, "Received reconfigure");
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
stream->restart_download = TRUE;
gst_task_start (stream->download_task);
}
GST_OBJECT_UNLOCK (mssdemux);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
gst_event_unref (event);
return TRUE;
}
}
}
break;
default:
break;
}
2013-04-16 22:35:03 +00:00
return gst_pad_event_default (pad, parent, event);
not_supported:
gst_event_unref (event);
return FALSE;
}
static gboolean
2013-04-16 22:35:03 +00:00
gst_mss_demux_src_query (GstPad * pad, GstObject * parent, GstQuery * query)
{
GstMssDemux *mssdemux;
gboolean ret = FALSE;
if (query == NULL)
return FALSE;
2013-04-16 22:35:03 +00:00
mssdemux = GST_MSS_DEMUX (parent);
switch (query->type) {
case GST_QUERY_DURATION:{
GstClockTime duration = -1;
GstFormat fmt;
gst_query_parse_duration (query, &fmt, NULL);
if (fmt == GST_FORMAT_TIME && mssdemux->manifest) {
/* TODO should we use the streams accumulated duration or the main manifest duration? */
duration = gst_mss_manifest_get_gst_duration (mssdemux->manifest);
if (GST_CLOCK_TIME_IS_VALID (duration) && duration > 0) {
gst_query_set_duration (query, GST_FORMAT_TIME, duration);
ret = TRUE;
}
}
GST_INFO_OBJECT (mssdemux, "GST_QUERY_DURATION returns %s with duration %"
GST_TIME_FORMAT, ret ? "TRUE" : "FALSE", GST_TIME_ARGS (duration));
break;
}
case GST_QUERY_LATENCY:{
gboolean live = FALSE;
live = mssdemux->manifest
&& gst_mss_manifest_is_live (mssdemux->manifest);
gst_query_set_latency (query, live, 0, -1);
ret = TRUE;
break;
}
case GST_QUERY_SEEKING:{
GstFormat fmt;
gint64 stop = -1;
if (!mssdemux->manifest) {
return FALSE; /* no seeking without manifest */
} else if (gst_mss_manifest_is_live (mssdemux->manifest)) {
return FALSE; /* no live seeking */
}
gst_query_parse_seeking (query, &fmt, NULL, NULL, NULL);
GST_INFO_OBJECT (mssdemux, "Received GST_QUERY_SEEKING with format %d",
fmt);
if (fmt == GST_FORMAT_TIME) {
GstClockTime duration;
duration = gst_mss_manifest_get_gst_duration (mssdemux->manifest);
if (GST_CLOCK_TIME_IS_VALID (duration) && duration > 0)
stop = duration;
gst_query_set_seeking (query, fmt, TRUE, 0, stop);
ret = TRUE;
GST_INFO_OBJECT (mssdemux, "GST_QUERY_SEEKING returning with stop : %"
GST_TIME_FORMAT, GST_TIME_ARGS (stop));
}
break;
}
default:
/* Don't fordward queries upstream because of the special nature of this
* "demuxer", which relies on the upstream element only to be fed
* the Manifest
*/
break;
}
return ret;
}
static void
_set_src_pad_functions (GstPad * pad)
{
gst_pad_set_query_function (pad, GST_DEBUG_FUNCPTR (gst_mss_demux_src_query));
gst_pad_set_event_function (pad, GST_DEBUG_FUNCPTR (gst_mss_demux_src_event));
}
static GstPad *
_create_pad (GstMssDemux * mssdemux, GstMssStream * manifeststream)
{
gchar *name;
GstPad *srcpad = NULL;
GstMssStreamType streamtype;
streamtype = gst_mss_stream_get_type (manifeststream);
GST_DEBUG_OBJECT (mssdemux, "Found stream of type: %s",
gst_mss_stream_type_name (streamtype));
/* TODO use stream's name/bitrate/index as the pad name? */
if (streamtype == MSS_STREAM_TYPE_VIDEO) {
name = g_strdup_printf ("video_%02u", mssdemux->n_videos++);
srcpad =
gst_pad_new_from_static_template (&gst_mss_demux_videosrc_template,
name);
g_free (name);
} else if (streamtype == MSS_STREAM_TYPE_AUDIO) {
name = g_strdup_printf ("audio_%02u", mssdemux->n_audios++);
srcpad =
gst_pad_new_from_static_template (&gst_mss_demux_audiosrc_template,
name);
g_free (name);
}
if (!srcpad) {
GST_WARNING_OBJECT (mssdemux, "Ignoring unknown type stream");
return NULL;
}
_set_src_pad_functions (srcpad);
return srcpad;
}
static void
gst_mss_demux_create_streams (GstMssDemux * mssdemux)
{
GSList *streams = gst_mss_manifest_get_streams (mssdemux->manifest);
GSList *iter;
if (streams == NULL) {
GST_INFO_OBJECT (mssdemux, "No streams found in the manifest");
GST_ELEMENT_ERROR (mssdemux, STREAM, DEMUX,
(_("This file contains no playable streams.")),
("no streams found at the Manifest"));
return;
}
for (iter = streams; iter; iter = g_slist_next (iter)) {
GstPad *srcpad = NULL;
GstMssDemuxStream *stream = NULL;
GstMssStream *manifeststream = iter->data;
srcpad = _create_pad (mssdemux, manifeststream);
if (!srcpad) {
continue;
}
stream = gst_mss_demux_stream_new (mssdemux, manifeststream, srcpad);
gst_mss_stream_set_active (manifeststream, TRUE);
mssdemux->streams = g_slist_append (mssdemux->streams, stream);
}
/* select initial bitrates */
GST_OBJECT_LOCK (mssdemux);
GST_INFO_OBJECT (mssdemux, "Changing max bitrate to %" G_GUINT64_FORMAT,
mssdemux->connection_speed);
gst_mss_manifest_change_bitrate (mssdemux->manifest,
mssdemux->connection_speed);
mssdemux->update_bitrates = FALSE;
GST_OBJECT_UNLOCK (mssdemux);
}
static GstCaps *
create_mss_caps (GstMssDemuxStream * stream, GstCaps * caps)
{
return gst_caps_new_simple ("video/quicktime", "variant", G_TYPE_STRING,
"mss-fragmented", "timescale", G_TYPE_UINT64,
gst_mss_stream_get_timescale (stream->manifest_stream), "media-caps",
GST_TYPE_CAPS, caps, NULL);
}
static gboolean
gst_mss_demux_expose_stream (GstMssDemux * mssdemux, GstMssDemuxStream * stream)
{
GstCaps *caps;
GstCaps *media_caps;
GstPad *pad = stream->pad;
media_caps = gst_mss_stream_get_caps (stream->manifest_stream);
if (media_caps) {
gchar *name = gst_pad_get_name (pad);
const gchar *lang;
GstEvent *event;
gchar *stream_id;
2013-04-16 22:35:03 +00:00
gst_pad_set_active (pad, TRUE);
caps = create_mss_caps (stream, media_caps);
gst_caps_unref (media_caps);
stream_id =
gst_pad_create_stream_id (pad, GST_ELEMENT_CAST (mssdemux), name);
event =
gst_pad_get_sticky_event (mssdemux->sinkpad, GST_EVENT_STREAM_START, 0);
if (event) {
if (gst_event_parse_group_id (event, &mssdemux->group_id))
mssdemux->have_group_id = TRUE;
else
mssdemux->have_group_id = FALSE;
gst_event_unref (event);
} else if (!mssdemux->have_group_id) {
mssdemux->have_group_id = TRUE;
mssdemux->group_id = gst_util_group_id_next ();
}
event = gst_event_new_stream_start (stream_id);
if (mssdemux->have_group_id)
gst_event_set_group_id (event, mssdemux->group_id);
gst_pad_push_event (pad, event);
g_free (stream_id);
g_free (name);
gst_pad_set_caps (pad, caps);
stream->caps = caps;
lang = gst_mss_stream_get_lang (stream->manifest_stream);
if (lang != NULL) {
GstTagList *tags;
tags = gst_tag_list_new (GST_TAG_LANGUAGE_CODE, lang, NULL);
gst_pad_push_event (stream->pad, gst_event_new_tag (tags));
}
gst_pad_push_event (stream->pad, gst_event_new_segment (&stream->segment));
GST_INFO_OBJECT (mssdemux, "Adding srcpad %s:%s with caps %" GST_PTR_FORMAT,
GST_DEBUG_PAD_NAME (pad), caps);
gst_object_ref (pad);
gst_element_add_pad (GST_ELEMENT_CAST (mssdemux), pad);
} else {
GST_WARNING_OBJECT (mssdemux,
"Couldn't get caps from manifest stream %p %s, not exposing it", stream,
GST_PAD_NAME (stream->pad));
return FALSE;
}
return TRUE;
}
static gboolean
gst_mss_demux_process_manifest (GstMssDemux * mssdemux)
{
GstQuery *query;
gchar *uri = NULL;
gboolean ret;
GSList *iter;
g_return_val_if_fail (mssdemux->manifest_buffer != NULL, FALSE);
g_return_val_if_fail (mssdemux->manifest == NULL, FALSE);
query = gst_query_new_uri ();
ret = gst_pad_peer_query (mssdemux->sinkpad, query);
if (ret) {
gchar *baseurl_end;
gst_query_parse_uri (query, &uri);
GST_INFO_OBJECT (mssdemux, "Upstream is using URI: %s", uri);
mssdemux->manifest_uri = g_strdup (uri);
baseurl_end = g_strrstr (uri, "/Manifest");
if (baseurl_end == NULL) {
/* second try */
baseurl_end = g_strrstr (uri, "/manifest");
}
if (baseurl_end) {
/* set the new end of the string */
baseurl_end[0] = '\0';
} else {
GST_WARNING_OBJECT (mssdemux, "Stream's URI didn't end with /manifest");
}
mssdemux->base_url = uri;
}
gst_query_unref (query);
if (mssdemux->base_url == NULL) {
GST_ELEMENT_ERROR (mssdemux, RESOURCE, NOT_FOUND,
(_("Couldn't get the Manifest's URI")),
("need to get the manifest's URI from upstream elements"));
return FALSE;
}
GST_INFO_OBJECT (mssdemux, "Received manifest: %i bytes",
(gint) gst_buffer_get_size (mssdemux->manifest_buffer));
mssdemux->manifest = gst_mss_manifest_new (mssdemux->manifest_buffer);
if (!mssdemux->manifest) {
GST_ELEMENT_ERROR (mssdemux, STREAM, FORMAT, ("Bad manifest file"),
("Xml manifest file couldn't be parsed"));
return FALSE;
}
GST_INFO_OBJECT (mssdemux, "Live stream: %d",
gst_mss_manifest_is_live (mssdemux->manifest));
gst_mss_demux_create_streams (mssdemux);
for (iter = mssdemux->streams; iter;) {
GSList *current = iter;
GstMssDemuxStream *stream = iter->data;
iter = g_slist_next (iter); /* do it ourselves as we want it done in the beginning of the loop */
if (!gst_mss_demux_expose_stream (mssdemux, stream)) {
gst_mss_demux_stream_free (stream);
mssdemux->streams = g_slist_delete_link (mssdemux->streams, current);
}
}
if (!mssdemux->streams) {
/* no streams */
GST_WARNING_OBJECT (mssdemux, "Couldn't identify the caps for any of the "
"streams found in the manifest");
GST_ELEMENT_ERROR (mssdemux, STREAM, DEMUX,
(_("This file contains no playable streams.")),
("No known stream formats found at the Manifest"));
return FALSE;
}
gst_element_no_more_pads (GST_ELEMENT_CAST (mssdemux));
return TRUE;
}
static void
gst_mss_demux_reload_manifest (GstMssDemux * mssdemux)
{
GstUriDownloader *downloader;
GstFragment *manifest_data;
GstBuffer *manifest_buffer;
downloader = gst_uri_downloader_new ();
manifest_data =
gst_uri_downloader_fetch_uri (downloader, mssdemux->manifest_uri, NULL,
TRUE, TRUE, TRUE, NULL);
manifest_buffer = gst_fragment_get_buffer (manifest_data);
g_object_unref (manifest_data);
gst_mss_manifest_reload_fragments (mssdemux->manifest, manifest_buffer);
gst_buffer_replace (&mssdemux->manifest_buffer, manifest_buffer);
gst_buffer_unref (manifest_buffer);
g_object_unref (downloader);
}
static GstEvent *
gst_mss_demux_reconfigure_stream (GstMssDemuxStream * stream)
{
GstEvent *capsevent = NULL;
GstMssDemux *mssdemux = stream->parent;
guint64 new_bitrate;
new_bitrate =
mssdemux->bitrate_limit *
gst_download_rate_get_current_rate (&stream->download_rate);
if (mssdemux->connection_speed) {
new_bitrate = MIN (mssdemux->connection_speed, new_bitrate);
}
GST_DEBUG_OBJECT (stream->pad,
"Current stream download bitrate %" G_GUINT64_FORMAT, new_bitrate);
if (gst_mss_stream_select_bitrate (stream->manifest_stream, new_bitrate)) {
GstCaps *caps;
caps = gst_mss_stream_get_caps (stream->manifest_stream);
GST_DEBUG_OBJECT (stream->pad,
"Starting streams reconfiguration due to bitrate changes");
if (stream->caps)
gst_caps_unref (stream->caps);
stream->caps = create_mss_caps (stream, caps);
gst_caps_unref (caps);
GST_DEBUG_OBJECT (stream->pad,
"Stream changed bitrate to %" G_GUINT64_FORMAT " caps: %"
GST_PTR_FORMAT,
gst_mss_stream_get_current_bitrate (stream->manifest_stream), caps);
2013-04-16 22:35:03 +00:00
capsevent = gst_event_new_caps (stream->caps);
GST_DEBUG_OBJECT (stream->pad, "Finished streams reconfiguration");
}
return capsevent;
}
static GstFlowReturn
gst_mss_demux_stream_download_fragment (GstMssDemuxStream * stream,
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
GstBuffer ** _buffer)
{
GstMssDemux *mssdemux = stream->parent;
gchar *path;
gchar *url;
GstFragment *fragment;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
GstBuffer *buffer;
GstFlowReturn ret = GST_FLOW_OK;
guint64 before_download, after_download;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
/* special case for not-linked streams */
if (stream->last_ret == GST_FLOW_NOT_LINKED) {
GST_DEBUG_OBJECT (stream->pad, "Skipping download for not-linked stream %p",
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
stream);
return GST_FLOW_NOT_LINKED;
}
before_download = g_get_real_time ();
GST_DEBUG_OBJECT (stream->pad, "Getting url for stream");
ret = gst_mss_stream_get_fragment_url (stream->manifest_stream, &path);
switch (ret) {
case GST_FLOW_OK:
break; /* all is good, let's go */
2013-04-16 22:35:03 +00:00
case GST_FLOW_EOS:
if (gst_mss_manifest_is_live (mssdemux->manifest)) {
gst_mss_demux_reload_manifest (mssdemux);
return GST_FLOW_OK;
}
2013-04-16 22:35:03 +00:00
return GST_FLOW_EOS;
case GST_FLOW_ERROR:
goto error;
default:
break;
}
if (!path) {
goto no_url_error;
}
GST_DEBUG_OBJECT (mssdemux, "Got url path '%s' for stream %p", path, stream);
url = g_strdup_printf ("%s/%s", mssdemux->base_url, path);
GST_DEBUG_OBJECT (mssdemux, "Got url '%s' for stream %p", url, stream);
fragment =
gst_uri_downloader_fetch_uri (stream->downloader, url,
mssdemux->manifest_uri, FALSE, FALSE, TRUE, NULL);
g_free (path);
g_free (url);
if (!fragment) {
GST_INFO_OBJECT (mssdemux, "No fragment downloaded");
/* TODO check if we are truly stoping */
if (gst_mss_manifest_is_live (mssdemux->manifest)) {
/* looks like there is no way of knowing when a live stream has ended
* Have to assume we are falling behind and cause a manifest reload */
return GST_FLOW_OK;
}
return GST_FLOW_ERROR;
}
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
buffer = gst_fragment_get_buffer (fragment);
*_buffer = buffer = gst_buffer_make_writable (buffer);
GST_BUFFER_TIMESTAMP (buffer) =
gst_mss_stream_get_fragment_gst_timestamp (stream->manifest_stream);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
GST_BUFFER_DURATION (buffer) =
gst_mss_stream_get_fragment_gst_duration (stream->manifest_stream);
2013-01-17 19:19:54 +00:00
g_object_unref (fragment);
after_download = g_get_real_time ();
{
#ifndef GST_DISABLE_GST_DEBUG
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
guint64 bitrate = (8 * gst_buffer_get_size (buffer) * 1000000LLU) /
(after_download - before_download);
#endif
GST_DEBUG_OBJECT (mssdemux,
"Measured download bitrate: %s %" G_GUINT64_FORMAT " bps",
GST_PAD_NAME (stream->pad), bitrate);
gst_download_rate_add_rate (&stream->download_rate,
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
gst_buffer_get_size (buffer),
2013-04-16 22:35:03 +00:00
1000 * (after_download - before_download));
}
return ret;
no_url_error:
{
GST_ELEMENT_ERROR (mssdemux, STREAM, DEMUX,
(_("Failed to get fragment URL.")),
("An error happened when getting fragment URL"));
gst_task_pause (stream->download_task);
return GST_FLOW_ERROR;
}
error:
{
GST_WARNING_OBJECT (mssdemux, "Error while pushing fragment");
gst_task_pause (stream->download_task);
return GST_FLOW_ERROR;
}
}
static void
gst_mss_demux_download_loop (GstMssDemuxStream * stream)
{
GstMssDemux *mssdemux = stream->parent;
GstFlowReturn ret;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
GstBuffer *buffer = NULL;
gboolean buffer_downloaded = FALSE;
GstEvent *gap = NULL;
GstEvent *capsevent = NULL;
GST_LOG_OBJECT (stream->pad, "download loop start");
GST_OBJECT_LOCK (mssdemux);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
if (G_UNLIKELY (stream->restart_download)) {
GstClockTime cur, ts;
gint64 pos;
GST_DEBUG_OBJECT (stream->pad,
"Activating stream due to reconfigure event");
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
cur = GST_CLOCK_TIME_IS_VALID (stream->next_timestamp) ?
stream->next_timestamp : 0;
if (gst_pad_peer_query_position (stream->pad, GST_FORMAT_TIME, &pos)) {
ts = (GstClockTime) pos;
GST_DEBUG_OBJECT (mssdemux, "Downstream position: %"
GST_TIME_FORMAT, GST_TIME_ARGS (ts));
} else {
GST_DEBUG_OBJECT (mssdemux, "Downstream position query failed, "
"failling back to segment position");
ts = stream->segment.position;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
}
/* we might have already pushed this data */
ts = MAX (ts, stream->next_timestamp);
GST_DEBUG_OBJECT (stream->pad, "Restarting stream at "
"position %" GST_TIME_FORMAT, GST_TIME_ARGS (ts));
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
if (GST_CLOCK_TIME_IS_VALID (ts)) {
gst_mss_stream_seek (stream->manifest_stream, ts);
if (cur < ts) {
gap = gst_event_new_gap (cur, ts - cur);
}
}
stream->restart_download = FALSE;
stream->last_ret = GST_FLOW_OK;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
}
capsevent = gst_mss_demux_reconfigure_stream (stream);
GST_OBJECT_UNLOCK (mssdemux);
if (G_UNLIKELY (gap != NULL))
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
gst_pad_push_event (stream->pad, gap);
if (G_UNLIKELY (capsevent != NULL))
gst_pad_push_event (stream->pad, capsevent);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
ret = gst_mss_demux_stream_download_fragment (stream, &buffer);
buffer_downloaded = buffer != NULL;
GST_OBJECT_LOCK (mssdemux);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
if (stream->cancelled) {
if (buffer)
gst_buffer_unref (buffer);
GST_OBJECT_UNLOCK (mssdemux);
goto cancelled;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
}
GST_OBJECT_UNLOCK (mssdemux);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
if (buffer) {
ret = gst_mss_demux_stream_push (stream, buffer);
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
}
GST_OBJECT_LOCK (mssdemux);
stream->last_ret = ret;
if (stream->cancelled) {
GST_OBJECT_UNLOCK (mssdemux);
goto cancelled;
}
switch (ret) {
case GST_FLOW_OK:
break; /* all is good, let's go */
2013-04-16 22:35:03 +00:00
case GST_FLOW_EOS:
GST_DEBUG_OBJECT (stream->pad, "EOS, stopping download loop");
/* we push the EOS after releasing the object lock */
gst_task_pause (stream->download_task);
break;
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
case GST_FLOW_NOT_LINKED:
gst_task_pause (stream->download_task);
if (gst_mss_demux_combine_flows (mssdemux) == GST_FLOW_NOT_LINKED) {
GST_ELEMENT_ERROR (mssdemux, STREAM, FAILED,
(_("Internal data stream error.")),
("stream stopped, reason %s",
gst_flow_get_name (GST_FLOW_NOT_LINKED)));
}
break;
case GST_FLOW_FLUSHING:{
GSList *iter;
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *other;
other = iter->data;
gst_task_pause (other->download_task);
}
}
break;
default:
if (ret <= GST_FLOW_ERROR) {
if (buffer_downloaded) {
GST_ERROR_OBJECT (mssdemux, "Error while pushing fragment");
} else {
GST_WARNING_OBJECT (mssdemux, "Error while downloading fragment");
if (++stream->download_error_count >=
DOWNLOAD_RATE_MAX_HISTORY_LENGTH) {
GST_ELEMENT_ERROR (mssdemux, RESOURCE, NOT_FOUND,
(_("Couldn't download fragments")),
("fragment downloading has failed too much consecutive times"));
}
}
}
break;
}
GST_OBJECT_UNLOCK (mssdemux);
if (G_UNLIKELY (ret == GST_FLOW_EOS)) {
gst_mss_demux_stream_push_event (stream, gst_event_new_eos ());
}
if (buffer_downloaded) {
stream->download_error_count = 0;
gst_mss_stream_advance_fragment (stream->manifest_stream);
}
GST_LOG_OBJECT (stream->pad, "download loop end");
return;
cancelled:
{
GST_DEBUG_OBJECT (stream->pad, "Stream has been cancelled");
return;
}
}
mssdemux: avoid downloading not-linked streams When a stream gets a not-linked return, it will be marked as so and won't download any more new fragments until a reconfigure event is received. This will make mssdemux expose all pads, but only download fragments for the streams that are actually being used. Relying on the pads being linked/unlinked isn't enough in this scenario as there might be an input-selector downstream that is actually discarding buffers for a given linked pad. When streams are switching, the old active stream can be blocked because input-selector will block not-linked streams. In case the mssdemux's stream loop is blocked pushing a buffer to a full queue downstream it will never unblock as the queue will not drain (input-selector is blocking). In this scenario, stream switching will deadlock as input-selector is waiting for the newly active stream data and the stream_loop that would push this data is blocked waiting for input-selector. To solve this issue, whenever an stream is reactivated on a reconfigure it will enter into the 'catch up mode', in this mode it can push buffers from its download thread until it reaches the currrent GstSegment's position. This works because this timestamp will always be behind or equal to the maximum timestamp pushed for all streams, after pushing data for this timestamp, the stream will go back to default and be pushed sequentially from the main streaming thread. By this time, the input-selector should have already released the thread. https://bugzilla.gnome.org/show_bug.cgi?id=711849
2013-11-12 12:58:31 +00:00
static GstFlowReturn
gst_mss_demux_combine_flows (GstMssDemux * mssdemux)
{
gboolean all_notlinked = TRUE;
GSList *iter;
for (iter = mssdemux->streams; iter; iter = g_slist_next (iter)) {
GstMssDemuxStream *stream = iter->data;
if (stream->last_ret != GST_FLOW_NOT_LINKED)
all_notlinked = FALSE;
if (stream->last_ret <= GST_FLOW_NOT_NEGOTIATED
|| stream->last_ret == GST_FLOW_FLUSHING)
return stream->last_ret;
}
if (all_notlinked)
return GST_FLOW_NOT_LINKED;
return GST_FLOW_OK;
}
static gboolean
gst_mss_demux_stream_push (GstMssDemuxStream * stream, GstBuffer * buf)
{
GstFlowReturn ret;
if (G_UNLIKELY (stream->pending_newsegment)) {
gst_pad_push_event (stream->pad, stream->pending_newsegment);
stream->pending_newsegment = NULL;
}
if (GST_BUFFER_TIMESTAMP (buf) != stream->next_timestamp) {
GST_DEBUG_OBJECT (stream->pad, "Marking buffer %p as discont buffer:%"
GST_TIME_FORMAT " != expected:%" GST_TIME_FORMAT, buf,
GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buf)),
GST_TIME_ARGS (stream->next_timestamp));
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DISCONT);
}
GST_DEBUG_OBJECT (stream->pad,
"Pushing buffer %p %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT
" discont:%d", buf,
GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buf)),
GST_TIME_ARGS (GST_BUFFER_DURATION (buf)),
GST_BUFFER_FLAG_IS_SET (buf, GST_BUFFER_FLAG_DISCONT));
stream->next_timestamp =
GST_BUFFER_TIMESTAMP (buf) + GST_BUFFER_DURATION (buf);
stream->have_data = TRUE;
stream->segment.position = GST_BUFFER_TIMESTAMP (buf);
ret = gst_pad_push (stream->pad, GST_BUFFER_CAST (buf));
GST_DEBUG_OBJECT (stream->pad, "Pushed. result: %d (%s)",
ret, gst_flow_get_name (ret));
return ret;
}
static gboolean
gst_mss_demux_stream_push_event (GstMssDemuxStream * stream, GstEvent * event)
{
gboolean ret;
if (GST_EVENT_TYPE (event) == GST_EVENT_EOS) {
stream->eos = TRUE;
}
GST_DEBUG_OBJECT (stream->pad, "Pushing event %" GST_PTR_FORMAT, event);
ret = gst_pad_push_event (stream->pad, event);
return ret;
}