recording-daemon: Insert silence frames as needed during TLS forwarding

If the forwarded streams of a call are later combined without the inserted
silence, the resulting mix may end up with them out of sync. This
is already handled when outputting mixed files so the same functionality
from mix.c has been added for tls forwarding

lib/common.Makefile

@@ -41,7 +41,7 @@ $(DAEMONSRCS) $(HASHSRCS):	$(patsubst %,../daemon/%,$(DAEMONSRCS)) $(patsubst %,
 			-M "date:$(shell date -I)" \
 			-o "$@"
 
-resample.c codeclib.strhash.c mix.c:	fix_frame_channel_layout.h
+resample.c codeclib.strhash.c mix.c packet.c:	fix_frame_channel_layout.h
 
 ifeq ($(with_transcoding),yes)
 codec.c:	dtmf_rx_fillin.h

recording-daemon/decoder.c

@@ -171,6 +171,11 @@ no_recording:
 			ssrc->sent_intro = 1;
 		}
 
+		ssrc_tls_fwd_silence_frames_upto(ssrc, dec_frame, dec_frame->pts);
+		uint64_t next_pts = dec_frame->pts + dec_frame->nb_samples;
+		if (next_pts > ssrc->tls_in_pts)
+			ssrc->tls_in_pts = next_pts;
+
 		int linesize = av_get_bytes_per_sample(dec_frame->format) * dec_frame->nb_samples;
 		dbg("Writing %u bytes PCM to TLS", linesize);
 		streambuf_write(ssrc->tls_fwd_stream, (char *) dec_frame->extended_data[0], linesize);

recording-daemon/packet.c

@@ -17,6 +17,7 @@
 #include "streambuf.h"
 #include "resample.h"
 #include "tag.h"
+#include "fix_frame_channel_layout.h"
 
 
 static ssize_t ssrc_tls_write(void *, const void *, size_t);
@@ -144,6 +145,49 @@ void ssrc_tls_state(ssrc_t *ssrc) {
 }
 
 
+void ssrc_tls_fwd_silence_frames_upto(ssrc_t *ssrc, AVFrame *frame, int64_t upto) {
+	unsigned int silence_samples = ssrc->tls_fwd_format.clockrate / 100;
+
+	while (ssrc->tls_in_pts < upto) {
+		if (G_UNLIKELY(upto - ssrc->tls_in_pts > ssrc->tls_fwd_format.clockrate * 30)) {
+			ilog(LOG_WARN, "More than 30 seconds of silence needed to fill mix buffer, resetting");
+			ssrc->tls_in_pts = upto;
+			break;
+		}
+		if (G_UNLIKELY(!ssrc->tls_silence_frame)) {
+			ssrc->tls_silence_frame = av_frame_alloc();
+			ssrc->tls_silence_frame->format = ssrc->tls_fwd_format.format;
+			DEF_CH_LAYOUT(&ssrc->tls_silence_frame->CH_LAYOUT, ssrc->tls_fwd_format.channels);
+			ssrc->tls_silence_frame->nb_samples = silence_samples;
+			ssrc->tls_silence_frame->sample_rate = ssrc->tls_fwd_format.clockrate;
+			if (av_frame_get_buffer(ssrc->tls_silence_frame, 0) < 0) {
+				ilog(LOG_ERR, "Failed to get silence frame buffers");
+				return;
+			}
+			int planes = av_sample_fmt_is_planar(ssrc->tls_silence_frame->format) ? ssrc->tls_fwd_format.channels : 1;
+			for (int i = 0; i < planes; i++)
+				memset(ssrc->tls_silence_frame->extended_data[i], 0, ssrc->tls_silence_frame->linesize[0]);
+		}
+
+		dbg("pushing silence frame into TLS-formward stream (%lli < %llu)",
+				(long long unsigned) ssrc->tls_in_pts,
+				(long long unsigned) upto);
+
+		ssrc->tls_silence_frame->pts = ssrc->tls_in_pts;
+		ssrc->tls_silence_frame->nb_samples = MIN(silence_samples, upto - ssrc->tls_in_pts);
+		ssrc->tls_in_pts += ssrc->tls_silence_frame->nb_samples;
+
+		CH_LAYOUT_T channel_layout;
+		DEF_CH_LAYOUT(&channel_layout, ssrc->tls_fwd_format.channels);
+		ssrc->tls_silence_frame->CH_LAYOUT = channel_layout;
+
+		int linesize = av_get_bytes_per_sample(frame->format) * ssrc->tls_silence_frame->nb_samples;
+		dbg("Writing %u bytes PCM to TLS", linesize);
+		streambuf_write(ssrc->tls_fwd_stream, (char *) ssrc->tls_silence_frame->extended_data[0], linesize);
+	}
+}
+
+
 void ssrc_close(ssrc_t *s) {
 	output_close(s->metafile, s->output, tag_get(s->metafile, s->stream->tag), s->metafile->discard);
 	s->output = NULL;
@@ -156,6 +200,7 @@ void ssrc_close(ssrc_t *s) {
 
 void ssrc_free(void *p) {
 	ssrc_t *s = p;
+	av_frame_free(&s->tls_silence_frame);
 	packet_sequencer_destroy(&s->sequencer);
 	ssrc_close(s);
 	g_slice_free1(sizeof(*s), s);

recording-daemon/packet.h

@@ -2,6 +2,7 @@
 #define _PACKET_H_
 
 #include "types.h"
+#include <libavutil/frame.h>
 
 void ssrc_close(ssrc_t *s);
 void ssrc_free(void *p);
@@ -9,5 +10,6 @@ void ssrc_free(void *p);
 void packet_process(stream_t *, unsigned char *, unsigned len);
 
 void ssrc_tls_state(ssrc_t *ssrc);
+void ssrc_tls_fwd_silence_frames_upto(ssrc_t *ssrc, AVFrame *frame, int64_t upto);
 
 #endif

recording-daemon/types.h

@@ -87,6 +87,8 @@ struct ssrc_s {
 	format_t tls_fwd_format;
 	resample_t tls_fwd_resampler;
 	socket_t tls_fwd_sock;
+	uint64_t tls_in_pts;
+	AVFrame *tls_silence_frame;
 	//BIO *bio;
 	SSL_CTX *ssl_ctx;
 	SSL *ssl;