Improve SCO socket read error handling

A combination of buggy adapter firmware and bugs in the BlueZ drivers,
especially with older kernels, means that SCO sockets can return
unexpected errors. This commit makes our SCO decoders more robust in
such cases.

src/sco.c

@@ -321,21 +321,29 @@ static void *sco_cvsd_dec_thread(struct ba_transport_thread *th) {
 
 		if ((size_t)len == buffer.nmemb * buffer.size) {
 			debug("Resizing CVSD read buffer: %zd -> %zd",
-					buffer.nmemb * buffer.size, buffer.nmemb * buffer.size * 2);
+					buffer.nmemb * buffer.size, buffer.nmemb * 2 * buffer.size);
 			if (ffb_init_int16_t(&buffer, buffer.nmemb * 2) == -1)
 				error("Couldn't resize CVSD read buffer: %s", strerror(errno));
 		}
 
 		if (!ba_transport_pcm_is_active(t_pcm))
 			continue;
 
-		ssize_t samples = len / sizeof(int16_t);
+		if (len > 0)
+			ffb_seek(&buffer, len / buffer.size);
+
+		ssize_t samples;
+		if ((samples = ffb_len_out(&buffer)) <= 0)
+			continue;
+
 		io_pcm_scale(t_pcm, buffer.data, samples);
 		if ((samples = io_pcm_write(t_pcm, buffer.data, samples)) == -1)
 			error("FIFO write error: %s", strerror(errno));
 		else if (samples == 0)
 			ba_transport_stop_if_no_clients(t);
 
+		ffb_shift(&buffer, samples);
+
 	}
 
 exit:
@@ -457,7 +465,8 @@ static void *sco_msbc_dec_thread(struct ba_transport_thread *th) {
 		if (!ba_transport_pcm_is_active(t_pcm))
 			continue;
 
-		ffb_seek(&msbc.data, len);
+		if (len > 0)
+			ffb_seek(&msbc.data, len);
 
 		int err;
 		if ((err = msbc_decode(&msbc)) < 0) {