struct PayloadContext {
AVIOContext *data;
- uint32_t timestamp;
+ uint32_t timestamp;
+ int is_keyframe;
+ /* If sequence_ok is set, we keep returning data (even if we might have
+ * lost some data, but we haven't lost any too critical data that would
+ * cause the decoder to desynchronize and output random garbage).
+ */
+ int sequence_ok;
+ int first_part_size;
+ uint16_t prev_seq;
+ int prev_pictureid;
+ int broken_frame;
+ /* If sequence_dirty is set, we have lost some data (critical or
+ * non-critical) and decoding will have some sort of artefacts, and
+ * we thus should request a new keyframe.
+ */
+ int sequence_dirty;
+ int got_keyframe;
};
-static int vp8_handle_packet(AVFormatContext *ctx,
- PayloadContext *vp8,
- AVStream *st,
- AVPacket *pkt,
- uint32_t *timestamp,
- const uint8_t *buf,
- int len, int flags)
+static void vp8_free_buffer(PayloadContext *vp8)
+{
+ uint8_t *tmp;
+ if (!vp8->data)
+ return;
+ avio_close_dyn_buf(vp8->data, &tmp);
+ av_free(tmp);
+ vp8->data = NULL;
+}
+
+static int vp8_broken_sequence(AVFormatContext *ctx, PayloadContext *vp8,
+ const char *msg)
+{
+ vp8->sequence_ok = 0;
+ av_log(ctx, AV_LOG_WARNING, "%s", msg);
+ vp8_free_buffer(vp8);
+ return AVERROR(EAGAIN);
+}
+
+static int vp8_handle_packet(AVFormatContext *ctx, PayloadContext *vp8,
+ AVStream *st, AVPacket *pkt, uint32_t *timestamp,
+ const uint8_t *buf, int len, uint16_t seq,
+ int flags)
{
int start_partition, end_packet;
int extended_bits, part_id;
int pictureid_present = 0, tl0picidx_present = 0, tid_present = 0,
keyidx_present = 0;
+ int pictureid = -1, pictureid_mask = 0;
+ int returned_old_frame = 0;
+ uint32_t old_timestamp;
+
+ if (!buf) {
+ if (vp8->data) {
+ int ret = ff_rtp_finalize_packet(pkt, &vp8->data, st->index);
+ if (ret < 0)
+ return ret;
+ *timestamp = vp8->timestamp;
+ if (vp8->sequence_dirty)
+ pkt->flags |= AV_PKT_FLAG_CORRUPT;
+ return 0;
+ }
+ return AVERROR(EAGAIN);
+ }
if (len < 1)
return AVERROR_INVALIDDATA;
len--;
}
if (pictureid_present) {
- int size;
if (len < 1)
return AVERROR_INVALIDDATA;
- size = buf[0] & 0x80 ? 2 : 1;
- buf += size;
- len -= size;
+ if (buf[0] & 0x80) {
+ if (len < 2)
+ return AVERROR_INVALIDDATA;
+ pictureid = AV_RB16(buf) & 0x7fff;
+ pictureid_mask = 0x7fff;
+ buf += 2;
+ len -= 2;
+ } else {
+ pictureid = buf[0] & 0x7f;
+ pictureid_mask = 0x7f;
+ buf++;
+ len--;
+ }
}
if (tl0picidx_present) {
// Ignoring temporal level zero index
if (len < 1)
return AVERROR_INVALIDDATA;
- if (start_partition && part_id == 0) {
+ if (start_partition && part_id == 0 && len >= 3) {
int res;
- if (vp8->data) {
- uint8_t *tmp;
- avio_close_dyn_buf(vp8->data, &tmp);
- av_free(tmp);
- vp8->data = NULL;
+ int non_key = buf[0] & 0x01;
+ if (!non_key) {
+ vp8_free_buffer(vp8);
+ // Keyframe, decoding ok again
+ vp8->sequence_ok = 1;
+ vp8->sequence_dirty = 0;
+ vp8->got_keyframe = 1;
+ } else {
+ int can_continue = vp8->data && !vp8->is_keyframe &&
+ avio_tell(vp8->data) >= vp8->first_part_size;
+ if (!vp8->sequence_ok)
+ return AVERROR(EAGAIN);
+ if (!vp8->got_keyframe)
+ return vp8_broken_sequence(ctx, vp8, "Keyframe missing\n");
+ if (pictureid >= 0) {
+ if (pictureid != ((vp8->prev_pictureid + 1) & pictureid_mask)) {
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed a picture, sequence broken\n");
+ } else {
+ if (vp8->data && !can_continue)
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed a picture, sequence broken\n");
+ }
+ } else {
+ uint16_t expected_seq = vp8->prev_seq + 1;
+ int16_t diff = seq - expected_seq;
+ if (vp8->data) {
+ // No picture id, so we can't know if missed packets
+ // contained any new frames. If diff == 0, we did get
+ // later packets from the same frame (matching timestamp),
+ // so we know we didn't miss any frame. If diff == 1 and
+ // we still have data (not flushed by the end of frame
+ // marker), the single missed packet must have been part
+ // of the same frame.
+ if ((diff == 0 || diff == 1) && can_continue) {
+ // Proceed with what we have
+ } else {
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed too much, sequence broken\n");
+ }
+ } else {
+ if (diff != 0)
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed unknown data, sequence broken\n");
+ }
+ }
+ if (vp8->data) {
+ vp8->sequence_dirty = 1;
+ if (avio_tell(vp8->data) >= vp8->first_part_size) {
+ int ret = ff_rtp_finalize_packet(pkt, &vp8->data, st->index);
+ if (ret < 0)
+ return ret;
+ pkt->flags |= AV_PKT_FLAG_CORRUPT;
+ returned_old_frame = 1;
+ old_timestamp = vp8->timestamp;
+ } else {
+ // Shouldn't happen
+ vp8_free_buffer(vp8);
+ }
+ }
}
+ vp8->first_part_size = (AV_RL16(&buf[1]) << 3 | buf[0] >> 5) + 3;
if ((res = avio_open_dyn_buf(&vp8->data)) < 0)
return res;
vp8->timestamp = *timestamp;
- }
+ vp8->broken_frame = 0;
+ vp8->prev_pictureid = pictureid;
+ vp8->is_keyframe = !non_key;
+ } else {
+ uint16_t expected_seq = vp8->prev_seq + 1;
- if (!vp8->data || vp8->timestamp != *timestamp) {
- av_log(ctx, AV_LOG_WARNING,
- "Received no start marker; dropping frame\n");
- return AVERROR(EAGAIN);
+ if (!vp8->sequence_ok)
+ return AVERROR(EAGAIN);
+
+ if (vp8->timestamp != *timestamp) {
+ // Missed the start of the new frame, sequence broken
+ return vp8_broken_sequence(ctx, vp8,
+ "Received no start marker; dropping frame\n");
+ }
+
+ if (seq != expected_seq) {
+ if (vp8->is_keyframe) {
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed part of a keyframe, sequence broken\n");
+ } else if (vp8->data && avio_tell(vp8->data) >= vp8->first_part_size) {
+ vp8->broken_frame = 1;
+ vp8->sequence_dirty = 1;
+ } else {
+ return vp8_broken_sequence(ctx, vp8,
+ "Missed part of the first partition, sequence broken\n");
+ }
+ }
}
- avio_write(vp8->data, buf, len);
+ if (!vp8->data)
+ return vp8_broken_sequence(ctx, vp8, "Received no start marker\n");
+
+ vp8->prev_seq = seq;
+ if (!vp8->broken_frame)
+ avio_write(vp8->data, buf, len);
+
+ if (returned_old_frame) {
+ *timestamp = old_timestamp;
+ return end_packet ? 1 : 0;
+ }
if (end_packet) {
- int ret = ff_rtp_finalize_packet(pkt, &vp8->data, st->index);
+ int ret;
+ ret = ff_rtp_finalize_packet(pkt, &vp8->data, st->index);
if (ret < 0)
return ret;
+ if (vp8->sequence_dirty)
+ pkt->flags |= AV_PKT_FLAG_CORRUPT;
return 0;
}
static PayloadContext *vp8_new_context(void)
{
- return av_mallocz(sizeof(PayloadContext));
+ PayloadContext *vp8 = av_mallocz(sizeof(PayloadContext));
+ if (!vp8)
+ return NULL;
+ vp8->sequence_ok = 1;
+ return vp8;
}
static void vp8_free_context(PayloadContext *vp8)
{
- if (vp8->data) {
- uint8_t *tmp;
- avio_close_dyn_buf(vp8->data, &tmp);
- av_free(tmp);
- }
+ vp8_free_buffer(vp8);
av_free(vp8);
}
+static int vp8_need_keyframe(PayloadContext *vp8)
+{
+ return vp8->sequence_dirty || !vp8->sequence_ok;
+}
+
RTPDynamicProtocolHandler ff_vp8_dynamic_handler = {
.enc_name = "VP8",
.codec_type = AVMEDIA_TYPE_VIDEO,
.alloc = vp8_new_context,
.free = vp8_free_context,
.parse_packet = vp8_handle_packet,
+ .need_keyframe = vp8_need_keyframe,
};