On Wed, Nov 19, 2003 at 02:03:41PM +0100, Lars Ruoff wrote:
> what exactly are
> typedef struct _frame_data {
> ...
> guint32 pkt_len; /* Packet length */
The length of the packet as it appeared "on the wire" (or, for 802.11,
"on the air" :-)), as reported by the driver.
> guint32 cap_len; /* Amount actually captured */
The amount of that packet data that was copied up to libpcap and thus to
Ethereal.
> struct _rtp_info {
> ...
> guint info_data_len;
> guint info_payload_offset;
> guint info_payload_len;
> };
>
> in packet-rtp.c, info_data_len is defined as
> rtp_info.info_data_len = tvb_reported_length_remaining( tvb, offset
);
That's based on the "reported length", which is, in turn, based on
"pkt_len", not "cap_len".
Note that if "tvb_reported_length_remaining(tvb, offset)" is greater
than "tvb_length_remaining(tvb, offset)", you do *NOT* have the full
payload.
> currently, i'm getting a pointer to the raw RTP packet by doing a
> sample.frame = cfile.pd + pinfo->fd->pkt_len - rtpinfo->info_data_len;
...
> And what about fragmentation?
If the "raw RTP packet" is constructed from multiple link-layer packets,
that won't work. "cfile.pd" points to a link-layer packet as read from
the capture file.
It should be possible to just put into the "struct _rtp_info" a "const
guint8 *" whose value is set in "dissect_rtp_data()" to
tvb_get_ptr(newtvb, 0, -1)
so that you get a *pointer* to the raw data, not some *offset* into
something you might not be able to get a handle on. That should be safe
because the "epan_dissect_t" constructed for the packet has not yet been
freed when the taps are called (destroying the "epan_dissect_t" will end
up freeing all the tvbuffs and hence invalidating pointers to their
data), because, in "add_packet_to_packet_list()", it calls
"tap_push_tapped_queue()" immediately after calling
"epan_dissect_run()", and well before calling "epan_dissect_free()".