Re: pcap DLT request for virtio-scsi SCSI transport

[Stefan Hajnoczi <stefanha () gmail com>]

On Tue, Mar 6, 2012 at 6:54 PM, Guy Harris <guy () alum mit edu> wrote:
> On Mar 3, 2012, at 3:38 AM, Stefan Hajnoczi wrote:
>
> > There are SCSI commands and responses.  Commands and responses are
> > separate pcap packets because there can be multiple outstanding
> > commands to multiple targets/LUNs.
> >
> > From the spec, commands have the following layout:
> >
> >    u8 lun[8];
> >    u64 id;
> >    u8 task_attr;
> >    u8 prio;
> >    u8 crn;
> >    char cdb[cdb_size];
>
> (Presumably the CDB size depends on the command.)
>
> >    char dataout[];
> >
> > Responses have the following layout:
> >
> >    u32 sense_len;
> >    u32 residual;
> >    u16 status_qualifier;
> >    u8 status;
> >    u8 response;
> >    u8 sense[sense_size];
> >    char datain[];
> >
> > The pcap mapping of this layout should be straightforward.
>
> So does a command packet begin with 8 bytes of LUN, and a response packet begin with 4 bytes of sense length?
>
> If so, how do you distinguish between a command packet and a response packet?
>
> Or is there additional metadata at the beginning of the packet to indicate whether a packet is a command or a 
> response?

You are right, additional metadata is needed.

We need a pcap encapsulation because the virtio-scsi device itself can
separate differences between command/response and other transport
events due to the nature of virtio.  The device itself has separate
queues and a notification mechanism, so that all commands are placed
on one specific queue, all responses are received on another queue,
etc.  This implicit information needs to be encoded for pcap - we need
a wrapper structure which properly identifies command/response/events.

Sorry if I've been vague about the pcap layout of commands and
responses, there is no defined mapping yet.

Ronnie has drafted up some recommendations and Cong is writing a
specification based on this to describe the pcap layout in detail.
This is work in progress but we'll get back to you with the pcap
specification.

Stefan
-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.