explanation of 2)
scsi cdbs are of variable length 6 to 16 bytes.
iscsi always transfer the cdb as 16 bytes and pads it to 16 bytes if too short.
in most implementations of a scsi api you MUST specify the correct cdb
length when you pass the cdb over to a device suing an ioctl() or
similar.
The length of the cdb is NOT present in the iscsi header :-(
my own personal hack iscsi2scsi bridge i use at home solves the
problem by probing unknown scsi cdb's by trying to tell the scsi layer
its cdb length is 16,14,...,6 bytes until the ioctl() doesnt complain
about incorrect length any more.
pretty ugly.
if they had just had someone building bridges involved they would have
avoided people like me having to implement such ugly hacks to overcome
design bugs in the protocol.
On 4/20/06, ronnie sahlberg <ronniesahlberg@xxxxxxxxx> wrote:
> one thing that would be neat but obsolete/useless
> would be if someone could build a few old linux kernels clients and
> servers that support hyperscsi and create a number of capture
> files and also update the wiki for hyperscsi accordingly.
>
> while hyperscsi is a deadish protocol with the emergence of the (too
> complex for its own good and lacking) iscsi protocol it would be
> neat if we had good support for it.
>
> it is a nice protocol in many ways and would imho opinion be wastly
> superior to iscsi if they had finished converting it to hyperscsi over
> ip instead of over ethernet.
>
>
> (iscsi has lots of spare space in its header but still lacks an indication
> of
> 1, whether any digests are enabled or not, 2, the actual length of
> the scsi cdb making 1, tools like ethereal more complex trying to
> autodetect the settings if the login commands are not present in the
> capture and 2, building a iscsi to iscsi bridge more complex than
> nessecary since it would by nessecity either need to probe the ioctl
> for the correct cdb size (iscsi cdbs are always 16 bytes) and the
> packets dont even contain what command set it used :-( )
>
>
> ((that is what happens when non-ip people try to build an ip-standard))
>
>
> On 4/19/06, Gerald Combs <gerald@xxxxxxxxxxxx> wrote:
> > Google is accepting applications from mentoring organizations for the
> > 2006 Summer of Code until May 1:
> >
> > http://code.google.com/soc/mentorfaq.html#33
> >
> > Is there anything from the wishlist that would be suitable for a
> > student-level developer to work on over a three-month timeframe?
> > _______________________________________________
> > Ethereal-dev mailing list
> > Ethereal-dev@xxxxxxxxxxxx
> > http://www.ethereal.com/mailman/listinfo/ethereal-dev
> >
>