On Fri, Apr 17, 2009 at 6:19 AM, Milos Muzik <[email protected]> wrote:

>
> Alastair Neil wrote:
> <snip>
> >
> > I succeeded in getting wireshark in stalled on the linux target server
> > and did a capture from there (attached).  I'm not an expert but this
> > jumped out at me as a likely cause:
> >
> >
> >     8.784473 192.168.0.135 -> 192.168.0.45 iSCSI SCSI: Persistent
> >     Reserve In LUN: 0x00
> >       8.784859 192.168.0.45 -> 192.168.0.135 iSCSI SCSI Response (Check
> >     Condition) LUN:0x00 [Unreassembled Packet [incorrect TCP checksum]]
> >       8.850272 192.168.0.135 -> 192.168.0.45 TCP 49622 > iscsi-target
> >     [ACK] Seq=1737 Ack=19641 Win=277384 Len=0 TSV=2601 TSER=798756713
> <snip>
> > It looks like a patch to support support for persistent reservations in
> > the linux kernel was submitted in January 30th this year so it is likely
> > that this is simply an unsupported feature at the moment.  My question
> > is now will this cause problems?  Is there a way to disable the
> > behaviour in the initiator?
> >
> >
> > Thanks, Alastair Neil
>
> Hi Alastair,
>
> those commands "Persistent Reserve In" are likely sent by the initiator to
> test
> if the targets support persistent reservations. I don't think it is a
> problem
> that your targets do not support it.
>
> I wonder what is the actual problem. Is it just appearance of those
> messages
> about not enough sense information? Are you eventually able to use the
> iSCSI
> disks? The net traffic dump you attached suggests that at least the login
> phase
> was successful.
>
> Anyway, to investigate if some sense data were corrupted you need more
> verbose
> output from wireshark. The information you are looking for is on SCSI
> layer. As
> there are no Request Sense commands then probably autosensing is used.
> Hence try
> to look at responses with Check Condition if they contain valid sense data.
>
> Wireshark also supports import from Solaris' snoop(1M) output. So you can
> analyze network traffic on the initiator side as well. Use this command to
> do
> that:
> snoop -x 0 -d <interface> -o <output file>
>
> Another option for tracing SCSI commands is to use nice scsi.d dtrace
> script
> http://blogs.sun.com/chrisg/entry/scsi_d_script
>
> Regards,
> Milos



Thanks for the tips,  yes the targets are functional, I worry when I see
error messages.  I'll try one or both of the things you suggest to see if i
can get to the root of the issue.

Thanks again,

Alastair
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to