RE: detection of failed sessions to allow re-login

Hi Paul,
 
The full clause states the "initiator session state MUST be FAILED". The
initiator's perspective is different to the target's in this scenario.
The initiator has no session state (it's just been powered on), but the
target is not aware the HBA has been reset.
 
The target catches up with the initiator's view by detecting the TSIH=0,
which means the initiator denies all knowledge of any previous session
state. This is one of the specific scenarios that session reinstatement
was invented to resolve.
 
Cheers
Ken


________________________________

From: Paul Hughes [mailto:[email protected]] 
Sent: Friday, 20 April 2007 11:37
To: Sandars, Ken; [email protected]
Subject: RE: [Ips] detection of failed sessions to allow re-login


Isn't session re-instatement possible only for sessions that are in the
failed state (no active connections)?  That was my understanding of
Section 5.3.5 of RFC 3720.
 
How can the target know that the session is in the failed state if it
hasn't detected that the only connection was dropped?
 
Thanks,
Paul
 

________________________________

From: Sandars, Ken [mailto:[email protected]] 
Sent: Thursday, April 19, 2007 6:56 PM
To: Paul Hughes; [email protected]
Subject: RE: [Ips] detection of failed sessions to allow re-login


Hi Paul,
 
That's a target problem, most likely a bug. The second login with TSIH=0
tells the target to perform session reinstatement. That's jargon for
"silently nuke the first session".
 
The target may be failing the login because it's internal cleanup
requires more time (I/O requests from the previous session are jammed
for instance). Your scenario suggests this is not likely.
 
HTH,
Ken

________________________________

From: Paul Hughes [mailto:[email protected]] 
Sent: Friday, 20 April 2007 09:14
To: [email protected]
Subject: [Ips] detection of failed sessions to allow re-login


I have a question about how a target can quickly detect session failures
so that a re-login can succeed.
 
Here's my scenario:
 
1) an initiator is booting from an iSCSI target
2) the initiator is using an iSCSI HBA to communicate with the iSCSI
target
3) the HBA BIOS creates the first session, discovers the boot LUN, and
reads the boot loader
4) the boot loader reads the kernel from the boot LUN
5) the kernel resets the iSCSI HBA while loading an HBA driver
6) the HBA driver attempts to create a new session
 
The problem I'm seeing is that the target is failing the login for the
new session because the target thinks the first session created by the
HBA BIOS is still valid (not in failed state).  The HBA reset was not
detected by the target soon enough for the target to know that the first
session is now in the failed state when the initiator attempts to login
and create the second session using the same InitiatorName, ISID,
TargetName, and TargetPortalGroupTag as the first session (with TSIH=0).
The target does not see a link down event because a switch is connected
between the HBA and the target port.  The target eventually detects that
the first session is failed when it sends a NOP-Out PDU and receives a
transport failure.  Unfortunately, this occurs too late and the boot
fails.
 
In my case the target is sending NOP-Out PDUs every 60 seconds.  I can
change that to 5 seconds, but I don't think that will fix every case.
Is there a better way for the target to determine that the first session
has failed so that a re-login will succeed on the first try?
 
Thanks,
Paul

_______________________________________________
Ips mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ips