[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
  Web www.spinics.net

RE: session re-instatement

Title: session re-instatement

You may have asynchronous events or I/O activity still taking place on the failed session, it may help you to have a flag or a state that marks that session as being in reinstatement.  Depending on your implementation, this may help you avoid taking action on an event that that no longer applicable nor has a chance of success on a failed session. 


You may even keep an internal pointer to the session block/state that is supplanting the failed session.


The key thing to marking the session as failed is to avoid race conditions between the failed session and the spanking new session that is about to replace it..



[ formerly at Silverback Systems, iSCSI pioneers.. :) ]

Carlos Rimola
Sr. Principal Software Engineer
Symantec - Next Generation Storage
Tel.: 650-527-4505
Cell: 408-858-5548

From: Eddy Quicksall [mailto:Quicksall_iSCSI@xxxxxxxxxxxxx]
Sent: Tuesday, February 27, 2007 8:15 AM
To: Sandars, Ken; Paul Hughes; IP Storage Mailing List (E-mail)
Subject: Re: session re-instatement


I think the correct way is what Ken says. In simple terms don't put the reinstatement logic in the place where you receive the first PDU; instead let it go through the full login and when that is finished check to see if this is a reinstatement. Temporarily you will have two sessions that will appear to conflict but before you go to FFP you will resolve the problem and end up with only one session.

----- Original Message -----

From: Sandars, Ken

Sent: Monday, February 26, 2007 9:26 PM

Subject: RE: session re-instatement


Hi Paul,


Section 5.3.5 [RFC3720] para 2 states:

The initiator session state MUST be FAILED (Section 7.3 Session State Diagrams) when the initiator attempts a session reinstatement.

In your scenario the initiator violates this rule since both are in state FREE (Q1).


The part of your question explores what happens at the target if this scenario does occur.


The connection and session state machines for the target show that the session begins life in the ACTIVE state (Q2) when the login phase of the leading connection starts (S4). Once the leading connection transits (T5) to LOGGED_IN (S5), the session performs the equivalent transit (N2) to LOGGED_IN(Q3).


In your scenario, there are two new sessions in ACTIVE state, racing towards LOGGED_IN. A sensible target implementation waits until a connection makes the T5 transition (corresponding to the session's N2 transition) before enacting the session reinstatement action. Afterall, it would be a shame to nuke a potentially viable session with a dodgy login attempt.


Some sort of implementation-specific mutual exclusion mechanism must be used on the target to ensure only one party manipulates the session state. Hence one connection reaches the T5 transition point and runs the session reinstatement algorithm.


At this point the other session is in Q2 and its connection is in S4 (albeit on the cusp of its transition). While not listed as a T7 reason, I'd think a resilient target would take that action on the connection. The session would make the N9 transition (again, not explicitly listed as a reason, but follows as a consequence of the T7 transit).


In other words, the first one to reach full feature phase survives and the other gets trashed. Sound fair?





From: Paul Hughes [mailto:phughes@xxxxxxxxxxxxxx]
Sent: Tuesday, 27 February 2007 03:23
To: IP Storage Mailing List (E-mail)
Subject: session re-instatement

Is there any text in RFC 3720 or the implementer's guide that prevents an initiator from simultaneously logging-in on two connections to the same target portal using the same InitiatorName, ISID, and TSIH=0?  If the logins were done sequentially, the second login would re-instate the session created by the first login, but I can see how a target might not do session re-instatement if the sessions are being created simultaneously.


Ips mailing list

Ips mailing list

[IETF]     [Linux iSCSI]     [Linux SCSI]     [Linux Resources]     [Yosemite News]     [IETF Announcements]     [IETF Discussion]     [SCSI]

Add to Google Powered by Linux