Re: [linux-iscsi-devel] wish list: logout upon first login

[Mike C. <mi...@us...>]

Dominik L. Borkowski wrote:
> Hello,
> Since the open-iscsi and linux-iscsi seem to be merging, I figured I'll send 
> this e-mail to both lists. I apologize if that's not the most proper way of 
> approaching this.
> 
> Long story short, from our short experience with linux-iscsi drivers we have 
> noticed an interesting behavior. It seems that unlike windows iscsi driver, 
> the linux-iscsi initiator when first attaching to the target it does not try 
> to logout any previous sessions. We have addressed this problem with the 
> vendor of our iscsi targets, for which they are providing us with new 
> firmware which contains a simple workaround. However, some other folks may 
> not be so lucky with their vendors, so this may be something that you could 
> consider to be a new item on a wish list.
> 
> Here's a sample scenario which affected us, and explains the problem in a bit 
> more detail. The steps are in chronological order:
> 
> 1) linux initiator running linux-iscsi drivers attaches to an iscsi target. 
> Everything goes fine, storage is available, everybody is happy.
> 
> 2) for whatever reason, say a kernel panic, the initiator does not have a 
> chance to properly log out from the target.
> 
> 3) initiator gets booted up, tries to attach to the target. In our case, the 
> target decides that there was a previous session with the initiator already, 
> and refuses any further sessions with this initiator. 
> 

Have you run the current 4.0.X.Y linux-iscsi? If you do

rc.iscsi start
rc.iscsi stop
rc.iscsi start

Does this work for you? For a completely clean (in that the kernel does not
crash) shutdown and restart like above, we do not send a logout PDU.  Have
you tried the open-iscsi/5.0.0.X tree and done a

iscsiadm blah blah --login
iscsiadm blah blah --logout
iscsiadm blah blah --login

I am pretty sure it does not yet do a logout during clean shutdowns too.
So both of these paths look similar to your kernel crash scenario above.

The major difference is that the sockets may get closed nicely
in some cases.


> That's pretty much the gist of it. Our vendor had more experience with the 
> windows iscsi drivers, and according to them the windows initiators would try 
> to perform a logout first, then login. Whether that's bad, or the issue 
> should always be solved on the target's side, I leave it up for discussion. I 
> figured some of you may be interested.
> 
> 
> Sincerely,