Re: still more CLID_INUSE

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: cburnett@us.ibm.com
Date: 07/19/02-12:30:24 PM Z


From: cburnett@us.ibm.com
Subject: Re: still more CLID_INUSE
Message-ID: <OF41B30A69.5D36F1C8-ON85256BFB.005E2EBF@us.ibm.com>
Date: Fri, 19 Jul 2002 12:30:24 -0500



>>Peter wrote:
>>I'm still not convinced that the current way of doing SETCLIENTID is
>>optimal.  But, since we have much to do, I will stop complaining about
it,
>>at least until 4.0 is finished :-)



I would agree that SETCLIENTID has some holes that should addressed. I
would prefer to see this get addressed in 4.0 instead of being deffered.
Given this is the foundation upon which NFS V4 state mangement is built, it
seems like it needs to be complete the first time out of the chute.

Independent of when it is handled, shouldn't an item be added to the NFS
issues list for this? Here is my attempt to capture what I think are the
main issues. I hoped I have included the original problem Rick reported.

  SETCLIENTID does not handle the case of two clients
  independently generating the same nfs_client_id4.id value. While the
  spec encourages implementations to create unique values, the potential
  exists in the absence of an RFC dictated algorithm. Unless servers
  further qualify the client's signature with additional SETCLIENTID
  information such as the RPC authentication and/or client network
  information (callback info, reverse binding), the potential for
  indefinite CLID_INUSE errors exists.

  The RPC authentication is probably not a good choice as
  a distinguishing client property. Some environments will use the
  same credential across all the clients to avoid the creation of
  per client principals and the associated administrative overhead. In
  fact I would guess that AUTH_SYS (UID=0) will be the predominant choice.

  If the server chooses to use client network information as a
distinguishing
  feature, then support for multiple callback paths to a client with
  more than one network address is prohibited. This would preclude NFS
  implementations from supporting a desirable availability characteristic.

  If a client does receive repeated CLID_INUSE errors because of
  accidental id collision, the specification does not provide guidance on
  how a client might react. It seems reasonable, that a client could make
  a decision to generate a new id value and retry the SETCLIENTID operation
  after some delay period.

  Ideally, there is probably an opportunity to modify the SETCLIENTID
  "handshaking" mechanism to handle the coincidental occurrence of
   duplicate nfs_client_id4.id values given it is possible.


Carl


Carl Burnett
AIX Kernel Architecture - Distributed File Systems
(512) 838-8498, TL 678-8498
(please reply to cburnett@us.ibm.com)



|---------+---------------------------------->
|         |           Peter Astrand          |
|         |           <astrand@lysator.liu.  |
|         |           se>                    |
|         |           Sent by: owner-nfsv4-  |
|         |           wg@sunroof.eng.sun.com |
|         |                                  |
|         |                                  |
|         |           07/19/2002 03:47 AM    |
|         |                                  |
|---------+---------------------------------->
  >---------------------------------------------------------------------------------------------------------------|
  |                                                                                                               |
  |       To:       nfsv4-wg@sunroof.eng.sun.com                                                                  |
  |       cc:                                                                                                     |
  |       Subject:  Re: still more CLID_INUSE                                                                     |
  |                                                                                                               |
  |                                                                                                               |
  >---------------------------------------------------------------------------------------------------------------|





.
.
.
.


I'm still not convinced that the current way of doing SETCLIENTID is
optimal.  But, since we have much to do, I will stop complaining about it,
at least until 4.0 is finished :-)


--
/Peter Åstrand <astrand@lysator.liu.se>







graycol.gif


ecblank.gif


New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:50:02 AM Z CST