From: Mike Eisler (mike@eisler.com)
Date: 12/20/02-12:36:01 PM Z
Message-ID: <3E036311.2060801@eisler.com>
Date: Fri, 20 Dec 2002 10:36:01 -0800
From: Mike Eisler <mike@eisler.com>
Subject: Re: V4 complexity (was re: Proposed charter addition)
Neil Brown wrote:
>On Thursday December 19, mike@eisler.com wrote:
>
>>For symlinks, named pipes, sockets, and special devices,
>>mostly (more on this modifier later) persistent file handles
>>are still a reasonable implementation choice if for no
>>other reason in that adding OPEN operations for those
>>objects is overkill (to me at least).
>>
>
>For symlinks there is no concept of OPEN. For the rest, the open is
>entirely implemented on the client and the server plays no part. So I
>think non-persistence is fine for them.
>
Clients will want to cache the file handles
if they can (and they won't bother if ANY volatile bit is on) in a name
cache. So for performance reasons, the aforementioned objects
ought to be mostly persistent. I understand that for the Linux NFS server,
there are security reasons that dictate the use of volatile file handles.
Given that symlinks are readable by all, and that access control
to the other objects is a client side thing, such security issues shouldn't
apply to the special objects. But I may not understand the Linux issues
correctly or completely.
>>
>>The faults of NFSv4 aren't in Sun (Inc.), they are in ourselves.
>>
>
>
>Please accept my apologies if it offended.
>
No offence here; I don't work for Sun.
-mre
This archive was generated by hypermail 2.1.2 : 03/04/05-01:50:44 AM Z CST