[182] in arla-drinkers
Re: patch to fix symlink-related GPFs on Linux
daemon@ATHENA.MIT.EDU (Magnus Ahltorp)
Tue Aug 4 18:52:07 1998
From owner-arla-drinkers@stacken.kth.se Tue Aug 04 22:52:07 1998
Return-Path: <owner-arla-drinkers@stacken.kth.se>
Delivered-To: arla-drinkers-mtg@bloom-picayune.mit.edu
Received: (qmail 19423 invoked from network); 4 Aug 1998 22:52:06 -0000
Received: from unknown (HELO sundance.stacken.kth.se) (130.237.234.41)
by bloom-picayune.mit.edu with SMTP; 4 Aug 1998 22:52:06 -0000
Received: (from majordom@localhost)
by sundance.stacken.kth.se (8.8.8/8.8.8) id AAA20550
for arla-drinkers-list; Wed, 5 Aug 1998 00:46:27 +0200 (MET DST)
Received: from yakko.stacken.kth.se (map@yakko.stacken.kth.se [130.237.234.52])
by sundance.stacken.kth.se (8.8.8/8.8.8) with ESMTP id AAA20546;
Wed, 5 Aug 1998 00:46:16 +0200 (MET DST)
Received: (from map@localhost)
by yakko.stacken.kth.se (8.8.8/8.8.7) id AAA21001;
Wed, 5 Aug 1998 00:45:24 +0200 (CEST)
To: amu@mit.edu (Aaron M. Ucko)
Cc: arla-drinkers@stacken.kth.se
Subject: Re: patch to fix symlink-related GPFs on Linux
References: <udlww8ssa67.fsf@opus.mit.edu>
From: Magnus Ahltorp <map@stacken.kth.se>
Date: 05 Aug 1998 00:45:23 +0200
In-Reply-To: amu@MIT.EDU's message of 01 Aug 1998 16:45:52 -0400
Message-ID: <lv1pvegnz7g.fsf@yakko.stacken.kth.se>
Lines: 33
X-Mailer: Gnus v5.3/Emacs 19.34
Sender: owner-arla-drinkers@stacken.kth.se
Precedence: bulk
> For the record, the readpage solution as implemented in 0.9 seems to
> work pretty well on my system. There are still some glitches,
> though. In particular:
>
> I can't unmount /afs either before or after killing arlad.
This is "normal". This is due to the xfs_cache module (it keeps a
reference count on all cached nodes). The unmounting sometimes works
when killing arlad, but obviously it doesn't work for you. I'll maybe
look into it some day, but I cannot say Linux 2.0 is very high on my
priority list. You or someone also maybe wants to have a look at it?
> Programs sometimes have trouble finding their current directories if
> they aren't world-listable.
This is also due to, or maybe I should say, in spite of, the cache. As
the cache size approaches infinite, the problem frequency should
approach 0. This problem only exists with Linux 2.0, since the inode
numbers must be (in some sense) "correct" for getcwd to work (because
of that horrid method of doing getcwd).
One way of solving that problem is to write a cache that keeps a
reference count on how many entries a certain directory has, and never
throw a directory that is "in use". The problem is that the now used
bucket caching will have to be thrown out, and a more general hash
caching with external linking will have to be used instead. As I have
said, Linux 2.0 support is not very much worked on.
I will try to look into some of these things now that I have come back
from my trip to France.
/Magnus
map@stacken.kth.se
