From owner-freebsd-hackers Fri Jan 13 07:45:18 1995 Return-Path: hackers-owner Received: (from root@localhost) by freefall.cdrom.com (8.6.9/8.6.6) id HAA12055 for hackers-outgoing; Fri, 13 Jan 1995 07:45:18 -0800 Received: from labinfo.iet.unipi.it (labinfo.iet.unipi.it [131.114.9.5]) by freefall.cdrom.com (8.6.9/8.6.6) with ESMTP id HAA12019 for ; Fri, 13 Jan 1995 07:45:07 -0800 Received: from localhost (luigi@localhost) by labinfo.iet.unipi.it (8.6.5/8.6.5) id QAA00703; Fri, 13 Jan 1995 16:41:27 +0100 From: Luigi Rizzo Message-Id: <199501131541.QAA00703@labinfo.iet.unipi.it> Subject: Re: About readonly root partition To: tinguely@plains.nodak.edu (Mark Tinguely) Date: Fri, 13 Jan 1995 16:41:27 +0100 (MET) Cc: hackers@FreeBSD.org In-Reply-To: <199501131503.AA12584@plains.NoDak.edu> from "Mark Tinguely" at Jan 13, 95 09:02:41 am X-Mailer: ELM [version 2.4 PL23] Content-Type: text Content-Length: 2479 Sender: hackers-owner@FreeBSD.org Precedence: bulk > > In the interest of better support for diskless and in general for > > multiple installations, it seems to me that it would be nice if > > /etc were *not* on the same filesystem as root, but rather in /var or > > some other location, so that each machine can have its own copy. > > IMHO this should be a local change not a distributed change. Also IMHO, we > should push for DATALESS environments and discourage DISKLESS enviroments. > disks are far too cheap to have a 40-60 meg drive for boot,swap,tmp,parts of > var to save the network which will be saturated with the applications as it is. My points are that 1) the proposed change (or something equivalent) helps in having a readonly root, which is good for a lot of uses, including running the system from a CD, running experimental kernels withouth risking to trash your system files, etc. etc.; as a side effect, it helps having DISKLESS environments, which at times are useful. 2) it does not break the world: /etc/rc is still there, it just gives you an additional option at the cost of a couple of lines of code in /sbin/init. DATALESS vs DISKLESS is another topic of discussion. I agree that swap, tmp, var *must* be local if possible, but that is already supported (I like it a lot having local swap, either on a dedicated partition or on the DOS filesystem[haven't tried the latter], and all the non-permanent, writable portion of the filesystem on mfs which in turn goes to the local swap). The usefulness of a local boot partition is questionable (to me at least), as the kernel is loaded only once. Coming to programs, it becomes hard to say what is going to be local and what not. Probably the VM system should know what's local and what is not, to choose the best policy: as an example, I may be wrong but I think that executable files are not usually saved to backing store because it is considered cheaper to recover the code from the filesystem. This might be a bad idea if swap is local (as it should always be) and the file is on an NFS filesystem. David and John might have some better comments on this subject. Luigi ==================================================================== Luigi Rizzo Dip. di Ingegneria dell'Informazione email: luigi@iet.unipi.it Universita' di Pisa tel: +39-50-568533 via Diotisalvi 2, 56126 PISA (Italy) fax: +39-50-568522 ====================================================================