From owner-freebsd-hackers  Wed Jan 15 16:35:59 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.4/8.8.4) id QAA10123
          for hackers-outgoing; Wed, 15 Jan 1997 16:35:59 -0800 (PST)
Received: from awfulhak.demon.co.uk (awfulhak.demon.co.uk [158.152.17.1])
          by freefall.freebsd.org (8.8.4/8.8.4) with ESMTP id QAA10110
          for <hackers@freebsd.org>; Wed, 15 Jan 1997 16:35:55 -0800 (PST)
Received: from awfulhak.demon.co.uk (localhost.coverform.lan [127.0.0.1]) by awfulhak.demon.co.uk (8.8.4/8.7.3) with ESMTP id XAA14961; Wed, 15 Jan 1997 23:02:41 GMT
Message-Id: <199701152302.XAA14961@awfulhak.demon.co.uk>
X-Mailer: exmh version 1.6.9 8/22/96
To: Terry Lambert <terry@lambert.org>
cc: stesin@gu.net, karpen@ocean.campus.luth.se, hackers@freebsd.org
Subject: Re: truss, trace ?? 
In-reply-to: Your message of "Tue, 14 Jan 1997 09:46:17 MST."
             <199701141646.JAA29888@phaeton.artisoft.com> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Wed, 15 Jan 1997 23:02:40 +0000
From: Brian Somers <brian@awfulhak.demon.co.uk>
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Terry Lambert wrote:
> > > Cyclic file types imply record orientation.
> > [.....]
> > 
> > Unless of course you don't mind having a truncated line at the start
> > of your log ;)
> > 
> > This (IMHO) would be more "natural", 'cos there shouldn't be any
> > "record" knowledge at that level.
> 
> "Natural" for text files, maybe.
> 
> For the wtmp file, it's not so natural... it's wierd, even, since you
> can't resync to a valid record boundry without sync data built into
> the data format.
> 
> For text files, the sync data can be "after the first \n", since it
> is a variable length record format with "\n" record seperators, but
> how do you resync wtmp?

I'd tend to approach it with a new fs driver.  Using this driver, the only way 
to change a file size is using truncate().  A write() will always succeed and 
will wrap when hitting EOF.  The fs holds an internal "start" offset 
("end"+1), and lseek will seek to pos%filesize.  read() can return zero when 
it hits the internal "start" offset.

With this, fixed record lengths can be dealt with by making sure you 
truncate() the file to a multiple of the record size, and variable records are 
not supported (you end up with front-truncated lines).  The alternative is to 
maintain a whopping great list of write() sizes so that you can maintain a 
different "start" and "end" pointer.

The only inconsistency is that the following code is broken:

    off_t pos = lseek (fd, 0, 1);
    close (fd);
    fd = open( file, O_RDWR );
    lseek( fd, pos, 0 );

but then, doing this with log files and newsyslog running around moving them 
is dodgy anyway.


-- 
Brian <brian@awfulhak.demon.co.uk>, <brian@freebsd.org>
      <http://www.awfulhak.demon.co.uk/>
Don't _EVER_ lose your sense of humour....