From: cgf AT cygnus DOT com (Christopher Faylor)
Subject: Re: Are lseek changes correct?
13 Jul 1998 07:39:31 -0700
Message-ID: <199807131422.KAA27088.cygnus.cygwin32.developers@kramden.cygnus.com>
To: cygwin32-developers AT cygnus DOT com, sos AT prospect DOT com DOT ru

>From: Sergey Okhapkin <sos AT prospect DOT com DOT ru>
>Date: Mon, 13 Jul 1998 10:33:33 +0400
>
>Wed Jul  8 15:05:10 1998  DJ Delorie  <dj AT cygnus DOT com>
>
>	* fhandler.cc (fhandler_base::lseek): Note lseek so that next
>	write() can check for the Win95 "gap" bug.
>	(fhandler_base::write): If Win95 and lseek past eof
>	followed by write, use WriteFile to force the "gap" to be filled
>	with zeros rather than left to the "undefined" data Win32 specifies.
>	(fhandler_base::fhandler_base): initialize check_win95_lseek_bug_.
>	* fhandler.h (class fhandler_base): Add check_win95_lseek_bug_
>	for bug: when seek past EOF and write, win95 fills with random
>	data (security hole).
>
>Shouldn't check_win95_lseek_bug_ be saved in linearize and restored in 
>delinearize? Lseek() may be performed in parent, but write() - in execed 
>child.

Nice catch, Sergey.  This should be fixed in the next snapshot.  It
will be caught automatically (I hope) by my new, improved version of
the linearize/de_linearize functions.

-Chris