full-disclosure - Bigger burger roll needed

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20051003200945.GA29951@linux.unixwiz.net>
Date: Mon Oct  3 21:09:53 2005
From: steve at unixwiz.net (Steve Friedl)
Subject: Bigger burger roll needed

On Mon, Oct 03, 2005 at 03:41:58PM -0400, TheGesus wrote:
> In NT4 they redesigned the GDI so that the user could bypass
> "userland" and talk straight to the kernel.
> 
> It's been so long I don't recall the exact details, but this re-hack
> paved the way for  DirectX and sped up the response of the new
> desktop, which of course was lifted from WIndows 95.
> 
> After NT4 anything that hooked into the GDI could BSOD.  New video
> driver?  BSOD.  New printer driver?  BSOD.  It was quite a mess.

This is only partly the case; this is the history.

In NT3.51, *all* GDI (printer and video) was done in userland, but
GDI calls involved an expensive context switch and/or local procedure
call. I guess for printer drivers this was not really a big deal,
but for video it matters a lot. Gamers care about this, right?

In NT4 all GDI dove into kernel space, and it provided a substantial
performance boost, but it completely sucked for print driver writers.
No thread support, no real support for floating point math, not any
performance difference to write home about, and a BSOD was as easy
as an assertion failure. Porting a complex user-mode driver to
kernel mode could be a daunting task.

Well, all that silly "but kernel-mode print drivers won't be as robust"
talk turned out to be true, so Windows 2000 supported both kernel mode
(version 2) and user mode (version 3) drivers. I assume that version 1
drivers were NT3.51 usermode.

XP is the same way, and in Server 2003 there is a Group Policy option
that disables kernel mode drivers, and I understand that Vista/Longhorn
will forbid kernel mode print drivers altogether.

Saying that the user "bypasses user mode and talks directly to the
kernel" is not really that meaningful: it doesn't talk "to the kernel",
just to the GDI, and it's not really any different from an IOCTL.

It wasn't terribly robust, but I don't think it was inherently insecure.

Steve (who writes print drivers too)

--- 
Stephen J Friedl | Security Consultant |  UNIX Wizard  |   +1 714 544-6561
www.unixwiz.net  | Tustin, Calif. USA  | Microsoft MVP | steve@...xwiz.net