From: mixmaster@remail.obscura.com (Mixmaster)
Date: Fri, 30 Aug 1996 21:41:09 +0800
To: cypherpunks@toad.com
Subject: Ian's Linux Filesystem Patches
Message-ID: <199608300740.AAA05386@sirius.infonex.com>
MIME-Version: 1.0
Content-Type: text/plain


I installed Ian's filesystem patches for Linux 2.0.11, and then patched the
kernel up to 2.0.15. Unfortunately I soon realized that the entire system 
grinds to a halt as the kernel performs cryptographic operations on a block
of data. 

I lack enough crypto expertise to fix this, but I believe it just needs to
be made preemptable, however that works. The filesystem patch, while a
definite step in the right direction, is all but useless in its present form.

For an idea of what I'm talking about, kill your turbo switch, or install on
a slow machine so that this is a bit more noticable. Users may want to kill
their internal chip cache. Patch the kernel up to 2.0.15, and install the 
kernel. Setup the loopback device as the README file states. Then make a 
filesystem on that device. The system will slow to a halt about the time that 
it actually needs to write some data to the filesystem, probably during the 
superblock write in mke2fs. Then, mount the filesystem and try to write a 
large block of data to the device. Again, it dies.

This will need to be fixed if everyone wants to install such a patch into
their kernel to reduce suspicion, or if it is to be included as standard in
the kernel source tree. I would fix this if I knew more about the way the
kernel is set up. I do fear making it preemptable may open a large can of
worms, at which point it may be more useful to implement this in userspace,
maybe by the creation of a seperate entry in the device hierarchy.