Mailer@SRI-CSL (03/07/83)
From: The Mailer Daemon <Mailer at SRI-CSL>
Message failed for the following:
unix-wiz at SRI-CSL: No such mailbox
------------
Received: from UDEL-RELAY by SRI-CSL at 7-Mar-83 1321-PST
Return-Path: <rehmi.umcp-cs@UDel-Relay>
Date: 4 Mar 83 12:18:15 EST (Fri)
From: Sneaker To Vaxen <mork.umcp-cs@UDel-Relay>
Subject: There are advantages to having /dev/*mem...
To: unix-wiz@sri-csl
Via: UMCP-CS; 7 Mar 83 5:30-EST
1) looking easily at and patching a running unix (while debugging new drivers,
playing with things, just having fun, etc.)
2) accessing any sundry table whatsoever.
3) lack of unnecessary standardization.
4) making it possible to do neat hacks like doing ioctls (e.g. FIONREAD) on
mpx channels.
5) and the list goes on...
As for security, just create a dummy group, say mem, and put *mem into that
group, mode 6[64]0, and have all programs wishing to twiddle it be setgid
to mem. People are too paranoid.
Speaking of such, I too like the idea of having system calls and devices for
certain things, but for system tables? That's just going too far. For example,
I'm working on a driver (/dev/pmem) that given a pid by ioctl, maps itself onto
that proc's address space. And a snarf() system call which in effect switches
two processes in place. But come on, drivers for tables is just a load of bs.
-------