Re: HEADS UP: /bin and /sbin are now dynamically linked

From: Tim Kientzle <kientzle_at_acm.org>
Date: Thu, 20 Nov 2003 16:31:10 -0800
Garance A Drosihn wrote:
> At 6:26 PM +0100 11/17/03, Julian Stacey wrote:
>> Seconded !  Better commit an improved switch with
>> default = Off.
> 
> The time for voting was months ago.  

Actually, the discussion started almost a year ago now.
That's when the new PAM/NSS libraries were first being
announced, which was a big driving factor for all-dynamic
linking.  I recall quite a bit of that discussion
happening right here on current_at_.

Many of us here have been hamstrung by systems that didn't
provide a static fallback.  I've personally been bitten by
unrecoverable Linux and Solaris systems due to hosed shared
libraries.  That's why I volunteered to build /rescue in the
first place, so that I'd never be faced with an unrecoverable
FreeBSD machine.

I'm pretty comfortable with the failsafes that we
have in place:
  * /sbin/init is static
  * If /bin/sh fails, /rescue/sh can be run
  * /rescue provides a complete set of statically-linked
    sysadmin utilities that should be sufficient
    for recovering a damaged system.

There are a few things I'd like to see:
  * It would be nice if the kernel noticed that /sbin/init
    failed too quickly and prompted the user for an alternate
    init.  That would open the door to a dynamic or just more
    ambitious /sbin/init, since you could always fall back
    to /rescue/init for recovery.
  * /rescue/vi is currently unusable if /usr is missing because
    the termcap database is in /usr.  One possibility
    would be to build a couple of default termcap entries
    into ncurses or into vi.

If there are still rough edges on some of this well,
that is what -CURRENT is all about, after all. ;-)


Tim Kientzle
Received on Thu Nov 20 2003 - 15:31:21 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:30 UTC