Re: itp: static bins / resolving static debian issues
Yes you're right. You need a static sshd to get into the system. I was
mostly responding to the first part of your message.
It's probably OK not to have a static getty, since several of them
should be already be linked and loaded--you will just have to get
your root password correct before you lose your getty. This is
probably acceptable, though I have had the experience where I've
watched my gettys go down one after another as I tried various
userids to get into a system.
I don't think you would need a static init unless you tried to reboot
the system, though I'm willing to be proven wrong. It should be linked
and loaded, and when it forks, it should carry that information.
Apache would fail to run most dynamic content if your C library
got hosed, but many other services can survive it (and if you were
using modperl or modphp you would come away unscathed). Other web servers
(such as Java based ones running servlets), databases, name service,
irc servers, and many other things don't launch sub-processes and
could survive it as well. Also, if you are running a file server, NFS
will keep exporting your existing mounts, though it will fail to
establish new ones.
So there are a substantial number of important services you might
be running where you might want live recovery.
Obviously losing your C library causes trouble, and you want to
get it back as soon as possible. On many systems live recovery is
going to be more of an "I can do it from remote and save a trip"
issue rather than a "I can keep my database and file server running"
issue. In either case, it's a win.
Not everyone needs live recovery, but I believe the cost is very small
for those who don't need it--and not everyone who needs it knows that
they do, until they need it.
Justin
On Thu, Aug 19, 1999 at 10:15:41AM +0200, Ruud de Rooij wrote:
> On 1999/08/19, Justin Wells wrote:
>
> > Some of you may have missed the mounds of earlier discussion. The
> > point of "live recovery" is to recover the system WITHOUT requiring
> > a reboot, therefore any boot disk or LILO solution is not acceptable.
> >
> > Suppose you screw something up, but your database and webserver are
> > still linked and loaded. Why bring them down?
> >
> > You need to get a root shell WITHOUT a reboot. So you need roots
> > shell to be static, and you need a static su, and a static sulogin.
>
> And how are you going to become root on a system if noone is logged in? You
> need a static getty, a static login, or, if you want to do it remotely, a
> static inetd, a static telnetd, or a static sshd. And so on. Like I said in
> my previous message.
>
> Besides, if it is your webserver you are worrying about, you do realize it
> can't run CGI scripts anymore, if libc is hosed. I doubt sendmail or any other
> mail server will continue to work if libc is hosed either.
>
> Could you please consider answering under the message you quoted? Most people
> read from top to bottom.
>
> - Ruud de Rooij.
> --
> ruud de rooij | ruud@ruud.org | http://ruud.org
>
>
Reply to: