Re: Re: Re: SSH host key rotation – known_hosts file not updated
Nico Kadel-Garcia
nkadel at gmail.com
Fri Oct 25 04:42:00 AEDT 2024
On Fri, Oct 18, 2024 at 1:31 PM Jan Eden <tech at eden.one> wrote:
>
> On 2024-10-17 19:26, Nico Kadel-Garcia wrote:
>
> > > Thank you! Increasing the verbosity revealed a known_hosts entry linked
> > > to serverA's IP address (I had forgotten that I had connected to it by
> > > IP address at some point). Deleting this entry solved the problem; the
> > > new host key was stored in known_hosts when I connected to serverA
> > > again.
> > >
> > > - Jan
> >
> > And... *THIS* is why so many people disable known_hosts entirely. The
> > chance of an IP address being reused for a distinct hostname is pretty
> > high in a DHCP environment without reservations, coupled with dynamic
> > DNS. It's also very common when servers get rebuilt from images and
> > fresh hostkeys generated automatically on the same hardware, even with
> > the same IP address. The popular solution is to simply disable
> > known_hosts in your ~/.ssh/config as needed:
> >
> > # Disable known_hosts to avoid IP re-use conflicts
> > Host *
> > UserKnownHostsFile /dev/null
> > StrictHostKeyChecking no
> > LogLevel ERROR
>
> Thanks for the hint. How would I verify a server's identity without
> known_hosts / StrictHostKeyChecking?
By verifying it with a TLS signature, if you feel the need to spend
the effort. Those don't scale well. Otherwise, known_hosts entries are
pretty much "first-come, first serve", and whoever's key is accepted
first can commit a quite successful denial of service attack against
the second host. That's actually the *idea*, but it presumes that the
host_keys are consistent and stable identifiers rather than merely
public/private key pairs to ensure protection of content. When the
world was much smaller, back in 1995, known_hosts was pretty useful.
IP addresses and DNS were pretty stable and not prone to modern drift,
such as occur in auto-scaling in cloud based VLANs. But re-use of IP
addresses has become commonplace, especially with the address space
limits of IPv4 and the refusal of so many to even bother with IPv6.
More information about the openssh-unix-dev
mailing list