Re: kern/60084: swap encryption: per-page tracking seems overengineered

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

> Date: Sun, 15 Mar 2026 14:05:57 +0900
> From: Takashi Yamamoto <imuwoto2%gmail.com@localhost>
> 
> On Sun, Mar 15, 2026 at 1:40 PM Taylor R Campbell via gnats
> <gnats-admin%netbsd.org@localhost> wrote:
> >
> > Requiring a reboot (or, worse, kernel rebuild) to test the performance
> > impact in a real workload would be annoying, so I made sure there is a
> > way to enable and disable it dynamically.
> 
> if you want to test the performance impact, you don't want to have
> encrypt/unencrypt pages mixed, i suppose.

For serious performance testing, sure.

For casually checking whether the impact is noticeable or not, it is
nice to be able to just flip a switch and see whether the system
starts to get sluggish, and whether the underlying swap file/device
has recognizable memory or gibberish in it.

> my proposal is just a global knob (vm.swap_encrypt) which can't be
> changed if any swap devices are configured.

What should happen if you try to set

vm.swap_encrypt=1

in /etc/sysctl.conf, but a swap device is already configured by the
time it is executed?

Here are some possibilities:

1. The system fails to boot to multi-user and drops to single-user.

2. The system boots but swap is all plaintext.

The status quo, in contrast, is:

3. The system boots and all memory swapped _after_ /etc/rc.d/sysctl is
   ciphertext.

This is still possibly problematic -- but I think it's unlikely for
any secrets to be swapped at this point.  It looks like the only
relevant rc.d scripts that run between /etc/rc.d/swap1 and
/etc/rc.d/sysctl are random_seed and ipsec, and except on extremely
low-memory systems, they aren't likely to swap (the big one likely
requiring swap is /etc/rc.d/fsck).

So I think (3) is better than (1) or (2) on most machines for most
users.

Of course it would be better for security to have never swapped
plaintext in the first place.  Perhaps, rather than vm.swap_encrypt=1
in sysctl.conf, it would be better to have swap_encrypt=YES in rc.conf
so another rc.d script which runs before /etc/rc.d/swap1 (and
therefore before /etc/rc.d/sysctl) can make the change before swap is
configured -- that way at least you don't need a kernel rebuild to
enforce it.  But, of course, changing things six years on is risky
because users may have started to rely on vm.swap_encrypt=1 in
sysctl.conf.

I don't object to changing how this works -- I just want to make sure
that we have a clear plan for ensuring users have an easy path to good
security, without regressions for existing users.