Re: psi_trigger_poll() is completely broken

[Linus Torvalds]

On Thu, Jan 6, 2022 at 4:14 PM Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
>
> I had to make the following changes to Linus's patch:

Ack. Thanks.

> This is one way to fix the use-after-free, but the fact that it allows anyone
> who can write to a /proc/pressure/* file to cause the kernel to allocate an
> unbounded number of 'struct psi_trigger' structs is still really broken.

Yeah, I agree. Very non-optimal - that patch really was trying to just
keep the status quo, and fixing the immediate problems.

Modifying that patch to only allow a previous NULL value in
psi_trigger_replace() would be fairly simple - it would basically just
get rid of the "stale_trigger" list (and the loops it creates).

You'd still want the psi_trigger_release() model to separate that
whole "release" from "new trigger".

But that does require that nobody ever does more than a single write
to one file.

Debian code search finds those "/proc/pressure/xyz" files mentioned at
least by systemd and the chromium chrome browser sources. Whether they
actually write triggers to them, I can't say.

Maybe we just need to try.

               Linus