Re: [PATCH] pstore: migrate to crypto acomp interface (take 2)

[Ard Biesheuvel]

On Mon, 17 Oct 2022 at 21:40, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>
> On Mon, Oct 17, 2022 at 09:33:06PM +0200, Ard Biesheuvel wrote:
> > On Mon, 17 Oct 2022 at 21:29, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> > >
> > > On Mon, Oct 17, 2022 at 08:14:14PM +0200, Ard Biesheuvel wrote:
> > > > So once we use the same size for input and output, I was curious
> > > > whether we could encrypt in place, and get rid of the big_oops_buf.
> > > > And the answer is 'yes', precisely because we have this horrid per-CPU
> > > > allocation which serves as a bounce buffer. And this is not specific
> > > > to acomp, the old comp algorithms get wrapped in scomps which receive
> > > > the same treatment.
> > >
> > > Ah, in the sense that "in place" is actually happening in the per-cpu
> > > allocation, and only if it succeeds does the input buffer get
> > > overwritten?
> >
> > Something like that IIRC.
> >
> > > > So at that point, I wondered what the point is of all this complexity.
> > > > Do we really need 6 different algorithms to compress a couple of K of
> > > > ASCII text on a code path that is ice cold by definition? Wouldn't it
> > > > be better to drop the crypto API altogether here, and just use GZIP
> > > > via the library interface?
> > >
> > > Well, my goal was to make the algo "pstore doesn't care". If someone
> > > picks deflate, do they still get all the per-cpu allocations?
> > >
> >
> > Not if you use the library interface directly.
> >
> > The issue with the percpu buffers is that they are only kept if any
> > scomp TFMs are active, but this is always the case for pstore, as you
> > don't want to allocate it on the panic path.
>
> Okay, so strictly speaking, eliminating the per-CPU allocation is an
> improvement. Keeping scomp and doing in-place compression will let
> pstore use "any" compressions method.
>

I'm not following the point you are making here.

> Is there a crypto API that does _not_ preallocate the per-CPU stuff?
> Because, as you say, it's a huge amount of memory on the bigger
> systems...
>

The library interface for each of the respective algorithms.