Re: [REGRESSION][PATCH] Revert "pstore: migrate to crypto acomp interface"

From: Kees Cook
Date: Fri Sep 30 2022 - 15:08:56 EST


On Fri, Sep 30, 2022 at 03:31:17PM -0300, Guilherme G. Piccoli wrote:
> On 30/09/2022 12:51, Ard Biesheuvel wrote:
> > [...]
> >
> > Does this help?
> >
> > diff --git a/fs/pstore/platform.c b/fs/pstore/platform.c
> > index b2fd3c20e7c2..c0b609d7d04e 100644
> > --- a/fs/pstore/platform.c
> > +++ b/fs/pstore/platform.c
> > @@ -292,7 +292,7 @@ static int pstore_compress(const void *in, void *out,
> > return ret;
> > }
> >
> > - return outlen;
> > + return creq->dlen;
> > }
> >
> > static void allocate_buf_for_compression(void)
> >
>
> Thanks a lot Ard, this seems to be the fix! Tested with lz4/zstd/deflate
> in both ramoops/efi backends, and all worked fine. It makes sense,
> outlen was modified in the previous API and not in the acomp thing, so
> it was a good catch =)
>
>
> >> Heheh you're right! But for something like this (pstore/dmesg
> >> compression broke for the most backends), I'd be glad if we could fix it
> >> before the release.
> >
> > Yeah better to revert - this was not a critical change anyway. But I
> > think the tweak above should fix things (it works for me here)
>
> Agreed - in fact seems it was reverted already. More than that, I found
> yet another small issue in the acomp refactor, a memory leak - attached
> is a patch with the fix, feel free to integrate in your acomp refactor
> when re-submitting (I mean, feel free to just integrate the code, don't
> need to send it as a separate patch/fix).
>
> I'm also working some fixes in implicit conversions in pstore that
> aren't great (unsigned -> int in many places), I'll send some stuff next
> week.

Awesome! Thank you both! I'll probably be busy for the next week with
the merge window, but after that I'll pull new stuff into -next for
pstore. I appreciate more people poking around in the code. :)

--
Kees Cook