[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230801172814.GD2607694@kernel.org>
Date: Tue, 1 Aug 2023 20:28:14 +0300
From: Mike Rapoport <rppt@...nel.org>
To: "Edgecombe, Rick P" <rick.p.edgecombe@...el.com>
Cc: "broonie@...nel.org" <broonie@...nel.org>,
"corbet@....net" <corbet@....net>,
"ardb@...nel.org" <ardb@...nel.org>,
"keescook@...omium.org" <keescook@...omium.org>,
"Szabolcs.Nagy@....com" <Szabolcs.Nagy@....com>,
"shuah@...nel.org" <shuah@...nel.org>,
"maz@...nel.org" <maz@...nel.org>,
"james.morse@....com" <james.morse@....com>,
"debug@...osinc.com" <debug@...osinc.com>,
"aou@...s.berkeley.edu" <aou@...s.berkeley.edu>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"catalin.marinas@....com" <catalin.marinas@....com>,
"hjl.tools@...il.com" <hjl.tools@...il.com>,
"paul.walmsley@...ive.com" <paul.walmsley@...ive.com>,
"linux-mm@...ck.org" <linux-mm@...ck.org>,
"akpm@...ux-foundation.org" <akpm@...ux-foundation.org>,
"oleg@...hat.com" <oleg@...hat.com>,
"arnd@...db.de" <arnd@...db.de>,
"ebiederm@...ssion.com" <ebiederm@...ssion.com>,
"will@...nel.org" <will@...nel.org>,
"suzuki.poulose@....com" <suzuki.poulose@....com>,
"kvmarm@...ts.linux.dev" <kvmarm@...ts.linux.dev>,
"linux-doc@...r.kernel.org" <linux-doc@...r.kernel.org>,
"linux-fsdevel@...r.kernel.org" <linux-fsdevel@...r.kernel.org>,
"linux-arm-kernel@...ts.infradead.org"
<linux-arm-kernel@...ts.infradead.org>,
"linux-kselftest@...r.kernel.org" <linux-kselftest@...r.kernel.org>,
"linux-arch@...r.kernel.org" <linux-arch@...r.kernel.org>,
"oliver.upton@...ux.dev" <oliver.upton@...ux.dev>,
"linux-riscv@...ts.infradead.org" <linux-riscv@...ts.infradead.org>,
"palmer@...belt.com" <palmer@...belt.com>
Subject: Re: [PATCH v3 21/36] arm64/mm: Implement map_shadow_stack()
On Tue, Aug 01, 2023 at 05:07:00PM +0000, Edgecombe, Rick P wrote:
> On Tue, 2023-08-01 at 15:01 +0100, Mark Brown wrote:
> > On Mon, Jul 31, 2023 at 11:19:34PM +0000, Edgecombe, Rick P wrote:
> >
> > > The thing I was trying to get at was, we have this shared syscall
> > > that
> > > means create shadow stack memory and prepopulate it like this flag
> > > says. On x86 we optionally support SHADOW_STACK_SET_TOKEN which
> > > means
> > > put a token right at the end of size. So maybe arm should have a
> > > different flag value that includes putting the marker and then the
> > > token, and x86 could match it someday if we get markers too.
> >
> > Oh, I see. My mental model was that this was controlling the whole
> > thing we put at the top rather than treating the terminator and the
> > cap
> > separately.
> >
> > > It could be a different flag, like SHADOW_STACK_SET_TOKEN_MARKER,
> > > or it
> > > could be SHADOW_STACK_SET_MARKER, and callers could pass
> > > (SHADOW_STACK_SET_TOKEN | SHADOW_STACK_SET_MARKER) to get what you
> > > have
> > > implemented here. What do you think?
> >
> > For arm64 code this would mean that it would be possible (and fairly
> > easy) to create stacks which don't have a termination record which
> > would
> > make life harder for unwinders to rely on. I don't think this is
> > insurmountable, creating manually shouldn't be the standard and it'll
> > already be an issue on x86 anyway.
>
> If you are going to support optionally writing to shadow stacks (which
> x86 needed for CRIU, and also seems like a nice thing for several other
> reasons), you are already at that point. Can't you also do a bunch of
> gcspopm's to the top of the GCS stack, and have no marker to hit before
> the end of the stack? (maybe not in GCS, I don't know...)
>
> >
> > The other minor issue is that the current arm64 marker is all bits 0
> > so by itself for arm64 _MARKER would have no perceptible impact, it
> > would only serve to push the token down a slot in the stack (I'm
> > guessing that's the intended meaning?).
>
> Pushing the token down a frame is what flags==0 does in this patch,
> right?
>
> You don't have to support all the flags actually, you could just
> support the one mode you already have and reject all other
> combinations... Then it matches between arch's, and you still have the
> guaranteed-ish end marker.
>
> So the question is not what mode should arm support, but should we have
> the flags match between x86 and ARM?
What if the flag will be called, say, SHADOW_STACK_DEFAULT_INIT?
Then each arch can push whatever it likes to and from the userspace
perspective the shadow stack will have some basic init state, no matter
what architecture it is.
> > I'm not sure that's a
> > particularly big deal though.
>
> Yea, it's not a big problem either way.
--
Sincerely yours,
Mike.
Powered by blists - more mailing lists