Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
MIME-Version: 1.0
References: <20231016143828.647848-1-jeffxu@chromium.org> <ZS1URCBgwGGj9JtM@casper.infradead.org>
 <CAKbZUD2A+=bp_sd+Q0Yif7NJqMu8p__eb4yguq0agEcmLH8SDQ@mail.gmail.com> <CALmYWFtOgMAQDGNoM6k2Ev4kMHD396wwH+rVDODaSjsyVMDogg@mail.gmail.com>
In-Reply-To: <CALmYWFtOgMAQDGNoM6k2Ev4kMHD396wwH+rVDODaSjsyVMDogg@mail.gmail.com>
From:   Pedro Falcato <pedro.falcato@gmail.com>
Date:   Tue, 17 Oct 2023 23:35:30 +0100
Message-ID: <CAKbZUD2j1jbomCAVxUX_JmG1rfa8udc=5SqVOpDgc-3GnSTbAQ@mail.gmail.com>
Subject: Re: [RFC PATCH v1 0/8] Introduce mseal() syscall
To:     Jeff Xu <jeffxu@google.com>
Cc:     Matthew Wilcox <willy@infradead.org>, jeffxu@chromium.org,
        akpm@linux-foundation.org, keescook@chromium.org,
        sroettger@google.com, jorgelo@chromium.org, groeck@chromium.org,
        linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org,
        linux-mm@kvack.org, jannh@google.com, surenb@google.com,
        alex.sierra@amd.com, apopple@nvidia.com,
        aneesh.kumar@linux.ibm.com, axelrasmussen@google.com,
        ben@decadent.org.uk, catalin.marinas@arm.com, david@redhat.com,
        dwmw@amazon.co.uk, ying.huang@intel.com, hughd@google.com,
        joey.gouly@arm.com, corbet@lwn.net, wangkefeng.wang@huawei.com,
        Liam.Howlett@oracle.com, torvalds@linux-foundation.org,
        lstoakes@gmail.com, mawupeng1@huawei.com, linmiaohe@huawei.com,
        namit@vmware.com, peterx@redhat.com, peterz@infradead.org,
        ryan.roberts@arm.com, shr@devkernel.io, vbabka@suse.cz,
        xiujianfeng@huawei.com, yu.ma@intel.com, zhangpeng362@huawei.com,
        dave.hansen@intel.com, luto@kernel.org,
        linux-hardening@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Oct 17, 2023 at 10:34=E2=80=AFPM Jeff Xu <jeffxu@google.com> wrote:
>
> On Tue, Oct 17, 2023 at 8:30=E2=80=AFAM Pedro Falcato <pedro.falcato@gmai=
l.com> wrote:
> >
> > On Mon, Oct 16, 2023 at 4:18=E2=80=AFPM Matthew Wilcox <willy@infradead=
.org> wrote:
> > >
> > > On Mon, Oct 16, 2023 at 02:38:19PM +0000, jeffxu@chromium.org wrote:
> > > > Modern CPUs support memory permissions such as RW and NX bits. Linu=
x has
> > > > supported NX since the release of kernel version 2.6.8 in August 20=
04 [1].
> > >
> > > This seems like a confusing way to introduce the subject.  Here, you'=
re
> > > talking about page permissions, whereas (as far as I can tell), mseal=
() is
> > > about making _virtual_ addresses immutable, for some value of immutab=
le.
> > >
> > > > Memory sealing additionally protects the mapping itself against
> > > > modifications. This is useful to mitigate memory corruption issues =
where
> > > > a corrupted pointer is passed to a memory management syscall. For e=
xample,
> > > > such an attacker primitive can break control-flow integrity guarant=
ees
> > > > since read-only memory that is supposed to be trusted can become wr=
itable
> > > > or .text pages can get remapped. Memory sealing can automatically b=
e
> > > > applied by the runtime loader to seal .text and .rodata pages and
> > > > applications can additionally seal security critical data at runtim=
e.
> > > > A similar feature already exists in the XNU kernel with the
> > > > VM_FLAGS_PERMANENT [3] flag and on OpenBSD with the mimmutable sysc=
all [4].
> > > > Also, Chrome wants to adopt this feature for their CFI work [2] and=
 this
> > > > patchset has been designed to be compatible with the Chrome use cas=
e.
> > >
> > > This [2] seems very generic and wide-ranging, not helpful.  [5] was m=
ore
> > > useful to understand what you're trying to do.
> > >
> > > > The new mseal() is an architecture independent syscall, and with
> > > > following signature:
> > > >
> > > > mseal(void addr, size_t len, unsigned int types, unsigned int flags=
)
> > > >
> > > > addr/len: memory range.  Must be continuous/allocated memory, or el=
se
> > > > mseal() will fail and no VMA is updated. For details on acceptable
> > > > arguments, please refer to comments in mseal.c. Those are also full=
y
> > > > covered by the selftest.
> > >
> > > Mmm.  So when you say "continuous/allocated" what you really mean is
> > > "Must have contiguous VMAs" rather than "All pages in this range must
> > > be populated", yes?
> > >
> > > > types: bit mask to specify which syscall to seal, currently they ar=
e:
> > > > MM_SEAL_MSEAL 0x1
> > > > MM_SEAL_MPROTECT 0x2
> > > > MM_SEAL_MUNMAP 0x4
> > > > MM_SEAL_MMAP 0x8
> > > > MM_SEAL_MREMAP 0x10
> > >
> > > I don't understand why we want this level of granularity.  The OpenBS=
D
> > > and XNU examples just say "This must be immutable*".  For values of
> > > immutable that allow downgrading access (eg RW to RO or RX to RO),
> > > but not upgrading access (RW->RX, RO->*, RX->RW).
> > >
> > > > Each bit represents sealing for one specific syscall type, e.g.
> > > > MM_SEAL_MPROTECT will deny mprotect syscall. The consideration of b=
itmask
> > > > is that the API is extendable, i.e. when needed, the sealing can be
> > > > extended to madvise, mlock, etc. Backward compatibility is also eas=
y.
> > >
> > > Honestly, it feels too flexible.  Why not just two flags to mprotect(=
)
> > > -- PROT_IMMUTABLE and PROT_DOWNGRADABLE.  I can see a use for that --
> > > maybe for some things we want to be able to downgrade and for other
> > > things, we don't.
> >
> > I think it's worth pointing out that this suggestion (with PROT_*)
> > could easily integrate with mmap() and as such allow for one-shot
> > mmap() + mseal().
> > If we consider the common case as 'addr =3D mmap(...); mseal(addr);', i=
t
> > definitely sounds like a performance win as we halve the number of
> > syscalls for a sealed mapping. And if we trivially look at e.g OpenBSD
> > ld.so code, mmap() + mimmutable() and mprotect() + mimmutable() seem
> > like common patterns.
> >
> Yes. mmap() can support sealing as well, and memory is allocated as
> immutable from begining.
> This is orthogonal to mseal() though.

I don't see how this can be orthogonal to mseal().
In the case we opt for adding PROT_ bits, we should more or less only
need to adapt calc_vm_prot_bits(), and the rest should work without
issues.
vma merging won't merge vmas with different prots. The current
interfaces (mmap and mprotect) would work just fine.
In this case, mseal() or mimmutable() would only be needed if you need
to set immutability over a range of VMAs with different permissions.

Note: modifications should look kinda like this: https://godbolt.org/z/Tbjj=
d14Pe
The only annoying wrench in my plans here is that we have effectively
run out of vm_flags bits in 32-bit architectures, so this approach as
I described is not compatible with 32-bit.

> In case of ld.so, iiuc, memory can be first allocated as W, then later
> changed to RO, for example, during symbol resolution.
> The important point is that the application can decide what type of
> sealing it wants, and when to apply it.  There needs to be an api(),
> that can be mseal() or mprotect2() or mimmutable(), the naming is not
> important to me.
>
> mprotect() in linux have the following signature:
> int mprotect(void addr[.len], size_t len, int prot);
> the prot bitmasks are all taken here.
> I have not checked the prot field in mmap(), there might be bits left,
> even not, we could have mmap2(), so that is not an issue.

I don't see what you mean. We have plenty of prot bits left (32-bits,
and we seem to have around 8 different bits used).
And even if we didn't, prot is the same in mprotect and mmap and mmap2 :)

The only issue seems to be that 32-bit ran out of vm_flags, but that
can probably be worked around if need be.

--=20
Pedro