From: Jisheng Zhang <jszhang@kernel.org>
To: Zi Yan <ziy@nvidia.com>
Cc: Paul Walmsley <pjw@kernel.org>,
Palmer Dabbelt <palmer@dabbelt.com>,
Albert Ou <aou@eecs.berkeley.edu>,
Alexandre Ghiti <alex@ghiti.fr>, Zong Li <zong.li@sifive.com>,
Deepak Gupta <debug@rivosinc.com>,
Palmer Dabbelt <palmer@rivosinc.com>,
Andrew Morton <akpm@linux-foundation.org>,
David Hildenbrand <david@kernel.org>,
John Hubbard <jhubbard@nvidia.com>, Peter Xu <peterx@redhat.com>,
linux-mm@kvack.org, linux-riscv@lists.infradead.org,
linux-kernel@vger.kernel.org, Jason Gunthorpe <jgg@ziepe.ca>,
stable@vger.kernel.org, Sashiko <sashiko-bot@kernel.org>
Subject: Re: [PATCH] riscv: mm: don't let GUP-fast follow entries without _PAGE_READ
Date: Tue, 20 Oct 2026 09:30:34 +0800 [thread overview]
Message-ID: <atbEOlTjwZ9nr7rG@xhacker> (raw)
In-Reply-To: <20261015-riscv-gup-read-fix-v1-1-8cf7f2504029@nvidia.com>
On Thu, Oct 15, 2026 at 10:37:44PM -0400, Zi Yan wrote:
> riscv maps PROT_EXEC alone as PAGE_EXEC, which has
> _PAGE_EXEC but not _PAGE_READ, and a writable shadow stack entry is
> XWR = 010, which has _PAGE_WRITE but not _PAGE_READ. riscv uses the
> generic pte_access_permitted(), pmd_access_permitted() and
> pud_access_permitted(), which only check that the entry is present and,
> for a write, writable. So GUP-fast follows both kinds of entries:
>
> - a read, e.g. from vmsplice() or an O_DIRECT write, can read an
> execute-only page, which userspace cannot load from;
> - a write, e.g. from an O_DIRECT read, can modify a shadow stack page,
> as pte_write() is true for it.
>
> Slow GUP rejects both: check_vma_flags() fails a read from a VMA without
> VM_READ and a write to a VM_SHADOW_STACK VMA, unless FOLL_FORCE is set.
>
> Fix it by requiring both _PAGE_PRESENT and _PAGE_READ in
> pte_access_permitted(), and using it for pmd_access_permitted() and
> pud_access_permitted() too.
>
> Requiring _PAGE_PRESENT also stops GUP-fast from following a THP PMD
> invalidated by pmdp_invalidate(). x86, arm64, and powerpc do the same
> thing.
>
> Fixes: 3f910b7a522e ("riscv: enable HAVE_FAST_GUP if MMU")
> Fixes: c68c2ef9d641 ("riscv/mm: teach pte_mkwrite to manufacture shadow stack PTEs")
> Cc: stable@vger.kernel.org
> Reported-by: Sashiko <sashiko-bot@kernel.org>
> Closes: https://sashiko.dev/#/patchset/20261015-anon_vma_uaf_fix-v1-0-7b977303b42b%40nvidia.com?part=2
> Assisted-by: LLM
> Signed-off-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: Jisheng Zhang <jszhang@kernel.org>
> ---
> Sashiko found that the riscv pmd_access_permitted() added by the
> lazyfree THP discard rework [1] lets GUP-fast bypass execute-only
> protection. The generic helpers riscv uses today have the same gap, at
> the PTE level too, and the generic pmd_access_permitted() also lets
> GUP-fast follow an invalidated PMD. This patch makes GUP-fast require
> both _PAGE_PRESENT and _PAGE_READ. Requiring _PAGE_PRESENT stops it from
> following a PMD invalidated by pmdp_invalidate(), as x86, arm64, and
> powerpc already do and as the rework relies on.
>
> While reviewing this patch, an LLM also found that riscv's
> pte_wrprotect() sets _PAGE_READ unconditionally, which turns an
> execute-only entry into a readable one [2], e.g. every execute-only
This issue can be fixed by Nickolai's patch:
https://lore.kernel.org/linux-riscv/20261017162229.796020-1-nickolai@csail.mit.edu/
> hugetlb mapping via make_huge_pte(). That needs a separate fix.
>
> Link: https://lore.kernel.org/all/20261015-anon_vma_uaf_fix-v1-0-7b977303b42b@nvidia.com/ [1]
> Link: https://elixir.bootlin.com/linux/v7.3-rc7/source/arch/riscv/include/asm/pgtable.h#L409 [2]
> ---
> arch/riscv/include/asm/pgtable.h | 27 +++++++++++++++++++++++++++
> 1 file changed, 27 insertions(+)
>
> diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
> index 40b1ed4f3ea89..f46b742d6dd93 100644
> --- a/arch/riscv/include/asm/pgtable.h
> +++ b/arch/riscv/include/asm/pgtable.h
> @@ -372,6 +372,21 @@ static inline int pte_write(pte_t pte)
> return pte_val(pte) & _PAGE_WRITE;
> }
>
> +/*
> + * GUP-fast only follows entries that are present and have _PAGE_READ,
> + * and leaves the rest to slow GUP, which checks the VMA: an execute-only
> + * entry and a writable shadow stack entry (XWR = 010) lack _PAGE_READ, a
> + * PROT_NONE or invalidated entry lacks _PAGE_PRESENT. Any other writable
> + * entry is readable too.
> + */
> +#define pte_access_permitted pte_access_permitted
> +static inline bool pte_access_permitted(pte_t pte, bool write)
> +{
> + const unsigned long need = _PAGE_PRESENT | _PAGE_READ;
> +
> + return (pte_val(pte) & need) == need && (!write || pte_write(pte));
> +}
> +
> static inline int pte_exec(pte_t pte)
> {
> return pte_val(pte) & _PAGE_EXEC;
> @@ -820,12 +835,24 @@ static inline int pmd_write(pmd_t pmd)
> return pte_write(pmd_pte(pmd));
> }
>
> +#define pmd_access_permitted pmd_access_permitted
> +static inline bool pmd_access_permitted(pmd_t pmd, bool write)
> +{
> + return pte_access_permitted(pmd_pte(pmd), write);
> +}
> +
> #define pud_write pud_write
> static inline int pud_write(pud_t pud)
> {
> return pte_write(pud_pte(pud));
> }
>
> +#define pud_access_permitted pud_access_permitted
> +static inline bool pud_access_permitted(pud_t pud, bool write)
> +{
> + return pte_access_permitted(pud_pte(pud), write);
> +}
> +
> #define pmd_dirty pmd_dirty
> static inline int pmd_dirty(pmd_t pmd)
> {
>
> ---
> base-commit: 24a19b953e2a3d0ea5a8c4951a0ab8f32720c907
> change-id: 20261015-riscv-gup-read-fix-d657d67d350a
>
> Best regards,
> --
> Yan, Zi
>
>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv
prev parent reply other threads:[~2026-10-20 1:50 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-16 2:37 Zi Yan
2026-10-20 1:30 ` Jisheng Zhang [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=atbEOlTjwZ9nr7rG@xhacker \
--to=jszhang@kernel.org \
--cc=akpm@linux-foundation.org \
--cc=alex@ghiti.fr \
--cc=aou@eecs.berkeley.edu \
--cc=david@kernel.org \
--cc=debug@rivosinc.com \
--cc=jgg@ziepe.ca \
--cc=jhubbard@nvidia.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-riscv@lists.infradead.org \
--cc=palmer@dabbelt.com \
--cc=palmer@rivosinc.com \
--cc=peterx@redhat.com \
--cc=pjw@kernel.org \
--cc=sashiko-bot@kernel.org \
--cc=stable@vger.kernel.org \
--cc=ziy@nvidia.com \
--cc=zong.li@sifive.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®