The userfaultfd calls copy_huge_page_from_user() which does not do
any cache flushing for the target page. Then the target page will
be mapped to the user space with a different address (user address),
which might have an alias issue with the kernel address used to copy
the data from the user to. Fix this issue by flushing dcache in
copy_huge_page_from_user().
Fixes: fa4d75c1de13 ("userfaultfd: hugetlbfs: add copy_huge_page_from_user for hugetlb userfaultfd support")
Signed-off-by: Muchun Song <[email protected]>
---
mm/memory.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/mm/memory.c b/mm/memory.c
index e8ce066be5f2..eb027da68aa7 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -5405,6 +5405,8 @@ long copy_huge_page_from_user(struct page *dst_page,
if (rc)
break;
+ flush_dcache_page(subpage);
+
cond_resched();
}
return ret_val;
--
2.11.0
On 2/7/22 23:36, Muchun Song wrote:
> The userfaultfd calls copy_huge_page_from_user() which does not do
> any cache flushing for the target page. Then the target page will
> be mapped to the user space with a different address (user address),
> which might have an alias issue with the kernel address used to copy
> the data from the user to. Fix this issue by flushing dcache in
> copy_huge_page_from_user().
Quick question.
Should this also be done for the non-hugetlb case? Take a look at the
routines __mcopy_atomic() and mcopy_atomic_pte(). Or, is that somehow
handled?
For this change,
Reviewed-by: Mike Kravetz <[email protected]>
--
Mike Kravetz
>
> Fixes: fa4d75c1de13 ("userfaultfd: hugetlbfs: add copy_huge_page_from_user for hugetlb userfaultfd support")
> Signed-off-by: Muchun Song <[email protected]>
> ---
> mm/memory.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index e8ce066be5f2..eb027da68aa7 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -5405,6 +5405,8 @@ long copy_huge_page_from_user(struct page *dst_page,
> if (rc)
> break;
>
> + flush_dcache_page(subpage);
> +
> cond_resched();
> }
> return ret_val;
On Thu, Feb 10, 2022 at 3:07 AM Mike Kravetz <[email protected]> wrote:
>
> On 2/7/22 23:36, Muchun Song wrote:
> > The userfaultfd calls copy_huge_page_from_user() which does not do
> > any cache flushing for the target page. Then the target page will
> > be mapped to the user space with a different address (user address),
> > which might have an alias issue with the kernel address used to copy
> > the data from the user to. Fix this issue by flushing dcache in
> > copy_huge_page_from_user().
>
> Quick question.
>
> Should this also be done for the non-hugetlb case? Take a look at the
> routines __mcopy_atomic() and mcopy_atomic_pte(). Or, is that somehow
> handled?
Actually, you are right. __mcopy_atomic() and mcopy_atomic_pte()
should also be fixed. And shmem_mfill_atomic_pte() also should
be fixed. I'll fix those places in the next version. Thanks.
>
> For this change,
> Reviewed-by: Mike Kravetz <[email protected]>
Thanks Mike.