diff mbox series

mm/uffd: Always wr-protect pte in pte|pmd_mkuffd_wp()

Message ID 20221208194628.766316-1-peterx@redhat.com (mailing list archive)
State New
Headers show
Series mm/uffd: Always wr-protect pte in pte|pmd_mkuffd_wp() | expand

Commit Message

Peter Xu Dec. 8, 2022, 7:46 p.m. UTC
This patch is a cleanup to always wr-protect pte/pmd in mkuffd_wp paths.

The reasons I still think this patch is worthwhile, are:

  (1) It is a cleanup already; diffstat tells.

  (2) It just feels natural after I thought about this, if the pte is uffd
      protected, let's remove the write bit no matter what it was.

  (2) Since x86 is the only arch that supports uffd-wp, it also redefines
      pte|pmd_mkuffd_wp() in that it should always contain removals of
      write bits.  It means any future arch that want to implement uffd-wp
      should naturally follow this rule too.  It's good to make it a
      default, even if with vm_page_prot changes on VM_UFFD_WP.

  (3) It covers more than vm_page_prot.  So no chance of any potential
      future "accident" (like pte_mkdirty() sparc64 or loongarch, even
      though it just got its pte_mkdirty fixed <1 month ago).  It'll be
      fairly clear when reading the code too that we don't worry anything
      before a pte_mkuffd_wp() on uncertainty of the write bit.

We may call pte_wrprotect() one more time in some paths (e.g. thp split),
but that should be fully local bitop instruction so the overhead should be
negligible.

Although this patch should logically also fix all the known issues on
uffd-wp too recently on either page migration or numa balancing, but this
is not the plan for that fix.  So no fixes, and stable doesn't need this.

Signed-off-by: Peter Xu <peterx@redhat.com>
---

Note: this patch should be able to apply cleanly with/without the other
mm/migrate patch, or David's vm_page_prot changes.
---
 arch/x86/include/asm/pgtable.h | 24 ++++++++++++------------
 include/asm-generic/hugetlb.h  | 16 ++++++++--------
 mm/huge_memory.c               |  8 +++-----
 mm/hugetlb.c                   |  4 ++--
 mm/memory.c                    |  8 +++-----
 mm/mprotect.c                  |  6 ++----
 mm/userfaultfd.c               | 18 ++----------------
 7 files changed, 32 insertions(+), 52 deletions(-)

Comments

David Hildenbrand Dec. 14, 2022, 10:59 a.m. UTC | #1
On 08.12.22 20:46, Peter Xu wrote:
> This patch is a cleanup to always wr-protect pte/pmd in mkuffd_wp paths.
> 
> The reasons I still think this patch is worthwhile, are:
> 
>    (1) It is a cleanup already; diffstat tells.
> 
>    (2) It just feels natural after I thought about this, if the pte is uffd
>        protected, let's remove the write bit no matter what it was.
> 
>    (2) Since x86 is the only arch that supports uffd-wp, it also redefines
>        pte|pmd_mkuffd_wp() in that it should always contain removals of
>        write bits.  It means any future arch that want to implement uffd-wp
>        should naturally follow this rule too.  It's good to make it a
>        default, even if with vm_page_prot changes on VM_UFFD_WP.
> 
>    (3) It covers more than vm_page_prot.  So no chance of any potential
>        future "accident" (like pte_mkdirty() sparc64 or loongarch, even
>        though it just got its pte_mkdirty fixed <1 month ago).  It'll be
>        fairly clear when reading the code too that we don't worry anything
>        before a pte_mkuffd_wp() on uncertainty of the write bit.

Don't necessarily agree with (3). If you'd have a broken pte_mkdirty() 
and do the pte_mkdirty() after pte_mkuffd_wp() it would still be broken. 
Because sparc64 and loongarch are simply broken.

> 
> We may call pte_wrprotect() one more time in some paths (e.g. thp split),
> but that should be fully local bitop instruction so the overhead should be
> negligible.
> 
> Although this patch should logically also fix all the known issues on
> uffd-wp too recently on either page migration or numa balancing, but this
> is not the plan for that fix.  So no fixes, and stable doesn't need this.

I don't see how this would fix do_numa_page(), where we only do a 
pte_modify().

> 
> Signed-off-by: Peter Xu <peterx@redhat.com>
> ---
> 
> Note: this patch should be able to apply cleanly with/without the other
> mm/migrate patch, or David's vm_page_prot changes.
> ---
>   arch/x86/include/asm/pgtable.h | 24 ++++++++++++------------
>   include/asm-generic/hugetlb.h  | 16 ++++++++--------
>   mm/huge_memory.c               |  8 +++-----
>   mm/hugetlb.c                   |  4 ++--
>   mm/memory.c                    |  8 +++-----
>   mm/mprotect.c                  |  6 ++----
>   mm/userfaultfd.c               | 18 ++----------------
>   7 files changed, 32 insertions(+), 52 deletions(-)

It's certainly a cleanup, even though we might unnecessarily wrprotect 
(I don't think we care).
Peter Xu Dec. 14, 2022, 2:26 p.m. UTC | #2
On Wed, Dec 14, 2022 at 11:59:35AM +0100, David Hildenbrand wrote:
> On 08.12.22 20:46, Peter Xu wrote:
> > This patch is a cleanup to always wr-protect pte/pmd in mkuffd_wp paths.
> > 
> > The reasons I still think this patch is worthwhile, are:
> > 
> >    (1) It is a cleanup already; diffstat tells.
> > 
> >    (2) It just feels natural after I thought about this, if the pte is uffd
> >        protected, let's remove the write bit no matter what it was.
> > 
> >    (2) Since x86 is the only arch that supports uffd-wp, it also redefines
> >        pte|pmd_mkuffd_wp() in that it should always contain removals of
> >        write bits.  It means any future arch that want to implement uffd-wp
> >        should naturally follow this rule too.  It's good to make it a
> >        default, even if with vm_page_prot changes on VM_UFFD_WP.
> > 
> >    (3) It covers more than vm_page_prot.  So no chance of any potential
> >        future "accident" (like pte_mkdirty() sparc64 or loongarch, even
> >        though it just got its pte_mkdirty fixed <1 month ago).  It'll be
> >        fairly clear when reading the code too that we don't worry anything
> >        before a pte_mkuffd_wp() on uncertainty of the write bit.
> 
> Don't necessarily agree with (3). If you'd have a broken pte_mkdirty() and
> do the pte_mkdirty() after pte_mkuffd_wp() it would still be broken. Because
> sparc64 and loongarch are simply broken.

That's why I mentioned on the order of operations matters.

> 
> > 
> > We may call pte_wrprotect() one more time in some paths (e.g. thp split),
> > but that should be fully local bitop instruction so the overhead should be
> > negligible.
> > 
> > Although this patch should logically also fix all the known issues on
> > uffd-wp too recently on either page migration or numa balancing, but this
> > is not the plan for that fix.  So no fixes, and stable doesn't need this.
> 
> I don't see how this would fix do_numa_page(), where we only do a
> pte_modify().

Yes, this patch won't, because it's a pure cleanup.  Otherwise we need
another line of wr-protect in numa recover path.

I can remove that sentence in v2 commit log.
David Hildenbrand Dec. 14, 2022, 2:27 p.m. UTC | #3
On 14.12.22 15:26, Peter Xu wrote:
> On Wed, Dec 14, 2022 at 11:59:35AM +0100, David Hildenbrand wrote:
>> On 08.12.22 20:46, Peter Xu wrote:
>>> This patch is a cleanup to always wr-protect pte/pmd in mkuffd_wp paths.
>>>
>>> The reasons I still think this patch is worthwhile, are:
>>>
>>>     (1) It is a cleanup already; diffstat tells.
>>>
>>>     (2) It just feels natural after I thought about this, if the pte is uffd
>>>         protected, let's remove the write bit no matter what it was.
>>>
>>>     (2) Since x86 is the only arch that supports uffd-wp, it also redefines
>>>         pte|pmd_mkuffd_wp() in that it should always contain removals of
>>>         write bits.  It means any future arch that want to implement uffd-wp
>>>         should naturally follow this rule too.  It's good to make it a
>>>         default, even if with vm_page_prot changes on VM_UFFD_WP.
>>>
>>>     (3) It covers more than vm_page_prot.  So no chance of any potential
>>>         future "accident" (like pte_mkdirty() sparc64 or loongarch, even
>>>         though it just got its pte_mkdirty fixed <1 month ago).  It'll be
>>>         fairly clear when reading the code too that we don't worry anything
>>>         before a pte_mkuffd_wp() on uncertainty of the write bit.
>>
>> Don't necessarily agree with (3). If you'd have a broken pte_mkdirty() and
>> do the pte_mkdirty() after pte_mkuffd_wp() it would still be broken. Because
>> sparc64 and loongarch are simply broken.
> 
> That's why I mentioned on the order of operations matters.
> 
>>
>>>
>>> We may call pte_wrprotect() one more time in some paths (e.g. thp split),
>>> but that should be fully local bitop instruction so the overhead should be
>>> negligible.
>>>
>>> Although this patch should logically also fix all the known issues on
>>> uffd-wp too recently on either page migration or numa balancing, but this
>>> is not the plan for that fix.  So no fixes, and stable doesn't need this.
>>
>> I don't see how this would fix do_numa_page(), where we only do a
>> pte_modify().
> 
> Yes, this patch won't, because it's a pure cleanup.  Otherwise we need
> another line of wr-protect in numa recover path.
> 
> I can remove that sentence in v2 commit log.

Feel free to add my

Acked-by: David Hildenbrand <david@redhat.com>

Nothing jumped at me.
diff mbox series

Patch

diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 0564edd24ffb..1c843395a8b3 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -289,6 +289,11 @@  static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
 	return native_make_pte(v & ~clear);
 }
 
+static inline pte_t pte_wrprotect(pte_t pte)
+{
+	return pte_clear_flags(pte, _PAGE_RW);
+}
+
 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
 static inline int pte_uffd_wp(pte_t pte)
 {
@@ -313,7 +318,7 @@  static inline int pte_uffd_wp(pte_t pte)
 
 static inline pte_t pte_mkuffd_wp(pte_t pte)
 {
-	return pte_set_flags(pte, _PAGE_UFFD_WP);
+	return pte_wrprotect(pte_set_flags(pte, _PAGE_UFFD_WP));
 }
 
 static inline pte_t pte_clear_uffd_wp(pte_t pte)
@@ -332,11 +337,6 @@  static inline pte_t pte_mkold(pte_t pte)
 	return pte_clear_flags(pte, _PAGE_ACCESSED);
 }
 
-static inline pte_t pte_wrprotect(pte_t pte)
-{
-	return pte_clear_flags(pte, _PAGE_RW);
-}
-
 static inline pte_t pte_mkexec(pte_t pte)
 {
 	return pte_clear_flags(pte, _PAGE_NX);
@@ -401,6 +401,11 @@  static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
 	return native_make_pmd(v & ~clear);
 }
 
+static inline pmd_t pmd_wrprotect(pmd_t pmd)
+{
+	return pmd_clear_flags(pmd, _PAGE_RW);
+}
+
 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
 static inline int pmd_uffd_wp(pmd_t pmd)
 {
@@ -409,7 +414,7 @@  static inline int pmd_uffd_wp(pmd_t pmd)
 
 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
 {
-	return pmd_set_flags(pmd, _PAGE_UFFD_WP);
+	return pmd_wrprotect(pmd_set_flags(pmd, _PAGE_UFFD_WP));
 }
 
 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
@@ -428,11 +433,6 @@  static inline pmd_t pmd_mkclean(pmd_t pmd)
 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
 }
 
-static inline pmd_t pmd_wrprotect(pmd_t pmd)
-{
-	return pmd_clear_flags(pmd, _PAGE_RW);
-}
-
 static inline pmd_t pmd_mkdirty(pmd_t pmd)
 {
 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
diff --git a/include/asm-generic/hugetlb.h b/include/asm-generic/hugetlb.h
index a57d667addd2..d7f6335d3999 100644
--- a/include/asm-generic/hugetlb.h
+++ b/include/asm-generic/hugetlb.h
@@ -25,6 +25,13 @@  static inline pte_t huge_pte_mkwrite(pte_t pte)
 	return pte_mkwrite(pte);
 }
 
+#ifndef __HAVE_ARCH_HUGE_PTE_WRPROTECT
+static inline pte_t huge_pte_wrprotect(pte_t pte)
+{
+	return pte_wrprotect(pte);
+}
+#endif
+
 static inline pte_t huge_pte_mkdirty(pte_t pte)
 {
 	return pte_mkdirty(pte);
@@ -37,7 +44,7 @@  static inline pte_t huge_pte_modify(pte_t pte, pgprot_t newprot)
 
 static inline pte_t huge_pte_mkuffd_wp(pte_t pte)
 {
-	return pte_mkuffd_wp(pte);
+	return huge_pte_wrprotect(pte_mkuffd_wp(pte));
 }
 
 static inline pte_t huge_pte_clear_uffd_wp(pte_t pte)
@@ -104,13 +111,6 @@  static inline int huge_pte_none_mostly(pte_t pte)
 	return huge_pte_none(pte) || is_pte_marker(pte);
 }
 
-#ifndef __HAVE_ARCH_HUGE_PTE_WRPROTECT
-static inline pte_t huge_pte_wrprotect(pte_t pte)
-{
-	return pte_wrprotect(pte);
-}
-#endif
-
 #ifndef __HAVE_ARCH_PREPARE_HUGEPAGE_RANGE
 static inline int prepare_hugepage_range(struct file *file,
 		unsigned long addr, unsigned long len)
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 1d9ad909c87c..86f1ecb0f9b4 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1913,17 +1913,15 @@  int change_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	oldpmd = pmdp_invalidate_ad(vma, addr, pmd);
 
 	entry = pmd_modify(oldpmd, newprot);
-	if (uffd_wp) {
-		entry = pmd_wrprotect(entry);
+	if (uffd_wp)
 		entry = pmd_mkuffd_wp(entry);
-	} else if (uffd_wp_resolve) {
+	else if (uffd_wp_resolve)
 		/*
 		 * Leave the write bit to be handled by PF interrupt
 		 * handler, then things like COW could be properly
 		 * handled.
 		 */
 		entry = pmd_clear_uffd_wp(entry);
-	}
 
 	/* See change_pte_range(). */
 	if ((cp_flags & MM_CP_TRY_CHANGE_WRITABLE) && !pmd_write(entry) &&
@@ -3267,7 +3265,7 @@  void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
 	if (is_writable_migration_entry(entry))
 		pmde = maybe_pmd_mkwrite(pmde, vma);
 	if (pmd_swp_uffd_wp(*pvmw->pmd))
-		pmde = pmd_wrprotect(pmd_mkuffd_wp(pmde));
+		pmde = pmd_mkuffd_wp(pmde);
 	if (!is_migration_entry_young(entry))
 		pmde = pmd_mkold(pmde);
 	/* NOTE: this may contain setting soft-dirty on some archs */
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 1088f2f41c88..a10cb7038a04 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -5749,7 +5749,7 @@  static vm_fault_t hugetlb_no_page(struct mm_struct *mm,
 	 * if populated.
 	 */
 	if (unlikely(pte_marker_uffd_wp(old_pte)))
-		new_pte = huge_pte_wrprotect(huge_pte_mkuffd_wp(new_pte));
+		new_pte = huge_pte_mkuffd_wp(new_pte);
 	set_huge_pte_at(mm, haddr, ptep, new_pte);
 
 	hugetlb_count_add(pages_per_huge_page(h), mm);
@@ -6550,7 +6550,7 @@  unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 			pte = huge_pte_modify(old_pte, newprot);
 			pte = arch_make_huge_pte(pte, shift, vma->vm_flags);
 			if (uffd_wp)
-				pte = huge_pte_mkuffd_wp(huge_pte_wrprotect(pte));
+				pte = huge_pte_mkuffd_wp(pte);
 			else if (uffd_wp_resolve)
 				pte = huge_pte_clear_uffd_wp(pte);
 			huge_ptep_modify_prot_commit(vma, address, ptep, old_pte, pte);
diff --git a/mm/memory.c b/mm/memory.c
index aad226daf41b..1e2628bf8de1 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -882,7 +882,7 @@  copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma
 	pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma);
 	if (userfaultfd_pte_wp(dst_vma, *src_pte))
 		/* Uffd-wp needs to be delivered to dest pte as well */
-		pte = pte_wrprotect(pte_mkuffd_wp(pte));
+		pte = pte_mkuffd_wp(pte);
 	set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte);
 	return 0;
 }
@@ -3950,10 +3950,8 @@  vm_fault_t do_swap_page(struct vm_fault *vmf)
 	flush_icache_page(vma, page);
 	if (pte_swp_soft_dirty(vmf->orig_pte))
 		pte = pte_mksoft_dirty(pte);
-	if (pte_swp_uffd_wp(vmf->orig_pte)) {
+	if (pte_swp_uffd_wp(vmf->orig_pte))
 		pte = pte_mkuffd_wp(pte);
-		pte = pte_wrprotect(pte);
-	}
 	vmf->orig_pte = pte;
 
 	/* ksm created a completely new copy */
@@ -4296,7 +4294,7 @@  void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr)
 	if (write)
 		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
 	if (unlikely(uffd_wp))
-		entry = pte_mkuffd_wp(pte_wrprotect(entry));
+		entry = pte_mkuffd_wp(entry);
 	/* copy-on-write page */
 	if (write && !(vma->vm_flags & VM_SHARED)) {
 		inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
diff --git a/mm/mprotect.c b/mm/mprotect.c
index 093cb50f2fc4..a816ec34c234 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -177,12 +177,10 @@  static unsigned long change_pte_range(struct mmu_gather *tlb,
 			oldpte = ptep_modify_prot_start(vma, addr, pte);
 			ptent = pte_modify(oldpte, newprot);
 
-			if (uffd_wp) {
-				ptent = pte_wrprotect(ptent);
+			if (uffd_wp)
 				ptent = pte_mkuffd_wp(ptent);
-			} else if (uffd_wp_resolve) {
+			else if (uffd_wp_resolve)
 				ptent = pte_clear_uffd_wp(ptent);
-			}
 
 			/*
 			 * In some writable, shared mappings, we might want
diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index b7a9479bece2..7d28f9f02892 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -74,24 +74,10 @@  int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
 	_dst_pte = pte_mkdirty(_dst_pte);
 	if (page_in_cache && !vm_shared)
 		writable = false;
-
-	/*
-	 * Always mark a PTE as write-protected when needed, regardless of
-	 * VM_WRITE, which the user might change.
-	 */
-	if (wp_copy) {
-		_dst_pte = pte_mkuffd_wp(_dst_pte);
-		writable = false;
-	}
-
 	if (writable)
 		_dst_pte = pte_mkwrite(_dst_pte);
-	else
-		/*
-		 * We need this to make sure write bit removed; as mk_pte()
-		 * could return a pte with write bit set.
-		 */
-		_dst_pte = pte_wrprotect(_dst_pte);
+	if (wp_copy)
+		_dst_pte = pte_mkuffd_wp(_dst_pte);
 
 	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);