[PATCH 0/2] fix for direct-I/O to DAX mappings

[PATCH 0/2] fix for direct-I/O to DAX mappings

[Dan Williams]

Hi Andrew,

While Ross was doing a review of a new mmap+DAX direct-I/O test case for
xfstests, from Xiong, he noticed occasions where it failed to trigger a
page dirty event.  Dave then spotted the problem fixed by patch1. The
pte_devmap() check is precluding pte_allows_gup(), i.e. bypassing
permission checks and dirty tracking.

Patch2 is a cleanup and clarifies that pte_unmap() only needs to be done
once per page-worth of ptes. It unifies the exit paths similar to the
generic gup_pte_range() in the __HAVE_ARCH_PTE_SPECIAL case.

I'm sending this through the -mm tree for a double-check from memory
management folks. It has a build success notification from the kbuild
robot.

---

Dan Williams (2):
      x86, mm: fix gup_pte_range() vs DAX mappings
      x86, mm: unify exit paths in gup_pte_range()


 arch/x86/mm/gup.c |   37 +++++++++++++++++++++----------------
 1 file changed, 21 insertions(+), 16 deletions(-)

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 1/2] x86, mm: fix gup_pte_range() vs DAX mappings

[Dan Williams]

gup_pte_range() fails to check pte_allows_gup() before translating a DAX
pte entry, pte_devmap(), to a page. This allows writes to read-only
mappings, and bypasses the DAX cacheline dirty tracking due to missed
'mkwrite' faults. The gup_huge_pmd() path and the gup_huge_pud() path
correctly check pte_allows_gup() before checking for _devmap() entries.

Fixes: 3565fce3a659 ("mm, x86: get_user_pages() for dax mappings")
Cc: <x86@kernel.org>
Cc: <stable@vger.kernel.org>
Cc: Xiong Zhou <xzhou@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Reported-by: Dave Hansen <dave.hansen@linux.intel.com>
Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/gup.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/gup.c b/arch/x86/mm/gup.c
index 0d4fb3ebbbac..1680768d392c 100644
--- a/arch/x86/mm/gup.c
+++ b/arch/x86/mm/gup.c
@@ -120,6 +120,11 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
 			return 0;
 		}
 
+		if (!pte_allows_gup(pte_val(pte), write)) {
+			pte_unmap(ptep);
+			return 0;
+		}
+
 		if (pte_devmap(pte)) {
 			pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
 			if (unlikely(!pgmap)) {
@@ -127,8 +132,7 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
 				pte_unmap(ptep);
 				return 0;
 			}
-		} else if (!pte_allows_gup(pte_val(pte), write) ||
-			   pte_special(pte)) {
+		} else if (pte_special(pte)) {
 			pte_unmap(ptep);
 			return 0;
 		}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 2/2] x86, mm: unify exit paths in gup_pte_range()

[Dan Williams]

All exit paths from gup_pte_range() require pte_unmap() of the original
pte page before returning. Refactor the code to have a single exit point
to do the unmap.

This mirrors the flow of the generic gup_pte_range() in mm/gup.c.

Cc: <x86@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/gup.c |   39 ++++++++++++++++++++-------------------
 1 file changed, 20 insertions(+), 19 deletions(-)

diff --git a/arch/x86/mm/gup.c b/arch/x86/mm/gup.c
index 1680768d392c..e703f09c1d78 100644
--- a/arch/x86/mm/gup.c
+++ b/arch/x86/mm/gup.c
@@ -106,36 +106,35 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
 		unsigned long end, int write, struct page **pages, int *nr)
 {
 	struct dev_pagemap *pgmap = NULL;
-	int nr_start = *nr;
-	pte_t *ptep;
+	int nr_start = *nr, ret = 0;
+	pte_t *ptep, *ptem;
 
-	ptep = pte_offset_map(&pmd, addr);
+	/*
+	 * Keep the original mapped PTE value (ptem) around since we
+	 * might increment ptep off the end of the page when finishing
+	 * our loop iteration.
+	 */
+	ptem = ptep = pte_offset_map(&pmd, addr);
 	do {
 		pte_t pte = gup_get_pte(ptep);
 		struct page *page;
 
 		/* Similar to the PMD case, NUMA hinting must take slow path */
-		if (pte_protnone(pte)) {
-			pte_unmap(ptep);
-			return 0;
-		}
+		if (pte_protnone(pte))
+			break;
 
-		if (!pte_allows_gup(pte_val(pte), write)) {
-			pte_unmap(ptep);
-			return 0;
-		}
+		if (!pte_allows_gup(pte_val(pte), write))
+			break;
 
 		if (pte_devmap(pte)) {
 			pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
 			if (unlikely(!pgmap)) {
 				undo_dev_pagemap(nr, nr_start, pages);
-				pte_unmap(ptep);
-				return 0;
+				break;
 			}
-		} else if (pte_special(pte)) {
-			pte_unmap(ptep);
-			return 0;
-		}
+		} else if (pte_special(pte))
+			break;
+
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		get_page(page);
@@ -145,9 +144,11 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
 		(*nr)++;
 
 	} while (ptep++, addr += PAGE_SIZE, addr != end);
-	pte_unmap(ptep - 1);
+	if (addr == end)
+		ret = 1;
+	pte_unmap(ptem);
 
-	return 1;
+	return ret;
 }
 
 static inline void get_head_page_multiple(struct page *page, int nr)

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 0/2] fix for direct-I/O to DAX mappings

[Linus Torvalds]

On Sat, Feb 25, 2017 at 9:08 AM, Dan Williams <dan.j.williams@intel.com> wrote:
>
> I'm sending this through the -mm tree for a double-check from memory
> management folks. It has a build success notification from the kbuild
> robot.

I'm just checking that this isn't lost - I didn't get it in the latest
patch-bomb from Andrew.

I'm assuming it's still percolating through your system, Andrew, but
if not, holler.

              Linus

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 0/2] fix for direct-I/O to DAX mappings

[Andrew Morton]

On Tue, 28 Feb 2017 09:10:39 -0800 Linus Torvalds <torvalds@linux-foundation.org> wrote:

> On Sat, Feb 25, 2017 at 9:08 AM, Dan Williams <dan.j.williams@intel.com> wrote:
> >
> > I'm sending this through the -mm tree for a double-check from memory
> > management folks. It has a build success notification from the kbuild
> > robot.
> 
> I'm just checking that this isn't lost - I didn't get it in the latest
> patch-bomb from Andrew.
> 
> I'm assuming it's still percolating through your system, Andrew, but
> if not, holler.
> 

Yup, I've got them.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 0/2] fix for direct-I/O to DAX mappings

[Xiong Zhou]

On Sat, Feb 25, 2017 at 09:08:28AM -0800, Dan Williams wrote:
> Hi Andrew,
> 
> While Ross was doing a review of a new mmap+DAX direct-I/O test case for
> xfstests, from Xiong, he noticed occasions where it failed to trigger a
> page dirty event.  Dave then spotted the problem fixed by patch1. The
> pte_devmap() check is precluding pte_allows_gup(), i.e. bypassing
> permission checks and dirty tracking.

This mmap-dax-dio case still fails with this patchset, while it makes
sense. It's the test case that need to be fixed.

BTW, this patchset fixes another xfsrestore issue, which i hit now
and then, xfs/301 w/ or wo/ DAX only on nvdimms. xfsrestore never
return but killable.

Thanks,
> 
> Patch2 is a cleanup and clarifies that pte_unmap() only needs to be done
> once per page-worth of ptes. It unifies the exit paths similar to the
> generic gup_pte_range() in the __HAVE_ARCH_PTE_SPECIAL case.
> 
> I'm sending this through the -mm tree for a double-check from memory
> management folks. It has a build success notification from the kbuild
> robot.
> 
> ---
> 
> Dan Williams (2):
>       x86, mm: fix gup_pte_range() vs DAX mappings
>       x86, mm: unify exit paths in gup_pte_range()
> 
> 
>  arch/x86/mm/gup.c |   37 +++++++++++++++++++++----------------
>  1 file changed, 21 insertions(+), 16 deletions(-)

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>