Re: [PATCH] mm/khugepaged: Detecting uffd-wp vma more efficiently

[Peter Xu <peterx@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1201 bytes --]

On Wed, Sep 22, 2021 at 06:22:45PM -0700, Hugh Dickins wrote:
> No, I think I misunderstood you before: thanks for re-explaining.
> (And Axel's !userfaultfd_minor() check before calling do_fault_around()
> plays an important part in making sure that it does reach shmem_fault().)

Still thanks for confirming this, Hugh.

Said that, Axel, I didn't mean I'm against doing something similar like
uffd-wp; it's just a heads-up that maybe you won't find a reproducer with real
issues with minor mode.

Even if I think minor mode should be fine with current code, we could still
choose to disable khugepaged from removing the pmd for VM_UFFD_MINOR vmas, just
like what we'll do with VM_UFFD_WP.  At least it can still reduce false
positives.

So far in my local branch I queued the patch which I attached, that's required
for uffd-wp shmem afaict.  If you think minor mode would like that too, I can
post it separately with minor mode added in.

Note that it's slightly different from what I pasted in reply to Yang Shi - I
made it slightly more complicated just to make sure there's no race.  I
mentioned the possible race (I think) in the commit log.

Let me know your preference.

Thanks,

-- 
Peter Xu

[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 2697 bytes --]

From 989d36914ac144177e17f9aacbf2785bb8f21420 Mon Sep 17 00:00:00 2001
From: Peter Xu <peterx@redhat.com>
Date: Wed, 22 Sep 2021 16:23:33 -0400
Subject: [PATCH] mm/khugepaged: Don't recycle vma pgtable if uffd-wp
 registered

When we're trying to collapse a 2M huge shmem page, don't retract pgtable pmd
page if it's registered with uffd-wp, because that pgtable could have pte
markers installed.  Recycling of that pgtable means we'll lose the pte markers.
That could cause data loss for an uffd-wp enabled application on shmem.

Instead of disabling khugepaged on these files, simply skip retracting these
special VMAs, then the page cache can still be merged into a huge thp, and
other mm/vma can still map the range of file with a huge thp when proper.

Note that checking VM_UFFD_WP needs to be done with mmap_sem held for write,
that avoids race like:

         khugepaged                             user thread
         ==========                             ===========
     check VM_UFFD_WP, not set
                                       UFFDIO_REGISTER with uffd-wp on shmem
                                       wr-protect some pages (install markers)
     take mmap_sem write lock
     erase pmd and free pmd page
      --> pte markers are dropped unnoticed!

Signed-off-by: Peter Xu <peterx@redhat.com>
---
 mm/khugepaged.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 045cc579f724..23e1d03156b3 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1451,6 +1451,10 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 	if (!hugepage_vma_check(vma, vma->vm_flags | VM_HUGEPAGE))
 		return;
 
+	/* Keep pmd pgtable for uffd-wp; see comment in retract_page_tables() */
+	if (userfaultfd_wp(vma))
+		return;
+
 	hpage = find_lock_page(vma->vm_file->f_mapping,
 			       linear_page_index(vma, haddr));
 	if (!hpage)
@@ -1591,7 +1595,15 @@ static void retract_page_tables(struct address_space *mapping, pgoff_t pgoff)
 		 * reverse order. Trylock is a way to avoid deadlock.
 		 */
 		if (mmap_write_trylock(mm)) {
-			if (!khugepaged_test_exit(mm)) {
+			/*
+			 * When a vma is registered with uffd-wp, we can't
+			 * recycle the pmd pgtable because there can be pte
+			 * markers installed.  Skip it only, so the rest mm/vma
+			 * can still have the same file mapped hugely, however
+			 * it'll always mapped in small page size for uffd-wp
+			 * registered ranges.
+			 */
+			if (!khugepaged_test_exit(mm) && !userfaultfd_wp(vma)) {
 				spinlock_t *ptl = pmd_lock(mm, pmd);
 				/* assume page table is clear */
 				_pmd = pmdp_collapse_flush(vma, addr, pmd);
-- 
2.31.1