[PATCH 5.14-rc1] mm/rmap: fix munlocking Anon THP with mlocked ptes

[Hugh Dickins]

Many thanks to Kirill for reminding that PageDoubleMap cannot be relied on
to warn of pte mappings in the Anon THP case; and a scan of subpages does
not seem appropriate here.  Note how follow_trans_huge_pmd() does not even
mark an Anon THP as mlocked when compound_mapcount != 1: multiple mlocking
of Anon THP is avoided, so simply return from page_mlock() in this case.

Link: https://lore.kernel.org/lkml/cfa154c-d595-406-eb7d-eb9df730f944@xxxxxxxxxx/
Fixes: d9770fcc1c0c ("mm/rmap: fix old bug: munlocking THP missed other mlocks")
Reported-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
Acked-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Alistair Popple <apopple@xxxxxxxxxx>
Cc: Jason Gunthorpe <jgg@xxxxxxxxxx>
Cc: Ralph Campbell <rcampbell@xxxxxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Yang Shi <shy828301@xxxxxxxxx>
Cc: Shakeel Butt <shakeelb@xxxxxxxxxx>
---
Linus, thanks a lot for last-minute hoovering up those four mm/rmap
patches, with 3/4 fixing the syzbot and 0day reports on munlocking ...

BUT

... the version of 2/4 in 5.14-rc1 is defective (PageDoubleMap is a
confusing flag which behaves differently on anon and file), Kirill had
spotted that, and what he Acked was the v2 which went into mmotm, rather
than the first version I posted.  This patch here converts the v1 in rc1
into the v2 Acked by Kirill.

What will go wrong with v1 in?  I don't actually know: nothing terrible,
can only affect people splitting and mlocking anon THPs, maybe nobody and
nobot will notice, maybe some VM_BUG_ONs or "Bad page"s will turn up.
I'll be on the lookout to point reporters to this fix (more lines than
strictly necessary, because it removes a level of indentation).

And sorry for putting 2/4 before the more urgent 3/4, but I couldn't
tell what to do in 3/4, without first fixing the older bug in 2/4.

Hugh

 mm/rmap.c |   39 ++++++++++++++++++++++-----------------
 1 file changed, 22 insertions(+), 17 deletions(-)

--- 5.14-rc1/mm/rmap.c
+++ linux/mm/rmap.c
@@ -1440,21 +1440,20 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		/*
 		 * If the page is mlock()d, we cannot swap it out.
 		 */
-		if (!(flags & TTU_IGNORE_MLOCK)) {
-			if (vma->vm_flags & VM_LOCKED) {
-				/* PTE-mapped THP are never marked as mlocked */
-				if (!PageTransCompound(page) ||
-				    (PageHead(page) && !PageDoubleMap(page))) {
-					/*
-					 * Holding pte lock, we do *not* need
-					 * mmap_lock here
-					 */
-					mlock_vma_page(page);
-				}
-				ret = false;
-				page_vma_mapped_walk_done(&pvmw);
-				break;
-			}
+		if (!(flags & TTU_IGNORE_MLOCK) &&
+		    (vma->vm_flags & VM_LOCKED)) {
+			/*
+			 * PTE-mapped THP are never marked as mlocked: so do
+			 * not set it on a DoubleMap THP, nor on an Anon THP
+			 * (which may still be PTE-mapped after DoubleMap was
+			 * cleared).  But stop unmapping even in those cases.
+			 */
+			if (!PageTransCompound(page) || (PageHead(page) &&
+			     !PageDoubleMap(page) && !PageAnon(page)))
+				mlock_vma_page(page);
+			page_vma_mapped_walk_done(&pvmw);
+			ret = false;
+			break;
 		}
 
 		/* Unexpected PMD-mapped THP? */
@@ -1986,8 +1985,10 @@ static bool page_mlock_one(struct page *page, struct vm_area_struct *vma,
 		 */
 		if (vma->vm_flags & VM_LOCKED) {
 			/*
-			 * PTE-mapped THP are never marked as mlocked, but
-			 * this function is never called when PageDoubleMap().
+			 * PTE-mapped THP are never marked as mlocked; but
+			 * this function is never called on a DoubleMap THP,
+			 * nor on an Anon THP (which may still be PTE-mapped
+			 * after DoubleMap was cleared).
 			 */
 			mlock_vma_page(page);
 			/*
@@ -2022,6 +2023,10 @@ void page_mlock(struct page *page)
 	VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
 	VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
 
+	/* Anon THP are only marked as mlocked when singly mapped */
+	if (PageTransCompound(page) && PageAnon(page))
+		return;
+
 	rmap_walk(page, &rwc);
 }