[PATCH 4.19 06/34] mm/thp: try_to_unmap() use TTU_SYNC for safe splitting

[Greg Kroah-Hartman]

From: Hugh Dickins <hughd@xxxxxxxxxx>

[ Upstream commit 732ed55823fc3ad998d43b86bf771887bcc5ec67 ]

Stressing huge tmpfs often crashed on unmap_page()'s VM_BUG_ON_PAGE
(!unmap_success): with dump_page() showing mapcount:1, but then its raw
struct page output showing _mapcount ffffffff i.e.  mapcount 0.

And even if that particular VM_BUG_ON_PAGE(!unmap_success) is removed,
it is immediately followed by a VM_BUG_ON_PAGE(compound_mapcount(head)),
and further down an IS_ENABLED(CONFIG_DEBUG_VM) total_mapcount BUG():
all indicative of some mapcount difficulty in development here perhaps.
But the !CONFIG_DEBUG_VM path handles the failures correctly and
silently.

I believe the problem is that once a racing unmap has cleared pte or
pmd, try_to_unmap_one() may skip taking the page table lock, and emerge
from try_to_unmap() before the racing task has reached decrementing
mapcount.

Instead of abandoning the unsafe VM_BUG_ON_PAGE(), and the ones that
follow, use PVMW_SYNC in try_to_unmap_one() in this case: adding
TTU_SYNC to the options, and passing that from unmap_page().

When CONFIG_DEBUG_VM, or for non-debug too? Consensus is to do the same
for both: the slight overhead added should rarely matter, except perhaps
if splitting sparsely-populated multiply-mapped shmem.  Once confident
that bugs are fixed, TTU_SYNC here can be removed, and the race
tolerated.

Link: https://lkml.kernel.org/r/c1e95853-8bcd-d8fd-55fa-e7f2488e78f@xxxxxxxxxx
Fixes: fec89c109f3a ("thp: rewrite freeze_page()/unfreeze_page() with generic rmap walkers")
Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
Cc: Alistair Popple <apopple@xxxxxxxxxx>
Cc: Jan Kara <jack@xxxxxxx>
Cc: Jue Wang <juew@xxxxxxxxxx>
Cc: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Cc: "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx>
Cc: Miaohe Lin <linmiaohe@xxxxxxxxxx>
Cc: Minchan Kim <minchan@xxxxxxxxxx>
Cc: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
Cc: Oscar Salvador <osalvador@xxxxxxx>
Cc: Peter Xu <peterx@xxxxxxxxxx>
Cc: Ralph Campbell <rcampbell@xxxxxxxxxx>
Cc: Shakeel Butt <shakeelb@xxxxxxxxxx>
Cc: Wang Yugui <wangyugui@xxxxxxxxxxxx>
Cc: Yang Shi <shy828301@xxxxxxxxx>
Cc: Zi Yan <ziy@xxxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>

Note on stable backport: upstream TTU_SYNC 0x10 takes the value which
5.11 commit 013339df116c ("mm/rmap: always do TTU_IGNORE_ACCESS") freed.
It is very tempting to backport that commit (as 5.10 already did) and
make no change here; but on reflection, good as that commit is, I'm
reluctant to include any possible side-effect of it in this series.

Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
 include/linux/rmap.h |  3 ++-
 mm/huge_memory.c     |  2 +-
 mm/page_vma_mapped.c | 11 +++++++++++
 mm/rmap.c            | 17 ++++++++++++++++-
 4 files changed, 30 insertions(+), 3 deletions(-)

diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index d7d6d4eb1794..91ccae946716 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -98,7 +98,8 @@ enum ttu_flags {
 					 * do a final flush if necessary */
 	TTU_RMAP_LOCKED		= 0x80,	/* do not grab rmap lock:
 					 * caller holds it */
-	TTU_SPLIT_FREEZE	= 0x100,		/* freeze pte under splitting thp */
+	TTU_SPLIT_FREEZE	= 0x100, /* freeze pte under splitting thp */
+	TTU_SYNC		= 0x200, /* avoid racy checks with PVMW_SYNC */
 };
 
 #ifdef CONFIG_MMU
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 82ed62775c00..78c1ad5f8109 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2430,7 +2430,7 @@ void vma_adjust_trans_huge(struct vm_area_struct *vma,
 static void unmap_page(struct page *page)
 {
 	enum ttu_flags ttu_flags = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS |
-		TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD;
+		TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD | TTU_SYNC;
 	bool unmap_success;
 
 	VM_BUG_ON_PAGE(!PageHead(page), page);
diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c
index 11df03e71288..08e283ad4660 100644
--- a/mm/page_vma_mapped.c
+++ b/mm/page_vma_mapped.c
@@ -208,6 +208,17 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
 			pvmw->ptl = NULL;
 		}
 	} else if (!pmd_present(pmde)) {
+		/*
+		 * If PVMW_SYNC, take and drop THP pmd lock so that we
+		 * cannot return prematurely, while zap_huge_pmd() has
+		 * cleared *pmd but not decremented compound_mapcount().
+		 */
+		if ((pvmw->flags & PVMW_SYNC) &&
+		    PageTransCompound(pvmw->page)) {
+			spinlock_t *ptl = pmd_lock(mm, pvmw->pmd);
+
+			spin_unlock(ptl);
+		}
 		return false;
 	}
 	if (!map_pte(pvmw))
diff --git a/mm/rmap.c b/mm/rmap.c
index 70872d5b203c..5df055654e63 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1348,6 +1348,15 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 	unsigned long start = address, end;
 	enum ttu_flags flags = (enum ttu_flags)arg;
 
+	/*
+	 * When racing against e.g. zap_pte_range() on another cpu,
+	 * in between its ptep_get_and_clear_full() and page_remove_rmap(),
+	 * try_to_unmap() may return false when it is about to become true,
+	 * if page table locking is skipped: use TTU_SYNC to wait for that.
+	 */
+	if (flags & TTU_SYNC)
+		pvmw.flags = PVMW_SYNC;
+
 	/* munlock has nothing to gain from examining un-locked vmas */
 	if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
 		return true;
@@ -1723,7 +1732,13 @@ bool try_to_unmap(struct page *page, enum ttu_flags flags)
 	else
 		rmap_walk(page, &rwc);
 
-	return !page_mapcount(page) ? true : false;
+	/*
+	 * When racing against e.g. zap_pte_range() on another cpu,
+	 * in between its ptep_get_and_clear_full() and page_remove_rmap(),
+	 * try_to_unmap() may return false when it is about to become true,
+	 * if page table locking is skipped: use TTU_SYNC to wait for that.
+	 */
+	return !page_mapcount(page);
 }
 
 /**
-- 
2.30.2