[PATCH 5/6] mremap: Check if it's possible to split original vma

[Dmitry Safonov <dima@arista.com>]

If original VMA can't be split at the desired address, do_munmap() will
fail and leave both new-copied VMA and old VMA. De-facto it's
MREMAP_DONTUNMAP behaviour, which is unexpected.

Currently, it may fail such way for hugetlbfs and dax device mappings.

Minimize such unpleasant situations to OOM by checking .may_split()
before attempting to create a VMA copy.

Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 mm/mremap.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/mm/mremap.c b/mm/mremap.c
index 898e9818ba6d..3c4047c23673 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -343,7 +343,7 @@ static unsigned long move_vma(struct vm_area_struct *vma,
 	unsigned long excess = 0;
 	unsigned long hiwater_vm;
 	int split = 0;
-	int err;
+	int err = 0;
 	bool need_rmap_locks;
 
 	/*
@@ -353,6 +353,15 @@ static unsigned long move_vma(struct vm_area_struct *vma,
 	if (mm->map_count >= sysctl_max_map_count - 3)
 		return -ENOMEM;
 
+	if (vma->vm_ops && vma->vm_ops->may_split) {
+		if (vma->vm_start != old_addr)
+			err = vma->vm_ops->may_split(vma, old_addr);
+		if (!err && vma->vm_end != old_addr + old_len)
+			err = vma->vm_ops->may_split(vma, old_addr + old_len);
+		if (err)
+			return err;
+	}
+
 	/*
 	 * Advise KSM to break any KSM pages in the area to be moved:
 	 * it would be confusing if they were to turn up at the new
-- 
2.28.0