hugetlb: fix memoryleak in hugetlb_mcopy_atomic_pte

commit da9a298 upstream.

When alloc_huge_page fails, *pagep is set to NULL without put_page first.
So the hugepage indicated by *pagep is leaked.

Link: https://lkml.kernel.org/r/20220709092629.54291-1-linmiaohe@huawei.com
Fixes: 8cc5fcb ("mm, hugetlb: fix racy resv_huge_pages underflow on UFFDIO_COPY")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Acked-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

mm/hugetlb.c

@@ -5314,6 +5314,7 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
 		page = alloc_huge_page(dst_vma, dst_addr, 0);
 		if (IS_ERR(page)) {
+			put_page(*pagep);
 			ret = -ENOMEM;
 			*pagep = NULL;
 			goto out;