mm/migration: fix potential page refcounts leak in migrate_pages

In -ENOMEM case, there might be some subpages of fail-to-migrate THPs left
in thp_split_pages list.  We should move them back to migration list so
that they could be put back to the right list by the caller otherwise the
page refcnt will be leaked here.  Also adjust nr_failed and nr_thp_failed
accordingly to make vm events account more accurate.

Link: https://lkml.kernel.org/r/20220318111709.60311-10-linmiaohe@huawei.com
Fixes: b5bade97 ("mm: migrate: fix the return value of migrate_pages()")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Cc: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/migrate.c

@@ -1434,6 +1434,14 @@ int migrate_pages(struct list_head *from, new_page_t get_new_page,
 				}
 
 				nr_failed_pages += nr_subpages;
+				/*
+				 * There might be some subpages of fail-to-migrate THPs
+				 * left in thp_split_pages list. Move them back to migration
+				 * list so that they could be put back to the right list by
+				 * the caller otherwise the page refcnt will be leaked.
+				 */
+				list_splice_init(&thp_split_pages, from);
+				nr_thp_failed += thp_retry;
 				goto out;
 			case -EAGAIN:
 				if (is_thp)