mm,hugetlb: take hugetlb_lock before decrementing h->resv_huge_pages

The h->*_huge_pages counters are protected by the hugetlb_lock, but
alloc_huge_page has a corner case where it can decrement the counter
outside of the lock.

This could lead to a corrupted value of h->resv_huge_pages, which we have
observed on our systems.

Take the hugetlb_lock before decrementing h->resv_huge_pages to avoid a
potential race.

Link: https://lkml.kernel.org/r/20221017202505.0e6a4fcd@imladris.surriel.com
Fixes: a88c76954804 ("mm: hugetlb: fix hugepage memory leak caused by wrong reserve count")
Signed-off-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Glen McCready <gkmccready@meta.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index b586cdd75930b93db9f293c5e190906cd30ffd3a..dede0337c07c7a2c4dcb8bede53f2040031f8e4b 100644 (file)
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2924,11 +2924,11 @@ struct page *alloc_huge_page(struct vm_area_struct *vma,
                page = alloc_buddy_huge_page_with_mpol(h, vma, addr);
                if (!page)
                        goto out_uncharge_cgroup;
+               spin_lock_irq(&hugetlb_lock);
                if (!avoid_reserve && vma_has_reserves(vma, gbl_chg)) {
                        SetHPageRestoreReserve(page);
                        h->resv_huge_pages--;
                }
-               spin_lock_irq(&hugetlb_lock);
                list_add(&page->lru, &h->hugepage_activelist);
                set_page_refcounted(page);
                /* Fall through */