fs/hugetlbfs/inode.c: fix hwpoison reserve accounting
authorMike Kravetz <mike.kravetz@oracle.com>
Thu, 2 Nov 2017 22:59:41 +0000 (15:59 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 3 Nov 2017 14:39:19 +0000 (07:39 -0700)
Calling madvise(MADV_HWPOISON) on a hugetlbfs page will result in bad
(negative) reserved huge page counts.  This may not happen immediately,
but may happen later when the underlying file is removed or filesystem
unmounted.  For example:

  AnonHugePages:         0 kB
  ShmemHugePages:        0 kB
  HugePages_Total:       1
  HugePages_Free:        0
  HugePages_Rsvd:    18446744073709551615
  HugePages_Surp:        0
  Hugepagesize:       2048 kB

In routine hugetlbfs_error_remove_page(), hugetlb_fix_reserve_counts is
called after remove_huge_page.  hugetlb_fix_reserve_counts is designed
to only be called/used only if a failure is returned from
hugetlb_unreserve_pages.  Therefore, call hugetlb_unreserve_pages as
required and only call hugetlb_fix_reserve_counts in the unlikely event
that hugetlb_unreserve_pages returns an error.

Link: http://lkml.kernel.org/r/20171019230007.17043-2-mike.kravetz@oracle.com
Fixes: 78bb920344b8 ("mm: hwpoison: dissolve in-use hugepage in unrecoverable memory error")
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

index 59073e9f01a410f29150a3abf0087ac28409ad5b..ed113ea17affe6475523434d475aa2d94f6ce9c7 100644 (file)
@@ -842,9 +842,12 @@ static int hugetlbfs_error_remove_page(struct address_space *mapping,
                                struct page *page)
        struct inode *inode = mapping->host;
+       pgoff_t index = page->index;
-       hugetlb_fix_reserve_counts(inode);
+       if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
+               hugetlb_fix_reserve_counts(inode);
        return 0;