mm/hugetlb: fix huge page reservation leak in private mapping error paths
authorMike Kravetz <mike.kravetz@oracle.com>
Thu, 10 Nov 2016 18:46:32 +0000 (10:46 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 11 Nov 2016 16:12:37 +0000 (08:12 -0800)
Error paths in hugetlb_cow() and hugetlb_no_page() may free a newly
allocated huge page.

If a reservation was associated with the huge page, alloc_huge_page()
consumed the reservation while allocating.  When the newly allocated
page is freed in free_huge_page(), it will increment the global
reservation count.  However, the reservation entry in the reserve map
will remain.

This is not an issue for shared mappings as the entry in the reserve map
indicates a reservation exists.  But, an entry in a private mapping
reserve map indicates the reservation was consumed and no longer exists.
This results in an inconsistency between the reserve map and the global
reservation count.  This 'leaks' a reserved huge page.

Create a new routine restore_reserve_on_error() to restore the reserve
entry in these specific error paths.  This routine makes use of a new
function vma_add_reservation() which will add a reserve entry for a
specific address/page.

In general, these error paths were rarely (if ever) taken on most
architectures.  However, powerpc contained arch specific code that that
resulted in an extra fault and execution of these error paths on all
private mappings.

Fixes: 67961f9db8c4 ("mm/hugetlb: fix huge page reserve accounting for private mappings)
Link: http://lkml.kernel.org/r/1476933077-23091-2-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reported-by: Jan Stancek <jstancek@redhat.com>
Tested-by: Jan Stancek <jstancek@redhat.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Kirill A . Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/hugetlb.c

index ec49d9ef1eefd0155f099d813ed547c16172d69e..418bf01a50ed1f9dde0ca6c083aafa86c1869f23 100644 (file)
@@ -1826,11 +1826,17 @@ static void return_unused_surplus_pages(struct hstate *h,
  * is not the case is if a reserve map was changed between calls.  It
  * is the responsibility of the caller to notice the difference and
  * take appropriate action.
+ *
+ * vma_add_reservation is used in error paths where a reservation must
+ * be restored when a newly allocated huge page must be freed.  It is
+ * to be called after calling vma_needs_reservation to determine if a
+ * reservation exists.
  */
 enum vma_resv_mode {
        VMA_NEEDS_RESV,
        VMA_COMMIT_RESV,
        VMA_END_RESV,
+       VMA_ADD_RESV,
 };
 static long __vma_reservation_common(struct hstate *h,
                                struct vm_area_struct *vma, unsigned long addr,
@@ -1856,6 +1862,14 @@ static long __vma_reservation_common(struct hstate *h,
                region_abort(resv, idx, idx + 1);
                ret = 0;
                break;
+       case VMA_ADD_RESV:
+               if (vma->vm_flags & VM_MAYSHARE)
+                       ret = region_add(resv, idx, idx + 1);
+               else {
+                       region_abort(resv, idx, idx + 1);
+                       ret = region_del(resv, idx, idx + 1);
+               }
+               break;
        default:
                BUG();
        }
@@ -1903,6 +1917,56 @@ static void vma_end_reservation(struct hstate *h,
        (void)__vma_reservation_common(h, vma, addr, VMA_END_RESV);
 }
 
+static long vma_add_reservation(struct hstate *h,
+                       struct vm_area_struct *vma, unsigned long addr)
+{
+       return __vma_reservation_common(h, vma, addr, VMA_ADD_RESV);
+}
+
+/*
+ * This routine is called to restore a reservation on error paths.  In the
+ * specific error paths, a huge page was allocated (via alloc_huge_page)
+ * and is about to be freed.  If a reservation for the page existed,
+ * alloc_huge_page would have consumed the reservation and set PagePrivate
+ * in the newly allocated page.  When the page is freed via free_huge_page,
+ * the global reservation count will be incremented if PagePrivate is set.
+ * However, free_huge_page can not adjust the reserve map.  Adjust the
+ * reserve map here to be consistent with global reserve count adjustments
+ * to be made by free_huge_page.
+ */
+static void restore_reserve_on_error(struct hstate *h,
+                       struct vm_area_struct *vma, unsigned long address,
+                       struct page *page)
+{
+       if (unlikely(PagePrivate(page))) {
+               long rc = vma_needs_reservation(h, vma, address);
+
+               if (unlikely(rc < 0)) {
+                       /*
+                        * Rare out of memory condition in reserve map
+                        * manipulation.  Clear PagePrivate so that
+                        * global reserve count will not be incremented
+                        * by free_huge_page.  This will make it appear
+                        * as though the reservation for this page was
+                        * consumed.  This may prevent the task from
+                        * faulting in the page at a later time.  This
+                        * is better than inconsistent global huge page
+                        * accounting of reserve counts.
+                        */
+                       ClearPagePrivate(page);
+               } else if (rc) {
+                       rc = vma_add_reservation(h, vma, address);
+                       if (unlikely(rc < 0))
+                               /*
+                                * See above comment about rare out of
+                                * memory condition.
+                                */
+                               ClearPagePrivate(page);
+               } else
+                       vma_end_reservation(h, vma, address);
+       }
+}
+
 struct page *alloc_huge_page(struct vm_area_struct *vma,
                                    unsigned long addr, int avoid_reserve)
 {
@@ -3498,6 +3562,7 @@ retry_avoidcopy:
        spin_unlock(ptl);
        mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 out_release_all:
+       restore_reserve_on_error(h, vma, address, new_page);
        put_page(new_page);
 out_release_old:
        put_page(old_page);
@@ -3680,6 +3745,7 @@ backout:
        spin_unlock(ptl);
 backout_unlocked:
        unlock_page(page);
+       restore_reserve_on_error(h, vma, address, page);
        put_page(page);
        goto out;
 }