mm: fix memory leak in copy_huge_pmd()
authorMatthew Wilcox <matthew.r.wilcox@intel.com>
Wed, 3 Feb 2016 00:57:57 +0000 (16:57 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 3 Feb 2016 16:28:43 +0000 (08:28 -0800)
We allocate a pgtable but do not attach it to anything if the PMD is in
a DAX VMA, causing it to leak.

We certainly try to not free pgtables associated with the huge zero page
if the zero page is in a DAX VMA, so I think this is the right solution.
This needs to be properly audited.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/huge_memory.c

index c1411961167e1257c1a56b000512f36d1766471e..36c070167b7109f0639ff4188b6b67772e832591 100644 (file)
@@ -858,7 +858,8 @@ static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
                return false;
        entry = mk_pmd(zero_page, vma->vm_page_prot);
        entry = pmd_mkhuge(entry);
-       pgtable_trans_huge_deposit(mm, pmd, pgtable);
+       if (pgtable)
+               pgtable_trans_huge_deposit(mm, pmd, pgtable);
        set_pmd_at(mm, haddr, pmd, entry);
        atomic_long_inc(&mm->nr_ptes);
        return true;
@@ -1036,13 +1037,15 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        spinlock_t *dst_ptl, *src_ptl;
        struct page *src_page;
        pmd_t pmd;
-       pgtable_t pgtable;
+       pgtable_t pgtable = NULL;
        int ret;
 
-       ret = -ENOMEM;
-       pgtable = pte_alloc_one(dst_mm, addr);
-       if (unlikely(!pgtable))
-               goto out;
+       if (!vma_is_dax(vma)) {
+               ret = -ENOMEM;
+               pgtable = pte_alloc_one(dst_mm, addr);
+               if (unlikely(!pgtable))
+                       goto out;
+       }
 
        dst_ptl = pmd_lock(dst_mm, dst_pmd);
        src_ptl = pmd_lockptr(src_mm, src_pmd);
@@ -1073,7 +1076,7 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
                goto out_unlock;
        }
 
-       if (pmd_trans_huge(pmd)) {
+       if (!vma_is_dax(vma)) {
                /* thp accounting separate from pmd_devmap accounting */
                src_page = pmd_page(pmd);
                VM_BUG_ON_PAGE(!PageHead(src_page), src_page);