[PATCH] mbind: fix verify_pages pte_page
Hugh Dickins [Wed, 22 Jun 2005 00:15:06 +0000 (17:15 -0700)]
Strict mbind's check that pages already mapped are on right node has been
using pte_page without checking if pfn_valid, and without page_table_lock
to prevent spurious failures when try_to_unmap_one intervenes between the
pte_present and the pte_page.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

mm/mempolicy.c

index 39252c7..c512cc9 100644 (file)
@@ -242,6 +242,9 @@ static int
 verify_pages(struct mm_struct *mm,
             unsigned long addr, unsigned long end, unsigned long *nodes)
 {
+       int err = 0;
+
+       spin_lock(&mm->page_table_lock);
        while (addr < end) {
                struct page *p;
                pte_t *pte;
@@ -268,17 +271,23 @@ verify_pages(struct mm_struct *mm,
                }
                p = NULL;
                pte = pte_offset_map(pmd, addr);
-               if (pte_present(*pte))
-                       p = pte_page(*pte);
+               if (pte_present(*pte)) {
+                       unsigned long pfn = pte_pfn(*pte);
+                       if (pfn_valid(pfn))
+                               p = pfn_to_page(pfn);
+               }
                pte_unmap(pte);
                if (p) {
                        unsigned nid = page_to_nid(p);
-                       if (!test_bit(nid, nodes))
-                               return -EIO;
+                       if (!test_bit(nid, nodes)) {
+                               err = -EIO;
+                               break;
+                       }
                }
                addr += PAGE_SIZE;
        }
-       return 0;
+       spin_unlock(&mm->page_table_lock);
+       return err;
 }
 
 /* Step 1: check the range */