mm: pagewalk: call pte_hole() for VM_PFNMAP during walk_page_range
authorShiraz Hashim <shashim@codeaurora.org>
Thu, 5 Feb 2015 20:25:06 +0000 (12:25 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 11 Feb 2015 06:48:16 +0000 (14:48 +0800)
commit 23aaed6659df9adfabe9c583e67a36b54e21df46 upstream.

walk_page_range() silently skips vma having VM_PFNMAP set, which leads
to undesirable behaviour at client end (who called walk_page_range).
Userspace applications get the wrong data, so the effect is like just
confusing users (if the applications just display the data) or sometimes
killing the processes (if the applications do something with
misunderstanding virtual addresses due to the wrong data.)

For example for pagemap_read, when no callbacks are called against
VM_PFNMAP vma, pagemap_read may prepare pagemap data for next virtual
address range at wrong index.

Eventually userspace may get wrong pagemap data for a task.
Corresponding to a VM_PFNMAP marked vma region, kernel may report
mappings from subsequent vma regions.  User space in turn may account
more pages (than really are) to the task.

In my case I was using procmem, procrack (Android utility) which uses
pagemap interface to account RSS pages of a task.  Due to this bug it
was giving a wrong picture for vmas (with VM_PFNMAP set).

Fixes: a9ff785e4437 ("mm/pagewalk.c: walk_page_range should avoid VM_PFNMAP areas")
Signed-off-by: Shiraz Hashim <shashim@codeaurora.org>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/pagewalk.c

index 2beeabf502c50f1ff3069981bd5b3a84b7463b6f..9056d22d28808839cc1a65b41353d3b0f52779dc 100644 (file)
@@ -199,7 +199,10 @@ int walk_page_range(unsigned long addr, unsigned long end,
                         */
                        if ((vma->vm_start <= addr) &&
                            (vma->vm_flags & VM_PFNMAP)) {
-                               next = vma->vm_end;
+                               if (walk->pte_hole)
+                                       err = walk->pte_hole(addr, next, walk);
+                               if (err)
+                                       break;
                                pgd = pgd_offset(walk->mm, next);
                                continue;
                        }