[PATCH] fix VmSize and VmData after mremap
Hugh Dickins [Thu, 4 Aug 2005 20:07:09 +0000 (13:07 -0700)]
mremap's move_vma is applying __vm_stat_account to the old vma which may
have already been freed: move it to just before the do_munmap.

mremapping to and fro with CONFIG_DEBUG_SLAB=y showed /proc/<pid>/status
VmSize and VmData wrapping just like in kernel bugzilla #4842, and fixed by
this patch - worth including in 2.6.13, though not yet confirmed that it
fixes that specific report from Frank van Maarseveen.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

mm/mremap.c

index ec7238a..fc45dc9 100644 (file)
@@ -229,6 +229,7 @@ static unsigned long move_vma(struct vm_area_struct *vma,
         * since do_munmap() will decrement it by old_len == new_len
         */
        mm->total_vm += new_len >> PAGE_SHIFT;
+       __vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
 
        if (do_munmap(mm, old_addr, old_len) < 0) {
                /* OOM: unable to split vma, just get accounts right */
@@ -243,7 +244,6 @@ static unsigned long move_vma(struct vm_area_struct *vma,
                        vma->vm_next->vm_flags |= VM_ACCOUNT;
        }
 
-       __vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
        if (vm_flags & VM_LOCKED) {
                mm->locked_vm += new_len >> PAGE_SHIFT;
                if (new_len > old_len)