Re: Multiple potential races on vma->vm_flags

[Oleg Nesterov]

On 09/11, Kirill A. Shutemov wrote:
>
> This one is tricky. I *assume* the mm cannot be generally accessible after
> mm_users drops to zero, but I'm not entirely sure about it.
> procfs? ptrace?

Well, all I can say is that proc/ptrace look fine afaics...

This is off-topic, but how about the patch below? Different threads can
expand different vma's at the same time under read_lock(mmap_sem), so
vma_lock_anon_vma() can't help to serialize "locked_vm += grow".

Oleg.

--- x/mm/mmap.c
+++ x/mm/mmap.c
@@ -2146,9 +2146,6 @@ static int acct_stack_growth(struct vm_a
 	if (security_vm_enough_memory_mm(mm, grow))
 		return -ENOMEM;
 
-	/* Ok, everything looks good - let it rip */
-	if (vma->vm_flags & VM_LOCKED)
-		mm->locked_vm += grow;
 	vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
 	return 0;
 }
@@ -2210,6 +2207,8 @@ int expand_upwards(struct vm_area_struct
 				 * against concurrent vma expansions.
 				 */
 				spin_lock(&vma->vm_mm->page_table_lock);
+				if (vma->vm_flags & VM_LOCKED)
+					mm->locked_vm += grow;
 				anon_vma_interval_tree_pre_update_vma(vma);
 				vma->vm_end = address;
 				anon_vma_interval_tree_post_update_vma(vma);
@@ -2281,6 +2280,8 @@ int expand_downwards(struct vm_area_stru
 				 * against concurrent vma expansions.
 				 */
 				spin_lock(&vma->vm_mm->page_table_lock);
+				if (vma->vm_flags & VM_LOCKED)
+					mm->locked_vm += grow;
 				anon_vma_interval_tree_pre_update_vma(vma);
 				vma->vm_start = address;
 				vma->vm_pgoff -= grow;

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/