Re: [PATCH] unserialized task->files changing (v2)

["Pavel V. Emelianov" <xemul@sw.ru>]

Eric Dumazet wrote:
> On Tuesday 08 August 2006 13:31, Kirill Korotaev wrote:
>> Fixed race on put_files_struct on exec with proc.
>> Restoring files on current on error path may lead
>> to proc having a pointer to already kfree-d files_struct.
>>
>> ->files changing at exit.c and khtread.c are safe as
>> exit_files() makes all things under lock.
>>
>> v2 patch changes:
>> - introduced reset_files_struct() as Christoph Hellwig suggested
>>
>> Found during OpenVZ stress testing.
>
> Sorry but there is something I dont understand. You ignored my point.
>
> +void reset_files_struct(struct task_struct *tsk, struct files_struct 
> *files)
> +{
> +       struct files_struct *old;
> +
> +       old = tsk->files;
> +       task_lock(tsk);
> +       tsk->files = files;
> +       task_unlock(tsk);
> +       put_files_struct(old);
> +}
>
> Its seems very strange to protect tsk->files = files with a
> task_lock()/task_unlock(). What is it supposed to guard against ???
>
> If this patch corrects the 'bug', then a simpler fix would be to use a 
> memory
> barrier between "tsk->files = files" and "put_files_struct(old);"
>
> No need to perform 2 atomics ops on the task lock.
>
> old = tsk->files;
> tsk->files = files;
> smp_mb();
> put_files_struct(old);

No. The race being discussed is:

proc code:                             resetting code:
=============================================================================
task_lock(tsk);
files = tsk->files;
                                       old = tsk->files;
                                       tsk->files = files;
                                       put_files_struct(old); /* dec to 0 */
                                            `- kmem_cache_free(files);
get_files_struct(file); /* already free */
task_unlock(tsk);

So having smp_mb() before put_files_struct() does not fix the problem.

>
> That would be enough to guard against proc code (because this code 
> only needs
> to read tsk->files of course)
>
> The same remark can be said for __exit_files() from kernel/exit.c
>
> If this task_lock()/task_unlock() patch is really needed, then a 
> comment in
> the source would be very fair.
>
> Eric