Re: Potential data race in SyS_swapon

[Cesar Eduardo Barros]

Em 07-08-2015 13:14, Andrey Konovalov escreveu:
> Hi!
>
> We are working on a dynamic data race detector for the Linux kernel
> called KernelThreadSanitizer (ktsan)
> (https://github.com/google/ktsan/wiki).
>
> While running ktsan on the upstream revision 21bdb584af8c with trinity
> we got a few reports from SyS_swapon, here is one of them:

[...]

> The race is happening when accessing the swap_file field of a
> swap_info_struct struct.
>
> 2392         for (i = 0; i < nr_swapfiles; i++) {
> 2393                 struct swap_info_struct *q = swap_info[i];
> 2394
> 2395                 if (q == p || !q->swap_file)
> 2396                         continue;
> 2397                 if (mapping == q->swap_file->f_mapping) {
> 2398                         error = -EBUSY;
> 2399                         goto bad_swap;
> 2400                 }
> 2401         }
>
> 2539         spin_lock(&swap_lock);
> 2540         p->swap_file = NULL;
> 2541         p->flags = 0;
> 2542         spin_unlock(&swap_lock);

There's another (more important) place which sets the swap_file field to 
NULL, it's within swapoff. It's also protected by swap_lock.

> Since the swap_lock lock is not taken in the first snippet, it's
> possible for q->swap_file to be assigned to NULL and reloaded between
> executing lines 2395 and 2397, which might lead to a null pointer
> dereference.

I agree with that analysis. It should be possible to hit by racing 
swapon of a file with swapoff of another.

> Looks like the swap_lock should be taken when iterating through the
> swap_info array on lines 2392 - 2401.

I'd take that lock a couple of lines earlier, so that every place that 
sets the swap_file field on a swap_info_struct is behind swap_lock, for 
simplicity.

--
Cesar Eduardo Barros
cesarb@xxxxxxxxxxxxx
--
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/