Re: Concurrent access to /dev/urandom

[Matt Mackall]

On Wed, Dec 08, 2004 at 08:57:05PM -0500, Theodore Ts'o wrote:
> On Wed, Dec 08, 2004 at 01:56:14PM -0800, Matt Mackall wrote:
> > 
> > Ted, I think this is a bit more straightforward than your patch, and
> > safer as it protects get_random_bytes() and internal extract_entropy()
> > users. And I'd be leery of your get_cpu() trick due to preempt
> > issues.
> > 
> 
> I'm concerned that turning off interrupts during even a single SHA-1
> transform will put us above the radar with respect to the preempt
> latency statistics again.  We could use a separate spinlock that only
> pretects the mix_ptr and mixing access to the pool, so we're at least
> not disabling interrupts, but we still are holding a spinlock across a
> cryptographic operation.

It's been suggested to me that a sequence lock might be the right
approach to this, which I'll try to take a look at this evening. Also,
I'm going to time the lock hold time in my previous more conventional
patch and see what kind of neighborhood we're in.

-- 
Mathematics is the supreme nostalgia of our time.
-
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/