[PATCH] Add -e encryption support to SMB3 client protocol support, plus documentation updates.

[Jeremy Allison]

On Wed, Aug 21, 2013 at 09:36:59AM -0700, Jeremy Allison wrote:
> On Wed, Aug 21, 2013 at 05:03:52PM +0200, Michael Adam wrote:
> > 
> > It seems to me from looking at captures that
> > with SMB1, I always get 0x4000 byte writes
> > and 0xFC00 reads, irrespective of the iosize
> > setting, but more parallelised when using a
> > higher value or 0.
> > (Observed running a thus patched smbclient
> > against a plain w2k8r2.)
> 
> The iosize is setting the "window" that
> is used for parallelization. I think setting
> it smaller than the minimum write size (0x4000)
> on SMB1 is probably being mapped to the minimum
> inside the clireadwrite.c code.
> 
> > Also using SMB2+, I don't see different write
> > or read lengths when using other values than 0,
> > but different parallelism. Am I asking stupid
> > questions?...
> 
> Nope - that's the fact that iosize is the window,
> not the actual read/write size.
> 
> > But you can still set it to a positive value
> > bigger than 0xFFFF00 for SMB2+ and with a
> > different effect than setting 0xFFFF00, right?
> 
> Yes, but setting it to anything other than
> zero for SMB2 is essentially slowing things
> down. Zero means the client dynamically adapts
> the window size based on the available credits.

I suppose if you want to be really accurate
you could prevent iosize from being set to
a value between the ranges 1 - (0x4000-1),
but this just seems a bit more complexity
than is worth it. Let me know if you want
to do that and I'll add a patch on top :-).

Jeremy.