HostAP in the kernel ?

[Jouni Malinen]

On Wed, Nov 20, 2002 at 12:21:41PM -0800, Jean Tourrilhes wrote:

> 	I would like to plan in advance to make sure we get the HostAP
> driver integrated in kernel 2.6.X. We still have until this summer,
> but I hate to do things last minute.

Yes, I agree with this completely. I think that the renamed files
(hostap prefix in every file) was enough to make you happy with using
linux/drivers/net/wireless. I have a patch for integrating the driver
into 2.5.27 kernel and I will update it to the latest one.

I think that I will also take couple of configuration items (like
hostapd support and possibly firmware download support) from
hostap_config.h and add them to kernel configuration. In addition,
hostapd and all user space tools could be separated to a separate
package for those using kernel version of the driver.

> 	Concerns :
> 	o 2.5.X stabilisation. My hope is that in a few month they
> will stop changing kernel APIs. Currently module support is broken.

Well, at least I'm now using 2.5.47 in my main development host, so the
driver gets tested on that. ChangeLog for 2.5.48 was too scare for me to
test it yet on this host ;-). Anyway, I think I will be following latest
2.5.x versions now, since my laptop kind of requires them to get all
needed functions working..

> 	o crypto : the kernel just included a new crypto API. Maybe
> you may want to hook host-wep to it, that would be an easier sell.

Current kernel crypto implementation does not seem to support RC4, so it
might not be best possible solutions at least yet. CRC32 code is
probably already in kernel, so that could be used. However, current WEP
code in Host AP driver interleaves RC4 and CRC32 operations can goes
through the payload in one pass. I would rather not change this unless
the new version is performing at least as well as the implementation in
the driver. Anyway, hostap_crypt_wep.c is not really that large a piece
of code..

As far as supporting other encryption algorithms, crypto API might be
good starting point. However, I would still like to be able to support
2.2 and 2.4 kernel versions, so there might be alternative
implementations. Modular encryption support in the current driver should
make this relatively easy, so this should be OK.

> 	On the Wireless Extension side, I believe that everything we
> discussed has been included. I'm planning to work on WE-16 (mostly
> reworking iwspy support), so if you need an extra feature, tell me
> about it.

I think that wireless extensions situation is quite good now. I don't
think I have any new requirements for that at the moment, but I would
probably need to go through the driver code at some point and take a
closer look at this.

-- 
Jouni Malinen                                            PGP id EFC895FA