From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jean Tourrilhes Subject: Re: Updated WE-18 (WPA) proposal Date: Mon, 30 Aug 2004 09:50:26 -0700 Sender: hostap-bounces+gldh-hostap-597=gmane.org@shmoo.com Message-ID: <20040830165026.GD29492@bougret.hpl.hp.com> References: <20040830045441.GA7415@jm.kir.nu> Reply-To: jt@hpl.hp.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: To: Jouni Malinen , netdev@oss.sgi.com, hostap@shmoo.com, Pedro Ramalhais Content-Disposition: inline In-Reply-To: <20040830045441.GA7415@jm.kir.nu> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: hostap-bounces+gldh-hostap-597=gmane.org@shmoo.com List-Id: netdev.vger.kernel.org On Sun, Aug 29, 2004 at 09:54:41PM -0700, Jouni Malinen wrote: > Finally, I had enough time to implement and test the proposed WE-18 > (WPA) changes with Host AP driver and wpa_supplicant. Great ! > Since WE-17 has apparently not yet been merged all the way into > linux-2.6 tree, the patch below is against Linux 2.6.8.1 that has been > patched with WE-17 patch (http://www.hpl.hp.com/personal/ > Jean_Tourrilhes/Linux/iw268_we17-10.diff). Don't worry, I'll fix that. Anyway, WE-17 is pending in Jeff's tree, and I don't think he will make major changes to it. > - replaced optional parameter (iw_point) to SIOCSIWSCAN with a new ioctl > (SIOCSIWSCANEXT) since the previous design was not really backwards > compatible (e.g., 'iwlist wlan0 scan' did not work) Latest Wireless Tools actually fixes that. Most distro seems to have adopted WT-27-preXX, and I plan to release WT-27 soon after WE-17, so I would not consider that a big issue. Having a separate ioctl has one advantage, you know if the driver support it or not. One the other hand, having a single ioctl may reduce bloat. > Question: is length field in struct iw_point in bytes or tokens > (token_size bytes)? I assumed it was in bytes, but this did not work > very well with WE ioctls that had token_size != 1; I made SIOCSIWSCANEXT > use token_size = 1 for now, but it could be replaced to be > sizeof(struct) and min_tokens=max_tokesn=1 once this question is > resolved. Originally, I was using length == num-tokens, with token-size != 1. However, after a while, I realised that having length == num-bytes was a much better option, so that's why the "newer" ioctls tend to all have token_size == 1. In the case of SIOCSIWSCANEXT, it's especially important as the struct may grow in the future, so the size would allow to distinguish the various additions. Thanks a lot ! Jean