|
← Revision 3 as of 2002-01-18 14:11:11
Size: 2234
Comment:
|
← Revision 4 as of 2002-07-08 20:27:54 →
Size: 4841
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 15: | Line 15: |
| Just a few notes from reading comments in the code... | 7/8/2002 |
| Line 17: | Line 17: |
| There is a PRISM2_MONITOR directive that enables a netlink socket for dumping/sniffing raw wlan frames. Will have to try this.. ;) | Well some time has gone on and this driver does some pretty neat things. |
| Line 19: | Line 19: |
| Some evidence of the driver supporting AP/AP bridging? | WDS aka WirelessDistributionSystem... allows access points to talk to other access points. The Prism chipsets are actually broken in this regard as they cannot send 4-address frames in HostAP mode. So, following on the original extensions done to the driver by the Instant802 OpenAP project, Jouni implemented a workaround by which 3-address frames are sent with a 4th address in an footer portion. |
| Line 21: | Line 21: |
|
{{{/* allow this for (pseudo) ad-hoc style communication * between APs; originating AP removes ToDS from * unicast packets, but does not currently know how to * do it for broadcast packets (not easy to know, * whether they are to associated STAs or * not-associated APs */ /* FIX: this accepts unwanted broadcast frames from * other BSSes and these frames should be droped if * pseudo ad-hoc inter-AP communication is not needed */}}} |
So essentially now HostAP access points can talk to each other. The interface works such that when WDS is enabled and another HostAP AP is in range, an interface of form wlanNwdsM appears for use. This is an altogether logically different ethernet that only these two APs share. These interfaces can be added to a Linux bridge group and STP run such that a mesh of HostAP APs can have and offer to their client STAs seamless layer 2 connectivity. Or, these interfaces can be routed between in a somewhat more efficient though harder to configure scheme. It is not abundantly clear what the best automated routing protocol for this arrangement would be. |
| Line 32: | Line 23: |
| There are comments that confirm an only partial support for Wireless Extensions and no support for WSPY, perhaps as related to below... | Remember that fundamentally the card is only on one RF channel at a time so in a group of HostAP APs all links are dependent on the openness of a single radio channel. Of note is that within a WDS mesh there is no resolution of any HiddenNode issues... it's the same topology as pure ad-hoc between the HostAP APs. One tip is to decrease the beacon intervals of the HostAP APs such that less radio time is monopolized such. Changing to 1s intervals from 1/100s allowed approximately 15% faster transfer rates between a two-node WDS test system. |
| Line 34: | Line 25: |
| Here's a clue about some limitations... I assume this functionality is standard fare in other "real" APs? He hardcodes the tx_rate to 11 Mb/s. | Also added this spring was a MAC access control list and functions to dynamically kick MACs from association. |
| Line 36: | Line 27: |
|
{{{/* FIX: set tx_rate if f/w does not know how to do it; * in Host AP mode this would require keeping records for each station * about their signal-level and maximal rate, in addition, the driver * would need to try to reduce the rate when packets are lost and * increase the rate, if signal quality improves */}}} |
Other news is that the open1x.org (UMD) implementation of the IEEE 802.1x authenticator works with the HostAP driver exclusively. This is a system whereby 509.x PKI certs are used as authentication tokens over the ethernet (LEAP), typically with a backend to a Radius server from the AP. Certain Intel and Cisco, etc. APs support this functionality out of the box. Available also is a supplicant for unix clients to authenticate to an 802.1x ethernet. (Note: I have never gotten 802.1x to work but that I'm sure is my fault ;-)) HostAP now supports WEP. Due to quirks in the Prism firmware decryption must be done on the host machine in HostAP mode, which creates a small engineering issue when determining host system requirements. Optional in the HostAP configuration recently is the hostapd implementation of a userspace authenticator. This will allow direct hooks to the plethora of authentication systems available on the unix platform HostAP drivers support PCI, PLX and PCMCIA versions of the Prism hardware. The make process for the drivers has changed this spring such that one no long must copy a tarball into the pcmcia-cs build tree to be made there. Define your KERNEL_PATH line in the Makefile, and IFF you wish to compile against an external (non-kernel-tree) pcmcia-cs the PCMCIA_PATH. Your choice of three make targets: "make pccard", "make pci", "make plx". Then "make install" if you are so inclined though this is rumored to not work rightaway on RedHat systems--I recommend just copying the driver/module/hostap_cs.o to the e.g. /lib/modules/2.4.18/pcmcia directory or the driver/module/hostap_pci.o to /lib/modules/2.4.18/kernel/drivers/net directory. HostAPMode functionality is standard in recent BSD kernel drivers for the Prism cards. Prism chipsets have an RFMonitor mode of operation which some drivers including the HostAP can take advantage of to listen to the raw 802.11b frames on the air. Something I didn't know until a little bit ago is that the Prism actually won't let you see certain malformed frames that a Cisco actually will. Just a sidenote. |
HostApMode is a special mode present only in Intersil's Prism2Card which enables a PcmciaCard to act as a true AccessPoint (eg. BssMode or InfrastructureMode) without requiring special firmware.
JouniMalinen wrote Linux drivers which utilize this feature which people are using to build LinuxAccessPoints.
The homepage for this software is:
http://people.ssh.com/jkm/Prism2 (used to be: http://194.241.248.126/Prism2/ and http://www.epitest.fi/Prism2/)
The announcement of his work this work was made (I think) first here:
There is a new [mailto:hostap@lists.ssh.com mailing list] available for discussion on the HostApMode drivers. You can [mailto:hostap-subscribe@lists.ssh.com subscribe] by sending mail and archives of the list are also [http://www.epitest.fi/hostap/ available].
7/8/2002
Well some time has gone on and this driver does some pretty neat things.
WDS aka WirelessDistributionSystem... allows access points to talk to other access points. The Prism chipsets are actually broken in this regard as they cannot send 4-address frames in HostAP mode. So, following on the original extensions done to the driver by the Instant802 OpenAP project, Jouni implemented a workaround by which 3-address frames are sent with a 4th address in an footer portion.
So essentially now HostAP access points can talk to each other. The interface works such that when WDS is enabled and another HostAP AP is in range, an interface of form wlanNwdsM appears for use. This is an altogether logically different ethernet that only these two APs share. These interfaces can be added to a Linux bridge group and STP run such that a mesh of HostAP APs can have and offer to their client STAs seamless layer 2 connectivity. Or, these interfaces can be routed between in a somewhat more efficient though harder to configure scheme. It is not abundantly clear what the best automated routing protocol for this arrangement would be.
Remember that fundamentally the card is only on one RF channel at a time so in a group of HostAP APs all links are dependent on the openness of a single radio channel. Of note is that within a WDS mesh there is no resolution of any HiddenNode issues... it's the same topology as pure ad-hoc between the HostAP APs. One tip is to decrease the beacon intervals of the HostAP APs such that less radio time is monopolized such. Changing to 1s intervals from 1/100s allowed approximately 15% faster transfer rates between a two-node WDS test system.
Also added this spring was a MAC access control list and functions to dynamically kick MACs from association.
Other news is that the open1x.org (UMD) implementation of the IEEE 802.1x authenticator works with the HostAP driver exclusively. This is a system whereby 509.x PKI certs are used as authentication tokens over the ethernet (LEAP), typically with a backend to a Radius server from the AP. Certain Intel and Cisco, etc. APs support this functionality out of the box. Available also is a supplicant for unix clients to authenticate to an 802.1x ethernet. (Note: I have never gotten 802.1x to work but that I'm sure is my fault ;-))
HostAP now supports WEP. Due to quirks in the Prism firmware decryption must be done on the host machine in HostAP mode, which creates a small engineering issue when determining host system requirements.
Optional in the HostAP configuration recently is the hostapd implementation of a userspace authenticator. This will allow direct hooks to the plethora of authentication systems available on the unix platform
HostAP drivers support PCI, PLX and PCMCIA versions of the Prism hardware. The make process for the drivers has changed this spring such that one no long must copy a tarball into the pcmcia-cs build tree to be made there. Define your KERNEL_PATH line in the Makefile, and IFF you wish to compile against an external (non-kernel-tree) pcmcia-cs the PCMCIA_PATH. Your choice of three make targets: "make pccard", "make pci", "make plx". Then "make install" if you are so inclined though this is rumored to not work rightaway on RedHat systems--I recommend just copying the driver/module/hostap_cs.o to the e.g. /lib/modules/2.4.18/pcmcia directory or the driver/module/hostap_pci.o to /lib/modules/2.4.18/kernel/drivers/net directory.
HostAPMode functionality is standard in recent BSD kernel drivers for the Prism cards.
Prism chipsets have an RFMonitor mode of operation which some drivers including the HostAP can take advantage of to listen to the raw 802.11b frames on the air. Something I didn't know until a little bit ago is that the Prism actually won't let you see certain malformed frames that a Cisco actually will. Just a sidenote.