Question on HostAP, IAPP link layer update and bridge station cache

Jon-Olov Vatn vatn
Mon Mar 24 04:38:25 PST 2003


Hi!

I have a problem in the situation when using hostAP as a bridging AP.
In my network I have two bridging APs and a fixed host (Sender)
connected via an Ethernet bridge, see below.
I am using the current cvs version of hostAP (as of March 24 2003) in
order to have the new AP submit a link layer update frame when a STA 
(re)associates. The problem I have is that if the STA is not
"actively" transmitting any traffic it will take a long time before
the new AP will forward packets to it. I describe the situation a
bit more in detail below:

1) The STA associates with AP1. AP1 will submit a link layer
   update frame with Dst-MAC=broadcast and Src-MAC="MACofSTA".
2) Both the Ethernet bridge and and AP2 will discover the link layer
   update and store an entry about "MACofSTA" in their station caches.
   In the figure below this means that the Eth-bridge learns that
   MACofSTA resides on port1 and (the bridge inside) AP2 learns that
   MACofSTA resides on interface eth0.
3) Now assume that the fixed host (Sender) is pinging the STA
   (receiver), i.e., the STA transmits data (Echo Replies), but only
   when it receives some data from the sender (Echo Requests).
4) The STA then reassociates with AP2. AP2 will submit a link layer
   update frame with Dst-MAC=broadcast and Src-MAC={MACofSTA}.
5) Both the Ethernet bridge and (the bridge inside) AP1 learns that
   the STA has moved. The Echo requests from the Sender is now
   forwarded onto port 3 of the Ethernet bridge and will reach AP2.
6) The bridge inside AP2 has not(1) learned that the STA has moved, i.e.,
   it still believes that MACofSTA resides on eth0. The Echo Request
   packets are dropped at AP2 instead of being forwarded to the STA.
7) When the entry in station cache (of the bridge) in AP2 ages out, or
   when the STA decides to "actively" transmit a packet, the bridge
   will start to forward the Echo Requests to the STA.
   This may take several seconds.

Have anyone else also noticed this behaviour?

                   +-------+
                   |       |
                   | Sender|
                   |       |
                   +---+---+
		       |
		       |2
                   +---+---+
                  1| Eth.  |3
            +------+ bridge+------+
	    |      |       |      |
	    |      +-------+      |
	    | eth0                | eth0
	+---+---+             +---+---+
        |       |             |       |
        |  AP1  |             |  AP2  |
        |       |             |       |
	+---+---+             +---+---+
	    | wlan0               | wlan0
	    o                     o
                o
                |
            +---+---+
            |       |
            |  STA  | =>
            | (Rec) |
            +-------+


Note that this problem would not occur if the entry for MACofSTA in
AP2 had already aged out at the time of the handover, however, it
would be good if one did not have to rely on the aging timeout in the
AP bridge. I should also mention that I uncommented some lines in
"iapp.c" (for initializing "iapp_udp_sock") in order to avoid some run
time errors (perror("ioctl(SIOCGIFADDR)").  I do not believe that this
affects the scenario above. (Anyhow I have attached my modified iapp.c
file in case anyone is interested.)

Now to my question. If one would like to fix this problem, I can see
two different approaches, and I wonder what way to go:
* "Interaction between hostAP and briding code"
  One could make the hostap daemon (hostapd) interact with bridging
  code to make sure that the station cache (of the bridge) in the AP
  has no stale entry stating that MACofSTA resides on eth0.
  Even better would be to add an entry that specifies that MACofSTA
  now resides on interface wlan0.
  That is, upon a station (re)associating with an AP, the AP should
  update its own station cache appropiately by "interacting" with the
  bridging software.
* "Inserting link layer update frame in wlan0 input queue"
  If possible, the link layer update frame should be inserted in the
  input queue of wlan0, instead of sending it directly out on eth0.
  (Would this be feasible?)
  Then the bridge in the AP will(?) believe that MACofSTA resides on
  wlan0 and update its station cache. Also, it will forward the link
  layer update packet on eth0 so that the other bridges will update
  any stale routing cache entries. 

Which way would be the best to go?

(Yet another alternative would of course be to "show some patience" and
wait until the stale entry in the AP times out or until the STA
actually transmits any packet.)


BW J-O

----------------------------------------------------------------------
Jon-Olov Vatn                         Email:    vatn at it.kth.se
Royal Institute of Technology         Address:  KTH-IMIT
Dep. of Microelectronics and                    Isafjordsgatan 39
Information Technology                		S-164 40 Kista  SWEDEN
Telecommunication System Lab.         Fax:      +46 8 751 17 93

-------------- next part --------------
/*
 * Host AP (software wireless LAN access point) user space daemon for
 * Host AP kernel driver / IEEE 802.11f Inter-Access Point Protocol (IAPP)
 * Copyright (c) 2002-2003, Jouni Malinen <jkmaline at cc.hut.fi>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation. See README and COPYING for
 * more details.
 */

/* TODO:
 * - add support for MOVE-notify and MOVE-response (this requires support for
 *   finding out IP address for previous AP using RADIUS)
 * - add support for Send- and ACK-Security-Block to speedup IEEE 802.1X during
 *   reassociation to another AP
 * - implement counters etc. for IAPP MIB
 * - verify endianness of fields in IAPP messages; are they big-endian as
 *   used here?
 */

#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <netinet/in.h>
#include <net/if.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <arpa/inet.h>
#ifdef USE_KERNEL_HEADERS
#include <linux/if_packet.h>
#else /* USE_KERNEL_HEADERS */
#include <netpacket/packet.h>
#endif /* USE_KERNEL_HEADERS */

#include "hostapd.h"
#include "ieee802_11.h"
#include "iapp.h"
#include "eloop.h"
#include "sta_info.h"
#include "driver.h"


static void iapp_send_add(hostapd *hapd, struct sta_info *sta)
{
	char buf[128];
	struct iapp_hdr *hdr;
	struct iapp_add_notify *add;
	struct ieee80211_mgmt *assoc;
	struct sockaddr_in addr;

	/* Send IAPP-ADD Packet to remove possible association from other APs
	 */

	hdr = (struct iapp_hdr *) buf;
	hdr->version = IAPP_VERSION;
	hdr->command = IAPP_CMD_ADD_notify;
	hdr->identifier = host_to_be16(hapd->iapp_identifier++);
	hdr->length = host_to_be16(sizeof(*hdr) + sizeof(*add));

	add = (struct iapp_add_notify *) (hdr + 1);
	add->addr_len = ETH_ALEN;
	add->reserved = 0;
	memcpy(add->mac_addr, sta->addr, ETH_ALEN);

	assoc = sta->last_assoc_req;
	if (assoc) {
		u16 seq = WLAN_GET_SEQ_SEQ(le_to_host16(assoc->seq_ctrl));
		add->seq_num = host_to_be16(seq);
	} else
		add->seq_num = 0;
	
	/* Send to local subnet address (UDP port IAPP_PORT) */
	memset(&addr, 0, sizeof(addr));
	addr.sin_family = AF_INET;
	addr.sin_addr.s_addr = hapd->iapp_broadcast.s_addr;
	addr.sin_port = htons(IAPP_PORT);
	if (sendto(hapd->iapp_udp_sock, buf, (char *) (add + 1) - buf, 0,
		   (struct sockaddr *) &addr, sizeof(addr)) < 0)
		perror("sendto[IAPP-ADD]");
}


static void iapp_send_layer2_update(hostapd *hapd, struct sta_info *sta)
{
	struct iapp_layer2_update msg;

	/* Send Level 2 Update Frame to update forwarding tables in layer 2
	 * bridge devices */

	/* 802.2 Type 1 Logical Link Control (LLC) Exchange Identifier (XID)
	 * Update response frame; IEEE Std 802.2-1998, 5.4.1.2.1 */

	memset(msg.da, 0xff, ETH_ALEN);
	memcpy(msg.sa, sta->addr, ETH_ALEN);
	msg.len = host_to_be16(8);
	msg.dsap = 0;
	msg.ssap = 0;
	msg.control = 0xaf; /* XID response lsb.1111F101.
			     * F=0 (no poll command; unsolicited frame) */
	msg.xid_info[0] = 0x81; /* XID format identifier */
	msg.xid_info[1] = 1; /* LLC types/classes: Type 1 LLC */
	msg.xid_info[2] = 1 << 1; /* XID sender's receive window size (RW)
				   * FIX: what is correct RW with 802.11? */

	if (send(hapd->iapp_packet_sock, &msg, sizeof(msg), 0) < 0)
		perror("send[L2 Update]");
}


void iapp_new_station(hostapd *hapd, struct sta_info *sta)
{
	struct ieee80211_mgmt *assoc;

	iapp_send_add(hapd, sta);
	iapp_send_layer2_update(hapd, sta);

	assoc = sta->last_assoc_req;
	if (assoc && WLAN_FC_GET_STYPE(le_to_host16(assoc->frame_control)) ==
	    WLAN_FC_STYPE_REASSOC_REQ) {
		/* Send IAPP-MOVE to old AP */
	}
}


static void iapp_process_add_notify(hostapd *hapd, struct sockaddr_in *from,
				    struct iapp_hdr *hdr, int len)
{
	struct iapp_add_notify *add = (struct iapp_add_notify *) (hdr + 1);
	struct sta_info *sta;

	if (len != sizeof(*add)) {
		printf("Invalid IAPP-ADD packet length %d (expected %d)\n",
		       len, sizeof(*add));
		return;
	}

	printf("Received IAPP-ADD for STA " MACSTR " (seq# %d) from %s:%d\n",
	       MAC2STR(add->mac_addr), be_to_host16(add->seq_num),
	       inet_ntoa(from->sin_addr), ntohs(from->sin_port));

	sta = ap_get_sta(hapd, add->mac_addr);
	if (!sta)
		return;

	/* TODO: could use seq_num to try to determine whether last association
	 * to this AP is newer than the one advertised in IAPP-ADD. Although,
	 * this is not really a reliable verification. */

	printf("Removing STA " MACSTR " due to IAPP-ADD notification from "
	       "%s\n", MAC2STR(sta->addr), inet_ntoa(from->sin_addr));
	sta->flags &= ~(WLAN_STA_AUTH | WLAN_STA_ASSOC);
	remove_sta(hapd, sta);
}


static void iapp_receive_udp(int sock, void *eloop_ctx, void *sock_ctx)
{
	hostapd *hapd = eloop_ctx;
	int len, hlen;
	unsigned char buf[128];
	struct sockaddr_in from;
	socklen_t fromlen;
	struct iapp_hdr *hdr;

	/* Handle incoming IAPP frames (over UDP/IP) */

	fromlen = sizeof(from);
	len = recvfrom(hapd->iapp_udp_sock, buf, sizeof(buf), 0,
		       (struct sockaddr *) &from, &fromlen);
	if (len < 0)
		perror("recvfrom");

	if (from.sin_addr.s_addr == hapd->iapp_own.s_addr)
		return; /* ignore own IAPP messages */

	HOSTAPD_DEBUG(HOSTAPD_DEBUG_MINIMAL,
		      "Received %d byte IAPP frame from %s\n",
		      len, inet_ntoa(from.sin_addr));

	if (len < sizeof(*hdr)) {
		printf("Too short IAPP frame (len=%d)\n", len);
		return;
	}

	hdr = (struct iapp_hdr *) buf;
	hlen = be_to_host16(hdr->length);
	HOSTAPD_DEBUG(HOSTAPD_DEBUG_MINIMAL,
		      "IAPP: version=%d command=%d id=%d len=%d\n",
		      hdr->version, hdr->command,
		      be_to_host16(hdr->identifier), hlen);
	if (hlen > len) {
		printf("Underflow IAPP frame (hlen=%d len=%d)\n", hlen, len);
		return;
	}
	if (hlen < len) {
		printf("Ignoring %d extra bytes from IAPP frame\n",
		       len - hlen);
		len = hlen;
	}

	if (hdr->command == IAPP_CMD_ADD_notify)
		iapp_process_add_notify(hapd, &from, hdr, hlen - sizeof(*hdr));
	else
		printf("Unknown IAPP command %d\n", hdr->command);
}


int iapp_init(hostapd *hapd)
{
	struct ifreq ifr;
	struct sockaddr_ll addr;
	int ifindex, one;
	struct sockaddr_in *paddr, uaddr;

	/* TODO:
	 * open socket for sending and receiving IAPP frames over TCP
	 */

	hapd->iapp_udp_sock = socket(PF_INET, SOCK_DGRAM, 0);
	if (hapd->iapp_udp_sock < 0) {
		perror("socket[PF_INET,SOCK_DGRAM]");
		return -1;
	}

	memset(&ifr, 0, sizeof(ifr));
	strncpy(ifr.ifr_name, hapd->conf->iapp_iface, sizeof(ifr.ifr_name));
	if (ioctl(hapd->iapp_udp_sock, SIOCGIFINDEX, &ifr) != 0) {
		perror("ioctl(SIOCGIFINDEX)");
		return -1;
	}

	ifindex = ifr.ifr_ifindex;
	/*
	if (ioctl(hapd->iapp_udp_sock, SIOCGIFADDR, &ifr) != 0) {
		perror("ioctl(SIOCGIFADDR)");
		return -1;
	}
	paddr = (struct sockaddr_in *) &ifr.ifr_addr;
	if (paddr->sin_family != AF_INET) {
		printf("Invalid address family %i (SIOCGIFADDR)\n",
		       paddr->sin_family);
		return -1;
	}
	hapd->iapp_own.s_addr = paddr->sin_addr.s_addr;

	if (ioctl(hapd->iapp_udp_sock, SIOCGIFBRDADDR, &ifr) != 0) {
		perror("ioctl(SIOCGIFBRDADDR)");
		return -1;
	}
	paddr = (struct sockaddr_in *) &ifr.ifr_addr;
	if (paddr->sin_family != AF_INET) {
		printf("Invalid address family %i (SIOCGIFBRDADDR)\n",
		       paddr->sin_family);
		return -1;
	}
	hapd->iapp_broadcast.s_addr = paddr->sin_addr.s_addr;

	one = 1;
	if (setsockopt(hapd->iapp_udp_sock, SOL_SOCKET, SO_BROADCAST,
		       (char *) &one, sizeof(one)) < 0) {
		perror("setsockopt[SOL_SOCKET,SO_BROADCAST]");
		return -1;
	}

	memset(&uaddr, 0, sizeof(uaddr));
	uaddr.sin_family = AF_INET;
	uaddr.sin_port = htons(IAPP_PORT);
	if (bind(hapd->iapp_udp_sock, (struct sockaddr *) &uaddr,
		 sizeof(uaddr)) < 0) {
		perror("bind[UDP]");
		return -1;
	}

*/
	hapd->iapp_packet_sock = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
	if (hapd->iapp_packet_sock < 0) {
		perror("socket[PF_PACKET,SOCK_RAW]");
		return -1;
	}

	memset(&addr, 0, sizeof(addr));
	addr.sll_family = AF_PACKET;
	addr.sll_ifindex = ifindex;
	if (bind(hapd->iapp_packet_sock, (struct sockaddr *) &addr,
		 sizeof(addr)) < 0) {
		perror("bind[PACKET]");
		return -1;
	}

	if (eloop_register_read_sock(hapd->iapp_udp_sock, iapp_receive_udp,
				     NULL)) {
		printf("Could not register read socket for IAPP.\n");
		return -1;
	}

	printf("IEEE 802.11f (IAPP) using interface %s and broadcast address "
	       "%s\n", hapd->conf->iapp_iface,
	       inet_ntoa(hapd->iapp_broadcast));

	return 0;
}


void iapp_deinit(hostapd *hapd)
{
	if (hapd->iapp_udp_sock >= 0)
		close(hapd->iapp_udp_sock);
	if (hapd->iapp_packet_sock >= 0)
		close(hapd->iapp_packet_sock);
}
-------------- next part --------------




More information about the Hostap mailing list