[v5,12/16] hostapd: FILS Extended Key ID support
diff mbox series

Message ID 20190825163521.22625-13-alexander@wetzel-home.de
State Superseded
Headers show
  • Support seamless PTK rekeys with Extended Key ID
Related show

Commit Message

Alexander Wetzel Aug. 25, 2019, 4:35 p.m. UTC
IEEE 802.11ai - 2016 is missing any instructions how to handle it in
combination with Extended Key ID. There are two obvious ways:

 1) FILS can only use keyid 0 and the STAs decide on rekey if they can
    use Extended Key ID or not.

 2) FILS also checks if Extended Key ID can be used by both STAs and
    adds the KeyID KDE when it's being used in the initial handshake.

The later seems to be closer to the intent of 802.11ai and since there
are no other implementations for Extended Key ID to consider this patch
implements option 2) for now.

Signed-off-by: Alexander Wetzel <alexander@wetzel-home.de>

Now this is a very free interpretation of how to handle Extended Key ID
in combination with FILS. Technically it's the same issue as we have for
FT, so I'm using the same (arguable) solution here:

We bypass the 4-way handshake and Extended Key ID is therefore mostly
irrelevant. Neither FILS nor FT discuss Extended Key ID at all but both
have a mechanism to hand over/get the GTK ID. Which of course can also
pass over the (unicast) KeyID required for Extended Key ID support...

Now the new patch series is rigorously sticking to the key install mode
used at the initial connect: When either the AP or the STA tries to use
anything else than for the connect we kill the connection. By also
adding the KeyID to the KDEs these checks work basically out of the box
and the Extended Key ID flag in the RSN capabilities serves a purpose.

Alternatively we could relax the checks and accept that we either still
set the Extended Key ID bit in RSN but just assume the keyid is always
zero for FT and FILS or even drop the bit in the RSN capabilities and
relax the sanity checks for FILS and FT accordingly.

Since not Extended Key ID capable STAs won't care either way and
there are zero implementations for Extended Key ID we have to stay
compatible with I decided to first try what I consider the cleanest way.

Therefore Unicast KeyIDs have been added to the frames transporting also
the GTK ID. Based on the feedback we either keep or change it.

 src/ap/wpa_auth.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff mbox series

diff --git a/src/ap/wpa_auth.c b/src/ap/wpa_auth.c
index 0213e97c2..af3b855fb 100644
--- a/src/ap/wpa_auth.c
+++ b/src/ap/wpa_auth.c
@@ -2692,6 +2692,15 @@  static struct wpabuf * fils_prepare_plainbuf(struct wpa_state_machine *sm,
 	wpabuf_put_u8(plain, WLAN_EID_EXT_KEY_DELIVERY);
 	wpa_auth_get_seqnum(sm->wpa_auth, NULL, gsm->GN,
 			    wpabuf_put(plain, WPA_KEY_RSC_LEN));
+	hdr[1] = 0;
+	if (sm->use_extended_key_id) {
+		hdr[0] = sm->keyidx_active & 0x01;
+		tmp = wpabuf_put(plain, 0);
+		tmp2 = wpa_add_kde(tmp, RSN_KEY_DATA_KEYID, hdr, 2, NULL, 0);
+		wpabuf_put(plain, tmp2 - tmp);
+	}
 	/* GTK KDE */
 	gtk = gsm->GTK[gsm->GN - 1];
 	gtk_len = gsm->GTK_len;
@@ -2708,7 +2717,6 @@  static struct wpabuf * fils_prepare_plainbuf(struct wpa_state_machine *sm,
 		gtk = dummy_gtk;
 	hdr[0] = gsm->GN & 0x03;
-	hdr[1] = 0;
 	tmp = wpabuf_put(plain, 0);
 	tmp2 = wpa_add_kde(tmp, RSN_KEY_DATA_GROUPKEY, hdr, 2,
 			   gtk, gtk_len);
@@ -2763,7 +2771,7 @@  int fils_set_tk(struct wpa_state_machine *sm)
 	klen = wpa_cipher_key_len(sm->pairwise);
 	wpa_printf(MSG_DEBUG, "FILS: Configure TK to the driver");
-	if (wpa_auth_set_key(sm->wpa_auth, 0, alg, sm->addr, 0,
+	if (wpa_auth_set_key(sm->wpa_auth, 0, alg, sm->addr, sm->keyidx_active,
 			     sm->PTK.tk, klen, KEY_TYPE_PAIRWISE)) {
 		wpa_printf(MSG_DEBUG, "FILS: Failed to set TK to the driver");
 		return -1;