[RFC PATCH] xfrm: Accept XFRM_STATE_AF_UNSPEC SAs on IPv4/IPv6 only hosts

[Martin Willi <martin@strongswan.org>]

Installing SAs using the XFRM_STATE_AF_UNSPEC fails on hosts with
support for one address family only. This patch accepts such SAs, even
if the processing of not supported packets will fail.

Signed-off-by: Martin Willi <martin@strongswan.org>

---
For me the meaning of that flag is not really clear. If it means
"I-want-to-run-v6-in-v4-or-vice-versa" then the current behavior
(reject SA) is probably ok.
However, in my understanding ("accept-any-inner-address-family"), the
kernel should accept such SAs. Otherwise userspace has to query the
supported address families.

diff --git a/net/xfrm/xfrm_state.c b/net/xfrm/xfrm_state.c
index 508337f..456782b 100644
--- a/net/xfrm/xfrm_state.c
+++ b/net/xfrm/xfrm_state.c
@@ -2032,8 +2032,9 @@ int xfrm_init_state(struct xfrm_state *x)
 		x->inner_mode = inner_mode;
 	} else {
 		struct xfrm_mode *inner_mode_iaf;
+		int iafamily = AF_INET;
 
-		inner_mode = xfrm_get_mode(x->props.mode, AF_INET);
+		inner_mode = xfrm_get_mode(x->props.mode, x->props.family);
 		if (inner_mode == NULL)
 			goto error;
 
@@ -2041,22 +2042,17 @@ int xfrm_init_state(struct xfrm_state *x)
 			xfrm_put_mode(inner_mode);
 			goto error;
 		}
+		x->inner_mode = inner_mode;
 
-		inner_mode_iaf = xfrm_get_mode(x->props.mode, AF_INET6);
-		if (inner_mode_iaf == NULL)
-			goto error;
+		if (x->props.family == AF_INET)
+			iafamily = AF_INET6;
 
-		if (!(inner_mode_iaf->flags & XFRM_MODE_FLAG_TUNNEL)) {
-			xfrm_put_mode(inner_mode_iaf);
-			goto error;
-		}
-
-		if (x->props.family == AF_INET) {
-			x->inner_mode = inner_mode;
-			x->inner_mode_iaf = inner_mode_iaf;
-		} else {
-			x->inner_mode = inner_mode_iaf;
-			x->inner_mode_iaf = inner_mode;
+		inner_mode_iaf = xfrm_get_mode(x->props.mode, iafamily);
+		if (inner_mode_iaf) {
+			if (inner_mode_iaf->flags & XFRM_MODE_FLAG_TUNNEL)
+				x->inner_mode_iaf = inner_mode_iaf;
+			else
+				xfrm_put_mode(inner_mode_iaf);
 		}
 	}