SMP locking for radix nodes.

5 files changed, 66 insertions, 17 deletions

diff --git a/sys/net/if.c b/sys/net/if.c
index 42f2590..1f58605 100644
--- a/sys/net/if.c
+++ b/sys/net/if.c
@@ -533,7 +533,9 @@ if_detach(ifp)
 	for (i = 1; i <= AF_MAX; i++) {
 		if ((rnh = rt_tables[i]) == NULL)
 			continue;
+		RADIX_NODE_HEAD_LOCK(rnh);
 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
+		RADIX_NODE_HEAD_UNLOCK(rnh);
 	}
 
 	/* Announce that the interface is gone. */
diff --git a/sys/net/radix.c b/sys/net/radix.c
index e1742f3..f6cdadb 100644
--- a/sys/net/radix.c
+++ b/sys/net/radix.c
@@ -1024,6 +1024,7 @@ rn_inithead(head, off)
 	if (rnh == 0)
 		return (0);
 	Bzero(rnh, sizeof (*rnh));
+	RADIX_NODE_HEAD_LOCK_INIT(rnh);
 	*head = rnh;
 	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
 	ttt = rnh->rnh_nodes + 2;
diff --git a/sys/net/radix.h b/sys/net/radix.h
index 3a65585..d9937ca 100644
--- a/sys/net/radix.h
+++ b/sys/net/radix.h
@@ -37,6 +37,9 @@
 #ifndef _RADIX_H_
 #define	_RADIX_H_
 
+#include <sys/lock.h>
+#include <sys/mutex.h>
+
 #ifdef MALLOC_DECLARE
 MALLOC_DECLARE(M_RTABLE);
 #endif
@@ -138,6 +141,7 @@ struct radix_node_head {
 	void	(*rnh_close)	/* do something when the last ref drops */
 		(struct radix_node *rn, struct radix_node_head *head);
 	struct	radix_node rnh_nodes[3];	/* empty tree for common case */
+	struct	mtx rnh_mtx;
 };
 
 #ifndef _KERNEL
@@ -152,6 +156,13 @@ struct radix_node_head {
 #define Bzero(p, n) bzero((caddr_t)(p), (unsigned)(n));
 #define R_Malloc(p, t, n) (p = (t) malloc((unsigned long)(n), M_RTABLE, M_NOWAIT))
 #define Free(p) free((caddr_t)p, M_RTABLE);
+
+
+#define RADIX_NODE_HEAD_LOCK_INIT(rnh)	\
+    mtx_init(&(rnh)->rnh_mtx, "radix node head", NULL, MTX_DEF | MTX_RECURSE)
+#define RADIX_NODE_HEAD_LOCK(rnh)	mtx_lock(&(rnh)->rnh_mtx)
+#define RADIX_NODE_HEAD_UNLOCK(rnh)	mtx_unlock(&(rnh)->rnh_mtx)
+#define RADIX_NODE_HEAD_DESTROY(rnh)	mtx_destroy(&(rnh)->rnh_mtx)
 #endif /* _KERNEL */
 
 void	 rn_init(void);
diff --git a/sys/net/route.c b/sys/net/route.c
index 1727d5b..260f375 100644
--- a/sys/net/route.c
+++ b/sys/net/route.c
@@ -132,7 +132,12 @@ rtalloc1(dst, report, ignflags)
 	/*
 	 * Look up the address in the table for that Address Family
 	 */
-	if (rnh && (rn = rnh->rnh_matchaddr((caddr_t)dst, rnh)) &&
+	if (rnh == NULL) {
+		rtstat.rts_unreach++;
+		goto miss2;
+	}
+	RADIX_NODE_HEAD_LOCK(rnh);
+	if ((rn = rnh->rnh_matchaddr((caddr_t)dst, rnh)) &&
 	    ((rn->rn_flags & RNF_ROOT) == 0)) {
 		/*
 		 * If we find it and it's not the root node, then
@@ -178,6 +183,7 @@ rtalloc1(dst, report, ignflags)
 			rt_missmsg(RTM_ADD, &info, rt->rt_flags, 0);
 		} else
 			rt->rt_refcnt++;
+		RADIX_NODE_HEAD_UNLOCK(rnh);
 	} else {
 		/*
 		 * Either we hit the root or couldn't find any match,
@@ -185,7 +191,9 @@ rtalloc1(dst, report, ignflags)
 		 * "caint get there frm here"
 		 */
 		rtstat.rts_unreach++;
-	miss:	if (report) {
+	miss:
+		RADIX_NODE_HEAD_UNLOCK(rnh);
+	miss2:	if (report) {
 			/*
 			 * If required, report the failure to the supervising
 			 * Authorities.
@@ -211,8 +219,7 @@ rtfree(rt)
 	/*
 	 * find the tree for that address family
 	 */
-	register struct radix_node_head *rnh =
-		rt_tables[rt_key(rt)->sa_family];
+	struct radix_node_head *rnh = rt_tables[rt_key(rt)->sa_family];
 
 	if (rt == 0 || rnh == 0)
 		panic("rtfree");
@@ -222,7 +229,7 @@ rtfree(rt)
 	 * and there is a close function defined, call the close function
 	 */
 	rt->rt_refcnt--;
-	if(rnh->rnh_close && rt->rt_refcnt == 0) {
+	if (rnh->rnh_close && rt->rt_refcnt == 0) {
 		rnh->rnh_close((struct radix_node *)rt, rnh);
 	}
 
@@ -270,7 +277,8 @@ rtfree(rt)
 	}
 }
 
-#define	equal(a1, a2) (bcmp((caddr_t)(a1), (caddr_t)(a2), (a1)->sa_len) == 0)
+/* compare two sockaddr structures */
+#define	sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
 
 /*
  * Force a routing table entry to the specified
@@ -306,7 +314,7 @@ rtredirect(dst, gateway, netmask, flags, src, rtp)
 	 * going down recently.
 	 */
 	if (!(flags & RTF_DONE) && rt &&
-	     (!equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
+	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
 		error = EINVAL;
 	else if (ifa_ifwithaddr(gateway))
 		error = EHOSTUNREACH;
@@ -538,6 +546,7 @@ rtrequest1(req, info, ret_nrt)
 	 */
 	if ((rnh = rt_tables[dst->sa_family]) == 0)
 		senderr(EAFNOSUPPORT);
+	RADIX_NODE_HEAD_LOCK(rnh);
 	/*
 	 * If we are adding a host route then we don't want to put
 	 * a netmask in the tree, nor do we want to clone it.
@@ -763,6 +772,7 @@ rtrequest1(req, info, ret_nrt)
 		error = EOPNOTSUPP;
 	}
 bad:
+	RADIX_NODE_HEAD_UNLOCK(rnh);
 	splx(s);
 	return (error);
 #undef dst
@@ -1002,8 +1012,10 @@ rt_setgate(rt0, dst, gate)
 		struct rtfc_arg arg;
 		arg.rnh = rnh;
 		arg.rt0 = rt;
+		RADIX_NODE_HEAD_LOCK(rnh);
 		rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt),
 				       rt_fixchange, &arg);
+		RADIX_NODE_HEAD_UNLOCK(rnh);
 	}
 
 	return 0;
@@ -1079,11 +1091,16 @@ rtinit(ifa, cmd, flags)
 		 * Look up an rtentry that is in the routing tree and
 		 * contains the correct info.
 		 */
-		if ((rnh = rt_tables[dst->sa_family]) == NULL ||
-		    (rn = rnh->rnh_lookup(dst, netmask, rnh)) == NULL ||
+		if ((rnh = rt_tables[dst->sa_family]) == NULL)
+			goto bad;
+		RADIX_NODE_HEAD_LOCK(rnh);
+		error = ((rn = rnh->rnh_lookup(dst, netmask, rnh)) == NULL ||
 		    (rn->rn_flags & RNF_ROOT) ||
 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
-		    !equal(SA(rn->rn_key), dst)) {
+		    !sa_equal(SA(rn->rn_key), dst));
+		RADIX_NODE_HEAD_UNLOCK(rnh);
+		if (error) {
+bad:
 			if (m)
 				(void) m_free(m);
 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
diff --git a/sys/net/rtsock.c b/sys/net/rtsock.c
index 40f2ee1..0c441ef 100644
--- a/sys/net/rtsock.c
+++ b/sys/net/rtsock.c
@@ -367,11 +367,15 @@ route_output(m, so)
 	case RTM_LOCK:
 		if ((rnh = rt_tables[dst->sa_family]) == 0) {
 			senderr(EAFNOSUPPORT);
-		} else if ((rt = (struct rtentry *)
-				rnh->rnh_lookup(dst, netmask, rnh)) != NULL)
+		}
+		RADIX_NODE_HEAD_LOCK(rnh);
+		rt = (struct rtentry *) rnh->rnh_lookup(dst, netmask, rnh);
+		RADIX_NODE_HEAD_UNLOCK(rnh);
+		if (rt != NULL)
 			rt->rt_refcnt++;
 		else
 			senderr(ESRCH);
+
 		switch(rtm->rtm_type) {
 
 		case RTM_GET:
@@ -1024,11 +1028,25 @@ sysctl_rtsock(SYSCTL_HANDLER_ARGS)
 
 	case NET_RT_DUMP:
 	case NET_RT_FLAGS:
-		for (i = 1; i <= AF_MAX; i++)
-			if ((rnh = rt_tables[i]) && (af == 0 || af == i) &&
-			    (error = rnh->rnh_walktree(rnh,
-							sysctl_dumpentry, &w)))
-				break;
+		if (af != 0) {
+			if ((rnh = rt_tables[af]) != NULL) {
+				RADIX_NODE_HEAD_LOCK(rnh);
+			    	error = rnh->rnh_walktree(rnh,
+				    sysctl_dumpentry, &w);
+				RADIX_NODE_HEAD_UNLOCK(rnh);
+			} else
+				error = EAFNOSUPPORT;
+		} else {
+			for (i = 1; i <= AF_MAX; i++)
+				if ((rnh = rt_tables[i]) != NULL) {
+					RADIX_NODE_HEAD_LOCK(rnh);
+					error = rnh->rnh_walktree(rnh,
+					    sysctl_dumpentry, &w);
+					RADIX_NODE_HEAD_UNLOCK(rnh);
+					if (error)
+						break;
+				}
+		}
 		break;
 
 	case NET_RT_IFLIST: