This is the file that actually implements the smarter behavior.

svn path=/head/; revision=4074

sys/netinet/in_rmx.c

@@ -0,0 +1,283 @@
+/*
+ * Copyright 1994, Massachusetts Institute of Technology.  All Rights Reserved.
+ *
+ * You may copy this file verbatim until I find the official 
+ * Institute boilerplate.
+ *
+ * $Id$
+ */
+
+/*
+ * This code does two things necessary for the enhanced TCP metrics to
+ * function in a useful manner:
+ *  1) It marks all non-host routes as `cloning', thus ensuring that
+ *     every actual reference to such a route actually gets turned
+ *     into a reference to a host route to the specific destination
+ *     requested.
+ *  2) When such routes lose all their references, it arranges for them
+ *     to be deleted in some random collection of circumstances, so that
+ *     a large quantity of stale routing data is not kept in kernel memory
+ *     indefinitely.  See in_rtqtimo() below for the exact mechanism.
+ *
+ * At least initially, we think that this should have lower overhead than
+ * using the existing `expire' mechanism and walking the radix tree
+ * periodically, deleting things as we go.  That method would be relatively
+ * easy to implement within the framework used here, and in the future
+ * we made code both ways, so that folks with large routing tables can use
+ * the external queue, and the majority with small routing tables can do
+ * the tree-walk.
+ */
+
+/*
+ * XXX - look for races
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/queue.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+#include <sys/mbuf.h>
+
+#include <net/if.h>
+#include <net/route.h>
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/in_var.h>
+
+/*
+ * The list of unreferenced IP routes...
+ */
+struct in_rtq {
+	TAILQ_ENTRY(in_rtq)	inr_entry;
+	struct rtentry		*inr_rt;
+	time_t			inr_whenadded;
+};
+
+TAILQ_HEAD(, in_rtq) inr_head;
+int inr_nelem = -1;
+
+/*
+ * Do what we need to do when inserting a route.
+ * Note that we don't automatically add the route to the queue, because
+ * our caller, rtrequest(), will immediately bump the refcount and give
+ * a the route to someone.
+ */
+static struct radix_node *
+in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
+	    struct radix_node *treenodes)
+{
+	struct rtentry *rt = (struct rtentry *)treenodes;
+	struct radix_node *rn;
+	struct in_rtq *inr;
+
+	/*
+	 * For IP, all non-host routes are automatically cloning.
+	 */
+	if(!(rt->rt_flags & RTF_HOST))
+		rt->rt_flags |= RTF_CLONING;
+
+	rn = rn_addroute(v_arg, n_arg, head, treenodes);
+	rt = (struct rtentry *)rn;
+	if(!rt
+	   || !(rt->rt_flags & RTF_HOST)
+	   || (rt->rt_flags & (RTF_LLINFO | RTF_STATIC | RTF_DYNAMIC)))
+		return rn;
+
+	return rn;
+}
+
+static struct radix_node *
+in_delroute(void *v_arg, void *netmask_arg, struct radix_node_head *head)
+{
+#ifdef DIAGNOSTIC
+	int nentries = 0;
+#endif
+	struct rtentry *rt = (struct rtentry *)rn_search(v_arg, 
+							 head->rnh_treetop);
+	if(rt && rt->rt_refcnt <= 0
+	   && ((rt->rt_flags & (RTF_HOST | RTF_LLINFO | RTF_STATIC))
+	       == RTF_HOST)) {
+		struct in_rtq *inr = inr_head.tqh_first;
+
+		while(inr) {
+			if(inr->inr_rt == rt) {
+				TAILQ_REMOVE(&inr_head, inr, inr_entry);
+				inr_nelem--;
+				FREE(inr, M_RTABLE);
+#ifdef DIAGNOSTIC
+				nentries++;
+#else
+				break;
+#endif
+			}
+			inr = inr->inr_entry.tqe_next;
+		}
+
+#ifdef DIAGNOSTIC
+		if(nentries != 1) {
+			log(LOG_DEBUG, "route %p had %d queue entries!\n",
+			    (void *)rt, nentries);
+		}
+#endif
+	}
+
+	return rn_delete(v_arg, netmask_arg, head);
+}
+
+/*
+ * Find something in the queue.
+ */
+static inline struct in_rtq *
+inr_findit(struct rtentry *rt)
+{
+	struct in_rtq *inr;
+
+#ifndef DIAGNOSTIC
+	if((rt->rt_flags & (RTF_HOST | RTF_LLINFO | RTF_STATIC)) != RTF_HOST)
+		return 0;
+#endif
+	inr = inr_head.tqh_first;
+
+	while(inr) {
+		if(inr->inr_rt == rt)
+			return inr;
+		inr = inr->inr_entry.tqe_next;
+	}
+
+	return 0;
+}
+	
+
+/*
+ * This code is the inverse of in_clsroute: on first reference, we remove the
+ * route from our queue (if it's in there)
+ */
+static struct radix_node *
+in_matroute(void *v_arg, struct radix_node_head *head)
+{
+	struct radix_node *rn = rn_match(v_arg, head);
+	struct rtentry *rt = (struct rtentry *)rn;
+
+	if(rt && rt->rt_refcnt == 0) { /* this is first reference */
+		struct in_rtq *inr = inr_findit(rt);
+		if(inr) {
+			TAILQ_REMOVE(&inr_head, inr, inr_entry);
+			inr_nelem--;
+			FREE(inr, M_RTABLE);
+		}
+	}
+	return rn;
+}
+
+/*
+ * On last reference drop, add the route to the queue so that it can be
+ * timed out.
+ */
+static void
+in_clsroute(struct radix_node *rn, struct radix_node_head *head)
+{
+	struct rtentry *rt = (struct rtentry *)rn;
+	struct in_rtq *inr;
+	
+	inr = inr_findit(rt);
+	if(inr) {
+		/*
+		 * In this case, we are probably in the process of deleting
+		 * this route; the code path in route.c results in rnh_close()
+		 * being called when during deletion of unreferenced routes.
+		 * In any case, don't allocate a new queue entry for this
+		 * route because there already is one.
+		 */
+		return;
+	}
+
+	MALLOC(inr, struct in_rtq *, sizeof *inr, M_RTABLE, M_NOWAIT);
+	if(!inr)		
+		return;		/* oops... no matter */
+
+	inr->inr_rt = rt;
+	inr->inr_whenadded = time.tv_sec;
+	TAILQ_INSERT_TAIL(&inr_head, inr, inr_entry);
+	inr_nelem++;
+}
+
+/*
+ * Get rid of everything in the queue, we are short on memory.
+ * Although this looks like an infinite loop, it really isn't;
+ * rtrequest() eventually calls in_delroute() which ends up deleting
+ * the node at the head (or so we hope).  This should be called from
+ * ip_drain().
+ */
+void
+in_rtqdrain(void)
+{
+	struct in_rtq *inr;
+
+	while(inr = inr_head.tqh_first) {
+		rtrequest(RTM_DELETE, rt_key(inr->inr_rt), 
+			  inr->inr_rt->rt_gateway, rt_mask(inr->inr_rt),
+			  RTF_HOST, 0);
+		/* KILL! KILL! KILL! */
+	}
+}
+
+#define RTQ_TIMEOUT	(60*hz)	/* run once a minute */
+#define RTQ_REALLYOLD	4*60*60	/* four hours is ``really old'' */
+#define RTQ_TOOMANY	128	/* > 128 routes is ``too many'' */
+
+/*
+ * Get rid of old routes.  We have two strategies here:
+ * 1) If there are more than RTQ_TOOMANY routes, delete half of them.
+ * 2) Delete all routes older than RTQ_REALLYOLD (the LRU nature of the
+ *    queue ensures that these are all at the front).
+ */
+static void
+in_rtqtimo(void *rock)
+{
+	int s = splnet();
+	struct in_rtq *inr = inr_head.tqh_first;
+	
+	if(inr_nelem > RTQ_TOOMANY) {
+		int ntodel = inr_nelem / 2;
+
+		for(; inr && ntodel; inr = inr_head.tqh_first, ntodel--) {
+			rtrequest(RTM_DELETE, rt_key(inr->inr_rt),
+				  inr->inr_rt->rt_gateway, 
+				  rt_mask(inr->inr_rt), RTF_HOST, 0);
+		}
+	}
+
+	while(inr && (time.tv_sec - inr->inr_whenadded) > RTQ_REALLYOLD) {
+		rtrequest(RTM_DELETE, rt_key(inr->inr_rt), 
+			  inr->inr_rt->rt_gateway, rt_mask(inr->inr_rt),
+			  RTF_HOST, 0);
+		inr = inr_head.tqh_first;
+	}
+
+	timeout(in_rtqtimo, (void *)0, RTQ_TIMEOUT);
+}
+
+/*
+ * Initialize our routing tree.
+ */
+int
+in_inithead(void **head, int off)
+{
+	struct radix_node_head *rnh;
+
+	if(!rn_inithead(head, off))
+		return 0;
+
+	rnh = *head;
+	rnh->rnh_addaddr = in_addroute;
+	rnh->rnh_deladdr = in_delroute;
+	rnh->rnh_matchaddr = in_matroute;
+	rnh->rnh_close = in_clsroute;
+	TAILQ_INIT(&inr_head);
+	inr_nelem = 0;
+	in_rtqtimo(0);		/* kick off timeout first time */
+	return 1;
+}
+