Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 5 Aug 2003 16:18:26 -0700 (PDT)
From:      Sam Leffler <sam@FreeBSD.org>
To:        Perforce Change Reviews <perforce@freebsd.org>
Subject:   PERFORCE change 35571 for review
Message-ID:  <200308052318.h75NIQ14073824@repoman.freebsd.org>

next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=35571

Change 35571 by sam@sam_ebb on 2003/08/05 16:18:18

	Checkpoint fast forwarding locking: we use a mutex per hash bucket.
	Need to revisit fast path processing to see if we can come up with
	a lock free scheme.
	
	Note this code is untested.

Affected files ...

.. //depot/projects/netperf/sys/netinet/ip_flow.c#2 edit
.. //depot/projects/netperf/sys/netinet/ip_flow.h#2 edit

Differences ...

==== //depot/projects/netperf/sys/netinet/ip_flow.c#2 (text+ko) ====

@@ -38,8 +38,10 @@
 
 #include <sys/param.h>
 #include <sys/systm.h>
+#include <sys/lock.h>
 #include <sys/malloc.h>
 #include <sys/mbuf.h>
+#include <sys/mutex.h>
 #include <sys/protosw.h>
 #include <sys/socket.h>
 #include <sys/kernel.h>
@@ -59,10 +61,31 @@
 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
 #define IPFLOW_HASHBITS		6	/* should not be a multiple of 8 */
 #define	IPFLOW_HASHSIZE		(1 << IPFLOW_HASHBITS)
-static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE];
+#if IPFLOW_HASHSIZE > 255
+#error "make ipf_hash larger"
+#endif
+static struct ipflow_head ipflows[IPFLOW_HASHSIZE];
 static int ipflow_inuse;
 #define	IPFLOW_MAX		256
 
+/*
+ * Each flow list has a lock that guards updates to the list and to
+ * all entries on the list.  Flow entries hold the hash index for
+ * finding the head of the list so the lock can be found quickly.
+ *
+ * ipflow_inuse holds a count of the number of flow entries present.
+ * This is used to bound the size of the table.  When IPFLOW_MAX entries
+ * are present and an additional entry is needed one is chosen for
+ * replacement.  We could use atomic ops for this counter but having it
+ * inconsistent doesn't appear to be a problem.
+ */
+#define	IPFLOW_HEAD_LOCK(_ipfh)		mtx_lock(&(_ipfh)->ipfh_mtx)
+#define	IPFLOW_HEAD_UNLOCK(_ipfh)	mtx_unlock(&(_ipfh)->ipfh_mtx)
+#define	IPFLOW_LOCK(_ipf) \
+	IPFLOW_HEAD_LOCK(&ipflows[(_ipf)->ipf_hash])
+#define	IPFLOW_UNLOCK(_ipf) \
+	IPFLOW_HEAD_UNLOCK(&ipflows[(_ipf)->ipf_hash])
+
 static int ipflow_active = 0;
 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
     &ipflow_active, 0, "Enable flow-based IP forwarding");
@@ -70,10 +93,7 @@
 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
 
 static unsigned
-ipflow_hash(
-	struct in_addr dst,
-	struct in_addr src,
-	unsigned tos)
+ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
 {
 	unsigned hash = tos;
 	int idx;
@@ -83,28 +103,30 @@
 }
 
 static struct ipflow *
-ipflow_lookup(
-	const struct ip *ip)
+ipflow_lookup(const struct ip *ip)
 {
 	unsigned hash;
+	struct ipflow_head *head;
 	struct ipflow *ipf;
 
 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
+	head = &ipflows[hash];
 
-	ipf = LIST_FIRST(&ipflows[hash]);
-	while (ipf != NULL) {
+	IPFLOW_HEAD_LOCK(head);
+	LIST_FOREACH(ipf, &head->ipfh_head, ipf_next) {
 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
-		    && ip->ip_tos == ipf->ipf_tos)
-			break;
-		ipf = LIST_NEXT(ipf, ipf_next);
+		    && ip->ip_tos == ipf->ipf_tos) {
+			/* NB: return head locked */
+			return ipf;
+		}
 	}
-	return ipf;
+	IPFLOW_HEAD_UNLOCK(head);
+	return NULL;
 }
 
 int
-ipflow_fastforward(
-	struct mbuf *m)
+ipflow_fastforward(struct mbuf *m)
 {
 	struct ip *ip;
 	struct ipflow *ipf;
@@ -134,14 +156,18 @@
 	 * Route and interface still up?
 	 */
 	rt = ipf->ipf_ro.ro_rt;
-	if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0)
+	if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0) {
+		IPFLOW_UNLOCK(ipf);
 		return 0;
+	}
 
 	/*
 	 * Packet size OK?  TTL?
 	 */
-	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
+	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC) {
+		IPFLOW_UNLOCK(ipf);
 		return 0;
+	}
 
 	/*
 	 * Everything checks out and so we can forward this packet.
@@ -170,12 +196,12 @@
 		else
 			ipf->ipf_errors++;
 	}
+	IPFLOW_UNLOCK(ipf);
 	return 1;
 }
-
+
 static void
-ipflow_addstats(
-	struct ipflow *ipf)
+ipflow_addstats(struct ipflow *ipf)
 {
 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
@@ -183,36 +209,21 @@
 	ipstat.ips_fastforward += ipf->ipf_uses;
 }
 
-static void
-ipflow_free(
-	struct ipflow *ipf)
-{
-	int s;
-	/*
-	 * Remove the flow from the hash table (at elevated IPL).
-	 * Once it's off the list, we can deal with it at normal
-	 * network IPL.
-	 */
-	s = splimp();
-	LIST_REMOVE(ipf, ipf_next);
-	splx(s);
-	ipflow_addstats(ipf);
-	RTFREE(ipf->ipf_ro.ro_rt);
-	ipflow_inuse--;
-	free(ipf, M_IPFLOW);
-}
-
+/*
+ * XXX the locking here makes reaping an entry very expensive...
+ */
 static struct ipflow *
-ipflow_reap(
-	void)
+ipflow_reap(void)
 {
-	struct ipflow *ipf, *maybe_ipf = NULL;
+	struct ipflow *victim = NULL;
+	struct ipflow *ipf;
 	int idx;
-	int s;
 
 	for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
-		ipf = LIST_FIRST(&ipflows[idx]);
-		while (ipf != NULL) {
+		struct ipflow_head *head = &ipflows[idx];
+
+		IPFLOW_HEAD_LOCK(head);
+		LIST_FOREACH(ipf, &head->ipfh_head, ipf_next) {
 			/*
 			 * If this no longer points to a valid route
 			 * reclaim it.
@@ -224,38 +235,58 @@
 			 * or has had the least uses in the last 1.5 
 			 * intervals.
 			 */
-			if (maybe_ipf == NULL
-			    || ipf->ipf_timer < maybe_ipf->ipf_timer
-			    || (ipf->ipf_timer == maybe_ipf->ipf_timer
+			if (victim == NULL)
+				victim = ipf;
+			else if (ipf->ipf_timer < victim->ipf_timer
+			    || (ipf->ipf_timer == victim->ipf_timer
 				&& ipf->ipf_last_uses + ipf->ipf_uses <
-				      maybe_ipf->ipf_last_uses +
-					maybe_ipf->ipf_uses))
-				maybe_ipf = ipf;
-			ipf = LIST_NEXT(ipf, ipf_next);
+				    victim->ipf_last_uses + victim->ipf_uses)) {
+				if (victim->ipf_hash != ipf->ipf_hash)
+					IPFLOW_UNLOCK(victim);
+				victim = ipf;
+			}
 		}
+		if (victim && victim->ipf_hash != idx)
+			IPFLOW_HEAD_UNLOCK(head);
 	}
-	ipf = maybe_ipf;
+	ipf = victim;
     done:
 	/*
 	 * Remove the entry from the flow table.
 	 */
-	s = splimp();
 	LIST_REMOVE(ipf, ipf_next);
-	splx(s);
+	IPFLOW_UNLOCK(ipf);
+
 	ipflow_addstats(ipf);
 	RTFREE(ipf->ipf_ro.ro_rt);
 	return ipf;
 }
 
+static void
+ipflow_free(struct ipflow *ipf)
+{
+	/*
+	 * Remove the flow from the hash table.
+	 */
+	LIST_REMOVE(ipf, ipf_next);
+
+	ipflow_addstats(ipf);
+	RTFREE(ipf->ipf_ro.ro_rt);
+	ipflow_inuse--;
+	free(ipf, M_IPFLOW);
+}
+
 void
-ipflow_slowtimo(
-	void)
+ipflow_slowtimo(void)
 {
 	struct ipflow *ipf;
 	int idx;
 
 	for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
-		ipf = LIST_FIRST(&ipflows[idx]);
+		struct ipflow_head *head = &ipflows[idx];
+
+		IPFLOW_HEAD_LOCK(head);
+		ipf = LIST_FIRST(&head->ipfh_head);
 		while (ipf != NULL) {
 			struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 			if (--ipf->ipf_timer == 0) {
@@ -269,18 +300,15 @@
 			}
 			ipf = next_ipf;
 		}
+		IPFLOW_HEAD_UNLOCK(head);
 	}
 }
 
 void
-ipflow_create(
-	const struct route *ro,
-	struct mbuf *m)
+ipflow_create(const struct route *ro, struct mbuf *m)
 {
 	const struct ip *const ip = mtod(m, struct ip *);
 	struct ipflow *ipf;
-	unsigned hash;
-	int s;
 
 	/*
 	 * Don't create cache entries for ICMP messages.
@@ -304,12 +332,18 @@
 			ipflow_inuse++;
 		}
 		bzero((caddr_t) ipf, sizeof(*ipf));
+
+		ipf->ipf_hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
+		ipf->ipf_dst = ip->ip_dst;
+		ipf->ipf_src = ip->ip_src;
+		ipf->ipf_tos = ip->ip_tos;
+
+		IPFLOW_LOCK(ipf);
 	} else {
-		s = splimp();
 		LIST_REMOVE(ipf, ipf_next);
-		splx(s);
-		ipflow_addstats(ipf);
-		RTFREE(ipf->ipf_ro.ro_rt);
+
+		ipflow_addstats(ipf);		/* add stats to old route */
+		RTFREE(ipf->ipf_ro.ro_rt);	/* clear reference */
 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
 		ipf->ipf_errors = ipf->ipf_dropped = 0;
 	}
@@ -318,16 +352,26 @@
 	 * Fill in the updated information.
 	 */
 	ipf->ipf_ro = *ro;
+	RT_LOCK(ro->ro_rt);
 	ro->ro_rt->rt_refcnt++;
-	ipf->ipf_dst = ip->ip_dst;
-	ipf->ipf_src = ip->ip_src;
-	ipf->ipf_tos = ip->ip_tos;
+	RT_UNLOCK(ro->ro_rt);
 	ipf->ipf_timer = IPFLOW_TIMER;
 	/*
 	 * Insert into the approriate bucket of the flow table.
 	 */
-	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
-	s = splimp();
-	LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
-	splx(s);
+	LIST_INSERT_HEAD(&ipflows[ipf->ipf_hash].ipfh_head, ipf, ipf_next);
+	IPFLOW_UNLOCK(ipf);
+}
+
+static void
+ipflow_init(void)
+{
+	int idx;
+
+	for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
+		struct ipflow_head *head = &ipflows[idx];
+		LIST_INIT(&head->ipfh_head);
+		mtx_init(&head->ipfh_mtx, "ipflow list head", NULL, MTX_DEF);
+	}
 }
+SYSINIT(ipflow, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);

==== //depot/projects/netperf/sys/netinet/ip_flow.h#2 (text+ko) ====

@@ -44,6 +44,8 @@
 	struct in_addr ipf_dst;		/* destination address */
 	struct in_addr ipf_src;		/* source address */
 
+	/* NB: this assumes the size of the list head hash table is <=256 */
+	u_int8_t ipf_hash;		/* index in list head table */
 	u_int8_t ipf_tos;		/* type-of-service */
 	struct route ipf_ro;		/* associated route entry */
 	u_long ipf_uses;		/* number of uses in this period */
@@ -54,4 +56,9 @@
 	u_long ipf_last_uses;		/* number of uses in last period */
 };
 
+struct ipflow_head {
+	LIST_HEAD(ipflowhead, ipflow) ipfh_head;
+	struct mtx ipfh_mtx;
+};
+
 #endif



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200308052318.h75NIQ14073824>