Add a queue to hold packets while we await an ARP reply.

When a fast machine first brings up some non TCP networking program
it is quite possible that we will drop packets due to the fact that
only one packet can be held per ARP entry.  This leads to packets
being missed when a program starts or restarts if the ARP data is
not currently in the ARP cache.

This code adds a new sysctl, net.link.ether.inet.maxhold, which defines
a system wide maximum number of packets to be held in each ARP entry.
Up to maxhold packets are queued until an ARP reply is received or
the ARP times out.  The default setting is the old value of 1
which has been part of the BSD networking code since time
immemorial.

Expose the time we hold an incomplete ARP entry by adding
the sysctl net.link.ether.inet.wait, which defaults to 20
seconds, the value used when the new ARP code was added..

Reviewed by:	bz, rpaulo
MFC after: 3 weeks

4 files changed, 85 insertions, 22 deletions

diff --git a/sys/net/if_llatbl.c b/sys/net/if_llatbl.c
index a0fb071..3444c97 100644
--- a/sys/net/if_llatbl.c
+++ b/sys/net/if_llatbl.c
@@ -100,18 +100,34 @@ done:
  * This function is called by the timer functions
  * such as arptimer() and nd6_llinfo_timer(), and
  * the caller does the locking.
+ *
+ * Returns the number of held packets, if any, that were dropped.
  */
-void
+size_t
 llentry_free(struct llentry *lle)
 {
-	
+	size_t pkts_dropped;
+	struct mbuf *next;
+
+	pkts_dropped = 0;
 	LLE_WLOCK_ASSERT(lle);
 	LIST_REMOVE(lle, lle_next);
 
-	if (lle->la_hold != NULL)
+	while ((lle->la_numheld > 0) && (lle->la_hold != NULL)) {
+		next = lle->la_hold->m_nextpkt;
 		m_freem(lle->la_hold);
+		lle->la_hold = next;
+		lle->la_numheld--;
+		pkts_dropped++;
+	}
+
+	KASSERT(lle->la_numheld == 0, 
+		("%s: la_numheld %d > 0, pkts_droped %ld", __func__, 
+		 lle->la_numheld, pkts_dropped));
 
 	LLE_FREE_LOCKED(lle);
+
+	return (pkts_dropped);
 }
 
 /*
@@ -412,6 +428,7 @@ llatbl_lle_show(struct llentry_sa *la)
 	db_printf(" lle_tbl=%p\n", lle->lle_tbl);
 	db_printf(" lle_head=%p\n", lle->lle_head);
 	db_printf(" la_hold=%p\n", lle->la_hold);
+	db_printf(" la_numheld=%d\n", lle->la_numheld);
 	db_printf(" la_expire=%ju\n", (uintmax_t)lle->la_expire);
 	db_printf(" la_flags=0x%04x\n", lle->la_flags);
 	db_printf(" la_asked=%u\n", lle->la_asked);
diff --git a/sys/net/if_llatbl.h b/sys/net/if_llatbl.h
index a4d02ab..6c5ad18 100644
--- a/sys/net/if_llatbl.h
+++ b/sys/net/if_llatbl.h
@@ -58,6 +58,7 @@ struct llentry {
 	struct lltable		 *lle_tbl;
 	struct llentries	 *lle_head;
 	struct mbuf		 *la_hold;
+	int     		 la_numheld;  /* # of packets currently held */
 	time_t			 la_expire;
 	uint16_t		 la_flags;    
 	uint16_t		 la_asked;
@@ -191,7 +192,7 @@ void		lltable_drain(int);
 #endif
 int		lltable_sysctl_dumparp(int, struct sysctl_req *);
 
-void		llentry_free(struct llentry *);
+size_t		llentry_free(struct llentry *);
 int		llentry_update(struct llentry **, struct lltable *,
                        struct sockaddr_storage *, struct ifnet *);
 
diff --git a/sys/netinet/if_ether.c b/sys/netinet/if_ether.c
index 6d38388..ef59cdb 100644
--- a/sys/netinet/if_ether.c
+++ b/sys/netinet/if_ether.c
@@ -89,13 +89,16 @@ VNET_DEFINE(int, useloopback) = 1;	/* use loopback interface for
 static VNET_DEFINE(int, arp_proxyall) = 0;
 static VNET_DEFINE(int, arpt_down) = 20;      /* keep incomplete entries for
 					       * 20 seconds */
-static VNET_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
+VNET_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
+
+static VNET_DEFINE(int, arp_maxhold) = 1;
 
 #define	V_arpt_keep		VNET(arpt_keep)
 #define	V_arpt_down		VNET(arpt_down)
 #define	V_arp_maxtries		VNET(arp_maxtries)
 #define	V_arp_proxyall		VNET(arp_proxyall)
 #define	V_arpstat		VNET(arpstat)
+#define	V_arp_maxhold		VNET(arp_maxhold)
 
 SYSCTL_VNET_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
 	&VNET_NAME(arpt_keep), 0,
@@ -109,9 +112,15 @@ SYSCTL_VNET_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
 SYSCTL_VNET_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
 	&VNET_NAME(arp_proxyall), 0,
 	"Enable proxy ARP for all suitable requests");
+SYSCTL_VNET_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_RW,
+	&VNET_NAME(arpt_down), 0,
+	"Incomplete ARP entry lifetime in seconds");
 SYSCTL_VNET_STRUCT(_net_link_ether_arp, OID_AUTO, stats, CTLFLAG_RW,
 	&VNET_NAME(arpstat), arpstat,
 	"ARP statistics (struct arpstat, net/if_arp.h)");
+SYSCTL_VNET_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_RW,
+	&VNET_NAME(arp_maxhold), 0, 
+	"Number of packets to hold per ARP entry");
 
 static void	arp_init(void);
 void		arprequest(struct ifnet *,
@@ -162,6 +171,7 @@ arptimer(void *arg)
 {
 	struct ifnet *ifp;
 	struct llentry   *lle;
+	int pkts_dropped;
 
 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
 	lle = (struct llentry *)arg;
@@ -176,7 +186,8 @@ arptimer(void *arg)
 		    callout_active(&lle->la_timer)) {
 			callout_stop(&lle->la_timer);
 			LLE_REMREF(lle);
-			(void) llentry_free(lle);
+			pkts_dropped = llentry_free(lle);
+			ARPSTAT_ADD(dropped, pkts_dropped);
 			ARPSTAT_INC(timeouts);
 		} else {
 #ifdef DIAGNOSTIC
@@ -275,6 +286,8 @@ arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
 {
 	struct llentry *la = 0;
 	u_int flags = 0;
+	struct mbuf *curr = NULL;
+	struct mbuf *next = NULL;
 	int error, renew;
 
 	*lle = NULL;
@@ -348,15 +361,28 @@ retry:
 	}
 	/*
 	 * There is an arptab entry, but no ethernet address
-	 * response yet.  Replace the held mbuf with this
-	 * latest one.
+	 * response yet.  Add the mbuf to the list, dropping
+	 * the oldest packet if we have exceeded the system
+	 * setting.
 	 */
 	if (m != NULL) {
+		if (la->la_numheld >= V_arp_maxhold) {
+			if (la->la_hold != NULL) {
+				next = la->la_hold->m_nextpkt;
+				m_freem(la->la_hold);
+				la->la_hold = next;
+				la->la_numheld--;
+				ARPSTAT_INC(dropped);
+			}
+		} 
 		if (la->la_hold != NULL) {
-			m_freem(la->la_hold);
-			ARPSTAT_INC(dropped);
-		}
-		la->la_hold = m;
+			curr = la->la_hold;
+			while (curr->m_nextpkt != NULL)
+				curr = curr->m_nextpkt;
+			curr->m_nextpkt = m;
+		} else 
+			la->la_hold = m;
+		la->la_numheld++;
 		if (renew == 0 && (flags & LLE_EXCLUSIVE)) {
 			flags &= ~LLE_EXCLUSIVE;
 			LLE_DOWNGRADE(la);
@@ -483,7 +509,6 @@ in_arpinput(struct mbuf *m)
 	struct rtentry *rt;
 	struct ifaddr *ifa;
 	struct in_ifaddr *ia;
-	struct mbuf *hold;
 	struct sockaddr sa;
 	struct in_addr isaddr, itaddr, myaddr;
 	u_int8_t *enaddr = NULL;
@@ -698,15 +723,29 @@ match:
 		}
 		la->la_asked = 0;
 		la->la_preempt = V_arp_maxtries;
-		hold = la->la_hold;
-		if (hold != NULL) {
-			la->la_hold = NULL;
+		/* 
+		 * The packets are all freed within the call to the output
+		 * routine.
+		 *
+		 * NB: The lock MUST be released before the call to the
+		 * output routine.
+		 */
+		if (la->la_hold != NULL) {
+			struct mbuf *m_hold, *m_hold_next;
+
 			memcpy(&sa, L3_ADDR(la), sizeof(sa));
-		}
-		LLE_WUNLOCK(la);
-		if (hold != NULL)
-			(*ifp->if_output)(ifp, hold, &sa, NULL);
-	}
+			LLE_WUNLOCK(la);
+			for (m_hold = la->la_hold, la->la_hold = NULL;
+			     m_hold != NULL; m_hold = m_hold_next) {
+				m_hold_next = m_hold->m_nextpkt;
+				m_hold->m_nextpkt = NULL;
+				(*ifp->if_output)(ifp, m_hold, &sa, NULL);
+			}
+		} else
+			LLE_WUNLOCK(la);
+		la->la_hold = NULL;
+		la->la_numheld = 0;
+	} /* end of FIB loop */
 reply:
 	if (op != ARPOP_REQUEST)
 		goto drop;
diff --git a/sys/netinet/in.c b/sys/netinet/in.c
index b2ae8d9..2ec54e2 100644
--- a/sys/netinet/in.c
+++ b/sys/netinet/in.c
@@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$");
 
 #include <net/if.h>
 #include <net/if_var.h>
+#include <net/if_arp.h>
 #include <net/if_dl.h>
 #include <net/if_llatbl.h>
 #include <net/if_types.h>
@@ -89,6 +90,9 @@ SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, same_prefix_carp_only, CTLFLAG_RW,
 VNET_DECLARE(struct inpcbinfo, ripcbinfo);
 #define	V_ripcbinfo			VNET(ripcbinfo)
 
+VNET_DECLARE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
+#define	V_arpstat		VNET(arpstat)
+
 /*
  * Return 1 if an internet address is for a ``local'' host
  * (one to which we have a connection).  If subnetsarelocal
@@ -1363,6 +1367,7 @@ in_lltable_prefix_free(struct lltable *llt,
 	const struct sockaddr_in *msk = (const struct sockaddr_in *)mask;
 	struct llentry *lle, *next;
 	register int i;
+	size_t pkts_dropped;
 
 	for (i=0; i < LLTBL_HASHTBL_SIZE; i++) {
 		LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) {
@@ -1375,7 +1380,8 @@ in_lltable_prefix_free(struct lltable *llt,
 				LLE_WLOCK(lle);
 				if (canceled)
 					LLE_REMREF(lle);
-				llentry_free(lle);
+				pkts_dropped = llentry_free(lle);
+				ARPSTAT_ADD(dropped, pkts_dropped);
 			}
 		}
 	}