Module Name: src
Committed By: ozaki-r
Date: Mon Aug 31 08:05:20 UTC 2015
Modified Files:
src/sys/net: if_arp.h if_ethersubr.c if_llatbl.c if_llatbl.h
if_tokensubr.c
src/sys/netinet: if_arp.c in.c in_proto.c in_var.h
Log Message:
Replace ARP cache (llinfo) with lltable/llentry
Highlights of the change are:
- Use llentry instead of llinfo to manage ARP caches
- ARP specific data are stored in the hashed list
of an interface instead of the global list (llinfo_arp)
- Fine-grain locking on llentry
- arptimer (callout) per ARP cache
- the global timer callout with the big locks can be
removed (though softnet_lock is still required for now)
- net.inet.arp.prune is now obsoleted
- it was the interval of the global timer callout
- net.inet.arp.refresh is now obsoleted
- it was a parameter that prevents expiration of active caches
- Removed to simplify the timer logic, but we may be able to
restore the feature if really needed
Proposed on tech-kern and tech-net.
To generate a diff of this commit:
cvs rdiff -u -r1.29 -r1.30 src/sys/net/if_arp.h
cvs rdiff -u -r1.212 -r1.213 src/sys/net/if_ethersubr.c
cvs rdiff -u -r1.1 -r1.2 src/sys/net/if_llatbl.c src/sys/net/if_llatbl.h
cvs rdiff -u -r1.70 -r1.71 src/sys/net/if_tokensubr.c
cvs rdiff -u -r1.173 -r1.174 src/sys/netinet/if_arp.c
cvs rdiff -u -r1.158 -r1.159 src/sys/netinet/in.c
cvs rdiff -u -r1.113 -r1.114 src/sys/netinet/in_proto.c
cvs rdiff -u -r1.73 -r1.74 src/sys/netinet/in_var.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/net/if_arp.h
diff -u src/sys/net/if_arp.h:1.29 src/sys/net/if_arp.h:1.30
--- src/sys/net/if_arp.h:1.29 Tue Apr 15 15:17:54 2008
+++ src/sys/net/if_arp.h Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_arp.h,v 1.29 2008/04/15 15:17:54 thorpej Exp $ */
+/* $NetBSD: if_arp.h,v 1.30 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (c) 1986, 1993
@@ -127,4 +127,6 @@ struct arpreq {
#define ARP_NSTATS 23
+void arp_stat_add(int, uint64_t);
+
#endif /* !_NET_IF_ARP_H_ */
Index: src/sys/net/if_ethersubr.c
diff -u src/sys/net/if_ethersubr.c:1.212 src/sys/net/if_ethersubr.c:1.213
--- src/sys/net/if_ethersubr.c:1.212 Mon Aug 24 22:21:26 2015
+++ src/sys/net/if_ethersubr.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_ethersubr.c,v 1.212 2015/08/24 22:21:26 pooka Exp $ */
+/* $NetBSD: if_ethersubr.c,v 1.213 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -61,7 +61,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: if_ethersubr.c,v 1.212 2015/08/24 22:21:26 pooka Exp $");
+__KERNEL_RCSID(0, "$NetBSD: if_ethersubr.c,v 1.213 2015/08/31 08:05:20 ozaki-r Exp $");
#ifdef _KERNEL_OPT
#include "opt_inet.h"
@@ -243,7 +243,7 @@ ether_output(struct ifnet * const ifp0,
else if (m->m_flags & M_MCAST)
ETHER_MAP_IP_MULTICAST(&satocsin(dst)->sin_addr, edst);
else if (!arpresolve(ifp, rt, m, dst, edst))
- return (0); /* if not yet resolved */
+ return 0; /* if not yet resolved */
/* If broadcasting on a simplex interface, loopback a copy */
if ((m->m_flags & M_BCAST) && (ifp->if_flags & IFF_SIMPLEX))
mcopy = m_copy(m, 0, (int)M_COPYALL);
Index: src/sys/net/if_llatbl.c
diff -u src/sys/net/if_llatbl.c:1.1 src/sys/net/if_llatbl.c:1.2
--- src/sys/net/if_llatbl.c:1.1 Mon Aug 31 07:56:58 2015
+++ src/sys/net/if_llatbl.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_llatbl.c,v 1.1 2015/08/31 07:56:58 ozaki-r Exp $ */
+/* $NetBSD: if_llatbl.c,v 1.2 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (c) 2004 Luigi Rizzo, Alessandro Cerri. All rights reserved.
* Copyright (c) 2004-2008 Qing Li. All rights reserved.
@@ -373,6 +373,11 @@ lltable_free(struct lltable *llt)
LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next) {
if (callout_stop(&lle->la_timer))
LLE_REMREF(lle);
+#if __NetBSD__
+ /* XXX should have callback? */
+ if (lle->la_rt != NULL)
+ rtfree(lle->la_rt);
+#endif
llentry_free(lle);
}
Index: src/sys/net/if_llatbl.h
diff -u src/sys/net/if_llatbl.h:1.1 src/sys/net/if_llatbl.h:1.2
--- src/sys/net/if_llatbl.h:1.1 Mon Aug 31 07:56:58 2015
+++ src/sys/net/if_llatbl.h Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_llatbl.h,v 1.1 2015/08/31 07:56:58 ozaki-r Exp $ */
+/* $NetBSD: if_llatbl.h,v 1.2 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (c) 2004 Luigi Rizzo, Alessandro Cerri. All rights reserved.
* Copyright (c) 2004-2008 Qing Li. All rights reserved.
@@ -94,10 +94,29 @@ struct llentry {
#endif
};
-#define LLE_WLOCK(lle) rw_enter(&(lle)->lle_lock, RW_WRITER)
-#define LLE_RLOCK(lle) rw_enter(&(lle)->lle_lock, RW_READER)
-#define LLE_WUNLOCK(lle) rw_exit(&(lle)->lle_lock)
-#define LLE_RUNLOCK(lle) rw_exit(&(lle)->lle_lock)
+
+#if 0
+#define LLE_LOCK_TRACE(n) aprint_normal("%s: " #n " line %d\n", __func__, __LINE__)
+#else
+#define LLE_LOCK_TRACE(n)
+#endif
+
+#define LLE_WLOCK(lle) do { \
+ LLE_LOCK_TRACE(WL); \
+ rw_enter(&(lle)->lle_lock, RW_WRITER); \
+ } while (0)
+#define LLE_RLOCK(lle) do { \
+ LLE_LOCK_TRACE(RL); \
+ rw_enter(&(lle)->lle_lock, RW_READER); \
+ } while (0)
+#define LLE_WUNLOCK(lle) do { \
+ LLE_LOCK_TRACE(WU); \
+ rw_exit(&(lle)->lle_lock); \
+ } while (0)
+#define LLE_RUNLOCK(lle) do { \
+ LLE_LOCK_TRACE(RU); \
+ rw_exit(&(lle)->lle_lock); \
+ } while (0)
#define LLE_DOWNGRADE(lle) rw_downgrade(&(lle)->lle_lock)
#define LLE_TRY_UPGRADE(lle) rw_tryupgrade(&(lle)->lle_lock)
#ifdef __FreeBSD__
@@ -106,7 +125,7 @@ struct llentry {
#define LLE_LOCK_INIT(lle) rw_init(&(lle)->lle_lock)
#endif
#define LLE_LOCK_DESTROY(lle) rw_destroy(&(lle)->lle_lock)
-#define LLE_WLOCK_ASSERT(lle) KASSERT(rw_lock_held(&(lle)->lle_lock))
+#define LLE_WLOCK_ASSERT(lle) KASSERT(rw_write_held(&(lle)->lle_lock))
#define LLE_IS_VALID(lle) (((lle) != NULL) && ((lle) != (void *)-1))
Index: src/sys/net/if_tokensubr.c
diff -u src/sys/net/if_tokensubr.c:1.70 src/sys/net/if_tokensubr.c:1.71
--- src/sys/net/if_tokensubr.c:1.70 Mon Aug 24 22:21:26 2015
+++ src/sys/net/if_tokensubr.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_tokensubr.c,v 1.70 2015/08/24 22:21:26 pooka Exp $ */
+/* $NetBSD: if_tokensubr.c,v 1.71 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (c) 1982, 1989, 1993
@@ -92,7 +92,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: if_tokensubr.c,v 1.70 2015/08/24 22:21:26 pooka Exp $");
+__KERNEL_RCSID(0, "$NetBSD: if_tokensubr.c,v 1.71 2015/08/31 08:05:20 ozaki-r Exp $");
#ifdef _KERNEL_OPT
#include "opt_inet.h"
@@ -114,11 +114,12 @@ __KERNEL_RCSID(0, "$NetBSD: if_tokensubr
#include <sys/cpu.h>
#include <net/if.h>
-#include <net/netisr.h>
-#include <net/route.h>
-#include <net/if_llc.h>
#include <net/if_dl.h>
+#include <net/if_llatbl.h>
+#include <net/if_llc.h>
#include <net/if_types.h>
+#include <net/netisr.h>
+#include <net/route.h>
#include <net/bpf.h>
@@ -217,9 +218,13 @@ token_output(struct ifnet *ifp0, struct
* XXX m->m_flags & M_MCAST IEEE802_MAP_IP_MULTICAST ??
*/
else {
+ struct llentry *la;
if (!arpresolve(ifp, rt, m, dst, edst))
return (0); /* if not yet resolved */
- rif = TOKEN_RIF((struct llinfo_arp *) rt->rt_llinfo);
+ la = rt->rt_llinfo;
+ KASSERT(la != NULL);
+ KASSERT(la->la_opaque != NULL);
+ rif = la->la_opaque;
riflen = (ntohs(rif->tr_rcf) & TOKEN_RCF_LEN_MASK) >> 8;
}
/* If broadcasting on a simplex interface, loopback a copy. */
Index: src/sys/netinet/if_arp.c
diff -u src/sys/netinet/if_arp.c:1.173 src/sys/netinet/if_arp.c:1.174
--- src/sys/netinet/if_arp.c:1.173 Mon Aug 24 22:21:26 2015
+++ src/sys/netinet/if_arp.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: if_arp.c,v 1.173 2015/08/24 22:21:26 pooka Exp $ */
+/* $NetBSD: if_arp.c,v 1.174 2015/08/31 08:05:20 ozaki-r Exp $ */
/*-
* Copyright (c) 1998, 2000, 2008 The NetBSD Foundation, Inc.
@@ -68,7 +68,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.173 2015/08/24 22:21:26 pooka Exp $");
+__KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.174 2015/08/31 08:05:20 ozaki-r Exp $");
#ifdef _KERNEL_OPT
#include "opt_ddb.h"
@@ -98,6 +98,7 @@ __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1
#include <sys/socketvar.h>
#include <sys/percpu.h>
#include <sys/cprng.h>
+#include <sys/kmem.h>
#include <net/ethertypes.h>
#include <net/if.h>
@@ -105,6 +106,7 @@ __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1
#include <net/if_token.h>
#include <net/if_types.h>
#include <net/if_ether.h>
+#include <net/if_llatbl.h>
#include <net/net_osdep.h>
#include <net/route.h>
#include <net/net_stats.h>
@@ -143,6 +145,7 @@ static int arpt_prune = (5*60*1); /* wal
static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
static int arpt_down = 20; /* once declared down, don't send for 20 secs */
static int arpt_refresh = (5*60); /* time left before refreshing */
+static int arp_maxhold = 1; /* number of packets to hold per ARP entry */
#define rt_expire rt_rmx.rmx_expire
#define rt_pksent rt_rmx.rmx_pksent
@@ -158,12 +161,10 @@ static void arp_init(void);
static struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
const struct sockaddr *);
-static void arptfree(struct llinfo_arp *);
+static void arptfree(struct llentry *);
static void arptimer(void *);
-static struct llinfo_arp *arplookup1(struct mbuf *, const struct in_addr *,
- int, int, struct rtentry *);
-static struct llinfo_arp *arplookup(struct mbuf *, const struct in_addr *,
- int, int);
+static struct llentry *arplookup(struct ifnet *, struct mbuf *,
+ const struct in_addr *, int, int, int, struct rtentry *);
static void in_arpinput(struct mbuf *);
static void in_revarpinput(struct mbuf *);
static void revarprequest(struct ifnet *);
@@ -175,7 +176,6 @@ static void arp_dad_start(struct ifaddr
static void arp_dad_stop(struct ifaddr *);
static void arp_dad_duplicated(struct ifaddr *);
-LIST_HEAD(llinfo_arpq, llinfo_arp) llinfo_arp;
struct ifqueue arpintrq = {
.ifq_head = NULL,
.ifq_tail = NULL,
@@ -186,7 +186,6 @@ struct ifqueue arpintrq = {
static int arp_inuse, arp_allocated;
static int arp_maxtries = 5;
static int useloopback = 1; /* use loopback interface for local traffic */
-static int arpinit_done = 0;
static percpu_t *arpstat_percpu;
@@ -196,8 +195,6 @@ static percpu_t *arpstat_percpu;
#define ARP_STATINC(x) _NET_STATINC(arpstat_percpu, x)
#define ARP_STATADD(x, v) _NET_STATADD(arpstat_percpu, x, v)
-struct callout arptimer_ch;
-
/* revarp state */
static struct in_addr myip, srv_ip;
static int myip_initialized = 0;
@@ -287,75 +284,6 @@ struct domain arpdomain = {
.dom_protoswNPROTOSW = &arpsw[__arraycount(arpsw)],
};
-/*
- * ARP table locking.
- *
- * to prevent lossage vs. the arp_drain routine (which may be called at
- * any time, including in a device driver context), we do two things:
- *
- * 1) manipulation of la->la_hold is done at splnet() (for all of
- * about two instructions).
- *
- * 2) manipulation of the arp table's linked list is done under the
- * protection of the ARP_LOCK; if arp_drain() or arptimer is called
- * while the arp table is locked, we punt and try again later.
- */
-
-static int arp_locked;
-static inline int arp_lock_try(int);
-static inline void arp_unlock(void);
-
-static inline int
-arp_lock_try(int recurse)
-{
- int s;
-
- /*
- * Use splvm() -- we're blocking things that would cause
- * mbuf allocation.
- */
- s = splvm();
- if (!recurse && arp_locked) {
- splx(s);
- return 0;
- }
- arp_locked++;
- splx(s);
- return 1;
-}
-
-static inline void
-arp_unlock(void)
-{
- int s;
-
- s = splvm();
- arp_locked--;
- splx(s);
-}
-
-#ifdef DIAGNOSTIC
-#define ARP_LOCK(recurse) \
-do { \
- if (arp_lock_try(recurse) == 0) { \
- printf("%s:%d: arp already locked\n", __FILE__, __LINE__); \
- panic("arp_lock"); \
- } \
-} while (/*CONSTCOND*/ 0)
-#define ARP_LOCK_CHECK() \
-do { \
- if (arp_locked == 0) { \
- printf("%s:%d: arp lock not held\n", __FILE__, __LINE__); \
- panic("arp lock check"); \
- } \
-} while (/*CONSTCOND*/ 0)
-#else
-#define ARP_LOCK(x) (void) arp_lock_try(x)
-#define ARP_LOCK_CHECK() /* nothing */
-#endif
-
-#define ARP_UNLOCK() arp_unlock()
-
static void sysctl_net_inet_arp_setup(struct sysctllog **);
void
@@ -380,74 +308,71 @@ arp_drainstub(void)
void
arp_drain(void)
{
- struct llinfo_arp *la;
- int count = 0;
- struct mbuf *mold;
-
- KERNEL_LOCK(1, NULL);
-
- if (arp_lock_try(0) == 0) {
- KERNEL_UNLOCK_ONE(NULL);
- return;
- }
-
- LIST_FOREACH(la, &llinfo_arp, la_list) {
- mold = la->la_hold;
- la->la_hold = NULL;
- if (mold) {
- m_freem(mold);
- count++;
- }
- }
- ARP_UNLOCK();
- ARP_STATADD(ARP_STAT_DFRDROPPED, count);
- KERNEL_UNLOCK_ONE(NULL);
+ lltable_drain(AF_INET);
}
-
-/*
- * Timeout routine. Age arp_tab entries periodically.
- */
-/* ARGSUSED */
static void
arptimer(void *arg)
{
- struct llinfo_arp *la, *nla;
+ struct llentry *lle = arg;
+ struct ifnet *ifp;
mutex_enter(softnet_lock);
- KERNEL_LOCK(1, NULL);
- if (arp_lock_try(0) == 0) {
- /* get it later.. */
- KERNEL_UNLOCK_ONE(NULL);
- mutex_exit(softnet_lock);
- return;
+ if (lle == NULL)
+ goto out;
+
+ if (lle->la_flags & LLE_STATIC)
+ goto out;
+
+ LLE_WLOCK(lle);
+ if (callout_pending(&lle->la_timer)) {
+ /*
+ * Here we are a bit odd here in the treatment of
+ * active/pending. If the pending bit is set, it got
+ * rescheduled before I ran. The active
+ * bit we ignore, since if it was stopped
+ * in ll_tablefree() and was currently running
+ * it would have return 0 so the code would
+ * not have deleted it since the callout could
+ * not be stopped so we want to go through
+ * with the delete here now. If the callout
+ * was restarted, the pending bit will be back on and
+ * we just want to bail since the callout_reset would
+ * return 1 and our reference would have been removed
+ * by arpresolve() below.
+ */
+ LLE_WUNLOCK(lle);
+ goto out;
}
+ ifp = lle->lle_tbl->llt_ifp;
- callout_reset(&arptimer_ch, arpt_prune * hz, arptimer, NULL);
- LIST_FOREACH_SAFE(la, &llinfo_arp, la_list, nla) {
- struct rtentry *rt = la->la_rt;
-
- if (rt->rt_expire == 0)
- continue;
- if ((rt->rt_expire - time_uptime) < arpt_refresh &&
- rt->rt_pksent > (time_uptime - arpt_keep)) {
- /*
- * If the entry has been used during since last
- * refresh, try to renew it before deleting.
- */
- arprequest(rt->rt_ifp,
- &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
- &satocsin(rt_getkey(rt))->sin_addr,
- CLLADDR(rt->rt_ifp->if_sadl));
- } else if (rt->rt_expire <= time_uptime)
- arptfree(la); /* timer has expired; clear */
+ callout_stop(&lle->la_timer);
+
+ /* XXX: LOR avoidance. We still have ref on lle. */
+ LLE_WUNLOCK(lle);
+
+ /* We have to call this w/o lock */
+ arptfree(lle);
+
+ IF_AFDATA_LOCK(ifp);
+ LLE_WLOCK(lle);
+
+ /* Guard against race with other llentry_free(). */
+ if (lle->la_flags & LLE_LINKED) {
+ size_t pkts_dropped;
+
+ LLE_REMREF(lle);
+ pkts_dropped = llentry_free(lle);
+ ARP_STATADD(ARP_STAT_DFRDROPPED, pkts_dropped);
+ } else {
+ LLE_FREE_LOCKED(lle);
}
- ARP_UNLOCK();
+ IF_AFDATA_UNLOCK(ifp);
- KERNEL_UNLOCK_ONE(NULL);
+out:
mutex_exit(softnet_lock);
}
@@ -498,19 +423,11 @@ void
arp_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
{
struct sockaddr *gate = rt->rt_gateway;
- struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
- size_t allocsize;
- struct mbuf *mold;
- int s;
+ struct llentry *la = NULL;
struct in_ifaddr *ia;
struct ifaddr *ifa;
struct ifnet *ifp = rt->rt_ifp;
-
- if (!arpinit_done) {
- arpinit_done = 1;
- callout_init(&arptimer_ch, CALLOUT_MPSAFE);
- callout_reset(&arptimer_ch, hz, arptimer, NULL);
- }
+ int flags = 0;
if (req == RTM_LLINFO_UPD) {
struct in_addr *in;
@@ -565,7 +482,9 @@ arp_rtrequest(int req, struct rtentry *r
return;
}
- ARP_LOCK(1); /* we may already be locked here. */
+ IF_AFDATA_RLOCK(ifp);
+ la = lla_lookup(LLTABLE(ifp), flags, rt_getkey(rt));
+ IF_AFDATA_RUNLOCK(ifp);
switch (req) {
case RTM_SETGATE:
@@ -660,26 +579,38 @@ arp_rtrequest(int req, struct rtentry *r
* Case 2: This route may come from cloning, or a manual route
* add with a LL address.
*/
+ flags = LLE_EXCLUSIVE;
+ if ((rt->rt_flags & RTF_CLONED) == 0)
+ flags |= LLE_IFADDR;
+
+ IF_AFDATA_WLOCK(ifp);
+ la = lla_create(LLTABLE(ifp), flags, rt_getkey(rt));
+ IF_AFDATA_WUNLOCK(ifp);
+
+ if (la == NULL) {
+ log(LOG_DEBUG, "%s: lla_create failed\n",
+ __func__);
+ rt->rt_llinfo = NULL;
+ break;
+ }
+ rt->rt_llinfo = la;
switch (ifp->if_type) {
#if NTOKEN > 0
case IFT_ISO88025:
- allocsize = sizeof(*la) + sizeof(struct token_rif);
+ la->la_opaque = kmem_alloc(sizeof(struct token_rif),
+ KM_SLEEP);
break;
#endif /* NTOKEN > 0 */
default:
- allocsize = sizeof(*la);
- }
- R_Malloc(la, struct llinfo_arp *, allocsize);
- rt->rt_llinfo = (void *)la;
- if (la == NULL) {
- log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
break;
}
- arp_inuse++, arp_allocated++;
- memset(la, 0, allocsize);
la->la_rt = rt;
+ rt->rt_refcnt++;
rt->rt_flags |= RTF_LLINFO;
- LIST_INSERT_HEAD(&llinfo_arp, la, la_list);
+ arp_inuse++, arp_allocated++;
+
+ LLE_WUNLOCK(la);
+ la = NULL;
INADDR_TO_IA(satocsin(rt_getkey(rt))->sin_addr, ia);
while (ia && ia->ia_ifp != ifp)
@@ -726,21 +657,54 @@ arp_rtrequest(int req, struct rtentry *r
if (la == NULL)
break;
arp_inuse--;
- LIST_REMOVE(la, la_list);
rt->rt_llinfo = NULL;
rt->rt_flags &= ~RTF_LLINFO;
- s = splnet();
- mold = la->la_hold;
- la->la_hold = NULL;
- splx(s);
+ LLE_RUNLOCK(la);
- if (mold)
- m_freem(mold);
+ flags |= LLE_EXCLUSIVE;
+ IF_AFDATA_WLOCK(ifp);
+
+ la = lla_lookup(LLTABLE(ifp), flags, rt_getkey(rt));
+ /* This shouldn't happen */
+ if (la == NULL) {
+ IF_AFDATA_WUNLOCK(ifp);
+ break;
+ }
- Free((void *)la);
+ if (la->la_opaque != NULL) {
+ switch (ifp->if_type) {
+#if NTOKEN > 0
+ case IFT_ISO88025:
+ kmem_free(la->la_opaque,
+ sizeof(struct token_rif));
+ break;
+#endif /* NTOKEN > 0 */
+ default:
+ break;
+ }
+ }
+
+ if (la->la_rt != NULL) {
+ /*
+ * Don't rtfree (may actually free objects) here.
+ * Leave it to rtrequest1.
+ */
+ la->la_rt->rt_refcnt--;
+ la->la_rt = NULL;
+ }
+ llentry_free(la);
+
+ IF_AFDATA_WUNLOCK(ifp);
+ la = NULL;
+ }
+
+ if (la != NULL) {
+ if (flags & LLE_EXCLUSIVE)
+ LLE_WUNLOCK(la);
+ else
+ LLE_RUNLOCK(la);
}
- ARP_UNLOCK();
}
/*
@@ -759,6 +723,10 @@ arprequest(struct ifnet *ifp,
struct sockaddr sa;
uint64_t *arps;
+ KASSERT(sip != NULL);
+ KASSERT(tip != NULL);
+ KASSERT(enaddr != NULL);
+
if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
return;
MCLAIM(m, &arpdomain.dom_mowner);
@@ -816,12 +784,15 @@ int
arpresolve(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
const struct sockaddr *dst, u_char *desten)
{
- struct llinfo_arp *la;
+ struct llentry *la;
const struct sockaddr_dl *sdl;
- struct mbuf *mold;
- int s;
+ int renew;
+ int flags = 0;
+ int error;
+ bool create;
- if ((la = arplookup1(m, &satocsin(dst)->sin_addr, 1, 0, rt)) != NULL)
+ la = arplookup(ifp, m, &satocsin(dst)->sin_addr, 1, 0, 0, rt);
+ if (la != NULL)
rt = la->la_rt;
if (la == NULL || rt == NULL) {
@@ -830,6 +801,8 @@ arpresolve(struct ifnet *ifp, struct rte
"arpresolve: can't allocate llinfo on %s for %s\n",
ifp->if_xname, in_fmtaddr(satocsin(dst)->sin_addr));
m_freem(m);
+ if (la != NULL)
+ LLE_RUNLOCK(la);
return 0;
}
sdl = satocsdl(rt->rt_gateway);
@@ -842,24 +815,9 @@ arpresolve(struct ifnet *ifp, struct rte
memcpy(desten, CLLADDR(sdl),
min(sdl->sdl_alen, ifp->if_addrlen));
rt->rt_pksent = time_uptime; /* Time for last pkt sent */
+ LLE_RUNLOCK(la);
return 1;
}
- /*
- * There is an arptab entry, but no ethernet address
- * response yet. Replace the held mbuf with this
- * latest one.
- */
-
- ARP_STATINC(ARP_STAT_DFRTOTAL);
- s = splnet();
- mold = la->la_hold;
- la->la_hold = m;
- splx(s);
-
- if (mold) {
- ARP_STATINC(ARP_STAT_DFRDROPPED);
- m_freem(mold);
- }
/*
* Re-send the ARP request when appropriate.
@@ -872,27 +830,158 @@ arpresolve(struct ifnet *ifp, struct rte
rt->rt_expire = time_uptime;
}
#endif
- if (rt->rt_expire) {
- rt->rt_flags &= ~RTF_REJECT;
- if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
- rt->rt_expire = time_uptime;
- if (la->la_asked++ < arp_maxtries) {
- arprequest(ifp,
- &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
- &satocsin(dst)->sin_addr,
+
+retry:
+ create = false;
+ if (la == NULL) {
+ IF_AFDATA_RLOCK(ifp);
+ la = lla_lookup(LLTABLE(ifp), flags, dst);
+ IF_AFDATA_RUNLOCK(ifp);
+ }
+
+ if ((la == NULL) && ((flags & LLE_EXCLUSIVE) == 0)
+#ifdef __FreeBSD__
+ && ((ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0)) {
+#else
+ && ((ifp->if_flags & IFF_NOARP) == 0)) {
+#endif
+ create = true;
+ flags |= LLE_EXCLUSIVE;
+ IF_AFDATA_WLOCK(ifp);
+ la = lla_create(LLTABLE(ifp), flags, dst);
+ IF_AFDATA_WUNLOCK(ifp);
+ }
+
+ if (la == NULL) {
+ if (create) {
+ log(LOG_DEBUG,
+ "%s: failed to create llentry for %s on %s\n",
+ __func__, inet_ntoa(satocsin(dst)->sin_addr),
+ ifp->if_xname);
+ }
+ m_freem(m);
+ return 0;
+ }
+
+ if ((la->la_flags & LLE_VALID) &&
+ ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
+ memcpy(desten, CLLADDR(sdl),
+ min(sdl->sdl_alen, ifp->if_addrlen));
+ renew = 0;
+ /*
+ * If entry has an expiry time and it is approaching,
+ * see if we need to send an ARP request within this
+ * arpt_down interval.
+ */
+ if (!(la->la_flags & LLE_STATIC) &&
+ time_uptime + la->la_preempt > la->la_expire) {
+ renew = 1;
+ la->la_preempt--;
+ }
+
+ if (flags & LLE_EXCLUSIVE)
+ LLE_WUNLOCK(la);
+ else
+ LLE_RUNLOCK(la);
+
+ if (renew == 1) {
+ const u_int8_t *enaddr =
#if NCARP > 0
- (rt->rt_ifp->if_type == IFT_CARP) ?
- CLLADDR(rt->rt_ifp->if_sadl):
+ (rt->rt_ifp->if_type == IFT_CARP) ?
+ CLLADDR(rt->rt_ifp->if_sadl):
#endif
- CLLADDR(ifp->if_sadl));
- } else {
- rt->rt_flags |= RTF_REJECT;
- rt->rt_expire += arpt_down;
- la->la_asked = 0;
+ CLLADDR(ifp->if_sadl);
+ arprequest(ifp, &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
+ &satocsin(dst)->sin_addr, enaddr);
+ }
+
+ return 1;
+ }
+
+ if (la->la_flags & LLE_STATIC) { /* should not happen! */
+ log(LOG_DEBUG, "arpresolve: ouch, empty static llinfo for %s\n",
+ inet_ntoa(satocsin(dst)->sin_addr));
+ m_freem(m);
+ error = EINVAL;
+ goto done;
+ }
+
+ renew = (la->la_asked == 0 || la->la_expire != time_uptime);
+ if ((renew || m != NULL) && (flags & LLE_EXCLUSIVE) == 0) {
+ flags |= LLE_EXCLUSIVE;
+ LLE_RUNLOCK(la);
+ la = NULL;
+ goto retry;
+ }
+ /*
+ * There is an arptab entry, but no ethernet address
+ * response yet. Add the mbuf to the list, dropping
+ * the oldest packet if we have exceeded the system
+ * setting.
+ */
+ if (m != NULL) {
+ LLE_WLOCK_ASSERT(la);
+ if (la->la_numheld >= arp_maxhold) {
+ if (la->la_hold != NULL) {
+ struct mbuf *next = la->la_hold->m_nextpkt;
+ m_freem(la->la_hold);
+ la->la_hold = next;
+ la->la_numheld--;
+ ARP_STATINC(ARP_STAT_DFRDROPPED);
}
}
+ if (la->la_hold != NULL) {
+ struct mbuf *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);
+ }
+
}
- return 0;
+ /*
+ * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
+ * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
+ * if we have already sent arp_maxtries ARP requests. Retransmit the
+ * ARP request, but not faster than one request per second.
+ */
+ if (la->la_asked < arp_maxtries)
+ error = EWOULDBLOCK; /* First request. */
+ else
+ error = (rt->rt_flags & RTF_GATEWAY) ?
+ EHOSTUNREACH : EHOSTDOWN;
+
+ if (renew) {
+ const u_int8_t *enaddr =
+#if NCARP > 0
+ (rt->rt_ifp->if_type == IFT_CARP) ?
+ CLLADDR(rt->rt_ifp->if_sadl):
+#endif
+ CLLADDR(ifp->if_sadl);
+ LLE_ADDREF(la);
+ la->la_expire = time_uptime;
+ callout_reset(&la->la_timer, hz * arpt_down,
+ arptimer, la);
+ la->la_asked++;
+ if (flags & LLE_EXCLUSIVE)
+ LLE_WUNLOCK(la);
+ else
+ LLE_RUNLOCK(la);
+ arprequest(ifp, &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
+ &satocsin(dst)->sin_addr, enaddr);
+ return error == 0;
+ }
+done:
+ if (flags & LLE_EXCLUSIVE)
+ LLE_WUNLOCK(la);
+ else
+ LLE_RUNLOCK(la);
+ return error == 0;
}
/*
@@ -976,8 +1065,8 @@ in_arpinput(struct mbuf *m)
{
struct arphdr *ah;
struct ifnet *ifp = m->m_pkthdr.rcvif;
- struct llinfo_arp *la = NULL;
- struct rtentry *rt;
+ struct llentry *la = NULL;
+ struct rtentry *rt = NULL;
struct in_ifaddr *ia;
#if NBRIDGE > 0
struct in_ifaddr *bridge_ia = NULL;
@@ -985,13 +1074,11 @@ in_arpinput(struct mbuf *m)
#if NCARP > 0
u_int32_t count = 0, index = 0;
#endif
- struct sockaddr_dl *sdl;
+ struct sockaddr_dl *sdl = NULL;
struct sockaddr sa;
struct in_addr isaddr, itaddr, myaddr;
int op;
- struct mbuf *mold;
void *tha;
- int s;
uint64_t *arps;
if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
@@ -1142,8 +1229,14 @@ in_arpinput(struct mbuf *m)
itaddr = myaddr;
goto reply;
}
- la = arplookup(m, &isaddr, in_hosteq(itaddr, myaddr), 0);
- if (la != NULL && (rt = la->la_rt) && (sdl = satosdl(rt->rt_gateway))) {
+
+ la = arplookup(ifp, m, &isaddr, in_hosteq(itaddr, myaddr), 0, 1, NULL);
+ if (la != NULL) {
+ rt = la->la_rt;
+ if (rt != NULL)
+ sdl = satosdl(rt->rt_gateway);
+ }
+ if (sdl != NULL) {
if (sdl->sdl_alen &&
memcmp(ar_sha(ah), CLLADDR(sdl), sdl->sdl_alen)) {
if (rt->rt_flags & RTF_STATIC) {
@@ -1226,22 +1319,43 @@ in_arpinput(struct mbuf *m)
#endif /* NTOKEN > 0 */
(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, ar_sha(ah),
ah->ar_hln);
- if (rt->rt_expire)
+ if (rt->rt_expire) {
rt->rt_expire = time_uptime + arpt_keep;
+
+ KASSERT((la->la_flags & LLE_STATIC) == 0);
+ LLE_ADDREF(la);
+ callout_reset(&la->la_timer, hz * arpt_keep, arptimer, la);
+ }
rt->rt_flags &= ~RTF_REJECT;
la->la_asked = 0;
- s = splnet();
- mold = la->la_hold;
- la->la_hold = NULL;
- splx(s);
-
- if (mold) {
- ARP_STATINC(ARP_STAT_DFRSENT);
- (*ifp->if_output)(ifp, mold, rt_getkey(rt), rt);
- }
+ if (la->la_hold != NULL) {
+ int n = la->la_numheld;
+ struct mbuf *m_hold, *m_hold_next;
+
+ m_hold = la->la_hold;
+ la->la_hold = NULL;
+ la->la_numheld = 0;
+ /*
+ * We have to unlock here because if_output would call
+ * arpresolve
+ */
+ LLE_WUNLOCK(la);
+ ARP_STATADD(ARP_STAT_DFRSENT, n);
+ for (; 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, rt_getkey(rt), rt);
+ }
+ } else
+ LLE_WUNLOCK(la);
+ la = NULL;
}
reply:
+ if (la != NULL) {
+ LLE_WUNLOCK(la);
+ la = NULL;
+ }
if (op != ARPOP_REQUEST) {
if (op == ARPOP_REPLY)
ARP_STATINC(ARP_STAT_RCVREPLY);
@@ -1258,10 +1372,12 @@ reply:
memcpy(tha, ar_sha(ah), ah->ar_hln);
memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
} else {
- la = arplookup(m, &itaddr, 0, SIN_PROXY);
+ la = arplookup(ifp, m, &itaddr, 0, SIN_PROXY, 0, NULL);
if (la == NULL)
goto out;
rt = la->la_rt;
+ LLE_RUNLOCK(la);
+ la = NULL;
if (rt->rt_ifp->if_type == IFT_CARP &&
m->m_pkthdr.rcvif->if_type != IFT_CARP)
goto out;
@@ -1300,47 +1416,38 @@ reply:
ARP_STAT_PUTREF();
(*ifp->if_output)(ifp, m, &sa, NULL);
return;
+
out:
+ if (la != NULL)
+ LLE_WUNLOCK(la);
m_freem(m);
}
/*
* Free an arp entry.
*/
-static void arptfree(struct llinfo_arp *la)
+static void arptfree(struct llentry *la)
{
struct rtentry *rt = la->la_rt;
- struct sockaddr_dl *sdl;
- ARP_LOCK_CHECK();
+ KASSERT(rt != NULL);
- if (rt == NULL)
- panic("arptfree");
- if (rt->rt_refcnt > 0 && (sdl = satosdl(rt->rt_gateway)) &&
- sdl->sdl_family == AF_LINK) {
- sdl->sdl_alen = 0;
- la->la_asked = 0;
- rt->rt_flags &= ~RTF_REJECT;
- return;
+ if (la->la_rt != NULL) {
+ rtfree(la->la_rt);
+ la->la_rt = NULL;
}
- rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL);
-}
-static struct llinfo_arp *
-arplookup(struct mbuf *m, const struct in_addr *addr, int create, int proxy)
-{
- return arplookup1(m, addr, create, proxy, NULL);
+ rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL);
}
/*
* Lookup or enter a new address in arptab.
*/
-static struct llinfo_arp *
-arplookup1(struct mbuf *m, const struct in_addr *addr, int create, int proxy,
- struct rtentry *rt0)
+static struct llentry *
+arplookup(struct ifnet *ifp, struct mbuf *m, const struct in_addr *addr,
+ int create, int proxy, int wlock, struct rtentry *rt0)
{
struct arphdr *ah;
- struct ifnet *ifp = m->m_pkthdr.rcvif;
struct rtentry *rt;
struct sockaddr_inarp sin;
const char *why = NULL;
@@ -1364,8 +1471,22 @@ arplookup1(struct mbuf *m, const struct
(__rt)->rt_gateway->sa_family == AF_LINK)
- if (IS_LLINFO(rt))
- return (struct llinfo_arp *)rt->rt_llinfo;
+ if (IS_LLINFO(rt)) {
+ struct llentry *la;
+ int flags = wlock ? LLE_EXCLUSIVE : 0;
+
+ if (create) {
+ IF_AFDATA_WLOCK(ifp);
+ la = lla_create(LLTABLE(ifp), flags, rt_getkey(rt));
+ IF_AFDATA_WUNLOCK(ifp);
+ } else {
+ IF_AFDATA_RLOCK(ifp);
+ la = lla_lookup(LLTABLE(ifp), flags, rt_getkey(rt));
+ IF_AFDATA_RUNLOCK(ifp);
+ }
+
+ return la;
+ }
if (create) {
if (rt->rt_flags & RTF_GATEWAY) {
@@ -1382,7 +1503,7 @@ arplookup1(struct mbuf *m, const struct
lla_snprintf(ar_sha(ah), ah->ar_hln),
(ifp) ? ifp->if_xname : "null", why);
}
- if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_CLONED) != 0) {
+ if ((rt->rt_flags & RTF_CLONED) != 0) {
rtrequest(RTM_DELETE, rt_getkey(rt),
rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
}
@@ -1997,6 +2118,12 @@ db_show_arptab(db_expr_t addr, bool have
}
#endif
+void
+arp_stat_add(int type, uint64_t count)
+{
+ ARP_STATADD(type, count);
+}
+
static int
sysctl_net_inet_arp_stats(SYSCTLFN_ARGS)
{
Index: src/sys/netinet/in.c
diff -u src/sys/netinet/in.c:1.158 src/sys/netinet/in.c:1.159
--- src/sys/netinet/in.c:1.158 Mon Aug 31 08:02:44 2015
+++ src/sys/netinet/in.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: in.c,v 1.158 2015/08/31 08:02:44 ozaki-r Exp $ */
+/* $NetBSD: in.c,v 1.159 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -91,7 +91,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: in.c,v 1.158 2015/08/31 08:02:44 ozaki-r Exp $");
+__KERNEL_RCSID(0, "$NetBSD: in.c,v 1.159 2015/08/31 08:05:20 ozaki-r Exp $");
#include "arp.h"
@@ -121,6 +121,7 @@ __KERNEL_RCSID(0, "$NetBSD: in.c,v 1.158
#include <net/route.h>
#include <net/pfil.h>
+#include <net/if_arp.h>
#include <net/if_ether.h>
#include <net/if_types.h>
#include <net/if_llatbl.h>
@@ -1574,7 +1575,7 @@ in_lltable_new(struct in_addr addr4, u_i
lle->base.lle_refcnt = 1;
lle->base.lle_free = in_lltable_destroy_lle;
LLE_LOCK_INIT(&lle->base);
- callout_init(&lle->base.la_timer, 1);
+ callout_init(&lle->base.la_timer, CALLOUT_MPSAFE);
return (&lle->base);
}
@@ -1622,11 +1623,7 @@ in_lltable_free_entry(struct lltable *ll
/* Drop hold queue */
pkts_dropped = llentry_free(lle);
-#ifdef __FreeBSD__
- ARPSTAT_ADD(dropped, pkts_dropped);
-#else
- (void) pkts_dropped; /* FIXME */
-#endif
+ arp_stat_add(ARP_STAT_DFRDROPPED, (uint64_t)pkts_dropped);
}
static int
@@ -1653,7 +1650,7 @@ in_lltable_rtcheck(struct ifnet *ifp, u_
rt->rt_ifp->if_type != IFT_ETHER ||
#ifdef __FreeBSD__
(rt->rt_ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) != 0 ||
-#else /* XXX */
+#else
(rt->rt_ifp->if_flags & IFF_NOARP) != 0 ||
#endif
memcmp(rt->rt_gateway->sa_data, l3addr->sa_data,
@@ -1699,9 +1696,6 @@ in_lltable_rtcheck(struct ifnet *ifp, u_
error = 0;
error:
-#ifdef __FreeBSD__
- RTFREE_LOCKED(rt);
-#endif
return error;
}
@@ -1773,9 +1767,6 @@ in_lltable_delete(struct lltable *llt, u
if (!(lle->la_flags & LLE_IFADDR) || (flags & LLE_IFADDR)) {
LLE_WLOCK(lle);
lle->la_flags |= LLE_DELETED;
-#ifdef __FreeBSD__
- EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_DELETED);
-#endif
#ifdef DIAGNOSTIC
log(LOG_INFO, "ifaddr cache = %p is deleted\n", lle);
#endif
Index: src/sys/netinet/in_proto.c
diff -u src/sys/netinet/in_proto.c:1.113 src/sys/netinet/in_proto.c:1.114
--- src/sys/netinet/in_proto.c:1.113 Mon Aug 24 22:21:26 2015
+++ src/sys/netinet/in_proto.c Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: in_proto.c,v 1.113 2015/08/24 22:21:26 pooka Exp $ */
+/* $NetBSD: in_proto.c,v 1.114 2015/08/31 08:05:20 ozaki-r Exp $ */
/*
* Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -61,7 +61,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: in_proto.c,v 1.113 2015/08/24 22:21:26 pooka Exp $");
+__KERNEL_RCSID(0, "$NetBSD: in_proto.c,v 1.114 2015/08/31 08:05:20 ozaki-r Exp $");
#ifdef _KERNEL_OPT
#include "opt_mrouting.h"
@@ -388,13 +388,8 @@ struct domain inetdomain = {
.dom_maxrtkey = sizeof(struct ip_pack4),
.dom_if_up = in_if_up,
.dom_if_down = in_if_down,
-#ifdef IPSELSRC
.dom_ifattach = in_domifattach,
.dom_ifdetach = in_domifdetach,
-#else
- .dom_ifattach = NULL,
- .dom_ifdetach = NULL,
-#endif
.dom_if_link_state_change = in_if_link_state_change,
.dom_ifqueues = { NULL, NULL },
.dom_link = { NULL },
Index: src/sys/netinet/in_var.h
diff -u src/sys/netinet/in_var.h:1.73 src/sys/netinet/in_var.h:1.74
--- src/sys/netinet/in_var.h:1.73 Mon Aug 31 08:02:44 2015
+++ src/sys/netinet/in_var.h Mon Aug 31 08:05:20 2015
@@ -1,4 +1,4 @@
-/* $NetBSD: in_var.h,v 1.73 2015/08/31 08:02:44 ozaki-r Exp $ */
+/* $NetBSD: in_var.h,v 1.74 2015/08/31 08:05:20 ozaki-r Exp $ */
/*-
* Copyright (c) 1998 The NetBSD Foundation, Inc.
@@ -323,6 +323,9 @@ ip_newid(const struct in_ifaddr *ia)
int sysctl_inpcblist(SYSCTLFN_PROTO);
#endif
+#define LLTABLE(ifp) \
+ ((struct in_ifinfo *)(ifp)->if_afdata[AF_INET])->ii_llt
+
#endif /* !_KERNEL */
/* INET6 stuff */
