X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/9bccf70c0258c7cac2dcb80011b2a964d884c552..060df5ea7c632b1ac8cc8aac1fb59758165c2084:/bsd/netinet6/nd6.c diff --git a/bsd/netinet6/nd6.c b/bsd/netinet6/nd6.c index 60d206276..d71746042 100644 --- a/bsd/netinet6/nd6.c +++ b/bsd/netinet6/nd6.c @@ -1,4 +1,32 @@ -/* $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.9 2001/07/11 09:39:04 ume Exp $ */ +/* + * Copyright (c) 2008-2009 Apple Inc. All rights reserved. + * + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ + * + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. + * + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, + * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. + * + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ + */ + +/* $FreeBSD: src/sys/netinet6/nd6.c,v 1.20 2002/08/02 20:49:14 rwatson Exp $ */ /* $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $ */ /* @@ -46,10 +74,13 @@ #include #include #include +#include #include #include #include +#include #include +#include #define DONT_WARN_OBSOLETE #include @@ -60,6 +91,7 @@ #include #include +#include #include #include #include @@ -76,8 +108,10 @@ #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ +#define SA(p) ((struct sockaddr *)(p)) #define SIN6(s) ((struct sockaddr_in6 *)s) #define SDL(s) ((struct sockaddr_dl *)s) +#define equal(a1, a2) (bcmp((caddr_t)(a1), (caddr_t)(a2), (a1)->sa_len) == 0) /* timer values */ int nd6_prune = 1; /* walk list every 1 seconds */ @@ -101,18 +135,87 @@ int nd6_debug = 0; /* for debugging? */ static int nd6_inuse, nd6_allocated; -struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6}; -static size_t nd_ifinfo_indexlim = 8; +/* + * Synchronization notes: + * + * The global list of ND entries are stored in llinfo_nd6; an entry + * gets inserted into the list when the route is created and gets + * removed from the list when it is deleted; this is done as part + * of RTM_ADD/RTM_RESOLVE/RTM_DELETE in nd6_rtrequest(). + * + * Because rnh_lock and rt_lock for the entry are held during those + * operations, the same locks (and thus lock ordering) must be used + * elsewhere to access the relevant data structure fields: + * + * ln_next, ln_prev, ln_rt + * + * - Routing lock (rnh_lock) + * + * ln_hold, ln_asked, ln_expire, ln_state, ln_router, ln_byhint, ln_flags + * + * - Routing entry lock (rt_lock) + * + * Due to the dependency on rt_lock, llinfo_nd6 has the same lifetime + * as the route entry itself. When a route is deleted (RTM_DELETE), + * it is simply removed from the global list but the memory is not + * freed until the route itself is freed. + */ +struct llinfo_nd6 llinfo_nd6 = { + &llinfo_nd6, &llinfo_nd6, NULL, NULL, 0, 0, 0, 0, 0, 0 +}; + +/* Protected by nd_if_rwlock */ +size_t nd_ifinfo_indexlim = 32; /* increased for 5589193 */ struct nd_ifinfo *nd_ifinfo = NULL; + +static lck_grp_attr_t *nd_if_rwlock_grp_attr; +static lck_grp_t *nd_if_rwlock_grp; +static lck_attr_t *nd_if_rwlock_attr; +lck_rw_t *nd_if_rwlock; + +/* Protected by nd6_mutex */ struct nd_drhead nd_defrouter; struct nd_prhead nd_prefix = { 0 }; int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL; static struct sockaddr_in6 all1_sa; -static void nd6_slowtimo_funneled __P((void *)); -static int regen_tmpaddr __P((struct in6_ifaddr *)); +static int regen_tmpaddr(struct in6_ifaddr *); +extern lck_mtx_t *ip6_mutex; +extern lck_mtx_t *nd6_mutex; +static void nd6_slowtimo(void *ignored_arg); +static struct llinfo_nd6 *nd6_llinfo_alloc(void); +static void nd6_llinfo_free(void *); + +static void nd6_siocgdrlst(void *, int); +static void nd6_siocgprlst(void *, int); + +/* + * Insertion and removal from llinfo_nd6 must be done with rnh_lock held. + */ +#define LN_DEQUEUE(_ln) do { \ + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); \ + RT_LOCK_ASSERT_HELD((_ln)->ln_rt); \ + (_ln)->ln_next->ln_prev = (_ln)->ln_prev; \ + (_ln)->ln_prev->ln_next = (_ln)->ln_next; \ + (_ln)->ln_prev = (_ln)->ln_next = NULL; \ + (_ln)->ln_flags &= ~ND6_LNF_IN_USE; \ +} while (0) + +#define LN_INSERTHEAD(_ln) do { \ + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); \ + RT_LOCK_ASSERT_HELD((_ln)->ln_rt); \ + (_ln)->ln_next = llinfo_nd6.ln_next; \ + llinfo_nd6.ln_next = (_ln); \ + (_ln)->ln_prev = &llinfo_nd6; \ + (_ln)->ln_next->ln_prev = (_ln); \ + (_ln)->ln_flags |= ND6_LNF_IN_USE; \ +} while (0) + +static struct zone *llinfo_nd6_zone; +#define LLINFO_ND6_ZONE_MAX 256 /* maximum elements in zone */ +#define LLINFO_ND6_ZONE_NAME "llinfo_nd6" /* name for zone */ void nd6_init() @@ -121,7 +224,7 @@ nd6_init() int i; if (nd6_init_done) { - log(LOG_NOTICE, "nd6_init called more than once(ignored)\n"); + log(LOG_NOTICE, "nd6_init called more than once (ignored)\n"); return; } @@ -133,38 +236,88 @@ nd6_init() /* initialization of the default router list */ TAILQ_INIT(&nd_defrouter); + nd_if_rwlock_grp_attr = lck_grp_attr_alloc_init(); + nd_if_rwlock_grp = lck_grp_alloc_init("nd_if_rwlock", + nd_if_rwlock_grp_attr); + nd_if_rwlock_attr = lck_attr_alloc_init(); + nd_if_rwlock = lck_rw_alloc_init(nd_if_rwlock_grp, nd_if_rwlock_attr); + + llinfo_nd6_zone = zinit(sizeof (struct llinfo_nd6), + LLINFO_ND6_ZONE_MAX * sizeof (struct llinfo_nd6), 0, + LLINFO_ND6_ZONE_NAME); + if (llinfo_nd6_zone == NULL) + panic("%s: failed allocating llinfo_nd6_zone", __func__); + + zone_change(llinfo_nd6_zone, Z_EXPAND, TRUE); + nd6_init_done = 1; /* start timer */ - timeout(nd6_slowtimo_funneled, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz); + timeout(nd6_slowtimo, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz); } -void -nd6_ifattach(ifp) - struct ifnet *ifp; +static struct llinfo_nd6 * +nd6_llinfo_alloc(void) +{ + return (zalloc(llinfo_nd6_zone)); +} + +static void +nd6_llinfo_free(void *arg) +{ + struct llinfo_nd6 *ln = arg; + + if (ln->ln_next != NULL || ln->ln_prev != NULL) { + panic("%s: trying to free %p when it is in use", __func__, ln); + /* NOTREACHED */ + } + + /* Just in case there's anything there, free it */ + if (ln->ln_hold != NULL) { + m_freem(ln->ln_hold); + ln->ln_hold = NULL; + } + + zfree(llinfo_nd6_zone, ln); +} + +int +nd6_ifattach(struct ifnet *ifp) { /* * We have some arrays that should be indexed by if_index. * since if_index will grow dynamically, they should grow too. */ + lck_rw_lock_exclusive(nd_if_rwlock); if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) { size_t n; caddr_t q; + size_t newlim = nd_ifinfo_indexlim; - while (if_index >= nd_ifinfo_indexlim) - nd_ifinfo_indexlim <<= 1; + while (if_index >= newlim) + newlim <<= 1; /* grow nd_ifinfo */ - n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo); + n = newlim * sizeof(struct nd_ifinfo); q = (caddr_t)_MALLOC(n, M_IP6NDP, M_WAITOK); + if (q == NULL) { + lck_rw_done(nd_if_rwlock); + return ENOBUFS; + } bzero(q, n); + nd_ifinfo_indexlim = newlim; if (nd_ifinfo) { bcopy((caddr_t)nd_ifinfo, q, n/2); + /* + * We might want to pattern fill the old + * array to catch use-after-free cases. + */ FREE((caddr_t)nd_ifinfo, M_IP6NDP); } nd_ifinfo = (struct nd_ifinfo *)q; } + lck_rw_done(nd_if_rwlock); #define ND nd_ifinfo[ifp->if_index] @@ -175,18 +328,23 @@ nd6_ifattach(ifp) * the linkmtu member, which was not suitable because it could be * initialized via "ifconfig mtu". */ - if (ND.basereachable) - return; - - ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu; + lck_rw_lock_shared(nd_if_rwlock); + if (ND.basereachable) { + lck_rw_done(nd_if_rwlock); + return 0; + } + ND.linkmtu = ifp->if_mtu; ND.chlim = IPV6_DEFHLIM; ND.basereachable = REACHABLE_TIME; ND.reachable = ND_COMPUTE_RTIME(ND.basereachable); ND.retrans = RETRANS_TIMER; ND.receivedra = 0; ND.flags = ND6_IFF_PERFORMNUD; + lck_rw_done(nd_if_rwlock); nd6_setmtu(ifp); #undef ND + + return 0; } /* @@ -194,73 +352,60 @@ nd6_ifattach(ifp) * changes, which means we might have to adjust the ND level MTU. */ void -nd6_setmtu(ifp) - struct ifnet *ifp; +nd6_setmtu(struct ifnet *ifp) { -#ifndef MIN -#define MIN(a,b) ((a) < (b) ? (a) : (b)) -#endif - struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index]; - u_long oldmaxmtu = ndi->maxmtu; - u_long oldlinkmtu = ndi->linkmtu; + struct nd_ifinfo *ndi; + u_int32_t oldmaxmtu, maxmtu; - switch (ifp->if_type) { - case IFT_ARCNET: /* XXX MTU handling needs more work */ - ndi->maxmtu = MIN(60480, ifp->if_mtu); - break; - case IFT_ETHER: - ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu); - break; - case IFT_FDDI: - ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); - break; - case IFT_ATM: - ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu); - break; - case IFT_IEEE1394: /* XXX should be IEEE1394MTU(1500) */ - ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu); - break; -#if IFT_IEEE80211 - case IFT_IEEE80211: /* XXX should be IEEE80211MTU(1500) */ - ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu); - break; -#endif - default: - ndi->maxmtu = ifp->if_mtu; - break; + /* + * Make sure IPv6 is enabled for the interface first, + * because this can be called directly from SIOCSIFMTU for IPv4 + */ + lck_rw_lock_shared(nd_if_rwlock); + if (ifp->if_index >= nd_ifinfo_indexlim) { + lck_rw_done(nd_if_rwlock); + return; /* we're out of bound for nd_ifinfo */ } - if (oldmaxmtu != ndi->maxmtu) { - /* - * If the ND level MTU is not set yet, or if the maxmtu - * is reset to a smaller value than the ND level MTU, - * also reset the ND level MTU. - */ - if (ndi->linkmtu == 0 || - ndi->maxmtu < ndi->linkmtu) { - ndi->linkmtu = ndi->maxmtu; - /* also adjust in6_maxmtu if necessary. */ - if (oldlinkmtu == 0) { - /* - * XXX: the case analysis is grotty, but - * it is not efficient to call in6_setmaxmtu() - * here when we are during the initialization - * procedure. - */ - if (in6_maxmtu < ndi->linkmtu) - in6_maxmtu = ndi->linkmtu; - } else - in6_setmaxmtu(); - } + ndi = &nd_ifinfo[ifp->if_index]; + oldmaxmtu = ndi->maxmtu; + + /* + * The ND level maxmtu is somewhat redundant to the interface MTU + * and is an implementation artifact of KAME. Instead of hard- + * limiting the maxmtu based on the interface type here, we simply + * take the if_mtu value since SIOCSIFMTU would have taken care of + * the sanity checks related to the maximum MTU allowed for the + * interface (a value that is known only by the interface layer), + * by sending the request down via ifnet_ioctl(). The use of the + * ND level maxmtu and linkmtu (the latter obtained via RA) are done + * via IN6_LINKMTU() which does further checking against if_mtu. + */ + maxmtu = ndi->maxmtu = ifp->if_mtu; + + /* + * Decreasing the interface MTU under IPV6 minimum MTU may cause + * undesirable situation. We thus notify the operator of the change + * explicitly. The check for oldmaxmtu is necessary to restrict the + * log to the case of changing the MTU, not initializing it. + */ + if (oldmaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { + log(LOG_NOTICE, "nd6_setmtu: " + "new link MTU on %s%d (%u) is too small for IPv6\n", + ifp->if_name, ifp->if_unit, (uint32_t)ndi->maxmtu); } -#undef MIN + lck_rw_done(nd_if_rwlock); + + /* also adjust in6_maxmtu if necessary. */ + if (maxmtu > in6_maxmtu) + in6_setmaxmtu(); } void -nd6_option_init(opt, icmp6len, ndopts) - void *opt; - int icmp6len; - union nd_opts *ndopts; +nd6_option_init( + void *opt, + int icmp6len, + union nd_opts *ndopts) { bzero(ndopts, sizeof(*ndopts)); ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; @@ -277,8 +422,8 @@ nd6_option_init(opt, icmp6len, ndopts) * Take one ND option. */ struct nd_opt_hdr * -nd6_option(ndopts) - union nd_opts *ndopts; +nd6_option( + union nd_opts *ndopts) { struct nd_opt_hdr *nd_opt; int olen; @@ -329,8 +474,8 @@ nd6_option(ndopts) * multiple options of the same type. */ int -nd6_options(ndopts) - union nd_opts *ndopts; +nd6_options( + union nd_opts *ndopts) { struct nd_opt_hdr *nd_opt; int i = 0; @@ -405,150 +550,228 @@ skip1: return 0; } -/* - * ND6 timer routine to expire default route list and prefix list - */ -void -nd6_timer_funneled(ignored_arg) - void *ignored_arg; -{ -#ifdef __APPLE__ - boolean_t funnel_state; - funnel_state = thread_funnel_set(network_flock, TRUE); -#endif - nd6_timer(ignored_arg); -#ifdef __APPLE__ - (void) thread_funnel_set(network_flock, FALSE); -#endif -} void -nd6_timer(ignored_arg) - void *ignored_arg; +nd6_drain(__unused void *ignored_arg) { - int s; struct llinfo_nd6 *ln; struct nd_defrouter *dr; struct nd_prefix *pr; - struct ifnet *ifp; + struct ifnet *ifp = NULL; struct in6_ifaddr *ia6, *nia6; struct in6_addrlifetime *lt6; - - s = splnet(); - - timeout(nd6_timer_funneled, (caddr_t)0, nd6_prune * hz); + struct timeval timenow; + getmicrotime(&timenow); +again: + /* + * The global list llinfo_nd6 is modified by nd6_request() and is + * therefore protected by rnh_lock. For obvious reasons, we cannot + * hold rnh_lock across calls that might lead to code paths which + * attempt to acquire rnh_lock, else we deadlock. Hence for such + * cases we drop rt_lock and rnh_lock, make the calls, and repeat the + * loop. To ensure that we don't process the same entry more than + * once in a single timeout, we mark the "already-seen" entries with + * ND6_LNF_TIMER_SKIP flag. At the end of the loop, we do a second + * pass thru the entries and clear the flag so they can be processed + * during the next timeout. + */ + lck_mtx_lock(rnh_lock); ln = llinfo_nd6.ln_next; - /* XXX BSD/OS separates this code -- itojun */ - while (ln && ln != &llinfo_nd6) { + while (ln != NULL && ln != &llinfo_nd6) { struct rtentry *rt; struct sockaddr_in6 *dst; - struct llinfo_nd6 *next = ln->ln_next; - /* XXX: used for the DELAY case only: */ - struct nd_ifinfo *ndi = NULL; + struct llinfo_nd6 *next; + struct nd_ifinfo ndi; + + /* ln_next/prev/rt is protected by rnh_lock */ + next = ln->ln_next; + rt = ln->ln_rt; + RT_LOCK(rt); - if ((rt = ln->ln_rt) == NULL) { + /* We've seen this already; skip it */ + if (ln->ln_flags & ND6_LNF_TIMER_SKIP) { + RT_UNLOCK(rt); ln = next; continue; } + + /* rt->rt_ifp should never be NULL */ if ((ifp = rt->rt_ifp) == NULL) { + panic("%s: ln(%p) rt(%p) rt_ifp == NULL", __func__, + ln, rt); + /* NOTREACHED */ + } + + /* rt_llinfo must always be equal to ln */ + if ((struct llinfo_nd6 *)rt->rt_llinfo != ln) { + panic("%s: rt_llinfo(%p) is not equal to ln(%p)", + __func__, rt->rt_llinfo, ln); + /* NOTREACHED */ + } + + /* rt_key should never be NULL */ + dst = (struct sockaddr_in6 *)rt_key(rt); + if (dst == NULL) { + panic("%s: rt(%p) key is NULL ln(%p)", __func__, + rt, ln); + /* NOTREACHED */ + } + + /* Set the flag in case we jump to "again" */ + ln->ln_flags |= ND6_LNF_TIMER_SKIP; + + if (ln->ln_expire > timenow.tv_sec) { + RT_UNLOCK(rt); ln = next; continue; } - ndi = &nd_ifinfo[ifp->if_index]; - dst = (struct sockaddr_in6 *)rt_key(rt); - if (ln->ln_expire > time_second) { + /* Make a copy (we're using it read-only anyway) */ + lck_rw_lock_shared(nd_if_rwlock); + if (ifp->if_index >= nd_ifinfo_indexlim) { + lck_rw_done(nd_if_rwlock); + RT_UNLOCK(rt); ln = next; continue; } - - /* sanity check */ - if (!rt) - panic("rt=0 in nd6_timer(ln=%p)\n", ln); - if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln) - panic("rt_llinfo(%p) is not equal to ln(%p)\n", - rt->rt_llinfo, ln); - if (!dst) - panic("dst=0 in nd6_timer(ln=%p)\n", ln); + ndi = nd_ifinfo[ifp->if_index]; + lck_rw_done(nd_if_rwlock); + + RT_LOCK_ASSERT_HELD(rt); switch (ln->ln_state) { case ND6_LLINFO_INCOMPLETE: if (ln->ln_asked < nd6_mmaxtries) { ln->ln_asked++; - ln->ln_expire = time_second + - nd_ifinfo[ifp->if_index].retrans / 1000; + ln->ln_expire = timenow.tv_sec + + ndi.retrans / 1000; + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); nd6_ns_output(ifp, NULL, &dst->sin6_addr, - ln, 0); + ln, 0, 0); + RT_REMREF(rt); } else { struct mbuf *m = ln->ln_hold; - if (m) { - if (rt->rt_ifp) { - /* - * Fake rcvif to make ICMP error - * more helpful in diagnosing - * for the receiver. - * XXX: should we consider - * older rcvif? - */ - m->m_pkthdr.rcvif = rt->rt_ifp; - } + ln->ln_hold = NULL; + if (m != NULL) { + /* + * Fake rcvif to make ICMP error + * more helpful in diagnosing + * for the receiver. + * XXX: should we consider + * older rcvif? + */ + m->m_pkthdr.rcvif = ifp; + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 0); - ln->ln_hold = NULL; + } else { + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); } - next = nd6_free(rt); + nd6_free(rt); } - break; + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + goto again; + case ND6_LLINFO_REACHABLE: if (ln->ln_expire) { ln->ln_state = ND6_LLINFO_STALE; - ln->ln_expire = time_second + nd6_gctimer; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, + nd6_gctimer); } + RT_UNLOCK(rt); break; case ND6_LLINFO_STALE: + case ND6_LLINFO_PURGE: /* Garbage Collection(RFC 2461 5.3) */ - if (ln->ln_expire) - next = nd6_free(rt); + if (ln->ln_expire) { + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); + nd6_free(rt); + lck_mtx_assert(rnh_lock, + LCK_MTX_ASSERT_NOTOWNED); + goto again; + } else { + RT_UNLOCK(rt); + } break; case ND6_LLINFO_DELAY: - if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { + if ((ndi.flags & ND6_IFF_PERFORMNUD) != 0) { /* We need NUD */ ln->ln_asked = 1; ln->ln_state = ND6_LLINFO_PROBE; - ln->ln_expire = time_second + - ndi->retrans / 1000; + ln->ln_expire = timenow.tv_sec + + ndi.retrans / 1000; + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); nd6_ns_output(ifp, &dst->sin6_addr, - &dst->sin6_addr, - ln, 0); - } else { - ln->ln_state = ND6_LLINFO_STALE; /* XXX */ - ln->ln_expire = time_second + nd6_gctimer; + &dst->sin6_addr, ln, 0, 0); + lck_mtx_assert(rnh_lock, + LCK_MTX_ASSERT_NOTOWNED); + RT_REMREF(rt); + goto again; } + ln->ln_state = ND6_LLINFO_STALE; /* XXX */ + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, + nd6_gctimer); + RT_UNLOCK(rt); break; + case ND6_LLINFO_PROBE: if (ln->ln_asked < nd6_umaxtries) { ln->ln_asked++; - ln->ln_expire = time_second + - nd_ifinfo[ifp->if_index].retrans / 1000; + ln->ln_expire = timenow.tv_sec + + ndi.retrans / 1000; + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); nd6_ns_output(ifp, &dst->sin6_addr, - &dst->sin6_addr, ln, 0); + &dst->sin6_addr, ln, 0, 0); + RT_REMREF(rt); } else { - next = nd6_free(rt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); + nd6_free(rt); } + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + goto again; + + default: + RT_UNLOCK(rt); break; } ln = next; } - + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); + + /* Now clear the flag from all entries */ + ln = llinfo_nd6.ln_next; + while (ln != NULL && ln != &llinfo_nd6) { + struct rtentry *rt = ln->ln_rt; + struct llinfo_nd6 *next = ln->ln_next; + + RT_LOCK_SPIN(rt); + if (ln->ln_flags & ND6_LNF_TIMER_SKIP) + ln->ln_flags &= ~ND6_LNF_TIMER_SKIP; + RT_UNLOCK(rt); + ln = next; + } + lck_mtx_unlock(rnh_lock); + /* expire default router list */ + lck_mtx_lock(nd6_mutex); dr = TAILQ_FIRST(&nd_defrouter); while (dr) { - if (dr->expire && dr->expire < time_second) { + if (dr->expire && dr->expire < timenow.tv_sec) { struct nd_defrouter *t; t = TAILQ_NEXT(dr, dr_entry); - defrtrlist_del(dr); + defrtrlist_del(dr, 1); dr = t; } else { dr = TAILQ_NEXT(dr, dr_entry); @@ -562,17 +785,25 @@ nd6_timer(ignored_arg) * rather separate address lifetimes and prefix lifetimes. */ addrloop: - for (ia6 = in6_ifaddr; ia6; ia6 = nia6) { + for (ia6 = in6_ifaddrs; ia6; ia6 = nia6) { nia6 = ia6->ia_next; /* check address lifetime */ lt6 = &ia6->ia6_lifetime; if (IFA6_IS_INVALID(ia6)) { int regen = 0; + /* + * Extra reference for ourselves; it's no-op if + * we don't have to regenerate temporary address, + * otherwise it protects the address from going + * away since we drop nd6_mutex below. + */ + ifaref(&ia6->ia_ifa); + /* * If the expiring address is temporary, try * regenerating a new one. This would be useful when - * we suspended a laptop PC, then turned on after a + * we suspended a laptop PC, then turned it on after a * period that could invalidate all temporary * addresses. Although we may have to restart the * loop (see below), it must be after purging the @@ -581,15 +812,27 @@ nd6_timer(ignored_arg) */ if (ip6_use_tempaddr && (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { - if (regen_tmpaddr(ia6) == 0) + /* NOTE: We have to drop the lock here because + * regen_tmpaddr() eventually calls in6_update_ifa(), + * which must take the lock and would otherwise cause a + * hang. This is safe because the goto addrloop + * leads to a reevaluation of the in6_ifaddrs list + */ + lck_mtx_unlock(nd6_mutex); + if (regen_tmpaddr(ia6) == 0) regen = 1; + lck_mtx_lock(nd6_mutex); } - in6_purgeaddr(&ia6->ia_ifa); + in6_purgeaddr(&ia6->ia_ifa, 1); + + /* Release extra reference taken above */ + ifafree(&ia6->ia_ifa); if (regen) goto addrloop; /* XXX: see below */ - } else if (IFA6_IS_DEPRECATED(ia6)) { + } + if (IFA6_IS_DEPRECATED(ia6)) { int oldflags = ia6->ia6_flags; ia6->ia6_flags |= IN6_IFF_DEPRECATED; @@ -602,6 +845,8 @@ nd6_timer(ignored_arg) (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && (oldflags & IN6_IFF_DEPRECATED) == 0) { + /* see NOTE above */ + lck_mtx_unlock(nd6_mutex); if (regen_tmpaddr(ia6) == 0) { /* * A new temporary address is @@ -610,15 +855,17 @@ nd6_timer(ignored_arg) * has changed while we are still in * the loop. Although the change * would not cause disaster (because - * it's not an addition, but a - * deletion,) we'd rather restart the + * it's not a deletion, but an + * addition,) we'd rather restart the * loop just for safety. Or does this * significantly reduce performance?? */ + lck_mtx_lock(nd6_mutex); goto addrloop; } + lck_mtx_lock(nd6_mutex); } - } else if (IFA6_IS_DEPRECATED(ia6)) { + } else { /* * A new RA might have made a deprecated address * preferred. @@ -634,16 +881,8 @@ nd6_timer(ignored_arg) * check prefix lifetime. * since pltime is just for autoconf, pltime processing for * prefix is not necessary. - * - * we offset expire time by NDPR_KEEP_EXPIRE, so that we - * can use the old prefix information to validate the - * next prefix information to come. See prelist_update() - * for actual validation. - * - * I don't think such an offset is necessary. - * (jinmei@kame.net, 20010130). */ - if (pr->ndpr_expire && pr->ndpr_expire < time_second) { + if (pr->ndpr_expire && pr->ndpr_expire < timenow.tv_sec) { struct nd_prefix *t; t = pr->ndpr_next; @@ -652,23 +891,37 @@ nd6_timer(ignored_arg) * separate. NEVER perform in6_purgeaddr here. */ - prelist_remove(pr); + prelist_remove(pr, 1); pr = t; } else pr = pr->ndpr_next; } - splx(s); + lck_mtx_unlock(nd6_mutex); +} + +/* + * ND6 timer routine to expire default route list and prefix list + */ +void +nd6_timer(__unused void *ignored_arg) +{ + nd6_drain(NULL); + timeout(nd6_timer, (caddr_t)0, nd6_prune * hz); } static int -regen_tmpaddr(ia6) - struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */ +regen_tmpaddr( + struct in6_ifaddr *ia6) /* deprecated/invalidated temporary address */ { struct ifaddr *ifa; struct ifnet *ifp; struct in6_ifaddr *public_ifa6 = NULL; + struct timeval timenow; + + getmicrotime(&timenow); ifp = ia6->ia_ifa.ifa_ifp; + ifnet_lock_exclusive(ifp); for (ifa = ifp->if_addrlist.tqh_first; ifa; ifa = ifa->ifa_list.tqe_next) { @@ -709,11 +962,12 @@ regen_tmpaddr(ia6) if (!IFA6_IS_DEPRECATED(it6)) public_ifa6 = it6; } + ifnet_lock_done(ifp); if (public_ifa6 != NULL) { int e; - if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) { + if ((e = in6_tmpifadd(public_ifa6, 0, M_WAITOK)) != 0) { log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" " tmp addr,errno=%d\n", e); return(-1); @@ -729,14 +983,15 @@ regen_tmpaddr(ia6) * ifp goes away. */ void -nd6_purge(ifp) - struct ifnet *ifp; +nd6_purge( + struct ifnet *ifp) { - struct llinfo_nd6 *ln, *nln; + struct llinfo_nd6 *ln; struct nd_defrouter *dr, *ndr, drany; struct nd_prefix *pr, *npr; /* Nuke default router list entries toward ifp */ + lck_mtx_lock(nd6_mutex); if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) { /* * The first entry of the list may be stored in @@ -745,11 +1000,11 @@ nd6_purge(ifp) for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) { ndr = TAILQ_NEXT(dr, dr_entry); if (dr->ifp == ifp) - defrtrlist_del(dr); + defrtrlist_del(dr, 1); } dr = TAILQ_FIRST(&nd_defrouter); if (dr->ifp == ifp) - defrtrlist_del(dr); + defrtrlist_del(dr, 1); } /* Nuke prefix list entries toward ifp */ @@ -764,20 +1019,27 @@ nd6_purge(ifp) * by itself. * (jinmei@kame.net 20010129) */ - prelist_remove(pr); + prelist_remove(pr, 1); } } /* cancel default outgoing interface setting */ - if (nd6_defifindex == ifp->if_index) + if (nd6_defifindex == ifp->if_index) { + /* Release nd6_mutex as it will be acquired + * during nd6_setdefaultiface again + */ + lck_mtx_unlock(nd6_mutex); nd6_setdefaultiface(0); + lck_mtx_lock(nd6_mutex); + } - if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */ + if (!ip6_forwarding && (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { /* refresh default router list */ bzero(&drany, sizeof(drany)); defrouter_delreq(&drany, 0); defrouter_select(); } + lck_mtx_unlock(nd6_mutex); /* * Nuke neighbor cache entries for the ifp. @@ -785,28 +1047,54 @@ nd6_purge(ifp) * due to KAME goto ours hack. See RTM_RESOLVE case in * nd6_rtrequest(), and ip6_input(). */ +again: + lck_mtx_lock(rnh_lock); ln = llinfo_nd6.ln_next; - while (ln && ln != &llinfo_nd6) { + while (ln != NULL && ln != &llinfo_nd6) { struct rtentry *rt; - struct sockaddr_dl *sdl; + struct llinfo_nd6 *nln; nln = ln->ln_next; rt = ln->ln_rt; - if (rt && rt->rt_gateway && - rt->rt_gateway->sa_family == AF_LINK) { - sdl = (struct sockaddr_dl *)rt->rt_gateway; - if (sdl->sdl_index == ifp->if_index) - nln = nd6_free(rt); + RT_LOCK(rt); + if (rt->rt_gateway != NULL && + rt->rt_gateway->sa_family == AF_LINK && + SDL(rt->rt_gateway)->sdl_index == ifp->if_index) { + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); + /* + * See comments on nd6_timer() for reasons why + * this loop is repeated; we bite the costs of + * going thru the same llinfo_nd6 more than once + * here, since this purge happens during detach, + * and that unlike the timer case, it's possible + * there's more than one purges happening at the + * same time (thus a flag wouldn't buy anything). + */ + nd6_free(rt); + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + goto again; + } else { + RT_UNLOCK(rt); } ln = nln; } + lck_mtx_unlock(rnh_lock); } +/* + * Upon success, the returned route will be locked and the caller is + * responsible for releasing the reference and doing RT_UNLOCK(rt). + * This routine does not require rnh_lock to be held by the caller, + * although it needs to be indicated of such a case in order to call + * the correct variant of the relevant routing routines. + */ struct rtentry * -nd6_lookup(addr6, create, ifp) - struct in6_addr *addr6; - int create; - struct ifnet *ifp; +nd6_lookup( + struct in6_addr *addr6, + int create, + struct ifnet *ifp, + int rt_locked) { struct rtentry *rt; struct sockaddr_in6 sin6; @@ -818,21 +1106,34 @@ nd6_lookup(addr6, create, ifp) #if SCOPEDROUTING sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6); #endif - rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL); - if (rt && (rt->rt_flags & RTF_LLINFO) == 0) { - /* - * This is the case for the default route. - * If we want to create a neighbor cache for the address, we - * should free the route for the destination and allocate an - * interface route. - */ - if (create) { - rtfree(rt); - rt = 0; + if (rt_locked) + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); + + rt = rt_locked ? rtalloc1_locked((struct sockaddr *)&sin6, create, 0) : + rtalloc1((struct sockaddr *)&sin6, create, 0); + + if (rt != NULL) { + RT_LOCK(rt); + if ((rt->rt_flags & RTF_LLINFO) == 0) { + /* + * This is the case for the default route. If we + * want to create a neighbor cache for the address, + * we should free the route for the destination and + * allocate an interface route. + */ + if (create) { + RT_UNLOCK(rt); + if (rt_locked) + rtfree_locked(rt); + else + rtfree(rt); + rt = NULL; + } } } - if (!rt) { + if (rt == NULL) { if (create && ifp) { + struct ifaddr *ifa; int e; /* @@ -842,56 +1143,74 @@ nd6_lookup(addr6, create, ifp) * This hack is necessary for a neighbor which can't * be covered by our own prefix. */ - struct ifaddr *ifa = - ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); + ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); if (ifa == NULL) return(NULL); /* - * Create a new route. RTF_LLINFO is necessary + * Create a new route. RTF_LLINFO is necessary * to create a Neighbor Cache entry for the * destination in nd6_rtrequest which will be - * called in rtequest via ifa->ifa_rtrequest. + * called in rtrequest via ifa->ifa_rtrequest. */ - if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6, - ifa->ifa_addr, - (struct sockaddr *)&all1_sa, - (ifa->ifa_flags | - RTF_HOST | RTF_LLINFO) & - ~RTF_CLONING, - &rt)) != 0) - log(LOG_ERR, - "nd6_lookup: failed to add route for a " - "neighbor(%s), errno=%d\n", - ip6_sprintf(addr6), e); + if (!rt_locked) + lck_mtx_lock(rnh_lock); + if ((e = rtrequest_locked(RTM_ADD, + (struct sockaddr *)&sin6, ifa->ifa_addr, + (struct sockaddr *)&all1_sa, + (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) & + ~RTF_CLONING, &rt)) != 0) { + if (e != EEXIST) + log(LOG_ERR, "%s: failed to add route " + "for a neighbor(%s), errno=%d\n", + __func__, ip6_sprintf(addr6), e); + } + if (!rt_locked) + lck_mtx_unlock(rnh_lock); + ifafree(ifa); if (rt == NULL) return(NULL); + + RT_LOCK(rt); if (rt->rt_llinfo) { - struct llinfo_nd6 *ln = - (struct llinfo_nd6 *)rt->rt_llinfo; + struct llinfo_nd6 *ln = rt->rt_llinfo; ln->ln_state = ND6_LLINFO_NOSTATE; } - } else + } else { return(NULL); + } } - rtunref(rt); + RT_LOCK_ASSERT_HELD(rt); /* * Validation for the entry. + * Note that the check for rt_llinfo is necessary because a cloned + * route from a parent route that has the L flag (e.g. the default + * route to a p2p interface) may have the flag, too, while the + * destination is not actually a neighbor. * XXX: we can't use rt->rt_ifp to check for the interface, since * it might be the loopback interface if the entry is for our * own address on a non-loopback interface. Instead, we should - * use rt->rt_ifa->ifa_ifp, which would specify the REAL interface. + * use rt->rt_ifa->ifa_ifp, which would specify the REAL + * interface. */ - if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || - rt->rt_gateway->sa_family != AF_LINK || + if (ifp == NULL || (ifp->if_type == IFT_PPP) || + (ifp->if_eflags & IFEF_NOAUTOIPV6LL) || + (rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || + rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL || (ifp && rt->rt_ifa->ifa_ifp != ifp)) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); if (create) { - log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n", - ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec"); + log(LOG_DEBUG, "%s: failed to lookup %s " + "(if = %s)\n", __func__, ip6_sprintf(addr6), + ifp ? if_name(ifp) : "unspec"); /* xxx more logs... kazu */ } - return(0); + return(NULL); } + /* + * Caller needs to release reference and call RT_UNLOCK(rt). + */ return(rt); } @@ -900,11 +1219,13 @@ nd6_lookup(addr6, create, ifp) * XXX: should take care of the destination of a p2p link? */ int -nd6_is_addr_neighbor(addr, ifp) - struct sockaddr_in6 *addr; - struct ifnet *ifp; +nd6_is_addr_neighbor( + struct sockaddr_in6 *addr, + struct ifnet *ifp, + int rt_locked) { struct ifaddr *ifa; + struct rtentry *rt; int i; #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr) @@ -923,28 +1244,36 @@ nd6_is_addr_neighbor(addr, ifp) * If the address matches one of our addresses, * it should be a neighbor. */ + ifnet_lock_shared(ifp); for (ifa = ifp->if_addrlist.tqh_first; ifa; ifa = ifa->ifa_list.tqe_next) { if (ifa->ifa_addr->sa_family != AF_INET6) - next: continue; + continue; for (i = 0; i < 4; i++) { if ((IFADDR6(ifa).s6_addr32[i] ^ addr->sin6_addr.s6_addr32[i]) & - IFMASK6(ifa).s6_addr32[i]) - goto next; + IFMASK6(ifa).s6_addr32[i]) + continue; } + ifnet_lock_done(ifp); return(1); } + ifnet_lock_done(ifp); /* * Even if the address matches none of our addresses, it might be - * in the neighbor cache. + * in the neighbor cache. Callee returns a locked route upon + * success. */ - if (nd6_lookup(&addr->sin6_addr, 0, ifp)) + if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp, rt_locked)) != NULL) { + RT_LOCK_ASSERT_HELD(rt); + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); return(1); + } return(0); #undef IFADDR6 @@ -954,32 +1283,50 @@ nd6_is_addr_neighbor(addr, ifp) /* * Free an nd6 llinfo entry. */ -struct llinfo_nd6 * -nd6_free(rt) - struct rtentry *rt; +void +nd6_free( + struct rtentry *rt) { - struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next; - struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; + struct llinfo_nd6 *ln; + struct in6_addr in6; struct nd_defrouter *dr; + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + RT_LOCK_ASSERT_NOTHELD(rt); + lck_mtx_lock(nd6_mutex); + + RT_LOCK(rt); + RT_ADDREF_LOCKED(rt); /* Extra ref */ + ln = rt->rt_llinfo; + in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; + + /* + * Prevent another thread from modifying rt_key, rt_gateway + * via rt_setgate() after the rt_lock is dropped by marking + * the route as defunct. + */ + rt->rt_flags |= RTF_CONDEMNED; + /* * we used to have pfctlinput(PRC_HOSTDEAD) here. * even though it is not harmful, it was not really necessary. */ - if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */ - int s; - s = splnet(); - dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr, - rt->rt_ifp); + if (!ip6_forwarding && (ip6_accept_rtadv || + (rt->rt_ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { + dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))-> + sin6_addr, rt->rt_ifp); - if (ln->ln_router || dr) { + if ((ln && ln->ln_router) || dr) { /* * rt6_flush must be called whether or not the neighbor * is in the Default Router List. * See a corresponding comment in nd6_na_input(). */ + RT_UNLOCK(rt); rt6_flush(&in6, rt->rt_ifp); + } else { + RT_UNLOCK(rt); } if (dr) { @@ -992,10 +1339,11 @@ nd6_free(rt) /* * Temporarily fake the state to choose a new default * router and to perform on-link determination of - * prefixes coreectly. + * prefixes correctly. * Below the state will be set correctly, * or the entry itself will be deleted. */ + RT_LOCK_SPIN(rt); ln->ln_state = ND6_LLINFO_INCOMPLETE; /* @@ -1004,7 +1352,8 @@ nd6_free(rt) * before the default router selection, we perform * the check now. */ - pfxlist_onlink_check(); + RT_UNLOCK(rt); + pfxlist_onlink_check(1); if (dr == TAILQ_FIRST(&nd_defrouter)) { /* @@ -1020,26 +1369,22 @@ nd6_free(rt) defrouter_select(); } } - splx(s); + RT_LOCK_ASSERT_NOTHELD(rt); + } else { + RT_UNLOCK(rt); } - /* - * Before deleting the entry, remember the next entry as the - * return value. We need this because pfxlist_onlink_check() above - * might have freed other entries (particularly the old next entry) as - * a side effect (XXX). - */ - next = ln->ln_next; - + lck_mtx_unlock(nd6_mutex); /* * Detach the route from the routing tree and the list of neighbor * caches, and disable the route entry not to be used in already * cached routes. */ - rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0, + (void) rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0, rt_mask(rt), 0, (struct rtentry **)0); - return(next); + /* Extra ref held above; now free it */ + rtfree(rt); } /* @@ -1048,12 +1393,15 @@ nd6_free(rt) * XXX cost-effective metods? */ void -nd6_nud_hint(rt, dst6, force) - struct rtentry *rt; - struct in6_addr *dst6; - int force; +nd6_nud_hint( + struct rtentry *rt, + struct in6_addr *dst6, + int force) { struct llinfo_nd6 *ln; + struct timeval timenow; + + getmicrotime(&timenow); /* * If the caller specified "rt", use that. Otherwise, resolve the @@ -1062,8 +1410,13 @@ nd6_nud_hint(rt, dst6, force) if (!rt) { if (!dst6) return; - if (!(rt = nd6_lookup(dst6, 0, NULL))) + /* Callee returns a locked route upon success */ + if ((rt = nd6_lookup(dst6, 0, NULL, 0)) == NULL) return; + RT_LOCK_ASSERT_HELD(rt); + } else { + RT_LOCK(rt); + RT_ADDREF_LOCKED(rt); } if ((rt->rt_flags & RTF_GATEWAY) != 0 || @@ -1071,12 +1424,12 @@ nd6_nud_hint(rt, dst6, force) !rt->rt_llinfo || !rt->rt_gateway || rt->rt_gateway->sa_family != AF_LINK) { /* This is not a host route. */ - return; + goto done; } - ln = (struct llinfo_nd6 *)rt->rt_llinfo; + ln = rt->rt_llinfo; if (ln->ln_state < ND6_LLINFO_REACHABLE) - return; + goto done; /* * if we get upper-layer reachability confirmation many times, @@ -1085,28 +1438,39 @@ nd6_nud_hint(rt, dst6, force) if (!force) { ln->ln_byhint++; if (ln->ln_byhint > nd6_maxnudhint) - return; + goto done; } ln->ln_state = ND6_LLINFO_REACHABLE; - if (ln->ln_expire) - ln->ln_expire = time_second + - nd_ifinfo[rt->rt_ifp->if_index].reachable; + if (ln->ln_expire) { + lck_rw_lock_shared(nd_if_rwlock); + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, + nd_ifinfo[rt->rt_ifp->if_index].reachable); + lck_rw_done(nd_if_rwlock); + } +done: + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); } void -nd6_rtrequest(req, rt, sa) - int req; - struct rtentry *rt; - struct sockaddr *sa; /* xxx unused */ +nd6_rtrequest( + int req, + struct rtentry *rt, + __unused struct sockaddr *sa) { struct sockaddr *gate = rt->rt_gateway; - struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo; - static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; + struct llinfo_nd6 *ln = rt->rt_llinfo; + static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK, 0, 0, 0, 0, 0, + {0,0,0,0,0,0,0,0,0,0,0,0,} }; struct ifnet *ifp = rt->rt_ifp; struct ifaddr *ifa; + struct timeval timenow; - if (rt->rt_flags & RTF_GATEWAY) + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); + RT_LOCK_ASSERT_HELD(rt); + + if ((rt->rt_flags & RTF_GATEWAY)) return; if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) { @@ -1120,6 +1484,48 @@ nd6_rtrequest(req, rt, sa) return; } + if (req == RTM_RESOLVE) { + int no_nd_cache; + + if (!nd6_need_cache(ifp)) { /* stf case */ + no_nd_cache = 1; + } else { + /* + * nd6_is_addr_neighbor() may call nd6_lookup(), + * therefore we drop rt_lock to avoid deadlock + * during the lookup. Using rt_key(rt) is still + * safe because it won't change while rnh_lock + * is held. + */ + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + no_nd_cache = !nd6_is_addr_neighbor( + (struct sockaddr_in6 *)rt_key(rt), ifp, 1); + RT_LOCK(rt); + RT_REMREF_LOCKED(rt); + } + + /* + * FreeBSD and BSD/OS often make a cloned host route based + * on a less-specific route (e.g. the default route). + * If the less specific route does not have a "gateway" + * (this is the case when the route just goes to a p2p or an + * stf interface), we'll mistakenly make a neighbor cache for + * the host route, and will see strange neighbor solicitation + * for the corresponding destination. In order to avoid the + * confusion, we check if the destination of the route is + * a neighbor in terms of neighbor discovery, and stop the + * process if not. Additionally, we remove the LLINFO flag + * so that ndp(8) will not try to get the neighbor information + * of the destination. + */ + if (no_nd_cache) { + rt->rt_flags &= ~RTF_LLINFO; + return; + } + } + + getmicrotime(&timenow); switch (req) { case RTM_ADD: /* @@ -1132,28 +1538,23 @@ nd6_rtrequest(req, rt, sa) if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) { /* * Case 1: This route should come from - * a route to interface. RTF_LLINFO flag is set + * a route to interface. RTF_LLINFO flag is set * for a host route whose destination should be * treated as on-link. */ - rt_setgate(rt, rt_key(rt), - (struct sockaddr *)&null_sdl); - gate = rt->rt_gateway; - SDL(gate)->sdl_type = ifp->if_type; - SDL(gate)->sdl_index = ifp->if_index; - if (ln) - ln->ln_expire = time_second; -#if 1 - if (ln && ln->ln_expire == 0) { - /* kludge for desktops */ -#if 0 - printf("nd6_request: time.tv_sec is zero; " - "treat it as 1\n"); -#endif - ln->ln_expire = 1; + if (rt_setgate(rt, rt_key(rt), + (struct sockaddr *)&null_sdl) == 0) { + gate = rt->rt_gateway; + SDL(gate)->sdl_type = ifp->if_type; + SDL(gate)->sdl_index = ifp->if_index; + /* + * In case we're called before 1.0 sec. + * has elapsed. + */ + if (ln != NULL) + ln->ln_expire = MAX(timenow.tv_sec, 1); } -#endif - if (rt->rt_flags & RTF_CLONING) + if ((rt->rt_flags & RTF_CLONING)) break; } /* @@ -1206,12 +1607,13 @@ nd6_rtrequest(req, rt, sa) * Case 2: This route may come from cloning, or a manual route * add with a LL address. */ - R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln)); - rt->rt_llinfo = (caddr_t)ln; - if (!ln) { + rt->rt_llinfo = ln = nd6_llinfo_alloc(); + if (ln == NULL) { log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n"); break; } + rt->rt_llinfo_free = nd6_llinfo_free; + nd6_inuse++; nd6_allocated++; Bzero(ln, sizeof(*ln)); @@ -1231,13 +1633,44 @@ nd6_rtrequest(req, rt, sa) * initialized in rtrequest(), so rt_expire is 0. */ ln->ln_state = ND6_LLINFO_NOSTATE; - ln->ln_expire = time_second; + /* In case we're called before 1.0 sec. has elapsed */ + ln->ln_expire = MAX(timenow.tv_sec, 1); } rt->rt_flags |= RTF_LLINFO; - ln->ln_next = llinfo_nd6.ln_next; - llinfo_nd6.ln_next = ln; - ln->ln_prev = &llinfo_nd6; - ln->ln_next->ln_prev = ln; + LN_INSERTHEAD(ln); + + /* + * If we have too many cache entries, initiate immediate + * purging for some "less recently used" entries. Note that + * we cannot directly call nd6_free() here because it would + * cause re-entering rtable related routines triggering an LOR + * problem. + */ + if (ip6_neighborgcthresh >= 0 && + nd6_inuse >= ip6_neighborgcthresh) { + int i; + + for (i = 0; i < 10 && llinfo_nd6.ln_prev != ln; i++) { + struct llinfo_nd6 *ln_end = llinfo_nd6.ln_prev; + struct rtentry *rt_end = ln_end->ln_rt; + + /* Move this entry to the head */ + RT_LOCK(rt_end); + LN_DEQUEUE(ln_end); + LN_INSERTHEAD(ln_end); + + if (ln_end->ln_expire == 0) { + RT_UNLOCK(rt_end); + continue; + } + if (ln_end->ln_state > ND6_LLINFO_INCOMPLETE) + ln_end->ln_state = ND6_LLINFO_STALE; + else + ln_end->ln_state = ND6_LLINFO_PURGE; + ln_end->ln_expire = timenow.tv_sec; + RT_UNLOCK(rt_end); + } + } /* * check if rt_key(rt) is one of my address assigned @@ -1255,7 +1688,15 @@ nd6_rtrequest(req, rt, sa) SDL(gate)->sdl_alen = ifp->if_addrlen; } if (nd6_useloopback) { - rt->rt_ifp = &loif[0]; /*XXX*/ +#if IFNET_ROUTE_REFCNT + /* Adjust route ref count for the interfaces */ + if (rt->rt_if_ref_fn != NULL && + rt->rt_ifp != lo_ifp) { + rt->rt_if_ref_fn(lo_ifp, 1); + rt->rt_if_ref_fn(rt->rt_ifp, -1); + } +#endif /* IFNET_ROUTE_REFCNT */ + rt->rt_ifp = lo_ifp; /* XXX */ /* * Make sure rt_ifa be equal to the ifaddr * corresponding to the address. @@ -1268,6 +1709,7 @@ nd6_rtrequest(req, rt, sa) rtsetifa(rt, ifa); } } + ifafree(ifa); } else if (rt->rt_flags & RTF_ANNOUNCE) { ln->ln_expire = 0; ln->ln_state = ND6_LLINFO_REACHABLE; @@ -1285,7 +1727,7 @@ nd6_rtrequest(req, rt, sa) llsol.s6_addr32[2] = htonl(1); llsol.s6_addr8[12] = 0xff; - if (!in6_addmulti(&llsol, ifp, &error)) { + if (!in6_addmulti(&llsol, ifp, &error, 0)) { nd6log((LOG_ERR, "%s: failed to join " "%s (errno=%d)\n", if_name(ifp), ip6_sprintf(&llsol), error)); @@ -1310,172 +1752,288 @@ nd6_rtrequest(req, rt, sa) llsol.s6_addr32[2] = htonl(1); llsol.s6_addr8[12] = 0xff; + ifnet_lock_shared(ifp); IN6_LOOKUP_MULTI(llsol, ifp, in6m); + ifnet_lock_done(ifp); if (in6m) - in6_delmulti(in6m); + in6_delmulti(in6m, 0); } nd6_inuse--; - ln->ln_next->ln_prev = ln->ln_prev; - ln->ln_prev->ln_next = ln->ln_next; - ln->ln_prev = NULL; - rt->rt_llinfo = 0; + /* + * Unchain it but defer the actual freeing until the route + * itself is to be freed. rt->rt_llinfo still points to + * llinfo_nd6, and likewise, ln->ln_rt stil points to this + * route entry, except that RTF_LLINFO is now cleared. + */ + if (ln->ln_flags & ND6_LNF_IN_USE) + LN_DEQUEUE(ln); rt->rt_flags &= ~RTF_LLINFO; - if (ln->ln_hold) + if (ln->ln_hold != NULL) m_freem(ln->ln_hold); - Free((caddr_t)ln); + ln->ln_hold = NULL; } } -int -nd6_ioctl(cmd, data, ifp) - u_long cmd; - caddr_t data; - struct ifnet *ifp; +static void +nd6_siocgdrlst(void *data, int data_is_64) { - struct in6_drlist *drl = (struct in6_drlist *)data; - struct in6_prlist *prl = (struct in6_prlist *)data; - struct in6_ndireq *ndi = (struct in6_ndireq *)data; - struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; - struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; - struct nd_defrouter *dr, any; - struct nd_prefix *pr; - struct rtentry *rt; - int i = 0, error = 0; - int s; + struct in6_drlist_64 *drl_64 = (struct in6_drlist_64 *)data; + struct in6_drlist_32 *drl_32 = (struct in6_drlist_32 *)data; + struct nd_defrouter *dr; + int i = 0; - switch (cmd) { - case SIOCGDRLST_IN6: - /* - * obsolete API, use sysctl under net.inet6.icmp6 - */ - bzero(drl, sizeof(*drl)); - s = splnet(); - dr = TAILQ_FIRST(&nd_defrouter); + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); + + bzero(data, data_is_64 ? sizeof (*drl_64) : sizeof (*drl_32)); + dr = TAILQ_FIRST(&nd_defrouter); + if (data_is_64) { + /* For 64-bit process */ while (dr && i < DRLSTSIZ) { - drl->defrouter[i].rtaddr = dr->rtaddr; - if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) { + drl_64->defrouter[i].rtaddr = dr->rtaddr; + if (IN6_IS_ADDR_LINKLOCAL(&drl_64->defrouter[i].rtaddr)) { /* XXX: need to this hack for KAME stack */ - drl->defrouter[i].rtaddr.s6_addr16[1] = 0; - } else + drl_64->defrouter[i].rtaddr.s6_addr16[1] = 0; + } else { log(LOG_ERR, "default router list contains a " "non-linklocal address(%s)\n", - ip6_sprintf(&drl->defrouter[i].rtaddr)); - - drl->defrouter[i].flags = dr->flags; - drl->defrouter[i].rtlifetime = dr->rtlifetime; - drl->defrouter[i].expire = dr->expire; - drl->defrouter[i].if_index = dr->ifp->if_index; + ip6_sprintf(&drl_64->defrouter[i].rtaddr)); + } + drl_64->defrouter[i].flags = dr->flags; + drl_64->defrouter[i].rtlifetime = dr->rtlifetime; + drl_64->defrouter[i].expire = dr->expire; + drl_64->defrouter[i].if_index = dr->ifp->if_index; i++; dr = TAILQ_NEXT(dr, dr_entry); } - splx(s); - break; - case SIOCGPRLST_IN6: - /* - * obsolete API, use sysctl under net.inet6.icmp6 - */ - /* - * XXX meaning of fields, especialy "raflags", is very - * differnet between RA prefix list and RR/static prefix list. - * how about separating ioctls into two? - */ - bzero(prl, sizeof(*prl)); - s = splnet(); - pr = nd_prefix.lh_first; - while (pr && i < PRLSTSIZ) { + return; + } + /* For 32-bit process */ + while (dr && i < DRLSTSIZ) { + drl_32->defrouter[i].rtaddr = dr->rtaddr; + if (IN6_IS_ADDR_LINKLOCAL(&drl_32->defrouter[i].rtaddr)) { + /* XXX: need to this hack for KAME stack */ + drl_32->defrouter[i].rtaddr.s6_addr16[1] = 0; + } else { + log(LOG_ERR, + "default router list contains a " + "non-linklocal address(%s)\n", + ip6_sprintf(&drl_32->defrouter[i].rtaddr)); + } + drl_32->defrouter[i].flags = dr->flags; + drl_32->defrouter[i].rtlifetime = dr->rtlifetime; + drl_32->defrouter[i].expire = dr->expire; + drl_32->defrouter[i].if_index = dr->ifp->if_index; + i++; + dr = TAILQ_NEXT(dr, dr_entry); + } +} + +static void +nd6_siocgprlst(void *data, int data_is_64) +{ + struct in6_prlist_64 *prl_64 = (struct in6_prlist_64 *)data; + struct in6_prlist_32 *prl_32 = (struct in6_prlist_32 *)data; + struct nd_prefix *pr; + struct rr_prefix *rpp; + int i = 0; + + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); + /* + * XXX meaning of fields, especialy "raflags", is very + * differnet between RA prefix list and RR/static prefix list. + * how about separating ioctls into two? + */ + bzero(data, data_is_64 ? sizeof (*prl_64) : sizeof (*prl_32)); + pr = nd_prefix.lh_first; + if (data_is_64) { + /* For 64-bit process */ + while (pr && i < PRLSTSIZ) { struct nd_pfxrouter *pfr; int j; - (void)in6_embedscope(&prl->prefix[i].prefix, + (void) in6_embedscope(&prl_64->prefix[i].prefix, &pr->ndpr_prefix, NULL, NULL); - prl->prefix[i].raflags = pr->ndpr_raf; - prl->prefix[i].prefixlen = pr->ndpr_plen; - prl->prefix[i].vltime = pr->ndpr_vltime; - prl->prefix[i].pltime = pr->ndpr_pltime; - prl->prefix[i].if_index = pr->ndpr_ifp->if_index; - prl->prefix[i].expire = pr->ndpr_expire; + prl_64->prefix[i].raflags = pr->ndpr_raf; + prl_64->prefix[i].prefixlen = pr->ndpr_plen; + prl_64->prefix[i].vltime = pr->ndpr_vltime; + prl_64->prefix[i].pltime = pr->ndpr_pltime; + prl_64->prefix[i].if_index = pr->ndpr_ifp->if_index; + prl_64->prefix[i].expire = pr->ndpr_expire; pfr = pr->ndpr_advrtrs.lh_first; j = 0; while (pfr) { if (j < DRLSTSIZ) { -#define RTRADDR prl->prefix[i].advrtr[j] +#define RTRADDR prl_64->prefix[i].advrtr[j] RTRADDR = pfr->router->rtaddr; if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) { /* XXX: hack for KAME */ RTRADDR.s6_addr16[1] = 0; - } else + } else { log(LOG_ERR, "a router(%s) advertises " "a prefix with " "non-link local address\n", ip6_sprintf(&RTRADDR)); + } #undef RTRADDR } j++; pfr = pfr->pfr_next; } - prl->prefix[i].advrtrs = j; - prl->prefix[i].origin = PR_ORIG_RA; + prl_64->prefix[i].advrtrs = j; + prl_64->prefix[i].origin = PR_ORIG_RA; i++; pr = pr->ndpr_next; } - { - struct rr_prefix *rpp; for (rpp = LIST_FIRST(&rr_prefix); rpp; rpp = LIST_NEXT(rpp, rp_entry)) { if (i >= PRLSTSIZ) break; - (void)in6_embedscope(&prl->prefix[i].prefix, + (void) in6_embedscope(&prl_64->prefix[i].prefix, &pr->ndpr_prefix, NULL, NULL); - prl->prefix[i].raflags = rpp->rp_raf; - prl->prefix[i].prefixlen = rpp->rp_plen; - prl->prefix[i].vltime = rpp->rp_vltime; - prl->prefix[i].pltime = rpp->rp_pltime; - prl->prefix[i].if_index = rpp->rp_ifp->if_index; - prl->prefix[i].expire = rpp->rp_expire; - prl->prefix[i].advrtrs = 0; - prl->prefix[i].origin = rpp->rp_origin; + prl_64->prefix[i].raflags = rpp->rp_raf; + prl_64->prefix[i].prefixlen = rpp->rp_plen; + prl_64->prefix[i].vltime = rpp->rp_vltime; + prl_64->prefix[i].pltime = rpp->rp_pltime; + prl_64->prefix[i].if_index = rpp->rp_ifp->if_index; + prl_64->prefix[i].expire = rpp->rp_expire; + prl_64->prefix[i].advrtrs = 0; + prl_64->prefix[i].origin = rpp->rp_origin; i++; } - } - splx(s); + return; + } + /* For 32-bit process */ + while (pr && i < PRLSTSIZ) { + struct nd_pfxrouter *pfr; + int j; + + (void) in6_embedscope(&prl_32->prefix[i].prefix, + &pr->ndpr_prefix, NULL, NULL); + prl_32->prefix[i].raflags = pr->ndpr_raf; + prl_32->prefix[i].prefixlen = pr->ndpr_plen; + prl_32->prefix[i].vltime = pr->ndpr_vltime; + prl_32->prefix[i].pltime = pr->ndpr_pltime; + prl_32->prefix[i].if_index = pr->ndpr_ifp->if_index; + prl_32->prefix[i].expire = pr->ndpr_expire; + + pfr = pr->ndpr_advrtrs.lh_first; + j = 0; + while (pfr) { + if (j < DRLSTSIZ) { +#define RTRADDR prl_32->prefix[i].advrtr[j] + RTRADDR = pfr->router->rtaddr; + if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) { + /* XXX: hack for KAME */ + RTRADDR.s6_addr16[1] = 0; + } else { + log(LOG_ERR, + "a router(%s) advertises " + "a prefix with " + "non-link local address\n", + ip6_sprintf(&RTRADDR)); + } +#undef RTRADDR + } + j++; + pfr = pfr->pfr_next; + } + prl_32->prefix[i].advrtrs = j; + prl_32->prefix[i].origin = PR_ORIG_RA; - break; - case OSIOCGIFINFO_IN6: - if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { - error = EINVAL; + i++; + pr = pr->ndpr_next; + } + + for (rpp = LIST_FIRST(&rr_prefix); rpp; + rpp = LIST_NEXT(rpp, rp_entry)) { + if (i >= PRLSTSIZ) break; - } - ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu; - ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu; - ndi->ndi.basereachable = - nd_ifinfo[ifp->if_index].basereachable; - ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable; - ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans; - ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags; - ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm; - ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim; - ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra; + (void) in6_embedscope(&prl_32->prefix[i].prefix, + &pr->ndpr_prefix, NULL, NULL); + prl_32->prefix[i].raflags = rpp->rp_raf; + prl_32->prefix[i].prefixlen = rpp->rp_plen; + prl_32->prefix[i].vltime = rpp->rp_vltime; + prl_32->prefix[i].pltime = rpp->rp_pltime; + prl_32->prefix[i].if_index = rpp->rp_ifp->if_index; + prl_32->prefix[i].expire = rpp->rp_expire; + prl_32->prefix[i].advrtrs = 0; + prl_32->prefix[i].origin = rpp->rp_origin; + i++; + } +} + +int +nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) +{ + struct in6_ndireq *ndi = (struct in6_ndireq *)data; + struct in6_ondireq *ondi = (struct in6_ondireq *)data; + struct nd_defrouter *dr, any; + struct nd_prefix *pr; + struct rtentry *rt; + int i = ifp->if_index, error = 0; + + switch (cmd) { + case SIOCGDRLST_IN6_32: + case SIOCGDRLST_IN6_64: + /* + * obsolete API, use sysctl under net.inet6.icmp6 + */ + lck_mtx_lock(nd6_mutex); + nd6_siocgdrlst(data, cmd == SIOCGDRLST_IN6_64); + lck_mtx_unlock(nd6_mutex); + break; + + case SIOCGPRLST_IN6_32: + case SIOCGPRLST_IN6_64: + /* + * obsolete API, use sysctl under net.inet6.icmp6 + */ + lck_mtx_lock(nd6_mutex); + nd6_siocgprlst(data, cmd == SIOCGPRLST_IN6_64); + lck_mtx_unlock(nd6_mutex); break; + + case OSIOCGIFINFO_IN6: case SIOCGIFINFO_IN6: + /* + * SIOCGIFINFO_IN6 ioctl is encoded with in6_ondireq + * instead of in6_ndireq, so we treat it as such. + */ + lck_rw_lock_shared(nd_if_rwlock); if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { + lck_rw_done(nd_if_rwlock); error = EINVAL; break; } - ndi->ndi = nd_ifinfo[ifp->if_index]; + ondi->ndi.linkmtu = IN6_LINKMTU(ifp); + ondi->ndi.maxmtu = nd_ifinfo[i].maxmtu; + ondi->ndi.basereachable = nd_ifinfo[i].basereachable; + ondi->ndi.reachable = nd_ifinfo[i].reachable; + ondi->ndi.retrans = nd_ifinfo[i].retrans; + ondi->ndi.flags = nd_ifinfo[i].flags; + ondi->ndi.recalctm = nd_ifinfo[i].recalctm; + ondi->ndi.chlim = nd_ifinfo[i].chlim; + ondi->ndi.receivedra = nd_ifinfo[i].receivedra; + lck_rw_done(nd_if_rwlock); break; + case SIOCSIFINFO_FLAGS: /* XXX: almost all other fields of ndi->ndi is unused */ + lck_rw_lock_shared(nd_if_rwlock); if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { + lck_rw_done(nd_if_rwlock); error = EINVAL; break; } - nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags; + nd_ifinfo[i].flags = ndi->ndi.flags; + lck_rw_done(nd_if_rwlock); break; + case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ /* flush default router list */ /* @@ -1483,16 +2041,18 @@ nd6_ioctl(cmd, data, ifp) * route equals to the top of default router list */ bzero(&any, sizeof(any)); - defrouter_delreq(&any, 0); + lck_mtx_lock(nd6_mutex); + defrouter_delreq(&any, 1); defrouter_select(); + lck_mtx_unlock(nd6_mutex); /* xxx sumikawa: flush prefix list */ break; - case SIOCSPFXFLUSH_IN6: - { + + case SIOCSPFXFLUSH_IN6: { /* flush all the prefix advertised by routers */ - struct nd_prefix *pr, *next; + struct nd_prefix *next; + lck_mtx_lock(nd6_mutex); - s = splnet(); for (pr = nd_prefix.lh_first; pr; pr = next) { struct in6_ifaddr *ia, *ia_next; @@ -1502,27 +2062,27 @@ nd6_ioctl(cmd, data, ifp) continue; /* XXX */ /* do we really have to remove addresses as well? */ - for (ia = in6_ifaddr; ia; ia = ia_next) { - /* ia might be removed. keep the next ptr. */ + for (ia = in6_ifaddrs; ia; ia = ia_next) { + /* ia might be removed. keep the next ptr. */ ia_next = ia->ia_next; if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) continue; if (ia->ia6_ndpr == pr) - in6_purgeaddr(&ia->ia_ifa); + in6_purgeaddr(&ia->ia_ifa, 1); } - prelist_remove(pr); + prelist_remove(pr, 1); } - splx(s); + lck_mtx_unlock(nd6_mutex); break; - } - case SIOCSRTRFLUSH_IN6: - { + } + + case SIOCSRTRFLUSH_IN6: { /* flush all the default routers */ - struct nd_defrouter *dr, *next; + struct nd_defrouter *next; - s = splnet(); + lck_mtx_lock(nd6_mutex); if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) { /* * The first entry of the list may be stored in @@ -1530,67 +2090,119 @@ nd6_ioctl(cmd, data, ifp) */ for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) { next = TAILQ_NEXT(dr, dr_entry); - defrtrlist_del(dr); + defrtrlist_del(dr, 1); } - defrtrlist_del(TAILQ_FIRST(&nd_defrouter)); + defrtrlist_del(TAILQ_FIRST(&nd_defrouter), 1); } - splx(s); + lck_mtx_unlock(nd6_mutex); break; - } - case SIOCGNBRINFO_IN6: - { + } + + case SIOCGNBRINFO_IN6_32: { struct llinfo_nd6 *ln; - struct in6_addr nb_addr = nbi->addr; /* make local for safety */ + struct in6_nbrinfo_32 *nbi_32 = (struct in6_nbrinfo_32 *)data; + /* make local for safety */ + struct in6_addr nb_addr = nbi_32->addr; /* * XXX: KAME specific hack for scoped addresses * XXXX: for other scopes than link-local? */ - if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) || - IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) { + if (IN6_IS_ADDR_LINKLOCAL(&nbi_32->addr) || + IN6_IS_ADDR_MC_LINKLOCAL(&nbi_32->addr)) { u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2]; if (*idp == 0) *idp = htons(ifp->if_index); } - s = splnet(); - if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) { + /* Callee returns a locked route upon success */ + if ((rt = nd6_lookup(&nb_addr, 0, ifp, 0)) == NULL) { error = EINVAL; - splx(s); break; } - ln = (struct llinfo_nd6 *)rt->rt_llinfo; - nbi->state = ln->ln_state; - nbi->asked = ln->ln_asked; - nbi->isrouter = ln->ln_router; - nbi->expire = ln->ln_expire; - splx(s); - + RT_LOCK_ASSERT_HELD(rt); + ln = rt->rt_llinfo; + nbi_32->state = ln->ln_state; + nbi_32->asked = ln->ln_asked; + nbi_32->isrouter = ln->ln_router; + nbi_32->expire = ln->ln_expire; + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); break; - } - case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ - ndif->ifindex = nd6_defifindex; + } + + case SIOCGNBRINFO_IN6_64: { + struct llinfo_nd6 *ln; + struct in6_nbrinfo_64 *nbi_64 = (struct in6_nbrinfo_64 *)data; + /* make local for safety */ + struct in6_addr nb_addr = nbi_64->addr; + + /* + * XXX: KAME specific hack for scoped addresses + * XXXX: for other scopes than link-local? + */ + if (IN6_IS_ADDR_LINKLOCAL(&nbi_64->addr) || + IN6_IS_ADDR_MC_LINKLOCAL(&nbi_64->addr)) { + u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2]; + + if (*idp == 0) + *idp = htons(ifp->if_index); + } + + /* Callee returns a locked route upon success */ + if ((rt = nd6_lookup(&nb_addr, 0, ifp, 0)) == NULL) { + error = EINVAL; + break; + } + RT_LOCK_ASSERT_HELD(rt); + ln = rt->rt_llinfo; + nbi_64->state = ln->ln_state; + nbi_64->asked = ln->ln_asked; + nbi_64->isrouter = ln->ln_router; + nbi_64->expire = ln->ln_expire; + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); break; - case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ - return(nd6_setdefaultiface(ndif->ifindex)); + } + + case SIOCGDEFIFACE_IN6_32: /* XXX: should be implemented as a sysctl? */ + case SIOCGDEFIFACE_IN6_64: { + struct in6_ndifreq_64 *ndif_64 = (struct in6_ndifreq_64 *)data; + struct in6_ndifreq_32 *ndif_32 = (struct in6_ndifreq_32 *)data; + + if (cmd == SIOCGDEFIFACE_IN6_64) + ndif_64->ifindex = nd6_defifindex; + else + ndif_32->ifindex = nd6_defifindex; break; } - return(error); + + case SIOCSDEFIFACE_IN6_32: /* XXX: should be implemented as a sysctl? */ + case SIOCSDEFIFACE_IN6_64: { + struct in6_ndifreq_64 *ndif_64 = (struct in6_ndifreq_64 *)data; + struct in6_ndifreq_32 *ndif_32 = (struct in6_ndifreq_32 *)data; + + return (nd6_setdefaultiface(cmd == SIOCSDEFIFACE_IN6_64 ? + ndif_64->ifindex : ndif_32->ifindex)); + /* NOTREACHED */ + } + } + return (error); } /* * Create neighbor cache entry and cache link-layer address, * on reception of inbound ND6 packets. (RS/RA/NS/redirect) */ -struct rtentry * -nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code) - struct ifnet *ifp; - struct in6_addr *from; - char *lladdr; - int lladdrlen; - int type; /* ICMP6 type */ - int code; /* type dependent information */ +void +nd6_cache_lladdr( + struct ifnet *ifp, + struct in6_addr *from, + char *lladdr, + __unused int lladdrlen, + int type, /* ICMP6 type */ + int code) /* type dependent information */ { struct rtentry *rt = NULL; struct llinfo_nd6 *ln = NULL; @@ -1600,6 +2212,7 @@ nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code) int olladdr; int llchange; int newstate = 0; + struct timeval timenow; if (!ifp) panic("ifp == NULL in nd6_cache_lladdr"); @@ -1608,7 +2221,7 @@ nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code) /* nothing must be updated for unspecified address */ if (IN6_IS_ADDR_UNSPECIFIED(from)) - return NULL; + return; /* * Validation about ifp->if_addrlen and lladdrlen must be done in @@ -1619,32 +2232,39 @@ nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code) * Spec says nothing in sections for RA, RS and NA. There's small * description on it in NS section (RFC 2461 7.2.3). */ + getmicrotime(&timenow); - rt = nd6_lookup(from, 0, ifp); - if (!rt) { + rt = nd6_lookup(from, 0, ifp, 0); + if (rt == NULL) { #if 0 /* nothing must be done if there's no lladdr */ if (!lladdr || !lladdrlen) - return NULL; + return; #endif - rt = nd6_lookup(from, 1, ifp); + if ((rt = nd6_lookup(from, 1, ifp, 0)) == NULL) + return; + RT_LOCK_ASSERT_HELD(rt); is_newentry = 1; } else { + RT_LOCK_ASSERT_HELD(rt); /* do nothing if static ndp is set */ - if (rt->rt_flags & RTF_STATIC) - return NULL; + if (rt->rt_flags & RTF_STATIC) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + return; + } is_newentry = 0; } - if (!rt) - return NULL; if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) { fail: - (void)nd6_free(rt); - return NULL; + RT_UNLOCK(rt); + nd6_free(rt); + rtfree(rt); + return; } - ln = (struct llinfo_nd6 *)rt->rt_llinfo; + ln = rt->rt_llinfo; if (!ln) goto fail; if (!rt->rt_gateway) @@ -1673,7 +2293,7 @@ fail: * 1 -- y -- (7) * STALE */ - if (lladdr) { /*(3-5) and (7)*/ + if (lladdr) { /* (3-5) and (7) */ /* * Record source link-layer address * XXX is it dependent to ifp->if_type? @@ -1683,17 +2303,17 @@ fail: } if (!is_newentry) { - if ((!olladdr && lladdr) /*(3)*/ - || (olladdr && lladdr && llchange)) { /*(5)*/ + if ((!olladdr && lladdr) /* (3) */ + || (olladdr && lladdr && llchange)) { /* (5) */ do_update = 1; newstate = ND6_LLINFO_STALE; - } else /*(1-2,4)*/ + } else /* (1-2,4) */ do_update = 0; } else { do_update = 1; - if (!lladdr) /*(6)*/ + if (!lladdr) /* (6) */ newstate = ND6_LLINFO_NOSTATE; - else /*(7)*/ + else /* (7) */ newstate = ND6_LLINFO_STALE; } @@ -1704,27 +2324,30 @@ fail: ln->ln_state = newstate; if (ln->ln_state == ND6_LLINFO_STALE) { + struct mbuf *m = ln->ln_hold; /* * XXX: since nd6_output() below will cause * state tansition to DELAY and reset the timer, * we must set the timer now, although it is actually * meaningless. */ - ln->ln_expire = time_second + nd6_gctimer; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, + nd6_gctimer); + ln->ln_hold = NULL; - if (ln->ln_hold) { + if (m != NULL) { /* * we assume ifp is not a p2p here, so just * set the 2nd argument as the 1st one. */ - nd6_output(ifp, ifp, ln->ln_hold, - (struct sockaddr_in6 *)rt_key(rt), - rt); - ln->ln_hold = NULL; + RT_UNLOCK(rt); + nd6_output(ifp, ifp, m, + (struct sockaddr_in6 *)rt_key(rt), rt, 0); + RT_LOCK(rt); } } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { /* probe right away */ - ln->ln_expire = time_second; + ln->ln_expire = timenow.tv_sec; } } @@ -1762,7 +2385,7 @@ fail: /* * New entry must have is_router flag cleared. */ - if (is_newentry) /*(6-7)*/ + if (is_newentry) /* (6-7) */ ln->ln_router = 0; break; case ND_REDIRECT: @@ -1773,7 +2396,7 @@ fail: */ if (code == ND_REDIRECT_ROUTER) ln->ln_router = 1; - else if (is_newentry) /*(6-7)*/ + else if (is_newentry) /* (6-7) */ ln->ln_router = 0; break; case ND_ROUTER_SOLICIT: @@ -1786,8 +2409,8 @@ fail: /* * Mark an entry with lladdr as a router. */ - if ((!is_newentry && (olladdr || lladdr)) /*(2-5)*/ - || (is_newentry && lladdr)) { /*(7)*/ + if ((!is_newentry && (olladdr || lladdr)) /* (2-5) */ + || (is_newentry && lladdr)) { /* (7) */ ln->ln_router = 1; } break; @@ -1808,26 +2431,30 @@ fail: * for those are not autoconfigured hosts, we explicitly avoid such * cases for safety. */ - if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv) + if (do_update && ln->ln_router && !ip6_forwarding && + (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + lck_mtx_lock(nd6_mutex); defrouter_select(); - - return rt; + lck_mtx_unlock(nd6_mutex); + } else { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + } } - static void -nd6_slowtimo(ignored_arg) - void *ignored_arg; +nd6_slowtimo( + __unused void *ignored_arg) { - int s = splnet(); int i; struct nd_ifinfo *nd6if; - s = splnet(); - timeout(nd6_slowtimo_funneled, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz); + lck_rw_lock_shared(nd_if_rwlock); for (i = 1; i < if_index + 1; i++) { if (!nd_ifinfo || i >= nd_ifinfo_indexlim) - continue; + break; nd6if = &nd_ifinfo[i]; if (nd6if->basereachable && /* already initialized */ (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { @@ -1841,98 +2468,207 @@ nd6_slowtimo(ignored_arg) nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); } } - splx(s); -} - -static void -nd6_slowtimo_funneled(ignored_arg) - void *ignored_arg; -{ -#ifdef __APPLE__ - boolean_t funnel_state; - funnel_state = thread_funnel_set(network_flock, TRUE); -#endif - nd6_slowtimo(ignored_arg); -#ifdef __APPLE__ - (void) thread_funnel_set(network_flock, FALSE); -#endif + lck_rw_done(nd_if_rwlock); + timeout(nd6_slowtimo, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz); } #define senderr(e) { error = (e); goto bad;} int -nd6_output(ifp, origifp, m0, dst, rt0) - struct ifnet *ifp; - struct ifnet *origifp; - struct mbuf *m0; - struct sockaddr_in6 *dst; - struct rtentry *rt0; +nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, + struct sockaddr_in6 *dst, struct rtentry *hint0, int locked) { struct mbuf *m = m0; - struct rtentry *rt = rt0; - struct sockaddr_in6 *gw6 = NULL; + struct rtentry *rt = hint0, *hint = hint0; struct llinfo_nd6 *ln = NULL; int error = 0; + struct timeval timenow; + struct rtentry *rtrele = NULL; - if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) - goto sendpkt; + if (rt != NULL) { + RT_LOCK_SPIN(rt); + RT_ADDREF_LOCKED(rt); + } - if (nd6_need_cache(ifp) == 0) + if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr) || !nd6_need_cache(ifp)) { + if (rt != NULL) + RT_UNLOCK(rt); goto sendpkt; + } /* - * next hop determination. This routine is derived from ether_outpout. + * Next hop determination. Because we may involve the gateway route + * in addition to the original route, locking is rather complicated. + * The general concept is that regardless of whether the route points + * to the original route or to the gateway route, this routine takes + * an extra reference on such a route. This extra reference will be + * released at the end. + * + * Care must be taken to ensure that the "hint0" route never gets freed + * via rtfree(), since the caller may have stored it inside a struct + * route with a reference held for that placeholder. + * + * This logic is similar to, though not exactly the same as the one + * used by arp_route_to_gateway_route(). */ - if (rt) { - if ((rt->rt_flags & RTF_UP) == 0) { - if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) != - NULL) - { - rtunref(rt); + if (rt != NULL) { + /* + * We have a reference to "rt" by now (or below via rtalloc1), + * which will either be released or freed at the end of this + * routine. + */ + RT_LOCK_ASSERT_HELD(rt); + if (!(rt->rt_flags & RTF_UP)) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + if ((hint = rt = rtalloc1((struct sockaddr *)dst, + 1, 0)) != NULL) { + RT_LOCK_SPIN(rt); if (rt->rt_ifp != ifp) { /* XXX: loop care? */ - return nd6_output(ifp, origifp, m0, - dst, rt); + RT_UNLOCK(rt); + error = nd6_output(ifp, origifp, m0, + dst, rt, locked); + rtfree(rt); + return (error); } - } else + } else { senderr(EHOSTUNREACH); + } } if (rt->rt_flags & RTF_GATEWAY) { - gw6 = (struct sockaddr_in6 *)rt->rt_gateway; + struct rtentry *gwrt; + struct in6_ifaddr *ia6 = NULL; + struct sockaddr_in6 gw6; + + gw6 = *((struct sockaddr_in6 *)rt->rt_gateway); + /* + * Must drop rt_lock since nd6_is_addr_neighbor() + * calls nd6_lookup() and acquires rnh_lock. + */ + RT_UNLOCK(rt); /* * We skip link-layer address resolution and NUD * if the gateway is not a neighbor from ND point - * of view, regardless the value of the - * nd_ifinfo.flags. - * The second condition is a bit tricky: we skip + * of view, regardless of the value of nd_ifinfo.flags. + * The second condition is a bit tricky; we skip * if the gateway is our own address, which is * sometimes used to install a route to a p2p link. */ - if (!nd6_is_addr_neighbor(gw6, ifp) || - in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) { + if (!nd6_is_addr_neighbor(&gw6, ifp, 0) || + (ia6 = in6ifa_ifpwithaddr(ifp, &gw6.sin6_addr))) { /* * We allow this kind of tricky route only * when the outgoing interface is p2p. * XXX: we may need a more generic rule here. */ + if (ia6 != NULL) + ifafree(&ia6->ia_ifa); if ((ifp->if_flags & IFF_POINTOPOINT) == 0) senderr(EHOSTUNREACH); - goto sendpkt; } - if (rt->rt_gwroute == 0) + RT_LOCK_SPIN(rt); + gw6 = *((struct sockaddr_in6 *)rt->rt_gateway); + + /* If hint is now down, give up */ + if (!(rt->rt_flags & RTF_UP)) { + RT_UNLOCK(rt); + senderr(EHOSTUNREACH); + } + + /* If there's no gateway route, look it up */ + if ((gwrt = rt->rt_gwroute) == NULL) { + RT_UNLOCK(rt); goto lookup; - if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) { - rtfree(rt); rt = rt0; - lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL); - if ((rt = rt->rt_gwroute) == 0) + } + /* Become a regular mutex */ + RT_CONVERT_LOCK(rt); + + /* + * Take gwrt's lock while holding route's lock; + * this is okay since gwrt never points back + * to rt, so no lock ordering issues. + */ + RT_LOCK_SPIN(gwrt); + if (!(gwrt->rt_flags & RTF_UP)) { + struct rtentry *ogwrt; + + rt->rt_gwroute = NULL; + RT_UNLOCK(gwrt); + RT_UNLOCK(rt); + rtfree(gwrt); +lookup: + gwrt = rtalloc1((struct sockaddr *)&gw6, 1, 0); + + RT_LOCK(rt); + /* + * Bail out if the route is down, no route + * to gateway, circular route, or if the + * gateway portion of "rt" has changed. + */ + if (!(rt->rt_flags & RTF_UP) || + gwrt == NULL || gwrt == rt || + !equal(SA(&gw6), rt->rt_gateway)) { + if (gwrt == rt) { + RT_REMREF_LOCKED(gwrt); + gwrt = NULL; + } + RT_UNLOCK(rt); + if (gwrt != NULL) + rtfree(gwrt); + senderr(EHOSTUNREACH); + } + + /* Remove any existing gwrt */ + ogwrt = rt->rt_gwroute; + if ((rt->rt_gwroute = gwrt) != NULL) + RT_ADDREF(gwrt); + + RT_UNLOCK(rt); + /* Now free the replaced gwrt */ + if (ogwrt != NULL) + rtfree(ogwrt); + /* If still no route to gateway, bail out */ + if (gwrt == NULL) + senderr(EHOSTUNREACH); + /* Remember to release/free "rt" at the end */ + rtrele = rt; + rt = gwrt; + RT_LOCK_SPIN(rt); + /* If gwrt is now down, give up */ + if (!(rt->rt_flags & RTF_UP)) { + RT_UNLOCK(rt); + rtfree(rt); + rt = NULL; + /* "rtrele" == original "rt" */ senderr(EHOSTUNREACH); + } + } else { + RT_ADDREF_LOCKED(gwrt); + RT_UNLOCK(gwrt); + RT_UNLOCK(rt); + RT_LOCK_SPIN(gwrt); + /* If gwrt is now down, give up */ + if (!(gwrt->rt_flags & RTF_UP)) { + RT_UNLOCK(gwrt); + rtfree(gwrt); + senderr(EHOSTUNREACH); + } + /* Remember to release/free "rt" at the end */ + rtrele = rt; + rt = gwrt; } } + /* Become a regular mutex */ + RT_CONVERT_LOCK(rt); } + if (rt != NULL) + RT_LOCK_ASSERT_HELD(rt); + /* * Address resolution or Neighbor Unreachability Detection * for the next hop. @@ -1941,36 +2677,63 @@ nd6_output(ifp, origifp, m0, dst, rt0) */ /* Look up the neighbor cache for the nexthop */ - if (rt && (rt->rt_flags & RTF_LLINFO) != 0) - ln = (struct llinfo_nd6 *)rt->rt_llinfo; - else { + if (rt && (rt->rt_flags & RTF_LLINFO) != 0) { + ln = rt->rt_llinfo; + } else { /* * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), - * the condition below is not very efficient. But we believe + * the condition below is not very efficient. But we believe * it is tolerable, because this should be a rare case. + * Must drop rt_lock since nd6_is_addr_neighbor() calls + * nd6_lookup() and acquires rnh_lock. */ - if (nd6_is_addr_neighbor(dst, ifp) && - (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL) - ln = (struct llinfo_nd6 *)rt->rt_llinfo; + if (rt != NULL) + RT_UNLOCK(rt); + if (nd6_is_addr_neighbor(dst, ifp, 0)) { + /* "rtrele" may have been used, so clean up "rt" now */ + if (rt != NULL) { + /* Don't free "hint0" */ + if (rt == hint0) + RT_REMREF(rt); + else + rtfree(rt); + } + /* Callee returns a locked route upon success */ + rt = nd6_lookup(&dst->sin6_addr, 1, ifp, 0); + if (rt != NULL) { + RT_LOCK_ASSERT_HELD(rt); + ln = rt->rt_llinfo; + } + } else if (rt != NULL) { + RT_LOCK(rt); + } } + if (!ln || !rt) { + if (rt != NULL) + RT_UNLOCK(rt); + lck_rw_lock_shared(nd_if_rwlock); if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) { + lck_rw_done(nd_if_rwlock); log(LOG_DEBUG, "nd6_output: can't allocate llinfo for %s " "(ln=%p, rt=%p)\n", ip6_sprintf(&dst->sin6_addr), ln, rt); senderr(EIO); /* XXX: good error? */ } + lck_rw_done(nd_if_rwlock); goto sendpkt; /* send anyway */ } + getmicrotime(&timenow); + /* We don't have to do link-layer address resolution on a p2p link. */ if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && ln->ln_state < ND6_LLINFO_REACHABLE) { ln->ln_state = ND6_LLINFO_STALE; - ln->ln_expire = time_second + nd6_gctimer; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_gctimer); } /* @@ -1983,73 +2746,165 @@ nd6_output(ifp, origifp, m0, dst, rt0) if (ln->ln_state == ND6_LLINFO_STALE) { ln->ln_asked = 0; ln->ln_state = ND6_LLINFO_DELAY; - ln->ln_expire = time_second + nd6_delay; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_delay); } /* * If the neighbor cache entry has a state other than INCOMPLETE - * (i.e. its link-layer address is already reloved), just + * (i.e. its link-layer address is already resolved), just * send the packet. */ - if (ln->ln_state > ND6_LLINFO_INCOMPLETE) + if (ln->ln_state > ND6_LLINFO_INCOMPLETE) { + RT_UNLOCK(rt); + /* + * Move this entry to the head of the queue so that it is + * less likely for this entry to be a target of forced + * garbage collection (see nd6_rtrequest()). + */ + lck_mtx_lock(rnh_lock); + RT_LOCK_SPIN(rt); + if (ln->ln_flags & ND6_LNF_IN_USE) { + LN_DEQUEUE(ln); + LN_INSERTHEAD(ln); + } + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); goto sendpkt; + } /* * There is a neighbor cache entry, but no ethernet address - * response yet. Replace the held mbuf (if any) with this + * response yet. Replace the held mbuf (if any) with this * latest one. * - * XXX Does the code conform to rate-limiting rule? - * (RFC 2461 7.2.2) + * This code conforms to the rate-limiting rule described in Section + * 7.2.2 of RFC 2461, because the timer is set correctly after sending + * an NS below. */ if (ln->ln_state == ND6_LLINFO_NOSTATE) ln->ln_state = ND6_LLINFO_INCOMPLETE; if (ln->ln_hold) m_freem(ln->ln_hold); ln->ln_hold = m; - if (ln->ln_expire) { - if (ln->ln_asked < nd6_mmaxtries && - ln->ln_expire < time_second) { - ln->ln_asked++; - ln->ln_expire = time_second + - nd_ifinfo[ifp->if_index].retrans / 1000; - nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); - } + if (ln->ln_expire && ln->ln_asked < nd6_mmaxtries && + ln->ln_expire < timenow.tv_sec) { + ln->ln_asked++; + lck_rw_lock_shared(nd_if_rwlock); + ln->ln_expire = timenow.tv_sec + + nd_ifinfo[ifp->if_index].retrans / 1000; + lck_rw_done(nd_if_rwlock); + RT_UNLOCK(rt); + /* We still have a reference on rt (for ln) */ + nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0, locked); + } else { + RT_UNLOCK(rt); } - return(0); - - sendpkt: -#ifdef __APPLE__ + /* + * Move this entry to the head of the queue so that it is + * less likely for this entry to be a target of forced + * garbage collection (see nd6_rtrequest()). + */ + lck_mtx_lock(rnh_lock); + RT_LOCK_SPIN(rt); + if (ln->ln_flags & ND6_LNF_IN_USE) { + LN_DEQUEUE(ln); + LN_INSERTHEAD(ln); + } + /* Clean up "rt" now while we can */ + if (rt == hint0) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + } else { + RT_UNLOCK(rt); + rtfree_locked(rt); + } + rt = NULL; /* "rt" has been taken care of */ + lck_mtx_unlock(rnh_lock); + + error = 0; + goto release; - /* Make sure the HW checksum flags are cleaned before sending the packet */ +sendpkt: + if (rt != NULL) + RT_LOCK_ASSERT_NOTHELD(rt); - m->m_pkthdr.rcvif = (struct ifnet *)0; + /* Clean up HW checksum flags before sending the packet */ m->m_pkthdr.csum_data = 0; m->m_pkthdr.csum_flags = 0; if ((ifp->if_flags & IFF_LOOPBACK) != 0) { - return (dlil_output(ifptodlt(origifp, PF_INET6), m, (caddr_t)rt, (struct sockaddr *)dst,0)); - } + /* forwarding rules require the original scope_id */ + m->m_pkthdr.rcvif = origifp; + if (locked) + lck_mtx_unlock(ip6_mutex); + error = dlil_output(origifp, PF_INET6, m, (caddr_t)rt, + (struct sockaddr *)dst, 0); + if (locked) + lck_mtx_lock(ip6_mutex); + goto release; + } else { + /* Do not allow loopback address to wind up on a wire */ + struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); - return (dlil_output(ifptodlt(ifp, PF_INET6), m, (caddr_t)rt, (struct sockaddr *)dst, 0)); -#else - if ((ifp->if_flags & IFF_LOOPBACK) != 0) { - return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, - rt)); + if ((IN6_IS_ADDR_LOOPBACK(&ip6->ip6_src) || + IN6_IS_ADDR_LOOPBACK(&ip6->ip6_dst))) { + ip6stat.ip6s_badscope++; + /* + * Do not simply drop the packet just like a + * firewall -- we want the the application to feel + * the pain. Return ENETUNREACH like ip6_output + * does in some similar cases. This can startle + * the otherwise clueless process that specifies + * loopback as the source address. + */ + error = ENETUNREACH; + goto bad; + } } - return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt)); -#endif - - bad: - if (m) + + m->m_pkthdr.rcvif = NULL; + if (locked) + lck_mtx_unlock(ip6_mutex); + error = dlil_output(ifp, PF_INET6, m, (caddr_t)rt, + (struct sockaddr *)dst, 0); + if (locked) + lck_mtx_lock(ip6_mutex); + goto release; + +bad: + if (m != NULL) m_freem(m); + +release: + /* Clean up "rt" unless it's already been done */ + if (rt != NULL) { + RT_LOCK_SPIN(rt); + if (rt == hint0) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + } else { + RT_UNLOCK(rt); + rtfree(rt); + } + } + /* And now clean up "rtrele" if there is any */ + if (rtrele != NULL) { + RT_LOCK_SPIN(rtrele); + if (rtrele == hint0) { + RT_REMREF_LOCKED(rtrele); + RT_UNLOCK(rtrele); + } else { + RT_UNLOCK(rtrele); + rtfree(rtrele); + } + } return (error); -} +} #undef senderr int -nd6_need_cache(ifp) - struct ifnet *ifp; +nd6_need_cache( + struct ifnet *ifp) { /* * XXX: we currently do not make neighbor cache on any interface @@ -2063,12 +2918,12 @@ nd6_need_cache(ifp) case IFT_ETHER: case IFT_FDDI: case IFT_IEEE1394: -#if IFT_L2VLAN case IFT_L2VLAN: -#endif + case IFT_IEEE8023ADLAG: #if IFT_IEEE80211 case IFT_IEEE80211: #endif + case IFT_BRIDGE: case IFT_GIF: /* XXX need more cases? */ return(1); default: @@ -2077,12 +2932,12 @@ nd6_need_cache(ifp) } int -nd6_storelladdr(ifp, rt, m, dst, desten) - struct ifnet *ifp; - struct rtentry *rt; - struct mbuf *m; - struct sockaddr *dst; - u_char *desten; +nd6_storelladdr( + struct ifnet *ifp, + struct rtentry *rt, + struct mbuf *m, + struct sockaddr *dst, + u_char *desten) { int i; struct sockaddr_dl *sdl; @@ -2091,12 +2946,12 @@ nd6_storelladdr(ifp, rt, m, dst, desten) switch (ifp->if_type) { case IFT_ETHER: case IFT_FDDI: -#if IFT_L2VLAN - case IFT_L2VLAN: -#endif + case IFT_L2VLAN: + case IFT_IEEE8023ADLAG: #if IFT_IEEE80211 case IFT_IEEE80211: #endif + case IFT_BRIDGE: ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, desten); return(1); @@ -2108,154 +2963,341 @@ nd6_storelladdr(ifp, rt, m, dst, desten) *desten = 0; return(1); default: - m_freem(m); - return(0); + return(0); /* caller will free mbuf */ } } if (rt == NULL) { /* this could happen, if we could not allocate memory */ - m_freem(m); - return(0); + return(0); /* caller will free mbuf */ } + RT_LOCK(rt); if (rt->rt_gateway->sa_family != AF_LINK) { printf("nd6_storelladdr: something odd happens\n"); - m_freem(m); - return(0); + RT_UNLOCK(rt); + return(0); /* caller will free mbuf */ } sdl = SDL(rt->rt_gateway); if (sdl->sdl_alen == 0) { /* this should be impossible, but we bark here for debugging */ printf("nd6_storelladdr: sdl_alen == 0\n"); - m_freem(m); - return(0); + RT_UNLOCK(rt); + return(0); /* caller will free mbuf */ } bcopy(LLADDR(sdl), desten, sdl->sdl_alen); + RT_UNLOCK(rt); return(1); } -#ifndef __APPLE__ -static int nd6_sysctl_drlist SYSCTL_HANDLER_ARGS; -static int nd6_sysctl_prlist SYSCTL_HANDLER_ARGS; + +/* + * This is the ND pre-output routine; care must be taken to ensure that + * the "hint" route never gets freed via rtfree(), since the caller may + * have stored it inside a struct route with a reference held for that + * placeholder. + */ +errno_t +nd6_lookup_ipv6(ifnet_t ifp, const struct sockaddr_in6 *ip6_dest, + struct sockaddr_dl *ll_dest, size_t ll_dest_len, route_t hint, + mbuf_t packet) +{ + route_t route = hint; + errno_t result = 0; + struct sockaddr_dl *sdl = NULL; + size_t copy_len; + + if (ip6_dest->sin6_family != AF_INET6) + return (EAFNOSUPPORT); + + if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING)) + return (ENETDOWN); + + if (hint != NULL) { + /* + * Callee holds a reference on the route and returns + * with the route entry locked, upon success. + */ + result = arp_route_to_gateway_route( + (const struct sockaddr*)ip6_dest, hint, &route); + if (result != 0) + return (result); + if (route != NULL) + RT_LOCK_ASSERT_HELD(route); + } + + if ((packet->m_flags & M_MCAST) != 0) { + if (route != NULL) + RT_UNLOCK(route); + result = dlil_resolve_multi(ifp, + (const struct sockaddr*)ip6_dest, + (struct sockaddr *)ll_dest, ll_dest_len); + if (route != NULL) + RT_LOCK(route); + goto release; + } + + if (route == NULL) { + /* + * This could happen, if we could not allocate memory or + * if arp_route_to_gateway_route() didn't return a route. + */ + result = ENOBUFS; + goto release; + } + + if (route->rt_gateway->sa_family != AF_LINK) { + printf("nd6_lookup_ipv6: gateway address not AF_LINK\n"); + result = EADDRNOTAVAIL; + goto release; + } + + sdl = SDL(route->rt_gateway); + if (sdl->sdl_alen == 0) { + /* this should be impossible, but we bark here for debugging */ + printf("nd6_lookup_ipv6: sdl_alen == 0\n"); + result = EHOSTUNREACH; + goto release; + } + + copy_len = sdl->sdl_len <= ll_dest_len ? sdl->sdl_len : ll_dest_len; + bcopy(sdl, ll_dest, copy_len); + +release: + if (route != NULL) { + if (route == hint) { + RT_REMREF_LOCKED(route); + RT_UNLOCK(route); + } else { + RT_UNLOCK(route); + rtfree(route); + } + } + return (result); +} + SYSCTL_DECL(_net_inet6_icmp6); -SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, - CTLFLAG_RD, nd6_sysctl_drlist, ""); -SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, - CTLFLAG_RD, nd6_sysctl_prlist, ""); static int -nd6_sysctl_drlist SYSCTL_HANDLER_ARGS +nd6_sysctl_drlist SYSCTL_HANDLER_ARGS { - int error; +#pragma unused(oidp, arg1, arg2) + int error = 0; char buf[1024]; - struct in6_defrouter *d, *de; struct nd_defrouter *dr; + int p64 = proc_is64bit(req->p); if (req->newptr) - return EPERM; - error = 0; - - for (dr = TAILQ_FIRST(&nd_defrouter); - dr; - dr = TAILQ_NEXT(dr, dr_entry)) { - d = (struct in6_defrouter *)buf; - de = (struct in6_defrouter *)(buf + sizeof(buf)); - - if (d + 1 <= de) { - bzero(d, sizeof(*d)); - d->rtaddr.sin6_family = AF_INET6; - d->rtaddr.sin6_len = sizeof(d->rtaddr); - if (in6_recoverscope(&d->rtaddr, &dr->rtaddr, - dr->ifp) != 0) - log(LOG_ERR, - "scope error in " - "default router list (%s)\n", - ip6_sprintf(&dr->rtaddr)); - d->flags = dr->flags; - d->rtlifetime = dr->rtlifetime; - d->expire = dr->expire; - d->if_index = dr->ifp->if_index; - } else - panic("buffer too short"); - - error = SYSCTL_OUT(req, buf, sizeof(*d)); - if (error) - break; + return (EPERM); + + lck_mtx_lock(nd6_mutex); + if (p64) { + struct in6_defrouter_64 *d, *de; + + for (dr = TAILQ_FIRST(&nd_defrouter); + dr; + dr = TAILQ_NEXT(dr, dr_entry)) { + d = (struct in6_defrouter_64 *)buf; + de = (struct in6_defrouter_64 *)(buf + sizeof (buf)); + + if (d + 1 <= de) { + bzero(d, sizeof (*d)); + d->rtaddr.sin6_family = AF_INET6; + d->rtaddr.sin6_len = sizeof (d->rtaddr); + if (in6_recoverscope(&d->rtaddr, &dr->rtaddr, + dr->ifp) != 0) + log(LOG_ERR, + "scope error in " + "default router list (%s)\n", + ip6_sprintf(&dr->rtaddr)); + d->flags = dr->flags; + d->rtlifetime = dr->rtlifetime; + d->expire = dr->expire; + d->if_index = dr->ifp->if_index; + } else { + panic("buffer too short"); + } + error = SYSCTL_OUT(req, buf, sizeof (*d)); + if (error) + break; + } + } else { + struct in6_defrouter_32 *d_32, *de_32; + + for (dr = TAILQ_FIRST(&nd_defrouter); + dr; + dr = TAILQ_NEXT(dr, dr_entry)) { + d_32 = (struct in6_defrouter_32 *)buf; + de_32 = (struct in6_defrouter_32 *)(buf + sizeof (buf)); + + if (d_32 + 1 <= de_32) { + bzero(d_32, sizeof (*d_32)); + d_32->rtaddr.sin6_family = AF_INET6; + d_32->rtaddr.sin6_len = sizeof (d_32->rtaddr); + if (in6_recoverscope(&d_32->rtaddr, &dr->rtaddr, + dr->ifp) != 0) + log(LOG_ERR, + "scope error in " + "default router list (%s)\n", + ip6_sprintf(&dr->rtaddr)); + d_32->flags = dr->flags; + d_32->rtlifetime = dr->rtlifetime; + d_32->expire = dr->expire; + d_32->if_index = dr->ifp->if_index; + } else { + panic("buffer too short"); + } + error = SYSCTL_OUT(req, buf, sizeof (*d_32)); + if (error) + break; + } } - return error; + lck_mtx_unlock(nd6_mutex); + return (error); } static int -nd6_sysctl_prlist SYSCTL_HANDLER_ARGS +nd6_sysctl_prlist SYSCTL_HANDLER_ARGS { - int error; +#pragma unused(oidp, arg1, arg2) + int error = 0; char buf[1024]; - struct in6_prefix *p, *pe; struct nd_prefix *pr; + int p64 = proc_is64bit(req->p); if (req->newptr) - return EPERM; - error = 0; + return (EPERM); - for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { - u_short advrtrs; - size_t advance; - struct sockaddr_in6 *sin6, *s6; - struct nd_pfxrouter *pfr; + lck_mtx_lock(nd6_mutex); + if (p64) { + struct in6_prefix_64 *p, *pe; - p = (struct in6_prefix *)buf; - pe = (struct in6_prefix *)(buf + sizeof(buf)); + for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { + u_short advrtrs = 0; + size_t advance; + struct sockaddr_in6 *sin6, *s6; + struct nd_pfxrouter *pfr; - if (p + 1 <= pe) { - bzero(p, sizeof(*p)); - sin6 = (struct sockaddr_in6 *)(p + 1); + p = (struct in6_prefix_64 *)buf; + pe = (struct in6_prefix_64 *)(buf + sizeof (buf)); - p->prefix = pr->ndpr_prefix; - if (in6_recoverscope(&p->prefix, - &p->prefix.sin6_addr, pr->ndpr_ifp) != 0) - log(LOG_ERR, - "scope error in prefix list (%s)\n", - ip6_sprintf(&p->prefix.sin6_addr)); - p->raflags = pr->ndpr_raf; - p->prefixlen = pr->ndpr_plen; - p->vltime = pr->ndpr_vltime; - p->pltime = pr->ndpr_pltime; - p->if_index = pr->ndpr_ifp->if_index; - p->expire = pr->ndpr_expire; - p->refcnt = pr->ndpr_refcnt; - p->flags = pr->ndpr_stateflags; - p->origin = PR_ORIG_RA; - advrtrs = 0; - for (pfr = pr->ndpr_advrtrs.lh_first; - pfr; - pfr = pfr->pfr_next) { - if ((void *)&sin6[advrtrs + 1] > - (void *)pe) { + if (p + 1 <= pe) { + bzero(p, sizeof (*p)); + sin6 = (struct sockaddr_in6 *)(p + 1); + + p->prefix = pr->ndpr_prefix; + if (in6_recoverscope(&p->prefix, + &p->prefix.sin6_addr, pr->ndpr_ifp) != 0) + log(LOG_ERR, + "scope error in prefix list (%s)\n", + ip6_sprintf(&p->prefix.sin6_addr)); + p->raflags = pr->ndpr_raf; + p->prefixlen = pr->ndpr_plen; + p->vltime = pr->ndpr_vltime; + p->pltime = pr->ndpr_pltime; + p->if_index = pr->ndpr_ifp->if_index; + p->expire = pr->ndpr_expire; + p->refcnt = pr->ndpr_refcnt; + p->flags = pr->ndpr_stateflags; + p->origin = PR_ORIG_RA; + advrtrs = 0; + for (pfr = pr->ndpr_advrtrs.lh_first; + pfr; + pfr = pfr->pfr_next) { + if ((void *)&sin6[advrtrs + 1] > + (void *)pe) { + advrtrs++; + continue; + } + s6 = &sin6[advrtrs]; + bzero(s6, sizeof (*s6)); + s6->sin6_family = AF_INET6; + s6->sin6_len = sizeof (*sin6); + if (in6_recoverscope(s6, + &pfr->router->rtaddr, + pfr->router->ifp) != 0) + log(LOG_ERR, "scope error in " + "prefix list (%s)\n", + ip6_sprintf(&pfr->router-> + rtaddr)); advrtrs++; - continue; } - s6 = &sin6[advrtrs]; - bzero(s6, sizeof(*s6)); - s6->sin6_family = AF_INET6; - s6->sin6_len = sizeof(*sin6); - if (in6_recoverscope(s6, - &pfr->router->rtaddr, - pfr->router->ifp) != 0) - log(LOG_ERR, - "scope error in " - "prefix list (%s)\n", - ip6_sprintf(&pfr->router->rtaddr)); - advrtrs++; + p->advrtrs = advrtrs; + } else { + panic("buffer too short"); } - p->advrtrs = advrtrs; - } else - panic("buffer too short"); + advance = sizeof (*p) + sizeof (*sin6) * advrtrs; + error = SYSCTL_OUT(req, buf, advance); + if (error) + break; + } + } else { + struct in6_prefix_32 *p_32, *pe_32; - advance = sizeof(*p) + sizeof(*sin6) * advrtrs; - error = SYSCTL_OUT(req, buf, advance); - if (error) - break; + for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { + u_short advrtrs = 0; + size_t advance; + struct sockaddr_in6 *sin6, *s6; + struct nd_pfxrouter *pfr; + + p_32 = (struct in6_prefix_32 *)buf; + pe_32 = (struct in6_prefix_32 *)(buf + sizeof (buf)); + + if (p_32 + 1 <= pe_32) { + bzero(p_32, sizeof (*p_32)); + sin6 = (struct sockaddr_in6 *)(p_32 + 1); + + p_32->prefix = pr->ndpr_prefix; + if (in6_recoverscope(&p_32->prefix, + &p_32->prefix.sin6_addr, pr->ndpr_ifp) != 0) + log(LOG_ERR, "scope error in prefix " + "list (%s)\n", ip6_sprintf(&p_32-> + prefix.sin6_addr)); + p_32->raflags = pr->ndpr_raf; + p_32->prefixlen = pr->ndpr_plen; + p_32->vltime = pr->ndpr_vltime; + p_32->pltime = pr->ndpr_pltime; + p_32->if_index = pr->ndpr_ifp->if_index; + p_32->expire = pr->ndpr_expire; + p_32->refcnt = pr->ndpr_refcnt; + p_32->flags = pr->ndpr_stateflags; + p_32->origin = PR_ORIG_RA; + advrtrs = 0; + for (pfr = pr->ndpr_advrtrs.lh_first; + pfr; + pfr = pfr->pfr_next) { + if ((void *)&sin6[advrtrs + 1] > + (void *)pe_32) { + advrtrs++; + continue; + } + s6 = &sin6[advrtrs]; + bzero(s6, sizeof (*s6)); + s6->sin6_family = AF_INET6; + s6->sin6_len = sizeof (*sin6); + if (in6_recoverscope(s6, + &pfr->router->rtaddr, + pfr->router->ifp) != 0) + log(LOG_ERR, "scope error in " + "prefix list (%s)\n", + ip6_sprintf(&pfr->router-> + rtaddr)); + advrtrs++; + } + p_32->advrtrs = advrtrs; + } else { + panic("buffer too short"); + } + advance = sizeof (*p_32) + sizeof (*sin6) * advrtrs; + error = SYSCTL_OUT(req, buf, advance); + if (error) + break; + } } - return error; + lck_mtx_unlock(nd6_mutex); + return (error); } -#endif +SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, + CTLFLAG_RD, 0, 0, nd6_sysctl_drlist, "S,in6_defrouter",""); +SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, + CTLFLAG_RD, 0, 0, nd6_sysctl_prlist, "S,in6_defrouter",""); +