X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..99c3a10404e5d1ef94397ab4df5a8b74711fc4d3:/bsd/netinet6/nd6.c?ds=sidebyside diff --git a/bsd/netinet6/nd6.c b/bsd/netinet6/nd6.c index 1bee938d4..7cae1d7e7 100644 --- a/bsd/netinet6/nd6.c +++ b/bsd/netinet6/nd6.c @@ -1,3 +1,31 @@ +/* + * Copyright (c) 2000-2012 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 $ */ @@ -50,25 +78,29 @@ #include #include #include +#include +#include + #include -#include +#include #define DONT_WARN_OBSOLETE #include #include #include -#include +#include #include #include +#include #include +#include #include -#include #include #include #include #include -#include +#include #include #include "loop.h" @@ -78,8 +110,7 @@ #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ -#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 */ @@ -93,6 +124,7 @@ int nd6_gctimer = (60 * 60 * 24); /* 1 day: garbage collection timer */ int nd6_maxndopt = 10; /* max # of ND options allowed */ int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */ +int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */ #if ND6_DEBUG int nd6_debug = 1; @@ -100,24 +132,105 @@ int nd6_debug = 1; int nd6_debug = 0; #endif +int nd6_optimistic_dad = + (ND6_OPTIMISTIC_DAD_LINKLOCAL|ND6_OPTIMISTIC_DAD_AUTOCONF| + ND6_OPTIMISTIC_DAD_TEMPORARY|ND6_OPTIMISTIC_DAD_DYNAMIC); + +static int nd6_is_new_addr_neighbor (struct sockaddr_in6 *, struct ifnet *); + /* for debugging? */ static int nd6_inuse, nd6_allocated; -struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6, NULL, NULL, 0, 0, 0, 0, 0 }; -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, + * ln_llreach, ln_lastused + * + * - 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, NULL, 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_lock_grp_attr; +static lck_grp_t *nd_if_lock_grp; +static lck_attr_t *nd_if_lock_attr; +decl_lck_rw_data(, nd_if_rwlock_data); +lck_rw_t *nd_if_rwlock = &nd_if_rwlock_data; + +/* Protected by nd6_mutex */ struct nd_drhead nd_defrouter; struct nd_prhead nd_prefix = { 0 }; +/* Serialization variables for nd6_drain() */ +static boolean_t nd6_drain_busy; +static void *nd6_drain_waitchan = &nd6_drain_busy; +static int nd6_drain_waiters = 0; + int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL; static struct sockaddr_in6 all1_sa; static int regen_tmpaddr(struct in6_ifaddr *); -extern lck_mtx_t *rt_mtx; -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_llinfo_purge(struct rtentry *); +static void nd6_llinfo_get_ri(struct rtentry *, struct rt_reach_info *); +static void nd6_llinfo_get_iflri(struct rtentry *, struct ifnet_llreach_info *); + +static int nd6_siocgdrlst(void *, int); +static int 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() @@ -126,7 +239,7 @@ 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; } @@ -138,60 +251,178 @@ nd6_init() /* initialization of the default router list */ TAILQ_INIT(&nd_defrouter); + nd_if_lock_grp_attr = lck_grp_attr_alloc_init(); + nd_if_lock_grp = lck_grp_alloc_init("nd_if_lock", nd_if_lock_grp_attr); + nd_if_lock_attr = lck_attr_alloc_init(); + lck_rw_init(nd_if_rwlock, nd_if_lock_grp, nd_if_lock_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); + zone_change(llinfo_nd6_zone, Z_CALLERACCT, FALSE); + + nd6_nbr_init(); + nd6_rtr_init(); + nd6_prproxy_init(); + nd6_init_done = 1; /* start timer */ timeout(nd6_slowtimo, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz); } -void -nd6_ifattach( - 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; + } + + /* Purge any link-layer info caching */ + VERIFY(ln->ln_rt->rt_llinfo == ln); + if (ln->ln_rt->rt_llinfo_purge != NULL) + ln->ln_rt->rt_llinfo_purge(ln->ln_rt); + + zfree(llinfo_nd6_zone, ln); +} + +static void +nd6_llinfo_purge(struct rtentry *rt) +{ + struct llinfo_nd6 *ln = rt->rt_llinfo; + + RT_LOCK_ASSERT_HELD(rt); + VERIFY(rt->rt_llinfo_purge == nd6_llinfo_purge && ln != NULL); + + if (ln->ln_llreach != NULL) { + RT_CONVERT_LOCK(rt); + ifnet_llreach_free(ln->ln_llreach); + ln->ln_llreach = NULL; + } + ln->ln_lastused = 0; +} + +static void +nd6_llinfo_get_ri(struct rtentry *rt, struct rt_reach_info *ri) +{ + struct llinfo_nd6 *ln = rt->rt_llinfo; + struct if_llreach *lr = ln->ln_llreach; + + if (lr == NULL) { + bzero(ri, sizeof (*ri)); + ri->ri_rssi = IFNET_RSSI_UNKNOWN; + ri->ri_lqm = IFNET_LQM_THRESH_OFF; + ri->ri_npm = IFNET_NPM_THRESH_UNKNOWN; + } else { + IFLR_LOCK(lr); + /* Export to rt_reach_info structure */ + ifnet_lr2ri(lr, ri); + /* Export ND6 send expiration (calendar) time */ + ri->ri_snd_expire = + ifnet_llreach_up2calexp(lr, ln->ln_lastused); + IFLR_UNLOCK(lr); + } +} + +static void +nd6_llinfo_get_iflri(struct rtentry *rt, struct ifnet_llreach_info *iflri) { + struct llinfo_nd6 *ln = rt->rt_llinfo; + struct if_llreach *lr = ln->ln_llreach; + + if (lr == NULL) { + bzero(iflri, sizeof (*iflri)); + iflri->iflri_rssi = IFNET_RSSI_UNKNOWN; + iflri->iflri_lqm = IFNET_LQM_THRESH_OFF; + iflri->iflri_npm = IFNET_NPM_THRESH_UNKNOWN; + } else { + IFLR_LOCK(lr); + /* Export to ifnet_llreach_info structure */ + ifnet_lr2iflri(lr, iflri); + /* Export ND6 send expiration (uptime) time */ + iflri->iflri_snd_expire = + ifnet_llreach_up2upexp(lr, ln->ln_lastused); + IFLR_UNLOCK(lr); + } +} +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; + nd_ifinfo = (struct nd_ifinfo *)(void *)q; } #define ND nd_ifinfo[ifp->if_index] - /* * Don't initialize if called twice. - * XXX: to detect this, we should choose a member that is never set - * before initialization of the ND structure itself. We formaly used - * 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; + if (ND.initialized) { + lck_rw_done(nd_if_rwlock); + return (0); + } + lck_mtx_init(&ND.lock, nd_if_lock_grp, nd_if_lock_attr); + ND.initialized = TRUE; + 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; - nd6_setmtu(ifp); + lck_rw_done(nd_if_rwlock); #undef ND + + nd6_setmtu(ifp); + + return (0); } /* @@ -202,18 +433,22 @@ void nd6_setmtu(struct ifnet *ifp) { struct nd_ifinfo *ndi; - u_long oldmaxmtu; + u_int32_t oldmaxmtu, maxmtu; /* * Make sure IPv6 is enabled for the interface first, * because this can be called directly from SIOCSIFMTU for IPv4 */ - - if (ifp->if_index >= nd_ifinfo_indexlim) { - return; /* we're out of bound for nd_ifinfo */ + lck_rw_lock_shared(nd_if_rwlock); + if (ifp->if_index >= nd_ifinfo_indexlim || + !nd_ifinfo[ifp->if_index].initialized) { + lck_rw_done(nd_if_rwlock); + return; /* nd_ifinfo out of bound, or not yet initialized */ } ndi = &nd_ifinfo[ifp->if_index]; + VERIFY(ndi->initialized); + lck_mtx_lock(&ndi->lock); oldmaxmtu = ndi->maxmtu; /* @@ -224,10 +459,10 @@ nd6_setmtu(struct ifnet *ifp) * 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. + * ND level maxmtu and linkmtu are done via IN6_LINKMTU() which + * does further checking against if_mtu. */ - ndi->maxmtu = ifp->if_mtu; + maxmtu = ndi->maxmtu = ifp->if_mtu; /* * Decreasing the interface MTU under IPV6 minimum MTU may cause @@ -237,12 +472,15 @@ nd6_setmtu(struct ifnet *ifp) */ if (oldmaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { log(LOG_NOTICE, "nd6_setmtu: " - "new link MTU on %s%d (%lu) is too small for IPv6\n", - ifp->if_name, ifp->if_unit, (unsigned long)ndi->maxmtu); + "new link MTU on %s%d (%u) is too small for IPv6\n", + ifp->if_name, ifp->if_unit, (uint32_t)ndi->maxmtu); } + ndi->linkmtu = ifp->if_mtu; + lck_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); /* also adjust in6_maxmtu if necessary. */ - if (ndi->maxmtu > in6_maxmtu) + if (maxmtu > in6_maxmtu) in6_setmaxmtu(); } @@ -325,16 +563,16 @@ nd6_options( struct nd_opt_hdr *nd_opt; int i = 0; - if (!ndopts) - panic("ndopts == NULL in nd6_options\n"); - if (!ndopts->nd_opts_last) - panic("uninitialized ndopts in nd6_options\n"); - if (!ndopts->nd_opts_search) + if (ndopts == NULL) + panic("ndopts == NULL in nd6_options"); + if (ndopts->nd_opts_last == NULL) + panic("uninitialized ndopts in nd6_options"); + if (ndopts->nd_opts_search == NULL) return 0; while (1) { nd_opt = nd6_option(ndopts); - if (!nd_opt && !ndopts->nd_opts_last) { + if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { /* * Message validation requires that all included * options have a length that is greater than zero. @@ -344,7 +582,7 @@ nd6_options( return -1; } - if (!nd_opt) + if (nd_opt == NULL) goto skip1; switch (nd_opt->nd_opt_type) { @@ -370,6 +608,9 @@ nd6_options( ndopts->nd_opts_pi_end = (struct nd_opt_prefix_info *)nd_opt; break; + case ND_OPT_RDNSS: + /* ignore */ + break; default: /* * Unknown options must be silently ignored, @@ -395,12 +636,8 @@ skip1: return 0; } -/* - * ND6 timer routine to expire default route list and prefix list - */ void -nd6_timer( - __unused void *ignored_arg) +nd6_drain(__unused void *ignored_arg) { struct llinfo_nd6 *ln; struct nd_defrouter *dr; @@ -411,123 +648,215 @@ nd6_timer( 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; - 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; + u_int32_t retrans, flags; + + /* 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) { - ln = next; - continue; + panic("%s: ln(%p) rt(%p) rt_ifp == NULL", __func__, + ln, rt); + /* NOTREACHED */ } - ndi = &nd_ifinfo[ifp->if_index]; - dst = (struct sockaddr_in6 *)rt_key(rt); - if (ln->ln_expire > timenow.tv_sec) { - ln = next; - continue; + /* 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 */ } - /* sanity check */ - if (!rt) { - printf("rt=0 in nd6_timer(ln=%p)\n", ln); - ln = next; - continue; + /* rt_key should never be NULL */ + dst = (struct sockaddr_in6 *)(void *)rt_key(rt); + if (dst == NULL) { + panic("%s: rt(%p) key is NULL ln(%p)", __func__, + rt, ln); + /* NOTREACHED */ } - if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln) { - printf("rt_llinfo(%p) is not equal to ln(%p)\n", - rt->rt_llinfo, ln); + + /* 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; + continue; } - if (!dst) { - printf("dst=0 in nd6_timer(ln=%p)\n", ln); + + 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; } + ndi = ND_IFINFO(ifp); + VERIFY(ndi->initialized); + lck_mtx_lock(&ndi->lock); + retrans = ndi->retrans; + flags = ndi->flags; + lck_mtx_unlock(&ndi->lock); + 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 = timenow.tv_sec + - nd_ifinfo[ifp->if_index].retrans / 1000; - nd6_ns_output(ifp, NULL, &dst->sin6_addr, - ln, 0, 0); + ln->ln_expire = timenow.tv_sec + retrans / 1000; + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); + if (ip6_forwarding) { + nd6_prproxy_ns_output(ifp, NULL, + &dst->sin6_addr, ln); + } else { + nd6_ns_output(ifp, NULL, + &dst->sin6_addr, ln, 0); + } + RT_REMREF(rt); } else { struct mbuf *m = ln->ln_hold; ln->ln_hold = NULL; - 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; - } + 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 = timenow.tv_sec + 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 ((flags & ND6_IFF_PERFORMNUD) != 0) { /* We need NUD */ ln->ln_asked = 1; ln->ln_state = ND6_LLINFO_PROBE; - ln->ln_expire = timenow.tv_sec + - ndi->retrans / 1000; + ln->ln_expire = timenow.tv_sec + 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, 0); - } else { - ln->ln_state = ND6_LLINFO_STALE; /* XXX */ - ln->ln_expire = timenow.tv_sec + nd6_gctimer; + &dst->sin6_addr, ln, 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 = timenow.tv_sec + - nd_ifinfo[ifp->if_index].retrans / 1000; + ln->ln_expire = timenow.tv_sec + 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, 0); + &dst->sin6_addr, ln, 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); @@ -535,12 +864,13 @@ nd6_timer( if (dr->expire && dr->expire < timenow.tv_sec) { struct nd_defrouter *t; t = TAILQ_NEXT(dr, dr_entry); - defrtrlist_del(dr, 1); + defrtrlist_del(dr); dr = t; } else { dr = TAILQ_NEXT(dr, dr_entry); } } + lck_mtx_unlock(nd6_mutex); /* * expire interface addresses. @@ -548,14 +878,21 @@ nd6_timer( * However, from a stricter speci-confrmance standpoint, we should * rather separate address lifetimes and prefix lifetimes. */ - addrloop: +addrloop: + lck_rw_lock_exclusive(&in6_ifaddr_rwlock); for (ia6 = in6_ifaddrs; ia6; ia6 = nia6) { nia6 = ia6->ia_next; + IFA_LOCK(&ia6->ia_ifa); + /* + * 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 in6_ifaddr_rwlock below. + */ + IFA_ADDREF_LOCKED(&ia6->ia_ifa); /* check address lifetime */ lt6 = &ia6->ia6_lifetime; if (IFA6_IS_INVALID(ia6)) { - int regen = 0; - /* * If the expiring address is temporary, try * regenerating a new one. This would be useful when @@ -564,27 +901,39 @@ nd6_timer( * addresses. Although we may have to restart the * loop (see below), it must be after purging the * address. Otherwise, we'd see an infinite loop of - * regeneration. + * regeneration. */ if (ip6_use_tempaddr && (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 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 + /* + * 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 + * re-evaluation of the in6_ifaddrs list */ - lck_mtx_unlock(nd6_mutex); - if (regen_tmpaddr(ia6) == 0) - regen = 1; - lck_mtx_lock(nd6_mutex); + IFA_UNLOCK(&ia6->ia_ifa); + lck_rw_done(&in6_ifaddr_rwlock); + (void) regen_tmpaddr(ia6); + } else { + IFA_UNLOCK(&ia6->ia_ifa); + lck_rw_done(&in6_ifaddr_rwlock); } - in6_purgeaddr(&ia6->ia_ifa, 1); + /* + * Purging the address would have caused + * in6_ifaddr_rwlock to be dropped and reacquired; + * therefore search again from the beginning + * of in6_ifaddrs list. + */ + in6_purgeaddr(&ia6->ia_ifa); - if (regen) - goto addrloop; /* XXX: see below */ + /* Release extra reference taken above */ + IFA_REMREF(&ia6->ia_ifa); + goto addrloop; } + IFA_LOCK_ASSERT_HELD(&ia6->ia_ifa); if (IFA6_IS_DEPRECATED(ia6)) { int oldflags = ia6->ia6_flags; @@ -599,7 +948,8 @@ nd6_timer( (oldflags & IN6_IFF_DEPRECATED) == 0) { /* see NOTE above */ - lck_mtx_unlock(nd6_mutex); + IFA_UNLOCK(&ia6->ia_ifa); + lck_rw_done(&in6_ifaddr_rwlock); if (regen_tmpaddr(ia6) == 0) { /* * A new temporary address is @@ -610,13 +960,16 @@ nd6_timer( * would not cause disaster (because * it's not a deletion, but an * addition,) we'd rather restart the - * loop just for safety. Or does this + * loop just for safety. Or does this * significantly reduce performance?? */ - lck_mtx_lock(nd6_mutex); + /* Release extra reference */ + IFA_REMREF(&ia6->ia_ifa); goto addrloop; } - lck_mtx_lock(nd6_mutex); + lck_rw_lock_exclusive(&in6_ifaddr_rwlock); + } else { + IFA_UNLOCK(&ia6->ia_ifa); } } else { /* @@ -624,8 +977,26 @@ nd6_timer( * preferred. */ ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; + IFA_UNLOCK(&ia6->ia_ifa); } + lck_rw_assert(&in6_ifaddr_rwlock, LCK_RW_ASSERT_EXCLUSIVE); + /* Release extra reference taken above */ + IFA_REMREF(&ia6->ia_ifa); + } + lck_rw_done(&in6_ifaddr_rwlock); + + lck_mtx_lock(nd6_mutex); + /* + * Since we drop the nd6_mutex in prelist_remove, we want to run this + * section single threaded. + */ + while (nd6_drain_busy) { + nd6_drain_waiters++; + msleep(nd6_drain_waitchan, nd6_mutex, (PZERO-1), + __func__, NULL); + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); } + nd6_drain_busy = TRUE; /* expire prefix list */ pr = nd_prefix.lh_first; @@ -635,21 +1006,107 @@ nd6_timer( * since pltime is just for autoconf, pltime processing for * prefix is not necessary. */ + NDPR_LOCK(pr); + if (pr->ndpr_stateflags & NDPRF_PROCESSED) { + NDPR_UNLOCK(pr); + pr = pr->ndpr_next; + continue; + } if (pr->ndpr_expire && pr->ndpr_expire < timenow.tv_sec) { - struct nd_prefix *t; - t = pr->ndpr_next; - /* * address expiration and prefix expiration are * separate. NEVER perform in6_purgeaddr here. */ - - prelist_remove(pr, 1); - pr = t; - } else + pr->ndpr_stateflags |= NDPRF_PROCESSED; + NDPR_ADDREF_LOCKED(pr); + prelist_remove(pr); + NDPR_UNLOCK(pr); + NDPR_REMREF(pr); + pr = nd_prefix.lh_first; + } else { + pr->ndpr_stateflags |= NDPRF_PROCESSED; + NDPR_UNLOCK(pr); pr = pr->ndpr_next; + } + } + LIST_FOREACH(pr, &nd_prefix, ndpr_entry) { + NDPR_LOCK(pr); + pr->ndpr_stateflags &= ~NDPRF_PROCESSED; + NDPR_UNLOCK(pr); + } + nd6_drain_busy = FALSE; + if (nd6_drain_waiters > 0) { + nd6_drain_waiters = 0; + wakeup(nd6_drain_waitchan); } lck_mtx_unlock(nd6_mutex); +} + +/* + * ND6 router advertisement kernel notification + */ +void +nd6_post_msg(u_int32_t code, struct nd_prefix_list *prefix_list, + u_int32_t list_length, u_int32_t mtu, char *dl_addr, u_int32_t dl_addr_len) +{ + struct kev_msg ev_msg; + struct kev_nd6_ra_data nd6_ra_msg_data; + struct nd_prefix_list *itr = prefix_list; + + bzero(&ev_msg, sizeof(struct kev_msg)); + ev_msg.vendor_code = KEV_VENDOR_APPLE; + ev_msg.kev_class = KEV_NETWORK_CLASS; + ev_msg.kev_subclass = KEV_ND6_SUBCLASS; + ev_msg.event_code = code; + + bzero(&nd6_ra_msg_data, sizeof(nd6_ra_msg_data)); + nd6_ra_msg_data.lladdrlen = (dl_addr_len <= ND6_ROUTER_LL_SIZE) ? + dl_addr_len : ND6_ROUTER_LL_SIZE; + bcopy(dl_addr, &nd6_ra_msg_data.lladdr, nd6_ra_msg_data.lladdrlen); + + if (mtu > 0 && mtu >= IPV6_MMTU) { + nd6_ra_msg_data.mtu = mtu; + nd6_ra_msg_data.flags |= KEV_ND6_DATA_VALID_MTU; + } + + if (list_length > 0 && prefix_list != NULL) { + nd6_ra_msg_data.list_length = list_length; + nd6_ra_msg_data.flags |= KEV_ND6_DATA_VALID_PREFIX; + } + + while (itr != NULL && nd6_ra_msg_data.list_index < list_length) { + bcopy(&itr->pr.ndpr_prefix, &nd6_ra_msg_data.prefix.prefix, + sizeof (nd6_ra_msg_data.prefix.prefix)); + nd6_ra_msg_data.prefix.raflags = itr->pr.ndpr_raf; + nd6_ra_msg_data.prefix.prefixlen = itr->pr.ndpr_plen; + nd6_ra_msg_data.prefix.origin = PR_ORIG_RA; + nd6_ra_msg_data.prefix.vltime = itr->pr.ndpr_vltime; + nd6_ra_msg_data.prefix.pltime = itr->pr.ndpr_pltime; + nd6_ra_msg_data.prefix.expire = itr->pr.ndpr_expire; + nd6_ra_msg_data.prefix.flags = itr->pr.ndpr_stateflags; + nd6_ra_msg_data.prefix.refcnt = itr->pr.ndpr_addrcnt; + nd6_ra_msg_data.prefix.if_index = itr->pr.ndpr_ifp->if_index; + + /* send the message up */ + ev_msg.dv[0].data_ptr = &nd6_ra_msg_data; + ev_msg.dv[0].data_length = sizeof(nd6_ra_msg_data); + ev_msg.dv[1].data_length = 0; + kev_post_msg(&ev_msg); + + /* clean up for the next prefix */ + bzero(&nd6_ra_msg_data.prefix, sizeof(nd6_ra_msg_data.prefix)); + itr = itr->next; + nd6_ra_msg_data.list_index++; + } +} + +/* + * 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); } @@ -665,25 +1122,29 @@ regen_tmpaddr( getmicrotime(&timenow); ifp = ia6->ia_ifa.ifa_ifp; - ifnet_lock_exclusive(ifp); + ifnet_lock_shared(ifp); for (ifa = ifp->if_addrlist.tqh_first; ifa; ifa = ifa->ifa_list.tqe_next) { struct in6_ifaddr *it6; - if (ifa->ifa_addr->sa_family != AF_INET6) + IFA_LOCK(ifa); + if (ifa->ifa_addr->sa_family != AF_INET6) { + IFA_UNLOCK(ifa); continue; - + } it6 = (struct in6_ifaddr *)ifa; /* ignore no autoconf addresses. */ - if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) + if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) { + IFA_UNLOCK(ifa); continue; - + } /* ignore autoconf addresses with different prefixes. */ - if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) + if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) { + IFA_UNLOCK(ifa); continue; - + } /* * Now we are looking at an autoconf address with the same * prefix as ours. If the address is temporary and is still @@ -693,6 +1154,9 @@ regen_tmpaddr( */ if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && !IFA6_IS_DEPRECATED(it6)) { + IFA_UNLOCK(ifa); + if (public_ifa6 != NULL) + IFA_REMREF(&public_ifa6->ia_ifa); public_ifa6 = NULL; break; } @@ -703,19 +1167,28 @@ regen_tmpaddr( * loop here, because there may be a still-preferred temporary * address with the prefix. */ - if (!IFA6_IS_DEPRECATED(it6)) - public_ifa6 = it6; + if (!IFA6_IS_DEPRECATED(it6)) { + IFA_ADDREF_LOCKED(ifa); /* for public_ifa6 */ + IFA_UNLOCK(ifa); + if (public_ifa6 != NULL) + IFA_REMREF(&public_ifa6->ia_ifa); + public_ifa6 = it6; + } else { + IFA_UNLOCK(ifa); + } } 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); + IFA_REMREF(&public_ifa6->ia_ifa); return(-1); } + IFA_REMREF(&public_ifa6->ia_ifa); return(0); } @@ -730,8 +1203,8 @@ void nd6_purge( struct ifnet *ifp) { - struct llinfo_nd6 *ln, *nln; - struct nd_defrouter *dr, *ndr, drany; + struct llinfo_nd6 *ln; + struct nd_defrouter *dr, *ndr; struct nd_prefix *pr, *npr; /* Nuke default router list entries toward ifp */ @@ -743,18 +1216,38 @@ nd6_purge( */ for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) { ndr = TAILQ_NEXT(dr, dr_entry); + if (dr->stateflags & NDDRF_INSTALLED) + continue; if (dr->ifp == ifp) - defrtrlist_del(dr, 1); + defrtrlist_del(dr); } dr = TAILQ_FIRST(&nd_defrouter); if (dr->ifp == ifp) - defrtrlist_del(dr, 1); + defrtrlist_del(dr); + } + + for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) { + ndr = TAILQ_NEXT(dr, dr_entry); + if (!(dr->stateflags & NDDRF_INSTALLED)) + continue; + + if (dr->ifp == ifp) + defrtrlist_del(dr); } /* Nuke prefix list entries toward ifp */ for (pr = nd_prefix.lh_first; pr; pr = npr) { npr = pr->ndpr_next; + NDPR_LOCK(pr); if (pr->ndpr_ifp == ifp) { + /* + * Because if_detach() does *not* release prefixes + * while purging addresses the reference count will + * still be above zero. We therefore reset it to + * make sure that the prefix really gets purged. + */ + pr->ndpr_addrcnt = 0; + /* * Previously, pr->ndpr_addr is removed as well, * but I strongly believe we don't have to do it. @@ -763,21 +1256,31 @@ nd6_purge( * by itself. * (jinmei@kame.net 20010129) */ - prelist_remove(pr, 1); + NDPR_ADDREF_LOCKED(pr); + prelist_remove(pr); + NDPR_UNLOCK(pr); + NDPR_REMREF(pr); + } else { + NDPR_UNLOCK(pr); } } + lck_mtx_unlock(nd6_mutex); /* cancel default outgoing interface setting */ - if (nd6_defifindex == ifp->if_index) + if (nd6_defifindex == ifp->if_index) { nd6_setdefaultiface(0); + } - if (!ip6_forwarding && (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { + /* + * Perform default router selection even when we are a router, + * if Scoped Routing is enabled. + */ + if (ip6_doscopedroute || !ip6_forwarding) { + lck_mtx_lock(nd6_mutex); /* refresh default router list */ - bzero(&drany, sizeof(drany)); - defrouter_delreq(&drany, 0); - defrouter_select(); + defrouter_select(ifp); + lck_mtx_unlock(nd6_mutex); } - lck_mtx_unlock(nd6_mutex); /* * Nuke neighbor cache entries for the ifp. @@ -785,23 +1288,48 @@ nd6_purge( * 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( struct in6_addr *addr6, @@ -811,31 +1339,46 @@ nd6_lookup( { struct rtentry *rt; struct sockaddr_in6 sin6; + unsigned int ifscope; bzero(&sin6, sizeof(sin6)); sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_family = AF_INET6; sin6.sin6_addr = *addr6; -#if SCOPEDROUTING - sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6); -#endif - if (!rt_locked) - lck_mtx_lock(rt_mtx); - rt = rtalloc1_locked((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_locked(rt); - rt = 0; + + ifscope = (ifp != NULL) ? ifp->if_index : IFSCOPE_NONE; + if (rt_locked) { + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); + rt = rtalloc1_scoped_locked((struct sockaddr *)&sin6, + create, 0, ifscope); + } else { + rt = rtalloc1_scoped((struct sockaddr *)&sin6, + create, 0, ifscope); + } + + 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; + u_int32_t ifa_flags; int e; /* @@ -845,13 +1388,9 @@ nd6_lookup( * 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); - if (ifa == NULL) { - if (!rt_locked) - lck_mtx_unlock(rt_mtx); + ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); + if (ifa == NULL) return(NULL); - } /* * Create a new route. RTF_LLINFO is necessary @@ -859,37 +1398,37 @@ nd6_lookup( * destination in nd6_rtrequest which will be * called in rtrequest via ifa->ifa_rtrequest. */ - 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 (!rt_locked) + lck_mtx_lock(rnh_lock); + IFA_LOCK_SPIN(ifa); + ifa_flags = ifa->ifa_flags; + IFA_UNLOCK(ifa); + if ((e = rtrequest_scoped_locked(RTM_ADD, + (struct sockaddr *)&sin6, ifa->ifa_addr, + (struct sockaddr *)&all1_sa, + (ifa_flags | RTF_HOST | RTF_LLINFO) & + ~RTF_CLONING, &rt, ifscope)) != 0) { if (e != EEXIST) - log(LOG_ERR, - "nd6_lookup: failed to add route for a " - "neighbor(%s), errno=%d\n", - ip6_sprintf(addr6), e); + log(LOG_ERR, "%s: failed to add route " + "for a neighbor(%s), errno=%d\n", + __func__, ip6_sprintf(addr6), e); } - ifafree(ifa); - if (rt == NULL) { - if (!rt_locked) - lck_mtx_unlock(rt_mtx); + if (!rt_locked) + lck_mtx_unlock(rnh_lock); + IFA_REMREF(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 { - if (!rt_locked) - lck_mtx_unlock(rt_mtx); 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 @@ -900,105 +1439,201 @@ nd6_lookup( * 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. + * interface. + * Note also that ifa_ifp and ifp may differ when we connect two + * interfaces to a same link, install a link prefix to an interface, + * and try to install a neighbor cache on an interface that does not + * have a route to the prefix. + * + * If the address is from a proxied prefix, the ifa_ifp and ifp might + * not match, because nd6_na_input() could have modified the ifp + * of the route to point to the interface where the NA arrived on, + * hence the test for RTF_PROXY. */ - if ((ifp && ifp->if_type !=IFT_PPP) && ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || + if ((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))) { - if (!rt_locked) - lck_mtx_unlock(rt_mtx); + (ifp && rt->rt_ifa->ifa_ifp != ifp && + !(rt->rt_flags & RTF_PROXY))) { + 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(NULL); - } - if (!rt_locked) - lck_mtx_unlock(rt_mtx); + } + /* + * Caller needs to release reference and call RT_UNLOCK(rt). + */ return(rt); } /* - * Detect if a given IPv6 address identifies a neighbor on a given link. - * XXX: should take care of the destination of a p2p link? + * Test whether a given IPv6 address is a neighbor or not, ignoring + * the actual neighbor cache. The neighbor cache is ignored in order + * to not reenter the routing code from within itself. */ -int -nd6_is_addr_neighbor( +static int +nd6_is_new_addr_neighbor( struct sockaddr_in6 *addr, - struct ifnet *ifp, - int rt_locked) + struct ifnet *ifp) { - struct ifaddr *ifa; - int i; + struct nd_prefix *pr; + struct ifaddr *dstaddr; -#define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr) -#define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr) + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); /* * A link-local address is always a neighbor. - * XXX: we should use the sin6_scope_id field rather than the embedded - * interface index. + * XXX: a link does not necessarily specify a single interface. */ - if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) && - ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index) - return(1); + if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { + struct sockaddr_in6 sin6_copy; + u_int32_t zone; + + /* + * We need sin6_copy since sa6_recoverscope() may modify the + * content (XXX). + */ + sin6_copy = *addr; + if (sa6_recoverscope(&sin6_copy, FALSE)) + return (0); /* XXX: should be impossible */ + if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) + return (0); + if (sin6_copy.sin6_scope_id == zone) + return (1); + else + return (0); + } /* * If the address matches one of our addresses, * it should be a neighbor. + * If the address matches one of our on-link prefixes, 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) + for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { + NDPR_LOCK(pr); + if (pr->ndpr_ifp != ifp) { + NDPR_UNLOCK(pr); + continue; + } + if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { + NDPR_UNLOCK(pr); continue; + } + if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, + &addr->sin6_addr, &pr->ndpr_mask)) { + NDPR_UNLOCK(pr); + return (1); + } + NDPR_UNLOCK(pr); + } - for (i = 0; i < 4; i++) { - if ((IFADDR6(ifa).s6_addr32[i] ^ - addr->sin6_addr.s6_addr32[i]) & - IFMASK6(ifa).s6_addr32[i]) - continue; + /* + * If the address is assigned on the node of the other side of + * a p2p interface, the address should be a neighbor. + */ + dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); + if (dstaddr != NULL) { + if (dstaddr->ifa_ifp == ifp) { + IFA_REMREF(dstaddr); + return (1); } - ifnet_lock_done(ifp); - return(1); + IFA_REMREF(dstaddr); + dstaddr = NULL; } - ifnet_lock_done(ifp); /* - * Even if the address matches none of our addresses, it might be - * in the neighbor cache. + * If the default router list is empty, all addresses are regarded + * as on-link, and thus, as a neighbor. + * XXX: we restrict the condition to hosts, because routers usually do + * not have the "default router list". + * XXX: this block should eventually be removed (it is disabled when + * Scoped Routing is in effect); treating all destinations as on-link + * in the absence of a router is rather harmful. */ - if (nd6_lookup(&addr->sin6_addr, 0, ifp, rt_locked) != NULL) - return(1); + if (!ip6_doscopedroute && !ip6_forwarding && + TAILQ_FIRST(&nd_defrouter) == NULL && + nd6_defifindex == ifp->if_index) { + return (1); + } - return(0); -#undef IFADDR6 -#undef IFMASK6 + return (0); +} + + +/* + * Detect if a given IPv6 address identifies a neighbor on a given link. + * XXX: should take care of the destination of a p2p link? + */ +int +nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp, int rt_locked) +{ + struct rtentry *rt; + + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_NOTOWNED); + lck_mtx_lock(nd6_mutex); + if (nd6_is_new_addr_neighbor(addr, ifp)) { + lck_mtx_unlock(nd6_mutex); + return (1); + } + lck_mtx_unlock(nd6_mutex); + + /* + * Even if the address matches none of our addresses, it might be + * in the neighbor cache. + */ + 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); } /* * Free an nd6 llinfo entry. + * Since the function would cause significant changes in the kernel, DO NOT + * make it global, unless you have a strong reason for the change, and are sure + * that the change is safe. */ -struct llinfo_nd6 * +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 *)(void *)rt_key(rt))->sin6_addr; + /* - * we used to have pfctlinput(PRC_HOSTDEAD) here. - * even though it is not harmful, it was not really necessary. + * 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; - if (!ip6_forwarding && (ip6_accept_rtadv || (rt->rt_ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { - lck_mtx_lock(nd6_mutex); - dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr, - rt->rt_ifp); + /* + * We used to have pfctlinput(PRC_HOSTDEAD) here. Even though it is + * not harmful, it was not really necessary. Perform default router + * selection even when we are a router, if Scoped Routing is enabled. + */ + if (ip6_doscopedroute || !ip6_forwarding) { + dr = defrouter_lookup(&((struct sockaddr_in6 *)(void *) + rt_key(rt))->sin6_addr, rt->rt_ifp); if ((ln && ln->ln_router) || dr) { /* @@ -1006,10 +1641,16 @@ nd6_free( * is in the Default Router List. * See a corresponding comment in nd6_na_input(). */ + RT_UNLOCK(rt); + lck_mtx_unlock(nd6_mutex); rt6_flush(&in6, rt->rt_ifp); + lck_mtx_lock(nd6_mutex); + } else { + RT_UNLOCK(rt); } if (dr) { + NDDR_REMREF(dr); /* * Unreachablity of a router might affect the default * router selection and on-link detection of advertised @@ -1023,6 +1664,7 @@ nd6_free( * Below the state will be set correctly, * or the entry itself will be deleted. */ + RT_LOCK_SPIN(rt); ln->ln_state = ND6_LLINFO_INCOMPLETE; /* @@ -1031,51 +1673,36 @@ nd6_free( * before the default router selection, we perform * the check now. */ - pfxlist_onlink_check(1); + RT_UNLOCK(rt); + pfxlist_onlink_check(); - if (dr == TAILQ_FIRST(&nd_defrouter)) { - /* - * It is used as the current default router, - * so we have to move it to the end of the - * list and choose a new one. - * XXX: it is not very efficient if this is - * the only router. - */ - TAILQ_REMOVE(&nd_defrouter, dr, dr_entry); - TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry); - - defrouter_select(); - } + /* + * refresh default router list + */ + defrouter_select(rt->rt_ifp); } - lck_mtx_unlock(nd6_mutex); + 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). - */ - if (ln) - next = ln->ln_next; - else - next = 0; - + 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); } /* * Upper-layer reachability hint for Neighbor Unreachability Detection. * - * XXX cost-effective metods? + * XXX cost-effective methods? */ void nd6_nud_hint( @@ -1095,8 +1722,13 @@ nd6_nud_hint( if (!rt) { if (!dst6) return; - if (!(rt = nd6_lookup(dst6, 0, NULL, 0))) + /* 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 || @@ -1104,12 +1736,12 @@ nd6_nud_hint( !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, @@ -1118,13 +1750,24 @@ nd6_nud_hint( 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 = timenow.tv_sec + - nd_ifinfo[rt->rt_ifp->if_index].reachable; + if (ln->ln_expire) { + struct nd_ifinfo *ndi; + + lck_rw_lock_shared(nd_if_rwlock); + ndi = ND_IFINFO(rt->rt_ifp); + VERIFY(ndi != NULL && ndi->initialized); + lck_mtx_lock(&ndi->lock); + ln->ln_expire = timenow.tv_sec + ndi->reachable; + lck_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); + } +done: + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); } void @@ -1134,13 +1777,15 @@ nd6_rtrequest( __unused struct sockaddr *sa) { struct sockaddr *gate = rt->rt_gateway; - struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo; + 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; + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED); + RT_LOCK_ASSERT_HELD(rt); if ((rt->rt_flags & RTF_GATEWAY)) return; @@ -1156,9 +1801,27 @@ nd6_rtrequest( return; } - if (req == RTM_RESOLVE && - (nd6_need_cache(ifp) == 0 || /* stf case */ - !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp, 1))) { + if (req == RTM_RESOLVE) { + int no_nd_cache; + + if (!nd6_need_cache(ifp)) { /* stf case */ + no_nd_cache = 1; + } else { + struct sockaddr_in6 sin6; + + rtkey_to_sa6(rt, &sin6); + /* + * nd6_is_addr_neighbor() may call nd6_lookup(), + * therefore we drop rt_lock to avoid deadlock + * during the lookup. + */ + RT_ADDREF_LOCKED(rt); + RT_UNLOCK(rt); + no_nd_cache = !nd6_is_addr_neighbor(&sin6, 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). @@ -1173,8 +1836,10 @@ nd6_rtrequest( * so that ndp(8) will not try to get the neighbor information * of the destination. */ - rt->rt_flags &= ~RTF_LLINFO; - return; + if (no_nd_cache) { + rt->rt_flags &= ~RTF_LLINFO; + return; + } } getmicrotime(&timenow); @@ -1187,30 +1852,26 @@ nd6_rtrequest( * SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff) * rt->rt_flags |= RTF_CLONING; */ - if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) { + if ((rt->rt_flags & RTF_CLONING) || + ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) { /* - * Case 1: This route should come from - * a route to interface. RTF_LLINFO flag is set - * for a host route whose destination should be - * treated as on-link. + * Case 1: This route should come from a route to + * interface (RTF_CLONING case) or the route should be + * treated as on-link but is currently not + * (RTF_LLINFO && ln == NULL case). */ - 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 = timenow.tv_sec; -#if 1 - if (ln && ln->ln_expire == 0) { - /* kludge for desktops */ -#if 0 - printf("nd6_rtequest: 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)) break; } @@ -1232,15 +1893,6 @@ nd6_rtrequest( * (or should we allow proxy ND configuration only for * routers? there's no mention about proxy ND from hosts) */ -#if 0 - /* XXX it does not work */ - if (rt->rt_flags & RTF_ANNOUNCE) - nd6_na_output(ifp, - &SIN6(rt_key(rt))->sin6_addr, - &SIN6(rt_key(rt))->sin6_addr, - ip6_forwarding ? ND_NA_FLAG_ROUTER : 0, - 1, NULL); -#endif /* FALLTHROUGH */ case RTM_RESOLVE: if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) { @@ -1250,9 +1902,12 @@ nd6_rtrequest( */ if (gate->sa_family != AF_LINK || gate->sa_len < sizeof(null_sdl)) { - log(LOG_DEBUG, - "nd6_rtrequest: bad gateway value: %s\n", - if_name(ifp)); + /* Don't complain in case of RTM_ADD */ + if (req == RTM_RESOLVE) { + log(LOG_DEBUG, + "nd6_rtrequest: bad gateway " + "value: %s\n", if_name(ifp)); + } break; } SDL(gate)->sdl_type = ifp->if_type; @@ -1264,12 +1919,16 @@ nd6_rtrequest( * 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_get_ri = nd6_llinfo_get_ri; + rt->rt_llinfo_get_iflri = nd6_llinfo_get_iflri; + rt->rt_llinfo_purge = nd6_llinfo_purge; + rt->rt_llinfo_free = nd6_llinfo_free; + nd6_inuse++; nd6_allocated++; Bzero(ln, sizeof(*ln)); @@ -1289,13 +1948,44 @@ nd6_rtrequest( * initialized in rtrequest(), so rt_expire is 0. */ ln->ln_state = ND6_LLINFO_NOSTATE; - ln->ln_expire = timenow.tv_sec; + /* 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 @@ -1313,6 +2003,22 @@ nd6_rtrequest( SDL(gate)->sdl_alen = ifp->if_addrlen; } if (nd6_useloopback) { + if (rt->rt_ifp != lo_ifp) { + /* + * Purge any link-layer info caching. + */ + if (rt->rt_llinfo_purge != NULL) + rt->rt_llinfo_purge(rt); + + /* + * Adjust route ref count for the + * interfaces. + */ + if (rt->rt_if_ref_fn != NULL) { + rt->rt_if_ref_fn(lo_ifp, 1); + rt->rt_if_ref_fn(rt->rt_ifp, -1); + } + } rt->rt_ifp = lo_ifp; /* XXX */ /* * Make sure rt_ifa be equal to the ifaddr @@ -1326,6 +2032,7 @@ nd6_rtrequest( rtsetifa(rt, ifa); } } + IFA_REMREF(ifa); } else if (rt->rt_flags & RTF_ANNOUNCE) { ln->ln_expire = 0; ln->ln_state = ND6_LLINFO_REACHABLE; @@ -1334,26 +2041,30 @@ nd6_rtrequest( /* join solicited node multicast for proxy ND */ if (ifp->if_flags & IFF_MULTICAST) { struct in6_addr llsol; + struct in6_multi *in6m; int error; llsol = SIN6(rt_key(rt))->sin6_addr; - llsol.s6_addr16[0] = htons(0xff02); - llsol.s6_addr16[1] = htons(ifp->if_index); + llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; llsol.s6_addr32[1] = 0; llsol.s6_addr32[2] = htonl(1); llsol.s6_addr8[12] = 0xff; - - if (!in6_addmulti(&llsol, ifp, &error, 0)) { + if (in6_setscope(&llsol, ifp, NULL)) + break; + error = in6_mc_join(ifp, &llsol, NULL, &in6m, 0); + if (error) { nd6log((LOG_ERR, "%s: failed to join " "%s (errno=%d)\n", if_name(ifp), ip6_sprintf(&llsol), error)); + } else { + IN6M_REMREF(in6m); } } } break; case RTM_DELETE: - if (!ln) + if (ln == NULL) break; /* leave from solicited node multicast for proxy ND */ if ((rt->rt_flags & RTF_ANNOUNCE) != 0 && @@ -1362,224 +2073,436 @@ nd6_rtrequest( struct in6_multi *in6m; llsol = SIN6(rt_key(rt))->sin6_addr; - llsol.s6_addr16[0] = htons(0xff02); - llsol.s6_addr16[1] = htons(ifp->if_index); + llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; llsol.s6_addr32[1] = 0; 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, 0); + if (in6_setscope(&llsol, ifp, NULL) == 0) { + in6_multihead_lock_shared(); + IN6_LOOKUP_MULTI(&llsol, ifp, in6m); + in6_multihead_lock_done(); + if (in6m != NULL) { + in6_mc_leave(in6m, NULL); + IN6M_REMREF(in6m); + } + } } 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); + + /* + * Purge any link-layer info caching. + */ + if (rt->rt_llinfo_purge != NULL) + rt->rt_llinfo_purge(rt); + rt->rt_flags &= ~RTF_LLINFO; - if (ln->ln_hold) + if (ln->ln_hold != NULL) { m_freem(ln->ln_hold); - ln->ln_hold = NULL; - R_Free((caddr_t)ln); + ln->ln_hold = NULL; + } } } -int -nd6_ioctl( - u_long cmd, - caddr_t data, - struct ifnet *ifp) +static int +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; + struct in6_drlist_32 *drl_32; + struct nd_defrouter *dr; + int i = 0; + + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); + + dr = TAILQ_FIRST(&nd_defrouter); + + /* For 64-bit process */ + if (data_is_64) { + struct in6_drlist_64 *drl_64; + + drl_64 = _MALLOC(sizeof (*drl_64), M_TEMP, M_WAITOK|M_ZERO); + if (drl_64 == NULL) + return (ENOMEM); + + /* preserve the interface name */ + bcopy(data, drl_64, sizeof (drl_64->ifname)); - switch (cmd) { - case SIOCGDRLST_IN6: - /* - * obsolete API, use sysctl under net.inet6.icmp6 - */ - lck_mtx_lock(nd6_mutex); - bzero(drl, sizeof(*drl)); - dr = TAILQ_FIRST(&nd_defrouter); 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); } - lck_mtx_unlock(nd6_mutex); - 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)); - lck_mtx_lock(nd6_mutex); - pr = nd_prefix.lh_first; + bcopy(drl_64, data, sizeof (*drl_64)); + _FREE(drl_64, M_TEMP); + return (0); + } + + /* For 32-bit process */ + drl_32 = _MALLOC(sizeof (*drl_32), M_TEMP, M_WAITOK|M_ZERO); + if (drl_32 == NULL) + return (ENOMEM); + + /* preserve the interface name */ + bcopy(data, drl_32, sizeof (drl_32->ifname)); + + 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); + } + bcopy(drl_32, data, sizeof (*drl_32)); + _FREE(drl_32, M_TEMP); + return (0); +} + +/* + * XXX meaning of fields, especialy "raflags", is very + * differnet between RA prefix list and RR/static prefix list. + * how about separating ioctls into two? + */ +static int +nd6_siocgprlst(void *data, int data_is_64) +{ + struct in6_prlist_32 *prl_32; + struct nd_prefix *pr; + int i = 0; + + lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED); + + pr = nd_prefix.lh_first; + + /* For 64-bit process */ + if (data_is_64) { + struct in6_prlist_64 *prl_64; + + prl_64 = _MALLOC(sizeof (*prl_64), M_TEMP, M_WAITOK|M_ZERO); + if (prl_64 == NULL) + return (ENOMEM); + + /* preserve the interface name */ + bcopy(data, prl_64, sizeof (prl_64->ifname)); + while (pr && i < PRLSTSIZ) { struct nd_pfxrouter *pfr; int j; - (void)in6_embedscope(&prl->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; + NDPR_LOCK(pr); + (void) in6_embedscope(&prl_64->prefix[i].prefix, + &pr->ndpr_prefix, NULL, NULL, NULL); + 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; + NDPR_UNLOCK(pr); i++; pr = pr->ndpr_next; } - { - struct rr_prefix *rpp; + bcopy(prl_64, data, sizeof (*prl_64)); + _FREE(prl_64, M_TEMP); + return (0); + } - for (rpp = LIST_FIRST(&rr_prefix); rpp; - rpp = LIST_NEXT(rpp, rp_entry)) { - if (i >= PRLSTSIZ) - break; - (void)in6_embedscope(&prl->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; - i++; + /* For 32-bit process */ + prl_32 = _MALLOC(sizeof (*prl_32), M_TEMP, M_WAITOK|M_ZERO); + if (prl_32 == NULL) + return (ENOMEM); + + /* preserve the interface name */ + bcopy(data, prl_32, sizeof (prl_32->ifname)); + + while (pr && i < PRLSTSIZ) { + struct nd_pfxrouter *pfr; + int j; + + NDPR_LOCK(pr); + (void) in6_embedscope(&prl_32->prefix[i].prefix, + &pr->ndpr_prefix, NULL, 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; + NDPR_UNLOCK(pr); + + i++; + pr = pr->ndpr_next; + } + bcopy(prl_32, data, sizeof (*prl_32)); + _FREE(prl_32, M_TEMP); + return (0); +} + +int +nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) +{ + struct nd_defrouter *dr; + struct nd_prefix *pr; + struct rtentry *rt; + int i = ifp->if_index, error = 0; + + switch (cmd) { + case SIOCGDRLST_IN6_32: /* struct in6_drlist_32 */ + case SIOCGDRLST_IN6_64: /* struct in6_drlist_64 */ + /* + * obsolete API, use sysctl under net.inet6.icmp6 + */ + lck_mtx_lock(nd6_mutex); + error = nd6_siocgdrlst(data, cmd == SIOCGDRLST_IN6_64); lck_mtx_unlock(nd6_mutex); break; - case OSIOCGIFINFO_IN6: - if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { - error = EINVAL; - break; - } - ndi->ndi.linkmtu = IN6_LINKMTU(ifp); - 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; + + case SIOCGPRLST_IN6_32: /* struct in6_prlist_32 */ + case SIOCGPRLST_IN6_64: /* struct in6_prlist_64 */ + /* + * obsolete API, use sysctl under net.inet6.icmp6 + */ + lck_mtx_lock(nd6_mutex); + error = nd6_siocgprlst(data, cmd == SIOCGPRLST_IN6_64); + lck_mtx_unlock(nd6_mutex); break; - case SIOCGIFINFO_IN6: - if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { + + case OSIOCGIFINFO_IN6: /* struct in6_ondireq */ + case SIOCGIFINFO_IN6: { /* struct in6_ondireq */ + u_int32_t linkmtu; + struct in6_ondireq *ondi = (struct in6_ondireq *)(void *)data; + struct nd_ifinfo *ndi; + /* + * 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); + ndi = ND_IFINFO(ifp); + if (!nd_ifinfo || i >= nd_ifinfo_indexlim || + !ndi->initialized) { + lck_rw_done(nd_if_rwlock); error = EINVAL; break; } - ndi->ndi = nd_ifinfo[ifp->if_index]; + lck_mtx_lock(&ndi->lock); + linkmtu = IN6_LINKMTU(ifp); + bcopy(&linkmtu, &ondi->ndi.linkmtu, sizeof (linkmtu)); + bcopy(&nd_ifinfo[i].maxmtu, &ondi->ndi.maxmtu, + sizeof (u_int32_t)); + bcopy(&nd_ifinfo[i].basereachable, &ondi->ndi.basereachable, + sizeof (u_int32_t)); + bcopy(&nd_ifinfo[i].reachable, &ondi->ndi.reachable, + sizeof (u_int32_t)); + bcopy(&nd_ifinfo[i].retrans, &ondi->ndi.retrans, + sizeof (u_int32_t)); + bcopy(&nd_ifinfo[i].flags, &ondi->ndi.flags, + sizeof (u_int32_t)); + bcopy(&nd_ifinfo[i].recalctm, &ondi->ndi.recalctm, + sizeof (int)); + ondi->ndi.chlim = nd_ifinfo[i].chlim; + ondi->ndi.receivedra = 0; + lck_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); break; - case SIOCSIFINFO_FLAGS: - /* XXX: almost all other fields of ndi->ndi is unused */ - if (!nd_ifinfo || i >= nd_ifinfo_indexlim) { + } + + case SIOCSIFINFO_FLAGS: { /* struct in6_ndireq */ + struct in6_ndireq *cndi = (struct in6_ndireq *)(void *)data; + u_int32_t oflags, flags; + struct nd_ifinfo *ndi; + + /* XXX: almost all other fields of cndi->ndi is unused */ + lck_rw_lock_shared(nd_if_rwlock); + ndi = ND_IFINFO(ifp); + if (!nd_ifinfo || i >= nd_ifinfo_indexlim || + !ndi->initialized) { + lck_rw_done(nd_if_rwlock); error = EINVAL; break; } - nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags; + lck_mtx_lock(&ndi->lock); + oflags = nd_ifinfo[i].flags; + bcopy(&cndi->ndi.flags, &nd_ifinfo[i].flags, sizeof (flags)); + flags = nd_ifinfo[i].flags; + lck_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); + + if (oflags == flags) + break; + + error = nd6_setifinfo(ifp, oflags, flags); break; - case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ + } + + case SIOCSNDFLUSH_IN6: /* struct in6_ifreq */ /* flush default router list */ /* * xxx sumikawa: should not delete route if default * route equals to the top of default router list */ - bzero(&any, sizeof(any)); lck_mtx_lock(nd6_mutex); - defrouter_delreq(&any, 1); - defrouter_select(); + defrouter_reset(); + defrouter_select(ifp); lck_mtx_unlock(nd6_mutex); /* xxx sumikawa: flush prefix list */ break; - case SIOCSPFXFLUSH_IN6: - { + + case SIOCSPFXFLUSH_IN6: { /* struct in6_ifreq */ /* flush all the prefix advertised by routers */ struct nd_prefix *next; - lck_mtx_lock(nd6_mutex); + lck_mtx_lock(nd6_mutex); for (pr = nd_prefix.lh_first; pr; pr = next) { - struct in6_ifaddr *ia, *ia_next; + struct in6_ifaddr *ia; next = pr->ndpr_next; - if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) + NDPR_LOCK(pr); + if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) { + NDPR_UNLOCK(pr); continue; /* XXX */ - + } + if (ifp != lo_ifp && pr->ndpr_ifp != ifp) { + NDPR_UNLOCK(pr); + continue; + } /* do we really have to remove addresses as well? */ - for (ia = in6_ifaddrs; ia; ia = ia_next) { - /* ia might be removed. keep the next ptr. */ - ia_next = ia->ia_next; + NDPR_ADDREF_LOCKED(pr); + NDPR_UNLOCK(pr); + lck_rw_lock_exclusive(&in6_ifaddr_rwlock); + ia = in6_ifaddrs; + while (ia != NULL) { + IFA_LOCK(&ia->ia_ifa); + if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) { + IFA_UNLOCK(&ia->ia_ifa); + ia = ia->ia_next; + continue; + } - if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) + if (ia->ia6_ndpr == pr) { + IFA_ADDREF_LOCKED(&ia->ia_ifa); + IFA_UNLOCK(&ia->ia_ifa); + lck_rw_done(&in6_ifaddr_rwlock); + lck_mtx_unlock(nd6_mutex); + in6_purgeaddr(&ia->ia_ifa); + IFA_REMREF(&ia->ia_ifa); + lck_mtx_lock(nd6_mutex); + lck_rw_lock_exclusive(&in6_ifaddr_rwlock); + /* + * Purging the address caused + * in6_ifaddr_rwlock to be + * dropped and + * reacquired; therefore search again + * from the beginning of in6_ifaddrs. + * The same applies for the prefix list. + */ + ia = in6_ifaddrs; + next = nd_prefix.lh_first; continue; - if (ia->ia6_ndpr == pr) - in6_purgeaddr(&ia->ia_ifa, 1); + } + IFA_UNLOCK(&ia->ia_ifa); + ia = ia->ia_next; } - prelist_remove(pr, 1); + lck_rw_done(&in6_ifaddr_rwlock); + NDPR_LOCK(pr); + prelist_remove(pr); + NDPR_UNLOCK(pr); + /* + * If we were trying to restart this loop + * above by changing the value of 'next', we might + * end up freeing the only element on the list + * when we call NDPR_REMREF(). + * When this happens, we also have get out of this + * loop because we have nothing else to do. + */ + if (pr == next) + next = NULL; + NDPR_REMREF(pr); } lck_mtx_unlock(nd6_mutex); break; - } - case SIOCSRTRFLUSH_IN6: - { + } + + case SIOCSRTRFLUSH_IN6: { /* struct in6_ifreq */ /* flush all the default routers */ struct nd_defrouter *next; @@ -1591,57 +2514,137 @@ nd6_ioctl( */ for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) { next = TAILQ_NEXT(dr, dr_entry); - defrtrlist_del(dr, 1); + if (ifp == lo_ifp || dr->ifp == ifp) + defrtrlist_del(dr); } - defrtrlist_del(TAILQ_FIRST(&nd_defrouter), 1); + if (ifp == lo_ifp || + TAILQ_FIRST(&nd_defrouter)->ifp == ifp) + defrtrlist_del(TAILQ_FIRST(&nd_defrouter)); } lck_mtx_unlock(nd6_mutex); break; - } - case SIOCGNBRINFO_IN6: - { + } + + case SIOCGNBRINFO_IN6_32: { /* struct in6_nbrinfo_32 */ struct llinfo_nd6 *ln; - struct in6_addr nb_addr = nbi->addr; /* make local for safety */ + struct in6_nbrinfo_32 nbi_32; + struct in6_addr nb_addr; /* make local for safety */ + bcopy(data, &nbi_32, sizeof (nbi_32)); + 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)) { - u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2]; + if (IN6_IS_ADDR_LINKLOCAL(&nbi_32.addr) || + IN6_IS_ADDR_MC_LINKLOCAL(&nbi_32.addr)) { + u_int16_t *idp = + (u_int16_t *)(void *)&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; } - 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; - + 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); + bcopy(&nbi_32, data, sizeof (nbi_32)); break; - } - case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ - ndif->ifindex = nd6_defifindex; + } + + case SIOCGNBRINFO_IN6_64: { /* struct in6_nbrinfo_64 */ + struct llinfo_nd6 *ln; + struct in6_nbrinfo_64 nbi_64; + struct in6_addr nb_addr; /* make local for safety */ + + bcopy(data, &nbi_64, sizeof (nbi_64)); + 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 *)(void *)&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); + bcopy(&nbi_64, data, sizeof (nbi_64)); break; - case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ - return(nd6_setdefaultiface(ndif->ifindex)); + } + + case SIOCGDEFIFACE_IN6_32: /* struct in6_ndifreq_32 */ + case SIOCGDEFIFACE_IN6_64: { /* struct in6_ndifreq_64 */ + struct in6_ndifreq_64 *ndif_64 = + (struct in6_ndifreq_64 *)(void *)data; + struct in6_ndifreq_32 *ndif_32 = + (struct in6_ndifreq_32 *)(void *)data; + + if (cmd == SIOCGDEFIFACE_IN6_64) { + u_int64_t j = nd6_defifindex; + bcopy(&j, &ndif_64->ifindex, sizeof (j)); + } else { + bcopy(&nd6_defifindex, &ndif_32->ifindex, + sizeof (u_int32_t)); + } break; } - return(error); + + case SIOCSDEFIFACE_IN6_32: /* struct in6_ndifreq_32 */ + case SIOCSDEFIFACE_IN6_64: { /* struct in6_ndifreq_64 */ + struct in6_ndifreq_64 *ndif_64 = + (struct in6_ndifreq_64 *)(void *)data; + struct in6_ndifreq_32 *ndif_32 = + (struct in6_ndifreq_32 *)(void *)data; + u_int32_t idx; + + if (cmd == SIOCSDEFIFACE_IN6_64) { + u_int64_t j; + bcopy(&ndif_64->ifindex, &j, sizeof (j)); + idx = (u_int32_t)j; + } else { + bcopy(&ndif_32->ifindex, &idx, sizeof (idx)); + } + + error = nd6_setdefaultiface(idx); + return (error); + /* NOTREACHED */ + } + } + return (error); } /* * Create neighbor cache entry and cache link-layer address, * on reception of inbound ND6 packets. (RS/RA/NS/redirect) */ -struct rtentry * +void nd6_cache_lladdr( struct ifnet *ifp, struct in6_addr *from, @@ -1660,14 +2663,14 @@ nd6_cache_lladdr( int newstate = 0; struct timeval timenow; - if (!ifp) + if (ifp == NULL) panic("ifp == NULL in nd6_cache_lladdr"); - if (!from) + if (from == NULL) panic("from == NULL in nd6_cache_lladdr"); /* 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 @@ -1680,39 +2683,36 @@ nd6_cache_lladdr( */ getmicrotime(&timenow); - lck_mtx_lock(rt_mtx); - rt = nd6_lookup(from, 0, ifp, 1); - if (!rt) { -#if 0 - /* nothing must be done if there's no lladdr */ - if (!lladdr || !lladdrlen) - return NULL; -#endif - - rt = nd6_lookup(from, 1, ifp, 1); + rt = nd6_lookup(from, 0, ifp, 0); + if (rt == NULL) { + 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) { - lck_mtx_unlock(rt_mtx); - return NULL; + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); + return; } is_newentry = 0; } - lck_mtx_unlock(rt_mtx); - - if (!rt) - return NULL; + if (rt == NULL) + return; 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; - if (!ln) + if (ln == NULL) goto fail; - if (!rt->rt_gateway) + if (rt->rt_gateway == NULL) goto fail; if (rt->rt_gateway->sa_family != AF_LINK) goto fail; @@ -1745,18 +2745,21 @@ fail: */ sdl->sdl_alen = ifp->if_addrlen; bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen); + + /* cache the gateway (sender HW) address */ + nd6_llreach_alloc(rt, ifp, LLADDR(sdl), sdl->sdl_alen, FALSE); } if (!is_newentry) { - if ((!olladdr && lladdr) /* (3) */ - || (olladdr && lladdr && llchange)) { /* (5) */ + if ((!olladdr && lladdr != NULL) || /* (3) */ + (olladdr && lladdr != NULL && llchange)) { /* (5) */ do_update = 1; newstate = ND6_LLINFO_STALE; } else /* (1-2,4) */ do_update = 0; } else { do_update = 1; - if (!lladdr) /* (6) */ + if (lladdr == NULL) /* (6) */ newstate = ND6_LLINFO_NOSTATE; else /* (7) */ newstate = ND6_LLINFO_STALE; @@ -1769,6 +2772,7 @@ 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, @@ -1776,16 +2780,19 @@ fail: * meaningless. */ ln->ln_expire = timenow.tv_sec + nd6_gctimer; + ln->ln_hold = NULL; + + if (m != NULL) { + struct sockaddr_in6 sin6; - if (ln->ln_hold) { + rtkey_to_sa6(rt, &sin6); /* * 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, 0); - ln->ln_hold = NULL; + RT_UNLOCK(rt); + nd6_output(ifp, ifp, m, &sin6, rt, NULL); + RT_LOCK(rt); } } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { /* probe right away */ @@ -1817,7 +2824,7 @@ fail: * 0 n y -- (3) c s s * 0 y y n (4) c s s * 0 y y y (5) c s s - * 1 -- n -- (6) c c c s + * 1 -- n -- (6) c c c s * 1 -- y -- (7) c c s c s * * (c=clear s=set) @@ -1833,8 +2840,8 @@ fail: case ND_REDIRECT: /* * If the icmp is a redirect to a better router, always set the - * is_router flag. Otherwise, if the entry is newly created, - * clear the flag. [RFC 2461, sec 8.3] + * is_router flag. Otherwise, if the entry is newly created, + * clear the flag. [RFC 2461, sec 8.3] */ if (code == ND_REDIRECT_ROUTER) ln->ln_router = 1; @@ -1851,8 +2858,8 @@ 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; @@ -1864,22 +2871,21 @@ fail: * created, it might affect the selection policy. * Question: can we restrict the first condition to the "is_newentry" * case? - * XXX: when we hear an RA from a new router with the link-layer - * address option, defrouter_select() is called twice, since - * defrtrlist_update called the function as well. However, I believe - * we can compromise the overhead, since it only happens the first - * time. - * XXX: although defrouter_select() should not have a bad effect - * for those are not autoconfigured hosts, we explicitly avoid such - * cases for safety. + * + * Note: Perform default router selection even when we are a router, + * if Scoped Routing is enabled. */ - if (do_update && ln->ln_router && !ip6_forwarding && (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { + if (do_update && ln->ln_router && + (ip6_doscopedroute || !ip6_forwarding)) { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); lck_mtx_lock(nd6_mutex); - defrouter_select(); + defrouter_select(ifp); lck_mtx_unlock(nd6_mutex); + } else { + RT_REMREF_LOCKED(rt); + RT_UNLOCK(rt); } - - return rt; } static void @@ -1889,11 +2895,14 @@ nd6_slowtimo( int i; struct nd_ifinfo *nd6if; - lck_mtx_lock(nd6_mutex); + 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->initialized) + break; + lck_mtx_lock(&nd6if->lock); if (nd6if->basereachable && /* already initialized */ (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { /* @@ -1905,59 +2914,88 @@ nd6_slowtimo( nd6if->recalctm = nd6_recalc_reachtm_interval; nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); } + lck_mtx_unlock(&nd6if->lock); } - lck_mtx_unlock(nd6_mutex); + 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( - struct ifnet *ifp, - struct ifnet *origifp, - struct mbuf *m0, - struct sockaddr_in6 *dst, - struct rtentry *rt0, - int locked) +nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, + struct sockaddr_in6 *dst, struct rtentry *hint0, struct flowadv *adv) { 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; + struct nd_ifinfo *ndi; - 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 route_to_gwroute(). */ - lck_mtx_lock(rt_mtx); - if (rt) { - if ((rt->rt_flags & RTF_UP) == 0) { - if ((rt0 = rt = rtalloc1_locked((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_scoped((struct sockaddr *)dst, + 1, 0, ifp->if_index)) != NULL) { + RT_LOCK_SPIN(rt); if (rt->rt_ifp != ifp) { /* XXX: loop care? */ - lck_mtx_unlock(rt_mtx); - return nd6_output(ifp, origifp, m0, - dst, rt, locked); + RT_UNLOCK(rt); + error = nd6_output(ifp, origifp, m0, + dst, rt, adv); + rtfree(rt); + return (error); } } else { - lck_mtx_unlock(rt_mtx); 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; + + rtgw_to_sa6(rt, &gw6); + /* + * 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 @@ -1967,31 +3005,129 @@ nd6_output( * 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, 1) || - 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. */ - lck_mtx_unlock(rt_mtx); + if (ia6 != NULL) + IFA_REMREF(&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 *)(void *)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_locked(rt); rt = rt0; - lookup: rt->rt_gwroute = rtalloc1_locked(rt->rt_gateway, 1, 0UL); - if ((rt = rt->rt_gwroute) == 0) { - lck_mtx_unlock(rt_mtx); + } + /* 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)) { + rt->rt_gwroute = NULL; + RT_UNLOCK(gwrt); + RT_UNLOCK(rt); + rtfree(gwrt); +lookup: + lck_mtx_lock(rnh_lock); + gwrt = rtalloc1_scoped_locked( + (struct sockaddr *)&gw6, 1, 0, + ifp->if_index); + + 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_locked(gwrt); + lck_mtx_unlock(rnh_lock); senderr(EHOSTUNREACH); } + VERIFY(gwrt != NULL); + /* + * Set gateway route; callee adds ref to gwrt; + * gwrt has an extra ref from rtalloc1() for + * this routine. + */ + rt_set_gwroute(rt, rt_key(rt), gwrt); + RT_UNLOCK(rt); + lck_mtx_unlock(rnh_lock); + /* Remember to release/free "rt" at the end */ + rtrele = rt; + rt = gwrt; + } else { + RT_ADDREF_LOCKED(gwrt); + RT_UNLOCK(gwrt); + RT_UNLOCK(rt); + /* Remember to release/free "rt" at the end */ + rtrele = rt; + rt = gwrt; + } + VERIFY(rt == gwrt); + + /* + * This is an opportunity to revalidate the parent + * route's gwroute, in case it now points to a dead + * route entry. Parent route won't go away since the + * clone (hint) holds a reference to it. rt == gwrt. + */ + RT_LOCK_SPIN(hint); + if ((hint->rt_flags & (RTF_WASCLONED | RTF_UP)) == + (RTF_WASCLONED | RTF_UP)) { + struct rtentry *prt = hint->rt_parent; + VERIFY(prt != NULL); + + RT_CONVERT_LOCK(hint); + RT_ADDREF(prt); + RT_UNLOCK(hint); + rt_revalidate_gwroute(prt, rt); + RT_REMREF(prt); + } else { + RT_UNLOCK(hint); + } + + RT_LOCK_SPIN(rt); + /* rt == gwrt; if it is now down, give up */ + if (!(rt->rt_flags & RTF_UP)) { + RT_UNLOCK(rt); + rtfree(rt); + rt = NULL; + /* "rtrele" == original "rt" */ + senderr(EHOSTUNREACH); } } + + /* Become a regular mutex */ + RT_CONVERT_LOCK(rt); } /* @@ -2002,28 +3138,63 @@ nd6_output( */ /* 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 { + struct sockaddr_in6 sin6; + /* + * Clear out Scope ID field in case it is set. + */ + sin6 = *dst; + sin6.sin6_scope_id = 0; /* * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), * 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, 1) && - (rt = nd6_lookup(&dst->sin6_addr, 1, ifp, 1)) != NULL) - ln = (struct llinfo_nd6 *)rt->rt_llinfo; + if (rt != NULL) + RT_UNLOCK(rt); + if (nd6_is_addr_neighbor(&sin6, 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); + } } - lck_mtx_unlock(rt_mtx); + if (!ln || !rt) { + if (rt != NULL) + RT_UNLOCK(rt); + lck_rw_lock_shared(nd_if_rwlock); + ndi = ND_IFINFO(ifp); + VERIFY(ndi != NULL && ndi->initialized); + lck_mtx_lock(&ndi->lock); if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && - !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) { + !(ndi->flags & ND6_IFF_PERFORMNUD)) { + lck_mtx_unlock(&ndi->lock); + 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_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); goto sendpkt; /* send anyway */ } @@ -2034,7 +3205,7 @@ nd6_output( if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && ln->ln_state < ND6_LLINFO_REACHABLE) { ln->ln_state = ND6_LLINFO_STALE; - ln->ln_expire = timenow.tv_sec + nd6_gctimer; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_gctimer); } /* @@ -2047,7 +3218,7 @@ nd6_output( if (ln->ln_state == ND6_LLINFO_STALE) { ln->ln_asked = 0; ln->ln_state = ND6_LLINFO_DELAY; - ln->ln_expire = timenow.tv_sec + nd6_delay; + ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_delay); } /* @@ -2055,8 +3226,23 @@ nd6_output( * (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 @@ -2072,45 +3258,85 @@ nd6_output( 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 < timenow.tv_sec) { - ln->ln_asked++; - ln->ln_expire = timenow.tv_sec + - nd_ifinfo[ifp->if_index].retrans / 1000; - nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0, locked); - } + 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); + ndi = ND_IFINFO(ifp); + VERIFY(ndi != NULL && ndi->initialized); + lck_mtx_lock(&ndi->lock); + ln->ln_expire = timenow.tv_sec + ndi->retrans / 1000; + lck_mtx_unlock(&ndi->lock); + lck_rw_done(nd_if_rwlock); + RT_UNLOCK(rt); + /* We still have a reference on rt (for ln) */ + if (ip6_forwarding) + nd6_prproxy_ns_output(ifp, NULL, &dst->sin6_addr, ln); + else + nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); + } else { + RT_UNLOCK(rt); } - return(0); - - sendpkt: -#ifdef __APPLE__ - - /* Make sure the HW checksum flags are cleaned before sending the packet */ + /* + * 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); - m->m_pkthdr.csum_data = 0; - m->m_pkthdr.csum_flags = 0; + error = 0; + goto release; + +sendpkt: + if (rt != NULL) + RT_LOCK_ASSERT_NOTHELD(rt); + + /* discard the packet if IPv6 operation is disabled on the interface */ + lck_rw_lock_shared(nd_if_rwlock); + ndi = ND_IFINFO(ifp); + VERIFY(ndi != NULL && ndi->initialized); + /* test is done here without holding ndi lock, for performance */ + if (ndi->flags & ND6_IFF_IFDISABLED) { + lck_rw_done(nd_if_rwlock); + error = ENETDOWN; /* better error? */ + goto bad; + } + lck_rw_done(nd_if_rwlock); if ((ifp->if_flags & IFF_LOOPBACK) != 0) { - m->m_pkthdr.rcvif = origifp; /* forwarding rules require the original scope_id */ - 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); - return error; + /* forwarding rules require the original scope_id */ + m->m_pkthdr.rcvif = origifp; + error = dlil_output(origifp, PF_INET6, m, (caddr_t)rt, + (struct sockaddr *)dst, 0, adv); + goto release; } else { /* Do not allow loopback address to wind up on a wire */ struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); - + 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 + /* + * 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; @@ -2118,26 +3344,51 @@ nd6_output( } } - m->m_pkthdr.rcvif = 0; - 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); - return(error); -#else - if ((ifp->if_flags & IFF_LOOPBACK) != 0) { - return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, - rt)); + if (rt != NULL) { + RT_LOCK_SPIN(rt); + /* Mark use timestamp */ + if (rt->rt_llinfo != NULL) + nd6_llreach_use(rt->rt_llinfo); + RT_UNLOCK(rt); } - return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt)); -#endif - - bad: - if (m) + + if (hint && nstat_collect) + nstat_route_tx(hint, 1, m->m_pkthdr.len, 0); + + m->m_pkthdr.rcvif = NULL; + error = dlil_output(ifp, PF_INET6, m, (caddr_t)rt, + (struct sockaddr *)dst, 0, adv); + 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 @@ -2162,6 +3413,12 @@ nd6_need_cache( case IFT_IEEE80211: #endif case IFT_GIF: /* XXX need more cases? */ + case IFT_PPP: +#if IFT_TUNNEL + case IFT_TUNNEL: +#endif + case IFT_BRIDGE: + case IFT_CELLULAR: return(1); default: return(0); @@ -2188,6 +3445,7 @@ nd6_storelladdr( #if IFT_IEEE80211 case IFT_IEEE80211: #endif + case IFT_BRIDGE: ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, desten); return(1); @@ -2207,209 +3465,360 @@ nd6_storelladdr( /* this could happen, if we could not allocate memory */ return(0); /* caller will free mbuf */ } + RT_LOCK(rt); if (rt->rt_gateway->sa_family != AF_LINK) { printf("nd6_storelladdr: something odd happens\n"); + 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"); + RT_UNLOCK(rt); return(0); /* caller will free mbuf */ } bcopy(LLADDR(sdl), desten, sdl->sdl_alen); + RT_UNLOCK(rt); return(1); } -extern errno_t arp_route_to_gateway_route(const struct sockaddr *net_dest, - route_t hint, route_t *out_route); - +/* + * 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) +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; - + return (EAFNOSUPPORT); + if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING)) - return ENETDOWN; - - if (hint) { - result = arp_route_to_gateway_route((const struct sockaddr*)ip6_dest, hint, &route); + return (ENETDOWN); + + if (hint != NULL) { + /* + * Callee holds a reference on the route and returns + * with the route entry locked, upon success. + */ + result = route_to_gwroute((const struct sockaddr *)ip6_dest, + hint, &route); if (result != 0) - return result; + return (result); + if (route != NULL) + RT_LOCK_ASSERT_HELD(route); } - + if ((packet->m_flags & M_MCAST) != 0) { - return dlil_resolve_multi(ifp, (const struct sockaddr*)ip6_dest, - (struct sockaddr *)ll_dest, ll_dest_len); + 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 */ - return ENOBUFS; + /* + * This could happen, if we could not allocate memory or + * if route_to_gwroute() didn't return a route. + */ + result = ENOBUFS; + goto release; } - - lck_mtx_lock(rt_mtx); - + if (route->rt_gateway->sa_family != AF_LINK) { printf("nd6_lookup_ipv6: gateway address not AF_LINK\n"); result = EADDRNOTAVAIL; - goto done; + goto release; } - + sdl = SDL(route->rt_gateway); if (sdl->sdl_alen == 0) { /* this should be impossible, but we bark here for debugging */ - printf("nd6_storelladdr: sdl_alen == 0\n"); + 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); -done: - lck_mtx_unlock(rt_mtx); - return result; +release: + if (route != NULL) { + if (route == hint) { + RT_REMREF_LOCKED(route); + RT_UNLOCK(route); + } else { + RT_UNLOCK(route); + rtfree(route); + } + } + return (result); +} + +int +nd6_setifinfo(struct ifnet *ifp, u_int32_t before, u_int32_t after) +{ + /* + * We only care about ND6_IFF_PROXY_PREFIXES for now. + */ + before &= ND6_IFF_PROXY_PREFIXES; + after &= ND6_IFF_PROXY_PREFIXES; + + if (before == after) + return (0); + + return (nd6_if_prproxy(ifp, ((int32_t)(after - before) > 0))); } SYSCTL_DECL(_net_inet6_icmp6); static int -nd6_sysctl_drlist SYSCTL_HANDLER_ARGS +nd6_sysctl_drlist SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) - int error; + 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; + return (EPERM); lck_mtx_lock(nd6_mutex); - 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; + 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 *)(void *)buf; + de = (struct in6_defrouter_64 *) + (void *)(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->stateflags = dr->stateflags; + d->stateflags &= ~NDDRF_PROCESSED; + 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 *)(void *)buf; + de_32 = (struct in6_defrouter_32 *) + (void *)(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->stateflags = dr->stateflags; + d_32->stateflags &= ~NDDRF_PROCESSED; + 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; + } } lck_mtx_unlock(nd6_mutex); - return error; + return (error); } static int -nd6_sysctl_prlist SYSCTL_HANDLER_ARGS +nd6_sysctl_prlist SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) - int error; + 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); lck_mtx_lock(nd6_mutex); + if (p64) { + struct in6_prefix_64 *p, *pe; - 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; + 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 = (struct in6_prefix *)buf; - pe = (struct in6_prefix *)(buf + sizeof(buf)); + p = (struct in6_prefix_64 *)(void *)buf; + pe = (struct in6_prefix_64 *) + (void *)(buf + sizeof (buf)); - if (p + 1 <= pe) { - bzero(p, sizeof(*p)); - sin6 = (struct sockaddr_in6 *)(p + 1); + 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) { + NDPR_LOCK(pr); + 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_addrcnt; + 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; + NDPR_UNLOCK(pr); + } 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 *)(void *)buf; + pe_32 = (struct in6_prefix_32 *) + (void *)(buf + sizeof (buf)); + + if (p_32 + 1 <= pe_32) { + bzero(p_32, sizeof (*p_32)); + sin6 = (struct sockaddr_in6 *)(p_32 + 1); + + NDPR_LOCK(pr); + 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_addrcnt; + 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; + NDPR_UNLOCK(pr); + } else { + panic("buffer too short"); + } + advance = sizeof (*p_32) + sizeof (*sin6) * advrtrs; + error = SYSCTL_OUT(req, buf, advance); + if (error) + break; + } } lck_mtx_unlock(nd6_mutex); - return error; + return (error); } SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, - CTLFLAG_RD, 0, 0, nd6_sysctl_drlist, "S,in6_defrouter",""); + CTLFLAG_RD | CTLFLAG_LOCKED, 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",""); + CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, nd6_sysctl_prlist, "S,in6_defrouter","");