Please note that diffs are not public domain; they are subject to the copyright notices on the relevant files. =================================================================== RCS file: /ftp/cvs/cvsroot/src/sys/netinet/ip_input.c,v rcsdiff: /ftp/cvs/cvsroot/src/sys/netinet/ip_input.c,v: warning: Unknown phrases like `commitid ...;' are present. retrieving revision 1.330 retrieving revision 1.355 diff -u -p -r1.330 -r1.355 --- src/sys/netinet/ip_input.c 2016/06/28 02:02:56 1.330 +++ src/sys/netinet/ip_input.c 2017/06/01 02:45:14 1.355 @@ -1,4 +1,4 @@ -/* $NetBSD: ip_input.c,v 1.330 2016/06/28 02:02:56 ozaki-r Exp $ */ +/* $NetBSD: ip_input.c,v 1.355 2017/06/01 02:45:14 chs Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -91,7 +91,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: ip_input.c,v 1.330 2016/06/28 02:02:56 ozaki-r Exp $"); +__KERNEL_RCSID(0, "$NetBSD: ip_input.c,v 1.355 2017/06/01 02:45:14 chs Exp $"); #ifdef _KERNEL_OPT #include "opt_inet.h" @@ -101,6 +101,7 @@ __KERNEL_RCSID(0, "$NetBSD: ip_input.c,v #include "opt_mrouting.h" #include "opt_mbuftrace.h" #include "opt_inet_csum.h" +#include "opt_net_mpsafe.h" #endif #include "arp.h" @@ -225,7 +226,7 @@ pfil_head_t * inet_pfil_hook __read_mo ipid_state_t * ip_ids __read_mostly; percpu_t * ipstat_percpu __read_mostly; -static struct route ipforward_rt __cacheline_aligned; +static percpu_t *ipforward_rt_percpu __cacheline_aligned; uint16_t ip_id; @@ -252,30 +253,23 @@ EVCNT_ATTACH_STATIC(ip_swcsum); #endif /* INET_CSUM_COUNTERS */ /* - * We need to save the IP options in case a protocol wants to respond + * Used to save the IP options in case a protocol wants to respond * to an incoming packet over the same route if the packet got here * using IP source routing. This allows connection establishment and * maintenance when the remote end is on a network that is not known * to us. */ - -static int ip_nhops = 0; - -static struct ip_srcrt { - struct in_addr dst; /* final destination */ - char nop; /* one NOP to align */ - char srcopt[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN and OFFSET */ - struct in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)]; -} ip_srcrt; +struct ip_srcrt { + int isr_nhops; /* number of hops */ + struct in_addr isr_dst; /* final destination */ + char isr_nop; /* one NOP to align */ + char isr_hdr[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN & OFFSET */ + struct in_addr isr_routes[MAX_IPOPTLEN/sizeof(struct in_addr)]; +}; static int ip_drainwanted; -struct sockaddr_in ipaddr = { - .sin_len = sizeof(ipaddr), - .sin_family = AF_INET, -}; - -static void save_rte(u_char *, struct in_addr); +static void save_rte(struct mbuf *, u_char *, struct in_addr); #ifdef MBUFTRACE struct mowner ip_rx_mowner = MOWNER_INIT("internet", "rx"); @@ -286,12 +280,21 @@ static void ipintr(void *); static void ip_input(struct mbuf *); static void ip_forward(struct mbuf *, int, struct ifnet *); static bool ip_dooptions(struct mbuf *); -static struct in_ifaddr *ip_rtaddr(struct in_addr); +static struct in_ifaddr *ip_rtaddr(struct in_addr, struct psref *); static void sysctl_net_inet_ip_setup(struct sysctllog **); -/* XXX: Not yet enabled. */ +static struct in_ifaddr *ip_match_our_address(struct ifnet *, struct ip *, + int *); +static struct in_ifaddr *ip_match_our_address_broadcast(struct ifnet *, + struct ip *); + +#ifdef NET_MPSAFE +#define SOFTNET_LOCK() mutex_enter(softnet_lock) +#define SOFTNET_UNLOCK() mutex_exit(softnet_lock) +#else #define SOFTNET_LOCK() KASSERT(mutex_owned(softnet_lock)) #define SOFTNET_UNLOCK() KASSERT(mutex_owned(softnet_lock)) +#endif /* * IP initialization: fill in IP protocol switch table. @@ -325,7 +328,6 @@ ip_init(void) ip_ids = ip_id_init(); ip_id = time_uptime & 0xfffff; - ip_mtudisc_timeout_q = rt_timer_queue_create(ip_mtudisc_timeout); #ifdef GATEWAY ipflow_init(); #endif @@ -340,6 +342,84 @@ ip_init(void) #endif /* MBUFTRACE */ ipstat_percpu = percpu_alloc(sizeof(uint64_t) * IP_NSTATS); + ipforward_rt_percpu = percpu_alloc(sizeof(struct route)); + ip_mtudisc_timeout_q = rt_timer_queue_create(ip_mtudisc_timeout); +} + +static struct in_ifaddr * +ip_match_our_address(struct ifnet *ifp, struct ip *ip, int *downmatch) +{ + struct in_ifaddr *ia = NULL; + int checkif; + + /* + * Enable a consistency check between the destination address + * and the arrival interface for a unicast packet (the RFC 1122 + * strong ES model) if IP forwarding is disabled and the packet + * is not locally generated. + * + * XXX - Checking also should be disabled if the destination + * address is ipnat'ed to a different interface. + * + * XXX - Checking is incompatible with IP aliases added + * to the loopback interface instead of the interface where + * the packets are received. + * + * XXX - We need to add a per ifaddr flag for this so that + * we get finer grain control. + */ + checkif = ip_checkinterface && (ipforwarding == 0) && + (ifp->if_flags & IFF_LOOPBACK) == 0; + + IN_ADDRHASH_READER_FOREACH(ia, ip->ip_dst.s_addr) { + if (in_hosteq(ia->ia_addr.sin_addr, ip->ip_dst)) { + if (ia->ia4_flags & IN_IFF_NOTREADY) + continue; + if (checkif && ia->ia_ifp != ifp) + continue; + if ((ia->ia_ifp->if_flags & IFF_UP) != 0 && + (ia->ia4_flags & IN_IFF_DETACHED) == 0) + break; + else + (*downmatch)++; + } + } + + return ia; +} + +static struct in_ifaddr * +ip_match_our_address_broadcast(struct ifnet *ifp, struct ip *ip) +{ + struct in_ifaddr *ia = NULL; + struct ifaddr *ifa; + + IFADDR_READER_FOREACH(ifa, ifp) { + if (ifa->ifa_addr->sa_family != AF_INET) + continue; + ia = ifatoia(ifa); + if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED)) + continue; + if (in_hosteq(ip->ip_dst, ia->ia_broadaddr.sin_addr) || + in_hosteq(ip->ip_dst, ia->ia_netbroadcast) || + /* + * Look for all-0's host part (old broadcast addr), + * either for subnet or net. + */ + ip->ip_dst.s_addr == ia->ia_subnet || + ip->ip_dst.s_addr == ia->ia_net) + goto matched; + /* + * An interface with IP address zero accepts + * all packets that arrive on that interface. + */ + if (in_nullhost(ia->ia_addr.sin_addr)) + goto matched; + } + ia = NULL; + +matched: + return ia; } /* @@ -352,11 +432,15 @@ ipintr(void *arg __unused) KASSERT(cpu_softintr_p()); +#ifndef NET_MPSAFE mutex_enter(softnet_lock); +#endif while ((m = pktq_dequeue(ip_pktq)) != NULL) { ip_input(m); } +#ifndef NET_MPSAFE mutex_exit(softnet_lock); +#endif } /* @@ -367,14 +451,13 @@ static void ip_input(struct mbuf *m) { struct ip *ip = NULL; - struct in_ifaddr *ia; - struct ifaddr *ifa; + struct in_ifaddr *ia = NULL; int hlen = 0, len; int downmatch; - int checkif; int srcrt = 0; ifnet_t *ifp; struct psref psref; + int s; KASSERTMSG(cpu_softintr_p(), "ip_input: not in the software " "interrupt handler; synchronization assumptions violated"); @@ -384,16 +467,15 @@ ip_input(struct mbuf *m) ifp = m_get_rcvif_psref(m, &psref); if (__predict_false(ifp == NULL)) - goto bad; + goto out; /* * If no IP addresses have been set yet but the interfaces * are receiving, can't do anything with incoming packets yet. * Note: we pre-check without locks held. */ - if (!TAILQ_FIRST(&in_ifaddrhead)) { - goto bad; - } + if (IN_ADDRLIST_READER_EMPTY()) + goto out; IP_STATINC(IP_STAT_TOTAL); /* @@ -418,12 +500,12 @@ ip_input(struct mbuf *m) ip = mtod(m, struct ip *); if (ip->ip_v != IPVERSION) { IP_STATINC(IP_STAT_BADVERS); - goto bad; + goto out; } hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) { /* minimum header length */ IP_STATINC(IP_STAT_BADHLEN); - goto bad; + goto out; } if (hlen > m->m_len) { if ((m = m_pullup(m, hlen)) == NULL) { @@ -439,7 +521,7 @@ ip_input(struct mbuf *m) */ if (IN_MULTICAST(ip->ip_src.s_addr)) { IP_STATINC(IP_STAT_BADADDR); - goto bad; + goto out; } /* 127/8 must not appear on wire - RFC1122 */ @@ -447,7 +529,7 @@ ip_input(struct mbuf *m) (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { if ((ifp->if_flags & IFF_LOOPBACK) == 0) { IP_STATINC(IP_STAT_BADADDR); - goto bad; + goto out; } } @@ -456,7 +538,8 @@ ip_input(struct mbuf *m) M_CSUM_IPv4_BAD)) { case M_CSUM_IPv4|M_CSUM_IPv4_BAD: INET_CSUM_COUNTER_INCR(&ip_hwcsum_bad); - goto badcsum; + IP_STATINC(IP_STAT_BADSUM); + goto out; case M_CSUM_IPv4: /* Checksum was okay. */ @@ -471,8 +554,10 @@ ip_input(struct mbuf *m) if (__predict_true(!(ifp->if_flags & IFF_LOOPBACK) || ip_do_loopback_cksum)) { INET_CSUM_COUNTER_INCR(&ip_swcsum); - if (in_cksum(m, hlen) != 0) - goto badcsum; + if (in_cksum(m, hlen) != 0) { + IP_STATINC(IP_STAT_BADSUM); + goto out; + } } break; } @@ -485,7 +570,7 @@ ip_input(struct mbuf *m) */ if (len < hlen) { IP_STATINC(IP_STAT_BADLEN); - goto bad; + goto out; } /* @@ -496,7 +581,7 @@ ip_input(struct mbuf *m) */ if (m->m_pkthdr.len < len) { IP_STATINC(IP_STAT_TOOSHORT); - goto bad; + goto out; } if (m->m_pkthdr.len > len) { if (m->m_len == m->m_pkthdr.len) { @@ -528,10 +613,9 @@ ip_input(struct mbuf *m) struct in_addr odst = ip->ip_dst; bool freed; - SOFTNET_LOCK(); freed = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_IN) != 0; - SOFTNET_UNLOCK(); if (freed || m == NULL) { + m = NULL; goto out; } ip = mtod(m, struct ip *); @@ -561,6 +645,7 @@ ip_input(struct mbuf *m) if ((*altq_input)(m, AF_INET) == 0) { /* Packet dropped by traffic conditioner. */ SOFTNET_UNLOCK(); + m = NULL; goto out; } SOFTNET_UNLOCK(); @@ -573,28 +658,10 @@ ip_input(struct mbuf *m) * error was detected (causing an icmp message * to be sent and the original packet to be freed). */ - ip_nhops = 0; /* for source routed packets */ - if (hlen > sizeof (struct ip) && ip_dooptions(m)) + if (hlen > sizeof (struct ip) && ip_dooptions(m)) { + m = NULL; goto out; - - /* - * Enable a consistency check between the destination address - * and the arrival interface for a unicast packet (the RFC 1122 - * strong ES model) if IP forwarding is disabled and the packet - * is not locally generated. - * - * XXX - Checking also should be disabled if the destination - * address is ipnat'ed to a different interface. - * - * XXX - Checking is incompatible with IP aliases added - * to the loopback interface instead of the interface where - * the packets are received. - * - * XXX - We need to add a per ifaddr flag for this so that - * we get finer grain control. - */ - checkif = ip_checkinterface && (ipforwarding == 0) && - (ifp->if_flags & IFF_LOOPBACK) == 0; + } /* * Check our list of addresses, to see if the packet is for us. @@ -604,44 +671,22 @@ ip_input(struct mbuf *m) * or IN_IFF_NOTREADY addresses as not mine. */ downmatch = 0; - LIST_FOREACH(ia, &IN_IFADDR_HASH(ip->ip_dst.s_addr), ia_hash) { - if (in_hosteq(ia->ia_addr.sin_addr, ip->ip_dst)) { - if (ia->ia4_flags & IN_IFF_NOTREADY) - continue; - if (checkif && ia->ia_ifp != ifp) - continue; - if ((ia->ia_ifp->if_flags & IFF_UP) != 0) - break; - else - downmatch++; - } - } - if (ia != NULL) + s = pserialize_read_enter(); + ia = ip_match_our_address(ifp, ip, &downmatch); + if (ia != NULL) { + pserialize_read_exit(s); goto ours; + } + if (ifp->if_flags & IFF_BROADCAST) { - IFADDR_FOREACH(ifa, ifp) { - if (ifa->ifa_addr->sa_family != AF_INET) - continue; - ia = ifatoia(ifa); - if (ia->ia4_flags & IN_IFF_NOTREADY) - continue; - if (in_hosteq(ip->ip_dst, ia->ia_broadaddr.sin_addr) || - in_hosteq(ip->ip_dst, ia->ia_netbroadcast) || - /* - * Look for all-0's host part (old broadcast addr), - * either for subnet or net. - */ - ip->ip_dst.s_addr == ia->ia_subnet || - ip->ip_dst.s_addr == ia->ia_net) - goto ours; - /* - * An interface with IP address zero accepts - * all packets that arrive on that interface. - */ - if (in_nullhost(ia->ia_addr.sin_addr)) - goto ours; + ia = ip_match_our_address_broadcast(ifp, ip); + if (ia != NULL) { + pserialize_read_exit(s); + goto ours; } } + pserialize_read_exit(s); + if (IN_MULTICAST(ip->ip_dst.s_addr)) { #ifdef MROUTING extern struct socket *ip_mrouter; @@ -663,7 +708,7 @@ ip_input(struct mbuf *m) if (ip_mforward(m, ifp) != 0) { SOFTNET_UNLOCK(); IP_STATINC(IP_STAT_CANTFORWARD); - goto bad; + goto out; } SOFTNET_UNLOCK(); @@ -684,7 +729,7 @@ ip_input(struct mbuf *m) */ if (!in_multi_group(ip->ip_dst, ifp, 0)) { IP_STATINC(IP_STAT_CANTFORWARD); - goto bad; + goto out; } goto ours; } @@ -719,7 +764,7 @@ ip_input(struct mbuf *m) if (ipsec4_input(m, IP_FORWARDING | (ip_directedbcast ? IP_ALLOWBROADCAST : 0)) != 0) { SOFTNET_UNLOCK(); - goto bad; + goto out; } SOFTNET_UNLOCK(); } @@ -731,6 +776,7 @@ ip_input(struct mbuf *m) ours: m_put_rcvif_psref(ifp, &psref); + ifp = NULL; /* * If offset or IP_MF are set, must reassemble. @@ -741,11 +787,11 @@ ours: */ if (ip_reass_packet(&m, ip) != 0) { /* Failed; invalid fragment(s) or packet. */ - goto bad; + goto out; } if (m == NULL) { /* More fragments should come; silently return. */ - return; + goto out; } /* * Reassembly is done, we have the final packet. @@ -766,7 +812,7 @@ ours: SOFTNET_LOCK(); if (ipsec4_input(m, 0) != 0) { SOFTNET_UNLOCK(); - goto bad; + goto out; } SOFTNET_UNLOCK(); } @@ -776,8 +822,17 @@ ours: * Switch out to protocol's input routine. */ #if IFA_STATS - if (ia && ip) - ia->ia_ifa.ifa_data.ifad_inbytes += ntohs(ip->ip_len); + if (ia && ip) { + struct in_ifaddr *_ia; + /* + * Keep a reference from ip_match_our_address with psref + * is expensive, so explore ia here again. + */ + s = pserialize_read_enter(); + _ia = in_get_ia(ip->ip_dst); + _ia->ia_ifa.ifa_data.ifad_inbytes += ntohs(ip->ip_len); + pserialize_read_exit(s); + } #endif IP_STATINC(IP_STAT_DELIVERED); @@ -787,18 +842,11 @@ ours: (*inetsw[ip_protox[nh]].pr_input)(m, off, nh); SOFTNET_UNLOCK(); return; -bad: - m_put_rcvif_psref(ifp, &psref); - m_freem(m); - return; -badcsum: - m_put_rcvif_psref(ifp, &psref); - IP_STATINC(IP_STAT_BADSUM); - m_freem(m); - return; out: m_put_rcvif_psref(ifp, &psref); + if (m != NULL) + m_freem(m); } /* @@ -808,13 +856,17 @@ void ip_slowtimo(void) { +#ifndef NET_MPSAFE mutex_enter(softnet_lock); KERNEL_LOCK(1, NULL); +#endif ip_reass_slowtimo(); +#ifndef NET_MPSAFE KERNEL_UNLOCK_ONE(NULL); mutex_exit(softnet_lock); +#endif } /* @@ -846,6 +898,8 @@ ip_dooptions(struct mbuf *m) int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB, forward = 0; struct in_addr dst; n_time ntime; + struct ifaddr *ifa = NULL; + int s; dst = ip->ip_dst; cp = (u_char *)(ip + 1); @@ -882,7 +936,13 @@ ip_dooptions(struct mbuf *m) * address is on directly accessible net. */ case IPOPT_LSRR: - case IPOPT_SSRR: + case IPOPT_SSRR: { + struct psref psref; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; + if (ip_allowsrcrt == 0) { type = ICMP_UNREACH; code = ICMP_UNREACH_NET_PROHIB; @@ -897,8 +957,11 @@ ip_dooptions(struct mbuf *m) goto bad; } ipaddr.sin_addr = ip->ip_dst; - ia = ifatoia(ifa_ifwithaddr(sintosa(&ipaddr))); - if (ia == 0) { + + s = pserialize_read_enter(); + ifa = ifa_ifwithaddr(sintosa(&ipaddr)); + if (ifa == NULL) { + pserialize_read_exit(s); if (opt == IPOPT_SSRR) { type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; @@ -910,12 +973,14 @@ ip_dooptions(struct mbuf *m) */ break; } + pserialize_read_exit(s); + off--; /* 0 origin */ if ((off + sizeof(struct in_addr)) > optlen) { /* * End of source route. Should be for us. */ - save_rte(cp, ip->ip_src); + save_rte(m, cp, ip->ip_src); break; } /* @@ -923,11 +988,17 @@ ip_dooptions(struct mbuf *m) */ memcpy((void *)&ipaddr.sin_addr, (void *)(cp + off), sizeof(ipaddr.sin_addr)); - if (opt == IPOPT_SSRR) - ia = ifatoia(ifa_ifwithladdr(sintosa(&ipaddr))); - else - ia = ip_rtaddr(ipaddr.sin_addr); - if (ia == 0) { + if (opt == IPOPT_SSRR) { + ifa = ifa_ifwithladdr_psref(sintosa(&ipaddr), + &psref); + if (ifa != NULL) + ia = ifatoia(ifa); + else + ia = NULL; + } else { + ia = ip_rtaddr(ipaddr.sin_addr, &psref); + } + if (ia == NULL) { type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; goto bad; @@ -935,14 +1006,22 @@ ip_dooptions(struct mbuf *m) ip->ip_dst = ipaddr.sin_addr; bcopy((void *)&ia->ia_addr.sin_addr, (void *)(cp + off), sizeof(struct in_addr)); + ia4_release(ia, &psref); cp[IPOPT_OFFSET] += sizeof(struct in_addr); /* * Let ip_intr's mcast routing check handle mcast pkts */ forward = !IN_MULTICAST(ip->ip_dst.s_addr); break; + } + + case IPOPT_RR: { + struct psref psref; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; - case IPOPT_RR: if (optlen < IPOPT_OFFSET + sizeof(*cp)) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; @@ -963,17 +1042,23 @@ ip_dooptions(struct mbuf *m) * locate outgoing interface; if we're the destination, * use the incoming interface (should be same). */ - if ((ia = ifatoia(ifa_ifwithaddr(sintosa(&ipaddr)))) - == NULL && - (ia = ip_rtaddr(ipaddr.sin_addr)) == NULL) { - type = ICMP_UNREACH; - code = ICMP_UNREACH_HOST; - goto bad; + ifa = ifa_ifwithaddr_psref(sintosa(&ipaddr), &psref); + if (ifa == NULL) { + ia = ip_rtaddr(ipaddr.sin_addr, &psref); + if (ia == NULL) { + type = ICMP_UNREACH; + code = ICMP_UNREACH_HOST; + goto bad; + } + } else { + ia = ifatoia(ifa); } bcopy((void *)&ia->ia_addr.sin_addr, (void *)(cp + off), sizeof(struct in_addr)); + ia4_release(ia, &psref); cp[IPOPT_OFFSET] += sizeof(struct in_addr); break; + } case IPOPT_TS: code = cp - (u_char *)ip; @@ -1002,7 +1087,11 @@ ip_dooptions(struct mbuf *m) case IPOPT_TS_TSANDADDR: { struct ifnet *rcvif; - int s; + int _s, _ss; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; if (ipt->ipt_ptr - 1 + sizeof(n_time) + sizeof(struct in_addr) > ipt->ipt_len) { @@ -1011,19 +1100,31 @@ ip_dooptions(struct mbuf *m) goto bad; } ipaddr.sin_addr = dst; - rcvif = m_get_rcvif(m, &s); - ia = ifatoia(ifaof_ifpforaddr(sintosa(&ipaddr), - rcvif)); - m_put_rcvif(rcvif, &s); - if (ia == 0) - continue; + _ss = pserialize_read_enter(); + rcvif = m_get_rcvif(m, &_s); + if (__predict_true(rcvif != NULL)) { + ifa = ifaof_ifpforaddr(sintosa(&ipaddr), + rcvif); + } + m_put_rcvif(rcvif, &_s); + if (ifa == NULL) { + pserialize_read_exit(_ss); + break; + } + ia = ifatoia(ifa); bcopy(&ia->ia_addr.sin_addr, cp0, sizeof(struct in_addr)); + pserialize_read_exit(_ss); ipt->ipt_ptr += sizeof(struct in_addr); break; } - case IPOPT_TS_PRESPEC: + case IPOPT_TS_PRESPEC: { + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; + if (ipt->ipt_ptr - 1 + sizeof(n_time) + sizeof(struct in_addr) > ipt->ipt_len) { code = (u_char *)&ipt->ipt_ptr - @@ -1032,11 +1133,16 @@ ip_dooptions(struct mbuf *m) } memcpy(&ipaddr.sin_addr, cp0, sizeof(struct in_addr)); - if (ifatoia(ifa_ifwithaddr(sintosa(&ipaddr))) - == NULL) + s = pserialize_read_enter(); + ifa = ifa_ifwithaddr(sintosa(&ipaddr)); + if (ifa == NULL) { + pserialize_read_exit(s); continue; + } + pserialize_read_exit(s); ipt->ipt_ptr += sizeof(struct in_addr); break; + } default: /* XXX can't take &ipt->ipt_flg */ @@ -1053,7 +1159,7 @@ ip_dooptions(struct mbuf *m) } if (forward) { struct ifnet *rcvif; - struct psref psref; + struct psref _psref; if (ip_forwsrcrt == 0) { type = ICMP_UNREACH; @@ -1061,14 +1167,14 @@ ip_dooptions(struct mbuf *m) goto bad; } - rcvif = m_get_rcvif_psref(m, &psref); + rcvif = m_get_rcvif_psref(m, &_psref); if (__predict_false(rcvif == NULL)) { type = ICMP_UNREACH; code = ICMP_UNREACH_HOST; goto bad; } ip_forward(m, 1, rcvif); - m_put_rcvif_psref(rcvif, &psref); + m_put_rcvif_psref(rcvif, &_psref); return true; } return false; @@ -1083,21 +1189,27 @@ bad: * return internet address info of interface to be used to get there. */ static struct in_ifaddr * -ip_rtaddr(struct in_addr dst) +ip_rtaddr(struct in_addr dst, struct psref *psref) { struct rtentry *rt; union { struct sockaddr dst; struct sockaddr_in dst4; } u; + struct route *ro; sockaddr_in_init(&u.dst4, &dst, 0); - SOFTNET_LOCK(); - rt = rtcache_lookup(&ipforward_rt, &u.dst); - SOFTNET_UNLOCK(); - if (rt == NULL) + ro = percpu_getref(ipforward_rt_percpu); + rt = rtcache_lookup(ro, &u.dst); + if (rt == NULL) { + percpu_putref(ipforward_rt_percpu); return NULL; + } + + ia4_acquire(ifatoia(rt->rt_ifa), psref); + rtcache_unref(rt, ro); + percpu_putref(ipforward_rt_percpu); return ifatoia(rt->rt_ifa); } @@ -1107,16 +1219,25 @@ ip_rtaddr(struct in_addr dst) * up later by ip_srcroute if the receiver is interested. */ static void -save_rte(u_char *option, struct in_addr dst) +save_rte(struct mbuf *m, u_char *option, struct in_addr dst) { + struct ip_srcrt *isr; + struct m_tag *mtag; unsigned olen; olen = option[IPOPT_OLEN]; - if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst))) + if (olen > sizeof(isr->isr_hdr) + sizeof(isr->isr_routes)) return; - memcpy((void *)ip_srcrt.srcopt, (void *)option, olen); - ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr); - ip_srcrt.dst = dst; + + mtag = m_tag_get(PACKET_TAG_SRCROUTE, sizeof(*isr), M_NOWAIT); + if (mtag == NULL) + return; + isr = (struct ip_srcrt *)(mtag + 1); + + memcpy(isr->isr_hdr, option, olen); + isr->isr_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr); + isr->isr_dst = dst; + m_tag_prepend(m, mtag); } /* @@ -1125,36 +1246,43 @@ save_rte(u_char *option, struct in_addr * The first hop is placed before the options, will be removed later. */ struct mbuf * -ip_srcroute(void) +ip_srcroute(struct mbuf *m0) { struct in_addr *p, *q; struct mbuf *m; + struct ip_srcrt *isr; + struct m_tag *mtag; + + mtag = m_tag_find(m0, PACKET_TAG_SRCROUTE, NULL); + if (mtag == NULL) + return NULL; + isr = (struct ip_srcrt *)(mtag + 1); - if (ip_nhops == 0) + if (isr->isr_nhops == 0) return NULL; + m = m_get(M_DONTWAIT, MT_SOOPTS); - if (m == 0) + if (m == NULL) return NULL; MCLAIM(m, &inetdomain.dom_mowner); -#define OPTSIZ (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt)) +#define OPTSIZ (sizeof(isr->isr_nop) + sizeof(isr->isr_hdr)) - /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */ - m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) + - OPTSIZ; + /* length is (nhops+1)*sizeof(addr) + sizeof(nop + header) */ + m->m_len = (isr->isr_nhops + 1) * sizeof(struct in_addr) + OPTSIZ; /* * First save first hop for return route */ - p = &ip_srcrt.route[ip_nhops - 1]; + p = &(isr->isr_routes[isr->isr_nhops - 1]); *(mtod(m, struct in_addr *)) = *p--; /* * Copy option fields and padding (nop) to mbuf. */ - ip_srcrt.nop = IPOPT_NOP; - ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF; - memmove(mtod(m, char *) + sizeof(struct in_addr), &ip_srcrt.nop, + isr->isr_nop = IPOPT_NOP; + isr->isr_hdr[IPOPT_OFFSET] = IPOPT_MINOFF; + memmove(mtod(m, char *) + sizeof(struct in_addr), &isr->isr_nop, OPTSIZ); q = (struct in_addr *)(mtod(m, char *) + sizeof(struct in_addr) + OPTSIZ); @@ -1163,14 +1291,15 @@ ip_srcroute(void) * Record return path as an IP source route, * reversing the path (pointers are now aligned). */ - while (p >= ip_srcrt.route) { + while (p >= isr->isr_routes) { *q++ = *p--; } /* * Last hop goes to final destination. */ - *q = ip_srcrt.dst; - return (m); + *q = isr->isr_dst; + m_tag_delete(m0, mtag); + return m; } const int inetctlerrmap[PRC_NCMDS] = { @@ -1227,6 +1356,7 @@ ip_forward(struct mbuf *m, int srcrt, st struct sockaddr_in dst4; } u; uint64_t *ips; + struct route *ro; KASSERTMSG(cpu_softintr_p(), "ip_forward: not in the software " "interrupt handler; synchronization assumptions violated"); @@ -1248,19 +1378,18 @@ ip_forward(struct mbuf *m, int srcrt, st return; } - SOFTNET_LOCK(); - if (ip->ip_ttl <= IPTTLDEC) { icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0); - SOFTNET_UNLOCK(); return; } sockaddr_in_init(&u.dst4, &ip->ip_dst, 0); - if ((rt = rtcache_lookup(&ipforward_rt, &u.dst)) == NULL) { + ro = percpu_getref(ipforward_rt_percpu); + rt = rtcache_lookup(ro, &u.dst); + if (rt == NULL) { + percpu_putref(ipforward_rt_percpu); icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NET, dest, 0); - SOFTNET_UNLOCK(); return; } @@ -1302,8 +1431,9 @@ ip_forward(struct mbuf *m, int srcrt, st code = ICMP_REDIRECT_HOST; } } + rtcache_unref(rt, ro); - error = ip_output(m, NULL, &ipforward_rt, + error = ip_output(m, NULL, ro, (IP_FORWARDING | (ip_directedbcast ? IP_ALLOWBROADCAST : 0)), NULL, NULL); @@ -1325,18 +1455,18 @@ ip_forward(struct mbuf *m, int srcrt, st if (mcopy) { #ifdef GATEWAY if (mcopy->m_flags & M_CANFASTFWD) - ipflow_create(&ipforward_rt, mcopy); + ipflow_create(ro, mcopy); #endif m_freem(mcopy); } - SOFTNET_UNLOCK(); + percpu_putref(ipforward_rt_percpu); return; redirect: error: if (mcopy == NULL) { - SOFTNET_UNLOCK(); + percpu_putref(ipforward_rt_percpu); return; } @@ -1359,8 +1489,10 @@ error: type = ICMP_UNREACH; code = ICMP_UNREACH_NEEDFRAG; - if ((rt = rtcache_validate(&ipforward_rt)) != NULL) + if ((rt = rtcache_validate(ro)) != NULL) { destmtu = rt->rt_ifp->if_mtu; + rtcache_unref(rt, ro); + } #ifdef IPSEC if (ipsec_used) (void)ipsec4_forward(mcopy, &destmtu); @@ -1377,11 +1509,11 @@ error: */ if (mcopy) m_freem(mcopy); - SOFTNET_UNLOCK(); + percpu_putref(ipforward_rt_percpu); return; } icmp_error(mcopy, type, code, dest, destmtu); - SOFTNET_UNLOCK(); + percpu_putref(ipforward_rt_percpu); } void @@ -1445,8 +1577,8 @@ ip_savecontrol(struct inpcb *inp, struct if (inpflags & INP_RECVIF) { struct sockaddr_dl sdl; - sockaddr_dl_init(&sdl, sizeof(sdl), ifp ? - ifp->if_index : 0, 0, NULL, 0, NULL, 0); + sockaddr_dl_init(&sdl, sizeof(sdl), ifp->if_index, 0, NULL, 0, + NULL, 0); *mp = sbcreatecontrol(&sdl, sdl.sdl_len, IP_RECVIF, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; @@ -1496,23 +1628,25 @@ sysctl_net_inet_ip_pmtudto(SYSCTLFN_ARGS int error, tmp; struct sysctlnode node; + icmp_mtudisc_lock(); + node = *rnode; tmp = ip_mtudisc_timeout; node.sysctl_data = &tmp; error = sysctl_lookup(SYSCTLFN_CALL(&node)); if (error || newp == NULL) - return (error); - if (tmp < 0) - return (EINVAL); - - mutex_enter(softnet_lock); + goto out; + if (tmp < 0) { + error = EINVAL; + goto out; + } ip_mtudisc_timeout = tmp; rt_timer_queue_change(ip_mtudisc_timeout_q, ip_mtudisc_timeout); - - mutex_exit(softnet_lock); - - return (0); + error = 0; +out: + icmp_mtudisc_unlock(); + return error; } static int