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/netinet6/ip6_input.c,v rcsdiff: /ftp/cvs/cvsroot/src/sys/netinet6/ip6_input.c,v: warning: Unknown phrases like `commitid ...;' are present. retrieving revision 1.164.2.1 retrieving revision 1.178.2.9 diff -u -p -r1.164.2.1 -r1.178.2.9 --- src/sys/netinet6/ip6_input.c 2016/08/06 00:19:10 1.164.2.1 +++ src/sys/netinet6/ip6_input.c 2019/09/24 18:27:09 1.178.2.9 @@ -1,4 +1,4 @@ -/* $NetBSD: ip6_input.c,v 1.164.2.1 2016/08/06 00:19:10 pgoyette Exp $ */ +/* $NetBSD: ip6_input.c,v 1.178.2.9 2019/09/24 18:27:09 martin Exp $ */ /* $KAME: ip6_input.c,v 1.188 2001/03/29 05:34:31 itojun Exp $ */ /* @@ -62,7 +62,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: ip6_input.c,v 1.164.2.1 2016/08/06 00:19:10 pgoyette Exp $"); +__KERNEL_RCSID(0, "$NetBSD: ip6_input.c,v 1.178.2.9 2019/09/24 18:27:09 martin Exp $"); #ifdef _KERNEL_OPT #include "opt_gateway.h" @@ -70,11 +70,11 @@ __KERNEL_RCSID(0, "$NetBSD: ip6_input.c, #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_compat_netbsd.h" +#include "opt_net_mpsafe.h" #endif #include #include -#include #include #include #include @@ -87,6 +87,7 @@ __KERNEL_RCSID(0, "$NetBSD: ip6_input.c, #include #include #include +#include #include #include @@ -143,6 +144,8 @@ pfil_head_t *inet6_pfil_hook; percpu_t *ip6stat_percpu; +percpu_t *ip6_forward_rt_percpu __cacheline_aligned; + static void ip6_init2(void); static void ip6intr(void *); static struct m_tag *ip6_setdstifaddr(struct mbuf *, const struct in6_ifaddr *); @@ -152,6 +155,14 @@ static int ip6_process_hopopts(struct mb static struct mbuf *ip6_pullexthdr(struct mbuf *, size_t, int); static void sysctl_net_inet6_ip6_setup(struct sysctllog **); +#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 + /* * IP6 initialization: fill in IP6 protocol switch table. * All protocols not implemented in kernel go to raw IP6 protocol handler. @@ -194,6 +205,7 @@ ip6_init(void) KASSERT(inet6_pfil_hook != NULL); ip6stat_percpu = percpu_alloc(sizeof(uint64_t) * IP6_NSTATS); + ip6_forward_rt_percpu = rtcache_percpu_alloc(); } static void @@ -216,7 +228,7 @@ ip6intr(void *arg __unused) { struct mbuf *m; - mutex_enter(softnet_lock); + SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE(); while ((m = pktq_dequeue(ip6_pktq)) != NULL) { struct psref psref; struct ifnet *rcvif = m_get_rcvif_psref(m, &psref); @@ -236,11 +248,9 @@ ip6intr(void *arg __unused) ip6_input(m, rcvif); m_put_rcvif_psref(rcvif, &psref); } - mutex_exit(softnet_lock); + SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); } -extern struct route ip6_forward_rt; - void ip6_input(struct mbuf *m, struct ifnet *rcvif) { @@ -248,14 +258,15 @@ ip6_input(struct mbuf *m, struct ifnet * int hit, off = sizeof(struct ip6_hdr), nest; u_int32_t plen; u_int32_t rtalert = ~0; - int nxt, ours = 0, rh_present = 0; + int nxt, ours = 0, rh_present = 0, frg_present; struct ifnet *deliverifp = NULL; int srcrt = 0; - const struct rtentry *rt; + struct rtentry *rt = NULL; union { struct sockaddr dst; struct sockaddr_in6 dst6; } u; + struct route *ro; /* * make sure we don't have onion peering information into m_tag. @@ -347,6 +358,13 @@ ip6_input(struct mbuf *m, struct ifnet * return; if (m == NULL) return; + if (__predict_false(m->m_len < sizeof(struct ip6_hdr))) { + if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { + IP6_STATINC(IP6_STAT_TOOSMALL); + in6_ifstat_inc(rcvif, ifs6_in_hdrerr); + return; + } + } ip6 = mtod(m, struct ip6_hdr *); srcrt = !IN6_ARE_ADDR_EQUAL(&odst, &ip6->ip6_dst); } @@ -354,9 +372,14 @@ ip6_input(struct mbuf *m, struct ifnet * IP6_STATINC(IP6_STAT_NXTHIST + ip6->ip6_nxt); #ifdef ALTQ - if (altq_input != NULL && (*altq_input)(m, AF_INET6) == 0) { - /* packet is dropped by traffic conditioner */ - return; + if (altq_input != NULL) { + SOFTNET_LOCK(); + if ((*altq_input)(m, AF_INET6) == 0) { + SOFTNET_UNLOCK(); + /* packet is dropped by traffic conditioner */ + return; + } + SOFTNET_UNLOCK(); } #endif @@ -430,19 +453,20 @@ ip6_input(struct mbuf *m, struct ifnet * goto bad; } + ro = rtcache_percpu_getref(ip6_forward_rt_percpu); /* * Multicast check */ if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { - struct in6_multi *in6m = 0; + bool ingroup; in6_ifstat_inc(rcvif, ifs6_in_mcast); /* * See if we belong to the destination multicast group on the * arrival interface. */ - IN6_LOOKUP_MULTI(ip6->ip6_dst, rcvif, in6m); - if (in6m) + ingroup = in6_multi_group(&ip6->ip6_dst, rcvif); + if (ingroup) ours = 1; else if (!ip6_mrouter) { uint64_t *ip6s = IP6_STAT_GETREF(); @@ -450,7 +474,7 @@ ip6_input(struct mbuf *m, struct ifnet * ip6s[IP6_STAT_CANTFORWARD]++; IP6_STAT_PUTREF(); in6_ifstat_inc(rcvif, ifs6_in_discard); - goto bad; + goto bad_unref; } deliverifp = rcvif; goto hbhcheck; @@ -461,7 +485,7 @@ ip6_input(struct mbuf *m, struct ifnet * /* * Unicast check */ - rt = rtcache_lookup2(&ip6_forward_rt, &u.dst, 1, &hit); + rt = rtcache_lookup2(ro, &u.dst, 1, &hit); if (hit) IP6_STATINC(IP6_STAT_FORWARD_CACHEHIT); else @@ -490,24 +514,46 @@ ip6_input(struct mbuf *m, struct ifnet * #endif rt->rt_ifp->if_type == IFT_LOOP) { struct in6_ifaddr *ia6 = (struct in6_ifaddr *)rt->rt_ifa; + int addrok; + if (ia6->ia6_flags & IN6_IFF_ANYCAST) m->m_flags |= M_ANYCAST6; /* * packets to a tentative, duplicated, or somehow invalid * address must not be accepted. */ - if (!(ia6->ia6_flags & IN6_IFF_NOTREADY)) { + if (ia6->ia6_flags & IN6_IFF_NOTREADY) + addrok = 0; + else if (ia6->ia6_flags & IN6_IFF_DETACHED && + !IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) + { + /* Allow internal traffic to DETACHED addresses */ + struct sockaddr_in6 sin6; + int s; + + memset(&sin6, 0, sizeof(sin6)); + sin6.sin6_family = AF_INET6; + sin6.sin6_len = sizeof(sin6); + sin6.sin6_addr = ip6->ip6_src; + s = pserialize_read_enter(); + addrok = (ifa_ifwithaddr(sin6tosa(&sin6)) != NULL); + pserialize_read_exit(s); + } else + addrok = 1; + if (addrok) { /* this address is ready */ ours = 1; deliverifp = ia6->ia_ifp; /* correct? */ goto hbhcheck; } else { /* address is not ready, so discard the packet. */ + char ip6bufs[INET6_ADDRSTRLEN]; + char ip6bufd[INET6_ADDRSTRLEN]; nd6log(LOG_INFO, "packet to an unready address %s->%s\n", - ip6_sprintf(&ip6->ip6_src), - ip6_sprintf(&ip6->ip6_dst)); + IN6_PRINT(ip6bufs, &ip6->ip6_src), + IN6_PRINT(ip6bufd, &ip6->ip6_dst)); - goto bad; + goto bad_unref; } } @@ -553,7 +599,7 @@ ip6_input(struct mbuf *m, struct ifnet * if (!ip6_forwarding) { IP6_STATINC(IP6_STAT_CANTFORWARD); in6_ifstat_inc(rcvif, ifs6_in_discard); - goto bad; + goto bad_unref; } hbhcheck: @@ -592,6 +638,8 @@ ip6_input(struct mbuf *m, struct ifnet * #if 0 /*touches NULL pointer*/ in6_ifstat_inc(rcvif, ifs6_in_discard); #endif + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); return; /* m have already been freed */ } @@ -615,12 +663,16 @@ ip6_input(struct mbuf *m, struct ifnet * icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, (char *)&ip6->ip6_plen - (char *)ip6); + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); return; } IP6_EXTHDR_GET(hbh, struct ip6_hbh *, m, sizeof(struct ip6_hdr), sizeof(struct ip6_hbh)); if (hbh == NULL) { IP6_STATINC(IP6_STAT_TOOSHORT); + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); return; } KASSERT(IP6_HDR_ALIGNED_P(hbh)); @@ -644,7 +696,7 @@ ip6_input(struct mbuf *m, struct ifnet * if (m->m_pkthdr.len - sizeof(struct ip6_hdr) < plen) { IP6_STATINC(IP6_STAT_TOOSHORT); in6_ifstat_inc(rcvif, ifs6_in_truncated); - goto bad; + goto bad_unref; } if (m->m_pkthdr.len > sizeof(struct ip6_hdr) + plen) { if (m->m_len == m->m_pkthdr.len) { @@ -666,16 +718,25 @@ ip6_input(struct mbuf *m, struct ifnet * * ip6_mforward() returns a non-zero value, the packet * must be discarded, else it may be accepted below. */ - if (ip6_mrouter && ip6_mforward(ip6, rcvif, m)) { - IP6_STATINC(IP6_STAT_CANTFORWARD); - m_freem(m); - return; - } - if (!ours) { - m_freem(m); - return; + if (ip6_mrouter != NULL) { + int error; + + SOFTNET_LOCK(); + error = ip6_mforward(ip6, rcvif, m); + SOFTNET_UNLOCK(); + + if (error != 0) { + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); + IP6_STATINC(IP6_STAT_CANTFORWARD); + goto bad; + } } + if (!ours) + goto bad_unref; } else if (!ours) { + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); ip6_forward(m, srcrt); return; } @@ -695,7 +756,7 @@ ip6_input(struct mbuf *m, struct ifnet * IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { IP6_STATINC(IP6_STAT_BADSCOPE); in6_ifstat_inc(rcvif, ifs6_in_addrerr); - goto bad; + goto bad_unref; } /* @@ -715,7 +776,14 @@ ip6_input(struct mbuf *m, struct ifnet * in6_ifstat_inc(deliverifp, ifs6_in_deliver); nest = 0; + if (rt != NULL) { + rtcache_unref(rt, ro); + rt = NULL; + } + rtcache_percpu_putref(ip6_forward_rt_percpu); + rh_present = 0; + frg_present = 0; while (nxt != IPPROTO_DONE) { if (ip6_hdrnestlimit && (++nest > ip6_hdrnestlimit)) { IP6_STATINC(IP6_STAT_TOOMANYHDR); @@ -739,6 +807,12 @@ ip6_input(struct mbuf *m, struct ifnet * IP6_STATINC(IP6_STAT_BADOPTIONS); goto bad; } + } else if (nxt == IPPROTO_FRAGMENT) { + if (frg_present++) { + in6_ifstat_inc(rcvif, ifs6_in_hdrerr); + IP6_STATINC(IP6_STAT_BADOPTIONS); + goto bad; + } } #ifdef IPSEC @@ -750,7 +824,9 @@ ip6_input(struct mbuf *m, struct ifnet * */ if ((inet6sw[ip_protox[nxt]].pr_flags & PR_LASTHDR) != 0) { - int error = ipsec6_input(m); + int error; + + error = ipsec6_input(m); if (error) goto bad; } @@ -760,8 +836,13 @@ ip6_input(struct mbuf *m, struct ifnet * nxt = (*inet6sw[ip6_protox[nxt]].pr_input)(&m, &off, nxt); } return; + + bad_unref: + rtcache_unref(rt, ro); + rtcache_percpu_putref(ip6_forward_rt_percpu); bad: m_freem(m); + return; } /* @@ -1264,10 +1345,7 @@ ip6_notify_pmtu(struct in6pcb *in6p, con if (mtu == NULL) return; -#ifdef DIAGNOSTIC - if (so == NULL) /* I believe this is impossible */ - panic("ip6_notify_pmtu: socket is NULL"); -#endif + KASSERT(so != NULL); memset(&mtuctl, 0, sizeof(mtuctl)); /* zero-clear for safety */ mtuctl.ip6m_mtu = *mtu; @@ -1281,8 +1359,8 @@ ip6_notify_pmtu(struct in6pcb *in6p, con if (sbappendaddr(&so->so_rcv, (const struct sockaddr *)dst, NULL, m_mtu) == 0) { + soroverflow(so); m_freem(m_mtu); - /* XXX: should count statistics */ } else sorwakeup(so); @@ -1341,50 +1419,44 @@ ip6_pullexthdr(struct mbuf *m, size_t of } /* - * Get pointer to the previous header followed by the header + * Get offset to the previous header followed by the header * currently processed. - * XXX: This function supposes that - * M includes all headers, - * the next header field and the header length field of each header - * are valid, and - * the sum of each header length equals to OFF. - * Because of these assumptions, this function must be called very - * carefully. Moreover, it will not be used in the near future when - * we develop `neater' mechanism to process extension headers. */ -u_int8_t * +int ip6_get_prevhdr(struct mbuf *m, int off) { struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); - if (off == sizeof(struct ip6_hdr)) - return (&ip6->ip6_nxt); - else { - int len, nxt; - struct ip6_ext *ip6e = NULL; + if (off == sizeof(struct ip6_hdr)) { + return offsetof(struct ip6_hdr, ip6_nxt); + } else if (off < sizeof(struct ip6_hdr)) { + panic("%s: off < sizeof(struct ip6_hdr)", __func__); + } else { + int len, nlen, nxt; + struct ip6_ext ip6e; nxt = ip6->ip6_nxt; len = sizeof(struct ip6_hdr); + nlen = 0; while (len < off) { - ip6e = (struct ip6_ext *)(mtod(m, char *) + len); + m_copydata(m, len, sizeof(ip6e), &ip6e); switch (nxt) { case IPPROTO_FRAGMENT: - len += sizeof(struct ip6_frag); + nlen = sizeof(struct ip6_frag); break; case IPPROTO_AH: - len += (ip6e->ip6e_len + 2) << 2; + nlen = (ip6e.ip6e_len + 2) << 2; break; default: - len += (ip6e->ip6e_len + 1) << 3; + nlen = (ip6e.ip6e_len + 1) << 3; break; } - nxt = ip6e->ip6e_nxt; + len += nlen; + nxt = ip6e.ip6e_nxt; } - if (ip6e) - return (&ip6e->ip6e_nxt); - else - return NULL; + + return (len - nlen); } }