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 1.149.4.7 retrieving revision 1.164.2.4 diff -u -p -r1.149.4.7 -r1.164.2.4 --- src/sys/netinet6/ip6_input.c 2016/07/09 20:25:22 1.149.4.7 +++ src/sys/netinet6/ip6_input.c 2017/03/20 06:57:51 1.164.2.4 @@ -1,4 +1,4 @@ -/* $NetBSD: ip6_input.c,v 1.149.4.7 2016/07/09 20:25:22 skrll Exp $ */ +/* $NetBSD: ip6_input.c,v 1.164.2.4 2017/03/20 06:57:51 pgoyette 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 1.149.4.7 2016/07/09 20:25:22 skrll Exp $"); +__KERNEL_RCSID(0, "$NetBSD: ip6_input.c,v 1.164.2.4 2017/03/20 06:57:51 pgoyette 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 @@ -152,6 +152,14 @@ 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. @@ -216,7 +224,9 @@ ip6intr(void *arg __unused) { struct mbuf *m; +#ifndef NET_MPSAFE mutex_enter(softnet_lock); +#endif while ((m = pktq_dequeue(ip6_pktq)) != NULL) { struct psref psref; struct ifnet *rcvif = m_get_rcvif_psref(m, &psref); @@ -236,7 +246,9 @@ ip6intr(void *arg __unused) ip6_input(m, rcvif); m_put_rcvif_psref(rcvif, &psref); } +#ifndef NET_MPSAFE mutex_exit(softnet_lock); +#endif } extern struct route ip6_forward_rt; @@ -251,7 +263,7 @@ ip6_input(struct mbuf *m, struct ifnet * int nxt, ours = 0, rh_present = 0; struct ifnet *deliverifp = NULL; int srcrt = 0; - const struct rtentry *rt; + struct rtentry *rt = NULL; union { struct sockaddr dst; struct sockaddr_in6 dst6; @@ -354,9 +366,14 @@ 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 @@ -414,7 +431,7 @@ ip6_input(struct mbuf *m, struct ifnet * * to the usage conflict. * in6_setscope() then also checks and rejects the cases where src or * dst are the loopback address and the receiving interface - * is not loopback. + * is not loopback. */ if (__predict_false( m_makewritable(&m, 0, sizeof(struct ip6_hdr), M_DONTWAIT))) @@ -430,19 +447,20 @@ ip6_input(struct mbuf *m, struct ifnet * goto bad; } + ro = 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 +468,7 @@ 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 +479,7 @@ 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 @@ -503,11 +521,13 @@ ip6_input(struct mbuf *m, struct ifnet * 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 +573,7 @@ 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: @@ -565,8 +585,10 @@ ip6_input(struct mbuf *m, struct ifnet * */ if (deliverifp && ip6_getdstifaddr(m) == NULL) { struct in6_ifaddr *ia6; + int s = pserialize_read_enter(); ia6 = in6_ifawithifp(deliverifp, &ip6->ip6_dst); + /* Depends on ip6_setdstifaddr never sleep */ if (ia6 != NULL && ip6_setdstifaddr(m, ia6) == NULL) { /* * XXX maybe we should drop the packet here, @@ -574,6 +596,7 @@ ip6_input(struct mbuf *m, struct ifnet * * to the upper layers. */ } + pserialize_read_exit(s); } /* @@ -589,6 +612,8 @@ ip6_input(struct mbuf *m, struct ifnet * #if 0 /*touches NULL pointer*/ in6_ifstat_inc(rcvif, ifs6_in_discard); #endif + rtcache_unref(rt, ro); + percpu_putref(ip6_forward_rt_percpu); return; /* m have already been freed */ } @@ -612,12 +637,16 @@ 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); + 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); + percpu_putref(ip6_forward_rt_percpu); return; } KASSERT(IP6_HDR_ALIGNED_P(hbh)); @@ -641,7 +670,7 @@ 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) { @@ -663,16 +692,25 @@ 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); + 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); + percpu_putref(ip6_forward_rt_percpu); ip6_forward(m, srcrt); return; } @@ -692,7 +730,7 @@ 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; } /* @@ -701,15 +739,23 @@ ip6_input(struct mbuf *m, struct ifnet * #ifdef IFA_STATS if (deliverifp != NULL) { struct in6_ifaddr *ia6; + int s = pserialize_read_enter(); ia6 = in6_ifawithifp(deliverifp, &ip6->ip6_dst); if (ia6) ia6->ia_ifa.ifa_data.ifad_inbytes += m->m_pkthdr.len; + pserialize_read_exit(s); } #endif IP6_STATINC(IP6_STAT_DELIVERED); in6_ifstat_inc(deliverifp, ifs6_in_deliver); nest = 0; + if (rt != NULL) { + rtcache_unref(rt, ro); + rt = NULL; + } + percpu_putref(ip6_forward_rt_percpu); + rh_present = 0; while (nxt != IPPROTO_DONE) { if (ip6_hdrnestlimit && (++nest > ip6_hdrnestlimit)) { @@ -745,18 +791,29 @@ ip6_input(struct mbuf *m, struct ifnet * */ if ((inet6sw[ip_protox[nxt]].pr_flags & PR_LASTHDR) != 0) { - int error = ipsec6_input(m); + int error; + + SOFTNET_LOCK(); + error = ipsec6_input(m); + SOFTNET_UNLOCK(); if (error) goto bad; } } #endif /* IPSEC */ + SOFTNET_LOCK(); nxt = (*inet6sw[ip6_protox[nxt]].pr_input)(&m, &off, nxt); + SOFTNET_UNLOCK(); } return; + + bad_unref: + rtcache_unref(rt, ro); + percpu_putref(ip6_forward_rt_percpu); bad: m_freem(m); + return; } /* @@ -802,7 +859,7 @@ ip6_getdstifaddr(struct mbuf *m) * rtalertp - XXX: should be stored more smart way */ int -ip6_hopopts_input(u_int32_t *plenp, u_int32_t *rtalertp, +ip6_hopopts_input(u_int32_t *plenp, u_int32_t *rtalertp, struct mbuf **mp, int *offp) { struct mbuf *m = *mp; @@ -847,7 +904,7 @@ ip6_hopopts_input(u_int32_t *plenp, u_in * opthead + hbhlen is located in continuous memory region. */ static int -ip6_process_hopopts(struct mbuf *m, u_int8_t *opthead, int hbhlen, +ip6_process_hopopts(struct mbuf *m, u_int8_t *opthead, int hbhlen, u_int32_t *rtalertp, u_int32_t *plenp) { struct ip6_hdr *ip6; @@ -1022,7 +1079,7 @@ ip6_unknown_opt(u_int8_t *optp, struct m * you are using IP6_EXTHDR_CHECK() not m_pulldown()) */ void -ip6_savecontrol(struct in6pcb *in6p, struct mbuf **mp, +ip6_savecontrol(struct in6pcb *in6p, struct mbuf **mp, struct ip6_hdr *ip6, struct mbuf *m) { #ifdef RFC2292 @@ -1259,10 +1316,7 @@ 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; @@ -1525,66 +1579,6 @@ ip6_delaux(struct mbuf *m) m_tag_delete(m, mtag); } -#ifdef GATEWAY -/* - * sysctl helper routine for net.inet.ip6.maxflows. Since - * we could reduce this value, call ip6flow_reap(); - */ -static int -sysctl_net_inet6_ip6_maxflows(SYSCTLFN_ARGS) -{ - int error; - - error = sysctl_lookup(SYSCTLFN_CALL(rnode)); - if (error || newp == NULL) - return (error); - - mutex_enter(softnet_lock); - KERNEL_LOCK(1, NULL); - - ip6flow_reap(0); - - KERNEL_UNLOCK_ONE(NULL); - mutex_exit(softnet_lock); - - return (0); -} - -static int -sysctl_net_inet6_ip6_hashsize(SYSCTLFN_ARGS) -{ - int error, tmp; - struct sysctlnode node; - - node = *rnode; - tmp = ip6_hashsize; - node.sysctl_data = &tmp; - error = sysctl_lookup(SYSCTLFN_CALL(&node)); - if (error || newp == NULL) - return (error); - - if ((tmp & (tmp - 1)) == 0 && tmp != 0) { - /* - * Can only fail due to malloc() - */ - mutex_enter(softnet_lock); - KERNEL_LOCK(1, NULL); - - error = ip6flow_invalidate_all(tmp); - - KERNEL_UNLOCK_ONE(NULL); - mutex_exit(softnet_lock); - } else { - /* - * EINVAL if not a power of 2 - */ - error = EINVAL; - } - - return error; -} -#endif /* GATEWAY */ - /* * System control for IP6 */ @@ -1903,22 +1897,6 @@ sysctl_net_inet6_ip6_setup(struct sysctl NULL, 0, &ip6_mcast_pmtu, 0, CTL_NET, PF_INET6, IPPROTO_IPV6, CTL_CREATE, CTL_EOL); -#ifdef GATEWAY - sysctl_createv(clog, 0, NULL, NULL, - CTLFLAG_PERMANENT|CTLFLAG_READWRITE, - CTLTYPE_INT, "maxflows", - SYSCTL_DESCR("Number of flows for fast forwarding (IPv6)"), - sysctl_net_inet6_ip6_maxflows, 0, &ip6_maxflows, 0, - CTL_NET, PF_INET6, IPPROTO_IPV6, - CTL_CREATE, CTL_EOL); - sysctl_createv(clog, 0, NULL, NULL, - CTLFLAG_PERMANENT|CTLFLAG_READWRITE, - CTLTYPE_INT, "hashsize", - SYSCTL_DESCR("Size of hash table for fast forwarding (IPv6)"), - sysctl_net_inet6_ip6_hashsize, 0, &ip6_hashsize, 0, - CTL_NET, PF_INET6, IPPROTO_IPV6, - CTL_CREATE, CTL_EOL); -#endif /* anonportalgo RFC6056 subtree */ const struct sysctlnode *portalgo_node; sysctl_createv(clog, 0, NULL, &portalgo_node,