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/udp_usrreq.c,v rcsdiff: /ftp/cvs/cvsroot/src/sys/netinet/udp_usrreq.c,v: warning: Unknown phrases like `commitid ...;' are present. retrieving revision 1.119 retrieving revision 1.173 diff -u -p -r1.119 -r1.173 --- src/sys/netinet/udp_usrreq.c 2004/04/18 23:35:56 1.119 +++ src/sys/netinet/udp_usrreq.c 2008/08/06 15:01:23 1.173 @@ -1,4 +1,4 @@ -/* $NetBSD: udp_usrreq.c,v 1.119 2004/04/18 23:35:56 matt Exp $ */ +/* $NetBSD: udp_usrreq.c,v 1.173 2008/08/06 15:01:23 plunky Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -61,7 +61,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.119 2004/04/18 23:35:56 matt Exp $"); +__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.173 2008/08/06 15:01:23 plunky Exp $"); #include "opt_inet.h" #include "opt_ipsec.h" @@ -94,13 +94,17 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #include #include #include +#include #ifdef INET6 #include #include #include +#include #include #include +#include +#include #endif #ifndef INET6 @@ -117,15 +121,20 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #ifdef FAST_IPSEC #include +#include +#include +#include #ifdef INET6 #include #endif -#endif /* FAST_IPSEC*/ +#endif /* FAST_IPSEC */ #ifdef IPSEC #include +#include +#include #include -#endif /*IPSEC*/ +#endif /* IPSEC */ #ifdef IPKDB #include @@ -135,26 +144,30 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c * UDP protocol implementation. * Per RFC 768, August, 1980. */ -#ifndef COMPAT_42 int udpcksum = 1; -#else -int udpcksum = 0; /* XXX */ -#endif +int udp_do_loopback_cksum = 0; struct inpcbtable udbtable; -struct udpstat udpstat; + +percpu_t *udpstat_percpu; #ifdef INET +#ifdef IPSEC_NAT_T +static int udp4_espinudp (struct mbuf **, int, struct sockaddr *, + struct socket *); +#endif static void udp4_sendup (struct mbuf *, int, struct sockaddr *, struct socket *); static int udp4_realinput (struct sockaddr_in *, struct sockaddr_in *, - struct mbuf *, int); + struct mbuf **, int); +static int udp4_input_checksum(struct mbuf *, const struct udphdr *, int, int); #endif #ifdef INET6 static void udp6_sendup (struct mbuf *, int, struct sockaddr *, struct socket *); static int udp6_realinput (int, struct sockaddr_in6 *, struct sockaddr_in6 *, struct mbuf *, int); +static int udp6_input_checksum(struct mbuf *, const struct udphdr *, int, int); #endif #ifdef INET static void udp_notify (struct inpcb *, int); @@ -166,14 +179,15 @@ static void udp_notify (struct inpcb *, int udbhashsize = UDBHASHSIZE; #ifdef MBUFTRACE -struct mowner udp_mowner = { "udp" }; -struct mowner udp_rx_mowner = { "udp", "rx" }; -struct mowner udp_tx_mowner = { "udp", "tx" }; +struct mowner udp_mowner = MOWNER_INIT("udp", ""); +struct mowner udp_rx_mowner = MOWNER_INIT("udp", "rx"); +struct mowner udp_tx_mowner = MOWNER_INIT("udp", "tx"); #endif #ifdef UDP_CSUM_COUNTERS #include +#if defined(INET) struct evcnt udp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "udp", "hwcsum bad"); struct evcnt udp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, @@ -183,6 +197,28 @@ struct evcnt udp_hwcsum_data = EVCNT_INI struct evcnt udp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "udp", "swcsum"); +EVCNT_ATTACH_STATIC(udp_hwcsum_bad); +EVCNT_ATTACH_STATIC(udp_hwcsum_ok); +EVCNT_ATTACH_STATIC(udp_hwcsum_data); +EVCNT_ATTACH_STATIC(udp_swcsum); +#endif /* defined(INET) */ + +#if defined(INET6) +struct evcnt udp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, + NULL, "udp6", "hwcsum bad"); +struct evcnt udp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, + NULL, "udp6", "hwcsum ok"); +struct evcnt udp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, + NULL, "udp6", "hwcsum data"); +struct evcnt udp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, + NULL, "udp6", "swcsum"); + +EVCNT_ATTACH_STATIC(udp6_hwcsum_bad); +EVCNT_ATTACH_STATIC(udp6_hwcsum_ok); +EVCNT_ATTACH_STATIC(udp6_hwcsum_data); +EVCNT_ATTACH_STATIC(udp6_swcsum); +#endif /* defined(INET6) */ + #define UDP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ #else @@ -197,19 +233,114 @@ udp_init(void) in_pcbinit(&udbtable, udbhashsize, udbhashsize); -#ifdef UDP_CSUM_COUNTERS - evcnt_attach_static(&udp_hwcsum_bad); - evcnt_attach_static(&udp_hwcsum_ok); - evcnt_attach_static(&udp_hwcsum_data); - evcnt_attach_static(&udp_swcsum); -#endif /* UDP_CSUM_COUNTERS */ - MOWNER_ATTACH(&udp_tx_mowner); MOWNER_ATTACH(&udp_rx_mowner); MOWNER_ATTACH(&udp_mowner); + +#ifdef INET + udpstat_percpu = percpu_alloc(sizeof(uint64_t) * UDP_NSTATS); +#endif +#ifdef INET6 + udp6stat_percpu = percpu_alloc(sizeof(uint64_t) * UDP6_NSTATS); +#endif +} + +/* + * Checksum extended UDP header and data. + */ + +int +udp_input_checksum(int af, struct mbuf *m, const struct udphdr *uh, + int iphlen, int len) +{ + + switch (af) { +#ifdef INET + case AF_INET: + return udp4_input_checksum(m, uh, iphlen, len); +#endif +#ifdef INET6 + case AF_INET6: + return udp6_input_checksum(m, uh, iphlen, len); +#endif + } +#ifdef DIAGNOSTIC + panic("udp_input_checksum: unknown af %d", af); +#endif + /* NOTREACHED */ + return -1; } #ifdef INET + +/* + * Checksum extended UDP header and data. + */ + +static int +udp4_input_checksum(struct mbuf *m, const struct udphdr *uh, + int iphlen, int len) +{ + + /* + * XXX it's better to record and check if this mbuf is + * already checked. + */ + + if (uh->uh_sum == 0) + return 0; + + switch (m->m_pkthdr.csum_flags & + ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_UDPv4) | + M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { + case M_CSUM_UDPv4|M_CSUM_TCP_UDP_BAD: + UDP_CSUM_COUNTER_INCR(&udp_hwcsum_bad); + goto badcsum; + + case M_CSUM_UDPv4|M_CSUM_DATA: { + u_int32_t hw_csum = m->m_pkthdr.csum_data; + + UDP_CSUM_COUNTER_INCR(&udp_hwcsum_data); + if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { + const struct ip *ip = + mtod(m, const struct ip *); + + hw_csum = in_cksum_phdr(ip->ip_src.s_addr, + ip->ip_dst.s_addr, + htons(hw_csum + len + IPPROTO_UDP)); + } + if ((hw_csum ^ 0xffff) != 0) + goto badcsum; + break; + } + + case M_CSUM_UDPv4: + /* Checksum was okay. */ + UDP_CSUM_COUNTER_INCR(&udp_hwcsum_ok); + break; + + default: + /* + * Need to compute it ourselves. Maybe skip checksum + * on loopback interfaces. + */ + if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & + IFF_LOOPBACK) || + udp_do_loopback_cksum)) { + UDP_CSUM_COUNTER_INCR(&udp_swcsum); + if (in4_cksum(m, IPPROTO_UDP, iphlen, len) != 0) + goto badcsum; + } + break; + } + + return 0; + +badcsum: + UDP_STATINC(UDP_STAT_BADSUM); + return -1; +} + void udp_input(struct mbuf *m, ...) { @@ -228,7 +359,7 @@ udp_input(struct mbuf *m, ...) va_end(ap); MCLAIM(m, &udp_rx_mowner); - udpstat.udps_ipackets++; + UDP_STATINC(UDP_STAT_IPACKETS); /* * Get IP and UDP header together in first mbuf. @@ -236,7 +367,7 @@ udp_input(struct mbuf *m, ...) ip = mtod(m, struct ip *); IP6_EXTHDR_GET(uh, struct udphdr *, m, iphlen, sizeof(struct udphdr)); if (uh == NULL) { - udpstat.udps_hdrops++; + UDP_STATINC(UDP_STAT_HDROPS); return; } KASSERT(UDP_HDR_ALIGNED_P(uh)); @@ -253,7 +384,7 @@ udp_input(struct mbuf *m, ...) len = ntohs((u_int16_t)uh->uh_ulen); if (ip_len != iphlen + len) { if (ip_len < iphlen + len || len < sizeof(struct udphdr)) { - udpstat.udps_badlen++; + UDP_STATINC(UDP_STAT_BADLEN); goto bad; } m_adj(m, iphlen + len - ip_len); @@ -262,53 +393,17 @@ udp_input(struct mbuf *m, ...) /* * Checksum extended UDP header and data. */ - if (uh->uh_sum) { - switch (m->m_pkthdr.csum_flags & - ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_UDPv4) | - M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { - case M_CSUM_UDPv4|M_CSUM_TCP_UDP_BAD: - UDP_CSUM_COUNTER_INCR(&udp_hwcsum_bad); - goto badcsum; - - case M_CSUM_UDPv4|M_CSUM_DATA: { - u_int32_t hw_csum = m->m_pkthdr.csum_data; - UDP_CSUM_COUNTER_INCR(&udp_hwcsum_data); - if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) - hw_csum = in_cksum_phdr(ip->ip_src.s_addr, - ip->ip_dst.s_addr, - htons(hw_csum + len + IPPROTO_UDP)); - if ((hw_csum ^ 0xffff) != 0) - goto badcsum; - break; - } - - case M_CSUM_UDPv4: - /* Checksum was okay. */ - UDP_CSUM_COUNTER_INCR(&udp_hwcsum_ok); - break; - - default: - /* Need to compute it ourselves. */ - UDP_CSUM_COUNTER_INCR(&udp_swcsum); - if (in4_cksum(m, IPPROTO_UDP, iphlen, len) != 0) - goto badcsum; - break; - } - } + if (udp4_input_checksum(m, uh, iphlen, len)) + goto badcsum; /* construct source and dst sockaddrs. */ - bzero(&src, sizeof(src)); - src.sin_family = AF_INET; - src.sin_len = sizeof(struct sockaddr_in); - bcopy(&ip->ip_src, &src.sin_addr, sizeof(src.sin_addr)); - src.sin_port = uh->uh_sport; - bzero(&dst, sizeof(dst)); - dst.sin_family = AF_INET; - dst.sin_len = sizeof(struct sockaddr_in); - bcopy(&ip->ip_dst, &dst.sin_addr, sizeof(dst.sin_addr)); - dst.sin_port = uh->uh_dport; + sockaddr_in_init(&src, &ip->ip_src, uh->uh_sport); + sockaddr_in_init(&dst, &ip->ip_dst, uh->uh_dport); - n = udp4_realinput(&src, &dst, m, iphlen); + if ((n = udp4_realinput(&src, &dst, &m, iphlen)) == -1) { + UDP_STATINC(UDP_STAT_HDROPS); + return; + } #ifdef INET6 if (IN_MULTICAST(ip->ip_dst.s_addr) || n == 0) { struct sockaddr_in6 src6, dst6; @@ -334,10 +429,10 @@ udp_input(struct mbuf *m, ...) if (n == 0) { if (m->m_flags & (M_BCAST | M_MCAST)) { - udpstat.udps_noportbcast++; + UDP_STATINC(UDP_STAT_NOPORTBCAST); goto bad; } - udpstat.udps_noport++; + UDP_STATINC(UDP_STAT_NOPORT); #ifdef IPKDB if (checkipkdb(&ip->ip_src, uh->uh_sport, uh->uh_dport, m, iphlen + sizeof(struct udphdr), @@ -360,11 +455,62 @@ bad: badcsum: m_freem(m); - udpstat.udps_badsum++; } #endif #ifdef INET6 +static int +udp6_input_checksum(struct mbuf *m, const struct udphdr *uh, int off, int len) +{ + + /* + * XXX it's better to record and check if this mbuf is + * already checked. + */ + + if (__predict_false((m->m_flags & M_LOOP) && !udp_do_loopback_cksum)) { + goto good; + } + if (uh->uh_sum == 0) { + UDP6_STATINC(UDP6_STAT_NOSUM); + goto bad; + } + + switch (m->m_pkthdr.csum_flags & + ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_UDPv6) | + M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { + case M_CSUM_UDPv6|M_CSUM_TCP_UDP_BAD: + UDP_CSUM_COUNTER_INCR(&udp6_hwcsum_bad); + UDP6_STATINC(UDP6_STAT_BADSUM); + goto bad; + +#if 0 /* notyet */ + case M_CSUM_UDPv6|M_CSUM_DATA: +#endif + + case M_CSUM_UDPv6: + /* Checksum was okay. */ + UDP_CSUM_COUNTER_INCR(&udp6_hwcsum_ok); + break; + + default: + /* + * Need to compute it ourselves. Maybe skip checksum + * on loopback interfaces. + */ + UDP_CSUM_COUNTER_INCR(&udp6_swcsum); + if (in6_cksum(m, IPPROTO_UDP, off, len) != 0) { + UDP6_STATINC(UDP6_STAT_BADSUM); + goto bad; + } + } + +good: + return 0; +bad: + return -1; +} + int udp6_input(struct mbuf **mp, int *offp, int proto) { @@ -385,13 +531,13 @@ udp6_input(struct mbuf **mp, int *offp, } #endif - udp6stat.udp6s_ipackets++; + UDP6_STATINC(UDP6_STAT_IPACKETS); /* check for jumbogram is done in ip6_input. we can trust pkthdr.len */ plen = m->m_pkthdr.len - off; IP6_EXTHDR_GET(uh, struct udphdr *, m, off, sizeof(struct udphdr)); if (uh == NULL) { - ip6stat.ip6s_tooshort++; + IP6_STATINC(IP6_STAT_TOOSHORT); return IPPROTO_DONE; } KASSERT(UDP_HDR_ALIGNED_P(uh)); @@ -404,7 +550,7 @@ udp6_input(struct mbuf **mp, int *offp, ulen = plen; if (plen != ulen) { - udp6stat.udp6s_badlen++; + UDP6_STATINC(UDP6_STAT_BADLEN); goto bad; } @@ -420,40 +566,32 @@ udp6_input(struct mbuf **mp, int *offp, } /* - * Checksum extended UDP header and data. + * Checksum extended UDP header and data. Maybe skip checksum + * on loopback interfaces. */ - if (uh->uh_sum == 0) { - udp6stat.udp6s_nosum++; - goto bad; - } - if (in6_cksum(m, IPPROTO_UDP, off, ulen) != 0) { - udp6stat.udp6s_badsum++; + if (udp6_input_checksum(m, uh, off, ulen)) goto bad; - } /* * Construct source and dst sockaddrs. - * Note that ifindex (s6_addr16[1]) is already filled. */ bzero(&src, sizeof(src)); src.sin6_family = AF_INET6; src.sin6_len = sizeof(struct sockaddr_in6); - /* KAME hack: recover scopeid */ - (void)in6_recoverscope(&src, &ip6->ip6_src, m->m_pkthdr.rcvif); + src.sin6_addr = ip6->ip6_src; src.sin6_port = uh->uh_sport; bzero(&dst, sizeof(dst)); dst.sin6_family = AF_INET6; dst.sin6_len = sizeof(struct sockaddr_in6); - /* KAME hack: recover scopeid */ - (void)in6_recoverscope(&dst, &ip6->ip6_dst, m->m_pkthdr.rcvif); + dst.sin6_addr = ip6->ip6_dst; dst.sin6_port = uh->uh_dport; if (udp6_realinput(AF_INET6, &src, &dst, m, off) == 0) { if (m->m_flags & M_MCAST) { - udp6stat.udp6s_noportmcast++; + UDP6_STATINC(UDP6_STAT_NOPORTMCAST); goto bad; } - udp6stat.udp6s_noport++; + UDP6_STATINC(UDP6_STAT_NOPORT); icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0); m = NULL; } @@ -491,15 +629,15 @@ udp4_sendup(struct mbuf *m, int off /* o #if defined(IPSEC) || defined(FAST_IPSEC) /* check AH/ESP integrity. */ if (so != NULL && ipsec4_in_reject_so(m, so)) { - ipsecstat.in_polvio++; - if ((n = m_copy(m, 0, M_COPYALL)) != NULL) + IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); + if ((n = m_copypacket(m, M_DONTWAIT)) != NULL) icmp_error(n, ICMP_UNREACH, ICMP_UNREACH_ADMIN_PROHIBIT, 0, 0); return; } #endif /*IPSEC*/ - if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { + if ((n = m_copypacket(m, M_DONTWAIT)) != NULL) { if (inp && (inp->inp_flags & INP_CONTROLOPTS || so->so_options & SO_TIMESTAMP)) { struct ip *ip = mtod(n, struct ip *); @@ -512,7 +650,8 @@ udp4_sendup(struct mbuf *m, int off /* o m_freem(n); if (opts) m_freem(opts); - udpstat.udps_fullsock++; + so->so_rcv.sb_overflowed++; + UDP_STATINC(UDP_STAT_FULLSOCK); } else sorwakeup(so); } @@ -537,15 +676,15 @@ udp6_sendup(struct mbuf *m, int off /* o #if defined(IPSEC) || defined(FAST_IPSEC) /* check AH/ESP integrity. */ if (so != NULL && ipsec6_in_reject_so(m, so)) { - ipsec6stat.in_polvio++; - if ((n = m_copy(m, 0, M_COPYALL)) != NULL) + IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); + if ((n = m_copypacket(m, M_DONTWAIT)) != NULL) icmp6_error(n, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADMIN, 0); return; } #endif /*IPSEC*/ - if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { + if ((n = m_copypacket(m, M_DONTWAIT)) != NULL) { if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS || in6p->in6p_socket->so_options & SO_TIMESTAMP)) { struct ip6_hdr *ip6 = mtod(n, struct ip6_hdr *); @@ -557,7 +696,8 @@ udp6_sendup(struct mbuf *m, int off /* o m_freem(n); if (opts) m_freem(opts); - udp6stat.udp6s_fullsock++; + so->so_rcv.sb_overflowed++; + UDP6_STATINC(UDP6_STAT_FULLSOCK); } else sorwakeup(so); } @@ -567,13 +707,14 @@ udp6_sendup(struct mbuf *m, int off /* o #ifdef INET static int udp4_realinput(struct sockaddr_in *src, struct sockaddr_in *dst, - struct mbuf *m, int off /* offset of udphdr */) + struct mbuf **mp, int off /* offset of udphdr */) { u_int16_t *sport, *dport; int rcvcnt; struct in_addr *src4, *dst4; struct inpcb_hdr *inph; struct inpcb *inp; + struct mbuf *m = *mp; rcvcnt = 0; off += sizeof(struct udphdr); /* now, offset of payload */ @@ -650,12 +791,40 @@ udp4_realinput(struct sockaddr_in *src, */ inp = in_pcblookup_connect(&udbtable, *src4, *sport, *dst4, *dport); if (inp == 0) { - ++udpstat.udps_pcbhashmiss; + UDP_STATINC(UDP_STAT_PCBHASHMISS); inp = in_pcblookup_bind(&udbtable, *dst4, *dport); if (inp == 0) return rcvcnt; } +#ifdef IPSEC_NAT_T + /* Handle ESP over UDP */ + if (inp->inp_flags & INP_ESPINUDP_ALL) { + struct sockaddr *sa = (struct sockaddr *)src; + + switch(udp4_espinudp(mp, off, sa, inp->inp_socket)) { + case -1: /* Error, m was freeed */ + rcvcnt = -1; + goto bad; + break; + + case 1: /* ESP over UDP */ + rcvcnt++; + goto bad; + break; + + case 0: /* plain UDP */ + default: /* Unexpected */ + /* + * Normal UDP processing will take place + * m may have changed. + */ + m = *mp; + break; + } + } +#endif + udp4_sendup(m, off, (struct sockaddr *)src, inp->inp_socket); rcvcnt++; } @@ -672,7 +841,7 @@ udp6_realinput(int af, struct sockaddr_i { u_int16_t sport, dport; int rcvcnt; - struct in6_addr src6, dst6; + struct in6_addr src6, *dst6; const struct in_addr *dst4; struct inpcb_hdr *inph; struct in6pcb *in6p; @@ -685,13 +854,18 @@ udp6_realinput(int af, struct sockaddr_i if (src->sin6_family != AF_INET6 || dst->sin6_family != AF_INET6) goto bad; - in6_embedscope(&src6, src, NULL, NULL); + src6 = src->sin6_addr; + if (sa6_recoverscope(src) != 0) { + /* XXX: should be impossible. */ + goto bad; + } sport = src->sin6_port; - in6_embedscope(&dst6, dst, NULL, NULL); + dport = dst->sin6_port; dst4 = (struct in_addr *)&dst->sin6_addr.s6_addr[12]; + dst6 = &dst->sin6_addr; - if (IN6_IS_ADDR_MULTICAST(&dst6) || + if (IN6_IS_ADDR_MULTICAST(dst6) || (af == AF_INET && IN_MULTICAST(dst4->s_addr))) { /* * Deliver a multicast or broadcast datagram to *all* sockets @@ -724,10 +898,11 @@ udp6_realinput(int af, struct sockaddr_i if (in6p->in6p_lport != dport) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_laddr)) { - if (!IN6_ARE_ADDR_EQUAL(&in6p->in6p_laddr, &dst6)) + if (!IN6_ARE_ADDR_EQUAL(&in6p->in6p_laddr, + dst6)) continue; } else { - if (IN6_IS_ADDR_V4MAPPED(&dst6) && + if (IN6_IS_ADDR_V4MAPPED(dst6) && (in6p->in6p_flags & IN6P_IPV6_V6ONLY)) continue; } @@ -761,11 +936,11 @@ udp6_realinput(int af, struct sockaddr_i /* * Locate pcb for datagram. */ - in6p = in6_pcblookup_connect(&udbtable, &src6, sport, - &dst6, dport, 0); + in6p = in6_pcblookup_connect(&udbtable, &src6, sport, dst6, + dport, 0); if (in6p == 0) { - ++udpstat.udps_pcbhashmiss; - in6p = in6_pcblookup_bind(&udbtable, &dst6, dport, 0); + UDP_STATINC(UDP_STAT_PCBHASHMISS); + in6p = in6_pcblookup_bind(&udbtable, dst6, dport, 0); if (in6p == 0) return rcvcnt; } @@ -793,7 +968,7 @@ udp_notify(struct inpcb *inp, int errno) } void * -udp_ctlinput(int cmd, struct sockaddr *sa, void *v) +udp_ctlinput(int cmd, const struct sockaddr *sa, void *v) { struct ip *ip = v; struct udphdr *uh; @@ -813,22 +988,107 @@ udp_ctlinput(int cmd, struct sockaddr *s else if (errno == 0) return NULL; if (ip) { - uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); - in_pcbnotify(&udbtable, satosin(sa)->sin_addr, uh->uh_dport, + uh = (struct udphdr *)((char *)ip + (ip->ip_hl << 2)); + in_pcbnotify(&udbtable, satocsin(sa)->sin_addr, uh->uh_dport, ip->ip_src, uh->uh_sport, errno, notify); /* XXX mapped address case */ } else - in_pcbnotifyall(&udbtable, satosin(sa)->sin_addr, errno, + in_pcbnotifyall(&udbtable, satocsin(sa)->sin_addr, errno, notify); return NULL; } int +udp_ctloutput(int op, struct socket *so, struct sockopt *sopt) +{ + int s; + int error = 0; + struct inpcb *inp; + int family; + int optval; + + family = so->so_proto->pr_domain->dom_family; + + s = splsoftnet(); + switch (family) { +#ifdef INET + case PF_INET: + if (sopt->sopt_level != IPPROTO_UDP) { + error = ip_ctloutput(op, so, sopt); + goto end; + } + break; +#endif +#ifdef INET6 + case PF_INET6: + if (sopt->sopt_level != IPPROTO_UDP) { + error = ip6_ctloutput(op, so, sopt); + goto end; + } + break; +#endif + default: + error = EAFNOSUPPORT; + goto end; + } + + + switch (op) { + case PRCO_SETOPT: + inp = sotoinpcb(so); + + switch (sopt->sopt_name) { + case UDP_ENCAP: + error = sockopt_getint(sopt, &optval); + if (error) + break; + + switch(optval) { +#ifdef IPSEC_NAT_T + case 0: + inp->inp_flags &= ~INP_ESPINUDP_ALL; + break; + + case UDP_ENCAP_ESPINUDP: + inp->inp_flags &= ~INP_ESPINUDP_ALL; + inp->inp_flags |= INP_ESPINUDP; + break; + + case UDP_ENCAP_ESPINUDP_NON_IKE: + inp->inp_flags &= ~INP_ESPINUDP_ALL; + inp->inp_flags |= INP_ESPINUDP_NON_IKE; + break; +#endif + default: + error = EINVAL; + break; + } + break; + + default: + error = ENOPROTOOPT; + break; + } + break; + + default: + error = EINVAL; + break; + } + +end: + splx(s); + return error; +} + + +int udp_output(struct mbuf *m, ...) { struct inpcb *inp; struct udpiphdr *ui; + struct route *ro; int len = m->m_pkthdr.len; int error = 0; va_list ap; @@ -869,6 +1129,8 @@ udp_output(struct mbuf *m, ...) ui->ui_dport = inp->inp_fport; ui->ui_ulen = htons((u_int16_t)len + sizeof(struct udphdr)); + ro = &inp->inp_route; + /* * Set up checksum and output datagram. */ @@ -887,9 +1149,9 @@ udp_output(struct mbuf *m, ...) ((struct ip *)ui)->ip_len = htons(sizeof (struct udpiphdr) + len); ((struct ip *)ui)->ip_ttl = inp->inp_ip.ip_ttl; /* XXX */ ((struct ip *)ui)->ip_tos = inp->inp_ip.ip_tos; /* XXX */ - udpstat.udps_opackets++; + UDP_STATINC(UDP_STAT_OPACKETS); - return (ip_output(m, inp->inp_options, &inp->inp_route, + return (ip_output(m, inp->inp_options, ro, inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST), inp->inp_moptions, inp->inp_socket)); @@ -905,30 +1167,36 @@ int udp_recvspace = 40 * (1024 + sizeof( /*ARGSUSED*/ int udp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, - struct mbuf *control, struct proc *p) + struct mbuf *control, struct lwp *l) { struct inpcb *inp; int s; int error = 0; if (req == PRU_CONTROL) - return (in_control(so, (long)m, (caddr_t)nam, - (struct ifnet *)control, p)); + return (in_control(so, (long)m, (void *)nam, + (struct ifnet *)control, l)); + + s = splsoftnet(); if (req == PRU_PURGEIF) { + mutex_enter(softnet_lock); in_pcbpurgeif0(&udbtable, (struct ifnet *)control); in_purgeif((struct ifnet *)control); in_pcbpurgeif(&udbtable, (struct ifnet *)control); + mutex_exit(softnet_lock); + splx(s); return (0); } - s = splsoftnet(); inp = sotoinpcb(so); #ifdef DIAGNOSTIC if (req != PRU_SEND && req != PRU_SENDOOB && control) panic("udp_usrreq: unexpected control mbuf"); #endif - if (inp == 0 && req != PRU_ATTACH) { + if (req == PRU_ATTACH) { + sosetlock(so); + } else if (inp == 0) { error = EINVAL; goto release; } @@ -966,7 +1234,7 @@ udp_usrreq(struct socket *so, int req, s break; case PRU_BIND: - error = in_pcbbind(inp, nam, p); + error = in_pcbbind(inp, nam, l); break; case PRU_LISTEN: @@ -974,7 +1242,7 @@ udp_usrreq(struct socket *so, int req, s break; case PRU_CONNECT: - error = in_pcbconnect(inp, nam); + error = in_pcbconnect(inp, nam, l); if (error) break; soisconnected(so); @@ -1016,7 +1284,7 @@ udp_usrreq(struct socket *so, int req, s error = EISCONN; goto die; } - error = in_pcbconnect(inp, nam); + error = in_pcbconnect(inp, nam, l); if (error) goto die; } else { @@ -1072,6 +1340,13 @@ release: return (error); } +static int +sysctl_net_inet_udp_stats(SYSCTLFN_ARGS) +{ + + return (NETSTAT_SYSCTL(udpstat_percpu, UDP_NSTATS)); +} + /* * Sysctl for udp variables. */ @@ -1090,27 +1365,201 @@ SYSCTL_SETUP(sysctl_net_inet_udp_setup, CTL_NET, PF_INET, CTL_EOL); sysctl_createv(clog, 0, NULL, NULL, CTLFLAG_PERMANENT, - CTLTYPE_NODE, "udp", NULL, + CTLTYPE_NODE, "udp", + SYSCTL_DESCR("UDPv4 related settings"), NULL, 0, NULL, 0, CTL_NET, PF_INET, IPPROTO_UDP, CTL_EOL); sysctl_createv(clog, 0, NULL, NULL, CTLFLAG_PERMANENT|CTLFLAG_READWRITE, - CTLTYPE_INT, "checksum", NULL, + CTLTYPE_INT, "checksum", + SYSCTL_DESCR("Compute UDP checksums"), NULL, 0, &udpcksum, 0, CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_CHECKSUM, CTL_EOL); sysctl_createv(clog, 0, NULL, NULL, CTLFLAG_PERMANENT|CTLFLAG_READWRITE, - CTLTYPE_INT, "sendspace", NULL, + CTLTYPE_INT, "sendspace", + SYSCTL_DESCR("Default UDP send buffer size"), NULL, 0, &udp_sendspace, 0, CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_SENDSPACE, CTL_EOL); sysctl_createv(clog, 0, NULL, NULL, CTLFLAG_PERMANENT|CTLFLAG_READWRITE, - CTLTYPE_INT, "recvspace", NULL, + CTLTYPE_INT, "recvspace", + SYSCTL_DESCR("Default UDP receive buffer size"), NULL, 0, &udp_recvspace, 0, CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_RECVSPACE, CTL_EOL); + sysctl_createv(clog, 0, NULL, NULL, + CTLFLAG_PERMANENT|CTLFLAG_READWRITE, + CTLTYPE_INT, "do_loopback_cksum", + SYSCTL_DESCR("Perform UDP checksum on loopback"), + NULL, 0, &udp_do_loopback_cksum, 0, + CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_LOOPBACKCKSUM, + CTL_EOL); + sysctl_createv(clog, 0, NULL, NULL, + CTLFLAG_PERMANENT, + CTLTYPE_STRUCT, "pcblist", + SYSCTL_DESCR("UDP protocol control block list"), + sysctl_inpcblist, 0, &udbtable, 0, + CTL_NET, PF_INET, IPPROTO_UDP, CTL_CREATE, + CTL_EOL); + sysctl_createv(clog, 0, NULL, NULL, + CTLFLAG_PERMANENT, + CTLTYPE_STRUCT, "stats", + SYSCTL_DESCR("UDP statistics"), + sysctl_net_inet_udp_stats, 0, NULL, 0, + CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_STATS, + CTL_EOL); +} +#endif + +void +udp_statinc(u_int stat) +{ + + KASSERT(stat < UDP_NSTATS); + UDP_STATINC(stat); +} + +#if (defined INET && defined IPSEC_NAT_T) +/* + * Returns: + * 1 if the packet was processed + * 0 if normal UDP processing should take place + * -1 if an error occurent and m was freed + */ +static int +udp4_espinudp(struct mbuf **mp, int off, struct sockaddr *src, + struct socket *so) +{ + size_t len; + void *data; + struct inpcb *inp; + size_t skip = 0; + size_t minlen; + size_t iphdrlen; + struct ip *ip; + struct mbuf *n; + struct m_tag *tag; + struct udphdr *udphdr; + u_int16_t sport, dport; + struct mbuf *m = *mp; + + /* + * Collapse the mbuf chain if the first mbuf is too short + * The longest case is: UDP + non ESP marker + ESP + */ + minlen = off + sizeof(u_int64_t) + sizeof(struct esp); + if (minlen > m->m_pkthdr.len) + minlen = m->m_pkthdr.len; + + if (m->m_len < minlen) { + if ((*mp = m_pullup(m, minlen)) == NULL) { + printf("udp4_espinudp: m_pullup failed\n"); + return -1; + } + m = *mp; + } + + len = m->m_len - off; + data = mtod(m, char *) + off; + inp = sotoinpcb(so); + + /* Ignore keepalive packets */ + if ((len == 1) && (*(unsigned char *)data == 0xff)) { + return 1; + } + + /* + * Check that the payload is long enough to hold + * an ESP header and compute the length of encapsulation + * header to remove + */ + if (inp->inp_flags & INP_ESPINUDP) { + u_int32_t *st = (u_int32_t *)data; + + if ((len <= sizeof(struct esp)) || (*st == 0)) + return 0; /* Normal UDP processing */ + + skip = sizeof(struct udphdr); + } + + if (inp->inp_flags & INP_ESPINUDP_NON_IKE) { + u_int32_t *st = (u_int32_t *)data; + + if ((len <= sizeof(u_int64_t) + sizeof(struct esp)) + || ((st[0] | st[1]) != 0)) + return 0; /* Normal UDP processing */ + + skip = sizeof(struct udphdr) + sizeof(u_int64_t); + } + + /* + * Get the UDP ports. They are handled in network + * order everywhere in IPSEC_NAT_T code. + */ + udphdr = (struct udphdr *)((char *)data - skip); + sport = udphdr->uh_sport; + dport = udphdr->uh_dport; + + /* + * Remove the UDP header (and possibly the non ESP marker) + * IP header lendth is iphdrlen + * Before: + * <--- off ---> + * +----+------+-----+ + * | IP | UDP | ESP | + * +----+------+-----+ + * <-skip-> + * After: + * +----+-----+ + * | IP | ESP | + * +----+-----+ + * <-skip-> + */ + iphdrlen = off - sizeof(struct udphdr); + memmove(mtod(m, char *) + skip, mtod(m, void *), iphdrlen); + m_adj(m, skip); + + ip = mtod(m, struct ip *); + ip->ip_len = htons(ntohs(ip->ip_len) - skip); + ip->ip_p = IPPROTO_ESP; + + /* + * Copy the mbuf to avoid multiple free, as both + * esp4_input (which we call) and udp_input (which + * called us) free the mbuf. + */ + if ((n = m_dup(m, 0, M_COPYALL, M_DONTWAIT)) == NULL) { + printf("udp4_espinudp: m_dup failed\n"); + return 0; + } + + /* + * Add a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember + * the source UDP port. This is required if we want + * to select the right SPD for multiple hosts behind + * same NAT + */ + if ((tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, + sizeof(sport) + sizeof(dport), M_DONTWAIT)) == NULL) { + printf("udp4_espinudp: m_tag_get failed\n"); + m_freem(n); + return 0; + } + ((u_int16_t *)(tag + 1))[0] = sport; + ((u_int16_t *)(tag + 1))[1] = dport; + m_tag_prepend(n, tag); + +#ifdef FAST_IPSEC + ipsec4_common_input(n, iphdrlen, IPPROTO_ESP); +#else + esp4_input(n, iphdrlen); +#endif + + /* We handled it, it shoudln't be handled by UDP */ + return 1; } #endif