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 1.103.2.9 retrieving revision 1.160.8.1 diff -u -p -r1.103.2.9 -r1.160.8.1 --- src/sys/netinet/udp_usrreq.c 2005/04/01 14:31:50 1.103.2.9 +++ src/sys/netinet/udp_usrreq.c 2007/11/06 23:33:54 1.160.8.1 @@ -1,4 +1,4 @@ -/* $NetBSD: udp_usrreq.c,v 1.103.2.9 2005/04/01 14:31:50 skrll Exp $ */ +/* $NetBSD: udp_usrreq.c,v 1.160.8.1 2007/11/06 23:33:54 matt Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -61,7 +61,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.103.2.9 2005/04/01 14:31:50 skrll Exp $"); +__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.160.8.1 2007/11/06 23:33:54 matt Exp $"); #include "opt_inet.h" #include "opt_ipsec.h" @@ -95,17 +95,13 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #include #include -#ifdef IPSEC_NAT_T -#include -#include -#endif - #ifdef INET6 #include #include #include #include #include +#include #endif #ifndef INET6 @@ -123,6 +119,7 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #ifdef FAST_IPSEC #include #include /* XXX ipsecstat namespace */ +#include #ifdef INET6 #include #endif @@ -130,6 +127,7 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #ifdef IPSEC #include +#include #include #endif /*IPSEC*/ @@ -141,11 +139,7 @@ __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; @@ -153,13 +147,13 @@ struct udpstat udpstat; #ifdef INET #ifdef IPSEC_NAT_T -static int udp4_espinudp (struct mbuf *, int, struct sockaddr *, +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 @@ -179,14 +173,15 @@ 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, @@ -196,12 +191,29 @@ struct evcnt udp_hwcsum_data = EVCNT_INI struct evcnt udp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "udp", "swcsum"); -#define UDP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ - 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 @@ -372,18 +384,13 @@ udp_input(struct mbuf *m, ...) 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) { + udpstat.udps_hdrops++; + return; + } #ifdef INET6 if (IN_MULTICAST(ip->ip_dst.s_addr) || n == 0) { struct sockaddr_in6 src6, dst6; @@ -443,6 +450,11 @@ 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; } @@ -450,9 +462,34 @@ udp6_input_checksum(struct mbuf *m, cons udp6stat.udp6s_nosum++; goto bad; } - if (in6_cksum(m, IPPROTO_UDP, off, len) != 0) { + + 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); udp6stat.udp6s_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) { + udp6stat.udp6s_badsum++; + goto bad; + } } good: @@ -524,19 +561,16 @@ udp6_input(struct mbuf **mp, int *offp, /* * 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) { @@ -583,14 +617,14 @@ udp4_sendup(struct mbuf *m, int off /* o /* 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) + 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 *); @@ -630,14 +664,14 @@ udp6_sendup(struct mbuf *m, int off /* o /* 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) + 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 *); @@ -660,13 +694,14 @@ 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 */ @@ -754,12 +789,26 @@ udp4_realinput(struct sockaddr_in *src, if (inp->inp_flags & INP_ESPINUDP_ALL) { struct sockaddr *sa = (struct sockaddr *)src; - if (udp4_espinudp(m, off, sa, inp->inp_socket) != 0) { + 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; - /* Normal UDP processing will take place */ + case 0: /* plain UDP */ + default: /* Unexpected */ + /* + * Normal UDP processing will take place + * m may have changed. + */ + m = *mp; + break; + } } #endif @@ -779,7 +828,7 @@ 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; @@ -792,13 +841,18 @@ 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 @@ -831,10 +885,11 @@ 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; } @@ -868,11 +923,11 @@ 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); + in6p = in6_pcblookup_bind(&udbtable, dst6, dport, 0); if (in6p == 0) return rcvcnt; } @@ -900,7 +955,7 @@ 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; @@ -920,23 +975,20 @@ 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(op, so, level, optname, mp) - int op; - struct socket *so; - int level, optname; - struct mbuf **mp; +udp_ctloutput(int op, struct socket *so, int level, int optname, + struct mbuf **mp) { int s; int error = 0; @@ -967,7 +1019,6 @@ udp_ctloutput(op, so, level, optname, mp default: error = EAFNOSUPPORT; goto end; - break; } @@ -980,7 +1031,7 @@ udp_ctloutput(op, so, level, optname, mp case UDP_ENCAP: if (m == NULL || m->m_len < sizeof (int)) { error = EINVAL; - goto end; + break; } switch(*mtod(m, int *)) { @@ -1001,21 +1052,21 @@ udp_ctloutput(op, so, level, optname, mp #endif default: error = EINVAL; - goto end; break; } break; default: error = ENOPROTOOPT; - goto end; break; } + if (m != NULL) { + m_free(m); + } break; default: error = EINVAL; - goto end; break; } @@ -1080,18 +1131,11 @@ udp_output(struct mbuf *m, ...) /* * XXX Cache pseudo-header checksum part for * XXX "connected" UDP sockets. - * Maybe skip checksums on loopback interfaces. */ ui->ui_sum = in_cksum_phdr(ui->ui_src.s_addr, ui->ui_dst.s_addr, htons((u_int16_t)len + sizeof(struct udphdr) + IPPROTO_UDP)); - if (__predict_true(ro->ro_rt == NULL || - !(ro->ro_rt->rt_ifp->if_flags & - IFF_LOOPBACK) || - udp_do_loopback_cksum)) - m->m_pkthdr.csum_flags = M_CSUM_UDPv4; - else - m->m_pkthdr.csum_flags = 0; + m->m_pkthdr.csum_flags = M_CSUM_UDPv4; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); } else ui->ui_sum = 0; @@ -1119,23 +1163,23 @@ udp_usrreq(struct socket *so, int req, s struct mbuf *control, struct lwp *l) { struct inpcb *inp; - struct proc *p; int s; int error = 0; - p = l ? l->l_proc : NULL; 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) { in_pcbpurgeif0(&udbtable, (struct ifnet *)control); in_purgeif((struct ifnet *)control); in_pcbpurgeif(&udbtable, (struct ifnet *)control); + splx(s); return (0); } - s = splsoftnet(); inp = sotoinpcb(so); #ifdef DIAGNOSTIC if (req != PRU_SEND && req != PRU_SENDOOB && control) @@ -1179,7 +1223,7 @@ 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: @@ -1187,7 +1231,7 @@ 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); @@ -1229,7 +1273,7 @@ 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 { @@ -1343,6 +1387,13 @@ SYSCTL_SETUP(sysctl_net_inet_udp_setup, 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"), + NULL, 0, &udpstat, sizeof(udpstat), + CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_STATS, + CTL_EOL); } #endif @@ -1351,22 +1402,24 @@ SYSCTL_SETUP(sysctl_net_inet_udp_setup, * 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(m, off, src, so) - struct mbuf *m; - int off; - struct sockaddr *src; - struct socket *so; +udp4_espinudp(struct mbuf **mp, int off, struct sockaddr *src, + struct socket *so) { size_t len; - caddr_t data; + 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 @@ -1377,18 +1430,19 @@ udp4_espinudp(m, off, src, so) minlen = m->m_pkthdr.len; if (m->m_len < minlen) { - if ((m = m_pullup(m, minlen)) == NULL) { + if ((*mp = m_pullup(m, minlen)) == NULL) { printf("udp4_espinudp: m_pullup failed\n"); - return 0; + return -1; } + m = *mp; } len = m->m_len - off; - data = mtod(m, caddr_t) + off; + data = mtod(m, char *) + off; inp = sotoinpcb(so); /* Ignore keepalive packets */ - if ((len == 1) && (data[0] == '\xff')) { + if ((len == 1) && (*(unsigned char *)data == 0xff)) { return 1; } @@ -1407,16 +1461,24 @@ udp4_espinudp(m, off, src, so) } if (inp->inp_flags & INP_ESPINUDP_NON_IKE) { - u_int64_t *st = (u_int64_t *)data; + u_int32_t *st = (u_int32_t *)data; if ((len <= sizeof(u_int64_t) + sizeof(struct esp)) - || (*st != 0)) + || ((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: @@ -1432,7 +1494,7 @@ udp4_espinudp(m, off, src, so) * <-skip-> */ iphdrlen = off - sizeof(struct udphdr); - memmove(mtod(m, caddr_t) + skip, mtod(m, caddr_t), iphdrlen); + memmove(mtod(m, char *) + skip, mtod(m, void *), iphdrlen); m_adj(m, skip); ip = mtod(m, struct ip *); @@ -1449,7 +1511,27 @@ udp4_espinudp(m, off, src, so) 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;