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 1.126 retrieving revision 1.140 diff -u -p -r1.126 -r1.140 --- src/sys/netinet/udp_usrreq.c 2004/12/17 22:54:52 1.126 +++ src/sys/netinet/udp_usrreq.c 2005/08/10 13:05:17 1.140 @@ -1,4 +1,4 @@ -/* $NetBSD: udp_usrreq.c,v 1.126 2004/12/17 22:54:52 christos Exp $ */ +/* $NetBSD: udp_usrreq.c,v 1.140 2005/08/10 13:05:17 yamt Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -61,7 +61,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.126 2004/12/17 22:54:52 christos Exp $"); +__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.140 2005/08/10 13:05:17 yamt Exp $"); #include "opt_inet.h" #include "opt_ipsec.h" @@ -95,6 +95,11 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #include #include +#ifdef IPSEC_NAT_T +#include +#include +#endif + #ifdef INET6 #include #include @@ -141,22 +146,27 @@ int udpcksum = 1; #else int udpcksum = 0; /* XXX */ #endif -int udp_do_loopback_cksum = 1; struct inpcbtable udbtable; struct udpstat udpstat; #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); +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); @@ -176,6 +186,7 @@ struct mowner udp_tx_mowner = { "udp", " #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, @@ -185,12 +196,29 @@ 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 @@ -209,7 +237,102 @@ udp_init(void) MOWNER_ATTACH(&udp_mowner); } +/* + * 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: + udpstat.udps_badsum++; + return -1; +} + void udp_input(struct mbuf *m, ...) { @@ -262,46 +385,8 @@ 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. 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; - } - } + if (udp4_input_checksum(m, uh, iphlen, len)) + goto badcsum; /* construct source and dst sockaddrs. */ bzero(&src, sizeof(src)); @@ -367,11 +452,62 @@ 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) { + udp6stat.udp6s_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); + 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: + return 0; +bad: + return -1; +} + int udp6_input(struct mbuf **mp, int *offp, int proto) { @@ -430,18 +566,8 @@ udp6_input(struct mbuf **mp, int *offp, * Checksum extended UDP header and data. Maybe skip checksum * on loopback interfaces. */ - if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & - IFF_LOOPBACK) || - udp_do_loopback_cksum)) { - if (uh->uh_sum == 0) { - udp6stat.udp6s_nosum++; - goto bad; - } - if (in6_cksum(m, IPPROTO_UDP, off, ulen) != 0) { - udp6stat.udp6s_badsum++; - goto bad; - } - } + if (udp6_input_checksum(m, uh, off, ulen)) + goto bad; /* * Construct source and dst sockaddrs. @@ -670,6 +796,20 @@ udp4_realinput(struct sockaddr_in *src, return rcvcnt; } +#ifdef IPSEC_NAT_T + /* Handle ESP over UDP */ + if (inp->inp_flags & INP_ESPINUDP_ALL) { + struct sockaddr *sa = (struct sockaddr *)src; + + if (udp4_espinudp(m, off, sa, inp->inp_socket) != 0) { + rcvcnt++; + goto bad; + } + + /* Normal UDP processing will take place */ + } +#endif + udp4_sendup(m, off, (struct sockaddr *)src, inp->inp_socket); rcvcnt++; } @@ -839,6 +979,100 @@ udp_ctlinput(int cmd, struct sockaddr *s } int +udp_ctloutput(op, so, level, optname, mp) + int op; + struct socket *so; + int level, optname; + struct mbuf **mp; +{ + int s; + int error = 0; + struct mbuf *m; + struct inpcb *inp; + int family; + + family = so->so_proto->pr_domain->dom_family; + + s = splsoftnet(); + switch (family) { +#ifdef INET + case PF_INET: + if (level != IPPROTO_UDP) { + error = ip_ctloutput(op, so, level, optname, mp); + goto end; + } + break; +#endif +#ifdef INET6 + case PF_INET6: + if (level != IPPROTO_UDP) { + error = ip6_ctloutput(op, so, level, optname, mp); + goto end; + } + break; +#endif + default: + error = EAFNOSUPPORT; + goto end; + break; + } + + + switch (op) { + case PRCO_SETOPT: + m = *mp; + inp = sotoinpcb(so); + + switch (optname) { + case UDP_ENCAP: + if (m == NULL || m->m_len < sizeof (int)) { + error = EINVAL; + goto end; + } + + switch(*mtod(m, int *)) { +#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; + goto end; + break; + } + break; + + default: + error = ENOPROTOOPT; + goto end; + break; + } + break; + + default: + error = EINVAL; + goto end; + break; + } + +end: + splx(s); + return error; +} + + +int udp_output(struct mbuf *m, ...) { struct inpcb *inp; @@ -893,18 +1127,11 @@ 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; @@ -1147,5 +1374,156 @@ SYSCTL_SETUP(sysctl_net_inet_udp_setup, 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"), + NULL, 0, &udpstat, sizeof(udpstat), + CTL_NET, PF_INET, IPPROTO_UDP, UDPCTL_STATS, + CTL_EOL); +} +#endif + +#if (defined INET && defined IPSEC_NAT_T) +/* + * Returns: + * 1 if the packet was processed + * 0 if normal UDP processing should take place + */ +static int +udp4_espinudp(m, off, src, so) + struct mbuf *m; + int off; + struct sockaddr *src; + struct socket *so; +{ + size_t len; + caddr_t 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; + + /* + * 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 ((m = m_pullup(m, minlen)) == NULL) { + printf("udp4_espinudp: m_pullup failed\n"); + return 0; + } + } + + len = m->m_len - off; + data = mtod(m, caddr_t) + off; + inp = sotoinpcb(so); + + /* Ignore keepalive packets */ + if ((len == 1) && (data[0] == '\xff')) { + 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_int64_t *st = (u_int64_t *)data; + + if ((len <= sizeof(u_int64_t) + sizeof(struct esp)) + || (*st != 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 *)(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, caddr_t) + skip, mtod(m, caddr_t), 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); + + esp4_input(n, iphdrlen); + + /* We handled it, it shoudln't be handled by UDP */ + return 1; } #endif