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 1.171 retrieving revision 1.190.2.4 diff -u -p -r1.171 -r1.190.2.4 --- src/sys/netinet/udp_usrreq.c 2008/04/26 08:13:59 1.171 +++ src/sys/netinet/udp_usrreq.c 2013/10/17 23:52:18 1.190.2.4 @@ -1,4 +1,4 @@ -/* $NetBSD: udp_usrreq.c,v 1.171 2008/04/26 08:13:59 yamt Exp $ */ +/* $NetBSD: udp_usrreq.c,v 1.190.2.4 2013/10/17 23:52:18 rmind Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -60,25 +60,28 @@ * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 */ +/* + * UDP protocol implementation. + * Per RFC 768, August, 1980. + */ + #include -__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.171 2008/04/26 08:13:59 yamt Exp $"); +__KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c,v 1.190.2.4 2013/10/17 23:52:18 rmind Exp $"); #include "opt_inet.h" +#include "opt_compat_netbsd.h" #include "opt_ipsec.h" #include "opt_inet_csum.h" #include "opt_ipkdb.h" #include "opt_mbuftrace.h" #include -#include #include #include #include #include -#include -#include #include -#include +#include #include #include @@ -104,7 +107,6 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #include #include #include -#include #endif #ifndef INET6 @@ -112,14 +114,7 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #include #endif -#include "faith.h" -#if defined(NFAITH) && NFAITH > 0 -#include -#endif - -#include - -#ifdef FAST_IPSEC +#ifdef IPSEC #include #include #include @@ -127,32 +122,24 @@ __KERNEL_RCSID(0, "$NetBSD: udp_usrreq.c #ifdef INET6 #include #endif -#endif /* FAST_IPSEC */ +#endif /* IPSEC */ -#ifdef IPSEC -#include -#include -#include -#include -#endif /* IPSEC */ +#ifdef COMPAT_50 +#include +#endif #ifdef IPKDB #include #endif -/* - * UDP protocol implementation. - * Per RFC 768, August, 1980. - */ int udpcksum = 1; int udp_do_loopback_cksum = 0; -struct inpcbtable udbtable; - -percpu_t *udpstat_percpu; +inpcbtable_t * udbtable __read_mostly; +percpu_t * udpstat_percpu; #ifdef INET -#ifdef IPSEC_NAT_T +#ifdef IPSEC static int udp4_espinudp (struct mbuf **, int, struct sockaddr *, struct socket *); #endif @@ -162,22 +149,15 @@ static int udp4_realinput (struct sockad 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); -#endif #ifndef UDBHASHSIZE #define UDBHASHSIZE 128 #endif int udbhashsize = UDBHASHSIZE; +static int udp_sendspace = 9216; +static int udp_recvspace = 40 * (1024 + sizeof(struct sockaddr_in)); + #ifdef MBUFTRACE struct mowner udp_mowner = MOWNER_INIT("udp", ""); struct mowner udp_rx_mowner = MOWNER_INIT("udp", "rx"); @@ -203,35 +183,18 @@ 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 - #define UDP_CSUM_COUNTER_INCR(ev) /* nothing */ - #endif /* UDP_CSUM_COUNTERS */ +static void sysctl_net_inet_udp_setup(struct sysctllog **); + void udp_init(void) { - - in_pcbinit(&udbtable, udbhashsize, udbhashsize); + udbtable = inpcb_init(udbhashsize, udbhashsize, 0); + sysctl_net_inet_udp_setup(NULL); MOWNER_ATTACH(&udp_tx_mowner); MOWNER_ATTACH(&udp_rx_mowner); @@ -240,9 +203,6 @@ udp_init(void) #ifdef INET udpstat_percpu = percpu_alloc(sizeof(uint64_t) * UDP_NSTATS); #endif -#ifdef INET6 - udp6stat_percpu = percpu_alloc(sizeof(uint64_t) * UDP6_NSTATS); -#endif } /* @@ -253,7 +213,6 @@ int udp_input_checksum(int af, struct mbuf *m, const struct udphdr *uh, int iphlen, int len) { - switch (af) { #ifdef INET case AF_INET: @@ -263,11 +222,9 @@ udp_input_checksum(int af, struct mbuf * case AF_INET6: return udp6_input_checksum(m, uh, iphlen, len); #endif + default: + KASSERT(false); } -#ifdef DIAGNOSTIC - panic("udp_input_checksum: unknown af %d", af); -#endif - /* NOTREACHED */ return -1; } @@ -348,10 +305,8 @@ udp_input(struct mbuf *m, ...) struct sockaddr_in src, dst; struct ip *ip; struct udphdr *uh; - int iphlen; - int len; - int n; - u_int16_t ip_len; + int iphlen, len, n; + uint16_t ip_len; va_start(ap, m); iphlen = va_arg(ap, int); @@ -372,7 +327,7 @@ udp_input(struct mbuf *m, ...) } KASSERT(UDP_HDR_ALIGNED_P(uh)); - /* destination port of 0 is illegal, based on RFC768. */ + /* Destination port of 0 is illegal, based on RFC 768. */ if (uh->uh_dport == 0) goto bad; @@ -381,7 +336,7 @@ udp_input(struct mbuf *m, ...) * If not enough data to reflect UDP length, drop. */ ip_len = ntohs(ip->ip_len); - len = ntohs((u_int16_t)uh->uh_ulen); + len = ntohs((uint16_t)uh->uh_ulen); if (ip_len != iphlen + len) { if (ip_len < iphlen + len || len < sizeof(struct udphdr)) { UDP_STATINC(UDP_STAT_BADLEN); @@ -394,7 +349,7 @@ udp_input(struct mbuf *m, ...) * Checksum extended UDP header and data. */ if (udp4_input_checksum(m, uh, iphlen, len)) - goto badcsum; + goto bad; /* construct source and dst sockaddrs. */ sockaddr_in_init(&src, &ip->ip_src, uh->uh_sport); @@ -404,22 +359,30 @@ udp_input(struct mbuf *m, ...) UDP_STATINC(UDP_STAT_HDROPS); return; } + if (m == NULL) { + /* + * packet has been processed by ESP stuff - + * e.g. dropped NAT-T-keep-alive-packet ... + */ + return; + } + ip = mtod(m, struct ip *); #ifdef INET6 if (IN_MULTICAST(ip->ip_dst.s_addr) || n == 0) { struct sockaddr_in6 src6, dst6; - bzero(&src6, sizeof(src6)); + memset(&src6, 0, sizeof(src6)); src6.sin6_family = AF_INET6; src6.sin6_len = sizeof(struct sockaddr_in6); src6.sin6_addr.s6_addr[10] = src6.sin6_addr.s6_addr[11] = 0xff; - bcopy(&ip->ip_src, &src6.sin6_addr.s6_addr[12], + memcpy(&src6.sin6_addr.s6_addr[12], &ip->ip_src, sizeof(ip->ip_src)); src6.sin6_port = uh->uh_sport; - bzero(&dst6, sizeof(dst6)); + memset(&dst6, 0, sizeof(dst6)); dst6.sin6_family = AF_INET6; dst6.sin6_len = sizeof(struct sockaddr_in6); dst6.sin6_addr.s6_addr[10] = dst6.sin6_addr.s6_addr[11] = 0xff; - bcopy(&ip->ip_dst, &dst6.sin6_addr.s6_addr[12], + memcpy(&dst6.sin6_addr.s6_addr[12], &ip->ip_dst, sizeof(ip->ip_dst)); dst6.sin6_port = uh->uh_dport; @@ -447,170 +410,20 @@ udp_input(struct mbuf *m, ...) icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); m = NULL; } - -bad: - if (m) - m_freem(m); - return; - -badcsum: - m_freem(m); -} -#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) -{ - struct mbuf *m = *mp; - int off = *offp; - struct sockaddr_in6 src, dst; - struct ip6_hdr *ip6; - struct udphdr *uh; - u_int32_t plen, ulen; - - ip6 = mtod(m, struct ip6_hdr *); - -#if defined(NFAITH) && 0 < NFAITH - if (faithprefix(&ip6->ip6_dst)) { - /* send icmp6 host unreach? */ + if (m) { m_freem(m); - return IPPROTO_DONE; - } -#endif - - 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) { - IP6_STATINC(IP6_STAT_TOOSHORT); - return IPPROTO_DONE; } - KASSERT(UDP_HDR_ALIGNED_P(uh)); - ulen = ntohs((u_short)uh->uh_ulen); - /* - * RFC2675 section 4: jumbograms will have 0 in the UDP header field, - * iff payload length > 0xffff. - */ - if (ulen == 0 && plen > 0xffff) - ulen = plen; - - if (plen != ulen) { - UDP6_STATINC(UDP6_STAT_BADLEN); - goto bad; - } - - /* destination port of 0 is illegal, based on RFC768. */ - if (uh->uh_dport == 0) - goto bad; - - /* Be proactive about malicious use of IPv4 mapped address */ - if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || - IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { - /* XXX stat */ - goto bad; - } - - /* - * Checksum extended UDP header and data. Maybe skip checksum - * on loopback interfaces. - */ - if (udp6_input_checksum(m, uh, off, ulen)) - goto bad; - - /* - * Construct source and dst sockaddrs. - */ - bzero(&src, sizeof(src)); - src.sin6_family = AF_INET6; - src.sin6_len = sizeof(struct sockaddr_in6); - 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); - 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) { - UDP6_STATINC(UDP6_STAT_NOPORTMCAST); - goto bad; - } - UDP6_STATINC(UDP6_STAT_NOPORT); - icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0); - m = NULL; - } - -bad: - if (m) - m_freem(m); - return IPPROTO_DONE; + return; } -#endif -#ifdef INET static void udp4_sendup(struct mbuf *m, int off /* offset of data portion */, struct sockaddr *src, struct socket *so) { struct mbuf *opts = NULL; struct mbuf *n; - struct inpcb *inp = NULL; + inpcb_t *inp = NULL; if (!so) return; @@ -626,7 +439,7 @@ udp4_sendup(struct mbuf *m, int off /* o return; } -#if defined(IPSEC) || defined(FAST_IPSEC) +#if defined(IPSEC) /* check AH/ESP integrity. */ if (so != NULL && ipsec4_in_reject_so(m, so)) { IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); @@ -638,7 +451,10 @@ udp4_sendup(struct mbuf *m, int off /* o #endif /*IPSEC*/ if ((n = m_copypacket(m, M_DONTWAIT)) != NULL) { - if (inp && (inp->inp_flags & INP_CONTROLOPTS + if (inp && ((inpcb_get_flags(inp) & INP_CONTROLOPTS) != 0 +#ifdef SO_OTIMESTAMP + || so->so_options & SO_OTIMESTAMP +#endif || so->so_options & SO_TIMESTAMP)) { struct ip *ip = mtod(n, struct ip *); ip_savecontrol(inp, &opts, ip, n); @@ -656,65 +472,68 @@ udp4_sendup(struct mbuf *m, int off /* o sorwakeup(so); } } -#endif -#ifdef INET6 -static void -udp6_sendup(struct mbuf *m, int off /* offset of data portion */, - struct sockaddr *src, struct socket *so) -{ - struct mbuf *opts = NULL; - struct mbuf *n; - struct in6pcb *in6p = NULL; +struct udp_pcb_ctx { + struct mbuf * mbuf; + struct sockaddr_in * src; + struct sockaddr_in * dst; + int off; + int rcvcnt; +}; - if (!so) - return; - if (so->so_proto->pr_domain->dom_family != AF_INET6) - return; - in6p = sotoin6pcb(so); +static int +udp4_pcb_process(inpcb_t *inp, void *arg) +{ + struct udp_pcb_ctx *uctx = arg; + struct in_addr dst4 = uctx->dst->sin_addr; + in_port_t dport = uctx->dst->sin_port; + struct in_addr laddr, faddr; + in_port_t lport, fport; + struct socket *so; -#if defined(IPSEC) || defined(FAST_IPSEC) - /* check AH/ESP integrity. */ - if (so != NULL && ipsec6_in_reject_so(m, so)) { - 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; + inpcb_get_ports(inp, &lport, &fport); + if (lport != dport) { + return 0; } -#endif /*IPSEC*/ + inpcb_get_addrs(inp, &laddr, &faddr); + if (!in_nullhost(laddr) && !in_hosteq(laddr, dst4)) { + return 0; + } + if (!in_nullhost(faddr)) { + struct in_addr src4 = uctx->src->sin_addr; + in_port_t sport = uctx->src->sin_port; - 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 *); - ip6_savecontrol(in6p, &opts, ip6, n); + if (!in_hosteq(faddr, src4) || fport != sport) { + return 0; } + } - m_adj(n, off); - if (sbappendaddr(&so->so_rcv, src, n, opts) == 0) { - m_freem(n); - if (opts) - m_freem(opts); - so->so_rcv.sb_overflowed++; - UDP6_STATINC(UDP6_STAT_FULLSOCK); - } else - sorwakeup(so); + so = inpcb_get_socket(inp); + udp4_sendup(uctx->mbuf, uctx->off, (struct sockaddr *)uctx->src, so); + uctx->rcvcnt++; + + /* + * Do not look for additional matches if this one does not have + * either the SO_REUSEPORT or SO_REUSEADDR socket options set. + * This heuristic avoids searching through all PCBs in the common + * case of a non-shared port. It assumes that an application will + * never clear these options after setting them. + */ + if ((so->so_options & (SO_REUSEPORT|SO_REUSEADDR)) == 0) { + return EJUSTRETURN; } + return 0; } -#endif -#ifdef INET static int udp4_realinput(struct sockaddr_in *src, struct sockaddr_in *dst, - struct mbuf **mp, int off /* offset of udphdr */) + struct mbuf **mp, int off /* offset of udphdr */) { - u_int16_t *sport, *dport; - int rcvcnt; + in_port_t *sport, *dport; struct in_addr *src4, *dst4; - struct inpcb_hdr *inph; - struct inpcb *inp; + inpcb_t *inp; struct mbuf *m = *mp; + int rcvcnt; rcvcnt = 0; off += sizeof(struct udphdr); /* now, offset of payload */ @@ -729,80 +548,43 @@ udp4_realinput(struct sockaddr_in *src, if (IN_MULTICAST(dst4->s_addr) || in_broadcast(*dst4, m->m_pkthdr.rcvif)) { + struct udp_pcb_ctx uctx = { + .mbuf = m, .src = src, .dst = dst, + .off = off, .rcvcnt = 0 + }; + int error; + /* * Deliver a multicast or broadcast datagram to *all* sockets * for which the local and remote addresses and ports match * those of the incoming datagram. This allows more than * one process to receive multi/broadcasts on the same port. - * (This really ought to be done for unicast datagrams as - * well, but that would cause problems with existing - * applications that open both address-specific sockets and - * a wildcard socket listening to the same port -- they would - * end up receiving duplicates of every unicast datagram. - * Those applications open the multiple sockets to overcome an - * inadequacy of the UDP socket interface, but for backwards - * compatibility we avoid the problem here rather than - * fixing the interface. Maybe 4.5BSD will remedy this?) - */ - - /* - * KAME note: traditionally we dropped udpiphdr from mbuf here. - * we need udpiphdr for IPsec processing so we do that later. */ - /* - * Locate pcb(s) for datagram. - */ - CIRCLEQ_FOREACH(inph, &udbtable.inpt_queue, inph_queue) { - inp = (struct inpcb *)inph; - if (inp->inp_af != AF_INET) - continue; - - if (inp->inp_lport != *dport) - continue; - if (!in_nullhost(inp->inp_laddr)) { - if (!in_hosteq(inp->inp_laddr, *dst4)) - continue; - } - if (!in_nullhost(inp->inp_faddr)) { - if (!in_hosteq(inp->inp_faddr, *src4) || - inp->inp_fport != *sport) - continue; - } - - udp4_sendup(m, off, (struct sockaddr *)src, - inp->inp_socket); - rcvcnt++; - - /* - * Don't look for additional matches if this one does - * not have either the SO_REUSEPORT or SO_REUSEADDR - * socket options set. This heuristic avoids searching - * through all pcbs in the common case of a non-shared - * port. It assumes that an application will never - * clear these options after setting them. - */ - if ((inp->inp_socket->so_options & - (SO_REUSEPORT|SO_REUSEADDR)) == 0) - break; - } + error = inpcb_foreach(udbtable, AF_INET, + udp4_pcb_process, &uctx); + KASSERT(error == 0 || error == EJUSTRETURN); + rcvcnt = uctx.rcvcnt; } else { /* - * Locate pcb for datagram. + * Locate PCB for datagram. */ - inp = in_pcblookup_connect(&udbtable, *src4, *sport, *dst4, *dport); - if (inp == 0) { + struct socket *so; + + inp = inpcb_lookup(udbtable, *src4, *sport, *dst4, *dport, NULL); + if (inp == NULL) { UDP_STATINC(UDP_STAT_PCBHASHMISS); - inp = in_pcblookup_bind(&udbtable, *dst4, *dport); - if (inp == 0) + inp = inpcb_lookup_bound(udbtable, *dst4, *dport); + if (inp == NULL) return rcvcnt; } + so = inpcb_get_socket(inp); -#ifdef IPSEC_NAT_T +#ifdef IPSEC /* Handle ESP over UDP */ - if (inp->inp_flags & INP_ESPINUDP_ALL) { + if (inpcb_get_flags(inp) & INP_ESPINUDP_ALL) { struct sockaddr *sa = (struct sockaddr *)src; - switch(udp4_espinudp(mp, off, sa, inp->inp_socket)) { + switch (udp4_espinudp(mp, off, sa, so)) { case -1: /* Error, m was freeed */ rcvcnt = -1; goto bad; @@ -825,127 +607,13 @@ udp4_realinput(struct sockaddr_in *src, } #endif - udp4_sendup(m, off, (struct sockaddr *)src, inp->inp_socket); - rcvcnt++; - } - -bad: - return rcvcnt; -} -#endif - -#ifdef INET6 -static int -udp6_realinput(int af, struct sockaddr_in6 *src, struct sockaddr_in6 *dst, - struct mbuf *m, int off) -{ - u_int16_t sport, dport; - int rcvcnt; - struct in6_addr src6, *dst6; - const struct in_addr *dst4; - struct inpcb_hdr *inph; - struct in6pcb *in6p; - - rcvcnt = 0; - off += sizeof(struct udphdr); /* now, offset of payload */ - - if (af != AF_INET && af != AF_INET6) - goto bad; - if (src->sin6_family != AF_INET6 || dst->sin6_family != AF_INET6) - goto bad; - - src6 = src->sin6_addr; - if (sa6_recoverscope(src) != 0) { - /* XXX: should be impossible. */ - goto bad; - } - sport = src->sin6_port; - - dport = dst->sin6_port; - dst4 = (struct in_addr *)&dst->sin6_addr.s6_addr[12]; - dst6 = &dst->sin6_addr; - - if (IN6_IS_ADDR_MULTICAST(dst6) || - (af == AF_INET && IN_MULTICAST(dst4->s_addr))) { - /* - * Deliver a multicast or broadcast datagram to *all* sockets - * for which the local and remote addresses and ports match - * those of the incoming datagram. This allows more than - * one process to receive multi/broadcasts on the same port. - * (This really ought to be done for unicast datagrams as - * well, but that would cause problems with existing - * applications that open both address-specific sockets and - * a wildcard socket listening to the same port -- they would - * end up receiving duplicates of every unicast datagram. - * Those applications open the multiple sockets to overcome an - * inadequacy of the UDP socket interface, but for backwards - * compatibility we avoid the problem here rather than - * fixing the interface. Maybe 4.5BSD will remedy this?) - */ - - /* - * KAME note: traditionally we dropped udpiphdr from mbuf here. - * we need udpiphdr for IPsec processing so we do that later. - */ - /* - * Locate pcb(s) for datagram. - */ - CIRCLEQ_FOREACH(inph, &udbtable.inpt_queue, inph_queue) { - in6p = (struct in6pcb *)inph; - if (in6p->in6p_af != AF_INET6) - continue; - - if (in6p->in6p_lport != dport) - continue; - if (!IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_laddr)) { - if (!IN6_ARE_ADDR_EQUAL(&in6p->in6p_laddr, - dst6)) - continue; - } else { - if (IN6_IS_ADDR_V4MAPPED(dst6) && - (in6p->in6p_flags & IN6P_IPV6_V6ONLY)) - continue; - } - if (!IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr)) { - if (!IN6_ARE_ADDR_EQUAL(&in6p->in6p_faddr, - &src6) || in6p->in6p_fport != sport) - continue; - } else { - if (IN6_IS_ADDR_V4MAPPED(&src6) && - (in6p->in6p_flags & IN6P_IPV6_V6ONLY)) - continue; - } - - udp6_sendup(m, off, (struct sockaddr *)src, - in6p->in6p_socket); - rcvcnt++; - - /* - * Don't look for additional matches if this one does - * not have either the SO_REUSEPORT or SO_REUSEADDR - * socket options set. This heuristic avoids searching - * through all pcbs in the common case of a non-shared - * port. It assumes that an application will never - * clear these options after setting them. - */ - if ((in6p->in6p_socket->so_options & - (SO_REUSEPORT|SO_REUSEADDR)) == 0) - break; - } - } else { /* - * Locate pcb for datagram. + * Check the minimum TTL for socket. */ - in6p = in6_pcblookup_connect(&udbtable, &src6, sport, dst6, - dport, 0); - if (in6p == 0) { - UDP_STATINC(UDP_STAT_PCBHASHMISS); - in6p = in6_pcblookup_bind(&udbtable, dst6, dport, 0); - if (in6p == 0) - return rcvcnt; + if (mtod(m, struct ip *)->ip_ttl < inpcb_get_minttl(inp)) { + goto bad; } - - udp6_sendup(m, off, (struct sockaddr *)src, in6p->in6p_socket); + udp4_sendup(m, off, (struct sockaddr *)src, so); rcvcnt++; } @@ -956,15 +624,17 @@ bad: #ifdef INET /* - * Notify a udp user of an asynchronous error; + * Notify a UDP user of an asynchronous error; * just wake up so that he can collect error status. */ static void -udp_notify(struct inpcb *inp, int errno) +udp_notify(inpcb_t *inp, int errno) { - inp->inp_socket->so_error = errno; - sorwakeup(inp->inp_socket); - sowwakeup(inp->inp_socket); + struct socket *so = inpcb_get_socket(inp); + + so->so_error = errno; + sorwakeup(so); + sowwakeup(so); } void * @@ -972,110 +642,97 @@ udp_ctlinput(int cmd, const struct socka { struct ip *ip = v; struct udphdr *uh; - void (*notify)(struct inpcb *, int) = udp_notify; int errno; + bool rdr; - if (sa->sa_family != AF_INET - || sa->sa_len != sizeof(struct sockaddr_in)) + if (sa->sa_family != AF_INET || + sa->sa_len != sizeof(struct sockaddr_in)) return NULL; if ((unsigned)cmd >= PRC_NCMDS) return NULL; errno = inetctlerrmap[cmd]; - if (PRC_IS_REDIRECT(cmd)) - notify = in_rtchange, ip = 0; - else if (cmd == PRC_HOSTDEAD) - ip = 0; - else if (errno == 0) + + rdr = PRC_IS_REDIRECT(cmd); + if (rdr || cmd == PRC_HOSTDEAD || ip == NULL) { + inpcb_notifyall(udbtable, satocsin(sa)->sin_addr, + errno, rdr ? inpcb_rtchange : udp_notify); + return NULL; + } else if (errno == 0) { return NULL; - if (ip) { - 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, satocsin(sa)->sin_addr, errno, - notify); + /* Note: mapped address case */ + uh = (struct udphdr *)((char *)ip + (ip->ip_hl << 2)); + inpcb_notify(udbtable, satocsin(sa)->sin_addr, uh->uh_dport, + ip->ip_src, uh->uh_sport, errno, udp_notify); return NULL; } int -udp_ctloutput(int op, struct socket *so, int level, int optname, - struct mbuf **mp) +udp_ctloutput(int op, struct socket *so, struct sockopt *sopt) { - int s; - int error = 0; - struct mbuf *m; - struct inpcb *inp; - int family; + int family, optval, inpflags, error = 0; + inpcb_t *inp; + + KASSERT(solocked(so)); 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; + if (sopt->sopt_level != IPPROTO_UDP) { + return ip_ctloutput(op, so, sopt); } break; #endif #ifdef INET6 case PF_INET6: - if (level != IPPROTO_UDP) { - error = ip6_ctloutput(op, so, level, optname, mp); - goto end; + if (sopt->sopt_level != IPPROTO_UDP) { + return ip6_ctloutput(op, so, sopt); } break; #endif default: - error = EAFNOSUPPORT; - goto end; + return EAFNOSUPPORT; } - switch (op) { case PRCO_SETOPT: - m = *mp; inp = sotoinpcb(so); - switch (optname) { + switch (sopt->sopt_name) { case UDP_ENCAP: - if (m == NULL || m->m_len != sizeof(int)) { - error = EINVAL; + error = sockopt_getint(sopt, &optval); + if (error) break; - } - switch(*mtod(m, int *)) { -#ifdef IPSEC_NAT_T + inpflags = inpcb_get_flags(inp); + switch(optval) { case 0: - inp->inp_flags &= ~INP_ESPINUDP_ALL; + inpflags &= ~INP_ESPINUDP_ALL; break; case UDP_ENCAP_ESPINUDP: - inp->inp_flags &= ~INP_ESPINUDP_ALL; - inp->inp_flags |= INP_ESPINUDP; + inpflags &= ~INP_ESPINUDP_ALL; + inpflags |= INP_ESPINUDP; break; case UDP_ENCAP_ESPINUDP_NON_IKE: - inp->inp_flags &= ~INP_ESPINUDP_ALL; - inp->inp_flags |= INP_ESPINUDP_NON_IKE; + inpflags &= ~INP_ESPINUDP_ALL; + inpflags |= INP_ESPINUDP_NON_IKE; break; -#endif default: error = EINVAL; break; } + inpcb_set_flags(inp, inpflags); break; default: error = ENOPROTOOPT; break; } - if (m != NULL) { - m_free(m); - } break; default: @@ -1083,33 +740,27 @@ udp_ctloutput(int op, struct socket *so, break; } -end: - splx(s); return error; } - -int -udp_output(struct mbuf *m, ...) +static int +udp_output(struct mbuf *m, inpcb_t *inp) { - struct inpcb *inp; + struct socket *so; struct udpiphdr *ui; struct route *ro; int len = m->m_pkthdr.len; int error = 0; - va_list ap; MCLAIM(m, &udp_tx_mowner); - va_start(ap, m); - inp = va_arg(ap, struct inpcb *); - va_end(ap); + so = inpcb_get_socket(inp); + KASSERT(solocked(so)); /* - * Calculate data length and get a mbuf - * for UDP and IP headers. + * Calculate data length and get a mbuf for UDP and IP headers. */ M_PREPEND(m, sizeof(struct udpiphdr), M_DONTWAIT); - if (m == 0) { + if (m == NULL) { error = ENOBUFS; goto release; } @@ -1124,18 +775,13 @@ udp_output(struct mbuf *m, ...) } /* - * Fill in mbuf with extended UDP header - * and addresses and length put into network format. + * Fill in mbuf with extended UDP header. */ ui = mtod(m, struct udpiphdr *); ui->ui_pr = IPPROTO_UDP; - ui->ui_src = inp->inp_laddr; - ui->ui_dst = inp->inp_faddr; - ui->ui_sport = inp->inp_lport; - ui->ui_dport = inp->inp_fport; - ui->ui_ulen = htons((u_int16_t)len + sizeof(struct udphdr)); - - ro = &inp->inp_route; + inpcb_get_addrs(inp, &ui->ui_src, &ui->ui_dst); + inpcb_get_ports(inp, &ui->ui_sport, &ui->ui_dport); + ui->ui_ulen = htons((uint16_t)len + sizeof(struct udphdr)); /* * Set up checksum and output datagram. @@ -1152,95 +798,151 @@ udp_output(struct mbuf *m, ...) m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); } else ui->ui_sum = 0; - ((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 */ + + struct ip *ui_ip = (struct ip *)ui; + ui_ip->ip_len = htons(sizeof (struct udpiphdr) + len); + + struct ip *inp_ip = in_getiphdr(inp); + ui_ip->ip_ttl = inp_ip->ip_ttl; /* XXX */ + ui_ip->ip_tos = inp_ip->ip_tos; /* XXX */ UDP_STATINC(UDP_STAT_OPACKETS); - return (ip_output(m, inp->inp_options, ro, - inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST), - inp->inp_moptions, inp->inp_socket)); + ro = inpcb_get_route(inp); + return ip_output(m, inpcb_get_options(inp), ro, + so->so_options & (SO_DONTROUTE | SO_BROADCAST), + inpcb_get_moptions(inp), so); release: m_freem(m); - return (error); + return error; } -int udp_sendspace = 9216; /* really max datagram size */ -int udp_recvspace = 40 * (1024 + sizeof(struct sockaddr_in)); - /* 40 1K datagrams */ +static int +udp_attach(struct socket *so, int proto) +{ + inpcb_t *inp; + struct ip *ip; + int error; + + KASSERT(sotoinpcb(so) == NULL); -/*ARGSUSED*/ -int + /* Assign the lock (must happen if we will error out). */ + sosetlock(so); + +#ifdef MBUFTRACE + so->so_mowner = &udp_mowner; + so->so_rcv.sb_mowner = &udp_rx_mowner; + so->so_snd.sb_mowner = &udp_tx_mowner; +#endif + if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { + error = soreserve(so, udp_sendspace, udp_recvspace); + if (error) { + return error; + } + } + + error = inpcb_create(so, udbtable); + if (error) { + return error; + } + inp = sotoinpcb(so); + ip = in_getiphdr(inp); + ip->ip_ttl = ip_defttl; + return error; +} + +static void +udp_detach(struct socket *so) +{ + inpcb_t *inp; + + KASSERT(solocked(so)); + inp = sotoinpcb(so); + KASSERT(inp != NULL); + inpcb_destroy(inp); +} + +static int +udp_send(struct socket *so, struct mbuf *m, struct mbuf *nam, + struct mbuf *control) +{ + inpcb_t *inp; + struct in_addr laddr; + int error; + + KASSERT(solocked(so)); + inp = sotoinpcb(so); + + if (control && control->m_len) { + m_freem(control); + m_freem(m); + return EINVAL; + } + + if ((so->so_state & SS_ISCONNECTED) != 0) { + m_freem(m); + return nam ? EISCONN : ENOTCONN; + } + + if (nam) { + /* + * XXX: sendto() case - temporarily connect the socket + * to the destination, send and then disconnect. Also, + * preserve the local address as it may be changed. + */ + inpcb_get_addrs(inp, &laddr, NULL); + if ((error = inpcb_connect(inp, nam, curlwp)) != 0) { + m_freem(m); + return error; + } + } + error = udp_output(m, inp); + if (nam) { + inpcb_disconnect(inp); + inpcb_set_addrs(inp, &laddr, NULL); + inpcb_set_state(inp, INP_BOUND); + } + return error; +} + +static int udp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, - struct mbuf *control, struct lwp *l) + struct mbuf *control, struct lwp *l) { - struct inpcb *inp; - int s; + inpcb_t *inp; int error = 0; - if (req == PRU_CONTROL) - return (in_control(so, (long)m, (void *)nam, - (struct ifnet *)control, l)); - - s = splsoftnet(); + KASSERT(req != PRU_ATTACH); + KASSERT(req != PRU_DETACH); + if (req == PRU_CONTROL) { + KERNEL_LOCK(1, NULL); + error = in_control(so, (long)m, nam, (ifnet_t *)control, l); + KERNEL_UNLOCK_ONE(NULL); + return error; + } if (req == PRU_PURGEIF) { + KERNEL_LOCK(1, NULL); mutex_enter(softnet_lock); - in_pcbpurgeif0(&udbtable, (struct ifnet *)control); - in_purgeif((struct ifnet *)control); - in_pcbpurgeif(&udbtable, (struct ifnet *)control); + inpcb_purgeif0(udbtable, (ifnet_t *)control); + in_purgeif((ifnet_t *)control); + inpcb_purgeif(udbtable, (ifnet_t *)control); mutex_exit(softnet_lock); - splx(s); - return (0); + KERNEL_UNLOCK_ONE(NULL); + return 0; } + KASSERT(solocked(so)); inp = sotoinpcb(so); -#ifdef DIAGNOSTIC - if (req != PRU_SEND && req != PRU_SENDOOB && control) - panic("udp_usrreq: unexpected control mbuf"); -#endif - if (req == PRU_ATTACH) { - sosetlock(so); - } else if (inp == 0) { - error = EINVAL; - goto release; + + KASSERT(!control || (req == PRU_SEND || req == PRU_SENDOOB)); + if (inp == NULL) { + return EINVAL; } - /* - * Note: need to block udp_input while changing - * the udp pcb queue and/or pcb addresses. - */ switch (req) { - - case PRU_ATTACH: - if (inp != 0) { - error = EISCONN; - break; - } -#ifdef MBUFTRACE - so->so_mowner = &udp_mowner; - so->so_rcv.sb_mowner = &udp_rx_mowner; - so->so_snd.sb_mowner = &udp_tx_mowner; -#endif - if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { - error = soreserve(so, udp_sendspace, udp_recvspace); - if (error) - break; - } - error = in_pcballoc(so, &udbtable); - if (error) - break; - inp = sotoinpcb(so); - inp->inp_ip.ip_ttl = ip_defttl; - break; - - case PRU_DETACH: - in_pcbdetach(inp); - break; - case PRU_BIND: - error = in_pcbbind(inp, nam, l); + error = inpcb_bind(inp, nam, l); break; case PRU_LISTEN: @@ -1248,7 +950,7 @@ udp_usrreq(struct socket *so, int req, s break; case PRU_CONNECT: - error = in_pcbconnect(inp, nam, l); + error = inpcb_connect(inp, nam, l); if (error) break; soisconnected(so); @@ -1260,10 +962,8 @@ udp_usrreq(struct socket *so, int req, s case PRU_DISCONNECT: /*soisdisconnected(so);*/ - so->so_state &= ~SS_ISCONNECTED; /* XXX */ - in_pcbdisconnect(inp); - inp->inp_laddr = zeroin_addr; /* XXX */ - in_pcbstate(inp, INP_BOUND); /* XXX */ + so->so_state &= ~SS_ISCONNECTED; /* XXX */ + inpcb_disconnect(inp); break; case PRU_SHUTDOWN: @@ -1275,49 +975,14 @@ udp_usrreq(struct socket *so, int req, s break; case PRU_SEND: - if (control && control->m_len) { - m_freem(control); - m_freem(m); - error = EINVAL; - break; - } - { - struct in_addr laddr; /* XXX */ - - if (nam) { - laddr = inp->inp_laddr; /* XXX */ - if ((so->so_state & SS_ISCONNECTED) != 0) { - error = EISCONN; - goto die; - } - error = in_pcbconnect(inp, nam, l); - if (error) - goto die; - } else { - if ((so->so_state & SS_ISCONNECTED) == 0) { - error = ENOTCONN; - goto die; - } - } - error = udp_output(m, inp); - m = NULL; - if (nam) { - in_pcbdisconnect(inp); - inp->inp_laddr = laddr; /* XXX */ - in_pcbstate(inp, INP_BOUND); /* XXX */ - } - die: - if (m) - m_freem(m); - } + error = udp_send(so, m, nam, control); break; case PRU_SENSE: /* * stat: don't bother with a blocksize. */ - splx(s); - return (0); + return 0; case PRU_RCVOOB: error = EOPNOTSUPP; @@ -1330,40 +995,33 @@ udp_usrreq(struct socket *so, int req, s break; case PRU_SOCKADDR: - in_setsockaddr(inp, nam); + inpcb_fetch_sockaddr(inp, nam); break; case PRU_PEERADDR: - in_setpeeraddr(inp, nam); + inpcb_fetch_peeraddr(inp, nam); break; default: panic("udp_usrreq"); } -release: - splx(s); - return (error); + return error; } static int sysctl_net_inet_udp_stats(SYSCTLFN_ARGS) { - netstat_sysctl_context ctx; - uint64_t udps[UDP_NSTATS]; - ctx.ctx_stat = udpstat_percpu; - ctx.ctx_counters = udps; - ctx.ctx_ncounters = UDP_NSTATS; - return (NETSTAT_SYSCTL(&ctx)); + return (NETSTAT_SYSCTL(udpstat_percpu, UDP_NSTATS)); } /* * Sysctl for udp variables. */ -SYSCTL_SETUP(sysctl_net_inet_udp_setup, "sysctl net.inet.udp subtree setup") +static void +sysctl_net_inet_udp_setup(struct sysctllog **clog) { - sysctl_createv(clog, 0, NULL, NULL, CTLFLAG_PERMANENT, CTLTYPE_NODE, "net", NULL, @@ -1413,7 +1071,7 @@ SYSCTL_SETUP(sysctl_net_inet_udp_setup, CTLFLAG_PERMANENT, CTLTYPE_STRUCT, "pcblist", SYSCTL_DESCR("UDP protocol control block list"), - sysctl_inpcblist, 0, &udbtable, 0, + sysctl_inpcblist, 0, udbtable, 0, CTL_NET, PF_INET, IPPROTO_UDP, CTL_CREATE, CTL_EOL); sysctl_createv(clog, 0, NULL, NULL, @@ -1434,7 +1092,7 @@ udp_statinc(u_int stat) UDP_STATINC(stat); } -#if (defined INET && defined IPSEC_NAT_T) +#if defined(INET) && defined(IPSEC) /* * Returns: * 1 if the packet was processed @@ -1447,16 +1105,16 @@ udp4_espinudp(struct mbuf **mp, int off, { size_t len; void *data; - struct inpcb *inp; + inpcb_t *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; + in_port_t sport, dport; struct mbuf *m = *mp; + int inpflags; /* * Collapse the mbuf chain if the first mbuf is too short @@ -1480,15 +1138,18 @@ udp4_espinudp(struct mbuf **mp, int off, /* Ignore keepalive packets */ if ((len == 1) && (*(unsigned char *)data == 0xff)) { + m_free(m); + *mp = NULL; /* avoid any further processiong by caller ... */ return 1; } + inpflags = inpcb_get_flags(inp); /* * 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) { + if (inpflags & INP_ESPINUDP) { u_int32_t *st = (u_int32_t *)data; if ((len <= sizeof(struct esp)) || (*st == 0)) @@ -1497,7 +1158,7 @@ udp4_espinudp(struct mbuf **mp, int off, skip = sizeof(struct udphdr); } - if (inp->inp_flags & INP_ESPINUDP_NON_IKE) { + if (inpflags & INP_ESPINUDP_NON_IKE) { u_int32_t *st = (u_int32_t *)data; if ((len <= sizeof(u_int64_t) + sizeof(struct esp)) @@ -1539,16 +1200,9 @@ udp4_espinudp(struct mbuf **mp, int off, 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; - } - - /* + * We have modified the packet - it is now ESP, so we should not + * return to UDP processing ... + * * 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 @@ -1557,20 +1211,27 @@ udp4_espinudp(struct mbuf **mp, int off, 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; + m_freem(m); + return -1; } ((u_int16_t *)(tag + 1))[0] = sport; ((u_int16_t *)(tag + 1))[1] = dport; - m_tag_prepend(n, tag); + m_tag_prepend(m, tag); -#ifdef FAST_IPSEC - ipsec4_common_input(n, iphdrlen, IPPROTO_ESP); +#ifdef IPSEC + ipsec4_common_input(m, iphdrlen, IPPROTO_ESP); #else - esp4_input(n, iphdrlen); + esp4_input(m, iphdrlen); #endif - /* We handled it, it shoudln't be handled by UDP */ + /* We handled it, it shouldn't be handled by UDP */ + *mp = NULL; /* avoid free by caller ... */ return 1; } #endif + +const struct pr_usrreqs udp_usrreqs = { + .pr_attach = udp_attach, + .pr_detach = udp_detach, + .pr_generic = udp_usrreq, +};