2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2008 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
6 * Copyright (c) 2014 Kevin Lo
9 * Portions of this software were developed by Robert N. M. Watson under
10 * contract to Juniper Networks, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
51 #include <sys/kernel.h>
53 #include <sys/malloc.h>
57 #include <sys/protosw.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 #include <net/rss_config.h>
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/in_systm.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip.h>
81 #include <netinet/ip6.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/icmp_var.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/ip_options.h>
88 #include <netinet6/ip6_var.h>
90 #include <netinet/udp.h>
91 #include <netinet/udp_var.h>
92 #include <netinet/udplite.h>
93 #include <netinet/in_rss.h>
96 #include <netipsec/ipsec.h>
97 #include <netipsec/esp.h>
100 #include <machine/in_cksum.h>
102 #include <security/mac/mac_framework.h>
105 * UDP and UDP-Lite protocols implementation.
106 * Per RFC 768, August, 1980.
107 * Per RFC 3828, July, 2004.
111 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
112 * removes the only data integrity mechanism for packets and malformed
113 * packets that would otherwise be discarded due to bad checksums, and may
114 * cause problems (especially for NFS data blocks).
116 VNET_DEFINE(int, udp_cksum) = 1;
117 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
118 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
120 int udp_log_in_vain = 0;
121 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
122 &udp_log_in_vain, 0, "Log all incoming UDP packets");
124 VNET_DEFINE(int, udp_blackhole) = 0;
125 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
126 &VNET_NAME(udp_blackhole), 0,
127 "Do not send port unreachables for refused connects");
129 static VNET_DEFINE(int, udp_require_l2_bcast) = 0;
130 #define V_udp_require_l2_bcast VNET(udp_require_l2_bcast)
131 SYSCTL_INT(_net_inet_udp, OID_AUTO, require_l2_bcast, CTLFLAG_VNET | CTLFLAG_RW,
132 &VNET_NAME(udp_require_l2_bcast), 0,
133 "Only treat packets sent to an L2 broadcast address as broadcast packets");
135 u_long udp_sendspace = 9216; /* really max datagram size */
136 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
137 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
139 u_long udp_recvspace = 40 * (1024 +
141 sizeof(struct sockaddr_in6)
143 sizeof(struct sockaddr_in)
145 ); /* 40 1K datagrams */
147 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
148 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
150 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
151 VNET_DEFINE(struct inpcbinfo, udbinfo);
152 VNET_DEFINE(struct inpcbhead, ulitecb);
153 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
154 static VNET_DEFINE(uma_zone_t, udpcb_zone);
155 #define V_udpcb_zone VNET(udpcb_zone)
158 #define UDBHASHSIZE 128
161 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
162 VNET_PCPUSTAT_SYSINIT(udpstat);
163 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
164 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
167 VNET_PCPUSTAT_SYSUNINIT(udpstat);
170 static void udp_detach(struct socket *so);
171 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
172 struct mbuf *, struct thread *);
177 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
179 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
181 #endif /* IPSEC_NAT_T */
185 udp_zone_change(void *tag)
188 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
189 uma_zone_set_max(V_udpcb_zone, maxsockets);
193 udp_inpcb_init(void *mem, int size, int flags)
198 INP_LOCK_INIT(inp, "inp", "udpinp");
203 udplite_inpcb_init(void *mem, int size, int flags)
208 INP_LOCK_INIT(inp, "inp", "udpliteinp");
217 * For now default to 2-tuple UDP hashing - until the fragment
218 * reassembly code can also update the flowid.
220 * Once we can calculate the flowid that way and re-establish
221 * a 4-tuple, flip this to 4-tuple.
223 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
224 "udp_inpcb", udp_inpcb_init, NULL, 0,
225 IPI_HASHFIELDS_2TUPLE);
226 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
227 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
228 uma_zone_set_max(V_udpcb_zone, maxsockets);
229 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
230 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
231 EVENTHANDLER_PRI_ANY);
238 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
239 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
240 0, IPI_HASHFIELDS_2TUPLE);
244 * Kernel module interface for updating udpstat. The argument is an index
245 * into udpstat treated as an array of u_long. While this encodes the
246 * general layout of udpstat into the caller, it doesn't encode its location,
247 * so that future changes to add, for example, per-CPU stats support won't
248 * cause binary compatibility problems for kernel modules.
251 kmod_udpstat_inc(int statnum)
254 counter_u64_add(VNET(udpstat)[statnum], 1);
258 udp_newudpcb(struct inpcb *inp)
262 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
270 udp_discardcb(struct udpcb *up)
273 uma_zfree(V_udpcb_zone, up);
278 udp_destroy(void *unused __unused)
281 in_pcbinfo_destroy(&V_udbinfo);
282 uma_zdestroy(V_udpcb_zone);
284 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
287 udplite_destroy(void *unused __unused)
290 in_pcbinfo_destroy(&V_ulitecbinfo);
292 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
298 * Subroutine of udp_input(), which appends the provided mbuf chain to the
299 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
300 * contains the source address. If the socket ends up being an IPv6 socket,
301 * udp_append() will convert to a sockaddr_in6 before passing the address
302 * into the socket code.
304 * In the normal case udp_append() will return 0, indicating that you
305 * must unlock the inp. However if a tunneling protocol is in place we increment
306 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
307 * then decrement the reference count. If the inp_rele returns 1, indicating the
308 * inp is gone, we return that to the caller to tell them *not* to unlock
309 * the inp. In the case of multi-cast this will cause the distribution
310 * to stop (though most tunneling protocols known currently do *not* use
314 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
315 struct sockaddr_in *udp_in)
317 struct sockaddr *append_sa;
319 struct mbuf *opts = NULL;
321 struct sockaddr_in6 udp_in6;
325 INP_LOCK_ASSERT(inp);
328 * Engage the tunneling protocol.
331 if (up->u_tun_func != NULL) {
334 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
337 return (in_pcbrele_rlocked(inp));
340 off += sizeof(struct udphdr);
343 /* Check AH/ESP integrity. */
344 if (ipsec4_in_reject(n, inp)) {
350 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
351 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
352 n = udp4_espdecap(inp, n, off);
353 if (n == NULL) /* Consumed. */
356 #endif /* IPSEC_NAT_T */
359 if (mac_inpcb_check_deliver(inp, n) != 0) {
364 if (inp->inp_flags & INP_CONTROLOPTS ||
365 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
367 if (inp->inp_vflag & INP_IPV6)
368 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
371 ip_savecontrol(inp, &opts, ip, n);
374 if (inp->inp_vflag & INP_IPV6) {
375 bzero(&udp_in6, sizeof(udp_in6));
376 udp_in6.sin6_len = sizeof(udp_in6);
377 udp_in6.sin6_family = AF_INET6;
378 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
379 append_sa = (struct sockaddr *)&udp_in6;
382 append_sa = (struct sockaddr *)udp_in;
385 so = inp->inp_socket;
386 SOCKBUF_LOCK(&so->so_rcv);
387 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
388 SOCKBUF_UNLOCK(&so->so_rcv);
392 UDPSTAT_INC(udps_fullsock);
394 sorwakeup_locked(so);
399 udp_input(struct mbuf **mp, int *offp, int proto)
405 uint16_t len, ip_len;
406 struct inpcbinfo *pcbinfo;
408 struct sockaddr_in udp_in;
410 struct m_tag *fwd_tag;
411 int cscov_partial, iphlen;
415 ifp = m->m_pkthdr.rcvif;
417 UDPSTAT_INC(udps_ipackets);
420 * Strip IP options, if any; should skip this, make available to
421 * user, and use on returned packets, but we don't yet have a way to
422 * check the checksum with options still present.
424 if (iphlen > sizeof (struct ip)) {
426 iphlen = sizeof(struct ip);
430 * Get IP and UDP header together in first mbuf.
432 ip = mtod(m, struct ip *);
433 if (m->m_len < iphlen + sizeof(struct udphdr)) {
434 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
435 UDPSTAT_INC(udps_hdrops);
436 return (IPPROTO_DONE);
438 ip = mtod(m, struct ip *);
440 uh = (struct udphdr *)((caddr_t)ip + iphlen);
441 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
444 * Destination port of 0 is illegal, based on RFC768.
446 if (uh->uh_dport == 0)
450 * Construct sockaddr format source address. Stuff source address
451 * and datagram in user buffer.
453 bzero(&udp_in, sizeof(udp_in));
454 udp_in.sin_len = sizeof(udp_in);
455 udp_in.sin_family = AF_INET;
456 udp_in.sin_port = uh->uh_sport;
457 udp_in.sin_addr = ip->ip_src;
460 * Make mbuf data length reflect UDP length. If not enough data to
461 * reflect UDP length, drop.
463 len = ntohs((u_short)uh->uh_ulen);
464 ip_len = ntohs(ip->ip_len) - iphlen;
465 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
466 /* Zero means checksum over the complete packet. */
472 if (len > ip_len || len < sizeof(struct udphdr)) {
473 UDPSTAT_INC(udps_badlen);
476 if (proto == IPPROTO_UDP)
477 m_adj(m, len - ip_len);
481 * Save a copy of the IP header in case we want restore it for
482 * sending an ICMP error message in response.
484 if (!V_udp_blackhole)
487 memset(&save_ip, 0, sizeof(save_ip));
490 * Checksum extended UDP header and data.
495 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
497 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
498 uh_sum = m->m_pkthdr.csum_data;
500 uh_sum = in_pseudo(ip->ip_src.s_addr,
501 ip->ip_dst.s_addr, htonl((u_short)len +
502 m->m_pkthdr.csum_data + proto));
507 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
508 bzero(((struct ipovly *)ip)->ih_x1, 9);
509 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
510 uh->uh_ulen : htons(ip_len);
511 uh_sum = in_cksum(m, len + sizeof (struct ip));
512 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
515 UDPSTAT_INC(udps_badsum);
517 return (IPPROTO_DONE);
520 if (proto == IPPROTO_UDP) {
521 UDPSTAT_INC(udps_nosum);
523 /* UDPLite requires a checksum */
524 /* XXX: What is the right UDPLite MIB counter here? */
526 return (IPPROTO_DONE);
530 pcbinfo = udp_get_inpcbinfo(proto);
531 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
532 ((!V_udp_require_l2_bcast || m->m_flags & M_BCAST) &&
533 in_broadcast(ip->ip_dst, ifp))) {
535 struct inpcbhead *pcblist;
536 struct ip_moptions *imo;
538 INP_INFO_RLOCK(pcbinfo);
539 pcblist = udp_get_pcblist(proto);
541 LIST_FOREACH(inp, pcblist, inp_list) {
542 if (inp->inp_lport != uh->uh_dport)
545 if ((inp->inp_vflag & INP_IPV4) == 0)
548 if (inp->inp_laddr.s_addr != INADDR_ANY &&
549 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
551 if (inp->inp_faddr.s_addr != INADDR_ANY &&
552 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
554 if (inp->inp_fport != 0 &&
555 inp->inp_fport != uh->uh_sport)
561 * XXXRW: Because we weren't holding either the inpcb
562 * or the hash lock when we checked for a match
563 * before, we should probably recheck now that the
564 * inpcb lock is held.
568 * Handle socket delivery policy for any-source
569 * and source-specific multicast. [RFC3678]
571 imo = inp->inp_moptions;
572 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
573 struct sockaddr_in group;
579 bzero(&group, sizeof(struct sockaddr_in));
580 group.sin_len = sizeof(struct sockaddr_in);
581 group.sin_family = AF_INET;
582 group.sin_addr = ip->ip_dst;
584 blocked = imo_multi_filter(imo, ifp,
585 (struct sockaddr *)&group,
586 (struct sockaddr *)&udp_in);
587 if (blocked != MCAST_PASS) {
588 if (blocked == MCAST_NOTGMEMBER)
589 IPSTAT_INC(ips_notmember);
590 if (blocked == MCAST_NOTSMEMBER ||
591 blocked == MCAST_MUTED)
592 UDPSTAT_INC(udps_filtermcast);
600 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
602 UDP_PROBE(receive, NULL, last, ip,
604 if (udp_append(last, ip, n, iphlen,
613 * Don't look for additional matches if this one does
614 * not have either the SO_REUSEPORT or SO_REUSEADDR
615 * socket options set. This heuristic avoids
616 * searching through all pcbs in the common case of a
617 * non-shared port. It assumes that an application
618 * will never clear these options after setting them.
620 if ((last->inp_socket->so_options &
621 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
627 * No matching pcb found; discard datagram. (No need
628 * to send an ICMP Port Unreachable for a broadcast
629 * or multicast datgram.)
631 UDPSTAT_INC(udps_noportbcast);
634 INP_INFO_RUNLOCK(pcbinfo);
637 UDP_PROBE(receive, NULL, last, ip, last, uh);
638 if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
641 INP_INFO_RUNLOCK(pcbinfo);
642 return (IPPROTO_DONE);
646 * Locate pcb for datagram.
650 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
652 if ((m->m_flags & M_IP_NEXTHOP) &&
653 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
654 struct sockaddr_in *next_hop;
656 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
659 * Transparently forwarded. Pretend to be the destination.
660 * Already got one like this?
662 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
663 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
666 * It's new. Try to find the ambushing socket.
667 * Because we've rewritten the destination address,
668 * any hardware-generated hash is ignored.
670 inp = in_pcblookup(pcbinfo, ip->ip_src,
671 uh->uh_sport, next_hop->sin_addr,
672 next_hop->sin_port ? htons(next_hop->sin_port) :
673 uh->uh_dport, INPLOOKUP_WILDCARD |
674 INPLOOKUP_RLOCKPCB, ifp);
676 /* Remove the tag from the packet. We don't need it anymore. */
677 m_tag_delete(m, fwd_tag);
678 m->m_flags &= ~M_IP_NEXTHOP;
680 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
681 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
682 INPLOOKUP_RLOCKPCB, ifp, m);
684 if (udp_log_in_vain) {
685 char buf[4*sizeof "123"];
687 strcpy(buf, inet_ntoa(ip->ip_dst));
689 "Connection attempt to UDP %s:%d from %s:%d\n",
690 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
691 ntohs(uh->uh_sport));
693 UDPSTAT_INC(udps_noport);
694 if (m->m_flags & (M_BCAST | M_MCAST)) {
695 UDPSTAT_INC(udps_noportbcast);
700 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
703 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
704 return (IPPROTO_DONE);
708 * Check the minimum TTL for socket.
710 INP_RLOCK_ASSERT(inp);
711 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
714 return (IPPROTO_DONE);
720 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
723 return (IPPROTO_DONE);
727 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
728 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
730 return (IPPROTO_DONE);
734 return (IPPROTO_DONE);
739 * Notify a udp user of an asynchronous error; just wake up so that they can
740 * collect error status.
743 udp_notify(struct inpcb *inp, int errno)
747 * While udp_ctlinput() always calls udp_notify() with a read lock
748 * when invoking it directly, in_pcbnotifyall() currently uses write
749 * locks due to sharing code with TCP. For now, accept either a read
750 * or a write lock, but a read lock is sufficient.
752 INP_LOCK_ASSERT(inp);
753 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
754 errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
755 RTFREE(inp->inp_route.ro_rt);
756 inp->inp_route.ro_rt = (struct rtentry *)NULL;
759 inp->inp_socket->so_error = errno;
760 sorwakeup(inp->inp_socket);
761 sowwakeup(inp->inp_socket);
767 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
768 struct inpcbinfo *pcbinfo)
772 struct in_addr faddr;
775 faddr = ((struct sockaddr_in *)sa)->sin_addr;
776 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
779 if (PRC_IS_REDIRECT(cmd)) {
780 /* signal EHOSTDOWN, as it flushes the cached route */
781 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
786 * Hostdead is ugly because it goes linearly through all PCBs.
788 * XXX: We never get this from ICMP, otherwise it makes an excellent
789 * DoS attack on machines with many connections.
791 if (cmd == PRC_HOSTDEAD)
793 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
796 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
797 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
798 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
800 INP_RLOCK_ASSERT(inp);
801 if (inp->inp_socket != NULL) {
802 udp_notify(inp, inetctlerrmap[cmd]);
806 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
807 ip->ip_src, uh->uh_sport,
808 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
813 if (up->u_icmp_func != NULL) {
815 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
822 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
826 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
829 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
833 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
836 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
841 udp_pcblist(SYSCTL_HANDLER_ARGS)
844 struct inpcb *inp, **inp_list;
849 * The process of preparing the PCB list is too time-consuming and
850 * resource-intensive to repeat twice on every request.
852 if (req->oldptr == 0) {
853 n = V_udbinfo.ipi_count;
854 n += imax(n / 8, 10);
855 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
859 if (req->newptr != 0)
863 * OK, now we're committed to doing something.
865 INP_INFO_RLOCK(&V_udbinfo);
866 gencnt = V_udbinfo.ipi_gencnt;
867 n = V_udbinfo.ipi_count;
868 INP_INFO_RUNLOCK(&V_udbinfo);
870 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
871 + n * sizeof(struct xinpcb));
875 xig.xig_len = sizeof xig;
877 xig.xig_gen = gencnt;
878 xig.xig_sogen = so_gencnt;
879 error = SYSCTL_OUT(req, &xig, sizeof xig);
883 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
884 if (inp_list == NULL)
887 INP_INFO_RLOCK(&V_udbinfo);
888 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
889 inp = LIST_NEXT(inp, inp_list)) {
891 if (inp->inp_gencnt <= gencnt &&
892 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
898 INP_INFO_RUNLOCK(&V_udbinfo);
902 for (i = 0; i < n; i++) {
905 if (inp->inp_gencnt <= gencnt) {
908 bzero(&xi, sizeof(xi));
909 xi.xi_len = sizeof xi;
910 /* XXX should avoid extra copy */
911 bcopy(inp, &xi.xi_inp, sizeof *inp);
913 sotoxsocket(inp->inp_socket, &xi.xi_socket);
914 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
916 error = SYSCTL_OUT(req, &xi, sizeof xi);
920 INP_INFO_WLOCK(&V_udbinfo);
921 for (i = 0; i < n; i++) {
924 if (!in_pcbrele_rlocked(inp))
927 INP_INFO_WUNLOCK(&V_udbinfo);
931 * Give the user an updated idea of our state. If the
932 * generation differs from what we told her before, she knows
933 * that something happened while we were processing this
934 * request, and it might be necessary to retry.
936 INP_INFO_RLOCK(&V_udbinfo);
937 xig.xig_gen = V_udbinfo.ipi_gencnt;
938 xig.xig_sogen = so_gencnt;
939 xig.xig_count = V_udbinfo.ipi_count;
940 INP_INFO_RUNLOCK(&V_udbinfo);
941 error = SYSCTL_OUT(req, &xig, sizeof xig);
943 free(inp_list, M_TEMP);
947 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
948 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
949 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
953 udp_getcred(SYSCTL_HANDLER_ARGS)
956 struct sockaddr_in addrs[2];
960 error = priv_check(req->td, PRIV_NETINET_GETCRED);
963 error = SYSCTL_IN(req, addrs, sizeof(addrs));
966 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
967 addrs[0].sin_addr, addrs[0].sin_port,
968 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
970 INP_RLOCK_ASSERT(inp);
971 if (inp->inp_socket == NULL)
974 error = cr_canseeinpcb(req->td->td_ucred, inp);
976 cru2x(inp->inp_cred, &xuc);
981 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
985 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
986 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
987 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
991 udp_ctloutput(struct socket *so, struct sockopt *sopt)
995 int isudplite, error, optval;
998 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
1000 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1002 if (sopt->sopt_level != so->so_proto->pr_protocol) {
1004 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
1006 error = ip6_ctloutput(so, sopt);
1009 #if defined(INET) && defined(INET6)
1015 error = ip_ctloutput(so, sopt);
1021 switch (sopt->sopt_dir) {
1023 switch (sopt->sopt_name) {
1026 error = sooptcopyin(sopt, &optval, sizeof optval,
1030 inp = sotoinpcb(so);
1031 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1034 up = intoudpcb(inp);
1035 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1039 /* Clear all UDP encap. */
1041 up->u_flags &= ~UF_ESPINUDP_ALL;
1045 case UDP_ENCAP_ESPINUDP:
1046 case UDP_ENCAP_ESPINUDP_NON_IKE:
1047 up->u_flags &= ~UF_ESPINUDP_ALL;
1048 if (optval == UDP_ENCAP_ESPINUDP)
1049 up->u_flags |= UF_ESPINUDP;
1050 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1051 up->u_flags |= UF_ESPINUDP_NON_IKE;
1060 case UDPLITE_SEND_CSCOV:
1061 case UDPLITE_RECV_CSCOV:
1064 error = ENOPROTOOPT;
1068 error = sooptcopyin(sopt, &optval, sizeof(optval),
1072 inp = sotoinpcb(so);
1073 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1075 up = intoudpcb(inp);
1076 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1077 if ((optval != 0 && optval < 8) || (optval > 65535)) {
1082 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1083 up->u_txcslen = optval;
1085 up->u_rxcslen = optval;
1090 error = ENOPROTOOPT;
1095 switch (sopt->sopt_name) {
1098 up = intoudpcb(inp);
1099 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1100 optval = up->u_flags & UF_ESPINUDP_ALL;
1102 error = sooptcopyout(sopt, &optval, sizeof optval);
1105 case UDPLITE_SEND_CSCOV:
1106 case UDPLITE_RECV_CSCOV:
1109 error = ENOPROTOOPT;
1112 up = intoudpcb(inp);
1113 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1114 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1115 optval = up->u_txcslen;
1117 optval = up->u_rxcslen;
1119 error = sooptcopyout(sopt, &optval, sizeof(optval));
1123 error = ENOPROTOOPT;
1132 #define UH_WLOCKED 2
1133 #define UH_RLOCKED 1
1134 #define UH_UNLOCKED 0
1136 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1137 struct mbuf *control, struct thread *td)
1139 struct udpiphdr *ui;
1140 int len = m->m_pkthdr.len;
1141 struct in_addr faddr, laddr;
1143 struct inpcbinfo *pcbinfo;
1144 struct sockaddr_in *sin, src;
1145 int cscov_partial = 0;
1148 u_short fport, lport;
1149 int unlock_udbinfo, unlock_inp;
1153 uint32_t flowid = 0;
1154 uint8_t flowtype = M_HASHTYPE_NONE;
1157 * udp_output() may need to temporarily bind or connect the current
1158 * inpcb. As such, we don't know up front whether we will need the
1159 * pcbinfo lock or not. Do any work to decide what is needed up
1160 * front before acquiring any locks.
1162 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1170 sin = (struct sockaddr_in *)addr;
1172 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1174 unlock_inp = UH_WLOCKED;
1177 unlock_inp = UH_RLOCKED;
1179 tos = inp->inp_ip_tos;
1180 if (control != NULL) {
1182 * XXX: Currently, we assume all the optional information is
1183 * stored in a single mbuf.
1185 if (control->m_next) {
1186 if (unlock_inp == UH_WLOCKED)
1194 for (; control->m_len > 0;
1195 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1196 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1197 cm = mtod(control, struct cmsghdr *);
1198 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1199 || cm->cmsg_len > control->m_len) {
1203 if (cm->cmsg_level != IPPROTO_IP)
1206 switch (cm->cmsg_type) {
1207 case IP_SENDSRCADDR:
1209 CMSG_LEN(sizeof(struct in_addr))) {
1213 bzero(&src, sizeof(src));
1214 src.sin_family = AF_INET;
1215 src.sin_len = sizeof(src);
1216 src.sin_port = inp->inp_lport;
1218 *(struct in_addr *)CMSG_DATA(cm);
1222 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1226 tos = *(u_char *)CMSG_DATA(cm);
1230 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1234 flowid = *(uint32_t *) CMSG_DATA(cm);
1238 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1242 flowtype = *(uint32_t *) CMSG_DATA(cm);
1246 case IP_RSSBUCKETID:
1247 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1251 /* This is just a placeholder for now */
1255 error = ENOPROTOOPT;
1264 if (unlock_inp == UH_WLOCKED)
1273 * Depending on whether or not the application has bound or connected
1274 * the socket, we may have to do varying levels of work. The optimal
1275 * case is for a connected UDP socket, as a global lock isn't
1278 * In order to decide which we need, we require stability of the
1279 * inpcb binding, which we ensure by acquiring a read lock on the
1280 * inpcb. This doesn't strictly follow the lock order, so we play
1281 * the trylock and retry game; note that we may end up with more
1282 * conservative locks than required the second time around, so later
1283 * assertions have to accept that. Further analysis of the number of
1284 * misses under contention is required.
1286 * XXXRW: Check that hash locking update here is correct.
1288 pr = inp->inp_socket->so_proto->pr_protocol;
1289 pcbinfo = udp_get_inpcbinfo(pr);
1290 sin = (struct sockaddr_in *)addr;
1292 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1293 INP_HASH_WLOCK(pcbinfo);
1294 unlock_udbinfo = UH_WLOCKED;
1295 } else if ((sin != NULL && (
1296 (sin->sin_addr.s_addr == INADDR_ANY) ||
1297 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1298 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1299 (inp->inp_lport == 0))) ||
1300 (src.sin_family == AF_INET)) {
1301 INP_HASH_RLOCK(pcbinfo);
1302 unlock_udbinfo = UH_RLOCKED;
1304 unlock_udbinfo = UH_UNLOCKED;
1307 * If the IP_SENDSRCADDR control message was specified, override the
1308 * source address for this datagram. Its use is invalidated if the
1309 * address thus specified is incomplete or clobbers other inpcbs.
1311 laddr = inp->inp_laddr;
1312 lport = inp->inp_lport;
1313 if (src.sin_family == AF_INET) {
1314 INP_HASH_LOCK_ASSERT(pcbinfo);
1316 (laddr.s_addr == INADDR_ANY &&
1317 src.sin_addr.s_addr == INADDR_ANY)) {
1321 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1322 &laddr.s_addr, &lport, td->td_ucred);
1328 * If a UDP socket has been connected, then a local address/port will
1329 * have been selected and bound.
1331 * If a UDP socket has not been connected to, then an explicit
1332 * destination address must be used, in which case a local
1333 * address/port may not have been selected and bound.
1336 INP_LOCK_ASSERT(inp);
1337 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1343 * Jail may rewrite the destination address, so let it do
1344 * that before we use it.
1346 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1351 * If a local address or port hasn't yet been selected, or if
1352 * the destination address needs to be rewritten due to using
1353 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1354 * to do the heavy lifting. Once a port is selected, we
1355 * commit the binding back to the socket; we also commit the
1356 * binding of the address if in jail.
1358 * If we already have a valid binding and we're not
1359 * requesting a destination address rewrite, use a fast path.
1361 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1362 inp->inp_lport == 0 ||
1363 sin->sin_addr.s_addr == INADDR_ANY ||
1364 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1365 INP_HASH_LOCK_ASSERT(pcbinfo);
1366 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1367 &lport, &faddr.s_addr, &fport, NULL,
1373 * XXXRW: Why not commit the port if the address is
1376 /* Commit the local port if newly assigned. */
1377 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1378 inp->inp_lport == 0) {
1379 INP_WLOCK_ASSERT(inp);
1380 INP_HASH_WLOCK_ASSERT(pcbinfo);
1382 * Remember addr if jailed, to prevent
1385 if (prison_flag(td->td_ucred, PR_IP4))
1386 inp->inp_laddr = laddr;
1387 inp->inp_lport = lport;
1388 if (in_pcbinshash(inp) != 0) {
1393 inp->inp_flags |= INP_ANONPORT;
1396 faddr = sin->sin_addr;
1397 fport = sin->sin_port;
1400 INP_LOCK_ASSERT(inp);
1401 faddr = inp->inp_faddr;
1402 fport = inp->inp_fport;
1403 if (faddr.s_addr == INADDR_ANY) {
1410 * Calculate data length and get a mbuf for UDP, IP, and possible
1411 * link-layer headers. Immediate slide the data pointer back forward
1412 * since we won't use that space at this layer.
1414 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1419 m->m_data += max_linkhdr;
1420 m->m_len -= max_linkhdr;
1421 m->m_pkthdr.len -= max_linkhdr;
1424 * Fill in mbuf with extended UDP header and addresses and length put
1425 * into network format.
1427 ui = mtod(m, struct udpiphdr *);
1428 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1432 ui->ui_sport = lport;
1433 ui->ui_dport = fport;
1434 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1435 if (pr == IPPROTO_UDPLITE) {
1439 up = intoudpcb(inp);
1440 cscov = up->u_txcslen;
1441 plen = (u_short)len + sizeof(struct udphdr);
1444 ui->ui_len = htons(plen);
1445 ui->ui_ulen = htons(cscov);
1447 * For UDP-Lite, checksum coverage length of zero means
1448 * the entire UDPLite packet is covered by the checksum.
1450 cscov_partial = (cscov == 0) ? 0 : 1;
1452 ui->ui_v = IPVERSION << 4;
1455 * Set the Don't Fragment bit in the IP header.
1457 if (inp->inp_flags & INP_DONTFRAG) {
1460 ip = (struct ip *)&ui->ui_i;
1461 ip->ip_off |= htons(IP_DF);
1465 if (inp->inp_socket->so_options & SO_DONTROUTE)
1466 ipflags |= IP_ROUTETOIF;
1467 if (inp->inp_socket->so_options & SO_BROADCAST)
1468 ipflags |= IP_ALLOWBROADCAST;
1469 if (inp->inp_flags & INP_ONESBCAST)
1470 ipflags |= IP_SENDONES;
1473 mac_inpcb_create_mbuf(inp, m);
1477 * Set up checksum and output datagram.
1480 if (pr == IPPROTO_UDPLITE) {
1481 if (inp->inp_flags & INP_ONESBCAST)
1482 faddr.s_addr = INADDR_BROADCAST;
1483 if (cscov_partial) {
1484 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1485 ui->ui_sum = 0xffff;
1487 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1488 ui->ui_sum = 0xffff;
1490 } else if (V_udp_cksum) {
1491 if (inp->inp_flags & INP_ONESBCAST)
1492 faddr.s_addr = INADDR_BROADCAST;
1493 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1494 htons((u_short)len + sizeof(struct udphdr) + pr));
1495 m->m_pkthdr.csum_flags = CSUM_UDP;
1496 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1498 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1499 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1500 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1501 UDPSTAT_INC(udps_opackets);
1504 * Setup flowid / RSS information for outbound socket.
1506 * Once the UDP code decides to set a flowid some other way,
1507 * this allows the flowid to be overridden by userland.
1509 if (flowtype != M_HASHTYPE_NONE) {
1510 m->m_pkthdr.flowid = flowid;
1511 M_HASHTYPE_SET(m, flowtype);
1514 uint32_t hash_val, hash_type;
1516 * Calculate an appropriate RSS hash for UDP and
1519 * The called function will take care of figuring out
1520 * whether a 2-tuple or 4-tuple hash is required based
1521 * on the currently configured scheme.
1523 * Later later on connected socket values should be
1524 * cached in the inpcb and reused, rather than constantly
1525 * re-calculating it.
1527 * UDP Lite is a different protocol number and will
1528 * likely end up being hashed as a 2-tuple until
1529 * RSS / NICs grow UDP Lite protocol awareness.
1531 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1532 pr, &hash_val, &hash_type) == 0) {
1533 m->m_pkthdr.flowid = hash_val;
1534 M_HASHTYPE_SET(m, hash_type);
1541 * Don't override with the inp cached flowid value.
1543 * Depending upon the kind of send being done, the inp
1544 * flowid/flowtype values may actually not be appropriate
1545 * for this particular socket send.
1547 * We should either leave the flowid at zero (which is what is
1548 * currently done) or set it to some software generated
1549 * hash value based on the packet contents.
1551 ipflags |= IP_NODEFAULTFLOWID;
1554 if (unlock_udbinfo == UH_WLOCKED)
1555 INP_HASH_WUNLOCK(pcbinfo);
1556 else if (unlock_udbinfo == UH_RLOCKED)
1557 INP_HASH_RUNLOCK(pcbinfo);
1558 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1559 error = ip_output(m, inp->inp_options,
1560 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1561 inp->inp_moptions, inp);
1562 if (unlock_inp == UH_WLOCKED)
1569 if (unlock_udbinfo == UH_WLOCKED) {
1570 KASSERT(unlock_inp == UH_WLOCKED,
1571 ("%s: excl udbinfo lock, shared inp lock", __func__));
1572 INP_HASH_WUNLOCK(pcbinfo);
1574 } else if (unlock_udbinfo == UH_RLOCKED) {
1575 KASSERT(unlock_inp == UH_RLOCKED,
1576 ("%s: shared udbinfo lock, excl inp lock", __func__));
1577 INP_HASH_RUNLOCK(pcbinfo);
1579 } else if (unlock_inp == UH_WLOCKED)
1588 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1590 * Potentially decap ESP in UDP frame. Check for an ESP header
1591 * and optional marker; if present, strip the UDP header and
1592 * push the result through IPSec.
1594 * Returns mbuf to be processed (potentially re-allocated) or
1595 * NULL if consumed and/or processed.
1597 static struct mbuf *
1598 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1600 size_t minlen, payload, skip, iphlen;
1604 struct udphdr *udphdr;
1607 INP_RLOCK_ASSERT(inp);
1610 * Pull up data so the longest case is contiguous:
1611 * IP/UDP hdr + non ESP marker + ESP hdr.
1613 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1614 if (minlen > m->m_pkthdr.len)
1615 minlen = m->m_pkthdr.len;
1616 if ((m = m_pullup(m, minlen)) == NULL) {
1617 IPSECSTAT_INC(ips_in_inval);
1618 return (NULL); /* Bypass caller processing. */
1620 data = mtod(m, caddr_t); /* Points to ip header. */
1621 payload = m->m_len - off; /* Size of payload. */
1623 if (payload == 1 && data[off] == '\xff')
1624 return (m); /* NB: keepalive packet, no decap. */
1626 up = intoudpcb(inp);
1627 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1628 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1629 ("u_flags 0x%x", up->u_flags));
1632 * Check that the payload is large enough to hold an
1633 * ESP header and compute the amount of data to remove.
1635 * NB: the caller has already done a pullup for us.
1636 * XXX can we assume alignment and eliminate bcopys?
1638 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1640 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1641 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1642 * possible AH mode non-IKE marker+non-ESP marker
1643 * from draft-ietf-ipsec-udp-encaps-00.txt.
1647 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1648 return (m); /* NB: no decap. */
1649 bcopy(data + off, &marker, sizeof(uint64_t));
1650 if (marker != 0) /* Non-IKE marker. */
1651 return (m); /* NB: no decap. */
1652 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1656 if (payload <= sizeof(struct esp)) {
1657 IPSECSTAT_INC(ips_in_inval);
1659 return (NULL); /* Discard. */
1661 bcopy(data + off, &spi, sizeof(uint32_t));
1662 if (spi == 0) /* Non-ESP marker. */
1663 return (m); /* NB: no decap. */
1664 skip = sizeof(struct udphdr);
1668 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1669 * the UDP ports. This is required if we want to select
1670 * the right SPD for multiple hosts behind same NAT.
1672 * NB: ports are maintained in network byte order everywhere
1673 * in the NAT-T code.
1675 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1676 2 * sizeof(uint16_t), M_NOWAIT);
1678 IPSECSTAT_INC(ips_in_nomem);
1680 return (NULL); /* Discard. */
1682 iphlen = off - sizeof(struct udphdr);
1683 udphdr = (struct udphdr *)(data + iphlen);
1684 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1685 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1686 m_tag_prepend(m, tag);
1689 * Remove the UDP header (and possibly the non ESP marker)
1690 * IP header length is iphlen
1693 * +----+------+-----+
1694 * | IP | UDP | ESP |
1695 * +----+------+-----+
1703 ovbcopy(data, data + skip, iphlen);
1706 ip = mtod(m, struct ip *);
1707 ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1708 ip->ip_p = IPPROTO_ESP;
1711 * We cannot yet update the cksums so clear any
1712 * h/w cksum flags as they are no longer valid.
1714 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1715 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1717 (void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1719 return (NULL); /* NB: consumed, bypass processing. */
1721 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1724 udp_abort(struct socket *so)
1727 struct inpcbinfo *pcbinfo;
1729 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1730 inp = sotoinpcb(so);
1731 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1733 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1734 INP_HASH_WLOCK(pcbinfo);
1735 in_pcbdisconnect(inp);
1736 inp->inp_laddr.s_addr = INADDR_ANY;
1737 INP_HASH_WUNLOCK(pcbinfo);
1738 soisdisconnected(so);
1744 udp_attach(struct socket *so, int proto, struct thread *td)
1747 struct inpcbinfo *pcbinfo;
1750 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1751 inp = sotoinpcb(so);
1752 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1753 error = soreserve(so, udp_sendspace, udp_recvspace);
1756 INP_INFO_WLOCK(pcbinfo);
1757 error = in_pcballoc(so, pcbinfo);
1759 INP_INFO_WUNLOCK(pcbinfo);
1763 inp = sotoinpcb(so);
1764 inp->inp_vflag |= INP_IPV4;
1765 inp->inp_ip_ttl = V_ip_defttl;
1767 error = udp_newudpcb(inp);
1771 INP_INFO_WUNLOCK(pcbinfo);
1776 INP_INFO_WUNLOCK(pcbinfo);
1782 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1787 KASSERT(so->so_type == SOCK_DGRAM,
1788 ("udp_set_kernel_tunneling: !dgram"));
1789 inp = sotoinpcb(so);
1790 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1792 up = intoudpcb(inp);
1793 if ((up->u_tun_func != NULL) ||
1794 (up->u_icmp_func != NULL)) {
1799 up->u_icmp_func = i;
1800 up->u_tun_ctx = ctx;
1807 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1810 struct inpcbinfo *pcbinfo;
1813 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1814 inp = sotoinpcb(so);
1815 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1817 INP_HASH_WLOCK(pcbinfo);
1818 error = in_pcbbind(inp, nam, td->td_ucred);
1819 INP_HASH_WUNLOCK(pcbinfo);
1825 udp_close(struct socket *so)
1828 struct inpcbinfo *pcbinfo;
1830 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1831 inp = sotoinpcb(so);
1832 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1834 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1835 INP_HASH_WLOCK(pcbinfo);
1836 in_pcbdisconnect(inp);
1837 inp->inp_laddr.s_addr = INADDR_ANY;
1838 INP_HASH_WUNLOCK(pcbinfo);
1839 soisdisconnected(so);
1845 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1848 struct inpcbinfo *pcbinfo;
1849 struct sockaddr_in *sin;
1852 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1853 inp = sotoinpcb(so);
1854 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1856 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1860 sin = (struct sockaddr_in *)nam;
1861 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1866 INP_HASH_WLOCK(pcbinfo);
1867 error = in_pcbconnect(inp, nam, td->td_ucred);
1868 INP_HASH_WUNLOCK(pcbinfo);
1876 udp_detach(struct socket *so)
1879 struct inpcbinfo *pcbinfo;
1882 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1883 inp = sotoinpcb(so);
1884 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1885 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1886 ("udp_detach: not disconnected"));
1887 INP_INFO_WLOCK(pcbinfo);
1889 up = intoudpcb(inp);
1890 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1891 inp->inp_ppcb = NULL;
1894 INP_INFO_WUNLOCK(pcbinfo);
1899 udp_disconnect(struct socket *so)
1902 struct inpcbinfo *pcbinfo;
1904 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1905 inp = sotoinpcb(so);
1906 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1908 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1912 INP_HASH_WLOCK(pcbinfo);
1913 in_pcbdisconnect(inp);
1914 inp->inp_laddr.s_addr = INADDR_ANY;
1915 INP_HASH_WUNLOCK(pcbinfo);
1917 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1924 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1925 struct mbuf *control, struct thread *td)
1929 inp = sotoinpcb(so);
1930 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1931 return (udp_output(inp, m, addr, control, td));
1936 udp_shutdown(struct socket *so)
1940 inp = sotoinpcb(so);
1941 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1949 struct pr_usrreqs udp_usrreqs = {
1950 .pru_abort = udp_abort,
1951 .pru_attach = udp_attach,
1952 .pru_bind = udp_bind,
1953 .pru_connect = udp_connect,
1954 .pru_control = in_control,
1955 .pru_detach = udp_detach,
1956 .pru_disconnect = udp_disconnect,
1957 .pru_peeraddr = in_getpeeraddr,
1958 .pru_send = udp_send,
1959 .pru_soreceive = soreceive_dgram,
1960 .pru_sosend = sosend_dgram,
1961 .pru_shutdown = udp_shutdown,
1962 .pru_sockaddr = in_getsockaddr,
1963 .pru_sosetlabel = in_pcbsosetlabel,
1964 .pru_close = udp_close,