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 u_long udp_sendspace = 9216; /* really max datagram size */
130 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
131 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
133 u_long udp_recvspace = 40 * (1024 +
135 sizeof(struct sockaddr_in6)
137 sizeof(struct sockaddr_in)
139 ); /* 40 1K datagrams */
141 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
142 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
144 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
145 VNET_DEFINE(struct inpcbinfo, udbinfo);
146 VNET_DEFINE(struct inpcbhead, ulitecb);
147 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
148 static VNET_DEFINE(uma_zone_t, udpcb_zone);
149 #define V_udpcb_zone VNET(udpcb_zone)
152 #define UDBHASHSIZE 128
155 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
156 VNET_PCPUSTAT_SYSINIT(udpstat);
157 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
158 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
161 VNET_PCPUSTAT_SYSUNINIT(udpstat);
164 static void udp_detach(struct socket *so);
165 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
166 struct mbuf *, struct thread *);
171 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
173 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
175 #endif /* IPSEC_NAT_T */
179 udp_zone_change(void *tag)
182 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
183 uma_zone_set_max(V_udpcb_zone, maxsockets);
187 udp_inpcb_init(void *mem, int size, int flags)
192 INP_LOCK_INIT(inp, "inp", "udpinp");
197 udplite_inpcb_init(void *mem, int size, int flags)
202 INP_LOCK_INIT(inp, "inp", "udpliteinp");
211 * For now default to 2-tuple UDP hashing - until the fragment
212 * reassembly code can also update the flowid.
214 * Once we can calculate the flowid that way and re-establish
215 * a 4-tuple, flip this to 4-tuple.
217 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
218 "udp_inpcb", udp_inpcb_init, NULL, 0,
219 IPI_HASHFIELDS_2TUPLE);
220 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
221 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
222 uma_zone_set_max(V_udpcb_zone, maxsockets);
223 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
224 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
225 EVENTHANDLER_PRI_ANY);
232 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
233 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
234 0, IPI_HASHFIELDS_2TUPLE);
238 * Kernel module interface for updating udpstat. The argument is an index
239 * into udpstat treated as an array of u_long. While this encodes the
240 * general layout of udpstat into the caller, it doesn't encode its location,
241 * so that future changes to add, for example, per-CPU stats support won't
242 * cause binary compatibility problems for kernel modules.
245 kmod_udpstat_inc(int statnum)
248 counter_u64_add(VNET(udpstat)[statnum], 1);
252 udp_newudpcb(struct inpcb *inp)
256 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
264 udp_discardcb(struct udpcb *up)
267 uma_zfree(V_udpcb_zone, up);
272 udp_destroy(void *unused __unused)
275 in_pcbinfo_destroy(&V_udbinfo);
276 uma_zdestroy(V_udpcb_zone);
278 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
281 udplite_destroy(void *unused __unused)
284 in_pcbinfo_destroy(&V_ulitecbinfo);
286 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
292 * Subroutine of udp_input(), which appends the provided mbuf chain to the
293 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
294 * contains the source address. If the socket ends up being an IPv6 socket,
295 * udp_append() will convert to a sockaddr_in6 before passing the address
296 * into the socket code.
298 * In the normal case udp_append() will return 0, indicating that you
299 * must unlock the inp. However if a tunneling protocol is in place we increment
300 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
301 * then decrement the reference count. If the inp_rele returns 1, indicating the
302 * inp is gone, we return that to the caller to tell them *not* to unlock
303 * the inp. In the case of multi-cast this will cause the distribution
304 * to stop (though most tunneling protocols known currently do *not* use
308 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
309 struct sockaddr_in *udp_in)
311 struct sockaddr *append_sa;
313 struct mbuf *opts = NULL;
315 struct sockaddr_in6 udp_in6;
319 INP_LOCK_ASSERT(inp);
322 * Engage the tunneling protocol.
325 if (up->u_tun_func != NULL) {
328 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
331 return (in_pcbrele_rlocked(inp));
334 off += sizeof(struct udphdr);
337 /* Check AH/ESP integrity. */
338 if (ipsec4_in_reject(n, inp)) {
344 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
345 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
346 n = udp4_espdecap(inp, n, off);
347 if (n == NULL) /* Consumed. */
350 #endif /* IPSEC_NAT_T */
353 if (mac_inpcb_check_deliver(inp, n) != 0) {
358 if (inp->inp_flags & INP_CONTROLOPTS ||
359 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
361 if (inp->inp_vflag & INP_IPV6)
362 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
365 ip_savecontrol(inp, &opts, ip, n);
368 if (inp->inp_vflag & INP_IPV6) {
369 bzero(&udp_in6, sizeof(udp_in6));
370 udp_in6.sin6_len = sizeof(udp_in6);
371 udp_in6.sin6_family = AF_INET6;
372 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
373 append_sa = (struct sockaddr *)&udp_in6;
376 append_sa = (struct sockaddr *)udp_in;
379 so = inp->inp_socket;
380 SOCKBUF_LOCK(&so->so_rcv);
381 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
382 SOCKBUF_UNLOCK(&so->so_rcv);
386 UDPSTAT_INC(udps_fullsock);
388 sorwakeup_locked(so);
393 udp_input(struct mbuf **mp, int *offp, int proto)
399 uint16_t len, ip_len;
400 struct inpcbinfo *pcbinfo;
402 struct sockaddr_in udp_in;
404 struct m_tag *fwd_tag;
405 int cscov_partial, iphlen;
409 ifp = m->m_pkthdr.rcvif;
411 UDPSTAT_INC(udps_ipackets);
414 * Strip IP options, if any; should skip this, make available to
415 * user, and use on returned packets, but we don't yet have a way to
416 * check the checksum with options still present.
418 if (iphlen > sizeof (struct ip)) {
420 iphlen = sizeof(struct ip);
424 * Get IP and UDP header together in first mbuf.
426 ip = mtod(m, struct ip *);
427 if (m->m_len < iphlen + sizeof(struct udphdr)) {
428 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
429 UDPSTAT_INC(udps_hdrops);
430 return (IPPROTO_DONE);
432 ip = mtod(m, struct ip *);
434 uh = (struct udphdr *)((caddr_t)ip + iphlen);
435 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
438 * Destination port of 0 is illegal, based on RFC768.
440 if (uh->uh_dport == 0)
444 * Construct sockaddr format source address. Stuff source address
445 * and datagram in user buffer.
447 bzero(&udp_in, sizeof(udp_in));
448 udp_in.sin_len = sizeof(udp_in);
449 udp_in.sin_family = AF_INET;
450 udp_in.sin_port = uh->uh_sport;
451 udp_in.sin_addr = ip->ip_src;
454 * Make mbuf data length reflect UDP length. If not enough data to
455 * reflect UDP length, drop.
457 len = ntohs((u_short)uh->uh_ulen);
458 ip_len = ntohs(ip->ip_len) - iphlen;
459 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
460 /* Zero means checksum over the complete packet. */
466 if (len > ip_len || len < sizeof(struct udphdr)) {
467 UDPSTAT_INC(udps_badlen);
470 if (proto == IPPROTO_UDP)
471 m_adj(m, len - ip_len);
475 * Save a copy of the IP header in case we want restore it for
476 * sending an ICMP error message in response.
478 if (!V_udp_blackhole)
481 memset(&save_ip, 0, sizeof(save_ip));
484 * Checksum extended UDP header and data.
489 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
491 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
492 uh_sum = m->m_pkthdr.csum_data;
494 uh_sum = in_pseudo(ip->ip_src.s_addr,
495 ip->ip_dst.s_addr, htonl((u_short)len +
496 m->m_pkthdr.csum_data + proto));
501 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
502 bzero(((struct ipovly *)ip)->ih_x1, 9);
503 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
504 uh->uh_ulen : htons(ip_len);
505 uh_sum = in_cksum(m, len + sizeof (struct ip));
506 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
509 UDPSTAT_INC(udps_badsum);
511 return (IPPROTO_DONE);
514 if (proto == IPPROTO_UDP) {
515 UDPSTAT_INC(udps_nosum);
517 /* UDPLite requires a checksum */
518 /* XXX: What is the right UDPLite MIB counter here? */
520 return (IPPROTO_DONE);
524 pcbinfo = udp_get_inpcbinfo(proto);
525 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
526 in_broadcast(ip->ip_dst, ifp)) {
528 struct inpcbhead *pcblist;
529 struct ip_moptions *imo;
531 INP_INFO_RLOCK(pcbinfo);
532 pcblist = udp_get_pcblist(proto);
534 LIST_FOREACH(inp, pcblist, inp_list) {
535 if (inp->inp_lport != uh->uh_dport)
538 if ((inp->inp_vflag & INP_IPV4) == 0)
541 if (inp->inp_laddr.s_addr != INADDR_ANY &&
542 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
544 if (inp->inp_faddr.s_addr != INADDR_ANY &&
545 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
547 if (inp->inp_fport != 0 &&
548 inp->inp_fport != uh->uh_sport)
554 * XXXRW: Because we weren't holding either the inpcb
555 * or the hash lock when we checked for a match
556 * before, we should probably recheck now that the
557 * inpcb lock is held.
561 * Handle socket delivery policy for any-source
562 * and source-specific multicast. [RFC3678]
564 imo = inp->inp_moptions;
565 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
566 struct sockaddr_in group;
572 bzero(&group, sizeof(struct sockaddr_in));
573 group.sin_len = sizeof(struct sockaddr_in);
574 group.sin_family = AF_INET;
575 group.sin_addr = ip->ip_dst;
577 blocked = imo_multi_filter(imo, ifp,
578 (struct sockaddr *)&group,
579 (struct sockaddr *)&udp_in);
580 if (blocked != MCAST_PASS) {
581 if (blocked == MCAST_NOTGMEMBER)
582 IPSTAT_INC(ips_notmember);
583 if (blocked == MCAST_NOTSMEMBER ||
584 blocked == MCAST_MUTED)
585 UDPSTAT_INC(udps_filtermcast);
593 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
594 UDP_PROBE(receive, NULL, last, ip,
596 if (udp_append(last, ip, n, iphlen,
605 * Don't look for additional matches if this one does
606 * not have either the SO_REUSEPORT or SO_REUSEADDR
607 * socket options set. This heuristic avoids
608 * searching through all pcbs in the common case of a
609 * non-shared port. It assumes that an application
610 * will never clear these options after setting them.
612 if ((last->inp_socket->so_options &
613 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
619 * No matching pcb found; discard datagram. (No need
620 * to send an ICMP Port Unreachable for a broadcast
621 * or multicast datgram.)
623 UDPSTAT_INC(udps_noportbcast);
626 INP_INFO_RUNLOCK(pcbinfo);
629 UDP_PROBE(receive, NULL, last, ip, last, uh);
630 if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
633 INP_INFO_RUNLOCK(pcbinfo);
634 return (IPPROTO_DONE);
638 * Locate pcb for datagram.
642 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
644 if ((m->m_flags & M_IP_NEXTHOP) &&
645 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
646 struct sockaddr_in *next_hop;
648 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
651 * Transparently forwarded. Pretend to be the destination.
652 * Already got one like this?
654 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
655 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
658 * It's new. Try to find the ambushing socket.
659 * Because we've rewritten the destination address,
660 * any hardware-generated hash is ignored.
662 inp = in_pcblookup(pcbinfo, ip->ip_src,
663 uh->uh_sport, next_hop->sin_addr,
664 next_hop->sin_port ? htons(next_hop->sin_port) :
665 uh->uh_dport, INPLOOKUP_WILDCARD |
666 INPLOOKUP_RLOCKPCB, ifp);
668 /* Remove the tag from the packet. We don't need it anymore. */
669 m_tag_delete(m, fwd_tag);
670 m->m_flags &= ~M_IP_NEXTHOP;
672 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
673 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
674 INPLOOKUP_RLOCKPCB, ifp, m);
676 if (udp_log_in_vain) {
677 char buf[4*sizeof "123"];
679 strcpy(buf, inet_ntoa(ip->ip_dst));
681 "Connection attempt to UDP %s:%d from %s:%d\n",
682 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
683 ntohs(uh->uh_sport));
685 UDPSTAT_INC(udps_noport);
686 if (m->m_flags & (M_BCAST | M_MCAST)) {
687 UDPSTAT_INC(udps_noportbcast);
692 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
695 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
696 return (IPPROTO_DONE);
700 * Check the minimum TTL for socket.
702 INP_RLOCK_ASSERT(inp);
703 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
706 return (IPPROTO_DONE);
712 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
715 return (IPPROTO_DONE);
719 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
720 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
722 return (IPPROTO_DONE);
726 return (IPPROTO_DONE);
731 * Notify a udp user of an asynchronous error; just wake up so that they can
732 * collect error status.
735 udp_notify(struct inpcb *inp, int errno)
739 * While udp_ctlinput() always calls udp_notify() with a read lock
740 * when invoking it directly, in_pcbnotifyall() currently uses write
741 * locks due to sharing code with TCP. For now, accept either a read
742 * or a write lock, but a read lock is sufficient.
744 INP_LOCK_ASSERT(inp);
745 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
746 errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
747 RTFREE(inp->inp_route.ro_rt);
748 inp->inp_route.ro_rt = (struct rtentry *)NULL;
751 inp->inp_socket->so_error = errno;
752 sorwakeup(inp->inp_socket);
753 sowwakeup(inp->inp_socket);
759 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
760 struct inpcbinfo *pcbinfo)
764 struct in_addr faddr;
767 faddr = ((struct sockaddr_in *)sa)->sin_addr;
768 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
771 if (PRC_IS_REDIRECT(cmd)) {
772 /* signal EHOSTDOWN, as it flushes the cached route */
773 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
778 * Hostdead is ugly because it goes linearly through all PCBs.
780 * XXX: We never get this from ICMP, otherwise it makes an excellent
781 * DoS attack on machines with many connections.
783 if (cmd == PRC_HOSTDEAD)
785 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
788 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
789 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
790 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
792 INP_RLOCK_ASSERT(inp);
793 if (inp->inp_socket != NULL) {
794 udp_notify(inp, inetctlerrmap[cmd]);
798 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
799 ip->ip_src, uh->uh_sport,
800 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
805 if (up->u_icmp_func != NULL) {
807 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
814 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
818 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
821 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
825 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
828 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
833 udp_pcblist(SYSCTL_HANDLER_ARGS)
836 struct inpcb *inp, **inp_list;
841 * The process of preparing the PCB list is too time-consuming and
842 * resource-intensive to repeat twice on every request.
844 if (req->oldptr == 0) {
845 n = V_udbinfo.ipi_count;
846 n += imax(n / 8, 10);
847 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
851 if (req->newptr != 0)
855 * OK, now we're committed to doing something.
857 INP_INFO_RLOCK(&V_udbinfo);
858 gencnt = V_udbinfo.ipi_gencnt;
859 n = V_udbinfo.ipi_count;
860 INP_INFO_RUNLOCK(&V_udbinfo);
862 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
863 + n * sizeof(struct xinpcb));
867 xig.xig_len = sizeof xig;
869 xig.xig_gen = gencnt;
870 xig.xig_sogen = so_gencnt;
871 error = SYSCTL_OUT(req, &xig, sizeof xig);
875 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
876 if (inp_list == NULL)
879 INP_INFO_RLOCK(&V_udbinfo);
880 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
881 inp = LIST_NEXT(inp, inp_list)) {
883 if (inp->inp_gencnt <= gencnt &&
884 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
890 INP_INFO_RUNLOCK(&V_udbinfo);
894 for (i = 0; i < n; i++) {
897 if (inp->inp_gencnt <= gencnt) {
900 bzero(&xi, sizeof(xi));
901 xi.xi_len = sizeof xi;
902 /* XXX should avoid extra copy */
903 bcopy(inp, &xi.xi_inp, sizeof *inp);
905 sotoxsocket(inp->inp_socket, &xi.xi_socket);
906 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
908 error = SYSCTL_OUT(req, &xi, sizeof xi);
912 INP_INFO_WLOCK(&V_udbinfo);
913 for (i = 0; i < n; i++) {
916 if (!in_pcbrele_rlocked(inp))
919 INP_INFO_WUNLOCK(&V_udbinfo);
923 * Give the user an updated idea of our state. If the
924 * generation differs from what we told her before, she knows
925 * that something happened while we were processing this
926 * request, and it might be necessary to retry.
928 INP_INFO_RLOCK(&V_udbinfo);
929 xig.xig_gen = V_udbinfo.ipi_gencnt;
930 xig.xig_sogen = so_gencnt;
931 xig.xig_count = V_udbinfo.ipi_count;
932 INP_INFO_RUNLOCK(&V_udbinfo);
933 error = SYSCTL_OUT(req, &xig, sizeof xig);
935 free(inp_list, M_TEMP);
939 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
940 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
941 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
945 udp_getcred(SYSCTL_HANDLER_ARGS)
948 struct sockaddr_in addrs[2];
952 error = priv_check(req->td, PRIV_NETINET_GETCRED);
955 error = SYSCTL_IN(req, addrs, sizeof(addrs));
958 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
959 addrs[0].sin_addr, addrs[0].sin_port,
960 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
962 INP_RLOCK_ASSERT(inp);
963 if (inp->inp_socket == NULL)
966 error = cr_canseeinpcb(req->td->td_ucred, inp);
968 cru2x(inp->inp_cred, &xuc);
973 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
977 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
978 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
979 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
983 udp_ctloutput(struct socket *so, struct sockopt *sopt)
987 int isudplite, error, optval;
990 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
992 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
994 if (sopt->sopt_level != so->so_proto->pr_protocol) {
996 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
998 error = ip6_ctloutput(so, sopt);
1001 #if defined(INET) && defined(INET6)
1007 error = ip_ctloutput(so, sopt);
1013 switch (sopt->sopt_dir) {
1015 switch (sopt->sopt_name) {
1018 error = sooptcopyin(sopt, &optval, sizeof optval,
1022 inp = sotoinpcb(so);
1023 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1026 up = intoudpcb(inp);
1027 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1031 /* Clear all UDP encap. */
1033 up->u_flags &= ~UF_ESPINUDP_ALL;
1037 case UDP_ENCAP_ESPINUDP:
1038 case UDP_ENCAP_ESPINUDP_NON_IKE:
1039 up->u_flags &= ~UF_ESPINUDP_ALL;
1040 if (optval == UDP_ENCAP_ESPINUDP)
1041 up->u_flags |= UF_ESPINUDP;
1042 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1043 up->u_flags |= UF_ESPINUDP_NON_IKE;
1052 case UDPLITE_SEND_CSCOV:
1053 case UDPLITE_RECV_CSCOV:
1056 error = ENOPROTOOPT;
1060 error = sooptcopyin(sopt, &optval, sizeof(optval),
1064 inp = sotoinpcb(so);
1065 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1067 up = intoudpcb(inp);
1068 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1069 if ((optval != 0 && optval < 8) || (optval > 65535)) {
1074 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1075 up->u_txcslen = optval;
1077 up->u_rxcslen = optval;
1082 error = ENOPROTOOPT;
1087 switch (sopt->sopt_name) {
1090 up = intoudpcb(inp);
1091 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1092 optval = up->u_flags & UF_ESPINUDP_ALL;
1094 error = sooptcopyout(sopt, &optval, sizeof optval);
1097 case UDPLITE_SEND_CSCOV:
1098 case UDPLITE_RECV_CSCOV:
1101 error = ENOPROTOOPT;
1104 up = intoudpcb(inp);
1105 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1106 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1107 optval = up->u_txcslen;
1109 optval = up->u_rxcslen;
1111 error = sooptcopyout(sopt, &optval, sizeof(optval));
1115 error = ENOPROTOOPT;
1124 #define UH_WLOCKED 2
1125 #define UH_RLOCKED 1
1126 #define UH_UNLOCKED 0
1128 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1129 struct mbuf *control, struct thread *td)
1131 struct udpiphdr *ui;
1132 int len = m->m_pkthdr.len;
1133 struct in_addr faddr, laddr;
1135 struct inpcbinfo *pcbinfo;
1136 struct sockaddr_in *sin, src;
1137 int cscov_partial = 0;
1140 u_short fport, lport;
1141 int unlock_udbinfo, unlock_inp;
1145 uint32_t flowid = 0;
1146 uint8_t flowtype = M_HASHTYPE_NONE;
1149 * udp_output() may need to temporarily bind or connect the current
1150 * inpcb. As such, we don't know up front whether we will need the
1151 * pcbinfo lock or not. Do any work to decide what is needed up
1152 * front before acquiring any locks.
1154 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1162 sin = (struct sockaddr_in *)addr;
1164 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1166 unlock_inp = UH_WLOCKED;
1169 unlock_inp = UH_RLOCKED;
1171 tos = inp->inp_ip_tos;
1172 if (control != NULL) {
1174 * XXX: Currently, we assume all the optional information is
1175 * stored in a single mbuf.
1177 if (control->m_next) {
1178 if (unlock_inp == UH_WLOCKED)
1186 for (; control->m_len > 0;
1187 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1188 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1189 cm = mtod(control, struct cmsghdr *);
1190 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1191 || cm->cmsg_len > control->m_len) {
1195 if (cm->cmsg_level != IPPROTO_IP)
1198 switch (cm->cmsg_type) {
1199 case IP_SENDSRCADDR:
1201 CMSG_LEN(sizeof(struct in_addr))) {
1205 bzero(&src, sizeof(src));
1206 src.sin_family = AF_INET;
1207 src.sin_len = sizeof(src);
1208 src.sin_port = inp->inp_lport;
1210 *(struct in_addr *)CMSG_DATA(cm);
1214 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1218 tos = *(u_char *)CMSG_DATA(cm);
1222 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1226 flowid = *(uint32_t *) CMSG_DATA(cm);
1230 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1234 flowtype = *(uint32_t *) CMSG_DATA(cm);
1238 case IP_RSSBUCKETID:
1239 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1243 /* This is just a placeholder for now */
1247 error = ENOPROTOOPT;
1256 if (unlock_inp == UH_WLOCKED)
1265 * Depending on whether or not the application has bound or connected
1266 * the socket, we may have to do varying levels of work. The optimal
1267 * case is for a connected UDP socket, as a global lock isn't
1270 * In order to decide which we need, we require stability of the
1271 * inpcb binding, which we ensure by acquiring a read lock on the
1272 * inpcb. This doesn't strictly follow the lock order, so we play
1273 * the trylock and retry game; note that we may end up with more
1274 * conservative locks than required the second time around, so later
1275 * assertions have to accept that. Further analysis of the number of
1276 * misses under contention is required.
1278 * XXXRW: Check that hash locking update here is correct.
1280 pr = inp->inp_socket->so_proto->pr_protocol;
1281 pcbinfo = udp_get_inpcbinfo(pr);
1282 sin = (struct sockaddr_in *)addr;
1284 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1285 INP_HASH_WLOCK(pcbinfo);
1286 unlock_udbinfo = UH_WLOCKED;
1287 } else if ((sin != NULL && (
1288 (sin->sin_addr.s_addr == INADDR_ANY) ||
1289 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1290 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1291 (inp->inp_lport == 0))) ||
1292 (src.sin_family == AF_INET)) {
1293 INP_HASH_RLOCK(pcbinfo);
1294 unlock_udbinfo = UH_RLOCKED;
1296 unlock_udbinfo = UH_UNLOCKED;
1299 * If the IP_SENDSRCADDR control message was specified, override the
1300 * source address for this datagram. Its use is invalidated if the
1301 * address thus specified is incomplete or clobbers other inpcbs.
1303 laddr = inp->inp_laddr;
1304 lport = inp->inp_lport;
1305 if (src.sin_family == AF_INET) {
1306 INP_HASH_LOCK_ASSERT(pcbinfo);
1308 (laddr.s_addr == INADDR_ANY &&
1309 src.sin_addr.s_addr == INADDR_ANY)) {
1313 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1314 &laddr.s_addr, &lport, td->td_ucred);
1320 * If a UDP socket has been connected, then a local address/port will
1321 * have been selected and bound.
1323 * If a UDP socket has not been connected to, then an explicit
1324 * destination address must be used, in which case a local
1325 * address/port may not have been selected and bound.
1328 INP_LOCK_ASSERT(inp);
1329 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1335 * Jail may rewrite the destination address, so let it do
1336 * that before we use it.
1338 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1343 * If a local address or port hasn't yet been selected, or if
1344 * the destination address needs to be rewritten due to using
1345 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1346 * to do the heavy lifting. Once a port is selected, we
1347 * commit the binding back to the socket; we also commit the
1348 * binding of the address if in jail.
1350 * If we already have a valid binding and we're not
1351 * requesting a destination address rewrite, use a fast path.
1353 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1354 inp->inp_lport == 0 ||
1355 sin->sin_addr.s_addr == INADDR_ANY ||
1356 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1357 INP_HASH_LOCK_ASSERT(pcbinfo);
1358 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1359 &lport, &faddr.s_addr, &fport, NULL,
1365 * XXXRW: Why not commit the port if the address is
1368 /* Commit the local port if newly assigned. */
1369 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1370 inp->inp_lport == 0) {
1371 INP_WLOCK_ASSERT(inp);
1372 INP_HASH_WLOCK_ASSERT(pcbinfo);
1374 * Remember addr if jailed, to prevent
1377 if (prison_flag(td->td_ucred, PR_IP4))
1378 inp->inp_laddr = laddr;
1379 inp->inp_lport = lport;
1380 if (in_pcbinshash(inp) != 0) {
1385 inp->inp_flags |= INP_ANONPORT;
1388 faddr = sin->sin_addr;
1389 fport = sin->sin_port;
1392 INP_LOCK_ASSERT(inp);
1393 faddr = inp->inp_faddr;
1394 fport = inp->inp_fport;
1395 if (faddr.s_addr == INADDR_ANY) {
1402 * Calculate data length and get a mbuf for UDP, IP, and possible
1403 * link-layer headers. Immediate slide the data pointer back forward
1404 * since we won't use that space at this layer.
1406 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1411 m->m_data += max_linkhdr;
1412 m->m_len -= max_linkhdr;
1413 m->m_pkthdr.len -= max_linkhdr;
1416 * Fill in mbuf with extended UDP header and addresses and length put
1417 * into network format.
1419 ui = mtod(m, struct udpiphdr *);
1420 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1424 ui->ui_sport = lport;
1425 ui->ui_dport = fport;
1426 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1427 if (pr == IPPROTO_UDPLITE) {
1431 up = intoudpcb(inp);
1432 cscov = up->u_txcslen;
1433 plen = (u_short)len + sizeof(struct udphdr);
1436 ui->ui_len = htons(plen);
1437 ui->ui_ulen = htons(cscov);
1439 * For UDP-Lite, checksum coverage length of zero means
1440 * the entire UDPLite packet is covered by the checksum.
1442 cscov_partial = (cscov == 0) ? 0 : 1;
1444 ui->ui_v = IPVERSION << 4;
1447 * Set the Don't Fragment bit in the IP header.
1449 if (inp->inp_flags & INP_DONTFRAG) {
1452 ip = (struct ip *)&ui->ui_i;
1453 ip->ip_off |= htons(IP_DF);
1457 if (inp->inp_socket->so_options & SO_DONTROUTE)
1458 ipflags |= IP_ROUTETOIF;
1459 if (inp->inp_socket->so_options & SO_BROADCAST)
1460 ipflags |= IP_ALLOWBROADCAST;
1461 if (inp->inp_flags & INP_ONESBCAST)
1462 ipflags |= IP_SENDONES;
1465 mac_inpcb_create_mbuf(inp, m);
1469 * Set up checksum and output datagram.
1472 if (pr == IPPROTO_UDPLITE) {
1473 if (inp->inp_flags & INP_ONESBCAST)
1474 faddr.s_addr = INADDR_BROADCAST;
1475 if (cscov_partial) {
1476 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1477 ui->ui_sum = 0xffff;
1479 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1480 ui->ui_sum = 0xffff;
1482 } else if (V_udp_cksum) {
1483 if (inp->inp_flags & INP_ONESBCAST)
1484 faddr.s_addr = INADDR_BROADCAST;
1485 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1486 htons((u_short)len + sizeof(struct udphdr) + pr));
1487 m->m_pkthdr.csum_flags = CSUM_UDP;
1488 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1490 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1491 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1492 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1493 UDPSTAT_INC(udps_opackets);
1496 * Setup flowid / RSS information for outbound socket.
1498 * Once the UDP code decides to set a flowid some other way,
1499 * this allows the flowid to be overridden by userland.
1501 if (flowtype != M_HASHTYPE_NONE) {
1502 m->m_pkthdr.flowid = flowid;
1503 M_HASHTYPE_SET(m, flowtype);
1506 uint32_t hash_val, hash_type;
1508 * Calculate an appropriate RSS hash for UDP and
1511 * The called function will take care of figuring out
1512 * whether a 2-tuple or 4-tuple hash is required based
1513 * on the currently configured scheme.
1515 * Later later on connected socket values should be
1516 * cached in the inpcb and reused, rather than constantly
1517 * re-calculating it.
1519 * UDP Lite is a different protocol number and will
1520 * likely end up being hashed as a 2-tuple until
1521 * RSS / NICs grow UDP Lite protocol awareness.
1523 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1524 pr, &hash_val, &hash_type) == 0) {
1525 m->m_pkthdr.flowid = hash_val;
1526 M_HASHTYPE_SET(m, hash_type);
1533 * Don't override with the inp cached flowid value.
1535 * Depending upon the kind of send being done, the inp
1536 * flowid/flowtype values may actually not be appropriate
1537 * for this particular socket send.
1539 * We should either leave the flowid at zero (which is what is
1540 * currently done) or set it to some software generated
1541 * hash value based on the packet contents.
1543 ipflags |= IP_NODEFAULTFLOWID;
1546 if (unlock_udbinfo == UH_WLOCKED)
1547 INP_HASH_WUNLOCK(pcbinfo);
1548 else if (unlock_udbinfo == UH_RLOCKED)
1549 INP_HASH_RUNLOCK(pcbinfo);
1550 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1551 error = ip_output(m, inp->inp_options,
1552 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1553 inp->inp_moptions, inp);
1554 if (unlock_inp == UH_WLOCKED)
1561 if (unlock_udbinfo == UH_WLOCKED) {
1562 INP_HASH_WUNLOCK(pcbinfo);
1564 } else if (unlock_udbinfo == UH_RLOCKED) {
1565 INP_HASH_RUNLOCK(pcbinfo);
1574 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1576 * Potentially decap ESP in UDP frame. Check for an ESP header
1577 * and optional marker; if present, strip the UDP header and
1578 * push the result through IPSec.
1580 * Returns mbuf to be processed (potentially re-allocated) or
1581 * NULL if consumed and/or processed.
1583 static struct mbuf *
1584 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1586 size_t minlen, payload, skip, iphlen;
1590 struct udphdr *udphdr;
1593 INP_RLOCK_ASSERT(inp);
1596 * Pull up data so the longest case is contiguous:
1597 * IP/UDP hdr + non ESP marker + ESP hdr.
1599 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1600 if (minlen > m->m_pkthdr.len)
1601 minlen = m->m_pkthdr.len;
1602 if ((m = m_pullup(m, minlen)) == NULL) {
1603 IPSECSTAT_INC(ips_in_inval);
1604 return (NULL); /* Bypass caller processing. */
1606 data = mtod(m, caddr_t); /* Points to ip header. */
1607 payload = m->m_len - off; /* Size of payload. */
1609 if (payload == 1 && data[off] == '\xff')
1610 return (m); /* NB: keepalive packet, no decap. */
1612 up = intoudpcb(inp);
1613 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1614 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1615 ("u_flags 0x%x", up->u_flags));
1618 * Check that the payload is large enough to hold an
1619 * ESP header and compute the amount of data to remove.
1621 * NB: the caller has already done a pullup for us.
1622 * XXX can we assume alignment and eliminate bcopys?
1624 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1626 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1627 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1628 * possible AH mode non-IKE marker+non-ESP marker
1629 * from draft-ietf-ipsec-udp-encaps-00.txt.
1633 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1634 return (m); /* NB: no decap. */
1635 bcopy(data + off, &marker, sizeof(uint64_t));
1636 if (marker != 0) /* Non-IKE marker. */
1637 return (m); /* NB: no decap. */
1638 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1642 if (payload <= sizeof(struct esp)) {
1643 IPSECSTAT_INC(ips_in_inval);
1645 return (NULL); /* Discard. */
1647 bcopy(data + off, &spi, sizeof(uint32_t));
1648 if (spi == 0) /* Non-ESP marker. */
1649 return (m); /* NB: no decap. */
1650 skip = sizeof(struct udphdr);
1654 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1655 * the UDP ports. This is required if we want to select
1656 * the right SPD for multiple hosts behind same NAT.
1658 * NB: ports are maintained in network byte order everywhere
1659 * in the NAT-T code.
1661 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1662 2 * sizeof(uint16_t), M_NOWAIT);
1664 IPSECSTAT_INC(ips_in_nomem);
1666 return (NULL); /* Discard. */
1668 iphlen = off - sizeof(struct udphdr);
1669 udphdr = (struct udphdr *)(data + iphlen);
1670 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1671 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1672 m_tag_prepend(m, tag);
1675 * Remove the UDP header (and possibly the non ESP marker)
1676 * IP header length is iphlen
1679 * +----+------+-----+
1680 * | IP | UDP | ESP |
1681 * +----+------+-----+
1689 ovbcopy(data, data + skip, iphlen);
1692 ip = mtod(m, struct ip *);
1693 ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1694 ip->ip_p = IPPROTO_ESP;
1697 * We cannot yet update the cksums so clear any
1698 * h/w cksum flags as they are no longer valid.
1700 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1701 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1703 (void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1705 return (NULL); /* NB: consumed, bypass processing. */
1707 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1710 udp_abort(struct socket *so)
1713 struct inpcbinfo *pcbinfo;
1715 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1716 inp = sotoinpcb(so);
1717 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1719 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1720 INP_HASH_WLOCK(pcbinfo);
1721 in_pcbdisconnect(inp);
1722 inp->inp_laddr.s_addr = INADDR_ANY;
1723 INP_HASH_WUNLOCK(pcbinfo);
1724 soisdisconnected(so);
1730 udp_attach(struct socket *so, int proto, struct thread *td)
1733 struct inpcbinfo *pcbinfo;
1736 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1737 inp = sotoinpcb(so);
1738 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1739 error = soreserve(so, udp_sendspace, udp_recvspace);
1742 INP_INFO_WLOCK(pcbinfo);
1743 error = in_pcballoc(so, pcbinfo);
1745 INP_INFO_WUNLOCK(pcbinfo);
1749 inp = sotoinpcb(so);
1750 inp->inp_vflag |= INP_IPV4;
1751 inp->inp_ip_ttl = V_ip_defttl;
1753 error = udp_newudpcb(inp);
1757 INP_INFO_WUNLOCK(pcbinfo);
1762 INP_INFO_WUNLOCK(pcbinfo);
1768 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1773 KASSERT(so->so_type == SOCK_DGRAM,
1774 ("udp_set_kernel_tunneling: !dgram"));
1775 inp = sotoinpcb(so);
1776 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1778 up = intoudpcb(inp);
1779 if ((up->u_tun_func != NULL) ||
1780 (up->u_icmp_func != NULL)) {
1785 up->u_icmp_func = i;
1786 up->u_tun_ctx = ctx;
1793 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1796 struct inpcbinfo *pcbinfo;
1799 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1800 inp = sotoinpcb(so);
1801 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1803 INP_HASH_WLOCK(pcbinfo);
1804 error = in_pcbbind(inp, nam, td->td_ucred);
1805 INP_HASH_WUNLOCK(pcbinfo);
1811 udp_close(struct socket *so)
1814 struct inpcbinfo *pcbinfo;
1816 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1817 inp = sotoinpcb(so);
1818 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1820 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1821 INP_HASH_WLOCK(pcbinfo);
1822 in_pcbdisconnect(inp);
1823 inp->inp_laddr.s_addr = INADDR_ANY;
1824 INP_HASH_WUNLOCK(pcbinfo);
1825 soisdisconnected(so);
1831 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1834 struct inpcbinfo *pcbinfo;
1835 struct sockaddr_in *sin;
1838 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1839 inp = sotoinpcb(so);
1840 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1842 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1846 sin = (struct sockaddr_in *)nam;
1847 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1852 INP_HASH_WLOCK(pcbinfo);
1853 error = in_pcbconnect(inp, nam, td->td_ucred);
1854 INP_HASH_WUNLOCK(pcbinfo);
1862 udp_detach(struct socket *so)
1865 struct inpcbinfo *pcbinfo;
1868 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1869 inp = sotoinpcb(so);
1870 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1871 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1872 ("udp_detach: not disconnected"));
1873 INP_INFO_WLOCK(pcbinfo);
1875 up = intoudpcb(inp);
1876 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1877 inp->inp_ppcb = NULL;
1880 INP_INFO_WUNLOCK(pcbinfo);
1885 udp_disconnect(struct socket *so)
1888 struct inpcbinfo *pcbinfo;
1890 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1891 inp = sotoinpcb(so);
1892 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1894 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1898 INP_HASH_WLOCK(pcbinfo);
1899 in_pcbdisconnect(inp);
1900 inp->inp_laddr.s_addr = INADDR_ANY;
1901 INP_HASH_WUNLOCK(pcbinfo);
1903 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1910 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1911 struct mbuf *control, struct thread *td)
1915 inp = sotoinpcb(so);
1916 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1917 return (udp_output(inp, m, addr, control, td));
1922 udp_shutdown(struct socket *so)
1926 inp = sotoinpcb(so);
1927 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1935 struct pr_usrreqs udp_usrreqs = {
1936 .pru_abort = udp_abort,
1937 .pru_attach = udp_attach,
1938 .pru_bind = udp_bind,
1939 .pru_connect = udp_connect,
1940 .pru_control = in_control,
1941 .pru_detach = udp_detach,
1942 .pru_disconnect = udp_disconnect,
1943 .pru_peeraddr = in_getpeeraddr,
1944 .pru_send = udp_send,
1945 .pru_soreceive = soreceive_dgram,
1946 .pru_sosend = sosend_dgram,
1947 .pru_shutdown = udp_shutdown,
1948 .pru_sockaddr = in_getsockaddr,
1949 .pru_sosetlabel = in_pcbsosetlabel,
1950 .pru_close = udp_close,