xref: /freebsd/sys/netinet/udp_usrreq.c (revision 884d26c84cba3ffc3d4e626306098fcdfe6a0c2b)
1 /*-
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
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Robert N. M. Watson under
10  * contract to Juniper Networks, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
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.
23  *
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
34  * SUCH DAMAGE.
35  *
36  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 #include "opt_rss.h"
46 
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/sdt.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
66 
67 #include <vm/uma.h>
68 
69 #include <net/if.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 #include <net/rss_config.h>
73 
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>
80 #ifdef INET6
81 #include <netinet/ip6.h>
82 #endif
83 #include <netinet/ip_icmp.h>
84 #include <netinet/icmp_var.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/ip_options.h>
87 #ifdef INET6
88 #include <netinet6/ip6_var.h>
89 #endif
90 #include <netinet/udp.h>
91 #include <netinet/udp_var.h>
92 #include <netinet/udplite.h>
93 #include <netinet/in_rss.h>
94 
95 #ifdef IPSEC
96 #include <netipsec/ipsec.h>
97 #include <netipsec/esp.h>
98 #endif
99 
100 #include <machine/in_cksum.h>
101 
102 #include <security/mac/mac_framework.h>
103 
104 /*
105  * UDP and UDP-Lite protocols implementation.
106  * Per RFC 768, August, 1980.
107  * Per RFC 3828, July, 2004.
108  */
109 
110 /*
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).
115  */
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");
119 
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");
123 
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");
128 
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");
132 
133 u_long	udp_recvspace = 40 * (1024 +
134 #ifdef INET6
135 				      sizeof(struct sockaddr_in6)
136 #else
137 				      sizeof(struct sockaddr_in)
138 #endif
139 				      );	/* 40 1K datagrams */
140 
141 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
142     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
143 
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)
150 
151 #ifndef UDBHASHSIZE
152 #define	UDBHASHSIZE	128
153 #endif
154 
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)");
159 
160 #ifdef VIMAGE
161 VNET_PCPUSTAT_SYSUNINIT(udpstat);
162 #endif /* VIMAGE */
163 #ifdef INET
164 static void	udp_detach(struct socket *so);
165 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
166 		    struct mbuf *, struct thread *);
167 #endif
168 
169 #ifdef IPSEC
170 #ifdef IPSEC_NAT_T
171 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
172 #ifdef INET
173 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
174 #endif
175 #endif /* IPSEC_NAT_T */
176 #endif /* IPSEC */
177 
178 static void
179 udp_zone_change(void *tag)
180 {
181 
182 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
183 	uma_zone_set_max(V_udpcb_zone, maxsockets);
184 }
185 
186 static int
187 udp_inpcb_init(void *mem, int size, int flags)
188 {
189 	struct inpcb *inp;
190 
191 	inp = mem;
192 	INP_LOCK_INIT(inp, "inp", "udpinp");
193 	return (0);
194 }
195 
196 static int
197 udplite_inpcb_init(void *mem, int size, int flags)
198 {
199 	struct inpcb *inp;
200 
201 	inp = mem;
202 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
203 	return (0);
204 }
205 
206 void
207 udp_init(void)
208 {
209 
210 	/*
211 	 * For now default to 2-tuple UDP hashing - until the fragment
212 	 * reassembly code can also update the flowid.
213 	 *
214 	 * Once we can calculate the flowid that way and re-establish
215 	 * a 4-tuple, flip this to 4-tuple.
216 	 */
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);
226 }
227 
228 void
229 udplite_init(void)
230 {
231 
232 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
233 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
234 	    0, IPI_HASHFIELDS_2TUPLE);
235 }
236 
237 /*
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.
243  */
244 void
245 kmod_udpstat_inc(int statnum)
246 {
247 
248 	counter_u64_add(VNET(udpstat)[statnum], 1);
249 }
250 
251 int
252 udp_newudpcb(struct inpcb *inp)
253 {
254 	struct udpcb *up;
255 
256 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
257 	if (up == NULL)
258 		return (ENOBUFS);
259 	inp->inp_ppcb = up;
260 	return (0);
261 }
262 
263 void
264 udp_discardcb(struct udpcb *up)
265 {
266 
267 	uma_zfree(V_udpcb_zone, up);
268 }
269 
270 #ifdef VIMAGE
271 static void
272 udp_destroy(void *unused __unused)
273 {
274 
275 	in_pcbinfo_destroy(&V_udbinfo);
276 	uma_zdestroy(V_udpcb_zone);
277 }
278 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
279 
280 static void
281 udplite_destroy(void *unused __unused)
282 {
283 
284 	in_pcbinfo_destroy(&V_ulitecbinfo);
285 }
286 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
287     NULL);
288 #endif
289 
290 #ifdef INET
291 /*
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.
297  *
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
305  * multicast).
306  */
307 static int
308 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
309     struct sockaddr_in *udp_in)
310 {
311 	struct sockaddr *append_sa;
312 	struct socket *so;
313 	struct mbuf *opts = NULL;
314 #ifdef INET6
315 	struct sockaddr_in6 udp_in6;
316 #endif
317 	struct udpcb *up;
318 
319 	INP_LOCK_ASSERT(inp);
320 
321 	/*
322 	 * Engage the tunneling protocol.
323 	 */
324 	up = intoudpcb(inp);
325 	if (up->u_tun_func != NULL) {
326 		in_pcbref(inp);
327 		INP_RUNLOCK(inp);
328 		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
329 		    up->u_tun_ctx);
330 		INP_RLOCK(inp);
331 		return (in_pcbrele_rlocked(inp));
332 	}
333 
334 	off += sizeof(struct udphdr);
335 
336 #ifdef IPSEC
337 	/* Check AH/ESP integrity. */
338 	if (ipsec4_in_reject(n, inp)) {
339 		m_freem(n);
340 		return (0);
341 	}
342 #ifdef IPSEC_NAT_T
343 	up = intoudpcb(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. */
348 			return (0);
349 	}
350 #endif /* IPSEC_NAT_T */
351 #endif /* IPSEC */
352 #ifdef MAC
353 	if (mac_inpcb_check_deliver(inp, n) != 0) {
354 		m_freem(n);
355 		return (0);
356 	}
357 #endif /* MAC */
358 	if (inp->inp_flags & INP_CONTROLOPTS ||
359 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
360 #ifdef INET6
361 		if (inp->inp_vflag & INP_IPV6)
362 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
363 		else
364 #endif /* INET6 */
365 			ip_savecontrol(inp, &opts, ip, n);
366 	}
367 #ifdef INET6
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;
374 	} else
375 #endif /* INET6 */
376 		append_sa = (struct sockaddr *)udp_in;
377 	m_adj(n, off);
378 
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);
383 		m_freem(n);
384 		if (opts)
385 			m_freem(opts);
386 		UDPSTAT_INC(udps_fullsock);
387 	} else
388 		sorwakeup_locked(so);
389 	return (0);
390 }
391 
392 int
393 udp_input(struct mbuf **mp, int *offp, int proto)
394 {
395 	struct ip *ip;
396 	struct udphdr *uh;
397 	struct ifnet *ifp;
398 	struct inpcb *inp;
399 	uint16_t len, ip_len;
400 	struct inpcbinfo *pcbinfo;
401 	struct ip save_ip;
402 	struct sockaddr_in udp_in;
403 	struct mbuf *m;
404 	struct m_tag *fwd_tag;
405 	int cscov_partial, iphlen;
406 
407 	m = *mp;
408 	iphlen = *offp;
409 	ifp = m->m_pkthdr.rcvif;
410 	*mp = NULL;
411 	UDPSTAT_INC(udps_ipackets);
412 
413 	/*
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.
417 	 */
418 	if (iphlen > sizeof (struct ip)) {
419 		ip_stripoptions(m);
420 		iphlen = sizeof(struct ip);
421 	}
422 
423 	/*
424 	 * Get IP and UDP header together in first mbuf.
425 	 */
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);
431 		}
432 		ip = mtod(m, struct ip *);
433 	}
434 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
435 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
436 
437 	/*
438 	 * Destination port of 0 is illegal, based on RFC768.
439 	 */
440 	if (uh->uh_dport == 0)
441 		goto badunlocked;
442 
443 	/*
444 	 * Construct sockaddr format source address.  Stuff source address
445 	 * and datagram in user buffer.
446 	 */
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;
452 
453 	/*
454 	 * Make mbuf data length reflect UDP length.  If not enough data to
455 	 * reflect UDP length, drop.
456 	 */
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. */
461 		if (len == 0)
462 			len = ip_len;
463 		cscov_partial = 0;
464 	}
465 	if (ip_len != len) {
466 		if (len > ip_len || len < sizeof(struct udphdr)) {
467 			UDPSTAT_INC(udps_badlen);
468 			goto badunlocked;
469 		}
470 		if (proto == IPPROTO_UDP)
471 			m_adj(m, len - ip_len);
472 	}
473 
474 	/*
475 	 * Save a copy of the IP header in case we want restore it for
476 	 * sending an ICMP error message in response.
477 	 */
478 	if (!V_udp_blackhole)
479 		save_ip = *ip;
480 	else
481 		memset(&save_ip, 0, sizeof(save_ip));
482 
483 	/*
484 	 * Checksum extended UDP header and data.
485 	 */
486 	if (uh->uh_sum) {
487 		u_short uh_sum;
488 
489 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
490 		    !cscov_partial) {
491 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
492 				uh_sum = m->m_pkthdr.csum_data;
493 			else
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));
497 			uh_sum ^= 0xffff;
498 		} else {
499 			char b[9];
500 
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);
507 		}
508 		if (uh_sum) {
509 			UDPSTAT_INC(udps_badsum);
510 			m_freem(m);
511 			return (IPPROTO_DONE);
512 		}
513 	} else {
514 		if (proto == IPPROTO_UDP) {
515 			UDPSTAT_INC(udps_nosum);
516 		} else {
517 			/* UDPLite requires a checksum */
518 			/* XXX: What is the right UDPLite MIB counter here? */
519 			m_freem(m);
520 			return (IPPROTO_DONE);
521 		}
522 	}
523 
524 	pcbinfo = udp_get_inpcbinfo(proto);
525 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
526 	    in_broadcast(ip->ip_dst, ifp)) {
527 		struct inpcb *last;
528 		struct inpcbhead *pcblist;
529 		struct ip_moptions *imo;
530 
531 		INP_INFO_RLOCK(pcbinfo);
532 		pcblist = udp_get_pcblist(proto);
533 		last = NULL;
534 		LIST_FOREACH(inp, pcblist, inp_list) {
535 			if (inp->inp_lport != uh->uh_dport)
536 				continue;
537 #ifdef INET6
538 			if ((inp->inp_vflag & INP_IPV4) == 0)
539 				continue;
540 #endif
541 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
542 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
543 				continue;
544 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
545 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
546 				continue;
547 			if (inp->inp_fport != 0 &&
548 			    inp->inp_fport != uh->uh_sport)
549 				continue;
550 
551 			INP_RLOCK(inp);
552 
553 			/*
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.
558 			 */
559 
560 			/*
561 			 * Handle socket delivery policy for any-source
562 			 * and source-specific multicast. [RFC3678]
563 			 */
564 			imo = inp->inp_moptions;
565 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
566 				struct sockaddr_in	 group;
567 				int			 blocked;
568 				if (imo == NULL) {
569 					INP_RUNLOCK(inp);
570 					continue;
571 				}
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;
576 
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);
586 					INP_RUNLOCK(inp);
587 					continue;
588 				}
589 			}
590 			if (last != NULL) {
591 				struct mbuf *n;
592 
593 				if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
594 					UDP_PROBE(receive, NULL, last, ip,
595 					    last, uh);
596 					if (udp_append(last, ip, n, iphlen,
597 						&udp_in)) {
598 						goto inp_lost;
599 					}
600 				}
601 				INP_RUNLOCK(last);
602 			}
603 			last = inp;
604 			/*
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.
611 			 */
612 			if ((last->inp_socket->so_options &
613 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
614 				break;
615 		}
616 
617 		if (last == NULL) {
618 			/*
619 			 * No matching pcb found; discard datagram.  (No need
620 			 * to send an ICMP Port Unreachable for a broadcast
621 			 * or multicast datgram.)
622 			 */
623 			UDPSTAT_INC(udps_noportbcast);
624 			if (inp)
625 				INP_RUNLOCK(inp);
626 			INP_INFO_RUNLOCK(pcbinfo);
627 			goto badunlocked;
628 		}
629 		UDP_PROBE(receive, NULL, last, ip, last, uh);
630 		if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
631 			INP_RUNLOCK(last);
632 	inp_lost:
633 		INP_INFO_RUNLOCK(pcbinfo);
634 		return (IPPROTO_DONE);
635 	}
636 
637 	/*
638 	 * Locate pcb for datagram.
639 	 */
640 
641 	/*
642 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
643 	 */
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;
647 
648 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
649 
650 		/*
651 		 * Transparently forwarded. Pretend to be the destination.
652 		 * Already got one like this?
653 		 */
654 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
655 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
656 		if (!inp) {
657 			/*
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.
661 			 */
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);
667 		}
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;
671 	} else
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);
675 	if (inp == NULL) {
676 		if (udp_log_in_vain) {
677 			char buf[4*sizeof "123"];
678 
679 			strcpy(buf, inet_ntoa(ip->ip_dst));
680 			log(LOG_INFO,
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));
684 		}
685 		UDPSTAT_INC(udps_noport);
686 		if (m->m_flags & (M_BCAST | M_MCAST)) {
687 			UDPSTAT_INC(udps_noportbcast);
688 			goto badunlocked;
689 		}
690 		if (V_udp_blackhole)
691 			goto badunlocked;
692 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
693 			goto badunlocked;
694 		*ip = save_ip;
695 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
696 		return (IPPROTO_DONE);
697 	}
698 
699 	/*
700 	 * Check the minimum TTL for socket.
701 	 */
702 	INP_RLOCK_ASSERT(inp);
703 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
704 		INP_RUNLOCK(inp);
705 		m_freem(m);
706 		return (IPPROTO_DONE);
707 	}
708 	if (cscov_partial) {
709 		struct udpcb *up;
710 
711 		up = intoudpcb(inp);
712 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
713 			INP_RUNLOCK(inp);
714 			m_freem(m);
715 			return (IPPROTO_DONE);
716 		}
717 	}
718 
719 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
720 	if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
721 		INP_RUNLOCK(inp);
722 	return (IPPROTO_DONE);
723 
724 badunlocked:
725 	m_freem(m);
726 	return (IPPROTO_DONE);
727 }
728 #endif /* INET */
729 
730 /*
731  * Notify a udp user of an asynchronous error; just wake up so that they can
732  * collect error status.
733  */
734 struct inpcb *
735 udp_notify(struct inpcb *inp, int errno)
736 {
737 
738 	/*
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.
743 	 */
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;
749 	}
750 
751 	inp->inp_socket->so_error = errno;
752 	sorwakeup(inp->inp_socket);
753 	sowwakeup(inp->inp_socket);
754 	return (inp);
755 }
756 
757 #ifdef INET
758 static void
759 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
760     struct inpcbinfo *pcbinfo)
761 {
762 	struct ip *ip = vip;
763 	struct udphdr *uh;
764 	struct in_addr faddr;
765 	struct inpcb *inp;
766 
767 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
768 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
769 		return;
770 
771 	if (PRC_IS_REDIRECT(cmd)) {
772 		/* signal EHOSTDOWN, as it flushes the cached route */
773 		in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
774 		return;
775 	}
776 
777 	/*
778 	 * Hostdead is ugly because it goes linearly through all PCBs.
779 	 *
780 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
781 	 * DoS attack on machines with many connections.
782 	 */
783 	if (cmd == PRC_HOSTDEAD)
784 		ip = NULL;
785 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
786 		return;
787 	if (ip != NULL) {
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);
791 		if (inp != NULL) {
792 			INP_RLOCK_ASSERT(inp);
793 			if (inp->inp_socket != NULL) {
794 				udp_notify(inp, inetctlerrmap[cmd]);
795 			}
796 			INP_RUNLOCK(inp);
797 		} else {
798 			inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
799 					   ip->ip_src, uh->uh_sport,
800 					   INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
801 			if (inp != NULL) {
802 				struct udpcb *up;
803 
804 				up = intoudpcb(inp);
805 				if (up->u_icmp_func != NULL) {
806 					INP_RUNLOCK(inp);
807 					(*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
808 				} else {
809 					INP_RUNLOCK(inp);
810 				}
811 			}
812 		}
813 	} else
814 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
815 		    udp_notify);
816 }
817 void
818 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
819 {
820 
821 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
822 }
823 
824 void
825 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
826 {
827 
828 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
829 }
830 #endif /* INET */
831 
832 static int
833 udp_pcblist(SYSCTL_HANDLER_ARGS)
834 {
835 	int error, i, n;
836 	struct inpcb *inp, **inp_list;
837 	inp_gen_t gencnt;
838 	struct xinpgen xig;
839 
840 	/*
841 	 * The process of preparing the PCB list is too time-consuming and
842 	 * resource-intensive to repeat twice on every request.
843 	 */
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);
848 		return (0);
849 	}
850 
851 	if (req->newptr != 0)
852 		return (EPERM);
853 
854 	/*
855 	 * OK, now we're committed to doing something.
856 	 */
857 	INP_INFO_RLOCK(&V_udbinfo);
858 	gencnt = V_udbinfo.ipi_gencnt;
859 	n = V_udbinfo.ipi_count;
860 	INP_INFO_RUNLOCK(&V_udbinfo);
861 
862 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
863 		+ n * sizeof(struct xinpcb));
864 	if (error != 0)
865 		return (error);
866 
867 	xig.xig_len = sizeof xig;
868 	xig.xig_count = n;
869 	xig.xig_gen = gencnt;
870 	xig.xig_sogen = so_gencnt;
871 	error = SYSCTL_OUT(req, &xig, sizeof xig);
872 	if (error)
873 		return (error);
874 
875 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
876 	if (inp_list == NULL)
877 		return (ENOMEM);
878 
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)) {
882 		INP_WLOCK(inp);
883 		if (inp->inp_gencnt <= gencnt &&
884 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
885 			in_pcbref(inp);
886 			inp_list[i++] = inp;
887 		}
888 		INP_WUNLOCK(inp);
889 	}
890 	INP_INFO_RUNLOCK(&V_udbinfo);
891 	n = i;
892 
893 	error = 0;
894 	for (i = 0; i < n; i++) {
895 		inp = inp_list[i];
896 		INP_RLOCK(inp);
897 		if (inp->inp_gencnt <= gencnt) {
898 			struct xinpcb xi;
899 
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);
904 			if (inp->inp_socket)
905 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
906 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
907 			INP_RUNLOCK(inp);
908 			error = SYSCTL_OUT(req, &xi, sizeof xi);
909 		} else
910 			INP_RUNLOCK(inp);
911 	}
912 	INP_INFO_WLOCK(&V_udbinfo);
913 	for (i = 0; i < n; i++) {
914 		inp = inp_list[i];
915 		INP_RLOCK(inp);
916 		if (!in_pcbrele_rlocked(inp))
917 			INP_RUNLOCK(inp);
918 	}
919 	INP_INFO_WUNLOCK(&V_udbinfo);
920 
921 	if (!error) {
922 		/*
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.
927 		 */
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);
934 	}
935 	free(inp_list, M_TEMP);
936 	return (error);
937 }
938 
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");
942 
943 #ifdef INET
944 static int
945 udp_getcred(SYSCTL_HANDLER_ARGS)
946 {
947 	struct xucred xuc;
948 	struct sockaddr_in addrs[2];
949 	struct inpcb *inp;
950 	int error;
951 
952 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
953 	if (error)
954 		return (error);
955 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
956 	if (error)
957 		return (error);
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);
961 	if (inp != NULL) {
962 		INP_RLOCK_ASSERT(inp);
963 		if (inp->inp_socket == NULL)
964 			error = ENOENT;
965 		if (error == 0)
966 			error = cr_canseeinpcb(req->td->td_ucred, inp);
967 		if (error == 0)
968 			cru2x(inp->inp_cred, &xuc);
969 		INP_RUNLOCK(inp);
970 	} else
971 		error = ENOENT;
972 	if (error == 0)
973 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
974 	return (error);
975 }
976 
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");
980 #endif /* INET */
981 
982 int
983 udp_ctloutput(struct socket *so, struct sockopt *sopt)
984 {
985 	struct inpcb *inp;
986 	struct udpcb *up;
987 	int isudplite, error, optval;
988 
989 	error = 0;
990 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
991 	inp = sotoinpcb(so);
992 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
993 	INP_WLOCK(inp);
994 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
995 #ifdef INET6
996 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
997 			INP_WUNLOCK(inp);
998 			error = ip6_ctloutput(so, sopt);
999 		}
1000 #endif
1001 #if defined(INET) && defined(INET6)
1002 		else
1003 #endif
1004 #ifdef INET
1005 		{
1006 			INP_WUNLOCK(inp);
1007 			error = ip_ctloutput(so, sopt);
1008 		}
1009 #endif
1010 		return (error);
1011 	}
1012 
1013 	switch (sopt->sopt_dir) {
1014 	case SOPT_SET:
1015 		switch (sopt->sopt_name) {
1016 		case UDP_ENCAP:
1017 			INP_WUNLOCK(inp);
1018 			error = sooptcopyin(sopt, &optval, sizeof optval,
1019 					    sizeof optval);
1020 			if (error)
1021 				break;
1022 			inp = sotoinpcb(so);
1023 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1024 			INP_WLOCK(inp);
1025 #ifdef IPSEC_NAT_T
1026 			up = intoudpcb(inp);
1027 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1028 #endif
1029 			switch (optval) {
1030 			case 0:
1031 				/* Clear all UDP encap. */
1032 #ifdef IPSEC_NAT_T
1033 				up->u_flags &= ~UF_ESPINUDP_ALL;
1034 #endif
1035 				break;
1036 #ifdef IPSEC_NAT_T
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;
1044 				break;
1045 #endif
1046 			default:
1047 				error = EINVAL;
1048 				break;
1049 			}
1050 			INP_WUNLOCK(inp);
1051 			break;
1052 		case UDPLITE_SEND_CSCOV:
1053 		case UDPLITE_RECV_CSCOV:
1054 			if (!isudplite) {
1055 				INP_WUNLOCK(inp);
1056 				error = ENOPROTOOPT;
1057 				break;
1058 			}
1059 			INP_WUNLOCK(inp);
1060 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1061 			    sizeof(optval));
1062 			if (error != 0)
1063 				break;
1064 			inp = sotoinpcb(so);
1065 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1066 			INP_WLOCK(inp);
1067 			up = intoudpcb(inp);
1068 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1069 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
1070 				INP_WUNLOCK(inp);
1071 				error = EINVAL;
1072 				break;
1073 			}
1074 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1075 				up->u_txcslen = optval;
1076 			else
1077 				up->u_rxcslen = optval;
1078 			INP_WUNLOCK(inp);
1079 			break;
1080 		default:
1081 			INP_WUNLOCK(inp);
1082 			error = ENOPROTOOPT;
1083 			break;
1084 		}
1085 		break;
1086 	case SOPT_GET:
1087 		switch (sopt->sopt_name) {
1088 #ifdef IPSEC_NAT_T
1089 		case UDP_ENCAP:
1090 			up = intoudpcb(inp);
1091 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1092 			optval = up->u_flags & UF_ESPINUDP_ALL;
1093 			INP_WUNLOCK(inp);
1094 			error = sooptcopyout(sopt, &optval, sizeof optval);
1095 			break;
1096 #endif
1097 		case UDPLITE_SEND_CSCOV:
1098 		case UDPLITE_RECV_CSCOV:
1099 			if (!isudplite) {
1100 				INP_WUNLOCK(inp);
1101 				error = ENOPROTOOPT;
1102 				break;
1103 			}
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;
1108 			else
1109 				optval = up->u_rxcslen;
1110 			INP_WUNLOCK(inp);
1111 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1112 			break;
1113 		default:
1114 			INP_WUNLOCK(inp);
1115 			error = ENOPROTOOPT;
1116 			break;
1117 		}
1118 		break;
1119 	}
1120 	return (error);
1121 }
1122 
1123 #ifdef INET
1124 #define	UH_WLOCKED	2
1125 #define	UH_RLOCKED	1
1126 #define	UH_UNLOCKED	0
1127 static int
1128 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1129     struct mbuf *control, struct thread *td)
1130 {
1131 	struct udpiphdr *ui;
1132 	int len = m->m_pkthdr.len;
1133 	struct in_addr faddr, laddr;
1134 	struct cmsghdr *cm;
1135 	struct inpcbinfo *pcbinfo;
1136 	struct sockaddr_in *sin, src;
1137 	int cscov_partial = 0;
1138 	int error = 0;
1139 	int ipflags;
1140 	u_short fport, lport;
1141 	int unlock_udbinfo, unlock_inp;
1142 	u_char tos;
1143 	uint8_t pr;
1144 	uint16_t cscov = 0;
1145 	uint32_t flowid = 0;
1146 	uint8_t flowtype = M_HASHTYPE_NONE;
1147 
1148 	/*
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.
1153 	 */
1154 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1155 		if (control)
1156 			m_freem(control);
1157 		m_freem(m);
1158 		return (EMSGSIZE);
1159 	}
1160 
1161 	src.sin_family = 0;
1162 	sin = (struct sockaddr_in *)addr;
1163 	if (sin == NULL ||
1164 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1165 		INP_WLOCK(inp);
1166 		unlock_inp = UH_WLOCKED;
1167 	} else {
1168 		INP_RLOCK(inp);
1169 		unlock_inp = UH_RLOCKED;
1170 	}
1171 	tos = inp->inp_ip_tos;
1172 	if (control != NULL) {
1173 		/*
1174 		 * XXX: Currently, we assume all the optional information is
1175 		 * stored in a single mbuf.
1176 		 */
1177 		if (control->m_next) {
1178 			if (unlock_inp == UH_WLOCKED)
1179 				INP_WUNLOCK(inp);
1180 			else
1181 				INP_RUNLOCK(inp);
1182 			m_freem(control);
1183 			m_freem(m);
1184 			return (EINVAL);
1185 		}
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) {
1192 				error = EINVAL;
1193 				break;
1194 			}
1195 			if (cm->cmsg_level != IPPROTO_IP)
1196 				continue;
1197 
1198 			switch (cm->cmsg_type) {
1199 			case IP_SENDSRCADDR:
1200 				if (cm->cmsg_len !=
1201 				    CMSG_LEN(sizeof(struct in_addr))) {
1202 					error = EINVAL;
1203 					break;
1204 				}
1205 				bzero(&src, sizeof(src));
1206 				src.sin_family = AF_INET;
1207 				src.sin_len = sizeof(src);
1208 				src.sin_port = inp->inp_lport;
1209 				src.sin_addr =
1210 				    *(struct in_addr *)CMSG_DATA(cm);
1211 				break;
1212 
1213 			case IP_TOS:
1214 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1215 					error = EINVAL;
1216 					break;
1217 				}
1218 				tos = *(u_char *)CMSG_DATA(cm);
1219 				break;
1220 
1221 			case IP_FLOWID:
1222 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1223 					error = EINVAL;
1224 					break;
1225 				}
1226 				flowid = *(uint32_t *) CMSG_DATA(cm);
1227 				break;
1228 
1229 			case IP_FLOWTYPE:
1230 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1231 					error = EINVAL;
1232 					break;
1233 				}
1234 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1235 				break;
1236 
1237 #ifdef	RSS
1238 			case IP_RSSBUCKETID:
1239 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1240 					error = EINVAL;
1241 					break;
1242 				}
1243 				/* This is just a placeholder for now */
1244 				break;
1245 #endif	/* RSS */
1246 			default:
1247 				error = ENOPROTOOPT;
1248 				break;
1249 			}
1250 			if (error)
1251 				break;
1252 		}
1253 		m_freem(control);
1254 	}
1255 	if (error) {
1256 		if (unlock_inp == UH_WLOCKED)
1257 			INP_WUNLOCK(inp);
1258 		else
1259 			INP_RUNLOCK(inp);
1260 		m_freem(m);
1261 		return (error);
1262 	}
1263 
1264 	/*
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
1268 	 * required at all.
1269 	 *
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.
1277 	 *
1278 	 * XXXRW: Check that hash locking update here is correct.
1279 	 */
1280 	pr = inp->inp_socket->so_proto->pr_protocol;
1281 	pcbinfo = udp_get_inpcbinfo(pr);
1282 	sin = (struct sockaddr_in *)addr;
1283 	if (sin != NULL &&
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;
1295 	} else
1296 		unlock_udbinfo = UH_UNLOCKED;
1297 
1298 	/*
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.
1302 	 */
1303 	laddr = inp->inp_laddr;
1304 	lport = inp->inp_lport;
1305 	if (src.sin_family == AF_INET) {
1306 		INP_HASH_LOCK_ASSERT(pcbinfo);
1307 		if ((lport == 0) ||
1308 		    (laddr.s_addr == INADDR_ANY &&
1309 		     src.sin_addr.s_addr == INADDR_ANY)) {
1310 			error = EINVAL;
1311 			goto release;
1312 		}
1313 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1314 		    &laddr.s_addr, &lport, td->td_ucred);
1315 		if (error)
1316 			goto release;
1317 	}
1318 
1319 	/*
1320 	 * If a UDP socket has been connected, then a local address/port will
1321 	 * have been selected and bound.
1322 	 *
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.
1326 	 */
1327 	if (sin != NULL) {
1328 		INP_LOCK_ASSERT(inp);
1329 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1330 			error = EISCONN;
1331 			goto release;
1332 		}
1333 
1334 		/*
1335 		 * Jail may rewrite the destination address, so let it do
1336 		 * that before we use it.
1337 		 */
1338 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1339 		if (error)
1340 			goto release;
1341 
1342 		/*
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.
1349 		 *
1350 		 * If we already have a valid binding and we're not
1351 		 * requesting a destination address rewrite, use a fast path.
1352 		 */
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,
1360 			    td->td_ucred);
1361 			if (error)
1362 				goto release;
1363 
1364 			/*
1365 			 * XXXRW: Why not commit the port if the address is
1366 			 * !INADDR_ANY?
1367 			 */
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);
1373 				/*
1374 				 * Remember addr if jailed, to prevent
1375 				 * rebinding.
1376 				 */
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) {
1381 					inp->inp_lport = 0;
1382 					error = EAGAIN;
1383 					goto release;
1384 				}
1385 				inp->inp_flags |= INP_ANONPORT;
1386 			}
1387 		} else {
1388 			faddr = sin->sin_addr;
1389 			fport = sin->sin_port;
1390 		}
1391 	} else {
1392 		INP_LOCK_ASSERT(inp);
1393 		faddr = inp->inp_faddr;
1394 		fport = inp->inp_fport;
1395 		if (faddr.s_addr == INADDR_ANY) {
1396 			error = ENOTCONN;
1397 			goto release;
1398 		}
1399 	}
1400 
1401 	/*
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.
1405 	 */
1406 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1407 	if (m == NULL) {
1408 		error = ENOBUFS;
1409 		goto release;
1410 	}
1411 	m->m_data += max_linkhdr;
1412 	m->m_len -= max_linkhdr;
1413 	m->m_pkthdr.len -= max_linkhdr;
1414 
1415 	/*
1416 	 * Fill in mbuf with extended UDP header and addresses and length put
1417 	 * into network format.
1418 	 */
1419 	ui = mtod(m, struct udpiphdr *);
1420 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1421 	ui->ui_pr = pr;
1422 	ui->ui_src = laddr;
1423 	ui->ui_dst = faddr;
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) {
1428 		struct udpcb *up;
1429 		uint16_t plen;
1430 
1431 		up = intoudpcb(inp);
1432 		cscov = up->u_txcslen;
1433 		plen = (u_short)len + sizeof(struct udphdr);
1434 		if (cscov >= plen)
1435 			cscov = 0;
1436 		ui->ui_len = htons(plen);
1437 		ui->ui_ulen = htons(cscov);
1438 		/*
1439 		 * For UDP-Lite, checksum coverage length of zero means
1440 		 * the entire UDPLite packet is covered by the checksum.
1441 		 */
1442 		cscov_partial = (cscov == 0) ? 0 : 1;
1443 	} else
1444 		ui->ui_v = IPVERSION << 4;
1445 
1446 	/*
1447 	 * Set the Don't Fragment bit in the IP header.
1448 	 */
1449 	if (inp->inp_flags & INP_DONTFRAG) {
1450 		struct ip *ip;
1451 
1452 		ip = (struct ip *)&ui->ui_i;
1453 		ip->ip_off |= htons(IP_DF);
1454 	}
1455 
1456 	ipflags = 0;
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;
1463 
1464 #ifdef MAC
1465 	mac_inpcb_create_mbuf(inp, m);
1466 #endif
1467 
1468 	/*
1469 	 * Set up checksum and output datagram.
1470 	 */
1471 	ui->ui_sum = 0;
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;
1478 		} else {
1479 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1480 				ui->ui_sum = 0xffff;
1481 		}
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);
1489 	}
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);
1494 
1495 	/*
1496 	 * Setup flowid / RSS information for outbound socket.
1497 	 *
1498 	 * Once the UDP code decides to set a flowid some other way,
1499 	 * this allows the flowid to be overridden by userland.
1500 	 */
1501 	if (flowtype != M_HASHTYPE_NONE) {
1502 		m->m_pkthdr.flowid = flowid;
1503 		M_HASHTYPE_SET(m, flowtype);
1504 #ifdef	RSS
1505 	} else {
1506 		uint32_t hash_val, hash_type;
1507 		/*
1508 		 * Calculate an appropriate RSS hash for UDP and
1509 		 * UDP Lite.
1510 		 *
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.
1514 		 *
1515 		 * Later later on connected socket values should be
1516 		 * cached in the inpcb and reused, rather than constantly
1517 		 * re-calculating it.
1518 		 *
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.
1522 		 */
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);
1527 		}
1528 #endif
1529 	}
1530 
1531 #ifdef	RSS
1532 	/*
1533 	 * Don't override with the inp cached flowid value.
1534 	 *
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.
1538 	 *
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.
1542 	 */
1543 	ipflags |= IP_NODEFAULTFLOWID;
1544 #endif	/* RSS */
1545 
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)
1555 		INP_WUNLOCK(inp);
1556 	else
1557 		INP_RUNLOCK(inp);
1558 	return (error);
1559 
1560 release:
1561 	if (unlock_udbinfo == UH_WLOCKED) {
1562 		INP_HASH_WUNLOCK(pcbinfo);
1563 		INP_WUNLOCK(inp);
1564 	} else if (unlock_udbinfo == UH_RLOCKED) {
1565 		INP_HASH_RUNLOCK(pcbinfo);
1566 		INP_RUNLOCK(inp);
1567 	} else
1568 		INP_RUNLOCK(inp);
1569 	m_freem(m);
1570 	return (error);
1571 }
1572 
1573 
1574 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1575 /*
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.
1579  *
1580  * Returns mbuf to be processed (potentially re-allocated) or
1581  * NULL if consumed and/or processed.
1582  */
1583 static struct mbuf *
1584 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1585 {
1586 	size_t minlen, payload, skip, iphlen;
1587 	caddr_t data;
1588 	struct udpcb *up;
1589 	struct m_tag *tag;
1590 	struct udphdr *udphdr;
1591 	struct ip *ip;
1592 
1593 	INP_RLOCK_ASSERT(inp);
1594 
1595 	/*
1596 	 * Pull up data so the longest case is contiguous:
1597 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1598 	 */
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. */
1605 	}
1606 	data = mtod(m, caddr_t);	/* Points to ip header. */
1607 	payload = m->m_len - off;	/* Size of payload. */
1608 
1609 	if (payload == 1 && data[off] == '\xff')
1610 		return (m);		/* NB: keepalive packet, no decap. */
1611 
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));
1616 
1617 	/*
1618 	 * Check that the payload is large enough to hold an
1619 	 * ESP header and compute the amount of data to remove.
1620 	 *
1621 	 * NB: the caller has already done a pullup for us.
1622 	 * XXX can we assume alignment and eliminate bcopys?
1623 	 */
1624 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1625 		/*
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.
1630 		 */
1631 		uint64_t marker;
1632 
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);
1639 	} else {
1640 		uint32_t spi;
1641 
1642 		if (payload <= sizeof(struct esp)) {
1643 			IPSECSTAT_INC(ips_in_inval);
1644 			m_freem(m);
1645 			return (NULL);	/* Discard. */
1646 		}
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);
1651 	}
1652 
1653 	/*
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.
1657 	 *
1658 	 * NB: ports are maintained in network byte order everywhere
1659 	 *     in the NAT-T code.
1660 	 */
1661 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1662 		2 * sizeof(uint16_t), M_NOWAIT);
1663 	if (tag == NULL) {
1664 		IPSECSTAT_INC(ips_in_nomem);
1665 		m_freem(m);
1666 		return (NULL);		/* Discard. */
1667 	}
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);
1673 
1674 	/*
1675 	 * Remove the UDP header (and possibly the non ESP marker)
1676 	 * IP header length is iphlen
1677 	 * Before:
1678 	 *   <--- off --->
1679 	 *   +----+------+-----+
1680 	 *   | IP |  UDP | ESP |
1681 	 *   +----+------+-----+
1682 	 *        <-skip->
1683 	 * After:
1684 	 *          +----+-----+
1685 	 *          | IP | ESP |
1686 	 *          +----+-----+
1687 	 *   <-skip->
1688 	 */
1689 	ovbcopy(data, data + skip, iphlen);
1690 	m_adj(m, skip);
1691 
1692 	ip = mtod(m, struct ip *);
1693 	ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1694 	ip->ip_p = IPPROTO_ESP;
1695 
1696 	/*
1697 	 * We cannot yet update the cksums so clear any
1698 	 * h/w cksum flags as they are no longer valid.
1699 	 */
1700 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1701 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1702 
1703 	(void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1704 				AF_INET, ip->ip_p);
1705 	return (NULL);			/* NB: consumed, bypass processing. */
1706 }
1707 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1708 
1709 static void
1710 udp_abort(struct socket *so)
1711 {
1712 	struct inpcb *inp;
1713 	struct inpcbinfo *pcbinfo;
1714 
1715 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1716 	inp = sotoinpcb(so);
1717 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1718 	INP_WLOCK(inp);
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);
1725 	}
1726 	INP_WUNLOCK(inp);
1727 }
1728 
1729 static int
1730 udp_attach(struct socket *so, int proto, struct thread *td)
1731 {
1732 	struct inpcb *inp;
1733 	struct inpcbinfo *pcbinfo;
1734 	int error;
1735 
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);
1740 	if (error)
1741 		return (error);
1742 	INP_INFO_WLOCK(pcbinfo);
1743 	error = in_pcballoc(so, pcbinfo);
1744 	if (error) {
1745 		INP_INFO_WUNLOCK(pcbinfo);
1746 		return (error);
1747 	}
1748 
1749 	inp = sotoinpcb(so);
1750 	inp->inp_vflag |= INP_IPV4;
1751 	inp->inp_ip_ttl = V_ip_defttl;
1752 
1753 	error = udp_newudpcb(inp);
1754 	if (error) {
1755 		in_pcbdetach(inp);
1756 		in_pcbfree(inp);
1757 		INP_INFO_WUNLOCK(pcbinfo);
1758 		return (error);
1759 	}
1760 
1761 	INP_WUNLOCK(inp);
1762 	INP_INFO_WUNLOCK(pcbinfo);
1763 	return (0);
1764 }
1765 #endif /* INET */
1766 
1767 int
1768 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1769 {
1770 	struct inpcb *inp;
1771 	struct udpcb *up;
1772 
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"));
1777 	INP_WLOCK(inp);
1778 	up = intoudpcb(inp);
1779 	if ((up->u_tun_func != NULL) ||
1780 	    (up->u_icmp_func != NULL)) {
1781 		INP_WUNLOCK(inp);
1782 		return (EBUSY);
1783 	}
1784 	up->u_tun_func = f;
1785 	up->u_icmp_func = i;
1786 	up->u_tun_ctx = ctx;
1787 	INP_WUNLOCK(inp);
1788 	return (0);
1789 }
1790 
1791 #ifdef INET
1792 static int
1793 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1794 {
1795 	struct inpcb *inp;
1796 	struct inpcbinfo *pcbinfo;
1797 	int error;
1798 
1799 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1800 	inp = sotoinpcb(so);
1801 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1802 	INP_WLOCK(inp);
1803 	INP_HASH_WLOCK(pcbinfo);
1804 	error = in_pcbbind(inp, nam, td->td_ucred);
1805 	INP_HASH_WUNLOCK(pcbinfo);
1806 	INP_WUNLOCK(inp);
1807 	return (error);
1808 }
1809 
1810 static void
1811 udp_close(struct socket *so)
1812 {
1813 	struct inpcb *inp;
1814 	struct inpcbinfo *pcbinfo;
1815 
1816 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1817 	inp = sotoinpcb(so);
1818 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1819 	INP_WLOCK(inp);
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);
1826 	}
1827 	INP_WUNLOCK(inp);
1828 }
1829 
1830 static int
1831 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1832 {
1833 	struct inpcb *inp;
1834 	struct inpcbinfo *pcbinfo;
1835 	struct sockaddr_in *sin;
1836 	int error;
1837 
1838 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1839 	inp = sotoinpcb(so);
1840 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1841 	INP_WLOCK(inp);
1842 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1843 		INP_WUNLOCK(inp);
1844 		return (EISCONN);
1845 	}
1846 	sin = (struct sockaddr_in *)nam;
1847 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1848 	if (error != 0) {
1849 		INP_WUNLOCK(inp);
1850 		return (error);
1851 	}
1852 	INP_HASH_WLOCK(pcbinfo);
1853 	error = in_pcbconnect(inp, nam, td->td_ucred);
1854 	INP_HASH_WUNLOCK(pcbinfo);
1855 	if (error == 0)
1856 		soisconnected(so);
1857 	INP_WUNLOCK(inp);
1858 	return (error);
1859 }
1860 
1861 static void
1862 udp_detach(struct socket *so)
1863 {
1864 	struct inpcb *inp;
1865 	struct inpcbinfo *pcbinfo;
1866 	struct udpcb *up;
1867 
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);
1874 	INP_WLOCK(inp);
1875 	up = intoudpcb(inp);
1876 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1877 	inp->inp_ppcb = NULL;
1878 	in_pcbdetach(inp);
1879 	in_pcbfree(inp);
1880 	INP_INFO_WUNLOCK(pcbinfo);
1881 	udp_discardcb(up);
1882 }
1883 
1884 static int
1885 udp_disconnect(struct socket *so)
1886 {
1887 	struct inpcb *inp;
1888 	struct inpcbinfo *pcbinfo;
1889 
1890 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1891 	inp = sotoinpcb(so);
1892 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1893 	INP_WLOCK(inp);
1894 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1895 		INP_WUNLOCK(inp);
1896 		return (ENOTCONN);
1897 	}
1898 	INP_HASH_WLOCK(pcbinfo);
1899 	in_pcbdisconnect(inp);
1900 	inp->inp_laddr.s_addr = INADDR_ANY;
1901 	INP_HASH_WUNLOCK(pcbinfo);
1902 	SOCK_LOCK(so);
1903 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1904 	SOCK_UNLOCK(so);
1905 	INP_WUNLOCK(inp);
1906 	return (0);
1907 }
1908 
1909 static int
1910 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1911     struct mbuf *control, struct thread *td)
1912 {
1913 	struct inpcb *inp;
1914 
1915 	inp = sotoinpcb(so);
1916 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1917 	return (udp_output(inp, m, addr, control, td));
1918 }
1919 #endif /* INET */
1920 
1921 int
1922 udp_shutdown(struct socket *so)
1923 {
1924 	struct inpcb *inp;
1925 
1926 	inp = sotoinpcb(so);
1927 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1928 	INP_WLOCK(inp);
1929 	socantsendmore(so);
1930 	INP_WUNLOCK(inp);
1931 	return (0);
1932 }
1933 
1934 #ifdef INET
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,
1951 };
1952 #endif /* INET */
1953