xref: /freebsd/sys/netinet/udp_usrreq.c (revision e27abb6689c5733dd08ce240d5402a0de3a42254)
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_ipfw.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
46 #include "opt_rss.h"
47 
48 #include <sys/param.h>
49 #include <sys/domain.h>
50 #include <sys/eventhandler.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/protosw.h>
59 #include <sys/sdt.h>
60 #include <sys/signalvar.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/syslog.h>
66 #include <sys/systm.h>
67 
68 #include <vm/uma.h>
69 
70 #include <net/if.h>
71 #include <net/if_var.h>
72 #include <net/route.h>
73 #include <net/rss_config.h>
74 
75 #include <netinet/in.h>
76 #include <netinet/in_kdtrace.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #endif
84 #include <netinet/ip_icmp.h>
85 #include <netinet/icmp_var.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_options.h>
88 #ifdef INET6
89 #include <netinet6/ip6_var.h>
90 #endif
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
93 #include <netinet/udplite.h>
94 #include <netinet/in_rss.h>
95 
96 #ifdef IPSEC
97 #include <netipsec/ipsec.h>
98 #include <netipsec/esp.h>
99 #endif
100 
101 #include <machine/in_cksum.h>
102 
103 #include <security/mac/mac_framework.h>
104 
105 /*
106  * UDP and UDP-Lite protocols implementation.
107  * Per RFC 768, August, 1980.
108  * Per RFC 3828, July, 2004.
109  */
110 
111 /*
112  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
113  * removes the only data integrity mechanism for packets and malformed
114  * packets that would otherwise be discarded due to bad checksums, and may
115  * cause problems (especially for NFS data blocks).
116  */
117 VNET_DEFINE(int, udp_cksum) = 1;
118 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
119     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
120 
121 int	udp_log_in_vain = 0;
122 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
123     &udp_log_in_vain, 0, "Log all incoming UDP packets");
124 
125 VNET_DEFINE(int, udp_blackhole) = 0;
126 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
127     &VNET_NAME(udp_blackhole), 0,
128     "Do not send port unreachables for refused connects");
129 
130 u_long	udp_sendspace = 9216;		/* really max datagram size */
131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
132     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
133 
134 u_long	udp_recvspace = 40 * (1024 +
135 #ifdef INET6
136 				      sizeof(struct sockaddr_in6)
137 #else
138 				      sizeof(struct sockaddr_in)
139 #endif
140 				      );	/* 40 1K datagrams */
141 
142 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
143     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
144 
145 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
146 VNET_DEFINE(struct inpcbinfo, udbinfo);
147 VNET_DEFINE(struct inpcbhead, ulitecb);
148 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
149 static VNET_DEFINE(uma_zone_t, udpcb_zone);
150 #define	V_udpcb_zone			VNET(udpcb_zone)
151 
152 #ifndef UDBHASHSIZE
153 #define	UDBHASHSIZE	128
154 #endif
155 
156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
157 VNET_PCPUSTAT_SYSINIT(udpstat);
158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
159     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
160 
161 #ifdef VIMAGE
162 VNET_PCPUSTAT_SYSUNINIT(udpstat);
163 #endif /* VIMAGE */
164 #ifdef INET
165 static void	udp_detach(struct socket *so);
166 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
167 		    struct mbuf *, struct thread *);
168 #endif
169 
170 #ifdef IPSEC
171 #ifdef IPSEC_NAT_T
172 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
173 #ifdef INET
174 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
175 #endif
176 #endif /* IPSEC_NAT_T */
177 #endif /* IPSEC */
178 
179 static void
180 udp_zone_change(void *tag)
181 {
182 
183 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
184 	uma_zone_set_max(V_udpcb_zone, maxsockets);
185 }
186 
187 static int
188 udp_inpcb_init(void *mem, int size, int flags)
189 {
190 	struct inpcb *inp;
191 
192 	inp = mem;
193 	INP_LOCK_INIT(inp, "inp", "udpinp");
194 	return (0);
195 }
196 
197 static int
198 udplite_inpcb_init(void *mem, int size, int flags)
199 {
200 	struct inpcb *inp;
201 
202 	inp = mem;
203 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
204 	return (0);
205 }
206 
207 void
208 udp_init(void)
209 {
210 
211 	/*
212 	 * For now default to 2-tuple UDP hashing - until the fragment
213 	 * reassembly code can also update the flowid.
214 	 *
215 	 * Once we can calculate the flowid that way and re-establish
216 	 * a 4-tuple, flip this to 4-tuple.
217 	 */
218 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
219 	    "udp_inpcb", udp_inpcb_init, NULL, 0,
220 	    IPI_HASHFIELDS_2TUPLE);
221 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
222 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
223 	uma_zone_set_max(V_udpcb_zone, maxsockets);
224 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
225 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
226 	    EVENTHANDLER_PRI_ANY);
227 }
228 
229 void
230 udplite_init(void)
231 {
232 
233 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
234 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
235 	    0, IPI_HASHFIELDS_2TUPLE);
236 }
237 
238 /*
239  * Kernel module interface for updating udpstat.  The argument is an index
240  * into udpstat treated as an array of u_long.  While this encodes the
241  * general layout of udpstat into the caller, it doesn't encode its location,
242  * so that future changes to add, for example, per-CPU stats support won't
243  * cause binary compatibility problems for kernel modules.
244  */
245 void
246 kmod_udpstat_inc(int statnum)
247 {
248 
249 	counter_u64_add(VNET(udpstat)[statnum], 1);
250 }
251 
252 int
253 udp_newudpcb(struct inpcb *inp)
254 {
255 	struct udpcb *up;
256 
257 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
258 	if (up == NULL)
259 		return (ENOBUFS);
260 	inp->inp_ppcb = up;
261 	return (0);
262 }
263 
264 void
265 udp_discardcb(struct udpcb *up)
266 {
267 
268 	uma_zfree(V_udpcb_zone, up);
269 }
270 
271 #ifdef VIMAGE
272 static void
273 udp_destroy(void *unused __unused)
274 {
275 
276 	in_pcbinfo_destroy(&V_udbinfo);
277 	uma_zdestroy(V_udpcb_zone);
278 }
279 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
280 
281 static void
282 udplite_destroy(void *unused __unused)
283 {
284 
285 	in_pcbinfo_destroy(&V_ulitecbinfo);
286 }
287 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
288     NULL);
289 #endif
290 
291 #ifdef INET
292 /*
293  * Subroutine of udp_input(), which appends the provided mbuf chain to the
294  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
295  * contains the source address.  If the socket ends up being an IPv6 socket,
296  * udp_append() will convert to a sockaddr_in6 before passing the address
297  * into the socket code.
298  *
299  * In the normal case udp_append() will return 0, indicating that you
300  * must unlock the inp. However if a tunneling protocol is in place we increment
301  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
302  * then decrement the reference count. If the inp_rele returns 1, indicating the
303  * inp is gone, we return that to the caller to tell them *not* to unlock
304  * the inp. In the case of multi-cast this will cause the distribution
305  * to stop (though most tunneling protocols known currently do *not* use
306  * multicast).
307  */
308 static int
309 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
310     struct sockaddr_in *udp_in)
311 {
312 	struct sockaddr *append_sa;
313 	struct socket *so;
314 	struct mbuf *opts = NULL;
315 #ifdef INET6
316 	struct sockaddr_in6 udp_in6;
317 #endif
318 	struct udpcb *up;
319 
320 	INP_LOCK_ASSERT(inp);
321 
322 	/*
323 	 * Engage the tunneling protocol.
324 	 */
325 	up = intoudpcb(inp);
326 	if (up->u_tun_func != NULL) {
327 		in_pcbref(inp);
328 		INP_RUNLOCK(inp);
329 		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
330 		    up->u_tun_ctx);
331 		INP_RLOCK(inp);
332 		return (in_pcbrele_rlocked(inp));
333 	}
334 
335 	off += sizeof(struct udphdr);
336 
337 #ifdef IPSEC
338 	/* Check AH/ESP integrity. */
339 	if (ipsec4_in_reject(n, inp)) {
340 		m_freem(n);
341 		return (0);
342 	}
343 #ifdef IPSEC_NAT_T
344 	up = intoudpcb(inp);
345 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
346 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
347 		n = udp4_espdecap(inp, n, off);
348 		if (n == NULL)				/* Consumed. */
349 			return (0);
350 	}
351 #endif /* IPSEC_NAT_T */
352 #endif /* IPSEC */
353 #ifdef MAC
354 	if (mac_inpcb_check_deliver(inp, n) != 0) {
355 		m_freem(n);
356 		return (0);
357 	}
358 #endif /* MAC */
359 	if (inp->inp_flags & INP_CONTROLOPTS ||
360 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
361 #ifdef INET6
362 		if (inp->inp_vflag & INP_IPV6)
363 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
364 		else
365 #endif /* INET6 */
366 			ip_savecontrol(inp, &opts, ip, n);
367 	}
368 #ifdef INET6
369 	if (inp->inp_vflag & INP_IPV6) {
370 		bzero(&udp_in6, sizeof(udp_in6));
371 		udp_in6.sin6_len = sizeof(udp_in6);
372 		udp_in6.sin6_family = AF_INET6;
373 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
374 		append_sa = (struct sockaddr *)&udp_in6;
375 	} else
376 #endif /* INET6 */
377 		append_sa = (struct sockaddr *)udp_in;
378 	m_adj(n, off);
379 
380 	so = inp->inp_socket;
381 	SOCKBUF_LOCK(&so->so_rcv);
382 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
383 		SOCKBUF_UNLOCK(&so->so_rcv);
384 		m_freem(n);
385 		if (opts)
386 			m_freem(opts);
387 		UDPSTAT_INC(udps_fullsock);
388 	} else
389 		sorwakeup_locked(so);
390 	return (0);
391 }
392 
393 int
394 udp_input(struct mbuf **mp, int *offp, int proto)
395 {
396 	struct ip *ip;
397 	struct udphdr *uh;
398 	struct ifnet *ifp;
399 	struct inpcb *inp;
400 	uint16_t len, ip_len;
401 	struct inpcbinfo *pcbinfo;
402 	struct ip save_ip;
403 	struct sockaddr_in udp_in;
404 	struct mbuf *m;
405 	struct m_tag *fwd_tag;
406 	int cscov_partial, iphlen;
407 
408 	m = *mp;
409 	iphlen = *offp;
410 	ifp = m->m_pkthdr.rcvif;
411 	*mp = NULL;
412 	UDPSTAT_INC(udps_ipackets);
413 
414 	/*
415 	 * Strip IP options, if any; should skip this, make available to
416 	 * user, and use on returned packets, but we don't yet have a way to
417 	 * check the checksum with options still present.
418 	 */
419 	if (iphlen > sizeof (struct ip)) {
420 		ip_stripoptions(m);
421 		iphlen = sizeof(struct ip);
422 	}
423 
424 	/*
425 	 * Get IP and UDP header together in first mbuf.
426 	 */
427 	ip = mtod(m, struct ip *);
428 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
429 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
430 			UDPSTAT_INC(udps_hdrops);
431 			return (IPPROTO_DONE);
432 		}
433 		ip = mtod(m, struct ip *);
434 	}
435 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
436 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
437 
438 	/*
439 	 * Destination port of 0 is illegal, based on RFC768.
440 	 */
441 	if (uh->uh_dport == 0)
442 		goto badunlocked;
443 
444 	/*
445 	 * Construct sockaddr format source address.  Stuff source address
446 	 * and datagram in user buffer.
447 	 */
448 	bzero(&udp_in, sizeof(udp_in));
449 	udp_in.sin_len = sizeof(udp_in);
450 	udp_in.sin_family = AF_INET;
451 	udp_in.sin_port = uh->uh_sport;
452 	udp_in.sin_addr = ip->ip_src;
453 
454 	/*
455 	 * Make mbuf data length reflect UDP length.  If not enough data to
456 	 * reflect UDP length, drop.
457 	 */
458 	len = ntohs((u_short)uh->uh_ulen);
459 	ip_len = ntohs(ip->ip_len) - iphlen;
460 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
461 		/* Zero means checksum over the complete packet. */
462 		if (len == 0)
463 			len = ip_len;
464 		cscov_partial = 0;
465 	}
466 	if (ip_len != len) {
467 		if (len > ip_len || len < sizeof(struct udphdr)) {
468 			UDPSTAT_INC(udps_badlen);
469 			goto badunlocked;
470 		}
471 		if (proto == IPPROTO_UDP)
472 			m_adj(m, len - ip_len);
473 	}
474 
475 	/*
476 	 * Save a copy of the IP header in case we want restore it for
477 	 * sending an ICMP error message in response.
478 	 */
479 	if (!V_udp_blackhole)
480 		save_ip = *ip;
481 	else
482 		memset(&save_ip, 0, sizeof(save_ip));
483 
484 	/*
485 	 * Checksum extended UDP header and data.
486 	 */
487 	if (uh->uh_sum) {
488 		u_short uh_sum;
489 
490 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
491 		    !cscov_partial) {
492 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
493 				uh_sum = m->m_pkthdr.csum_data;
494 			else
495 				uh_sum = in_pseudo(ip->ip_src.s_addr,
496 				    ip->ip_dst.s_addr, htonl((u_short)len +
497 				    m->m_pkthdr.csum_data + proto));
498 			uh_sum ^= 0xffff;
499 		} else {
500 			char b[9];
501 
502 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
503 			bzero(((struct ipovly *)ip)->ih_x1, 9);
504 			((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
505 			    uh->uh_ulen : htons(ip_len);
506 			uh_sum = in_cksum(m, len + sizeof (struct ip));
507 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
508 		}
509 		if (uh_sum) {
510 			UDPSTAT_INC(udps_badsum);
511 			m_freem(m);
512 			return (IPPROTO_DONE);
513 		}
514 	} else {
515 		if (proto == IPPROTO_UDP) {
516 			UDPSTAT_INC(udps_nosum);
517 		} else {
518 			/* UDPLite requires a checksum */
519 			/* XXX: What is the right UDPLite MIB counter here? */
520 			m_freem(m);
521 			return (IPPROTO_DONE);
522 		}
523 	}
524 
525 	pcbinfo = udp_get_inpcbinfo(proto);
526 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
527 	    in_broadcast(ip->ip_dst, ifp)) {
528 		struct inpcb *last;
529 		struct inpcbhead *pcblist;
530 		struct ip_moptions *imo;
531 
532 		INP_INFO_RLOCK(pcbinfo);
533 		pcblist = udp_get_pcblist(proto);
534 		last = NULL;
535 		LIST_FOREACH(inp, pcblist, inp_list) {
536 			if (inp->inp_lport != uh->uh_dport)
537 				continue;
538 #ifdef INET6
539 			if ((inp->inp_vflag & INP_IPV4) == 0)
540 				continue;
541 #endif
542 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
543 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
544 				continue;
545 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
546 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
547 				continue;
548 			if (inp->inp_fport != 0 &&
549 			    inp->inp_fport != uh->uh_sport)
550 				continue;
551 
552 			INP_RLOCK(inp);
553 
554 			/*
555 			 * XXXRW: Because we weren't holding either the inpcb
556 			 * or the hash lock when we checked for a match
557 			 * before, we should probably recheck now that the
558 			 * inpcb lock is held.
559 			 */
560 
561 			/*
562 			 * Handle socket delivery policy for any-source
563 			 * and source-specific multicast. [RFC3678]
564 			 */
565 			imo = inp->inp_moptions;
566 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
567 				struct sockaddr_in	 group;
568 				int			 blocked;
569 				if (imo == NULL) {
570 					INP_RUNLOCK(inp);
571 					continue;
572 				}
573 				bzero(&group, sizeof(struct sockaddr_in));
574 				group.sin_len = sizeof(struct sockaddr_in);
575 				group.sin_family = AF_INET;
576 				group.sin_addr = ip->ip_dst;
577 
578 				blocked = imo_multi_filter(imo, ifp,
579 					(struct sockaddr *)&group,
580 					(struct sockaddr *)&udp_in);
581 				if (blocked != MCAST_PASS) {
582 					if (blocked == MCAST_NOTGMEMBER)
583 						IPSTAT_INC(ips_notmember);
584 					if (blocked == MCAST_NOTSMEMBER ||
585 					    blocked == MCAST_MUTED)
586 						UDPSTAT_INC(udps_filtermcast);
587 					INP_RUNLOCK(inp);
588 					continue;
589 				}
590 			}
591 			if (last != NULL) {
592 				struct mbuf *n;
593 
594 				if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
595 					UDP_PROBE(receive, NULL, last, ip,
596 					    last, uh);
597 					if (udp_append(last, ip, n, iphlen,
598 						&udp_in)) {
599 						goto inp_lost;
600 					}
601 				}
602 				INP_RUNLOCK(last);
603 			}
604 			last = inp;
605 			/*
606 			 * Don't look for additional matches if this one does
607 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
608 			 * socket options set.  This heuristic avoids
609 			 * searching through all pcbs in the common case of a
610 			 * non-shared port.  It assumes that an application
611 			 * will never clear these options after setting them.
612 			 */
613 			if ((last->inp_socket->so_options &
614 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
615 				break;
616 		}
617 
618 		if (last == NULL) {
619 			/*
620 			 * No matching pcb found; discard datagram.  (No need
621 			 * to send an ICMP Port Unreachable for a broadcast
622 			 * or multicast datgram.)
623 			 */
624 			UDPSTAT_INC(udps_noportbcast);
625 			if (inp)
626 				INP_RUNLOCK(inp);
627 			INP_INFO_RUNLOCK(pcbinfo);
628 			goto badunlocked;
629 		}
630 		UDP_PROBE(receive, NULL, last, ip, last, uh);
631 		if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
632 			INP_RUNLOCK(last);
633 	inp_lost:
634 		INP_INFO_RUNLOCK(pcbinfo);
635 		return (IPPROTO_DONE);
636 	}
637 
638 	/*
639 	 * Locate pcb for datagram.
640 	 */
641 
642 	/*
643 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
644 	 */
645 	if ((m->m_flags & M_IP_NEXTHOP) &&
646 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
647 		struct sockaddr_in *next_hop;
648 
649 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
650 
651 		/*
652 		 * Transparently forwarded. Pretend to be the destination.
653 		 * Already got one like this?
654 		 */
655 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
656 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
657 		if (!inp) {
658 			/*
659 			 * It's new.  Try to find the ambushing socket.
660 			 * Because we've rewritten the destination address,
661 			 * any hardware-generated hash is ignored.
662 			 */
663 			inp = in_pcblookup(pcbinfo, ip->ip_src,
664 			    uh->uh_sport, next_hop->sin_addr,
665 			    next_hop->sin_port ? htons(next_hop->sin_port) :
666 			    uh->uh_dport, INPLOOKUP_WILDCARD |
667 			    INPLOOKUP_RLOCKPCB, ifp);
668 		}
669 		/* Remove the tag from the packet. We don't need it anymore. */
670 		m_tag_delete(m, fwd_tag);
671 		m->m_flags &= ~M_IP_NEXTHOP;
672 	} else
673 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
674 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
675 		    INPLOOKUP_RLOCKPCB, ifp, m);
676 	if (inp == NULL) {
677 		if (udp_log_in_vain) {
678 			char buf[4*sizeof "123"];
679 
680 			strcpy(buf, inet_ntoa(ip->ip_dst));
681 			log(LOG_INFO,
682 			    "Connection attempt to UDP %s:%d from %s:%d\n",
683 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
684 			    ntohs(uh->uh_sport));
685 		}
686 		UDPSTAT_INC(udps_noport);
687 		if (m->m_flags & (M_BCAST | M_MCAST)) {
688 			UDPSTAT_INC(udps_noportbcast);
689 			goto badunlocked;
690 		}
691 		if (V_udp_blackhole)
692 			goto badunlocked;
693 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
694 			goto badunlocked;
695 		*ip = save_ip;
696 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
697 		return (IPPROTO_DONE);
698 	}
699 
700 	/*
701 	 * Check the minimum TTL for socket.
702 	 */
703 	INP_RLOCK_ASSERT(inp);
704 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
705 		INP_RUNLOCK(inp);
706 		m_freem(m);
707 		return (IPPROTO_DONE);
708 	}
709 	if (cscov_partial) {
710 		struct udpcb *up;
711 
712 		up = intoudpcb(inp);
713 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
714 			INP_RUNLOCK(inp);
715 			m_freem(m);
716 			return (IPPROTO_DONE);
717 		}
718 	}
719 
720 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
721 	if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
722 		INP_RUNLOCK(inp);
723 	return (IPPROTO_DONE);
724 
725 badunlocked:
726 	m_freem(m);
727 	return (IPPROTO_DONE);
728 }
729 #endif /* INET */
730 
731 /*
732  * Notify a udp user of an asynchronous error; just wake up so that they can
733  * collect error status.
734  */
735 struct inpcb *
736 udp_notify(struct inpcb *inp, int errno)
737 {
738 
739 	/*
740 	 * While udp_ctlinput() always calls udp_notify() with a read lock
741 	 * when invoking it directly, in_pcbnotifyall() currently uses write
742 	 * locks due to sharing code with TCP.  For now, accept either a read
743 	 * or a write lock, but a read lock is sufficient.
744 	 */
745 	INP_LOCK_ASSERT(inp);
746 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
747 	     errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
748 		RTFREE(inp->inp_route.ro_rt);
749 		inp->inp_route.ro_rt = (struct rtentry *)NULL;
750 	}
751 
752 	inp->inp_socket->so_error = errno;
753 	sorwakeup(inp->inp_socket);
754 	sowwakeup(inp->inp_socket);
755 	return (inp);
756 }
757 
758 #ifdef INET
759 static void
760 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
761     struct inpcbinfo *pcbinfo)
762 {
763 	struct ip *ip = vip;
764 	struct udphdr *uh;
765 	struct in_addr faddr;
766 	struct inpcb *inp;
767 
768 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
769 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
770 		return;
771 
772 	if (PRC_IS_REDIRECT(cmd)) {
773 		/* signal EHOSTDOWN, as it flushes the cached route */
774 		in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
775 		return;
776 	}
777 
778 	/*
779 	 * Hostdead is ugly because it goes linearly through all PCBs.
780 	 *
781 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
782 	 * DoS attack on machines with many connections.
783 	 */
784 	if (cmd == PRC_HOSTDEAD)
785 		ip = NULL;
786 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
787 		return;
788 	if (ip != NULL) {
789 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
790 		inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
791 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
792 		if (inp != NULL) {
793 			INP_RLOCK_ASSERT(inp);
794 			if (inp->inp_socket != NULL) {
795 				udp_notify(inp, inetctlerrmap[cmd]);
796 			}
797 			INP_RUNLOCK(inp);
798 		} else {
799 			inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
800 					   ip->ip_src, uh->uh_sport,
801 					   INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
802 			if (inp != NULL) {
803 				struct udpcb *up;
804 
805 				up = intoudpcb(inp);
806 				if (up->u_icmp_func != NULL) {
807 					INP_RUNLOCK(inp);
808 					(*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
809 				} else {
810 					INP_RUNLOCK(inp);
811 				}
812 			}
813 		}
814 	} else
815 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
816 		    udp_notify);
817 }
818 void
819 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
820 {
821 
822 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
823 }
824 
825 void
826 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
827 {
828 
829 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
830 }
831 #endif /* INET */
832 
833 static int
834 udp_pcblist(SYSCTL_HANDLER_ARGS)
835 {
836 	int error, i, n;
837 	struct inpcb *inp, **inp_list;
838 	inp_gen_t gencnt;
839 	struct xinpgen xig;
840 
841 	/*
842 	 * The process of preparing the PCB list is too time-consuming and
843 	 * resource-intensive to repeat twice on every request.
844 	 */
845 	if (req->oldptr == 0) {
846 		n = V_udbinfo.ipi_count;
847 		n += imax(n / 8, 10);
848 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
849 		return (0);
850 	}
851 
852 	if (req->newptr != 0)
853 		return (EPERM);
854 
855 	/*
856 	 * OK, now we're committed to doing something.
857 	 */
858 	INP_INFO_RLOCK(&V_udbinfo);
859 	gencnt = V_udbinfo.ipi_gencnt;
860 	n = V_udbinfo.ipi_count;
861 	INP_INFO_RUNLOCK(&V_udbinfo);
862 
863 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
864 		+ n * sizeof(struct xinpcb));
865 	if (error != 0)
866 		return (error);
867 
868 	xig.xig_len = sizeof xig;
869 	xig.xig_count = n;
870 	xig.xig_gen = gencnt;
871 	xig.xig_sogen = so_gencnt;
872 	error = SYSCTL_OUT(req, &xig, sizeof xig);
873 	if (error)
874 		return (error);
875 
876 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
877 	if (inp_list == NULL)
878 		return (ENOMEM);
879 
880 	INP_INFO_RLOCK(&V_udbinfo);
881 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
882 	     inp = LIST_NEXT(inp, inp_list)) {
883 		INP_WLOCK(inp);
884 		if (inp->inp_gencnt <= gencnt &&
885 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
886 			in_pcbref(inp);
887 			inp_list[i++] = inp;
888 		}
889 		INP_WUNLOCK(inp);
890 	}
891 	INP_INFO_RUNLOCK(&V_udbinfo);
892 	n = i;
893 
894 	error = 0;
895 	for (i = 0; i < n; i++) {
896 		inp = inp_list[i];
897 		INP_RLOCK(inp);
898 		if (inp->inp_gencnt <= gencnt) {
899 			struct xinpcb xi;
900 
901 			bzero(&xi, sizeof(xi));
902 			xi.xi_len = sizeof xi;
903 			/* XXX should avoid extra copy */
904 			bcopy(inp, &xi.xi_inp, sizeof *inp);
905 			if (inp->inp_socket)
906 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
907 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
908 			INP_RUNLOCK(inp);
909 			error = SYSCTL_OUT(req, &xi, sizeof xi);
910 		} else
911 			INP_RUNLOCK(inp);
912 	}
913 	INP_INFO_WLOCK(&V_udbinfo);
914 	for (i = 0; i < n; i++) {
915 		inp = inp_list[i];
916 		INP_RLOCK(inp);
917 		if (!in_pcbrele_rlocked(inp))
918 			INP_RUNLOCK(inp);
919 	}
920 	INP_INFO_WUNLOCK(&V_udbinfo);
921 
922 	if (!error) {
923 		/*
924 		 * Give the user an updated idea of our state.  If the
925 		 * generation differs from what we told her before, she knows
926 		 * that something happened while we were processing this
927 		 * request, and it might be necessary to retry.
928 		 */
929 		INP_INFO_RLOCK(&V_udbinfo);
930 		xig.xig_gen = V_udbinfo.ipi_gencnt;
931 		xig.xig_sogen = so_gencnt;
932 		xig.xig_count = V_udbinfo.ipi_count;
933 		INP_INFO_RUNLOCK(&V_udbinfo);
934 		error = SYSCTL_OUT(req, &xig, sizeof xig);
935 	}
936 	free(inp_list, M_TEMP);
937 	return (error);
938 }
939 
940 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
941     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
942     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
943 
944 #ifdef INET
945 static int
946 udp_getcred(SYSCTL_HANDLER_ARGS)
947 {
948 	struct xucred xuc;
949 	struct sockaddr_in addrs[2];
950 	struct inpcb *inp;
951 	int error;
952 
953 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
954 	if (error)
955 		return (error);
956 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
957 	if (error)
958 		return (error);
959 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
960 	    addrs[0].sin_addr, addrs[0].sin_port,
961 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
962 	if (inp != NULL) {
963 		INP_RLOCK_ASSERT(inp);
964 		if (inp->inp_socket == NULL)
965 			error = ENOENT;
966 		if (error == 0)
967 			error = cr_canseeinpcb(req->td->td_ucred, inp);
968 		if (error == 0)
969 			cru2x(inp->inp_cred, &xuc);
970 		INP_RUNLOCK(inp);
971 	} else
972 		error = ENOENT;
973 	if (error == 0)
974 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
975 	return (error);
976 }
977 
978 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
979     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
980     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
981 #endif /* INET */
982 
983 int
984 udp_ctloutput(struct socket *so, struct sockopt *sopt)
985 {
986 	struct inpcb *inp;
987 	struct udpcb *up;
988 	int isudplite, error, optval;
989 
990 	error = 0;
991 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
992 	inp = sotoinpcb(so);
993 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
994 	INP_WLOCK(inp);
995 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
996 #ifdef INET6
997 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
998 			INP_WUNLOCK(inp);
999 			error = ip6_ctloutput(so, sopt);
1000 		}
1001 #endif
1002 #if defined(INET) && defined(INET6)
1003 		else
1004 #endif
1005 #ifdef INET
1006 		{
1007 			INP_WUNLOCK(inp);
1008 			error = ip_ctloutput(so, sopt);
1009 		}
1010 #endif
1011 		return (error);
1012 	}
1013 
1014 	switch (sopt->sopt_dir) {
1015 	case SOPT_SET:
1016 		switch (sopt->sopt_name) {
1017 		case UDP_ENCAP:
1018 			INP_WUNLOCK(inp);
1019 			error = sooptcopyin(sopt, &optval, sizeof optval,
1020 					    sizeof optval);
1021 			if (error)
1022 				break;
1023 			inp = sotoinpcb(so);
1024 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1025 			INP_WLOCK(inp);
1026 #ifdef IPSEC_NAT_T
1027 			up = intoudpcb(inp);
1028 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1029 #endif
1030 			switch (optval) {
1031 			case 0:
1032 				/* Clear all UDP encap. */
1033 #ifdef IPSEC_NAT_T
1034 				up->u_flags &= ~UF_ESPINUDP_ALL;
1035 #endif
1036 				break;
1037 #ifdef IPSEC_NAT_T
1038 			case UDP_ENCAP_ESPINUDP:
1039 			case UDP_ENCAP_ESPINUDP_NON_IKE:
1040 				up->u_flags &= ~UF_ESPINUDP_ALL;
1041 				if (optval == UDP_ENCAP_ESPINUDP)
1042 					up->u_flags |= UF_ESPINUDP;
1043 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1044 					up->u_flags |= UF_ESPINUDP_NON_IKE;
1045 				break;
1046 #endif
1047 			default:
1048 				error = EINVAL;
1049 				break;
1050 			}
1051 			INP_WUNLOCK(inp);
1052 			break;
1053 		case UDPLITE_SEND_CSCOV:
1054 		case UDPLITE_RECV_CSCOV:
1055 			if (!isudplite) {
1056 				INP_WUNLOCK(inp);
1057 				error = ENOPROTOOPT;
1058 				break;
1059 			}
1060 			INP_WUNLOCK(inp);
1061 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1062 			    sizeof(optval));
1063 			if (error != 0)
1064 				break;
1065 			inp = sotoinpcb(so);
1066 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1067 			INP_WLOCK(inp);
1068 			up = intoudpcb(inp);
1069 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1070 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
1071 				INP_WUNLOCK(inp);
1072 				error = EINVAL;
1073 				break;
1074 			}
1075 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1076 				up->u_txcslen = optval;
1077 			else
1078 				up->u_rxcslen = optval;
1079 			INP_WUNLOCK(inp);
1080 			break;
1081 		default:
1082 			INP_WUNLOCK(inp);
1083 			error = ENOPROTOOPT;
1084 			break;
1085 		}
1086 		break;
1087 	case SOPT_GET:
1088 		switch (sopt->sopt_name) {
1089 #ifdef IPSEC_NAT_T
1090 		case UDP_ENCAP:
1091 			up = intoudpcb(inp);
1092 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1093 			optval = up->u_flags & UF_ESPINUDP_ALL;
1094 			INP_WUNLOCK(inp);
1095 			error = sooptcopyout(sopt, &optval, sizeof optval);
1096 			break;
1097 #endif
1098 		case UDPLITE_SEND_CSCOV:
1099 		case UDPLITE_RECV_CSCOV:
1100 			if (!isudplite) {
1101 				INP_WUNLOCK(inp);
1102 				error = ENOPROTOOPT;
1103 				break;
1104 			}
1105 			up = intoudpcb(inp);
1106 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1107 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1108 				optval = up->u_txcslen;
1109 			else
1110 				optval = up->u_rxcslen;
1111 			INP_WUNLOCK(inp);
1112 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1113 			break;
1114 		default:
1115 			INP_WUNLOCK(inp);
1116 			error = ENOPROTOOPT;
1117 			break;
1118 		}
1119 		break;
1120 	}
1121 	return (error);
1122 }
1123 
1124 #ifdef INET
1125 #define	UH_WLOCKED	2
1126 #define	UH_RLOCKED	1
1127 #define	UH_UNLOCKED	0
1128 static int
1129 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1130     struct mbuf *control, struct thread *td)
1131 {
1132 	struct udpiphdr *ui;
1133 	int len = m->m_pkthdr.len;
1134 	struct in_addr faddr, laddr;
1135 	struct cmsghdr *cm;
1136 	struct inpcbinfo *pcbinfo;
1137 	struct sockaddr_in *sin, src;
1138 	int cscov_partial = 0;
1139 	int error = 0;
1140 	int ipflags;
1141 	u_short fport, lport;
1142 	int unlock_udbinfo, unlock_inp;
1143 	u_char tos;
1144 	uint8_t pr;
1145 	uint16_t cscov = 0;
1146 	uint32_t flowid = 0;
1147 	uint8_t flowtype = M_HASHTYPE_NONE;
1148 
1149 	/*
1150 	 * udp_output() may need to temporarily bind or connect the current
1151 	 * inpcb.  As such, we don't know up front whether we will need the
1152 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1153 	 * front before acquiring any locks.
1154 	 */
1155 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1156 		if (control)
1157 			m_freem(control);
1158 		m_freem(m);
1159 		return (EMSGSIZE);
1160 	}
1161 
1162 	src.sin_family = 0;
1163 	sin = (struct sockaddr_in *)addr;
1164 	if (sin == NULL ||
1165 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1166 		INP_WLOCK(inp);
1167 		unlock_inp = UH_WLOCKED;
1168 	} else {
1169 		INP_RLOCK(inp);
1170 		unlock_inp = UH_RLOCKED;
1171 	}
1172 	tos = inp->inp_ip_tos;
1173 	if (control != NULL) {
1174 		/*
1175 		 * XXX: Currently, we assume all the optional information is
1176 		 * stored in a single mbuf.
1177 		 */
1178 		if (control->m_next) {
1179 			if (unlock_inp == UH_WLOCKED)
1180 				INP_WUNLOCK(inp);
1181 			else
1182 				INP_RUNLOCK(inp);
1183 			m_freem(control);
1184 			m_freem(m);
1185 			return (EINVAL);
1186 		}
1187 		for (; control->m_len > 0;
1188 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1189 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1190 			cm = mtod(control, struct cmsghdr *);
1191 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1192 			    || cm->cmsg_len > control->m_len) {
1193 				error = EINVAL;
1194 				break;
1195 			}
1196 			if (cm->cmsg_level != IPPROTO_IP)
1197 				continue;
1198 
1199 			switch (cm->cmsg_type) {
1200 			case IP_SENDSRCADDR:
1201 				if (cm->cmsg_len !=
1202 				    CMSG_LEN(sizeof(struct in_addr))) {
1203 					error = EINVAL;
1204 					break;
1205 				}
1206 				bzero(&src, sizeof(src));
1207 				src.sin_family = AF_INET;
1208 				src.sin_len = sizeof(src);
1209 				src.sin_port = inp->inp_lport;
1210 				src.sin_addr =
1211 				    *(struct in_addr *)CMSG_DATA(cm);
1212 				break;
1213 
1214 			case IP_TOS:
1215 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1216 					error = EINVAL;
1217 					break;
1218 				}
1219 				tos = *(u_char *)CMSG_DATA(cm);
1220 				break;
1221 
1222 			case IP_FLOWID:
1223 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1224 					error = EINVAL;
1225 					break;
1226 				}
1227 				flowid = *(uint32_t *) CMSG_DATA(cm);
1228 				break;
1229 
1230 			case IP_FLOWTYPE:
1231 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1232 					error = EINVAL;
1233 					break;
1234 				}
1235 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1236 				break;
1237 
1238 #ifdef	RSS
1239 			case IP_RSSBUCKETID:
1240 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1241 					error = EINVAL;
1242 					break;
1243 				}
1244 				/* This is just a placeholder for now */
1245 				break;
1246 #endif	/* RSS */
1247 			default:
1248 				error = ENOPROTOOPT;
1249 				break;
1250 			}
1251 			if (error)
1252 				break;
1253 		}
1254 		m_freem(control);
1255 	}
1256 	if (error) {
1257 		if (unlock_inp == UH_WLOCKED)
1258 			INP_WUNLOCK(inp);
1259 		else
1260 			INP_RUNLOCK(inp);
1261 		m_freem(m);
1262 		return (error);
1263 	}
1264 
1265 	/*
1266 	 * Depending on whether or not the application has bound or connected
1267 	 * the socket, we may have to do varying levels of work.  The optimal
1268 	 * case is for a connected UDP socket, as a global lock isn't
1269 	 * required at all.
1270 	 *
1271 	 * In order to decide which we need, we require stability of the
1272 	 * inpcb binding, which we ensure by acquiring a read lock on the
1273 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1274 	 * the trylock and retry game; note that we may end up with more
1275 	 * conservative locks than required the second time around, so later
1276 	 * assertions have to accept that.  Further analysis of the number of
1277 	 * misses under contention is required.
1278 	 *
1279 	 * XXXRW: Check that hash locking update here is correct.
1280 	 */
1281 	pr = inp->inp_socket->so_proto->pr_protocol;
1282 	pcbinfo = udp_get_inpcbinfo(pr);
1283 	sin = (struct sockaddr_in *)addr;
1284 	if (sin != NULL &&
1285 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1286 		INP_HASH_WLOCK(pcbinfo);
1287 		unlock_udbinfo = UH_WLOCKED;
1288 	} else if ((sin != NULL && (
1289 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1290 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1291 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1292 	    (inp->inp_lport == 0))) ||
1293 	    (src.sin_family == AF_INET)) {
1294 		INP_HASH_RLOCK(pcbinfo);
1295 		unlock_udbinfo = UH_RLOCKED;
1296 	} else
1297 		unlock_udbinfo = UH_UNLOCKED;
1298 
1299 	/*
1300 	 * If the IP_SENDSRCADDR control message was specified, override the
1301 	 * source address for this datagram.  Its use is invalidated if the
1302 	 * address thus specified is incomplete or clobbers other inpcbs.
1303 	 */
1304 	laddr = inp->inp_laddr;
1305 	lport = inp->inp_lport;
1306 	if (src.sin_family == AF_INET) {
1307 		INP_HASH_LOCK_ASSERT(pcbinfo);
1308 		if ((lport == 0) ||
1309 		    (laddr.s_addr == INADDR_ANY &&
1310 		     src.sin_addr.s_addr == INADDR_ANY)) {
1311 			error = EINVAL;
1312 			goto release;
1313 		}
1314 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1315 		    &laddr.s_addr, &lport, td->td_ucred);
1316 		if (error)
1317 			goto release;
1318 	}
1319 
1320 	/*
1321 	 * If a UDP socket has been connected, then a local address/port will
1322 	 * have been selected and bound.
1323 	 *
1324 	 * If a UDP socket has not been connected to, then an explicit
1325 	 * destination address must be used, in which case a local
1326 	 * address/port may not have been selected and bound.
1327 	 */
1328 	if (sin != NULL) {
1329 		INP_LOCK_ASSERT(inp);
1330 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1331 			error = EISCONN;
1332 			goto release;
1333 		}
1334 
1335 		/*
1336 		 * Jail may rewrite the destination address, so let it do
1337 		 * that before we use it.
1338 		 */
1339 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1340 		if (error)
1341 			goto release;
1342 
1343 		/*
1344 		 * If a local address or port hasn't yet been selected, or if
1345 		 * the destination address needs to be rewritten due to using
1346 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1347 		 * to do the heavy lifting.  Once a port is selected, we
1348 		 * commit the binding back to the socket; we also commit the
1349 		 * binding of the address if in jail.
1350 		 *
1351 		 * If we already have a valid binding and we're not
1352 		 * requesting a destination address rewrite, use a fast path.
1353 		 */
1354 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1355 		    inp->inp_lport == 0 ||
1356 		    sin->sin_addr.s_addr == INADDR_ANY ||
1357 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1358 			INP_HASH_LOCK_ASSERT(pcbinfo);
1359 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1360 			    &lport, &faddr.s_addr, &fport, NULL,
1361 			    td->td_ucred);
1362 			if (error)
1363 				goto release;
1364 
1365 			/*
1366 			 * XXXRW: Why not commit the port if the address is
1367 			 * !INADDR_ANY?
1368 			 */
1369 			/* Commit the local port if newly assigned. */
1370 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1371 			    inp->inp_lport == 0) {
1372 				INP_WLOCK_ASSERT(inp);
1373 				INP_HASH_WLOCK_ASSERT(pcbinfo);
1374 				/*
1375 				 * Remember addr if jailed, to prevent
1376 				 * rebinding.
1377 				 */
1378 				if (prison_flag(td->td_ucred, PR_IP4))
1379 					inp->inp_laddr = laddr;
1380 				inp->inp_lport = lport;
1381 				if (in_pcbinshash(inp) != 0) {
1382 					inp->inp_lport = 0;
1383 					error = EAGAIN;
1384 					goto release;
1385 				}
1386 				inp->inp_flags |= INP_ANONPORT;
1387 			}
1388 		} else {
1389 			faddr = sin->sin_addr;
1390 			fport = sin->sin_port;
1391 		}
1392 	} else {
1393 		INP_LOCK_ASSERT(inp);
1394 		faddr = inp->inp_faddr;
1395 		fport = inp->inp_fport;
1396 		if (faddr.s_addr == INADDR_ANY) {
1397 			error = ENOTCONN;
1398 			goto release;
1399 		}
1400 	}
1401 
1402 	/*
1403 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1404 	 * link-layer headers.  Immediate slide the data pointer back forward
1405 	 * since we won't use that space at this layer.
1406 	 */
1407 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1408 	if (m == NULL) {
1409 		error = ENOBUFS;
1410 		goto release;
1411 	}
1412 	m->m_data += max_linkhdr;
1413 	m->m_len -= max_linkhdr;
1414 	m->m_pkthdr.len -= max_linkhdr;
1415 
1416 	/*
1417 	 * Fill in mbuf with extended UDP header and addresses and length put
1418 	 * into network format.
1419 	 */
1420 	ui = mtod(m, struct udpiphdr *);
1421 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1422 	ui->ui_pr = pr;
1423 	ui->ui_src = laddr;
1424 	ui->ui_dst = faddr;
1425 	ui->ui_sport = lport;
1426 	ui->ui_dport = fport;
1427 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1428 	if (pr == IPPROTO_UDPLITE) {
1429 		struct udpcb *up;
1430 		uint16_t plen;
1431 
1432 		up = intoudpcb(inp);
1433 		cscov = up->u_txcslen;
1434 		plen = (u_short)len + sizeof(struct udphdr);
1435 		if (cscov >= plen)
1436 			cscov = 0;
1437 		ui->ui_len = htons(plen);
1438 		ui->ui_ulen = htons(cscov);
1439 		/*
1440 		 * For UDP-Lite, checksum coverage length of zero means
1441 		 * the entire UDPLite packet is covered by the checksum.
1442 		 */
1443 		cscov_partial = (cscov == 0) ? 0 : 1;
1444 	} else
1445 		ui->ui_v = IPVERSION << 4;
1446 
1447 	/*
1448 	 * Set the Don't Fragment bit in the IP header.
1449 	 */
1450 	if (inp->inp_flags & INP_DONTFRAG) {
1451 		struct ip *ip;
1452 
1453 		ip = (struct ip *)&ui->ui_i;
1454 		ip->ip_off |= htons(IP_DF);
1455 	}
1456 
1457 	ipflags = 0;
1458 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1459 		ipflags |= IP_ROUTETOIF;
1460 	if (inp->inp_socket->so_options & SO_BROADCAST)
1461 		ipflags |= IP_ALLOWBROADCAST;
1462 	if (inp->inp_flags & INP_ONESBCAST)
1463 		ipflags |= IP_SENDONES;
1464 
1465 #ifdef MAC
1466 	mac_inpcb_create_mbuf(inp, m);
1467 #endif
1468 
1469 	/*
1470 	 * Set up checksum and output datagram.
1471 	 */
1472 	ui->ui_sum = 0;
1473 	if (pr == IPPROTO_UDPLITE) {
1474 		if (inp->inp_flags & INP_ONESBCAST)
1475 			faddr.s_addr = INADDR_BROADCAST;
1476 		if (cscov_partial) {
1477 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1478 				ui->ui_sum = 0xffff;
1479 		} else {
1480 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1481 				ui->ui_sum = 0xffff;
1482 		}
1483 	} else if (V_udp_cksum) {
1484 		if (inp->inp_flags & INP_ONESBCAST)
1485 			faddr.s_addr = INADDR_BROADCAST;
1486 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1487 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1488 		m->m_pkthdr.csum_flags = CSUM_UDP;
1489 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1490 	}
1491 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1492 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1493 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1494 	UDPSTAT_INC(udps_opackets);
1495 
1496 	/*
1497 	 * Setup flowid / RSS information for outbound socket.
1498 	 *
1499 	 * Once the UDP code decides to set a flowid some other way,
1500 	 * this allows the flowid to be overridden by userland.
1501 	 */
1502 	if (flowtype != M_HASHTYPE_NONE) {
1503 		m->m_pkthdr.flowid = flowid;
1504 		M_HASHTYPE_SET(m, flowtype);
1505 #ifdef	RSS
1506 	} else {
1507 		uint32_t hash_val, hash_type;
1508 		/*
1509 		 * Calculate an appropriate RSS hash for UDP and
1510 		 * UDP Lite.
1511 		 *
1512 		 * The called function will take care of figuring out
1513 		 * whether a 2-tuple or 4-tuple hash is required based
1514 		 * on the currently configured scheme.
1515 		 *
1516 		 * Later later on connected socket values should be
1517 		 * cached in the inpcb and reused, rather than constantly
1518 		 * re-calculating it.
1519 		 *
1520 		 * UDP Lite is a different protocol number and will
1521 		 * likely end up being hashed as a 2-tuple until
1522 		 * RSS / NICs grow UDP Lite protocol awareness.
1523 		 */
1524 		if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1525 		    pr, &hash_val, &hash_type) == 0) {
1526 			m->m_pkthdr.flowid = hash_val;
1527 			M_HASHTYPE_SET(m, hash_type);
1528 		}
1529 #endif
1530 	}
1531 
1532 #ifdef	RSS
1533 	/*
1534 	 * Don't override with the inp cached flowid value.
1535 	 *
1536 	 * Depending upon the kind of send being done, the inp
1537 	 * flowid/flowtype values may actually not be appropriate
1538 	 * for this particular socket send.
1539 	 *
1540 	 * We should either leave the flowid at zero (which is what is
1541 	 * currently done) or set it to some software generated
1542 	 * hash value based on the packet contents.
1543 	 */
1544 	ipflags |= IP_NODEFAULTFLOWID;
1545 #endif	/* RSS */
1546 
1547 	if (unlock_udbinfo == UH_WLOCKED)
1548 		INP_HASH_WUNLOCK(pcbinfo);
1549 	else if (unlock_udbinfo == UH_RLOCKED)
1550 		INP_HASH_RUNLOCK(pcbinfo);
1551 	UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1552 	error = ip_output(m, inp->inp_options,
1553 	    (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1554 	    inp->inp_moptions, inp);
1555 	if (unlock_inp == UH_WLOCKED)
1556 		INP_WUNLOCK(inp);
1557 	else
1558 		INP_RUNLOCK(inp);
1559 	return (error);
1560 
1561 release:
1562 	if (unlock_udbinfo == UH_WLOCKED) {
1563 		INP_HASH_WUNLOCK(pcbinfo);
1564 		INP_WUNLOCK(inp);
1565 	} else if (unlock_udbinfo == UH_RLOCKED) {
1566 		INP_HASH_RUNLOCK(pcbinfo);
1567 		INP_RUNLOCK(inp);
1568 	} else
1569 		INP_RUNLOCK(inp);
1570 	m_freem(m);
1571 	return (error);
1572 }
1573 
1574 
1575 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1576 /*
1577  * Potentially decap ESP in UDP frame.  Check for an ESP header
1578  * and optional marker; if present, strip the UDP header and
1579  * push the result through IPSec.
1580  *
1581  * Returns mbuf to be processed (potentially re-allocated) or
1582  * NULL if consumed and/or processed.
1583  */
1584 static struct mbuf *
1585 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1586 {
1587 	size_t minlen, payload, skip, iphlen;
1588 	caddr_t data;
1589 	struct udpcb *up;
1590 	struct m_tag *tag;
1591 	struct udphdr *udphdr;
1592 	struct ip *ip;
1593 
1594 	INP_RLOCK_ASSERT(inp);
1595 
1596 	/*
1597 	 * Pull up data so the longest case is contiguous:
1598 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1599 	 */
1600 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1601 	if (minlen > m->m_pkthdr.len)
1602 		minlen = m->m_pkthdr.len;
1603 	if ((m = m_pullup(m, minlen)) == NULL) {
1604 		IPSECSTAT_INC(ips_in_inval);
1605 		return (NULL);		/* Bypass caller processing. */
1606 	}
1607 	data = mtod(m, caddr_t);	/* Points to ip header. */
1608 	payload = m->m_len - off;	/* Size of payload. */
1609 
1610 	if (payload == 1 && data[off] == '\xff')
1611 		return (m);		/* NB: keepalive packet, no decap. */
1612 
1613 	up = intoudpcb(inp);
1614 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1615 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1616 	    ("u_flags 0x%x", up->u_flags));
1617 
1618 	/*
1619 	 * Check that the payload is large enough to hold an
1620 	 * ESP header and compute the amount of data to remove.
1621 	 *
1622 	 * NB: the caller has already done a pullup for us.
1623 	 * XXX can we assume alignment and eliminate bcopys?
1624 	 */
1625 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1626 		/*
1627 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1628 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1629 		 * possible AH mode non-IKE marker+non-ESP marker
1630 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1631 		 */
1632 		uint64_t marker;
1633 
1634 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1635 			return (m);	/* NB: no decap. */
1636 		bcopy(data + off, &marker, sizeof(uint64_t));
1637 		if (marker != 0)	/* Non-IKE marker. */
1638 			return (m);	/* NB: no decap. */
1639 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1640 	} else {
1641 		uint32_t spi;
1642 
1643 		if (payload <= sizeof(struct esp)) {
1644 			IPSECSTAT_INC(ips_in_inval);
1645 			m_freem(m);
1646 			return (NULL);	/* Discard. */
1647 		}
1648 		bcopy(data + off, &spi, sizeof(uint32_t));
1649 		if (spi == 0)		/* Non-ESP marker. */
1650 			return (m);	/* NB: no decap. */
1651 		skip = sizeof(struct udphdr);
1652 	}
1653 
1654 	/*
1655 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1656 	 * the UDP ports. This is required if we want to select
1657 	 * the right SPD for multiple hosts behind same NAT.
1658 	 *
1659 	 * NB: ports are maintained in network byte order everywhere
1660 	 *     in the NAT-T code.
1661 	 */
1662 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1663 		2 * sizeof(uint16_t), M_NOWAIT);
1664 	if (tag == NULL) {
1665 		IPSECSTAT_INC(ips_in_nomem);
1666 		m_freem(m);
1667 		return (NULL);		/* Discard. */
1668 	}
1669 	iphlen = off - sizeof(struct udphdr);
1670 	udphdr = (struct udphdr *)(data + iphlen);
1671 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1672 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1673 	m_tag_prepend(m, tag);
1674 
1675 	/*
1676 	 * Remove the UDP header (and possibly the non ESP marker)
1677 	 * IP header length is iphlen
1678 	 * Before:
1679 	 *   <--- off --->
1680 	 *   +----+------+-----+
1681 	 *   | IP |  UDP | ESP |
1682 	 *   +----+------+-----+
1683 	 *        <-skip->
1684 	 * After:
1685 	 *          +----+-----+
1686 	 *          | IP | ESP |
1687 	 *          +----+-----+
1688 	 *   <-skip->
1689 	 */
1690 	ovbcopy(data, data + skip, iphlen);
1691 	m_adj(m, skip);
1692 
1693 	ip = mtod(m, struct ip *);
1694 	ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1695 	ip->ip_p = IPPROTO_ESP;
1696 
1697 	/*
1698 	 * We cannot yet update the cksums so clear any
1699 	 * h/w cksum flags as they are no longer valid.
1700 	 */
1701 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1702 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1703 
1704 	(void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1705 				AF_INET, ip->ip_p);
1706 	return (NULL);			/* NB: consumed, bypass processing. */
1707 }
1708 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1709 
1710 static void
1711 udp_abort(struct socket *so)
1712 {
1713 	struct inpcb *inp;
1714 	struct inpcbinfo *pcbinfo;
1715 
1716 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1717 	inp = sotoinpcb(so);
1718 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1719 	INP_WLOCK(inp);
1720 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1721 		INP_HASH_WLOCK(pcbinfo);
1722 		in_pcbdisconnect(inp);
1723 		inp->inp_laddr.s_addr = INADDR_ANY;
1724 		INP_HASH_WUNLOCK(pcbinfo);
1725 		soisdisconnected(so);
1726 	}
1727 	INP_WUNLOCK(inp);
1728 }
1729 
1730 static int
1731 udp_attach(struct socket *so, int proto, struct thread *td)
1732 {
1733 	struct inpcb *inp;
1734 	struct inpcbinfo *pcbinfo;
1735 	int error;
1736 
1737 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1738 	inp = sotoinpcb(so);
1739 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1740 	error = soreserve(so, udp_sendspace, udp_recvspace);
1741 	if (error)
1742 		return (error);
1743 	INP_INFO_WLOCK(pcbinfo);
1744 	error = in_pcballoc(so, pcbinfo);
1745 	if (error) {
1746 		INP_INFO_WUNLOCK(pcbinfo);
1747 		return (error);
1748 	}
1749 
1750 	inp = sotoinpcb(so);
1751 	inp->inp_vflag |= INP_IPV4;
1752 	inp->inp_ip_ttl = V_ip_defttl;
1753 
1754 	error = udp_newudpcb(inp);
1755 	if (error) {
1756 		in_pcbdetach(inp);
1757 		in_pcbfree(inp);
1758 		INP_INFO_WUNLOCK(pcbinfo);
1759 		return (error);
1760 	}
1761 
1762 	INP_WUNLOCK(inp);
1763 	INP_INFO_WUNLOCK(pcbinfo);
1764 	return (0);
1765 }
1766 #endif /* INET */
1767 
1768 int
1769 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1770 {
1771 	struct inpcb *inp;
1772 	struct udpcb *up;
1773 
1774 	KASSERT(so->so_type == SOCK_DGRAM,
1775 	    ("udp_set_kernel_tunneling: !dgram"));
1776 	inp = sotoinpcb(so);
1777 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1778 	INP_WLOCK(inp);
1779 	up = intoudpcb(inp);
1780 	if ((up->u_tun_func != NULL) ||
1781 	    (up->u_icmp_func != NULL)) {
1782 		INP_WUNLOCK(inp);
1783 		return (EBUSY);
1784 	}
1785 	up->u_tun_func = f;
1786 	up->u_icmp_func = i;
1787 	up->u_tun_ctx = ctx;
1788 	INP_WUNLOCK(inp);
1789 	return (0);
1790 }
1791 
1792 #ifdef INET
1793 static int
1794 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1795 {
1796 	struct inpcb *inp;
1797 	struct inpcbinfo *pcbinfo;
1798 	int error;
1799 
1800 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1801 	inp = sotoinpcb(so);
1802 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1803 	INP_WLOCK(inp);
1804 	INP_HASH_WLOCK(pcbinfo);
1805 	error = in_pcbbind(inp, nam, td->td_ucred);
1806 	INP_HASH_WUNLOCK(pcbinfo);
1807 	INP_WUNLOCK(inp);
1808 	return (error);
1809 }
1810 
1811 static void
1812 udp_close(struct socket *so)
1813 {
1814 	struct inpcb *inp;
1815 	struct inpcbinfo *pcbinfo;
1816 
1817 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1818 	inp = sotoinpcb(so);
1819 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1820 	INP_WLOCK(inp);
1821 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1822 		INP_HASH_WLOCK(pcbinfo);
1823 		in_pcbdisconnect(inp);
1824 		inp->inp_laddr.s_addr = INADDR_ANY;
1825 		INP_HASH_WUNLOCK(pcbinfo);
1826 		soisdisconnected(so);
1827 	}
1828 	INP_WUNLOCK(inp);
1829 }
1830 
1831 static int
1832 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1833 {
1834 	struct inpcb *inp;
1835 	struct inpcbinfo *pcbinfo;
1836 	struct sockaddr_in *sin;
1837 	int error;
1838 
1839 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1840 	inp = sotoinpcb(so);
1841 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1842 	INP_WLOCK(inp);
1843 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1844 		INP_WUNLOCK(inp);
1845 		return (EISCONN);
1846 	}
1847 	sin = (struct sockaddr_in *)nam;
1848 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1849 	if (error != 0) {
1850 		INP_WUNLOCK(inp);
1851 		return (error);
1852 	}
1853 	INP_HASH_WLOCK(pcbinfo);
1854 	error = in_pcbconnect(inp, nam, td->td_ucred);
1855 	INP_HASH_WUNLOCK(pcbinfo);
1856 	if (error == 0)
1857 		soisconnected(so);
1858 	INP_WUNLOCK(inp);
1859 	return (error);
1860 }
1861 
1862 static void
1863 udp_detach(struct socket *so)
1864 {
1865 	struct inpcb *inp;
1866 	struct inpcbinfo *pcbinfo;
1867 	struct udpcb *up;
1868 
1869 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1870 	inp = sotoinpcb(so);
1871 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1872 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1873 	    ("udp_detach: not disconnected"));
1874 	INP_INFO_WLOCK(pcbinfo);
1875 	INP_WLOCK(inp);
1876 	up = intoudpcb(inp);
1877 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1878 	inp->inp_ppcb = NULL;
1879 	in_pcbdetach(inp);
1880 	in_pcbfree(inp);
1881 	INP_INFO_WUNLOCK(pcbinfo);
1882 	udp_discardcb(up);
1883 }
1884 
1885 static int
1886 udp_disconnect(struct socket *so)
1887 {
1888 	struct inpcb *inp;
1889 	struct inpcbinfo *pcbinfo;
1890 
1891 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1892 	inp = sotoinpcb(so);
1893 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1894 	INP_WLOCK(inp);
1895 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1896 		INP_WUNLOCK(inp);
1897 		return (ENOTCONN);
1898 	}
1899 	INP_HASH_WLOCK(pcbinfo);
1900 	in_pcbdisconnect(inp);
1901 	inp->inp_laddr.s_addr = INADDR_ANY;
1902 	INP_HASH_WUNLOCK(pcbinfo);
1903 	SOCK_LOCK(so);
1904 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1905 	SOCK_UNLOCK(so);
1906 	INP_WUNLOCK(inp);
1907 	return (0);
1908 }
1909 
1910 static int
1911 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1912     struct mbuf *control, struct thread *td)
1913 {
1914 	struct inpcb *inp;
1915 
1916 	inp = sotoinpcb(so);
1917 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1918 	return (udp_output(inp, m, addr, control, td));
1919 }
1920 #endif /* INET */
1921 
1922 int
1923 udp_shutdown(struct socket *so)
1924 {
1925 	struct inpcb *inp;
1926 
1927 	inp = sotoinpcb(so);
1928 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1929 	INP_WLOCK(inp);
1930 	socantsendmore(so);
1931 	INP_WUNLOCK(inp);
1932 	return (0);
1933 }
1934 
1935 #ifdef INET
1936 struct pr_usrreqs udp_usrreqs = {
1937 	.pru_abort =		udp_abort,
1938 	.pru_attach =		udp_attach,
1939 	.pru_bind =		udp_bind,
1940 	.pru_connect =		udp_connect,
1941 	.pru_control =		in_control,
1942 	.pru_detach =		udp_detach,
1943 	.pru_disconnect =	udp_disconnect,
1944 	.pru_peeraddr =		in_getpeeraddr,
1945 	.pru_send =		udp_send,
1946 	.pru_soreceive =	soreceive_dgram,
1947 	.pru_sosend =		sosend_dgram,
1948 	.pru_shutdown =		udp_shutdown,
1949 	.pru_sockaddr =		in_getsockaddr,
1950 	.pru_sosetlabel =	in_pcbsosetlabel,
1951 	.pru_close =		udp_close,
1952 };
1953 #endif /* INET */
1954