xref: /freebsd/sys/netinet/udp_usrreq.c (revision 7cd2dcf07629713e5a3d60472cfe4701b705a167)
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  * All rights reserved.
7  *
8  * Portions of this software were developed by Robert N. M. Watson under
9  * contract to Juniper Networks, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ipfw.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/sx.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/systm.h>
64 
65 #include <vm/uma.h>
66 
67 #include <net/if.h>
68 #include <net/route.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
75 #ifdef INET6
76 #include <netinet/ip6.h>
77 #endif
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/ip_options.h>
82 #ifdef INET6
83 #include <netinet6/ip6_var.h>
84 #endif
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
87 
88 #ifdef IPSEC
89 #include <netipsec/ipsec.h>
90 #include <netipsec/esp.h>
91 #endif
92 
93 #include <machine/in_cksum.h>
94 
95 #include <security/mac/mac_framework.h>
96 
97 /*
98  * UDP protocol implementation.
99  * Per RFC 768, August, 1980.
100  */
101 
102 /*
103  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
104  * removes the only data integrity mechanism for packets and malformed
105  * packets that would otherwise be discarded due to bad checksums, and may
106  * cause problems (especially for NFS data blocks).
107  */
108 VNET_DEFINE(int, udp_cksum) = 1;
109 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
110     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
111 
112 int	udp_log_in_vain = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114     &udp_log_in_vain, 0, "Log all incoming UDP packets");
115 
116 VNET_DEFINE(int, udp_blackhole) = 0;
117 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
118     &VNET_NAME(udp_blackhole), 0,
119     "Do not send port unreachables for refused connects");
120 
121 u_long	udp_sendspace = 9216;		/* really max datagram size */
122 					/* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
125 
126 u_long	udp_recvspace = 40 * (1024 +
127 #ifdef INET6
128 				      sizeof(struct sockaddr_in6)
129 #else
130 				      sizeof(struct sockaddr_in)
131 #endif
132 				      );
133 
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
136 
137 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
138 VNET_DEFINE(struct inpcbinfo, udbinfo);
139 static VNET_DEFINE(uma_zone_t, udpcb_zone);
140 #define	V_udpcb_zone			VNET(udpcb_zone)
141 
142 #ifndef UDBHASHSIZE
143 #define	UDBHASHSIZE	128
144 #endif
145 
146 VNET_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
147 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
148     &VNET_NAME(udpstat), udpstat,
149     "UDP statistics (struct udpstat, netinet/udp_var.h)");
150 
151 #ifdef INET
152 static void	udp_detach(struct socket *so);
153 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
154 		    struct mbuf *, struct thread *);
155 #endif
156 
157 #ifdef IPSEC
158 #ifdef IPSEC_NAT_T
159 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
160 #ifdef INET
161 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
162 #endif
163 #endif /* IPSEC_NAT_T */
164 #endif /* IPSEC */
165 
166 static void
167 udp_zone_change(void *tag)
168 {
169 
170 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
171 	uma_zone_set_max(V_udpcb_zone, maxsockets);
172 }
173 
174 static int
175 udp_inpcb_init(void *mem, int size, int flags)
176 {
177 	struct inpcb *inp;
178 
179 	inp = mem;
180 	INP_LOCK_INIT(inp, "inp", "udpinp");
181 	return (0);
182 }
183 
184 void
185 udp_init(void)
186 {
187 
188 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
189 	    "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
190 	    IPI_HASHFIELDS_2TUPLE);
191 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
192 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 	uma_zone_set_max(V_udpcb_zone, maxsockets);
194 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
195 	    EVENTHANDLER_PRI_ANY);
196 }
197 
198 /*
199  * Kernel module interface for updating udpstat.  The argument is an index
200  * into udpstat treated as an array of u_long.  While this encodes the
201  * general layout of udpstat into the caller, it doesn't encode its location,
202  * so that future changes to add, for example, per-CPU stats support won't
203  * cause binary compatibility problems for kernel modules.
204  */
205 void
206 kmod_udpstat_inc(int statnum)
207 {
208 
209 	(*((u_long *)&V_udpstat + statnum))++;
210 }
211 
212 int
213 udp_newudpcb(struct inpcb *inp)
214 {
215 	struct udpcb *up;
216 
217 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
218 	if (up == NULL)
219 		return (ENOBUFS);
220 	inp->inp_ppcb = up;
221 	return (0);
222 }
223 
224 void
225 udp_discardcb(struct udpcb *up)
226 {
227 
228 	uma_zfree(V_udpcb_zone, up);
229 }
230 
231 #ifdef VIMAGE
232 void
233 udp_destroy(void)
234 {
235 
236 	in_pcbinfo_destroy(&V_udbinfo);
237 	uma_zdestroy(V_udpcb_zone);
238 }
239 #endif
240 
241 #ifdef INET
242 /*
243  * Subroutine of udp_input(), which appends the provided mbuf chain to the
244  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
245  * contains the source address.  If the socket ends up being an IPv6 socket,
246  * udp_append() will convert to a sockaddr_in6 before passing the address
247  * into the socket code.
248  */
249 static void
250 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
251     struct sockaddr_in *udp_in)
252 {
253 	struct sockaddr *append_sa;
254 	struct socket *so;
255 	struct mbuf *opts = 0;
256 #ifdef INET6
257 	struct sockaddr_in6 udp_in6;
258 #endif
259 	struct udpcb *up;
260 
261 	INP_LOCK_ASSERT(inp);
262 
263 	/*
264 	 * Engage the tunneling protocol.
265 	 */
266 	up = intoudpcb(inp);
267 	if (up->u_tun_func != NULL) {
268 		(*up->u_tun_func)(n, off, inp);
269 		return;
270 	}
271 
272 	if (n == NULL)
273 		return;
274 
275 	off += sizeof(struct udphdr);
276 
277 #ifdef IPSEC
278 	/* Check AH/ESP integrity. */
279 	if (ipsec4_in_reject(n, inp)) {
280 		m_freem(n);
281 		V_ipsec4stat.in_polvio++;
282 		return;
283 	}
284 #ifdef IPSEC_NAT_T
285 	up = intoudpcb(inp);
286 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
287 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
288 		n = udp4_espdecap(inp, n, off);
289 		if (n == NULL)				/* Consumed. */
290 			return;
291 	}
292 #endif /* IPSEC_NAT_T */
293 #endif /* IPSEC */
294 #ifdef MAC
295 	if (mac_inpcb_check_deliver(inp, n) != 0) {
296 		m_freem(n);
297 		return;
298 	}
299 #endif /* MAC */
300 	if (inp->inp_flags & INP_CONTROLOPTS ||
301 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
302 #ifdef INET6
303 		if (inp->inp_vflag & INP_IPV6)
304 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
305 		else
306 #endif /* INET6 */
307 			ip_savecontrol(inp, &opts, ip, n);
308 	}
309 #ifdef INET6
310 	if (inp->inp_vflag & INP_IPV6) {
311 		bzero(&udp_in6, sizeof(udp_in6));
312 		udp_in6.sin6_len = sizeof(udp_in6);
313 		udp_in6.sin6_family = AF_INET6;
314 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
315 		append_sa = (struct sockaddr *)&udp_in6;
316 	} else
317 #endif /* INET6 */
318 		append_sa = (struct sockaddr *)udp_in;
319 	m_adj(n, off);
320 
321 	so = inp->inp_socket;
322 	SOCKBUF_LOCK(&so->so_rcv);
323 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
324 		SOCKBUF_UNLOCK(&so->so_rcv);
325 		m_freem(n);
326 		if (opts)
327 			m_freem(opts);
328 		UDPSTAT_INC(udps_fullsock);
329 	} else
330 		sorwakeup_locked(so);
331 }
332 
333 void
334 udp_input(struct mbuf *m, int off)
335 {
336 	int iphlen = off;
337 	struct ip *ip;
338 	struct udphdr *uh;
339 	struct ifnet *ifp;
340 	struct inpcb *inp;
341 	uint16_t len, ip_len;
342 	struct ip save_ip;
343 	struct sockaddr_in udp_in;
344 	struct m_tag *fwd_tag;
345 
346 	ifp = m->m_pkthdr.rcvif;
347 	UDPSTAT_INC(udps_ipackets);
348 
349 	/*
350 	 * Strip IP options, if any; should skip this, make available to
351 	 * user, and use on returned packets, but we don't yet have a way to
352 	 * check the checksum with options still present.
353 	 */
354 	if (iphlen > sizeof (struct ip)) {
355 		ip_stripoptions(m);
356 		iphlen = sizeof(struct ip);
357 	}
358 
359 	/*
360 	 * Get IP and UDP header together in first mbuf.
361 	 */
362 	ip = mtod(m, struct ip *);
363 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
364 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
365 			UDPSTAT_INC(udps_hdrops);
366 			return;
367 		}
368 		ip = mtod(m, struct ip *);
369 	}
370 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
371 
372 	/*
373 	 * Destination port of 0 is illegal, based on RFC768.
374 	 */
375 	if (uh->uh_dport == 0)
376 		goto badunlocked;
377 
378 	/*
379 	 * Construct sockaddr format source address.  Stuff source address
380 	 * and datagram in user buffer.
381 	 */
382 	bzero(&udp_in, sizeof(udp_in));
383 	udp_in.sin_len = sizeof(udp_in);
384 	udp_in.sin_family = AF_INET;
385 	udp_in.sin_port = uh->uh_sport;
386 	udp_in.sin_addr = ip->ip_src;
387 
388 	/*
389 	 * Make mbuf data length reflect UDP length.  If not enough data to
390 	 * reflect UDP length, drop.
391 	 */
392 	len = ntohs((u_short)uh->uh_ulen);
393 	ip_len = ntohs(ip->ip_len) - iphlen;
394 	if (ip_len != len) {
395 		if (len > ip_len || len < sizeof(struct udphdr)) {
396 			UDPSTAT_INC(udps_badlen);
397 			goto badunlocked;
398 		}
399 		m_adj(m, len - ip_len);
400 	}
401 
402 	/*
403 	 * Save a copy of the IP header in case we want restore it for
404 	 * sending an ICMP error message in response.
405 	 */
406 	if (!V_udp_blackhole)
407 		save_ip = *ip;
408 	else
409 		memset(&save_ip, 0, sizeof(save_ip));
410 
411 	/*
412 	 * Checksum extended UDP header and data.
413 	 */
414 	if (uh->uh_sum) {
415 		u_short uh_sum;
416 
417 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
418 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
419 				uh_sum = m->m_pkthdr.csum_data;
420 			else
421 				uh_sum = in_pseudo(ip->ip_src.s_addr,
422 				    ip->ip_dst.s_addr, htonl((u_short)len +
423 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
424 			uh_sum ^= 0xffff;
425 		} else {
426 			char b[9];
427 
428 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
429 			bzero(((struct ipovly *)ip)->ih_x1, 9);
430 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
431 			uh_sum = in_cksum(m, len + sizeof (struct ip));
432 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
433 		}
434 		if (uh_sum) {
435 			UDPSTAT_INC(udps_badsum);
436 			m_freem(m);
437 			return;
438 		}
439 	} else
440 		UDPSTAT_INC(udps_nosum);
441 
442 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
443 	    in_broadcast(ip->ip_dst, ifp)) {
444 		struct inpcb *last;
445 		struct ip_moptions *imo;
446 
447 		INP_INFO_RLOCK(&V_udbinfo);
448 		last = NULL;
449 		LIST_FOREACH(inp, &V_udb, inp_list) {
450 			if (inp->inp_lport != uh->uh_dport)
451 				continue;
452 #ifdef INET6
453 			if ((inp->inp_vflag & INP_IPV4) == 0)
454 				continue;
455 #endif
456 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
457 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
458 				continue;
459 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
460 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
461 				continue;
462 			if (inp->inp_fport != 0 &&
463 			    inp->inp_fport != uh->uh_sport)
464 				continue;
465 
466 			INP_RLOCK(inp);
467 
468 			/*
469 			 * XXXRW: Because we weren't holding either the inpcb
470 			 * or the hash lock when we checked for a match
471 			 * before, we should probably recheck now that the
472 			 * inpcb lock is held.
473 			 */
474 
475 			/*
476 			 * Handle socket delivery policy for any-source
477 			 * and source-specific multicast. [RFC3678]
478 			 */
479 			imo = inp->inp_moptions;
480 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
481 				struct sockaddr_in	 group;
482 				int			 blocked;
483 				if (imo == NULL) {
484 					INP_RUNLOCK(inp);
485 					continue;
486 				}
487 				bzero(&group, sizeof(struct sockaddr_in));
488 				group.sin_len = sizeof(struct sockaddr_in);
489 				group.sin_family = AF_INET;
490 				group.sin_addr = ip->ip_dst;
491 
492 				blocked = imo_multi_filter(imo, ifp,
493 					(struct sockaddr *)&group,
494 					(struct sockaddr *)&udp_in);
495 				if (blocked != MCAST_PASS) {
496 					if (blocked == MCAST_NOTGMEMBER)
497 						IPSTAT_INC(ips_notmember);
498 					if (blocked == MCAST_NOTSMEMBER ||
499 					    blocked == MCAST_MUTED)
500 						UDPSTAT_INC(udps_filtermcast);
501 					INP_RUNLOCK(inp);
502 					continue;
503 				}
504 			}
505 			if (last != NULL) {
506 				struct mbuf *n;
507 
508 				n = m_copy(m, 0, M_COPYALL);
509 				udp_append(last, ip, n, iphlen, &udp_in);
510 				INP_RUNLOCK(last);
511 			}
512 			last = inp;
513 			/*
514 			 * Don't look for additional matches if this one does
515 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
516 			 * socket options set.  This heuristic avoids
517 			 * searching through all pcbs in the common case of a
518 			 * non-shared port.  It assumes that an application
519 			 * will never clear these options after setting them.
520 			 */
521 			if ((last->inp_socket->so_options &
522 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
523 				break;
524 		}
525 
526 		if (last == NULL) {
527 			/*
528 			 * No matching pcb found; discard datagram.  (No need
529 			 * to send an ICMP Port Unreachable for a broadcast
530 			 * or multicast datgram.)
531 			 */
532 			UDPSTAT_INC(udps_noportbcast);
533 			if (inp)
534 				INP_RUNLOCK(inp);
535 			INP_INFO_RUNLOCK(&V_udbinfo);
536 			goto badunlocked;
537 		}
538 		udp_append(last, ip, m, iphlen, &udp_in);
539 		INP_RUNLOCK(last);
540 		INP_INFO_RUNLOCK(&V_udbinfo);
541 		return;
542 	}
543 
544 	/*
545 	 * Locate pcb for datagram.
546 	 */
547 
548 	/*
549 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
550 	 */
551 	if ((m->m_flags & M_IP_NEXTHOP) &&
552 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
553 		struct sockaddr_in *next_hop;
554 
555 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
556 
557 		/*
558 		 * Transparently forwarded. Pretend to be the destination.
559 		 * Already got one like this?
560 		 */
561 		inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
562 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
563 		if (!inp) {
564 			/*
565 			 * It's new.  Try to find the ambushing socket.
566 			 * Because we've rewritten the destination address,
567 			 * any hardware-generated hash is ignored.
568 			 */
569 			inp = in_pcblookup(&V_udbinfo, ip->ip_src,
570 			    uh->uh_sport, next_hop->sin_addr,
571 			    next_hop->sin_port ? htons(next_hop->sin_port) :
572 			    uh->uh_dport, INPLOOKUP_WILDCARD |
573 			    INPLOOKUP_RLOCKPCB, ifp);
574 		}
575 		/* Remove the tag from the packet. We don't need it anymore. */
576 		m_tag_delete(m, fwd_tag);
577 		m->m_flags &= ~M_IP_NEXTHOP;
578 	} else
579 		inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
580 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
581 		    INPLOOKUP_RLOCKPCB, ifp, m);
582 	if (inp == NULL) {
583 		if (udp_log_in_vain) {
584 			char buf[4*sizeof "123"];
585 
586 			strcpy(buf, inet_ntoa(ip->ip_dst));
587 			log(LOG_INFO,
588 			    "Connection attempt to UDP %s:%d from %s:%d\n",
589 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
590 			    ntohs(uh->uh_sport));
591 		}
592 		UDPSTAT_INC(udps_noport);
593 		if (m->m_flags & (M_BCAST | M_MCAST)) {
594 			UDPSTAT_INC(udps_noportbcast);
595 			goto badunlocked;
596 		}
597 		if (V_udp_blackhole)
598 			goto badunlocked;
599 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
600 			goto badunlocked;
601 		*ip = save_ip;
602 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
603 		return;
604 	}
605 
606 	/*
607 	 * Check the minimum TTL for socket.
608 	 */
609 	INP_RLOCK_ASSERT(inp);
610 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
611 		INP_RUNLOCK(inp);
612 		m_freem(m);
613 		return;
614 	}
615 	udp_append(inp, ip, m, iphlen, &udp_in);
616 	INP_RUNLOCK(inp);
617 	return;
618 
619 badunlocked:
620 	m_freem(m);
621 }
622 #endif /* INET */
623 
624 /*
625  * Notify a udp user of an asynchronous error; just wake up so that they can
626  * collect error status.
627  */
628 struct inpcb *
629 udp_notify(struct inpcb *inp, int errno)
630 {
631 
632 	/*
633 	 * While udp_ctlinput() always calls udp_notify() with a read lock
634 	 * when invoking it directly, in_pcbnotifyall() currently uses write
635 	 * locks due to sharing code with TCP.  For now, accept either a read
636 	 * or a write lock, but a read lock is sufficient.
637 	 */
638 	INP_LOCK_ASSERT(inp);
639 
640 	inp->inp_socket->so_error = errno;
641 	sorwakeup(inp->inp_socket);
642 	sowwakeup(inp->inp_socket);
643 	return (inp);
644 }
645 
646 #ifdef INET
647 void
648 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
649 {
650 	struct ip *ip = vip;
651 	struct udphdr *uh;
652 	struct in_addr faddr;
653 	struct inpcb *inp;
654 
655 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
656 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
657 		return;
658 
659 	/*
660 	 * Redirects don't need to be handled up here.
661 	 */
662 	if (PRC_IS_REDIRECT(cmd))
663 		return;
664 
665 	/*
666 	 * Hostdead is ugly because it goes linearly through all PCBs.
667 	 *
668 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
669 	 * DoS attack on machines with many connections.
670 	 */
671 	if (cmd == PRC_HOSTDEAD)
672 		ip = NULL;
673 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
674 		return;
675 	if (ip != NULL) {
676 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
677 		inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
678 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
679 		if (inp != NULL) {
680 			INP_RLOCK_ASSERT(inp);
681 			if (inp->inp_socket != NULL) {
682 				udp_notify(inp, inetctlerrmap[cmd]);
683 			}
684 			INP_RUNLOCK(inp);
685 		}
686 	} else
687 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
688 		    udp_notify);
689 }
690 #endif /* INET */
691 
692 static int
693 udp_pcblist(SYSCTL_HANDLER_ARGS)
694 {
695 	int error, i, n;
696 	struct inpcb *inp, **inp_list;
697 	inp_gen_t gencnt;
698 	struct xinpgen xig;
699 
700 	/*
701 	 * The process of preparing the PCB list is too time-consuming and
702 	 * resource-intensive to repeat twice on every request.
703 	 */
704 	if (req->oldptr == 0) {
705 		n = V_udbinfo.ipi_count;
706 		n += imax(n / 8, 10);
707 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
708 		return (0);
709 	}
710 
711 	if (req->newptr != 0)
712 		return (EPERM);
713 
714 	/*
715 	 * OK, now we're committed to doing something.
716 	 */
717 	INP_INFO_RLOCK(&V_udbinfo);
718 	gencnt = V_udbinfo.ipi_gencnt;
719 	n = V_udbinfo.ipi_count;
720 	INP_INFO_RUNLOCK(&V_udbinfo);
721 
722 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
723 		+ n * sizeof(struct xinpcb));
724 	if (error != 0)
725 		return (error);
726 
727 	xig.xig_len = sizeof xig;
728 	xig.xig_count = n;
729 	xig.xig_gen = gencnt;
730 	xig.xig_sogen = so_gencnt;
731 	error = SYSCTL_OUT(req, &xig, sizeof xig);
732 	if (error)
733 		return (error);
734 
735 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
736 	if (inp_list == 0)
737 		return (ENOMEM);
738 
739 	INP_INFO_RLOCK(&V_udbinfo);
740 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
741 	     inp = LIST_NEXT(inp, inp_list)) {
742 		INP_WLOCK(inp);
743 		if (inp->inp_gencnt <= gencnt &&
744 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
745 			in_pcbref(inp);
746 			inp_list[i++] = inp;
747 		}
748 		INP_WUNLOCK(inp);
749 	}
750 	INP_INFO_RUNLOCK(&V_udbinfo);
751 	n = i;
752 
753 	error = 0;
754 	for (i = 0; i < n; i++) {
755 		inp = inp_list[i];
756 		INP_RLOCK(inp);
757 		if (inp->inp_gencnt <= gencnt) {
758 			struct xinpcb xi;
759 
760 			bzero(&xi, sizeof(xi));
761 			xi.xi_len = sizeof xi;
762 			/* XXX should avoid extra copy */
763 			bcopy(inp, &xi.xi_inp, sizeof *inp);
764 			if (inp->inp_socket)
765 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
766 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
767 			INP_RUNLOCK(inp);
768 			error = SYSCTL_OUT(req, &xi, sizeof xi);
769 		} else
770 			INP_RUNLOCK(inp);
771 	}
772 	INP_INFO_WLOCK(&V_udbinfo);
773 	for (i = 0; i < n; i++) {
774 		inp = inp_list[i];
775 		INP_RLOCK(inp);
776 		if (!in_pcbrele_rlocked(inp))
777 			INP_RUNLOCK(inp);
778 	}
779 	INP_INFO_WUNLOCK(&V_udbinfo);
780 
781 	if (!error) {
782 		/*
783 		 * Give the user an updated idea of our state.  If the
784 		 * generation differs from what we told her before, she knows
785 		 * that something happened while we were processing this
786 		 * request, and it might be necessary to retry.
787 		 */
788 		INP_INFO_RLOCK(&V_udbinfo);
789 		xig.xig_gen = V_udbinfo.ipi_gencnt;
790 		xig.xig_sogen = so_gencnt;
791 		xig.xig_count = V_udbinfo.ipi_count;
792 		INP_INFO_RUNLOCK(&V_udbinfo);
793 		error = SYSCTL_OUT(req, &xig, sizeof xig);
794 	}
795 	free(inp_list, M_TEMP);
796 	return (error);
797 }
798 
799 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
800     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
801     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
802 
803 #ifdef INET
804 static int
805 udp_getcred(SYSCTL_HANDLER_ARGS)
806 {
807 	struct xucred xuc;
808 	struct sockaddr_in addrs[2];
809 	struct inpcb *inp;
810 	int error;
811 
812 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
813 	if (error)
814 		return (error);
815 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
816 	if (error)
817 		return (error);
818 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
819 	    addrs[0].sin_addr, addrs[0].sin_port,
820 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
821 	if (inp != NULL) {
822 		INP_RLOCK_ASSERT(inp);
823 		if (inp->inp_socket == NULL)
824 			error = ENOENT;
825 		if (error == 0)
826 			error = cr_canseeinpcb(req->td->td_ucred, inp);
827 		if (error == 0)
828 			cru2x(inp->inp_cred, &xuc);
829 		INP_RUNLOCK(inp);
830 	} else
831 		error = ENOENT;
832 	if (error == 0)
833 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
834 	return (error);
835 }
836 
837 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
838     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
839     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
840 #endif /* INET */
841 
842 int
843 udp_ctloutput(struct socket *so, struct sockopt *sopt)
844 {
845 	int error = 0, optval;
846 	struct inpcb *inp;
847 #ifdef IPSEC_NAT_T
848 	struct udpcb *up;
849 #endif
850 
851 	inp = sotoinpcb(so);
852 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
853 	INP_WLOCK(inp);
854 	if (sopt->sopt_level != IPPROTO_UDP) {
855 #ifdef INET6
856 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
857 			INP_WUNLOCK(inp);
858 			error = ip6_ctloutput(so, sopt);
859 		}
860 #endif
861 #if defined(INET) && defined(INET6)
862 		else
863 #endif
864 #ifdef INET
865 		{
866 			INP_WUNLOCK(inp);
867 			error = ip_ctloutput(so, sopt);
868 		}
869 #endif
870 		return (error);
871 	}
872 
873 	switch (sopt->sopt_dir) {
874 	case SOPT_SET:
875 		switch (sopt->sopt_name) {
876 		case UDP_ENCAP:
877 			INP_WUNLOCK(inp);
878 			error = sooptcopyin(sopt, &optval, sizeof optval,
879 					    sizeof optval);
880 			if (error)
881 				break;
882 			inp = sotoinpcb(so);
883 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
884 			INP_WLOCK(inp);
885 #ifdef IPSEC_NAT_T
886 			up = intoudpcb(inp);
887 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
888 #endif
889 			switch (optval) {
890 			case 0:
891 				/* Clear all UDP encap. */
892 #ifdef IPSEC_NAT_T
893 				up->u_flags &= ~UF_ESPINUDP_ALL;
894 #endif
895 				break;
896 #ifdef IPSEC_NAT_T
897 			case UDP_ENCAP_ESPINUDP:
898 			case UDP_ENCAP_ESPINUDP_NON_IKE:
899 				up->u_flags &= ~UF_ESPINUDP_ALL;
900 				if (optval == UDP_ENCAP_ESPINUDP)
901 					up->u_flags |= UF_ESPINUDP;
902 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
903 					up->u_flags |= UF_ESPINUDP_NON_IKE;
904 				break;
905 #endif
906 			default:
907 				error = EINVAL;
908 				break;
909 			}
910 			INP_WUNLOCK(inp);
911 			break;
912 		default:
913 			INP_WUNLOCK(inp);
914 			error = ENOPROTOOPT;
915 			break;
916 		}
917 		break;
918 	case SOPT_GET:
919 		switch (sopt->sopt_name) {
920 #ifdef IPSEC_NAT_T
921 		case UDP_ENCAP:
922 			up = intoudpcb(inp);
923 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
924 			optval = up->u_flags & UF_ESPINUDP_ALL;
925 			INP_WUNLOCK(inp);
926 			error = sooptcopyout(sopt, &optval, sizeof optval);
927 			break;
928 #endif
929 		default:
930 			INP_WUNLOCK(inp);
931 			error = ENOPROTOOPT;
932 			break;
933 		}
934 		break;
935 	}
936 	return (error);
937 }
938 
939 #ifdef INET
940 #define	UH_WLOCKED	2
941 #define	UH_RLOCKED	1
942 #define	UH_UNLOCKED	0
943 static int
944 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
945     struct mbuf *control, struct thread *td)
946 {
947 	struct udpiphdr *ui;
948 	int len = m->m_pkthdr.len;
949 	struct in_addr faddr, laddr;
950 	struct cmsghdr *cm;
951 	struct sockaddr_in *sin, src;
952 	int error = 0;
953 	int ipflags;
954 	u_short fport, lport;
955 	int unlock_udbinfo;
956 	u_char tos;
957 
958 	/*
959 	 * udp_output() may need to temporarily bind or connect the current
960 	 * inpcb.  As such, we don't know up front whether we will need the
961 	 * pcbinfo lock or not.  Do any work to decide what is needed up
962 	 * front before acquiring any locks.
963 	 */
964 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
965 		if (control)
966 			m_freem(control);
967 		m_freem(m);
968 		return (EMSGSIZE);
969 	}
970 
971 	src.sin_family = 0;
972 	INP_RLOCK(inp);
973 	tos = inp->inp_ip_tos;
974 	if (control != NULL) {
975 		/*
976 		 * XXX: Currently, we assume all the optional information is
977 		 * stored in a single mbuf.
978 		 */
979 		if (control->m_next) {
980 			INP_RUNLOCK(inp);
981 			m_freem(control);
982 			m_freem(m);
983 			return (EINVAL);
984 		}
985 		for (; control->m_len > 0;
986 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
987 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
988 			cm = mtod(control, struct cmsghdr *);
989 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
990 			    || cm->cmsg_len > control->m_len) {
991 				error = EINVAL;
992 				break;
993 			}
994 			if (cm->cmsg_level != IPPROTO_IP)
995 				continue;
996 
997 			switch (cm->cmsg_type) {
998 			case IP_SENDSRCADDR:
999 				if (cm->cmsg_len !=
1000 				    CMSG_LEN(sizeof(struct in_addr))) {
1001 					error = EINVAL;
1002 					break;
1003 				}
1004 				bzero(&src, sizeof(src));
1005 				src.sin_family = AF_INET;
1006 				src.sin_len = sizeof(src);
1007 				src.sin_port = inp->inp_lport;
1008 				src.sin_addr =
1009 				    *(struct in_addr *)CMSG_DATA(cm);
1010 				break;
1011 
1012 			case IP_TOS:
1013 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1014 					error = EINVAL;
1015 					break;
1016 				}
1017 				tos = *(u_char *)CMSG_DATA(cm);
1018 				break;
1019 
1020 			default:
1021 				error = ENOPROTOOPT;
1022 				break;
1023 			}
1024 			if (error)
1025 				break;
1026 		}
1027 		m_freem(control);
1028 	}
1029 	if (error) {
1030 		INP_RUNLOCK(inp);
1031 		m_freem(m);
1032 		return (error);
1033 	}
1034 
1035 	/*
1036 	 * Depending on whether or not the application has bound or connected
1037 	 * the socket, we may have to do varying levels of work.  The optimal
1038 	 * case is for a connected UDP socket, as a global lock isn't
1039 	 * required at all.
1040 	 *
1041 	 * In order to decide which we need, we require stability of the
1042 	 * inpcb binding, which we ensure by acquiring a read lock on the
1043 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1044 	 * the trylock and retry game; note that we may end up with more
1045 	 * conservative locks than required the second time around, so later
1046 	 * assertions have to accept that.  Further analysis of the number of
1047 	 * misses under contention is required.
1048 	 *
1049 	 * XXXRW: Check that hash locking update here is correct.
1050 	 */
1051 	sin = (struct sockaddr_in *)addr;
1052 	if (sin != NULL &&
1053 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1054 		INP_RUNLOCK(inp);
1055 		INP_WLOCK(inp);
1056 		INP_HASH_WLOCK(&V_udbinfo);
1057 		unlock_udbinfo = UH_WLOCKED;
1058 	} else if ((sin != NULL && (
1059 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1060 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1061 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1062 	    (inp->inp_lport == 0))) ||
1063 	    (src.sin_family == AF_INET)) {
1064 		INP_HASH_RLOCK(&V_udbinfo);
1065 		unlock_udbinfo = UH_RLOCKED;
1066 	} else
1067 		unlock_udbinfo = UH_UNLOCKED;
1068 
1069 	/*
1070 	 * If the IP_SENDSRCADDR control message was specified, override the
1071 	 * source address for this datagram.  Its use is invalidated if the
1072 	 * address thus specified is incomplete or clobbers other inpcbs.
1073 	 */
1074 	laddr = inp->inp_laddr;
1075 	lport = inp->inp_lport;
1076 	if (src.sin_family == AF_INET) {
1077 		INP_HASH_LOCK_ASSERT(&V_udbinfo);
1078 		if ((lport == 0) ||
1079 		    (laddr.s_addr == INADDR_ANY &&
1080 		     src.sin_addr.s_addr == INADDR_ANY)) {
1081 			error = EINVAL;
1082 			goto release;
1083 		}
1084 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1085 		    &laddr.s_addr, &lport, td->td_ucred);
1086 		if (error)
1087 			goto release;
1088 	}
1089 
1090 	/*
1091 	 * If a UDP socket has been connected, then a local address/port will
1092 	 * have been selected and bound.
1093 	 *
1094 	 * If a UDP socket has not been connected to, then an explicit
1095 	 * destination address must be used, in which case a local
1096 	 * address/port may not have been selected and bound.
1097 	 */
1098 	if (sin != NULL) {
1099 		INP_LOCK_ASSERT(inp);
1100 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1101 			error = EISCONN;
1102 			goto release;
1103 		}
1104 
1105 		/*
1106 		 * Jail may rewrite the destination address, so let it do
1107 		 * that before we use it.
1108 		 */
1109 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1110 		if (error)
1111 			goto release;
1112 
1113 		/*
1114 		 * If a local address or port hasn't yet been selected, or if
1115 		 * the destination address needs to be rewritten due to using
1116 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1117 		 * to do the heavy lifting.  Once a port is selected, we
1118 		 * commit the binding back to the socket; we also commit the
1119 		 * binding of the address if in jail.
1120 		 *
1121 		 * If we already have a valid binding and we're not
1122 		 * requesting a destination address rewrite, use a fast path.
1123 		 */
1124 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1125 		    inp->inp_lport == 0 ||
1126 		    sin->sin_addr.s_addr == INADDR_ANY ||
1127 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1128 			INP_HASH_LOCK_ASSERT(&V_udbinfo);
1129 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1130 			    &lport, &faddr.s_addr, &fport, NULL,
1131 			    td->td_ucred);
1132 			if (error)
1133 				goto release;
1134 
1135 			/*
1136 			 * XXXRW: Why not commit the port if the address is
1137 			 * !INADDR_ANY?
1138 			 */
1139 			/* Commit the local port if newly assigned. */
1140 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1141 			    inp->inp_lport == 0) {
1142 				INP_WLOCK_ASSERT(inp);
1143 				INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1144 				/*
1145 				 * Remember addr if jailed, to prevent
1146 				 * rebinding.
1147 				 */
1148 				if (prison_flag(td->td_ucred, PR_IP4))
1149 					inp->inp_laddr = laddr;
1150 				inp->inp_lport = lport;
1151 				if (in_pcbinshash(inp) != 0) {
1152 					inp->inp_lport = 0;
1153 					error = EAGAIN;
1154 					goto release;
1155 				}
1156 				inp->inp_flags |= INP_ANONPORT;
1157 			}
1158 		} else {
1159 			faddr = sin->sin_addr;
1160 			fport = sin->sin_port;
1161 		}
1162 	} else {
1163 		INP_LOCK_ASSERT(inp);
1164 		faddr = inp->inp_faddr;
1165 		fport = inp->inp_fport;
1166 		if (faddr.s_addr == INADDR_ANY) {
1167 			error = ENOTCONN;
1168 			goto release;
1169 		}
1170 	}
1171 
1172 	/*
1173 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1174 	 * link-layer headers.  Immediate slide the data pointer back forward
1175 	 * since we won't use that space at this layer.
1176 	 */
1177 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1178 	if (m == NULL) {
1179 		error = ENOBUFS;
1180 		goto release;
1181 	}
1182 	m->m_data += max_linkhdr;
1183 	m->m_len -= max_linkhdr;
1184 	m->m_pkthdr.len -= max_linkhdr;
1185 
1186 	/*
1187 	 * Fill in mbuf with extended UDP header and addresses and length put
1188 	 * into network format.
1189 	 */
1190 	ui = mtod(m, struct udpiphdr *);
1191 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1192 	ui->ui_pr = IPPROTO_UDP;
1193 	ui->ui_src = laddr;
1194 	ui->ui_dst = faddr;
1195 	ui->ui_sport = lport;
1196 	ui->ui_dport = fport;
1197 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1198 
1199 	/*
1200 	 * Set the Don't Fragment bit in the IP header.
1201 	 */
1202 	if (inp->inp_flags & INP_DONTFRAG) {
1203 		struct ip *ip;
1204 
1205 		ip = (struct ip *)&ui->ui_i;
1206 		ip->ip_off |= htons(IP_DF);
1207 	}
1208 
1209 	ipflags = 0;
1210 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1211 		ipflags |= IP_ROUTETOIF;
1212 	if (inp->inp_socket->so_options & SO_BROADCAST)
1213 		ipflags |= IP_ALLOWBROADCAST;
1214 	if (inp->inp_flags & INP_ONESBCAST)
1215 		ipflags |= IP_SENDONES;
1216 
1217 #ifdef MAC
1218 	mac_inpcb_create_mbuf(inp, m);
1219 #endif
1220 
1221 	/*
1222 	 * Set up checksum and output datagram.
1223 	 */
1224 	if (V_udp_cksum) {
1225 		if (inp->inp_flags & INP_ONESBCAST)
1226 			faddr.s_addr = INADDR_BROADCAST;
1227 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1228 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1229 		m->m_pkthdr.csum_flags = CSUM_UDP;
1230 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1231 	} else
1232 		ui->ui_sum = 0;
1233 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1234 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1235 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1236 	UDPSTAT_INC(udps_opackets);
1237 
1238 	if (unlock_udbinfo == UH_WLOCKED)
1239 		INP_HASH_WUNLOCK(&V_udbinfo);
1240 	else if (unlock_udbinfo == UH_RLOCKED)
1241 		INP_HASH_RUNLOCK(&V_udbinfo);
1242 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1243 	    inp->inp_moptions, inp);
1244 	if (unlock_udbinfo == UH_WLOCKED)
1245 		INP_WUNLOCK(inp);
1246 	else
1247 		INP_RUNLOCK(inp);
1248 	return (error);
1249 
1250 release:
1251 	if (unlock_udbinfo == UH_WLOCKED) {
1252 		INP_HASH_WUNLOCK(&V_udbinfo);
1253 		INP_WUNLOCK(inp);
1254 	} else if (unlock_udbinfo == UH_RLOCKED) {
1255 		INP_HASH_RUNLOCK(&V_udbinfo);
1256 		INP_RUNLOCK(inp);
1257 	} else
1258 		INP_RUNLOCK(inp);
1259 	m_freem(m);
1260 	return (error);
1261 }
1262 
1263 
1264 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1265 /*
1266  * Potentially decap ESP in UDP frame.  Check for an ESP header
1267  * and optional marker; if present, strip the UDP header and
1268  * push the result through IPSec.
1269  *
1270  * Returns mbuf to be processed (potentially re-allocated) or
1271  * NULL if consumed and/or processed.
1272  */
1273 static struct mbuf *
1274 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1275 {
1276 	size_t minlen, payload, skip, iphlen;
1277 	caddr_t data;
1278 	struct udpcb *up;
1279 	struct m_tag *tag;
1280 	struct udphdr *udphdr;
1281 	struct ip *ip;
1282 
1283 	INP_RLOCK_ASSERT(inp);
1284 
1285 	/*
1286 	 * Pull up data so the longest case is contiguous:
1287 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1288 	 */
1289 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1290 	if (minlen > m->m_pkthdr.len)
1291 		minlen = m->m_pkthdr.len;
1292 	if ((m = m_pullup(m, minlen)) == NULL) {
1293 		V_ipsec4stat.in_inval++;
1294 		return (NULL);		/* Bypass caller processing. */
1295 	}
1296 	data = mtod(m, caddr_t);	/* Points to ip header. */
1297 	payload = m->m_len - off;	/* Size of payload. */
1298 
1299 	if (payload == 1 && data[off] == '\xff')
1300 		return (m);		/* NB: keepalive packet, no decap. */
1301 
1302 	up = intoudpcb(inp);
1303 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1304 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1305 	    ("u_flags 0x%x", up->u_flags));
1306 
1307 	/*
1308 	 * Check that the payload is large enough to hold an
1309 	 * ESP header and compute the amount of data to remove.
1310 	 *
1311 	 * NB: the caller has already done a pullup for us.
1312 	 * XXX can we assume alignment and eliminate bcopys?
1313 	 */
1314 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1315 		/*
1316 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1317 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1318 		 * possible AH mode non-IKE marker+non-ESP marker
1319 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1320 		 */
1321 		uint64_t marker;
1322 
1323 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1324 			return (m);	/* NB: no decap. */
1325 		bcopy(data + off, &marker, sizeof(uint64_t));
1326 		if (marker != 0)	/* Non-IKE marker. */
1327 			return (m);	/* NB: no decap. */
1328 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1329 	} else {
1330 		uint32_t spi;
1331 
1332 		if (payload <= sizeof(struct esp)) {
1333 			V_ipsec4stat.in_inval++;
1334 			m_freem(m);
1335 			return (NULL);	/* Discard. */
1336 		}
1337 		bcopy(data + off, &spi, sizeof(uint32_t));
1338 		if (spi == 0)		/* Non-ESP marker. */
1339 			return (m);	/* NB: no decap. */
1340 		skip = sizeof(struct udphdr);
1341 	}
1342 
1343 	/*
1344 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1345 	 * the UDP ports. This is required if we want to select
1346 	 * the right SPD for multiple hosts behind same NAT.
1347 	 *
1348 	 * NB: ports are maintained in network byte order everywhere
1349 	 *     in the NAT-T code.
1350 	 */
1351 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1352 		2 * sizeof(uint16_t), M_NOWAIT);
1353 	if (tag == NULL) {
1354 		V_ipsec4stat.in_nomem++;
1355 		m_freem(m);
1356 		return (NULL);		/* Discard. */
1357 	}
1358 	iphlen = off - sizeof(struct udphdr);
1359 	udphdr = (struct udphdr *)(data + iphlen);
1360 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1361 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1362 	m_tag_prepend(m, tag);
1363 
1364 	/*
1365 	 * Remove the UDP header (and possibly the non ESP marker)
1366 	 * IP header length is iphlen
1367 	 * Before:
1368 	 *   <--- off --->
1369 	 *   +----+------+-----+
1370 	 *   | IP |  UDP | ESP |
1371 	 *   +----+------+-----+
1372 	 *        <-skip->
1373 	 * After:
1374 	 *          +----+-----+
1375 	 *          | IP | ESP |
1376 	 *          +----+-----+
1377 	 *   <-skip->
1378 	 */
1379 	ovbcopy(data, data + skip, iphlen);
1380 	m_adj(m, skip);
1381 
1382 	ip = mtod(m, struct ip *);
1383 	ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1384 	ip->ip_p = IPPROTO_ESP;
1385 
1386 	/*
1387 	 * We cannot yet update the cksums so clear any
1388 	 * h/w cksum flags as they are no longer valid.
1389 	 */
1390 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1391 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1392 
1393 	(void) ipsec4_common_input(m, iphlen, ip->ip_p);
1394 	return (NULL);			/* NB: consumed, bypass processing. */
1395 }
1396 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1397 
1398 static void
1399 udp_abort(struct socket *so)
1400 {
1401 	struct inpcb *inp;
1402 
1403 	inp = sotoinpcb(so);
1404 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1405 	INP_WLOCK(inp);
1406 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1407 		INP_HASH_WLOCK(&V_udbinfo);
1408 		in_pcbdisconnect(inp);
1409 		inp->inp_laddr.s_addr = INADDR_ANY;
1410 		INP_HASH_WUNLOCK(&V_udbinfo);
1411 		soisdisconnected(so);
1412 	}
1413 	INP_WUNLOCK(inp);
1414 }
1415 
1416 static int
1417 udp_attach(struct socket *so, int proto, struct thread *td)
1418 {
1419 	struct inpcb *inp;
1420 	int error;
1421 
1422 	inp = sotoinpcb(so);
1423 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1424 	error = soreserve(so, udp_sendspace, udp_recvspace);
1425 	if (error)
1426 		return (error);
1427 	INP_INFO_WLOCK(&V_udbinfo);
1428 	error = in_pcballoc(so, &V_udbinfo);
1429 	if (error) {
1430 		INP_INFO_WUNLOCK(&V_udbinfo);
1431 		return (error);
1432 	}
1433 
1434 	inp = sotoinpcb(so);
1435 	inp->inp_vflag |= INP_IPV4;
1436 	inp->inp_ip_ttl = V_ip_defttl;
1437 
1438 	error = udp_newudpcb(inp);
1439 	if (error) {
1440 		in_pcbdetach(inp);
1441 		in_pcbfree(inp);
1442 		INP_INFO_WUNLOCK(&V_udbinfo);
1443 		return (error);
1444 	}
1445 
1446 	INP_WUNLOCK(inp);
1447 	INP_INFO_WUNLOCK(&V_udbinfo);
1448 	return (0);
1449 }
1450 #endif /* INET */
1451 
1452 int
1453 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1454 {
1455 	struct inpcb *inp;
1456 	struct udpcb *up;
1457 
1458 	KASSERT(so->so_type == SOCK_DGRAM,
1459 	    ("udp_set_kernel_tunneling: !dgram"));
1460 	inp = sotoinpcb(so);
1461 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1462 	INP_WLOCK(inp);
1463 	up = intoudpcb(inp);
1464 	if (up->u_tun_func != NULL) {
1465 		INP_WUNLOCK(inp);
1466 		return (EBUSY);
1467 	}
1468 	up->u_tun_func = f;
1469 	INP_WUNLOCK(inp);
1470 	return (0);
1471 }
1472 
1473 #ifdef INET
1474 static int
1475 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1476 {
1477 	struct inpcb *inp;
1478 	int error;
1479 
1480 	inp = sotoinpcb(so);
1481 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1482 	INP_WLOCK(inp);
1483 	INP_HASH_WLOCK(&V_udbinfo);
1484 	error = in_pcbbind(inp, nam, td->td_ucred);
1485 	INP_HASH_WUNLOCK(&V_udbinfo);
1486 	INP_WUNLOCK(inp);
1487 	return (error);
1488 }
1489 
1490 static void
1491 udp_close(struct socket *so)
1492 {
1493 	struct inpcb *inp;
1494 
1495 	inp = sotoinpcb(so);
1496 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1497 	INP_WLOCK(inp);
1498 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1499 		INP_HASH_WLOCK(&V_udbinfo);
1500 		in_pcbdisconnect(inp);
1501 		inp->inp_laddr.s_addr = INADDR_ANY;
1502 		INP_HASH_WUNLOCK(&V_udbinfo);
1503 		soisdisconnected(so);
1504 	}
1505 	INP_WUNLOCK(inp);
1506 }
1507 
1508 static int
1509 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1510 {
1511 	struct inpcb *inp;
1512 	int error;
1513 	struct sockaddr_in *sin;
1514 
1515 	inp = sotoinpcb(so);
1516 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1517 	INP_WLOCK(inp);
1518 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1519 		INP_WUNLOCK(inp);
1520 		return (EISCONN);
1521 	}
1522 	sin = (struct sockaddr_in *)nam;
1523 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1524 	if (error != 0) {
1525 		INP_WUNLOCK(inp);
1526 		return (error);
1527 	}
1528 	INP_HASH_WLOCK(&V_udbinfo);
1529 	error = in_pcbconnect(inp, nam, td->td_ucred);
1530 	INP_HASH_WUNLOCK(&V_udbinfo);
1531 	if (error == 0)
1532 		soisconnected(so);
1533 	INP_WUNLOCK(inp);
1534 	return (error);
1535 }
1536 
1537 static void
1538 udp_detach(struct socket *so)
1539 {
1540 	struct inpcb *inp;
1541 	struct udpcb *up;
1542 
1543 	inp = sotoinpcb(so);
1544 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1545 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1546 	    ("udp_detach: not disconnected"));
1547 	INP_INFO_WLOCK(&V_udbinfo);
1548 	INP_WLOCK(inp);
1549 	up = intoudpcb(inp);
1550 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1551 	inp->inp_ppcb = NULL;
1552 	in_pcbdetach(inp);
1553 	in_pcbfree(inp);
1554 	INP_INFO_WUNLOCK(&V_udbinfo);
1555 	udp_discardcb(up);
1556 }
1557 
1558 static int
1559 udp_disconnect(struct socket *so)
1560 {
1561 	struct inpcb *inp;
1562 
1563 	inp = sotoinpcb(so);
1564 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1565 	INP_WLOCK(inp);
1566 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1567 		INP_WUNLOCK(inp);
1568 		return (ENOTCONN);
1569 	}
1570 	INP_HASH_WLOCK(&V_udbinfo);
1571 	in_pcbdisconnect(inp);
1572 	inp->inp_laddr.s_addr = INADDR_ANY;
1573 	INP_HASH_WUNLOCK(&V_udbinfo);
1574 	SOCK_LOCK(so);
1575 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1576 	SOCK_UNLOCK(so);
1577 	INP_WUNLOCK(inp);
1578 	return (0);
1579 }
1580 
1581 static int
1582 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1583     struct mbuf *control, struct thread *td)
1584 {
1585 	struct inpcb *inp;
1586 
1587 	inp = sotoinpcb(so);
1588 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1589 	return (udp_output(inp, m, addr, control, td));
1590 }
1591 #endif /* INET */
1592 
1593 int
1594 udp_shutdown(struct socket *so)
1595 {
1596 	struct inpcb *inp;
1597 
1598 	inp = sotoinpcb(so);
1599 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1600 	INP_WLOCK(inp);
1601 	socantsendmore(so);
1602 	INP_WUNLOCK(inp);
1603 	return (0);
1604 }
1605 
1606 #ifdef INET
1607 struct pr_usrreqs udp_usrreqs = {
1608 	.pru_abort =		udp_abort,
1609 	.pru_attach =		udp_attach,
1610 	.pru_bind =		udp_bind,
1611 	.pru_connect =		udp_connect,
1612 	.pru_control =		in_control,
1613 	.pru_detach =		udp_detach,
1614 	.pru_disconnect =	udp_disconnect,
1615 	.pru_peeraddr =		in_getpeeraddr,
1616 	.pru_send =		udp_send,
1617 	.pru_soreceive =	soreceive_dgram,
1618 	.pru_sosend =		sosend_dgram,
1619 	.pru_shutdown =		udp_shutdown,
1620 	.pru_sockaddr =		in_getsockaddr,
1621 	.pru_sosetlabel =	in_pcbsosetlabel,
1622 	.pru_close =		udp_close,
1623 };
1624 #endif /* INET */
1625