xref: /freebsd/sys/netinet/udp_usrreq.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_ipfw.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 
41 #include <sys/param.h>
42 #include <sys/domain.h>
43 #include <sys/eventhandler.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/protosw.h>
52 #include <sys/signalvar.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/systm.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #endif
73 #include <netinet/ip_icmp.h>
74 #include <netinet/icmp_var.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/ip_options.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #endif
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
82 
83 #ifdef IPSEC
84 #include <netipsec/ipsec.h>
85 #include <netipsec/esp.h>
86 #endif
87 
88 #include <machine/in_cksum.h>
89 
90 #include <security/mac/mac_framework.h>
91 
92 /*
93  * UDP protocol implementation.
94  * Per RFC 768, August, 1980.
95  */
96 
97 VNET_DEFINE(int, udp_blackhole);
98 
99 /*
100  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
101  * removes the only data integrity mechanism for packets and malformed
102  * packets that would otherwise be discarded due to bad checksums, and may
103  * cause problems (especially for NFS data blocks).
104  */
105 static int	udp_cksum = 1;
106 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
107     0, "compute udp checksum");
108 
109 int	udp_log_in_vain = 0;
110 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
111     &udp_log_in_vain, 0, "Log all incoming UDP packets");
112 
113 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
114     &VNET_NAME(udp_blackhole), 0,
115     "Do not send port unreachables for refused connects");
116 
117 u_long	udp_sendspace = 9216;		/* really max datagram size */
118 					/* 40 1K datagrams */
119 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
120     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
121 
122 u_long	udp_recvspace = 40 * (1024 +
123 #ifdef INET6
124 				      sizeof(struct sockaddr_in6)
125 #else
126 				      sizeof(struct sockaddr_in)
127 #endif
128 				      );
129 
130 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
131     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
132 
133 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
134 VNET_DEFINE(struct inpcbinfo, udbinfo);
135 static VNET_DEFINE(uma_zone_t, udpcb_zone);
136 VNET_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
137 
138 #define	V_udpcb_zone			VNET(udpcb_zone)
139 
140 #ifndef UDBHASHSIZE
141 #define	UDBHASHSIZE	128
142 #endif
143 
144 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
145     &VNET_NAME(udpstat), udpstat,
146     "UDP statistics (struct udpstat, netinet/udp_var.h)");
147 
148 static void	udp_detach(struct socket *so);
149 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
150 		    struct mbuf *, struct thread *);
151 #ifdef IPSEC
152 #ifdef IPSEC_NAT_T
153 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
154 #ifdef INET
155 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
156 #endif
157 #endif /* IPSEC_NAT_T */
158 #endif /* IPSEC */
159 
160 static void
161 udp_zone_change(void *tag)
162 {
163 
164 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
165 	uma_zone_set_max(V_udpcb_zone, maxsockets);
166 }
167 
168 static int
169 udp_inpcb_init(void *mem, int size, int flags)
170 {
171 	struct inpcb *inp;
172 
173 	inp = mem;
174 	INP_LOCK_INIT(inp, "inp", "udpinp");
175 	return (0);
176 }
177 
178 void
179 udp_init(void)
180 {
181 
182 	V_udp_blackhole = 0;
183 
184 	INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
185 	LIST_INIT(&V_udb);
186 #ifdef VIMAGE
187 	V_udbinfo.ipi_vnet = curvnet;
188 #endif
189 	V_udbinfo.ipi_listhead = &V_udb;
190 	V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
191 	    &V_udbinfo.ipi_hashmask);
192 	V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
193 	    &V_udbinfo.ipi_porthashmask);
194 	V_udbinfo.ipi_zone = uma_zcreate("udp_inpcb", sizeof(struct inpcb),
195 	    NULL, NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
197 
198 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
199 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
200 	uma_zone_set_max(V_udpcb_zone, maxsockets);
201 
202 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
203 	    EVENTHANDLER_PRI_ANY);
204 }
205 
206 /*
207  * Kernel module interface for updating udpstat.  The argument is an index
208  * into udpstat treated as an array of u_long.  While this encodes the
209  * general layout of udpstat into the caller, it doesn't encode its location,
210  * so that future changes to add, for example, per-CPU stats support won't
211  * cause binary compatibility problems for kernel modules.
212  */
213 void
214 kmod_udpstat_inc(int statnum)
215 {
216 
217 	(*((u_long *)&V_udpstat + statnum))++;
218 }
219 
220 int
221 udp_newudpcb(struct inpcb *inp)
222 {
223 	struct udpcb *up;
224 
225 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
226 	if (up == NULL)
227 		return (ENOBUFS);
228 	inp->inp_ppcb = up;
229 	return (0);
230 }
231 
232 void
233 udp_discardcb(struct udpcb *up)
234 {
235 
236 	uma_zfree(V_udpcb_zone, up);
237 }
238 
239 #ifdef VIMAGE
240 void
241 udp_destroy(void)
242 {
243 
244 	hashdestroy(V_udbinfo.ipi_hashbase, M_PCB,
245 	    V_udbinfo.ipi_hashmask);
246 	hashdestroy(V_udbinfo.ipi_porthashbase, M_PCB,
247 	    V_udbinfo.ipi_porthashmask);
248 	INP_INFO_LOCK_DESTROY(&V_udbinfo);
249 }
250 #endif
251 
252 /*
253  * Subroutine of udp_input(), which appends the provided mbuf chain to the
254  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
255  * contains the source address.  If the socket ends up being an IPv6 socket,
256  * udp_append() will convert to a sockaddr_in6 before passing the address
257  * into the socket code.
258  */
259 static void
260 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
261     struct sockaddr_in *udp_in)
262 {
263 	struct sockaddr *append_sa;
264 	struct socket *so;
265 	struct mbuf *opts = 0;
266 #ifdef INET6
267 	struct sockaddr_in6 udp_in6;
268 #endif
269 #ifdef IPSEC
270 #ifdef IPSEC_NAT_T
271 #ifdef INET
272 	struct udpcb *up;
273 #endif
274 #endif
275 #endif
276 
277 	INP_RLOCK_ASSERT(inp);
278 
279 #ifdef IPSEC
280 	/* Check AH/ESP integrity. */
281 	if (ipsec4_in_reject(n, inp)) {
282 		m_freem(n);
283 		V_ipsec4stat.in_polvio++;
284 		return;
285 	}
286 #ifdef IPSEC_NAT_T
287 #ifdef INET
288 	up = intoudpcb(inp);
289 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
290 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
291 		n = udp4_espdecap(inp, n, off);
292 		if (n == NULL)				/* Consumed. */
293 			return;
294 	}
295 #endif /* INET */
296 #endif /* IPSEC_NAT_T */
297 #endif /* IPSEC */
298 #ifdef MAC
299 	if (mac_inpcb_check_deliver(inp, n) != 0) {
300 		m_freem(n);
301 		return;
302 	}
303 #endif
304 	if (inp->inp_flags & INP_CONTROLOPTS ||
305 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
306 #ifdef INET6
307 		if (inp->inp_vflag & INP_IPV6)
308 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
309 		else
310 #endif
311 			ip_savecontrol(inp, &opts, ip, n);
312 	}
313 #ifdef INET6
314 	if (inp->inp_vflag & INP_IPV6) {
315 		bzero(&udp_in6, sizeof(udp_in6));
316 		udp_in6.sin6_len = sizeof(udp_in6);
317 		udp_in6.sin6_family = AF_INET6;
318 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
319 		append_sa = (struct sockaddr *)&udp_in6;
320 	} else
321 #endif
322 		append_sa = (struct sockaddr *)udp_in;
323 	m_adj(n, off);
324 
325 	so = inp->inp_socket;
326 	SOCKBUF_LOCK(&so->so_rcv);
327 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
328 		SOCKBUF_UNLOCK(&so->so_rcv);
329 		m_freem(n);
330 		if (opts)
331 			m_freem(opts);
332 		UDPSTAT_INC(udps_fullsock);
333 	} else
334 		sorwakeup_locked(so);
335 }
336 
337 void
338 udp_input(struct mbuf *m, int off)
339 {
340 	int iphlen = off;
341 	struct ip *ip;
342 	struct udphdr *uh;
343 	struct ifnet *ifp;
344 	struct inpcb *inp;
345 	struct udpcb *up;
346 	int len;
347 	struct ip save_ip;
348 	struct sockaddr_in udp_in;
349 #ifdef IPFIREWALL_FORWARD
350 	struct m_tag *fwd_tag;
351 #endif
352 
353 	ifp = m->m_pkthdr.rcvif;
354 	UDPSTAT_INC(udps_ipackets);
355 
356 	/*
357 	 * Strip IP options, if any; should skip this, make available to
358 	 * user, and use on returned packets, but we don't yet have a way to
359 	 * check the checksum with options still present.
360 	 */
361 	if (iphlen > sizeof (struct ip)) {
362 		ip_stripoptions(m, (struct mbuf *)0);
363 		iphlen = sizeof(struct ip);
364 	}
365 
366 	/*
367 	 * Get IP and UDP header together in first mbuf.
368 	 */
369 	ip = mtod(m, struct ip *);
370 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
371 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
372 			UDPSTAT_INC(udps_hdrops);
373 			return;
374 		}
375 		ip = mtod(m, struct ip *);
376 	}
377 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
378 
379 	/*
380 	 * Destination port of 0 is illegal, based on RFC768.
381 	 */
382 	if (uh->uh_dport == 0)
383 		goto badunlocked;
384 
385 	/*
386 	 * Construct sockaddr format source address.  Stuff source address
387 	 * and datagram in user buffer.
388 	 */
389 	bzero(&udp_in, sizeof(udp_in));
390 	udp_in.sin_len = sizeof(udp_in);
391 	udp_in.sin_family = AF_INET;
392 	udp_in.sin_port = uh->uh_sport;
393 	udp_in.sin_addr = ip->ip_src;
394 
395 	/*
396 	 * Make mbuf data length reflect UDP length.  If not enough data to
397 	 * reflect UDP length, drop.
398 	 */
399 	len = ntohs((u_short)uh->uh_ulen);
400 	if (ip->ip_len != len) {
401 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
402 			UDPSTAT_INC(udps_badlen);
403 			goto badunlocked;
404 		}
405 		m_adj(m, len - ip->ip_len);
406 		/* ip->ip_len = len; */
407 	}
408 
409 	/*
410 	 * Save a copy of the IP header in case we want restore it for
411 	 * sending an ICMP error message in response.
412 	 */
413 	if (!V_udp_blackhole)
414 		save_ip = *ip;
415 	else
416 		memset(&save_ip, 0, sizeof(save_ip));
417 
418 	/*
419 	 * Checksum extended UDP header and data.
420 	 */
421 	if (uh->uh_sum) {
422 		u_short uh_sum;
423 
424 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
425 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
426 				uh_sum = m->m_pkthdr.csum_data;
427 			else
428 				uh_sum = in_pseudo(ip->ip_src.s_addr,
429 				    ip->ip_dst.s_addr, htonl((u_short)len +
430 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
431 			uh_sum ^= 0xffff;
432 		} else {
433 			char b[9];
434 
435 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
436 			bzero(((struct ipovly *)ip)->ih_x1, 9);
437 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
438 			uh_sum = in_cksum(m, len + sizeof (struct ip));
439 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
440 		}
441 		if (uh_sum) {
442 			UDPSTAT_INC(udps_badsum);
443 			m_freem(m);
444 			return;
445 		}
446 	} else
447 		UDPSTAT_INC(udps_nosum);
448 
449 #ifdef IPFIREWALL_FORWARD
450 	/*
451 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
452 	 */
453 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
454 	if (fwd_tag != NULL) {
455 		struct sockaddr_in *next_hop;
456 
457 		/*
458 		 * Do the hack.
459 		 */
460 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
461 		ip->ip_dst = next_hop->sin_addr;
462 		uh->uh_dport = ntohs(next_hop->sin_port);
463 
464 		/*
465 		 * Remove the tag from the packet.  We don't need it anymore.
466 		 */
467 		m_tag_delete(m, fwd_tag);
468 	}
469 #endif
470 
471 	INP_INFO_RLOCK(&V_udbinfo);
472 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
473 	    in_broadcast(ip->ip_dst, ifp)) {
474 		struct inpcb *last;
475 		struct ip_moptions *imo;
476 
477 		last = NULL;
478 		LIST_FOREACH(inp, &V_udb, inp_list) {
479 			if (inp->inp_lport != uh->uh_dport)
480 				continue;
481 #ifdef INET6
482 			if ((inp->inp_vflag & INP_IPV4) == 0)
483 				continue;
484 #endif
485 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
486 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
487 				continue;
488 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
489 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
490 				continue;
491 			if (inp->inp_fport != 0 &&
492 			    inp->inp_fport != uh->uh_sport)
493 				continue;
494 
495 			INP_RLOCK(inp);
496 
497 			/*
498 			 * Handle socket delivery policy for any-source
499 			 * and source-specific multicast. [RFC3678]
500 			 */
501 			imo = inp->inp_moptions;
502 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
503 			    imo != NULL) {
504 				struct sockaddr_in	 group;
505 				int			 blocked;
506 
507 				bzero(&group, sizeof(struct sockaddr_in));
508 				group.sin_len = sizeof(struct sockaddr_in);
509 				group.sin_family = AF_INET;
510 				group.sin_addr = ip->ip_dst;
511 
512 				blocked = imo_multi_filter(imo, ifp,
513 					(struct sockaddr *)&group,
514 					(struct sockaddr *)&udp_in);
515 				if (blocked != MCAST_PASS) {
516 					if (blocked == MCAST_NOTGMEMBER)
517 						IPSTAT_INC(ips_notmember);
518 					if (blocked == MCAST_NOTSMEMBER ||
519 					    blocked == MCAST_MUTED)
520 						UDPSTAT_INC(udps_filtermcast);
521 					INP_RUNLOCK(inp);
522 					continue;
523 				}
524 			}
525 			if (last != NULL) {
526 				struct mbuf *n;
527 
528 				n = m_copy(m, 0, M_COPYALL);
529 				up = intoudpcb(last);
530 				if (up->u_tun_func == NULL) {
531 					if (n != NULL)
532 						udp_append(last,
533 						    ip, n,
534 						    iphlen +
535 						    sizeof(struct udphdr),
536 						    &udp_in);
537 				} else {
538 					/*
539 					 * Engage the tunneling protocol we
540 					 * will have to leave the info_lock
541 					 * up, since we are hunting through
542 					 * multiple UDP's.
543 					 */
544 
545 					(*up->u_tun_func)(n, iphlen, last);
546 				}
547 				INP_RUNLOCK(last);
548 			}
549 			last = inp;
550 			/*
551 			 * Don't look for additional matches if this one does
552 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
553 			 * socket options set.  This heuristic avoids
554 			 * searching through all pcbs in the common case of a
555 			 * non-shared port.  It assumes that an application
556 			 * will never clear these options after setting them.
557 			 */
558 			if ((last->inp_socket->so_options &
559 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
560 				break;
561 		}
562 
563 		if (last == NULL) {
564 			/*
565 			 * No matching pcb found; discard datagram.  (No need
566 			 * to send an ICMP Port Unreachable for a broadcast
567 			 * or multicast datgram.)
568 			 */
569 			UDPSTAT_INC(udps_noportbcast);
570 			goto badheadlocked;
571 		}
572 		up = intoudpcb(last);
573 		if (up->u_tun_func == NULL) {
574 			udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
575 			    &udp_in);
576 		} else {
577 			/*
578 			 * Engage the tunneling protocol.
579 			 */
580 			(*up->u_tun_func)(m, iphlen, last);
581 		}
582 		INP_RUNLOCK(last);
583 		INP_INFO_RUNLOCK(&V_udbinfo);
584 		return;
585 	}
586 
587 	/*
588 	 * Locate pcb for datagram.
589 	 */
590 	inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
591 	    ip->ip_dst, uh->uh_dport, 1, ifp);
592 	if (inp == NULL) {
593 		if (udp_log_in_vain) {
594 			char buf[4*sizeof "123"];
595 
596 			strcpy(buf, inet_ntoa(ip->ip_dst));
597 			log(LOG_INFO,
598 			    "Connection attempt to UDP %s:%d from %s:%d\n",
599 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
600 			    ntohs(uh->uh_sport));
601 		}
602 		UDPSTAT_INC(udps_noport);
603 		if (m->m_flags & (M_BCAST | M_MCAST)) {
604 			UDPSTAT_INC(udps_noportbcast);
605 			goto badheadlocked;
606 		}
607 		if (V_udp_blackhole)
608 			goto badheadlocked;
609 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
610 			goto badheadlocked;
611 		*ip = save_ip;
612 		ip->ip_len += iphlen;
613 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
614 		INP_INFO_RUNLOCK(&V_udbinfo);
615 		return;
616 	}
617 
618 	/*
619 	 * Check the minimum TTL for socket.
620 	 */
621 	INP_RLOCK(inp);
622 	INP_INFO_RUNLOCK(&V_udbinfo);
623 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
624 		INP_RUNLOCK(inp);
625 		goto badunlocked;
626 	}
627 	up = intoudpcb(inp);
628 	if (up->u_tun_func == NULL) {
629 		udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
630 	} else {
631 		/*
632 		 * Engage the tunneling protocol.
633 		 */
634 
635 		(*up->u_tun_func)(m, iphlen, inp);
636 	}
637 	INP_RUNLOCK(inp);
638 	return;
639 
640 badheadlocked:
641 	if (inp)
642 		INP_RUNLOCK(inp);
643 	INP_INFO_RUNLOCK(&V_udbinfo);
644 badunlocked:
645 	m_freem(m);
646 }
647 
648 /*
649  * Notify a udp user of an asynchronous error; just wake up so that they can
650  * collect error status.
651  */
652 struct inpcb *
653 udp_notify(struct inpcb *inp, int errno)
654 {
655 
656 	/*
657 	 * While udp_ctlinput() always calls udp_notify() with a read lock
658 	 * when invoking it directly, in_pcbnotifyall() currently uses write
659 	 * locks due to sharing code with TCP.  For now, accept either a read
660 	 * or a write lock, but a read lock is sufficient.
661 	 */
662 	INP_LOCK_ASSERT(inp);
663 
664 	inp->inp_socket->so_error = errno;
665 	sorwakeup(inp->inp_socket);
666 	sowwakeup(inp->inp_socket);
667 	return (inp);
668 }
669 
670 void
671 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
672 {
673 	struct ip *ip = vip;
674 	struct udphdr *uh;
675 	struct in_addr faddr;
676 	struct inpcb *inp;
677 
678 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
679 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
680 		return;
681 
682 	/*
683 	 * Redirects don't need to be handled up here.
684 	 */
685 	if (PRC_IS_REDIRECT(cmd))
686 		return;
687 
688 	/*
689 	 * Hostdead is ugly because it goes linearly through all PCBs.
690 	 *
691 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
692 	 * DoS attack on machines with many connections.
693 	 */
694 	if (cmd == PRC_HOSTDEAD)
695 		ip = NULL;
696 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
697 		return;
698 	if (ip != NULL) {
699 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
700 		INP_INFO_RLOCK(&V_udbinfo);
701 		inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
702 		    ip->ip_src, uh->uh_sport, 0, NULL);
703 		if (inp != NULL) {
704 			INP_RLOCK(inp);
705 			if (inp->inp_socket != NULL) {
706 				udp_notify(inp, inetctlerrmap[cmd]);
707 			}
708 			INP_RUNLOCK(inp);
709 		}
710 		INP_INFO_RUNLOCK(&V_udbinfo);
711 	} else
712 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
713 		    udp_notify);
714 }
715 
716 static int
717 udp_pcblist(SYSCTL_HANDLER_ARGS)
718 {
719 	int error, i, n;
720 	struct inpcb *inp, **inp_list;
721 	inp_gen_t gencnt;
722 	struct xinpgen xig;
723 
724 	/*
725 	 * The process of preparing the PCB list is too time-consuming and
726 	 * resource-intensive to repeat twice on every request.
727 	 */
728 	if (req->oldptr == 0) {
729 		n = V_udbinfo.ipi_count;
730 		req->oldidx = 2 * (sizeof xig)
731 			+ (n + n/8) * sizeof(struct xinpcb);
732 		return (0);
733 	}
734 
735 	if (req->newptr != 0)
736 		return (EPERM);
737 
738 	/*
739 	 * OK, now we're committed to doing something.
740 	 */
741 	INP_INFO_RLOCK(&V_udbinfo);
742 	gencnt = V_udbinfo.ipi_gencnt;
743 	n = V_udbinfo.ipi_count;
744 	INP_INFO_RUNLOCK(&V_udbinfo);
745 
746 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
747 		+ n * sizeof(struct xinpcb));
748 	if (error != 0)
749 		return (error);
750 
751 	xig.xig_len = sizeof xig;
752 	xig.xig_count = n;
753 	xig.xig_gen = gencnt;
754 	xig.xig_sogen = so_gencnt;
755 	error = SYSCTL_OUT(req, &xig, sizeof xig);
756 	if (error)
757 		return (error);
758 
759 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
760 	if (inp_list == 0)
761 		return (ENOMEM);
762 
763 	INP_INFO_RLOCK(&V_udbinfo);
764 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
765 	     inp = LIST_NEXT(inp, inp_list)) {
766 		INP_RLOCK(inp);
767 		if (inp->inp_gencnt <= gencnt &&
768 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
769 			inp_list[i++] = inp;
770 		INP_RUNLOCK(inp);
771 	}
772 	INP_INFO_RUNLOCK(&V_udbinfo);
773 	n = i;
774 
775 	error = 0;
776 	for (i = 0; i < n; i++) {
777 		inp = inp_list[i];
778 		INP_RLOCK(inp);
779 		if (inp->inp_gencnt <= gencnt) {
780 			struct xinpcb xi;
781 			bzero(&xi, sizeof(xi));
782 			xi.xi_len = sizeof xi;
783 			/* XXX should avoid extra copy */
784 			bcopy(inp, &xi.xi_inp, sizeof *inp);
785 			if (inp->inp_socket)
786 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
787 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
788 			INP_RUNLOCK(inp);
789 			error = SYSCTL_OUT(req, &xi, sizeof xi);
790 		} else
791 			INP_RUNLOCK(inp);
792 	}
793 	if (!error) {
794 		/*
795 		 * Give the user an updated idea of our state.  If the
796 		 * generation differs from what we told her before, she knows
797 		 * that something happened while we were processing this
798 		 * request, and it might be necessary to retry.
799 		 */
800 		INP_INFO_RLOCK(&V_udbinfo);
801 		xig.xig_gen = V_udbinfo.ipi_gencnt;
802 		xig.xig_sogen = so_gencnt;
803 		xig.xig_count = V_udbinfo.ipi_count;
804 		INP_INFO_RUNLOCK(&V_udbinfo);
805 		error = SYSCTL_OUT(req, &xig, sizeof xig);
806 	}
807 	free(inp_list, M_TEMP);
808 	return (error);
809 }
810 
811 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
812     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
813 
814 static int
815 udp_getcred(SYSCTL_HANDLER_ARGS)
816 {
817 	struct xucred xuc;
818 	struct sockaddr_in addrs[2];
819 	struct inpcb *inp;
820 	int error;
821 
822 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
823 	if (error)
824 		return (error);
825 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
826 	if (error)
827 		return (error);
828 	INP_INFO_RLOCK(&V_udbinfo);
829 	inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
830 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
831 	if (inp != NULL) {
832 		INP_RLOCK(inp);
833 		INP_INFO_RUNLOCK(&V_udbinfo);
834 		if (inp->inp_socket == NULL)
835 			error = ENOENT;
836 		if (error == 0)
837 			error = cr_canseeinpcb(req->td->td_ucred, inp);
838 		if (error == 0)
839 			cru2x(inp->inp_cred, &xuc);
840 		INP_RUNLOCK(inp);
841 	} else {
842 		INP_INFO_RUNLOCK(&V_udbinfo);
843 		error = ENOENT;
844 	}
845 	if (error == 0)
846 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
847 	return (error);
848 }
849 
850 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
851     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
852     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
853 
854 int
855 udp_ctloutput(struct socket *so, struct sockopt *sopt)
856 {
857 	int error = 0, optval;
858 	struct inpcb *inp;
859 #ifdef IPSEC_NAT_T
860 	struct udpcb *up;
861 #endif
862 
863 	inp = sotoinpcb(so);
864 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
865 	INP_WLOCK(inp);
866 	if (sopt->sopt_level != IPPROTO_UDP) {
867 #ifdef INET6
868 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
869 			INP_WUNLOCK(inp);
870 			error = ip6_ctloutput(so, sopt);
871 		} else {
872 #endif
873 			INP_WUNLOCK(inp);
874 			error = ip_ctloutput(so, sopt);
875 #ifdef INET6
876 		}
877 #endif
878 		return (error);
879 	}
880 
881 	switch (sopt->sopt_dir) {
882 	case SOPT_SET:
883 		switch (sopt->sopt_name) {
884 		case UDP_ENCAP:
885 			INP_WUNLOCK(inp);
886 			error = sooptcopyin(sopt, &optval, sizeof optval,
887 					    sizeof optval);
888 			if (error)
889 				break;
890 			inp = sotoinpcb(so);
891 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
892 			INP_WLOCK(inp);
893 #ifdef IPSEC_NAT_T
894 			up = intoudpcb(inp);
895 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
896 #endif
897 			switch (optval) {
898 			case 0:
899 				/* Clear all UDP encap. */
900 #ifdef IPSEC_NAT_T
901 				up->u_flags &= ~UF_ESPINUDP_ALL;
902 #endif
903 				break;
904 #ifdef IPSEC_NAT_T
905 			case UDP_ENCAP_ESPINUDP:
906 			case UDP_ENCAP_ESPINUDP_NON_IKE:
907 				up->u_flags &= ~UF_ESPINUDP_ALL;
908 				if (optval == UDP_ENCAP_ESPINUDP)
909 					up->u_flags |= UF_ESPINUDP;
910 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
911 					up->u_flags |= UF_ESPINUDP_NON_IKE;
912 				break;
913 #endif
914 			default:
915 				error = EINVAL;
916 				break;
917 			}
918 			INP_WUNLOCK(inp);
919 			break;
920 		default:
921 			INP_WUNLOCK(inp);
922 			error = ENOPROTOOPT;
923 			break;
924 		}
925 		break;
926 	case SOPT_GET:
927 		switch (sopt->sopt_name) {
928 #ifdef IPSEC_NAT_T
929 		case UDP_ENCAP:
930 			up = intoudpcb(inp);
931 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
932 			optval = up->u_flags & UF_ESPINUDP_ALL;
933 			INP_WUNLOCK(inp);
934 			error = sooptcopyout(sopt, &optval, sizeof optval);
935 			break;
936 #endif
937 		default:
938 			INP_WUNLOCK(inp);
939 			error = ENOPROTOOPT;
940 			break;
941 		}
942 		break;
943 	}
944 	return (error);
945 }
946 
947 static int
948 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
949     struct mbuf *control, struct thread *td)
950 {
951 	struct udpiphdr *ui;
952 	int len = m->m_pkthdr.len;
953 	struct in_addr faddr, laddr;
954 	struct cmsghdr *cm;
955 	struct sockaddr_in *sin, src;
956 	int error = 0;
957 	int ipflags;
958 	u_short fport, lport;
959 	int unlock_udbinfo;
960 
961 	/*
962 	 * udp_output() may need to temporarily bind or connect the current
963 	 * inpcb.  As such, we don't know up front whether we will need the
964 	 * pcbinfo lock or not.  Do any work to decide what is needed up
965 	 * front before acquiring any locks.
966 	 */
967 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
968 		if (control)
969 			m_freem(control);
970 		m_freem(m);
971 		return (EMSGSIZE);
972 	}
973 
974 	src.sin_family = 0;
975 	if (control != NULL) {
976 		/*
977 		 * XXX: Currently, we assume all the optional information is
978 		 * stored in a single mbuf.
979 		 */
980 		if (control->m_next) {
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 			default:
1013 				error = ENOPROTOOPT;
1014 				break;
1015 			}
1016 			if (error)
1017 				break;
1018 		}
1019 		m_freem(control);
1020 	}
1021 	if (error) {
1022 		m_freem(m);
1023 		return (error);
1024 	}
1025 
1026 	/*
1027 	 * Depending on whether or not the application has bound or connected
1028 	 * the socket, we may have to do varying levels of work.  The optimal
1029 	 * case is for a connected UDP socket, as a global lock isn't
1030 	 * required at all.
1031 	 *
1032 	 * In order to decide which we need, we require stability of the
1033 	 * inpcb binding, which we ensure by acquiring a read lock on the
1034 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1035 	 * the trylock and retry game; note that we may end up with more
1036 	 * conservative locks than required the second time around, so later
1037 	 * assertions have to accept that.  Further analysis of the number of
1038 	 * misses under contention is required.
1039 	 */
1040 	sin = (struct sockaddr_in *)addr;
1041 	INP_RLOCK(inp);
1042 	if (sin != NULL &&
1043 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1044 		INP_RUNLOCK(inp);
1045 		INP_INFO_WLOCK(&V_udbinfo);
1046 		INP_WLOCK(inp);
1047 		unlock_udbinfo = 2;
1048 	} else if ((sin != NULL && (
1049 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1050 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1051 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1052 	    (inp->inp_lport == 0))) ||
1053 	    (src.sin_family == AF_INET)) {
1054 		if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
1055 			INP_RUNLOCK(inp);
1056 			INP_INFO_RLOCK(&V_udbinfo);
1057 			INP_RLOCK(inp);
1058 		}
1059 		unlock_udbinfo = 1;
1060 	} else
1061 		unlock_udbinfo = 0;
1062 
1063 	/*
1064 	 * If the IP_SENDSRCADDR control message was specified, override the
1065 	 * source address for this datagram.  Its use is invalidated if the
1066 	 * address thus specified is incomplete or clobbers other inpcbs.
1067 	 */
1068 	laddr = inp->inp_laddr;
1069 	lport = inp->inp_lport;
1070 	if (src.sin_family == AF_INET) {
1071 		INP_INFO_LOCK_ASSERT(&V_udbinfo);
1072 		if ((lport == 0) ||
1073 		    (laddr.s_addr == INADDR_ANY &&
1074 		     src.sin_addr.s_addr == INADDR_ANY)) {
1075 			error = EINVAL;
1076 			goto release;
1077 		}
1078 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1079 		    &laddr.s_addr, &lport, td->td_ucred);
1080 		if (error)
1081 			goto release;
1082 	}
1083 
1084 	/*
1085 	 * If a UDP socket has been connected, then a local address/port will
1086 	 * have been selected and bound.
1087 	 *
1088 	 * If a UDP socket has not been connected to, then an explicit
1089 	 * destination address must be used, in which case a local
1090 	 * address/port may not have been selected and bound.
1091 	 */
1092 	if (sin != NULL) {
1093 		INP_LOCK_ASSERT(inp);
1094 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1095 			error = EISCONN;
1096 			goto release;
1097 		}
1098 
1099 		/*
1100 		 * Jail may rewrite the destination address, so let it do
1101 		 * that before we use it.
1102 		 */
1103 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1104 		if (error)
1105 			goto release;
1106 
1107 		/*
1108 		 * If a local address or port hasn't yet been selected, or if
1109 		 * the destination address needs to be rewritten due to using
1110 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1111 		 * to do the heavy lifting.  Once a port is selected, we
1112 		 * commit the binding back to the socket; we also commit the
1113 		 * binding of the address if in jail.
1114 		 *
1115 		 * If we already have a valid binding and we're not
1116 		 * requesting a destination address rewrite, use a fast path.
1117 		 */
1118 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1119 		    inp->inp_lport == 0 ||
1120 		    sin->sin_addr.s_addr == INADDR_ANY ||
1121 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1122 			INP_INFO_LOCK_ASSERT(&V_udbinfo);
1123 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1124 			    &lport, &faddr.s_addr, &fport, NULL,
1125 			    td->td_ucred);
1126 			if (error)
1127 				goto release;
1128 
1129 			/*
1130 			 * XXXRW: Why not commit the port if the address is
1131 			 * !INADDR_ANY?
1132 			 */
1133 			/* Commit the local port if newly assigned. */
1134 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1135 			    inp->inp_lport == 0) {
1136 				INP_INFO_WLOCK_ASSERT(&V_udbinfo);
1137 				INP_WLOCK_ASSERT(inp);
1138 				/*
1139 				 * Remember addr if jailed, to prevent
1140 				 * rebinding.
1141 				 */
1142 				if (prison_flag(td->td_ucred, PR_IP4))
1143 					inp->inp_laddr = laddr;
1144 				inp->inp_lport = lport;
1145 				if (in_pcbinshash(inp) != 0) {
1146 					inp->inp_lport = 0;
1147 					error = EAGAIN;
1148 					goto release;
1149 				}
1150 				inp->inp_flags |= INP_ANONPORT;
1151 			}
1152 		} else {
1153 			faddr = sin->sin_addr;
1154 			fport = sin->sin_port;
1155 		}
1156 	} else {
1157 		INP_LOCK_ASSERT(inp);
1158 		faddr = inp->inp_faddr;
1159 		fport = inp->inp_fport;
1160 		if (faddr.s_addr == INADDR_ANY) {
1161 			error = ENOTCONN;
1162 			goto release;
1163 		}
1164 	}
1165 
1166 	/*
1167 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1168 	 * link-layer headers.  Immediate slide the data pointer back forward
1169 	 * since we won't use that space at this layer.
1170 	 */
1171 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1172 	if (m == NULL) {
1173 		error = ENOBUFS;
1174 		goto release;
1175 	}
1176 	m->m_data += max_linkhdr;
1177 	m->m_len -= max_linkhdr;
1178 	m->m_pkthdr.len -= max_linkhdr;
1179 
1180 	/*
1181 	 * Fill in mbuf with extended UDP header and addresses and length put
1182 	 * into network format.
1183 	 */
1184 	ui = mtod(m, struct udpiphdr *);
1185 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1186 	ui->ui_pr = IPPROTO_UDP;
1187 	ui->ui_src = laddr;
1188 	ui->ui_dst = faddr;
1189 	ui->ui_sport = lport;
1190 	ui->ui_dport = fport;
1191 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1192 
1193 	/*
1194 	 * Set the Don't Fragment bit in the IP header.
1195 	 */
1196 	if (inp->inp_flags & INP_DONTFRAG) {
1197 		struct ip *ip;
1198 
1199 		ip = (struct ip *)&ui->ui_i;
1200 		ip->ip_off |= IP_DF;
1201 	}
1202 
1203 	ipflags = 0;
1204 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1205 		ipflags |= IP_ROUTETOIF;
1206 	if (inp->inp_socket->so_options & SO_BROADCAST)
1207 		ipflags |= IP_ALLOWBROADCAST;
1208 	if (inp->inp_flags & INP_ONESBCAST)
1209 		ipflags |= IP_SENDONES;
1210 
1211 #ifdef MAC
1212 	mac_inpcb_create_mbuf(inp, m);
1213 #endif
1214 
1215 	/*
1216 	 * Set up checksum and output datagram.
1217 	 */
1218 	if (udp_cksum) {
1219 		if (inp->inp_flags & INP_ONESBCAST)
1220 			faddr.s_addr = INADDR_BROADCAST;
1221 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1222 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1223 		m->m_pkthdr.csum_flags = CSUM_UDP;
1224 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1225 	} else
1226 		ui->ui_sum = 0;
1227 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1228 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1229 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
1230 	UDPSTAT_INC(udps_opackets);
1231 
1232 	if (unlock_udbinfo == 2)
1233 		INP_INFO_WUNLOCK(&V_udbinfo);
1234 	else if (unlock_udbinfo == 1)
1235 		INP_INFO_RUNLOCK(&V_udbinfo);
1236 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1237 	    inp->inp_moptions, inp);
1238 	if (unlock_udbinfo == 2)
1239 		INP_WUNLOCK(inp);
1240 	else
1241 		INP_RUNLOCK(inp);
1242 	return (error);
1243 
1244 release:
1245 	if (unlock_udbinfo == 2) {
1246 		INP_WUNLOCK(inp);
1247 		INP_INFO_WUNLOCK(&V_udbinfo);
1248 	} else if (unlock_udbinfo == 1) {
1249 		INP_RUNLOCK(inp);
1250 		INP_INFO_RUNLOCK(&V_udbinfo);
1251 	} else
1252 		INP_RUNLOCK(inp);
1253 	m_freem(m);
1254 	return (error);
1255 }
1256 
1257 
1258 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1259 #ifdef INET
1260 /*
1261  * Potentially decap ESP in UDP frame.  Check for an ESP header
1262  * and optional marker; if present, strip the UDP header and
1263  * push the result through IPSec.
1264  *
1265  * Returns mbuf to be processed (potentially re-allocated) or
1266  * NULL if consumed and/or processed.
1267  */
1268 static struct mbuf *
1269 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1270 {
1271 	size_t minlen, payload, skip, iphlen;
1272 	caddr_t data;
1273 	struct udpcb *up;
1274 	struct m_tag *tag;
1275 	struct udphdr *udphdr;
1276 	struct ip *ip;
1277 
1278 	INP_RLOCK_ASSERT(inp);
1279 
1280 	/*
1281 	 * Pull up data so the longest case is contiguous:
1282 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1283 	 */
1284 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1285 	if (minlen > m->m_pkthdr.len)
1286 		minlen = m->m_pkthdr.len;
1287 	if ((m = m_pullup(m, minlen)) == NULL) {
1288 		V_ipsec4stat.in_inval++;
1289 		return (NULL);		/* Bypass caller processing. */
1290 	}
1291 	data = mtod(m, caddr_t);	/* Points to ip header. */
1292 	payload = m->m_len - off;	/* Size of payload. */
1293 
1294 	if (payload == 1 && data[off] == '\xff')
1295 		return (m);		/* NB: keepalive packet, no decap. */
1296 
1297 	up = intoudpcb(inp);
1298 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1299 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1300 	    ("u_flags 0x%x", up->u_flags));
1301 
1302 	/*
1303 	 * Check that the payload is large enough to hold an
1304 	 * ESP header and compute the amount of data to remove.
1305 	 *
1306 	 * NB: the caller has already done a pullup for us.
1307 	 * XXX can we assume alignment and eliminate bcopys?
1308 	 */
1309 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1310 		/*
1311 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1312 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1313 		 * possible AH mode non-IKE marker+non-ESP marker
1314 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1315 		 */
1316 		uint64_t marker;
1317 
1318 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1319 			return (m);	/* NB: no decap. */
1320 		bcopy(data + off, &marker, sizeof(uint64_t));
1321 		if (marker != 0)	/* Non-IKE marker. */
1322 			return (m);	/* NB: no decap. */
1323 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1324 	} else {
1325 		uint32_t spi;
1326 
1327 		if (payload <= sizeof(struct esp)) {
1328 			V_ipsec4stat.in_inval++;
1329 			m_freem(m);
1330 			return (NULL);	/* Discard. */
1331 		}
1332 		bcopy(data + off, &spi, sizeof(uint32_t));
1333 		if (spi == 0)		/* Non-ESP marker. */
1334 			return (m);	/* NB: no decap. */
1335 		skip = sizeof(struct udphdr);
1336 	}
1337 
1338 	/*
1339 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1340 	 * the UDP ports. This is required if we want to select
1341 	 * the right SPD for multiple hosts behind same NAT.
1342 	 *
1343 	 * NB: ports are maintained in network byte order everywhere
1344 	 *     in the NAT-T code.
1345 	 */
1346 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1347 		2 * sizeof(uint16_t), M_NOWAIT);
1348 	if (tag == NULL) {
1349 		V_ipsec4stat.in_nomem++;
1350 		m_freem(m);
1351 		return (NULL);		/* Discard. */
1352 	}
1353 	iphlen = off - sizeof(struct udphdr);
1354 	udphdr = (struct udphdr *)(data + iphlen);
1355 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1356 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1357 	m_tag_prepend(m, tag);
1358 
1359 	/*
1360 	 * Remove the UDP header (and possibly the non ESP marker)
1361 	 * IP header length is iphlen
1362 	 * Before:
1363 	 *   <--- off --->
1364 	 *   +----+------+-----+
1365 	 *   | IP |  UDP | ESP |
1366 	 *   +----+------+-----+
1367 	 *        <-skip->
1368 	 * After:
1369 	 *          +----+-----+
1370 	 *          | IP | ESP |
1371 	 *          +----+-----+
1372 	 *   <-skip->
1373 	 */
1374 	ovbcopy(data, data + skip, iphlen);
1375 	m_adj(m, skip);
1376 
1377 	ip = mtod(m, struct ip *);
1378 	ip->ip_len -= skip;
1379 	ip->ip_p = IPPROTO_ESP;
1380 
1381 	/*
1382 	 * We cannot yet update the cksums so clear any
1383 	 * h/w cksum flags as they are no longer valid.
1384 	 */
1385 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1386 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1387 
1388 	(void) ipsec4_common_input(m, iphlen, ip->ip_p);
1389 	return (NULL);			/* NB: consumed, bypass processing. */
1390 }
1391 #endif /* INET */
1392 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1393 
1394 static void
1395 udp_abort(struct socket *so)
1396 {
1397 	struct inpcb *inp;
1398 
1399 	inp = sotoinpcb(so);
1400 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1401 	INP_INFO_WLOCK(&V_udbinfo);
1402 	INP_WLOCK(inp);
1403 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1404 		in_pcbdisconnect(inp);
1405 		inp->inp_laddr.s_addr = INADDR_ANY;
1406 		soisdisconnected(so);
1407 	}
1408 	INP_WUNLOCK(inp);
1409 	INP_INFO_WUNLOCK(&V_udbinfo);
1410 }
1411 
1412 static int
1413 udp_attach(struct socket *so, int proto, struct thread *td)
1414 {
1415 	struct inpcb *inp;
1416 	int error;
1417 
1418 	inp = sotoinpcb(so);
1419 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1420 	error = soreserve(so, udp_sendspace, udp_recvspace);
1421 	if (error)
1422 		return (error);
1423 	INP_INFO_WLOCK(&V_udbinfo);
1424 	error = in_pcballoc(so, &V_udbinfo);
1425 	if (error) {
1426 		INP_INFO_WUNLOCK(&V_udbinfo);
1427 		return (error);
1428 	}
1429 
1430 	inp = (struct inpcb *)so->so_pcb;
1431 	inp->inp_vflag |= INP_IPV4;
1432 	inp->inp_ip_ttl = V_ip_defttl;
1433 
1434 	error = udp_newudpcb(inp);
1435 	if (error) {
1436 		in_pcbdetach(inp);
1437 		in_pcbfree(inp);
1438 		INP_INFO_WUNLOCK(&V_udbinfo);
1439 		return (error);
1440 	}
1441 
1442 	INP_WUNLOCK(inp);
1443 	INP_INFO_WUNLOCK(&V_udbinfo);
1444 	return (0);
1445 }
1446 
1447 int
1448 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1449 {
1450 	struct inpcb *inp;
1451 	struct udpcb *up;
1452 
1453 	KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram"));
1454 	KASSERT(so->so_pcb != NULL, ("udp_set_kernel_tunneling: NULL inp"));
1455 	if (so->so_type != SOCK_DGRAM) {
1456 		/* Not UDP socket... sorry! */
1457 		return (ENOTSUP);
1458 	}
1459 	inp = (struct inpcb *)so->so_pcb;
1460 	if (inp == NULL) {
1461 		/* NULL INP? */
1462 		return (EINVAL);
1463 	}
1464 	INP_WLOCK(inp);
1465 	up = intoudpcb(inp);
1466 	if (up->u_tun_func != NULL) {
1467 		INP_WUNLOCK(inp);
1468 		return (EBUSY);
1469 	}
1470 	up->u_tun_func = f;
1471 	INP_WUNLOCK(inp);
1472 	return (0);
1473 }
1474 
1475 static int
1476 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1477 {
1478 	struct inpcb *inp;
1479 	int error;
1480 
1481 	inp = sotoinpcb(so);
1482 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1483 	INP_INFO_WLOCK(&V_udbinfo);
1484 	INP_WLOCK(inp);
1485 	error = in_pcbbind(inp, nam, td->td_ucred);
1486 	INP_WUNLOCK(inp);
1487 	INP_INFO_WUNLOCK(&V_udbinfo);
1488 	return (error);
1489 }
1490 
1491 static void
1492 udp_close(struct socket *so)
1493 {
1494 	struct inpcb *inp;
1495 
1496 	inp = sotoinpcb(so);
1497 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1498 	INP_INFO_WLOCK(&V_udbinfo);
1499 	INP_WLOCK(inp);
1500 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1501 		in_pcbdisconnect(inp);
1502 		inp->inp_laddr.s_addr = INADDR_ANY;
1503 		soisdisconnected(so);
1504 	}
1505 	INP_WUNLOCK(inp);
1506 	INP_INFO_WUNLOCK(&V_udbinfo);
1507 }
1508 
1509 static int
1510 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1511 {
1512 	struct inpcb *inp;
1513 	int error;
1514 	struct sockaddr_in *sin;
1515 
1516 	inp = sotoinpcb(so);
1517 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1518 	INP_INFO_WLOCK(&V_udbinfo);
1519 	INP_WLOCK(inp);
1520 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1521 		INP_WUNLOCK(inp);
1522 		INP_INFO_WUNLOCK(&V_udbinfo);
1523 		return (EISCONN);
1524 	}
1525 	sin = (struct sockaddr_in *)nam;
1526 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1527 	if (error != 0) {
1528 		INP_WUNLOCK(inp);
1529 		INP_INFO_WUNLOCK(&V_udbinfo);
1530 		return (error);
1531 	}
1532 	error = in_pcbconnect(inp, nam, td->td_ucred);
1533 	if (error == 0)
1534 		soisconnected(so);
1535 	INP_WUNLOCK(inp);
1536 	INP_INFO_WUNLOCK(&V_udbinfo);
1537 	return (error);
1538 }
1539 
1540 static void
1541 udp_detach(struct socket *so)
1542 {
1543 	struct inpcb *inp;
1544 	struct udpcb *up;
1545 
1546 	inp = sotoinpcb(so);
1547 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1548 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1549 	    ("udp_detach: not disconnected"));
1550 	INP_INFO_WLOCK(&V_udbinfo);
1551 	INP_WLOCK(inp);
1552 	up = intoudpcb(inp);
1553 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1554 	inp->inp_ppcb = NULL;
1555 	in_pcbdetach(inp);
1556 	in_pcbfree(inp);
1557 	INP_INFO_WUNLOCK(&V_udbinfo);
1558 	udp_discardcb(up);
1559 }
1560 
1561 static int
1562 udp_disconnect(struct socket *so)
1563 {
1564 	struct inpcb *inp;
1565 
1566 	inp = sotoinpcb(so);
1567 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1568 	INP_INFO_WLOCK(&V_udbinfo);
1569 	INP_WLOCK(inp);
1570 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1571 		INP_WUNLOCK(inp);
1572 		INP_INFO_WUNLOCK(&V_udbinfo);
1573 		return (ENOTCONN);
1574 	}
1575 
1576 	in_pcbdisconnect(inp);
1577 	inp->inp_laddr.s_addr = INADDR_ANY;
1578 	SOCK_LOCK(so);
1579 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1580 	SOCK_UNLOCK(so);
1581 	INP_WUNLOCK(inp);
1582 	INP_INFO_WUNLOCK(&V_udbinfo);
1583 	return (0);
1584 }
1585 
1586 static int
1587 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1588     struct mbuf *control, struct thread *td)
1589 {
1590 	struct inpcb *inp;
1591 
1592 	inp = sotoinpcb(so);
1593 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1594 	return (udp_output(inp, m, addr, control, td));
1595 }
1596 
1597 int
1598 udp_shutdown(struct socket *so)
1599 {
1600 	struct inpcb *inp;
1601 
1602 	inp = sotoinpcb(so);
1603 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1604 	INP_WLOCK(inp);
1605 	socantsendmore(so);
1606 	INP_WUNLOCK(inp);
1607 	return (0);
1608 }
1609 
1610 struct pr_usrreqs udp_usrreqs = {
1611 	.pru_abort =		udp_abort,
1612 	.pru_attach =		udp_attach,
1613 	.pru_bind =		udp_bind,
1614 	.pru_connect =		udp_connect,
1615 	.pru_control =		in_control,
1616 	.pru_detach =		udp_detach,
1617 	.pru_disconnect =	udp_disconnect,
1618 	.pru_peeraddr =		in_getpeeraddr,
1619 	.pru_send =		udp_send,
1620 	.pru_soreceive =	soreceive_dgram,
1621 	.pru_sosend =		sosend_dgram,
1622 	.pru_shutdown =		udp_shutdown,
1623 	.pru_sockaddr =		in_getsockaddr,
1624 	.pru_sosetlabel =	in_pcbsosetlabel,
1625 	.pru_close =		udp_close,
1626 };
1627