xref: /freebsd/sys/netinet/udp_usrreq.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
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 	int len;
342 	struct ip save_ip;
343 	struct sockaddr_in udp_in;
344 #ifdef IPFIREWALL_FORWARD
345 	struct m_tag *fwd_tag;
346 #endif
347 
348 	ifp = m->m_pkthdr.rcvif;
349 	UDPSTAT_INC(udps_ipackets);
350 
351 	/*
352 	 * Strip IP options, if any; should skip this, make available to
353 	 * user, and use on returned packets, but we don't yet have a way to
354 	 * check the checksum with options still present.
355 	 */
356 	if (iphlen > sizeof (struct ip)) {
357 		ip_stripoptions(m);
358 		iphlen = sizeof(struct ip);
359 	}
360 
361 	/*
362 	 * Get IP and UDP header together in first mbuf.
363 	 */
364 	ip = mtod(m, struct ip *);
365 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
366 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
367 			UDPSTAT_INC(udps_hdrops);
368 			return;
369 		}
370 		ip = mtod(m, struct ip *);
371 	}
372 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
373 
374 	/*
375 	 * Destination port of 0 is illegal, based on RFC768.
376 	 */
377 	if (uh->uh_dport == 0)
378 		goto badunlocked;
379 
380 	/*
381 	 * Construct sockaddr format source address.  Stuff source address
382 	 * and datagram in user buffer.
383 	 */
384 	bzero(&udp_in, sizeof(udp_in));
385 	udp_in.sin_len = sizeof(udp_in);
386 	udp_in.sin_family = AF_INET;
387 	udp_in.sin_port = uh->uh_sport;
388 	udp_in.sin_addr = ip->ip_src;
389 
390 	/*
391 	 * Make mbuf data length reflect UDP length.  If not enough data to
392 	 * reflect UDP length, drop.
393 	 */
394 	len = ntohs((u_short)uh->uh_ulen);
395 	if (ip->ip_len != len) {
396 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
397 			UDPSTAT_INC(udps_badlen);
398 			goto badunlocked;
399 		}
400 		m_adj(m, len - ip->ip_len);
401 		/* ip->ip_len = len; */
402 	}
403 
404 	/*
405 	 * Save a copy of the IP header in case we want restore it for
406 	 * sending an ICMP error message in response.
407 	 */
408 	if (!V_udp_blackhole)
409 		save_ip = *ip;
410 	else
411 		memset(&save_ip, 0, sizeof(save_ip));
412 
413 	/*
414 	 * Checksum extended UDP header and data.
415 	 */
416 	if (uh->uh_sum) {
417 		u_short uh_sum;
418 
419 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
420 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
421 				uh_sum = m->m_pkthdr.csum_data;
422 			else
423 				uh_sum = in_pseudo(ip->ip_src.s_addr,
424 				    ip->ip_dst.s_addr, htonl((u_short)len +
425 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
426 			uh_sum ^= 0xffff;
427 		} else {
428 			char b[9];
429 
430 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
431 			bzero(((struct ipovly *)ip)->ih_x1, 9);
432 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
433 			uh_sum = in_cksum(m, len + sizeof (struct ip));
434 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
435 		}
436 		if (uh_sum) {
437 			UDPSTAT_INC(udps_badsum);
438 			m_freem(m);
439 			return;
440 		}
441 	} else
442 		UDPSTAT_INC(udps_nosum);
443 
444 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
445 	    in_broadcast(ip->ip_dst, ifp)) {
446 		struct inpcb *last;
447 		struct ip_moptions *imo;
448 
449 		INP_INFO_RLOCK(&V_udbinfo);
450 		last = NULL;
451 		LIST_FOREACH(inp, &V_udb, inp_list) {
452 			if (inp->inp_lport != uh->uh_dport)
453 				continue;
454 #ifdef INET6
455 			if ((inp->inp_vflag & INP_IPV4) == 0)
456 				continue;
457 #endif
458 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
459 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
460 				continue;
461 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
462 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
463 				continue;
464 			if (inp->inp_fport != 0 &&
465 			    inp->inp_fport != uh->uh_sport)
466 				continue;
467 
468 			INP_RLOCK(inp);
469 
470 			/*
471 			 * XXXRW: Because we weren't holding either the inpcb
472 			 * or the hash lock when we checked for a match
473 			 * before, we should probably recheck now that the
474 			 * inpcb lock is held.
475 			 */
476 
477 			/*
478 			 * Handle socket delivery policy for any-source
479 			 * and source-specific multicast. [RFC3678]
480 			 */
481 			imo = inp->inp_moptions;
482 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
483 				struct sockaddr_in	 group;
484 				int			 blocked;
485 				if (imo == NULL) {
486 					INP_RUNLOCK(inp);
487 					continue;
488 				}
489 				bzero(&group, sizeof(struct sockaddr_in));
490 				group.sin_len = sizeof(struct sockaddr_in);
491 				group.sin_family = AF_INET;
492 				group.sin_addr = ip->ip_dst;
493 
494 				blocked = imo_multi_filter(imo, ifp,
495 					(struct sockaddr *)&group,
496 					(struct sockaddr *)&udp_in);
497 				if (blocked != MCAST_PASS) {
498 					if (blocked == MCAST_NOTGMEMBER)
499 						IPSTAT_INC(ips_notmember);
500 					if (blocked == MCAST_NOTSMEMBER ||
501 					    blocked == MCAST_MUTED)
502 						UDPSTAT_INC(udps_filtermcast);
503 					INP_RUNLOCK(inp);
504 					continue;
505 				}
506 			}
507 			if (last != NULL) {
508 				struct mbuf *n;
509 
510 				n = m_copy(m, 0, M_COPYALL);
511 				udp_append(last, ip, n, iphlen, &udp_in);
512 				INP_RUNLOCK(last);
513 			}
514 			last = inp;
515 			/*
516 			 * Don't look for additional matches if this one does
517 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
518 			 * socket options set.  This heuristic avoids
519 			 * searching through all pcbs in the common case of a
520 			 * non-shared port.  It assumes that an application
521 			 * will never clear these options after setting them.
522 			 */
523 			if ((last->inp_socket->so_options &
524 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
525 				break;
526 		}
527 
528 		if (last == NULL) {
529 			/*
530 			 * No matching pcb found; discard datagram.  (No need
531 			 * to send an ICMP Port Unreachable for a broadcast
532 			 * or multicast datgram.)
533 			 */
534 			UDPSTAT_INC(udps_noportbcast);
535 			if (inp)
536 				INP_RUNLOCK(inp);
537 			INP_INFO_RUNLOCK(&V_udbinfo);
538 			goto badunlocked;
539 		}
540 		udp_append(last, ip, m, iphlen, &udp_in);
541 		INP_RUNLOCK(last);
542 		INP_INFO_RUNLOCK(&V_udbinfo);
543 		return;
544 	}
545 
546 	/*
547 	 * Locate pcb for datagram.
548 	 */
549 #ifdef IPFIREWALL_FORWARD
550 	/*
551 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
552 	 */
553 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
554 	if (fwd_tag != NULL) {
555 		struct sockaddr_in *next_hop;
556 
557 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
558 
559 		/*
560 		 * Transparently forwarded. Pretend to be the destination.
561 		 * Already got one like this?
562 		 */
563 		inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
564 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
565 		if (!inp) {
566 			/*
567 			 * It's new.  Try to find the ambushing socket.
568 			 * Because we've rewritten the destination address,
569 			 * any hardware-generated hash is ignored.
570 			 */
571 			inp = in_pcblookup(&V_udbinfo, ip->ip_src,
572 			    uh->uh_sport, next_hop->sin_addr,
573 			    next_hop->sin_port ? htons(next_hop->sin_port) :
574 			    uh->uh_dport, INPLOOKUP_WILDCARD |
575 			    INPLOOKUP_RLOCKPCB, ifp);
576 		}
577 		/* Remove the tag from the packet. We don't need it anymore. */
578 		m_tag_delete(m, fwd_tag);
579 	} else
580 #endif /* IPFIREWALL_FORWARD */
581 		inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
582 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
583 		    INPLOOKUP_RLOCKPCB, ifp, m);
584 	if (inp == NULL) {
585 		if (udp_log_in_vain) {
586 			char buf[4*sizeof "123"];
587 
588 			strcpy(buf, inet_ntoa(ip->ip_dst));
589 			log(LOG_INFO,
590 			    "Connection attempt to UDP %s:%d from %s:%d\n",
591 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
592 			    ntohs(uh->uh_sport));
593 		}
594 		UDPSTAT_INC(udps_noport);
595 		if (m->m_flags & (M_BCAST | M_MCAST)) {
596 			UDPSTAT_INC(udps_noportbcast);
597 			goto badunlocked;
598 		}
599 		if (V_udp_blackhole)
600 			goto badunlocked;
601 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
602 			goto badunlocked;
603 		*ip = save_ip;
604 		ip->ip_len += iphlen;
605 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
606 		return;
607 	}
608 
609 	/*
610 	 * Check the minimum TTL for socket.
611 	 */
612 	INP_RLOCK_ASSERT(inp);
613 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
614 		INP_RUNLOCK(inp);
615 		m_freem(m);
616 		return;
617 	}
618 	udp_append(inp, ip, m, iphlen, &udp_in);
619 	INP_RUNLOCK(inp);
620 	return;
621 
622 badunlocked:
623 	m_freem(m);
624 }
625 #endif /* INET */
626 
627 /*
628  * Notify a udp user of an asynchronous error; just wake up so that they can
629  * collect error status.
630  */
631 struct inpcb *
632 udp_notify(struct inpcb *inp, int errno)
633 {
634 
635 	/*
636 	 * While udp_ctlinput() always calls udp_notify() with a read lock
637 	 * when invoking it directly, in_pcbnotifyall() currently uses write
638 	 * locks due to sharing code with TCP.  For now, accept either a read
639 	 * or a write lock, but a read lock is sufficient.
640 	 */
641 	INP_LOCK_ASSERT(inp);
642 
643 	inp->inp_socket->so_error = errno;
644 	sorwakeup(inp->inp_socket);
645 	sowwakeup(inp->inp_socket);
646 	return (inp);
647 }
648 
649 #ifdef INET
650 void
651 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
652 {
653 	struct ip *ip = vip;
654 	struct udphdr *uh;
655 	struct in_addr faddr;
656 	struct inpcb *inp;
657 
658 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
659 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
660 		return;
661 
662 	/*
663 	 * Redirects don't need to be handled up here.
664 	 */
665 	if (PRC_IS_REDIRECT(cmd))
666 		return;
667 
668 	/*
669 	 * Hostdead is ugly because it goes linearly through all PCBs.
670 	 *
671 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
672 	 * DoS attack on machines with many connections.
673 	 */
674 	if (cmd == PRC_HOSTDEAD)
675 		ip = NULL;
676 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
677 		return;
678 	if (ip != NULL) {
679 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
680 		inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
681 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
682 		if (inp != NULL) {
683 			INP_RLOCK_ASSERT(inp);
684 			if (inp->inp_socket != NULL) {
685 				udp_notify(inp, inetctlerrmap[cmd]);
686 			}
687 			INP_RUNLOCK(inp);
688 		}
689 	} else
690 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
691 		    udp_notify);
692 }
693 #endif /* INET */
694 
695 static int
696 udp_pcblist(SYSCTL_HANDLER_ARGS)
697 {
698 	int error, i, n;
699 	struct inpcb *inp, **inp_list;
700 	inp_gen_t gencnt;
701 	struct xinpgen xig;
702 
703 	/*
704 	 * The process of preparing the PCB list is too time-consuming and
705 	 * resource-intensive to repeat twice on every request.
706 	 */
707 	if (req->oldptr == 0) {
708 		n = V_udbinfo.ipi_count;
709 		n += imax(n / 8, 10);
710 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
711 		return (0);
712 	}
713 
714 	if (req->newptr != 0)
715 		return (EPERM);
716 
717 	/*
718 	 * OK, now we're committed to doing something.
719 	 */
720 	INP_INFO_RLOCK(&V_udbinfo);
721 	gencnt = V_udbinfo.ipi_gencnt;
722 	n = V_udbinfo.ipi_count;
723 	INP_INFO_RUNLOCK(&V_udbinfo);
724 
725 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
726 		+ n * sizeof(struct xinpcb));
727 	if (error != 0)
728 		return (error);
729 
730 	xig.xig_len = sizeof xig;
731 	xig.xig_count = n;
732 	xig.xig_gen = gencnt;
733 	xig.xig_sogen = so_gencnt;
734 	error = SYSCTL_OUT(req, &xig, sizeof xig);
735 	if (error)
736 		return (error);
737 
738 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
739 	if (inp_list == 0)
740 		return (ENOMEM);
741 
742 	INP_INFO_RLOCK(&V_udbinfo);
743 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
744 	     inp = LIST_NEXT(inp, inp_list)) {
745 		INP_WLOCK(inp);
746 		if (inp->inp_gencnt <= gencnt &&
747 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
748 			in_pcbref(inp);
749 			inp_list[i++] = inp;
750 		}
751 		INP_WUNLOCK(inp);
752 	}
753 	INP_INFO_RUNLOCK(&V_udbinfo);
754 	n = i;
755 
756 	error = 0;
757 	for (i = 0; i < n; i++) {
758 		inp = inp_list[i];
759 		INP_RLOCK(inp);
760 		if (inp->inp_gencnt <= gencnt) {
761 			struct xinpcb xi;
762 
763 			bzero(&xi, sizeof(xi));
764 			xi.xi_len = sizeof xi;
765 			/* XXX should avoid extra copy */
766 			bcopy(inp, &xi.xi_inp, sizeof *inp);
767 			if (inp->inp_socket)
768 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
769 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
770 			INP_RUNLOCK(inp);
771 			error = SYSCTL_OUT(req, &xi, sizeof xi);
772 		} else
773 			INP_RUNLOCK(inp);
774 	}
775 	INP_INFO_WLOCK(&V_udbinfo);
776 	for (i = 0; i < n; i++) {
777 		inp = inp_list[i];
778 		INP_RLOCK(inp);
779 		if (!in_pcbrele_rlocked(inp))
780 			INP_RUNLOCK(inp);
781 	}
782 	INP_INFO_WUNLOCK(&V_udbinfo);
783 
784 	if (!error) {
785 		/*
786 		 * Give the user an updated idea of our state.  If the
787 		 * generation differs from what we told her before, she knows
788 		 * that something happened while we were processing this
789 		 * request, and it might be necessary to retry.
790 		 */
791 		INP_INFO_RLOCK(&V_udbinfo);
792 		xig.xig_gen = V_udbinfo.ipi_gencnt;
793 		xig.xig_sogen = so_gencnt;
794 		xig.xig_count = V_udbinfo.ipi_count;
795 		INP_INFO_RUNLOCK(&V_udbinfo);
796 		error = SYSCTL_OUT(req, &xig, sizeof xig);
797 	}
798 	free(inp_list, M_TEMP);
799 	return (error);
800 }
801 
802 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
803     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
804     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
805 
806 #ifdef INET
807 static int
808 udp_getcred(SYSCTL_HANDLER_ARGS)
809 {
810 	struct xucred xuc;
811 	struct sockaddr_in addrs[2];
812 	struct inpcb *inp;
813 	int error;
814 
815 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
816 	if (error)
817 		return (error);
818 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
819 	if (error)
820 		return (error);
821 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
822 	    addrs[0].sin_addr, addrs[0].sin_port,
823 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
824 	if (inp != NULL) {
825 		INP_RLOCK_ASSERT(inp);
826 		if (inp->inp_socket == NULL)
827 			error = ENOENT;
828 		if (error == 0)
829 			error = cr_canseeinpcb(req->td->td_ucred, inp);
830 		if (error == 0)
831 			cru2x(inp->inp_cred, &xuc);
832 		INP_RUNLOCK(inp);
833 	} else
834 		error = ENOENT;
835 	if (error == 0)
836 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
837 	return (error);
838 }
839 
840 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
841     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
842     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
843 #endif /* INET */
844 
845 int
846 udp_ctloutput(struct socket *so, struct sockopt *sopt)
847 {
848 	int error = 0, optval;
849 	struct inpcb *inp;
850 #ifdef IPSEC_NAT_T
851 	struct udpcb *up;
852 #endif
853 
854 	inp = sotoinpcb(so);
855 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
856 	INP_WLOCK(inp);
857 	if (sopt->sopt_level != IPPROTO_UDP) {
858 #ifdef INET6
859 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
860 			INP_WUNLOCK(inp);
861 			error = ip6_ctloutput(so, sopt);
862 		}
863 #endif
864 #if defined(INET) && defined(INET6)
865 		else
866 #endif
867 #ifdef INET
868 		{
869 			INP_WUNLOCK(inp);
870 			error = ip_ctloutput(so, sopt);
871 		}
872 #endif
873 		return (error);
874 	}
875 
876 	switch (sopt->sopt_dir) {
877 	case SOPT_SET:
878 		switch (sopt->sopt_name) {
879 		case UDP_ENCAP:
880 			INP_WUNLOCK(inp);
881 			error = sooptcopyin(sopt, &optval, sizeof optval,
882 					    sizeof optval);
883 			if (error)
884 				break;
885 			inp = sotoinpcb(so);
886 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
887 			INP_WLOCK(inp);
888 #ifdef IPSEC_NAT_T
889 			up = intoudpcb(inp);
890 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
891 #endif
892 			switch (optval) {
893 			case 0:
894 				/* Clear all UDP encap. */
895 #ifdef IPSEC_NAT_T
896 				up->u_flags &= ~UF_ESPINUDP_ALL;
897 #endif
898 				break;
899 #ifdef IPSEC_NAT_T
900 			case UDP_ENCAP_ESPINUDP:
901 			case UDP_ENCAP_ESPINUDP_NON_IKE:
902 				up->u_flags &= ~UF_ESPINUDP_ALL;
903 				if (optval == UDP_ENCAP_ESPINUDP)
904 					up->u_flags |= UF_ESPINUDP;
905 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
906 					up->u_flags |= UF_ESPINUDP_NON_IKE;
907 				break;
908 #endif
909 			default:
910 				error = EINVAL;
911 				break;
912 			}
913 			INP_WUNLOCK(inp);
914 			break;
915 		default:
916 			INP_WUNLOCK(inp);
917 			error = ENOPROTOOPT;
918 			break;
919 		}
920 		break;
921 	case SOPT_GET:
922 		switch (sopt->sopt_name) {
923 #ifdef IPSEC_NAT_T
924 		case UDP_ENCAP:
925 			up = intoudpcb(inp);
926 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
927 			optval = up->u_flags & UF_ESPINUDP_ALL;
928 			INP_WUNLOCK(inp);
929 			error = sooptcopyout(sopt, &optval, sizeof optval);
930 			break;
931 #endif
932 		default:
933 			INP_WUNLOCK(inp);
934 			error = ENOPROTOOPT;
935 			break;
936 		}
937 		break;
938 	}
939 	return (error);
940 }
941 
942 #ifdef INET
943 #define	UH_WLOCKED	2
944 #define	UH_RLOCKED	1
945 #define	UH_UNLOCKED	0
946 static int
947 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
948     struct mbuf *control, struct thread *td)
949 {
950 	struct udpiphdr *ui;
951 	int len = m->m_pkthdr.len;
952 	struct in_addr faddr, laddr;
953 	struct cmsghdr *cm;
954 	struct sockaddr_in *sin, src;
955 	int error = 0;
956 	int ipflags;
957 	u_short fport, lport;
958 	int unlock_udbinfo;
959 	u_char tos;
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 	INP_RLOCK(inp);
976 	tos = inp->inp_ip_tos;
977 	if (control != NULL) {
978 		/*
979 		 * XXX: Currently, we assume all the optional information is
980 		 * stored in a single mbuf.
981 		 */
982 		if (control->m_next) {
983 			INP_RUNLOCK(inp);
984 			m_freem(control);
985 			m_freem(m);
986 			return (EINVAL);
987 		}
988 		for (; control->m_len > 0;
989 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
990 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
991 			cm = mtod(control, struct cmsghdr *);
992 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
993 			    || cm->cmsg_len > control->m_len) {
994 				error = EINVAL;
995 				break;
996 			}
997 			if (cm->cmsg_level != IPPROTO_IP)
998 				continue;
999 
1000 			switch (cm->cmsg_type) {
1001 			case IP_SENDSRCADDR:
1002 				if (cm->cmsg_len !=
1003 				    CMSG_LEN(sizeof(struct in_addr))) {
1004 					error = EINVAL;
1005 					break;
1006 				}
1007 				bzero(&src, sizeof(src));
1008 				src.sin_family = AF_INET;
1009 				src.sin_len = sizeof(src);
1010 				src.sin_port = inp->inp_lport;
1011 				src.sin_addr =
1012 				    *(struct in_addr *)CMSG_DATA(cm);
1013 				break;
1014 
1015 			case IP_TOS:
1016 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1017 					error = EINVAL;
1018 					break;
1019 				}
1020 				tos = *(u_char *)CMSG_DATA(cm);
1021 				break;
1022 
1023 			default:
1024 				error = ENOPROTOOPT;
1025 				break;
1026 			}
1027 			if (error)
1028 				break;
1029 		}
1030 		m_freem(control);
1031 	}
1032 	if (error) {
1033 		INP_RUNLOCK(inp);
1034 		m_freem(m);
1035 		return (error);
1036 	}
1037 
1038 	/*
1039 	 * Depending on whether or not the application has bound or connected
1040 	 * the socket, we may have to do varying levels of work.  The optimal
1041 	 * case is for a connected UDP socket, as a global lock isn't
1042 	 * required at all.
1043 	 *
1044 	 * In order to decide which we need, we require stability of the
1045 	 * inpcb binding, which we ensure by acquiring a read lock on the
1046 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1047 	 * the trylock and retry game; note that we may end up with more
1048 	 * conservative locks than required the second time around, so later
1049 	 * assertions have to accept that.  Further analysis of the number of
1050 	 * misses under contention is required.
1051 	 *
1052 	 * XXXRW: Check that hash locking update here is correct.
1053 	 */
1054 	sin = (struct sockaddr_in *)addr;
1055 	if (sin != NULL &&
1056 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1057 		INP_RUNLOCK(inp);
1058 		INP_WLOCK(inp);
1059 		INP_HASH_WLOCK(&V_udbinfo);
1060 		unlock_udbinfo = UH_WLOCKED;
1061 	} else if ((sin != NULL && (
1062 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1063 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1064 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1065 	    (inp->inp_lport == 0))) ||
1066 	    (src.sin_family == AF_INET)) {
1067 		INP_HASH_RLOCK(&V_udbinfo);
1068 		unlock_udbinfo = UH_RLOCKED;
1069 	} else
1070 		unlock_udbinfo = UH_UNLOCKED;
1071 
1072 	/*
1073 	 * If the IP_SENDSRCADDR control message was specified, override the
1074 	 * source address for this datagram.  Its use is invalidated if the
1075 	 * address thus specified is incomplete or clobbers other inpcbs.
1076 	 */
1077 	laddr = inp->inp_laddr;
1078 	lport = inp->inp_lport;
1079 	if (src.sin_family == AF_INET) {
1080 		INP_HASH_LOCK_ASSERT(&V_udbinfo);
1081 		if ((lport == 0) ||
1082 		    (laddr.s_addr == INADDR_ANY &&
1083 		     src.sin_addr.s_addr == INADDR_ANY)) {
1084 			error = EINVAL;
1085 			goto release;
1086 		}
1087 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1088 		    &laddr.s_addr, &lport, td->td_ucred);
1089 		if (error)
1090 			goto release;
1091 	}
1092 
1093 	/*
1094 	 * If a UDP socket has been connected, then a local address/port will
1095 	 * have been selected and bound.
1096 	 *
1097 	 * If a UDP socket has not been connected to, then an explicit
1098 	 * destination address must be used, in which case a local
1099 	 * address/port may not have been selected and bound.
1100 	 */
1101 	if (sin != NULL) {
1102 		INP_LOCK_ASSERT(inp);
1103 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1104 			error = EISCONN;
1105 			goto release;
1106 		}
1107 
1108 		/*
1109 		 * Jail may rewrite the destination address, so let it do
1110 		 * that before we use it.
1111 		 */
1112 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1113 		if (error)
1114 			goto release;
1115 
1116 		/*
1117 		 * If a local address or port hasn't yet been selected, or if
1118 		 * the destination address needs to be rewritten due to using
1119 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1120 		 * to do the heavy lifting.  Once a port is selected, we
1121 		 * commit the binding back to the socket; we also commit the
1122 		 * binding of the address if in jail.
1123 		 *
1124 		 * If we already have a valid binding and we're not
1125 		 * requesting a destination address rewrite, use a fast path.
1126 		 */
1127 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1128 		    inp->inp_lport == 0 ||
1129 		    sin->sin_addr.s_addr == INADDR_ANY ||
1130 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1131 			INP_HASH_LOCK_ASSERT(&V_udbinfo);
1132 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1133 			    &lport, &faddr.s_addr, &fport, NULL,
1134 			    td->td_ucred);
1135 			if (error)
1136 				goto release;
1137 
1138 			/*
1139 			 * XXXRW: Why not commit the port if the address is
1140 			 * !INADDR_ANY?
1141 			 */
1142 			/* Commit the local port if newly assigned. */
1143 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1144 			    inp->inp_lport == 0) {
1145 				INP_WLOCK_ASSERT(inp);
1146 				INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1147 				/*
1148 				 * Remember addr if jailed, to prevent
1149 				 * rebinding.
1150 				 */
1151 				if (prison_flag(td->td_ucred, PR_IP4))
1152 					inp->inp_laddr = laddr;
1153 				inp->inp_lport = lport;
1154 				if (in_pcbinshash(inp) != 0) {
1155 					inp->inp_lport = 0;
1156 					error = EAGAIN;
1157 					goto release;
1158 				}
1159 				inp->inp_flags |= INP_ANONPORT;
1160 			}
1161 		} else {
1162 			faddr = sin->sin_addr;
1163 			fport = sin->sin_port;
1164 		}
1165 	} else {
1166 		INP_LOCK_ASSERT(inp);
1167 		faddr = inp->inp_faddr;
1168 		fport = inp->inp_fport;
1169 		if (faddr.s_addr == INADDR_ANY) {
1170 			error = ENOTCONN;
1171 			goto release;
1172 		}
1173 	}
1174 
1175 	/*
1176 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1177 	 * link-layer headers.  Immediate slide the data pointer back forward
1178 	 * since we won't use that space at this layer.
1179 	 */
1180 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1181 	if (m == NULL) {
1182 		error = ENOBUFS;
1183 		goto release;
1184 	}
1185 	m->m_data += max_linkhdr;
1186 	m->m_len -= max_linkhdr;
1187 	m->m_pkthdr.len -= max_linkhdr;
1188 
1189 	/*
1190 	 * Fill in mbuf with extended UDP header and addresses and length put
1191 	 * into network format.
1192 	 */
1193 	ui = mtod(m, struct udpiphdr *);
1194 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1195 	ui->ui_pr = IPPROTO_UDP;
1196 	ui->ui_src = laddr;
1197 	ui->ui_dst = faddr;
1198 	ui->ui_sport = lport;
1199 	ui->ui_dport = fport;
1200 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1201 
1202 	/*
1203 	 * Set the Don't Fragment bit in the IP header.
1204 	 */
1205 	if (inp->inp_flags & INP_DONTFRAG) {
1206 		struct ip *ip;
1207 
1208 		ip = (struct ip *)&ui->ui_i;
1209 		ip->ip_off |= IP_DF;
1210 	}
1211 
1212 	ipflags = 0;
1213 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1214 		ipflags |= IP_ROUTETOIF;
1215 	if (inp->inp_socket->so_options & SO_BROADCAST)
1216 		ipflags |= IP_ALLOWBROADCAST;
1217 	if (inp->inp_flags & INP_ONESBCAST)
1218 		ipflags |= IP_SENDONES;
1219 
1220 #ifdef MAC
1221 	mac_inpcb_create_mbuf(inp, m);
1222 #endif
1223 
1224 	/*
1225 	 * Set up checksum and output datagram.
1226 	 */
1227 	if (V_udp_cksum) {
1228 		if (inp->inp_flags & INP_ONESBCAST)
1229 			faddr.s_addr = INADDR_BROADCAST;
1230 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1231 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1232 		m->m_pkthdr.csum_flags = CSUM_UDP;
1233 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1234 	} else
1235 		ui->ui_sum = 0;
1236 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1237 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1238 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1239 	UDPSTAT_INC(udps_opackets);
1240 
1241 	if (unlock_udbinfo == UH_WLOCKED)
1242 		INP_HASH_WUNLOCK(&V_udbinfo);
1243 	else if (unlock_udbinfo == UH_RLOCKED)
1244 		INP_HASH_RUNLOCK(&V_udbinfo);
1245 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1246 	    inp->inp_moptions, inp);
1247 	if (unlock_udbinfo == UH_WLOCKED)
1248 		INP_WUNLOCK(inp);
1249 	else
1250 		INP_RUNLOCK(inp);
1251 	return (error);
1252 
1253 release:
1254 	if (unlock_udbinfo == UH_WLOCKED) {
1255 		INP_HASH_WUNLOCK(&V_udbinfo);
1256 		INP_WUNLOCK(inp);
1257 	} else if (unlock_udbinfo == UH_RLOCKED) {
1258 		INP_HASH_RUNLOCK(&V_udbinfo);
1259 		INP_RUNLOCK(inp);
1260 	} else
1261 		INP_RUNLOCK(inp);
1262 	m_freem(m);
1263 	return (error);
1264 }
1265 
1266 
1267 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1268 /*
1269  * Potentially decap ESP in UDP frame.  Check for an ESP header
1270  * and optional marker; if present, strip the UDP header and
1271  * push the result through IPSec.
1272  *
1273  * Returns mbuf to be processed (potentially re-allocated) or
1274  * NULL if consumed and/or processed.
1275  */
1276 static struct mbuf *
1277 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1278 {
1279 	size_t minlen, payload, skip, iphlen;
1280 	caddr_t data;
1281 	struct udpcb *up;
1282 	struct m_tag *tag;
1283 	struct udphdr *udphdr;
1284 	struct ip *ip;
1285 
1286 	INP_RLOCK_ASSERT(inp);
1287 
1288 	/*
1289 	 * Pull up data so the longest case is contiguous:
1290 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1291 	 */
1292 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1293 	if (minlen > m->m_pkthdr.len)
1294 		minlen = m->m_pkthdr.len;
1295 	if ((m = m_pullup(m, minlen)) == NULL) {
1296 		V_ipsec4stat.in_inval++;
1297 		return (NULL);		/* Bypass caller processing. */
1298 	}
1299 	data = mtod(m, caddr_t);	/* Points to ip header. */
1300 	payload = m->m_len - off;	/* Size of payload. */
1301 
1302 	if (payload == 1 && data[off] == '\xff')
1303 		return (m);		/* NB: keepalive packet, no decap. */
1304 
1305 	up = intoudpcb(inp);
1306 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1307 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1308 	    ("u_flags 0x%x", up->u_flags));
1309 
1310 	/*
1311 	 * Check that the payload is large enough to hold an
1312 	 * ESP header and compute the amount of data to remove.
1313 	 *
1314 	 * NB: the caller has already done a pullup for us.
1315 	 * XXX can we assume alignment and eliminate bcopys?
1316 	 */
1317 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1318 		/*
1319 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1320 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1321 		 * possible AH mode non-IKE marker+non-ESP marker
1322 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1323 		 */
1324 		uint64_t marker;
1325 
1326 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1327 			return (m);	/* NB: no decap. */
1328 		bcopy(data + off, &marker, sizeof(uint64_t));
1329 		if (marker != 0)	/* Non-IKE marker. */
1330 			return (m);	/* NB: no decap. */
1331 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1332 	} else {
1333 		uint32_t spi;
1334 
1335 		if (payload <= sizeof(struct esp)) {
1336 			V_ipsec4stat.in_inval++;
1337 			m_freem(m);
1338 			return (NULL);	/* Discard. */
1339 		}
1340 		bcopy(data + off, &spi, sizeof(uint32_t));
1341 		if (spi == 0)		/* Non-ESP marker. */
1342 			return (m);	/* NB: no decap. */
1343 		skip = sizeof(struct udphdr);
1344 	}
1345 
1346 	/*
1347 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1348 	 * the UDP ports. This is required if we want to select
1349 	 * the right SPD for multiple hosts behind same NAT.
1350 	 *
1351 	 * NB: ports are maintained in network byte order everywhere
1352 	 *     in the NAT-T code.
1353 	 */
1354 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1355 		2 * sizeof(uint16_t), M_NOWAIT);
1356 	if (tag == NULL) {
1357 		V_ipsec4stat.in_nomem++;
1358 		m_freem(m);
1359 		return (NULL);		/* Discard. */
1360 	}
1361 	iphlen = off - sizeof(struct udphdr);
1362 	udphdr = (struct udphdr *)(data + iphlen);
1363 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1364 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1365 	m_tag_prepend(m, tag);
1366 
1367 	/*
1368 	 * Remove the UDP header (and possibly the non ESP marker)
1369 	 * IP header length is iphlen
1370 	 * Before:
1371 	 *   <--- off --->
1372 	 *   +----+------+-----+
1373 	 *   | IP |  UDP | ESP |
1374 	 *   +----+------+-----+
1375 	 *        <-skip->
1376 	 * After:
1377 	 *          +----+-----+
1378 	 *          | IP | ESP |
1379 	 *          +----+-----+
1380 	 *   <-skip->
1381 	 */
1382 	ovbcopy(data, data + skip, iphlen);
1383 	m_adj(m, skip);
1384 
1385 	ip = mtod(m, struct ip *);
1386 	ip->ip_len -= skip;
1387 	ip->ip_p = IPPROTO_ESP;
1388 
1389 	/*
1390 	 * We cannot yet update the cksums so clear any
1391 	 * h/w cksum flags as they are no longer valid.
1392 	 */
1393 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1394 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1395 
1396 	(void) ipsec4_common_input(m, iphlen, ip->ip_p);
1397 	return (NULL);			/* NB: consumed, bypass processing. */
1398 }
1399 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1400 
1401 static void
1402 udp_abort(struct socket *so)
1403 {
1404 	struct inpcb *inp;
1405 
1406 	inp = sotoinpcb(so);
1407 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1408 	INP_WLOCK(inp);
1409 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1410 		INP_HASH_WLOCK(&V_udbinfo);
1411 		in_pcbdisconnect(inp);
1412 		inp->inp_laddr.s_addr = INADDR_ANY;
1413 		INP_HASH_WUNLOCK(&V_udbinfo);
1414 		soisdisconnected(so);
1415 	}
1416 	INP_WUNLOCK(inp);
1417 }
1418 
1419 static int
1420 udp_attach(struct socket *so, int proto, struct thread *td)
1421 {
1422 	struct inpcb *inp;
1423 	int error;
1424 
1425 	inp = sotoinpcb(so);
1426 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1427 	error = soreserve(so, udp_sendspace, udp_recvspace);
1428 	if (error)
1429 		return (error);
1430 	INP_INFO_WLOCK(&V_udbinfo);
1431 	error = in_pcballoc(so, &V_udbinfo);
1432 	if (error) {
1433 		INP_INFO_WUNLOCK(&V_udbinfo);
1434 		return (error);
1435 	}
1436 
1437 	inp = sotoinpcb(so);
1438 	inp->inp_vflag |= INP_IPV4;
1439 	inp->inp_ip_ttl = V_ip_defttl;
1440 
1441 	error = udp_newudpcb(inp);
1442 	if (error) {
1443 		in_pcbdetach(inp);
1444 		in_pcbfree(inp);
1445 		INP_INFO_WUNLOCK(&V_udbinfo);
1446 		return (error);
1447 	}
1448 
1449 	INP_WUNLOCK(inp);
1450 	INP_INFO_WUNLOCK(&V_udbinfo);
1451 	return (0);
1452 }
1453 #endif /* INET */
1454 
1455 int
1456 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1457 {
1458 	struct inpcb *inp;
1459 	struct udpcb *up;
1460 
1461 	KASSERT(so->so_type == SOCK_DGRAM,
1462 	    ("udp_set_kernel_tunneling: !dgram"));
1463 	inp = sotoinpcb(so);
1464 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1465 	INP_WLOCK(inp);
1466 	up = intoudpcb(inp);
1467 	if (up->u_tun_func != NULL) {
1468 		INP_WUNLOCK(inp);
1469 		return (EBUSY);
1470 	}
1471 	up->u_tun_func = f;
1472 	INP_WUNLOCK(inp);
1473 	return (0);
1474 }
1475 
1476 #ifdef INET
1477 static int
1478 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1479 {
1480 	struct inpcb *inp;
1481 	int error;
1482 
1483 	inp = sotoinpcb(so);
1484 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1485 	INP_WLOCK(inp);
1486 	INP_HASH_WLOCK(&V_udbinfo);
1487 	error = in_pcbbind(inp, nam, td->td_ucred);
1488 	INP_HASH_WUNLOCK(&V_udbinfo);
1489 	INP_WUNLOCK(inp);
1490 	return (error);
1491 }
1492 
1493 static void
1494 udp_close(struct socket *so)
1495 {
1496 	struct inpcb *inp;
1497 
1498 	inp = sotoinpcb(so);
1499 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1500 	INP_WLOCK(inp);
1501 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1502 		INP_HASH_WLOCK(&V_udbinfo);
1503 		in_pcbdisconnect(inp);
1504 		inp->inp_laddr.s_addr = INADDR_ANY;
1505 		INP_HASH_WUNLOCK(&V_udbinfo);
1506 		soisdisconnected(so);
1507 	}
1508 	INP_WUNLOCK(inp);
1509 }
1510 
1511 static int
1512 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1513 {
1514 	struct inpcb *inp;
1515 	int error;
1516 	struct sockaddr_in *sin;
1517 
1518 	inp = sotoinpcb(so);
1519 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1520 	INP_WLOCK(inp);
1521 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1522 		INP_WUNLOCK(inp);
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 		return (error);
1530 	}
1531 	INP_HASH_WLOCK(&V_udbinfo);
1532 	error = in_pcbconnect(inp, nam, td->td_ucred);
1533 	INP_HASH_WUNLOCK(&V_udbinfo);
1534 	if (error == 0)
1535 		soisconnected(so);
1536 	INP_WUNLOCK(inp);
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_WLOCK(inp);
1569 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1570 		INP_WUNLOCK(inp);
1571 		return (ENOTCONN);
1572 	}
1573 	INP_HASH_WLOCK(&V_udbinfo);
1574 	in_pcbdisconnect(inp);
1575 	inp->inp_laddr.s_addr = INADDR_ANY;
1576 	INP_HASH_WUNLOCK(&V_udbinfo);
1577 	SOCK_LOCK(so);
1578 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1579 	SOCK_UNLOCK(so);
1580 	INP_WUNLOCK(inp);
1581 	return (0);
1582 }
1583 
1584 static int
1585 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1586     struct mbuf *control, struct thread *td)
1587 {
1588 	struct inpcb *inp;
1589 
1590 	inp = sotoinpcb(so);
1591 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1592 	return (udp_output(inp, m, addr, control, td));
1593 }
1594 #endif /* INET */
1595 
1596 int
1597 udp_shutdown(struct socket *so)
1598 {
1599 	struct inpcb *inp;
1600 
1601 	inp = sotoinpcb(so);
1602 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1603 	INP_WLOCK(inp);
1604 	socantsendmore(so);
1605 	INP_WUNLOCK(inp);
1606 	return (0);
1607 }
1608 
1609 #ifdef INET
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 #endif /* INET */
1628