xref: /freebsd/sys/netinet/udp_usrreq.c (revision 11b249f99e4d4891bd2a21ef796d40073ee15508)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2008 Robert N. M. Watson
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * Copyright (c) 2014 Kevin Lo
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Robert N. M. Watson under
10  * contract to Juniper Networks, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ipfw.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
46 #include "opt_rss.h"
47 
48 #include <sys/param.h>
49 #include <sys/domain.h>
50 #include <sys/eventhandler.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/protosw.h>
59 #include <sys/sdt.h>
60 #include <sys/signalvar.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/syslog.h>
66 #include <sys/systm.h>
67 
68 #include <vm/uma.h>
69 
70 #include <net/if.h>
71 #include <net/if_var.h>
72 #include <net/route.h>
73 #include <net/rss_config.h>
74 
75 #include <netinet/in.h>
76 #include <netinet/in_kdtrace.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #endif
84 #include <netinet/ip_icmp.h>
85 #include <netinet/icmp_var.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_options.h>
88 #ifdef INET6
89 #include <netinet6/ip6_var.h>
90 #endif
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
93 #include <netinet/udplite.h>
94 #include <netinet/in_rss.h>
95 
96 #ifdef IPSEC
97 #include <netipsec/ipsec.h>
98 #include <netipsec/esp.h>
99 #endif
100 
101 #include <machine/in_cksum.h>
102 
103 #include <security/mac/mac_framework.h>
104 
105 /*
106  * UDP and UDP-Lite protocols implementation.
107  * Per RFC 768, August, 1980.
108  * Per RFC 3828, July, 2004.
109  */
110 
111 /*
112  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
113  * removes the only data integrity mechanism for packets and malformed
114  * packets that would otherwise be discarded due to bad checksums, and may
115  * cause problems (especially for NFS data blocks).
116  */
117 VNET_DEFINE(int, udp_cksum) = 1;
118 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
119     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
120 
121 int	udp_log_in_vain = 0;
122 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
123     &udp_log_in_vain, 0, "Log all incoming UDP packets");
124 
125 VNET_DEFINE(int, udp_blackhole) = 0;
126 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
127     &VNET_NAME(udp_blackhole), 0,
128     "Do not send port unreachables for refused connects");
129 
130 u_long	udp_sendspace = 9216;		/* really max datagram size */
131 					/* 40 1K datagrams */
132 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
133     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
134 
135 u_long	udp_recvspace = 40 * (1024 +
136 #ifdef INET6
137 				      sizeof(struct sockaddr_in6)
138 #else
139 				      sizeof(struct sockaddr_in)
140 #endif
141 				      );
142 
143 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
144     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
145 
146 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
147 VNET_DEFINE(struct inpcbinfo, udbinfo);
148 VNET_DEFINE(struct inpcbhead, ulitecb);
149 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
150 static VNET_DEFINE(uma_zone_t, udpcb_zone);
151 #define	V_udpcb_zone			VNET(udpcb_zone)
152 
153 #ifndef UDBHASHSIZE
154 #define	UDBHASHSIZE	128
155 #endif
156 
157 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
158 VNET_PCPUSTAT_SYSINIT(udpstat);
159 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
160     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
161 
162 #ifdef VIMAGE
163 VNET_PCPUSTAT_SYSUNINIT(udpstat);
164 #endif /* VIMAGE */
165 #ifdef INET
166 static void	udp_detach(struct socket *so);
167 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
168 		    struct mbuf *, struct thread *);
169 #endif
170 
171 #ifdef IPSEC
172 #ifdef IPSEC_NAT_T
173 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
174 #ifdef INET
175 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
176 #endif
177 #endif /* IPSEC_NAT_T */
178 #endif /* IPSEC */
179 
180 static void
181 udp_zone_change(void *tag)
182 {
183 
184 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
185 	uma_zone_set_max(V_udpcb_zone, maxsockets);
186 }
187 
188 static int
189 udp_inpcb_init(void *mem, int size, int flags)
190 {
191 	struct inpcb *inp;
192 
193 	inp = mem;
194 	INP_LOCK_INIT(inp, "inp", "udpinp");
195 	return (0);
196 }
197 
198 static int
199 udplite_inpcb_init(void *mem, int size, int flags)
200 {
201 	struct inpcb *inp;
202 
203 	inp = mem;
204 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
205 	return (0);
206 }
207 
208 void
209 udp_init(void)
210 {
211 
212 	/*
213 	 * For now default to 2-tuple UDP hashing - until the fragment
214 	 * reassembly code can also update the flowid.
215 	 *
216 	 * Once we can calculate the flowid that way and re-establish
217 	 * a 4-tuple, flip this to 4-tuple.
218 	 */
219 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
220 	    "udp_inpcb", udp_inpcb_init, NULL, 0,
221 	    IPI_HASHFIELDS_2TUPLE);
222 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
223 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
224 	uma_zone_set_max(V_udpcb_zone, maxsockets);
225 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
226 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
227 	    EVENTHANDLER_PRI_ANY);
228 }
229 
230 void
231 udplite_init(void)
232 {
233 
234 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
235 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
236 	    0, IPI_HASHFIELDS_2TUPLE);
237 }
238 
239 /*
240  * Kernel module interface for updating udpstat.  The argument is an index
241  * into udpstat treated as an array of u_long.  While this encodes the
242  * general layout of udpstat into the caller, it doesn't encode its location,
243  * so that future changes to add, for example, per-CPU stats support won't
244  * cause binary compatibility problems for kernel modules.
245  */
246 void
247 kmod_udpstat_inc(int statnum)
248 {
249 
250 	counter_u64_add(VNET(udpstat)[statnum], 1);
251 }
252 
253 int
254 udp_newudpcb(struct inpcb *inp)
255 {
256 	struct udpcb *up;
257 
258 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
259 	if (up == NULL)
260 		return (ENOBUFS);
261 	inp->inp_ppcb = up;
262 	return (0);
263 }
264 
265 void
266 udp_discardcb(struct udpcb *up)
267 {
268 
269 	uma_zfree(V_udpcb_zone, up);
270 }
271 
272 #ifdef VIMAGE
273 void
274 udp_destroy(void)
275 {
276 
277 	in_pcbinfo_destroy(&V_udbinfo);
278 	uma_zdestroy(V_udpcb_zone);
279 }
280 
281 void
282 udplite_destroy(void)
283 {
284 
285 	in_pcbinfo_destroy(&V_ulitecbinfo);
286 }
287 #endif
288 
289 #ifdef INET
290 /*
291  * Subroutine of udp_input(), which appends the provided mbuf chain to the
292  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
293  * contains the source address.  If the socket ends up being an IPv6 socket,
294  * udp_append() will convert to a sockaddr_in6 before passing the address
295  * into the socket code.
296  *
297  * In the normal case udp_append() will return 0, indicating that you
298  * must unlock the inp. However if a tunneling protocol is in place we increment
299  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
300  * then decrement the reference count. If the inp_rele returns 1, indicating the
301  * inp is gone, we return that to the caller to tell them *not* to unlock
302  * the inp. In the case of multi-cast this will cause the distribution
303  * to stop (though most tunneling protocols known currently do *not* use
304  * multicast).
305  */
306 static int
307 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
308     struct sockaddr_in *udp_in)
309 {
310 	struct sockaddr *append_sa;
311 	struct socket *so;
312 	struct mbuf *opts = 0;
313 #ifdef INET6
314 	struct sockaddr_in6 udp_in6;
315 #endif
316 	struct udpcb *up;
317 
318 	INP_LOCK_ASSERT(inp);
319 
320 	/*
321 	 * Engage the tunneling protocol.
322 	 */
323 	up = intoudpcb(inp);
324 	if (up->u_tun_func != NULL) {
325 		in_pcbref(inp);
326 		INP_RUNLOCK(inp);
327 		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
328 		    up->u_tun_ctx);
329 		INP_RLOCK(inp);
330 		return (in_pcbrele_rlocked(inp));
331 	}
332 
333 	off += sizeof(struct udphdr);
334 
335 #ifdef IPSEC
336 	/* Check AH/ESP integrity. */
337 	if (ipsec4_in_reject(n, inp)) {
338 		m_freem(n);
339 		return (0);
340 	}
341 #ifdef IPSEC_NAT_T
342 	up = intoudpcb(inp);
343 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
344 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
345 		n = udp4_espdecap(inp, n, off);
346 		if (n == NULL)				/* Consumed. */
347 			return (0);
348 	}
349 #endif /* IPSEC_NAT_T */
350 #endif /* IPSEC */
351 #ifdef MAC
352 	if (mac_inpcb_check_deliver(inp, n) != 0) {
353 		m_freem(n);
354 		return (0);
355 	}
356 #endif /* MAC */
357 	if (inp->inp_flags & INP_CONTROLOPTS ||
358 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
359 #ifdef INET6
360 		if (inp->inp_vflag & INP_IPV6)
361 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
362 		else
363 #endif /* INET6 */
364 			ip_savecontrol(inp, &opts, ip, n);
365 	}
366 #ifdef INET6
367 	if (inp->inp_vflag & INP_IPV6) {
368 		bzero(&udp_in6, sizeof(udp_in6));
369 		udp_in6.sin6_len = sizeof(udp_in6);
370 		udp_in6.sin6_family = AF_INET6;
371 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
372 		append_sa = (struct sockaddr *)&udp_in6;
373 	} else
374 #endif /* INET6 */
375 		append_sa = (struct sockaddr *)udp_in;
376 	m_adj(n, off);
377 
378 	so = inp->inp_socket;
379 	SOCKBUF_LOCK(&so->so_rcv);
380 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
381 		SOCKBUF_UNLOCK(&so->so_rcv);
382 		m_freem(n);
383 		if (opts)
384 			m_freem(opts);
385 		UDPSTAT_INC(udps_fullsock);
386 	} else
387 		sorwakeup_locked(so);
388 	return (0);
389 }
390 
391 int
392 udp_input(struct mbuf **mp, int *offp, int proto)
393 {
394 	struct ip *ip;
395 	struct udphdr *uh;
396 	struct ifnet *ifp;
397 	struct inpcb *inp;
398 	uint16_t len, ip_len;
399 	struct inpcbinfo *pcbinfo;
400 	struct ip save_ip;
401 	struct sockaddr_in udp_in;
402 	struct mbuf *m;
403 	struct m_tag *fwd_tag;
404 	int cscov_partial, iphlen;
405 
406 	m = *mp;
407 	iphlen = *offp;
408 	ifp = m->m_pkthdr.rcvif;
409 	*mp = NULL;
410 	UDPSTAT_INC(udps_ipackets);
411 
412 	/*
413 	 * Strip IP options, if any; should skip this, make available to
414 	 * user, and use on returned packets, but we don't yet have a way to
415 	 * check the checksum with options still present.
416 	 */
417 	if (iphlen > sizeof (struct ip)) {
418 		ip_stripoptions(m);
419 		iphlen = sizeof(struct ip);
420 	}
421 
422 	/*
423 	 * Get IP and UDP header together in first mbuf.
424 	 */
425 	ip = mtod(m, struct ip *);
426 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
427 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
428 			UDPSTAT_INC(udps_hdrops);
429 			return (IPPROTO_DONE);
430 		}
431 		ip = mtod(m, struct ip *);
432 	}
433 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
434 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
435 
436 	/*
437 	 * Destination port of 0 is illegal, based on RFC768.
438 	 */
439 	if (uh->uh_dport == 0)
440 		goto badunlocked;
441 
442 	/*
443 	 * Construct sockaddr format source address.  Stuff source address
444 	 * and datagram in user buffer.
445 	 */
446 	bzero(&udp_in, sizeof(udp_in));
447 	udp_in.sin_len = sizeof(udp_in);
448 	udp_in.sin_family = AF_INET;
449 	udp_in.sin_port = uh->uh_sport;
450 	udp_in.sin_addr = ip->ip_src;
451 
452 	/*
453 	 * Make mbuf data length reflect UDP length.  If not enough data to
454 	 * reflect UDP length, drop.
455 	 */
456 	len = ntohs((u_short)uh->uh_ulen);
457 	ip_len = ntohs(ip->ip_len) - iphlen;
458 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
459 		/* Zero means checksum over the complete packet. */
460 		if (len == 0)
461 			len = ip_len;
462 		cscov_partial = 0;
463 	}
464 	if (ip_len != len) {
465 		if (len > ip_len || len < sizeof(struct udphdr)) {
466 			UDPSTAT_INC(udps_badlen);
467 			goto badunlocked;
468 		}
469 		if (proto == IPPROTO_UDP)
470 			m_adj(m, len - ip_len);
471 	}
472 
473 	/*
474 	 * Save a copy of the IP header in case we want restore it for
475 	 * sending an ICMP error message in response.
476 	 */
477 	if (!V_udp_blackhole)
478 		save_ip = *ip;
479 	else
480 		memset(&save_ip, 0, sizeof(save_ip));
481 
482 	/*
483 	 * Checksum extended UDP header and data.
484 	 */
485 	if (uh->uh_sum) {
486 		u_short uh_sum;
487 
488 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
489 		    !cscov_partial) {
490 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
491 				uh_sum = m->m_pkthdr.csum_data;
492 			else
493 				uh_sum = in_pseudo(ip->ip_src.s_addr,
494 				    ip->ip_dst.s_addr, htonl((u_short)len +
495 				    m->m_pkthdr.csum_data + proto));
496 			uh_sum ^= 0xffff;
497 		} else {
498 			char b[9];
499 
500 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
501 			bzero(((struct ipovly *)ip)->ih_x1, 9);
502 			((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
503 			    uh->uh_ulen : htons(ip_len);
504 			uh_sum = in_cksum(m, len + sizeof (struct ip));
505 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
506 		}
507 		if (uh_sum) {
508 			UDPSTAT_INC(udps_badsum);
509 			m_freem(m);
510 			return (IPPROTO_DONE);
511 		}
512 	} else {
513 		if (proto == IPPROTO_UDP) {
514 			UDPSTAT_INC(udps_nosum);
515 		} else {
516 			/* UDPLite requires a checksum */
517 			/* XXX: What is the right UDPLite MIB counter here? */
518 			m_freem(m);
519 			return (IPPROTO_DONE);
520 		}
521 	}
522 
523 	pcbinfo = udp_get_inpcbinfo(proto);
524 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
525 	    in_broadcast(ip->ip_dst, ifp)) {
526 		struct inpcb *last;
527 		struct inpcbhead *pcblist;
528 		struct ip_moptions *imo;
529 
530 		INP_INFO_RLOCK(pcbinfo);
531 		pcblist = udp_get_pcblist(proto);
532 		last = NULL;
533 		LIST_FOREACH(inp, pcblist, inp_list) {
534 			if (inp->inp_lport != uh->uh_dport)
535 				continue;
536 #ifdef INET6
537 			if ((inp->inp_vflag & INP_IPV4) == 0)
538 				continue;
539 #endif
540 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
541 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
542 				continue;
543 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
544 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
545 				continue;
546 			if (inp->inp_fport != 0 &&
547 			    inp->inp_fport != uh->uh_sport)
548 				continue;
549 
550 			INP_RLOCK(inp);
551 
552 			/*
553 			 * XXXRW: Because we weren't holding either the inpcb
554 			 * or the hash lock when we checked for a match
555 			 * before, we should probably recheck now that the
556 			 * inpcb lock is held.
557 			 */
558 
559 			/*
560 			 * Handle socket delivery policy for any-source
561 			 * and source-specific multicast. [RFC3678]
562 			 */
563 			imo = inp->inp_moptions;
564 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
565 				struct sockaddr_in	 group;
566 				int			 blocked;
567 				if (imo == NULL) {
568 					INP_RUNLOCK(inp);
569 					continue;
570 				}
571 				bzero(&group, sizeof(struct sockaddr_in));
572 				group.sin_len = sizeof(struct sockaddr_in);
573 				group.sin_family = AF_INET;
574 				group.sin_addr = ip->ip_dst;
575 
576 				blocked = imo_multi_filter(imo, ifp,
577 					(struct sockaddr *)&group,
578 					(struct sockaddr *)&udp_in);
579 				if (blocked != MCAST_PASS) {
580 					if (blocked == MCAST_NOTGMEMBER)
581 						IPSTAT_INC(ips_notmember);
582 					if (blocked == MCAST_NOTSMEMBER ||
583 					    blocked == MCAST_MUTED)
584 						UDPSTAT_INC(udps_filtermcast);
585 					INP_RUNLOCK(inp);
586 					continue;
587 				}
588 			}
589 			if (last != NULL) {
590 				struct mbuf *n;
591 
592 				if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
593 					UDP_PROBE(receive, NULL, last, ip,
594 					    last, uh);
595 					if (udp_append(last, ip, n, iphlen,
596 						&udp_in)) {
597 						goto inp_lost;
598 					}
599 				}
600 				INP_RUNLOCK(last);
601 			}
602 			last = inp;
603 			/*
604 			 * Don't look for additional matches if this one does
605 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
606 			 * socket options set.  This heuristic avoids
607 			 * searching through all pcbs in the common case of a
608 			 * non-shared port.  It assumes that an application
609 			 * will never clear these options after setting them.
610 			 */
611 			if ((last->inp_socket->so_options &
612 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
613 				break;
614 		}
615 
616 		if (last == NULL) {
617 			/*
618 			 * No matching pcb found; discard datagram.  (No need
619 			 * to send an ICMP Port Unreachable for a broadcast
620 			 * or multicast datgram.)
621 			 */
622 			UDPSTAT_INC(udps_noportbcast);
623 			if (inp)
624 				INP_RUNLOCK(inp);
625 			INP_INFO_RUNLOCK(pcbinfo);
626 			goto badunlocked;
627 		}
628 		UDP_PROBE(receive, NULL, last, ip, last, uh);
629 		if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
630 			INP_RUNLOCK(last);
631 	inp_lost:
632 		INP_INFO_RUNLOCK(pcbinfo);
633 		return (IPPROTO_DONE);
634 	}
635 
636 	/*
637 	 * Locate pcb for datagram.
638 	 */
639 
640 	/*
641 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
642 	 */
643 	if ((m->m_flags & M_IP_NEXTHOP) &&
644 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
645 		struct sockaddr_in *next_hop;
646 
647 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
648 
649 		/*
650 		 * Transparently forwarded. Pretend to be the destination.
651 		 * Already got one like this?
652 		 */
653 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
654 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
655 		if (!inp) {
656 			/*
657 			 * It's new.  Try to find the ambushing socket.
658 			 * Because we've rewritten the destination address,
659 			 * any hardware-generated hash is ignored.
660 			 */
661 			inp = in_pcblookup(pcbinfo, ip->ip_src,
662 			    uh->uh_sport, next_hop->sin_addr,
663 			    next_hop->sin_port ? htons(next_hop->sin_port) :
664 			    uh->uh_dport, INPLOOKUP_WILDCARD |
665 			    INPLOOKUP_RLOCKPCB, ifp);
666 		}
667 		/* Remove the tag from the packet. We don't need it anymore. */
668 		m_tag_delete(m, fwd_tag);
669 		m->m_flags &= ~M_IP_NEXTHOP;
670 	} else
671 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
672 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
673 		    INPLOOKUP_RLOCKPCB, ifp, m);
674 	if (inp == NULL) {
675 		if (udp_log_in_vain) {
676 			char buf[4*sizeof "123"];
677 
678 			strcpy(buf, inet_ntoa(ip->ip_dst));
679 			log(LOG_INFO,
680 			    "Connection attempt to UDP %s:%d from %s:%d\n",
681 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
682 			    ntohs(uh->uh_sport));
683 		}
684 		UDPSTAT_INC(udps_noport);
685 		if (m->m_flags & (M_BCAST | M_MCAST)) {
686 			UDPSTAT_INC(udps_noportbcast);
687 			goto badunlocked;
688 		}
689 		if (V_udp_blackhole)
690 			goto badunlocked;
691 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
692 			goto badunlocked;
693 		*ip = save_ip;
694 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
695 		return (IPPROTO_DONE);
696 	}
697 
698 	/*
699 	 * Check the minimum TTL for socket.
700 	 */
701 	INP_RLOCK_ASSERT(inp);
702 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
703 		INP_RUNLOCK(inp);
704 		m_freem(m);
705 		return (IPPROTO_DONE);
706 	}
707 	if (cscov_partial) {
708 		struct udpcb *up;
709 
710 		up = intoudpcb(inp);
711 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
712 			INP_RUNLOCK(inp);
713 			m_freem(m);
714 			return (IPPROTO_DONE);
715 		}
716 	}
717 
718 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
719 	if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
720 		INP_RUNLOCK(inp);
721 	return (IPPROTO_DONE);
722 
723 badunlocked:
724 	m_freem(m);
725 	return (IPPROTO_DONE);
726 }
727 #endif /* INET */
728 
729 /*
730  * Notify a udp user of an asynchronous error; just wake up so that they can
731  * collect error status.
732  */
733 struct inpcb *
734 udp_notify(struct inpcb *inp, int errno)
735 {
736 
737 	/*
738 	 * While udp_ctlinput() always calls udp_notify() with a read lock
739 	 * when invoking it directly, in_pcbnotifyall() currently uses write
740 	 * locks due to sharing code with TCP.  For now, accept either a read
741 	 * or a write lock, but a read lock is sufficient.
742 	 */
743 	INP_LOCK_ASSERT(inp);
744 
745 	inp->inp_socket->so_error = errno;
746 	sorwakeup(inp->inp_socket);
747 	sowwakeup(inp->inp_socket);
748 	return (inp);
749 }
750 
751 #ifdef INET
752 static void
753 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
754     struct inpcbinfo *pcbinfo)
755 {
756 	struct ip *ip = vip;
757 	struct udphdr *uh;
758 	struct in_addr faddr;
759 	struct inpcb *inp;
760 
761 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
762 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
763 		return;
764 
765 	/*
766 	 * Redirects don't need to be handled up here.
767 	 */
768 	if (PRC_IS_REDIRECT(cmd))
769 		return;
770 
771 	/*
772 	 * Hostdead is ugly because it goes linearly through all PCBs.
773 	 *
774 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
775 	 * DoS attack on machines with many connections.
776 	 */
777 	if (cmd == PRC_HOSTDEAD)
778 		ip = NULL;
779 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
780 		return;
781 	if (ip != NULL) {
782 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
783 		inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
784 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
785 		if (inp != NULL) {
786 			INP_RLOCK_ASSERT(inp);
787 			if (inp->inp_socket != NULL) {
788 				udp_notify(inp, inetctlerrmap[cmd]);
789 			}
790 			INP_RUNLOCK(inp);
791 		}
792 	} else
793 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
794 		    udp_notify);
795 }
796 void
797 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
798 {
799 
800 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
801 }
802 
803 void
804 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
805 {
806 
807 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
808 }
809 #endif /* INET */
810 
811 static int
812 udp_pcblist(SYSCTL_HANDLER_ARGS)
813 {
814 	int error, i, n;
815 	struct inpcb *inp, **inp_list;
816 	inp_gen_t gencnt;
817 	struct xinpgen xig;
818 
819 	/*
820 	 * The process of preparing the PCB list is too time-consuming and
821 	 * resource-intensive to repeat twice on every request.
822 	 */
823 	if (req->oldptr == 0) {
824 		n = V_udbinfo.ipi_count;
825 		n += imax(n / 8, 10);
826 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
827 		return (0);
828 	}
829 
830 	if (req->newptr != 0)
831 		return (EPERM);
832 
833 	/*
834 	 * OK, now we're committed to doing something.
835 	 */
836 	INP_INFO_RLOCK(&V_udbinfo);
837 	gencnt = V_udbinfo.ipi_gencnt;
838 	n = V_udbinfo.ipi_count;
839 	INP_INFO_RUNLOCK(&V_udbinfo);
840 
841 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
842 		+ n * sizeof(struct xinpcb));
843 	if (error != 0)
844 		return (error);
845 
846 	xig.xig_len = sizeof xig;
847 	xig.xig_count = n;
848 	xig.xig_gen = gencnt;
849 	xig.xig_sogen = so_gencnt;
850 	error = SYSCTL_OUT(req, &xig, sizeof xig);
851 	if (error)
852 		return (error);
853 
854 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
855 	if (inp_list == 0)
856 		return (ENOMEM);
857 
858 	INP_INFO_RLOCK(&V_udbinfo);
859 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
860 	     inp = LIST_NEXT(inp, inp_list)) {
861 		INP_WLOCK(inp);
862 		if (inp->inp_gencnt <= gencnt &&
863 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
864 			in_pcbref(inp);
865 			inp_list[i++] = inp;
866 		}
867 		INP_WUNLOCK(inp);
868 	}
869 	INP_INFO_RUNLOCK(&V_udbinfo);
870 	n = i;
871 
872 	error = 0;
873 	for (i = 0; i < n; i++) {
874 		inp = inp_list[i];
875 		INP_RLOCK(inp);
876 		if (inp->inp_gencnt <= gencnt) {
877 			struct xinpcb xi;
878 
879 			bzero(&xi, sizeof(xi));
880 			xi.xi_len = sizeof xi;
881 			/* XXX should avoid extra copy */
882 			bcopy(inp, &xi.xi_inp, sizeof *inp);
883 			if (inp->inp_socket)
884 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
885 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
886 			INP_RUNLOCK(inp);
887 			error = SYSCTL_OUT(req, &xi, sizeof xi);
888 		} else
889 			INP_RUNLOCK(inp);
890 	}
891 	INP_INFO_WLOCK(&V_udbinfo);
892 	for (i = 0; i < n; i++) {
893 		inp = inp_list[i];
894 		INP_RLOCK(inp);
895 		if (!in_pcbrele_rlocked(inp))
896 			INP_RUNLOCK(inp);
897 	}
898 	INP_INFO_WUNLOCK(&V_udbinfo);
899 
900 	if (!error) {
901 		/*
902 		 * Give the user an updated idea of our state.  If the
903 		 * generation differs from what we told her before, she knows
904 		 * that something happened while we were processing this
905 		 * request, and it might be necessary to retry.
906 		 */
907 		INP_INFO_RLOCK(&V_udbinfo);
908 		xig.xig_gen = V_udbinfo.ipi_gencnt;
909 		xig.xig_sogen = so_gencnt;
910 		xig.xig_count = V_udbinfo.ipi_count;
911 		INP_INFO_RUNLOCK(&V_udbinfo);
912 		error = SYSCTL_OUT(req, &xig, sizeof xig);
913 	}
914 	free(inp_list, M_TEMP);
915 	return (error);
916 }
917 
918 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
919     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
920     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
921 
922 #ifdef INET
923 static int
924 udp_getcred(SYSCTL_HANDLER_ARGS)
925 {
926 	struct xucred xuc;
927 	struct sockaddr_in addrs[2];
928 	struct inpcb *inp;
929 	int error;
930 
931 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
932 	if (error)
933 		return (error);
934 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
935 	if (error)
936 		return (error);
937 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
938 	    addrs[0].sin_addr, addrs[0].sin_port,
939 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
940 	if (inp != NULL) {
941 		INP_RLOCK_ASSERT(inp);
942 		if (inp->inp_socket == NULL)
943 			error = ENOENT;
944 		if (error == 0)
945 			error = cr_canseeinpcb(req->td->td_ucred, inp);
946 		if (error == 0)
947 			cru2x(inp->inp_cred, &xuc);
948 		INP_RUNLOCK(inp);
949 	} else
950 		error = ENOENT;
951 	if (error == 0)
952 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
953 	return (error);
954 }
955 
956 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
957     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
958     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
959 #endif /* INET */
960 
961 int
962 udp_ctloutput(struct socket *so, struct sockopt *sopt)
963 {
964 	struct inpcb *inp;
965 	struct udpcb *up;
966 	int isudplite, error, optval;
967 
968 	error = 0;
969 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
970 	inp = sotoinpcb(so);
971 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
972 	INP_WLOCK(inp);
973 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
974 #ifdef INET6
975 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
976 			INP_WUNLOCK(inp);
977 			error = ip6_ctloutput(so, sopt);
978 		}
979 #endif
980 #if defined(INET) && defined(INET6)
981 		else
982 #endif
983 #ifdef INET
984 		{
985 			INP_WUNLOCK(inp);
986 			error = ip_ctloutput(so, sopt);
987 		}
988 #endif
989 		return (error);
990 	}
991 
992 	switch (sopt->sopt_dir) {
993 	case SOPT_SET:
994 		switch (sopt->sopt_name) {
995 		case UDP_ENCAP:
996 			INP_WUNLOCK(inp);
997 			error = sooptcopyin(sopt, &optval, sizeof optval,
998 					    sizeof optval);
999 			if (error)
1000 				break;
1001 			inp = sotoinpcb(so);
1002 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1003 			INP_WLOCK(inp);
1004 #ifdef IPSEC_NAT_T
1005 			up = intoudpcb(inp);
1006 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1007 #endif
1008 			switch (optval) {
1009 			case 0:
1010 				/* Clear all UDP encap. */
1011 #ifdef IPSEC_NAT_T
1012 				up->u_flags &= ~UF_ESPINUDP_ALL;
1013 #endif
1014 				break;
1015 #ifdef IPSEC_NAT_T
1016 			case UDP_ENCAP_ESPINUDP:
1017 			case UDP_ENCAP_ESPINUDP_NON_IKE:
1018 				up->u_flags &= ~UF_ESPINUDP_ALL;
1019 				if (optval == UDP_ENCAP_ESPINUDP)
1020 					up->u_flags |= UF_ESPINUDP;
1021 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1022 					up->u_flags |= UF_ESPINUDP_NON_IKE;
1023 				break;
1024 #endif
1025 			default:
1026 				error = EINVAL;
1027 				break;
1028 			}
1029 			INP_WUNLOCK(inp);
1030 			break;
1031 		case UDPLITE_SEND_CSCOV:
1032 		case UDPLITE_RECV_CSCOV:
1033 			if (!isudplite) {
1034 				INP_WUNLOCK(inp);
1035 				error = ENOPROTOOPT;
1036 				break;
1037 			}
1038 			INP_WUNLOCK(inp);
1039 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1040 			    sizeof(optval));
1041 			if (error != 0)
1042 				break;
1043 			inp = sotoinpcb(so);
1044 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1045 			INP_WLOCK(inp);
1046 			up = intoudpcb(inp);
1047 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1048 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
1049 				INP_WUNLOCK(inp);
1050 				error = EINVAL;
1051 				break;
1052 			}
1053 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1054 				up->u_txcslen = optval;
1055 			else
1056 				up->u_rxcslen = optval;
1057 			INP_WUNLOCK(inp);
1058 			break;
1059 		default:
1060 			INP_WUNLOCK(inp);
1061 			error = ENOPROTOOPT;
1062 			break;
1063 		}
1064 		break;
1065 	case SOPT_GET:
1066 		switch (sopt->sopt_name) {
1067 #ifdef IPSEC_NAT_T
1068 		case UDP_ENCAP:
1069 			up = intoudpcb(inp);
1070 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1071 			optval = up->u_flags & UF_ESPINUDP_ALL;
1072 			INP_WUNLOCK(inp);
1073 			error = sooptcopyout(sopt, &optval, sizeof optval);
1074 			break;
1075 #endif
1076 		case UDPLITE_SEND_CSCOV:
1077 		case UDPLITE_RECV_CSCOV:
1078 			if (!isudplite) {
1079 				INP_WUNLOCK(inp);
1080 				error = ENOPROTOOPT;
1081 				break;
1082 			}
1083 			up = intoudpcb(inp);
1084 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1085 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1086 				optval = up->u_txcslen;
1087 			else
1088 				optval = up->u_rxcslen;
1089 			INP_WUNLOCK(inp);
1090 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1091 			break;
1092 		default:
1093 			INP_WUNLOCK(inp);
1094 			error = ENOPROTOOPT;
1095 			break;
1096 		}
1097 		break;
1098 	}
1099 	return (error);
1100 }
1101 
1102 #ifdef INET
1103 #define	UH_WLOCKED	2
1104 #define	UH_RLOCKED	1
1105 #define	UH_UNLOCKED	0
1106 static int
1107 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1108     struct mbuf *control, struct thread *td)
1109 {
1110 	struct udpiphdr *ui;
1111 	int len = m->m_pkthdr.len;
1112 	struct in_addr faddr, laddr;
1113 	struct cmsghdr *cm;
1114 	struct inpcbinfo *pcbinfo;
1115 	struct sockaddr_in *sin, src;
1116 	int cscov_partial = 0;
1117 	int error = 0;
1118 	int ipflags;
1119 	u_short fport, lport;
1120 	int unlock_udbinfo;
1121 	u_char tos;
1122 	uint8_t pr;
1123 	uint16_t cscov = 0;
1124 	uint32_t flowid = 0;
1125 	uint8_t flowtype = M_HASHTYPE_NONE;
1126 
1127 	/*
1128 	 * udp_output() may need to temporarily bind or connect the current
1129 	 * inpcb.  As such, we don't know up front whether we will need the
1130 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1131 	 * front before acquiring any locks.
1132 	 */
1133 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1134 		if (control)
1135 			m_freem(control);
1136 		m_freem(m);
1137 		return (EMSGSIZE);
1138 	}
1139 
1140 	src.sin_family = 0;
1141 	INP_RLOCK(inp);
1142 	tos = inp->inp_ip_tos;
1143 	if (control != NULL) {
1144 		/*
1145 		 * XXX: Currently, we assume all the optional information is
1146 		 * stored in a single mbuf.
1147 		 */
1148 		if (control->m_next) {
1149 			INP_RUNLOCK(inp);
1150 			m_freem(control);
1151 			m_freem(m);
1152 			return (EINVAL);
1153 		}
1154 		for (; control->m_len > 0;
1155 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1156 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1157 			cm = mtod(control, struct cmsghdr *);
1158 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1159 			    || cm->cmsg_len > control->m_len) {
1160 				error = EINVAL;
1161 				break;
1162 			}
1163 			if (cm->cmsg_level != IPPROTO_IP)
1164 				continue;
1165 
1166 			switch (cm->cmsg_type) {
1167 			case IP_SENDSRCADDR:
1168 				if (cm->cmsg_len !=
1169 				    CMSG_LEN(sizeof(struct in_addr))) {
1170 					error = EINVAL;
1171 					break;
1172 				}
1173 				bzero(&src, sizeof(src));
1174 				src.sin_family = AF_INET;
1175 				src.sin_len = sizeof(src);
1176 				src.sin_port = inp->inp_lport;
1177 				src.sin_addr =
1178 				    *(struct in_addr *)CMSG_DATA(cm);
1179 				break;
1180 
1181 			case IP_TOS:
1182 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1183 					error = EINVAL;
1184 					break;
1185 				}
1186 				tos = *(u_char *)CMSG_DATA(cm);
1187 				break;
1188 
1189 			case IP_FLOWID:
1190 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1191 					error = EINVAL;
1192 					break;
1193 				}
1194 				flowid = *(uint32_t *) CMSG_DATA(cm);
1195 				break;
1196 
1197 			case IP_FLOWTYPE:
1198 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1199 					error = EINVAL;
1200 					break;
1201 				}
1202 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1203 				break;
1204 
1205 #ifdef	RSS
1206 			case IP_RSSBUCKETID:
1207 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1208 					error = EINVAL;
1209 					break;
1210 				}
1211 				/* This is just a placeholder for now */
1212 				break;
1213 #endif	/* RSS */
1214 			default:
1215 				error = ENOPROTOOPT;
1216 				break;
1217 			}
1218 			if (error)
1219 				break;
1220 		}
1221 		m_freem(control);
1222 	}
1223 	if (error) {
1224 		INP_RUNLOCK(inp);
1225 		m_freem(m);
1226 		return (error);
1227 	}
1228 
1229 	/*
1230 	 * Depending on whether or not the application has bound or connected
1231 	 * the socket, we may have to do varying levels of work.  The optimal
1232 	 * case is for a connected UDP socket, as a global lock isn't
1233 	 * required at all.
1234 	 *
1235 	 * In order to decide which we need, we require stability of the
1236 	 * inpcb binding, which we ensure by acquiring a read lock on the
1237 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1238 	 * the trylock and retry game; note that we may end up with more
1239 	 * conservative locks than required the second time around, so later
1240 	 * assertions have to accept that.  Further analysis of the number of
1241 	 * misses under contention is required.
1242 	 *
1243 	 * XXXRW: Check that hash locking update here is correct.
1244 	 */
1245 	pr = inp->inp_socket->so_proto->pr_protocol;
1246 	pcbinfo = udp_get_inpcbinfo(pr);
1247 	sin = (struct sockaddr_in *)addr;
1248 	if (sin != NULL &&
1249 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1250 		INP_RUNLOCK(inp);
1251 		INP_WLOCK(inp);
1252 		INP_HASH_WLOCK(pcbinfo);
1253 		unlock_udbinfo = UH_WLOCKED;
1254 	} else if ((sin != NULL && (
1255 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1256 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1257 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1258 	    (inp->inp_lport == 0))) ||
1259 	    (src.sin_family == AF_INET)) {
1260 		INP_HASH_RLOCK(pcbinfo);
1261 		unlock_udbinfo = UH_RLOCKED;
1262 	} else
1263 		unlock_udbinfo = UH_UNLOCKED;
1264 
1265 	/*
1266 	 * If the IP_SENDSRCADDR control message was specified, override the
1267 	 * source address for this datagram.  Its use is invalidated if the
1268 	 * address thus specified is incomplete or clobbers other inpcbs.
1269 	 */
1270 	laddr = inp->inp_laddr;
1271 	lport = inp->inp_lport;
1272 	if (src.sin_family == AF_INET) {
1273 		INP_HASH_LOCK_ASSERT(pcbinfo);
1274 		if ((lport == 0) ||
1275 		    (laddr.s_addr == INADDR_ANY &&
1276 		     src.sin_addr.s_addr == INADDR_ANY)) {
1277 			error = EINVAL;
1278 			goto release;
1279 		}
1280 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1281 		    &laddr.s_addr, &lport, td->td_ucred);
1282 		if (error)
1283 			goto release;
1284 	}
1285 
1286 	/*
1287 	 * If a UDP socket has been connected, then a local address/port will
1288 	 * have been selected and bound.
1289 	 *
1290 	 * If a UDP socket has not been connected to, then an explicit
1291 	 * destination address must be used, in which case a local
1292 	 * address/port may not have been selected and bound.
1293 	 */
1294 	if (sin != NULL) {
1295 		INP_LOCK_ASSERT(inp);
1296 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1297 			error = EISCONN;
1298 			goto release;
1299 		}
1300 
1301 		/*
1302 		 * Jail may rewrite the destination address, so let it do
1303 		 * that before we use it.
1304 		 */
1305 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1306 		if (error)
1307 			goto release;
1308 
1309 		/*
1310 		 * If a local address or port hasn't yet been selected, or if
1311 		 * the destination address needs to be rewritten due to using
1312 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1313 		 * to do the heavy lifting.  Once a port is selected, we
1314 		 * commit the binding back to the socket; we also commit the
1315 		 * binding of the address if in jail.
1316 		 *
1317 		 * If we already have a valid binding and we're not
1318 		 * requesting a destination address rewrite, use a fast path.
1319 		 */
1320 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1321 		    inp->inp_lport == 0 ||
1322 		    sin->sin_addr.s_addr == INADDR_ANY ||
1323 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1324 			INP_HASH_LOCK_ASSERT(pcbinfo);
1325 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1326 			    &lport, &faddr.s_addr, &fport, NULL,
1327 			    td->td_ucred);
1328 			if (error)
1329 				goto release;
1330 
1331 			/*
1332 			 * XXXRW: Why not commit the port if the address is
1333 			 * !INADDR_ANY?
1334 			 */
1335 			/* Commit the local port if newly assigned. */
1336 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1337 			    inp->inp_lport == 0) {
1338 				INP_WLOCK_ASSERT(inp);
1339 				INP_HASH_WLOCK_ASSERT(pcbinfo);
1340 				/*
1341 				 * Remember addr if jailed, to prevent
1342 				 * rebinding.
1343 				 */
1344 				if (prison_flag(td->td_ucred, PR_IP4))
1345 					inp->inp_laddr = laddr;
1346 				inp->inp_lport = lport;
1347 				if (in_pcbinshash(inp) != 0) {
1348 					inp->inp_lport = 0;
1349 					error = EAGAIN;
1350 					goto release;
1351 				}
1352 				inp->inp_flags |= INP_ANONPORT;
1353 			}
1354 		} else {
1355 			faddr = sin->sin_addr;
1356 			fport = sin->sin_port;
1357 		}
1358 	} else {
1359 		INP_LOCK_ASSERT(inp);
1360 		faddr = inp->inp_faddr;
1361 		fport = inp->inp_fport;
1362 		if (faddr.s_addr == INADDR_ANY) {
1363 			error = ENOTCONN;
1364 			goto release;
1365 		}
1366 	}
1367 
1368 	/*
1369 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1370 	 * link-layer headers.  Immediate slide the data pointer back forward
1371 	 * since we won't use that space at this layer.
1372 	 */
1373 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1374 	if (m == NULL) {
1375 		error = ENOBUFS;
1376 		goto release;
1377 	}
1378 	m->m_data += max_linkhdr;
1379 	m->m_len -= max_linkhdr;
1380 	m->m_pkthdr.len -= max_linkhdr;
1381 
1382 	/*
1383 	 * Fill in mbuf with extended UDP header and addresses and length put
1384 	 * into network format.
1385 	 */
1386 	ui = mtod(m, struct udpiphdr *);
1387 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1388 	ui->ui_pr = pr;
1389 	ui->ui_src = laddr;
1390 	ui->ui_dst = faddr;
1391 	ui->ui_sport = lport;
1392 	ui->ui_dport = fport;
1393 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1394 	if (pr == IPPROTO_UDPLITE) {
1395 		struct udpcb *up;
1396 		uint16_t plen;
1397 
1398 		up = intoudpcb(inp);
1399 		cscov = up->u_txcslen;
1400 		plen = (u_short)len + sizeof(struct udphdr);
1401 		if (cscov >= plen)
1402 			cscov = 0;
1403 		ui->ui_len = htons(plen);
1404 		ui->ui_ulen = htons(cscov);
1405 		/*
1406 		 * For UDP-Lite, checksum coverage length of zero means
1407 		 * the entire UDPLite packet is covered by the checksum.
1408 		 */
1409 		cscov_partial = (cscov == 0) ? 0 : 1;
1410 	} else
1411 		ui->ui_v = IPVERSION << 4;
1412 
1413 	/*
1414 	 * Set the Don't Fragment bit in the IP header.
1415 	 */
1416 	if (inp->inp_flags & INP_DONTFRAG) {
1417 		struct ip *ip;
1418 
1419 		ip = (struct ip *)&ui->ui_i;
1420 		ip->ip_off |= htons(IP_DF);
1421 	}
1422 
1423 	ipflags = 0;
1424 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1425 		ipflags |= IP_ROUTETOIF;
1426 	if (inp->inp_socket->so_options & SO_BROADCAST)
1427 		ipflags |= IP_ALLOWBROADCAST;
1428 	if (inp->inp_flags & INP_ONESBCAST)
1429 		ipflags |= IP_SENDONES;
1430 
1431 #ifdef MAC
1432 	mac_inpcb_create_mbuf(inp, m);
1433 #endif
1434 
1435 	/*
1436 	 * Set up checksum and output datagram.
1437 	 */
1438 	ui->ui_sum = 0;
1439 	if (pr == IPPROTO_UDPLITE) {
1440 		if (inp->inp_flags & INP_ONESBCAST)
1441 			faddr.s_addr = INADDR_BROADCAST;
1442 		if (cscov_partial) {
1443 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1444 				ui->ui_sum = 0xffff;
1445 		} else {
1446 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1447 				ui->ui_sum = 0xffff;
1448 		}
1449 	} else if (V_udp_cksum) {
1450 		if (inp->inp_flags & INP_ONESBCAST)
1451 			faddr.s_addr = INADDR_BROADCAST;
1452 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1453 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1454 		m->m_pkthdr.csum_flags = CSUM_UDP;
1455 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1456 	}
1457 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1458 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1459 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1460 	UDPSTAT_INC(udps_opackets);
1461 
1462 	/*
1463 	 * Setup flowid / RSS information for outbound socket.
1464 	 *
1465 	 * Once the UDP code decides to set a flowid some other way,
1466 	 * this allows the flowid to be overridden by userland.
1467 	 */
1468 	if (flowtype != M_HASHTYPE_NONE) {
1469 		m->m_pkthdr.flowid = flowid;
1470 		M_HASHTYPE_SET(m, flowtype);
1471 #ifdef	RSS
1472 	} else {
1473 		uint32_t hash_val, hash_type;
1474 		/*
1475 		 * Calculate an appropriate RSS hash for UDP and
1476 		 * UDP Lite.
1477 		 *
1478 		 * The called function will take care of figuring out
1479 		 * whether a 2-tuple or 4-tuple hash is required based
1480 		 * on the currently configured scheme.
1481 		 *
1482 		 * Later later on connected socket values should be
1483 		 * cached in the inpcb and reused, rather than constantly
1484 		 * re-calculating it.
1485 		 *
1486 		 * UDP Lite is a different protocol number and will
1487 		 * likely end up being hashed as a 2-tuple until
1488 		 * RSS / NICs grow UDP Lite protocol awareness.
1489 		 */
1490 		if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1491 		    pr, &hash_val, &hash_type) == 0) {
1492 			m->m_pkthdr.flowid = hash_val;
1493 			M_HASHTYPE_SET(m, hash_type);
1494 		}
1495 #endif
1496 	}
1497 
1498 #ifdef	RSS
1499 	/*
1500 	 * Don't override with the inp cached flowid value.
1501 	 *
1502 	 * Depending upon the kind of send being done, the inp
1503 	 * flowid/flowtype values may actually not be appropriate
1504 	 * for this particular socket send.
1505 	 *
1506 	 * We should either leave the flowid at zero (which is what is
1507 	 * currently done) or set it to some software generated
1508 	 * hash value based on the packet contents.
1509 	 */
1510 	ipflags |= IP_NODEFAULTFLOWID;
1511 #endif	/* RSS */
1512 
1513 	if (unlock_udbinfo == UH_WLOCKED)
1514 		INP_HASH_WUNLOCK(pcbinfo);
1515 	else if (unlock_udbinfo == UH_RLOCKED)
1516 		INP_HASH_RUNLOCK(pcbinfo);
1517 	UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1518 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1519 	    inp->inp_moptions, inp);
1520 	if (unlock_udbinfo == UH_WLOCKED)
1521 		INP_WUNLOCK(inp);
1522 	else
1523 		INP_RUNLOCK(inp);
1524 	return (error);
1525 
1526 release:
1527 	if (unlock_udbinfo == UH_WLOCKED) {
1528 		INP_HASH_WUNLOCK(pcbinfo);
1529 		INP_WUNLOCK(inp);
1530 	} else if (unlock_udbinfo == UH_RLOCKED) {
1531 		INP_HASH_RUNLOCK(pcbinfo);
1532 		INP_RUNLOCK(inp);
1533 	} else
1534 		INP_RUNLOCK(inp);
1535 	m_freem(m);
1536 	return (error);
1537 }
1538 
1539 
1540 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1541 /*
1542  * Potentially decap ESP in UDP frame.  Check for an ESP header
1543  * and optional marker; if present, strip the UDP header and
1544  * push the result through IPSec.
1545  *
1546  * Returns mbuf to be processed (potentially re-allocated) or
1547  * NULL if consumed and/or processed.
1548  */
1549 static struct mbuf *
1550 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1551 {
1552 	size_t minlen, payload, skip, iphlen;
1553 	caddr_t data;
1554 	struct udpcb *up;
1555 	struct m_tag *tag;
1556 	struct udphdr *udphdr;
1557 	struct ip *ip;
1558 
1559 	INP_RLOCK_ASSERT(inp);
1560 
1561 	/*
1562 	 * Pull up data so the longest case is contiguous:
1563 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1564 	 */
1565 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1566 	if (minlen > m->m_pkthdr.len)
1567 		minlen = m->m_pkthdr.len;
1568 	if ((m = m_pullup(m, minlen)) == NULL) {
1569 		IPSECSTAT_INC(ips_in_inval);
1570 		return (NULL);		/* Bypass caller processing. */
1571 	}
1572 	data = mtod(m, caddr_t);	/* Points to ip header. */
1573 	payload = m->m_len - off;	/* Size of payload. */
1574 
1575 	if (payload == 1 && data[off] == '\xff')
1576 		return (m);		/* NB: keepalive packet, no decap. */
1577 
1578 	up = intoudpcb(inp);
1579 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1580 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1581 	    ("u_flags 0x%x", up->u_flags));
1582 
1583 	/*
1584 	 * Check that the payload is large enough to hold an
1585 	 * ESP header and compute the amount of data to remove.
1586 	 *
1587 	 * NB: the caller has already done a pullup for us.
1588 	 * XXX can we assume alignment and eliminate bcopys?
1589 	 */
1590 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1591 		/*
1592 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1593 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1594 		 * possible AH mode non-IKE marker+non-ESP marker
1595 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1596 		 */
1597 		uint64_t marker;
1598 
1599 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1600 			return (m);	/* NB: no decap. */
1601 		bcopy(data + off, &marker, sizeof(uint64_t));
1602 		if (marker != 0)	/* Non-IKE marker. */
1603 			return (m);	/* NB: no decap. */
1604 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1605 	} else {
1606 		uint32_t spi;
1607 
1608 		if (payload <= sizeof(struct esp)) {
1609 			IPSECSTAT_INC(ips_in_inval);
1610 			m_freem(m);
1611 			return (NULL);	/* Discard. */
1612 		}
1613 		bcopy(data + off, &spi, sizeof(uint32_t));
1614 		if (spi == 0)		/* Non-ESP marker. */
1615 			return (m);	/* NB: no decap. */
1616 		skip = sizeof(struct udphdr);
1617 	}
1618 
1619 	/*
1620 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1621 	 * the UDP ports. This is required if we want to select
1622 	 * the right SPD for multiple hosts behind same NAT.
1623 	 *
1624 	 * NB: ports are maintained in network byte order everywhere
1625 	 *     in the NAT-T code.
1626 	 */
1627 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1628 		2 * sizeof(uint16_t), M_NOWAIT);
1629 	if (tag == NULL) {
1630 		IPSECSTAT_INC(ips_in_nomem);
1631 		m_freem(m);
1632 		return (NULL);		/* Discard. */
1633 	}
1634 	iphlen = off - sizeof(struct udphdr);
1635 	udphdr = (struct udphdr *)(data + iphlen);
1636 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1637 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1638 	m_tag_prepend(m, tag);
1639 
1640 	/*
1641 	 * Remove the UDP header (and possibly the non ESP marker)
1642 	 * IP header length is iphlen
1643 	 * Before:
1644 	 *   <--- off --->
1645 	 *   +----+------+-----+
1646 	 *   | IP |  UDP | ESP |
1647 	 *   +----+------+-----+
1648 	 *        <-skip->
1649 	 * After:
1650 	 *          +----+-----+
1651 	 *          | IP | ESP |
1652 	 *          +----+-----+
1653 	 *   <-skip->
1654 	 */
1655 	ovbcopy(data, data + skip, iphlen);
1656 	m_adj(m, skip);
1657 
1658 	ip = mtod(m, struct ip *);
1659 	ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1660 	ip->ip_p = IPPROTO_ESP;
1661 
1662 	/*
1663 	 * We cannot yet update the cksums so clear any
1664 	 * h/w cksum flags as they are no longer valid.
1665 	 */
1666 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1667 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1668 
1669 	(void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1670 				AF_INET, ip->ip_p);
1671 	return (NULL);			/* NB: consumed, bypass processing. */
1672 }
1673 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1674 
1675 static void
1676 udp_abort(struct socket *so)
1677 {
1678 	struct inpcb *inp;
1679 	struct inpcbinfo *pcbinfo;
1680 
1681 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1682 	inp = sotoinpcb(so);
1683 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1684 	INP_WLOCK(inp);
1685 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1686 		INP_HASH_WLOCK(pcbinfo);
1687 		in_pcbdisconnect(inp);
1688 		inp->inp_laddr.s_addr = INADDR_ANY;
1689 		INP_HASH_WUNLOCK(pcbinfo);
1690 		soisdisconnected(so);
1691 	}
1692 	INP_WUNLOCK(inp);
1693 }
1694 
1695 static int
1696 udp_attach(struct socket *so, int proto, struct thread *td)
1697 {
1698 	struct inpcb *inp;
1699 	struct inpcbinfo *pcbinfo;
1700 	int error;
1701 
1702 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1703 	inp = sotoinpcb(so);
1704 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1705 	error = soreserve(so, udp_sendspace, udp_recvspace);
1706 	if (error)
1707 		return (error);
1708 	INP_INFO_WLOCK(pcbinfo);
1709 	error = in_pcballoc(so, pcbinfo);
1710 	if (error) {
1711 		INP_INFO_WUNLOCK(pcbinfo);
1712 		return (error);
1713 	}
1714 
1715 	inp = sotoinpcb(so);
1716 	inp->inp_vflag |= INP_IPV4;
1717 	inp->inp_ip_ttl = V_ip_defttl;
1718 
1719 	error = udp_newudpcb(inp);
1720 	if (error) {
1721 		in_pcbdetach(inp);
1722 		in_pcbfree(inp);
1723 		INP_INFO_WUNLOCK(pcbinfo);
1724 		return (error);
1725 	}
1726 
1727 	INP_WUNLOCK(inp);
1728 	INP_INFO_WUNLOCK(pcbinfo);
1729 	return (0);
1730 }
1731 #endif /* INET */
1732 
1733 int
1734 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, void *ctx)
1735 {
1736 	struct inpcb *inp;
1737 	struct udpcb *up;
1738 
1739 	KASSERT(so->so_type == SOCK_DGRAM,
1740 	    ("udp_set_kernel_tunneling: !dgram"));
1741 	inp = sotoinpcb(so);
1742 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1743 	INP_WLOCK(inp);
1744 	up = intoudpcb(inp);
1745 	if (up->u_tun_func != NULL) {
1746 		INP_WUNLOCK(inp);
1747 		return (EBUSY);
1748 	}
1749 	up->u_tun_func = f;
1750 	up->u_tun_ctx = ctx;
1751 	INP_WUNLOCK(inp);
1752 	return (0);
1753 }
1754 
1755 #ifdef INET
1756 static int
1757 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1758 {
1759 	struct inpcb *inp;
1760 	struct inpcbinfo *pcbinfo;
1761 	int error;
1762 
1763 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1764 	inp = sotoinpcb(so);
1765 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1766 	INP_WLOCK(inp);
1767 	INP_HASH_WLOCK(pcbinfo);
1768 	error = in_pcbbind(inp, nam, td->td_ucred);
1769 	INP_HASH_WUNLOCK(pcbinfo);
1770 	INP_WUNLOCK(inp);
1771 	return (error);
1772 }
1773 
1774 static void
1775 udp_close(struct socket *so)
1776 {
1777 	struct inpcb *inp;
1778 	struct inpcbinfo *pcbinfo;
1779 
1780 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1781 	inp = sotoinpcb(so);
1782 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1783 	INP_WLOCK(inp);
1784 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1785 		INP_HASH_WLOCK(pcbinfo);
1786 		in_pcbdisconnect(inp);
1787 		inp->inp_laddr.s_addr = INADDR_ANY;
1788 		INP_HASH_WUNLOCK(pcbinfo);
1789 		soisdisconnected(so);
1790 	}
1791 	INP_WUNLOCK(inp);
1792 }
1793 
1794 static int
1795 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1796 {
1797 	struct inpcb *inp;
1798 	struct inpcbinfo *pcbinfo;
1799 	struct sockaddr_in *sin;
1800 	int error;
1801 
1802 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1803 	inp = sotoinpcb(so);
1804 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1805 	INP_WLOCK(inp);
1806 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1807 		INP_WUNLOCK(inp);
1808 		return (EISCONN);
1809 	}
1810 	sin = (struct sockaddr_in *)nam;
1811 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1812 	if (error != 0) {
1813 		INP_WUNLOCK(inp);
1814 		return (error);
1815 	}
1816 	INP_HASH_WLOCK(pcbinfo);
1817 	error = in_pcbconnect(inp, nam, td->td_ucred);
1818 	INP_HASH_WUNLOCK(pcbinfo);
1819 	if (error == 0)
1820 		soisconnected(so);
1821 	INP_WUNLOCK(inp);
1822 	return (error);
1823 }
1824 
1825 static void
1826 udp_detach(struct socket *so)
1827 {
1828 	struct inpcb *inp;
1829 	struct inpcbinfo *pcbinfo;
1830 	struct udpcb *up;
1831 
1832 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1833 	inp = sotoinpcb(so);
1834 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1835 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1836 	    ("udp_detach: not disconnected"));
1837 	INP_INFO_WLOCK(pcbinfo);
1838 	INP_WLOCK(inp);
1839 	up = intoudpcb(inp);
1840 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1841 	inp->inp_ppcb = NULL;
1842 	in_pcbdetach(inp);
1843 	in_pcbfree(inp);
1844 	INP_INFO_WUNLOCK(pcbinfo);
1845 	udp_discardcb(up);
1846 }
1847 
1848 static int
1849 udp_disconnect(struct socket *so)
1850 {
1851 	struct inpcb *inp;
1852 	struct inpcbinfo *pcbinfo;
1853 
1854 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1855 	inp = sotoinpcb(so);
1856 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1857 	INP_WLOCK(inp);
1858 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1859 		INP_WUNLOCK(inp);
1860 		return (ENOTCONN);
1861 	}
1862 	INP_HASH_WLOCK(pcbinfo);
1863 	in_pcbdisconnect(inp);
1864 	inp->inp_laddr.s_addr = INADDR_ANY;
1865 	INP_HASH_WUNLOCK(pcbinfo);
1866 	SOCK_LOCK(so);
1867 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1868 	SOCK_UNLOCK(so);
1869 	INP_WUNLOCK(inp);
1870 	return (0);
1871 }
1872 
1873 static int
1874 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1875     struct mbuf *control, struct thread *td)
1876 {
1877 	struct inpcb *inp;
1878 
1879 	inp = sotoinpcb(so);
1880 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1881 	return (udp_output(inp, m, addr, control, td));
1882 }
1883 #endif /* INET */
1884 
1885 int
1886 udp_shutdown(struct socket *so)
1887 {
1888 	struct inpcb *inp;
1889 
1890 	inp = sotoinpcb(so);
1891 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1892 	INP_WLOCK(inp);
1893 	socantsendmore(so);
1894 	INP_WUNLOCK(inp);
1895 	return (0);
1896 }
1897 
1898 #ifdef INET
1899 struct pr_usrreqs udp_usrreqs = {
1900 	.pru_abort =		udp_abort,
1901 	.pru_attach =		udp_attach,
1902 	.pru_bind =		udp_bind,
1903 	.pru_connect =		udp_connect,
1904 	.pru_control =		in_control,
1905 	.pru_detach =		udp_detach,
1906 	.pru_disconnect =	udp_disconnect,
1907 	.pru_peeraddr =		in_getpeeraddr,
1908 	.pru_send =		udp_send,
1909 	.pru_soreceive =	soreceive_dgram,
1910 	.pru_sosend =		sosend_dgram,
1911 	.pru_shutdown =		udp_shutdown,
1912 	.pru_sockaddr =		in_getsockaddr,
1913 	.pru_sosetlabel =	in_pcbsosetlabel,
1914 	.pru_close =		udp_close,
1915 };
1916 #endif /* INET */
1917