xref: /freebsd/sys/netinet/udp_usrreq.c (revision f0574f5cf69e168cc4ea71ebbe5fdec9ec9a3dfe)
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  * 3. 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_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 #include "opt_rss.h"
46 
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/sdt.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
66 
67 #include <vm/uma.h>
68 
69 #include <net/if.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 #include <net/rss_config.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/in_systm.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip.h>
80 #ifdef INET6
81 #include <netinet/ip6.h>
82 #endif
83 #include <netinet/ip_icmp.h>
84 #include <netinet/icmp_var.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/ip_options.h>
87 #ifdef INET6
88 #include <netinet6/ip6_var.h>
89 #endif
90 #include <netinet/udp.h>
91 #include <netinet/udp_var.h>
92 #include <netinet/udplite.h>
93 #include <netinet/in_rss.h>
94 
95 #include <netipsec/ipsec_support.h>
96 
97 #include <machine/in_cksum.h>
98 
99 #include <security/mac/mac_framework.h>
100 
101 /*
102  * UDP and UDP-Lite protocols implementation.
103  * Per RFC 768, August, 1980.
104  * Per RFC 3828, July, 2004.
105  */
106 
107 /*
108  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
109  * removes the only data integrity mechanism for packets and malformed
110  * packets that would otherwise be discarded due to bad checksums, and may
111  * cause problems (especially for NFS data blocks).
112  */
113 VNET_DEFINE(int, udp_cksum) = 1;
114 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
115     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
116 
117 int	udp_log_in_vain = 0;
118 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
119     &udp_log_in_vain, 0, "Log all incoming UDP packets");
120 
121 VNET_DEFINE(int, udp_blackhole) = 0;
122 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
123     &VNET_NAME(udp_blackhole), 0,
124     "Do not send port unreachables for refused connects");
125 
126 static VNET_DEFINE(int, udp_require_l2_bcast) = 0;
127 #define	V_udp_require_l2_bcast		VNET(udp_require_l2_bcast)
128 SYSCTL_INT(_net_inet_udp, OID_AUTO, require_l2_bcast, CTLFLAG_VNET | CTLFLAG_RW,
129     &VNET_NAME(udp_require_l2_bcast), 0,
130     "Only treat packets sent to an L2 broadcast address as broadcast packets");
131 
132 u_long	udp_sendspace = 9216;		/* really max datagram size */
133 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
134     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
135 
136 u_long	udp_recvspace = 40 * (1024 +
137 #ifdef INET6
138 				      sizeof(struct sockaddr_in6)
139 #else
140 				      sizeof(struct sockaddr_in)
141 #endif
142 				      );	/* 40 1K datagrams */
143 
144 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
145     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
146 
147 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
148 VNET_DEFINE(struct inpcbinfo, udbinfo);
149 VNET_DEFINE(struct inpcbhead, ulitecb);
150 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
151 static VNET_DEFINE(uma_zone_t, udpcb_zone);
152 #define	V_udpcb_zone			VNET(udpcb_zone)
153 
154 #ifndef UDBHASHSIZE
155 #define	UDBHASHSIZE	128
156 #endif
157 
158 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
159 VNET_PCPUSTAT_SYSINIT(udpstat);
160 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
161     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
162 
163 #ifdef VIMAGE
164 VNET_PCPUSTAT_SYSUNINIT(udpstat);
165 #endif /* VIMAGE */
166 #ifdef INET
167 static void	udp_detach(struct socket *so);
168 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
169 		    struct mbuf *, struct thread *);
170 #endif
171 
172 static void
173 udp_zone_change(void *tag)
174 {
175 
176 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
177 	uma_zone_set_max(V_udpcb_zone, maxsockets);
178 }
179 
180 static int
181 udp_inpcb_init(void *mem, int size, int flags)
182 {
183 	struct inpcb *inp;
184 
185 	inp = mem;
186 	INP_LOCK_INIT(inp, "inp", "udpinp");
187 	return (0);
188 }
189 
190 static int
191 udplite_inpcb_init(void *mem, int size, int flags)
192 {
193 	struct inpcb *inp;
194 
195 	inp = mem;
196 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
197 	return (0);
198 }
199 
200 void
201 udp_init(void)
202 {
203 
204 	/*
205 	 * For now default to 2-tuple UDP hashing - until the fragment
206 	 * reassembly code can also update the flowid.
207 	 *
208 	 * Once we can calculate the flowid that way and re-establish
209 	 * a 4-tuple, flip this to 4-tuple.
210 	 */
211 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
212 	    "udp_inpcb", udp_inpcb_init, NULL, 0,
213 	    IPI_HASHFIELDS_2TUPLE);
214 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
215 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
216 	uma_zone_set_max(V_udpcb_zone, maxsockets);
217 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
218 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
219 	    EVENTHANDLER_PRI_ANY);
220 }
221 
222 void
223 udplite_init(void)
224 {
225 
226 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
227 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
228 	    0, IPI_HASHFIELDS_2TUPLE);
229 }
230 
231 /*
232  * Kernel module interface for updating udpstat.  The argument is an index
233  * into udpstat treated as an array of u_long.  While this encodes the
234  * general layout of udpstat into the caller, it doesn't encode its location,
235  * so that future changes to add, for example, per-CPU stats support won't
236  * cause binary compatibility problems for kernel modules.
237  */
238 void
239 kmod_udpstat_inc(int statnum)
240 {
241 
242 	counter_u64_add(VNET(udpstat)[statnum], 1);
243 }
244 
245 int
246 udp_newudpcb(struct inpcb *inp)
247 {
248 	struct udpcb *up;
249 
250 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
251 	if (up == NULL)
252 		return (ENOBUFS);
253 	inp->inp_ppcb = up;
254 	return (0);
255 }
256 
257 void
258 udp_discardcb(struct udpcb *up)
259 {
260 
261 	uma_zfree(V_udpcb_zone, up);
262 }
263 
264 #ifdef VIMAGE
265 static void
266 udp_destroy(void *unused __unused)
267 {
268 
269 	in_pcbinfo_destroy(&V_udbinfo);
270 	uma_zdestroy(V_udpcb_zone);
271 }
272 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
273 
274 static void
275 udplite_destroy(void *unused __unused)
276 {
277 
278 	in_pcbinfo_destroy(&V_ulitecbinfo);
279 }
280 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
281     NULL);
282 #endif
283 
284 #ifdef INET
285 /*
286  * Subroutine of udp_input(), which appends the provided mbuf chain to the
287  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
288  * contains the source address.  If the socket ends up being an IPv6 socket,
289  * udp_append() will convert to a sockaddr_in6 before passing the address
290  * into the socket code.
291  *
292  * In the normal case udp_append() will return 0, indicating that you
293  * must unlock the inp. However if a tunneling protocol is in place we increment
294  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
295  * then decrement the reference count. If the inp_rele returns 1, indicating the
296  * inp is gone, we return that to the caller to tell them *not* to unlock
297  * the inp. In the case of multi-cast this will cause the distribution
298  * to stop (though most tunneling protocols known currently do *not* use
299  * multicast).
300  */
301 static int
302 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
303     struct sockaddr_in *udp_in)
304 {
305 	struct sockaddr *append_sa;
306 	struct socket *so;
307 	struct mbuf *tmpopts, *opts = NULL;
308 #ifdef INET6
309 	struct sockaddr_in6 udp_in6;
310 #endif
311 	struct udpcb *up;
312 
313 	INP_LOCK_ASSERT(inp);
314 
315 	/*
316 	 * Engage the tunneling protocol.
317 	 */
318 	up = intoudpcb(inp);
319 	if (up->u_tun_func != NULL) {
320 		in_pcbref(inp);
321 		INP_RUNLOCK(inp);
322 		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
323 		    up->u_tun_ctx);
324 		INP_RLOCK(inp);
325 		return (in_pcbrele_rlocked(inp));
326 	}
327 
328 	off += sizeof(struct udphdr);
329 
330 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
331 	/* Check AH/ESP integrity. */
332 	if (IPSEC_ENABLED(ipv4) &&
333 	    IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
334 		m_freem(n);
335 		return (0);
336 	}
337 	if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
338 		if (IPSEC_ENABLED(ipv4) &&
339 		    UDPENCAP_INPUT(n, off, AF_INET) != 0)
340 			return (0);	/* Consumed. */
341 	}
342 #endif /* IPSEC */
343 #ifdef MAC
344 	if (mac_inpcb_check_deliver(inp, n) != 0) {
345 		m_freem(n);
346 		return (0);
347 	}
348 #endif /* MAC */
349 	if (inp->inp_flags & INP_CONTROLOPTS ||
350 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
351 #ifdef INET6
352 		if (inp->inp_vflag & INP_IPV6)
353 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
354 		else
355 #endif /* INET6 */
356 			ip_savecontrol(inp, &opts, ip, n);
357 	}
358 	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
359 		tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
360 			sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
361 		if (tmpopts) {
362 			if (opts) {
363 				tmpopts->m_next = opts;
364 				opts = tmpopts;
365 			} else
366 				opts = tmpopts;
367 		}
368 	}
369 #ifdef INET6
370 	if (inp->inp_vflag & INP_IPV6) {
371 		bzero(&udp_in6, sizeof(udp_in6));
372 		udp_in6.sin6_len = sizeof(udp_in6);
373 		udp_in6.sin6_family = AF_INET6;
374 		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
375 		append_sa = (struct sockaddr *)&udp_in6;
376 	} else
377 #endif /* INET6 */
378 		append_sa = (struct sockaddr *)&udp_in[0];
379 	m_adj(n, off);
380 
381 	so = inp->inp_socket;
382 	SOCKBUF_LOCK(&so->so_rcv);
383 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
384 		SOCKBUF_UNLOCK(&so->so_rcv);
385 		m_freem(n);
386 		if (opts)
387 			m_freem(opts);
388 		UDPSTAT_INC(udps_fullsock);
389 	} else
390 		sorwakeup_locked(so);
391 	return (0);
392 }
393 
394 int
395 udp_input(struct mbuf **mp, int *offp, int proto)
396 {
397 	struct ip *ip;
398 	struct udphdr *uh;
399 	struct ifnet *ifp;
400 	struct inpcb *inp;
401 	uint16_t len, ip_len;
402 	struct inpcbinfo *pcbinfo;
403 	struct ip save_ip;
404 	struct sockaddr_in udp_in[2];
405 	struct mbuf *m;
406 	struct m_tag *fwd_tag;
407 	int cscov_partial, iphlen;
408 
409 	m = *mp;
410 	iphlen = *offp;
411 	ifp = m->m_pkthdr.rcvif;
412 	*mp = NULL;
413 	UDPSTAT_INC(udps_ipackets);
414 
415 	/*
416 	 * Strip IP options, if any; should skip this, make available to
417 	 * user, and use on returned packets, but we don't yet have a way to
418 	 * check the checksum with options still present.
419 	 */
420 	if (iphlen > sizeof (struct ip)) {
421 		ip_stripoptions(m);
422 		iphlen = sizeof(struct ip);
423 	}
424 
425 	/*
426 	 * Get IP and UDP header together in first mbuf.
427 	 */
428 	ip = mtod(m, struct ip *);
429 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
430 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
431 			UDPSTAT_INC(udps_hdrops);
432 			return (IPPROTO_DONE);
433 		}
434 		ip = mtod(m, struct ip *);
435 	}
436 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
437 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
438 
439 	/*
440 	 * Destination port of 0 is illegal, based on RFC768.
441 	 */
442 	if (uh->uh_dport == 0)
443 		goto badunlocked;
444 
445 	/*
446 	 * Construct sockaddr format source address.  Stuff source address
447 	 * and datagram in user buffer.
448 	 */
449 	bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
450 	udp_in[0].sin_len = sizeof(struct sockaddr_in);
451 	udp_in[0].sin_family = AF_INET;
452 	udp_in[0].sin_port = uh->uh_sport;
453 	udp_in[0].sin_addr = ip->ip_src;
454 	udp_in[1].sin_len = sizeof(struct sockaddr_in);
455 	udp_in[1].sin_family = AF_INET;
456 	udp_in[1].sin_port = uh->uh_dport;
457 	udp_in[1].sin_addr = ip->ip_dst;
458 
459 	/*
460 	 * Make mbuf data length reflect UDP length.  If not enough data to
461 	 * reflect UDP length, drop.
462 	 */
463 	len = ntohs((u_short)uh->uh_ulen);
464 	ip_len = ntohs(ip->ip_len) - iphlen;
465 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
466 		/* Zero means checksum over the complete packet. */
467 		if (len == 0)
468 			len = ip_len;
469 		cscov_partial = 0;
470 	}
471 	if (ip_len != len) {
472 		if (len > ip_len || len < sizeof(struct udphdr)) {
473 			UDPSTAT_INC(udps_badlen);
474 			goto badunlocked;
475 		}
476 		if (proto == IPPROTO_UDP)
477 			m_adj(m, len - ip_len);
478 	}
479 
480 	/*
481 	 * Save a copy of the IP header in case we want restore it for
482 	 * sending an ICMP error message in response.
483 	 */
484 	if (!V_udp_blackhole)
485 		save_ip = *ip;
486 	else
487 		memset(&save_ip, 0, sizeof(save_ip));
488 
489 	/*
490 	 * Checksum extended UDP header and data.
491 	 */
492 	if (uh->uh_sum) {
493 		u_short uh_sum;
494 
495 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
496 		    !cscov_partial) {
497 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
498 				uh_sum = m->m_pkthdr.csum_data;
499 			else
500 				uh_sum = in_pseudo(ip->ip_src.s_addr,
501 				    ip->ip_dst.s_addr, htonl((u_short)len +
502 				    m->m_pkthdr.csum_data + proto));
503 			uh_sum ^= 0xffff;
504 		} else {
505 			char b[9];
506 
507 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
508 			bzero(((struct ipovly *)ip)->ih_x1, 9);
509 			((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
510 			    uh->uh_ulen : htons(ip_len);
511 			uh_sum = in_cksum(m, len + sizeof (struct ip));
512 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
513 		}
514 		if (uh_sum) {
515 			UDPSTAT_INC(udps_badsum);
516 			m_freem(m);
517 			return (IPPROTO_DONE);
518 		}
519 	} else {
520 		if (proto == IPPROTO_UDP) {
521 			UDPSTAT_INC(udps_nosum);
522 		} else {
523 			/* UDPLite requires a checksum */
524 			/* XXX: What is the right UDPLite MIB counter here? */
525 			m_freem(m);
526 			return (IPPROTO_DONE);
527 		}
528 	}
529 
530 	pcbinfo = udp_get_inpcbinfo(proto);
531 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
532 	    ((!V_udp_require_l2_bcast || m->m_flags & M_BCAST) &&
533 	    in_broadcast(ip->ip_dst, ifp))) {
534 		struct inpcb *last;
535 		struct inpcbhead *pcblist;
536 		struct ip_moptions *imo;
537 
538 		INP_INFO_RLOCK(pcbinfo);
539 		pcblist = udp_get_pcblist(proto);
540 		last = NULL;
541 		LIST_FOREACH(inp, pcblist, inp_list) {
542 			if (inp->inp_lport != uh->uh_dport)
543 				continue;
544 #ifdef INET6
545 			if ((inp->inp_vflag & INP_IPV4) == 0)
546 				continue;
547 #endif
548 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
549 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
550 				continue;
551 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
552 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
553 				continue;
554 			if (inp->inp_fport != 0 &&
555 			    inp->inp_fport != uh->uh_sport)
556 				continue;
557 
558 			INP_RLOCK(inp);
559 
560 			/*
561 			 * XXXRW: Because we weren't holding either the inpcb
562 			 * or the hash lock when we checked for a match
563 			 * before, we should probably recheck now that the
564 			 * inpcb lock is held.
565 			 */
566 
567 			/*
568 			 * Handle socket delivery policy for any-source
569 			 * and source-specific multicast. [RFC3678]
570 			 */
571 			imo = inp->inp_moptions;
572 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
573 				struct sockaddr_in	 group;
574 				int			 blocked;
575 				if (imo == NULL) {
576 					INP_RUNLOCK(inp);
577 					continue;
578 				}
579 				bzero(&group, sizeof(struct sockaddr_in));
580 				group.sin_len = sizeof(struct sockaddr_in);
581 				group.sin_family = AF_INET;
582 				group.sin_addr = ip->ip_dst;
583 
584 				blocked = imo_multi_filter(imo, ifp,
585 					(struct sockaddr *)&group,
586 					(struct sockaddr *)&udp_in[0]);
587 				if (blocked != MCAST_PASS) {
588 					if (blocked == MCAST_NOTGMEMBER)
589 						IPSTAT_INC(ips_notmember);
590 					if (blocked == MCAST_NOTSMEMBER ||
591 					    blocked == MCAST_MUTED)
592 						UDPSTAT_INC(udps_filtermcast);
593 					INP_RUNLOCK(inp);
594 					continue;
595 				}
596 			}
597 			if (last != NULL) {
598 				struct mbuf *n;
599 
600 				if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
601 				    NULL) {
602 					UDP_PROBE(receive, NULL, last, ip,
603 					    last, uh);
604 					if (udp_append(last, ip, n, iphlen,
605 						udp_in)) {
606 						goto inp_lost;
607 					}
608 				}
609 				INP_RUNLOCK(last);
610 			}
611 			last = inp;
612 			/*
613 			 * Don't look for additional matches if this one does
614 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
615 			 * socket options set.  This heuristic avoids
616 			 * searching through all pcbs in the common case of a
617 			 * non-shared port.  It assumes that an application
618 			 * will never clear these options after setting them.
619 			 */
620 			if ((last->inp_socket->so_options &
621 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
622 				break;
623 		}
624 
625 		if (last == NULL) {
626 			/*
627 			 * No matching pcb found; discard datagram.  (No need
628 			 * to send an ICMP Port Unreachable for a broadcast
629 			 * or multicast datgram.)
630 			 */
631 			UDPSTAT_INC(udps_noportbcast);
632 			if (inp)
633 				INP_RUNLOCK(inp);
634 			INP_INFO_RUNLOCK(pcbinfo);
635 			goto badunlocked;
636 		}
637 		UDP_PROBE(receive, NULL, last, ip, last, uh);
638 		if (udp_append(last, ip, m, iphlen, udp_in) == 0)
639 			INP_RUNLOCK(last);
640 	inp_lost:
641 		INP_INFO_RUNLOCK(pcbinfo);
642 		return (IPPROTO_DONE);
643 	}
644 
645 	/*
646 	 * Locate pcb for datagram.
647 	 */
648 
649 	/*
650 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
651 	 */
652 	if ((m->m_flags & M_IP_NEXTHOP) &&
653 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
654 		struct sockaddr_in *next_hop;
655 
656 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
657 
658 		/*
659 		 * Transparently forwarded. Pretend to be the destination.
660 		 * Already got one like this?
661 		 */
662 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
663 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
664 		if (!inp) {
665 			/*
666 			 * It's new.  Try to find the ambushing socket.
667 			 * Because we've rewritten the destination address,
668 			 * any hardware-generated hash is ignored.
669 			 */
670 			inp = in_pcblookup(pcbinfo, ip->ip_src,
671 			    uh->uh_sport, next_hop->sin_addr,
672 			    next_hop->sin_port ? htons(next_hop->sin_port) :
673 			    uh->uh_dport, INPLOOKUP_WILDCARD |
674 			    INPLOOKUP_RLOCKPCB, ifp);
675 		}
676 		/* Remove the tag from the packet. We don't need it anymore. */
677 		m_tag_delete(m, fwd_tag);
678 		m->m_flags &= ~M_IP_NEXTHOP;
679 	} else
680 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
681 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
682 		    INPLOOKUP_RLOCKPCB, ifp, m);
683 	if (inp == NULL) {
684 		if (udp_log_in_vain) {
685 			char src[INET_ADDRSTRLEN];
686 			char dst[INET_ADDRSTRLEN];
687 
688 			log(LOG_INFO,
689 			    "Connection attempt to UDP %s:%d from %s:%d\n",
690 			    inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
691 			    inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
692 		}
693 		UDPSTAT_INC(udps_noport);
694 		if (m->m_flags & (M_BCAST | M_MCAST)) {
695 			UDPSTAT_INC(udps_noportbcast);
696 			goto badunlocked;
697 		}
698 		if (V_udp_blackhole)
699 			goto badunlocked;
700 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
701 			goto badunlocked;
702 		*ip = save_ip;
703 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
704 		return (IPPROTO_DONE);
705 	}
706 
707 	/*
708 	 * Check the minimum TTL for socket.
709 	 */
710 	INP_RLOCK_ASSERT(inp);
711 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
712 		INP_RUNLOCK(inp);
713 		m_freem(m);
714 		return (IPPROTO_DONE);
715 	}
716 	if (cscov_partial) {
717 		struct udpcb *up;
718 
719 		up = intoudpcb(inp);
720 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
721 			INP_RUNLOCK(inp);
722 			m_freem(m);
723 			return (IPPROTO_DONE);
724 		}
725 	}
726 
727 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
728 	if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
729 		INP_RUNLOCK(inp);
730 	return (IPPROTO_DONE);
731 
732 badunlocked:
733 	m_freem(m);
734 	return (IPPROTO_DONE);
735 }
736 #endif /* INET */
737 
738 /*
739  * Notify a udp user of an asynchronous error; just wake up so that they can
740  * collect error status.
741  */
742 struct inpcb *
743 udp_notify(struct inpcb *inp, int errno)
744 {
745 
746 	/*
747 	 * While udp_ctlinput() always calls udp_notify() with a read lock
748 	 * when invoking it directly, in_pcbnotifyall() currently uses write
749 	 * locks due to sharing code with TCP.  For now, accept either a read
750 	 * or a write lock, but a read lock is sufficient.
751 	 */
752 	INP_LOCK_ASSERT(inp);
753 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
754 	     errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
755 		RTFREE(inp->inp_route.ro_rt);
756 		inp->inp_route.ro_rt = (struct rtentry *)NULL;
757 	}
758 
759 	inp->inp_socket->so_error = errno;
760 	sorwakeup(inp->inp_socket);
761 	sowwakeup(inp->inp_socket);
762 	return (inp);
763 }
764 
765 #ifdef INET
766 static void
767 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
768     struct inpcbinfo *pcbinfo)
769 {
770 	struct ip *ip = vip;
771 	struct udphdr *uh;
772 	struct in_addr faddr;
773 	struct inpcb *inp;
774 
775 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
776 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
777 		return;
778 
779 	if (PRC_IS_REDIRECT(cmd)) {
780 		/* signal EHOSTDOWN, as it flushes the cached route */
781 		in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
782 		return;
783 	}
784 
785 	/*
786 	 * Hostdead is ugly because it goes linearly through all PCBs.
787 	 *
788 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
789 	 * DoS attack on machines with many connections.
790 	 */
791 	if (cmd == PRC_HOSTDEAD)
792 		ip = NULL;
793 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
794 		return;
795 	if (ip != NULL) {
796 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
797 		inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
798 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
799 		if (inp != NULL) {
800 			INP_RLOCK_ASSERT(inp);
801 			if (inp->inp_socket != NULL) {
802 				udp_notify(inp, inetctlerrmap[cmd]);
803 			}
804 			INP_RUNLOCK(inp);
805 		} else {
806 			inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
807 					   ip->ip_src, uh->uh_sport,
808 					   INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
809 			if (inp != NULL) {
810 				struct udpcb *up;
811 
812 				up = intoudpcb(inp);
813 				if (up->u_icmp_func != NULL) {
814 					INP_RUNLOCK(inp);
815 					(*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
816 				} else {
817 					INP_RUNLOCK(inp);
818 				}
819 			}
820 		}
821 	} else
822 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
823 		    udp_notify);
824 }
825 void
826 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
827 {
828 
829 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
830 }
831 
832 void
833 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
834 {
835 
836 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
837 }
838 #endif /* INET */
839 
840 static int
841 udp_pcblist(SYSCTL_HANDLER_ARGS)
842 {
843 	int error, i, n;
844 	struct inpcb *inp, **inp_list;
845 	inp_gen_t gencnt;
846 	struct xinpgen xig;
847 
848 	/*
849 	 * The process of preparing the PCB list is too time-consuming and
850 	 * resource-intensive to repeat twice on every request.
851 	 */
852 	if (req->oldptr == 0) {
853 		n = V_udbinfo.ipi_count;
854 		n += imax(n / 8, 10);
855 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
856 		return (0);
857 	}
858 
859 	if (req->newptr != 0)
860 		return (EPERM);
861 
862 	/*
863 	 * OK, now we're committed to doing something.
864 	 */
865 	INP_INFO_RLOCK(&V_udbinfo);
866 	gencnt = V_udbinfo.ipi_gencnt;
867 	n = V_udbinfo.ipi_count;
868 	INP_INFO_RUNLOCK(&V_udbinfo);
869 
870 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
871 		+ n * sizeof(struct xinpcb));
872 	if (error != 0)
873 		return (error);
874 
875 	xig.xig_len = sizeof xig;
876 	xig.xig_count = n;
877 	xig.xig_gen = gencnt;
878 	xig.xig_sogen = so_gencnt;
879 	error = SYSCTL_OUT(req, &xig, sizeof xig);
880 	if (error)
881 		return (error);
882 
883 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
884 	if (inp_list == NULL)
885 		return (ENOMEM);
886 
887 	INP_INFO_RLOCK(&V_udbinfo);
888 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
889 	     inp = LIST_NEXT(inp, inp_list)) {
890 		INP_WLOCK(inp);
891 		if (inp->inp_gencnt <= gencnt &&
892 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
893 			in_pcbref(inp);
894 			inp_list[i++] = inp;
895 		}
896 		INP_WUNLOCK(inp);
897 	}
898 	INP_INFO_RUNLOCK(&V_udbinfo);
899 	n = i;
900 
901 	error = 0;
902 	for (i = 0; i < n; i++) {
903 		inp = inp_list[i];
904 		INP_RLOCK(inp);
905 		if (inp->inp_gencnt <= gencnt) {
906 			struct xinpcb xi;
907 
908 			in_pcbtoxinpcb(inp, &xi);
909 			INP_RUNLOCK(inp);
910 			error = SYSCTL_OUT(req, &xi, sizeof xi);
911 		} else
912 			INP_RUNLOCK(inp);
913 	}
914 	INP_INFO_WLOCK(&V_udbinfo);
915 	for (i = 0; i < n; i++) {
916 		inp = inp_list[i];
917 		INP_RLOCK(inp);
918 		if (!in_pcbrele_rlocked(inp))
919 			INP_RUNLOCK(inp);
920 	}
921 	INP_INFO_WUNLOCK(&V_udbinfo);
922 
923 	if (!error) {
924 		/*
925 		 * Give the user an updated idea of our state.  If the
926 		 * generation differs from what we told her before, she knows
927 		 * that something happened while we were processing this
928 		 * request, and it might be necessary to retry.
929 		 */
930 		INP_INFO_RLOCK(&V_udbinfo);
931 		xig.xig_gen = V_udbinfo.ipi_gencnt;
932 		xig.xig_sogen = so_gencnt;
933 		xig.xig_count = V_udbinfo.ipi_count;
934 		INP_INFO_RUNLOCK(&V_udbinfo);
935 		error = SYSCTL_OUT(req, &xig, sizeof xig);
936 	}
937 	free(inp_list, M_TEMP);
938 	return (error);
939 }
940 
941 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
942     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
943     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
944 
945 #ifdef INET
946 static int
947 udp_getcred(SYSCTL_HANDLER_ARGS)
948 {
949 	struct xucred xuc;
950 	struct sockaddr_in addrs[2];
951 	struct inpcb *inp;
952 	int error;
953 
954 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
955 	if (error)
956 		return (error);
957 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
958 	if (error)
959 		return (error);
960 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
961 	    addrs[0].sin_addr, addrs[0].sin_port,
962 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
963 	if (inp != NULL) {
964 		INP_RLOCK_ASSERT(inp);
965 		if (inp->inp_socket == NULL)
966 			error = ENOENT;
967 		if (error == 0)
968 			error = cr_canseeinpcb(req->td->td_ucred, inp);
969 		if (error == 0)
970 			cru2x(inp->inp_cred, &xuc);
971 		INP_RUNLOCK(inp);
972 	} else
973 		error = ENOENT;
974 	if (error == 0)
975 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
976 	return (error);
977 }
978 
979 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
980     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
981     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
982 #endif /* INET */
983 
984 int
985 udp_ctloutput(struct socket *so, struct sockopt *sopt)
986 {
987 	struct inpcb *inp;
988 	struct udpcb *up;
989 	int isudplite, error, optval;
990 
991 	error = 0;
992 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
993 	inp = sotoinpcb(so);
994 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
995 	INP_WLOCK(inp);
996 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
997 #ifdef INET6
998 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
999 			INP_WUNLOCK(inp);
1000 			error = ip6_ctloutput(so, sopt);
1001 		}
1002 #endif
1003 #if defined(INET) && defined(INET6)
1004 		else
1005 #endif
1006 #ifdef INET
1007 		{
1008 			INP_WUNLOCK(inp);
1009 			error = ip_ctloutput(so, sopt);
1010 		}
1011 #endif
1012 		return (error);
1013 	}
1014 
1015 	switch (sopt->sopt_dir) {
1016 	case SOPT_SET:
1017 		switch (sopt->sopt_name) {
1018 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1019 #ifdef INET
1020 		case UDP_ENCAP:
1021 			if (!IPSEC_ENABLED(ipv4)) {
1022 				INP_WUNLOCK(inp);
1023 				return (ENOPROTOOPT);
1024 			}
1025 			error = UDPENCAP_PCBCTL(inp, sopt);
1026 			break;
1027 #endif /* INET */
1028 #endif /* IPSEC */
1029 		case UDPLITE_SEND_CSCOV:
1030 		case UDPLITE_RECV_CSCOV:
1031 			if (!isudplite) {
1032 				INP_WUNLOCK(inp);
1033 				error = ENOPROTOOPT;
1034 				break;
1035 			}
1036 			INP_WUNLOCK(inp);
1037 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1038 			    sizeof(optval));
1039 			if (error != 0)
1040 				break;
1041 			inp = sotoinpcb(so);
1042 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1043 			INP_WLOCK(inp);
1044 			up = intoudpcb(inp);
1045 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1046 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
1047 				INP_WUNLOCK(inp);
1048 				error = EINVAL;
1049 				break;
1050 			}
1051 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1052 				up->u_txcslen = optval;
1053 			else
1054 				up->u_rxcslen = optval;
1055 			INP_WUNLOCK(inp);
1056 			break;
1057 		default:
1058 			INP_WUNLOCK(inp);
1059 			error = ENOPROTOOPT;
1060 			break;
1061 		}
1062 		break;
1063 	case SOPT_GET:
1064 		switch (sopt->sopt_name) {
1065 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1066 #ifdef INET
1067 		case UDP_ENCAP:
1068 			if (!IPSEC_ENABLED(ipv4)) {
1069 				INP_WUNLOCK(inp);
1070 				return (ENOPROTOOPT);
1071 			}
1072 			error = UDPENCAP_PCBCTL(inp, sopt);
1073 			break;
1074 #endif /* INET */
1075 #endif /* IPSEC */
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, unlock_inp;
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 	sin = (struct sockaddr_in *)addr;
1142 	if (sin == NULL ||
1143 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1144 		INP_WLOCK(inp);
1145 		unlock_inp = UH_WLOCKED;
1146 	} else {
1147 		INP_RLOCK(inp);
1148 		unlock_inp = UH_RLOCKED;
1149 	}
1150 	tos = inp->inp_ip_tos;
1151 	if (control != NULL) {
1152 		/*
1153 		 * XXX: Currently, we assume all the optional information is
1154 		 * stored in a single mbuf.
1155 		 */
1156 		if (control->m_next) {
1157 			if (unlock_inp == UH_WLOCKED)
1158 				INP_WUNLOCK(inp);
1159 			else
1160 				INP_RUNLOCK(inp);
1161 			m_freem(control);
1162 			m_freem(m);
1163 			return (EINVAL);
1164 		}
1165 		for (; control->m_len > 0;
1166 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1167 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1168 			cm = mtod(control, struct cmsghdr *);
1169 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1170 			    || cm->cmsg_len > control->m_len) {
1171 				error = EINVAL;
1172 				break;
1173 			}
1174 			if (cm->cmsg_level != IPPROTO_IP)
1175 				continue;
1176 
1177 			switch (cm->cmsg_type) {
1178 			case IP_SENDSRCADDR:
1179 				if (cm->cmsg_len !=
1180 				    CMSG_LEN(sizeof(struct in_addr))) {
1181 					error = EINVAL;
1182 					break;
1183 				}
1184 				bzero(&src, sizeof(src));
1185 				src.sin_family = AF_INET;
1186 				src.sin_len = sizeof(src);
1187 				src.sin_port = inp->inp_lport;
1188 				src.sin_addr =
1189 				    *(struct in_addr *)CMSG_DATA(cm);
1190 				break;
1191 
1192 			case IP_TOS:
1193 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1194 					error = EINVAL;
1195 					break;
1196 				}
1197 				tos = *(u_char *)CMSG_DATA(cm);
1198 				break;
1199 
1200 			case IP_FLOWID:
1201 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1202 					error = EINVAL;
1203 					break;
1204 				}
1205 				flowid = *(uint32_t *) CMSG_DATA(cm);
1206 				break;
1207 
1208 			case IP_FLOWTYPE:
1209 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1210 					error = EINVAL;
1211 					break;
1212 				}
1213 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1214 				break;
1215 
1216 #ifdef	RSS
1217 			case IP_RSSBUCKETID:
1218 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1219 					error = EINVAL;
1220 					break;
1221 				}
1222 				/* This is just a placeholder for now */
1223 				break;
1224 #endif	/* RSS */
1225 			default:
1226 				error = ENOPROTOOPT;
1227 				break;
1228 			}
1229 			if (error)
1230 				break;
1231 		}
1232 		m_freem(control);
1233 	}
1234 	if (error) {
1235 		if (unlock_inp == UH_WLOCKED)
1236 			INP_WUNLOCK(inp);
1237 		else
1238 			INP_RUNLOCK(inp);
1239 		m_freem(m);
1240 		return (error);
1241 	}
1242 
1243 	/*
1244 	 * Depending on whether or not the application has bound or connected
1245 	 * the socket, we may have to do varying levels of work.  The optimal
1246 	 * case is for a connected UDP socket, as a global lock isn't
1247 	 * required at all.
1248 	 *
1249 	 * In order to decide which we need, we require stability of the
1250 	 * inpcb binding, which we ensure by acquiring a read lock on the
1251 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1252 	 * the trylock and retry game; note that we may end up with more
1253 	 * conservative locks than required the second time around, so later
1254 	 * assertions have to accept that.  Further analysis of the number of
1255 	 * misses under contention is required.
1256 	 *
1257 	 * XXXRW: Check that hash locking update here is correct.
1258 	 */
1259 	pr = inp->inp_socket->so_proto->pr_protocol;
1260 	pcbinfo = udp_get_inpcbinfo(pr);
1261 	sin = (struct sockaddr_in *)addr;
1262 	if (sin != NULL &&
1263 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1264 		INP_HASH_WLOCK(pcbinfo);
1265 		unlock_udbinfo = UH_WLOCKED;
1266 	} else if ((sin != NULL && (
1267 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1268 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1269 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1270 	    (inp->inp_lport == 0))) ||
1271 	    (src.sin_family == AF_INET)) {
1272 		INP_HASH_RLOCK(pcbinfo);
1273 		unlock_udbinfo = UH_RLOCKED;
1274 	} else
1275 		unlock_udbinfo = UH_UNLOCKED;
1276 
1277 	/*
1278 	 * If the IP_SENDSRCADDR control message was specified, override the
1279 	 * source address for this datagram.  Its use is invalidated if the
1280 	 * address thus specified is incomplete or clobbers other inpcbs.
1281 	 */
1282 	laddr = inp->inp_laddr;
1283 	lport = inp->inp_lport;
1284 	if (src.sin_family == AF_INET) {
1285 		INP_HASH_LOCK_ASSERT(pcbinfo);
1286 		if ((lport == 0) ||
1287 		    (laddr.s_addr == INADDR_ANY &&
1288 		     src.sin_addr.s_addr == INADDR_ANY)) {
1289 			error = EINVAL;
1290 			goto release;
1291 		}
1292 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1293 		    &laddr.s_addr, &lport, td->td_ucred);
1294 		if (error)
1295 			goto release;
1296 	}
1297 
1298 	/*
1299 	 * If a UDP socket has been connected, then a local address/port will
1300 	 * have been selected and bound.
1301 	 *
1302 	 * If a UDP socket has not been connected to, then an explicit
1303 	 * destination address must be used, in which case a local
1304 	 * address/port may not have been selected and bound.
1305 	 */
1306 	if (sin != NULL) {
1307 		INP_LOCK_ASSERT(inp);
1308 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1309 			error = EISCONN;
1310 			goto release;
1311 		}
1312 
1313 		/*
1314 		 * Jail may rewrite the destination address, so let it do
1315 		 * that before we use it.
1316 		 */
1317 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1318 		if (error)
1319 			goto release;
1320 
1321 		/*
1322 		 * If a local address or port hasn't yet been selected, or if
1323 		 * the destination address needs to be rewritten due to using
1324 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1325 		 * to do the heavy lifting.  Once a port is selected, we
1326 		 * commit the binding back to the socket; we also commit the
1327 		 * binding of the address if in jail.
1328 		 *
1329 		 * If we already have a valid binding and we're not
1330 		 * requesting a destination address rewrite, use a fast path.
1331 		 */
1332 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1333 		    inp->inp_lport == 0 ||
1334 		    sin->sin_addr.s_addr == INADDR_ANY ||
1335 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1336 			INP_HASH_LOCK_ASSERT(pcbinfo);
1337 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1338 			    &lport, &faddr.s_addr, &fport, NULL,
1339 			    td->td_ucred);
1340 			if (error)
1341 				goto release;
1342 
1343 			/*
1344 			 * XXXRW: Why not commit the port if the address is
1345 			 * !INADDR_ANY?
1346 			 */
1347 			/* Commit the local port if newly assigned. */
1348 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1349 			    inp->inp_lport == 0) {
1350 				INP_WLOCK_ASSERT(inp);
1351 				INP_HASH_WLOCK_ASSERT(pcbinfo);
1352 				/*
1353 				 * Remember addr if jailed, to prevent
1354 				 * rebinding.
1355 				 */
1356 				if (prison_flag(td->td_ucred, PR_IP4))
1357 					inp->inp_laddr = laddr;
1358 				inp->inp_lport = lport;
1359 				if (in_pcbinshash(inp) != 0) {
1360 					inp->inp_lport = 0;
1361 					error = EAGAIN;
1362 					goto release;
1363 				}
1364 				inp->inp_flags |= INP_ANONPORT;
1365 			}
1366 		} else {
1367 			faddr = sin->sin_addr;
1368 			fport = sin->sin_port;
1369 		}
1370 	} else {
1371 		INP_LOCK_ASSERT(inp);
1372 		faddr = inp->inp_faddr;
1373 		fport = inp->inp_fport;
1374 		if (faddr.s_addr == INADDR_ANY) {
1375 			error = ENOTCONN;
1376 			goto release;
1377 		}
1378 	}
1379 
1380 	/*
1381 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1382 	 * link-layer headers.  Immediate slide the data pointer back forward
1383 	 * since we won't use that space at this layer.
1384 	 */
1385 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1386 	if (m == NULL) {
1387 		error = ENOBUFS;
1388 		goto release;
1389 	}
1390 	m->m_data += max_linkhdr;
1391 	m->m_len -= max_linkhdr;
1392 	m->m_pkthdr.len -= max_linkhdr;
1393 
1394 	/*
1395 	 * Fill in mbuf with extended UDP header and addresses and length put
1396 	 * into network format.
1397 	 */
1398 	ui = mtod(m, struct udpiphdr *);
1399 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1400 	ui->ui_pr = pr;
1401 	ui->ui_src = laddr;
1402 	ui->ui_dst = faddr;
1403 	ui->ui_sport = lport;
1404 	ui->ui_dport = fport;
1405 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1406 	if (pr == IPPROTO_UDPLITE) {
1407 		struct udpcb *up;
1408 		uint16_t plen;
1409 
1410 		up = intoudpcb(inp);
1411 		cscov = up->u_txcslen;
1412 		plen = (u_short)len + sizeof(struct udphdr);
1413 		if (cscov >= plen)
1414 			cscov = 0;
1415 		ui->ui_len = htons(plen);
1416 		ui->ui_ulen = htons(cscov);
1417 		/*
1418 		 * For UDP-Lite, checksum coverage length of zero means
1419 		 * the entire UDPLite packet is covered by the checksum.
1420 		 */
1421 		cscov_partial = (cscov == 0) ? 0 : 1;
1422 	} else
1423 		ui->ui_v = IPVERSION << 4;
1424 
1425 	/*
1426 	 * Set the Don't Fragment bit in the IP header.
1427 	 */
1428 	if (inp->inp_flags & INP_DONTFRAG) {
1429 		struct ip *ip;
1430 
1431 		ip = (struct ip *)&ui->ui_i;
1432 		ip->ip_off |= htons(IP_DF);
1433 	}
1434 
1435 	ipflags = 0;
1436 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1437 		ipflags |= IP_ROUTETOIF;
1438 	if (inp->inp_socket->so_options & SO_BROADCAST)
1439 		ipflags |= IP_ALLOWBROADCAST;
1440 	if (inp->inp_flags & INP_ONESBCAST)
1441 		ipflags |= IP_SENDONES;
1442 
1443 #ifdef MAC
1444 	mac_inpcb_create_mbuf(inp, m);
1445 #endif
1446 
1447 	/*
1448 	 * Set up checksum and output datagram.
1449 	 */
1450 	ui->ui_sum = 0;
1451 	if (pr == IPPROTO_UDPLITE) {
1452 		if (inp->inp_flags & INP_ONESBCAST)
1453 			faddr.s_addr = INADDR_BROADCAST;
1454 		if (cscov_partial) {
1455 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1456 				ui->ui_sum = 0xffff;
1457 		} else {
1458 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1459 				ui->ui_sum = 0xffff;
1460 		}
1461 	} else if (V_udp_cksum) {
1462 		if (inp->inp_flags & INP_ONESBCAST)
1463 			faddr.s_addr = INADDR_BROADCAST;
1464 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1465 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1466 		m->m_pkthdr.csum_flags = CSUM_UDP;
1467 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1468 	}
1469 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1470 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1471 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1472 	UDPSTAT_INC(udps_opackets);
1473 
1474 	/*
1475 	 * Setup flowid / RSS information for outbound socket.
1476 	 *
1477 	 * Once the UDP code decides to set a flowid some other way,
1478 	 * this allows the flowid to be overridden by userland.
1479 	 */
1480 	if (flowtype != M_HASHTYPE_NONE) {
1481 		m->m_pkthdr.flowid = flowid;
1482 		M_HASHTYPE_SET(m, flowtype);
1483 #ifdef	RSS
1484 	} else {
1485 		uint32_t hash_val, hash_type;
1486 		/*
1487 		 * Calculate an appropriate RSS hash for UDP and
1488 		 * UDP Lite.
1489 		 *
1490 		 * The called function will take care of figuring out
1491 		 * whether a 2-tuple or 4-tuple hash is required based
1492 		 * on the currently configured scheme.
1493 		 *
1494 		 * Later later on connected socket values should be
1495 		 * cached in the inpcb and reused, rather than constantly
1496 		 * re-calculating it.
1497 		 *
1498 		 * UDP Lite is a different protocol number and will
1499 		 * likely end up being hashed as a 2-tuple until
1500 		 * RSS / NICs grow UDP Lite protocol awareness.
1501 		 */
1502 		if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1503 		    pr, &hash_val, &hash_type) == 0) {
1504 			m->m_pkthdr.flowid = hash_val;
1505 			M_HASHTYPE_SET(m, hash_type);
1506 		}
1507 #endif
1508 	}
1509 
1510 #ifdef	RSS
1511 	/*
1512 	 * Don't override with the inp cached flowid value.
1513 	 *
1514 	 * Depending upon the kind of send being done, the inp
1515 	 * flowid/flowtype values may actually not be appropriate
1516 	 * for this particular socket send.
1517 	 *
1518 	 * We should either leave the flowid at zero (which is what is
1519 	 * currently done) or set it to some software generated
1520 	 * hash value based on the packet contents.
1521 	 */
1522 	ipflags |= IP_NODEFAULTFLOWID;
1523 #endif	/* RSS */
1524 
1525 	if (unlock_udbinfo == UH_WLOCKED)
1526 		INP_HASH_WUNLOCK(pcbinfo);
1527 	else if (unlock_udbinfo == UH_RLOCKED)
1528 		INP_HASH_RUNLOCK(pcbinfo);
1529 	UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1530 	error = ip_output(m, inp->inp_options,
1531 	    (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1532 	    inp->inp_moptions, inp);
1533 	if (unlock_inp == UH_WLOCKED)
1534 		INP_WUNLOCK(inp);
1535 	else
1536 		INP_RUNLOCK(inp);
1537 	return (error);
1538 
1539 release:
1540 	if (unlock_udbinfo == UH_WLOCKED) {
1541 		KASSERT(unlock_inp == UH_WLOCKED,
1542 		    ("%s: excl udbinfo lock, shared inp lock", __func__));
1543 		INP_HASH_WUNLOCK(pcbinfo);
1544 		INP_WUNLOCK(inp);
1545 	} else if (unlock_udbinfo == UH_RLOCKED) {
1546 		KASSERT(unlock_inp == UH_RLOCKED,
1547 		    ("%s: shared udbinfo lock, excl inp lock", __func__));
1548 		INP_HASH_RUNLOCK(pcbinfo);
1549 		INP_RUNLOCK(inp);
1550 	} else if (unlock_inp == UH_WLOCKED)
1551 		INP_WUNLOCK(inp);
1552 	else
1553 		INP_RUNLOCK(inp);
1554 	m_freem(m);
1555 	return (error);
1556 }
1557 
1558 static void
1559 udp_abort(struct socket *so)
1560 {
1561 	struct inpcb *inp;
1562 	struct inpcbinfo *pcbinfo;
1563 
1564 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1565 	inp = sotoinpcb(so);
1566 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1567 	INP_WLOCK(inp);
1568 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1569 		INP_HASH_WLOCK(pcbinfo);
1570 		in_pcbdisconnect(inp);
1571 		inp->inp_laddr.s_addr = INADDR_ANY;
1572 		INP_HASH_WUNLOCK(pcbinfo);
1573 		soisdisconnected(so);
1574 	}
1575 	INP_WUNLOCK(inp);
1576 }
1577 
1578 static int
1579 udp_attach(struct socket *so, int proto, struct thread *td)
1580 {
1581 	struct inpcb *inp;
1582 	struct inpcbinfo *pcbinfo;
1583 	int error;
1584 
1585 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1586 	inp = sotoinpcb(so);
1587 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1588 	error = soreserve(so, udp_sendspace, udp_recvspace);
1589 	if (error)
1590 		return (error);
1591 	INP_INFO_WLOCK(pcbinfo);
1592 	error = in_pcballoc(so, pcbinfo);
1593 	if (error) {
1594 		INP_INFO_WUNLOCK(pcbinfo);
1595 		return (error);
1596 	}
1597 
1598 	inp = sotoinpcb(so);
1599 	inp->inp_vflag |= INP_IPV4;
1600 	inp->inp_ip_ttl = V_ip_defttl;
1601 
1602 	error = udp_newudpcb(inp);
1603 	if (error) {
1604 		in_pcbdetach(inp);
1605 		in_pcbfree(inp);
1606 		INP_INFO_WUNLOCK(pcbinfo);
1607 		return (error);
1608 	}
1609 
1610 	INP_WUNLOCK(inp);
1611 	INP_INFO_WUNLOCK(pcbinfo);
1612 	return (0);
1613 }
1614 #endif /* INET */
1615 
1616 int
1617 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1618 {
1619 	struct inpcb *inp;
1620 	struct udpcb *up;
1621 
1622 	KASSERT(so->so_type == SOCK_DGRAM,
1623 	    ("udp_set_kernel_tunneling: !dgram"));
1624 	inp = sotoinpcb(so);
1625 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1626 	INP_WLOCK(inp);
1627 	up = intoudpcb(inp);
1628 	if ((up->u_tun_func != NULL) ||
1629 	    (up->u_icmp_func != NULL)) {
1630 		INP_WUNLOCK(inp);
1631 		return (EBUSY);
1632 	}
1633 	up->u_tun_func = f;
1634 	up->u_icmp_func = i;
1635 	up->u_tun_ctx = ctx;
1636 	INP_WUNLOCK(inp);
1637 	return (0);
1638 }
1639 
1640 #ifdef INET
1641 static int
1642 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1643 {
1644 	struct inpcb *inp;
1645 	struct inpcbinfo *pcbinfo;
1646 	int error;
1647 
1648 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1649 	inp = sotoinpcb(so);
1650 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1651 	INP_WLOCK(inp);
1652 	INP_HASH_WLOCK(pcbinfo);
1653 	error = in_pcbbind(inp, nam, td->td_ucred);
1654 	INP_HASH_WUNLOCK(pcbinfo);
1655 	INP_WUNLOCK(inp);
1656 	return (error);
1657 }
1658 
1659 static void
1660 udp_close(struct socket *so)
1661 {
1662 	struct inpcb *inp;
1663 	struct inpcbinfo *pcbinfo;
1664 
1665 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1666 	inp = sotoinpcb(so);
1667 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1668 	INP_WLOCK(inp);
1669 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1670 		INP_HASH_WLOCK(pcbinfo);
1671 		in_pcbdisconnect(inp);
1672 		inp->inp_laddr.s_addr = INADDR_ANY;
1673 		INP_HASH_WUNLOCK(pcbinfo);
1674 		soisdisconnected(so);
1675 	}
1676 	INP_WUNLOCK(inp);
1677 }
1678 
1679 static int
1680 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1681 {
1682 	struct inpcb *inp;
1683 	struct inpcbinfo *pcbinfo;
1684 	struct sockaddr_in *sin;
1685 	int error;
1686 
1687 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1688 	inp = sotoinpcb(so);
1689 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1690 	INP_WLOCK(inp);
1691 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1692 		INP_WUNLOCK(inp);
1693 		return (EISCONN);
1694 	}
1695 	sin = (struct sockaddr_in *)nam;
1696 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1697 	if (error != 0) {
1698 		INP_WUNLOCK(inp);
1699 		return (error);
1700 	}
1701 	INP_HASH_WLOCK(pcbinfo);
1702 	error = in_pcbconnect(inp, nam, td->td_ucred);
1703 	INP_HASH_WUNLOCK(pcbinfo);
1704 	if (error == 0)
1705 		soisconnected(so);
1706 	INP_WUNLOCK(inp);
1707 	return (error);
1708 }
1709 
1710 static void
1711 udp_detach(struct socket *so)
1712 {
1713 	struct inpcb *inp;
1714 	struct inpcbinfo *pcbinfo;
1715 	struct udpcb *up;
1716 
1717 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1718 	inp = sotoinpcb(so);
1719 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1720 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1721 	    ("udp_detach: not disconnected"));
1722 	INP_INFO_WLOCK(pcbinfo);
1723 	INP_WLOCK(inp);
1724 	up = intoudpcb(inp);
1725 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1726 	inp->inp_ppcb = NULL;
1727 	in_pcbdetach(inp);
1728 	in_pcbfree(inp);
1729 	INP_INFO_WUNLOCK(pcbinfo);
1730 	udp_discardcb(up);
1731 }
1732 
1733 static int
1734 udp_disconnect(struct socket *so)
1735 {
1736 	struct inpcb *inp;
1737 	struct inpcbinfo *pcbinfo;
1738 
1739 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1740 	inp = sotoinpcb(so);
1741 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1742 	INP_WLOCK(inp);
1743 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1744 		INP_WUNLOCK(inp);
1745 		return (ENOTCONN);
1746 	}
1747 	INP_HASH_WLOCK(pcbinfo);
1748 	in_pcbdisconnect(inp);
1749 	inp->inp_laddr.s_addr = INADDR_ANY;
1750 	INP_HASH_WUNLOCK(pcbinfo);
1751 	SOCK_LOCK(so);
1752 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1753 	SOCK_UNLOCK(so);
1754 	INP_WUNLOCK(inp);
1755 	return (0);
1756 }
1757 
1758 static int
1759 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1760     struct mbuf *control, struct thread *td)
1761 {
1762 	struct inpcb *inp;
1763 
1764 	inp = sotoinpcb(so);
1765 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1766 	return (udp_output(inp, m, addr, control, td));
1767 }
1768 #endif /* INET */
1769 
1770 int
1771 udp_shutdown(struct socket *so)
1772 {
1773 	struct inpcb *inp;
1774 
1775 	inp = sotoinpcb(so);
1776 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1777 	INP_WLOCK(inp);
1778 	socantsendmore(so);
1779 	INP_WUNLOCK(inp);
1780 	return (0);
1781 }
1782 
1783 #ifdef INET
1784 struct pr_usrreqs udp_usrreqs = {
1785 	.pru_abort =		udp_abort,
1786 	.pru_attach =		udp_attach,
1787 	.pru_bind =		udp_bind,
1788 	.pru_connect =		udp_connect,
1789 	.pru_control =		in_control,
1790 	.pru_detach =		udp_detach,
1791 	.pru_disconnect =	udp_disconnect,
1792 	.pru_peeraddr =		in_getpeeraddr,
1793 	.pru_send =		udp_send,
1794 	.pru_soreceive =	soreceive_dgram,
1795 	.pru_sosend =		sosend_dgram,
1796 	.pru_shutdown =		udp_shutdown,
1797 	.pru_sockaddr =		in_getsockaddr,
1798 	.pru_sosetlabel =	in_pcbsosetlabel,
1799 	.pru_close =		udp_close,
1800 };
1801 #endif /* INET */
1802