xref: /freebsd/sys/netinet/udp_usrreq.c (revision 7791ecf04b48a0c365b003447f479ec890115dfc)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.
6  * Copyright (c) 2008 Robert N. M. Watson
7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
8  * Copyright (c) 2014 Kevin Lo
9  * All rights reserved.
10  *
11  * Portions of this software were developed by Robert N. M. Watson under
12  * contract to Juniper Networks, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_route.h"
48 #include "opt_rss.h"
49 
50 #include <sys/param.h>
51 #include <sys/domain.h>
52 #include <sys/eventhandler.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/protosw.h>
61 #include <sys/sdt.h>
62 #include <sys/signalvar.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/systm.h>
69 
70 #include <vm/uma.h>
71 
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/route.h>
75 #include <net/route/nhop.h>
76 #include <net/rss_config.h>
77 
78 #include <netinet/in.h>
79 #include <netinet/in_kdtrace.h>
80 #include <netinet/in_fib.h>
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/in_var.h>
84 #include <netinet/ip.h>
85 #ifdef INET6
86 #include <netinet/ip6.h>
87 #endif
88 #include <netinet/ip_icmp.h>
89 #include <netinet/icmp_var.h>
90 #include <netinet/ip_var.h>
91 #include <netinet/ip_options.h>
92 #ifdef INET6
93 #include <netinet6/ip6_var.h>
94 #endif
95 #include <netinet/udp.h>
96 #include <netinet/udp_var.h>
97 #include <netinet/udplite.h>
98 #include <netinet/in_rss.h>
99 
100 #include <netipsec/ipsec_support.h>
101 
102 #include <machine/in_cksum.h>
103 
104 #include <security/mac/mac_framework.h>
105 
106 /*
107  * UDP and UDP-Lite protocols implementation.
108  * Per RFC 768, August, 1980.
109  * Per RFC 3828, July, 2004.
110  */
111 
112 /*
113  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
114  * removes the only data integrity mechanism for packets and malformed
115  * packets that would otherwise be discarded due to bad checksums, and may
116  * cause problems (especially for NFS data blocks).
117  */
118 VNET_DEFINE(int, udp_cksum) = 1;
119 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
120     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
121 
122 VNET_DEFINE(int, udp_log_in_vain) = 0;
123 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
124     &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets");
125 
126 VNET_DEFINE(int, udp_blackhole) = 0;
127 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
128     &VNET_NAME(udp_blackhole), 0,
129     "Do not send port unreachables for refused connects");
130 
131 u_long	udp_sendspace = 9216;		/* really max datagram size */
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 				      );	/* 40 1K datagrams */
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 VNET_DEFINE_STATIC(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 *, int);
169 #endif
170 
171 static void
172 udp_zone_change(void *tag)
173 {
174 
175 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
176 	uma_zone_set_max(V_udpcb_zone, maxsockets);
177 }
178 
179 static int
180 udp_inpcb_init(void *mem, int size, int flags)
181 {
182 	struct inpcb *inp;
183 
184 	inp = mem;
185 	INP_LOCK_INIT(inp, "inp", "udpinp");
186 	return (0);
187 }
188 
189 static int
190 udplite_inpcb_init(void *mem, int size, int flags)
191 {
192 	struct inpcb *inp;
193 
194 	inp = mem;
195 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
196 	return (0);
197 }
198 
199 void
200 udp_init(void)
201 {
202 
203 	/*
204 	 * For now default to 2-tuple UDP hashing - until the fragment
205 	 * reassembly code can also update the flowid.
206 	 *
207 	 * Once we can calculate the flowid that way and re-establish
208 	 * a 4-tuple, flip this to 4-tuple.
209 	 */
210 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
211 	    "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE);
212 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
213 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
214 	uma_zone_set_max(V_udpcb_zone, maxsockets);
215 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
216 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
217 	    EVENTHANDLER_PRI_ANY);
218 }
219 
220 void
221 udplite_init(void)
222 {
223 
224 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
225 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init,
226 	    IPI_HASHFIELDS_2TUPLE);
227 }
228 
229 /*
230  * Kernel module interface for updating udpstat.  The argument is an index
231  * into udpstat treated as an array of u_long.  While this encodes the
232  * general layout of udpstat into the caller, it doesn't encode its location,
233  * so that future changes to add, for example, per-CPU stats support won't
234  * cause binary compatibility problems for kernel modules.
235  */
236 void
237 kmod_udpstat_inc(int statnum)
238 {
239 
240 	counter_u64_add(VNET(udpstat)[statnum], 1);
241 }
242 
243 int
244 udp_newudpcb(struct inpcb *inp)
245 {
246 	struct udpcb *up;
247 
248 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
249 	if (up == NULL)
250 		return (ENOBUFS);
251 	inp->inp_ppcb = up;
252 	return (0);
253 }
254 
255 void
256 udp_discardcb(struct udpcb *up)
257 {
258 
259 	uma_zfree(V_udpcb_zone, up);
260 }
261 
262 #ifdef VIMAGE
263 static void
264 udp_destroy(void *unused __unused)
265 {
266 
267 	in_pcbinfo_destroy(&V_udbinfo);
268 	uma_zdestroy(V_udpcb_zone);
269 }
270 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
271 
272 static void
273 udplite_destroy(void *unused __unused)
274 {
275 
276 	in_pcbinfo_destroy(&V_ulitecbinfo);
277 }
278 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
279     NULL);
280 #endif
281 
282 #ifdef INET
283 /*
284  * Subroutine of udp_input(), which appends the provided mbuf chain to the
285  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
286  * contains the source address.  If the socket ends up being an IPv6 socket,
287  * udp_append() will convert to a sockaddr_in6 before passing the address
288  * into the socket code.
289  *
290  * In the normal case udp_append() will return 0, indicating that you
291  * must unlock the inp. However if a tunneling protocol is in place we increment
292  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
293  * then decrement the reference count. If the inp_rele returns 1, indicating the
294  * inp is gone, we return that to the caller to tell them *not* to unlock
295  * the inp. In the case of multi-cast this will cause the distribution
296  * to stop (though most tunneling protocols known currently do *not* use
297  * multicast).
298  */
299 static int
300 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
301     struct sockaddr_in *udp_in)
302 {
303 	struct sockaddr *append_sa;
304 	struct socket *so;
305 	struct mbuf *tmpopts, *opts = NULL;
306 #ifdef INET6
307 	struct sockaddr_in6 udp_in6;
308 #endif
309 	struct udpcb *up;
310 
311 	INP_LOCK_ASSERT(inp);
312 
313 	/*
314 	 * Engage the tunneling protocol.
315 	 */
316 	up = intoudpcb(inp);
317 	if (up->u_tun_func != NULL) {
318 		in_pcbref(inp);
319 		INP_RUNLOCK(inp);
320 		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
321 		    up->u_tun_ctx);
322 		INP_RLOCK(inp);
323 		return (in_pcbrele_rlocked(inp));
324 	}
325 
326 	off += sizeof(struct udphdr);
327 
328 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
329 	/* Check AH/ESP integrity. */
330 	if (IPSEC_ENABLED(ipv4) &&
331 	    IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
332 		m_freem(n);
333 		return (0);
334 	}
335 	if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
336 		if (IPSEC_ENABLED(ipv4) &&
337 		    UDPENCAP_INPUT(n, off, AF_INET) != 0)
338 			return (0);	/* Consumed. */
339 	}
340 #endif /* IPSEC */
341 #ifdef MAC
342 	if (mac_inpcb_check_deliver(inp, n) != 0) {
343 		m_freem(n);
344 		return (0);
345 	}
346 #endif /* MAC */
347 	if (inp->inp_flags & INP_CONTROLOPTS ||
348 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
349 #ifdef INET6
350 		if (inp->inp_vflag & INP_IPV6)
351 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
352 		else
353 #endif /* INET6 */
354 			ip_savecontrol(inp, &opts, ip, n);
355 	}
356 	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
357 		tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
358 			sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
359 		if (tmpopts) {
360 			if (opts) {
361 				tmpopts->m_next = opts;
362 				opts = tmpopts;
363 			} else
364 				opts = tmpopts;
365 		}
366 	}
367 #ifdef INET6
368 	if (inp->inp_vflag & INP_IPV6) {
369 		bzero(&udp_in6, sizeof(udp_in6));
370 		udp_in6.sin6_len = sizeof(udp_in6);
371 		udp_in6.sin6_family = AF_INET6;
372 		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
373 		append_sa = (struct sockaddr *)&udp_in6;
374 	} else
375 #endif /* INET6 */
376 		append_sa = (struct sockaddr *)&udp_in[0];
377 	m_adj(n, off);
378 
379 	so = inp->inp_socket;
380 	SOCKBUF_LOCK(&so->so_rcv);
381 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
382 		SOCKBUF_UNLOCK(&so->so_rcv);
383 		m_freem(n);
384 		if (opts)
385 			m_freem(opts);
386 		UDPSTAT_INC(udps_fullsock);
387 	} else
388 		sorwakeup_locked(so);
389 	return (0);
390 }
391 
392 int
393 udp_input(struct mbuf **mp, int *offp, int proto)
394 {
395 	struct ip *ip;
396 	struct udphdr *uh;
397 	struct ifnet *ifp;
398 	struct inpcb *inp;
399 	uint16_t len, ip_len;
400 	struct inpcbinfo *pcbinfo;
401 	struct ip save_ip;
402 	struct sockaddr_in udp_in[2];
403 	struct mbuf *m;
404 	struct m_tag *fwd_tag;
405 	int cscov_partial, iphlen;
406 
407 	m = *mp;
408 	iphlen = *offp;
409 	ifp = m->m_pkthdr.rcvif;
410 	*mp = NULL;
411 	UDPSTAT_INC(udps_ipackets);
412 
413 	/*
414 	 * Strip IP options, if any; should skip this, make available to
415 	 * user, and use on returned packets, but we don't yet have a way to
416 	 * check the checksum with options still present.
417 	 */
418 	if (iphlen > sizeof (struct ip)) {
419 		ip_stripoptions(m);
420 		iphlen = sizeof(struct ip);
421 	}
422 
423 	/*
424 	 * Get IP and UDP header together in first mbuf.
425 	 */
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 	}
432 	ip = mtod(m, struct ip *);
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[0], sizeof(struct sockaddr_in) * 2);
447 	udp_in[0].sin_len = sizeof(struct sockaddr_in);
448 	udp_in[0].sin_family = AF_INET;
449 	udp_in[0].sin_port = uh->uh_sport;
450 	udp_in[0].sin_addr = ip->ip_src;
451 	udp_in[1].sin_len = sizeof(struct sockaddr_in);
452 	udp_in[1].sin_family = AF_INET;
453 	udp_in[1].sin_port = uh->uh_dport;
454 	udp_in[1].sin_addr = ip->ip_dst;
455 
456 	/*
457 	 * Make mbuf data length reflect UDP length.  If not enough data to
458 	 * reflect UDP length, drop.
459 	 */
460 	len = ntohs((u_short)uh->uh_ulen);
461 	ip_len = ntohs(ip->ip_len) - iphlen;
462 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
463 		/* Zero means checksum over the complete packet. */
464 		if (len == 0)
465 			len = ip_len;
466 		cscov_partial = 0;
467 	}
468 	if (ip_len != len) {
469 		if (len > ip_len || len < sizeof(struct udphdr)) {
470 			UDPSTAT_INC(udps_badlen);
471 			goto badunlocked;
472 		}
473 		if (proto == IPPROTO_UDP)
474 			m_adj(m, len - ip_len);
475 	}
476 
477 	/*
478 	 * Save a copy of the IP header in case we want restore it for
479 	 * sending an ICMP error message in response.
480 	 */
481 	if (!V_udp_blackhole)
482 		save_ip = *ip;
483 	else
484 		memset(&save_ip, 0, sizeof(save_ip));
485 
486 	/*
487 	 * Checksum extended UDP header and data.
488 	 */
489 	if (uh->uh_sum) {
490 		u_short uh_sum;
491 
492 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
493 		    !cscov_partial) {
494 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
495 				uh_sum = m->m_pkthdr.csum_data;
496 			else
497 				uh_sum = in_pseudo(ip->ip_src.s_addr,
498 				    ip->ip_dst.s_addr, htonl((u_short)len +
499 				    m->m_pkthdr.csum_data + proto));
500 			uh_sum ^= 0xffff;
501 		} else {
502 			char b[9];
503 
504 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
505 			bzero(((struct ipovly *)ip)->ih_x1, 9);
506 			((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
507 			    uh->uh_ulen : htons(ip_len);
508 			uh_sum = in_cksum(m, len + sizeof (struct ip));
509 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
510 		}
511 		if (uh_sum) {
512 			UDPSTAT_INC(udps_badsum);
513 			m_freem(m);
514 			return (IPPROTO_DONE);
515 		}
516 	} else {
517 		if (proto == IPPROTO_UDP) {
518 			UDPSTAT_INC(udps_nosum);
519 		} else {
520 			/* UDPLite requires a checksum */
521 			/* XXX: What is the right UDPLite MIB counter here? */
522 			m_freem(m);
523 			return (IPPROTO_DONE);
524 		}
525 	}
526 
527 	pcbinfo = udp_get_inpcbinfo(proto);
528 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
529 	    in_broadcast(ip->ip_dst, ifp)) {
530 		struct inpcb *last;
531 		struct inpcbhead *pcblist;
532 
533 		NET_EPOCH_ASSERT();
534 
535 		pcblist = udp_get_pcblist(proto);
536 		last = NULL;
537 		CK_LIST_FOREACH(inp, pcblist, inp_list) {
538 			if (inp->inp_lport != uh->uh_dport)
539 				continue;
540 #ifdef INET6
541 			if ((inp->inp_vflag & INP_IPV4) == 0)
542 				continue;
543 #endif
544 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
545 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
546 				continue;
547 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
548 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
549 				continue;
550 			if (inp->inp_fport != 0 &&
551 			    inp->inp_fport != uh->uh_sport)
552 				continue;
553 
554 			INP_RLOCK(inp);
555 
556 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
557 				INP_RUNLOCK(inp);
558 				continue;
559 			}
560 
561 			/*
562 			 * XXXRW: Because we weren't holding either the inpcb
563 			 * or the hash lock when we checked for a match
564 			 * before, we should probably recheck now that the
565 			 * inpcb lock is held.
566 			 */
567 
568 			/*
569 			 * Handle socket delivery policy for any-source
570 			 * and source-specific multicast. [RFC3678]
571 			 */
572 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
573 				struct ip_moptions	*imo;
574 				struct sockaddr_in	 group;
575 				int			 blocked;
576 
577 				imo = inp->inp_moptions;
578 				if (imo == NULL) {
579 					INP_RUNLOCK(inp);
580 					continue;
581 				}
582 				bzero(&group, sizeof(struct sockaddr_in));
583 				group.sin_len = sizeof(struct sockaddr_in);
584 				group.sin_family = AF_INET;
585 				group.sin_addr = ip->ip_dst;
586 
587 				blocked = imo_multi_filter(imo, ifp,
588 					(struct sockaddr *)&group,
589 					(struct sockaddr *)&udp_in[0]);
590 				if (blocked != MCAST_PASS) {
591 					if (blocked == MCAST_NOTGMEMBER)
592 						IPSTAT_INC(ips_notmember);
593 					if (blocked == MCAST_NOTSMEMBER ||
594 					    blocked == MCAST_MUTED)
595 						UDPSTAT_INC(udps_filtermcast);
596 					INP_RUNLOCK(inp);
597 					continue;
598 				}
599 			}
600 			if (last != NULL) {
601 				struct mbuf *n;
602 
603 				if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
604 				    NULL) {
605 					if (proto == IPPROTO_UDPLITE)
606 						UDPLITE_PROBE(receive, NULL, last, ip,
607 						    last, uh);
608 					else
609 						UDP_PROBE(receive, NULL, last, ip, last,
610 						    uh);
611 					if (udp_append(last, ip, n, iphlen,
612 						udp_in)) {
613 						goto inp_lost;
614 					}
615 				}
616 				INP_RUNLOCK(last);
617 			}
618 			last = inp;
619 			/*
620 			 * Don't look for additional matches if this one does
621 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
622 			 * socket options set.  This heuristic avoids
623 			 * searching through all pcbs in the common case of a
624 			 * non-shared port.  It assumes that an application
625 			 * will never clear these options after setting them.
626 			 */
627 			if ((last->inp_socket->so_options &
628 			    (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
629 				break;
630 		}
631 
632 		if (last == NULL) {
633 			/*
634 			 * No matching pcb found; discard datagram.  (No need
635 			 * to send an ICMP Port Unreachable for a broadcast
636 			 * or multicast datgram.)
637 			 */
638 			UDPSTAT_INC(udps_noportbcast);
639 			if (inp)
640 				INP_RUNLOCK(inp);
641 			goto badunlocked;
642 		}
643 		if (proto == IPPROTO_UDPLITE)
644 			UDPLITE_PROBE(receive, NULL, last, ip, last, uh);
645 		else
646 			UDP_PROBE(receive, NULL, last, ip, last, uh);
647 		if (udp_append(last, ip, m, iphlen, udp_in) == 0)
648 			INP_RUNLOCK(last);
649 	inp_lost:
650 		return (IPPROTO_DONE);
651 	}
652 
653 	/*
654 	 * Locate pcb for datagram.
655 	 */
656 
657 	/*
658 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
659 	 */
660 	if ((m->m_flags & M_IP_NEXTHOP) &&
661 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
662 		struct sockaddr_in *next_hop;
663 
664 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
665 
666 		/*
667 		 * Transparently forwarded. Pretend to be the destination.
668 		 * Already got one like this?
669 		 */
670 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
671 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
672 		if (!inp) {
673 			/*
674 			 * It's new.  Try to find the ambushing socket.
675 			 * Because we've rewritten the destination address,
676 			 * any hardware-generated hash is ignored.
677 			 */
678 			inp = in_pcblookup(pcbinfo, ip->ip_src,
679 			    uh->uh_sport, next_hop->sin_addr,
680 			    next_hop->sin_port ? htons(next_hop->sin_port) :
681 			    uh->uh_dport, INPLOOKUP_WILDCARD |
682 			    INPLOOKUP_RLOCKPCB, ifp);
683 		}
684 		/* Remove the tag from the packet. We don't need it anymore. */
685 		m_tag_delete(m, fwd_tag);
686 		m->m_flags &= ~M_IP_NEXTHOP;
687 	} else
688 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
689 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
690 		    INPLOOKUP_RLOCKPCB, ifp, m);
691 	if (inp == NULL) {
692 		if (V_udp_log_in_vain) {
693 			char src[INET_ADDRSTRLEN];
694 			char dst[INET_ADDRSTRLEN];
695 
696 			log(LOG_INFO,
697 			    "Connection attempt to UDP %s:%d from %s:%d\n",
698 			    inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
699 			    inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
700 		}
701 		if (proto == IPPROTO_UDPLITE)
702 			UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
703 		else
704 			UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
705 		UDPSTAT_INC(udps_noport);
706 		if (m->m_flags & (M_BCAST | M_MCAST)) {
707 			UDPSTAT_INC(udps_noportbcast);
708 			goto badunlocked;
709 		}
710 		if (V_udp_blackhole)
711 			goto badunlocked;
712 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
713 			goto badunlocked;
714 		*ip = save_ip;
715 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
716 		return (IPPROTO_DONE);
717 	}
718 
719 	/*
720 	 * Check the minimum TTL for socket.
721 	 */
722 	INP_RLOCK_ASSERT(inp);
723 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
724 		if (proto == IPPROTO_UDPLITE)
725 			UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
726 		else
727 			UDP_PROBE(receive, NULL, inp, ip, inp, uh);
728 		INP_RUNLOCK(inp);
729 		m_freem(m);
730 		return (IPPROTO_DONE);
731 	}
732 	if (cscov_partial) {
733 		struct udpcb *up;
734 
735 		up = intoudpcb(inp);
736 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
737 			INP_RUNLOCK(inp);
738 			m_freem(m);
739 			return (IPPROTO_DONE);
740 		}
741 	}
742 
743 	if (proto == IPPROTO_UDPLITE)
744 		UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
745 	else
746 		UDP_PROBE(receive, NULL, inp, ip, inp, uh);
747 	if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
748 		INP_RUNLOCK(inp);
749 	return (IPPROTO_DONE);
750 
751 badunlocked:
752 	m_freem(m);
753 	return (IPPROTO_DONE);
754 }
755 #endif /* INET */
756 
757 /*
758  * Notify a udp user of an asynchronous error; just wake up so that they can
759  * collect error status.
760  */
761 struct inpcb *
762 udp_notify(struct inpcb *inp, int errno)
763 {
764 
765 	INP_WLOCK_ASSERT(inp);
766 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
767 	     errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
768 		NH_FREE(inp->inp_route.ro_nh);
769 		inp->inp_route.ro_nh = (struct nhop_object *)NULL;
770 	}
771 
772 	inp->inp_socket->so_error = errno;
773 	sorwakeup(inp->inp_socket);
774 	sowwakeup(inp->inp_socket);
775 	return (inp);
776 }
777 
778 #ifdef INET
779 static void
780 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
781     struct inpcbinfo *pcbinfo)
782 {
783 	struct ip *ip = vip;
784 	struct udphdr *uh;
785 	struct in_addr faddr;
786 	struct inpcb *inp;
787 
788 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
789 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
790 		return;
791 
792 	if (PRC_IS_REDIRECT(cmd)) {
793 		/* signal EHOSTDOWN, as it flushes the cached route */
794 		in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
795 		return;
796 	}
797 
798 	/*
799 	 * Hostdead is ugly because it goes linearly through all PCBs.
800 	 *
801 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
802 	 * DoS attack on machines with many connections.
803 	 */
804 	if (cmd == PRC_HOSTDEAD)
805 		ip = NULL;
806 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
807 		return;
808 	if (ip != NULL) {
809 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
810 		inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
811 		    ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
812 		if (inp != NULL) {
813 			INP_WLOCK_ASSERT(inp);
814 			if (inp->inp_socket != NULL) {
815 				udp_notify(inp, inetctlerrmap[cmd]);
816 			}
817 			INP_WUNLOCK(inp);
818 		} else {
819 			inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
820 					   ip->ip_src, uh->uh_sport,
821 					   INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
822 			if (inp != NULL) {
823 				struct udpcb *up;
824 				void *ctx;
825 				udp_tun_icmp_t func;
826 
827 				up = intoudpcb(inp);
828 				ctx = up->u_tun_ctx;
829 				func = up->u_icmp_func;
830 				INP_RUNLOCK(inp);
831 				if (func != NULL)
832 					(*func)(cmd, sa, vip, ctx);
833 			}
834 		}
835 	} else
836 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
837 		    udp_notify);
838 }
839 void
840 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
841 {
842 
843 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
844 }
845 
846 void
847 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
848 {
849 
850 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
851 }
852 #endif /* INET */
853 
854 static int
855 udp_pcblist(SYSCTL_HANDLER_ARGS)
856 {
857 	struct xinpgen xig;
858 	struct epoch_tracker et;
859 	struct inpcb *inp;
860 	int error;
861 
862 	if (req->newptr != 0)
863 		return (EPERM);
864 
865 	if (req->oldptr == 0) {
866 		int n;
867 
868 		n = V_udbinfo.ipi_count;
869 		n += imax(n / 8, 10);
870 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
871 		return (0);
872 	}
873 
874 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
875 		return (error);
876 
877 	bzero(&xig, sizeof(xig));
878 	xig.xig_len = sizeof xig;
879 	xig.xig_count = V_udbinfo.ipi_count;
880 	xig.xig_gen = V_udbinfo.ipi_gencnt;
881 	xig.xig_sogen = so_gencnt;
882 	error = SYSCTL_OUT(req, &xig, sizeof xig);
883 	if (error)
884 		return (error);
885 
886 	NET_EPOCH_ENTER(et);
887 	for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead);
888 	    inp != NULL;
889 	    inp = CK_LIST_NEXT(inp, inp_list)) {
890 		INP_RLOCK(inp);
891 		if (inp->inp_gencnt <= xig.xig_gen &&
892 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
893 			struct xinpcb xi;
894 
895 			in_pcbtoxinpcb(inp, &xi);
896 			INP_RUNLOCK(inp);
897 			error = SYSCTL_OUT(req, &xi, sizeof xi);
898 			if (error)
899 				break;
900 		} else
901 			INP_RUNLOCK(inp);
902 	}
903 	NET_EPOCH_EXIT(et);
904 
905 	if (!error) {
906 		/*
907 		 * Give the user an updated idea of our state.  If the
908 		 * generation differs from what we told her before, she knows
909 		 * that something happened while we were processing this
910 		 * request, and it might be necessary to retry.
911 		 */
912 		xig.xig_gen = V_udbinfo.ipi_gencnt;
913 		xig.xig_sogen = so_gencnt;
914 		xig.xig_count = V_udbinfo.ipi_count;
915 		error = SYSCTL_OUT(req, &xig, sizeof xig);
916 	}
917 
918 	return (error);
919 }
920 
921 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
922     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
923     udp_pcblist, "S,xinpcb",
924     "List of active UDP sockets");
925 
926 #ifdef INET
927 static int
928 udp_getcred(SYSCTL_HANDLER_ARGS)
929 {
930 	struct xucred xuc;
931 	struct sockaddr_in addrs[2];
932 	struct epoch_tracker et;
933 	struct inpcb *inp;
934 	int error;
935 
936 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
937 	if (error)
938 		return (error);
939 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
940 	if (error)
941 		return (error);
942 	NET_EPOCH_ENTER(et);
943 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
944 	    addrs[0].sin_addr, addrs[0].sin_port,
945 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
946 	NET_EPOCH_EXIT(et);
947 	if (inp != NULL) {
948 		INP_RLOCK_ASSERT(inp);
949 		if (inp->inp_socket == NULL)
950 			error = ENOENT;
951 		if (error == 0)
952 			error = cr_canseeinpcb(req->td->td_ucred, inp);
953 		if (error == 0)
954 			cru2x(inp->inp_cred, &xuc);
955 		INP_RUNLOCK(inp);
956 	} else
957 		error = ENOENT;
958 	if (error == 0)
959 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
960 	return (error);
961 }
962 
963 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
964     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
965     0, 0, udp_getcred, "S,xucred",
966     "Get the xucred of a UDP connection");
967 #endif /* INET */
968 
969 int
970 udp_ctloutput(struct socket *so, struct sockopt *sopt)
971 {
972 	struct inpcb *inp;
973 	struct udpcb *up;
974 	int isudplite, error, optval;
975 
976 	error = 0;
977 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
978 	inp = sotoinpcb(so);
979 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
980 	INP_WLOCK(inp);
981 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
982 #ifdef INET6
983 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
984 			INP_WUNLOCK(inp);
985 			error = ip6_ctloutput(so, sopt);
986 		}
987 #endif
988 #if defined(INET) && defined(INET6)
989 		else
990 #endif
991 #ifdef INET
992 		{
993 			INP_WUNLOCK(inp);
994 			error = ip_ctloutput(so, sopt);
995 		}
996 #endif
997 		return (error);
998 	}
999 
1000 	switch (sopt->sopt_dir) {
1001 	case SOPT_SET:
1002 		switch (sopt->sopt_name) {
1003 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1004 #ifdef INET
1005 		case UDP_ENCAP:
1006 			if (!IPSEC_ENABLED(ipv4)) {
1007 				INP_WUNLOCK(inp);
1008 				return (ENOPROTOOPT);
1009 			}
1010 			error = UDPENCAP_PCBCTL(inp, sopt);
1011 			break;
1012 #endif /* INET */
1013 #endif /* IPSEC */
1014 		case UDPLITE_SEND_CSCOV:
1015 		case UDPLITE_RECV_CSCOV:
1016 			if (!isudplite) {
1017 				INP_WUNLOCK(inp);
1018 				error = ENOPROTOOPT;
1019 				break;
1020 			}
1021 			INP_WUNLOCK(inp);
1022 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1023 			    sizeof(optval));
1024 			if (error != 0)
1025 				break;
1026 			inp = sotoinpcb(so);
1027 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1028 			INP_WLOCK(inp);
1029 			up = intoudpcb(inp);
1030 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1031 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
1032 				INP_WUNLOCK(inp);
1033 				error = EINVAL;
1034 				break;
1035 			}
1036 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1037 				up->u_txcslen = optval;
1038 			else
1039 				up->u_rxcslen = optval;
1040 			INP_WUNLOCK(inp);
1041 			break;
1042 		default:
1043 			INP_WUNLOCK(inp);
1044 			error = ENOPROTOOPT;
1045 			break;
1046 		}
1047 		break;
1048 	case SOPT_GET:
1049 		switch (sopt->sopt_name) {
1050 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1051 #ifdef INET
1052 		case UDP_ENCAP:
1053 			if (!IPSEC_ENABLED(ipv4)) {
1054 				INP_WUNLOCK(inp);
1055 				return (ENOPROTOOPT);
1056 			}
1057 			error = UDPENCAP_PCBCTL(inp, sopt);
1058 			break;
1059 #endif /* INET */
1060 #endif /* IPSEC */
1061 		case UDPLITE_SEND_CSCOV:
1062 		case UDPLITE_RECV_CSCOV:
1063 			if (!isudplite) {
1064 				INP_WUNLOCK(inp);
1065 				error = ENOPROTOOPT;
1066 				break;
1067 			}
1068 			up = intoudpcb(inp);
1069 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1070 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1071 				optval = up->u_txcslen;
1072 			else
1073 				optval = up->u_rxcslen;
1074 			INP_WUNLOCK(inp);
1075 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1076 			break;
1077 		default:
1078 			INP_WUNLOCK(inp);
1079 			error = ENOPROTOOPT;
1080 			break;
1081 		}
1082 		break;
1083 	}
1084 	return (error);
1085 }
1086 
1087 #ifdef INET
1088 #ifdef INET6
1089 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1090 static int
1091 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1092     struct inpcb *inp, int flags)
1093 {
1094 	struct ifnet *ifp;
1095 	struct in6_pktinfo *pktinfo;
1096 	struct in_addr ia;
1097 
1098 	if ((flags & PRUS_IPV6) == 0)
1099 		return (0);
1100 
1101 	if (cm->cmsg_level != IPPROTO_IPV6)
1102 		return (0);
1103 
1104 	if  (cm->cmsg_type != IPV6_2292PKTINFO &&
1105 	    cm->cmsg_type != IPV6_PKTINFO)
1106 		return (0);
1107 
1108 	if (cm->cmsg_len !=
1109 	    CMSG_LEN(sizeof(struct in6_pktinfo)))
1110 		return (EINVAL);
1111 
1112 	pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1113 	if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1114 	    !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1115 		return (EINVAL);
1116 
1117 	/* Validate the interface index if specified. */
1118 	if (pktinfo->ipi6_ifindex > V_if_index)
1119 		return (ENXIO);
1120 
1121 	ifp = NULL;
1122 	if (pktinfo->ipi6_ifindex) {
1123 		ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1124 		if (ifp == NULL)
1125 			return (ENXIO);
1126 	}
1127 	if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1128 		ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1129 		if (in_ifhasaddr(ifp, ia) == 0)
1130 			return (EADDRNOTAVAIL);
1131 	}
1132 
1133 	bzero(src, sizeof(*src));
1134 	src->sin_family = AF_INET;
1135 	src->sin_len = sizeof(*src);
1136 	src->sin_port = inp->inp_lport;
1137 	src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1138 
1139 	return (0);
1140 }
1141 #endif
1142 
1143 static int
1144 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1145     struct mbuf *control, struct thread *td, int flags)
1146 {
1147 	struct udpiphdr *ui;
1148 	int len = m->m_pkthdr.len;
1149 	struct in_addr faddr, laddr;
1150 	struct cmsghdr *cm;
1151 	struct inpcbinfo *pcbinfo;
1152 	struct sockaddr_in *sin, src;
1153 	struct epoch_tracker et;
1154 	int cscov_partial = 0;
1155 	int error = 0;
1156 	int ipflags = 0;
1157 	u_short fport, lport;
1158 	u_char tos;
1159 	uint8_t pr;
1160 	uint16_t cscov = 0;
1161 	uint32_t flowid = 0;
1162 	uint8_t flowtype = M_HASHTYPE_NONE;
1163 
1164 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1165 		if (control)
1166 			m_freem(control);
1167 		m_freem(m);
1168 		return (EMSGSIZE);
1169 	}
1170 
1171 	src.sin_family = 0;
1172 	sin = (struct sockaddr_in *)addr;
1173 
1174 	/*
1175 	 * udp_output() may need to temporarily bind or connect the current
1176 	 * inpcb.  As such, we don't know up front whether we will need the
1177 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1178 	 * front before acquiring any locks.
1179 	 *
1180 	 * We will need network epoch in either case, to safely lookup into
1181 	 * pcb hash.
1182 	 */
1183 	if (sin == NULL ||
1184 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0))
1185 		INP_WLOCK(inp);
1186 	else
1187 		INP_RLOCK(inp);
1188 	NET_EPOCH_ENTER(et);
1189 	tos = inp->inp_ip_tos;
1190 	if (control != NULL) {
1191 		/*
1192 		 * XXX: Currently, we assume all the optional information is
1193 		 * stored in a single mbuf.
1194 		 */
1195 		if (control->m_next) {
1196 			m_freem(control);
1197 			error = EINVAL;
1198 			goto release;
1199 		}
1200 		for (; control->m_len > 0;
1201 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1202 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1203 			cm = mtod(control, struct cmsghdr *);
1204 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1205 			    || cm->cmsg_len > control->m_len) {
1206 				error = EINVAL;
1207 				break;
1208 			}
1209 #ifdef INET6
1210 			error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1211 			if (error != 0)
1212 				break;
1213 #endif
1214 			if (cm->cmsg_level != IPPROTO_IP)
1215 				continue;
1216 
1217 			switch (cm->cmsg_type) {
1218 			case IP_SENDSRCADDR:
1219 				if (cm->cmsg_len !=
1220 				    CMSG_LEN(sizeof(struct in_addr))) {
1221 					error = EINVAL;
1222 					break;
1223 				}
1224 				bzero(&src, sizeof(src));
1225 				src.sin_family = AF_INET;
1226 				src.sin_len = sizeof(src);
1227 				src.sin_port = inp->inp_lport;
1228 				src.sin_addr =
1229 				    *(struct in_addr *)CMSG_DATA(cm);
1230 				break;
1231 
1232 			case IP_TOS:
1233 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1234 					error = EINVAL;
1235 					break;
1236 				}
1237 				tos = *(u_char *)CMSG_DATA(cm);
1238 				break;
1239 
1240 			case IP_FLOWID:
1241 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1242 					error = EINVAL;
1243 					break;
1244 				}
1245 				flowid = *(uint32_t *) CMSG_DATA(cm);
1246 				break;
1247 
1248 			case IP_FLOWTYPE:
1249 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1250 					error = EINVAL;
1251 					break;
1252 				}
1253 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1254 				break;
1255 
1256 #ifdef	RSS
1257 			case IP_RSSBUCKETID:
1258 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1259 					error = EINVAL;
1260 					break;
1261 				}
1262 				/* This is just a placeholder for now */
1263 				break;
1264 #endif	/* RSS */
1265 			default:
1266 				error = ENOPROTOOPT;
1267 				break;
1268 			}
1269 			if (error)
1270 				break;
1271 		}
1272 		m_freem(control);
1273 	}
1274 	if (error)
1275 		goto release;
1276 
1277 	pr = inp->inp_socket->so_proto->pr_protocol;
1278 	pcbinfo = udp_get_inpcbinfo(pr);
1279 
1280 	/*
1281 	 * If the IP_SENDSRCADDR control message was specified, override the
1282 	 * source address for this datagram.  Its use is invalidated if the
1283 	 * address thus specified is incomplete or clobbers other inpcbs.
1284 	 */
1285 	laddr = inp->inp_laddr;
1286 	lport = inp->inp_lport;
1287 	if (src.sin_family == AF_INET) {
1288 		INP_HASH_LOCK_ASSERT(pcbinfo);
1289 		if ((lport == 0) ||
1290 		    (laddr.s_addr == INADDR_ANY &&
1291 		     src.sin_addr.s_addr == INADDR_ANY)) {
1292 			error = EINVAL;
1293 			goto release;
1294 		}
1295 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1296 		    &laddr.s_addr, &lport, td->td_ucred);
1297 		if (error)
1298 			goto release;
1299 	}
1300 
1301 	/*
1302 	 * If a UDP socket has been connected, then a local address/port will
1303 	 * have been selected and bound.
1304 	 *
1305 	 * If a UDP socket has not been connected to, then an explicit
1306 	 * destination address must be used, in which case a local
1307 	 * address/port may not have been selected and bound.
1308 	 */
1309 	if (sin != NULL) {
1310 		INP_LOCK_ASSERT(inp);
1311 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1312 			error = EISCONN;
1313 			goto release;
1314 		}
1315 
1316 		/*
1317 		 * Jail may rewrite the destination address, so let it do
1318 		 * that before we use it.
1319 		 */
1320 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1321 		if (error)
1322 			goto release;
1323 
1324 		/*
1325 		 * If a local address or port hasn't yet been selected, or if
1326 		 * the destination address needs to be rewritten due to using
1327 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1328 		 * to do the heavy lifting.  Once a port is selected, we
1329 		 * commit the binding back to the socket; we also commit the
1330 		 * binding of the address if in jail.
1331 		 *
1332 		 * If we already have a valid binding and we're not
1333 		 * requesting a destination address rewrite, use a fast path.
1334 		 */
1335 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1336 		    inp->inp_lport == 0 ||
1337 		    sin->sin_addr.s_addr == INADDR_ANY ||
1338 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1339 			INP_HASH_LOCK_ASSERT(pcbinfo);
1340 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1341 			    &lport, &faddr.s_addr, &fport, NULL,
1342 			    td->td_ucred);
1343 			if (error)
1344 				goto release;
1345 
1346 			/*
1347 			 * XXXRW: Why not commit the port if the address is
1348 			 * !INADDR_ANY?
1349 			 */
1350 			/* Commit the local port if newly assigned. */
1351 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1352 			    inp->inp_lport == 0) {
1353 				INP_WLOCK_ASSERT(inp);
1354 				/*
1355 				 * Remember addr if jailed, to prevent
1356 				 * rebinding.
1357 				 */
1358 				if (prison_flag(td->td_ucred, PR_IP4))
1359 					inp->inp_laddr = laddr;
1360 				inp->inp_lport = lport;
1361 				INP_HASH_WLOCK(pcbinfo);
1362 				error = in_pcbinshash(inp);
1363 				INP_HASH_WUNLOCK(pcbinfo);
1364 				if (error != 0) {
1365 					inp->inp_lport = 0;
1366 					error = EAGAIN;
1367 					goto release;
1368 				}
1369 				inp->inp_flags |= INP_ANONPORT;
1370 			}
1371 		} else {
1372 			faddr = sin->sin_addr;
1373 			fport = sin->sin_port;
1374 		}
1375 	} else {
1376 		INP_LOCK_ASSERT(inp);
1377 		faddr = inp->inp_faddr;
1378 		fport = inp->inp_fport;
1379 		if (faddr.s_addr == INADDR_ANY) {
1380 			error = ENOTCONN;
1381 			goto release;
1382 		}
1383 	}
1384 
1385 	/*
1386 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1387 	 * link-layer headers.  Immediate slide the data pointer back forward
1388 	 * since we won't use that space at this layer.
1389 	 */
1390 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1391 	if (m == NULL) {
1392 		error = ENOBUFS;
1393 		goto release;
1394 	}
1395 	m->m_data += max_linkhdr;
1396 	m->m_len -= max_linkhdr;
1397 	m->m_pkthdr.len -= max_linkhdr;
1398 
1399 	/*
1400 	 * Fill in mbuf with extended UDP header and addresses and length put
1401 	 * into network format.
1402 	 */
1403 	ui = mtod(m, struct udpiphdr *);
1404 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1405 	ui->ui_v = IPVERSION << 4;
1406 	ui->ui_pr = pr;
1407 	ui->ui_src = laddr;
1408 	ui->ui_dst = faddr;
1409 	ui->ui_sport = lport;
1410 	ui->ui_dport = fport;
1411 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1412 	if (pr == IPPROTO_UDPLITE) {
1413 		struct udpcb *up;
1414 		uint16_t plen;
1415 
1416 		up = intoudpcb(inp);
1417 		cscov = up->u_txcslen;
1418 		plen = (u_short)len + sizeof(struct udphdr);
1419 		if (cscov >= plen)
1420 			cscov = 0;
1421 		ui->ui_len = htons(plen);
1422 		ui->ui_ulen = htons(cscov);
1423 		/*
1424 		 * For UDP-Lite, checksum coverage length of zero means
1425 		 * the entire UDPLite packet is covered by the checksum.
1426 		 */
1427 		cscov_partial = (cscov == 0) ? 0 : 1;
1428 	}
1429 
1430 	/*
1431 	 * Set the Don't Fragment bit in the IP header.
1432 	 */
1433 	if (inp->inp_flags & INP_DONTFRAG) {
1434 		struct ip *ip;
1435 
1436 		ip = (struct ip *)&ui->ui_i;
1437 		ip->ip_off |= htons(IP_DF);
1438 	}
1439 
1440 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1441 		ipflags |= IP_ROUTETOIF;
1442 	if (inp->inp_socket->so_options & SO_BROADCAST)
1443 		ipflags |= IP_ALLOWBROADCAST;
1444 	if (inp->inp_flags & INP_ONESBCAST)
1445 		ipflags |= IP_SENDONES;
1446 
1447 #ifdef MAC
1448 	mac_inpcb_create_mbuf(inp, m);
1449 #endif
1450 
1451 	/*
1452 	 * Set up checksum and output datagram.
1453 	 */
1454 	ui->ui_sum = 0;
1455 	if (pr == IPPROTO_UDPLITE) {
1456 		if (inp->inp_flags & INP_ONESBCAST)
1457 			faddr.s_addr = INADDR_BROADCAST;
1458 		if (cscov_partial) {
1459 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1460 				ui->ui_sum = 0xffff;
1461 		} else {
1462 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1463 				ui->ui_sum = 0xffff;
1464 		}
1465 	} else if (V_udp_cksum) {
1466 		if (inp->inp_flags & INP_ONESBCAST)
1467 			faddr.s_addr = INADDR_BROADCAST;
1468 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1469 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1470 		m->m_pkthdr.csum_flags = CSUM_UDP;
1471 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1472 	}
1473 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1474 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1475 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1476 	UDPSTAT_INC(udps_opackets);
1477 
1478 	/*
1479 	 * Setup flowid / RSS information for outbound socket.
1480 	 *
1481 	 * Once the UDP code decides to set a flowid some other way,
1482 	 * this allows the flowid to be overridden by userland.
1483 	 */
1484 	if (flowtype != M_HASHTYPE_NONE) {
1485 		m->m_pkthdr.flowid = flowid;
1486 		M_HASHTYPE_SET(m, flowtype);
1487 	}
1488 #if defined(ROUTE_MPATH) || defined(RSS)
1489 	else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1490 		uint32_t hash_val, hash_type;
1491 
1492 		hash_val = fib4_calc_packet_hash(laddr, faddr,
1493 		    lport, fport, pr, &hash_type);
1494 		m->m_pkthdr.flowid = hash_val;
1495 		M_HASHTYPE_SET(m, hash_type);
1496 	}
1497 
1498 	/*
1499 	 * Don't override with the inp cached flowid value.
1500 	 *
1501 	 * Depending upon the kind of send being done, the inp
1502 	 * flowid/flowtype values may actually not be appropriate
1503 	 * for this particular socket send.
1504 	 *
1505 	 * We should either leave the flowid at zero (which is what is
1506 	 * currently done) or set it to some software generated
1507 	 * hash value based on the packet contents.
1508 	 */
1509 	ipflags |= IP_NODEFAULTFLOWID;
1510 #endif	/* RSS */
1511 
1512 	if (pr == IPPROTO_UDPLITE)
1513 		UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1514 	else
1515 		UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1516 	error = ip_output(m, inp->inp_options,
1517 	    INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags,
1518 	    inp->inp_moptions, inp);
1519 	INP_UNLOCK(inp);
1520 	NET_EPOCH_EXIT(et);
1521 	return (error);
1522 
1523 release:
1524 	INP_UNLOCK(inp);
1525 	NET_EPOCH_EXIT(et);
1526 	m_freem(m);
1527 	return (error);
1528 }
1529 
1530 static void
1531 udp_abort(struct socket *so)
1532 {
1533 	struct inpcb *inp;
1534 	struct inpcbinfo *pcbinfo;
1535 
1536 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1537 	inp = sotoinpcb(so);
1538 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1539 	INP_WLOCK(inp);
1540 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1541 		INP_HASH_WLOCK(pcbinfo);
1542 		in_pcbdisconnect(inp);
1543 		inp->inp_laddr.s_addr = INADDR_ANY;
1544 		INP_HASH_WUNLOCK(pcbinfo);
1545 		soisdisconnected(so);
1546 	}
1547 	INP_WUNLOCK(inp);
1548 }
1549 
1550 static int
1551 udp_attach(struct socket *so, int proto, struct thread *td)
1552 {
1553 	static uint32_t udp_flowid;
1554 	struct inpcb *inp;
1555 	struct inpcbinfo *pcbinfo;
1556 	int error;
1557 
1558 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1559 	inp = sotoinpcb(so);
1560 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1561 	error = soreserve(so, udp_sendspace, udp_recvspace);
1562 	if (error)
1563 		return (error);
1564 	INP_INFO_WLOCK(pcbinfo);
1565 	error = in_pcballoc(so, pcbinfo);
1566 	if (error) {
1567 		INP_INFO_WUNLOCK(pcbinfo);
1568 		return (error);
1569 	}
1570 
1571 	inp = sotoinpcb(so);
1572 	inp->inp_vflag |= INP_IPV4;
1573 	inp->inp_ip_ttl = V_ip_defttl;
1574 	inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1575 	inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1576 
1577 	error = udp_newudpcb(inp);
1578 	if (error) {
1579 		in_pcbdetach(inp);
1580 		in_pcbfree(inp);
1581 		INP_INFO_WUNLOCK(pcbinfo);
1582 		return (error);
1583 	}
1584 
1585 	INP_WUNLOCK(inp);
1586 	INP_INFO_WUNLOCK(pcbinfo);
1587 	return (0);
1588 }
1589 #endif /* INET */
1590 
1591 int
1592 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1593 {
1594 	struct inpcb *inp;
1595 	struct udpcb *up;
1596 
1597 	KASSERT(so->so_type == SOCK_DGRAM,
1598 	    ("udp_set_kernel_tunneling: !dgram"));
1599 	inp = sotoinpcb(so);
1600 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1601 	INP_WLOCK(inp);
1602 	up = intoudpcb(inp);
1603 	if ((up->u_tun_func != NULL) ||
1604 	    (up->u_icmp_func != NULL)) {
1605 		INP_WUNLOCK(inp);
1606 		return (EBUSY);
1607 	}
1608 	up->u_tun_func = f;
1609 	up->u_icmp_func = i;
1610 	up->u_tun_ctx = ctx;
1611 	INP_WUNLOCK(inp);
1612 	return (0);
1613 }
1614 
1615 #ifdef INET
1616 static int
1617 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1618 {
1619 	struct inpcb *inp;
1620 	struct inpcbinfo *pcbinfo;
1621 	int error;
1622 
1623 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1624 	inp = sotoinpcb(so);
1625 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1626 	INP_WLOCK(inp);
1627 	INP_HASH_WLOCK(pcbinfo);
1628 	error = in_pcbbind(inp, nam, td->td_ucred);
1629 	INP_HASH_WUNLOCK(pcbinfo);
1630 	INP_WUNLOCK(inp);
1631 	return (error);
1632 }
1633 
1634 static void
1635 udp_close(struct socket *so)
1636 {
1637 	struct inpcb *inp;
1638 	struct inpcbinfo *pcbinfo;
1639 
1640 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1641 	inp = sotoinpcb(so);
1642 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1643 	INP_WLOCK(inp);
1644 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1645 		INP_HASH_WLOCK(pcbinfo);
1646 		in_pcbdisconnect(inp);
1647 		inp->inp_laddr.s_addr = INADDR_ANY;
1648 		INP_HASH_WUNLOCK(pcbinfo);
1649 		soisdisconnected(so);
1650 	}
1651 	INP_WUNLOCK(inp);
1652 }
1653 
1654 static int
1655 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1656 {
1657 	struct epoch_tracker et;
1658 	struct inpcb *inp;
1659 	struct inpcbinfo *pcbinfo;
1660 	struct sockaddr_in *sin;
1661 	int error;
1662 
1663 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1664 	inp = sotoinpcb(so);
1665 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1666 	INP_WLOCK(inp);
1667 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1668 		INP_WUNLOCK(inp);
1669 		return (EISCONN);
1670 	}
1671 	sin = (struct sockaddr_in *)nam;
1672 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1673 	if (error != 0) {
1674 		INP_WUNLOCK(inp);
1675 		return (error);
1676 	}
1677 	NET_EPOCH_ENTER(et);
1678 	INP_HASH_WLOCK(pcbinfo);
1679 	error = in_pcbconnect(inp, nam, td->td_ucred);
1680 	INP_HASH_WUNLOCK(pcbinfo);
1681 	NET_EPOCH_EXIT(et);
1682 	if (error == 0)
1683 		soisconnected(so);
1684 	INP_WUNLOCK(inp);
1685 	return (error);
1686 }
1687 
1688 static void
1689 udp_detach(struct socket *so)
1690 {
1691 	struct inpcb *inp;
1692 	struct inpcbinfo *pcbinfo;
1693 	struct udpcb *up;
1694 
1695 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1696 	inp = sotoinpcb(so);
1697 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1698 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1699 	    ("udp_detach: not disconnected"));
1700 	INP_INFO_WLOCK(pcbinfo);
1701 	INP_WLOCK(inp);
1702 	up = intoudpcb(inp);
1703 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1704 	inp->inp_ppcb = NULL;
1705 	in_pcbdetach(inp);
1706 	in_pcbfree(inp);
1707 	INP_INFO_WUNLOCK(pcbinfo);
1708 	udp_discardcb(up);
1709 }
1710 
1711 static int
1712 udp_disconnect(struct socket *so)
1713 {
1714 	struct inpcb *inp;
1715 	struct inpcbinfo *pcbinfo;
1716 
1717 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1718 	inp = sotoinpcb(so);
1719 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1720 	INP_WLOCK(inp);
1721 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1722 		INP_WUNLOCK(inp);
1723 		return (ENOTCONN);
1724 	}
1725 	INP_HASH_WLOCK(pcbinfo);
1726 	in_pcbdisconnect(inp);
1727 	inp->inp_laddr.s_addr = INADDR_ANY;
1728 	INP_HASH_WUNLOCK(pcbinfo);
1729 	SOCK_LOCK(so);
1730 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1731 	SOCK_UNLOCK(so);
1732 	INP_WUNLOCK(inp);
1733 	return (0);
1734 }
1735 
1736 static int
1737 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1738     struct mbuf *control, struct thread *td)
1739 {
1740 	struct inpcb *inp;
1741 
1742 	inp = sotoinpcb(so);
1743 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1744 	return (udp_output(inp, m, addr, control, td, flags));
1745 }
1746 #endif /* INET */
1747 
1748 int
1749 udp_shutdown(struct socket *so)
1750 {
1751 	struct inpcb *inp;
1752 
1753 	inp = sotoinpcb(so);
1754 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1755 	INP_WLOCK(inp);
1756 	socantsendmore(so);
1757 	INP_WUNLOCK(inp);
1758 	return (0);
1759 }
1760 
1761 #ifdef INET
1762 struct pr_usrreqs udp_usrreqs = {
1763 	.pru_abort =		udp_abort,
1764 	.pru_attach =		udp_attach,
1765 	.pru_bind =		udp_bind,
1766 	.pru_connect =		udp_connect,
1767 	.pru_control =		in_control,
1768 	.pru_detach =		udp_detach,
1769 	.pru_disconnect =	udp_disconnect,
1770 	.pru_peeraddr =		in_getpeeraddr,
1771 	.pru_send =		udp_send,
1772 	.pru_soreceive =	soreceive_dgram,
1773 	.pru_sosend =		sosend_dgram,
1774 	.pru_shutdown =		udp_shutdown,
1775 	.pru_sockaddr =		in_getsockaddr,
1776 	.pru_sosetlabel =	in_pcbsosetlabel,
1777 	.pru_close =		udp_close,
1778 };
1779 #endif /* INET */
1780