xref: /freebsd/sys/netinet/udp_usrreq.c (revision 328110da2661a8841f12000b99fea27ceacdd5b2)
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 
39 #include <sys/cdefs.h>
40 #include "opt_inet.h"
41 #include "opt_inet6.h"
42 #include "opt_ipsec.h"
43 #include "opt_route.h"
44 #include "opt_rss.h"
45 
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/sdt.h>
58 #include <sys/signalvar.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sx.h>
62 #include <sys/sysctl.h>
63 #include <sys/syslog.h>
64 #include <sys/systm.h>
65 
66 #include <vm/uma.h>
67 
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/route.h>
71 #include <net/route/nhop.h>
72 #include <net/rss_config.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #endif
84 #include <netinet/ip_icmp.h>
85 #include <netinet/icmp_var.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_options.h>
88 #ifdef INET6
89 #include <netinet6/ip6_var.h>
90 #endif
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
93 #include <netinet/udplite.h>
94 #include <netinet/in_rss.h>
95 
96 #include <netipsec/ipsec_support.h>
97 
98 #include <machine/in_cksum.h>
99 
100 #include <security/mac/mac_framework.h>
101 
102 /*
103  * UDP and UDP-Lite protocols implementation.
104  * Per RFC 768, August, 1980.
105  * Per RFC 3828, July, 2004.
106  */
107 
108 VNET_DEFINE(int, udp_bind_all_fibs) = 1;
109 SYSCTL_INT(_net_inet_udp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN,
110     &VNET_NAME(udp_bind_all_fibs), 0,
111     "Bound sockets receive traffic from all FIBs");
112 
113 /*
114  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
115  * removes the only data integrity mechanism for packets and malformed
116  * packets that would otherwise be discarded due to bad checksums, and may
117  * cause problems (especially for NFS data blocks).
118  */
119 VNET_DEFINE(int, udp_cksum) = 1;
120 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
121     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
122 
123 VNET_DEFINE(int, udp_log_in_vain) = 0;
124 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
125     &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets");
126 
127 VNET_DEFINE(int, udp_blackhole) = 0;
128 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
129     &VNET_NAME(udp_blackhole), 0,
130     "Do not send port unreachables for refused connects");
131 VNET_DEFINE(bool, udp_blackhole_local) = false;
132 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
133     CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false,
134     "Enforce net.inet.udp.blackhole for locally originated packets");
135 
136 u_long	udp_sendspace = 9216;		/* really max datagram size */
137 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
138     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
139 
140 u_long	udp_recvspace = 40 * (1024 +
141 #ifdef INET6
142 				      sizeof(struct sockaddr_in6)
143 #else
144 				      sizeof(struct sockaddr_in)
145 #endif
146 				      );	/* 40 1K datagrams */
147 
148 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
149     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
150 
151 VNET_DEFINE(struct inpcbinfo, udbinfo);
152 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
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 #endif
169 
170 INPCBSTORAGE_DEFINE(udpcbstor, udpcb, "udpinp", "udp_inpcb", "udp", "udphash");
171 INPCBSTORAGE_DEFINE(udplitecbstor, udpcb, "udpliteinp", "udplite_inpcb",
172     "udplite", "udplitehash");
173 
174 static void
175 udp_vnet_init(void *arg __unused)
176 {
177 
178 	/*
179 	 * For now default to 2-tuple UDP hashing - until the fragment
180 	 * reassembly code can also update the flowid.
181 	 *
182 	 * Once we can calculate the flowid that way and re-establish
183 	 * a 4-tuple, flip this to 4-tuple.
184 	 */
185 	in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE);
186 	/* Additional pcbinfo for UDP-Lite */
187 	in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE,
188 	    UDBHASHSIZE);
189 }
190 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
191     udp_vnet_init, NULL);
192 
193 /*
194  * Kernel module interface for updating udpstat.  The argument is an index
195  * into udpstat treated as an array of u_long.  While this encodes the
196  * general layout of udpstat into the caller, it doesn't encode its location,
197  * so that future changes to add, for example, per-CPU stats support won't
198  * cause binary compatibility problems for kernel modules.
199  */
200 void
201 kmod_udpstat_inc(int statnum)
202 {
203 
204 	counter_u64_add(VNET(udpstat)[statnum], 1);
205 }
206 
207 #ifdef VIMAGE
208 static void
209 udp_destroy(void *unused __unused)
210 {
211 
212 	in_pcbinfo_destroy(&V_udbinfo);
213 	in_pcbinfo_destroy(&V_ulitecbinfo);
214 }
215 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
216 #endif
217 
218 #ifdef INET
219 /*
220  * Subroutine of udp_input(), which appends the provided mbuf chain to the
221  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
222  * contains the source address.  If the socket ends up being an IPv6 socket,
223  * udp_append() will convert to a sockaddr_in6 before passing the address
224  * into the socket code.
225  *
226  * In the normal case udp_append() will return 0, indicating that you
227  * must unlock the inp. However if a tunneling protocol is in place we increment
228  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
229  * then decrement the reference count. If the inp_rele returns 1, indicating the
230  * inp is gone, we return that to the caller to tell them *not* to unlock
231  * the inp. In the case of multi-cast this will cause the distribution
232  * to stop (though most tunneling protocols known currently do *not* use
233  * multicast).
234  */
235 static int
236 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
237     struct sockaddr_in *udp_in)
238 {
239 	struct sockaddr *append_sa;
240 	struct socket *so;
241 	struct mbuf *tmpopts, *opts = NULL;
242 #ifdef INET6
243 	struct sockaddr_in6 udp_in6;
244 #endif
245 	struct udpcb *up;
246 
247 	INP_LOCK_ASSERT(inp);
248 
249 	/*
250 	 * Engage the tunneling protocol.
251 	 */
252 	up = intoudpcb(inp);
253 	if (up->u_tun_func != NULL) {
254 		bool filtered;
255 
256 		in_pcbref(inp);
257 		INP_RUNLOCK(inp);
258 		filtered = (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
259 		    up->u_tun_ctx);
260 		INP_RLOCK(inp);
261 		if (in_pcbrele_rlocked(inp))
262 			return (1);
263 		if (filtered) {
264 			INP_RUNLOCK(inp);
265 			return (1);
266 		}
267 	}
268 
269 	off += sizeof(struct udphdr);
270 
271 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
272 	/* Check AH/ESP integrity. */
273 	if (IPSEC_ENABLED(ipv4) &&
274 	    IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
275 		m_freem(n);
276 		return (0);
277 	}
278 	if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
279 		if (IPSEC_ENABLED(ipv4) &&
280 		    UDPENCAP_INPUT(ipv4, n, off, AF_INET) != 0)
281 			return (0);	/* Consumed. */
282 	}
283 #endif /* IPSEC */
284 #ifdef MAC
285 	if (mac_inpcb_check_deliver(inp, n) != 0) {
286 		m_freem(n);
287 		return (0);
288 	}
289 #endif /* MAC */
290 	if (inp->inp_flags & INP_CONTROLOPTS ||
291 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
292 #ifdef INET6
293 		if (inp->inp_vflag & INP_IPV6)
294 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
295 		else
296 #endif /* INET6 */
297 			ip_savecontrol(inp, &opts, ip, n);
298 	}
299 	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
300 		tmpopts = sbcreatecontrol(&udp_in[1],
301 		    sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP,
302 		    M_NOWAIT);
303 		if (tmpopts) {
304 			if (opts) {
305 				tmpopts->m_next = opts;
306 				opts = tmpopts;
307 			} else
308 				opts = tmpopts;
309 		}
310 	}
311 #ifdef INET6
312 	if (inp->inp_vflag & INP_IPV6) {
313 		bzero(&udp_in6, sizeof(udp_in6));
314 		udp_in6.sin6_len = sizeof(udp_in6);
315 		udp_in6.sin6_family = AF_INET6;
316 		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
317 		append_sa = (struct sockaddr *)&udp_in6;
318 	} else
319 #endif /* INET6 */
320 		append_sa = (struct sockaddr *)&udp_in[0];
321 	m_adj(n, off);
322 
323 	so = inp->inp_socket;
324 	SOCKBUF_LOCK(&so->so_rcv);
325 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
326 		soroverflow_locked(so);
327 		m_freem(n);
328 		if (opts)
329 			m_freem(opts);
330 		UDPSTAT_INC(udps_fullsock);
331 	} else
332 		sorwakeup_locked(so);
333 	return (0);
334 }
335 
336 static bool
337 udp_multi_match(const struct inpcb *inp, void *v)
338 {
339 	struct ip *ip = v;
340 	struct udphdr *uh = (struct udphdr *)(ip + 1);
341 
342 	if (inp->inp_lport != uh->uh_dport)
343 		return (false);
344 #ifdef INET6
345 	if ((inp->inp_vflag & INP_IPV4) == 0)
346 		return (false);
347 #endif
348 	if (inp->inp_laddr.s_addr != INADDR_ANY &&
349 	    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
350 		return (false);
351 	if (inp->inp_faddr.s_addr != INADDR_ANY &&
352 	    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
353 		return (false);
354 	if (inp->inp_fport != 0 &&
355 	    inp->inp_fport != uh->uh_sport)
356 		return (false);
357 
358 	return (true);
359 }
360 
361 static int
362 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in)
363 {
364 	struct ip *ip = mtod(m, struct ip *);
365 	struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto),
366 	    INPLOOKUP_RLOCKPCB, udp_multi_match, ip);
367 #ifdef KDTRACE_HOOKS
368 	struct udphdr *uh = (struct udphdr *)(ip + 1);
369 #endif
370 	struct inpcb *inp;
371 	struct mbuf *n;
372 	int appends = 0, fib;
373 
374 	MPASS(ip->ip_hl == sizeof(struct ip) >> 2);
375 
376 	fib = M_GETFIB(m);
377 
378 	while ((inp = inp_next(&inpi)) != NULL) {
379 		/*
380 		 * XXXRW: Because we weren't holding either the inpcb
381 		 * or the hash lock when we checked for a match
382 		 * before, we should probably recheck now that the
383 		 * inpcb lock is held.
384 		 */
385 
386 		if (V_udp_bind_all_fibs == 0 && fib != inp->inp_inc.inc_fibnum)
387 			/*
388 			 * Sockets bound to a specific FIB can only receive
389 			 * packets from that FIB.
390 			 */
391 			continue;
392 
393 		/*
394 		 * Handle socket delivery policy for any-source
395 		 * and source-specific multicast. [RFC3678]
396 		 */
397 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
398 			struct ip_moptions	*imo;
399 			struct sockaddr_in	 group;
400 			int			 blocked;
401 
402 			imo = inp->inp_moptions;
403 			if (imo == NULL)
404 				continue;
405 			bzero(&group, sizeof(struct sockaddr_in));
406 			group.sin_len = sizeof(struct sockaddr_in);
407 			group.sin_family = AF_INET;
408 			group.sin_addr = ip->ip_dst;
409 
410 			blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif,
411 				(struct sockaddr *)&group,
412 				(struct sockaddr *)&udp_in[0]);
413 			if (blocked != MCAST_PASS) {
414 				if (blocked == MCAST_NOTGMEMBER)
415 					IPSTAT_INC(ips_notmember);
416 				if (blocked == MCAST_NOTSMEMBER ||
417 				    blocked == MCAST_MUTED)
418 					UDPSTAT_INC(udps_filtermcast);
419 				continue;
420 			}
421 		}
422 		if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) {
423 			if (proto == IPPROTO_UDPLITE)
424 				UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
425 			else
426 				UDP_PROBE(receive, NULL, inp, ip, inp, uh);
427 			if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) {
428 				break;
429 			} else
430 				appends++;
431 		}
432 		/*
433 		 * Don't look for additional matches if this one does
434 		 * not have either the SO_REUSEPORT or SO_REUSEADDR
435 		 * socket options set.  This heuristic avoids
436 		 * searching through all pcbs in the common case of a
437 		 * non-shared port.  It assumes that an application
438 		 * will never clear these options after setting them.
439 		 */
440 		if ((inp->inp_socket->so_options &
441 		    (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) {
442 			INP_RUNLOCK(inp);
443 			break;
444 		}
445 	}
446 
447 	if (appends == 0) {
448 		/*
449 		 * No matching pcb found; discard datagram.  (No need
450 		 * to send an ICMP Port Unreachable for a broadcast
451 		 * or multicast datgram.)
452 		 */
453 		UDPSTAT_INC(udps_noport);
454 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
455 			UDPSTAT_INC(udps_noportmcast);
456 		else
457 			UDPSTAT_INC(udps_noportbcast);
458 	}
459 	m_freem(m);
460 
461 	return (IPPROTO_DONE);
462 }
463 
464 static int
465 udp_input(struct mbuf **mp, int *offp, int proto)
466 {
467 	struct ip *ip;
468 	struct udphdr *uh;
469 	struct ifnet *ifp;
470 	struct inpcb *inp;
471 	uint16_t len, ip_len;
472 	struct inpcbinfo *pcbinfo;
473 	struct sockaddr_in udp_in[2];
474 	struct mbuf *m;
475 	struct m_tag *fwd_tag;
476 	int cscov_partial, iphlen, lookupflags;
477 
478 	m = *mp;
479 	iphlen = *offp;
480 	ifp = m->m_pkthdr.rcvif;
481 	*mp = NULL;
482 	UDPSTAT_INC(udps_ipackets);
483 
484 	/*
485 	 * Strip IP options, if any; should skip this, make available to
486 	 * user, and use on returned packets, but we don't yet have a way to
487 	 * check the checksum with options still present.
488 	 */
489 	if (iphlen > sizeof (struct ip)) {
490 		ip_stripoptions(m);
491 		iphlen = sizeof(struct ip);
492 	}
493 
494 	/*
495 	 * Get IP and UDP header together in first mbuf.
496 	 */
497 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
498 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
499 			UDPSTAT_INC(udps_hdrops);
500 			return (IPPROTO_DONE);
501 		}
502 	}
503 	ip = mtod(m, struct ip *);
504 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
505 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
506 
507 	/*
508 	 * Destination port of 0 is illegal, based on RFC768.
509 	 */
510 	if (uh->uh_dport == 0)
511 		goto badunlocked;
512 
513 	/*
514 	 * Construct sockaddr format source address.  Stuff source address
515 	 * and datagram in user buffer.
516 	 */
517 	bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
518 	udp_in[0].sin_len = sizeof(struct sockaddr_in);
519 	udp_in[0].sin_family = AF_INET;
520 	udp_in[0].sin_port = uh->uh_sport;
521 	udp_in[0].sin_addr = ip->ip_src;
522 	udp_in[1].sin_len = sizeof(struct sockaddr_in);
523 	udp_in[1].sin_family = AF_INET;
524 	udp_in[1].sin_port = uh->uh_dport;
525 	udp_in[1].sin_addr = ip->ip_dst;
526 
527 	/*
528 	 * Make mbuf data length reflect UDP length.  If not enough data to
529 	 * reflect UDP length, drop.
530 	 */
531 	len = ntohs((u_short)uh->uh_ulen);
532 	ip_len = ntohs(ip->ip_len) - iphlen;
533 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
534 		/* Zero means checksum over the complete packet. */
535 		if (len == 0)
536 			len = ip_len;
537 		cscov_partial = 0;
538 	}
539 	if (ip_len != len) {
540 		if (len > ip_len || len < sizeof(struct udphdr)) {
541 			UDPSTAT_INC(udps_badlen);
542 			goto badunlocked;
543 		}
544 		if (proto == IPPROTO_UDP)
545 			m_adj(m, len - ip_len);
546 	}
547 
548 	/*
549 	 * Checksum extended UDP header and data.
550 	 */
551 	if (uh->uh_sum) {
552 		u_short uh_sum;
553 
554 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
555 		    !cscov_partial) {
556 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
557 				uh_sum = m->m_pkthdr.csum_data;
558 			else
559 				uh_sum = in_pseudo(ip->ip_src.s_addr,
560 				    ip->ip_dst.s_addr, htonl((u_short)len +
561 				    m->m_pkthdr.csum_data + proto));
562 			uh_sum ^= 0xffff;
563 		} else {
564 			char b[offsetof(struct ipovly, ih_src)];
565 			struct ipovly *ipov = (struct ipovly *)ip;
566 
567 			memcpy(b, ipov, sizeof(b));
568 			bzero(ipov, sizeof(ipov->ih_x1));
569 			ipov->ih_len = (proto == IPPROTO_UDP) ?
570 			    uh->uh_ulen : htons(ip_len);
571 			uh_sum = in_cksum(m, len + sizeof (struct ip));
572 			memcpy(ipov, b, sizeof(b));
573 		}
574 		if (uh_sum) {
575 			UDPSTAT_INC(udps_badsum);
576 			m_freem(m);
577 			return (IPPROTO_DONE);
578 		}
579 	} else {
580 		if (proto == IPPROTO_UDP) {
581 			UDPSTAT_INC(udps_nosum);
582 		} else {
583 			/* UDPLite requires a checksum */
584 			/* XXX: What is the right UDPLite MIB counter here? */
585 			m_freem(m);
586 			return (IPPROTO_DONE);
587 		}
588 	}
589 
590 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
591 	    in_ifnet_broadcast(ip->ip_dst, ifp))
592 		return (udp_multi_input(m, proto, udp_in));
593 
594 	pcbinfo = udp_get_inpcbinfo(proto);
595 
596 	/*
597 	 * Locate pcb for datagram.
598 	 */
599 	lookupflags = INPLOOKUP_RLOCKPCB |
600 	    (V_udp_bind_all_fibs ? 0 : INPLOOKUP_FIB);
601 
602 	/*
603 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
604 	 */
605 	if ((m->m_flags & M_IP_NEXTHOP) &&
606 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
607 		struct sockaddr_in *next_hop;
608 
609 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
610 
611 		/*
612 		 * Transparently forwarded. Pretend to be the destination.
613 		 * Already got one like this?
614 		 */
615 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
616 		    ip->ip_dst, uh->uh_dport, lookupflags, ifp, m);
617 		if (!inp) {
618 			/*
619 			 * It's new.  Try to find the ambushing socket.
620 			 * Because we've rewritten the destination address,
621 			 * any hardware-generated hash is ignored.
622 			 */
623 			inp = in_pcblookup(pcbinfo, ip->ip_src,
624 			    uh->uh_sport, next_hop->sin_addr,
625 			    next_hop->sin_port ? htons(next_hop->sin_port) :
626 			    uh->uh_dport, INPLOOKUP_WILDCARD | lookupflags,
627 			    ifp);
628 		}
629 		/* Remove the tag from the packet. We don't need it anymore. */
630 		m_tag_delete(m, fwd_tag);
631 		m->m_flags &= ~M_IP_NEXTHOP;
632 	} else
633 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
634 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
635 		    lookupflags, ifp, m);
636 	if (inp == NULL) {
637 		if (V_udp_log_in_vain) {
638 			char src[INET_ADDRSTRLEN];
639 			char dst[INET_ADDRSTRLEN];
640 
641 			log(LOG_INFO,
642 			    "Connection attempt to UDP %s:%d from %s:%d\n",
643 			    inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
644 			    inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
645 		}
646 		if (proto == IPPROTO_UDPLITE)
647 			UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
648 		else
649 			UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
650 		UDPSTAT_INC(udps_noport);
651 		if (m->m_flags & (M_BCAST | M_MCAST)) {
652 			UDPSTAT_INC(udps_noportbcast);
653 			goto badunlocked;
654 		}
655 		if (V_udp_blackhole && (V_udp_blackhole_local ||
656 		    !in_localip(ip->ip_src)))
657 			goto badunlocked;
658 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
659 			goto badunlocked;
660 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
661 		return (IPPROTO_DONE);
662 	}
663 
664 	/*
665 	 * Check the minimum TTL for socket.
666 	 */
667 	INP_RLOCK_ASSERT(inp);
668 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
669 		if (proto == IPPROTO_UDPLITE)
670 			UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
671 		else
672 			UDP_PROBE(receive, NULL, inp, ip, inp, uh);
673 		INP_RUNLOCK(inp);
674 		m_freem(m);
675 		return (IPPROTO_DONE);
676 	}
677 	if (cscov_partial) {
678 		struct udpcb *up;
679 
680 		up = intoudpcb(inp);
681 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
682 			INP_RUNLOCK(inp);
683 			m_freem(m);
684 			return (IPPROTO_DONE);
685 		}
686 	}
687 
688 	if (proto == IPPROTO_UDPLITE)
689 		UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
690 	else
691 		UDP_PROBE(receive, NULL, inp, ip, inp, uh);
692 	if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
693 		INP_RUNLOCK(inp);
694 	return (IPPROTO_DONE);
695 
696 badunlocked:
697 	m_freem(m);
698 	return (IPPROTO_DONE);
699 }
700 #endif /* INET */
701 
702 /*
703  * Notify a udp user of an asynchronous error; just wake up so that they can
704  * collect error status.
705  */
706 struct inpcb *
707 udp_notify(struct inpcb *inp, int errno)
708 {
709 
710 	INP_WLOCK_ASSERT(inp);
711 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
712 	     errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
713 		NH_FREE(inp->inp_route.ro_nh);
714 		inp->inp_route.ro_nh = (struct nhop_object *)NULL;
715 	}
716 
717 	inp->inp_socket->so_error = errno;
718 	sorwakeup(inp->inp_socket);
719 	sowwakeup(inp->inp_socket);
720 	return (inp);
721 }
722 
723 #ifdef INET
724 static void
725 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo)
726 {
727 	struct ip *ip = &icmp->icmp_ip;
728 	struct udphdr *uh;
729 	struct inpcb *inp;
730 
731 	if (icmp_errmap(icmp) == 0)
732 		return;
733 
734 	uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
735 	inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src,
736 	    uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
737 	if (inp != NULL) {
738 		INP_WLOCK_ASSERT(inp);
739 		if (inp->inp_socket != NULL)
740 			udp_notify(inp, icmp_errmap(icmp));
741 		INP_WUNLOCK(inp);
742 	} else {
743 		inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport,
744 		    ip->ip_src, uh->uh_sport,
745 		    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
746 		if (inp != NULL) {
747 			struct udpcb *up;
748 			udp_tun_icmp_t *func;
749 
750 			up = intoudpcb(inp);
751 			func = up->u_icmp_func;
752 			INP_RUNLOCK(inp);
753 			if (func != NULL)
754 				func(icmp);
755 		}
756 	}
757 }
758 
759 static void
760 udp_ctlinput(struct icmp *icmp)
761 {
762 
763 	return (udp_common_ctlinput(icmp, &V_udbinfo));
764 }
765 
766 static void
767 udplite_ctlinput(struct icmp *icmp)
768 {
769 
770 	return (udp_common_ctlinput(icmp, &V_ulitecbinfo));
771 }
772 #endif /* INET */
773 
774 static int
775 udp_pcblist(SYSCTL_HANDLER_ARGS)
776 {
777 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
778 	    INPLOOKUP_RLOCKPCB);
779 	struct xinpgen xig;
780 	struct inpcb *inp;
781 	int error;
782 
783 	if (req->newptr != 0)
784 		return (EPERM);
785 
786 	if (req->oldptr == 0) {
787 		int n;
788 
789 		n = V_udbinfo.ipi_count;
790 		n += imax(n / 8, 10);
791 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
792 		return (0);
793 	}
794 
795 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
796 		return (error);
797 
798 	bzero(&xig, sizeof(xig));
799 	xig.xig_len = sizeof xig;
800 	xig.xig_count = V_udbinfo.ipi_count;
801 	xig.xig_gen = V_udbinfo.ipi_gencnt;
802 	xig.xig_sogen = so_gencnt;
803 	error = SYSCTL_OUT(req, &xig, sizeof xig);
804 	if (error)
805 		return (error);
806 
807 	while ((inp = inp_next(&inpi)) != NULL) {
808 		if (inp->inp_gencnt <= xig.xig_gen &&
809 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
810 			struct xinpcb xi;
811 
812 			in_pcbtoxinpcb(inp, &xi);
813 			error = SYSCTL_OUT(req, &xi, sizeof xi);
814 			if (error) {
815 				INP_RUNLOCK(inp);
816 				break;
817 			}
818 		}
819 	}
820 
821 	if (!error) {
822 		/*
823 		 * Give the user an updated idea of our state.  If the
824 		 * generation differs from what we told her before, she knows
825 		 * that something happened while we were processing this
826 		 * request, and it might be necessary to retry.
827 		 */
828 		xig.xig_gen = V_udbinfo.ipi_gencnt;
829 		xig.xig_sogen = so_gencnt;
830 		xig.xig_count = V_udbinfo.ipi_count;
831 		error = SYSCTL_OUT(req, &xig, sizeof xig);
832 	}
833 
834 	return (error);
835 }
836 
837 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
838     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
839     udp_pcblist, "S,xinpcb",
840     "List of active UDP sockets");
841 
842 #ifdef INET
843 static int
844 udp_getcred(SYSCTL_HANDLER_ARGS)
845 {
846 	struct xucred xuc;
847 	struct sockaddr_in addrs[2];
848 	struct epoch_tracker et;
849 	struct inpcb *inp;
850 	int error;
851 
852 	if (req->newptr == NULL)
853 		return (EINVAL);
854 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
855 	if (error)
856 		return (error);
857 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
858 	if (error)
859 		return (error);
860 	NET_EPOCH_ENTER(et);
861 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
862 	    addrs[0].sin_addr, addrs[0].sin_port,
863 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
864 	NET_EPOCH_EXIT(et);
865 	if (inp != NULL) {
866 		INP_RLOCK_ASSERT(inp);
867 		if (inp->inp_socket == NULL)
868 			error = ENOENT;
869 		if (error == 0)
870 			error = cr_canseeinpcb(req->td->td_ucred, inp);
871 		if (error == 0)
872 			cru2x(inp->inp_cred, &xuc);
873 		INP_RUNLOCK(inp);
874 	} else
875 		error = ENOENT;
876 	if (error == 0)
877 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
878 	return (error);
879 }
880 
881 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
882     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
883     0, 0, udp_getcred, "S,xucred",
884     "Get the xucred of a UDP connection");
885 #endif /* INET */
886 
887 int
888 udp_ctloutput(struct socket *so, struct sockopt *sopt)
889 {
890 	struct inpcb *inp;
891 	struct udpcb *up;
892 	int isudplite, error, optval;
893 
894 	error = 0;
895 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
896 	inp = sotoinpcb(so);
897 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
898 	INP_WLOCK(inp);
899 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
900 #ifdef INET6
901 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
902 			INP_WUNLOCK(inp);
903 			error = ip6_ctloutput(so, sopt);
904 		}
905 #endif
906 #if defined(INET) && defined(INET6)
907 		else
908 #endif
909 #ifdef INET
910 		{
911 			INP_WUNLOCK(inp);
912 			error = ip_ctloutput(so, sopt);
913 		}
914 #endif
915 		return (error);
916 	}
917 
918 	switch (sopt->sopt_dir) {
919 	case SOPT_SET:
920 		switch (sopt->sopt_name) {
921 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
922 #if defined(INET) || defined(INET6)
923 		case UDP_ENCAP:
924 #ifdef INET
925 			if (INP_SOCKAF(so) == AF_INET) {
926 				if (!IPSEC_ENABLED(ipv4)) {
927 					INP_WUNLOCK(inp);
928 					return (ENOPROTOOPT);
929 				}
930 				error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
931 				break;
932 			}
933 #endif /* INET */
934 #ifdef INET6
935 			if (INP_SOCKAF(so) == AF_INET6) {
936 				if (!IPSEC_ENABLED(ipv6)) {
937 					INP_WUNLOCK(inp);
938 					return (ENOPROTOOPT);
939 				}
940 				error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
941 				break;
942 			}
943 #endif /* INET6 */
944 			INP_WUNLOCK(inp);
945 			return (EINVAL);
946 #endif /* INET || INET6 */
947 
948 #endif /* IPSEC */
949 		case UDPLITE_SEND_CSCOV:
950 		case UDPLITE_RECV_CSCOV:
951 			if (!isudplite) {
952 				INP_WUNLOCK(inp);
953 				error = ENOPROTOOPT;
954 				break;
955 			}
956 			INP_WUNLOCK(inp);
957 			error = sooptcopyin(sopt, &optval, sizeof(optval),
958 			    sizeof(optval));
959 			if (error != 0)
960 				break;
961 			inp = sotoinpcb(so);
962 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
963 			INP_WLOCK(inp);
964 			up = intoudpcb(inp);
965 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
966 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
967 				INP_WUNLOCK(inp);
968 				error = EINVAL;
969 				break;
970 			}
971 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
972 				up->u_txcslen = optval;
973 			else
974 				up->u_rxcslen = optval;
975 			INP_WUNLOCK(inp);
976 			break;
977 		default:
978 			INP_WUNLOCK(inp);
979 			error = ENOPROTOOPT;
980 			break;
981 		}
982 		break;
983 	case SOPT_GET:
984 		switch (sopt->sopt_name) {
985 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
986 #if defined(INET) || defined(INET6)
987 		case UDP_ENCAP:
988 #ifdef INET
989 			if (INP_SOCKAF(so) == AF_INET) {
990 				if (!IPSEC_ENABLED(ipv4)) {
991 					INP_WUNLOCK(inp);
992 					return (ENOPROTOOPT);
993 				}
994 				error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
995 				break;
996 			}
997 #endif /* INET */
998 #ifdef INET6
999 			if (INP_SOCKAF(so) == AF_INET6) {
1000 				if (!IPSEC_ENABLED(ipv6)) {
1001 					INP_WUNLOCK(inp);
1002 					return (ENOPROTOOPT);
1003 				}
1004 				error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
1005 				break;
1006 			}
1007 #endif /* INET6 */
1008 			INP_WUNLOCK(inp);
1009 			return (EINVAL);
1010 #endif /* INET || INET6 */
1011 
1012 #endif /* IPSEC */
1013 		case UDPLITE_SEND_CSCOV:
1014 		case UDPLITE_RECV_CSCOV:
1015 			if (!isudplite) {
1016 				INP_WUNLOCK(inp);
1017 				error = ENOPROTOOPT;
1018 				break;
1019 			}
1020 			up = intoudpcb(inp);
1021 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1022 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1023 				optval = up->u_txcslen;
1024 			else
1025 				optval = up->u_rxcslen;
1026 			INP_WUNLOCK(inp);
1027 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1028 			break;
1029 		default:
1030 			INP_WUNLOCK(inp);
1031 			error = ENOPROTOOPT;
1032 			break;
1033 		}
1034 		break;
1035 	}
1036 	return (error);
1037 }
1038 
1039 #ifdef INET
1040 #ifdef INET6
1041 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1042 static int
1043 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1044     struct inpcb *inp, int flags)
1045 {
1046 	struct ifnet *ifp;
1047 	struct in6_pktinfo *pktinfo;
1048 	struct in_addr ia;
1049 
1050 	NET_EPOCH_ASSERT();
1051 
1052 	if ((flags & PRUS_IPV6) == 0)
1053 		return (0);
1054 
1055 	if (cm->cmsg_level != IPPROTO_IPV6)
1056 		return (0);
1057 
1058 	if  (cm->cmsg_type != IPV6_2292PKTINFO &&
1059 	    cm->cmsg_type != IPV6_PKTINFO)
1060 		return (0);
1061 
1062 	if (cm->cmsg_len !=
1063 	    CMSG_LEN(sizeof(struct in6_pktinfo)))
1064 		return (EINVAL);
1065 
1066 	pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1067 	if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1068 	    !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1069 		return (EINVAL);
1070 
1071 	/* Validate the interface index if specified. */
1072 	if (pktinfo->ipi6_ifindex) {
1073 		ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1074 		if (ifp == NULL)
1075 			return (ENXIO);
1076 	} else
1077 		ifp = NULL;
1078 	if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1079 		ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1080 		if (!in_ifhasaddr(ifp, ia))
1081 			return (EADDRNOTAVAIL);
1082 	}
1083 
1084 	bzero(src, sizeof(*src));
1085 	src->sin_family = AF_INET;
1086 	src->sin_len = sizeof(*src);
1087 	src->sin_port = inp->inp_lport;
1088 	src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1089 
1090 	return (0);
1091 }
1092 #endif	/* INET6 */
1093 
1094 int
1095 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1096     struct mbuf *control, struct thread *td)
1097 {
1098 	struct inpcb *inp;
1099 	struct udpiphdr *ui;
1100 	int len, error = 0;
1101 	struct in_addr faddr, laddr;
1102 	struct cmsghdr *cm;
1103 	struct inpcbinfo *pcbinfo;
1104 	struct sockaddr_in *sin, src;
1105 	struct epoch_tracker et;
1106 	int cscov_partial = 0;
1107 	int ipflags = 0;
1108 	u_short fport, lport;
1109 	u_char tos, vflagsav;
1110 	uint8_t pr;
1111 	uint16_t cscov = 0;
1112 	uint32_t flowid = 0;
1113 	uint8_t flowtype = M_HASHTYPE_NONE;
1114 	bool use_cached_route;
1115 
1116 	inp = sotoinpcb(so);
1117 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1118 
1119 	if (addr != NULL) {
1120 		if (addr->sa_family != AF_INET)
1121 			error = EAFNOSUPPORT;
1122 		else if (addr->sa_len != sizeof(struct sockaddr_in))
1123 			error = EINVAL;
1124 		if (__predict_false(error != 0)) {
1125 			m_freem(control);
1126 			m_freem(m);
1127 			return (error);
1128 		}
1129 	}
1130 
1131 	len = m->m_pkthdr.len;
1132 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1133 		if (control)
1134 			m_freem(control);
1135 		m_freem(m);
1136 		return (EMSGSIZE);
1137 	}
1138 
1139 	src.sin_family = 0;
1140 	sin = (struct sockaddr_in *)addr;
1141 
1142 	/*
1143 	 * udp_send() may need to bind the current inpcb.  As such, we don't
1144 	 * know up front whether we will need the pcbinfo lock or not.  Do any
1145 	 * work to decide what is needed up front before acquiring any locks.
1146 	 *
1147 	 * We will need network epoch in either case, to safely lookup into
1148 	 * pcb hash.
1149 	 */
1150 	use_cached_route = sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0);
1151 	if (use_cached_route || (flags & PRUS_IPV6) != 0)
1152 		INP_WLOCK(inp);
1153 	else
1154 		INP_RLOCK(inp);
1155 	NET_EPOCH_ENTER(et);
1156 	tos = inp->inp_ip_tos;
1157 	if (control != NULL) {
1158 		/*
1159 		 * XXX: Currently, we assume all the optional information is
1160 		 * stored in a single mbuf.
1161 		 */
1162 		if (control->m_next) {
1163 			m_freem(control);
1164 			error = EINVAL;
1165 			goto release;
1166 		}
1167 		for (; control->m_len > 0;
1168 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1169 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1170 			cm = mtod(control, struct cmsghdr *);
1171 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1172 			    || cm->cmsg_len > control->m_len) {
1173 				error = EINVAL;
1174 				break;
1175 			}
1176 #ifdef INET6
1177 			error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1178 			if (error != 0)
1179 				break;
1180 #endif
1181 			if (cm->cmsg_level != IPPROTO_IP)
1182 				continue;
1183 
1184 			switch (cm->cmsg_type) {
1185 			case IP_SENDSRCADDR:
1186 				if (cm->cmsg_len !=
1187 				    CMSG_LEN(sizeof(struct in_addr))) {
1188 					error = EINVAL;
1189 					break;
1190 				}
1191 				bzero(&src, sizeof(src));
1192 				src.sin_family = AF_INET;
1193 				src.sin_len = sizeof(src);
1194 				src.sin_port = inp->inp_lport;
1195 				src.sin_addr =
1196 				    *(struct in_addr *)CMSG_DATA(cm);
1197 				break;
1198 
1199 			case IP_TOS:
1200 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1201 					error = EINVAL;
1202 					break;
1203 				}
1204 				tos = *(u_char *)CMSG_DATA(cm);
1205 				break;
1206 
1207 			case IP_FLOWID:
1208 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1209 					error = EINVAL;
1210 					break;
1211 				}
1212 				flowid = *(uint32_t *) CMSG_DATA(cm);
1213 				break;
1214 
1215 			case IP_FLOWTYPE:
1216 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1217 					error = EINVAL;
1218 					break;
1219 				}
1220 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1221 				break;
1222 
1223 #ifdef	RSS
1224 			case IP_RSSBUCKETID:
1225 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1226 					error = EINVAL;
1227 					break;
1228 				}
1229 				/* This is just a placeholder for now */
1230 				break;
1231 #endif	/* RSS */
1232 			default:
1233 				error = ENOPROTOOPT;
1234 				break;
1235 			}
1236 			if (error)
1237 				break;
1238 		}
1239 		m_freem(control);
1240 		control = NULL;
1241 	}
1242 	if (error)
1243 		goto release;
1244 
1245 	pr = inp->inp_socket->so_proto->pr_protocol;
1246 	pcbinfo = udp_get_inpcbinfo(pr);
1247 
1248 	/*
1249 	 * If the IP_SENDSRCADDR control message was specified, override the
1250 	 * source address for this datagram.  Its use is invalidated if the
1251 	 * address thus specified is incomplete or clobbers other inpcbs.
1252 	 */
1253 	laddr = inp->inp_laddr;
1254 	lport = inp->inp_lport;
1255 	if (src.sin_family == AF_INET) {
1256 		if ((lport == 0) ||
1257 		    (laddr.s_addr == INADDR_ANY &&
1258 		     src.sin_addr.s_addr == INADDR_ANY)) {
1259 			error = EINVAL;
1260 			goto release;
1261 		}
1262 		if ((flags & PRUS_IPV6) != 0) {
1263 			vflagsav = inp->inp_vflag;
1264 			inp->inp_vflag |= INP_IPV4;
1265 			inp->inp_vflag &= ~INP_IPV6;
1266 		}
1267 		INP_HASH_WLOCK(pcbinfo);
1268 		error = in_pcbbind_setup(inp, &src, &laddr.s_addr, &lport,
1269 		    V_udp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred);
1270 		INP_HASH_WUNLOCK(pcbinfo);
1271 		if ((flags & PRUS_IPV6) != 0)
1272 			inp->inp_vflag = vflagsav;
1273 		if (error)
1274 			goto release;
1275 	}
1276 
1277 	/*
1278 	 * If a UDP socket has been connected, then a local address/port will
1279 	 * have been selected and bound.
1280 	 *
1281 	 * If a UDP socket has not been connected to, then an explicit
1282 	 * destination address must be used, in which case a local
1283 	 * address/port may not have been selected and bound.
1284 	 */
1285 	if (sin != NULL) {
1286 		INP_LOCK_ASSERT(inp);
1287 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1288 			error = EISCONN;
1289 			goto release;
1290 		}
1291 
1292 		/*
1293 		 * Jail may rewrite the destination address, so let it do
1294 		 * that before we use it.
1295 		 */
1296 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1297 		if (error)
1298 			goto release;
1299 		/*
1300 		 * sendto(2) on unconnected UDP socket results in implicit
1301 		 * binding to INADDR_ANY and anonymous port.  This has two
1302 		 * side effects:
1303 		 * 1) after first sendto(2) the socket will receive datagrams
1304 		 *    destined to the selected port.
1305 		 * 2) subsequent sendto(2) calls will use the same source port.
1306 		 */
1307 		if (inp->inp_lport == 0) {
1308 			struct sockaddr_in wild = {
1309 				.sin_family = AF_INET,
1310 				.sin_len = sizeof(struct sockaddr_in),
1311 			};
1312 
1313 			INP_HASH_WLOCK(pcbinfo);
1314 			error = in_pcbbind(inp, &wild, V_udp_bind_all_fibs ?
1315 			    0 : INPBIND_FIB, td->td_ucred);
1316 			INP_HASH_WUNLOCK(pcbinfo);
1317 			if (error)
1318 				goto release;
1319 			lport = inp->inp_lport;
1320 			laddr = inp->inp_laddr;
1321 		}
1322 		if (laddr.s_addr == INADDR_ANY) {
1323 			error = in_pcbladdr(inp, &sin->sin_addr, &laddr,
1324 			    td->td_ucred);
1325 			if (error)
1326 				goto release;
1327 		}
1328 		faddr = sin->sin_addr;
1329 		fport = sin->sin_port;
1330 	} else {
1331 		INP_LOCK_ASSERT(inp);
1332 		faddr = inp->inp_faddr;
1333 		fport = inp->inp_fport;
1334 		if (faddr.s_addr == INADDR_ANY) {
1335 			error = ENOTCONN;
1336 			goto release;
1337 		}
1338 	}
1339 
1340 	/*
1341 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1342 	 * link-layer headers.  Immediate slide the data pointer back forward
1343 	 * since we won't use that space at this layer.
1344 	 */
1345 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1346 	if (m == NULL) {
1347 		error = ENOBUFS;
1348 		goto release;
1349 	}
1350 	m->m_data += max_linkhdr;
1351 	m->m_len -= max_linkhdr;
1352 	m->m_pkthdr.len -= max_linkhdr;
1353 
1354 	/*
1355 	 * Fill in mbuf with extended UDP header and addresses and length put
1356 	 * into network format.
1357 	 */
1358 	ui = mtod(m, struct udpiphdr *);
1359 	/*
1360 	 * Filling only those fields of udpiphdr that participate in the
1361 	 * checksum calculation. The rest must be zeroed and will be filled
1362 	 * later.
1363 	 */
1364 	bzero(ui->ui_x1, sizeof(ui->ui_x1));
1365 	ui->ui_pr = pr;
1366 	ui->ui_src = laddr;
1367 	ui->ui_dst = faddr;
1368 	ui->ui_sport = lport;
1369 	ui->ui_dport = fport;
1370 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1371 	if (pr == IPPROTO_UDPLITE) {
1372 		struct udpcb *up;
1373 		uint16_t plen;
1374 
1375 		up = intoudpcb(inp);
1376 		cscov = up->u_txcslen;
1377 		plen = (u_short)len + sizeof(struct udphdr);
1378 		if (cscov >= plen)
1379 			cscov = 0;
1380 		ui->ui_len = htons(plen);
1381 		ui->ui_ulen = htons(cscov);
1382 		/*
1383 		 * For UDP-Lite, checksum coverage length of zero means
1384 		 * the entire UDPLite packet is covered by the checksum.
1385 		 */
1386 		cscov_partial = (cscov == 0) ? 0 : 1;
1387 	}
1388 
1389 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1390 		ipflags |= IP_ROUTETOIF;
1391 	if (inp->inp_socket->so_options & SO_BROADCAST)
1392 		ipflags |= IP_ALLOWBROADCAST;
1393 	if (inp->inp_flags & INP_ONESBCAST)
1394 		ipflags |= IP_SENDONES;
1395 
1396 #ifdef MAC
1397 	mac_inpcb_create_mbuf(inp, m);
1398 #endif
1399 
1400 	/*
1401 	 * Set up checksum and output datagram.
1402 	 */
1403 	ui->ui_sum = 0;
1404 	if (pr == IPPROTO_UDPLITE) {
1405 		if (inp->inp_flags & INP_ONESBCAST)
1406 			faddr.s_addr = INADDR_BROADCAST;
1407 		if (cscov_partial) {
1408 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1409 				ui->ui_sum = 0xffff;
1410 		} else {
1411 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1412 				ui->ui_sum = 0xffff;
1413 		}
1414 	} else if (V_udp_cksum) {
1415 		if (inp->inp_flags & INP_ONESBCAST)
1416 			faddr.s_addr = INADDR_BROADCAST;
1417 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1418 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1419 		m->m_pkthdr.csum_flags = CSUM_UDP;
1420 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1421 	}
1422 	/*
1423 	 * After finishing the checksum computation, fill the remaining fields
1424 	 * of udpiphdr.
1425 	 */
1426 	((struct ip *)ui)->ip_v = IPVERSION;
1427 	((struct ip *)ui)->ip_tos = tos;
1428 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1429 	if (inp->inp_flags & INP_DONTFRAG)
1430 		((struct ip *)ui)->ip_off |= htons(IP_DF);
1431 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;
1432 	UDPSTAT_INC(udps_opackets);
1433 
1434 	/*
1435 	 * Setup flowid / RSS information for outbound socket.
1436 	 *
1437 	 * Once the UDP code decides to set a flowid some other way,
1438 	 * this allows the flowid to be overridden by userland.
1439 	 */
1440 	if (flowtype != M_HASHTYPE_NONE) {
1441 		m->m_pkthdr.flowid = flowid;
1442 		M_HASHTYPE_SET(m, flowtype);
1443 	}
1444 #if defined(ROUTE_MPATH) || defined(RSS)
1445 	else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1446 		uint32_t hash_val, hash_type;
1447 
1448 		hash_val = fib4_calc_packet_hash(laddr, faddr,
1449 		    lport, fport, pr, &hash_type);
1450 		m->m_pkthdr.flowid = hash_val;
1451 		M_HASHTYPE_SET(m, hash_type);
1452 	}
1453 
1454 	/*
1455 	 * Don't override with the inp cached flowid value.
1456 	 *
1457 	 * Depending upon the kind of send being done, the inp
1458 	 * flowid/flowtype values may actually not be appropriate
1459 	 * for this particular socket send.
1460 	 *
1461 	 * We should either leave the flowid at zero (which is what is
1462 	 * currently done) or set it to some software generated
1463 	 * hash value based on the packet contents.
1464 	 */
1465 	ipflags |= IP_NODEFAULTFLOWID;
1466 #endif	/* RSS */
1467 
1468 	if (pr == IPPROTO_UDPLITE)
1469 		UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1470 	else
1471 		UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1472 	error = ip_output(m, inp->inp_options,
1473 	    use_cached_route ? &inp->inp_route : NULL, ipflags,
1474 	    inp->inp_moptions, inp);
1475 	INP_UNLOCK(inp);
1476 	NET_EPOCH_EXIT(et);
1477 	return (error);
1478 
1479 release:
1480 	INP_UNLOCK(inp);
1481 	NET_EPOCH_EXIT(et);
1482 	m_freem(m);
1483 	return (error);
1484 }
1485 
1486 void
1487 udp_abort(struct socket *so)
1488 {
1489 	struct inpcb *inp;
1490 	struct inpcbinfo *pcbinfo;
1491 
1492 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1493 	inp = sotoinpcb(so);
1494 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1495 	INP_WLOCK(inp);
1496 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1497 		INP_HASH_WLOCK(pcbinfo);
1498 		in_pcbdisconnect(inp);
1499 		INP_HASH_WUNLOCK(pcbinfo);
1500 		soisdisconnected(so);
1501 	}
1502 	INP_WUNLOCK(inp);
1503 }
1504 
1505 static int
1506 udp_attach(struct socket *so, int proto, struct thread *td)
1507 {
1508 	static uint32_t udp_flowid;
1509 	struct inpcbinfo *pcbinfo;
1510 	struct inpcb *inp;
1511 	struct udpcb *up;
1512 	int error;
1513 
1514 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1515 	inp = sotoinpcb(so);
1516 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1517 	error = soreserve(so, udp_sendspace, udp_recvspace);
1518 	if (error)
1519 		return (error);
1520 	error = in_pcballoc(so, pcbinfo);
1521 	if (error)
1522 		return (error);
1523 
1524 	inp = sotoinpcb(so);
1525 	inp->inp_ip_ttl = V_ip_defttl;
1526 	inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1527 	inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1528 	up = intoudpcb(inp);
1529 	bzero(&up->u_start_zero, u_zero_size);
1530 	INP_WUNLOCK(inp);
1531 
1532 	return (0);
1533 }
1534 #endif /* INET */
1535 
1536 int
1537 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1538 {
1539 	struct inpcb *inp;
1540 	struct udpcb *up;
1541 
1542 	KASSERT(so->so_type == SOCK_DGRAM,
1543 	    ("udp_set_kernel_tunneling: !dgram"));
1544 	inp = sotoinpcb(so);
1545 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1546 	INP_WLOCK(inp);
1547 	up = intoudpcb(inp);
1548 	if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1549 	    (up->u_icmp_func != NULL))) {
1550 		INP_WUNLOCK(inp);
1551 		return (EBUSY);
1552 	}
1553 	up->u_tun_func = f;
1554 	up->u_icmp_func = i;
1555 	up->u_tun_ctx = ctx;
1556 	INP_WUNLOCK(inp);
1557 	return (0);
1558 }
1559 
1560 #ifdef INET
1561 static int
1562 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1563 {
1564 	struct inpcb *inp;
1565 	struct inpcbinfo *pcbinfo;
1566 	struct sockaddr_in *sinp;
1567 	int error;
1568 
1569 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1570 	inp = sotoinpcb(so);
1571 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1572 
1573 	sinp = (struct sockaddr_in *)nam;
1574 	if (nam->sa_family != AF_INET) {
1575 		/*
1576 		 * Preserve compatibility with old programs.
1577 		 */
1578 		if (nam->sa_family != AF_UNSPEC ||
1579 		    nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1580 		    sinp->sin_addr.s_addr != INADDR_ANY)
1581 			return (EAFNOSUPPORT);
1582 		nam->sa_family = AF_INET;
1583 	}
1584 	if (nam->sa_len != sizeof(struct sockaddr_in))
1585 		return (EINVAL);
1586 
1587 	INP_WLOCK(inp);
1588 	INP_HASH_WLOCK(pcbinfo);
1589 	error = in_pcbbind(inp, sinp, V_udp_bind_all_fibs ? 0 : INPBIND_FIB,
1590 	    td->td_ucred);
1591 	INP_HASH_WUNLOCK(pcbinfo);
1592 	INP_WUNLOCK(inp);
1593 	return (error);
1594 }
1595 
1596 static void
1597 udp_close(struct socket *so)
1598 {
1599 	struct inpcb *inp;
1600 	struct inpcbinfo *pcbinfo;
1601 
1602 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1603 	inp = sotoinpcb(so);
1604 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1605 	INP_WLOCK(inp);
1606 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1607 		INP_HASH_WLOCK(pcbinfo);
1608 		in_pcbdisconnect(inp);
1609 		INP_HASH_WUNLOCK(pcbinfo);
1610 		soisdisconnected(so);
1611 	}
1612 	INP_WUNLOCK(inp);
1613 }
1614 
1615 static int
1616 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1617 {
1618 	struct epoch_tracker et;
1619 	struct inpcb *inp;
1620 	struct inpcbinfo *pcbinfo;
1621 	struct sockaddr_in *sin;
1622 	int error;
1623 
1624 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1625 	inp = sotoinpcb(so);
1626 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1627 
1628 	sin = (struct sockaddr_in *)nam;
1629 	if (sin->sin_family != AF_INET)
1630 		return (EAFNOSUPPORT);
1631 	if (sin->sin_len != sizeof(*sin))
1632 		return (EINVAL);
1633 
1634 	INP_WLOCK(inp);
1635 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1636 		INP_WUNLOCK(inp);
1637 		return (EISCONN);
1638 	}
1639 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1640 	if (error != 0) {
1641 		INP_WUNLOCK(inp);
1642 		return (error);
1643 	}
1644 	NET_EPOCH_ENTER(et);
1645 	INP_HASH_WLOCK(pcbinfo);
1646 	error = in_pcbconnect(inp, sin, td->td_ucred);
1647 	INP_HASH_WUNLOCK(pcbinfo);
1648 	NET_EPOCH_EXIT(et);
1649 	if (error == 0)
1650 		soisconnected(so);
1651 	INP_WUNLOCK(inp);
1652 	return (error);
1653 }
1654 
1655 static void
1656 udp_detach(struct socket *so)
1657 {
1658 	struct inpcb *inp;
1659 
1660 	inp = sotoinpcb(so);
1661 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1662 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1663 	    ("udp_detach: not disconnected"));
1664 	INP_WLOCK(inp);
1665 	in_pcbfree(inp);
1666 }
1667 
1668 int
1669 udp_disconnect(struct socket *so)
1670 {
1671 	struct inpcb *inp;
1672 	struct inpcbinfo *pcbinfo;
1673 
1674 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1675 	inp = sotoinpcb(so);
1676 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1677 	INP_WLOCK(inp);
1678 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1679 		INP_WUNLOCK(inp);
1680 		return (ENOTCONN);
1681 	}
1682 	INP_HASH_WLOCK(pcbinfo);
1683 	in_pcbdisconnect(inp);
1684 	INP_HASH_WUNLOCK(pcbinfo);
1685 	SOCK_LOCK(so);
1686 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1687 	SOCK_UNLOCK(so);
1688 	INP_WUNLOCK(inp);
1689 	return (0);
1690 }
1691 #endif /* INET */
1692 
1693 int
1694 udp_shutdown(struct socket *so, enum shutdown_how how)
1695 {
1696 	int error;
1697 
1698 	SOCK_LOCK(so);
1699 	if (!(so->so_state & SS_ISCONNECTED))
1700 		/*
1701 		 * POSIX mandates us to just return ENOTCONN when shutdown(2) is
1702 		 * invoked on a datagram sockets, however historically we would
1703 		 * actually tear socket down.  This is known to be leveraged by
1704 		 * some applications to unblock process waiting in recv(2) by
1705 		 * other process that it shares that socket with.  Try to meet
1706 		 * both backward-compatibility and POSIX requirements by forcing
1707 		 * ENOTCONN but still flushing buffers and performing wakeup(9).
1708 		 *
1709 		 * XXXGL: it remains unknown what applications expect this
1710 		 * behavior and is this isolated to unix/dgram or inet/dgram or
1711 		 * both.  See: D10351, D3039.
1712 		 */
1713 		error = ENOTCONN;
1714 	else
1715 		error = 0;
1716 	SOCK_UNLOCK(so);
1717 
1718 	switch (how) {
1719 	case SHUT_RD:
1720 		sorflush(so);
1721 		break;
1722 	case SHUT_RDWR:
1723 		sorflush(so);
1724 		/* FALLTHROUGH */
1725 	case SHUT_WR:
1726 		socantsendmore(so);
1727 	}
1728 
1729 	return (error);
1730 }
1731 
1732 #ifdef INET
1733 #define	UDP_PROTOSW							\
1734 	.pr_type =		SOCK_DGRAM,				\
1735 	.pr_flags =		PR_ATOMIC | PR_ADDR | PR_CAPATTACH,	\
1736 	.pr_ctloutput =		udp_ctloutput,				\
1737 	.pr_abort =		udp_abort,				\
1738 	.pr_attach =		udp_attach,				\
1739 	.pr_bind =		udp_bind,				\
1740 	.pr_connect =		udp_connect,				\
1741 	.pr_control =		in_control,				\
1742 	.pr_detach =		udp_detach,				\
1743 	.pr_disconnect =	udp_disconnect,				\
1744 	.pr_peeraddr =		in_getpeeraddr,				\
1745 	.pr_send =		udp_send,				\
1746 	.pr_soreceive =		soreceive_dgram,			\
1747 	.pr_sosend =		sosend_dgram,				\
1748 	.pr_shutdown =		udp_shutdown,				\
1749 	.pr_sockaddr =		in_getsockaddr,				\
1750 	.pr_sosetlabel =	in_pcbsosetlabel,			\
1751 	.pr_close =		udp_close
1752 
1753 struct protosw udp_protosw = {
1754 	.pr_protocol =		IPPROTO_UDP,
1755 	UDP_PROTOSW
1756 };
1757 
1758 struct protosw udplite_protosw = {
1759 	.pr_protocol =		IPPROTO_UDPLITE,
1760 	UDP_PROTOSW
1761 };
1762 
1763 static void
1764 udp_init(void *arg __unused)
1765 {
1766 
1767 	IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput);
1768 	IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput);
1769 }
1770 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
1771 #endif /* INET */
1772