xref: /freebsd/sys/netinet/udp_usrreq.c (revision b11aebff4d4a301bfeac516b5be45080c8a11142)
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 VNET_DEFINE(bool, udp_blackhole_local) = false;
131 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
132     CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false,
133     "Enforce net.inet.udp.blackhole for locally originated packets");
134 
135 u_long	udp_sendspace = 9216;		/* really max datagram size */
136 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
137     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
138 
139 u_long	udp_recvspace = 40 * (1024 +
140 #ifdef INET6
141 				      sizeof(struct sockaddr_in6)
142 #else
143 				      sizeof(struct sockaddr_in)
144 #endif
145 				      );	/* 40 1K datagrams */
146 
147 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
148     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
149 
150 VNET_DEFINE(struct inpcbinfo, udbinfo);
151 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
152 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone);
153 #define	V_udpcb_zone			VNET(udpcb_zone)
154 
155 #ifndef UDBHASHSIZE
156 #define	UDBHASHSIZE	128
157 #endif
158 
159 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
160 VNET_PCPUSTAT_SYSINIT(udpstat);
161 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
162     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
163 
164 #ifdef VIMAGE
165 VNET_PCPUSTAT_SYSUNINIT(udpstat);
166 #endif /* VIMAGE */
167 #ifdef INET
168 static void	udp_detach(struct socket *so);
169 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
170 		    struct mbuf *, struct thread *, int);
171 #endif
172 
173 INPCBSTORAGE_DEFINE(udpcbstor, inpcb, "udpinp", "udp_inpcb", "udp", "udphash");
174 INPCBSTORAGE_DEFINE(udplitecbstor, inpcb, "udpliteinp", "udplite_inpcb",
175     "udplite", "udplitehash");
176 
177 static void
178 udp_vnet_init(void *arg __unused)
179 {
180 
181 	/*
182 	 * For now default to 2-tuple UDP hashing - until the fragment
183 	 * reassembly code can also update the flowid.
184 	 *
185 	 * Once we can calculate the flowid that way and re-establish
186 	 * a 4-tuple, flip this to 4-tuple.
187 	 */
188 	in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE);
189 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
190 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
191 	uma_zone_set_max(V_udpcb_zone, maxsockets);
192 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
193 
194 	/* Additional pcbinfo for UDP-Lite */
195 	in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE,
196 	    UDBHASHSIZE);
197 }
198 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
199     udp_vnet_init, NULL);
200 
201 /*
202  * Kernel module interface for updating udpstat.  The argument is an index
203  * into udpstat treated as an array of u_long.  While this encodes the
204  * general layout of udpstat into the caller, it doesn't encode its location,
205  * so that future changes to add, for example, per-CPU stats support won't
206  * cause binary compatibility problems for kernel modules.
207  */
208 void
209 kmod_udpstat_inc(int statnum)
210 {
211 
212 	counter_u64_add(VNET(udpstat)[statnum], 1);
213 }
214 
215 int
216 udp_newudpcb(struct inpcb *inp)
217 {
218 	struct udpcb *up;
219 
220 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
221 	if (up == NULL)
222 		return (ENOBUFS);
223 	inp->inp_ppcb = up;
224 	return (0);
225 }
226 
227 void
228 udp_discardcb(struct udpcb *up)
229 {
230 
231 	uma_zfree(V_udpcb_zone, up);
232 }
233 
234 #ifdef VIMAGE
235 static void
236 udp_destroy(void *unused __unused)
237 {
238 
239 	in_pcbinfo_destroy(&V_udbinfo);
240 	in_pcbinfo_destroy(&V_ulitecbinfo);
241 	uma_zdestroy(V_udpcb_zone);
242 }
243 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
244 #endif
245 
246 #ifdef INET
247 /*
248  * Subroutine of udp_input(), which appends the provided mbuf chain to the
249  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
250  * contains the source address.  If the socket ends up being an IPv6 socket,
251  * udp_append() will convert to a sockaddr_in6 before passing the address
252  * into the socket code.
253  *
254  * In the normal case udp_append() will return 0, indicating that you
255  * must unlock the inp. However if a tunneling protocol is in place we increment
256  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
257  * then decrement the reference count. If the inp_rele returns 1, indicating the
258  * inp is gone, we return that to the caller to tell them *not* to unlock
259  * the inp. In the case of multi-cast this will cause the distribution
260  * to stop (though most tunneling protocols known currently do *not* use
261  * multicast).
262  */
263 static int
264 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
265     struct sockaddr_in *udp_in)
266 {
267 	struct sockaddr *append_sa;
268 	struct socket *so;
269 	struct mbuf *tmpopts, *opts = NULL;
270 #ifdef INET6
271 	struct sockaddr_in6 udp_in6;
272 #endif
273 	struct udpcb *up;
274 	bool filtered;
275 
276 	INP_LOCK_ASSERT(inp);
277 
278 	/*
279 	 * Engage the tunneling protocol.
280 	 */
281 	up = intoudpcb(inp);
282 	if (up->u_tun_func != NULL) {
283 		in_pcbref(inp);
284 		INP_RUNLOCK(inp);
285 		filtered = (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
286 		    up->u_tun_ctx);
287 		INP_RLOCK(inp);
288 		if (filtered)
289 			return (in_pcbrele_rlocked(inp));
290 	}
291 
292 	off += sizeof(struct udphdr);
293 
294 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
295 	/* Check AH/ESP integrity. */
296 	if (IPSEC_ENABLED(ipv4) &&
297 	    IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
298 		m_freem(n);
299 		return (0);
300 	}
301 	if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
302 		if (IPSEC_ENABLED(ipv4) &&
303 		    UDPENCAP_INPUT(n, off, AF_INET) != 0)
304 			return (0);	/* Consumed. */
305 	}
306 #endif /* IPSEC */
307 #ifdef MAC
308 	if (mac_inpcb_check_deliver(inp, n) != 0) {
309 		m_freem(n);
310 		return (0);
311 	}
312 #endif /* MAC */
313 	if (inp->inp_flags & INP_CONTROLOPTS ||
314 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
315 #ifdef INET6
316 		if (inp->inp_vflag & INP_IPV6)
317 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
318 		else
319 #endif /* INET6 */
320 			ip_savecontrol(inp, &opts, ip, n);
321 	}
322 	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
323 		tmpopts = sbcreatecontrol(&udp_in[1],
324 		    sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP,
325 		    M_NOWAIT);
326 		if (tmpopts) {
327 			if (opts) {
328 				tmpopts->m_next = opts;
329 				opts = tmpopts;
330 			} else
331 				opts = tmpopts;
332 		}
333 	}
334 #ifdef INET6
335 	if (inp->inp_vflag & INP_IPV6) {
336 		bzero(&udp_in6, sizeof(udp_in6));
337 		udp_in6.sin6_len = sizeof(udp_in6);
338 		udp_in6.sin6_family = AF_INET6;
339 		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
340 		append_sa = (struct sockaddr *)&udp_in6;
341 	} else
342 #endif /* INET6 */
343 		append_sa = (struct sockaddr *)&udp_in[0];
344 	m_adj(n, off);
345 
346 	so = inp->inp_socket;
347 	SOCKBUF_LOCK(&so->so_rcv);
348 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
349 		soroverflow_locked(so);
350 		m_freem(n);
351 		if (opts)
352 			m_freem(opts);
353 		UDPSTAT_INC(udps_fullsock);
354 	} else
355 		sorwakeup_locked(so);
356 	return (0);
357 }
358 
359 static bool
360 udp_multi_match(const struct inpcb *inp, void *v)
361 {
362 	struct ip *ip = v;
363 	struct udphdr *uh = (struct udphdr *)(ip + 1);
364 
365 	if (inp->inp_lport != uh->uh_dport)
366 		return (false);
367 #ifdef INET6
368 	if ((inp->inp_vflag & INP_IPV4) == 0)
369 		return (false);
370 #endif
371 	if (inp->inp_laddr.s_addr != INADDR_ANY &&
372 	    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
373 		return (false);
374 	if (inp->inp_faddr.s_addr != INADDR_ANY &&
375 	    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
376 		return (false);
377 	if (inp->inp_fport != 0 &&
378 	    inp->inp_fport != uh->uh_sport)
379 		return (false);
380 
381 	return (true);
382 }
383 
384 static int
385 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in)
386 {
387 	struct ip *ip = mtod(m, struct ip *);
388 	struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto),
389 	    INPLOOKUP_RLOCKPCB, udp_multi_match, ip);
390 #ifdef KDTRACE_HOOKS
391 	struct udphdr *uh = (struct udphdr *)(ip + 1);
392 #endif
393 	struct inpcb *inp;
394 	struct mbuf *n;
395 	int appends = 0;
396 
397 	MPASS(ip->ip_hl == sizeof(struct ip) >> 2);
398 
399 	while ((inp = inp_next(&inpi)) != NULL) {
400 		/*
401 		 * XXXRW: Because we weren't holding either the inpcb
402 		 * or the hash lock when we checked for a match
403 		 * before, we should probably recheck now that the
404 		 * inpcb lock is held.
405 		 */
406 		/*
407 		 * Handle socket delivery policy for any-source
408 		 * and source-specific multicast. [RFC3678]
409 		 */
410 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
411 			struct ip_moptions	*imo;
412 			struct sockaddr_in	 group;
413 			int			 blocked;
414 
415 			imo = inp->inp_moptions;
416 			if (imo == NULL)
417 				continue;
418 			bzero(&group, sizeof(struct sockaddr_in));
419 			group.sin_len = sizeof(struct sockaddr_in);
420 			group.sin_family = AF_INET;
421 			group.sin_addr = ip->ip_dst;
422 
423 			blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif,
424 				(struct sockaddr *)&group,
425 				(struct sockaddr *)&udp_in[0]);
426 			if (blocked != MCAST_PASS) {
427 				if (blocked == MCAST_NOTGMEMBER)
428 					IPSTAT_INC(ips_notmember);
429 				if (blocked == MCAST_NOTSMEMBER ||
430 				    blocked == MCAST_MUTED)
431 					UDPSTAT_INC(udps_filtermcast);
432 				continue;
433 			}
434 		}
435 		if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) {
436 			if (proto == IPPROTO_UDPLITE)
437 				UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
438 			else
439 				UDP_PROBE(receive, NULL, inp, ip, inp, uh);
440 			if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) {
441 				break;
442 			} else
443 				appends++;
444 		}
445 		/*
446 		 * Don't look for additional matches if this one does
447 		 * not have either the SO_REUSEPORT or SO_REUSEADDR
448 		 * socket options set.  This heuristic avoids
449 		 * searching through all pcbs in the common case of a
450 		 * non-shared port.  It assumes that an application
451 		 * will never clear these options after setting them.
452 		 */
453 		if ((inp->inp_socket->so_options &
454 		    (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) {
455 			INP_RUNLOCK(inp);
456 			break;
457 		}
458 	}
459 
460 	if (appends == 0) {
461 		/*
462 		 * No matching pcb found; discard datagram.  (No need
463 		 * to send an ICMP Port Unreachable for a broadcast
464 		 * or multicast datgram.)
465 		 */
466 		UDPSTAT_INC(udps_noport);
467 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
468 			UDPSTAT_INC(udps_noportmcast);
469 		else
470 			UDPSTAT_INC(udps_noportbcast);
471 	}
472 	m_freem(m);
473 
474 	return (IPPROTO_DONE);
475 }
476 
477 static int
478 udp_input(struct mbuf **mp, int *offp, int proto)
479 {
480 	struct ip *ip;
481 	struct udphdr *uh;
482 	struct ifnet *ifp;
483 	struct inpcb *inp;
484 	uint16_t len, ip_len;
485 	struct inpcbinfo *pcbinfo;
486 	struct sockaddr_in udp_in[2];
487 	struct mbuf *m;
488 	struct m_tag *fwd_tag;
489 	int cscov_partial, iphlen;
490 
491 	m = *mp;
492 	iphlen = *offp;
493 	ifp = m->m_pkthdr.rcvif;
494 	*mp = NULL;
495 	UDPSTAT_INC(udps_ipackets);
496 
497 	/*
498 	 * Strip IP options, if any; should skip this, make available to
499 	 * user, and use on returned packets, but we don't yet have a way to
500 	 * check the checksum with options still present.
501 	 */
502 	if (iphlen > sizeof (struct ip)) {
503 		ip_stripoptions(m);
504 		iphlen = sizeof(struct ip);
505 	}
506 
507 	/*
508 	 * Get IP and UDP header together in first mbuf.
509 	 */
510 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
511 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
512 			UDPSTAT_INC(udps_hdrops);
513 			return (IPPROTO_DONE);
514 		}
515 	}
516 	ip = mtod(m, struct ip *);
517 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
518 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
519 
520 	/*
521 	 * Destination port of 0 is illegal, based on RFC768.
522 	 */
523 	if (uh->uh_dport == 0)
524 		goto badunlocked;
525 
526 	/*
527 	 * Construct sockaddr format source address.  Stuff source address
528 	 * and datagram in user buffer.
529 	 */
530 	bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
531 	udp_in[0].sin_len = sizeof(struct sockaddr_in);
532 	udp_in[0].sin_family = AF_INET;
533 	udp_in[0].sin_port = uh->uh_sport;
534 	udp_in[0].sin_addr = ip->ip_src;
535 	udp_in[1].sin_len = sizeof(struct sockaddr_in);
536 	udp_in[1].sin_family = AF_INET;
537 	udp_in[1].sin_port = uh->uh_dport;
538 	udp_in[1].sin_addr = ip->ip_dst;
539 
540 	/*
541 	 * Make mbuf data length reflect UDP length.  If not enough data to
542 	 * reflect UDP length, drop.
543 	 */
544 	len = ntohs((u_short)uh->uh_ulen);
545 	ip_len = ntohs(ip->ip_len) - iphlen;
546 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
547 		/* Zero means checksum over the complete packet. */
548 		if (len == 0)
549 			len = ip_len;
550 		cscov_partial = 0;
551 	}
552 	if (ip_len != len) {
553 		if (len > ip_len || len < sizeof(struct udphdr)) {
554 			UDPSTAT_INC(udps_badlen);
555 			goto badunlocked;
556 		}
557 		if (proto == IPPROTO_UDP)
558 			m_adj(m, len - ip_len);
559 	}
560 
561 	/*
562 	 * Checksum extended UDP header and data.
563 	 */
564 	if (uh->uh_sum) {
565 		u_short uh_sum;
566 
567 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
568 		    !cscov_partial) {
569 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
570 				uh_sum = m->m_pkthdr.csum_data;
571 			else
572 				uh_sum = in_pseudo(ip->ip_src.s_addr,
573 				    ip->ip_dst.s_addr, htonl((u_short)len +
574 				    m->m_pkthdr.csum_data + proto));
575 			uh_sum ^= 0xffff;
576 		} else {
577 			char b[offsetof(struct ipovly, ih_src)];
578 			struct ipovly *ipov = (struct ipovly *)ip;
579 
580 			bcopy(ipov, b, sizeof(b));
581 			bzero(ipov, sizeof(ipov->ih_x1));
582 			ipov->ih_len = (proto == IPPROTO_UDP) ?
583 			    uh->uh_ulen : htons(ip_len);
584 			uh_sum = in_cksum(m, len + sizeof (struct ip));
585 			bcopy(b, ipov, sizeof(b));
586 		}
587 		if (uh_sum) {
588 			UDPSTAT_INC(udps_badsum);
589 			m_freem(m);
590 			return (IPPROTO_DONE);
591 		}
592 	} else {
593 		if (proto == IPPROTO_UDP) {
594 			UDPSTAT_INC(udps_nosum);
595 		} else {
596 			/* UDPLite requires a checksum */
597 			/* XXX: What is the right UDPLite MIB counter here? */
598 			m_freem(m);
599 			return (IPPROTO_DONE);
600 		}
601 	}
602 
603 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
604 	    in_broadcast(ip->ip_dst, ifp))
605 		return (udp_multi_input(m, proto, udp_in));
606 
607 	pcbinfo = udp_get_inpcbinfo(proto);
608 
609 	/*
610 	 * Locate pcb for datagram.
611 	 *
612 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
613 	 */
614 	if ((m->m_flags & M_IP_NEXTHOP) &&
615 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
616 		struct sockaddr_in *next_hop;
617 
618 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
619 
620 		/*
621 		 * Transparently forwarded. Pretend to be the destination.
622 		 * Already got one like this?
623 		 */
624 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
625 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
626 		if (!inp) {
627 			/*
628 			 * It's new.  Try to find the ambushing socket.
629 			 * Because we've rewritten the destination address,
630 			 * any hardware-generated hash is ignored.
631 			 */
632 			inp = in_pcblookup(pcbinfo, ip->ip_src,
633 			    uh->uh_sport, next_hop->sin_addr,
634 			    next_hop->sin_port ? htons(next_hop->sin_port) :
635 			    uh->uh_dport, INPLOOKUP_WILDCARD |
636 			    INPLOOKUP_RLOCKPCB, ifp);
637 		}
638 		/* Remove the tag from the packet. We don't need it anymore. */
639 		m_tag_delete(m, fwd_tag);
640 		m->m_flags &= ~M_IP_NEXTHOP;
641 	} else
642 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
643 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
644 		    INPLOOKUP_RLOCKPCB, ifp, m);
645 	if (inp == NULL) {
646 		if (V_udp_log_in_vain) {
647 			char src[INET_ADDRSTRLEN];
648 			char dst[INET_ADDRSTRLEN];
649 
650 			log(LOG_INFO,
651 			    "Connection attempt to UDP %s:%d from %s:%d\n",
652 			    inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
653 			    inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
654 		}
655 		if (proto == IPPROTO_UDPLITE)
656 			UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
657 		else
658 			UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
659 		UDPSTAT_INC(udps_noport);
660 		if (m->m_flags & (M_BCAST | M_MCAST)) {
661 			UDPSTAT_INC(udps_noportbcast);
662 			goto badunlocked;
663 		}
664 		if (V_udp_blackhole && (V_udp_blackhole_local ||
665 		    !in_localip(ip->ip_src)))
666 			goto badunlocked;
667 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
668 			goto badunlocked;
669 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
670 		return (IPPROTO_DONE);
671 	}
672 
673 	/*
674 	 * Check the minimum TTL for socket.
675 	 */
676 	INP_RLOCK_ASSERT(inp);
677 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
678 		if (proto == IPPROTO_UDPLITE)
679 			UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
680 		else
681 			UDP_PROBE(receive, NULL, inp, ip, inp, uh);
682 		INP_RUNLOCK(inp);
683 		m_freem(m);
684 		return (IPPROTO_DONE);
685 	}
686 	if (cscov_partial) {
687 		struct udpcb *up;
688 
689 		up = intoudpcb(inp);
690 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
691 			INP_RUNLOCK(inp);
692 			m_freem(m);
693 			return (IPPROTO_DONE);
694 		}
695 	}
696 
697 	if (proto == IPPROTO_UDPLITE)
698 		UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
699 	else
700 		UDP_PROBE(receive, NULL, inp, ip, inp, uh);
701 	if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
702 		INP_RUNLOCK(inp);
703 	return (IPPROTO_DONE);
704 
705 badunlocked:
706 	m_freem(m);
707 	return (IPPROTO_DONE);
708 }
709 #endif /* INET */
710 
711 /*
712  * Notify a udp user of an asynchronous error; just wake up so that they can
713  * collect error status.
714  */
715 struct inpcb *
716 udp_notify(struct inpcb *inp, int errno)
717 {
718 
719 	INP_WLOCK_ASSERT(inp);
720 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
721 	     errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
722 		NH_FREE(inp->inp_route.ro_nh);
723 		inp->inp_route.ro_nh = (struct nhop_object *)NULL;
724 	}
725 
726 	inp->inp_socket->so_error = errno;
727 	sorwakeup(inp->inp_socket);
728 	sowwakeup(inp->inp_socket);
729 	return (inp);
730 }
731 
732 #ifdef INET
733 static void
734 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo)
735 {
736 	struct ip *ip = &icmp->icmp_ip;
737 	struct udphdr *uh;
738 	struct inpcb *inp;
739 
740 	if (icmp_errmap(icmp) == 0)
741 		return;
742 
743 	uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
744 	inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src,
745 	    uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
746 	if (inp != NULL) {
747 		INP_WLOCK_ASSERT(inp);
748 		if (inp->inp_socket != NULL)
749 			udp_notify(inp, icmp_errmap(icmp));
750 		INP_WUNLOCK(inp);
751 	} else {
752 		inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport,
753 		    ip->ip_src, uh->uh_sport,
754 		    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
755 		if (inp != NULL) {
756 			struct udpcb *up;
757 			udp_tun_icmp_t *func;
758 
759 			up = intoudpcb(inp);
760 			func = up->u_icmp_func;
761 			INP_RUNLOCK(inp);
762 			if (func != NULL)
763 				func(icmp);
764 		}
765 	}
766 }
767 
768 static void
769 udp_ctlinput(struct icmp *icmp)
770 {
771 
772 	return (udp_common_ctlinput(icmp, &V_udbinfo));
773 }
774 
775 static void
776 udplite_ctlinput(struct icmp *icmp)
777 {
778 
779 	return (udp_common_ctlinput(icmp, &V_ulitecbinfo));
780 }
781 #endif /* INET */
782 
783 static int
784 udp_pcblist(SYSCTL_HANDLER_ARGS)
785 {
786 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
787 	    INPLOOKUP_RLOCKPCB);
788 	struct xinpgen xig;
789 	struct inpcb *inp;
790 	int error;
791 
792 	if (req->newptr != 0)
793 		return (EPERM);
794 
795 	if (req->oldptr == 0) {
796 		int n;
797 
798 		n = V_udbinfo.ipi_count;
799 		n += imax(n / 8, 10);
800 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
801 		return (0);
802 	}
803 
804 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
805 		return (error);
806 
807 	bzero(&xig, sizeof(xig));
808 	xig.xig_len = sizeof xig;
809 	xig.xig_count = V_udbinfo.ipi_count;
810 	xig.xig_gen = V_udbinfo.ipi_gencnt;
811 	xig.xig_sogen = so_gencnt;
812 	error = SYSCTL_OUT(req, &xig, sizeof xig);
813 	if (error)
814 		return (error);
815 
816 	while ((inp = inp_next(&inpi)) != NULL) {
817 		if (inp->inp_gencnt <= xig.xig_gen &&
818 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
819 			struct xinpcb xi;
820 
821 			in_pcbtoxinpcb(inp, &xi);
822 			error = SYSCTL_OUT(req, &xi, sizeof xi);
823 			if (error) {
824 				INP_RUNLOCK(inp);
825 				break;
826 			}
827 		}
828 	}
829 
830 	if (!error) {
831 		/*
832 		 * Give the user an updated idea of our state.  If the
833 		 * generation differs from what we told her before, she knows
834 		 * that something happened while we were processing this
835 		 * request, and it might be necessary to retry.
836 		 */
837 		xig.xig_gen = V_udbinfo.ipi_gencnt;
838 		xig.xig_sogen = so_gencnt;
839 		xig.xig_count = V_udbinfo.ipi_count;
840 		error = SYSCTL_OUT(req, &xig, sizeof xig);
841 	}
842 
843 	return (error);
844 }
845 
846 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
847     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
848     udp_pcblist, "S,xinpcb",
849     "List of active UDP sockets");
850 
851 #ifdef INET
852 static int
853 udp_getcred(SYSCTL_HANDLER_ARGS)
854 {
855 	struct xucred xuc;
856 	struct sockaddr_in addrs[2];
857 	struct epoch_tracker et;
858 	struct inpcb *inp;
859 	int error;
860 
861 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
862 	if (error)
863 		return (error);
864 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
865 	if (error)
866 		return (error);
867 	NET_EPOCH_ENTER(et);
868 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
869 	    addrs[0].sin_addr, addrs[0].sin_port,
870 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
871 	NET_EPOCH_EXIT(et);
872 	if (inp != NULL) {
873 		INP_RLOCK_ASSERT(inp);
874 		if (inp->inp_socket == NULL)
875 			error = ENOENT;
876 		if (error == 0)
877 			error = cr_canseeinpcb(req->td->td_ucred, inp);
878 		if (error == 0)
879 			cru2x(inp->inp_cred, &xuc);
880 		INP_RUNLOCK(inp);
881 	} else
882 		error = ENOENT;
883 	if (error == 0)
884 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
885 	return (error);
886 }
887 
888 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
889     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
890     0, 0, udp_getcred, "S,xucred",
891     "Get the xucred of a UDP connection");
892 #endif /* INET */
893 
894 int
895 udp_ctloutput(struct socket *so, struct sockopt *sopt)
896 {
897 	struct inpcb *inp;
898 	struct udpcb *up;
899 	int isudplite, error, optval;
900 
901 	error = 0;
902 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
903 	inp = sotoinpcb(so);
904 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
905 	INP_WLOCK(inp);
906 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
907 #ifdef INET6
908 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
909 			INP_WUNLOCK(inp);
910 			error = ip6_ctloutput(so, sopt);
911 		}
912 #endif
913 #if defined(INET) && defined(INET6)
914 		else
915 #endif
916 #ifdef INET
917 		{
918 			INP_WUNLOCK(inp);
919 			error = ip_ctloutput(so, sopt);
920 		}
921 #endif
922 		return (error);
923 	}
924 
925 	switch (sopt->sopt_dir) {
926 	case SOPT_SET:
927 		switch (sopt->sopt_name) {
928 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
929 #ifdef INET
930 		case UDP_ENCAP:
931 			if (!IPSEC_ENABLED(ipv4)) {
932 				INP_WUNLOCK(inp);
933 				return (ENOPROTOOPT);
934 			}
935 			error = UDPENCAP_PCBCTL(inp, sopt);
936 			break;
937 #endif /* INET */
938 #endif /* IPSEC */
939 		case UDPLITE_SEND_CSCOV:
940 		case UDPLITE_RECV_CSCOV:
941 			if (!isudplite) {
942 				INP_WUNLOCK(inp);
943 				error = ENOPROTOOPT;
944 				break;
945 			}
946 			INP_WUNLOCK(inp);
947 			error = sooptcopyin(sopt, &optval, sizeof(optval),
948 			    sizeof(optval));
949 			if (error != 0)
950 				break;
951 			inp = sotoinpcb(so);
952 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
953 			INP_WLOCK(inp);
954 			up = intoudpcb(inp);
955 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
956 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
957 				INP_WUNLOCK(inp);
958 				error = EINVAL;
959 				break;
960 			}
961 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
962 				up->u_txcslen = optval;
963 			else
964 				up->u_rxcslen = optval;
965 			INP_WUNLOCK(inp);
966 			break;
967 		default:
968 			INP_WUNLOCK(inp);
969 			error = ENOPROTOOPT;
970 			break;
971 		}
972 		break;
973 	case SOPT_GET:
974 		switch (sopt->sopt_name) {
975 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
976 #ifdef INET
977 		case UDP_ENCAP:
978 			if (!IPSEC_ENABLED(ipv4)) {
979 				INP_WUNLOCK(inp);
980 				return (ENOPROTOOPT);
981 			}
982 			error = UDPENCAP_PCBCTL(inp, sopt);
983 			break;
984 #endif /* INET */
985 #endif /* IPSEC */
986 		case UDPLITE_SEND_CSCOV:
987 		case UDPLITE_RECV_CSCOV:
988 			if (!isudplite) {
989 				INP_WUNLOCK(inp);
990 				error = ENOPROTOOPT;
991 				break;
992 			}
993 			up = intoudpcb(inp);
994 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
995 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
996 				optval = up->u_txcslen;
997 			else
998 				optval = up->u_rxcslen;
999 			INP_WUNLOCK(inp);
1000 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1001 			break;
1002 		default:
1003 			INP_WUNLOCK(inp);
1004 			error = ENOPROTOOPT;
1005 			break;
1006 		}
1007 		break;
1008 	}
1009 	return (error);
1010 }
1011 
1012 #ifdef INET
1013 #ifdef INET6
1014 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1015 static int
1016 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1017     struct inpcb *inp, int flags)
1018 {
1019 	struct ifnet *ifp;
1020 	struct in6_pktinfo *pktinfo;
1021 	struct in_addr ia;
1022 
1023 	if ((flags & PRUS_IPV6) == 0)
1024 		return (0);
1025 
1026 	if (cm->cmsg_level != IPPROTO_IPV6)
1027 		return (0);
1028 
1029 	if  (cm->cmsg_type != IPV6_2292PKTINFO &&
1030 	    cm->cmsg_type != IPV6_PKTINFO)
1031 		return (0);
1032 
1033 	if (cm->cmsg_len !=
1034 	    CMSG_LEN(sizeof(struct in6_pktinfo)))
1035 		return (EINVAL);
1036 
1037 	pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1038 	if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1039 	    !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1040 		return (EINVAL);
1041 
1042 	/* Validate the interface index if specified. */
1043 	if (pktinfo->ipi6_ifindex) {
1044 		struct epoch_tracker et;
1045 
1046 		NET_EPOCH_ENTER(et);
1047 		ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1048 		NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
1049 		if (ifp == NULL)
1050 			return (ENXIO);
1051 	} else
1052 		ifp = NULL;
1053 	if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1054 		ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1055 		if (in_ifhasaddr(ifp, ia) == 0)
1056 			return (EADDRNOTAVAIL);
1057 	}
1058 
1059 	bzero(src, sizeof(*src));
1060 	src->sin_family = AF_INET;
1061 	src->sin_len = sizeof(*src);
1062 	src->sin_port = inp->inp_lport;
1063 	src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1064 
1065 	return (0);
1066 }
1067 #endif
1068 
1069 static int
1070 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1071     struct mbuf *control, struct thread *td, int flags)
1072 {
1073 	struct udpiphdr *ui;
1074 	int len = m->m_pkthdr.len;
1075 	struct in_addr faddr, laddr;
1076 	struct cmsghdr *cm;
1077 	struct inpcbinfo *pcbinfo;
1078 	struct sockaddr_in *sin, src;
1079 	struct epoch_tracker et;
1080 	int cscov_partial = 0;
1081 	int error = 0;
1082 	int ipflags = 0;
1083 	u_short fport, lport;
1084 	u_char tos;
1085 	uint8_t pr;
1086 	uint16_t cscov = 0;
1087 	uint32_t flowid = 0;
1088 	uint8_t flowtype = M_HASHTYPE_NONE;
1089 
1090 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1091 		if (control)
1092 			m_freem(control);
1093 		m_freem(m);
1094 		return (EMSGSIZE);
1095 	}
1096 
1097 	src.sin_family = 0;
1098 	sin = (struct sockaddr_in *)addr;
1099 
1100 	/*
1101 	 * udp_output() may need to temporarily bind or connect the current
1102 	 * inpcb.  As such, we don't know up front whether we will need the
1103 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1104 	 * front before acquiring any locks.
1105 	 *
1106 	 * We will need network epoch in either case, to safely lookup into
1107 	 * pcb hash.
1108 	 */
1109 	if (sin == NULL ||
1110 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0))
1111 		INP_WLOCK(inp);
1112 	else
1113 		INP_RLOCK(inp);
1114 	NET_EPOCH_ENTER(et);
1115 	tos = inp->inp_ip_tos;
1116 	if (control != NULL) {
1117 		/*
1118 		 * XXX: Currently, we assume all the optional information is
1119 		 * stored in a single mbuf.
1120 		 */
1121 		if (control->m_next) {
1122 			m_freem(control);
1123 			error = EINVAL;
1124 			goto release;
1125 		}
1126 		for (; control->m_len > 0;
1127 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1128 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1129 			cm = mtod(control, struct cmsghdr *);
1130 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1131 			    || cm->cmsg_len > control->m_len) {
1132 				error = EINVAL;
1133 				break;
1134 			}
1135 #ifdef INET6
1136 			error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1137 			if (error != 0)
1138 				break;
1139 #endif
1140 			if (cm->cmsg_level != IPPROTO_IP)
1141 				continue;
1142 
1143 			switch (cm->cmsg_type) {
1144 			case IP_SENDSRCADDR:
1145 				if (cm->cmsg_len !=
1146 				    CMSG_LEN(sizeof(struct in_addr))) {
1147 					error = EINVAL;
1148 					break;
1149 				}
1150 				bzero(&src, sizeof(src));
1151 				src.sin_family = AF_INET;
1152 				src.sin_len = sizeof(src);
1153 				src.sin_port = inp->inp_lport;
1154 				src.sin_addr =
1155 				    *(struct in_addr *)CMSG_DATA(cm);
1156 				break;
1157 
1158 			case IP_TOS:
1159 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1160 					error = EINVAL;
1161 					break;
1162 				}
1163 				tos = *(u_char *)CMSG_DATA(cm);
1164 				break;
1165 
1166 			case IP_FLOWID:
1167 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1168 					error = EINVAL;
1169 					break;
1170 				}
1171 				flowid = *(uint32_t *) CMSG_DATA(cm);
1172 				break;
1173 
1174 			case IP_FLOWTYPE:
1175 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1176 					error = EINVAL;
1177 					break;
1178 				}
1179 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1180 				break;
1181 
1182 #ifdef	RSS
1183 			case IP_RSSBUCKETID:
1184 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1185 					error = EINVAL;
1186 					break;
1187 				}
1188 				/* This is just a placeholder for now */
1189 				break;
1190 #endif	/* RSS */
1191 			default:
1192 				error = ENOPROTOOPT;
1193 				break;
1194 			}
1195 			if (error)
1196 				break;
1197 		}
1198 		m_freem(control);
1199 		control = NULL;
1200 	}
1201 	if (error)
1202 		goto release;
1203 
1204 	pr = inp->inp_socket->so_proto->pr_protocol;
1205 	pcbinfo = udp_get_inpcbinfo(pr);
1206 
1207 	/*
1208 	 * If the IP_SENDSRCADDR control message was specified, override the
1209 	 * source address for this datagram.  Its use is invalidated if the
1210 	 * address thus specified is incomplete or clobbers other inpcbs.
1211 	 */
1212 	laddr = inp->inp_laddr;
1213 	lport = inp->inp_lport;
1214 	if (src.sin_family == AF_INET) {
1215 		if ((lport == 0) ||
1216 		    (laddr.s_addr == INADDR_ANY &&
1217 		     src.sin_addr.s_addr == INADDR_ANY)) {
1218 			error = EINVAL;
1219 			goto release;
1220 		}
1221 		INP_HASH_WLOCK(pcbinfo);
1222 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1223 		    &laddr.s_addr, &lport, td->td_ucred);
1224 		INP_HASH_WUNLOCK(pcbinfo);
1225 		if (error)
1226 			goto release;
1227 	}
1228 
1229 	/*
1230 	 * If a UDP socket has been connected, then a local address/port will
1231 	 * have been selected and bound.
1232 	 *
1233 	 * If a UDP socket has not been connected to, then an explicit
1234 	 * destination address must be used, in which case a local
1235 	 * address/port may not have been selected and bound.
1236 	 */
1237 	if (sin != NULL) {
1238 		INP_LOCK_ASSERT(inp);
1239 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1240 			error = EISCONN;
1241 			goto release;
1242 		}
1243 
1244 		/*
1245 		 * Jail may rewrite the destination address, so let it do
1246 		 * that before we use it.
1247 		 */
1248 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1249 		if (error)
1250 			goto release;
1251 
1252 		/*
1253 		 * If a local address or port hasn't yet been selected, or if
1254 		 * the destination address needs to be rewritten due to using
1255 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1256 		 * to do the heavy lifting.  Once a port is selected, we
1257 		 * commit the binding back to the socket; we also commit the
1258 		 * binding of the address if in jail.
1259 		 *
1260 		 * If we already have a valid binding and we're not
1261 		 * requesting a destination address rewrite, use a fast path.
1262 		 */
1263 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1264 		    inp->inp_lport == 0 ||
1265 		    sin->sin_addr.s_addr == INADDR_ANY ||
1266 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1267 			INP_HASH_WLOCK(pcbinfo);
1268 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1269 			    &lport, &faddr.s_addr, &fport, NULL,
1270 			    td->td_ucred);
1271 			if (error) {
1272 				INP_HASH_WUNLOCK(pcbinfo);
1273 				goto release;
1274 			}
1275 
1276 			/*
1277 			 * XXXRW: Why not commit the port if the address is
1278 			 * !INADDR_ANY?
1279 			 */
1280 			/* Commit the local port if newly assigned. */
1281 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1282 			    inp->inp_lport == 0) {
1283 				INP_WLOCK_ASSERT(inp);
1284 				/*
1285 				 * Remember addr if jailed, to prevent
1286 				 * rebinding.
1287 				 */
1288 				if (prison_flag(td->td_ucred, PR_IP4))
1289 					inp->inp_laddr = laddr;
1290 				inp->inp_lport = lport;
1291 				error = in_pcbinshash(inp);
1292 				INP_HASH_WUNLOCK(pcbinfo);
1293 				if (error != 0) {
1294 					inp->inp_lport = 0;
1295 					error = EAGAIN;
1296 					goto release;
1297 				}
1298 				inp->inp_flags |= INP_ANONPORT;
1299 			} else
1300 				INP_HASH_WUNLOCK(pcbinfo);
1301 		} else {
1302 			faddr = sin->sin_addr;
1303 			fport = sin->sin_port;
1304 		}
1305 	} else {
1306 		INP_LOCK_ASSERT(inp);
1307 		faddr = inp->inp_faddr;
1308 		fport = inp->inp_fport;
1309 		if (faddr.s_addr == INADDR_ANY) {
1310 			error = ENOTCONN;
1311 			goto release;
1312 		}
1313 	}
1314 
1315 	/*
1316 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1317 	 * link-layer headers.  Immediate slide the data pointer back forward
1318 	 * since we won't use that space at this layer.
1319 	 */
1320 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1321 	if (m == NULL) {
1322 		error = ENOBUFS;
1323 		goto release;
1324 	}
1325 	m->m_data += max_linkhdr;
1326 	m->m_len -= max_linkhdr;
1327 	m->m_pkthdr.len -= max_linkhdr;
1328 
1329 	/*
1330 	 * Fill in mbuf with extended UDP header and addresses and length put
1331 	 * into network format.
1332 	 */
1333 	ui = mtod(m, struct udpiphdr *);
1334 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1335 	ui->ui_v = IPVERSION << 4;
1336 	ui->ui_pr = pr;
1337 	ui->ui_src = laddr;
1338 	ui->ui_dst = faddr;
1339 	ui->ui_sport = lport;
1340 	ui->ui_dport = fport;
1341 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1342 	if (pr == IPPROTO_UDPLITE) {
1343 		struct udpcb *up;
1344 		uint16_t plen;
1345 
1346 		up = intoudpcb(inp);
1347 		cscov = up->u_txcslen;
1348 		plen = (u_short)len + sizeof(struct udphdr);
1349 		if (cscov >= plen)
1350 			cscov = 0;
1351 		ui->ui_len = htons(plen);
1352 		ui->ui_ulen = htons(cscov);
1353 		/*
1354 		 * For UDP-Lite, checksum coverage length of zero means
1355 		 * the entire UDPLite packet is covered by the checksum.
1356 		 */
1357 		cscov_partial = (cscov == 0) ? 0 : 1;
1358 	}
1359 
1360 	/*
1361 	 * Set the Don't Fragment bit in the IP header.
1362 	 */
1363 	if (inp->inp_flags & INP_DONTFRAG) {
1364 		struct ip *ip;
1365 
1366 		ip = (struct ip *)&ui->ui_i;
1367 		ip->ip_off |= htons(IP_DF);
1368 	}
1369 
1370 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1371 		ipflags |= IP_ROUTETOIF;
1372 	if (inp->inp_socket->so_options & SO_BROADCAST)
1373 		ipflags |= IP_ALLOWBROADCAST;
1374 	if (inp->inp_flags & INP_ONESBCAST)
1375 		ipflags |= IP_SENDONES;
1376 
1377 #ifdef MAC
1378 	mac_inpcb_create_mbuf(inp, m);
1379 #endif
1380 
1381 	/*
1382 	 * Set up checksum and output datagram.
1383 	 */
1384 	ui->ui_sum = 0;
1385 	if (pr == IPPROTO_UDPLITE) {
1386 		if (inp->inp_flags & INP_ONESBCAST)
1387 			faddr.s_addr = INADDR_BROADCAST;
1388 		if (cscov_partial) {
1389 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1390 				ui->ui_sum = 0xffff;
1391 		} else {
1392 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1393 				ui->ui_sum = 0xffff;
1394 		}
1395 	} else if (V_udp_cksum) {
1396 		if (inp->inp_flags & INP_ONESBCAST)
1397 			faddr.s_addr = INADDR_BROADCAST;
1398 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1399 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1400 		m->m_pkthdr.csum_flags = CSUM_UDP;
1401 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1402 	}
1403 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1404 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1405 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1406 	UDPSTAT_INC(udps_opackets);
1407 
1408 	/*
1409 	 * Setup flowid / RSS information for outbound socket.
1410 	 *
1411 	 * Once the UDP code decides to set a flowid some other way,
1412 	 * this allows the flowid to be overridden by userland.
1413 	 */
1414 	if (flowtype != M_HASHTYPE_NONE) {
1415 		m->m_pkthdr.flowid = flowid;
1416 		M_HASHTYPE_SET(m, flowtype);
1417 	}
1418 #if defined(ROUTE_MPATH) || defined(RSS)
1419 	else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1420 		uint32_t hash_val, hash_type;
1421 
1422 		hash_val = fib4_calc_packet_hash(laddr, faddr,
1423 		    lport, fport, pr, &hash_type);
1424 		m->m_pkthdr.flowid = hash_val;
1425 		M_HASHTYPE_SET(m, hash_type);
1426 	}
1427 
1428 	/*
1429 	 * Don't override with the inp cached flowid value.
1430 	 *
1431 	 * Depending upon the kind of send being done, the inp
1432 	 * flowid/flowtype values may actually not be appropriate
1433 	 * for this particular socket send.
1434 	 *
1435 	 * We should either leave the flowid at zero (which is what is
1436 	 * currently done) or set it to some software generated
1437 	 * hash value based on the packet contents.
1438 	 */
1439 	ipflags |= IP_NODEFAULTFLOWID;
1440 #endif	/* RSS */
1441 
1442 	if (pr == IPPROTO_UDPLITE)
1443 		UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1444 	else
1445 		UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1446 	error = ip_output(m, inp->inp_options,
1447 	    INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags,
1448 	    inp->inp_moptions, inp);
1449 	INP_UNLOCK(inp);
1450 	NET_EPOCH_EXIT(et);
1451 	return (error);
1452 
1453 release:
1454 	INP_UNLOCK(inp);
1455 	NET_EPOCH_EXIT(et);
1456 	m_freem(m);
1457 	return (error);
1458 }
1459 
1460 pr_abort_t udp_abort;			/* shared with udp6_usrreq.c */
1461 void
1462 udp_abort(struct socket *so)
1463 {
1464 	struct inpcb *inp;
1465 	struct inpcbinfo *pcbinfo;
1466 
1467 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1468 	inp = sotoinpcb(so);
1469 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1470 	INP_WLOCK(inp);
1471 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1472 		INP_HASH_WLOCK(pcbinfo);
1473 		in_pcbdisconnect(inp);
1474 		inp->inp_laddr.s_addr = INADDR_ANY;
1475 		INP_HASH_WUNLOCK(pcbinfo);
1476 		soisdisconnected(so);
1477 	}
1478 	INP_WUNLOCK(inp);
1479 }
1480 
1481 static int
1482 udp_attach(struct socket *so, int proto, struct thread *td)
1483 {
1484 	static uint32_t udp_flowid;
1485 	struct inpcb *inp;
1486 	struct inpcbinfo *pcbinfo;
1487 	int error;
1488 
1489 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1490 	inp = sotoinpcb(so);
1491 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1492 	error = soreserve(so, udp_sendspace, udp_recvspace);
1493 	if (error)
1494 		return (error);
1495 	error = in_pcballoc(so, pcbinfo);
1496 	if (error)
1497 		return (error);
1498 
1499 	inp = sotoinpcb(so);
1500 	inp->inp_ip_ttl = V_ip_defttl;
1501 	inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1502 	inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1503 
1504 	error = udp_newudpcb(inp);
1505 	if (error) {
1506 		in_pcbdetach(inp);
1507 		in_pcbfree(inp);
1508 		return (error);
1509 	}
1510 	INP_WUNLOCK(inp);
1511 
1512 	return (0);
1513 }
1514 #endif /* INET */
1515 
1516 int
1517 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1518 {
1519 	struct inpcb *inp;
1520 	struct udpcb *up;
1521 
1522 	KASSERT(so->so_type == SOCK_DGRAM,
1523 	    ("udp_set_kernel_tunneling: !dgram"));
1524 	inp = sotoinpcb(so);
1525 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1526 	INP_WLOCK(inp);
1527 	up = intoudpcb(inp);
1528 	if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1529 	    (up->u_icmp_func != NULL))) {
1530 		INP_WUNLOCK(inp);
1531 		return (EBUSY);
1532 	}
1533 	up->u_tun_func = f;
1534 	up->u_icmp_func = i;
1535 	up->u_tun_ctx = ctx;
1536 	INP_WUNLOCK(inp);
1537 	return (0);
1538 }
1539 
1540 #ifdef INET
1541 static int
1542 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1543 {
1544 	struct inpcb *inp;
1545 	struct inpcbinfo *pcbinfo;
1546 	struct sockaddr_in *sinp;
1547 	int error;
1548 
1549 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1550 	inp = sotoinpcb(so);
1551 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1552 
1553 	sinp = (struct sockaddr_in *)nam;
1554 	if (nam->sa_family != AF_INET) {
1555 		/*
1556 		 * Preserve compatibility with old programs.
1557 		 */
1558 		if (nam->sa_family != AF_UNSPEC ||
1559 		    nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1560 		    sinp->sin_addr.s_addr != INADDR_ANY)
1561 			return (EAFNOSUPPORT);
1562 		nam->sa_family = AF_INET;
1563 	}
1564 	if (nam->sa_len != sizeof(struct sockaddr_in))
1565 		return (EINVAL);
1566 
1567 	INP_WLOCK(inp);
1568 	INP_HASH_WLOCK(pcbinfo);
1569 	error = in_pcbbind(inp, nam, td->td_ucred);
1570 	INP_HASH_WUNLOCK(pcbinfo);
1571 	INP_WUNLOCK(inp);
1572 	return (error);
1573 }
1574 
1575 static void
1576 udp_close(struct socket *so)
1577 {
1578 	struct inpcb *inp;
1579 	struct inpcbinfo *pcbinfo;
1580 
1581 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1582 	inp = sotoinpcb(so);
1583 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1584 	INP_WLOCK(inp);
1585 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1586 		INP_HASH_WLOCK(pcbinfo);
1587 		in_pcbdisconnect(inp);
1588 		inp->inp_laddr.s_addr = INADDR_ANY;
1589 		INP_HASH_WUNLOCK(pcbinfo);
1590 		soisdisconnected(so);
1591 	}
1592 	INP_WUNLOCK(inp);
1593 }
1594 
1595 static int
1596 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1597 {
1598 	struct epoch_tracker et;
1599 	struct inpcb *inp;
1600 	struct inpcbinfo *pcbinfo;
1601 	struct sockaddr_in *sin;
1602 	int error;
1603 
1604 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1605 	inp = sotoinpcb(so);
1606 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1607 
1608 	sin = (struct sockaddr_in *)nam;
1609 	if (sin->sin_family != AF_INET)
1610 		return (EAFNOSUPPORT);
1611 	if (sin->sin_len != sizeof(*sin))
1612 		return (EINVAL);
1613 
1614 	INP_WLOCK(inp);
1615 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1616 		INP_WUNLOCK(inp);
1617 		return (EISCONN);
1618 	}
1619 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1620 	if (error != 0) {
1621 		INP_WUNLOCK(inp);
1622 		return (error);
1623 	}
1624 	NET_EPOCH_ENTER(et);
1625 	INP_HASH_WLOCK(pcbinfo);
1626 	error = in_pcbconnect(inp, nam, td->td_ucred, true);
1627 	INP_HASH_WUNLOCK(pcbinfo);
1628 	NET_EPOCH_EXIT(et);
1629 	if (error == 0)
1630 		soisconnected(so);
1631 	INP_WUNLOCK(inp);
1632 	return (error);
1633 }
1634 
1635 static void
1636 udp_detach(struct socket *so)
1637 {
1638 	struct inpcb *inp;
1639 	struct udpcb *up;
1640 
1641 	inp = sotoinpcb(so);
1642 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1643 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1644 	    ("udp_detach: not disconnected"));
1645 	INP_WLOCK(inp);
1646 	up = intoudpcb(inp);
1647 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1648 	inp->inp_ppcb = NULL;
1649 	in_pcbdetach(inp);
1650 	in_pcbfree(inp);
1651 	udp_discardcb(up);
1652 }
1653 
1654 pr_disconnect_t udp_disconnect;		/* shared with udp6_usrreq.c */
1655 int
1656 udp_disconnect(struct socket *so)
1657 {
1658 	struct inpcb *inp;
1659 	struct inpcbinfo *pcbinfo;
1660 
1661 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1662 	inp = sotoinpcb(so);
1663 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1664 	INP_WLOCK(inp);
1665 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1666 		INP_WUNLOCK(inp);
1667 		return (ENOTCONN);
1668 	}
1669 	INP_HASH_WLOCK(pcbinfo);
1670 	in_pcbdisconnect(inp);
1671 	inp->inp_laddr.s_addr = INADDR_ANY;
1672 	INP_HASH_WUNLOCK(pcbinfo);
1673 	SOCK_LOCK(so);
1674 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1675 	SOCK_UNLOCK(so);
1676 	INP_WUNLOCK(inp);
1677 	return (0);
1678 }
1679 
1680 pr_send_t udp_send;			/* shared with udp6_usrreq.c */
1681 int
1682 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1683     struct mbuf *control, struct thread *td)
1684 {
1685 	struct inpcb *inp;
1686 	int error;
1687 
1688 	inp = sotoinpcb(so);
1689 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1690 
1691 	if (addr != NULL) {
1692 		error = 0;
1693 		if (addr->sa_family != AF_INET)
1694 			error = EAFNOSUPPORT;
1695 		else if (addr->sa_len != sizeof(struct sockaddr_in))
1696 			error = EINVAL;
1697 		if (__predict_false(error != 0)) {
1698 			m_freem(control);
1699 			m_freem(m);
1700 			return (error);
1701 		}
1702 	}
1703 	return (udp_output(inp, m, addr, control, td, flags));
1704 }
1705 #endif /* INET */
1706 
1707 int
1708 udp_shutdown(struct socket *so)
1709 {
1710 	struct inpcb *inp;
1711 
1712 	inp = sotoinpcb(so);
1713 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1714 	INP_WLOCK(inp);
1715 	socantsendmore(so);
1716 	INP_WUNLOCK(inp);
1717 	return (0);
1718 }
1719 
1720 #ifdef INET
1721 #define	UDP_PROTOSW							\
1722 	.pr_type =		SOCK_DGRAM,				\
1723 	.pr_flags =		PR_ATOMIC | PR_ADDR | PR_CAPATTACH,	\
1724 	.pr_ctloutput =		udp_ctloutput,				\
1725 	.pr_abort =		udp_abort,				\
1726 	.pr_attach =		udp_attach,				\
1727 	.pr_bind =		udp_bind,				\
1728 	.pr_connect =		udp_connect,				\
1729 	.pr_control =		in_control,				\
1730 	.pr_detach =		udp_detach,				\
1731 	.pr_disconnect =	udp_disconnect,				\
1732 	.pr_peeraddr =		in_getpeeraddr,				\
1733 	.pr_send =		udp_send,				\
1734 	.pr_soreceive =		soreceive_dgram,			\
1735 	.pr_sosend =		sosend_dgram,				\
1736 	.pr_shutdown =		udp_shutdown,				\
1737 	.pr_sockaddr =		in_getsockaddr,				\
1738 	.pr_sosetlabel =	in_pcbsosetlabel,			\
1739 	.pr_close =		udp_close
1740 
1741 struct protosw udp_protosw = {
1742 	.pr_protocol =		IPPROTO_UDP,
1743 	UDP_PROTOSW
1744 };
1745 
1746 struct protosw udplite_protosw = {
1747 	.pr_protocol =		IPPROTO_UDPLITE,
1748 	UDP_PROTOSW
1749 };
1750 
1751 static void
1752 udp_init(void *arg __unused)
1753 {
1754 
1755 	IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput);
1756 	IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput);
1757 }
1758 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
1759 #endif /* INET */
1760