xref: /freebsd/sys/netinet/udp_usrreq.c (revision 298f5fdc242b760e70cd3494e3a4f1f50b20664d)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2008 Robert N. M. Watson
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * Copyright (c) 2014 Kevin Lo
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Robert N. M. Watson under
10  * contract to Juniper Networks, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ipfw.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
46 
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/sdt.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
66 
67 #include <vm/uma.h>
68 
69 #include <net/if.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 
73 #include <netinet/in.h>
74 #include <netinet/in_kdtrace.h>
75 #include <netinet/in_pcb.h>
76 #include <netinet/in_systm.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip.h>
79 #ifdef INET6
80 #include <netinet/ip6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/icmp_var.h>
84 #include <netinet/ip_var.h>
85 #include <netinet/ip_options.h>
86 #ifdef INET6
87 #include <netinet6/ip6_var.h>
88 #endif
89 #include <netinet/udp.h>
90 #include <netinet/udp_var.h>
91 #include <netinet/udplite.h>
92 
93 #ifdef IPSEC
94 #include <netipsec/ipsec.h>
95 #include <netipsec/esp.h>
96 #endif
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 /*
109  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
110  * removes the only data integrity mechanism for packets and malformed
111  * packets that would otherwise be discarded due to bad checksums, and may
112  * cause problems (especially for NFS data blocks).
113  */
114 VNET_DEFINE(int, udp_cksum) = 1;
115 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
116     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
117 
118 int	udp_log_in_vain = 0;
119 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
120     &udp_log_in_vain, 0, "Log all incoming UDP packets");
121 
122 VNET_DEFINE(int, udp_blackhole) = 0;
123 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
124     &VNET_NAME(udp_blackhole), 0,
125     "Do not send port unreachables for refused connects");
126 
127 u_long	udp_sendspace = 9216;		/* really max datagram size */
128 					/* 40 1K datagrams */
129 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
130     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
131 
132 u_long	udp_recvspace = 40 * (1024 +
133 #ifdef INET6
134 				      sizeof(struct sockaddr_in6)
135 #else
136 				      sizeof(struct sockaddr_in)
137 #endif
138 				      );
139 
140 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
141     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
142 
143 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
144 VNET_DEFINE(struct inpcbinfo, udbinfo);
145 VNET_DEFINE(struct inpcbhead, ulitecb);
146 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
147 static VNET_DEFINE(uma_zone_t, udpcb_zone);
148 #define	V_udpcb_zone			VNET(udpcb_zone)
149 
150 #ifndef UDBHASHSIZE
151 #define	UDBHASHSIZE	128
152 #endif
153 
154 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
155 VNET_PCPUSTAT_SYSINIT(udpstat);
156 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
157     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
158 
159 #ifdef VIMAGE
160 VNET_PCPUSTAT_SYSUNINIT(udpstat);
161 #endif /* VIMAGE */
162 #ifdef INET
163 static void	udp_detach(struct socket *so);
164 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
165 		    struct mbuf *, struct thread *);
166 #endif
167 
168 #ifdef IPSEC
169 #ifdef IPSEC_NAT_T
170 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
171 #ifdef INET
172 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
173 #endif
174 #endif /* IPSEC_NAT_T */
175 #endif /* IPSEC */
176 
177 static void
178 udp_zone_change(void *tag)
179 {
180 
181 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
182 	uma_zone_set_max(V_udpcb_zone, maxsockets);
183 }
184 
185 static int
186 udp_inpcb_init(void *mem, int size, int flags)
187 {
188 	struct inpcb *inp;
189 
190 	inp = mem;
191 	INP_LOCK_INIT(inp, "inp", "udpinp");
192 	return (0);
193 }
194 
195 static int
196 udplite_inpcb_init(void *mem, int size, int flags)
197 {
198 	struct inpcb *inp;
199 
200 	inp = mem;
201 	INP_LOCK_INIT(inp, "inp", "udpliteinp");
202 	return (0);
203 }
204 
205 void
206 udp_init(void)
207 {
208 
209 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
210 	    "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
211 	    IPI_HASHFIELDS_2TUPLE);
212 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
213 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
214 	uma_zone_set_max(V_udpcb_zone, maxsockets);
215 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
216 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
217 	    EVENTHANDLER_PRI_ANY);
218 }
219 
220 void
221 udplite_init(void)
222 {
223 
224 	in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
225 	    UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
226 	    UMA_ZONE_NOFREE, IPI_HASHFIELDS_2TUPLE);
227 }
228 
229 /*
230  * Kernel module interface for updating udpstat.  The argument is an index
231  * into udpstat treated as an array of u_long.  While this encodes the
232  * general layout of udpstat into the caller, it doesn't encode its location,
233  * so that future changes to add, for example, per-CPU stats support won't
234  * cause binary compatibility problems for kernel modules.
235  */
236 void
237 kmod_udpstat_inc(int statnum)
238 {
239 
240 	counter_u64_add(VNET(udpstat)[statnum], 1);
241 }
242 
243 int
244 udp_newudpcb(struct inpcb *inp)
245 {
246 	struct udpcb *up;
247 
248 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
249 	if (up == NULL)
250 		return (ENOBUFS);
251 	inp->inp_ppcb = up;
252 	return (0);
253 }
254 
255 void
256 udp_discardcb(struct udpcb *up)
257 {
258 
259 	uma_zfree(V_udpcb_zone, up);
260 }
261 
262 #ifdef VIMAGE
263 void
264 udp_destroy(void)
265 {
266 
267 	in_pcbinfo_destroy(&V_udbinfo);
268 	uma_zdestroy(V_udpcb_zone);
269 }
270 
271 void
272 udplite_destroy(void)
273 {
274 
275 	in_pcbinfo_destroy(&V_ulitecbinfo);
276 }
277 #endif
278 
279 #ifdef INET
280 /*
281  * Subroutine of udp_input(), which appends the provided mbuf chain to the
282  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
283  * contains the source address.  If the socket ends up being an IPv6 socket,
284  * udp_append() will convert to a sockaddr_in6 before passing the address
285  * into the socket code.
286  */
287 static void
288 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
289     struct sockaddr_in *udp_in)
290 {
291 	struct sockaddr *append_sa;
292 	struct socket *so;
293 	struct mbuf *opts = 0;
294 #ifdef INET6
295 	struct sockaddr_in6 udp_in6;
296 #endif
297 	struct udpcb *up;
298 
299 	INP_LOCK_ASSERT(inp);
300 
301 	/*
302 	 * Engage the tunneling protocol.
303 	 */
304 	up = intoudpcb(inp);
305 	if (up->u_tun_func != NULL) {
306 		(*up->u_tun_func)(n, off, inp);
307 		return;
308 	}
309 
310 	if (n == NULL)
311 		return;
312 
313 	off += sizeof(struct udphdr);
314 
315 #ifdef IPSEC
316 	/* Check AH/ESP integrity. */
317 	if (ipsec4_in_reject(n, inp)) {
318 		m_freem(n);
319 		IPSECSTAT_INC(ips_in_polvio);
320 		return;
321 	}
322 #ifdef IPSEC_NAT_T
323 	up = intoudpcb(inp);
324 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
325 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
326 		n = udp4_espdecap(inp, n, off);
327 		if (n == NULL)				/* Consumed. */
328 			return;
329 	}
330 #endif /* IPSEC_NAT_T */
331 #endif /* IPSEC */
332 #ifdef MAC
333 	if (mac_inpcb_check_deliver(inp, n) != 0) {
334 		m_freem(n);
335 		return;
336 	}
337 #endif /* MAC */
338 	if (inp->inp_flags & INP_CONTROLOPTS ||
339 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
340 #ifdef INET6
341 		if (inp->inp_vflag & INP_IPV6)
342 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
343 		else
344 #endif /* INET6 */
345 			ip_savecontrol(inp, &opts, ip, n);
346 	}
347 #ifdef INET6
348 	if (inp->inp_vflag & INP_IPV6) {
349 		bzero(&udp_in6, sizeof(udp_in6));
350 		udp_in6.sin6_len = sizeof(udp_in6);
351 		udp_in6.sin6_family = AF_INET6;
352 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
353 		append_sa = (struct sockaddr *)&udp_in6;
354 	} else
355 #endif /* INET6 */
356 		append_sa = (struct sockaddr *)udp_in;
357 	m_adj(n, off);
358 
359 	so = inp->inp_socket;
360 	SOCKBUF_LOCK(&so->so_rcv);
361 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
362 		SOCKBUF_UNLOCK(&so->so_rcv);
363 		m_freem(n);
364 		if (opts)
365 			m_freem(opts);
366 		UDPSTAT_INC(udps_fullsock);
367 	} else
368 		sorwakeup_locked(so);
369 }
370 
371 void
372 udp_input(struct mbuf *m, int off)
373 {
374 	int iphlen = off;
375 	struct ip *ip;
376 	struct udphdr *uh;
377 	struct ifnet *ifp;
378 	struct inpcb *inp;
379 	uint16_t len, ip_len;
380 	struct inpcbinfo *pcbinfo;
381 	struct ip save_ip;
382 	struct sockaddr_in udp_in;
383 	struct m_tag *fwd_tag;
384 	int cscov_partial;
385 	uint8_t pr;
386 
387 	ifp = m->m_pkthdr.rcvif;
388 	UDPSTAT_INC(udps_ipackets);
389 
390 	/*
391 	 * Strip IP options, if any; should skip this, make available to
392 	 * user, and use on returned packets, but we don't yet have a way to
393 	 * check the checksum with options still present.
394 	 */
395 	if (iphlen > sizeof (struct ip)) {
396 		ip_stripoptions(m);
397 		iphlen = sizeof(struct ip);
398 	}
399 
400 	/*
401 	 * Get IP and UDP header together in first mbuf.
402 	 */
403 	ip = mtod(m, struct ip *);
404 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
405 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
406 			UDPSTAT_INC(udps_hdrops);
407 			return;
408 		}
409 		ip = mtod(m, struct ip *);
410 	}
411 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
412 	pr = ip->ip_p;
413 	cscov_partial = (pr == IPPROTO_UDPLITE) ? 1 : 0;
414 
415 	/*
416 	 * Destination port of 0 is illegal, based on RFC768.
417 	 */
418 	if (uh->uh_dport == 0)
419 		goto badunlocked;
420 
421 	/*
422 	 * Construct sockaddr format source address.  Stuff source address
423 	 * and datagram in user buffer.
424 	 */
425 	bzero(&udp_in, sizeof(udp_in));
426 	udp_in.sin_len = sizeof(udp_in);
427 	udp_in.sin_family = AF_INET;
428 	udp_in.sin_port = uh->uh_sport;
429 	udp_in.sin_addr = ip->ip_src;
430 
431 	/*
432 	 * Make mbuf data length reflect UDP length.  If not enough data to
433 	 * reflect UDP length, drop.
434 	 */
435 	len = ntohs((u_short)uh->uh_ulen);
436 	ip_len = ntohs(ip->ip_len) - iphlen;
437 	if (pr == IPPROTO_UDPLITE && len == 0) {
438 		/* Zero means checksum over the complete packet. */
439 		len = ip_len;
440 		cscov_partial = 0;
441 	}
442 	if (ip_len != len) {
443 		if (len > ip_len || len < sizeof(struct udphdr)) {
444 			UDPSTAT_INC(udps_badlen);
445 			goto badunlocked;
446 		}
447 		if (pr == IPPROTO_UDP)
448 			m_adj(m, len - ip_len);
449 	}
450 
451 	/*
452 	 * Save a copy of the IP header in case we want restore it for
453 	 * sending an ICMP error message in response.
454 	 */
455 	if (!V_udp_blackhole)
456 		save_ip = *ip;
457 	else
458 		memset(&save_ip, 0, sizeof(save_ip));
459 
460 	/*
461 	 * Checksum extended UDP header and data.
462 	 */
463 	if (uh->uh_sum) {
464 		u_short uh_sum;
465 
466 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
467 		    !cscov_partial) {
468 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
469 				uh_sum = m->m_pkthdr.csum_data;
470 			else
471 				uh_sum = in_pseudo(ip->ip_src.s_addr,
472 				    ip->ip_dst.s_addr, htonl((u_short)len +
473 				    m->m_pkthdr.csum_data + pr));
474 			uh_sum ^= 0xffff;
475 		} else {
476 			char b[9];
477 
478 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
479 			bzero(((struct ipovly *)ip)->ih_x1, 9);
480 			((struct ipovly *)ip)->ih_len = (pr == IPPROTO_UDP) ?
481 			    uh->uh_ulen : htons(ip_len);
482 			uh_sum = in_cksum(m, len + sizeof (struct ip));
483 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
484 		}
485 		if (uh_sum) {
486 			UDPSTAT_INC(udps_badsum);
487 			m_freem(m);
488 			return;
489 		}
490 	} else
491 		UDPSTAT_INC(udps_nosum);
492 
493 	pcbinfo = get_inpcbinfo(pr);
494 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
495 	    in_broadcast(ip->ip_dst, ifp)) {
496 		struct inpcb *last;
497 		struct inpcbhead *pcblist;
498 		struct ip_moptions *imo;
499 
500 		INP_INFO_RLOCK(pcbinfo);
501 		pcblist = get_pcblist(pr);
502 		last = NULL;
503 		LIST_FOREACH(inp, pcblist, inp_list) {
504 			if (inp->inp_lport != uh->uh_dport)
505 				continue;
506 #ifdef INET6
507 			if ((inp->inp_vflag & INP_IPV4) == 0)
508 				continue;
509 #endif
510 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
511 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
512 				continue;
513 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
514 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
515 				continue;
516 			if (inp->inp_fport != 0 &&
517 			    inp->inp_fport != uh->uh_sport)
518 				continue;
519 
520 			INP_RLOCK(inp);
521 
522 			/*
523 			 * XXXRW: Because we weren't holding either the inpcb
524 			 * or the hash lock when we checked for a match
525 			 * before, we should probably recheck now that the
526 			 * inpcb lock is held.
527 			 */
528 
529 			/*
530 			 * Handle socket delivery policy for any-source
531 			 * and source-specific multicast. [RFC3678]
532 			 */
533 			imo = inp->inp_moptions;
534 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
535 				struct sockaddr_in	 group;
536 				int			 blocked;
537 				if (imo == NULL) {
538 					INP_RUNLOCK(inp);
539 					continue;
540 				}
541 				bzero(&group, sizeof(struct sockaddr_in));
542 				group.sin_len = sizeof(struct sockaddr_in);
543 				group.sin_family = AF_INET;
544 				group.sin_addr = ip->ip_dst;
545 
546 				blocked = imo_multi_filter(imo, ifp,
547 					(struct sockaddr *)&group,
548 					(struct sockaddr *)&udp_in);
549 				if (blocked != MCAST_PASS) {
550 					if (blocked == MCAST_NOTGMEMBER)
551 						IPSTAT_INC(ips_notmember);
552 					if (blocked == MCAST_NOTSMEMBER ||
553 					    blocked == MCAST_MUTED)
554 						UDPSTAT_INC(udps_filtermcast);
555 					INP_RUNLOCK(inp);
556 					continue;
557 				}
558 			}
559 			if (last != NULL) {
560 				struct mbuf *n;
561 
562 				n = m_copy(m, 0, M_COPYALL);
563 				udp_append(last, ip, n, iphlen, &udp_in);
564 				INP_RUNLOCK(last);
565 			}
566 			last = inp;
567 			/*
568 			 * Don't look for additional matches if this one does
569 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
570 			 * socket options set.  This heuristic avoids
571 			 * searching through all pcbs in the common case of a
572 			 * non-shared port.  It assumes that an application
573 			 * will never clear these options after setting them.
574 			 */
575 			if ((last->inp_socket->so_options &
576 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
577 				break;
578 		}
579 
580 		if (last == NULL) {
581 			/*
582 			 * No matching pcb found; discard datagram.  (No need
583 			 * to send an ICMP Port Unreachable for a broadcast
584 			 * or multicast datgram.)
585 			 */
586 			UDPSTAT_INC(udps_noportbcast);
587 			if (inp)
588 				INP_RUNLOCK(inp);
589 			INP_INFO_RUNLOCK(pcbinfo);
590 			goto badunlocked;
591 		}
592 		udp_append(last, ip, m, iphlen, &udp_in);
593 		INP_RUNLOCK(last);
594 		INP_INFO_RUNLOCK(pcbinfo);
595 		return;
596 	}
597 
598 	/*
599 	 * Locate pcb for datagram.
600 	 */
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, INPLOOKUP_RLOCKPCB, 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 |
627 			    INPLOOKUP_RLOCKPCB, 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 		    INPLOOKUP_RLOCKPCB, ifp, m);
636 	if (inp == NULL) {
637 		if (udp_log_in_vain) {
638 			char buf[4*sizeof "123"];
639 
640 			strcpy(buf, inet_ntoa(ip->ip_dst));
641 			log(LOG_INFO,
642 			    "Connection attempt to UDP %s:%d from %s:%d\n",
643 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
644 			    ntohs(uh->uh_sport));
645 		}
646 		UDPSTAT_INC(udps_noport);
647 		if (m->m_flags & (M_BCAST | M_MCAST)) {
648 			UDPSTAT_INC(udps_noportbcast);
649 			goto badunlocked;
650 		}
651 		if (V_udp_blackhole)
652 			goto badunlocked;
653 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
654 			goto badunlocked;
655 		*ip = save_ip;
656 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
657 		return;
658 	}
659 
660 	/*
661 	 * Check the minimum TTL for socket.
662 	 */
663 	INP_RLOCK_ASSERT(inp);
664 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
665 		INP_RUNLOCK(inp);
666 		m_freem(m);
667 		return;
668 	}
669 	if (cscov_partial) {
670 		struct udpcb *up;
671 
672 		up = intoudpcb(inp);
673 		if (up->u_rxcslen > len) {
674 			INP_RUNLOCK(inp);
675 			m_freem(m);
676 			return;
677 		}
678 	}
679 
680 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
681 	udp_append(inp, ip, m, iphlen, &udp_in);
682 	INP_RUNLOCK(inp);
683 	return;
684 
685 badunlocked:
686 	m_freem(m);
687 }
688 #endif /* INET */
689 
690 /*
691  * Notify a udp user of an asynchronous error; just wake up so that they can
692  * collect error status.
693  */
694 struct inpcb *
695 udp_notify(struct inpcb *inp, int errno)
696 {
697 
698 	/*
699 	 * While udp_ctlinput() always calls udp_notify() with a read lock
700 	 * when invoking it directly, in_pcbnotifyall() currently uses write
701 	 * locks due to sharing code with TCP.  For now, accept either a read
702 	 * or a write lock, but a read lock is sufficient.
703 	 */
704 	INP_LOCK_ASSERT(inp);
705 
706 	inp->inp_socket->so_error = errno;
707 	sorwakeup(inp->inp_socket);
708 	sowwakeup(inp->inp_socket);
709 	return (inp);
710 }
711 
712 #ifdef INET
713 static void
714 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
715     struct inpcbinfo *pcbinfo)
716 {
717 	struct ip *ip = vip;
718 	struct udphdr *uh;
719 	struct in_addr faddr;
720 	struct inpcb *inp;
721 
722 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
723 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
724 		return;
725 
726 	/*
727 	 * Redirects don't need to be handled up here.
728 	 */
729 	if (PRC_IS_REDIRECT(cmd))
730 		return;
731 
732 	/*
733 	 * Hostdead is ugly because it goes linearly through all PCBs.
734 	 *
735 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
736 	 * DoS attack on machines with many connections.
737 	 */
738 	if (cmd == PRC_HOSTDEAD)
739 		ip = NULL;
740 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
741 		return;
742 	if (ip != NULL) {
743 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
744 		inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
745 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
746 		if (inp != NULL) {
747 			INP_RLOCK_ASSERT(inp);
748 			if (inp->inp_socket != NULL) {
749 				udp_notify(inp, inetctlerrmap[cmd]);
750 			}
751 			INP_RUNLOCK(inp);
752 		}
753 	} else
754 		in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
755 		    udp_notify);
756 }
757 void
758 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
759 {
760 
761 	return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
762 }
763 
764 void
765 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
766 {
767 
768 	return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
769 }
770 #endif /* INET */
771 
772 static int
773 udp_pcblist(SYSCTL_HANDLER_ARGS)
774 {
775 	int error, i, n;
776 	struct inpcb *inp, **inp_list;
777 	inp_gen_t gencnt;
778 	struct xinpgen xig;
779 
780 	/*
781 	 * The process of preparing the PCB list is too time-consuming and
782 	 * resource-intensive to repeat twice on every request.
783 	 */
784 	if (req->oldptr == 0) {
785 		n = V_udbinfo.ipi_count;
786 		n += imax(n / 8, 10);
787 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
788 		return (0);
789 	}
790 
791 	if (req->newptr != 0)
792 		return (EPERM);
793 
794 	/*
795 	 * OK, now we're committed to doing something.
796 	 */
797 	INP_INFO_RLOCK(&V_udbinfo);
798 	gencnt = V_udbinfo.ipi_gencnt;
799 	n = V_udbinfo.ipi_count;
800 	INP_INFO_RUNLOCK(&V_udbinfo);
801 
802 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
803 		+ n * sizeof(struct xinpcb));
804 	if (error != 0)
805 		return (error);
806 
807 	xig.xig_len = sizeof xig;
808 	xig.xig_count = n;
809 	xig.xig_gen = gencnt;
810 	xig.xig_sogen = so_gencnt;
811 	error = SYSCTL_OUT(req, &xig, sizeof xig);
812 	if (error)
813 		return (error);
814 
815 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
816 	if (inp_list == 0)
817 		return (ENOMEM);
818 
819 	INP_INFO_RLOCK(&V_udbinfo);
820 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
821 	     inp = LIST_NEXT(inp, inp_list)) {
822 		INP_WLOCK(inp);
823 		if (inp->inp_gencnt <= gencnt &&
824 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
825 			in_pcbref(inp);
826 			inp_list[i++] = inp;
827 		}
828 		INP_WUNLOCK(inp);
829 	}
830 	INP_INFO_RUNLOCK(&V_udbinfo);
831 	n = i;
832 
833 	error = 0;
834 	for (i = 0; i < n; i++) {
835 		inp = inp_list[i];
836 		INP_RLOCK(inp);
837 		if (inp->inp_gencnt <= gencnt) {
838 			struct xinpcb xi;
839 
840 			bzero(&xi, sizeof(xi));
841 			xi.xi_len = sizeof xi;
842 			/* XXX should avoid extra copy */
843 			bcopy(inp, &xi.xi_inp, sizeof *inp);
844 			if (inp->inp_socket)
845 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
846 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
847 			INP_RUNLOCK(inp);
848 			error = SYSCTL_OUT(req, &xi, sizeof xi);
849 		} else
850 			INP_RUNLOCK(inp);
851 	}
852 	INP_INFO_WLOCK(&V_udbinfo);
853 	for (i = 0; i < n; i++) {
854 		inp = inp_list[i];
855 		INP_RLOCK(inp);
856 		if (!in_pcbrele_rlocked(inp))
857 			INP_RUNLOCK(inp);
858 	}
859 	INP_INFO_WUNLOCK(&V_udbinfo);
860 
861 	if (!error) {
862 		/*
863 		 * Give the user an updated idea of our state.  If the
864 		 * generation differs from what we told her before, she knows
865 		 * that something happened while we were processing this
866 		 * request, and it might be necessary to retry.
867 		 */
868 		INP_INFO_RLOCK(&V_udbinfo);
869 		xig.xig_gen = V_udbinfo.ipi_gencnt;
870 		xig.xig_sogen = so_gencnt;
871 		xig.xig_count = V_udbinfo.ipi_count;
872 		INP_INFO_RUNLOCK(&V_udbinfo);
873 		error = SYSCTL_OUT(req, &xig, sizeof xig);
874 	}
875 	free(inp_list, M_TEMP);
876 	return (error);
877 }
878 
879 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
880     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
881     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
882 
883 #ifdef INET
884 static int
885 udp_getcred(SYSCTL_HANDLER_ARGS)
886 {
887 	struct xucred xuc;
888 	struct sockaddr_in addrs[2];
889 	struct inpcb *inp;
890 	int error;
891 
892 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
893 	if (error)
894 		return (error);
895 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
896 	if (error)
897 		return (error);
898 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
899 	    addrs[0].sin_addr, addrs[0].sin_port,
900 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
901 	if (inp != NULL) {
902 		INP_RLOCK_ASSERT(inp);
903 		if (inp->inp_socket == NULL)
904 			error = ENOENT;
905 		if (error == 0)
906 			error = cr_canseeinpcb(req->td->td_ucred, inp);
907 		if (error == 0)
908 			cru2x(inp->inp_cred, &xuc);
909 		INP_RUNLOCK(inp);
910 	} else
911 		error = ENOENT;
912 	if (error == 0)
913 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
914 	return (error);
915 }
916 
917 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
918     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
919     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
920 #endif /* INET */
921 
922 int
923 udp_ctloutput(struct socket *so, struct sockopt *sopt)
924 {
925 	struct inpcb *inp;
926 	struct udpcb *up;
927 	int isudplite, error, optval;
928 
929 	error = 0;
930 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
931 	inp = sotoinpcb(so);
932 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
933 	INP_WLOCK(inp);
934 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
935 #ifdef INET6
936 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
937 			INP_WUNLOCK(inp);
938 			error = ip6_ctloutput(so, sopt);
939 		}
940 #endif
941 #if defined(INET) && defined(INET6)
942 		else
943 #endif
944 #ifdef INET
945 		{
946 			INP_WUNLOCK(inp);
947 			error = ip_ctloutput(so, sopt);
948 		}
949 #endif
950 		return (error);
951 	}
952 
953 	switch (sopt->sopt_dir) {
954 	case SOPT_SET:
955 		switch (sopt->sopt_name) {
956 		case UDP_ENCAP:
957 			INP_WUNLOCK(inp);
958 			error = sooptcopyin(sopt, &optval, sizeof optval,
959 					    sizeof optval);
960 			if (error)
961 				break;
962 			inp = sotoinpcb(so);
963 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
964 			INP_WLOCK(inp);
965 #ifdef IPSEC_NAT_T
966 			up = intoudpcb(inp);
967 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
968 #endif
969 			switch (optval) {
970 			case 0:
971 				/* Clear all UDP encap. */
972 #ifdef IPSEC_NAT_T
973 				up->u_flags &= ~UF_ESPINUDP_ALL;
974 #endif
975 				break;
976 #ifdef IPSEC_NAT_T
977 			case UDP_ENCAP_ESPINUDP:
978 			case UDP_ENCAP_ESPINUDP_NON_IKE:
979 				up->u_flags &= ~UF_ESPINUDP_ALL;
980 				if (optval == UDP_ENCAP_ESPINUDP)
981 					up->u_flags |= UF_ESPINUDP;
982 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
983 					up->u_flags |= UF_ESPINUDP_NON_IKE;
984 				break;
985 #endif
986 			default:
987 				error = EINVAL;
988 				break;
989 			}
990 			INP_WUNLOCK(inp);
991 			break;
992 		case UDPLITE_SEND_CSCOV:
993 		case UDPLITE_RECV_CSCOV:
994 			if (!isudplite) {
995 				INP_WUNLOCK(inp);
996 				error = ENOPROTOOPT;
997 				break;
998 			}
999 			INP_WUNLOCK(inp);
1000 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1001 			    sizeof(optval));
1002 			if (error != 0)
1003 				break;
1004 			inp = sotoinpcb(so);
1005 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1006 			INP_WLOCK(inp);
1007 			up = intoudpcb(inp);
1008 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1009 			if (optval != 0 && optval < 8) {
1010 				INP_WUNLOCK(inp);
1011 				error = EINVAL;
1012 				break;
1013 			}
1014 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1015 				up->u_txcslen = optval;
1016 			else
1017 				up->u_rxcslen = optval;
1018 			INP_WUNLOCK(inp);
1019 			break;
1020 		default:
1021 			INP_WUNLOCK(inp);
1022 			error = ENOPROTOOPT;
1023 			break;
1024 		}
1025 		break;
1026 	case SOPT_GET:
1027 		switch (sopt->sopt_name) {
1028 #ifdef IPSEC_NAT_T
1029 		case UDP_ENCAP:
1030 			up = intoudpcb(inp);
1031 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1032 			optval = up->u_flags & UF_ESPINUDP_ALL;
1033 			INP_WUNLOCK(inp);
1034 			error = sooptcopyout(sopt, &optval, sizeof optval);
1035 			break;
1036 #endif
1037 		case UDPLITE_SEND_CSCOV:
1038 		case UDPLITE_RECV_CSCOV:
1039 			if (!isudplite) {
1040 				INP_WUNLOCK(inp);
1041 				error = ENOPROTOOPT;
1042 				break;
1043 			}
1044 			up = intoudpcb(inp);
1045 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1046 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1047 				optval = up->u_txcslen;
1048 			else
1049 				optval = up->u_rxcslen;
1050 			INP_WUNLOCK(inp);
1051 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1052 			break;
1053 		default:
1054 			INP_WUNLOCK(inp);
1055 			error = ENOPROTOOPT;
1056 			break;
1057 		}
1058 		break;
1059 	}
1060 	return (error);
1061 }
1062 
1063 #ifdef INET
1064 #define	UH_WLOCKED	2
1065 #define	UH_RLOCKED	1
1066 #define	UH_UNLOCKED	0
1067 static int
1068 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1069     struct mbuf *control, struct thread *td)
1070 {
1071 	struct udpiphdr *ui;
1072 	int len = m->m_pkthdr.len;
1073 	struct in_addr faddr, laddr;
1074 	struct cmsghdr *cm;
1075 	struct inpcbinfo *pcbinfo;
1076 	struct sockaddr_in *sin, src;
1077 	int cscov_partial = 0;
1078 	int error = 0;
1079 	int ipflags;
1080 	u_short fport, lport;
1081 	int unlock_udbinfo;
1082 	u_char tos;
1083 	uint8_t pr;
1084 	uint16_t cscov = 0;
1085 
1086 	/*
1087 	 * udp_output() may need to temporarily bind or connect the current
1088 	 * inpcb.  As such, we don't know up front whether we will need the
1089 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1090 	 * front before acquiring any locks.
1091 	 */
1092 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1093 		if (control)
1094 			m_freem(control);
1095 		m_freem(m);
1096 		return (EMSGSIZE);
1097 	}
1098 
1099 	src.sin_family = 0;
1100 	INP_RLOCK(inp);
1101 	tos = inp->inp_ip_tos;
1102 	if (control != NULL) {
1103 		/*
1104 		 * XXX: Currently, we assume all the optional information is
1105 		 * stored in a single mbuf.
1106 		 */
1107 		if (control->m_next) {
1108 			INP_RUNLOCK(inp);
1109 			m_freem(control);
1110 			m_freem(m);
1111 			return (EINVAL);
1112 		}
1113 		for (; control->m_len > 0;
1114 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1115 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1116 			cm = mtod(control, struct cmsghdr *);
1117 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1118 			    || cm->cmsg_len > control->m_len) {
1119 				error = EINVAL;
1120 				break;
1121 			}
1122 			if (cm->cmsg_level != IPPROTO_IP)
1123 				continue;
1124 
1125 			switch (cm->cmsg_type) {
1126 			case IP_SENDSRCADDR:
1127 				if (cm->cmsg_len !=
1128 				    CMSG_LEN(sizeof(struct in_addr))) {
1129 					error = EINVAL;
1130 					break;
1131 				}
1132 				bzero(&src, sizeof(src));
1133 				src.sin_family = AF_INET;
1134 				src.sin_len = sizeof(src);
1135 				src.sin_port = inp->inp_lport;
1136 				src.sin_addr =
1137 				    *(struct in_addr *)CMSG_DATA(cm);
1138 				break;
1139 
1140 			case IP_TOS:
1141 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1142 					error = EINVAL;
1143 					break;
1144 				}
1145 				tos = *(u_char *)CMSG_DATA(cm);
1146 				break;
1147 
1148 			default:
1149 				error = ENOPROTOOPT;
1150 				break;
1151 			}
1152 			if (error)
1153 				break;
1154 		}
1155 		m_freem(control);
1156 	}
1157 	if (error) {
1158 		INP_RUNLOCK(inp);
1159 		m_freem(m);
1160 		return (error);
1161 	}
1162 
1163 	/*
1164 	 * Depending on whether or not the application has bound or connected
1165 	 * the socket, we may have to do varying levels of work.  The optimal
1166 	 * case is for a connected UDP socket, as a global lock isn't
1167 	 * required at all.
1168 	 *
1169 	 * In order to decide which we need, we require stability of the
1170 	 * inpcb binding, which we ensure by acquiring a read lock on the
1171 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1172 	 * the trylock and retry game; note that we may end up with more
1173 	 * conservative locks than required the second time around, so later
1174 	 * assertions have to accept that.  Further analysis of the number of
1175 	 * misses under contention is required.
1176 	 *
1177 	 * XXXRW: Check that hash locking update here is correct.
1178 	 */
1179 	pr = inp->inp_socket->so_proto->pr_protocol;
1180 	pcbinfo = get_inpcbinfo(pr);
1181 	sin = (struct sockaddr_in *)addr;
1182 	if (sin != NULL &&
1183 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1184 		INP_RUNLOCK(inp);
1185 		INP_WLOCK(inp);
1186 		INP_HASH_WLOCK(pcbinfo);
1187 		unlock_udbinfo = UH_WLOCKED;
1188 	} else if ((sin != NULL && (
1189 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1190 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1191 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1192 	    (inp->inp_lport == 0))) ||
1193 	    (src.sin_family == AF_INET)) {
1194 		INP_HASH_RLOCK(pcbinfo);
1195 		unlock_udbinfo = UH_RLOCKED;
1196 	} else
1197 		unlock_udbinfo = UH_UNLOCKED;
1198 
1199 	/*
1200 	 * If the IP_SENDSRCADDR control message was specified, override the
1201 	 * source address for this datagram.  Its use is invalidated if the
1202 	 * address thus specified is incomplete or clobbers other inpcbs.
1203 	 */
1204 	laddr = inp->inp_laddr;
1205 	lport = inp->inp_lport;
1206 	if (src.sin_family == AF_INET) {
1207 		INP_HASH_LOCK_ASSERT(pcbinfo);
1208 		if ((lport == 0) ||
1209 		    (laddr.s_addr == INADDR_ANY &&
1210 		     src.sin_addr.s_addr == INADDR_ANY)) {
1211 			error = EINVAL;
1212 			goto release;
1213 		}
1214 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1215 		    &laddr.s_addr, &lport, td->td_ucred);
1216 		if (error)
1217 			goto release;
1218 	}
1219 
1220 	/*
1221 	 * If a UDP socket has been connected, then a local address/port will
1222 	 * have been selected and bound.
1223 	 *
1224 	 * If a UDP socket has not been connected to, then an explicit
1225 	 * destination address must be used, in which case a local
1226 	 * address/port may not have been selected and bound.
1227 	 */
1228 	if (sin != NULL) {
1229 		INP_LOCK_ASSERT(inp);
1230 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1231 			error = EISCONN;
1232 			goto release;
1233 		}
1234 
1235 		/*
1236 		 * Jail may rewrite the destination address, so let it do
1237 		 * that before we use it.
1238 		 */
1239 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1240 		if (error)
1241 			goto release;
1242 
1243 		/*
1244 		 * If a local address or port hasn't yet been selected, or if
1245 		 * the destination address needs to be rewritten due to using
1246 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1247 		 * to do the heavy lifting.  Once a port is selected, we
1248 		 * commit the binding back to the socket; we also commit the
1249 		 * binding of the address if in jail.
1250 		 *
1251 		 * If we already have a valid binding and we're not
1252 		 * requesting a destination address rewrite, use a fast path.
1253 		 */
1254 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1255 		    inp->inp_lport == 0 ||
1256 		    sin->sin_addr.s_addr == INADDR_ANY ||
1257 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1258 			INP_HASH_LOCK_ASSERT(pcbinfo);
1259 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1260 			    &lport, &faddr.s_addr, &fport, NULL,
1261 			    td->td_ucred);
1262 			if (error)
1263 				goto release;
1264 
1265 			/*
1266 			 * XXXRW: Why not commit the port if the address is
1267 			 * !INADDR_ANY?
1268 			 */
1269 			/* Commit the local port if newly assigned. */
1270 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1271 			    inp->inp_lport == 0) {
1272 				INP_WLOCK_ASSERT(inp);
1273 				INP_HASH_WLOCK_ASSERT(pcbinfo);
1274 				/*
1275 				 * Remember addr if jailed, to prevent
1276 				 * rebinding.
1277 				 */
1278 				if (prison_flag(td->td_ucred, PR_IP4))
1279 					inp->inp_laddr = laddr;
1280 				inp->inp_lport = lport;
1281 				if (in_pcbinshash(inp) != 0) {
1282 					inp->inp_lport = 0;
1283 					error = EAGAIN;
1284 					goto release;
1285 				}
1286 				inp->inp_flags |= INP_ANONPORT;
1287 			}
1288 		} else {
1289 			faddr = sin->sin_addr;
1290 			fport = sin->sin_port;
1291 		}
1292 	} else {
1293 		INP_LOCK_ASSERT(inp);
1294 		faddr = inp->inp_faddr;
1295 		fport = inp->inp_fport;
1296 		if (faddr.s_addr == INADDR_ANY) {
1297 			error = ENOTCONN;
1298 			goto release;
1299 		}
1300 	}
1301 
1302 	/*
1303 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1304 	 * link-layer headers.  Immediate slide the data pointer back forward
1305 	 * since we won't use that space at this layer.
1306 	 */
1307 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1308 	if (m == NULL) {
1309 		error = ENOBUFS;
1310 		goto release;
1311 	}
1312 	m->m_data += max_linkhdr;
1313 	m->m_len -= max_linkhdr;
1314 	m->m_pkthdr.len -= max_linkhdr;
1315 
1316 	/*
1317 	 * Fill in mbuf with extended UDP header and addresses and length put
1318 	 * into network format.
1319 	 */
1320 	ui = mtod(m, struct udpiphdr *);
1321 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1322 	ui->ui_pr = pr;
1323 	ui->ui_src = laddr;
1324 	ui->ui_dst = faddr;
1325 	ui->ui_sport = lport;
1326 	ui->ui_dport = fport;
1327 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1328 	if (pr == IPPROTO_UDPLITE) {
1329 		struct udpcb *up;
1330 		uint16_t plen;
1331 
1332 		up = intoudpcb(inp);
1333 		cscov = up->u_txcslen;
1334 		plen = (u_short)len + sizeof(struct udphdr);
1335 		if (cscov >= plen)
1336 			cscov = 0;
1337 		ui->ui_len = htons(plen);
1338 		ui->ui_ulen = htons(cscov);
1339 		/*
1340 		 * For UDP-Lite, checksum coverage length of zero means
1341 		 * the entire UDPLite packet is covered by the checksum.
1342 		 */
1343 		cscov_partial = (cscov == 0) ? 0 : 1;
1344 	} else
1345 		ui->ui_v = IPVERSION << 4;
1346 
1347 	/*
1348 	 * Set the Don't Fragment bit in the IP header.
1349 	 */
1350 	if (inp->inp_flags & INP_DONTFRAG) {
1351 		struct ip *ip;
1352 
1353 		ip = (struct ip *)&ui->ui_i;
1354 		ip->ip_off |= htons(IP_DF);
1355 	}
1356 
1357 	ipflags = 0;
1358 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1359 		ipflags |= IP_ROUTETOIF;
1360 	if (inp->inp_socket->so_options & SO_BROADCAST)
1361 		ipflags |= IP_ALLOWBROADCAST;
1362 	if (inp->inp_flags & INP_ONESBCAST)
1363 		ipflags |= IP_SENDONES;
1364 
1365 #ifdef MAC
1366 	mac_inpcb_create_mbuf(inp, m);
1367 #endif
1368 
1369 	/*
1370 	 * Set up checksum and output datagram.
1371 	 */
1372 	ui->ui_sum = 0;
1373 	if (pr == IPPROTO_UDPLITE) {
1374 		if (inp->inp_flags & INP_ONESBCAST)
1375 			faddr.s_addr = INADDR_BROADCAST;
1376 		if (cscov_partial) {
1377 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1378 				ui->ui_sum = 0xffff;
1379 		} else {
1380 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1381 				ui->ui_sum = 0xffff;
1382 		}
1383 	} else if (V_udp_cksum) {
1384 		if (inp->inp_flags & INP_ONESBCAST)
1385 			faddr.s_addr = INADDR_BROADCAST;
1386 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1387 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1388 		m->m_pkthdr.csum_flags = CSUM_UDP;
1389 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1390 	}
1391 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1392 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1393 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1394 	UDPSTAT_INC(udps_opackets);
1395 
1396 	if (unlock_udbinfo == UH_WLOCKED)
1397 		INP_HASH_WUNLOCK(pcbinfo);
1398 	else if (unlock_udbinfo == UH_RLOCKED)
1399 		INP_HASH_RUNLOCK(pcbinfo);
1400 	UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1401 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1402 	    inp->inp_moptions, inp);
1403 	if (unlock_udbinfo == UH_WLOCKED)
1404 		INP_WUNLOCK(inp);
1405 	else
1406 		INP_RUNLOCK(inp);
1407 	return (error);
1408 
1409 release:
1410 	if (unlock_udbinfo == UH_WLOCKED) {
1411 		INP_HASH_WUNLOCK(pcbinfo);
1412 		INP_WUNLOCK(inp);
1413 	} else if (unlock_udbinfo == UH_RLOCKED) {
1414 		INP_HASH_RUNLOCK(pcbinfo);
1415 		INP_RUNLOCK(inp);
1416 	} else
1417 		INP_RUNLOCK(inp);
1418 	m_freem(m);
1419 	return (error);
1420 }
1421 
1422 
1423 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1424 /*
1425  * Potentially decap ESP in UDP frame.  Check for an ESP header
1426  * and optional marker; if present, strip the UDP header and
1427  * push the result through IPSec.
1428  *
1429  * Returns mbuf to be processed (potentially re-allocated) or
1430  * NULL if consumed and/or processed.
1431  */
1432 static struct mbuf *
1433 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1434 {
1435 	size_t minlen, payload, skip, iphlen;
1436 	caddr_t data;
1437 	struct udpcb *up;
1438 	struct m_tag *tag;
1439 	struct udphdr *udphdr;
1440 	struct ip *ip;
1441 
1442 	INP_RLOCK_ASSERT(inp);
1443 
1444 	/*
1445 	 * Pull up data so the longest case is contiguous:
1446 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1447 	 */
1448 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1449 	if (minlen > m->m_pkthdr.len)
1450 		minlen = m->m_pkthdr.len;
1451 	if ((m = m_pullup(m, minlen)) == NULL) {
1452 		IPSECSTAT_INC(ips_in_inval);
1453 		return (NULL);		/* Bypass caller processing. */
1454 	}
1455 	data = mtod(m, caddr_t);	/* Points to ip header. */
1456 	payload = m->m_len - off;	/* Size of payload. */
1457 
1458 	if (payload == 1 && data[off] == '\xff')
1459 		return (m);		/* NB: keepalive packet, no decap. */
1460 
1461 	up = intoudpcb(inp);
1462 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1463 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1464 	    ("u_flags 0x%x", up->u_flags));
1465 
1466 	/*
1467 	 * Check that the payload is large enough to hold an
1468 	 * ESP header and compute the amount of data to remove.
1469 	 *
1470 	 * NB: the caller has already done a pullup for us.
1471 	 * XXX can we assume alignment and eliminate bcopys?
1472 	 */
1473 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1474 		/*
1475 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1476 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1477 		 * possible AH mode non-IKE marker+non-ESP marker
1478 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1479 		 */
1480 		uint64_t marker;
1481 
1482 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1483 			return (m);	/* NB: no decap. */
1484 		bcopy(data + off, &marker, sizeof(uint64_t));
1485 		if (marker != 0)	/* Non-IKE marker. */
1486 			return (m);	/* NB: no decap. */
1487 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1488 	} else {
1489 		uint32_t spi;
1490 
1491 		if (payload <= sizeof(struct esp)) {
1492 			IPSECSTAT_INC(ips_in_inval);
1493 			m_freem(m);
1494 			return (NULL);	/* Discard. */
1495 		}
1496 		bcopy(data + off, &spi, sizeof(uint32_t));
1497 		if (spi == 0)		/* Non-ESP marker. */
1498 			return (m);	/* NB: no decap. */
1499 		skip = sizeof(struct udphdr);
1500 	}
1501 
1502 	/*
1503 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1504 	 * the UDP ports. This is required if we want to select
1505 	 * the right SPD for multiple hosts behind same NAT.
1506 	 *
1507 	 * NB: ports are maintained in network byte order everywhere
1508 	 *     in the NAT-T code.
1509 	 */
1510 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1511 		2 * sizeof(uint16_t), M_NOWAIT);
1512 	if (tag == NULL) {
1513 		IPSECSTAT_INC(ips_in_nomem);
1514 		m_freem(m);
1515 		return (NULL);		/* Discard. */
1516 	}
1517 	iphlen = off - sizeof(struct udphdr);
1518 	udphdr = (struct udphdr *)(data + iphlen);
1519 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1520 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1521 	m_tag_prepend(m, tag);
1522 
1523 	/*
1524 	 * Remove the UDP header (and possibly the non ESP marker)
1525 	 * IP header length is iphlen
1526 	 * Before:
1527 	 *   <--- off --->
1528 	 *   +----+------+-----+
1529 	 *   | IP |  UDP | ESP |
1530 	 *   +----+------+-----+
1531 	 *        <-skip->
1532 	 * After:
1533 	 *          +----+-----+
1534 	 *          | IP | ESP |
1535 	 *          +----+-----+
1536 	 *   <-skip->
1537 	 */
1538 	ovbcopy(data, data + skip, iphlen);
1539 	m_adj(m, skip);
1540 
1541 	ip = mtod(m, struct ip *);
1542 	ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1543 	ip->ip_p = IPPROTO_ESP;
1544 
1545 	/*
1546 	 * We cannot yet update the cksums so clear any
1547 	 * h/w cksum flags as they are no longer valid.
1548 	 */
1549 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1550 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1551 
1552 	(void) ipsec4_common_input(m, iphlen, ip->ip_p);
1553 	return (NULL);			/* NB: consumed, bypass processing. */
1554 }
1555 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1556 
1557 static void
1558 udp_abort(struct socket *so)
1559 {
1560 	struct inpcb *inp;
1561 	struct inpcbinfo *pcbinfo;
1562 
1563 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1564 	inp = sotoinpcb(so);
1565 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1566 	INP_WLOCK(inp);
1567 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1568 		INP_HASH_WLOCK(pcbinfo);
1569 		in_pcbdisconnect(inp);
1570 		inp->inp_laddr.s_addr = INADDR_ANY;
1571 		INP_HASH_WUNLOCK(pcbinfo);
1572 		soisdisconnected(so);
1573 	}
1574 	INP_WUNLOCK(inp);
1575 }
1576 
1577 static int
1578 udp_attach(struct socket *so, int proto, struct thread *td)
1579 {
1580 	struct inpcb *inp;
1581 	struct inpcbinfo *pcbinfo;
1582 	int error;
1583 
1584 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1585 	inp = sotoinpcb(so);
1586 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1587 	error = soreserve(so, udp_sendspace, udp_recvspace);
1588 	if (error)
1589 		return (error);
1590 	INP_INFO_WLOCK(pcbinfo);
1591 	error = in_pcballoc(so, pcbinfo);
1592 	if (error) {
1593 		INP_INFO_WUNLOCK(pcbinfo);
1594 		return (error);
1595 	}
1596 
1597 	inp = sotoinpcb(so);
1598 	inp->inp_vflag |= INP_IPV4;
1599 	inp->inp_ip_ttl = V_ip_defttl;
1600 
1601 	error = udp_newudpcb(inp);
1602 	if (error) {
1603 		in_pcbdetach(inp);
1604 		in_pcbfree(inp);
1605 		INP_INFO_WUNLOCK(pcbinfo);
1606 		return (error);
1607 	}
1608 
1609 	INP_WUNLOCK(inp);
1610 	INP_INFO_WUNLOCK(pcbinfo);
1611 	return (0);
1612 }
1613 #endif /* INET */
1614 
1615 int
1616 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1617 {
1618 	struct inpcb *inp;
1619 	struct udpcb *up;
1620 
1621 	KASSERT(so->so_type == SOCK_DGRAM,
1622 	    ("udp_set_kernel_tunneling: !dgram"));
1623 	inp = sotoinpcb(so);
1624 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1625 	INP_WLOCK(inp);
1626 	up = intoudpcb(inp);
1627 	if (up->u_tun_func != NULL) {
1628 		INP_WUNLOCK(inp);
1629 		return (EBUSY);
1630 	}
1631 	up->u_tun_func = f;
1632 	INP_WUNLOCK(inp);
1633 	return (0);
1634 }
1635 
1636 #ifdef INET
1637 static int
1638 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1639 {
1640 	struct inpcb *inp;
1641 	struct inpcbinfo *pcbinfo;
1642 	int error;
1643 
1644 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1645 	inp = sotoinpcb(so);
1646 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1647 	INP_WLOCK(inp);
1648 	INP_HASH_WLOCK(pcbinfo);
1649 	error = in_pcbbind(inp, nam, td->td_ucred);
1650 	INP_HASH_WUNLOCK(pcbinfo);
1651 	INP_WUNLOCK(inp);
1652 	return (error);
1653 }
1654 
1655 static void
1656 udp_close(struct socket *so)
1657 {
1658 	struct inpcb *inp;
1659 	struct inpcbinfo *pcbinfo;
1660 
1661 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1662 	inp = sotoinpcb(so);
1663 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1664 	INP_WLOCK(inp);
1665 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1666 		INP_HASH_WLOCK(pcbinfo);
1667 		in_pcbdisconnect(inp);
1668 		inp->inp_laddr.s_addr = INADDR_ANY;
1669 		INP_HASH_WUNLOCK(pcbinfo);
1670 		soisdisconnected(so);
1671 	}
1672 	INP_WUNLOCK(inp);
1673 }
1674 
1675 static int
1676 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1677 {
1678 	struct inpcb *inp;
1679 	struct inpcbinfo *pcbinfo;
1680 	struct sockaddr_in *sin;
1681 	int error;
1682 
1683 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1684 	inp = sotoinpcb(so);
1685 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1686 	INP_WLOCK(inp);
1687 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1688 		INP_WUNLOCK(inp);
1689 		return (EISCONN);
1690 	}
1691 	sin = (struct sockaddr_in *)nam;
1692 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1693 	if (error != 0) {
1694 		INP_WUNLOCK(inp);
1695 		return (error);
1696 	}
1697 	INP_HASH_WLOCK(pcbinfo);
1698 	error = in_pcbconnect(inp, nam, td->td_ucred);
1699 	INP_HASH_WUNLOCK(pcbinfo);
1700 	if (error == 0)
1701 		soisconnected(so);
1702 	INP_WUNLOCK(inp);
1703 	return (error);
1704 }
1705 
1706 static void
1707 udp_detach(struct socket *so)
1708 {
1709 	struct inpcb *inp;
1710 	struct inpcbinfo *pcbinfo;
1711 	struct udpcb *up;
1712 
1713 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1714 	inp = sotoinpcb(so);
1715 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1716 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1717 	    ("udp_detach: not disconnected"));
1718 	INP_INFO_WLOCK(pcbinfo);
1719 	INP_WLOCK(inp);
1720 	up = intoudpcb(inp);
1721 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1722 	inp->inp_ppcb = NULL;
1723 	in_pcbdetach(inp);
1724 	in_pcbfree(inp);
1725 	INP_INFO_WUNLOCK(pcbinfo);
1726 	udp_discardcb(up);
1727 }
1728 
1729 static int
1730 udp_disconnect(struct socket *so)
1731 {
1732 	struct inpcb *inp;
1733 	struct inpcbinfo *pcbinfo;
1734 
1735 	pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol);
1736 	inp = sotoinpcb(so);
1737 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1738 	INP_WLOCK(inp);
1739 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1740 		INP_WUNLOCK(inp);
1741 		return (ENOTCONN);
1742 	}
1743 	INP_HASH_WLOCK(pcbinfo);
1744 	in_pcbdisconnect(inp);
1745 	inp->inp_laddr.s_addr = INADDR_ANY;
1746 	INP_HASH_WUNLOCK(pcbinfo);
1747 	SOCK_LOCK(so);
1748 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1749 	SOCK_UNLOCK(so);
1750 	INP_WUNLOCK(inp);
1751 	return (0);
1752 }
1753 
1754 static int
1755 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1756     struct mbuf *control, struct thread *td)
1757 {
1758 	struct inpcb *inp;
1759 
1760 	inp = sotoinpcb(so);
1761 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1762 	return (udp_output(inp, m, addr, control, td));
1763 }
1764 #endif /* INET */
1765 
1766 int
1767 udp_shutdown(struct socket *so)
1768 {
1769 	struct inpcb *inp;
1770 
1771 	inp = sotoinpcb(so);
1772 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1773 	INP_WLOCK(inp);
1774 	socantsendmore(so);
1775 	INP_WUNLOCK(inp);
1776 	return (0);
1777 }
1778 
1779 #ifdef INET
1780 struct pr_usrreqs udp_usrreqs = {
1781 	.pru_abort =		udp_abort,
1782 	.pru_attach =		udp_attach,
1783 	.pru_bind =		udp_bind,
1784 	.pru_connect =		udp_connect,
1785 	.pru_control =		in_control,
1786 	.pru_detach =		udp_detach,
1787 	.pru_disconnect =	udp_disconnect,
1788 	.pru_peeraddr =		in_getpeeraddr,
1789 	.pru_send =		udp_send,
1790 	.pru_soreceive =	soreceive_dgram,
1791 	.pru_sosend =		sosend_dgram,
1792 	.pru_shutdown =		udp_shutdown,
1793 	.pru_sockaddr =		in_getsockaddr,
1794 	.pru_sosetlabel =	in_pcbsosetlabel,
1795 	.pru_close =		udp_close,
1796 };
1797 #endif /* INET */
1798