xref: /freebsd/sys/netinet/udp_usrreq.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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  * All rights reserved.
7  *
8  * Portions of this software were developed by Robert N. M. Watson under
9  * contract to Juniper Networks, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ipfw.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/sx.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/systm.h>
64 
65 #include <vm/uma.h>
66 
67 #include <net/if.h>
68 #include <net/route.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
75 #ifdef INET6
76 #include <netinet/ip6.h>
77 #endif
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/ip_options.h>
82 #ifdef INET6
83 #include <netinet6/ip6_var.h>
84 #endif
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
87 
88 #ifdef IPSEC
89 #include <netipsec/ipsec.h>
90 #include <netipsec/esp.h>
91 #endif
92 
93 #include <machine/in_cksum.h>
94 
95 #include <security/mac/mac_framework.h>
96 
97 /*
98  * UDP protocol implementation.
99  * Per RFC 768, August, 1980.
100  */
101 
102 /*
103  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
104  * removes the only data integrity mechanism for packets and malformed
105  * packets that would otherwise be discarded due to bad checksums, and may
106  * cause problems (especially for NFS data blocks).
107  */
108 static int	udp_cksum = 1;
109 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
110     0, "compute udp checksum");
111 
112 int	udp_log_in_vain = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114     &udp_log_in_vain, 0, "Log all incoming UDP packets");
115 
116 VNET_DEFINE(int, udp_blackhole) = 0;
117 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
118     &VNET_NAME(udp_blackhole), 0,
119     "Do not send port unreachables for refused connects");
120 
121 u_long	udp_sendspace = 9216;		/* really max datagram size */
122 					/* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
125 
126 u_long	udp_recvspace = 40 * (1024 +
127 #ifdef INET6
128 				      sizeof(struct sockaddr_in6)
129 #else
130 				      sizeof(struct sockaddr_in)
131 #endif
132 				      );
133 
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
136 
137 VNET_DEFINE(struct inpcbhead, udb);		/* from udp_var.h */
138 VNET_DEFINE(struct inpcbinfo, udbinfo);
139 static VNET_DEFINE(uma_zone_t, udpcb_zone);
140 #define	V_udpcb_zone			VNET(udpcb_zone)
141 
142 #ifndef UDBHASHSIZE
143 #define	UDBHASHSIZE	128
144 #endif
145 
146 VNET_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
147 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
148     &VNET_NAME(udpstat), udpstat,
149     "UDP statistics (struct udpstat, netinet/udp_var.h)");
150 
151 #ifdef INET
152 static void	udp_detach(struct socket *so);
153 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
154 		    struct mbuf *, struct thread *);
155 #endif
156 
157 #ifdef IPSEC
158 #ifdef IPSEC_NAT_T
159 #define	UF_ESPINUDP_ALL	(UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
160 #ifdef INET
161 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
162 #endif
163 #endif /* IPSEC_NAT_T */
164 #endif /* IPSEC */
165 
166 static void
167 udp_zone_change(void *tag)
168 {
169 
170 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
171 	uma_zone_set_max(V_udpcb_zone, maxsockets);
172 }
173 
174 static int
175 udp_inpcb_init(void *mem, int size, int flags)
176 {
177 	struct inpcb *inp;
178 
179 	inp = mem;
180 	INP_LOCK_INIT(inp, "inp", "udpinp");
181 	return (0);
182 }
183 
184 void
185 udp_init(void)
186 {
187 
188 	in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
189 	    "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE);
190 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
191 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
192 	uma_zone_set_max(V_udpcb_zone, maxsockets);
193 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
194 	    EVENTHANDLER_PRI_ANY);
195 }
196 
197 /*
198  * Kernel module interface for updating udpstat.  The argument is an index
199  * into udpstat treated as an array of u_long.  While this encodes the
200  * general layout of udpstat into the caller, it doesn't encode its location,
201  * so that future changes to add, for example, per-CPU stats support won't
202  * cause binary compatibility problems for kernel modules.
203  */
204 void
205 kmod_udpstat_inc(int statnum)
206 {
207 
208 	(*((u_long *)&V_udpstat + statnum))++;
209 }
210 
211 int
212 udp_newudpcb(struct inpcb *inp)
213 {
214 	struct udpcb *up;
215 
216 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
217 	if (up == NULL)
218 		return (ENOBUFS);
219 	inp->inp_ppcb = up;
220 	return (0);
221 }
222 
223 void
224 udp_discardcb(struct udpcb *up)
225 {
226 
227 	uma_zfree(V_udpcb_zone, up);
228 }
229 
230 #ifdef VIMAGE
231 void
232 udp_destroy(void)
233 {
234 
235 	in_pcbinfo_destroy(&V_udbinfo);
236 	uma_zdestroy(V_udpcb_zone);
237 }
238 #endif
239 
240 #ifdef INET
241 /*
242  * Subroutine of udp_input(), which appends the provided mbuf chain to the
243  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
244  * contains the source address.  If the socket ends up being an IPv6 socket,
245  * udp_append() will convert to a sockaddr_in6 before passing the address
246  * into the socket code.
247  */
248 static void
249 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
250     struct sockaddr_in *udp_in)
251 {
252 	struct sockaddr *append_sa;
253 	struct socket *so;
254 	struct mbuf *opts = 0;
255 #ifdef INET6
256 	struct sockaddr_in6 udp_in6;
257 #endif
258 	struct udpcb *up;
259 
260 	INP_LOCK_ASSERT(inp);
261 
262 	/*
263 	 * Engage the tunneling protocol.
264 	 */
265 	up = intoudpcb(inp);
266 	if (up->u_tun_func != NULL) {
267 		(*up->u_tun_func)(n, off, inp);
268 		return;
269 	}
270 
271 	if (n == NULL)
272 		return;
273 
274 	off += sizeof(struct udphdr);
275 
276 #ifdef IPSEC
277 	/* Check AH/ESP integrity. */
278 	if (ipsec4_in_reject(n, inp)) {
279 		m_freem(n);
280 		V_ipsec4stat.in_polvio++;
281 		return;
282 	}
283 #ifdef IPSEC_NAT_T
284 	up = intoudpcb(inp);
285 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
286 	if (up->u_flags & UF_ESPINUDP_ALL) {	/* IPSec UDP encaps. */
287 		n = udp4_espdecap(inp, n, off);
288 		if (n == NULL)				/* Consumed. */
289 			return;
290 	}
291 #endif /* IPSEC_NAT_T */
292 #endif /* IPSEC */
293 #ifdef MAC
294 	if (mac_inpcb_check_deliver(inp, n) != 0) {
295 		m_freem(n);
296 		return;
297 	}
298 #endif /* MAC */
299 	if (inp->inp_flags & INP_CONTROLOPTS ||
300 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
301 #ifdef INET6
302 		if (inp->inp_vflag & INP_IPV6)
303 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
304 		else
305 #endif /* INET6 */
306 			ip_savecontrol(inp, &opts, ip, n);
307 	}
308 #ifdef INET6
309 	if (inp->inp_vflag & INP_IPV6) {
310 		bzero(&udp_in6, sizeof(udp_in6));
311 		udp_in6.sin6_len = sizeof(udp_in6);
312 		udp_in6.sin6_family = AF_INET6;
313 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
314 		append_sa = (struct sockaddr *)&udp_in6;
315 	} else
316 #endif /* INET6 */
317 		append_sa = (struct sockaddr *)udp_in;
318 	m_adj(n, off);
319 
320 	so = inp->inp_socket;
321 	SOCKBUF_LOCK(&so->so_rcv);
322 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
323 		SOCKBUF_UNLOCK(&so->so_rcv);
324 		m_freem(n);
325 		if (opts)
326 			m_freem(opts);
327 		UDPSTAT_INC(udps_fullsock);
328 	} else
329 		sorwakeup_locked(so);
330 }
331 
332 void
333 udp_input(struct mbuf *m, int off)
334 {
335 	int iphlen = off;
336 	struct ip *ip;
337 	struct udphdr *uh;
338 	struct ifnet *ifp;
339 	struct inpcb *inp;
340 	int len;
341 	struct ip save_ip;
342 	struct sockaddr_in udp_in;
343 #ifdef IPFIREWALL_FORWARD
344 	struct m_tag *fwd_tag;
345 #endif
346 
347 	ifp = m->m_pkthdr.rcvif;
348 	UDPSTAT_INC(udps_ipackets);
349 
350 	/*
351 	 * Strip IP options, if any; should skip this, make available to
352 	 * user, and use on returned packets, but we don't yet have a way to
353 	 * check the checksum with options still present.
354 	 */
355 	if (iphlen > sizeof (struct ip)) {
356 		ip_stripoptions(m, (struct mbuf *)0);
357 		iphlen = sizeof(struct ip);
358 	}
359 
360 	/*
361 	 * Get IP and UDP header together in first mbuf.
362 	 */
363 	ip = mtod(m, struct ip *);
364 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
365 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
366 			UDPSTAT_INC(udps_hdrops);
367 			return;
368 		}
369 		ip = mtod(m, struct ip *);
370 	}
371 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
372 
373 	/*
374 	 * Destination port of 0 is illegal, based on RFC768.
375 	 */
376 	if (uh->uh_dport == 0)
377 		goto badunlocked;
378 
379 	/*
380 	 * Construct sockaddr format source address.  Stuff source address
381 	 * and datagram in user buffer.
382 	 */
383 	bzero(&udp_in, sizeof(udp_in));
384 	udp_in.sin_len = sizeof(udp_in);
385 	udp_in.sin_family = AF_INET;
386 	udp_in.sin_port = uh->uh_sport;
387 	udp_in.sin_addr = ip->ip_src;
388 
389 	/*
390 	 * Make mbuf data length reflect UDP length.  If not enough data to
391 	 * reflect UDP length, drop.
392 	 */
393 	len = ntohs((u_short)uh->uh_ulen);
394 	if (ip->ip_len != len) {
395 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
396 			UDPSTAT_INC(udps_badlen);
397 			goto badunlocked;
398 		}
399 		m_adj(m, len - ip->ip_len);
400 		/* ip->ip_len = len; */
401 	}
402 
403 	/*
404 	 * Save a copy of the IP header in case we want restore it for
405 	 * sending an ICMP error message in response.
406 	 */
407 	if (!V_udp_blackhole)
408 		save_ip = *ip;
409 	else
410 		memset(&save_ip, 0, sizeof(save_ip));
411 
412 	/*
413 	 * Checksum extended UDP header and data.
414 	 */
415 	if (uh->uh_sum) {
416 		u_short uh_sum;
417 
418 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
419 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
420 				uh_sum = m->m_pkthdr.csum_data;
421 			else
422 				uh_sum = in_pseudo(ip->ip_src.s_addr,
423 				    ip->ip_dst.s_addr, htonl((u_short)len +
424 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
425 			uh_sum ^= 0xffff;
426 		} else {
427 			char b[9];
428 
429 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
430 			bzero(((struct ipovly *)ip)->ih_x1, 9);
431 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
432 			uh_sum = in_cksum(m, len + sizeof (struct ip));
433 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
434 		}
435 		if (uh_sum) {
436 			UDPSTAT_INC(udps_badsum);
437 			m_freem(m);
438 			return;
439 		}
440 	} else
441 		UDPSTAT_INC(udps_nosum);
442 
443 #ifdef IPFIREWALL_FORWARD
444 	/*
445 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
446 	 */
447 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
448 	if (fwd_tag != NULL) {
449 		struct sockaddr_in *next_hop;
450 
451 		/*
452 		 * Do the hack.
453 		 */
454 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
455 		ip->ip_dst = next_hop->sin_addr;
456 		uh->uh_dport = ntohs(next_hop->sin_port);
457 
458 		/*
459 		 * Remove the tag from the packet.  We don't need it anymore.
460 		 */
461 		m_tag_delete(m, fwd_tag);
462 	}
463 #endif
464 
465 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
466 	    in_broadcast(ip->ip_dst, ifp)) {
467 		struct inpcb *last;
468 		struct ip_moptions *imo;
469 
470 		INP_INFO_RLOCK(&V_udbinfo);
471 		last = NULL;
472 		LIST_FOREACH(inp, &V_udb, inp_list) {
473 			if (inp->inp_lport != uh->uh_dport)
474 				continue;
475 #ifdef INET6
476 			if ((inp->inp_vflag & INP_IPV4) == 0)
477 				continue;
478 #endif
479 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
480 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
481 				continue;
482 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
483 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
484 				continue;
485 			if (inp->inp_fport != 0 &&
486 			    inp->inp_fport != uh->uh_sport)
487 				continue;
488 
489 			INP_RLOCK(inp);
490 
491 			/*
492 			 * XXXRW: Because we weren't holding either the inpcb
493 			 * or the hash lock when we checked for a match
494 			 * before, we should probably recheck now that the
495 			 * inpcb lock is held.
496 			 */
497 
498 			/*
499 			 * Handle socket delivery policy for any-source
500 			 * and source-specific multicast. [RFC3678]
501 			 */
502 			imo = inp->inp_moptions;
503 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
504 				struct sockaddr_in	 group;
505 				int			 blocked;
506 				if (imo == NULL) {
507 					INP_RUNLOCK(inp);
508 					continue;
509 				}
510 				bzero(&group, sizeof(struct sockaddr_in));
511 				group.sin_len = sizeof(struct sockaddr_in);
512 				group.sin_family = AF_INET;
513 				group.sin_addr = ip->ip_dst;
514 
515 				blocked = imo_multi_filter(imo, ifp,
516 					(struct sockaddr *)&group,
517 					(struct sockaddr *)&udp_in);
518 				if (blocked != MCAST_PASS) {
519 					if (blocked == MCAST_NOTGMEMBER)
520 						IPSTAT_INC(ips_notmember);
521 					if (blocked == MCAST_NOTSMEMBER ||
522 					    blocked == MCAST_MUTED)
523 						UDPSTAT_INC(udps_filtermcast);
524 					INP_RUNLOCK(inp);
525 					continue;
526 				}
527 			}
528 			if (last != NULL) {
529 				struct mbuf *n;
530 
531 				n = m_copy(m, 0, M_COPYALL);
532 				udp_append(last, ip, n, iphlen, &udp_in);
533 				INP_RUNLOCK(last);
534 			}
535 			last = inp;
536 			/*
537 			 * Don't look for additional matches if this one does
538 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
539 			 * socket options set.  This heuristic avoids
540 			 * searching through all pcbs in the common case of a
541 			 * non-shared port.  It assumes that an application
542 			 * will never clear these options after setting them.
543 			 */
544 			if ((last->inp_socket->so_options &
545 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
546 				break;
547 		}
548 
549 		if (last == NULL) {
550 			/*
551 			 * No matching pcb found; discard datagram.  (No need
552 			 * to send an ICMP Port Unreachable for a broadcast
553 			 * or multicast datgram.)
554 			 */
555 			UDPSTAT_INC(udps_noportbcast);
556 			if (inp)
557 				INP_RUNLOCK(inp);
558 			INP_INFO_RUNLOCK(&V_udbinfo);
559 			goto badunlocked;
560 		}
561 		udp_append(last, ip, m, iphlen, &udp_in);
562 		INP_RUNLOCK(last);
563 		INP_INFO_RUNLOCK(&V_udbinfo);
564 		return;
565 	}
566 
567 	/*
568 	 * Locate pcb for datagram.
569 	 */
570 	inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
571 	    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
572 	    ifp, m);
573 	if (inp == NULL) {
574 		if (udp_log_in_vain) {
575 			char buf[4*sizeof "123"];
576 
577 			strcpy(buf, inet_ntoa(ip->ip_dst));
578 			log(LOG_INFO,
579 			    "Connection attempt to UDP %s:%d from %s:%d\n",
580 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
581 			    ntohs(uh->uh_sport));
582 		}
583 		UDPSTAT_INC(udps_noport);
584 		if (m->m_flags & (M_BCAST | M_MCAST)) {
585 			UDPSTAT_INC(udps_noportbcast);
586 			goto badunlocked;
587 		}
588 		if (V_udp_blackhole)
589 			goto badunlocked;
590 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
591 			goto badunlocked;
592 		*ip = save_ip;
593 		ip->ip_len += iphlen;
594 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
595 		return;
596 	}
597 
598 	/*
599 	 * Check the minimum TTL for socket.
600 	 */
601 	INP_RLOCK_ASSERT(inp);
602 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
603 		INP_RUNLOCK(inp);
604 		m_freem(m);
605 		return;
606 	}
607 	udp_append(inp, ip, m, iphlen, &udp_in);
608 	INP_RUNLOCK(inp);
609 	return;
610 
611 badunlocked:
612 	m_freem(m);
613 }
614 #endif /* INET */
615 
616 /*
617  * Notify a udp user of an asynchronous error; just wake up so that they can
618  * collect error status.
619  */
620 struct inpcb *
621 udp_notify(struct inpcb *inp, int errno)
622 {
623 
624 	/*
625 	 * While udp_ctlinput() always calls udp_notify() with a read lock
626 	 * when invoking it directly, in_pcbnotifyall() currently uses write
627 	 * locks due to sharing code with TCP.  For now, accept either a read
628 	 * or a write lock, but a read lock is sufficient.
629 	 */
630 	INP_LOCK_ASSERT(inp);
631 
632 	inp->inp_socket->so_error = errno;
633 	sorwakeup(inp->inp_socket);
634 	sowwakeup(inp->inp_socket);
635 	return (inp);
636 }
637 
638 #ifdef INET
639 void
640 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
641 {
642 	struct ip *ip = vip;
643 	struct udphdr *uh;
644 	struct in_addr faddr;
645 	struct inpcb *inp;
646 
647 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
648 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
649 		return;
650 
651 	/*
652 	 * Redirects don't need to be handled up here.
653 	 */
654 	if (PRC_IS_REDIRECT(cmd))
655 		return;
656 
657 	/*
658 	 * Hostdead is ugly because it goes linearly through all PCBs.
659 	 *
660 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
661 	 * DoS attack on machines with many connections.
662 	 */
663 	if (cmd == PRC_HOSTDEAD)
664 		ip = NULL;
665 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
666 		return;
667 	if (ip != NULL) {
668 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
669 		inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
670 		    ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
671 		if (inp != NULL) {
672 			INP_RLOCK_ASSERT(inp);
673 			if (inp->inp_socket != NULL) {
674 				udp_notify(inp, inetctlerrmap[cmd]);
675 			}
676 			INP_RUNLOCK(inp);
677 		}
678 	} else
679 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
680 		    udp_notify);
681 }
682 #endif /* INET */
683 
684 static int
685 udp_pcblist(SYSCTL_HANDLER_ARGS)
686 {
687 	int error, i, n;
688 	struct inpcb *inp, **inp_list;
689 	inp_gen_t gencnt;
690 	struct xinpgen xig;
691 
692 	/*
693 	 * The process of preparing the PCB list is too time-consuming and
694 	 * resource-intensive to repeat twice on every request.
695 	 */
696 	if (req->oldptr == 0) {
697 		n = V_udbinfo.ipi_count;
698 		n += imax(n / 8, 10);
699 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
700 		return (0);
701 	}
702 
703 	if (req->newptr != 0)
704 		return (EPERM);
705 
706 	/*
707 	 * OK, now we're committed to doing something.
708 	 */
709 	INP_INFO_RLOCK(&V_udbinfo);
710 	gencnt = V_udbinfo.ipi_gencnt;
711 	n = V_udbinfo.ipi_count;
712 	INP_INFO_RUNLOCK(&V_udbinfo);
713 
714 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
715 		+ n * sizeof(struct xinpcb));
716 	if (error != 0)
717 		return (error);
718 
719 	xig.xig_len = sizeof xig;
720 	xig.xig_count = n;
721 	xig.xig_gen = gencnt;
722 	xig.xig_sogen = so_gencnt;
723 	error = SYSCTL_OUT(req, &xig, sizeof xig);
724 	if (error)
725 		return (error);
726 
727 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
728 	if (inp_list == 0)
729 		return (ENOMEM);
730 
731 	INP_INFO_RLOCK(&V_udbinfo);
732 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
733 	     inp = LIST_NEXT(inp, inp_list)) {
734 		INP_WLOCK(inp);
735 		if (inp->inp_gencnt <= gencnt &&
736 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
737 			in_pcbref(inp);
738 			inp_list[i++] = inp;
739 		}
740 		INP_WUNLOCK(inp);
741 	}
742 	INP_INFO_RUNLOCK(&V_udbinfo);
743 	n = i;
744 
745 	error = 0;
746 	for (i = 0; i < n; i++) {
747 		inp = inp_list[i];
748 		INP_RLOCK(inp);
749 		if (inp->inp_gencnt <= gencnt) {
750 			struct xinpcb xi;
751 
752 			bzero(&xi, sizeof(xi));
753 			xi.xi_len = sizeof xi;
754 			/* XXX should avoid extra copy */
755 			bcopy(inp, &xi.xi_inp, sizeof *inp);
756 			if (inp->inp_socket)
757 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
758 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
759 			INP_RUNLOCK(inp);
760 			error = SYSCTL_OUT(req, &xi, sizeof xi);
761 		} else
762 			INP_RUNLOCK(inp);
763 	}
764 	INP_INFO_WLOCK(&V_udbinfo);
765 	for (i = 0; i < n; i++) {
766 		inp = inp_list[i];
767 		INP_RLOCK(inp);
768 		if (!in_pcbrele_rlocked(inp))
769 			INP_RUNLOCK(inp);
770 	}
771 	INP_INFO_WUNLOCK(&V_udbinfo);
772 
773 	if (!error) {
774 		/*
775 		 * Give the user an updated idea of our state.  If the
776 		 * generation differs from what we told her before, she knows
777 		 * that something happened while we were processing this
778 		 * request, and it might be necessary to retry.
779 		 */
780 		INP_INFO_RLOCK(&V_udbinfo);
781 		xig.xig_gen = V_udbinfo.ipi_gencnt;
782 		xig.xig_sogen = so_gencnt;
783 		xig.xig_count = V_udbinfo.ipi_count;
784 		INP_INFO_RUNLOCK(&V_udbinfo);
785 		error = SYSCTL_OUT(req, &xig, sizeof xig);
786 	}
787 	free(inp_list, M_TEMP);
788 	return (error);
789 }
790 
791 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
792     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
793     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
794 
795 #ifdef INET
796 static int
797 udp_getcred(SYSCTL_HANDLER_ARGS)
798 {
799 	struct xucred xuc;
800 	struct sockaddr_in addrs[2];
801 	struct inpcb *inp;
802 	int error;
803 
804 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
805 	if (error)
806 		return (error);
807 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
808 	if (error)
809 		return (error);
810 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
811 	    addrs[0].sin_addr, addrs[0].sin_port,
812 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
813 	if (inp != NULL) {
814 		INP_RLOCK_ASSERT(inp);
815 		if (inp->inp_socket == NULL)
816 			error = ENOENT;
817 		if (error == 0)
818 			error = cr_canseeinpcb(req->td->td_ucred, inp);
819 		if (error == 0)
820 			cru2x(inp->inp_cred, &xuc);
821 		INP_RUNLOCK(inp);
822 	} else
823 		error = ENOENT;
824 	if (error == 0)
825 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
826 	return (error);
827 }
828 
829 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
830     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
831     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
832 #endif /* INET */
833 
834 int
835 udp_ctloutput(struct socket *so, struct sockopt *sopt)
836 {
837 	int error = 0, optval;
838 	struct inpcb *inp;
839 #ifdef IPSEC_NAT_T
840 	struct udpcb *up;
841 #endif
842 
843 	inp = sotoinpcb(so);
844 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
845 	INP_WLOCK(inp);
846 	if (sopt->sopt_level != IPPROTO_UDP) {
847 #ifdef INET6
848 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
849 			INP_WUNLOCK(inp);
850 			error = ip6_ctloutput(so, sopt);
851 		}
852 #endif
853 #if defined(INET) && defined(INET6)
854 		else
855 #endif
856 #ifdef INET
857 		{
858 			INP_WUNLOCK(inp);
859 			error = ip_ctloutput(so, sopt);
860 		}
861 #endif
862 		return (error);
863 	}
864 
865 	switch (sopt->sopt_dir) {
866 	case SOPT_SET:
867 		switch (sopt->sopt_name) {
868 		case UDP_ENCAP:
869 			INP_WUNLOCK(inp);
870 			error = sooptcopyin(sopt, &optval, sizeof optval,
871 					    sizeof optval);
872 			if (error)
873 				break;
874 			inp = sotoinpcb(so);
875 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
876 			INP_WLOCK(inp);
877 #ifdef IPSEC_NAT_T
878 			up = intoudpcb(inp);
879 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
880 #endif
881 			switch (optval) {
882 			case 0:
883 				/* Clear all UDP encap. */
884 #ifdef IPSEC_NAT_T
885 				up->u_flags &= ~UF_ESPINUDP_ALL;
886 #endif
887 				break;
888 #ifdef IPSEC_NAT_T
889 			case UDP_ENCAP_ESPINUDP:
890 			case UDP_ENCAP_ESPINUDP_NON_IKE:
891 				up->u_flags &= ~UF_ESPINUDP_ALL;
892 				if (optval == UDP_ENCAP_ESPINUDP)
893 					up->u_flags |= UF_ESPINUDP;
894 				else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
895 					up->u_flags |= UF_ESPINUDP_NON_IKE;
896 				break;
897 #endif
898 			default:
899 				error = EINVAL;
900 				break;
901 			}
902 			INP_WUNLOCK(inp);
903 			break;
904 		default:
905 			INP_WUNLOCK(inp);
906 			error = ENOPROTOOPT;
907 			break;
908 		}
909 		break;
910 	case SOPT_GET:
911 		switch (sopt->sopt_name) {
912 #ifdef IPSEC_NAT_T
913 		case UDP_ENCAP:
914 			up = intoudpcb(inp);
915 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
916 			optval = up->u_flags & UF_ESPINUDP_ALL;
917 			INP_WUNLOCK(inp);
918 			error = sooptcopyout(sopt, &optval, sizeof optval);
919 			break;
920 #endif
921 		default:
922 			INP_WUNLOCK(inp);
923 			error = ENOPROTOOPT;
924 			break;
925 		}
926 		break;
927 	}
928 	return (error);
929 }
930 
931 #ifdef INET
932 #define	UH_WLOCKED	2
933 #define	UH_RLOCKED	1
934 #define	UH_UNLOCKED	0
935 static int
936 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
937     struct mbuf *control, struct thread *td)
938 {
939 	struct udpiphdr *ui;
940 	int len = m->m_pkthdr.len;
941 	struct in_addr faddr, laddr;
942 	struct cmsghdr *cm;
943 	struct sockaddr_in *sin, src;
944 	int error = 0;
945 	int ipflags;
946 	u_short fport, lport;
947 	int unlock_udbinfo;
948 
949 	/*
950 	 * udp_output() may need to temporarily bind or connect the current
951 	 * inpcb.  As such, we don't know up front whether we will need the
952 	 * pcbinfo lock or not.  Do any work to decide what is needed up
953 	 * front before acquiring any locks.
954 	 */
955 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
956 		if (control)
957 			m_freem(control);
958 		m_freem(m);
959 		return (EMSGSIZE);
960 	}
961 
962 	src.sin_family = 0;
963 	if (control != NULL) {
964 		/*
965 		 * XXX: Currently, we assume all the optional information is
966 		 * stored in a single mbuf.
967 		 */
968 		if (control->m_next) {
969 			m_freem(control);
970 			m_freem(m);
971 			return (EINVAL);
972 		}
973 		for (; control->m_len > 0;
974 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
975 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
976 			cm = mtod(control, struct cmsghdr *);
977 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
978 			    || cm->cmsg_len > control->m_len) {
979 				error = EINVAL;
980 				break;
981 			}
982 			if (cm->cmsg_level != IPPROTO_IP)
983 				continue;
984 
985 			switch (cm->cmsg_type) {
986 			case IP_SENDSRCADDR:
987 				if (cm->cmsg_len !=
988 				    CMSG_LEN(sizeof(struct in_addr))) {
989 					error = EINVAL;
990 					break;
991 				}
992 				bzero(&src, sizeof(src));
993 				src.sin_family = AF_INET;
994 				src.sin_len = sizeof(src);
995 				src.sin_port = inp->inp_lport;
996 				src.sin_addr =
997 				    *(struct in_addr *)CMSG_DATA(cm);
998 				break;
999 
1000 			default:
1001 				error = ENOPROTOOPT;
1002 				break;
1003 			}
1004 			if (error)
1005 				break;
1006 		}
1007 		m_freem(control);
1008 	}
1009 	if (error) {
1010 		m_freem(m);
1011 		return (error);
1012 	}
1013 
1014 	/*
1015 	 * Depending on whether or not the application has bound or connected
1016 	 * the socket, we may have to do varying levels of work.  The optimal
1017 	 * case is for a connected UDP socket, as a global lock isn't
1018 	 * required at all.
1019 	 *
1020 	 * In order to decide which we need, we require stability of the
1021 	 * inpcb binding, which we ensure by acquiring a read lock on the
1022 	 * inpcb.  This doesn't strictly follow the lock order, so we play
1023 	 * the trylock and retry game; note that we may end up with more
1024 	 * conservative locks than required the second time around, so later
1025 	 * assertions have to accept that.  Further analysis of the number of
1026 	 * misses under contention is required.
1027 	 *
1028 	 * XXXRW: Check that hash locking update here is correct.
1029 	 */
1030 	sin = (struct sockaddr_in *)addr;
1031 	INP_RLOCK(inp);
1032 	if (sin != NULL &&
1033 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1034 		INP_RUNLOCK(inp);
1035 		INP_WLOCK(inp);
1036 		INP_HASH_WLOCK(&V_udbinfo);
1037 		unlock_udbinfo = UH_WLOCKED;
1038 	} else if ((sin != NULL && (
1039 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
1040 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1041 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
1042 	    (inp->inp_lport == 0))) ||
1043 	    (src.sin_family == AF_INET)) {
1044 		INP_HASH_RLOCK(&V_udbinfo);
1045 		unlock_udbinfo = UH_RLOCKED;
1046 	} else
1047 		unlock_udbinfo = UH_UNLOCKED;
1048 
1049 	/*
1050 	 * If the IP_SENDSRCADDR control message was specified, override the
1051 	 * source address for this datagram.  Its use is invalidated if the
1052 	 * address thus specified is incomplete or clobbers other inpcbs.
1053 	 */
1054 	laddr = inp->inp_laddr;
1055 	lport = inp->inp_lport;
1056 	if (src.sin_family == AF_INET) {
1057 		INP_HASH_LOCK_ASSERT(&V_udbinfo);
1058 		if ((lport == 0) ||
1059 		    (laddr.s_addr == INADDR_ANY &&
1060 		     src.sin_addr.s_addr == INADDR_ANY)) {
1061 			error = EINVAL;
1062 			goto release;
1063 		}
1064 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1065 		    &laddr.s_addr, &lport, td->td_ucred);
1066 		if (error)
1067 			goto release;
1068 	}
1069 
1070 	/*
1071 	 * If a UDP socket has been connected, then a local address/port will
1072 	 * have been selected and bound.
1073 	 *
1074 	 * If a UDP socket has not been connected to, then an explicit
1075 	 * destination address must be used, in which case a local
1076 	 * address/port may not have been selected and bound.
1077 	 */
1078 	if (sin != NULL) {
1079 		INP_LOCK_ASSERT(inp);
1080 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1081 			error = EISCONN;
1082 			goto release;
1083 		}
1084 
1085 		/*
1086 		 * Jail may rewrite the destination address, so let it do
1087 		 * that before we use it.
1088 		 */
1089 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1090 		if (error)
1091 			goto release;
1092 
1093 		/*
1094 		 * If a local address or port hasn't yet been selected, or if
1095 		 * the destination address needs to be rewritten due to using
1096 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1097 		 * to do the heavy lifting.  Once a port is selected, we
1098 		 * commit the binding back to the socket; we also commit the
1099 		 * binding of the address if in jail.
1100 		 *
1101 		 * If we already have a valid binding and we're not
1102 		 * requesting a destination address rewrite, use a fast path.
1103 		 */
1104 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1105 		    inp->inp_lport == 0 ||
1106 		    sin->sin_addr.s_addr == INADDR_ANY ||
1107 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1108 			INP_HASH_LOCK_ASSERT(&V_udbinfo);
1109 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1110 			    &lport, &faddr.s_addr, &fport, NULL,
1111 			    td->td_ucred);
1112 			if (error)
1113 				goto release;
1114 
1115 			/*
1116 			 * XXXRW: Why not commit the port if the address is
1117 			 * !INADDR_ANY?
1118 			 */
1119 			/* Commit the local port if newly assigned. */
1120 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1121 			    inp->inp_lport == 0) {
1122 				INP_WLOCK_ASSERT(inp);
1123 				INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1124 				/*
1125 				 * Remember addr if jailed, to prevent
1126 				 * rebinding.
1127 				 */
1128 				if (prison_flag(td->td_ucred, PR_IP4))
1129 					inp->inp_laddr = laddr;
1130 				inp->inp_lport = lport;
1131 				if (in_pcbinshash(inp) != 0) {
1132 					inp->inp_lport = 0;
1133 					error = EAGAIN;
1134 					goto release;
1135 				}
1136 				inp->inp_flags |= INP_ANONPORT;
1137 			}
1138 		} else {
1139 			faddr = sin->sin_addr;
1140 			fport = sin->sin_port;
1141 		}
1142 	} else {
1143 		INP_LOCK_ASSERT(inp);
1144 		faddr = inp->inp_faddr;
1145 		fport = inp->inp_fport;
1146 		if (faddr.s_addr == INADDR_ANY) {
1147 			error = ENOTCONN;
1148 			goto release;
1149 		}
1150 	}
1151 
1152 	/*
1153 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1154 	 * link-layer headers.  Immediate slide the data pointer back forward
1155 	 * since we won't use that space at this layer.
1156 	 */
1157 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1158 	if (m == NULL) {
1159 		error = ENOBUFS;
1160 		goto release;
1161 	}
1162 	m->m_data += max_linkhdr;
1163 	m->m_len -= max_linkhdr;
1164 	m->m_pkthdr.len -= max_linkhdr;
1165 
1166 	/*
1167 	 * Fill in mbuf with extended UDP header and addresses and length put
1168 	 * into network format.
1169 	 */
1170 	ui = mtod(m, struct udpiphdr *);
1171 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1172 	ui->ui_pr = IPPROTO_UDP;
1173 	ui->ui_src = laddr;
1174 	ui->ui_dst = faddr;
1175 	ui->ui_sport = lport;
1176 	ui->ui_dport = fport;
1177 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1178 
1179 	/*
1180 	 * Set the Don't Fragment bit in the IP header.
1181 	 */
1182 	if (inp->inp_flags & INP_DONTFRAG) {
1183 		struct ip *ip;
1184 
1185 		ip = (struct ip *)&ui->ui_i;
1186 		ip->ip_off |= IP_DF;
1187 	}
1188 
1189 	ipflags = 0;
1190 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1191 		ipflags |= IP_ROUTETOIF;
1192 	if (inp->inp_socket->so_options & SO_BROADCAST)
1193 		ipflags |= IP_ALLOWBROADCAST;
1194 	if (inp->inp_flags & INP_ONESBCAST)
1195 		ipflags |= IP_SENDONES;
1196 
1197 #ifdef MAC
1198 	mac_inpcb_create_mbuf(inp, m);
1199 #endif
1200 
1201 	/*
1202 	 * Set up checksum and output datagram.
1203 	 */
1204 	if (udp_cksum) {
1205 		if (inp->inp_flags & INP_ONESBCAST)
1206 			faddr.s_addr = INADDR_BROADCAST;
1207 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1208 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1209 		m->m_pkthdr.csum_flags = CSUM_UDP;
1210 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1211 	} else
1212 		ui->ui_sum = 0;
1213 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1214 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1215 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
1216 	UDPSTAT_INC(udps_opackets);
1217 
1218 	if (unlock_udbinfo == UH_WLOCKED)
1219 		INP_HASH_WUNLOCK(&V_udbinfo);
1220 	else if (unlock_udbinfo == UH_RLOCKED)
1221 		INP_HASH_RUNLOCK(&V_udbinfo);
1222 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1223 	    inp->inp_moptions, inp);
1224 	if (unlock_udbinfo == UH_WLOCKED)
1225 		INP_WUNLOCK(inp);
1226 	else
1227 		INP_RUNLOCK(inp);
1228 	return (error);
1229 
1230 release:
1231 	if (unlock_udbinfo == UH_WLOCKED) {
1232 		INP_HASH_WUNLOCK(&V_udbinfo);
1233 		INP_WUNLOCK(inp);
1234 	} else if (unlock_udbinfo == UH_RLOCKED) {
1235 		INP_HASH_RUNLOCK(&V_udbinfo);
1236 		INP_RUNLOCK(inp);
1237 	} else
1238 		INP_RUNLOCK(inp);
1239 	m_freem(m);
1240 	return (error);
1241 }
1242 
1243 
1244 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1245 /*
1246  * Potentially decap ESP in UDP frame.  Check for an ESP header
1247  * and optional marker; if present, strip the UDP header and
1248  * push the result through IPSec.
1249  *
1250  * Returns mbuf to be processed (potentially re-allocated) or
1251  * NULL if consumed and/or processed.
1252  */
1253 static struct mbuf *
1254 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1255 {
1256 	size_t minlen, payload, skip, iphlen;
1257 	caddr_t data;
1258 	struct udpcb *up;
1259 	struct m_tag *tag;
1260 	struct udphdr *udphdr;
1261 	struct ip *ip;
1262 
1263 	INP_RLOCK_ASSERT(inp);
1264 
1265 	/*
1266 	 * Pull up data so the longest case is contiguous:
1267 	 *    IP/UDP hdr + non ESP marker + ESP hdr.
1268 	 */
1269 	minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1270 	if (minlen > m->m_pkthdr.len)
1271 		minlen = m->m_pkthdr.len;
1272 	if ((m = m_pullup(m, minlen)) == NULL) {
1273 		V_ipsec4stat.in_inval++;
1274 		return (NULL);		/* Bypass caller processing. */
1275 	}
1276 	data = mtod(m, caddr_t);	/* Points to ip header. */
1277 	payload = m->m_len - off;	/* Size of payload. */
1278 
1279 	if (payload == 1 && data[off] == '\xff')
1280 		return (m);		/* NB: keepalive packet, no decap. */
1281 
1282 	up = intoudpcb(inp);
1283 	KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1284 	KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1285 	    ("u_flags 0x%x", up->u_flags));
1286 
1287 	/*
1288 	 * Check that the payload is large enough to hold an
1289 	 * ESP header and compute the amount of data to remove.
1290 	 *
1291 	 * NB: the caller has already done a pullup for us.
1292 	 * XXX can we assume alignment and eliminate bcopys?
1293 	 */
1294 	if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1295 		/*
1296 		 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1297 		 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1298 		 * possible AH mode non-IKE marker+non-ESP marker
1299 		 * from draft-ietf-ipsec-udp-encaps-00.txt.
1300 		 */
1301 		uint64_t marker;
1302 
1303 		if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1304 			return (m);	/* NB: no decap. */
1305 		bcopy(data + off, &marker, sizeof(uint64_t));
1306 		if (marker != 0)	/* Non-IKE marker. */
1307 			return (m);	/* NB: no decap. */
1308 		skip = sizeof(uint64_t) + sizeof(struct udphdr);
1309 	} else {
1310 		uint32_t spi;
1311 
1312 		if (payload <= sizeof(struct esp)) {
1313 			V_ipsec4stat.in_inval++;
1314 			m_freem(m);
1315 			return (NULL);	/* Discard. */
1316 		}
1317 		bcopy(data + off, &spi, sizeof(uint32_t));
1318 		if (spi == 0)		/* Non-ESP marker. */
1319 			return (m);	/* NB: no decap. */
1320 		skip = sizeof(struct udphdr);
1321 	}
1322 
1323 	/*
1324 	 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1325 	 * the UDP ports. This is required if we want to select
1326 	 * the right SPD for multiple hosts behind same NAT.
1327 	 *
1328 	 * NB: ports are maintained in network byte order everywhere
1329 	 *     in the NAT-T code.
1330 	 */
1331 	tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1332 		2 * sizeof(uint16_t), M_NOWAIT);
1333 	if (tag == NULL) {
1334 		V_ipsec4stat.in_nomem++;
1335 		m_freem(m);
1336 		return (NULL);		/* Discard. */
1337 	}
1338 	iphlen = off - sizeof(struct udphdr);
1339 	udphdr = (struct udphdr *)(data + iphlen);
1340 	((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1341 	((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1342 	m_tag_prepend(m, tag);
1343 
1344 	/*
1345 	 * Remove the UDP header (and possibly the non ESP marker)
1346 	 * IP header length is iphlen
1347 	 * Before:
1348 	 *   <--- off --->
1349 	 *   +----+------+-----+
1350 	 *   | IP |  UDP | ESP |
1351 	 *   +----+------+-----+
1352 	 *        <-skip->
1353 	 * After:
1354 	 *          +----+-----+
1355 	 *          | IP | ESP |
1356 	 *          +----+-----+
1357 	 *   <-skip->
1358 	 */
1359 	ovbcopy(data, data + skip, iphlen);
1360 	m_adj(m, skip);
1361 
1362 	ip = mtod(m, struct ip *);
1363 	ip->ip_len -= skip;
1364 	ip->ip_p = IPPROTO_ESP;
1365 
1366 	/*
1367 	 * We cannot yet update the cksums so clear any
1368 	 * h/w cksum flags as they are no longer valid.
1369 	 */
1370 	if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1371 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1372 
1373 	(void) ipsec4_common_input(m, iphlen, ip->ip_p);
1374 	return (NULL);			/* NB: consumed, bypass processing. */
1375 }
1376 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1377 
1378 static void
1379 udp_abort(struct socket *so)
1380 {
1381 	struct inpcb *inp;
1382 
1383 	inp = sotoinpcb(so);
1384 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1385 	INP_WLOCK(inp);
1386 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1387 		INP_HASH_WLOCK(&V_udbinfo);
1388 		in_pcbdisconnect(inp);
1389 		inp->inp_laddr.s_addr = INADDR_ANY;
1390 		INP_HASH_WUNLOCK(&V_udbinfo);
1391 		soisdisconnected(so);
1392 	}
1393 	INP_WUNLOCK(inp);
1394 }
1395 
1396 static int
1397 udp_attach(struct socket *so, int proto, struct thread *td)
1398 {
1399 	struct inpcb *inp;
1400 	int error;
1401 
1402 	inp = sotoinpcb(so);
1403 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1404 	error = soreserve(so, udp_sendspace, udp_recvspace);
1405 	if (error)
1406 		return (error);
1407 	INP_INFO_WLOCK(&V_udbinfo);
1408 	error = in_pcballoc(so, &V_udbinfo);
1409 	if (error) {
1410 		INP_INFO_WUNLOCK(&V_udbinfo);
1411 		return (error);
1412 	}
1413 
1414 	inp = sotoinpcb(so);
1415 	inp->inp_vflag |= INP_IPV4;
1416 	inp->inp_ip_ttl = V_ip_defttl;
1417 
1418 	error = udp_newudpcb(inp);
1419 	if (error) {
1420 		in_pcbdetach(inp);
1421 		in_pcbfree(inp);
1422 		INP_INFO_WUNLOCK(&V_udbinfo);
1423 		return (error);
1424 	}
1425 
1426 	INP_WUNLOCK(inp);
1427 	INP_INFO_WUNLOCK(&V_udbinfo);
1428 	return (0);
1429 }
1430 #endif /* INET */
1431 
1432 int
1433 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1434 {
1435 	struct inpcb *inp;
1436 	struct udpcb *up;
1437 
1438 	KASSERT(so->so_type == SOCK_DGRAM,
1439 	    ("udp_set_kernel_tunneling: !dgram"));
1440 	inp = sotoinpcb(so);
1441 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1442 	INP_WLOCK(inp);
1443 	up = intoudpcb(inp);
1444 	if (up->u_tun_func != NULL) {
1445 		INP_WUNLOCK(inp);
1446 		return (EBUSY);
1447 	}
1448 	up->u_tun_func = f;
1449 	INP_WUNLOCK(inp);
1450 	return (0);
1451 }
1452 
1453 #ifdef INET
1454 static int
1455 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1456 {
1457 	struct inpcb *inp;
1458 	int error;
1459 
1460 	inp = sotoinpcb(so);
1461 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1462 	INP_WLOCK(inp);
1463 	INP_HASH_WLOCK(&V_udbinfo);
1464 	error = in_pcbbind(inp, nam, td->td_ucred);
1465 	INP_HASH_WUNLOCK(&V_udbinfo);
1466 	INP_WUNLOCK(inp);
1467 	return (error);
1468 }
1469 
1470 static void
1471 udp_close(struct socket *so)
1472 {
1473 	struct inpcb *inp;
1474 
1475 	inp = sotoinpcb(so);
1476 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1477 	INP_WLOCK(inp);
1478 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1479 		INP_HASH_WLOCK(&V_udbinfo);
1480 		in_pcbdisconnect(inp);
1481 		inp->inp_laddr.s_addr = INADDR_ANY;
1482 		INP_HASH_WUNLOCK(&V_udbinfo);
1483 		soisdisconnected(so);
1484 	}
1485 	INP_WUNLOCK(inp);
1486 }
1487 
1488 static int
1489 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1490 {
1491 	struct inpcb *inp;
1492 	int error;
1493 	struct sockaddr_in *sin;
1494 
1495 	inp = sotoinpcb(so);
1496 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1497 	INP_WLOCK(inp);
1498 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1499 		INP_WUNLOCK(inp);
1500 		return (EISCONN);
1501 	}
1502 	sin = (struct sockaddr_in *)nam;
1503 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1504 	if (error != 0) {
1505 		INP_WUNLOCK(inp);
1506 		return (error);
1507 	}
1508 	INP_HASH_WLOCK(&V_udbinfo);
1509 	error = in_pcbconnect(inp, nam, td->td_ucred);
1510 	INP_HASH_WUNLOCK(&V_udbinfo);
1511 	if (error == 0)
1512 		soisconnected(so);
1513 	INP_WUNLOCK(inp);
1514 	return (error);
1515 }
1516 
1517 static void
1518 udp_detach(struct socket *so)
1519 {
1520 	struct inpcb *inp;
1521 	struct udpcb *up;
1522 
1523 	inp = sotoinpcb(so);
1524 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1525 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1526 	    ("udp_detach: not disconnected"));
1527 	INP_INFO_WLOCK(&V_udbinfo);
1528 	INP_WLOCK(inp);
1529 	up = intoudpcb(inp);
1530 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1531 	inp->inp_ppcb = NULL;
1532 	in_pcbdetach(inp);
1533 	in_pcbfree(inp);
1534 	INP_INFO_WUNLOCK(&V_udbinfo);
1535 	udp_discardcb(up);
1536 }
1537 
1538 static int
1539 udp_disconnect(struct socket *so)
1540 {
1541 	struct inpcb *inp;
1542 
1543 	inp = sotoinpcb(so);
1544 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1545 	INP_WLOCK(inp);
1546 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1547 		INP_WUNLOCK(inp);
1548 		return (ENOTCONN);
1549 	}
1550 	INP_HASH_WLOCK(&V_udbinfo);
1551 	in_pcbdisconnect(inp);
1552 	inp->inp_laddr.s_addr = INADDR_ANY;
1553 	INP_HASH_WUNLOCK(&V_udbinfo);
1554 	SOCK_LOCK(so);
1555 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1556 	SOCK_UNLOCK(so);
1557 	INP_WUNLOCK(inp);
1558 	return (0);
1559 }
1560 
1561 static int
1562 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1563     struct mbuf *control, struct thread *td)
1564 {
1565 	struct inpcb *inp;
1566 
1567 	inp = sotoinpcb(so);
1568 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1569 	return (udp_output(inp, m, addr, control, td));
1570 }
1571 #endif /* INET */
1572 
1573 int
1574 udp_shutdown(struct socket *so)
1575 {
1576 	struct inpcb *inp;
1577 
1578 	inp = sotoinpcb(so);
1579 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1580 	INP_WLOCK(inp);
1581 	socantsendmore(so);
1582 	INP_WUNLOCK(inp);
1583 	return (0);
1584 }
1585 
1586 #ifdef INET
1587 struct pr_usrreqs udp_usrreqs = {
1588 	.pru_abort =		udp_abort,
1589 	.pru_attach =		udp_attach,
1590 	.pru_bind =		udp_bind,
1591 	.pru_connect =		udp_connect,
1592 	.pru_control =		in_control,
1593 	.pru_detach =		udp_detach,
1594 	.pru_disconnect =	udp_disconnect,
1595 	.pru_peeraddr =		in_getpeeraddr,
1596 	.pru_send =		udp_send,
1597 	.pru_soreceive =	soreceive_dgram,
1598 	.pru_sosend =		sosend_dgram,
1599 	.pru_shutdown =		udp_shutdown,
1600 	.pru_sockaddr =		in_getsockaddr,
1601 	.pru_sosetlabel =	in_pcbsosetlabel,
1602 	.pru_close =		udp_close,
1603 };
1604 #endif /* INET */
1605