xref: /dragonfly/sys/netinet/udp_usrreq.c (revision 82730a9c)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
63  * $FreeBSD: src/sys/netinet/udp_usrreq.c,v 1.64.2.18 2003/01/24 05:11:34 sam Exp $
64  */
65 
66 #include "opt_ipsec.h"
67 #include "opt_inet6.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.h>
72 #include <sys/malloc.h>
73 #include <sys/mbuf.h>
74 #include <sys/domain.h>
75 #include <sys/proc.h>
76 #include <sys/priv.h>
77 #include <sys/protosw.h>
78 #include <sys/socket.h>
79 #include <sys/socketvar.h>
80 #include <sys/sysctl.h>
81 #include <sys/syslog.h>
82 #include <sys/in_cksum.h>
83 
84 #include <sys/thread2.h>
85 #include <sys/socketvar2.h>
86 #include <sys/serialize.h>
87 
88 #include <machine/stdarg.h>
89 
90 #include <net/if.h>
91 #include <net/route.h>
92 #include <net/netmsg2.h>
93 #include <net/netisr2.h>
94 
95 #include <netinet/in.h>
96 #include <netinet/in_systm.h>
97 #include <netinet/ip.h>
98 #ifdef INET6
99 #include <netinet/ip6.h>
100 #endif
101 #include <netinet/in_pcb.h>
102 #include <netinet/in_var.h>
103 #include <netinet/ip_var.h>
104 #ifdef INET6
105 #include <netinet6/ip6_var.h>
106 #endif
107 #include <netinet/ip_icmp.h>
108 #include <netinet/icmp_var.h>
109 #include <netinet/udp.h>
110 #include <netinet/udp_var.h>
111 
112 #ifdef FAST_IPSEC
113 #include <netproto/ipsec/ipsec.h>
114 #endif
115 
116 #ifdef IPSEC
117 #include <netinet6/ipsec.h>
118 #endif
119 
120 /*
121  * UDP protocol implementation.
122  * Per RFC 768, August, 1980.
123  */
124 #ifndef	COMPAT_42
125 static int	udpcksum = 1;
126 #else
127 static int	udpcksum = 0;		/* XXX */
128 #endif
129 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
130     &udpcksum, 0, "Enable checksumming of UDP packets");
131 
132 int	log_in_vain = 0;
133 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
134     &log_in_vain, 0, "Log all incoming UDP packets");
135 
136 static int	blackhole = 0;
137 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
138 	&blackhole, 0, "Do not send port unreachables for refused connects");
139 
140 static int	strict_mcast_mship = 1;
141 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
142 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
143 
144 int	udp_sosend_async = 1;
145 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_async, CTLFLAG_RW,
146 	&udp_sosend_async, 0, "UDP asynchronized pru_send");
147 
148 int	udp_sosend_prepend = 1;
149 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_prepend, CTLFLAG_RW,
150 	&udp_sosend_prepend, 0,
151 	"Prepend enough space for proto and link header in pru_send");
152 
153 static int udp_reuseport_ext = 1;
154 SYSCTL_INT(_net_inet_udp, OID_AUTO, reuseport_ext, CTLFLAG_RW,
155 	&udp_reuseport_ext, 0, "SO_REUSEPORT extension");
156 
157 struct	inpcbinfo udbinfo;
158 
159 static struct netisr_barrier *udbinfo_br;
160 static struct lwkt_serialize udbinfo_slize = LWKT_SERIALIZE_INITIALIZER;
161 
162 #ifndef UDBHASHSIZE
163 #define UDBHASHSIZE 16
164 #endif
165 
166 struct	udpstat udpstat_percpu[MAXCPU] __cachealign;
167 
168 #ifdef INET6
169 struct udp_in6 {
170 	struct sockaddr_in6	uin6_sin;
171 	u_char			uin6_init_done : 1;
172 };
173 struct udp_ip6 {
174 	struct ip6_hdr		uip6_ip6;
175 	u_char			uip6_init_done : 1;
176 };
177 #else
178 struct udp_in6;
179 struct udp_ip6;
180 #endif /* INET6 */
181 
182 static void udp_append (struct inpcb *last, struct ip *ip,
183     struct mbuf *n, int off, struct sockaddr_in *udp_in,
184     struct udp_in6 *, struct udp_ip6 *);
185 #ifdef INET6
186 static void ip_2_ip6_hdr (struct ip6_hdr *ip6, struct ip *ip);
187 #endif
188 
189 static int udp_connect_oncpu(struct socket *so, struct thread *td,
190 			struct sockaddr_in *sin, struct sockaddr_in *if_sin);
191 static int udp_output (struct inpcb *, struct mbuf *, struct sockaddr *,
192 			struct thread *, int, boolean_t);
193 
194 void
195 udp_init(void)
196 {
197 	int cpu;
198 
199 	in_pcbinfo_init(&udbinfo);
200 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
201 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
202 					&udbinfo.porthashmask);
203 	udbinfo.wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB,
204 					    &udbinfo.wildcardhashmask);
205 	udbinfo.localgrphashbase = hashinit(UDBHASHSIZE, M_PCB,
206 					    &udbinfo.localgrphashmask);
207 	udbinfo.ipi_size = sizeof(struct inpcb);
208 
209 	udbinfo_br = netisr_barrier_create();
210 
211 	/*
212 	 * Initialize UDP statistics counters for each CPU.
213 	 */
214 	for (cpu = 0; cpu < ncpus; ++cpu)
215 		bzero(&udpstat_percpu[cpu], sizeof(struct udpstat));
216 }
217 
218 static int
219 sysctl_udpstat(SYSCTL_HANDLER_ARGS)
220 {
221 	int cpu, error = 0;
222 
223 	for (cpu = 0; cpu < ncpus; ++cpu) {
224 		if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu],
225 					sizeof(struct udpstat))))
226 			break;
227 		if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu],
228 				       sizeof(struct udpstat))))
229 			break;
230 	}
231 
232 	return (error);
233 }
234 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
235     0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics");
236 
237 /*
238  * Check multicast packets to make sure they are only sent to sockets with
239  * multicast memberships for the packet's destination address and arrival
240  * interface.  Multicast packets to multicast-unaware sockets are also
241  * disallowed.
242  *
243  * Returns 0 if the packet is acceptable, -1 if it is not.
244  */
245 static __inline int
246 check_multicast_membership(struct ip *ip, struct inpcb *inp, struct mbuf *m)
247 {
248 	int mshipno;
249 	struct ip_moptions *mopt;
250 
251 	if (strict_mcast_mship == 0 ||
252 	    !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
253 		return (0);
254 	}
255 	mopt = inp->inp_moptions;
256 	if (mopt == NULL)
257 		return (-1);
258 	for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) {
259 		struct in_multi *maddr = mopt->imo_membership[mshipno];
260 
261 		if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr &&
262 		    m->m_pkthdr.rcvif == maddr->inm_ifp) {
263 			return (0);
264 		}
265 	}
266 	return (-1);
267 }
268 
269 int
270 udp_input(struct mbuf **mp, int *offp, int proto)
271 {
272 	struct sockaddr_in udp_in = { sizeof udp_in, AF_INET };
273 #ifdef INET6
274 	struct udp_in6 udp_in6 = {
275 		{ sizeof udp_in6.uin6_sin, AF_INET6 }, 0
276 	};
277 	struct udp_ip6 udp_ip6;
278 #endif
279 
280 	int iphlen;
281 	struct ip *ip;
282 	struct udphdr *uh;
283 	struct inpcb *inp;
284 	struct mbuf *m;
285 	struct mbuf *opts = NULL;
286 	int len, off;
287 	struct ip save_ip;
288 	struct sockaddr *append_sa;
289 
290 	off = *offp;
291 	m = *mp;
292 	*mp = NULL;
293 
294 	iphlen = off;
295 	udp_stat.udps_ipackets++;
296 
297 	/*
298 	 * Strip IP options, if any; should skip this,
299 	 * make available to user, and use on returned packets,
300 	 * but we don't yet have a way to check the checksum
301 	 * with options still present.
302 	 */
303 	if (iphlen > sizeof(struct ip)) {
304 		ip_stripoptions(m);
305 		iphlen = sizeof(struct ip);
306 	}
307 
308 	/*
309 	 * IP and UDP headers are together in first mbuf.
310 	 * Already checked and pulled up in ip_demux().
311 	 */
312 	KASSERT(m->m_len >= iphlen + sizeof(struct udphdr),
313 	    ("UDP header not in one mbuf"));
314 
315 	ip = mtod(m, struct ip *);
316 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
317 
318 	/* destination port of 0 is illegal, based on RFC768. */
319 	if (uh->uh_dport == 0)
320 		goto bad;
321 
322 	/*
323 	 * Make mbuf data length reflect UDP length.
324 	 * If not enough data to reflect UDP length, drop.
325 	 */
326 	len = ntohs((u_short)uh->uh_ulen);
327 	if (ip->ip_len != len) {
328 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
329 			udp_stat.udps_badlen++;
330 			goto bad;
331 		}
332 		m_adj(m, len - ip->ip_len);
333 		/* ip->ip_len = len; */
334 	}
335 	/*
336 	 * Save a copy of the IP header in case we want restore it
337 	 * for sending an ICMP error message in response.
338 	 */
339 	save_ip = *ip;
340 
341 	/*
342 	 * Checksum extended UDP header and data.
343 	 */
344 	if (uh->uh_sum) {
345 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
346 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
347 				uh->uh_sum = m->m_pkthdr.csum_data;
348 			else
349 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
350 				    ip->ip_dst.s_addr, htonl((u_short)len +
351 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
352 			uh->uh_sum ^= 0xffff;
353 		} else {
354 			char b[9];
355 
356 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
357 			bzero(((struct ipovly *)ip)->ih_x1, 9);
358 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
359 			uh->uh_sum = in_cksum(m, len + sizeof(struct ip));
360 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
361 		}
362 		if (uh->uh_sum) {
363 			udp_stat.udps_badsum++;
364 			m_freem(m);
365 			return(IPPROTO_DONE);
366 		}
367 	} else
368 		udp_stat.udps_nosum++;
369 
370 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
371 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
372 		struct inpcb *last;
373 
374 		/*
375 		 * Deliver a multicast or broadcast datagram to *all* sockets
376 		 * for which the local and remote addresses and ports match
377 		 * those of the incoming datagram.  This allows more than
378 		 * one process to receive multi/broadcasts on the same port.
379 		 * (This really ought to be done for unicast datagrams as
380 		 * well, but that would cause problems with existing
381 		 * applications that open both address-specific sockets and
382 		 * a wildcard socket listening to the same port -- they would
383 		 * end up receiving duplicates of every unicast datagram.
384 		 * Those applications open the multiple sockets to overcome an
385 		 * inadequacy of the UDP socket interface, but for backwards
386 		 * compatibility we avoid the problem here rather than
387 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
388 		 */
389 
390 		/*
391 		 * Construct sockaddr format source address.
392 		 */
393 		udp_in.sin_port = uh->uh_sport;
394 		udp_in.sin_addr = ip->ip_src;
395 		/*
396 		 * Locate pcb(s) for datagram.
397 		 * (Algorithm copied from raw_intr().)
398 		 */
399 		last = NULL;
400 #ifdef INET6
401 		udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0;
402 #endif
403 		LIST_FOREACH(inp, &udbinfo.pcblisthead, inp_list) {
404 			KKASSERT((inp->inp_flags & INP_PLACEMARKER) == 0);
405 #ifdef INET6
406 			if (!(inp->inp_vflag & INP_IPV4))
407 				continue;
408 #endif
409 			if (inp->inp_lport != uh->uh_dport)
410 				continue;
411 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
412 				if (inp->inp_laddr.s_addr !=
413 				    ip->ip_dst.s_addr)
414 					continue;
415 			}
416 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
417 				if (inp->inp_faddr.s_addr !=
418 				    ip->ip_src.s_addr ||
419 				    inp->inp_fport != uh->uh_sport)
420 					continue;
421 			}
422 
423 			if (check_multicast_membership(ip, inp, m) < 0)
424 				continue;
425 
426 			if (last != NULL) {
427 				struct mbuf *n;
428 
429 #ifdef IPSEC
430 				/* check AH/ESP integrity. */
431 				if (ipsec4_in_reject_so(m, last->inp_socket))
432 					ipsecstat.in_polvio++;
433 					/* do not inject data to pcb */
434 				else
435 #endif /*IPSEC*/
436 #ifdef FAST_IPSEC
437 				/* check AH/ESP integrity. */
438 				if (ipsec4_in_reject(m, last))
439 					;
440 				else
441 #endif /*FAST_IPSEC*/
442 				if ((n = m_copypacket(m, MB_DONTWAIT)) != NULL)
443 					udp_append(last, ip, n,
444 					    iphlen + sizeof(struct udphdr),
445 					    &udp_in,
446 #ifdef INET6
447 					    &udp_in6, &udp_ip6
448 #else
449 				            NULL, NULL
450 #endif
451 					    );
452 			}
453 			last = inp;
454 			/*
455 			 * Don't look for additional matches if this one does
456 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
457 			 * socket options set.  This heuristic avoids searching
458 			 * through all pcbs in the common case of a non-shared
459 			 * port.  It * assumes that an application will never
460 			 * clear these options after setting them.
461 			 */
462 			if (!(last->inp_socket->so_options &
463 			    (SO_REUSEPORT | SO_REUSEADDR)))
464 				break;
465 		}
466 
467 		if (last == NULL) {
468 			/*
469 			 * No matching pcb found; discard datagram.
470 			 * (No need to send an ICMP Port Unreachable
471 			 * for a broadcast or multicast datgram.)
472 			 */
473 			udp_stat.udps_noportbcast++;
474 			goto bad;
475 		}
476 #ifdef IPSEC
477 		/* check AH/ESP integrity. */
478 		if (ipsec4_in_reject_so(m, last->inp_socket)) {
479 			ipsecstat.in_polvio++;
480 			goto bad;
481 		}
482 #endif /*IPSEC*/
483 #ifdef FAST_IPSEC
484 		/* check AH/ESP integrity. */
485 		if (ipsec4_in_reject(m, last))
486 			goto bad;
487 #endif /*FAST_IPSEC*/
488 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
489 		    &udp_in,
490 #ifdef INET6
491 		    &udp_in6, &udp_ip6
492 #else
493 		    NULL, NULL
494 #endif
495 		    );
496 		return(IPPROTO_DONE);
497 	}
498 	/*
499 	 * Locate pcb for datagram.
500 	 */
501 	inp = in_pcblookup_pkthash(&udbinfo, ip->ip_src, uh->uh_sport,
502 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif,
503 	    udp_reuseport_ext ? m : NULL);
504 	if (inp == NULL) {
505 		if (log_in_vain) {
506 			char buf[sizeof "aaa.bbb.ccc.ddd"];
507 
508 			strcpy(buf, inet_ntoa(ip->ip_dst));
509 			log(LOG_INFO,
510 			    "Connection attempt to UDP %s:%d from %s:%d\n",
511 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
512 			    ntohs(uh->uh_sport));
513 		}
514 		udp_stat.udps_noport++;
515 		if (m->m_flags & (M_BCAST | M_MCAST)) {
516 			udp_stat.udps_noportbcast++;
517 			goto bad;
518 		}
519 		if (blackhole)
520 			goto bad;
521 #ifdef ICMP_BANDLIM
522 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
523 			goto bad;
524 #endif
525 		*ip = save_ip;
526 		ip->ip_len += iphlen;
527 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
528 		return(IPPROTO_DONE);
529 	}
530 #ifdef IPSEC
531 	if (ipsec4_in_reject_so(m, inp->inp_socket)) {
532 		ipsecstat.in_polvio++;
533 		goto bad;
534 	}
535 #endif /*IPSEC*/
536 #ifdef FAST_IPSEC
537 	if (ipsec4_in_reject(m, inp))
538 		goto bad;
539 #endif /*FAST_IPSEC*/
540 	/*
541 	 * Check the minimum TTL for socket.
542 	 */
543 	if (ip->ip_ttl < inp->inp_ip_minttl)
544 		goto bad;
545 
546 	/*
547 	 * Construct sockaddr format source address.
548 	 * Stuff source address and datagram in user buffer.
549 	 */
550 	udp_in.sin_port = uh->uh_sport;
551 	udp_in.sin_addr = ip->ip_src;
552 	if ((inp->inp_flags & INP_CONTROLOPTS) ||
553 	    (inp->inp_socket->so_options & SO_TIMESTAMP)) {
554 #ifdef INET6
555 		if (inp->inp_vflag & INP_IPV6) {
556 			int savedflags;
557 
558 			ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
559 			savedflags = inp->inp_flags;
560 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
561 			ip6_savecontrol(inp, &opts, &udp_ip6.uip6_ip6, m);
562 			inp->inp_flags = savedflags;
563 		} else
564 #endif
565 		ip_savecontrol(inp, &opts, ip, m);
566 	}
567 	m_adj(m, iphlen + sizeof(struct udphdr));
568 #ifdef INET6
569 	if (inp->inp_vflag & INP_IPV6) {
570 		in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
571 		append_sa = (struct sockaddr *)&udp_in6;
572 	} else
573 #endif
574 		append_sa = (struct sockaddr *)&udp_in;
575 
576 	lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token);
577 	if (ssb_appendaddr(&inp->inp_socket->so_rcv, append_sa, m, opts) == 0) {
578 		udp_stat.udps_fullsock++;
579 		lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
580 		goto bad;
581 	}
582 	lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
583 	sorwakeup(inp->inp_socket);
584 	return(IPPROTO_DONE);
585 bad:
586 	m_freem(m);
587 	if (opts)
588 		m_freem(opts);
589 	return(IPPROTO_DONE);
590 }
591 
592 #ifdef INET6
593 static void
594 ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip)
595 {
596 	bzero(ip6, sizeof *ip6);
597 
598 	ip6->ip6_vfc = IPV6_VERSION;
599 	ip6->ip6_plen = ip->ip_len;
600 	ip6->ip6_nxt = ip->ip_p;
601 	ip6->ip6_hlim = ip->ip_ttl;
602 	ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] =
603 		IPV6_ADDR_INT32_SMP;
604 	ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr;
605 	ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr;
606 }
607 #endif
608 
609 /*
610  * subroutine of udp_input(), mainly for source code readability.
611  * caller must properly init udp_ip6 and udp_in6 beforehand.
612  */
613 static void
614 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off,
615     struct sockaddr_in *udp_in,
616     struct udp_in6 *udp_in6, struct udp_ip6 *udp_ip6)
617 {
618 	struct sockaddr *append_sa;
619 	struct mbuf *opts = NULL;
620 
621 	if (last->inp_flags & INP_CONTROLOPTS ||
622 	    last->inp_socket->so_options & SO_TIMESTAMP) {
623 #ifdef INET6
624 		if (last->inp_vflag & INP_IPV6) {
625 			int savedflags;
626 
627 			if (udp_ip6->uip6_init_done == 0) {
628 				ip_2_ip6_hdr(&udp_ip6->uip6_ip6, ip);
629 				udp_ip6->uip6_init_done = 1;
630 			}
631 			savedflags = last->inp_flags;
632 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
633 			ip6_savecontrol(last, &opts, &udp_ip6->uip6_ip6, n);
634 			last->inp_flags = savedflags;
635 		} else
636 #endif
637 		ip_savecontrol(last, &opts, ip, n);
638 	}
639 #ifdef INET6
640 	if (last->inp_vflag & INP_IPV6) {
641 		if (udp_in6->uin6_init_done == 0) {
642 			in6_sin_2_v4mapsin6(udp_in, &udp_in6->uin6_sin);
643 			udp_in6->uin6_init_done = 1;
644 		}
645 		append_sa = (struct sockaddr *)&udp_in6->uin6_sin;
646 	} else
647 #endif
648 		append_sa = (struct sockaddr *)udp_in;
649 	m_adj(n, off);
650 	lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token);
651 	if (ssb_appendaddr(&last->inp_socket->so_rcv, append_sa, n, opts) == 0) {
652 		m_freem(n);
653 		if (opts)
654 			m_freem(opts);
655 		udp_stat.udps_fullsock++;
656 	} else {
657 		sorwakeup(last->inp_socket);
658 	}
659 	lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token);
660 }
661 
662 /*
663  * Notify a udp user of an asynchronous error;
664  * just wake up so that he can collect error status.
665  */
666 void
667 udp_notify(struct inpcb *inp, int error)
668 {
669 	inp->inp_socket->so_error = error;
670 	sorwakeup(inp->inp_socket);
671 	sowwakeup(inp->inp_socket);
672 }
673 
674 struct netmsg_udp_notify {
675 	struct netmsg_base base;
676 	void		(*nm_notify)(struct inpcb *, int);
677 	struct in_addr	nm_faddr;
678 	int		nm_arg;
679 };
680 
681 static void
682 udp_notifyall_oncpu(netmsg_t msg)
683 {
684 	struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg;
685 #if 0
686 	int nextcpu;
687 #endif
688 
689 	in_pcbnotifyall(&udbinfo.pcblisthead, nm->nm_faddr,
690 			nm->nm_arg, nm->nm_notify);
691 	lwkt_replymsg(&nm->base.lmsg, 0);
692 
693 #if 0
694 	/* XXX currently udp only runs on cpu 0 */
695 	nextcpu = mycpuid + 1;
696 	if (nextcpu < ncpus2)
697 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
698 	else
699 		lwkt_replymsg(&nmsg->base.lmsg, 0);
700 #endif
701 }
702 
703 static void
704 udp_rtchange(struct inpcb *inp, int err)
705 {
706 	/* XXX Nuke this, once UDP inpcbs are CPU localized */
707 	if (inp->inp_route.ro_rt && inp->inp_route.ro_rt->rt_cpuid == mycpuid) {
708 		rtfree(inp->inp_route.ro_rt);
709 		inp->inp_route.ro_rt = NULL;
710 		/*
711 		 * A new route can be allocated the next time
712 		 * output is attempted.
713 		 */
714 	}
715 }
716 
717 void
718 udp_ctlinput(netmsg_t msg)
719 {
720 	struct sockaddr *sa = msg->ctlinput.nm_arg;
721 	struct ip *ip = msg->ctlinput.nm_extra;
722 	int cmd = msg->ctlinput.nm_cmd;
723 	struct udphdr *uh;
724 	void (*notify) (struct inpcb *, int) = udp_notify;
725 	struct in_addr faddr;
726 	struct inpcb *inp;
727 
728 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
729 
730 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
731 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
732 		goto done;
733 
734 	if (PRC_IS_REDIRECT(cmd)) {
735 		ip = NULL;
736 		notify = udp_rtchange;
737 	} else if (cmd == PRC_HOSTDEAD) {
738 		ip = NULL;
739 	} else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
740 		goto done;
741 	}
742 
743 	if (ip) {
744 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
745 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
746 					ip->ip_src, uh->uh_sport, 0, NULL);
747 		if (inp != NULL && inp->inp_socket != NULL)
748 			(*notify)(inp, inetctlerrmap[cmd]);
749 	} else if (PRC_IS_REDIRECT(cmd)) {
750 		struct netmsg_udp_notify *nm;
751 
752 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
753 		nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
754 		netmsg_init(&nm->base, NULL, &netisr_afree_rport,
755 			    0, udp_notifyall_oncpu);
756 		nm->nm_faddr = faddr;
757 		nm->nm_arg = inetctlerrmap[cmd];
758 		nm->nm_notify = notify;
759 		lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
760 	} else {
761 		/*
762 		 * XXX We should forward msg upon PRC_HOSTHEAD and ip == NULL,
763 		 * once UDP inpcbs are CPU localized
764 		 */
765 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
766 		in_pcbnotifyall(&udbinfo.pcblisthead, faddr, inetctlerrmap[cmd],
767 				notify);
768 	}
769 done:
770 	lwkt_replymsg(&msg->lmsg, 0);
771 }
772 
773 static int
774 udp_pcblist(SYSCTL_HANDLER_ARGS)
775 {
776 	struct xinpcb *xi;
777 	int error, nxi, i;
778 
779 	udbinfo_lock();
780 	error = in_pcblist_global_nomarker(oidp, arg1, arg2, req, &xi, &nxi);
781 	udbinfo_unlock();
782 
783 	if (error) {
784 		KKASSERT(xi == NULL);
785 		return error;
786 	}
787 	if (nxi == 0) {
788 		KKASSERT(xi == NULL);
789 		return 0;
790 	}
791 
792 	for (i = 0; i < nxi; ++i) {
793 		error = SYSCTL_OUT(req, &xi[i], sizeof(xi[i]));
794 		if (error)
795 			break;
796 	}
797 	kfree(xi, M_TEMP);
798 
799 	return error;
800 }
801 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, &udbinfo, 0,
802 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
803 
804 static int
805 udp_getcred(SYSCTL_HANDLER_ARGS)
806 {
807 	struct sockaddr_in addrs[2];
808 	struct ucred cred0, *cred = NULL;
809 	struct inpcb *inp;
810 	int error;
811 
812 	error = priv_check(req->td, PRIV_ROOT);
813 	if (error)
814 		return (error);
815 	error = SYSCTL_IN(req, addrs, sizeof addrs);
816 	if (error)
817 		return (error);
818 
819 	udbinfo_lock();
820 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
821 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
822 	if (inp == NULL || inp->inp_socket == NULL) {
823 		error = ENOENT;
824 	} else {
825 		if (inp->inp_socket->so_cred != NULL) {
826 			cred0 = *(inp->inp_socket->so_cred);
827 			cred = &cred0;
828 		}
829 	}
830 	udbinfo_unlock();
831 
832 	if (error)
833 		return error;
834 
835 	return SYSCTL_OUT(req, cred, sizeof(struct ucred));
836 }
837 
838 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
839     0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection");
840 
841 static int
842 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *dstaddr,
843     struct thread *td, int flags, boolean_t held_td)
844 {
845 	struct udpiphdr *ui;
846 	int len = m->m_pkthdr.len;
847 	struct sockaddr_in *sin;	/* really is initialized before use */
848 	int error = 0, lport_any = 0;
849 
850 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
851 		error = EMSGSIZE;
852 		goto release;
853 	}
854 
855 	if (inp->inp_lport == 0) {	/* unbound socket */
856 		error = in_pcbbind(inp, NULL, td);
857 		if (error)
858 			goto release;
859 
860 		udbinfo_barrier_set();
861 		in_pcbinswildcardhash(inp);
862 		udbinfo_barrier_rem();
863 		lport_any = 1;
864 	}
865 
866 	if (dstaddr != NULL) {		/* destination address specified */
867 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
868 			/* already connected */
869 			error = EISCONN;
870 			goto release;
871 		}
872 		sin = (struct sockaddr_in *)dstaddr;
873 		if (!prison_remote_ip(td, (struct sockaddr *)&sin)) {
874 			error = EAFNOSUPPORT; /* IPv6 only jail */
875 			goto release;
876 		}
877 	} else {
878 		if (inp->inp_faddr.s_addr == INADDR_ANY) {
879 			/* no destination specified and not already connected */
880 			error = ENOTCONN;
881 			goto release;
882 		}
883 		sin = NULL;
884 	}
885 
886 	/*
887 	 * Calculate data length and get a mbuf
888 	 * for UDP and IP headers.
889 	 */
890 	M_PREPEND(m, sizeof(struct udpiphdr), MB_DONTWAIT);
891 	if (m == NULL) {
892 		error = ENOBUFS;
893 		goto release;
894 	}
895 
896 	/*
897 	 * Fill in mbuf with extended UDP header
898 	 * and addresses and length put into network format.
899 	 */
900 	ui = mtod(m, struct udpiphdr *);
901 	bzero(ui->ui_x1, sizeof ui->ui_x1);	/* XXX still needed? */
902 	ui->ui_pr = IPPROTO_UDP;
903 
904 	/*
905 	 * Set destination address.
906 	 */
907 	if (dstaddr != NULL) {			/* use specified destination */
908 		ui->ui_dst = sin->sin_addr;
909 		ui->ui_dport = sin->sin_port;
910 	} else {				/* use connected destination */
911 		ui->ui_dst = inp->inp_faddr;
912 		ui->ui_dport = inp->inp_fport;
913 	}
914 
915 	/*
916 	 * Set source address.
917 	 */
918 	if (inp->inp_laddr.s_addr == INADDR_ANY ||
919 	    IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) {
920 		struct sockaddr_in *if_sin;
921 
922 		if (dstaddr == NULL) {
923 			/*
924 			 * connect() had (or should have) failed because
925 			 * the interface had no IP address, but the
926 			 * application proceeded to call send() anyways.
927 			 */
928 			error = ENOTCONN;
929 			goto release;
930 		}
931 
932 		/* Look up outgoing interface. */
933 		error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1);
934 		if (error)
935 			goto release;
936 		ui->ui_src = if_sin->sin_addr;	/* use address of interface */
937 	} else {
938 		ui->ui_src = inp->inp_laddr;	/* use non-null bound address */
939 	}
940 	ui->ui_sport = inp->inp_lport;
941 	KASSERT(inp->inp_lport != 0, ("inp lport should have been bound"));
942 
943 	/*
944 	 * Release the original thread, since it is no longer used
945 	 */
946 	if (held_td) {
947 		lwkt_rele(td);
948 		held_td = FALSE;
949 	}
950 
951 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
952 
953 	/*
954 	 * Set up checksum and output datagram.
955 	 */
956 	if (udpcksum) {
957 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr,
958 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
959 		m->m_pkthdr.csum_flags = CSUM_UDP;
960 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
961 		m->m_pkthdr.csum_thlen = sizeof(struct udphdr);
962 	} else {
963 		ui->ui_sum = 0;
964 	}
965 	((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len;
966 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
967 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
968 	udp_stat.udps_opackets++;
969 
970 	error = ip_output(m, inp->inp_options, &inp->inp_route,
971 	    (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)) |
972 	    flags | IP_DEBUGROUTE,
973 	    inp->inp_moptions, inp);
974 
975 	/*
976 	 * If this is the first data gram sent on an unbound and unconnected
977 	 * UDP socket, lport will be changed in this function.  If target
978 	 * CPU after this lport changing is no longer the current CPU, then
979 	 * free the route entry allocated on the current CPU.
980 	 */
981 	if (lport_any) {
982 		if (udp_addrcpu(inp->inp_faddr.s_addr, inp->inp_fport,
983 		    inp->inp_laddr.s_addr, inp->inp_lport) != mycpuid) {
984 #ifdef notyet
985 			struct route *ro = &inp->inp_route;
986 
987 			if (ro->ro_rt != NULL)
988 				RTFREE(ro->ro_rt);
989 			bzero(ro, sizeof(*ro));
990 #else
991 			panic("UDP activity should only be in netisr0");
992 #endif
993 		}
994 	}
995 	return (error);
996 
997 release:
998 	if (held_td)
999 		lwkt_rele(td);
1000 	m_freem(m);
1001 	return (error);
1002 }
1003 
1004 u_long	udp_sendspace = 9216;		/* really max datagram size */
1005 					/* 40 1K datagrams */
1006 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
1007     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
1008 
1009 u_long	udp_recvspace = 40 * (1024 +
1010 #ifdef INET6
1011 				      sizeof(struct sockaddr_in6)
1012 #else
1013 				      sizeof(struct sockaddr_in)
1014 #endif
1015 				      );
1016 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1017     &udp_recvspace, 0, "Maximum incoming UDP datagram size");
1018 
1019 /*
1020  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
1021  *	 will sofree() it when we return.
1022  */
1023 static void
1024 udp_abort(netmsg_t msg)
1025 {
1026 	struct socket *so = msg->abort.base.nm_so;
1027 	struct inpcb *inp;
1028 	int error;
1029 
1030 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1031 
1032 	inp = so->so_pcb;
1033 	if (inp) {
1034 		soisdisconnected(so);
1035 
1036 		udbinfo_barrier_set();
1037 		in_pcbdetach(inp);
1038 		udbinfo_barrier_rem();
1039 		error = 0;
1040 	} else {
1041 		error = EINVAL;
1042 	}
1043 	lwkt_replymsg(&msg->abort.base.lmsg, error);
1044 }
1045 
1046 static void
1047 udp_attach(netmsg_t msg)
1048 {
1049 	struct socket *so = msg->attach.base.nm_so;
1050 	struct pru_attach_info *ai = msg->attach.nm_ai;
1051 	struct inpcb *inp;
1052 	int error;
1053 
1054 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1055 
1056 	inp = so->so_pcb;
1057 	if (inp != NULL) {
1058 		error = EINVAL;
1059 		goto out;
1060 	}
1061 	error = soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit);
1062 	if (error)
1063 		goto out;
1064 
1065 	udbinfo_barrier_set();
1066 	error = in_pcballoc(so, &udbinfo);
1067 	udbinfo_barrier_rem();
1068 
1069 	if (error)
1070 		goto out;
1071 
1072 	/*
1073 	 * Set default port for protocol processing prior to bind/connect.
1074 	 */
1075 	sosetport(so, netisr_cpuport(0));
1076 
1077 	inp = (struct inpcb *)so->so_pcb;
1078 	inp->inp_vflag |= INP_IPV4;
1079 	inp->inp_ip_ttl = ip_defttl;
1080 	error = 0;
1081 out:
1082 	lwkt_replymsg(&msg->attach.base.lmsg, error);
1083 }
1084 
1085 static void
1086 udp_bind(netmsg_t msg)
1087 {
1088 	struct socket *so = msg->bind.base.nm_so;
1089 	struct sockaddr *nam = msg->bind.nm_nam;
1090 	struct thread *td = msg->bind.nm_td;
1091 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1092 	struct inpcb *inp;
1093 	int error;
1094 
1095 	inp = so->so_pcb;
1096 	if (inp) {
1097 		error = in_pcbbind(inp, nam, td);
1098 		if (error == 0) {
1099 			if (sin->sin_addr.s_addr != INADDR_ANY)
1100 				inp->inp_flags |= INP_WASBOUND_NOTANY;
1101 
1102 			udbinfo_barrier_set();
1103 			in_pcbinswildcardhash(inp);
1104 			udbinfo_barrier_rem();
1105 		}
1106 	} else {
1107 		error = EINVAL;
1108 	}
1109 	lwkt_replymsg(&msg->bind.base.lmsg, error);
1110 }
1111 
1112 static void
1113 udp_connect(netmsg_t msg)
1114 {
1115 	struct socket *so = msg->connect.base.nm_so;
1116 	struct sockaddr *nam = msg->connect.nm_nam;
1117 	struct thread *td = msg->connect.nm_td;
1118 	struct inpcb *inp;
1119 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1120 	struct sockaddr_in *if_sin;
1121 	lwkt_port_t port;
1122 	int error;
1123 
1124 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1125 
1126 	inp = so->so_pcb;
1127 	if (inp == NULL) {
1128 		error = EINVAL;
1129 		goto out;
1130 	}
1131 
1132 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1133 		panic("UDP does not support RECONNECT");
1134 #ifdef notyet
1135 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1136 		in_pcblink(inp, &udbinfo);
1137 #endif
1138 	}
1139 
1140 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1141 		error = EISCONN;
1142 		goto out;
1143 	}
1144 	error = 0;
1145 
1146 	/*
1147 	 * Bind if we have to
1148 	 */
1149 	if (td->td_proc && td->td_proc->p_ucred->cr_prison != NULL &&
1150 	    inp->inp_laddr.s_addr == INADDR_ANY) {
1151 		error = in_pcbbind(inp, NULL, td);
1152 		if (error)
1153 			goto out;
1154 	}
1155 
1156 	/*
1157 	 * Calculate the correct protocol processing thread.  The connect
1158 	 * operation must run there.
1159 	 */
1160 	error = in_pcbladdr(inp, nam, &if_sin, td);
1161 	if (error)
1162 		goto out;
1163 	if (!prison_remote_ip(td, nam)) {
1164 		error = EAFNOSUPPORT; /* IPv6 only jail */
1165 		goto out;
1166 	}
1167 
1168 	port = udp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1169 			    inp->inp_laddr.s_addr, inp->inp_lport);
1170 	if (port != &curthread->td_msgport) {
1171 #ifdef notyet
1172 		struct route *ro = &inp->inp_route;
1173 
1174 		/*
1175 		 * in_pcbladdr() may have allocated a route entry for us
1176 		 * on the current CPU, but we need a route entry on the
1177 		 * inpcb's owner CPU, so free it here.
1178 		 */
1179 		if (ro->ro_rt != NULL)
1180 			RTFREE(ro->ro_rt);
1181 		bzero(ro, sizeof(*ro));
1182 
1183 		/*
1184 		 * We are moving the protocol processing port the socket
1185 		 * is on, we have to unlink here and re-link on the
1186 		 * target cpu.
1187 		 */
1188 		in_pcbunlink(so->so_pcb, &udbinfo);
1189 		/* in_pcbunlink(so->so_pcb, &udbinfo[mycpu->gd_cpuid]); */
1190 		sosetport(so, port);
1191 		msg->connect.nm_flags |= PRUC_RECONNECT;
1192 		msg->connect.base.nm_dispatch = udp_connect;
1193 
1194 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1195 		/* msg invalid now */
1196 		return;
1197 #else
1198 		panic("UDP activity should only be in netisr0");
1199 #endif
1200 	}
1201 	KKASSERT(port == &curthread->td_msgport);
1202 	error = udp_connect_oncpu(so, td, sin, if_sin);
1203 out:
1204 	KKASSERT(msg->connect.nm_m == NULL);
1205 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1206 }
1207 
1208 static int
1209 udp_connect_oncpu(struct socket *so, struct thread *td,
1210 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
1211 {
1212 	struct inpcb *inp;
1213 	int error;
1214 
1215 	udbinfo_barrier_set();
1216 
1217 	inp = so->so_pcb;
1218 	if (inp->inp_flags & INP_WILDCARD)
1219 		in_pcbremwildcardhash(inp);
1220 	error = in_pcbconnect(inp, (struct sockaddr *)sin, td);
1221 
1222 	if (error == 0) {
1223 		/*
1224 		 * No more errors can occur, finish adjusting the socket
1225 		 * and change the processing port to reflect the connected
1226 		 * socket.  Once set we can no longer safely mess with the
1227 		 * socket.
1228 		 */
1229 		soisconnected(so);
1230 	} else if (error == EAFNOSUPPORT) {	/* connection dissolved */
1231 		/*
1232 		 * Follow traditional BSD behavior and retain
1233 		 * the local port binding.  But, fix the old misbehavior
1234 		 * of overwriting any previously bound local address.
1235 		 */
1236 		if (!(inp->inp_flags & INP_WASBOUND_NOTANY))
1237 			inp->inp_laddr.s_addr = INADDR_ANY;
1238 		in_pcbinswildcardhash(inp);
1239 	}
1240 
1241 	udbinfo_barrier_rem();
1242 	return error;
1243 }
1244 
1245 static void
1246 udp_detach(netmsg_t msg)
1247 {
1248 	struct socket *so = msg->detach.base.nm_so;
1249 	struct inpcb *inp;
1250 	int error;
1251 
1252 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1253 
1254 	inp = so->so_pcb;
1255 	if (inp) {
1256 		udbinfo_barrier_set();
1257 		in_pcbdetach(inp);
1258 		udbinfo_barrier_rem();
1259 		error = 0;
1260 	} else {
1261 		error = EINVAL;
1262 	}
1263 	lwkt_replymsg(&msg->detach.base.lmsg, error);
1264 }
1265 
1266 static void
1267 udp_disconnect(netmsg_t msg)
1268 {
1269 	struct socket *so = msg->disconnect.base.nm_so;
1270 	struct route *ro;
1271 	struct inpcb *inp;
1272 	int error;
1273 
1274 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1275 
1276 	inp = so->so_pcb;
1277 	if (inp == NULL) {
1278 		error = EINVAL;
1279 		goto out;
1280 	}
1281 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1282 		error = ENOTCONN;
1283 		goto out;
1284 	}
1285 
1286 	soreference(so);
1287 
1288 	udbinfo_barrier_set();
1289 	in_pcbdisconnect(inp);
1290 	udbinfo_barrier_rem();
1291 
1292 	soclrstate(so, SS_ISCONNECTED);		/* XXX */
1293 	sofree(so);
1294 
1295 	ro = &inp->inp_route;
1296 	if (ro->ro_rt != NULL)
1297 		RTFREE(ro->ro_rt);
1298 	bzero(ro, sizeof(*ro));
1299 	error = 0;
1300 out:
1301 	lwkt_replymsg(&msg->disconnect.base.lmsg, error);
1302 }
1303 
1304 static void
1305 udp_send(netmsg_t msg)
1306 {
1307 	struct socket *so = msg->send.base.nm_so;
1308 	struct mbuf *m = msg->send.nm_m;
1309 	struct sockaddr *addr = msg->send.nm_addr;
1310 	int pru_flags = msg->send.nm_flags;
1311 	struct inpcb *inp;
1312 	int error;
1313 
1314 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1315 	KKASSERT(msg->send.nm_control == NULL);
1316 
1317 	inp = so->so_pcb;
1318 	if (inp) {
1319 		struct thread *td = msg->send.nm_td;
1320 		int flags = 0;
1321 
1322 		if (pru_flags & PRUS_DONTROUTE)
1323 			flags |= SO_DONTROUTE;
1324 		error = udp_output(inp, m, addr, td, flags,
1325 		    (pru_flags & PRUS_HELDTD) ? TRUE : FALSE);
1326 	} else {
1327 		if (pru_flags & PRUS_HELDTD)
1328 			lwkt_rele(msg->send.nm_td);
1329 		m_freem(m);
1330 		error = EINVAL;
1331 	}
1332 
1333 	if (pru_flags & PRUS_FREEADDR)
1334 		kfree(addr, M_SONAME);
1335 
1336 	if ((pru_flags & PRUS_NOREPLY) == 0)
1337 		lwkt_replymsg(&msg->send.base.lmsg, error);
1338 }
1339 
1340 void
1341 udp_shutdown(netmsg_t msg)
1342 {
1343 	struct socket *so = msg->shutdown.base.nm_so;
1344 	struct inpcb *inp;
1345 	int error;
1346 
1347 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1348 
1349 	inp = so->so_pcb;
1350 	if (inp) {
1351 		socantsendmore(so);
1352 		error = 0;
1353 	} else {
1354 		error = EINVAL;
1355 	}
1356 	lwkt_replymsg(&msg->shutdown.base.lmsg, error);
1357 }
1358 
1359 void
1360 udbinfo_lock(void)
1361 {
1362 	lwkt_serialize_enter(&udbinfo_slize);
1363 }
1364 
1365 void
1366 udbinfo_unlock(void)
1367 {
1368 	lwkt_serialize_exit(&udbinfo_slize);
1369 }
1370 
1371 void
1372 udbinfo_barrier_set(void)
1373 {
1374 	netisr_barrier_set(udbinfo_br);
1375 	udbinfo_lock();
1376 }
1377 
1378 void
1379 udbinfo_barrier_rem(void)
1380 {
1381 	udbinfo_unlock();
1382 	netisr_barrier_rem(udbinfo_br);
1383 }
1384 
1385 struct pr_usrreqs udp_usrreqs = {
1386 	.pru_abort = udp_abort,
1387 	.pru_accept = pr_generic_notsupp,
1388 	.pru_attach = udp_attach,
1389 	.pru_bind = udp_bind,
1390 	.pru_connect = udp_connect,
1391 	.pru_connect2 = pr_generic_notsupp,
1392 	.pru_control = in_control_dispatch,
1393 	.pru_detach = udp_detach,
1394 	.pru_disconnect = udp_disconnect,
1395 	.pru_listen = pr_generic_notsupp,
1396 	.pru_peeraddr = in_setpeeraddr_dispatch,
1397 	.pru_rcvd = pr_generic_notsupp,
1398 	.pru_rcvoob = pr_generic_notsupp,
1399 	.pru_send = udp_send,
1400 	.pru_sense = pru_sense_null,
1401 	.pru_shutdown = udp_shutdown,
1402 	.pru_sockaddr = in_setsockaddr_dispatch,
1403 	.pru_sosend = sosendudp,
1404 	.pru_soreceive = soreceive
1405 };
1406 
1407