xref: /dragonfly/sys/netinet/udp_usrreq.c (revision 61c0377f)
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 struct	inpcbportinfo udbportinfo;
159 
160 static struct netisr_barrier *udbinfo_br;
161 static struct lwkt_serialize udbinfo_slize = LWKT_SERIALIZE_INITIALIZER;
162 
163 #ifndef UDBHASHSIZE
164 #define UDBHASHSIZE 16
165 #endif
166 
167 struct	udpstat udpstat_percpu[MAXCPU] __cachealign;
168 
169 #ifdef INET6
170 struct udp_in6 {
171 	struct sockaddr_in6	uin6_sin;
172 	u_char			uin6_init_done : 1;
173 };
174 struct udp_ip6 {
175 	struct ip6_hdr		uip6_ip6;
176 	u_char			uip6_init_done : 1;
177 };
178 #else
179 struct udp_in6;
180 struct udp_ip6;
181 #endif /* INET6 */
182 
183 static void udp_append (struct inpcb *last, struct ip *ip,
184     struct mbuf *n, int off, struct sockaddr_in *udp_in,
185     struct udp_in6 *, struct udp_ip6 *);
186 #ifdef INET6
187 static void ip_2_ip6_hdr (struct ip6_hdr *ip6, struct ip *ip);
188 #endif
189 
190 static int udp_connect_oncpu(struct socket *so, struct thread *td,
191 			struct sockaddr_in *sin, struct sockaddr_in *if_sin);
192 static int udp_output (struct inpcb *, struct mbuf *, struct sockaddr *,
193 			struct thread *, int, boolean_t);
194 
195 void
196 udp_init(void)
197 {
198 	int cpu;
199 
200 	in_pcbinfo_init(&udbinfo);
201 	in_pcbportinfo_init(&udbportinfo, UDBHASHSIZE, FALSE, 0);
202 
203 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
204 	udbinfo.portinfo = &udbportinfo;
205 	udbinfo.wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB,
206 					    &udbinfo.wildcardhashmask);
207 	udbinfo.localgrphashbase = hashinit(UDBHASHSIZE, M_PCB,
208 					    &udbinfo.localgrphashmask);
209 	udbinfo.ipi_size = sizeof(struct inpcb);
210 
211 	udbinfo_br = netisr_barrier_create();
212 
213 	/*
214 	 * Initialize UDP statistics counters for each CPU.
215 	 */
216 	for (cpu = 0; cpu < ncpus; ++cpu)
217 		bzero(&udpstat_percpu[cpu], sizeof(struct udpstat));
218 }
219 
220 static int
221 sysctl_udpstat(SYSCTL_HANDLER_ARGS)
222 {
223 	int cpu, error = 0;
224 
225 	for (cpu = 0; cpu < ncpus; ++cpu) {
226 		if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu],
227 					sizeof(struct udpstat))))
228 			break;
229 		if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu],
230 				       sizeof(struct udpstat))))
231 			break;
232 	}
233 
234 	return (error);
235 }
236 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
237     0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics");
238 
239 /*
240  * Check multicast packets to make sure they are only sent to sockets with
241  * multicast memberships for the packet's destination address and arrival
242  * interface.  Multicast packets to multicast-unaware sockets are also
243  * disallowed.
244  *
245  * Returns 0 if the packet is acceptable, -1 if it is not.
246  */
247 static __inline int
248 check_multicast_membership(struct ip *ip, struct inpcb *inp, struct mbuf *m)
249 {
250 	int mshipno;
251 	struct ip_moptions *mopt;
252 
253 	if (strict_mcast_mship == 0 ||
254 	    !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
255 		return (0);
256 	}
257 	mopt = inp->inp_moptions;
258 	if (mopt == NULL)
259 		return (-1);
260 	for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) {
261 		struct in_multi *maddr = mopt->imo_membership[mshipno];
262 
263 		if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr &&
264 		    m->m_pkthdr.rcvif == maddr->inm_ifp) {
265 			return (0);
266 		}
267 	}
268 	return (-1);
269 }
270 
271 int
272 udp_input(struct mbuf **mp, int *offp, int proto)
273 {
274 	struct sockaddr_in udp_in = { sizeof udp_in, AF_INET };
275 #ifdef INET6
276 	struct udp_in6 udp_in6 = {
277 		{ sizeof udp_in6.uin6_sin, AF_INET6 }, 0
278 	};
279 	struct udp_ip6 udp_ip6;
280 #endif
281 
282 	int iphlen;
283 	struct ip *ip;
284 	struct udphdr *uh;
285 	struct inpcb *inp;
286 	struct mbuf *m;
287 	struct mbuf *opts = NULL;
288 	int len, off;
289 	struct ip save_ip;
290 	struct sockaddr *append_sa;
291 
292 	off = *offp;
293 	m = *mp;
294 	*mp = NULL;
295 
296 	iphlen = off;
297 	udp_stat.udps_ipackets++;
298 
299 	/*
300 	 * Strip IP options, if any; should skip this,
301 	 * make available to user, and use on returned packets,
302 	 * but we don't yet have a way to check the checksum
303 	 * with options still present.
304 	 */
305 	if (iphlen > sizeof(struct ip)) {
306 		ip_stripoptions(m);
307 		iphlen = sizeof(struct ip);
308 	}
309 
310 	/*
311 	 * IP and UDP headers are together in first mbuf.
312 	 * Already checked and pulled up in ip_demux().
313 	 */
314 	KASSERT(m->m_len >= iphlen + sizeof(struct udphdr),
315 	    ("UDP header not in one mbuf"));
316 
317 	ip = mtod(m, struct ip *);
318 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
319 
320 	/* destination port of 0 is illegal, based on RFC768. */
321 	if (uh->uh_dport == 0)
322 		goto bad;
323 
324 	/*
325 	 * Make mbuf data length reflect UDP length.
326 	 * If not enough data to reflect UDP length, drop.
327 	 */
328 	len = ntohs((u_short)uh->uh_ulen);
329 	if (ip->ip_len != len) {
330 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
331 			udp_stat.udps_badlen++;
332 			goto bad;
333 		}
334 		m_adj(m, len - ip->ip_len);
335 		/* ip->ip_len = len; */
336 	}
337 	/*
338 	 * Save a copy of the IP header in case we want restore it
339 	 * for sending an ICMP error message in response.
340 	 */
341 	save_ip = *ip;
342 
343 	/*
344 	 * Checksum extended UDP header and data.
345 	 */
346 	if (uh->uh_sum) {
347 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
348 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
349 				uh->uh_sum = m->m_pkthdr.csum_data;
350 			else
351 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
352 				    ip->ip_dst.s_addr, htonl((u_short)len +
353 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
354 			uh->uh_sum ^= 0xffff;
355 		} else {
356 			char b[9];
357 
358 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
359 			bzero(((struct ipovly *)ip)->ih_x1, 9);
360 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
361 			uh->uh_sum = in_cksum(m, len + sizeof(struct ip));
362 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
363 		}
364 		if (uh->uh_sum) {
365 			udp_stat.udps_badsum++;
366 			m_freem(m);
367 			return(IPPROTO_DONE);
368 		}
369 	} else
370 		udp_stat.udps_nosum++;
371 
372 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
373 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
374 		struct inpcb *last;
375 
376 		/*
377 		 * Deliver a multicast or broadcast datagram to *all* sockets
378 		 * for which the local and remote addresses and ports match
379 		 * those of the incoming datagram.  This allows more than
380 		 * one process to receive multi/broadcasts on the same port.
381 		 * (This really ought to be done for unicast datagrams as
382 		 * well, but that would cause problems with existing
383 		 * applications that open both address-specific sockets and
384 		 * a wildcard socket listening to the same port -- they would
385 		 * end up receiving duplicates of every unicast datagram.
386 		 * Those applications open the multiple sockets to overcome an
387 		 * inadequacy of the UDP socket interface, but for backwards
388 		 * compatibility we avoid the problem here rather than
389 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
390 		 */
391 
392 		/*
393 		 * Construct sockaddr format source address.
394 		 */
395 		udp_in.sin_port = uh->uh_sport;
396 		udp_in.sin_addr = ip->ip_src;
397 		/*
398 		 * Locate pcb(s) for datagram.
399 		 * (Algorithm copied from raw_intr().)
400 		 */
401 		last = NULL;
402 #ifdef INET6
403 		udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0;
404 #endif
405 		LIST_FOREACH(inp, &udbinfo.pcblisthead, inp_list) {
406 			KKASSERT((inp->inp_flags & INP_PLACEMARKER) == 0);
407 #ifdef INET6
408 			if (!(inp->inp_vflag & INP_IPV4))
409 				continue;
410 #endif
411 			if (inp->inp_lport != uh->uh_dport)
412 				continue;
413 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
414 				if (inp->inp_laddr.s_addr !=
415 				    ip->ip_dst.s_addr)
416 					continue;
417 			}
418 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
419 				if (inp->inp_faddr.s_addr !=
420 				    ip->ip_src.s_addr ||
421 				    inp->inp_fport != uh->uh_sport)
422 					continue;
423 			}
424 
425 			if (check_multicast_membership(ip, inp, m) < 0)
426 				continue;
427 
428 			if (last != NULL) {
429 				struct mbuf *n;
430 
431 #ifdef IPSEC
432 				/* check AH/ESP integrity. */
433 				if (ipsec4_in_reject_so(m, last->inp_socket))
434 					ipsecstat.in_polvio++;
435 					/* do not inject data to pcb */
436 				else
437 #endif /*IPSEC*/
438 #ifdef FAST_IPSEC
439 				/* check AH/ESP integrity. */
440 				if (ipsec4_in_reject(m, last))
441 					;
442 				else
443 #endif /*FAST_IPSEC*/
444 				if ((n = m_copypacket(m, MB_DONTWAIT)) != NULL)
445 					udp_append(last, ip, n,
446 					    iphlen + sizeof(struct udphdr),
447 					    &udp_in,
448 #ifdef INET6
449 					    &udp_in6, &udp_ip6
450 #else
451 				            NULL, NULL
452 #endif
453 					    );
454 			}
455 			last = inp;
456 			/*
457 			 * Don't look for additional matches if this one does
458 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
459 			 * socket options set.  This heuristic avoids searching
460 			 * through all pcbs in the common case of a non-shared
461 			 * port.  It * assumes that an application will never
462 			 * clear these options after setting them.
463 			 */
464 			if (!(last->inp_socket->so_options &
465 			    (SO_REUSEPORT | SO_REUSEADDR)))
466 				break;
467 		}
468 
469 		if (last == NULL) {
470 			/*
471 			 * No matching pcb found; discard datagram.
472 			 * (No need to send an ICMP Port Unreachable
473 			 * for a broadcast or multicast datgram.)
474 			 */
475 			udp_stat.udps_noportbcast++;
476 			goto bad;
477 		}
478 #ifdef IPSEC
479 		/* check AH/ESP integrity. */
480 		if (ipsec4_in_reject_so(m, last->inp_socket)) {
481 			ipsecstat.in_polvio++;
482 			goto bad;
483 		}
484 #endif /*IPSEC*/
485 #ifdef FAST_IPSEC
486 		/* check AH/ESP integrity. */
487 		if (ipsec4_in_reject(m, last))
488 			goto bad;
489 #endif /*FAST_IPSEC*/
490 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
491 		    &udp_in,
492 #ifdef INET6
493 		    &udp_in6, &udp_ip6
494 #else
495 		    NULL, NULL
496 #endif
497 		    );
498 		return(IPPROTO_DONE);
499 	}
500 	/*
501 	 * Locate pcb for datagram.
502 	 */
503 	inp = in_pcblookup_pkthash(&udbinfo, ip->ip_src, uh->uh_sport,
504 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif,
505 	    udp_reuseport_ext ? m : NULL);
506 	if (inp == NULL) {
507 		if (log_in_vain) {
508 			char buf[sizeof "aaa.bbb.ccc.ddd"];
509 
510 			strcpy(buf, inet_ntoa(ip->ip_dst));
511 			log(LOG_INFO,
512 			    "Connection attempt to UDP %s:%d from %s:%d\n",
513 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
514 			    ntohs(uh->uh_sport));
515 		}
516 		udp_stat.udps_noport++;
517 		if (m->m_flags & (M_BCAST | M_MCAST)) {
518 			udp_stat.udps_noportbcast++;
519 			goto bad;
520 		}
521 		if (blackhole)
522 			goto bad;
523 #ifdef ICMP_BANDLIM
524 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
525 			goto bad;
526 #endif
527 		*ip = save_ip;
528 		ip->ip_len += iphlen;
529 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
530 		return(IPPROTO_DONE);
531 	}
532 #ifdef IPSEC
533 	if (ipsec4_in_reject_so(m, inp->inp_socket)) {
534 		ipsecstat.in_polvio++;
535 		goto bad;
536 	}
537 #endif /*IPSEC*/
538 #ifdef FAST_IPSEC
539 	if (ipsec4_in_reject(m, inp))
540 		goto bad;
541 #endif /*FAST_IPSEC*/
542 	/*
543 	 * Check the minimum TTL for socket.
544 	 */
545 	if (ip->ip_ttl < inp->inp_ip_minttl)
546 		goto bad;
547 
548 	/*
549 	 * Construct sockaddr format source address.
550 	 * Stuff source address and datagram in user buffer.
551 	 */
552 	udp_in.sin_port = uh->uh_sport;
553 	udp_in.sin_addr = ip->ip_src;
554 	if ((inp->inp_flags & INP_CONTROLOPTS) ||
555 	    (inp->inp_socket->so_options & SO_TIMESTAMP)) {
556 #ifdef INET6
557 		if (inp->inp_vflag & INP_IPV6) {
558 			int savedflags;
559 
560 			ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
561 			savedflags = inp->inp_flags;
562 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
563 			ip6_savecontrol(inp, &opts, &udp_ip6.uip6_ip6, m);
564 			inp->inp_flags = savedflags;
565 		} else
566 #endif
567 		ip_savecontrol(inp, &opts, ip, m);
568 	}
569 	m_adj(m, iphlen + sizeof(struct udphdr));
570 #ifdef INET6
571 	if (inp->inp_vflag & INP_IPV6) {
572 		in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
573 		append_sa = (struct sockaddr *)&udp_in6;
574 	} else
575 #endif
576 		append_sa = (struct sockaddr *)&udp_in;
577 
578 	lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token);
579 	if (ssb_appendaddr(&inp->inp_socket->so_rcv, append_sa, m, opts) == 0) {
580 		udp_stat.udps_fullsock++;
581 		lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
582 		goto bad;
583 	}
584 	lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
585 	sorwakeup(inp->inp_socket);
586 	return(IPPROTO_DONE);
587 bad:
588 	m_freem(m);
589 	if (opts)
590 		m_freem(opts);
591 	return(IPPROTO_DONE);
592 }
593 
594 #ifdef INET6
595 static void
596 ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip)
597 {
598 	bzero(ip6, sizeof *ip6);
599 
600 	ip6->ip6_vfc = IPV6_VERSION;
601 	ip6->ip6_plen = ip->ip_len;
602 	ip6->ip6_nxt = ip->ip_p;
603 	ip6->ip6_hlim = ip->ip_ttl;
604 	ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] =
605 		IPV6_ADDR_INT32_SMP;
606 	ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr;
607 	ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr;
608 }
609 #endif
610 
611 /*
612  * subroutine of udp_input(), mainly for source code readability.
613  * caller must properly init udp_ip6 and udp_in6 beforehand.
614  */
615 static void
616 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off,
617     struct sockaddr_in *udp_in,
618     struct udp_in6 *udp_in6, struct udp_ip6 *udp_ip6)
619 {
620 	struct sockaddr *append_sa;
621 	struct mbuf *opts = NULL;
622 
623 	if (last->inp_flags & INP_CONTROLOPTS ||
624 	    last->inp_socket->so_options & SO_TIMESTAMP) {
625 #ifdef INET6
626 		if (last->inp_vflag & INP_IPV6) {
627 			int savedflags;
628 
629 			if (udp_ip6->uip6_init_done == 0) {
630 				ip_2_ip6_hdr(&udp_ip6->uip6_ip6, ip);
631 				udp_ip6->uip6_init_done = 1;
632 			}
633 			savedflags = last->inp_flags;
634 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
635 			ip6_savecontrol(last, &opts, &udp_ip6->uip6_ip6, n);
636 			last->inp_flags = savedflags;
637 		} else
638 #endif
639 		ip_savecontrol(last, &opts, ip, n);
640 	}
641 #ifdef INET6
642 	if (last->inp_vflag & INP_IPV6) {
643 		if (udp_in6->uin6_init_done == 0) {
644 			in6_sin_2_v4mapsin6(udp_in, &udp_in6->uin6_sin);
645 			udp_in6->uin6_init_done = 1;
646 		}
647 		append_sa = (struct sockaddr *)&udp_in6->uin6_sin;
648 	} else
649 #endif
650 		append_sa = (struct sockaddr *)udp_in;
651 	m_adj(n, off);
652 	lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token);
653 	if (ssb_appendaddr(&last->inp_socket->so_rcv, append_sa, n, opts) == 0) {
654 		m_freem(n);
655 		if (opts)
656 			m_freem(opts);
657 		udp_stat.udps_fullsock++;
658 	} else {
659 		sorwakeup(last->inp_socket);
660 	}
661 	lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token);
662 }
663 
664 /*
665  * Notify a udp user of an asynchronous error;
666  * just wake up so that he can collect error status.
667  */
668 void
669 udp_notify(struct inpcb *inp, int error)
670 {
671 	inp->inp_socket->so_error = error;
672 	sorwakeup(inp->inp_socket);
673 	sowwakeup(inp->inp_socket);
674 }
675 
676 struct netmsg_udp_notify {
677 	struct netmsg_base base;
678 	void		(*nm_notify)(struct inpcb *, int);
679 	struct in_addr	nm_faddr;
680 	int		nm_arg;
681 };
682 
683 static void
684 udp_notifyall_oncpu(netmsg_t msg)
685 {
686 	struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg;
687 #if 0
688 	int nextcpu;
689 #endif
690 
691 	in_pcbnotifyall(&udbinfo.pcblisthead, nm->nm_faddr,
692 			nm->nm_arg, nm->nm_notify);
693 	lwkt_replymsg(&nm->base.lmsg, 0);
694 
695 #if 0
696 	/* XXX currently udp only runs on cpu 0 */
697 	nextcpu = mycpuid + 1;
698 	if (nextcpu < ncpus2)
699 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
700 	else
701 		lwkt_replymsg(&nmsg->base.lmsg, 0);
702 #endif
703 }
704 
705 static void
706 udp_rtchange(struct inpcb *inp, int err)
707 {
708 	/* XXX Nuke this, once UDP inpcbs are CPU localized */
709 	if (inp->inp_route.ro_rt && inp->inp_route.ro_rt->rt_cpuid == mycpuid) {
710 		rtfree(inp->inp_route.ro_rt);
711 		inp->inp_route.ro_rt = NULL;
712 		/*
713 		 * A new route can be allocated the next time
714 		 * output is attempted.
715 		 */
716 	}
717 }
718 
719 void
720 udp_ctlinput(netmsg_t msg)
721 {
722 	struct sockaddr *sa = msg->ctlinput.nm_arg;
723 	struct ip *ip = msg->ctlinput.nm_extra;
724 	int cmd = msg->ctlinput.nm_cmd;
725 	struct udphdr *uh;
726 	void (*notify) (struct inpcb *, int) = udp_notify;
727 	struct in_addr faddr;
728 	struct inpcb *inp;
729 
730 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
731 
732 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
733 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
734 		goto done;
735 
736 	if (PRC_IS_REDIRECT(cmd)) {
737 		ip = NULL;
738 		notify = udp_rtchange;
739 	} else if (cmd == PRC_HOSTDEAD) {
740 		ip = NULL;
741 	} else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
742 		goto done;
743 	}
744 
745 	if (ip) {
746 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
747 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
748 					ip->ip_src, uh->uh_sport, 0, NULL);
749 		if (inp != NULL && inp->inp_socket != NULL)
750 			(*notify)(inp, inetctlerrmap[cmd]);
751 	} else if (PRC_IS_REDIRECT(cmd)) {
752 		struct netmsg_udp_notify *nm;
753 
754 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
755 		nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
756 		netmsg_init(&nm->base, NULL, &netisr_afree_rport,
757 			    0, udp_notifyall_oncpu);
758 		nm->nm_faddr = faddr;
759 		nm->nm_arg = inetctlerrmap[cmd];
760 		nm->nm_notify = notify;
761 		lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
762 	} else {
763 		/*
764 		 * XXX We should forward msg upon PRC_HOSTHEAD and ip == NULL,
765 		 * once UDP inpcbs are CPU localized
766 		 */
767 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
768 		in_pcbnotifyall(&udbinfo.pcblisthead, faddr, inetctlerrmap[cmd],
769 				notify);
770 	}
771 done:
772 	lwkt_replymsg(&msg->lmsg, 0);
773 }
774 
775 static int
776 udp_pcblist(SYSCTL_HANDLER_ARGS)
777 {
778 	struct xinpcb *xi;
779 	int error, nxi, i;
780 
781 	udbinfo_lock();
782 	error = in_pcblist_global_nomarker(oidp, arg1, arg2, req, &xi, &nxi);
783 	udbinfo_unlock();
784 
785 	if (error) {
786 		KKASSERT(xi == NULL);
787 		return error;
788 	}
789 	if (nxi == 0) {
790 		KKASSERT(xi == NULL);
791 		return 0;
792 	}
793 
794 	for (i = 0; i < nxi; ++i) {
795 		error = SYSCTL_OUT(req, &xi[i], sizeof(xi[i]));
796 		if (error)
797 			break;
798 	}
799 	kfree(xi, M_TEMP);
800 
801 	return error;
802 }
803 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, &udbinfo, 0,
804 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
805 
806 static int
807 udp_getcred(SYSCTL_HANDLER_ARGS)
808 {
809 	struct sockaddr_in addrs[2];
810 	struct ucred cred0, *cred = NULL;
811 	struct inpcb *inp;
812 	int error;
813 
814 	error = priv_check(req->td, PRIV_ROOT);
815 	if (error)
816 		return (error);
817 	error = SYSCTL_IN(req, addrs, sizeof addrs);
818 	if (error)
819 		return (error);
820 
821 	udbinfo_lock();
822 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
823 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
824 	if (inp == NULL || inp->inp_socket == NULL) {
825 		error = ENOENT;
826 	} else {
827 		if (inp->inp_socket->so_cred != NULL) {
828 			cred0 = *(inp->inp_socket->so_cred);
829 			cred = &cred0;
830 		}
831 	}
832 	udbinfo_unlock();
833 
834 	if (error)
835 		return error;
836 
837 	return SYSCTL_OUT(req, cred, sizeof(struct ucred));
838 }
839 
840 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
841     0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection");
842 
843 static int
844 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *dstaddr,
845     struct thread *td, int flags, boolean_t held_td)
846 {
847 	struct udpiphdr *ui;
848 	int len = m->m_pkthdr.len;
849 	struct sockaddr_in *sin;	/* really is initialized before use */
850 	int error = 0, lport_any = 0;
851 
852 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
853 		error = EMSGSIZE;
854 		goto release;
855 	}
856 
857 	if (inp->inp_lport == 0) {	/* unbound socket */
858 		error = in_pcbbind(inp, NULL, td);
859 		if (error)
860 			goto release;
861 
862 		udbinfo_barrier_set();
863 		in_pcbinswildcardhash(inp);
864 		udbinfo_barrier_rem();
865 		lport_any = 1;
866 	}
867 
868 	if (dstaddr != NULL) {		/* destination address specified */
869 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
870 			/* already connected */
871 			error = EISCONN;
872 			goto release;
873 		}
874 		sin = (struct sockaddr_in *)dstaddr;
875 		if (!prison_remote_ip(td, (struct sockaddr *)&sin)) {
876 			error = EAFNOSUPPORT; /* IPv6 only jail */
877 			goto release;
878 		}
879 	} else {
880 		if (inp->inp_faddr.s_addr == INADDR_ANY) {
881 			/* no destination specified and not already connected */
882 			error = ENOTCONN;
883 			goto release;
884 		}
885 		sin = NULL;
886 	}
887 
888 	/*
889 	 * Calculate data length and get a mbuf
890 	 * for UDP and IP headers.
891 	 */
892 	M_PREPEND(m, sizeof(struct udpiphdr), MB_DONTWAIT);
893 	if (m == NULL) {
894 		error = ENOBUFS;
895 		goto release;
896 	}
897 
898 	/*
899 	 * Fill in mbuf with extended UDP header
900 	 * and addresses and length put into network format.
901 	 */
902 	ui = mtod(m, struct udpiphdr *);
903 	bzero(ui->ui_x1, sizeof ui->ui_x1);	/* XXX still needed? */
904 	ui->ui_pr = IPPROTO_UDP;
905 
906 	/*
907 	 * Set destination address.
908 	 */
909 	if (dstaddr != NULL) {			/* use specified destination */
910 		ui->ui_dst = sin->sin_addr;
911 		ui->ui_dport = sin->sin_port;
912 	} else {				/* use connected destination */
913 		ui->ui_dst = inp->inp_faddr;
914 		ui->ui_dport = inp->inp_fport;
915 	}
916 
917 	/*
918 	 * Set source address.
919 	 */
920 	if (inp->inp_laddr.s_addr == INADDR_ANY ||
921 	    IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) {
922 		struct sockaddr_in *if_sin;
923 
924 		if (dstaddr == NULL) {
925 			/*
926 			 * connect() had (or should have) failed because
927 			 * the interface had no IP address, but the
928 			 * application proceeded to call send() anyways.
929 			 */
930 			error = ENOTCONN;
931 			goto release;
932 		}
933 
934 		/* Look up outgoing interface. */
935 		error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1);
936 		if (error)
937 			goto release;
938 		ui->ui_src = if_sin->sin_addr;	/* use address of interface */
939 	} else {
940 		ui->ui_src = inp->inp_laddr;	/* use non-null bound address */
941 	}
942 	ui->ui_sport = inp->inp_lport;
943 	KASSERT(inp->inp_lport != 0, ("inp lport should have been bound"));
944 
945 	/*
946 	 * Release the original thread, since it is no longer used
947 	 */
948 	if (held_td) {
949 		lwkt_rele(td);
950 		held_td = FALSE;
951 	}
952 
953 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
954 
955 	/*
956 	 * Set up checksum and output datagram.
957 	 */
958 	if (udpcksum) {
959 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr,
960 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
961 		m->m_pkthdr.csum_flags = CSUM_UDP;
962 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
963 		m->m_pkthdr.csum_thlen = sizeof(struct udphdr);
964 	} else {
965 		ui->ui_sum = 0;
966 	}
967 	((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len;
968 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
969 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
970 	udp_stat.udps_opackets++;
971 
972 	error = ip_output(m, inp->inp_options, &inp->inp_route,
973 	    (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)) |
974 	    flags | IP_DEBUGROUTE,
975 	    inp->inp_moptions, inp);
976 
977 	/*
978 	 * If this is the first data gram sent on an unbound and unconnected
979 	 * UDP socket, lport will be changed in this function.  If target
980 	 * CPU after this lport changing is no longer the current CPU, then
981 	 * free the route entry allocated on the current CPU.
982 	 */
983 	if (lport_any) {
984 		if (udp_addrcpu(inp->inp_faddr.s_addr, inp->inp_fport,
985 		    inp->inp_laddr.s_addr, inp->inp_lport) != mycpuid) {
986 #ifdef notyet
987 			struct route *ro = &inp->inp_route;
988 
989 			if (ro->ro_rt != NULL)
990 				RTFREE(ro->ro_rt);
991 			bzero(ro, sizeof(*ro));
992 #else
993 			panic("UDP activity should only be in netisr0");
994 #endif
995 		}
996 	}
997 	return (error);
998 
999 release:
1000 	if (held_td)
1001 		lwkt_rele(td);
1002 	m_freem(m);
1003 	return (error);
1004 }
1005 
1006 u_long	udp_sendspace = 9216;		/* really max datagram size */
1007 					/* 40 1K datagrams */
1008 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
1009     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
1010 
1011 u_long	udp_recvspace = 40 * (1024 +
1012 #ifdef INET6
1013 				      sizeof(struct sockaddr_in6)
1014 #else
1015 				      sizeof(struct sockaddr_in)
1016 #endif
1017 				      );
1018 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1019     &udp_recvspace, 0, "Maximum incoming UDP datagram size");
1020 
1021 /*
1022  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
1023  *	 will sofree() it when we return.
1024  */
1025 static void
1026 udp_abort(netmsg_t msg)
1027 {
1028 	struct socket *so = msg->abort.base.nm_so;
1029 	struct inpcb *inp;
1030 	int error;
1031 
1032 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1033 
1034 	inp = so->so_pcb;
1035 	if (inp) {
1036 		soisdisconnected(so);
1037 
1038 		udbinfo_barrier_set();
1039 		in_pcbdetach(inp);
1040 		udbinfo_barrier_rem();
1041 		error = 0;
1042 	} else {
1043 		error = EINVAL;
1044 	}
1045 	lwkt_replymsg(&msg->abort.base.lmsg, error);
1046 }
1047 
1048 static void
1049 udp_attach(netmsg_t msg)
1050 {
1051 	struct socket *so = msg->attach.base.nm_so;
1052 	struct pru_attach_info *ai = msg->attach.nm_ai;
1053 	struct inpcb *inp;
1054 	int error;
1055 
1056 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1057 
1058 	inp = so->so_pcb;
1059 	if (inp != NULL) {
1060 		error = EINVAL;
1061 		goto out;
1062 	}
1063 	error = soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit);
1064 	if (error)
1065 		goto out;
1066 
1067 	udbinfo_barrier_set();
1068 	error = in_pcballoc(so, &udbinfo);
1069 	udbinfo_barrier_rem();
1070 
1071 	if (error)
1072 		goto out;
1073 
1074 	/*
1075 	 * Set default port for protocol processing prior to bind/connect.
1076 	 */
1077 	sosetport(so, netisr_cpuport(0));
1078 
1079 	inp = (struct inpcb *)so->so_pcb;
1080 	inp->inp_vflag |= INP_IPV4;
1081 	inp->inp_ip_ttl = ip_defttl;
1082 	error = 0;
1083 out:
1084 	lwkt_replymsg(&msg->attach.base.lmsg, error);
1085 }
1086 
1087 static void
1088 udp_bind(netmsg_t msg)
1089 {
1090 	struct socket *so = msg->bind.base.nm_so;
1091 	struct sockaddr *nam = msg->bind.nm_nam;
1092 	struct thread *td = msg->bind.nm_td;
1093 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1094 	struct inpcb *inp;
1095 	int error;
1096 
1097 	inp = so->so_pcb;
1098 	if (inp) {
1099 		error = in_pcbbind(inp, nam, td);
1100 		if (error == 0) {
1101 			if (sin->sin_addr.s_addr != INADDR_ANY)
1102 				inp->inp_flags |= INP_WASBOUND_NOTANY;
1103 
1104 			udbinfo_barrier_set();
1105 			in_pcbinswildcardhash(inp);
1106 			udbinfo_barrier_rem();
1107 		}
1108 	} else {
1109 		error = EINVAL;
1110 	}
1111 	lwkt_replymsg(&msg->bind.base.lmsg, error);
1112 }
1113 
1114 static void
1115 udp_connect(netmsg_t msg)
1116 {
1117 	struct socket *so = msg->connect.base.nm_so;
1118 	struct sockaddr *nam = msg->connect.nm_nam;
1119 	struct thread *td = msg->connect.nm_td;
1120 	struct inpcb *inp;
1121 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1122 	struct sockaddr_in *if_sin;
1123 	lwkt_port_t port;
1124 	int error;
1125 
1126 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1127 
1128 	inp = so->so_pcb;
1129 	if (inp == NULL) {
1130 		error = EINVAL;
1131 		goto out;
1132 	}
1133 
1134 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1135 		panic("UDP does not support RECONNECT");
1136 #ifdef notyet
1137 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1138 		in_pcblink(inp, &udbinfo);
1139 #endif
1140 	}
1141 
1142 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1143 		error = EISCONN;
1144 		goto out;
1145 	}
1146 	error = 0;
1147 
1148 	/*
1149 	 * Bind if we have to
1150 	 */
1151 	if (td->td_proc && td->td_proc->p_ucred->cr_prison != NULL &&
1152 	    inp->inp_laddr.s_addr == INADDR_ANY) {
1153 		error = in_pcbbind(inp, NULL, td);
1154 		if (error)
1155 			goto out;
1156 	}
1157 
1158 	/*
1159 	 * Calculate the correct protocol processing thread.  The connect
1160 	 * operation must run there.
1161 	 */
1162 	error = in_pcbladdr(inp, nam, &if_sin, td);
1163 	if (error)
1164 		goto out;
1165 	if (!prison_remote_ip(td, nam)) {
1166 		error = EAFNOSUPPORT; /* IPv6 only jail */
1167 		goto out;
1168 	}
1169 
1170 	port = udp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1171 			    inp->inp_laddr.s_addr, inp->inp_lport);
1172 	if (port != &curthread->td_msgport) {
1173 #ifdef notyet
1174 		struct route *ro = &inp->inp_route;
1175 
1176 		/*
1177 		 * in_pcbladdr() may have allocated a route entry for us
1178 		 * on the current CPU, but we need a route entry on the
1179 		 * inpcb's owner CPU, so free it here.
1180 		 */
1181 		if (ro->ro_rt != NULL)
1182 			RTFREE(ro->ro_rt);
1183 		bzero(ro, sizeof(*ro));
1184 
1185 		/*
1186 		 * We are moving the protocol processing port the socket
1187 		 * is on, we have to unlink here and re-link on the
1188 		 * target cpu.
1189 		 */
1190 		in_pcbunlink(so->so_pcb, &udbinfo);
1191 		/* in_pcbunlink(so->so_pcb, &udbinfo[mycpu->gd_cpuid]); */
1192 		sosetport(so, port);
1193 		msg->connect.nm_flags |= PRUC_RECONNECT;
1194 		msg->connect.base.nm_dispatch = udp_connect;
1195 
1196 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1197 		/* msg invalid now */
1198 		return;
1199 #else
1200 		panic("UDP activity should only be in netisr0");
1201 #endif
1202 	}
1203 	KKASSERT(port == &curthread->td_msgport);
1204 	error = udp_connect_oncpu(so, td, sin, if_sin);
1205 out:
1206 	KKASSERT(msg->connect.nm_m == NULL);
1207 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1208 }
1209 
1210 static int
1211 udp_connect_oncpu(struct socket *so, struct thread *td,
1212 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
1213 {
1214 	struct inpcb *inp;
1215 	int error;
1216 
1217 	udbinfo_barrier_set();
1218 
1219 	inp = so->so_pcb;
1220 	if (inp->inp_flags & INP_WILDCARD)
1221 		in_pcbremwildcardhash(inp);
1222 	error = in_pcbconnect(inp, (struct sockaddr *)sin, td);
1223 
1224 	if (error == 0) {
1225 		/*
1226 		 * No more errors can occur, finish adjusting the socket
1227 		 * and change the processing port to reflect the connected
1228 		 * socket.  Once set we can no longer safely mess with the
1229 		 * socket.
1230 		 */
1231 		soisconnected(so);
1232 	} else if (error == EAFNOSUPPORT) {	/* connection dissolved */
1233 		/*
1234 		 * Follow traditional BSD behavior and retain
1235 		 * the local port binding.  But, fix the old misbehavior
1236 		 * of overwriting any previously bound local address.
1237 		 */
1238 		if (!(inp->inp_flags & INP_WASBOUND_NOTANY))
1239 			inp->inp_laddr.s_addr = INADDR_ANY;
1240 		in_pcbinswildcardhash(inp);
1241 	}
1242 
1243 	udbinfo_barrier_rem();
1244 	return error;
1245 }
1246 
1247 static void
1248 udp_detach(netmsg_t msg)
1249 {
1250 	struct socket *so = msg->detach.base.nm_so;
1251 	struct inpcb *inp;
1252 	int error;
1253 
1254 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1255 
1256 	inp = so->so_pcb;
1257 	if (inp) {
1258 		udbinfo_barrier_set();
1259 		in_pcbdetach(inp);
1260 		udbinfo_barrier_rem();
1261 		error = 0;
1262 	} else {
1263 		error = EINVAL;
1264 	}
1265 	lwkt_replymsg(&msg->detach.base.lmsg, error);
1266 }
1267 
1268 static void
1269 udp_disconnect(netmsg_t msg)
1270 {
1271 	struct socket *so = msg->disconnect.base.nm_so;
1272 	struct route *ro;
1273 	struct inpcb *inp;
1274 	int error;
1275 
1276 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1277 
1278 	inp = so->so_pcb;
1279 	if (inp == NULL) {
1280 		error = EINVAL;
1281 		goto out;
1282 	}
1283 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1284 		error = ENOTCONN;
1285 		goto out;
1286 	}
1287 
1288 	soreference(so);
1289 
1290 	udbinfo_barrier_set();
1291 	in_pcbdisconnect(inp);
1292 	udbinfo_barrier_rem();
1293 
1294 	soclrstate(so, SS_ISCONNECTED);		/* XXX */
1295 	sofree(so);
1296 
1297 	ro = &inp->inp_route;
1298 	if (ro->ro_rt != NULL)
1299 		RTFREE(ro->ro_rt);
1300 	bzero(ro, sizeof(*ro));
1301 	error = 0;
1302 out:
1303 	lwkt_replymsg(&msg->disconnect.base.lmsg, error);
1304 }
1305 
1306 static void
1307 udp_send(netmsg_t msg)
1308 {
1309 	struct socket *so = msg->send.base.nm_so;
1310 	struct mbuf *m = msg->send.nm_m;
1311 	struct sockaddr *addr = msg->send.nm_addr;
1312 	int pru_flags = msg->send.nm_flags;
1313 	struct inpcb *inp;
1314 	int error;
1315 
1316 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1317 	KKASSERT(msg->send.nm_control == NULL);
1318 
1319 	inp = so->so_pcb;
1320 	if (inp) {
1321 		struct thread *td = msg->send.nm_td;
1322 		int flags = 0;
1323 
1324 		if (pru_flags & PRUS_DONTROUTE)
1325 			flags |= SO_DONTROUTE;
1326 		error = udp_output(inp, m, addr, td, flags,
1327 		    (pru_flags & PRUS_HELDTD) ? TRUE : FALSE);
1328 	} else {
1329 		if (pru_flags & PRUS_HELDTD)
1330 			lwkt_rele(msg->send.nm_td);
1331 		m_freem(m);
1332 		error = EINVAL;
1333 	}
1334 
1335 	if (pru_flags & PRUS_FREEADDR)
1336 		kfree(addr, M_SONAME);
1337 
1338 	if ((pru_flags & PRUS_NOREPLY) == 0)
1339 		lwkt_replymsg(&msg->send.base.lmsg, error);
1340 }
1341 
1342 void
1343 udp_shutdown(netmsg_t msg)
1344 {
1345 	struct socket *so = msg->shutdown.base.nm_so;
1346 	struct inpcb *inp;
1347 	int error;
1348 
1349 	KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1350 
1351 	inp = so->so_pcb;
1352 	if (inp) {
1353 		socantsendmore(so);
1354 		error = 0;
1355 	} else {
1356 		error = EINVAL;
1357 	}
1358 	lwkt_replymsg(&msg->shutdown.base.lmsg, error);
1359 }
1360 
1361 void
1362 udbinfo_lock(void)
1363 {
1364 	lwkt_serialize_enter(&udbinfo_slize);
1365 }
1366 
1367 void
1368 udbinfo_unlock(void)
1369 {
1370 	lwkt_serialize_exit(&udbinfo_slize);
1371 }
1372 
1373 void
1374 udbinfo_barrier_set(void)
1375 {
1376 	netisr_barrier_set(udbinfo_br);
1377 	udbinfo_lock();
1378 }
1379 
1380 void
1381 udbinfo_barrier_rem(void)
1382 {
1383 	udbinfo_unlock();
1384 	netisr_barrier_rem(udbinfo_br);
1385 }
1386 
1387 struct pr_usrreqs udp_usrreqs = {
1388 	.pru_abort = udp_abort,
1389 	.pru_accept = pr_generic_notsupp,
1390 	.pru_attach = udp_attach,
1391 	.pru_bind = udp_bind,
1392 	.pru_connect = udp_connect,
1393 	.pru_connect2 = pr_generic_notsupp,
1394 	.pru_control = in_control_dispatch,
1395 	.pru_detach = udp_detach,
1396 	.pru_disconnect = udp_disconnect,
1397 	.pru_listen = pr_generic_notsupp,
1398 	.pru_peeraddr = in_setpeeraddr_dispatch,
1399 	.pru_rcvd = pr_generic_notsupp,
1400 	.pru_rcvoob = pr_generic_notsupp,
1401 	.pru_send = udp_send,
1402 	.pru_sense = pru_sense_null,
1403 	.pru_shutdown = udp_shutdown,
1404 	.pru_sockaddr = in_setsockaddr_dispatch,
1405 	.pru_sosend = sosendudp,
1406 	.pru_soreceive = soreceive
1407 };
1408 
1409