xref: /freebsd/sys/netinet/udp_usrreq.c (revision 657729a89dd578d8cfc70d6616f5c65a48a8b33a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.
6  * Copyright (c) 2008 Robert N. M. Watson
7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
8  * Copyright (c) 2014 Kevin Lo
9  * All rights reserved.
10  *
11  * Portions of this software were developed by Robert N. M. Watson under
12  * contract to Juniper Networks, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_route.h"
48 #include "opt_rss.h"
49 
50 #include <sys/param.h>
51 #include <sys/domain.h>
52 #include <sys/eventhandler.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/protosw.h>
61 #include <sys/sdt.h>
62 #include <sys/signalvar.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/systm.h>
69 
70 #include <vm/uma.h>
71 
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/route.h>
75 #include <net/route/nhop.h>
76 #include <net/rss_config.h>
77 
78 #include <netinet/in.h>
79 #include <netinet/in_kdtrace.h>
80 #include <netinet/in_fib.h>
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/in_var.h>
84 #include <netinet/ip.h>
85 #ifdef INET6
86 #include <netinet/ip6.h>
87 #endif
88 #include <netinet/ip_icmp.h>
89 #include <netinet/icmp_var.h>
90 #include <netinet/ip_var.h>
91 #include <netinet/ip_options.h>
92 #ifdef INET6
93 #include <netinet6/ip6_var.h>
94 #endif
95 #include <netinet/udp.h>
96 #include <netinet/udp_var.h>
97 #include <netinet/udplite.h>
98 #include <netinet/in_rss.h>
99 
100 #include <netipsec/ipsec_support.h>
101 
102 #include <machine/in_cksum.h>
103 
104 #include <security/mac/mac_framework.h>
105 
106 /*
107  * UDP and UDP-Lite protocols implementation.
108  * Per RFC 768, August, 1980.
109  * Per RFC 3828, July, 2004.
110  */
111 
112 /*
113  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
114  * removes the only data integrity mechanism for packets and malformed
115  * packets that would otherwise be discarded due to bad checksums, and may
116  * cause problems (especially for NFS data blocks).
117  */
118 VNET_DEFINE(int, udp_cksum) = 1;
119 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
120     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
121 
122 VNET_DEFINE(int, udp_log_in_vain) = 0;
123 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
124     &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets");
125 
126 VNET_DEFINE(int, udp_blackhole) = 0;
127 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
128     &VNET_NAME(udp_blackhole), 0,
129     "Do not send port unreachables for refused connects");
130 VNET_DEFINE(bool, udp_blackhole_local) = false;
131 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
132     CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false,
133     "Enforce net.inet.udp.blackhole for locally originated packets");
134 
135 u_long	udp_sendspace = 9216;		/* really max datagram size */
136 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
137     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
138 
139 u_long	udp_recvspace = 40 * (1024 +
140 #ifdef INET6
141 				      sizeof(struct sockaddr_in6)
142 #else
143 				      sizeof(struct sockaddr_in)
144 #endif
145 				      );	/* 40 1K datagrams */
146 
147 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
148     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
149 
150 VNET_DEFINE(struct inpcbinfo, udbinfo);
151 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
152 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone);
153 #define	V_udpcb_zone			VNET(udpcb_zone)
154 
155 #ifndef UDBHASHSIZE
156 #define	UDBHASHSIZE	128
157 #endif
158 
159 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);		/* from udp_var.h */
160 VNET_PCPUSTAT_SYSINIT(udpstat);
161 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
162     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
163 
164 #ifdef VIMAGE
165 VNET_PCPUSTAT_SYSUNINIT(udpstat);
166 #endif /* VIMAGE */
167 #ifdef INET
168 static void	udp_detach(struct socket *so);
169 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
170 		    struct mbuf *, struct thread *, int);
171 #endif
172 
173 INPCBSTORAGE_DEFINE(udpcbstor, inpcb, "udpinp", "udp_inpcb", "udp", "udphash");
174 INPCBSTORAGE_DEFINE(udplitecbstor, inpcb, "udpliteinp", "udplite_inpcb",
175     "udplite", "udplitehash");
176 
177 static void
178 udp_vnet_init(void *arg __unused)
179 {
180 
181 	/*
182 	 * For now default to 2-tuple UDP hashing - until the fragment
183 	 * reassembly code can also update the flowid.
184 	 *
185 	 * Once we can calculate the flowid that way and re-establish
186 	 * a 4-tuple, flip this to 4-tuple.
187 	 */
188 	in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE);
189 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
190 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
191 	uma_zone_set_max(V_udpcb_zone, maxsockets);
192 	uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
193 
194 	/* Additional pcbinfo for UDP-Lite */
195 	in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE,
196 	    UDBHASHSIZE);
197 }
198 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
199     udp_vnet_init, NULL);
200 
201 /*
202  * Kernel module interface for updating udpstat.  The argument is an index
203  * into udpstat treated as an array of u_long.  While this encodes the
204  * general layout of udpstat into the caller, it doesn't encode its location,
205  * so that future changes to add, for example, per-CPU stats support won't
206  * cause binary compatibility problems for kernel modules.
207  */
208 void
209 kmod_udpstat_inc(int statnum)
210 {
211 
212 	counter_u64_add(VNET(udpstat)[statnum], 1);
213 }
214 
215 int
216 udp_newudpcb(struct inpcb *inp)
217 {
218 	struct udpcb *up;
219 
220 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
221 	if (up == NULL)
222 		return (ENOBUFS);
223 	inp->inp_ppcb = up;
224 	return (0);
225 }
226 
227 void
228 udp_discardcb(struct udpcb *up)
229 {
230 
231 	uma_zfree(V_udpcb_zone, up);
232 }
233 
234 #ifdef VIMAGE
235 static void
236 udp_destroy(void *unused __unused)
237 {
238 
239 	in_pcbinfo_destroy(&V_udbinfo);
240 	uma_zdestroy(V_udpcb_zone);
241 }
242 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
243 
244 static void
245 udplite_destroy(void *unused __unused)
246 {
247 
248 	in_pcbinfo_destroy(&V_ulitecbinfo);
249 }
250 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
251     NULL);
252 #endif
253 
254 #ifdef INET
255 /*
256  * Subroutine of udp_input(), which appends the provided mbuf chain to the
257  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
258  * contains the source address.  If the socket ends up being an IPv6 socket,
259  * udp_append() will convert to a sockaddr_in6 before passing the address
260  * into the socket code.
261  *
262  * In the normal case udp_append() will return 0, indicating that you
263  * must unlock the inp. However if a tunneling protocol is in place we increment
264  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
265  * then decrement the reference count. If the inp_rele returns 1, indicating the
266  * inp is gone, we return that to the caller to tell them *not* to unlock
267  * the inp. In the case of multi-cast this will cause the distribution
268  * to stop (though most tunneling protocols known currently do *not* use
269  * multicast).
270  */
271 static int
272 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
273     struct sockaddr_in *udp_in)
274 {
275 	struct sockaddr *append_sa;
276 	struct socket *so;
277 	struct mbuf *tmpopts, *opts = NULL;
278 #ifdef INET6
279 	struct sockaddr_in6 udp_in6;
280 #endif
281 	struct udpcb *up;
282 	bool filtered;
283 
284 	INP_LOCK_ASSERT(inp);
285 
286 	/*
287 	 * Engage the tunneling protocol.
288 	 */
289 	up = intoudpcb(inp);
290 	if (up->u_tun_func != NULL) {
291 		in_pcbref(inp);
292 		INP_RUNLOCK(inp);
293 		filtered = (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
294 		    up->u_tun_ctx);
295 		INP_RLOCK(inp);
296 		if (filtered)
297 			return (in_pcbrele_rlocked(inp));
298 	}
299 
300 	off += sizeof(struct udphdr);
301 
302 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
303 	/* Check AH/ESP integrity. */
304 	if (IPSEC_ENABLED(ipv4) &&
305 	    IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
306 		m_freem(n);
307 		return (0);
308 	}
309 	if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
310 		if (IPSEC_ENABLED(ipv4) &&
311 		    UDPENCAP_INPUT(n, off, AF_INET) != 0)
312 			return (0);	/* Consumed. */
313 	}
314 #endif /* IPSEC */
315 #ifdef MAC
316 	if (mac_inpcb_check_deliver(inp, n) != 0) {
317 		m_freem(n);
318 		return (0);
319 	}
320 #endif /* MAC */
321 	if (inp->inp_flags & INP_CONTROLOPTS ||
322 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
323 #ifdef INET6
324 		if (inp->inp_vflag & INP_IPV6)
325 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
326 		else
327 #endif /* INET6 */
328 			ip_savecontrol(inp, &opts, ip, n);
329 	}
330 	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
331 		tmpopts = sbcreatecontrol(&udp_in[1],
332 		    sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP,
333 		    M_NOWAIT);
334 		if (tmpopts) {
335 			if (opts) {
336 				tmpopts->m_next = opts;
337 				opts = tmpopts;
338 			} else
339 				opts = tmpopts;
340 		}
341 	}
342 #ifdef INET6
343 	if (inp->inp_vflag & INP_IPV6) {
344 		bzero(&udp_in6, sizeof(udp_in6));
345 		udp_in6.sin6_len = sizeof(udp_in6);
346 		udp_in6.sin6_family = AF_INET6;
347 		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
348 		append_sa = (struct sockaddr *)&udp_in6;
349 	} else
350 #endif /* INET6 */
351 		append_sa = (struct sockaddr *)&udp_in[0];
352 	m_adj(n, off);
353 
354 	so = inp->inp_socket;
355 	SOCKBUF_LOCK(&so->so_rcv);
356 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
357 		soroverflow_locked(so);
358 		m_freem(n);
359 		if (opts)
360 			m_freem(opts);
361 		UDPSTAT_INC(udps_fullsock);
362 	} else
363 		sorwakeup_locked(so);
364 	return (0);
365 }
366 
367 static bool
368 udp_multi_match(const struct inpcb *inp, void *v)
369 {
370 	struct ip *ip = v;
371 	struct udphdr *uh = (struct udphdr *)(ip + 1);
372 
373 	if (inp->inp_lport != uh->uh_dport)
374 		return (false);
375 #ifdef INET6
376 	if ((inp->inp_vflag & INP_IPV4) == 0)
377 		return (false);
378 #endif
379 	if (inp->inp_laddr.s_addr != INADDR_ANY &&
380 	    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
381 		return (false);
382 	if (inp->inp_faddr.s_addr != INADDR_ANY &&
383 	    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
384 		return (false);
385 	if (inp->inp_fport != 0 &&
386 	    inp->inp_fport != uh->uh_sport)
387 		return (false);
388 
389 	return (true);
390 }
391 
392 static int
393 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in)
394 {
395 	struct ip *ip = mtod(m, struct ip *);
396 	struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto),
397 	    INPLOOKUP_RLOCKPCB, udp_multi_match, ip);
398 #ifdef KDTRACE_HOOKS
399 	struct udphdr *uh = (struct udphdr *)(ip + 1);
400 #endif
401 	struct inpcb *inp;
402 	struct mbuf *n;
403 	int appends = 0;
404 
405 	MPASS(ip->ip_hl == sizeof(struct ip) >> 2);
406 
407 	while ((inp = inp_next(&inpi)) != NULL) {
408 		/*
409 		 * XXXRW: Because we weren't holding either the inpcb
410 		 * or the hash lock when we checked for a match
411 		 * before, we should probably recheck now that the
412 		 * inpcb lock is held.
413 		 */
414 		/*
415 		 * Handle socket delivery policy for any-source
416 		 * and source-specific multicast. [RFC3678]
417 		 */
418 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
419 			struct ip_moptions	*imo;
420 			struct sockaddr_in	 group;
421 			int			 blocked;
422 
423 			imo = inp->inp_moptions;
424 			if (imo == NULL)
425 				continue;
426 			bzero(&group, sizeof(struct sockaddr_in));
427 			group.sin_len = sizeof(struct sockaddr_in);
428 			group.sin_family = AF_INET;
429 			group.sin_addr = ip->ip_dst;
430 
431 			blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif,
432 				(struct sockaddr *)&group,
433 				(struct sockaddr *)&udp_in[0]);
434 			if (blocked != MCAST_PASS) {
435 				if (blocked == MCAST_NOTGMEMBER)
436 					IPSTAT_INC(ips_notmember);
437 				if (blocked == MCAST_NOTSMEMBER ||
438 				    blocked == MCAST_MUTED)
439 					UDPSTAT_INC(udps_filtermcast);
440 				continue;
441 			}
442 		}
443 		if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) {
444 			if (proto == IPPROTO_UDPLITE)
445 				UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
446 			else
447 				UDP_PROBE(receive, NULL, inp, ip, inp, uh);
448 			if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) {
449 				break;
450 			} else
451 				appends++;
452 		}
453 		/*
454 		 * Don't look for additional matches if this one does
455 		 * not have either the SO_REUSEPORT or SO_REUSEADDR
456 		 * socket options set.  This heuristic avoids
457 		 * searching through all pcbs in the common case of a
458 		 * non-shared port.  It assumes that an application
459 		 * will never clear these options after setting them.
460 		 */
461 		if ((inp->inp_socket->so_options &
462 		    (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) {
463 			INP_RUNLOCK(inp);
464 			break;
465 		}
466 	}
467 
468 	if (appends == 0) {
469 		/*
470 		 * No matching pcb found; discard datagram.  (No need
471 		 * to send an ICMP Port Unreachable for a broadcast
472 		 * or multicast datgram.)
473 		 */
474 		UDPSTAT_INC(udps_noport);
475 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
476 			UDPSTAT_INC(udps_noportmcast);
477 		else
478 			UDPSTAT_INC(udps_noportbcast);
479 	}
480 	m_freem(m);
481 
482 	return (IPPROTO_DONE);
483 }
484 
485 static int
486 udp_input(struct mbuf **mp, int *offp, int proto)
487 {
488 	struct ip *ip;
489 	struct udphdr *uh;
490 	struct ifnet *ifp;
491 	struct inpcb *inp;
492 	uint16_t len, ip_len;
493 	struct inpcbinfo *pcbinfo;
494 	struct sockaddr_in udp_in[2];
495 	struct mbuf *m;
496 	struct m_tag *fwd_tag;
497 	int cscov_partial, iphlen;
498 
499 	m = *mp;
500 	iphlen = *offp;
501 	ifp = m->m_pkthdr.rcvif;
502 	*mp = NULL;
503 	UDPSTAT_INC(udps_ipackets);
504 
505 	/*
506 	 * Strip IP options, if any; should skip this, make available to
507 	 * user, and use on returned packets, but we don't yet have a way to
508 	 * check the checksum with options still present.
509 	 */
510 	if (iphlen > sizeof (struct ip)) {
511 		ip_stripoptions(m);
512 		iphlen = sizeof(struct ip);
513 	}
514 
515 	/*
516 	 * Get IP and UDP header together in first mbuf.
517 	 */
518 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
519 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
520 			UDPSTAT_INC(udps_hdrops);
521 			return (IPPROTO_DONE);
522 		}
523 	}
524 	ip = mtod(m, struct ip *);
525 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
526 	cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
527 
528 	/*
529 	 * Destination port of 0 is illegal, based on RFC768.
530 	 */
531 	if (uh->uh_dport == 0)
532 		goto badunlocked;
533 
534 	/*
535 	 * Construct sockaddr format source address.  Stuff source address
536 	 * and datagram in user buffer.
537 	 */
538 	bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
539 	udp_in[0].sin_len = sizeof(struct sockaddr_in);
540 	udp_in[0].sin_family = AF_INET;
541 	udp_in[0].sin_port = uh->uh_sport;
542 	udp_in[0].sin_addr = ip->ip_src;
543 	udp_in[1].sin_len = sizeof(struct sockaddr_in);
544 	udp_in[1].sin_family = AF_INET;
545 	udp_in[1].sin_port = uh->uh_dport;
546 	udp_in[1].sin_addr = ip->ip_dst;
547 
548 	/*
549 	 * Make mbuf data length reflect UDP length.  If not enough data to
550 	 * reflect UDP length, drop.
551 	 */
552 	len = ntohs((u_short)uh->uh_ulen);
553 	ip_len = ntohs(ip->ip_len) - iphlen;
554 	if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
555 		/* Zero means checksum over the complete packet. */
556 		if (len == 0)
557 			len = ip_len;
558 		cscov_partial = 0;
559 	}
560 	if (ip_len != len) {
561 		if (len > ip_len || len < sizeof(struct udphdr)) {
562 			UDPSTAT_INC(udps_badlen);
563 			goto badunlocked;
564 		}
565 		if (proto == IPPROTO_UDP)
566 			m_adj(m, len - ip_len);
567 	}
568 
569 	/*
570 	 * Checksum extended UDP header and data.
571 	 */
572 	if (uh->uh_sum) {
573 		u_short uh_sum;
574 
575 		if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
576 		    !cscov_partial) {
577 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
578 				uh_sum = m->m_pkthdr.csum_data;
579 			else
580 				uh_sum = in_pseudo(ip->ip_src.s_addr,
581 				    ip->ip_dst.s_addr, htonl((u_short)len +
582 				    m->m_pkthdr.csum_data + proto));
583 			uh_sum ^= 0xffff;
584 		} else {
585 			char b[offsetof(struct ipovly, ih_src)];
586 			struct ipovly *ipov = (struct ipovly *)ip;
587 
588 			bcopy(ipov, b, sizeof(b));
589 			bzero(ipov, sizeof(ipov->ih_x1));
590 			ipov->ih_len = (proto == IPPROTO_UDP) ?
591 			    uh->uh_ulen : htons(ip_len);
592 			uh_sum = in_cksum(m, len + sizeof (struct ip));
593 			bcopy(b, ipov, sizeof(b));
594 		}
595 		if (uh_sum) {
596 			UDPSTAT_INC(udps_badsum);
597 			m_freem(m);
598 			return (IPPROTO_DONE);
599 		}
600 	} else {
601 		if (proto == IPPROTO_UDP) {
602 			UDPSTAT_INC(udps_nosum);
603 		} else {
604 			/* UDPLite requires a checksum */
605 			/* XXX: What is the right UDPLite MIB counter here? */
606 			m_freem(m);
607 			return (IPPROTO_DONE);
608 		}
609 	}
610 
611 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
612 	    in_broadcast(ip->ip_dst, ifp))
613 		return (udp_multi_input(m, proto, udp_in));
614 
615 	pcbinfo = udp_get_inpcbinfo(proto);
616 
617 	/*
618 	 * Locate pcb for datagram.
619 	 *
620 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
621 	 */
622 	if ((m->m_flags & M_IP_NEXTHOP) &&
623 	    (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
624 		struct sockaddr_in *next_hop;
625 
626 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
627 
628 		/*
629 		 * Transparently forwarded. Pretend to be the destination.
630 		 * Already got one like this?
631 		 */
632 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
633 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
634 		if (!inp) {
635 			/*
636 			 * It's new.  Try to find the ambushing socket.
637 			 * Because we've rewritten the destination address,
638 			 * any hardware-generated hash is ignored.
639 			 */
640 			inp = in_pcblookup(pcbinfo, ip->ip_src,
641 			    uh->uh_sport, next_hop->sin_addr,
642 			    next_hop->sin_port ? htons(next_hop->sin_port) :
643 			    uh->uh_dport, INPLOOKUP_WILDCARD |
644 			    INPLOOKUP_RLOCKPCB, ifp);
645 		}
646 		/* Remove the tag from the packet. We don't need it anymore. */
647 		m_tag_delete(m, fwd_tag);
648 		m->m_flags &= ~M_IP_NEXTHOP;
649 	} else
650 		inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
651 		    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
652 		    INPLOOKUP_RLOCKPCB, ifp, m);
653 	if (inp == NULL) {
654 		if (V_udp_log_in_vain) {
655 			char src[INET_ADDRSTRLEN];
656 			char dst[INET_ADDRSTRLEN];
657 
658 			log(LOG_INFO,
659 			    "Connection attempt to UDP %s:%d from %s:%d\n",
660 			    inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
661 			    inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
662 		}
663 		if (proto == IPPROTO_UDPLITE)
664 			UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
665 		else
666 			UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
667 		UDPSTAT_INC(udps_noport);
668 		if (m->m_flags & (M_BCAST | M_MCAST)) {
669 			UDPSTAT_INC(udps_noportbcast);
670 			goto badunlocked;
671 		}
672 		if (V_udp_blackhole && (V_udp_blackhole_local ||
673 		    !in_localip(ip->ip_src)))
674 			goto badunlocked;
675 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
676 			goto badunlocked;
677 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
678 		return (IPPROTO_DONE);
679 	}
680 
681 	/*
682 	 * Check the minimum TTL for socket.
683 	 */
684 	INP_RLOCK_ASSERT(inp);
685 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
686 		if (proto == IPPROTO_UDPLITE)
687 			UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
688 		else
689 			UDP_PROBE(receive, NULL, inp, ip, inp, uh);
690 		INP_RUNLOCK(inp);
691 		m_freem(m);
692 		return (IPPROTO_DONE);
693 	}
694 	if (cscov_partial) {
695 		struct udpcb *up;
696 
697 		up = intoudpcb(inp);
698 		if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
699 			INP_RUNLOCK(inp);
700 			m_freem(m);
701 			return (IPPROTO_DONE);
702 		}
703 	}
704 
705 	if (proto == IPPROTO_UDPLITE)
706 		UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
707 	else
708 		UDP_PROBE(receive, NULL, inp, ip, inp, uh);
709 	if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
710 		INP_RUNLOCK(inp);
711 	return (IPPROTO_DONE);
712 
713 badunlocked:
714 	m_freem(m);
715 	return (IPPROTO_DONE);
716 }
717 #endif /* INET */
718 
719 /*
720  * Notify a udp user of an asynchronous error; just wake up so that they can
721  * collect error status.
722  */
723 struct inpcb *
724 udp_notify(struct inpcb *inp, int errno)
725 {
726 
727 	INP_WLOCK_ASSERT(inp);
728 	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
729 	     errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
730 		NH_FREE(inp->inp_route.ro_nh);
731 		inp->inp_route.ro_nh = (struct nhop_object *)NULL;
732 	}
733 
734 	inp->inp_socket->so_error = errno;
735 	sorwakeup(inp->inp_socket);
736 	sowwakeup(inp->inp_socket);
737 	return (inp);
738 }
739 
740 #ifdef INET
741 static void
742 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo)
743 {
744 	struct ip *ip = &icmp->icmp_ip;
745 	struct udphdr *uh;
746 	struct inpcb *inp;
747 
748 	if (icmp_errmap(icmp) == 0)
749 		return;
750 
751 	uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
752 	inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src,
753 	    uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
754 	if (inp != NULL) {
755 		INP_WLOCK_ASSERT(inp);
756 		if (inp->inp_socket != NULL)
757 			udp_notify(inp, icmp_errmap(icmp));
758 		INP_WUNLOCK(inp);
759 	} else {
760 		inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport,
761 		    ip->ip_src, uh->uh_sport,
762 		    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
763 		if (inp != NULL) {
764 			struct udpcb *up;
765 			udp_tun_icmp_t *func;
766 
767 			up = intoudpcb(inp);
768 			func = up->u_icmp_func;
769 			INP_RUNLOCK(inp);
770 			if (func != NULL)
771 				func(icmp);
772 		}
773 	}
774 }
775 
776 static void
777 udp_ctlinput(struct icmp *icmp)
778 {
779 
780 	return (udp_common_ctlinput(icmp, &V_udbinfo));
781 }
782 
783 static void
784 udplite_ctlinput(struct icmp *icmp)
785 {
786 
787 	return (udp_common_ctlinput(icmp, &V_ulitecbinfo));
788 }
789 #endif /* INET */
790 
791 static int
792 udp_pcblist(SYSCTL_HANDLER_ARGS)
793 {
794 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
795 	    INPLOOKUP_RLOCKPCB);
796 	struct xinpgen xig;
797 	struct inpcb *inp;
798 	int error;
799 
800 	if (req->newptr != 0)
801 		return (EPERM);
802 
803 	if (req->oldptr == 0) {
804 		int n;
805 
806 		n = V_udbinfo.ipi_count;
807 		n += imax(n / 8, 10);
808 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
809 		return (0);
810 	}
811 
812 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
813 		return (error);
814 
815 	bzero(&xig, sizeof(xig));
816 	xig.xig_len = sizeof xig;
817 	xig.xig_count = V_udbinfo.ipi_count;
818 	xig.xig_gen = V_udbinfo.ipi_gencnt;
819 	xig.xig_sogen = so_gencnt;
820 	error = SYSCTL_OUT(req, &xig, sizeof xig);
821 	if (error)
822 		return (error);
823 
824 	while ((inp = inp_next(&inpi)) != NULL) {
825 		if (inp->inp_gencnt <= xig.xig_gen &&
826 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
827 			struct xinpcb xi;
828 
829 			in_pcbtoxinpcb(inp, &xi);
830 			error = SYSCTL_OUT(req, &xi, sizeof xi);
831 			if (error) {
832 				INP_RUNLOCK(inp);
833 				break;
834 			}
835 		}
836 	}
837 
838 	if (!error) {
839 		/*
840 		 * Give the user an updated idea of our state.  If the
841 		 * generation differs from what we told her before, she knows
842 		 * that something happened while we were processing this
843 		 * request, and it might be necessary to retry.
844 		 */
845 		xig.xig_gen = V_udbinfo.ipi_gencnt;
846 		xig.xig_sogen = so_gencnt;
847 		xig.xig_count = V_udbinfo.ipi_count;
848 		error = SYSCTL_OUT(req, &xig, sizeof xig);
849 	}
850 
851 	return (error);
852 }
853 
854 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
855     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
856     udp_pcblist, "S,xinpcb",
857     "List of active UDP sockets");
858 
859 #ifdef INET
860 static int
861 udp_getcred(SYSCTL_HANDLER_ARGS)
862 {
863 	struct xucred xuc;
864 	struct sockaddr_in addrs[2];
865 	struct epoch_tracker et;
866 	struct inpcb *inp;
867 	int error;
868 
869 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
870 	if (error)
871 		return (error);
872 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
873 	if (error)
874 		return (error);
875 	NET_EPOCH_ENTER(et);
876 	inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
877 	    addrs[0].sin_addr, addrs[0].sin_port,
878 	    INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
879 	NET_EPOCH_EXIT(et);
880 	if (inp != NULL) {
881 		INP_RLOCK_ASSERT(inp);
882 		if (inp->inp_socket == NULL)
883 			error = ENOENT;
884 		if (error == 0)
885 			error = cr_canseeinpcb(req->td->td_ucred, inp);
886 		if (error == 0)
887 			cru2x(inp->inp_cred, &xuc);
888 		INP_RUNLOCK(inp);
889 	} else
890 		error = ENOENT;
891 	if (error == 0)
892 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
893 	return (error);
894 }
895 
896 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
897     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
898     0, 0, udp_getcred, "S,xucred",
899     "Get the xucred of a UDP connection");
900 #endif /* INET */
901 
902 int
903 udp_ctloutput(struct socket *so, struct sockopt *sopt)
904 {
905 	struct inpcb *inp;
906 	struct udpcb *up;
907 	int isudplite, error, optval;
908 
909 	error = 0;
910 	isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
911 	inp = sotoinpcb(so);
912 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
913 	INP_WLOCK(inp);
914 	if (sopt->sopt_level != so->so_proto->pr_protocol) {
915 #ifdef INET6
916 		if (INP_CHECK_SOCKAF(so, AF_INET6)) {
917 			INP_WUNLOCK(inp);
918 			error = ip6_ctloutput(so, sopt);
919 		}
920 #endif
921 #if defined(INET) && defined(INET6)
922 		else
923 #endif
924 #ifdef INET
925 		{
926 			INP_WUNLOCK(inp);
927 			error = ip_ctloutput(so, sopt);
928 		}
929 #endif
930 		return (error);
931 	}
932 
933 	switch (sopt->sopt_dir) {
934 	case SOPT_SET:
935 		switch (sopt->sopt_name) {
936 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
937 #ifdef INET
938 		case UDP_ENCAP:
939 			if (!IPSEC_ENABLED(ipv4)) {
940 				INP_WUNLOCK(inp);
941 				return (ENOPROTOOPT);
942 			}
943 			error = UDPENCAP_PCBCTL(inp, sopt);
944 			break;
945 #endif /* INET */
946 #endif /* IPSEC */
947 		case UDPLITE_SEND_CSCOV:
948 		case UDPLITE_RECV_CSCOV:
949 			if (!isudplite) {
950 				INP_WUNLOCK(inp);
951 				error = ENOPROTOOPT;
952 				break;
953 			}
954 			INP_WUNLOCK(inp);
955 			error = sooptcopyin(sopt, &optval, sizeof(optval),
956 			    sizeof(optval));
957 			if (error != 0)
958 				break;
959 			inp = sotoinpcb(so);
960 			KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
961 			INP_WLOCK(inp);
962 			up = intoudpcb(inp);
963 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
964 			if ((optval != 0 && optval < 8) || (optval > 65535)) {
965 				INP_WUNLOCK(inp);
966 				error = EINVAL;
967 				break;
968 			}
969 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
970 				up->u_txcslen = optval;
971 			else
972 				up->u_rxcslen = optval;
973 			INP_WUNLOCK(inp);
974 			break;
975 		default:
976 			INP_WUNLOCK(inp);
977 			error = ENOPROTOOPT;
978 			break;
979 		}
980 		break;
981 	case SOPT_GET:
982 		switch (sopt->sopt_name) {
983 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
984 #ifdef INET
985 		case UDP_ENCAP:
986 			if (!IPSEC_ENABLED(ipv4)) {
987 				INP_WUNLOCK(inp);
988 				return (ENOPROTOOPT);
989 			}
990 			error = UDPENCAP_PCBCTL(inp, sopt);
991 			break;
992 #endif /* INET */
993 #endif /* IPSEC */
994 		case UDPLITE_SEND_CSCOV:
995 		case UDPLITE_RECV_CSCOV:
996 			if (!isudplite) {
997 				INP_WUNLOCK(inp);
998 				error = ENOPROTOOPT;
999 				break;
1000 			}
1001 			up = intoudpcb(inp);
1002 			KASSERT(up != NULL, ("%s: up == NULL", __func__));
1003 			if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1004 				optval = up->u_txcslen;
1005 			else
1006 				optval = up->u_rxcslen;
1007 			INP_WUNLOCK(inp);
1008 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1009 			break;
1010 		default:
1011 			INP_WUNLOCK(inp);
1012 			error = ENOPROTOOPT;
1013 			break;
1014 		}
1015 		break;
1016 	}
1017 	return (error);
1018 }
1019 
1020 #ifdef INET
1021 #ifdef INET6
1022 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1023 static int
1024 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1025     struct inpcb *inp, int flags)
1026 {
1027 	struct ifnet *ifp;
1028 	struct in6_pktinfo *pktinfo;
1029 	struct in_addr ia;
1030 
1031 	if ((flags & PRUS_IPV6) == 0)
1032 		return (0);
1033 
1034 	if (cm->cmsg_level != IPPROTO_IPV6)
1035 		return (0);
1036 
1037 	if  (cm->cmsg_type != IPV6_2292PKTINFO &&
1038 	    cm->cmsg_type != IPV6_PKTINFO)
1039 		return (0);
1040 
1041 	if (cm->cmsg_len !=
1042 	    CMSG_LEN(sizeof(struct in6_pktinfo)))
1043 		return (EINVAL);
1044 
1045 	pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1046 	if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1047 	    !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1048 		return (EINVAL);
1049 
1050 	/* Validate the interface index if specified. */
1051 	if (pktinfo->ipi6_ifindex) {
1052 		struct epoch_tracker et;
1053 
1054 		NET_EPOCH_ENTER(et);
1055 		ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1056 		NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
1057 		if (ifp == NULL)
1058 			return (ENXIO);
1059 	} else
1060 		ifp = NULL;
1061 	if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1062 		ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1063 		if (in_ifhasaddr(ifp, ia) == 0)
1064 			return (EADDRNOTAVAIL);
1065 	}
1066 
1067 	bzero(src, sizeof(*src));
1068 	src->sin_family = AF_INET;
1069 	src->sin_len = sizeof(*src);
1070 	src->sin_port = inp->inp_lport;
1071 	src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1072 
1073 	return (0);
1074 }
1075 #endif
1076 
1077 static int
1078 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1079     struct mbuf *control, struct thread *td, int flags)
1080 {
1081 	struct udpiphdr *ui;
1082 	int len = m->m_pkthdr.len;
1083 	struct in_addr faddr, laddr;
1084 	struct cmsghdr *cm;
1085 	struct inpcbinfo *pcbinfo;
1086 	struct sockaddr_in *sin, src;
1087 	struct epoch_tracker et;
1088 	int cscov_partial = 0;
1089 	int error = 0;
1090 	int ipflags = 0;
1091 	u_short fport, lport;
1092 	u_char tos;
1093 	uint8_t pr;
1094 	uint16_t cscov = 0;
1095 	uint32_t flowid = 0;
1096 	uint8_t flowtype = M_HASHTYPE_NONE;
1097 
1098 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1099 		if (control)
1100 			m_freem(control);
1101 		m_freem(m);
1102 		return (EMSGSIZE);
1103 	}
1104 
1105 	src.sin_family = 0;
1106 	sin = (struct sockaddr_in *)addr;
1107 
1108 	/*
1109 	 * udp_output() may need to temporarily bind or connect the current
1110 	 * inpcb.  As such, we don't know up front whether we will need the
1111 	 * pcbinfo lock or not.  Do any work to decide what is needed up
1112 	 * front before acquiring any locks.
1113 	 *
1114 	 * We will need network epoch in either case, to safely lookup into
1115 	 * pcb hash.
1116 	 */
1117 	if (sin == NULL ||
1118 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0))
1119 		INP_WLOCK(inp);
1120 	else
1121 		INP_RLOCK(inp);
1122 	NET_EPOCH_ENTER(et);
1123 	tos = inp->inp_ip_tos;
1124 	if (control != NULL) {
1125 		/*
1126 		 * XXX: Currently, we assume all the optional information is
1127 		 * stored in a single mbuf.
1128 		 */
1129 		if (control->m_next) {
1130 			m_freem(control);
1131 			error = EINVAL;
1132 			goto release;
1133 		}
1134 		for (; control->m_len > 0;
1135 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
1136 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1137 			cm = mtod(control, struct cmsghdr *);
1138 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1139 			    || cm->cmsg_len > control->m_len) {
1140 				error = EINVAL;
1141 				break;
1142 			}
1143 #ifdef INET6
1144 			error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1145 			if (error != 0)
1146 				break;
1147 #endif
1148 			if (cm->cmsg_level != IPPROTO_IP)
1149 				continue;
1150 
1151 			switch (cm->cmsg_type) {
1152 			case IP_SENDSRCADDR:
1153 				if (cm->cmsg_len !=
1154 				    CMSG_LEN(sizeof(struct in_addr))) {
1155 					error = EINVAL;
1156 					break;
1157 				}
1158 				bzero(&src, sizeof(src));
1159 				src.sin_family = AF_INET;
1160 				src.sin_len = sizeof(src);
1161 				src.sin_port = inp->inp_lport;
1162 				src.sin_addr =
1163 				    *(struct in_addr *)CMSG_DATA(cm);
1164 				break;
1165 
1166 			case IP_TOS:
1167 				if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1168 					error = EINVAL;
1169 					break;
1170 				}
1171 				tos = *(u_char *)CMSG_DATA(cm);
1172 				break;
1173 
1174 			case IP_FLOWID:
1175 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1176 					error = EINVAL;
1177 					break;
1178 				}
1179 				flowid = *(uint32_t *) CMSG_DATA(cm);
1180 				break;
1181 
1182 			case IP_FLOWTYPE:
1183 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1184 					error = EINVAL;
1185 					break;
1186 				}
1187 				flowtype = *(uint32_t *) CMSG_DATA(cm);
1188 				break;
1189 
1190 #ifdef	RSS
1191 			case IP_RSSBUCKETID:
1192 				if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1193 					error = EINVAL;
1194 					break;
1195 				}
1196 				/* This is just a placeholder for now */
1197 				break;
1198 #endif	/* RSS */
1199 			default:
1200 				error = ENOPROTOOPT;
1201 				break;
1202 			}
1203 			if (error)
1204 				break;
1205 		}
1206 		m_freem(control);
1207 		control = NULL;
1208 	}
1209 	if (error)
1210 		goto release;
1211 
1212 	pr = inp->inp_socket->so_proto->pr_protocol;
1213 	pcbinfo = udp_get_inpcbinfo(pr);
1214 
1215 	/*
1216 	 * If the IP_SENDSRCADDR control message was specified, override the
1217 	 * source address for this datagram.  Its use is invalidated if the
1218 	 * address thus specified is incomplete or clobbers other inpcbs.
1219 	 */
1220 	laddr = inp->inp_laddr;
1221 	lport = inp->inp_lport;
1222 	if (src.sin_family == AF_INET) {
1223 		if ((lport == 0) ||
1224 		    (laddr.s_addr == INADDR_ANY &&
1225 		     src.sin_addr.s_addr == INADDR_ANY)) {
1226 			error = EINVAL;
1227 			goto release;
1228 		}
1229 		INP_HASH_WLOCK(pcbinfo);
1230 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1231 		    &laddr.s_addr, &lport, td->td_ucred);
1232 		INP_HASH_WUNLOCK(pcbinfo);
1233 		if (error)
1234 			goto release;
1235 	}
1236 
1237 	/*
1238 	 * If a UDP socket has been connected, then a local address/port will
1239 	 * have been selected and bound.
1240 	 *
1241 	 * If a UDP socket has not been connected to, then an explicit
1242 	 * destination address must be used, in which case a local
1243 	 * address/port may not have been selected and bound.
1244 	 */
1245 	if (sin != NULL) {
1246 		INP_LOCK_ASSERT(inp);
1247 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1248 			error = EISCONN;
1249 			goto release;
1250 		}
1251 
1252 		/*
1253 		 * Jail may rewrite the destination address, so let it do
1254 		 * that before we use it.
1255 		 */
1256 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1257 		if (error)
1258 			goto release;
1259 
1260 		/*
1261 		 * If a local address or port hasn't yet been selected, or if
1262 		 * the destination address needs to be rewritten due to using
1263 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1264 		 * to do the heavy lifting.  Once a port is selected, we
1265 		 * commit the binding back to the socket; we also commit the
1266 		 * binding of the address if in jail.
1267 		 *
1268 		 * If we already have a valid binding and we're not
1269 		 * requesting a destination address rewrite, use a fast path.
1270 		 */
1271 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1272 		    inp->inp_lport == 0 ||
1273 		    sin->sin_addr.s_addr == INADDR_ANY ||
1274 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1275 			INP_HASH_WLOCK(pcbinfo);
1276 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1277 			    &lport, &faddr.s_addr, &fport, NULL,
1278 			    td->td_ucred);
1279 			if (error) {
1280 				INP_HASH_WUNLOCK(pcbinfo);
1281 				goto release;
1282 			}
1283 
1284 			/*
1285 			 * XXXRW: Why not commit the port if the address is
1286 			 * !INADDR_ANY?
1287 			 */
1288 			/* Commit the local port if newly assigned. */
1289 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1290 			    inp->inp_lport == 0) {
1291 				INP_WLOCK_ASSERT(inp);
1292 				/*
1293 				 * Remember addr if jailed, to prevent
1294 				 * rebinding.
1295 				 */
1296 				if (prison_flag(td->td_ucred, PR_IP4))
1297 					inp->inp_laddr = laddr;
1298 				inp->inp_lport = lport;
1299 				error = in_pcbinshash(inp);
1300 				INP_HASH_WUNLOCK(pcbinfo);
1301 				if (error != 0) {
1302 					inp->inp_lport = 0;
1303 					error = EAGAIN;
1304 					goto release;
1305 				}
1306 				inp->inp_flags |= INP_ANONPORT;
1307 			} else
1308 				INP_HASH_WUNLOCK(pcbinfo);
1309 		} else {
1310 			faddr = sin->sin_addr;
1311 			fport = sin->sin_port;
1312 		}
1313 	} else {
1314 		INP_LOCK_ASSERT(inp);
1315 		faddr = inp->inp_faddr;
1316 		fport = inp->inp_fport;
1317 		if (faddr.s_addr == INADDR_ANY) {
1318 			error = ENOTCONN;
1319 			goto release;
1320 		}
1321 	}
1322 
1323 	/*
1324 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1325 	 * link-layer headers.  Immediate slide the data pointer back forward
1326 	 * since we won't use that space at this layer.
1327 	 */
1328 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1329 	if (m == NULL) {
1330 		error = ENOBUFS;
1331 		goto release;
1332 	}
1333 	m->m_data += max_linkhdr;
1334 	m->m_len -= max_linkhdr;
1335 	m->m_pkthdr.len -= max_linkhdr;
1336 
1337 	/*
1338 	 * Fill in mbuf with extended UDP header and addresses and length put
1339 	 * into network format.
1340 	 */
1341 	ui = mtod(m, struct udpiphdr *);
1342 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1343 	ui->ui_v = IPVERSION << 4;
1344 	ui->ui_pr = pr;
1345 	ui->ui_src = laddr;
1346 	ui->ui_dst = faddr;
1347 	ui->ui_sport = lport;
1348 	ui->ui_dport = fport;
1349 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1350 	if (pr == IPPROTO_UDPLITE) {
1351 		struct udpcb *up;
1352 		uint16_t plen;
1353 
1354 		up = intoudpcb(inp);
1355 		cscov = up->u_txcslen;
1356 		plen = (u_short)len + sizeof(struct udphdr);
1357 		if (cscov >= plen)
1358 			cscov = 0;
1359 		ui->ui_len = htons(plen);
1360 		ui->ui_ulen = htons(cscov);
1361 		/*
1362 		 * For UDP-Lite, checksum coverage length of zero means
1363 		 * the entire UDPLite packet is covered by the checksum.
1364 		 */
1365 		cscov_partial = (cscov == 0) ? 0 : 1;
1366 	}
1367 
1368 	/*
1369 	 * Set the Don't Fragment bit in the IP header.
1370 	 */
1371 	if (inp->inp_flags & INP_DONTFRAG) {
1372 		struct ip *ip;
1373 
1374 		ip = (struct ip *)&ui->ui_i;
1375 		ip->ip_off |= htons(IP_DF);
1376 	}
1377 
1378 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1379 		ipflags |= IP_ROUTETOIF;
1380 	if (inp->inp_socket->so_options & SO_BROADCAST)
1381 		ipflags |= IP_ALLOWBROADCAST;
1382 	if (inp->inp_flags & INP_ONESBCAST)
1383 		ipflags |= IP_SENDONES;
1384 
1385 #ifdef MAC
1386 	mac_inpcb_create_mbuf(inp, m);
1387 #endif
1388 
1389 	/*
1390 	 * Set up checksum and output datagram.
1391 	 */
1392 	ui->ui_sum = 0;
1393 	if (pr == IPPROTO_UDPLITE) {
1394 		if (inp->inp_flags & INP_ONESBCAST)
1395 			faddr.s_addr = INADDR_BROADCAST;
1396 		if (cscov_partial) {
1397 			if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1398 				ui->ui_sum = 0xffff;
1399 		} else {
1400 			if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1401 				ui->ui_sum = 0xffff;
1402 		}
1403 	} else if (V_udp_cksum) {
1404 		if (inp->inp_flags & INP_ONESBCAST)
1405 			faddr.s_addr = INADDR_BROADCAST;
1406 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1407 		    htons((u_short)len + sizeof(struct udphdr) + pr));
1408 		m->m_pkthdr.csum_flags = CSUM_UDP;
1409 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1410 	}
1411 	((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1412 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1413 	((struct ip *)ui)->ip_tos = tos;		/* XXX */
1414 	UDPSTAT_INC(udps_opackets);
1415 
1416 	/*
1417 	 * Setup flowid / RSS information for outbound socket.
1418 	 *
1419 	 * Once the UDP code decides to set a flowid some other way,
1420 	 * this allows the flowid to be overridden by userland.
1421 	 */
1422 	if (flowtype != M_HASHTYPE_NONE) {
1423 		m->m_pkthdr.flowid = flowid;
1424 		M_HASHTYPE_SET(m, flowtype);
1425 	}
1426 #if defined(ROUTE_MPATH) || defined(RSS)
1427 	else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1428 		uint32_t hash_val, hash_type;
1429 
1430 		hash_val = fib4_calc_packet_hash(laddr, faddr,
1431 		    lport, fport, pr, &hash_type);
1432 		m->m_pkthdr.flowid = hash_val;
1433 		M_HASHTYPE_SET(m, hash_type);
1434 	}
1435 
1436 	/*
1437 	 * Don't override with the inp cached flowid value.
1438 	 *
1439 	 * Depending upon the kind of send being done, the inp
1440 	 * flowid/flowtype values may actually not be appropriate
1441 	 * for this particular socket send.
1442 	 *
1443 	 * We should either leave the flowid at zero (which is what is
1444 	 * currently done) or set it to some software generated
1445 	 * hash value based on the packet contents.
1446 	 */
1447 	ipflags |= IP_NODEFAULTFLOWID;
1448 #endif	/* RSS */
1449 
1450 	if (pr == IPPROTO_UDPLITE)
1451 		UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1452 	else
1453 		UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1454 	error = ip_output(m, inp->inp_options,
1455 	    INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags,
1456 	    inp->inp_moptions, inp);
1457 	INP_UNLOCK(inp);
1458 	NET_EPOCH_EXIT(et);
1459 	return (error);
1460 
1461 release:
1462 	INP_UNLOCK(inp);
1463 	NET_EPOCH_EXIT(et);
1464 	m_freem(m);
1465 	return (error);
1466 }
1467 
1468 pr_abort_t udp_abort;			/* shared with udp6_usrreq.c */
1469 void
1470 udp_abort(struct socket *so)
1471 {
1472 	struct inpcb *inp;
1473 	struct inpcbinfo *pcbinfo;
1474 
1475 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1476 	inp = sotoinpcb(so);
1477 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1478 	INP_WLOCK(inp);
1479 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1480 		INP_HASH_WLOCK(pcbinfo);
1481 		in_pcbdisconnect(inp);
1482 		inp->inp_laddr.s_addr = INADDR_ANY;
1483 		INP_HASH_WUNLOCK(pcbinfo);
1484 		soisdisconnected(so);
1485 	}
1486 	INP_WUNLOCK(inp);
1487 }
1488 
1489 static int
1490 udp_attach(struct socket *so, int proto, struct thread *td)
1491 {
1492 	static uint32_t udp_flowid;
1493 	struct inpcb *inp;
1494 	struct inpcbinfo *pcbinfo;
1495 	int error;
1496 
1497 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1498 	inp = sotoinpcb(so);
1499 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1500 	error = soreserve(so, udp_sendspace, udp_recvspace);
1501 	if (error)
1502 		return (error);
1503 	error = in_pcballoc(so, pcbinfo);
1504 	if (error)
1505 		return (error);
1506 
1507 	inp = sotoinpcb(so);
1508 	inp->inp_ip_ttl = V_ip_defttl;
1509 	inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1510 	inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1511 
1512 	error = udp_newudpcb(inp);
1513 	if (error) {
1514 		in_pcbdetach(inp);
1515 		in_pcbfree(inp);
1516 		return (error);
1517 	}
1518 	INP_WUNLOCK(inp);
1519 
1520 	return (0);
1521 }
1522 #endif /* INET */
1523 
1524 int
1525 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1526 {
1527 	struct inpcb *inp;
1528 	struct udpcb *up;
1529 
1530 	KASSERT(so->so_type == SOCK_DGRAM,
1531 	    ("udp_set_kernel_tunneling: !dgram"));
1532 	inp = sotoinpcb(so);
1533 	KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1534 	INP_WLOCK(inp);
1535 	up = intoudpcb(inp);
1536 	if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1537 	    (up->u_icmp_func != NULL))) {
1538 		INP_WUNLOCK(inp);
1539 		return (EBUSY);
1540 	}
1541 	up->u_tun_func = f;
1542 	up->u_icmp_func = i;
1543 	up->u_tun_ctx = ctx;
1544 	INP_WUNLOCK(inp);
1545 	return (0);
1546 }
1547 
1548 #ifdef INET
1549 static int
1550 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1551 {
1552 	struct inpcb *inp;
1553 	struct inpcbinfo *pcbinfo;
1554 	struct sockaddr_in *sinp;
1555 	int error;
1556 
1557 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1558 	inp = sotoinpcb(so);
1559 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1560 
1561 	sinp = (struct sockaddr_in *)nam;
1562 	if (nam->sa_family != AF_INET) {
1563 		/*
1564 		 * Preserve compatibility with old programs.
1565 		 */
1566 		if (nam->sa_family != AF_UNSPEC ||
1567 		    nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1568 		    sinp->sin_addr.s_addr != INADDR_ANY)
1569 			return (EAFNOSUPPORT);
1570 		nam->sa_family = AF_INET;
1571 	}
1572 	if (nam->sa_len != sizeof(struct sockaddr_in))
1573 		return (EINVAL);
1574 
1575 	INP_WLOCK(inp);
1576 	INP_HASH_WLOCK(pcbinfo);
1577 	error = in_pcbbind(inp, nam, td->td_ucred);
1578 	INP_HASH_WUNLOCK(pcbinfo);
1579 	INP_WUNLOCK(inp);
1580 	return (error);
1581 }
1582 
1583 static void
1584 udp_close(struct socket *so)
1585 {
1586 	struct inpcb *inp;
1587 	struct inpcbinfo *pcbinfo;
1588 
1589 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1590 	inp = sotoinpcb(so);
1591 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1592 	INP_WLOCK(inp);
1593 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1594 		INP_HASH_WLOCK(pcbinfo);
1595 		in_pcbdisconnect(inp);
1596 		inp->inp_laddr.s_addr = INADDR_ANY;
1597 		INP_HASH_WUNLOCK(pcbinfo);
1598 		soisdisconnected(so);
1599 	}
1600 	INP_WUNLOCK(inp);
1601 }
1602 
1603 static int
1604 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1605 {
1606 	struct epoch_tracker et;
1607 	struct inpcb *inp;
1608 	struct inpcbinfo *pcbinfo;
1609 	struct sockaddr_in *sin;
1610 	int error;
1611 
1612 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1613 	inp = sotoinpcb(so);
1614 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1615 
1616 	sin = (struct sockaddr_in *)nam;
1617 	if (sin->sin_family != AF_INET)
1618 		return (EAFNOSUPPORT);
1619 	if (sin->sin_len != sizeof(*sin))
1620 		return (EINVAL);
1621 
1622 	INP_WLOCK(inp);
1623 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1624 		INP_WUNLOCK(inp);
1625 		return (EISCONN);
1626 	}
1627 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1628 	if (error != 0) {
1629 		INP_WUNLOCK(inp);
1630 		return (error);
1631 	}
1632 	NET_EPOCH_ENTER(et);
1633 	INP_HASH_WLOCK(pcbinfo);
1634 	error = in_pcbconnect(inp, nam, td->td_ucred, true);
1635 	INP_HASH_WUNLOCK(pcbinfo);
1636 	NET_EPOCH_EXIT(et);
1637 	if (error == 0)
1638 		soisconnected(so);
1639 	INP_WUNLOCK(inp);
1640 	return (error);
1641 }
1642 
1643 static void
1644 udp_detach(struct socket *so)
1645 {
1646 	struct inpcb *inp;
1647 	struct udpcb *up;
1648 
1649 	inp = sotoinpcb(so);
1650 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1651 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1652 	    ("udp_detach: not disconnected"));
1653 	INP_WLOCK(inp);
1654 	up = intoudpcb(inp);
1655 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1656 	inp->inp_ppcb = NULL;
1657 	in_pcbdetach(inp);
1658 	in_pcbfree(inp);
1659 	udp_discardcb(up);
1660 }
1661 
1662 pr_disconnect_t udp_disconnect;		/* shared with udp6_usrreq.c */
1663 int
1664 udp_disconnect(struct socket *so)
1665 {
1666 	struct inpcb *inp;
1667 	struct inpcbinfo *pcbinfo;
1668 
1669 	pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1670 	inp = sotoinpcb(so);
1671 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1672 	INP_WLOCK(inp);
1673 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1674 		INP_WUNLOCK(inp);
1675 		return (ENOTCONN);
1676 	}
1677 	INP_HASH_WLOCK(pcbinfo);
1678 	in_pcbdisconnect(inp);
1679 	inp->inp_laddr.s_addr = INADDR_ANY;
1680 	INP_HASH_WUNLOCK(pcbinfo);
1681 	SOCK_LOCK(so);
1682 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1683 	SOCK_UNLOCK(so);
1684 	INP_WUNLOCK(inp);
1685 	return (0);
1686 }
1687 
1688 pr_send_t udp_send;			/* shared with udp6_usrreq.c */
1689 int
1690 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1691     struct mbuf *control, struct thread *td)
1692 {
1693 	struct inpcb *inp;
1694 	int error;
1695 
1696 	inp = sotoinpcb(so);
1697 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1698 
1699 	if (addr != NULL) {
1700 		error = 0;
1701 		if (addr->sa_family != AF_INET)
1702 			error = EAFNOSUPPORT;
1703 		else if (addr->sa_len != sizeof(struct sockaddr_in))
1704 			error = EINVAL;
1705 		if (__predict_false(error != 0)) {
1706 			m_freem(control);
1707 			m_freem(m);
1708 			return (error);
1709 		}
1710 	}
1711 	return (udp_output(inp, m, addr, control, td, flags));
1712 }
1713 #endif /* INET */
1714 
1715 int
1716 udp_shutdown(struct socket *so)
1717 {
1718 	struct inpcb *inp;
1719 
1720 	inp = sotoinpcb(so);
1721 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1722 	INP_WLOCK(inp);
1723 	socantsendmore(so);
1724 	INP_WUNLOCK(inp);
1725 	return (0);
1726 }
1727 
1728 #ifdef INET
1729 #define	UDP_PROTOSW							\
1730 	.pr_type =		SOCK_DGRAM,				\
1731 	.pr_flags =		PR_ATOMIC | PR_ADDR | PR_CAPATTACH,	\
1732 	.pr_ctloutput =		udp_ctloutput,				\
1733 	.pr_abort =		udp_abort,				\
1734 	.pr_attach =		udp_attach,				\
1735 	.pr_bind =		udp_bind,				\
1736 	.pr_connect =		udp_connect,				\
1737 	.pr_control =		in_control,				\
1738 	.pr_detach =		udp_detach,				\
1739 	.pr_disconnect =	udp_disconnect,				\
1740 	.pr_peeraddr =		in_getpeeraddr,				\
1741 	.pr_send =		udp_send,				\
1742 	.pr_soreceive =		soreceive_dgram,			\
1743 	.pr_sosend =		sosend_dgram,				\
1744 	.pr_shutdown =		udp_shutdown,				\
1745 	.pr_sockaddr =		in_getsockaddr,				\
1746 	.pr_sosetlabel =	in_pcbsosetlabel,			\
1747 	.pr_close =		udp_close
1748 
1749 struct protosw udp_protosw = {
1750 	.pr_protocol =		IPPROTO_UDP,
1751 	UDP_PROTOSW
1752 };
1753 
1754 struct protosw udplite_protosw = {
1755 	.pr_protocol =		IPPROTO_UDPLITE,
1756 	UDP_PROTOSW
1757 };
1758 
1759 static void
1760 udp_init(void *arg __unused)
1761 {
1762 
1763 	IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput);
1764 	IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput);
1765 }
1766 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
1767 #endif /* INET */
1768