]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet/udp_usrreq.c
zfs: merge openzfs/zfs@bc3f12bfa (master) into main
[FreeBSD/FreeBSD.git] / sys / netinet / udp_usrreq.c
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, "udpinp", "udp_inpcb", "udp", "udphash");
174 INPCBSTORAGE_DEFINE(udplitecbstor, "udpliteinp", "udplite_inpcb", "udplite",
175     "udplitehash");
176
177 static void
178 udp_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_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
199
200 /*
201  * Kernel module interface for updating udpstat.  The argument is an index
202  * into udpstat treated as an array of u_long.  While this encodes the
203  * general layout of udpstat into the caller, it doesn't encode its location,
204  * so that future changes to add, for example, per-CPU stats support won't
205  * cause binary compatibility problems for kernel modules.
206  */
207 void
208 kmod_udpstat_inc(int statnum)
209 {
210
211         counter_u64_add(VNET(udpstat)[statnum], 1);
212 }
213
214 int
215 udp_newudpcb(struct inpcb *inp)
216 {
217         struct udpcb *up;
218
219         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
220         if (up == NULL)
221                 return (ENOBUFS);
222         inp->inp_ppcb = up;
223         return (0);
224 }
225
226 void
227 udp_discardcb(struct udpcb *up)
228 {
229
230         uma_zfree(V_udpcb_zone, up);
231 }
232
233 #ifdef VIMAGE
234 static void
235 udp_destroy(void *unused __unused)
236 {
237
238         in_pcbinfo_destroy(&V_udbinfo);
239         uma_zdestroy(V_udpcb_zone);
240 }
241 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
242
243 static void
244 udplite_destroy(void *unused __unused)
245 {
246
247         in_pcbinfo_destroy(&V_ulitecbinfo);
248 }
249 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
250     NULL);
251 #endif
252
253 #ifdef INET
254 /*
255  * Subroutine of udp_input(), which appends the provided mbuf chain to the
256  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
257  * contains the source address.  If the socket ends up being an IPv6 socket,
258  * udp_append() will convert to a sockaddr_in6 before passing the address
259  * into the socket code.
260  *
261  * In the normal case udp_append() will return 0, indicating that you
262  * must unlock the inp. However if a tunneling protocol is in place we increment
263  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
264  * then decrement the reference count. If the inp_rele returns 1, indicating the
265  * inp is gone, we return that to the caller to tell them *not* to unlock
266  * the inp. In the case of multi-cast this will cause the distribution
267  * to stop (though most tunneling protocols known currently do *not* use
268  * multicast).
269  */
270 static int
271 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
272     struct sockaddr_in *udp_in)
273 {
274         struct sockaddr *append_sa;
275         struct socket *so;
276         struct mbuf *tmpopts, *opts = NULL;
277 #ifdef INET6
278         struct sockaddr_in6 udp_in6;
279 #endif
280         struct udpcb *up;
281
282         INP_LOCK_ASSERT(inp);
283
284         /*
285          * Engage the tunneling protocol.
286          */
287         up = intoudpcb(inp);
288         if (up->u_tun_func != NULL) {
289                 in_pcbref(inp);
290                 INP_RUNLOCK(inp);
291                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
292                     up->u_tun_ctx);
293                 INP_RLOCK(inp);
294                 return (in_pcbrele_rlocked(inp));
295         }
296
297         off += sizeof(struct udphdr);
298
299 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
300         /* Check AH/ESP integrity. */
301         if (IPSEC_ENABLED(ipv4) &&
302             IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
303                 m_freem(n);
304                 return (0);
305         }
306         if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
307                 if (IPSEC_ENABLED(ipv4) &&
308                     UDPENCAP_INPUT(n, off, AF_INET) != 0)
309                         return (0);     /* Consumed. */
310         }
311 #endif /* IPSEC */
312 #ifdef MAC
313         if (mac_inpcb_check_deliver(inp, n) != 0) {
314                 m_freem(n);
315                 return (0);
316         }
317 #endif /* MAC */
318         if (inp->inp_flags & INP_CONTROLOPTS ||
319             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
320 #ifdef INET6
321                 if (inp->inp_vflag & INP_IPV6)
322                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
323                 else
324 #endif /* INET6 */
325                         ip_savecontrol(inp, &opts, ip, n);
326         }
327         if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
328                 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
329                         sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
330                 if (tmpopts) {
331                         if (opts) {
332                                 tmpopts->m_next = opts;
333                                 opts = tmpopts;
334                         } else
335                                 opts = tmpopts;
336                 }
337         }
338 #ifdef INET6
339         if (inp->inp_vflag & INP_IPV6) {
340                 bzero(&udp_in6, sizeof(udp_in6));
341                 udp_in6.sin6_len = sizeof(udp_in6);
342                 udp_in6.sin6_family = AF_INET6;
343                 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
344                 append_sa = (struct sockaddr *)&udp_in6;
345         } else
346 #endif /* INET6 */
347                 append_sa = (struct sockaddr *)&udp_in[0];
348         m_adj(n, off);
349
350         so = inp->inp_socket;
351         SOCKBUF_LOCK(&so->so_rcv);
352         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
353                 soroverflow_locked(so);
354                 m_freem(n);
355                 if (opts)
356                         m_freem(opts);
357                 UDPSTAT_INC(udps_fullsock);
358         } else
359                 sorwakeup_locked(so);
360         return (0);
361 }
362
363 static bool
364 udp_multi_match(const struct inpcb *inp, void *v)
365 {
366         struct ip *ip = v;
367         struct udphdr *uh = (struct udphdr *)(ip + 1);
368
369         if (inp->inp_lport != uh->uh_dport)
370                 return (false);
371 #ifdef INET6
372         if ((inp->inp_vflag & INP_IPV4) == 0)
373                 return (false);
374 #endif
375         if (inp->inp_laddr.s_addr != INADDR_ANY &&
376             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
377                 return (false);
378         if (inp->inp_faddr.s_addr != INADDR_ANY &&
379             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
380                 return (false);
381         if (inp->inp_fport != 0 &&
382             inp->inp_fport != uh->uh_sport)
383                 return (false);
384
385         return (true);
386 }
387
388 static int
389 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in)
390 {
391         struct ip *ip = mtod(m, struct ip *);
392         struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto),
393             INPLOOKUP_RLOCKPCB, udp_multi_match, ip);
394 #ifdef KDTRACE_HOOKS
395         struct udphdr *uh = (struct udphdr *)(ip + 1);
396 #endif
397         struct inpcb *inp;
398         struct mbuf *n;
399         int appends = 0;
400
401         MPASS(ip->ip_hl == sizeof(struct ip) >> 2);
402
403         while ((inp = inp_next(&inpi)) != NULL) {
404                 /*
405                  * XXXRW: Because we weren't holding either the inpcb
406                  * or the hash lock when we checked for a match
407                  * before, we should probably recheck now that the
408                  * inpcb lock is held.
409                  */
410                 /*
411                  * Handle socket delivery policy for any-source
412                  * and source-specific multicast. [RFC3678]
413                  */
414                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
415                         struct ip_moptions      *imo;
416                         struct sockaddr_in       group;
417                         int                      blocked;
418
419                         imo = inp->inp_moptions;
420                         if (imo == NULL)
421                                 continue;
422                         bzero(&group, sizeof(struct sockaddr_in));
423                         group.sin_len = sizeof(struct sockaddr_in);
424                         group.sin_family = AF_INET;
425                         group.sin_addr = ip->ip_dst;
426
427                         blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif,
428                                 (struct sockaddr *)&group,
429                                 (struct sockaddr *)&udp_in[0]);
430                         if (blocked != MCAST_PASS) {
431                                 if (blocked == MCAST_NOTGMEMBER)
432                                         IPSTAT_INC(ips_notmember);
433                                 if (blocked == MCAST_NOTSMEMBER ||
434                                     blocked == MCAST_MUTED)
435                                         UDPSTAT_INC(udps_filtermcast);
436                                 continue;
437                         }
438                 }
439                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) {
440                         if (proto == IPPROTO_UDPLITE)
441                                 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
442                         else
443                                 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
444                         if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) {
445                                 INP_RUNLOCK(inp);
446                                 break;
447                         } else
448                                 appends++;
449                 }
450                 /*
451                  * Don't look for additional matches if this one does
452                  * not have either the SO_REUSEPORT or SO_REUSEADDR
453                  * socket options set.  This heuristic avoids
454                  * searching through all pcbs in the common case of a
455                  * non-shared port.  It assumes that an application
456                  * will never clear these options after setting them.
457                  */
458                 if ((inp->inp_socket->so_options &
459                     (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) {
460                         INP_RUNLOCK(inp);
461                         break;
462                 }
463         }
464
465         if (appends == 0) {
466                 /*
467                  * No matching pcb found; discard datagram.  (No need
468                  * to send an ICMP Port Unreachable for a broadcast
469                  * or multicast datgram.)
470                  */
471                 UDPSTAT_INC(udps_noport);
472                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
473                         UDPSTAT_INC(udps_noportmcast);
474                 else
475                         UDPSTAT_INC(udps_noportbcast);
476         }
477         m_freem(m);
478
479         return (IPPROTO_DONE);
480 }
481
482 int
483 udp_input(struct mbuf **mp, int *offp, int proto)
484 {
485         struct ip *ip;
486         struct udphdr *uh;
487         struct ifnet *ifp;
488         struct inpcb *inp;
489         uint16_t len, ip_len;
490         struct inpcbinfo *pcbinfo;
491         struct sockaddr_in udp_in[2];
492         struct mbuf *m;
493         struct m_tag *fwd_tag;
494         int cscov_partial, iphlen;
495
496         m = *mp;
497         iphlen = *offp;
498         ifp = m->m_pkthdr.rcvif;
499         *mp = NULL;
500         UDPSTAT_INC(udps_ipackets);
501
502         /*
503          * Strip IP options, if any; should skip this, make available to
504          * user, and use on returned packets, but we don't yet have a way to
505          * check the checksum with options still present.
506          */
507         if (iphlen > sizeof (struct ip)) {
508                 ip_stripoptions(m);
509                 iphlen = sizeof(struct ip);
510         }
511
512         /*
513          * Get IP and UDP header together in first mbuf.
514          */
515         if (m->m_len < iphlen + sizeof(struct udphdr)) {
516                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
517                         UDPSTAT_INC(udps_hdrops);
518                         return (IPPROTO_DONE);
519                 }
520         }
521         ip = mtod(m, struct ip *);
522         uh = (struct udphdr *)((caddr_t)ip + iphlen);
523         cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
524
525         /*
526          * Destination port of 0 is illegal, based on RFC768.
527          */
528         if (uh->uh_dport == 0)
529                 goto badunlocked;
530
531         /*
532          * Construct sockaddr format source address.  Stuff source address
533          * and datagram in user buffer.
534          */
535         bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
536         udp_in[0].sin_len = sizeof(struct sockaddr_in);
537         udp_in[0].sin_family = AF_INET;
538         udp_in[0].sin_port = uh->uh_sport;
539         udp_in[0].sin_addr = ip->ip_src;
540         udp_in[1].sin_len = sizeof(struct sockaddr_in);
541         udp_in[1].sin_family = AF_INET;
542         udp_in[1].sin_port = uh->uh_dport;
543         udp_in[1].sin_addr = ip->ip_dst;
544
545         /*
546          * Make mbuf data length reflect UDP length.  If not enough data to
547          * reflect UDP length, drop.
548          */
549         len = ntohs((u_short)uh->uh_ulen);
550         ip_len = ntohs(ip->ip_len) - iphlen;
551         if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
552                 /* Zero means checksum over the complete packet. */
553                 if (len == 0)
554                         len = ip_len;
555                 cscov_partial = 0;
556         }
557         if (ip_len != len) {
558                 if (len > ip_len || len < sizeof(struct udphdr)) {
559                         UDPSTAT_INC(udps_badlen);
560                         goto badunlocked;
561                 }
562                 if (proto == IPPROTO_UDP)
563                         m_adj(m, len - ip_len);
564         }
565
566         /*
567          * Checksum extended UDP header and data.
568          */
569         if (uh->uh_sum) {
570                 u_short uh_sum;
571
572                 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
573                     !cscov_partial) {
574                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
575                                 uh_sum = m->m_pkthdr.csum_data;
576                         else
577                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
578                                     ip->ip_dst.s_addr, htonl((u_short)len +
579                                     m->m_pkthdr.csum_data + proto));
580                         uh_sum ^= 0xffff;
581                 } else {
582                         char b[offsetof(struct ipovly, ih_src)];
583                         struct ipovly *ipov = (struct ipovly *)ip;
584
585                         bcopy(ipov, b, sizeof(b));
586                         bzero(ipov, sizeof(ipov->ih_x1));
587                         ipov->ih_len = (proto == IPPROTO_UDP) ?
588                             uh->uh_ulen : htons(ip_len);
589                         uh_sum = in_cksum(m, len + sizeof (struct ip));
590                         bcopy(b, ipov, sizeof(b));
591                 }
592                 if (uh_sum) {
593                         UDPSTAT_INC(udps_badsum);
594                         m_freem(m);
595                         return (IPPROTO_DONE);
596                 }
597         } else {
598                 if (proto == IPPROTO_UDP) {
599                         UDPSTAT_INC(udps_nosum);
600                 } else {
601                         /* UDPLite requires a checksum */
602                         /* XXX: What is the right UDPLite MIB counter here? */
603                         m_freem(m);
604                         return (IPPROTO_DONE);
605                 }
606         }
607
608         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
609             in_broadcast(ip->ip_dst, ifp))
610                 return (udp_multi_input(m, proto, udp_in));
611
612         pcbinfo = udp_get_inpcbinfo(proto);
613
614         /*
615          * Locate pcb for datagram.
616          *
617          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
618          */
619         if ((m->m_flags & M_IP_NEXTHOP) &&
620             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
621                 struct sockaddr_in *next_hop;
622
623                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
624
625                 /*
626                  * Transparently forwarded. Pretend to be the destination.
627                  * Already got one like this?
628                  */
629                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
630                     ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
631                 if (!inp) {
632                         /*
633                          * It's new.  Try to find the ambushing socket.
634                          * Because we've rewritten the destination address,
635                          * any hardware-generated hash is ignored.
636                          */
637                         inp = in_pcblookup(pcbinfo, ip->ip_src,
638                             uh->uh_sport, next_hop->sin_addr,
639                             next_hop->sin_port ? htons(next_hop->sin_port) :
640                             uh->uh_dport, INPLOOKUP_WILDCARD |
641                             INPLOOKUP_RLOCKPCB, ifp);
642                 }
643                 /* Remove the tag from the packet. We don't need it anymore. */
644                 m_tag_delete(m, fwd_tag);
645                 m->m_flags &= ~M_IP_NEXTHOP;
646         } else
647                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
648                     ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
649                     INPLOOKUP_RLOCKPCB, ifp, m);
650         if (inp == NULL) {
651                 if (V_udp_log_in_vain) {
652                         char src[INET_ADDRSTRLEN];
653                         char dst[INET_ADDRSTRLEN];
654
655                         log(LOG_INFO,
656                             "Connection attempt to UDP %s:%d from %s:%d\n",
657                             inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
658                             inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
659                 }
660                 if (proto == IPPROTO_UDPLITE)
661                         UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
662                 else
663                         UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
664                 UDPSTAT_INC(udps_noport);
665                 if (m->m_flags & (M_BCAST | M_MCAST)) {
666                         UDPSTAT_INC(udps_noportbcast);
667                         goto badunlocked;
668                 }
669                 if (V_udp_blackhole && (V_udp_blackhole_local ||
670                     !in_localip(ip->ip_src)))
671                         goto badunlocked;
672                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
673                         goto badunlocked;
674                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
675                 return (IPPROTO_DONE);
676         }
677
678         /*
679          * Check the minimum TTL for socket.
680          */
681         INP_RLOCK_ASSERT(inp);
682         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
683                 if (proto == IPPROTO_UDPLITE)
684                         UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
685                 else
686                         UDP_PROBE(receive, NULL, inp, ip, inp, uh);
687                 INP_RUNLOCK(inp);
688                 m_freem(m);
689                 return (IPPROTO_DONE);
690         }
691         if (cscov_partial) {
692                 struct udpcb *up;
693
694                 up = intoudpcb(inp);
695                 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
696                         INP_RUNLOCK(inp);
697                         m_freem(m);
698                         return (IPPROTO_DONE);
699                 }
700         }
701
702         if (proto == IPPROTO_UDPLITE)
703                 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
704         else
705                 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
706         if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
707                 INP_RUNLOCK(inp);
708         return (IPPROTO_DONE);
709
710 badunlocked:
711         m_freem(m);
712         return (IPPROTO_DONE);
713 }
714 #endif /* INET */
715
716 /*
717  * Notify a udp user of an asynchronous error; just wake up so that they can
718  * collect error status.
719  */
720 struct inpcb *
721 udp_notify(struct inpcb *inp, int errno)
722 {
723
724         INP_WLOCK_ASSERT(inp);
725         if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
726              errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
727                 NH_FREE(inp->inp_route.ro_nh);
728                 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
729         }
730
731         inp->inp_socket->so_error = errno;
732         sorwakeup(inp->inp_socket);
733         sowwakeup(inp->inp_socket);
734         return (inp);
735 }
736
737 #ifdef INET
738 static void
739 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
740     struct inpcbinfo *pcbinfo)
741 {
742         struct ip *ip = vip;
743         struct udphdr *uh;
744         struct in_addr faddr;
745         struct inpcb *inp;
746
747         faddr = ((struct sockaddr_in *)sa)->sin_addr;
748         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
749                 return;
750
751         if (PRC_IS_REDIRECT(cmd)) {
752                 /* signal EHOSTDOWN, as it flushes the cached route */
753                 in_pcbnotifyall(pcbinfo, faddr, EHOSTDOWN, udp_notify);
754                 return;
755         }
756
757         /*
758          * Hostdead is ugly because it goes linearly through all PCBs.
759          *
760          * XXX: We never get this from ICMP, otherwise it makes an excellent
761          * DoS attack on machines with many connections.
762          */
763         if (cmd == PRC_HOSTDEAD)
764                 ip = NULL;
765         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
766                 return;
767         if (ip != NULL) {
768                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
769                 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
770                     ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
771                 if (inp != NULL) {
772                         INP_WLOCK_ASSERT(inp);
773                         if (inp->inp_socket != NULL) {
774                                 udp_notify(inp, inetctlerrmap[cmd]);
775                         }
776                         INP_WUNLOCK(inp);
777                 } else {
778                         inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
779                                            ip->ip_src, uh->uh_sport,
780                                            INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
781                         if (inp != NULL) {
782                                 struct udpcb *up;
783                                 void *ctx;
784                                 udp_tun_icmp_t func;
785
786                                 up = intoudpcb(inp);
787                                 ctx = up->u_tun_ctx;
788                                 func = up->u_icmp_func;
789                                 INP_RUNLOCK(inp);
790                                 if (func != NULL)
791                                         (*func)(cmd, sa, vip, ctx);
792                         }
793                 }
794         } else
795                 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
796                     udp_notify);
797 }
798 void
799 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
800 {
801
802         return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
803 }
804
805 void
806 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
807 {
808
809         return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
810 }
811 #endif /* INET */
812
813 static int
814 udp_pcblist(SYSCTL_HANDLER_ARGS)
815 {
816         struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
817             INPLOOKUP_RLOCKPCB);
818         struct xinpgen xig;
819         struct inpcb *inp;
820         int error;
821
822         if (req->newptr != 0)
823                 return (EPERM);
824
825         if (req->oldptr == 0) {
826                 int n;
827
828                 n = V_udbinfo.ipi_count;
829                 n += imax(n / 8, 10);
830                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
831                 return (0);
832         }
833
834         if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
835                 return (error);
836
837         bzero(&xig, sizeof(xig));
838         xig.xig_len = sizeof xig;
839         xig.xig_count = V_udbinfo.ipi_count;
840         xig.xig_gen = V_udbinfo.ipi_gencnt;
841         xig.xig_sogen = so_gencnt;
842         error = SYSCTL_OUT(req, &xig, sizeof xig);
843         if (error)
844                 return (error);
845
846         while ((inp = inp_next(&inpi)) != NULL) {
847                 if (inp->inp_gencnt <= xig.xig_gen &&
848                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
849                         struct xinpcb xi;
850
851                         in_pcbtoxinpcb(inp, &xi);
852                         error = SYSCTL_OUT(req, &xi, sizeof xi);
853                         if (error) {
854                                 INP_RUNLOCK(inp);
855                                 break;
856                         }
857                 }
858         }
859
860         if (!error) {
861                 /*
862                  * Give the user an updated idea of our state.  If the
863                  * generation differs from what we told her before, she knows
864                  * that something happened while we were processing this
865                  * request, and it might be necessary to retry.
866                  */
867                 xig.xig_gen = V_udbinfo.ipi_gencnt;
868                 xig.xig_sogen = so_gencnt;
869                 xig.xig_count = V_udbinfo.ipi_count;
870                 error = SYSCTL_OUT(req, &xig, sizeof xig);
871         }
872
873         return (error);
874 }
875
876 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
877     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
878     udp_pcblist, "S,xinpcb",
879     "List of active UDP sockets");
880
881 #ifdef INET
882 static int
883 udp_getcred(SYSCTL_HANDLER_ARGS)
884 {
885         struct xucred xuc;
886         struct sockaddr_in addrs[2];
887         struct epoch_tracker et;
888         struct inpcb *inp;
889         int error;
890
891         error = priv_check(req->td, PRIV_NETINET_GETCRED);
892         if (error)
893                 return (error);
894         error = SYSCTL_IN(req, addrs, sizeof(addrs));
895         if (error)
896                 return (error);
897         NET_EPOCH_ENTER(et);
898         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
899             addrs[0].sin_addr, addrs[0].sin_port,
900             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
901         NET_EPOCH_EXIT(et);
902         if (inp != NULL) {
903                 INP_RLOCK_ASSERT(inp);
904                 if (inp->inp_socket == NULL)
905                         error = ENOENT;
906                 if (error == 0)
907                         error = cr_canseeinpcb(req->td->td_ucred, inp);
908                 if (error == 0)
909                         cru2x(inp->inp_cred, &xuc);
910                 INP_RUNLOCK(inp);
911         } else
912                 error = ENOENT;
913         if (error == 0)
914                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
915         return (error);
916 }
917
918 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
919     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
920     0, 0, udp_getcred, "S,xucred",
921     "Get the xucred of a UDP connection");
922 #endif /* INET */
923
924 int
925 udp_ctloutput(struct socket *so, struct sockopt *sopt)
926 {
927         struct inpcb *inp;
928         struct udpcb *up;
929         int isudplite, error, optval;
930
931         error = 0;
932         isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
933         inp = sotoinpcb(so);
934         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
935         INP_WLOCK(inp);
936         if (sopt->sopt_level != so->so_proto->pr_protocol) {
937 #ifdef INET6
938                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
939                         INP_WUNLOCK(inp);
940                         error = ip6_ctloutput(so, sopt);
941                 }
942 #endif
943 #if defined(INET) && defined(INET6)
944                 else
945 #endif
946 #ifdef INET
947                 {
948                         INP_WUNLOCK(inp);
949                         error = ip_ctloutput(so, sopt);
950                 }
951 #endif
952                 return (error);
953         }
954
955         switch (sopt->sopt_dir) {
956         case SOPT_SET:
957                 switch (sopt->sopt_name) {
958 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
959 #ifdef INET
960                 case UDP_ENCAP:
961                         if (!IPSEC_ENABLED(ipv4)) {
962                                 INP_WUNLOCK(inp);
963                                 return (ENOPROTOOPT);
964                         }
965                         error = UDPENCAP_PCBCTL(inp, sopt);
966                         break;
967 #endif /* INET */
968 #endif /* IPSEC */
969                 case UDPLITE_SEND_CSCOV:
970                 case UDPLITE_RECV_CSCOV:
971                         if (!isudplite) {
972                                 INP_WUNLOCK(inp);
973                                 error = ENOPROTOOPT;
974                                 break;
975                         }
976                         INP_WUNLOCK(inp);
977                         error = sooptcopyin(sopt, &optval, sizeof(optval),
978                             sizeof(optval));
979                         if (error != 0)
980                                 break;
981                         inp = sotoinpcb(so);
982                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
983                         INP_WLOCK(inp);
984                         up = intoudpcb(inp);
985                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
986                         if ((optval != 0 && optval < 8) || (optval > 65535)) {
987                                 INP_WUNLOCK(inp);
988                                 error = EINVAL;
989                                 break;
990                         }
991                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
992                                 up->u_txcslen = optval;
993                         else
994                                 up->u_rxcslen = optval;
995                         INP_WUNLOCK(inp);
996                         break;
997                 default:
998                         INP_WUNLOCK(inp);
999                         error = ENOPROTOOPT;
1000                         break;
1001                 }
1002                 break;
1003         case SOPT_GET:
1004                 switch (sopt->sopt_name) {
1005 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1006 #ifdef INET
1007                 case UDP_ENCAP:
1008                         if (!IPSEC_ENABLED(ipv4)) {
1009                                 INP_WUNLOCK(inp);
1010                                 return (ENOPROTOOPT);
1011                         }
1012                         error = UDPENCAP_PCBCTL(inp, sopt);
1013                         break;
1014 #endif /* INET */
1015 #endif /* IPSEC */
1016                 case UDPLITE_SEND_CSCOV:
1017                 case UDPLITE_RECV_CSCOV:
1018                         if (!isudplite) {
1019                                 INP_WUNLOCK(inp);
1020                                 error = ENOPROTOOPT;
1021                                 break;
1022                         }
1023                         up = intoudpcb(inp);
1024                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1025                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1026                                 optval = up->u_txcslen;
1027                         else
1028                                 optval = up->u_rxcslen;
1029                         INP_WUNLOCK(inp);
1030                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1031                         break;
1032                 default:
1033                         INP_WUNLOCK(inp);
1034                         error = ENOPROTOOPT;
1035                         break;
1036                 }
1037                 break;
1038         }
1039         return (error);
1040 }
1041
1042 #ifdef INET
1043 #ifdef INET6
1044 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1045 static int
1046 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1047     struct inpcb *inp, int flags)
1048 {
1049         struct ifnet *ifp;
1050         struct in6_pktinfo *pktinfo;
1051         struct in_addr ia;
1052
1053         if ((flags & PRUS_IPV6) == 0)
1054                 return (0);
1055
1056         if (cm->cmsg_level != IPPROTO_IPV6)
1057                 return (0);
1058
1059         if  (cm->cmsg_type != IPV6_2292PKTINFO &&
1060             cm->cmsg_type != IPV6_PKTINFO)
1061                 return (0);
1062
1063         if (cm->cmsg_len !=
1064             CMSG_LEN(sizeof(struct in6_pktinfo)))
1065                 return (EINVAL);
1066
1067         pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1068         if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1069             !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1070                 return (EINVAL);
1071
1072         /* Validate the interface index if specified. */
1073         if (pktinfo->ipi6_ifindex) {
1074                 struct epoch_tracker et;
1075
1076                 NET_EPOCH_ENTER(et);
1077                 ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1078                 NET_EPOCH_EXIT(et);     /* XXXGL: unsafe ifp */
1079                 if (ifp == NULL)
1080                         return (ENXIO);
1081         } else
1082                 ifp = NULL;
1083         if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1084                 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1085                 if (in_ifhasaddr(ifp, ia) == 0)
1086                         return (EADDRNOTAVAIL);
1087         }
1088
1089         bzero(src, sizeof(*src));
1090         src->sin_family = AF_INET;
1091         src->sin_len = sizeof(*src);
1092         src->sin_port = inp->inp_lport;
1093         src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1094
1095         return (0);
1096 }
1097 #endif
1098
1099 static int
1100 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1101     struct mbuf *control, struct thread *td, int flags)
1102 {
1103         struct udpiphdr *ui;
1104         int len = m->m_pkthdr.len;
1105         struct in_addr faddr, laddr;
1106         struct cmsghdr *cm;
1107         struct inpcbinfo *pcbinfo;
1108         struct sockaddr_in *sin, src;
1109         struct epoch_tracker et;
1110         int cscov_partial = 0;
1111         int error = 0;
1112         int ipflags = 0;
1113         u_short fport, lport;
1114         u_char tos;
1115         uint8_t pr;
1116         uint16_t cscov = 0;
1117         uint32_t flowid = 0;
1118         uint8_t flowtype = M_HASHTYPE_NONE;
1119
1120         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1121                 if (control)
1122                         m_freem(control);
1123                 m_freem(m);
1124                 return (EMSGSIZE);
1125         }
1126
1127         src.sin_family = 0;
1128         sin = (struct sockaddr_in *)addr;
1129
1130         /*
1131          * udp_output() may need to temporarily bind or connect the current
1132          * inpcb.  As such, we don't know up front whether we will need the
1133          * pcbinfo lock or not.  Do any work to decide what is needed up
1134          * front before acquiring any locks.
1135          *
1136          * We will need network epoch in either case, to safely lookup into
1137          * pcb hash.
1138          */
1139         if (sin == NULL ||
1140             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0))
1141                 INP_WLOCK(inp);
1142         else
1143                 INP_RLOCK(inp);
1144         NET_EPOCH_ENTER(et);
1145         tos = inp->inp_ip_tos;
1146         if (control != NULL) {
1147                 /*
1148                  * XXX: Currently, we assume all the optional information is
1149                  * stored in a single mbuf.
1150                  */
1151                 if (control->m_next) {
1152                         m_freem(control);
1153                         error = EINVAL;
1154                         goto release;
1155                 }
1156                 for (; control->m_len > 0;
1157                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1158                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1159                         cm = mtod(control, struct cmsghdr *);
1160                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1161                             || cm->cmsg_len > control->m_len) {
1162                                 error = EINVAL;
1163                                 break;
1164                         }
1165 #ifdef INET6
1166                         error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1167                         if (error != 0)
1168                                 break;
1169 #endif
1170                         if (cm->cmsg_level != IPPROTO_IP)
1171                                 continue;
1172
1173                         switch (cm->cmsg_type) {
1174                         case IP_SENDSRCADDR:
1175                                 if (cm->cmsg_len !=
1176                                     CMSG_LEN(sizeof(struct in_addr))) {
1177                                         error = EINVAL;
1178                                         break;
1179                                 }
1180                                 bzero(&src, sizeof(src));
1181                                 src.sin_family = AF_INET;
1182                                 src.sin_len = sizeof(src);
1183                                 src.sin_port = inp->inp_lport;
1184                                 src.sin_addr =
1185                                     *(struct in_addr *)CMSG_DATA(cm);
1186                                 break;
1187
1188                         case IP_TOS:
1189                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1190                                         error = EINVAL;
1191                                         break;
1192                                 }
1193                                 tos = *(u_char *)CMSG_DATA(cm);
1194                                 break;
1195
1196                         case IP_FLOWID:
1197                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1198                                         error = EINVAL;
1199                                         break;
1200                                 }
1201                                 flowid = *(uint32_t *) CMSG_DATA(cm);
1202                                 break;
1203
1204                         case IP_FLOWTYPE:
1205                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1206                                         error = EINVAL;
1207                                         break;
1208                                 }
1209                                 flowtype = *(uint32_t *) CMSG_DATA(cm);
1210                                 break;
1211
1212 #ifdef  RSS
1213                         case IP_RSSBUCKETID:
1214                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1215                                         error = EINVAL;
1216                                         break;
1217                                 }
1218                                 /* This is just a placeholder for now */
1219                                 break;
1220 #endif  /* RSS */
1221                         default:
1222                                 error = ENOPROTOOPT;
1223                                 break;
1224                         }
1225                         if (error)
1226                                 break;
1227                 }
1228                 m_freem(control);
1229                 control = NULL;
1230         }
1231         if (error)
1232                 goto release;
1233
1234         pr = inp->inp_socket->so_proto->pr_protocol;
1235         pcbinfo = udp_get_inpcbinfo(pr);
1236
1237         /*
1238          * If the IP_SENDSRCADDR control message was specified, override the
1239          * source address for this datagram.  Its use is invalidated if the
1240          * address thus specified is incomplete or clobbers other inpcbs.
1241          */
1242         laddr = inp->inp_laddr;
1243         lport = inp->inp_lport;
1244         if (src.sin_family == AF_INET) {
1245                 if ((lport == 0) ||
1246                     (laddr.s_addr == INADDR_ANY &&
1247                      src.sin_addr.s_addr == INADDR_ANY)) {
1248                         error = EINVAL;
1249                         goto release;
1250                 }
1251                 INP_HASH_WLOCK(pcbinfo);
1252                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1253                     &laddr.s_addr, &lport, td->td_ucred);
1254                 INP_HASH_WUNLOCK(pcbinfo);
1255                 if (error)
1256                         goto release;
1257         }
1258
1259         /*
1260          * If a UDP socket has been connected, then a local address/port will
1261          * have been selected and bound.
1262          *
1263          * If a UDP socket has not been connected to, then an explicit
1264          * destination address must be used, in which case a local
1265          * address/port may not have been selected and bound.
1266          */
1267         if (sin != NULL) {
1268                 INP_LOCK_ASSERT(inp);
1269                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1270                         error = EISCONN;
1271                         goto release;
1272                 }
1273
1274                 /*
1275                  * Jail may rewrite the destination address, so let it do
1276                  * that before we use it.
1277                  */
1278                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1279                 if (error)
1280                         goto release;
1281
1282                 /*
1283                  * If a local address or port hasn't yet been selected, or if
1284                  * the destination address needs to be rewritten due to using
1285                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1286                  * to do the heavy lifting.  Once a port is selected, we
1287                  * commit the binding back to the socket; we also commit the
1288                  * binding of the address if in jail.
1289                  *
1290                  * If we already have a valid binding and we're not
1291                  * requesting a destination address rewrite, use a fast path.
1292                  */
1293                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1294                     inp->inp_lport == 0 ||
1295                     sin->sin_addr.s_addr == INADDR_ANY ||
1296                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1297                         INP_HASH_WLOCK(pcbinfo);
1298                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1299                             &lport, &faddr.s_addr, &fport, NULL,
1300                             td->td_ucred);
1301                         if (error) {
1302                                 INP_HASH_WUNLOCK(pcbinfo);
1303                                 goto release;
1304                         }
1305
1306                         /*
1307                          * XXXRW: Why not commit the port if the address is
1308                          * !INADDR_ANY?
1309                          */
1310                         /* Commit the local port if newly assigned. */
1311                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1312                             inp->inp_lport == 0) {
1313                                 INP_WLOCK_ASSERT(inp);
1314                                 /*
1315                                  * Remember addr if jailed, to prevent
1316                                  * rebinding.
1317                                  */
1318                                 if (prison_flag(td->td_ucred, PR_IP4))
1319                                         inp->inp_laddr = laddr;
1320                                 inp->inp_lport = lport;
1321                                 error = in_pcbinshash(inp);
1322                                 INP_HASH_WUNLOCK(pcbinfo);
1323                                 if (error != 0) {
1324                                         inp->inp_lport = 0;
1325                                         error = EAGAIN;
1326                                         goto release;
1327                                 }
1328                                 inp->inp_flags |= INP_ANONPORT;
1329                         } else
1330                                 INP_HASH_WUNLOCK(pcbinfo);
1331                 } else {
1332                         faddr = sin->sin_addr;
1333                         fport = sin->sin_port;
1334                 }
1335         } else {
1336                 INP_LOCK_ASSERT(inp);
1337                 faddr = inp->inp_faddr;
1338                 fport = inp->inp_fport;
1339                 if (faddr.s_addr == INADDR_ANY) {
1340                         error = ENOTCONN;
1341                         goto release;
1342                 }
1343         }
1344
1345         /*
1346          * Calculate data length and get a mbuf for UDP, IP, and possible
1347          * link-layer headers.  Immediate slide the data pointer back forward
1348          * since we won't use that space at this layer.
1349          */
1350         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1351         if (m == NULL) {
1352                 error = ENOBUFS;
1353                 goto release;
1354         }
1355         m->m_data += max_linkhdr;
1356         m->m_len -= max_linkhdr;
1357         m->m_pkthdr.len -= max_linkhdr;
1358
1359         /*
1360          * Fill in mbuf with extended UDP header and addresses and length put
1361          * into network format.
1362          */
1363         ui = mtod(m, struct udpiphdr *);
1364         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1365         ui->ui_v = IPVERSION << 4;
1366         ui->ui_pr = pr;
1367         ui->ui_src = laddr;
1368         ui->ui_dst = faddr;
1369         ui->ui_sport = lport;
1370         ui->ui_dport = fport;
1371         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1372         if (pr == IPPROTO_UDPLITE) {
1373                 struct udpcb *up;
1374                 uint16_t plen;
1375
1376                 up = intoudpcb(inp);
1377                 cscov = up->u_txcslen;
1378                 plen = (u_short)len + sizeof(struct udphdr);
1379                 if (cscov >= plen)
1380                         cscov = 0;
1381                 ui->ui_len = htons(plen);
1382                 ui->ui_ulen = htons(cscov);
1383                 /*
1384                  * For UDP-Lite, checksum coverage length of zero means
1385                  * the entire UDPLite packet is covered by the checksum.
1386                  */
1387                 cscov_partial = (cscov == 0) ? 0 : 1;
1388         }
1389
1390         /*
1391          * Set the Don't Fragment bit in the IP header.
1392          */
1393         if (inp->inp_flags & INP_DONTFRAG) {
1394                 struct ip *ip;
1395
1396                 ip = (struct ip *)&ui->ui_i;
1397                 ip->ip_off |= htons(IP_DF);
1398         }
1399
1400         if (inp->inp_socket->so_options & SO_DONTROUTE)
1401                 ipflags |= IP_ROUTETOIF;
1402         if (inp->inp_socket->so_options & SO_BROADCAST)
1403                 ipflags |= IP_ALLOWBROADCAST;
1404         if (inp->inp_flags & INP_ONESBCAST)
1405                 ipflags |= IP_SENDONES;
1406
1407 #ifdef MAC
1408         mac_inpcb_create_mbuf(inp, m);
1409 #endif
1410
1411         /*
1412          * Set up checksum and output datagram.
1413          */
1414         ui->ui_sum = 0;
1415         if (pr == IPPROTO_UDPLITE) {
1416                 if (inp->inp_flags & INP_ONESBCAST)
1417                         faddr.s_addr = INADDR_BROADCAST;
1418                 if (cscov_partial) {
1419                         if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1420                                 ui->ui_sum = 0xffff;
1421                 } else {
1422                         if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1423                                 ui->ui_sum = 0xffff;
1424                 }
1425         } else if (V_udp_cksum) {
1426                 if (inp->inp_flags & INP_ONESBCAST)
1427                         faddr.s_addr = INADDR_BROADCAST;
1428                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1429                     htons((u_short)len + sizeof(struct udphdr) + pr));
1430                 m->m_pkthdr.csum_flags = CSUM_UDP;
1431                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1432         }
1433         ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1434         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1435         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1436         UDPSTAT_INC(udps_opackets);
1437
1438         /*
1439          * Setup flowid / RSS information for outbound socket.
1440          *
1441          * Once the UDP code decides to set a flowid some other way,
1442          * this allows the flowid to be overridden by userland.
1443          */
1444         if (flowtype != M_HASHTYPE_NONE) {
1445                 m->m_pkthdr.flowid = flowid;
1446                 M_HASHTYPE_SET(m, flowtype);
1447         }
1448 #if defined(ROUTE_MPATH) || defined(RSS)
1449         else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1450                 uint32_t hash_val, hash_type;
1451
1452                 hash_val = fib4_calc_packet_hash(laddr, faddr,
1453                     lport, fport, pr, &hash_type);
1454                 m->m_pkthdr.flowid = hash_val;
1455                 M_HASHTYPE_SET(m, hash_type);
1456         }
1457
1458         /*
1459          * Don't override with the inp cached flowid value.
1460          *
1461          * Depending upon the kind of send being done, the inp
1462          * flowid/flowtype values may actually not be appropriate
1463          * for this particular socket send.
1464          *
1465          * We should either leave the flowid at zero (which is what is
1466          * currently done) or set it to some software generated
1467          * hash value based on the packet contents.
1468          */
1469         ipflags |= IP_NODEFAULTFLOWID;
1470 #endif  /* RSS */
1471
1472         if (pr == IPPROTO_UDPLITE)
1473                 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1474         else
1475                 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1476         error = ip_output(m, inp->inp_options,
1477             INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags,
1478             inp->inp_moptions, inp);
1479         INP_UNLOCK(inp);
1480         NET_EPOCH_EXIT(et);
1481         return (error);
1482
1483 release:
1484         INP_UNLOCK(inp);
1485         NET_EPOCH_EXIT(et);
1486         m_freem(m);
1487         return (error);
1488 }
1489
1490 static void
1491 udp_abort(struct socket *so)
1492 {
1493         struct inpcb *inp;
1494         struct inpcbinfo *pcbinfo;
1495
1496         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1497         inp = sotoinpcb(so);
1498         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1499         INP_WLOCK(inp);
1500         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1501                 INP_HASH_WLOCK(pcbinfo);
1502                 in_pcbdisconnect(inp);
1503                 inp->inp_laddr.s_addr = INADDR_ANY;
1504                 INP_HASH_WUNLOCK(pcbinfo);
1505                 soisdisconnected(so);
1506         }
1507         INP_WUNLOCK(inp);
1508 }
1509
1510 static int
1511 udp_attach(struct socket *so, int proto, struct thread *td)
1512 {
1513         static uint32_t udp_flowid;
1514         struct inpcb *inp;
1515         struct inpcbinfo *pcbinfo;
1516         int error;
1517
1518         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1519         inp = sotoinpcb(so);
1520         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1521         error = soreserve(so, udp_sendspace, udp_recvspace);
1522         if (error)
1523                 return (error);
1524         error = in_pcballoc(so, pcbinfo);
1525         if (error)
1526                 return (error);
1527
1528         inp = sotoinpcb(so);
1529         inp->inp_vflag |= INP_IPV4;
1530         inp->inp_ip_ttl = V_ip_defttl;
1531         inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1532         inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1533
1534         error = udp_newudpcb(inp);
1535         if (error) {
1536                 in_pcbdetach(inp);
1537                 in_pcbfree(inp);
1538                 return (error);
1539         }
1540         INP_WUNLOCK(inp);
1541
1542         return (0);
1543 }
1544 #endif /* INET */
1545
1546 int
1547 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1548 {
1549         struct inpcb *inp;
1550         struct udpcb *up;
1551
1552         KASSERT(so->so_type == SOCK_DGRAM,
1553             ("udp_set_kernel_tunneling: !dgram"));
1554         inp = sotoinpcb(so);
1555         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1556         INP_WLOCK(inp);
1557         up = intoudpcb(inp);
1558         if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1559             (up->u_icmp_func != NULL))) {
1560                 INP_WUNLOCK(inp);
1561                 return (EBUSY);
1562         }
1563         up->u_tun_func = f;
1564         up->u_icmp_func = i;
1565         up->u_tun_ctx = ctx;
1566         INP_WUNLOCK(inp);
1567         return (0);
1568 }
1569
1570 #ifdef INET
1571 static int
1572 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1573 {
1574         struct inpcb *inp;
1575         struct inpcbinfo *pcbinfo;
1576         struct sockaddr_in *sinp;
1577         int error;
1578
1579         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1580         inp = sotoinpcb(so);
1581         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1582
1583         sinp = (struct sockaddr_in *)nam;
1584         if (nam->sa_family != AF_INET) {
1585                 /*
1586                  * Preserve compatibility with old programs.
1587                  */
1588                 if (nam->sa_family != AF_UNSPEC ||
1589                     nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1590                     sinp->sin_addr.s_addr != INADDR_ANY)
1591                         return (EAFNOSUPPORT);
1592                 nam->sa_family = AF_INET;
1593         }
1594         if (nam->sa_len != sizeof(struct sockaddr_in))
1595                 return (EINVAL);
1596
1597         INP_WLOCK(inp);
1598         INP_HASH_WLOCK(pcbinfo);
1599         error = in_pcbbind(inp, nam, td->td_ucred);
1600         INP_HASH_WUNLOCK(pcbinfo);
1601         INP_WUNLOCK(inp);
1602         return (error);
1603 }
1604
1605 static void
1606 udp_close(struct socket *so)
1607 {
1608         struct inpcb *inp;
1609         struct inpcbinfo *pcbinfo;
1610
1611         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1612         inp = sotoinpcb(so);
1613         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1614         INP_WLOCK(inp);
1615         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1616                 INP_HASH_WLOCK(pcbinfo);
1617                 in_pcbdisconnect(inp);
1618                 inp->inp_laddr.s_addr = INADDR_ANY;
1619                 INP_HASH_WUNLOCK(pcbinfo);
1620                 soisdisconnected(so);
1621         }
1622         INP_WUNLOCK(inp);
1623 }
1624
1625 static int
1626 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1627 {
1628         struct epoch_tracker et;
1629         struct inpcb *inp;
1630         struct inpcbinfo *pcbinfo;
1631         struct sockaddr_in *sin;
1632         int error;
1633
1634         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1635         inp = sotoinpcb(so);
1636         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1637
1638         sin = (struct sockaddr_in *)nam;
1639         if (sin->sin_family != AF_INET)
1640                 return (EAFNOSUPPORT);
1641         if (sin->sin_len != sizeof(*sin))
1642                 return (EINVAL);
1643
1644         INP_WLOCK(inp);
1645         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1646                 INP_WUNLOCK(inp);
1647                 return (EISCONN);
1648         }
1649         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1650         if (error != 0) {
1651                 INP_WUNLOCK(inp);
1652                 return (error);
1653         }
1654         NET_EPOCH_ENTER(et);
1655         INP_HASH_WLOCK(pcbinfo);
1656         error = in_pcbconnect(inp, nam, td->td_ucred, true);
1657         INP_HASH_WUNLOCK(pcbinfo);
1658         NET_EPOCH_EXIT(et);
1659         if (error == 0)
1660                 soisconnected(so);
1661         INP_WUNLOCK(inp);
1662         return (error);
1663 }
1664
1665 static void
1666 udp_detach(struct socket *so)
1667 {
1668         struct inpcb *inp;
1669         struct udpcb *up;
1670
1671         inp = sotoinpcb(so);
1672         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1673         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1674             ("udp_detach: not disconnected"));
1675         INP_WLOCK(inp);
1676         up = intoudpcb(inp);
1677         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1678         inp->inp_ppcb = NULL;
1679         in_pcbdetach(inp);
1680         in_pcbfree(inp);
1681         udp_discardcb(up);
1682 }
1683
1684 static int
1685 udp_disconnect(struct socket *so)
1686 {
1687         struct inpcb *inp;
1688         struct inpcbinfo *pcbinfo;
1689
1690         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1691         inp = sotoinpcb(so);
1692         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1693         INP_WLOCK(inp);
1694         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1695                 INP_WUNLOCK(inp);
1696                 return (ENOTCONN);
1697         }
1698         INP_HASH_WLOCK(pcbinfo);
1699         in_pcbdisconnect(inp);
1700         inp->inp_laddr.s_addr = INADDR_ANY;
1701         INP_HASH_WUNLOCK(pcbinfo);
1702         SOCK_LOCK(so);
1703         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1704         SOCK_UNLOCK(so);
1705         INP_WUNLOCK(inp);
1706         return (0);
1707 }
1708
1709 static int
1710 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1711     struct mbuf *control, struct thread *td)
1712 {
1713         struct inpcb *inp;
1714         int error;
1715
1716         inp = sotoinpcb(so);
1717         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1718
1719         if (addr != NULL) {
1720                 error = 0;
1721                 if (addr->sa_family != AF_INET)
1722                         error = EAFNOSUPPORT;
1723                 else if (addr->sa_len != sizeof(struct sockaddr_in))
1724                         error = EINVAL;
1725                 if (__predict_false(error != 0)) {
1726                         m_freem(control);
1727                         m_freem(m);
1728                         return (error);
1729                 }
1730         }
1731         return (udp_output(inp, m, addr, control, td, flags));
1732 }
1733 #endif /* INET */
1734
1735 int
1736 udp_shutdown(struct socket *so)
1737 {
1738         struct inpcb *inp;
1739
1740         inp = sotoinpcb(so);
1741         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1742         INP_WLOCK(inp);
1743         socantsendmore(so);
1744         INP_WUNLOCK(inp);
1745         return (0);
1746 }
1747
1748 #ifdef INET
1749 struct pr_usrreqs udp_usrreqs = {
1750         .pru_abort =            udp_abort,
1751         .pru_attach =           udp_attach,
1752         .pru_bind =             udp_bind,
1753         .pru_connect =          udp_connect,
1754         .pru_control =          in_control,
1755         .pru_detach =           udp_detach,
1756         .pru_disconnect =       udp_disconnect,
1757         .pru_peeraddr =         in_getpeeraddr,
1758         .pru_send =             udp_send,
1759         .pru_soreceive =        soreceive_dgram,
1760         .pru_sosend =           sosend_dgram,
1761         .pru_shutdown =         udp_shutdown,
1762         .pru_sockaddr =         in_getsockaddr,
1763         .pru_sosetlabel =       in_pcbsosetlabel,
1764         .pru_close =            udp_close,
1765 };
1766 #endif /* INET */