udp: Unify ctlinput and ctlport cmd/faddr/icmp_ip processing
[dragonfly.git] / sys / netinet / udp_usrreq.c
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
63  * $FreeBSD: src/sys/netinet/udp_usrreq.c,v 1.64.2.18 2003/01/24 05:11:34 sam Exp $
64  */
65
66 #include "opt_ipsec.h"
67 #include "opt_inet6.h"
68
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.h>
72 #include <sys/malloc.h>
73 #include <sys/mbuf.h>
74 #include <sys/domain.h>
75 #include <sys/proc.h>
76 #include <sys/priv.h>
77 #include <sys/protosw.h>
78 #include <sys/socket.h>
79 #include <sys/socketvar.h>
80 #include <sys/sysctl.h>
81 #include <sys/syslog.h>
82 #include <sys/in_cksum.h>
83 #include <sys/ktr.h>
84
85 #include <sys/thread2.h>
86 #include <sys/socketvar2.h>
87 #include <sys/serialize.h>
88
89 #include <machine/stdarg.h>
90
91 #include <net/if.h>
92 #include <net/route.h>
93 #include <net/netmsg2.h>
94 #include <net/netisr2.h>
95
96 #include <netinet/in.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #endif
102 #include <netinet/in_pcb.h>
103 #include <netinet/in_var.h>
104 #include <netinet/ip_var.h>
105 #ifdef INET6
106 #include <netinet6/ip6_var.h>
107 #endif
108 #include <netinet/ip_icmp.h>
109 #include <netinet/icmp_var.h>
110 #include <netinet/udp.h>
111 #include <netinet/udp_var.h>
112
113 #ifdef FAST_IPSEC
114 #include <netproto/ipsec/ipsec.h>
115 #endif
116
117 #ifdef IPSEC
118 #include <netinet6/ipsec.h>
119 #endif
120
121 #define MSGF_UDP_SEND           MSGF_PROTO1
122
123 #define INP_DIRECT_DETACH       INP_FLAG_PROTO2
124
125 #define UDP_KTR_STRING          "inp=%p"
126 #define UDP_KTR_ARGS            struct inpcb *inp
127
128 #ifndef KTR_UDP
129 #define KTR_UDP                 KTR_ALL
130 #endif
131
132 KTR_INFO_MASTER(udp);
133 KTR_INFO(KTR_UDP, udp, send_beg, 0, UDP_KTR_STRING, UDP_KTR_ARGS);
134 KTR_INFO(KTR_UDP, udp, send_end, 1, UDP_KTR_STRING, UDP_KTR_ARGS);
135 KTR_INFO(KTR_UDP, udp, send_ipout, 2, UDP_KTR_STRING, UDP_KTR_ARGS);
136 KTR_INFO(KTR_UDP, udp, redisp_ipout_beg, 3, UDP_KTR_STRING, UDP_KTR_ARGS);
137 KTR_INFO(KTR_UDP, udp, redisp_ipout_end, 4, UDP_KTR_STRING, UDP_KTR_ARGS);
138 KTR_INFO(KTR_UDP, udp, send_redisp, 5, UDP_KTR_STRING, UDP_KTR_ARGS);
139 KTR_INFO(KTR_UDP, udp, send_inswildcard, 6, UDP_KTR_STRING, UDP_KTR_ARGS);
140
141 #define logudp(name, inp)       KTR_LOG(udp_##name, inp)
142
143 /*
144  * UDP protocol implementation.
145  * Per RFC 768, August, 1980.
146  */
147 #ifndef COMPAT_42
148 static int      udpcksum = 1;
149 #else
150 static int      udpcksum = 0;           /* XXX */
151 #endif
152 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
153     &udpcksum, 0, "Enable checksumming of UDP packets");
154
155 int     log_in_vain = 0;
156 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
157     &log_in_vain, 0, "Log all incoming UDP packets");
158
159 static int      blackhole = 0;
160 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
161         &blackhole, 0, "Do not send port unreachables for refused connects");
162
163 static int      strict_mcast_mship = 1;
164 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
165         &strict_mcast_mship, 0, "Only send multicast to member sockets");
166
167 int     udp_sosend_async = 1;
168 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_async, CTLFLAG_RW,
169         &udp_sosend_async, 0, "UDP asynchronized pru_send");
170
171 int     udp_sosend_prepend = 1;
172 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_prepend, CTLFLAG_RW,
173         &udp_sosend_prepend, 0,
174         "Prepend enough space for proto and link header in pru_send");
175
176 static int udp_reuseport_ext = 1;
177 SYSCTL_INT(_net_inet_udp, OID_AUTO, reuseport_ext, CTLFLAG_RW,
178         &udp_reuseport_ext, 0, "SO_REUSEPORT extension");
179
180 struct  inpcbinfo udbinfo[MAXCPU];
181
182 #ifndef UDBHASHSIZE
183 #define UDBHASHSIZE 16
184 #endif
185
186 struct  udpstat udpstat_percpu[MAXCPU] __cachealign;
187
188 #ifdef INET6
189 struct udp_in6 {
190         struct sockaddr_in6     uin6_sin;
191         u_char                  uin6_init_done : 1;
192 };
193 struct udp_ip6 {
194         struct ip6_hdr          uip6_ip6;
195         u_char                  uip6_init_done : 1;
196 };
197 #else
198 struct udp_in6;
199 struct udp_ip6;
200 #endif /* INET6 */
201
202 static void udp_append (struct inpcb *last, struct ip *ip,
203     struct mbuf *n, int off, struct sockaddr_in *udp_in,
204     struct udp_in6 *, struct udp_ip6 *);
205 #ifdef INET6
206 static void ip_2_ip6_hdr (struct ip6_hdr *ip6, struct ip *ip);
207 #endif
208
209 static int udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin,
210     struct sockaddr_in *if_sin);
211
212 static boolean_t udp_inswildcardhash(struct inpcb *inp,
213     struct netmsg_base *msg, int error);
214 static void udp_remwildcardhash(struct inpcb *inp);
215
216 void
217 udp_init(void)
218 {
219         struct inpcbportinfo *portinfo;
220         int cpu;
221
222         portinfo = kmalloc_cachealign(sizeof(*portinfo) * ncpus2, M_PCB,
223             M_WAITOK);
224
225         for (cpu = 0; cpu < ncpus2; cpu++) {
226                 struct inpcbinfo *uicb = &udbinfo[cpu];
227
228                 /*
229                  * NOTE:
230                  * UDP pcb list, wildcard hash table and localgroup hash
231                  * table are shared.
232                  */
233                 in_pcbinfo_init(uicb, cpu, TRUE);
234                 uicb->hashbase = hashinit(UDBHASHSIZE, M_PCB, &uicb->hashmask);
235
236                 in_pcbportinfo_init(&portinfo[cpu], UDBHASHSIZE, TRUE, cpu);
237                 uicb->portinfo = portinfo;
238                 uicb->portinfo_mask = ncpus2_mask;
239
240                 uicb->wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB,
241                     &uicb->wildcardhashmask);
242                 uicb->localgrphashbase = hashinit(UDBHASHSIZE, M_PCB,
243                     &uicb->localgrphashmask);
244
245                 uicb->ipi_size = sizeof(struct inpcb);
246         }
247
248         /*
249          * Initialize UDP statistics counters for each CPU.
250          */
251         for (cpu = 0; cpu < ncpus; ++cpu)
252                 bzero(&udpstat_percpu[cpu], sizeof(struct udpstat));
253 }
254
255 static int
256 sysctl_udpstat(SYSCTL_HANDLER_ARGS)
257 {
258         int cpu, error = 0;
259
260         for (cpu = 0; cpu < ncpus; ++cpu) {
261                 if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu],
262                                         sizeof(struct udpstat))))
263                         break;
264                 if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu],
265                                        sizeof(struct udpstat))))
266                         break;
267         }
268
269         return (error);
270 }
271 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
272     0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics");
273
274 void
275 udp_ctloutput(netmsg_t msg)
276 {
277         struct socket *so = msg->base.nm_so;
278         struct sockopt *sopt = msg->ctloutput.nm_sopt;
279         struct  inpcb *inp = so->so_pcb;
280
281         if (sopt->sopt_level == IPPROTO_IP) {
282                 switch (sopt->sopt_name) {
283                 case IP_MULTICAST_IF:
284                 case IP_MULTICAST_VIF:
285                 case IP_MULTICAST_TTL:
286                 case IP_MULTICAST_LOOP:
287                 case IP_ADD_MEMBERSHIP:
288                 case IP_DROP_MEMBERSHIP:
289                         if (&curthread->td_msgport != netisr_cpuport(0)) {
290                                 /*
291                                  * This pr_ctloutput msg will be forwarded
292                                  * to netisr0 to run; we can't do direct
293                                  * detaching anymore.
294                                  */
295                                 inp->inp_flags &= ~INP_DIRECT_DETACH;
296                         }
297                         break;
298                 }
299         }
300         return ip_ctloutput(msg);
301 }
302
303 /*
304  * Check multicast packets to make sure they are only sent to sockets with
305  * multicast memberships for the packet's destination address and arrival
306  * interface.  Multicast packets to multicast-unaware sockets are also
307  * disallowed.
308  *
309  * Returns 0 if the packet is acceptable, -1 if it is not.
310  */
311 static __inline int
312 check_multicast_membership(const struct ip *ip, const struct inpcb *inp,
313     const struct mbuf *m)
314 {
315         const struct ip_moptions *mopt;
316         int mshipno;
317
318         if (strict_mcast_mship == 0 ||
319             !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
320                 return (0);
321         }
322
323         KASSERT(&curthread->td_msgport == netisr_cpuport(0),
324             ("multicast input not in netisr0"));
325
326         mopt = inp->inp_moptions;
327         if (mopt == NULL)
328                 return (-1);
329         for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) {
330                 const struct in_multi *maddr = mopt->imo_membership[mshipno];
331
332                 if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr &&
333                     m->m_pkthdr.rcvif == maddr->inm_ifp) {
334                         return (0);
335                 }
336         }
337         return (-1);
338 }
339
340 struct udp_mcast_arg {
341         struct inpcb    *inp;
342         struct inpcb    *last;
343         struct ip       *ip;
344         struct mbuf     *m;
345         int             iphlen;
346         struct sockaddr_in *udp_in;
347 #ifdef INET6
348         struct udp_in6  *udp_in6;
349         struct udp_ip6  *udp_ip6;
350 #endif
351 };
352
353 static int
354 udp_mcast_input(struct udp_mcast_arg *arg)
355 {
356         struct inpcb *inp = arg->inp;
357         struct inpcb *last = arg->last;
358         struct ip *ip = arg->ip;
359         struct mbuf *m = arg->m;
360
361         if (check_multicast_membership(ip, inp, m) < 0)
362                 return ERESTART; /* caller continue */
363
364         if (last != NULL) {
365                 struct mbuf *n;
366
367 #ifdef IPSEC
368                 /* check AH/ESP integrity. */
369                 if (ipsec4_in_reject_so(m, last->inp_socket))
370                         ipsecstat.in_polvio++;
371                         /* do not inject data to pcb */
372                 else
373 #endif /*IPSEC*/
374 #ifdef FAST_IPSEC
375                 /* check AH/ESP integrity. */
376                 if (ipsec4_in_reject(m, last))
377                         ;
378                 else
379 #endif /*FAST_IPSEC*/
380                 if ((n = m_copypacket(m, MB_DONTWAIT)) != NULL)
381                         udp_append(last, ip, n,
382                             arg->iphlen + sizeof(struct udphdr),
383                             arg->udp_in,
384 #ifdef INET6
385                             arg->udp_in6, arg->udp_ip6
386 #else
387                             NULL, NULL
388 #endif
389                             );
390         }
391         arg->last = last = inp;
392
393         /*
394          * Don't look for additional matches if this one does
395          * not have either the SO_REUSEPORT or SO_REUSEADDR
396          * socket options set.  This heuristic avoids searching
397          * through all pcbs in the common case of a non-shared
398          * port.  It * assumes that an application will never
399          * clear these options after setting them.
400          */
401         if (!(last->inp_socket->so_options &
402             (SO_REUSEPORT | SO_REUSEADDR)))
403                 return EJUSTRETURN; /* caller stop */
404         return 0;
405 }
406
407 int
408 udp_input(struct mbuf **mp, int *offp, int proto)
409 {
410         struct sockaddr_in udp_in = { sizeof udp_in, AF_INET };
411 #ifdef INET6
412         struct udp_in6 udp_in6 = {
413                 { sizeof udp_in6.uin6_sin, AF_INET6 }, 0
414         };
415         struct udp_ip6 udp_ip6;
416 #endif
417
418         int iphlen;
419         struct ip *ip;
420         struct udphdr *uh;
421         struct inpcb *inp;
422         struct mbuf *m;
423         struct mbuf *opts = NULL;
424         int len, off;
425         struct ip save_ip;
426         struct sockaddr *append_sa;
427         struct inpcbinfo *pcbinfo = &udbinfo[mycpuid];
428
429         off = *offp;
430         m = *mp;
431         *mp = NULL;
432
433         iphlen = off;
434         udp_stat.udps_ipackets++;
435
436         /*
437          * Strip IP options, if any; should skip this,
438          * make available to user, and use on returned packets,
439          * but we don't yet have a way to check the checksum
440          * with options still present.
441          */
442         if (iphlen > sizeof(struct ip)) {
443                 ip_stripoptions(m);
444                 iphlen = sizeof(struct ip);
445         }
446
447         /*
448          * IP and UDP headers are together in first mbuf.
449          * Already checked and pulled up in ip_demux().
450          */
451         KASSERT(m->m_len >= iphlen + sizeof(struct udphdr),
452             ("UDP header not in one mbuf"));
453
454         ip = mtod(m, struct ip *);
455         uh = (struct udphdr *)((caddr_t)ip + iphlen);
456
457         /* destination port of 0 is illegal, based on RFC768. */
458         if (uh->uh_dport == 0)
459                 goto bad;
460
461         /*
462          * Make mbuf data length reflect UDP length.
463          * If not enough data to reflect UDP length, drop.
464          */
465         len = ntohs((u_short)uh->uh_ulen);
466         if (ip->ip_len != len) {
467                 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
468                         udp_stat.udps_badlen++;
469                         goto bad;
470                 }
471                 m_adj(m, len - ip->ip_len);
472                 /* ip->ip_len = len; */
473         }
474         /*
475          * Save a copy of the IP header in case we want restore it
476          * for sending an ICMP error message in response.
477          */
478         save_ip = *ip;
479
480         /*
481          * Checksum extended UDP header and data.
482          */
483         if (uh->uh_sum) {
484                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
485                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
486                                 uh->uh_sum = m->m_pkthdr.csum_data;
487                         else
488                                 uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
489                                     ip->ip_dst.s_addr, htonl((u_short)len +
490                                     m->m_pkthdr.csum_data + IPPROTO_UDP));
491                         uh->uh_sum ^= 0xffff;
492                 } else {
493                         char b[9];
494
495                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
496                         bzero(((struct ipovly *)ip)->ih_x1, 9);
497                         ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
498                         uh->uh_sum = in_cksum(m, len + sizeof(struct ip));
499                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
500                 }
501                 if (uh->uh_sum) {
502                         udp_stat.udps_badsum++;
503                         m_freem(m);
504                         return(IPPROTO_DONE);
505                 }
506         } else
507                 udp_stat.udps_nosum++;
508
509         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
510             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
511                 struct inpcbhead *connhead;
512                 struct inpcontainer *ic, *ic_marker;
513                 struct inpcontainerhead *ichead;
514                 struct udp_mcast_arg arg;
515                 struct inpcb *last;
516                 int error;
517
518                 /*
519                  * Deliver a multicast or broadcast datagram to *all* sockets
520                  * for which the local and remote addresses and ports match
521                  * those of the incoming datagram.  This allows more than
522                  * one process to receive multi/broadcasts on the same port.
523                  * (This really ought to be done for unicast datagrams as
524                  * well, but that would cause problems with existing
525                  * applications that open both address-specific sockets and
526                  * a wildcard socket listening to the same port -- they would
527                  * end up receiving duplicates of every unicast datagram.
528                  * Those applications open the multiple sockets to overcome an
529                  * inadequacy of the UDP socket interface, but for backwards
530                  * compatibility we avoid the problem here rather than
531                  * fixing the interface.  Maybe 4.5BSD will remedy this?)
532                  */
533
534                 /*
535                  * Construct sockaddr format source address.
536                  */
537                 udp_in.sin_port = uh->uh_sport;
538                 udp_in.sin_addr = ip->ip_src;
539                 arg.udp_in = &udp_in;
540                 /*
541                  * Locate pcb(s) for datagram.
542                  * (Algorithm copied from raw_intr().)
543                  */
544                 last = NULL;
545 #ifdef INET6
546                 udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0;
547                 arg.udp_in6 = &udp_in6;
548                 arg.udp_ip6 = &udp_ip6;
549 #endif
550                 arg.iphlen = iphlen;
551
552                 connhead = &pcbinfo->hashbase[
553                     INP_PCBCONNHASH(ip->ip_src.s_addr, uh->uh_sport,
554                     ip->ip_dst.s_addr, uh->uh_dport, pcbinfo->hashmask)];
555                 LIST_FOREACH(inp, connhead, inp_hash) {
556 #ifdef INET6
557                         if (!(inp->inp_vflag & INP_IPV4))
558                                 continue;
559 #endif
560                         if (!in_hosteq(inp->inp_faddr, ip->ip_src) ||
561                             !in_hosteq(inp->inp_laddr, ip->ip_dst) ||
562                             inp->inp_fport != uh->uh_sport ||
563                             inp->inp_lport != uh->uh_dport)
564                                 continue;
565
566                         arg.inp = inp;
567                         arg.last = last;
568                         arg.ip = ip;
569                         arg.m = m;
570
571                         error = udp_mcast_input(&arg);
572                         if (error == ERESTART)
573                                 continue;
574                         last = arg.last;
575
576                         if (error == EJUSTRETURN)
577                                 goto done;
578                 }
579
580                 ichead = &pcbinfo->wildcardhashbase[
581                     INP_PCBWILDCARDHASH(uh->uh_dport,
582                     pcbinfo->wildcardhashmask)];
583                 ic_marker = in_pcbcontainer_marker(mycpuid);
584
585                 GET_PCBINFO_TOKEN(pcbinfo);
586                 LIST_INSERT_HEAD(ichead, ic_marker, ic_list);
587                 while ((ic = LIST_NEXT(ic_marker, ic_list)) != NULL) {
588                         LIST_REMOVE(ic_marker, ic_list);
589                         LIST_INSERT_AFTER(ic, ic_marker, ic_list);
590
591                         inp = ic->ic_inp;
592                         if (inp->inp_flags & INP_PLACEMARKER)
593                                 continue;
594 #ifdef INET6
595                         if (!(inp->inp_vflag & INP_IPV4))
596                                 continue;
597 #endif
598                         if (inp->inp_lport != uh->uh_dport)
599                                 continue;
600                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
601                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
602                                 continue;
603
604                         arg.inp = inp;
605                         arg.last = last;
606                         arg.ip = ip;
607                         arg.m = m;
608
609                         error = udp_mcast_input(&arg);
610                         if (error == ERESTART)
611                                 continue;
612                         last = arg.last;
613
614                         if (error == EJUSTRETURN)
615                                 break;
616                 }
617                 LIST_REMOVE(ic_marker, ic_list);
618                 REL_PCBINFO_TOKEN(pcbinfo);
619 done:
620                 if (last == NULL) {
621                         /*
622                          * No matching pcb found; discard datagram.
623                          * (No need to send an ICMP Port Unreachable
624                          * for a broadcast or multicast datgram.)
625                          */
626                         udp_stat.udps_noportbcast++;
627                         goto bad;
628                 }
629 #ifdef IPSEC
630                 /* check AH/ESP integrity. */
631                 if (ipsec4_in_reject_so(m, last->inp_socket)) {
632                         ipsecstat.in_polvio++;
633                         goto bad;
634                 }
635 #endif /*IPSEC*/
636 #ifdef FAST_IPSEC
637                 /* check AH/ESP integrity. */
638                 if (ipsec4_in_reject(m, last))
639                         goto bad;
640 #endif /*FAST_IPSEC*/
641                 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
642                     &udp_in,
643 #ifdef INET6
644                     &udp_in6, &udp_ip6
645 #else
646                     NULL, NULL
647 #endif
648                     );
649                 return(IPPROTO_DONE);
650         }
651         /*
652          * Locate pcb for datagram.
653          */
654         inp = in_pcblookup_pkthash(pcbinfo, ip->ip_src, uh->uh_sport,
655             ip->ip_dst, uh->uh_dport, TRUE, m->m_pkthdr.rcvif,
656             udp_reuseport_ext ? m : NULL);
657         if (inp == NULL) {
658                 if (log_in_vain) {
659                         char buf[sizeof "aaa.bbb.ccc.ddd"];
660
661                         strcpy(buf, inet_ntoa(ip->ip_dst));
662                         log(LOG_INFO,
663                             "Connection attempt to UDP %s:%d from %s:%d\n",
664                             buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
665                             ntohs(uh->uh_sport));
666                 }
667                 udp_stat.udps_noport++;
668                 if (m->m_flags & (M_BCAST | M_MCAST)) {
669                         udp_stat.udps_noportbcast++;
670                         goto bad;
671                 }
672                 if (blackhole)
673                         goto bad;
674 #ifdef ICMP_BANDLIM
675                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
676                         goto bad;
677 #endif
678                 *ip = save_ip;
679                 ip->ip_len += iphlen;
680                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
681                 return(IPPROTO_DONE);
682         }
683 #ifdef IPSEC
684         if (ipsec4_in_reject_so(m, inp->inp_socket)) {
685                 ipsecstat.in_polvio++;
686                 goto bad;
687         }
688 #endif /*IPSEC*/
689 #ifdef FAST_IPSEC
690         if (ipsec4_in_reject(m, inp))
691                 goto bad;
692 #endif /*FAST_IPSEC*/
693         /*
694          * Check the minimum TTL for socket.
695          */
696         if (ip->ip_ttl < inp->inp_ip_minttl)
697                 goto bad;
698
699         /*
700          * Construct sockaddr format source address.
701          * Stuff source address and datagram in user buffer.
702          */
703         udp_in.sin_port = uh->uh_sport;
704         udp_in.sin_addr = ip->ip_src;
705         if ((inp->inp_flags & INP_CONTROLOPTS) ||
706             (inp->inp_socket->so_options & SO_TIMESTAMP)) {
707 #ifdef INET6
708                 if (inp->inp_vflag & INP_IPV6) {
709                         int savedflags;
710
711                         ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
712                         savedflags = inp->inp_flags;
713                         inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
714                         ip6_savecontrol(inp, &opts, &udp_ip6.uip6_ip6, m);
715                         inp->inp_flags = savedflags;
716                 } else
717 #endif
718                 ip_savecontrol(inp, &opts, ip, m);
719         }
720         m_adj(m, iphlen + sizeof(struct udphdr));
721 #ifdef INET6
722         if (inp->inp_vflag & INP_IPV6) {
723                 in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
724                 append_sa = (struct sockaddr *)&udp_in6;
725         } else
726 #endif
727                 append_sa = (struct sockaddr *)&udp_in;
728
729         lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token);
730         if (ssb_appendaddr(&inp->inp_socket->so_rcv, append_sa, m, opts) == 0) {
731                 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
732                 udp_stat.udps_fullsock++;
733                 goto bad;
734         }
735         lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
736         sorwakeup(inp->inp_socket);
737         return(IPPROTO_DONE);
738 bad:
739         m_freem(m);
740         if (opts)
741                 m_freem(opts);
742         return(IPPROTO_DONE);
743 }
744
745 #ifdef INET6
746 static void
747 ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip)
748 {
749         bzero(ip6, sizeof *ip6);
750
751         ip6->ip6_vfc = IPV6_VERSION;
752         ip6->ip6_plen = ip->ip_len;
753         ip6->ip6_nxt = ip->ip_p;
754         ip6->ip6_hlim = ip->ip_ttl;
755         ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] =
756                 IPV6_ADDR_INT32_SMP;
757         ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr;
758         ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr;
759 }
760 #endif
761
762 /*
763  * subroutine of udp_input(), mainly for source code readability.
764  * caller must properly init udp_ip6 and udp_in6 beforehand.
765  */
766 static void
767 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off,
768     struct sockaddr_in *udp_in,
769     struct udp_in6 *udp_in6, struct udp_ip6 *udp_ip6)
770 {
771         struct sockaddr *append_sa;
772         struct mbuf *opts = NULL;
773         int ret;
774
775         if (last->inp_flags & INP_CONTROLOPTS ||
776             last->inp_socket->so_options & SO_TIMESTAMP) {
777 #ifdef INET6
778                 if (last->inp_vflag & INP_IPV6) {
779                         int savedflags;
780
781                         if (udp_ip6->uip6_init_done == 0) {
782                                 ip_2_ip6_hdr(&udp_ip6->uip6_ip6, ip);
783                                 udp_ip6->uip6_init_done = 1;
784                         }
785                         savedflags = last->inp_flags;
786                         last->inp_flags &= ~INP_UNMAPPABLEOPTS;
787                         ip6_savecontrol(last, &opts, &udp_ip6->uip6_ip6, n);
788                         last->inp_flags = savedflags;
789                 } else
790 #endif
791                 ip_savecontrol(last, &opts, ip, n);
792         }
793 #ifdef INET6
794         if (last->inp_vflag & INP_IPV6) {
795                 if (udp_in6->uin6_init_done == 0) {
796                         in6_sin_2_v4mapsin6(udp_in, &udp_in6->uin6_sin);
797                         udp_in6->uin6_init_done = 1;
798                 }
799                 append_sa = (struct sockaddr *)&udp_in6->uin6_sin;
800         } else
801 #endif
802                 append_sa = (struct sockaddr *)udp_in;
803         m_adj(n, off);
804
805         lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token);
806         ret = ssb_appendaddr(&last->inp_socket->so_rcv, append_sa, n, opts);
807         lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token);
808         if (ret == 0) {
809                 m_freem(n);
810                 if (opts)
811                         m_freem(opts);
812                 udp_stat.udps_fullsock++;
813         } else {
814                 sorwakeup(last->inp_socket);
815         }
816 }
817
818 /*
819  * Notify a udp user of an asynchronous error;
820  * just wake up so that he can collect error status.
821  */
822 void
823 udp_notify(struct inpcb *inp, int error)
824 {
825         inp->inp_socket->so_error = error;
826         sorwakeup(inp->inp_socket);
827         sowwakeup(inp->inp_socket);
828 }
829
830 struct netmsg_udp_notify {
831         struct netmsg_base base;
832         inp_notify_t    nm_notify;
833         struct in_addr  nm_faddr;
834         int             nm_arg;
835 };
836
837 static void
838 udp_notifyall_oncpu(netmsg_t msg)
839 {
840         struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg;
841         int nextcpu, cpu = mycpuid;
842
843         in_pcbnotifyall(&udbinfo[cpu], nm->nm_faddr, nm->nm_arg, nm->nm_notify);
844
845         nextcpu = cpu + 1;
846         if (nextcpu < ncpus2)
847                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
848         else
849                 lwkt_replymsg(&nm->base.lmsg, 0);
850 }
851
852 inp_notify_t
853 udp_get_inpnotify(int cmd, const struct sockaddr *sa,
854     struct ip **ip0, int *cpuid)
855 {
856         struct in_addr faddr;
857         struct ip *ip = *ip0;
858         inp_notify_t notify = udp_notify;
859
860         faddr = ((const struct sockaddr_in *)sa)->sin_addr;
861         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
862                 return NULL;
863
864         if (PRC_IS_REDIRECT(cmd)) {
865                 ip = NULL;
866                 notify = in_rtchange;
867         } else if (cmd == PRC_HOSTDEAD) {
868                 ip = NULL;
869         } else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
870                 return NULL;
871         }
872
873         if (cpuid != NULL) {
874                 if (ip == NULL) {
875                         /* Go through all CPUs */
876                         *cpuid = ncpus;
877                 } else {
878                         const struct udphdr *uh;
879
880                         uh = (const struct udphdr *)
881                             ((caddr_t)ip + (ip->ip_hl << 2));
882                         *cpuid = udp_addrcpu(faddr.s_addr, ip->ip_src.s_addr,
883                             uh->uh_dport, uh->uh_sport);
884                 }
885         }
886
887         *ip0 = ip;
888         return notify;
889 }
890
891 void
892 udp_ctlinput(netmsg_t msg)
893 {
894         struct sockaddr *sa = msg->ctlinput.nm_arg;
895         struct ip *ip = msg->ctlinput.nm_extra;
896         int cmd = msg->ctlinput.nm_cmd;
897         inp_notify_t notify;
898         struct in_addr faddr;
899
900         notify = udp_get_inpnotify(cmd, sa, &ip, NULL);
901         if (notify == NULL)
902                 goto done;
903
904         faddr = ((struct sockaddr_in *)sa)->sin_addr;
905         if (ip) {
906                 const struct udphdr *uh;
907                 struct inpcb *inp;
908
909                 uh = (const struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
910                 inp = in_pcblookup_hash(&udbinfo[mycpuid], faddr, uh->uh_dport,
911                                         ip->ip_src, uh->uh_sport, 0, NULL);
912                 if (inp != NULL && inp->inp_socket != NULL)
913                         notify(inp, inetctlerrmap[cmd]);
914         } else {
915                 struct netmsg_udp_notify *nm;
916
917                 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
918                 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
919                 netmsg_init(&nm->base, NULL, &netisr_afree_rport,
920                             0, udp_notifyall_oncpu);
921                 nm->nm_faddr = faddr;
922                 nm->nm_arg = inetctlerrmap[cmd];
923                 nm->nm_notify = notify;
924                 lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
925         }
926 done:
927         lwkt_replymsg(&msg->lmsg, 0);
928 }
929
930 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, udbinfo, 0,
931             in_pcblist_global_ncpus2, "S,xinpcb", "List of active UDP sockets");
932
933 static int
934 udp_getcred(SYSCTL_HANDLER_ARGS)
935 {
936         struct sockaddr_in addrs[2];
937         struct ucred cred0, *cred = NULL;
938         struct inpcb *inp;
939         int error, cpu, origcpu;
940
941         error = priv_check(req->td, PRIV_ROOT);
942         if (error)
943                 return (error);
944         error = SYSCTL_IN(req, addrs, sizeof addrs);
945         if (error)
946                 return (error);
947
948         origcpu = mycpuid;
949         cpu = udp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
950             addrs[0].sin_addr.s_addr, addrs[0].sin_port);
951
952         lwkt_migratecpu(cpu);
953
954         inp = in_pcblookup_hash(&udbinfo[cpu],
955             addrs[1].sin_addr, addrs[1].sin_port,
956             addrs[0].sin_addr, addrs[0].sin_port, TRUE, NULL);
957         if (inp == NULL || inp->inp_socket == NULL) {
958                 error = ENOENT;
959         } else if (inp->inp_socket->so_cred != NULL) {
960                 cred0 = *(inp->inp_socket->so_cred);
961                 cred = &cred0;
962         }
963
964         lwkt_migratecpu(origcpu);
965
966         if (error)
967                 return error;
968
969         return SYSCTL_OUT(req, cred, sizeof(struct ucred));
970 }
971 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
972     0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection");
973
974 static void
975 udp_send_redispatch(netmsg_t msg)
976 {
977         struct mbuf *m = msg->send.nm_m;
978         int pru_flags = msg->send.nm_flags;
979         struct inpcb *inp = msg->send.base.nm_so->so_pcb;
980         struct mbuf *m_opt = msg->send.nm_control; /* XXX save ipopt */
981         int flags = msg->send.nm_priv; /* ip_output flags */
982         int error;
983
984         logudp(redisp_ipout_beg, inp);
985
986         /*
987          * - Don't use inp route cache.  It should only be used in the
988          *   inp owner netisr.
989          * - Access to inp_moptions should be safe, since multicast UDP
990          *   datagrams are redispatched to netisr0 and inp_moptions is
991          *   changed only in netisr0.
992          */
993         error = ip_output(m, m_opt, NULL, flags, inp->inp_moptions, inp);
994         if ((pru_flags & PRUS_NOREPLY) == 0)
995                 lwkt_replymsg(&msg->send.base.lmsg, error);
996
997         if (m_opt != NULL) {
998                 /* Free saved ip options, if any */
999                 m_freem(m_opt);
1000         }
1001
1002         logudp(redisp_ipout_end, inp);
1003 }
1004
1005 static void
1006 udp_send(netmsg_t msg)
1007 {
1008         struct socket *so = msg->send.base.nm_so;
1009         struct mbuf *m = msg->send.nm_m;
1010         struct sockaddr *dstaddr = msg->send.nm_addr;
1011         int pru_flags = msg->send.nm_flags;
1012         struct inpcb *inp = so->so_pcb;
1013         struct thread *td = msg->send.nm_td;
1014         int flags;
1015
1016         struct udpiphdr *ui;
1017         int len = m->m_pkthdr.len;
1018         struct sockaddr_in *sin;        /* really is initialized before use */
1019         int error = 0, cpu;
1020
1021         KKASSERT(msg->send.nm_control == NULL);
1022
1023         logudp(send_beg, inp);
1024
1025         if (inp == NULL) {
1026                 error = EINVAL;
1027                 goto release;
1028         }
1029
1030         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1031                 error = EMSGSIZE;
1032                 goto release;
1033         }
1034
1035         if (inp->inp_lport == 0) {      /* unbound socket */
1036                 boolean_t forwarded;
1037
1038                 error = in_pcbbind(inp, NULL, td);
1039                 if (error)
1040                         goto release;
1041
1042                 /*
1043                  * Need to call udp_send again, after this inpcb is
1044                  * inserted into wildcard hash table.
1045                  */
1046                 msg->send.base.lmsg.ms_flags |= MSGF_UDP_SEND;
1047                 forwarded = udp_inswildcardhash(inp, &msg->send.base, 0);
1048                 if (forwarded) {
1049                         /*
1050                          * The message is further forwarded, so we are
1051                          * done here.
1052                          */
1053                         logudp(send_inswildcard, inp);
1054                         return;
1055                 }
1056         }
1057
1058         if (dstaddr != NULL) {          /* destination address specified */
1059                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1060                         /* already connected */
1061                         error = EISCONN;
1062                         goto release;
1063                 }
1064                 sin = (struct sockaddr_in *)dstaddr;
1065                 if (!prison_remote_ip(td, (struct sockaddr *)&sin)) {
1066                         error = EAFNOSUPPORT; /* IPv6 only jail */
1067                         goto release;
1068                 }
1069         } else {
1070                 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1071                         /* no destination specified and not already connected */
1072                         error = ENOTCONN;
1073                         goto release;
1074                 }
1075                 sin = NULL;
1076         }
1077
1078         /*
1079          * Calculate data length and get a mbuf
1080          * for UDP and IP headers.
1081          */
1082         M_PREPEND(m, sizeof(struct udpiphdr), MB_DONTWAIT);
1083         if (m == NULL) {
1084                 error = ENOBUFS;
1085                 goto release;
1086         }
1087
1088         /*
1089          * Fill in mbuf with extended UDP header
1090          * and addresses and length put into network format.
1091          */
1092         ui = mtod(m, struct udpiphdr *);
1093         bzero(ui->ui_x1, sizeof ui->ui_x1);     /* XXX still needed? */
1094         ui->ui_pr = IPPROTO_UDP;
1095
1096         /*
1097          * Set destination address.
1098          */
1099         if (dstaddr != NULL) {                  /* use specified destination */
1100                 ui->ui_dst = sin->sin_addr;
1101                 ui->ui_dport = sin->sin_port;
1102         } else {                                /* use connected destination */
1103                 ui->ui_dst = inp->inp_faddr;
1104                 ui->ui_dport = inp->inp_fport;
1105         }
1106
1107         /*
1108          * Set source address.
1109          */
1110         if (inp->inp_laddr.s_addr == INADDR_ANY ||
1111             IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) {
1112                 struct sockaddr_in *if_sin;
1113
1114                 if (dstaddr == NULL) {  
1115                         /*
1116                          * connect() had (or should have) failed because
1117                          * the interface had no IP address, but the
1118                          * application proceeded to call send() anyways.
1119                          */
1120                         error = ENOTCONN;
1121                         goto release;
1122                 }
1123
1124                 /* Look up outgoing interface. */
1125                 error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1);
1126                 if (error)
1127                         goto release;
1128                 ui->ui_src = if_sin->sin_addr;  /* use address of interface */
1129         } else {
1130                 ui->ui_src = inp->inp_laddr;    /* use non-null bound address */
1131         }
1132         ui->ui_sport = inp->inp_lport;
1133         KASSERT(inp->inp_lport != 0, ("inp lport should have been bound"));
1134
1135         /*
1136          * Release the original thread, since it is no longer used
1137          */
1138         if (pru_flags & PRUS_HELDTD) {
1139                 lwkt_rele(td);
1140                 pru_flags &= ~PRUS_HELDTD;
1141         }
1142         /*
1143          * Free the dest address, since it is no longer needed
1144          */
1145         if (pru_flags & PRUS_FREEADDR) {
1146                 kfree(dstaddr, M_SONAME);
1147                 pru_flags &= ~PRUS_FREEADDR;
1148         }
1149
1150         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1151
1152         /*
1153          * Set up checksum and output datagram.
1154          */
1155         if (udpcksum) {
1156                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr,
1157                     htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1158                 m->m_pkthdr.csum_flags = CSUM_UDP;
1159                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1160                 m->m_pkthdr.csum_thlen = sizeof(struct udphdr);
1161         } else {
1162                 ui->ui_sum = 0;
1163         }
1164         ((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len;
1165         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1166         ((struct ip *)ui)->ip_tos = inp->inp_ip_tos;    /* XXX */
1167         udp_stat.udps_opackets++;
1168
1169         flags = IP_DEBUGROUTE |
1170             (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST));
1171         if (pru_flags & PRUS_DONTROUTE)
1172                 flags |= SO_DONTROUTE;
1173
1174         if (inp->inp_flags & INP_CONNECTED) {
1175                 /*
1176                  * For connected socket, this datagram has already
1177                  * been in the correct netisr; no need to rehash.
1178                  */
1179                 goto sendit;
1180         }
1181
1182         cpu = udp_addrcpu(ui->ui_dst.s_addr, ui->ui_dport,
1183             ui->ui_src.s_addr, ui->ui_sport);
1184         if (cpu != mycpuid) {
1185                 struct mbuf *m_opt = NULL;
1186                 struct netmsg_pru_send *smsg;
1187                 struct lwkt_port *port = netisr_cpuport(cpu);
1188
1189                 /*
1190                  * Not on the CPU that matches this UDP datagram hash;
1191                  * redispatch to the correct CPU to do the ip_output().
1192                  */
1193                 if (inp->inp_options != NULL) {
1194                         /*
1195                          * If there are ip options, then save a copy,
1196                          * since accessing inp_options on other CPUs'
1197                          * is not safe.
1198                          *
1199                          * XXX optimize this?
1200                          */
1201                         m_opt = m_copym(inp->inp_options, 0, M_COPYALL,
1202                             MB_WAIT);
1203                 }
1204                 if ((pru_flags & PRUS_NOREPLY) == 0) {
1205                         /*
1206                          * Change some parts of the original netmsg and
1207                          * forward it to the target netisr.
1208                          *
1209                          * NOTE: so_port MUST NOT be checked in the target
1210                          * netisr.
1211                          */
1212                         smsg = &msg->send;
1213                         smsg->nm_priv = flags; /* ip_output flags */
1214                         smsg->nm_m = m;
1215                         smsg->nm_control = m_opt; /* XXX save ipopt */
1216                         smsg->base.lmsg.ms_flags |= MSGF_IGNSOPORT;
1217                         smsg->base.nm_dispatch = udp_send_redispatch;
1218                         lwkt_forwardmsg(port, &smsg->base.lmsg);
1219                 } else {
1220                         /*
1221                          * Recreate the netmsg, since the original mbuf
1222                          * could have been changed.  And send it to the
1223                          * target netisr.
1224                          *
1225                          * NOTE: so_port MUST NOT be checked in the target
1226                          * netisr.
1227                          */
1228                         smsg = &m->m_hdr.mh_sndmsg;
1229                         netmsg_init(&smsg->base, so, &netisr_apanic_rport,
1230                             MSGF_IGNSOPORT, udp_send_redispatch);
1231                         smsg->nm_priv = flags; /* ip_output flags */
1232                         smsg->nm_flags = pru_flags;
1233                         smsg->nm_m = m;
1234                         smsg->nm_control = m_opt; /* XXX save ipopt */
1235                         lwkt_sendmsg(port, &smsg->base.lmsg);
1236                 }
1237
1238                 /* This UDP datagram is redispatched; done */
1239                 logudp(send_redisp, inp);
1240                 return;
1241         }
1242
1243 sendit:
1244         logudp(send_ipout, inp);
1245         error = ip_output(m, inp->inp_options, &inp->inp_route, flags,
1246             inp->inp_moptions, inp);
1247         m = NULL;
1248
1249 release:
1250         if (m != NULL)
1251                 m_freem(m);
1252
1253         if (pru_flags & PRUS_HELDTD)
1254                 lwkt_rele(td);
1255         if (pru_flags & PRUS_FREEADDR)
1256                 kfree(dstaddr, M_SONAME);
1257         if ((pru_flags & PRUS_NOREPLY) == 0)
1258                 lwkt_replymsg(&msg->send.base.lmsg, error);
1259
1260         logudp(send_end, inp);
1261 }
1262
1263 u_long  udp_sendspace = 9216;           /* really max datagram size */
1264                                         /* 40 1K datagrams */
1265 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
1266     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
1267
1268 u_long  udp_recvspace = 40 * (1024 +
1269 #ifdef INET6
1270                                       sizeof(struct sockaddr_in6)
1271 #else
1272                                       sizeof(struct sockaddr_in)
1273 #endif
1274                                       );
1275 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1276     &udp_recvspace, 0, "Maximum incoming UDP datagram size");
1277
1278 /*
1279  * This should never happen, since UDP socket does not support
1280  * connection acception (SO_ACCEPTCONN, i.e. listen(2)).
1281  */
1282 static void
1283 udp_abort(netmsg_t msg __unused)
1284 {
1285         panic("udp_abort is called");
1286 }
1287
1288 static void
1289 udp_attach(netmsg_t msg)
1290 {
1291         struct socket *so = msg->attach.base.nm_so;
1292         struct pru_attach_info *ai = msg->attach.nm_ai;
1293         struct inpcb *inp;
1294         int error;
1295
1296         inp = so->so_pcb;
1297         if (inp != NULL) {
1298                 error = EINVAL;
1299                 goto out;
1300         }
1301         error = soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit);
1302         if (error)
1303                 goto out;
1304
1305         error = in_pcballoc(so, &udbinfo[mycpuid]);
1306         if (error)
1307                 goto out;
1308
1309         inp = (struct inpcb *)so->so_pcb;
1310         inp->inp_flags |= INP_DIRECT_DETACH;
1311         inp->inp_vflag |= INP_IPV4;
1312         inp->inp_ip_ttl = ip_defttl;
1313         error = 0;
1314 out:
1315         lwkt_replymsg(&msg->attach.base.lmsg, error);
1316 }
1317
1318 static void
1319 udp_inswildcard_replymsg(netmsg_t msg)
1320 {
1321         lwkt_msg_t lmsg = &msg->lmsg;
1322
1323         if (lmsg->ms_flags & MSGF_UDP_SEND) {
1324                 udp_send(msg);
1325                 /* msg is replied by udp_send() */
1326         } else {
1327                 lwkt_replymsg(lmsg, lmsg->ms_error);
1328         }
1329 }
1330
1331 static void
1332 udp_soreuseport_dispatch(netmsg_t msg)
1333 {
1334         /* This inpcb has already been in the wildcard hash. */
1335         in_pcblink_flags(msg->base.nm_so->so_pcb, &udbinfo[mycpuid], 0);
1336         udp_inswildcard_replymsg(msg);
1337 }
1338
1339 static void
1340 udp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
1341 {
1342         sosetport(((struct netmsg_base *)msg)->nm_so, port);
1343 }
1344
1345 static boolean_t
1346 udp_inswildcardhash_oncpu(struct inpcb *inp, struct netmsg_base *msg)
1347 {
1348         int cpu;
1349
1350         KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid],
1351             ("not on owner cpu"));
1352
1353         in_pcbinswildcardhash(inp);
1354         for (cpu = 0; cpu < ncpus2; ++cpu) {
1355                 if (cpu == mycpuid) {
1356                         /*
1357                          * This inpcb has been inserted by the above
1358                          * in_pcbinswildcardhash().
1359                          */
1360                         continue;
1361                 }
1362                 in_pcbinswildcardhash_oncpu(inp, &udbinfo[cpu]);
1363         }
1364
1365         if (inp->inp_socket->so_options & SO_REUSEPORT) {
1366                 /*
1367                  * For SO_REUSEPORT socket, redistribute it based on its
1368                  * local group index.
1369                  */
1370                 cpu = inp->inp_lgrpindex & ncpus2_mask;
1371                 if (cpu != mycpuid) {
1372                         struct lwkt_port *port = netisr_cpuport(cpu);
1373                         lwkt_msg_t lmsg = &msg->lmsg;
1374
1375                         /*
1376                          * We are moving the protocol processing port the
1377                          * socket is on, we have to unlink here and re-link
1378                          * on the target cpu (this inpcb is still left in
1379                          * the wildcard hash).
1380                          */
1381                         in_pcbunlink_flags(inp, &udbinfo[mycpuid], 0);
1382                         msg->nm_dispatch = udp_soreuseport_dispatch;
1383
1384                         /*
1385                          * See the related comment in tcp_usrreq.c
1386                          * tcp_connect()
1387                          */
1388                         lwkt_setmsg_receipt(lmsg, udp_sosetport);
1389                         lwkt_forwardmsg(port, lmsg);
1390                         return TRUE; /* forwarded */
1391                 }
1392         }
1393         return FALSE;
1394 }
1395
1396 static void
1397 udp_inswildcardhash_dispatch(netmsg_t msg)
1398 {
1399         struct inpcb *inp = msg->base.nm_so->so_pcb;
1400         boolean_t forwarded;
1401
1402         KASSERT(inp->inp_lport != 0, ("local port not set yet"));
1403         KASSERT((ntohs(inp->inp_lport) & ncpus2_mask) == mycpuid,
1404             ("not target cpu"));
1405
1406         in_pcblink(inp, &udbinfo[mycpuid]);
1407
1408         forwarded = udp_inswildcardhash_oncpu(inp, &msg->base);
1409         if (forwarded) {
1410                 /* The message is further forwarded, so we are done here. */
1411                 return;
1412         }
1413         udp_inswildcard_replymsg(msg);
1414 }
1415
1416 static boolean_t
1417 udp_inswildcardhash(struct inpcb *inp, struct netmsg_base *msg, int error)
1418 {
1419         lwkt_msg_t lmsg = &msg->lmsg;
1420         int cpu;
1421
1422         ASSERT_INP_NOTINHASH(inp);
1423
1424         /* This inpcb could no longer be directly detached */
1425         inp->inp_flags &= ~INP_DIRECT_DETACH;
1426
1427         /*
1428          * Always clear the route cache, so we don't need to
1429          * worry about any owner CPU changes later.
1430          */
1431         in_pcbresetroute(inp);
1432
1433         KASSERT(inp->inp_lport != 0, ("local port not set yet"));
1434         cpu = ntohs(inp->inp_lport) & ncpus2_mask;
1435
1436         lmsg->ms_error = error;
1437         if (cpu != mycpuid) {
1438                 struct lwkt_port *port = netisr_cpuport(cpu);
1439
1440                 /*
1441                  * We are moving the protocol processing port the socket
1442                  * is on, we have to unlink here and re-link on the
1443                  * target cpu.
1444                  */
1445                 in_pcbunlink(inp, &udbinfo[mycpuid]);
1446                 msg->nm_dispatch = udp_inswildcardhash_dispatch;
1447
1448                 /* See the related comment in tcp_usrreq.c tcp_connect() */
1449                 lwkt_setmsg_receipt(lmsg, udp_sosetport);
1450                 lwkt_forwardmsg(port, lmsg);
1451                 return TRUE; /* forwarded */
1452         }
1453
1454         return udp_inswildcardhash_oncpu(inp, msg);
1455 }
1456
1457 static void
1458 udp_bind(netmsg_t msg)
1459 {
1460         struct socket *so = msg->bind.base.nm_so;
1461         struct inpcb *inp;
1462         int error;
1463
1464         inp = so->so_pcb;
1465         if (inp) {
1466                 struct sockaddr *nam = msg->bind.nm_nam;
1467                 struct thread *td = msg->bind.nm_td;
1468
1469                 error = in_pcbbind(inp, nam, td);
1470                 if (error == 0) {
1471                         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1472                         boolean_t forwarded;
1473
1474                         if (sin->sin_addr.s_addr != INADDR_ANY)
1475                                 inp->inp_flags |= INP_WASBOUND_NOTANY;
1476
1477                         forwarded = udp_inswildcardhash(inp,
1478                             &msg->bind.base, 0);
1479                         if (forwarded) {
1480                                 /*
1481                                  * The message is further forwarded, so
1482                                  * we are done here.
1483                                  */
1484                                 return;
1485                         }
1486                 }
1487         } else {
1488                 error = EINVAL;
1489         }
1490         lwkt_replymsg(&msg->bind.base.lmsg, error);
1491 }
1492
1493 static void
1494 udp_connect(netmsg_t msg)
1495 {
1496         struct socket *so = msg->connect.base.nm_so;
1497         struct sockaddr *nam = msg->connect.nm_nam;
1498         struct thread *td = msg->connect.nm_td;
1499         struct inpcb *inp;
1500         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1501         struct sockaddr_in *if_sin;
1502         struct lwkt_port *port;
1503         int error;
1504
1505         KKASSERT(msg->connect.nm_m == NULL);
1506
1507         inp = so->so_pcb;
1508         if (inp == NULL) {
1509                 error = EINVAL;
1510                 goto out;
1511         }
1512
1513         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1514                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1515                 in_pcblink(inp, &udbinfo[mycpuid]);
1516         }
1517
1518         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1519                 error = EISCONN;
1520                 goto out;
1521         }
1522         error = 0;
1523
1524         /*
1525          * Bind if we have to
1526          */
1527         if (inp->inp_lport == 0) {
1528                 error = in_pcbbind(inp, NULL, td);
1529                 if (error)
1530                         goto out;
1531         }
1532
1533         /*
1534          * Calculate the correct protocol processing thread.  The connect
1535          * operation must run there.
1536          */
1537         error = in_pcbladdr(inp, nam, &if_sin, td);
1538         if (error)
1539                 goto out;
1540         if (!prison_remote_ip(td, nam)) {
1541                 error = EAFNOSUPPORT; /* IPv6 only jail */
1542                 goto out;
1543         }
1544
1545         port = udp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1546             inp->inp_laddr.s_addr != INADDR_ANY ?
1547             inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, inp->inp_lport);
1548         if (port != &curthread->td_msgport) {
1549                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1550                 int nm_flags = PRUC_RECONNECT;
1551
1552                 /*
1553                  * in_pcbladdr() may have allocated a route entry for us
1554                  * on the current CPU, but we need a route entry on the
1555                  * inpcb's owner CPU, so free it here.
1556                  */
1557                 in_pcbresetroute(inp);
1558
1559                 if (inp->inp_flags & INP_WILDCARD) {
1560                         /*
1561                          * Remove this inpcb from the wildcard hash before
1562                          * the socket's msgport changes.
1563                          */
1564                         udp_remwildcardhash(inp);
1565                 }
1566
1567                 /*
1568                  * We are moving the protocol processing port the socket
1569                  * is on, we have to unlink here and re-link on the
1570                  * target cpu.
1571                  */
1572                 in_pcbunlink(inp, &udbinfo[mycpuid]);
1573                 msg->connect.nm_flags |= nm_flags;
1574
1575                 /* See the related comment in tcp_usrreq.c tcp_connect() */
1576                 lwkt_setmsg_receipt(lmsg, udp_sosetport);
1577                 lwkt_forwardmsg(port, lmsg);
1578                 /* msg invalid now */
1579                 return;
1580         }
1581         error = udp_connect_oncpu(inp, sin, if_sin);
1582 out:
1583         if (error && inp != NULL && inp->inp_lport != 0 &&
1584             (inp->inp_flags & INP_WILDCARD) == 0) {
1585                 boolean_t forwarded;
1586
1587                 /* Connect failed; put it to wildcard hash. */
1588                 forwarded = udp_inswildcardhash(inp, &msg->connect.base,
1589                     error);
1590                 if (forwarded) {
1591                         /*
1592                          * The message is further forwarded, so we are done
1593                          * here.
1594                          */
1595                         return;
1596                 }
1597         }
1598         lwkt_replymsg(&msg->connect.base.lmsg, error);
1599 }
1600
1601 static void
1602 udp_remwildcardhash(struct inpcb *inp)
1603 {
1604         int cpu;
1605
1606         KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid],
1607             ("not on owner cpu"));
1608
1609         for (cpu = 0; cpu < ncpus2; ++cpu) {
1610                 if (cpu == mycpuid) {
1611                         /*
1612                          * This inpcb will be removed by the later
1613                          * in_pcbremwildcardhash().
1614                          */
1615                         continue;
1616                 }
1617                 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpu]);
1618         }
1619         in_pcbremwildcardhash(inp);
1620 }
1621
1622 static int
1623 udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin,
1624     struct sockaddr_in *if_sin)
1625 {
1626         struct socket *so = inp->inp_socket;
1627         struct inpcb *oinp;
1628
1629         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1630             sin->sin_addr, sin->sin_port,
1631             inp->inp_laddr.s_addr != INADDR_ANY ?
1632             inp->inp_laddr : if_sin->sin_addr, inp->inp_lport, FALSE, NULL);
1633         if (oinp != NULL)
1634                 return EADDRINUSE;
1635
1636         /*
1637          * No more errors can occur, finish adjusting the socket
1638          * and change the processing port to reflect the connected
1639          * socket.  Once set we can no longer safely mess with the
1640          * socket.
1641          */
1642
1643         if (inp->inp_flags & INP_WILDCARD)
1644                 udp_remwildcardhash(inp);
1645
1646         if (inp->inp_laddr.s_addr == INADDR_ANY)
1647                 inp->inp_laddr = if_sin->sin_addr;
1648         inp->inp_faddr = sin->sin_addr;
1649         inp->inp_fport = sin->sin_port;
1650         in_pcbinsconnhash(inp);
1651
1652         soisconnected(so);
1653
1654         return 0;
1655 }
1656
1657 static void
1658 udp_detach2(struct socket *so)
1659 {
1660         in_pcbdetach(so->so_pcb);
1661         sodiscard(so);
1662         sofree(so);
1663 }
1664
1665 static void
1666 udp_detach_final_dispatch(netmsg_t msg)
1667 {
1668         udp_detach2(msg->base.nm_so);
1669 }
1670
1671 static void
1672 udp_detach_oncpu_dispatch(netmsg_t msg)
1673 {
1674         struct netmsg_base *clomsg = &msg->base;
1675         struct socket *so = clomsg->nm_so;
1676         struct inpcb *inp = so->so_pcb;
1677         struct thread *td = curthread;
1678         int nextcpu, cpuid = mycpuid;
1679
1680         KASSERT(td->td_type == TD_TYPE_NETISR, ("not in netisr"));
1681
1682         if (inp->inp_flags & INP_WILDCARD) {
1683                 /*
1684                  * This inp will be removed on the inp's
1685                  * owner CPU later, so don't do it now.
1686                  */
1687                 if (&td->td_msgport != so->so_port)
1688                         in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpuid]);
1689         }
1690
1691         if (cpuid == 0) {
1692                 /*
1693                  * Free and clear multicast socket option,
1694                  * which is only accessed in netisr0.
1695                  */
1696                 ip_freemoptions(inp->inp_moptions);
1697                 inp->inp_moptions = NULL;
1698         }
1699
1700         nextcpu = cpuid + 1;
1701         if (nextcpu < ncpus2) {
1702                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &clomsg->lmsg);
1703         } else {
1704                 /*
1705                  * No one could see this inpcb now; destroy this
1706                  * inpcb in its owner netisr.
1707                  */
1708                 netmsg_init(clomsg, so, &netisr_apanic_rport, 0,
1709                     udp_detach_final_dispatch);
1710                 lwkt_sendmsg(so->so_port, &clomsg->lmsg);
1711         }
1712 }
1713
1714 static void
1715 udp_detach(netmsg_t msg)
1716 {
1717         struct socket *so = msg->detach.base.nm_so;
1718         struct netmsg_base *clomsg;
1719         struct inpcb *inp;
1720
1721         inp = so->so_pcb;
1722         if (inp == NULL) {
1723                 lwkt_replymsg(&msg->detach.base.lmsg, EINVAL);
1724                 return;
1725         }
1726
1727         /*
1728          * Reply EJUSTRETURN ASAP, we will call sodiscard() and
1729          * sofree() later.
1730          */
1731         lwkt_replymsg(&msg->detach.base.lmsg, EJUSTRETURN);
1732
1733         if (ncpus2 == 1) {
1734                 /* Only one CPU, detach the inpcb directly. */
1735                 udp_detach2(so);
1736                 return;
1737         }
1738
1739         /*
1740          * Remove this inpcb from the inpcb list first, so that
1741          * no one could find this inpcb from the inpcb list.
1742          */
1743         in_pcbofflist(inp);
1744
1745         if (inp->inp_flags & INP_DIRECT_DETACH) {
1746                 /*
1747                  * Direct detaching is allowed
1748                  */
1749                 KASSERT((inp->inp_flags & INP_WILDCARD) == 0,
1750                     ("in the wildcardhash"));
1751                 KASSERT(inp->inp_moptions == NULL, ("has mcast options"));
1752                 udp_detach2(so);
1753                 return;
1754         }
1755
1756         /*
1757          * Go through netisrs which process UDP to make sure
1758          * no one could find this inpcb anymore.
1759          */
1760         clomsg = &so->so_clomsg;
1761         netmsg_init(clomsg, so, &netisr_apanic_rport, MSGF_IGNSOPORT,
1762             udp_detach_oncpu_dispatch);
1763         lwkt_sendmsg(netisr_cpuport(0), &clomsg->lmsg);
1764 }
1765
1766 static void
1767 udp_disconnect(netmsg_t msg)
1768 {
1769         struct socket *so = msg->disconnect.base.nm_so;
1770         struct inpcb *inp;
1771         boolean_t forwarded;
1772         int error = 0;
1773
1774         inp = so->so_pcb;
1775         if (inp == NULL) {
1776                 error = EINVAL;
1777                 goto out;
1778         }
1779         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1780                 error = ENOTCONN;
1781                 goto out;
1782         }
1783
1784         soclrstate(so, SS_ISCONNECTED);         /* XXX */
1785
1786         in_pcbdisconnect(inp);
1787
1788         /*
1789          * Follow traditional BSD behavior and retain the local port
1790          * binding.  But, fix the old misbehavior of overwriting any
1791          * previously bound local address.
1792          */
1793         if (!(inp->inp_flags & INP_WASBOUND_NOTANY))
1794                 inp->inp_laddr.s_addr = INADDR_ANY;
1795
1796         if (so->so_state & SS_ISCLOSING) {
1797                 /*
1798                  * If this socket is being closed, there is no need
1799                  * to put this socket back into wildcard hash table.
1800                  */
1801                 error = 0;
1802                 goto out;
1803         }
1804
1805         forwarded = udp_inswildcardhash(inp, &msg->disconnect.base, 0);
1806         if (forwarded) {
1807                 /*
1808                  * The message is further forwarded, so we are done
1809                  * here.
1810                  */
1811                 return;
1812         }
1813 out:
1814         lwkt_replymsg(&msg->disconnect.base.lmsg, error);
1815 }
1816
1817 void
1818 udp_shutdown(netmsg_t msg)
1819 {
1820         struct socket *so = msg->shutdown.base.nm_so;
1821         struct inpcb *inp;
1822         int error;
1823
1824         inp = so->so_pcb;
1825         if (inp) {
1826                 socantsendmore(so);
1827                 error = 0;
1828         } else {
1829                 error = EINVAL;
1830         }
1831         lwkt_replymsg(&msg->shutdown.base.lmsg, error);
1832 }
1833
1834 struct pr_usrreqs udp_usrreqs = {
1835         .pru_abort = udp_abort,
1836         .pru_accept = pr_generic_notsupp,
1837         .pru_attach = udp_attach,
1838         .pru_bind = udp_bind,
1839         .pru_connect = udp_connect,
1840         .pru_connect2 = pr_generic_notsupp,
1841         .pru_control = in_control_dispatch,
1842         .pru_detach = udp_detach,
1843         .pru_disconnect = udp_disconnect,
1844         .pru_listen = pr_generic_notsupp,
1845         .pru_peeraddr = in_setpeeraddr_dispatch,
1846         .pru_rcvd = pr_generic_notsupp,
1847         .pru_rcvoob = pr_generic_notsupp,
1848         .pru_send = udp_send,
1849         .pru_sense = pru_sense_null,
1850         .pru_shutdown = udp_shutdown,
1851         .pru_sockaddr = in_setsockaddr_dispatch,
1852         .pru_sosend = sosendudp,
1853         .pru_soreceive = soreceive
1854 };