socket: introduce SO_RERROR to detect receive buffer overflow
[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_inet6.h"
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/domain.h>
74 #include <sys/proc.h>
75 #include <sys/priv.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/sysctl.h>
80 #include <sys/syslog.h>
81 #include <sys/in_cksum.h>
82 #include <sys/ktr.h>
83 #include <sys/jail.h>
84
85 #include <sys/socketvar2.h>
86 #include <sys/serialize.h>
87
88 #include <machine/stdarg.h>
89
90 #include <net/if.h>
91 #include <net/route.h>
92 #include <net/netmsg2.h>
93 #include <net/netisr2.h>
94
95 #include <netinet/in.h>
96 #include <netinet/in_systm.h>
97 #include <netinet/ip.h>
98 #ifdef INET6
99 #include <netinet/ip6.h>
100 #endif
101 #include <netinet/in_pcb.h>
102 #include <netinet/in_var.h>
103 #include <netinet/ip_var.h>
104 #ifdef INET6
105 #include <netinet6/ip6_var.h>
106 #endif
107 #include <netinet/ip_icmp.h>
108 #include <netinet/icmp_var.h>
109 #include <netinet/udp.h>
110 #include <netinet/udp_var.h>
111
112 #define MSGF_UDP_SEND           MSGF_PROTO1
113
114 #define INP_DIRECT_DETACH       INP_FLAG_PROTO2
115
116 #define UDP_KTR_STRING          "inp=%p"
117 #define UDP_KTR_ARGS            struct inpcb *inp
118
119 #ifndef KTR_UDP
120 #define KTR_UDP                 KTR_ALL
121 #endif
122
123 KTR_INFO_MASTER(udp);
124 KTR_INFO(KTR_UDP, udp, send_beg, 0, UDP_KTR_STRING, UDP_KTR_ARGS);
125 KTR_INFO(KTR_UDP, udp, send_end, 1, UDP_KTR_STRING, UDP_KTR_ARGS);
126 KTR_INFO(KTR_UDP, udp, send_ipout, 2, UDP_KTR_STRING, UDP_KTR_ARGS);
127 KTR_INFO(KTR_UDP, udp, redisp_ipout_beg, 3, UDP_KTR_STRING, UDP_KTR_ARGS);
128 KTR_INFO(KTR_UDP, udp, redisp_ipout_end, 4, UDP_KTR_STRING, UDP_KTR_ARGS);
129 KTR_INFO(KTR_UDP, udp, send_redisp, 5, UDP_KTR_STRING, UDP_KTR_ARGS);
130 KTR_INFO(KTR_UDP, udp, send_inswildcard, 6, UDP_KTR_STRING, UDP_KTR_ARGS);
131
132 #define logudp(name, inp)       KTR_LOG(udp_##name, inp)
133
134 /*
135  * UDP protocol implementation.
136  * Per RFC 768, August, 1980.
137  */
138 #ifndef COMPAT_42
139 static int      udpcksum = 1;
140 #else
141 static int      udpcksum = 0;           /* XXX */
142 #endif
143 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
144     &udpcksum, 0, "Enable checksumming of UDP packets");
145
146 int     log_in_vain = 0;
147 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
148     &log_in_vain, 0, "Log all incoming UDP packets");
149
150 static int      blackhole = 0;
151 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
152         &blackhole, 0, "Do not send port unreachables for refused connects");
153
154 static int      strict_mcast_mship = 1;
155 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
156         &strict_mcast_mship, 0, "Only send multicast to member sockets");
157
158 int     udp_sosend_async = 1;
159 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_async, CTLFLAG_RW,
160         &udp_sosend_async, 0, "UDP asynchronized pru_send");
161
162 int     udp_sosend_prepend = 1;
163 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_prepend, CTLFLAG_RW,
164         &udp_sosend_prepend, 0,
165         "Prepend enough space for proto and link header in pru_send");
166
167 static int udp_reuseport_ext = 1;
168 SYSCTL_INT(_net_inet_udp, OID_AUTO, reuseport_ext, CTLFLAG_RW,
169         &udp_reuseport_ext, 0, "SO_REUSEPORT extension");
170
171 struct  inpcbinfo udbinfo[MAXCPU];
172
173 #ifndef UDBHASHSIZE
174 #define UDBHASHSIZE 16
175 #endif
176 CTASSERT((UDBHASHSIZE & (UDBHASHSIZE - 1)) == 0);
177
178 struct  udpstat udpstat_percpu[MAXCPU] __cachealign;
179
180 static void udp_append(struct inpcb *last, struct ip *ip,
181     struct mbuf *n, int off, struct sockaddr_in *udp_in);
182
183 static int udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin,
184     struct sockaddr_in *if_sin, uint16_t hash);
185
186 static boolean_t udp_inswildcardhash(struct inpcb *inp,
187     struct netmsg_base *msg, int error);
188 static void udp_remwildcardhash(struct inpcb *inp);
189
190 static __inline int
191 udp_lportcpu(short lport)
192 {
193         return (ntohs(lport) % netisr_ncpus);
194 }
195
196 void
197 udp_init(void)
198 {
199         struct inpcbportinfo *portinfo;
200         int cpu;
201
202         portinfo = kmalloc(sizeof(*portinfo) * netisr_ncpus, M_PCB,
203                            M_WAITOK | M_CACHEALIGN);
204
205         for (cpu = 0; cpu < netisr_ncpus; cpu++) {
206                 struct inpcbinfo *uicb = &udbinfo[cpu];
207
208                 /*
209                  * NOTE:
210                  * UDP pcb list, wildcard hash table and localgroup hash
211                  * table are shared.
212                  */
213                 in_pcbinfo_init(uicb, cpu, TRUE);
214                 uicb->hashbase = hashinit(UDBHASHSIZE, M_PCB, &uicb->hashmask);
215
216                 in_pcbportinfo_init(&portinfo[cpu], UDBHASHSIZE, cpu);
217                 in_pcbportinfo_set(uicb, portinfo, netisr_ncpus);
218
219                 uicb->wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB,
220                     &uicb->wildcardhashmask);
221                 uicb->localgrphashbase = hashinit(UDBHASHSIZE, M_PCB,
222                     &uicb->localgrphashmask);
223
224                 uicb->ipi_size = sizeof(struct inpcb);
225         }
226
227         /*
228          * Initialize UDP statistics counters for each CPU.
229          */
230         for (cpu = 0; cpu < netisr_ncpus; ++cpu)
231                 bzero(&udpstat_percpu[cpu], sizeof(struct udpstat));
232 }
233
234 static int
235 sysctl_udpstat(SYSCTL_HANDLER_ARGS)
236 {
237         int cpu, error = 0;
238
239         for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
240                 if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu],
241                                         sizeof(struct udpstat))))
242                         break;
243                 if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu],
244                                        sizeof(struct udpstat))))
245                         break;
246         }
247
248         return (error);
249 }
250 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
251     0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics");
252
253 void
254 udp_ctloutput(netmsg_t msg)
255 {
256         struct socket *so = msg->base.nm_so;
257         struct sockopt *sopt = msg->ctloutput.nm_sopt;
258         struct inpcb *inp = so->so_pcb;
259
260         if (inp == NULL) {
261                 lwkt_replymsg(&msg->lmsg, EINVAL);
262                 return;
263         }
264
265         if (sopt->sopt_level == IPPROTO_IP && sopt->sopt_dir == SOPT_SET) {
266                 switch (sopt->sopt_name) {
267                 case IP_MULTICAST_IF:
268                 case IP_MULTICAST_VIF:
269                 case IP_MULTICAST_TTL:
270                 case IP_MULTICAST_LOOP:
271                 case IP_ADD_MEMBERSHIP:
272                 case IP_DROP_MEMBERSHIP:
273                         /*
274                          * This pr_ctloutput msg will be forwarded
275                          * to netisr0 to run; we can't do direct
276                          * detaching anymore.
277                          *
278                          * NOTE:
279                          * Don't optimize for the sockets whose
280                          * current so_port is netisr0's msgport.
281                          * These sockets could be connect(2)'ed
282                          * later and the so_port will be changed.
283                          */
284                         inp->inp_flags &= ~INP_DIRECT_DETACH;
285                         break;
286                 }
287         }
288         return ip_ctloutput(msg);
289 }
290
291 /*
292  * Check multicast packets to make sure they are only sent to sockets with
293  * multicast memberships for the packet's destination address and arrival
294  * interface.  Multicast packets to multicast-unaware sockets are also
295  * disallowed.
296  *
297  * Returns 0 if the packet is acceptable, -1 if it is not.
298  */
299 static __inline int
300 check_multicast_membership(const struct ip *ip, const struct inpcb *inp,
301     const struct mbuf *m)
302 {
303         const struct ip_moptions *mopt;
304         int mshipno;
305
306         if (strict_mcast_mship == 0 ||
307             !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
308                 return (0);
309         }
310
311         ASSERT_NETISR0;
312
313         mopt = inp->inp_moptions;
314         if (mopt == NULL)
315                 return (-1);
316         for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) {
317                 const struct in_multi *maddr = mopt->imo_membership[mshipno];
318
319                 if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr &&
320                     m->m_pkthdr.rcvif == maddr->inm_ifp) {
321                         return (0);
322                 }
323         }
324         return (-1);
325 }
326
327 struct udp_mcast_arg {
328         struct inpcb    *inp;
329         struct inpcb    *last;
330         struct ip       *ip;
331         struct mbuf     *m;
332         int             iphlen;
333         struct sockaddr_in *udp_in;
334 };
335
336 static int
337 udp_mcast_input(struct udp_mcast_arg *arg)
338 {
339         struct inpcb *inp = arg->inp;
340         struct inpcb *last = arg->last;
341         struct ip *ip = arg->ip;
342         struct mbuf *m = arg->m;
343
344         if (check_multicast_membership(ip, inp, m) < 0)
345                 return ERESTART; /* caller continue */
346
347         if (last != NULL) {
348                 struct mbuf *n;
349
350                 if ((n = m_copypacket(m, M_NOWAIT)) != NULL)
351                         udp_append(last, ip, n,
352                             arg->iphlen + sizeof(struct udphdr),
353                             arg->udp_in);
354         }
355         arg->last = last = inp;
356
357         /*
358          * Don't look for additional matches if this one does
359          * not have either the SO_REUSEPORT or SO_REUSEADDR
360          * socket options set.  This heuristic avoids searching
361          * through all pcbs in the common case of a non-shared
362          * port.  It * assumes that an application will never
363          * clear these options after setting them.
364          */
365         if (!(last->inp_socket->so_options &
366             (SO_REUSEPORT | SO_REUSEADDR)))
367                 return EJUSTRETURN; /* caller stop */
368         return 0;
369 }
370
371 int
372 udp_input(struct mbuf **mp, int *offp, int proto)
373 {
374         struct sockaddr_in udp_in = { sizeof udp_in, AF_INET };
375         int iphlen;
376         struct ip *ip;
377         struct udphdr *uh;
378         struct inpcb *inp;
379         struct mbuf *m;
380         struct mbuf *opts = NULL;
381         int len, off;
382         struct ip save_ip;
383         struct inpcbinfo *pcbinfo = &udbinfo[mycpuid];
384
385         off = *offp;
386         m = *mp;
387         *mp = NULL;
388
389         iphlen = off;
390         udp_stat.udps_ipackets++;
391
392         /*
393          * Strip IP options, if any; should skip this,
394          * make available to user, and use on returned packets,
395          * but we don't yet have a way to check the checksum
396          * with options still present.
397          */
398         if (iphlen > sizeof(struct ip)) {
399                 ip_stripoptions(m);
400                 iphlen = sizeof(struct ip);
401         }
402
403         /*
404          * IP and UDP headers are together in first mbuf.
405          * Already checked and pulled up in ip_demux().
406          */
407         KASSERT(m->m_len >= iphlen + sizeof(struct udphdr),
408             ("UDP header not in one mbuf"));
409
410         ip = mtod(m, struct ip *);
411         uh = (struct udphdr *)((caddr_t)ip + iphlen);
412
413         /* destination port of 0 is illegal, based on RFC768. */
414         if (uh->uh_dport == 0)
415                 goto bad;
416
417         /*
418          * Make mbuf data length reflect UDP length.
419          * If not enough data to reflect UDP length, drop.
420          */
421         len = ntohs((u_short)uh->uh_ulen);
422         if (ip->ip_len != len) {
423                 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
424                         udp_stat.udps_badlen++;
425                         goto bad;
426                 }
427                 m_adj(m, len - ip->ip_len);
428                 /* ip->ip_len = len; */
429         }
430         /*
431          * Save a copy of the IP header in case we want restore it
432          * for sending an ICMP error message in response.
433          */
434         save_ip = *ip;
435
436         /*
437          * Checksum extended UDP header and data.
438          */
439         if (uh->uh_sum) {
440                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
441                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
442                                 uh->uh_sum = m->m_pkthdr.csum_data;
443                         else
444                                 uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
445                                     ip->ip_dst.s_addr, htonl((u_short)len +
446                                     m->m_pkthdr.csum_data + IPPROTO_UDP));
447                         uh->uh_sum ^= 0xffff;
448                 } else {
449                         char b[9];
450
451                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
452                         bzero(((struct ipovly *)ip)->ih_x1, 9);
453                         ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
454                         uh->uh_sum = in_cksum(m, len + sizeof(struct ip));
455                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
456                 }
457                 if (uh->uh_sum) {
458                         udp_stat.udps_badsum++;
459                         m_freem(m);
460                         return(IPPROTO_DONE);
461                 }
462         } else
463                 udp_stat.udps_nosum++;
464
465         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
466             in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
467                 struct inpcbhead *connhead;
468                 struct inpcontainer *ic, *ic_marker;
469                 struct inpcontainerhead *ichead;
470                 struct udp_mcast_arg arg;
471                 struct inpcb *last;
472                 int error;
473
474                 /*
475                  * Deliver a multicast or broadcast datagram to *all* sockets
476                  * for which the local and remote addresses and ports match
477                  * those of the incoming datagram.  This allows more than
478                  * one process to receive multi/broadcasts on the same port.
479                  * (This really ought to be done for unicast datagrams as
480                  * well, but that would cause problems with existing
481                  * applications that open both address-specific sockets and
482                  * a wildcard socket listening to the same port -- they would
483                  * end up receiving duplicates of every unicast datagram.
484                  * Those applications open the multiple sockets to overcome an
485                  * inadequacy of the UDP socket interface, but for backwards
486                  * compatibility we avoid the problem here rather than
487                  * fixing the interface.  Maybe 4.5BSD will remedy this?)
488                  */
489
490                 /*
491                  * Construct sockaddr format source address.
492                  */
493                 udp_in.sin_port = uh->uh_sport;
494                 udp_in.sin_addr = ip->ip_src;
495                 arg.udp_in = &udp_in;
496                 /*
497                  * Locate pcb(s) for datagram.
498                  * (Algorithm copied from raw_intr().)
499                  */
500                 last = NULL;
501                 arg.iphlen = iphlen;
502
503                 connhead = &pcbinfo->hashbase[
504                     INP_PCBCONNHASH(ip->ip_src.s_addr, uh->uh_sport,
505                     ip->ip_dst.s_addr, uh->uh_dport, pcbinfo->hashmask)];
506                 LIST_FOREACH(inp, connhead, inp_hash) {
507 #ifdef INET6
508                         if (!INP_ISIPV4(inp))
509                                 continue;
510 #endif
511                         if (!in_hosteq(inp->inp_faddr, ip->ip_src) ||
512                             !in_hosteq(inp->inp_laddr, ip->ip_dst) ||
513                             inp->inp_fport != uh->uh_sport ||
514                             inp->inp_lport != uh->uh_dport)
515                                 continue;
516
517                         arg.inp = inp;
518                         arg.last = last;
519                         arg.ip = ip;
520                         arg.m = m;
521
522                         error = udp_mcast_input(&arg);
523                         if (error == ERESTART)
524                                 continue;
525                         last = arg.last;
526
527                         if (error == EJUSTRETURN)
528                                 goto done;
529                 }
530
531                 ichead = &pcbinfo->wildcardhashbase[
532                     INP_PCBWILDCARDHASH(uh->uh_dport,
533                     pcbinfo->wildcardhashmask)];
534                 ic_marker = in_pcbcontainer_marker();
535
536                 GET_PCBINFO_TOKEN(pcbinfo);
537                 LIST_INSERT_HEAD(ichead, ic_marker, ic_list);
538                 while ((ic = LIST_NEXT(ic_marker, ic_list)) != NULL) {
539                         LIST_REMOVE(ic_marker, ic_list);
540                         LIST_INSERT_AFTER(ic, ic_marker, ic_list);
541
542                         inp = ic->ic_inp;
543                         if (inp->inp_flags & INP_PLACEMARKER)
544                                 continue;
545 #ifdef INET6
546                         if (!INP_ISIPV4(inp))
547                                 continue;
548 #endif
549                         if (inp->inp_lport != uh->uh_dport)
550                                 continue;
551                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
552                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
553                                 continue;
554
555                         arg.inp = inp;
556                         arg.last = last;
557                         arg.ip = ip;
558                         arg.m = m;
559
560                         error = udp_mcast_input(&arg);
561                         if (error == ERESTART)
562                                 continue;
563                         last = arg.last;
564
565                         if (error == EJUSTRETURN)
566                                 break;
567                 }
568                 LIST_REMOVE(ic_marker, ic_list);
569                 REL_PCBINFO_TOKEN(pcbinfo);
570 done:
571                 if (last == NULL) {
572                         /*
573                          * No matching pcb found; discard datagram.
574                          * (No need to send an ICMP Port Unreachable
575                          * for a broadcast or multicast datgram.)
576                          */
577                         udp_stat.udps_noportbcast++;
578                         goto bad;
579                 }
580                 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
581                     &udp_in);
582                 return(IPPROTO_DONE);
583         }
584         /*
585          * Locate pcb for datagram.
586          */
587         inp = in_pcblookup_pkthash(pcbinfo, ip->ip_src, uh->uh_sport,
588             ip->ip_dst, uh->uh_dport, TRUE, m->m_pkthdr.rcvif,
589             udp_reuseport_ext ? m : NULL);
590         if (inp == NULL) {
591                 if (log_in_vain) {
592                         char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
593
594                         log(LOG_INFO,
595                             "Connection attempt to UDP %s:%d from %s:%d\n",
596                             kinet_ntoa(ip->ip_dst, dst), ntohs(uh->uh_dport),
597                             kinet_ntoa(ip->ip_src, src), ntohs(uh->uh_sport));
598                 }
599                 udp_stat.udps_noport++;
600                 if (m->m_flags & (M_BCAST | M_MCAST)) {
601                         udp_stat.udps_noportbcast++;
602                         goto bad;
603                 }
604                 if (blackhole)
605                         goto bad;
606 #ifdef ICMP_BANDLIM
607                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
608                         goto bad;
609 #endif
610                 *ip = save_ip;
611                 ip->ip_len += iphlen;
612                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
613                 return(IPPROTO_DONE);
614         }
615         KASSERT(INP_ISIPV4(inp), ("not inet inpcb"));
616         /*
617          * Check the minimum TTL for socket.
618          */
619         if (ip->ip_ttl < inp->inp_ip_minttl)
620                 goto bad;
621
622         /*
623          * Construct sockaddr format source address.
624          * Stuff source address and datagram in user buffer.
625          */
626         udp_in.sin_port = uh->uh_sport;
627         udp_in.sin_addr = ip->ip_src;
628         if ((inp->inp_flags & INP_CONTROLOPTS) ||
629             (inp->inp_socket->so_options & SO_TIMESTAMP))
630                 ip_savecontrol(inp, &opts, ip, m);
631         m_adj(m, iphlen + sizeof(struct udphdr));
632
633         lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token);
634         if (ssb_appendaddr(&inp->inp_socket->so_rcv,
635             (struct sockaddr *)&udp_in, m, opts) == 0) {
636                 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
637                 udp_stat.udps_fullsock++;
638                 soroverflow(inp->inp_socket);
639                 goto bad;
640         }
641         lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token);
642         sorwakeup(inp->inp_socket);
643         return(IPPROTO_DONE);
644 bad:
645         m_freem(m);
646         if (opts)
647                 m_freem(opts);
648         return(IPPROTO_DONE);
649 }
650
651 /*
652  * subroutine of udp_input(), mainly for source code readability.
653  * caller must properly init udp_ip6 and udp_in6 beforehand.
654  */
655 static void
656 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off,
657     struct sockaddr_in *udp_in)
658 {
659         struct mbuf *opts = NULL;
660         int ret;
661
662         KASSERT(INP_ISIPV4(last), ("not inet inpcb"));
663
664         if (last->inp_flags & INP_CONTROLOPTS ||
665             last->inp_socket->so_options & SO_TIMESTAMP)
666                 ip_savecontrol(last, &opts, ip, n);
667         m_adj(n, off);
668
669         lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token);
670         ret = ssb_appendaddr(&last->inp_socket->so_rcv,
671             (struct sockaddr *)udp_in, n, opts);
672         lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token);
673         if (ret == 0) {
674                 m_freem(n);
675                 if (opts)
676                         m_freem(opts);
677                 udp_stat.udps_fullsock++;
678         } else {
679                 sorwakeup(last->inp_socket);
680         }
681 }
682
683 /*
684  * Notify a udp user of an asynchronous error;
685  * just wake up so that he can collect error status.
686  */
687 void
688 udp_notify(struct inpcb *inp, int error)
689 {
690         inp->inp_socket->so_error = error;
691         sorwakeup(inp->inp_socket);
692         sowwakeup(inp->inp_socket);
693 }
694
695 struct netmsg_udp_notify {
696         struct netmsg_base base;
697         inp_notify_t    nm_notify;
698         struct in_addr  nm_faddr;
699         int             nm_arg;
700 };
701
702 static void
703 udp_notifyall_oncpu(netmsg_t msg)
704 {
705         struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg;
706         int nextcpu, cpu = mycpuid;
707
708         ASSERT_NETISR_NCPUS(cpu);
709
710         in_pcbnotifyall(&udbinfo[cpu], nm->nm_faddr, nm->nm_arg, nm->nm_notify);
711
712         nextcpu = cpu + 1;
713         if (nextcpu < netisr_ncpus)
714                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
715         else
716                 lwkt_replymsg(&nm->base.lmsg, 0);
717 }
718
719 inp_notify_t
720 udp_get_inpnotify(int cmd, const struct sockaddr *sa,
721     struct ip **ip0, int *cpuid)
722 {
723         struct in_addr faddr;
724         struct ip *ip = *ip0;
725         inp_notify_t notify = udp_notify;
726
727         faddr = ((const struct sockaddr_in *)sa)->sin_addr;
728         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
729                 return NULL;
730
731         if (PRC_IS_REDIRECT(cmd)) {
732                 ip = NULL;
733                 notify = in_rtchange;
734         } else if (cmd == PRC_HOSTDEAD) {
735                 ip = NULL;
736         } else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
737                 return NULL;
738         }
739
740         if (cpuid != NULL) {
741                 if (ip == NULL) {
742                         /* Go through all effective netisr CPUs. */
743                         *cpuid = netisr_ncpus;
744                 } else {
745                         const struct udphdr *uh;
746
747                         uh = (const struct udphdr *)
748                             ((caddr_t)ip + (ip->ip_hl << 2));
749                         *cpuid = udp_addrcpu(faddr.s_addr, uh->uh_dport,
750                             ip->ip_src.s_addr, uh->uh_sport);
751                 }
752         }
753
754         *ip0 = ip;
755         return notify;
756 }
757
758 void
759 udp_ctlinput(netmsg_t msg)
760 {
761         struct sockaddr *sa = msg->ctlinput.nm_arg;
762         struct ip *ip = msg->ctlinput.nm_extra;
763         int cmd = msg->ctlinput.nm_cmd, cpuid;
764         inp_notify_t notify;
765         struct in_addr faddr;
766
767         ASSERT_NETISR_NCPUS(mycpuid);
768
769         notify = udp_get_inpnotify(cmd, sa, &ip, &cpuid);
770         if (notify == NULL)
771                 goto done;
772
773         faddr = ((struct sockaddr_in *)sa)->sin_addr;
774         if (ip) {
775                 const struct udphdr *uh;
776                 struct inpcb *inp;
777
778                 if (cpuid != mycpuid)
779                         goto done;
780
781                 uh = (const struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
782                 inp = in_pcblookup_hash(&udbinfo[mycpuid], faddr, uh->uh_dport,
783                                         ip->ip_src, uh->uh_sport, 0, NULL);
784                 if (inp != NULL && inp->inp_socket != NULL)
785                         notify(inp, inetctlerrmap[cmd]);
786         } else if (msg->ctlinput.nm_direct) {
787                 if (cpuid != netisr_ncpus && cpuid != mycpuid)
788                         goto done;
789
790                 in_pcbnotifyall(&udbinfo[mycpuid], faddr, inetctlerrmap[cmd],
791                     notify);
792         } else {
793                 struct netmsg_udp_notify *nm;
794
795                 ASSERT_NETISR0;
796                 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
797                 netmsg_init(&nm->base, NULL, &netisr_afree_rport,
798                             0, udp_notifyall_oncpu);
799                 nm->nm_faddr = faddr;
800                 nm->nm_arg = inetctlerrmap[cmd];
801                 nm->nm_notify = notify;
802                 lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
803         }
804 done:
805         lwkt_replymsg(&msg->lmsg, 0);
806 }
807
808 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, udbinfo, 0,
809             in_pcblist_ncpus, "S,xinpcb", "List of active UDP sockets");
810
811 static int
812 udp_getcred(SYSCTL_HANDLER_ARGS)
813 {
814         struct sockaddr_in addrs[2];
815         struct ucred cred0, *cred = NULL;
816         struct inpcb *inp;
817         int error, cpu, origcpu;
818
819         error = priv_check(req->td, PRIV_ROOT);
820         if (error)
821                 return (error);
822         error = SYSCTL_IN(req, addrs, sizeof addrs);
823         if (error)
824                 return (error);
825
826         origcpu = mycpuid;
827         cpu = udp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
828             addrs[0].sin_addr.s_addr, addrs[0].sin_port);
829
830         lwkt_migratecpu(cpu);
831
832         inp = in_pcblookup_hash(&udbinfo[cpu],
833             addrs[1].sin_addr, addrs[1].sin_port,
834             addrs[0].sin_addr, addrs[0].sin_port, TRUE, NULL);
835         if (inp == NULL || inp->inp_socket == NULL) {
836                 error = ENOENT;
837         } else if (inp->inp_socket->so_cred != NULL) {
838                 cred0 = *(inp->inp_socket->so_cred);
839                 cred = &cred0;
840         }
841
842         lwkt_migratecpu(origcpu);
843
844         if (error)
845                 return error;
846
847         return SYSCTL_OUT(req, cred, sizeof(struct ucred));
848 }
849 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
850     0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection");
851
852 static void
853 udp_send_redispatch(netmsg_t msg)
854 {
855         struct mbuf *m = msg->send.nm_m;
856         int pru_flags = msg->send.nm_flags;
857         struct inpcb *inp = msg->send.base.nm_so->so_pcb;
858         struct mbuf *m_opt = msg->send.nm_control; /* XXX save ipopt */
859         int flags = msg->send.nm_priv; /* ip_output flags */
860         int error;
861
862         logudp(redisp_ipout_beg, inp);
863
864         /*
865          * - Don't use inp route cache.  It should only be used in the
866          *   inp owner netisr.
867          * - Access to inp_moptions should be safe, since multicast UDP
868          *   datagrams are redispatched to netisr0 and inp_moptions is
869          *   changed only in netisr0.
870          */
871         error = ip_output(m, m_opt, NULL, flags, inp->inp_moptions, inp);
872         if ((pru_flags & PRUS_NOREPLY) == 0)
873                 lwkt_replymsg(&msg->send.base.lmsg, error);
874
875         if (m_opt != NULL) {
876                 /* Free saved ip options, if any */
877                 m_freem(m_opt);
878         }
879
880         logudp(redisp_ipout_end, inp);
881 }
882
883 static void
884 udp_send(netmsg_t msg)
885 {
886         struct socket *so = msg->send.base.nm_so;
887         struct mbuf *m = msg->send.nm_m;
888         struct sockaddr *dstaddr = msg->send.nm_addr;
889         int pru_flags = msg->send.nm_flags;
890         struct inpcb *inp = so->so_pcb;
891         struct thread *td = msg->send.nm_td;
892         uint16_t hash;
893         int flags;
894
895         struct udpiphdr *ui;
896         int len = m->m_pkthdr.len;
897         struct sockaddr_in *sin;        /* really is initialized before use */
898         int error = 0, cpu;
899
900         KKASSERT(msg->send.nm_control == NULL);
901
902         logudp(send_beg, inp);
903
904         if (inp == NULL) {
905                 error = EINVAL;
906                 goto release;
907         }
908
909         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
910                 error = EMSGSIZE;
911                 goto release;
912         }
913
914         if (inp->inp_lport == 0) {      /* unbound socket */
915                 boolean_t forwarded;
916
917                 error = in_pcbbind(inp, NULL, td);
918                 if (error)
919                         goto release;
920
921                 /*
922                  * Need to call udp_send again, after this inpcb is
923                  * inserted into wildcard hash table.
924                  */
925                 msg->send.base.lmsg.ms_flags |= MSGF_UDP_SEND;
926                 forwarded = udp_inswildcardhash(inp, &msg->send.base, 0);
927                 if (forwarded) {
928                         /*
929                          * The message is further forwarded, so we are
930                          * done here.
931                          */
932                         logudp(send_inswildcard, inp);
933                         return;
934                 }
935         }
936
937         if (dstaddr != NULL) {          /* destination address specified */
938                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
939                         /* already connected */
940                         error = EISCONN;
941                         goto release;
942                 }
943                 sin = (struct sockaddr_in *)dstaddr;
944                 if (!prison_remote_ip(td, (struct sockaddr *)&sin)) {
945                         error = EAFNOSUPPORT; /* IPv6 only jail */
946                         goto release;
947                 }
948         } else {
949                 if (inp->inp_faddr.s_addr == INADDR_ANY) {
950                         /* no destination specified and not already connected */
951                         error = ENOTCONN;
952                         goto release;
953                 }
954                 sin = NULL;
955         }
956
957         /*
958          * Calculate data length and get a mbuf
959          * for UDP and IP headers.
960          */
961         M_PREPEND(m, sizeof(struct udpiphdr), M_NOWAIT);
962         if (m == NULL) {
963                 error = ENOBUFS;
964                 goto release;
965         }
966
967         /*
968          * Fill in mbuf with extended UDP header
969          * and addresses and length put into network format.
970          */
971         ui = mtod(m, struct udpiphdr *);
972         bzero(ui->ui_x1, sizeof ui->ui_x1);     /* XXX still needed? */
973         ui->ui_pr = IPPROTO_UDP;
974
975         /*
976          * Set destination address.
977          */
978         if (dstaddr != NULL) {                  /* use specified destination */
979                 ui->ui_dst = sin->sin_addr;
980                 ui->ui_dport = sin->sin_port;
981         } else {                                /* use connected destination */
982                 ui->ui_dst = inp->inp_faddr;
983                 ui->ui_dport = inp->inp_fport;
984         }
985
986         /*
987          * Set source address.
988          */
989         if (inp->inp_laddr.s_addr == INADDR_ANY ||
990             IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) {
991                 struct sockaddr_in *if_sin;
992
993                 if (dstaddr == NULL) {  
994                         /*
995                          * connect() had (or should have) failed because
996                          * the interface had no IP address, but the
997                          * application proceeded to call send() anyways.
998                          */
999                         error = ENOTCONN;
1000                         goto release;
1001                 }
1002
1003                 /* Look up outgoing interface. */
1004                 error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1);
1005                 if (error)
1006                         goto release;
1007                 ui->ui_src = if_sin->sin_addr;  /* use address of interface */
1008         } else {
1009                 ui->ui_src = inp->inp_laddr;    /* use non-null bound address */
1010         }
1011         ui->ui_sport = inp->inp_lport;
1012         KASSERT(inp->inp_lport != 0, ("inp lport should have been bound"));
1013
1014         /*
1015          * Release the original thread, since it is no longer used
1016          */
1017         if (pru_flags & PRUS_HELDTD) {
1018                 lwkt_rele(td);
1019                 pru_flags &= ~PRUS_HELDTD;
1020         }
1021         /*
1022          * Free the dest address, since it is no longer needed
1023          */
1024         if (pru_flags & PRUS_FREEADDR) {
1025                 kfree(dstaddr, M_SONAME);
1026                 pru_flags &= ~PRUS_FREEADDR;
1027         }
1028
1029         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1030
1031         /*
1032          * Set up checksum and output datagram.
1033          */
1034         if (udpcksum) {
1035                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr,
1036                     htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1037                 m->m_pkthdr.csum_flags = CSUM_UDP;
1038                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1039                 m->m_pkthdr.csum_thlen = sizeof(struct udphdr);
1040         } else {
1041                 ui->ui_sum = 0;
1042         }
1043         ((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len;
1044         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1045         ((struct ip *)ui)->ip_tos = inp->inp_ip_tos;    /* XXX */
1046         udp_stat.udps_opackets++;
1047
1048         flags = IP_DEBUGROUTE |
1049             (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST));
1050         if (pru_flags & PRUS_DONTROUTE)
1051                 flags |= SO_DONTROUTE;
1052
1053         if (inp->inp_flags & INP_CONNECTED) {
1054                 /*
1055                  * For connected socket, this datagram has already
1056                  * been in the correct netisr; no need to rehash.
1057                  */
1058                 KASSERT(inp->inp_flags & INP_HASH, ("inpcb has no hash"));
1059                 m_sethash(m, inp->inp_hashval);
1060                 goto sendit;
1061         }
1062
1063         hash = udp_addrhash(ui->ui_dst.s_addr, ui->ui_dport,
1064             ui->ui_src.s_addr, ui->ui_sport);
1065         m_sethash(m, hash);
1066
1067         cpu = netisr_hashcpu(hash);
1068         if (cpu != mycpuid) {
1069                 struct mbuf *m_opt = NULL;
1070                 struct netmsg_pru_send *smsg;
1071                 struct lwkt_port *port = netisr_cpuport(cpu);
1072
1073                 /*
1074                  * Not on the CPU that matches this UDP datagram hash;
1075                  * redispatch to the correct CPU to do the ip_output().
1076                  */
1077                 if (inp->inp_options != NULL) {
1078                         /*
1079                          * If there are ip options, then save a copy,
1080                          * since accessing inp_options on other CPUs'
1081                          * is not safe.
1082                          *
1083                          * XXX optimize this?
1084                          */
1085                         m_opt = m_copym(inp->inp_options, 0, M_COPYALL,
1086                             M_WAITOK);
1087                 }
1088                 if ((pru_flags & PRUS_NOREPLY) == 0) {
1089                         /*
1090                          * Change some parts of the original netmsg and
1091                          * forward it to the target netisr.
1092                          *
1093                          * NOTE: so_port MUST NOT be checked in the target
1094                          * netisr.
1095                          */
1096                         smsg = &msg->send;
1097                         smsg->nm_priv = flags; /* ip_output flags */
1098                         smsg->nm_m = m;
1099                         smsg->nm_control = m_opt; /* XXX save ipopt */
1100                         smsg->base.lmsg.ms_flags |= MSGF_IGNSOPORT;
1101                         smsg->base.nm_dispatch = udp_send_redispatch;
1102                         lwkt_forwardmsg(port, &smsg->base.lmsg);
1103                 } else {
1104                         /*
1105                          * Recreate the netmsg, since the original mbuf
1106                          * could have been changed.  And send it to the
1107                          * target netisr.
1108                          *
1109                          * NOTE: so_port MUST NOT be checked in the target
1110                          * netisr.
1111                          */
1112                         smsg = &m->m_hdr.mh_sndmsg;
1113                         netmsg_init(&smsg->base, so, &netisr_apanic_rport,
1114                             MSGF_IGNSOPORT, udp_send_redispatch);
1115                         smsg->nm_priv = flags; /* ip_output flags */
1116                         smsg->nm_flags = pru_flags;
1117                         smsg->nm_m = m;
1118                         smsg->nm_control = m_opt; /* XXX save ipopt */
1119                         lwkt_sendmsg(port, &smsg->base.lmsg);
1120                 }
1121
1122                 /* This UDP datagram is redispatched; done */
1123                 logudp(send_redisp, inp);
1124                 return;
1125         }
1126
1127 sendit:
1128         logudp(send_ipout, inp);
1129         error = ip_output(m, inp->inp_options, &inp->inp_route, flags,
1130             inp->inp_moptions, inp);
1131         m = NULL;
1132
1133 release:
1134         if (m != NULL)
1135                 m_freem(m);
1136
1137         if (pru_flags & PRUS_HELDTD)
1138                 lwkt_rele(td);
1139         if (pru_flags & PRUS_FREEADDR)
1140                 kfree(dstaddr, M_SONAME);
1141         if ((pru_flags & PRUS_NOREPLY) == 0)
1142                 lwkt_replymsg(&msg->send.base.lmsg, error);
1143
1144         logudp(send_end, inp);
1145 }
1146
1147 u_long  udp_sendspace = 9216;           /* really max datagram size */
1148                                         /* 40 1K datagrams */
1149 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
1150     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
1151
1152 u_long  udp_recvspace = 40 * (1024 +
1153 #ifdef INET6
1154                                       sizeof(struct sockaddr_in6)
1155 #else
1156                                       sizeof(struct sockaddr_in)
1157 #endif
1158                                       );
1159 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1160     &udp_recvspace, 0, "Maximum incoming UDP datagram size");
1161
1162 /*
1163  * This should never happen, since UDP socket does not support
1164  * connection acception (SO_ACCEPTCONN, i.e. listen(2)).
1165  */
1166 static void
1167 udp_abort(netmsg_t msg __unused)
1168 {
1169         panic("udp_abort is called");
1170 }
1171
1172 static int
1173 udp_preattach(struct socket *so, int proto __unused, struct pru_attach_info *ai)
1174 {
1175         return soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit);
1176 }
1177
1178 static void
1179 udp_attach(netmsg_t msg)
1180 {
1181         struct socket *so = msg->attach.base.nm_so;
1182         struct pru_attach_info *ai = msg->attach.nm_ai;
1183         struct inpcb *inp;
1184         int error;
1185
1186         KASSERT(so->so_pcb == NULL, ("udp socket attached"));
1187
1188         if (ai != NULL) {
1189                 error = udp_preattach(so, 0 /* don't care */, ai);
1190                 if (error)
1191                         goto out;
1192         } else {
1193                 /* Post attach; do nothing */
1194         }
1195
1196         error = in_pcballoc(so, &udbinfo[mycpuid]);
1197         if (error)
1198                 goto out;
1199
1200         inp = so->so_pcb;
1201         inp->inp_flags |= INP_DIRECT_DETACH;
1202         inp->inp_ip_ttl = ip_defttl;
1203         error = 0;
1204 out:
1205         lwkt_replymsg(&msg->attach.base.lmsg, error);
1206 }
1207
1208 static void
1209 udp_inswildcard_replymsg(netmsg_t msg)
1210 {
1211         lwkt_msg_t lmsg = &msg->lmsg;
1212
1213         if (lmsg->ms_flags & MSGF_UDP_SEND) {
1214                 udp_send(msg);
1215                 /* msg is replied by udp_send() */
1216         } else {
1217                 lwkt_replymsg(lmsg, lmsg->ms_error);
1218         }
1219 }
1220
1221 static void
1222 udp_soreuseport_dispatch(netmsg_t msg)
1223 {
1224         /* This inpcb has already been in the wildcard hash. */
1225         in_pcblink_flags(msg->base.nm_so->so_pcb, &udbinfo[mycpuid], 0);
1226         udp_inswildcard_replymsg(msg);
1227 }
1228
1229 static void
1230 udp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
1231 {
1232         sosetport(((struct netmsg_base *)msg)->nm_so, port);
1233 }
1234
1235 static boolean_t
1236 udp_inswildcardhash_oncpu(struct inpcb *inp, struct netmsg_base *msg)
1237 {
1238         int cpu;
1239
1240         KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid],
1241             ("not on owner cpu"));
1242
1243         in_pcbinswildcardhash(inp);
1244         for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
1245                 if (cpu == mycpuid) {
1246                         /*
1247                          * This inpcb has been inserted by the above
1248                          * in_pcbinswildcardhash().
1249                          */
1250                         continue;
1251                 }
1252                 in_pcbinswildcardhash_oncpu(inp, &udbinfo[cpu]);
1253         }
1254
1255         /* NOTE: inp_lgrpindex is _not_ assigned in jail. */
1256         if ((inp->inp_socket->so_options & SO_REUSEPORT) &&
1257             inp->inp_lgrpindex >= 0) {
1258                 /*
1259                  * For SO_REUSEPORT socket, redistribute it based on its
1260                  * local group index.
1261                  */
1262                 cpu = inp->inp_lgrpindex % netisr_ncpus;
1263                 if (cpu != mycpuid) {
1264                         struct lwkt_port *port = netisr_cpuport(cpu);
1265                         lwkt_msg_t lmsg = &msg->lmsg;
1266
1267                         /*
1268                          * We are moving the protocol processing port the
1269                          * socket is on, we have to unlink here and re-link
1270                          * on the target cpu (this inpcb is still left in
1271                          * the wildcard hash).
1272                          */
1273                         in_pcbunlink_flags(inp, &udbinfo[mycpuid], 0);
1274                         msg->nm_dispatch = udp_soreuseport_dispatch;
1275
1276                         /*
1277                          * See the related comment in tcp_usrreq.c
1278                          * tcp_connect()
1279                          */
1280                         lwkt_setmsg_receipt(lmsg, udp_sosetport);
1281                         lwkt_forwardmsg(port, lmsg);
1282                         return TRUE; /* forwarded */
1283                 }
1284         }
1285         return FALSE;
1286 }
1287
1288 static void
1289 udp_inswildcardhash_dispatch(netmsg_t msg)
1290 {
1291         struct inpcb *inp = msg->base.nm_so->so_pcb;
1292         boolean_t forwarded;
1293
1294         KASSERT(inp->inp_lport != 0, ("local port not set yet"));
1295         KASSERT(udp_lportcpu(inp->inp_lport) == mycpuid, ("not target cpu"));
1296
1297         in_pcblink(inp, &udbinfo[mycpuid]);
1298
1299         forwarded = udp_inswildcardhash_oncpu(inp, &msg->base);
1300         if (forwarded) {
1301                 /* The message is further forwarded, so we are done here. */
1302                 return;
1303         }
1304         udp_inswildcard_replymsg(msg);
1305 }
1306
1307 static boolean_t
1308 udp_inswildcardhash(struct inpcb *inp, struct netmsg_base *msg, int error)
1309 {
1310         lwkt_msg_t lmsg = &msg->lmsg;
1311         int cpu;
1312
1313         ASSERT_INP_NOTINHASH(inp);
1314
1315         /* This inpcb could no longer be directly detached */
1316         inp->inp_flags &= ~INP_DIRECT_DETACH;
1317
1318         /*
1319          * Always clear the route cache, so we don't need to
1320          * worry about any owner CPU changes later.
1321          */
1322         in_pcbresetroute(inp);
1323
1324         KASSERT(inp->inp_lport != 0, ("local port not set yet"));
1325         cpu = udp_lportcpu(inp->inp_lport);
1326
1327         lmsg->ms_error = error;
1328         if (cpu != mycpuid) {
1329                 struct lwkt_port *port = netisr_cpuport(cpu);
1330
1331                 /*
1332                  * We are moving the protocol processing port the socket
1333                  * is on, we have to unlink here and re-link on the
1334                  * target cpu.
1335                  */
1336                 in_pcbunlink(inp, &udbinfo[mycpuid]);
1337                 msg->nm_dispatch = udp_inswildcardhash_dispatch;
1338
1339                 /* See the related comment in tcp_usrreq.c tcp_connect() */
1340                 lwkt_setmsg_receipt(lmsg, udp_sosetport);
1341                 lwkt_forwardmsg(port, lmsg);
1342                 return TRUE; /* forwarded */
1343         }
1344
1345         return udp_inswildcardhash_oncpu(inp, msg);
1346 }
1347
1348 static void
1349 udp_bind(netmsg_t msg)
1350 {
1351         struct socket *so = msg->bind.base.nm_so;
1352         struct inpcb *inp;
1353         int error;
1354
1355         inp = so->so_pcb;
1356         if (inp) {
1357                 struct sockaddr *nam = msg->bind.nm_nam;
1358                 struct thread *td = msg->bind.nm_td;
1359                 struct sockaddr_in *sin;
1360                 lwkt_port_t port;
1361                 int cpu;
1362
1363                 /*
1364                  * Check "already bound" here (in_pcbbind() does the same
1365                  * check though), so we don't forward a connected/bound
1366                  * socket randomly which would panic in the following
1367                  * in_pcbunlink().
1368                  */
1369                 if (inp->inp_lport != 0 ||
1370                     inp->inp_laddr.s_addr != INADDR_ANY) {
1371                         error = EINVAL; /* already bound */
1372                         goto done;
1373                 }
1374
1375                 if (nam->sa_len != sizeof(*sin)) {
1376                         error = EINVAL;
1377                         goto done;
1378                 }
1379                 sin = (struct sockaddr_in *)nam;
1380
1381                 cpu = udp_lportcpu(sin->sin_port);
1382                 port = netisr_cpuport(cpu);
1383
1384                 /*
1385                  * See the related comment in tcp_usrreq.c tcp_usr_bind().
1386                  * The exception is that we use local port based netisr
1387                  * to serialize in_pcbbind().
1388                  */
1389                 if (&curthread->td_msgport != port) {
1390                         lwkt_msg_t lmsg = &msg->bind.base.lmsg;
1391
1392                         KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0,
1393                             ("already asked to relink"));
1394
1395                         in_pcbunlink(so->so_pcb, &udbinfo[mycpuid]);
1396                         msg->bind.nm_flags |= PRUB_RELINK;
1397
1398                         /*
1399                          * See the related comment in tcp_usrreq.c
1400                          * tcp_connect().
1401                          */
1402                         lwkt_setmsg_receipt(lmsg, udp_sosetport);
1403                         lwkt_forwardmsg(port, lmsg);
1404                         /* msg invalid now */
1405                         return;
1406                 }
1407                 KASSERT(so->so_port == port, ("so_port is not netisr%d", cpu));
1408
1409                 if (msg->bind.nm_flags & PRUB_RELINK) {
1410                         msg->bind.nm_flags &= ~PRUB_RELINK;
1411                         in_pcblink(so->so_pcb, &udbinfo[mycpuid]);
1412                 }
1413                 KASSERT(inp->inp_pcbinfo == &udbinfo[cpu],
1414                     ("pcbinfo is not udbinfo%d", cpu));
1415
1416                 error = in_pcbbind(inp, nam, td);
1417                 if (error == 0) {
1418                         boolean_t forwarded;
1419
1420                         if (sin->sin_addr.s_addr != INADDR_ANY)
1421                                 inp->inp_flags |= INP_WASBOUND_NOTANY;
1422
1423                         forwarded = udp_inswildcardhash(inp,
1424                             &msg->bind.base, 0);
1425                         if (forwarded) {
1426                                 /*
1427                                  * The message is further forwarded, so
1428                                  * we are done here.
1429                                  */
1430                                 return;
1431                         }
1432                 }
1433         } else {
1434                 error = EINVAL;
1435         }
1436 done:
1437         lwkt_replymsg(&msg->bind.base.lmsg, error);
1438 }
1439
1440 static int
1441 udp_preconnect(struct socket *so, const struct sockaddr *nam __unused,
1442     struct thread *td __unused)
1443 {
1444         sosetstate(so, SS_ISCONNECTED);         /* XXX */
1445         return 0;
1446 }
1447
1448 static void
1449 udp_connect(netmsg_t msg)
1450 {
1451         struct socket *so = msg->connect.base.nm_so;
1452         struct sockaddr *nam = msg->connect.nm_nam;
1453         struct thread *td = msg->connect.nm_td;
1454         struct inpcb *inp;
1455         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1456         struct sockaddr_in *if_sin;
1457         struct lwkt_port *port;
1458         uint16_t hash;
1459         int error;
1460
1461         KKASSERT(msg->connect.nm_m == NULL);
1462
1463         inp = so->so_pcb;
1464         if (inp == NULL) {
1465                 error = EINVAL;
1466                 goto out;
1467         }
1468
1469         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1470                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1471                 in_pcblink(inp, &udbinfo[mycpuid]);
1472         }
1473
1474         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1475                 error = EISCONN;
1476                 goto out;
1477         }
1478         error = 0;
1479
1480         /*
1481          * Bind if we have to
1482          */
1483         if (inp->inp_lport == 0) {
1484                 error = in_pcbbind(inp, NULL, td);
1485                 if (error)
1486                         goto out;
1487         }
1488
1489         /*
1490          * Calculate the correct protocol processing thread.  The connect
1491          * operation must run there.
1492          */
1493         error = in_pcbladdr(inp, nam, &if_sin, td);
1494         if (error)
1495                 goto out;
1496         if (!prison_remote_ip(td, nam)) {
1497                 error = EAFNOSUPPORT; /* IPv6 only jail */
1498                 goto out;
1499         }
1500
1501         hash = udp_addrhash(sin->sin_addr.s_addr, sin->sin_port,
1502             inp->inp_laddr.s_addr != INADDR_ANY ?
1503             inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, inp->inp_lport);
1504         port = netisr_hashport(hash);
1505         if (port != &curthread->td_msgport) {
1506                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1507                 int nm_flags = PRUC_RECONNECT;
1508
1509                 /*
1510                  * in_pcbladdr() may have allocated a route entry for us
1511                  * on the current CPU, but we need a route entry on the
1512                  * inpcb's owner CPU, so free it here.
1513                  */
1514                 in_pcbresetroute(inp);
1515
1516                 if (inp->inp_flags & INP_WILDCARD) {
1517                         /*
1518                          * Remove this inpcb from the wildcard hash before
1519                          * the socket's msgport changes.
1520                          */
1521                         udp_remwildcardhash(inp);
1522                 }
1523
1524                 if (so->so_orig_port == NULL) {
1525                         /*
1526                          * First time change protocol processing port.
1527                          * Save the current port for synchronization upon
1528                          * udp_detach.
1529                          */
1530                         so->so_orig_port = &curthread->td_msgport;
1531                 } else {
1532                         /*
1533                          * We have changed protocol processing port more
1534                          * than once.  We could not do direct detach
1535                          * anymore, because we lose the track of the
1536                          * original protocol processing ports to perform
1537                          * synchronization upon udp_detach.  This should
1538                          * be rare though.
1539                          */
1540                         inp->inp_flags &= ~INP_DIRECT_DETACH;
1541                 }
1542
1543                 /*
1544                  * We are moving the protocol processing port the socket
1545                  * is on, we have to unlink here and re-link on the
1546                  * target cpu.
1547                  */
1548                 in_pcbunlink(inp, &udbinfo[mycpuid]);
1549                 msg->connect.nm_flags |= nm_flags;
1550
1551                 /* See the related comment in tcp_usrreq.c tcp_connect() */
1552                 lwkt_setmsg_receipt(lmsg, udp_sosetport);
1553                 lwkt_forwardmsg(port, lmsg);
1554                 /* msg invalid now */
1555                 return;
1556         }
1557         error = udp_connect_oncpu(inp, sin, if_sin, hash);
1558 out:
1559         if (msg->connect.nm_flags & PRUC_HELDTD)
1560                 lwkt_rele(td);
1561         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1562                 if (inp->inp_lport == 0) {
1563                         /*
1564                          * As long as we have the local port, it is fine
1565                          * for connect to fail, e.g. disconnect.
1566                          */
1567                         so->so_error = error;
1568                 }
1569                 soclrstate(so, SS_ISCONNECTED);
1570                 /*
1571                  * Wake up callers blocked on this socket to make sure
1572                  * that they can see this error.
1573                  *
1574                  * NOTE:
1575                  * sodisconnected() can't be used here, which bricks
1576                  * sending and receiving.
1577                  */
1578                 wakeup(&so->so_timeo);
1579                 sowwakeup(so);
1580                 sorwakeup(so);
1581         }
1582         if (error && inp != NULL && inp->inp_lport != 0 &&
1583             (inp->inp_flags & INP_WILDCARD) == 0) {
1584                 boolean_t forwarded;
1585
1586                 /* Connect failed; put it to wildcard hash. */
1587                 forwarded = udp_inswildcardhash(inp, &msg->connect.base,
1588                     error);
1589                 if (forwarded) {
1590                         /*
1591                          * The message is further forwarded, so we are done
1592                          * here.
1593                          */
1594                         return;
1595                 }
1596         }
1597         lwkt_replymsg(&msg->connect.base.lmsg, error);
1598 }
1599
1600 static void
1601 udp_remwildcardhash(struct inpcb *inp)
1602 {
1603         int cpu;
1604
1605         KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid],
1606             ("not on owner cpu"));
1607
1608         for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
1609                 if (cpu == mycpuid) {
1610                         /*
1611                          * This inpcb will be removed by the later
1612                          * in_pcbremwildcardhash().
1613                          */
1614                         continue;
1615                 }
1616                 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpu]);
1617         }
1618         in_pcbremwildcardhash(inp);
1619 }
1620
1621 static int
1622 udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin,
1623     struct sockaddr_in *if_sin, uint16_t hash)
1624 {
1625         struct socket *so = inp->inp_socket;
1626         struct inpcb *oinp;
1627
1628         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1629             sin->sin_addr, sin->sin_port,
1630             inp->inp_laddr.s_addr != INADDR_ANY ?
1631             inp->inp_laddr : if_sin->sin_addr, inp->inp_lport, FALSE, NULL);
1632         if (oinp != NULL)
1633                 return EADDRINUSE;
1634
1635         /*
1636          * No more errors can occur, finish adjusting the socket
1637          * and change the processing port to reflect the connected
1638          * socket.  Once set we can no longer safely mess with the
1639          * socket.
1640          */
1641
1642         if (inp->inp_flags & INP_WILDCARD)
1643                 udp_remwildcardhash(inp);
1644
1645         if (inp->inp_laddr.s_addr == INADDR_ANY)
1646                 inp->inp_laddr = if_sin->sin_addr;
1647         inp->inp_faddr = sin->sin_addr;
1648         inp->inp_fport = sin->sin_port;
1649         in_pcbinsconnhash(inp);
1650
1651         inp->inp_flags |= INP_HASH;
1652         inp->inp_hashval = hash;
1653
1654         soisconnected(so);
1655
1656         return 0;
1657 }
1658
1659 static void
1660 udp_detach2(struct socket *so)
1661 {
1662         in_pcbdetach(so->so_pcb);
1663         sodiscard(so);
1664         sofree(so);
1665 }
1666
1667 static void
1668 udp_detach_final_dispatch(netmsg_t msg)
1669 {
1670         udp_detach2(msg->base.nm_so);
1671 }
1672
1673 static void
1674 udp_detach_oncpu_dispatch(netmsg_t msg)
1675 {
1676         struct netmsg_base *clomsg = &msg->base;
1677         struct socket *so = clomsg->nm_so;
1678         struct inpcb *inp = so->so_pcb;
1679         struct thread *td = curthread;
1680         int nextcpu, cpuid = mycpuid;
1681
1682         KASSERT(td->td_type == TD_TYPE_NETISR, ("not in netisr"));
1683
1684         if (inp->inp_flags & INP_WILDCARD) {
1685                 /*
1686                  * This inp will be removed on the inp's
1687                  * owner CPU later, so don't do it now.
1688                  */
1689                 if (&td->td_msgport != so->so_port)
1690                         in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpuid]);
1691         }
1692
1693         if (cpuid == 0) {
1694                 /*
1695                  * Free and clear multicast socket option,
1696                  * which is only accessed in netisr0.
1697                  */
1698                 ip_freemoptions(inp->inp_moptions);
1699                 inp->inp_moptions = NULL;
1700         }
1701
1702         nextcpu = cpuid + 1;
1703         if (nextcpu < netisr_ncpus) {
1704                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &clomsg->lmsg);
1705         } else {
1706                 /*
1707                  * No one could see this inpcb now; destroy this
1708                  * inpcb in its owner netisr.
1709                  */
1710                 netmsg_init(clomsg, so, &netisr_apanic_rport, 0,
1711                     udp_detach_final_dispatch);
1712                 lwkt_sendmsg(so->so_port, &clomsg->lmsg);
1713         }
1714 }
1715
1716 static void
1717 udp_detach_syncorig_dispatch(netmsg_t msg)
1718 {
1719         struct netmsg_base *clomsg = &msg->base;
1720         struct socket *so = clomsg->nm_so;
1721
1722         /*
1723          * Original protocol processing port is synchronized;
1724          * destroy this inpcb in its owner netisr.
1725          */
1726         netmsg_init(clomsg, so, &netisr_apanic_rport, 0,
1727             udp_detach_final_dispatch);
1728         lwkt_sendmsg(so->so_port, &clomsg->lmsg);
1729 }
1730
1731 static void
1732 udp_detach(netmsg_t msg)
1733 {
1734         struct socket *so = msg->detach.base.nm_so;
1735         struct netmsg_base *clomsg;
1736         struct inpcb *inp;
1737
1738         inp = so->so_pcb;
1739         if (inp == NULL) {
1740                 lwkt_replymsg(&msg->detach.base.lmsg, EINVAL);
1741                 return;
1742         }
1743
1744         /*
1745          * Reply EJUSTRETURN ASAP, we will call sodiscard() and
1746          * sofree() later.
1747          */
1748         lwkt_replymsg(&msg->detach.base.lmsg, EJUSTRETURN);
1749
1750         if (netisr_ncpus == 1) {
1751                 /* Only one CPU, detach the inpcb directly. */
1752                 udp_detach2(so);
1753                 return;
1754         }
1755
1756         /*
1757          * Remove this inpcb from the inpcb list first, so that
1758          * no one could find this inpcb from the inpcb list.
1759          */
1760         in_pcbofflist(inp);
1761
1762         /*
1763          * Remove this inpcb from the local port hash directly
1764          * here, so that its bound local port could be recycled
1765          * timely.
1766          */
1767         in_pcbremporthash(inp);
1768
1769         if (inp->inp_flags & INP_DIRECT_DETACH) {
1770                 /*
1771                  * Direct detaching is allowed
1772                  */
1773                 KASSERT((inp->inp_flags & INP_WILDCARD) == 0,
1774                     ("in the wildcardhash"));
1775                 KASSERT(inp->inp_moptions == NULL, ("has mcast options"));
1776                 if (so->so_orig_port == NULL) {
1777                         udp_detach2(so);
1778                 } else {
1779                         /*
1780                          * Protocol processing port changed once, so
1781                          * we need to make sure that there are nothing
1782                          * left on the original protocol processing
1783                          * port before we destroy this socket and inpcb.
1784                          * This is more lightweight than going through
1785                          * all UDP processing netisrs.
1786                          */
1787                         clomsg = &so->so_clomsg;
1788                         netmsg_init(clomsg, so, &netisr_apanic_rport,
1789                             MSGF_IGNSOPORT, udp_detach_syncorig_dispatch);
1790                         lwkt_sendmsg(so->so_orig_port, &clomsg->lmsg);
1791                 }
1792                 return;
1793         }
1794
1795         /*
1796          * Go through netisrs which process UDP to make sure
1797          * no one could find this inpcb anymore.
1798          */
1799         clomsg = &so->so_clomsg;
1800         netmsg_init(clomsg, so, &netisr_apanic_rport, MSGF_IGNSOPORT,
1801             udp_detach_oncpu_dispatch);
1802         lwkt_sendmsg(netisr_cpuport(0), &clomsg->lmsg);
1803 }
1804
1805 static void
1806 udp_disconnect(netmsg_t msg)
1807 {
1808         struct socket *so = msg->disconnect.base.nm_so;
1809         struct inpcb *inp;
1810         boolean_t forwarded;
1811         int error = 0;
1812
1813         inp = so->so_pcb;
1814         if (inp == NULL) {
1815                 error = EINVAL;
1816                 goto out;
1817         }
1818         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1819                 error = ENOTCONN;
1820                 goto out;
1821         }
1822
1823         soclrstate(so, SS_ISCONNECTED);         /* XXX */
1824
1825         in_pcbdisconnect(inp);
1826         inp->inp_flags &= ~INP_HASH;
1827
1828         /*
1829          * Follow traditional BSD behavior and retain the local port
1830          * binding.  But, fix the old misbehavior of overwriting any
1831          * previously bound local address.
1832          */
1833         if (!(inp->inp_flags & INP_WASBOUND_NOTANY))
1834                 inp->inp_laddr.s_addr = INADDR_ANY;
1835
1836         if (so->so_state & SS_ISCLOSING) {
1837                 /*
1838                  * If this socket is being closed, there is no need
1839                  * to put this socket back into wildcard hash table.
1840                  */
1841                 error = 0;
1842                 goto out;
1843         }
1844
1845         forwarded = udp_inswildcardhash(inp, &msg->disconnect.base, 0);
1846         if (forwarded) {
1847                 /*
1848                  * The message is further forwarded, so we are done
1849                  * here.
1850                  */
1851                 return;
1852         }
1853 out:
1854         lwkt_replymsg(&msg->disconnect.base.lmsg, error);
1855 }
1856
1857 void
1858 udp_shutdown(netmsg_t msg)
1859 {
1860         struct socket *so = msg->shutdown.base.nm_so;
1861         struct inpcb *inp;
1862         int error;
1863
1864         inp = so->so_pcb;
1865         if (inp) {
1866                 socantsendmore(so);
1867                 error = 0;
1868         } else {
1869                 error = EINVAL;
1870         }
1871         lwkt_replymsg(&msg->shutdown.base.lmsg, error);
1872 }
1873
1874 struct pr_usrreqs udp_usrreqs = {
1875         .pru_abort = udp_abort,
1876         .pru_accept = pr_generic_notsupp,
1877         .pru_attach = udp_attach,
1878         .pru_bind = udp_bind,
1879         .pru_connect = udp_connect,
1880         .pru_connect2 = pr_generic_notsupp,
1881         .pru_control = in_control_dispatch,
1882         .pru_detach = udp_detach,
1883         .pru_disconnect = udp_disconnect,
1884         .pru_listen = pr_generic_notsupp,
1885         .pru_peeraddr = in_setpeeraddr_dispatch,
1886         .pru_rcvd = pr_generic_notsupp,
1887         .pru_rcvoob = pr_generic_notsupp,
1888         .pru_send = udp_send,
1889         .pru_sense = pru_sense_null,
1890         .pru_shutdown = udp_shutdown,
1891         .pru_sockaddr = in_setsockaddr_dispatch,
1892         .pru_sosend = sosendudp,
1893         .pru_soreceive = soreceive,
1894         .pru_preconnect = udp_preconnect,
1895         .pru_preattach = udp_preattach
1896 };