2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2008 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
8 * Portions of this software were developed by Robert N. M. Watson under
9 * contract to Juniper Networks, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 #include "opt_kdtrace.h"
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
51 #include <sys/kernel.h>
53 #include <sys/malloc.h>
57 #include <sys/protosw.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
70 #include <net/route.h>
72 #include <netinet/in.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
79 #include <netinet/ip6.h>
81 #include <netinet/ip_icmp.h>
82 #include <netinet/icmp_var.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/ip_options.h>
86 #include <netinet6/ip6_var.h>
88 #include <netinet/udp.h>
89 #include <netinet/udp_var.h>
92 #include <netipsec/ipsec.h>
93 #include <netipsec/esp.h>
96 #include <machine/in_cksum.h>
98 #include <security/mac/mac_framework.h>
101 * UDP protocol implementation.
102 * Per RFC 768, August, 1980.
106 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
107 * removes the only data integrity mechanism for packets and malformed
108 * packets that would otherwise be discarded due to bad checksums, and may
109 * cause problems (especially for NFS data blocks).
111 VNET_DEFINE(int, udp_cksum) = 1;
112 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
113 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
115 int udp_log_in_vain = 0;
116 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
117 &udp_log_in_vain, 0, "Log all incoming UDP packets");
119 VNET_DEFINE(int, udp_blackhole) = 0;
120 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
121 &VNET_NAME(udp_blackhole), 0,
122 "Do not send port unreachables for refused connects");
124 u_long udp_sendspace = 9216; /* really max datagram size */
125 /* 40 1K datagrams */
126 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
127 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
129 u_long udp_recvspace = 40 * (1024 +
131 sizeof(struct sockaddr_in6)
133 sizeof(struct sockaddr_in)
137 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
138 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
140 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
141 VNET_DEFINE(struct inpcbinfo, udbinfo);
142 static VNET_DEFINE(uma_zone_t, udpcb_zone);
143 #define V_udpcb_zone VNET(udpcb_zone)
146 #define UDBHASHSIZE 128
149 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
150 VNET_PCPUSTAT_SYSINIT(udpstat);
151 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
152 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
155 VNET_PCPUSTAT_SYSUNINIT(udpstat);
158 static void udp_detach(struct socket *so);
159 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
160 struct mbuf *, struct thread *);
165 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
167 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
169 #endif /* IPSEC_NAT_T */
173 udp_zone_change(void *tag)
176 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
177 uma_zone_set_max(V_udpcb_zone, maxsockets);
181 udp_inpcb_init(void *mem, int size, int flags)
186 INP_LOCK_INIT(inp, "inp", "udpinp");
194 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
195 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
196 IPI_HASHFIELDS_2TUPLE);
197 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
199 uma_zone_set_max(V_udpcb_zone, maxsockets);
200 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
201 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
202 EVENTHANDLER_PRI_ANY);
206 * Kernel module interface for updating udpstat. The argument is an index
207 * into udpstat treated as an array of u_long. While this encodes the
208 * general layout of udpstat into the caller, it doesn't encode its location,
209 * so that future changes to add, for example, per-CPU stats support won't
210 * cause binary compatibility problems for kernel modules.
213 kmod_udpstat_inc(int statnum)
216 counter_u64_add(VNET(udpstat)[statnum], 1);
220 udp_newudpcb(struct inpcb *inp)
224 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
232 udp_discardcb(struct udpcb *up)
235 uma_zfree(V_udpcb_zone, up);
243 in_pcbinfo_destroy(&V_udbinfo);
244 uma_zdestroy(V_udpcb_zone);
250 * Subroutine of udp_input(), which appends the provided mbuf chain to the
251 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
252 * contains the source address. If the socket ends up being an IPv6 socket,
253 * udp_append() will convert to a sockaddr_in6 before passing the address
254 * into the socket code.
257 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
258 struct sockaddr_in *udp_in)
260 struct sockaddr *append_sa;
262 struct mbuf *opts = 0;
264 struct sockaddr_in6 udp_in6;
268 INP_LOCK_ASSERT(inp);
271 * Engage the tunneling protocol.
274 if (up->u_tun_func != NULL) {
275 (*up->u_tun_func)(n, off, inp);
282 off += sizeof(struct udphdr);
285 /* Check AH/ESP integrity. */
286 if (ipsec4_in_reject(n, inp)) {
288 IPSECSTAT_INC(ips_in_polvio);
293 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
294 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
295 n = udp4_espdecap(inp, n, off);
296 if (n == NULL) /* Consumed. */
299 #endif /* IPSEC_NAT_T */
302 if (mac_inpcb_check_deliver(inp, n) != 0) {
307 if (inp->inp_flags & INP_CONTROLOPTS ||
308 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
310 if (inp->inp_vflag & INP_IPV6)
311 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
314 ip_savecontrol(inp, &opts, ip, n);
317 if (inp->inp_vflag & INP_IPV6) {
318 bzero(&udp_in6, sizeof(udp_in6));
319 udp_in6.sin6_len = sizeof(udp_in6);
320 udp_in6.sin6_family = AF_INET6;
321 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
322 append_sa = (struct sockaddr *)&udp_in6;
325 append_sa = (struct sockaddr *)udp_in;
328 so = inp->inp_socket;
329 SOCKBUF_LOCK(&so->so_rcv);
330 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
331 SOCKBUF_UNLOCK(&so->so_rcv);
335 UDPSTAT_INC(udps_fullsock);
337 sorwakeup_locked(so);
341 udp_input(struct mbuf *m, int off)
348 uint16_t len, ip_len;
350 struct sockaddr_in udp_in;
351 struct m_tag *fwd_tag;
353 ifp = m->m_pkthdr.rcvif;
354 UDPSTAT_INC(udps_ipackets);
357 * Strip IP options, if any; should skip this, make available to
358 * user, and use on returned packets, but we don't yet have a way to
359 * check the checksum with options still present.
361 if (iphlen > sizeof (struct ip)) {
363 iphlen = sizeof(struct ip);
367 * Get IP and UDP header together in first mbuf.
369 ip = mtod(m, struct ip *);
370 if (m->m_len < iphlen + sizeof(struct udphdr)) {
371 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
372 UDPSTAT_INC(udps_hdrops);
375 ip = mtod(m, struct ip *);
377 uh = (struct udphdr *)((caddr_t)ip + iphlen);
380 * Destination port of 0 is illegal, based on RFC768.
382 if (uh->uh_dport == 0)
386 * Construct sockaddr format source address. Stuff source address
387 * and datagram in user buffer.
389 bzero(&udp_in, sizeof(udp_in));
390 udp_in.sin_len = sizeof(udp_in);
391 udp_in.sin_family = AF_INET;
392 udp_in.sin_port = uh->uh_sport;
393 udp_in.sin_addr = ip->ip_src;
396 * Make mbuf data length reflect UDP length. If not enough data to
397 * reflect UDP length, drop.
399 len = ntohs((u_short)uh->uh_ulen);
400 ip_len = ntohs(ip->ip_len) - iphlen;
402 if (len > ip_len || len < sizeof(struct udphdr)) {
403 UDPSTAT_INC(udps_badlen);
406 m_adj(m, len - ip_len);
410 * Save a copy of the IP header in case we want restore it for
411 * sending an ICMP error message in response.
413 if (!V_udp_blackhole)
416 memset(&save_ip, 0, sizeof(save_ip));
419 * Checksum extended UDP header and data.
424 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
425 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
426 uh_sum = m->m_pkthdr.csum_data;
428 uh_sum = in_pseudo(ip->ip_src.s_addr,
429 ip->ip_dst.s_addr, htonl((u_short)len +
430 m->m_pkthdr.csum_data + IPPROTO_UDP));
435 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
436 bzero(((struct ipovly *)ip)->ih_x1, 9);
437 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
438 uh_sum = in_cksum(m, len + sizeof (struct ip));
439 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
442 UDPSTAT_INC(udps_badsum);
447 UDPSTAT_INC(udps_nosum);
449 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
450 in_broadcast(ip->ip_dst, ifp)) {
452 struct ip_moptions *imo;
454 INP_INFO_RLOCK(&V_udbinfo);
456 LIST_FOREACH(inp, &V_udb, inp_list) {
457 if (inp->inp_lport != uh->uh_dport)
460 if ((inp->inp_vflag & INP_IPV4) == 0)
463 if (inp->inp_laddr.s_addr != INADDR_ANY &&
464 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
466 if (inp->inp_faddr.s_addr != INADDR_ANY &&
467 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
469 if (inp->inp_fport != 0 &&
470 inp->inp_fport != uh->uh_sport)
476 * XXXRW: Because we weren't holding either the inpcb
477 * or the hash lock when we checked for a match
478 * before, we should probably recheck now that the
479 * inpcb lock is held.
483 * Handle socket delivery policy for any-source
484 * and source-specific multicast. [RFC3678]
486 imo = inp->inp_moptions;
487 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
488 struct sockaddr_in group;
494 bzero(&group, sizeof(struct sockaddr_in));
495 group.sin_len = sizeof(struct sockaddr_in);
496 group.sin_family = AF_INET;
497 group.sin_addr = ip->ip_dst;
499 blocked = imo_multi_filter(imo, ifp,
500 (struct sockaddr *)&group,
501 (struct sockaddr *)&udp_in);
502 if (blocked != MCAST_PASS) {
503 if (blocked == MCAST_NOTGMEMBER)
504 IPSTAT_INC(ips_notmember);
505 if (blocked == MCAST_NOTSMEMBER ||
506 blocked == MCAST_MUTED)
507 UDPSTAT_INC(udps_filtermcast);
515 n = m_copy(m, 0, M_COPYALL);
516 udp_append(last, ip, n, iphlen, &udp_in);
521 * Don't look for additional matches if this one does
522 * not have either the SO_REUSEPORT or SO_REUSEADDR
523 * socket options set. This heuristic avoids
524 * searching through all pcbs in the common case of a
525 * non-shared port. It assumes that an application
526 * will never clear these options after setting them.
528 if ((last->inp_socket->so_options &
529 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
535 * No matching pcb found; discard datagram. (No need
536 * to send an ICMP Port Unreachable for a broadcast
537 * or multicast datgram.)
539 UDPSTAT_INC(udps_noportbcast);
542 INP_INFO_RUNLOCK(&V_udbinfo);
545 udp_append(last, ip, m, iphlen, &udp_in);
547 INP_INFO_RUNLOCK(&V_udbinfo);
552 * Locate pcb for datagram.
556 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
558 if ((m->m_flags & M_IP_NEXTHOP) &&
559 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
560 struct sockaddr_in *next_hop;
562 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
565 * Transparently forwarded. Pretend to be the destination.
566 * Already got one like this?
568 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
569 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
572 * It's new. Try to find the ambushing socket.
573 * Because we've rewritten the destination address,
574 * any hardware-generated hash is ignored.
576 inp = in_pcblookup(&V_udbinfo, ip->ip_src,
577 uh->uh_sport, next_hop->sin_addr,
578 next_hop->sin_port ? htons(next_hop->sin_port) :
579 uh->uh_dport, INPLOOKUP_WILDCARD |
580 INPLOOKUP_RLOCKPCB, ifp);
582 /* Remove the tag from the packet. We don't need it anymore. */
583 m_tag_delete(m, fwd_tag);
584 m->m_flags &= ~M_IP_NEXTHOP;
586 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
587 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
588 INPLOOKUP_RLOCKPCB, ifp, m);
590 if (udp_log_in_vain) {
591 char buf[4*sizeof "123"];
593 strcpy(buf, inet_ntoa(ip->ip_dst));
595 "Connection attempt to UDP %s:%d from %s:%d\n",
596 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
597 ntohs(uh->uh_sport));
599 UDPSTAT_INC(udps_noport);
600 if (m->m_flags & (M_BCAST | M_MCAST)) {
601 UDPSTAT_INC(udps_noportbcast);
606 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
609 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
614 * Check the minimum TTL for socket.
616 INP_RLOCK_ASSERT(inp);
617 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
623 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
624 udp_append(inp, ip, m, iphlen, &udp_in);
634 * Notify a udp user of an asynchronous error; just wake up so that they can
635 * collect error status.
638 udp_notify(struct inpcb *inp, int errno)
642 * While udp_ctlinput() always calls udp_notify() with a read lock
643 * when invoking it directly, in_pcbnotifyall() currently uses write
644 * locks due to sharing code with TCP. For now, accept either a read
645 * or a write lock, but a read lock is sufficient.
647 INP_LOCK_ASSERT(inp);
649 inp->inp_socket->so_error = errno;
650 sorwakeup(inp->inp_socket);
651 sowwakeup(inp->inp_socket);
657 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
661 struct in_addr faddr;
664 faddr = ((struct sockaddr_in *)sa)->sin_addr;
665 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
669 * Redirects don't need to be handled up here.
671 if (PRC_IS_REDIRECT(cmd))
675 * Hostdead is ugly because it goes linearly through all PCBs.
677 * XXX: We never get this from ICMP, otherwise it makes an excellent
678 * DoS attack on machines with many connections.
680 if (cmd == PRC_HOSTDEAD)
682 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
685 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
686 inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
687 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
689 INP_RLOCK_ASSERT(inp);
690 if (inp->inp_socket != NULL) {
691 udp_notify(inp, inetctlerrmap[cmd]);
696 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
702 udp_pcblist(SYSCTL_HANDLER_ARGS)
705 struct inpcb *inp, **inp_list;
710 * The process of preparing the PCB list is too time-consuming and
711 * resource-intensive to repeat twice on every request.
713 if (req->oldptr == 0) {
714 n = V_udbinfo.ipi_count;
715 n += imax(n / 8, 10);
716 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
720 if (req->newptr != 0)
724 * OK, now we're committed to doing something.
726 INP_INFO_RLOCK(&V_udbinfo);
727 gencnt = V_udbinfo.ipi_gencnt;
728 n = V_udbinfo.ipi_count;
729 INP_INFO_RUNLOCK(&V_udbinfo);
731 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
732 + n * sizeof(struct xinpcb));
736 xig.xig_len = sizeof xig;
738 xig.xig_gen = gencnt;
739 xig.xig_sogen = so_gencnt;
740 error = SYSCTL_OUT(req, &xig, sizeof xig);
744 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
748 INP_INFO_RLOCK(&V_udbinfo);
749 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
750 inp = LIST_NEXT(inp, inp_list)) {
752 if (inp->inp_gencnt <= gencnt &&
753 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
759 INP_INFO_RUNLOCK(&V_udbinfo);
763 for (i = 0; i < n; i++) {
766 if (inp->inp_gencnt <= gencnt) {
769 bzero(&xi, sizeof(xi));
770 xi.xi_len = sizeof xi;
771 /* XXX should avoid extra copy */
772 bcopy(inp, &xi.xi_inp, sizeof *inp);
774 sotoxsocket(inp->inp_socket, &xi.xi_socket);
775 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
777 error = SYSCTL_OUT(req, &xi, sizeof xi);
781 INP_INFO_WLOCK(&V_udbinfo);
782 for (i = 0; i < n; i++) {
785 if (!in_pcbrele_rlocked(inp))
788 INP_INFO_WUNLOCK(&V_udbinfo);
792 * Give the user an updated idea of our state. If the
793 * generation differs from what we told her before, she knows
794 * that something happened while we were processing this
795 * request, and it might be necessary to retry.
797 INP_INFO_RLOCK(&V_udbinfo);
798 xig.xig_gen = V_udbinfo.ipi_gencnt;
799 xig.xig_sogen = so_gencnt;
800 xig.xig_count = V_udbinfo.ipi_count;
801 INP_INFO_RUNLOCK(&V_udbinfo);
802 error = SYSCTL_OUT(req, &xig, sizeof xig);
804 free(inp_list, M_TEMP);
808 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
809 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
810 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
814 udp_getcred(SYSCTL_HANDLER_ARGS)
817 struct sockaddr_in addrs[2];
821 error = priv_check(req->td, PRIV_NETINET_GETCRED);
824 error = SYSCTL_IN(req, addrs, sizeof(addrs));
827 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
828 addrs[0].sin_addr, addrs[0].sin_port,
829 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
831 INP_RLOCK_ASSERT(inp);
832 if (inp->inp_socket == NULL)
835 error = cr_canseeinpcb(req->td->td_ucred, inp);
837 cru2x(inp->inp_cred, &xuc);
842 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
846 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
847 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
848 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
852 udp_ctloutput(struct socket *so, struct sockopt *sopt)
854 int error = 0, optval;
861 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
863 if (sopt->sopt_level != IPPROTO_UDP) {
865 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
867 error = ip6_ctloutput(so, sopt);
870 #if defined(INET) && defined(INET6)
876 error = ip_ctloutput(so, sopt);
882 switch (sopt->sopt_dir) {
884 switch (sopt->sopt_name) {
887 error = sooptcopyin(sopt, &optval, sizeof optval,
892 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
896 KASSERT(up != NULL, ("%s: up == NULL", __func__));
900 /* Clear all UDP encap. */
902 up->u_flags &= ~UF_ESPINUDP_ALL;
906 case UDP_ENCAP_ESPINUDP:
907 case UDP_ENCAP_ESPINUDP_NON_IKE:
908 up->u_flags &= ~UF_ESPINUDP_ALL;
909 if (optval == UDP_ENCAP_ESPINUDP)
910 up->u_flags |= UF_ESPINUDP;
911 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
912 up->u_flags |= UF_ESPINUDP_NON_IKE;
928 switch (sopt->sopt_name) {
932 KASSERT(up != NULL, ("%s: up == NULL", __func__));
933 optval = up->u_flags & UF_ESPINUDP_ALL;
935 error = sooptcopyout(sopt, &optval, sizeof optval);
951 #define UH_UNLOCKED 0
953 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
954 struct mbuf *control, struct thread *td)
957 int len = m->m_pkthdr.len;
958 struct in_addr faddr, laddr;
960 struct sockaddr_in *sin, src;
963 u_short fport, lport;
968 * udp_output() may need to temporarily bind or connect the current
969 * inpcb. As such, we don't know up front whether we will need the
970 * pcbinfo lock or not. Do any work to decide what is needed up
971 * front before acquiring any locks.
973 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
982 tos = inp->inp_ip_tos;
983 if (control != NULL) {
985 * XXX: Currently, we assume all the optional information is
986 * stored in a single mbuf.
988 if (control->m_next) {
994 for (; control->m_len > 0;
995 control->m_data += CMSG_ALIGN(cm->cmsg_len),
996 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
997 cm = mtod(control, struct cmsghdr *);
998 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
999 || cm->cmsg_len > control->m_len) {
1003 if (cm->cmsg_level != IPPROTO_IP)
1006 switch (cm->cmsg_type) {
1007 case IP_SENDSRCADDR:
1009 CMSG_LEN(sizeof(struct in_addr))) {
1013 bzero(&src, sizeof(src));
1014 src.sin_family = AF_INET;
1015 src.sin_len = sizeof(src);
1016 src.sin_port = inp->inp_lport;
1018 *(struct in_addr *)CMSG_DATA(cm);
1022 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1026 tos = *(u_char *)CMSG_DATA(cm);
1030 error = ENOPROTOOPT;
1045 * Depending on whether or not the application has bound or connected
1046 * the socket, we may have to do varying levels of work. The optimal
1047 * case is for a connected UDP socket, as a global lock isn't
1050 * In order to decide which we need, we require stability of the
1051 * inpcb binding, which we ensure by acquiring a read lock on the
1052 * inpcb. This doesn't strictly follow the lock order, so we play
1053 * the trylock and retry game; note that we may end up with more
1054 * conservative locks than required the second time around, so later
1055 * assertions have to accept that. Further analysis of the number of
1056 * misses under contention is required.
1058 * XXXRW: Check that hash locking update here is correct.
1060 sin = (struct sockaddr_in *)addr;
1062 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1065 INP_HASH_WLOCK(&V_udbinfo);
1066 unlock_udbinfo = UH_WLOCKED;
1067 } else if ((sin != NULL && (
1068 (sin->sin_addr.s_addr == INADDR_ANY) ||
1069 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1070 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1071 (inp->inp_lport == 0))) ||
1072 (src.sin_family == AF_INET)) {
1073 INP_HASH_RLOCK(&V_udbinfo);
1074 unlock_udbinfo = UH_RLOCKED;
1076 unlock_udbinfo = UH_UNLOCKED;
1079 * If the IP_SENDSRCADDR control message was specified, override the
1080 * source address for this datagram. Its use is invalidated if the
1081 * address thus specified is incomplete or clobbers other inpcbs.
1083 laddr = inp->inp_laddr;
1084 lport = inp->inp_lport;
1085 if (src.sin_family == AF_INET) {
1086 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1088 (laddr.s_addr == INADDR_ANY &&
1089 src.sin_addr.s_addr == INADDR_ANY)) {
1093 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1094 &laddr.s_addr, &lport, td->td_ucred);
1100 * If a UDP socket has been connected, then a local address/port will
1101 * have been selected and bound.
1103 * If a UDP socket has not been connected to, then an explicit
1104 * destination address must be used, in which case a local
1105 * address/port may not have been selected and bound.
1108 INP_LOCK_ASSERT(inp);
1109 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1115 * Jail may rewrite the destination address, so let it do
1116 * that before we use it.
1118 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1123 * If a local address or port hasn't yet been selected, or if
1124 * the destination address needs to be rewritten due to using
1125 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1126 * to do the heavy lifting. Once a port is selected, we
1127 * commit the binding back to the socket; we also commit the
1128 * binding of the address if in jail.
1130 * If we already have a valid binding and we're not
1131 * requesting a destination address rewrite, use a fast path.
1133 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1134 inp->inp_lport == 0 ||
1135 sin->sin_addr.s_addr == INADDR_ANY ||
1136 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1137 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1138 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1139 &lport, &faddr.s_addr, &fport, NULL,
1145 * XXXRW: Why not commit the port if the address is
1148 /* Commit the local port if newly assigned. */
1149 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1150 inp->inp_lport == 0) {
1151 INP_WLOCK_ASSERT(inp);
1152 INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1154 * Remember addr if jailed, to prevent
1157 if (prison_flag(td->td_ucred, PR_IP4))
1158 inp->inp_laddr = laddr;
1159 inp->inp_lport = lport;
1160 if (in_pcbinshash(inp) != 0) {
1165 inp->inp_flags |= INP_ANONPORT;
1168 faddr = sin->sin_addr;
1169 fport = sin->sin_port;
1172 INP_LOCK_ASSERT(inp);
1173 faddr = inp->inp_faddr;
1174 fport = inp->inp_fport;
1175 if (faddr.s_addr == INADDR_ANY) {
1182 * Calculate data length and get a mbuf for UDP, IP, and possible
1183 * link-layer headers. Immediate slide the data pointer back forward
1184 * since we won't use that space at this layer.
1186 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1191 m->m_data += max_linkhdr;
1192 m->m_len -= max_linkhdr;
1193 m->m_pkthdr.len -= max_linkhdr;
1196 * Fill in mbuf with extended UDP header and addresses and length put
1197 * into network format.
1199 ui = mtod(m, struct udpiphdr *);
1200 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1201 ui->ui_v = IPVERSION << 4;
1202 ui->ui_pr = IPPROTO_UDP;
1205 ui->ui_sport = lport;
1206 ui->ui_dport = fport;
1207 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1210 * Set the Don't Fragment bit in the IP header.
1212 if (inp->inp_flags & INP_DONTFRAG) {
1215 ip = (struct ip *)&ui->ui_i;
1216 ip->ip_off |= htons(IP_DF);
1220 if (inp->inp_socket->so_options & SO_DONTROUTE)
1221 ipflags |= IP_ROUTETOIF;
1222 if (inp->inp_socket->so_options & SO_BROADCAST)
1223 ipflags |= IP_ALLOWBROADCAST;
1224 if (inp->inp_flags & INP_ONESBCAST)
1225 ipflags |= IP_SENDONES;
1228 mac_inpcb_create_mbuf(inp, m);
1232 * Set up checksum and output datagram.
1235 if (inp->inp_flags & INP_ONESBCAST)
1236 faddr.s_addr = INADDR_BROADCAST;
1237 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1238 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1239 m->m_pkthdr.csum_flags = CSUM_UDP;
1240 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1243 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1244 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1245 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1246 UDPSTAT_INC(udps_opackets);
1248 if (unlock_udbinfo == UH_WLOCKED)
1249 INP_HASH_WUNLOCK(&V_udbinfo);
1250 else if (unlock_udbinfo == UH_RLOCKED)
1251 INP_HASH_RUNLOCK(&V_udbinfo);
1252 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1253 error = ip_output(m, inp->inp_options, NULL, ipflags,
1254 inp->inp_moptions, inp);
1255 if (unlock_udbinfo == UH_WLOCKED)
1262 if (unlock_udbinfo == UH_WLOCKED) {
1263 INP_HASH_WUNLOCK(&V_udbinfo);
1265 } else if (unlock_udbinfo == UH_RLOCKED) {
1266 INP_HASH_RUNLOCK(&V_udbinfo);
1275 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1277 * Potentially decap ESP in UDP frame. Check for an ESP header
1278 * and optional marker; if present, strip the UDP header and
1279 * push the result through IPSec.
1281 * Returns mbuf to be processed (potentially re-allocated) or
1282 * NULL if consumed and/or processed.
1284 static struct mbuf *
1285 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1287 size_t minlen, payload, skip, iphlen;
1291 struct udphdr *udphdr;
1294 INP_RLOCK_ASSERT(inp);
1297 * Pull up data so the longest case is contiguous:
1298 * IP/UDP hdr + non ESP marker + ESP hdr.
1300 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1301 if (minlen > m->m_pkthdr.len)
1302 minlen = m->m_pkthdr.len;
1303 if ((m = m_pullup(m, minlen)) == NULL) {
1304 IPSECSTAT_INC(ips_in_inval);
1305 return (NULL); /* Bypass caller processing. */
1307 data = mtod(m, caddr_t); /* Points to ip header. */
1308 payload = m->m_len - off; /* Size of payload. */
1310 if (payload == 1 && data[off] == '\xff')
1311 return (m); /* NB: keepalive packet, no decap. */
1313 up = intoudpcb(inp);
1314 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1315 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1316 ("u_flags 0x%x", up->u_flags));
1319 * Check that the payload is large enough to hold an
1320 * ESP header and compute the amount of data to remove.
1322 * NB: the caller has already done a pullup for us.
1323 * XXX can we assume alignment and eliminate bcopys?
1325 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1327 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1328 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1329 * possible AH mode non-IKE marker+non-ESP marker
1330 * from draft-ietf-ipsec-udp-encaps-00.txt.
1334 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1335 return (m); /* NB: no decap. */
1336 bcopy(data + off, &marker, sizeof(uint64_t));
1337 if (marker != 0) /* Non-IKE marker. */
1338 return (m); /* NB: no decap. */
1339 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1343 if (payload <= sizeof(struct esp)) {
1344 IPSECSTAT_INC(ips_in_inval);
1346 return (NULL); /* Discard. */
1348 bcopy(data + off, &spi, sizeof(uint32_t));
1349 if (spi == 0) /* Non-ESP marker. */
1350 return (m); /* NB: no decap. */
1351 skip = sizeof(struct udphdr);
1355 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1356 * the UDP ports. This is required if we want to select
1357 * the right SPD for multiple hosts behind same NAT.
1359 * NB: ports are maintained in network byte order everywhere
1360 * in the NAT-T code.
1362 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1363 2 * sizeof(uint16_t), M_NOWAIT);
1365 IPSECSTAT_INC(ips_in_nomem);
1367 return (NULL); /* Discard. */
1369 iphlen = off - sizeof(struct udphdr);
1370 udphdr = (struct udphdr *)(data + iphlen);
1371 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1372 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1373 m_tag_prepend(m, tag);
1376 * Remove the UDP header (and possibly the non ESP marker)
1377 * IP header length is iphlen
1380 * +----+------+-----+
1381 * | IP | UDP | ESP |
1382 * +----+------+-----+
1390 ovbcopy(data, data + skip, iphlen);
1393 ip = mtod(m, struct ip *);
1394 ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1395 ip->ip_p = IPPROTO_ESP;
1398 * We cannot yet update the cksums so clear any
1399 * h/w cksum flags as they are no longer valid.
1401 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1402 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1404 (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1405 return (NULL); /* NB: consumed, bypass processing. */
1407 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1410 udp_abort(struct socket *so)
1414 inp = sotoinpcb(so);
1415 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1417 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1418 INP_HASH_WLOCK(&V_udbinfo);
1419 in_pcbdisconnect(inp);
1420 inp->inp_laddr.s_addr = INADDR_ANY;
1421 INP_HASH_WUNLOCK(&V_udbinfo);
1422 soisdisconnected(so);
1428 udp_attach(struct socket *so, int proto, struct thread *td)
1433 inp = sotoinpcb(so);
1434 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1435 error = soreserve(so, udp_sendspace, udp_recvspace);
1438 INP_INFO_WLOCK(&V_udbinfo);
1439 error = in_pcballoc(so, &V_udbinfo);
1441 INP_INFO_WUNLOCK(&V_udbinfo);
1445 inp = sotoinpcb(so);
1446 inp->inp_vflag |= INP_IPV4;
1447 inp->inp_ip_ttl = V_ip_defttl;
1449 error = udp_newudpcb(inp);
1453 INP_INFO_WUNLOCK(&V_udbinfo);
1458 INP_INFO_WUNLOCK(&V_udbinfo);
1464 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1469 KASSERT(so->so_type == SOCK_DGRAM,
1470 ("udp_set_kernel_tunneling: !dgram"));
1471 inp = sotoinpcb(so);
1472 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1474 up = intoudpcb(inp);
1475 if (up->u_tun_func != NULL) {
1486 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1491 inp = sotoinpcb(so);
1492 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1494 INP_HASH_WLOCK(&V_udbinfo);
1495 error = in_pcbbind(inp, nam, td->td_ucred);
1496 INP_HASH_WUNLOCK(&V_udbinfo);
1502 udp_close(struct socket *so)
1506 inp = sotoinpcb(so);
1507 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1509 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1510 INP_HASH_WLOCK(&V_udbinfo);
1511 in_pcbdisconnect(inp);
1512 inp->inp_laddr.s_addr = INADDR_ANY;
1513 INP_HASH_WUNLOCK(&V_udbinfo);
1514 soisdisconnected(so);
1520 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1524 struct sockaddr_in *sin;
1526 inp = sotoinpcb(so);
1527 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1529 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1533 sin = (struct sockaddr_in *)nam;
1534 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1539 INP_HASH_WLOCK(&V_udbinfo);
1540 error = in_pcbconnect(inp, nam, td->td_ucred);
1541 INP_HASH_WUNLOCK(&V_udbinfo);
1549 udp_detach(struct socket *so)
1554 inp = sotoinpcb(so);
1555 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1556 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1557 ("udp_detach: not disconnected"));
1558 INP_INFO_WLOCK(&V_udbinfo);
1560 up = intoudpcb(inp);
1561 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1562 inp->inp_ppcb = NULL;
1565 INP_INFO_WUNLOCK(&V_udbinfo);
1570 udp_disconnect(struct socket *so)
1574 inp = sotoinpcb(so);
1575 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1577 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1581 INP_HASH_WLOCK(&V_udbinfo);
1582 in_pcbdisconnect(inp);
1583 inp->inp_laddr.s_addr = INADDR_ANY;
1584 INP_HASH_WUNLOCK(&V_udbinfo);
1586 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1593 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1594 struct mbuf *control, struct thread *td)
1598 inp = sotoinpcb(so);
1599 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1600 return (udp_output(inp, m, addr, control, td));
1605 udp_shutdown(struct socket *so)
1609 inp = sotoinpcb(so);
1610 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1618 struct pr_usrreqs udp_usrreqs = {
1619 .pru_abort = udp_abort,
1620 .pru_attach = udp_attach,
1621 .pru_bind = udp_bind,
1622 .pru_connect = udp_connect,
1623 .pru_control = in_control,
1624 .pru_detach = udp_detach,
1625 .pru_disconnect = udp_disconnect,
1626 .pru_peeraddr = in_getpeeraddr,
1627 .pru_send = udp_send,
1628 .pru_soreceive = soreceive_dgram,
1629 .pru_sosend = sosend_dgram,
1630 .pru_shutdown = udp_shutdown,
1631 .pru_sockaddr = in_getsockaddr,
1632 .pru_sosetlabel = in_pcbsosetlabel,
1633 .pru_close = udp_close,