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"
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
52 #include <sys/malloc.h>
56 #include <sys/protosw.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/systm.h>
68 #include <net/route.h>
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
76 #include <netinet/ip6.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/ip_options.h>
83 #include <netinet6/ip6_var.h>
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
89 #include <netipsec/ipsec.h>
90 #include <netipsec/esp.h>
93 #include <machine/in_cksum.h>
95 #include <security/mac/mac_framework.h>
98 * UDP protocol implementation.
99 * Per RFC 768, August, 1980.
103 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
104 * removes the only data integrity mechanism for packets and malformed
105 * packets that would otherwise be discarded due to bad checksums, and may
106 * cause problems (especially for NFS data blocks).
108 static int udp_cksum = 1;
109 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
110 0, "compute udp checksum");
112 int udp_log_in_vain = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114 &udp_log_in_vain, 0, "Log all incoming UDP packets");
116 VNET_DEFINE(int, udp_blackhole) = 0;
117 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
118 &VNET_NAME(udp_blackhole), 0,
119 "Do not send port unreachables for refused connects");
121 u_long udp_sendspace = 9216; /* really max datagram size */
122 /* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
126 u_long udp_recvspace = 40 * (1024 +
128 sizeof(struct sockaddr_in6)
130 sizeof(struct sockaddr_in)
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
137 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
138 VNET_DEFINE(struct inpcbinfo, udbinfo);
139 static VNET_DEFINE(uma_zone_t, udpcb_zone);
140 #define V_udpcb_zone VNET(udpcb_zone)
143 #define UDBHASHSIZE 128
146 VNET_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
147 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
148 &VNET_NAME(udpstat), udpstat,
149 "UDP statistics (struct udpstat, netinet/udp_var.h)");
152 static void udp_detach(struct socket *so);
153 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
154 struct mbuf *, struct thread *);
159 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
161 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
163 #endif /* IPSEC_NAT_T */
167 udp_zone_change(void *tag)
170 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
171 uma_zone_set_max(V_udpcb_zone, maxsockets);
175 udp_inpcb_init(void *mem, int size, int flags)
180 INP_LOCK_INIT(inp, "inp", "udpinp");
188 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
189 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
190 IPI_HASHFIELDS_2TUPLE);
191 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 uma_zone_set_max(V_udpcb_zone, maxsockets);
194 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
195 EVENTHANDLER_PRI_ANY);
199 * Kernel module interface for updating udpstat. The argument is an index
200 * into udpstat treated as an array of u_long. While this encodes the
201 * general layout of udpstat into the caller, it doesn't encode its location,
202 * so that future changes to add, for example, per-CPU stats support won't
203 * cause binary compatibility problems for kernel modules.
206 kmod_udpstat_inc(int statnum)
209 (*((u_long *)&V_udpstat + statnum))++;
213 udp_newudpcb(struct inpcb *inp)
217 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
225 udp_discardcb(struct udpcb *up)
228 uma_zfree(V_udpcb_zone, up);
236 in_pcbinfo_destroy(&V_udbinfo);
237 uma_zdestroy(V_udpcb_zone);
243 * Subroutine of udp_input(), which appends the provided mbuf chain to the
244 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
245 * contains the source address. If the socket ends up being an IPv6 socket,
246 * udp_append() will convert to a sockaddr_in6 before passing the address
247 * into the socket code.
250 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
251 struct sockaddr_in *udp_in)
253 struct sockaddr *append_sa;
255 struct mbuf *opts = 0;
257 struct sockaddr_in6 udp_in6;
261 INP_LOCK_ASSERT(inp);
264 * Engage the tunneling protocol.
267 if (up->u_tun_func != NULL) {
268 (*up->u_tun_func)(n, off, inp);
275 off += sizeof(struct udphdr);
278 /* Check AH/ESP integrity. */
279 if (ipsec4_in_reject(n, inp)) {
281 V_ipsec4stat.in_polvio++;
286 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
287 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
288 n = udp4_espdecap(inp, n, off);
289 if (n == NULL) /* Consumed. */
292 #endif /* IPSEC_NAT_T */
295 if (mac_inpcb_check_deliver(inp, n) != 0) {
300 if (inp->inp_flags & INP_CONTROLOPTS ||
301 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
303 if (inp->inp_vflag & INP_IPV6)
304 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
307 ip_savecontrol(inp, &opts, ip, n);
310 if (inp->inp_vflag & INP_IPV6) {
311 bzero(&udp_in6, sizeof(udp_in6));
312 udp_in6.sin6_len = sizeof(udp_in6);
313 udp_in6.sin6_family = AF_INET6;
314 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
315 append_sa = (struct sockaddr *)&udp_in6;
318 append_sa = (struct sockaddr *)udp_in;
321 so = inp->inp_socket;
322 SOCKBUF_LOCK(&so->so_rcv);
323 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
324 SOCKBUF_UNLOCK(&so->so_rcv);
328 UDPSTAT_INC(udps_fullsock);
330 sorwakeup_locked(so);
334 udp_input(struct mbuf *m, int off)
343 struct sockaddr_in udp_in;
344 #ifdef IPFIREWALL_FORWARD
345 struct m_tag *fwd_tag;
348 ifp = m->m_pkthdr.rcvif;
349 UDPSTAT_INC(udps_ipackets);
352 * Strip IP options, if any; should skip this, make available to
353 * user, and use on returned packets, but we don't yet have a way to
354 * check the checksum with options still present.
356 if (iphlen > sizeof (struct ip)) {
357 ip_stripoptions(m, (struct mbuf *)0);
358 iphlen = sizeof(struct ip);
362 * Get IP and UDP header together in first mbuf.
364 ip = mtod(m, struct ip *);
365 if (m->m_len < iphlen + sizeof(struct udphdr)) {
366 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
367 UDPSTAT_INC(udps_hdrops);
370 ip = mtod(m, struct ip *);
372 uh = (struct udphdr *)((caddr_t)ip + iphlen);
375 * Destination port of 0 is illegal, based on RFC768.
377 if (uh->uh_dport == 0)
381 * Construct sockaddr format source address. Stuff source address
382 * and datagram in user buffer.
384 bzero(&udp_in, sizeof(udp_in));
385 udp_in.sin_len = sizeof(udp_in);
386 udp_in.sin_family = AF_INET;
387 udp_in.sin_port = uh->uh_sport;
388 udp_in.sin_addr = ip->ip_src;
391 * Make mbuf data length reflect UDP length. If not enough data to
392 * reflect UDP length, drop.
394 len = ntohs((u_short)uh->uh_ulen);
395 if (ip->ip_len != len) {
396 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
397 UDPSTAT_INC(udps_badlen);
400 m_adj(m, len - ip->ip_len);
401 /* ip->ip_len = len; */
405 * Save a copy of the IP header in case we want restore it for
406 * sending an ICMP error message in response.
408 if (!V_udp_blackhole)
411 memset(&save_ip, 0, sizeof(save_ip));
414 * Checksum extended UDP header and data.
419 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
420 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
421 uh_sum = m->m_pkthdr.csum_data;
423 uh_sum = in_pseudo(ip->ip_src.s_addr,
424 ip->ip_dst.s_addr, htonl((u_short)len +
425 m->m_pkthdr.csum_data + IPPROTO_UDP));
430 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
431 bzero(((struct ipovly *)ip)->ih_x1, 9);
432 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
433 uh_sum = in_cksum(m, len + sizeof (struct ip));
434 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
437 UDPSTAT_INC(udps_badsum);
442 UDPSTAT_INC(udps_nosum);
444 #ifdef IPFIREWALL_FORWARD
446 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
448 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
449 if (fwd_tag != NULL) {
450 struct sockaddr_in *next_hop;
455 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
456 ip->ip_dst = next_hop->sin_addr;
457 uh->uh_dport = ntohs(next_hop->sin_port);
460 * Remove the tag from the packet. We don't need it anymore.
462 m_tag_delete(m, fwd_tag);
466 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
467 in_broadcast(ip->ip_dst, ifp)) {
469 struct ip_moptions *imo;
471 INP_INFO_RLOCK(&V_udbinfo);
473 LIST_FOREACH(inp, &V_udb, inp_list) {
474 if (inp->inp_lport != uh->uh_dport)
477 if ((inp->inp_vflag & INP_IPV4) == 0)
480 if (inp->inp_laddr.s_addr != INADDR_ANY &&
481 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
483 if (inp->inp_faddr.s_addr != INADDR_ANY &&
484 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
486 if (inp->inp_fport != 0 &&
487 inp->inp_fport != uh->uh_sport)
493 * XXXRW: Because we weren't holding either the inpcb
494 * or the hash lock when we checked for a match
495 * before, we should probably recheck now that the
496 * inpcb lock is held.
500 * Handle socket delivery policy for any-source
501 * and source-specific multicast. [RFC3678]
503 imo = inp->inp_moptions;
504 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
505 struct sockaddr_in group;
511 bzero(&group, sizeof(struct sockaddr_in));
512 group.sin_len = sizeof(struct sockaddr_in);
513 group.sin_family = AF_INET;
514 group.sin_addr = ip->ip_dst;
516 blocked = imo_multi_filter(imo, ifp,
517 (struct sockaddr *)&group,
518 (struct sockaddr *)&udp_in);
519 if (blocked != MCAST_PASS) {
520 if (blocked == MCAST_NOTGMEMBER)
521 IPSTAT_INC(ips_notmember);
522 if (blocked == MCAST_NOTSMEMBER ||
523 blocked == MCAST_MUTED)
524 UDPSTAT_INC(udps_filtermcast);
532 n = m_copy(m, 0, M_COPYALL);
533 udp_append(last, ip, n, iphlen, &udp_in);
538 * Don't look for additional matches if this one does
539 * not have either the SO_REUSEPORT or SO_REUSEADDR
540 * socket options set. This heuristic avoids
541 * searching through all pcbs in the common case of a
542 * non-shared port. It assumes that an application
543 * will never clear these options after setting them.
545 if ((last->inp_socket->so_options &
546 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
552 * No matching pcb found; discard datagram. (No need
553 * to send an ICMP Port Unreachable for a broadcast
554 * or multicast datgram.)
556 UDPSTAT_INC(udps_noportbcast);
559 INP_INFO_RUNLOCK(&V_udbinfo);
562 udp_append(last, ip, m, iphlen, &udp_in);
564 INP_INFO_RUNLOCK(&V_udbinfo);
569 * Locate pcb for datagram.
571 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
572 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
575 if (udp_log_in_vain) {
576 char buf[4*sizeof "123"];
578 strcpy(buf, inet_ntoa(ip->ip_dst));
580 "Connection attempt to UDP %s:%d from %s:%d\n",
581 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
582 ntohs(uh->uh_sport));
584 UDPSTAT_INC(udps_noport);
585 if (m->m_flags & (M_BCAST | M_MCAST)) {
586 UDPSTAT_INC(udps_noportbcast);
591 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
594 ip->ip_len += iphlen;
595 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
600 * Check the minimum TTL for socket.
602 INP_RLOCK_ASSERT(inp);
603 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
608 udp_append(inp, ip, m, iphlen, &udp_in);
618 * Notify a udp user of an asynchronous error; just wake up so that they can
619 * collect error status.
622 udp_notify(struct inpcb *inp, int errno)
626 * While udp_ctlinput() always calls udp_notify() with a read lock
627 * when invoking it directly, in_pcbnotifyall() currently uses write
628 * locks due to sharing code with TCP. For now, accept either a read
629 * or a write lock, but a read lock is sufficient.
631 INP_LOCK_ASSERT(inp);
633 inp->inp_socket->so_error = errno;
634 sorwakeup(inp->inp_socket);
635 sowwakeup(inp->inp_socket);
641 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
645 struct in_addr faddr;
648 faddr = ((struct sockaddr_in *)sa)->sin_addr;
649 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
653 * Redirects don't need to be handled up here.
655 if (PRC_IS_REDIRECT(cmd))
659 * Hostdead is ugly because it goes linearly through all PCBs.
661 * XXX: We never get this from ICMP, otherwise it makes an excellent
662 * DoS attack on machines with many connections.
664 if (cmd == PRC_HOSTDEAD)
666 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
669 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
670 inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
671 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
673 INP_RLOCK_ASSERT(inp);
674 if (inp->inp_socket != NULL) {
675 udp_notify(inp, inetctlerrmap[cmd]);
680 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
686 udp_pcblist(SYSCTL_HANDLER_ARGS)
689 struct inpcb *inp, **inp_list;
694 * The process of preparing the PCB list is too time-consuming and
695 * resource-intensive to repeat twice on every request.
697 if (req->oldptr == 0) {
698 n = V_udbinfo.ipi_count;
699 n += imax(n / 8, 10);
700 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
704 if (req->newptr != 0)
708 * OK, now we're committed to doing something.
710 INP_INFO_RLOCK(&V_udbinfo);
711 gencnt = V_udbinfo.ipi_gencnt;
712 n = V_udbinfo.ipi_count;
713 INP_INFO_RUNLOCK(&V_udbinfo);
715 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
716 + n * sizeof(struct xinpcb));
720 xig.xig_len = sizeof xig;
722 xig.xig_gen = gencnt;
723 xig.xig_sogen = so_gencnt;
724 error = SYSCTL_OUT(req, &xig, sizeof xig);
728 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
732 INP_INFO_RLOCK(&V_udbinfo);
733 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
734 inp = LIST_NEXT(inp, inp_list)) {
736 if (inp->inp_gencnt <= gencnt &&
737 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
743 INP_INFO_RUNLOCK(&V_udbinfo);
747 for (i = 0; i < n; i++) {
750 if (inp->inp_gencnt <= gencnt) {
753 bzero(&xi, sizeof(xi));
754 xi.xi_len = sizeof xi;
755 /* XXX should avoid extra copy */
756 bcopy(inp, &xi.xi_inp, sizeof *inp);
758 sotoxsocket(inp->inp_socket, &xi.xi_socket);
759 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
761 error = SYSCTL_OUT(req, &xi, sizeof xi);
765 INP_INFO_WLOCK(&V_udbinfo);
766 for (i = 0; i < n; i++) {
769 if (!in_pcbrele_rlocked(inp))
772 INP_INFO_WUNLOCK(&V_udbinfo);
776 * Give the user an updated idea of our state. If the
777 * generation differs from what we told her before, she knows
778 * that something happened while we were processing this
779 * request, and it might be necessary to retry.
781 INP_INFO_RLOCK(&V_udbinfo);
782 xig.xig_gen = V_udbinfo.ipi_gencnt;
783 xig.xig_sogen = so_gencnt;
784 xig.xig_count = V_udbinfo.ipi_count;
785 INP_INFO_RUNLOCK(&V_udbinfo);
786 error = SYSCTL_OUT(req, &xig, sizeof xig);
788 free(inp_list, M_TEMP);
792 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
793 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
794 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
798 udp_getcred(SYSCTL_HANDLER_ARGS)
801 struct sockaddr_in addrs[2];
805 error = priv_check(req->td, PRIV_NETINET_GETCRED);
808 error = SYSCTL_IN(req, addrs, sizeof(addrs));
811 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
812 addrs[0].sin_addr, addrs[0].sin_port,
813 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
815 INP_RLOCK_ASSERT(inp);
816 if (inp->inp_socket == NULL)
819 error = cr_canseeinpcb(req->td->td_ucred, inp);
821 cru2x(inp->inp_cred, &xuc);
826 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
830 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
831 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
832 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
836 udp_ctloutput(struct socket *so, struct sockopt *sopt)
838 int error = 0, optval;
845 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
847 if (sopt->sopt_level != IPPROTO_UDP) {
849 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
851 error = ip6_ctloutput(so, sopt);
854 #if defined(INET) && defined(INET6)
860 error = ip_ctloutput(so, sopt);
866 switch (sopt->sopt_dir) {
868 switch (sopt->sopt_name) {
871 error = sooptcopyin(sopt, &optval, sizeof optval,
876 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
880 KASSERT(up != NULL, ("%s: up == NULL", __func__));
884 /* Clear all UDP encap. */
886 up->u_flags &= ~UF_ESPINUDP_ALL;
890 case UDP_ENCAP_ESPINUDP:
891 case UDP_ENCAP_ESPINUDP_NON_IKE:
892 up->u_flags &= ~UF_ESPINUDP_ALL;
893 if (optval == UDP_ENCAP_ESPINUDP)
894 up->u_flags |= UF_ESPINUDP;
895 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
896 up->u_flags |= UF_ESPINUDP_NON_IKE;
912 switch (sopt->sopt_name) {
916 KASSERT(up != NULL, ("%s: up == NULL", __func__));
917 optval = up->u_flags & UF_ESPINUDP_ALL;
919 error = sooptcopyout(sopt, &optval, sizeof optval);
935 #define UH_UNLOCKED 0
937 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
938 struct mbuf *control, struct thread *td)
941 int len = m->m_pkthdr.len;
942 struct in_addr faddr, laddr;
944 struct sockaddr_in *sin, src;
947 u_short fport, lport;
951 * udp_output() may need to temporarily bind or connect the current
952 * inpcb. As such, we don't know up front whether we will need the
953 * pcbinfo lock or not. Do any work to decide what is needed up
954 * front before acquiring any locks.
956 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
964 if (control != NULL) {
966 * XXX: Currently, we assume all the optional information is
967 * stored in a single mbuf.
969 if (control->m_next) {
974 for (; control->m_len > 0;
975 control->m_data += CMSG_ALIGN(cm->cmsg_len),
976 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
977 cm = mtod(control, struct cmsghdr *);
978 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
979 || cm->cmsg_len > control->m_len) {
983 if (cm->cmsg_level != IPPROTO_IP)
986 switch (cm->cmsg_type) {
989 CMSG_LEN(sizeof(struct in_addr))) {
993 bzero(&src, sizeof(src));
994 src.sin_family = AF_INET;
995 src.sin_len = sizeof(src);
996 src.sin_port = inp->inp_lport;
998 *(struct in_addr *)CMSG_DATA(cm);
1002 error = ENOPROTOOPT;
1016 * Depending on whether or not the application has bound or connected
1017 * the socket, we may have to do varying levels of work. The optimal
1018 * case is for a connected UDP socket, as a global lock isn't
1021 * In order to decide which we need, we require stability of the
1022 * inpcb binding, which we ensure by acquiring a read lock on the
1023 * inpcb. This doesn't strictly follow the lock order, so we play
1024 * the trylock and retry game; note that we may end up with more
1025 * conservative locks than required the second time around, so later
1026 * assertions have to accept that. Further analysis of the number of
1027 * misses under contention is required.
1029 * XXXRW: Check that hash locking update here is correct.
1031 sin = (struct sockaddr_in *)addr;
1034 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1037 INP_HASH_WLOCK(&V_udbinfo);
1038 unlock_udbinfo = UH_WLOCKED;
1039 } else if ((sin != NULL && (
1040 (sin->sin_addr.s_addr == INADDR_ANY) ||
1041 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1042 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1043 (inp->inp_lport == 0))) ||
1044 (src.sin_family == AF_INET)) {
1045 INP_HASH_RLOCK(&V_udbinfo);
1046 unlock_udbinfo = UH_RLOCKED;
1048 unlock_udbinfo = UH_UNLOCKED;
1051 * If the IP_SENDSRCADDR control message was specified, override the
1052 * source address for this datagram. Its use is invalidated if the
1053 * address thus specified is incomplete or clobbers other inpcbs.
1055 laddr = inp->inp_laddr;
1056 lport = inp->inp_lport;
1057 if (src.sin_family == AF_INET) {
1058 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1060 (laddr.s_addr == INADDR_ANY &&
1061 src.sin_addr.s_addr == INADDR_ANY)) {
1065 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1066 &laddr.s_addr, &lport, td->td_ucred);
1072 * If a UDP socket has been connected, then a local address/port will
1073 * have been selected and bound.
1075 * If a UDP socket has not been connected to, then an explicit
1076 * destination address must be used, in which case a local
1077 * address/port may not have been selected and bound.
1080 INP_LOCK_ASSERT(inp);
1081 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1087 * Jail may rewrite the destination address, so let it do
1088 * that before we use it.
1090 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1095 * If a local address or port hasn't yet been selected, or if
1096 * the destination address needs to be rewritten due to using
1097 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1098 * to do the heavy lifting. Once a port is selected, we
1099 * commit the binding back to the socket; we also commit the
1100 * binding of the address if in jail.
1102 * If we already have a valid binding and we're not
1103 * requesting a destination address rewrite, use a fast path.
1105 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1106 inp->inp_lport == 0 ||
1107 sin->sin_addr.s_addr == INADDR_ANY ||
1108 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1109 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1110 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1111 &lport, &faddr.s_addr, &fport, NULL,
1117 * XXXRW: Why not commit the port if the address is
1120 /* Commit the local port if newly assigned. */
1121 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1122 inp->inp_lport == 0) {
1123 INP_WLOCK_ASSERT(inp);
1124 INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1126 * Remember addr if jailed, to prevent
1129 if (prison_flag(td->td_ucred, PR_IP4))
1130 inp->inp_laddr = laddr;
1131 inp->inp_lport = lport;
1132 if (in_pcbinshash(inp) != 0) {
1137 inp->inp_flags |= INP_ANONPORT;
1140 faddr = sin->sin_addr;
1141 fport = sin->sin_port;
1144 INP_LOCK_ASSERT(inp);
1145 faddr = inp->inp_faddr;
1146 fport = inp->inp_fport;
1147 if (faddr.s_addr == INADDR_ANY) {
1154 * Calculate data length and get a mbuf for UDP, IP, and possible
1155 * link-layer headers. Immediate slide the data pointer back forward
1156 * since we won't use that space at this layer.
1158 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1163 m->m_data += max_linkhdr;
1164 m->m_len -= max_linkhdr;
1165 m->m_pkthdr.len -= max_linkhdr;
1168 * Fill in mbuf with extended UDP header and addresses and length put
1169 * into network format.
1171 ui = mtod(m, struct udpiphdr *);
1172 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1173 ui->ui_pr = IPPROTO_UDP;
1176 ui->ui_sport = lport;
1177 ui->ui_dport = fport;
1178 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1181 * Set the Don't Fragment bit in the IP header.
1183 if (inp->inp_flags & INP_DONTFRAG) {
1186 ip = (struct ip *)&ui->ui_i;
1187 ip->ip_off |= IP_DF;
1191 if (inp->inp_socket->so_options & SO_DONTROUTE)
1192 ipflags |= IP_ROUTETOIF;
1193 if (inp->inp_socket->so_options & SO_BROADCAST)
1194 ipflags |= IP_ALLOWBROADCAST;
1195 if (inp->inp_flags & INP_ONESBCAST)
1196 ipflags |= IP_SENDONES;
1199 mac_inpcb_create_mbuf(inp, m);
1203 * Set up checksum and output datagram.
1206 if (inp->inp_flags & INP_ONESBCAST)
1207 faddr.s_addr = INADDR_BROADCAST;
1208 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1209 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1210 m->m_pkthdr.csum_flags = CSUM_UDP;
1211 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1214 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1215 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1216 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */
1217 UDPSTAT_INC(udps_opackets);
1219 if (unlock_udbinfo == UH_WLOCKED)
1220 INP_HASH_WUNLOCK(&V_udbinfo);
1221 else if (unlock_udbinfo == UH_RLOCKED)
1222 INP_HASH_RUNLOCK(&V_udbinfo);
1223 error = ip_output(m, inp->inp_options, NULL, ipflags,
1224 inp->inp_moptions, inp);
1225 if (unlock_udbinfo == UH_WLOCKED)
1232 if (unlock_udbinfo == UH_WLOCKED) {
1233 INP_HASH_WUNLOCK(&V_udbinfo);
1235 } else if (unlock_udbinfo == UH_RLOCKED) {
1236 INP_HASH_RUNLOCK(&V_udbinfo);
1245 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1247 * Potentially decap ESP in UDP frame. Check for an ESP header
1248 * and optional marker; if present, strip the UDP header and
1249 * push the result through IPSec.
1251 * Returns mbuf to be processed (potentially re-allocated) or
1252 * NULL if consumed and/or processed.
1254 static struct mbuf *
1255 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1257 size_t minlen, payload, skip, iphlen;
1261 struct udphdr *udphdr;
1264 INP_RLOCK_ASSERT(inp);
1267 * Pull up data so the longest case is contiguous:
1268 * IP/UDP hdr + non ESP marker + ESP hdr.
1270 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1271 if (minlen > m->m_pkthdr.len)
1272 minlen = m->m_pkthdr.len;
1273 if ((m = m_pullup(m, minlen)) == NULL) {
1274 V_ipsec4stat.in_inval++;
1275 return (NULL); /* Bypass caller processing. */
1277 data = mtod(m, caddr_t); /* Points to ip header. */
1278 payload = m->m_len - off; /* Size of payload. */
1280 if (payload == 1 && data[off] == '\xff')
1281 return (m); /* NB: keepalive packet, no decap. */
1283 up = intoudpcb(inp);
1284 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1285 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1286 ("u_flags 0x%x", up->u_flags));
1289 * Check that the payload is large enough to hold an
1290 * ESP header and compute the amount of data to remove.
1292 * NB: the caller has already done a pullup for us.
1293 * XXX can we assume alignment and eliminate bcopys?
1295 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1297 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1298 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1299 * possible AH mode non-IKE marker+non-ESP marker
1300 * from draft-ietf-ipsec-udp-encaps-00.txt.
1304 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1305 return (m); /* NB: no decap. */
1306 bcopy(data + off, &marker, sizeof(uint64_t));
1307 if (marker != 0) /* Non-IKE marker. */
1308 return (m); /* NB: no decap. */
1309 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1313 if (payload <= sizeof(struct esp)) {
1314 V_ipsec4stat.in_inval++;
1316 return (NULL); /* Discard. */
1318 bcopy(data + off, &spi, sizeof(uint32_t));
1319 if (spi == 0) /* Non-ESP marker. */
1320 return (m); /* NB: no decap. */
1321 skip = sizeof(struct udphdr);
1325 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1326 * the UDP ports. This is required if we want to select
1327 * the right SPD for multiple hosts behind same NAT.
1329 * NB: ports are maintained in network byte order everywhere
1330 * in the NAT-T code.
1332 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1333 2 * sizeof(uint16_t), M_NOWAIT);
1335 V_ipsec4stat.in_nomem++;
1337 return (NULL); /* Discard. */
1339 iphlen = off - sizeof(struct udphdr);
1340 udphdr = (struct udphdr *)(data + iphlen);
1341 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1342 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1343 m_tag_prepend(m, tag);
1346 * Remove the UDP header (and possibly the non ESP marker)
1347 * IP header length is iphlen
1350 * +----+------+-----+
1351 * | IP | UDP | ESP |
1352 * +----+------+-----+
1360 ovbcopy(data, data + skip, iphlen);
1363 ip = mtod(m, struct ip *);
1365 ip->ip_p = IPPROTO_ESP;
1368 * We cannot yet update the cksums so clear any
1369 * h/w cksum flags as they are no longer valid.
1371 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1372 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1374 (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1375 return (NULL); /* NB: consumed, bypass processing. */
1377 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1380 udp_abort(struct socket *so)
1384 inp = sotoinpcb(so);
1385 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1387 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1388 INP_HASH_WLOCK(&V_udbinfo);
1389 in_pcbdisconnect(inp);
1390 inp->inp_laddr.s_addr = INADDR_ANY;
1391 INP_HASH_WUNLOCK(&V_udbinfo);
1392 soisdisconnected(so);
1398 udp_attach(struct socket *so, int proto, struct thread *td)
1403 inp = sotoinpcb(so);
1404 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1405 error = soreserve(so, udp_sendspace, udp_recvspace);
1408 INP_INFO_WLOCK(&V_udbinfo);
1409 error = in_pcballoc(so, &V_udbinfo);
1411 INP_INFO_WUNLOCK(&V_udbinfo);
1415 inp = sotoinpcb(so);
1416 inp->inp_vflag |= INP_IPV4;
1417 inp->inp_ip_ttl = V_ip_defttl;
1419 error = udp_newudpcb(inp);
1423 INP_INFO_WUNLOCK(&V_udbinfo);
1428 INP_INFO_WUNLOCK(&V_udbinfo);
1434 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1439 KASSERT(so->so_type == SOCK_DGRAM,
1440 ("udp_set_kernel_tunneling: !dgram"));
1441 inp = sotoinpcb(so);
1442 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1444 up = intoudpcb(inp);
1445 if (up->u_tun_func != NULL) {
1456 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1461 inp = sotoinpcb(so);
1462 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1464 INP_HASH_WLOCK(&V_udbinfo);
1465 error = in_pcbbind(inp, nam, td->td_ucred);
1466 INP_HASH_WUNLOCK(&V_udbinfo);
1472 udp_close(struct socket *so)
1476 inp = sotoinpcb(so);
1477 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1479 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1480 INP_HASH_WLOCK(&V_udbinfo);
1481 in_pcbdisconnect(inp);
1482 inp->inp_laddr.s_addr = INADDR_ANY;
1483 INP_HASH_WUNLOCK(&V_udbinfo);
1484 soisdisconnected(so);
1490 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1494 struct sockaddr_in *sin;
1496 inp = sotoinpcb(so);
1497 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1499 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1503 sin = (struct sockaddr_in *)nam;
1504 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1509 INP_HASH_WLOCK(&V_udbinfo);
1510 error = in_pcbconnect(inp, nam, td->td_ucred);
1511 INP_HASH_WUNLOCK(&V_udbinfo);
1519 udp_detach(struct socket *so)
1524 inp = sotoinpcb(so);
1525 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1526 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1527 ("udp_detach: not disconnected"));
1528 INP_INFO_WLOCK(&V_udbinfo);
1530 up = intoudpcb(inp);
1531 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1532 inp->inp_ppcb = NULL;
1535 INP_INFO_WUNLOCK(&V_udbinfo);
1540 udp_disconnect(struct socket *so)
1544 inp = sotoinpcb(so);
1545 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1547 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1551 INP_HASH_WLOCK(&V_udbinfo);
1552 in_pcbdisconnect(inp);
1553 inp->inp_laddr.s_addr = INADDR_ANY;
1554 INP_HASH_WUNLOCK(&V_udbinfo);
1556 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1563 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1564 struct mbuf *control, struct thread *td)
1568 inp = sotoinpcb(so);
1569 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1570 return (udp_output(inp, m, addr, control, td));
1575 udp_shutdown(struct socket *so)
1579 inp = sotoinpcb(so);
1580 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1588 struct pr_usrreqs udp_usrreqs = {
1589 .pru_abort = udp_abort,
1590 .pru_attach = udp_attach,
1591 .pru_bind = udp_bind,
1592 .pru_connect = udp_connect,
1593 .pru_control = in_control,
1594 .pru_detach = udp_detach,
1595 .pru_disconnect = udp_disconnect,
1596 .pru_peeraddr = in_getpeeraddr,
1597 .pru_send = udp_send,
1598 .pru_soreceive = soreceive_dgram,
1599 .pru_sosend = sosend_dgram,
1600 .pru_shutdown = udp_shutdown,
1601 .pru_sockaddr = in_getsockaddr,
1602 .pru_sosetlabel = in_pcbsosetlabel,
1603 .pru_close = udp_close,