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
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
41 #include <sys/param.h>
42 #include <sys/domain.h>
43 #include <sys/eventhandler.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
51 #include <sys/protosw.h>
52 #include <sys/signalvar.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/systm.h>
63 #include <net/route.h>
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/icmp_var.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/ip_options.h>
78 #include <netinet6/ip6_var.h>
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
84 #include <netipsec/ipsec.h>
85 #include <netipsec/esp.h>
88 #include <machine/in_cksum.h>
90 #include <security/mac/mac_framework.h>
93 * UDP protocol implementation.
94 * Per RFC 768, August, 1980.
97 VNET_DEFINE(int, udp_blackhole);
100 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
101 * removes the only data integrity mechanism for packets and malformed
102 * packets that would otherwise be discarded due to bad checksums, and may
103 * cause problems (especially for NFS data blocks).
105 static int udp_cksum = 1;
106 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
107 0, "compute udp checksum");
109 int udp_log_in_vain = 0;
110 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
111 &udp_log_in_vain, 0, "Log all incoming UDP packets");
113 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
114 &VNET_NAME(udp_blackhole), 0,
115 "Do not send port unreachables for refused connects");
117 u_long udp_sendspace = 9216; /* really max datagram size */
118 /* 40 1K datagrams */
119 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
120 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
122 u_long udp_recvspace = 40 * (1024 +
124 sizeof(struct sockaddr_in6)
126 sizeof(struct sockaddr_in)
130 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
131 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
133 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
134 VNET_DEFINE(struct inpcbinfo, udbinfo);
135 static VNET_DEFINE(uma_zone_t, udpcb_zone);
136 VNET_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
138 #define V_udpcb_zone VNET(udpcb_zone)
141 #define UDBHASHSIZE 128
144 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
145 &VNET_NAME(udpstat), udpstat,
146 "UDP statistics (struct udpstat, netinet/udp_var.h)");
148 static void udp_detach(struct socket *so);
149 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
150 struct mbuf *, struct thread *);
153 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
155 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
157 #endif /* IPSEC_NAT_T */
161 udp_zone_change(void *tag)
164 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
165 uma_zone_set_max(V_udpcb_zone, maxsockets);
169 udp_inpcb_init(void *mem, int size, int flags)
174 INP_LOCK_INIT(inp, "inp", "udpinp");
184 INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
187 V_udbinfo.ipi_vnet = curvnet;
189 V_udbinfo.ipi_listhead = &V_udb;
190 V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
191 &V_udbinfo.ipi_hashmask);
192 V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
193 &V_udbinfo.ipi_porthashmask);
194 V_udbinfo.ipi_zone = uma_zcreate("udp_inpcb", sizeof(struct inpcb),
195 NULL, NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
198 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
200 uma_zone_set_max(V_udpcb_zone, maxsockets);
202 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
203 EVENTHANDLER_PRI_ANY);
207 * Kernel module interface for updating udpstat. The argument is an index
208 * into udpstat treated as an array of u_long. While this encodes the
209 * general layout of udpstat into the caller, it doesn't encode its location,
210 * so that future changes to add, for example, per-CPU stats support won't
211 * cause binary compatibility problems for kernel modules.
214 kmod_udpstat_inc(int statnum)
217 (*((u_long *)&V_udpstat + statnum))++;
221 udp_newudpcb(struct inpcb *inp)
225 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
233 udp_discardcb(struct udpcb *up)
236 uma_zfree(V_udpcb_zone, up);
244 hashdestroy(V_udbinfo.ipi_hashbase, M_PCB,
245 V_udbinfo.ipi_hashmask);
246 hashdestroy(V_udbinfo.ipi_porthashbase, M_PCB,
247 V_udbinfo.ipi_porthashmask);
248 INP_INFO_LOCK_DESTROY(&V_udbinfo);
253 * Subroutine of udp_input(), which appends the provided mbuf chain to the
254 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
255 * contains the source address. If the socket ends up being an IPv6 socket,
256 * udp_append() will convert to a sockaddr_in6 before passing the address
257 * into the socket code.
260 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
261 struct sockaddr_in *udp_in)
263 struct sockaddr *append_sa;
265 struct mbuf *opts = 0;
267 struct sockaddr_in6 udp_in6;
277 INP_RLOCK_ASSERT(inp);
280 /* Check AH/ESP integrity. */
281 if (ipsec4_in_reject(n, inp)) {
283 V_ipsec4stat.in_polvio++;
289 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
290 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
291 n = udp4_espdecap(inp, n, off);
292 if (n == NULL) /* Consumed. */
296 #endif /* IPSEC_NAT_T */
299 if (mac_inpcb_check_deliver(inp, n) != 0) {
304 if (inp->inp_flags & INP_CONTROLOPTS ||
305 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
307 if (inp->inp_vflag & INP_IPV6)
308 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
311 ip_savecontrol(inp, &opts, ip, n);
314 if (inp->inp_vflag & INP_IPV6) {
315 bzero(&udp_in6, sizeof(udp_in6));
316 udp_in6.sin6_len = sizeof(udp_in6);
317 udp_in6.sin6_family = AF_INET6;
318 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
319 append_sa = (struct sockaddr *)&udp_in6;
322 append_sa = (struct sockaddr *)udp_in;
325 so = inp->inp_socket;
326 SOCKBUF_LOCK(&so->so_rcv);
327 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
328 SOCKBUF_UNLOCK(&so->so_rcv);
332 UDPSTAT_INC(udps_fullsock);
334 sorwakeup_locked(so);
338 udp_input(struct mbuf *m, int off)
348 struct sockaddr_in udp_in;
349 #ifdef IPFIREWALL_FORWARD
350 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)) {
362 ip_stripoptions(m, (struct mbuf *)0);
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 if (ip->ip_len != len) {
401 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
402 UDPSTAT_INC(udps_badlen);
405 m_adj(m, len - ip->ip_len);
406 /* ip->ip_len = 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 #ifdef IPFIREWALL_FORWARD
451 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
453 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
454 if (fwd_tag != NULL) {
455 struct sockaddr_in *next_hop;
460 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
461 ip->ip_dst = next_hop->sin_addr;
462 uh->uh_dport = ntohs(next_hop->sin_port);
465 * Remove the tag from the packet. We don't need it anymore.
467 m_tag_delete(m, fwd_tag);
471 INP_INFO_RLOCK(&V_udbinfo);
472 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
473 in_broadcast(ip->ip_dst, ifp)) {
475 struct ip_moptions *imo;
478 LIST_FOREACH(inp, &V_udb, inp_list) {
479 if (inp->inp_lport != uh->uh_dport)
482 if ((inp->inp_vflag & INP_IPV4) == 0)
485 if (inp->inp_laddr.s_addr != INADDR_ANY &&
486 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
488 if (inp->inp_faddr.s_addr != INADDR_ANY &&
489 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
491 if (inp->inp_fport != 0 &&
492 inp->inp_fport != uh->uh_sport)
498 * Handle socket delivery policy for any-source
499 * and source-specific multicast. [RFC3678]
501 imo = inp->inp_moptions;
502 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
504 struct sockaddr_in group;
507 bzero(&group, sizeof(struct sockaddr_in));
508 group.sin_len = sizeof(struct sockaddr_in);
509 group.sin_family = AF_INET;
510 group.sin_addr = ip->ip_dst;
512 blocked = imo_multi_filter(imo, ifp,
513 (struct sockaddr *)&group,
514 (struct sockaddr *)&udp_in);
515 if (blocked != MCAST_PASS) {
516 if (blocked == MCAST_NOTGMEMBER)
517 IPSTAT_INC(ips_notmember);
518 if (blocked == MCAST_NOTSMEMBER ||
519 blocked == MCAST_MUTED)
520 UDPSTAT_INC(udps_filtermcast);
528 n = m_copy(m, 0, M_COPYALL);
529 up = intoudpcb(last);
530 if (up->u_tun_func == NULL) {
535 sizeof(struct udphdr),
539 * Engage the tunneling protocol we
540 * will have to leave the info_lock
541 * up, since we are hunting through
545 (*up->u_tun_func)(n, iphlen, last);
551 * Don't look for additional matches if this one does
552 * not have either the SO_REUSEPORT or SO_REUSEADDR
553 * socket options set. This heuristic avoids
554 * searching through all pcbs in the common case of a
555 * non-shared port. It assumes that an application
556 * will never clear these options after setting them.
558 if ((last->inp_socket->so_options &
559 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
565 * No matching pcb found; discard datagram. (No need
566 * to send an ICMP Port Unreachable for a broadcast
567 * or multicast datgram.)
569 UDPSTAT_INC(udps_noportbcast);
572 up = intoudpcb(last);
573 if (up->u_tun_func == NULL) {
574 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
578 * Engage the tunneling protocol.
580 (*up->u_tun_func)(m, iphlen, last);
583 INP_INFO_RUNLOCK(&V_udbinfo);
588 * Locate pcb for datagram.
590 inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
591 ip->ip_dst, uh->uh_dport, 1, ifp);
593 if (udp_log_in_vain) {
594 char buf[4*sizeof "123"];
596 strcpy(buf, inet_ntoa(ip->ip_dst));
598 "Connection attempt to UDP %s:%d from %s:%d\n",
599 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
600 ntohs(uh->uh_sport));
602 UDPSTAT_INC(udps_noport);
603 if (m->m_flags & (M_BCAST | M_MCAST)) {
604 UDPSTAT_INC(udps_noportbcast);
609 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
612 ip->ip_len += iphlen;
613 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
614 INP_INFO_RUNLOCK(&V_udbinfo);
619 * Check the minimum TTL for socket.
622 INP_INFO_RUNLOCK(&V_udbinfo);
623 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
628 if (up->u_tun_func == NULL) {
629 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
632 * Engage the tunneling protocol.
635 (*up->u_tun_func)(m, iphlen, inp);
643 INP_INFO_RUNLOCK(&V_udbinfo);
649 * Notify a udp user of an asynchronous error; just wake up so that they can
650 * collect error status.
653 udp_notify(struct inpcb *inp, int errno)
657 * While udp_ctlinput() always calls udp_notify() with a read lock
658 * when invoking it directly, in_pcbnotifyall() currently uses write
659 * locks due to sharing code with TCP. For now, accept either a read
660 * or a write lock, but a read lock is sufficient.
662 INP_LOCK_ASSERT(inp);
664 inp->inp_socket->so_error = errno;
665 sorwakeup(inp->inp_socket);
666 sowwakeup(inp->inp_socket);
671 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
675 struct in_addr faddr;
678 faddr = ((struct sockaddr_in *)sa)->sin_addr;
679 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
683 * Redirects don't need to be handled up here.
685 if (PRC_IS_REDIRECT(cmd))
689 * Hostdead is ugly because it goes linearly through all PCBs.
691 * XXX: We never get this from ICMP, otherwise it makes an excellent
692 * DoS attack on machines with many connections.
694 if (cmd == PRC_HOSTDEAD)
696 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
699 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
700 INP_INFO_RLOCK(&V_udbinfo);
701 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
702 ip->ip_src, uh->uh_sport, 0, NULL);
705 if (inp->inp_socket != NULL) {
706 udp_notify(inp, inetctlerrmap[cmd]);
710 INP_INFO_RUNLOCK(&V_udbinfo);
712 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
717 udp_pcblist(SYSCTL_HANDLER_ARGS)
720 struct inpcb *inp, **inp_list;
725 * The process of preparing the PCB list is too time-consuming and
726 * resource-intensive to repeat twice on every request.
728 if (req->oldptr == 0) {
729 n = V_udbinfo.ipi_count;
730 req->oldidx = 2 * (sizeof xig)
731 + (n + n/8) * sizeof(struct xinpcb);
735 if (req->newptr != 0)
739 * OK, now we're committed to doing something.
741 INP_INFO_RLOCK(&V_udbinfo);
742 gencnt = V_udbinfo.ipi_gencnt;
743 n = V_udbinfo.ipi_count;
744 INP_INFO_RUNLOCK(&V_udbinfo);
746 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
747 + n * sizeof(struct xinpcb));
751 xig.xig_len = sizeof xig;
753 xig.xig_gen = gencnt;
754 xig.xig_sogen = so_gencnt;
755 error = SYSCTL_OUT(req, &xig, sizeof xig);
759 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
763 INP_INFO_RLOCK(&V_udbinfo);
764 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
765 inp = LIST_NEXT(inp, inp_list)) {
767 if (inp->inp_gencnt <= gencnt &&
768 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
772 INP_INFO_RUNLOCK(&V_udbinfo);
776 for (i = 0; i < n; i++) {
779 if (inp->inp_gencnt <= gencnt) {
781 bzero(&xi, sizeof(xi));
782 xi.xi_len = sizeof xi;
783 /* XXX should avoid extra copy */
784 bcopy(inp, &xi.xi_inp, sizeof *inp);
786 sotoxsocket(inp->inp_socket, &xi.xi_socket);
787 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
789 error = SYSCTL_OUT(req, &xi, sizeof xi);
795 * Give the user an updated idea of our state. If the
796 * generation differs from what we told her before, she knows
797 * that something happened while we were processing this
798 * request, and it might be necessary to retry.
800 INP_INFO_RLOCK(&V_udbinfo);
801 xig.xig_gen = V_udbinfo.ipi_gencnt;
802 xig.xig_sogen = so_gencnt;
803 xig.xig_count = V_udbinfo.ipi_count;
804 INP_INFO_RUNLOCK(&V_udbinfo);
805 error = SYSCTL_OUT(req, &xig, sizeof xig);
807 free(inp_list, M_TEMP);
811 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
812 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
815 udp_getcred(SYSCTL_HANDLER_ARGS)
818 struct sockaddr_in addrs[2];
822 error = priv_check(req->td, PRIV_NETINET_GETCRED);
825 error = SYSCTL_IN(req, addrs, sizeof(addrs));
828 INP_INFO_RLOCK(&V_udbinfo);
829 inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
830 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
833 INP_INFO_RUNLOCK(&V_udbinfo);
834 if (inp->inp_socket == NULL)
837 error = cr_canseeinpcb(req->td->td_ucred, inp);
839 cru2x(inp->inp_cred, &xuc);
842 INP_INFO_RUNLOCK(&V_udbinfo);
846 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
850 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
851 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
852 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
855 udp_ctloutput(struct socket *so, struct sockopt *sopt)
857 int error = 0, optval;
864 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
866 if (sopt->sopt_level != IPPROTO_UDP) {
868 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
870 error = ip6_ctloutput(so, sopt);
874 error = ip_ctloutput(so, sopt);
881 switch (sopt->sopt_dir) {
883 switch (sopt->sopt_name) {
886 error = sooptcopyin(sopt, &optval, sizeof optval,
891 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
895 KASSERT(up != NULL, ("%s: up == NULL", __func__));
899 /* Clear all UDP encap. */
901 up->u_flags &= ~UF_ESPINUDP_ALL;
905 case UDP_ENCAP_ESPINUDP:
906 case UDP_ENCAP_ESPINUDP_NON_IKE:
907 up->u_flags &= ~UF_ESPINUDP_ALL;
908 if (optval == UDP_ENCAP_ESPINUDP)
909 up->u_flags |= UF_ESPINUDP;
910 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
911 up->u_flags |= UF_ESPINUDP_NON_IKE;
927 switch (sopt->sopt_name) {
931 KASSERT(up != NULL, ("%s: up == NULL", __func__));
932 optval = up->u_flags & UF_ESPINUDP_ALL;
934 error = sooptcopyout(sopt, &optval, sizeof optval);
948 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
949 struct mbuf *control, struct thread *td)
952 int len = m->m_pkthdr.len;
953 struct in_addr faddr, laddr;
955 struct sockaddr_in *sin, src;
958 u_short fport, lport;
962 * udp_output() may need to temporarily bind or connect the current
963 * inpcb. As such, we don't know up front whether we will need the
964 * pcbinfo lock or not. Do any work to decide what is needed up
965 * front before acquiring any locks.
967 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
975 if (control != NULL) {
977 * XXX: Currently, we assume all the optional information is
978 * stored in a single mbuf.
980 if (control->m_next) {
985 for (; control->m_len > 0;
986 control->m_data += CMSG_ALIGN(cm->cmsg_len),
987 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
988 cm = mtod(control, struct cmsghdr *);
989 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
990 || cm->cmsg_len > control->m_len) {
994 if (cm->cmsg_level != IPPROTO_IP)
997 switch (cm->cmsg_type) {
1000 CMSG_LEN(sizeof(struct in_addr))) {
1004 bzero(&src, sizeof(src));
1005 src.sin_family = AF_INET;
1006 src.sin_len = sizeof(src);
1007 src.sin_port = inp->inp_lport;
1009 *(struct in_addr *)CMSG_DATA(cm);
1013 error = ENOPROTOOPT;
1027 * Depending on whether or not the application has bound or connected
1028 * the socket, we may have to do varying levels of work. The optimal
1029 * case is for a connected UDP socket, as a global lock isn't
1032 * In order to decide which we need, we require stability of the
1033 * inpcb binding, which we ensure by acquiring a read lock on the
1034 * inpcb. This doesn't strictly follow the lock order, so we play
1035 * the trylock and retry game; note that we may end up with more
1036 * conservative locks than required the second time around, so later
1037 * assertions have to accept that. Further analysis of the number of
1038 * misses under contention is required.
1040 sin = (struct sockaddr_in *)addr;
1043 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1045 INP_INFO_WLOCK(&V_udbinfo);
1048 } else if ((sin != NULL && (
1049 (sin->sin_addr.s_addr == INADDR_ANY) ||
1050 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1051 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1052 (inp->inp_lport == 0))) ||
1053 (src.sin_family == AF_INET)) {
1054 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
1056 INP_INFO_RLOCK(&V_udbinfo);
1064 * If the IP_SENDSRCADDR control message was specified, override the
1065 * source address for this datagram. Its use is invalidated if the
1066 * address thus specified is incomplete or clobbers other inpcbs.
1068 laddr = inp->inp_laddr;
1069 lport = inp->inp_lport;
1070 if (src.sin_family == AF_INET) {
1071 INP_INFO_LOCK_ASSERT(&V_udbinfo);
1073 (laddr.s_addr == INADDR_ANY &&
1074 src.sin_addr.s_addr == INADDR_ANY)) {
1078 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1079 &laddr.s_addr, &lport, td->td_ucred);
1085 * If a UDP socket has been connected, then a local address/port will
1086 * have been selected and bound.
1088 * If a UDP socket has not been connected to, then an explicit
1089 * destination address must be used, in which case a local
1090 * address/port may not have been selected and bound.
1093 INP_LOCK_ASSERT(inp);
1094 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1100 * Jail may rewrite the destination address, so let it do
1101 * that before we use it.
1103 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1108 * If a local address or port hasn't yet been selected, or if
1109 * the destination address needs to be rewritten due to using
1110 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1111 * to do the heavy lifting. Once a port is selected, we
1112 * commit the binding back to the socket; we also commit the
1113 * binding of the address if in jail.
1115 * If we already have a valid binding and we're not
1116 * requesting a destination address rewrite, use a fast path.
1118 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1119 inp->inp_lport == 0 ||
1120 sin->sin_addr.s_addr == INADDR_ANY ||
1121 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1122 INP_INFO_LOCK_ASSERT(&V_udbinfo);
1123 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1124 &lport, &faddr.s_addr, &fport, NULL,
1130 * XXXRW: Why not commit the port if the address is
1133 /* Commit the local port if newly assigned. */
1134 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1135 inp->inp_lport == 0) {
1136 INP_INFO_WLOCK_ASSERT(&V_udbinfo);
1137 INP_WLOCK_ASSERT(inp);
1139 * Remember addr if jailed, to prevent
1142 if (prison_flag(td->td_ucred, PR_IP4))
1143 inp->inp_laddr = laddr;
1144 inp->inp_lport = lport;
1145 if (in_pcbinshash(inp) != 0) {
1150 inp->inp_flags |= INP_ANONPORT;
1153 faddr = sin->sin_addr;
1154 fport = sin->sin_port;
1157 INP_LOCK_ASSERT(inp);
1158 faddr = inp->inp_faddr;
1159 fport = inp->inp_fport;
1160 if (faddr.s_addr == INADDR_ANY) {
1167 * Calculate data length and get a mbuf for UDP, IP, and possible
1168 * link-layer headers. Immediate slide the data pointer back forward
1169 * since we won't use that space at this layer.
1171 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1176 m->m_data += max_linkhdr;
1177 m->m_len -= max_linkhdr;
1178 m->m_pkthdr.len -= max_linkhdr;
1181 * Fill in mbuf with extended UDP header and addresses and length put
1182 * into network format.
1184 ui = mtod(m, struct udpiphdr *);
1185 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1186 ui->ui_pr = IPPROTO_UDP;
1189 ui->ui_sport = lport;
1190 ui->ui_dport = fport;
1191 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1194 * Set the Don't Fragment bit in the IP header.
1196 if (inp->inp_flags & INP_DONTFRAG) {
1199 ip = (struct ip *)&ui->ui_i;
1200 ip->ip_off |= IP_DF;
1204 if (inp->inp_socket->so_options & SO_DONTROUTE)
1205 ipflags |= IP_ROUTETOIF;
1206 if (inp->inp_socket->so_options & SO_BROADCAST)
1207 ipflags |= IP_ALLOWBROADCAST;
1208 if (inp->inp_flags & INP_ONESBCAST)
1209 ipflags |= IP_SENDONES;
1212 mac_inpcb_create_mbuf(inp, m);
1216 * Set up checksum and output datagram.
1219 if (inp->inp_flags & INP_ONESBCAST)
1220 faddr.s_addr = INADDR_BROADCAST;
1221 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1222 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1223 m->m_pkthdr.csum_flags = CSUM_UDP;
1224 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1227 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1228 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1229 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */
1230 UDPSTAT_INC(udps_opackets);
1232 if (unlock_udbinfo == 2)
1233 INP_INFO_WUNLOCK(&V_udbinfo);
1234 else if (unlock_udbinfo == 1)
1235 INP_INFO_RUNLOCK(&V_udbinfo);
1236 error = ip_output(m, inp->inp_options, NULL, ipflags,
1237 inp->inp_moptions, inp);
1238 if (unlock_udbinfo == 2)
1245 if (unlock_udbinfo == 2) {
1247 INP_INFO_WUNLOCK(&V_udbinfo);
1248 } else if (unlock_udbinfo == 1) {
1250 INP_INFO_RUNLOCK(&V_udbinfo);
1258 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1261 * Potentially decap ESP in UDP frame. Check for an ESP header
1262 * and optional marker; if present, strip the UDP header and
1263 * push the result through IPSec.
1265 * Returns mbuf to be processed (potentially re-allocated) or
1266 * NULL if consumed and/or processed.
1268 static struct mbuf *
1269 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1271 size_t minlen, payload, skip, iphlen;
1275 struct udphdr *udphdr;
1278 INP_RLOCK_ASSERT(inp);
1281 * Pull up data so the longest case is contiguous:
1282 * IP/UDP hdr + non ESP marker + ESP hdr.
1284 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1285 if (minlen > m->m_pkthdr.len)
1286 minlen = m->m_pkthdr.len;
1287 if ((m = m_pullup(m, minlen)) == NULL) {
1288 V_ipsec4stat.in_inval++;
1289 return (NULL); /* Bypass caller processing. */
1291 data = mtod(m, caddr_t); /* Points to ip header. */
1292 payload = m->m_len - off; /* Size of payload. */
1294 if (payload == 1 && data[off] == '\xff')
1295 return (m); /* NB: keepalive packet, no decap. */
1297 up = intoudpcb(inp);
1298 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1299 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1300 ("u_flags 0x%x", up->u_flags));
1303 * Check that the payload is large enough to hold an
1304 * ESP header and compute the amount of data to remove.
1306 * NB: the caller has already done a pullup for us.
1307 * XXX can we assume alignment and eliminate bcopys?
1309 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1311 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1312 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1313 * possible AH mode non-IKE marker+non-ESP marker
1314 * from draft-ietf-ipsec-udp-encaps-00.txt.
1318 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1319 return (m); /* NB: no decap. */
1320 bcopy(data + off, &marker, sizeof(uint64_t));
1321 if (marker != 0) /* Non-IKE marker. */
1322 return (m); /* NB: no decap. */
1323 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1327 if (payload <= sizeof(struct esp)) {
1328 V_ipsec4stat.in_inval++;
1330 return (NULL); /* Discard. */
1332 bcopy(data + off, &spi, sizeof(uint32_t));
1333 if (spi == 0) /* Non-ESP marker. */
1334 return (m); /* NB: no decap. */
1335 skip = sizeof(struct udphdr);
1339 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1340 * the UDP ports. This is required if we want to select
1341 * the right SPD for multiple hosts behind same NAT.
1343 * NB: ports are maintained in network byte order everywhere
1344 * in the NAT-T code.
1346 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1347 2 * sizeof(uint16_t), M_NOWAIT);
1349 V_ipsec4stat.in_nomem++;
1351 return (NULL); /* Discard. */
1353 iphlen = off - sizeof(struct udphdr);
1354 udphdr = (struct udphdr *)(data + iphlen);
1355 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1356 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1357 m_tag_prepend(m, tag);
1360 * Remove the UDP header (and possibly the non ESP marker)
1361 * IP header length is iphlen
1364 * +----+------+-----+
1365 * | IP | UDP | ESP |
1366 * +----+------+-----+
1374 ovbcopy(data, data + skip, iphlen);
1377 ip = mtod(m, struct ip *);
1379 ip->ip_p = IPPROTO_ESP;
1382 * We cannot yet update the cksums so clear any
1383 * h/w cksum flags as they are no longer valid.
1385 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1386 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1388 (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1389 return (NULL); /* NB: consumed, bypass processing. */
1392 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1395 udp_abort(struct socket *so)
1399 inp = sotoinpcb(so);
1400 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1401 INP_INFO_WLOCK(&V_udbinfo);
1403 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1404 in_pcbdisconnect(inp);
1405 inp->inp_laddr.s_addr = INADDR_ANY;
1406 soisdisconnected(so);
1409 INP_INFO_WUNLOCK(&V_udbinfo);
1413 udp_attach(struct socket *so, int proto, struct thread *td)
1418 inp = sotoinpcb(so);
1419 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1420 error = soreserve(so, udp_sendspace, udp_recvspace);
1423 INP_INFO_WLOCK(&V_udbinfo);
1424 error = in_pcballoc(so, &V_udbinfo);
1426 INP_INFO_WUNLOCK(&V_udbinfo);
1430 inp = (struct inpcb *)so->so_pcb;
1431 inp->inp_vflag |= INP_IPV4;
1432 inp->inp_ip_ttl = V_ip_defttl;
1434 error = udp_newudpcb(inp);
1438 INP_INFO_WUNLOCK(&V_udbinfo);
1443 INP_INFO_WUNLOCK(&V_udbinfo);
1448 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1453 KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram"));
1454 KASSERT(so->so_pcb != NULL, ("udp_set_kernel_tunneling: NULL inp"));
1455 if (so->so_type != SOCK_DGRAM) {
1456 /* Not UDP socket... sorry! */
1459 inp = (struct inpcb *)so->so_pcb;
1465 up = intoudpcb(inp);
1466 if (up->u_tun_func != NULL) {
1476 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1481 inp = sotoinpcb(so);
1482 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1483 INP_INFO_WLOCK(&V_udbinfo);
1485 error = in_pcbbind(inp, nam, td->td_ucred);
1487 INP_INFO_WUNLOCK(&V_udbinfo);
1492 udp_close(struct socket *so)
1496 inp = sotoinpcb(so);
1497 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1498 INP_INFO_WLOCK(&V_udbinfo);
1500 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1501 in_pcbdisconnect(inp);
1502 inp->inp_laddr.s_addr = INADDR_ANY;
1503 soisdisconnected(so);
1506 INP_INFO_WUNLOCK(&V_udbinfo);
1510 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1514 struct sockaddr_in *sin;
1516 inp = sotoinpcb(so);
1517 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1518 INP_INFO_WLOCK(&V_udbinfo);
1520 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1522 INP_INFO_WUNLOCK(&V_udbinfo);
1525 sin = (struct sockaddr_in *)nam;
1526 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1529 INP_INFO_WUNLOCK(&V_udbinfo);
1532 error = in_pcbconnect(inp, nam, td->td_ucred);
1536 INP_INFO_WUNLOCK(&V_udbinfo);
1541 udp_detach(struct socket *so)
1546 inp = sotoinpcb(so);
1547 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1548 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1549 ("udp_detach: not disconnected"));
1550 INP_INFO_WLOCK(&V_udbinfo);
1552 up = intoudpcb(inp);
1553 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1554 inp->inp_ppcb = NULL;
1557 INP_INFO_WUNLOCK(&V_udbinfo);
1562 udp_disconnect(struct socket *so)
1566 inp = sotoinpcb(so);
1567 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1568 INP_INFO_WLOCK(&V_udbinfo);
1570 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1572 INP_INFO_WUNLOCK(&V_udbinfo);
1576 in_pcbdisconnect(inp);
1577 inp->inp_laddr.s_addr = INADDR_ANY;
1579 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1582 INP_INFO_WUNLOCK(&V_udbinfo);
1587 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1588 struct mbuf *control, struct thread *td)
1592 inp = sotoinpcb(so);
1593 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1594 return (udp_output(inp, m, addr, control, td));
1598 udp_shutdown(struct socket *so)
1602 inp = sotoinpcb(so);
1603 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1610 struct pr_usrreqs udp_usrreqs = {
1611 .pru_abort = udp_abort,
1612 .pru_attach = udp_attach,
1613 .pru_bind = udp_bind,
1614 .pru_connect = udp_connect,
1615 .pru_control = in_control,
1616 .pru_detach = udp_detach,
1617 .pru_disconnect = udp_disconnect,
1618 .pru_peeraddr = in_getpeeraddr,
1619 .pru_send = udp_send,
1620 .pru_soreceive = soreceive_dgram,
1621 .pru_sosend = sosend_dgram,
1622 .pru_shutdown = udp_shutdown,
1623 .pru_sockaddr = in_getsockaddr,
1624 .pru_sosetlabel = in_pcbsosetlabel,
1625 .pru_close = udp_close,