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 VNET_DEFINE(int, udp_cksum) = 1;
109 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
110 &VNET_NAME(udp_cksum), 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_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
147 VNET_PCPUSTAT_SYSINIT(udpstat);
148 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
149 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
152 VNET_PCPUSTAT_SYSUNINIT(udpstat);
155 static void udp_detach(struct socket *so);
156 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
157 struct mbuf *, struct thread *);
162 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
164 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
166 #endif /* IPSEC_NAT_T */
170 udp_zone_change(void *tag)
173 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
174 uma_zone_set_max(V_udpcb_zone, maxsockets);
178 udp_inpcb_init(void *mem, int size, int flags)
183 INP_LOCK_INIT(inp, "inp", "udpinp");
191 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
192 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
193 IPI_HASHFIELDS_2TUPLE);
194 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
195 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 uma_zone_set_max(V_udpcb_zone, maxsockets);
197 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
198 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
199 EVENTHANDLER_PRI_ANY);
203 * Kernel module interface for updating udpstat. The argument is an index
204 * into udpstat treated as an array of u_long. While this encodes the
205 * general layout of udpstat into the caller, it doesn't encode its location,
206 * so that future changes to add, for example, per-CPU stats support won't
207 * cause binary compatibility problems for kernel modules.
210 kmod_udpstat_inc(int statnum)
213 counter_u64_add(VNET(udpstat)[statnum], 1);
217 udp_newudpcb(struct inpcb *inp)
221 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
229 udp_discardcb(struct udpcb *up)
232 uma_zfree(V_udpcb_zone, up);
240 in_pcbinfo_destroy(&V_udbinfo);
241 uma_zdestroy(V_udpcb_zone);
247 * Subroutine of udp_input(), which appends the provided mbuf chain to the
248 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
249 * contains the source address. If the socket ends up being an IPv6 socket,
250 * udp_append() will convert to a sockaddr_in6 before passing the address
251 * into the socket code.
254 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
255 struct sockaddr_in *udp_in)
257 struct sockaddr *append_sa;
259 struct mbuf *opts = 0;
261 struct sockaddr_in6 udp_in6;
265 INP_LOCK_ASSERT(inp);
268 * Engage the tunneling protocol.
271 if (up->u_tun_func != NULL) {
272 (*up->u_tun_func)(n, off, inp);
279 off += sizeof(struct udphdr);
282 /* Check AH/ESP integrity. */
283 if (ipsec4_in_reject(n, inp)) {
285 IPSECSTAT_INC(in_polvio);
290 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
291 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
292 n = udp4_espdecap(inp, n, off);
293 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)
345 uint16_t len, ip_len;
347 struct sockaddr_in udp_in;
348 struct m_tag *fwd_tag;
350 ifp = m->m_pkthdr.rcvif;
351 UDPSTAT_INC(udps_ipackets);
354 * Strip IP options, if any; should skip this, make available to
355 * user, and use on returned packets, but we don't yet have a way to
356 * check the checksum with options still present.
358 if (iphlen > sizeof (struct ip)) {
360 iphlen = sizeof(struct ip);
364 * Get IP and UDP header together in first mbuf.
366 ip = mtod(m, struct ip *);
367 if (m->m_len < iphlen + sizeof(struct udphdr)) {
368 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
369 UDPSTAT_INC(udps_hdrops);
372 ip = mtod(m, struct ip *);
374 uh = (struct udphdr *)((caddr_t)ip + iphlen);
377 * Destination port of 0 is illegal, based on RFC768.
379 if (uh->uh_dport == 0)
383 * Construct sockaddr format source address. Stuff source address
384 * and datagram in user buffer.
386 bzero(&udp_in, sizeof(udp_in));
387 udp_in.sin_len = sizeof(udp_in);
388 udp_in.sin_family = AF_INET;
389 udp_in.sin_port = uh->uh_sport;
390 udp_in.sin_addr = ip->ip_src;
393 * Make mbuf data length reflect UDP length. If not enough data to
394 * reflect UDP length, drop.
396 len = ntohs((u_short)uh->uh_ulen);
397 ip_len = ntohs(ip->ip_len) - iphlen;
399 if (len > ip_len || len < sizeof(struct udphdr)) {
400 UDPSTAT_INC(udps_badlen);
403 m_adj(m, len - ip_len);
407 * Save a copy of the IP header in case we want restore it for
408 * sending an ICMP error message in response.
410 if (!V_udp_blackhole)
413 memset(&save_ip, 0, sizeof(save_ip));
416 * Checksum extended UDP header and data.
421 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
422 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
423 uh_sum = m->m_pkthdr.csum_data;
425 uh_sum = in_pseudo(ip->ip_src.s_addr,
426 ip->ip_dst.s_addr, htonl((u_short)len +
427 m->m_pkthdr.csum_data + IPPROTO_UDP));
432 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
433 bzero(((struct ipovly *)ip)->ih_x1, 9);
434 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
435 uh_sum = in_cksum(m, len + sizeof (struct ip));
436 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
439 UDPSTAT_INC(udps_badsum);
444 UDPSTAT_INC(udps_nosum);
446 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
447 in_broadcast(ip->ip_dst, ifp)) {
449 struct ip_moptions *imo;
451 INP_INFO_RLOCK(&V_udbinfo);
453 LIST_FOREACH(inp, &V_udb, inp_list) {
454 if (inp->inp_lport != uh->uh_dport)
457 if ((inp->inp_vflag & INP_IPV4) == 0)
460 if (inp->inp_laddr.s_addr != INADDR_ANY &&
461 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
463 if (inp->inp_faddr.s_addr != INADDR_ANY &&
464 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
466 if (inp->inp_fport != 0 &&
467 inp->inp_fport != uh->uh_sport)
473 * XXXRW: Because we weren't holding either the inpcb
474 * or the hash lock when we checked for a match
475 * before, we should probably recheck now that the
476 * inpcb lock is held.
480 * Handle socket delivery policy for any-source
481 * and source-specific multicast. [RFC3678]
483 imo = inp->inp_moptions;
484 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
485 struct sockaddr_in group;
491 bzero(&group, sizeof(struct sockaddr_in));
492 group.sin_len = sizeof(struct sockaddr_in);
493 group.sin_family = AF_INET;
494 group.sin_addr = ip->ip_dst;
496 blocked = imo_multi_filter(imo, ifp,
497 (struct sockaddr *)&group,
498 (struct sockaddr *)&udp_in);
499 if (blocked != MCAST_PASS) {
500 if (blocked == MCAST_NOTGMEMBER)
501 IPSTAT_INC(ips_notmember);
502 if (blocked == MCAST_NOTSMEMBER ||
503 blocked == MCAST_MUTED)
504 UDPSTAT_INC(udps_filtermcast);
512 n = m_copy(m, 0, M_COPYALL);
513 udp_append(last, ip, n, iphlen, &udp_in);
518 * Don't look for additional matches if this one does
519 * not have either the SO_REUSEPORT or SO_REUSEADDR
520 * socket options set. This heuristic avoids
521 * searching through all pcbs in the common case of a
522 * non-shared port. It assumes that an application
523 * will never clear these options after setting them.
525 if ((last->inp_socket->so_options &
526 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
532 * No matching pcb found; discard datagram. (No need
533 * to send an ICMP Port Unreachable for a broadcast
534 * or multicast datgram.)
536 UDPSTAT_INC(udps_noportbcast);
539 INP_INFO_RUNLOCK(&V_udbinfo);
542 udp_append(last, ip, m, iphlen, &udp_in);
544 INP_INFO_RUNLOCK(&V_udbinfo);
549 * Locate pcb for datagram.
553 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
555 if ((m->m_flags & M_IP_NEXTHOP) &&
556 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
557 struct sockaddr_in *next_hop;
559 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
562 * Transparently forwarded. Pretend to be the destination.
563 * Already got one like this?
565 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
566 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
569 * It's new. Try to find the ambushing socket.
570 * Because we've rewritten the destination address,
571 * any hardware-generated hash is ignored.
573 inp = in_pcblookup(&V_udbinfo, ip->ip_src,
574 uh->uh_sport, next_hop->sin_addr,
575 next_hop->sin_port ? htons(next_hop->sin_port) :
576 uh->uh_dport, INPLOOKUP_WILDCARD |
577 INPLOOKUP_RLOCKPCB, ifp);
579 /* Remove the tag from the packet. We don't need it anymore. */
580 m_tag_delete(m, fwd_tag);
581 m->m_flags &= ~M_IP_NEXTHOP;
583 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
584 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
585 INPLOOKUP_RLOCKPCB, ifp, m);
587 if (udp_log_in_vain) {
588 char buf[4*sizeof "123"];
590 strcpy(buf, inet_ntoa(ip->ip_dst));
592 "Connection attempt to UDP %s:%d from %s:%d\n",
593 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
594 ntohs(uh->uh_sport));
596 UDPSTAT_INC(udps_noport);
597 if (m->m_flags & (M_BCAST | M_MCAST)) {
598 UDPSTAT_INC(udps_noportbcast);
603 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
606 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
611 * Check the minimum TTL for socket.
613 INP_RLOCK_ASSERT(inp);
614 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
619 udp_append(inp, ip, m, iphlen, &udp_in);
629 * Notify a udp user of an asynchronous error; just wake up so that they can
630 * collect error status.
633 udp_notify(struct inpcb *inp, int errno)
637 * While udp_ctlinput() always calls udp_notify() with a read lock
638 * when invoking it directly, in_pcbnotifyall() currently uses write
639 * locks due to sharing code with TCP. For now, accept either a read
640 * or a write lock, but a read lock is sufficient.
642 INP_LOCK_ASSERT(inp);
644 inp->inp_socket->so_error = errno;
645 sorwakeup(inp->inp_socket);
646 sowwakeup(inp->inp_socket);
652 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
656 struct in_addr faddr;
659 faddr = ((struct sockaddr_in *)sa)->sin_addr;
660 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
664 * Redirects don't need to be handled up here.
666 if (PRC_IS_REDIRECT(cmd))
670 * Hostdead is ugly because it goes linearly through all PCBs.
672 * XXX: We never get this from ICMP, otherwise it makes an excellent
673 * DoS attack on machines with many connections.
675 if (cmd == PRC_HOSTDEAD)
677 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
680 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
681 inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
682 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
684 INP_RLOCK_ASSERT(inp);
685 if (inp->inp_socket != NULL) {
686 udp_notify(inp, inetctlerrmap[cmd]);
691 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
697 udp_pcblist(SYSCTL_HANDLER_ARGS)
700 struct inpcb *inp, **inp_list;
705 * The process of preparing the PCB list is too time-consuming and
706 * resource-intensive to repeat twice on every request.
708 if (req->oldptr == 0) {
709 n = V_udbinfo.ipi_count;
710 n += imax(n / 8, 10);
711 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
715 if (req->newptr != 0)
719 * OK, now we're committed to doing something.
721 INP_INFO_RLOCK(&V_udbinfo);
722 gencnt = V_udbinfo.ipi_gencnt;
723 n = V_udbinfo.ipi_count;
724 INP_INFO_RUNLOCK(&V_udbinfo);
726 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
727 + n * sizeof(struct xinpcb));
731 xig.xig_len = sizeof xig;
733 xig.xig_gen = gencnt;
734 xig.xig_sogen = so_gencnt;
735 error = SYSCTL_OUT(req, &xig, sizeof xig);
739 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
743 INP_INFO_RLOCK(&V_udbinfo);
744 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
745 inp = LIST_NEXT(inp, inp_list)) {
747 if (inp->inp_gencnt <= gencnt &&
748 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
754 INP_INFO_RUNLOCK(&V_udbinfo);
758 for (i = 0; i < n; i++) {
761 if (inp->inp_gencnt <= gencnt) {
764 bzero(&xi, sizeof(xi));
765 xi.xi_len = sizeof xi;
766 /* XXX should avoid extra copy */
767 bcopy(inp, &xi.xi_inp, sizeof *inp);
769 sotoxsocket(inp->inp_socket, &xi.xi_socket);
770 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
772 error = SYSCTL_OUT(req, &xi, sizeof xi);
776 INP_INFO_WLOCK(&V_udbinfo);
777 for (i = 0; i < n; i++) {
780 if (!in_pcbrele_rlocked(inp))
783 INP_INFO_WUNLOCK(&V_udbinfo);
787 * Give the user an updated idea of our state. If the
788 * generation differs from what we told her before, she knows
789 * that something happened while we were processing this
790 * request, and it might be necessary to retry.
792 INP_INFO_RLOCK(&V_udbinfo);
793 xig.xig_gen = V_udbinfo.ipi_gencnt;
794 xig.xig_sogen = so_gencnt;
795 xig.xig_count = V_udbinfo.ipi_count;
796 INP_INFO_RUNLOCK(&V_udbinfo);
797 error = SYSCTL_OUT(req, &xig, sizeof xig);
799 free(inp_list, M_TEMP);
803 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
804 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
805 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
809 udp_getcred(SYSCTL_HANDLER_ARGS)
812 struct sockaddr_in addrs[2];
816 error = priv_check(req->td, PRIV_NETINET_GETCRED);
819 error = SYSCTL_IN(req, addrs, sizeof(addrs));
822 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
823 addrs[0].sin_addr, addrs[0].sin_port,
824 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
826 INP_RLOCK_ASSERT(inp);
827 if (inp->inp_socket == NULL)
830 error = cr_canseeinpcb(req->td->td_ucred, inp);
832 cru2x(inp->inp_cred, &xuc);
837 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
841 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
842 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
843 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
847 udp_ctloutput(struct socket *so, struct sockopt *sopt)
849 int error = 0, optval;
856 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
858 if (sopt->sopt_level != IPPROTO_UDP) {
860 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
862 error = ip6_ctloutput(so, sopt);
865 #if defined(INET) && defined(INET6)
871 error = ip_ctloutput(so, sopt);
877 switch (sopt->sopt_dir) {
879 switch (sopt->sopt_name) {
882 error = sooptcopyin(sopt, &optval, sizeof optval,
887 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
891 KASSERT(up != NULL, ("%s: up == NULL", __func__));
895 /* Clear all UDP encap. */
897 up->u_flags &= ~UF_ESPINUDP_ALL;
901 case UDP_ENCAP_ESPINUDP:
902 case UDP_ENCAP_ESPINUDP_NON_IKE:
903 up->u_flags &= ~UF_ESPINUDP_ALL;
904 if (optval == UDP_ENCAP_ESPINUDP)
905 up->u_flags |= UF_ESPINUDP;
906 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
907 up->u_flags |= UF_ESPINUDP_NON_IKE;
923 switch (sopt->sopt_name) {
927 KASSERT(up != NULL, ("%s: up == NULL", __func__));
928 optval = up->u_flags & UF_ESPINUDP_ALL;
930 error = sooptcopyout(sopt, &optval, sizeof optval);
946 #define UH_UNLOCKED 0
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;
963 * udp_output() may need to temporarily bind or connect the current
964 * inpcb. As such, we don't know up front whether we will need the
965 * pcbinfo lock or not. Do any work to decide what is needed up
966 * front before acquiring any locks.
968 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
977 tos = inp->inp_ip_tos;
978 if (control != NULL) {
980 * XXX: Currently, we assume all the optional information is
981 * stored in a single mbuf.
983 if (control->m_next) {
989 for (; control->m_len > 0;
990 control->m_data += CMSG_ALIGN(cm->cmsg_len),
991 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
992 cm = mtod(control, struct cmsghdr *);
993 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
994 || cm->cmsg_len > control->m_len) {
998 if (cm->cmsg_level != IPPROTO_IP)
1001 switch (cm->cmsg_type) {
1002 case IP_SENDSRCADDR:
1004 CMSG_LEN(sizeof(struct in_addr))) {
1008 bzero(&src, sizeof(src));
1009 src.sin_family = AF_INET;
1010 src.sin_len = sizeof(src);
1011 src.sin_port = inp->inp_lport;
1013 *(struct in_addr *)CMSG_DATA(cm);
1017 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1021 tos = *(u_char *)CMSG_DATA(cm);
1025 error = ENOPROTOOPT;
1040 * Depending on whether or not the application has bound or connected
1041 * the socket, we may have to do varying levels of work. The optimal
1042 * case is for a connected UDP socket, as a global lock isn't
1045 * In order to decide which we need, we require stability of the
1046 * inpcb binding, which we ensure by acquiring a read lock on the
1047 * inpcb. This doesn't strictly follow the lock order, so we play
1048 * the trylock and retry game; note that we may end up with more
1049 * conservative locks than required the second time around, so later
1050 * assertions have to accept that. Further analysis of the number of
1051 * misses under contention is required.
1053 * XXXRW: Check that hash locking update here is correct.
1055 sin = (struct sockaddr_in *)addr;
1057 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1060 INP_HASH_WLOCK(&V_udbinfo);
1061 unlock_udbinfo = UH_WLOCKED;
1062 } else if ((sin != NULL && (
1063 (sin->sin_addr.s_addr == INADDR_ANY) ||
1064 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1065 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1066 (inp->inp_lport == 0))) ||
1067 (src.sin_family == AF_INET)) {
1068 INP_HASH_RLOCK(&V_udbinfo);
1069 unlock_udbinfo = UH_RLOCKED;
1071 unlock_udbinfo = UH_UNLOCKED;
1074 * If the IP_SENDSRCADDR control message was specified, override the
1075 * source address for this datagram. Its use is invalidated if the
1076 * address thus specified is incomplete or clobbers other inpcbs.
1078 laddr = inp->inp_laddr;
1079 lport = inp->inp_lport;
1080 if (src.sin_family == AF_INET) {
1081 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1083 (laddr.s_addr == INADDR_ANY &&
1084 src.sin_addr.s_addr == INADDR_ANY)) {
1088 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1089 &laddr.s_addr, &lport, td->td_ucred);
1095 * If a UDP socket has been connected, then a local address/port will
1096 * have been selected and bound.
1098 * If a UDP socket has not been connected to, then an explicit
1099 * destination address must be used, in which case a local
1100 * address/port may not have been selected and bound.
1103 INP_LOCK_ASSERT(inp);
1104 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1110 * Jail may rewrite the destination address, so let it do
1111 * that before we use it.
1113 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1118 * If a local address or port hasn't yet been selected, or if
1119 * the destination address needs to be rewritten due to using
1120 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1121 * to do the heavy lifting. Once a port is selected, we
1122 * commit the binding back to the socket; we also commit the
1123 * binding of the address if in jail.
1125 * If we already have a valid binding and we're not
1126 * requesting a destination address rewrite, use a fast path.
1128 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1129 inp->inp_lport == 0 ||
1130 sin->sin_addr.s_addr == INADDR_ANY ||
1131 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1132 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1133 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1134 &lport, &faddr.s_addr, &fport, NULL,
1140 * XXXRW: Why not commit the port if the address is
1143 /* Commit the local port if newly assigned. */
1144 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1145 inp->inp_lport == 0) {
1146 INP_WLOCK_ASSERT(inp);
1147 INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1149 * Remember addr if jailed, to prevent
1152 if (prison_flag(td->td_ucred, PR_IP4))
1153 inp->inp_laddr = laddr;
1154 inp->inp_lport = lport;
1155 if (in_pcbinshash(inp) != 0) {
1160 inp->inp_flags |= INP_ANONPORT;
1163 faddr = sin->sin_addr;
1164 fport = sin->sin_port;
1167 INP_LOCK_ASSERT(inp);
1168 faddr = inp->inp_faddr;
1169 fport = inp->inp_fport;
1170 if (faddr.s_addr == INADDR_ANY) {
1177 * Calculate data length and get a mbuf for UDP, IP, and possible
1178 * link-layer headers. Immediate slide the data pointer back forward
1179 * since we won't use that space at this layer.
1181 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1186 m->m_data += max_linkhdr;
1187 m->m_len -= max_linkhdr;
1188 m->m_pkthdr.len -= max_linkhdr;
1191 * Fill in mbuf with extended UDP header and addresses and length put
1192 * into network format.
1194 ui = mtod(m, struct udpiphdr *);
1195 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1196 ui->ui_pr = IPPROTO_UDP;
1199 ui->ui_sport = lport;
1200 ui->ui_dport = fport;
1201 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1204 * Set the Don't Fragment bit in the IP header.
1206 if (inp->inp_flags & INP_DONTFRAG) {
1209 ip = (struct ip *)&ui->ui_i;
1210 ip->ip_off |= htons(IP_DF);
1214 if (inp->inp_socket->so_options & SO_DONTROUTE)
1215 ipflags |= IP_ROUTETOIF;
1216 if (inp->inp_socket->so_options & SO_BROADCAST)
1217 ipflags |= IP_ALLOWBROADCAST;
1218 if (inp->inp_flags & INP_ONESBCAST)
1219 ipflags |= IP_SENDONES;
1222 mac_inpcb_create_mbuf(inp, m);
1226 * Set up checksum and output datagram.
1229 if (inp->inp_flags & INP_ONESBCAST)
1230 faddr.s_addr = INADDR_BROADCAST;
1231 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1232 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1233 m->m_pkthdr.csum_flags = CSUM_UDP;
1234 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1237 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1238 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1239 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1240 UDPSTAT_INC(udps_opackets);
1242 if (unlock_udbinfo == UH_WLOCKED)
1243 INP_HASH_WUNLOCK(&V_udbinfo);
1244 else if (unlock_udbinfo == UH_RLOCKED)
1245 INP_HASH_RUNLOCK(&V_udbinfo);
1246 error = ip_output(m, inp->inp_options, NULL, ipflags,
1247 inp->inp_moptions, inp);
1248 if (unlock_udbinfo == UH_WLOCKED)
1255 if (unlock_udbinfo == UH_WLOCKED) {
1256 INP_HASH_WUNLOCK(&V_udbinfo);
1258 } else if (unlock_udbinfo == UH_RLOCKED) {
1259 INP_HASH_RUNLOCK(&V_udbinfo);
1268 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1270 * Potentially decap ESP in UDP frame. Check for an ESP header
1271 * and optional marker; if present, strip the UDP header and
1272 * push the result through IPSec.
1274 * Returns mbuf to be processed (potentially re-allocated) or
1275 * NULL if consumed and/or processed.
1277 static struct mbuf *
1278 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1280 size_t minlen, payload, skip, iphlen;
1284 struct udphdr *udphdr;
1287 INP_RLOCK_ASSERT(inp);
1290 * Pull up data so the longest case is contiguous:
1291 * IP/UDP hdr + non ESP marker + ESP hdr.
1293 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1294 if (minlen > m->m_pkthdr.len)
1295 minlen = m->m_pkthdr.len;
1296 if ((m = m_pullup(m, minlen)) == NULL) {
1297 IPSECSTAT_INC(in_inval);
1298 return (NULL); /* Bypass caller processing. */
1300 data = mtod(m, caddr_t); /* Points to ip header. */
1301 payload = m->m_len - off; /* Size of payload. */
1303 if (payload == 1 && data[off] == '\xff')
1304 return (m); /* NB: keepalive packet, no decap. */
1306 up = intoudpcb(inp);
1307 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1308 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1309 ("u_flags 0x%x", up->u_flags));
1312 * Check that the payload is large enough to hold an
1313 * ESP header and compute the amount of data to remove.
1315 * NB: the caller has already done a pullup for us.
1316 * XXX can we assume alignment and eliminate bcopys?
1318 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1320 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1321 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1322 * possible AH mode non-IKE marker+non-ESP marker
1323 * from draft-ietf-ipsec-udp-encaps-00.txt.
1327 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1328 return (m); /* NB: no decap. */
1329 bcopy(data + off, &marker, sizeof(uint64_t));
1330 if (marker != 0) /* Non-IKE marker. */
1331 return (m); /* NB: no decap. */
1332 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1336 if (payload <= sizeof(struct esp)) {
1337 IPSECSTAT_INC(in_inval);
1339 return (NULL); /* Discard. */
1341 bcopy(data + off, &spi, sizeof(uint32_t));
1342 if (spi == 0) /* Non-ESP marker. */
1343 return (m); /* NB: no decap. */
1344 skip = sizeof(struct udphdr);
1348 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1349 * the UDP ports. This is required if we want to select
1350 * the right SPD for multiple hosts behind same NAT.
1352 * NB: ports are maintained in network byte order everywhere
1353 * in the NAT-T code.
1355 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1356 2 * sizeof(uint16_t), M_NOWAIT);
1358 IPSECSTAT_INC(in_nomem);
1360 return (NULL); /* Discard. */
1362 iphlen = off - sizeof(struct udphdr);
1363 udphdr = (struct udphdr *)(data + iphlen);
1364 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1365 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1366 m_tag_prepend(m, tag);
1369 * Remove the UDP header (and possibly the non ESP marker)
1370 * IP header length is iphlen
1373 * +----+------+-----+
1374 * | IP | UDP | ESP |
1375 * +----+------+-----+
1383 ovbcopy(data, data + skip, iphlen);
1386 ip = mtod(m, struct ip *);
1387 ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1388 ip->ip_p = IPPROTO_ESP;
1391 * We cannot yet update the cksums so clear any
1392 * h/w cksum flags as they are no longer valid.
1394 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1395 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1397 (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1398 return (NULL); /* NB: consumed, bypass processing. */
1400 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1403 udp_abort(struct socket *so)
1407 inp = sotoinpcb(so);
1408 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1410 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1411 INP_HASH_WLOCK(&V_udbinfo);
1412 in_pcbdisconnect(inp);
1413 inp->inp_laddr.s_addr = INADDR_ANY;
1414 INP_HASH_WUNLOCK(&V_udbinfo);
1415 soisdisconnected(so);
1421 udp_attach(struct socket *so, int proto, struct thread *td)
1426 inp = sotoinpcb(so);
1427 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1428 error = soreserve(so, udp_sendspace, udp_recvspace);
1431 INP_INFO_WLOCK(&V_udbinfo);
1432 error = in_pcballoc(so, &V_udbinfo);
1434 INP_INFO_WUNLOCK(&V_udbinfo);
1438 inp = sotoinpcb(so);
1439 inp->inp_vflag |= INP_IPV4;
1440 inp->inp_ip_ttl = V_ip_defttl;
1442 error = udp_newudpcb(inp);
1446 INP_INFO_WUNLOCK(&V_udbinfo);
1451 INP_INFO_WUNLOCK(&V_udbinfo);
1457 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1462 KASSERT(so->so_type == SOCK_DGRAM,
1463 ("udp_set_kernel_tunneling: !dgram"));
1464 inp = sotoinpcb(so);
1465 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1467 up = intoudpcb(inp);
1468 if (up->u_tun_func != NULL) {
1479 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1484 inp = sotoinpcb(so);
1485 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1487 INP_HASH_WLOCK(&V_udbinfo);
1488 error = in_pcbbind(inp, nam, td->td_ucred);
1489 INP_HASH_WUNLOCK(&V_udbinfo);
1495 udp_close(struct socket *so)
1499 inp = sotoinpcb(so);
1500 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1502 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1503 INP_HASH_WLOCK(&V_udbinfo);
1504 in_pcbdisconnect(inp);
1505 inp->inp_laddr.s_addr = INADDR_ANY;
1506 INP_HASH_WUNLOCK(&V_udbinfo);
1507 soisdisconnected(so);
1513 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1517 struct sockaddr_in *sin;
1519 inp = sotoinpcb(so);
1520 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1522 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1526 sin = (struct sockaddr_in *)nam;
1527 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1532 INP_HASH_WLOCK(&V_udbinfo);
1533 error = in_pcbconnect(inp, nam, td->td_ucred);
1534 INP_HASH_WUNLOCK(&V_udbinfo);
1542 udp_detach(struct socket *so)
1547 inp = sotoinpcb(so);
1548 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1549 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1550 ("udp_detach: not disconnected"));
1551 INP_INFO_WLOCK(&V_udbinfo);
1553 up = intoudpcb(inp);
1554 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1555 inp->inp_ppcb = NULL;
1558 INP_INFO_WUNLOCK(&V_udbinfo);
1563 udp_disconnect(struct socket *so)
1567 inp = sotoinpcb(so);
1568 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1570 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1574 INP_HASH_WLOCK(&V_udbinfo);
1575 in_pcbdisconnect(inp);
1576 inp->inp_laddr.s_addr = INADDR_ANY;
1577 INP_HASH_WUNLOCK(&V_udbinfo);
1579 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1586 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1587 struct mbuf *control, struct thread *td)
1591 inp = sotoinpcb(so);
1592 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1593 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"));
1611 struct pr_usrreqs udp_usrreqs = {
1612 .pru_abort = udp_abort,
1613 .pru_attach = udp_attach,
1614 .pru_bind = udp_bind,
1615 .pru_connect = udp_connect,
1616 .pru_control = in_control,
1617 .pru_detach = udp_detach,
1618 .pru_disconnect = udp_disconnect,
1619 .pru_peeraddr = in_getpeeraddr,
1620 .pru_send = udp_send,
1621 .pru_soreceive = soreceive_dgram,
1622 .pru_sosend = sosend_dgram,
1623 .pru_shutdown = udp_shutdown,
1624 .pru_sockaddr = in_getsockaddr,
1625 .pru_sosetlabel = in_pcbsosetlabel,
1626 .pru_close = udp_close,