2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
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 * 3. 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
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
39 #error "IPDIVERT requires INET"
42 #include <sys/param.h>
43 #include <sys/eventhandler.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
48 #include <sys/module.h>
49 #include <sys/kernel.h>
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
59 #include <net/if_var.h>
60 #include <net/netisr.h>
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
72 #if defined(SCTP) || defined(SCTP_SUPPORT)
73 #include <netinet/sctp_crc32.h>
76 #include <security/mac/mac_framework.h>
82 * Allocate enough space to hold a full IP packet
84 #define DIVSNDQ (65536 + 100)
85 #define DIVRCVQ (65536 + 100)
88 * Divert sockets work in conjunction with ipfw or other packet filters,
89 * see the divert(4) manpage for features.
90 * Packets are selected by the packet filter and tagged with an
91 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
92 * the packet filter) and information on the matching filter rule for
93 * subsequent reinjection. The divert_port is used to put the packet
94 * on the corresponding divert socket, while the rule number is passed
95 * up (at least partially) as the sin_port in the struct sockaddr.
97 * Packets written to the divert socket carry in sin_addr a
98 * destination address, and in sin_port the number of the filter rule
99 * after which to continue processing.
100 * If the destination address is INADDR_ANY, the packet is treated as
101 * as outgoing and sent to ip_output(); otherwise it is treated as
102 * incoming and sent to ip_input().
103 * Further, sin_zero carries some information on the interface,
104 * which can be used in the reinject -- see comments in the code.
106 * On reinjection, processing in ip_input() and ip_output()
107 * will be exactly the same as for the original packet, except that
108 * packet filter processing will start at the rule number after the one
109 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
110 * will apply the entire ruleset to the packet).
113 /* Internal variables. */
114 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
115 #define V_divcbinfo VNET(divcbinfo)
117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
120 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
121 struct sockaddr_in *sin);
122 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
125 * Initialize divert connection block queue.
127 INPCBSTORAGE_DEFINE(divcbstor, "divinp", "divcb", "div", "divhash");
130 div_init(void *arg __unused)
134 * XXX We don't use the hash list for divert IP, but it's easier to
135 * allocate one-entry hash lists than it is to check all over the
136 * place for hashbase == NULL.
138 in_pcbinfo_init(&V_divcbinfo, &divcbstor, 1, 1);
140 VNET_SYSINIT(div_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_init, NULL);
143 div_destroy(void *unused __unused)
146 in_pcbinfo_destroy(&V_divcbinfo);
148 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_destroy, NULL);
151 * IPPROTO_DIVERT is not in the real IP protocol number space; this
152 * function should never be called. Just in case, drop any packets.
155 div_input(struct mbuf **mp, int *offp, int proto)
157 struct mbuf *m = *mp;
159 KMOD_IPSTAT_INC(ips_noproto);
161 return (IPPROTO_DONE);
165 div_port_match(const struct inpcb *inp, void *v)
167 uint16_t nport = *(uint16_t *)v;
169 return (inp->inp_lport == nport);
173 * Divert a packet by passing it up to the divert socket at port 'port'.
175 * Setup generic address and protocol structures for div_input routine,
176 * then pass them along with mbuf chain.
179 divert_packet(struct mbuf *m, bool incoming)
181 #if defined(SCTP) || defined(SCTP_SUPPORT)
187 struct sockaddr_in divsrc;
188 struct inpcb_iterator inpi = INP_ITERATOR(&V_divcbinfo,
189 INPLOOKUP_RLOCKPCB, div_port_match, &nport);
194 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
200 if (m->m_len < sizeof(struct ip) &&
201 (m = m_pullup(m, sizeof(struct ip))) == NULL)
204 /* Delayed checksums are currently not compatible with divert. */
205 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
207 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
209 #if defined(SCTP) || defined(SCTP_SUPPORT)
210 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
211 ip = mtod(m, struct ip *);
212 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
213 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
217 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
218 in6_delayed_cksum(m, m->m_pkthdr.len -
219 sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
220 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
222 #if defined(SCTP) || defined(SCTP_SUPPORT)
223 if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
224 sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
225 m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
229 bzero(&divsrc, sizeof(divsrc));
230 divsrc.sin_len = sizeof(divsrc);
231 divsrc.sin_family = AF_INET;
232 /* record matching rule, in host format */
233 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
235 * Record receive interface address, if any.
236 * But only for incoming packets.
245 /* Find IP address for receive interface */
246 ifp = m->m_pkthdr.rcvif;
247 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
248 if (ifa->ifa_addr->sa_family != AF_INET)
251 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
256 * Record the incoming interface name whenever we have one.
258 if (m->m_pkthdr.rcvif) {
260 * Hide the actual interface name in there in the
261 * sin_zero array. XXX This needs to be moved to a
262 * different sockaddr type for divert, e.g.
263 * sockaddr_div with multiple fields like
264 * sockaddr_dl. Presently we have only 7 bytes
265 * but that will do for now as most interfaces
266 * are 4 or less + 2 or less bytes for unit.
267 * There is probably a faster way of doing this,
268 * possibly taking it from the sockaddr_dl on the iface.
269 * This solves the problem of a P2P link and a LAN interface
270 * having the same address, which can result in the wrong
271 * interface being assigned to the packet when fed back
272 * into the divert socket. Theoretically if the daemon saves
273 * and re-uses the sockaddr_in as suggested in the man pages,
274 * this iface name will come along for the ride.
275 * (see div_output for the other half of this.)
277 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
278 sizeof(divsrc.sin_zero));
281 /* Put packet on socket queue, if any */
283 /* nport is inp_next's context. */
284 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
285 while ((inp = inp_next(&inpi)) != NULL) {
286 sa = inp->inp_socket;
287 SOCKBUF_LOCK(&sa->so_rcv);
288 if (sbappendaddr_locked(&sa->so_rcv,
289 (struct sockaddr *)&divsrc, m, NULL) == 0) {
290 soroverflow_locked(sa);
291 sa = NULL; /* force mbuf reclaim below */
293 sorwakeup_locked(sa);
294 /* XXX why does only one socket match? */
300 KMOD_IPSTAT_INC(ips_noproto);
301 KMOD_IPSTAT_DEC(ips_delivered);
306 * Deliver packet back into the IP processing machinery.
308 * If no address specified, or address is 0.0.0.0, send to ip_output();
309 * otherwise, send to ip_input() and mark as having been received on
310 * the interface with that address.
313 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
314 struct mbuf *control)
316 struct epoch_tracker et;
319 struct ipfw_rule_ref *dt;
323 m_freem(control); /* XXX */
328 if (sin->sin_family != AF_INET) {
330 return (EAFNOSUPPORT);
332 if (sin->sin_len != sizeof(*sin)) {
339 * An mbuf may hasn't come from userland, but we pretend
342 m->m_pkthdr.rcvif = NULL;
344 M_SETFIB(m, so->so_fibnum);
346 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
348 /* this should be normal */
349 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
350 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
355 m_tag_prepend(m, mtag);
357 dt = (struct ipfw_rule_ref *)(mtag+1);
359 /* Loopback avoidance and state recovery */
363 /* set the starting point. We provide a non-zero slot,
364 * but a non_matching chain_id to skip that info and use
365 * the rulenum/rule_id.
367 dt->slot = 1; /* dummy, chain_id is invalid */
369 dt->rulenum = sin->sin_port+1; /* host format ? */
371 /* XXX: broken for IPv6 */
373 * Find receive interface with the given name, stuffed
374 * (if it exists) in the sin_zero[] field.
375 * The name is user supplied data so don't trust its size
376 * or that it is zero terminated.
378 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
380 if ( i > 0 && i < sizeof(sin->sin_zero))
381 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
384 ip = mtod(m, struct ip *);
390 case IPV6_VERSION >> 4:
396 return (EAFNOSUPPORT);
399 /* Reinject packet into the system as incoming or outgoing */
401 if (!sin || sin->sin_addr.s_addr == 0) {
402 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
403 error = div_output_outbound(family, so, m);
405 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
406 error = div_output_inbound(family, so, m, sin);
414 * Sends mbuf @m to the wire via ip[6]_output().
416 * Returns 0 on success or an errno value on failure. @m is always consumed.
419 div_output_outbound(int family, struct socket *so, struct mbuf *m)
421 struct ip *const ip = mtod(m, struct ip *);
422 struct mbuf *options;
431 * Don't allow both user specified and setsockopt
432 * options, and don't allow packet length sizes that
435 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
436 inp->inp_options != NULL) ||
437 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
446 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
448 /* Don't allow packet length sizes that will crash */
449 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
459 /* Send packet to output processing */
460 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
463 mac_inpcb_create_mbuf(inp, m);
466 * Get ready to inject the packet into ip_output().
467 * Just in case socket options were specified on the
468 * divert socket, we duplicate them. This is done
469 * to avoid having to hold the PCB locks over the call
470 * to ip_output(), as doing this results in a number of
471 * lock ordering complexities.
473 * Note that we set the multicast options argument for
474 * ip_output() to NULL since it should be invariant that
475 * they are not present.
477 KASSERT(inp->inp_moptions == NULL,
478 ("multicast options set on a divert socket"));
480 * XXXCSJP: It is unclear to me whether or not it makes
481 * sense for divert sockets to have options. However,
482 * for now we will duplicate them with the INP locks
483 * held so we can use them in ip_output() without
484 * requring a reference to the pcb.
487 if (inp->inp_options != NULL) {
488 options = m_dup(inp->inp_options, M_NOWAIT);
489 if (options == NULL) {
500 error = ip_output(m, options, NULL,
501 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
502 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
506 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
517 * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
519 * Returns 0 on success or an errno value on failure. @m is always consumed.
522 div_output_inbound(int family, struct socket *so, struct mbuf *m,
523 struct sockaddr_in *sin)
528 if (m->m_pkthdr.rcvif == NULL) {
530 * No luck with the name, check by IP address.
531 * Clear the port and the ifname to make sure
532 * there are no distractions for ifa_ifwithaddr.
535 /* XXX: broken for IPv6 */
536 bzero(sin->sin_zero, sizeof(sin->sin_zero));
538 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
541 return (EADDRNOTAVAIL);
543 m->m_pkthdr.rcvif = ifa->ifa_ifp;
546 mac_socket_create_mbuf(so, m);
548 /* Send packet to input processing via netisr */
551 ip = mtod(m, struct ip *);
553 * Restore M_BCAST flag when destination address is
554 * broadcast. It is expected by ip_tryforward().
556 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
557 m->m_flags |= M_MCAST;
558 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
559 m->m_flags |= M_BCAST;
560 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
564 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
576 div_attach(struct socket *so, int proto, struct thread *td)
582 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
584 error = priv_check(td, PRIV_NETINET_DIVERT);
588 error = soreserve(so, div_sendspace, div_recvspace);
591 error = in_pcballoc(so, &V_divcbinfo);
594 inp = (struct inpcb *)so->so_pcb;
595 inp->inp_ip_p = proto;
596 inp->inp_flags |= INP_HDRINCL;
602 div_detach(struct socket *so)
607 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
614 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
620 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
621 /* in_pcbbind assumes that nam is a sockaddr_in
622 * and in_pcbbind requires a valid address. Since divert
623 * sockets don't we need to make sure the address is
624 * filled in properly.
625 * XXX -- divert should not be abusing in_pcbind
626 * and should probably have its own family.
628 if (nam->sa_family != AF_INET)
630 if (nam->sa_len != sizeof(struct sockaddr_in))
632 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
634 INP_HASH_WLOCK(&V_divcbinfo);
635 error = in_pcbbind(inp, nam, td->td_ucred);
636 INP_HASH_WUNLOCK(&V_divcbinfo);
642 div_shutdown(struct socket *so)
647 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
655 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
656 struct mbuf *control, struct thread *td)
659 /* Packet must have a header (but that's about it) */
660 if (m->m_len < sizeof (struct ip) &&
661 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
662 KMOD_IPSTAT_INC(ips_toosmall);
670 return div_output(so, m, (struct sockaddr_in *)nam, control);
674 div_pcblist(SYSCTL_HANDLER_ARGS)
676 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_divcbinfo,
682 if (req->newptr != 0)
685 if (req->oldptr == 0) {
688 n = V_divcbinfo.ipi_count;
689 n += imax(n / 8, 10);
690 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
694 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
697 bzero(&xig, sizeof(xig));
698 xig.xig_len = sizeof xig;
699 xig.xig_count = V_divcbinfo.ipi_count;
700 xig.xig_gen = V_divcbinfo.ipi_gencnt;
701 xig.xig_sogen = so_gencnt;
702 error = SYSCTL_OUT(req, &xig, sizeof xig);
706 while ((inp = inp_next(&inpi)) != NULL) {
707 if (inp->inp_gencnt <= xig.xig_gen) {
710 in_pcbtoxinpcb(inp, &xi);
711 error = SYSCTL_OUT(req, &xi, sizeof xi);
721 * Give the user an updated idea of our state.
722 * If the generation differs from what we told
723 * her before, she knows that something happened
724 * while we were processing this request, and it
725 * might be necessary to retry.
727 xig.xig_gen = V_divcbinfo.ipi_gencnt;
728 xig.xig_sogen = so_gencnt;
729 xig.xig_count = V_divcbinfo.ipi_count;
730 error = SYSCTL_OUT(req, &xig, sizeof xig);
737 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
738 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
740 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
741 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
742 NULL, 0, div_pcblist, "S,xinpcb",
743 "List of active divert sockets");
746 struct pr_usrreqs div_usrreqs = {
747 .pru_attach = div_attach,
748 .pru_bind = div_bind,
749 .pru_control = in_control,
750 .pru_detach = div_detach,
751 .pru_peeraddr = in_getpeeraddr,
752 .pru_send = div_send,
753 .pru_shutdown = div_shutdown,
754 .pru_sockaddr = in_getsockaddr,
755 .pru_sosetlabel = in_pcbsosetlabel
758 struct protosw div_protosw = {
760 .pr_protocol = IPPROTO_DIVERT,
761 .pr_flags = PR_ATOMIC|PR_ADDR,
762 .pr_input = div_input,
763 .pr_usrreqs = &div_usrreqs
767 div_modevent(module_t mod, int type, void *unused)
774 * Protocol will be initialized by pf_proto_register().
775 * We don't have to register ip_protox because we are not
776 * a true IP protocol that goes over the wire.
778 err = pf_proto_register(PF_INET, &div_protosw);
781 ip_divert_ptr = divert_packet;
785 * IPDIVERT may normally not be unloaded because of the
786 * potential race conditions. Tell kldunload we can't be
787 * unloaded unless the unload is forced.
795 * Module ipdivert can only be unloaded if no sockets are
796 * connected. Maybe this can be changed later to forcefully
797 * disconnect any open sockets.
799 * XXXRW: Note that there is a slight race here, as a new
800 * socket open request could be spinning on the lock and then
801 * we destroy the lock.
803 INP_INFO_WLOCK(&V_divcbinfo);
804 if (V_divcbinfo.ipi_count != 0) {
806 INP_INFO_WUNLOCK(&V_divcbinfo);
809 ip_divert_ptr = NULL;
810 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
811 INP_INFO_WUNLOCK(&V_divcbinfo);
823 static moduledata_t ipdivertmod = {
829 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
830 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
831 MODULE_VERSION(ipdivert, 1);