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 inpcbhead, divcb);
115 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
117 #define V_divcb VNET(divcb)
118 #define V_divcbinfo VNET(divcbinfo)
120 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
121 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
123 static eventhandler_tag ip_divert_event_tag;
126 * Initialize divert connection block queue.
129 div_zone_change(void *tag)
132 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
136 div_inpcb_init(void *mem, int size, int flags)
138 struct inpcb *inp = mem;
140 INP_LOCK_INIT(inp, "inp", "divinp");
149 * XXX We don't use the hash list for divert IP, but it's easier to
150 * allocate one-entry hash lists than it is to check all over the
151 * place for hashbase == NULL.
153 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
154 div_inpcb_init, IPI_HASHFIELDS_NONE);
158 div_destroy(void *unused __unused)
161 in_pcbinfo_destroy(&V_divcbinfo);
163 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
167 * IPPROTO_DIVERT is not in the real IP protocol number space; this
168 * function should never be called. Just in case, drop any packets.
171 div_input(struct mbuf **mp, int *offp, int proto)
173 struct mbuf *m = *mp;
175 KMOD_IPSTAT_INC(ips_noproto);
177 return (IPPROTO_DONE);
181 * Divert a packet by passing it up to the divert socket at port 'port'.
183 * Setup generic address and protocol structures for div_input routine,
184 * then pass them along with mbuf chain.
187 divert_packet(struct mbuf *m, int incoming)
193 struct sockaddr_in divsrc;
195 struct epoch_tracker et;
197 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
203 if (m->m_len < sizeof(struct ip) &&
204 (m = m_pullup(m, sizeof(struct ip))) == NULL)
206 ip = mtod(m, struct ip *);
208 /* Delayed checksums are currently not compatible with divert. */
209 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
211 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
213 #if defined(SCTP) || defined(SCTP_SUPPORT)
214 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
215 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
216 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
219 bzero(&divsrc, sizeof(divsrc));
220 divsrc.sin_len = sizeof(divsrc);
221 divsrc.sin_family = AF_INET;
222 /* record matching rule, in host format */
223 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
225 * Record receive interface address, if any.
226 * But only for incoming packets.
235 /* Find IP address for receive interface */
236 ifp = m->m_pkthdr.rcvif;
238 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
239 if (ifa->ifa_addr->sa_family != AF_INET)
242 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
245 if_addr_runlock(ifp);
248 * Record the incoming interface name whenever we have one.
250 if (m->m_pkthdr.rcvif) {
252 * Hide the actual interface name in there in the
253 * sin_zero array. XXX This needs to be moved to a
254 * different sockaddr type for divert, e.g.
255 * sockaddr_div with multiple fields like
256 * sockaddr_dl. Presently we have only 7 bytes
257 * but that will do for now as most interfaces
258 * are 4 or less + 2 or less bytes for unit.
259 * There is probably a faster way of doing this,
260 * possibly taking it from the sockaddr_dl on the iface.
261 * This solves the problem of a P2P link and a LAN interface
262 * having the same address, which can result in the wrong
263 * interface being assigned to the packet when fed back
264 * into the divert socket. Theoretically if the daemon saves
265 * and re-uses the sockaddr_in as suggested in the man pages,
266 * this iface name will come along for the ride.
267 * (see div_output for the other half of this.)
269 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
270 sizeof(divsrc.sin_zero));
273 /* Put packet on socket queue, if any */
275 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
276 INP_INFO_RLOCK_ET(&V_divcbinfo, et);
277 CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
278 /* XXX why does only one socket match? */
279 if (inp->inp_lport == nport) {
281 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
285 sa = inp->inp_socket;
286 SOCKBUF_LOCK(&sa->so_rcv);
287 if (sbappendaddr_locked(&sa->so_rcv,
288 (struct sockaddr *)&divsrc, m,
289 (struct mbuf *)0) == 0) {
290 soroverflow_locked(sa);
291 sa = NULL; /* force mbuf reclaim below */
293 sorwakeup_locked(sa);
298 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
301 KMOD_IPSTAT_INC(ips_noproto);
302 KMOD_IPSTAT_DEC(ips_delivered);
307 * Deliver packet back into the IP processing machinery.
309 * If no address specified, or address is 0.0.0.0, send to ip_output();
310 * otherwise, send to ip_input() and mark as having been received on
311 * the interface with that address.
314 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
315 struct mbuf *control)
317 struct ip *const ip = mtod(m, struct ip *);
319 struct ipfw_rule_ref *dt;
323 * An mbuf may hasn't come from userland, but we pretend
326 m->m_pkthdr.rcvif = NULL;
328 M_SETFIB(m, so->so_fibnum);
331 m_freem(control); /* XXX */
333 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
335 /* this should be normal */
336 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
337 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
342 m_tag_prepend(m, mtag);
344 dt = (struct ipfw_rule_ref *)(mtag+1);
346 /* Loopback avoidance and state recovery */
350 /* set the starting point. We provide a non-zero slot,
351 * but a non_matching chain_id to skip that info and use
352 * the rulenum/rule_id.
354 dt->slot = 1; /* dummy, chain_id is invalid */
356 dt->rulenum = sin->sin_port+1; /* host format ? */
359 * Find receive interface with the given name, stuffed
360 * (if it exists) in the sin_zero[] field.
361 * The name is user supplied data so don't trust its size
362 * or that it is zero terminated.
364 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
366 if ( i > 0 && i < sizeof(sin->sin_zero))
367 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
370 /* Reinject packet into the system as incoming or outgoing */
371 if (!sin || sin->sin_addr.s_addr == 0) {
372 struct mbuf *options = NULL;
375 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
381 * Don't allow both user specified and setsockopt
382 * options, and don't allow packet length sizes that
385 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
386 inp->inp_options != NULL) ||
387 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
394 case IPV6_VERSION >> 4:
396 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
398 /* Don't allow packet length sizes that will crash */
399 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
413 /* Send packet to output processing */
414 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
417 mac_inpcb_create_mbuf(inp, m);
420 * Get ready to inject the packet into ip_output().
421 * Just in case socket options were specified on the
422 * divert socket, we duplicate them. This is done
423 * to avoid having to hold the PCB locks over the call
424 * to ip_output(), as doing this results in a number of
425 * lock ordering complexities.
427 * Note that we set the multicast options argument for
428 * ip_output() to NULL since it should be invariant that
429 * they are not present.
431 KASSERT(inp->inp_moptions == NULL,
432 ("multicast options set on a divert socket"));
434 * XXXCSJP: It is unclear to me whether or not it makes
435 * sense for divert sockets to have options. However,
436 * for now we will duplicate them with the INP locks
437 * held so we can use them in ip_output() without
438 * requring a reference to the pcb.
440 if (inp->inp_options != NULL) {
441 options = m_dup(inp->inp_options, M_NOWAIT);
442 if (options == NULL) {
452 error = ip_output(m, options, NULL,
453 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
454 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
457 case IPV6_VERSION >> 4:
458 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
465 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
466 if (m->m_pkthdr.rcvif == NULL) {
468 * No luck with the name, check by IP address.
469 * Clear the port and the ifname to make sure
470 * there are no distractions for ifa_ifwithaddr.
474 bzero(sin->sin_zero, sizeof(sin->sin_zero));
477 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
479 error = EADDRNOTAVAIL;
483 m->m_pkthdr.rcvif = ifa->ifa_ifp;
487 mac_socket_create_mbuf(so, m);
489 /* Send packet to input processing via netisr */
493 * Restore M_BCAST flag when destination address is
494 * broadcast. It is expected by ip_tryforward().
496 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
497 m->m_flags |= M_MCAST;
498 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
499 m->m_flags |= M_BCAST;
500 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
503 case IPV6_VERSION >> 4:
504 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
521 div_attach(struct socket *so, int proto, struct thread *td)
527 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
529 error = priv_check(td, PRIV_NETINET_DIVERT);
533 error = soreserve(so, div_sendspace, div_recvspace);
536 INP_INFO_WLOCK(&V_divcbinfo);
537 error = in_pcballoc(so, &V_divcbinfo);
539 INP_INFO_WUNLOCK(&V_divcbinfo);
542 inp = (struct inpcb *)so->so_pcb;
543 INP_INFO_WUNLOCK(&V_divcbinfo);
544 inp->inp_ip_p = proto;
545 inp->inp_vflag |= INP_IPV4;
546 inp->inp_flags |= INP_HDRINCL;
552 div_detach(struct socket *so)
557 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
558 INP_INFO_WLOCK(&V_divcbinfo);
562 INP_INFO_WUNLOCK(&V_divcbinfo);
566 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
572 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
573 /* in_pcbbind assumes that nam is a sockaddr_in
574 * and in_pcbbind requires a valid address. Since divert
575 * sockets don't we need to make sure the address is
576 * filled in properly.
577 * XXX -- divert should not be abusing in_pcbind
578 * and should probably have its own family.
580 if (nam->sa_family != AF_INET)
582 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
583 INP_INFO_WLOCK(&V_divcbinfo);
585 INP_HASH_WLOCK(&V_divcbinfo);
586 error = in_pcbbind(inp, nam, td->td_ucred);
587 INP_HASH_WUNLOCK(&V_divcbinfo);
589 INP_INFO_WUNLOCK(&V_divcbinfo);
594 div_shutdown(struct socket *so)
599 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
607 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
608 struct mbuf *control, struct thread *td)
611 /* Packet must have a header (but that's about it) */
612 if (m->m_len < sizeof (struct ip) &&
613 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
614 KMOD_IPSTAT_INC(ips_toosmall);
620 return div_output(so, m, (struct sockaddr_in *)nam, control);
624 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
626 struct in_addr faddr;
628 faddr = ((struct sockaddr_in *)sa)->sin_addr;
629 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
631 if (PRC_IS_REDIRECT(cmd))
636 div_pcblist(SYSCTL_HANDLER_ARGS)
639 struct inpcb *inp, **inp_list;
642 struct epoch_tracker et;
645 * The process of preparing the TCB list is too time-consuming and
646 * resource-intensive to repeat twice on every request.
648 if (req->oldptr == 0) {
649 n = V_divcbinfo.ipi_count;
650 n += imax(n / 8, 10);
651 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
655 if (req->newptr != 0)
659 * OK, now we're committed to doing something.
661 INP_INFO_WLOCK(&V_divcbinfo);
662 gencnt = V_divcbinfo.ipi_gencnt;
663 n = V_divcbinfo.ipi_count;
664 INP_INFO_WUNLOCK(&V_divcbinfo);
666 error = sysctl_wire_old_buffer(req,
667 2 * sizeof(xig) + n*sizeof(struct xinpcb));
671 bzero(&xig, sizeof(xig));
672 xig.xig_len = sizeof xig;
674 xig.xig_gen = gencnt;
675 xig.xig_sogen = so_gencnt;
676 error = SYSCTL_OUT(req, &xig, sizeof xig);
680 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
681 if (inp_list == NULL)
684 INP_INFO_RLOCK_ET(&V_divcbinfo, et);
685 for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
686 inp = CK_LIST_NEXT(inp, inp_list)) {
688 if (inp->inp_gencnt <= gencnt &&
689 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
695 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
699 for (i = 0; i < n; i++) {
702 if (inp->inp_gencnt <= gencnt) {
705 in_pcbtoxinpcb(inp, &xi);
707 error = SYSCTL_OUT(req, &xi, sizeof xi);
711 INP_INFO_WLOCK(&V_divcbinfo);
712 for (i = 0; i < n; i++) {
715 if (!in_pcbrele_rlocked(inp))
718 INP_INFO_WUNLOCK(&V_divcbinfo);
721 struct epoch_tracker et;
723 * Give the user an updated idea of our state.
724 * If the generation differs from what we told
725 * her before, she knows that something happened
726 * while we were processing this request, and it
727 * might be necessary to retry.
729 INP_INFO_RLOCK_ET(&V_divcbinfo, et);
730 xig.xig_gen = V_divcbinfo.ipi_gencnt;
731 xig.xig_sogen = so_gencnt;
732 xig.xig_count = V_divcbinfo.ipi_count;
733 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et);
734 error = SYSCTL_OUT(req, &xig, sizeof xig);
736 free(inp_list, M_TEMP);
741 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
743 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
744 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
747 struct pr_usrreqs div_usrreqs = {
748 .pru_attach = div_attach,
749 .pru_bind = div_bind,
750 .pru_control = in_control,
751 .pru_detach = div_detach,
752 .pru_peeraddr = in_getpeeraddr,
753 .pru_send = div_send,
754 .pru_shutdown = div_shutdown,
755 .pru_sockaddr = in_getsockaddr,
756 .pru_sosetlabel = in_pcbsosetlabel
759 struct protosw div_protosw = {
761 .pr_protocol = IPPROTO_DIVERT,
762 .pr_flags = PR_ATOMIC|PR_ADDR,
763 .pr_input = div_input,
764 .pr_ctlinput = div_ctlinput,
765 .pr_ctloutput = ip_ctloutput,
767 .pr_usrreqs = &div_usrreqs
771 div_modevent(module_t mod, int type, void *unused)
778 * Protocol will be initialized by pf_proto_register().
779 * We don't have to register ip_protox because we are not
780 * a true IP protocol that goes over the wire.
782 err = pf_proto_register(PF_INET, &div_protosw);
785 ip_divert_ptr = divert_packet;
786 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
787 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
791 * IPDIVERT may normally not be unloaded because of the
792 * potential race conditions. Tell kldunload we can't be
793 * unloaded unless the unload is forced.
801 * Module ipdivert can only be unloaded if no sockets are
802 * connected. Maybe this can be changed later to forcefully
803 * disconnect any open sockets.
805 * XXXRW: Note that there is a slight race here, as a new
806 * socket open request could be spinning on the lock and then
807 * we destroy the lock.
809 INP_INFO_WLOCK(&V_divcbinfo);
810 if (V_divcbinfo.ipi_count != 0) {
812 INP_INFO_WUNLOCK(&V_divcbinfo);
815 ip_divert_ptr = NULL;
816 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
817 INP_INFO_WUNLOCK(&V_divcbinfo);
821 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
830 static moduledata_t ipdivertmod = {
836 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
837 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
838 MODULE_VERSION(ipdivert, 1);