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 #include <sys/param.h>
40 #include <sys/eventhandler.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/kernel.h>
49 #include <sys/domain.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
57 #include <net/if_var.h>
58 #include <net/if_private.h>
59 #include <net/netisr.h>
61 #include <netinet/in.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_divert.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 * Usually a system has very few divert ports. Previous implementation
91 #define DIVHASHSIZE (1 << 3) /* 8 entries, one cache line. */
92 #define DIVHASH(port) (port % DIVHASHSIZE)
93 #define DCBHASH(dcb) ((dcb)->dcb_port % DIVHASHSIZE)
96 * Divert sockets work in conjunction with ipfw or other packet filters,
97 * see the divert(4) manpage for features.
98 * Packets are selected by the packet filter and tagged with an
99 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
100 * the packet filter) and information on the matching filter rule for
101 * subsequent reinjection. The divert_port is used to put the packet
102 * on the corresponding divert socket, while the rule number is passed
103 * up (at least partially) as the sin_port in the struct sockaddr.
105 * Packets written to the divert socket carry in sin_addr a
106 * destination address, and in sin_port the number of the filter rule
107 * after which to continue processing.
108 * If the destination address is INADDR_ANY, the packet is treated as
109 * as outgoing and sent to ip_output(); otherwise it is treated as
110 * incoming and sent to ip_input().
111 * Further, sin_zero carries some information on the interface,
112 * which can be used in the reinject -- see comments in the code.
114 * On reinjection, processing in ip_input() and ip_output()
115 * will be exactly the same as for the original packet, except that
116 * packet filter processing will start at the rule number after the one
117 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
118 * will apply the entire ruleset to the packet).
120 static SYSCTL_NODE(_net_inet, OID_AUTO, divert, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
123 VNET_PCPUSTAT_DEFINE_STATIC(struct divstat, divstat);
124 VNET_PCPUSTAT_SYSINIT(divstat);
126 VNET_PCPUSTAT_SYSUNINIT(divstat);
128 SYSCTL_VNET_PCPUSTAT(_net_inet_divert, OID_AUTO, stats, struct divstat,
129 divstat, "divert(4) socket statistics");
130 #define DIVSTAT_INC(name) \
131 VNET_PCPUSTAT_ADD(struct divstat, divstat, div_ ## name, 1)
133 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
134 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
136 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
137 struct sockaddr_in *sin);
138 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
142 SLIST_ENTRY(divcb) dcb_next;
144 #define DCB_UNBOUND ((intptr_t)-1)
146 struct socket *dcb_socket;
149 struct epoch_context dcb_epochctx;
152 SLIST_HEAD(divhashhead, divcb);
154 VNET_DEFINE_STATIC(struct divhashhead, divhash[DIVHASHSIZE]) = {};
155 #define V_divhash VNET(divhash)
156 VNET_DEFINE_STATIC(uint64_t, dcb_count) = 0;
157 #define V_dcb_count VNET(dcb_count)
158 VNET_DEFINE_STATIC(uint64_t, dcb_gencnt) = 0;
159 #define V_dcb_gencnt VNET(dcb_gencnt)
161 static struct mtx divert_mtx;
162 MTX_SYSINIT(divert, &divert_mtx, "divert(4) socket pcb lists", MTX_DEF);
163 #define DIVERT_LOCK() mtx_lock(&divert_mtx)
164 #define DIVERT_UNLOCK() mtx_unlock(&divert_mtx)
167 * Divert a packet by passing it up to the divert socket at port 'port'.
170 divert_packet(struct mbuf *m, bool incoming)
174 struct sockaddr_in divsrc;
179 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
185 if (m->m_len < sizeof(struct ip) &&
186 (m = m_pullup(m, sizeof(struct ip))) == NULL)
189 /* Delayed checksums are currently not compatible with divert. */
190 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
192 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
194 #if defined(SCTP) || defined(SCTP_SUPPORT)
195 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
198 ip = mtod(m, struct ip *);
199 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
200 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
205 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
206 in6_delayed_cksum(m, m->m_pkthdr.len -
207 sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
208 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
210 #if defined(SCTP) || defined(SCTP_SUPPORT)
211 if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
212 sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
213 m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
217 bzero(&divsrc, sizeof(divsrc));
218 divsrc.sin_len = sizeof(divsrc);
219 divsrc.sin_family = AF_INET;
220 /* record matching rule, in host format */
221 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
223 * Record receive interface address, if any.
224 * But only for incoming packets.
233 /* Find IP address for receive interface */
234 ifp = m->m_pkthdr.rcvif;
235 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
236 if (ifa->ifa_addr->sa_family != AF_INET)
239 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
244 * Record the incoming interface name whenever we have one.
246 if (m->m_pkthdr.rcvif) {
248 * Hide the actual interface name in there in the
249 * sin_zero array. XXX This needs to be moved to a
250 * different sockaddr type for divert, e.g.
251 * sockaddr_div with multiple fields like
252 * sockaddr_dl. Presently we have only 7 bytes
253 * but that will do for now as most interfaces
254 * are 4 or less + 2 or less bytes for unit.
255 * There is probably a faster way of doing this,
256 * possibly taking it from the sockaddr_dl on the iface.
257 * This solves the problem of a P2P link and a LAN interface
258 * having the same address, which can result in the wrong
259 * interface being assigned to the packet when fed back
260 * into the divert socket. Theoretically if the daemon saves
261 * and re-uses the sockaddr_in as suggested in the man pages,
262 * this iface name will come along for the ride.
263 * (see div_output for the other half of this.)
265 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
266 sizeof(divsrc.sin_zero));
269 /* Put packet on socket queue, if any */
270 nport = htons((uint16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
271 SLIST_FOREACH(dcb, &V_divhash[DIVHASH(nport)], dcb_next)
272 if (dcb->dcb_port == nport)
276 struct socket *sa = dcb->dcb_socket;
278 SOCKBUF_LOCK(&sa->so_rcv);
279 if (sbappendaddr_locked(&sa->so_rcv,
280 (struct sockaddr *)&divsrc, m, NULL) == 0) {
281 soroverflow_locked(sa);
284 sorwakeup_locked(sa);
285 DIVSTAT_INC(diverted);
294 * Deliver packet back into the IP processing machinery.
296 * If no address specified, or address is 0.0.0.0, send to ip_output();
297 * otherwise, send to ip_input() and mark as having been received on
298 * the interface with that address.
301 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
302 struct mbuf *control, struct thread *td)
304 struct epoch_tracker et;
305 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
308 struct ipfw_rule_ref *dt;
314 /* Packet must have a header (but that's about it) */
315 if (m->m_len < sizeof (struct ip) &&
316 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
322 if (sin->sin_family != AF_INET) {
324 return (EAFNOSUPPORT);
326 if (sin->sin_len != sizeof(*sin)) {
333 * An mbuf may hasn't come from userland, but we pretend
336 m->m_pkthdr.rcvif = NULL;
338 M_SETFIB(m, so->so_fibnum);
340 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
342 /* this should be normal */
343 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
344 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
349 m_tag_prepend(m, mtag);
351 dt = (struct ipfw_rule_ref *)(mtag+1);
353 /* Loopback avoidance and state recovery */
357 /* set the starting point. We provide a non-zero slot,
358 * but a non_matching chain_id to skip that info and use
359 * the rulenum/rule_id.
361 dt->slot = 1; /* dummy, chain_id is invalid */
363 dt->rulenum = sin->sin_port+1; /* host format ? */
365 /* XXX: broken for IPv6 */
367 * Find receive interface with the given name, stuffed
368 * (if it exists) in the sin_zero[] field.
369 * The name is user supplied data so don't trust its size
370 * or that it is zero terminated.
372 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
374 if ( i > 0 && i < sizeof(sin->sin_zero))
375 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
378 ip = mtod(m, struct ip *);
386 case IPV6_VERSION >> 4:
392 return (EAFNOSUPPORT);
395 /* Reinject packet into the system as incoming or outgoing */
397 if (!sin || sin->sin_addr.s_addr == 0) {
398 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
399 error = div_output_outbound(family, so, m);
401 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
402 error = div_output_inbound(family, so, m, sin);
410 * Sends mbuf @m to the wire via ip[6]_output().
412 * Returns 0 on success or an errno value on failure. @m is always consumed.
415 div_output_outbound(int family, struct socket *so, struct mbuf *m)
423 struct ip *const ip = mtod(m, struct ip *);
425 /* Don't allow packet length sizes that will crash. */
426 if (((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
436 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
438 /* Don't allow packet length sizes that will crash */
439 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
449 mac_socket_create_mbuf(so, m);
456 error = ip_output(m, NULL, NULL,
457 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
458 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
463 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
468 DIVSTAT_INC(outbound);
474 * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
476 * Returns 0 on success or an errno value on failure. @m is always consumed.
479 div_output_inbound(int family, struct socket *so, struct mbuf *m,
480 struct sockaddr_in *sin)
484 if (m->m_pkthdr.rcvif == NULL) {
486 * No luck with the name, check by IP address.
487 * Clear the port and the ifname to make sure
488 * there are no distractions for ifa_ifwithaddr.
491 /* XXX: broken for IPv6 */
492 bzero(sin->sin_zero, sizeof(sin->sin_zero));
494 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
497 return (EADDRNOTAVAIL);
499 m->m_pkthdr.rcvif = ifa->ifa_ifp;
502 mac_socket_create_mbuf(so, m);
504 /* Send packet to input processing via netisr */
511 ip = mtod(m, struct ip *);
513 * Restore M_BCAST flag when destination address is
514 * broadcast. It is expected by ip_tryforward().
516 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
517 m->m_flags |= M_MCAST;
518 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
519 m->m_flags |= M_BCAST;
520 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
521 DIVSTAT_INC(inbound);
527 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
528 DIVSTAT_INC(inbound);
540 div_attach(struct socket *so, int proto, struct thread *td)
546 error = priv_check(td, PRIV_NETINET_DIVERT);
550 error = soreserve(so, div_sendspace, div_recvspace);
553 dcb = malloc(sizeof(*dcb), M_PCB, M_WAITOK);
554 dcb->dcb_bound = DCB_UNBOUND;
555 dcb->dcb_socket = so;
558 dcb->dcb_gencnt = ++V_dcb_gencnt;
566 div_free(epoch_context_t ctx)
568 struct divcb *dcb = __containerof(ctx, struct divcb, dcb_epochctx);
574 div_detach(struct socket *so)
576 struct divcb *dcb = so->so_pcb;
580 if (dcb->dcb_bound != DCB_UNBOUND)
581 SLIST_REMOVE(&V_divhash[DCBHASH(dcb)], dcb, divcb, dcb_next);
585 NET_EPOCH_CALL(div_free, &dcb->dcb_epochctx);
589 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
594 if (nam->sa_family != AF_INET)
596 if (nam->sa_len != sizeof(struct sockaddr_in))
598 port = ((struct sockaddr_in *)nam)->sin_port;
600 SLIST_FOREACH(dcb, &V_divhash[DIVHASH(port)], dcb_next)
601 if (dcb->dcb_port == port) {
606 if (dcb->dcb_bound != DCB_UNBOUND)
607 SLIST_REMOVE(&V_divhash[DCBHASH(dcb)], dcb, divcb, dcb_next);
608 dcb->dcb_port = port;
609 SLIST_INSERT_HEAD(&V_divhash[DIVHASH(port)], dcb, dcb_next);
616 div_shutdown(struct socket *so)
624 div_pcblist(SYSCTL_HANDLER_ARGS)
630 if (req->newptr != 0)
633 if (req->oldptr == 0) {
637 n += imax(n / 8, 10);
638 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
642 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
645 bzero(&xig, sizeof(xig));
646 xig.xig_len = sizeof xig;
647 xig.xig_count = V_dcb_count;
648 xig.xig_gen = V_dcb_gencnt;
649 xig.xig_sogen = so_gencnt;
650 error = SYSCTL_OUT(req, &xig, sizeof xig);
655 for (int i = 0; i < DIVHASHSIZE; i++)
656 SLIST_FOREACH(dcb, &V_divhash[i], dcb_next) {
657 if (dcb->dcb_gencnt <= xig.xig_gen) {
660 bzero(&xi, sizeof(xi));
661 xi.xi_len = sizeof(struct xinpcb);
662 sotoxsocket(dcb->dcb_socket, &xi.xi_socket);
663 xi.inp_gencnt = dcb->dcb_gencnt;
664 xi.inp_vflag = INP_IPV4; /* XXX: netstat(1) */
665 xi.inp_inc.inc_ie.ie_lport = dcb->dcb_port;
666 error = SYSCTL_OUT(req, &xi, sizeof xi);
673 * Give the user an updated idea of our state.
674 * If the generation differs from what we told
675 * her before, she knows that something happened
676 * while we were processing this request, and it
677 * might be necessary to retry.
679 xig.xig_gen = V_dcb_gencnt;
680 xig.xig_sogen = so_gencnt;
681 xig.xig_count = V_dcb_count;
682 error = SYSCTL_OUT(req, &xig, sizeof xig);
689 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
690 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, div_pcblist,
691 "S,xinpcb", "List of active divert sockets");
693 static struct protosw div_protosw = {
695 .pr_flags = PR_ATOMIC|PR_ADDR,
696 .pr_attach = div_attach,
698 .pr_detach = div_detach,
700 .pr_shutdown = div_shutdown,
703 static struct domain divertdomain = {
704 .dom_family = PF_DIVERT,
705 .dom_name = "divert",
707 .dom_protosw = { &div_protosw },
711 div_modevent(module_t mod, int type, void *unused)
717 domain_add(&divertdomain);
718 ip_divert_ptr = divert_packet;
722 * IPDIVERT may normally not be unloaded because of the
723 * potential race conditions. Tell kldunload we can't be
724 * unloaded unless the unload is forced.
732 * Module ipdivert can only be unloaded if no sockets are
733 * connected. Maybe this can be changed later to forcefully
734 * disconnect any open sockets.
736 * XXXRW: Note that there is a slight race here, as a new
737 * socket open request could be spinning on the lock and then
738 * we destroy the lock.
740 * XXXGL: One more reason this code is incorrect is that it
741 * checks only the current vnet.
744 if (V_dcb_count != 0) {
750 ip_divert_ptr = NULL;
751 domain_remove(&divertdomain);
760 static moduledata_t ipdivertmod = {
766 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
767 MODULE_VERSION(ipdivert, 1);