2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #if !defined(KLD_MODULE)
37 #error "IPDIVERT requires INET."
41 #include <sys/param.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/kernel.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
57 #include <net/netisr.h>
59 #include <netinet/in.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_var.h>
66 #include <netinet/sctp_crc32.h>
69 #include <security/mac/mac_framework.h>
76 * Allocate enough space to hold a full IP packet
78 #define DIVSNDQ (65536 + 100)
79 #define DIVRCVQ (65536 + 100)
82 * Divert sockets work in conjunction with ipfw or other packet filters,
83 * see the divert(4) manpage for features.
84 * Packets are selected by the packet filter and tagged with an
85 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
86 * the packet filter) and information on the matching filter rule for
87 * subsequent reinjection. The divert_port is used to put the packet
88 * on the corresponding divert socket, while the rule number is passed
89 * up (at least partially) as the sin_port in the struct sockaddr.
91 * Packets written to the divert socket carry in sin_addr a
92 * destination address, and in sin_port the number of the filter rule
93 * after which to continue processing.
94 * If the destination address is INADDR_ANY, the packet is treated as
95 * as outgoing and sent to ip_output(); otherwise it is treated as
96 * incoming and sent to ip_input().
97 * Further, sin_zero carries some information on the interface,
98 * which can be used in the reinject -- see comments in the code.
100 * On reinjection, processing in ip_input() and ip_output()
101 * will be exactly the same as for the original packet, except that
102 * packet filter processing will start at the rule number after the one
103 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
104 * will apply the entire ruleset to the packet).
107 /* Internal variables. */
108 static VNET_DEFINE(struct inpcbhead, divcb);
109 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
111 #define V_divcb VNET(divcb)
112 #define V_divcbinfo VNET(divcbinfo)
114 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
115 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
117 static eventhandler_tag ip_divert_event_tag;
120 * Initialize divert connection block queue.
123 div_zone_change(void *tag)
126 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
130 div_inpcb_init(void *mem, int size, int flags)
132 struct inpcb *inp = mem;
134 INP_LOCK_INIT(inp, "inp", "divinp");
139 div_inpcb_fini(void *mem, int size)
141 struct inpcb *inp = mem;
143 INP_LOCK_DESTROY(inp);
151 * XXX We don't use the hash list for divert IP, but it's easier to
152 * allocate one-entry hash lists than it is to check all over the
153 * place for hashbase == NULL.
155 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
156 div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE,
157 IPI_HASHFIELDS_NONE);
164 in_pcbinfo_destroy(&V_divcbinfo);
168 * IPPROTO_DIVERT is not in the real IP protocol number space; this
169 * function should never be called. Just in case, drop any packets.
172 div_input(struct mbuf *m, int off)
175 KMOD_IPSTAT_INC(ips_noproto);
180 * Divert a packet by passing it up to the divert socket at port 'port'.
182 * Setup generic address and protocol structures for div_input routine,
183 * then pass them along with mbuf chain.
186 divert_packet(struct mbuf *m, int incoming)
192 struct sockaddr_in divsrc;
195 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
201 if (m->m_len < sizeof(struct ip) &&
202 (m = m_pullup(m, sizeof(struct ip))) == 0)
204 ip = mtod(m, struct ip *);
206 /* Delayed checksums are currently not compatible with divert. */
207 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
208 ip->ip_len = ntohs(ip->ip_len);
210 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
211 ip->ip_len = htons(ip->ip_len);
214 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
215 ip->ip_len = ntohs(ip->ip_len);
216 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
217 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
218 ip->ip_len = htons(ip->ip_len);
221 bzero(&divsrc, sizeof(divsrc));
222 divsrc.sin_len = sizeof(divsrc);
223 divsrc.sin_family = AF_INET;
224 /* record matching rule, in host format */
225 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
227 * Record receive interface address, if any.
228 * But only for incoming packets.
237 /* Find IP address for receive interface */
238 ifp = m->m_pkthdr.rcvif;
240 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
241 if (ifa->ifa_addr->sa_family != AF_INET)
244 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
247 if_addr_runlock(ifp);
250 * Record the incoming interface name whenever we have one.
252 if (m->m_pkthdr.rcvif) {
254 * Hide the actual interface name in there in the
255 * sin_zero array. XXX This needs to be moved to a
256 * different sockaddr type for divert, e.g.
257 * sockaddr_div with multiple fields like
258 * sockaddr_dl. Presently we have only 7 bytes
259 * but that will do for now as most interfaces
260 * are 4 or less + 2 or less bytes for unit.
261 * There is probably a faster way of doing this,
262 * possibly taking it from the sockaddr_dl on the iface.
263 * This solves the problem of a P2P link and a LAN interface
264 * having the same address, which can result in the wrong
265 * interface being assigned to the packet when fed back
266 * into the divert socket. Theoretically if the daemon saves
267 * and re-uses the sockaddr_in as suggested in the man pages,
268 * this iface name will come along for the ride.
269 * (see div_output for the other half of this.)
271 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
272 sizeof(divsrc.sin_zero));
275 /* Put packet on socket queue, if any */
277 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
278 INP_INFO_RLOCK(&V_divcbinfo);
279 LIST_FOREACH(inp, &V_divcb, inp_list) {
280 /* XXX why does only one socket match? */
281 if (inp->inp_lport == nport) {
283 sa = inp->inp_socket;
284 SOCKBUF_LOCK(&sa->so_rcv);
285 if (sbappendaddr_locked(&sa->so_rcv,
286 (struct sockaddr *)&divsrc, m,
287 (struct mbuf *)0) == 0) {
288 SOCKBUF_UNLOCK(&sa->so_rcv);
289 sa = NULL; /* force mbuf reclaim below */
291 sorwakeup_locked(sa);
296 INP_INFO_RUNLOCK(&V_divcbinfo);
299 KMOD_IPSTAT_INC(ips_noproto);
300 KMOD_IPSTAT_DEC(ips_delivered);
305 * Deliver packet back into the IP processing machinery.
307 * If no address specified, or address is 0.0.0.0, send to ip_output();
308 * otherwise, send to ip_input() and mark as having been received on
309 * the interface with that address.
312 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
313 struct mbuf *control)
316 struct ipfw_rule_ref *dt;
318 struct mbuf *options;
321 * An mbuf may hasn't come from userland, but we pretend
324 m->m_pkthdr.rcvif = NULL;
326 M_SETFIB(m, so->so_fibnum);
329 m_freem(control); /* XXX */
331 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
333 /* this should be normal */
334 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
335 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
340 m_tag_prepend(m, mtag);
342 dt = (struct ipfw_rule_ref *)(mtag+1);
344 /* Loopback avoidance and state recovery */
348 /* set the starting point. We provide a non-zero slot,
349 * but a non_matching chain_id to skip that info and use
350 * the rulenum/rule_id.
352 dt->slot = 1; /* dummy, chain_id is invalid */
354 dt->rulenum = sin->sin_port+1; /* host format ? */
357 * Find receive interface with the given name, stuffed
358 * (if it exists) in the sin_zero[] field.
359 * The name is user supplied data so don't trust its size
360 * or that it is zero terminated.
362 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
364 if ( i > 0 && i < sizeof(sin->sin_zero))
365 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
368 /* Reinject packet into the system as incoming or outgoing */
369 if (!sin || sin->sin_addr.s_addr == 0) {
370 struct ip *const ip = mtod(m, struct ip *);
373 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
377 * Don't allow both user specified and setsockopt options,
378 * and don't allow packet length sizes that will crash
380 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
381 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
386 /* Convert fields to host order for ip_output() */
387 ip->ip_len = ntohs(ip->ip_len);
388 ip->ip_off = ntohs(ip->ip_off);
390 /* Send packet to output processing */
391 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
394 mac_inpcb_create_mbuf(inp, m);
397 * Get ready to inject the packet into ip_output().
398 * Just in case socket options were specified on the
399 * divert socket, we duplicate them. This is done
400 * to avoid having to hold the PCB locks over the call
401 * to ip_output(), as doing this results in a number of
402 * lock ordering complexities.
404 * Note that we set the multicast options argument for
405 * ip_output() to NULL since it should be invariant that
406 * they are not present.
408 KASSERT(inp->inp_moptions == NULL,
409 ("multicast options set on a divert socket"));
412 * XXXCSJP: It is unclear to me whether or not it makes
413 * sense for divert sockets to have options. However,
414 * for now we will duplicate them with the INP locks
415 * held so we can use them in ip_output() without
416 * requring a reference to the pcb.
418 if (inp->inp_options != NULL) {
419 options = m_dup(inp->inp_options, M_DONTWAIT);
424 if (error == ENOBUFS) {
428 error = ip_output(m, options, NULL,
429 ((so->so_options & SO_DONTROUTE) ?
430 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
431 IP_RAWOUTPUT, NULL, NULL);
436 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
437 if (m->m_pkthdr.rcvif == NULL) {
439 * No luck with the name, check by IP address.
440 * Clear the port and the ifname to make sure
441 * there are no distractions for ifa_ifwithaddr.
445 bzero(sin->sin_zero, sizeof(sin->sin_zero));
447 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
449 error = EADDRNOTAVAIL;
452 m->m_pkthdr.rcvif = ifa->ifa_ifp;
456 mac_socket_create_mbuf(so, m);
458 /* Send packet to input processing via netisr */
459 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
470 div_attach(struct socket *so, int proto, struct thread *td)
476 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
478 error = priv_check(td, PRIV_NETINET_DIVERT);
482 error = soreserve(so, div_sendspace, div_recvspace);
485 INP_INFO_WLOCK(&V_divcbinfo);
486 error = in_pcballoc(so, &V_divcbinfo);
488 INP_INFO_WUNLOCK(&V_divcbinfo);
491 inp = (struct inpcb *)so->so_pcb;
492 INP_INFO_WUNLOCK(&V_divcbinfo);
493 inp->inp_ip_p = proto;
494 inp->inp_vflag |= INP_IPV4;
495 inp->inp_flags |= INP_HDRINCL;
501 div_detach(struct socket *so)
506 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
507 INP_INFO_WLOCK(&V_divcbinfo);
511 INP_INFO_WUNLOCK(&V_divcbinfo);
515 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
521 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
522 /* in_pcbbind assumes that nam is a sockaddr_in
523 * and in_pcbbind requires a valid address. Since divert
524 * sockets don't we need to make sure the address is
525 * filled in properly.
526 * XXX -- divert should not be abusing in_pcbind
527 * and should probably have its own family.
529 if (nam->sa_family != AF_INET)
531 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
532 INP_INFO_WLOCK(&V_divcbinfo);
534 INP_HASH_WLOCK(&V_divcbinfo);
535 error = in_pcbbind(inp, nam, td->td_ucred);
536 INP_HASH_WUNLOCK(&V_divcbinfo);
538 INP_INFO_WUNLOCK(&V_divcbinfo);
543 div_shutdown(struct socket *so)
548 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
556 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
557 struct mbuf *control, struct thread *td)
560 /* Packet must have a header (but that's about it) */
561 if (m->m_len < sizeof (struct ip) &&
562 (m = m_pullup(m, sizeof (struct ip))) == 0) {
563 KMOD_IPSTAT_INC(ips_toosmall);
569 return div_output(so, m, (struct sockaddr_in *)nam, control);
573 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
575 struct in_addr faddr;
577 faddr = ((struct sockaddr_in *)sa)->sin_addr;
578 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
580 if (PRC_IS_REDIRECT(cmd))
585 div_pcblist(SYSCTL_HANDLER_ARGS)
588 struct inpcb *inp, **inp_list;
593 * The process of preparing the TCB list is too time-consuming and
594 * resource-intensive to repeat twice on every request.
596 if (req->oldptr == 0) {
597 n = V_divcbinfo.ipi_count;
598 n += imax(n / 8, 10);
599 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
603 if (req->newptr != 0)
607 * OK, now we're committed to doing something.
609 INP_INFO_RLOCK(&V_divcbinfo);
610 gencnt = V_divcbinfo.ipi_gencnt;
611 n = V_divcbinfo.ipi_count;
612 INP_INFO_RUNLOCK(&V_divcbinfo);
614 error = sysctl_wire_old_buffer(req,
615 2 * sizeof(xig) + n*sizeof(struct xinpcb));
619 xig.xig_len = sizeof xig;
621 xig.xig_gen = gencnt;
622 xig.xig_sogen = so_gencnt;
623 error = SYSCTL_OUT(req, &xig, sizeof xig);
627 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
631 INP_INFO_RLOCK(&V_divcbinfo);
632 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
633 inp = LIST_NEXT(inp, inp_list)) {
635 if (inp->inp_gencnt <= gencnt &&
636 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
642 INP_INFO_RUNLOCK(&V_divcbinfo);
646 for (i = 0; i < n; i++) {
649 if (inp->inp_gencnt <= gencnt) {
651 bzero(&xi, sizeof(xi));
652 xi.xi_len = sizeof xi;
653 /* XXX should avoid extra copy */
654 bcopy(inp, &xi.xi_inp, sizeof *inp);
656 sotoxsocket(inp->inp_socket, &xi.xi_socket);
658 error = SYSCTL_OUT(req, &xi, sizeof xi);
662 INP_INFO_WLOCK(&V_divcbinfo);
663 for (i = 0; i < n; i++) {
666 if (!in_pcbrele_rlocked(inp))
669 INP_INFO_WUNLOCK(&V_divcbinfo);
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 INP_INFO_RLOCK(&V_divcbinfo);
680 xig.xig_gen = V_divcbinfo.ipi_gencnt;
681 xig.xig_sogen = so_gencnt;
682 xig.xig_count = V_divcbinfo.ipi_count;
683 INP_INFO_RUNLOCK(&V_divcbinfo);
684 error = SYSCTL_OUT(req, &xig, sizeof xig);
686 free(inp_list, M_TEMP);
691 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
692 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
693 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
696 struct pr_usrreqs div_usrreqs = {
697 .pru_attach = div_attach,
698 .pru_bind = div_bind,
699 .pru_control = in_control,
700 .pru_detach = div_detach,
701 .pru_peeraddr = in_getpeeraddr,
702 .pru_send = div_send,
703 .pru_shutdown = div_shutdown,
704 .pru_sockaddr = in_getsockaddr,
705 .pru_sosetlabel = in_pcbsosetlabel
708 struct protosw div_protosw = {
710 .pr_protocol = IPPROTO_DIVERT,
711 .pr_flags = PR_ATOMIC|PR_ADDR,
712 .pr_input = div_input,
713 .pr_ctlinput = div_ctlinput,
714 .pr_ctloutput = ip_ctloutput,
717 .pr_destroy = div_destroy,
719 .pr_usrreqs = &div_usrreqs
723 div_modevent(module_t mod, int type, void *unused)
733 * Protocol will be initialized by pf_proto_register().
734 * We don't have to register ip_protox because we are not
735 * a true IP protocol that goes over the wire.
737 err = pf_proto_register(PF_INET, &div_protosw);
740 ip_divert_ptr = divert_packet;
741 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
742 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
746 * IPDIVERT may normally not be unloaded because of the
747 * potential race conditions. Tell kldunload we can't be
748 * unloaded unless the unload is forced.
760 * Module ipdivert can only be unloaded if no sockets are
761 * connected. Maybe this can be changed later to forcefully
762 * disconnect any open sockets.
764 * XXXRW: Note that there is a slight race here, as a new
765 * socket open request could be spinning on the lock and then
766 * we destroy the lock.
768 INP_INFO_WLOCK(&V_divcbinfo);
769 n = V_divcbinfo.ipi_count;
772 INP_INFO_WUNLOCK(&V_divcbinfo);
775 ip_divert_ptr = NULL;
776 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
777 INP_INFO_WUNLOCK(&V_divcbinfo);
779 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
789 static moduledata_t ipdivertmod = {
795 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
796 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
797 MODULE_VERSION(ipdivert, 1);