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)
38 #error "IPDIVERT requires INET."
41 #error "IPDIVERT requires IPFIREWALL"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/kernel.h>
54 #include <sys/protosw.h>
55 #include <sys/rwlock.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
60 #include <sys/sysctl.h>
61 #include <sys/systm.h>
66 #include <net/netisr.h>
67 #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>
75 #include <netinet/ip_divert.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_fw.h>
79 #include <netinet/sctp_crc32.h>
82 #include <security/mac/mac_framework.h>
89 * Allocate enough space to hold a full IP packet
91 #define DIVSNDQ (65536 + 100)
92 #define DIVRCVQ (65536 + 100)
95 * Divert sockets work in conjunction with ipfw, see the divert(4)
96 * manpage for features.
97 * Internally, packets selected by ipfw in ip_input() or ip_output(),
98 * and never diverted before, are passed to the input queue of the
99 * divert socket with a given 'divert_port' number (as specified in
100 * the matching ipfw rule), and they are tagged with a 16 bit cookie
101 * (representing the rule number of the matching ipfw rule), which
102 * is passed to process reading from the socket.
104 * Packets written to the divert socket are again tagged with a cookie
105 * (usually the same as above) and a destination address.
106 * If the destination address is INADDR_ANY then the packet is
107 * treated as outgoing and sent to ip_output(), otherwise it is
108 * treated as incoming and sent to ip_input().
109 * In both cases, the packet is tagged with the cookie.
111 * On reinjection, processing in ip_input() and ip_output()
112 * will be exactly the same as for the original packet, except that
113 * ipfw processing will start at the rule number after the one
114 * written in the cookie (so, tagging a packet with a cookie of 0
115 * will cause it to be effectively considered as a standard packet).
118 /* Internal variables. */
119 static VNET_DEFINE(struct inpcbhead, divcb);
120 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
122 #define V_divcb VNET(divcb)
123 #define V_divcbinfo VNET(divcbinfo)
125 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
126 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
129 * Initialize divert connection block queue.
132 div_zone_change(void *tag)
135 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
139 div_inpcb_init(void *mem, int size, int flags)
141 struct inpcb *inp = mem;
143 INP_LOCK_INIT(inp, "inp", "divinp");
148 div_inpcb_fini(void *mem, int size)
150 struct inpcb *inp = mem;
152 INP_LOCK_DESTROY(inp);
159 INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
161 V_divcbinfo.ipi_listhead = &V_divcb;
163 V_divcbinfo.ipi_vnet = curvnet;
166 * XXX We don't use the hash list for divert IP, but it's easier
167 * to allocate a one entry hash list than it is to check all
168 * over the place for hashbase == NULL.
170 V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
171 V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
172 &V_divcbinfo.ipi_porthashmask);
173 V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
174 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
176 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
177 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
178 NULL, EVENTHANDLER_PRI_ANY);
182 * IPPROTO_DIVERT is not in the real IP protocol number space; this
183 * function should never be called. Just in case, drop any packets.
186 div_input(struct mbuf *m, int off)
189 KMOD_IPSTAT_INC(ips_noproto);
194 * Divert a packet by passing it up to the divert socket at port 'port'.
196 * Setup generic address and protocol structures for div_input routine,
197 * then pass them along with mbuf chain.
200 divert_packet(struct mbuf *m, int incoming)
206 struct sockaddr_in divsrc;
209 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
211 printf("%s: no divert tag\n", __func__);
216 if (m->m_len < sizeof(struct ip) &&
217 (m = m_pullup(m, sizeof(struct ip))) == 0)
219 ip = mtod(m, struct ip *);
221 /* Delayed checksums are currently not compatible with divert. */
222 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
223 ip->ip_len = ntohs(ip->ip_len);
225 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
226 ip->ip_len = htons(ip->ip_len);
229 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
230 ip->ip_len = ntohs(ip->ip_len);
231 sctp_delayed_cksum(m);
232 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
233 ip->ip_len = htons(ip->ip_len);
237 * Record receive interface address, if any.
238 * But only for incoming packets.
240 bzero(&divsrc, sizeof(divsrc));
241 divsrc.sin_len = sizeof(divsrc);
242 divsrc.sin_family = AF_INET;
243 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
251 /* Find IP address for receive interface */
252 ifp = m->m_pkthdr.rcvif;
254 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
255 if (ifa->ifa_addr->sa_family != AF_INET)
258 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
261 if_addr_runlock(ifp);
264 * Record the incoming interface name whenever we have one.
266 if (m->m_pkthdr.rcvif) {
268 * Hide the actual interface name in there in the
269 * sin_zero array. XXX This needs to be moved to a
270 * different sockaddr type for divert, e.g.
271 * sockaddr_div with multiple fields like
272 * sockaddr_dl. Presently we have only 7 bytes
273 * but that will do for now as most interfaces
274 * are 4 or less + 2 or less bytes for unit.
275 * There is probably a faster way of doing this,
276 * possibly taking it from the sockaddr_dl on the iface.
277 * This solves the problem of a P2P link and a LAN interface
278 * having the same address, which can result in the wrong
279 * interface being assigned to the packet when fed back
280 * into the divert socket. Theoretically if the daemon saves
281 * and re-uses the sockaddr_in as suggested in the man pages,
282 * this iface name will come along for the ride.
283 * (see div_output for the other half of this.)
285 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
286 sizeof(divsrc.sin_zero));
289 /* Put packet on socket queue, if any */
291 nport = htons((u_int16_t)divert_info(mtag));
292 INP_INFO_RLOCK(&V_divcbinfo);
293 LIST_FOREACH(inp, &V_divcb, inp_list) {
294 /* XXX why does only one socket match? */
295 if (inp->inp_lport == nport) {
297 sa = inp->inp_socket;
298 SOCKBUF_LOCK(&sa->so_rcv);
299 if (sbappendaddr_locked(&sa->so_rcv,
300 (struct sockaddr *)&divsrc, m,
301 (struct mbuf *)0) == 0) {
302 SOCKBUF_UNLOCK(&sa->so_rcv);
303 sa = NULL; /* force mbuf reclaim below */
305 sorwakeup_locked(sa);
310 INP_INFO_RUNLOCK(&V_divcbinfo);
313 KMOD_IPSTAT_INC(ips_noproto);
314 KMOD_IPSTAT_DEC(ips_delivered);
319 * Deliver packet back into the IP processing machinery.
321 * If no address specified, or address is 0.0.0.0, send to ip_output();
322 * otherwise, send to ip_input() and mark as having been received on
323 * the interface with that address.
326 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
327 struct mbuf *control)
330 struct divert_tag *dt;
332 struct mbuf *options;
335 * An mbuf may hasn't come from userland, but we pretend
338 m->m_pkthdr.rcvif = NULL;
340 M_SETFIB(m, so->so_fibnum);
343 m_freem(control); /* XXX */
345 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
346 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
352 dt = (struct divert_tag *)(mtag+1);
353 m_tag_prepend(m, mtag);
355 dt = (struct divert_tag *)(mtag+1);
357 /* Loopback avoidance and state recovery */
361 dt->cookie = sin->sin_port;
363 * Find receive interface with the given name, stuffed
364 * (if it exists) in the sin_zero[] field.
365 * The name is user supplied data so don't trust its size
366 * or that it is zero terminated.
368 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
370 if ( i > 0 && i < sizeof(sin->sin_zero))
371 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
374 /* Reinject packet into the system as incoming or outgoing */
375 if (!sin || sin->sin_addr.s_addr == 0) {
376 struct ip *const ip = mtod(m, struct ip *);
379 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
380 INP_INFO_WLOCK(&V_divcbinfo);
384 * Don't allow both user specified and setsockopt options,
385 * and don't allow packet length sizes that will crash
387 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
388 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
391 INP_INFO_WUNLOCK(&V_divcbinfo);
394 /* Convert fields to host order for ip_output() */
395 ip->ip_len = ntohs(ip->ip_len);
396 ip->ip_off = ntohs(ip->ip_off);
398 /* Send packet to output processing */
399 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
402 mac_inpcb_create_mbuf(inp, m);
405 * Get ready to inject the packet into ip_output().
406 * Just in case socket options were specified on the
407 * divert socket, we duplicate them. This is done
408 * to avoid having to hold the PCB locks over the call
409 * to ip_output(), as doing this results in a number of
410 * lock ordering complexities.
412 * Note that we set the multicast options argument for
413 * ip_output() to NULL since it should be invariant that
414 * they are not present.
416 KASSERT(inp->inp_moptions == NULL,
417 ("multicast options set on a divert socket"));
420 * XXXCSJP: It is unclear to me whether or not it makes
421 * sense for divert sockets to have options. However,
422 * for now we will duplicate them with the INP locks
423 * held so we can use them in ip_output() without
424 * requring a reference to the pcb.
426 if (inp->inp_options != NULL) {
427 options = m_dup(inp->inp_options, M_DONTWAIT);
432 INP_INFO_WUNLOCK(&V_divcbinfo);
433 if (error == ENOBUFS) {
437 error = ip_output(m, options, NULL,
438 ((so->so_options & SO_DONTROUTE) ?
439 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
440 IP_RAWOUTPUT, NULL, NULL);
445 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
446 if (m->m_pkthdr.rcvif == NULL) {
448 * No luck with the name, check by IP address.
449 * Clear the port and the ifname to make sure
450 * there are no distractions for ifa_ifwithaddr.
454 bzero(sin->sin_zero, sizeof(sin->sin_zero));
456 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
458 error = EADDRNOTAVAIL;
461 m->m_pkthdr.rcvif = ifa->ifa_ifp;
465 mac_socket_create_mbuf(so, m);
467 /* Send packet to input processing via netisr */
468 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
479 div_attach(struct socket *so, int proto, struct thread *td)
485 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
487 error = priv_check(td, PRIV_NETINET_DIVERT);
491 error = soreserve(so, div_sendspace, div_recvspace);
494 INP_INFO_WLOCK(&V_divcbinfo);
495 error = in_pcballoc(so, &V_divcbinfo);
497 INP_INFO_WUNLOCK(&V_divcbinfo);
500 inp = (struct inpcb *)so->so_pcb;
501 INP_INFO_WUNLOCK(&V_divcbinfo);
502 inp->inp_ip_p = proto;
503 inp->inp_vflag |= INP_IPV4;
504 inp->inp_flags |= INP_HDRINCL;
510 div_detach(struct socket *so)
515 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
516 INP_INFO_WLOCK(&V_divcbinfo);
520 INP_INFO_WUNLOCK(&V_divcbinfo);
524 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
530 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
531 /* in_pcbbind assumes that nam is a sockaddr_in
532 * and in_pcbbind requires a valid address. Since divert
533 * sockets don't we need to make sure the address is
534 * filled in properly.
535 * XXX -- divert should not be abusing in_pcbind
536 * and should probably have its own family.
538 if (nam->sa_family != AF_INET)
540 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
541 INP_INFO_WLOCK(&V_divcbinfo);
543 error = in_pcbbind(inp, nam, td->td_ucred);
545 INP_INFO_WUNLOCK(&V_divcbinfo);
550 div_shutdown(struct socket *so)
555 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
563 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
564 struct mbuf *control, struct thread *td)
567 /* Packet must have a header (but that's about it) */
568 if (m->m_len < sizeof (struct ip) &&
569 (m = m_pullup(m, sizeof (struct ip))) == 0) {
570 KMOD_IPSTAT_INC(ips_toosmall);
576 return div_output(so, m, (struct sockaddr_in *)nam, control);
580 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
582 struct in_addr faddr;
584 faddr = ((struct sockaddr_in *)sa)->sin_addr;
585 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
587 if (PRC_IS_REDIRECT(cmd))
592 div_pcblist(SYSCTL_HANDLER_ARGS)
595 struct inpcb *inp, **inp_list;
600 * The process of preparing the TCB list is too time-consuming and
601 * resource-intensive to repeat twice on every request.
603 if (req->oldptr == 0) {
604 n = V_divcbinfo.ipi_count;
605 req->oldidx = 2 * (sizeof xig)
606 + (n + n/8) * sizeof(struct xinpcb);
610 if (req->newptr != 0)
614 * OK, now we're committed to doing something.
616 INP_INFO_RLOCK(&V_divcbinfo);
617 gencnt = V_divcbinfo.ipi_gencnt;
618 n = V_divcbinfo.ipi_count;
619 INP_INFO_RUNLOCK(&V_divcbinfo);
621 error = sysctl_wire_old_buffer(req,
622 2 * sizeof(xig) + n*sizeof(struct xinpcb));
626 xig.xig_len = sizeof xig;
628 xig.xig_gen = gencnt;
629 xig.xig_sogen = so_gencnt;
630 error = SYSCTL_OUT(req, &xig, sizeof xig);
634 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
638 INP_INFO_RLOCK(&V_divcbinfo);
639 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
640 inp = LIST_NEXT(inp, inp_list)) {
642 if (inp->inp_gencnt <= gencnt &&
643 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
647 INP_INFO_RUNLOCK(&V_divcbinfo);
651 for (i = 0; i < n; i++) {
654 if (inp->inp_gencnt <= gencnt) {
656 bzero(&xi, sizeof(xi));
657 xi.xi_len = sizeof xi;
658 /* XXX should avoid extra copy */
659 bcopy(inp, &xi.xi_inp, sizeof *inp);
661 sotoxsocket(inp->inp_socket, &xi.xi_socket);
663 error = SYSCTL_OUT(req, &xi, sizeof xi);
669 * Give the user an updated idea of our state.
670 * If the generation differs from what we told
671 * her before, she knows that something happened
672 * while we were processing this request, and it
673 * might be necessary to retry.
675 INP_INFO_RLOCK(&V_divcbinfo);
676 xig.xig_gen = V_divcbinfo.ipi_gencnt;
677 xig.xig_sogen = so_gencnt;
678 xig.xig_count = V_divcbinfo.ipi_count;
679 INP_INFO_RUNLOCK(&V_divcbinfo);
680 error = SYSCTL_OUT(req, &xig, sizeof xig);
682 free(inp_list, M_TEMP);
687 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
688 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
689 div_pcblist, "S,xinpcb", "List of active divert sockets");
692 struct pr_usrreqs div_usrreqs = {
693 .pru_attach = div_attach,
694 .pru_bind = div_bind,
695 .pru_control = in_control,
696 .pru_detach = div_detach,
697 .pru_peeraddr = in_getpeeraddr,
698 .pru_send = div_send,
699 .pru_shutdown = div_shutdown,
700 .pru_sockaddr = in_getsockaddr,
701 .pru_sosetlabel = in_pcbsosetlabel
704 struct protosw div_protosw = {
706 .pr_protocol = IPPROTO_DIVERT,
707 .pr_flags = PR_ATOMIC|PR_ADDR,
708 .pr_input = div_input,
709 .pr_ctlinput = div_ctlinput,
710 .pr_ctloutput = ip_ctloutput,
712 .pr_usrreqs = &div_usrreqs
716 div_modevent(module_t mod, int type, void *unused)
724 * Protocol will be initialized by pf_proto_register().
725 * We don't have to register ip_protox because we are not
726 * a true IP protocol that goes over the wire.
728 err = pf_proto_register(PF_INET, &div_protosw);
729 ip_divert_ptr = divert_packet;
733 * IPDIVERT may normally not be unloaded because of the
734 * potential race conditions. Tell kldunload we can't be
735 * unloaded unless the unload is forced.
743 * Module ipdivert can only be unloaded if no sockets are
744 * connected. Maybe this can be changed later to forcefully
745 * disconnect any open sockets.
747 * XXXRW: Note that there is a slight race here, as a new
748 * socket open request could be spinning on the lock and then
749 * we destroy the lock.
751 INP_INFO_WLOCK(&V_divcbinfo);
752 n = V_divcbinfo.ipi_count;
755 INP_INFO_WUNLOCK(&V_divcbinfo);
758 ip_divert_ptr = NULL;
759 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
760 INP_INFO_WUNLOCK(&V_divcbinfo);
761 INP_INFO_LOCK_DESTROY(&V_divcbinfo);
762 uma_zdestroy(V_divcbinfo.ipi_zone);
771 static moduledata_t ipdivertmod = {
777 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
778 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
779 MODULE_VERSION(ipdivert, 1);