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)
39 #error "IPDIVERT requires INET."
42 #error "IPDIVERT requires IPFIREWALL"
46 #include <sys/param.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/module.h>
52 #include <sys/kernel.h>
55 #include <sys/protosw.h>
56 #include <sys/rwlock.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/systm.h>
63 #include <sys/vimage.h>
68 #include <net/netisr.h>
69 #include <net/route.h>
71 #include <netinet/in.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/in_var.h>
75 #include <netinet/ip.h>
76 #include <netinet/ip_divert.h>
77 #include <netinet/ip_var.h>
78 #include <netinet/ip_fw.h>
79 #include <netinet/vinet.h>
81 #include <netinet/sctp_crc32.h>
84 #include <security/mac/mac_framework.h>
91 * Allocate enough space to hold a full IP packet
93 #define DIVSNDQ (65536 + 100)
94 #define DIVRCVQ (65536 + 100)
97 * Divert sockets work in conjunction with ipfw, see the divert(4)
98 * manpage for features.
99 * Internally, packets selected by ipfw in ip_input() or ip_output(),
100 * and never diverted before, are passed to the input queue of the
101 * divert socket with a given 'divert_port' number (as specified in
102 * the matching ipfw rule), and they are tagged with a 16 bit cookie
103 * (representing the rule number of the matching ipfw rule), which
104 * is passed to process reading from the socket.
106 * Packets written to the divert socket are again tagged with a cookie
107 * (usually the same as above) and a destination address.
108 * If the destination address is INADDR_ANY then the packet is
109 * treated as outgoing and sent to ip_output(), otherwise it is
110 * treated as incoming and sent to ip_input().
111 * In both cases, the packet is tagged with the cookie.
113 * On reinjection, processing in ip_input() and ip_output()
114 * will be exactly the same as for the original packet, except that
115 * ipfw processing will start at the rule number after the one
116 * written in the cookie (so, tagging a packet with a cookie of 0
117 * will cause it to be effectively considered as a standard packet).
120 /* Internal variables. */
121 #ifdef VIMAGE_GLOBALS
122 static struct inpcbhead divcb;
123 static struct inpcbinfo divcbinfo;
126 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
127 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
130 * Initialize divert connection block queue.
133 div_zone_change(void *tag)
136 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
140 div_inpcb_init(void *mem, int size, int flags)
142 struct inpcb *inp = mem;
144 INP_LOCK_INIT(inp, "inp", "divinp");
149 div_inpcb_fini(void *mem, int size)
151 struct inpcb *inp = mem;
153 INP_LOCK_DESTROY(inp);
159 INIT_VNET_INET(curvnet);
161 INP_INFO_LOCK_INIT(&V_divcbinfo, "div");
163 V_divcbinfo.ipi_listhead = &V_divcb;
165 * XXX We don't use the hash list for divert IP, but it's easier
166 * to allocate a one entry hash list than it is to check all
167 * over the place for hashbase == NULL.
169 V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask);
170 V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
171 &V_divcbinfo.ipi_porthashmask);
172 V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
173 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
175 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
176 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
177 NULL, EVENTHANDLER_PRI_ANY);
181 * IPPROTO_DIVERT is not in the real IP protocol number space; this
182 * function should never be called. Just in case, drop any packets.
185 div_input(struct mbuf *m, int off)
187 INIT_VNET_INET(curvnet);
189 V_ipstat.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)
202 INIT_VNET_INET(curvnet);
207 struct sockaddr_in divsrc;
210 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
212 printf("%s: no divert tag\n", __func__);
217 if (m->m_len < sizeof(struct ip) &&
218 (m = m_pullup(m, sizeof(struct ip))) == 0)
220 ip = mtod(m, struct ip *);
222 /* Delayed checksums are currently not compatible with divert. */
223 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
224 ip->ip_len = ntohs(ip->ip_len);
226 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
227 ip->ip_len = htons(ip->ip_len);
230 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
231 ip->ip_len = ntohs(ip->ip_len);
232 sctp_delayed_cksum(m);
233 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
234 ip->ip_len = htons(ip->ip_len);
238 * Record receive interface address, if any.
239 * But only for incoming packets.
241 bzero(&divsrc, sizeof(divsrc));
242 divsrc.sin_len = sizeof(divsrc);
243 divsrc.sin_family = AF_INET;
244 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
251 /* Find IP address for receive interface */
252 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
253 if (ifa->ifa_addr->sa_family != AF_INET)
256 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
261 * Record the incoming interface name whenever we have one.
263 if (m->m_pkthdr.rcvif) {
265 * Hide the actual interface name in there in the
266 * sin_zero array. XXX This needs to be moved to a
267 * different sockaddr type for divert, e.g.
268 * sockaddr_div with multiple fields like
269 * sockaddr_dl. Presently we have only 7 bytes
270 * but that will do for now as most interfaces
271 * are 4 or less + 2 or less bytes for unit.
272 * There is probably a faster way of doing this,
273 * possibly taking it from the sockaddr_dl on the iface.
274 * This solves the problem of a P2P link and a LAN interface
275 * having the same address, which can result in the wrong
276 * interface being assigned to the packet when fed back
277 * into the divert socket. Theoretically if the daemon saves
278 * and re-uses the sockaddr_in as suggested in the man pages,
279 * this iface name will come along for the ride.
280 * (see div_output for the other half of this.)
282 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
283 sizeof(divsrc.sin_zero));
286 /* Put packet on socket queue, if any */
288 nport = htons((u_int16_t)divert_info(mtag));
289 INP_INFO_RLOCK(&V_divcbinfo);
290 LIST_FOREACH(inp, &V_divcb, inp_list) {
291 /* XXX why does only one socket match? */
292 if (inp->inp_lport == nport) {
294 sa = inp->inp_socket;
295 SOCKBUF_LOCK(&sa->so_rcv);
296 if (sbappendaddr_locked(&sa->so_rcv,
297 (struct sockaddr *)&divsrc, m,
298 (struct mbuf *)0) == 0) {
299 SOCKBUF_UNLOCK(&sa->so_rcv);
300 sa = NULL; /* force mbuf reclaim below */
302 sorwakeup_locked(sa);
307 INP_INFO_RUNLOCK(&V_divcbinfo);
310 V_ipstat.ips_noproto++;
311 V_ipstat.ips_delivered--;
316 * Deliver packet back into the IP processing machinery.
318 * If no address specified, or address is 0.0.0.0, send to ip_output();
319 * otherwise, send to ip_input() and mark as having been received on
320 * the interface with that address.
323 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
324 struct mbuf *control)
326 INIT_VNET_INET(curvnet);
328 struct divert_tag *dt;
330 struct mbuf *options;
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);
341 m_freem(control); /* XXX */
343 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
344 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
350 dt = (struct divert_tag *)(mtag+1);
351 m_tag_prepend(m, mtag);
353 dt = (struct divert_tag *)(mtag+1);
355 /* Loopback avoidance and state recovery */
359 dt->cookie = sin->sin_port;
361 * Find receive interface with the given name, stuffed
362 * (if it exists) in the sin_zero[] field.
363 * The name is user supplied data so don't trust its size
364 * or that it is zero terminated.
366 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
368 if ( i > 0 && i < sizeof(sin->sin_zero))
369 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
372 /* Reinject packet into the system as incoming or outgoing */
373 if (!sin || sin->sin_addr.s_addr == 0) {
374 struct ip *const ip = mtod(m, struct ip *);
377 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
378 INP_INFO_WLOCK(&V_divcbinfo);
382 * Don't allow both user specified and setsockopt options,
383 * and don't allow packet length sizes that will crash
385 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
386 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
389 INP_INFO_WUNLOCK(&V_divcbinfo);
392 /* Convert fields to host order for ip_output() */
393 ip->ip_len = ntohs(ip->ip_len);
394 ip->ip_off = ntohs(ip->ip_off);
396 /* Send packet to output processing */
397 V_ipstat.ips_rawout++; /* XXX */
400 mac_inpcb_create_mbuf(inp, m);
403 * Get ready to inject the packet into ip_output().
404 * Just in case socket options were specified on the
405 * divert socket, we duplicate them. This is done
406 * to avoid having to hold the PCB locks over the call
407 * to ip_output(), as doing this results in a number of
408 * lock ordering complexities.
410 * Note that we set the multicast options argument for
411 * ip_output() to NULL since it should be invariant that
412 * they are not present.
414 KASSERT(inp->inp_moptions == NULL,
415 ("multicast options set on a divert socket"));
418 * XXXCSJP: It is unclear to me whether or not it makes
419 * sense for divert sockets to have options. However,
420 * for now we will duplicate them with the INP locks
421 * held so we can use them in ip_output() without
422 * requring a reference to the pcb.
424 if (inp->inp_options != NULL) {
425 options = m_dup(inp->inp_options, M_DONTWAIT);
430 INP_INFO_WUNLOCK(&V_divcbinfo);
431 if (error == ENOBUFS) {
435 error = ip_output(m, options, NULL,
436 ((so->so_options & SO_DONTROUTE) ?
437 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
438 IP_RAWOUTPUT, NULL, NULL);
443 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
444 if (m->m_pkthdr.rcvif == NULL) {
446 * No luck with the name, check by IP address.
447 * Clear the port and the ifname to make sure
448 * there are no distractions for ifa_ifwithaddr.
452 bzero(sin->sin_zero, sizeof(sin->sin_zero));
454 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
456 error = EADDRNOTAVAIL;
459 m->m_pkthdr.rcvif = ifa->ifa_ifp;
463 mac_socket_create_mbuf(so, m);
466 /* Send packet to input processing via netisr */
467 netisr_queue(NETISR_IP, m);
478 div_attach(struct socket *so, int proto, struct thread *td)
480 INIT_VNET_INET(so->so_vnet);
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)
512 INIT_VNET_INET(so->so_vnet);
516 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
517 INP_INFO_WLOCK(&V_divcbinfo);
521 INP_INFO_WUNLOCK(&V_divcbinfo);
525 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
527 INIT_VNET_INET(so->so_vnet);
532 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
533 /* in_pcbbind assumes that nam is a sockaddr_in
534 * and in_pcbbind requires a valid address. Since divert
535 * sockets don't we need to make sure the address is
536 * filled in properly.
537 * XXX -- divert should not be abusing in_pcbind
538 * and should probably have its own family.
540 if (nam->sa_family != AF_INET)
542 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
543 INP_INFO_WLOCK(&V_divcbinfo);
545 error = in_pcbbind(inp, nam, td->td_ucred);
547 INP_INFO_WUNLOCK(&V_divcbinfo);
552 div_shutdown(struct socket *so)
557 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
565 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
566 struct mbuf *control, struct thread *td)
568 INIT_VNET_INET(so->so_vnet);
570 /* Packet must have a header (but that's about it) */
571 if (m->m_len < sizeof (struct ip) &&
572 (m = m_pullup(m, sizeof (struct ip))) == 0) {
573 V_ipstat.ips_toosmall++;
579 return div_output(so, m, (struct sockaddr_in *)nam, control);
583 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
585 struct in_addr faddr;
587 faddr = ((struct sockaddr_in *)sa)->sin_addr;
588 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
590 if (PRC_IS_REDIRECT(cmd))
595 div_pcblist(SYSCTL_HANDLER_ARGS)
597 INIT_VNET_INET(curvnet);
599 struct inpcb *inp, **inp_list;
604 * The process of preparing the TCB list is too time-consuming and
605 * resource-intensive to repeat twice on every request.
607 if (req->oldptr == 0) {
608 n = V_divcbinfo.ipi_count;
609 req->oldidx = 2 * (sizeof xig)
610 + (n + n/8) * sizeof(struct xinpcb);
614 if (req->newptr != 0)
618 * OK, now we're committed to doing something.
620 INP_INFO_RLOCK(&V_divcbinfo);
621 gencnt = V_divcbinfo.ipi_gencnt;
622 n = V_divcbinfo.ipi_count;
623 INP_INFO_RUNLOCK(&V_divcbinfo);
625 error = sysctl_wire_old_buffer(req,
626 2 * sizeof(xig) + n*sizeof(struct xinpcb));
630 xig.xig_len = sizeof xig;
632 xig.xig_gen = gencnt;
633 xig.xig_sogen = so_gencnt;
634 error = SYSCTL_OUT(req, &xig, sizeof xig);
638 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
642 INP_INFO_RLOCK(&V_divcbinfo);
643 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
644 inp = LIST_NEXT(inp, inp_list)) {
646 if (inp->inp_gencnt <= gencnt &&
647 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
651 INP_INFO_RUNLOCK(&V_divcbinfo);
655 for (i = 0; i < n; i++) {
658 if (inp->inp_gencnt <= gencnt) {
660 bzero(&xi, sizeof(xi));
661 xi.xi_len = sizeof xi;
662 /* XXX should avoid extra copy */
663 bcopy(inp, &xi.xi_inp, sizeof *inp);
665 sotoxsocket(inp->inp_socket, &xi.xi_socket);
667 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 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, CTLFLAG_RD, 0, 0,
693 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,
716 .pr_usrreqs = &div_usrreqs
720 div_modevent(module_t mod, int type, void *unused)
728 * Protocol will be initialized by pf_proto_register().
729 * We don't have to register ip_protox because we are not
730 * a true IP protocol that goes over the wire.
732 err = pf_proto_register(PF_INET, &div_protosw);
733 ip_divert_ptr = divert_packet;
737 * IPDIVERT may normally not be unloaded because of the
738 * potential race conditions. Tell kldunload we can't be
739 * unloaded unless the unload is forced.
747 * Module ipdivert can only be unloaded if no sockets are
748 * connected. Maybe this can be changed later to forcefully
749 * disconnect any open sockets.
751 * XXXRW: Note that there is a slight race here, as a new
752 * socket open request could be spinning on the lock and then
753 * we destroy the lock.
755 INP_INFO_WLOCK(&V_divcbinfo);
756 n = V_divcbinfo.ipi_count;
759 INP_INFO_WUNLOCK(&V_divcbinfo);
762 ip_divert_ptr = NULL;
763 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
764 INP_INFO_WUNLOCK(&V_divcbinfo);
765 INP_INFO_LOCK_DESTROY(&V_divcbinfo);
766 uma_zdestroy(V_divcbinfo.ipi_zone);
775 static moduledata_t ipdivertmod = {
781 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
782 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
783 MODULE_VERSION(ipdivert, 1);