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
32 #if !defined(KLD_MODULE)
37 #error "IPDIVERT requires INET."
40 #error "IPDIVERT requires IPFIREWALL"
44 #include <sys/param.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/signalvar.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
64 #include <net/route.h>
66 #include <netinet/in.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip_divert.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/ip_fw.h>
80 * Allocate enough space to hold a full IP packet
82 #define DIVSNDQ (65536 + 100)
83 #define DIVRCVQ (65536 + 100)
86 * Divert sockets work in conjunction with ipfw, see the divert(4)
87 * manpage for features.
88 * Internally, packets selected by ipfw in ip_input() or ip_output(),
89 * and never diverted before, are passed to the input queue of the
90 * divert socket with a given 'divert_port' number (as specified in
91 * the matching ipfw rule), and they are tagged with a 16 bit cookie
92 * (representing the rule number of the matching ipfw rule), which
93 * is passed to process reading from the socket.
95 * Packets written to the divert socket are again tagged with a cookie
96 * (usually the same as above) and a destination address.
97 * If the destination address is INADDR_ANY then the packet is
98 * treated as outgoing and sent to ip_output(), otherwise it is
99 * treated as incoming and sent to ip_input().
100 * In both cases, the packet is tagged with the cookie.
102 * On reinjection, processing in ip_input() and ip_output()
103 * will be exactly the same as for the original packet, except that
104 * ipfw processing will start at the rule number after the one
105 * written in the cookie (so, tagging a packet with a cookie of 0
106 * will cause it to be effectively considered as a standard packet).
109 /* Internal variables. */
110 static struct inpcbhead divcb;
111 static struct inpcbinfo divcbinfo;
113 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
114 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
117 * Initialize divert connection block queue.
122 INP_INFO_LOCK_INIT(&divcbinfo, "div");
124 divcbinfo.listhead = &divcb;
126 * XXX We don't use the hash list for divert IP, but it's easier
127 * to allocate a one entry hash list than it is to check all
128 * over the place for hashbase == NULL.
130 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
131 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
132 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
133 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
134 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
138 * IPPROTO_DIVERT is not in the real IP protocol number space; this
139 * function should never be called. Just in case, drop any packets.
142 div_input(struct mbuf *m, int off)
144 ipstat.ips_noproto++;
149 * Divert a packet by passing it up to the divert socket at port 'port'.
151 * Setup generic address and protocol structures for div_input routine,
152 * then pass them along with mbuf chain.
155 divert_packet(struct mbuf *m, int incoming)
161 struct sockaddr_in divsrc;
164 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
166 printf("%s: no divert tag\n", __func__);
171 if (m->m_len < sizeof(struct ip) &&
172 (m = m_pullup(m, sizeof(struct ip))) == 0)
174 ip = mtod(m, struct ip *);
176 /* Delayed checksums are currently not compatible with divert. */
177 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
178 ip->ip_len = ntohs(ip->ip_len);
180 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
181 ip->ip_len = htons(ip->ip_len);
185 * Record receive interface address, if any.
186 * But only for incoming packets.
188 bzero(&divsrc, sizeof(divsrc));
189 divsrc.sin_len = sizeof(divsrc);
190 divsrc.sin_family = AF_INET;
191 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
198 /* Find IP address for receive interface */
199 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
200 if (ifa->ifa_addr == NULL)
202 if (ifa->ifa_addr->sa_family != AF_INET)
205 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
210 * Record the incoming interface name whenever we have one.
212 if (m->m_pkthdr.rcvif) {
214 * Hide the actual interface name in there in the
215 * sin_zero array. XXX This needs to be moved to a
216 * different sockaddr type for divert, e.g.
217 * sockaddr_div with multiple fields like
218 * sockaddr_dl. Presently we have only 7 bytes
219 * but that will do for now as most interfaces
220 * are 4 or less + 2 or less bytes for unit.
221 * There is probably a faster way of doing this,
222 * possibly taking it from the sockaddr_dl on the iface.
223 * This solves the problem of a P2P link and a LAN interface
224 * having the same address, which can result in the wrong
225 * interface being assigned to the packet when fed back
226 * into the divert socket. Theoretically if the daemon saves
227 * and re-uses the sockaddr_in as suggested in the man pages,
228 * this iface name will come along for the ride.
229 * (see div_output for the other half of this.)
231 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
232 sizeof(divsrc.sin_zero));
235 /* Put packet on socket queue, if any */
237 nport = htons((u_int16_t)divert_info(mtag));
238 INP_INFO_RLOCK(&divcbinfo);
239 LIST_FOREACH(inp, &divcb, inp_list) {
241 /* XXX why does only one socket match? */
242 if (inp->inp_lport == nport) {
243 sa = inp->inp_socket;
244 SOCKBUF_LOCK(&sa->so_rcv);
245 if (sbappendaddr_locked(&sa->so_rcv,
246 (struct sockaddr *)&divsrc, m,
247 (struct mbuf *)0) == 0) {
248 SOCKBUF_UNLOCK(&sa->so_rcv);
249 sa = NULL; /* force mbuf reclaim below */
251 sorwakeup_locked(sa);
257 INP_INFO_RUNLOCK(&divcbinfo);
260 ipstat.ips_noproto++;
261 ipstat.ips_delivered--;
266 * Deliver packet back into the IP processing machinery.
268 * If no address specified, or address is 0.0.0.0, send to ip_output();
269 * otherwise, send to ip_input() and mark as having been received on
270 * the interface with that address.
273 div_output(struct socket *so, struct mbuf *m,
274 struct sockaddr_in *sin, struct mbuf *control)
277 struct divert_tag *dt;
280 m->m_pkthdr.rcvif = NULL;
283 m_freem(control); /* XXX */
285 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
286 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
292 dt = (struct divert_tag *)(mtag+1);
293 m_tag_prepend(m, mtag);
295 dt = (struct divert_tag *)(mtag+1);
297 /* Loopback avoidance and state recovery */
301 dt->cookie = sin->sin_port;
303 * Find receive interface with the given name, stuffed
304 * (if it exists) in the sin_zero[] field.
305 * The name is user supplied data so don't trust its size
306 * or that it is zero terminated.
308 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
310 if ( i > 0 && i < sizeof(sin->sin_zero))
311 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
314 /* Reinject packet into the system as incoming or outgoing */
315 if (!sin || sin->sin_addr.s_addr == 0) {
316 struct ip *const ip = mtod(m, struct ip *);
319 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
320 INP_INFO_WLOCK(&divcbinfo);
324 * Don't allow both user specified and setsockopt options,
325 * and don't allow packet length sizes that will crash
327 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
328 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
332 /* Convert fields to host order for ip_output() */
333 ip->ip_len = ntohs(ip->ip_len);
334 ip->ip_off = ntohs(ip->ip_off);
336 /* Send packet to output processing */
337 ipstat.ips_rawout++; /* XXX */
340 mac_create_mbuf_from_inpcb(inp, m);
343 inp->inp_options, NULL,
344 ((so->so_options & SO_DONTROUTE) ?
346 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
347 inp->inp_moptions, NULL);
350 INP_INFO_WUNLOCK(&divcbinfo);
352 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
353 if (m->m_pkthdr.rcvif == NULL) {
355 * No luck with the name, check by IP address.
356 * Clear the port and the ifname to make sure
357 * there are no distractions for ifa_ifwithaddr.
361 bzero(sin->sin_zero, sizeof(sin->sin_zero));
363 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
365 error = EADDRNOTAVAIL;
368 m->m_pkthdr.rcvif = ifa->ifa_ifp;
372 mac_create_mbuf_from_socket(so, m);
375 /* Send packet to input processing */
387 div_attach(struct socket *so, int proto, struct thread *td)
392 INP_INFO_WLOCK(&divcbinfo);
395 INP_INFO_WUNLOCK(&divcbinfo);
398 if (td && (error = suser(td)) != 0) {
399 INP_INFO_WUNLOCK(&divcbinfo);
402 error = soreserve(so, div_sendspace, div_recvspace);
404 INP_INFO_WUNLOCK(&divcbinfo);
407 error = in_pcballoc(so, &divcbinfo, "divinp");
409 INP_INFO_WUNLOCK(&divcbinfo);
412 inp = (struct inpcb *)so->so_pcb;
414 INP_INFO_WUNLOCK(&divcbinfo);
415 inp->inp_ip_p = proto;
416 inp->inp_vflag |= INP_IPV4;
417 inp->inp_flags |= INP_HDRINCL;
423 div_detach(struct socket *so)
427 INP_INFO_WLOCK(&divcbinfo);
430 INP_INFO_WUNLOCK(&divcbinfo);
435 INP_INFO_WUNLOCK(&divcbinfo);
440 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
445 INP_INFO_WLOCK(&divcbinfo);
448 INP_INFO_WUNLOCK(&divcbinfo);
451 /* in_pcbbind assumes that nam is a sockaddr_in
452 * and in_pcbbind requires a valid address. Since divert
453 * sockets don't we need to make sure the address is
454 * filled in properly.
455 * XXX -- divert should not be abusing in_pcbind
456 * and should probably have its own family.
458 if (nam->sa_family != AF_INET)
459 error = EAFNOSUPPORT;
461 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
463 error = in_pcbbind(inp, nam, td->td_ucred);
466 INP_INFO_WUNLOCK(&divcbinfo);
471 div_shutdown(struct socket *so)
475 INP_INFO_RLOCK(&divcbinfo);
478 INP_INFO_RUNLOCK(&divcbinfo);
482 INP_INFO_RUNLOCK(&divcbinfo);
489 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
490 struct mbuf *control, struct thread *td)
492 /* Packet must have a header (but that's about it) */
493 if (m->m_len < sizeof (struct ip) &&
494 (m = m_pullup(m, sizeof (struct ip))) == 0) {
495 ipstat.ips_toosmall++;
501 return div_output(so, m, (struct sockaddr_in *)nam, control);
505 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
507 struct in_addr faddr;
509 faddr = ((struct sockaddr_in *)sa)->sin_addr;
510 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
512 if (PRC_IS_REDIRECT(cmd))
517 div_pcblist(SYSCTL_HANDLER_ARGS)
520 struct inpcb *inp, **inp_list;
525 * The process of preparing the TCB list is too time-consuming and
526 * resource-intensive to repeat twice on every request.
528 if (req->oldptr == 0) {
529 n = divcbinfo.ipi_count;
530 req->oldidx = 2 * (sizeof xig)
531 + (n + n/8) * sizeof(struct xinpcb);
535 if (req->newptr != 0)
539 * OK, now we're committed to doing something.
541 INP_INFO_RLOCK(&divcbinfo);
542 gencnt = divcbinfo.ipi_gencnt;
543 n = divcbinfo.ipi_count;
544 INP_INFO_RUNLOCK(&divcbinfo);
546 error = sysctl_wire_old_buffer(req,
547 2 * sizeof(xig) + n*sizeof(struct xinpcb));
551 xig.xig_len = sizeof xig;
553 xig.xig_gen = gencnt;
554 xig.xig_sogen = so_gencnt;
555 error = SYSCTL_OUT(req, &xig, sizeof xig);
559 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
563 INP_INFO_RLOCK(&divcbinfo);
564 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
565 inp = LIST_NEXT(inp, inp_list)) {
567 if (inp->inp_gencnt <= gencnt &&
568 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
572 INP_INFO_RUNLOCK(&divcbinfo);
576 for (i = 0; i < n; i++) {
578 if (inp->inp_gencnt <= gencnt) {
580 xi.xi_len = sizeof xi;
581 /* XXX should avoid extra copy */
582 bcopy(inp, &xi.xi_inp, sizeof *inp);
584 sotoxsocket(inp->inp_socket, &xi.xi_socket);
585 error = SYSCTL_OUT(req, &xi, sizeof xi);
590 * Give the user an updated idea of our state.
591 * If the generation differs from what we told
592 * her before, she knows that something happened
593 * while we were processing this request, and it
594 * might be necessary to retry.
596 INP_INFO_RLOCK(&divcbinfo);
597 xig.xig_gen = divcbinfo.ipi_gencnt;
598 xig.xig_sogen = so_gencnt;
599 xig.xig_count = divcbinfo.ipi_count;
600 INP_INFO_RUNLOCK(&divcbinfo);
601 error = SYSCTL_OUT(req, &xig, sizeof xig);
603 free(inp_list, M_TEMP);
608 * This is the wrapper function for in_setsockaddr. We just pass down
609 * the pcbinfo for in_setpeeraddr to lock.
612 div_sockaddr(struct socket *so, struct sockaddr **nam)
614 return (in_setsockaddr(so, nam, &divcbinfo));
618 * This is the wrapper function for in_setpeeraddr. We just pass down
619 * the pcbinfo for in_setpeeraddr to lock.
622 div_peeraddr(struct socket *so, struct sockaddr **nam)
624 return (in_setpeeraddr(so, nam, &divcbinfo));
628 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
629 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
630 div_pcblist, "S,xinpcb", "List of active divert sockets");
633 struct pr_usrreqs div_usrreqs = {
634 .pru_attach = div_attach,
635 .pru_bind = div_bind,
636 .pru_control = in_control,
637 .pru_detach = div_detach,
638 .pru_peeraddr = div_peeraddr,
639 .pru_send = div_send,
640 .pru_shutdown = div_shutdown,
641 .pru_sockaddr = div_sockaddr,
642 .pru_sosetlabel = in_pcbsosetlabel
645 struct protosw div_protosw = {
646 SOCK_RAW, NULL, IPPROTO_DIVERT, PR_ATOMIC|PR_ADDR,
647 div_input, NULL, div_ctlinput, ip_ctloutput,
649 div_init, NULL, NULL, NULL,
654 div_modevent(module_t mod, int type, void *unused)
662 * Protocol will be initialized by pf_proto_register().
663 * We don't have to register ip_protox because we are not
664 * a true IP protocol that goes over the wire.
666 err = pf_proto_register(PF_INET, &div_protosw);
667 ip_divert_ptr = divert_packet;
671 * IPDIVERT may normally not be unloaded because of the
672 * potential race conditions. Tell kldunload we can't be
673 * unloaded unless the unload is forced.
681 * Module ipdivert can only be unloaded if no sockets are
682 * connected. Maybe this can be changed later to forcefully
683 * disconnect any open sockets.
685 * XXXRW: Note that there is a slight race here, as a new
686 * socket open request could be spinning on the lock and then
687 * we destroy the lock.
689 INP_INFO_WLOCK(&divcbinfo);
690 n = divcbinfo.ipi_count;
693 INP_INFO_WUNLOCK(&divcbinfo);
696 ip_divert_ptr = NULL;
697 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
698 INP_INFO_WUNLOCK(&divcbinfo);
699 INP_INFO_LOCK_DESTROY(&divcbinfo);
700 uma_zdestroy(divcbinfo.ipi_zone);
709 static moduledata_t ipdivertmod = {
715 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
716 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
717 MODULE_VERSION(ipdivert, 1);