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.
120 div_zone_change(void *tag)
123 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
127 div_inpcb_init(void *mem, int size, int flags)
129 struct inpcb *inp = (struct inpcb *) mem;
130 INP_LOCK_INIT(inp, "inp", "divinp");
135 div_inpcb_fini(void *mem, int size)
137 struct inpcb *inp = (struct inpcb *) mem;
138 INP_LOCK_DESTROY(inp);
145 INP_INFO_LOCK_INIT(&divcbinfo, "div");
147 divcbinfo.listhead = &divcb;
149 * XXX We don't use the hash list for divert IP, but it's easier
150 * to allocate a one entry hash list than it is to check all
151 * over the place for hashbase == NULL.
153 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
154 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
155 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
156 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
157 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
158 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
159 NULL, EVENTHANDLER_PRI_ANY);
163 * IPPROTO_DIVERT is not in the real IP protocol number space; this
164 * function should never be called. Just in case, drop any packets.
167 div_input(struct mbuf *m, int off)
169 ipstat.ips_noproto++;
174 * Divert a packet by passing it up to the divert socket at port 'port'.
176 * Setup generic address and protocol structures for div_input routine,
177 * then pass them along with mbuf chain.
180 divert_packet(struct mbuf *m, int incoming)
186 struct sockaddr_in divsrc;
189 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
191 printf("%s: no divert tag\n", __func__);
196 if (m->m_len < sizeof(struct ip) &&
197 (m = m_pullup(m, sizeof(struct ip))) == 0)
199 ip = mtod(m, struct ip *);
201 /* Delayed checksums are currently not compatible with divert. */
202 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
203 ip->ip_len = ntohs(ip->ip_len);
205 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
206 ip->ip_len = htons(ip->ip_len);
210 * Record receive interface address, if any.
211 * But only for incoming packets.
213 bzero(&divsrc, sizeof(divsrc));
214 divsrc.sin_len = sizeof(divsrc);
215 divsrc.sin_family = AF_INET;
216 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
223 /* Find IP address for receive interface */
224 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
225 if (ifa->ifa_addr->sa_family != AF_INET)
228 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
233 * Record the incoming interface name whenever we have one.
235 if (m->m_pkthdr.rcvif) {
237 * Hide the actual interface name in there in the
238 * sin_zero array. XXX This needs to be moved to a
239 * different sockaddr type for divert, e.g.
240 * sockaddr_div with multiple fields like
241 * sockaddr_dl. Presently we have only 7 bytes
242 * but that will do for now as most interfaces
243 * are 4 or less + 2 or less bytes for unit.
244 * There is probably a faster way of doing this,
245 * possibly taking it from the sockaddr_dl on the iface.
246 * This solves the problem of a P2P link and a LAN interface
247 * having the same address, which can result in the wrong
248 * interface being assigned to the packet when fed back
249 * into the divert socket. Theoretically if the daemon saves
250 * and re-uses the sockaddr_in as suggested in the man pages,
251 * this iface name will come along for the ride.
252 * (see div_output for the other half of this.)
254 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
255 sizeof(divsrc.sin_zero));
258 /* Put packet on socket queue, if any */
260 nport = htons((u_int16_t)divert_info(mtag));
261 INP_INFO_RLOCK(&divcbinfo);
262 LIST_FOREACH(inp, &divcb, inp_list) {
264 /* XXX why does only one socket match? */
265 if (inp->inp_lport == nport) {
266 sa = inp->inp_socket;
267 SOCKBUF_LOCK(&sa->so_rcv);
268 if (sbappendaddr_locked(&sa->so_rcv,
269 (struct sockaddr *)&divsrc, m,
270 (struct mbuf *)0) == 0) {
271 SOCKBUF_UNLOCK(&sa->so_rcv);
272 sa = NULL; /* force mbuf reclaim below */
274 sorwakeup_locked(sa);
280 INP_INFO_RUNLOCK(&divcbinfo);
283 ipstat.ips_noproto++;
284 ipstat.ips_delivered--;
289 * Deliver packet back into the IP processing machinery.
291 * If no address specified, or address is 0.0.0.0, send to ip_output();
292 * otherwise, send to ip_input() and mark as having been received on
293 * the interface with that address.
296 div_output(struct socket *so, struct mbuf *m,
297 struct sockaddr_in *sin, struct mbuf *control)
300 struct divert_tag *dt;
304 * An mbuf may hasn't come from userland, but we pretend
307 m->m_pkthdr.rcvif = NULL;
311 m_freem(control); /* XXX */
313 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
314 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
320 dt = (struct divert_tag *)(mtag+1);
321 m_tag_prepend(m, mtag);
323 dt = (struct divert_tag *)(mtag+1);
325 /* Loopback avoidance and state recovery */
329 dt->cookie = sin->sin_port;
331 * Find receive interface with the given name, stuffed
332 * (if it exists) in the sin_zero[] field.
333 * The name is user supplied data so don't trust its size
334 * or that it is zero terminated.
336 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
338 if ( i > 0 && i < sizeof(sin->sin_zero))
339 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
342 /* Reinject packet into the system as incoming or outgoing */
343 if (!sin || sin->sin_addr.s_addr == 0) {
344 struct ip *const ip = mtod(m, struct ip *);
347 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
348 INP_INFO_WLOCK(&divcbinfo);
352 * Don't allow both user specified and setsockopt options,
353 * and don't allow packet length sizes that will crash
355 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
356 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
360 /* Convert fields to host order for ip_output() */
361 ip->ip_len = ntohs(ip->ip_len);
362 ip->ip_off = ntohs(ip->ip_off);
364 /* Send packet to output processing */
365 ipstat.ips_rawout++; /* XXX */
368 mac_create_mbuf_from_inpcb(inp, m);
371 inp->inp_options, NULL,
372 ((so->so_options & SO_DONTROUTE) ?
374 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
375 inp->inp_moptions, NULL);
378 INP_INFO_WUNLOCK(&divcbinfo);
380 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
381 if (m->m_pkthdr.rcvif == NULL) {
383 * No luck with the name, check by IP address.
384 * Clear the port and the ifname to make sure
385 * there are no distractions for ifa_ifwithaddr.
389 bzero(sin->sin_zero, sizeof(sin->sin_zero));
391 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
393 error = EADDRNOTAVAIL;
396 m->m_pkthdr.rcvif = ifa->ifa_ifp;
400 mac_create_mbuf_from_socket(so, m);
403 /* Send packet to input processing */
415 div_attach(struct socket *so, int proto, struct thread *td)
421 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
422 if (td && (error = suser(td)) != 0)
424 error = soreserve(so, div_sendspace, div_recvspace);
427 INP_INFO_WLOCK(&divcbinfo);
428 error = in_pcballoc(so, &divcbinfo);
430 INP_INFO_WUNLOCK(&divcbinfo);
433 inp = (struct inpcb *)so->so_pcb;
434 INP_INFO_WUNLOCK(&divcbinfo);
435 inp->inp_ip_p = proto;
436 inp->inp_vflag |= INP_IPV4;
437 inp->inp_flags |= INP_HDRINCL;
443 div_detach(struct socket *so)
448 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
449 INP_INFO_WLOCK(&divcbinfo);
453 INP_INFO_WUNLOCK(&divcbinfo);
457 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
463 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
464 /* in_pcbbind assumes that nam is a sockaddr_in
465 * and in_pcbbind requires a valid address. Since divert
466 * sockets don't we need to make sure the address is
467 * filled in properly.
468 * XXX -- divert should not be abusing in_pcbind
469 * and should probably have its own family.
471 if (nam->sa_family != AF_INET)
473 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
474 INP_INFO_WLOCK(&divcbinfo);
476 error = in_pcbbind(inp, nam, td->td_ucred);
478 INP_INFO_WUNLOCK(&divcbinfo);
483 div_shutdown(struct socket *so)
488 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
496 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
497 struct mbuf *control, struct thread *td)
499 /* Packet must have a header (but that's about it) */
500 if (m->m_len < sizeof (struct ip) &&
501 (m = m_pullup(m, sizeof (struct ip))) == 0) {
502 ipstat.ips_toosmall++;
508 return div_output(so, m, (struct sockaddr_in *)nam, control);
512 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
514 struct in_addr faddr;
516 faddr = ((struct sockaddr_in *)sa)->sin_addr;
517 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
519 if (PRC_IS_REDIRECT(cmd))
524 div_pcblist(SYSCTL_HANDLER_ARGS)
527 struct inpcb *inp, **inp_list;
532 * The process of preparing the TCB list is too time-consuming and
533 * resource-intensive to repeat twice on every request.
535 if (req->oldptr == 0) {
536 n = divcbinfo.ipi_count;
537 req->oldidx = 2 * (sizeof xig)
538 + (n + n/8) * sizeof(struct xinpcb);
542 if (req->newptr != 0)
546 * OK, now we're committed to doing something.
548 INP_INFO_RLOCK(&divcbinfo);
549 gencnt = divcbinfo.ipi_gencnt;
550 n = divcbinfo.ipi_count;
551 INP_INFO_RUNLOCK(&divcbinfo);
553 error = sysctl_wire_old_buffer(req,
554 2 * sizeof(xig) + n*sizeof(struct xinpcb));
558 xig.xig_len = sizeof xig;
560 xig.xig_gen = gencnt;
561 xig.xig_sogen = so_gencnt;
562 error = SYSCTL_OUT(req, &xig, sizeof xig);
566 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
570 INP_INFO_RLOCK(&divcbinfo);
571 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
572 inp = LIST_NEXT(inp, inp_list)) {
574 if (inp->inp_gencnt <= gencnt &&
575 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
579 INP_INFO_RUNLOCK(&divcbinfo);
583 for (i = 0; i < n; i++) {
586 if (inp->inp_gencnt <= gencnt) {
588 bzero(&xi, sizeof(xi));
589 xi.xi_len = sizeof xi;
590 /* XXX should avoid extra copy */
591 bcopy(inp, &xi.xi_inp, sizeof *inp);
593 sotoxsocket(inp->inp_socket, &xi.xi_socket);
595 error = SYSCTL_OUT(req, &xi, sizeof xi);
601 * Give the user an updated idea of our state.
602 * If the generation differs from what we told
603 * her before, she knows that something happened
604 * while we were processing this request, and it
605 * might be necessary to retry.
607 INP_INFO_RLOCK(&divcbinfo);
608 xig.xig_gen = divcbinfo.ipi_gencnt;
609 xig.xig_sogen = so_gencnt;
610 xig.xig_count = divcbinfo.ipi_count;
611 INP_INFO_RUNLOCK(&divcbinfo);
612 error = SYSCTL_OUT(req, &xig, sizeof xig);
614 free(inp_list, M_TEMP);
619 * This is the wrapper function for in_setsockaddr. We just pass down
620 * the pcbinfo for in_setpeeraddr to lock.
623 div_sockaddr(struct socket *so, struct sockaddr **nam)
625 return (in_setsockaddr(so, nam, &divcbinfo));
629 * This is the wrapper function for in_setpeeraddr. We just pass down
630 * the pcbinfo for in_setpeeraddr to lock.
633 div_peeraddr(struct socket *so, struct sockaddr **nam)
635 return (in_setpeeraddr(so, nam, &divcbinfo));
639 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
640 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
641 div_pcblist, "S,xinpcb", "List of active divert sockets");
644 struct pr_usrreqs div_usrreqs = {
645 .pru_attach = div_attach,
646 .pru_bind = div_bind,
647 .pru_control = in_control,
648 .pru_detach = div_detach,
649 .pru_peeraddr = div_peeraddr,
650 .pru_send = div_send,
651 .pru_shutdown = div_shutdown,
652 .pru_sockaddr = div_sockaddr,
653 .pru_sosetlabel = in_pcbsosetlabel
656 struct protosw div_protosw = {
658 .pr_protocol = IPPROTO_DIVERT,
659 .pr_flags = PR_ATOMIC|PR_ADDR,
660 .pr_input = div_input,
661 .pr_ctlinput = div_ctlinput,
662 .pr_ctloutput = ip_ctloutput,
664 .pr_usrreqs = &div_usrreqs
668 div_modevent(module_t mod, int type, void *unused)
676 * Protocol will be initialized by pf_proto_register().
677 * We don't have to register ip_protox because we are not
678 * a true IP protocol that goes over the wire.
680 err = pf_proto_register(PF_INET, &div_protosw);
681 ip_divert_ptr = divert_packet;
685 * IPDIVERT may normally not be unloaded because of the
686 * potential race conditions. Tell kldunload we can't be
687 * unloaded unless the unload is forced.
695 * Module ipdivert can only be unloaded if no sockets are
696 * connected. Maybe this can be changed later to forcefully
697 * disconnect any open sockets.
699 * XXXRW: Note that there is a slight race here, as a new
700 * socket open request could be spinning on the lock and then
701 * we destroy the lock.
703 INP_INFO_WLOCK(&divcbinfo);
704 n = divcbinfo.ipi_count;
707 INP_INFO_WUNLOCK(&divcbinfo);
710 ip_divert_ptr = NULL;
711 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
712 INP_INFO_WUNLOCK(&divcbinfo);
713 INP_INFO_LOCK_DESTROY(&divcbinfo);
714 uma_zdestroy(divcbinfo.ipi_zone);
723 static moduledata_t ipdivertmod = {
729 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
730 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
731 MODULE_VERSION(ipdivert, 1);