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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
34 * $DragonFly: src/sys/netinet/ip_divert.c,v 1.40 2008/10/21 13:51:01 sephe Exp $
41 #include "opt_ipdivert.h"
42 #include "opt_ipsec.h"
45 #error "IPDIVERT requires INET."
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/socket.h>
53 #include <sys/protosw.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <sys/systm.h>
59 #include <sys/in_cksum.h>
62 #include <sys/msgport.h>
66 #include <net/route.h>
69 #include <net/netmsg2.h>
71 #include <sys/thread2.h>
72 #include <sys/mplock2.h>
74 #include <netinet/in.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/ip.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip_var.h>
80 #include <netinet/ip_divert.h>
87 * Allocate enough space to hold a full IP packet
89 #define DIVSNDQ (65536 + 100)
90 #define DIVRCVQ (65536 + 100)
92 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
94 #define DIV_OUTPUT 0x10000
95 #define DIV_INPUT 0x20000
98 * Divert sockets work in conjunction with ipfw, see the divert(4)
99 * manpage for features.
100 * Internally, packets selected by ipfw in ip_input() or ip_output(),
101 * and never diverted before, are passed to the input queue of the
102 * divert socket with a given 'divert_port' number (as specified in
103 * the matching ipfw rule), and they are tagged with a 16 bit cookie
104 * (representing the rule number of the matching ipfw rule), which
105 * is passed to process reading from the socket.
107 * Packets written to the divert socket are again tagged with a cookie
108 * (usually the same as above) and a destination address.
109 * If the destination address is INADDR_ANY then the packet is
110 * treated as outgoing and sent to ip_output(), otherwise it is
111 * treated as incoming and sent to ip_input().
112 * In both cases, the packet is tagged with the cookie.
114 * On reinjection, processing in ip_input() and ip_output()
115 * will be exactly the same as for the original packet, except that
116 * ipfw processing will start at the rule number after the one
117 * written in the cookie (so, tagging a packet with a cookie of 0
118 * will cause it to be effectively considered as a standard packet).
121 /* Internal variables */
122 static struct inpcbinfo divcbinfo;
124 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
125 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
127 static struct mbuf *ip_divert(struct mbuf *, int, int);
129 static struct lwkt_token div_token = LWKT_TOKEN_MP_INITIALIZER(div_token);
132 * Initialize divert connection block queue.
137 in_pcbinfo_init(&divcbinfo);
139 * XXX We don't use the hash list for divert IP, but it's easier
140 * to allocate a one entry hash list than it is to check all
141 * over the place for hashbase == NULL.
143 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
144 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
145 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
146 &divcbinfo.wildcardhashmask);
147 divcbinfo.ipi_size = sizeof(struct inpcb);
148 ip_divert_p = ip_divert;
152 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
153 * with that protocol number to enter the system from the outside.
156 div_input(struct mbuf *m, ...)
158 ipstat.ips_noproto++;
163 div_soport(struct socket *so, struct sockaddr *nam, struct mbuf **mptr)
165 struct sockaddr_in *sin;
169 sin = (struct sockaddr_in *)nam;
173 m->m_pkthdr.rcvif = NULL;
174 dir = DIV_IS_OUTPUT(sin) ? IP_MPORT_OUT : IP_MPORT_IN;
180 * Try locating the interface, if we originally had one.
181 * This is done even for outgoing packets, since for a
182 * forwarded packet, there must be an interface attached.
184 * Find receive interface with the given name, stuffed
185 * (if it exists) in the sin_zero[] field.
186 * The name is user supplied data so don't trust its size
187 * or that it is zero terminated.
189 for (i = 0; sin->sin_zero[i] && i < sizeof(sin->sin_zero); i++)
191 if (i > 0 && i < sizeof(sin->sin_zero))
192 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
195 if (dir == IP_MPORT_IN && m->m_pkthdr.rcvif == NULL) {
197 * No luck with the name, check by IP address.
198 * Clear the port and the ifname to make sure
199 * there are no distractions for ifa_ifwithaddr.
201 * Be careful not to trash sin->sin_port; it will
202 * be used later in div_output().
207 bzero(sin->sin_zero, sizeof(sin->sin_zero));
208 sin_port = sin->sin_port; /* save */
210 ifa = ifa_ifwithaddr((struct sockaddr *)sin);
216 sin->sin_port = sin_port; /* restore */
217 m->m_pkthdr.rcvif = ifa->ifa_ifp;
221 * Recalculate the protocol thread.
223 ip_cpufn(mptr, 0, dir);
226 KKASSERT(m->m_flash & M_HASH);
227 return(cpu_portfn(m->m_pkthdr.hash));
233 * Divert a packet by passing it up to the divert socket at port 'port'.
235 * Setup generic address and protocol structures for div_input routine,
236 * then pass them along with mbuf chain.
239 div_packet(struct mbuf *m, int incoming, int port)
241 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
245 struct divert_info *divinfo;
248 /* Locate the divert info */
249 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
250 divinfo = m_tag_data(mtag);
251 divsrc.sin_port = divinfo->skipto;
254 * Record receive interface address, if any.
255 * But only for incoming packets.
257 divsrc.sin_addr.s_addr = 0;
259 struct ifaddr_container *ifac;
261 /* Find IP address for receive interface */
262 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
264 struct ifaddr *ifa = ifac->ifa;
266 if (ifa->ifa_addr == NULL)
268 if (ifa->ifa_addr->sa_family != AF_INET)
271 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
276 * Record the incoming interface name whenever we have one.
278 if (m->m_pkthdr.rcvif) {
280 * Hide the actual interface name in there in the
281 * sin_zero array. XXX This needs to be moved to a
282 * different sockaddr type for divert, e.g.
283 * sockaddr_div with multiple fields like
284 * sockaddr_dl. Presently we have only 7 bytes
285 * but that will do for now as most interfaces
286 * are 4 or less + 2 or less bytes for unit.
287 * There is probably a faster way of doing this,
288 * possibly taking it from the sockaddr_dl on the iface.
289 * This solves the problem of a P2P link and a LAN interface
290 * having the same address, which can result in the wrong
291 * interface being assigned to the packet when fed back
292 * into the divert socket. Theoretically if the daemon saves
293 * and re-uses the sockaddr_in as suggested in the man pages,
294 * this iface name will come along for the ride.
295 * (see div_output for the other half of this.)
297 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero,
298 m->m_pkthdr.rcvif->if_xname);
301 /* Put packet on socket queue, if any */
303 nport = htons((u_int16_t)port);
307 * Following loop to locate the inpcb is MPSAFE since the inpcb
308 * insertion/removal happens on the same CPU (CPU0), however,
309 * saving/testing the socket pointer is not MPSAFE. So we still
310 * need to hold BGL here.
312 lwkt_gettoken(&div_token);
313 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
314 if (inp->inp_flags & INP_PLACEMARKER)
316 if (inp->inp_lport == nport)
317 sa = inp->inp_socket;
320 lwkt_gettoken(&sa->so_rcv.ssb_token);
321 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
325 lwkt_reltoken(&sa->so_rcv.ssb_token);
328 ipstat.ips_noproto++;
329 ipstat.ips_delivered--;
331 lwkt_reltoken(&div_token);
336 div_packet_handler(struct netmsg *nmsg)
338 struct netmsg_packet *nmp;
339 struct lwkt_msg *msg;
341 int port, incoming = 0;
343 nmp = (struct netmsg_packet *)nmsg;
346 msg = &nmsg->nm_lmsg;
347 port = msg->u.ms_result32 & 0xffff;
348 if (msg->u.ms_result32 & DIV_INPUT)
351 div_packet(m, incoming, port);
356 divert_packet(struct mbuf *m, int incoming)
359 struct divert_info *divinfo;
365 if (m->m_len < sizeof(struct ip) &&
366 (m = m_pullup(m, sizeof(struct ip))) == NULL)
369 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
370 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
371 divinfo = m_tag_data(mtag);
373 port = divinfo->port;
374 KASSERT(port != 0, ("%s: port=0", __func__));
378 struct netmsg_packet *nmp;
379 struct lwkt_msg *msg;
381 nmp = &m->m_hdr.mh_netmsg;
382 netmsg_init(&nmp->nm_netmsg, NULL, &netisr_apanic_rport,
383 0, div_packet_handler);
386 msg = &nmp->nm_netmsg.nm_lmsg;
387 msg->u.ms_result32 = port; /* port is 16bits */
389 msg->u.ms_result32 |= DIV_INPUT;
391 msg->u.ms_result32 |= DIV_OUTPUT;
393 lwkt_sendmsg(cpu_portfn(0), &nmp->nm_netmsg.nm_lmsg);
396 div_packet(m, incoming, port);
400 * Deliver packet back into the IP processing machinery.
402 * If no address specified, or address is 0.0.0.0, send to ip_output();
403 * otherwise, send to ip_input() and mark as having been received on
404 * the interface with that address.
407 div_output(struct socket *so, struct mbuf *m,
408 struct sockaddr_in *sin, struct mbuf *control)
412 struct divert_info *divinfo;
415 m_freem(control); /* XXX */
418 * Prepare the tag for divert info. Note that a packet
419 * with a 0 tag in mh_data is effectively untagged,
420 * so we could optimize that case.
422 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
427 m_tag_prepend(m, mtag);
429 /* Loopback avoidance and state recovery */
430 divinfo = m_tag_data(mtag);
432 divinfo->skipto = sin->sin_port;
436 /* Reinject packet into the system as incoming or outgoing */
437 if (DIV_IS_OUTPUT(sin)) {
438 struct ip *const ip = mtod(m, struct ip *);
440 /* Don't allow packet length sizes that will crash */
441 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
446 /* Convert fields to host order for ip_output() */
447 ip->ip_len = ntohs(ip->ip_len);
448 ip->ip_off = ntohs(ip->ip_off);
450 /* Send packet to output processing */
451 ipstat.ips_rawout++; /* XXX */
452 error = ip_output(m, NULL, NULL,
453 (so->so_options & SO_DONTROUTE) |
454 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
467 div_attach(struct socket *so, int proto, struct pru_attach_info *ai)
475 if ((error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY)) != 0)
478 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
481 lwkt_gettoken(&div_token);
482 error = in_pcballoc(so, &divcbinfo);
484 lwkt_reltoken(&div_token);
487 inp = (struct inpcb *)so->so_pcb;
488 inp->inp_ip_p = proto;
489 inp->inp_vflag |= INP_IPV4;
490 inp->inp_flags |= INP_HDRINCL;
492 * The socket is always "connected" because
493 * we always know "where" to send the packet.
495 so->so_port = cpu0_soport(so, NULL, NULL);
496 sosetstate(so, SS_ISCONNECTED);
497 lwkt_reltoken(&div_token);
502 div_detach(struct socket *so)
514 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
515 * will sofree() it when we return.
518 div_abort(struct socket *so)
522 soisdisconnected(so);
523 error = div_detach(so);
529 div_disconnect(struct socket *so)
533 if (!(so->so_state & SS_ISCONNECTED))
536 error = div_abort(so);
543 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
548 * in_pcbbind assumes that nam is a sockaddr_in
549 * and in_pcbbind requires a valid address. Since divert
550 * sockets don't we need to make sure the address is
551 * filled in properly.
552 * XXX -- divert should not be abusing in_pcbind
553 * and should probably have its own family.
555 if (nam->sa_family != AF_INET) {
556 error = EAFNOSUPPORT;
558 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
559 error = in_pcbbind(so->so_pcb, nam, td);
565 div_shutdown(struct socket *so)
572 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
573 struct mbuf *control, struct thread *td)
575 /* Length check already done in ip_cpufn() */
576 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
579 return div_output(so, m, (struct sockaddr_in *)nam, control);
582 SYSCTL_DECL(_net_inet_divert);
583 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
584 in_pcblist_global, "S,xinpcb", "List of active divert sockets");
586 struct pr_usrreqs div_usrreqs = {
587 .pru_abort = div_abort,
588 .pru_accept = pru_accept_notsupp,
589 .pru_attach = div_attach,
590 .pru_bind = div_bind,
591 .pru_connect = pru_connect_notsupp,
592 .pru_connect2 = pru_connect2_notsupp,
593 .pru_control = in_control,
594 .pru_detach = div_detach,
595 .pru_disconnect = div_disconnect,
596 .pru_listen = pru_listen_notsupp,
597 .pru_peeraddr = in_setpeeraddr,
598 .pru_rcvd = pru_rcvd_notsupp,
599 .pru_rcvoob = pru_rcvoob_notsupp,
600 .pru_send = div_send,
601 .pru_sense = pru_sense_null,
602 .pru_shutdown = div_shutdown,
603 .pru_sockaddr = in_setsockaddr,
604 .pru_sosend = sosend,
605 .pru_soreceive = soreceive
609 ip_divert_out(struct mbuf *m, int tee)
611 struct mbuf *clone = NULL;
612 struct ip *ip = mtod(m, struct ip *);
614 /* Clone packet if we're doing a 'tee' */
616 clone = m_dup(m, MB_DONTWAIT);
620 * delayed checksums are not currently compatible
621 * with divert sockets.
623 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
625 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
628 /* Restore packet header fields to original values */
629 ip->ip_len = htons(ip->ip_len);
630 ip->ip_off = htons(ip->ip_off);
632 /* Deliver packet to divert input routine */
635 /* If 'tee', continue with original packet */
640 ip_divert_in(struct mbuf *m, int tee)
642 struct mbuf *clone = NULL;
643 struct ip *ip = mtod(m, struct ip *);
646 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
647 const struct divert_info *divinfo;
652 * Only trust divert info in the fragment
655 frag_off = ip->ip_off << 3;
657 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
658 m_tag_delete(m, mtag);
662 * Attempt reassembly; if it succeeds, proceed.
663 * ip_reass() will return a different mbuf.
668 ip = mtod(m, struct ip *);
670 /* Caller need to redispatch the packet, if it is for us */
671 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
674 * Get the header length of the reassembled
677 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
680 * Restore original checksum before diverting
684 ip->ip_len = htons(ip->ip_len);
685 ip->ip_off = htons(ip->ip_off);
687 if (hlen == sizeof(struct ip))
688 ip->ip_sum = in_cksum_hdr(ip);
690 ip->ip_sum = in_cksum(m, hlen);
691 ip->ip_off = ntohs(ip->ip_off);
692 ip->ip_len = ntohs(ip->ip_len);
695 * Only use the saved divert info
697 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
699 /* Wrongly configured ipfw */
700 kprintf("ip_input no divert info\n");
704 divinfo = m_tag_data(mtag);
709 * Divert or tee packet to the divert protocol if
713 /* Clone packet if we're doing a 'tee' */
715 clone = m_dup(m, MB_DONTWAIT);
718 * Restore packet header fields to original
721 ip->ip_len = htons(ip->ip_len);
722 ip->ip_off = htons(ip->ip_off);
724 /* Deliver packet to divert input routine */
727 /* Catch invalid reference */
731 ipstat.ips_delivered++;
733 /* If 'tee', continue with original packet */
736 * Complete processing of the packet.
737 * XXX Better safe than sorry, remove the DIVERT tag.
739 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
740 KKASSERT(mtag != NULL);
741 m_tag_delete(clone, mtag);
747 ip_divert(struct mbuf *m, int tee, int incoming)
752 ret = ip_divert_in(m, tee);
754 ret = ip_divert_out(m, tee);