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/socketvar2.h>
56 #include <sys/sysctl.h>
57 #include <sys/systm.h>
60 #include <sys/in_cksum.h>
62 #include <sys/msgport.h>
65 #include <net/route.h>
67 #include <net/netmsg2.h>
68 #include <sys/thread2.h>
69 #include <sys/mplock2.h>
71 #include <netinet/in.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/ip.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_divert.h>
84 * Allocate enough space to hold a full IP packet
86 #define DIVSNDQ (65536 + 100)
87 #define DIVRCVQ (65536 + 100)
89 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
91 #define DIV_OUTPUT 0x10000
92 #define DIV_INPUT 0x20000
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 struct inpcbinfo divcbinfo;
121 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
122 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
124 static struct mbuf *ip_divert(struct mbuf *, int, int);
126 static struct lwkt_token div_token = LWKT_TOKEN_INITIALIZER(div_token);
129 * Initialize divert connection block queue.
134 in_pcbinfo_init(&divcbinfo);
136 * XXX We don't use the hash list for divert IP, but it's easier
137 * to allocate a one entry hash list than it is to check all
138 * over the place for hashbase == NULL.
140 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
141 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
142 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
143 &divcbinfo.wildcardhashmask);
144 divcbinfo.ipi_size = sizeof(struct inpcb);
145 ip_divert_p = ip_divert;
149 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
150 * with that protocol number to enter the system from the outside.
153 div_input(struct mbuf **mp, int *offp, int proto)
155 struct mbuf *m = *mp;
157 ipstat.ips_noproto++;
159 return(IPPROTO_DONE);
163 * Divert a packet by passing it up to the divert socket at port 'port'.
165 * Setup generic address and protocol structures for div_input routine,
166 * then pass them along with mbuf chain.
169 div_packet(struct mbuf *m, int incoming, int port)
171 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
175 struct divert_info *divinfo;
178 /* Locate the divert info */
179 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
180 divinfo = m_tag_data(mtag);
181 divsrc.sin_port = divinfo->skipto;
184 * Record receive interface address, if any.
185 * But only for incoming packets.
187 divsrc.sin_addr.s_addr = 0;
189 struct ifaddr_container *ifac;
191 /* Find IP address for receive interface */
192 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
194 struct ifaddr *ifa = ifac->ifa;
196 if (ifa->ifa_addr == NULL)
198 if (ifa->ifa_addr->sa_family != AF_INET)
201 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
206 * Record the incoming interface name whenever we have one.
208 if (m->m_pkthdr.rcvif) {
210 * Hide the actual interface name in there in the
211 * sin_zero array. XXX This needs to be moved to a
212 * different sockaddr type for divert, e.g.
213 * sockaddr_div with multiple fields like
214 * sockaddr_dl. Presently we have only 7 bytes
215 * but that will do for now as most interfaces
216 * are 4 or less + 2 or less bytes for unit.
217 * There is probably a faster way of doing this,
218 * possibly taking it from the sockaddr_dl on the iface.
219 * This solves the problem of a P2P link and a LAN interface
220 * having the same address, which can result in the wrong
221 * interface being assigned to the packet when fed back
222 * into the divert socket. Theoretically if the daemon saves
223 * and re-uses the sockaddr_in as suggested in the man pages,
224 * this iface name will come along for the ride.
225 * (see div_output for the other half of this.)
227 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero,
228 m->m_pkthdr.rcvif->if_xname);
231 /* Put packet on socket queue, if any */
233 nport = htons((u_int16_t)port);
237 * Following loop to locate the inpcb is MPSAFE since the inpcb
238 * insertion/removal happens on the same CPU (CPU0), however,
239 * saving/testing the socket pointer is not MPSAFE. So we still
240 * need to hold BGL here.
242 lwkt_gettoken(&div_token);
243 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
244 if (inp->inp_flags & INP_PLACEMARKER)
246 if (inp->inp_lport == nport)
247 sa = inp->inp_socket;
250 lwkt_gettoken(&sa->so_rcv.ssb_token);
251 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
255 lwkt_reltoken(&sa->so_rcv.ssb_token);
258 ipstat.ips_noproto++;
259 ipstat.ips_delivered--;
261 lwkt_reltoken(&div_token);
267 div_packet_handler(netmsg_t msg)
270 int port, incoming = 0;
272 m = msg->packet.nm_packet;
274 port = msg->lmsg.u.ms_result32 & 0xffff;
275 if (msg->lmsg.u.ms_result32 & DIV_INPUT)
277 div_packet(m, incoming, port);
278 /* no reply, msg embedded in mbuf */
284 divert_packet(struct mbuf *m, int incoming)
287 struct divert_info *divinfo;
293 if (m->m_len < sizeof(struct ip) &&
294 (m = m_pullup(m, sizeof(struct ip))) == NULL)
297 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
298 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
299 divinfo = m_tag_data(mtag);
301 port = divinfo->port;
302 KASSERT(port != 0, ("%s: port=0", __func__));
306 struct netmsg_packet *nmp;
308 nmp = &m->m_hdr.mh_netmsg;
309 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
310 0, div_packet_handler);
313 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
315 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
317 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
319 lwkt_sendmsg(cpu_portfn(0), &nmp->base.lmsg);
321 div_packet(m, incoming, port);
324 div_packet(m, incoming, port);
329 * Deliver packet back into the IP processing machinery.
331 * If no address specified, or address is 0.0.0.0, send to ip_output();
332 * otherwise, send to ip_input() and mark as having been received on
333 * the interface with that address.
336 div_output(struct socket *so, struct mbuf *m,
337 struct sockaddr_in *sin, struct mbuf *control)
341 struct divert_info *divinfo;
344 m_freem(control); /* XXX */
347 * Prepare the tag for divert info. Note that a packet
348 * with a 0 tag in mh_data is effectively untagged,
349 * so we could optimize that case.
351 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
356 m_tag_prepend(m, mtag);
358 /* Loopback avoidance and state recovery */
359 divinfo = m_tag_data(mtag);
361 divinfo->skipto = sin->sin_port;
365 /* Reinject packet into the system as incoming or outgoing */
366 if (DIV_IS_OUTPUT(sin)) {
367 struct ip *const ip = mtod(m, struct ip *);
369 /* Don't allow packet length sizes that will crash */
370 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
375 /* Convert fields to host order for ip_output() */
376 ip->ip_len = ntohs(ip->ip_len);
377 ip->ip_off = ntohs(ip->ip_off);
379 /* Send packet to output processing */
380 ipstat.ips_rawout++; /* XXX */
381 error = ip_output(m, NULL, NULL,
382 (so->so_options & SO_DONTROUTE) |
383 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
396 div_attach(netmsg_t msg)
398 struct socket *so = msg->attach.base.nm_so;
399 int proto = msg->attach.nm_proto;
400 struct pru_attach_info *ai = msg->attach.nm_ai;
407 error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
411 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
414 lwkt_gettoken(&div_token);
415 sosetport(so, cpu_portfn(0));
416 error = in_pcballoc(so, &divcbinfo);
418 lwkt_reltoken(&div_token);
421 inp = (struct inpcb *)so->so_pcb;
422 inp->inp_ip_p = proto;
423 inp->inp_vflag |= INP_IPV4;
424 inp->inp_flags |= INP_HDRINCL;
426 * The socket is always "connected" because
427 * we always know "where" to send the packet.
429 sosetstate(so, SS_ISCONNECTED);
430 lwkt_reltoken(&div_token);
433 lwkt_replymsg(&msg->attach.base.lmsg, error);
437 div_detach(netmsg_t msg)
439 struct socket *so = msg->detach.base.nm_so;
446 lwkt_replymsg(&msg->detach.base.lmsg, 0);
450 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
451 * will sofree() it when we return.
454 div_abort(netmsg_t msg)
456 struct socket *so = msg->abort.base.nm_so;
458 soisdisconnected(so);
460 /* msg invalid now */
464 div_disconnect(netmsg_t msg)
466 struct socket *so = msg->disconnect.base.nm_so;
469 if (so->so_state & SS_ISCONNECTED) {
472 /* msg invalid now */
477 lwkt_replymsg(&msg->disconnect.base.lmsg, error);
481 div_bind(netmsg_t msg)
483 struct socket *so = msg->bind.base.nm_so;
484 struct sockaddr *nam = msg->bind.nm_nam;
488 * in_pcbbind assumes that nam is a sockaddr_in
489 * and in_pcbbind requires a valid address. Since divert
490 * sockets don't we need to make sure the address is
491 * filled in properly.
492 * XXX -- divert should not be abusing in_pcbind
493 * and should probably have its own family.
495 if (nam->sa_family != AF_INET) {
496 error = EAFNOSUPPORT;
498 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
499 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
501 lwkt_replymsg(&msg->bind.base.lmsg, error);
505 div_shutdown(netmsg_t msg)
507 struct socket *so = msg->shutdown.base.nm_so;
511 lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
515 div_send(netmsg_t msg)
517 struct socket *so = msg->send.base.nm_so;
518 struct mbuf *m = msg->send.nm_m;
519 struct sockaddr *nam = msg->send.nm_addr;
520 struct mbuf *control = msg->send.nm_control;
523 /* Length check already done in ip_cpufn() */
524 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
527 error = div_output(so, m, (struct sockaddr_in *)nam, control);
528 lwkt_replymsg(&msg->send.base.lmsg, error);
531 SYSCTL_DECL(_net_inet_divert);
532 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
533 in_pcblist_global, "S,xinpcb", "List of active divert sockets");
535 struct pr_usrreqs div_usrreqs = {
536 .pru_abort = div_abort,
537 .pru_accept = pr_generic_notsupp,
538 .pru_attach = div_attach,
539 .pru_bind = div_bind,
540 .pru_connect = pr_generic_notsupp,
541 .pru_connect2 = pr_generic_notsupp,
542 .pru_control = in_control_dispatch,
543 .pru_detach = div_detach,
544 .pru_disconnect = div_disconnect,
545 .pru_listen = pr_generic_notsupp,
546 .pru_peeraddr = in_setpeeraddr_dispatch,
547 .pru_rcvd = pr_generic_notsupp,
548 .pru_rcvoob = pr_generic_notsupp,
549 .pru_send = div_send,
550 .pru_sense = pru_sense_null,
551 .pru_shutdown = div_shutdown,
552 .pru_sockaddr = in_setsockaddr_dispatch,
553 .pru_sosend = sosend,
554 .pru_soreceive = soreceive
558 ip_divert_out(struct mbuf *m, int tee)
560 struct mbuf *clone = NULL;
561 struct ip *ip = mtod(m, struct ip *);
563 /* Clone packet if we're doing a 'tee' */
565 clone = m_dup(m, MB_DONTWAIT);
569 * delayed checksums are not currently compatible
570 * with divert sockets.
572 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
574 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
577 /* Restore packet header fields to original values */
578 ip->ip_len = htons(ip->ip_len);
579 ip->ip_off = htons(ip->ip_off);
581 /* Deliver packet to divert input routine */
584 /* If 'tee', continue with original packet */
589 ip_divert_in(struct mbuf *m, int tee)
591 struct mbuf *clone = NULL;
592 struct ip *ip = mtod(m, struct ip *);
595 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
596 const struct divert_info *divinfo;
601 * Only trust divert info in the fragment
604 frag_off = ip->ip_off << 3;
606 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
607 m_tag_delete(m, mtag);
611 * Attempt reassembly; if it succeeds, proceed.
612 * ip_reass() will return a different mbuf.
617 ip = mtod(m, struct ip *);
619 /* Caller need to redispatch the packet, if it is for us */
620 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
623 * Get the header length of the reassembled
626 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
629 * Restore original checksum before diverting
633 ip->ip_len = htons(ip->ip_len);
634 ip->ip_off = htons(ip->ip_off);
636 if (hlen == sizeof(struct ip))
637 ip->ip_sum = in_cksum_hdr(ip);
639 ip->ip_sum = in_cksum(m, hlen);
640 ip->ip_off = ntohs(ip->ip_off);
641 ip->ip_len = ntohs(ip->ip_len);
644 * Only use the saved divert info
646 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
648 /* Wrongly configured ipfw */
649 kprintf("ip_input no divert info\n");
653 divinfo = m_tag_data(mtag);
658 * Divert or tee packet to the divert protocol if
662 /* Clone packet if we're doing a 'tee' */
664 clone = m_dup(m, MB_DONTWAIT);
667 * Restore packet header fields to original
670 ip->ip_len = htons(ip->ip_len);
671 ip->ip_off = htons(ip->ip_off);
673 /* Deliver packet to divert input routine */
676 /* Catch invalid reference */
680 ipstat.ips_delivered++;
682 /* If 'tee', continue with original packet */
685 * Complete processing of the packet.
686 * XXX Better safe than sorry, remove the DIVERT tag.
688 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
689 KKASSERT(mtag != NULL);
690 m_tag_delete(clone, mtag);
696 ip_divert(struct mbuf *m, int tee, int incoming)
701 ret = ip_divert_in(m, tee);
703 ret = ip_divert_out(m, tee);