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. 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
29 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
35 #include "opt_ipdivert.h"
36 #include "opt_ipsec.h"
39 #error "IPDIVERT requires INET."
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/socket.h>
47 #include <sys/protosw.h>
48 #include <sys/socketvar.h>
49 #include <sys/socketvar2.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
54 #include <sys/in_cksum.h>
56 #include <sys/msgport.h>
59 #include <net/route.h>
61 #include <net/netmsg2.h>
62 #include <net/netisr2.h>
63 #include <sys/thread2.h>
64 #include <sys/mplock2.h>
66 #include <netinet/in.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_divert.h>
79 * Allocate enough space to hold a full IP packet
81 #define DIVSNDQ (65536 + 100)
82 #define DIVRCVQ (65536 + 100)
84 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
86 #define DIV_OUTPUT 0x10000
87 #define DIV_INPUT 0x20000
90 * Divert sockets work in conjunction with ipfw, see the divert(4)
91 * manpage for features.
92 * Internally, packets selected by ipfw in ip_input() or ip_output(),
93 * and never diverted before, are passed to the input queue of the
94 * divert socket with a given 'divert_port' number (as specified in
95 * the matching ipfw rule), and they are tagged with a 16 bit cookie
96 * (representing the rule number of the matching ipfw rule), which
97 * is passed to process reading from the socket.
99 * Packets written to the divert socket are again tagged with a cookie
100 * (usually the same as above) and a destination address.
101 * If the destination address is INADDR_ANY then the packet is
102 * treated as outgoing and sent to ip_output(), otherwise it is
103 * treated as incoming and sent to ip_input().
104 * In both cases, the packet is tagged with the cookie.
106 * On reinjection, processing in ip_input() and ip_output()
107 * will be exactly the same as for the original packet, except that
108 * ipfw processing will start at the rule number after the one
109 * written in the cookie (so, tagging a packet with a cookie of 0
110 * will cause it to be effectively considered as a standard packet).
113 /* Internal variables */
114 static struct inpcbinfo divcbinfo;
115 static struct inpcbportinfo divcbportinfo;
117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
120 static struct mbuf *ip_divert(struct mbuf *, int, int);
122 static struct lwkt_token div_token = LWKT_TOKEN_INITIALIZER(div_token);
125 * Initialize divert connection block queue.
130 in_pcbinfo_init(&divcbinfo);
131 in_pcbportinfo_init(&divcbportinfo, 1, FALSE, 0);
133 * XXX We don't use the hash list for divert IP, but it's easier
134 * to allocate a one entry hash list than it is to check all
135 * over the place for hashbase == NULL.
137 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
138 divcbinfo.portinfo = &divcbportinfo;
139 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
140 &divcbinfo.wildcardhashmask);
141 divcbinfo.ipi_size = sizeof(struct inpcb);
142 ip_divert_p = ip_divert;
146 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
147 * with that protocol number to enter the system from the outside.
150 div_input(struct mbuf **mp, int *offp, int proto)
152 struct mbuf *m = *mp;
154 ipstat.ips_noproto++;
156 return(IPPROTO_DONE);
160 * Divert a packet by passing it up to the divert socket at port 'port'.
162 * Setup generic address and protocol structures for div_input routine,
163 * then pass them along with mbuf chain.
166 div_packet(struct mbuf *m, int incoming, int port)
168 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
172 struct divert_info *divinfo;
175 /* Locate the divert info */
176 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
177 divinfo = m_tag_data(mtag);
178 divsrc.sin_port = divinfo->skipto;
181 * Record receive interface address, if any.
182 * But only for incoming packets.
184 divsrc.sin_addr.s_addr = 0;
186 struct ifaddr_container *ifac;
188 /* Find IP address for receive interface */
189 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
191 struct ifaddr *ifa = ifac->ifa;
193 if (ifa->ifa_addr == NULL)
195 if (ifa->ifa_addr->sa_family != AF_INET)
198 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
203 * Record the incoming interface name whenever we have one.
205 if (m->m_pkthdr.rcvif) {
207 * Hide the actual interface name in there in the
208 * sin_zero array. XXX This needs to be moved to a
209 * different sockaddr type for divert, e.g.
210 * sockaddr_div with multiple fields like
211 * sockaddr_dl. Presently we have only 7 bytes
212 * but that will do for now as most interfaces
213 * are 4 or less + 2 or less bytes for unit.
214 * There is probably a faster way of doing this,
215 * possibly taking it from the sockaddr_dl on the iface.
216 * This solves the problem of a P2P link and a LAN interface
217 * having the same address, which can result in the wrong
218 * interface being assigned to the packet when fed back
219 * into the divert socket. Theoretically if the daemon saves
220 * and re-uses the sockaddr_in as suggested in the man pages,
221 * this iface name will come along for the ride.
222 * (see div_output for the other half of this.)
224 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s",
225 m->m_pkthdr.rcvif->if_xname);
228 /* Put packet on socket queue, if any */
230 nport = htons((u_int16_t)port);
234 * Following loop to locate the inpcb is MPSAFE since the inpcb
235 * insertion/removal happens on the same CPU (CPU0), however,
236 * saving/testing the socket pointer is not MPSAFE. So we still
237 * need to hold BGL here.
239 lwkt_gettoken(&div_token);
240 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
241 if (inp->inp_flags & INP_PLACEMARKER)
243 if (inp->inp_lport == nport)
244 sa = inp->inp_socket;
247 lwkt_gettoken(&sa->so_rcv.ssb_token);
248 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
252 lwkt_reltoken(&sa->so_rcv.ssb_token);
255 ipstat.ips_noproto++;
256 ipstat.ips_delivered--;
258 lwkt_reltoken(&div_token);
262 div_packet_handler(netmsg_t msg)
265 int port, incoming = 0;
267 m = msg->packet.nm_packet;
269 port = msg->lmsg.u.ms_result32 & 0xffff;
270 if (msg->lmsg.u.ms_result32 & DIV_INPUT)
272 div_packet(m, incoming, port);
273 /* no reply, msg embedded in mbuf */
277 divert_packet(struct mbuf *m, int incoming)
280 struct divert_info *divinfo;
286 if (m->m_len < sizeof(struct ip) &&
287 (m = m_pullup(m, sizeof(struct ip))) == NULL)
290 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
291 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
292 divinfo = m_tag_data(mtag);
294 port = divinfo->port;
295 KASSERT(port != 0, ("%s: port=0", __func__));
298 struct netmsg_packet *nmp;
300 nmp = &m->m_hdr.mh_netmsg;
301 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
302 0, div_packet_handler);
305 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
307 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
309 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
311 lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg);
313 div_packet(m, incoming, port);
318 * Deliver packet back into the IP processing machinery.
320 * If no address specified, or address is 0.0.0.0, send to ip_output();
321 * otherwise, send to ip_input() and mark as having been received on
322 * the interface with that address.
325 div_output(struct socket *so, struct mbuf *m,
326 struct sockaddr_in *sin, struct mbuf *control)
330 struct divert_info *divinfo;
333 m_freem(control); /* XXX */
336 * Prepare the tag for divert info. Note that a packet
337 * with a 0 tag in mh_data is effectively untagged,
338 * so we could optimize that case.
340 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
345 m_tag_prepend(m, mtag);
347 /* Loopback avoidance and state recovery */
348 divinfo = m_tag_data(mtag);
350 divinfo->skipto = sin->sin_port;
354 /* Reinject packet into the system as incoming or outgoing */
355 if (DIV_IS_OUTPUT(sin)) {
356 struct ip *const ip = mtod(m, struct ip *);
358 /* Don't allow packet length sizes that will crash */
359 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
364 /* Convert fields to host order for ip_output() */
365 ip->ip_len = ntohs(ip->ip_len);
366 ip->ip_off = ntohs(ip->ip_off);
368 /* Send packet to output processing */
369 ipstat.ips_rawout++; /* XXX */
370 error = ip_output(m, NULL, NULL,
371 (so->so_options & SO_DONTROUTE) |
372 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
385 div_attach(netmsg_t msg)
387 struct socket *so = msg->attach.base.nm_so;
388 int proto = msg->attach.nm_proto;
389 struct pru_attach_info *ai = msg->attach.nm_ai;
396 error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
400 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
403 lwkt_gettoken(&div_token);
404 sosetport(so, netisr_cpuport(0));
405 error = in_pcballoc(so, &divcbinfo);
407 lwkt_reltoken(&div_token);
410 inp = (struct inpcb *)so->so_pcb;
411 inp->inp_ip_p = proto;
412 inp->inp_vflag |= INP_IPV4;
413 inp->inp_flags |= INP_HDRINCL;
415 * The socket is always "connected" because
416 * we always know "where" to send the packet.
418 sosetstate(so, SS_ISCONNECTED);
419 lwkt_reltoken(&div_token);
422 lwkt_replymsg(&msg->attach.base.lmsg, error);
426 div_detach(netmsg_t msg)
428 struct socket *so = msg->detach.base.nm_so;
435 lwkt_replymsg(&msg->detach.base.lmsg, 0);
439 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
440 * will sofree() it when we return.
443 div_abort(netmsg_t msg)
445 struct socket *so = msg->abort.base.nm_so;
447 soisdisconnected(so);
449 /* msg invalid now */
453 div_disconnect(netmsg_t msg)
455 struct socket *so = msg->disconnect.base.nm_so;
458 if (so->so_state & SS_ISCONNECTED) {
461 /* msg invalid now */
466 lwkt_replymsg(&msg->disconnect.base.lmsg, error);
470 div_bind(netmsg_t msg)
472 struct socket *so = msg->bind.base.nm_so;
473 struct sockaddr *nam = msg->bind.nm_nam;
477 * in_pcbbind assumes that nam is a sockaddr_in
478 * and in_pcbbind requires a valid address. Since divert
479 * sockets don't we need to make sure the address is
480 * filled in properly.
481 * XXX -- divert should not be abusing in_pcbind
482 * and should probably have its own family.
484 if (nam->sa_family != AF_INET) {
485 error = EAFNOSUPPORT;
487 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
488 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
490 lwkt_replymsg(&msg->bind.base.lmsg, error);
494 div_shutdown(netmsg_t msg)
496 struct socket *so = msg->shutdown.base.nm_so;
500 lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
504 div_send(netmsg_t msg)
506 struct socket *so = msg->send.base.nm_so;
507 struct mbuf *m = msg->send.nm_m;
508 struct sockaddr *nam = msg->send.nm_addr;
509 struct mbuf *control = msg->send.nm_control;
512 /* Length check already done in ip_hashfn() */
513 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
516 error = div_output(so, m, (struct sockaddr_in *)nam, control);
517 lwkt_replymsg(&msg->send.base.lmsg, error);
520 SYSCTL_DECL(_net_inet_divert);
521 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
522 in_pcblist_global_cpu0, "S,xinpcb", "List of active divert sockets");
524 struct pr_usrreqs div_usrreqs = {
525 .pru_abort = div_abort,
526 .pru_accept = pr_generic_notsupp,
527 .pru_attach = div_attach,
528 .pru_bind = div_bind,
529 .pru_connect = pr_generic_notsupp,
530 .pru_connect2 = pr_generic_notsupp,
531 .pru_control = in_control_dispatch,
532 .pru_detach = div_detach,
533 .pru_disconnect = div_disconnect,
534 .pru_listen = pr_generic_notsupp,
535 .pru_peeraddr = in_setpeeraddr_dispatch,
536 .pru_rcvd = pr_generic_notsupp,
537 .pru_rcvoob = pr_generic_notsupp,
538 .pru_send = div_send,
539 .pru_sense = pru_sense_null,
540 .pru_shutdown = div_shutdown,
541 .pru_sockaddr = in_setsockaddr_dispatch,
542 .pru_sosend = sosend,
543 .pru_soreceive = soreceive
547 ip_divert_out(struct mbuf *m, int tee)
549 struct mbuf *clone = NULL;
550 struct ip *ip = mtod(m, struct ip *);
552 /* Clone packet if we're doing a 'tee' */
554 clone = m_dup(m, MB_DONTWAIT);
558 * delayed checksums are not currently compatible
559 * with divert sockets.
561 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
563 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
566 /* Restore packet header fields to original values */
567 ip->ip_len = htons(ip->ip_len);
568 ip->ip_off = htons(ip->ip_off);
570 /* Deliver packet to divert input routine */
573 /* If 'tee', continue with original packet */
578 ip_divert_in(struct mbuf *m, int tee)
580 struct mbuf *clone = NULL;
581 struct ip *ip = mtod(m, struct ip *);
584 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
585 const struct divert_info *divinfo;
590 * Only trust divert info in the fragment
593 frag_off = ip->ip_off << 3;
595 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
596 m_tag_delete(m, mtag);
600 * Attempt reassembly; if it succeeds, proceed.
601 * ip_reass() will return a different mbuf.
606 ip = mtod(m, struct ip *);
608 /* Caller need to redispatch the packet, if it is for us */
609 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
612 * Get the header length of the reassembled
615 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
618 * Restore original checksum before diverting
622 ip->ip_len = htons(ip->ip_len);
623 ip->ip_off = htons(ip->ip_off);
625 if (hlen == sizeof(struct ip))
626 ip->ip_sum = in_cksum_hdr(ip);
628 ip->ip_sum = in_cksum(m, hlen);
629 ip->ip_off = ntohs(ip->ip_off);
630 ip->ip_len = ntohs(ip->ip_len);
633 * Only use the saved divert info
635 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
637 /* Wrongly configured ipfw */
638 kprintf("ip_input no divert info\n");
642 divinfo = m_tag_data(mtag);
647 * Divert or tee packet to the divert protocol if
651 /* Clone packet if we're doing a 'tee' */
653 clone = m_dup(m, MB_DONTWAIT);
656 * Restore packet header fields to original
659 ip->ip_len = htons(ip->ip_len);
660 ip->ip_off = htons(ip->ip_off);
662 /* Deliver packet to divert input routine */
665 /* Catch invalid reference */
669 ipstat.ips_delivered++;
671 /* If 'tee', continue with original packet */
674 * Complete processing of the packet.
675 * XXX Better safe than sorry, remove the DIVERT tag.
677 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
678 KKASSERT(mtag != NULL);
679 m_tag_delete(clone, mtag);
685 ip_divert(struct mbuf *m, int tee, int incoming)
690 ret = ip_divert_in(m, tee);
692 ret = ip_divert_out(m, tee);