xref: /dflybsd-src/sys/netinet/ip_divert.c (revision 3cf94ecf50c1958c7b10de185222fd2816ccfff0)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
20  *
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
31  * SUCH DAMAGE.
32  *
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 1.23 2005/05/08 11:01:26 joerg Exp $
35  */
36 
37 #include "opt_inet.h"
38 #include "opt_ipfw.h"
39 #include "opt_ipdivert.h"
40 #include "opt_ipsec.h"
41 
42 #ifndef INET
43 #error "IPDIVERT requires INET."
44 #endif
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/protosw.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
54 #include <sys/systm.h>
55 #include <sys/proc.h>
56 
57 #include <vm/vm_zone.h>
58 
59 #include <net/if.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68 
69 /*
70  * Divert sockets
71  */
72 
73 /*
74  * Allocate enough space to hold a full IP packet
75  */
76 #define	DIVSNDQ		(65536 + 100)
77 #define	DIVRCVQ		(65536 + 100)
78 
79 /*
80  * Divert sockets work in conjunction with ipfw, see the divert(4)
81  * manpage for features.
82  * Internally, packets selected by ipfw in ip_input() or ip_output(),
83  * and never diverted before, are passed to the input queue of the
84  * divert socket with a given 'divert_port' number (as specified in
85  * the matching ipfw rule), and they are tagged with a 16 bit cookie
86  * (representing the rule number of the matching ipfw rule), which
87  * is passed to process reading from the socket.
88  *
89  * Packets written to the divert socket are again tagged with a cookie
90  * (usually the same as above) and a destination address.
91  * If the destination address is INADDR_ANY then the packet is
92  * treated as outgoing and sent to ip_output(), otherwise it is
93  * treated as incoming and sent to ip_input().
94  * In both cases, the packet is tagged with the cookie.
95  *
96  * On reinjection, processing in ip_input() and ip_output()
97  * will be exactly the same as for the original packet, except that
98  * ipfw processing will start at the rule number after the one
99  * written in the cookie (so, tagging a packet with a cookie of 0
100  * will cause it to be effectively considered as a standard packet).
101  */
102 
103 /* Internal variables */
104 static struct inpcbinfo divcbinfo;
105 
106 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
107 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
108 
109 /*
110  * Initialize divert connection block queue.
111  */
112 void
113 div_init(void)
114 {
115 	in_pcbinfo_init(&divcbinfo);
116 	/*
117 	 * XXX We don't use the hash list for divert IP, but it's easier
118 	 * to allocate a one entry hash list than it is to check all
119 	 * over the place for hashbase == NULL.
120 	 */
121 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
122 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
123 	divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
124 					      &divcbinfo.wildcardhashmask);
125 	divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb),
126 				   maxsockets, ZONE_INTERRUPT, 0);
127 }
128 
129 /*
130  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
131  * with that protocol number to enter the system from the outside.
132  */
133 void
134 div_input(struct mbuf *m, ...)
135 {
136 	ipstat.ips_noproto++;
137 	m_freem(m);
138 }
139 
140 /*
141  * Divert a packet by passing it up to the divert socket at port 'port'.
142  *
143  * Setup generic address and protocol structures for div_input routine,
144  * then pass them along with mbuf chain.
145  */
146 void
147 divert_packet(struct mbuf *m, int incoming, int port)
148 {
149 	struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
150 	struct ip *ip;
151 	struct inpcb *inp;
152 	struct socket *sa;
153 	struct m_tag *mtag;
154 	u_int16_t nport;
155 
156 	/* Sanity check */
157 	KASSERT(port != 0, ("%s: port=0", __func__));
158 
159 	if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
160 		divsrc.sin_port = *(u_int16_t *)(mtag + 1);
161 	else
162 		divsrc.sin_port = 0;
163 
164 	/* Assure header */
165 	if (m->m_len < sizeof(struct ip) &&
166 	    (m = m_pullup(m, sizeof(struct ip))) == NULL)
167 		return;
168 	ip = mtod(m, struct ip *);
169 
170 	/*
171 	 * Record receive interface address, if any.
172 	 * But only for incoming packets.
173 	 */
174 	divsrc.sin_addr.s_addr = 0;
175 	if (incoming) {
176 		struct ifaddr *ifa;
177 
178 		/* Sanity check */
179 		KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __func__));
180 
181 		/* Find IP address for receive interface */
182 		TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
183 			if (ifa->ifa_addr == NULL)
184 				continue;
185 			if (ifa->ifa_addr->sa_family != AF_INET)
186 				continue;
187 			divsrc.sin_addr =
188 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
189 			break;
190 		}
191 	}
192 	/*
193 	 * Record the incoming interface name whenever we have one.
194 	 */
195 	if (m->m_pkthdr.rcvif) {
196 		/*
197 		 * Hide the actual interface name in there in the
198 		 * sin_zero array. XXX This needs to be moved to a
199 		 * different sockaddr type for divert, e.g.
200 		 * sockaddr_div with multiple fields like
201 		 * sockaddr_dl. Presently we have only 7 bytes
202 		 * but that will do for now as most interfaces
203 		 * are 4 or less + 2 or less bytes for unit.
204 		 * There is probably a faster way of doing this,
205 		 * possibly taking it from the sockaddr_dl on the iface.
206 		 * This solves the problem of a P2P link and a LAN interface
207 		 * having the same address, which can result in the wrong
208 		 * interface being assigned to the packet when fed back
209 		 * into the divert socket. Theoretically if the daemon saves
210 		 * and re-uses the sockaddr_in as suggested in the man pages,
211 		 * this iface name will come along for the ride.
212 		 * (see div_output for the other half of this.)
213 		 */
214 		snprintf(divsrc.sin_zero, sizeof divsrc.sin_zero,
215 			 m->m_pkthdr.rcvif->if_xname);
216 	}
217 
218 	/* Put packet on socket queue, if any */
219 	sa = NULL;
220 	nport = htons((u_int16_t)port);
221 	LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
222 		if (inp->inp_flags & INP_PLACEMARKER)
223 			continue;
224 		if (inp->inp_lport == nport)
225 			sa = inp->inp_socket;
226 	}
227 	if (sa) {
228 		if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m,
229 				 (struct mbuf *)NULL) == 0)
230 			m_freem(m);
231 		else
232 			sorwakeup(sa);
233 	} else {
234 		m_freem(m);
235 		ipstat.ips_noproto++;
236 		ipstat.ips_delivered--;
237 	}
238 }
239 
240 /*
241  * Deliver packet back into the IP processing machinery.
242  *
243  * If no address specified, or address is 0.0.0.0, send to ip_output();
244  * otherwise, send to ip_input() and mark as having been received on
245  * the interface with that address.
246  */
247 static int
248 div_output(struct socket *so, struct mbuf *m,
249 	struct sockaddr_in *sin, struct mbuf *control)
250 {
251 	int error = 0;
252 	struct m_tag *mtag;
253 
254 	/*
255 	 * Prepare the tag for divert info. Note that a packet
256 	 * with a 0 tag in mh_data is effectively untagged,
257 	 * so we could optimize that case.
258 	 */
259 	mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(u_int16_t), M_NOWAIT);
260 	if (mtag == NULL) {
261 		error = ENOBUFS;
262 		goto cantsend;
263 	}
264 	m_tag_prepend(m, mtag);
265 	m->m_pkthdr.rcvif = NULL;	/* XXX is it necessary ? */
266 
267 	if (control)
268 		m_freem(control);		/* XXX */
269 
270 	/* Loopback avoidance and state recovery */
271 	if (sin) {
272 		int i;
273 
274 		*(u_int16_t *)mtag = sin->sin_port;
275 		/*
276 		 * Find receive interface with the given name, stuffed
277 		 * (if it exists) in the sin_zero[] field.
278 		 * The name is user supplied data so don't trust its size
279 		 * or that it is zero terminated.
280 		 */
281 		for (i = 0; sin->sin_zero[i] && i < sizeof sin->sin_zero; i++)
282 			;
283 		if ( i > 0 && i < sizeof sin->sin_zero)
284 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
285 	}
286 
287 	/* Reinject packet into the system as incoming or outgoing */
288 	if (!sin || sin->sin_addr.s_addr == 0) {
289 		struct inpcb *const inp = so->so_pcb;
290 		struct ip *const ip = mtod(m, struct ip *);
291 
292 		/*
293 		 * Don't allow both user specified and setsockopt options,
294 		 * and don't allow packet length sizes that will crash
295 		 */
296 		if (((ip->ip_hl != (sizeof *ip) >> 2) && inp->inp_options) ||
297 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
298 			error = EINVAL;
299 			goto cantsend;
300 		}
301 
302 		/* Convert fields to host order for ip_output() */
303 		ip->ip_len = ntohs(ip->ip_len);
304 		ip->ip_off = ntohs(ip->ip_off);
305 
306 		/* Send packet to output processing */
307 		ipstat.ips_rawout++;			/* XXX */
308 		error = ip_output(m,
309 			    inp->inp_options, &inp->inp_route,
310 			    (so->so_options & SO_DONTROUTE) |
311 			    IP_ALLOWBROADCAST | IP_RAWOUTPUT,
312 			    inp->inp_moptions, NULL);
313 	} else {
314 		if (m->m_pkthdr.rcvif == NULL) {
315 			/*
316 			 * No luck with the name, check by IP address.
317 			 * Clear the port and the ifname to make sure
318 			 * there are no distractions for ifa_ifwithaddr.
319 			 */
320 			struct	ifaddr *ifa;
321 
322 			bzero(sin->sin_zero, sizeof sin->sin_zero);
323 			sin->sin_port = 0;
324 			ifa = ifa_ifwithaddr((struct sockaddr *) sin);
325 			if (ifa == NULL) {
326 				error = EADDRNOTAVAIL;
327 				goto cantsend;
328 			}
329 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
330 		}
331 		ip_input(m);
332 	}
333 
334 	return error;
335 
336 cantsend:
337 	m_freem(m);
338 	return error;
339 }
340 
341 static int
342 div_attach(struct socket *so, int proto, struct pru_attach_info *ai)
343 {
344 	struct inpcb *inp;
345 	int error, s;
346 
347 	inp  = so->so_pcb;
348 	if (inp)
349 		panic("div_attach");
350 	if ((error = suser_cred(ai->p_ucred, NULL_CRED_OKAY)) != 0)
351 		return error;
352 
353 	error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
354 	if (error)
355 		return error;
356 	s = splnet();
357 	error = in_pcballoc(so, &divcbinfo);
358 	splx(s);
359 	if (error)
360 		return error;
361 	inp = (struct inpcb *)so->so_pcb;
362 	inp->inp_ip_p = proto;
363 	inp->inp_vflag |= INP_IPV4;
364 	inp->inp_flags |= INP_HDRINCL;
365 	/*
366 	 * The socket is always "connected" because
367 	 * we always know "where" to send the packet.
368 	 */
369 	so->so_state |= SS_ISCONNECTED;
370 	return 0;
371 }
372 
373 static int
374 div_detach(struct socket *so)
375 {
376 	struct inpcb *inp;
377 
378 	inp = so->so_pcb;
379 	if (inp == NULL)
380 		panic("div_detach");
381 	in_pcbdetach(inp);
382 	return 0;
383 }
384 
385 static int
386 div_abort(struct socket *so)
387 {
388 	soisdisconnected(so);
389 	return div_detach(so);
390 }
391 
392 static int
393 div_disconnect(struct socket *so)
394 {
395 	if (!(so->so_state & SS_ISCONNECTED))
396 		return ENOTCONN;
397 	return div_abort(so);
398 }
399 
400 static int
401 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
402 {
403 	struct inpcb *inp;
404 	int s;
405 	int error;
406 
407 	s = splnet();
408 	inp = so->so_pcb;
409 	/* in_pcbbind assumes that nam is a sockaddr_in
410 	 * and in_pcbbind requires a valid address. Since divert
411 	 * sockets don't we need to make sure the address is
412 	 * filled in properly.
413 	 * XXX -- divert should not be abusing in_pcbind
414 	 * and should probably have its own family.
415 	 */
416 	if (nam->sa_family != AF_INET)
417 		error = EAFNOSUPPORT;
418 	else {
419 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
420 		error = in_pcbbind(inp, nam, td);
421 	}
422 	splx(s);
423 	return error;
424 }
425 
426 static int
427 div_shutdown(struct socket *so)
428 {
429 	socantsendmore(so);
430 	return 0;
431 }
432 
433 static int
434 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
435 	 struct mbuf *control, struct thread *td)
436 {
437 	/* Packet must have a header (but that's about it) */
438 	if (m->m_len < sizeof(struct ip) &&
439 	    (m = m_pullup(m, sizeof(struct ip))) == NULL) {
440 		ipstat.ips_toosmall++;
441 		m_freem(m);
442 		return EINVAL;
443 	}
444 
445 	/* Send packet */
446 	return div_output(so, m, (struct sockaddr_in *)nam, control);
447 }
448 
449 SYSCTL_DECL(_net_inet_divert);
450 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
451 	    in_pcblist_global, "S,xinpcb", "List of active divert sockets");
452 
453 struct pr_usrreqs div_usrreqs = {
454 	div_abort, pru_accept_notsupp, div_attach, div_bind,
455 	pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
456 	div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp,
457 	pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
458 	in_setsockaddr, sosend, soreceive, sopoll
459 };
460