xref: /freebsd/sys/netinet/ip_divert.c (revision 817420dc8eac7df799c78f5309b75092b7f7cd40)
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  * $FreeBSD$
34  */
35 
36 #include "opt_inet.h"
37 #include "opt_ipfw.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipsec.h"
40 
41 #ifndef INET
42 #error "IPDIVERT requires INET."
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/protosw.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/systm.h>
54 
55 #include <vm/vm_zone.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/ip.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 
67 /*
68  * Divert sockets
69  */
70 
71 /*
72  * Allocate enough space to hold a full IP packet
73  */
74 #define	DIVSNDQ		(65536 + 100)
75 #define	DIVRCVQ		(65536 + 100)
76 
77 /*
78  * A 16 bit cookie is passed to and from the user process.
79  * The user process can send it back to help the caller know
80  * something about where the packet originally came from.
81  *
82  * In the case of ipfw, then the cookie is the rule that sent
83  * us here. On reinjection is is the rule after which processing
84  * should continue. Leaving it the same will make processing start
85  * at the rule number after that which sent it here. Setting it to
86  * 0 will restart processing at the beginning.
87  *
88  * For divert_packet(), ip_divert_cookie is an input value only.
89  * For div_output(), ip_divert_cookie is an output value only.
90  */
91 u_int16_t ip_divert_cookie;
92 
93 /* Internal variables */
94 static struct inpcbhead divcb;
95 static struct inpcbinfo divcbinfo;
96 
97 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
98 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
99 
100 /* Optimization: have this preinitialized */
101 static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET };
102 
103 /* Internal functions */
104 static int div_output(struct socket *so,
105 		struct mbuf *m, struct sockaddr *addr, struct mbuf *control);
106 
107 /*
108  * Initialize divert connection block queue.
109  */
110 void
111 div_init(void)
112 {
113 	LIST_INIT(&divcb);
114 	divcbinfo.listhead = &divcb;
115 	/*
116 	 * XXX We don't use the hash list for divert IP, but it's easier
117 	 * to allocate a one entry hash list than it is to check all
118 	 * over the place for hashbase == NULL.
119 	 */
120 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
121 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
122 	divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb),
123 				   maxsockets, ZONE_INTERRUPT, 0);
124 }
125 
126 /*
127  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
128  * with that protocol number to enter the system from the outside.
129  */
130 void
131 div_input(struct mbuf *m, int off, int proto)
132 {
133 	ipstat.ips_noproto++;
134 	m_freem(m);
135 }
136 
137 /*
138  * Divert a packet by passing it up to the divert socket at port 'port'.
139  *
140  * Setup generic address and protocol structures for div_input routine,
141  * then pass them along with mbuf chain.
142  */
143 void
144 divert_packet(struct mbuf *m, int incoming, int port)
145 {
146 	struct ip *ip;
147 	struct inpcb *inp;
148 	struct socket *sa;
149 	u_int16_t nport;
150 
151 	/* Sanity check */
152 	KASSERT(port != 0, ("%s: port=0", __FUNCTION__));
153 
154 	/* Record and reset divert cookie */
155 	divsrc.sin_port = ip_divert_cookie;
156 	ip_divert_cookie = 0;
157 
158 	/* Assure header */
159 	if (m->m_len < sizeof(struct ip) &&
160 	    (m = m_pullup(m, sizeof(struct ip))) == 0) {
161 		return;
162 	}
163 	ip = mtod(m, struct ip *);
164 
165 	/*
166 	 * Record receive interface address, if any.
167 	 * But only for incoming packets.
168 	 */
169 	divsrc.sin_addr.s_addr = 0;
170 	if (incoming) {
171 		struct ifaddr *ifa;
172 
173 		/* Sanity check */
174 		KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __FUNCTION__));
175 
176 		/* Find IP address for receive interface */
177 		for (ifa = m->m_pkthdr.rcvif->if_addrhead.tqh_first;
178 		    ifa != NULL; ifa = ifa->ifa_link.tqe_next) {
179 			if (ifa->ifa_addr == NULL)
180 				continue;
181 			if (ifa->ifa_addr->sa_family != AF_INET)
182 				continue;
183 			divsrc.sin_addr =
184 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
185 			break;
186 		}
187 	}
188 	/*
189 	 * Record the incoming interface name whenever we have one.
190 	 */
191 	bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero));
192 	if (m->m_pkthdr.rcvif) {
193 		/*
194 		 * Hide the actual interface name in there in the
195 		 * sin_zero array. XXX This needs to be moved to a
196 		 * different sockaddr type for divert, e.g.
197 		 * sockaddr_div with multiple fields like
198 		 * sockaddr_dl. Presently we have only 7 bytes
199 		 * but that will do for now as most interfaces
200 		 * are 4 or less + 2 or less bytes for unit.
201 		 * There is probably a faster way of doing this,
202 		 * possibly taking it from the sockaddr_dl on the iface.
203 		 * This solves the problem of a P2P link and a LAN interface
204 		 * having the same address, which can result in the wrong
205 		 * interface being assigned to the packet when fed back
206 		 * into the divert socket. Theoretically if the daemon saves
207 		 * and re-uses the sockaddr_in as suggested in the man pages,
208 		 * this iface name will come along for the ride.
209 		 * (see div_output for the other half of this.)
210 		 */
211 		snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero),
212 			"%s%d", m->m_pkthdr.rcvif->if_name,
213 			m->m_pkthdr.rcvif->if_unit);
214 	}
215 
216 	/* Put packet on socket queue, if any */
217 	sa = NULL;
218 	nport = htons((u_int16_t)port);
219 	for (inp = divcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
220 		if (inp->inp_lport == nport)
221 			sa = inp->inp_socket;
222 	}
223 	if (sa) {
224 		if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc,
225 				m, (struct mbuf *)0) == 0)
226 			m_freem(m);
227 		else
228 			sorwakeup(sa);
229 	} else {
230 		m_freem(m);
231 		ipstat.ips_noproto++;
232 		ipstat.ips_delivered--;
233         }
234 }
235 
236 /*
237  * Deliver packet back into the IP processing machinery.
238  *
239  * If no address specified, or address is 0.0.0.0, send to ip_output();
240  * otherwise, send to ip_input() and mark as having been received on
241  * the interface with that address.
242  */
243 static int
244 div_output(so, m, addr, control)
245 	struct socket *so;
246 	register struct mbuf *m;
247 	struct sockaddr *addr;
248 	struct mbuf *control;
249 {
250 	register struct inpcb *const inp = sotoinpcb(so);
251 	register struct ip *const ip = mtod(m, struct ip *);
252 	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
253 	int error = 0;
254 
255 	if (control)
256 		m_freem(control);		/* XXX */
257 
258 	/* Loopback avoidance and state recovery */
259 	if (sin) {
260 		int	len = 0;
261 		char	*c = sin->sin_zero;
262 
263 		ip_divert_cookie = sin->sin_port;
264 
265 		/*
266 		 * Find receive interface with the given name or IP address.
267 		 * The name is user supplied data so don't trust it's size or
268 		 * that it is zero terminated. The name has priority.
269 		 * We are presently assuming that the sockaddr_in
270 		 * has not been replaced by a sockaddr_div, so we limit it
271 		 * to 16 bytes in total. the name is stuffed (if it exists)
272 		 * in the sin_zero[] field.
273 		 */
274 		while (*c++ && (len++ < sizeof(sin->sin_zero)));
275 		if ((len > 0) && (len < sizeof(sin->sin_zero)))
276 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
277 	} else {
278 		ip_divert_cookie = 0;
279 	}
280 
281 	/* Reinject packet into the system as incoming or outgoing */
282 	if (!sin || sin->sin_addr.s_addr == 0) {
283 		/*
284 		 * Don't allow both user specified and setsockopt options,
285 		 * and don't allow packet length sizes that will crash
286 		 */
287 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
288 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
289 			error = EINVAL;
290 			goto cantsend;
291 		}
292 
293 		/* Convert fields to host order for ip_output() */
294 		NTOHS(ip->ip_len);
295 		NTOHS(ip->ip_off);
296 
297 		/* Send packet to output processing */
298 		ipstat.ips_rawout++;			/* XXX */
299 		error = ip_output(m, inp->inp_options, &inp->inp_route,
300 			(so->so_options & SO_DONTROUTE) |
301 			IP_ALLOWBROADCAST | IP_RAWOUTPUT,
302 			inp->inp_moptions);
303 	} else {
304 		struct	ifaddr *ifa;
305 
306 		/* If no luck with the name above. check by IP address.  */
307 		if (m->m_pkthdr.rcvif == NULL) {
308 			/*
309 			 * Make sure there are no distractions
310 			 * for ifa_ifwithaddr. Clear the port and the ifname.
311 			 * Maybe zap all 8 bytes at once using a 64bit write?
312 			 */
313 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
314 			/* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */
315 			sin->sin_port = 0;
316 			if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) {
317 				error = EADDRNOTAVAIL;
318 				goto cantsend;
319 			}
320 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
321 		}
322 
323 		/* Send packet to input processing */
324 		ip_input(m);
325 	}
326 
327 	/* paranoid: Reset for next time (and other packets) */
328 	/* almost definitly already done in the ipfw filter but.. */
329 	ip_divert_cookie = 0;
330 	return error;
331 
332 cantsend:
333 	m_freem(m);
334 	ip_divert_cookie = 0;
335 	return error;
336 }
337 
338 static int
339 div_attach(struct socket *so, int proto, struct proc *p)
340 {
341 	struct inpcb *inp;
342 	int error, s;
343 
344 	inp  = sotoinpcb(so);
345 	if (inp)
346 		panic("div_attach");
347 	if (p && (error = suser(p)) != 0)
348 		return error;
349 
350 	error = soreserve(so, div_sendspace, div_recvspace);
351 	if (error)
352 		return error;
353 	s = splnet();
354 	error = in_pcballoc(so, &divcbinfo, p);
355 	splx(s);
356 	if (error)
357 		return error;
358 	inp = (struct inpcb *)so->so_pcb;
359 	inp->inp_ip_p = proto;
360 	inp->inp_vflag |= INP_IPV4;
361 	inp->inp_flags |= INP_HDRINCL;
362 	/* The socket is always "connected" because
363 	   we always know "where" to send the packet */
364 	so->so_state |= SS_ISCONNECTED;
365 #ifdef IPSEC
366 	error = ipsec_init_policy(so, &inp->inp_sp);
367 	if (error != 0) {
368 		in_pcbdetach(inp);
369 		return error;
370 	}
371 #endif /*IPSEC*/
372 	return 0;
373 }
374 
375 static int
376 div_detach(struct socket *so)
377 {
378 	struct inpcb *inp;
379 
380 	inp = sotoinpcb(so);
381 	if (inp == 0)
382 		panic("div_detach");
383 	in_pcbdetach(inp);
384 	return 0;
385 }
386 
387 static int
388 div_abort(struct socket *so)
389 {
390 	soisdisconnected(so);
391 	return div_detach(so);
392 }
393 
394 static int
395 div_disconnect(struct socket *so)
396 {
397 	if ((so->so_state & SS_ISCONNECTED) == 0)
398 		return ENOTCONN;
399 	return div_abort(so);
400 }
401 
402 static int
403 div_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
404 {
405 	struct inpcb *inp;
406 	int s;
407 	int error;
408 
409 	s = splnet();
410 	inp = sotoinpcb(so);
411 	/* in_pcbbind assumes that the socket is a sockaddr_in
412 	 * and in_pcbbind requires a valid address. Since divert
413 	 * sockets don't we need to make sure the address is
414 	 * filled in properly.
415 	 * XXX -- divert should not be abusing in_pcbind
416 	 * and should probably have its own family.
417 	 */
418 	if (nam->sa_family != AF_INET) {
419 		error = EAFNOSUPPORT;
420 	} else {
421 		((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
422 		error = in_pcbbind(inp, nam, p);
423 	}
424 	splx(s);
425 	return error;
426 }
427 
428 static int
429 div_shutdown(struct socket *so)
430 {
431 	socantsendmore(so);
432 	return 0;
433 }
434 
435 static int
436 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
437 	 struct mbuf *control, struct proc *p)
438 {
439 	/* Packet must have a header (but that's about it) */
440 	if (m->m_len < sizeof (struct ip) &&
441 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
442 		ipstat.ips_toosmall++;
443 		m_freem(m);
444 		return EINVAL;
445 	}
446 
447 	/* Send packet */
448 	return div_output(so, m, nam, control);
449 }
450 
451 static int
452 div_pcblist(SYSCTL_HANDLER_ARGS)
453 {
454 	int error, i, n, s;
455 	struct inpcb *inp, **inp_list;
456 	inp_gen_t gencnt;
457 	struct xinpgen xig;
458 
459 	/*
460 	 * The process of preparing the TCB list is too time-consuming and
461 	 * resource-intensive to repeat twice on every request.
462 	 */
463 	if (req->oldptr == 0) {
464 		n = divcbinfo.ipi_count;
465 		req->oldidx = 2 * (sizeof xig)
466 			+ (n + n/8) * sizeof(struct xinpcb);
467 		return 0;
468 	}
469 
470 	if (req->newptr != 0)
471 		return EPERM;
472 
473 	/*
474 	 * OK, now we're committed to doing something.
475 	 */
476 	s = splnet();
477 	gencnt = divcbinfo.ipi_gencnt;
478 	n = divcbinfo.ipi_count;
479 	splx(s);
480 
481 	xig.xig_len = sizeof xig;
482 	xig.xig_count = n;
483 	xig.xig_gen = gencnt;
484 	xig.xig_sogen = so_gencnt;
485 	error = SYSCTL_OUT(req, &xig, sizeof xig);
486 	if (error)
487 		return error;
488 
489 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
490 	if (inp_list == 0)
491 		return ENOMEM;
492 
493 	s = splnet();
494 	for (inp = divcbinfo.listhead->lh_first, i = 0; inp && i < n;
495 	     inp = inp->inp_list.le_next) {
496 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
497 			inp_list[i++] = inp;
498 	}
499 	splx(s);
500 	n = i;
501 
502 	error = 0;
503 	for (i = 0; i < n; i++) {
504 		inp = inp_list[i];
505 		if (inp->inp_gencnt <= gencnt) {
506 			struct xinpcb xi;
507 			xi.xi_len = sizeof xi;
508 			/* XXX should avoid extra copy */
509 			bcopy(inp, &xi.xi_inp, sizeof *inp);
510 			if (inp->inp_socket)
511 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
512 			error = SYSCTL_OUT(req, &xi, sizeof xi);
513 		}
514 	}
515 	if (!error) {
516 		/*
517 		 * Give the user an updated idea of our state.
518 		 * If the generation differs from what we told
519 		 * her before, she knows that something happened
520 		 * while we were processing this request, and it
521 		 * might be necessary to retry.
522 		 */
523 		s = splnet();
524 		xig.xig_gen = divcbinfo.ipi_gencnt;
525 		xig.xig_sogen = so_gencnt;
526 		xig.xig_count = divcbinfo.ipi_count;
527 		splx(s);
528 		error = SYSCTL_OUT(req, &xig, sizeof xig);
529 	}
530 	free(inp_list, M_TEMP);
531 	return error;
532 }
533 
534 SYSCTL_DECL(_net_inet_divert);
535 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
536 	    div_pcblist, "S,xinpcb", "List of active divert sockets");
537 
538 struct pr_usrreqs div_usrreqs = {
539 	div_abort, pru_accept_notsupp, div_attach, div_bind,
540 	pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
541 	div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp,
542 	pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
543 	in_setsockaddr, sosend, soreceive, sopoll
544 };
545