xref: /titanic_41/usr/src/uts/common/inet/sctp/sctp_conn.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/systm.h>
31 #include <sys/stream.h>
32 #include <sys/cmn_err.h>
33 #include <sys/kmem.h>
34 #define	_SUN_TPI_VERSION 2
35 #include <sys/tihdr.h>
36 #include <sys/stropts.h>
37 #include <sys/strsubr.h>
38 #include <sys/socket.h>
39 
40 #include <netinet/in.h>
41 #include <netinet/ip6.h>
42 
43 #include <inet/common.h>
44 #include <inet/ip.h>
45 #include <inet/ip6.h>
46 #include <inet/ipclassifier.h>
47 #include <inet/ipsec_impl.h>
48 
49 #include "sctp_impl.h"
50 #include "sctp_addr.h"
51 
52 /*
53  * Common accept code.  Called by sctp_conn_request.
54  * cr_pkt is the INIT / INIT ACK packet.
55  */
56 static int
57 sctp_accept_comm(sctp_t *listener, sctp_t *acceptor, mblk_t *cr_pkt,
58     uint_t ip_hdr_len, sctp_init_chunk_t *iack)
59 {
60 
61 	sctp_hdr_t		*sctph;
62 	sctp_chunk_hdr_t	*ich;
63 	sctp_init_chunk_t	*init;
64 	int			err;
65 	uint_t			sctp_options;
66 
67 	sctph = (sctp_hdr_t *)(cr_pkt->b_rptr + ip_hdr_len);
68 	ASSERT(OK_32PTR(sctph));
69 
70 	acceptor->sctp_lport = listener->sctp_lport;
71 	acceptor->sctp_fport = sctph->sh_sport;
72 
73 	ich = (sctp_chunk_hdr_t *)(iack + 1);
74 	init = (sctp_init_chunk_t *)(ich + 1);
75 
76 	/* acceptor isn't in any fanouts yet, so don't need to hold locks */
77 	ASSERT(acceptor->sctp_faddrs == NULL);
78 	err = sctp_get_addrparams(acceptor, listener, cr_pkt, ich,
79 	    &sctp_options);
80 	if (err != 0)
81 		return (err);
82 
83 	if ((sctp_options & SCTP_PRSCTP_OPTION) &&
84 	    listener->sctp_prsctp_aware && sctp_prsctp_enabled) {
85 		acceptor->sctp_prsctp_aware = B_TRUE;
86 	} else {
87 		acceptor->sctp_prsctp_aware = B_FALSE;
88 	}
89 	/* The new sctp_t is fully bound now. */
90 	acceptor->sctp_connp->conn_fully_bound = B_TRUE;
91 
92 	sctp_set_hdraddrs(acceptor);
93 
94 	/* Get  initial TSNs */
95 	acceptor->sctp_ltsn = ntohl(iack->sic_inittsn);
96 	acceptor->sctp_recovery_tsn = acceptor->sctp_lastack_rxd =
97 	    acceptor->sctp_ltsn - 1;
98 	acceptor->sctp_adv_pap = acceptor->sctp_lastack_rxd;
99 	/* Serial numbers are initialized to the same value as the TSNs */
100 	acceptor->sctp_lcsn = acceptor->sctp_ltsn;
101 
102 	if (!sctp_initialize_params(acceptor, init, iack))
103 		return (ENOMEM);
104 
105 	/*
106 	 * Copy sctp_secret from the listener in case we need to validate
107 	 * a possibly delayed cookie.
108 	 */
109 	bcopy(listener->sctp_secret, acceptor->sctp_secret, SCTP_SECRET_LEN);
110 	bcopy(listener->sctp_old_secret, acceptor->sctp_old_secret,
111 	    SCTP_SECRET_LEN);
112 	acceptor->sctp_last_secret_update = lbolt64;
113 
114 	/*
115 	 * After acceptor is inserted in the hash list, it can be found.
116 	 * So we need to lock it here.
117 	 */
118 	RUN_SCTP(acceptor);
119 
120 	sctp_conn_hash_insert(&sctp_conn_fanout[
121 	    SCTP_CONN_HASH(acceptor->sctp_ports)], acceptor, 0);
122 	sctp_bind_hash_insert(&sctp_bind_fanout[
123 	    SCTP_BIND_HASH(ntohs(acceptor->sctp_lport))], acceptor, 0);
124 
125 	/*
126 	 * No need to check for multicast destination since ip will only pass
127 	 * up multicasts to those that have expressed interest
128 	 * TODO: what about rejecting broadcasts?
129 	 * Also check that source is not a multicast or broadcast address.
130 	 */
131 	/* XXXSCTP */
132 	acceptor->sctp_state = SCTPS_ESTABLISHED;
133 	acceptor->sctp_assoc_start_time = (uint32_t)lbolt;
134 	/*
135 	 * listener->sctp_rwnd should be the default window size or a
136 	 * window size changed via SO_RCVBUF option.
137 	 */
138 	acceptor->sctp_rwnd = MSS_ROUNDUP(listener->sctp_rwnd,
139 	    (acceptor->sctp_mss - sizeof (sctp_data_hdr_t)));
140 
141 	bcopy(&listener->sctp_upcalls, &acceptor->sctp_upcalls,
142 	    sizeof (sctp_upcalls_t));
143 
144 	return (0);
145 }
146 
147 /* Process the COOKIE packet, mp, directed at the listener 'sctp' */
148 sctp_t *
149 sctp_conn_request(sctp_t *sctp, mblk_t *mp, uint_t ifindex, uint_t ip_hdr_len,
150     sctp_init_chunk_t *iack, mblk_t *ipsec_mp)
151 {
152 	sctp_t	*eager;
153 	uint_t	ipvers;
154 	ip6_t	*ip6h;
155 	int	err;
156 	conn_t	*connp, *econnp;
157 
158 	/*
159 	 * No need to check for duplicate as this is the listener
160 	 * and we are holding the lock.  This means that no new
161 	 * connection can be created out of it.  And since the
162 	 * fanout already done cannot find a match, it means that
163 	 * there is no duplicate.
164 	 */
165 	ipvers = IPH_HDR_VERSION(mp->b_rptr);
166 	ASSERT(ipvers == IPV6_VERSION || ipvers == IPV4_VERSION);
167 	ASSERT(OK_32PTR(mp->b_rptr));
168 
169 	if ((eager = sctp_create_eager(sctp)) == NULL) {
170 		return (NULL);
171 	}
172 
173 	if (ipvers != IPV4_VERSION) {
174 		ip6h = (ip6_t *)mp->b_rptr;
175 		/*
176 		 * Record ifindex (might be zero) to tie this connection to
177 		 * that interface if either the listener was bound or
178 		 * if the connection is using link-local addresses.
179 		 */
180 		if (sctp->sctp_bound_if == ifindex ||
181 		    IN6_IS_ADDR_LINKLOCAL(&ip6h->ip6_src))
182 			eager->sctp_bound_if = ifindex;
183 		/*
184 		 * XXX broken. bound_if is always overwritten by statement
185 		 * below. What is the right thing to do here?
186 		 */
187 		eager->sctp_bound_if = sctp->sctp_bound_if;
188 	}
189 
190 	connp = sctp->sctp_connp;
191 	econnp = eager->sctp_connp;
192 
193 	if (connp->conn_policy != NULL) {
194 		ipsec_in_t *ii;
195 
196 		ASSERT(ipsec_mp != NULL);
197 		ii = (ipsec_in_t *)(ipsec_mp->b_rptr);
198 		ASSERT(ii->ipsec_in_policy == NULL);
199 		IPPH_REFHOLD(connp->conn_policy);
200 		ii->ipsec_in_policy = connp->conn_policy;
201 
202 		ipsec_mp->b_datap->db_type = IPSEC_POLICY_SET;
203 		if (!ip_bind_ipsec_policy_set(econnp, ipsec_mp)) {
204 			sctp_close_eager(eager);
205 			BUMP_MIB(&sctp_mib, sctpListenDrop);
206 			return (NULL);
207 		}
208 	}
209 
210 	if (ipsec_mp != NULL) {
211 		/*
212 		 * XXX need to fix the cached policy issue here.
213 		 * We temporarily set the conn_src/conn_rem here so
214 		 * that IPsec can use it for the latched policy
215 		 * selector.  This is obvioursly wrong as SCTP can
216 		 * use different addresses...
217 		 */
218 		if (ipvers == IPV4_VERSION) {
219 			ipha_t	*ipha;
220 
221 			ipha = (ipha_t *)mp->b_rptr;
222 			econnp->conn_src = ipha->ipha_dst;
223 			econnp->conn_rem = ipha->ipha_src;
224 		} else {
225 			econnp->conn_srcv6 = ip6h->ip6_dst;
226 			econnp->conn_remv6 = ip6h->ip6_src;
227 		}
228 	}
229 	if (ipsec_conn_cache_policy(econnp, ipvers == IPV4_VERSION) != 0) {
230 		sctp_close_eager(eager);
231 		BUMP_MIB(&sctp_mib, sctpListenDrop);
232 		return (NULL);
233 	}
234 
235 	err = sctp_accept_comm(sctp, eager, mp, ip_hdr_len, iack);
236 	if (err) {
237 		sctp_close_eager(eager);
238 		BUMP_MIB(&sctp_mib, sctpListenDrop);
239 		return (NULL);
240 	}
241 
242 	/* Connection established, so send up the conn_ind */
243 	if ((eager->sctp_ulpd = sctp->sctp_ulp_newconn(sctp->sctp_ulpd,
244 	    eager)) == NULL) {
245 		sctp_close_eager(eager);
246 		BUMP_MIB(&sctp_mib, sctpListenDrop);
247 		return (NULL);
248 	}
249 	ASSERT(SCTP_IS_DETACHED(eager));
250 	eager->sctp_detached = B_FALSE;
251 	if (eager->sctp_family == AF_INET) {
252 		eager->sctp_ulp_prop(eager->sctp_ulpd,
253 		    sctp_wroff_xtra + sizeof (sctp_data_hdr_t) +
254 		    sctp->sctp_hdr_len, strmsgsz);
255 	} else {
256 		eager->sctp_ulp_prop(eager->sctp_ulpd,
257 		    sctp_wroff_xtra + sizeof (sctp_data_hdr_t) +
258 		    sctp->sctp_hdr6_len, strmsgsz);
259 	}
260 	return (eager);
261 }
262 
263 /*
264  * Connect to a peer - this function inserts the sctp in the
265  * bind and conn fanouts, sends the INIT, and replies to the client
266  * with an OK ack.
267  */
268 /* ARGSUSED */
269 int
270 sctp_connect(sctp_t *sctp, const struct sockaddr *dst, uint32_t addrlen)
271 {
272 	sin_t		*sin;
273 	sin6_t		*sin6;
274 	in_port_t	lport;
275 	in6_addr_t	dstaddr;
276 	in_port_t	dstport;
277 	mblk_t		*initmp;
278 	sctp_tf_t	*tbf;
279 	sctp_t		*lsctp;
280 	char		buf[INET6_ADDRSTRLEN];
281 	int		sleep = sctp->sctp_cansleep ? KM_SLEEP : KM_NOSLEEP;
282 	int 		hdrlen;
283 	ip6_rthdr_t	*rth;
284 	sctp_faddr_t	*cur_fp;
285 
286 	/*
287 	 * Determine packet type based on type of address passed in
288 	 * the request should contain an IPv4 or IPv6 address.
289 	 * Make sure that address family matches the type of
290 	 * family of the the address passed down
291 	 */
292 	if (addrlen < sizeof (sin_t)) {
293 		return (EINVAL);
294 	}
295 	switch (dst->sa_family) {
296 	case AF_INET:
297 		sin = (sin_t *)dst;
298 
299 		/* Check for attempt to connect to non-unicast */
300 		if (IN_MULTICAST(sin->sin_addr.s_addr) ||
301 		    (sin->sin_addr.s_addr == INADDR_BROADCAST)) {
302 			ip0dbg(("sctp_connect: non-unicast\n"));
303 			return (EINVAL);
304 		}
305 		if (sctp->sctp_connp->conn_ipv6_v6only)
306 			return (EAFNOSUPPORT);
307 
308 		/* convert to v6 mapped */
309 		/* Check for attempt to connect to INADDR_ANY */
310 		if (sin->sin_addr.s_addr == INADDR_ANY)  {
311 			struct in_addr v4_addr;
312 			/*
313 			 * SunOS 4.x and 4.3 BSD allow an application
314 			 * to connect a TCP socket to INADDR_ANY.
315 			 * When they do this, the kernel picks the
316 			 * address of one interface and uses it
317 			 * instead.  The kernel usually ends up
318 			 * picking the address of the loopback
319 			 * interface.  This is an undocumented feature.
320 			 * However, we provide the same thing here
321 			 * in case any TCP apps that use this feature
322 			 * are being ported to SCTP...
323 			 */
324 			v4_addr.s_addr = htonl(INADDR_LOOPBACK);
325 			IN6_INADDR_TO_V4MAPPED(&v4_addr, &dstaddr);
326 		} else {
327 			IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &dstaddr);
328 		}
329 		dstport = sin->sin_port;
330 		if (sin->sin_family == AF_INET) {
331 			hdrlen = sctp->sctp_hdr_len;
332 		} else {
333 			hdrlen = sctp->sctp_hdr6_len;
334 		}
335 		break;
336 	case AF_INET6:
337 		sin6 = (sin6_t *)dst;
338 		/* Check for attempt to connect to non-unicast. */
339 		if ((addrlen < sizeof (sin6_t)) ||
340 		    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
341 			ip0dbg(("sctp_connect: non-unicast\n"));
342 			return (EINVAL);
343 		}
344 		if (sctp->sctp_connp->conn_ipv6_v6only &&
345 		    IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
346 			return (EAFNOSUPPORT);
347 		}
348 		/* check for attempt to connect to unspec */
349 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
350 			dstaddr = ipv6_loopback;
351 		} else {
352 			dstaddr = sin6->sin6_addr;
353 		}
354 		dstport = sin6->sin6_port;
355 		hdrlen = sctp->sctp_hdr6_len;
356 		break;
357 	default:
358 		dprint(1, ("sctp_connect: unknown family %d\n",
359 			dst->sa_family));
360 		return (EAFNOSUPPORT);
361 	}
362 
363 	(void) inet_ntop(AF_INET6, &dstaddr, buf, sizeof (buf));
364 	dprint(1, ("sctp_connect: attempting connect to %s...\n", buf));
365 
366 	RUN_SCTP(sctp);
367 
368 	if (sctp->sctp_family != dst->sa_family) {
369 		WAKE_SCTP(sctp);
370 		return (EINVAL);
371 	}
372 
373 	switch (sctp->sctp_state) {
374 	case SCTPS_IDLE: {
375 		int	err;
376 		/*
377 		 * We support a quick connect capability here, allowing
378 		 * clients to transition directly from IDLE to COOKIE_WAIT.
379 		 * sctp_bindi will pick an unused port, insert the connection
380 		 * in the bind hash and transition to BOUND state. SCTP
381 		 * picks and uses what it considers the optimal local address
382 		 * set (just like specifiying INADDR_ANY to bind()).
383 		 */
384 		dprint(1, ("sctp_connect: idle, attempting bind...\n"));
385 		ASSERT(sctp->sctp_nsaddrs == 0);
386 
387 		err = sctp_dup_saddrs(NULL, sctp, sleep);
388 		if (err != 0) {
389 			WAKE_SCTP(sctp);
390 			return (err);
391 		}
392 		lport = sctp_update_next_port(sctp_next_port_to_try);
393 		lport = sctp_bindi(sctp, lport, 0, 0);
394 		if (lport == 0) {
395 			WAKE_SCTP(sctp);
396 			sctp_free_saddrs(sctp);
397 			return (EADDRNOTAVAIL);
398 		}
399 		sctp->sctp_bound_to_all = 1;
400 		/* FALLTHRU */
401 	}
402 
403 	case SCTPS_BOUND:
404 		ASSERT(sctp->sctp_nsaddrs > 0);
405 
406 		/* do the connect */
407 		/* XXX check for attempt to connect to self */
408 		sctp->sctp_fport = dstport;
409 
410 		ASSERT(sctp->sctp_iphc);
411 		ASSERT(sctp->sctp_iphc6);
412 
413 		/*
414 		 * Don't allow this connection to completely duplicate
415 		 * an existing connection.
416 		 *
417 		 * Ensure that the duplicate check and insertion is atomic.
418 		 */
419 		sctp_conn_hash_remove(sctp);
420 		tbf = &sctp_conn_fanout[SCTP_CONN_HASH(sctp->sctp_ports)];
421 		mutex_enter(&tbf->tf_lock);
422 		lsctp = sctp_lookup(sctp, &dstaddr, tbf, &sctp->sctp_ports,
423 		    SCTPS_COOKIE_WAIT);
424 		if (lsctp != NULL) {
425 			/* found a duplicate connection */
426 			mutex_exit(&tbf->tf_lock);
427 			SCTP_REFRELE(lsctp);
428 			WAKE_SCTP(sctp);
429 			return (EADDRINUSE);
430 		}
431 		/*
432 		 * OK; set up the peer addr (this may grow after we get
433 		 * the INIT ACK from the peer with additional addresses).
434 		 */
435 		if (sctp_add_faddr(sctp, &dstaddr, sleep) < 0) {
436 			mutex_exit(&tbf->tf_lock);
437 			WAKE_SCTP(sctp);
438 			return (ENOMEM);
439 		}
440 		/* No valid src addr, return. */
441 		if (sctp->sctp_faddrs->state == SCTP_FADDRS_UNREACH) {
442 			mutex_exit(&tbf->tf_lock);
443 			WAKE_SCTP(sctp);
444 			return (EADDRNOTAVAIL);
445 		}
446 		sctp->sctp_primary = sctp->sctp_faddrs;
447 		sctp->sctp_current = sctp->sctp_faddrs;
448 		cur_fp = sctp->sctp_current;
449 		sctp->sctp_mss = sctp->sctp_faddrs->sfa_pmss;
450 		sctp_conn_hash_insert(tbf, sctp, 1);
451 		mutex_exit(&tbf->tf_lock);
452 
453 		/* initialize composite headers */
454 		sctp_set_hdraddrs(sctp);
455 
456 		/*
457 		 * Massage a routing header (if present) putting the first hop
458 		 * in ip6_dst.
459 		 */
460 		rth = ip_find_rthdr_v6(sctp->sctp_ip6h,
461 		    (uint8_t *)sctp->sctp_sctph6);
462 		if (rth != NULL)
463 			(void) ip_massage_options_v6(sctp->sctp_ip6h, rth);
464 
465 		/*
466 		 * Turn off the don't fragment bit on the (only) faddr,
467 		 * so that if one of the messages exchanged during the
468 		 * initialization sequence exceeds the path mtu, it
469 		 * at least has a chance to get there. SCTP does no
470 		 * fragmentation of initialization messages.  The DF bit
471 		 * will be turned on again in sctp_send_cookie_echo()
472 		 * (but the cookie echo will still be sent with the df bit
473 		 * off).
474 		 */
475 		cur_fp->df = B_FALSE;
476 
477 		/* Mark this address as alive */
478 		cur_fp->state = SCTP_FADDRS_ALIVE;
479 
480 		/* This sctp_t is fully bound now. */
481 		sctp->sctp_connp->conn_fully_bound = B_TRUE;
482 
483 		/* Send the INIT to the peer */
484 		SCTP_FADDR_TIMER_RESTART(sctp, cur_fp, cur_fp->rto);
485 		initmp = sctp_init_mp(sctp);
486 		if (initmp == NULL) {
487 			WAKE_SCTP(sctp);
488 			/* let timer retry */
489 			return (0);
490 		}
491 		sctp->sctp_state = SCTPS_COOKIE_WAIT;
492 		WAKE_SCTP(sctp);
493 		/* OK to call IP_PUT() here instead of sctp_add_sendq(). */
494 		CONN_INC_REF(sctp->sctp_connp);
495 		initmp->b_flag |= MSGHASREF;
496 		IP_PUT(initmp, sctp->sctp_connp, sctp->sctp_current->isv4);
497 		BUMP_LOCAL(sctp->sctp_opkts);
498 
499 		sctp->sctp_ulp_prop(sctp->sctp_ulpd,
500 		    sctp_wroff_xtra + hdrlen + sizeof (sctp_data_hdr_t), 0);
501 
502 		return (0);
503 	default:
504 		ip0dbg(("sctp_connect: invalid state. %d\n", sctp->sctp_state));
505 		WAKE_SCTP(sctp);
506 		return (EINVAL);
507 	}
508 }
509