xref: /illumos-gate/usr/src/uts/common/inet/tcp/tcp_bind.c (revision 3299f39fdcbdab4be7a9c70daa3873f2b78a398d)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2016 by Delphix. All rights reserved.
26  */
27 
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/strsun.h>
31 #include <sys/strsubr.h>
32 #include <sys/stropts.h>
33 #include <sys/strlog.h>
34 #define	_SUN_TPI_VERSION 2
35 #include <sys/tihdr.h>
36 #include <sys/suntpi.h>
37 #include <sys/xti_inet.h>
38 #include <sys/policy.h>
39 #include <sys/squeue_impl.h>
40 #include <sys/squeue.h>
41 #include <sys/tsol/tnet.h>
42 
43 #include <rpc/pmap_prot.h>
44 
45 #include <inet/common.h>
46 #include <inet/ip.h>
47 #include <inet/tcp.h>
48 #include <inet/tcp_impl.h>
49 #include <inet/proto_set.h>
50 #include <inet/ipsec_impl.h>
51 
52 /* Setable in /etc/system */
53 /* If set to 0, pick ephemeral port sequentially; otherwise randomly. */
54 static uint32_t tcp_random_anon_port = 1;
55 
56 static int	tcp_bind_select_lport(tcp_t *, in_port_t *, boolean_t,
57 		    cred_t *cr);
58 static in_port_t	tcp_get_next_priv_port(const tcp_t *);
59 
60 /*
61  * Hash list insertion routine for tcp_t structures. Each hash bucket
62  * contains a list of tcp_t entries, and each entry is bound to a unique
63  * port. If there are multiple tcp_t's that are bound to the same port, then
64  * one of them will be linked into the hash bucket list, and the rest will
65  * hang off of that one entry. For each port, entries bound to a specific IP
66  * address will be inserted before those those bound to INADDR_ANY.
67  */
68 void
69 tcp_bind_hash_insert(tf_t *tbf, tcp_t *tcp, int caller_holds_lock)
70 {
71 	tcp_t	**tcpp;
72 	tcp_t	*tcpnext;
73 	tcp_t	*tcphash;
74 	conn_t	*connp = tcp->tcp_connp;
75 	conn_t	*connext;
76 
77 	if (tcp->tcp_ptpbhn != NULL) {
78 		ASSERT(!caller_holds_lock);
79 		tcp_bind_hash_remove(tcp);
80 	}
81 	tcpp = &tbf->tf_tcp;
82 	if (!caller_holds_lock) {
83 		mutex_enter(&tbf->tf_lock);
84 	} else {
85 		ASSERT(MUTEX_HELD(&tbf->tf_lock));
86 	}
87 	tcphash = tcpp[0];
88 	tcpnext = NULL;
89 	if (tcphash != NULL) {
90 		/* Look for an entry using the same port */
91 		while ((tcphash = tcpp[0]) != NULL &&
92 		    connp->conn_lport != tcphash->tcp_connp->conn_lport)
93 			tcpp = &(tcphash->tcp_bind_hash);
94 
95 		/* The port was not found, just add to the end */
96 		if (tcphash == NULL)
97 			goto insert;
98 
99 		/*
100 		 * OK, there already exists an entry bound to the
101 		 * same port.
102 		 *
103 		 * If the new tcp bound to the INADDR_ANY address
104 		 * and the first one in the list is not bound to
105 		 * INADDR_ANY we skip all entries until we find the
106 		 * first one bound to INADDR_ANY.
107 		 * This makes sure that applications binding to a
108 		 * specific address get preference over those binding to
109 		 * INADDR_ANY.
110 		 */
111 		tcpnext = tcphash;
112 		connext = tcpnext->tcp_connp;
113 		tcphash = NULL;
114 		if (V6_OR_V4_INADDR_ANY(connp->conn_bound_addr_v6) &&
115 		    !V6_OR_V4_INADDR_ANY(connext->conn_bound_addr_v6)) {
116 			while ((tcpnext = tcpp[0]) != NULL) {
117 				connext = tcpnext->tcp_connp;
118 				if (!V6_OR_V4_INADDR_ANY(
119 				    connext->conn_bound_addr_v6))
120 					tcpp = &(tcpnext->tcp_bind_hash_port);
121 				else
122 					break;
123 			}
124 			if (tcpnext != NULL) {
125 				tcpnext->tcp_ptpbhn = &tcp->tcp_bind_hash_port;
126 				tcphash = tcpnext->tcp_bind_hash;
127 				if (tcphash != NULL) {
128 					tcphash->tcp_ptpbhn =
129 					    &(tcp->tcp_bind_hash);
130 					tcpnext->tcp_bind_hash = NULL;
131 				}
132 			}
133 		} else {
134 			tcpnext->tcp_ptpbhn = &tcp->tcp_bind_hash_port;
135 			tcphash = tcpnext->tcp_bind_hash;
136 			if (tcphash != NULL) {
137 				tcphash->tcp_ptpbhn =
138 				    &(tcp->tcp_bind_hash);
139 				tcpnext->tcp_bind_hash = NULL;
140 			}
141 		}
142 	}
143 insert:
144 	tcp->tcp_bind_hash_port = tcpnext;
145 	tcp->tcp_bind_hash = tcphash;
146 	tcp->tcp_ptpbhn = tcpp;
147 	tcpp[0] = tcp;
148 	if (!caller_holds_lock)
149 		mutex_exit(&tbf->tf_lock);
150 }
151 
152 /*
153  * Hash list removal routine for tcp_t structures.
154  */
155 void
156 tcp_bind_hash_remove(tcp_t *tcp)
157 {
158 	tcp_t	*tcpnext;
159 	kmutex_t *lockp;
160 	tcp_stack_t	*tcps = tcp->tcp_tcps;
161 	conn_t		*connp = tcp->tcp_connp;
162 
163 	if (tcp->tcp_ptpbhn == NULL)
164 		return;
165 
166 	/*
167 	 * Extract the lock pointer in case there are concurrent
168 	 * hash_remove's for this instance.
169 	 */
170 	ASSERT(connp->conn_lport != 0);
171 	lockp = &tcps->tcps_bind_fanout[TCP_BIND_HASH(
172 	    connp->conn_lport)].tf_lock;
173 
174 	ASSERT(lockp != NULL);
175 	mutex_enter(lockp);
176 	if (tcp->tcp_ptpbhn) {
177 		tcpnext = tcp->tcp_bind_hash_port;
178 		if (tcpnext != NULL) {
179 			tcp->tcp_bind_hash_port = NULL;
180 			tcpnext->tcp_ptpbhn = tcp->tcp_ptpbhn;
181 			tcpnext->tcp_bind_hash = tcp->tcp_bind_hash;
182 			if (tcpnext->tcp_bind_hash != NULL) {
183 				tcpnext->tcp_bind_hash->tcp_ptpbhn =
184 				    &(tcpnext->tcp_bind_hash);
185 				tcp->tcp_bind_hash = NULL;
186 			}
187 		} else if ((tcpnext = tcp->tcp_bind_hash) != NULL) {
188 			tcpnext->tcp_ptpbhn = tcp->tcp_ptpbhn;
189 			tcp->tcp_bind_hash = NULL;
190 		}
191 		*tcp->tcp_ptpbhn = tcpnext;
192 		tcp->tcp_ptpbhn = NULL;
193 	}
194 	mutex_exit(lockp);
195 }
196 
197 /*
198  * Don't let port fall into the privileged range.
199  * Since the extra privileged ports can be arbitrary we also
200  * ensure that we exclude those from consideration.
201  * tcp_g_epriv_ports is not sorted thus we loop over it until
202  * there are no changes.
203  *
204  * Note: No locks are held when inspecting tcp_g_*epriv_ports
205  * but instead the code relies on:
206  * - the fact that the address of the array and its size never changes
207  * - the atomic assignment of the elements of the array
208  *
209  * Returns 0 if there are no more ports available.
210  *
211  * TS note: skip multilevel ports.
212  */
213 in_port_t
214 tcp_update_next_port(in_port_t port, const tcp_t *tcp, boolean_t random)
215 {
216 	int i, bump;
217 	boolean_t restart = B_FALSE;
218 	tcp_stack_t *tcps = tcp->tcp_tcps;
219 
220 	if (random && tcp_random_anon_port != 0) {
221 		(void) random_get_pseudo_bytes((uint8_t *)&port,
222 		    sizeof (in_port_t));
223 		/*
224 		 * Unless changed by a sys admin, the smallest anon port
225 		 * is 32768 and the largest anon port is 65535.  It is
226 		 * very likely (50%) for the random port to be smaller
227 		 * than the smallest anon port.  When that happens,
228 		 * add port % (anon port range) to the smallest anon
229 		 * port to get the random port.  It should fall into the
230 		 * valid anon port range.
231 		 */
232 		if ((port < tcps->tcps_smallest_anon_port) ||
233 		    (port > tcps->tcps_largest_anon_port)) {
234 			if (tcps->tcps_smallest_anon_port ==
235 			    tcps->tcps_largest_anon_port) {
236 				bump = 0;
237 			} else {
238 				bump = port % (tcps->tcps_largest_anon_port -
239 				    tcps->tcps_smallest_anon_port);
240 			}
241 			port = tcps->tcps_smallest_anon_port + bump;
242 		}
243 	}
244 
245 retry:
246 	if (port < tcps->tcps_smallest_anon_port)
247 		port = (in_port_t)tcps->tcps_smallest_anon_port;
248 
249 	if (port > tcps->tcps_largest_anon_port) {
250 		if (restart)
251 			return (0);
252 		restart = B_TRUE;
253 		port = (in_port_t)tcps->tcps_smallest_anon_port;
254 	}
255 
256 	if (port < tcps->tcps_smallest_nonpriv_port)
257 		port = (in_port_t)tcps->tcps_smallest_nonpriv_port;
258 
259 	for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
260 		if (port == tcps->tcps_g_epriv_ports[i]) {
261 			port++;
262 			/*
263 			 * Make sure whether the port is in the
264 			 * valid range.
265 			 */
266 			goto retry;
267 		}
268 	}
269 	if (is_system_labeled() &&
270 	    (i = tsol_next_port(crgetzone(tcp->tcp_connp->conn_cred), port,
271 	    IPPROTO_TCP, B_TRUE)) != 0) {
272 		port = i;
273 		goto retry;
274 	}
275 	return (port);
276 }
277 
278 /*
279  * Return the next anonymous port in the privileged port range for
280  * bind checking.  It starts at IPPORT_RESERVED - 1 and goes
281  * downwards.  This is the same behavior as documented in the userland
282  * library call rresvport(3N).
283  *
284  * TS note: skip multilevel ports.
285  */
286 static in_port_t
287 tcp_get_next_priv_port(const tcp_t *tcp)
288 {
289 	static in_port_t next_priv_port = IPPORT_RESERVED - 1;
290 	in_port_t nextport;
291 	boolean_t restart = B_FALSE;
292 	tcp_stack_t *tcps = tcp->tcp_tcps;
293 retry:
294 	if (next_priv_port < tcps->tcps_min_anonpriv_port ||
295 	    next_priv_port >= IPPORT_RESERVED) {
296 		next_priv_port = IPPORT_RESERVED - 1;
297 		if (restart)
298 			return (0);
299 		restart = B_TRUE;
300 	}
301 	if (is_system_labeled() &&
302 	    (nextport = tsol_next_port(crgetzone(tcp->tcp_connp->conn_cred),
303 	    next_priv_port, IPPROTO_TCP, B_FALSE)) != 0) {
304 		next_priv_port = nextport;
305 		goto retry;
306 	}
307 	return (next_priv_port--);
308 }
309 
310 static int
311 tcp_bind_select_lport(tcp_t *tcp, in_port_t *requested_port_ptr,
312     boolean_t bind_to_req_port_only, cred_t *cr)
313 {
314 	in_port_t	mlp_port;
315 	mlp_type_t 	addrtype, mlptype;
316 	boolean_t	user_specified;
317 	in_port_t	allocated_port;
318 	in_port_t	requested_port = *requested_port_ptr;
319 	conn_t		*connp = tcp->tcp_connp;
320 	zone_t		*zone;
321 	tcp_stack_t	*tcps = tcp->tcp_tcps;
322 	in6_addr_t	v6addr = connp->conn_laddr_v6;
323 
324 	/*
325 	 * XXX It's up to the caller to specify bind_to_req_port_only or not.
326 	 */
327 	ASSERT(cr != NULL);
328 
329 	/*
330 	 * Get a valid port (within the anonymous range and should not
331 	 * be a privileged one) to use if the user has not given a port.
332 	 * If multiple threads are here, they may all start with
333 	 * with the same initial port. But, it should be fine as long as
334 	 * tcp_bindi will ensure that no two threads will be assigned
335 	 * the same port.
336 	 *
337 	 * NOTE: XXX If a privileged process asks for an anonymous port, we
338 	 * still check for ports only in the range > tcp_smallest_non_priv_port,
339 	 * unless TCP_ANONPRIVBIND option is set.
340 	 */
341 	mlptype = mlptSingle;
342 	mlp_port = requested_port;
343 	if (requested_port == 0) {
344 		requested_port = connp->conn_anon_priv_bind ?
345 		    tcp_get_next_priv_port(tcp) :
346 		    tcp_update_next_port(tcps->tcps_next_port_to_try,
347 		    tcp, B_TRUE);
348 		if (requested_port == 0) {
349 			return (-TNOADDR);
350 		}
351 		user_specified = B_FALSE;
352 
353 		/*
354 		 * If the user went through one of the RPC interfaces to create
355 		 * this socket and RPC is MLP in this zone, then give them an
356 		 * anonymous MLP.
357 		 */
358 		if (connp->conn_anon_mlp && is_system_labeled()) {
359 			zone = crgetzone(cr);
360 			addrtype = tsol_mlp_addr_type(
361 			    connp->conn_allzones ? ALL_ZONES : zone->zone_id,
362 			    IPV6_VERSION, &v6addr,
363 			    tcps->tcps_netstack->netstack_ip);
364 			if (addrtype == mlptSingle) {
365 				return (-TNOADDR);
366 			}
367 			mlptype = tsol_mlp_port_type(zone, IPPROTO_TCP,
368 			    PMAPPORT, addrtype);
369 			mlp_port = PMAPPORT;
370 		}
371 	} else {
372 		int i;
373 		boolean_t priv = B_FALSE;
374 
375 		/*
376 		 * If the requested_port is in the well-known privileged range,
377 		 * verify that the stream was opened by a privileged user.
378 		 * Note: No locks are held when inspecting tcp_g_*epriv_ports
379 		 * but instead the code relies on:
380 		 * - the fact that the address of the array and its size never
381 		 *   changes
382 		 * - the atomic assignment of the elements of the array
383 		 */
384 		if (requested_port < tcps->tcps_smallest_nonpriv_port) {
385 			priv = B_TRUE;
386 		} else {
387 			for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
388 				if (requested_port ==
389 				    tcps->tcps_g_epriv_ports[i]) {
390 					priv = B_TRUE;
391 					break;
392 				}
393 			}
394 		}
395 		if (priv) {
396 			if (secpolicy_net_privaddr(cr, requested_port,
397 			    IPPROTO_TCP) != 0) {
398 				if (connp->conn_debug) {
399 					(void) strlog(TCP_MOD_ID, 0, 1,
400 					    SL_ERROR|SL_TRACE,
401 					    "tcp_bind: no priv for port %d",
402 					    requested_port);
403 				}
404 				return (-TACCES);
405 			}
406 		}
407 		user_specified = B_TRUE;
408 
409 		connp = tcp->tcp_connp;
410 		if (is_system_labeled()) {
411 			zone = crgetzone(cr);
412 			addrtype = tsol_mlp_addr_type(
413 			    connp->conn_allzones ? ALL_ZONES : zone->zone_id,
414 			    IPV6_VERSION, &v6addr,
415 			    tcps->tcps_netstack->netstack_ip);
416 			if (addrtype == mlptSingle) {
417 				return (-TNOADDR);
418 			}
419 			mlptype = tsol_mlp_port_type(zone, IPPROTO_TCP,
420 			    requested_port, addrtype);
421 		}
422 	}
423 
424 	if (mlptype != mlptSingle) {
425 		if (secpolicy_net_bindmlp(cr) != 0) {
426 			if (connp->conn_debug) {
427 				(void) strlog(TCP_MOD_ID, 0, 1,
428 				    SL_ERROR|SL_TRACE,
429 				    "tcp_bind: no priv for multilevel port %d",
430 				    requested_port);
431 			}
432 			return (-TACCES);
433 		}
434 
435 		/*
436 		 * If we're specifically binding a shared IP address and the
437 		 * port is MLP on shared addresses, then check to see if this
438 		 * zone actually owns the MLP.  Reject if not.
439 		 */
440 		if (mlptype == mlptShared && addrtype == mlptShared) {
441 			/*
442 			 * No need to handle exclusive-stack zones since
443 			 * ALL_ZONES only applies to the shared stack.
444 			 */
445 			zoneid_t mlpzone;
446 
447 			mlpzone = tsol_mlp_findzone(IPPROTO_TCP,
448 			    htons(mlp_port));
449 			if (connp->conn_zoneid != mlpzone) {
450 				if (connp->conn_debug) {
451 					(void) strlog(TCP_MOD_ID, 0, 1,
452 					    SL_ERROR|SL_TRACE,
453 					    "tcp_bind: attempt to bind port "
454 					    "%d on shared addr in zone %d "
455 					    "(should be %d)",
456 					    mlp_port, connp->conn_zoneid,
457 					    mlpzone);
458 				}
459 				return (-TACCES);
460 			}
461 		}
462 
463 		if (!user_specified) {
464 			int err;
465 			err = tsol_mlp_anon(zone, mlptype, connp->conn_proto,
466 			    requested_port, B_TRUE);
467 			if (err != 0) {
468 				if (connp->conn_debug) {
469 					(void) strlog(TCP_MOD_ID, 0, 1,
470 					    SL_ERROR|SL_TRACE,
471 					    "tcp_bind: cannot establish anon "
472 					    "MLP for port %d",
473 					    requested_port);
474 				}
475 				return (err);
476 			}
477 			connp->conn_anon_port = B_TRUE;
478 		}
479 		connp->conn_mlp_type = mlptype;
480 	}
481 
482 	allocated_port = tcp_bindi(tcp, requested_port, &v6addr,
483 	    connp->conn_reuseaddr, B_FALSE, bind_to_req_port_only,
484 	    user_specified);
485 
486 	if (allocated_port == 0) {
487 		connp->conn_mlp_type = mlptSingle;
488 		if (connp->conn_anon_port) {
489 			connp->conn_anon_port = B_FALSE;
490 			(void) tsol_mlp_anon(zone, mlptype, connp->conn_proto,
491 			    requested_port, B_FALSE);
492 		}
493 		if (bind_to_req_port_only) {
494 			if (connp->conn_debug) {
495 				(void) strlog(TCP_MOD_ID, 0, 1,
496 				    SL_ERROR|SL_TRACE,
497 				    "tcp_bind: requested addr busy");
498 			}
499 			return (-TADDRBUSY);
500 		} else {
501 			/* If we are out of ports, fail the bind. */
502 			if (connp->conn_debug) {
503 				(void) strlog(TCP_MOD_ID, 0, 1,
504 				    SL_ERROR|SL_TRACE,
505 				    "tcp_bind: out of ports?");
506 			}
507 			return (-TNOADDR);
508 		}
509 	}
510 
511 	/* Pass the allocated port back */
512 	*requested_port_ptr = allocated_port;
513 	return (0);
514 }
515 
516 /*
517  * Check the address and check/pick a local port number.
518  */
519 int
520 tcp_bind_check(conn_t *connp, struct sockaddr *sa, socklen_t len, cred_t *cr,
521     boolean_t bind_to_req_port_only)
522 {
523 	tcp_t	*tcp = connp->conn_tcp;
524 	sin_t	*sin;
525 	sin6_t  *sin6;
526 	in_port_t	requested_port;
527 	ipaddr_t	v4addr;
528 	in6_addr_t	v6addr;
529 	ip_laddr_t	laddr_type = IPVL_UNICAST_UP;	/* INADDR_ANY */
530 	zoneid_t	zoneid = IPCL_ZONEID(connp);
531 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
532 	uint_t		scopeid = 0;
533 	int		error = 0;
534 	ip_xmit_attr_t	*ixa = connp->conn_ixa;
535 
536 	ASSERT((uintptr_t)len <= (uintptr_t)INT_MAX);
537 
538 	if (tcp->tcp_state == TCPS_BOUND) {
539 		return (0);
540 	} else if (tcp->tcp_state > TCPS_BOUND) {
541 		if (connp->conn_debug) {
542 			(void) strlog(TCP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
543 			    "tcp_bind: bad state, %d", tcp->tcp_state);
544 		}
545 		return (-TOUTSTATE);
546 	}
547 
548 	ASSERT(sa != NULL && len != 0);
549 
550 	if (!OK_32PTR((char *)sa)) {
551 		if (connp->conn_debug) {
552 			(void) strlog(TCP_MOD_ID, 0, 1,
553 			    SL_ERROR|SL_TRACE,
554 			    "tcp_bind: bad address parameter, "
555 			    "address %p, len %d",
556 			    (void *)sa, len);
557 		}
558 		return (-TPROTO);
559 	}
560 
561 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
562 	if (error != 0) {
563 		return (error);
564 	}
565 
566 	switch (len) {
567 	case sizeof (sin_t):	/* Complete IPv4 address */
568 		sin = (sin_t *)sa;
569 		requested_port = ntohs(sin->sin_port);
570 		v4addr = sin->sin_addr.s_addr;
571 		IN6_IPADDR_TO_V4MAPPED(v4addr, &v6addr);
572 		if (v4addr != INADDR_ANY) {
573 			laddr_type = ip_laddr_verify_v4(v4addr, zoneid, ipst,
574 			    B_FALSE);
575 		}
576 		break;
577 
578 	case sizeof (sin6_t): /* Complete IPv6 address */
579 		sin6 = (sin6_t *)sa;
580 		v6addr = sin6->sin6_addr;
581 		requested_port = ntohs(sin6->sin6_port);
582 		if (IN6_IS_ADDR_V4MAPPED(&v6addr)) {
583 			if (connp->conn_ipv6_v6only)
584 				return (EADDRNOTAVAIL);
585 
586 			IN6_V4MAPPED_TO_IPADDR(&v6addr, v4addr);
587 			if (v4addr != INADDR_ANY) {
588 				laddr_type = ip_laddr_verify_v4(v4addr,
589 				    zoneid, ipst, B_FALSE);
590 			}
591 		} else {
592 			if (!IN6_IS_ADDR_UNSPECIFIED(&v6addr)) {
593 				if (IN6_IS_ADDR_LINKSCOPE(&v6addr))
594 					scopeid = sin6->sin6_scope_id;
595 				laddr_type = ip_laddr_verify_v6(&v6addr,
596 				    zoneid, ipst, B_FALSE, scopeid);
597 			}
598 		}
599 		break;
600 
601 	default:
602 		if (connp->conn_debug) {
603 			(void) strlog(TCP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
604 			    "tcp_bind: bad address length, %d", len);
605 		}
606 		return (EAFNOSUPPORT);
607 		/* return (-TBADADDR); */
608 	}
609 
610 	/* Is the local address a valid unicast address? */
611 	if (laddr_type == IPVL_BAD)
612 		return (EADDRNOTAVAIL);
613 
614 	connp->conn_bound_addr_v6 = v6addr;
615 	if (scopeid != 0) {
616 		ixa->ixa_flags |= IXAF_SCOPEID_SET;
617 		ixa->ixa_scopeid = scopeid;
618 		connp->conn_incoming_ifindex = scopeid;
619 	} else {
620 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
621 		connp->conn_incoming_ifindex = connp->conn_bound_if;
622 	}
623 
624 	connp->conn_laddr_v6 = v6addr;
625 	connp->conn_saddr_v6 = v6addr;
626 
627 	bind_to_req_port_only = requested_port != 0 && bind_to_req_port_only;
628 
629 	error = tcp_bind_select_lport(tcp, &requested_port,
630 	    bind_to_req_port_only, cr);
631 	if (error != 0) {
632 		connp->conn_laddr_v6 = ipv6_all_zeros;
633 		connp->conn_saddr_v6 = ipv6_all_zeros;
634 		connp->conn_bound_addr_v6 = ipv6_all_zeros;
635 	}
636 	return (error);
637 }
638 
639 /*
640  * If the "bind_to_req_port_only" parameter is set, if the requested port
641  * number is available, return it, If not return 0
642  *
643  * If "bind_to_req_port_only" parameter is not set and
644  * If the requested port number is available, return it.  If not, return
645  * the first anonymous port we happen across.  If no anonymous ports are
646  * available, return 0. addr is the requested local address, if any.
647  *
648  * In either case, when succeeding update the tcp_t to record the port number
649  * and insert it in the bind hash table.
650  *
651  * Note that TCP over IPv4 and IPv6 sockets can use the same port number
652  * without setting SO_REUSEADDR. This is needed so that they
653  * can be viewed as two independent transport protocols.
654  */
655 in_port_t
656 tcp_bindi(tcp_t *tcp, in_port_t port, const in6_addr_t *laddr,
657     int reuseaddr, boolean_t quick_connect,
658     boolean_t bind_to_req_port_only, boolean_t user_specified)
659 {
660 	/* number of times we have run around the loop */
661 	int count = 0;
662 	/* maximum number of times to run around the loop */
663 	int loopmax;
664 	conn_t *connp = tcp->tcp_connp;
665 	tcp_stack_t	*tcps = tcp->tcp_tcps;
666 
667 	/*
668 	 * Lookup for free addresses is done in a loop and "loopmax"
669 	 * influences how long we spin in the loop
670 	 */
671 	if (bind_to_req_port_only) {
672 		/*
673 		 * If the requested port is busy, don't bother to look
674 		 * for a new one. Setting loop maximum count to 1 has
675 		 * that effect.
676 		 */
677 		loopmax = 1;
678 	} else {
679 		/*
680 		 * If the requested port is busy, look for a free one
681 		 * in the anonymous port range.
682 		 * Set loopmax appropriately so that one does not look
683 		 * forever in the case all of the anonymous ports are in use.
684 		 */
685 		if (connp->conn_anon_priv_bind) {
686 			/*
687 			 * loopmax =
688 			 * 	(IPPORT_RESERVED-1) - tcp_min_anonpriv_port + 1
689 			 */
690 			loopmax = IPPORT_RESERVED -
691 			    tcps->tcps_min_anonpriv_port;
692 		} else {
693 			loopmax = (tcps->tcps_largest_anon_port -
694 			    tcps->tcps_smallest_anon_port + 1);
695 		}
696 	}
697 	do {
698 		uint16_t	lport;
699 		tf_t		*tbf;
700 		tcp_t		*ltcp;
701 		conn_t		*lconnp;
702 
703 		lport = htons(port);
704 
705 		/*
706 		 * Ensure that the tcp_t is not currently in the bind hash.
707 		 * Hold the lock on the hash bucket to ensure that
708 		 * the duplicate check plus the insertion is an atomic
709 		 * operation.
710 		 *
711 		 * This function does an inline lookup on the bind hash list
712 		 * Make sure that we access only members of tcp_t
713 		 * and that we don't look at tcp_tcp, since we are not
714 		 * doing a CONN_INC_REF.
715 		 */
716 		tcp_bind_hash_remove(tcp);
717 		tbf = &tcps->tcps_bind_fanout[TCP_BIND_HASH(lport)];
718 		mutex_enter(&tbf->tf_lock);
719 		for (ltcp = tbf->tf_tcp; ltcp != NULL;
720 		    ltcp = ltcp->tcp_bind_hash) {
721 			if (lport == ltcp->tcp_connp->conn_lport)
722 				break;
723 		}
724 
725 		for (; ltcp != NULL; ltcp = ltcp->tcp_bind_hash_port) {
726 			boolean_t not_socket;
727 			boolean_t exclbind;
728 
729 			lconnp = ltcp->tcp_connp;
730 
731 			/*
732 			 * On a labeled system, we must treat bindings to ports
733 			 * on shared IP addresses by sockets with MAC exemption
734 			 * privilege as being in all zones, as there's
735 			 * otherwise no way to identify the right receiver.
736 			 */
737 			if (!IPCL_BIND_ZONE_MATCH(lconnp, connp))
738 				continue;
739 
740 			/*
741 			 * If TCP_EXCLBIND is set for either the bound or
742 			 * binding endpoint, the semantics of bind
743 			 * is changed according to the following.
744 			 *
745 			 * spec = specified address (v4 or v6)
746 			 * unspec = unspecified address (v4 or v6)
747 			 * A = specified addresses are different for endpoints
748 			 *
749 			 * bound	bind to		allowed
750 			 * -------------------------------------
751 			 * unspec	unspec		no
752 			 * unspec	spec		no
753 			 * spec		unspec		no
754 			 * spec		spec		yes if A
755 			 *
756 			 * For labeled systems, SO_MAC_EXEMPT behaves the same
757 			 * as TCP_EXCLBIND, except that zoneid is ignored.
758 			 *
759 			 * Note:
760 			 *
761 			 * 1. Because of TLI semantics, an endpoint can go
762 			 * back from, say TCP_ESTABLISHED to TCPS_LISTEN or
763 			 * TCPS_BOUND, depending on whether it is originally
764 			 * a listener or not.  That is why we need to check
765 			 * for states greater than or equal to TCPS_BOUND
766 			 * here.
767 			 *
768 			 * 2. Ideally, we should only check for state equals
769 			 * to TCPS_LISTEN. And the following check should be
770 			 * added.
771 			 *
772 			 * if (ltcp->tcp_state == TCPS_LISTEN ||
773 			 *	!reuseaddr || !lconnp->conn_reuseaddr) {
774 			 *		...
775 			 * }
776 			 *
777 			 * The semantics will be changed to this.  If the
778 			 * endpoint on the list is in state not equal to
779 			 * TCPS_LISTEN and both endpoints have SO_REUSEADDR
780 			 * set, let the bind succeed.
781 			 *
782 			 * Because of (1), we cannot do that for TLI
783 			 * endpoints.  But we can do that for socket endpoints.
784 			 * If in future, we can change this going back
785 			 * semantics, we can use the above check for TLI also.
786 			 */
787 			not_socket = !(TCP_IS_SOCKET(ltcp) &&
788 			    TCP_IS_SOCKET(tcp));
789 			exclbind = lconnp->conn_exclbind ||
790 			    connp->conn_exclbind;
791 
792 			if ((lconnp->conn_mac_mode != CONN_MAC_DEFAULT) ||
793 			    (connp->conn_mac_mode != CONN_MAC_DEFAULT) ||
794 			    (exclbind && (not_socket ||
795 			    ltcp->tcp_state <= TCPS_ESTABLISHED))) {
796 				if (V6_OR_V4_INADDR_ANY(
797 				    lconnp->conn_bound_addr_v6) ||
798 				    V6_OR_V4_INADDR_ANY(*laddr) ||
799 				    IN6_ARE_ADDR_EQUAL(laddr,
800 				    &lconnp->conn_bound_addr_v6)) {
801 					break;
802 				}
803 				continue;
804 			}
805 
806 			/*
807 			 * Check ipversion to allow IPv4 and IPv6 sockets to
808 			 * have disjoint port number spaces, if *_EXCLBIND
809 			 * is not set and only if the application binds to a
810 			 * specific port. We use the same autoassigned port
811 			 * number space for IPv4 and IPv6 sockets.
812 			 */
813 			if (connp->conn_ipversion != lconnp->conn_ipversion &&
814 			    bind_to_req_port_only)
815 				continue;
816 
817 			/*
818 			 * Ideally, we should make sure that the source
819 			 * address, remote address, and remote port in the
820 			 * four tuple for this tcp-connection is unique.
821 			 * However, trying to find out the local source
822 			 * address would require too much code duplication
823 			 * with IP, since IP needs needs to have that code
824 			 * to support userland TCP implementations.
825 			 */
826 			if (quick_connect &&
827 			    (ltcp->tcp_state > TCPS_LISTEN) &&
828 			    ((connp->conn_fport != lconnp->conn_fport) ||
829 			    !IN6_ARE_ADDR_EQUAL(&connp->conn_faddr_v6,
830 			    &lconnp->conn_faddr_v6)))
831 				continue;
832 
833 			if (!reuseaddr) {
834 				/*
835 				 * No socket option SO_REUSEADDR.
836 				 * If existing port is bound to
837 				 * a non-wildcard IP address
838 				 * and the requesting stream is
839 				 * bound to a distinct
840 				 * different IP addresses
841 				 * (non-wildcard, also), keep
842 				 * going.
843 				 */
844 				if (!V6_OR_V4_INADDR_ANY(*laddr) &&
845 				    !V6_OR_V4_INADDR_ANY(
846 				    lconnp->conn_bound_addr_v6) &&
847 				    !IN6_ARE_ADDR_EQUAL(laddr,
848 				    &lconnp->conn_bound_addr_v6))
849 					continue;
850 				if (ltcp->tcp_state >= TCPS_BOUND) {
851 					/*
852 					 * This port is being used and
853 					 * its state is >= TCPS_BOUND,
854 					 * so we can't bind to it.
855 					 */
856 					break;
857 				}
858 			} else {
859 				/*
860 				 * socket option SO_REUSEADDR is set on the
861 				 * binding tcp_t.
862 				 *
863 				 * If two streams are bound to
864 				 * same IP address or both addr
865 				 * and bound source are wildcards
866 				 * (INADDR_ANY), we want to stop
867 				 * searching.
868 				 * We have found a match of IP source
869 				 * address and source port, which is
870 				 * refused regardless of the
871 				 * SO_REUSEADDR setting, so we break.
872 				 */
873 				if (IN6_ARE_ADDR_EQUAL(laddr,
874 				    &lconnp->conn_bound_addr_v6) &&
875 				    (ltcp->tcp_state == TCPS_LISTEN ||
876 				    ltcp->tcp_state == TCPS_BOUND))
877 					break;
878 			}
879 		}
880 		if (ltcp != NULL) {
881 			/* The port number is busy */
882 			mutex_exit(&tbf->tf_lock);
883 		} else {
884 			/*
885 			 * This port is ours. Insert in fanout and mark as
886 			 * bound to prevent others from getting the port
887 			 * number.
888 			 */
889 			tcp->tcp_state = TCPS_BOUND;
890 			DTRACE_TCP6(state__change, void, NULL,
891 			    ip_xmit_attr_t *, connp->conn_ixa,
892 			    void, NULL, tcp_t *, tcp, void, NULL,
893 			    int32_t, TCPS_IDLE);
894 
895 			connp->conn_lport = htons(port);
896 
897 			ASSERT(&tcps->tcps_bind_fanout[TCP_BIND_HASH(
898 			    connp->conn_lport)] == tbf);
899 			tcp_bind_hash_insert(tbf, tcp, 1);
900 
901 			mutex_exit(&tbf->tf_lock);
902 
903 			/*
904 			 * We don't want tcp_next_port_to_try to "inherit"
905 			 * a port number supplied by the user in a bind.
906 			 */
907 			if (user_specified)
908 				return (port);
909 
910 			/*
911 			 * This is the only place where tcp_next_port_to_try
912 			 * is updated. After the update, it may or may not
913 			 * be in the valid range.
914 			 */
915 			if (!connp->conn_anon_priv_bind)
916 				tcps->tcps_next_port_to_try = port + 1;
917 			return (port);
918 		}
919 
920 		if (connp->conn_anon_priv_bind) {
921 			port = tcp_get_next_priv_port(tcp);
922 		} else {
923 			if (count == 0 && user_specified) {
924 				/*
925 				 * We may have to return an anonymous port. So
926 				 * get one to start with.
927 				 */
928 				port =
929 				    tcp_update_next_port(
930 				    tcps->tcps_next_port_to_try,
931 				    tcp, B_TRUE);
932 				user_specified = B_FALSE;
933 			} else {
934 				port = tcp_update_next_port(port + 1, tcp,
935 				    B_FALSE);
936 			}
937 		}
938 		if (port == 0)
939 			break;
940 
941 		/*
942 		 * Don't let this loop run forever in the case where
943 		 * all of the anonymous ports are in use.
944 		 */
945 	} while (++count < loopmax);
946 	return (0);
947 }
948