xref: /freebsd/sys/netinet/tcp_subr.c (revision e91d723ad446b5429318b24f4578a4f7a160f65e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/eventhandler.h>
46 #ifdef TCP_HHOOK
47 #include <sys/hhook.h>
48 #endif
49 #include <sys/kernel.h>
50 #ifdef TCP_HHOOK
51 #include <sys/khelp.h>
52 #endif
53 #include <sys/sysctl.h>
54 #include <sys/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/refcount.h>
57 #include <sys/mbuf.h>
58 #ifdef INET6
59 #include <sys/domain.h>
60 #endif
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/sdt.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/random.h>
68 
69 #include <vm/uma.h>
70 
71 #include <net/route.h>
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/vnet.h>
75 
76 #include <netinet/in.h>
77 #include <netinet/in_fib.h>
78 #include <netinet/in_kdtrace.h>
79 #include <netinet/in_pcb.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/in_var.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/ip_var.h>
85 #ifdef INET6
86 #include <netinet/icmp6.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_fib.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet6/nd6.h>
93 #endif
94 
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
98 #include <netinet/tcp_timer.h>
99 #include <netinet/tcp_var.h>
100 #include <netinet/tcp_log_buf.h>
101 #include <netinet/tcp_syncache.h>
102 #include <netinet/tcp_hpts.h>
103 #include <netinet/cc/cc.h>
104 #ifdef INET6
105 #include <netinet6/tcp6_var.h>
106 #endif
107 #include <netinet/tcpip.h>
108 #include <netinet/tcp_fastopen.h>
109 #ifdef TCPPCAP
110 #include <netinet/tcp_pcap.h>
111 #endif
112 #ifdef TCPDEBUG
113 #include <netinet/tcp_debug.h>
114 #endif
115 #ifdef INET6
116 #include <netinet6/ip6protosw.h>
117 #endif
118 #ifdef TCP_OFFLOAD
119 #include <netinet/tcp_offload.h>
120 #endif
121 
122 #include <netipsec/ipsec_support.h>
123 
124 #include <machine/in_cksum.h>
125 #include <sys/md5.h>
126 
127 #include <security/mac/mac_framework.h>
128 
129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 #ifdef INET6
131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
132 #endif
133 
134 struct rwlock tcp_function_lock;
135 
136 static int
137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
138 {
139 	int error, new;
140 
141 	new = V_tcp_mssdflt;
142 	error = sysctl_handle_int(oidp, &new, 0, req);
143 	if (error == 0 && req->newptr) {
144 		if (new < TCP_MINMSS)
145 			error = EINVAL;
146 		else
147 			V_tcp_mssdflt = new;
148 	}
149 	return (error);
150 }
151 
152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154     &sysctl_net_inet_tcp_mss_check, "I",
155     "Default TCP Maximum Segment Size");
156 
157 #ifdef INET6
158 static int
159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
160 {
161 	int error, new;
162 
163 	new = V_tcp_v6mssdflt;
164 	error = sysctl_handle_int(oidp, &new, 0, req);
165 	if (error == 0 && req->newptr) {
166 		if (new < TCP_MINMSS)
167 			error = EINVAL;
168 		else
169 			V_tcp_v6mssdflt = new;
170 	}
171 	return (error);
172 }
173 
174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176     &sysctl_net_inet_tcp_mss_v6_check, "I",
177    "Default TCP Maximum Segment Size for IPv6");
178 #endif /* INET6 */
179 
180 /*
181  * Minimum MSS we accept and use. This prevents DoS attacks where
182  * we are forced to a ridiculous low MSS like 20 and send hundreds
183  * of packets instead of one. The effect scales with the available
184  * bandwidth and quickly saturates the CPU and network interface
185  * with packet generation and sending. Set to zero to disable MINMSS
186  * checking. This setting prevents us from sending too small packets.
187  */
188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190      &VNET_NAME(tcp_minmss), 0,
191     "Minimum TCP Maximum Segment Size");
192 
193 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195     &VNET_NAME(tcp_do_rfc1323), 0,
196     "Enable rfc1323 (high performance TCP) extensions");
197 
198 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
200     &VNET_NAME(tcp_ts_offset_per_conn), 0,
201     "Initialize TCP timestamps per connection instead of per host pair");
202 
203 static int	tcp_log_debug = 0;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
205     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
206 
207 static int	tcp_tcbhashsize;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
209     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
210 
211 static int	do_tcpdrain = 1;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
213     "Enable tcp_drain routine for extra help when low on mbufs");
214 
215 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
216     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
217 
218 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
219 #define	V_icmp_may_rst			VNET(icmp_may_rst)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
221     &VNET_NAME(icmp_may_rst), 0,
222     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
223 
224 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
225 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
227     &VNET_NAME(tcp_isn_reseed_interval), 0,
228     "Seconds between reseeding of ISN secret");
229 
230 static int	tcp_soreceive_stream;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
232     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
233 
234 VNET_DEFINE(uma_zone_t, sack_hole_zone);
235 #define	V_sack_hole_zone		VNET(sack_hole_zone)
236 
237 #ifdef TCP_HHOOK
238 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
239 #endif
240 
241 #define TS_OFFSET_SECRET_LENGTH 32
242 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
243 #define	V_ts_offset_secret	VNET(ts_offset_secret)
244 
245 static int	tcp_default_fb_init(struct tcpcb *tp);
246 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
247 static int	tcp_default_handoff_ok(struct tcpcb *tp);
248 static struct inpcb *tcp_notify(struct inpcb *, int);
249 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
250 static void tcp_mtudisc(struct inpcb *, int);
251 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
252 		    void *ip4hdr, const void *ip6hdr);
253 
254 
255 static struct tcp_function_block tcp_def_funcblk = {
256 	.tfb_tcp_block_name = "freebsd",
257 	.tfb_tcp_output = tcp_output,
258 	.tfb_tcp_do_segment = tcp_do_segment,
259 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
260 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
261 	.tfb_tcp_fb_init = tcp_default_fb_init,
262 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
263 };
264 
265 static int tcp_fb_cnt = 0;
266 struct tcp_funchead t_functions;
267 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
268 
269 static struct tcp_function_block *
270 find_tcp_functions_locked(struct tcp_function_set *fs)
271 {
272 	struct tcp_function *f;
273 	struct tcp_function_block *blk=NULL;
274 
275 	TAILQ_FOREACH(f, &t_functions, tf_next) {
276 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
277 			blk = f->tf_fb;
278 			break;
279 		}
280 	}
281 	return(blk);
282 }
283 
284 static struct tcp_function_block *
285 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
286 {
287 	struct tcp_function_block *rblk=NULL;
288 	struct tcp_function *f;
289 
290 	TAILQ_FOREACH(f, &t_functions, tf_next) {
291 		if (f->tf_fb == blk) {
292 			rblk = blk;
293 			if (s) {
294 				*s = f;
295 			}
296 			break;
297 		}
298 	}
299 	return (rblk);
300 }
301 
302 struct tcp_function_block *
303 find_and_ref_tcp_functions(struct tcp_function_set *fs)
304 {
305 	struct tcp_function_block *blk;
306 
307 	rw_rlock(&tcp_function_lock);
308 	blk = find_tcp_functions_locked(fs);
309 	if (blk)
310 		refcount_acquire(&blk->tfb_refcnt);
311 	rw_runlock(&tcp_function_lock);
312 	return(blk);
313 }
314 
315 struct tcp_function_block *
316 find_and_ref_tcp_fb(struct tcp_function_block *blk)
317 {
318 	struct tcp_function_block *rblk;
319 
320 	rw_rlock(&tcp_function_lock);
321 	rblk = find_tcp_fb_locked(blk, NULL);
322 	if (rblk)
323 		refcount_acquire(&rblk->tfb_refcnt);
324 	rw_runlock(&tcp_function_lock);
325 	return(rblk);
326 }
327 
328 static struct tcp_function_block *
329 find_and_ref_tcp_default_fb(void)
330 {
331 	struct tcp_function_block *rblk;
332 
333 	rw_rlock(&tcp_function_lock);
334 	rblk = tcp_func_set_ptr;
335 	refcount_acquire(&rblk->tfb_refcnt);
336 	rw_runlock(&tcp_function_lock);
337 	return (rblk);
338 }
339 
340 void
341 tcp_switch_back_to_default(struct tcpcb *tp)
342 {
343 	struct tcp_function_block *tfb;
344 
345 	KASSERT(tp->t_fb != &tcp_def_funcblk,
346 	    ("%s: called by the built-in default stack", __func__));
347 
348 	/*
349 	 * Release the old stack. This function will either find a new one
350 	 * or panic.
351 	 */
352 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
353 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
354 	refcount_release(&tp->t_fb->tfb_refcnt);
355 
356 	/*
357 	 * Now, we'll find a new function block to use.
358 	 * Start by trying the current user-selected
359 	 * default, unless this stack is the user-selected
360 	 * default.
361 	 */
362 	tfb = find_and_ref_tcp_default_fb();
363 	if (tfb == tp->t_fb) {
364 		refcount_release(&tfb->tfb_refcnt);
365 		tfb = NULL;
366 	}
367 	/* Does the stack accept this connection? */
368 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
369 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
370 		refcount_release(&tfb->tfb_refcnt);
371 		tfb = NULL;
372 	}
373 	/* Try to use that stack. */
374 	if (tfb != NULL) {
375 		/* Initialize the new stack. If it succeeds, we are done. */
376 		tp->t_fb = tfb;
377 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
378 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
379 			return;
380 
381 		/*
382 		 * Initialization failed. Release the reference count on
383 		 * the stack.
384 		 */
385 		refcount_release(&tfb->tfb_refcnt);
386 	}
387 
388 	/*
389 	 * If that wasn't feasible, use the built-in default
390 	 * stack which is not allowed to reject anyone.
391 	 */
392 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
393 	if (tfb == NULL) {
394 		/* there always should be a default */
395 		panic("Can't refer to tcp_def_funcblk");
396 	}
397 	if (tfb->tfb_tcp_handoff_ok != NULL) {
398 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
399 			/* The default stack cannot say no */
400 			panic("Default stack rejects a new session?");
401 		}
402 	}
403 	tp->t_fb = tfb;
404 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
405 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
406 		/* The default stack cannot fail */
407 		panic("Default stack initialization failed");
408 	}
409 }
410 
411 static int
412 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
413 {
414 	int error=ENOENT;
415 	struct tcp_function_set fs;
416 	struct tcp_function_block *blk;
417 
418 	memset(&fs, 0, sizeof(fs));
419 	rw_rlock(&tcp_function_lock);
420 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
421 	if (blk) {
422 		/* Found him */
423 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
424 		fs.pcbcnt = blk->tfb_refcnt;
425 	}
426 	rw_runlock(&tcp_function_lock);
427 	error = sysctl_handle_string(oidp, fs.function_set_name,
428 				     sizeof(fs.function_set_name), req);
429 
430 	/* Check for error or no change */
431 	if (error != 0 || req->newptr == NULL)
432 		return(error);
433 
434 	rw_wlock(&tcp_function_lock);
435 	blk = find_tcp_functions_locked(&fs);
436 	if ((blk == NULL) ||
437 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
438 		error = ENOENT;
439 		goto done;
440 	}
441 	tcp_func_set_ptr = blk;
442 done:
443 	rw_wunlock(&tcp_function_lock);
444 	return (error);
445 }
446 
447 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
448 	    CTLTYPE_STRING | CTLFLAG_RW,
449 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
450 	    "Set/get the default TCP functions");
451 
452 static int
453 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
454 {
455 	int error, cnt, linesz;
456 	struct tcp_function *f;
457 	char *buffer, *cp;
458 	size_t bufsz, outsz;
459 	bool alias;
460 
461 	cnt = 0;
462 	rw_rlock(&tcp_function_lock);
463 	TAILQ_FOREACH(f, &t_functions, tf_next) {
464 		cnt++;
465 	}
466 	rw_runlock(&tcp_function_lock);
467 
468 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
469 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
470 
471 	error = 0;
472 	cp = buffer;
473 
474 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
475 	    "Alias", "PCB count");
476 	cp += linesz;
477 	bufsz -= linesz;
478 	outsz = linesz;
479 
480 	rw_rlock(&tcp_function_lock);
481 	TAILQ_FOREACH(f, &t_functions, tf_next) {
482 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
483 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
484 		    f->tf_fb->tfb_tcp_block_name,
485 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
486 		    alias ? f->tf_name : "-",
487 		    f->tf_fb->tfb_refcnt);
488 		if (linesz >= bufsz) {
489 			error = EOVERFLOW;
490 			break;
491 		}
492 		cp += linesz;
493 		bufsz -= linesz;
494 		outsz += linesz;
495 	}
496 	rw_runlock(&tcp_function_lock);
497 	if (error == 0)
498 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
499 	free(buffer, M_TEMP);
500 	return (error);
501 }
502 
503 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
504 	    CTLTYPE_STRING|CTLFLAG_RD,
505 	    NULL, 0, sysctl_net_inet_list_available, "A",
506 	    "list available TCP Function sets");
507 
508 /*
509  * Exports one (struct tcp_function_info) for each alias/name.
510  */
511 static int
512 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
513 {
514 	int cnt, error;
515 	struct tcp_function *f;
516 	struct tcp_function_info tfi;
517 
518 	/*
519 	 * We don't allow writes.
520 	 */
521 	if (req->newptr != NULL)
522 		return (EINVAL);
523 
524 	/*
525 	 * Wire the old buffer so we can directly copy the functions to
526 	 * user space without dropping the lock.
527 	 */
528 	if (req->oldptr != NULL) {
529 		error = sysctl_wire_old_buffer(req, 0);
530 		if (error)
531 			return (error);
532 	}
533 
534 	/*
535 	 * Walk the list and copy out matching entries. If INVARIANTS
536 	 * is compiled in, also walk the list to verify the length of
537 	 * the list matches what we have recorded.
538 	 */
539 	rw_rlock(&tcp_function_lock);
540 
541 	cnt = 0;
542 #ifndef INVARIANTS
543 	if (req->oldptr == NULL) {
544 		cnt = tcp_fb_cnt;
545 		goto skip_loop;
546 	}
547 #endif
548 	TAILQ_FOREACH(f, &t_functions, tf_next) {
549 #ifdef INVARIANTS
550 		cnt++;
551 #endif
552 		if (req->oldptr != NULL) {
553 			bzero(&tfi, sizeof(tfi));
554 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
555 			tfi.tfi_id = f->tf_fb->tfb_id;
556 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
557 			    sizeof(tfi.tfi_alias));
558 			(void)strlcpy(tfi.tfi_name,
559 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
560 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
561 			/*
562 			 * Don't stop on error, as that is the
563 			 * mechanism we use to accumulate length
564 			 * information if the buffer was too short.
565 			 */
566 		}
567 	}
568 	KASSERT(cnt == tcp_fb_cnt,
569 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
570 #ifndef INVARIANTS
571 skip_loop:
572 #endif
573 	rw_runlock(&tcp_function_lock);
574 	if (req->oldptr == NULL)
575 		error = SYSCTL_OUT(req, NULL,
576 		    (cnt + 1) * sizeof(struct tcp_function_info));
577 
578 	return (error);
579 }
580 
581 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
582 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
583 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
584 	    "List TCP function block name-to-ID mappings");
585 
586 /*
587  * tfb_tcp_handoff_ok() function for the default stack.
588  * Note that we'll basically try to take all comers.
589  */
590 static int
591 tcp_default_handoff_ok(struct tcpcb *tp)
592 {
593 
594 	return (0);
595 }
596 
597 /*
598  * tfb_tcp_fb_init() function for the default stack.
599  *
600  * This handles making sure we have appropriate timers set if you are
601  * transitioning a socket that has some amount of setup done.
602  *
603  * The init() fuction from the default can *never* return non-zero i.e.
604  * it is required to always succeed since it is the stack of last resort!
605  */
606 static int
607 tcp_default_fb_init(struct tcpcb *tp)
608 {
609 
610 	struct socket *so;
611 
612 	INP_WLOCK_ASSERT(tp->t_inpcb);
613 
614 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
615 	    ("%s: connection %p in unexpected state %d", __func__, tp,
616 	    tp->t_state));
617 
618 	/*
619 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
620 	 * know what to do for unexpected states (which includes TIME_WAIT).
621 	 */
622 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
623 		return (0);
624 
625 	/*
626 	 * Make sure some kind of transmission timer is set if there is
627 	 * outstanding data.
628 	 */
629 	so = tp->t_inpcb->inp_socket;
630 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
631 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
632 	    tcp_timer_active(tp, TT_PERSIST))) {
633 		/*
634 		 * If the session has established and it looks like it should
635 		 * be in the persist state, set the persist timer. Otherwise,
636 		 * set the retransmit timer.
637 		 */
638 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
639 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
640 		    (int32_t)sbavail(&so->so_snd))
641 			tcp_setpersist(tp);
642 		else
643 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
644 	}
645 
646 	/* All non-embryonic sessions get a keepalive timer. */
647 	if (!tcp_timer_active(tp, TT_KEEP))
648 		tcp_timer_activate(tp, TT_KEEP,
649 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
650 		    TP_KEEPINIT(tp));
651 
652 	return (0);
653 }
654 
655 /*
656  * tfb_tcp_fb_fini() function for the default stack.
657  *
658  * This changes state as necessary (or prudent) to prepare for another stack
659  * to assume responsibility for the connection.
660  */
661 static void
662 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
663 {
664 
665 	INP_WLOCK_ASSERT(tp->t_inpcb);
666 	return;
667 }
668 
669 /*
670  * Target size of TCP PCB hash tables. Must be a power of two.
671  *
672  * Note that this can be overridden by the kernel environment
673  * variable net.inet.tcp.tcbhashsize
674  */
675 #ifndef TCBHASHSIZE
676 #define TCBHASHSIZE	0
677 #endif
678 
679 /*
680  * XXX
681  * Callouts should be moved into struct tcp directly.  They are currently
682  * separate because the tcpcb structure is exported to userland for sysctl
683  * parsing purposes, which do not know about callouts.
684  */
685 struct tcpcb_mem {
686 	struct	tcpcb		tcb;
687 	struct	tcp_timer	tt;
688 	struct	cc_var		ccv;
689 #ifdef TCP_HHOOK
690 	struct	osd		osd;
691 #endif
692 };
693 
694 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
695 #define	V_tcpcb_zone			VNET(tcpcb_zone)
696 
697 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
698 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
699 
700 static struct mtx isn_mtx;
701 
702 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
703 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
704 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
705 
706 /*
707  * TCP initialization.
708  */
709 static void
710 tcp_zone_change(void *tag)
711 {
712 
713 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
714 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
715 	tcp_tw_zone_change();
716 }
717 
718 static int
719 tcp_inpcb_init(void *mem, int size, int flags)
720 {
721 	struct inpcb *inp = mem;
722 
723 	INP_LOCK_INIT(inp, "inp", "tcpinp");
724 	return (0);
725 }
726 
727 /*
728  * Take a value and get the next power of 2 that doesn't overflow.
729  * Used to size the tcp_inpcb hash buckets.
730  */
731 static int
732 maketcp_hashsize(int size)
733 {
734 	int hashsize;
735 
736 	/*
737 	 * auto tune.
738 	 * get the next power of 2 higher than maxsockets.
739 	 */
740 	hashsize = 1 << fls(size);
741 	/* catch overflow, and just go one power of 2 smaller */
742 	if (hashsize < size) {
743 		hashsize = 1 << (fls(size) - 1);
744 	}
745 	return (hashsize);
746 }
747 
748 static volatile int next_tcp_stack_id = 1;
749 
750 /*
751  * Register a TCP function block with the name provided in the names
752  * array.  (Note that this function does NOT automatically register
753  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
754  * explicitly include blk->tfb_tcp_block_name in the list of names if
755  * you wish to register the stack with that name.)
756  *
757  * Either all name registrations will succeed or all will fail.  If
758  * a name registration fails, the function will update the num_names
759  * argument to point to the array index of the name that encountered
760  * the failure.
761  *
762  * Returns 0 on success, or an error code on failure.
763  */
764 int
765 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
766     const char *names[], int *num_names)
767 {
768 	struct tcp_function *n;
769 	struct tcp_function_set fs;
770 	int error, i;
771 
772 	KASSERT(names != NULL && *num_names > 0,
773 	    ("%s: Called with 0-length name list", __func__));
774 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
775 	KASSERT(rw_initialized(&tcp_function_lock),
776 	    ("%s: called too early", __func__));
777 
778 	if ((blk->tfb_tcp_output == NULL) ||
779 	    (blk->tfb_tcp_do_segment == NULL) ||
780 	    (blk->tfb_tcp_ctloutput == NULL) ||
781 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
782 		/*
783 		 * These functions are required and you
784 		 * need a name.
785 		 */
786 		*num_names = 0;
787 		return (EINVAL);
788 	}
789 	if (blk->tfb_tcp_timer_stop_all ||
790 	    blk->tfb_tcp_timer_activate ||
791 	    blk->tfb_tcp_timer_active ||
792 	    blk->tfb_tcp_timer_stop) {
793 		/*
794 		 * If you define one timer function you
795 		 * must have them all.
796 		 */
797 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
798 		    (blk->tfb_tcp_timer_activate == NULL) ||
799 		    (blk->tfb_tcp_timer_active == NULL) ||
800 		    (blk->tfb_tcp_timer_stop == NULL)) {
801 			*num_names = 0;
802 			return (EINVAL);
803 		}
804 	}
805 
806 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
807 		*num_names = 0;
808 		return (EINVAL);
809 	}
810 
811 	refcount_init(&blk->tfb_refcnt, 0);
812 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
813 	for (i = 0; i < *num_names; i++) {
814 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
815 		if (n == NULL) {
816 			error = ENOMEM;
817 			goto cleanup;
818 		}
819 		n->tf_fb = blk;
820 
821 		(void)strlcpy(fs.function_set_name, names[i],
822 		    sizeof(fs.function_set_name));
823 		rw_wlock(&tcp_function_lock);
824 		if (find_tcp_functions_locked(&fs) != NULL) {
825 			/* Duplicate name space not allowed */
826 			rw_wunlock(&tcp_function_lock);
827 			free(n, M_TCPFUNCTIONS);
828 			error = EALREADY;
829 			goto cleanup;
830 		}
831 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
832 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
833 		tcp_fb_cnt++;
834 		rw_wunlock(&tcp_function_lock);
835 	}
836 	return(0);
837 
838 cleanup:
839 	/*
840 	 * Deregister the names we just added. Because registration failed
841 	 * for names[i], we don't need to deregister that name.
842 	 */
843 	*num_names = i;
844 	rw_wlock(&tcp_function_lock);
845 	while (--i >= 0) {
846 		TAILQ_FOREACH(n, &t_functions, tf_next) {
847 			if (!strncmp(n->tf_name, names[i],
848 			    TCP_FUNCTION_NAME_LEN_MAX)) {
849 				TAILQ_REMOVE(&t_functions, n, tf_next);
850 				tcp_fb_cnt--;
851 				n->tf_fb = NULL;
852 				free(n, M_TCPFUNCTIONS);
853 				break;
854 			}
855 		}
856 	}
857 	rw_wunlock(&tcp_function_lock);
858 	return (error);
859 }
860 
861 /*
862  * Register a TCP function block using the name provided in the name
863  * argument.
864  *
865  * Returns 0 on success, or an error code on failure.
866  */
867 int
868 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
869     int wait)
870 {
871 	const char *name_list[1];
872 	int num_names, rv;
873 
874 	num_names = 1;
875 	if (name != NULL)
876 		name_list[0] = name;
877 	else
878 		name_list[0] = blk->tfb_tcp_block_name;
879 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
880 	return (rv);
881 }
882 
883 /*
884  * Register a TCP function block using the name defined in
885  * blk->tfb_tcp_block_name.
886  *
887  * Returns 0 on success, or an error code on failure.
888  */
889 int
890 register_tcp_functions(struct tcp_function_block *blk, int wait)
891 {
892 
893 	return (register_tcp_functions_as_name(blk, NULL, wait));
894 }
895 
896 /*
897  * Deregister all names associated with a function block. This
898  * functionally removes the function block from use within the system.
899  *
900  * When called with a true quiesce argument, mark the function block
901  * as being removed so no more stacks will use it and determine
902  * whether the removal would succeed.
903  *
904  * When called with a false quiesce argument, actually attempt the
905  * removal.
906  *
907  * When called with a force argument, attempt to switch all TCBs to
908  * use the default stack instead of returning EBUSY.
909  *
910  * Returns 0 on success (or if the removal would succeed, or an error
911  * code on failure.
912  */
913 int
914 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
915     bool force)
916 {
917 	struct tcp_function *f;
918 
919 	if (blk == &tcp_def_funcblk) {
920 		/* You can't un-register the default */
921 		return (EPERM);
922 	}
923 	rw_wlock(&tcp_function_lock);
924 	if (blk == tcp_func_set_ptr) {
925 		/* You can't free the current default */
926 		rw_wunlock(&tcp_function_lock);
927 		return (EBUSY);
928 	}
929 	/* Mark the block so no more stacks can use it. */
930 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
931 	/*
932 	 * If TCBs are still attached to the stack, attempt to switch them
933 	 * to the default stack.
934 	 */
935 	if (force && blk->tfb_refcnt) {
936 		struct inpcb *inp;
937 		struct tcpcb *tp;
938 		VNET_ITERATOR_DECL(vnet_iter);
939 
940 		rw_wunlock(&tcp_function_lock);
941 
942 		VNET_LIST_RLOCK();
943 		VNET_FOREACH(vnet_iter) {
944 			CURVNET_SET(vnet_iter);
945 			INP_INFO_WLOCK(&V_tcbinfo);
946 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
947 				INP_WLOCK(inp);
948 				if (inp->inp_flags & INP_TIMEWAIT) {
949 					INP_WUNLOCK(inp);
950 					continue;
951 				}
952 				tp = intotcpcb(inp);
953 				if (tp == NULL || tp->t_fb != blk) {
954 					INP_WUNLOCK(inp);
955 					continue;
956 				}
957 				tcp_switch_back_to_default(tp);
958 				INP_WUNLOCK(inp);
959 			}
960 			INP_INFO_WUNLOCK(&V_tcbinfo);
961 			CURVNET_RESTORE();
962 		}
963 		VNET_LIST_RUNLOCK();
964 
965 		rw_wlock(&tcp_function_lock);
966 	}
967 	if (blk->tfb_refcnt) {
968 		/* TCBs still attached. */
969 		rw_wunlock(&tcp_function_lock);
970 		return (EBUSY);
971 	}
972 	if (quiesce) {
973 		/* Skip removal. */
974 		rw_wunlock(&tcp_function_lock);
975 		return (0);
976 	}
977 	/* Remove any function names that map to this function block. */
978 	while (find_tcp_fb_locked(blk, &f) != NULL) {
979 		TAILQ_REMOVE(&t_functions, f, tf_next);
980 		tcp_fb_cnt--;
981 		f->tf_fb = NULL;
982 		free(f, M_TCPFUNCTIONS);
983 	}
984 	rw_wunlock(&tcp_function_lock);
985 	return (0);
986 }
987 
988 void
989 tcp_init(void)
990 {
991 	const char *tcbhash_tuneable;
992 	int hashsize;
993 
994 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
995 
996 #ifdef TCP_HHOOK
997 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
998 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
999 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1000 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1001 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1002 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1003 #endif
1004 	hashsize = TCBHASHSIZE;
1005 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1006 	if (hashsize == 0) {
1007 		/*
1008 		 * Auto tune the hash size based on maxsockets.
1009 		 * A perfect hash would have a 1:1 mapping
1010 		 * (hashsize = maxsockets) however it's been
1011 		 * suggested that O(2) average is better.
1012 		 */
1013 		hashsize = maketcp_hashsize(maxsockets / 4);
1014 		/*
1015 		 * Our historical default is 512,
1016 		 * do not autotune lower than this.
1017 		 */
1018 		if (hashsize < 512)
1019 			hashsize = 512;
1020 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1021 			printf("%s: %s auto tuned to %d\n", __func__,
1022 			    tcbhash_tuneable, hashsize);
1023 	}
1024 	/*
1025 	 * We require a hashsize to be a power of two.
1026 	 * Previously if it was not a power of two we would just reset it
1027 	 * back to 512, which could be a nasty surprise if you did not notice
1028 	 * the error message.
1029 	 * Instead what we do is clip it to the closest power of two lower
1030 	 * than the specified hash value.
1031 	 */
1032 	if (!powerof2(hashsize)) {
1033 		int oldhashsize = hashsize;
1034 
1035 		hashsize = maketcp_hashsize(hashsize);
1036 		/* prevent absurdly low value */
1037 		if (hashsize < 16)
1038 			hashsize = 16;
1039 		printf("%s: WARNING: TCB hash size not a power of 2, "
1040 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1041 		    hashsize);
1042 	}
1043 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1044 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1045 
1046 	/*
1047 	 * These have to be type stable for the benefit of the timers.
1048 	 */
1049 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1050 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1051 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1052 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1053 
1054 	tcp_tw_init();
1055 	syncache_init();
1056 	tcp_hc_init();
1057 
1058 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1059 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1060 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1061 
1062 	tcp_fastopen_init();
1063 
1064 	/* Skip initialization of globals for non-default instances. */
1065 	if (!IS_DEFAULT_VNET(curvnet))
1066 		return;
1067 
1068 	tcp_reass_global_init();
1069 
1070 	/* XXX virtualize those bellow? */
1071 	tcp_delacktime = TCPTV_DELACK;
1072 	tcp_keepinit = TCPTV_KEEP_INIT;
1073 	tcp_keepidle = TCPTV_KEEP_IDLE;
1074 	tcp_keepintvl = TCPTV_KEEPINTVL;
1075 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1076 	tcp_msl = TCPTV_MSL;
1077 	tcp_rexmit_initial = TCPTV_RTOBASE;
1078 	if (tcp_rexmit_initial < 1)
1079 		tcp_rexmit_initial = 1;
1080 	tcp_rexmit_min = TCPTV_MIN;
1081 	if (tcp_rexmit_min < 1)
1082 		tcp_rexmit_min = 1;
1083 	tcp_persmin = TCPTV_PERSMIN;
1084 	tcp_persmax = TCPTV_PERSMAX;
1085 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1086 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1087 	tcp_tcbhashsize = hashsize;
1088 
1089 	/* Setup the tcp function block list */
1090 	TAILQ_INIT(&t_functions);
1091 	rw_init(&tcp_function_lock, "tcp_func_lock");
1092 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1093 #ifdef TCP_BLACKBOX
1094 	/* Initialize the TCP logging data. */
1095 	tcp_log_init();
1096 #endif
1097 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1098 
1099 	if (tcp_soreceive_stream) {
1100 #ifdef INET
1101 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1102 #endif
1103 #ifdef INET6
1104 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1105 #endif /* INET6 */
1106 	}
1107 
1108 #ifdef INET6
1109 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1110 #else /* INET6 */
1111 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1112 #endif /* INET6 */
1113 	if (max_protohdr < TCP_MINPROTOHDR)
1114 		max_protohdr = TCP_MINPROTOHDR;
1115 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1116 		panic("tcp_init");
1117 #undef TCP_MINPROTOHDR
1118 
1119 	ISN_LOCK_INIT();
1120 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1121 		SHUTDOWN_PRI_DEFAULT);
1122 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1123 		EVENTHANDLER_PRI_ANY);
1124 #ifdef TCPPCAP
1125 	tcp_pcap_init();
1126 #endif
1127 }
1128 
1129 #ifdef VIMAGE
1130 static void
1131 tcp_destroy(void *unused __unused)
1132 {
1133 	int n;
1134 #ifdef TCP_HHOOK
1135 	int error;
1136 #endif
1137 
1138 	/*
1139 	 * All our processes are gone, all our sockets should be cleaned
1140 	 * up, which means, we should be past the tcp_discardcb() calls.
1141 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1142 	 */
1143 	for (;;) {
1144 		INP_LIST_RLOCK(&V_tcbinfo);
1145 		n = V_tcbinfo.ipi_count;
1146 		INP_LIST_RUNLOCK(&V_tcbinfo);
1147 		if (n == 0)
1148 			break;
1149 		pause("tcpdes", hz / 10);
1150 	}
1151 	tcp_hc_destroy();
1152 	syncache_destroy();
1153 	tcp_tw_destroy();
1154 	in_pcbinfo_destroy(&V_tcbinfo);
1155 	/* tcp_discardcb() clears the sack_holes up. */
1156 	uma_zdestroy(V_sack_hole_zone);
1157 	uma_zdestroy(V_tcpcb_zone);
1158 
1159 	/*
1160 	 * Cannot free the zone until all tcpcbs are released as we attach
1161 	 * the allocations to them.
1162 	 */
1163 	tcp_fastopen_destroy();
1164 
1165 #ifdef TCP_HHOOK
1166 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1167 	if (error != 0) {
1168 		printf("%s: WARNING: unable to deregister helper hook "
1169 		    "type=%d, id=%d: error %d returned\n", __func__,
1170 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1171 	}
1172 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1173 	if (error != 0) {
1174 		printf("%s: WARNING: unable to deregister helper hook "
1175 		    "type=%d, id=%d: error %d returned\n", __func__,
1176 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1177 	}
1178 #endif
1179 }
1180 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1181 #endif
1182 
1183 void
1184 tcp_fini(void *xtp)
1185 {
1186 
1187 }
1188 
1189 /*
1190  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1191  * tcp_template used to store this data in mbufs, but we now recopy it out
1192  * of the tcpcb each time to conserve mbufs.
1193  */
1194 void
1195 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1196 {
1197 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1198 
1199 	INP_WLOCK_ASSERT(inp);
1200 
1201 #ifdef INET6
1202 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1203 		struct ip6_hdr *ip6;
1204 
1205 		ip6 = (struct ip6_hdr *)ip_ptr;
1206 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1207 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1208 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1209 			(IPV6_VERSION & IPV6_VERSION_MASK);
1210 		ip6->ip6_nxt = IPPROTO_TCP;
1211 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1212 		ip6->ip6_src = inp->in6p_laddr;
1213 		ip6->ip6_dst = inp->in6p_faddr;
1214 	}
1215 #endif /* INET6 */
1216 #if defined(INET6) && defined(INET)
1217 	else
1218 #endif
1219 #ifdef INET
1220 	{
1221 		struct ip *ip;
1222 
1223 		ip = (struct ip *)ip_ptr;
1224 		ip->ip_v = IPVERSION;
1225 		ip->ip_hl = 5;
1226 		ip->ip_tos = inp->inp_ip_tos;
1227 		ip->ip_len = 0;
1228 		ip->ip_id = 0;
1229 		ip->ip_off = 0;
1230 		ip->ip_ttl = inp->inp_ip_ttl;
1231 		ip->ip_sum = 0;
1232 		ip->ip_p = IPPROTO_TCP;
1233 		ip->ip_src = inp->inp_laddr;
1234 		ip->ip_dst = inp->inp_faddr;
1235 	}
1236 #endif /* INET */
1237 	th->th_sport = inp->inp_lport;
1238 	th->th_dport = inp->inp_fport;
1239 	th->th_seq = 0;
1240 	th->th_ack = 0;
1241 	th->th_x2 = 0;
1242 	th->th_off = 5;
1243 	th->th_flags = 0;
1244 	th->th_win = 0;
1245 	th->th_urp = 0;
1246 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1247 }
1248 
1249 /*
1250  * Create template to be used to send tcp packets on a connection.
1251  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1252  * use for this function is in keepalives, which use tcp_respond.
1253  */
1254 struct tcptemp *
1255 tcpip_maketemplate(struct inpcb *inp)
1256 {
1257 	struct tcptemp *t;
1258 
1259 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1260 	if (t == NULL)
1261 		return (NULL);
1262 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1263 	return (t);
1264 }
1265 
1266 /*
1267  * Send a single message to the TCP at address specified by
1268  * the given TCP/IP header.  If m == NULL, then we make a copy
1269  * of the tcpiphdr at th and send directly to the addressed host.
1270  * This is used to force keep alive messages out using the TCP
1271  * template for a connection.  If flags are given then we send
1272  * a message back to the TCP which originated the segment th,
1273  * and discard the mbuf containing it and any other attached mbufs.
1274  *
1275  * In any case the ack and sequence number of the transmitted
1276  * segment are as specified by the parameters.
1277  *
1278  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1279  */
1280 void
1281 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1282     tcp_seq ack, tcp_seq seq, int flags)
1283 {
1284 	struct tcpopt to;
1285 	struct inpcb *inp;
1286 	struct ip *ip;
1287 	struct mbuf *optm;
1288 	struct tcphdr *nth;
1289 	u_char *optp;
1290 #ifdef INET6
1291 	struct ip6_hdr *ip6;
1292 	int isipv6;
1293 #endif /* INET6 */
1294 	int optlen, tlen, win;
1295 	bool incl_opts;
1296 
1297 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1298 
1299 #ifdef INET6
1300 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1301 	ip6 = ipgen;
1302 #endif /* INET6 */
1303 	ip = ipgen;
1304 
1305 	if (tp != NULL) {
1306 		inp = tp->t_inpcb;
1307 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1308 		INP_WLOCK_ASSERT(inp);
1309 	} else
1310 		inp = NULL;
1311 
1312 	incl_opts = false;
1313 	win = 0;
1314 	if (tp != NULL) {
1315 		if (!(flags & TH_RST)) {
1316 			win = sbspace(&inp->inp_socket->so_rcv);
1317 			if (win > TCP_MAXWIN << tp->rcv_scale)
1318 				win = TCP_MAXWIN << tp->rcv_scale;
1319 		}
1320 		if ((tp->t_flags & TF_NOOPT) == 0)
1321 			incl_opts = true;
1322 	}
1323 	if (m == NULL) {
1324 		m = m_gethdr(M_NOWAIT, MT_DATA);
1325 		if (m == NULL)
1326 			return;
1327 		m->m_data += max_linkhdr;
1328 #ifdef INET6
1329 		if (isipv6) {
1330 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1331 			      sizeof(struct ip6_hdr));
1332 			ip6 = mtod(m, struct ip6_hdr *);
1333 			nth = (struct tcphdr *)(ip6 + 1);
1334 		} else
1335 #endif /* INET6 */
1336 		{
1337 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1338 			ip = mtod(m, struct ip *);
1339 			nth = (struct tcphdr *)(ip + 1);
1340 		}
1341 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1342 		flags = TH_ACK;
1343 	} else if (!M_WRITABLE(m)) {
1344 		struct mbuf *n;
1345 
1346 		/* Can't reuse 'm', allocate a new mbuf. */
1347 		n = m_gethdr(M_NOWAIT, MT_DATA);
1348 		if (n == NULL) {
1349 			m_freem(m);
1350 			return;
1351 		}
1352 
1353 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1354 			m_freem(m);
1355 			m_freem(n);
1356 			return;
1357 		}
1358 
1359 		n->m_data += max_linkhdr;
1360 		/* m_len is set later */
1361 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1362 #ifdef INET6
1363 		if (isipv6) {
1364 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1365 			      sizeof(struct ip6_hdr));
1366 			ip6 = mtod(n, struct ip6_hdr *);
1367 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1368 			nth = (struct tcphdr *)(ip6 + 1);
1369 		} else
1370 #endif /* INET6 */
1371 		{
1372 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1373 			ip = mtod(n, struct ip *);
1374 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1375 			nth = (struct tcphdr *)(ip + 1);
1376 		}
1377 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1378 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1379 		th = nth;
1380 		m_freem(m);
1381 		m = n;
1382 	} else {
1383 		/*
1384 		 *  reuse the mbuf.
1385 		 * XXX MRT We inherit the FIB, which is lucky.
1386 		 */
1387 		m_freem(m->m_next);
1388 		m->m_next = NULL;
1389 		m->m_data = (caddr_t)ipgen;
1390 		/* m_len is set later */
1391 #ifdef INET6
1392 		if (isipv6) {
1393 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1394 			nth = (struct tcphdr *)(ip6 + 1);
1395 		} else
1396 #endif /* INET6 */
1397 		{
1398 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1399 			nth = (struct tcphdr *)(ip + 1);
1400 		}
1401 		if (th != nth) {
1402 			/*
1403 			 * this is usually a case when an extension header
1404 			 * exists between the IPv6 header and the
1405 			 * TCP header.
1406 			 */
1407 			nth->th_sport = th->th_sport;
1408 			nth->th_dport = th->th_dport;
1409 		}
1410 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1411 #undef xchg
1412 	}
1413 	tlen = 0;
1414 #ifdef INET6
1415 	if (isipv6)
1416 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1417 #endif
1418 #if defined(INET) && defined(INET6)
1419 	else
1420 #endif
1421 #ifdef INET
1422 		tlen = sizeof (struct tcpiphdr);
1423 #endif
1424 #ifdef INVARIANTS
1425 	m->m_len = 0;
1426 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1427 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1428 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1429 #endif
1430 	m->m_len = tlen;
1431 	to.to_flags = 0;
1432 	if (incl_opts) {
1433 		/* Make sure we have room. */
1434 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1435 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1436 			if (m->m_next) {
1437 				optp = mtod(m->m_next, u_char *);
1438 				optm = m->m_next;
1439 			} else
1440 				incl_opts = false;
1441 		} else {
1442 			optp = (u_char *) (nth + 1);
1443 			optm = m;
1444 		}
1445 	}
1446 	if (incl_opts) {
1447 		/* Timestamps. */
1448 		if (tp->t_flags & TF_RCVD_TSTMP) {
1449 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1450 			to.to_tsecr = tp->ts_recent;
1451 			to.to_flags |= TOF_TS;
1452 		}
1453 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1454 		/* TCP-MD5 (RFC2385). */
1455 		if (tp->t_flags & TF_SIGNATURE)
1456 			to.to_flags |= TOF_SIGNATURE;
1457 #endif
1458 		/* Add the options. */
1459 		tlen += optlen = tcp_addoptions(&to, optp);
1460 
1461 		/* Update m_len in the correct mbuf. */
1462 		optm->m_len += optlen;
1463 	} else
1464 		optlen = 0;
1465 #ifdef INET6
1466 	if (isipv6) {
1467 		ip6->ip6_flow = 0;
1468 		ip6->ip6_vfc = IPV6_VERSION;
1469 		ip6->ip6_nxt = IPPROTO_TCP;
1470 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1471 	}
1472 #endif
1473 #if defined(INET) && defined(INET6)
1474 	else
1475 #endif
1476 #ifdef INET
1477 	{
1478 		ip->ip_len = htons(tlen);
1479 		ip->ip_ttl = V_ip_defttl;
1480 		if (V_path_mtu_discovery)
1481 			ip->ip_off |= htons(IP_DF);
1482 	}
1483 #endif
1484 	m->m_pkthdr.len = tlen;
1485 	m->m_pkthdr.rcvif = NULL;
1486 #ifdef MAC
1487 	if (inp != NULL) {
1488 		/*
1489 		 * Packet is associated with a socket, so allow the
1490 		 * label of the response to reflect the socket label.
1491 		 */
1492 		INP_WLOCK_ASSERT(inp);
1493 		mac_inpcb_create_mbuf(inp, m);
1494 	} else {
1495 		/*
1496 		 * Packet is not associated with a socket, so possibly
1497 		 * update the label in place.
1498 		 */
1499 		mac_netinet_tcp_reply(m);
1500 	}
1501 #endif
1502 	nth->th_seq = htonl(seq);
1503 	nth->th_ack = htonl(ack);
1504 	nth->th_x2 = 0;
1505 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1506 	nth->th_flags = flags;
1507 	if (tp != NULL)
1508 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1509 	else
1510 		nth->th_win = htons((u_short)win);
1511 	nth->th_urp = 0;
1512 
1513 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1514 	if (to.to_flags & TOF_SIGNATURE) {
1515 		if (!TCPMD5_ENABLED() ||
1516 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1517 			m_freem(m);
1518 			return;
1519 		}
1520 	}
1521 #endif
1522 
1523 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1524 #ifdef INET6
1525 	if (isipv6) {
1526 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1527 		nth->th_sum = in6_cksum_pseudo(ip6,
1528 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1529 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1530 		    NULL, NULL);
1531 	}
1532 #endif /* INET6 */
1533 #if defined(INET6) && defined(INET)
1534 	else
1535 #endif
1536 #ifdef INET
1537 	{
1538 		m->m_pkthdr.csum_flags = CSUM_TCP;
1539 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1540 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1541 	}
1542 #endif /* INET */
1543 #ifdef TCPDEBUG
1544 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1545 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1546 #endif
1547 	TCP_PROBE3(debug__output, tp, th, m);
1548 	if (flags & TH_RST)
1549 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1550 
1551 #ifdef INET6
1552 	if (isipv6) {
1553 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1554 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1555 	}
1556 #endif /* INET6 */
1557 #if defined(INET) && defined(INET6)
1558 	else
1559 #endif
1560 #ifdef INET
1561 	{
1562 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1563 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1564 	}
1565 #endif
1566 }
1567 
1568 /*
1569  * Create a new TCP control block, making an
1570  * empty reassembly queue and hooking it to the argument
1571  * protocol control block.  The `inp' parameter must have
1572  * come from the zone allocator set up in tcp_init().
1573  */
1574 struct tcpcb *
1575 tcp_newtcpcb(struct inpcb *inp)
1576 {
1577 	struct tcpcb_mem *tm;
1578 	struct tcpcb *tp;
1579 #ifdef INET6
1580 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1581 #endif /* INET6 */
1582 
1583 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1584 	if (tm == NULL)
1585 		return (NULL);
1586 	tp = &tm->tcb;
1587 
1588 	/* Initialise cc_var struct for this tcpcb. */
1589 	tp->ccv = &tm->ccv;
1590 	tp->ccv->type = IPPROTO_TCP;
1591 	tp->ccv->ccvc.tcp = tp;
1592 	rw_rlock(&tcp_function_lock);
1593 	tp->t_fb = tcp_func_set_ptr;
1594 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1595 	rw_runlock(&tcp_function_lock);
1596 	/*
1597 	 * Use the current system default CC algorithm.
1598 	 */
1599 	CC_LIST_RLOCK();
1600 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1601 	CC_ALGO(tp) = CC_DEFAULT();
1602 	CC_LIST_RUNLOCK();
1603 
1604 	if (CC_ALGO(tp)->cb_init != NULL)
1605 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1606 			if (tp->t_fb->tfb_tcp_fb_fini)
1607 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1608 			refcount_release(&tp->t_fb->tfb_refcnt);
1609 			uma_zfree(V_tcpcb_zone, tm);
1610 			return (NULL);
1611 		}
1612 
1613 #ifdef TCP_HHOOK
1614 	tp->osd = &tm->osd;
1615 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1616 		if (tp->t_fb->tfb_tcp_fb_fini)
1617 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1618 		refcount_release(&tp->t_fb->tfb_refcnt);
1619 		uma_zfree(V_tcpcb_zone, tm);
1620 		return (NULL);
1621 	}
1622 #endif
1623 
1624 #ifdef VIMAGE
1625 	tp->t_vnet = inp->inp_vnet;
1626 #endif
1627 	tp->t_timers = &tm->tt;
1628 	TAILQ_INIT(&tp->t_segq);
1629 	tp->t_maxseg =
1630 #ifdef INET6
1631 		isipv6 ? V_tcp_v6mssdflt :
1632 #endif /* INET6 */
1633 		V_tcp_mssdflt;
1634 
1635 	/* Set up our timeouts. */
1636 	callout_init(&tp->t_timers->tt_rexmt, 1);
1637 	callout_init(&tp->t_timers->tt_persist, 1);
1638 	callout_init(&tp->t_timers->tt_keep, 1);
1639 	callout_init(&tp->t_timers->tt_2msl, 1);
1640 	callout_init(&tp->t_timers->tt_delack, 1);
1641 
1642 	if (V_tcp_do_rfc1323)
1643 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1644 	if (V_tcp_do_sack)
1645 		tp->t_flags |= TF_SACK_PERMIT;
1646 	TAILQ_INIT(&tp->snd_holes);
1647 	/*
1648 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1649 	 * is called.
1650 	 */
1651 	in_pcbref(inp);	/* Reference for tcpcb */
1652 	tp->t_inpcb = inp;
1653 
1654 	/*
1655 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1656 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1657 	 * reasonable initial retransmit time.
1658 	 */
1659 	tp->t_srtt = TCPTV_SRTTBASE;
1660 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1661 	tp->t_rttmin = tcp_rexmit_min;
1662 	tp->t_rxtcur = tcp_rexmit_initial;
1663 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1664 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1665 	tp->t_rcvtime = ticks;
1666 	/*
1667 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1668 	 * because the socket may be bound to an IPv6 wildcard address,
1669 	 * which may match an IPv4-mapped IPv6 address.
1670 	 */
1671 	inp->inp_ip_ttl = V_ip_defttl;
1672 	inp->inp_ppcb = tp;
1673 #ifdef TCPPCAP
1674 	/*
1675 	 * Init the TCP PCAP queues.
1676 	 */
1677 	tcp_pcap_tcpcb_init(tp);
1678 #endif
1679 #ifdef TCP_BLACKBOX
1680 	/* Initialize the per-TCPCB log data. */
1681 	tcp_log_tcpcbinit(tp);
1682 #endif
1683 	if (tp->t_fb->tfb_tcp_fb_init) {
1684 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1685 	}
1686 	return (tp);		/* XXX */
1687 }
1688 
1689 /*
1690  * Switch the congestion control algorithm back to NewReno for any active
1691  * control blocks using an algorithm which is about to go away.
1692  * This ensures the CC framework can allow the unload to proceed without leaving
1693  * any dangling pointers which would trigger a panic.
1694  * Returning non-zero would inform the CC framework that something went wrong
1695  * and it would be unsafe to allow the unload to proceed. However, there is no
1696  * way for this to occur with this implementation so we always return zero.
1697  */
1698 int
1699 tcp_ccalgounload(struct cc_algo *unload_algo)
1700 {
1701 	struct cc_algo *tmpalgo;
1702 	struct inpcb *inp;
1703 	struct tcpcb *tp;
1704 	VNET_ITERATOR_DECL(vnet_iter);
1705 
1706 	/*
1707 	 * Check all active control blocks across all network stacks and change
1708 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1709 	 * requires cleanup code to be run, call it.
1710 	 */
1711 	VNET_LIST_RLOCK();
1712 	VNET_FOREACH(vnet_iter) {
1713 		CURVNET_SET(vnet_iter);
1714 		INP_INFO_WLOCK(&V_tcbinfo);
1715 		/*
1716 		 * New connections already part way through being initialised
1717 		 * with the CC algo we're removing will not race with this code
1718 		 * because the INP_INFO_WLOCK is held during initialisation. We
1719 		 * therefore don't enter the loop below until the connection
1720 		 * list has stabilised.
1721 		 */
1722 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1723 			INP_WLOCK(inp);
1724 			/* Important to skip tcptw structs. */
1725 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1726 			    (tp = intotcpcb(inp)) != NULL) {
1727 				/*
1728 				 * By holding INP_WLOCK here, we are assured
1729 				 * that the connection is not currently
1730 				 * executing inside the CC module's functions
1731 				 * i.e. it is safe to make the switch back to
1732 				 * NewReno.
1733 				 */
1734 				if (CC_ALGO(tp) == unload_algo) {
1735 					tmpalgo = CC_ALGO(tp);
1736 					if (tmpalgo->cb_destroy != NULL)
1737 						tmpalgo->cb_destroy(tp->ccv);
1738 					CC_DATA(tp) = NULL;
1739 					/*
1740 					 * NewReno may allocate memory on
1741 					 * demand for certain stateful
1742 					 * configuration as needed, but is
1743 					 * coded to never fail on memory
1744 					 * allocation failure so it is a safe
1745 					 * fallback.
1746 					 */
1747 					CC_ALGO(tp) = &newreno_cc_algo;
1748 				}
1749 			}
1750 			INP_WUNLOCK(inp);
1751 		}
1752 		INP_INFO_WUNLOCK(&V_tcbinfo);
1753 		CURVNET_RESTORE();
1754 	}
1755 	VNET_LIST_RUNLOCK();
1756 
1757 	return (0);
1758 }
1759 
1760 /*
1761  * Drop a TCP connection, reporting
1762  * the specified error.  If connection is synchronized,
1763  * then send a RST to peer.
1764  */
1765 struct tcpcb *
1766 tcp_drop(struct tcpcb *tp, int errno)
1767 {
1768 	struct socket *so = tp->t_inpcb->inp_socket;
1769 
1770 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1771 	INP_WLOCK_ASSERT(tp->t_inpcb);
1772 
1773 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1774 		tcp_state_change(tp, TCPS_CLOSED);
1775 		(void) tp->t_fb->tfb_tcp_output(tp);
1776 		TCPSTAT_INC(tcps_drops);
1777 	} else
1778 		TCPSTAT_INC(tcps_conndrops);
1779 	if (errno == ETIMEDOUT && tp->t_softerror)
1780 		errno = tp->t_softerror;
1781 	so->so_error = errno;
1782 	return (tcp_close(tp));
1783 }
1784 
1785 void
1786 tcp_discardcb(struct tcpcb *tp)
1787 {
1788 	struct inpcb *inp = tp->t_inpcb;
1789 	struct socket *so = inp->inp_socket;
1790 #ifdef INET6
1791 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1792 #endif /* INET6 */
1793 	int released __unused;
1794 
1795 	INP_WLOCK_ASSERT(inp);
1796 
1797 	/*
1798 	 * Make sure that all of our timers are stopped before we delete the
1799 	 * PCB.
1800 	 *
1801 	 * If stopping a timer fails, we schedule a discard function in same
1802 	 * callout, and the last discard function called will take care of
1803 	 * deleting the tcpcb.
1804 	 */
1805 	tp->t_timers->tt_draincnt = 0;
1806 	tcp_timer_stop(tp, TT_REXMT);
1807 	tcp_timer_stop(tp, TT_PERSIST);
1808 	tcp_timer_stop(tp, TT_KEEP);
1809 	tcp_timer_stop(tp, TT_2MSL);
1810 	tcp_timer_stop(tp, TT_DELACK);
1811 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1812 		/*
1813 		 * Call the stop-all function of the methods,
1814 		 * this function should call the tcp_timer_stop()
1815 		 * method with each of the function specific timeouts.
1816 		 * That stop will be called via the tfb_tcp_timer_stop()
1817 		 * which should use the async drain function of the
1818 		 * callout system (see tcp_var.h).
1819 		 */
1820 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1821 	}
1822 
1823 	/*
1824 	 * If we got enough samples through the srtt filter,
1825 	 * save the rtt and rttvar in the routing entry.
1826 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1827 	 * 4 samples is enough for the srtt filter to converge
1828 	 * to within enough % of the correct value; fewer samples
1829 	 * and we could save a bogus rtt. The danger is not high
1830 	 * as tcp quickly recovers from everything.
1831 	 * XXX: Works very well but needs some more statistics!
1832 	 */
1833 	if (tp->t_rttupdated >= 4) {
1834 		struct hc_metrics_lite metrics;
1835 		uint32_t ssthresh;
1836 
1837 		bzero(&metrics, sizeof(metrics));
1838 		/*
1839 		 * Update the ssthresh always when the conditions below
1840 		 * are satisfied. This gives us better new start value
1841 		 * for the congestion avoidance for new connections.
1842 		 * ssthresh is only set if packet loss occurred on a session.
1843 		 *
1844 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1845 		 * being torn down.  Ideally this code would not use 'so'.
1846 		 */
1847 		ssthresh = tp->snd_ssthresh;
1848 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1849 			/*
1850 			 * convert the limit from user data bytes to
1851 			 * packets then to packet data bytes.
1852 			 */
1853 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1854 			if (ssthresh < 2)
1855 				ssthresh = 2;
1856 			ssthresh *= (tp->t_maxseg +
1857 #ifdef INET6
1858 			    (isipv6 ? sizeof (struct ip6_hdr) +
1859 				sizeof (struct tcphdr) :
1860 #endif
1861 				sizeof (struct tcpiphdr)
1862 #ifdef INET6
1863 			    )
1864 #endif
1865 			    );
1866 		} else
1867 			ssthresh = 0;
1868 		metrics.rmx_ssthresh = ssthresh;
1869 
1870 		metrics.rmx_rtt = tp->t_srtt;
1871 		metrics.rmx_rttvar = tp->t_rttvar;
1872 		metrics.rmx_cwnd = tp->snd_cwnd;
1873 		metrics.rmx_sendpipe = 0;
1874 		metrics.rmx_recvpipe = 0;
1875 
1876 		tcp_hc_update(&inp->inp_inc, &metrics);
1877 	}
1878 
1879 	/* free the reassembly queue, if any */
1880 	tcp_reass_flush(tp);
1881 
1882 #ifdef TCP_OFFLOAD
1883 	/* Disconnect offload device, if any. */
1884 	if (tp->t_flags & TF_TOE)
1885 		tcp_offload_detach(tp);
1886 #endif
1887 
1888 	tcp_free_sackholes(tp);
1889 
1890 #ifdef TCPPCAP
1891 	/* Free the TCP PCAP queues. */
1892 	tcp_pcap_drain(&(tp->t_inpkts));
1893 	tcp_pcap_drain(&(tp->t_outpkts));
1894 #endif
1895 
1896 	/* Allow the CC algorithm to clean up after itself. */
1897 	if (CC_ALGO(tp)->cb_destroy != NULL)
1898 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1899 	CC_DATA(tp) = NULL;
1900 
1901 #ifdef TCP_HHOOK
1902 	khelp_destroy_osd(tp->osd);
1903 #endif
1904 
1905 	CC_ALGO(tp) = NULL;
1906 	inp->inp_ppcb = NULL;
1907 	if (tp->t_timers->tt_draincnt == 0) {
1908 		/* We own the last reference on tcpcb, let's free it. */
1909 #ifdef TCP_BLACKBOX
1910 		tcp_log_tcpcbfini(tp);
1911 #endif
1912 		TCPSTATES_DEC(tp->t_state);
1913 		if (tp->t_fb->tfb_tcp_fb_fini)
1914 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1915 		refcount_release(&tp->t_fb->tfb_refcnt);
1916 		tp->t_inpcb = NULL;
1917 		uma_zfree(V_tcpcb_zone, tp);
1918 		released = in_pcbrele_wlocked(inp);
1919 		KASSERT(!released, ("%s: inp %p should not have been released "
1920 			"here", __func__, inp));
1921 	}
1922 }
1923 
1924 void
1925 tcp_timer_discard(void *ptp)
1926 {
1927 	struct inpcb *inp;
1928 	struct tcpcb *tp;
1929 	struct epoch_tracker et;
1930 
1931 	tp = (struct tcpcb *)ptp;
1932 	CURVNET_SET(tp->t_vnet);
1933 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1934 	inp = tp->t_inpcb;
1935 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1936 		__func__, tp));
1937 	INP_WLOCK(inp);
1938 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1939 		("%s: tcpcb has to be stopped here", __func__));
1940 	tp->t_timers->tt_draincnt--;
1941 	if (tp->t_timers->tt_draincnt == 0) {
1942 		/* We own the last reference on this tcpcb, let's free it. */
1943 #ifdef TCP_BLACKBOX
1944 		tcp_log_tcpcbfini(tp);
1945 #endif
1946 		TCPSTATES_DEC(tp->t_state);
1947 		if (tp->t_fb->tfb_tcp_fb_fini)
1948 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1949 		refcount_release(&tp->t_fb->tfb_refcnt);
1950 		tp->t_inpcb = NULL;
1951 		uma_zfree(V_tcpcb_zone, tp);
1952 		if (in_pcbrele_wlocked(inp)) {
1953 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1954 			CURVNET_RESTORE();
1955 			return;
1956 		}
1957 	}
1958 	INP_WUNLOCK(inp);
1959 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1960 	CURVNET_RESTORE();
1961 }
1962 
1963 /*
1964  * Attempt to close a TCP control block, marking it as dropped, and freeing
1965  * the socket if we hold the only reference.
1966  */
1967 struct tcpcb *
1968 tcp_close(struct tcpcb *tp)
1969 {
1970 	struct inpcb *inp = tp->t_inpcb;
1971 	struct socket *so;
1972 
1973 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1974 	INP_WLOCK_ASSERT(inp);
1975 
1976 #ifdef TCP_OFFLOAD
1977 	if (tp->t_state == TCPS_LISTEN)
1978 		tcp_offload_listen_stop(tp);
1979 #endif
1980 	/*
1981 	 * This releases the TFO pending counter resource for TFO listen
1982 	 * sockets as well as passively-created TFO sockets that transition
1983 	 * from SYN_RECEIVED to CLOSED.
1984 	 */
1985 	if (tp->t_tfo_pending) {
1986 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1987 		tp->t_tfo_pending = NULL;
1988 	}
1989 	in_pcbdrop(inp);
1990 	TCPSTAT_INC(tcps_closed);
1991 	if (tp->t_state != TCPS_CLOSED)
1992 		tcp_state_change(tp, TCPS_CLOSED);
1993 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1994 	so = inp->inp_socket;
1995 	soisdisconnected(so);
1996 	if (inp->inp_flags & INP_SOCKREF) {
1997 		KASSERT(so->so_state & SS_PROTOREF,
1998 		    ("tcp_close: !SS_PROTOREF"));
1999 		inp->inp_flags &= ~INP_SOCKREF;
2000 		INP_WUNLOCK(inp);
2001 		SOCK_LOCK(so);
2002 		so->so_state &= ~SS_PROTOREF;
2003 		sofree(so);
2004 		return (NULL);
2005 	}
2006 	return (tp);
2007 }
2008 
2009 void
2010 tcp_drain(void)
2011 {
2012 	VNET_ITERATOR_DECL(vnet_iter);
2013 
2014 	if (!do_tcpdrain)
2015 		return;
2016 
2017 	VNET_LIST_RLOCK_NOSLEEP();
2018 	VNET_FOREACH(vnet_iter) {
2019 		CURVNET_SET(vnet_iter);
2020 		struct inpcb *inpb;
2021 		struct tcpcb *tcpb;
2022 
2023 	/*
2024 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2025 	 * if there is one...
2026 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2027 	 *      reassembly queue should be flushed, but in a situation
2028 	 *	where we're really low on mbufs, this is potentially
2029 	 *	useful.
2030 	 */
2031 		INP_INFO_WLOCK(&V_tcbinfo);
2032 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2033 			INP_WLOCK(inpb);
2034 			if (inpb->inp_flags & INP_TIMEWAIT) {
2035 				INP_WUNLOCK(inpb);
2036 				continue;
2037 			}
2038 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2039 				tcp_reass_flush(tcpb);
2040 				tcp_clean_sackreport(tcpb);
2041 #ifdef TCP_BLACKBOX
2042 				tcp_log_drain(tcpb);
2043 #endif
2044 #ifdef TCPPCAP
2045 				if (tcp_pcap_aggressive_free) {
2046 					/* Free the TCP PCAP queues. */
2047 					tcp_pcap_drain(&(tcpb->t_inpkts));
2048 					tcp_pcap_drain(&(tcpb->t_outpkts));
2049 				}
2050 #endif
2051 			}
2052 			INP_WUNLOCK(inpb);
2053 		}
2054 		INP_INFO_WUNLOCK(&V_tcbinfo);
2055 		CURVNET_RESTORE();
2056 	}
2057 	VNET_LIST_RUNLOCK_NOSLEEP();
2058 }
2059 
2060 /*
2061  * Notify a tcp user of an asynchronous error;
2062  * store error as soft error, but wake up user
2063  * (for now, won't do anything until can select for soft error).
2064  *
2065  * Do not wake up user since there currently is no mechanism for
2066  * reporting soft errors (yet - a kqueue filter may be added).
2067  */
2068 static struct inpcb *
2069 tcp_notify(struct inpcb *inp, int error)
2070 {
2071 	struct tcpcb *tp;
2072 
2073 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2074 	INP_WLOCK_ASSERT(inp);
2075 
2076 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2077 	    (inp->inp_flags & INP_DROPPED))
2078 		return (inp);
2079 
2080 	tp = intotcpcb(inp);
2081 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2082 
2083 	/*
2084 	 * Ignore some errors if we are hooked up.
2085 	 * If connection hasn't completed, has retransmitted several times,
2086 	 * and receives a second error, give up now.  This is better
2087 	 * than waiting a long time to establish a connection that
2088 	 * can never complete.
2089 	 */
2090 	if (tp->t_state == TCPS_ESTABLISHED &&
2091 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2092 	     error == EHOSTDOWN)) {
2093 		if (inp->inp_route.ro_rt) {
2094 			RTFREE(inp->inp_route.ro_rt);
2095 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2096 		}
2097 		return (inp);
2098 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2099 	    tp->t_softerror) {
2100 		tp = tcp_drop(tp, error);
2101 		if (tp != NULL)
2102 			return (inp);
2103 		else
2104 			return (NULL);
2105 	} else {
2106 		tp->t_softerror = error;
2107 		return (inp);
2108 	}
2109 #if 0
2110 	wakeup( &so->so_timeo);
2111 	sorwakeup(so);
2112 	sowwakeup(so);
2113 #endif
2114 }
2115 
2116 static int
2117 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2118 {
2119 	int error, i, m, n, pcb_count;
2120 	struct inpcb *inp, **inp_list;
2121 	inp_gen_t gencnt;
2122 	struct xinpgen xig;
2123 	struct epoch_tracker et;
2124 
2125 	/*
2126 	 * The process of preparing the TCB list is too time-consuming and
2127 	 * resource-intensive to repeat twice on every request.
2128 	 */
2129 	if (req->oldptr == NULL) {
2130 		n = V_tcbinfo.ipi_count +
2131 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2132 		n += imax(n / 8, 10);
2133 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2134 		return (0);
2135 	}
2136 
2137 	if (req->newptr != NULL)
2138 		return (EPERM);
2139 
2140 	/*
2141 	 * OK, now we're committed to doing something.
2142 	 */
2143 	INP_LIST_RLOCK(&V_tcbinfo);
2144 	gencnt = V_tcbinfo.ipi_gencnt;
2145 	n = V_tcbinfo.ipi_count;
2146 	INP_LIST_RUNLOCK(&V_tcbinfo);
2147 
2148 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2149 
2150 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2151 		+ (n + m) * sizeof(struct xtcpcb));
2152 	if (error != 0)
2153 		return (error);
2154 
2155 	bzero(&xig, sizeof(xig));
2156 	xig.xig_len = sizeof xig;
2157 	xig.xig_count = n + m;
2158 	xig.xig_gen = gencnt;
2159 	xig.xig_sogen = so_gencnt;
2160 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2161 	if (error)
2162 		return (error);
2163 
2164 	error = syncache_pcblist(req, m, &pcb_count);
2165 	if (error)
2166 		return (error);
2167 
2168 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2169 
2170 	INP_INFO_WLOCK(&V_tcbinfo);
2171 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2172 	    inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2173 		INP_WLOCK(inp);
2174 		if (inp->inp_gencnt <= gencnt) {
2175 			/*
2176 			 * XXX: This use of cr_cansee(), introduced with
2177 			 * TCP state changes, is not quite right, but for
2178 			 * now, better than nothing.
2179 			 */
2180 			if (inp->inp_flags & INP_TIMEWAIT) {
2181 				if (intotw(inp) != NULL)
2182 					error = cr_cansee(req->td->td_ucred,
2183 					    intotw(inp)->tw_cred);
2184 				else
2185 					error = EINVAL;	/* Skip this inp. */
2186 			} else
2187 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2188 			if (error == 0) {
2189 				in_pcbref(inp);
2190 				inp_list[i++] = inp;
2191 			}
2192 		}
2193 		INP_WUNLOCK(inp);
2194 	}
2195 	INP_INFO_WUNLOCK(&V_tcbinfo);
2196 	n = i;
2197 
2198 	error = 0;
2199 	for (i = 0; i < n; i++) {
2200 		inp = inp_list[i];
2201 		INP_RLOCK(inp);
2202 		if (inp->inp_gencnt <= gencnt) {
2203 			struct xtcpcb xt;
2204 
2205 			tcp_inptoxtp(inp, &xt);
2206 			INP_RUNLOCK(inp);
2207 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2208 		} else
2209 			INP_RUNLOCK(inp);
2210 	}
2211 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2212 	for (i = 0; i < n; i++) {
2213 		inp = inp_list[i];
2214 		INP_RLOCK(inp);
2215 		if (!in_pcbrele_rlocked(inp))
2216 			INP_RUNLOCK(inp);
2217 	}
2218 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2219 
2220 	if (!error) {
2221 		/*
2222 		 * Give the user an updated idea of our state.
2223 		 * If the generation differs from what we told
2224 		 * her before, she knows that something happened
2225 		 * while we were processing this request, and it
2226 		 * might be necessary to retry.
2227 		 */
2228 		INP_LIST_RLOCK(&V_tcbinfo);
2229 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2230 		xig.xig_sogen = so_gencnt;
2231 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2232 		INP_LIST_RUNLOCK(&V_tcbinfo);
2233 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2234 	}
2235 	free(inp_list, M_TEMP);
2236 	return (error);
2237 }
2238 
2239 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2240     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2241     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2242 
2243 #ifdef INET
2244 static int
2245 tcp_getcred(SYSCTL_HANDLER_ARGS)
2246 {
2247 	struct xucred xuc;
2248 	struct sockaddr_in addrs[2];
2249 	struct inpcb *inp;
2250 	int error;
2251 
2252 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2253 	if (error)
2254 		return (error);
2255 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2256 	if (error)
2257 		return (error);
2258 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2259 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2260 	if (inp != NULL) {
2261 		if (inp->inp_socket == NULL)
2262 			error = ENOENT;
2263 		if (error == 0)
2264 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2265 		if (error == 0)
2266 			cru2x(inp->inp_cred, &xuc);
2267 		INP_RUNLOCK(inp);
2268 	} else
2269 		error = ENOENT;
2270 	if (error == 0)
2271 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2272 	return (error);
2273 }
2274 
2275 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2276     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2277     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2278 #endif /* INET */
2279 
2280 #ifdef INET6
2281 static int
2282 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2283 {
2284 	struct xucred xuc;
2285 	struct sockaddr_in6 addrs[2];
2286 	struct inpcb *inp;
2287 	int error;
2288 #ifdef INET
2289 	int mapped = 0;
2290 #endif
2291 
2292 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2293 	if (error)
2294 		return (error);
2295 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2296 	if (error)
2297 		return (error);
2298 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2299 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2300 		return (error);
2301 	}
2302 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2303 #ifdef INET
2304 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2305 			mapped = 1;
2306 		else
2307 #endif
2308 			return (EINVAL);
2309 	}
2310 
2311 #ifdef INET
2312 	if (mapped == 1)
2313 		inp = in_pcblookup(&V_tcbinfo,
2314 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2315 			addrs[1].sin6_port,
2316 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2317 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2318 	else
2319 #endif
2320 		inp = in6_pcblookup(&V_tcbinfo,
2321 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2322 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2323 			INPLOOKUP_RLOCKPCB, NULL);
2324 	if (inp != NULL) {
2325 		if (inp->inp_socket == NULL)
2326 			error = ENOENT;
2327 		if (error == 0)
2328 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2329 		if (error == 0)
2330 			cru2x(inp->inp_cred, &xuc);
2331 		INP_RUNLOCK(inp);
2332 	} else
2333 		error = ENOENT;
2334 	if (error == 0)
2335 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2336 	return (error);
2337 }
2338 
2339 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2340     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2341     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2342 #endif /* INET6 */
2343 
2344 
2345 #ifdef INET
2346 void
2347 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2348 {
2349 	struct ip *ip = vip;
2350 	struct tcphdr *th;
2351 	struct in_addr faddr;
2352 	struct inpcb *inp;
2353 	struct tcpcb *tp;
2354 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2355 	struct icmp *icp;
2356 	struct in_conninfo inc;
2357 	struct epoch_tracker et;
2358 	tcp_seq icmp_tcp_seq;
2359 	int mtu;
2360 
2361 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2362 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2363 		return;
2364 
2365 	if (cmd == PRC_MSGSIZE)
2366 		notify = tcp_mtudisc_notify;
2367 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2368 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2369 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2370 		notify = tcp_drop_syn_sent;
2371 
2372 	/*
2373 	 * Hostdead is ugly because it goes linearly through all PCBs.
2374 	 * XXX: We never get this from ICMP, otherwise it makes an
2375 	 * excellent DoS attack on machines with many connections.
2376 	 */
2377 	else if (cmd == PRC_HOSTDEAD)
2378 		ip = NULL;
2379 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2380 		return;
2381 
2382 	if (ip == NULL) {
2383 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2384 		return;
2385 	}
2386 
2387 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2388 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2389 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2390 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2391 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2392 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2393 		/* signal EHOSTDOWN, as it flushes the cached route */
2394 		inp = (*notify)(inp, EHOSTDOWN);
2395 		goto out;
2396 	}
2397 	icmp_tcp_seq = th->th_seq;
2398 	if (inp != NULL)  {
2399 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2400 		    !(inp->inp_flags & INP_DROPPED) &&
2401 		    !(inp->inp_socket == NULL)) {
2402 			tp = intotcpcb(inp);
2403 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2404 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2405 				if (cmd == PRC_MSGSIZE) {
2406 					/*
2407 					 * MTU discovery:
2408 					 * If we got a needfrag set the MTU
2409 					 * in the route to the suggested new
2410 					 * value (if given) and then notify.
2411 					 */
2412 					mtu = ntohs(icp->icmp_nextmtu);
2413 					/*
2414 					 * If no alternative MTU was
2415 					 * proposed, try the next smaller
2416 					 * one.
2417 					 */
2418 					if (!mtu)
2419 						mtu = ip_next_mtu(
2420 						    ntohs(ip->ip_len), 1);
2421 					if (mtu < V_tcp_minmss +
2422 					    sizeof(struct tcpiphdr))
2423 						mtu = V_tcp_minmss +
2424 						    sizeof(struct tcpiphdr);
2425 					/*
2426 					 * Only process the offered MTU if it
2427 					 * is smaller than the current one.
2428 					 */
2429 					if (mtu < tp->t_maxseg +
2430 					    sizeof(struct tcpiphdr)) {
2431 						bzero(&inc, sizeof(inc));
2432 						inc.inc_faddr = faddr;
2433 						inc.inc_fibnum =
2434 						    inp->inp_inc.inc_fibnum;
2435 						tcp_hc_updatemtu(&inc, mtu);
2436 						tcp_mtudisc(inp, mtu);
2437 					}
2438 				} else
2439 					inp = (*notify)(inp,
2440 					    inetctlerrmap[cmd]);
2441 			}
2442 		}
2443 	} else {
2444 		bzero(&inc, sizeof(inc));
2445 		inc.inc_fport = th->th_dport;
2446 		inc.inc_lport = th->th_sport;
2447 		inc.inc_faddr = faddr;
2448 		inc.inc_laddr = ip->ip_src;
2449 		syncache_unreach(&inc, icmp_tcp_seq);
2450 	}
2451 out:
2452 	if (inp != NULL)
2453 		INP_WUNLOCK(inp);
2454 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2455 }
2456 #endif /* INET */
2457 
2458 #ifdef INET6
2459 void
2460 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2461 {
2462 	struct in6_addr *dst;
2463 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2464 	struct ip6_hdr *ip6;
2465 	struct mbuf *m;
2466 	struct inpcb *inp;
2467 	struct tcpcb *tp;
2468 	struct icmp6_hdr *icmp6;
2469 	struct ip6ctlparam *ip6cp = NULL;
2470 	const struct sockaddr_in6 *sa6_src = NULL;
2471 	struct in_conninfo inc;
2472 	struct epoch_tracker et;
2473 	struct tcp_ports {
2474 		uint16_t th_sport;
2475 		uint16_t th_dport;
2476 	} t_ports;
2477 	tcp_seq icmp_tcp_seq;
2478 	unsigned int mtu;
2479 	unsigned int off;
2480 
2481 	if (sa->sa_family != AF_INET6 ||
2482 	    sa->sa_len != sizeof(struct sockaddr_in6))
2483 		return;
2484 
2485 	/* if the parameter is from icmp6, decode it. */
2486 	if (d != NULL) {
2487 		ip6cp = (struct ip6ctlparam *)d;
2488 		icmp6 = ip6cp->ip6c_icmp6;
2489 		m = ip6cp->ip6c_m;
2490 		ip6 = ip6cp->ip6c_ip6;
2491 		off = ip6cp->ip6c_off;
2492 		sa6_src = ip6cp->ip6c_src;
2493 		dst = ip6cp->ip6c_finaldst;
2494 	} else {
2495 		m = NULL;
2496 		ip6 = NULL;
2497 		off = 0;	/* fool gcc */
2498 		sa6_src = &sa6_any;
2499 		dst = NULL;
2500 	}
2501 
2502 	if (cmd == PRC_MSGSIZE)
2503 		notify = tcp_mtudisc_notify;
2504 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2505 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2506 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2507 		notify = tcp_drop_syn_sent;
2508 
2509 	/*
2510 	 * Hostdead is ugly because it goes linearly through all PCBs.
2511 	 * XXX: We never get this from ICMP, otherwise it makes an
2512 	 * excellent DoS attack on machines with many connections.
2513 	 */
2514 	else if (cmd == PRC_HOSTDEAD)
2515 		ip6 = NULL;
2516 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2517 		return;
2518 
2519 	if (ip6 == NULL) {
2520 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2521 			      (const struct sockaddr *)sa6_src,
2522 			      0, cmd, NULL, notify);
2523 		return;
2524 	}
2525 
2526 	/* Check if we can safely get the ports from the tcp hdr */
2527 	if (m == NULL ||
2528 	    (m->m_pkthdr.len <
2529 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2530 		return;
2531 	}
2532 	bzero(&t_ports, sizeof(struct tcp_ports));
2533 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2534 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2535 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2536 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2537 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2538 		/* signal EHOSTDOWN, as it flushes the cached route */
2539 		inp = (*notify)(inp, EHOSTDOWN);
2540 		goto out;
2541 	}
2542 	off += sizeof(struct tcp_ports);
2543 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2544 		goto out;
2545 	}
2546 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2547 	if (inp != NULL)  {
2548 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2549 		    !(inp->inp_flags & INP_DROPPED) &&
2550 		    !(inp->inp_socket == NULL)) {
2551 			tp = intotcpcb(inp);
2552 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2553 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2554 				if (cmd == PRC_MSGSIZE) {
2555 					/*
2556 					 * MTU discovery:
2557 					 * If we got a needfrag set the MTU
2558 					 * in the route to the suggested new
2559 					 * value (if given) and then notify.
2560 					 */
2561 					mtu = ntohl(icmp6->icmp6_mtu);
2562 					/*
2563 					 * If no alternative MTU was
2564 					 * proposed, or the proposed
2565 					 * MTU was too small, set to
2566 					 * the min.
2567 					 */
2568 					if (mtu < IPV6_MMTU)
2569 						mtu = IPV6_MMTU - 8;
2570 					bzero(&inc, sizeof(inc));
2571 					inc.inc_fibnum = M_GETFIB(m);
2572 					inc.inc_flags |= INC_ISIPV6;
2573 					inc.inc6_faddr = *dst;
2574 					if (in6_setscope(&inc.inc6_faddr,
2575 						m->m_pkthdr.rcvif, NULL))
2576 						goto out;
2577 					/*
2578 					 * Only process the offered MTU if it
2579 					 * is smaller than the current one.
2580 					 */
2581 					if (mtu < tp->t_maxseg +
2582 					    sizeof (struct tcphdr) +
2583 					    sizeof (struct ip6_hdr)) {
2584 						tcp_hc_updatemtu(&inc, mtu);
2585 						tcp_mtudisc(inp, mtu);
2586 						ICMP6STAT_INC(icp6s_pmtuchg);
2587 					}
2588 				} else
2589 					inp = (*notify)(inp,
2590 					    inet6ctlerrmap[cmd]);
2591 			}
2592 		}
2593 	} else {
2594 		bzero(&inc, sizeof(inc));
2595 		inc.inc_fibnum = M_GETFIB(m);
2596 		inc.inc_flags |= INC_ISIPV6;
2597 		inc.inc_fport = t_ports.th_dport;
2598 		inc.inc_lport = t_ports.th_sport;
2599 		inc.inc6_faddr = *dst;
2600 		inc.inc6_laddr = ip6->ip6_src;
2601 		syncache_unreach(&inc, icmp_tcp_seq);
2602 	}
2603 out:
2604 	if (inp != NULL)
2605 		INP_WUNLOCK(inp);
2606 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2607 }
2608 #endif /* INET6 */
2609 
2610 static uint32_t
2611 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2612 {
2613 	MD5_CTX ctx;
2614 	uint32_t hash[4];
2615 
2616 	MD5Init(&ctx);
2617 	MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2618 	MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2619 	switch (inc->inc_flags & INC_ISIPV6) {
2620 #ifdef INET
2621 	case 0:
2622 		MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2623 		MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2624 		break;
2625 #endif
2626 #ifdef INET6
2627 	case INC_ISIPV6:
2628 		MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2629 		MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2630 		break;
2631 #endif
2632 	}
2633 	MD5Update(&ctx, key, len);
2634 	MD5Final((unsigned char *)hash, &ctx);
2635 
2636 	return (hash[0]);
2637 }
2638 
2639 uint32_t
2640 tcp_new_ts_offset(struct in_conninfo *inc)
2641 {
2642 	struct in_conninfo inc_store, *local_inc;
2643 
2644 	if (!V_tcp_ts_offset_per_conn) {
2645 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2646 		inc_store.inc_lport = 0;
2647 		inc_store.inc_fport = 0;
2648 		local_inc = &inc_store;
2649 	} else {
2650 		local_inc = inc;
2651 	}
2652 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2653 	    sizeof(V_ts_offset_secret)));
2654 }
2655 
2656 /*
2657  * Following is where TCP initial sequence number generation occurs.
2658  *
2659  * There are two places where we must use initial sequence numbers:
2660  * 1.  In SYN-ACK packets.
2661  * 2.  In SYN packets.
2662  *
2663  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2664  * tcp_syncache.c for details.
2665  *
2666  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2667  * depends on this property.  In addition, these ISNs should be
2668  * unguessable so as to prevent connection hijacking.  To satisfy
2669  * the requirements of this situation, the algorithm outlined in
2670  * RFC 1948 is used, with only small modifications.
2671  *
2672  * Implementation details:
2673  *
2674  * Time is based off the system timer, and is corrected so that it
2675  * increases by one megabyte per second.  This allows for proper
2676  * recycling on high speed LANs while still leaving over an hour
2677  * before rollover.
2678  *
2679  * As reading the *exact* system time is too expensive to be done
2680  * whenever setting up a TCP connection, we increment the time
2681  * offset in two ways.  First, a small random positive increment
2682  * is added to isn_offset for each connection that is set up.
2683  * Second, the function tcp_isn_tick fires once per clock tick
2684  * and increments isn_offset as necessary so that sequence numbers
2685  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2686  * random positive increments serve only to ensure that the same
2687  * exact sequence number is never sent out twice (as could otherwise
2688  * happen when a port is recycled in less than the system tick
2689  * interval.)
2690  *
2691  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2692  * between seeding of isn_secret.  This is normally set to zero,
2693  * as reseeding should not be necessary.
2694  *
2695  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2696  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2697  * general, this means holding an exclusive (write) lock.
2698  */
2699 
2700 #define ISN_BYTES_PER_SECOND 1048576
2701 #define ISN_STATIC_INCREMENT 4096
2702 #define ISN_RANDOM_INCREMENT (4096 - 1)
2703 #define ISN_SECRET_LENGTH    32
2704 
2705 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2706 VNET_DEFINE_STATIC(int, isn_last);
2707 VNET_DEFINE_STATIC(int, isn_last_reseed);
2708 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2709 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2710 
2711 #define	V_isn_secret			VNET(isn_secret)
2712 #define	V_isn_last			VNET(isn_last)
2713 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2714 #define	V_isn_offset			VNET(isn_offset)
2715 #define	V_isn_offset_old		VNET(isn_offset_old)
2716 
2717 tcp_seq
2718 tcp_new_isn(struct in_conninfo *inc)
2719 {
2720 	tcp_seq new_isn;
2721 	u_int32_t projected_offset;
2722 
2723 	ISN_LOCK();
2724 	/* Seed if this is the first use, reseed if requested. */
2725 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2726 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2727 		< (u_int)ticks))) {
2728 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2729 		V_isn_last_reseed = ticks;
2730 	}
2731 
2732 	/* Compute the md5 hash and return the ISN. */
2733 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2734 	    sizeof(V_isn_secret));
2735 	V_isn_offset += ISN_STATIC_INCREMENT +
2736 		(arc4random() & ISN_RANDOM_INCREMENT);
2737 	if (ticks != V_isn_last) {
2738 		projected_offset = V_isn_offset_old +
2739 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2740 		if (SEQ_GT(projected_offset, V_isn_offset))
2741 			V_isn_offset = projected_offset;
2742 		V_isn_offset_old = V_isn_offset;
2743 		V_isn_last = ticks;
2744 	}
2745 	new_isn += V_isn_offset;
2746 	ISN_UNLOCK();
2747 	return (new_isn);
2748 }
2749 
2750 /*
2751  * When a specific ICMP unreachable message is received and the
2752  * connection state is SYN-SENT, drop the connection.  This behavior
2753  * is controlled by the icmp_may_rst sysctl.
2754  */
2755 struct inpcb *
2756 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2757 {
2758 	struct tcpcb *tp;
2759 
2760 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2761 	INP_WLOCK_ASSERT(inp);
2762 
2763 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2764 	    (inp->inp_flags & INP_DROPPED))
2765 		return (inp);
2766 
2767 	tp = intotcpcb(inp);
2768 	if (tp->t_state != TCPS_SYN_SENT)
2769 		return (inp);
2770 
2771 	if (IS_FASTOPEN(tp->t_flags))
2772 		tcp_fastopen_disable_path(tp);
2773 
2774 	tp = tcp_drop(tp, errno);
2775 	if (tp != NULL)
2776 		return (inp);
2777 	else
2778 		return (NULL);
2779 }
2780 
2781 /*
2782  * When `need fragmentation' ICMP is received, update our idea of the MSS
2783  * based on the new value. Also nudge TCP to send something, since we
2784  * know the packet we just sent was dropped.
2785  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2786  */
2787 static struct inpcb *
2788 tcp_mtudisc_notify(struct inpcb *inp, int error)
2789 {
2790 
2791 	tcp_mtudisc(inp, -1);
2792 	return (inp);
2793 }
2794 
2795 static void
2796 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2797 {
2798 	struct tcpcb *tp;
2799 	struct socket *so;
2800 
2801 	INP_WLOCK_ASSERT(inp);
2802 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2803 	    (inp->inp_flags & INP_DROPPED))
2804 		return;
2805 
2806 	tp = intotcpcb(inp);
2807 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2808 
2809 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2810 
2811 	so = inp->inp_socket;
2812 	SOCKBUF_LOCK(&so->so_snd);
2813 	/* If the mss is larger than the socket buffer, decrease the mss. */
2814 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2815 		tp->t_maxseg = so->so_snd.sb_hiwat;
2816 	SOCKBUF_UNLOCK(&so->so_snd);
2817 
2818 	TCPSTAT_INC(tcps_mturesent);
2819 	tp->t_rtttime = 0;
2820 	tp->snd_nxt = tp->snd_una;
2821 	tcp_free_sackholes(tp);
2822 	tp->snd_recover = tp->snd_max;
2823 	if (tp->t_flags & TF_SACK_PERMIT)
2824 		EXIT_FASTRECOVERY(tp->t_flags);
2825 	tp->t_fb->tfb_tcp_output(tp);
2826 }
2827 
2828 #ifdef INET
2829 /*
2830  * Look-up the routing entry to the peer of this inpcb.  If no route
2831  * is found and it cannot be allocated, then return 0.  This routine
2832  * is called by TCP routines that access the rmx structure and by
2833  * tcp_mss_update to get the peer/interface MTU.
2834  */
2835 uint32_t
2836 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2837 {
2838 	struct nhop4_extended nh4;
2839 	struct ifnet *ifp;
2840 	uint32_t maxmtu = 0;
2841 
2842 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2843 
2844 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2845 
2846 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2847 		    NHR_REF, 0, &nh4) != 0)
2848 			return (0);
2849 
2850 		ifp = nh4.nh_ifp;
2851 		maxmtu = nh4.nh_mtu;
2852 
2853 		/* Report additional interface capabilities. */
2854 		if (cap != NULL) {
2855 			if (ifp->if_capenable & IFCAP_TSO4 &&
2856 			    ifp->if_hwassist & CSUM_TSO) {
2857 				cap->ifcap |= CSUM_TSO;
2858 				cap->tsomax = ifp->if_hw_tsomax;
2859 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2860 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2861 			}
2862 		}
2863 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2864 	}
2865 	return (maxmtu);
2866 }
2867 #endif /* INET */
2868 
2869 #ifdef INET6
2870 uint32_t
2871 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2872 {
2873 	struct nhop6_extended nh6;
2874 	struct in6_addr dst6;
2875 	uint32_t scopeid;
2876 	struct ifnet *ifp;
2877 	uint32_t maxmtu = 0;
2878 
2879 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2880 
2881 	if (inc->inc_flags & INC_IPV6MINMTU)
2882 		return (IPV6_MMTU);
2883 
2884 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2885 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2886 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2887 		    0, &nh6) != 0)
2888 			return (0);
2889 
2890 		ifp = nh6.nh_ifp;
2891 		maxmtu = nh6.nh_mtu;
2892 
2893 		/* Report additional interface capabilities. */
2894 		if (cap != NULL) {
2895 			if (ifp->if_capenable & IFCAP_TSO6 &&
2896 			    ifp->if_hwassist & CSUM_TSO) {
2897 				cap->ifcap |= CSUM_TSO;
2898 				cap->tsomax = ifp->if_hw_tsomax;
2899 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2900 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2901 			}
2902 		}
2903 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2904 	}
2905 
2906 	return (maxmtu);
2907 }
2908 #endif /* INET6 */
2909 
2910 /*
2911  * Calculate effective SMSS per RFC5681 definition for a given TCP
2912  * connection at its current state, taking into account SACK and etc.
2913  */
2914 u_int
2915 tcp_maxseg(const struct tcpcb *tp)
2916 {
2917 	u_int optlen;
2918 
2919 	if (tp->t_flags & TF_NOOPT)
2920 		return (tp->t_maxseg);
2921 
2922 	/*
2923 	 * Here we have a simplified code from tcp_addoptions(),
2924 	 * without a proper loop, and having most of paddings hardcoded.
2925 	 * We might make mistakes with padding here in some edge cases,
2926 	 * but this is harmless, since result of tcp_maxseg() is used
2927 	 * only in cwnd and ssthresh estimations.
2928 	 */
2929 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2930 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2931 		if (tp->t_flags & TF_RCVD_TSTMP)
2932 			optlen = TCPOLEN_TSTAMP_APPA;
2933 		else
2934 			optlen = 0;
2935 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2936 		if (tp->t_flags & TF_SIGNATURE)
2937 			optlen += PAD(TCPOLEN_SIGNATURE);
2938 #endif
2939 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2940 			optlen += TCPOLEN_SACKHDR;
2941 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2942 			optlen = PAD(optlen);
2943 		}
2944 	} else {
2945 		if (tp->t_flags & TF_REQ_TSTMP)
2946 			optlen = TCPOLEN_TSTAMP_APPA;
2947 		else
2948 			optlen = PAD(TCPOLEN_MAXSEG);
2949 		if (tp->t_flags & TF_REQ_SCALE)
2950 			optlen += PAD(TCPOLEN_WINDOW);
2951 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2952 		if (tp->t_flags & TF_SIGNATURE)
2953 			optlen += PAD(TCPOLEN_SIGNATURE);
2954 #endif
2955 		if (tp->t_flags & TF_SACK_PERMIT)
2956 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2957 	}
2958 #undef PAD
2959 	optlen = min(optlen, TCP_MAXOLEN);
2960 	return (tp->t_maxseg - optlen);
2961 }
2962 
2963 static int
2964 sysctl_drop(SYSCTL_HANDLER_ARGS)
2965 {
2966 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2967 	struct sockaddr_storage addrs[2];
2968 	struct inpcb *inp;
2969 	struct tcpcb *tp;
2970 	struct tcptw *tw;
2971 	struct sockaddr_in *fin, *lin;
2972 	struct epoch_tracker et;
2973 #ifdef INET6
2974 	struct sockaddr_in6 *fin6, *lin6;
2975 #endif
2976 	int error;
2977 
2978 	inp = NULL;
2979 	fin = lin = NULL;
2980 #ifdef INET6
2981 	fin6 = lin6 = NULL;
2982 #endif
2983 	error = 0;
2984 
2985 	if (req->oldptr != NULL || req->oldlen != 0)
2986 		return (EINVAL);
2987 	if (req->newptr == NULL)
2988 		return (EPERM);
2989 	if (req->newlen < sizeof(addrs))
2990 		return (ENOMEM);
2991 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2992 	if (error)
2993 		return (error);
2994 
2995 	switch (addrs[0].ss_family) {
2996 #ifdef INET6
2997 	case AF_INET6:
2998 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2999 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3000 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3001 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3002 			return (EINVAL);
3003 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3004 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3005 				return (EINVAL);
3006 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3007 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3008 			fin = (struct sockaddr_in *)&addrs[0];
3009 			lin = (struct sockaddr_in *)&addrs[1];
3010 			break;
3011 		}
3012 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3013 		if (error)
3014 			return (error);
3015 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3016 		if (error)
3017 			return (error);
3018 		break;
3019 #endif
3020 #ifdef INET
3021 	case AF_INET:
3022 		fin = (struct sockaddr_in *)&addrs[0];
3023 		lin = (struct sockaddr_in *)&addrs[1];
3024 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3025 		    lin->sin_len != sizeof(struct sockaddr_in))
3026 			return (EINVAL);
3027 		break;
3028 #endif
3029 	default:
3030 		return (EINVAL);
3031 	}
3032 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3033 	switch (addrs[0].ss_family) {
3034 #ifdef INET6
3035 	case AF_INET6:
3036 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3037 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3038 		    INPLOOKUP_WLOCKPCB, NULL);
3039 		break;
3040 #endif
3041 #ifdef INET
3042 	case AF_INET:
3043 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3044 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3045 		break;
3046 #endif
3047 	}
3048 	if (inp != NULL) {
3049 		if (inp->inp_flags & INP_TIMEWAIT) {
3050 			/*
3051 			 * XXXRW: There currently exists a state where an
3052 			 * inpcb is present, but its timewait state has been
3053 			 * discarded.  For now, don't allow dropping of this
3054 			 * type of inpcb.
3055 			 */
3056 			tw = intotw(inp);
3057 			if (tw != NULL)
3058 				tcp_twclose(tw, 0);
3059 			else
3060 				INP_WUNLOCK(inp);
3061 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3062 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3063 			tp = intotcpcb(inp);
3064 			tp = tcp_drop(tp, ECONNABORTED);
3065 			if (tp != NULL)
3066 				INP_WUNLOCK(inp);
3067 		} else
3068 			INP_WUNLOCK(inp);
3069 	} else
3070 		error = ESRCH;
3071 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3072 	return (error);
3073 }
3074 
3075 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3076     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3077     0, sysctl_drop, "", "Drop TCP connection");
3078 
3079 /*
3080  * Generate a standardized TCP log line for use throughout the
3081  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3082  * allow use in the interrupt context.
3083  *
3084  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3085  * NB: The function may return NULL if memory allocation failed.
3086  *
3087  * Due to header inclusion and ordering limitations the struct ip
3088  * and ip6_hdr pointers have to be passed as void pointers.
3089  */
3090 char *
3091 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3092     const void *ip6hdr)
3093 {
3094 
3095 	/* Is logging enabled? */
3096 	if (tcp_log_in_vain == 0)
3097 		return (NULL);
3098 
3099 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3100 }
3101 
3102 char *
3103 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3104     const void *ip6hdr)
3105 {
3106 
3107 	/* Is logging enabled? */
3108 	if (tcp_log_debug == 0)
3109 		return (NULL);
3110 
3111 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3112 }
3113 
3114 static char *
3115 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3116     const void *ip6hdr)
3117 {
3118 	char *s, *sp;
3119 	size_t size;
3120 	struct ip *ip;
3121 #ifdef INET6
3122 	const struct ip6_hdr *ip6;
3123 
3124 	ip6 = (const struct ip6_hdr *)ip6hdr;
3125 #endif /* INET6 */
3126 	ip = (struct ip *)ip4hdr;
3127 
3128 	/*
3129 	 * The log line looks like this:
3130 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3131 	 */
3132 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3133 	    sizeof(PRINT_TH_FLAGS) + 1 +
3134 #ifdef INET6
3135 	    2 * INET6_ADDRSTRLEN;
3136 #else
3137 	    2 * INET_ADDRSTRLEN;
3138 #endif /* INET6 */
3139 
3140 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3141 	if (s == NULL)
3142 		return (NULL);
3143 
3144 	strcat(s, "TCP: [");
3145 	sp = s + strlen(s);
3146 
3147 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3148 		inet_ntoa_r(inc->inc_faddr, sp);
3149 		sp = s + strlen(s);
3150 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3151 		sp = s + strlen(s);
3152 		inet_ntoa_r(inc->inc_laddr, sp);
3153 		sp = s + strlen(s);
3154 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3155 #ifdef INET6
3156 	} else if (inc) {
3157 		ip6_sprintf(sp, &inc->inc6_faddr);
3158 		sp = s + strlen(s);
3159 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3160 		sp = s + strlen(s);
3161 		ip6_sprintf(sp, &inc->inc6_laddr);
3162 		sp = s + strlen(s);
3163 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3164 	} else if (ip6 && th) {
3165 		ip6_sprintf(sp, &ip6->ip6_src);
3166 		sp = s + strlen(s);
3167 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3168 		sp = s + strlen(s);
3169 		ip6_sprintf(sp, &ip6->ip6_dst);
3170 		sp = s + strlen(s);
3171 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3172 #endif /* INET6 */
3173 #ifdef INET
3174 	} else if (ip && th) {
3175 		inet_ntoa_r(ip->ip_src, sp);
3176 		sp = s + strlen(s);
3177 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3178 		sp = s + strlen(s);
3179 		inet_ntoa_r(ip->ip_dst, sp);
3180 		sp = s + strlen(s);
3181 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3182 #endif /* INET */
3183 	} else {
3184 		free(s, M_TCPLOG);
3185 		return (NULL);
3186 	}
3187 	sp = s + strlen(s);
3188 	if (th)
3189 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3190 	if (*(s + size - 1) != '\0')
3191 		panic("%s: string too long", __func__);
3192 	return (s);
3193 }
3194 
3195 /*
3196  * A subroutine which makes it easy to track TCP state changes with DTrace.
3197  * This function shouldn't be called for t_state initializations that don't
3198  * correspond to actual TCP state transitions.
3199  */
3200 void
3201 tcp_state_change(struct tcpcb *tp, int newstate)
3202 {
3203 #if defined(KDTRACE_HOOKS)
3204 	int pstate = tp->t_state;
3205 #endif
3206 
3207 	TCPSTATES_DEC(tp->t_state);
3208 	TCPSTATES_INC(newstate);
3209 	tp->t_state = newstate;
3210 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3211 }
3212 
3213 /*
3214  * Create an external-format (``xtcpcb'') structure using the information in
3215  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3216  * reduce the spew of irrelevant information over this interface, to isolate
3217  * user code from changes in the kernel structure, and potentially to provide
3218  * information-hiding if we decide that some of this information should be
3219  * hidden from users.
3220  */
3221 void
3222 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3223 {
3224 	struct tcpcb *tp = intotcpcb(inp);
3225 	sbintime_t now;
3226 
3227 	bzero(xt, sizeof(*xt));
3228 	if (inp->inp_flags & INP_TIMEWAIT) {
3229 		xt->t_state = TCPS_TIME_WAIT;
3230 	} else {
3231 		xt->t_state = tp->t_state;
3232 		xt->t_logstate = tp->t_logstate;
3233 		xt->t_flags = tp->t_flags;
3234 		xt->t_sndzerowin = tp->t_sndzerowin;
3235 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3236 		xt->t_rcvoopack = tp->t_rcvoopack;
3237 
3238 		now = getsbinuptime();
3239 #define	COPYTIMER(ttt)	do {						\
3240 		if (callout_active(&tp->t_timers->ttt))			\
3241 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3242 			    SBT_1MS;					\
3243 		else							\
3244 			xt->ttt = 0;					\
3245 } while (0)
3246 		COPYTIMER(tt_delack);
3247 		COPYTIMER(tt_rexmt);
3248 		COPYTIMER(tt_persist);
3249 		COPYTIMER(tt_keep);
3250 		COPYTIMER(tt_2msl);
3251 #undef COPYTIMER
3252 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3253 
3254 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3255 		    TCP_FUNCTION_NAME_LEN_MAX);
3256 #ifdef TCP_BLACKBOX
3257 		(void)tcp_log_get_id(tp, xt->xt_logid);
3258 #endif
3259 	}
3260 
3261 	xt->xt_len = sizeof(struct xtcpcb);
3262 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3263 	if (inp->inp_socket == NULL)
3264 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3265 }
3266