xref: /freebsd/sys/netinet/tcp_subr.c (revision 13ea0450a9c8742119d36f3bf8f47accdce46e54)
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 static int	tcp_log_debug = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 
202 static int	tcp_tcbhashsize;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 
206 static int	do_tcpdrain = 1;
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208     "Enable tcp_drain routine for extra help when low on mbufs");
209 
210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 
213 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
214 #define	V_icmp_may_rst			VNET(icmp_may_rst)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216     &VNET_NAME(icmp_may_rst), 0,
217     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 
219 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
220 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_isn_reseed_interval), 0,
223     "Seconds between reseeding of ISN secret");
224 
225 static int	tcp_soreceive_stream;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 
229 VNET_DEFINE(uma_zone_t, sack_hole_zone);
230 #define	V_sack_hole_zone		VNET(sack_hole_zone)
231 
232 #ifdef TCP_HHOOK
233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
234 #endif
235 
236 #define TS_OFFSET_SECRET_LENGTH 32
237 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
238 #define	V_ts_offset_secret	VNET(ts_offset_secret)
239 
240 static int	tcp_default_fb_init(struct tcpcb *tp);
241 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
242 static int	tcp_default_handoff_ok(struct tcpcb *tp);
243 static struct inpcb *tcp_notify(struct inpcb *, int);
244 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
245 static void tcp_mtudisc(struct inpcb *, int);
246 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
247 		    void *ip4hdr, const void *ip6hdr);
248 
249 
250 static struct tcp_function_block tcp_def_funcblk = {
251 	.tfb_tcp_block_name = "freebsd",
252 	.tfb_tcp_output = tcp_output,
253 	.tfb_tcp_do_segment = tcp_do_segment,
254 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
255 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
256 	.tfb_tcp_fb_init = tcp_default_fb_init,
257 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
258 };
259 
260 static int tcp_fb_cnt = 0;
261 struct tcp_funchead t_functions;
262 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
263 
264 static struct tcp_function_block *
265 find_tcp_functions_locked(struct tcp_function_set *fs)
266 {
267 	struct tcp_function *f;
268 	struct tcp_function_block *blk=NULL;
269 
270 	TAILQ_FOREACH(f, &t_functions, tf_next) {
271 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
272 			blk = f->tf_fb;
273 			break;
274 		}
275 	}
276 	return(blk);
277 }
278 
279 static struct tcp_function_block *
280 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
281 {
282 	struct tcp_function_block *rblk=NULL;
283 	struct tcp_function *f;
284 
285 	TAILQ_FOREACH(f, &t_functions, tf_next) {
286 		if (f->tf_fb == blk) {
287 			rblk = blk;
288 			if (s) {
289 				*s = f;
290 			}
291 			break;
292 		}
293 	}
294 	return (rblk);
295 }
296 
297 struct tcp_function_block *
298 find_and_ref_tcp_functions(struct tcp_function_set *fs)
299 {
300 	struct tcp_function_block *blk;
301 
302 	rw_rlock(&tcp_function_lock);
303 	blk = find_tcp_functions_locked(fs);
304 	if (blk)
305 		refcount_acquire(&blk->tfb_refcnt);
306 	rw_runlock(&tcp_function_lock);
307 	return(blk);
308 }
309 
310 struct tcp_function_block *
311 find_and_ref_tcp_fb(struct tcp_function_block *blk)
312 {
313 	struct tcp_function_block *rblk;
314 
315 	rw_rlock(&tcp_function_lock);
316 	rblk = find_tcp_fb_locked(blk, NULL);
317 	if (rblk)
318 		refcount_acquire(&rblk->tfb_refcnt);
319 	rw_runlock(&tcp_function_lock);
320 	return(rblk);
321 }
322 
323 static struct tcp_function_block *
324 find_and_ref_tcp_default_fb(void)
325 {
326 	struct tcp_function_block *rblk;
327 
328 	rw_rlock(&tcp_function_lock);
329 	rblk = tcp_func_set_ptr;
330 	refcount_acquire(&rblk->tfb_refcnt);
331 	rw_runlock(&tcp_function_lock);
332 	return (rblk);
333 }
334 
335 void
336 tcp_switch_back_to_default(struct tcpcb *tp)
337 {
338 	struct tcp_function_block *tfb;
339 
340 	KASSERT(tp->t_fb != &tcp_def_funcblk,
341 	    ("%s: called by the built-in default stack", __func__));
342 
343 	/*
344 	 * Release the old stack. This function will either find a new one
345 	 * or panic.
346 	 */
347 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
348 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
349 	refcount_release(&tp->t_fb->tfb_refcnt);
350 
351 	/*
352 	 * Now, we'll find a new function block to use.
353 	 * Start by trying the current user-selected
354 	 * default, unless this stack is the user-selected
355 	 * default.
356 	 */
357 	tfb = find_and_ref_tcp_default_fb();
358 	if (tfb == tp->t_fb) {
359 		refcount_release(&tfb->tfb_refcnt);
360 		tfb = NULL;
361 	}
362 	/* Does the stack accept this connection? */
363 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
364 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
365 		refcount_release(&tfb->tfb_refcnt);
366 		tfb = NULL;
367 	}
368 	/* Try to use that stack. */
369 	if (tfb != NULL) {
370 		/* Initialize the new stack. If it succeeds, we are done. */
371 		tp->t_fb = tfb;
372 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
373 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
374 			return;
375 
376 		/*
377 		 * Initialization failed. Release the reference count on
378 		 * the stack.
379 		 */
380 		refcount_release(&tfb->tfb_refcnt);
381 	}
382 
383 	/*
384 	 * If that wasn't feasible, use the built-in default
385 	 * stack which is not allowed to reject anyone.
386 	 */
387 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
388 	if (tfb == NULL) {
389 		/* there always should be a default */
390 		panic("Can't refer to tcp_def_funcblk");
391 	}
392 	if (tfb->tfb_tcp_handoff_ok != NULL) {
393 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
394 			/* The default stack cannot say no */
395 			panic("Default stack rejects a new session?");
396 		}
397 	}
398 	tp->t_fb = tfb;
399 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
400 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
401 		/* The default stack cannot fail */
402 		panic("Default stack initialization failed");
403 	}
404 }
405 
406 static int
407 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
408 {
409 	int error=ENOENT;
410 	struct tcp_function_set fs;
411 	struct tcp_function_block *blk;
412 
413 	memset(&fs, 0, sizeof(fs));
414 	rw_rlock(&tcp_function_lock);
415 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
416 	if (blk) {
417 		/* Found him */
418 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
419 		fs.pcbcnt = blk->tfb_refcnt;
420 	}
421 	rw_runlock(&tcp_function_lock);
422 	error = sysctl_handle_string(oidp, fs.function_set_name,
423 				     sizeof(fs.function_set_name), req);
424 
425 	/* Check for error or no change */
426 	if (error != 0 || req->newptr == NULL)
427 		return(error);
428 
429 	rw_wlock(&tcp_function_lock);
430 	blk = find_tcp_functions_locked(&fs);
431 	if ((blk == NULL) ||
432 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
433 		error = ENOENT;
434 		goto done;
435 	}
436 	tcp_func_set_ptr = blk;
437 done:
438 	rw_wunlock(&tcp_function_lock);
439 	return (error);
440 }
441 
442 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
443 	    CTLTYPE_STRING | CTLFLAG_RW,
444 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
445 	    "Set/get the default TCP functions");
446 
447 static int
448 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
449 {
450 	int error, cnt, linesz;
451 	struct tcp_function *f;
452 	char *buffer, *cp;
453 	size_t bufsz, outsz;
454 	bool alias;
455 
456 	cnt = 0;
457 	rw_rlock(&tcp_function_lock);
458 	TAILQ_FOREACH(f, &t_functions, tf_next) {
459 		cnt++;
460 	}
461 	rw_runlock(&tcp_function_lock);
462 
463 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
464 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
465 
466 	error = 0;
467 	cp = buffer;
468 
469 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
470 	    "Alias", "PCB count");
471 	cp += linesz;
472 	bufsz -= linesz;
473 	outsz = linesz;
474 
475 	rw_rlock(&tcp_function_lock);
476 	TAILQ_FOREACH(f, &t_functions, tf_next) {
477 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
478 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
479 		    f->tf_fb->tfb_tcp_block_name,
480 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
481 		    alias ? f->tf_name : "-",
482 		    f->tf_fb->tfb_refcnt);
483 		if (linesz >= bufsz) {
484 			error = EOVERFLOW;
485 			break;
486 		}
487 		cp += linesz;
488 		bufsz -= linesz;
489 		outsz += linesz;
490 	}
491 	rw_runlock(&tcp_function_lock);
492 	if (error == 0)
493 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
494 	free(buffer, M_TEMP);
495 	return (error);
496 }
497 
498 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
499 	    CTLTYPE_STRING|CTLFLAG_RD,
500 	    NULL, 0, sysctl_net_inet_list_available, "A",
501 	    "list available TCP Function sets");
502 
503 /*
504  * Exports one (struct tcp_function_info) for each alias/name.
505  */
506 static int
507 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
508 {
509 	int cnt, error;
510 	struct tcp_function *f;
511 	struct tcp_function_info tfi;
512 
513 	/*
514 	 * We don't allow writes.
515 	 */
516 	if (req->newptr != NULL)
517 		return (EINVAL);
518 
519 	/*
520 	 * Wire the old buffer so we can directly copy the functions to
521 	 * user space without dropping the lock.
522 	 */
523 	if (req->oldptr != NULL) {
524 		error = sysctl_wire_old_buffer(req, 0);
525 		if (error)
526 			return (error);
527 	}
528 
529 	/*
530 	 * Walk the list and copy out matching entries. If INVARIANTS
531 	 * is compiled in, also walk the list to verify the length of
532 	 * the list matches what we have recorded.
533 	 */
534 	rw_rlock(&tcp_function_lock);
535 
536 	cnt = 0;
537 #ifndef INVARIANTS
538 	if (req->oldptr == NULL) {
539 		cnt = tcp_fb_cnt;
540 		goto skip_loop;
541 	}
542 #endif
543 	TAILQ_FOREACH(f, &t_functions, tf_next) {
544 #ifdef INVARIANTS
545 		cnt++;
546 #endif
547 		if (req->oldptr != NULL) {
548 			bzero(&tfi, sizeof(tfi));
549 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
550 			tfi.tfi_id = f->tf_fb->tfb_id;
551 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
552 			    sizeof(tfi.tfi_alias));
553 			(void)strlcpy(tfi.tfi_name,
554 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
555 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
556 			/*
557 			 * Don't stop on error, as that is the
558 			 * mechanism we use to accumulate length
559 			 * information if the buffer was too short.
560 			 */
561 		}
562 	}
563 	KASSERT(cnt == tcp_fb_cnt,
564 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
565 #ifndef INVARIANTS
566 skip_loop:
567 #endif
568 	rw_runlock(&tcp_function_lock);
569 	if (req->oldptr == NULL)
570 		error = SYSCTL_OUT(req, NULL,
571 		    (cnt + 1) * sizeof(struct tcp_function_info));
572 
573 	return (error);
574 }
575 
576 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
577 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
578 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
579 	    "List TCP function block name-to-ID mappings");
580 
581 /*
582  * tfb_tcp_handoff_ok() function for the default stack.
583  * Note that we'll basically try to take all comers.
584  */
585 static int
586 tcp_default_handoff_ok(struct tcpcb *tp)
587 {
588 
589 	return (0);
590 }
591 
592 /*
593  * tfb_tcp_fb_init() function for the default stack.
594  *
595  * This handles making sure we have appropriate timers set if you are
596  * transitioning a socket that has some amount of setup done.
597  *
598  * The init() fuction from the default can *never* return non-zero i.e.
599  * it is required to always succeed since it is the stack of last resort!
600  */
601 static int
602 tcp_default_fb_init(struct tcpcb *tp)
603 {
604 
605 	struct socket *so;
606 
607 	INP_WLOCK_ASSERT(tp->t_inpcb);
608 
609 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
610 	    ("%s: connection %p in unexpected state %d", __func__, tp,
611 	    tp->t_state));
612 
613 	/*
614 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
615 	 * know what to do for unexpected states (which includes TIME_WAIT).
616 	 */
617 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
618 		return (0);
619 
620 	/*
621 	 * Make sure some kind of transmission timer is set if there is
622 	 * outstanding data.
623 	 */
624 	so = tp->t_inpcb->inp_socket;
625 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
626 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
627 	    tcp_timer_active(tp, TT_PERSIST))) {
628 		/*
629 		 * If the session has established and it looks like it should
630 		 * be in the persist state, set the persist timer. Otherwise,
631 		 * set the retransmit timer.
632 		 */
633 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
634 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
635 		    (int32_t)sbavail(&so->so_snd))
636 			tcp_setpersist(tp);
637 		else
638 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
639 	}
640 
641 	/* All non-embryonic sessions get a keepalive timer. */
642 	if (!tcp_timer_active(tp, TT_KEEP))
643 		tcp_timer_activate(tp, TT_KEEP,
644 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
645 		    TP_KEEPINIT(tp));
646 
647 	return (0);
648 }
649 
650 /*
651  * tfb_tcp_fb_fini() function for the default stack.
652  *
653  * This changes state as necessary (or prudent) to prepare for another stack
654  * to assume responsibility for the connection.
655  */
656 static void
657 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
658 {
659 
660 	INP_WLOCK_ASSERT(tp->t_inpcb);
661 	return;
662 }
663 
664 /*
665  * Target size of TCP PCB hash tables. Must be a power of two.
666  *
667  * Note that this can be overridden by the kernel environment
668  * variable net.inet.tcp.tcbhashsize
669  */
670 #ifndef TCBHASHSIZE
671 #define TCBHASHSIZE	0
672 #endif
673 
674 /*
675  * XXX
676  * Callouts should be moved into struct tcp directly.  They are currently
677  * separate because the tcpcb structure is exported to userland for sysctl
678  * parsing purposes, which do not know about callouts.
679  */
680 struct tcpcb_mem {
681 	struct	tcpcb		tcb;
682 	struct	tcp_timer	tt;
683 	struct	cc_var		ccv;
684 #ifdef TCP_HHOOK
685 	struct	osd		osd;
686 #endif
687 };
688 
689 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
690 #define	V_tcpcb_zone			VNET(tcpcb_zone)
691 
692 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
693 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
694 
695 static struct mtx isn_mtx;
696 
697 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
698 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
699 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
700 
701 /*
702  * TCP initialization.
703  */
704 static void
705 tcp_zone_change(void *tag)
706 {
707 
708 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
709 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
710 	tcp_tw_zone_change();
711 }
712 
713 static int
714 tcp_inpcb_init(void *mem, int size, int flags)
715 {
716 	struct inpcb *inp = mem;
717 
718 	INP_LOCK_INIT(inp, "inp", "tcpinp");
719 	return (0);
720 }
721 
722 /*
723  * Take a value and get the next power of 2 that doesn't overflow.
724  * Used to size the tcp_inpcb hash buckets.
725  */
726 static int
727 maketcp_hashsize(int size)
728 {
729 	int hashsize;
730 
731 	/*
732 	 * auto tune.
733 	 * get the next power of 2 higher than maxsockets.
734 	 */
735 	hashsize = 1 << fls(size);
736 	/* catch overflow, and just go one power of 2 smaller */
737 	if (hashsize < size) {
738 		hashsize = 1 << (fls(size) - 1);
739 	}
740 	return (hashsize);
741 }
742 
743 static volatile int next_tcp_stack_id = 1;
744 
745 /*
746  * Register a TCP function block with the name provided in the names
747  * array.  (Note that this function does NOT automatically register
748  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
749  * explicitly include blk->tfb_tcp_block_name in the list of names if
750  * you wish to register the stack with that name.)
751  *
752  * Either all name registrations will succeed or all will fail.  If
753  * a name registration fails, the function will update the num_names
754  * argument to point to the array index of the name that encountered
755  * the failure.
756  *
757  * Returns 0 on success, or an error code on failure.
758  */
759 int
760 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
761     const char *names[], int *num_names)
762 {
763 	struct tcp_function *n;
764 	struct tcp_function_set fs;
765 	int error, i;
766 
767 	KASSERT(names != NULL && *num_names > 0,
768 	    ("%s: Called with 0-length name list", __func__));
769 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
770 	KASSERT(rw_initialized(&tcp_function_lock),
771 	    ("%s: called too early", __func__));
772 
773 	if ((blk->tfb_tcp_output == NULL) ||
774 	    (blk->tfb_tcp_do_segment == NULL) ||
775 	    (blk->tfb_tcp_ctloutput == NULL) ||
776 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
777 		/*
778 		 * These functions are required and you
779 		 * need a name.
780 		 */
781 		*num_names = 0;
782 		return (EINVAL);
783 	}
784 	if (blk->tfb_tcp_timer_stop_all ||
785 	    blk->tfb_tcp_timer_activate ||
786 	    blk->tfb_tcp_timer_active ||
787 	    blk->tfb_tcp_timer_stop) {
788 		/*
789 		 * If you define one timer function you
790 		 * must have them all.
791 		 */
792 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
793 		    (blk->tfb_tcp_timer_activate == NULL) ||
794 		    (blk->tfb_tcp_timer_active == NULL) ||
795 		    (blk->tfb_tcp_timer_stop == NULL)) {
796 			*num_names = 0;
797 			return (EINVAL);
798 		}
799 	}
800 
801 	refcount_init(&blk->tfb_refcnt, 0);
802 	blk->tfb_flags = 0;
803 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
804 	for (i = 0; i < *num_names; i++) {
805 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
806 		if (n == NULL) {
807 			error = ENOMEM;
808 			goto cleanup;
809 		}
810 		n->tf_fb = blk;
811 
812 		(void)strlcpy(fs.function_set_name, names[i],
813 		    sizeof(fs.function_set_name));
814 		rw_wlock(&tcp_function_lock);
815 		if (find_tcp_functions_locked(&fs) != NULL) {
816 			/* Duplicate name space not allowed */
817 			rw_wunlock(&tcp_function_lock);
818 			free(n, M_TCPFUNCTIONS);
819 			error = EALREADY;
820 			goto cleanup;
821 		}
822 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
823 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
824 		tcp_fb_cnt++;
825 		rw_wunlock(&tcp_function_lock);
826 	}
827 	return(0);
828 
829 cleanup:
830 	/*
831 	 * Deregister the names we just added. Because registration failed
832 	 * for names[i], we don't need to deregister that name.
833 	 */
834 	*num_names = i;
835 	rw_wlock(&tcp_function_lock);
836 	while (--i >= 0) {
837 		TAILQ_FOREACH(n, &t_functions, tf_next) {
838 			if (!strncmp(n->tf_name, names[i],
839 			    TCP_FUNCTION_NAME_LEN_MAX)) {
840 				TAILQ_REMOVE(&t_functions, n, tf_next);
841 				tcp_fb_cnt--;
842 				n->tf_fb = NULL;
843 				free(n, M_TCPFUNCTIONS);
844 				break;
845 			}
846 		}
847 	}
848 	rw_wunlock(&tcp_function_lock);
849 	return (error);
850 }
851 
852 /*
853  * Register a TCP function block using the name provided in the name
854  * argument.
855  *
856  * Returns 0 on success, or an error code on failure.
857  */
858 int
859 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
860     int wait)
861 {
862 	const char *name_list[1];
863 	int num_names, rv;
864 
865 	num_names = 1;
866 	if (name != NULL)
867 		name_list[0] = name;
868 	else
869 		name_list[0] = blk->tfb_tcp_block_name;
870 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
871 	return (rv);
872 }
873 
874 /*
875  * Register a TCP function block using the name defined in
876  * blk->tfb_tcp_block_name.
877  *
878  * Returns 0 on success, or an error code on failure.
879  */
880 int
881 register_tcp_functions(struct tcp_function_block *blk, int wait)
882 {
883 
884 	return (register_tcp_functions_as_name(blk, NULL, wait));
885 }
886 
887 /*
888  * Deregister all names associated with a function block. This
889  * functionally removes the function block from use within the system.
890  *
891  * When called with a true quiesce argument, mark the function block
892  * as being removed so no more stacks will use it and determine
893  * whether the removal would succeed.
894  *
895  * When called with a false quiesce argument, actually attempt the
896  * removal.
897  *
898  * When called with a force argument, attempt to switch all TCBs to
899  * use the default stack instead of returning EBUSY.
900  *
901  * Returns 0 on success (or if the removal would succeed, or an error
902  * code on failure.
903  */
904 int
905 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
906     bool force)
907 {
908 	struct tcp_function *f;
909 
910 	if (blk == &tcp_def_funcblk) {
911 		/* You can't un-register the default */
912 		return (EPERM);
913 	}
914 	rw_wlock(&tcp_function_lock);
915 	if (blk == tcp_func_set_ptr) {
916 		/* You can't free the current default */
917 		rw_wunlock(&tcp_function_lock);
918 		return (EBUSY);
919 	}
920 	/* Mark the block so no more stacks can use it. */
921 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
922 	/*
923 	 * If TCBs are still attached to the stack, attempt to switch them
924 	 * to the default stack.
925 	 */
926 	if (force && blk->tfb_refcnt) {
927 		struct inpcb *inp;
928 		struct tcpcb *tp;
929 		VNET_ITERATOR_DECL(vnet_iter);
930 
931 		rw_wunlock(&tcp_function_lock);
932 
933 		VNET_LIST_RLOCK();
934 		VNET_FOREACH(vnet_iter) {
935 			CURVNET_SET(vnet_iter);
936 			INP_INFO_WLOCK(&V_tcbinfo);
937 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
938 				INP_WLOCK(inp);
939 				if (inp->inp_flags & INP_TIMEWAIT) {
940 					INP_WUNLOCK(inp);
941 					continue;
942 				}
943 				tp = intotcpcb(inp);
944 				if (tp == NULL || tp->t_fb != blk) {
945 					INP_WUNLOCK(inp);
946 					continue;
947 				}
948 				tcp_switch_back_to_default(tp);
949 				INP_WUNLOCK(inp);
950 			}
951 			INP_INFO_WUNLOCK(&V_tcbinfo);
952 			CURVNET_RESTORE();
953 		}
954 		VNET_LIST_RUNLOCK();
955 
956 		rw_wlock(&tcp_function_lock);
957 	}
958 	if (blk->tfb_refcnt) {
959 		/* TCBs still attached. */
960 		rw_wunlock(&tcp_function_lock);
961 		return (EBUSY);
962 	}
963 	if (quiesce) {
964 		/* Skip removal. */
965 		rw_wunlock(&tcp_function_lock);
966 		return (0);
967 	}
968 	/* Remove any function names that map to this function block. */
969 	while (find_tcp_fb_locked(blk, &f) != NULL) {
970 		TAILQ_REMOVE(&t_functions, f, tf_next);
971 		tcp_fb_cnt--;
972 		f->tf_fb = NULL;
973 		free(f, M_TCPFUNCTIONS);
974 	}
975 	rw_wunlock(&tcp_function_lock);
976 	return (0);
977 }
978 
979 void
980 tcp_init(void)
981 {
982 	const char *tcbhash_tuneable;
983 	int hashsize;
984 
985 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
986 
987 #ifdef TCP_HHOOK
988 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
989 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
990 		printf("%s: WARNING: unable to register helper hook\n", __func__);
991 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
992 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
993 		printf("%s: WARNING: unable to register helper hook\n", __func__);
994 #endif
995 	hashsize = TCBHASHSIZE;
996 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
997 	if (hashsize == 0) {
998 		/*
999 		 * Auto tune the hash size based on maxsockets.
1000 		 * A perfect hash would have a 1:1 mapping
1001 		 * (hashsize = maxsockets) however it's been
1002 		 * suggested that O(2) average is better.
1003 		 */
1004 		hashsize = maketcp_hashsize(maxsockets / 4);
1005 		/*
1006 		 * Our historical default is 512,
1007 		 * do not autotune lower than this.
1008 		 */
1009 		if (hashsize < 512)
1010 			hashsize = 512;
1011 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1012 			printf("%s: %s auto tuned to %d\n", __func__,
1013 			    tcbhash_tuneable, hashsize);
1014 	}
1015 	/*
1016 	 * We require a hashsize to be a power of two.
1017 	 * Previously if it was not a power of two we would just reset it
1018 	 * back to 512, which could be a nasty surprise if you did not notice
1019 	 * the error message.
1020 	 * Instead what we do is clip it to the closest power of two lower
1021 	 * than the specified hash value.
1022 	 */
1023 	if (!powerof2(hashsize)) {
1024 		int oldhashsize = hashsize;
1025 
1026 		hashsize = maketcp_hashsize(hashsize);
1027 		/* prevent absurdly low value */
1028 		if (hashsize < 16)
1029 			hashsize = 16;
1030 		printf("%s: WARNING: TCB hash size not a power of 2, "
1031 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1032 		    hashsize);
1033 	}
1034 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1035 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1036 
1037 	/*
1038 	 * These have to be type stable for the benefit of the timers.
1039 	 */
1040 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1041 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1042 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1043 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1044 
1045 	tcp_tw_init();
1046 	syncache_init();
1047 	tcp_hc_init();
1048 
1049 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1050 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1051 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1052 
1053 	tcp_fastopen_init();
1054 
1055 	/* Skip initialization of globals for non-default instances. */
1056 	if (!IS_DEFAULT_VNET(curvnet))
1057 		return;
1058 
1059 	tcp_reass_global_init();
1060 
1061 	/* XXX virtualize those bellow? */
1062 	tcp_delacktime = TCPTV_DELACK;
1063 	tcp_keepinit = TCPTV_KEEP_INIT;
1064 	tcp_keepidle = TCPTV_KEEP_IDLE;
1065 	tcp_keepintvl = TCPTV_KEEPINTVL;
1066 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1067 	tcp_msl = TCPTV_MSL;
1068 	tcp_rexmit_min = TCPTV_MIN;
1069 	if (tcp_rexmit_min < 1)
1070 		tcp_rexmit_min = 1;
1071 	tcp_persmin = TCPTV_PERSMIN;
1072 	tcp_persmax = TCPTV_PERSMAX;
1073 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1074 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1075 	tcp_tcbhashsize = hashsize;
1076 
1077 	/* Setup the tcp function block list */
1078 	TAILQ_INIT(&t_functions);
1079 	rw_init(&tcp_function_lock, "tcp_func_lock");
1080 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1081 #ifdef TCP_BLACKBOX
1082 	/* Initialize the TCP logging data. */
1083 	tcp_log_init();
1084 #endif
1085 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1086 
1087 	if (tcp_soreceive_stream) {
1088 #ifdef INET
1089 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1090 #endif
1091 #ifdef INET6
1092 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1093 #endif /* INET6 */
1094 	}
1095 
1096 #ifdef INET6
1097 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1098 #else /* INET6 */
1099 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1100 #endif /* INET6 */
1101 	if (max_protohdr < TCP_MINPROTOHDR)
1102 		max_protohdr = TCP_MINPROTOHDR;
1103 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1104 		panic("tcp_init");
1105 #undef TCP_MINPROTOHDR
1106 
1107 	ISN_LOCK_INIT();
1108 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1109 		SHUTDOWN_PRI_DEFAULT);
1110 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1111 		EVENTHANDLER_PRI_ANY);
1112 #ifdef TCPPCAP
1113 	tcp_pcap_init();
1114 #endif
1115 }
1116 
1117 #ifdef VIMAGE
1118 static void
1119 tcp_destroy(void *unused __unused)
1120 {
1121 	int n;
1122 #ifdef TCP_HHOOK
1123 	int error;
1124 #endif
1125 
1126 	/*
1127 	 * All our processes are gone, all our sockets should be cleaned
1128 	 * up, which means, we should be past the tcp_discardcb() calls.
1129 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1130 	 */
1131 	for (;;) {
1132 		INP_LIST_RLOCK(&V_tcbinfo);
1133 		n = V_tcbinfo.ipi_count;
1134 		INP_LIST_RUNLOCK(&V_tcbinfo);
1135 		if (n == 0)
1136 			break;
1137 		pause("tcpdes", hz / 10);
1138 	}
1139 	tcp_hc_destroy();
1140 	syncache_destroy();
1141 	tcp_tw_destroy();
1142 	in_pcbinfo_destroy(&V_tcbinfo);
1143 	/* tcp_discardcb() clears the sack_holes up. */
1144 	uma_zdestroy(V_sack_hole_zone);
1145 	uma_zdestroy(V_tcpcb_zone);
1146 
1147 	/*
1148 	 * Cannot free the zone until all tcpcbs are released as we attach
1149 	 * the allocations to them.
1150 	 */
1151 	tcp_fastopen_destroy();
1152 
1153 #ifdef TCP_HHOOK
1154 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1155 	if (error != 0) {
1156 		printf("%s: WARNING: unable to deregister helper hook "
1157 		    "type=%d, id=%d: error %d returned\n", __func__,
1158 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1159 	}
1160 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1161 	if (error != 0) {
1162 		printf("%s: WARNING: unable to deregister helper hook "
1163 		    "type=%d, id=%d: error %d returned\n", __func__,
1164 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1165 	}
1166 #endif
1167 }
1168 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1169 #endif
1170 
1171 void
1172 tcp_fini(void *xtp)
1173 {
1174 
1175 }
1176 
1177 /*
1178  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1179  * tcp_template used to store this data in mbufs, but we now recopy it out
1180  * of the tcpcb each time to conserve mbufs.
1181  */
1182 void
1183 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1184 {
1185 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1186 
1187 	INP_WLOCK_ASSERT(inp);
1188 
1189 #ifdef INET6
1190 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1191 		struct ip6_hdr *ip6;
1192 
1193 		ip6 = (struct ip6_hdr *)ip_ptr;
1194 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1195 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1196 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1197 			(IPV6_VERSION & IPV6_VERSION_MASK);
1198 		ip6->ip6_nxt = IPPROTO_TCP;
1199 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1200 		ip6->ip6_src = inp->in6p_laddr;
1201 		ip6->ip6_dst = inp->in6p_faddr;
1202 	}
1203 #endif /* INET6 */
1204 #if defined(INET6) && defined(INET)
1205 	else
1206 #endif
1207 #ifdef INET
1208 	{
1209 		struct ip *ip;
1210 
1211 		ip = (struct ip *)ip_ptr;
1212 		ip->ip_v = IPVERSION;
1213 		ip->ip_hl = 5;
1214 		ip->ip_tos = inp->inp_ip_tos;
1215 		ip->ip_len = 0;
1216 		ip->ip_id = 0;
1217 		ip->ip_off = 0;
1218 		ip->ip_ttl = inp->inp_ip_ttl;
1219 		ip->ip_sum = 0;
1220 		ip->ip_p = IPPROTO_TCP;
1221 		ip->ip_src = inp->inp_laddr;
1222 		ip->ip_dst = inp->inp_faddr;
1223 	}
1224 #endif /* INET */
1225 	th->th_sport = inp->inp_lport;
1226 	th->th_dport = inp->inp_fport;
1227 	th->th_seq = 0;
1228 	th->th_ack = 0;
1229 	th->th_x2 = 0;
1230 	th->th_off = 5;
1231 	th->th_flags = 0;
1232 	th->th_win = 0;
1233 	th->th_urp = 0;
1234 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1235 }
1236 
1237 /*
1238  * Create template to be used to send tcp packets on a connection.
1239  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1240  * use for this function is in keepalives, which use tcp_respond.
1241  */
1242 struct tcptemp *
1243 tcpip_maketemplate(struct inpcb *inp)
1244 {
1245 	struct tcptemp *t;
1246 
1247 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1248 	if (t == NULL)
1249 		return (NULL);
1250 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1251 	return (t);
1252 }
1253 
1254 /*
1255  * Send a single message to the TCP at address specified by
1256  * the given TCP/IP header.  If m == NULL, then we make a copy
1257  * of the tcpiphdr at th and send directly to the addressed host.
1258  * This is used to force keep alive messages out using the TCP
1259  * template for a connection.  If flags are given then we send
1260  * a message back to the TCP which originated the segment th,
1261  * and discard the mbuf containing it and any other attached mbufs.
1262  *
1263  * In any case the ack and sequence number of the transmitted
1264  * segment are as specified by the parameters.
1265  *
1266  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1267  */
1268 void
1269 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1270     tcp_seq ack, tcp_seq seq, int flags)
1271 {
1272 	struct tcpopt to;
1273 	struct inpcb *inp;
1274 	struct ip *ip;
1275 	struct mbuf *optm;
1276 	struct tcphdr *nth;
1277 	u_char *optp;
1278 #ifdef INET6
1279 	struct ip6_hdr *ip6;
1280 	int isipv6;
1281 #endif /* INET6 */
1282 	int optlen, tlen, win;
1283 	bool incl_opts;
1284 
1285 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1286 
1287 #ifdef INET6
1288 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1289 	ip6 = ipgen;
1290 #endif /* INET6 */
1291 	ip = ipgen;
1292 
1293 	if (tp != NULL) {
1294 		inp = tp->t_inpcb;
1295 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1296 		INP_WLOCK_ASSERT(inp);
1297 	} else
1298 		inp = NULL;
1299 
1300 	incl_opts = false;
1301 	win = 0;
1302 	if (tp != NULL) {
1303 		if (!(flags & TH_RST)) {
1304 			win = sbspace(&inp->inp_socket->so_rcv);
1305 			if (win > TCP_MAXWIN << tp->rcv_scale)
1306 				win = TCP_MAXWIN << tp->rcv_scale;
1307 		}
1308 		if ((tp->t_flags & TF_NOOPT) == 0)
1309 			incl_opts = true;
1310 	}
1311 	if (m == NULL) {
1312 		m = m_gethdr(M_NOWAIT, MT_DATA);
1313 		if (m == NULL)
1314 			return;
1315 		m->m_data += max_linkhdr;
1316 #ifdef INET6
1317 		if (isipv6) {
1318 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1319 			      sizeof(struct ip6_hdr));
1320 			ip6 = mtod(m, struct ip6_hdr *);
1321 			nth = (struct tcphdr *)(ip6 + 1);
1322 		} else
1323 #endif /* INET6 */
1324 		{
1325 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1326 			ip = mtod(m, struct ip *);
1327 			nth = (struct tcphdr *)(ip + 1);
1328 		}
1329 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1330 		flags = TH_ACK;
1331 	} else if (!M_WRITABLE(m)) {
1332 		struct mbuf *n;
1333 
1334 		/* Can't reuse 'm', allocate a new mbuf. */
1335 		n = m_gethdr(M_NOWAIT, MT_DATA);
1336 		if (n == NULL) {
1337 			m_freem(m);
1338 			return;
1339 		}
1340 
1341 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1342 			m_freem(m);
1343 			m_freem(n);
1344 			return;
1345 		}
1346 
1347 		n->m_data += max_linkhdr;
1348 		/* m_len is set later */
1349 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1350 #ifdef INET6
1351 		if (isipv6) {
1352 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1353 			      sizeof(struct ip6_hdr));
1354 			ip6 = mtod(n, struct ip6_hdr *);
1355 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1356 			nth = (struct tcphdr *)(ip6 + 1);
1357 		} else
1358 #endif /* INET6 */
1359 		{
1360 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1361 			ip = mtod(n, struct ip *);
1362 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1363 			nth = (struct tcphdr *)(ip + 1);
1364 		}
1365 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1366 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1367 		th = nth;
1368 		m_freem(m);
1369 		m = n;
1370 	} else {
1371 		/*
1372 		 *  reuse the mbuf.
1373 		 * XXX MRT We inherit the FIB, which is lucky.
1374 		 */
1375 		m_freem(m->m_next);
1376 		m->m_next = NULL;
1377 		m->m_data = (caddr_t)ipgen;
1378 		/* m_len is set later */
1379 #ifdef INET6
1380 		if (isipv6) {
1381 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1382 			nth = (struct tcphdr *)(ip6 + 1);
1383 		} else
1384 #endif /* INET6 */
1385 		{
1386 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1387 			nth = (struct tcphdr *)(ip + 1);
1388 		}
1389 		if (th != nth) {
1390 			/*
1391 			 * this is usually a case when an extension header
1392 			 * exists between the IPv6 header and the
1393 			 * TCP header.
1394 			 */
1395 			nth->th_sport = th->th_sport;
1396 			nth->th_dport = th->th_dport;
1397 		}
1398 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1399 #undef xchg
1400 	}
1401 	tlen = 0;
1402 #ifdef INET6
1403 	if (isipv6)
1404 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1405 #endif
1406 #if defined(INET) && defined(INET6)
1407 	else
1408 #endif
1409 #ifdef INET
1410 		tlen = sizeof (struct tcpiphdr);
1411 #endif
1412 #ifdef INVARIANTS
1413 	m->m_len = 0;
1414 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1415 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1416 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1417 #endif
1418 	m->m_len = tlen;
1419 	to.to_flags = 0;
1420 	if (incl_opts) {
1421 		/* Make sure we have room. */
1422 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1423 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1424 			if (m->m_next) {
1425 				optp = mtod(m->m_next, u_char *);
1426 				optm = m->m_next;
1427 			} else
1428 				incl_opts = false;
1429 		} else {
1430 			optp = (u_char *) (nth + 1);
1431 			optm = m;
1432 		}
1433 	}
1434 	if (incl_opts) {
1435 		/* Timestamps. */
1436 		if (tp->t_flags & TF_RCVD_TSTMP) {
1437 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1438 			to.to_tsecr = tp->ts_recent;
1439 			to.to_flags |= TOF_TS;
1440 		}
1441 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1442 		/* TCP-MD5 (RFC2385). */
1443 		if (tp->t_flags & TF_SIGNATURE)
1444 			to.to_flags |= TOF_SIGNATURE;
1445 #endif
1446 		/* Add the options. */
1447 		tlen += optlen = tcp_addoptions(&to, optp);
1448 
1449 		/* Update m_len in the correct mbuf. */
1450 		optm->m_len += optlen;
1451 	} else
1452 		optlen = 0;
1453 #ifdef INET6
1454 	if (isipv6) {
1455 		ip6->ip6_flow = 0;
1456 		ip6->ip6_vfc = IPV6_VERSION;
1457 		ip6->ip6_nxt = IPPROTO_TCP;
1458 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1459 	}
1460 #endif
1461 #if defined(INET) && defined(INET6)
1462 	else
1463 #endif
1464 #ifdef INET
1465 	{
1466 		ip->ip_len = htons(tlen);
1467 		ip->ip_ttl = V_ip_defttl;
1468 		if (V_path_mtu_discovery)
1469 			ip->ip_off |= htons(IP_DF);
1470 	}
1471 #endif
1472 	m->m_pkthdr.len = tlen;
1473 	m->m_pkthdr.rcvif = NULL;
1474 #ifdef MAC
1475 	if (inp != NULL) {
1476 		/*
1477 		 * Packet is associated with a socket, so allow the
1478 		 * label of the response to reflect the socket label.
1479 		 */
1480 		INP_WLOCK_ASSERT(inp);
1481 		mac_inpcb_create_mbuf(inp, m);
1482 	} else {
1483 		/*
1484 		 * Packet is not associated with a socket, so possibly
1485 		 * update the label in place.
1486 		 */
1487 		mac_netinet_tcp_reply(m);
1488 	}
1489 #endif
1490 	nth->th_seq = htonl(seq);
1491 	nth->th_ack = htonl(ack);
1492 	nth->th_x2 = 0;
1493 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1494 	nth->th_flags = flags;
1495 	if (tp != NULL)
1496 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1497 	else
1498 		nth->th_win = htons((u_short)win);
1499 	nth->th_urp = 0;
1500 
1501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1502 	if (to.to_flags & TOF_SIGNATURE) {
1503 		if (!TCPMD5_ENABLED() ||
1504 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1505 			m_freem(m);
1506 			return;
1507 		}
1508 	}
1509 #endif
1510 
1511 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1512 #ifdef INET6
1513 	if (isipv6) {
1514 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1515 		nth->th_sum = in6_cksum_pseudo(ip6,
1516 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1517 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1518 		    NULL, NULL);
1519 	}
1520 #endif /* INET6 */
1521 #if defined(INET6) && defined(INET)
1522 	else
1523 #endif
1524 #ifdef INET
1525 	{
1526 		m->m_pkthdr.csum_flags = CSUM_TCP;
1527 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1528 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1529 	}
1530 #endif /* INET */
1531 #ifdef TCPDEBUG
1532 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1533 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1534 #endif
1535 	TCP_PROBE3(debug__output, tp, th, m);
1536 	if (flags & TH_RST)
1537 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1538 
1539 #ifdef INET6
1540 	if (isipv6) {
1541 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1542 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1543 	}
1544 #endif /* INET6 */
1545 #if defined(INET) && defined(INET6)
1546 	else
1547 #endif
1548 #ifdef INET
1549 	{
1550 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1551 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1552 	}
1553 #endif
1554 }
1555 
1556 /*
1557  * Create a new TCP control block, making an
1558  * empty reassembly queue and hooking it to the argument
1559  * protocol control block.  The `inp' parameter must have
1560  * come from the zone allocator set up in tcp_init().
1561  */
1562 struct tcpcb *
1563 tcp_newtcpcb(struct inpcb *inp)
1564 {
1565 	struct tcpcb_mem *tm;
1566 	struct tcpcb *tp;
1567 #ifdef INET6
1568 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1569 #endif /* INET6 */
1570 
1571 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1572 	if (tm == NULL)
1573 		return (NULL);
1574 	tp = &tm->tcb;
1575 
1576 	/* Initialise cc_var struct for this tcpcb. */
1577 	tp->ccv = &tm->ccv;
1578 	tp->ccv->type = IPPROTO_TCP;
1579 	tp->ccv->ccvc.tcp = tp;
1580 	rw_rlock(&tcp_function_lock);
1581 	tp->t_fb = tcp_func_set_ptr;
1582 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1583 	rw_runlock(&tcp_function_lock);
1584 	/*
1585 	 * Use the current system default CC algorithm.
1586 	 */
1587 	CC_LIST_RLOCK();
1588 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1589 	CC_ALGO(tp) = CC_DEFAULT();
1590 	CC_LIST_RUNLOCK();
1591 
1592 	if (CC_ALGO(tp)->cb_init != NULL)
1593 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1594 			if (tp->t_fb->tfb_tcp_fb_fini)
1595 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1596 			refcount_release(&tp->t_fb->tfb_refcnt);
1597 			uma_zfree(V_tcpcb_zone, tm);
1598 			return (NULL);
1599 		}
1600 
1601 #ifdef TCP_HHOOK
1602 	tp->osd = &tm->osd;
1603 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1604 		if (tp->t_fb->tfb_tcp_fb_fini)
1605 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1606 		refcount_release(&tp->t_fb->tfb_refcnt);
1607 		uma_zfree(V_tcpcb_zone, tm);
1608 		return (NULL);
1609 	}
1610 #endif
1611 
1612 #ifdef VIMAGE
1613 	tp->t_vnet = inp->inp_vnet;
1614 #endif
1615 	tp->t_timers = &tm->tt;
1616 	TAILQ_INIT(&tp->t_segq);
1617 	tp->t_maxseg =
1618 #ifdef INET6
1619 		isipv6 ? V_tcp_v6mssdflt :
1620 #endif /* INET6 */
1621 		V_tcp_mssdflt;
1622 
1623 	/* Set up our timeouts. */
1624 	callout_init(&tp->t_timers->tt_rexmt, 1);
1625 	callout_init(&tp->t_timers->tt_persist, 1);
1626 	callout_init(&tp->t_timers->tt_keep, 1);
1627 	callout_init(&tp->t_timers->tt_2msl, 1);
1628 	callout_init(&tp->t_timers->tt_delack, 1);
1629 
1630 	if (V_tcp_do_rfc1323)
1631 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1632 	if (V_tcp_do_sack)
1633 		tp->t_flags |= TF_SACK_PERMIT;
1634 	TAILQ_INIT(&tp->snd_holes);
1635 	/*
1636 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1637 	 * is called.
1638 	 */
1639 	in_pcbref(inp);	/* Reference for tcpcb */
1640 	tp->t_inpcb = inp;
1641 
1642 	/*
1643 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1644 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1645 	 * reasonable initial retransmit time.
1646 	 */
1647 	tp->t_srtt = TCPTV_SRTTBASE;
1648 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1649 	tp->t_rttmin = tcp_rexmit_min;
1650 	tp->t_rxtcur = TCPTV_RTOBASE;
1651 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1652 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1653 	tp->t_rcvtime = ticks;
1654 	/*
1655 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1656 	 * because the socket may be bound to an IPv6 wildcard address,
1657 	 * which may match an IPv4-mapped IPv6 address.
1658 	 */
1659 	inp->inp_ip_ttl = V_ip_defttl;
1660 	inp->inp_ppcb = tp;
1661 #ifdef TCPPCAP
1662 	/*
1663 	 * Init the TCP PCAP queues.
1664 	 */
1665 	tcp_pcap_tcpcb_init(tp);
1666 #endif
1667 #ifdef TCP_BLACKBOX
1668 	/* Initialize the per-TCPCB log data. */
1669 	tcp_log_tcpcbinit(tp);
1670 #endif
1671 	if (tp->t_fb->tfb_tcp_fb_init) {
1672 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1673 	}
1674 	return (tp);		/* XXX */
1675 }
1676 
1677 /*
1678  * Switch the congestion control algorithm back to NewReno for any active
1679  * control blocks using an algorithm which is about to go away.
1680  * This ensures the CC framework can allow the unload to proceed without leaving
1681  * any dangling pointers which would trigger a panic.
1682  * Returning non-zero would inform the CC framework that something went wrong
1683  * and it would be unsafe to allow the unload to proceed. However, there is no
1684  * way for this to occur with this implementation so we always return zero.
1685  */
1686 int
1687 tcp_ccalgounload(struct cc_algo *unload_algo)
1688 {
1689 	struct cc_algo *tmpalgo;
1690 	struct inpcb *inp;
1691 	struct tcpcb *tp;
1692 	VNET_ITERATOR_DECL(vnet_iter);
1693 
1694 	/*
1695 	 * Check all active control blocks across all network stacks and change
1696 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1697 	 * requires cleanup code to be run, call it.
1698 	 */
1699 	VNET_LIST_RLOCK();
1700 	VNET_FOREACH(vnet_iter) {
1701 		CURVNET_SET(vnet_iter);
1702 		INP_INFO_WLOCK(&V_tcbinfo);
1703 		/*
1704 		 * New connections already part way through being initialised
1705 		 * with the CC algo we're removing will not race with this code
1706 		 * because the INP_INFO_WLOCK is held during initialisation. We
1707 		 * therefore don't enter the loop below until the connection
1708 		 * list has stabilised.
1709 		 */
1710 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1711 			INP_WLOCK(inp);
1712 			/* Important to skip tcptw structs. */
1713 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1714 			    (tp = intotcpcb(inp)) != NULL) {
1715 				/*
1716 				 * By holding INP_WLOCK here, we are assured
1717 				 * that the connection is not currently
1718 				 * executing inside the CC module's functions
1719 				 * i.e. it is safe to make the switch back to
1720 				 * NewReno.
1721 				 */
1722 				if (CC_ALGO(tp) == unload_algo) {
1723 					tmpalgo = CC_ALGO(tp);
1724 					if (tmpalgo->cb_destroy != NULL)
1725 						tmpalgo->cb_destroy(tp->ccv);
1726 					CC_DATA(tp) = NULL;
1727 					/*
1728 					 * NewReno may allocate memory on
1729 					 * demand for certain stateful
1730 					 * configuration as needed, but is
1731 					 * coded to never fail on memory
1732 					 * allocation failure so it is a safe
1733 					 * fallback.
1734 					 */
1735 					CC_ALGO(tp) = &newreno_cc_algo;
1736 				}
1737 			}
1738 			INP_WUNLOCK(inp);
1739 		}
1740 		INP_INFO_WUNLOCK(&V_tcbinfo);
1741 		CURVNET_RESTORE();
1742 	}
1743 	VNET_LIST_RUNLOCK();
1744 
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Drop a TCP connection, reporting
1750  * the specified error.  If connection is synchronized,
1751  * then send a RST to peer.
1752  */
1753 struct tcpcb *
1754 tcp_drop(struct tcpcb *tp, int errno)
1755 {
1756 	struct socket *so = tp->t_inpcb->inp_socket;
1757 
1758 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1759 	INP_WLOCK_ASSERT(tp->t_inpcb);
1760 
1761 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1762 		tcp_state_change(tp, TCPS_CLOSED);
1763 		(void) tp->t_fb->tfb_tcp_output(tp);
1764 		TCPSTAT_INC(tcps_drops);
1765 	} else
1766 		TCPSTAT_INC(tcps_conndrops);
1767 	if (errno == ETIMEDOUT && tp->t_softerror)
1768 		errno = tp->t_softerror;
1769 	so->so_error = errno;
1770 	return (tcp_close(tp));
1771 }
1772 
1773 void
1774 tcp_discardcb(struct tcpcb *tp)
1775 {
1776 	struct inpcb *inp = tp->t_inpcb;
1777 	struct socket *so = inp->inp_socket;
1778 #ifdef INET6
1779 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1780 #endif /* INET6 */
1781 	int released __unused;
1782 
1783 	INP_WLOCK_ASSERT(inp);
1784 
1785 	/*
1786 	 * Make sure that all of our timers are stopped before we delete the
1787 	 * PCB.
1788 	 *
1789 	 * If stopping a timer fails, we schedule a discard function in same
1790 	 * callout, and the last discard function called will take care of
1791 	 * deleting the tcpcb.
1792 	 */
1793 	tp->t_timers->tt_draincnt = 0;
1794 	tcp_timer_stop(tp, TT_REXMT);
1795 	tcp_timer_stop(tp, TT_PERSIST);
1796 	tcp_timer_stop(tp, TT_KEEP);
1797 	tcp_timer_stop(tp, TT_2MSL);
1798 	tcp_timer_stop(tp, TT_DELACK);
1799 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1800 		/*
1801 		 * Call the stop-all function of the methods,
1802 		 * this function should call the tcp_timer_stop()
1803 		 * method with each of the function specific timeouts.
1804 		 * That stop will be called via the tfb_tcp_timer_stop()
1805 		 * which should use the async drain function of the
1806 		 * callout system (see tcp_var.h).
1807 		 */
1808 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1809 	}
1810 
1811 	/*
1812 	 * If we got enough samples through the srtt filter,
1813 	 * save the rtt and rttvar in the routing entry.
1814 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1815 	 * 4 samples is enough for the srtt filter to converge
1816 	 * to within enough % of the correct value; fewer samples
1817 	 * and we could save a bogus rtt. The danger is not high
1818 	 * as tcp quickly recovers from everything.
1819 	 * XXX: Works very well but needs some more statistics!
1820 	 */
1821 	if (tp->t_rttupdated >= 4) {
1822 		struct hc_metrics_lite metrics;
1823 		uint32_t ssthresh;
1824 
1825 		bzero(&metrics, sizeof(metrics));
1826 		/*
1827 		 * Update the ssthresh always when the conditions below
1828 		 * are satisfied. This gives us better new start value
1829 		 * for the congestion avoidance for new connections.
1830 		 * ssthresh is only set if packet loss occurred on a session.
1831 		 *
1832 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1833 		 * being torn down.  Ideally this code would not use 'so'.
1834 		 */
1835 		ssthresh = tp->snd_ssthresh;
1836 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1837 			/*
1838 			 * convert the limit from user data bytes to
1839 			 * packets then to packet data bytes.
1840 			 */
1841 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1842 			if (ssthresh < 2)
1843 				ssthresh = 2;
1844 			ssthresh *= (tp->t_maxseg +
1845 #ifdef INET6
1846 			    (isipv6 ? sizeof (struct ip6_hdr) +
1847 				sizeof (struct tcphdr) :
1848 #endif
1849 				sizeof (struct tcpiphdr)
1850 #ifdef INET6
1851 			    )
1852 #endif
1853 			    );
1854 		} else
1855 			ssthresh = 0;
1856 		metrics.rmx_ssthresh = ssthresh;
1857 
1858 		metrics.rmx_rtt = tp->t_srtt;
1859 		metrics.rmx_rttvar = tp->t_rttvar;
1860 		metrics.rmx_cwnd = tp->snd_cwnd;
1861 		metrics.rmx_sendpipe = 0;
1862 		metrics.rmx_recvpipe = 0;
1863 
1864 		tcp_hc_update(&inp->inp_inc, &metrics);
1865 	}
1866 
1867 	/* free the reassembly queue, if any */
1868 	tcp_reass_flush(tp);
1869 
1870 #ifdef TCP_OFFLOAD
1871 	/* Disconnect offload device, if any. */
1872 	if (tp->t_flags & TF_TOE)
1873 		tcp_offload_detach(tp);
1874 #endif
1875 
1876 	tcp_free_sackholes(tp);
1877 
1878 #ifdef TCPPCAP
1879 	/* Free the TCP PCAP queues. */
1880 	tcp_pcap_drain(&(tp->t_inpkts));
1881 	tcp_pcap_drain(&(tp->t_outpkts));
1882 #endif
1883 
1884 	/* Allow the CC algorithm to clean up after itself. */
1885 	if (CC_ALGO(tp)->cb_destroy != NULL)
1886 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1887 	CC_DATA(tp) = NULL;
1888 
1889 #ifdef TCP_HHOOK
1890 	khelp_destroy_osd(tp->osd);
1891 #endif
1892 
1893 	CC_ALGO(tp) = NULL;
1894 	inp->inp_ppcb = NULL;
1895 	if (tp->t_timers->tt_draincnt == 0) {
1896 		/* We own the last reference on tcpcb, let's free it. */
1897 #ifdef TCP_BLACKBOX
1898 		tcp_log_tcpcbfini(tp);
1899 #endif
1900 		TCPSTATES_DEC(tp->t_state);
1901 		if (tp->t_fb->tfb_tcp_fb_fini)
1902 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1903 		refcount_release(&tp->t_fb->tfb_refcnt);
1904 		tp->t_inpcb = NULL;
1905 		uma_zfree(V_tcpcb_zone, tp);
1906 		released = in_pcbrele_wlocked(inp);
1907 		KASSERT(!released, ("%s: inp %p should not have been released "
1908 			"here", __func__, inp));
1909 	}
1910 }
1911 
1912 void
1913 tcp_timer_discard(void *ptp)
1914 {
1915 	struct inpcb *inp;
1916 	struct tcpcb *tp;
1917 	struct epoch_tracker et;
1918 
1919 	tp = (struct tcpcb *)ptp;
1920 	CURVNET_SET(tp->t_vnet);
1921 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1922 	inp = tp->t_inpcb;
1923 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1924 		__func__, tp));
1925 	INP_WLOCK(inp);
1926 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1927 		("%s: tcpcb has to be stopped here", __func__));
1928 	tp->t_timers->tt_draincnt--;
1929 	if (tp->t_timers->tt_draincnt == 0) {
1930 		/* We own the last reference on this tcpcb, let's free it. */
1931 #ifdef TCP_BLACKBOX
1932 		tcp_log_tcpcbfini(tp);
1933 #endif
1934 		TCPSTATES_DEC(tp->t_state);
1935 		if (tp->t_fb->tfb_tcp_fb_fini)
1936 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1937 		refcount_release(&tp->t_fb->tfb_refcnt);
1938 		tp->t_inpcb = NULL;
1939 		uma_zfree(V_tcpcb_zone, tp);
1940 		if (in_pcbrele_wlocked(inp)) {
1941 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1942 			CURVNET_RESTORE();
1943 			return;
1944 		}
1945 	}
1946 	INP_WUNLOCK(inp);
1947 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1948 	CURVNET_RESTORE();
1949 }
1950 
1951 /*
1952  * Attempt to close a TCP control block, marking it as dropped, and freeing
1953  * the socket if we hold the only reference.
1954  */
1955 struct tcpcb *
1956 tcp_close(struct tcpcb *tp)
1957 {
1958 	struct inpcb *inp = tp->t_inpcb;
1959 	struct socket *so;
1960 
1961 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1962 	INP_WLOCK_ASSERT(inp);
1963 
1964 #ifdef TCP_OFFLOAD
1965 	if (tp->t_state == TCPS_LISTEN)
1966 		tcp_offload_listen_stop(tp);
1967 #endif
1968 	/*
1969 	 * This releases the TFO pending counter resource for TFO listen
1970 	 * sockets as well as passively-created TFO sockets that transition
1971 	 * from SYN_RECEIVED to CLOSED.
1972 	 */
1973 	if (tp->t_tfo_pending) {
1974 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1975 		tp->t_tfo_pending = NULL;
1976 	}
1977 	in_pcbdrop(inp);
1978 	TCPSTAT_INC(tcps_closed);
1979 	if (tp->t_state != TCPS_CLOSED)
1980 		tcp_state_change(tp, TCPS_CLOSED);
1981 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1982 	so = inp->inp_socket;
1983 	soisdisconnected(so);
1984 	if (inp->inp_flags & INP_SOCKREF) {
1985 		KASSERT(so->so_state & SS_PROTOREF,
1986 		    ("tcp_close: !SS_PROTOREF"));
1987 		inp->inp_flags &= ~INP_SOCKREF;
1988 		INP_WUNLOCK(inp);
1989 		SOCK_LOCK(so);
1990 		so->so_state &= ~SS_PROTOREF;
1991 		sofree(so);
1992 		return (NULL);
1993 	}
1994 	return (tp);
1995 }
1996 
1997 void
1998 tcp_drain(void)
1999 {
2000 	VNET_ITERATOR_DECL(vnet_iter);
2001 
2002 	if (!do_tcpdrain)
2003 		return;
2004 
2005 	VNET_LIST_RLOCK_NOSLEEP();
2006 	VNET_FOREACH(vnet_iter) {
2007 		CURVNET_SET(vnet_iter);
2008 		struct inpcb *inpb;
2009 		struct tcpcb *tcpb;
2010 
2011 	/*
2012 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2013 	 * if there is one...
2014 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2015 	 *      reassembly queue should be flushed, but in a situation
2016 	 *	where we're really low on mbufs, this is potentially
2017 	 *	useful.
2018 	 */
2019 		INP_INFO_WLOCK(&V_tcbinfo);
2020 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2021 			INP_WLOCK(inpb);
2022 			if (inpb->inp_flags & INP_TIMEWAIT) {
2023 				INP_WUNLOCK(inpb);
2024 				continue;
2025 			}
2026 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2027 				tcp_reass_flush(tcpb);
2028 				tcp_clean_sackreport(tcpb);
2029 #ifdef TCP_BLACKBOX
2030 				tcp_log_drain(tcpb);
2031 #endif
2032 #ifdef TCPPCAP
2033 				if (tcp_pcap_aggressive_free) {
2034 					/* Free the TCP PCAP queues. */
2035 					tcp_pcap_drain(&(tcpb->t_inpkts));
2036 					tcp_pcap_drain(&(tcpb->t_outpkts));
2037 				}
2038 #endif
2039 			}
2040 			INP_WUNLOCK(inpb);
2041 		}
2042 		INP_INFO_WUNLOCK(&V_tcbinfo);
2043 		CURVNET_RESTORE();
2044 	}
2045 	VNET_LIST_RUNLOCK_NOSLEEP();
2046 }
2047 
2048 /*
2049  * Notify a tcp user of an asynchronous error;
2050  * store error as soft error, but wake up user
2051  * (for now, won't do anything until can select for soft error).
2052  *
2053  * Do not wake up user since there currently is no mechanism for
2054  * reporting soft errors (yet - a kqueue filter may be added).
2055  */
2056 static struct inpcb *
2057 tcp_notify(struct inpcb *inp, int error)
2058 {
2059 	struct tcpcb *tp;
2060 
2061 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2062 	INP_WLOCK_ASSERT(inp);
2063 
2064 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2065 	    (inp->inp_flags & INP_DROPPED))
2066 		return (inp);
2067 
2068 	tp = intotcpcb(inp);
2069 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2070 
2071 	/*
2072 	 * Ignore some errors if we are hooked up.
2073 	 * If connection hasn't completed, has retransmitted several times,
2074 	 * and receives a second error, give up now.  This is better
2075 	 * than waiting a long time to establish a connection that
2076 	 * can never complete.
2077 	 */
2078 	if (tp->t_state == TCPS_ESTABLISHED &&
2079 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2080 	     error == EHOSTDOWN)) {
2081 		if (inp->inp_route.ro_rt) {
2082 			RTFREE(inp->inp_route.ro_rt);
2083 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2084 		}
2085 		return (inp);
2086 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2087 	    tp->t_softerror) {
2088 		tp = tcp_drop(tp, error);
2089 		if (tp != NULL)
2090 			return (inp);
2091 		else
2092 			return (NULL);
2093 	} else {
2094 		tp->t_softerror = error;
2095 		return (inp);
2096 	}
2097 #if 0
2098 	wakeup( &so->so_timeo);
2099 	sorwakeup(so);
2100 	sowwakeup(so);
2101 #endif
2102 }
2103 
2104 static int
2105 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2106 {
2107 	int error, i, m, n, pcb_count;
2108 	struct inpcb *inp, **inp_list;
2109 	inp_gen_t gencnt;
2110 	struct xinpgen xig;
2111 	struct epoch_tracker et;
2112 
2113 	/*
2114 	 * The process of preparing the TCB list is too time-consuming and
2115 	 * resource-intensive to repeat twice on every request.
2116 	 */
2117 	if (req->oldptr == NULL) {
2118 		n = V_tcbinfo.ipi_count +
2119 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2120 		n += imax(n / 8, 10);
2121 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2122 		return (0);
2123 	}
2124 
2125 	if (req->newptr != NULL)
2126 		return (EPERM);
2127 
2128 	/*
2129 	 * OK, now we're committed to doing something.
2130 	 */
2131 	INP_LIST_RLOCK(&V_tcbinfo);
2132 	gencnt = V_tcbinfo.ipi_gencnt;
2133 	n = V_tcbinfo.ipi_count;
2134 	INP_LIST_RUNLOCK(&V_tcbinfo);
2135 
2136 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2137 
2138 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2139 		+ (n + m) * sizeof(struct xtcpcb));
2140 	if (error != 0)
2141 		return (error);
2142 
2143 	bzero(&xig, sizeof(xig));
2144 	xig.xig_len = sizeof xig;
2145 	xig.xig_count = n + m;
2146 	xig.xig_gen = gencnt;
2147 	xig.xig_sogen = so_gencnt;
2148 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2149 	if (error)
2150 		return (error);
2151 
2152 	error = syncache_pcblist(req, m, &pcb_count);
2153 	if (error)
2154 		return (error);
2155 
2156 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2157 
2158 	INP_INFO_WLOCK(&V_tcbinfo);
2159 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2160 	    inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2161 		INP_WLOCK(inp);
2162 		if (inp->inp_gencnt <= gencnt) {
2163 			/*
2164 			 * XXX: This use of cr_cansee(), introduced with
2165 			 * TCP state changes, is not quite right, but for
2166 			 * now, better than nothing.
2167 			 */
2168 			if (inp->inp_flags & INP_TIMEWAIT) {
2169 				if (intotw(inp) != NULL)
2170 					error = cr_cansee(req->td->td_ucred,
2171 					    intotw(inp)->tw_cred);
2172 				else
2173 					error = EINVAL;	/* Skip this inp. */
2174 			} else
2175 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2176 			if (error == 0) {
2177 				in_pcbref(inp);
2178 				inp_list[i++] = inp;
2179 			}
2180 		}
2181 		INP_WUNLOCK(inp);
2182 	}
2183 	INP_INFO_WUNLOCK(&V_tcbinfo);
2184 	n = i;
2185 
2186 	error = 0;
2187 	for (i = 0; i < n; i++) {
2188 		inp = inp_list[i];
2189 		INP_RLOCK(inp);
2190 		if (inp->inp_gencnt <= gencnt) {
2191 			struct xtcpcb xt;
2192 
2193 			tcp_inptoxtp(inp, &xt);
2194 			INP_RUNLOCK(inp);
2195 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2196 		} else
2197 			INP_RUNLOCK(inp);
2198 	}
2199 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2200 	for (i = 0; i < n; i++) {
2201 		inp = inp_list[i];
2202 		INP_RLOCK(inp);
2203 		if (!in_pcbrele_rlocked(inp))
2204 			INP_RUNLOCK(inp);
2205 	}
2206 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2207 
2208 	if (!error) {
2209 		/*
2210 		 * Give the user an updated idea of our state.
2211 		 * If the generation differs from what we told
2212 		 * her before, she knows that something happened
2213 		 * while we were processing this request, and it
2214 		 * might be necessary to retry.
2215 		 */
2216 		INP_LIST_RLOCK(&V_tcbinfo);
2217 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2218 		xig.xig_sogen = so_gencnt;
2219 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2220 		INP_LIST_RUNLOCK(&V_tcbinfo);
2221 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2222 	}
2223 	free(inp_list, M_TEMP);
2224 	return (error);
2225 }
2226 
2227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2228     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2229     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2230 
2231 #ifdef INET
2232 static int
2233 tcp_getcred(SYSCTL_HANDLER_ARGS)
2234 {
2235 	struct xucred xuc;
2236 	struct sockaddr_in addrs[2];
2237 	struct inpcb *inp;
2238 	int error;
2239 
2240 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2241 	if (error)
2242 		return (error);
2243 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2244 	if (error)
2245 		return (error);
2246 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2247 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2248 	if (inp != NULL) {
2249 		if (inp->inp_socket == NULL)
2250 			error = ENOENT;
2251 		if (error == 0)
2252 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2253 		if (error == 0)
2254 			cru2x(inp->inp_cred, &xuc);
2255 		INP_RUNLOCK(inp);
2256 	} else
2257 		error = ENOENT;
2258 	if (error == 0)
2259 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2260 	return (error);
2261 }
2262 
2263 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2264     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2265     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2266 #endif /* INET */
2267 
2268 #ifdef INET6
2269 static int
2270 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2271 {
2272 	struct xucred xuc;
2273 	struct sockaddr_in6 addrs[2];
2274 	struct inpcb *inp;
2275 	int error;
2276 #ifdef INET
2277 	int mapped = 0;
2278 #endif
2279 
2280 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2281 	if (error)
2282 		return (error);
2283 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2284 	if (error)
2285 		return (error);
2286 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2287 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2288 		return (error);
2289 	}
2290 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2291 #ifdef INET
2292 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2293 			mapped = 1;
2294 		else
2295 #endif
2296 			return (EINVAL);
2297 	}
2298 
2299 #ifdef INET
2300 	if (mapped == 1)
2301 		inp = in_pcblookup(&V_tcbinfo,
2302 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2303 			addrs[1].sin6_port,
2304 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2305 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2306 	else
2307 #endif
2308 		inp = in6_pcblookup(&V_tcbinfo,
2309 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2310 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2311 			INPLOOKUP_RLOCKPCB, NULL);
2312 	if (inp != NULL) {
2313 		if (inp->inp_socket == NULL)
2314 			error = ENOENT;
2315 		if (error == 0)
2316 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2317 		if (error == 0)
2318 			cru2x(inp->inp_cred, &xuc);
2319 		INP_RUNLOCK(inp);
2320 	} else
2321 		error = ENOENT;
2322 	if (error == 0)
2323 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2324 	return (error);
2325 }
2326 
2327 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2328     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2329     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2330 #endif /* INET6 */
2331 
2332 
2333 #ifdef INET
2334 void
2335 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2336 {
2337 	struct ip *ip = vip;
2338 	struct tcphdr *th;
2339 	struct in_addr faddr;
2340 	struct inpcb *inp;
2341 	struct tcpcb *tp;
2342 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2343 	struct icmp *icp;
2344 	struct in_conninfo inc;
2345 	struct epoch_tracker et;
2346 	tcp_seq icmp_tcp_seq;
2347 	int mtu;
2348 
2349 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2350 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2351 		return;
2352 
2353 	if (cmd == PRC_MSGSIZE)
2354 		notify = tcp_mtudisc_notify;
2355 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2356 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2357 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2358 		notify = tcp_drop_syn_sent;
2359 
2360 	/*
2361 	 * Hostdead is ugly because it goes linearly through all PCBs.
2362 	 * XXX: We never get this from ICMP, otherwise it makes an
2363 	 * excellent DoS attack on machines with many connections.
2364 	 */
2365 	else if (cmd == PRC_HOSTDEAD)
2366 		ip = NULL;
2367 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2368 		return;
2369 
2370 	if (ip == NULL) {
2371 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2372 		return;
2373 	}
2374 
2375 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2376 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2377 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2378 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2379 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2380 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2381 		/* signal EHOSTDOWN, as it flushes the cached route */
2382 		inp = (*notify)(inp, EHOSTDOWN);
2383 		goto out;
2384 	}
2385 	icmp_tcp_seq = th->th_seq;
2386 	if (inp != NULL)  {
2387 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2388 		    !(inp->inp_flags & INP_DROPPED) &&
2389 		    !(inp->inp_socket == NULL)) {
2390 			tp = intotcpcb(inp);
2391 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2392 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2393 				if (cmd == PRC_MSGSIZE) {
2394 					/*
2395 					 * MTU discovery:
2396 					 * If we got a needfrag set the MTU
2397 					 * in the route to the suggested new
2398 					 * value (if given) and then notify.
2399 					 */
2400 					mtu = ntohs(icp->icmp_nextmtu);
2401 					/*
2402 					 * If no alternative MTU was
2403 					 * proposed, try the next smaller
2404 					 * one.
2405 					 */
2406 					if (!mtu)
2407 						mtu = ip_next_mtu(
2408 						    ntohs(ip->ip_len), 1);
2409 					if (mtu < V_tcp_minmss +
2410 					    sizeof(struct tcpiphdr))
2411 						mtu = V_tcp_minmss +
2412 						    sizeof(struct tcpiphdr);
2413 					/*
2414 					 * Only process the offered MTU if it
2415 					 * is smaller than the current one.
2416 					 */
2417 					if (mtu < tp->t_maxseg +
2418 					    sizeof(struct tcpiphdr)) {
2419 						bzero(&inc, sizeof(inc));
2420 						inc.inc_faddr = faddr;
2421 						inc.inc_fibnum =
2422 						    inp->inp_inc.inc_fibnum;
2423 						tcp_hc_updatemtu(&inc, mtu);
2424 						tcp_mtudisc(inp, mtu);
2425 					}
2426 				} else
2427 					inp = (*notify)(inp,
2428 					    inetctlerrmap[cmd]);
2429 			}
2430 		}
2431 	} else {
2432 		bzero(&inc, sizeof(inc));
2433 		inc.inc_fport = th->th_dport;
2434 		inc.inc_lport = th->th_sport;
2435 		inc.inc_faddr = faddr;
2436 		inc.inc_laddr = ip->ip_src;
2437 		syncache_unreach(&inc, icmp_tcp_seq);
2438 	}
2439 out:
2440 	if (inp != NULL)
2441 		INP_WUNLOCK(inp);
2442 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2443 }
2444 #endif /* INET */
2445 
2446 #ifdef INET6
2447 void
2448 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2449 {
2450 	struct in6_addr *dst;
2451 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2452 	struct ip6_hdr *ip6;
2453 	struct mbuf *m;
2454 	struct inpcb *inp;
2455 	struct tcpcb *tp;
2456 	struct icmp6_hdr *icmp6;
2457 	struct ip6ctlparam *ip6cp = NULL;
2458 	const struct sockaddr_in6 *sa6_src = NULL;
2459 	struct in_conninfo inc;
2460 	struct epoch_tracker et;
2461 	struct tcp_ports {
2462 		uint16_t th_sport;
2463 		uint16_t th_dport;
2464 	} t_ports;
2465 	tcp_seq icmp_tcp_seq;
2466 	unsigned int mtu;
2467 	unsigned int off;
2468 
2469 	if (sa->sa_family != AF_INET6 ||
2470 	    sa->sa_len != sizeof(struct sockaddr_in6))
2471 		return;
2472 
2473 	/* if the parameter is from icmp6, decode it. */
2474 	if (d != NULL) {
2475 		ip6cp = (struct ip6ctlparam *)d;
2476 		icmp6 = ip6cp->ip6c_icmp6;
2477 		m = ip6cp->ip6c_m;
2478 		ip6 = ip6cp->ip6c_ip6;
2479 		off = ip6cp->ip6c_off;
2480 		sa6_src = ip6cp->ip6c_src;
2481 		dst = ip6cp->ip6c_finaldst;
2482 	} else {
2483 		m = NULL;
2484 		ip6 = NULL;
2485 		off = 0;	/* fool gcc */
2486 		sa6_src = &sa6_any;
2487 		dst = NULL;
2488 	}
2489 
2490 	if (cmd == PRC_MSGSIZE)
2491 		notify = tcp_mtudisc_notify;
2492 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2493 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2494 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2495 		notify = tcp_drop_syn_sent;
2496 
2497 	/*
2498 	 * Hostdead is ugly because it goes linearly through all PCBs.
2499 	 * XXX: We never get this from ICMP, otherwise it makes an
2500 	 * excellent DoS attack on machines with many connections.
2501 	 */
2502 	else if (cmd == PRC_HOSTDEAD)
2503 		ip6 = NULL;
2504 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2505 		return;
2506 
2507 	if (ip6 == NULL) {
2508 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2509 			      (const struct sockaddr *)sa6_src,
2510 			      0, cmd, NULL, notify);
2511 		return;
2512 	}
2513 
2514 	/* Check if we can safely get the ports from the tcp hdr */
2515 	if (m == NULL ||
2516 	    (m->m_pkthdr.len <
2517 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2518 		return;
2519 	}
2520 	bzero(&t_ports, sizeof(struct tcp_ports));
2521 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2522 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2523 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2524 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2525 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2526 		/* signal EHOSTDOWN, as it flushes the cached route */
2527 		inp = (*notify)(inp, EHOSTDOWN);
2528 		goto out;
2529 	}
2530 	off += sizeof(struct tcp_ports);
2531 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2532 		goto out;
2533 	}
2534 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2535 	if (inp != NULL)  {
2536 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2537 		    !(inp->inp_flags & INP_DROPPED) &&
2538 		    !(inp->inp_socket == NULL)) {
2539 			tp = intotcpcb(inp);
2540 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2541 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2542 				if (cmd == PRC_MSGSIZE) {
2543 					/*
2544 					 * MTU discovery:
2545 					 * If we got a needfrag set the MTU
2546 					 * in the route to the suggested new
2547 					 * value (if given) and then notify.
2548 					 */
2549 					mtu = ntohl(icmp6->icmp6_mtu);
2550 					/*
2551 					 * If no alternative MTU was
2552 					 * proposed, or the proposed
2553 					 * MTU was too small, set to
2554 					 * the min.
2555 					 */
2556 					if (mtu < IPV6_MMTU)
2557 						mtu = IPV6_MMTU - 8;
2558 					bzero(&inc, sizeof(inc));
2559 					inc.inc_fibnum = M_GETFIB(m);
2560 					inc.inc_flags |= INC_ISIPV6;
2561 					inc.inc6_faddr = *dst;
2562 					if (in6_setscope(&inc.inc6_faddr,
2563 						m->m_pkthdr.rcvif, NULL))
2564 						goto out;
2565 					/*
2566 					 * Only process the offered MTU if it
2567 					 * is smaller than the current one.
2568 					 */
2569 					if (mtu < tp->t_maxseg +
2570 					    sizeof (struct tcphdr) +
2571 					    sizeof (struct ip6_hdr)) {
2572 						tcp_hc_updatemtu(&inc, mtu);
2573 						tcp_mtudisc(inp, mtu);
2574 						ICMP6STAT_INC(icp6s_pmtuchg);
2575 					}
2576 				} else
2577 					inp = (*notify)(inp,
2578 					    inet6ctlerrmap[cmd]);
2579 			}
2580 		}
2581 	} else {
2582 		bzero(&inc, sizeof(inc));
2583 		inc.inc_fibnum = M_GETFIB(m);
2584 		inc.inc_flags |= INC_ISIPV6;
2585 		inc.inc_fport = t_ports.th_dport;
2586 		inc.inc_lport = t_ports.th_sport;
2587 		inc.inc6_faddr = *dst;
2588 		inc.inc6_laddr = ip6->ip6_src;
2589 		syncache_unreach(&inc, icmp_tcp_seq);
2590 	}
2591 out:
2592 	if (inp != NULL)
2593 		INP_WUNLOCK(inp);
2594 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2595 }
2596 #endif /* INET6 */
2597 
2598 static uint32_t
2599 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2600 {
2601 	MD5_CTX ctx;
2602 	uint32_t hash[4];
2603 
2604 	MD5Init(&ctx);
2605 	MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2606 	MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2607 	switch (inc->inc_flags & INC_ISIPV6) {
2608 #ifdef INET
2609 	case 0:
2610 		MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2611 		MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2612 		break;
2613 #endif
2614 #ifdef INET6
2615 	case INC_ISIPV6:
2616 		MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2617 		MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2618 		break;
2619 #endif
2620 	}
2621 	MD5Update(&ctx, key, len);
2622 	MD5Final((unsigned char *)hash, &ctx);
2623 
2624 	return (hash[0]);
2625 }
2626 
2627 uint32_t
2628 tcp_new_ts_offset(struct in_conninfo *inc)
2629 {
2630 	return (tcp_keyed_hash(inc, V_ts_offset_secret,
2631 	    sizeof(V_ts_offset_secret)));
2632 }
2633 
2634 /*
2635  * Following is where TCP initial sequence number generation occurs.
2636  *
2637  * There are two places where we must use initial sequence numbers:
2638  * 1.  In SYN-ACK packets.
2639  * 2.  In SYN packets.
2640  *
2641  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2642  * tcp_syncache.c for details.
2643  *
2644  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2645  * depends on this property.  In addition, these ISNs should be
2646  * unguessable so as to prevent connection hijacking.  To satisfy
2647  * the requirements of this situation, the algorithm outlined in
2648  * RFC 1948 is used, with only small modifications.
2649  *
2650  * Implementation details:
2651  *
2652  * Time is based off the system timer, and is corrected so that it
2653  * increases by one megabyte per second.  This allows for proper
2654  * recycling on high speed LANs while still leaving over an hour
2655  * before rollover.
2656  *
2657  * As reading the *exact* system time is too expensive to be done
2658  * whenever setting up a TCP connection, we increment the time
2659  * offset in two ways.  First, a small random positive increment
2660  * is added to isn_offset for each connection that is set up.
2661  * Second, the function tcp_isn_tick fires once per clock tick
2662  * and increments isn_offset as necessary so that sequence numbers
2663  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2664  * random positive increments serve only to ensure that the same
2665  * exact sequence number is never sent out twice (as could otherwise
2666  * happen when a port is recycled in less than the system tick
2667  * interval.)
2668  *
2669  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2670  * between seeding of isn_secret.  This is normally set to zero,
2671  * as reseeding should not be necessary.
2672  *
2673  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2674  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2675  * general, this means holding an exclusive (write) lock.
2676  */
2677 
2678 #define ISN_BYTES_PER_SECOND 1048576
2679 #define ISN_STATIC_INCREMENT 4096
2680 #define ISN_RANDOM_INCREMENT (4096 - 1)
2681 #define ISN_SECRET_LENGTH    32
2682 
2683 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2684 VNET_DEFINE_STATIC(int, isn_last);
2685 VNET_DEFINE_STATIC(int, isn_last_reseed);
2686 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2687 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2688 
2689 #define	V_isn_secret			VNET(isn_secret)
2690 #define	V_isn_last			VNET(isn_last)
2691 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2692 #define	V_isn_offset			VNET(isn_offset)
2693 #define	V_isn_offset_old		VNET(isn_offset_old)
2694 
2695 tcp_seq
2696 tcp_new_isn(struct in_conninfo *inc)
2697 {
2698 	tcp_seq new_isn;
2699 	u_int32_t projected_offset;
2700 
2701 	ISN_LOCK();
2702 	/* Seed if this is the first use, reseed if requested. */
2703 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2704 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2705 		< (u_int)ticks))) {
2706 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2707 		V_isn_last_reseed = ticks;
2708 	}
2709 
2710 	/* Compute the md5 hash and return the ISN. */
2711 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2712 	    sizeof(V_isn_secret));
2713 	V_isn_offset += ISN_STATIC_INCREMENT +
2714 		(arc4random() & ISN_RANDOM_INCREMENT);
2715 	if (ticks != V_isn_last) {
2716 		projected_offset = V_isn_offset_old +
2717 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2718 		if (SEQ_GT(projected_offset, V_isn_offset))
2719 			V_isn_offset = projected_offset;
2720 		V_isn_offset_old = V_isn_offset;
2721 		V_isn_last = ticks;
2722 	}
2723 	new_isn += V_isn_offset;
2724 	ISN_UNLOCK();
2725 	return (new_isn);
2726 }
2727 
2728 /*
2729  * When a specific ICMP unreachable message is received and the
2730  * connection state is SYN-SENT, drop the connection.  This behavior
2731  * is controlled by the icmp_may_rst sysctl.
2732  */
2733 struct inpcb *
2734 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2735 {
2736 	struct tcpcb *tp;
2737 
2738 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2739 	INP_WLOCK_ASSERT(inp);
2740 
2741 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2742 	    (inp->inp_flags & INP_DROPPED))
2743 		return (inp);
2744 
2745 	tp = intotcpcb(inp);
2746 	if (tp->t_state != TCPS_SYN_SENT)
2747 		return (inp);
2748 
2749 	if (IS_FASTOPEN(tp->t_flags))
2750 		tcp_fastopen_disable_path(tp);
2751 
2752 	tp = tcp_drop(tp, errno);
2753 	if (tp != NULL)
2754 		return (inp);
2755 	else
2756 		return (NULL);
2757 }
2758 
2759 /*
2760  * When `need fragmentation' ICMP is received, update our idea of the MSS
2761  * based on the new value. Also nudge TCP to send something, since we
2762  * know the packet we just sent was dropped.
2763  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2764  */
2765 static struct inpcb *
2766 tcp_mtudisc_notify(struct inpcb *inp, int error)
2767 {
2768 
2769 	tcp_mtudisc(inp, -1);
2770 	return (inp);
2771 }
2772 
2773 static void
2774 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2775 {
2776 	struct tcpcb *tp;
2777 	struct socket *so;
2778 
2779 	INP_WLOCK_ASSERT(inp);
2780 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2781 	    (inp->inp_flags & INP_DROPPED))
2782 		return;
2783 
2784 	tp = intotcpcb(inp);
2785 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2786 
2787 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2788 
2789 	so = inp->inp_socket;
2790 	SOCKBUF_LOCK(&so->so_snd);
2791 	/* If the mss is larger than the socket buffer, decrease the mss. */
2792 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2793 		tp->t_maxseg = so->so_snd.sb_hiwat;
2794 	SOCKBUF_UNLOCK(&so->so_snd);
2795 
2796 	TCPSTAT_INC(tcps_mturesent);
2797 	tp->t_rtttime = 0;
2798 	tp->snd_nxt = tp->snd_una;
2799 	tcp_free_sackholes(tp);
2800 	tp->snd_recover = tp->snd_max;
2801 	if (tp->t_flags & TF_SACK_PERMIT)
2802 		EXIT_FASTRECOVERY(tp->t_flags);
2803 	tp->t_fb->tfb_tcp_output(tp);
2804 }
2805 
2806 #ifdef INET
2807 /*
2808  * Look-up the routing entry to the peer of this inpcb.  If no route
2809  * is found and it cannot be allocated, then return 0.  This routine
2810  * is called by TCP routines that access the rmx structure and by
2811  * tcp_mss_update to get the peer/interface MTU.
2812  */
2813 uint32_t
2814 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2815 {
2816 	struct nhop4_extended nh4;
2817 	struct ifnet *ifp;
2818 	uint32_t maxmtu = 0;
2819 
2820 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2821 
2822 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2823 
2824 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2825 		    NHR_REF, 0, &nh4) != 0)
2826 			return (0);
2827 
2828 		ifp = nh4.nh_ifp;
2829 		maxmtu = nh4.nh_mtu;
2830 
2831 		/* Report additional interface capabilities. */
2832 		if (cap != NULL) {
2833 			if (ifp->if_capenable & IFCAP_TSO4 &&
2834 			    ifp->if_hwassist & CSUM_TSO) {
2835 				cap->ifcap |= CSUM_TSO;
2836 				cap->tsomax = ifp->if_hw_tsomax;
2837 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2838 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2839 			}
2840 		}
2841 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2842 	}
2843 	return (maxmtu);
2844 }
2845 #endif /* INET */
2846 
2847 #ifdef INET6
2848 uint32_t
2849 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2850 {
2851 	struct nhop6_extended nh6;
2852 	struct in6_addr dst6;
2853 	uint32_t scopeid;
2854 	struct ifnet *ifp;
2855 	uint32_t maxmtu = 0;
2856 
2857 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2858 
2859 	if (inc->inc_flags & INC_IPV6MINMTU)
2860 		return (IPV6_MMTU);
2861 
2862 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2863 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2864 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2865 		    0, &nh6) != 0)
2866 			return (0);
2867 
2868 		ifp = nh6.nh_ifp;
2869 		maxmtu = nh6.nh_mtu;
2870 
2871 		/* Report additional interface capabilities. */
2872 		if (cap != NULL) {
2873 			if (ifp->if_capenable & IFCAP_TSO6 &&
2874 			    ifp->if_hwassist & CSUM_TSO) {
2875 				cap->ifcap |= CSUM_TSO;
2876 				cap->tsomax = ifp->if_hw_tsomax;
2877 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2878 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2879 			}
2880 		}
2881 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2882 	}
2883 
2884 	return (maxmtu);
2885 }
2886 #endif /* INET6 */
2887 
2888 /*
2889  * Calculate effective SMSS per RFC5681 definition for a given TCP
2890  * connection at its current state, taking into account SACK and etc.
2891  */
2892 u_int
2893 tcp_maxseg(const struct tcpcb *tp)
2894 {
2895 	u_int optlen;
2896 
2897 	if (tp->t_flags & TF_NOOPT)
2898 		return (tp->t_maxseg);
2899 
2900 	/*
2901 	 * Here we have a simplified code from tcp_addoptions(),
2902 	 * without a proper loop, and having most of paddings hardcoded.
2903 	 * We might make mistakes with padding here in some edge cases,
2904 	 * but this is harmless, since result of tcp_maxseg() is used
2905 	 * only in cwnd and ssthresh estimations.
2906 	 */
2907 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2908 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2909 		if (tp->t_flags & TF_RCVD_TSTMP)
2910 			optlen = TCPOLEN_TSTAMP_APPA;
2911 		else
2912 			optlen = 0;
2913 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2914 		if (tp->t_flags & TF_SIGNATURE)
2915 			optlen += PAD(TCPOLEN_SIGNATURE);
2916 #endif
2917 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2918 			optlen += TCPOLEN_SACKHDR;
2919 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2920 			optlen = PAD(optlen);
2921 		}
2922 	} else {
2923 		if (tp->t_flags & TF_REQ_TSTMP)
2924 			optlen = TCPOLEN_TSTAMP_APPA;
2925 		else
2926 			optlen = PAD(TCPOLEN_MAXSEG);
2927 		if (tp->t_flags & TF_REQ_SCALE)
2928 			optlen += PAD(TCPOLEN_WINDOW);
2929 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2930 		if (tp->t_flags & TF_SIGNATURE)
2931 			optlen += PAD(TCPOLEN_SIGNATURE);
2932 #endif
2933 		if (tp->t_flags & TF_SACK_PERMIT)
2934 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2935 	}
2936 #undef PAD
2937 	optlen = min(optlen, TCP_MAXOLEN);
2938 	return (tp->t_maxseg - optlen);
2939 }
2940 
2941 static int
2942 sysctl_drop(SYSCTL_HANDLER_ARGS)
2943 {
2944 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2945 	struct sockaddr_storage addrs[2];
2946 	struct inpcb *inp;
2947 	struct tcpcb *tp;
2948 	struct tcptw *tw;
2949 	struct sockaddr_in *fin, *lin;
2950 	struct epoch_tracker et;
2951 #ifdef INET6
2952 	struct sockaddr_in6 *fin6, *lin6;
2953 #endif
2954 	int error;
2955 
2956 	inp = NULL;
2957 	fin = lin = NULL;
2958 #ifdef INET6
2959 	fin6 = lin6 = NULL;
2960 #endif
2961 	error = 0;
2962 
2963 	if (req->oldptr != NULL || req->oldlen != 0)
2964 		return (EINVAL);
2965 	if (req->newptr == NULL)
2966 		return (EPERM);
2967 	if (req->newlen < sizeof(addrs))
2968 		return (ENOMEM);
2969 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2970 	if (error)
2971 		return (error);
2972 
2973 	switch (addrs[0].ss_family) {
2974 #ifdef INET6
2975 	case AF_INET6:
2976 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2977 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2978 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2979 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2980 			return (EINVAL);
2981 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2982 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2983 				return (EINVAL);
2984 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2985 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2986 			fin = (struct sockaddr_in *)&addrs[0];
2987 			lin = (struct sockaddr_in *)&addrs[1];
2988 			break;
2989 		}
2990 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2991 		if (error)
2992 			return (error);
2993 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2994 		if (error)
2995 			return (error);
2996 		break;
2997 #endif
2998 #ifdef INET
2999 	case AF_INET:
3000 		fin = (struct sockaddr_in *)&addrs[0];
3001 		lin = (struct sockaddr_in *)&addrs[1];
3002 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3003 		    lin->sin_len != sizeof(struct sockaddr_in))
3004 			return (EINVAL);
3005 		break;
3006 #endif
3007 	default:
3008 		return (EINVAL);
3009 	}
3010 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3011 	switch (addrs[0].ss_family) {
3012 #ifdef INET6
3013 	case AF_INET6:
3014 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3015 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3016 		    INPLOOKUP_WLOCKPCB, NULL);
3017 		break;
3018 #endif
3019 #ifdef INET
3020 	case AF_INET:
3021 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3022 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3023 		break;
3024 #endif
3025 	}
3026 	if (inp != NULL) {
3027 		if (inp->inp_flags & INP_TIMEWAIT) {
3028 			/*
3029 			 * XXXRW: There currently exists a state where an
3030 			 * inpcb is present, but its timewait state has been
3031 			 * discarded.  For now, don't allow dropping of this
3032 			 * type of inpcb.
3033 			 */
3034 			tw = intotw(inp);
3035 			if (tw != NULL)
3036 				tcp_twclose(tw, 0);
3037 			else
3038 				INP_WUNLOCK(inp);
3039 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3040 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3041 			tp = intotcpcb(inp);
3042 			tp = tcp_drop(tp, ECONNABORTED);
3043 			if (tp != NULL)
3044 				INP_WUNLOCK(inp);
3045 		} else
3046 			INP_WUNLOCK(inp);
3047 	} else
3048 		error = ESRCH;
3049 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3050 	return (error);
3051 }
3052 
3053 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3054     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3055     0, sysctl_drop, "", "Drop TCP connection");
3056 
3057 /*
3058  * Generate a standardized TCP log line for use throughout the
3059  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3060  * allow use in the interrupt context.
3061  *
3062  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3063  * NB: The function may return NULL if memory allocation failed.
3064  *
3065  * Due to header inclusion and ordering limitations the struct ip
3066  * and ip6_hdr pointers have to be passed as void pointers.
3067  */
3068 char *
3069 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3070     const void *ip6hdr)
3071 {
3072 
3073 	/* Is logging enabled? */
3074 	if (tcp_log_in_vain == 0)
3075 		return (NULL);
3076 
3077 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3078 }
3079 
3080 char *
3081 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3082     const void *ip6hdr)
3083 {
3084 
3085 	/* Is logging enabled? */
3086 	if (tcp_log_debug == 0)
3087 		return (NULL);
3088 
3089 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3090 }
3091 
3092 static char *
3093 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3094     const void *ip6hdr)
3095 {
3096 	char *s, *sp;
3097 	size_t size;
3098 	struct ip *ip;
3099 #ifdef INET6
3100 	const struct ip6_hdr *ip6;
3101 
3102 	ip6 = (const struct ip6_hdr *)ip6hdr;
3103 #endif /* INET6 */
3104 	ip = (struct ip *)ip4hdr;
3105 
3106 	/*
3107 	 * The log line looks like this:
3108 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3109 	 */
3110 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3111 	    sizeof(PRINT_TH_FLAGS) + 1 +
3112 #ifdef INET6
3113 	    2 * INET6_ADDRSTRLEN;
3114 #else
3115 	    2 * INET_ADDRSTRLEN;
3116 #endif /* INET6 */
3117 
3118 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3119 	if (s == NULL)
3120 		return (NULL);
3121 
3122 	strcat(s, "TCP: [");
3123 	sp = s + strlen(s);
3124 
3125 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3126 		inet_ntoa_r(inc->inc_faddr, sp);
3127 		sp = s + strlen(s);
3128 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3129 		sp = s + strlen(s);
3130 		inet_ntoa_r(inc->inc_laddr, sp);
3131 		sp = s + strlen(s);
3132 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3133 #ifdef INET6
3134 	} else if (inc) {
3135 		ip6_sprintf(sp, &inc->inc6_faddr);
3136 		sp = s + strlen(s);
3137 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3138 		sp = s + strlen(s);
3139 		ip6_sprintf(sp, &inc->inc6_laddr);
3140 		sp = s + strlen(s);
3141 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3142 	} else if (ip6 && th) {
3143 		ip6_sprintf(sp, &ip6->ip6_src);
3144 		sp = s + strlen(s);
3145 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3146 		sp = s + strlen(s);
3147 		ip6_sprintf(sp, &ip6->ip6_dst);
3148 		sp = s + strlen(s);
3149 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3150 #endif /* INET6 */
3151 #ifdef INET
3152 	} else if (ip && th) {
3153 		inet_ntoa_r(ip->ip_src, sp);
3154 		sp = s + strlen(s);
3155 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3156 		sp = s + strlen(s);
3157 		inet_ntoa_r(ip->ip_dst, sp);
3158 		sp = s + strlen(s);
3159 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3160 #endif /* INET */
3161 	} else {
3162 		free(s, M_TCPLOG);
3163 		return (NULL);
3164 	}
3165 	sp = s + strlen(s);
3166 	if (th)
3167 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3168 	if (*(s + size - 1) != '\0')
3169 		panic("%s: string too long", __func__);
3170 	return (s);
3171 }
3172 
3173 /*
3174  * A subroutine which makes it easy to track TCP state changes with DTrace.
3175  * This function shouldn't be called for t_state initializations that don't
3176  * correspond to actual TCP state transitions.
3177  */
3178 void
3179 tcp_state_change(struct tcpcb *tp, int newstate)
3180 {
3181 #if defined(KDTRACE_HOOKS)
3182 	int pstate = tp->t_state;
3183 #endif
3184 
3185 	TCPSTATES_DEC(tp->t_state);
3186 	TCPSTATES_INC(newstate);
3187 	tp->t_state = newstate;
3188 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3189 }
3190 
3191 /*
3192  * Create an external-format (``xtcpcb'') structure using the information in
3193  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3194  * reduce the spew of irrelevant information over this interface, to isolate
3195  * user code from changes in the kernel structure, and potentially to provide
3196  * information-hiding if we decide that some of this information should be
3197  * hidden from users.
3198  */
3199 void
3200 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3201 {
3202 	struct tcpcb *tp = intotcpcb(inp);
3203 	sbintime_t now;
3204 
3205 	bzero(xt, sizeof(*xt));
3206 	if (inp->inp_flags & INP_TIMEWAIT) {
3207 		xt->t_state = TCPS_TIME_WAIT;
3208 	} else {
3209 		xt->t_state = tp->t_state;
3210 		xt->t_logstate = tp->t_logstate;
3211 		xt->t_flags = tp->t_flags;
3212 		xt->t_sndzerowin = tp->t_sndzerowin;
3213 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3214 		xt->t_rcvoopack = tp->t_rcvoopack;
3215 
3216 		now = getsbinuptime();
3217 #define	COPYTIMER(ttt)	do {						\
3218 		if (callout_active(&tp->t_timers->ttt))			\
3219 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3220 			    SBT_1MS;					\
3221 		else							\
3222 			xt->ttt = 0;					\
3223 } while (0)
3224 		COPYTIMER(tt_delack);
3225 		COPYTIMER(tt_rexmt);
3226 		COPYTIMER(tt_persist);
3227 		COPYTIMER(tt_keep);
3228 		COPYTIMER(tt_2msl);
3229 #undef COPYTIMER
3230 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3231 
3232 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3233 		    TCP_FUNCTION_NAME_LEN_MAX);
3234 #ifdef TCP_BLACKBOX
3235 		(void)tcp_log_get_id(tp, xt->xt_logid);
3236 #endif
3237 	}
3238 
3239 	xt->xt_len = sizeof(struct xtcpcb);
3240 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3241 	if (inp->inp_socket == NULL)
3242 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3243 }
3244