xref: /freebsd/sys/netinet/tcp_subr.c (revision c1cdf6a42f0d951ba720688dfc6ce07608b02f6e)
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 int t_functions_inited = 0;
261 static int tcp_fb_cnt = 0;
262 struct tcp_funchead t_functions;
263 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
264 
265 static void
266 init_tcp_functions(void)
267 {
268 	if (t_functions_inited == 0) {
269 		TAILQ_INIT(&t_functions);
270 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
271 		t_functions_inited = 1;
272 	}
273 }
274 
275 static struct tcp_function_block *
276 find_tcp_functions_locked(struct tcp_function_set *fs)
277 {
278 	struct tcp_function *f;
279 	struct tcp_function_block *blk=NULL;
280 
281 	TAILQ_FOREACH(f, &t_functions, tf_next) {
282 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
283 			blk = f->tf_fb;
284 			break;
285 		}
286 	}
287 	return(blk);
288 }
289 
290 static struct tcp_function_block *
291 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
292 {
293 	struct tcp_function_block *rblk=NULL;
294 	struct tcp_function *f;
295 
296 	TAILQ_FOREACH(f, &t_functions, tf_next) {
297 		if (f->tf_fb == blk) {
298 			rblk = blk;
299 			if (s) {
300 				*s = f;
301 			}
302 			break;
303 		}
304 	}
305 	return (rblk);
306 }
307 
308 struct tcp_function_block *
309 find_and_ref_tcp_functions(struct tcp_function_set *fs)
310 {
311 	struct tcp_function_block *blk;
312 
313 	rw_rlock(&tcp_function_lock);
314 	blk = find_tcp_functions_locked(fs);
315 	if (blk)
316 		refcount_acquire(&blk->tfb_refcnt);
317 	rw_runlock(&tcp_function_lock);
318 	return(blk);
319 }
320 
321 struct tcp_function_block *
322 find_and_ref_tcp_fb(struct tcp_function_block *blk)
323 {
324 	struct tcp_function_block *rblk;
325 
326 	rw_rlock(&tcp_function_lock);
327 	rblk = find_tcp_fb_locked(blk, NULL);
328 	if (rblk)
329 		refcount_acquire(&rblk->tfb_refcnt);
330 	rw_runlock(&tcp_function_lock);
331 	return(rblk);
332 }
333 
334 static struct tcp_function_block *
335 find_and_ref_tcp_default_fb(void)
336 {
337 	struct tcp_function_block *rblk;
338 
339 	rw_rlock(&tcp_function_lock);
340 	rblk = tcp_func_set_ptr;
341 	refcount_acquire(&rblk->tfb_refcnt);
342 	rw_runlock(&tcp_function_lock);
343 	return (rblk);
344 }
345 
346 void
347 tcp_switch_back_to_default(struct tcpcb *tp)
348 {
349 	struct tcp_function_block *tfb;
350 
351 	KASSERT(tp->t_fb != &tcp_def_funcblk,
352 	    ("%s: called by the built-in default stack", __func__));
353 
354 	/*
355 	 * Release the old stack. This function will either find a new one
356 	 * or panic.
357 	 */
358 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
359 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
360 	refcount_release(&tp->t_fb->tfb_refcnt);
361 
362 	/*
363 	 * Now, we'll find a new function block to use.
364 	 * Start by trying the current user-selected
365 	 * default, unless this stack is the user-selected
366 	 * default.
367 	 */
368 	tfb = find_and_ref_tcp_default_fb();
369 	if (tfb == tp->t_fb) {
370 		refcount_release(&tfb->tfb_refcnt);
371 		tfb = NULL;
372 	}
373 	/* Does the stack accept this connection? */
374 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
375 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
376 		refcount_release(&tfb->tfb_refcnt);
377 		tfb = NULL;
378 	}
379 	/* Try to use that stack. */
380 	if (tfb != NULL) {
381 		/* Initialize the new stack. If it succeeds, we are done. */
382 		tp->t_fb = tfb;
383 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
384 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
385 			return;
386 
387 		/*
388 		 * Initialization failed. Release the reference count on
389 		 * the stack.
390 		 */
391 		refcount_release(&tfb->tfb_refcnt);
392 	}
393 
394 	/*
395 	 * If that wasn't feasible, use the built-in default
396 	 * stack which is not allowed to reject anyone.
397 	 */
398 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
399 	if (tfb == NULL) {
400 		/* there always should be a default */
401 		panic("Can't refer to tcp_def_funcblk");
402 	}
403 	if (tfb->tfb_tcp_handoff_ok != NULL) {
404 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
405 			/* The default stack cannot say no */
406 			panic("Default stack rejects a new session?");
407 		}
408 	}
409 	tp->t_fb = tfb;
410 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
411 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
412 		/* The default stack cannot fail */
413 		panic("Default stack initialization failed");
414 	}
415 }
416 
417 static int
418 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
419 {
420 	int error=ENOENT;
421 	struct tcp_function_set fs;
422 	struct tcp_function_block *blk;
423 
424 	memset(&fs, 0, sizeof(fs));
425 	rw_rlock(&tcp_function_lock);
426 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
427 	if (blk) {
428 		/* Found him */
429 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
430 		fs.pcbcnt = blk->tfb_refcnt;
431 	}
432 	rw_runlock(&tcp_function_lock);
433 	error = sysctl_handle_string(oidp, fs.function_set_name,
434 				     sizeof(fs.function_set_name), req);
435 
436 	/* Check for error or no change */
437 	if (error != 0 || req->newptr == NULL)
438 		return(error);
439 
440 	rw_wlock(&tcp_function_lock);
441 	blk = find_tcp_functions_locked(&fs);
442 	if ((blk == NULL) ||
443 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
444 		error = ENOENT;
445 		goto done;
446 	}
447 	tcp_func_set_ptr = blk;
448 done:
449 	rw_wunlock(&tcp_function_lock);
450 	return (error);
451 }
452 
453 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
454 	    CTLTYPE_STRING | CTLFLAG_RW,
455 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
456 	    "Set/get the default TCP functions");
457 
458 static int
459 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
460 {
461 	int error, cnt, linesz;
462 	struct tcp_function *f;
463 	char *buffer, *cp;
464 	size_t bufsz, outsz;
465 	bool alias;
466 
467 	cnt = 0;
468 	rw_rlock(&tcp_function_lock);
469 	TAILQ_FOREACH(f, &t_functions, tf_next) {
470 		cnt++;
471 	}
472 	rw_runlock(&tcp_function_lock);
473 
474 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
475 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
476 
477 	error = 0;
478 	cp = buffer;
479 
480 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
481 	    "Alias", "PCB count");
482 	cp += linesz;
483 	bufsz -= linesz;
484 	outsz = linesz;
485 
486 	rw_rlock(&tcp_function_lock);
487 	TAILQ_FOREACH(f, &t_functions, tf_next) {
488 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
489 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
490 		    f->tf_fb->tfb_tcp_block_name,
491 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
492 		    alias ? f->tf_name : "-",
493 		    f->tf_fb->tfb_refcnt);
494 		if (linesz >= bufsz) {
495 			error = EOVERFLOW;
496 			break;
497 		}
498 		cp += linesz;
499 		bufsz -= linesz;
500 		outsz += linesz;
501 	}
502 	rw_runlock(&tcp_function_lock);
503 	if (error == 0)
504 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
505 	free(buffer, M_TEMP);
506 	return (error);
507 }
508 
509 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
510 	    CTLTYPE_STRING|CTLFLAG_RD,
511 	    NULL, 0, sysctl_net_inet_list_available, "A",
512 	    "list available TCP Function sets");
513 
514 /*
515  * Exports one (struct tcp_function_info) for each alias/name.
516  */
517 static int
518 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
519 {
520 	int cnt, error;
521 	struct tcp_function *f;
522 	struct tcp_function_info tfi;
523 
524 	/*
525 	 * We don't allow writes.
526 	 */
527 	if (req->newptr != NULL)
528 		return (EINVAL);
529 
530 	/*
531 	 * Wire the old buffer so we can directly copy the functions to
532 	 * user space without dropping the lock.
533 	 */
534 	if (req->oldptr != NULL) {
535 		error = sysctl_wire_old_buffer(req, 0);
536 		if (error)
537 			return (error);
538 	}
539 
540 	/*
541 	 * Walk the list and copy out matching entries. If INVARIANTS
542 	 * is compiled in, also walk the list to verify the length of
543 	 * the list matches what we have recorded.
544 	 */
545 	rw_rlock(&tcp_function_lock);
546 
547 	cnt = 0;
548 #ifndef INVARIANTS
549 	if (req->oldptr == NULL) {
550 		cnt = tcp_fb_cnt;
551 		goto skip_loop;
552 	}
553 #endif
554 	TAILQ_FOREACH(f, &t_functions, tf_next) {
555 #ifdef INVARIANTS
556 		cnt++;
557 #endif
558 		if (req->oldptr != NULL) {
559 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
560 			tfi.tfi_id = f->tf_fb->tfb_id;
561 			(void)strncpy(tfi.tfi_alias, f->tf_name,
562 			    TCP_FUNCTION_NAME_LEN_MAX);
563 			tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
564 			(void)strncpy(tfi.tfi_name,
565 			    f->tf_fb->tfb_tcp_block_name,
566 			    TCP_FUNCTION_NAME_LEN_MAX);
567 			tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
568 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
569 			/*
570 			 * Don't stop on error, as that is the
571 			 * mechanism we use to accumulate length
572 			 * information if the buffer was too short.
573 			 */
574 		}
575 	}
576 	KASSERT(cnt == tcp_fb_cnt,
577 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
578 #ifndef INVARIANTS
579 skip_loop:
580 #endif
581 	rw_runlock(&tcp_function_lock);
582 	if (req->oldptr == NULL)
583 		error = SYSCTL_OUT(req, NULL,
584 		    (cnt + 1) * sizeof(struct tcp_function_info));
585 
586 	return (error);
587 }
588 
589 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
590 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
591 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
592 	    "List TCP function block name-to-ID mappings");
593 
594 /*
595  * tfb_tcp_handoff_ok() function for the default stack.
596  * Note that we'll basically try to take all comers.
597  */
598 static int
599 tcp_default_handoff_ok(struct tcpcb *tp)
600 {
601 
602 	return (0);
603 }
604 
605 /*
606  * tfb_tcp_fb_init() function for the default stack.
607  *
608  * This handles making sure we have appropriate timers set if you are
609  * transitioning a socket that has some amount of setup done.
610  *
611  * The init() fuction from the default can *never* return non-zero i.e.
612  * it is required to always succeed since it is the stack of last resort!
613  */
614 static int
615 tcp_default_fb_init(struct tcpcb *tp)
616 {
617 
618 	struct socket *so;
619 
620 	INP_WLOCK_ASSERT(tp->t_inpcb);
621 
622 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
623 	    ("%s: connection %p in unexpected state %d", __func__, tp,
624 	    tp->t_state));
625 
626 	/*
627 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
628 	 * know what to do for unexpected states (which includes TIME_WAIT).
629 	 */
630 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
631 		return (0);
632 
633 	/*
634 	 * Make sure some kind of transmission timer is set if there is
635 	 * outstanding data.
636 	 */
637 	so = tp->t_inpcb->inp_socket;
638 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
639 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
640 	    tcp_timer_active(tp, TT_PERSIST))) {
641 		/*
642 		 * If the session has established and it looks like it should
643 		 * be in the persist state, set the persist timer. Otherwise,
644 		 * set the retransmit timer.
645 		 */
646 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
647 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
648 		    (int32_t)sbavail(&so->so_snd))
649 			tcp_setpersist(tp);
650 		else
651 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
652 	}
653 
654 	/* All non-embryonic sessions get a keepalive timer. */
655 	if (!tcp_timer_active(tp, TT_KEEP))
656 		tcp_timer_activate(tp, TT_KEEP,
657 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
658 		    TP_KEEPINIT(tp));
659 
660 	return (0);
661 }
662 
663 /*
664  * tfb_tcp_fb_fini() function for the default stack.
665  *
666  * This changes state as necessary (or prudent) to prepare for another stack
667  * to assume responsibility for the connection.
668  */
669 static void
670 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
671 {
672 
673 	INP_WLOCK_ASSERT(tp->t_inpcb);
674 	return;
675 }
676 
677 /*
678  * Target size of TCP PCB hash tables. Must be a power of two.
679  *
680  * Note that this can be overridden by the kernel environment
681  * variable net.inet.tcp.tcbhashsize
682  */
683 #ifndef TCBHASHSIZE
684 #define TCBHASHSIZE	0
685 #endif
686 
687 /*
688  * XXX
689  * Callouts should be moved into struct tcp directly.  They are currently
690  * separate because the tcpcb structure is exported to userland for sysctl
691  * parsing purposes, which do not know about callouts.
692  */
693 struct tcpcb_mem {
694 	struct	tcpcb		tcb;
695 	struct	tcp_timer	tt;
696 	struct	cc_var		ccv;
697 #ifdef TCP_HHOOK
698 	struct	osd		osd;
699 #endif
700 };
701 
702 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
703 #define	V_tcpcb_zone			VNET(tcpcb_zone)
704 
705 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
706 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
707 
708 static struct mtx isn_mtx;
709 
710 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
711 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
712 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
713 
714 /*
715  * TCP initialization.
716  */
717 static void
718 tcp_zone_change(void *tag)
719 {
720 
721 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
722 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
723 	tcp_tw_zone_change();
724 }
725 
726 static int
727 tcp_inpcb_init(void *mem, int size, int flags)
728 {
729 	struct inpcb *inp = mem;
730 
731 	INP_LOCK_INIT(inp, "inp", "tcpinp");
732 	return (0);
733 }
734 
735 /*
736  * Take a value and get the next power of 2 that doesn't overflow.
737  * Used to size the tcp_inpcb hash buckets.
738  */
739 static int
740 maketcp_hashsize(int size)
741 {
742 	int hashsize;
743 
744 	/*
745 	 * auto tune.
746 	 * get the next power of 2 higher than maxsockets.
747 	 */
748 	hashsize = 1 << fls(size);
749 	/* catch overflow, and just go one power of 2 smaller */
750 	if (hashsize < size) {
751 		hashsize = 1 << (fls(size) - 1);
752 	}
753 	return (hashsize);
754 }
755 
756 static volatile int next_tcp_stack_id = 1;
757 
758 /*
759  * Register a TCP function block with the name provided in the names
760  * array.  (Note that this function does NOT automatically register
761  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
762  * explicitly include blk->tfb_tcp_block_name in the list of names if
763  * you wish to register the stack with that name.)
764  *
765  * Either all name registrations will succeed or all will fail.  If
766  * a name registration fails, the function will update the num_names
767  * argument to point to the array index of the name that encountered
768  * the failure.
769  *
770  * Returns 0 on success, or an error code on failure.
771  */
772 int
773 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
774     const char *names[], int *num_names)
775 {
776 	struct tcp_function *n;
777 	struct tcp_function_set fs;
778 	int error, i;
779 
780 	KASSERT(names != NULL && *num_names > 0,
781 	    ("%s: Called with 0-length name list", __func__));
782 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
783 
784 	if (t_functions_inited == 0) {
785 		init_tcp_functions();
786 	}
787 	if ((blk->tfb_tcp_output == NULL) ||
788 	    (blk->tfb_tcp_do_segment == NULL) ||
789 	    (blk->tfb_tcp_ctloutput == NULL) ||
790 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
791 		/*
792 		 * These functions are required and you
793 		 * need a name.
794 		 */
795 		*num_names = 0;
796 		return (EINVAL);
797 	}
798 	if (blk->tfb_tcp_timer_stop_all ||
799 	    blk->tfb_tcp_timer_activate ||
800 	    blk->tfb_tcp_timer_active ||
801 	    blk->tfb_tcp_timer_stop) {
802 		/*
803 		 * If you define one timer function you
804 		 * must have them all.
805 		 */
806 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
807 		    (blk->tfb_tcp_timer_activate == NULL) ||
808 		    (blk->tfb_tcp_timer_active == NULL) ||
809 		    (blk->tfb_tcp_timer_stop == NULL)) {
810 			*num_names = 0;
811 			return (EINVAL);
812 		}
813 	}
814 
815 	refcount_init(&blk->tfb_refcnt, 0);
816 	blk->tfb_flags = 0;
817 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
818 	for (i = 0; i < *num_names; i++) {
819 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
820 		if (n == NULL) {
821 			error = ENOMEM;
822 			goto cleanup;
823 		}
824 		n->tf_fb = blk;
825 
826 		(void)strncpy(fs.function_set_name, names[i],
827 		    TCP_FUNCTION_NAME_LEN_MAX);
828 		fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
829 		rw_wlock(&tcp_function_lock);
830 		if (find_tcp_functions_locked(&fs) != NULL) {
831 			/* Duplicate name space not allowed */
832 			rw_wunlock(&tcp_function_lock);
833 			free(n, M_TCPFUNCTIONS);
834 			error = EALREADY;
835 			goto cleanup;
836 		}
837 		(void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
838 		n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
839 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
840 		tcp_fb_cnt++;
841 		rw_wunlock(&tcp_function_lock);
842 	}
843 	return(0);
844 
845 cleanup:
846 	/*
847 	 * Deregister the names we just added. Because registration failed
848 	 * for names[i], we don't need to deregister that name.
849 	 */
850 	*num_names = i;
851 	rw_wlock(&tcp_function_lock);
852 	while (--i >= 0) {
853 		TAILQ_FOREACH(n, &t_functions, tf_next) {
854 			if (!strncmp(n->tf_name, names[i],
855 			    TCP_FUNCTION_NAME_LEN_MAX)) {
856 				TAILQ_REMOVE(&t_functions, n, tf_next);
857 				tcp_fb_cnt--;
858 				n->tf_fb = NULL;
859 				free(n, M_TCPFUNCTIONS);
860 				break;
861 			}
862 		}
863 	}
864 	rw_wunlock(&tcp_function_lock);
865 	return (error);
866 }
867 
868 /*
869  * Register a TCP function block using the name provided in the name
870  * argument.
871  *
872  * Returns 0 on success, or an error code on failure.
873  */
874 int
875 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
876     int wait)
877 {
878 	const char *name_list[1];
879 	int num_names, rv;
880 
881 	num_names = 1;
882 	if (name != NULL)
883 		name_list[0] = name;
884 	else
885 		name_list[0] = blk->tfb_tcp_block_name;
886 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
887 	return (rv);
888 }
889 
890 /*
891  * Register a TCP function block using the name defined in
892  * blk->tfb_tcp_block_name.
893  *
894  * Returns 0 on success, or an error code on failure.
895  */
896 int
897 register_tcp_functions(struct tcp_function_block *blk, int wait)
898 {
899 
900 	return (register_tcp_functions_as_name(blk, NULL, wait));
901 }
902 
903 /*
904  * Deregister all names associated with a function block. This
905  * functionally removes the function block from use within the system.
906  *
907  * When called with a true quiesce argument, mark the function block
908  * as being removed so no more stacks will use it and determine
909  * whether the removal would succeed.
910  *
911  * When called with a false quiesce argument, actually attempt the
912  * removal.
913  *
914  * When called with a force argument, attempt to switch all TCBs to
915  * use the default stack instead of returning EBUSY.
916  *
917  * Returns 0 on success (or if the removal would succeed, or an error
918  * code on failure.
919  */
920 int
921 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
922     bool force)
923 {
924 	struct tcp_function *f;
925 
926 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
927 		/* You can't un-register the default */
928 		return (EPERM);
929 	}
930 	rw_wlock(&tcp_function_lock);
931 	if (blk == tcp_func_set_ptr) {
932 		/* You can't free the current default */
933 		rw_wunlock(&tcp_function_lock);
934 		return (EBUSY);
935 	}
936 	/* Mark the block so no more stacks can use it. */
937 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
938 	/*
939 	 * If TCBs are still attached to the stack, attempt to switch them
940 	 * to the default stack.
941 	 */
942 	if (force && blk->tfb_refcnt) {
943 		struct inpcb *inp;
944 		struct tcpcb *tp;
945 		VNET_ITERATOR_DECL(vnet_iter);
946 
947 		rw_wunlock(&tcp_function_lock);
948 
949 		VNET_LIST_RLOCK();
950 		VNET_FOREACH(vnet_iter) {
951 			CURVNET_SET(vnet_iter);
952 			INP_INFO_WLOCK(&V_tcbinfo);
953 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
954 				INP_WLOCK(inp);
955 				if (inp->inp_flags & INP_TIMEWAIT) {
956 					INP_WUNLOCK(inp);
957 					continue;
958 				}
959 				tp = intotcpcb(inp);
960 				if (tp == NULL || tp->t_fb != blk) {
961 					INP_WUNLOCK(inp);
962 					continue;
963 				}
964 				tcp_switch_back_to_default(tp);
965 				INP_WUNLOCK(inp);
966 			}
967 			INP_INFO_WUNLOCK(&V_tcbinfo);
968 			CURVNET_RESTORE();
969 		}
970 		VNET_LIST_RUNLOCK();
971 
972 		rw_wlock(&tcp_function_lock);
973 	}
974 	if (blk->tfb_refcnt) {
975 		/* TCBs still attached. */
976 		rw_wunlock(&tcp_function_lock);
977 		return (EBUSY);
978 	}
979 	if (quiesce) {
980 		/* Skip removal. */
981 		rw_wunlock(&tcp_function_lock);
982 		return (0);
983 	}
984 	/* Remove any function names that map to this function block. */
985 	while (find_tcp_fb_locked(blk, &f) != NULL) {
986 		TAILQ_REMOVE(&t_functions, f, tf_next);
987 		tcp_fb_cnt--;
988 		f->tf_fb = NULL;
989 		free(f, M_TCPFUNCTIONS);
990 	}
991 	rw_wunlock(&tcp_function_lock);
992 	return (0);
993 }
994 
995 void
996 tcp_init(void)
997 {
998 	const char *tcbhash_tuneable;
999 	int hashsize;
1000 
1001 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1002 
1003 #ifdef TCP_HHOOK
1004 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1005 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1006 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1007 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1008 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1009 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1010 #endif
1011 	hashsize = TCBHASHSIZE;
1012 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1013 	if (hashsize == 0) {
1014 		/*
1015 		 * Auto tune the hash size based on maxsockets.
1016 		 * A perfect hash would have a 1:1 mapping
1017 		 * (hashsize = maxsockets) however it's been
1018 		 * suggested that O(2) average is better.
1019 		 */
1020 		hashsize = maketcp_hashsize(maxsockets / 4);
1021 		/*
1022 		 * Our historical default is 512,
1023 		 * do not autotune lower than this.
1024 		 */
1025 		if (hashsize < 512)
1026 			hashsize = 512;
1027 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1028 			printf("%s: %s auto tuned to %d\n", __func__,
1029 			    tcbhash_tuneable, hashsize);
1030 	}
1031 	/*
1032 	 * We require a hashsize to be a power of two.
1033 	 * Previously if it was not a power of two we would just reset it
1034 	 * back to 512, which could be a nasty surprise if you did not notice
1035 	 * the error message.
1036 	 * Instead what we do is clip it to the closest power of two lower
1037 	 * than the specified hash value.
1038 	 */
1039 	if (!powerof2(hashsize)) {
1040 		int oldhashsize = hashsize;
1041 
1042 		hashsize = maketcp_hashsize(hashsize);
1043 		/* prevent absurdly low value */
1044 		if (hashsize < 16)
1045 			hashsize = 16;
1046 		printf("%s: WARNING: TCB hash size not a power of 2, "
1047 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1048 		    hashsize);
1049 	}
1050 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1051 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1052 
1053 	/*
1054 	 * These have to be type stable for the benefit of the timers.
1055 	 */
1056 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1057 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1058 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1059 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1060 
1061 	tcp_tw_init();
1062 	syncache_init();
1063 	tcp_hc_init();
1064 
1065 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1066 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1067 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1068 
1069 	tcp_fastopen_init();
1070 
1071 	/* Skip initialization of globals for non-default instances. */
1072 	if (!IS_DEFAULT_VNET(curvnet))
1073 		return;
1074 
1075 	tcp_reass_global_init();
1076 
1077 	/* XXX virtualize those bellow? */
1078 	tcp_delacktime = TCPTV_DELACK;
1079 	tcp_keepinit = TCPTV_KEEP_INIT;
1080 	tcp_keepidle = TCPTV_KEEP_IDLE;
1081 	tcp_keepintvl = TCPTV_KEEPINTVL;
1082 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1083 	tcp_msl = TCPTV_MSL;
1084 	tcp_rexmit_min = TCPTV_MIN;
1085 	if (tcp_rexmit_min < 1)
1086 		tcp_rexmit_min = 1;
1087 	tcp_persmin = TCPTV_PERSMIN;
1088 	tcp_persmax = TCPTV_PERSMAX;
1089 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1090 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1091 	tcp_tcbhashsize = hashsize;
1092 	/* Setup the tcp function block list */
1093 	init_tcp_functions();
1094 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1095 #ifdef TCP_BLACKBOX
1096 	/* Initialize the TCP logging data. */
1097 	tcp_log_init();
1098 #endif
1099 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1100 
1101 	if (tcp_soreceive_stream) {
1102 #ifdef INET
1103 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1104 #endif
1105 #ifdef INET6
1106 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1107 #endif /* INET6 */
1108 	}
1109 
1110 #ifdef INET6
1111 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1112 #else /* INET6 */
1113 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1114 #endif /* INET6 */
1115 	if (max_protohdr < TCP_MINPROTOHDR)
1116 		max_protohdr = TCP_MINPROTOHDR;
1117 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1118 		panic("tcp_init");
1119 #undef TCP_MINPROTOHDR
1120 
1121 	ISN_LOCK_INIT();
1122 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1123 		SHUTDOWN_PRI_DEFAULT);
1124 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1125 		EVENTHANDLER_PRI_ANY);
1126 #ifdef TCPPCAP
1127 	tcp_pcap_init();
1128 #endif
1129 }
1130 
1131 #ifdef VIMAGE
1132 static void
1133 tcp_destroy(void *unused __unused)
1134 {
1135 	int n;
1136 #ifdef TCP_HHOOK
1137 	int error;
1138 #endif
1139 
1140 	/*
1141 	 * All our processes are gone, all our sockets should be cleaned
1142 	 * up, which means, we should be past the tcp_discardcb() calls.
1143 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1144 	 */
1145 	for (;;) {
1146 		INP_LIST_RLOCK(&V_tcbinfo);
1147 		n = V_tcbinfo.ipi_count;
1148 		INP_LIST_RUNLOCK(&V_tcbinfo);
1149 		if (n == 0)
1150 			break;
1151 		pause("tcpdes", hz / 10);
1152 	}
1153 	tcp_hc_destroy();
1154 	syncache_destroy();
1155 	tcp_tw_destroy();
1156 	in_pcbinfo_destroy(&V_tcbinfo);
1157 	/* tcp_discardcb() clears the sack_holes up. */
1158 	uma_zdestroy(V_sack_hole_zone);
1159 	uma_zdestroy(V_tcpcb_zone);
1160 
1161 	/*
1162 	 * Cannot free the zone until all tcpcbs are released as we attach
1163 	 * the allocations to them.
1164 	 */
1165 	tcp_fastopen_destroy();
1166 
1167 #ifdef TCP_HHOOK
1168 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1169 	if (error != 0) {
1170 		printf("%s: WARNING: unable to deregister helper hook "
1171 		    "type=%d, id=%d: error %d returned\n", __func__,
1172 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1173 	}
1174 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1175 	if (error != 0) {
1176 		printf("%s: WARNING: unable to deregister helper hook "
1177 		    "type=%d, id=%d: error %d returned\n", __func__,
1178 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1179 	}
1180 #endif
1181 }
1182 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1183 #endif
1184 
1185 void
1186 tcp_fini(void *xtp)
1187 {
1188 
1189 }
1190 
1191 /*
1192  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1193  * tcp_template used to store this data in mbufs, but we now recopy it out
1194  * of the tcpcb each time to conserve mbufs.
1195  */
1196 void
1197 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1198 {
1199 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1200 
1201 	INP_WLOCK_ASSERT(inp);
1202 
1203 #ifdef INET6
1204 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1205 		struct ip6_hdr *ip6;
1206 
1207 		ip6 = (struct ip6_hdr *)ip_ptr;
1208 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1209 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1210 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1211 			(IPV6_VERSION & IPV6_VERSION_MASK);
1212 		ip6->ip6_nxt = IPPROTO_TCP;
1213 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1214 		ip6->ip6_src = inp->in6p_laddr;
1215 		ip6->ip6_dst = inp->in6p_faddr;
1216 	}
1217 #endif /* INET6 */
1218 #if defined(INET6) && defined(INET)
1219 	else
1220 #endif
1221 #ifdef INET
1222 	{
1223 		struct ip *ip;
1224 
1225 		ip = (struct ip *)ip_ptr;
1226 		ip->ip_v = IPVERSION;
1227 		ip->ip_hl = 5;
1228 		ip->ip_tos = inp->inp_ip_tos;
1229 		ip->ip_len = 0;
1230 		ip->ip_id = 0;
1231 		ip->ip_off = 0;
1232 		ip->ip_ttl = inp->inp_ip_ttl;
1233 		ip->ip_sum = 0;
1234 		ip->ip_p = IPPROTO_TCP;
1235 		ip->ip_src = inp->inp_laddr;
1236 		ip->ip_dst = inp->inp_faddr;
1237 	}
1238 #endif /* INET */
1239 	th->th_sport = inp->inp_lport;
1240 	th->th_dport = inp->inp_fport;
1241 	th->th_seq = 0;
1242 	th->th_ack = 0;
1243 	th->th_x2 = 0;
1244 	th->th_off = 5;
1245 	th->th_flags = 0;
1246 	th->th_win = 0;
1247 	th->th_urp = 0;
1248 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1249 }
1250 
1251 /*
1252  * Create template to be used to send tcp packets on a connection.
1253  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1254  * use for this function is in keepalives, which use tcp_respond.
1255  */
1256 struct tcptemp *
1257 tcpip_maketemplate(struct inpcb *inp)
1258 {
1259 	struct tcptemp *t;
1260 
1261 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1262 	if (t == NULL)
1263 		return (NULL);
1264 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1265 	return (t);
1266 }
1267 
1268 /*
1269  * Send a single message to the TCP at address specified by
1270  * the given TCP/IP header.  If m == NULL, then we make a copy
1271  * of the tcpiphdr at th and send directly to the addressed host.
1272  * This is used to force keep alive messages out using the TCP
1273  * template for a connection.  If flags are given then we send
1274  * a message back to the TCP which originated the segment th,
1275  * and discard the mbuf containing it and any other attached mbufs.
1276  *
1277  * In any case the ack and sequence number of the transmitted
1278  * segment are as specified by the parameters.
1279  *
1280  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1281  */
1282 void
1283 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1284     tcp_seq ack, tcp_seq seq, int flags)
1285 {
1286 	struct tcpopt to;
1287 	struct inpcb *inp;
1288 	struct ip *ip;
1289 	struct mbuf *optm;
1290 	struct tcphdr *nth;
1291 	u_char *optp;
1292 #ifdef INET6
1293 	struct ip6_hdr *ip6;
1294 	int isipv6;
1295 #endif /* INET6 */
1296 	int optlen, tlen, win;
1297 	bool incl_opts;
1298 
1299 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1300 
1301 #ifdef INET6
1302 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1303 	ip6 = ipgen;
1304 #endif /* INET6 */
1305 	ip = ipgen;
1306 
1307 	if (tp != NULL) {
1308 		inp = tp->t_inpcb;
1309 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1310 		INP_WLOCK_ASSERT(inp);
1311 	} else
1312 		inp = NULL;
1313 
1314 	incl_opts = false;
1315 	win = 0;
1316 	if (tp != NULL) {
1317 		if (!(flags & TH_RST)) {
1318 			win = sbspace(&inp->inp_socket->so_rcv);
1319 			if (win > TCP_MAXWIN << tp->rcv_scale)
1320 				win = TCP_MAXWIN << tp->rcv_scale;
1321 		}
1322 		if ((tp->t_flags & TF_NOOPT) == 0)
1323 			incl_opts = true;
1324 	}
1325 	if (m == NULL) {
1326 		m = m_gethdr(M_NOWAIT, MT_DATA);
1327 		if (m == NULL)
1328 			return;
1329 		m->m_data += max_linkhdr;
1330 #ifdef INET6
1331 		if (isipv6) {
1332 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1333 			      sizeof(struct ip6_hdr));
1334 			ip6 = mtod(m, struct ip6_hdr *);
1335 			nth = (struct tcphdr *)(ip6 + 1);
1336 		} else
1337 #endif /* INET6 */
1338 		{
1339 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1340 			ip = mtod(m, struct ip *);
1341 			nth = (struct tcphdr *)(ip + 1);
1342 		}
1343 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1344 		flags = TH_ACK;
1345 	} else if (!M_WRITABLE(m)) {
1346 		struct mbuf *n;
1347 
1348 		/* Can't reuse 'm', allocate a new mbuf. */
1349 		n = m_gethdr(M_NOWAIT, MT_DATA);
1350 		if (n == NULL) {
1351 			m_freem(m);
1352 			return;
1353 		}
1354 
1355 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1356 			m_freem(m);
1357 			m_freem(n);
1358 			return;
1359 		}
1360 
1361 		n->m_data += max_linkhdr;
1362 		/* m_len is set later */
1363 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1364 #ifdef INET6
1365 		if (isipv6) {
1366 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1367 			      sizeof(struct ip6_hdr));
1368 			ip6 = mtod(n, struct ip6_hdr *);
1369 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1370 			nth = (struct tcphdr *)(ip6 + 1);
1371 		} else
1372 #endif /* INET6 */
1373 		{
1374 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1375 			ip = mtod(n, struct ip *);
1376 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1377 			nth = (struct tcphdr *)(ip + 1);
1378 		}
1379 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1380 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1381 		th = nth;
1382 		m_freem(m);
1383 		m = n;
1384 	} else {
1385 		/*
1386 		 *  reuse the mbuf.
1387 		 * XXX MRT We inherit the FIB, which is lucky.
1388 		 */
1389 		m_freem(m->m_next);
1390 		m->m_next = NULL;
1391 		m->m_data = (caddr_t)ipgen;
1392 		/* m_len is set later */
1393 #ifdef INET6
1394 		if (isipv6) {
1395 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1396 			nth = (struct tcphdr *)(ip6 + 1);
1397 		} else
1398 #endif /* INET6 */
1399 		{
1400 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1401 			nth = (struct tcphdr *)(ip + 1);
1402 		}
1403 		if (th != nth) {
1404 			/*
1405 			 * this is usually a case when an extension header
1406 			 * exists between the IPv6 header and the
1407 			 * TCP header.
1408 			 */
1409 			nth->th_sport = th->th_sport;
1410 			nth->th_dport = th->th_dport;
1411 		}
1412 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1413 #undef xchg
1414 	}
1415 	tlen = 0;
1416 #ifdef INET6
1417 	if (isipv6)
1418 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1419 #endif
1420 #if defined(INET) && defined(INET6)
1421 	else
1422 #endif
1423 #ifdef INET
1424 		tlen = sizeof (struct tcpiphdr);
1425 #endif
1426 #ifdef INVARIANTS
1427 	m->m_len = 0;
1428 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1429 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1430 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1431 #endif
1432 	m->m_len = tlen;
1433 	to.to_flags = 0;
1434 	if (incl_opts) {
1435 		/* Make sure we have room. */
1436 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1437 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1438 			if (m->m_next) {
1439 				optp = mtod(m->m_next, u_char *);
1440 				optm = m->m_next;
1441 			} else
1442 				incl_opts = false;
1443 		} else {
1444 			optp = (u_char *) (nth + 1);
1445 			optm = m;
1446 		}
1447 	}
1448 	if (incl_opts) {
1449 		/* Timestamps. */
1450 		if (tp->t_flags & TF_RCVD_TSTMP) {
1451 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1452 			to.to_tsecr = tp->ts_recent;
1453 			to.to_flags |= TOF_TS;
1454 		}
1455 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1456 		/* TCP-MD5 (RFC2385). */
1457 		if (tp->t_flags & TF_SIGNATURE)
1458 			to.to_flags |= TOF_SIGNATURE;
1459 #endif
1460 		/* Add the options. */
1461 		tlen += optlen = tcp_addoptions(&to, optp);
1462 
1463 		/* Update m_len in the correct mbuf. */
1464 		optm->m_len += optlen;
1465 	} else
1466 		optlen = 0;
1467 #ifdef INET6
1468 	if (isipv6) {
1469 		ip6->ip6_flow = 0;
1470 		ip6->ip6_vfc = IPV6_VERSION;
1471 		ip6->ip6_nxt = IPPROTO_TCP;
1472 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1473 	}
1474 #endif
1475 #if defined(INET) && defined(INET6)
1476 	else
1477 #endif
1478 #ifdef INET
1479 	{
1480 		ip->ip_len = htons(tlen);
1481 		ip->ip_ttl = V_ip_defttl;
1482 		if (V_path_mtu_discovery)
1483 			ip->ip_off |= htons(IP_DF);
1484 	}
1485 #endif
1486 	m->m_pkthdr.len = tlen;
1487 	m->m_pkthdr.rcvif = NULL;
1488 #ifdef MAC
1489 	if (inp != NULL) {
1490 		/*
1491 		 * Packet is associated with a socket, so allow the
1492 		 * label of the response to reflect the socket label.
1493 		 */
1494 		INP_WLOCK_ASSERT(inp);
1495 		mac_inpcb_create_mbuf(inp, m);
1496 	} else {
1497 		/*
1498 		 * Packet is not associated with a socket, so possibly
1499 		 * update the label in place.
1500 		 */
1501 		mac_netinet_tcp_reply(m);
1502 	}
1503 #endif
1504 	nth->th_seq = htonl(seq);
1505 	nth->th_ack = htonl(ack);
1506 	nth->th_x2 = 0;
1507 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1508 	nth->th_flags = flags;
1509 	if (tp != NULL)
1510 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1511 	else
1512 		nth->th_win = htons((u_short)win);
1513 	nth->th_urp = 0;
1514 
1515 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1516 	if (to.to_flags & TOF_SIGNATURE) {
1517 		if (!TCPMD5_ENABLED() ||
1518 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1519 			m_freem(m);
1520 			return;
1521 		}
1522 	}
1523 #endif
1524 
1525 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1526 #ifdef INET6
1527 	if (isipv6) {
1528 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1529 		nth->th_sum = in6_cksum_pseudo(ip6,
1530 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1531 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1532 		    NULL, NULL);
1533 	}
1534 #endif /* INET6 */
1535 #if defined(INET6) && defined(INET)
1536 	else
1537 #endif
1538 #ifdef INET
1539 	{
1540 		m->m_pkthdr.csum_flags = CSUM_TCP;
1541 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1542 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1543 	}
1544 #endif /* INET */
1545 #ifdef TCPDEBUG
1546 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1547 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1548 #endif
1549 	TCP_PROBE3(debug__output, tp, th, m);
1550 	if (flags & TH_RST)
1551 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1552 
1553 #ifdef INET6
1554 	if (isipv6) {
1555 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1556 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1557 	}
1558 #endif /* INET6 */
1559 #if defined(INET) && defined(INET6)
1560 	else
1561 #endif
1562 #ifdef INET
1563 	{
1564 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1565 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1566 	}
1567 #endif
1568 }
1569 
1570 /*
1571  * Create a new TCP control block, making an
1572  * empty reassembly queue and hooking it to the argument
1573  * protocol control block.  The `inp' parameter must have
1574  * come from the zone allocator set up in tcp_init().
1575  */
1576 struct tcpcb *
1577 tcp_newtcpcb(struct inpcb *inp)
1578 {
1579 	struct tcpcb_mem *tm;
1580 	struct tcpcb *tp;
1581 #ifdef INET6
1582 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1583 #endif /* INET6 */
1584 
1585 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1586 	if (tm == NULL)
1587 		return (NULL);
1588 	tp = &tm->tcb;
1589 
1590 	/* Initialise cc_var struct for this tcpcb. */
1591 	tp->ccv = &tm->ccv;
1592 	tp->ccv->type = IPPROTO_TCP;
1593 	tp->ccv->ccvc.tcp = tp;
1594 	rw_rlock(&tcp_function_lock);
1595 	tp->t_fb = tcp_func_set_ptr;
1596 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1597 	rw_runlock(&tcp_function_lock);
1598 	/*
1599 	 * Use the current system default CC algorithm.
1600 	 */
1601 	CC_LIST_RLOCK();
1602 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1603 	CC_ALGO(tp) = CC_DEFAULT();
1604 	CC_LIST_RUNLOCK();
1605 
1606 	if (CC_ALGO(tp)->cb_init != NULL)
1607 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1608 			if (tp->t_fb->tfb_tcp_fb_fini)
1609 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1610 			refcount_release(&tp->t_fb->tfb_refcnt);
1611 			uma_zfree(V_tcpcb_zone, tm);
1612 			return (NULL);
1613 		}
1614 
1615 #ifdef TCP_HHOOK
1616 	tp->osd = &tm->osd;
1617 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1618 		if (tp->t_fb->tfb_tcp_fb_fini)
1619 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1620 		refcount_release(&tp->t_fb->tfb_refcnt);
1621 		uma_zfree(V_tcpcb_zone, tm);
1622 		return (NULL);
1623 	}
1624 #endif
1625 
1626 #ifdef VIMAGE
1627 	tp->t_vnet = inp->inp_vnet;
1628 #endif
1629 	tp->t_timers = &tm->tt;
1630 	TAILQ_INIT(&tp->t_segq);
1631 	tp->t_maxseg =
1632 #ifdef INET6
1633 		isipv6 ? V_tcp_v6mssdflt :
1634 #endif /* INET6 */
1635 		V_tcp_mssdflt;
1636 
1637 	/* Set up our timeouts. */
1638 	callout_init(&tp->t_timers->tt_rexmt, 1);
1639 	callout_init(&tp->t_timers->tt_persist, 1);
1640 	callout_init(&tp->t_timers->tt_keep, 1);
1641 	callout_init(&tp->t_timers->tt_2msl, 1);
1642 	callout_init(&tp->t_timers->tt_delack, 1);
1643 
1644 	if (V_tcp_do_rfc1323)
1645 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1646 	if (V_tcp_do_sack)
1647 		tp->t_flags |= TF_SACK_PERMIT;
1648 	TAILQ_INIT(&tp->snd_holes);
1649 	/*
1650 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1651 	 * is called.
1652 	 */
1653 	in_pcbref(inp);	/* Reference for tcpcb */
1654 	tp->t_inpcb = inp;
1655 
1656 	/*
1657 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1658 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1659 	 * reasonable initial retransmit time.
1660 	 */
1661 	tp->t_srtt = TCPTV_SRTTBASE;
1662 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1663 	tp->t_rttmin = tcp_rexmit_min;
1664 	tp->t_rxtcur = TCPTV_RTOBASE;
1665 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1666 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1667 	tp->t_rcvtime = ticks;
1668 	/*
1669 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1670 	 * because the socket may be bound to an IPv6 wildcard address,
1671 	 * which may match an IPv4-mapped IPv6 address.
1672 	 */
1673 	inp->inp_ip_ttl = V_ip_defttl;
1674 	inp->inp_ppcb = tp;
1675 #ifdef TCPPCAP
1676 	/*
1677 	 * Init the TCP PCAP queues.
1678 	 */
1679 	tcp_pcap_tcpcb_init(tp);
1680 #endif
1681 #ifdef TCP_BLACKBOX
1682 	/* Initialize the per-TCPCB log data. */
1683 	tcp_log_tcpcbinit(tp);
1684 #endif
1685 	if (tp->t_fb->tfb_tcp_fb_init) {
1686 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1687 	}
1688 	return (tp);		/* XXX */
1689 }
1690 
1691 /*
1692  * Switch the congestion control algorithm back to NewReno for any active
1693  * control blocks using an algorithm which is about to go away.
1694  * This ensures the CC framework can allow the unload to proceed without leaving
1695  * any dangling pointers which would trigger a panic.
1696  * Returning non-zero would inform the CC framework that something went wrong
1697  * and it would be unsafe to allow the unload to proceed. However, there is no
1698  * way for this to occur with this implementation so we always return zero.
1699  */
1700 int
1701 tcp_ccalgounload(struct cc_algo *unload_algo)
1702 {
1703 	struct cc_algo *tmpalgo;
1704 	struct inpcb *inp;
1705 	struct tcpcb *tp;
1706 	VNET_ITERATOR_DECL(vnet_iter);
1707 
1708 	/*
1709 	 * Check all active control blocks across all network stacks and change
1710 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1711 	 * requires cleanup code to be run, call it.
1712 	 */
1713 	VNET_LIST_RLOCK();
1714 	VNET_FOREACH(vnet_iter) {
1715 		CURVNET_SET(vnet_iter);
1716 		INP_INFO_WLOCK(&V_tcbinfo);
1717 		/*
1718 		 * New connections already part way through being initialised
1719 		 * with the CC algo we're removing will not race with this code
1720 		 * because the INP_INFO_WLOCK is held during initialisation. We
1721 		 * therefore don't enter the loop below until the connection
1722 		 * list has stabilised.
1723 		 */
1724 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1725 			INP_WLOCK(inp);
1726 			/* Important to skip tcptw structs. */
1727 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1728 			    (tp = intotcpcb(inp)) != NULL) {
1729 				/*
1730 				 * By holding INP_WLOCK here, we are assured
1731 				 * that the connection is not currently
1732 				 * executing inside the CC module's functions
1733 				 * i.e. it is safe to make the switch back to
1734 				 * NewReno.
1735 				 */
1736 				if (CC_ALGO(tp) == unload_algo) {
1737 					tmpalgo = CC_ALGO(tp);
1738 					if (tmpalgo->cb_destroy != NULL)
1739 						tmpalgo->cb_destroy(tp->ccv);
1740 					CC_DATA(tp) = NULL;
1741 					/*
1742 					 * NewReno may allocate memory on
1743 					 * demand for certain stateful
1744 					 * configuration as needed, but is
1745 					 * coded to never fail on memory
1746 					 * allocation failure so it is a safe
1747 					 * fallback.
1748 					 */
1749 					CC_ALGO(tp) = &newreno_cc_algo;
1750 				}
1751 			}
1752 			INP_WUNLOCK(inp);
1753 		}
1754 		INP_INFO_WUNLOCK(&V_tcbinfo);
1755 		CURVNET_RESTORE();
1756 	}
1757 	VNET_LIST_RUNLOCK();
1758 
1759 	return (0);
1760 }
1761 
1762 /*
1763  * Drop a TCP connection, reporting
1764  * the specified error.  If connection is synchronized,
1765  * then send a RST to peer.
1766  */
1767 struct tcpcb *
1768 tcp_drop(struct tcpcb *tp, int errno)
1769 {
1770 	struct socket *so = tp->t_inpcb->inp_socket;
1771 
1772 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1773 	INP_WLOCK_ASSERT(tp->t_inpcb);
1774 
1775 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1776 		tcp_state_change(tp, TCPS_CLOSED);
1777 		(void) tp->t_fb->tfb_tcp_output(tp);
1778 		TCPSTAT_INC(tcps_drops);
1779 	} else
1780 		TCPSTAT_INC(tcps_conndrops);
1781 	if (errno == ETIMEDOUT && tp->t_softerror)
1782 		errno = tp->t_softerror;
1783 	so->so_error = errno;
1784 	return (tcp_close(tp));
1785 }
1786 
1787 void
1788 tcp_discardcb(struct tcpcb *tp)
1789 {
1790 	struct inpcb *inp = tp->t_inpcb;
1791 	struct socket *so = inp->inp_socket;
1792 #ifdef INET6
1793 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1794 #endif /* INET6 */
1795 	int released __unused;
1796 
1797 	INP_WLOCK_ASSERT(inp);
1798 
1799 	/*
1800 	 * Make sure that all of our timers are stopped before we delete the
1801 	 * PCB.
1802 	 *
1803 	 * If stopping a timer fails, we schedule a discard function in same
1804 	 * callout, and the last discard function called will take care of
1805 	 * deleting the tcpcb.
1806 	 */
1807 	tp->t_timers->tt_draincnt = 0;
1808 	tcp_timer_stop(tp, TT_REXMT);
1809 	tcp_timer_stop(tp, TT_PERSIST);
1810 	tcp_timer_stop(tp, TT_KEEP);
1811 	tcp_timer_stop(tp, TT_2MSL);
1812 	tcp_timer_stop(tp, TT_DELACK);
1813 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1814 		/*
1815 		 * Call the stop-all function of the methods,
1816 		 * this function should call the tcp_timer_stop()
1817 		 * method with each of the function specific timeouts.
1818 		 * That stop will be called via the tfb_tcp_timer_stop()
1819 		 * which should use the async drain function of the
1820 		 * callout system (see tcp_var.h).
1821 		 */
1822 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1823 	}
1824 
1825 	/*
1826 	 * If we got enough samples through the srtt filter,
1827 	 * save the rtt and rttvar in the routing entry.
1828 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1829 	 * 4 samples is enough for the srtt filter to converge
1830 	 * to within enough % of the correct value; fewer samples
1831 	 * and we could save a bogus rtt. The danger is not high
1832 	 * as tcp quickly recovers from everything.
1833 	 * XXX: Works very well but needs some more statistics!
1834 	 */
1835 	if (tp->t_rttupdated >= 4) {
1836 		struct hc_metrics_lite metrics;
1837 		uint32_t ssthresh;
1838 
1839 		bzero(&metrics, sizeof(metrics));
1840 		/*
1841 		 * Update the ssthresh always when the conditions below
1842 		 * are satisfied. This gives us better new start value
1843 		 * for the congestion avoidance for new connections.
1844 		 * ssthresh is only set if packet loss occurred on a session.
1845 		 *
1846 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1847 		 * being torn down.  Ideally this code would not use 'so'.
1848 		 */
1849 		ssthresh = tp->snd_ssthresh;
1850 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1851 			/*
1852 			 * convert the limit from user data bytes to
1853 			 * packets then to packet data bytes.
1854 			 */
1855 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1856 			if (ssthresh < 2)
1857 				ssthresh = 2;
1858 			ssthresh *= (tp->t_maxseg +
1859 #ifdef INET6
1860 			    (isipv6 ? sizeof (struct ip6_hdr) +
1861 				sizeof (struct tcphdr) :
1862 #endif
1863 				sizeof (struct tcpiphdr)
1864 #ifdef INET6
1865 			    )
1866 #endif
1867 			    );
1868 		} else
1869 			ssthresh = 0;
1870 		metrics.rmx_ssthresh = ssthresh;
1871 
1872 		metrics.rmx_rtt = tp->t_srtt;
1873 		metrics.rmx_rttvar = tp->t_rttvar;
1874 		metrics.rmx_cwnd = tp->snd_cwnd;
1875 		metrics.rmx_sendpipe = 0;
1876 		metrics.rmx_recvpipe = 0;
1877 
1878 		tcp_hc_update(&inp->inp_inc, &metrics);
1879 	}
1880 
1881 	/* free the reassembly queue, if any */
1882 	tcp_reass_flush(tp);
1883 
1884 #ifdef TCP_OFFLOAD
1885 	/* Disconnect offload device, if any. */
1886 	if (tp->t_flags & TF_TOE)
1887 		tcp_offload_detach(tp);
1888 #endif
1889 
1890 	tcp_free_sackholes(tp);
1891 
1892 #ifdef TCPPCAP
1893 	/* Free the TCP PCAP queues. */
1894 	tcp_pcap_drain(&(tp->t_inpkts));
1895 	tcp_pcap_drain(&(tp->t_outpkts));
1896 #endif
1897 
1898 	/* Allow the CC algorithm to clean up after itself. */
1899 	if (CC_ALGO(tp)->cb_destroy != NULL)
1900 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1901 	CC_DATA(tp) = NULL;
1902 
1903 #ifdef TCP_HHOOK
1904 	khelp_destroy_osd(tp->osd);
1905 #endif
1906 
1907 	CC_ALGO(tp) = NULL;
1908 	inp->inp_ppcb = NULL;
1909 	if (tp->t_timers->tt_draincnt == 0) {
1910 		/* We own the last reference on tcpcb, let's free it. */
1911 #ifdef TCP_BLACKBOX
1912 		tcp_log_tcpcbfini(tp);
1913 #endif
1914 		TCPSTATES_DEC(tp->t_state);
1915 		if (tp->t_fb->tfb_tcp_fb_fini)
1916 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1917 		refcount_release(&tp->t_fb->tfb_refcnt);
1918 		tp->t_inpcb = NULL;
1919 		uma_zfree(V_tcpcb_zone, tp);
1920 		released = in_pcbrele_wlocked(inp);
1921 		KASSERT(!released, ("%s: inp %p should not have been released "
1922 			"here", __func__, inp));
1923 	}
1924 }
1925 
1926 void
1927 tcp_timer_discard(void *ptp)
1928 {
1929 	struct inpcb *inp;
1930 	struct tcpcb *tp;
1931 	struct epoch_tracker et;
1932 
1933 	tp = (struct tcpcb *)ptp;
1934 	CURVNET_SET(tp->t_vnet);
1935 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1936 	inp = tp->t_inpcb;
1937 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1938 		__func__, tp));
1939 	INP_WLOCK(inp);
1940 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1941 		("%s: tcpcb has to be stopped here", __func__));
1942 	tp->t_timers->tt_draincnt--;
1943 	if (tp->t_timers->tt_draincnt == 0) {
1944 		/* We own the last reference on this tcpcb, let's free it. */
1945 #ifdef TCP_BLACKBOX
1946 		tcp_log_tcpcbfini(tp);
1947 #endif
1948 		TCPSTATES_DEC(tp->t_state);
1949 		if (tp->t_fb->tfb_tcp_fb_fini)
1950 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1951 		refcount_release(&tp->t_fb->tfb_refcnt);
1952 		tp->t_inpcb = NULL;
1953 		uma_zfree(V_tcpcb_zone, tp);
1954 		if (in_pcbrele_wlocked(inp)) {
1955 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1956 			CURVNET_RESTORE();
1957 			return;
1958 		}
1959 	}
1960 	INP_WUNLOCK(inp);
1961 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1962 	CURVNET_RESTORE();
1963 }
1964 
1965 /*
1966  * Attempt to close a TCP control block, marking it as dropped, and freeing
1967  * the socket if we hold the only reference.
1968  */
1969 struct tcpcb *
1970 tcp_close(struct tcpcb *tp)
1971 {
1972 	struct inpcb *inp = tp->t_inpcb;
1973 	struct socket *so;
1974 
1975 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1976 	INP_WLOCK_ASSERT(inp);
1977 
1978 #ifdef TCP_OFFLOAD
1979 	if (tp->t_state == TCPS_LISTEN)
1980 		tcp_offload_listen_stop(tp);
1981 #endif
1982 	/*
1983 	 * This releases the TFO pending counter resource for TFO listen
1984 	 * sockets as well as passively-created TFO sockets that transition
1985 	 * from SYN_RECEIVED to CLOSED.
1986 	 */
1987 	if (tp->t_tfo_pending) {
1988 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1989 		tp->t_tfo_pending = NULL;
1990 	}
1991 	in_pcbdrop(inp);
1992 	TCPSTAT_INC(tcps_closed);
1993 	if (tp->t_state != TCPS_CLOSED)
1994 		tcp_state_change(tp, TCPS_CLOSED);
1995 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1996 	so = inp->inp_socket;
1997 	soisdisconnected(so);
1998 	if (inp->inp_flags & INP_SOCKREF) {
1999 		KASSERT(so->so_state & SS_PROTOREF,
2000 		    ("tcp_close: !SS_PROTOREF"));
2001 		inp->inp_flags &= ~INP_SOCKREF;
2002 		INP_WUNLOCK(inp);
2003 		SOCK_LOCK(so);
2004 		so->so_state &= ~SS_PROTOREF;
2005 		sofree(so);
2006 		return (NULL);
2007 	}
2008 	return (tp);
2009 }
2010 
2011 void
2012 tcp_drain(void)
2013 {
2014 	VNET_ITERATOR_DECL(vnet_iter);
2015 
2016 	if (!do_tcpdrain)
2017 		return;
2018 
2019 	VNET_LIST_RLOCK_NOSLEEP();
2020 	VNET_FOREACH(vnet_iter) {
2021 		CURVNET_SET(vnet_iter);
2022 		struct inpcb *inpb;
2023 		struct tcpcb *tcpb;
2024 
2025 	/*
2026 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2027 	 * if there is one...
2028 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2029 	 *      reassembly queue should be flushed, but in a situation
2030 	 *	where we're really low on mbufs, this is potentially
2031 	 *	useful.
2032 	 */
2033 		INP_INFO_WLOCK(&V_tcbinfo);
2034 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2035 			INP_WLOCK(inpb);
2036 			if (inpb->inp_flags & INP_TIMEWAIT) {
2037 				INP_WUNLOCK(inpb);
2038 				continue;
2039 			}
2040 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2041 				tcp_reass_flush(tcpb);
2042 				tcp_clean_sackreport(tcpb);
2043 #ifdef TCP_BLACKBOX
2044 				tcp_log_drain(tcpb);
2045 #endif
2046 #ifdef TCPPCAP
2047 				if (tcp_pcap_aggressive_free) {
2048 					/* Free the TCP PCAP queues. */
2049 					tcp_pcap_drain(&(tcpb->t_inpkts));
2050 					tcp_pcap_drain(&(tcpb->t_outpkts));
2051 				}
2052 #endif
2053 			}
2054 			INP_WUNLOCK(inpb);
2055 		}
2056 		INP_INFO_WUNLOCK(&V_tcbinfo);
2057 		CURVNET_RESTORE();
2058 	}
2059 	VNET_LIST_RUNLOCK_NOSLEEP();
2060 }
2061 
2062 /*
2063  * Notify a tcp user of an asynchronous error;
2064  * store error as soft error, but wake up user
2065  * (for now, won't do anything until can select for soft error).
2066  *
2067  * Do not wake up user since there currently is no mechanism for
2068  * reporting soft errors (yet - a kqueue filter may be added).
2069  */
2070 static struct inpcb *
2071 tcp_notify(struct inpcb *inp, int error)
2072 {
2073 	struct tcpcb *tp;
2074 
2075 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2076 	INP_WLOCK_ASSERT(inp);
2077 
2078 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2079 	    (inp->inp_flags & INP_DROPPED))
2080 		return (inp);
2081 
2082 	tp = intotcpcb(inp);
2083 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2084 
2085 	/*
2086 	 * Ignore some errors if we are hooked up.
2087 	 * If connection hasn't completed, has retransmitted several times,
2088 	 * and receives a second error, give up now.  This is better
2089 	 * than waiting a long time to establish a connection that
2090 	 * can never complete.
2091 	 */
2092 	if (tp->t_state == TCPS_ESTABLISHED &&
2093 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2094 	     error == EHOSTDOWN)) {
2095 		if (inp->inp_route.ro_rt) {
2096 			RTFREE(inp->inp_route.ro_rt);
2097 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2098 		}
2099 		return (inp);
2100 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2101 	    tp->t_softerror) {
2102 		tp = tcp_drop(tp, error);
2103 		if (tp != NULL)
2104 			return (inp);
2105 		else
2106 			return (NULL);
2107 	} else {
2108 		tp->t_softerror = error;
2109 		return (inp);
2110 	}
2111 #if 0
2112 	wakeup( &so->so_timeo);
2113 	sorwakeup(so);
2114 	sowwakeup(so);
2115 #endif
2116 }
2117 
2118 static int
2119 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2120 {
2121 	int error, i, m, n, pcb_count;
2122 	struct inpcb *inp, **inp_list;
2123 	inp_gen_t gencnt;
2124 	struct xinpgen xig;
2125 	struct epoch_tracker et;
2126 
2127 	/*
2128 	 * The process of preparing the TCB list is too time-consuming and
2129 	 * resource-intensive to repeat twice on every request.
2130 	 */
2131 	if (req->oldptr == NULL) {
2132 		n = V_tcbinfo.ipi_count +
2133 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2134 		n += imax(n / 8, 10);
2135 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2136 		return (0);
2137 	}
2138 
2139 	if (req->newptr != NULL)
2140 		return (EPERM);
2141 
2142 	/*
2143 	 * OK, now we're committed to doing something.
2144 	 */
2145 	INP_LIST_RLOCK(&V_tcbinfo);
2146 	gencnt = V_tcbinfo.ipi_gencnt;
2147 	n = V_tcbinfo.ipi_count;
2148 	INP_LIST_RUNLOCK(&V_tcbinfo);
2149 
2150 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2151 
2152 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2153 		+ (n + m) * sizeof(struct xtcpcb));
2154 	if (error != 0)
2155 		return (error);
2156 
2157 	xig.xig_len = sizeof xig;
2158 	xig.xig_count = n + m;
2159 	xig.xig_gen = gencnt;
2160 	xig.xig_sogen = so_gencnt;
2161 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2162 	if (error)
2163 		return (error);
2164 
2165 	error = syncache_pcblist(req, m, &pcb_count);
2166 	if (error)
2167 		return (error);
2168 
2169 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2170 
2171 	INP_INFO_WLOCK(&V_tcbinfo);
2172 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2173 	    inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2174 		INP_WLOCK(inp);
2175 		if (inp->inp_gencnt <= gencnt) {
2176 			/*
2177 			 * XXX: This use of cr_cansee(), introduced with
2178 			 * TCP state changes, is not quite right, but for
2179 			 * now, better than nothing.
2180 			 */
2181 			if (inp->inp_flags & INP_TIMEWAIT) {
2182 				if (intotw(inp) != NULL)
2183 					error = cr_cansee(req->td->td_ucred,
2184 					    intotw(inp)->tw_cred);
2185 				else
2186 					error = EINVAL;	/* Skip this inp. */
2187 			} else
2188 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2189 			if (error == 0) {
2190 				in_pcbref(inp);
2191 				inp_list[i++] = inp;
2192 			}
2193 		}
2194 		INP_WUNLOCK(inp);
2195 	}
2196 	INP_INFO_WUNLOCK(&V_tcbinfo);
2197 	n = i;
2198 
2199 	error = 0;
2200 	for (i = 0; i < n; i++) {
2201 		inp = inp_list[i];
2202 		INP_RLOCK(inp);
2203 		if (inp->inp_gencnt <= gencnt) {
2204 			struct xtcpcb xt;
2205 
2206 			tcp_inptoxtp(inp, &xt);
2207 			INP_RUNLOCK(inp);
2208 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2209 		} else
2210 			INP_RUNLOCK(inp);
2211 	}
2212 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2213 	for (i = 0; i < n; i++) {
2214 		inp = inp_list[i];
2215 		INP_RLOCK(inp);
2216 		if (!in_pcbrele_rlocked(inp))
2217 			INP_RUNLOCK(inp);
2218 	}
2219 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2220 
2221 	if (!error) {
2222 		/*
2223 		 * Give the user an updated idea of our state.
2224 		 * If the generation differs from what we told
2225 		 * her before, she knows that something happened
2226 		 * while we were processing this request, and it
2227 		 * might be necessary to retry.
2228 		 */
2229 		INP_LIST_RLOCK(&V_tcbinfo);
2230 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2231 		xig.xig_sogen = so_gencnt;
2232 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2233 		INP_LIST_RUNLOCK(&V_tcbinfo);
2234 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2235 	}
2236 	free(inp_list, M_TEMP);
2237 	return (error);
2238 }
2239 
2240 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2241     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2242     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2243 
2244 #ifdef INET
2245 static int
2246 tcp_getcred(SYSCTL_HANDLER_ARGS)
2247 {
2248 	struct xucred xuc;
2249 	struct sockaddr_in addrs[2];
2250 	struct inpcb *inp;
2251 	int error;
2252 
2253 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2254 	if (error)
2255 		return (error);
2256 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2257 	if (error)
2258 		return (error);
2259 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2260 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2261 	if (inp != NULL) {
2262 		if (inp->inp_socket == NULL)
2263 			error = ENOENT;
2264 		if (error == 0)
2265 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2266 		if (error == 0)
2267 			cru2x(inp->inp_cred, &xuc);
2268 		INP_RUNLOCK(inp);
2269 	} else
2270 		error = ENOENT;
2271 	if (error == 0)
2272 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2273 	return (error);
2274 }
2275 
2276 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2277     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2278     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2279 #endif /* INET */
2280 
2281 #ifdef INET6
2282 static int
2283 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2284 {
2285 	struct xucred xuc;
2286 	struct sockaddr_in6 addrs[2];
2287 	struct inpcb *inp;
2288 	int error;
2289 #ifdef INET
2290 	int mapped = 0;
2291 #endif
2292 
2293 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2294 	if (error)
2295 		return (error);
2296 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2297 	if (error)
2298 		return (error);
2299 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2300 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2301 		return (error);
2302 	}
2303 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2304 #ifdef INET
2305 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2306 			mapped = 1;
2307 		else
2308 #endif
2309 			return (EINVAL);
2310 	}
2311 
2312 #ifdef INET
2313 	if (mapped == 1)
2314 		inp = in_pcblookup(&V_tcbinfo,
2315 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2316 			addrs[1].sin6_port,
2317 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2318 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2319 	else
2320 #endif
2321 		inp = in6_pcblookup(&V_tcbinfo,
2322 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2323 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2324 			INPLOOKUP_RLOCKPCB, NULL);
2325 	if (inp != NULL) {
2326 		if (inp->inp_socket == NULL)
2327 			error = ENOENT;
2328 		if (error == 0)
2329 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2330 		if (error == 0)
2331 			cru2x(inp->inp_cred, &xuc);
2332 		INP_RUNLOCK(inp);
2333 	} else
2334 		error = ENOENT;
2335 	if (error == 0)
2336 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2337 	return (error);
2338 }
2339 
2340 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2341     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2342     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2343 #endif /* INET6 */
2344 
2345 
2346 #ifdef INET
2347 void
2348 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2349 {
2350 	struct ip *ip = vip;
2351 	struct tcphdr *th;
2352 	struct in_addr faddr;
2353 	struct inpcb *inp;
2354 	struct tcpcb *tp;
2355 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2356 	struct icmp *icp;
2357 	struct in_conninfo inc;
2358 	struct epoch_tracker et;
2359 	tcp_seq icmp_tcp_seq;
2360 	int mtu;
2361 
2362 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2363 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2364 		return;
2365 
2366 	if (cmd == PRC_MSGSIZE)
2367 		notify = tcp_mtudisc_notify;
2368 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2369 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2370 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2371 		notify = tcp_drop_syn_sent;
2372 
2373 	/*
2374 	 * Hostdead is ugly because it goes linearly through all PCBs.
2375 	 * XXX: We never get this from ICMP, otherwise it makes an
2376 	 * excellent DoS attack on machines with many connections.
2377 	 */
2378 	else if (cmd == PRC_HOSTDEAD)
2379 		ip = NULL;
2380 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2381 		return;
2382 
2383 	if (ip == NULL) {
2384 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2385 		return;
2386 	}
2387 
2388 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2389 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2390 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2391 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2392 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2393 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2394 		/* signal EHOSTDOWN, as it flushes the cached route */
2395 		inp = (*notify)(inp, EHOSTDOWN);
2396 		goto out;
2397 	}
2398 	icmp_tcp_seq = th->th_seq;
2399 	if (inp != NULL)  {
2400 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2401 		    !(inp->inp_flags & INP_DROPPED) &&
2402 		    !(inp->inp_socket == NULL)) {
2403 			tp = intotcpcb(inp);
2404 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2405 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2406 				if (cmd == PRC_MSGSIZE) {
2407 					/*
2408 					 * MTU discovery:
2409 					 * If we got a needfrag set the MTU
2410 					 * in the route to the suggested new
2411 					 * value (if given) and then notify.
2412 					 */
2413 					mtu = ntohs(icp->icmp_nextmtu);
2414 					/*
2415 					 * If no alternative MTU was
2416 					 * proposed, try the next smaller
2417 					 * one.
2418 					 */
2419 					if (!mtu)
2420 						mtu = ip_next_mtu(
2421 						    ntohs(ip->ip_len), 1);
2422 					if (mtu < V_tcp_minmss +
2423 					    sizeof(struct tcpiphdr))
2424 						mtu = V_tcp_minmss +
2425 						    sizeof(struct tcpiphdr);
2426 					/*
2427 					 * Only process the offered MTU if it
2428 					 * is smaller than the current one.
2429 					 */
2430 					if (mtu < tp->t_maxseg +
2431 					    sizeof(struct tcpiphdr)) {
2432 						bzero(&inc, sizeof(inc));
2433 						inc.inc_faddr = faddr;
2434 						inc.inc_fibnum =
2435 						    inp->inp_inc.inc_fibnum;
2436 						tcp_hc_updatemtu(&inc, mtu);
2437 						tcp_mtudisc(inp, mtu);
2438 					}
2439 				} else
2440 					inp = (*notify)(inp,
2441 					    inetctlerrmap[cmd]);
2442 			}
2443 		}
2444 	} else {
2445 		bzero(&inc, sizeof(inc));
2446 		inc.inc_fport = th->th_dport;
2447 		inc.inc_lport = th->th_sport;
2448 		inc.inc_faddr = faddr;
2449 		inc.inc_laddr = ip->ip_src;
2450 		syncache_unreach(&inc, icmp_tcp_seq);
2451 	}
2452 out:
2453 	if (inp != NULL)
2454 		INP_WUNLOCK(inp);
2455 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2456 }
2457 #endif /* INET */
2458 
2459 #ifdef INET6
2460 void
2461 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2462 {
2463 	struct in6_addr *dst;
2464 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2465 	struct ip6_hdr *ip6;
2466 	struct mbuf *m;
2467 	struct inpcb *inp;
2468 	struct tcpcb *tp;
2469 	struct icmp6_hdr *icmp6;
2470 	struct ip6ctlparam *ip6cp = NULL;
2471 	const struct sockaddr_in6 *sa6_src = NULL;
2472 	struct in_conninfo inc;
2473 	struct epoch_tracker et;
2474 	struct tcp_ports {
2475 		uint16_t th_sport;
2476 		uint16_t th_dport;
2477 	} t_ports;
2478 	tcp_seq icmp_tcp_seq;
2479 	unsigned int mtu;
2480 	unsigned int off;
2481 
2482 	if (sa->sa_family != AF_INET6 ||
2483 	    sa->sa_len != sizeof(struct sockaddr_in6))
2484 		return;
2485 
2486 	/* if the parameter is from icmp6, decode it. */
2487 	if (d != NULL) {
2488 		ip6cp = (struct ip6ctlparam *)d;
2489 		icmp6 = ip6cp->ip6c_icmp6;
2490 		m = ip6cp->ip6c_m;
2491 		ip6 = ip6cp->ip6c_ip6;
2492 		off = ip6cp->ip6c_off;
2493 		sa6_src = ip6cp->ip6c_src;
2494 		dst = ip6cp->ip6c_finaldst;
2495 	} else {
2496 		m = NULL;
2497 		ip6 = NULL;
2498 		off = 0;	/* fool gcc */
2499 		sa6_src = &sa6_any;
2500 		dst = NULL;
2501 	}
2502 
2503 	if (cmd == PRC_MSGSIZE)
2504 		notify = tcp_mtudisc_notify;
2505 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2506 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2507 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2508 		notify = tcp_drop_syn_sent;
2509 
2510 	/*
2511 	 * Hostdead is ugly because it goes linearly through all PCBs.
2512 	 * XXX: We never get this from ICMP, otherwise it makes an
2513 	 * excellent DoS attack on machines with many connections.
2514 	 */
2515 	else if (cmd == PRC_HOSTDEAD)
2516 		ip6 = NULL;
2517 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2518 		return;
2519 
2520 	if (ip6 == NULL) {
2521 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2522 			      (const struct sockaddr *)sa6_src,
2523 			      0, cmd, NULL, notify);
2524 		return;
2525 	}
2526 
2527 	/* Check if we can safely get the ports from the tcp hdr */
2528 	if (m == NULL ||
2529 	    (m->m_pkthdr.len <
2530 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2531 		return;
2532 	}
2533 	bzero(&t_ports, sizeof(struct tcp_ports));
2534 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2535 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2536 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2537 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2538 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2539 		/* signal EHOSTDOWN, as it flushes the cached route */
2540 		inp = (*notify)(inp, EHOSTDOWN);
2541 		goto out;
2542 	}
2543 	off += sizeof(struct tcp_ports);
2544 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2545 		goto out;
2546 	}
2547 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2548 	if (inp != NULL)  {
2549 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2550 		    !(inp->inp_flags & INP_DROPPED) &&
2551 		    !(inp->inp_socket == NULL)) {
2552 			tp = intotcpcb(inp);
2553 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2554 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2555 				if (cmd == PRC_MSGSIZE) {
2556 					/*
2557 					 * MTU discovery:
2558 					 * If we got a needfrag set the MTU
2559 					 * in the route to the suggested new
2560 					 * value (if given) and then notify.
2561 					 */
2562 					mtu = ntohl(icmp6->icmp6_mtu);
2563 					/*
2564 					 * If no alternative MTU was
2565 					 * proposed, or the proposed
2566 					 * MTU was too small, set to
2567 					 * the min.
2568 					 */
2569 					if (mtu < IPV6_MMTU)
2570 						mtu = IPV6_MMTU - 8;
2571 					bzero(&inc, sizeof(inc));
2572 					inc.inc_fibnum = M_GETFIB(m);
2573 					inc.inc_flags |= INC_ISIPV6;
2574 					inc.inc6_faddr = *dst;
2575 					if (in6_setscope(&inc.inc6_faddr,
2576 						m->m_pkthdr.rcvif, NULL))
2577 						goto out;
2578 					/*
2579 					 * Only process the offered MTU if it
2580 					 * is smaller than the current one.
2581 					 */
2582 					if (mtu < tp->t_maxseg +
2583 					    sizeof (struct tcphdr) +
2584 					    sizeof (struct ip6_hdr)) {
2585 						tcp_hc_updatemtu(&inc, mtu);
2586 						tcp_mtudisc(inp, mtu);
2587 						ICMP6STAT_INC(icp6s_pmtuchg);
2588 					}
2589 				} else
2590 					inp = (*notify)(inp,
2591 					    inet6ctlerrmap[cmd]);
2592 			}
2593 		}
2594 	} else {
2595 		bzero(&inc, sizeof(inc));
2596 		inc.inc_fibnum = M_GETFIB(m);
2597 		inc.inc_flags |= INC_ISIPV6;
2598 		inc.inc_fport = t_ports.th_dport;
2599 		inc.inc_lport = t_ports.th_sport;
2600 		inc.inc6_faddr = *dst;
2601 		inc.inc6_laddr = ip6->ip6_src;
2602 		syncache_unreach(&inc, icmp_tcp_seq);
2603 	}
2604 out:
2605 	if (inp != NULL)
2606 		INP_WUNLOCK(inp);
2607 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2608 }
2609 #endif /* INET6 */
2610 
2611 static uint32_t
2612 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2613 {
2614 	MD5_CTX ctx;
2615 	uint32_t hash[4];
2616 
2617 	MD5Init(&ctx);
2618 	MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2619 	MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2620 	switch (inc->inc_flags & INC_ISIPV6) {
2621 #ifdef INET
2622 	case 0:
2623 		MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2624 		MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2625 		break;
2626 #endif
2627 #ifdef INET6
2628 	case INC_ISIPV6:
2629 		MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2630 		MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2631 		break;
2632 #endif
2633 	}
2634 	MD5Update(&ctx, key, len);
2635 	MD5Final((unsigned char *)hash, &ctx);
2636 
2637 	return (hash[0]);
2638 }
2639 
2640 uint32_t
2641 tcp_new_ts_offset(struct in_conninfo *inc)
2642 {
2643 	return (tcp_keyed_hash(inc, V_ts_offset_secret,
2644 	    sizeof(V_ts_offset_secret)));
2645 }
2646 
2647 /*
2648  * Following is where TCP initial sequence number generation occurs.
2649  *
2650  * There are two places where we must use initial sequence numbers:
2651  * 1.  In SYN-ACK packets.
2652  * 2.  In SYN packets.
2653  *
2654  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2655  * tcp_syncache.c for details.
2656  *
2657  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2658  * depends on this property.  In addition, these ISNs should be
2659  * unguessable so as to prevent connection hijacking.  To satisfy
2660  * the requirements of this situation, the algorithm outlined in
2661  * RFC 1948 is used, with only small modifications.
2662  *
2663  * Implementation details:
2664  *
2665  * Time is based off the system timer, and is corrected so that it
2666  * increases by one megabyte per second.  This allows for proper
2667  * recycling on high speed LANs while still leaving over an hour
2668  * before rollover.
2669  *
2670  * As reading the *exact* system time is too expensive to be done
2671  * whenever setting up a TCP connection, we increment the time
2672  * offset in two ways.  First, a small random positive increment
2673  * is added to isn_offset for each connection that is set up.
2674  * Second, the function tcp_isn_tick fires once per clock tick
2675  * and increments isn_offset as necessary so that sequence numbers
2676  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2677  * random positive increments serve only to ensure that the same
2678  * exact sequence number is never sent out twice (as could otherwise
2679  * happen when a port is recycled in less than the system tick
2680  * interval.)
2681  *
2682  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2683  * between seeding of isn_secret.  This is normally set to zero,
2684  * as reseeding should not be necessary.
2685  *
2686  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2687  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2688  * general, this means holding an exclusive (write) lock.
2689  */
2690 
2691 #define ISN_BYTES_PER_SECOND 1048576
2692 #define ISN_STATIC_INCREMENT 4096
2693 #define ISN_RANDOM_INCREMENT (4096 - 1)
2694 #define ISN_SECRET_LENGTH    32
2695 
2696 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2697 VNET_DEFINE_STATIC(int, isn_last);
2698 VNET_DEFINE_STATIC(int, isn_last_reseed);
2699 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2700 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2701 
2702 #define	V_isn_secret			VNET(isn_secret)
2703 #define	V_isn_last			VNET(isn_last)
2704 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2705 #define	V_isn_offset			VNET(isn_offset)
2706 #define	V_isn_offset_old		VNET(isn_offset_old)
2707 
2708 tcp_seq
2709 tcp_new_isn(struct in_conninfo *inc)
2710 {
2711 	tcp_seq new_isn;
2712 	u_int32_t projected_offset;
2713 
2714 	ISN_LOCK();
2715 	/* Seed if this is the first use, reseed if requested. */
2716 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2717 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2718 		< (u_int)ticks))) {
2719 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2720 		V_isn_last_reseed = ticks;
2721 	}
2722 
2723 	/* Compute the md5 hash and return the ISN. */
2724 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2725 	    sizeof(V_isn_secret));
2726 	V_isn_offset += ISN_STATIC_INCREMENT +
2727 		(arc4random() & ISN_RANDOM_INCREMENT);
2728 	if (ticks != V_isn_last) {
2729 		projected_offset = V_isn_offset_old +
2730 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2731 		if (SEQ_GT(projected_offset, V_isn_offset))
2732 			V_isn_offset = projected_offset;
2733 		V_isn_offset_old = V_isn_offset;
2734 		V_isn_last = ticks;
2735 	}
2736 	new_isn += V_isn_offset;
2737 	ISN_UNLOCK();
2738 	return (new_isn);
2739 }
2740 
2741 /*
2742  * When a specific ICMP unreachable message is received and the
2743  * connection state is SYN-SENT, drop the connection.  This behavior
2744  * is controlled by the icmp_may_rst sysctl.
2745  */
2746 struct inpcb *
2747 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2748 {
2749 	struct tcpcb *tp;
2750 
2751 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2752 	INP_WLOCK_ASSERT(inp);
2753 
2754 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2755 	    (inp->inp_flags & INP_DROPPED))
2756 		return (inp);
2757 
2758 	tp = intotcpcb(inp);
2759 	if (tp->t_state != TCPS_SYN_SENT)
2760 		return (inp);
2761 
2762 	if (IS_FASTOPEN(tp->t_flags))
2763 		tcp_fastopen_disable_path(tp);
2764 
2765 	tp = tcp_drop(tp, errno);
2766 	if (tp != NULL)
2767 		return (inp);
2768 	else
2769 		return (NULL);
2770 }
2771 
2772 /*
2773  * When `need fragmentation' ICMP is received, update our idea of the MSS
2774  * based on the new value. Also nudge TCP to send something, since we
2775  * know the packet we just sent was dropped.
2776  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2777  */
2778 static struct inpcb *
2779 tcp_mtudisc_notify(struct inpcb *inp, int error)
2780 {
2781 
2782 	tcp_mtudisc(inp, -1);
2783 	return (inp);
2784 }
2785 
2786 static void
2787 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2788 {
2789 	struct tcpcb *tp;
2790 	struct socket *so;
2791 
2792 	INP_WLOCK_ASSERT(inp);
2793 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2794 	    (inp->inp_flags & INP_DROPPED))
2795 		return;
2796 
2797 	tp = intotcpcb(inp);
2798 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2799 
2800 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2801 
2802 	so = inp->inp_socket;
2803 	SOCKBUF_LOCK(&so->so_snd);
2804 	/* If the mss is larger than the socket buffer, decrease the mss. */
2805 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2806 		tp->t_maxseg = so->so_snd.sb_hiwat;
2807 	SOCKBUF_UNLOCK(&so->so_snd);
2808 
2809 	TCPSTAT_INC(tcps_mturesent);
2810 	tp->t_rtttime = 0;
2811 	tp->snd_nxt = tp->snd_una;
2812 	tcp_free_sackholes(tp);
2813 	tp->snd_recover = tp->snd_max;
2814 	if (tp->t_flags & TF_SACK_PERMIT)
2815 		EXIT_FASTRECOVERY(tp->t_flags);
2816 	tp->t_fb->tfb_tcp_output(tp);
2817 }
2818 
2819 #ifdef INET
2820 /*
2821  * Look-up the routing entry to the peer of this inpcb.  If no route
2822  * is found and it cannot be allocated, then return 0.  This routine
2823  * is called by TCP routines that access the rmx structure and by
2824  * tcp_mss_update to get the peer/interface MTU.
2825  */
2826 uint32_t
2827 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2828 {
2829 	struct nhop4_extended nh4;
2830 	struct ifnet *ifp;
2831 	uint32_t maxmtu = 0;
2832 
2833 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2834 
2835 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2836 
2837 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2838 		    NHR_REF, 0, &nh4) != 0)
2839 			return (0);
2840 
2841 		ifp = nh4.nh_ifp;
2842 		maxmtu = nh4.nh_mtu;
2843 
2844 		/* Report additional interface capabilities. */
2845 		if (cap != NULL) {
2846 			if (ifp->if_capenable & IFCAP_TSO4 &&
2847 			    ifp->if_hwassist & CSUM_TSO) {
2848 				cap->ifcap |= CSUM_TSO;
2849 				cap->tsomax = ifp->if_hw_tsomax;
2850 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2851 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2852 			}
2853 		}
2854 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2855 	}
2856 	return (maxmtu);
2857 }
2858 #endif /* INET */
2859 
2860 #ifdef INET6
2861 uint32_t
2862 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2863 {
2864 	struct nhop6_extended nh6;
2865 	struct in6_addr dst6;
2866 	uint32_t scopeid;
2867 	struct ifnet *ifp;
2868 	uint32_t maxmtu = 0;
2869 
2870 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2871 
2872 	if (inc->inc_flags & INC_IPV6MINMTU)
2873 		return (IPV6_MMTU);
2874 
2875 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2876 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2877 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2878 		    0, &nh6) != 0)
2879 			return (0);
2880 
2881 		ifp = nh6.nh_ifp;
2882 		maxmtu = nh6.nh_mtu;
2883 
2884 		/* Report additional interface capabilities. */
2885 		if (cap != NULL) {
2886 			if (ifp->if_capenable & IFCAP_TSO6 &&
2887 			    ifp->if_hwassist & CSUM_TSO) {
2888 				cap->ifcap |= CSUM_TSO;
2889 				cap->tsomax = ifp->if_hw_tsomax;
2890 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2891 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2892 			}
2893 		}
2894 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2895 	}
2896 
2897 	return (maxmtu);
2898 }
2899 #endif /* INET6 */
2900 
2901 /*
2902  * Calculate effective SMSS per RFC5681 definition for a given TCP
2903  * connection at its current state, taking into account SACK and etc.
2904  */
2905 u_int
2906 tcp_maxseg(const struct tcpcb *tp)
2907 {
2908 	u_int optlen;
2909 
2910 	if (tp->t_flags & TF_NOOPT)
2911 		return (tp->t_maxseg);
2912 
2913 	/*
2914 	 * Here we have a simplified code from tcp_addoptions(),
2915 	 * without a proper loop, and having most of paddings hardcoded.
2916 	 * We might make mistakes with padding here in some edge cases,
2917 	 * but this is harmless, since result of tcp_maxseg() is used
2918 	 * only in cwnd and ssthresh estimations.
2919 	 */
2920 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2921 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2922 		if (tp->t_flags & TF_RCVD_TSTMP)
2923 			optlen = TCPOLEN_TSTAMP_APPA;
2924 		else
2925 			optlen = 0;
2926 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2927 		if (tp->t_flags & TF_SIGNATURE)
2928 			optlen += PAD(TCPOLEN_SIGNATURE);
2929 #endif
2930 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2931 			optlen += TCPOLEN_SACKHDR;
2932 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2933 			optlen = PAD(optlen);
2934 		}
2935 	} else {
2936 		if (tp->t_flags & TF_REQ_TSTMP)
2937 			optlen = TCPOLEN_TSTAMP_APPA;
2938 		else
2939 			optlen = PAD(TCPOLEN_MAXSEG);
2940 		if (tp->t_flags & TF_REQ_SCALE)
2941 			optlen += PAD(TCPOLEN_WINDOW);
2942 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2943 		if (tp->t_flags & TF_SIGNATURE)
2944 			optlen += PAD(TCPOLEN_SIGNATURE);
2945 #endif
2946 		if (tp->t_flags & TF_SACK_PERMIT)
2947 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2948 	}
2949 #undef PAD
2950 	optlen = min(optlen, TCP_MAXOLEN);
2951 	return (tp->t_maxseg - optlen);
2952 }
2953 
2954 static int
2955 sysctl_drop(SYSCTL_HANDLER_ARGS)
2956 {
2957 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2958 	struct sockaddr_storage addrs[2];
2959 	struct inpcb *inp;
2960 	struct tcpcb *tp;
2961 	struct tcptw *tw;
2962 	struct sockaddr_in *fin, *lin;
2963 	struct epoch_tracker et;
2964 #ifdef INET6
2965 	struct sockaddr_in6 *fin6, *lin6;
2966 #endif
2967 	int error;
2968 
2969 	inp = NULL;
2970 	fin = lin = NULL;
2971 #ifdef INET6
2972 	fin6 = lin6 = NULL;
2973 #endif
2974 	error = 0;
2975 
2976 	if (req->oldptr != NULL || req->oldlen != 0)
2977 		return (EINVAL);
2978 	if (req->newptr == NULL)
2979 		return (EPERM);
2980 	if (req->newlen < sizeof(addrs))
2981 		return (ENOMEM);
2982 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2983 	if (error)
2984 		return (error);
2985 
2986 	switch (addrs[0].ss_family) {
2987 #ifdef INET6
2988 	case AF_INET6:
2989 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2990 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2991 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2992 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2993 			return (EINVAL);
2994 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2995 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2996 				return (EINVAL);
2997 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2998 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2999 			fin = (struct sockaddr_in *)&addrs[0];
3000 			lin = (struct sockaddr_in *)&addrs[1];
3001 			break;
3002 		}
3003 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3004 		if (error)
3005 			return (error);
3006 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3007 		if (error)
3008 			return (error);
3009 		break;
3010 #endif
3011 #ifdef INET
3012 	case AF_INET:
3013 		fin = (struct sockaddr_in *)&addrs[0];
3014 		lin = (struct sockaddr_in *)&addrs[1];
3015 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3016 		    lin->sin_len != sizeof(struct sockaddr_in))
3017 			return (EINVAL);
3018 		break;
3019 #endif
3020 	default:
3021 		return (EINVAL);
3022 	}
3023 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3024 	switch (addrs[0].ss_family) {
3025 #ifdef INET6
3026 	case AF_INET6:
3027 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3028 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3029 		    INPLOOKUP_WLOCKPCB, NULL);
3030 		break;
3031 #endif
3032 #ifdef INET
3033 	case AF_INET:
3034 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3035 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3036 		break;
3037 #endif
3038 	}
3039 	if (inp != NULL) {
3040 		if (inp->inp_flags & INP_TIMEWAIT) {
3041 			/*
3042 			 * XXXRW: There currently exists a state where an
3043 			 * inpcb is present, but its timewait state has been
3044 			 * discarded.  For now, don't allow dropping of this
3045 			 * type of inpcb.
3046 			 */
3047 			tw = intotw(inp);
3048 			if (tw != NULL)
3049 				tcp_twclose(tw, 0);
3050 			else
3051 				INP_WUNLOCK(inp);
3052 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3053 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3054 			tp = intotcpcb(inp);
3055 			tp = tcp_drop(tp, ECONNABORTED);
3056 			if (tp != NULL)
3057 				INP_WUNLOCK(inp);
3058 		} else
3059 			INP_WUNLOCK(inp);
3060 	} else
3061 		error = ESRCH;
3062 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3063 	return (error);
3064 }
3065 
3066 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3067     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3068     0, sysctl_drop, "", "Drop TCP connection");
3069 
3070 /*
3071  * Generate a standardized TCP log line for use throughout the
3072  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3073  * allow use in the interrupt context.
3074  *
3075  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3076  * NB: The function may return NULL if memory allocation failed.
3077  *
3078  * Due to header inclusion and ordering limitations the struct ip
3079  * and ip6_hdr pointers have to be passed as void pointers.
3080  */
3081 char *
3082 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3083     const void *ip6hdr)
3084 {
3085 
3086 	/* Is logging enabled? */
3087 	if (tcp_log_in_vain == 0)
3088 		return (NULL);
3089 
3090 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3091 }
3092 
3093 char *
3094 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3095     const void *ip6hdr)
3096 {
3097 
3098 	/* Is logging enabled? */
3099 	if (tcp_log_debug == 0)
3100 		return (NULL);
3101 
3102 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3103 }
3104 
3105 static char *
3106 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3107     const void *ip6hdr)
3108 {
3109 	char *s, *sp;
3110 	size_t size;
3111 	struct ip *ip;
3112 #ifdef INET6
3113 	const struct ip6_hdr *ip6;
3114 
3115 	ip6 = (const struct ip6_hdr *)ip6hdr;
3116 #endif /* INET6 */
3117 	ip = (struct ip *)ip4hdr;
3118 
3119 	/*
3120 	 * The log line looks like this:
3121 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3122 	 */
3123 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3124 	    sizeof(PRINT_TH_FLAGS) + 1 +
3125 #ifdef INET6
3126 	    2 * INET6_ADDRSTRLEN;
3127 #else
3128 	    2 * INET_ADDRSTRLEN;
3129 #endif /* INET6 */
3130 
3131 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3132 	if (s == NULL)
3133 		return (NULL);
3134 
3135 	strcat(s, "TCP: [");
3136 	sp = s + strlen(s);
3137 
3138 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3139 		inet_ntoa_r(inc->inc_faddr, sp);
3140 		sp = s + strlen(s);
3141 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3142 		sp = s + strlen(s);
3143 		inet_ntoa_r(inc->inc_laddr, sp);
3144 		sp = s + strlen(s);
3145 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3146 #ifdef INET6
3147 	} else if (inc) {
3148 		ip6_sprintf(sp, &inc->inc6_faddr);
3149 		sp = s + strlen(s);
3150 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3151 		sp = s + strlen(s);
3152 		ip6_sprintf(sp, &inc->inc6_laddr);
3153 		sp = s + strlen(s);
3154 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3155 	} else if (ip6 && th) {
3156 		ip6_sprintf(sp, &ip6->ip6_src);
3157 		sp = s + strlen(s);
3158 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3159 		sp = s + strlen(s);
3160 		ip6_sprintf(sp, &ip6->ip6_dst);
3161 		sp = s + strlen(s);
3162 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3163 #endif /* INET6 */
3164 #ifdef INET
3165 	} else if (ip && th) {
3166 		inet_ntoa_r(ip->ip_src, sp);
3167 		sp = s + strlen(s);
3168 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3169 		sp = s + strlen(s);
3170 		inet_ntoa_r(ip->ip_dst, sp);
3171 		sp = s + strlen(s);
3172 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3173 #endif /* INET */
3174 	} else {
3175 		free(s, M_TCPLOG);
3176 		return (NULL);
3177 	}
3178 	sp = s + strlen(s);
3179 	if (th)
3180 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3181 	if (*(s + size - 1) != '\0')
3182 		panic("%s: string too long", __func__);
3183 	return (s);
3184 }
3185 
3186 /*
3187  * A subroutine which makes it easy to track TCP state changes with DTrace.
3188  * This function shouldn't be called for t_state initializations that don't
3189  * correspond to actual TCP state transitions.
3190  */
3191 void
3192 tcp_state_change(struct tcpcb *tp, int newstate)
3193 {
3194 #if defined(KDTRACE_HOOKS)
3195 	int pstate = tp->t_state;
3196 #endif
3197 
3198 	TCPSTATES_DEC(tp->t_state);
3199 	TCPSTATES_INC(newstate);
3200 	tp->t_state = newstate;
3201 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3202 }
3203 
3204 /*
3205  * Create an external-format (``xtcpcb'') structure using the information in
3206  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3207  * reduce the spew of irrelevant information over this interface, to isolate
3208  * user code from changes in the kernel structure, and potentially to provide
3209  * information-hiding if we decide that some of this information should be
3210  * hidden from users.
3211  */
3212 void
3213 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3214 {
3215 	struct tcpcb *tp = intotcpcb(inp);
3216 	sbintime_t now;
3217 
3218 	if (inp->inp_flags & INP_TIMEWAIT) {
3219 		bzero(xt, sizeof(struct xtcpcb));
3220 		xt->t_state = TCPS_TIME_WAIT;
3221 	} else {
3222 		xt->t_state = tp->t_state;
3223 		xt->t_logstate = tp->t_logstate;
3224 		xt->t_flags = tp->t_flags;
3225 		xt->t_sndzerowin = tp->t_sndzerowin;
3226 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3227 		xt->t_rcvoopack = tp->t_rcvoopack;
3228 
3229 		now = getsbinuptime();
3230 #define	COPYTIMER(ttt)	do {						\
3231 		if (callout_active(&tp->t_timers->ttt))			\
3232 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3233 			    SBT_1MS;					\
3234 		else							\
3235 			xt->ttt = 0;					\
3236 } while (0)
3237 		COPYTIMER(tt_delack);
3238 		COPYTIMER(tt_rexmt);
3239 		COPYTIMER(tt_persist);
3240 		COPYTIMER(tt_keep);
3241 		COPYTIMER(tt_2msl);
3242 #undef COPYTIMER
3243 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3244 
3245 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3246 		    TCP_FUNCTION_NAME_LEN_MAX);
3247 		bzero(xt->xt_logid, TCP_LOG_ID_LEN);
3248 #ifdef TCP_BLACKBOX
3249 		(void)tcp_log_get_id(tp, xt->xt_logid);
3250 #endif
3251 	}
3252 
3253 	xt->xt_len = sizeof(struct xtcpcb);
3254 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3255 	if (inp->inp_socket == NULL)
3256 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3257 }
3258