xref: /freebsd/sys/netinet/tcp_subr.c (revision ddc0daea20280c3a06a910b72b14ffe3f624df71)
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_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
47 #ifdef TCP_HHOOK
48 #include <sys/hhook.h>
49 #endif
50 #include <sys/kernel.h>
51 #ifdef TCP_HHOOK
52 #include <sys/khelp.h>
53 #endif
54 #ifdef KERN_TLS
55 #include <sys/ktls.h>
56 #endif
57 #include <sys/sysctl.h>
58 #include <sys/jail.h>
59 #include <sys/malloc.h>
60 #include <sys/refcount.h>
61 #include <sys/mbuf.h>
62 #ifdef INET6
63 #include <sys/domain.h>
64 #endif
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/sdt.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/protosw.h>
71 #include <sys/random.h>
72 
73 #include <vm/uma.h>
74 
75 #include <net/route.h>
76 #include <net/if.h>
77 #include <net/if_var.h>
78 #include <net/vnet.h>
79 
80 #include <netinet/in.h>
81 #include <netinet/in_fib.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip.h>
87 #include <netinet/ip_icmp.h>
88 #include <netinet/ip_var.h>
89 #ifdef INET6
90 #include <netinet/icmp6.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/in6_fib.h>
93 #include <netinet6/in6_pcb.h>
94 #include <netinet6/ip6_var.h>
95 #include <netinet6/scope6_var.h>
96 #include <netinet6/nd6.h>
97 #endif
98 
99 #include <netinet/tcp.h>
100 #include <netinet/tcp_fsm.h>
101 #include <netinet/tcp_seq.h>
102 #include <netinet/tcp_timer.h>
103 #include <netinet/tcp_var.h>
104 #include <netinet/tcp_log_buf.h>
105 #include <netinet/tcp_syncache.h>
106 #include <netinet/tcp_hpts.h>
107 #include <netinet/cc/cc.h>
108 #ifdef INET6
109 #include <netinet6/tcp6_var.h>
110 #endif
111 #include <netinet/tcpip.h>
112 #include <netinet/tcp_fastopen.h>
113 #ifdef TCPPCAP
114 #include <netinet/tcp_pcap.h>
115 #endif
116 #ifdef TCPDEBUG
117 #include <netinet/tcp_debug.h>
118 #endif
119 #ifdef INET6
120 #include <netinet6/ip6protosw.h>
121 #endif
122 #ifdef TCP_OFFLOAD
123 #include <netinet/tcp_offload.h>
124 #endif
125 
126 #include <netipsec/ipsec_support.h>
127 
128 #include <machine/in_cksum.h>
129 #include <sys/md5.h>
130 
131 #include <security/mac/mac_framework.h>
132 
133 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
134 #ifdef INET6
135 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
136 #endif
137 
138 struct rwlock tcp_function_lock;
139 
140 static int
141 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
142 {
143 	int error, new;
144 
145 	new = V_tcp_mssdflt;
146 	error = sysctl_handle_int(oidp, &new, 0, req);
147 	if (error == 0 && req->newptr) {
148 		if (new < TCP_MINMSS)
149 			error = EINVAL;
150 		else
151 			V_tcp_mssdflt = new;
152 	}
153 	return (error);
154 }
155 
156 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
157     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
158     &sysctl_net_inet_tcp_mss_check, "I",
159     "Default TCP Maximum Segment Size");
160 
161 #ifdef INET6
162 static int
163 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
164 {
165 	int error, new;
166 
167 	new = V_tcp_v6mssdflt;
168 	error = sysctl_handle_int(oidp, &new, 0, req);
169 	if (error == 0 && req->newptr) {
170 		if (new < TCP_MINMSS)
171 			error = EINVAL;
172 		else
173 			V_tcp_v6mssdflt = new;
174 	}
175 	return (error);
176 }
177 
178 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
179     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
180     &sysctl_net_inet_tcp_mss_v6_check, "I",
181    "Default TCP Maximum Segment Size for IPv6");
182 #endif /* INET6 */
183 
184 /*
185  * Minimum MSS we accept and use. This prevents DoS attacks where
186  * we are forced to a ridiculous low MSS like 20 and send hundreds
187  * of packets instead of one. The effect scales with the available
188  * bandwidth and quickly saturates the CPU and network interface
189  * with packet generation and sending. Set to zero to disable MINMSS
190  * checking. This setting prevents us from sending too small packets.
191  */
192 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
194      &VNET_NAME(tcp_minmss), 0,
195     "Minimum TCP Maximum Segment Size");
196 
197 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
198 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
199     &VNET_NAME(tcp_do_rfc1323), 0,
200     "Enable rfc1323 (high performance TCP) extensions");
201 
202 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
204     &VNET_NAME(tcp_ts_offset_per_conn), 0,
205     "Initialize TCP timestamps per connection instead of per host pair");
206 
207 static int	tcp_log_debug = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
209     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
210 
211 static int	tcp_tcbhashsize;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
213     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
214 
215 static int	do_tcpdrain = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
217     "Enable tcp_drain routine for extra help when low on mbufs");
218 
219 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
220     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
221 
222 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
223 #define	V_icmp_may_rst			VNET(icmp_may_rst)
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
225     &VNET_NAME(icmp_may_rst), 0,
226     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
227 
228 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
229 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
231     &VNET_NAME(tcp_isn_reseed_interval), 0,
232     "Seconds between reseeding of ISN secret");
233 
234 static int	tcp_soreceive_stream;
235 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
236     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
237 
238 VNET_DEFINE(uma_zone_t, sack_hole_zone);
239 #define	V_sack_hole_zone		VNET(sack_hole_zone)
240 
241 #ifdef TCP_HHOOK
242 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
243 #endif
244 
245 #define TS_OFFSET_SECRET_LENGTH 32
246 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
247 #define	V_ts_offset_secret	VNET(ts_offset_secret)
248 
249 static int	tcp_default_fb_init(struct tcpcb *tp);
250 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
251 static int	tcp_default_handoff_ok(struct tcpcb *tp);
252 static struct inpcb *tcp_notify(struct inpcb *, int);
253 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
254 static void tcp_mtudisc(struct inpcb *, int);
255 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
256 		    void *ip4hdr, const void *ip6hdr);
257 
258 
259 static struct tcp_function_block tcp_def_funcblk = {
260 	.tfb_tcp_block_name = "freebsd",
261 	.tfb_tcp_output = tcp_output,
262 	.tfb_tcp_do_segment = tcp_do_segment,
263 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
264 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
265 	.tfb_tcp_fb_init = tcp_default_fb_init,
266 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
267 };
268 
269 static int tcp_fb_cnt = 0;
270 struct tcp_funchead t_functions;
271 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
272 
273 static struct tcp_function_block *
274 find_tcp_functions_locked(struct tcp_function_set *fs)
275 {
276 	struct tcp_function *f;
277 	struct tcp_function_block *blk=NULL;
278 
279 	TAILQ_FOREACH(f, &t_functions, tf_next) {
280 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
281 			blk = f->tf_fb;
282 			break;
283 		}
284 	}
285 	return(blk);
286 }
287 
288 static struct tcp_function_block *
289 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
290 {
291 	struct tcp_function_block *rblk=NULL;
292 	struct tcp_function *f;
293 
294 	TAILQ_FOREACH(f, &t_functions, tf_next) {
295 		if (f->tf_fb == blk) {
296 			rblk = blk;
297 			if (s) {
298 				*s = f;
299 			}
300 			break;
301 		}
302 	}
303 	return (rblk);
304 }
305 
306 struct tcp_function_block *
307 find_and_ref_tcp_functions(struct tcp_function_set *fs)
308 {
309 	struct tcp_function_block *blk;
310 
311 	rw_rlock(&tcp_function_lock);
312 	blk = find_tcp_functions_locked(fs);
313 	if (blk)
314 		refcount_acquire(&blk->tfb_refcnt);
315 	rw_runlock(&tcp_function_lock);
316 	return(blk);
317 }
318 
319 struct tcp_function_block *
320 find_and_ref_tcp_fb(struct tcp_function_block *blk)
321 {
322 	struct tcp_function_block *rblk;
323 
324 	rw_rlock(&tcp_function_lock);
325 	rblk = find_tcp_fb_locked(blk, NULL);
326 	if (rblk)
327 		refcount_acquire(&rblk->tfb_refcnt);
328 	rw_runlock(&tcp_function_lock);
329 	return(rblk);
330 }
331 
332 static struct tcp_function_block *
333 find_and_ref_tcp_default_fb(void)
334 {
335 	struct tcp_function_block *rblk;
336 
337 	rw_rlock(&tcp_function_lock);
338 	rblk = tcp_func_set_ptr;
339 	refcount_acquire(&rblk->tfb_refcnt);
340 	rw_runlock(&tcp_function_lock);
341 	return (rblk);
342 }
343 
344 void
345 tcp_switch_back_to_default(struct tcpcb *tp)
346 {
347 	struct tcp_function_block *tfb;
348 
349 	KASSERT(tp->t_fb != &tcp_def_funcblk,
350 	    ("%s: called by the built-in default stack", __func__));
351 
352 	/*
353 	 * Release the old stack. This function will either find a new one
354 	 * or panic.
355 	 */
356 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
357 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
358 	refcount_release(&tp->t_fb->tfb_refcnt);
359 
360 	/*
361 	 * Now, we'll find a new function block to use.
362 	 * Start by trying the current user-selected
363 	 * default, unless this stack is the user-selected
364 	 * default.
365 	 */
366 	tfb = find_and_ref_tcp_default_fb();
367 	if (tfb == tp->t_fb) {
368 		refcount_release(&tfb->tfb_refcnt);
369 		tfb = NULL;
370 	}
371 	/* Does the stack accept this connection? */
372 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
373 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
374 		refcount_release(&tfb->tfb_refcnt);
375 		tfb = NULL;
376 	}
377 	/* Try to use that stack. */
378 	if (tfb != NULL) {
379 		/* Initialize the new stack. If it succeeds, we are done. */
380 		tp->t_fb = tfb;
381 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
382 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
383 			return;
384 
385 		/*
386 		 * Initialization failed. Release the reference count on
387 		 * the stack.
388 		 */
389 		refcount_release(&tfb->tfb_refcnt);
390 	}
391 
392 	/*
393 	 * If that wasn't feasible, use the built-in default
394 	 * stack which is not allowed to reject anyone.
395 	 */
396 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
397 	if (tfb == NULL) {
398 		/* there always should be a default */
399 		panic("Can't refer to tcp_def_funcblk");
400 	}
401 	if (tfb->tfb_tcp_handoff_ok != NULL) {
402 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
403 			/* The default stack cannot say no */
404 			panic("Default stack rejects a new session?");
405 		}
406 	}
407 	tp->t_fb = tfb;
408 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
409 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
410 		/* The default stack cannot fail */
411 		panic("Default stack initialization failed");
412 	}
413 }
414 
415 static int
416 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
417 {
418 	int error=ENOENT;
419 	struct tcp_function_set fs;
420 	struct tcp_function_block *blk;
421 
422 	memset(&fs, 0, sizeof(fs));
423 	rw_rlock(&tcp_function_lock);
424 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
425 	if (blk) {
426 		/* Found him */
427 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
428 		fs.pcbcnt = blk->tfb_refcnt;
429 	}
430 	rw_runlock(&tcp_function_lock);
431 	error = sysctl_handle_string(oidp, fs.function_set_name,
432 				     sizeof(fs.function_set_name), req);
433 
434 	/* Check for error or no change */
435 	if (error != 0 || req->newptr == NULL)
436 		return(error);
437 
438 	rw_wlock(&tcp_function_lock);
439 	blk = find_tcp_functions_locked(&fs);
440 	if ((blk == NULL) ||
441 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
442 		error = ENOENT;
443 		goto done;
444 	}
445 	tcp_func_set_ptr = blk;
446 done:
447 	rw_wunlock(&tcp_function_lock);
448 	return (error);
449 }
450 
451 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
452 	    CTLTYPE_STRING | CTLFLAG_RW,
453 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
454 	    "Set/get the default TCP functions");
455 
456 static int
457 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
458 {
459 	int error, cnt, linesz;
460 	struct tcp_function *f;
461 	char *buffer, *cp;
462 	size_t bufsz, outsz;
463 	bool alias;
464 
465 	cnt = 0;
466 	rw_rlock(&tcp_function_lock);
467 	TAILQ_FOREACH(f, &t_functions, tf_next) {
468 		cnt++;
469 	}
470 	rw_runlock(&tcp_function_lock);
471 
472 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
473 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
474 
475 	error = 0;
476 	cp = buffer;
477 
478 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
479 	    "Alias", "PCB count");
480 	cp += linesz;
481 	bufsz -= linesz;
482 	outsz = linesz;
483 
484 	rw_rlock(&tcp_function_lock);
485 	TAILQ_FOREACH(f, &t_functions, tf_next) {
486 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
487 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
488 		    f->tf_fb->tfb_tcp_block_name,
489 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
490 		    alias ? f->tf_name : "-",
491 		    f->tf_fb->tfb_refcnt);
492 		if (linesz >= bufsz) {
493 			error = EOVERFLOW;
494 			break;
495 		}
496 		cp += linesz;
497 		bufsz -= linesz;
498 		outsz += linesz;
499 	}
500 	rw_runlock(&tcp_function_lock);
501 	if (error == 0)
502 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
503 	free(buffer, M_TEMP);
504 	return (error);
505 }
506 
507 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
508 	    CTLTYPE_STRING|CTLFLAG_RD,
509 	    NULL, 0, sysctl_net_inet_list_available, "A",
510 	    "list available TCP Function sets");
511 
512 /*
513  * Exports one (struct tcp_function_info) for each alias/name.
514  */
515 static int
516 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
517 {
518 	int cnt, error;
519 	struct tcp_function *f;
520 	struct tcp_function_info tfi;
521 
522 	/*
523 	 * We don't allow writes.
524 	 */
525 	if (req->newptr != NULL)
526 		return (EINVAL);
527 
528 	/*
529 	 * Wire the old buffer so we can directly copy the functions to
530 	 * user space without dropping the lock.
531 	 */
532 	if (req->oldptr != NULL) {
533 		error = sysctl_wire_old_buffer(req, 0);
534 		if (error)
535 			return (error);
536 	}
537 
538 	/*
539 	 * Walk the list and copy out matching entries. If INVARIANTS
540 	 * is compiled in, also walk the list to verify the length of
541 	 * the list matches what we have recorded.
542 	 */
543 	rw_rlock(&tcp_function_lock);
544 
545 	cnt = 0;
546 #ifndef INVARIANTS
547 	if (req->oldptr == NULL) {
548 		cnt = tcp_fb_cnt;
549 		goto skip_loop;
550 	}
551 #endif
552 	TAILQ_FOREACH(f, &t_functions, tf_next) {
553 #ifdef INVARIANTS
554 		cnt++;
555 #endif
556 		if (req->oldptr != NULL) {
557 			bzero(&tfi, sizeof(tfi));
558 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
559 			tfi.tfi_id = f->tf_fb->tfb_id;
560 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
561 			    sizeof(tfi.tfi_alias));
562 			(void)strlcpy(tfi.tfi_name,
563 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
564 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
565 			/*
566 			 * Don't stop on error, as that is the
567 			 * mechanism we use to accumulate length
568 			 * information if the buffer was too short.
569 			 */
570 		}
571 	}
572 	KASSERT(cnt == tcp_fb_cnt,
573 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
574 #ifndef INVARIANTS
575 skip_loop:
576 #endif
577 	rw_runlock(&tcp_function_lock);
578 	if (req->oldptr == NULL)
579 		error = SYSCTL_OUT(req, NULL,
580 		    (cnt + 1) * sizeof(struct tcp_function_info));
581 
582 	return (error);
583 }
584 
585 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
586 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
587 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
588 	    "List TCP function block name-to-ID mappings");
589 
590 /*
591  * tfb_tcp_handoff_ok() function for the default stack.
592  * Note that we'll basically try to take all comers.
593  */
594 static int
595 tcp_default_handoff_ok(struct tcpcb *tp)
596 {
597 
598 	return (0);
599 }
600 
601 /*
602  * tfb_tcp_fb_init() function for the default stack.
603  *
604  * This handles making sure we have appropriate timers set if you are
605  * transitioning a socket that has some amount of setup done.
606  *
607  * The init() fuction from the default can *never* return non-zero i.e.
608  * it is required to always succeed since it is the stack of last resort!
609  */
610 static int
611 tcp_default_fb_init(struct tcpcb *tp)
612 {
613 
614 	struct socket *so;
615 
616 	INP_WLOCK_ASSERT(tp->t_inpcb);
617 
618 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
619 	    ("%s: connection %p in unexpected state %d", __func__, tp,
620 	    tp->t_state));
621 
622 	/*
623 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
624 	 * know what to do for unexpected states (which includes TIME_WAIT).
625 	 */
626 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
627 		return (0);
628 
629 	/*
630 	 * Make sure some kind of transmission timer is set if there is
631 	 * outstanding data.
632 	 */
633 	so = tp->t_inpcb->inp_socket;
634 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
635 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
636 	    tcp_timer_active(tp, TT_PERSIST))) {
637 		/*
638 		 * If the session has established and it looks like it should
639 		 * be in the persist state, set the persist timer. Otherwise,
640 		 * set the retransmit timer.
641 		 */
642 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
643 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
644 		    (int32_t)sbavail(&so->so_snd))
645 			tcp_setpersist(tp);
646 		else
647 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
648 	}
649 
650 	/* All non-embryonic sessions get a keepalive timer. */
651 	if (!tcp_timer_active(tp, TT_KEEP))
652 		tcp_timer_activate(tp, TT_KEEP,
653 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
654 		    TP_KEEPINIT(tp));
655 
656 	return (0);
657 }
658 
659 /*
660  * tfb_tcp_fb_fini() function for the default stack.
661  *
662  * This changes state as necessary (or prudent) to prepare for another stack
663  * to assume responsibility for the connection.
664  */
665 static void
666 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
667 {
668 
669 	INP_WLOCK_ASSERT(tp->t_inpcb);
670 	return;
671 }
672 
673 /*
674  * Target size of TCP PCB hash tables. Must be a power of two.
675  *
676  * Note that this can be overridden by the kernel environment
677  * variable net.inet.tcp.tcbhashsize
678  */
679 #ifndef TCBHASHSIZE
680 #define TCBHASHSIZE	0
681 #endif
682 
683 /*
684  * XXX
685  * Callouts should be moved into struct tcp directly.  They are currently
686  * separate because the tcpcb structure is exported to userland for sysctl
687  * parsing purposes, which do not know about callouts.
688  */
689 struct tcpcb_mem {
690 	struct	tcpcb		tcb;
691 	struct	tcp_timer	tt;
692 	struct	cc_var		ccv;
693 #ifdef TCP_HHOOK
694 	struct	osd		osd;
695 #endif
696 };
697 
698 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
699 #define	V_tcpcb_zone			VNET(tcpcb_zone)
700 
701 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
702 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
703 
704 static struct mtx isn_mtx;
705 
706 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
707 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
708 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
709 
710 /*
711  * TCP initialization.
712  */
713 static void
714 tcp_zone_change(void *tag)
715 {
716 
717 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
718 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
719 	tcp_tw_zone_change();
720 }
721 
722 static int
723 tcp_inpcb_init(void *mem, int size, int flags)
724 {
725 	struct inpcb *inp = mem;
726 
727 	INP_LOCK_INIT(inp, "inp", "tcpinp");
728 	return (0);
729 }
730 
731 /*
732  * Take a value and get the next power of 2 that doesn't overflow.
733  * Used to size the tcp_inpcb hash buckets.
734  */
735 static int
736 maketcp_hashsize(int size)
737 {
738 	int hashsize;
739 
740 	/*
741 	 * auto tune.
742 	 * get the next power of 2 higher than maxsockets.
743 	 */
744 	hashsize = 1 << fls(size);
745 	/* catch overflow, and just go one power of 2 smaller */
746 	if (hashsize < size) {
747 		hashsize = 1 << (fls(size) - 1);
748 	}
749 	return (hashsize);
750 }
751 
752 static volatile int next_tcp_stack_id = 1;
753 
754 /*
755  * Register a TCP function block with the name provided in the names
756  * array.  (Note that this function does NOT automatically register
757  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
758  * explicitly include blk->tfb_tcp_block_name in the list of names if
759  * you wish to register the stack with that name.)
760  *
761  * Either all name registrations will succeed or all will fail.  If
762  * a name registration fails, the function will update the num_names
763  * argument to point to the array index of the name that encountered
764  * the failure.
765  *
766  * Returns 0 on success, or an error code on failure.
767  */
768 int
769 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
770     const char *names[], int *num_names)
771 {
772 	struct tcp_function *n;
773 	struct tcp_function_set fs;
774 	int error, i;
775 
776 	KASSERT(names != NULL && *num_names > 0,
777 	    ("%s: Called with 0-length name list", __func__));
778 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
779 	KASSERT(rw_initialized(&tcp_function_lock),
780 	    ("%s: called too early", __func__));
781 
782 	if ((blk->tfb_tcp_output == NULL) ||
783 	    (blk->tfb_tcp_do_segment == NULL) ||
784 	    (blk->tfb_tcp_ctloutput == NULL) ||
785 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
786 		/*
787 		 * These functions are required and you
788 		 * need a name.
789 		 */
790 		*num_names = 0;
791 		return (EINVAL);
792 	}
793 	if (blk->tfb_tcp_timer_stop_all ||
794 	    blk->tfb_tcp_timer_activate ||
795 	    blk->tfb_tcp_timer_active ||
796 	    blk->tfb_tcp_timer_stop) {
797 		/*
798 		 * If you define one timer function you
799 		 * must have them all.
800 		 */
801 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
802 		    (blk->tfb_tcp_timer_activate == NULL) ||
803 		    (blk->tfb_tcp_timer_active == NULL) ||
804 		    (blk->tfb_tcp_timer_stop == NULL)) {
805 			*num_names = 0;
806 			return (EINVAL);
807 		}
808 	}
809 
810 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
811 		*num_names = 0;
812 		return (EINVAL);
813 	}
814 
815 	refcount_init(&blk->tfb_refcnt, 0);
816 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
817 	for (i = 0; i < *num_names; i++) {
818 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
819 		if (n == NULL) {
820 			error = ENOMEM;
821 			goto cleanup;
822 		}
823 		n->tf_fb = blk;
824 
825 		(void)strlcpy(fs.function_set_name, names[i],
826 		    sizeof(fs.function_set_name));
827 		rw_wlock(&tcp_function_lock);
828 		if (find_tcp_functions_locked(&fs) != NULL) {
829 			/* Duplicate name space not allowed */
830 			rw_wunlock(&tcp_function_lock);
831 			free(n, M_TCPFUNCTIONS);
832 			error = EALREADY;
833 			goto cleanup;
834 		}
835 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
836 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
837 		tcp_fb_cnt++;
838 		rw_wunlock(&tcp_function_lock);
839 	}
840 	return(0);
841 
842 cleanup:
843 	/*
844 	 * Deregister the names we just added. Because registration failed
845 	 * for names[i], we don't need to deregister that name.
846 	 */
847 	*num_names = i;
848 	rw_wlock(&tcp_function_lock);
849 	while (--i >= 0) {
850 		TAILQ_FOREACH(n, &t_functions, tf_next) {
851 			if (!strncmp(n->tf_name, names[i],
852 			    TCP_FUNCTION_NAME_LEN_MAX)) {
853 				TAILQ_REMOVE(&t_functions, n, tf_next);
854 				tcp_fb_cnt--;
855 				n->tf_fb = NULL;
856 				free(n, M_TCPFUNCTIONS);
857 				break;
858 			}
859 		}
860 	}
861 	rw_wunlock(&tcp_function_lock);
862 	return (error);
863 }
864 
865 /*
866  * Register a TCP function block using the name provided in the name
867  * argument.
868  *
869  * Returns 0 on success, or an error code on failure.
870  */
871 int
872 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
873     int wait)
874 {
875 	const char *name_list[1];
876 	int num_names, rv;
877 
878 	num_names = 1;
879 	if (name != NULL)
880 		name_list[0] = name;
881 	else
882 		name_list[0] = blk->tfb_tcp_block_name;
883 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
884 	return (rv);
885 }
886 
887 /*
888  * Register a TCP function block using the name defined in
889  * blk->tfb_tcp_block_name.
890  *
891  * Returns 0 on success, or an error code on failure.
892  */
893 int
894 register_tcp_functions(struct tcp_function_block *blk, int wait)
895 {
896 
897 	return (register_tcp_functions_as_name(blk, NULL, wait));
898 }
899 
900 /*
901  * Deregister all names associated with a function block. This
902  * functionally removes the function block from use within the system.
903  *
904  * When called with a true quiesce argument, mark the function block
905  * as being removed so no more stacks will use it and determine
906  * whether the removal would succeed.
907  *
908  * When called with a false quiesce argument, actually attempt the
909  * removal.
910  *
911  * When called with a force argument, attempt to switch all TCBs to
912  * use the default stack instead of returning EBUSY.
913  *
914  * Returns 0 on success (or if the removal would succeed, or an error
915  * code on failure.
916  */
917 int
918 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
919     bool force)
920 {
921 	struct tcp_function *f;
922 
923 	if (blk == &tcp_def_funcblk) {
924 		/* You can't un-register the default */
925 		return (EPERM);
926 	}
927 	rw_wlock(&tcp_function_lock);
928 	if (blk == tcp_func_set_ptr) {
929 		/* You can't free the current default */
930 		rw_wunlock(&tcp_function_lock);
931 		return (EBUSY);
932 	}
933 	/* Mark the block so no more stacks can use it. */
934 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
935 	/*
936 	 * If TCBs are still attached to the stack, attempt to switch them
937 	 * to the default stack.
938 	 */
939 	if (force && blk->tfb_refcnt) {
940 		struct inpcb *inp;
941 		struct tcpcb *tp;
942 		VNET_ITERATOR_DECL(vnet_iter);
943 
944 		rw_wunlock(&tcp_function_lock);
945 
946 		VNET_LIST_RLOCK();
947 		VNET_FOREACH(vnet_iter) {
948 			CURVNET_SET(vnet_iter);
949 			INP_INFO_WLOCK(&V_tcbinfo);
950 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
951 				INP_WLOCK(inp);
952 				if (inp->inp_flags & INP_TIMEWAIT) {
953 					INP_WUNLOCK(inp);
954 					continue;
955 				}
956 				tp = intotcpcb(inp);
957 				if (tp == NULL || tp->t_fb != blk) {
958 					INP_WUNLOCK(inp);
959 					continue;
960 				}
961 				tcp_switch_back_to_default(tp);
962 				INP_WUNLOCK(inp);
963 			}
964 			INP_INFO_WUNLOCK(&V_tcbinfo);
965 			CURVNET_RESTORE();
966 		}
967 		VNET_LIST_RUNLOCK();
968 
969 		rw_wlock(&tcp_function_lock);
970 	}
971 	if (blk->tfb_refcnt) {
972 		/* TCBs still attached. */
973 		rw_wunlock(&tcp_function_lock);
974 		return (EBUSY);
975 	}
976 	if (quiesce) {
977 		/* Skip removal. */
978 		rw_wunlock(&tcp_function_lock);
979 		return (0);
980 	}
981 	/* Remove any function names that map to this function block. */
982 	while (find_tcp_fb_locked(blk, &f) != NULL) {
983 		TAILQ_REMOVE(&t_functions, f, tf_next);
984 		tcp_fb_cnt--;
985 		f->tf_fb = NULL;
986 		free(f, M_TCPFUNCTIONS);
987 	}
988 	rw_wunlock(&tcp_function_lock);
989 	return (0);
990 }
991 
992 void
993 tcp_init(void)
994 {
995 	const char *tcbhash_tuneable;
996 	int hashsize;
997 
998 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
999 
1000 #ifdef TCP_HHOOK
1001 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1002 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1003 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1004 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1005 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1006 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1007 #endif
1008 	hashsize = TCBHASHSIZE;
1009 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1010 	if (hashsize == 0) {
1011 		/*
1012 		 * Auto tune the hash size based on maxsockets.
1013 		 * A perfect hash would have a 1:1 mapping
1014 		 * (hashsize = maxsockets) however it's been
1015 		 * suggested that O(2) average is better.
1016 		 */
1017 		hashsize = maketcp_hashsize(maxsockets / 4);
1018 		/*
1019 		 * Our historical default is 512,
1020 		 * do not autotune lower than this.
1021 		 */
1022 		if (hashsize < 512)
1023 			hashsize = 512;
1024 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1025 			printf("%s: %s auto tuned to %d\n", __func__,
1026 			    tcbhash_tuneable, hashsize);
1027 	}
1028 	/*
1029 	 * We require a hashsize to be a power of two.
1030 	 * Previously if it was not a power of two we would just reset it
1031 	 * back to 512, which could be a nasty surprise if you did not notice
1032 	 * the error message.
1033 	 * Instead what we do is clip it to the closest power of two lower
1034 	 * than the specified hash value.
1035 	 */
1036 	if (!powerof2(hashsize)) {
1037 		int oldhashsize = hashsize;
1038 
1039 		hashsize = maketcp_hashsize(hashsize);
1040 		/* prevent absurdly low value */
1041 		if (hashsize < 16)
1042 			hashsize = 16;
1043 		printf("%s: WARNING: TCB hash size not a power of 2, "
1044 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1045 		    hashsize);
1046 	}
1047 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1048 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1049 
1050 	/*
1051 	 * These have to be type stable for the benefit of the timers.
1052 	 */
1053 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1054 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1055 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1056 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1057 
1058 	tcp_tw_init();
1059 	syncache_init();
1060 	tcp_hc_init();
1061 
1062 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1063 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1064 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1065 
1066 	tcp_fastopen_init();
1067 
1068 	/* Skip initialization of globals for non-default instances. */
1069 	if (!IS_DEFAULT_VNET(curvnet))
1070 		return;
1071 
1072 	tcp_reass_global_init();
1073 
1074 	/* XXX virtualize those bellow? */
1075 	tcp_delacktime = TCPTV_DELACK;
1076 	tcp_keepinit = TCPTV_KEEP_INIT;
1077 	tcp_keepidle = TCPTV_KEEP_IDLE;
1078 	tcp_keepintvl = TCPTV_KEEPINTVL;
1079 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1080 	tcp_msl = TCPTV_MSL;
1081 	tcp_rexmit_initial = TCPTV_RTOBASE;
1082 	if (tcp_rexmit_initial < 1)
1083 		tcp_rexmit_initial = 1;
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 
1093 	/* Setup the tcp function block list */
1094 	TAILQ_INIT(&t_functions);
1095 	rw_init(&tcp_function_lock, "tcp_func_lock");
1096 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1097 #ifdef TCP_BLACKBOX
1098 	/* Initialize the TCP logging data. */
1099 	tcp_log_init();
1100 #endif
1101 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1102 
1103 	if (tcp_soreceive_stream) {
1104 #ifdef INET
1105 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1106 #endif
1107 #ifdef INET6
1108 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1109 #endif /* INET6 */
1110 	}
1111 
1112 #ifdef INET6
1113 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1114 #else /* INET6 */
1115 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1116 #endif /* INET6 */
1117 	if (max_protohdr < TCP_MINPROTOHDR)
1118 		max_protohdr = TCP_MINPROTOHDR;
1119 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1120 		panic("tcp_init");
1121 #undef TCP_MINPROTOHDR
1122 
1123 	ISN_LOCK_INIT();
1124 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1125 		SHUTDOWN_PRI_DEFAULT);
1126 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1127 		EVENTHANDLER_PRI_ANY);
1128 #ifdef TCPPCAP
1129 	tcp_pcap_init();
1130 #endif
1131 }
1132 
1133 #ifdef VIMAGE
1134 static void
1135 tcp_destroy(void *unused __unused)
1136 {
1137 	int n;
1138 #ifdef TCP_HHOOK
1139 	int error;
1140 #endif
1141 
1142 	/*
1143 	 * All our processes are gone, all our sockets should be cleaned
1144 	 * up, which means, we should be past the tcp_discardcb() calls.
1145 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1146 	 */
1147 	for (;;) {
1148 		INP_LIST_RLOCK(&V_tcbinfo);
1149 		n = V_tcbinfo.ipi_count;
1150 		INP_LIST_RUNLOCK(&V_tcbinfo);
1151 		if (n == 0)
1152 			break;
1153 		pause("tcpdes", hz / 10);
1154 	}
1155 	tcp_hc_destroy();
1156 	syncache_destroy();
1157 	tcp_tw_destroy();
1158 	in_pcbinfo_destroy(&V_tcbinfo);
1159 	/* tcp_discardcb() clears the sack_holes up. */
1160 	uma_zdestroy(V_sack_hole_zone);
1161 	uma_zdestroy(V_tcpcb_zone);
1162 
1163 	/*
1164 	 * Cannot free the zone until all tcpcbs are released as we attach
1165 	 * the allocations to them.
1166 	 */
1167 	tcp_fastopen_destroy();
1168 
1169 #ifdef TCP_HHOOK
1170 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1171 	if (error != 0) {
1172 		printf("%s: WARNING: unable to deregister helper hook "
1173 		    "type=%d, id=%d: error %d returned\n", __func__,
1174 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1175 	}
1176 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1177 	if (error != 0) {
1178 		printf("%s: WARNING: unable to deregister helper hook "
1179 		    "type=%d, id=%d: error %d returned\n", __func__,
1180 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1181 	}
1182 #endif
1183 }
1184 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1185 #endif
1186 
1187 void
1188 tcp_fini(void *xtp)
1189 {
1190 
1191 }
1192 
1193 /*
1194  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1195  * tcp_template used to store this data in mbufs, but we now recopy it out
1196  * of the tcpcb each time to conserve mbufs.
1197  */
1198 void
1199 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1200 {
1201 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1202 
1203 	INP_WLOCK_ASSERT(inp);
1204 
1205 #ifdef INET6
1206 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1207 		struct ip6_hdr *ip6;
1208 
1209 		ip6 = (struct ip6_hdr *)ip_ptr;
1210 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1211 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1212 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1213 			(IPV6_VERSION & IPV6_VERSION_MASK);
1214 		ip6->ip6_nxt = IPPROTO_TCP;
1215 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1216 		ip6->ip6_src = inp->in6p_laddr;
1217 		ip6->ip6_dst = inp->in6p_faddr;
1218 	}
1219 #endif /* INET6 */
1220 #if defined(INET6) && defined(INET)
1221 	else
1222 #endif
1223 #ifdef INET
1224 	{
1225 		struct ip *ip;
1226 
1227 		ip = (struct ip *)ip_ptr;
1228 		ip->ip_v = IPVERSION;
1229 		ip->ip_hl = 5;
1230 		ip->ip_tos = inp->inp_ip_tos;
1231 		ip->ip_len = 0;
1232 		ip->ip_id = 0;
1233 		ip->ip_off = 0;
1234 		ip->ip_ttl = inp->inp_ip_ttl;
1235 		ip->ip_sum = 0;
1236 		ip->ip_p = IPPROTO_TCP;
1237 		ip->ip_src = inp->inp_laddr;
1238 		ip->ip_dst = inp->inp_faddr;
1239 	}
1240 #endif /* INET */
1241 	th->th_sport = inp->inp_lport;
1242 	th->th_dport = inp->inp_fport;
1243 	th->th_seq = 0;
1244 	th->th_ack = 0;
1245 	th->th_x2 = 0;
1246 	th->th_off = 5;
1247 	th->th_flags = 0;
1248 	th->th_win = 0;
1249 	th->th_urp = 0;
1250 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1251 }
1252 
1253 /*
1254  * Create template to be used to send tcp packets on a connection.
1255  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1256  * use for this function is in keepalives, which use tcp_respond.
1257  */
1258 struct tcptemp *
1259 tcpip_maketemplate(struct inpcb *inp)
1260 {
1261 	struct tcptemp *t;
1262 
1263 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1264 	if (t == NULL)
1265 		return (NULL);
1266 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1267 	return (t);
1268 }
1269 
1270 /*
1271  * Send a single message to the TCP at address specified by
1272  * the given TCP/IP header.  If m == NULL, then we make a copy
1273  * of the tcpiphdr at th and send directly to the addressed host.
1274  * This is used to force keep alive messages out using the TCP
1275  * template for a connection.  If flags are given then we send
1276  * a message back to the TCP which originated the segment th,
1277  * and discard the mbuf containing it and any other attached mbufs.
1278  *
1279  * In any case the ack and sequence number of the transmitted
1280  * segment are as specified by the parameters.
1281  *
1282  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1283  */
1284 void
1285 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1286     tcp_seq ack, tcp_seq seq, int flags)
1287 {
1288 	struct tcpopt to;
1289 	struct inpcb *inp;
1290 	struct ip *ip;
1291 	struct mbuf *optm;
1292 	struct tcphdr *nth;
1293 	u_char *optp;
1294 #ifdef INET6
1295 	struct ip6_hdr *ip6;
1296 	int isipv6;
1297 #endif /* INET6 */
1298 	int optlen, tlen, win;
1299 	bool incl_opts;
1300 
1301 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1302 
1303 #ifdef INET6
1304 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1305 	ip6 = ipgen;
1306 #endif /* INET6 */
1307 	ip = ipgen;
1308 
1309 	if (tp != NULL) {
1310 		inp = tp->t_inpcb;
1311 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1312 		INP_WLOCK_ASSERT(inp);
1313 	} else
1314 		inp = NULL;
1315 
1316 	incl_opts = false;
1317 	win = 0;
1318 	if (tp != NULL) {
1319 		if (!(flags & TH_RST)) {
1320 			win = sbspace(&inp->inp_socket->so_rcv);
1321 			if (win > TCP_MAXWIN << tp->rcv_scale)
1322 				win = TCP_MAXWIN << tp->rcv_scale;
1323 		}
1324 		if ((tp->t_flags & TF_NOOPT) == 0)
1325 			incl_opts = true;
1326 	}
1327 	if (m == NULL) {
1328 		m = m_gethdr(M_NOWAIT, MT_DATA);
1329 		if (m == NULL)
1330 			return;
1331 		m->m_data += max_linkhdr;
1332 #ifdef INET6
1333 		if (isipv6) {
1334 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1335 			      sizeof(struct ip6_hdr));
1336 			ip6 = mtod(m, struct ip6_hdr *);
1337 			nth = (struct tcphdr *)(ip6 + 1);
1338 		} else
1339 #endif /* INET6 */
1340 		{
1341 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1342 			ip = mtod(m, struct ip *);
1343 			nth = (struct tcphdr *)(ip + 1);
1344 		}
1345 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1346 		flags = TH_ACK;
1347 	} else if (!M_WRITABLE(m)) {
1348 		struct mbuf *n;
1349 
1350 		/* Can't reuse 'm', allocate a new mbuf. */
1351 		n = m_gethdr(M_NOWAIT, MT_DATA);
1352 		if (n == NULL) {
1353 			m_freem(m);
1354 			return;
1355 		}
1356 
1357 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1358 			m_freem(m);
1359 			m_freem(n);
1360 			return;
1361 		}
1362 
1363 		n->m_data += max_linkhdr;
1364 		/* m_len is set later */
1365 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1366 #ifdef INET6
1367 		if (isipv6) {
1368 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1369 			      sizeof(struct ip6_hdr));
1370 			ip6 = mtod(n, struct ip6_hdr *);
1371 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1372 			nth = (struct tcphdr *)(ip6 + 1);
1373 		} else
1374 #endif /* INET6 */
1375 		{
1376 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1377 			ip = mtod(n, struct ip *);
1378 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1379 			nth = (struct tcphdr *)(ip + 1);
1380 		}
1381 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1382 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1383 		th = nth;
1384 		m_freem(m);
1385 		m = n;
1386 	} else {
1387 		/*
1388 		 *  reuse the mbuf.
1389 		 * XXX MRT We inherit the FIB, which is lucky.
1390 		 */
1391 		m_freem(m->m_next);
1392 		m->m_next = NULL;
1393 		m->m_data = (caddr_t)ipgen;
1394 		/* m_len is set later */
1395 #ifdef INET6
1396 		if (isipv6) {
1397 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1398 			nth = (struct tcphdr *)(ip6 + 1);
1399 		} else
1400 #endif /* INET6 */
1401 		{
1402 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1403 			nth = (struct tcphdr *)(ip + 1);
1404 		}
1405 		if (th != nth) {
1406 			/*
1407 			 * this is usually a case when an extension header
1408 			 * exists between the IPv6 header and the
1409 			 * TCP header.
1410 			 */
1411 			nth->th_sport = th->th_sport;
1412 			nth->th_dport = th->th_dport;
1413 		}
1414 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1415 #undef xchg
1416 	}
1417 	tlen = 0;
1418 #ifdef INET6
1419 	if (isipv6)
1420 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1421 #endif
1422 #if defined(INET) && defined(INET6)
1423 	else
1424 #endif
1425 #ifdef INET
1426 		tlen = sizeof (struct tcpiphdr);
1427 #endif
1428 #ifdef INVARIANTS
1429 	m->m_len = 0;
1430 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1431 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1432 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1433 #endif
1434 	m->m_len = tlen;
1435 	to.to_flags = 0;
1436 	if (incl_opts) {
1437 		/* Make sure we have room. */
1438 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1439 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1440 			if (m->m_next) {
1441 				optp = mtod(m->m_next, u_char *);
1442 				optm = m->m_next;
1443 			} else
1444 				incl_opts = false;
1445 		} else {
1446 			optp = (u_char *) (nth + 1);
1447 			optm = m;
1448 		}
1449 	}
1450 	if (incl_opts) {
1451 		/* Timestamps. */
1452 		if (tp->t_flags & TF_RCVD_TSTMP) {
1453 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1454 			to.to_tsecr = tp->ts_recent;
1455 			to.to_flags |= TOF_TS;
1456 		}
1457 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1458 		/* TCP-MD5 (RFC2385). */
1459 		if (tp->t_flags & TF_SIGNATURE)
1460 			to.to_flags |= TOF_SIGNATURE;
1461 #endif
1462 		/* Add the options. */
1463 		tlen += optlen = tcp_addoptions(&to, optp);
1464 
1465 		/* Update m_len in the correct mbuf. */
1466 		optm->m_len += optlen;
1467 	} else
1468 		optlen = 0;
1469 #ifdef INET6
1470 	if (isipv6) {
1471 		ip6->ip6_flow = 0;
1472 		ip6->ip6_vfc = IPV6_VERSION;
1473 		ip6->ip6_nxt = IPPROTO_TCP;
1474 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1475 	}
1476 #endif
1477 #if defined(INET) && defined(INET6)
1478 	else
1479 #endif
1480 #ifdef INET
1481 	{
1482 		ip->ip_len = htons(tlen);
1483 		ip->ip_ttl = V_ip_defttl;
1484 		if (V_path_mtu_discovery)
1485 			ip->ip_off |= htons(IP_DF);
1486 	}
1487 #endif
1488 	m->m_pkthdr.len = tlen;
1489 	m->m_pkthdr.rcvif = NULL;
1490 #ifdef MAC
1491 	if (inp != NULL) {
1492 		/*
1493 		 * Packet is associated with a socket, so allow the
1494 		 * label of the response to reflect the socket label.
1495 		 */
1496 		INP_WLOCK_ASSERT(inp);
1497 		mac_inpcb_create_mbuf(inp, m);
1498 	} else {
1499 		/*
1500 		 * Packet is not associated with a socket, so possibly
1501 		 * update the label in place.
1502 		 */
1503 		mac_netinet_tcp_reply(m);
1504 	}
1505 #endif
1506 	nth->th_seq = htonl(seq);
1507 	nth->th_ack = htonl(ack);
1508 	nth->th_x2 = 0;
1509 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1510 	nth->th_flags = flags;
1511 	if (tp != NULL)
1512 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1513 	else
1514 		nth->th_win = htons((u_short)win);
1515 	nth->th_urp = 0;
1516 
1517 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1518 	if (to.to_flags & TOF_SIGNATURE) {
1519 		if (!TCPMD5_ENABLED() ||
1520 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1521 			m_freem(m);
1522 			return;
1523 		}
1524 	}
1525 #endif
1526 
1527 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1528 #ifdef INET6
1529 	if (isipv6) {
1530 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1531 		nth->th_sum = in6_cksum_pseudo(ip6,
1532 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1533 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1534 		    NULL, NULL);
1535 	}
1536 #endif /* INET6 */
1537 #if defined(INET6) && defined(INET)
1538 	else
1539 #endif
1540 #ifdef INET
1541 	{
1542 		m->m_pkthdr.csum_flags = CSUM_TCP;
1543 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1544 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1545 	}
1546 #endif /* INET */
1547 #ifdef TCPDEBUG
1548 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1549 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1550 #endif
1551 	TCP_PROBE3(debug__output, tp, th, m);
1552 	if (flags & TH_RST)
1553 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1554 
1555 #ifdef INET6
1556 	if (isipv6) {
1557 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1558 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1559 	}
1560 #endif /* INET6 */
1561 #if defined(INET) && defined(INET6)
1562 	else
1563 #endif
1564 #ifdef INET
1565 	{
1566 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1567 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1568 	}
1569 #endif
1570 }
1571 
1572 /*
1573  * Create a new TCP control block, making an
1574  * empty reassembly queue and hooking it to the argument
1575  * protocol control block.  The `inp' parameter must have
1576  * come from the zone allocator set up in tcp_init().
1577  */
1578 struct tcpcb *
1579 tcp_newtcpcb(struct inpcb *inp)
1580 {
1581 	struct tcpcb_mem *tm;
1582 	struct tcpcb *tp;
1583 #ifdef INET6
1584 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1585 #endif /* INET6 */
1586 
1587 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1588 	if (tm == NULL)
1589 		return (NULL);
1590 	tp = &tm->tcb;
1591 
1592 	/* Initialise cc_var struct for this tcpcb. */
1593 	tp->ccv = &tm->ccv;
1594 	tp->ccv->type = IPPROTO_TCP;
1595 	tp->ccv->ccvc.tcp = tp;
1596 	rw_rlock(&tcp_function_lock);
1597 	tp->t_fb = tcp_func_set_ptr;
1598 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1599 	rw_runlock(&tcp_function_lock);
1600 	/*
1601 	 * Use the current system default CC algorithm.
1602 	 */
1603 	CC_LIST_RLOCK();
1604 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1605 	CC_ALGO(tp) = CC_DEFAULT();
1606 	CC_LIST_RUNLOCK();
1607 
1608 	if (CC_ALGO(tp)->cb_init != NULL)
1609 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1610 			if (tp->t_fb->tfb_tcp_fb_fini)
1611 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1612 			refcount_release(&tp->t_fb->tfb_refcnt);
1613 			uma_zfree(V_tcpcb_zone, tm);
1614 			return (NULL);
1615 		}
1616 
1617 #ifdef TCP_HHOOK
1618 	tp->osd = &tm->osd;
1619 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1620 		if (tp->t_fb->tfb_tcp_fb_fini)
1621 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1622 		refcount_release(&tp->t_fb->tfb_refcnt);
1623 		uma_zfree(V_tcpcb_zone, tm);
1624 		return (NULL);
1625 	}
1626 #endif
1627 
1628 #ifdef VIMAGE
1629 	tp->t_vnet = inp->inp_vnet;
1630 #endif
1631 	tp->t_timers = &tm->tt;
1632 	TAILQ_INIT(&tp->t_segq);
1633 	tp->t_maxseg =
1634 #ifdef INET6
1635 		isipv6 ? V_tcp_v6mssdflt :
1636 #endif /* INET6 */
1637 		V_tcp_mssdflt;
1638 
1639 	/* Set up our timeouts. */
1640 	callout_init(&tp->t_timers->tt_rexmt, 1);
1641 	callout_init(&tp->t_timers->tt_persist, 1);
1642 	callout_init(&tp->t_timers->tt_keep, 1);
1643 	callout_init(&tp->t_timers->tt_2msl, 1);
1644 	callout_init(&tp->t_timers->tt_delack, 1);
1645 
1646 	if (V_tcp_do_rfc1323)
1647 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1648 	if (V_tcp_do_sack)
1649 		tp->t_flags |= TF_SACK_PERMIT;
1650 	TAILQ_INIT(&tp->snd_holes);
1651 	/*
1652 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1653 	 * is called.
1654 	 */
1655 	in_pcbref(inp);	/* Reference for tcpcb */
1656 	tp->t_inpcb = inp;
1657 
1658 	/*
1659 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1660 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1661 	 * reasonable initial retransmit time.
1662 	 */
1663 	tp->t_srtt = TCPTV_SRTTBASE;
1664 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1665 	tp->t_rttmin = tcp_rexmit_min;
1666 	tp->t_rxtcur = tcp_rexmit_initial;
1667 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1668 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1669 	tp->t_rcvtime = ticks;
1670 	/*
1671 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1672 	 * because the socket may be bound to an IPv6 wildcard address,
1673 	 * which may match an IPv4-mapped IPv6 address.
1674 	 */
1675 	inp->inp_ip_ttl = V_ip_defttl;
1676 	inp->inp_ppcb = tp;
1677 #ifdef TCPPCAP
1678 	/*
1679 	 * Init the TCP PCAP queues.
1680 	 */
1681 	tcp_pcap_tcpcb_init(tp);
1682 #endif
1683 #ifdef TCP_BLACKBOX
1684 	/* Initialize the per-TCPCB log data. */
1685 	tcp_log_tcpcbinit(tp);
1686 #endif
1687 	if (tp->t_fb->tfb_tcp_fb_init) {
1688 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1689 	}
1690 	return (tp);		/* XXX */
1691 }
1692 
1693 /*
1694  * Switch the congestion control algorithm back to NewReno for any active
1695  * control blocks using an algorithm which is about to go away.
1696  * This ensures the CC framework can allow the unload to proceed without leaving
1697  * any dangling pointers which would trigger a panic.
1698  * Returning non-zero would inform the CC framework that something went wrong
1699  * and it would be unsafe to allow the unload to proceed. However, there is no
1700  * way for this to occur with this implementation so we always return zero.
1701  */
1702 int
1703 tcp_ccalgounload(struct cc_algo *unload_algo)
1704 {
1705 	struct cc_algo *tmpalgo;
1706 	struct inpcb *inp;
1707 	struct tcpcb *tp;
1708 	VNET_ITERATOR_DECL(vnet_iter);
1709 
1710 	/*
1711 	 * Check all active control blocks across all network stacks and change
1712 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1713 	 * requires cleanup code to be run, call it.
1714 	 */
1715 	VNET_LIST_RLOCK();
1716 	VNET_FOREACH(vnet_iter) {
1717 		CURVNET_SET(vnet_iter);
1718 		INP_INFO_WLOCK(&V_tcbinfo);
1719 		/*
1720 		 * New connections already part way through being initialised
1721 		 * with the CC algo we're removing will not race with this code
1722 		 * because the INP_INFO_WLOCK is held during initialisation. We
1723 		 * therefore don't enter the loop below until the connection
1724 		 * list has stabilised.
1725 		 */
1726 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1727 			INP_WLOCK(inp);
1728 			/* Important to skip tcptw structs. */
1729 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1730 			    (tp = intotcpcb(inp)) != NULL) {
1731 				/*
1732 				 * By holding INP_WLOCK here, we are assured
1733 				 * that the connection is not currently
1734 				 * executing inside the CC module's functions
1735 				 * i.e. it is safe to make the switch back to
1736 				 * NewReno.
1737 				 */
1738 				if (CC_ALGO(tp) == unload_algo) {
1739 					tmpalgo = CC_ALGO(tp);
1740 					if (tmpalgo->cb_destroy != NULL)
1741 						tmpalgo->cb_destroy(tp->ccv);
1742 					CC_DATA(tp) = NULL;
1743 					/*
1744 					 * NewReno may allocate memory on
1745 					 * demand for certain stateful
1746 					 * configuration as needed, but is
1747 					 * coded to never fail on memory
1748 					 * allocation failure so it is a safe
1749 					 * fallback.
1750 					 */
1751 					CC_ALGO(tp) = &newreno_cc_algo;
1752 				}
1753 			}
1754 			INP_WUNLOCK(inp);
1755 		}
1756 		INP_INFO_WUNLOCK(&V_tcbinfo);
1757 		CURVNET_RESTORE();
1758 	}
1759 	VNET_LIST_RUNLOCK();
1760 
1761 	return (0);
1762 }
1763 
1764 /*
1765  * Drop a TCP connection, reporting
1766  * the specified error.  If connection is synchronized,
1767  * then send a RST to peer.
1768  */
1769 struct tcpcb *
1770 tcp_drop(struct tcpcb *tp, int errno)
1771 {
1772 	struct socket *so = tp->t_inpcb->inp_socket;
1773 
1774 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1775 	INP_WLOCK_ASSERT(tp->t_inpcb);
1776 
1777 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1778 		tcp_state_change(tp, TCPS_CLOSED);
1779 		(void) tp->t_fb->tfb_tcp_output(tp);
1780 		TCPSTAT_INC(tcps_drops);
1781 	} else
1782 		TCPSTAT_INC(tcps_conndrops);
1783 	if (errno == ETIMEDOUT && tp->t_softerror)
1784 		errno = tp->t_softerror;
1785 	so->so_error = errno;
1786 	return (tcp_close(tp));
1787 }
1788 
1789 void
1790 tcp_discardcb(struct tcpcb *tp)
1791 {
1792 	struct inpcb *inp = tp->t_inpcb;
1793 	struct socket *so = inp->inp_socket;
1794 #ifdef INET6
1795 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1796 #endif /* INET6 */
1797 	int released __unused;
1798 
1799 	INP_WLOCK_ASSERT(inp);
1800 
1801 	/*
1802 	 * Make sure that all of our timers are stopped before we delete the
1803 	 * PCB.
1804 	 *
1805 	 * If stopping a timer fails, we schedule a discard function in same
1806 	 * callout, and the last discard function called will take care of
1807 	 * deleting the tcpcb.
1808 	 */
1809 	tp->t_timers->tt_draincnt = 0;
1810 	tcp_timer_stop(tp, TT_REXMT);
1811 	tcp_timer_stop(tp, TT_PERSIST);
1812 	tcp_timer_stop(tp, TT_KEEP);
1813 	tcp_timer_stop(tp, TT_2MSL);
1814 	tcp_timer_stop(tp, TT_DELACK);
1815 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1816 		/*
1817 		 * Call the stop-all function of the methods,
1818 		 * this function should call the tcp_timer_stop()
1819 		 * method with each of the function specific timeouts.
1820 		 * That stop will be called via the tfb_tcp_timer_stop()
1821 		 * which should use the async drain function of the
1822 		 * callout system (see tcp_var.h).
1823 		 */
1824 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1825 	}
1826 
1827 	/*
1828 	 * If we got enough samples through the srtt filter,
1829 	 * save the rtt and rttvar in the routing entry.
1830 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1831 	 * 4 samples is enough for the srtt filter to converge
1832 	 * to within enough % of the correct value; fewer samples
1833 	 * and we could save a bogus rtt. The danger is not high
1834 	 * as tcp quickly recovers from everything.
1835 	 * XXX: Works very well but needs some more statistics!
1836 	 */
1837 	if (tp->t_rttupdated >= 4) {
1838 		struct hc_metrics_lite metrics;
1839 		uint32_t ssthresh;
1840 
1841 		bzero(&metrics, sizeof(metrics));
1842 		/*
1843 		 * Update the ssthresh always when the conditions below
1844 		 * are satisfied. This gives us better new start value
1845 		 * for the congestion avoidance for new connections.
1846 		 * ssthresh is only set if packet loss occurred on a session.
1847 		 *
1848 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1849 		 * being torn down.  Ideally this code would not use 'so'.
1850 		 */
1851 		ssthresh = tp->snd_ssthresh;
1852 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1853 			/*
1854 			 * convert the limit from user data bytes to
1855 			 * packets then to packet data bytes.
1856 			 */
1857 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1858 			if (ssthresh < 2)
1859 				ssthresh = 2;
1860 			ssthresh *= (tp->t_maxseg +
1861 #ifdef INET6
1862 			    (isipv6 ? sizeof (struct ip6_hdr) +
1863 				sizeof (struct tcphdr) :
1864 #endif
1865 				sizeof (struct tcpiphdr)
1866 #ifdef INET6
1867 			    )
1868 #endif
1869 			    );
1870 		} else
1871 			ssthresh = 0;
1872 		metrics.rmx_ssthresh = ssthresh;
1873 
1874 		metrics.rmx_rtt = tp->t_srtt;
1875 		metrics.rmx_rttvar = tp->t_rttvar;
1876 		metrics.rmx_cwnd = tp->snd_cwnd;
1877 		metrics.rmx_sendpipe = 0;
1878 		metrics.rmx_recvpipe = 0;
1879 
1880 		tcp_hc_update(&inp->inp_inc, &metrics);
1881 	}
1882 
1883 	/* free the reassembly queue, if any */
1884 	tcp_reass_flush(tp);
1885 
1886 #ifdef TCP_OFFLOAD
1887 	/* Disconnect offload device, if any. */
1888 	if (tp->t_flags & TF_TOE)
1889 		tcp_offload_detach(tp);
1890 #endif
1891 
1892 	tcp_free_sackholes(tp);
1893 
1894 #ifdef TCPPCAP
1895 	/* Free the TCP PCAP queues. */
1896 	tcp_pcap_drain(&(tp->t_inpkts));
1897 	tcp_pcap_drain(&(tp->t_outpkts));
1898 #endif
1899 
1900 	/* Allow the CC algorithm to clean up after itself. */
1901 	if (CC_ALGO(tp)->cb_destroy != NULL)
1902 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1903 	CC_DATA(tp) = NULL;
1904 
1905 #ifdef TCP_HHOOK
1906 	khelp_destroy_osd(tp->osd);
1907 #endif
1908 
1909 	CC_ALGO(tp) = NULL;
1910 	inp->inp_ppcb = NULL;
1911 	if (tp->t_timers->tt_draincnt == 0) {
1912 		/* We own the last reference on tcpcb, let's free it. */
1913 #ifdef TCP_BLACKBOX
1914 		tcp_log_tcpcbfini(tp);
1915 #endif
1916 		TCPSTATES_DEC(tp->t_state);
1917 		if (tp->t_fb->tfb_tcp_fb_fini)
1918 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1919 		refcount_release(&tp->t_fb->tfb_refcnt);
1920 		tp->t_inpcb = NULL;
1921 		uma_zfree(V_tcpcb_zone, tp);
1922 		released = in_pcbrele_wlocked(inp);
1923 		KASSERT(!released, ("%s: inp %p should not have been released "
1924 			"here", __func__, inp));
1925 	}
1926 }
1927 
1928 void
1929 tcp_timer_discard(void *ptp)
1930 {
1931 	struct inpcb *inp;
1932 	struct tcpcb *tp;
1933 	struct epoch_tracker et;
1934 
1935 	tp = (struct tcpcb *)ptp;
1936 	CURVNET_SET(tp->t_vnet);
1937 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1938 	inp = tp->t_inpcb;
1939 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1940 		__func__, tp));
1941 	INP_WLOCK(inp);
1942 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1943 		("%s: tcpcb has to be stopped here", __func__));
1944 	tp->t_timers->tt_draincnt--;
1945 	if (tp->t_timers->tt_draincnt == 0) {
1946 		/* We own the last reference on this tcpcb, let's free it. */
1947 #ifdef TCP_BLACKBOX
1948 		tcp_log_tcpcbfini(tp);
1949 #endif
1950 		TCPSTATES_DEC(tp->t_state);
1951 		if (tp->t_fb->tfb_tcp_fb_fini)
1952 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1953 		refcount_release(&tp->t_fb->tfb_refcnt);
1954 		tp->t_inpcb = NULL;
1955 		uma_zfree(V_tcpcb_zone, tp);
1956 		if (in_pcbrele_wlocked(inp)) {
1957 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1958 			CURVNET_RESTORE();
1959 			return;
1960 		}
1961 	}
1962 	INP_WUNLOCK(inp);
1963 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1964 	CURVNET_RESTORE();
1965 }
1966 
1967 /*
1968  * Attempt to close a TCP control block, marking it as dropped, and freeing
1969  * the socket if we hold the only reference.
1970  */
1971 struct tcpcb *
1972 tcp_close(struct tcpcb *tp)
1973 {
1974 	struct inpcb *inp = tp->t_inpcb;
1975 	struct socket *so;
1976 
1977 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1978 	INP_WLOCK_ASSERT(inp);
1979 
1980 #ifdef TCP_OFFLOAD
1981 	if (tp->t_state == TCPS_LISTEN)
1982 		tcp_offload_listen_stop(tp);
1983 #endif
1984 	/*
1985 	 * This releases the TFO pending counter resource for TFO listen
1986 	 * sockets as well as passively-created TFO sockets that transition
1987 	 * from SYN_RECEIVED to CLOSED.
1988 	 */
1989 	if (tp->t_tfo_pending) {
1990 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1991 		tp->t_tfo_pending = NULL;
1992 	}
1993 	in_pcbdrop(inp);
1994 	TCPSTAT_INC(tcps_closed);
1995 	if (tp->t_state != TCPS_CLOSED)
1996 		tcp_state_change(tp, TCPS_CLOSED);
1997 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1998 	so = inp->inp_socket;
1999 	soisdisconnected(so);
2000 	if (inp->inp_flags & INP_SOCKREF) {
2001 		KASSERT(so->so_state & SS_PROTOREF,
2002 		    ("tcp_close: !SS_PROTOREF"));
2003 		inp->inp_flags &= ~INP_SOCKREF;
2004 		INP_WUNLOCK(inp);
2005 		SOCK_LOCK(so);
2006 		so->so_state &= ~SS_PROTOREF;
2007 		sofree(so);
2008 		return (NULL);
2009 	}
2010 	return (tp);
2011 }
2012 
2013 void
2014 tcp_drain(void)
2015 {
2016 	VNET_ITERATOR_DECL(vnet_iter);
2017 
2018 	if (!do_tcpdrain)
2019 		return;
2020 
2021 	VNET_LIST_RLOCK_NOSLEEP();
2022 	VNET_FOREACH(vnet_iter) {
2023 		CURVNET_SET(vnet_iter);
2024 		struct inpcb *inpb;
2025 		struct tcpcb *tcpb;
2026 
2027 	/*
2028 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2029 	 * if there is one...
2030 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2031 	 *      reassembly queue should be flushed, but in a situation
2032 	 *	where we're really low on mbufs, this is potentially
2033 	 *	useful.
2034 	 */
2035 		INP_INFO_WLOCK(&V_tcbinfo);
2036 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2037 			INP_WLOCK(inpb);
2038 			if (inpb->inp_flags & INP_TIMEWAIT) {
2039 				INP_WUNLOCK(inpb);
2040 				continue;
2041 			}
2042 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2043 				tcp_reass_flush(tcpb);
2044 				tcp_clean_sackreport(tcpb);
2045 #ifdef TCP_BLACKBOX
2046 				tcp_log_drain(tcpb);
2047 #endif
2048 #ifdef TCPPCAP
2049 				if (tcp_pcap_aggressive_free) {
2050 					/* Free the TCP PCAP queues. */
2051 					tcp_pcap_drain(&(tcpb->t_inpkts));
2052 					tcp_pcap_drain(&(tcpb->t_outpkts));
2053 				}
2054 #endif
2055 			}
2056 			INP_WUNLOCK(inpb);
2057 		}
2058 		INP_INFO_WUNLOCK(&V_tcbinfo);
2059 		CURVNET_RESTORE();
2060 	}
2061 	VNET_LIST_RUNLOCK_NOSLEEP();
2062 }
2063 
2064 /*
2065  * Notify a tcp user of an asynchronous error;
2066  * store error as soft error, but wake up user
2067  * (for now, won't do anything until can select for soft error).
2068  *
2069  * Do not wake up user since there currently is no mechanism for
2070  * reporting soft errors (yet - a kqueue filter may be added).
2071  */
2072 static struct inpcb *
2073 tcp_notify(struct inpcb *inp, int error)
2074 {
2075 	struct tcpcb *tp;
2076 
2077 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2078 	INP_WLOCK_ASSERT(inp);
2079 
2080 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2081 	    (inp->inp_flags & INP_DROPPED))
2082 		return (inp);
2083 
2084 	tp = intotcpcb(inp);
2085 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2086 
2087 	/*
2088 	 * Ignore some errors if we are hooked up.
2089 	 * If connection hasn't completed, has retransmitted several times,
2090 	 * and receives a second error, give up now.  This is better
2091 	 * than waiting a long time to establish a connection that
2092 	 * can never complete.
2093 	 */
2094 	if (tp->t_state == TCPS_ESTABLISHED &&
2095 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2096 	     error == EHOSTDOWN)) {
2097 		if (inp->inp_route.ro_rt) {
2098 			RTFREE(inp->inp_route.ro_rt);
2099 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2100 		}
2101 		return (inp);
2102 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2103 	    tp->t_softerror) {
2104 		tp = tcp_drop(tp, error);
2105 		if (tp != NULL)
2106 			return (inp);
2107 		else
2108 			return (NULL);
2109 	} else {
2110 		tp->t_softerror = error;
2111 		return (inp);
2112 	}
2113 #if 0
2114 	wakeup( &so->so_timeo);
2115 	sorwakeup(so);
2116 	sowwakeup(so);
2117 #endif
2118 }
2119 
2120 static int
2121 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2122 {
2123 	int error, i, m, n, pcb_count;
2124 	struct inpcb *inp, **inp_list;
2125 	inp_gen_t gencnt;
2126 	struct xinpgen xig;
2127 	struct epoch_tracker et;
2128 
2129 	/*
2130 	 * The process of preparing the TCB list is too time-consuming and
2131 	 * resource-intensive to repeat twice on every request.
2132 	 */
2133 	if (req->oldptr == NULL) {
2134 		n = V_tcbinfo.ipi_count +
2135 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2136 		n += imax(n / 8, 10);
2137 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2138 		return (0);
2139 	}
2140 
2141 	if (req->newptr != NULL)
2142 		return (EPERM);
2143 
2144 	/*
2145 	 * OK, now we're committed to doing something.
2146 	 */
2147 	INP_LIST_RLOCK(&V_tcbinfo);
2148 	gencnt = V_tcbinfo.ipi_gencnt;
2149 	n = V_tcbinfo.ipi_count;
2150 	INP_LIST_RUNLOCK(&V_tcbinfo);
2151 
2152 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2153 
2154 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2155 		+ (n + m) * sizeof(struct xtcpcb));
2156 	if (error != 0)
2157 		return (error);
2158 
2159 	bzero(&xig, sizeof(xig));
2160 	xig.xig_len = sizeof xig;
2161 	xig.xig_count = n + m;
2162 	xig.xig_gen = gencnt;
2163 	xig.xig_sogen = so_gencnt;
2164 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2165 	if (error)
2166 		return (error);
2167 
2168 	error = syncache_pcblist(req, m, &pcb_count);
2169 	if (error)
2170 		return (error);
2171 
2172 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2173 
2174 	INP_INFO_WLOCK(&V_tcbinfo);
2175 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2176 	    inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2177 		INP_WLOCK(inp);
2178 		if (inp->inp_gencnt <= gencnt) {
2179 			/*
2180 			 * XXX: This use of cr_cansee(), introduced with
2181 			 * TCP state changes, is not quite right, but for
2182 			 * now, better than nothing.
2183 			 */
2184 			if (inp->inp_flags & INP_TIMEWAIT) {
2185 				if (intotw(inp) != NULL)
2186 					error = cr_cansee(req->td->td_ucred,
2187 					    intotw(inp)->tw_cred);
2188 				else
2189 					error = EINVAL;	/* Skip this inp. */
2190 			} else
2191 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2192 			if (error == 0) {
2193 				in_pcbref(inp);
2194 				inp_list[i++] = inp;
2195 			}
2196 		}
2197 		INP_WUNLOCK(inp);
2198 	}
2199 	INP_INFO_WUNLOCK(&V_tcbinfo);
2200 	n = i;
2201 
2202 	error = 0;
2203 	for (i = 0; i < n; i++) {
2204 		inp = inp_list[i];
2205 		INP_RLOCK(inp);
2206 		if (inp->inp_gencnt <= gencnt) {
2207 			struct xtcpcb xt;
2208 
2209 			tcp_inptoxtp(inp, &xt);
2210 			INP_RUNLOCK(inp);
2211 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2212 		} else
2213 			INP_RUNLOCK(inp);
2214 	}
2215 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2216 	for (i = 0; i < n; i++) {
2217 		inp = inp_list[i];
2218 		INP_RLOCK(inp);
2219 		if (!in_pcbrele_rlocked(inp))
2220 			INP_RUNLOCK(inp);
2221 	}
2222 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2223 
2224 	if (!error) {
2225 		/*
2226 		 * Give the user an updated idea of our state.
2227 		 * If the generation differs from what we told
2228 		 * her before, she knows that something happened
2229 		 * while we were processing this request, and it
2230 		 * might be necessary to retry.
2231 		 */
2232 		INP_LIST_RLOCK(&V_tcbinfo);
2233 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2234 		xig.xig_sogen = so_gencnt;
2235 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2236 		INP_LIST_RUNLOCK(&V_tcbinfo);
2237 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2238 	}
2239 	free(inp_list, M_TEMP);
2240 	return (error);
2241 }
2242 
2243 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2244     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2245     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2246 
2247 #ifdef INET
2248 static int
2249 tcp_getcred(SYSCTL_HANDLER_ARGS)
2250 {
2251 	struct xucred xuc;
2252 	struct sockaddr_in addrs[2];
2253 	struct inpcb *inp;
2254 	int error;
2255 
2256 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2257 	if (error)
2258 		return (error);
2259 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2260 	if (error)
2261 		return (error);
2262 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2263 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2264 	if (inp != NULL) {
2265 		if (inp->inp_socket == NULL)
2266 			error = ENOENT;
2267 		if (error == 0)
2268 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2269 		if (error == 0)
2270 			cru2x(inp->inp_cred, &xuc);
2271 		INP_RUNLOCK(inp);
2272 	} else
2273 		error = ENOENT;
2274 	if (error == 0)
2275 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2276 	return (error);
2277 }
2278 
2279 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2280     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2281     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2282 #endif /* INET */
2283 
2284 #ifdef INET6
2285 static int
2286 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2287 {
2288 	struct xucred xuc;
2289 	struct sockaddr_in6 addrs[2];
2290 	struct inpcb *inp;
2291 	int error;
2292 #ifdef INET
2293 	int mapped = 0;
2294 #endif
2295 
2296 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2297 	if (error)
2298 		return (error);
2299 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2300 	if (error)
2301 		return (error);
2302 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2303 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2304 		return (error);
2305 	}
2306 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2307 #ifdef INET
2308 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2309 			mapped = 1;
2310 		else
2311 #endif
2312 			return (EINVAL);
2313 	}
2314 
2315 #ifdef INET
2316 	if (mapped == 1)
2317 		inp = in_pcblookup(&V_tcbinfo,
2318 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2319 			addrs[1].sin6_port,
2320 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2321 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2322 	else
2323 #endif
2324 		inp = in6_pcblookup(&V_tcbinfo,
2325 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2326 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2327 			INPLOOKUP_RLOCKPCB, NULL);
2328 	if (inp != NULL) {
2329 		if (inp->inp_socket == NULL)
2330 			error = ENOENT;
2331 		if (error == 0)
2332 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2333 		if (error == 0)
2334 			cru2x(inp->inp_cred, &xuc);
2335 		INP_RUNLOCK(inp);
2336 	} else
2337 		error = ENOENT;
2338 	if (error == 0)
2339 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2340 	return (error);
2341 }
2342 
2343 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2344     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2345     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2346 #endif /* INET6 */
2347 
2348 
2349 #ifdef INET
2350 void
2351 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2352 {
2353 	struct ip *ip = vip;
2354 	struct tcphdr *th;
2355 	struct in_addr faddr;
2356 	struct inpcb *inp;
2357 	struct tcpcb *tp;
2358 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2359 	struct icmp *icp;
2360 	struct in_conninfo inc;
2361 	struct epoch_tracker et;
2362 	tcp_seq icmp_tcp_seq;
2363 	int mtu;
2364 
2365 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2366 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2367 		return;
2368 
2369 	if (cmd == PRC_MSGSIZE)
2370 		notify = tcp_mtudisc_notify;
2371 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2372 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2373 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2374 		notify = tcp_drop_syn_sent;
2375 
2376 	/*
2377 	 * Hostdead is ugly because it goes linearly through all PCBs.
2378 	 * XXX: We never get this from ICMP, otherwise it makes an
2379 	 * excellent DoS attack on machines with many connections.
2380 	 */
2381 	else if (cmd == PRC_HOSTDEAD)
2382 		ip = NULL;
2383 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2384 		return;
2385 
2386 	if (ip == NULL) {
2387 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2388 		return;
2389 	}
2390 
2391 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2392 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2393 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2394 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2395 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2396 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2397 		/* signal EHOSTDOWN, as it flushes the cached route */
2398 		inp = (*notify)(inp, EHOSTDOWN);
2399 		goto out;
2400 	}
2401 	icmp_tcp_seq = th->th_seq;
2402 	if (inp != NULL)  {
2403 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2404 		    !(inp->inp_flags & INP_DROPPED) &&
2405 		    !(inp->inp_socket == NULL)) {
2406 			tp = intotcpcb(inp);
2407 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2408 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2409 				if (cmd == PRC_MSGSIZE) {
2410 					/*
2411 					 * MTU discovery:
2412 					 * If we got a needfrag set the MTU
2413 					 * in the route to the suggested new
2414 					 * value (if given) and then notify.
2415 					 */
2416 					mtu = ntohs(icp->icmp_nextmtu);
2417 					/*
2418 					 * If no alternative MTU was
2419 					 * proposed, try the next smaller
2420 					 * one.
2421 					 */
2422 					if (!mtu)
2423 						mtu = ip_next_mtu(
2424 						    ntohs(ip->ip_len), 1);
2425 					if (mtu < V_tcp_minmss +
2426 					    sizeof(struct tcpiphdr))
2427 						mtu = V_tcp_minmss +
2428 						    sizeof(struct tcpiphdr);
2429 					/*
2430 					 * Only process the offered MTU if it
2431 					 * is smaller than the current one.
2432 					 */
2433 					if (mtu < tp->t_maxseg +
2434 					    sizeof(struct tcpiphdr)) {
2435 						bzero(&inc, sizeof(inc));
2436 						inc.inc_faddr = faddr;
2437 						inc.inc_fibnum =
2438 						    inp->inp_inc.inc_fibnum;
2439 						tcp_hc_updatemtu(&inc, mtu);
2440 						tcp_mtudisc(inp, mtu);
2441 					}
2442 				} else
2443 					inp = (*notify)(inp,
2444 					    inetctlerrmap[cmd]);
2445 			}
2446 		}
2447 	} else {
2448 		bzero(&inc, sizeof(inc));
2449 		inc.inc_fport = th->th_dport;
2450 		inc.inc_lport = th->th_sport;
2451 		inc.inc_faddr = faddr;
2452 		inc.inc_laddr = ip->ip_src;
2453 		syncache_unreach(&inc, icmp_tcp_seq);
2454 	}
2455 out:
2456 	if (inp != NULL)
2457 		INP_WUNLOCK(inp);
2458 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2459 }
2460 #endif /* INET */
2461 
2462 #ifdef INET6
2463 void
2464 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2465 {
2466 	struct in6_addr *dst;
2467 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2468 	struct ip6_hdr *ip6;
2469 	struct mbuf *m;
2470 	struct inpcb *inp;
2471 	struct tcpcb *tp;
2472 	struct icmp6_hdr *icmp6;
2473 	struct ip6ctlparam *ip6cp = NULL;
2474 	const struct sockaddr_in6 *sa6_src = NULL;
2475 	struct in_conninfo inc;
2476 	struct epoch_tracker et;
2477 	struct tcp_ports {
2478 		uint16_t th_sport;
2479 		uint16_t th_dport;
2480 	} t_ports;
2481 	tcp_seq icmp_tcp_seq;
2482 	unsigned int mtu;
2483 	unsigned int off;
2484 
2485 	if (sa->sa_family != AF_INET6 ||
2486 	    sa->sa_len != sizeof(struct sockaddr_in6))
2487 		return;
2488 
2489 	/* if the parameter is from icmp6, decode it. */
2490 	if (d != NULL) {
2491 		ip6cp = (struct ip6ctlparam *)d;
2492 		icmp6 = ip6cp->ip6c_icmp6;
2493 		m = ip6cp->ip6c_m;
2494 		ip6 = ip6cp->ip6c_ip6;
2495 		off = ip6cp->ip6c_off;
2496 		sa6_src = ip6cp->ip6c_src;
2497 		dst = ip6cp->ip6c_finaldst;
2498 	} else {
2499 		m = NULL;
2500 		ip6 = NULL;
2501 		off = 0;	/* fool gcc */
2502 		sa6_src = &sa6_any;
2503 		dst = NULL;
2504 	}
2505 
2506 	if (cmd == PRC_MSGSIZE)
2507 		notify = tcp_mtudisc_notify;
2508 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2509 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2510 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2511 		notify = tcp_drop_syn_sent;
2512 
2513 	/*
2514 	 * Hostdead is ugly because it goes linearly through all PCBs.
2515 	 * XXX: We never get this from ICMP, otherwise it makes an
2516 	 * excellent DoS attack on machines with many connections.
2517 	 */
2518 	else if (cmd == PRC_HOSTDEAD)
2519 		ip6 = NULL;
2520 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2521 		return;
2522 
2523 	if (ip6 == NULL) {
2524 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2525 			      (const struct sockaddr *)sa6_src,
2526 			      0, cmd, NULL, notify);
2527 		return;
2528 	}
2529 
2530 	/* Check if we can safely get the ports from the tcp hdr */
2531 	if (m == NULL ||
2532 	    (m->m_pkthdr.len <
2533 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2534 		return;
2535 	}
2536 	bzero(&t_ports, sizeof(struct tcp_ports));
2537 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2538 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2539 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2540 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2541 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2542 		/* signal EHOSTDOWN, as it flushes the cached route */
2543 		inp = (*notify)(inp, EHOSTDOWN);
2544 		goto out;
2545 	}
2546 	off += sizeof(struct tcp_ports);
2547 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2548 		goto out;
2549 	}
2550 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2551 	if (inp != NULL)  {
2552 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2553 		    !(inp->inp_flags & INP_DROPPED) &&
2554 		    !(inp->inp_socket == NULL)) {
2555 			tp = intotcpcb(inp);
2556 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2557 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2558 				if (cmd == PRC_MSGSIZE) {
2559 					/*
2560 					 * MTU discovery:
2561 					 * If we got a needfrag set the MTU
2562 					 * in the route to the suggested new
2563 					 * value (if given) and then notify.
2564 					 */
2565 					mtu = ntohl(icmp6->icmp6_mtu);
2566 					/*
2567 					 * If no alternative MTU was
2568 					 * proposed, or the proposed
2569 					 * MTU was too small, set to
2570 					 * the min.
2571 					 */
2572 					if (mtu < IPV6_MMTU)
2573 						mtu = IPV6_MMTU - 8;
2574 					bzero(&inc, sizeof(inc));
2575 					inc.inc_fibnum = M_GETFIB(m);
2576 					inc.inc_flags |= INC_ISIPV6;
2577 					inc.inc6_faddr = *dst;
2578 					if (in6_setscope(&inc.inc6_faddr,
2579 						m->m_pkthdr.rcvif, NULL))
2580 						goto out;
2581 					/*
2582 					 * Only process the offered MTU if it
2583 					 * is smaller than the current one.
2584 					 */
2585 					if (mtu < tp->t_maxseg +
2586 					    sizeof (struct tcphdr) +
2587 					    sizeof (struct ip6_hdr)) {
2588 						tcp_hc_updatemtu(&inc, mtu);
2589 						tcp_mtudisc(inp, mtu);
2590 						ICMP6STAT_INC(icp6s_pmtuchg);
2591 					}
2592 				} else
2593 					inp = (*notify)(inp,
2594 					    inet6ctlerrmap[cmd]);
2595 			}
2596 		}
2597 	} else {
2598 		bzero(&inc, sizeof(inc));
2599 		inc.inc_fibnum = M_GETFIB(m);
2600 		inc.inc_flags |= INC_ISIPV6;
2601 		inc.inc_fport = t_ports.th_dport;
2602 		inc.inc_lport = t_ports.th_sport;
2603 		inc.inc6_faddr = *dst;
2604 		inc.inc6_laddr = ip6->ip6_src;
2605 		syncache_unreach(&inc, icmp_tcp_seq);
2606 	}
2607 out:
2608 	if (inp != NULL)
2609 		INP_WUNLOCK(inp);
2610 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2611 }
2612 #endif /* INET6 */
2613 
2614 static uint32_t
2615 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2616 {
2617 	MD5_CTX ctx;
2618 	uint32_t hash[4];
2619 
2620 	MD5Init(&ctx);
2621 	MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2622 	MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2623 	switch (inc->inc_flags & INC_ISIPV6) {
2624 #ifdef INET
2625 	case 0:
2626 		MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2627 		MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2628 		break;
2629 #endif
2630 #ifdef INET6
2631 	case INC_ISIPV6:
2632 		MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2633 		MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2634 		break;
2635 #endif
2636 	}
2637 	MD5Update(&ctx, key, len);
2638 	MD5Final((unsigned char *)hash, &ctx);
2639 
2640 	return (hash[0]);
2641 }
2642 
2643 uint32_t
2644 tcp_new_ts_offset(struct in_conninfo *inc)
2645 {
2646 	struct in_conninfo inc_store, *local_inc;
2647 
2648 	if (!V_tcp_ts_offset_per_conn) {
2649 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2650 		inc_store.inc_lport = 0;
2651 		inc_store.inc_fport = 0;
2652 		local_inc = &inc_store;
2653 	} else {
2654 		local_inc = inc;
2655 	}
2656 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2657 	    sizeof(V_ts_offset_secret)));
2658 }
2659 
2660 /*
2661  * Following is where TCP initial sequence number generation occurs.
2662  *
2663  * There are two places where we must use initial sequence numbers:
2664  * 1.  In SYN-ACK packets.
2665  * 2.  In SYN packets.
2666  *
2667  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2668  * tcp_syncache.c for details.
2669  *
2670  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2671  * depends on this property.  In addition, these ISNs should be
2672  * unguessable so as to prevent connection hijacking.  To satisfy
2673  * the requirements of this situation, the algorithm outlined in
2674  * RFC 1948 is used, with only small modifications.
2675  *
2676  * Implementation details:
2677  *
2678  * Time is based off the system timer, and is corrected so that it
2679  * increases by one megabyte per second.  This allows for proper
2680  * recycling on high speed LANs while still leaving over an hour
2681  * before rollover.
2682  *
2683  * As reading the *exact* system time is too expensive to be done
2684  * whenever setting up a TCP connection, we increment the time
2685  * offset in two ways.  First, a small random positive increment
2686  * is added to isn_offset for each connection that is set up.
2687  * Second, the function tcp_isn_tick fires once per clock tick
2688  * and increments isn_offset as necessary so that sequence numbers
2689  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2690  * random positive increments serve only to ensure that the same
2691  * exact sequence number is never sent out twice (as could otherwise
2692  * happen when a port is recycled in less than the system tick
2693  * interval.)
2694  *
2695  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2696  * between seeding of isn_secret.  This is normally set to zero,
2697  * as reseeding should not be necessary.
2698  *
2699  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2700  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2701  * general, this means holding an exclusive (write) lock.
2702  */
2703 
2704 #define ISN_BYTES_PER_SECOND 1048576
2705 #define ISN_STATIC_INCREMENT 4096
2706 #define ISN_RANDOM_INCREMENT (4096 - 1)
2707 #define ISN_SECRET_LENGTH    32
2708 
2709 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2710 VNET_DEFINE_STATIC(int, isn_last);
2711 VNET_DEFINE_STATIC(int, isn_last_reseed);
2712 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2713 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2714 
2715 #define	V_isn_secret			VNET(isn_secret)
2716 #define	V_isn_last			VNET(isn_last)
2717 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2718 #define	V_isn_offset			VNET(isn_offset)
2719 #define	V_isn_offset_old		VNET(isn_offset_old)
2720 
2721 tcp_seq
2722 tcp_new_isn(struct in_conninfo *inc)
2723 {
2724 	tcp_seq new_isn;
2725 	u_int32_t projected_offset;
2726 
2727 	ISN_LOCK();
2728 	/* Seed if this is the first use, reseed if requested. */
2729 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2730 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2731 		< (u_int)ticks))) {
2732 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2733 		V_isn_last_reseed = ticks;
2734 	}
2735 
2736 	/* Compute the md5 hash and return the ISN. */
2737 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2738 	    sizeof(V_isn_secret));
2739 	V_isn_offset += ISN_STATIC_INCREMENT +
2740 		(arc4random() & ISN_RANDOM_INCREMENT);
2741 	if (ticks != V_isn_last) {
2742 		projected_offset = V_isn_offset_old +
2743 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2744 		if (SEQ_GT(projected_offset, V_isn_offset))
2745 			V_isn_offset = projected_offset;
2746 		V_isn_offset_old = V_isn_offset;
2747 		V_isn_last = ticks;
2748 	}
2749 	new_isn += V_isn_offset;
2750 	ISN_UNLOCK();
2751 	return (new_isn);
2752 }
2753 
2754 /*
2755  * When a specific ICMP unreachable message is received and the
2756  * connection state is SYN-SENT, drop the connection.  This behavior
2757  * is controlled by the icmp_may_rst sysctl.
2758  */
2759 struct inpcb *
2760 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2761 {
2762 	struct tcpcb *tp;
2763 
2764 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2765 	INP_WLOCK_ASSERT(inp);
2766 
2767 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2768 	    (inp->inp_flags & INP_DROPPED))
2769 		return (inp);
2770 
2771 	tp = intotcpcb(inp);
2772 	if (tp->t_state != TCPS_SYN_SENT)
2773 		return (inp);
2774 
2775 	if (IS_FASTOPEN(tp->t_flags))
2776 		tcp_fastopen_disable_path(tp);
2777 
2778 	tp = tcp_drop(tp, errno);
2779 	if (tp != NULL)
2780 		return (inp);
2781 	else
2782 		return (NULL);
2783 }
2784 
2785 /*
2786  * When `need fragmentation' ICMP is received, update our idea of the MSS
2787  * based on the new value. Also nudge TCP to send something, since we
2788  * know the packet we just sent was dropped.
2789  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2790  */
2791 static struct inpcb *
2792 tcp_mtudisc_notify(struct inpcb *inp, int error)
2793 {
2794 
2795 	tcp_mtudisc(inp, -1);
2796 	return (inp);
2797 }
2798 
2799 static void
2800 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2801 {
2802 	struct tcpcb *tp;
2803 	struct socket *so;
2804 
2805 	INP_WLOCK_ASSERT(inp);
2806 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2807 	    (inp->inp_flags & INP_DROPPED))
2808 		return;
2809 
2810 	tp = intotcpcb(inp);
2811 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2812 
2813 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2814 
2815 	so = inp->inp_socket;
2816 	SOCKBUF_LOCK(&so->so_snd);
2817 	/* If the mss is larger than the socket buffer, decrease the mss. */
2818 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2819 		tp->t_maxseg = so->so_snd.sb_hiwat;
2820 	SOCKBUF_UNLOCK(&so->so_snd);
2821 
2822 	TCPSTAT_INC(tcps_mturesent);
2823 	tp->t_rtttime = 0;
2824 	tp->snd_nxt = tp->snd_una;
2825 	tcp_free_sackholes(tp);
2826 	tp->snd_recover = tp->snd_max;
2827 	if (tp->t_flags & TF_SACK_PERMIT)
2828 		EXIT_FASTRECOVERY(tp->t_flags);
2829 	tp->t_fb->tfb_tcp_output(tp);
2830 }
2831 
2832 #ifdef INET
2833 /*
2834  * Look-up the routing entry to the peer of this inpcb.  If no route
2835  * is found and it cannot be allocated, then return 0.  This routine
2836  * is called by TCP routines that access the rmx structure and by
2837  * tcp_mss_update to get the peer/interface MTU.
2838  */
2839 uint32_t
2840 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2841 {
2842 	struct nhop4_extended nh4;
2843 	struct ifnet *ifp;
2844 	uint32_t maxmtu = 0;
2845 
2846 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2847 
2848 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2849 
2850 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2851 		    NHR_REF, 0, &nh4) != 0)
2852 			return (0);
2853 
2854 		ifp = nh4.nh_ifp;
2855 		maxmtu = nh4.nh_mtu;
2856 
2857 		/* Report additional interface capabilities. */
2858 		if (cap != NULL) {
2859 			if (ifp->if_capenable & IFCAP_TSO4 &&
2860 			    ifp->if_hwassist & CSUM_TSO) {
2861 				cap->ifcap |= CSUM_TSO;
2862 				cap->tsomax = ifp->if_hw_tsomax;
2863 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2864 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2865 			}
2866 		}
2867 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2868 	}
2869 	return (maxmtu);
2870 }
2871 #endif /* INET */
2872 
2873 #ifdef INET6
2874 uint32_t
2875 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2876 {
2877 	struct nhop6_extended nh6;
2878 	struct in6_addr dst6;
2879 	uint32_t scopeid;
2880 	struct ifnet *ifp;
2881 	uint32_t maxmtu = 0;
2882 
2883 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2884 
2885 	if (inc->inc_flags & INC_IPV6MINMTU)
2886 		return (IPV6_MMTU);
2887 
2888 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2889 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2890 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2891 		    0, &nh6) != 0)
2892 			return (0);
2893 
2894 		ifp = nh6.nh_ifp;
2895 		maxmtu = nh6.nh_mtu;
2896 
2897 		/* Report additional interface capabilities. */
2898 		if (cap != NULL) {
2899 			if (ifp->if_capenable & IFCAP_TSO6 &&
2900 			    ifp->if_hwassist & CSUM_TSO) {
2901 				cap->ifcap |= CSUM_TSO;
2902 				cap->tsomax = ifp->if_hw_tsomax;
2903 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2904 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2905 			}
2906 		}
2907 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2908 	}
2909 
2910 	return (maxmtu);
2911 }
2912 #endif /* INET6 */
2913 
2914 /*
2915  * Calculate effective SMSS per RFC5681 definition for a given TCP
2916  * connection at its current state, taking into account SACK and etc.
2917  */
2918 u_int
2919 tcp_maxseg(const struct tcpcb *tp)
2920 {
2921 	u_int optlen;
2922 
2923 	if (tp->t_flags & TF_NOOPT)
2924 		return (tp->t_maxseg);
2925 
2926 	/*
2927 	 * Here we have a simplified code from tcp_addoptions(),
2928 	 * without a proper loop, and having most of paddings hardcoded.
2929 	 * We might make mistakes with padding here in some edge cases,
2930 	 * but this is harmless, since result of tcp_maxseg() is used
2931 	 * only in cwnd and ssthresh estimations.
2932 	 */
2933 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2934 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2935 		if (tp->t_flags & TF_RCVD_TSTMP)
2936 			optlen = TCPOLEN_TSTAMP_APPA;
2937 		else
2938 			optlen = 0;
2939 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2940 		if (tp->t_flags & TF_SIGNATURE)
2941 			optlen += PAD(TCPOLEN_SIGNATURE);
2942 #endif
2943 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2944 			optlen += TCPOLEN_SACKHDR;
2945 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2946 			optlen = PAD(optlen);
2947 		}
2948 	} else {
2949 		if (tp->t_flags & TF_REQ_TSTMP)
2950 			optlen = TCPOLEN_TSTAMP_APPA;
2951 		else
2952 			optlen = PAD(TCPOLEN_MAXSEG);
2953 		if (tp->t_flags & TF_REQ_SCALE)
2954 			optlen += PAD(TCPOLEN_WINDOW);
2955 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2956 		if (tp->t_flags & TF_SIGNATURE)
2957 			optlen += PAD(TCPOLEN_SIGNATURE);
2958 #endif
2959 		if (tp->t_flags & TF_SACK_PERMIT)
2960 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2961 	}
2962 #undef PAD
2963 	optlen = min(optlen, TCP_MAXOLEN);
2964 	return (tp->t_maxseg - optlen);
2965 }
2966 
2967 static int
2968 sysctl_drop(SYSCTL_HANDLER_ARGS)
2969 {
2970 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2971 	struct sockaddr_storage addrs[2];
2972 	struct inpcb *inp;
2973 	struct tcpcb *tp;
2974 	struct tcptw *tw;
2975 	struct sockaddr_in *fin, *lin;
2976 	struct epoch_tracker et;
2977 #ifdef INET6
2978 	struct sockaddr_in6 *fin6, *lin6;
2979 #endif
2980 	int error;
2981 
2982 	inp = NULL;
2983 	fin = lin = NULL;
2984 #ifdef INET6
2985 	fin6 = lin6 = NULL;
2986 #endif
2987 	error = 0;
2988 
2989 	if (req->oldptr != NULL || req->oldlen != 0)
2990 		return (EINVAL);
2991 	if (req->newptr == NULL)
2992 		return (EPERM);
2993 	if (req->newlen < sizeof(addrs))
2994 		return (ENOMEM);
2995 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2996 	if (error)
2997 		return (error);
2998 
2999 	switch (addrs[0].ss_family) {
3000 #ifdef INET6
3001 	case AF_INET6:
3002 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3003 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3004 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3005 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3006 			return (EINVAL);
3007 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3008 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3009 				return (EINVAL);
3010 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3011 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3012 			fin = (struct sockaddr_in *)&addrs[0];
3013 			lin = (struct sockaddr_in *)&addrs[1];
3014 			break;
3015 		}
3016 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3017 		if (error)
3018 			return (error);
3019 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3020 		if (error)
3021 			return (error);
3022 		break;
3023 #endif
3024 #ifdef INET
3025 	case AF_INET:
3026 		fin = (struct sockaddr_in *)&addrs[0];
3027 		lin = (struct sockaddr_in *)&addrs[1];
3028 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3029 		    lin->sin_len != sizeof(struct sockaddr_in))
3030 			return (EINVAL);
3031 		break;
3032 #endif
3033 	default:
3034 		return (EINVAL);
3035 	}
3036 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3037 	switch (addrs[0].ss_family) {
3038 #ifdef INET6
3039 	case AF_INET6:
3040 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3041 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3042 		    INPLOOKUP_WLOCKPCB, NULL);
3043 		break;
3044 #endif
3045 #ifdef INET
3046 	case AF_INET:
3047 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3048 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3049 		break;
3050 #endif
3051 	}
3052 	if (inp != NULL) {
3053 		if (inp->inp_flags & INP_TIMEWAIT) {
3054 			/*
3055 			 * XXXRW: There currently exists a state where an
3056 			 * inpcb is present, but its timewait state has been
3057 			 * discarded.  For now, don't allow dropping of this
3058 			 * type of inpcb.
3059 			 */
3060 			tw = intotw(inp);
3061 			if (tw != NULL)
3062 				tcp_twclose(tw, 0);
3063 			else
3064 				INP_WUNLOCK(inp);
3065 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3066 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3067 			tp = intotcpcb(inp);
3068 			tp = tcp_drop(tp, ECONNABORTED);
3069 			if (tp != NULL)
3070 				INP_WUNLOCK(inp);
3071 		} else
3072 			INP_WUNLOCK(inp);
3073 	} else
3074 		error = ESRCH;
3075 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3076 	return (error);
3077 }
3078 
3079 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3080     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3081     0, sysctl_drop, "", "Drop TCP connection");
3082 
3083 #ifdef KERN_TLS
3084 static int
3085 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3086 {
3087 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3088 	struct sockaddr_storage addrs[2];
3089 	struct inpcb *inp;
3090 	struct sockaddr_in *fin, *lin;
3091 	struct epoch_tracker et;
3092 #ifdef INET6
3093 	struct sockaddr_in6 *fin6, *lin6;
3094 #endif
3095 	int error;
3096 
3097 	inp = NULL;
3098 	fin = lin = NULL;
3099 #ifdef INET6
3100 	fin6 = lin6 = NULL;
3101 #endif
3102 	error = 0;
3103 
3104 	if (req->oldptr != NULL || req->oldlen != 0)
3105 		return (EINVAL);
3106 	if (req->newptr == NULL)
3107 		return (EPERM);
3108 	if (req->newlen < sizeof(addrs))
3109 		return (ENOMEM);
3110 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3111 	if (error)
3112 		return (error);
3113 
3114 	switch (addrs[0].ss_family) {
3115 #ifdef INET6
3116 	case AF_INET6:
3117 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3118 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3119 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3120 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3121 			return (EINVAL);
3122 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3123 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3124 				return (EINVAL);
3125 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3126 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3127 			fin = (struct sockaddr_in *)&addrs[0];
3128 			lin = (struct sockaddr_in *)&addrs[1];
3129 			break;
3130 		}
3131 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3132 		if (error)
3133 			return (error);
3134 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3135 		if (error)
3136 			return (error);
3137 		break;
3138 #endif
3139 #ifdef INET
3140 	case AF_INET:
3141 		fin = (struct sockaddr_in *)&addrs[0];
3142 		lin = (struct sockaddr_in *)&addrs[1];
3143 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3144 		    lin->sin_len != sizeof(struct sockaddr_in))
3145 			return (EINVAL);
3146 		break;
3147 #endif
3148 	default:
3149 		return (EINVAL);
3150 	}
3151 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3152 	switch (addrs[0].ss_family) {
3153 #ifdef INET6
3154 	case AF_INET6:
3155 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3156 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3157 		    INPLOOKUP_WLOCKPCB, NULL);
3158 		break;
3159 #endif
3160 #ifdef INET
3161 	case AF_INET:
3162 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3163 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3164 		break;
3165 #endif
3166 	}
3167 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3168 	if (inp != NULL) {
3169 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3170 		    inp->inp_socket == NULL) {
3171 			error = ECONNRESET;
3172 			INP_WUNLOCK(inp);
3173 		} else {
3174 			struct socket *so;
3175 
3176 			so = inp->inp_socket;
3177 			soref(so);
3178 			error = ktls_set_tx_mode(so,
3179 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3180 			INP_WUNLOCK(inp);
3181 			SOCK_LOCK(so);
3182 			sorele(so);
3183 		}
3184 	} else
3185 		error = ESRCH;
3186 	return (error);
3187 }
3188 
3189 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3190     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3191     0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS");
3192 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3193     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3194     1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS");
3195 #endif
3196 
3197 /*
3198  * Generate a standardized TCP log line for use throughout the
3199  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3200  * allow use in the interrupt context.
3201  *
3202  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3203  * NB: The function may return NULL if memory allocation failed.
3204  *
3205  * Due to header inclusion and ordering limitations the struct ip
3206  * and ip6_hdr pointers have to be passed as void pointers.
3207  */
3208 char *
3209 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3210     const void *ip6hdr)
3211 {
3212 
3213 	/* Is logging enabled? */
3214 	if (tcp_log_in_vain == 0)
3215 		return (NULL);
3216 
3217 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3218 }
3219 
3220 char *
3221 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3222     const void *ip6hdr)
3223 {
3224 
3225 	/* Is logging enabled? */
3226 	if (tcp_log_debug == 0)
3227 		return (NULL);
3228 
3229 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3230 }
3231 
3232 static char *
3233 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3234     const void *ip6hdr)
3235 {
3236 	char *s, *sp;
3237 	size_t size;
3238 	struct ip *ip;
3239 #ifdef INET6
3240 	const struct ip6_hdr *ip6;
3241 
3242 	ip6 = (const struct ip6_hdr *)ip6hdr;
3243 #endif /* INET6 */
3244 	ip = (struct ip *)ip4hdr;
3245 
3246 	/*
3247 	 * The log line looks like this:
3248 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3249 	 */
3250 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3251 	    sizeof(PRINT_TH_FLAGS) + 1 +
3252 #ifdef INET6
3253 	    2 * INET6_ADDRSTRLEN;
3254 #else
3255 	    2 * INET_ADDRSTRLEN;
3256 #endif /* INET6 */
3257 
3258 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3259 	if (s == NULL)
3260 		return (NULL);
3261 
3262 	strcat(s, "TCP: [");
3263 	sp = s + strlen(s);
3264 
3265 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3266 		inet_ntoa_r(inc->inc_faddr, sp);
3267 		sp = s + strlen(s);
3268 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3269 		sp = s + strlen(s);
3270 		inet_ntoa_r(inc->inc_laddr, sp);
3271 		sp = s + strlen(s);
3272 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3273 #ifdef INET6
3274 	} else if (inc) {
3275 		ip6_sprintf(sp, &inc->inc6_faddr);
3276 		sp = s + strlen(s);
3277 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3278 		sp = s + strlen(s);
3279 		ip6_sprintf(sp, &inc->inc6_laddr);
3280 		sp = s + strlen(s);
3281 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3282 	} else if (ip6 && th) {
3283 		ip6_sprintf(sp, &ip6->ip6_src);
3284 		sp = s + strlen(s);
3285 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3286 		sp = s + strlen(s);
3287 		ip6_sprintf(sp, &ip6->ip6_dst);
3288 		sp = s + strlen(s);
3289 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3290 #endif /* INET6 */
3291 #ifdef INET
3292 	} else if (ip && th) {
3293 		inet_ntoa_r(ip->ip_src, sp);
3294 		sp = s + strlen(s);
3295 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3296 		sp = s + strlen(s);
3297 		inet_ntoa_r(ip->ip_dst, sp);
3298 		sp = s + strlen(s);
3299 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3300 #endif /* INET */
3301 	} else {
3302 		free(s, M_TCPLOG);
3303 		return (NULL);
3304 	}
3305 	sp = s + strlen(s);
3306 	if (th)
3307 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3308 	if (*(s + size - 1) != '\0')
3309 		panic("%s: string too long", __func__);
3310 	return (s);
3311 }
3312 
3313 /*
3314  * A subroutine which makes it easy to track TCP state changes with DTrace.
3315  * This function shouldn't be called for t_state initializations that don't
3316  * correspond to actual TCP state transitions.
3317  */
3318 void
3319 tcp_state_change(struct tcpcb *tp, int newstate)
3320 {
3321 #if defined(KDTRACE_HOOKS)
3322 	int pstate = tp->t_state;
3323 #endif
3324 
3325 	TCPSTATES_DEC(tp->t_state);
3326 	TCPSTATES_INC(newstate);
3327 	tp->t_state = newstate;
3328 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3329 }
3330 
3331 /*
3332  * Create an external-format (``xtcpcb'') structure using the information in
3333  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3334  * reduce the spew of irrelevant information over this interface, to isolate
3335  * user code from changes in the kernel structure, and potentially to provide
3336  * information-hiding if we decide that some of this information should be
3337  * hidden from users.
3338  */
3339 void
3340 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3341 {
3342 	struct tcpcb *tp = intotcpcb(inp);
3343 	sbintime_t now;
3344 
3345 	bzero(xt, sizeof(*xt));
3346 	if (inp->inp_flags & INP_TIMEWAIT) {
3347 		xt->t_state = TCPS_TIME_WAIT;
3348 	} else {
3349 		xt->t_state = tp->t_state;
3350 		xt->t_logstate = tp->t_logstate;
3351 		xt->t_flags = tp->t_flags;
3352 		xt->t_sndzerowin = tp->t_sndzerowin;
3353 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3354 		xt->t_rcvoopack = tp->t_rcvoopack;
3355 
3356 		now = getsbinuptime();
3357 #define	COPYTIMER(ttt)	do {						\
3358 		if (callout_active(&tp->t_timers->ttt))			\
3359 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3360 			    SBT_1MS;					\
3361 		else							\
3362 			xt->ttt = 0;					\
3363 } while (0)
3364 		COPYTIMER(tt_delack);
3365 		COPYTIMER(tt_rexmt);
3366 		COPYTIMER(tt_persist);
3367 		COPYTIMER(tt_keep);
3368 		COPYTIMER(tt_2msl);
3369 #undef COPYTIMER
3370 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3371 
3372 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3373 		    TCP_FUNCTION_NAME_LEN_MAX);
3374 #ifdef TCP_BLACKBOX
3375 		(void)tcp_log_get_id(tp, xt->xt_logid);
3376 #endif
3377 	}
3378 
3379 	xt->xt_len = sizeof(struct xtcpcb);
3380 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3381 	if (inp->inp_socket == NULL)
3382 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3383 }
3384