xref: /freebsd/sys/netinet/tcp_subr.c (revision 54c1a65736ec012b583ade1d53c477e182c574e4)
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/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/route/nhop.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
93 #ifdef INET6
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
101 #endif
102 
103 #include <netinet/tcp.h>
104 #ifdef INVARIANTS
105 #define TCPSTATES
106 #endif
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_seq.h>
109 #include <netinet/tcp_timer.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet/tcp_log_buf.h>
112 #include <netinet/tcp_syncache.h>
113 #include <netinet/tcp_hpts.h>
114 #include <netinet/cc/cc.h>
115 #ifdef INET6
116 #include <netinet6/tcp6_var.h>
117 #endif
118 #include <netinet/tcpip.h>
119 #include <netinet/tcp_fastopen.h>
120 #ifdef TCPPCAP
121 #include <netinet/tcp_pcap.h>
122 #endif
123 #ifdef TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126 #ifdef INET6
127 #include <netinet6/ip6protosw.h>
128 #endif
129 #ifdef TCP_OFFLOAD
130 #include <netinet/tcp_offload.h>
131 #endif
132 #include <netinet/udp.h>
133 #include <netinet/udp_var.h>
134 
135 #include <netipsec/ipsec_support.h>
136 
137 #include <machine/in_cksum.h>
138 #include <crypto/siphash/siphash.h>
139 
140 #include <security/mac/mac_framework.h>
141 
142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
143 #ifdef INET6
144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
145 #endif
146 
147 #ifdef NETFLIX_EXP_DETECTION
148 /*  Sack attack detection thresholds and such */
149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
150     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
151     "Sack Attack detection thresholds");
152 int32_t tcp_force_detection = 0;
153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
154     CTLFLAG_RW,
155     &tcp_force_detection, 0,
156     "Do we force detection even if the INP has it off?");
157 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
159     CTLFLAG_RW,
160     &tcp_sack_to_ack_thresh, 700,
161     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
162 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
164     CTLFLAG_RW,
165     &tcp_sack_to_move_thresh, 600,
166     "Percentage of sack moves we must see above (10.1 percent is 101)");
167 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
169     CTLFLAG_RW,
170     &tcp_restoral_thresh, 550,
171     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
172 int32_t tcp_sad_decay_val = 800;
173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
174     CTLFLAG_RW,
175     &tcp_sad_decay_val, 800,
176     "The decay percentage (10.1 percent equals 101 )");
177 int32_t tcp_map_minimum = 500;
178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
179     CTLFLAG_RW,
180     &tcp_map_minimum, 500,
181     "Number of Map enteries before we start detection");
182 int32_t tcp_attack_on_turns_on_logging = 0;
183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
184     CTLFLAG_RW,
185     &tcp_attack_on_turns_on_logging, 0,
186    "When we have a positive hit on attack, do we turn on logging?");
187 int32_t tcp_sad_pacing_interval = 2000;
188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
189     CTLFLAG_RW,
190     &tcp_sad_pacing_interval, 2000,
191     "What is the minimum pacing interval for a classified attacker?");
192 
193 int32_t tcp_sad_low_pps = 100;
194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
195     CTLFLAG_RW,
196     &tcp_sad_low_pps, 100,
197     "What is the input pps that below which we do not decay?");
198 #endif
199 uint32_t tcp_ack_war_time_window = 1000;
200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
201     CTLFLAG_RW,
202     &tcp_ack_war_time_window, 1000,
203    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
204 uint32_t tcp_ack_war_cnt = 5;
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
206     CTLFLAG_RW,
207     &tcp_ack_war_cnt, 5,
208    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
209 
210 struct rwlock tcp_function_lock;
211 
212 static int
213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
214 {
215 	int error, new;
216 
217 	new = V_tcp_mssdflt;
218 	error = sysctl_handle_int(oidp, &new, 0, req);
219 	if (error == 0 && req->newptr) {
220 		if (new < TCP_MINMSS)
221 			error = EINVAL;
222 		else
223 			V_tcp_mssdflt = new;
224 	}
225 	return (error);
226 }
227 
228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
229     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
230     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
231     "Default TCP Maximum Segment Size");
232 
233 #ifdef INET6
234 static int
235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
236 {
237 	int error, new;
238 
239 	new = V_tcp_v6mssdflt;
240 	error = sysctl_handle_int(oidp, &new, 0, req);
241 	if (error == 0 && req->newptr) {
242 		if (new < TCP_MINMSS)
243 			error = EINVAL;
244 		else
245 			V_tcp_v6mssdflt = new;
246 	}
247 	return (error);
248 }
249 
250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
251     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
252     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
253    "Default TCP Maximum Segment Size for IPv6");
254 #endif /* INET6 */
255 
256 /*
257  * Minimum MSS we accept and use. This prevents DoS attacks where
258  * we are forced to a ridiculous low MSS like 20 and send hundreds
259  * of packets instead of one. The effect scales with the available
260  * bandwidth and quickly saturates the CPU and network interface
261  * with packet generation and sending. Set to zero to disable MINMSS
262  * checking. This setting prevents us from sending too small packets.
263  */
264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
266      &VNET_NAME(tcp_minmss), 0,
267     "Minimum TCP Maximum Segment Size");
268 
269 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
271     &VNET_NAME(tcp_do_rfc1323), 0,
272     "Enable rfc1323 (high performance TCP) extensions");
273 
274 /*
275  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
276  * Some stacks negotiate TS, but never send them after connection setup. Some
277  * stacks negotiate TS, but don't send them when sending keep-alive segments.
278  * These include modern widely deployed TCP stacks.
279  * Therefore tolerating violations for now...
280  */
281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
283     &VNET_NAME(tcp_tolerate_missing_ts), 0,
284     "Tolerate missing TCP timestamps");
285 
286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
288     &VNET_NAME(tcp_ts_offset_per_conn), 0,
289     "Initialize TCP timestamps per connection instead of per host pair");
290 
291 /* How many connections are pacing */
292 static volatile uint32_t number_of_tcp_connections_pacing = 0;
293 static uint32_t shadow_num_connections = 0;
294 
295 static int tcp_pacing_limit = 10000;
296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
297     &tcp_pacing_limit, 1000,
298     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
299 
300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
301     &shadow_num_connections, 0, "Number of TCP connections being paced");
302 
303 static int	tcp_log_debug = 0;
304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
305     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
306 
307 static int	tcp_tcbhashsize;
308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
309     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
310 
311 static int	do_tcpdrain = 1;
312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
313     "Enable tcp_drain routine for extra help when low on mbufs");
314 
315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
316     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
317 
318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
319 #define	V_icmp_may_rst			VNET(icmp_may_rst)
320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
321     &VNET_NAME(icmp_may_rst), 0,
322     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
323 
324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
325 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
327     &VNET_NAME(tcp_isn_reseed_interval), 0,
328     "Seconds between reseeding of ISN secret");
329 
330 static int	tcp_soreceive_stream;
331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
332     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
333 
334 VNET_DEFINE(uma_zone_t, sack_hole_zone);
335 #define	V_sack_hole_zone		VNET(sack_hole_zone)
336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
337 static int
338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
339 {
340 	int error;
341 	uint32_t new;
342 
343 	new = V_tcp_map_entries_limit;
344 	error = sysctl_handle_int(oidp, &new, 0, req);
345 	if (error == 0 && req->newptr) {
346 		/* only allow "0" and value > minimum */
347 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
348 			error = EINVAL;
349 		else
350 			V_tcp_map_entries_limit = new;
351 	}
352 	return (error);
353 }
354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
355     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
356     &VNET_NAME(tcp_map_entries_limit), 0,
357     &sysctl_net_inet_tcp_map_limit_check, "IU",
358     "Total sendmap entries limit");
359 
360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
362      &VNET_NAME(tcp_map_split_limit), 0,
363     "Total sendmap split entries limit");
364 
365 #ifdef TCP_HHOOK
366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
367 #endif
368 
369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
371 #define	V_ts_offset_secret	VNET(ts_offset_secret)
372 
373 static int	tcp_default_fb_init(struct tcpcb *tp);
374 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
375 static int	tcp_default_handoff_ok(struct tcpcb *tp);
376 static struct inpcb *tcp_notify(struct inpcb *, int);
377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
378 static void tcp_mtudisc(struct inpcb *, int);
379 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
380 		    void *ip4hdr, const void *ip6hdr);
381 
382 static struct tcp_function_block tcp_def_funcblk = {
383 	.tfb_tcp_block_name = "freebsd",
384 	.tfb_tcp_output = tcp_output,
385 	.tfb_tcp_do_segment = tcp_do_segment,
386 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
387 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
388 	.tfb_tcp_fb_init = tcp_default_fb_init,
389 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
390 };
391 
392 static int tcp_fb_cnt = 0;
393 struct tcp_funchead t_functions;
394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
395 
396 void
397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
398 {
399 	TCPSTAT_INC(tcps_dsack_count);
400 	tp->t_dsack_pack++;
401 	if (tlp == 0) {
402 		if (SEQ_GT(end, start)) {
403 			tp->t_dsack_bytes += (end - start);
404 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
405 		} else {
406 			tp->t_dsack_tlp_bytes += (start - end);
407 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
408 		}
409 	} else {
410 		if (SEQ_GT(end, start)) {
411 			tp->t_dsack_bytes += (end - start);
412 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
413 		} else {
414 			tp->t_dsack_tlp_bytes += (start - end);
415 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
416 		}
417 	}
418 }
419 
420 static struct tcp_function_block *
421 find_tcp_functions_locked(struct tcp_function_set *fs)
422 {
423 	struct tcp_function *f;
424 	struct tcp_function_block *blk=NULL;
425 
426 	TAILQ_FOREACH(f, &t_functions, tf_next) {
427 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
428 			blk = f->tf_fb;
429 			break;
430 		}
431 	}
432 	return(blk);
433 }
434 
435 static struct tcp_function_block *
436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
437 {
438 	struct tcp_function_block *rblk=NULL;
439 	struct tcp_function *f;
440 
441 	TAILQ_FOREACH(f, &t_functions, tf_next) {
442 		if (f->tf_fb == blk) {
443 			rblk = blk;
444 			if (s) {
445 				*s = f;
446 			}
447 			break;
448 		}
449 	}
450 	return (rblk);
451 }
452 
453 struct tcp_function_block *
454 find_and_ref_tcp_functions(struct tcp_function_set *fs)
455 {
456 	struct tcp_function_block *blk;
457 
458 	rw_rlock(&tcp_function_lock);
459 	blk = find_tcp_functions_locked(fs);
460 	if (blk)
461 		refcount_acquire(&blk->tfb_refcnt);
462 	rw_runlock(&tcp_function_lock);
463 	return(blk);
464 }
465 
466 struct tcp_function_block *
467 find_and_ref_tcp_fb(struct tcp_function_block *blk)
468 {
469 	struct tcp_function_block *rblk;
470 
471 	rw_rlock(&tcp_function_lock);
472 	rblk = find_tcp_fb_locked(blk, NULL);
473 	if (rblk)
474 		refcount_acquire(&rblk->tfb_refcnt);
475 	rw_runlock(&tcp_function_lock);
476 	return(rblk);
477 }
478 
479 /* Find a matching alias for the given tcp_function_block. */
480 int
481 find_tcp_function_alias(struct tcp_function_block *blk,
482     struct tcp_function_set *fs)
483 {
484 	struct tcp_function *f;
485 	int found;
486 
487 	found = 0;
488 	rw_rlock(&tcp_function_lock);
489 	TAILQ_FOREACH(f, &t_functions, tf_next) {
490 		if ((f->tf_fb == blk) &&
491 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
492 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
493 			/* Matching function block with different name. */
494 			strncpy(fs->function_set_name, f->tf_name,
495 			    TCP_FUNCTION_NAME_LEN_MAX);
496 			found = 1;
497 			break;
498 		}
499 	}
500 	/* Null terminate the string appropriately. */
501 	if (found) {
502 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
503 	} else {
504 		fs->function_set_name[0] = '\0';
505 	}
506 	rw_runlock(&tcp_function_lock);
507 	return (found);
508 }
509 
510 static struct tcp_function_block *
511 find_and_ref_tcp_default_fb(void)
512 {
513 	struct tcp_function_block *rblk;
514 
515 	rw_rlock(&tcp_function_lock);
516 	rblk = tcp_func_set_ptr;
517 	refcount_acquire(&rblk->tfb_refcnt);
518 	rw_runlock(&tcp_function_lock);
519 	return (rblk);
520 }
521 
522 void
523 tcp_switch_back_to_default(struct tcpcb *tp)
524 {
525 	struct tcp_function_block *tfb;
526 
527 	KASSERT(tp->t_fb != &tcp_def_funcblk,
528 	    ("%s: called by the built-in default stack", __func__));
529 
530 	/*
531 	 * Release the old stack. This function will either find a new one
532 	 * or panic.
533 	 */
534 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
535 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
536 	refcount_release(&tp->t_fb->tfb_refcnt);
537 
538 	/*
539 	 * Now, we'll find a new function block to use.
540 	 * Start by trying the current user-selected
541 	 * default, unless this stack is the user-selected
542 	 * default.
543 	 */
544 	tfb = find_and_ref_tcp_default_fb();
545 	if (tfb == tp->t_fb) {
546 		refcount_release(&tfb->tfb_refcnt);
547 		tfb = NULL;
548 	}
549 	/* Does the stack accept this connection? */
550 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 		refcount_release(&tfb->tfb_refcnt);
553 		tfb = NULL;
554 	}
555 	/* Try to use that stack. */
556 	if (tfb != NULL) {
557 		/* Initialize the new stack. If it succeeds, we are done. */
558 		tp->t_fb = tfb;
559 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
560 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
561 			return;
562 
563 		/*
564 		 * Initialization failed. Release the reference count on
565 		 * the stack.
566 		 */
567 		refcount_release(&tfb->tfb_refcnt);
568 	}
569 
570 	/*
571 	 * If that wasn't feasible, use the built-in default
572 	 * stack which is not allowed to reject anyone.
573 	 */
574 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
575 	if (tfb == NULL) {
576 		/* there always should be a default */
577 		panic("Can't refer to tcp_def_funcblk");
578 	}
579 	if (tfb->tfb_tcp_handoff_ok != NULL) {
580 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
581 			/* The default stack cannot say no */
582 			panic("Default stack rejects a new session?");
583 		}
584 	}
585 	tp->t_fb = tfb;
586 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
587 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
588 		/* The default stack cannot fail */
589 		panic("Default stack initialization failed");
590 	}
591 }
592 
593 static void
594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
595     const struct sockaddr *sa, void *ctx)
596 {
597 	struct ip *iph;
598 #ifdef INET6
599 	struct ip6_hdr *ip6;
600 #endif
601 	struct udphdr *uh;
602 	struct tcphdr *th;
603 	int thlen;
604 	uint16_t port;
605 
606 	TCPSTAT_INC(tcps_tunneled_pkts);
607 	if ((m->m_flags & M_PKTHDR) == 0) {
608 		/* Can't handle one that is not a pkt hdr */
609 		TCPSTAT_INC(tcps_tunneled_errs);
610 		goto out;
611 	}
612 	thlen = sizeof(struct tcphdr);
613 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
614 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
615 		TCPSTAT_INC(tcps_tunneled_errs);
616 		goto out;
617 	}
618 	iph = mtod(m, struct ip *);
619 	uh = (struct udphdr *)((caddr_t)iph + off);
620 	th = (struct tcphdr *)(uh + 1);
621 	thlen = th->th_off << 2;
622 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
623 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
624 		if (m == NULL) {
625 			TCPSTAT_INC(tcps_tunneled_errs);
626 			goto out;
627 		} else {
628 			iph = mtod(m, struct ip *);
629 			uh = (struct udphdr *)((caddr_t)iph + off);
630 			th = (struct tcphdr *)(uh + 1);
631 		}
632 	}
633 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
634 	bcopy(th, uh, m->m_len - off);
635 	m->m_len -= sizeof(struct udphdr);
636 	m->m_pkthdr.len -= sizeof(struct udphdr);
637 	/*
638 	 * We use the same algorithm for
639 	 * both UDP and TCP for c-sum. So
640 	 * the code in tcp_input will skip
641 	 * the checksum. So we do nothing
642 	 * with the flag (m->m_pkthdr.csum_flags).
643 	 */
644 	switch (iph->ip_v) {
645 #ifdef INET
646 	case IPVERSION:
647 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
648 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
649 		break;
650 #endif
651 #ifdef INET6
652 	case IPV6_VERSION >> 4:
653 		ip6 = mtod(m, struct ip6_hdr *);
654 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
655 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
656 		break;
657 #endif
658 	default:
659 		goto out;
660 		break;
661 	}
662 	return;
663 out:
664 	m_freem(m);
665 }
666 
667 static int
668 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
669 {
670 	int error=ENOENT;
671 	struct tcp_function_set fs;
672 	struct tcp_function_block *blk;
673 
674 	memset(&fs, 0, sizeof(fs));
675 	rw_rlock(&tcp_function_lock);
676 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
677 	if (blk) {
678 		/* Found him */
679 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
680 		fs.pcbcnt = blk->tfb_refcnt;
681 	}
682 	rw_runlock(&tcp_function_lock);
683 	error = sysctl_handle_string(oidp, fs.function_set_name,
684 				     sizeof(fs.function_set_name), req);
685 
686 	/* Check for error or no change */
687 	if (error != 0 || req->newptr == NULL)
688 		return(error);
689 
690 	rw_wlock(&tcp_function_lock);
691 	blk = find_tcp_functions_locked(&fs);
692 	if ((blk == NULL) ||
693 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
694 		error = ENOENT;
695 		goto done;
696 	}
697 	tcp_func_set_ptr = blk;
698 done:
699 	rw_wunlock(&tcp_function_lock);
700 	return (error);
701 }
702 
703 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
704     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
705     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
706     "Set/get the default TCP functions");
707 
708 static int
709 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
710 {
711 	int error, cnt, linesz;
712 	struct tcp_function *f;
713 	char *buffer, *cp;
714 	size_t bufsz, outsz;
715 	bool alias;
716 
717 	cnt = 0;
718 	rw_rlock(&tcp_function_lock);
719 	TAILQ_FOREACH(f, &t_functions, tf_next) {
720 		cnt++;
721 	}
722 	rw_runlock(&tcp_function_lock);
723 
724 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
725 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
726 
727 	error = 0;
728 	cp = buffer;
729 
730 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
731 	    "Alias", "PCB count");
732 	cp += linesz;
733 	bufsz -= linesz;
734 	outsz = linesz;
735 
736 	rw_rlock(&tcp_function_lock);
737 	TAILQ_FOREACH(f, &t_functions, tf_next) {
738 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
739 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
740 		    f->tf_fb->tfb_tcp_block_name,
741 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
742 		    alias ? f->tf_name : "-",
743 		    f->tf_fb->tfb_refcnt);
744 		if (linesz >= bufsz) {
745 			error = EOVERFLOW;
746 			break;
747 		}
748 		cp += linesz;
749 		bufsz -= linesz;
750 		outsz += linesz;
751 	}
752 	rw_runlock(&tcp_function_lock);
753 	if (error == 0)
754 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
755 	free(buffer, M_TEMP);
756 	return (error);
757 }
758 
759 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
760     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
761     NULL, 0, sysctl_net_inet_list_available, "A",
762     "list available TCP Function sets");
763 
764 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
765 
766 #ifdef INET
767 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
768 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
769 #endif
770 #ifdef INET6
771 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
772 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
773 #endif
774 
775 static void
776 tcp_over_udp_stop(void)
777 {
778 	/*
779 	 * This function assumes sysctl caller holds inp_rinfo_lock()
780 	 * for writing!
781 	 */
782 #ifdef INET
783 	if (V_udp4_tun_socket != NULL) {
784 		soclose(V_udp4_tun_socket);
785 		V_udp4_tun_socket = NULL;
786 	}
787 #endif
788 #ifdef INET6
789 	if (V_udp6_tun_socket != NULL) {
790 		soclose(V_udp6_tun_socket);
791 		V_udp6_tun_socket = NULL;
792 	}
793 #endif
794 }
795 
796 static int
797 tcp_over_udp_start(void)
798 {
799 	uint16_t port;
800 	int ret;
801 #ifdef INET
802 	struct sockaddr_in sin;
803 #endif
804 #ifdef INET6
805 	struct sockaddr_in6 sin6;
806 #endif
807 	/*
808 	 * This function assumes sysctl caller holds inp_info_rlock()
809 	 * for writing!
810 	 */
811 	port = V_tcp_udp_tunneling_port;
812 	if (ntohs(port) == 0) {
813 		/* Must have a port set */
814 		return (EINVAL);
815 	}
816 #ifdef INET
817 	if (V_udp4_tun_socket != NULL) {
818 		/* Already running -- must stop first */
819 		return (EALREADY);
820 	}
821 #endif
822 #ifdef INET6
823 	if (V_udp6_tun_socket != NULL) {
824 		/* Already running -- must stop first */
825 		return (EALREADY);
826 	}
827 #endif
828 #ifdef INET
829 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
830 	    SOCK_DGRAM, IPPROTO_UDP,
831 	    curthread->td_ucred, curthread))) {
832 		tcp_over_udp_stop();
833 		return (ret);
834 	}
835 	/* Call the special UDP hook. */
836 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
837 	    tcp_recv_udp_tunneled_packet,
838 	    tcp_ctlinput_viaudp,
839 	    NULL))) {
840 		tcp_over_udp_stop();
841 		return (ret);
842 	}
843 	/* Ok, we have a socket, bind it to the port. */
844 	memset(&sin, 0, sizeof(struct sockaddr_in));
845 	sin.sin_len = sizeof(struct sockaddr_in);
846 	sin.sin_family = AF_INET;
847 	sin.sin_port = htons(port);
848 	if ((ret = sobind(V_udp4_tun_socket,
849 	    (struct sockaddr *)&sin, curthread))) {
850 		tcp_over_udp_stop();
851 		return (ret);
852 	}
853 #endif
854 #ifdef INET6
855 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
856 	    SOCK_DGRAM, IPPROTO_UDP,
857 	    curthread->td_ucred, curthread))) {
858 		tcp_over_udp_stop();
859 		return (ret);
860 	}
861 	/* Call the special UDP hook. */
862 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
863 	    tcp_recv_udp_tunneled_packet,
864 	    tcp6_ctlinput_viaudp,
865 	    NULL))) {
866 		tcp_over_udp_stop();
867 		return (ret);
868 	}
869 	/* Ok, we have a socket, bind it to the port. */
870 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
871 	sin6.sin6_len = sizeof(struct sockaddr_in6);
872 	sin6.sin6_family = AF_INET6;
873 	sin6.sin6_port = htons(port);
874 	if ((ret = sobind(V_udp6_tun_socket,
875 	    (struct sockaddr *)&sin6, curthread))) {
876 		tcp_over_udp_stop();
877 		return (ret);
878 	}
879 #endif
880 	return (0);
881 }
882 
883 static int
884 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
885 {
886 	int error;
887 	uint32_t old, new;
888 
889 	old = V_tcp_udp_tunneling_port;
890 	new = old;
891 	error = sysctl_handle_int(oidp, &new, 0, req);
892 	if ((error == 0) &&
893 	    (req->newptr != NULL)) {
894 		if ((new < TCP_TUNNELING_PORT_MIN) ||
895 		    (new > TCP_TUNNELING_PORT_MAX)) {
896 			error = EINVAL;
897 		} else {
898 			V_tcp_udp_tunneling_port = new;
899 			if (old != 0) {
900 				tcp_over_udp_stop();
901 			}
902 			if (new != 0) {
903 				error = tcp_over_udp_start();
904 			}
905 		}
906 	}
907 	return (error);
908 }
909 
910 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
911     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
912     &VNET_NAME(tcp_udp_tunneling_port),
913     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
914     "Tunneling port for tcp over udp");
915 
916 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
917 
918 static int
919 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
920 {
921 	int error, new;
922 
923 	new = V_tcp_udp_tunneling_overhead;
924 	error = sysctl_handle_int(oidp, &new, 0, req);
925 	if (error == 0 && req->newptr) {
926 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
927 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
928 			error = EINVAL;
929 		else
930 			V_tcp_udp_tunneling_overhead = new;
931 	}
932 	return (error);
933 }
934 
935 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
936     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
937     &VNET_NAME(tcp_udp_tunneling_overhead),
938     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
939     "MSS reduction when using tcp over udp");
940 
941 /*
942  * Exports one (struct tcp_function_info) for each alias/name.
943  */
944 static int
945 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
946 {
947 	int cnt, error;
948 	struct tcp_function *f;
949 	struct tcp_function_info tfi;
950 
951 	/*
952 	 * We don't allow writes.
953 	 */
954 	if (req->newptr != NULL)
955 		return (EINVAL);
956 
957 	/*
958 	 * Wire the old buffer so we can directly copy the functions to
959 	 * user space without dropping the lock.
960 	 */
961 	if (req->oldptr != NULL) {
962 		error = sysctl_wire_old_buffer(req, 0);
963 		if (error)
964 			return (error);
965 	}
966 
967 	/*
968 	 * Walk the list and copy out matching entries. If INVARIANTS
969 	 * is compiled in, also walk the list to verify the length of
970 	 * the list matches what we have recorded.
971 	 */
972 	rw_rlock(&tcp_function_lock);
973 
974 	cnt = 0;
975 #ifndef INVARIANTS
976 	if (req->oldptr == NULL) {
977 		cnt = tcp_fb_cnt;
978 		goto skip_loop;
979 	}
980 #endif
981 	TAILQ_FOREACH(f, &t_functions, tf_next) {
982 #ifdef INVARIANTS
983 		cnt++;
984 #endif
985 		if (req->oldptr != NULL) {
986 			bzero(&tfi, sizeof(tfi));
987 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
988 			tfi.tfi_id = f->tf_fb->tfb_id;
989 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
990 			    sizeof(tfi.tfi_alias));
991 			(void)strlcpy(tfi.tfi_name,
992 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
993 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
994 			/*
995 			 * Don't stop on error, as that is the
996 			 * mechanism we use to accumulate length
997 			 * information if the buffer was too short.
998 			 */
999 		}
1000 	}
1001 	KASSERT(cnt == tcp_fb_cnt,
1002 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1003 #ifndef INVARIANTS
1004 skip_loop:
1005 #endif
1006 	rw_runlock(&tcp_function_lock);
1007 	if (req->oldptr == NULL)
1008 		error = SYSCTL_OUT(req, NULL,
1009 		    (cnt + 1) * sizeof(struct tcp_function_info));
1010 
1011 	return (error);
1012 }
1013 
1014 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1015 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1016 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1017 	    "List TCP function block name-to-ID mappings");
1018 
1019 /*
1020  * tfb_tcp_handoff_ok() function for the default stack.
1021  * Note that we'll basically try to take all comers.
1022  */
1023 static int
1024 tcp_default_handoff_ok(struct tcpcb *tp)
1025 {
1026 
1027 	return (0);
1028 }
1029 
1030 /*
1031  * tfb_tcp_fb_init() function for the default stack.
1032  *
1033  * This handles making sure we have appropriate timers set if you are
1034  * transitioning a socket that has some amount of setup done.
1035  *
1036  * The init() fuction from the default can *never* return non-zero i.e.
1037  * it is required to always succeed since it is the stack of last resort!
1038  */
1039 static int
1040 tcp_default_fb_init(struct tcpcb *tp)
1041 {
1042 
1043 	struct socket *so;
1044 
1045 	INP_WLOCK_ASSERT(tp->t_inpcb);
1046 
1047 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1048 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1049 	    tp->t_state));
1050 
1051 	/*
1052 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1053 	 * know what to do for unexpected states (which includes TIME_WAIT).
1054 	 */
1055 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1056 		return (0);
1057 
1058 	/*
1059 	 * Make sure some kind of transmission timer is set if there is
1060 	 * outstanding data.
1061 	 */
1062 	so = tp->t_inpcb->inp_socket;
1063 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1064 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1065 	    tcp_timer_active(tp, TT_PERSIST))) {
1066 		/*
1067 		 * If the session has established and it looks like it should
1068 		 * be in the persist state, set the persist timer. Otherwise,
1069 		 * set the retransmit timer.
1070 		 */
1071 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1072 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1073 		    (int32_t)sbavail(&so->so_snd))
1074 			tcp_setpersist(tp);
1075 		else
1076 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1077 	}
1078 
1079 	/* All non-embryonic sessions get a keepalive timer. */
1080 	if (!tcp_timer_active(tp, TT_KEEP))
1081 		tcp_timer_activate(tp, TT_KEEP,
1082 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1083 		    TP_KEEPINIT(tp));
1084 
1085 	/*
1086 	 * Make sure critical variables are initialized
1087 	 * if transitioning while in Recovery.
1088 	 */
1089 	if IN_FASTRECOVERY(tp->t_flags) {
1090 		if (tp->sackhint.recover_fs == 0)
1091 			tp->sackhint.recover_fs = max(1,
1092 			    tp->snd_nxt - tp->snd_una);
1093 	}
1094 
1095 	return (0);
1096 }
1097 
1098 /*
1099  * tfb_tcp_fb_fini() function for the default stack.
1100  *
1101  * This changes state as necessary (or prudent) to prepare for another stack
1102  * to assume responsibility for the connection.
1103  */
1104 static void
1105 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1106 {
1107 
1108 	INP_WLOCK_ASSERT(tp->t_inpcb);
1109 	return;
1110 }
1111 
1112 /*
1113  * Target size of TCP PCB hash tables. Must be a power of two.
1114  *
1115  * Note that this can be overridden by the kernel environment
1116  * variable net.inet.tcp.tcbhashsize
1117  */
1118 #ifndef TCBHASHSIZE
1119 #define TCBHASHSIZE	0
1120 #endif
1121 
1122 /*
1123  * XXX
1124  * Callouts should be moved into struct tcp directly.  They are currently
1125  * separate because the tcpcb structure is exported to userland for sysctl
1126  * parsing purposes, which do not know about callouts.
1127  */
1128 struct tcpcb_mem {
1129 	struct	tcpcb		tcb;
1130 	struct	tcp_timer	tt;
1131 	struct	cc_var		ccv;
1132 #ifdef TCP_HHOOK
1133 	struct	osd		osd;
1134 #endif
1135 };
1136 
1137 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1138 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1139 
1140 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1141 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1142 
1143 static struct mtx isn_mtx;
1144 
1145 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1146 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1147 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1148 
1149 /*
1150  * TCP initialization.
1151  */
1152 static void
1153 tcp_zone_change(void *tag)
1154 {
1155 
1156 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
1157 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1158 	tcp_tw_zone_change();
1159 }
1160 
1161 static int
1162 tcp_inpcb_init(void *mem, int size, int flags)
1163 {
1164 	struct inpcb *inp = mem;
1165 
1166 	INP_LOCK_INIT(inp, "inp", "tcpinp");
1167 	return (0);
1168 }
1169 
1170 /*
1171  * Take a value and get the next power of 2 that doesn't overflow.
1172  * Used to size the tcp_inpcb hash buckets.
1173  */
1174 static int
1175 maketcp_hashsize(int size)
1176 {
1177 	int hashsize;
1178 
1179 	/*
1180 	 * auto tune.
1181 	 * get the next power of 2 higher than maxsockets.
1182 	 */
1183 	hashsize = 1 << fls(size);
1184 	/* catch overflow, and just go one power of 2 smaller */
1185 	if (hashsize < size) {
1186 		hashsize = 1 << (fls(size) - 1);
1187 	}
1188 	return (hashsize);
1189 }
1190 
1191 static volatile int next_tcp_stack_id = 1;
1192 
1193 /*
1194  * Register a TCP function block with the name provided in the names
1195  * array.  (Note that this function does NOT automatically register
1196  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1197  * explicitly include blk->tfb_tcp_block_name in the list of names if
1198  * you wish to register the stack with that name.)
1199  *
1200  * Either all name registrations will succeed or all will fail.  If
1201  * a name registration fails, the function will update the num_names
1202  * argument to point to the array index of the name that encountered
1203  * the failure.
1204  *
1205  * Returns 0 on success, or an error code on failure.
1206  */
1207 int
1208 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1209     const char *names[], int *num_names)
1210 {
1211 	struct tcp_function *n;
1212 	struct tcp_function_set fs;
1213 	int error, i;
1214 
1215 	KASSERT(names != NULL && *num_names > 0,
1216 	    ("%s: Called with 0-length name list", __func__));
1217 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1218 	KASSERT(rw_initialized(&tcp_function_lock),
1219 	    ("%s: called too early", __func__));
1220 
1221 	if ((blk->tfb_tcp_output == NULL) ||
1222 	    (blk->tfb_tcp_do_segment == NULL) ||
1223 	    (blk->tfb_tcp_ctloutput == NULL) ||
1224 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1225 		/*
1226 		 * These functions are required and you
1227 		 * need a name.
1228 		 */
1229 		*num_names = 0;
1230 		return (EINVAL);
1231 	}
1232 	if (blk->tfb_tcp_timer_stop_all ||
1233 	    blk->tfb_tcp_timer_activate ||
1234 	    blk->tfb_tcp_timer_active ||
1235 	    blk->tfb_tcp_timer_stop) {
1236 		/*
1237 		 * If you define one timer function you
1238 		 * must have them all.
1239 		 */
1240 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1241 		    (blk->tfb_tcp_timer_activate == NULL) ||
1242 		    (blk->tfb_tcp_timer_active == NULL) ||
1243 		    (blk->tfb_tcp_timer_stop == NULL)) {
1244 			*num_names = 0;
1245 			return (EINVAL);
1246 		}
1247 	}
1248 
1249 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1250 		*num_names = 0;
1251 		return (EINVAL);
1252 	}
1253 
1254 	refcount_init(&blk->tfb_refcnt, 0);
1255 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1256 	for (i = 0; i < *num_names; i++) {
1257 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1258 		if (n == NULL) {
1259 			error = ENOMEM;
1260 			goto cleanup;
1261 		}
1262 		n->tf_fb = blk;
1263 
1264 		(void)strlcpy(fs.function_set_name, names[i],
1265 		    sizeof(fs.function_set_name));
1266 		rw_wlock(&tcp_function_lock);
1267 		if (find_tcp_functions_locked(&fs) != NULL) {
1268 			/* Duplicate name space not allowed */
1269 			rw_wunlock(&tcp_function_lock);
1270 			free(n, M_TCPFUNCTIONS);
1271 			error = EALREADY;
1272 			goto cleanup;
1273 		}
1274 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1275 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1276 		tcp_fb_cnt++;
1277 		rw_wunlock(&tcp_function_lock);
1278 	}
1279 	return(0);
1280 
1281 cleanup:
1282 	/*
1283 	 * Deregister the names we just added. Because registration failed
1284 	 * for names[i], we don't need to deregister that name.
1285 	 */
1286 	*num_names = i;
1287 	rw_wlock(&tcp_function_lock);
1288 	while (--i >= 0) {
1289 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1290 			if (!strncmp(n->tf_name, names[i],
1291 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1292 				TAILQ_REMOVE(&t_functions, n, tf_next);
1293 				tcp_fb_cnt--;
1294 				n->tf_fb = NULL;
1295 				free(n, M_TCPFUNCTIONS);
1296 				break;
1297 			}
1298 		}
1299 	}
1300 	rw_wunlock(&tcp_function_lock);
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Register a TCP function block using the name provided in the name
1306  * argument.
1307  *
1308  * Returns 0 on success, or an error code on failure.
1309  */
1310 int
1311 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1312     int wait)
1313 {
1314 	const char *name_list[1];
1315 	int num_names, rv;
1316 
1317 	num_names = 1;
1318 	if (name != NULL)
1319 		name_list[0] = name;
1320 	else
1321 		name_list[0] = blk->tfb_tcp_block_name;
1322 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1323 	return (rv);
1324 }
1325 
1326 /*
1327  * Register a TCP function block using the name defined in
1328  * blk->tfb_tcp_block_name.
1329  *
1330  * Returns 0 on success, or an error code on failure.
1331  */
1332 int
1333 register_tcp_functions(struct tcp_function_block *blk, int wait)
1334 {
1335 
1336 	return (register_tcp_functions_as_name(blk, NULL, wait));
1337 }
1338 
1339 /*
1340  * Deregister all names associated with a function block. This
1341  * functionally removes the function block from use within the system.
1342  *
1343  * When called with a true quiesce argument, mark the function block
1344  * as being removed so no more stacks will use it and determine
1345  * whether the removal would succeed.
1346  *
1347  * When called with a false quiesce argument, actually attempt the
1348  * removal.
1349  *
1350  * When called with a force argument, attempt to switch all TCBs to
1351  * use the default stack instead of returning EBUSY.
1352  *
1353  * Returns 0 on success (or if the removal would succeed, or an error
1354  * code on failure.
1355  */
1356 int
1357 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1358     bool force)
1359 {
1360 	struct tcp_function *f;
1361 
1362 	if (blk == &tcp_def_funcblk) {
1363 		/* You can't un-register the default */
1364 		return (EPERM);
1365 	}
1366 	rw_wlock(&tcp_function_lock);
1367 	if (blk == tcp_func_set_ptr) {
1368 		/* You can't free the current default */
1369 		rw_wunlock(&tcp_function_lock);
1370 		return (EBUSY);
1371 	}
1372 	/* Mark the block so no more stacks can use it. */
1373 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1374 	/*
1375 	 * If TCBs are still attached to the stack, attempt to switch them
1376 	 * to the default stack.
1377 	 */
1378 	if (force && blk->tfb_refcnt) {
1379 		struct inpcb *inp;
1380 		struct tcpcb *tp;
1381 		VNET_ITERATOR_DECL(vnet_iter);
1382 
1383 		rw_wunlock(&tcp_function_lock);
1384 
1385 		VNET_LIST_RLOCK();
1386 		VNET_FOREACH(vnet_iter) {
1387 			CURVNET_SET(vnet_iter);
1388 			INP_INFO_WLOCK(&V_tcbinfo);
1389 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1390 				INP_WLOCK(inp);
1391 				if (inp->inp_flags & INP_TIMEWAIT) {
1392 					INP_WUNLOCK(inp);
1393 					continue;
1394 				}
1395 				tp = intotcpcb(inp);
1396 				if (tp == NULL || tp->t_fb != blk) {
1397 					INP_WUNLOCK(inp);
1398 					continue;
1399 				}
1400 				tcp_switch_back_to_default(tp);
1401 				INP_WUNLOCK(inp);
1402 			}
1403 			INP_INFO_WUNLOCK(&V_tcbinfo);
1404 			CURVNET_RESTORE();
1405 		}
1406 		VNET_LIST_RUNLOCK();
1407 
1408 		rw_wlock(&tcp_function_lock);
1409 	}
1410 	if (blk->tfb_refcnt) {
1411 		/* TCBs still attached. */
1412 		rw_wunlock(&tcp_function_lock);
1413 		return (EBUSY);
1414 	}
1415 	if (quiesce) {
1416 		/* Skip removal. */
1417 		rw_wunlock(&tcp_function_lock);
1418 		return (0);
1419 	}
1420 	/* Remove any function names that map to this function block. */
1421 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1422 		TAILQ_REMOVE(&t_functions, f, tf_next);
1423 		tcp_fb_cnt--;
1424 		f->tf_fb = NULL;
1425 		free(f, M_TCPFUNCTIONS);
1426 	}
1427 	rw_wunlock(&tcp_function_lock);
1428 	return (0);
1429 }
1430 
1431 void
1432 tcp_init(void)
1433 {
1434 	const char *tcbhash_tuneable;
1435 	int hashsize;
1436 
1437 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1438 
1439 #ifdef TCP_HHOOK
1440 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1441 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1442 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1443 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1444 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1445 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1446 #endif
1447 #ifdef STATS
1448 	if (tcp_stats_init())
1449 		printf("%s: WARNING: unable to initialise TCP stats\n",
1450 		    __func__);
1451 #endif
1452 	hashsize = TCBHASHSIZE;
1453 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1454 	if (hashsize == 0) {
1455 		/*
1456 		 * Auto tune the hash size based on maxsockets.
1457 		 * A perfect hash would have a 1:1 mapping
1458 		 * (hashsize = maxsockets) however it's been
1459 		 * suggested that O(2) average is better.
1460 		 */
1461 		hashsize = maketcp_hashsize(maxsockets / 4);
1462 		/*
1463 		 * Our historical default is 512,
1464 		 * do not autotune lower than this.
1465 		 */
1466 		if (hashsize < 512)
1467 			hashsize = 512;
1468 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1469 			printf("%s: %s auto tuned to %d\n", __func__,
1470 			    tcbhash_tuneable, hashsize);
1471 	}
1472 	/*
1473 	 * We require a hashsize to be a power of two.
1474 	 * Previously if it was not a power of two we would just reset it
1475 	 * back to 512, which could be a nasty surprise if you did not notice
1476 	 * the error message.
1477 	 * Instead what we do is clip it to the closest power of two lower
1478 	 * than the specified hash value.
1479 	 */
1480 	if (!powerof2(hashsize)) {
1481 		int oldhashsize = hashsize;
1482 
1483 		hashsize = maketcp_hashsize(hashsize);
1484 		/* prevent absurdly low value */
1485 		if (hashsize < 16)
1486 			hashsize = 16;
1487 		printf("%s: WARNING: TCB hash size not a power of 2, "
1488 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1489 		    hashsize);
1490 	}
1491 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1492 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1493 
1494 	/*
1495 	 * These have to be type stable for the benefit of the timers.
1496 	 */
1497 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1498 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1499 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1500 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1501 
1502 	tcp_tw_init();
1503 	syncache_init();
1504 	tcp_hc_init();
1505 
1506 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1507 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1508 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1509 
1510 	tcp_fastopen_init();
1511 
1512 	/* Skip initialization of globals for non-default instances. */
1513 	if (!IS_DEFAULT_VNET(curvnet))
1514 		return;
1515 
1516 	tcp_reass_global_init();
1517 
1518 	/* XXX virtualize those bellow? */
1519 	tcp_delacktime = TCPTV_DELACK;
1520 	tcp_keepinit = TCPTV_KEEP_INIT;
1521 	tcp_keepidle = TCPTV_KEEP_IDLE;
1522 	tcp_keepintvl = TCPTV_KEEPINTVL;
1523 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1524 	tcp_msl = TCPTV_MSL;
1525 	tcp_rexmit_initial = TCPTV_RTOBASE;
1526 	if (tcp_rexmit_initial < 1)
1527 		tcp_rexmit_initial = 1;
1528 	tcp_rexmit_min = TCPTV_MIN;
1529 	if (tcp_rexmit_min < 1)
1530 		tcp_rexmit_min = 1;
1531 	tcp_persmin = TCPTV_PERSMIN;
1532 	tcp_persmax = TCPTV_PERSMAX;
1533 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1534 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1535 	tcp_tcbhashsize = hashsize;
1536 
1537 	/* Setup the tcp function block list */
1538 	TAILQ_INIT(&t_functions);
1539 	rw_init(&tcp_function_lock, "tcp_func_lock");
1540 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1541 #ifdef TCP_BLACKBOX
1542 	/* Initialize the TCP logging data. */
1543 	tcp_log_init();
1544 #endif
1545 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1546 
1547 	if (tcp_soreceive_stream) {
1548 #ifdef INET
1549 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1550 #endif
1551 #ifdef INET6
1552 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1553 #endif /* INET6 */
1554 	}
1555 
1556 #ifdef INET6
1557 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1558 #else /* INET6 */
1559 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1560 #endif /* INET6 */
1561 	if (max_protohdr < TCP_MINPROTOHDR)
1562 		max_protohdr = TCP_MINPROTOHDR;
1563 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1564 		panic("tcp_init");
1565 #undef TCP_MINPROTOHDR
1566 
1567 	ISN_LOCK_INIT();
1568 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1569 		SHUTDOWN_PRI_DEFAULT);
1570 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1571 		EVENTHANDLER_PRI_ANY);
1572 
1573 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1574 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1575 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1576 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1577 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1578 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1579 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1580 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1581 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1582 #ifdef TCPPCAP
1583 	tcp_pcap_init();
1584 #endif
1585 }
1586 
1587 #ifdef VIMAGE
1588 static void
1589 tcp_destroy(void *unused __unused)
1590 {
1591 	int n;
1592 #ifdef TCP_HHOOK
1593 	int error;
1594 #endif
1595 
1596 	/*
1597 	 * All our processes are gone, all our sockets should be cleaned
1598 	 * up, which means, we should be past the tcp_discardcb() calls.
1599 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1600 	 */
1601 	for (;;) {
1602 		INP_LIST_RLOCK(&V_tcbinfo);
1603 		n = V_tcbinfo.ipi_count;
1604 		INP_LIST_RUNLOCK(&V_tcbinfo);
1605 		if (n == 0)
1606 			break;
1607 		pause("tcpdes", hz / 10);
1608 	}
1609 	tcp_hc_destroy();
1610 	syncache_destroy();
1611 	tcp_tw_destroy();
1612 	in_pcbinfo_destroy(&V_tcbinfo);
1613 	/* tcp_discardcb() clears the sack_holes up. */
1614 	uma_zdestroy(V_sack_hole_zone);
1615 	uma_zdestroy(V_tcpcb_zone);
1616 
1617 	/*
1618 	 * Cannot free the zone until all tcpcbs are released as we attach
1619 	 * the allocations to them.
1620 	 */
1621 	tcp_fastopen_destroy();
1622 
1623 #ifdef TCP_HHOOK
1624 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1625 	if (error != 0) {
1626 		printf("%s: WARNING: unable to deregister helper hook "
1627 		    "type=%d, id=%d: error %d returned\n", __func__,
1628 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1629 	}
1630 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1631 	if (error != 0) {
1632 		printf("%s: WARNING: unable to deregister helper hook "
1633 		    "type=%d, id=%d: error %d returned\n", __func__,
1634 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1635 	}
1636 #endif
1637 }
1638 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1639 #endif
1640 
1641 void
1642 tcp_fini(void *xtp)
1643 {
1644 
1645 }
1646 
1647 /*
1648  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1649  * tcp_template used to store this data in mbufs, but we now recopy it out
1650  * of the tcpcb each time to conserve mbufs.
1651  */
1652 void
1653 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1654 {
1655 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1656 
1657 	INP_WLOCK_ASSERT(inp);
1658 
1659 #ifdef INET6
1660 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1661 		struct ip6_hdr *ip6;
1662 
1663 		ip6 = (struct ip6_hdr *)ip_ptr;
1664 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1665 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1666 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1667 			(IPV6_VERSION & IPV6_VERSION_MASK);
1668 		if (port == 0)
1669 			ip6->ip6_nxt = IPPROTO_TCP;
1670 		else
1671 			ip6->ip6_nxt = IPPROTO_UDP;
1672 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1673 		ip6->ip6_src = inp->in6p_laddr;
1674 		ip6->ip6_dst = inp->in6p_faddr;
1675 	}
1676 #endif /* INET6 */
1677 #if defined(INET6) && defined(INET)
1678 	else
1679 #endif
1680 #ifdef INET
1681 	{
1682 		struct ip *ip;
1683 
1684 		ip = (struct ip *)ip_ptr;
1685 		ip->ip_v = IPVERSION;
1686 		ip->ip_hl = 5;
1687 		ip->ip_tos = inp->inp_ip_tos;
1688 		ip->ip_len = 0;
1689 		ip->ip_id = 0;
1690 		ip->ip_off = 0;
1691 		ip->ip_ttl = inp->inp_ip_ttl;
1692 		ip->ip_sum = 0;
1693 		if (port == 0)
1694 			ip->ip_p = IPPROTO_TCP;
1695 		else
1696 			ip->ip_p = IPPROTO_UDP;
1697 		ip->ip_src = inp->inp_laddr;
1698 		ip->ip_dst = inp->inp_faddr;
1699 	}
1700 #endif /* INET */
1701 	th->th_sport = inp->inp_lport;
1702 	th->th_dport = inp->inp_fport;
1703 	th->th_seq = 0;
1704 	th->th_ack = 0;
1705 	th->th_x2 = 0;
1706 	th->th_off = 5;
1707 	th->th_flags = 0;
1708 	th->th_win = 0;
1709 	th->th_urp = 0;
1710 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1711 }
1712 
1713 /*
1714  * Create template to be used to send tcp packets on a connection.
1715  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1716  * use for this function is in keepalives, which use tcp_respond.
1717  */
1718 struct tcptemp *
1719 tcpip_maketemplate(struct inpcb *inp)
1720 {
1721 	struct tcptemp *t;
1722 
1723 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1724 	if (t == NULL)
1725 		return (NULL);
1726 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1727 	return (t);
1728 }
1729 
1730 /*
1731  * Send a single message to the TCP at address specified by
1732  * the given TCP/IP header.  If m == NULL, then we make a copy
1733  * of the tcpiphdr at th and send directly to the addressed host.
1734  * This is used to force keep alive messages out using the TCP
1735  * template for a connection.  If flags are given then we send
1736  * a message back to the TCP which originated the segment th,
1737  * and discard the mbuf containing it and any other attached mbufs.
1738  *
1739  * In any case the ack and sequence number of the transmitted
1740  * segment are as specified by the parameters.
1741  *
1742  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1743  */
1744 void
1745 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1746     tcp_seq ack, tcp_seq seq, int flags)
1747 {
1748 	struct tcpopt to;
1749 	struct inpcb *inp;
1750 	struct ip *ip;
1751 	struct mbuf *optm;
1752 	struct udphdr *uh = NULL;
1753 	struct tcphdr *nth;
1754 	struct tcp_log_buffer *lgb;
1755 	u_char *optp;
1756 #ifdef INET6
1757 	struct ip6_hdr *ip6;
1758 	int isipv6;
1759 #endif /* INET6 */
1760 	int optlen, tlen, win, ulen;
1761 	bool incl_opts;
1762 	uint16_t port;
1763 	int output_ret;
1764 #ifdef INVARIANTS
1765 	int thflags = th->th_flags;
1766 #endif
1767 
1768 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1769 	NET_EPOCH_ASSERT();
1770 
1771 #ifdef INET6
1772 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1773 	ip6 = ipgen;
1774 #endif /* INET6 */
1775 	ip = ipgen;
1776 
1777 	if (tp != NULL) {
1778 		inp = tp->t_inpcb;
1779 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1780 		INP_LOCK_ASSERT(inp);
1781 	} else
1782 		inp = NULL;
1783 
1784 	if (m != NULL) {
1785 #ifdef INET6
1786 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1787 			port = m->m_pkthdr.tcp_tun_port;
1788 		else
1789 #endif
1790 		if (ip && (ip->ip_p == IPPROTO_UDP))
1791 			port = m->m_pkthdr.tcp_tun_port;
1792 		else
1793 			port = 0;
1794 	} else
1795 		port = tp->t_port;
1796 
1797 	incl_opts = false;
1798 	win = 0;
1799 	if (tp != NULL) {
1800 		if (!(flags & TH_RST)) {
1801 			win = sbspace(&inp->inp_socket->so_rcv);
1802 			if (win > TCP_MAXWIN << tp->rcv_scale)
1803 				win = TCP_MAXWIN << tp->rcv_scale;
1804 		}
1805 		if ((tp->t_flags & TF_NOOPT) == 0)
1806 			incl_opts = true;
1807 	}
1808 	if (m == NULL) {
1809 		m = m_gethdr(M_NOWAIT, MT_DATA);
1810 		if (m == NULL)
1811 			return;
1812 		m->m_data += max_linkhdr;
1813 #ifdef INET6
1814 		if (isipv6) {
1815 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1816 			      sizeof(struct ip6_hdr));
1817 			ip6 = mtod(m, struct ip6_hdr *);
1818 			nth = (struct tcphdr *)(ip6 + 1);
1819 			if (port) {
1820 				/* Insert a UDP header */
1821 				uh = (struct udphdr *)nth;
1822 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1823 				uh->uh_dport = port;
1824 				nth = (struct tcphdr *)(uh + 1);
1825 			}
1826 		} else
1827 #endif /* INET6 */
1828 		{
1829 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1830 			ip = mtod(m, struct ip *);
1831 			nth = (struct tcphdr *)(ip + 1);
1832 			if (port) {
1833 				/* Insert a UDP header */
1834 				uh = (struct udphdr *)nth;
1835 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1836 				uh->uh_dport = port;
1837 				nth = (struct tcphdr *)(uh + 1);
1838 			}
1839 		}
1840 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1841 		flags = TH_ACK;
1842 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1843 		struct mbuf *n;
1844 
1845 		/* Can't reuse 'm', allocate a new mbuf. */
1846 		n = m_gethdr(M_NOWAIT, MT_DATA);
1847 		if (n == NULL) {
1848 			m_freem(m);
1849 			return;
1850 		}
1851 
1852 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1853 			m_freem(m);
1854 			m_freem(n);
1855 			return;
1856 		}
1857 
1858 		n->m_data += max_linkhdr;
1859 		/* m_len is set later */
1860 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1861 #ifdef INET6
1862 		if (isipv6) {
1863 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1864 			      sizeof(struct ip6_hdr));
1865 			ip6 = mtod(n, struct ip6_hdr *);
1866 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1867 			nth = (struct tcphdr *)(ip6 + 1);
1868 			if (port) {
1869 				/* Insert a UDP header */
1870 				uh = (struct udphdr *)nth;
1871 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1872 				uh->uh_dport = port;
1873 				nth = (struct tcphdr *)(uh + 1);
1874 			}
1875 		} else
1876 #endif /* INET6 */
1877 		{
1878 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1879 			ip = mtod(n, struct ip *);
1880 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1881 			nth = (struct tcphdr *)(ip + 1);
1882 			if (port) {
1883 				/* Insert a UDP header */
1884 				uh = (struct udphdr *)nth;
1885 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1886 				uh->uh_dport = port;
1887 				nth = (struct tcphdr *)(uh + 1);
1888 			}
1889 		}
1890 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1891 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1892 		th = nth;
1893 		m_freem(m);
1894 		m = n;
1895 	} else {
1896 		/*
1897 		 *  reuse the mbuf.
1898 		 * XXX MRT We inherit the FIB, which is lucky.
1899 		 */
1900 		m_freem(m->m_next);
1901 		m->m_next = NULL;
1902 		m->m_data = (caddr_t)ipgen;
1903 		/* m_len is set later */
1904 #ifdef INET6
1905 		if (isipv6) {
1906 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1907 			nth = (struct tcphdr *)(ip6 + 1);
1908 		} else
1909 #endif /* INET6 */
1910 		{
1911 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1912 			nth = (struct tcphdr *)(ip + 1);
1913 		}
1914 		if (th != nth) {
1915 			/*
1916 			 * this is usually a case when an extension header
1917 			 * exists between the IPv6 header and the
1918 			 * TCP header.
1919 			 */
1920 			nth->th_sport = th->th_sport;
1921 			nth->th_dport = th->th_dport;
1922 		}
1923 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1924 #undef xchg
1925 	}
1926 	tlen = 0;
1927 #ifdef INET6
1928 	if (isipv6)
1929 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1930 #endif
1931 #if defined(INET) && defined(INET6)
1932 	else
1933 #endif
1934 #ifdef INET
1935 		tlen = sizeof (struct tcpiphdr);
1936 #endif
1937 	if (port)
1938 		tlen += sizeof (struct udphdr);
1939 #ifdef INVARIANTS
1940 	m->m_len = 0;
1941 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1942 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1943 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1944 #endif
1945 	m->m_len = tlen;
1946 	to.to_flags = 0;
1947 	if (incl_opts) {
1948 		/* Make sure we have room. */
1949 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1950 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1951 			if (m->m_next) {
1952 				optp = mtod(m->m_next, u_char *);
1953 				optm = m->m_next;
1954 			} else
1955 				incl_opts = false;
1956 		} else {
1957 			optp = (u_char *) (nth + 1);
1958 			optm = m;
1959 		}
1960 	}
1961 	if (incl_opts) {
1962 		/* Timestamps. */
1963 		if (tp->t_flags & TF_RCVD_TSTMP) {
1964 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1965 			to.to_tsecr = tp->ts_recent;
1966 			to.to_flags |= TOF_TS;
1967 		}
1968 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1969 		/* TCP-MD5 (RFC2385). */
1970 		if (tp->t_flags & TF_SIGNATURE)
1971 			to.to_flags |= TOF_SIGNATURE;
1972 #endif
1973 		/* Add the options. */
1974 		tlen += optlen = tcp_addoptions(&to, optp);
1975 
1976 		/* Update m_len in the correct mbuf. */
1977 		optm->m_len += optlen;
1978 	} else
1979 		optlen = 0;
1980 #ifdef INET6
1981 	if (isipv6) {
1982 		if (uh) {
1983 			ulen = tlen - sizeof(struct ip6_hdr);
1984 			uh->uh_ulen = htons(ulen);
1985 		}
1986 		ip6->ip6_flow = 0;
1987 		ip6->ip6_vfc = IPV6_VERSION;
1988 		if (port)
1989 			ip6->ip6_nxt = IPPROTO_UDP;
1990 		else
1991 			ip6->ip6_nxt = IPPROTO_TCP;
1992 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1993 	}
1994 #endif
1995 #if defined(INET) && defined(INET6)
1996 	else
1997 #endif
1998 #ifdef INET
1999 	{
2000 		if (uh) {
2001 			ulen = tlen - sizeof(struct ip);
2002 			uh->uh_ulen = htons(ulen);
2003 		}
2004 		ip->ip_len = htons(tlen);
2005 		ip->ip_ttl = V_ip_defttl;
2006 		if (port) {
2007 			ip->ip_p = IPPROTO_UDP;
2008 		} else {
2009 			ip->ip_p = IPPROTO_TCP;
2010 		}
2011 		if (V_path_mtu_discovery)
2012 			ip->ip_off |= htons(IP_DF);
2013 	}
2014 #endif
2015 	m->m_pkthdr.len = tlen;
2016 	m->m_pkthdr.rcvif = NULL;
2017 #ifdef MAC
2018 	if (inp != NULL) {
2019 		/*
2020 		 * Packet is associated with a socket, so allow the
2021 		 * label of the response to reflect the socket label.
2022 		 */
2023 		INP_LOCK_ASSERT(inp);
2024 		mac_inpcb_create_mbuf(inp, m);
2025 	} else {
2026 		/*
2027 		 * Packet is not associated with a socket, so possibly
2028 		 * update the label in place.
2029 		 */
2030 		mac_netinet_tcp_reply(m);
2031 	}
2032 #endif
2033 	nth->th_seq = htonl(seq);
2034 	nth->th_ack = htonl(ack);
2035 	nth->th_x2 = 0;
2036 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2037 	nth->th_flags = flags;
2038 	if (tp != NULL)
2039 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2040 	else
2041 		nth->th_win = htons((u_short)win);
2042 	nth->th_urp = 0;
2043 
2044 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2045 	if (to.to_flags & TOF_SIGNATURE) {
2046 		if (!TCPMD5_ENABLED() ||
2047 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2048 			m_freem(m);
2049 			return;
2050 		}
2051 	}
2052 #endif
2053 
2054 #ifdef INET6
2055 	if (isipv6) {
2056 		if (port) {
2057 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2058 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2059 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2060 			nth->th_sum = 0;
2061 		} else {
2062 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2063 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2064 			nth->th_sum = in6_cksum_pseudo(ip6,
2065 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2066 		}
2067 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2068 		    NULL, NULL);
2069 	}
2070 #endif /* INET6 */
2071 #if defined(INET6) && defined(INET)
2072 	else
2073 #endif
2074 #ifdef INET
2075 	{
2076 		if (port) {
2077 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2078 			    htons(ulen + IPPROTO_UDP));
2079 			m->m_pkthdr.csum_flags = CSUM_UDP;
2080 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2081 			nth->th_sum = 0;
2082 		} else {
2083 			m->m_pkthdr.csum_flags = CSUM_TCP;
2084 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2085 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2086 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2087 		}
2088 	}
2089 #endif /* INET */
2090 #ifdef TCPDEBUG
2091 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2092 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2093 #endif
2094 	TCP_PROBE3(debug__output, tp, th, m);
2095 	if (flags & TH_RST)
2096 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2097 	lgb = NULL;
2098 	if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2099 		if (INP_WLOCKED(inp)) {
2100 			union tcp_log_stackspecific log;
2101 			struct timeval tv;
2102 
2103 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2104 			log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2105 			log.u_bbr.ininput = tp->t_inpcb->inp_in_input;
2106 			log.u_bbr.flex8 = 4;
2107 			log.u_bbr.pkts_out = tp->t_maxseg;
2108 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2109 			log.u_bbr.delivered = 0;
2110 			lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2111 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2112 		} else {
2113 			/*
2114 			 * We can not log the packet, since we only own the
2115 			 * read lock, but a write lock is needed. The read lock
2116 			 * is not upgraded to a write lock, since only getting
2117 			 * the read lock was done intentionally to improve the
2118 			 * handling of SYN flooding attacks.
2119 			 * This happens only for pure SYN segments received in
2120 			 * the initial CLOSED state, or received in a more
2121 			 * advanced state than listen and the UDP encapsulation
2122 			 * port is unexpected.
2123 			 * The incoming SYN segments do not really belong to
2124 			 * the TCP connection and the handling does not change
2125 			 * the state of the TCP connection. Therefore, the
2126 			 * sending of the RST segments is not logged. Please
2127 			 * note that also the incoming SYN segments are not
2128 			 * logged.
2129 			 *
2130 			 * The following code ensures that the above description
2131 			 * is and stays correct.
2132 			 */
2133 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2134 			    (tp->t_state == TCPS_CLOSED ||
2135 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2136 			    ("%s: Logging of TCP segment with flags 0x%b and "
2137 			    "UDP encapsulation port %u skipped in state %s",
2138 			    __func__, thflags, PRINT_TH_FLAGS,
2139 			    ntohs(port), tcpstates[tp->t_state]));
2140 		}
2141 	}
2142 
2143 #ifdef INET6
2144 	if (isipv6) {
2145 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2146 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2147 	}
2148 #endif /* INET6 */
2149 #if defined(INET) && defined(INET6)
2150 	else
2151 #endif
2152 #ifdef INET
2153 	{
2154 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2155 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2156 	}
2157 #endif
2158 	if (lgb != NULL)
2159 		lgb->tlb_errno = output_ret;
2160 }
2161 
2162 /*
2163  * Create a new TCP control block, making an
2164  * empty reassembly queue and hooking it to the argument
2165  * protocol control block.  The `inp' parameter must have
2166  * come from the zone allocator set up in tcp_init().
2167  */
2168 struct tcpcb *
2169 tcp_newtcpcb(struct inpcb *inp)
2170 {
2171 	struct tcpcb_mem *tm;
2172 	struct tcpcb *tp;
2173 #ifdef INET6
2174 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2175 #endif /* INET6 */
2176 
2177 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2178 	if (tm == NULL)
2179 		return (NULL);
2180 	tp = &tm->tcb;
2181 
2182 	/* Initialise cc_var struct for this tcpcb. */
2183 	tp->ccv = &tm->ccv;
2184 	tp->ccv->type = IPPROTO_TCP;
2185 	tp->ccv->ccvc.tcp = tp;
2186 	rw_rlock(&tcp_function_lock);
2187 	tp->t_fb = tcp_func_set_ptr;
2188 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2189 	rw_runlock(&tcp_function_lock);
2190 	/*
2191 	 * Use the current system default CC algorithm.
2192 	 */
2193 	CC_LIST_RLOCK();
2194 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2195 	CC_ALGO(tp) = CC_DEFAULT_ALGO();
2196 	CC_LIST_RUNLOCK();
2197 
2198 	/*
2199 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2200 	 * is called.
2201 	 */
2202 	in_pcbref(inp);	/* Reference for tcpcb */
2203 	tp->t_inpcb = inp;
2204 
2205 	if (CC_ALGO(tp)->cb_init != NULL)
2206 		if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2207 			if (tp->t_fb->tfb_tcp_fb_fini)
2208 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2209 			in_pcbrele_wlocked(inp);
2210 			refcount_release(&tp->t_fb->tfb_refcnt);
2211 			uma_zfree(V_tcpcb_zone, tm);
2212 			return (NULL);
2213 		}
2214 
2215 #ifdef TCP_HHOOK
2216 	tp->osd = &tm->osd;
2217 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2218 		if (tp->t_fb->tfb_tcp_fb_fini)
2219 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2220 		in_pcbrele_wlocked(inp);
2221 		refcount_release(&tp->t_fb->tfb_refcnt);
2222 		uma_zfree(V_tcpcb_zone, tm);
2223 		return (NULL);
2224 	}
2225 #endif
2226 
2227 #ifdef VIMAGE
2228 	tp->t_vnet = inp->inp_vnet;
2229 #endif
2230 	tp->t_timers = &tm->tt;
2231 	TAILQ_INIT(&tp->t_segq);
2232 	tp->t_maxseg =
2233 #ifdef INET6
2234 		isipv6 ? V_tcp_v6mssdflt :
2235 #endif /* INET6 */
2236 		V_tcp_mssdflt;
2237 
2238 	/* Set up our timeouts. */
2239 	callout_init(&tp->t_timers->tt_rexmt, 1);
2240 	callout_init(&tp->t_timers->tt_persist, 1);
2241 	callout_init(&tp->t_timers->tt_keep, 1);
2242 	callout_init(&tp->t_timers->tt_2msl, 1);
2243 	callout_init(&tp->t_timers->tt_delack, 1);
2244 
2245 	if (V_tcp_do_rfc1323)
2246 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2247 	if (V_tcp_do_sack)
2248 		tp->t_flags |= TF_SACK_PERMIT;
2249 	TAILQ_INIT(&tp->snd_holes);
2250 
2251 	/*
2252 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2253 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2254 	 * reasonable initial retransmit time.
2255 	 */
2256 	tp->t_srtt = TCPTV_SRTTBASE;
2257 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2258 	tp->t_rttmin = tcp_rexmit_min;
2259 	tp->t_rxtcur = tcp_rexmit_initial;
2260 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2261 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2262 	tp->t_rcvtime = ticks;
2263 	/*
2264 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2265 	 * because the socket may be bound to an IPv6 wildcard address,
2266 	 * which may match an IPv4-mapped IPv6 address.
2267 	 */
2268 	inp->inp_ip_ttl = V_ip_defttl;
2269 	inp->inp_ppcb = tp;
2270 #ifdef TCPPCAP
2271 	/*
2272 	 * Init the TCP PCAP queues.
2273 	 */
2274 	tcp_pcap_tcpcb_init(tp);
2275 #endif
2276 #ifdef TCP_BLACKBOX
2277 	/* Initialize the per-TCPCB log data. */
2278 	tcp_log_tcpcbinit(tp);
2279 #endif
2280 	tp->t_pacing_rate = -1;
2281 	if (tp->t_fb->tfb_tcp_fb_init) {
2282 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2283 			refcount_release(&tp->t_fb->tfb_refcnt);
2284 			in_pcbrele_wlocked(inp);
2285 			uma_zfree(V_tcpcb_zone, tm);
2286 			return (NULL);
2287 		}
2288 	}
2289 #ifdef STATS
2290 	if (V_tcp_perconn_stats_enable == 1)
2291 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2292 #endif
2293 	if (V_tcp_do_lrd)
2294 		tp->t_flags |= TF_LRD;
2295 	return (tp);		/* XXX */
2296 }
2297 
2298 /*
2299  * Switch the congestion control algorithm back to Vnet default for any active
2300  * control blocks using an algorithm which is about to go away. If the algorithm
2301  * has a cb_init function and it fails (no memory) then the operation fails and
2302  * the unload will not succeed.
2303  *
2304  */
2305 int
2306 tcp_ccalgounload(struct cc_algo *unload_algo)
2307 {
2308 	struct cc_algo *oldalgo, *newalgo;
2309 	struct inpcb *inp;
2310 	struct tcpcb *tp;
2311 	VNET_ITERATOR_DECL(vnet_iter);
2312 
2313 	/*
2314 	 * Check all active control blocks across all network stacks and change
2315 	 * any that are using "unload_algo" back to its default. If "unload_algo"
2316 	 * requires cleanup code to be run, call it.
2317 	 */
2318 	VNET_LIST_RLOCK();
2319 	VNET_FOREACH(vnet_iter) {
2320 		CURVNET_SET(vnet_iter);
2321 		INP_INFO_WLOCK(&V_tcbinfo);
2322 		/*
2323 		 * New connections already part way through being initialised
2324 		 * with the CC algo we're removing will not race with this code
2325 		 * because the INP_INFO_WLOCK is held during initialisation. We
2326 		 * therefore don't enter the loop below until the connection
2327 		 * list has stabilised.
2328 		 */
2329 		newalgo = CC_DEFAULT_ALGO();
2330 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
2331 			INP_WLOCK(inp);
2332 			/* Important to skip tcptw structs. */
2333 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
2334 			    (tp = intotcpcb(inp)) != NULL) {
2335 				/*
2336 				 * By holding INP_WLOCK here, we are assured
2337 				 * that the connection is not currently
2338 				 * executing inside the CC module's functions.
2339 				 * We attempt to switch to the Vnets default,
2340 				 * if the init fails then we fail the whole
2341 				 * operation and the module unload will fail.
2342 				 */
2343 				if (CC_ALGO(tp) == unload_algo) {
2344 					struct cc_var cc_mem;
2345 					int err;
2346 
2347 					oldalgo = CC_ALGO(tp);
2348 					memset(&cc_mem, 0, sizeof(cc_mem));
2349 					cc_mem.ccvc.tcp = tp;
2350 					if (newalgo->cb_init == NULL) {
2351 						/*
2352 						 * No init we can skip the
2353 						 * dance around a possible failure.
2354 						 */
2355 						CC_DATA(tp) = NULL;
2356 						goto proceed;
2357 					}
2358 					err = (newalgo->cb_init)(&cc_mem, NULL);
2359 					if (err) {
2360 						/*
2361 						 * Presumably no memory the caller will
2362 						 * need to try again.
2363 						 */
2364 						INP_WUNLOCK(inp);
2365 						INP_INFO_WUNLOCK(&V_tcbinfo);
2366 						CURVNET_RESTORE();
2367 						VNET_LIST_RUNLOCK();
2368 						return (err);
2369 					}
2370 proceed:
2371 					if (oldalgo->cb_destroy != NULL)
2372 						oldalgo->cb_destroy(tp->ccv);
2373 					CC_ALGO(tp) = newalgo;
2374 					memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2375 					if (TCPS_HAVEESTABLISHED(tp->t_state) &&
2376 					    (CC_ALGO(tp)->conn_init != NULL)) {
2377 						/* Yep run the connection init for the new CC */
2378 						CC_ALGO(tp)->conn_init(tp->ccv);
2379 					}
2380 				}
2381 			}
2382 			INP_WUNLOCK(inp);
2383 		}
2384 		INP_INFO_WUNLOCK(&V_tcbinfo);
2385 		CURVNET_RESTORE();
2386 	}
2387 	VNET_LIST_RUNLOCK();
2388 	return (0);
2389 }
2390 
2391 /*
2392  * Drop a TCP connection, reporting
2393  * the specified error.  If connection is synchronized,
2394  * then send a RST to peer.
2395  */
2396 struct tcpcb *
2397 tcp_drop(struct tcpcb *tp, int errno)
2398 {
2399 	struct socket *so = tp->t_inpcb->inp_socket;
2400 
2401 	NET_EPOCH_ASSERT();
2402 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2403 	INP_WLOCK_ASSERT(tp->t_inpcb);
2404 
2405 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2406 		tcp_state_change(tp, TCPS_CLOSED);
2407 		(void) tp->t_fb->tfb_tcp_output(tp);
2408 		TCPSTAT_INC(tcps_drops);
2409 	} else
2410 		TCPSTAT_INC(tcps_conndrops);
2411 	if (errno == ETIMEDOUT && tp->t_softerror)
2412 		errno = tp->t_softerror;
2413 	so->so_error = errno;
2414 	return (tcp_close(tp));
2415 }
2416 
2417 void
2418 tcp_discardcb(struct tcpcb *tp)
2419 {
2420 	struct inpcb *inp = tp->t_inpcb;
2421 
2422 	INP_WLOCK_ASSERT(inp);
2423 
2424 	/*
2425 	 * Make sure that all of our timers are stopped before we delete the
2426 	 * PCB.
2427 	 *
2428 	 * If stopping a timer fails, we schedule a discard function in same
2429 	 * callout, and the last discard function called will take care of
2430 	 * deleting the tcpcb.
2431 	 */
2432 	tp->t_timers->tt_draincnt = 0;
2433 	tcp_timer_stop(tp, TT_REXMT);
2434 	tcp_timer_stop(tp, TT_PERSIST);
2435 	tcp_timer_stop(tp, TT_KEEP);
2436 	tcp_timer_stop(tp, TT_2MSL);
2437 	tcp_timer_stop(tp, TT_DELACK);
2438 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2439 		/*
2440 		 * Call the stop-all function of the methods,
2441 		 * this function should call the tcp_timer_stop()
2442 		 * method with each of the function specific timeouts.
2443 		 * That stop will be called via the tfb_tcp_timer_stop()
2444 		 * which should use the async drain function of the
2445 		 * callout system (see tcp_var.h).
2446 		 */
2447 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2448 	}
2449 
2450 	/* free the reassembly queue, if any */
2451 	tcp_reass_flush(tp);
2452 
2453 #ifdef TCP_OFFLOAD
2454 	/* Disconnect offload device, if any. */
2455 	if (tp->t_flags & TF_TOE)
2456 		tcp_offload_detach(tp);
2457 #endif
2458 
2459 	tcp_free_sackholes(tp);
2460 
2461 #ifdef TCPPCAP
2462 	/* Free the TCP PCAP queues. */
2463 	tcp_pcap_drain(&(tp->t_inpkts));
2464 	tcp_pcap_drain(&(tp->t_outpkts));
2465 #endif
2466 
2467 	/* Allow the CC algorithm to clean up after itself. */
2468 	if (CC_ALGO(tp)->cb_destroy != NULL)
2469 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2470 	CC_DATA(tp) = NULL;
2471 
2472 #ifdef TCP_HHOOK
2473 	khelp_destroy_osd(tp->osd);
2474 #endif
2475 #ifdef STATS
2476 	stats_blob_destroy(tp->t_stats);
2477 #endif
2478 
2479 	CC_ALGO(tp) = NULL;
2480 	inp->inp_ppcb = NULL;
2481 	if (tp->t_timers->tt_draincnt == 0) {
2482 		bool released __diagused;
2483 
2484 		released = tcp_freecb(tp);
2485 		KASSERT(!released, ("%s: inp %p should not have been released "
2486 		    "here", __func__, inp));
2487 	}
2488 }
2489 
2490 bool
2491 tcp_freecb(struct tcpcb *tp)
2492 {
2493 	struct inpcb *inp = tp->t_inpcb;
2494 	struct socket *so = inp->inp_socket;
2495 #ifdef INET6
2496 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2497 #endif
2498 
2499 	INP_WLOCK_ASSERT(inp);
2500 	MPASS(tp->t_timers->tt_draincnt == 0);
2501 
2502 	/* We own the last reference on tcpcb, let's free it. */
2503 #ifdef TCP_BLACKBOX
2504 	tcp_log_tcpcbfini(tp);
2505 #endif
2506 	TCPSTATES_DEC(tp->t_state);
2507 	if (tp->t_fb->tfb_tcp_fb_fini)
2508 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2509 
2510 	/*
2511 	 * If we got enough samples through the srtt filter,
2512 	 * save the rtt and rttvar in the routing entry.
2513 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2514 	 * 4 samples is enough for the srtt filter to converge
2515 	 * to within enough % of the correct value; fewer samples
2516 	 * and we could save a bogus rtt. The danger is not high
2517 	 * as tcp quickly recovers from everything.
2518 	 * XXX: Works very well but needs some more statistics!
2519 	 *
2520 	 * XXXRRS: Updating must be after the stack fini() since
2521 	 * that may be converting some internal representation of
2522 	 * say srtt etc into the general one used by other stacks.
2523 	 * Lets also at least protect against the so being NULL
2524 	 * as RW stated below.
2525 	 */
2526 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2527 		struct hc_metrics_lite metrics;
2528 		uint32_t ssthresh;
2529 
2530 		bzero(&metrics, sizeof(metrics));
2531 		/*
2532 		 * Update the ssthresh always when the conditions below
2533 		 * are satisfied. This gives us better new start value
2534 		 * for the congestion avoidance for new connections.
2535 		 * ssthresh is only set if packet loss occurred on a session.
2536 		 *
2537 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2538 		 * being torn down.  Ideally this code would not use 'so'.
2539 		 */
2540 		ssthresh = tp->snd_ssthresh;
2541 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2542 			/*
2543 			 * convert the limit from user data bytes to
2544 			 * packets then to packet data bytes.
2545 			 */
2546 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2547 			if (ssthresh < 2)
2548 				ssthresh = 2;
2549 			ssthresh *= (tp->t_maxseg +
2550 #ifdef INET6
2551 			    (isipv6 ? sizeof (struct ip6_hdr) +
2552 			    sizeof (struct tcphdr) :
2553 #endif
2554 			    sizeof (struct tcpiphdr)
2555 #ifdef INET6
2556 			    )
2557 #endif
2558 			    );
2559 		} else
2560 			ssthresh = 0;
2561 		metrics.rmx_ssthresh = ssthresh;
2562 
2563 		metrics.rmx_rtt = tp->t_srtt;
2564 		metrics.rmx_rttvar = tp->t_rttvar;
2565 		metrics.rmx_cwnd = tp->snd_cwnd;
2566 		metrics.rmx_sendpipe = 0;
2567 		metrics.rmx_recvpipe = 0;
2568 
2569 		tcp_hc_update(&inp->inp_inc, &metrics);
2570 	}
2571 
2572 	refcount_release(&tp->t_fb->tfb_refcnt);
2573 	uma_zfree(V_tcpcb_zone, tp);
2574 
2575 	return (in_pcbrele_wlocked(inp));
2576 }
2577 
2578 /*
2579  * Attempt to close a TCP control block, marking it as dropped, and freeing
2580  * the socket if we hold the only reference.
2581  */
2582 struct tcpcb *
2583 tcp_close(struct tcpcb *tp)
2584 {
2585 	struct inpcb *inp = tp->t_inpcb;
2586 	struct socket *so;
2587 
2588 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2589 	INP_WLOCK_ASSERT(inp);
2590 
2591 #ifdef TCP_OFFLOAD
2592 	if (tp->t_state == TCPS_LISTEN)
2593 		tcp_offload_listen_stop(tp);
2594 #endif
2595 	/*
2596 	 * This releases the TFO pending counter resource for TFO listen
2597 	 * sockets as well as passively-created TFO sockets that transition
2598 	 * from SYN_RECEIVED to CLOSED.
2599 	 */
2600 	if (tp->t_tfo_pending) {
2601 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2602 		tp->t_tfo_pending = NULL;
2603 	}
2604 	in_pcbdrop(inp);
2605 	TCPSTAT_INC(tcps_closed);
2606 	if (tp->t_state != TCPS_CLOSED)
2607 		tcp_state_change(tp, TCPS_CLOSED);
2608 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2609 	so = inp->inp_socket;
2610 	soisdisconnected(so);
2611 	if (inp->inp_flags & INP_SOCKREF) {
2612 		KASSERT(so->so_state & SS_PROTOREF,
2613 		    ("tcp_close: !SS_PROTOREF"));
2614 		inp->inp_flags &= ~INP_SOCKREF;
2615 		INP_WUNLOCK(inp);
2616 		SOCK_LOCK(so);
2617 		so->so_state &= ~SS_PROTOREF;
2618 		sofree(so);
2619 		return (NULL);
2620 	}
2621 	return (tp);
2622 }
2623 
2624 void
2625 tcp_drain(void)
2626 {
2627 	VNET_ITERATOR_DECL(vnet_iter);
2628 
2629 	if (!do_tcpdrain)
2630 		return;
2631 
2632 	VNET_LIST_RLOCK_NOSLEEP();
2633 	VNET_FOREACH(vnet_iter) {
2634 		CURVNET_SET(vnet_iter);
2635 		struct inpcb *inpb;
2636 		struct tcpcb *tcpb;
2637 
2638 	/*
2639 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2640 	 * if there is one...
2641 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2642 	 *      reassembly queue should be flushed, but in a situation
2643 	 *	where we're really low on mbufs, this is potentially
2644 	 *	useful.
2645 	 */
2646 		INP_INFO_WLOCK(&V_tcbinfo);
2647 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2648 			INP_WLOCK(inpb);
2649 			if (inpb->inp_flags & INP_TIMEWAIT) {
2650 				INP_WUNLOCK(inpb);
2651 				continue;
2652 			}
2653 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2654 				tcp_reass_flush(tcpb);
2655 				tcp_clean_sackreport(tcpb);
2656 #ifdef TCP_BLACKBOX
2657 				tcp_log_drain(tcpb);
2658 #endif
2659 #ifdef TCPPCAP
2660 				if (tcp_pcap_aggressive_free) {
2661 					/* Free the TCP PCAP queues. */
2662 					tcp_pcap_drain(&(tcpb->t_inpkts));
2663 					tcp_pcap_drain(&(tcpb->t_outpkts));
2664 				}
2665 #endif
2666 			}
2667 			INP_WUNLOCK(inpb);
2668 		}
2669 		INP_INFO_WUNLOCK(&V_tcbinfo);
2670 		CURVNET_RESTORE();
2671 	}
2672 	VNET_LIST_RUNLOCK_NOSLEEP();
2673 }
2674 
2675 /*
2676  * Notify a tcp user of an asynchronous error;
2677  * store error as soft error, but wake up user
2678  * (for now, won't do anything until can select for soft error).
2679  *
2680  * Do not wake up user since there currently is no mechanism for
2681  * reporting soft errors (yet - a kqueue filter may be added).
2682  */
2683 static struct inpcb *
2684 tcp_notify(struct inpcb *inp, int error)
2685 {
2686 	struct tcpcb *tp;
2687 
2688 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2689 	INP_WLOCK_ASSERT(inp);
2690 
2691 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2692 	    (inp->inp_flags & INP_DROPPED))
2693 		return (inp);
2694 
2695 	tp = intotcpcb(inp);
2696 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2697 
2698 	/*
2699 	 * Ignore some errors if we are hooked up.
2700 	 * If connection hasn't completed, has retransmitted several times,
2701 	 * and receives a second error, give up now.  This is better
2702 	 * than waiting a long time to establish a connection that
2703 	 * can never complete.
2704 	 */
2705 	if (tp->t_state == TCPS_ESTABLISHED &&
2706 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2707 	     error == EHOSTDOWN)) {
2708 		if (inp->inp_route.ro_nh) {
2709 			NH_FREE(inp->inp_route.ro_nh);
2710 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2711 		}
2712 		return (inp);
2713 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2714 	    tp->t_softerror) {
2715 		tp = tcp_drop(tp, error);
2716 		if (tp != NULL)
2717 			return (inp);
2718 		else
2719 			return (NULL);
2720 	} else {
2721 		tp->t_softerror = error;
2722 		return (inp);
2723 	}
2724 #if 0
2725 	wakeup( &so->so_timeo);
2726 	sorwakeup(so);
2727 	sowwakeup(so);
2728 #endif
2729 }
2730 
2731 static int
2732 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2733 {
2734 	struct epoch_tracker et;
2735 	struct inpcb *inp;
2736 	struct xinpgen xig;
2737 	int error;
2738 
2739 	if (req->newptr != NULL)
2740 		return (EPERM);
2741 
2742 	if (req->oldptr == NULL) {
2743 		int n;
2744 
2745 		n = V_tcbinfo.ipi_count +
2746 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2747 		n += imax(n / 8, 10);
2748 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2749 		return (0);
2750 	}
2751 
2752 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2753 		return (error);
2754 
2755 	bzero(&xig, sizeof(xig));
2756 	xig.xig_len = sizeof xig;
2757 	xig.xig_count = V_tcbinfo.ipi_count +
2758 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2759 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2760 	xig.xig_sogen = so_gencnt;
2761 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2762 	if (error)
2763 		return (error);
2764 
2765 	error = syncache_pcblist(req);
2766 	if (error)
2767 		return (error);
2768 
2769 	NET_EPOCH_ENTER(et);
2770 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2771 	    inp != NULL;
2772 	    inp = CK_LIST_NEXT(inp, inp_list)) {
2773 		INP_RLOCK(inp);
2774 		if (inp->inp_gencnt <= xig.xig_gen) {
2775 			int crerr;
2776 
2777 			/*
2778 			 * XXX: This use of cr_cansee(), introduced with
2779 			 * TCP state changes, is not quite right, but for
2780 			 * now, better than nothing.
2781 			 */
2782 			if (inp->inp_flags & INP_TIMEWAIT) {
2783 				if (intotw(inp) != NULL)
2784 					crerr = cr_cansee(req->td->td_ucred,
2785 					    intotw(inp)->tw_cred);
2786 				else
2787 					crerr = EINVAL;	/* Skip this inp. */
2788 			} else
2789 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2790 			if (crerr == 0) {
2791 				struct xtcpcb xt;
2792 
2793 				tcp_inptoxtp(inp, &xt);
2794 				INP_RUNLOCK(inp);
2795 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2796 				if (error)
2797 					break;
2798 				else
2799 					continue;
2800 			}
2801 		}
2802 		INP_RUNLOCK(inp);
2803 	}
2804 	NET_EPOCH_EXIT(et);
2805 
2806 	if (!error) {
2807 		/*
2808 		 * Give the user an updated idea of our state.
2809 		 * If the generation differs from what we told
2810 		 * her before, she knows that something happened
2811 		 * while we were processing this request, and it
2812 		 * might be necessary to retry.
2813 		 */
2814 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2815 		xig.xig_sogen = so_gencnt;
2816 		xig.xig_count = V_tcbinfo.ipi_count +
2817 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2818 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2819 	}
2820 
2821 	return (error);
2822 }
2823 
2824 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2825     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2826     NULL, 0, tcp_pcblist, "S,xtcpcb",
2827     "List of active TCP connections");
2828 
2829 #ifdef INET
2830 static int
2831 tcp_getcred(SYSCTL_HANDLER_ARGS)
2832 {
2833 	struct xucred xuc;
2834 	struct sockaddr_in addrs[2];
2835 	struct epoch_tracker et;
2836 	struct inpcb *inp;
2837 	int error;
2838 
2839 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2840 	if (error)
2841 		return (error);
2842 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2843 	if (error)
2844 		return (error);
2845 	NET_EPOCH_ENTER(et);
2846 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2847 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2848 	NET_EPOCH_EXIT(et);
2849 	if (inp != NULL) {
2850 		if (inp->inp_socket == NULL)
2851 			error = ENOENT;
2852 		if (error == 0)
2853 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2854 		if (error == 0)
2855 			cru2x(inp->inp_cred, &xuc);
2856 		INP_RUNLOCK(inp);
2857 	} else
2858 		error = ENOENT;
2859 	if (error == 0)
2860 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2861 	return (error);
2862 }
2863 
2864 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2865     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2866     0, 0, tcp_getcred, "S,xucred",
2867     "Get the xucred of a TCP connection");
2868 #endif /* INET */
2869 
2870 #ifdef INET6
2871 static int
2872 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2873 {
2874 	struct epoch_tracker et;
2875 	struct xucred xuc;
2876 	struct sockaddr_in6 addrs[2];
2877 	struct inpcb *inp;
2878 	int error;
2879 #ifdef INET
2880 	int mapped = 0;
2881 #endif
2882 
2883 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2884 	if (error)
2885 		return (error);
2886 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2887 	if (error)
2888 		return (error);
2889 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2890 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2891 		return (error);
2892 	}
2893 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2894 #ifdef INET
2895 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2896 			mapped = 1;
2897 		else
2898 #endif
2899 			return (EINVAL);
2900 	}
2901 
2902 	NET_EPOCH_ENTER(et);
2903 #ifdef INET
2904 	if (mapped == 1)
2905 		inp = in_pcblookup(&V_tcbinfo,
2906 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2907 			addrs[1].sin6_port,
2908 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2909 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2910 	else
2911 #endif
2912 		inp = in6_pcblookup(&V_tcbinfo,
2913 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2914 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2915 			INPLOOKUP_RLOCKPCB, NULL);
2916 	NET_EPOCH_EXIT(et);
2917 	if (inp != NULL) {
2918 		if (inp->inp_socket == NULL)
2919 			error = ENOENT;
2920 		if (error == 0)
2921 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2922 		if (error == 0)
2923 			cru2x(inp->inp_cred, &xuc);
2924 		INP_RUNLOCK(inp);
2925 	} else
2926 		error = ENOENT;
2927 	if (error == 0)
2928 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2929 	return (error);
2930 }
2931 
2932 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2933     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2934     0, 0, tcp6_getcred, "S,xucred",
2935     "Get the xucred of a TCP6 connection");
2936 #endif /* INET6 */
2937 
2938 #ifdef INET
2939 /* Path MTU to try next when a fragmentation-needed message is received. */
2940 static inline int
2941 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2942 {
2943 	int mtu = ntohs(icp->icmp_nextmtu);
2944 
2945 	/* If no alternative MTU was proposed, try the next smaller one. */
2946 	if (!mtu)
2947 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2948 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2949 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2950 
2951 	return (mtu);
2952 }
2953 
2954 static void
2955 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2956 {
2957 	struct ip *ip = vip;
2958 	struct tcphdr *th;
2959 	struct in_addr faddr;
2960 	struct inpcb *inp;
2961 	struct tcpcb *tp;
2962 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2963 	struct icmp *icp;
2964 	struct in_conninfo inc;
2965 	tcp_seq icmp_tcp_seq;
2966 	int mtu;
2967 
2968 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2969 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2970 		return;
2971 
2972 	if (cmd == PRC_MSGSIZE)
2973 		notify = tcp_mtudisc_notify;
2974 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2975 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2976 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2977 		notify = tcp_drop_syn_sent;
2978 
2979 	/*
2980 	 * Hostdead is ugly because it goes linearly through all PCBs.
2981 	 * XXX: We never get this from ICMP, otherwise it makes an
2982 	 * excellent DoS attack on machines with many connections.
2983 	 */
2984 	else if (cmd == PRC_HOSTDEAD)
2985 		ip = NULL;
2986 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2987 		return;
2988 
2989 	if (ip == NULL) {
2990 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2991 		return;
2992 	}
2993 
2994 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2995 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2996 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2997 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2998 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2999 		/* signal EHOSTDOWN, as it flushes the cached route */
3000 		inp = (*notify)(inp, EHOSTDOWN);
3001 		goto out;
3002 	}
3003 	icmp_tcp_seq = th->th_seq;
3004 	if (inp != NULL)  {
3005 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3006 		    !(inp->inp_flags & INP_DROPPED) &&
3007 		    !(inp->inp_socket == NULL)) {
3008 			tp = intotcpcb(inp);
3009 #ifdef TCP_OFFLOAD
3010 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3011 				/*
3012 				 * MTU discovery for offloaded connections.  Let
3013 				 * the TOE driver verify seq# and process it.
3014 				 */
3015 				mtu = tcp_next_pmtu(icp, ip);
3016 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3017 				goto out;
3018 			}
3019 #endif
3020 			if (tp->t_port != port) {
3021 				goto out;
3022 			}
3023 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3024 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3025 				if (cmd == PRC_MSGSIZE) {
3026 					/*
3027 					 * MTU discovery: we got a needfrag and
3028 					 * will potentially try a lower MTU.
3029 					 */
3030 					mtu = tcp_next_pmtu(icp, ip);
3031 
3032 					/*
3033 					 * Only process the offered MTU if it
3034 					 * is smaller than the current one.
3035 					 */
3036 					if (mtu < tp->t_maxseg +
3037 					    sizeof(struct tcpiphdr)) {
3038 						bzero(&inc, sizeof(inc));
3039 						inc.inc_faddr = faddr;
3040 						inc.inc_fibnum =
3041 						    inp->inp_inc.inc_fibnum;
3042 						tcp_hc_updatemtu(&inc, mtu);
3043 						tcp_mtudisc(inp, mtu);
3044 					}
3045 				} else
3046 					inp = (*notify)(inp,
3047 					    inetctlerrmap[cmd]);
3048 			}
3049 		}
3050 	} else {
3051 		bzero(&inc, sizeof(inc));
3052 		inc.inc_fport = th->th_dport;
3053 		inc.inc_lport = th->th_sport;
3054 		inc.inc_faddr = faddr;
3055 		inc.inc_laddr = ip->ip_src;
3056 		syncache_unreach(&inc, icmp_tcp_seq, port);
3057 	}
3058 out:
3059 	if (inp != NULL)
3060 		INP_WUNLOCK(inp);
3061 }
3062 
3063 void
3064 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
3065 {
3066 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
3067 }
3068 
3069 void
3070 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
3071 {
3072 	/* Its a tunneled TCP over UDP icmp */
3073 	struct ip *outer_ip, *inner_ip;
3074 	struct icmp *icmp;
3075 	struct udphdr *udp;
3076 	struct tcphdr *th, ttemp;
3077 	int i_hlen, o_len;
3078 	uint16_t port;
3079 
3080 	inner_ip = (struct ip *)vip;
3081 	icmp = (struct icmp *)((caddr_t)inner_ip -
3082 	    (sizeof(struct icmp) - sizeof(struct ip)));
3083 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
3084 	i_hlen = inner_ip->ip_hl << 2;
3085 	o_len = ntohs(outer_ip->ip_len);
3086 	if (o_len <
3087 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
3088 		/* Not enough data present */
3089 		return;
3090 	}
3091 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
3092 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
3093 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3094 		return;
3095 	}
3096 	port = udp->uh_dport;
3097 	th = (struct tcphdr *)(udp + 1);
3098 	memcpy(&ttemp, th, sizeof(struct tcphdr));
3099 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
3100 	/* Now adjust down the size of the outer IP header */
3101 	o_len -= sizeof(struct udphdr);
3102 	outer_ip->ip_len = htons(o_len);
3103 	/* Now call in to the normal handling code */
3104 	tcp_ctlinput_with_port(cmd, sa, vip, port);
3105 }
3106 #endif /* INET */
3107 
3108 #ifdef INET6
3109 static inline int
3110 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3111 {
3112 	int mtu = ntohl(icmp6->icmp6_mtu);
3113 
3114 	/*
3115 	 * If no alternative MTU was proposed, or the proposed MTU was too
3116 	 * small, set to the min.
3117 	 */
3118 	if (mtu < IPV6_MMTU)
3119 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
3120 	return (mtu);
3121 }
3122 
3123 static void
3124 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3125 {
3126 	struct in6_addr *dst;
3127 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3128 	struct ip6_hdr *ip6;
3129 	struct mbuf *m;
3130 	struct inpcb *inp;
3131 	struct tcpcb *tp;
3132 	struct icmp6_hdr *icmp6;
3133 	struct ip6ctlparam *ip6cp = NULL;
3134 	const struct sockaddr_in6 *sa6_src = NULL;
3135 	struct in_conninfo inc;
3136 	struct tcp_ports {
3137 		uint16_t th_sport;
3138 		uint16_t th_dport;
3139 	} t_ports;
3140 	tcp_seq icmp_tcp_seq;
3141 	unsigned int mtu;
3142 	unsigned int off;
3143 
3144 	if (sa->sa_family != AF_INET6 ||
3145 	    sa->sa_len != sizeof(struct sockaddr_in6))
3146 		return;
3147 
3148 	/* if the parameter is from icmp6, decode it. */
3149 	if (d != NULL) {
3150 		ip6cp = (struct ip6ctlparam *)d;
3151 		icmp6 = ip6cp->ip6c_icmp6;
3152 		m = ip6cp->ip6c_m;
3153 		ip6 = ip6cp->ip6c_ip6;
3154 		off = ip6cp->ip6c_off;
3155 		sa6_src = ip6cp->ip6c_src;
3156 		dst = ip6cp->ip6c_finaldst;
3157 	} else {
3158 		m = NULL;
3159 		ip6 = NULL;
3160 		off = 0;	/* fool gcc */
3161 		sa6_src = &sa6_any;
3162 		dst = NULL;
3163 	}
3164 
3165 	if (cmd == PRC_MSGSIZE)
3166 		notify = tcp_mtudisc_notify;
3167 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3168 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3169 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3170 		notify = tcp_drop_syn_sent;
3171 
3172 	/*
3173 	 * Hostdead is ugly because it goes linearly through all PCBs.
3174 	 * XXX: We never get this from ICMP, otherwise it makes an
3175 	 * excellent DoS attack on machines with many connections.
3176 	 */
3177 	else if (cmd == PRC_HOSTDEAD)
3178 		ip6 = NULL;
3179 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3180 		return;
3181 
3182 	if (ip6 == NULL) {
3183 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3184 			      (const struct sockaddr *)sa6_src,
3185 			      0, cmd, NULL, notify);
3186 		return;
3187 	}
3188 
3189 	/* Check if we can safely get the ports from the tcp hdr */
3190 	if (m == NULL ||
3191 	    (m->m_pkthdr.len <
3192 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3193 		return;
3194 	}
3195 	bzero(&t_ports, sizeof(struct tcp_ports));
3196 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3197 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3198 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3199 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3200 		/* signal EHOSTDOWN, as it flushes the cached route */
3201 		inp = (*notify)(inp, EHOSTDOWN);
3202 		goto out;
3203 	}
3204 	off += sizeof(struct tcp_ports);
3205 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3206 		goto out;
3207 	}
3208 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3209 	if (inp != NULL)  {
3210 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3211 		    !(inp->inp_flags & INP_DROPPED) &&
3212 		    !(inp->inp_socket == NULL)) {
3213 			tp = intotcpcb(inp);
3214 #ifdef TCP_OFFLOAD
3215 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3216 				/* MTU discovery for offloaded connections. */
3217 				mtu = tcp6_next_pmtu(icmp6);
3218 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3219 				goto out;
3220 			}
3221 #endif
3222 			if (tp->t_port != port) {
3223 				goto out;
3224 			}
3225 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3226 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3227 				if (cmd == PRC_MSGSIZE) {
3228 					/*
3229 					 * MTU discovery:
3230 					 * If we got a needfrag set the MTU
3231 					 * in the route to the suggested new
3232 					 * value (if given) and then notify.
3233 					 */
3234 					mtu = tcp6_next_pmtu(icmp6);
3235 
3236 					bzero(&inc, sizeof(inc));
3237 					inc.inc_fibnum = M_GETFIB(m);
3238 					inc.inc_flags |= INC_ISIPV6;
3239 					inc.inc6_faddr = *dst;
3240 					if (in6_setscope(&inc.inc6_faddr,
3241 						m->m_pkthdr.rcvif, NULL))
3242 						goto out;
3243 					/*
3244 					 * Only process the offered MTU if it
3245 					 * is smaller than the current one.
3246 					 */
3247 					if (mtu < tp->t_maxseg +
3248 					    sizeof (struct tcphdr) +
3249 					    sizeof (struct ip6_hdr)) {
3250 						tcp_hc_updatemtu(&inc, mtu);
3251 						tcp_mtudisc(inp, mtu);
3252 						ICMP6STAT_INC(icp6s_pmtuchg);
3253 					}
3254 				} else
3255 					inp = (*notify)(inp,
3256 					    inet6ctlerrmap[cmd]);
3257 			}
3258 		}
3259 	} else {
3260 		bzero(&inc, sizeof(inc));
3261 		inc.inc_fibnum = M_GETFIB(m);
3262 		inc.inc_flags |= INC_ISIPV6;
3263 		inc.inc_fport = t_ports.th_dport;
3264 		inc.inc_lport = t_ports.th_sport;
3265 		inc.inc6_faddr = *dst;
3266 		inc.inc6_laddr = ip6->ip6_src;
3267 		syncache_unreach(&inc, icmp_tcp_seq, port);
3268 	}
3269 out:
3270 	if (inp != NULL)
3271 		INP_WUNLOCK(inp);
3272 }
3273 
3274 void
3275 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3276 {
3277 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3278 }
3279 
3280 void
3281 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3282 {
3283 	struct ip6ctlparam *ip6cp;
3284 	struct mbuf *m;
3285 	struct udphdr *udp;
3286 	uint16_t port;
3287 
3288 	ip6cp = (struct ip6ctlparam *)d;
3289 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3290 	if (m == NULL) {
3291 		return;
3292 	}
3293 	udp = mtod(m, struct udphdr *);
3294 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3295 		return;
3296 	}
3297 	port = udp->uh_dport;
3298 	m_adj(m, sizeof(struct udphdr));
3299 	if ((m->m_flags & M_PKTHDR) == 0) {
3300 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3301 	}
3302 	/* Now call in to the normal handling code */
3303 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3304 }
3305 
3306 #endif /* INET6 */
3307 
3308 static uint32_t
3309 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3310 {
3311 	SIPHASH_CTX ctx;
3312 	uint32_t hash[2];
3313 
3314 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3315 	    ("%s: keylen %u too short ", __func__, len));
3316 	SipHash24_Init(&ctx);
3317 	SipHash_SetKey(&ctx, (uint8_t *)key);
3318 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3319 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3320 	switch (inc->inc_flags & INC_ISIPV6) {
3321 #ifdef INET
3322 	case 0:
3323 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3324 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3325 		break;
3326 #endif
3327 #ifdef INET6
3328 	case INC_ISIPV6:
3329 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3330 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3331 		break;
3332 #endif
3333 	}
3334 	SipHash_Final((uint8_t *)hash, &ctx);
3335 
3336 	return (hash[0] ^ hash[1]);
3337 }
3338 
3339 uint32_t
3340 tcp_new_ts_offset(struct in_conninfo *inc)
3341 {
3342 	struct in_conninfo inc_store, *local_inc;
3343 
3344 	if (!V_tcp_ts_offset_per_conn) {
3345 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3346 		inc_store.inc_lport = 0;
3347 		inc_store.inc_fport = 0;
3348 		local_inc = &inc_store;
3349 	} else {
3350 		local_inc = inc;
3351 	}
3352 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3353 	    sizeof(V_ts_offset_secret)));
3354 }
3355 
3356 /*
3357  * Following is where TCP initial sequence number generation occurs.
3358  *
3359  * There are two places where we must use initial sequence numbers:
3360  * 1.  In SYN-ACK packets.
3361  * 2.  In SYN packets.
3362  *
3363  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3364  * tcp_syncache.c for details.
3365  *
3366  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3367  * depends on this property.  In addition, these ISNs should be
3368  * unguessable so as to prevent connection hijacking.  To satisfy
3369  * the requirements of this situation, the algorithm outlined in
3370  * RFC 1948 is used, with only small modifications.
3371  *
3372  * Implementation details:
3373  *
3374  * Time is based off the system timer, and is corrected so that it
3375  * increases by one megabyte per second.  This allows for proper
3376  * recycling on high speed LANs while still leaving over an hour
3377  * before rollover.
3378  *
3379  * As reading the *exact* system time is too expensive to be done
3380  * whenever setting up a TCP connection, we increment the time
3381  * offset in two ways.  First, a small random positive increment
3382  * is added to isn_offset for each connection that is set up.
3383  * Second, the function tcp_isn_tick fires once per clock tick
3384  * and increments isn_offset as necessary so that sequence numbers
3385  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3386  * random positive increments serve only to ensure that the same
3387  * exact sequence number is never sent out twice (as could otherwise
3388  * happen when a port is recycled in less than the system tick
3389  * interval.)
3390  *
3391  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3392  * between seeding of isn_secret.  This is normally set to zero,
3393  * as reseeding should not be necessary.
3394  *
3395  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3396  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3397  * general, this means holding an exclusive (write) lock.
3398  */
3399 
3400 #define ISN_BYTES_PER_SECOND 1048576
3401 #define ISN_STATIC_INCREMENT 4096
3402 #define ISN_RANDOM_INCREMENT (4096 - 1)
3403 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3404 
3405 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3406 VNET_DEFINE_STATIC(int, isn_last);
3407 VNET_DEFINE_STATIC(int, isn_last_reseed);
3408 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3409 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3410 
3411 #define	V_isn_secret			VNET(isn_secret)
3412 #define	V_isn_last			VNET(isn_last)
3413 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3414 #define	V_isn_offset			VNET(isn_offset)
3415 #define	V_isn_offset_old		VNET(isn_offset_old)
3416 
3417 tcp_seq
3418 tcp_new_isn(struct in_conninfo *inc)
3419 {
3420 	tcp_seq new_isn;
3421 	u_int32_t projected_offset;
3422 
3423 	ISN_LOCK();
3424 	/* Seed if this is the first use, reseed if requested. */
3425 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3426 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3427 		< (u_int)ticks))) {
3428 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3429 		V_isn_last_reseed = ticks;
3430 	}
3431 
3432 	/* Compute the hash and return the ISN. */
3433 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3434 	    sizeof(V_isn_secret));
3435 	V_isn_offset += ISN_STATIC_INCREMENT +
3436 		(arc4random() & ISN_RANDOM_INCREMENT);
3437 	if (ticks != V_isn_last) {
3438 		projected_offset = V_isn_offset_old +
3439 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3440 		if (SEQ_GT(projected_offset, V_isn_offset))
3441 			V_isn_offset = projected_offset;
3442 		V_isn_offset_old = V_isn_offset;
3443 		V_isn_last = ticks;
3444 	}
3445 	new_isn += V_isn_offset;
3446 	ISN_UNLOCK();
3447 	return (new_isn);
3448 }
3449 
3450 /*
3451  * When a specific ICMP unreachable message is received and the
3452  * connection state is SYN-SENT, drop the connection.  This behavior
3453  * is controlled by the icmp_may_rst sysctl.
3454  */
3455 struct inpcb *
3456 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3457 {
3458 	struct tcpcb *tp;
3459 
3460 	NET_EPOCH_ASSERT();
3461 	INP_WLOCK_ASSERT(inp);
3462 
3463 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3464 	    (inp->inp_flags & INP_DROPPED))
3465 		return (inp);
3466 
3467 	tp = intotcpcb(inp);
3468 	if (tp->t_state != TCPS_SYN_SENT)
3469 		return (inp);
3470 
3471 	if (IS_FASTOPEN(tp->t_flags))
3472 		tcp_fastopen_disable_path(tp);
3473 
3474 	tp = tcp_drop(tp, errno);
3475 	if (tp != NULL)
3476 		return (inp);
3477 	else
3478 		return (NULL);
3479 }
3480 
3481 /*
3482  * When `need fragmentation' ICMP is received, update our idea of the MSS
3483  * based on the new value. Also nudge TCP to send something, since we
3484  * know the packet we just sent was dropped.
3485  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3486  */
3487 static struct inpcb *
3488 tcp_mtudisc_notify(struct inpcb *inp, int error)
3489 {
3490 
3491 	tcp_mtudisc(inp, -1);
3492 	return (inp);
3493 }
3494 
3495 static void
3496 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3497 {
3498 	struct tcpcb *tp;
3499 	struct socket *so;
3500 
3501 	INP_WLOCK_ASSERT(inp);
3502 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3503 	    (inp->inp_flags & INP_DROPPED))
3504 		return;
3505 
3506 	tp = intotcpcb(inp);
3507 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3508 
3509 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3510 
3511 	so = inp->inp_socket;
3512 	SOCKBUF_LOCK(&so->so_snd);
3513 	/* If the mss is larger than the socket buffer, decrease the mss. */
3514 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3515 		tp->t_maxseg = so->so_snd.sb_hiwat;
3516 	SOCKBUF_UNLOCK(&so->so_snd);
3517 
3518 	TCPSTAT_INC(tcps_mturesent);
3519 	tp->t_rtttime = 0;
3520 	tp->snd_nxt = tp->snd_una;
3521 	tcp_free_sackholes(tp);
3522 	tp->snd_recover = tp->snd_max;
3523 	if (tp->t_flags & TF_SACK_PERMIT)
3524 		EXIT_FASTRECOVERY(tp->t_flags);
3525 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3526 		/*
3527 		 * Conceptually the snd_nxt setting
3528 		 * and freeing sack holes should
3529 		 * be done by the default stacks
3530 		 * own tfb_tcp_mtu_chg().
3531 		 */
3532 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3533 	}
3534 	tp->t_fb->tfb_tcp_output(tp);
3535 }
3536 
3537 #ifdef INET
3538 /*
3539  * Look-up the routing entry to the peer of this inpcb.  If no route
3540  * is found and it cannot be allocated, then return 0.  This routine
3541  * is called by TCP routines that access the rmx structure and by
3542  * tcp_mss_update to get the peer/interface MTU.
3543  */
3544 uint32_t
3545 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3546 {
3547 	struct nhop_object *nh;
3548 	struct ifnet *ifp;
3549 	uint32_t maxmtu = 0;
3550 
3551 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3552 
3553 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3554 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3555 		if (nh == NULL)
3556 			return (0);
3557 
3558 		ifp = nh->nh_ifp;
3559 		maxmtu = nh->nh_mtu;
3560 
3561 		/* Report additional interface capabilities. */
3562 		if (cap != NULL) {
3563 			if (ifp->if_capenable & IFCAP_TSO4 &&
3564 			    ifp->if_hwassist & CSUM_TSO) {
3565 				cap->ifcap |= CSUM_TSO;
3566 				cap->tsomax = ifp->if_hw_tsomax;
3567 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3568 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3569 			}
3570 		}
3571 	}
3572 	return (maxmtu);
3573 }
3574 #endif /* INET */
3575 
3576 #ifdef INET6
3577 uint32_t
3578 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3579 {
3580 	struct nhop_object *nh;
3581 	struct in6_addr dst6;
3582 	uint32_t scopeid;
3583 	struct ifnet *ifp;
3584 	uint32_t maxmtu = 0;
3585 
3586 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3587 
3588 	if (inc->inc_flags & INC_IPV6MINMTU)
3589 		return (IPV6_MMTU);
3590 
3591 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3592 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3593 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3594 		if (nh == NULL)
3595 			return (0);
3596 
3597 		ifp = nh->nh_ifp;
3598 		maxmtu = nh->nh_mtu;
3599 
3600 		/* Report additional interface capabilities. */
3601 		if (cap != NULL) {
3602 			if (ifp->if_capenable & IFCAP_TSO6 &&
3603 			    ifp->if_hwassist & CSUM_TSO) {
3604 				cap->ifcap |= CSUM_TSO;
3605 				cap->tsomax = ifp->if_hw_tsomax;
3606 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3607 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3608 			}
3609 		}
3610 	}
3611 
3612 	return (maxmtu);
3613 }
3614 
3615 /*
3616  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3617  *
3618  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3619  * The right place to do that is ip6_setpktopt() that has just been
3620  * executed.  By the way it just filled ip6po_minmtu for us.
3621  */
3622 void
3623 tcp6_use_min_mtu(struct tcpcb *tp)
3624 {
3625 	struct inpcb *inp = tp->t_inpcb;
3626 
3627 	INP_WLOCK_ASSERT(inp);
3628 	/*
3629 	 * In case of the IPV6_USE_MIN_MTU socket
3630 	 * option, the INC_IPV6MINMTU flag to announce
3631 	 * a corresponding MSS during the initial
3632 	 * handshake.  If the TCP connection is not in
3633 	 * the front states, just reduce the MSS being
3634 	 * used.  This avoids the sending of TCP
3635 	 * segments which will be fragmented at the
3636 	 * IPv6 layer.
3637 	 */
3638 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3639 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3640 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3641 		struct ip6_pktopts *opt;
3642 
3643 		opt = inp->in6p_outputopts;
3644 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3645 		    tp->t_maxseg > TCP6_MSS)
3646 			tp->t_maxseg = TCP6_MSS;
3647 	}
3648 }
3649 #endif /* INET6 */
3650 
3651 /*
3652  * Calculate effective SMSS per RFC5681 definition for a given TCP
3653  * connection at its current state, taking into account SACK and etc.
3654  */
3655 u_int
3656 tcp_maxseg(const struct tcpcb *tp)
3657 {
3658 	u_int optlen;
3659 
3660 	if (tp->t_flags & TF_NOOPT)
3661 		return (tp->t_maxseg);
3662 
3663 	/*
3664 	 * Here we have a simplified code from tcp_addoptions(),
3665 	 * without a proper loop, and having most of paddings hardcoded.
3666 	 * We might make mistakes with padding here in some edge cases,
3667 	 * but this is harmless, since result of tcp_maxseg() is used
3668 	 * only in cwnd and ssthresh estimations.
3669 	 */
3670 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3671 		if (tp->t_flags & TF_RCVD_TSTMP)
3672 			optlen = TCPOLEN_TSTAMP_APPA;
3673 		else
3674 			optlen = 0;
3675 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3676 		if (tp->t_flags & TF_SIGNATURE)
3677 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3678 #endif
3679 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3680 			optlen += TCPOLEN_SACKHDR;
3681 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3682 			optlen = PADTCPOLEN(optlen);
3683 		}
3684 	} else {
3685 		if (tp->t_flags & TF_REQ_TSTMP)
3686 			optlen = TCPOLEN_TSTAMP_APPA;
3687 		else
3688 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3689 		if (tp->t_flags & TF_REQ_SCALE)
3690 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3691 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3692 		if (tp->t_flags & TF_SIGNATURE)
3693 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3694 #endif
3695 		if (tp->t_flags & TF_SACK_PERMIT)
3696 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3697 	}
3698 #undef PAD
3699 	optlen = min(optlen, TCP_MAXOLEN);
3700 	return (tp->t_maxseg - optlen);
3701 }
3702 
3703 
3704 u_int
3705 tcp_fixed_maxseg(const struct tcpcb *tp)
3706 {
3707 	int optlen;
3708 
3709 	if (tp->t_flags & TF_NOOPT)
3710 		return (tp->t_maxseg);
3711 
3712 	/*
3713 	 * Here we have a simplified code from tcp_addoptions(),
3714 	 * without a proper loop, and having most of paddings hardcoded.
3715 	 * We only consider fixed options that we would send every
3716 	 * time I.e. SACK is not considered. This is important
3717 	 * for cc modules to figure out what the modulo of the
3718 	 * cwnd should be.
3719 	 */
3720 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3721 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3722 		if (tp->t_flags & TF_RCVD_TSTMP)
3723 			optlen = TCPOLEN_TSTAMP_APPA;
3724 		else
3725 			optlen = 0;
3726 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3727 		if (tp->t_flags & TF_SIGNATURE)
3728 			optlen += PAD(TCPOLEN_SIGNATURE);
3729 #endif
3730 	} else {
3731 		if (tp->t_flags & TF_REQ_TSTMP)
3732 			optlen = TCPOLEN_TSTAMP_APPA;
3733 		else
3734 			optlen = PAD(TCPOLEN_MAXSEG);
3735 		if (tp->t_flags & TF_REQ_SCALE)
3736 			optlen += PAD(TCPOLEN_WINDOW);
3737 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3738 		if (tp->t_flags & TF_SIGNATURE)
3739 			optlen += PAD(TCPOLEN_SIGNATURE);
3740 #endif
3741 		if (tp->t_flags & TF_SACK_PERMIT)
3742 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3743 	}
3744 #undef PAD
3745 	optlen = min(optlen, TCP_MAXOLEN);
3746 	return (tp->t_maxseg - optlen);
3747 }
3748 
3749 
3750 
3751 static int
3752 sysctl_drop(SYSCTL_HANDLER_ARGS)
3753 {
3754 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3755 	struct sockaddr_storage addrs[2];
3756 	struct inpcb *inp;
3757 	struct tcpcb *tp;
3758 	struct tcptw *tw;
3759 	struct sockaddr_in *fin, *lin;
3760 	struct epoch_tracker et;
3761 #ifdef INET6
3762 	struct sockaddr_in6 *fin6, *lin6;
3763 #endif
3764 	int error;
3765 
3766 	inp = NULL;
3767 	fin = lin = NULL;
3768 #ifdef INET6
3769 	fin6 = lin6 = NULL;
3770 #endif
3771 	error = 0;
3772 
3773 	if (req->oldptr != NULL || req->oldlen != 0)
3774 		return (EINVAL);
3775 	if (req->newptr == NULL)
3776 		return (EPERM);
3777 	if (req->newlen < sizeof(addrs))
3778 		return (ENOMEM);
3779 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3780 	if (error)
3781 		return (error);
3782 
3783 	switch (addrs[0].ss_family) {
3784 #ifdef INET6
3785 	case AF_INET6:
3786 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3787 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3788 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3789 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3790 			return (EINVAL);
3791 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3792 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3793 				return (EINVAL);
3794 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3795 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3796 			fin = (struct sockaddr_in *)&addrs[0];
3797 			lin = (struct sockaddr_in *)&addrs[1];
3798 			break;
3799 		}
3800 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3801 		if (error)
3802 			return (error);
3803 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3804 		if (error)
3805 			return (error);
3806 		break;
3807 #endif
3808 #ifdef INET
3809 	case AF_INET:
3810 		fin = (struct sockaddr_in *)&addrs[0];
3811 		lin = (struct sockaddr_in *)&addrs[1];
3812 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3813 		    lin->sin_len != sizeof(struct sockaddr_in))
3814 			return (EINVAL);
3815 		break;
3816 #endif
3817 	default:
3818 		return (EINVAL);
3819 	}
3820 	NET_EPOCH_ENTER(et);
3821 	switch (addrs[0].ss_family) {
3822 #ifdef INET6
3823 	case AF_INET6:
3824 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3825 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3826 		    INPLOOKUP_WLOCKPCB, NULL);
3827 		break;
3828 #endif
3829 #ifdef INET
3830 	case AF_INET:
3831 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3832 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3833 		break;
3834 #endif
3835 	}
3836 	if (inp != NULL) {
3837 		if (inp->inp_flags & INP_TIMEWAIT) {
3838 			/*
3839 			 * XXXRW: There currently exists a state where an
3840 			 * inpcb is present, but its timewait state has been
3841 			 * discarded.  For now, don't allow dropping of this
3842 			 * type of inpcb.
3843 			 */
3844 			tw = intotw(inp);
3845 			if (tw != NULL)
3846 				tcp_twclose(tw, 0);
3847 			else
3848 				INP_WUNLOCK(inp);
3849 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3850 		    !SOLISTENING(inp->inp_socket)) {
3851 			tp = intotcpcb(inp);
3852 			tp = tcp_drop(tp, ECONNABORTED);
3853 			if (tp != NULL)
3854 				INP_WUNLOCK(inp);
3855 		} else
3856 			INP_WUNLOCK(inp);
3857 	} else
3858 		error = ESRCH;
3859 	NET_EPOCH_EXIT(et);
3860 	return (error);
3861 }
3862 
3863 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3864     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3865     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3866     "Drop TCP connection");
3867 
3868 #ifdef KERN_TLS
3869 static int
3870 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3871 {
3872 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3873 	struct sockaddr_storage addrs[2];
3874 	struct inpcb *inp;
3875 	struct sockaddr_in *fin, *lin;
3876 	struct epoch_tracker et;
3877 #ifdef INET6
3878 	struct sockaddr_in6 *fin6, *lin6;
3879 #endif
3880 	int error;
3881 
3882 	inp = NULL;
3883 	fin = lin = NULL;
3884 #ifdef INET6
3885 	fin6 = lin6 = NULL;
3886 #endif
3887 	error = 0;
3888 
3889 	if (req->oldptr != NULL || req->oldlen != 0)
3890 		return (EINVAL);
3891 	if (req->newptr == NULL)
3892 		return (EPERM);
3893 	if (req->newlen < sizeof(addrs))
3894 		return (ENOMEM);
3895 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3896 	if (error)
3897 		return (error);
3898 
3899 	switch (addrs[0].ss_family) {
3900 #ifdef INET6
3901 	case AF_INET6:
3902 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3903 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3904 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3905 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3906 			return (EINVAL);
3907 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3908 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3909 				return (EINVAL);
3910 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3911 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3912 			fin = (struct sockaddr_in *)&addrs[0];
3913 			lin = (struct sockaddr_in *)&addrs[1];
3914 			break;
3915 		}
3916 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3917 		if (error)
3918 			return (error);
3919 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3920 		if (error)
3921 			return (error);
3922 		break;
3923 #endif
3924 #ifdef INET
3925 	case AF_INET:
3926 		fin = (struct sockaddr_in *)&addrs[0];
3927 		lin = (struct sockaddr_in *)&addrs[1];
3928 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3929 		    lin->sin_len != sizeof(struct sockaddr_in))
3930 			return (EINVAL);
3931 		break;
3932 #endif
3933 	default:
3934 		return (EINVAL);
3935 	}
3936 	NET_EPOCH_ENTER(et);
3937 	switch (addrs[0].ss_family) {
3938 #ifdef INET6
3939 	case AF_INET6:
3940 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3941 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3942 		    INPLOOKUP_WLOCKPCB, NULL);
3943 		break;
3944 #endif
3945 #ifdef INET
3946 	case AF_INET:
3947 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3948 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3949 		break;
3950 #endif
3951 	}
3952 	NET_EPOCH_EXIT(et);
3953 	if (inp != NULL) {
3954 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3955 		    inp->inp_socket == NULL) {
3956 			error = ECONNRESET;
3957 			INP_WUNLOCK(inp);
3958 		} else {
3959 			struct socket *so;
3960 
3961 			so = inp->inp_socket;
3962 			soref(so);
3963 			error = ktls_set_tx_mode(so,
3964 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3965 			INP_WUNLOCK(inp);
3966 			sorele(so);
3967 		}
3968 	} else
3969 		error = ESRCH;
3970 	return (error);
3971 }
3972 
3973 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3974     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3975     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3976     "Switch TCP connection to SW TLS");
3977 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3978     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3979     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3980     "Switch TCP connection to ifnet TLS");
3981 #endif
3982 
3983 /*
3984  * Generate a standardized TCP log line for use throughout the
3985  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3986  * allow use in the interrupt context.
3987  *
3988  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3989  * NB: The function may return NULL if memory allocation failed.
3990  *
3991  * Due to header inclusion and ordering limitations the struct ip
3992  * and ip6_hdr pointers have to be passed as void pointers.
3993  */
3994 char *
3995 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3996     const void *ip6hdr)
3997 {
3998 
3999 	/* Is logging enabled? */
4000 	if (V_tcp_log_in_vain == 0)
4001 		return (NULL);
4002 
4003 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
4004 }
4005 
4006 char *
4007 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
4008     const void *ip6hdr)
4009 {
4010 
4011 	/* Is logging enabled? */
4012 	if (tcp_log_debug == 0)
4013 		return (NULL);
4014 
4015 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
4016 }
4017 
4018 static char *
4019 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
4020     const void *ip6hdr)
4021 {
4022 	char *s, *sp;
4023 	size_t size;
4024 	struct ip *ip;
4025 #ifdef INET6
4026 	const struct ip6_hdr *ip6;
4027 
4028 	ip6 = (const struct ip6_hdr *)ip6hdr;
4029 #endif /* INET6 */
4030 	ip = (struct ip *)ip4hdr;
4031 
4032 	/*
4033 	 * The log line looks like this:
4034 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
4035 	 */
4036 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
4037 	    sizeof(PRINT_TH_FLAGS) + 1 +
4038 #ifdef INET6
4039 	    2 * INET6_ADDRSTRLEN;
4040 #else
4041 	    2 * INET_ADDRSTRLEN;
4042 #endif /* INET6 */
4043 
4044 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
4045 	if (s == NULL)
4046 		return (NULL);
4047 
4048 	strcat(s, "TCP: [");
4049 	sp = s + strlen(s);
4050 
4051 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
4052 		inet_ntoa_r(inc->inc_faddr, sp);
4053 		sp = s + strlen(s);
4054 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4055 		sp = s + strlen(s);
4056 		inet_ntoa_r(inc->inc_laddr, sp);
4057 		sp = s + strlen(s);
4058 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4059 #ifdef INET6
4060 	} else if (inc) {
4061 		ip6_sprintf(sp, &inc->inc6_faddr);
4062 		sp = s + strlen(s);
4063 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4064 		sp = s + strlen(s);
4065 		ip6_sprintf(sp, &inc->inc6_laddr);
4066 		sp = s + strlen(s);
4067 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4068 	} else if (ip6 && th) {
4069 		ip6_sprintf(sp, &ip6->ip6_src);
4070 		sp = s + strlen(s);
4071 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4072 		sp = s + strlen(s);
4073 		ip6_sprintf(sp, &ip6->ip6_dst);
4074 		sp = s + strlen(s);
4075 		sprintf(sp, "]:%i", ntohs(th->th_dport));
4076 #endif /* INET6 */
4077 #ifdef INET
4078 	} else if (ip && th) {
4079 		inet_ntoa_r(ip->ip_src, sp);
4080 		sp = s + strlen(s);
4081 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4082 		sp = s + strlen(s);
4083 		inet_ntoa_r(ip->ip_dst, sp);
4084 		sp = s + strlen(s);
4085 		sprintf(sp, "]:%i", ntohs(th->th_dport));
4086 #endif /* INET */
4087 	} else {
4088 		free(s, M_TCPLOG);
4089 		return (NULL);
4090 	}
4091 	sp = s + strlen(s);
4092 	if (th)
4093 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
4094 	if (*(s + size - 1) != '\0')
4095 		panic("%s: string too long", __func__);
4096 	return (s);
4097 }
4098 
4099 /*
4100  * A subroutine which makes it easy to track TCP state changes with DTrace.
4101  * This function shouldn't be called for t_state initializations that don't
4102  * correspond to actual TCP state transitions.
4103  */
4104 void
4105 tcp_state_change(struct tcpcb *tp, int newstate)
4106 {
4107 #if defined(KDTRACE_HOOKS)
4108 	int pstate = tp->t_state;
4109 #endif
4110 
4111 	TCPSTATES_DEC(tp->t_state);
4112 	TCPSTATES_INC(newstate);
4113 	tp->t_state = newstate;
4114 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4115 }
4116 
4117 /*
4118  * Create an external-format (``xtcpcb'') structure using the information in
4119  * the kernel-format tcpcb structure pointed to by tp.  This is done to
4120  * reduce the spew of irrelevant information over this interface, to isolate
4121  * user code from changes in the kernel structure, and potentially to provide
4122  * information-hiding if we decide that some of this information should be
4123  * hidden from users.
4124  */
4125 void
4126 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4127 {
4128 	struct tcpcb *tp = intotcpcb(inp);
4129 	struct tcptw *tw = intotw(inp);
4130 	sbintime_t now;
4131 
4132 	bzero(xt, sizeof(*xt));
4133 	if (inp->inp_flags & INP_TIMEWAIT) {
4134 		xt->t_state = TCPS_TIME_WAIT;
4135 		xt->xt_encaps_port = tw->t_port;
4136 	} else {
4137 		xt->t_state = tp->t_state;
4138 		xt->t_logstate = tp->t_logstate;
4139 		xt->t_flags = tp->t_flags;
4140 		xt->t_sndzerowin = tp->t_sndzerowin;
4141 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4142 		xt->t_rcvoopack = tp->t_rcvoopack;
4143 		xt->t_rcv_wnd = tp->rcv_wnd;
4144 		xt->t_snd_wnd = tp->snd_wnd;
4145 		xt->t_snd_cwnd = tp->snd_cwnd;
4146 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4147 		xt->t_dsack_bytes = tp->t_dsack_bytes;
4148 		xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4149 		xt->t_dsack_pack = tp->t_dsack_pack;
4150 		xt->t_maxseg = tp->t_maxseg;
4151 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4152 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4153 
4154 		now = getsbinuptime();
4155 #define	COPYTIMER(ttt)	do {						\
4156 		if (callout_active(&tp->t_timers->ttt))			\
4157 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4158 			    SBT_1MS;					\
4159 		else							\
4160 			xt->ttt = 0;					\
4161 } while (0)
4162 		COPYTIMER(tt_delack);
4163 		COPYTIMER(tt_rexmt);
4164 		COPYTIMER(tt_persist);
4165 		COPYTIMER(tt_keep);
4166 		COPYTIMER(tt_2msl);
4167 #undef COPYTIMER
4168 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4169 
4170 		xt->xt_encaps_port = tp->t_port;
4171 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4172 		    TCP_FUNCTION_NAME_LEN_MAX);
4173 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4174 		    TCP_CA_NAME_MAX);
4175 #ifdef TCP_BLACKBOX
4176 		(void)tcp_log_get_id(tp, xt->xt_logid);
4177 #endif
4178 	}
4179 
4180 	xt->xt_len = sizeof(struct xtcpcb);
4181 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4182 	if (inp->inp_socket == NULL)
4183 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4184 }
4185 
4186 void
4187 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4188 {
4189 	uint32_t bit, i;
4190 
4191 	if ((tp == NULL) ||
4192 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4193 	    (status == 0)) {
4194 		/* Invalid */
4195 		return;
4196 	}
4197 	if (status > (sizeof(uint32_t) * 8)) {
4198 		/* Should this be a KASSERT? */
4199 		return;
4200 	}
4201 	bit = 1U << (status - 1);
4202 	if (bit & tp->t_end_info_status) {
4203 		/* already logged */
4204 		return;
4205 	}
4206 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4207 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4208 			tp->t_end_info_bytes[i] = status;
4209 			tp->t_end_info_status |= bit;
4210 			break;
4211 		}
4212 	}
4213 }
4214 
4215 int
4216 tcp_can_enable_pacing(void)
4217 {
4218 
4219 	if ((tcp_pacing_limit == -1) ||
4220 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4221 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4222 		shadow_num_connections = number_of_tcp_connections_pacing;
4223 		return (1);
4224 	} else {
4225 		return (0);
4226 	}
4227 }
4228 
4229 static uint8_t tcp_pacing_warning = 0;
4230 
4231 void
4232 tcp_decrement_paced_conn(void)
4233 {
4234 	uint32_t ret;
4235 
4236 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4237 	shadow_num_connections = number_of_tcp_connections_pacing;
4238 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4239 	if (ret == 0) {
4240 		if (tcp_pacing_limit != -1) {
4241 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4242 			tcp_pacing_limit = 0;
4243 		} else if (tcp_pacing_warning == 0) {
4244 			printf("Warning pacing count is invalid, invalid decrement\n");
4245 			tcp_pacing_warning = 1;
4246 		}
4247 	}
4248 }
4249