xref: /freebsd/sys/netinet/tcp_subr.c (revision c94c8223bd444fa38ded3797060110c590f422f4)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
49 #include <sys/jail.h>
50 #include <sys/malloc.h>
51 #include <sys/refcount.h>
52 #include <sys/mbuf.h>
53 #ifdef INET6
54 #include <sys/domain.h>
55 #endif
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/sdt.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/protosw.h>
62 #include <sys/random.h>
63 
64 #include <vm/uma.h>
65 
66 #include <net/route.h>
67 #include <net/if.h>
68 #include <net/if_var.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/in_fib.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
80 #ifdef INET6
81 #include <netinet/ip6.h>
82 #include <netinet6/in6_fib.h>
83 #include <netinet6/in6_pcb.h>
84 #include <netinet6/ip6_var.h>
85 #include <netinet6/scope6_var.h>
86 #include <netinet6/nd6.h>
87 #endif
88 
89 #ifdef TCP_RFC7413
90 #include <netinet/tcp_fastopen.h>
91 #endif
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcp_syncache.h>
98 #include <netinet/cc/cc.h>
99 #ifdef INET6
100 #include <netinet6/tcp6_var.h>
101 #endif
102 #include <netinet/tcpip.h>
103 #ifdef TCPPCAP
104 #include <netinet/tcp_pcap.h>
105 #endif
106 #ifdef TCPDEBUG
107 #include <netinet/tcp_debug.h>
108 #endif
109 #ifdef INET6
110 #include <netinet6/ip6protosw.h>
111 #endif
112 #ifdef TCP_OFFLOAD
113 #include <netinet/tcp_offload.h>
114 #endif
115 
116 #ifdef IPSEC
117 #include <netipsec/ipsec.h>
118 #include <netipsec/xform.h>
119 #ifdef INET6
120 #include <netipsec/ipsec6.h>
121 #endif
122 #include <netipsec/key.h>
123 #include <sys/syslog.h>
124 #endif /*IPSEC*/
125 
126 #include <machine/in_cksum.h>
127 #include <sys/md5.h>
128 
129 #include <security/mac/mac_framework.h>
130 
131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
132 #ifdef INET6
133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
134 #endif
135 
136 struct rwlock tcp_function_lock;
137 
138 static int
139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
140 {
141 	int error, new;
142 
143 	new = V_tcp_mssdflt;
144 	error = sysctl_handle_int(oidp, &new, 0, req);
145 	if (error == 0 && req->newptr) {
146 		if (new < TCP_MINMSS)
147 			error = EINVAL;
148 		else
149 			V_tcp_mssdflt = new;
150 	}
151 	return (error);
152 }
153 
154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
155     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
156     &sysctl_net_inet_tcp_mss_check, "I",
157     "Default TCP Maximum Segment Size");
158 
159 #ifdef INET6
160 static int
161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
162 {
163 	int error, new;
164 
165 	new = V_tcp_v6mssdflt;
166 	error = sysctl_handle_int(oidp, &new, 0, req);
167 	if (error == 0 && req->newptr) {
168 		if (new < TCP_MINMSS)
169 			error = EINVAL;
170 		else
171 			V_tcp_v6mssdflt = new;
172 	}
173 	return (error);
174 }
175 
176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
177     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
178     &sysctl_net_inet_tcp_mss_v6_check, "I",
179    "Default TCP Maximum Segment Size for IPv6");
180 #endif /* INET6 */
181 
182 /*
183  * Minimum MSS we accept and use. This prevents DoS attacks where
184  * we are forced to a ridiculous low MSS like 20 and send hundreds
185  * of packets instead of one. The effect scales with the available
186  * bandwidth and quickly saturates the CPU and network interface
187  * with packet generation and sending. Set to zero to disable MINMSS
188  * checking. This setting prevents us from sending too small packets.
189  */
190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
192      &VNET_NAME(tcp_minmss), 0,
193     "Minimum TCP Maximum Segment Size");
194 
195 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
197     &VNET_NAME(tcp_do_rfc1323), 0,
198     "Enable rfc1323 (high performance TCP) extensions");
199 
200 static int	tcp_log_debug = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
202     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
203 
204 static int	tcp_tcbhashsize;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
206     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
207 
208 static int	do_tcpdrain = 1;
209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
210     "Enable tcp_drain routine for extra help when low on mbufs");
211 
212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
213     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
214 
215 static VNET_DEFINE(int, icmp_may_rst) = 1;
216 #define	V_icmp_may_rst			VNET(icmp_may_rst)
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
218     &VNET_NAME(icmp_may_rst), 0,
219     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
220 
221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
222 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
224     &VNET_NAME(tcp_isn_reseed_interval), 0,
225     "Seconds between reseeding of ISN secret");
226 
227 static int	tcp_soreceive_stream;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
229     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
230 
231 #ifdef TCP_SIGNATURE
232 static int	tcp_sig_checksigs = 1;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
234     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
235 #endif
236 
237 VNET_DEFINE(uma_zone_t, sack_hole_zone);
238 #define	V_sack_hole_zone		VNET(sack_hole_zone)
239 
240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
241 
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
244 static void tcp_mtudisc(struct inpcb *, int);
245 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
246 		    void *ip4hdr, const void *ip6hdr);
247 static void	tcp_timer_discard(struct tcpcb *, uint32_t);
248 
249 
250 static struct tcp_function_block tcp_def_funcblk = {
251 	"default",
252 	tcp_output,
253 	tcp_do_segment,
254 	tcp_default_ctloutput,
255 	NULL,
256 	NULL,
257 	NULL,
258 	NULL,
259 	NULL,
260 	NULL,
261 	NULL,
262 	0,
263 	0
264 };
265 
266 int t_functions_inited = 0;
267 struct tcp_funchead t_functions;
268 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
269 
270 static void
271 init_tcp_functions(void)
272 {
273 	if (t_functions_inited == 0) {
274 		TAILQ_INIT(&t_functions);
275 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
276 		t_functions_inited = 1;
277 	}
278 }
279 
280 static struct tcp_function_block *
281 find_tcp_functions_locked(struct tcp_function_set *fs)
282 {
283 	struct tcp_function *f;
284 	struct tcp_function_block *blk=NULL;
285 
286 	TAILQ_FOREACH(f, &t_functions, tf_next) {
287 		if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
288 			blk = f->tf_fb;
289 			break;
290 		}
291 	}
292 	return(blk);
293 }
294 
295 static struct tcp_function_block *
296 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
297 {
298 	struct tcp_function_block *rblk=NULL;
299 	struct tcp_function *f;
300 
301 	TAILQ_FOREACH(f, &t_functions, tf_next) {
302 		if (f->tf_fb == blk) {
303 			rblk = blk;
304 			if (s) {
305 				*s = f;
306 			}
307 			break;
308 		}
309 	}
310 	return (rblk);
311 }
312 
313 struct tcp_function_block *
314 find_and_ref_tcp_functions(struct tcp_function_set *fs)
315 {
316 	struct tcp_function_block *blk;
317 
318 	rw_rlock(&tcp_function_lock);
319 	blk = find_tcp_functions_locked(fs);
320 	if (blk)
321 		refcount_acquire(&blk->tfb_refcnt);
322 	rw_runlock(&tcp_function_lock);
323 	return(blk);
324 }
325 
326 struct tcp_function_block *
327 find_and_ref_tcp_fb(struct tcp_function_block *blk)
328 {
329 	struct tcp_function_block *rblk;
330 
331 	rw_rlock(&tcp_function_lock);
332 	rblk = find_tcp_fb_locked(blk, NULL);
333 	if (rblk)
334 		refcount_acquire(&rblk->tfb_refcnt);
335 	rw_runlock(&tcp_function_lock);
336 	return(rblk);
337 }
338 
339 
340 static int
341 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
342 {
343 	int error=ENOENT;
344 	struct tcp_function_set fs;
345 	struct tcp_function_block *blk;
346 
347 	memset(&fs, 0, sizeof(fs));
348 	rw_rlock(&tcp_function_lock);
349 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
350 	if (blk) {
351 		/* Found him */
352 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
353 		fs.pcbcnt = blk->tfb_refcnt;
354 	}
355 	rw_runlock(&tcp_function_lock);
356 	error = sysctl_handle_string(oidp, fs.function_set_name,
357 				     sizeof(fs.function_set_name), req);
358 
359 	/* Check for error or no change */
360 	if (error != 0 || req->newptr == NULL)
361 		return(error);
362 
363 	rw_wlock(&tcp_function_lock);
364 	blk = find_tcp_functions_locked(&fs);
365 	if ((blk == NULL) ||
366 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
367 		error = ENOENT;
368 		goto done;
369 	}
370 	tcp_func_set_ptr = blk;
371 done:
372 	rw_wunlock(&tcp_function_lock);
373 	return (error);
374 }
375 
376 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
377 	    CTLTYPE_STRING | CTLFLAG_RW,
378 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
379 	    "Set/get the default TCP functions");
380 
381 static int
382 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
383 {
384 	int error, cnt, linesz;
385 	struct tcp_function *f;
386 	char *buffer, *cp;
387 	size_t bufsz, outsz;
388 
389 	cnt = 0;
390 	rw_rlock(&tcp_function_lock);
391 	TAILQ_FOREACH(f, &t_functions, tf_next) {
392 		cnt++;
393 	}
394 	rw_runlock(&tcp_function_lock);
395 
396 	bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
397 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
398 
399 	error = 0;
400 	cp = buffer;
401 
402 	linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
403 	cp += linesz;
404 	bufsz -= linesz;
405 	outsz = linesz;
406 
407 	rw_rlock(&tcp_function_lock);
408 	TAILQ_FOREACH(f, &t_functions, tf_next) {
409 		linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
410 		    f->tf_fb->tfb_tcp_block_name,
411 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
412 		    f->tf_fb->tfb_refcnt);
413 		if (linesz >= bufsz) {
414 			error = EOVERFLOW;
415 			break;
416 		}
417 		cp += linesz;
418 		bufsz -= linesz;
419 		outsz += linesz;
420 	}
421 	rw_runlock(&tcp_function_lock);
422 	if (error == 0)
423 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
424 	free(buffer, M_TEMP);
425 	return (error);
426 }
427 
428 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
429 	    CTLTYPE_STRING|CTLFLAG_RD,
430 	    NULL, 0, sysctl_net_inet_list_available, "A",
431 	    "list available TCP Function sets");
432 
433 /*
434  * Target size of TCP PCB hash tables. Must be a power of two.
435  *
436  * Note that this can be overridden by the kernel environment
437  * variable net.inet.tcp.tcbhashsize
438  */
439 #ifndef TCBHASHSIZE
440 #define TCBHASHSIZE	0
441 #endif
442 
443 /*
444  * XXX
445  * Callouts should be moved into struct tcp directly.  They are currently
446  * separate because the tcpcb structure is exported to userland for sysctl
447  * parsing purposes, which do not know about callouts.
448  */
449 struct tcpcb_mem {
450 	struct	tcpcb		tcb;
451 	struct	tcp_timer	tt;
452 	struct	cc_var		ccv;
453 	struct	osd		osd;
454 };
455 
456 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
457 #define	V_tcpcb_zone			VNET(tcpcb_zone)
458 
459 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
460 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
461 
462 static struct mtx isn_mtx;
463 
464 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
465 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
466 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
467 
468 /*
469  * TCP initialization.
470  */
471 static void
472 tcp_zone_change(void *tag)
473 {
474 
475 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
476 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
477 	tcp_tw_zone_change();
478 }
479 
480 static int
481 tcp_inpcb_init(void *mem, int size, int flags)
482 {
483 	struct inpcb *inp = mem;
484 
485 	INP_LOCK_INIT(inp, "inp", "tcpinp");
486 	return (0);
487 }
488 
489 /*
490  * Take a value and get the next power of 2 that doesn't overflow.
491  * Used to size the tcp_inpcb hash buckets.
492  */
493 static int
494 maketcp_hashsize(int size)
495 {
496 	int hashsize;
497 
498 	/*
499 	 * auto tune.
500 	 * get the next power of 2 higher than maxsockets.
501 	 */
502 	hashsize = 1 << fls(size);
503 	/* catch overflow, and just go one power of 2 smaller */
504 	if (hashsize < size) {
505 		hashsize = 1 << (fls(size) - 1);
506 	}
507 	return (hashsize);
508 }
509 
510 int
511 register_tcp_functions(struct tcp_function_block *blk, int wait)
512 {
513 	struct tcp_function_block *lblk;
514 	struct tcp_function *n;
515 	struct tcp_function_set fs;
516 
517 	if (t_functions_inited == 0) {
518 		init_tcp_functions();
519 	}
520 	if ((blk->tfb_tcp_output == NULL) ||
521 	    (blk->tfb_tcp_do_segment == NULL) ||
522 	    (blk->tfb_tcp_ctloutput == NULL) ||
523 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
524 		/*
525 		 * These functions are required and you
526 		 * need a name.
527 		 */
528 		return (EINVAL);
529 	}
530 	if (blk->tfb_tcp_timer_stop_all ||
531 	    blk->tfb_tcp_timers_left ||
532 	    blk->tfb_tcp_timer_activate ||
533 	    blk->tfb_tcp_timer_active ||
534 	    blk->tfb_tcp_timer_stop) {
535 		/*
536 		 * If you define one timer function you
537 		 * must have them all.
538 		 */
539 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
540 		    (blk->tfb_tcp_timers_left  == NULL) ||
541 		    (blk->tfb_tcp_timer_activate == NULL) ||
542 		    (blk->tfb_tcp_timer_active == NULL) ||
543 		    (blk->tfb_tcp_timer_stop == NULL)) {
544 			return (EINVAL);
545 		}
546 	}
547 	n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
548 	if (n == NULL) {
549 		return (ENOMEM);
550 	}
551 	n->tf_fb = blk;
552 	strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
553 	rw_wlock(&tcp_function_lock);
554 	lblk = find_tcp_functions_locked(&fs);
555 	if (lblk) {
556 		/* Duplicate name space not allowed */
557 		rw_wunlock(&tcp_function_lock);
558 		free(n, M_TCPFUNCTIONS);
559 		return (EALREADY);
560 	}
561 	refcount_init(&blk->tfb_refcnt, 0);
562 	blk->tfb_flags = 0;
563 	TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
564 	rw_wunlock(&tcp_function_lock);
565 	return(0);
566 }
567 
568 int
569 deregister_tcp_functions(struct tcp_function_block *blk)
570 {
571 	struct tcp_function_block *lblk;
572 	struct tcp_function *f;
573 	int error=ENOENT;
574 
575 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
576 		/* You can't un-register the default */
577 		return (EPERM);
578 	}
579 	rw_wlock(&tcp_function_lock);
580 	if (blk == tcp_func_set_ptr) {
581 		/* You can't free the current default */
582 		rw_wunlock(&tcp_function_lock);
583 		return (EBUSY);
584 	}
585 	if (blk->tfb_refcnt) {
586 		/* Still tcb attached, mark it. */
587 		blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
588 		rw_wunlock(&tcp_function_lock);
589 		return (EBUSY);
590 	}
591 	lblk = find_tcp_fb_locked(blk, &f);
592 	if (lblk) {
593 		/* Found */
594 		TAILQ_REMOVE(&t_functions, f, tf_next);
595 		f->tf_fb = NULL;
596 		free(f, M_TCPFUNCTIONS);
597 		error = 0;
598 	}
599 	rw_wunlock(&tcp_function_lock);
600 	return (error);
601 }
602 
603 void
604 tcp_init(void)
605 {
606 	const char *tcbhash_tuneable;
607 	int hashsize;
608 
609 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
610 
611 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
612 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
613 		printf("%s: WARNING: unable to register helper hook\n", __func__);
614 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
615 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
616 		printf("%s: WARNING: unable to register helper hook\n", __func__);
617 	hashsize = TCBHASHSIZE;
618 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
619 	if (hashsize == 0) {
620 		/*
621 		 * Auto tune the hash size based on maxsockets.
622 		 * A perfect hash would have a 1:1 mapping
623 		 * (hashsize = maxsockets) however it's been
624 		 * suggested that O(2) average is better.
625 		 */
626 		hashsize = maketcp_hashsize(maxsockets / 4);
627 		/*
628 		 * Our historical default is 512,
629 		 * do not autotune lower than this.
630 		 */
631 		if (hashsize < 512)
632 			hashsize = 512;
633 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
634 			printf("%s: %s auto tuned to %d\n", __func__,
635 			    tcbhash_tuneable, hashsize);
636 	}
637 	/*
638 	 * We require a hashsize to be a power of two.
639 	 * Previously if it was not a power of two we would just reset it
640 	 * back to 512, which could be a nasty surprise if you did not notice
641 	 * the error message.
642 	 * Instead what we do is clip it to the closest power of two lower
643 	 * than the specified hash value.
644 	 */
645 	if (!powerof2(hashsize)) {
646 		int oldhashsize = hashsize;
647 
648 		hashsize = maketcp_hashsize(hashsize);
649 		/* prevent absurdly low value */
650 		if (hashsize < 16)
651 			hashsize = 16;
652 		printf("%s: WARNING: TCB hash size not a power of 2, "
653 		    "clipped from %d to %d.\n", __func__, oldhashsize,
654 		    hashsize);
655 	}
656 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
657 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
658 	    IPI_HASHFIELDS_4TUPLE);
659 
660 	/*
661 	 * These have to be type stable for the benefit of the timers.
662 	 */
663 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
664 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
665 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
666 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
667 
668 	tcp_tw_init();
669 	syncache_init();
670 	tcp_hc_init();
671 
672 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
673 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
674 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
675 
676 	/* Skip initialization of globals for non-default instances. */
677 	if (!IS_DEFAULT_VNET(curvnet))
678 		return;
679 
680 	tcp_reass_global_init();
681 
682 	/* XXX virtualize those bellow? */
683 	tcp_delacktime = TCPTV_DELACK;
684 	tcp_keepinit = TCPTV_KEEP_INIT;
685 	tcp_keepidle = TCPTV_KEEP_IDLE;
686 	tcp_keepintvl = TCPTV_KEEPINTVL;
687 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
688 	tcp_msl = TCPTV_MSL;
689 	tcp_rexmit_min = TCPTV_MIN;
690 	if (tcp_rexmit_min < 1)
691 		tcp_rexmit_min = 1;
692 	tcp_persmin = TCPTV_PERSMIN;
693 	tcp_persmax = TCPTV_PERSMAX;
694 	tcp_rexmit_slop = TCPTV_CPU_VAR;
695 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
696 	tcp_tcbhashsize = hashsize;
697 	/* Setup the tcp function block list */
698 	init_tcp_functions();
699 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
700 
701 	if (tcp_soreceive_stream) {
702 #ifdef INET
703 		tcp_usrreqs.pru_soreceive = soreceive_stream;
704 #endif
705 #ifdef INET6
706 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
707 #endif /* INET6 */
708 	}
709 
710 #ifdef INET6
711 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
712 #else /* INET6 */
713 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
714 #endif /* INET6 */
715 	if (max_protohdr < TCP_MINPROTOHDR)
716 		max_protohdr = TCP_MINPROTOHDR;
717 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
718 		panic("tcp_init");
719 #undef TCP_MINPROTOHDR
720 
721 	ISN_LOCK_INIT();
722 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
723 		SHUTDOWN_PRI_DEFAULT);
724 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
725 		EVENTHANDLER_PRI_ANY);
726 #ifdef TCPPCAP
727 	tcp_pcap_init();
728 #endif
729 
730 #ifdef TCP_RFC7413
731 	tcp_fastopen_init();
732 #endif
733 }
734 
735 #ifdef VIMAGE
736 void
737 tcp_destroy(void)
738 {
739 	int error;
740 
741 #ifdef TCP_RFC7413
742 	tcp_fastopen_destroy();
743 #endif
744 	tcp_hc_destroy();
745 	syncache_destroy();
746 	tcp_tw_destroy();
747 	in_pcbinfo_destroy(&V_tcbinfo);
748 	uma_zdestroy(V_sack_hole_zone);
749 	uma_zdestroy(V_tcpcb_zone);
750 
751 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
752 	if (error != 0) {
753 		printf("%s: WARNING: unable to deregister helper hook "
754 		    "type=%d, id=%d: error %d returned\n", __func__,
755 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
756 	}
757 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
758 	if (error != 0) {
759 		printf("%s: WARNING: unable to deregister helper hook "
760 		    "type=%d, id=%d: error %d returned\n", __func__,
761 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
762 	}
763 }
764 #endif
765 
766 void
767 tcp_fini(void *xtp)
768 {
769 
770 }
771 
772 /*
773  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
774  * tcp_template used to store this data in mbufs, but we now recopy it out
775  * of the tcpcb each time to conserve mbufs.
776  */
777 void
778 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
779 {
780 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
781 
782 	INP_WLOCK_ASSERT(inp);
783 
784 #ifdef INET6
785 	if ((inp->inp_vflag & INP_IPV6) != 0) {
786 		struct ip6_hdr *ip6;
787 
788 		ip6 = (struct ip6_hdr *)ip_ptr;
789 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
790 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
791 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
792 			(IPV6_VERSION & IPV6_VERSION_MASK);
793 		ip6->ip6_nxt = IPPROTO_TCP;
794 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
795 		ip6->ip6_src = inp->in6p_laddr;
796 		ip6->ip6_dst = inp->in6p_faddr;
797 	}
798 #endif /* INET6 */
799 #if defined(INET6) && defined(INET)
800 	else
801 #endif
802 #ifdef INET
803 	{
804 		struct ip *ip;
805 
806 		ip = (struct ip *)ip_ptr;
807 		ip->ip_v = IPVERSION;
808 		ip->ip_hl = 5;
809 		ip->ip_tos = inp->inp_ip_tos;
810 		ip->ip_len = 0;
811 		ip->ip_id = 0;
812 		ip->ip_off = 0;
813 		ip->ip_ttl = inp->inp_ip_ttl;
814 		ip->ip_sum = 0;
815 		ip->ip_p = IPPROTO_TCP;
816 		ip->ip_src = inp->inp_laddr;
817 		ip->ip_dst = inp->inp_faddr;
818 	}
819 #endif /* INET */
820 	th->th_sport = inp->inp_lport;
821 	th->th_dport = inp->inp_fport;
822 	th->th_seq = 0;
823 	th->th_ack = 0;
824 	th->th_x2 = 0;
825 	th->th_off = 5;
826 	th->th_flags = 0;
827 	th->th_win = 0;
828 	th->th_urp = 0;
829 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
830 }
831 
832 /*
833  * Create template to be used to send tcp packets on a connection.
834  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
835  * use for this function is in keepalives, which use tcp_respond.
836  */
837 struct tcptemp *
838 tcpip_maketemplate(struct inpcb *inp)
839 {
840 	struct tcptemp *t;
841 
842 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
843 	if (t == NULL)
844 		return (NULL);
845 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
846 	return (t);
847 }
848 
849 /*
850  * Send a single message to the TCP at address specified by
851  * the given TCP/IP header.  If m == NULL, then we make a copy
852  * of the tcpiphdr at th and send directly to the addressed host.
853  * This is used to force keep alive messages out using the TCP
854  * template for a connection.  If flags are given then we send
855  * a message back to the TCP which originated the segment th,
856  * and discard the mbuf containing it and any other attached mbufs.
857  *
858  * In any case the ack and sequence number of the transmitted
859  * segment are as specified by the parameters.
860  *
861  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
862  */
863 void
864 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
865     tcp_seq ack, tcp_seq seq, int flags)
866 {
867 	struct tcpopt to;
868 	struct inpcb *inp;
869 	struct ip *ip;
870 	struct mbuf *optm;
871 	struct tcphdr *nth;
872 	u_char *optp;
873 #ifdef INET6
874 	struct ip6_hdr *ip6;
875 	int isipv6;
876 #endif /* INET6 */
877 	int optlen, tlen, win;
878 	bool incl_opts;
879 
880 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
881 
882 #ifdef INET6
883 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
884 	ip6 = ipgen;
885 #endif /* INET6 */
886 	ip = ipgen;
887 
888 	if (tp != NULL) {
889 		inp = tp->t_inpcb;
890 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
891 		INP_WLOCK_ASSERT(inp);
892 	} else
893 		inp = NULL;
894 
895 	incl_opts = false;
896 	win = 0;
897 	if (tp != NULL) {
898 		if (!(flags & TH_RST)) {
899 			win = sbspace(&inp->inp_socket->so_rcv);
900 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
901 				win = (long)TCP_MAXWIN << tp->rcv_scale;
902 		}
903 		if ((tp->t_flags & TF_NOOPT) == 0)
904 			incl_opts = true;
905 	}
906 	if (m == NULL) {
907 		m = m_gethdr(M_NOWAIT, MT_DATA);
908 		if (m == NULL)
909 			return;
910 		m->m_data += max_linkhdr;
911 #ifdef INET6
912 		if (isipv6) {
913 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
914 			      sizeof(struct ip6_hdr));
915 			ip6 = mtod(m, struct ip6_hdr *);
916 			nth = (struct tcphdr *)(ip6 + 1);
917 		} else
918 #endif /* INET6 */
919 		{
920 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
921 			ip = mtod(m, struct ip *);
922 			nth = (struct tcphdr *)(ip + 1);
923 		}
924 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
925 		flags = TH_ACK;
926 	} else {
927 		/*
928 		 *  reuse the mbuf.
929 		 * XXX MRT We inherrit the FIB, which is lucky.
930 		 */
931 		m_freem(m->m_next);
932 		m->m_next = NULL;
933 		m->m_data = (caddr_t)ipgen;
934 		/* m_len is set later */
935 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
936 #ifdef INET6
937 		if (isipv6) {
938 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
939 			nth = (struct tcphdr *)(ip6 + 1);
940 		} else
941 #endif /* INET6 */
942 		{
943 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
944 			nth = (struct tcphdr *)(ip + 1);
945 		}
946 		if (th != nth) {
947 			/*
948 			 * this is usually a case when an extension header
949 			 * exists between the IPv6 header and the
950 			 * TCP header.
951 			 */
952 			nth->th_sport = th->th_sport;
953 			nth->th_dport = th->th_dport;
954 		}
955 		xchg(nth->th_dport, nth->th_sport, uint16_t);
956 #undef xchg
957 	}
958 	tlen = 0;
959 #ifdef INET6
960 	if (isipv6)
961 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
962 #endif
963 #if defined(INET) && defined(INET6)
964 	else
965 #endif
966 #ifdef INET
967 		tlen = sizeof (struct tcpiphdr);
968 #endif
969 #ifdef INVARIANTS
970 	m->m_len = 0;
971 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
972 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
973 	    m, tlen, (long)M_TRAILINGSPACE(m)));
974 #endif
975 	m->m_len = tlen;
976 	to.to_flags = 0;
977 	if (incl_opts) {
978 		/* Make sure we have room. */
979 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
980 			m->m_next = m_get(M_NOWAIT, MT_DATA);
981 			if (m->m_next) {
982 				optp = mtod(m->m_next, u_char *);
983 				optm = m->m_next;
984 			} else
985 				incl_opts = false;
986 		} else {
987 			optp = (u_char *) (nth + 1);
988 			optm = m;
989 		}
990 	}
991 	if (incl_opts) {
992 		/* Timestamps. */
993 		if (tp->t_flags & TF_RCVD_TSTMP) {
994 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
995 			to.to_tsecr = tp->ts_recent;
996 			to.to_flags |= TOF_TS;
997 		}
998 #ifdef TCP_SIGNATURE
999 		/* TCP-MD5 (RFC2385). */
1000 		if (tp->t_flags & TF_SIGNATURE)
1001 			to.to_flags |= TOF_SIGNATURE;
1002 #endif
1003 
1004 		/* Add the options. */
1005 		tlen += optlen = tcp_addoptions(&to, optp);
1006 
1007 		/* Update m_len in the correct mbuf. */
1008 		optm->m_len += optlen;
1009 	} else
1010 		optlen = 0;
1011 #ifdef INET6
1012 	if (isipv6) {
1013 		ip6->ip6_flow = 0;
1014 		ip6->ip6_vfc = IPV6_VERSION;
1015 		ip6->ip6_nxt = IPPROTO_TCP;
1016 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1017 	}
1018 #endif
1019 #if defined(INET) && defined(INET6)
1020 	else
1021 #endif
1022 #ifdef INET
1023 	{
1024 		ip->ip_len = htons(tlen);
1025 		ip->ip_ttl = V_ip_defttl;
1026 		if (V_path_mtu_discovery)
1027 			ip->ip_off |= htons(IP_DF);
1028 	}
1029 #endif
1030 	m->m_pkthdr.len = tlen;
1031 	m->m_pkthdr.rcvif = NULL;
1032 #ifdef MAC
1033 	if (inp != NULL) {
1034 		/*
1035 		 * Packet is associated with a socket, so allow the
1036 		 * label of the response to reflect the socket label.
1037 		 */
1038 		INP_WLOCK_ASSERT(inp);
1039 		mac_inpcb_create_mbuf(inp, m);
1040 	} else {
1041 		/*
1042 		 * Packet is not associated with a socket, so possibly
1043 		 * update the label in place.
1044 		 */
1045 		mac_netinet_tcp_reply(m);
1046 	}
1047 #endif
1048 	nth->th_seq = htonl(seq);
1049 	nth->th_ack = htonl(ack);
1050 	nth->th_x2 = 0;
1051 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1052 	nth->th_flags = flags;
1053 	if (tp != NULL)
1054 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1055 	else
1056 		nth->th_win = htons((u_short)win);
1057 	nth->th_urp = 0;
1058 
1059 #ifdef TCP_SIGNATURE
1060 	if (to.to_flags & TOF_SIGNATURE) {
1061 		tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1062 		    IPSEC_DIR_OUTBOUND);
1063 	}
1064 #endif
1065 
1066 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1067 #ifdef INET6
1068 	if (isipv6) {
1069 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1070 		nth->th_sum = in6_cksum_pseudo(ip6,
1071 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1072 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1073 		    NULL, NULL);
1074 	}
1075 #endif /* INET6 */
1076 #if defined(INET6) && defined(INET)
1077 	else
1078 #endif
1079 #ifdef INET
1080 	{
1081 		m->m_pkthdr.csum_flags = CSUM_TCP;
1082 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1083 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1084 	}
1085 #endif /* INET */
1086 #ifdef TCPDEBUG
1087 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1088 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1089 #endif
1090 	TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1091 	if (flags & TH_RST)
1092 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1093 		    tp, nth);
1094 
1095 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1096 #ifdef INET6
1097 	if (isipv6)
1098 		(void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1099 #endif /* INET6 */
1100 #if defined(INET) && defined(INET6)
1101 	else
1102 #endif
1103 #ifdef INET
1104 		(void) ip_output(m, NULL, NULL, 0, NULL, inp);
1105 #endif
1106 }
1107 
1108 /*
1109  * Create a new TCP control block, making an
1110  * empty reassembly queue and hooking it to the argument
1111  * protocol control block.  The `inp' parameter must have
1112  * come from the zone allocator set up in tcp_init().
1113  */
1114 struct tcpcb *
1115 tcp_newtcpcb(struct inpcb *inp)
1116 {
1117 	struct tcpcb_mem *tm;
1118 	struct tcpcb *tp;
1119 #ifdef INET6
1120 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1121 #endif /* INET6 */
1122 
1123 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1124 	if (tm == NULL)
1125 		return (NULL);
1126 	tp = &tm->tcb;
1127 
1128 	/* Initialise cc_var struct for this tcpcb. */
1129 	tp->ccv = &tm->ccv;
1130 	tp->ccv->type = IPPROTO_TCP;
1131 	tp->ccv->ccvc.tcp = tp;
1132 	rw_rlock(&tcp_function_lock);
1133 	tp->t_fb = tcp_func_set_ptr;
1134 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1135 	rw_runlock(&tcp_function_lock);
1136 	if (tp->t_fb->tfb_tcp_fb_init) {
1137 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1138 	}
1139 	/*
1140 	 * Use the current system default CC algorithm.
1141 	 */
1142 	CC_LIST_RLOCK();
1143 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1144 	CC_ALGO(tp) = CC_DEFAULT();
1145 	CC_LIST_RUNLOCK();
1146 
1147 	if (CC_ALGO(tp)->cb_init != NULL)
1148 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1149 			if (tp->t_fb->tfb_tcp_fb_fini)
1150 				(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1151 			refcount_release(&tp->t_fb->tfb_refcnt);
1152 			uma_zfree(V_tcpcb_zone, tm);
1153 			return (NULL);
1154 		}
1155 
1156 	tp->osd = &tm->osd;
1157 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1158 		if (tp->t_fb->tfb_tcp_fb_fini)
1159 			(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1160 		refcount_release(&tp->t_fb->tfb_refcnt);
1161 		uma_zfree(V_tcpcb_zone, tm);
1162 		return (NULL);
1163 	}
1164 
1165 #ifdef VIMAGE
1166 	tp->t_vnet = inp->inp_vnet;
1167 #endif
1168 	tp->t_timers = &tm->tt;
1169 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
1170 	tp->t_maxseg =
1171 #ifdef INET6
1172 		isipv6 ? V_tcp_v6mssdflt :
1173 #endif /* INET6 */
1174 		V_tcp_mssdflt;
1175 
1176 	/* Set up our timeouts. */
1177 	callout_init(&tp->t_timers->tt_rexmt, 1);
1178 	callout_init(&tp->t_timers->tt_persist, 1);
1179 	callout_init(&tp->t_timers->tt_keep, 1);
1180 	callout_init(&tp->t_timers->tt_2msl, 1);
1181 	callout_init(&tp->t_timers->tt_delack, 1);
1182 
1183 	if (V_tcp_do_rfc1323)
1184 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1185 	if (V_tcp_do_sack)
1186 		tp->t_flags |= TF_SACK_PERMIT;
1187 	TAILQ_INIT(&tp->snd_holes);
1188 	/*
1189 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1190 	 * is called.
1191 	 */
1192 	in_pcbref(inp);	/* Reference for tcpcb */
1193 	tp->t_inpcb = inp;
1194 
1195 	/*
1196 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1197 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1198 	 * reasonable initial retransmit time.
1199 	 */
1200 	tp->t_srtt = TCPTV_SRTTBASE;
1201 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1202 	tp->t_rttmin = tcp_rexmit_min;
1203 	tp->t_rxtcur = TCPTV_RTOBASE;
1204 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1205 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1206 	tp->t_rcvtime = ticks;
1207 	/*
1208 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1209 	 * because the socket may be bound to an IPv6 wildcard address,
1210 	 * which may match an IPv4-mapped IPv6 address.
1211 	 */
1212 	inp->inp_ip_ttl = V_ip_defttl;
1213 	inp->inp_ppcb = tp;
1214 #ifdef TCPPCAP
1215 	/*
1216 	 * Init the TCP PCAP queues.
1217 	 */
1218 	tcp_pcap_tcpcb_init(tp);
1219 #endif
1220 	return (tp);		/* XXX */
1221 }
1222 
1223 /*
1224  * Switch the congestion control algorithm back to NewReno for any active
1225  * control blocks using an algorithm which is about to go away.
1226  * This ensures the CC framework can allow the unload to proceed without leaving
1227  * any dangling pointers which would trigger a panic.
1228  * Returning non-zero would inform the CC framework that something went wrong
1229  * and it would be unsafe to allow the unload to proceed. However, there is no
1230  * way for this to occur with this implementation so we always return zero.
1231  */
1232 int
1233 tcp_ccalgounload(struct cc_algo *unload_algo)
1234 {
1235 	struct cc_algo *tmpalgo;
1236 	struct inpcb *inp;
1237 	struct tcpcb *tp;
1238 	VNET_ITERATOR_DECL(vnet_iter);
1239 
1240 	/*
1241 	 * Check all active control blocks across all network stacks and change
1242 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1243 	 * requires cleanup code to be run, call it.
1244 	 */
1245 	VNET_LIST_RLOCK();
1246 	VNET_FOREACH(vnet_iter) {
1247 		CURVNET_SET(vnet_iter);
1248 		INP_INFO_WLOCK(&V_tcbinfo);
1249 		/*
1250 		 * New connections already part way through being initialised
1251 		 * with the CC algo we're removing will not race with this code
1252 		 * because the INP_INFO_WLOCK is held during initialisation. We
1253 		 * therefore don't enter the loop below until the connection
1254 		 * list has stabilised.
1255 		 */
1256 		LIST_FOREACH(inp, &V_tcb, inp_list) {
1257 			INP_WLOCK(inp);
1258 			/* Important to skip tcptw structs. */
1259 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1260 			    (tp = intotcpcb(inp)) != NULL) {
1261 				/*
1262 				 * By holding INP_WLOCK here, we are assured
1263 				 * that the connection is not currently
1264 				 * executing inside the CC module's functions
1265 				 * i.e. it is safe to make the switch back to
1266 				 * NewReno.
1267 				 */
1268 				if (CC_ALGO(tp) == unload_algo) {
1269 					tmpalgo = CC_ALGO(tp);
1270 					/* NewReno does not require any init. */
1271 					CC_ALGO(tp) = &newreno_cc_algo;
1272 					if (tmpalgo->cb_destroy != NULL)
1273 						tmpalgo->cb_destroy(tp->ccv);
1274 				}
1275 			}
1276 			INP_WUNLOCK(inp);
1277 		}
1278 		INP_INFO_WUNLOCK(&V_tcbinfo);
1279 		CURVNET_RESTORE();
1280 	}
1281 	VNET_LIST_RUNLOCK();
1282 
1283 	return (0);
1284 }
1285 
1286 /*
1287  * Drop a TCP connection, reporting
1288  * the specified error.  If connection is synchronized,
1289  * then send a RST to peer.
1290  */
1291 struct tcpcb *
1292 tcp_drop(struct tcpcb *tp, int errno)
1293 {
1294 	struct socket *so = tp->t_inpcb->inp_socket;
1295 
1296 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1297 	INP_WLOCK_ASSERT(tp->t_inpcb);
1298 
1299 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1300 		tcp_state_change(tp, TCPS_CLOSED);
1301 		(void) tp->t_fb->tfb_tcp_output(tp);
1302 		TCPSTAT_INC(tcps_drops);
1303 	} else
1304 		TCPSTAT_INC(tcps_conndrops);
1305 	if (errno == ETIMEDOUT && tp->t_softerror)
1306 		errno = tp->t_softerror;
1307 	so->so_error = errno;
1308 	return (tcp_close(tp));
1309 }
1310 
1311 void
1312 tcp_discardcb(struct tcpcb *tp)
1313 {
1314 	struct inpcb *inp = tp->t_inpcb;
1315 	struct socket *so = inp->inp_socket;
1316 #ifdef INET6
1317 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1318 #endif /* INET6 */
1319 	int released;
1320 
1321 	INP_WLOCK_ASSERT(inp);
1322 
1323 	/*
1324 	 * Make sure that all of our timers are stopped before we delete the
1325 	 * PCB.
1326 	 *
1327 	 * If stopping a timer fails, we schedule a discard function in same
1328 	 * callout, and the last discard function called will take care of
1329 	 * deleting the tcpcb.
1330 	 */
1331 	tcp_timer_stop(tp, TT_REXMT);
1332 	tcp_timer_stop(tp, TT_PERSIST);
1333 	tcp_timer_stop(tp, TT_KEEP);
1334 	tcp_timer_stop(tp, TT_2MSL);
1335 	tcp_timer_stop(tp, TT_DELACK);
1336 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1337 		/* Call the stop-all function of the methods */
1338 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1339 	}
1340 
1341 	/*
1342 	 * If we got enough samples through the srtt filter,
1343 	 * save the rtt and rttvar in the routing entry.
1344 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1345 	 * 4 samples is enough for the srtt filter to converge
1346 	 * to within enough % of the correct value; fewer samples
1347 	 * and we could save a bogus rtt. The danger is not high
1348 	 * as tcp quickly recovers from everything.
1349 	 * XXX: Works very well but needs some more statistics!
1350 	 */
1351 	if (tp->t_rttupdated >= 4) {
1352 		struct hc_metrics_lite metrics;
1353 		u_long ssthresh;
1354 
1355 		bzero(&metrics, sizeof(metrics));
1356 		/*
1357 		 * Update the ssthresh always when the conditions below
1358 		 * are satisfied. This gives us better new start value
1359 		 * for the congestion avoidance for new connections.
1360 		 * ssthresh is only set if packet loss occured on a session.
1361 		 *
1362 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1363 		 * being torn down.  Ideally this code would not use 'so'.
1364 		 */
1365 		ssthresh = tp->snd_ssthresh;
1366 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1367 			/*
1368 			 * convert the limit from user data bytes to
1369 			 * packets then to packet data bytes.
1370 			 */
1371 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1372 			if (ssthresh < 2)
1373 				ssthresh = 2;
1374 			ssthresh *= (u_long)(tp->t_maxseg +
1375 #ifdef INET6
1376 			    (isipv6 ? sizeof (struct ip6_hdr) +
1377 				sizeof (struct tcphdr) :
1378 #endif
1379 				sizeof (struct tcpiphdr)
1380 #ifdef INET6
1381 			    )
1382 #endif
1383 			    );
1384 		} else
1385 			ssthresh = 0;
1386 		metrics.rmx_ssthresh = ssthresh;
1387 
1388 		metrics.rmx_rtt = tp->t_srtt;
1389 		metrics.rmx_rttvar = tp->t_rttvar;
1390 		metrics.rmx_cwnd = tp->snd_cwnd;
1391 		metrics.rmx_sendpipe = 0;
1392 		metrics.rmx_recvpipe = 0;
1393 
1394 		tcp_hc_update(&inp->inp_inc, &metrics);
1395 	}
1396 
1397 	/* free the reassembly queue, if any */
1398 	tcp_reass_flush(tp);
1399 
1400 #ifdef TCP_OFFLOAD
1401 	/* Disconnect offload device, if any. */
1402 	if (tp->t_flags & TF_TOE)
1403 		tcp_offload_detach(tp);
1404 #endif
1405 
1406 	tcp_free_sackholes(tp);
1407 
1408 #ifdef TCPPCAP
1409 	/* Free the TCP PCAP queues. */
1410 	tcp_pcap_drain(&(tp->t_inpkts));
1411 	tcp_pcap_drain(&(tp->t_outpkts));
1412 #endif
1413 
1414 	/* Allow the CC algorithm to clean up after itself. */
1415 	if (CC_ALGO(tp)->cb_destroy != NULL)
1416 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1417 
1418 	khelp_destroy_osd(tp->osd);
1419 
1420 	CC_ALGO(tp) = NULL;
1421 	inp->inp_ppcb = NULL;
1422 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1423 		/* We own the last reference on tcpcb, let's free it. */
1424 		if ((tp->t_fb->tfb_tcp_timers_left) &&
1425 		    (tp->t_fb->tfb_tcp_timers_left(tp))) {
1426 			    /* Some fb timers left running! */
1427 			    return;
1428 		}
1429 		if (tp->t_fb->tfb_tcp_fb_fini)
1430 			(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1431 		refcount_release(&tp->t_fb->tfb_refcnt);
1432 		tp->t_inpcb = NULL;
1433 		uma_zfree(V_tcpcb_zone, tp);
1434 		released = in_pcbrele_wlocked(inp);
1435 		KASSERT(!released, ("%s: inp %p should not have been released "
1436 			"here", __func__, inp));
1437 	}
1438 }
1439 
1440 void
1441 tcp_timer_2msl_discard(void *xtp)
1442 {
1443 
1444 	tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1445 }
1446 
1447 void
1448 tcp_timer_keep_discard(void *xtp)
1449 {
1450 
1451 	tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1452 }
1453 
1454 void
1455 tcp_timer_persist_discard(void *xtp)
1456 {
1457 
1458 	tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1459 }
1460 
1461 void
1462 tcp_timer_rexmt_discard(void *xtp)
1463 {
1464 
1465 	tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1466 }
1467 
1468 void
1469 tcp_timer_delack_discard(void *xtp)
1470 {
1471 
1472 	tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1473 }
1474 
1475 void
1476 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1477 {
1478 	struct inpcb *inp;
1479 
1480 	CURVNET_SET(tp->t_vnet);
1481 	INP_INFO_RLOCK(&V_tcbinfo);
1482 	inp = tp->t_inpcb;
1483 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1484 		__func__, tp));
1485 	INP_WLOCK(inp);
1486 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1487 		("%s: tcpcb has to be stopped here", __func__));
1488 	KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1489 		("%s: discard callout should be running", __func__));
1490 	tp->t_timers->tt_flags &= ~timer_type;
1491 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1492 		/* We own the last reference on this tcpcb, let's free it. */
1493 		if ((tp->t_fb->tfb_tcp_timers_left) &&
1494 		    (tp->t_fb->tfb_tcp_timers_left(tp))) {
1495 			    /* Some fb timers left running! */
1496 			    goto leave;
1497 		}
1498 		if (tp->t_fb->tfb_tcp_fb_fini)
1499 			(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1500 		refcount_release(&tp->t_fb->tfb_refcnt);
1501 		tp->t_inpcb = NULL;
1502 		uma_zfree(V_tcpcb_zone, tp);
1503 		if (in_pcbrele_wlocked(inp)) {
1504 			INP_INFO_RUNLOCK(&V_tcbinfo);
1505 			CURVNET_RESTORE();
1506 			return;
1507 		}
1508 	}
1509 leave:
1510 	INP_WUNLOCK(inp);
1511 	INP_INFO_RUNLOCK(&V_tcbinfo);
1512 	CURVNET_RESTORE();
1513 }
1514 
1515 /*
1516  * Attempt to close a TCP control block, marking it as dropped, and freeing
1517  * the socket if we hold the only reference.
1518  */
1519 struct tcpcb *
1520 tcp_close(struct tcpcb *tp)
1521 {
1522 	struct inpcb *inp = tp->t_inpcb;
1523 	struct socket *so;
1524 
1525 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1526 	INP_WLOCK_ASSERT(inp);
1527 
1528 #ifdef TCP_OFFLOAD
1529 	if (tp->t_state == TCPS_LISTEN)
1530 		tcp_offload_listen_stop(tp);
1531 #endif
1532 #ifdef TCP_RFC7413
1533 	/*
1534 	 * This releases the TFO pending counter resource for TFO listen
1535 	 * sockets as well as passively-created TFO sockets that transition
1536 	 * from SYN_RECEIVED to CLOSED.
1537 	 */
1538 	if (tp->t_tfo_pending) {
1539 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1540 		tp->t_tfo_pending = NULL;
1541 	}
1542 #endif
1543 	in_pcbdrop(inp);
1544 	TCPSTAT_INC(tcps_closed);
1545 	TCPSTATES_DEC(tp->t_state);
1546 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1547 	so = inp->inp_socket;
1548 	soisdisconnected(so);
1549 	if (inp->inp_flags & INP_SOCKREF) {
1550 		KASSERT(so->so_state & SS_PROTOREF,
1551 		    ("tcp_close: !SS_PROTOREF"));
1552 		inp->inp_flags &= ~INP_SOCKREF;
1553 		INP_WUNLOCK(inp);
1554 		ACCEPT_LOCK();
1555 		SOCK_LOCK(so);
1556 		so->so_state &= ~SS_PROTOREF;
1557 		sofree(so);
1558 		return (NULL);
1559 	}
1560 	return (tp);
1561 }
1562 
1563 void
1564 tcp_drain(void)
1565 {
1566 	VNET_ITERATOR_DECL(vnet_iter);
1567 
1568 	if (!do_tcpdrain)
1569 		return;
1570 
1571 	VNET_LIST_RLOCK_NOSLEEP();
1572 	VNET_FOREACH(vnet_iter) {
1573 		CURVNET_SET(vnet_iter);
1574 		struct inpcb *inpb;
1575 		struct tcpcb *tcpb;
1576 
1577 	/*
1578 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1579 	 * if there is one...
1580 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1581 	 *      reassembly queue should be flushed, but in a situation
1582 	 *	where we're really low on mbufs, this is potentially
1583 	 *	useful.
1584 	 */
1585 		INP_INFO_WLOCK(&V_tcbinfo);
1586 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1587 			if (inpb->inp_flags & INP_TIMEWAIT)
1588 				continue;
1589 			INP_WLOCK(inpb);
1590 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1591 				tcp_reass_flush(tcpb);
1592 				tcp_clean_sackreport(tcpb);
1593 			}
1594 			INP_WUNLOCK(inpb);
1595 		}
1596 		INP_INFO_WUNLOCK(&V_tcbinfo);
1597 		CURVNET_RESTORE();
1598 	}
1599 	VNET_LIST_RUNLOCK_NOSLEEP();
1600 }
1601 
1602 /*
1603  * Notify a tcp user of an asynchronous error;
1604  * store error as soft error, but wake up user
1605  * (for now, won't do anything until can select for soft error).
1606  *
1607  * Do not wake up user since there currently is no mechanism for
1608  * reporting soft errors (yet - a kqueue filter may be added).
1609  */
1610 static struct inpcb *
1611 tcp_notify(struct inpcb *inp, int error)
1612 {
1613 	struct tcpcb *tp;
1614 
1615 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1616 	INP_WLOCK_ASSERT(inp);
1617 
1618 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1619 	    (inp->inp_flags & INP_DROPPED))
1620 		return (inp);
1621 
1622 	tp = intotcpcb(inp);
1623 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1624 
1625 	/*
1626 	 * Ignore some errors if we are hooked up.
1627 	 * If connection hasn't completed, has retransmitted several times,
1628 	 * and receives a second error, give up now.  This is better
1629 	 * than waiting a long time to establish a connection that
1630 	 * can never complete.
1631 	 */
1632 	if (tp->t_state == TCPS_ESTABLISHED &&
1633 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1634 	     error == EHOSTDOWN)) {
1635 		if (inp->inp_route.ro_rt) {
1636 			RTFREE(inp->inp_route.ro_rt);
1637 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
1638 		}
1639 		return (inp);
1640 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1641 	    tp->t_softerror) {
1642 		tp = tcp_drop(tp, error);
1643 		if (tp != NULL)
1644 			return (inp);
1645 		else
1646 			return (NULL);
1647 	} else {
1648 		tp->t_softerror = error;
1649 		return (inp);
1650 	}
1651 #if 0
1652 	wakeup( &so->so_timeo);
1653 	sorwakeup(so);
1654 	sowwakeup(so);
1655 #endif
1656 }
1657 
1658 static int
1659 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1660 {
1661 	int error, i, m, n, pcb_count;
1662 	struct inpcb *inp, **inp_list;
1663 	inp_gen_t gencnt;
1664 	struct xinpgen xig;
1665 
1666 	/*
1667 	 * The process of preparing the TCB list is too time-consuming and
1668 	 * resource-intensive to repeat twice on every request.
1669 	 */
1670 	if (req->oldptr == NULL) {
1671 		n = V_tcbinfo.ipi_count +
1672 		    counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]);
1673 		n += imax(n / 8, 10);
1674 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1675 		return (0);
1676 	}
1677 
1678 	if (req->newptr != NULL)
1679 		return (EPERM);
1680 
1681 	/*
1682 	 * OK, now we're committed to doing something.
1683 	 */
1684 	INP_LIST_RLOCK(&V_tcbinfo);
1685 	gencnt = V_tcbinfo.ipi_gencnt;
1686 	n = V_tcbinfo.ipi_count;
1687 	INP_LIST_RUNLOCK(&V_tcbinfo);
1688 
1689 	m = counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]);
1690 
1691 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1692 		+ (n + m) * sizeof(struct xtcpcb));
1693 	if (error != 0)
1694 		return (error);
1695 
1696 	xig.xig_len = sizeof xig;
1697 	xig.xig_count = n + m;
1698 	xig.xig_gen = gencnt;
1699 	xig.xig_sogen = so_gencnt;
1700 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1701 	if (error)
1702 		return (error);
1703 
1704 	error = syncache_pcblist(req, m, &pcb_count);
1705 	if (error)
1706 		return (error);
1707 
1708 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1709 	if (inp_list == NULL)
1710 		return (ENOMEM);
1711 
1712 	INP_INFO_WLOCK(&V_tcbinfo);
1713 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1714 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1715 		INP_WLOCK(inp);
1716 		if (inp->inp_gencnt <= gencnt) {
1717 			/*
1718 			 * XXX: This use of cr_cansee(), introduced with
1719 			 * TCP state changes, is not quite right, but for
1720 			 * now, better than nothing.
1721 			 */
1722 			if (inp->inp_flags & INP_TIMEWAIT) {
1723 				if (intotw(inp) != NULL)
1724 					error = cr_cansee(req->td->td_ucred,
1725 					    intotw(inp)->tw_cred);
1726 				else
1727 					error = EINVAL;	/* Skip this inp. */
1728 			} else
1729 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1730 			if (error == 0) {
1731 				in_pcbref(inp);
1732 				inp_list[i++] = inp;
1733 			}
1734 		}
1735 		INP_WUNLOCK(inp);
1736 	}
1737 	INP_INFO_WUNLOCK(&V_tcbinfo);
1738 	n = i;
1739 
1740 	error = 0;
1741 	for (i = 0; i < n; i++) {
1742 		inp = inp_list[i];
1743 		INP_RLOCK(inp);
1744 		if (inp->inp_gencnt <= gencnt) {
1745 			struct xtcpcb xt;
1746 			void *inp_ppcb;
1747 
1748 			bzero(&xt, sizeof(xt));
1749 			xt.xt_len = sizeof xt;
1750 			/* XXX should avoid extra copy */
1751 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1752 			inp_ppcb = inp->inp_ppcb;
1753 			if (inp_ppcb == NULL)
1754 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1755 			else if (inp->inp_flags & INP_TIMEWAIT) {
1756 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1757 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1758 			} else {
1759 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1760 				if (xt.xt_tp.t_timers)
1761 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1762 			}
1763 			if (inp->inp_socket != NULL)
1764 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1765 			else {
1766 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1767 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1768 			}
1769 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1770 			INP_RUNLOCK(inp);
1771 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1772 		} else
1773 			INP_RUNLOCK(inp);
1774 	}
1775 	INP_INFO_RLOCK(&V_tcbinfo);
1776 	for (i = 0; i < n; i++) {
1777 		inp = inp_list[i];
1778 		INP_RLOCK(inp);
1779 		if (!in_pcbrele_rlocked(inp))
1780 			INP_RUNLOCK(inp);
1781 	}
1782 	INP_INFO_RUNLOCK(&V_tcbinfo);
1783 
1784 	if (!error) {
1785 		/*
1786 		 * Give the user an updated idea of our state.
1787 		 * If the generation differs from what we told
1788 		 * her before, she knows that something happened
1789 		 * while we were processing this request, and it
1790 		 * might be necessary to retry.
1791 		 */
1792 		INP_LIST_RLOCK(&V_tcbinfo);
1793 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1794 		xig.xig_sogen = so_gencnt;
1795 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1796 		INP_LIST_RUNLOCK(&V_tcbinfo);
1797 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1798 	}
1799 	free(inp_list, M_TEMP);
1800 	return (error);
1801 }
1802 
1803 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1804     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1805     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1806 
1807 #ifdef INET
1808 static int
1809 tcp_getcred(SYSCTL_HANDLER_ARGS)
1810 {
1811 	struct xucred xuc;
1812 	struct sockaddr_in addrs[2];
1813 	struct inpcb *inp;
1814 	int error;
1815 
1816 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1817 	if (error)
1818 		return (error);
1819 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1820 	if (error)
1821 		return (error);
1822 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1823 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1824 	if (inp != NULL) {
1825 		if (inp->inp_socket == NULL)
1826 			error = ENOENT;
1827 		if (error == 0)
1828 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1829 		if (error == 0)
1830 			cru2x(inp->inp_cred, &xuc);
1831 		INP_RUNLOCK(inp);
1832 	} else
1833 		error = ENOENT;
1834 	if (error == 0)
1835 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1836 	return (error);
1837 }
1838 
1839 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1840     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1841     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1842 #endif /* INET */
1843 
1844 #ifdef INET6
1845 static int
1846 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1847 {
1848 	struct xucred xuc;
1849 	struct sockaddr_in6 addrs[2];
1850 	struct inpcb *inp;
1851 	int error;
1852 #ifdef INET
1853 	int mapped = 0;
1854 #endif
1855 
1856 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1857 	if (error)
1858 		return (error);
1859 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1860 	if (error)
1861 		return (error);
1862 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1863 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1864 		return (error);
1865 	}
1866 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1867 #ifdef INET
1868 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1869 			mapped = 1;
1870 		else
1871 #endif
1872 			return (EINVAL);
1873 	}
1874 
1875 #ifdef INET
1876 	if (mapped == 1)
1877 		inp = in_pcblookup(&V_tcbinfo,
1878 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1879 			addrs[1].sin6_port,
1880 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1881 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1882 	else
1883 #endif
1884 		inp = in6_pcblookup(&V_tcbinfo,
1885 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1886 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1887 			INPLOOKUP_RLOCKPCB, NULL);
1888 	if (inp != NULL) {
1889 		if (inp->inp_socket == NULL)
1890 			error = ENOENT;
1891 		if (error == 0)
1892 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1893 		if (error == 0)
1894 			cru2x(inp->inp_cred, &xuc);
1895 		INP_RUNLOCK(inp);
1896 	} else
1897 		error = ENOENT;
1898 	if (error == 0)
1899 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1900 	return (error);
1901 }
1902 
1903 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1904     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1905     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1906 #endif /* INET6 */
1907 
1908 
1909 #ifdef INET
1910 void
1911 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1912 {
1913 	struct ip *ip = vip;
1914 	struct tcphdr *th;
1915 	struct in_addr faddr;
1916 	struct inpcb *inp;
1917 	struct tcpcb *tp;
1918 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1919 	struct icmp *icp;
1920 	struct in_conninfo inc;
1921 	tcp_seq icmp_tcp_seq;
1922 	int mtu;
1923 
1924 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1925 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1926 		return;
1927 
1928 	if (cmd == PRC_MSGSIZE)
1929 		notify = tcp_mtudisc_notify;
1930 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1931 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1932 		notify = tcp_drop_syn_sent;
1933 	else if (PRC_IS_REDIRECT(cmd)) {
1934 		/* signal EHOSTDOWN, as it flushes the cached route */
1935 		in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify);
1936 		return;
1937 	}
1938 	/*
1939 	 * Hostdead is ugly because it goes linearly through all PCBs.
1940 	 * XXX: We never get this from ICMP, otherwise it makes an
1941 	 * excellent DoS attack on machines with many connections.
1942 	 */
1943 	else if (cmd == PRC_HOSTDEAD)
1944 		ip = NULL;
1945 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1946 		return;
1947 
1948 	if (ip == NULL) {
1949 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1950 		return;
1951 	}
1952 
1953 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1954 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1955 	INP_INFO_RLOCK(&V_tcbinfo);
1956 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1957 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1958 	if (inp != NULL)  {
1959 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
1960 		    !(inp->inp_flags & INP_DROPPED) &&
1961 		    !(inp->inp_socket == NULL)) {
1962 			icmp_tcp_seq = ntohl(th->th_seq);
1963 			tp = intotcpcb(inp);
1964 			if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1965 			    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1966 				if (cmd == PRC_MSGSIZE) {
1967 					/*
1968 					 * MTU discovery:
1969 					 * If we got a needfrag set the MTU
1970 					 * in the route to the suggested new
1971 					 * value (if given) and then notify.
1972 					 */
1973 				    	mtu = ntohs(icp->icmp_nextmtu);
1974 					/*
1975 					 * If no alternative MTU was
1976 					 * proposed, try the next smaller
1977 					 * one.
1978 					 */
1979 					if (!mtu)
1980 						mtu = ip_next_mtu(
1981 						    ntohs(ip->ip_len), 1);
1982 					if (mtu < V_tcp_minmss +
1983 					    sizeof(struct tcpiphdr))
1984 						mtu = V_tcp_minmss +
1985 						    sizeof(struct tcpiphdr);
1986 					/*
1987 					 * Only process the offered MTU if it
1988 					 * is smaller than the current one.
1989 					 */
1990 					if (mtu < tp->t_maxseg +
1991 					    sizeof(struct tcpiphdr)) {
1992 						bzero(&inc, sizeof(inc));
1993 						inc.inc_faddr = faddr;
1994 						inc.inc_fibnum =
1995 						    inp->inp_inc.inc_fibnum;
1996 						tcp_hc_updatemtu(&inc, mtu);
1997 						tcp_mtudisc(inp, mtu);
1998 					}
1999 				} else
2000 					inp = (*notify)(inp,
2001 					    inetctlerrmap[cmd]);
2002 			}
2003 		}
2004 		if (inp != NULL)
2005 			INP_WUNLOCK(inp);
2006 	} else {
2007 		bzero(&inc, sizeof(inc));
2008 		inc.inc_fport = th->th_dport;
2009 		inc.inc_lport = th->th_sport;
2010 		inc.inc_faddr = faddr;
2011 		inc.inc_laddr = ip->ip_src;
2012 		syncache_unreach(&inc, th);
2013 	}
2014 	INP_INFO_RUNLOCK(&V_tcbinfo);
2015 }
2016 #endif /* INET */
2017 
2018 #ifdef INET6
2019 void
2020 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2021 {
2022 	struct tcphdr th;
2023 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2024 	struct ip6_hdr *ip6;
2025 	struct mbuf *m;
2026 	struct ip6ctlparam *ip6cp = NULL;
2027 	const struct sockaddr_in6 *sa6_src = NULL;
2028 	int off;
2029 	struct tcp_portonly {
2030 		u_int16_t th_sport;
2031 		u_int16_t th_dport;
2032 	} *thp;
2033 
2034 	if (sa->sa_family != AF_INET6 ||
2035 	    sa->sa_len != sizeof(struct sockaddr_in6))
2036 		return;
2037 
2038 	if (cmd == PRC_MSGSIZE)
2039 		notify = tcp_mtudisc_notify;
2040 	else if (!PRC_IS_REDIRECT(cmd) &&
2041 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
2042 		return;
2043 
2044 	/* if the parameter is from icmp6, decode it. */
2045 	if (d != NULL) {
2046 		ip6cp = (struct ip6ctlparam *)d;
2047 		m = ip6cp->ip6c_m;
2048 		ip6 = ip6cp->ip6c_ip6;
2049 		off = ip6cp->ip6c_off;
2050 		sa6_src = ip6cp->ip6c_src;
2051 	} else {
2052 		m = NULL;
2053 		ip6 = NULL;
2054 		off = 0;	/* fool gcc */
2055 		sa6_src = &sa6_any;
2056 	}
2057 
2058 	if (ip6 != NULL) {
2059 		struct in_conninfo inc;
2060 		/*
2061 		 * XXX: We assume that when IPV6 is non NULL,
2062 		 * M and OFF are valid.
2063 		 */
2064 
2065 		/* check if we can safely examine src and dst ports */
2066 		if (m->m_pkthdr.len < off + sizeof(*thp))
2067 			return;
2068 
2069 		bzero(&th, sizeof(th));
2070 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2071 
2072 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
2073 		    (struct sockaddr *)ip6cp->ip6c_src,
2074 		    th.th_sport, cmd, NULL, notify);
2075 
2076 		bzero(&inc, sizeof(inc));
2077 		inc.inc_fport = th.th_dport;
2078 		inc.inc_lport = th.th_sport;
2079 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
2080 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
2081 		inc.inc_flags |= INC_ISIPV6;
2082 		INP_INFO_RLOCK(&V_tcbinfo);
2083 		syncache_unreach(&inc, &th);
2084 		INP_INFO_RUNLOCK(&V_tcbinfo);
2085 	} else
2086 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2087 			      0, cmd, NULL, notify);
2088 }
2089 #endif /* INET6 */
2090 
2091 
2092 /*
2093  * Following is where TCP initial sequence number generation occurs.
2094  *
2095  * There are two places where we must use initial sequence numbers:
2096  * 1.  In SYN-ACK packets.
2097  * 2.  In SYN packets.
2098  *
2099  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2100  * tcp_syncache.c for details.
2101  *
2102  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2103  * depends on this property.  In addition, these ISNs should be
2104  * unguessable so as to prevent connection hijacking.  To satisfy
2105  * the requirements of this situation, the algorithm outlined in
2106  * RFC 1948 is used, with only small modifications.
2107  *
2108  * Implementation details:
2109  *
2110  * Time is based off the system timer, and is corrected so that it
2111  * increases by one megabyte per second.  This allows for proper
2112  * recycling on high speed LANs while still leaving over an hour
2113  * before rollover.
2114  *
2115  * As reading the *exact* system time is too expensive to be done
2116  * whenever setting up a TCP connection, we increment the time
2117  * offset in two ways.  First, a small random positive increment
2118  * is added to isn_offset for each connection that is set up.
2119  * Second, the function tcp_isn_tick fires once per clock tick
2120  * and increments isn_offset as necessary so that sequence numbers
2121  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2122  * random positive increments serve only to ensure that the same
2123  * exact sequence number is never sent out twice (as could otherwise
2124  * happen when a port is recycled in less than the system tick
2125  * interval.)
2126  *
2127  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2128  * between seeding of isn_secret.  This is normally set to zero,
2129  * as reseeding should not be necessary.
2130  *
2131  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2132  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2133  * general, this means holding an exclusive (write) lock.
2134  */
2135 
2136 #define ISN_BYTES_PER_SECOND 1048576
2137 #define ISN_STATIC_INCREMENT 4096
2138 #define ISN_RANDOM_INCREMENT (4096 - 1)
2139 
2140 static VNET_DEFINE(u_char, isn_secret[32]);
2141 static VNET_DEFINE(int, isn_last);
2142 static VNET_DEFINE(int, isn_last_reseed);
2143 static VNET_DEFINE(u_int32_t, isn_offset);
2144 static VNET_DEFINE(u_int32_t, isn_offset_old);
2145 
2146 #define	V_isn_secret			VNET(isn_secret)
2147 #define	V_isn_last			VNET(isn_last)
2148 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2149 #define	V_isn_offset			VNET(isn_offset)
2150 #define	V_isn_offset_old		VNET(isn_offset_old)
2151 
2152 tcp_seq
2153 tcp_new_isn(struct tcpcb *tp)
2154 {
2155 	MD5_CTX isn_ctx;
2156 	u_int32_t md5_buffer[4];
2157 	tcp_seq new_isn;
2158 	u_int32_t projected_offset;
2159 
2160 	INP_WLOCK_ASSERT(tp->t_inpcb);
2161 
2162 	ISN_LOCK();
2163 	/* Seed if this is the first use, reseed if requested. */
2164 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2165 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2166 		< (u_int)ticks))) {
2167 		read_random(&V_isn_secret, sizeof(V_isn_secret));
2168 		V_isn_last_reseed = ticks;
2169 	}
2170 
2171 	/* Compute the md5 hash and return the ISN. */
2172 	MD5Init(&isn_ctx);
2173 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2174 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2175 #ifdef INET6
2176 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2177 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2178 			  sizeof(struct in6_addr));
2179 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2180 			  sizeof(struct in6_addr));
2181 	} else
2182 #endif
2183 	{
2184 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2185 			  sizeof(struct in_addr));
2186 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2187 			  sizeof(struct in_addr));
2188 	}
2189 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2190 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
2191 	new_isn = (tcp_seq) md5_buffer[0];
2192 	V_isn_offset += ISN_STATIC_INCREMENT +
2193 		(arc4random() & ISN_RANDOM_INCREMENT);
2194 	if (ticks != V_isn_last) {
2195 		projected_offset = V_isn_offset_old +
2196 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2197 		if (SEQ_GT(projected_offset, V_isn_offset))
2198 			V_isn_offset = projected_offset;
2199 		V_isn_offset_old = V_isn_offset;
2200 		V_isn_last = ticks;
2201 	}
2202 	new_isn += V_isn_offset;
2203 	ISN_UNLOCK();
2204 	return (new_isn);
2205 }
2206 
2207 /*
2208  * When a specific ICMP unreachable message is received and the
2209  * connection state is SYN-SENT, drop the connection.  This behavior
2210  * is controlled by the icmp_may_rst sysctl.
2211  */
2212 struct inpcb *
2213 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2214 {
2215 	struct tcpcb *tp;
2216 
2217 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2218 	INP_WLOCK_ASSERT(inp);
2219 
2220 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2221 	    (inp->inp_flags & INP_DROPPED))
2222 		return (inp);
2223 
2224 	tp = intotcpcb(inp);
2225 	if (tp->t_state != TCPS_SYN_SENT)
2226 		return (inp);
2227 
2228 	tp = tcp_drop(tp, errno);
2229 	if (tp != NULL)
2230 		return (inp);
2231 	else
2232 		return (NULL);
2233 }
2234 
2235 /*
2236  * When `need fragmentation' ICMP is received, update our idea of the MSS
2237  * based on the new value. Also nudge TCP to send something, since we
2238  * know the packet we just sent was dropped.
2239  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2240  */
2241 static struct inpcb *
2242 tcp_mtudisc_notify(struct inpcb *inp, int error)
2243 {
2244 
2245 	tcp_mtudisc(inp, -1);
2246 	return (inp);
2247 }
2248 
2249 static void
2250 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2251 {
2252 	struct tcpcb *tp;
2253 	struct socket *so;
2254 
2255 	INP_WLOCK_ASSERT(inp);
2256 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2257 	    (inp->inp_flags & INP_DROPPED))
2258 		return;
2259 
2260 	tp = intotcpcb(inp);
2261 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2262 
2263 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2264 
2265 	so = inp->inp_socket;
2266 	SOCKBUF_LOCK(&so->so_snd);
2267 	/* If the mss is larger than the socket buffer, decrease the mss. */
2268 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2269 		tp->t_maxseg = so->so_snd.sb_hiwat;
2270 	SOCKBUF_UNLOCK(&so->so_snd);
2271 
2272 	TCPSTAT_INC(tcps_mturesent);
2273 	tp->t_rtttime = 0;
2274 	tp->snd_nxt = tp->snd_una;
2275 	tcp_free_sackholes(tp);
2276 	tp->snd_recover = tp->snd_max;
2277 	if (tp->t_flags & TF_SACK_PERMIT)
2278 		EXIT_FASTRECOVERY(tp->t_flags);
2279 	tp->t_fb->tfb_tcp_output(tp);
2280 }
2281 
2282 #ifdef INET
2283 /*
2284  * Look-up the routing entry to the peer of this inpcb.  If no route
2285  * is found and it cannot be allocated, then return 0.  This routine
2286  * is called by TCP routines that access the rmx structure and by
2287  * tcp_mss_update to get the peer/interface MTU.
2288  */
2289 u_long
2290 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2291 {
2292 	struct nhop4_extended nh4;
2293 	struct ifnet *ifp;
2294 	u_long maxmtu = 0;
2295 
2296 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2297 
2298 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2299 
2300 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2301 		    NHR_REF, 0, &nh4) != 0)
2302 			return (0);
2303 
2304 		ifp = nh4.nh_ifp;
2305 		maxmtu = nh4.nh_mtu;
2306 
2307 		/* Report additional interface capabilities. */
2308 		if (cap != NULL) {
2309 			if (ifp->if_capenable & IFCAP_TSO4 &&
2310 			    ifp->if_hwassist & CSUM_TSO) {
2311 				cap->ifcap |= CSUM_TSO;
2312 				cap->tsomax = ifp->if_hw_tsomax;
2313 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2314 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2315 			}
2316 		}
2317 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2318 	}
2319 	return (maxmtu);
2320 }
2321 #endif /* INET */
2322 
2323 #ifdef INET6
2324 u_long
2325 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2326 {
2327 	struct nhop6_extended nh6;
2328 	struct in6_addr dst6;
2329 	uint32_t scopeid;
2330 	struct ifnet *ifp;
2331 	u_long maxmtu = 0;
2332 
2333 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2334 
2335 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2336 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2337 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2338 		    0, &nh6) != 0)
2339 			return (0);
2340 
2341 		ifp = nh6.nh_ifp;
2342 		maxmtu = nh6.nh_mtu;
2343 
2344 		/* Report additional interface capabilities. */
2345 		if (cap != NULL) {
2346 			if (ifp->if_capenable & IFCAP_TSO6 &&
2347 			    ifp->if_hwassist & CSUM_TSO) {
2348 				cap->ifcap |= CSUM_TSO;
2349 				cap->tsomax = ifp->if_hw_tsomax;
2350 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2351 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2352 			}
2353 		}
2354 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2355 	}
2356 
2357 	return (maxmtu);
2358 }
2359 #endif /* INET6 */
2360 
2361 /*
2362  * Calculate effective SMSS per RFC5681 definition for a given TCP
2363  * connection at its current state, taking into account SACK and etc.
2364  */
2365 u_int
2366 tcp_maxseg(const struct tcpcb *tp)
2367 {
2368 	u_int optlen;
2369 
2370 	if (tp->t_flags & TF_NOOPT)
2371 		return (tp->t_maxseg);
2372 
2373 	/*
2374 	 * Here we have a simplified code from tcp_addoptions(),
2375 	 * without a proper loop, and having most of paddings hardcoded.
2376 	 * We might make mistakes with padding here in some edge cases,
2377 	 * but this is harmless, since result of tcp_maxseg() is used
2378 	 * only in cwnd and ssthresh estimations.
2379 	 */
2380 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2381 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2382 		if (tp->t_flags & TF_RCVD_TSTMP)
2383 			optlen = TCPOLEN_TSTAMP_APPA;
2384 		else
2385 			optlen = 0;
2386 #ifdef TCP_SIGNATURE
2387 		if (tp->t_flags & TF_SIGNATURE)
2388 			optlen += PAD(TCPOLEN_SIGNATURE);
2389 #endif
2390 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2391 			optlen += TCPOLEN_SACKHDR;
2392 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2393 			optlen = PAD(optlen);
2394 		}
2395 	} else {
2396 		if (tp->t_flags & TF_REQ_TSTMP)
2397 			optlen = TCPOLEN_TSTAMP_APPA;
2398 		else
2399 			optlen = PAD(TCPOLEN_MAXSEG);
2400 		if (tp->t_flags & TF_REQ_SCALE)
2401 			optlen += PAD(TCPOLEN_WINDOW);
2402 #ifdef TCP_SIGNATURE
2403 		if (tp->t_flags & TF_SIGNATURE)
2404 			optlen += PAD(TCPOLEN_SIGNATURE);
2405 #endif
2406 		if (tp->t_flags & TF_SACK_PERMIT)
2407 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2408 	}
2409 #undef PAD
2410 	optlen = min(optlen, TCP_MAXOLEN);
2411 	return (tp->t_maxseg - optlen);
2412 }
2413 
2414 #ifdef IPSEC
2415 /* compute ESP/AH header size for TCP, including outer IP header. */
2416 size_t
2417 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2418 {
2419 	struct inpcb *inp;
2420 	struct mbuf *m;
2421 	size_t hdrsiz;
2422 	struct ip *ip;
2423 #ifdef INET6
2424 	struct ip6_hdr *ip6;
2425 #endif
2426 	struct tcphdr *th;
2427 
2428 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2429 		(!key_havesp(IPSEC_DIR_OUTBOUND)))
2430 		return (0);
2431 	m = m_gethdr(M_NOWAIT, MT_DATA);
2432 	if (!m)
2433 		return (0);
2434 
2435 #ifdef INET6
2436 	if ((inp->inp_vflag & INP_IPV6) != 0) {
2437 		ip6 = mtod(m, struct ip6_hdr *);
2438 		th = (struct tcphdr *)(ip6 + 1);
2439 		m->m_pkthdr.len = m->m_len =
2440 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2441 		tcpip_fillheaders(inp, ip6, th);
2442 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2443 	} else
2444 #endif /* INET6 */
2445 	{
2446 		ip = mtod(m, struct ip *);
2447 		th = (struct tcphdr *)(ip + 1);
2448 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2449 		tcpip_fillheaders(inp, ip, th);
2450 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2451 	}
2452 
2453 	m_free(m);
2454 	return (hdrsiz);
2455 }
2456 #endif /* IPSEC */
2457 
2458 #ifdef TCP_SIGNATURE
2459 /*
2460  * Callback function invoked by m_apply() to digest TCP segment data
2461  * contained within an mbuf chain.
2462  */
2463 static int
2464 tcp_signature_apply(void *fstate, void *data, u_int len)
2465 {
2466 
2467 	MD5Update(fstate, (u_char *)data, len);
2468 	return (0);
2469 }
2470 
2471 /*
2472  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2473  * search with the destination IP address, and a 'magic SPI' to be
2474  * determined by the application. This is hardcoded elsewhere to 1179
2475 */
2476 struct secasvar *
2477 tcp_get_sav(struct mbuf *m, u_int direction)
2478 {
2479 	union sockaddr_union dst;
2480 	struct secasvar *sav;
2481 	struct ip *ip;
2482 #ifdef INET6
2483 	struct ip6_hdr *ip6;
2484 	char ip6buf[INET6_ADDRSTRLEN];
2485 #endif
2486 
2487 	/* Extract the destination from the IP header in the mbuf. */
2488 	bzero(&dst, sizeof(union sockaddr_union));
2489 	ip = mtod(m, struct ip *);
2490 #ifdef INET6
2491 	ip6 = NULL;	/* Make the compiler happy. */
2492 #endif
2493 	switch (ip->ip_v) {
2494 #ifdef INET
2495 	case IPVERSION:
2496 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2497 		dst.sa.sa_family = AF_INET;
2498 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2499 		    ip->ip_src : ip->ip_dst;
2500 		break;
2501 #endif
2502 #ifdef INET6
2503 	case (IPV6_VERSION >> 4):
2504 		ip6 = mtod(m, struct ip6_hdr *);
2505 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2506 		dst.sa.sa_family = AF_INET6;
2507 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2508 		    ip6->ip6_src : ip6->ip6_dst;
2509 		break;
2510 #endif
2511 	default:
2512 		return (NULL);
2513 		/* NOTREACHED */
2514 		break;
2515 	}
2516 
2517 	/* Look up an SADB entry which matches the address of the peer. */
2518 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2519 	if (sav == NULL) {
2520 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2521 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2522 #ifdef INET6
2523 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2524 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2525 #endif
2526 			"(unsupported)"));
2527 	}
2528 
2529 	return (sav);
2530 }
2531 
2532 /*
2533  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2534  *
2535  * Parameters:
2536  * m		pointer to head of mbuf chain
2537  * len		length of TCP segment data, excluding options
2538  * optlen	length of TCP segment options
2539  * buf		pointer to storage for computed MD5 digest
2540  * sav		pointer to security assosiation
2541  *
2542  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2543  * When called from tcp_input(), we can be sure that th_sum has been
2544  * zeroed out and verified already.
2545  *
2546  * Releases reference to SADB key before return.
2547  *
2548  * Return 0 if successful, otherwise return -1.
2549  *
2550  */
2551 int
2552 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2553     u_char *buf, struct secasvar *sav)
2554 {
2555 #ifdef INET
2556 	struct ippseudo ippseudo;
2557 #endif
2558 	MD5_CTX ctx;
2559 	int doff;
2560 	struct ip *ip;
2561 #ifdef INET
2562 	struct ipovly *ipovly;
2563 #endif
2564 	struct tcphdr *th;
2565 #ifdef INET6
2566 	struct ip6_hdr *ip6;
2567 	struct in6_addr in6;
2568 	uint32_t plen;
2569 	uint16_t nhdr;
2570 #endif
2571 	u_short savecsum;
2572 
2573 	KASSERT(m != NULL, ("NULL mbuf chain"));
2574 	KASSERT(buf != NULL, ("NULL signature pointer"));
2575 
2576 	/* Extract the destination from the IP header in the mbuf. */
2577 	ip = mtod(m, struct ip *);
2578 #ifdef INET6
2579 	ip6 = NULL;	/* Make the compiler happy. */
2580 #endif
2581 
2582 	MD5Init(&ctx);
2583 	/*
2584 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2585 	 *
2586 	 * XXX The ippseudo header MUST be digested in network byte order,
2587 	 * or else we'll fail the regression test. Assume all fields we've
2588 	 * been doing arithmetic on have been in host byte order.
2589 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2590 	 * tcp_output(), the underlying ip_len member has not yet been set.
2591 	 */
2592 	switch (ip->ip_v) {
2593 #ifdef INET
2594 	case IPVERSION:
2595 		ipovly = (struct ipovly *)ip;
2596 		ippseudo.ippseudo_src = ipovly->ih_src;
2597 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2598 		ippseudo.ippseudo_pad = 0;
2599 		ippseudo.ippseudo_p = IPPROTO_TCP;
2600 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2601 		    optlen);
2602 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2603 
2604 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2605 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2606 		break;
2607 #endif
2608 #ifdef INET6
2609 	/*
2610 	 * RFC 2385, 2.0  Proposal
2611 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2612 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2613 	 * extended next header value (to form 32 bits), and 32-bit segment
2614 	 * length.
2615 	 * Note: Upper-Layer Packet Length comes before Next Header.
2616 	 */
2617 	case (IPV6_VERSION >> 4):
2618 		in6 = ip6->ip6_src;
2619 		in6_clearscope(&in6);
2620 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2621 		in6 = ip6->ip6_dst;
2622 		in6_clearscope(&in6);
2623 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2624 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2625 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2626 		nhdr = 0;
2627 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2628 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2629 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2630 		nhdr = IPPROTO_TCP;
2631 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2632 
2633 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2634 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2635 		break;
2636 #endif
2637 	default:
2638 		KEY_FREESAV(&sav);
2639 		return (-1);
2640 		/* NOTREACHED */
2641 		break;
2642 	}
2643 
2644 
2645 	/*
2646 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2647 	 * The TCP checksum must be set to zero.
2648 	 */
2649 	savecsum = th->th_sum;
2650 	th->th_sum = 0;
2651 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2652 	th->th_sum = savecsum;
2653 
2654 	/*
2655 	 * Step 3: Update MD5 hash with TCP segment data.
2656 	 *         Use m_apply() to avoid an early m_pullup().
2657 	 */
2658 	if (len > 0)
2659 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2660 
2661 	/*
2662 	 * Step 4: Update MD5 hash with shared secret.
2663 	 */
2664 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2665 	MD5Final(buf, &ctx);
2666 
2667 	key_sa_recordxfer(sav, m);
2668 	KEY_FREESAV(&sav);
2669 	return (0);
2670 }
2671 
2672 /*
2673  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2674  *
2675  * Return 0 if successful, otherwise return -1.
2676  */
2677 int
2678 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2679     u_char *buf, u_int direction)
2680 {
2681 	struct secasvar *sav;
2682 
2683 	if ((sav = tcp_get_sav(m, direction)) == NULL)
2684 		return (-1);
2685 
2686 	return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2687 }
2688 
2689 /*
2690  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2691  *
2692  * Parameters:
2693  * m		pointer to head of mbuf chain
2694  * len		length of TCP segment data, excluding options
2695  * optlen	length of TCP segment options
2696  * buf		pointer to storage for computed MD5 digest
2697  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2698  *
2699  * Return 1 if successful, otherwise return 0.
2700  */
2701 int
2702 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2703     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2704 {
2705 	char tmpdigest[TCP_SIGLEN];
2706 
2707 	if (tcp_sig_checksigs == 0)
2708 		return (1);
2709 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2710 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2711 
2712 			/*
2713 			 * If this socket is not expecting signature but
2714 			 * the segment contains signature just fail.
2715 			 */
2716 			TCPSTAT_INC(tcps_sig_err_sigopt);
2717 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2718 			return (0);
2719 		}
2720 
2721 		/* Signature is not expected, and not present in segment. */
2722 		return (1);
2723 	}
2724 
2725 	/*
2726 	 * If this socket is expecting signature but the segment does not
2727 	 * contain any just fail.
2728 	 */
2729 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2730 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2731 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2732 		return (0);
2733 	}
2734 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2735 	    IPSEC_DIR_INBOUND) == -1) {
2736 		TCPSTAT_INC(tcps_sig_err_buildsig);
2737 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2738 		return (0);
2739 	}
2740 
2741 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2742 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2743 		return (0);
2744 	}
2745 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2746 	return (1);
2747 }
2748 #endif /* TCP_SIGNATURE */
2749 
2750 static int
2751 sysctl_drop(SYSCTL_HANDLER_ARGS)
2752 {
2753 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2754 	struct sockaddr_storage addrs[2];
2755 	struct inpcb *inp;
2756 	struct tcpcb *tp;
2757 	struct tcptw *tw;
2758 	struct sockaddr_in *fin, *lin;
2759 #ifdef INET6
2760 	struct sockaddr_in6 *fin6, *lin6;
2761 #endif
2762 	int error;
2763 
2764 	inp = NULL;
2765 	fin = lin = NULL;
2766 #ifdef INET6
2767 	fin6 = lin6 = NULL;
2768 #endif
2769 	error = 0;
2770 
2771 	if (req->oldptr != NULL || req->oldlen != 0)
2772 		return (EINVAL);
2773 	if (req->newptr == NULL)
2774 		return (EPERM);
2775 	if (req->newlen < sizeof(addrs))
2776 		return (ENOMEM);
2777 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2778 	if (error)
2779 		return (error);
2780 
2781 	switch (addrs[0].ss_family) {
2782 #ifdef INET6
2783 	case AF_INET6:
2784 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2785 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2786 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2787 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2788 			return (EINVAL);
2789 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2790 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2791 				return (EINVAL);
2792 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2793 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2794 			fin = (struct sockaddr_in *)&addrs[0];
2795 			lin = (struct sockaddr_in *)&addrs[1];
2796 			break;
2797 		}
2798 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2799 		if (error)
2800 			return (error);
2801 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2802 		if (error)
2803 			return (error);
2804 		break;
2805 #endif
2806 #ifdef INET
2807 	case AF_INET:
2808 		fin = (struct sockaddr_in *)&addrs[0];
2809 		lin = (struct sockaddr_in *)&addrs[1];
2810 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2811 		    lin->sin_len != sizeof(struct sockaddr_in))
2812 			return (EINVAL);
2813 		break;
2814 #endif
2815 	default:
2816 		return (EINVAL);
2817 	}
2818 	INP_INFO_RLOCK(&V_tcbinfo);
2819 	switch (addrs[0].ss_family) {
2820 #ifdef INET6
2821 	case AF_INET6:
2822 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2823 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2824 		    INPLOOKUP_WLOCKPCB, NULL);
2825 		break;
2826 #endif
2827 #ifdef INET
2828 	case AF_INET:
2829 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2830 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2831 		break;
2832 #endif
2833 	}
2834 	if (inp != NULL) {
2835 		if (inp->inp_flags & INP_TIMEWAIT) {
2836 			/*
2837 			 * XXXRW: There currently exists a state where an
2838 			 * inpcb is present, but its timewait state has been
2839 			 * discarded.  For now, don't allow dropping of this
2840 			 * type of inpcb.
2841 			 */
2842 			tw = intotw(inp);
2843 			if (tw != NULL)
2844 				tcp_twclose(tw, 0);
2845 			else
2846 				INP_WUNLOCK(inp);
2847 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2848 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2849 			tp = intotcpcb(inp);
2850 			tp = tcp_drop(tp, ECONNABORTED);
2851 			if (tp != NULL)
2852 				INP_WUNLOCK(inp);
2853 		} else
2854 			INP_WUNLOCK(inp);
2855 	} else
2856 		error = ESRCH;
2857 	INP_INFO_RUNLOCK(&V_tcbinfo);
2858 	return (error);
2859 }
2860 
2861 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2862     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2863     0, sysctl_drop, "", "Drop TCP connection");
2864 
2865 /*
2866  * Generate a standardized TCP log line for use throughout the
2867  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2868  * allow use in the interrupt context.
2869  *
2870  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2871  * NB: The function may return NULL if memory allocation failed.
2872  *
2873  * Due to header inclusion and ordering limitations the struct ip
2874  * and ip6_hdr pointers have to be passed as void pointers.
2875  */
2876 char *
2877 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2878     const void *ip6hdr)
2879 {
2880 
2881 	/* Is logging enabled? */
2882 	if (tcp_log_in_vain == 0)
2883 		return (NULL);
2884 
2885 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2886 }
2887 
2888 char *
2889 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2890     const void *ip6hdr)
2891 {
2892 
2893 	/* Is logging enabled? */
2894 	if (tcp_log_debug == 0)
2895 		return (NULL);
2896 
2897 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2898 }
2899 
2900 static char *
2901 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2902     const void *ip6hdr)
2903 {
2904 	char *s, *sp;
2905 	size_t size;
2906 	struct ip *ip;
2907 #ifdef INET6
2908 	const struct ip6_hdr *ip6;
2909 
2910 	ip6 = (const struct ip6_hdr *)ip6hdr;
2911 #endif /* INET6 */
2912 	ip = (struct ip *)ip4hdr;
2913 
2914 	/*
2915 	 * The log line looks like this:
2916 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2917 	 */
2918 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2919 	    sizeof(PRINT_TH_FLAGS) + 1 +
2920 #ifdef INET6
2921 	    2 * INET6_ADDRSTRLEN;
2922 #else
2923 	    2 * INET_ADDRSTRLEN;
2924 #endif /* INET6 */
2925 
2926 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2927 	if (s == NULL)
2928 		return (NULL);
2929 
2930 	strcat(s, "TCP: [");
2931 	sp = s + strlen(s);
2932 
2933 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2934 		inet_ntoa_r(inc->inc_faddr, sp);
2935 		sp = s + strlen(s);
2936 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2937 		sp = s + strlen(s);
2938 		inet_ntoa_r(inc->inc_laddr, sp);
2939 		sp = s + strlen(s);
2940 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2941 #ifdef INET6
2942 	} else if (inc) {
2943 		ip6_sprintf(sp, &inc->inc6_faddr);
2944 		sp = s + strlen(s);
2945 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2946 		sp = s + strlen(s);
2947 		ip6_sprintf(sp, &inc->inc6_laddr);
2948 		sp = s + strlen(s);
2949 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2950 	} else if (ip6 && th) {
2951 		ip6_sprintf(sp, &ip6->ip6_src);
2952 		sp = s + strlen(s);
2953 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2954 		sp = s + strlen(s);
2955 		ip6_sprintf(sp, &ip6->ip6_dst);
2956 		sp = s + strlen(s);
2957 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2958 #endif /* INET6 */
2959 #ifdef INET
2960 	} else if (ip && th) {
2961 		inet_ntoa_r(ip->ip_src, sp);
2962 		sp = s + strlen(s);
2963 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2964 		sp = s + strlen(s);
2965 		inet_ntoa_r(ip->ip_dst, sp);
2966 		sp = s + strlen(s);
2967 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2968 #endif /* INET */
2969 	} else {
2970 		free(s, M_TCPLOG);
2971 		return (NULL);
2972 	}
2973 	sp = s + strlen(s);
2974 	if (th)
2975 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2976 	if (*(s + size - 1) != '\0')
2977 		panic("%s: string too long", __func__);
2978 	return (s);
2979 }
2980 
2981 /*
2982  * A subroutine which makes it easy to track TCP state changes with DTrace.
2983  * This function shouldn't be called for t_state initializations that don't
2984  * correspond to actual TCP state transitions.
2985  */
2986 void
2987 tcp_state_change(struct tcpcb *tp, int newstate)
2988 {
2989 #if defined(KDTRACE_HOOKS)
2990 	int pstate = tp->t_state;
2991 #endif
2992 
2993 	TCPSTATES_DEC(tp->t_state);
2994 	TCPSTATES_INC(newstate);
2995 	tp->t_state = newstate;
2996 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2997 }
2998