2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
46 #include <sys/hhook.h>
48 #include <sys/kernel.h>
50 #include <sys/khelp.h>
52 #include <sys/sysctl.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
58 #include <sys/domain.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/protosw.h>
66 #include <sys/random.h>
70 #include <net/route.h>
72 #include <net/if_var.h>
75 #include <netinet/in.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_kdtrace.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_icmp.h>
83 #include <netinet/ip_var.h>
85 #include <netinet/icmp6.h>
86 #include <netinet/ip6.h>
87 #include <netinet6/in6_fib.h>
88 #include <netinet6/in6_pcb.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/scope6_var.h>
91 #include <netinet6/nd6.h>
95 #include <netinet/tcp_fastopen.h>
97 #include <netinet/tcp.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 #include <netinet/tcp_syncache.h>
103 #include <netinet/cc/cc.h>
105 #include <netinet6/tcp6_var.h>
107 #include <netinet/tcpip.h>
109 #include <netinet/tcp_pcap.h>
112 #include <netinet/tcp_debug.h>
115 #include <netinet6/ip6protosw.h>
118 #include <netinet/tcp_offload.h>
122 #include <netipsec/ipsec.h>
123 #include <netipsec/xform.h>
125 #include <netipsec/ipsec6.h>
127 #include <netipsec/key.h>
128 #include <sys/syslog.h>
131 #include <machine/in_cksum.h>
134 #include <security/mac/mac_framework.h>
136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
141 struct rwlock tcp_function_lock;
144 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
149 error = sysctl_handle_int(oidp, &new, 0, req);
150 if (error == 0 && req->newptr) {
151 if (new < TCP_MINMSS)
159 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
160 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
161 &sysctl_net_inet_tcp_mss_check, "I",
162 "Default TCP Maximum Segment Size");
166 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
170 new = V_tcp_v6mssdflt;
171 error = sysctl_handle_int(oidp, &new, 0, req);
172 if (error == 0 && req->newptr) {
173 if (new < TCP_MINMSS)
176 V_tcp_v6mssdflt = new;
181 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
183 &sysctl_net_inet_tcp_mss_v6_check, "I",
184 "Default TCP Maximum Segment Size for IPv6");
188 * Minimum MSS we accept and use. This prevents DoS attacks where
189 * we are forced to a ridiculous low MSS like 20 and send hundreds
190 * of packets instead of one. The effect scales with the available
191 * bandwidth and quickly saturates the CPU and network interface
192 * with packet generation and sending. Set to zero to disable MINMSS
193 * checking. This setting prevents us from sending too small packets.
195 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_minmss), 0,
198 "Minimum TCP Maximum Segment Size");
200 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
201 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_do_rfc1323), 0,
203 "Enable rfc1323 (high performance TCP) extensions");
205 static int tcp_log_debug = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
207 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
209 static int tcp_tcbhashsize;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
211 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
213 static int do_tcpdrain = 1;
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
215 "Enable tcp_drain routine for extra help when low on mbufs");
217 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
218 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
220 static VNET_DEFINE(int, icmp_may_rst) = 1;
221 #define V_icmp_may_rst VNET(icmp_may_rst)
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
223 &VNET_NAME(icmp_may_rst), 0,
224 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
226 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
227 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
229 &VNET_NAME(tcp_isn_reseed_interval), 0,
230 "Seconds between reseeding of ISN secret");
232 static int tcp_soreceive_stream;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
234 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
237 static int tcp_sig_checksigs = 1;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
239 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
242 VNET_DEFINE(uma_zone_t, sack_hole_zone);
243 #define V_sack_hole_zone VNET(sack_hole_zone)
246 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
249 static struct inpcb *tcp_notify(struct inpcb *, int);
250 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
251 static void tcp_mtudisc(struct inpcb *, int);
252 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
253 void *ip4hdr, const void *ip6hdr);
256 static struct tcp_function_block tcp_def_funcblk = {
260 tcp_default_ctloutput,
271 int t_functions_inited = 0;
272 struct tcp_funchead t_functions;
273 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
276 init_tcp_functions(void)
278 if (t_functions_inited == 0) {
279 TAILQ_INIT(&t_functions);
280 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
281 t_functions_inited = 1;
285 static struct tcp_function_block *
286 find_tcp_functions_locked(struct tcp_function_set *fs)
288 struct tcp_function *f;
289 struct tcp_function_block *blk=NULL;
291 TAILQ_FOREACH(f, &t_functions, tf_next) {
292 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
300 static struct tcp_function_block *
301 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
303 struct tcp_function_block *rblk=NULL;
304 struct tcp_function *f;
306 TAILQ_FOREACH(f, &t_functions, tf_next) {
307 if (f->tf_fb == blk) {
318 struct tcp_function_block *
319 find_and_ref_tcp_functions(struct tcp_function_set *fs)
321 struct tcp_function_block *blk;
323 rw_rlock(&tcp_function_lock);
324 blk = find_tcp_functions_locked(fs);
326 refcount_acquire(&blk->tfb_refcnt);
327 rw_runlock(&tcp_function_lock);
331 struct tcp_function_block *
332 find_and_ref_tcp_fb(struct tcp_function_block *blk)
334 struct tcp_function_block *rblk;
336 rw_rlock(&tcp_function_lock);
337 rblk = find_tcp_fb_locked(blk, NULL);
339 refcount_acquire(&rblk->tfb_refcnt);
340 rw_runlock(&tcp_function_lock);
346 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
349 struct tcp_function_set fs;
350 struct tcp_function_block *blk;
352 memset(&fs, 0, sizeof(fs));
353 rw_rlock(&tcp_function_lock);
354 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
357 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
358 fs.pcbcnt = blk->tfb_refcnt;
360 rw_runlock(&tcp_function_lock);
361 error = sysctl_handle_string(oidp, fs.function_set_name,
362 sizeof(fs.function_set_name), req);
364 /* Check for error or no change */
365 if (error != 0 || req->newptr == NULL)
368 rw_wlock(&tcp_function_lock);
369 blk = find_tcp_functions_locked(&fs);
371 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
375 tcp_func_set_ptr = blk;
377 rw_wunlock(&tcp_function_lock);
381 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
382 CTLTYPE_STRING | CTLFLAG_RW,
383 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
384 "Set/get the default TCP functions");
387 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
389 int error, cnt, linesz;
390 struct tcp_function *f;
395 rw_rlock(&tcp_function_lock);
396 TAILQ_FOREACH(f, &t_functions, tf_next) {
399 rw_runlock(&tcp_function_lock);
401 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
402 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
407 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
412 rw_rlock(&tcp_function_lock);
413 TAILQ_FOREACH(f, &t_functions, tf_next) {
414 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
415 f->tf_fb->tfb_tcp_block_name,
416 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
417 f->tf_fb->tfb_refcnt);
418 if (linesz >= bufsz) {
426 rw_runlock(&tcp_function_lock);
428 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
429 free(buffer, M_TEMP);
433 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
434 CTLTYPE_STRING|CTLFLAG_RD,
435 NULL, 0, sysctl_net_inet_list_available, "A",
436 "list available TCP Function sets");
439 * Target size of TCP PCB hash tables. Must be a power of two.
441 * Note that this can be overridden by the kernel environment
442 * variable net.inet.tcp.tcbhashsize
445 #define TCBHASHSIZE 0
450 * Callouts should be moved into struct tcp directly. They are currently
451 * separate because the tcpcb structure is exported to userland for sysctl
452 * parsing purposes, which do not know about callouts.
463 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
464 #define V_tcpcb_zone VNET(tcpcb_zone)
466 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
467 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
469 static struct mtx isn_mtx;
471 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
472 #define ISN_LOCK() mtx_lock(&isn_mtx)
473 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
476 * TCP initialization.
479 tcp_zone_change(void *tag)
482 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
483 uma_zone_set_max(V_tcpcb_zone, maxsockets);
484 tcp_tw_zone_change();
488 tcp_inpcb_init(void *mem, int size, int flags)
490 struct inpcb *inp = mem;
492 INP_LOCK_INIT(inp, "inp", "tcpinp");
497 * Take a value and get the next power of 2 that doesn't overflow.
498 * Used to size the tcp_inpcb hash buckets.
501 maketcp_hashsize(int size)
507 * get the next power of 2 higher than maxsockets.
509 hashsize = 1 << fls(size);
510 /* catch overflow, and just go one power of 2 smaller */
511 if (hashsize < size) {
512 hashsize = 1 << (fls(size) - 1);
518 register_tcp_functions(struct tcp_function_block *blk, int wait)
520 struct tcp_function_block *lblk;
521 struct tcp_function *n;
522 struct tcp_function_set fs;
524 if (t_functions_inited == 0) {
525 init_tcp_functions();
527 if ((blk->tfb_tcp_output == NULL) ||
528 (blk->tfb_tcp_do_segment == NULL) ||
529 (blk->tfb_tcp_ctloutput == NULL) ||
530 (strlen(blk->tfb_tcp_block_name) == 0)) {
532 * These functions are required and you
537 if (blk->tfb_tcp_timer_stop_all ||
538 blk->tfb_tcp_timer_activate ||
539 blk->tfb_tcp_timer_active ||
540 blk->tfb_tcp_timer_stop) {
542 * If you define one timer function you
543 * must have them all.
545 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
546 (blk->tfb_tcp_timer_activate == NULL) ||
547 (blk->tfb_tcp_timer_active == NULL) ||
548 (blk->tfb_tcp_timer_stop == NULL)) {
552 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
557 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
558 rw_wlock(&tcp_function_lock);
559 lblk = find_tcp_functions_locked(&fs);
561 /* Duplicate name space not allowed */
562 rw_wunlock(&tcp_function_lock);
563 free(n, M_TCPFUNCTIONS);
566 refcount_init(&blk->tfb_refcnt, 0);
568 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
569 rw_wunlock(&tcp_function_lock);
574 deregister_tcp_functions(struct tcp_function_block *blk)
576 struct tcp_function_block *lblk;
577 struct tcp_function *f;
580 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
581 /* You can't un-register the default */
584 rw_wlock(&tcp_function_lock);
585 if (blk == tcp_func_set_ptr) {
586 /* You can't free the current default */
587 rw_wunlock(&tcp_function_lock);
590 if (blk->tfb_refcnt) {
591 /* Still tcb attached, mark it. */
592 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
593 rw_wunlock(&tcp_function_lock);
596 lblk = find_tcp_fb_locked(blk, &f);
599 TAILQ_REMOVE(&t_functions, f, tf_next);
601 free(f, M_TCPFUNCTIONS);
604 rw_wunlock(&tcp_function_lock);
611 const char *tcbhash_tuneable;
614 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
617 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
618 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
619 printf("%s: WARNING: unable to register helper hook\n", __func__);
620 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
621 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
622 printf("%s: WARNING: unable to register helper hook\n", __func__);
624 hashsize = TCBHASHSIZE;
625 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
628 * Auto tune the hash size based on maxsockets.
629 * A perfect hash would have a 1:1 mapping
630 * (hashsize = maxsockets) however it's been
631 * suggested that O(2) average is better.
633 hashsize = maketcp_hashsize(maxsockets / 4);
635 * Our historical default is 512,
636 * do not autotune lower than this.
640 if (bootverbose && IS_DEFAULT_VNET(curvnet))
641 printf("%s: %s auto tuned to %d\n", __func__,
642 tcbhash_tuneable, hashsize);
645 * We require a hashsize to be a power of two.
646 * Previously if it was not a power of two we would just reset it
647 * back to 512, which could be a nasty surprise if you did not notice
649 * Instead what we do is clip it to the closest power of two lower
650 * than the specified hash value.
652 if (!powerof2(hashsize)) {
653 int oldhashsize = hashsize;
655 hashsize = maketcp_hashsize(hashsize);
656 /* prevent absurdly low value */
659 printf("%s: WARNING: TCB hash size not a power of 2, "
660 "clipped from %d to %d.\n", __func__, oldhashsize,
663 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
664 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
667 * These have to be type stable for the benefit of the timers.
669 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
670 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
671 uma_zone_set_max(V_tcpcb_zone, maxsockets);
672 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
678 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
679 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
680 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
682 /* Skip initialization of globals for non-default instances. */
683 if (!IS_DEFAULT_VNET(curvnet))
686 tcp_reass_global_init();
688 /* XXX virtualize those bellow? */
689 tcp_delacktime = TCPTV_DELACK;
690 tcp_keepinit = TCPTV_KEEP_INIT;
691 tcp_keepidle = TCPTV_KEEP_IDLE;
692 tcp_keepintvl = TCPTV_KEEPINTVL;
693 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
695 tcp_rexmit_min = TCPTV_MIN;
696 if (tcp_rexmit_min < 1)
698 tcp_persmin = TCPTV_PERSMIN;
699 tcp_persmax = TCPTV_PERSMAX;
700 tcp_rexmit_slop = TCPTV_CPU_VAR;
701 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
702 tcp_tcbhashsize = hashsize;
703 /* Setup the tcp function block list */
704 init_tcp_functions();
705 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
707 if (tcp_soreceive_stream) {
709 tcp_usrreqs.pru_soreceive = soreceive_stream;
712 tcp6_usrreqs.pru_soreceive = soreceive_stream;
717 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
719 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
721 if (max_protohdr < TCP_MINPROTOHDR)
722 max_protohdr = TCP_MINPROTOHDR;
723 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
725 #undef TCP_MINPROTOHDR
728 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
729 SHUTDOWN_PRI_DEFAULT);
730 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
731 EVENTHANDLER_PRI_ANY);
743 tcp_destroy(void *unused __unused)
751 * All our processes are gone, all our sockets should be cleaned
752 * up, which means, we should be past the tcp_discardcb() calls.
753 * Sleep to let all tcpcb timers really disappear and cleanup.
756 INP_LIST_RLOCK(&V_tcbinfo);
757 n = V_tcbinfo.ipi_count;
758 INP_LIST_RUNLOCK(&V_tcbinfo);
761 pause("tcpdes", hz / 10);
766 in_pcbinfo_destroy(&V_tcbinfo);
767 /* tcp_discardcb() clears the sack_holes up. */
768 uma_zdestroy(V_sack_hole_zone);
769 uma_zdestroy(V_tcpcb_zone);
773 * Cannot free the zone until all tcpcbs are released as we attach
774 * the allocations to them.
776 tcp_fastopen_destroy();
780 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
782 printf("%s: WARNING: unable to deregister helper hook "
783 "type=%d, id=%d: error %d returned\n", __func__,
784 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
786 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
788 printf("%s: WARNING: unable to deregister helper hook "
789 "type=%d, id=%d: error %d returned\n", __func__,
790 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
794 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
804 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
805 * tcp_template used to store this data in mbufs, but we now recopy it out
806 * of the tcpcb each time to conserve mbufs.
809 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
811 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
813 INP_WLOCK_ASSERT(inp);
816 if ((inp->inp_vflag & INP_IPV6) != 0) {
819 ip6 = (struct ip6_hdr *)ip_ptr;
820 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
821 (inp->inp_flow & IPV6_FLOWINFO_MASK);
822 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
823 (IPV6_VERSION & IPV6_VERSION_MASK);
824 ip6->ip6_nxt = IPPROTO_TCP;
825 ip6->ip6_plen = htons(sizeof(struct tcphdr));
826 ip6->ip6_src = inp->in6p_laddr;
827 ip6->ip6_dst = inp->in6p_faddr;
830 #if defined(INET6) && defined(INET)
837 ip = (struct ip *)ip_ptr;
838 ip->ip_v = IPVERSION;
840 ip->ip_tos = inp->inp_ip_tos;
844 ip->ip_ttl = inp->inp_ip_ttl;
846 ip->ip_p = IPPROTO_TCP;
847 ip->ip_src = inp->inp_laddr;
848 ip->ip_dst = inp->inp_faddr;
851 th->th_sport = inp->inp_lport;
852 th->th_dport = inp->inp_fport;
860 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
864 * Create template to be used to send tcp packets on a connection.
865 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
866 * use for this function is in keepalives, which use tcp_respond.
869 tcpip_maketemplate(struct inpcb *inp)
873 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
876 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
881 * Send a single message to the TCP at address specified by
882 * the given TCP/IP header. If m == NULL, then we make a copy
883 * of the tcpiphdr at th and send directly to the addressed host.
884 * This is used to force keep alive messages out using the TCP
885 * template for a connection. If flags are given then we send
886 * a message back to the TCP which originated the segment th,
887 * and discard the mbuf containing it and any other attached mbufs.
889 * In any case the ack and sequence number of the transmitted
890 * segment are as specified by the parameters.
892 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
895 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
896 tcp_seq ack, tcp_seq seq, int flags)
908 int optlen, tlen, win;
911 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
914 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
921 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
922 INP_WLOCK_ASSERT(inp);
929 if (!(flags & TH_RST)) {
930 win = sbspace(&inp->inp_socket->so_rcv);
931 if (win > TCP_MAXWIN << tp->rcv_scale)
932 win = TCP_MAXWIN << tp->rcv_scale;
934 if ((tp->t_flags & TF_NOOPT) == 0)
938 m = m_gethdr(M_NOWAIT, MT_DATA);
941 m->m_data += max_linkhdr;
944 bcopy((caddr_t)ip6, mtod(m, caddr_t),
945 sizeof(struct ip6_hdr));
946 ip6 = mtod(m, struct ip6_hdr *);
947 nth = (struct tcphdr *)(ip6 + 1);
951 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
952 ip = mtod(m, struct ip *);
953 nth = (struct tcphdr *)(ip + 1);
955 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
957 } else if (!M_WRITABLE(m)) {
960 /* Can't reuse 'm', allocate a new mbuf. */
961 n = m_gethdr(M_NOWAIT, MT_DATA);
967 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
973 n->m_data += max_linkhdr;
974 /* m_len is set later */
975 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
978 bcopy((caddr_t)ip6, mtod(n, caddr_t),
979 sizeof(struct ip6_hdr));
980 ip6 = mtod(n, struct ip6_hdr *);
981 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
982 nth = (struct tcphdr *)(ip6 + 1);
986 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
987 ip = mtod(n, struct ip *);
988 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
989 nth = (struct tcphdr *)(ip + 1);
991 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
992 xchg(nth->th_dport, nth->th_sport, uint16_t);
999 * XXX MRT We inherit the FIB, which is lucky.
1003 m->m_data = (caddr_t)ipgen;
1004 /* m_len is set later */
1007 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1008 nth = (struct tcphdr *)(ip6 + 1);
1012 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1013 nth = (struct tcphdr *)(ip + 1);
1017 * this is usually a case when an extension header
1018 * exists between the IPv6 header and the
1021 nth->th_sport = th->th_sport;
1022 nth->th_dport = th->th_dport;
1024 xchg(nth->th_dport, nth->th_sport, uint16_t);
1030 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1032 #if defined(INET) && defined(INET6)
1036 tlen = sizeof (struct tcpiphdr);
1040 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1041 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1042 m, tlen, (long)M_TRAILINGSPACE(m)));
1047 /* Make sure we have room. */
1048 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1049 m->m_next = m_get(M_NOWAIT, MT_DATA);
1051 optp = mtod(m->m_next, u_char *);
1056 optp = (u_char *) (nth + 1);
1062 if (tp->t_flags & TF_RCVD_TSTMP) {
1063 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1064 to.to_tsecr = tp->ts_recent;
1065 to.to_flags |= TOF_TS;
1067 #ifdef TCP_SIGNATURE
1068 /* TCP-MD5 (RFC2385). */
1069 if (tp->t_flags & TF_SIGNATURE)
1070 to.to_flags |= TOF_SIGNATURE;
1073 /* Add the options. */
1074 tlen += optlen = tcp_addoptions(&to, optp);
1076 /* Update m_len in the correct mbuf. */
1077 optm->m_len += optlen;
1083 ip6->ip6_vfc = IPV6_VERSION;
1084 ip6->ip6_nxt = IPPROTO_TCP;
1085 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1088 #if defined(INET) && defined(INET6)
1093 ip->ip_len = htons(tlen);
1094 ip->ip_ttl = V_ip_defttl;
1095 if (V_path_mtu_discovery)
1096 ip->ip_off |= htons(IP_DF);
1099 m->m_pkthdr.len = tlen;
1100 m->m_pkthdr.rcvif = NULL;
1104 * Packet is associated with a socket, so allow the
1105 * label of the response to reflect the socket label.
1107 INP_WLOCK_ASSERT(inp);
1108 mac_inpcb_create_mbuf(inp, m);
1111 * Packet is not associated with a socket, so possibly
1112 * update the label in place.
1114 mac_netinet_tcp_reply(m);
1117 nth->th_seq = htonl(seq);
1118 nth->th_ack = htonl(ack);
1120 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1121 nth->th_flags = flags;
1123 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1125 nth->th_win = htons((u_short)win);
1128 #ifdef TCP_SIGNATURE
1129 if (to.to_flags & TOF_SIGNATURE) {
1130 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1131 IPSEC_DIR_OUTBOUND);
1135 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1138 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1139 nth->th_sum = in6_cksum_pseudo(ip6,
1140 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1141 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1145 #if defined(INET6) && defined(INET)
1150 m->m_pkthdr.csum_flags = CSUM_TCP;
1151 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1152 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1156 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1157 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1159 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1161 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1164 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1167 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1169 #if defined(INET) && defined(INET6)
1173 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1178 * Create a new TCP control block, making an
1179 * empty reassembly queue and hooking it to the argument
1180 * protocol control block. The `inp' parameter must have
1181 * come from the zone allocator set up in tcp_init().
1184 tcp_newtcpcb(struct inpcb *inp)
1186 struct tcpcb_mem *tm;
1189 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1192 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1197 /* Initialise cc_var struct for this tcpcb. */
1199 tp->ccv->type = IPPROTO_TCP;
1200 tp->ccv->ccvc.tcp = tp;
1201 rw_rlock(&tcp_function_lock);
1202 tp->t_fb = tcp_func_set_ptr;
1203 refcount_acquire(&tp->t_fb->tfb_refcnt);
1204 rw_runlock(&tcp_function_lock);
1206 * Use the current system default CC algorithm.
1209 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1210 CC_ALGO(tp) = CC_DEFAULT();
1213 if (CC_ALGO(tp)->cb_init != NULL)
1214 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1215 if (tp->t_fb->tfb_tcp_fb_fini)
1216 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1217 refcount_release(&tp->t_fb->tfb_refcnt);
1218 uma_zfree(V_tcpcb_zone, tm);
1224 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1225 if (tp->t_fb->tfb_tcp_fb_fini)
1226 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1227 refcount_release(&tp->t_fb->tfb_refcnt);
1228 uma_zfree(V_tcpcb_zone, tm);
1234 tp->t_vnet = inp->inp_vnet;
1236 tp->t_timers = &tm->tt;
1237 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1240 isipv6 ? V_tcp_v6mssdflt :
1244 /* Set up our timeouts. */
1245 callout_init(&tp->t_timers->tt_rexmt, 1);
1246 callout_init(&tp->t_timers->tt_persist, 1);
1247 callout_init(&tp->t_timers->tt_keep, 1);
1248 callout_init(&tp->t_timers->tt_2msl, 1);
1249 callout_init(&tp->t_timers->tt_delack, 1);
1251 if (V_tcp_do_rfc1323)
1252 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1254 tp->t_flags |= TF_SACK_PERMIT;
1255 TAILQ_INIT(&tp->snd_holes);
1257 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1260 in_pcbref(inp); /* Reference for tcpcb */
1264 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1265 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1266 * reasonable initial retransmit time.
1268 tp->t_srtt = TCPTV_SRTTBASE;
1269 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1270 tp->t_rttmin = tcp_rexmit_min;
1271 tp->t_rxtcur = TCPTV_RTOBASE;
1272 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1273 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1274 tp->t_rcvtime = ticks;
1276 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1277 * because the socket may be bound to an IPv6 wildcard address,
1278 * which may match an IPv4-mapped IPv6 address.
1280 inp->inp_ip_ttl = V_ip_defttl;
1284 * Init the TCP PCAP queues.
1286 tcp_pcap_tcpcb_init(tp);
1288 if (tp->t_fb->tfb_tcp_fb_init) {
1289 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1291 return (tp); /* XXX */
1295 * Switch the congestion control algorithm back to NewReno for any active
1296 * control blocks using an algorithm which is about to go away.
1297 * This ensures the CC framework can allow the unload to proceed without leaving
1298 * any dangling pointers which would trigger a panic.
1299 * Returning non-zero would inform the CC framework that something went wrong
1300 * and it would be unsafe to allow the unload to proceed. However, there is no
1301 * way for this to occur with this implementation so we always return zero.
1304 tcp_ccalgounload(struct cc_algo *unload_algo)
1306 struct cc_algo *tmpalgo;
1309 VNET_ITERATOR_DECL(vnet_iter);
1312 * Check all active control blocks across all network stacks and change
1313 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1314 * requires cleanup code to be run, call it.
1317 VNET_FOREACH(vnet_iter) {
1318 CURVNET_SET(vnet_iter);
1319 INP_INFO_WLOCK(&V_tcbinfo);
1321 * New connections already part way through being initialised
1322 * with the CC algo we're removing will not race with this code
1323 * because the INP_INFO_WLOCK is held during initialisation. We
1324 * therefore don't enter the loop below until the connection
1325 * list has stabilised.
1327 LIST_FOREACH(inp, &V_tcb, inp_list) {
1329 /* Important to skip tcptw structs. */
1330 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1331 (tp = intotcpcb(inp)) != NULL) {
1333 * By holding INP_WLOCK here, we are assured
1334 * that the connection is not currently
1335 * executing inside the CC module's functions
1336 * i.e. it is safe to make the switch back to
1339 if (CC_ALGO(tp) == unload_algo) {
1340 tmpalgo = CC_ALGO(tp);
1341 /* NewReno does not require any init. */
1342 CC_ALGO(tp) = &newreno_cc_algo;
1343 if (tmpalgo->cb_destroy != NULL)
1344 tmpalgo->cb_destroy(tp->ccv);
1349 INP_INFO_WUNLOCK(&V_tcbinfo);
1352 VNET_LIST_RUNLOCK();
1358 * Drop a TCP connection, reporting
1359 * the specified error. If connection is synchronized,
1360 * then send a RST to peer.
1363 tcp_drop(struct tcpcb *tp, int errno)
1365 struct socket *so = tp->t_inpcb->inp_socket;
1367 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1368 INP_WLOCK_ASSERT(tp->t_inpcb);
1370 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1371 tcp_state_change(tp, TCPS_CLOSED);
1372 (void) tp->t_fb->tfb_tcp_output(tp);
1373 TCPSTAT_INC(tcps_drops);
1375 TCPSTAT_INC(tcps_conndrops);
1376 if (errno == ETIMEDOUT && tp->t_softerror)
1377 errno = tp->t_softerror;
1378 so->so_error = errno;
1379 return (tcp_close(tp));
1383 tcp_discardcb(struct tcpcb *tp)
1385 struct inpcb *inp = tp->t_inpcb;
1386 struct socket *so = inp->inp_socket;
1388 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1392 INP_WLOCK_ASSERT(inp);
1395 * Make sure that all of our timers are stopped before we delete the
1398 * If stopping a timer fails, we schedule a discard function in same
1399 * callout, and the last discard function called will take care of
1400 * deleting the tcpcb.
1402 tp->t_timers->tt_draincnt = 0;
1403 tcp_timer_stop(tp, TT_REXMT);
1404 tcp_timer_stop(tp, TT_PERSIST);
1405 tcp_timer_stop(tp, TT_KEEP);
1406 tcp_timer_stop(tp, TT_2MSL);
1407 tcp_timer_stop(tp, TT_DELACK);
1408 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1410 * Call the stop-all function of the methods,
1411 * this function should call the tcp_timer_stop()
1412 * method with each of the function specific timeouts.
1413 * That stop will be called via the tfb_tcp_timer_stop()
1414 * which should use the async drain function of the
1415 * callout system (see tcp_var.h).
1417 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1421 * If we got enough samples through the srtt filter,
1422 * save the rtt and rttvar in the routing entry.
1423 * 'Enough' is arbitrarily defined as 4 rtt samples.
1424 * 4 samples is enough for the srtt filter to converge
1425 * to within enough % of the correct value; fewer samples
1426 * and we could save a bogus rtt. The danger is not high
1427 * as tcp quickly recovers from everything.
1428 * XXX: Works very well but needs some more statistics!
1430 if (tp->t_rttupdated >= 4) {
1431 struct hc_metrics_lite metrics;
1434 bzero(&metrics, sizeof(metrics));
1436 * Update the ssthresh always when the conditions below
1437 * are satisfied. This gives us better new start value
1438 * for the congestion avoidance for new connections.
1439 * ssthresh is only set if packet loss occurred on a session.
1441 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1442 * being torn down. Ideally this code would not use 'so'.
1444 ssthresh = tp->snd_ssthresh;
1445 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1447 * convert the limit from user data bytes to
1448 * packets then to packet data bytes.
1450 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1453 ssthresh *= (tp->t_maxseg +
1455 (isipv6 ? sizeof (struct ip6_hdr) +
1456 sizeof (struct tcphdr) :
1458 sizeof (struct tcpiphdr)
1465 metrics.rmx_ssthresh = ssthresh;
1467 metrics.rmx_rtt = tp->t_srtt;
1468 metrics.rmx_rttvar = tp->t_rttvar;
1469 metrics.rmx_cwnd = tp->snd_cwnd;
1470 metrics.rmx_sendpipe = 0;
1471 metrics.rmx_recvpipe = 0;
1473 tcp_hc_update(&inp->inp_inc, &metrics);
1476 /* free the reassembly queue, if any */
1477 tcp_reass_flush(tp);
1480 /* Disconnect offload device, if any. */
1481 if (tp->t_flags & TF_TOE)
1482 tcp_offload_detach(tp);
1485 tcp_free_sackholes(tp);
1488 /* Free the TCP PCAP queues. */
1489 tcp_pcap_drain(&(tp->t_inpkts));
1490 tcp_pcap_drain(&(tp->t_outpkts));
1493 /* Allow the CC algorithm to clean up after itself. */
1494 if (CC_ALGO(tp)->cb_destroy != NULL)
1495 CC_ALGO(tp)->cb_destroy(tp->ccv);
1498 khelp_destroy_osd(tp->osd);
1502 inp->inp_ppcb = NULL;
1503 if (tp->t_timers->tt_draincnt == 0) {
1504 /* We own the last reference on tcpcb, let's free it. */
1505 if (tp->t_fb->tfb_tcp_fb_fini)
1506 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1507 refcount_release(&tp->t_fb->tfb_refcnt);
1509 uma_zfree(V_tcpcb_zone, tp);
1510 released = in_pcbrele_wlocked(inp);
1511 KASSERT(!released, ("%s: inp %p should not have been released "
1512 "here", __func__, inp));
1517 tcp_timer_discard(void *ptp)
1522 tp = (struct tcpcb *)ptp;
1523 CURVNET_SET(tp->t_vnet);
1524 INP_INFO_RLOCK(&V_tcbinfo);
1526 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1529 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1530 ("%s: tcpcb has to be stopped here", __func__));
1531 tp->t_timers->tt_draincnt--;
1532 if (tp->t_timers->tt_draincnt == 0) {
1533 /* We own the last reference on this tcpcb, let's free it. */
1534 if (tp->t_fb->tfb_tcp_fb_fini)
1535 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1536 refcount_release(&tp->t_fb->tfb_refcnt);
1538 uma_zfree(V_tcpcb_zone, tp);
1539 if (in_pcbrele_wlocked(inp)) {
1540 INP_INFO_RUNLOCK(&V_tcbinfo);
1546 INP_INFO_RUNLOCK(&V_tcbinfo);
1551 * Attempt to close a TCP control block, marking it as dropped, and freeing
1552 * the socket if we hold the only reference.
1555 tcp_close(struct tcpcb *tp)
1557 struct inpcb *inp = tp->t_inpcb;
1560 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1561 INP_WLOCK_ASSERT(inp);
1564 if (tp->t_state == TCPS_LISTEN)
1565 tcp_offload_listen_stop(tp);
1569 * This releases the TFO pending counter resource for TFO listen
1570 * sockets as well as passively-created TFO sockets that transition
1571 * from SYN_RECEIVED to CLOSED.
1573 if (tp->t_tfo_pending) {
1574 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1575 tp->t_tfo_pending = NULL;
1579 TCPSTAT_INC(tcps_closed);
1580 TCPSTATES_DEC(tp->t_state);
1581 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1582 so = inp->inp_socket;
1583 soisdisconnected(so);
1584 if (inp->inp_flags & INP_SOCKREF) {
1585 KASSERT(so->so_state & SS_PROTOREF,
1586 ("tcp_close: !SS_PROTOREF"));
1587 inp->inp_flags &= ~INP_SOCKREF;
1591 so->so_state &= ~SS_PROTOREF;
1601 VNET_ITERATOR_DECL(vnet_iter);
1606 VNET_LIST_RLOCK_NOSLEEP();
1607 VNET_FOREACH(vnet_iter) {
1608 CURVNET_SET(vnet_iter);
1613 * Walk the tcpbs, if existing, and flush the reassembly queue,
1614 * if there is one...
1615 * XXX: The "Net/3" implementation doesn't imply that the TCP
1616 * reassembly queue should be flushed, but in a situation
1617 * where we're really low on mbufs, this is potentially
1620 INP_INFO_WLOCK(&V_tcbinfo);
1621 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1622 if (inpb->inp_flags & INP_TIMEWAIT)
1625 if ((tcpb = intotcpcb(inpb)) != NULL) {
1626 tcp_reass_flush(tcpb);
1627 tcp_clean_sackreport(tcpb);
1629 if (tcp_pcap_aggressive_free) {
1630 /* Free the TCP PCAP queues. */
1631 tcp_pcap_drain(&(tcpb->t_inpkts));
1632 tcp_pcap_drain(&(tcpb->t_outpkts));
1638 INP_INFO_WUNLOCK(&V_tcbinfo);
1641 VNET_LIST_RUNLOCK_NOSLEEP();
1645 * Notify a tcp user of an asynchronous error;
1646 * store error as soft error, but wake up user
1647 * (for now, won't do anything until can select for soft error).
1649 * Do not wake up user since there currently is no mechanism for
1650 * reporting soft errors (yet - a kqueue filter may be added).
1652 static struct inpcb *
1653 tcp_notify(struct inpcb *inp, int error)
1657 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1658 INP_WLOCK_ASSERT(inp);
1660 if ((inp->inp_flags & INP_TIMEWAIT) ||
1661 (inp->inp_flags & INP_DROPPED))
1664 tp = intotcpcb(inp);
1665 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1668 * Ignore some errors if we are hooked up.
1669 * If connection hasn't completed, has retransmitted several times,
1670 * and receives a second error, give up now. This is better
1671 * than waiting a long time to establish a connection that
1672 * can never complete.
1674 if (tp->t_state == TCPS_ESTABLISHED &&
1675 (error == EHOSTUNREACH || error == ENETUNREACH ||
1676 error == EHOSTDOWN)) {
1677 if (inp->inp_route.ro_rt) {
1678 RTFREE(inp->inp_route.ro_rt);
1679 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1682 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1684 tp = tcp_drop(tp, error);
1690 tp->t_softerror = error;
1694 wakeup( &so->so_timeo);
1701 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1703 int error, i, m, n, pcb_count;
1704 struct inpcb *inp, **inp_list;
1709 * The process of preparing the TCB list is too time-consuming and
1710 * resource-intensive to repeat twice on every request.
1712 if (req->oldptr == NULL) {
1713 n = V_tcbinfo.ipi_count +
1714 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1715 n += imax(n / 8, 10);
1716 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1720 if (req->newptr != NULL)
1724 * OK, now we're committed to doing something.
1726 INP_LIST_RLOCK(&V_tcbinfo);
1727 gencnt = V_tcbinfo.ipi_gencnt;
1728 n = V_tcbinfo.ipi_count;
1729 INP_LIST_RUNLOCK(&V_tcbinfo);
1731 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1733 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1734 + (n + m) * sizeof(struct xtcpcb));
1738 xig.xig_len = sizeof xig;
1739 xig.xig_count = n + m;
1740 xig.xig_gen = gencnt;
1741 xig.xig_sogen = so_gencnt;
1742 error = SYSCTL_OUT(req, &xig, sizeof xig);
1746 error = syncache_pcblist(req, m, &pcb_count);
1750 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1752 INP_INFO_WLOCK(&V_tcbinfo);
1753 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1754 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1756 if (inp->inp_gencnt <= gencnt) {
1758 * XXX: This use of cr_cansee(), introduced with
1759 * TCP state changes, is not quite right, but for
1760 * now, better than nothing.
1762 if (inp->inp_flags & INP_TIMEWAIT) {
1763 if (intotw(inp) != NULL)
1764 error = cr_cansee(req->td->td_ucred,
1765 intotw(inp)->tw_cred);
1767 error = EINVAL; /* Skip this inp. */
1769 error = cr_canseeinpcb(req->td->td_ucred, inp);
1772 inp_list[i++] = inp;
1777 INP_INFO_WUNLOCK(&V_tcbinfo);
1781 for (i = 0; i < n; i++) {
1784 if (inp->inp_gencnt <= gencnt) {
1788 bzero(&xt, sizeof(xt));
1789 xt.xt_len = sizeof xt;
1790 /* XXX should avoid extra copy */
1791 bcopy(inp, &xt.xt_inp, sizeof *inp);
1792 inp_ppcb = inp->inp_ppcb;
1793 if (inp_ppcb == NULL)
1794 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1795 else if (inp->inp_flags & INP_TIMEWAIT) {
1796 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1797 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1799 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1800 if (xt.xt_tp.t_timers)
1801 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1803 if (inp->inp_socket != NULL)
1804 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1806 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1807 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1809 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1811 error = SYSCTL_OUT(req, &xt, sizeof xt);
1815 INP_INFO_RLOCK(&V_tcbinfo);
1816 for (i = 0; i < n; i++) {
1819 if (!in_pcbrele_rlocked(inp))
1822 INP_INFO_RUNLOCK(&V_tcbinfo);
1826 * Give the user an updated idea of our state.
1827 * If the generation differs from what we told
1828 * her before, she knows that something happened
1829 * while we were processing this request, and it
1830 * might be necessary to retry.
1832 INP_LIST_RLOCK(&V_tcbinfo);
1833 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1834 xig.xig_sogen = so_gencnt;
1835 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1836 INP_LIST_RUNLOCK(&V_tcbinfo);
1837 error = SYSCTL_OUT(req, &xig, sizeof xig);
1839 free(inp_list, M_TEMP);
1843 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1844 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1845 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1849 tcp_getcred(SYSCTL_HANDLER_ARGS)
1852 struct sockaddr_in addrs[2];
1856 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1859 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1862 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1863 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1865 if (inp->inp_socket == NULL)
1868 error = cr_canseeinpcb(req->td->td_ucred, inp);
1870 cru2x(inp->inp_cred, &xuc);
1875 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1879 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1880 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1881 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1886 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1889 struct sockaddr_in6 addrs[2];
1896 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1899 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1902 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1903 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1906 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1908 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1917 inp = in_pcblookup(&V_tcbinfo,
1918 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1920 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1921 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1924 inp = in6_pcblookup(&V_tcbinfo,
1925 &addrs[1].sin6_addr, addrs[1].sin6_port,
1926 &addrs[0].sin6_addr, addrs[0].sin6_port,
1927 INPLOOKUP_RLOCKPCB, NULL);
1929 if (inp->inp_socket == NULL)
1932 error = cr_canseeinpcb(req->td->td_ucred, inp);
1934 cru2x(inp->inp_cred, &xuc);
1939 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1943 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1944 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1945 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1951 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1953 struct ip *ip = vip;
1955 struct in_addr faddr;
1958 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1960 struct in_conninfo inc;
1961 tcp_seq icmp_tcp_seq;
1964 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1965 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1968 if (cmd == PRC_MSGSIZE)
1969 notify = tcp_mtudisc_notify;
1970 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1971 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1972 cmd == PRC_TIMXCEED_INTRANS) && ip)
1973 notify = tcp_drop_syn_sent;
1976 * Hostdead is ugly because it goes linearly through all PCBs.
1977 * XXX: We never get this from ICMP, otherwise it makes an
1978 * excellent DoS attack on machines with many connections.
1980 else if (cmd == PRC_HOSTDEAD)
1982 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1986 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1990 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1991 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1992 INP_INFO_RLOCK(&V_tcbinfo);
1993 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1994 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1995 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1996 /* signal EHOSTDOWN, as it flushes the cached route */
1997 inp = (*notify)(inp, EHOSTDOWN);
2000 } else if (inp != NULL) {
2001 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2002 !(inp->inp_flags & INP_DROPPED) &&
2003 !(inp->inp_socket == NULL)) {
2004 icmp_tcp_seq = ntohl(th->th_seq);
2005 tp = intotcpcb(inp);
2006 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2007 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2008 if (cmd == PRC_MSGSIZE) {
2011 * If we got a needfrag set the MTU
2012 * in the route to the suggested new
2013 * value (if given) and then notify.
2015 mtu = ntohs(icp->icmp_nextmtu);
2017 * If no alternative MTU was
2018 * proposed, try the next smaller
2023 ntohs(ip->ip_len), 1);
2024 if (mtu < V_tcp_minmss +
2025 sizeof(struct tcpiphdr))
2026 mtu = V_tcp_minmss +
2027 sizeof(struct tcpiphdr);
2029 * Only process the offered MTU if it
2030 * is smaller than the current one.
2032 if (mtu < tp->t_maxseg +
2033 sizeof(struct tcpiphdr)) {
2034 bzero(&inc, sizeof(inc));
2035 inc.inc_faddr = faddr;
2037 inp->inp_inc.inc_fibnum;
2038 tcp_hc_updatemtu(&inc, mtu);
2039 tcp_mtudisc(inp, mtu);
2042 inp = (*notify)(inp,
2043 inetctlerrmap[cmd]);
2049 bzero(&inc, sizeof(inc));
2050 inc.inc_fport = th->th_dport;
2051 inc.inc_lport = th->th_sport;
2052 inc.inc_faddr = faddr;
2053 inc.inc_laddr = ip->ip_src;
2054 syncache_unreach(&inc, th);
2056 INP_INFO_RUNLOCK(&V_tcbinfo);
2062 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2064 struct in6_addr *dst;
2066 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2067 struct ip6_hdr *ip6;
2071 struct icmp6_hdr *icmp6;
2072 struct ip6ctlparam *ip6cp = NULL;
2073 const struct sockaddr_in6 *sa6_src = NULL;
2074 struct in_conninfo inc;
2075 tcp_seq icmp_tcp_seq;
2080 if (sa->sa_family != AF_INET6 ||
2081 sa->sa_len != sizeof(struct sockaddr_in6))
2084 /* if the parameter is from icmp6, decode it. */
2086 ip6cp = (struct ip6ctlparam *)d;
2087 icmp6 = ip6cp->ip6c_icmp6;
2089 ip6 = ip6cp->ip6c_ip6;
2090 off = ip6cp->ip6c_off;
2091 sa6_src = ip6cp->ip6c_src;
2092 dst = ip6cp->ip6c_finaldst;
2096 off = 0; /* fool gcc */
2101 if (cmd == PRC_MSGSIZE)
2102 notify = tcp_mtudisc_notify;
2103 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2104 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2105 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2106 notify = tcp_drop_syn_sent;
2109 * Hostdead is ugly because it goes linearly through all PCBs.
2110 * XXX: We never get this from ICMP, otherwise it makes an
2111 * excellent DoS attack on machines with many connections.
2113 else if (cmd == PRC_HOSTDEAD)
2115 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2119 in6_pcbnotify(&V_tcbinfo, sa, 0,
2120 (const struct sockaddr *)sa6_src,
2121 0, cmd, NULL, notify);
2125 /* Check if we can safely get the ports from the tcp hdr */
2128 (int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2132 th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2133 INP_INFO_RLOCK(&V_tcbinfo);
2134 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2135 &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2136 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2137 /* signal EHOSTDOWN, as it flushes the cached route */
2138 inp = (*notify)(inp, EHOSTDOWN);
2141 } else if (inp != NULL) {
2142 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2143 !(inp->inp_flags & INP_DROPPED) &&
2144 !(inp->inp_socket == NULL)) {
2145 icmp_tcp_seq = ntohl(th->th_seq);
2146 tp = intotcpcb(inp);
2147 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2148 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2149 if (cmd == PRC_MSGSIZE) {
2152 * If we got a needfrag set the MTU
2153 * in the route to the suggested new
2154 * value (if given) and then notify.
2156 mtu = ntohl(icmp6->icmp6_mtu);
2158 * If no alternative MTU was
2159 * proposed, or the proposed
2160 * MTU was too small, set to
2163 if (mtu < IPV6_MMTU)
2164 mtu = IPV6_MMTU - 8;
2167 bzero(&inc, sizeof(inc));
2168 inc.inc_fibnum = M_GETFIB(m);
2169 inc.inc_flags |= INC_ISIPV6;
2170 inc.inc6_faddr = *dst;
2171 if (in6_setscope(&inc.inc6_faddr,
2172 m->m_pkthdr.rcvif, NULL))
2176 * Only process the offered MTU if it
2177 * is smaller than the current one.
2179 if (mtu < tp->t_maxseg +
2180 (sizeof (*th) + sizeof (*ip6))) {
2181 tcp_hc_updatemtu(&inc, mtu);
2182 tcp_mtudisc(inp, mtu);
2183 ICMP6STAT_INC(icp6s_pmtuchg);
2186 inp = (*notify)(inp,
2187 inet6ctlerrmap[cmd]);
2194 bzero(&inc, sizeof(inc));
2195 inc.inc_fibnum = M_GETFIB(m);
2196 inc.inc_flags |= INC_ISIPV6;
2197 inc.inc_fport = th->th_dport;
2198 inc.inc_lport = th->th_sport;
2199 inc.inc6_faddr = *dst;
2200 inc.inc6_laddr = ip6->ip6_src;
2201 syncache_unreach(&inc, th);
2203 INP_INFO_RUNLOCK(&V_tcbinfo);
2209 * Following is where TCP initial sequence number generation occurs.
2211 * There are two places where we must use initial sequence numbers:
2212 * 1. In SYN-ACK packets.
2213 * 2. In SYN packets.
2215 * All ISNs for SYN-ACK packets are generated by the syncache. See
2216 * tcp_syncache.c for details.
2218 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2219 * depends on this property. In addition, these ISNs should be
2220 * unguessable so as to prevent connection hijacking. To satisfy
2221 * the requirements of this situation, the algorithm outlined in
2222 * RFC 1948 is used, with only small modifications.
2224 * Implementation details:
2226 * Time is based off the system timer, and is corrected so that it
2227 * increases by one megabyte per second. This allows for proper
2228 * recycling on high speed LANs while still leaving over an hour
2231 * As reading the *exact* system time is too expensive to be done
2232 * whenever setting up a TCP connection, we increment the time
2233 * offset in two ways. First, a small random positive increment
2234 * is added to isn_offset for each connection that is set up.
2235 * Second, the function tcp_isn_tick fires once per clock tick
2236 * and increments isn_offset as necessary so that sequence numbers
2237 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2238 * random positive increments serve only to ensure that the same
2239 * exact sequence number is never sent out twice (as could otherwise
2240 * happen when a port is recycled in less than the system tick
2243 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2244 * between seeding of isn_secret. This is normally set to zero,
2245 * as reseeding should not be necessary.
2247 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2248 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2249 * general, this means holding an exclusive (write) lock.
2252 #define ISN_BYTES_PER_SECOND 1048576
2253 #define ISN_STATIC_INCREMENT 4096
2254 #define ISN_RANDOM_INCREMENT (4096 - 1)
2256 static VNET_DEFINE(u_char, isn_secret[32]);
2257 static VNET_DEFINE(int, isn_last);
2258 static VNET_DEFINE(int, isn_last_reseed);
2259 static VNET_DEFINE(u_int32_t, isn_offset);
2260 static VNET_DEFINE(u_int32_t, isn_offset_old);
2262 #define V_isn_secret VNET(isn_secret)
2263 #define V_isn_last VNET(isn_last)
2264 #define V_isn_last_reseed VNET(isn_last_reseed)
2265 #define V_isn_offset VNET(isn_offset)
2266 #define V_isn_offset_old VNET(isn_offset_old)
2269 tcp_new_isn(struct tcpcb *tp)
2272 u_int32_t md5_buffer[4];
2274 u_int32_t projected_offset;
2276 INP_WLOCK_ASSERT(tp->t_inpcb);
2279 /* Seed if this is the first use, reseed if requested. */
2280 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2281 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2283 read_random(&V_isn_secret, sizeof(V_isn_secret));
2284 V_isn_last_reseed = ticks;
2287 /* Compute the md5 hash and return the ISN. */
2289 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2290 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2292 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2293 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2294 sizeof(struct in6_addr));
2295 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2296 sizeof(struct in6_addr));
2300 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2301 sizeof(struct in_addr));
2302 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2303 sizeof(struct in_addr));
2305 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2306 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2307 new_isn = (tcp_seq) md5_buffer[0];
2308 V_isn_offset += ISN_STATIC_INCREMENT +
2309 (arc4random() & ISN_RANDOM_INCREMENT);
2310 if (ticks != V_isn_last) {
2311 projected_offset = V_isn_offset_old +
2312 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2313 if (SEQ_GT(projected_offset, V_isn_offset))
2314 V_isn_offset = projected_offset;
2315 V_isn_offset_old = V_isn_offset;
2318 new_isn += V_isn_offset;
2324 * When a specific ICMP unreachable message is received and the
2325 * connection state is SYN-SENT, drop the connection. This behavior
2326 * is controlled by the icmp_may_rst sysctl.
2329 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2333 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2334 INP_WLOCK_ASSERT(inp);
2336 if ((inp->inp_flags & INP_TIMEWAIT) ||
2337 (inp->inp_flags & INP_DROPPED))
2340 tp = intotcpcb(inp);
2341 if (tp->t_state != TCPS_SYN_SENT)
2344 tp = tcp_drop(tp, errno);
2352 * When `need fragmentation' ICMP is received, update our idea of the MSS
2353 * based on the new value. Also nudge TCP to send something, since we
2354 * know the packet we just sent was dropped.
2355 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2357 static struct inpcb *
2358 tcp_mtudisc_notify(struct inpcb *inp, int error)
2361 tcp_mtudisc(inp, -1);
2366 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2371 INP_WLOCK_ASSERT(inp);
2372 if ((inp->inp_flags & INP_TIMEWAIT) ||
2373 (inp->inp_flags & INP_DROPPED))
2376 tp = intotcpcb(inp);
2377 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2379 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2381 so = inp->inp_socket;
2382 SOCKBUF_LOCK(&so->so_snd);
2383 /* If the mss is larger than the socket buffer, decrease the mss. */
2384 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2385 tp->t_maxseg = so->so_snd.sb_hiwat;
2386 SOCKBUF_UNLOCK(&so->so_snd);
2388 TCPSTAT_INC(tcps_mturesent);
2390 tp->snd_nxt = tp->snd_una;
2391 tcp_free_sackholes(tp);
2392 tp->snd_recover = tp->snd_max;
2393 if (tp->t_flags & TF_SACK_PERMIT)
2394 EXIT_FASTRECOVERY(tp->t_flags);
2395 tp->t_fb->tfb_tcp_output(tp);
2400 * Look-up the routing entry to the peer of this inpcb. If no route
2401 * is found and it cannot be allocated, then return 0. This routine
2402 * is called by TCP routines that access the rmx structure and by
2403 * tcp_mss_update to get the peer/interface MTU.
2406 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2408 struct nhop4_extended nh4;
2410 uint32_t maxmtu = 0;
2412 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2414 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2416 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2417 NHR_REF, 0, &nh4) != 0)
2421 maxmtu = nh4.nh_mtu;
2423 /* Report additional interface capabilities. */
2425 if (ifp->if_capenable & IFCAP_TSO4 &&
2426 ifp->if_hwassist & CSUM_TSO) {
2427 cap->ifcap |= CSUM_TSO;
2428 cap->tsomax = ifp->if_hw_tsomax;
2429 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2430 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2433 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2441 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2443 struct nhop6_extended nh6;
2444 struct in6_addr dst6;
2447 uint32_t maxmtu = 0;
2449 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2451 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2452 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2453 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2458 maxmtu = nh6.nh_mtu;
2460 /* Report additional interface capabilities. */
2462 if (ifp->if_capenable & IFCAP_TSO6 &&
2463 ifp->if_hwassist & CSUM_TSO) {
2464 cap->ifcap |= CSUM_TSO;
2465 cap->tsomax = ifp->if_hw_tsomax;
2466 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2467 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2470 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2478 * Calculate effective SMSS per RFC5681 definition for a given TCP
2479 * connection at its current state, taking into account SACK and etc.
2482 tcp_maxseg(const struct tcpcb *tp)
2486 if (tp->t_flags & TF_NOOPT)
2487 return (tp->t_maxseg);
2490 * Here we have a simplified code from tcp_addoptions(),
2491 * without a proper loop, and having most of paddings hardcoded.
2492 * We might make mistakes with padding here in some edge cases,
2493 * but this is harmless, since result of tcp_maxseg() is used
2494 * only in cwnd and ssthresh estimations.
2496 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2497 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2498 if (tp->t_flags & TF_RCVD_TSTMP)
2499 optlen = TCPOLEN_TSTAMP_APPA;
2502 #ifdef TCP_SIGNATURE
2503 if (tp->t_flags & TF_SIGNATURE)
2504 optlen += PAD(TCPOLEN_SIGNATURE);
2506 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2507 optlen += TCPOLEN_SACKHDR;
2508 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2509 optlen = PAD(optlen);
2512 if (tp->t_flags & TF_REQ_TSTMP)
2513 optlen = TCPOLEN_TSTAMP_APPA;
2515 optlen = PAD(TCPOLEN_MAXSEG);
2516 if (tp->t_flags & TF_REQ_SCALE)
2517 optlen += PAD(TCPOLEN_WINDOW);
2518 #ifdef TCP_SIGNATURE
2519 if (tp->t_flags & TF_SIGNATURE)
2520 optlen += PAD(TCPOLEN_SIGNATURE);
2522 if (tp->t_flags & TF_SACK_PERMIT)
2523 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2526 optlen = min(optlen, TCP_MAXOLEN);
2527 return (tp->t_maxseg - optlen);
2531 /* compute ESP/AH header size for TCP, including outer IP header. */
2533 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2540 struct ip6_hdr *ip6;
2544 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2545 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2547 m = m_gethdr(M_NOWAIT, MT_DATA);
2552 if ((inp->inp_vflag & INP_IPV6) != 0) {
2553 ip6 = mtod(m, struct ip6_hdr *);
2554 th = (struct tcphdr *)(ip6 + 1);
2555 m->m_pkthdr.len = m->m_len =
2556 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2557 tcpip_fillheaders(inp, ip6, th);
2558 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2562 ip = mtod(m, struct ip *);
2563 th = (struct tcphdr *)(ip + 1);
2564 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2565 tcpip_fillheaders(inp, ip, th);
2566 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2574 #ifdef TCP_SIGNATURE
2576 * Callback function invoked by m_apply() to digest TCP segment data
2577 * contained within an mbuf chain.
2580 tcp_signature_apply(void *fstate, void *data, u_int len)
2583 MD5Update(fstate, (u_char *)data, len);
2588 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2589 * search with the destination IP address, and a 'magic SPI' to be
2590 * determined by the application. This is hardcoded elsewhere to 1179
2593 tcp_get_sav(struct mbuf *m, u_int direction)
2595 union sockaddr_union dst;
2596 struct secasvar *sav;
2599 struct ip6_hdr *ip6;
2600 char ip6buf[INET6_ADDRSTRLEN];
2603 /* Extract the destination from the IP header in the mbuf. */
2604 bzero(&dst, sizeof(union sockaddr_union));
2605 ip = mtod(m, struct ip *);
2607 ip6 = NULL; /* Make the compiler happy. */
2612 dst.sa.sa_len = sizeof(struct sockaddr_in);
2613 dst.sa.sa_family = AF_INET;
2614 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2615 ip->ip_src : ip->ip_dst;
2619 case (IPV6_VERSION >> 4):
2620 ip6 = mtod(m, struct ip6_hdr *);
2621 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2622 dst.sa.sa_family = AF_INET6;
2623 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2624 ip6->ip6_src : ip6->ip6_dst;
2633 /* Look up an SADB entry which matches the address of the peer. */
2634 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2636 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2637 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2639 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2640 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2649 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2652 * m pointer to head of mbuf chain
2653 * len length of TCP segment data, excluding options
2654 * optlen length of TCP segment options
2655 * buf pointer to storage for computed MD5 digest
2656 * sav pointer to security assosiation
2658 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2659 * When called from tcp_input(), we can be sure that th_sum has been
2660 * zeroed out and verified already.
2662 * Releases reference to SADB key before return.
2664 * Return 0 if successful, otherwise return -1.
2668 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2669 u_char *buf, struct secasvar *sav)
2672 struct ippseudo ippseudo;
2678 struct ipovly *ipovly;
2682 struct ip6_hdr *ip6;
2683 struct in6_addr in6;
2689 KASSERT(m != NULL, ("NULL mbuf chain"));
2690 KASSERT(buf != NULL, ("NULL signature pointer"));
2692 /* Extract the destination from the IP header in the mbuf. */
2693 ip = mtod(m, struct ip *);
2695 ip6 = NULL; /* Make the compiler happy. */
2700 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2702 * XXX The ippseudo header MUST be digested in network byte order,
2703 * or else we'll fail the regression test. Assume all fields we've
2704 * been doing arithmetic on have been in host byte order.
2705 * XXX One cannot depend on ipovly->ih_len here. When called from
2706 * tcp_output(), the underlying ip_len member has not yet been set.
2711 ipovly = (struct ipovly *)ip;
2712 ippseudo.ippseudo_src = ipovly->ih_src;
2713 ippseudo.ippseudo_dst = ipovly->ih_dst;
2714 ippseudo.ippseudo_pad = 0;
2715 ippseudo.ippseudo_p = IPPROTO_TCP;
2716 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2718 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2720 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2721 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2726 * RFC 2385, 2.0 Proposal
2727 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2728 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2729 * extended next header value (to form 32 bits), and 32-bit segment
2731 * Note: Upper-Layer Packet Length comes before Next Header.
2733 case (IPV6_VERSION >> 4):
2734 ip6 = mtod(m, struct ip6_hdr *);
2736 in6_clearscope(&in6);
2737 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2739 in6_clearscope(&in6);
2740 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2741 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2742 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2744 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2745 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2746 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2748 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2750 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2751 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2763 * Step 2: Update MD5 hash with TCP header, excluding options.
2764 * The TCP checksum must be set to zero.
2766 savecsum = th->th_sum;
2768 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2769 th->th_sum = savecsum;
2772 * Step 3: Update MD5 hash with TCP segment data.
2773 * Use m_apply() to avoid an early m_pullup().
2776 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2779 * Step 4: Update MD5 hash with shared secret.
2781 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2782 MD5Final(buf, &ctx);
2784 key_sa_recordxfer(sav, m);
2790 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2792 * Return 0 if successful, otherwise return -1.
2795 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2796 u_char *buf, u_int direction)
2798 struct secasvar *sav;
2800 if ((sav = tcp_get_sav(m, direction)) == NULL)
2803 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2807 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2810 * m pointer to head of mbuf chain
2811 * len length of TCP segment data, excluding options
2812 * optlen length of TCP segment options
2813 * buf pointer to storage for computed MD5 digest
2814 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2816 * Return 1 if successful, otherwise return 0.
2819 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2820 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2822 char tmpdigest[TCP_SIGLEN];
2824 if (tcp_sig_checksigs == 0)
2826 if ((tcpbflag & TF_SIGNATURE) == 0) {
2827 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2830 * If this socket is not expecting signature but
2831 * the segment contains signature just fail.
2833 TCPSTAT_INC(tcps_sig_err_sigopt);
2834 TCPSTAT_INC(tcps_sig_rcvbadsig);
2838 /* Signature is not expected, and not present in segment. */
2843 * If this socket is expecting signature but the segment does not
2844 * contain any just fail.
2846 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2847 TCPSTAT_INC(tcps_sig_err_nosigopt);
2848 TCPSTAT_INC(tcps_sig_rcvbadsig);
2851 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2852 IPSEC_DIR_INBOUND) == -1) {
2853 TCPSTAT_INC(tcps_sig_err_buildsig);
2854 TCPSTAT_INC(tcps_sig_rcvbadsig);
2858 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2859 TCPSTAT_INC(tcps_sig_rcvbadsig);
2862 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2865 #endif /* TCP_SIGNATURE */
2868 sysctl_drop(SYSCTL_HANDLER_ARGS)
2870 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2871 struct sockaddr_storage addrs[2];
2875 struct sockaddr_in *fin, *lin;
2877 struct sockaddr_in6 *fin6, *lin6;
2888 if (req->oldptr != NULL || req->oldlen != 0)
2890 if (req->newptr == NULL)
2892 if (req->newlen < sizeof(addrs))
2894 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2898 switch (addrs[0].ss_family) {
2901 fin6 = (struct sockaddr_in6 *)&addrs[0];
2902 lin6 = (struct sockaddr_in6 *)&addrs[1];
2903 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2904 lin6->sin6_len != sizeof(struct sockaddr_in6))
2906 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2907 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2909 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2910 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2911 fin = (struct sockaddr_in *)&addrs[0];
2912 lin = (struct sockaddr_in *)&addrs[1];
2915 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2918 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2925 fin = (struct sockaddr_in *)&addrs[0];
2926 lin = (struct sockaddr_in *)&addrs[1];
2927 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2928 lin->sin_len != sizeof(struct sockaddr_in))
2935 INP_INFO_RLOCK(&V_tcbinfo);
2936 switch (addrs[0].ss_family) {
2939 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2940 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2941 INPLOOKUP_WLOCKPCB, NULL);
2946 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2947 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2952 if (inp->inp_flags & INP_TIMEWAIT) {
2954 * XXXRW: There currently exists a state where an
2955 * inpcb is present, but its timewait state has been
2956 * discarded. For now, don't allow dropping of this
2964 } else if (!(inp->inp_flags & INP_DROPPED) &&
2965 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2966 tp = intotcpcb(inp);
2967 tp = tcp_drop(tp, ECONNABORTED);
2974 INP_INFO_RUNLOCK(&V_tcbinfo);
2978 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2979 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2980 0, sysctl_drop, "", "Drop TCP connection");
2983 * Generate a standardized TCP log line for use throughout the
2984 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2985 * allow use in the interrupt context.
2987 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2988 * NB: The function may return NULL if memory allocation failed.
2990 * Due to header inclusion and ordering limitations the struct ip
2991 * and ip6_hdr pointers have to be passed as void pointers.
2994 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2998 /* Is logging enabled? */
2999 if (tcp_log_in_vain == 0)
3002 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3006 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3010 /* Is logging enabled? */
3011 if (tcp_log_debug == 0)
3014 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3018 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3025 const struct ip6_hdr *ip6;
3027 ip6 = (const struct ip6_hdr *)ip6hdr;
3029 ip = (struct ip *)ip4hdr;
3032 * The log line looks like this:
3033 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3035 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3036 sizeof(PRINT_TH_FLAGS) + 1 +
3038 2 * INET6_ADDRSTRLEN;
3040 2 * INET_ADDRSTRLEN;
3043 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3047 strcat(s, "TCP: [");
3050 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3051 inet_ntoa_r(inc->inc_faddr, sp);
3053 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3055 inet_ntoa_r(inc->inc_laddr, sp);
3057 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3060 ip6_sprintf(sp, &inc->inc6_faddr);
3062 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3064 ip6_sprintf(sp, &inc->inc6_laddr);
3066 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3067 } else if (ip6 && th) {
3068 ip6_sprintf(sp, &ip6->ip6_src);
3070 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3072 ip6_sprintf(sp, &ip6->ip6_dst);
3074 sprintf(sp, "]:%i", ntohs(th->th_dport));
3077 } else if (ip && th) {
3078 inet_ntoa_r(ip->ip_src, sp);
3080 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3082 inet_ntoa_r(ip->ip_dst, sp);
3084 sprintf(sp, "]:%i", ntohs(th->th_dport));
3092 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3093 if (*(s + size - 1) != '\0')
3094 panic("%s: string too long", __func__);
3099 * A subroutine which makes it easy to track TCP state changes with DTrace.
3100 * This function shouldn't be called for t_state initializations that don't
3101 * correspond to actual TCP state transitions.
3104 tcp_state_change(struct tcpcb *tp, int newstate)
3106 #if defined(KDTRACE_HOOKS)
3107 int pstate = tp->t_state;
3110 TCPSTATES_DEC(tp->t_state);
3111 TCPSTATES_INC(newstate);
3112 tp->t_state = newstate;
3113 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);