2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
48 #include <sys/hhook.h>
50 #include <sys/kernel.h>
52 #include <sys/khelp.h>
57 #include <sys/qmath.h>
58 #include <sys/stats.h>
59 #include <sys/sysctl.h>
61 #include <sys/malloc.h>
62 #include <sys/refcount.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/protosw.h>
70 #include <sys/random.h>
74 #include <net/route.h>
75 #include <net/route/nhop.h>
77 #include <net/if_var.h>
78 #include <net/if_private.h>
81 #include <netinet/in.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/in_kdtrace.h>
84 #include <netinet/in_pcb.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/in_var.h>
87 #include <netinet/ip.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/ip_var.h>
91 #include <netinet/icmp6.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/in6_fib.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet6/nd6.h>
100 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_ecn.h>
109 #include <netinet/tcp_log_buf.h>
110 #include <netinet/tcp_syncache.h>
111 #include <netinet/tcp_hpts.h>
112 #include <netinet/cc/cc.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/tcp_fastopen.h>
116 #include <netinet/tcp_pcap.h>
119 #include <netinet/tcp_offload.h>
121 #include <netinet/udp.h>
122 #include <netinet/udp_var.h>
124 #include <netinet6/tcp6_var.h>
127 #include <netipsec/ipsec_support.h>
129 #include <machine/in_cksum.h>
130 #include <crypto/siphash/siphash.h>
132 #include <security/mac/mac_framework.h>
135 static ip6proto_ctlinput_t tcp6_ctlinput;
136 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
144 #ifdef NETFLIX_EXP_DETECTION
145 /* Sack attack detection thresholds and such */
146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
147 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
148 "Sack Attack detection thresholds");
149 int32_t tcp_force_detection = 0;
150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
152 &tcp_force_detection, 0,
153 "Do we force detection even if the INP has it off?");
154 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
157 &tcp_sack_to_ack_thresh, 700,
158 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
159 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
162 &tcp_sack_to_move_thresh, 600,
163 "Percentage of sack moves we must see above (10.1 percent is 101)");
164 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
167 &tcp_restoral_thresh, 550,
168 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
169 int32_t tcp_sad_decay_val = 800;
170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
172 &tcp_sad_decay_val, 800,
173 "The decay percentage (10.1 percent equals 101 )");
174 int32_t tcp_map_minimum = 500;
175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
177 &tcp_map_minimum, 500,
178 "Number of Map enteries before we start detection");
179 int32_t tcp_attack_on_turns_on_logging = 0;
180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
182 &tcp_attack_on_turns_on_logging, 0,
183 "When we have a positive hit on attack, do we turn on logging?");
184 int32_t tcp_sad_pacing_interval = 2000;
185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
187 &tcp_sad_pacing_interval, 2000,
188 "What is the minimum pacing interval for a classified attacker?");
190 int32_t tcp_sad_low_pps = 100;
191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
193 &tcp_sad_low_pps, 100,
194 "What is the input pps that below which we do not decay?");
196 uint32_t tcp_ack_war_time_window = 1000;
197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
199 &tcp_ack_war_time_window, 1000,
200 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
201 uint32_t tcp_ack_war_cnt = 5;
202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
205 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
207 struct rwlock tcp_function_lock;
210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
215 error = sysctl_handle_int(oidp, &new, 0, req);
216 if (error == 0 && req->newptr) {
217 if (new < TCP_MINMSS)
225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
226 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
227 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
228 "Default TCP Maximum Segment Size");
232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
236 new = V_tcp_v6mssdflt;
237 error = sysctl_handle_int(oidp, &new, 0, req);
238 if (error == 0 && req->newptr) {
239 if (new < TCP_MINMSS)
242 V_tcp_v6mssdflt = new;
247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
248 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
250 "Default TCP Maximum Segment Size for IPv6");
254 * Minimum MSS we accept and use. This prevents DoS attacks where
255 * we are forced to a ridiculous low MSS like 20 and send hundreds
256 * of packets instead of one. The effect scales with the available
257 * bandwidth and quickly saturates the CPU and network interface
258 * with packet generation and sending. Set to zero to disable MINMSS
259 * checking. This setting prevents us from sending too small packets.
261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
263 &VNET_NAME(tcp_minmss), 0,
264 "Minimum TCP Maximum Segment Size");
266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
268 &VNET_NAME(tcp_do_rfc1323), 0,
269 "Enable rfc1323 (high performance TCP) extensions");
272 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
273 * Some stacks negotiate TS, but never send them after connection setup. Some
274 * stacks negotiate TS, but don't send them when sending keep-alive segments.
275 * These include modern widely deployed TCP stacks.
276 * Therefore tolerating violations for now...
278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
280 &VNET_NAME(tcp_tolerate_missing_ts), 0,
281 "Tolerate missing TCP timestamps");
283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
285 &VNET_NAME(tcp_ts_offset_per_conn), 0,
286 "Initialize TCP timestamps per connection instead of per host pair");
288 /* How many connections are pacing */
289 static volatile uint32_t number_of_tcp_connections_pacing = 0;
290 static uint32_t shadow_num_connections = 0;
292 static int tcp_pacing_limit = 10000;
293 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
294 &tcp_pacing_limit, 1000,
295 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
297 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
298 &shadow_num_connections, 0, "Number of TCP connections being paced");
300 static int tcp_log_debug = 0;
301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
302 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
304 static int tcp_tcbhashsize;
305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
306 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
308 static int do_tcpdrain = 1;
309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
310 "Enable tcp_drain routine for extra help when low on mbufs");
312 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
313 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
315 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
316 #define V_icmp_may_rst VNET(icmp_may_rst)
317 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
318 &VNET_NAME(icmp_may_rst), 0,
319 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
321 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
322 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
324 &VNET_NAME(tcp_isn_reseed_interval), 0,
325 "Seconds between reseeding of ISN secret");
327 static int tcp_soreceive_stream;
328 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
329 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
331 VNET_DEFINE(uma_zone_t, sack_hole_zone);
332 #define V_sack_hole_zone VNET(sack_hole_zone)
333 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
335 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
340 new = V_tcp_map_entries_limit;
341 error = sysctl_handle_int(oidp, &new, 0, req);
342 if (error == 0 && req->newptr) {
343 /* only allow "0" and value > minimum */
344 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
347 V_tcp_map_entries_limit = new;
351 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
352 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
353 &VNET_NAME(tcp_map_entries_limit), 0,
354 &sysctl_net_inet_tcp_map_limit_check, "IU",
355 "Total sendmap entries limit");
357 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
358 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
359 &VNET_NAME(tcp_map_split_limit), 0,
360 "Total sendmap split entries limit");
363 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
366 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
367 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
368 #define V_ts_offset_secret VNET(ts_offset_secret)
370 static int tcp_default_fb_init(struct tcpcb *tp);
371 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
372 static int tcp_default_handoff_ok(struct tcpcb *tp);
373 static struct inpcb *tcp_notify(struct inpcb *, int);
374 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
375 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
376 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
377 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
378 const void *ip4hdr, const void *ip6hdr);
379 static ipproto_ctlinput_t tcp_ctlinput;
380 static udp_tun_icmp_t tcp_ctlinput_viaudp;
382 static struct tcp_function_block tcp_def_funcblk = {
383 .tfb_tcp_block_name = "freebsd",
384 .tfb_tcp_output = tcp_default_output,
385 .tfb_tcp_do_segment = tcp_do_segment,
386 .tfb_tcp_ctloutput = tcp_default_ctloutput,
387 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
388 .tfb_tcp_fb_init = tcp_default_fb_init,
389 .tfb_tcp_fb_fini = tcp_default_fb_fini,
392 static int tcp_fb_cnt = 0;
393 struct tcp_funchead t_functions;
394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
399 TCPSTAT_INC(tcps_dsack_count);
402 if (SEQ_GT(end, start)) {
403 tp->t_dsack_bytes += (end - start);
404 TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
406 tp->t_dsack_tlp_bytes += (start - end);
407 TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
410 if (SEQ_GT(end, start)) {
411 tp->t_dsack_bytes += (end - start);
412 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
414 tp->t_dsack_tlp_bytes += (start - end);
415 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
420 static struct tcp_function_block *
421 find_tcp_functions_locked(struct tcp_function_set *fs)
423 struct tcp_function *f;
424 struct tcp_function_block *blk=NULL;
426 TAILQ_FOREACH(f, &t_functions, tf_next) {
427 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
435 static struct tcp_function_block *
436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
438 struct tcp_function_block *rblk=NULL;
439 struct tcp_function *f;
441 TAILQ_FOREACH(f, &t_functions, tf_next) {
442 if (f->tf_fb == blk) {
453 struct tcp_function_block *
454 find_and_ref_tcp_functions(struct tcp_function_set *fs)
456 struct tcp_function_block *blk;
458 rw_rlock(&tcp_function_lock);
459 blk = find_tcp_functions_locked(fs);
461 refcount_acquire(&blk->tfb_refcnt);
462 rw_runlock(&tcp_function_lock);
466 struct tcp_function_block *
467 find_and_ref_tcp_fb(struct tcp_function_block *blk)
469 struct tcp_function_block *rblk;
471 rw_rlock(&tcp_function_lock);
472 rblk = find_tcp_fb_locked(blk, NULL);
474 refcount_acquire(&rblk->tfb_refcnt);
475 rw_runlock(&tcp_function_lock);
479 /* Find a matching alias for the given tcp_function_block. */
481 find_tcp_function_alias(struct tcp_function_block *blk,
482 struct tcp_function_set *fs)
484 struct tcp_function *f;
488 rw_rlock(&tcp_function_lock);
489 TAILQ_FOREACH(f, &t_functions, tf_next) {
490 if ((f->tf_fb == blk) &&
491 (strncmp(f->tf_name, blk->tfb_tcp_block_name,
492 TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
493 /* Matching function block with different name. */
494 strncpy(fs->function_set_name, f->tf_name,
495 TCP_FUNCTION_NAME_LEN_MAX);
500 /* Null terminate the string appropriately. */
502 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
504 fs->function_set_name[0] = '\0';
506 rw_runlock(&tcp_function_lock);
510 static struct tcp_function_block *
511 find_and_ref_tcp_default_fb(void)
513 struct tcp_function_block *rblk;
515 rw_rlock(&tcp_function_lock);
516 rblk = tcp_func_set_ptr;
517 refcount_acquire(&rblk->tfb_refcnt);
518 rw_runlock(&tcp_function_lock);
523 tcp_switch_back_to_default(struct tcpcb *tp)
525 struct tcp_function_block *tfb;
527 KASSERT(tp->t_fb != &tcp_def_funcblk,
528 ("%s: called by the built-in default stack", __func__));
531 * Release the old stack. This function will either find a new one
534 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
535 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
536 refcount_release(&tp->t_fb->tfb_refcnt);
539 * Now, we'll find a new function block to use.
540 * Start by trying the current user-selected
541 * default, unless this stack is the user-selected
544 tfb = find_and_ref_tcp_default_fb();
545 if (tfb == tp->t_fb) {
546 refcount_release(&tfb->tfb_refcnt);
549 /* Does the stack accept this connection? */
550 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 refcount_release(&tfb->tfb_refcnt);
555 /* Try to use that stack. */
557 /* Initialize the new stack. If it succeeds, we are done. */
559 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
560 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
564 * Initialization failed. Release the reference count on
567 refcount_release(&tfb->tfb_refcnt);
571 * If that wasn't feasible, use the built-in default
572 * stack which is not allowed to reject anyone.
574 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
576 /* there always should be a default */
577 panic("Can't refer to tcp_def_funcblk");
579 if (tfb->tfb_tcp_handoff_ok != NULL) {
580 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
581 /* The default stack cannot say no */
582 panic("Default stack rejects a new session?");
586 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
587 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
588 /* The default stack cannot fail */
589 panic("Default stack initialization failed");
594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
595 const struct sockaddr *sa, void *ctx)
606 TCPSTAT_INC(tcps_tunneled_pkts);
607 if ((m->m_flags & M_PKTHDR) == 0) {
608 /* Can't handle one that is not a pkt hdr */
609 TCPSTAT_INC(tcps_tunneled_errs);
612 thlen = sizeof(struct tcphdr);
613 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
614 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
615 TCPSTAT_INC(tcps_tunneled_errs);
618 iph = mtod(m, struct ip *);
619 uh = (struct udphdr *)((caddr_t)iph + off);
620 th = (struct tcphdr *)(uh + 1);
621 thlen = th->th_off << 2;
622 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
623 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
625 TCPSTAT_INC(tcps_tunneled_errs);
628 iph = mtod(m, struct ip *);
629 uh = (struct udphdr *)((caddr_t)iph + off);
630 th = (struct tcphdr *)(uh + 1);
633 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
634 bcopy(th, uh, m->m_len - off);
635 m->m_len -= sizeof(struct udphdr);
636 m->m_pkthdr.len -= sizeof(struct udphdr);
638 * We use the same algorithm for
639 * both UDP and TCP for c-sum. So
640 * the code in tcp_input will skip
641 * the checksum. So we do nothing
642 * with the flag (m->m_pkthdr.csum_flags).
647 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
648 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
652 case IPV6_VERSION >> 4:
653 ip6 = mtod(m, struct ip6_hdr *);
654 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
655 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
670 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
673 struct tcp_function_set fs;
674 struct tcp_function_block *blk;
676 memset(&fs, 0, sizeof(fs));
677 rw_rlock(&tcp_function_lock);
678 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
681 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
682 fs.pcbcnt = blk->tfb_refcnt;
684 rw_runlock(&tcp_function_lock);
685 error = sysctl_handle_string(oidp, fs.function_set_name,
686 sizeof(fs.function_set_name), req);
688 /* Check for error or no change */
689 if (error != 0 || req->newptr == NULL)
692 rw_wlock(&tcp_function_lock);
693 blk = find_tcp_functions_locked(&fs);
695 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
699 tcp_func_set_ptr = blk;
701 rw_wunlock(&tcp_function_lock);
705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
706 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
707 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
708 "Set/get the default TCP functions");
711 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
713 int error, cnt, linesz;
714 struct tcp_function *f;
720 rw_rlock(&tcp_function_lock);
721 TAILQ_FOREACH(f, &t_functions, tf_next) {
724 rw_runlock(&tcp_function_lock);
726 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
727 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
732 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
733 "Alias", "PCB count");
738 rw_rlock(&tcp_function_lock);
739 TAILQ_FOREACH(f, &t_functions, tf_next) {
740 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
741 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
742 f->tf_fb->tfb_tcp_block_name,
743 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
744 alias ? f->tf_name : "-",
745 f->tf_fb->tfb_refcnt);
746 if (linesz >= bufsz) {
754 rw_runlock(&tcp_function_lock);
756 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
757 free(buffer, M_TEMP);
761 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
762 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
763 NULL, 0, sysctl_net_inet_list_available, "A",
764 "list available TCP Function sets");
766 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
769 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
770 #define V_udp4_tun_socket VNET(udp4_tun_socket)
773 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
774 #define V_udp6_tun_socket VNET(udp6_tun_socket)
777 static struct sx tcpoudp_lock;
780 tcp_over_udp_stop(void)
783 sx_assert(&tcpoudp_lock, SA_XLOCKED);
786 if (V_udp4_tun_socket != NULL) {
787 soclose(V_udp4_tun_socket);
788 V_udp4_tun_socket = NULL;
792 if (V_udp6_tun_socket != NULL) {
793 soclose(V_udp6_tun_socket);
794 V_udp6_tun_socket = NULL;
800 tcp_over_udp_start(void)
805 struct sockaddr_in sin;
808 struct sockaddr_in6 sin6;
811 sx_assert(&tcpoudp_lock, SA_XLOCKED);
813 port = V_tcp_udp_tunneling_port;
814 if (ntohs(port) == 0) {
815 /* Must have a port set */
819 if (V_udp4_tun_socket != NULL) {
820 /* Already running -- must stop first */
825 if (V_udp6_tun_socket != NULL) {
826 /* Already running -- must stop first */
831 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
832 SOCK_DGRAM, IPPROTO_UDP,
833 curthread->td_ucred, curthread))) {
837 /* Call the special UDP hook. */
838 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
839 tcp_recv_udp_tunneled_packet,
845 /* Ok, we have a socket, bind it to the port. */
846 memset(&sin, 0, sizeof(struct sockaddr_in));
847 sin.sin_len = sizeof(struct sockaddr_in);
848 sin.sin_family = AF_INET;
849 sin.sin_port = htons(port);
850 if ((ret = sobind(V_udp4_tun_socket,
851 (struct sockaddr *)&sin, curthread))) {
857 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
858 SOCK_DGRAM, IPPROTO_UDP,
859 curthread->td_ucred, curthread))) {
863 /* Call the special UDP hook. */
864 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
865 tcp_recv_udp_tunneled_packet,
866 tcp6_ctlinput_viaudp,
871 /* Ok, we have a socket, bind it to the port. */
872 memset(&sin6, 0, sizeof(struct sockaddr_in6));
873 sin6.sin6_len = sizeof(struct sockaddr_in6);
874 sin6.sin6_family = AF_INET6;
875 sin6.sin6_port = htons(port);
876 if ((ret = sobind(V_udp6_tun_socket,
877 (struct sockaddr *)&sin6, curthread))) {
886 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
891 old = V_tcp_udp_tunneling_port;
893 error = sysctl_handle_int(oidp, &new, 0, req);
895 (req->newptr != NULL)) {
896 if ((new < TCP_TUNNELING_PORT_MIN) ||
897 (new > TCP_TUNNELING_PORT_MAX)) {
900 sx_xlock(&tcpoudp_lock);
901 V_tcp_udp_tunneling_port = new;
906 error = tcp_over_udp_start();
908 V_tcp_udp_tunneling_port = 0;
911 sx_xunlock(&tcpoudp_lock);
917 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
918 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
919 &VNET_NAME(tcp_udp_tunneling_port),
920 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
921 "Tunneling port for tcp over udp");
923 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
926 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
930 new = V_tcp_udp_tunneling_overhead;
931 error = sysctl_handle_int(oidp, &new, 0, req);
932 if (error == 0 && req->newptr) {
933 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
934 (new > TCP_TUNNELING_OVERHEAD_MAX))
937 V_tcp_udp_tunneling_overhead = new;
942 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
943 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
944 &VNET_NAME(tcp_udp_tunneling_overhead),
945 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
946 "MSS reduction when using tcp over udp");
949 * Exports one (struct tcp_function_info) for each alias/name.
952 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
955 struct tcp_function *f;
956 struct tcp_function_info tfi;
959 * We don't allow writes.
961 if (req->newptr != NULL)
965 * Wire the old buffer so we can directly copy the functions to
966 * user space without dropping the lock.
968 if (req->oldptr != NULL) {
969 error = sysctl_wire_old_buffer(req, 0);
975 * Walk the list and copy out matching entries. If INVARIANTS
976 * is compiled in, also walk the list to verify the length of
977 * the list matches what we have recorded.
979 rw_rlock(&tcp_function_lock);
983 if (req->oldptr == NULL) {
988 TAILQ_FOREACH(f, &t_functions, tf_next) {
992 if (req->oldptr != NULL) {
993 bzero(&tfi, sizeof(tfi));
994 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
995 tfi.tfi_id = f->tf_fb->tfb_id;
996 (void)strlcpy(tfi.tfi_alias, f->tf_name,
997 sizeof(tfi.tfi_alias));
998 (void)strlcpy(tfi.tfi_name,
999 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
1000 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
1002 * Don't stop on error, as that is the
1003 * mechanism we use to accumulate length
1004 * information if the buffer was too short.
1008 KASSERT(cnt == tcp_fb_cnt,
1009 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1013 rw_runlock(&tcp_function_lock);
1014 if (req->oldptr == NULL)
1015 error = SYSCTL_OUT(req, NULL,
1016 (cnt + 1) * sizeof(struct tcp_function_info));
1021 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1022 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1023 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1024 "List TCP function block name-to-ID mappings");
1027 * tfb_tcp_handoff_ok() function for the default stack.
1028 * Note that we'll basically try to take all comers.
1031 tcp_default_handoff_ok(struct tcpcb *tp)
1038 * tfb_tcp_fb_init() function for the default stack.
1040 * This handles making sure we have appropriate timers set if you are
1041 * transitioning a socket that has some amount of setup done.
1043 * The init() fuction from the default can *never* return non-zero i.e.
1044 * it is required to always succeed since it is the stack of last resort!
1047 tcp_default_fb_init(struct tcpcb *tp)
1049 struct socket *so = tptosocket(tp);
1051 INP_WLOCK_ASSERT(tptoinpcb(tp));
1053 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1054 ("%s: connection %p in unexpected state %d", __func__, tp,
1058 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1059 * know what to do for unexpected states (which includes TIME_WAIT).
1061 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1065 * Make sure some kind of transmission timer is set if there is
1068 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1069 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1070 tcp_timer_active(tp, TT_PERSIST))) {
1072 * If the session has established and it looks like it should
1073 * be in the persist state, set the persist timer. Otherwise,
1074 * set the retransmit timer.
1076 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1077 (int32_t)(tp->snd_nxt - tp->snd_una) <
1078 (int32_t)sbavail(&so->so_snd))
1081 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1084 /* All non-embryonic sessions get a keepalive timer. */
1085 if (!tcp_timer_active(tp, TT_KEEP))
1086 tcp_timer_activate(tp, TT_KEEP,
1087 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1091 * Make sure critical variables are initialized
1092 * if transitioning while in Recovery.
1094 if IN_FASTRECOVERY(tp->t_flags) {
1095 if (tp->sackhint.recover_fs == 0)
1096 tp->sackhint.recover_fs = max(1,
1097 tp->snd_nxt - tp->snd_una);
1104 * tfb_tcp_fb_fini() function for the default stack.
1106 * This changes state as necessary (or prudent) to prepare for another stack
1107 * to assume responsibility for the connection.
1110 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1113 INP_WLOCK_ASSERT(tptoinpcb(tp));
1117 * Target size of TCP PCB hash tables. Must be a power of two.
1119 * Note that this can be overridden by the kernel environment
1120 * variable net.inet.tcp.tcbhashsize
1123 #define TCBHASHSIZE 0
1126 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1127 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1129 static struct mtx isn_mtx;
1131 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1132 #define ISN_LOCK() mtx_lock(&isn_mtx)
1133 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1135 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1138 * Take a value and get the next power of 2 that doesn't overflow.
1139 * Used to size the tcp_inpcb hash buckets.
1142 maketcp_hashsize(int size)
1148 * get the next power of 2 higher than maxsockets.
1150 hashsize = 1 << fls(size);
1151 /* catch overflow, and just go one power of 2 smaller */
1152 if (hashsize < size) {
1153 hashsize = 1 << (fls(size) - 1);
1158 static volatile int next_tcp_stack_id = 1;
1161 * Register a TCP function block with the name provided in the names
1162 * array. (Note that this function does NOT automatically register
1163 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1164 * explicitly include blk->tfb_tcp_block_name in the list of names if
1165 * you wish to register the stack with that name.)
1167 * Either all name registrations will succeed or all will fail. If
1168 * a name registration fails, the function will update the num_names
1169 * argument to point to the array index of the name that encountered
1172 * Returns 0 on success, or an error code on failure.
1175 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1176 const char *names[], int *num_names)
1178 struct tcp_function *n;
1179 struct tcp_function_set fs;
1182 KASSERT(names != NULL && *num_names > 0,
1183 ("%s: Called with 0-length name list", __func__));
1184 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1185 KASSERT(rw_initialized(&tcp_function_lock),
1186 ("%s: called too early", __func__));
1188 if ((blk->tfb_tcp_output == NULL) ||
1189 (blk->tfb_tcp_do_segment == NULL) ||
1190 (blk->tfb_tcp_ctloutput == NULL) ||
1191 (strlen(blk->tfb_tcp_block_name) == 0)) {
1193 * These functions are required and you
1200 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1205 refcount_init(&blk->tfb_refcnt, 0);
1206 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1207 for (i = 0; i < *num_names; i++) {
1208 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1215 (void)strlcpy(fs.function_set_name, names[i],
1216 sizeof(fs.function_set_name));
1217 rw_wlock(&tcp_function_lock);
1218 if (find_tcp_functions_locked(&fs) != NULL) {
1219 /* Duplicate name space not allowed */
1220 rw_wunlock(&tcp_function_lock);
1221 free(n, M_TCPFUNCTIONS);
1225 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1226 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1228 rw_wunlock(&tcp_function_lock);
1234 * Deregister the names we just added. Because registration failed
1235 * for names[i], we don't need to deregister that name.
1238 rw_wlock(&tcp_function_lock);
1240 TAILQ_FOREACH(n, &t_functions, tf_next) {
1241 if (!strncmp(n->tf_name, names[i],
1242 TCP_FUNCTION_NAME_LEN_MAX)) {
1243 TAILQ_REMOVE(&t_functions, n, tf_next);
1246 free(n, M_TCPFUNCTIONS);
1251 rw_wunlock(&tcp_function_lock);
1256 * Register a TCP function block using the name provided in the name
1259 * Returns 0 on success, or an error code on failure.
1262 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1265 const char *name_list[1];
1270 name_list[0] = name;
1272 name_list[0] = blk->tfb_tcp_block_name;
1273 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1278 * Register a TCP function block using the name defined in
1279 * blk->tfb_tcp_block_name.
1281 * Returns 0 on success, or an error code on failure.
1284 register_tcp_functions(struct tcp_function_block *blk, int wait)
1287 return (register_tcp_functions_as_name(blk, NULL, wait));
1291 * Deregister all names associated with a function block. This
1292 * functionally removes the function block from use within the system.
1294 * When called with a true quiesce argument, mark the function block
1295 * as being removed so no more stacks will use it and determine
1296 * whether the removal would succeed.
1298 * When called with a false quiesce argument, actually attempt the
1301 * When called with a force argument, attempt to switch all TCBs to
1302 * use the default stack instead of returning EBUSY.
1304 * Returns 0 on success (or if the removal would succeed, or an error
1308 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1311 struct tcp_function *f;
1313 if (blk == &tcp_def_funcblk) {
1314 /* You can't un-register the default */
1317 rw_wlock(&tcp_function_lock);
1318 if (blk == tcp_func_set_ptr) {
1319 /* You can't free the current default */
1320 rw_wunlock(&tcp_function_lock);
1323 /* Mark the block so no more stacks can use it. */
1324 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1326 * If TCBs are still attached to the stack, attempt to switch them
1327 * to the default stack.
1329 if (force && blk->tfb_refcnt) {
1330 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1331 INPLOOKUP_WLOCKPCB);
1334 VNET_ITERATOR_DECL(vnet_iter);
1336 rw_wunlock(&tcp_function_lock);
1339 VNET_FOREACH(vnet_iter) {
1340 CURVNET_SET(vnet_iter);
1341 while ((inp = inp_next(&inpi)) != NULL) {
1342 tp = intotcpcb(inp);
1343 if (tp == NULL || tp->t_fb != blk)
1345 tcp_switch_back_to_default(tp);
1349 VNET_LIST_RUNLOCK();
1351 rw_wlock(&tcp_function_lock);
1353 if (blk->tfb_refcnt) {
1354 /* TCBs still attached. */
1355 rw_wunlock(&tcp_function_lock);
1360 rw_wunlock(&tcp_function_lock);
1363 /* Remove any function names that map to this function block. */
1364 while (find_tcp_fb_locked(blk, &f) != NULL) {
1365 TAILQ_REMOVE(&t_functions, f, tf_next);
1368 free(f, M_TCPFUNCTIONS);
1370 rw_wunlock(&tcp_function_lock);
1377 struct epoch_tracker et;
1378 VNET_ITERATOR_DECL(vnet_iter);
1383 NET_EPOCH_ENTER(et);
1384 VNET_LIST_RLOCK_NOSLEEP();
1385 VNET_FOREACH(vnet_iter) {
1386 CURVNET_SET(vnet_iter);
1387 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1388 INPLOOKUP_WLOCKPCB);
1393 * Walk the tcpbs, if existing, and flush the reassembly queue,
1394 * if there is one...
1395 * XXX: The "Net/3" implementation doesn't imply that the TCP
1396 * reassembly queue should be flushed, but in a situation
1397 * where we're really low on mbufs, this is potentially
1400 while ((inpb = inp_next(&inpi)) != NULL) {
1401 if ((tcpb = intotcpcb(inpb)) != NULL) {
1402 tcp_reass_flush(tcpb);
1403 tcp_clean_sackreport(tcpb);
1405 tcp_log_drain(tcpb);
1408 if (tcp_pcap_aggressive_free) {
1409 /* Free the TCP PCAP queues. */
1410 tcp_pcap_drain(&(tcpb->t_inpkts));
1411 tcp_pcap_drain(&(tcpb->t_outpkts));
1418 VNET_LIST_RUNLOCK_NOSLEEP();
1423 tcp_vnet_init(void *arg __unused)
1427 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1428 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1429 printf("%s: WARNING: unable to register helper hook\n", __func__);
1430 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1431 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1432 printf("%s: WARNING: unable to register helper hook\n", __func__);
1435 if (tcp_stats_init())
1436 printf("%s: WARNING: unable to initialise TCP stats\n",
1439 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1445 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1446 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1447 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1449 tcp_fastopen_init();
1451 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1452 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1454 V_tcp_msl = TCPTV_MSL;
1456 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1457 tcp_vnet_init, NULL);
1460 tcp_init(void *arg __unused)
1462 const char *tcbhash_tuneable;
1465 tcp_reass_global_init();
1467 /* XXX virtualize those below? */
1468 tcp_delacktime = TCPTV_DELACK;
1469 tcp_keepinit = TCPTV_KEEP_INIT;
1470 tcp_keepidle = TCPTV_KEEP_IDLE;
1471 tcp_keepintvl = TCPTV_KEEPINTVL;
1472 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1473 tcp_rexmit_initial = TCPTV_RTOBASE;
1474 if (tcp_rexmit_initial < 1)
1475 tcp_rexmit_initial = 1;
1476 tcp_rexmit_min = TCPTV_MIN;
1477 if (tcp_rexmit_min < 1)
1479 tcp_persmin = TCPTV_PERSMIN;
1480 tcp_persmax = TCPTV_PERSMAX;
1481 tcp_rexmit_slop = TCPTV_CPU_VAR;
1482 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1484 /* Setup the tcp function block list */
1485 TAILQ_INIT(&t_functions);
1486 rw_init(&tcp_function_lock, "tcp_func_lock");
1487 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1488 sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1490 /* Initialize the TCP logging data. */
1493 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1495 if (tcp_soreceive_stream) {
1497 tcp_protosw.pr_soreceive = soreceive_stream;
1500 tcp6_protosw.pr_soreceive = soreceive_stream;
1505 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1507 max_protohdr_grow(sizeof(struct tcpiphdr));
1511 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1512 SHUTDOWN_PRI_DEFAULT);
1513 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1514 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1516 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1517 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1518 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1519 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1520 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1521 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1522 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1523 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1524 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1529 hashsize = TCBHASHSIZE;
1530 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1531 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1532 if (hashsize == 0) {
1534 * Auto tune the hash size based on maxsockets.
1535 * A perfect hash would have a 1:1 mapping
1536 * (hashsize = maxsockets) however it's been
1537 * suggested that O(2) average is better.
1539 hashsize = maketcp_hashsize(maxsockets / 4);
1541 * Our historical default is 512,
1542 * do not autotune lower than this.
1547 printf("%s: %s auto tuned to %d\n", __func__,
1548 tcbhash_tuneable, hashsize);
1551 * We require a hashsize to be a power of two.
1552 * Previously if it was not a power of two we would just reset it
1553 * back to 512, which could be a nasty surprise if you did not notice
1554 * the error message.
1555 * Instead what we do is clip it to the closest power of two lower
1556 * than the specified hash value.
1558 if (!powerof2(hashsize)) {
1559 int oldhashsize = hashsize;
1561 hashsize = maketcp_hashsize(hashsize);
1562 /* prevent absurdly low value */
1565 printf("%s: WARNING: TCB hash size not a power of 2, "
1566 "clipped from %d to %d.\n", __func__, oldhashsize,
1569 tcp_tcbhashsize = hashsize;
1572 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1575 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1578 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1582 tcp_destroy(void *unused __unused)
1590 * All our processes are gone, all our sockets should be cleaned
1591 * up, which means, we should be past the tcp_discardcb() calls.
1592 * Sleep to let all tcpcb timers really disappear and cleanup.
1595 INP_INFO_WLOCK(&V_tcbinfo);
1596 n = V_tcbinfo.ipi_count;
1597 INP_INFO_WUNLOCK(&V_tcbinfo);
1600 pause("tcpdes", hz / 10);
1604 in_pcbinfo_destroy(&V_tcbinfo);
1605 /* tcp_discardcb() clears the sack_holes up. */
1606 uma_zdestroy(V_sack_hole_zone);
1609 * Cannot free the zone until all tcpcbs are released as we attach
1610 * the allocations to them.
1612 tcp_fastopen_destroy();
1614 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1615 VNET_PCPUSTAT_FREE(tcpstat);
1618 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1620 printf("%s: WARNING: unable to deregister helper hook "
1621 "type=%d, id=%d: error %d returned\n", __func__,
1622 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1624 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1626 printf("%s: WARNING: unable to deregister helper hook "
1627 "type=%d, id=%d: error %d returned\n", __func__,
1628 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1632 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1642 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1643 * tcp_template used to store this data in mbufs, but we now recopy it out
1644 * of the tcpcb each time to conserve mbufs.
1647 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1649 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1651 INP_WLOCK_ASSERT(inp);
1654 if ((inp->inp_vflag & INP_IPV6) != 0) {
1655 struct ip6_hdr *ip6;
1657 ip6 = (struct ip6_hdr *)ip_ptr;
1658 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1659 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1660 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1661 (IPV6_VERSION & IPV6_VERSION_MASK);
1663 ip6->ip6_nxt = IPPROTO_TCP;
1665 ip6->ip6_nxt = IPPROTO_UDP;
1666 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1667 ip6->ip6_src = inp->in6p_laddr;
1668 ip6->ip6_dst = inp->in6p_faddr;
1671 #if defined(INET6) && defined(INET)
1678 ip = (struct ip *)ip_ptr;
1679 ip->ip_v = IPVERSION;
1681 ip->ip_tos = inp->inp_ip_tos;
1685 ip->ip_ttl = inp->inp_ip_ttl;
1688 ip->ip_p = IPPROTO_TCP;
1690 ip->ip_p = IPPROTO_UDP;
1691 ip->ip_src = inp->inp_laddr;
1692 ip->ip_dst = inp->inp_faddr;
1695 th->th_sport = inp->inp_lport;
1696 th->th_dport = inp->inp_fport;
1700 tcp_set_flags(th, 0);
1703 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1707 * Create template to be used to send tcp packets on a connection.
1708 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1709 * use for this function is in keepalives, which use tcp_respond.
1712 tcpip_maketemplate(struct inpcb *inp)
1716 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1719 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1724 * Send a single message to the TCP at address specified by
1725 * the given TCP/IP header. If m == NULL, then we make a copy
1726 * of the tcpiphdr at th and send directly to the addressed host.
1727 * This is used to force keep alive messages out using the TCP
1728 * template for a connection. If flags are given then we send
1729 * a message back to the TCP which originated the segment th,
1730 * and discard the mbuf containing it and any other attached mbufs.
1732 * In any case the ack and sequence number of the transmitted
1733 * segment are as specified by the parameters.
1735 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1738 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1739 tcp_seq ack, tcp_seq seq, uint16_t flags)
1745 struct udphdr *uh = NULL;
1747 struct tcp_log_buffer *lgb;
1750 struct ip6_hdr *ip6;
1753 int optlen, tlen, win, ulen;
1759 int thflags = tcp_get_flags(th);
1762 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1766 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1772 inp = tptoinpcb(tp);
1773 INP_LOCK_ASSERT(inp);
1779 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1780 port = m->m_pkthdr.tcp_tun_port;
1783 if (ip && (ip->ip_p == IPPROTO_UDP))
1784 port = m->m_pkthdr.tcp_tun_port;
1793 if (!(flags & TH_RST)) {
1794 win = sbspace(&inp->inp_socket->so_rcv);
1795 if (win > TCP_MAXWIN << tp->rcv_scale)
1796 win = TCP_MAXWIN << tp->rcv_scale;
1798 if ((tp->t_flags & TF_NOOPT) == 0)
1802 m = m_gethdr(M_NOWAIT, MT_DATA);
1805 m->m_data += max_linkhdr;
1808 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1809 sizeof(struct ip6_hdr));
1810 ip6 = mtod(m, struct ip6_hdr *);
1811 nth = (struct tcphdr *)(ip6 + 1);
1813 /* Insert a UDP header */
1814 uh = (struct udphdr *)nth;
1815 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1816 uh->uh_dport = port;
1817 nth = (struct tcphdr *)(uh + 1);
1822 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1823 ip = mtod(m, struct ip *);
1824 nth = (struct tcphdr *)(ip + 1);
1826 /* Insert a UDP header */
1827 uh = (struct udphdr *)nth;
1828 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1829 uh->uh_dport = port;
1830 nth = (struct tcphdr *)(uh + 1);
1833 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1835 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1838 /* Can't reuse 'm', allocate a new mbuf. */
1839 n = m_gethdr(M_NOWAIT, MT_DATA);
1845 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1851 n->m_data += max_linkhdr;
1852 /* m_len is set later */
1853 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1856 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1857 sizeof(struct ip6_hdr));
1858 ip6 = mtod(n, struct ip6_hdr *);
1859 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1860 nth = (struct tcphdr *)(ip6 + 1);
1862 /* Insert a UDP header */
1863 uh = (struct udphdr *)nth;
1864 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1865 uh->uh_dport = port;
1866 nth = (struct tcphdr *)(uh + 1);
1871 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1872 ip = mtod(n, struct ip *);
1873 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1874 nth = (struct tcphdr *)(ip + 1);
1876 /* Insert a UDP header */
1877 uh = (struct udphdr *)nth;
1878 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1879 uh->uh_dport = port;
1880 nth = (struct tcphdr *)(uh + 1);
1883 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1884 xchg(nth->th_dport, nth->th_sport, uint16_t);
1891 * XXX MRT We inherit the FIB, which is lucky.
1895 m->m_data = (caddr_t)ipgen;
1896 /* m_len is set later */
1899 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1900 nth = (struct tcphdr *)(ip6 + 1);
1904 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1905 nth = (struct tcphdr *)(ip + 1);
1909 * this is usually a case when an extension header
1910 * exists between the IPv6 header and the
1913 nth->th_sport = th->th_sport;
1914 nth->th_dport = th->th_dport;
1916 xchg(nth->th_dport, nth->th_sport, uint16_t);
1922 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1924 #if defined(INET) && defined(INET6)
1928 tlen = sizeof (struct tcpiphdr);
1931 tlen += sizeof (struct udphdr);
1934 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1935 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1936 m, tlen, (long)M_TRAILINGSPACE(m)));
1941 ect = tcp_ecn_output_established(tp, &flags, 0, false);
1942 /* Make sure we have room. */
1943 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1944 m->m_next = m_get(M_NOWAIT, MT_DATA);
1946 optp = mtod(m->m_next, u_char *);
1951 optp = (u_char *) (nth + 1);
1957 if (tp->t_flags & TF_RCVD_TSTMP) {
1958 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1959 to.to_tsecr = tp->ts_recent;
1960 to.to_flags |= TOF_TS;
1962 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1963 /* TCP-MD5 (RFC2385). */
1964 if (tp->t_flags & TF_SIGNATURE)
1965 to.to_flags |= TOF_SIGNATURE;
1967 /* Add the options. */
1968 tlen += optlen = tcp_addoptions(&to, optp);
1970 /* Update m_len in the correct mbuf. */
1971 optm->m_len += optlen;
1977 ulen = tlen - sizeof(struct ip6_hdr);
1978 uh->uh_ulen = htons(ulen);
1980 ip6->ip6_flow = htonl(ect << 20);
1981 ip6->ip6_vfc = IPV6_VERSION;
1983 ip6->ip6_nxt = IPPROTO_UDP;
1985 ip6->ip6_nxt = IPPROTO_TCP;
1986 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1989 #if defined(INET) && defined(INET6)
1995 ulen = tlen - sizeof(struct ip);
1996 uh->uh_ulen = htons(ulen);
1999 ip->ip_len = htons(tlen);
2000 ip->ip_ttl = V_ip_defttl;
2002 ip->ip_p = IPPROTO_UDP;
2004 ip->ip_p = IPPROTO_TCP;
2006 if (V_path_mtu_discovery)
2007 ip->ip_off |= htons(IP_DF);
2010 m->m_pkthdr.len = tlen;
2011 m->m_pkthdr.rcvif = NULL;
2015 * Packet is associated with a socket, so allow the
2016 * label of the response to reflect the socket label.
2018 INP_LOCK_ASSERT(inp);
2019 mac_inpcb_create_mbuf(inp, m);
2022 * Packet is not associated with a socket, so possibly
2023 * update the label in place.
2025 mac_netinet_tcp_reply(m);
2028 nth->th_seq = htonl(seq);
2029 nth->th_ack = htonl(ack);
2030 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2031 tcp_set_flags(nth, flags);
2033 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2035 nth->th_win = htons((u_short)win);
2038 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2039 if (to.to_flags & TOF_SIGNATURE) {
2040 if (!TCPMD5_ENABLED() ||
2041 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2051 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2052 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2053 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2056 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2057 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2058 nth->th_sum = in6_cksum_pseudo(ip6,
2059 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2061 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2064 #if defined(INET6) && defined(INET)
2070 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2071 htons(ulen + IPPROTO_UDP));
2072 m->m_pkthdr.csum_flags = CSUM_UDP;
2073 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2076 m->m_pkthdr.csum_flags = CSUM_TCP;
2077 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2078 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2079 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2083 TCP_PROBE3(debug__output, tp, th, m);
2085 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2087 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2088 if (INP_WLOCKED(inp)) {
2089 union tcp_log_stackspecific log;
2092 memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2093 log.u_bbr.inhpts = inp->inp_in_hpts;
2094 log.u_bbr.flex8 = 4;
2095 log.u_bbr.pkts_out = tp->t_maxseg;
2096 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2097 log.u_bbr.delivered = 0;
2098 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2099 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2102 * We can not log the packet, since we only own the
2103 * read lock, but a write lock is needed. The read lock
2104 * is not upgraded to a write lock, since only getting
2105 * the read lock was done intentionally to improve the
2106 * handling of SYN flooding attacks.
2107 * This happens only for pure SYN segments received in
2108 * the initial CLOSED state, or received in a more
2109 * advanced state than listen and the UDP encapsulation
2110 * port is unexpected.
2111 * The incoming SYN segments do not really belong to
2112 * the TCP connection and the handling does not change
2113 * the state of the TCP connection. Therefore, the
2114 * sending of the RST segments is not logged. Please
2115 * note that also the incoming SYN segments are not
2118 * The following code ensures that the above description
2119 * is and stays correct.
2121 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2122 (tp->t_state == TCPS_CLOSED ||
2123 (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2124 ("%s: Logging of TCP segment with flags 0x%b and "
2125 "UDP encapsulation port %u skipped in state %s",
2126 __func__, thflags, PRINT_TH_FLAGS,
2127 ntohs(port), tcpstates[tp->t_state]));
2132 TCPSTAT_INC(tcps_sndacks);
2133 else if (flags & (TH_SYN|TH_FIN|TH_RST))
2134 TCPSTAT_INC(tcps_sndctrl);
2135 TCPSTAT_INC(tcps_sndtotal);
2139 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2140 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2143 #if defined(INET) && defined(INET6)
2148 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2149 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2153 lgb->tlb_errno = output_ret;
2157 * Create a new TCP control block, making an empty reassembly queue and hooking
2158 * it to the argument protocol control block. The `inp' parameter must have
2159 * come from the zone allocator set up by tcpcbstor declaration.
2162 tcp_newtcpcb(struct inpcb *inp)
2164 struct tcpcb *tp = intotcpcb(inp);
2166 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2170 * Historically allocation was done with M_ZERO. There is a lot of
2171 * code that rely on that. For now take safe approach and zero whole
2172 * tcpcb. This definitely can be optimized.
2174 bzero(&tp->t_start_zero, t_zero_size);
2176 /* Initialise cc_var struct for this tcpcb. */
2177 tp->t_ccv.type = IPPROTO_TCP;
2178 tp->t_ccv.ccvc.tcp = tp;
2179 rw_rlock(&tcp_function_lock);
2180 tp->t_fb = tcp_func_set_ptr;
2181 refcount_acquire(&tp->t_fb->tfb_refcnt);
2182 rw_runlock(&tcp_function_lock);
2184 * Use the current system default CC algorithm.
2186 cc_attach(tp, CC_DEFAULT_ALGO());
2188 if (CC_ALGO(tp)->cb_init != NULL)
2189 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2191 if (tp->t_fb->tfb_tcp_fb_fini)
2192 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2193 refcount_release(&tp->t_fb->tfb_refcnt);
2198 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2199 if (tp->t_fb->tfb_tcp_fb_fini)
2200 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2201 refcount_release(&tp->t_fb->tfb_refcnt);
2206 TAILQ_INIT(&tp->t_segq);
2209 isipv6 ? V_tcp_v6mssdflt :
2213 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2214 for (int i = 0; i < TT_N; i++)
2215 tp->t_timers[i] = SBT_MAX;
2217 switch (V_tcp_do_rfc1323) {
2222 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2225 tp->t_flags = TF_REQ_SCALE;
2228 tp->t_flags = TF_REQ_TSTMP;
2232 tp->t_flags |= TF_SACK_PERMIT;
2233 TAILQ_INIT(&tp->snd_holes);
2236 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2237 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2238 * reasonable initial retransmit time.
2240 tp->t_srtt = TCPTV_SRTTBASE;
2241 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2242 tp->t_rttmin = tcp_rexmit_min;
2243 tp->t_rxtcur = tcp_rexmit_initial;
2244 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2245 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2246 tp->t_rcvtime = ticks;
2248 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2249 * because the socket may be bound to an IPv6 wildcard address,
2250 * which may match an IPv4-mapped IPv6 address.
2252 inp->inp_ip_ttl = V_ip_defttl;
2255 * If using hpts lets drop a random number in so
2256 * not all new connections fall on the same CPU.
2258 inp->inp_hpts_cpu = hpts_random_cpu(inp);
2262 * Init the TCP PCAP queues.
2264 tcp_pcap_tcpcb_init(tp);
2267 /* Initialize the per-TCPCB log data. */
2268 tcp_log_tcpcbinit(tp);
2270 tp->t_pacing_rate = -1;
2271 if (tp->t_fb->tfb_tcp_fb_init) {
2272 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2273 refcount_release(&tp->t_fb->tfb_refcnt);
2278 if (V_tcp_perconn_stats_enable == 1)
2279 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2282 tp->t_flags |= TF_LRD;
2288 * Drop a TCP connection, reporting
2289 * the specified error. If connection is synchronized,
2290 * then send a RST to peer.
2293 tcp_drop(struct tcpcb *tp, int errno)
2295 struct socket *so = tptosocket(tp);
2298 INP_WLOCK_ASSERT(tptoinpcb(tp));
2300 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2301 tcp_state_change(tp, TCPS_CLOSED);
2302 /* Don't use tcp_output() here due to possible recursion. */
2303 (void)tcp_output_nodrop(tp);
2304 TCPSTAT_INC(tcps_drops);
2306 TCPSTAT_INC(tcps_conndrops);
2307 if (errno == ETIMEDOUT && tp->t_softerror)
2308 errno = tp->t_softerror;
2309 so->so_error = errno;
2310 return (tcp_close(tp));
2314 tcp_discardcb(struct tcpcb *tp)
2316 struct inpcb *inp = tptoinpcb(tp);
2317 struct socket *so = tptosocket(tp);
2319 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2322 INP_WLOCK_ASSERT(inp);
2325 if (tp->t_fb->tfb_tcp_timer_stop_all) {
2326 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2329 /* free the reassembly queue, if any */
2330 tcp_reass_flush(tp);
2333 /* Disconnect offload device, if any. */
2334 if (tp->t_flags & TF_TOE)
2335 tcp_offload_detach(tp);
2338 tcp_free_sackholes(tp);
2341 /* Free the TCP PCAP queues. */
2342 tcp_pcap_drain(&(tp->t_inpkts));
2343 tcp_pcap_drain(&(tp->t_outpkts));
2346 /* Allow the CC algorithm to clean up after itself. */
2347 if (CC_ALGO(tp)->cb_destroy != NULL)
2348 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2350 /* Detach from the CC algorithm */
2354 khelp_destroy_osd(&tp->t_osd);
2357 stats_blob_destroy(tp->t_stats);
2363 tcp_log_tcpcbfini(tp);
2365 TCPSTATES_DEC(tp->t_state);
2366 if (tp->t_fb->tfb_tcp_fb_fini)
2367 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2370 * If we got enough samples through the srtt filter,
2371 * save the rtt and rttvar in the routing entry.
2372 * 'Enough' is arbitrarily defined as 4 rtt samples.
2373 * 4 samples is enough for the srtt filter to converge
2374 * to within enough % of the correct value; fewer samples
2375 * and we could save a bogus rtt. The danger is not high
2376 * as tcp quickly recovers from everything.
2377 * XXX: Works very well but needs some more statistics!
2379 * XXXRRS: Updating must be after the stack fini() since
2380 * that may be converting some internal representation of
2381 * say srtt etc into the general one used by other stacks.
2382 * Lets also at least protect against the so being NULL
2383 * as RW stated below.
2385 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2386 struct hc_metrics_lite metrics;
2389 bzero(&metrics, sizeof(metrics));
2391 * Update the ssthresh always when the conditions below
2392 * are satisfied. This gives us better new start value
2393 * for the congestion avoidance for new connections.
2394 * ssthresh is only set if packet loss occurred on a session.
2396 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2397 * being torn down. Ideally this code would not use 'so'.
2399 ssthresh = tp->snd_ssthresh;
2400 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2402 * convert the limit from user data bytes to
2403 * packets then to packet data bytes.
2405 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2408 ssthresh *= (tp->t_maxseg +
2410 (isipv6 ? sizeof (struct ip6_hdr) +
2411 sizeof (struct tcphdr) :
2413 sizeof (struct tcpiphdr)
2420 metrics.rmx_ssthresh = ssthresh;
2422 metrics.rmx_rtt = tp->t_srtt;
2423 metrics.rmx_rttvar = tp->t_rttvar;
2424 metrics.rmx_cwnd = tp->snd_cwnd;
2425 metrics.rmx_sendpipe = 0;
2426 metrics.rmx_recvpipe = 0;
2428 tcp_hc_update(&inp->inp_inc, &metrics);
2431 refcount_release(&tp->t_fb->tfb_refcnt);
2435 * Attempt to close a TCP control block, marking it as dropped, and freeing
2436 * the socket if we hold the only reference.
2439 tcp_close(struct tcpcb *tp)
2441 struct inpcb *inp = tptoinpcb(tp);
2442 struct socket *so = tptosocket(tp);
2444 INP_WLOCK_ASSERT(inp);
2447 if (tp->t_state == TCPS_LISTEN)
2448 tcp_offload_listen_stop(tp);
2451 * This releases the TFO pending counter resource for TFO listen
2452 * sockets as well as passively-created TFO sockets that transition
2453 * from SYN_RECEIVED to CLOSED.
2455 if (tp->t_tfo_pending) {
2456 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2457 tp->t_tfo_pending = NULL;
2460 tcp_hpts_remove(inp);
2463 TCPSTAT_INC(tcps_closed);
2464 if (tp->t_state != TCPS_CLOSED)
2465 tcp_state_change(tp, TCPS_CLOSED);
2466 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2467 soisdisconnected(so);
2468 if (inp->inp_flags & INP_SOCKREF) {
2469 inp->inp_flags &= ~INP_SOCKREF;
2478 * Notify a tcp user of an asynchronous error;
2479 * store error as soft error, but wake up user
2480 * (for now, won't do anything until can select for soft error).
2482 * Do not wake up user since there currently is no mechanism for
2483 * reporting soft errors (yet - a kqueue filter may be added).
2485 static struct inpcb *
2486 tcp_notify(struct inpcb *inp, int error)
2490 INP_WLOCK_ASSERT(inp);
2492 tp = intotcpcb(inp);
2493 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2496 * Ignore some errors if we are hooked up.
2497 * If connection hasn't completed, has retransmitted several times,
2498 * and receives a second error, give up now. This is better
2499 * than waiting a long time to establish a connection that
2500 * can never complete.
2502 if (tp->t_state == TCPS_ESTABLISHED &&
2503 (error == EHOSTUNREACH || error == ENETUNREACH ||
2504 error == EHOSTDOWN)) {
2505 if (inp->inp_route.ro_nh) {
2506 NH_FREE(inp->inp_route.ro_nh);
2507 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2510 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2512 tp = tcp_drop(tp, error);
2518 tp->t_softerror = error;
2522 wakeup( &so->so_timeo);
2529 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2531 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2532 INPLOOKUP_RLOCKPCB);
2537 if (req->newptr != NULL)
2540 if (req->oldptr == NULL) {
2543 n = V_tcbinfo.ipi_count +
2544 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2545 n += imax(n / 8, 10);
2546 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2550 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2553 bzero(&xig, sizeof(xig));
2554 xig.xig_len = sizeof xig;
2555 xig.xig_count = V_tcbinfo.ipi_count +
2556 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2557 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2558 xig.xig_sogen = so_gencnt;
2559 error = SYSCTL_OUT(req, &xig, sizeof xig);
2563 error = syncache_pcblist(req);
2567 while ((inp = inp_next(&inpi)) != NULL) {
2568 if (inp->inp_gencnt <= xig.xig_gen &&
2569 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2572 tcp_inptoxtp(inp, &xt);
2573 error = SYSCTL_OUT(req, &xt, sizeof xt);
2584 * Give the user an updated idea of our state.
2585 * If the generation differs from what we told
2586 * her before, she knows that something happened
2587 * while we were processing this request, and it
2588 * might be necessary to retry.
2590 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2591 xig.xig_sogen = so_gencnt;
2592 xig.xig_count = V_tcbinfo.ipi_count +
2593 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2594 error = SYSCTL_OUT(req, &xig, sizeof xig);
2600 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2601 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2602 NULL, 0, tcp_pcblist, "S,xtcpcb",
2603 "List of active TCP connections");
2607 tcp_getcred(SYSCTL_HANDLER_ARGS)
2610 struct sockaddr_in addrs[2];
2611 struct epoch_tracker et;
2615 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2618 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2621 NET_EPOCH_ENTER(et);
2622 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2623 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2627 error = cr_canseeinpcb(req->td->td_ucred, inp);
2629 cru2x(inp->inp_cred, &xuc);
2634 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2638 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2639 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2640 0, 0, tcp_getcred, "S,xucred",
2641 "Get the xucred of a TCP connection");
2646 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2648 struct epoch_tracker et;
2650 struct sockaddr_in6 addrs[2];
2657 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2660 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2663 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2664 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2667 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2669 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2676 NET_EPOCH_ENTER(et);
2679 inp = in_pcblookup(&V_tcbinfo,
2680 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2682 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2683 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2686 inp = in6_pcblookup(&V_tcbinfo,
2687 &addrs[1].sin6_addr, addrs[1].sin6_port,
2688 &addrs[0].sin6_addr, addrs[0].sin6_port,
2689 INPLOOKUP_RLOCKPCB, NULL);
2693 error = cr_canseeinpcb(req->td->td_ucred, inp);
2695 cru2x(inp->inp_cred, &xuc);
2700 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2704 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2705 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2706 0, 0, tcp6_getcred, "S,xucred",
2707 "Get the xucred of a TCP6 connection");
2711 /* Path MTU to try next when a fragmentation-needed message is received. */
2713 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2715 int mtu = ntohs(icp->icmp_nextmtu);
2717 /* If no alternative MTU was proposed, try the next smaller one. */
2719 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2720 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2721 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2727 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2733 struct inpcb *(*notify)(struct inpcb *, int);
2734 struct in_conninfo inc;
2735 tcp_seq icmp_tcp_seq;
2738 errno = icmp_errmap(icp);
2743 notify = tcp_mtudisc_notify;
2747 notify = tcp_drop_syn_sent;
2749 notify = tcp_notify;
2752 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2753 notify = tcp_drop_syn_sent;
2755 notify = tcp_notify;
2758 notify = tcp_notify;
2762 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2763 icmp_tcp_seq = th->th_seq;
2764 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2765 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2767 tp = intotcpcb(inp);
2769 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2771 * MTU discovery for offloaded connections. Let
2772 * the TOE driver verify seq# and process it.
2774 mtu = tcp_next_pmtu(icp, ip);
2775 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2779 if (tp->t_port != port)
2781 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2782 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2783 if (errno == EMSGSIZE) {
2785 * MTU discovery: we got a needfrag and
2786 * will potentially try a lower MTU.
2788 mtu = tcp_next_pmtu(icp, ip);
2791 * Only process the offered MTU if it
2792 * is smaller than the current one.
2794 if (mtu < tp->t_maxseg +
2795 sizeof(struct tcpiphdr)) {
2796 bzero(&inc, sizeof(inc));
2797 inc.inc_faddr = ip->ip_dst;
2799 inp->inp_inc.inc_fibnum;
2800 tcp_hc_updatemtu(&inc, mtu);
2801 inp = tcp_mtudisc(inp, mtu);
2804 inp = (*notify)(inp, errno);
2807 bzero(&inc, sizeof(inc));
2808 inc.inc_fport = th->th_dport;
2809 inc.inc_lport = th->th_sport;
2810 inc.inc_faddr = ip->ip_dst;
2811 inc.inc_laddr = ip->ip_src;
2812 syncache_unreach(&inc, icmp_tcp_seq, port);
2820 tcp_ctlinput(struct icmp *icmp)
2822 tcp_ctlinput_with_port(icmp, htons(0));
2826 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2828 /* Its a tunneled TCP over UDP icmp */
2829 struct icmp *icmp = param.icmp;
2830 struct ip *outer_ip, *inner_ip;
2832 struct tcphdr *th, ttemp;
2836 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2837 inner_ip = &icmp->icmp_ip;
2838 i_hlen = inner_ip->ip_hl << 2;
2839 o_len = ntohs(outer_ip->ip_len);
2841 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2842 /* Not enough data present */
2845 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2846 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2847 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2850 port = udp->uh_dport;
2851 th = (struct tcphdr *)(udp + 1);
2852 memcpy(&ttemp, th, sizeof(struct tcphdr));
2853 memcpy(udp, &ttemp, sizeof(struct tcphdr));
2854 /* Now adjust down the size of the outer IP header */
2855 o_len -= sizeof(struct udphdr);
2856 outer_ip->ip_len = htons(o_len);
2857 /* Now call in to the normal handling code */
2858 tcp_ctlinput_with_port(icmp, port);
2864 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2866 int mtu = ntohl(icmp6->icmp6_mtu);
2869 * If no alternative MTU was proposed, or the proposed MTU was too
2870 * small, set to the min.
2872 if (mtu < IPV6_MMTU)
2873 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */
2878 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2880 struct in6_addr *dst;
2881 struct inpcb *(*notify)(struct inpcb *, int);
2882 struct ip6_hdr *ip6;
2886 struct icmp6_hdr *icmp6;
2887 struct in_conninfo inc;
2892 tcp_seq icmp_tcp_seq;
2897 icmp6 = ip6cp->ip6c_icmp6;
2899 ip6 = ip6cp->ip6c_ip6;
2900 off = ip6cp->ip6c_off;
2901 dst = &ip6cp->ip6c_finaldst->sin6_addr;
2903 errno = icmp6_errmap(icmp6);
2908 notify = tcp_mtudisc_notify;
2912 notify = tcp_drop_syn_sent;
2914 notify = tcp_notify;
2918 * There are only four ICMPs that may reset connection:
2919 * - administratively prohibited
2920 * - port unreachable
2921 * - time exceeded in transit
2922 * - unknown next header
2924 if (V_icmp_may_rst &&
2925 ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
2926 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
2927 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
2928 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
2929 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
2930 (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
2931 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
2932 notify = tcp_drop_syn_sent;
2934 notify = tcp_notify;
2937 notify = tcp_notify;
2940 /* Check if we can safely get the ports from the tcp hdr */
2943 (int32_t) (off + sizeof(struct tcp_ports)))) {
2946 bzero(&t_ports, sizeof(struct tcp_ports));
2947 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2948 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2949 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2950 off += sizeof(struct tcp_ports);
2951 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2954 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2956 tp = intotcpcb(inp);
2958 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2959 /* MTU discovery for offloaded connections. */
2960 mtu = tcp6_next_pmtu(icmp6);
2961 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2965 if (tp->t_port != port)
2967 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2968 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2969 if (errno == EMSGSIZE) {
2972 * If we got a needfrag set the MTU
2973 * in the route to the suggested new
2974 * value (if given) and then notify.
2976 mtu = tcp6_next_pmtu(icmp6);
2978 bzero(&inc, sizeof(inc));
2979 inc.inc_fibnum = M_GETFIB(m);
2980 inc.inc_flags |= INC_ISIPV6;
2981 inc.inc6_faddr = *dst;
2982 if (in6_setscope(&inc.inc6_faddr,
2983 m->m_pkthdr.rcvif, NULL))
2986 * Only process the offered MTU if it
2987 * is smaller than the current one.
2989 if (mtu < tp->t_maxseg +
2990 sizeof (struct tcphdr) +
2991 sizeof (struct ip6_hdr)) {
2992 tcp_hc_updatemtu(&inc, mtu);
2993 tcp_mtudisc(inp, mtu);
2994 ICMP6STAT_INC(icp6s_pmtuchg);
2997 inp = (*notify)(inp, errno);
3000 bzero(&inc, sizeof(inc));
3001 inc.inc_fibnum = M_GETFIB(m);
3002 inc.inc_flags |= INC_ISIPV6;
3003 inc.inc_fport = t_ports.th_dport;
3004 inc.inc_lport = t_ports.th_sport;
3005 inc.inc6_faddr = *dst;
3006 inc.inc6_laddr = ip6->ip6_src;
3007 syncache_unreach(&inc, icmp_tcp_seq, port);
3015 tcp6_ctlinput(struct ip6ctlparam *ctl)
3017 tcp6_ctlinput_with_port(ctl, htons(0));
3021 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3023 struct ip6ctlparam *ip6cp = param.ip6cp;
3028 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3032 udp = mtod(m, struct udphdr *);
3033 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3036 port = udp->uh_dport;
3037 m_adj(m, sizeof(struct udphdr));
3038 if ((m->m_flags & M_PKTHDR) == 0) {
3039 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3041 /* Now call in to the normal handling code */
3042 tcp6_ctlinput_with_port(ip6cp, port);
3048 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3053 KASSERT(len >= SIPHASH_KEY_LENGTH,
3054 ("%s: keylen %u too short ", __func__, len));
3055 SipHash24_Init(&ctx);
3056 SipHash_SetKey(&ctx, (uint8_t *)key);
3057 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3058 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3059 switch (inc->inc_flags & INC_ISIPV6) {
3062 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3063 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3068 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3069 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3073 SipHash_Final((uint8_t *)hash, &ctx);
3075 return (hash[0] ^ hash[1]);
3079 tcp_new_ts_offset(struct in_conninfo *inc)
3081 struct in_conninfo inc_store, *local_inc;
3083 if (!V_tcp_ts_offset_per_conn) {
3084 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3085 inc_store.inc_lport = 0;
3086 inc_store.inc_fport = 0;
3087 local_inc = &inc_store;
3091 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3092 sizeof(V_ts_offset_secret)));
3096 * Following is where TCP initial sequence number generation occurs.
3098 * There are two places where we must use initial sequence numbers:
3099 * 1. In SYN-ACK packets.
3100 * 2. In SYN packets.
3102 * All ISNs for SYN-ACK packets are generated by the syncache. See
3103 * tcp_syncache.c for details.
3105 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3106 * depends on this property. In addition, these ISNs should be
3107 * unguessable so as to prevent connection hijacking. To satisfy
3108 * the requirements of this situation, the algorithm outlined in
3109 * RFC 1948 is used, with only small modifications.
3111 * Implementation details:
3113 * Time is based off the system timer, and is corrected so that it
3114 * increases by one megabyte per second. This allows for proper
3115 * recycling on high speed LANs while still leaving over an hour
3118 * As reading the *exact* system time is too expensive to be done
3119 * whenever setting up a TCP connection, we increment the time
3120 * offset in two ways. First, a small random positive increment
3121 * is added to isn_offset for each connection that is set up.
3122 * Second, the function tcp_isn_tick fires once per clock tick
3123 * and increments isn_offset as necessary so that sequence numbers
3124 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3125 * random positive increments serve only to ensure that the same
3126 * exact sequence number is never sent out twice (as could otherwise
3127 * happen when a port is recycled in less than the system tick
3130 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3131 * between seeding of isn_secret. This is normally set to zero,
3132 * as reseeding should not be necessary.
3134 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3135 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3136 * general, this means holding an exclusive (write) lock.
3139 #define ISN_BYTES_PER_SECOND 1048576
3140 #define ISN_STATIC_INCREMENT 4096
3141 #define ISN_RANDOM_INCREMENT (4096 - 1)
3142 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3144 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3145 VNET_DEFINE_STATIC(int, isn_last);
3146 VNET_DEFINE_STATIC(int, isn_last_reseed);
3147 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3148 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3150 #define V_isn_secret VNET(isn_secret)
3151 #define V_isn_last VNET(isn_last)
3152 #define V_isn_last_reseed VNET(isn_last_reseed)
3153 #define V_isn_offset VNET(isn_offset)
3154 #define V_isn_offset_old VNET(isn_offset_old)
3157 tcp_new_isn(struct in_conninfo *inc)
3160 u_int32_t projected_offset;
3163 /* Seed if this is the first use, reseed if requested. */
3164 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3165 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3167 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3168 V_isn_last_reseed = ticks;
3171 /* Compute the hash and return the ISN. */
3172 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3173 sizeof(V_isn_secret));
3174 V_isn_offset += ISN_STATIC_INCREMENT +
3175 (arc4random() & ISN_RANDOM_INCREMENT);
3176 if (ticks != V_isn_last) {
3177 projected_offset = V_isn_offset_old +
3178 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3179 if (SEQ_GT(projected_offset, V_isn_offset))
3180 V_isn_offset = projected_offset;
3181 V_isn_offset_old = V_isn_offset;
3184 new_isn += V_isn_offset;
3190 * When a specific ICMP unreachable message is received and the
3191 * connection state is SYN-SENT, drop the connection. This behavior
3192 * is controlled by the icmp_may_rst sysctl.
3194 static struct inpcb *
3195 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3200 INP_WLOCK_ASSERT(inp);
3202 tp = intotcpcb(inp);
3203 if (tp->t_state != TCPS_SYN_SENT)
3206 if (IS_FASTOPEN(tp->t_flags))
3207 tcp_fastopen_disable_path(tp);
3209 tp = tcp_drop(tp, errno);
3217 * When `need fragmentation' ICMP is received, update our idea of the MSS
3218 * based on the new value. Also nudge TCP to send something, since we
3219 * know the packet we just sent was dropped.
3220 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3222 static struct inpcb *
3223 tcp_mtudisc_notify(struct inpcb *inp, int error)
3226 return (tcp_mtudisc(inp, -1));
3229 static struct inpcb *
3230 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3235 INP_WLOCK_ASSERT(inp);
3237 tp = intotcpcb(inp);
3238 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3240 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3242 so = inp->inp_socket;
3243 SOCKBUF_LOCK(&so->so_snd);
3244 /* If the mss is larger than the socket buffer, decrease the mss. */
3245 if (so->so_snd.sb_hiwat < tp->t_maxseg)
3246 tp->t_maxseg = so->so_snd.sb_hiwat;
3247 SOCKBUF_UNLOCK(&so->so_snd);
3249 TCPSTAT_INC(tcps_mturesent);
3251 tp->snd_nxt = tp->snd_una;
3252 tcp_free_sackholes(tp);
3253 tp->snd_recover = tp->snd_max;
3254 if (tp->t_flags & TF_SACK_PERMIT)
3255 EXIT_FASTRECOVERY(tp->t_flags);
3256 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3258 * Conceptually the snd_nxt setting
3259 * and freeing sack holes should
3260 * be done by the default stacks
3261 * own tfb_tcp_mtu_chg().
3263 tp->t_fb->tfb_tcp_mtu_chg(tp);
3265 if (tcp_output(tp) < 0)
3273 * Look-up the routing entry to the peer of this inpcb. If no route
3274 * is found and it cannot be allocated, then return 0. This routine
3275 * is called by TCP routines that access the rmx structure and by
3276 * tcp_mss_update to get the peer/interface MTU.
3279 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3281 struct nhop_object *nh;
3283 uint32_t maxmtu = 0;
3285 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3287 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3288 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3293 maxmtu = nh->nh_mtu;
3295 /* Report additional interface capabilities. */
3297 if (ifp->if_capenable & IFCAP_TSO4 &&
3298 ifp->if_hwassist & CSUM_TSO) {
3299 cap->ifcap |= CSUM_TSO;
3300 cap->tsomax = ifp->if_hw_tsomax;
3301 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3302 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3312 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3314 struct nhop_object *nh;
3315 struct in6_addr dst6;
3318 uint32_t maxmtu = 0;
3320 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3322 if (inc->inc_flags & INC_IPV6MINMTU)
3325 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3326 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3327 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3332 maxmtu = nh->nh_mtu;
3334 /* Report additional interface capabilities. */
3336 if (ifp->if_capenable & IFCAP_TSO6 &&
3337 ifp->if_hwassist & CSUM_TSO) {
3338 cap->ifcap |= CSUM_TSO;
3339 cap->tsomax = ifp->if_hw_tsomax;
3340 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3341 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3350 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3352 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3353 * The right place to do that is ip6_setpktopt() that has just been
3354 * executed. By the way it just filled ip6po_minmtu for us.
3357 tcp6_use_min_mtu(struct tcpcb *tp)
3359 struct inpcb *inp = tptoinpcb(tp);
3361 INP_WLOCK_ASSERT(inp);
3363 * In case of the IPV6_USE_MIN_MTU socket
3364 * option, the INC_IPV6MINMTU flag to announce
3365 * a corresponding MSS during the initial
3366 * handshake. If the TCP connection is not in
3367 * the front states, just reduce the MSS being
3368 * used. This avoids the sending of TCP
3369 * segments which will be fragmented at the
3372 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3373 if ((tp->t_state >= TCPS_SYN_SENT) &&
3374 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3375 struct ip6_pktopts *opt;
3377 opt = inp->in6p_outputopts;
3378 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3379 tp->t_maxseg > TCP6_MSS)
3380 tp->t_maxseg = TCP6_MSS;
3386 * Calculate effective SMSS per RFC5681 definition for a given TCP
3387 * connection at its current state, taking into account SACK and etc.
3390 tcp_maxseg(const struct tcpcb *tp)
3394 if (tp->t_flags & TF_NOOPT)
3395 return (tp->t_maxseg);
3398 * Here we have a simplified code from tcp_addoptions(),
3399 * without a proper loop, and having most of paddings hardcoded.
3400 * We might make mistakes with padding here in some edge cases,
3401 * but this is harmless, since result of tcp_maxseg() is used
3402 * only in cwnd and ssthresh estimations.
3404 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3405 if (tp->t_flags & TF_RCVD_TSTMP)
3406 optlen = TCPOLEN_TSTAMP_APPA;
3409 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3410 if (tp->t_flags & TF_SIGNATURE)
3411 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3413 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3414 optlen += TCPOLEN_SACKHDR;
3415 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3416 optlen = PADTCPOLEN(optlen);
3419 if (tp->t_flags & TF_REQ_TSTMP)
3420 optlen = TCPOLEN_TSTAMP_APPA;
3422 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3423 if (tp->t_flags & TF_REQ_SCALE)
3424 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3425 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3426 if (tp->t_flags & TF_SIGNATURE)
3427 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3429 if (tp->t_flags & TF_SACK_PERMIT)
3430 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3433 optlen = min(optlen, TCP_MAXOLEN);
3434 return (tp->t_maxseg - optlen);
3439 tcp_fixed_maxseg(const struct tcpcb *tp)
3443 if (tp->t_flags & TF_NOOPT)
3444 return (tp->t_maxseg);
3447 * Here we have a simplified code from tcp_addoptions(),
3448 * without a proper loop, and having most of paddings hardcoded.
3449 * We only consider fixed options that we would send every
3450 * time I.e. SACK is not considered. This is important
3451 * for cc modules to figure out what the modulo of the
3454 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3455 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3456 if (tp->t_flags & TF_RCVD_TSTMP)
3457 optlen = TCPOLEN_TSTAMP_APPA;
3460 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3461 if (tp->t_flags & TF_SIGNATURE)
3462 optlen += PAD(TCPOLEN_SIGNATURE);
3465 if (tp->t_flags & TF_REQ_TSTMP)
3466 optlen = TCPOLEN_TSTAMP_APPA;
3468 optlen = PAD(TCPOLEN_MAXSEG);
3469 if (tp->t_flags & TF_REQ_SCALE)
3470 optlen += PAD(TCPOLEN_WINDOW);
3471 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3472 if (tp->t_flags & TF_SIGNATURE)
3473 optlen += PAD(TCPOLEN_SIGNATURE);
3475 if (tp->t_flags & TF_SACK_PERMIT)
3476 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3479 optlen = min(optlen, TCP_MAXOLEN);
3480 return (tp->t_maxseg - optlen);
3486 sysctl_drop(SYSCTL_HANDLER_ARGS)
3488 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3489 struct sockaddr_storage addrs[2];
3493 struct sockaddr_in *fin = NULL, *lin = NULL;
3495 struct epoch_tracker et;
3497 struct sockaddr_in6 *fin6, *lin6;
3507 if (req->oldptr != NULL || req->oldlen != 0)
3509 if (req->newptr == NULL)
3511 if (req->newlen < sizeof(addrs))
3513 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3517 switch (addrs[0].ss_family) {
3520 fin6 = (struct sockaddr_in6 *)&addrs[0];
3521 lin6 = (struct sockaddr_in6 *)&addrs[1];
3522 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3523 lin6->sin6_len != sizeof(struct sockaddr_in6))
3525 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3526 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3528 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3529 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3531 fin = (struct sockaddr_in *)&addrs[0];
3532 lin = (struct sockaddr_in *)&addrs[1];
3536 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3539 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3546 fin = (struct sockaddr_in *)&addrs[0];
3547 lin = (struct sockaddr_in *)&addrs[1];
3548 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3549 lin->sin_len != sizeof(struct sockaddr_in))
3556 NET_EPOCH_ENTER(et);
3557 switch (addrs[0].ss_family) {
3560 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3561 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3562 INPLOOKUP_WLOCKPCB, NULL);
3567 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3568 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3573 if (!SOLISTENING(inp->inp_socket)) {
3574 tp = intotcpcb(inp);
3575 tp = tcp_drop(tp, ECONNABORTED);
3586 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3587 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3588 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3589 "Drop TCP connection");
3592 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3594 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3595 &tcp_ctloutput_set));
3598 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3599 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3600 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3601 "Set socket option for TCP endpoint");
3605 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3607 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3608 struct sockaddr_storage addrs[2];
3611 struct sockaddr_in *fin = NULL, *lin = NULL;
3613 struct epoch_tracker et;
3615 struct sockaddr_in6 *fin6, *lin6;
3625 if (req->oldptr != NULL || req->oldlen != 0)
3627 if (req->newptr == NULL)
3629 if (req->newlen < sizeof(addrs))
3631 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3635 switch (addrs[0].ss_family) {
3638 fin6 = (struct sockaddr_in6 *)&addrs[0];
3639 lin6 = (struct sockaddr_in6 *)&addrs[1];
3640 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3641 lin6->sin6_len != sizeof(struct sockaddr_in6))
3643 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3644 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3646 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3647 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3649 fin = (struct sockaddr_in *)&addrs[0];
3650 lin = (struct sockaddr_in *)&addrs[1];
3654 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3657 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3664 fin = (struct sockaddr_in *)&addrs[0];
3665 lin = (struct sockaddr_in *)&addrs[1];
3666 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3667 lin->sin_len != sizeof(struct sockaddr_in))
3674 NET_EPOCH_ENTER(et);
3675 switch (addrs[0].ss_family) {
3678 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3679 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3680 INPLOOKUP_WLOCKPCB, NULL);
3685 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3686 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3694 so = inp->inp_socket;
3696 error = ktls_set_tx_mode(so,
3697 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3706 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3707 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3708 "Switch TCP connection to SW TLS");
3709 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3710 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3711 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3712 "Switch TCP connection to ifnet TLS");
3716 * Generate a standardized TCP log line for use throughout the
3717 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3718 * allow use in the interrupt context.
3720 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3721 * NB: The function may return NULL if memory allocation failed.
3723 * Due to header inclusion and ordering limitations the struct ip
3724 * and ip6_hdr pointers have to be passed as void pointers.
3727 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3731 /* Is logging enabled? */
3732 if (V_tcp_log_in_vain == 0)
3735 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3739 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3743 /* Is logging enabled? */
3744 if (tcp_log_debug == 0)
3747 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3751 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3757 const struct ip *ip = (const struct ip *)ip4hdr;
3760 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3764 * The log line looks like this:
3765 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3767 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3768 sizeof(PRINT_TH_FLAGS) + 1 +
3770 2 * INET6_ADDRSTRLEN;
3772 2 * INET_ADDRSTRLEN;
3775 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3779 strcat(s, "TCP: [");
3782 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3783 inet_ntoa_r(inc->inc_faddr, sp);
3785 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3787 inet_ntoa_r(inc->inc_laddr, sp);
3789 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3792 ip6_sprintf(sp, &inc->inc6_faddr);
3794 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3796 ip6_sprintf(sp, &inc->inc6_laddr);
3798 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3799 } else if (ip6 && th) {
3800 ip6_sprintf(sp, &ip6->ip6_src);
3802 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3804 ip6_sprintf(sp, &ip6->ip6_dst);
3806 sprintf(sp, "]:%i", ntohs(th->th_dport));
3809 } else if (ip && th) {
3810 inet_ntoa_r(ip->ip_src, sp);
3812 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3814 inet_ntoa_r(ip->ip_dst, sp);
3816 sprintf(sp, "]:%i", ntohs(th->th_dport));
3824 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3825 if (*(s + size - 1) != '\0')
3826 panic("%s: string too long", __func__);
3831 * A subroutine which makes it easy to track TCP state changes with DTrace.
3832 * This function shouldn't be called for t_state initializations that don't
3833 * correspond to actual TCP state transitions.
3836 tcp_state_change(struct tcpcb *tp, int newstate)
3838 #if defined(KDTRACE_HOOKS)
3839 int pstate = tp->t_state;
3842 TCPSTATES_DEC(tp->t_state);
3843 TCPSTATES_INC(newstate);
3844 tp->t_state = newstate;
3845 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3849 * Create an external-format (``xtcpcb'') structure using the information in
3850 * the kernel-format tcpcb structure pointed to by tp. This is done to
3851 * reduce the spew of irrelevant information over this interface, to isolate
3852 * user code from changes in the kernel structure, and potentially to provide
3853 * information-hiding if we decide that some of this information should be
3854 * hidden from users.
3857 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3859 struct tcpcb *tp = intotcpcb(inp);
3862 bzero(xt, sizeof(*xt));
3863 xt->t_state = tp->t_state;
3864 xt->t_logstate = tp->t_logstate;
3865 xt->t_flags = tp->t_flags;
3866 xt->t_sndzerowin = tp->t_sndzerowin;
3867 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3868 xt->t_rcvoopack = tp->t_rcvoopack;
3869 xt->t_rcv_wnd = tp->rcv_wnd;
3870 xt->t_snd_wnd = tp->snd_wnd;
3871 xt->t_snd_cwnd = tp->snd_cwnd;
3872 xt->t_snd_ssthresh = tp->snd_ssthresh;
3873 xt->t_dsack_bytes = tp->t_dsack_bytes;
3874 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3875 xt->t_dsack_pack = tp->t_dsack_pack;
3876 xt->t_maxseg = tp->t_maxseg;
3877 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3878 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3880 now = getsbinuptime();
3881 #define COPYTIMER(which,where) do { \
3882 if (tp->t_timers[which] != SBT_MAX) \
3883 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \
3887 COPYTIMER(TT_DELACK, tt_delack);
3888 COPYTIMER(TT_REXMT, tt_rexmt);
3889 COPYTIMER(TT_PERSIST, tt_persist);
3890 COPYTIMER(TT_KEEP, tt_keep);
3891 COPYTIMER(TT_2MSL, tt_2msl);
3893 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3895 xt->xt_encaps_port = tp->t_port;
3896 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3897 TCP_FUNCTION_NAME_LEN_MAX);
3898 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3900 (void)tcp_log_get_id(tp, xt->xt_logid);
3903 xt->xt_len = sizeof(struct xtcpcb);
3904 in_pcbtoxinpcb(inp, &xt->xt_inp);
3908 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3913 (status > TCP_EI_STATUS_MAX_VALUE) ||
3918 if (status > (sizeof(uint32_t) * 8)) {
3919 /* Should this be a KASSERT? */
3922 bit = 1U << (status - 1);
3923 if (bit & tp->t_end_info_status) {
3924 /* already logged */
3927 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3928 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3929 tp->t_end_info_bytes[i] = status;
3930 tp->t_end_info_status |= bit;
3937 tcp_can_enable_pacing(void)
3940 if ((tcp_pacing_limit == -1) ||
3941 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
3942 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
3943 shadow_num_connections = number_of_tcp_connections_pacing;
3950 static uint8_t tcp_pacing_warning = 0;
3953 tcp_decrement_paced_conn(void)
3957 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
3958 shadow_num_connections = number_of_tcp_connections_pacing;
3959 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
3961 if (tcp_pacing_limit != -1) {
3962 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
3963 tcp_pacing_limit = 0;
3964 } else if (tcp_pacing_warning == 0) {
3965 printf("Warning pacing count is invalid, invalid decrement\n");
3966 tcp_pacing_warning = 1;