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>
80 #include <netinet/in.h>
81 #include <netinet/in_fib.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip.h>
87 #include <netinet/ip_icmp.h>
88 #include <netinet/ip_var.h>
90 #include <netinet/icmp6.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/in6_fib.h>
93 #include <netinet6/in6_pcb.h>
94 #include <netinet6/ip6_var.h>
95 #include <netinet6/scope6_var.h>
96 #include <netinet6/nd6.h>
99 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcp_ecn.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_syncache.h>
110 #include <netinet/tcp_hpts.h>
111 #include <netinet/cc/cc.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/tcp_fastopen.h>
115 #include <netinet/tcp_pcap.h>
118 #include <netinet/tcp_offload.h>
120 #include <netinet/udp.h>
121 #include <netinet/udp_var.h>
123 #include <netinet6/tcp6_var.h>
126 #include <netipsec/ipsec_support.h>
128 #include <machine/in_cksum.h>
129 #include <crypto/siphash/siphash.h>
131 #include <security/mac/mac_framework.h>
134 static ip6proto_ctlinput_t tcp6_ctlinput;
135 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
138 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
140 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
143 #ifdef NETFLIX_EXP_DETECTION
144 /* Sack attack detection thresholds and such */
145 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
146 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
147 "Sack Attack detection thresholds");
148 int32_t tcp_force_detection = 0;
149 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
151 &tcp_force_detection, 0,
152 "Do we force detection even if the INP has it off?");
153 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
154 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
156 &tcp_sack_to_ack_thresh, 700,
157 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
158 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
159 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
161 &tcp_sack_to_move_thresh, 600,
162 "Percentage of sack moves we must see above (10.1 percent is 101)");
163 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
164 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
166 &tcp_restoral_thresh, 550,
167 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
168 int32_t tcp_sad_decay_val = 800;
169 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
171 &tcp_sad_decay_val, 800,
172 "The decay percentage (10.1 percent equals 101 )");
173 int32_t tcp_map_minimum = 500;
174 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
176 &tcp_map_minimum, 500,
177 "Number of Map enteries before we start detection");
178 int32_t tcp_attack_on_turns_on_logging = 0;
179 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
181 &tcp_attack_on_turns_on_logging, 0,
182 "When we have a positive hit on attack, do we turn on logging?");
183 int32_t tcp_sad_pacing_interval = 2000;
184 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
186 &tcp_sad_pacing_interval, 2000,
187 "What is the minimum pacing interval for a classified attacker?");
189 int32_t tcp_sad_low_pps = 100;
190 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
192 &tcp_sad_low_pps, 100,
193 "What is the input pps that below which we do not decay?");
195 uint32_t tcp_ack_war_time_window = 1000;
196 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
198 &tcp_ack_war_time_window, 1000,
199 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
200 uint32_t tcp_ack_war_cnt = 5;
201 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
204 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
206 struct rwlock tcp_function_lock;
209 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
214 error = sysctl_handle_int(oidp, &new, 0, req);
215 if (error == 0 && req->newptr) {
216 if (new < TCP_MINMSS)
224 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
225 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
226 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
227 "Default TCP Maximum Segment Size");
231 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
235 new = V_tcp_v6mssdflt;
236 error = sysctl_handle_int(oidp, &new, 0, req);
237 if (error == 0 && req->newptr) {
238 if (new < TCP_MINMSS)
241 V_tcp_v6mssdflt = new;
246 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
247 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
248 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
249 "Default TCP Maximum Segment Size for IPv6");
253 * Minimum MSS we accept and use. This prevents DoS attacks where
254 * we are forced to a ridiculous low MSS like 20 and send hundreds
255 * of packets instead of one. The effect scales with the available
256 * bandwidth and quickly saturates the CPU and network interface
257 * with packet generation and sending. Set to zero to disable MINMSS
258 * checking. This setting prevents us from sending too small packets.
260 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
261 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
262 &VNET_NAME(tcp_minmss), 0,
263 "Minimum TCP Maximum Segment Size");
265 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
266 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
267 &VNET_NAME(tcp_do_rfc1323), 0,
268 "Enable rfc1323 (high performance TCP) extensions");
271 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
272 * Some stacks negotiate TS, but never send them after connection setup. Some
273 * stacks negotiate TS, but don't send them when sending keep-alive segments.
274 * These include modern widely deployed TCP stacks.
275 * Therefore tolerating violations for now...
277 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
278 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
279 &VNET_NAME(tcp_tolerate_missing_ts), 0,
280 "Tolerate missing TCP timestamps");
282 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
283 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
284 &VNET_NAME(tcp_ts_offset_per_conn), 0,
285 "Initialize TCP timestamps per connection instead of per host pair");
287 /* How many connections are pacing */
288 static volatile uint32_t number_of_tcp_connections_pacing = 0;
289 static uint32_t shadow_num_connections = 0;
291 static int tcp_pacing_limit = 10000;
292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
293 &tcp_pacing_limit, 1000,
294 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
296 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
297 &shadow_num_connections, 0, "Number of TCP connections being paced");
299 static int tcp_log_debug = 0;
300 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
301 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
303 static int tcp_tcbhashsize;
304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
305 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
307 static int do_tcpdrain = 1;
308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
309 "Enable tcp_drain routine for extra help when low on mbufs");
311 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
312 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
314 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
315 #define V_icmp_may_rst VNET(icmp_may_rst)
316 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
317 &VNET_NAME(icmp_may_rst), 0,
318 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
320 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
321 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
322 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
323 &VNET_NAME(tcp_isn_reseed_interval), 0,
324 "Seconds between reseeding of ISN secret");
326 static int tcp_soreceive_stream;
327 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
328 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
330 VNET_DEFINE(uma_zone_t, sack_hole_zone);
331 #define V_sack_hole_zone VNET(sack_hole_zone)
332 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
334 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
339 new = V_tcp_map_entries_limit;
340 error = sysctl_handle_int(oidp, &new, 0, req);
341 if (error == 0 && req->newptr) {
342 /* only allow "0" and value > minimum */
343 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
346 V_tcp_map_entries_limit = new;
350 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
351 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
352 &VNET_NAME(tcp_map_entries_limit), 0,
353 &sysctl_net_inet_tcp_map_limit_check, "IU",
354 "Total sendmap entries limit");
356 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
357 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
358 &VNET_NAME(tcp_map_split_limit), 0,
359 "Total sendmap split entries limit");
362 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
365 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
366 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
367 #define V_ts_offset_secret VNET(ts_offset_secret)
369 static int tcp_default_fb_init(struct tcpcb *tp);
370 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
371 static int tcp_default_handoff_ok(struct tcpcb *tp);
372 static struct inpcb *tcp_notify(struct inpcb *, int);
373 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
374 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
375 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
376 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
377 const void *ip4hdr, const void *ip6hdr);
378 static ipproto_ctlinput_t tcp_ctlinput;
379 static udp_tun_icmp_t tcp_ctlinput_viaudp;
381 static struct tcp_function_block tcp_def_funcblk = {
382 .tfb_tcp_block_name = "freebsd",
383 .tfb_tcp_output = tcp_default_output,
384 .tfb_tcp_do_segment = tcp_do_segment,
385 .tfb_tcp_ctloutput = tcp_default_ctloutput,
386 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
387 .tfb_tcp_fb_init = tcp_default_fb_init,
388 .tfb_tcp_fb_fini = tcp_default_fb_fini,
391 static int tcp_fb_cnt = 0;
392 struct tcp_funchead t_functions;
393 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
396 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
398 TCPSTAT_INC(tcps_dsack_count);
401 if (SEQ_GT(end, start)) {
402 tp->t_dsack_bytes += (end - start);
403 TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
405 tp->t_dsack_tlp_bytes += (start - end);
406 TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
409 if (SEQ_GT(end, start)) {
410 tp->t_dsack_bytes += (end - start);
411 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
413 tp->t_dsack_tlp_bytes += (start - end);
414 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
419 static struct tcp_function_block *
420 find_tcp_functions_locked(struct tcp_function_set *fs)
422 struct tcp_function *f;
423 struct tcp_function_block *blk=NULL;
425 TAILQ_FOREACH(f, &t_functions, tf_next) {
426 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
434 static struct tcp_function_block *
435 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
437 struct tcp_function_block *rblk=NULL;
438 struct tcp_function *f;
440 TAILQ_FOREACH(f, &t_functions, tf_next) {
441 if (f->tf_fb == blk) {
452 struct tcp_function_block *
453 find_and_ref_tcp_functions(struct tcp_function_set *fs)
455 struct tcp_function_block *blk;
457 rw_rlock(&tcp_function_lock);
458 blk = find_tcp_functions_locked(fs);
460 refcount_acquire(&blk->tfb_refcnt);
461 rw_runlock(&tcp_function_lock);
465 struct tcp_function_block *
466 find_and_ref_tcp_fb(struct tcp_function_block *blk)
468 struct tcp_function_block *rblk;
470 rw_rlock(&tcp_function_lock);
471 rblk = find_tcp_fb_locked(blk, NULL);
473 refcount_acquire(&rblk->tfb_refcnt);
474 rw_runlock(&tcp_function_lock);
478 /* Find a matching alias for the given tcp_function_block. */
480 find_tcp_function_alias(struct tcp_function_block *blk,
481 struct tcp_function_set *fs)
483 struct tcp_function *f;
487 rw_rlock(&tcp_function_lock);
488 TAILQ_FOREACH(f, &t_functions, tf_next) {
489 if ((f->tf_fb == blk) &&
490 (strncmp(f->tf_name, blk->tfb_tcp_block_name,
491 TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
492 /* Matching function block with different name. */
493 strncpy(fs->function_set_name, f->tf_name,
494 TCP_FUNCTION_NAME_LEN_MAX);
499 /* Null terminate the string appropriately. */
501 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
503 fs->function_set_name[0] = '\0';
505 rw_runlock(&tcp_function_lock);
509 static struct tcp_function_block *
510 find_and_ref_tcp_default_fb(void)
512 struct tcp_function_block *rblk;
514 rw_rlock(&tcp_function_lock);
515 rblk = tcp_func_set_ptr;
516 refcount_acquire(&rblk->tfb_refcnt);
517 rw_runlock(&tcp_function_lock);
522 tcp_switch_back_to_default(struct tcpcb *tp)
524 struct tcp_function_block *tfb;
526 KASSERT(tp->t_fb != &tcp_def_funcblk,
527 ("%s: called by the built-in default stack", __func__));
530 * Release the old stack. This function will either find a new one
533 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
534 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
535 refcount_release(&tp->t_fb->tfb_refcnt);
538 * Now, we'll find a new function block to use.
539 * Start by trying the current user-selected
540 * default, unless this stack is the user-selected
543 tfb = find_and_ref_tcp_default_fb();
544 if (tfb == tp->t_fb) {
545 refcount_release(&tfb->tfb_refcnt);
548 /* Does the stack accept this connection? */
549 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
550 (*tfb->tfb_tcp_handoff_ok)(tp)) {
551 refcount_release(&tfb->tfb_refcnt);
554 /* Try to use that stack. */
556 /* Initialize the new stack. If it succeeds, we are done. */
558 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
559 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
563 * Initialization failed. Release the reference count on
566 refcount_release(&tfb->tfb_refcnt);
570 * If that wasn't feasible, use the built-in default
571 * stack which is not allowed to reject anyone.
573 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
575 /* there always should be a default */
576 panic("Can't refer to tcp_def_funcblk");
578 if (tfb->tfb_tcp_handoff_ok != NULL) {
579 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
580 /* The default stack cannot say no */
581 panic("Default stack rejects a new session?");
585 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
586 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
587 /* The default stack cannot fail */
588 panic("Default stack initialization failed");
593 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
594 const struct sockaddr *sa, void *ctx)
605 TCPSTAT_INC(tcps_tunneled_pkts);
606 if ((m->m_flags & M_PKTHDR) == 0) {
607 /* Can't handle one that is not a pkt hdr */
608 TCPSTAT_INC(tcps_tunneled_errs);
611 thlen = sizeof(struct tcphdr);
612 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
613 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
614 TCPSTAT_INC(tcps_tunneled_errs);
617 iph = mtod(m, struct ip *);
618 uh = (struct udphdr *)((caddr_t)iph + off);
619 th = (struct tcphdr *)(uh + 1);
620 thlen = th->th_off << 2;
621 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
622 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
624 TCPSTAT_INC(tcps_tunneled_errs);
627 iph = mtod(m, struct ip *);
628 uh = (struct udphdr *)((caddr_t)iph + off);
629 th = (struct tcphdr *)(uh + 1);
632 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
633 bcopy(th, uh, m->m_len - off);
634 m->m_len -= sizeof(struct udphdr);
635 m->m_pkthdr.len -= sizeof(struct udphdr);
637 * We use the same algorithm for
638 * both UDP and TCP for c-sum. So
639 * the code in tcp_input will skip
640 * the checksum. So we do nothing
641 * with the flag (m->m_pkthdr.csum_flags).
646 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
647 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
651 case IPV6_VERSION >> 4:
652 ip6 = mtod(m, struct ip6_hdr *);
653 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
654 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
669 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
672 struct tcp_function_set fs;
673 struct tcp_function_block *blk;
675 memset(&fs, 0, sizeof(fs));
676 rw_rlock(&tcp_function_lock);
677 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
680 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
681 fs.pcbcnt = blk->tfb_refcnt;
683 rw_runlock(&tcp_function_lock);
684 error = sysctl_handle_string(oidp, fs.function_set_name,
685 sizeof(fs.function_set_name), req);
687 /* Check for error or no change */
688 if (error != 0 || req->newptr == NULL)
691 rw_wlock(&tcp_function_lock);
692 blk = find_tcp_functions_locked(&fs);
694 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
698 tcp_func_set_ptr = blk;
700 rw_wunlock(&tcp_function_lock);
704 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
705 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
706 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
707 "Set/get the default TCP functions");
710 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
712 int error, cnt, linesz;
713 struct tcp_function *f;
719 rw_rlock(&tcp_function_lock);
720 TAILQ_FOREACH(f, &t_functions, tf_next) {
723 rw_runlock(&tcp_function_lock);
725 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
726 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
731 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
732 "Alias", "PCB count");
737 rw_rlock(&tcp_function_lock);
738 TAILQ_FOREACH(f, &t_functions, tf_next) {
739 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
740 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
741 f->tf_fb->tfb_tcp_block_name,
742 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
743 alias ? f->tf_name : "-",
744 f->tf_fb->tfb_refcnt);
745 if (linesz >= bufsz) {
753 rw_runlock(&tcp_function_lock);
755 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
756 free(buffer, M_TEMP);
760 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
761 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
762 NULL, 0, sysctl_net_inet_list_available, "A",
763 "list available TCP Function sets");
765 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
768 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
769 #define V_udp4_tun_socket VNET(udp4_tun_socket)
772 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
773 #define V_udp6_tun_socket VNET(udp6_tun_socket)
777 tcp_over_udp_stop(void)
780 * This function assumes sysctl caller holds inp_rinfo_lock()
784 if (V_udp4_tun_socket != NULL) {
785 soclose(V_udp4_tun_socket);
786 V_udp4_tun_socket = NULL;
790 if (V_udp6_tun_socket != NULL) {
791 soclose(V_udp6_tun_socket);
792 V_udp6_tun_socket = NULL;
798 tcp_over_udp_start(void)
803 struct sockaddr_in sin;
806 struct sockaddr_in6 sin6;
809 * This function assumes sysctl caller holds inp_info_rlock()
812 port = V_tcp_udp_tunneling_port;
813 if (ntohs(port) == 0) {
814 /* Must have a port set */
818 if (V_udp4_tun_socket != NULL) {
819 /* Already running -- must stop first */
824 if (V_udp6_tun_socket != NULL) {
825 /* Already running -- must stop first */
830 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
831 SOCK_DGRAM, IPPROTO_UDP,
832 curthread->td_ucred, curthread))) {
836 /* Call the special UDP hook. */
837 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
838 tcp_recv_udp_tunneled_packet,
844 /* Ok, we have a socket, bind it to the port. */
845 memset(&sin, 0, sizeof(struct sockaddr_in));
846 sin.sin_len = sizeof(struct sockaddr_in);
847 sin.sin_family = AF_INET;
848 sin.sin_port = htons(port);
849 if ((ret = sobind(V_udp4_tun_socket,
850 (struct sockaddr *)&sin, curthread))) {
856 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
857 SOCK_DGRAM, IPPROTO_UDP,
858 curthread->td_ucred, curthread))) {
862 /* Call the special UDP hook. */
863 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
864 tcp_recv_udp_tunneled_packet,
865 tcp6_ctlinput_viaudp,
870 /* Ok, we have a socket, bind it to the port. */
871 memset(&sin6, 0, sizeof(struct sockaddr_in6));
872 sin6.sin6_len = sizeof(struct sockaddr_in6);
873 sin6.sin6_family = AF_INET6;
874 sin6.sin6_port = htons(port);
875 if ((ret = sobind(V_udp6_tun_socket,
876 (struct sockaddr *)&sin6, curthread))) {
885 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
890 old = V_tcp_udp_tunneling_port;
892 error = sysctl_handle_int(oidp, &new, 0, req);
894 (req->newptr != NULL)) {
895 if ((new < TCP_TUNNELING_PORT_MIN) ||
896 (new > TCP_TUNNELING_PORT_MAX)) {
899 V_tcp_udp_tunneling_port = new;
904 error = tcp_over_udp_start();
911 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
912 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
913 &VNET_NAME(tcp_udp_tunneling_port),
914 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
915 "Tunneling port for tcp over udp");
917 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
920 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
924 new = V_tcp_udp_tunneling_overhead;
925 error = sysctl_handle_int(oidp, &new, 0, req);
926 if (error == 0 && req->newptr) {
927 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
928 (new > TCP_TUNNELING_OVERHEAD_MAX))
931 V_tcp_udp_tunneling_overhead = new;
936 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
937 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
938 &VNET_NAME(tcp_udp_tunneling_overhead),
939 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
940 "MSS reduction when using tcp over udp");
943 * Exports one (struct tcp_function_info) for each alias/name.
946 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
949 struct tcp_function *f;
950 struct tcp_function_info tfi;
953 * We don't allow writes.
955 if (req->newptr != NULL)
959 * Wire the old buffer so we can directly copy the functions to
960 * user space without dropping the lock.
962 if (req->oldptr != NULL) {
963 error = sysctl_wire_old_buffer(req, 0);
969 * Walk the list and copy out matching entries. If INVARIANTS
970 * is compiled in, also walk the list to verify the length of
971 * the list matches what we have recorded.
973 rw_rlock(&tcp_function_lock);
977 if (req->oldptr == NULL) {
982 TAILQ_FOREACH(f, &t_functions, tf_next) {
986 if (req->oldptr != NULL) {
987 bzero(&tfi, sizeof(tfi));
988 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
989 tfi.tfi_id = f->tf_fb->tfb_id;
990 (void)strlcpy(tfi.tfi_alias, f->tf_name,
991 sizeof(tfi.tfi_alias));
992 (void)strlcpy(tfi.tfi_name,
993 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
994 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
996 * Don't stop on error, as that is the
997 * mechanism we use to accumulate length
998 * information if the buffer was too short.
1002 KASSERT(cnt == tcp_fb_cnt,
1003 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1007 rw_runlock(&tcp_function_lock);
1008 if (req->oldptr == NULL)
1009 error = SYSCTL_OUT(req, NULL,
1010 (cnt + 1) * sizeof(struct tcp_function_info));
1015 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1016 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1017 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1018 "List TCP function block name-to-ID mappings");
1021 * tfb_tcp_handoff_ok() function for the default stack.
1022 * Note that we'll basically try to take all comers.
1025 tcp_default_handoff_ok(struct tcpcb *tp)
1032 * tfb_tcp_fb_init() function for the default stack.
1034 * This handles making sure we have appropriate timers set if you are
1035 * transitioning a socket that has some amount of setup done.
1037 * The init() fuction from the default can *never* return non-zero i.e.
1038 * it is required to always succeed since it is the stack of last resort!
1041 tcp_default_fb_init(struct tcpcb *tp)
1043 struct socket *so = tptosocket(tp);
1045 INP_WLOCK_ASSERT(tptoinpcb(tp));
1047 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1048 ("%s: connection %p in unexpected state %d", __func__, tp,
1052 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1053 * know what to do for unexpected states (which includes TIME_WAIT).
1055 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1059 * Make sure some kind of transmission timer is set if there is
1062 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1063 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1064 tcp_timer_active(tp, TT_PERSIST))) {
1066 * If the session has established and it looks like it should
1067 * be in the persist state, set the persist timer. Otherwise,
1068 * set the retransmit timer.
1070 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1071 (int32_t)(tp->snd_nxt - tp->snd_una) <
1072 (int32_t)sbavail(&so->so_snd))
1075 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1078 /* All non-embryonic sessions get a keepalive timer. */
1079 if (!tcp_timer_active(tp, TT_KEEP))
1080 tcp_timer_activate(tp, TT_KEEP,
1081 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1085 * Make sure critical variables are initialized
1086 * if transitioning while in Recovery.
1088 if IN_FASTRECOVERY(tp->t_flags) {
1089 if (tp->sackhint.recover_fs == 0)
1090 tp->sackhint.recover_fs = max(1,
1091 tp->snd_nxt - tp->snd_una);
1098 * tfb_tcp_fb_fini() function for the default stack.
1100 * This changes state as necessary (or prudent) to prepare for another stack
1101 * to assume responsibility for the connection.
1104 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1107 INP_WLOCK_ASSERT(tptoinpcb(tp));
1111 * Target size of TCP PCB hash tables. Must be a power of two.
1113 * Note that this can be overridden by the kernel environment
1114 * variable net.inet.tcp.tcbhashsize
1117 #define TCBHASHSIZE 0
1120 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1121 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1123 static struct mtx isn_mtx;
1125 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1126 #define ISN_LOCK() mtx_lock(&isn_mtx)
1127 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1129 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1132 * Take a value and get the next power of 2 that doesn't overflow.
1133 * Used to size the tcp_inpcb hash buckets.
1136 maketcp_hashsize(int size)
1142 * get the next power of 2 higher than maxsockets.
1144 hashsize = 1 << fls(size);
1145 /* catch overflow, and just go one power of 2 smaller */
1146 if (hashsize < size) {
1147 hashsize = 1 << (fls(size) - 1);
1152 static volatile int next_tcp_stack_id = 1;
1155 * Register a TCP function block with the name provided in the names
1156 * array. (Note that this function does NOT automatically register
1157 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1158 * explicitly include blk->tfb_tcp_block_name in the list of names if
1159 * you wish to register the stack with that name.)
1161 * Either all name registrations will succeed or all will fail. If
1162 * a name registration fails, the function will update the num_names
1163 * argument to point to the array index of the name that encountered
1166 * Returns 0 on success, or an error code on failure.
1169 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1170 const char *names[], int *num_names)
1172 struct tcp_function *n;
1173 struct tcp_function_set fs;
1176 KASSERT(names != NULL && *num_names > 0,
1177 ("%s: Called with 0-length name list", __func__));
1178 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1179 KASSERT(rw_initialized(&tcp_function_lock),
1180 ("%s: called too early", __func__));
1182 if ((blk->tfb_tcp_output == NULL) ||
1183 (blk->tfb_tcp_do_segment == NULL) ||
1184 (blk->tfb_tcp_ctloutput == NULL) ||
1185 (strlen(blk->tfb_tcp_block_name) == 0)) {
1187 * These functions are required and you
1194 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1199 refcount_init(&blk->tfb_refcnt, 0);
1200 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1201 for (i = 0; i < *num_names; i++) {
1202 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1209 (void)strlcpy(fs.function_set_name, names[i],
1210 sizeof(fs.function_set_name));
1211 rw_wlock(&tcp_function_lock);
1212 if (find_tcp_functions_locked(&fs) != NULL) {
1213 /* Duplicate name space not allowed */
1214 rw_wunlock(&tcp_function_lock);
1215 free(n, M_TCPFUNCTIONS);
1219 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1220 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1222 rw_wunlock(&tcp_function_lock);
1228 * Deregister the names we just added. Because registration failed
1229 * for names[i], we don't need to deregister that name.
1232 rw_wlock(&tcp_function_lock);
1234 TAILQ_FOREACH(n, &t_functions, tf_next) {
1235 if (!strncmp(n->tf_name, names[i],
1236 TCP_FUNCTION_NAME_LEN_MAX)) {
1237 TAILQ_REMOVE(&t_functions, n, tf_next);
1240 free(n, M_TCPFUNCTIONS);
1245 rw_wunlock(&tcp_function_lock);
1250 * Register a TCP function block using the name provided in the name
1253 * Returns 0 on success, or an error code on failure.
1256 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1259 const char *name_list[1];
1264 name_list[0] = name;
1266 name_list[0] = blk->tfb_tcp_block_name;
1267 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1272 * Register a TCP function block using the name defined in
1273 * blk->tfb_tcp_block_name.
1275 * Returns 0 on success, or an error code on failure.
1278 register_tcp_functions(struct tcp_function_block *blk, int wait)
1281 return (register_tcp_functions_as_name(blk, NULL, wait));
1285 * Deregister all names associated with a function block. This
1286 * functionally removes the function block from use within the system.
1288 * When called with a true quiesce argument, mark the function block
1289 * as being removed so no more stacks will use it and determine
1290 * whether the removal would succeed.
1292 * When called with a false quiesce argument, actually attempt the
1295 * When called with a force argument, attempt to switch all TCBs to
1296 * use the default stack instead of returning EBUSY.
1298 * Returns 0 on success (or if the removal would succeed, or an error
1302 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1305 struct tcp_function *f;
1307 if (blk == &tcp_def_funcblk) {
1308 /* You can't un-register the default */
1311 rw_wlock(&tcp_function_lock);
1312 if (blk == tcp_func_set_ptr) {
1313 /* You can't free the current default */
1314 rw_wunlock(&tcp_function_lock);
1317 /* Mark the block so no more stacks can use it. */
1318 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1320 * If TCBs are still attached to the stack, attempt to switch them
1321 * to the default stack.
1323 if (force && blk->tfb_refcnt) {
1324 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1325 INPLOOKUP_WLOCKPCB);
1328 VNET_ITERATOR_DECL(vnet_iter);
1330 rw_wunlock(&tcp_function_lock);
1333 VNET_FOREACH(vnet_iter) {
1334 CURVNET_SET(vnet_iter);
1335 while ((inp = inp_next(&inpi)) != NULL) {
1336 tp = intotcpcb(inp);
1337 if (tp == NULL || tp->t_fb != blk)
1339 tcp_switch_back_to_default(tp);
1343 VNET_LIST_RUNLOCK();
1345 rw_wlock(&tcp_function_lock);
1347 if (blk->tfb_refcnt) {
1348 /* TCBs still attached. */
1349 rw_wunlock(&tcp_function_lock);
1354 rw_wunlock(&tcp_function_lock);
1357 /* Remove any function names that map to this function block. */
1358 while (find_tcp_fb_locked(blk, &f) != NULL) {
1359 TAILQ_REMOVE(&t_functions, f, tf_next);
1362 free(f, M_TCPFUNCTIONS);
1364 rw_wunlock(&tcp_function_lock);
1371 struct epoch_tracker et;
1372 VNET_ITERATOR_DECL(vnet_iter);
1377 NET_EPOCH_ENTER(et);
1378 VNET_LIST_RLOCK_NOSLEEP();
1379 VNET_FOREACH(vnet_iter) {
1380 CURVNET_SET(vnet_iter);
1381 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1382 INPLOOKUP_WLOCKPCB);
1387 * Walk the tcpbs, if existing, and flush the reassembly queue,
1388 * if there is one...
1389 * XXX: The "Net/3" implementation doesn't imply that the TCP
1390 * reassembly queue should be flushed, but in a situation
1391 * where we're really low on mbufs, this is potentially
1394 while ((inpb = inp_next(&inpi)) != NULL) {
1395 if ((tcpb = intotcpcb(inpb)) != NULL) {
1396 tcp_reass_flush(tcpb);
1397 tcp_clean_sackreport(tcpb);
1399 tcp_log_drain(tcpb);
1402 if (tcp_pcap_aggressive_free) {
1403 /* Free the TCP PCAP queues. */
1404 tcp_pcap_drain(&(tcpb->t_inpkts));
1405 tcp_pcap_drain(&(tcpb->t_outpkts));
1412 VNET_LIST_RUNLOCK_NOSLEEP();
1417 tcp_vnet_init(void *arg __unused)
1421 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1422 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1423 printf("%s: WARNING: unable to register helper hook\n", __func__);
1424 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1425 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1426 printf("%s: WARNING: unable to register helper hook\n", __func__);
1429 if (tcp_stats_init())
1430 printf("%s: WARNING: unable to initialise TCP stats\n",
1433 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1439 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1440 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1441 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1443 tcp_fastopen_init();
1445 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1446 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1448 V_tcp_msl = TCPTV_MSL;
1450 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1451 tcp_vnet_init, NULL);
1454 tcp_init(void *arg __unused)
1456 const char *tcbhash_tuneable;
1459 tcp_reass_global_init();
1461 /* XXX virtualize those below? */
1462 tcp_delacktime = TCPTV_DELACK;
1463 tcp_keepinit = TCPTV_KEEP_INIT;
1464 tcp_keepidle = TCPTV_KEEP_IDLE;
1465 tcp_keepintvl = TCPTV_KEEPINTVL;
1466 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1467 tcp_rexmit_initial = TCPTV_RTOBASE;
1468 if (tcp_rexmit_initial < 1)
1469 tcp_rexmit_initial = 1;
1470 tcp_rexmit_min = TCPTV_MIN;
1471 if (tcp_rexmit_min < 1)
1473 tcp_persmin = TCPTV_PERSMIN;
1474 tcp_persmax = TCPTV_PERSMAX;
1475 tcp_rexmit_slop = TCPTV_CPU_VAR;
1476 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1478 /* Setup the tcp function block list */
1479 TAILQ_INIT(&t_functions);
1480 rw_init(&tcp_function_lock, "tcp_func_lock");
1481 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1483 /* Initialize the TCP logging data. */
1486 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1488 if (tcp_soreceive_stream) {
1490 tcp_protosw.pr_soreceive = soreceive_stream;
1493 tcp6_protosw.pr_soreceive = soreceive_stream;
1498 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1500 max_protohdr_grow(sizeof(struct tcpiphdr));
1504 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1505 SHUTDOWN_PRI_DEFAULT);
1506 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1507 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1509 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1510 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1511 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1512 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1513 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1514 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1515 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1516 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1517 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1522 hashsize = TCBHASHSIZE;
1523 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1524 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1525 if (hashsize == 0) {
1527 * Auto tune the hash size based on maxsockets.
1528 * A perfect hash would have a 1:1 mapping
1529 * (hashsize = maxsockets) however it's been
1530 * suggested that O(2) average is better.
1532 hashsize = maketcp_hashsize(maxsockets / 4);
1534 * Our historical default is 512,
1535 * do not autotune lower than this.
1540 printf("%s: %s auto tuned to %d\n", __func__,
1541 tcbhash_tuneable, hashsize);
1544 * We require a hashsize to be a power of two.
1545 * Previously if it was not a power of two we would just reset it
1546 * back to 512, which could be a nasty surprise if you did not notice
1547 * the error message.
1548 * Instead what we do is clip it to the closest power of two lower
1549 * than the specified hash value.
1551 if (!powerof2(hashsize)) {
1552 int oldhashsize = hashsize;
1554 hashsize = maketcp_hashsize(hashsize);
1555 /* prevent absurdly low value */
1558 printf("%s: WARNING: TCB hash size not a power of 2, "
1559 "clipped from %d to %d.\n", __func__, oldhashsize,
1562 tcp_tcbhashsize = hashsize;
1565 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1568 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1571 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1575 tcp_destroy(void *unused __unused)
1583 * All our processes are gone, all our sockets should be cleaned
1584 * up, which means, we should be past the tcp_discardcb() calls.
1585 * Sleep to let all tcpcb timers really disappear and cleanup.
1588 INP_INFO_WLOCK(&V_tcbinfo);
1589 n = V_tcbinfo.ipi_count;
1590 INP_INFO_WUNLOCK(&V_tcbinfo);
1593 pause("tcpdes", hz / 10);
1597 in_pcbinfo_destroy(&V_tcbinfo);
1598 /* tcp_discardcb() clears the sack_holes up. */
1599 uma_zdestroy(V_sack_hole_zone);
1602 * Cannot free the zone until all tcpcbs are released as we attach
1603 * the allocations to them.
1605 tcp_fastopen_destroy();
1607 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1608 VNET_PCPUSTAT_FREE(tcpstat);
1611 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1613 printf("%s: WARNING: unable to deregister helper hook "
1614 "type=%d, id=%d: error %d returned\n", __func__,
1615 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1617 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1619 printf("%s: WARNING: unable to deregister helper hook "
1620 "type=%d, id=%d: error %d returned\n", __func__,
1621 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1625 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1635 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1636 * tcp_template used to store this data in mbufs, but we now recopy it out
1637 * of the tcpcb each time to conserve mbufs.
1640 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1642 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1644 INP_WLOCK_ASSERT(inp);
1647 if ((inp->inp_vflag & INP_IPV6) != 0) {
1648 struct ip6_hdr *ip6;
1650 ip6 = (struct ip6_hdr *)ip_ptr;
1651 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1652 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1653 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1654 (IPV6_VERSION & IPV6_VERSION_MASK);
1656 ip6->ip6_nxt = IPPROTO_TCP;
1658 ip6->ip6_nxt = IPPROTO_UDP;
1659 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1660 ip6->ip6_src = inp->in6p_laddr;
1661 ip6->ip6_dst = inp->in6p_faddr;
1664 #if defined(INET6) && defined(INET)
1671 ip = (struct ip *)ip_ptr;
1672 ip->ip_v = IPVERSION;
1674 ip->ip_tos = inp->inp_ip_tos;
1678 ip->ip_ttl = inp->inp_ip_ttl;
1681 ip->ip_p = IPPROTO_TCP;
1683 ip->ip_p = IPPROTO_UDP;
1684 ip->ip_src = inp->inp_laddr;
1685 ip->ip_dst = inp->inp_faddr;
1688 th->th_sport = inp->inp_lport;
1689 th->th_dport = inp->inp_fport;
1693 tcp_set_flags(th, 0);
1696 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1700 * Create template to be used to send tcp packets on a connection.
1701 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1702 * use for this function is in keepalives, which use tcp_respond.
1705 tcpip_maketemplate(struct inpcb *inp)
1709 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1712 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1717 * Send a single message to the TCP at address specified by
1718 * the given TCP/IP header. If m == NULL, then we make a copy
1719 * of the tcpiphdr at th and send directly to the addressed host.
1720 * This is used to force keep alive messages out using the TCP
1721 * template for a connection. If flags are given then we send
1722 * a message back to the TCP which originated the segment th,
1723 * and discard the mbuf containing it and any other attached mbufs.
1725 * In any case the ack and sequence number of the transmitted
1726 * segment are as specified by the parameters.
1728 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1731 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1732 tcp_seq ack, tcp_seq seq, uint16_t flags)
1738 struct udphdr *uh = NULL;
1740 struct tcp_log_buffer *lgb;
1743 struct ip6_hdr *ip6;
1746 int optlen, tlen, win, ulen;
1752 int thflags = tcp_get_flags(th);
1755 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1759 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1765 inp = tptoinpcb(tp);
1766 INP_LOCK_ASSERT(inp);
1772 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1773 port = m->m_pkthdr.tcp_tun_port;
1776 if (ip && (ip->ip_p == IPPROTO_UDP))
1777 port = m->m_pkthdr.tcp_tun_port;
1786 if (!(flags & TH_RST)) {
1787 win = sbspace(&inp->inp_socket->so_rcv);
1788 if (win > TCP_MAXWIN << tp->rcv_scale)
1789 win = TCP_MAXWIN << tp->rcv_scale;
1791 if ((tp->t_flags & TF_NOOPT) == 0)
1795 m = m_gethdr(M_NOWAIT, MT_DATA);
1798 m->m_data += max_linkhdr;
1801 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1802 sizeof(struct ip6_hdr));
1803 ip6 = mtod(m, struct ip6_hdr *);
1804 nth = (struct tcphdr *)(ip6 + 1);
1806 /* Insert a UDP header */
1807 uh = (struct udphdr *)nth;
1808 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1809 uh->uh_dport = port;
1810 nth = (struct tcphdr *)(uh + 1);
1815 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1816 ip = mtod(m, struct ip *);
1817 nth = (struct tcphdr *)(ip + 1);
1819 /* Insert a UDP header */
1820 uh = (struct udphdr *)nth;
1821 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1822 uh->uh_dport = port;
1823 nth = (struct tcphdr *)(uh + 1);
1826 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1828 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1831 /* Can't reuse 'm', allocate a new mbuf. */
1832 n = m_gethdr(M_NOWAIT, MT_DATA);
1838 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1844 n->m_data += max_linkhdr;
1845 /* m_len is set later */
1846 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1849 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1850 sizeof(struct ip6_hdr));
1851 ip6 = mtod(n, struct ip6_hdr *);
1852 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1853 nth = (struct tcphdr *)(ip6 + 1);
1855 /* Insert a UDP header */
1856 uh = (struct udphdr *)nth;
1857 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1858 uh->uh_dport = port;
1859 nth = (struct tcphdr *)(uh + 1);
1864 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1865 ip = mtod(n, struct ip *);
1866 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1867 nth = (struct tcphdr *)(ip + 1);
1869 /* Insert a UDP header */
1870 uh = (struct udphdr *)nth;
1871 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1872 uh->uh_dport = port;
1873 nth = (struct tcphdr *)(uh + 1);
1876 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1877 xchg(nth->th_dport, nth->th_sport, uint16_t);
1884 * XXX MRT We inherit the FIB, which is lucky.
1888 m->m_data = (caddr_t)ipgen;
1889 /* m_len is set later */
1892 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1893 nth = (struct tcphdr *)(ip6 + 1);
1897 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1898 nth = (struct tcphdr *)(ip + 1);
1902 * this is usually a case when an extension header
1903 * exists between the IPv6 header and the
1906 nth->th_sport = th->th_sport;
1907 nth->th_dport = th->th_dport;
1909 xchg(nth->th_dport, nth->th_sport, uint16_t);
1915 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1917 #if defined(INET) && defined(INET6)
1921 tlen = sizeof (struct tcpiphdr);
1924 tlen += sizeof (struct udphdr);
1927 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1928 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1929 m, tlen, (long)M_TRAILINGSPACE(m)));
1934 ect = tcp_ecn_output_established(tp, &flags, 0, false);
1935 /* Make sure we have room. */
1936 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1937 m->m_next = m_get(M_NOWAIT, MT_DATA);
1939 optp = mtod(m->m_next, u_char *);
1944 optp = (u_char *) (nth + 1);
1950 if (tp->t_flags & TF_RCVD_TSTMP) {
1951 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1952 to.to_tsecr = tp->ts_recent;
1953 to.to_flags |= TOF_TS;
1955 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1956 /* TCP-MD5 (RFC2385). */
1957 if (tp->t_flags & TF_SIGNATURE)
1958 to.to_flags |= TOF_SIGNATURE;
1960 /* Add the options. */
1961 tlen += optlen = tcp_addoptions(&to, optp);
1963 /* Update m_len in the correct mbuf. */
1964 optm->m_len += optlen;
1970 ulen = tlen - sizeof(struct ip6_hdr);
1971 uh->uh_ulen = htons(ulen);
1973 ip6->ip6_flow = htonl(ect << 20);
1974 ip6->ip6_vfc = IPV6_VERSION;
1976 ip6->ip6_nxt = IPPROTO_UDP;
1978 ip6->ip6_nxt = IPPROTO_TCP;
1979 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1982 #if defined(INET) && defined(INET6)
1988 ulen = tlen - sizeof(struct ip);
1989 uh->uh_ulen = htons(ulen);
1992 ip->ip_len = htons(tlen);
1993 ip->ip_ttl = V_ip_defttl;
1995 ip->ip_p = IPPROTO_UDP;
1997 ip->ip_p = IPPROTO_TCP;
1999 if (V_path_mtu_discovery)
2000 ip->ip_off |= htons(IP_DF);
2003 m->m_pkthdr.len = tlen;
2004 m->m_pkthdr.rcvif = NULL;
2008 * Packet is associated with a socket, so allow the
2009 * label of the response to reflect the socket label.
2011 INP_LOCK_ASSERT(inp);
2012 mac_inpcb_create_mbuf(inp, m);
2015 * Packet is not associated with a socket, so possibly
2016 * update the label in place.
2018 mac_netinet_tcp_reply(m);
2021 nth->th_seq = htonl(seq);
2022 nth->th_ack = htonl(ack);
2023 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2024 tcp_set_flags(nth, flags);
2026 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2028 nth->th_win = htons((u_short)win);
2031 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2032 if (to.to_flags & TOF_SIGNATURE) {
2033 if (!TCPMD5_ENABLED() ||
2034 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2044 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2045 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2046 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2049 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2050 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2051 nth->th_sum = in6_cksum_pseudo(ip6,
2052 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2054 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2057 #if defined(INET6) && defined(INET)
2063 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2064 htons(ulen + IPPROTO_UDP));
2065 m->m_pkthdr.csum_flags = CSUM_UDP;
2066 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2069 m->m_pkthdr.csum_flags = CSUM_TCP;
2070 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2071 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2072 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2076 TCP_PROBE3(debug__output, tp, th, m);
2078 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2080 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2081 if (INP_WLOCKED(inp)) {
2082 union tcp_log_stackspecific log;
2085 memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2086 log.u_bbr.inhpts = inp->inp_in_hpts;
2087 log.u_bbr.flex8 = 4;
2088 log.u_bbr.pkts_out = tp->t_maxseg;
2089 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2090 log.u_bbr.delivered = 0;
2091 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2092 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2095 * We can not log the packet, since we only own the
2096 * read lock, but a write lock is needed. The read lock
2097 * is not upgraded to a write lock, since only getting
2098 * the read lock was done intentionally to improve the
2099 * handling of SYN flooding attacks.
2100 * This happens only for pure SYN segments received in
2101 * the initial CLOSED state, or received in a more
2102 * advanced state than listen and the UDP encapsulation
2103 * port is unexpected.
2104 * The incoming SYN segments do not really belong to
2105 * the TCP connection and the handling does not change
2106 * the state of the TCP connection. Therefore, the
2107 * sending of the RST segments is not logged. Please
2108 * note that also the incoming SYN segments are not
2111 * The following code ensures that the above description
2112 * is and stays correct.
2114 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2115 (tp->t_state == TCPS_CLOSED ||
2116 (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2117 ("%s: Logging of TCP segment with flags 0x%b and "
2118 "UDP encapsulation port %u skipped in state %s",
2119 __func__, thflags, PRINT_TH_FLAGS,
2120 ntohs(port), tcpstates[tp->t_state]));
2125 TCPSTAT_INC(tcps_sndacks);
2126 else if (flags & (TH_SYN|TH_FIN|TH_RST))
2127 TCPSTAT_INC(tcps_sndctrl);
2128 TCPSTAT_INC(tcps_sndtotal);
2132 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2133 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2136 #if defined(INET) && defined(INET6)
2141 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2142 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2146 lgb->tlb_errno = output_ret;
2150 * Create a new TCP control block, making an empty reassembly queue and hooking
2151 * it to the argument protocol control block. The `inp' parameter must have
2152 * come from the zone allocator set up by tcpcbstor declaration.
2155 tcp_newtcpcb(struct inpcb *inp)
2157 struct tcpcb *tp = intotcpcb(inp);
2159 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2163 * Historically allocation was done with M_ZERO. There is a lot of
2164 * code that rely on that. For now take safe approach and zero whole
2165 * tcpcb. This definitely can be optimized.
2167 bzero(&tp->t_start_zero, t_zero_size);
2169 /* Initialise cc_var struct for this tcpcb. */
2170 tp->t_ccv.type = IPPROTO_TCP;
2171 tp->t_ccv.ccvc.tcp = tp;
2172 rw_rlock(&tcp_function_lock);
2173 tp->t_fb = tcp_func_set_ptr;
2174 refcount_acquire(&tp->t_fb->tfb_refcnt);
2175 rw_runlock(&tcp_function_lock);
2177 * Use the current system default CC algorithm.
2179 cc_attach(tp, CC_DEFAULT_ALGO());
2181 if (CC_ALGO(tp)->cb_init != NULL)
2182 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2184 if (tp->t_fb->tfb_tcp_fb_fini)
2185 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2186 refcount_release(&tp->t_fb->tfb_refcnt);
2191 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2192 if (tp->t_fb->tfb_tcp_fb_fini)
2193 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2194 refcount_release(&tp->t_fb->tfb_refcnt);
2199 TAILQ_INIT(&tp->t_segq);
2202 isipv6 ? V_tcp_v6mssdflt :
2206 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2207 for (int i = 0; i < TT_N; i++)
2208 tp->t_timers[i] = SBT_MAX;
2210 switch (V_tcp_do_rfc1323) {
2215 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2218 tp->t_flags = TF_REQ_SCALE;
2221 tp->t_flags = TF_REQ_TSTMP;
2225 tp->t_flags |= TF_SACK_PERMIT;
2226 TAILQ_INIT(&tp->snd_holes);
2229 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2230 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2231 * reasonable initial retransmit time.
2233 tp->t_srtt = TCPTV_SRTTBASE;
2234 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2235 tp->t_rttmin = tcp_rexmit_min;
2236 tp->t_rxtcur = tcp_rexmit_initial;
2237 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2238 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2239 tp->t_rcvtime = ticks;
2241 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2242 * because the socket may be bound to an IPv6 wildcard address,
2243 * which may match an IPv4-mapped IPv6 address.
2245 inp->inp_ip_ttl = V_ip_defttl;
2248 * If using hpts lets drop a random number in so
2249 * not all new connections fall on the same CPU.
2251 inp->inp_hpts_cpu = hpts_random_cpu(inp);
2255 * Init the TCP PCAP queues.
2257 tcp_pcap_tcpcb_init(tp);
2260 /* Initialize the per-TCPCB log data. */
2261 tcp_log_tcpcbinit(tp);
2263 tp->t_pacing_rate = -1;
2264 if (tp->t_fb->tfb_tcp_fb_init) {
2265 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2266 refcount_release(&tp->t_fb->tfb_refcnt);
2271 if (V_tcp_perconn_stats_enable == 1)
2272 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2275 tp->t_flags |= TF_LRD;
2281 * Drop a TCP connection, reporting
2282 * the specified error. If connection is synchronized,
2283 * then send a RST to peer.
2286 tcp_drop(struct tcpcb *tp, int errno)
2288 struct socket *so = tptosocket(tp);
2291 INP_WLOCK_ASSERT(tptoinpcb(tp));
2293 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2294 tcp_state_change(tp, TCPS_CLOSED);
2295 /* Don't use tcp_output() here due to possible recursion. */
2296 (void)tcp_output_nodrop(tp);
2297 TCPSTAT_INC(tcps_drops);
2299 TCPSTAT_INC(tcps_conndrops);
2300 if (errno == ETIMEDOUT && tp->t_softerror)
2301 errno = tp->t_softerror;
2302 so->so_error = errno;
2303 return (tcp_close(tp));
2307 tcp_discardcb(struct tcpcb *tp)
2309 struct inpcb *inp = tptoinpcb(tp);
2310 struct socket *so = tptosocket(tp);
2312 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2315 INP_WLOCK_ASSERT(inp);
2318 if (tp->t_fb->tfb_tcp_timer_stop_all) {
2319 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2322 /* free the reassembly queue, if any */
2323 tcp_reass_flush(tp);
2326 /* Disconnect offload device, if any. */
2327 if (tp->t_flags & TF_TOE)
2328 tcp_offload_detach(tp);
2331 tcp_free_sackholes(tp);
2334 /* Free the TCP PCAP queues. */
2335 tcp_pcap_drain(&(tp->t_inpkts));
2336 tcp_pcap_drain(&(tp->t_outpkts));
2339 /* Allow the CC algorithm to clean up after itself. */
2340 if (CC_ALGO(tp)->cb_destroy != NULL)
2341 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2343 /* Detach from the CC algorithm */
2347 khelp_destroy_osd(&tp->t_osd);
2350 stats_blob_destroy(tp->t_stats);
2356 tcp_log_tcpcbfini(tp);
2358 TCPSTATES_DEC(tp->t_state);
2359 if (tp->t_fb->tfb_tcp_fb_fini)
2360 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2363 * If we got enough samples through the srtt filter,
2364 * save the rtt and rttvar in the routing entry.
2365 * 'Enough' is arbitrarily defined as 4 rtt samples.
2366 * 4 samples is enough for the srtt filter to converge
2367 * to within enough % of the correct value; fewer samples
2368 * and we could save a bogus rtt. The danger is not high
2369 * as tcp quickly recovers from everything.
2370 * XXX: Works very well but needs some more statistics!
2372 * XXXRRS: Updating must be after the stack fini() since
2373 * that may be converting some internal representation of
2374 * say srtt etc into the general one used by other stacks.
2375 * Lets also at least protect against the so being NULL
2376 * as RW stated below.
2378 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2379 struct hc_metrics_lite metrics;
2382 bzero(&metrics, sizeof(metrics));
2384 * Update the ssthresh always when the conditions below
2385 * are satisfied. This gives us better new start value
2386 * for the congestion avoidance for new connections.
2387 * ssthresh is only set if packet loss occurred on a session.
2389 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2390 * being torn down. Ideally this code would not use 'so'.
2392 ssthresh = tp->snd_ssthresh;
2393 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2395 * convert the limit from user data bytes to
2396 * packets then to packet data bytes.
2398 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2401 ssthresh *= (tp->t_maxseg +
2403 (isipv6 ? sizeof (struct ip6_hdr) +
2404 sizeof (struct tcphdr) :
2406 sizeof (struct tcpiphdr)
2413 metrics.rmx_ssthresh = ssthresh;
2415 metrics.rmx_rtt = tp->t_srtt;
2416 metrics.rmx_rttvar = tp->t_rttvar;
2417 metrics.rmx_cwnd = tp->snd_cwnd;
2418 metrics.rmx_sendpipe = 0;
2419 metrics.rmx_recvpipe = 0;
2421 tcp_hc_update(&inp->inp_inc, &metrics);
2424 refcount_release(&tp->t_fb->tfb_refcnt);
2428 * Attempt to close a TCP control block, marking it as dropped, and freeing
2429 * the socket if we hold the only reference.
2432 tcp_close(struct tcpcb *tp)
2434 struct inpcb *inp = tptoinpcb(tp);
2435 struct socket *so = tptosocket(tp);
2437 INP_WLOCK_ASSERT(inp);
2440 if (tp->t_state == TCPS_LISTEN)
2441 tcp_offload_listen_stop(tp);
2444 * This releases the TFO pending counter resource for TFO listen
2445 * sockets as well as passively-created TFO sockets that transition
2446 * from SYN_RECEIVED to CLOSED.
2448 if (tp->t_tfo_pending) {
2449 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2450 tp->t_tfo_pending = NULL;
2453 tcp_hpts_remove(inp);
2456 TCPSTAT_INC(tcps_closed);
2457 if (tp->t_state != TCPS_CLOSED)
2458 tcp_state_change(tp, TCPS_CLOSED);
2459 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2460 soisdisconnected(so);
2461 if (inp->inp_flags & INP_SOCKREF) {
2462 inp->inp_flags &= ~INP_SOCKREF;
2471 * Notify a tcp user of an asynchronous error;
2472 * store error as soft error, but wake up user
2473 * (for now, won't do anything until can select for soft error).
2475 * Do not wake up user since there currently is no mechanism for
2476 * reporting soft errors (yet - a kqueue filter may be added).
2478 static struct inpcb *
2479 tcp_notify(struct inpcb *inp, int error)
2483 INP_WLOCK_ASSERT(inp);
2485 tp = intotcpcb(inp);
2486 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2489 * Ignore some errors if we are hooked up.
2490 * If connection hasn't completed, has retransmitted several times,
2491 * and receives a second error, give up now. This is better
2492 * than waiting a long time to establish a connection that
2493 * can never complete.
2495 if (tp->t_state == TCPS_ESTABLISHED &&
2496 (error == EHOSTUNREACH || error == ENETUNREACH ||
2497 error == EHOSTDOWN)) {
2498 if (inp->inp_route.ro_nh) {
2499 NH_FREE(inp->inp_route.ro_nh);
2500 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2503 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2505 tp = tcp_drop(tp, error);
2511 tp->t_softerror = error;
2515 wakeup( &so->so_timeo);
2522 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2524 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2525 INPLOOKUP_RLOCKPCB);
2530 if (req->newptr != NULL)
2533 if (req->oldptr == NULL) {
2536 n = V_tcbinfo.ipi_count +
2537 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2538 n += imax(n / 8, 10);
2539 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2543 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2546 bzero(&xig, sizeof(xig));
2547 xig.xig_len = sizeof xig;
2548 xig.xig_count = V_tcbinfo.ipi_count +
2549 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2550 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2551 xig.xig_sogen = so_gencnt;
2552 error = SYSCTL_OUT(req, &xig, sizeof xig);
2556 error = syncache_pcblist(req);
2560 while ((inp = inp_next(&inpi)) != NULL) {
2561 if (inp->inp_gencnt <= xig.xig_gen &&
2562 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2565 tcp_inptoxtp(inp, &xt);
2566 error = SYSCTL_OUT(req, &xt, sizeof xt);
2577 * Give the user an updated idea of our state.
2578 * If the generation differs from what we told
2579 * her before, she knows that something happened
2580 * while we were processing this request, and it
2581 * might be necessary to retry.
2583 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2584 xig.xig_sogen = so_gencnt;
2585 xig.xig_count = V_tcbinfo.ipi_count +
2586 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2587 error = SYSCTL_OUT(req, &xig, sizeof xig);
2593 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2594 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2595 NULL, 0, tcp_pcblist, "S,xtcpcb",
2596 "List of active TCP connections");
2600 tcp_getcred(SYSCTL_HANDLER_ARGS)
2603 struct sockaddr_in addrs[2];
2604 struct epoch_tracker et;
2608 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2611 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2614 NET_EPOCH_ENTER(et);
2615 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2616 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2620 error = cr_canseeinpcb(req->td->td_ucred, inp);
2622 cru2x(inp->inp_cred, &xuc);
2627 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2631 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2632 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2633 0, 0, tcp_getcred, "S,xucred",
2634 "Get the xucred of a TCP connection");
2639 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2641 struct epoch_tracker et;
2643 struct sockaddr_in6 addrs[2];
2650 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2653 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2656 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2657 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2660 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2662 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2669 NET_EPOCH_ENTER(et);
2672 inp = in_pcblookup(&V_tcbinfo,
2673 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2675 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2676 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2679 inp = in6_pcblookup(&V_tcbinfo,
2680 &addrs[1].sin6_addr, addrs[1].sin6_port,
2681 &addrs[0].sin6_addr, addrs[0].sin6_port,
2682 INPLOOKUP_RLOCKPCB, NULL);
2686 error = cr_canseeinpcb(req->td->td_ucred, inp);
2688 cru2x(inp->inp_cred, &xuc);
2693 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2697 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2698 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2699 0, 0, tcp6_getcred, "S,xucred",
2700 "Get the xucred of a TCP6 connection");
2704 /* Path MTU to try next when a fragmentation-needed message is received. */
2706 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2708 int mtu = ntohs(icp->icmp_nextmtu);
2710 /* If no alternative MTU was proposed, try the next smaller one. */
2712 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2713 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2714 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2720 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2726 struct inpcb *(*notify)(struct inpcb *, int);
2727 struct in_conninfo inc;
2728 tcp_seq icmp_tcp_seq;
2731 errno = icmp_errmap(icp);
2736 notify = tcp_mtudisc_notify;
2740 notify = tcp_drop_syn_sent;
2742 notify = tcp_notify;
2745 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2746 notify = tcp_drop_syn_sent;
2748 notify = tcp_notify;
2751 notify = tcp_notify;
2755 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2756 icmp_tcp_seq = th->th_seq;
2757 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2758 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2760 tp = intotcpcb(inp);
2762 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2764 * MTU discovery for offloaded connections. Let
2765 * the TOE driver verify seq# and process it.
2767 mtu = tcp_next_pmtu(icp, ip);
2768 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2772 if (tp->t_port != port)
2774 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2775 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2776 if (errno == EMSGSIZE) {
2778 * MTU discovery: we got a needfrag and
2779 * will potentially try a lower MTU.
2781 mtu = tcp_next_pmtu(icp, ip);
2784 * Only process the offered MTU if it
2785 * is smaller than the current one.
2787 if (mtu < tp->t_maxseg +
2788 sizeof(struct tcpiphdr)) {
2789 bzero(&inc, sizeof(inc));
2790 inc.inc_faddr = ip->ip_dst;
2792 inp->inp_inc.inc_fibnum;
2793 tcp_hc_updatemtu(&inc, mtu);
2794 inp = tcp_mtudisc(inp, mtu);
2797 inp = (*notify)(inp, errno);
2800 bzero(&inc, sizeof(inc));
2801 inc.inc_fport = th->th_dport;
2802 inc.inc_lport = th->th_sport;
2803 inc.inc_faddr = ip->ip_dst;
2804 inc.inc_laddr = ip->ip_src;
2805 syncache_unreach(&inc, icmp_tcp_seq, port);
2813 tcp_ctlinput(struct icmp *icmp)
2815 tcp_ctlinput_with_port(icmp, htons(0));
2819 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2821 /* Its a tunneled TCP over UDP icmp */
2822 struct icmp *icmp = param.icmp;
2823 struct ip *outer_ip, *inner_ip;
2825 struct tcphdr *th, ttemp;
2829 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2830 inner_ip = &icmp->icmp_ip;
2831 i_hlen = inner_ip->ip_hl << 2;
2832 o_len = ntohs(outer_ip->ip_len);
2834 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2835 /* Not enough data present */
2838 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2839 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2840 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2843 port = udp->uh_dport;
2844 th = (struct tcphdr *)(udp + 1);
2845 memcpy(&ttemp, th, sizeof(struct tcphdr));
2846 memcpy(udp, &ttemp, sizeof(struct tcphdr));
2847 /* Now adjust down the size of the outer IP header */
2848 o_len -= sizeof(struct udphdr);
2849 outer_ip->ip_len = htons(o_len);
2850 /* Now call in to the normal handling code */
2851 tcp_ctlinput_with_port(icmp, port);
2857 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2859 int mtu = ntohl(icmp6->icmp6_mtu);
2862 * If no alternative MTU was proposed, or the proposed MTU was too
2863 * small, set to the min.
2865 if (mtu < IPV6_MMTU)
2866 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */
2871 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2873 struct in6_addr *dst;
2874 struct inpcb *(*notify)(struct inpcb *, int);
2875 struct ip6_hdr *ip6;
2879 struct icmp6_hdr *icmp6;
2880 struct in_conninfo inc;
2885 tcp_seq icmp_tcp_seq;
2890 icmp6 = ip6cp->ip6c_icmp6;
2892 ip6 = ip6cp->ip6c_ip6;
2893 off = ip6cp->ip6c_off;
2894 dst = &ip6cp->ip6c_finaldst->sin6_addr;
2896 errno = icmp6_errmap(icmp6);
2901 notify = tcp_mtudisc_notify;
2905 notify = tcp_drop_syn_sent;
2907 notify = tcp_notify;
2911 * There are only four ICMPs that may reset connection:
2912 * - administratively prohibited
2913 * - port unreachable
2914 * - time exceeded in transit
2915 * - unknown next header
2917 if (V_icmp_may_rst &&
2918 ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
2919 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
2920 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
2921 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
2922 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
2923 (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
2924 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
2925 notify = tcp_drop_syn_sent;
2927 notify = tcp_notify;
2930 notify = tcp_notify;
2933 /* Check if we can safely get the ports from the tcp hdr */
2936 (int32_t) (off + sizeof(struct tcp_ports)))) {
2939 bzero(&t_ports, sizeof(struct tcp_ports));
2940 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2941 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2942 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2943 off += sizeof(struct tcp_ports);
2944 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2947 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2949 tp = intotcpcb(inp);
2951 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2952 /* MTU discovery for offloaded connections. */
2953 mtu = tcp6_next_pmtu(icmp6);
2954 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2958 if (tp->t_port != port)
2960 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2961 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2962 if (errno == EMSGSIZE) {
2965 * If we got a needfrag set the MTU
2966 * in the route to the suggested new
2967 * value (if given) and then notify.
2969 mtu = tcp6_next_pmtu(icmp6);
2971 bzero(&inc, sizeof(inc));
2972 inc.inc_fibnum = M_GETFIB(m);
2973 inc.inc_flags |= INC_ISIPV6;
2974 inc.inc6_faddr = *dst;
2975 if (in6_setscope(&inc.inc6_faddr,
2976 m->m_pkthdr.rcvif, NULL))
2979 * Only process the offered MTU if it
2980 * is smaller than the current one.
2982 if (mtu < tp->t_maxseg +
2983 sizeof (struct tcphdr) +
2984 sizeof (struct ip6_hdr)) {
2985 tcp_hc_updatemtu(&inc, mtu);
2986 tcp_mtudisc(inp, mtu);
2987 ICMP6STAT_INC(icp6s_pmtuchg);
2990 inp = (*notify)(inp, errno);
2993 bzero(&inc, sizeof(inc));
2994 inc.inc_fibnum = M_GETFIB(m);
2995 inc.inc_flags |= INC_ISIPV6;
2996 inc.inc_fport = t_ports.th_dport;
2997 inc.inc_lport = t_ports.th_sport;
2998 inc.inc6_faddr = *dst;
2999 inc.inc6_laddr = ip6->ip6_src;
3000 syncache_unreach(&inc, icmp_tcp_seq, port);
3008 tcp6_ctlinput(struct ip6ctlparam *ctl)
3010 tcp6_ctlinput_with_port(ctl, htons(0));
3014 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3016 struct ip6ctlparam *ip6cp = param.ip6cp;
3021 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3025 udp = mtod(m, struct udphdr *);
3026 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3029 port = udp->uh_dport;
3030 m_adj(m, sizeof(struct udphdr));
3031 if ((m->m_flags & M_PKTHDR) == 0) {
3032 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3034 /* Now call in to the normal handling code */
3035 tcp6_ctlinput_with_port(ip6cp, port);
3041 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3046 KASSERT(len >= SIPHASH_KEY_LENGTH,
3047 ("%s: keylen %u too short ", __func__, len));
3048 SipHash24_Init(&ctx);
3049 SipHash_SetKey(&ctx, (uint8_t *)key);
3050 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3051 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3052 switch (inc->inc_flags & INC_ISIPV6) {
3055 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3056 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3061 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3062 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3066 SipHash_Final((uint8_t *)hash, &ctx);
3068 return (hash[0] ^ hash[1]);
3072 tcp_new_ts_offset(struct in_conninfo *inc)
3074 struct in_conninfo inc_store, *local_inc;
3076 if (!V_tcp_ts_offset_per_conn) {
3077 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3078 inc_store.inc_lport = 0;
3079 inc_store.inc_fport = 0;
3080 local_inc = &inc_store;
3084 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3085 sizeof(V_ts_offset_secret)));
3089 * Following is where TCP initial sequence number generation occurs.
3091 * There are two places where we must use initial sequence numbers:
3092 * 1. In SYN-ACK packets.
3093 * 2. In SYN packets.
3095 * All ISNs for SYN-ACK packets are generated by the syncache. See
3096 * tcp_syncache.c for details.
3098 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3099 * depends on this property. In addition, these ISNs should be
3100 * unguessable so as to prevent connection hijacking. To satisfy
3101 * the requirements of this situation, the algorithm outlined in
3102 * RFC 1948 is used, with only small modifications.
3104 * Implementation details:
3106 * Time is based off the system timer, and is corrected so that it
3107 * increases by one megabyte per second. This allows for proper
3108 * recycling on high speed LANs while still leaving over an hour
3111 * As reading the *exact* system time is too expensive to be done
3112 * whenever setting up a TCP connection, we increment the time
3113 * offset in two ways. First, a small random positive increment
3114 * is added to isn_offset for each connection that is set up.
3115 * Second, the function tcp_isn_tick fires once per clock tick
3116 * and increments isn_offset as necessary so that sequence numbers
3117 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3118 * random positive increments serve only to ensure that the same
3119 * exact sequence number is never sent out twice (as could otherwise
3120 * happen when a port is recycled in less than the system tick
3123 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3124 * between seeding of isn_secret. This is normally set to zero,
3125 * as reseeding should not be necessary.
3127 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3128 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3129 * general, this means holding an exclusive (write) lock.
3132 #define ISN_BYTES_PER_SECOND 1048576
3133 #define ISN_STATIC_INCREMENT 4096
3134 #define ISN_RANDOM_INCREMENT (4096 - 1)
3135 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3137 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3138 VNET_DEFINE_STATIC(int, isn_last);
3139 VNET_DEFINE_STATIC(int, isn_last_reseed);
3140 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3141 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3143 #define V_isn_secret VNET(isn_secret)
3144 #define V_isn_last VNET(isn_last)
3145 #define V_isn_last_reseed VNET(isn_last_reseed)
3146 #define V_isn_offset VNET(isn_offset)
3147 #define V_isn_offset_old VNET(isn_offset_old)
3150 tcp_new_isn(struct in_conninfo *inc)
3153 u_int32_t projected_offset;
3156 /* Seed if this is the first use, reseed if requested. */
3157 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3158 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3160 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3161 V_isn_last_reseed = ticks;
3164 /* Compute the hash and return the ISN. */
3165 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3166 sizeof(V_isn_secret));
3167 V_isn_offset += ISN_STATIC_INCREMENT +
3168 (arc4random() & ISN_RANDOM_INCREMENT);
3169 if (ticks != V_isn_last) {
3170 projected_offset = V_isn_offset_old +
3171 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3172 if (SEQ_GT(projected_offset, V_isn_offset))
3173 V_isn_offset = projected_offset;
3174 V_isn_offset_old = V_isn_offset;
3177 new_isn += V_isn_offset;
3183 * When a specific ICMP unreachable message is received and the
3184 * connection state is SYN-SENT, drop the connection. This behavior
3185 * is controlled by the icmp_may_rst sysctl.
3187 static struct inpcb *
3188 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3193 INP_WLOCK_ASSERT(inp);
3195 tp = intotcpcb(inp);
3196 if (tp->t_state != TCPS_SYN_SENT)
3199 if (IS_FASTOPEN(tp->t_flags))
3200 tcp_fastopen_disable_path(tp);
3202 tp = tcp_drop(tp, errno);
3210 * When `need fragmentation' ICMP is received, update our idea of the MSS
3211 * based on the new value. Also nudge TCP to send something, since we
3212 * know the packet we just sent was dropped.
3213 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3215 static struct inpcb *
3216 tcp_mtudisc_notify(struct inpcb *inp, int error)
3219 return (tcp_mtudisc(inp, -1));
3222 static struct inpcb *
3223 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3228 INP_WLOCK_ASSERT(inp);
3230 tp = intotcpcb(inp);
3231 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3233 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3235 so = inp->inp_socket;
3236 SOCKBUF_LOCK(&so->so_snd);
3237 /* If the mss is larger than the socket buffer, decrease the mss. */
3238 if (so->so_snd.sb_hiwat < tp->t_maxseg)
3239 tp->t_maxseg = so->so_snd.sb_hiwat;
3240 SOCKBUF_UNLOCK(&so->so_snd);
3242 TCPSTAT_INC(tcps_mturesent);
3244 tp->snd_nxt = tp->snd_una;
3245 tcp_free_sackholes(tp);
3246 tp->snd_recover = tp->snd_max;
3247 if (tp->t_flags & TF_SACK_PERMIT)
3248 EXIT_FASTRECOVERY(tp->t_flags);
3249 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3251 * Conceptually the snd_nxt setting
3252 * and freeing sack holes should
3253 * be done by the default stacks
3254 * own tfb_tcp_mtu_chg().
3256 tp->t_fb->tfb_tcp_mtu_chg(tp);
3258 if (tcp_output(tp) < 0)
3266 * Look-up the routing entry to the peer of this inpcb. If no route
3267 * is found and it cannot be allocated, then return 0. This routine
3268 * is called by TCP routines that access the rmx structure and by
3269 * tcp_mss_update to get the peer/interface MTU.
3272 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3274 struct nhop_object *nh;
3276 uint32_t maxmtu = 0;
3278 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3280 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3281 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3286 maxmtu = nh->nh_mtu;
3288 /* Report additional interface capabilities. */
3290 if (ifp->if_capenable & IFCAP_TSO4 &&
3291 ifp->if_hwassist & CSUM_TSO) {
3292 cap->ifcap |= CSUM_TSO;
3293 cap->tsomax = ifp->if_hw_tsomax;
3294 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3295 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3305 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3307 struct nhop_object *nh;
3308 struct in6_addr dst6;
3311 uint32_t maxmtu = 0;
3313 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3315 if (inc->inc_flags & INC_IPV6MINMTU)
3318 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3319 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3320 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3325 maxmtu = nh->nh_mtu;
3327 /* Report additional interface capabilities. */
3329 if (ifp->if_capenable & IFCAP_TSO6 &&
3330 ifp->if_hwassist & CSUM_TSO) {
3331 cap->ifcap |= CSUM_TSO;
3332 cap->tsomax = ifp->if_hw_tsomax;
3333 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3334 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3343 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3345 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3346 * The right place to do that is ip6_setpktopt() that has just been
3347 * executed. By the way it just filled ip6po_minmtu for us.
3350 tcp6_use_min_mtu(struct tcpcb *tp)
3352 struct inpcb *inp = tptoinpcb(tp);
3354 INP_WLOCK_ASSERT(inp);
3356 * In case of the IPV6_USE_MIN_MTU socket
3357 * option, the INC_IPV6MINMTU flag to announce
3358 * a corresponding MSS during the initial
3359 * handshake. If the TCP connection is not in
3360 * the front states, just reduce the MSS being
3361 * used. This avoids the sending of TCP
3362 * segments which will be fragmented at the
3365 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3366 if ((tp->t_state >= TCPS_SYN_SENT) &&
3367 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3368 struct ip6_pktopts *opt;
3370 opt = inp->in6p_outputopts;
3371 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3372 tp->t_maxseg > TCP6_MSS)
3373 tp->t_maxseg = TCP6_MSS;
3379 * Calculate effective SMSS per RFC5681 definition for a given TCP
3380 * connection at its current state, taking into account SACK and etc.
3383 tcp_maxseg(const struct tcpcb *tp)
3387 if (tp->t_flags & TF_NOOPT)
3388 return (tp->t_maxseg);
3391 * Here we have a simplified code from tcp_addoptions(),
3392 * without a proper loop, and having most of paddings hardcoded.
3393 * We might make mistakes with padding here in some edge cases,
3394 * but this is harmless, since result of tcp_maxseg() is used
3395 * only in cwnd and ssthresh estimations.
3397 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3398 if (tp->t_flags & TF_RCVD_TSTMP)
3399 optlen = TCPOLEN_TSTAMP_APPA;
3402 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3403 if (tp->t_flags & TF_SIGNATURE)
3404 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3406 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3407 optlen += TCPOLEN_SACKHDR;
3408 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3409 optlen = PADTCPOLEN(optlen);
3412 if (tp->t_flags & TF_REQ_TSTMP)
3413 optlen = TCPOLEN_TSTAMP_APPA;
3415 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3416 if (tp->t_flags & TF_REQ_SCALE)
3417 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3418 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3419 if (tp->t_flags & TF_SIGNATURE)
3420 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3422 if (tp->t_flags & TF_SACK_PERMIT)
3423 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3426 optlen = min(optlen, TCP_MAXOLEN);
3427 return (tp->t_maxseg - optlen);
3432 tcp_fixed_maxseg(const struct tcpcb *tp)
3436 if (tp->t_flags & TF_NOOPT)
3437 return (tp->t_maxseg);
3440 * Here we have a simplified code from tcp_addoptions(),
3441 * without a proper loop, and having most of paddings hardcoded.
3442 * We only consider fixed options that we would send every
3443 * time I.e. SACK is not considered. This is important
3444 * for cc modules to figure out what the modulo of the
3447 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3448 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3449 if (tp->t_flags & TF_RCVD_TSTMP)
3450 optlen = TCPOLEN_TSTAMP_APPA;
3453 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3454 if (tp->t_flags & TF_SIGNATURE)
3455 optlen += PAD(TCPOLEN_SIGNATURE);
3458 if (tp->t_flags & TF_REQ_TSTMP)
3459 optlen = TCPOLEN_TSTAMP_APPA;
3461 optlen = PAD(TCPOLEN_MAXSEG);
3462 if (tp->t_flags & TF_REQ_SCALE)
3463 optlen += PAD(TCPOLEN_WINDOW);
3464 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3465 if (tp->t_flags & TF_SIGNATURE)
3466 optlen += PAD(TCPOLEN_SIGNATURE);
3468 if (tp->t_flags & TF_SACK_PERMIT)
3469 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3472 optlen = min(optlen, TCP_MAXOLEN);
3473 return (tp->t_maxseg - optlen);
3479 sysctl_drop(SYSCTL_HANDLER_ARGS)
3481 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3482 struct sockaddr_storage addrs[2];
3486 struct sockaddr_in *fin = NULL, *lin = NULL;
3488 struct epoch_tracker et;
3490 struct sockaddr_in6 *fin6, *lin6;
3500 if (req->oldptr != NULL || req->oldlen != 0)
3502 if (req->newptr == NULL)
3504 if (req->newlen < sizeof(addrs))
3506 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3510 switch (addrs[0].ss_family) {
3513 fin6 = (struct sockaddr_in6 *)&addrs[0];
3514 lin6 = (struct sockaddr_in6 *)&addrs[1];
3515 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3516 lin6->sin6_len != sizeof(struct sockaddr_in6))
3518 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3519 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3521 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3522 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3524 fin = (struct sockaddr_in *)&addrs[0];
3525 lin = (struct sockaddr_in *)&addrs[1];
3529 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3532 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3539 fin = (struct sockaddr_in *)&addrs[0];
3540 lin = (struct sockaddr_in *)&addrs[1];
3541 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3542 lin->sin_len != sizeof(struct sockaddr_in))
3549 NET_EPOCH_ENTER(et);
3550 switch (addrs[0].ss_family) {
3553 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3554 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3555 INPLOOKUP_WLOCKPCB, NULL);
3560 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3561 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3566 if (!SOLISTENING(inp->inp_socket)) {
3567 tp = intotcpcb(inp);
3568 tp = tcp_drop(tp, ECONNABORTED);
3579 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3580 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3581 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3582 "Drop TCP connection");
3585 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3587 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3588 &tcp_ctloutput_set));
3591 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3592 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3593 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3594 "Set socket option for TCP endpoint");
3598 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3600 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3601 struct sockaddr_storage addrs[2];
3604 struct sockaddr_in *fin = NULL, *lin = NULL;
3606 struct epoch_tracker et;
3608 struct sockaddr_in6 *fin6, *lin6;
3618 if (req->oldptr != NULL || req->oldlen != 0)
3620 if (req->newptr == NULL)
3622 if (req->newlen < sizeof(addrs))
3624 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3628 switch (addrs[0].ss_family) {
3631 fin6 = (struct sockaddr_in6 *)&addrs[0];
3632 lin6 = (struct sockaddr_in6 *)&addrs[1];
3633 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3634 lin6->sin6_len != sizeof(struct sockaddr_in6))
3636 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3637 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3639 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3640 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3642 fin = (struct sockaddr_in *)&addrs[0];
3643 lin = (struct sockaddr_in *)&addrs[1];
3647 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3650 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3657 fin = (struct sockaddr_in *)&addrs[0];
3658 lin = (struct sockaddr_in *)&addrs[1];
3659 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3660 lin->sin_len != sizeof(struct sockaddr_in))
3667 NET_EPOCH_ENTER(et);
3668 switch (addrs[0].ss_family) {
3671 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3672 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3673 INPLOOKUP_WLOCKPCB, NULL);
3678 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3679 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3687 so = inp->inp_socket;
3689 error = ktls_set_tx_mode(so,
3690 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3698 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3699 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3700 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3701 "Switch TCP connection to SW TLS");
3702 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3703 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3704 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3705 "Switch TCP connection to ifnet TLS");
3709 * Generate a standardized TCP log line for use throughout the
3710 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3711 * allow use in the interrupt context.
3713 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3714 * NB: The function may return NULL if memory allocation failed.
3716 * Due to header inclusion and ordering limitations the struct ip
3717 * and ip6_hdr pointers have to be passed as void pointers.
3720 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3724 /* Is logging enabled? */
3725 if (V_tcp_log_in_vain == 0)
3728 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3732 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3736 /* Is logging enabled? */
3737 if (tcp_log_debug == 0)
3740 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3744 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3750 const struct ip *ip = (const struct ip *)ip4hdr;
3753 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3757 * The log line looks like this:
3758 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3760 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3761 sizeof(PRINT_TH_FLAGS) + 1 +
3763 2 * INET6_ADDRSTRLEN;
3765 2 * INET_ADDRSTRLEN;
3768 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3772 strcat(s, "TCP: [");
3775 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3776 inet_ntoa_r(inc->inc_faddr, sp);
3778 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3780 inet_ntoa_r(inc->inc_laddr, sp);
3782 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3785 ip6_sprintf(sp, &inc->inc6_faddr);
3787 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3789 ip6_sprintf(sp, &inc->inc6_laddr);
3791 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3792 } else if (ip6 && th) {
3793 ip6_sprintf(sp, &ip6->ip6_src);
3795 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3797 ip6_sprintf(sp, &ip6->ip6_dst);
3799 sprintf(sp, "]:%i", ntohs(th->th_dport));
3802 } else if (ip && th) {
3803 inet_ntoa_r(ip->ip_src, sp);
3805 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3807 inet_ntoa_r(ip->ip_dst, sp);
3809 sprintf(sp, "]:%i", ntohs(th->th_dport));
3817 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3818 if (*(s + size - 1) != '\0')
3819 panic("%s: string too long", __func__);
3824 * A subroutine which makes it easy to track TCP state changes with DTrace.
3825 * This function shouldn't be called for t_state initializations that don't
3826 * correspond to actual TCP state transitions.
3829 tcp_state_change(struct tcpcb *tp, int newstate)
3831 #if defined(KDTRACE_HOOKS)
3832 int pstate = tp->t_state;
3835 TCPSTATES_DEC(tp->t_state);
3836 TCPSTATES_INC(newstate);
3837 tp->t_state = newstate;
3838 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3842 * Create an external-format (``xtcpcb'') structure using the information in
3843 * the kernel-format tcpcb structure pointed to by tp. This is done to
3844 * reduce the spew of irrelevant information over this interface, to isolate
3845 * user code from changes in the kernel structure, and potentially to provide
3846 * information-hiding if we decide that some of this information should be
3847 * hidden from users.
3850 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3852 struct tcpcb *tp = intotcpcb(inp);
3855 bzero(xt, sizeof(*xt));
3856 xt->t_state = tp->t_state;
3857 xt->t_logstate = tp->t_logstate;
3858 xt->t_flags = tp->t_flags;
3859 xt->t_sndzerowin = tp->t_sndzerowin;
3860 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3861 xt->t_rcvoopack = tp->t_rcvoopack;
3862 xt->t_rcv_wnd = tp->rcv_wnd;
3863 xt->t_snd_wnd = tp->snd_wnd;
3864 xt->t_snd_cwnd = tp->snd_cwnd;
3865 xt->t_snd_ssthresh = tp->snd_ssthresh;
3866 xt->t_dsack_bytes = tp->t_dsack_bytes;
3867 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3868 xt->t_dsack_pack = tp->t_dsack_pack;
3869 xt->t_maxseg = tp->t_maxseg;
3870 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3871 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3873 now = getsbinuptime();
3874 #define COPYTIMER(which,where) do { \
3875 if (tp->t_timers[which] != SBT_MAX) \
3876 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \
3880 COPYTIMER(TT_DELACK, tt_delack);
3881 COPYTIMER(TT_REXMT, tt_rexmt);
3882 COPYTIMER(TT_PERSIST, tt_persist);
3883 COPYTIMER(TT_KEEP, tt_keep);
3884 COPYTIMER(TT_2MSL, tt_2msl);
3886 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3888 xt->xt_encaps_port = tp->t_port;
3889 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3890 TCP_FUNCTION_NAME_LEN_MAX);
3891 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3893 (void)tcp_log_get_id(tp, xt->xt_logid);
3896 xt->xt_len = sizeof(struct xtcpcb);
3897 in_pcbtoxinpcb(inp, &xt->xt_inp);
3901 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3906 (status > TCP_EI_STATUS_MAX_VALUE) ||
3911 if (status > (sizeof(uint32_t) * 8)) {
3912 /* Should this be a KASSERT? */
3915 bit = 1U << (status - 1);
3916 if (bit & tp->t_end_info_status) {
3917 /* already logged */
3920 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3921 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3922 tp->t_end_info_bytes[i] = status;
3923 tp->t_end_info_status |= bit;
3930 tcp_can_enable_pacing(void)
3933 if ((tcp_pacing_limit == -1) ||
3934 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
3935 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
3936 shadow_num_connections = number_of_tcp_connections_pacing;
3943 static uint8_t tcp_pacing_warning = 0;
3946 tcp_decrement_paced_conn(void)
3950 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
3951 shadow_num_connections = number_of_tcp_connections_pacing;
3952 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
3954 if (tcp_pacing_limit != -1) {
3955 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
3956 tcp_pacing_limit = 0;
3957 } else if (tcp_pacing_warning == 0) {
3958 printf("Warning pacing count is invalid, invalid decrement\n");
3959 tcp_pacing_warning = 1;