2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
48 #include <sys/hhook.h>
50 #include <sys/kernel.h>
52 #include <sys/khelp.h>
57 #include <sys/qmath.h>
58 #include <sys/stats.h>
59 #include <sys/sysctl.h>
61 #include <sys/malloc.h>
62 #include <sys/refcount.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/protosw.h>
70 #include <sys/random.h>
74 #include <net/route.h>
75 #include <net/route/nhop.h>
77 #include <net/if_var.h>
78 #include <net/if_private.h>
81 #include <netinet/in.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/in_kdtrace.h>
84 #include <netinet/in_pcb.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/in_var.h>
87 #include <netinet/ip.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/ip_var.h>
91 #include <netinet/icmp6.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/in6_fib.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet6/nd6.h>
100 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_ecn.h>
109 #include <netinet/tcp_log_buf.h>
110 #include <netinet/tcp_syncache.h>
111 #include <netinet/tcp_hpts.h>
112 #include <netinet/cc/cc.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/tcp_fastopen.h>
115 #include <netinet/tcp_accounting.h>
117 #include <netinet/tcp_pcap.h>
120 #include <netinet/tcp_offload.h>
122 #include <netinet/udp.h>
123 #include <netinet/udp_var.h>
125 #include <netinet6/tcp6_var.h>
128 #include <netipsec/ipsec_support.h>
130 #include <machine/in_cksum.h>
131 #include <crypto/siphash/siphash.h>
133 #include <security/mac/mac_framework.h>
136 static ip6proto_ctlinput_t tcp6_ctlinput;
137 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
140 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
142 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
145 #ifdef NETFLIX_EXP_DETECTION
146 /* Sack attack detection thresholds and such */
147 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
148 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
149 "Sack Attack detection thresholds");
150 int32_t tcp_force_detection = 0;
151 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
153 &tcp_force_detection, 0,
154 "Do we force detection even if the INP has it off?");
155 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
156 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
158 &tcp_sack_to_ack_thresh, 700,
159 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
160 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
161 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
163 &tcp_sack_to_move_thresh, 600,
164 "Percentage of sack moves we must see above (10.1 percent is 101)");
165 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
166 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
168 &tcp_restoral_thresh, 550,
169 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
170 int32_t tcp_sad_decay_val = 800;
171 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
173 &tcp_sad_decay_val, 800,
174 "The decay percentage (10.1 percent equals 101 )");
175 int32_t tcp_map_minimum = 500;
176 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
178 &tcp_map_minimum, 500,
179 "Number of Map enteries before we start detection");
180 int32_t tcp_sad_pacing_interval = 2000;
181 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
183 &tcp_sad_pacing_interval, 2000,
184 "What is the minimum pacing interval for a classified attacker?");
186 int32_t tcp_sad_low_pps = 100;
187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
189 &tcp_sad_low_pps, 100,
190 "What is the input pps that below which we do not decay?");
192 uint32_t tcp_ack_war_time_window = 1000;
193 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
195 &tcp_ack_war_time_window, 1000,
196 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
197 uint32_t tcp_ack_war_cnt = 5;
198 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
201 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
203 struct rwlock tcp_function_lock;
206 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
211 error = sysctl_handle_int(oidp, &new, 0, req);
212 if (error == 0 && req->newptr) {
213 if (new < TCP_MINMSS)
221 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
222 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
223 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
224 "Default TCP Maximum Segment Size");
228 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
232 new = V_tcp_v6mssdflt;
233 error = sysctl_handle_int(oidp, &new, 0, req);
234 if (error == 0 && req->newptr) {
235 if (new < TCP_MINMSS)
238 V_tcp_v6mssdflt = new;
243 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
244 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
245 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
246 "Default TCP Maximum Segment Size for IPv6");
250 * Minimum MSS we accept and use. This prevents DoS attacks where
251 * we are forced to a ridiculous low MSS like 20 and send hundreds
252 * of packets instead of one. The effect scales with the available
253 * bandwidth and quickly saturates the CPU and network interface
254 * with packet generation and sending. Set to zero to disable MINMSS
255 * checking. This setting prevents us from sending too small packets.
257 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
258 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
259 &VNET_NAME(tcp_minmss), 0,
260 "Minimum TCP Maximum Segment Size");
262 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
263 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
264 &VNET_NAME(tcp_do_rfc1323), 0,
265 "Enable rfc1323 (high performance TCP) extensions");
268 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
269 * Some stacks negotiate TS, but never send them after connection setup. Some
270 * stacks negotiate TS, but don't send them when sending keep-alive segments.
271 * These include modern widely deployed TCP stacks.
272 * Therefore tolerating violations for now...
274 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
275 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
276 &VNET_NAME(tcp_tolerate_missing_ts), 0,
277 "Tolerate missing TCP timestamps");
279 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
280 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
281 &VNET_NAME(tcp_ts_offset_per_conn), 0,
282 "Initialize TCP timestamps per connection instead of per host pair");
284 /* How many connections are pacing */
285 static volatile uint32_t number_of_tcp_connections_pacing = 0;
286 static uint32_t shadow_num_connections = 0;
288 static int tcp_pacing_limit = 10000;
289 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
290 &tcp_pacing_limit, 1000,
291 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
293 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
294 &shadow_num_connections, 0, "Number of TCP connections being paced");
296 static int tcp_log_debug = 0;
297 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
298 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
300 static int tcp_tcbhashsize;
301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
302 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
304 static int do_tcpdrain = 1;
305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
306 "Enable tcp_drain routine for extra help when low on mbufs");
308 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
309 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
311 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
312 #define V_icmp_may_rst VNET(icmp_may_rst)
313 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
314 &VNET_NAME(icmp_may_rst), 0,
315 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
317 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
318 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
319 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
320 &VNET_NAME(tcp_isn_reseed_interval), 0,
321 "Seconds between reseeding of ISN secret");
323 static int tcp_soreceive_stream;
324 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
325 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
327 VNET_DEFINE(uma_zone_t, sack_hole_zone);
328 #define V_sack_hole_zone VNET(sack_hole_zone)
329 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
331 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
336 new = V_tcp_map_entries_limit;
337 error = sysctl_handle_int(oidp, &new, 0, req);
338 if (error == 0 && req->newptr) {
339 /* only allow "0" and value > minimum */
340 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
343 V_tcp_map_entries_limit = new;
347 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
348 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
349 &VNET_NAME(tcp_map_entries_limit), 0,
350 &sysctl_net_inet_tcp_map_limit_check, "IU",
351 "Total sendmap entries limit");
353 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
354 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
355 &VNET_NAME(tcp_map_split_limit), 0,
356 "Total sendmap split entries limit");
359 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
362 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
363 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
364 #define V_ts_offset_secret VNET(ts_offset_secret)
366 static int tcp_default_fb_init(struct tcpcb *tp);
367 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
368 static int tcp_default_handoff_ok(struct tcpcb *tp);
369 static struct inpcb *tcp_notify(struct inpcb *, int);
370 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
371 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
372 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
373 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
374 const void *ip4hdr, const void *ip6hdr);
375 static ipproto_ctlinput_t tcp_ctlinput;
376 static udp_tun_icmp_t tcp_ctlinput_viaudp;
378 static struct tcp_function_block tcp_def_funcblk = {
379 .tfb_tcp_block_name = "freebsd",
380 .tfb_tcp_output = tcp_default_output,
381 .tfb_tcp_do_segment = tcp_do_segment,
382 .tfb_tcp_ctloutput = tcp_default_ctloutput,
383 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
384 .tfb_tcp_fb_init = tcp_default_fb_init,
385 .tfb_tcp_fb_fini = tcp_default_fb_fini,
388 static int tcp_fb_cnt = 0;
389 struct tcp_funchead t_functions;
390 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
393 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
395 TCPSTAT_INC(tcps_dsack_count);
398 if (SEQ_GT(end, start)) {
399 tp->t_dsack_bytes += (end - start);
400 TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
402 tp->t_dsack_tlp_bytes += (start - end);
403 TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
406 if (SEQ_GT(end, start)) {
407 tp->t_dsack_bytes += (end - start);
408 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
410 tp->t_dsack_tlp_bytes += (start - end);
411 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
416 static struct tcp_function_block *
417 find_tcp_functions_locked(struct tcp_function_set *fs)
419 struct tcp_function *f;
420 struct tcp_function_block *blk=NULL;
422 TAILQ_FOREACH(f, &t_functions, tf_next) {
423 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
431 static struct tcp_function_block *
432 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
434 struct tcp_function_block *rblk=NULL;
435 struct tcp_function *f;
437 TAILQ_FOREACH(f, &t_functions, tf_next) {
438 if (f->tf_fb == blk) {
449 struct tcp_function_block *
450 find_and_ref_tcp_functions(struct tcp_function_set *fs)
452 struct tcp_function_block *blk;
454 rw_rlock(&tcp_function_lock);
455 blk = find_tcp_functions_locked(fs);
457 refcount_acquire(&blk->tfb_refcnt);
458 rw_runlock(&tcp_function_lock);
462 struct tcp_function_block *
463 find_and_ref_tcp_fb(struct tcp_function_block *blk)
465 struct tcp_function_block *rblk;
467 rw_rlock(&tcp_function_lock);
468 rblk = find_tcp_fb_locked(blk, NULL);
470 refcount_acquire(&rblk->tfb_refcnt);
471 rw_runlock(&tcp_function_lock);
475 /* Find a matching alias for the given tcp_function_block. */
477 find_tcp_function_alias(struct tcp_function_block *blk,
478 struct tcp_function_set *fs)
480 struct tcp_function *f;
484 rw_rlock(&tcp_function_lock);
485 TAILQ_FOREACH(f, &t_functions, tf_next) {
486 if ((f->tf_fb == blk) &&
487 (strncmp(f->tf_name, blk->tfb_tcp_block_name,
488 TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
489 /* Matching function block with different name. */
490 strncpy(fs->function_set_name, f->tf_name,
491 TCP_FUNCTION_NAME_LEN_MAX);
496 /* Null terminate the string appropriately. */
498 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
500 fs->function_set_name[0] = '\0';
502 rw_runlock(&tcp_function_lock);
506 static struct tcp_function_block *
507 find_and_ref_tcp_default_fb(void)
509 struct tcp_function_block *rblk;
511 rw_rlock(&tcp_function_lock);
512 rblk = tcp_func_set_ptr;
513 refcount_acquire(&rblk->tfb_refcnt);
514 rw_runlock(&tcp_function_lock);
519 tcp_switch_back_to_default(struct tcpcb *tp)
521 struct tcp_function_block *tfb;
523 KASSERT(tp->t_fb != &tcp_def_funcblk,
524 ("%s: called by the built-in default stack", __func__));
527 * Release the old stack. This function will either find a new one
530 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
531 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
532 refcount_release(&tp->t_fb->tfb_refcnt);
535 * Now, we'll find a new function block to use.
536 * Start by trying the current user-selected
537 * default, unless this stack is the user-selected
540 tfb = find_and_ref_tcp_default_fb();
541 if (tfb == tp->t_fb) {
542 refcount_release(&tfb->tfb_refcnt);
545 /* Does the stack accept this connection? */
546 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
547 (*tfb->tfb_tcp_handoff_ok)(tp)) {
548 refcount_release(&tfb->tfb_refcnt);
551 /* Try to use that stack. */
553 /* Initialize the new stack. If it succeeds, we are done. */
555 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
556 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
560 * Initialization failed. Release the reference count on
563 refcount_release(&tfb->tfb_refcnt);
567 * If that wasn't feasible, use the built-in default
568 * stack which is not allowed to reject anyone.
570 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
572 /* there always should be a default */
573 panic("Can't refer to tcp_def_funcblk");
575 if (tfb->tfb_tcp_handoff_ok != NULL) {
576 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
577 /* The default stack cannot say no */
578 panic("Default stack rejects a new session?");
582 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
583 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
584 /* The default stack cannot fail */
585 panic("Default stack initialization failed");
590 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
591 const struct sockaddr *sa, void *ctx)
602 TCPSTAT_INC(tcps_tunneled_pkts);
603 if ((m->m_flags & M_PKTHDR) == 0) {
604 /* Can't handle one that is not a pkt hdr */
605 TCPSTAT_INC(tcps_tunneled_errs);
608 thlen = sizeof(struct tcphdr);
609 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
610 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
611 TCPSTAT_INC(tcps_tunneled_errs);
614 iph = mtod(m, struct ip *);
615 uh = (struct udphdr *)((caddr_t)iph + off);
616 th = (struct tcphdr *)(uh + 1);
617 thlen = th->th_off << 2;
618 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
619 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
621 TCPSTAT_INC(tcps_tunneled_errs);
624 iph = mtod(m, struct ip *);
625 uh = (struct udphdr *)((caddr_t)iph + off);
626 th = (struct tcphdr *)(uh + 1);
629 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
630 bcopy(th, uh, m->m_len - off);
631 m->m_len -= sizeof(struct udphdr);
632 m->m_pkthdr.len -= sizeof(struct udphdr);
634 * We use the same algorithm for
635 * both UDP and TCP for c-sum. So
636 * the code in tcp_input will skip
637 * the checksum. So we do nothing
638 * with the flag (m->m_pkthdr.csum_flags).
643 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
644 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
648 case IPV6_VERSION >> 4:
649 ip6 = mtod(m, struct ip6_hdr *);
650 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
651 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
666 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
669 struct tcp_function_set fs;
670 struct tcp_function_block *blk;
672 memset(&fs, 0, sizeof(fs));
673 rw_rlock(&tcp_function_lock);
674 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
677 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
678 fs.pcbcnt = blk->tfb_refcnt;
680 rw_runlock(&tcp_function_lock);
681 error = sysctl_handle_string(oidp, fs.function_set_name,
682 sizeof(fs.function_set_name), req);
684 /* Check for error or no change */
685 if (error != 0 || req->newptr == NULL)
688 rw_wlock(&tcp_function_lock);
689 blk = find_tcp_functions_locked(&fs);
691 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
695 tcp_func_set_ptr = blk;
697 rw_wunlock(&tcp_function_lock);
701 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
702 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
703 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
704 "Set/get the default TCP functions");
707 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
709 int error, cnt, linesz;
710 struct tcp_function *f;
716 rw_rlock(&tcp_function_lock);
717 TAILQ_FOREACH(f, &t_functions, tf_next) {
720 rw_runlock(&tcp_function_lock);
722 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
723 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
728 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
729 "Alias", "PCB count");
734 rw_rlock(&tcp_function_lock);
735 TAILQ_FOREACH(f, &t_functions, tf_next) {
736 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
737 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
738 f->tf_fb->tfb_tcp_block_name,
739 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
740 alias ? f->tf_name : "-",
741 f->tf_fb->tfb_refcnt);
742 if (linesz >= bufsz) {
750 rw_runlock(&tcp_function_lock);
752 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
753 free(buffer, M_TEMP);
757 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
758 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
759 NULL, 0, sysctl_net_inet_list_available, "A",
760 "list available TCP Function sets");
762 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
765 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
766 #define V_udp4_tun_socket VNET(udp4_tun_socket)
769 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
770 #define V_udp6_tun_socket VNET(udp6_tun_socket)
773 static struct sx tcpoudp_lock;
776 tcp_over_udp_stop(void)
779 sx_assert(&tcpoudp_lock, SA_XLOCKED);
782 if (V_udp4_tun_socket != NULL) {
783 soclose(V_udp4_tun_socket);
784 V_udp4_tun_socket = NULL;
788 if (V_udp6_tun_socket != NULL) {
789 soclose(V_udp6_tun_socket);
790 V_udp6_tun_socket = NULL;
796 tcp_over_udp_start(void)
801 struct sockaddr_in sin;
804 struct sockaddr_in6 sin6;
807 sx_assert(&tcpoudp_lock, SA_XLOCKED);
809 port = V_tcp_udp_tunneling_port;
810 if (ntohs(port) == 0) {
811 /* Must have a port set */
815 if (V_udp4_tun_socket != NULL) {
816 /* Already running -- must stop first */
821 if (V_udp6_tun_socket != NULL) {
822 /* Already running -- must stop first */
827 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
828 SOCK_DGRAM, IPPROTO_UDP,
829 curthread->td_ucred, curthread))) {
833 /* Call the special UDP hook. */
834 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
835 tcp_recv_udp_tunneled_packet,
841 /* Ok, we have a socket, bind it to the port. */
842 memset(&sin, 0, sizeof(struct sockaddr_in));
843 sin.sin_len = sizeof(struct sockaddr_in);
844 sin.sin_family = AF_INET;
845 sin.sin_port = htons(port);
846 if ((ret = sobind(V_udp4_tun_socket,
847 (struct sockaddr *)&sin, curthread))) {
853 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
854 SOCK_DGRAM, IPPROTO_UDP,
855 curthread->td_ucred, curthread))) {
859 /* Call the special UDP hook. */
860 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
861 tcp_recv_udp_tunneled_packet,
862 tcp6_ctlinput_viaudp,
867 /* Ok, we have a socket, bind it to the port. */
868 memset(&sin6, 0, sizeof(struct sockaddr_in6));
869 sin6.sin6_len = sizeof(struct sockaddr_in6);
870 sin6.sin6_family = AF_INET6;
871 sin6.sin6_port = htons(port);
872 if ((ret = sobind(V_udp6_tun_socket,
873 (struct sockaddr *)&sin6, curthread))) {
882 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
887 old = V_tcp_udp_tunneling_port;
889 error = sysctl_handle_int(oidp, &new, 0, req);
891 (req->newptr != NULL)) {
892 if ((new < TCP_TUNNELING_PORT_MIN) ||
893 (new > TCP_TUNNELING_PORT_MAX)) {
896 sx_xlock(&tcpoudp_lock);
897 V_tcp_udp_tunneling_port = new;
902 error = tcp_over_udp_start();
904 V_tcp_udp_tunneling_port = 0;
907 sx_xunlock(&tcpoudp_lock);
913 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
914 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
915 &VNET_NAME(tcp_udp_tunneling_port),
916 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
917 "Tunneling port for tcp over udp");
919 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
922 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
926 new = V_tcp_udp_tunneling_overhead;
927 error = sysctl_handle_int(oidp, &new, 0, req);
928 if (error == 0 && req->newptr) {
929 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
930 (new > TCP_TUNNELING_OVERHEAD_MAX))
933 V_tcp_udp_tunneling_overhead = new;
938 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
939 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
940 &VNET_NAME(tcp_udp_tunneling_overhead),
941 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
942 "MSS reduction when using tcp over udp");
945 * Exports one (struct tcp_function_info) for each alias/name.
948 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
951 struct tcp_function *f;
952 struct tcp_function_info tfi;
955 * We don't allow writes.
957 if (req->newptr != NULL)
961 * Wire the old buffer so we can directly copy the functions to
962 * user space without dropping the lock.
964 if (req->oldptr != NULL) {
965 error = sysctl_wire_old_buffer(req, 0);
971 * Walk the list and copy out matching entries. If INVARIANTS
972 * is compiled in, also walk the list to verify the length of
973 * the list matches what we have recorded.
975 rw_rlock(&tcp_function_lock);
979 if (req->oldptr == NULL) {
984 TAILQ_FOREACH(f, &t_functions, tf_next) {
988 if (req->oldptr != NULL) {
989 bzero(&tfi, sizeof(tfi));
990 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
991 tfi.tfi_id = f->tf_fb->tfb_id;
992 (void)strlcpy(tfi.tfi_alias, f->tf_name,
993 sizeof(tfi.tfi_alias));
994 (void)strlcpy(tfi.tfi_name,
995 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
996 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
998 * Don't stop on error, as that is the
999 * mechanism we use to accumulate length
1000 * information if the buffer was too short.
1004 KASSERT(cnt == tcp_fb_cnt,
1005 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1009 rw_runlock(&tcp_function_lock);
1010 if (req->oldptr == NULL)
1011 error = SYSCTL_OUT(req, NULL,
1012 (cnt + 1) * sizeof(struct tcp_function_info));
1017 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1018 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1019 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1020 "List TCP function block name-to-ID mappings");
1023 * tfb_tcp_handoff_ok() function for the default stack.
1024 * Note that we'll basically try to take all comers.
1027 tcp_default_handoff_ok(struct tcpcb *tp)
1034 * tfb_tcp_fb_init() function for the default stack.
1036 * This handles making sure we have appropriate timers set if you are
1037 * transitioning a socket that has some amount of setup done.
1039 * The init() fuction from the default can *never* return non-zero i.e.
1040 * it is required to always succeed since it is the stack of last resort!
1043 tcp_default_fb_init(struct tcpcb *tp)
1045 struct socket *so = tptosocket(tp);
1047 INP_WLOCK_ASSERT(tptoinpcb(tp));
1049 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1050 ("%s: connection %p in unexpected state %d", __func__, tp,
1054 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1055 * know what to do for unexpected states (which includes TIME_WAIT).
1057 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1061 * Make sure some kind of transmission timer is set if there is
1064 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1065 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1066 tcp_timer_active(tp, TT_PERSIST))) {
1068 * If the session has established and it looks like it should
1069 * be in the persist state, set the persist timer. Otherwise,
1070 * set the retransmit timer.
1072 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1073 (int32_t)(tp->snd_nxt - tp->snd_una) <
1074 (int32_t)sbavail(&so->so_snd))
1077 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1080 /* All non-embryonic sessions get a keepalive timer. */
1081 if (!tcp_timer_active(tp, TT_KEEP))
1082 tcp_timer_activate(tp, TT_KEEP,
1083 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1087 * Make sure critical variables are initialized
1088 * if transitioning while in Recovery.
1090 if IN_FASTRECOVERY(tp->t_flags) {
1091 if (tp->sackhint.recover_fs == 0)
1092 tp->sackhint.recover_fs = max(1,
1093 tp->snd_nxt - tp->snd_una);
1100 * tfb_tcp_fb_fini() function for the default stack.
1102 * This changes state as necessary (or prudent) to prepare for another stack
1103 * to assume responsibility for the connection.
1106 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1109 INP_WLOCK_ASSERT(tptoinpcb(tp));
1113 * Target size of TCP PCB hash tables. Must be a power of two.
1115 * Note that this can be overridden by the kernel environment
1116 * variable net.inet.tcp.tcbhashsize
1119 #define TCBHASHSIZE 0
1122 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1123 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1125 static struct mtx isn_mtx;
1127 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1128 #define ISN_LOCK() mtx_lock(&isn_mtx)
1129 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1131 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1134 * Take a value and get the next power of 2 that doesn't overflow.
1135 * Used to size the tcp_inpcb hash buckets.
1138 maketcp_hashsize(int size)
1144 * get the next power of 2 higher than maxsockets.
1146 hashsize = 1 << fls(size);
1147 /* catch overflow, and just go one power of 2 smaller */
1148 if (hashsize < size) {
1149 hashsize = 1 << (fls(size) - 1);
1154 static volatile int next_tcp_stack_id = 1;
1157 * Register a TCP function block with the name provided in the names
1158 * array. (Note that this function does NOT automatically register
1159 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1160 * explicitly include blk->tfb_tcp_block_name in the list of names if
1161 * you wish to register the stack with that name.)
1163 * Either all name registrations will succeed or all will fail. If
1164 * a name registration fails, the function will update the num_names
1165 * argument to point to the array index of the name that encountered
1168 * Returns 0 on success, or an error code on failure.
1171 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1172 const char *names[], int *num_names)
1174 struct tcp_function *n;
1175 struct tcp_function_set fs;
1178 KASSERT(names != NULL && *num_names > 0,
1179 ("%s: Called with 0-length name list", __func__));
1180 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1181 KASSERT(rw_initialized(&tcp_function_lock),
1182 ("%s: called too early", __func__));
1184 if ((blk->tfb_tcp_output == NULL) ||
1185 (blk->tfb_tcp_do_segment == NULL) ||
1186 (blk->tfb_tcp_ctloutput == NULL) ||
1187 (strlen(blk->tfb_tcp_block_name) == 0)) {
1189 * These functions are required and you
1196 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1201 refcount_init(&blk->tfb_refcnt, 0);
1202 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1203 for (i = 0; i < *num_names; i++) {
1204 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1211 (void)strlcpy(fs.function_set_name, names[i],
1212 sizeof(fs.function_set_name));
1213 rw_wlock(&tcp_function_lock);
1214 if (find_tcp_functions_locked(&fs) != NULL) {
1215 /* Duplicate name space not allowed */
1216 rw_wunlock(&tcp_function_lock);
1217 free(n, M_TCPFUNCTIONS);
1221 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1222 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1224 rw_wunlock(&tcp_function_lock);
1230 * Deregister the names we just added. Because registration failed
1231 * for names[i], we don't need to deregister that name.
1234 rw_wlock(&tcp_function_lock);
1236 TAILQ_FOREACH(n, &t_functions, tf_next) {
1237 if (!strncmp(n->tf_name, names[i],
1238 TCP_FUNCTION_NAME_LEN_MAX)) {
1239 TAILQ_REMOVE(&t_functions, n, tf_next);
1242 free(n, M_TCPFUNCTIONS);
1247 rw_wunlock(&tcp_function_lock);
1252 * Register a TCP function block using the name provided in the name
1255 * Returns 0 on success, or an error code on failure.
1258 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1261 const char *name_list[1];
1266 name_list[0] = name;
1268 name_list[0] = blk->tfb_tcp_block_name;
1269 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1274 * Register a TCP function block using the name defined in
1275 * blk->tfb_tcp_block_name.
1277 * Returns 0 on success, or an error code on failure.
1280 register_tcp_functions(struct tcp_function_block *blk, int wait)
1283 return (register_tcp_functions_as_name(blk, NULL, wait));
1287 * Deregister all names associated with a function block. This
1288 * functionally removes the function block from use within the system.
1290 * When called with a true quiesce argument, mark the function block
1291 * as being removed so no more stacks will use it and determine
1292 * whether the removal would succeed.
1294 * When called with a false quiesce argument, actually attempt the
1297 * When called with a force argument, attempt to switch all TCBs to
1298 * use the default stack instead of returning EBUSY.
1300 * Returns 0 on success (or if the removal would succeed, or an error
1304 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1307 struct tcp_function *f;
1309 if (blk == &tcp_def_funcblk) {
1310 /* You can't un-register the default */
1313 rw_wlock(&tcp_function_lock);
1314 if (blk == tcp_func_set_ptr) {
1315 /* You can't free the current default */
1316 rw_wunlock(&tcp_function_lock);
1319 /* Mark the block so no more stacks can use it. */
1320 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1322 * If TCBs are still attached to the stack, attempt to switch them
1323 * to the default stack.
1325 if (force && blk->tfb_refcnt) {
1326 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1327 INPLOOKUP_WLOCKPCB);
1330 VNET_ITERATOR_DECL(vnet_iter);
1332 rw_wunlock(&tcp_function_lock);
1335 VNET_FOREACH(vnet_iter) {
1336 CURVNET_SET(vnet_iter);
1337 while ((inp = inp_next(&inpi)) != NULL) {
1338 tp = intotcpcb(inp);
1339 if (tp == NULL || tp->t_fb != blk)
1341 tcp_switch_back_to_default(tp);
1345 VNET_LIST_RUNLOCK();
1347 rw_wlock(&tcp_function_lock);
1349 if (blk->tfb_refcnt) {
1350 /* TCBs still attached. */
1351 rw_wunlock(&tcp_function_lock);
1356 rw_wunlock(&tcp_function_lock);
1359 /* Remove any function names that map to this function block. */
1360 while (find_tcp_fb_locked(blk, &f) != NULL) {
1361 TAILQ_REMOVE(&t_functions, f, tf_next);
1364 free(f, M_TCPFUNCTIONS);
1366 rw_wunlock(&tcp_function_lock);
1373 struct epoch_tracker et;
1374 VNET_ITERATOR_DECL(vnet_iter);
1379 NET_EPOCH_ENTER(et);
1380 VNET_LIST_RLOCK_NOSLEEP();
1381 VNET_FOREACH(vnet_iter) {
1382 CURVNET_SET(vnet_iter);
1383 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1384 INPLOOKUP_WLOCKPCB);
1389 * Walk the tcpbs, if existing, and flush the reassembly queue,
1390 * if there is one...
1391 * XXX: The "Net/3" implementation doesn't imply that the TCP
1392 * reassembly queue should be flushed, but in a situation
1393 * where we're really low on mbufs, this is potentially
1396 while ((inpb = inp_next(&inpi)) != NULL) {
1397 if ((tcpb = intotcpcb(inpb)) != NULL) {
1398 tcp_reass_flush(tcpb);
1399 tcp_clean_sackreport(tcpb);
1401 tcp_log_drain(tcpb);
1404 if (tcp_pcap_aggressive_free) {
1405 /* Free the TCP PCAP queues. */
1406 tcp_pcap_drain(&(tcpb->t_inpkts));
1407 tcp_pcap_drain(&(tcpb->t_outpkts));
1414 VNET_LIST_RUNLOCK_NOSLEEP();
1419 tcp_vnet_init(void *arg __unused)
1423 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1424 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1425 printf("%s: WARNING: unable to register helper hook\n", __func__);
1426 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1427 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1428 printf("%s: WARNING: unable to register helper hook\n", __func__);
1431 if (tcp_stats_init())
1432 printf("%s: WARNING: unable to initialise TCP stats\n",
1435 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1441 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1442 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1443 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1445 tcp_fastopen_init();
1447 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1448 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1450 V_tcp_msl = TCPTV_MSL;
1452 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1453 tcp_vnet_init, NULL);
1456 tcp_init(void *arg __unused)
1458 const char *tcbhash_tuneable;
1461 tcp_reass_global_init();
1463 /* XXX virtualize those below? */
1464 tcp_delacktime = TCPTV_DELACK;
1465 tcp_keepinit = TCPTV_KEEP_INIT;
1466 tcp_keepidle = TCPTV_KEEP_IDLE;
1467 tcp_keepintvl = TCPTV_KEEPINTVL;
1468 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1469 tcp_rexmit_initial = TCPTV_RTOBASE;
1470 if (tcp_rexmit_initial < 1)
1471 tcp_rexmit_initial = 1;
1472 tcp_rexmit_min = TCPTV_MIN;
1473 if (tcp_rexmit_min < 1)
1475 tcp_persmin = TCPTV_PERSMIN;
1476 tcp_persmax = TCPTV_PERSMAX;
1477 tcp_rexmit_slop = TCPTV_CPU_VAR;
1478 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1480 /* Setup the tcp function block list */
1481 TAILQ_INIT(&t_functions);
1482 rw_init(&tcp_function_lock, "tcp_func_lock");
1483 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1484 sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1486 /* Initialize the TCP logging data. */
1489 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1491 if (tcp_soreceive_stream) {
1493 tcp_protosw.pr_soreceive = soreceive_stream;
1496 tcp6_protosw.pr_soreceive = soreceive_stream;
1501 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1503 max_protohdr_grow(sizeof(struct tcpiphdr));
1507 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1508 SHUTDOWN_PRI_DEFAULT);
1509 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1510 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1512 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1513 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1514 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1515 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1516 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1517 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1518 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1519 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1520 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1525 hashsize = TCBHASHSIZE;
1526 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1527 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1528 if (hashsize == 0) {
1530 * Auto tune the hash size based on maxsockets.
1531 * A perfect hash would have a 1:1 mapping
1532 * (hashsize = maxsockets) however it's been
1533 * suggested that O(2) average is better.
1535 hashsize = maketcp_hashsize(maxsockets / 4);
1537 * Our historical default is 512,
1538 * do not autotune lower than this.
1543 printf("%s: %s auto tuned to %d\n", __func__,
1544 tcbhash_tuneable, hashsize);
1547 * We require a hashsize to be a power of two.
1548 * Previously if it was not a power of two we would just reset it
1549 * back to 512, which could be a nasty surprise if you did not notice
1550 * the error message.
1551 * Instead what we do is clip it to the closest power of two lower
1552 * than the specified hash value.
1554 if (!powerof2(hashsize)) {
1555 int oldhashsize = hashsize;
1557 hashsize = maketcp_hashsize(hashsize);
1558 /* prevent absurdly low value */
1561 printf("%s: WARNING: TCB hash size not a power of 2, "
1562 "clipped from %d to %d.\n", __func__, oldhashsize,
1565 tcp_tcbhashsize = hashsize;
1568 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1571 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1574 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1578 tcp_destroy(void *unused __unused)
1586 * All our processes are gone, all our sockets should be cleaned
1587 * up, which means, we should be past the tcp_discardcb() calls.
1588 * Sleep to let all tcpcb timers really disappear and cleanup.
1591 INP_INFO_WLOCK(&V_tcbinfo);
1592 n = V_tcbinfo.ipi_count;
1593 INP_INFO_WUNLOCK(&V_tcbinfo);
1596 pause("tcpdes", hz / 10);
1600 in_pcbinfo_destroy(&V_tcbinfo);
1601 /* tcp_discardcb() clears the sack_holes up. */
1602 uma_zdestroy(V_sack_hole_zone);
1605 * Cannot free the zone until all tcpcbs are released as we attach
1606 * the allocations to them.
1608 tcp_fastopen_destroy();
1610 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1611 VNET_PCPUSTAT_FREE(tcpstat);
1614 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1616 printf("%s: WARNING: unable to deregister helper hook "
1617 "type=%d, id=%d: error %d returned\n", __func__,
1618 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1620 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1622 printf("%s: WARNING: unable to deregister helper hook "
1623 "type=%d, id=%d: error %d returned\n", __func__,
1624 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1628 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1638 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1639 * tcp_template used to store this data in mbufs, but we now recopy it out
1640 * of the tcpcb each time to conserve mbufs.
1643 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1645 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1647 INP_WLOCK_ASSERT(inp);
1650 if ((inp->inp_vflag & INP_IPV6) != 0) {
1651 struct ip6_hdr *ip6;
1653 ip6 = (struct ip6_hdr *)ip_ptr;
1654 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1655 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1656 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1657 (IPV6_VERSION & IPV6_VERSION_MASK);
1659 ip6->ip6_nxt = IPPROTO_TCP;
1661 ip6->ip6_nxt = IPPROTO_UDP;
1662 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1663 ip6->ip6_src = inp->in6p_laddr;
1664 ip6->ip6_dst = inp->in6p_faddr;
1667 #if defined(INET6) && defined(INET)
1674 ip = (struct ip *)ip_ptr;
1675 ip->ip_v = IPVERSION;
1677 ip->ip_tos = inp->inp_ip_tos;
1681 ip->ip_ttl = inp->inp_ip_ttl;
1684 ip->ip_p = IPPROTO_TCP;
1686 ip->ip_p = IPPROTO_UDP;
1687 ip->ip_src = inp->inp_laddr;
1688 ip->ip_dst = inp->inp_faddr;
1691 th->th_sport = inp->inp_lport;
1692 th->th_dport = inp->inp_fport;
1696 tcp_set_flags(th, 0);
1699 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1703 * Create template to be used to send tcp packets on a connection.
1704 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1705 * use for this function is in keepalives, which use tcp_respond.
1708 tcpip_maketemplate(struct inpcb *inp)
1712 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1715 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1720 * Send a single message to the TCP at address specified by
1721 * the given TCP/IP header. If m == NULL, then we make a copy
1722 * of the tcpiphdr at th and send directly to the addressed host.
1723 * This is used to force keep alive messages out using the TCP
1724 * template for a connection. If flags are given then we send
1725 * a message back to the TCP which originated the segment th,
1726 * and discard the mbuf containing it and any other attached mbufs.
1728 * In any case the ack and sequence number of the transmitted
1729 * segment are as specified by the parameters.
1731 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1734 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1735 tcp_seq ack, tcp_seq seq, uint16_t flags)
1741 struct udphdr *uh = NULL;
1743 struct tcp_log_buffer *lgb;
1746 struct ip6_hdr *ip6;
1749 int optlen, tlen, win, ulen;
1755 int thflags = tcp_get_flags(th);
1758 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1762 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1768 inp = tptoinpcb(tp);
1769 INP_LOCK_ASSERT(inp);
1775 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1776 port = m->m_pkthdr.tcp_tun_port;
1779 if (ip && (ip->ip_p == IPPROTO_UDP))
1780 port = m->m_pkthdr.tcp_tun_port;
1789 if (!(flags & TH_RST)) {
1790 win = sbspace(&inp->inp_socket->so_rcv);
1791 if (win > TCP_MAXWIN << tp->rcv_scale)
1792 win = TCP_MAXWIN << tp->rcv_scale;
1794 if ((tp->t_flags & TF_NOOPT) == 0)
1798 m = m_gethdr(M_NOWAIT, MT_DATA);
1801 m->m_data += max_linkhdr;
1804 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1805 sizeof(struct ip6_hdr));
1806 ip6 = mtod(m, struct ip6_hdr *);
1807 nth = (struct tcphdr *)(ip6 + 1);
1809 /* Insert a UDP header */
1810 uh = (struct udphdr *)nth;
1811 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1812 uh->uh_dport = port;
1813 nth = (struct tcphdr *)(uh + 1);
1818 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1819 ip = mtod(m, struct ip *);
1820 nth = (struct tcphdr *)(ip + 1);
1822 /* Insert a UDP header */
1823 uh = (struct udphdr *)nth;
1824 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1825 uh->uh_dport = port;
1826 nth = (struct tcphdr *)(uh + 1);
1829 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1831 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1834 /* Can't reuse 'm', allocate a new mbuf. */
1835 n = m_gethdr(M_NOWAIT, MT_DATA);
1841 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1847 n->m_data += max_linkhdr;
1848 /* m_len is set later */
1849 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1852 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1853 sizeof(struct ip6_hdr));
1854 ip6 = mtod(n, struct ip6_hdr *);
1855 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1856 nth = (struct tcphdr *)(ip6 + 1);
1858 /* Insert a UDP header */
1859 uh = (struct udphdr *)nth;
1860 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1861 uh->uh_dport = port;
1862 nth = (struct tcphdr *)(uh + 1);
1867 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1868 ip = mtod(n, struct ip *);
1869 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1870 nth = (struct tcphdr *)(ip + 1);
1872 /* Insert a UDP header */
1873 uh = (struct udphdr *)nth;
1874 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1875 uh->uh_dport = port;
1876 nth = (struct tcphdr *)(uh + 1);
1879 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1880 xchg(nth->th_dport, nth->th_sport, uint16_t);
1887 * XXX MRT We inherit the FIB, which is lucky.
1891 m->m_data = (caddr_t)ipgen;
1892 /* m_len is set later */
1895 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1896 nth = (struct tcphdr *)(ip6 + 1);
1900 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1901 nth = (struct tcphdr *)(ip + 1);
1905 * this is usually a case when an extension header
1906 * exists between the IPv6 header and the
1909 nth->th_sport = th->th_sport;
1910 nth->th_dport = th->th_dport;
1912 xchg(nth->th_dport, nth->th_sport, uint16_t);
1918 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1920 #if defined(INET) && defined(INET6)
1924 tlen = sizeof (struct tcpiphdr);
1927 tlen += sizeof (struct udphdr);
1930 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1931 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1932 m, tlen, (long)M_TRAILINGSPACE(m)));
1937 ect = tcp_ecn_output_established(tp, &flags, 0, false);
1938 /* Make sure we have room. */
1939 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1940 m->m_next = m_get(M_NOWAIT, MT_DATA);
1942 optp = mtod(m->m_next, u_char *);
1947 optp = (u_char *) (nth + 1);
1953 if (tp->t_flags & TF_RCVD_TSTMP) {
1954 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1955 to.to_tsecr = tp->ts_recent;
1956 to.to_flags |= TOF_TS;
1958 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1959 /* TCP-MD5 (RFC2385). */
1960 if (tp->t_flags & TF_SIGNATURE)
1961 to.to_flags |= TOF_SIGNATURE;
1963 /* Add the options. */
1964 tlen += optlen = tcp_addoptions(&to, optp);
1966 /* Update m_len in the correct mbuf. */
1967 optm->m_len += optlen;
1973 ulen = tlen - sizeof(struct ip6_hdr);
1974 uh->uh_ulen = htons(ulen);
1976 ip6->ip6_flow = htonl(ect << 20);
1977 ip6->ip6_vfc = IPV6_VERSION;
1979 ip6->ip6_nxt = IPPROTO_UDP;
1981 ip6->ip6_nxt = IPPROTO_TCP;
1982 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1985 #if defined(INET) && defined(INET6)
1991 ulen = tlen - sizeof(struct ip);
1992 uh->uh_ulen = htons(ulen);
1995 ip->ip_len = htons(tlen);
1996 ip->ip_ttl = V_ip_defttl;
1998 ip->ip_p = IPPROTO_UDP;
2000 ip->ip_p = IPPROTO_TCP;
2002 if (V_path_mtu_discovery)
2003 ip->ip_off |= htons(IP_DF);
2006 m->m_pkthdr.len = tlen;
2007 m->m_pkthdr.rcvif = NULL;
2011 * Packet is associated with a socket, so allow the
2012 * label of the response to reflect the socket label.
2014 INP_LOCK_ASSERT(inp);
2015 mac_inpcb_create_mbuf(inp, m);
2018 * Packet is not associated with a socket, so possibly
2019 * update the label in place.
2021 mac_netinet_tcp_reply(m);
2024 nth->th_seq = htonl(seq);
2025 nth->th_ack = htonl(ack);
2026 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2027 tcp_set_flags(nth, flags);
2029 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2031 nth->th_win = htons((u_short)win);
2034 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2035 if (to.to_flags & TOF_SIGNATURE) {
2036 if (!TCPMD5_ENABLED() ||
2037 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2047 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2048 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2049 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2052 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2053 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2054 nth->th_sum = in6_cksum_pseudo(ip6,
2055 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2057 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2060 #if defined(INET6) && defined(INET)
2066 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2067 htons(ulen + IPPROTO_UDP));
2068 m->m_pkthdr.csum_flags = CSUM_UDP;
2069 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2072 m->m_pkthdr.csum_flags = CSUM_TCP;
2073 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2074 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2075 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2079 TCP_PROBE3(debug__output, tp, th, m);
2081 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2083 if ((tp != NULL) && tcp_bblogging_on(tp)) {
2084 if (INP_WLOCKED(inp)) {
2085 union tcp_log_stackspecific log;
2088 memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2089 log.u_bbr.inhpts = inp->inp_in_hpts;
2090 log.u_bbr.flex8 = 4;
2091 log.u_bbr.pkts_out = tp->t_maxseg;
2092 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2093 log.u_bbr.delivered = 0;
2094 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT,
2095 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2098 * We can not log the packet, since we only own the
2099 * read lock, but a write lock is needed. The read lock
2100 * is not upgraded to a write lock, since only getting
2101 * the read lock was done intentionally to improve the
2102 * handling of SYN flooding attacks.
2103 * This happens only for pure SYN segments received in
2104 * the initial CLOSED state, or received in a more
2105 * advanced state than listen and the UDP encapsulation
2106 * port is unexpected.
2107 * The incoming SYN segments do not really belong to
2108 * the TCP connection and the handling does not change
2109 * the state of the TCP connection. Therefore, the
2110 * sending of the RST segments is not logged. Please
2111 * note that also the incoming SYN segments are not
2114 * The following code ensures that the above description
2115 * is and stays correct.
2117 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2118 (tp->t_state == TCPS_CLOSED ||
2119 (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2120 ("%s: Logging of TCP segment with flags 0x%b and "
2121 "UDP encapsulation port %u skipped in state %s",
2122 __func__, thflags, PRINT_TH_FLAGS,
2123 ntohs(port), tcpstates[tp->t_state]));
2128 TCPSTAT_INC(tcps_sndacks);
2129 else if (flags & (TH_SYN|TH_FIN|TH_RST))
2130 TCPSTAT_INC(tcps_sndctrl);
2131 TCPSTAT_INC(tcps_sndtotal);
2135 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2136 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2139 #if defined(INET) && defined(INET6)
2144 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2145 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2149 lgb->tlb_errno = output_ret;
2153 * Create a new TCP control block, making an empty reassembly queue and hooking
2154 * it to the argument protocol control block. The `inp' parameter must have
2155 * come from the zone allocator set up by tcpcbstor declaration.
2158 tcp_newtcpcb(struct inpcb *inp)
2160 struct tcpcb *tp = intotcpcb(inp);
2162 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2166 * Historically allocation was done with M_ZERO. There is a lot of
2167 * code that rely on that. For now take safe approach and zero whole
2168 * tcpcb. This definitely can be optimized.
2170 bzero(&tp->t_start_zero, t_zero_size);
2172 /* Initialise cc_var struct for this tcpcb. */
2173 tp->t_ccv.type = IPPROTO_TCP;
2174 tp->t_ccv.ccvc.tcp = tp;
2175 rw_rlock(&tcp_function_lock);
2176 tp->t_fb = tcp_func_set_ptr;
2177 refcount_acquire(&tp->t_fb->tfb_refcnt);
2178 rw_runlock(&tcp_function_lock);
2180 * Use the current system default CC algorithm.
2182 cc_attach(tp, CC_DEFAULT_ALGO());
2184 if (CC_ALGO(tp)->cb_init != NULL)
2185 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2187 if (tp->t_fb->tfb_tcp_fb_fini)
2188 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2189 refcount_release(&tp->t_fb->tfb_refcnt);
2194 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2195 if (tp->t_fb->tfb_tcp_fb_fini)
2196 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2197 refcount_release(&tp->t_fb->tfb_refcnt);
2202 TAILQ_INIT(&tp->t_segq);
2205 isipv6 ? V_tcp_v6mssdflt :
2209 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2210 for (int i = 0; i < TT_N; i++)
2211 tp->t_timers[i] = SBT_MAX;
2213 switch (V_tcp_do_rfc1323) {
2218 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2221 tp->t_flags = TF_REQ_SCALE;
2224 tp->t_flags = TF_REQ_TSTMP;
2228 tp->t_flags |= TF_SACK_PERMIT;
2229 TAILQ_INIT(&tp->snd_holes);
2232 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2233 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2234 * reasonable initial retransmit time.
2236 tp->t_srtt = TCPTV_SRTTBASE;
2237 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2238 tp->t_rttmin = tcp_rexmit_min;
2239 tp->t_rxtcur = tcp_rexmit_initial;
2240 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2241 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2242 tp->t_rcvtime = ticks;
2244 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2245 * because the socket may be bound to an IPv6 wildcard address,
2246 * which may match an IPv4-mapped IPv6 address.
2248 inp->inp_ip_ttl = V_ip_defttl;
2251 * If using hpts lets drop a random number in so
2252 * not all new connections fall on the same CPU.
2254 inp->inp_hpts_cpu = hpts_random_cpu(inp);
2258 * Init the TCP PCAP queues.
2260 tcp_pcap_tcpcb_init(tp);
2263 /* Initialize the per-TCPCB log data. */
2264 tcp_log_tcpcbinit(tp);
2266 tp->t_pacing_rate = -1;
2267 if (tp->t_fb->tfb_tcp_fb_init) {
2268 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2269 refcount_release(&tp->t_fb->tfb_refcnt);
2274 if (V_tcp_perconn_stats_enable == 1)
2275 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2278 tp->t_flags |= TF_LRD;
2284 * Drop a TCP connection, reporting
2285 * the specified error. If connection is synchronized,
2286 * then send a RST to peer.
2289 tcp_drop(struct tcpcb *tp, int errno)
2291 struct socket *so = tptosocket(tp);
2294 INP_WLOCK_ASSERT(tptoinpcb(tp));
2296 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2297 tcp_state_change(tp, TCPS_CLOSED);
2298 /* Don't use tcp_output() here due to possible recursion. */
2299 (void)tcp_output_nodrop(tp);
2300 TCPSTAT_INC(tcps_drops);
2302 TCPSTAT_INC(tcps_conndrops);
2303 if (errno == ETIMEDOUT && tp->t_softerror)
2304 errno = tp->t_softerror;
2305 so->so_error = errno;
2306 return (tcp_close(tp));
2310 tcp_discardcb(struct tcpcb *tp)
2312 struct inpcb *inp = tptoinpcb(tp);
2313 struct socket *so = tptosocket(tp);
2315 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2318 INP_WLOCK_ASSERT(inp);
2321 if (tp->t_fb->tfb_tcp_timer_stop_all) {
2322 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2325 /* free the reassembly queue, if any */
2326 tcp_reass_flush(tp);
2329 /* Disconnect offload device, if any. */
2330 if (tp->t_flags & TF_TOE)
2331 tcp_offload_detach(tp);
2334 tcp_free_sackholes(tp);
2337 /* Free the TCP PCAP queues. */
2338 tcp_pcap_drain(&(tp->t_inpkts));
2339 tcp_pcap_drain(&(tp->t_outpkts));
2342 /* Allow the CC algorithm to clean up after itself. */
2343 if (CC_ALGO(tp)->cb_destroy != NULL)
2344 CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2346 /* Detach from the CC algorithm */
2350 khelp_destroy_osd(&tp->t_osd);
2353 stats_blob_destroy(tp->t_stats);
2359 tcp_log_tcpcbfini(tp);
2361 TCPSTATES_DEC(tp->t_state);
2362 if (tp->t_fb->tfb_tcp_fb_fini)
2363 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2366 * If we got enough samples through the srtt filter,
2367 * save the rtt and rttvar in the routing entry.
2368 * 'Enough' is arbitrarily defined as 4 rtt samples.
2369 * 4 samples is enough for the srtt filter to converge
2370 * to within enough % of the correct value; fewer samples
2371 * and we could save a bogus rtt. The danger is not high
2372 * as tcp quickly recovers from everything.
2373 * XXX: Works very well but needs some more statistics!
2375 * XXXRRS: Updating must be after the stack fini() since
2376 * that may be converting some internal representation of
2377 * say srtt etc into the general one used by other stacks.
2378 * Lets also at least protect against the so being NULL
2379 * as RW stated below.
2381 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2382 struct hc_metrics_lite metrics;
2385 bzero(&metrics, sizeof(metrics));
2387 * Update the ssthresh always when the conditions below
2388 * are satisfied. This gives us better new start value
2389 * for the congestion avoidance for new connections.
2390 * ssthresh is only set if packet loss occurred on a session.
2392 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2393 * being torn down. Ideally this code would not use 'so'.
2395 ssthresh = tp->snd_ssthresh;
2396 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2398 * convert the limit from user data bytes to
2399 * packets then to packet data bytes.
2401 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2404 ssthresh *= (tp->t_maxseg +
2406 (isipv6 ? sizeof (struct ip6_hdr) +
2407 sizeof (struct tcphdr) :
2409 sizeof (struct tcpiphdr)
2416 metrics.rmx_ssthresh = ssthresh;
2418 metrics.rmx_rtt = tp->t_srtt;
2419 metrics.rmx_rttvar = tp->t_rttvar;
2420 metrics.rmx_cwnd = tp->snd_cwnd;
2421 metrics.rmx_sendpipe = 0;
2422 metrics.rmx_recvpipe = 0;
2424 tcp_hc_update(&inp->inp_inc, &metrics);
2427 refcount_release(&tp->t_fb->tfb_refcnt);
2431 * Attempt to close a TCP control block, marking it as dropped, and freeing
2432 * the socket if we hold the only reference.
2435 tcp_close(struct tcpcb *tp)
2437 struct inpcb *inp = tptoinpcb(tp);
2438 struct socket *so = tptosocket(tp);
2440 INP_WLOCK_ASSERT(inp);
2443 if (tp->t_state == TCPS_LISTEN)
2444 tcp_offload_listen_stop(tp);
2447 * This releases the TFO pending counter resource for TFO listen
2448 * sockets as well as passively-created TFO sockets that transition
2449 * from SYN_RECEIVED to CLOSED.
2451 if (tp->t_tfo_pending) {
2452 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2453 tp->t_tfo_pending = NULL;
2456 tcp_hpts_remove(inp);
2459 TCPSTAT_INC(tcps_closed);
2460 if (tp->t_state != TCPS_CLOSED)
2461 tcp_state_change(tp, TCPS_CLOSED);
2462 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2463 soisdisconnected(so);
2464 if (inp->inp_flags & INP_SOCKREF) {
2465 inp->inp_flags &= ~INP_SOCKREF;
2474 * Notify a tcp user of an asynchronous error;
2475 * store error as soft error, but wake up user
2476 * (for now, won't do anything until can select for soft error).
2478 * Do not wake up user since there currently is no mechanism for
2479 * reporting soft errors (yet - a kqueue filter may be added).
2481 static struct inpcb *
2482 tcp_notify(struct inpcb *inp, int error)
2486 INP_WLOCK_ASSERT(inp);
2488 tp = intotcpcb(inp);
2489 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2492 * Ignore some errors if we are hooked up.
2493 * If connection hasn't completed, has retransmitted several times,
2494 * and receives a second error, give up now. This is better
2495 * than waiting a long time to establish a connection that
2496 * can never complete.
2498 if (tp->t_state == TCPS_ESTABLISHED &&
2499 (error == EHOSTUNREACH || error == ENETUNREACH ||
2500 error == EHOSTDOWN)) {
2501 if (inp->inp_route.ro_nh) {
2502 NH_FREE(inp->inp_route.ro_nh);
2503 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2506 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2508 tp = tcp_drop(tp, error);
2514 tp->t_softerror = error;
2518 wakeup( &so->so_timeo);
2525 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2527 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2528 INPLOOKUP_RLOCKPCB);
2533 if (req->newptr != NULL)
2536 if (req->oldptr == NULL) {
2539 n = V_tcbinfo.ipi_count +
2540 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2541 n += imax(n / 8, 10);
2542 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2546 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2549 bzero(&xig, sizeof(xig));
2550 xig.xig_len = sizeof xig;
2551 xig.xig_count = V_tcbinfo.ipi_count +
2552 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2553 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2554 xig.xig_sogen = so_gencnt;
2555 error = SYSCTL_OUT(req, &xig, sizeof xig);
2559 error = syncache_pcblist(req);
2563 while ((inp = inp_next(&inpi)) != NULL) {
2564 if (inp->inp_gencnt <= xig.xig_gen &&
2565 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2568 tcp_inptoxtp(inp, &xt);
2569 error = SYSCTL_OUT(req, &xt, sizeof xt);
2580 * Give the user an updated idea of our state.
2581 * If the generation differs from what we told
2582 * her before, she knows that something happened
2583 * while we were processing this request, and it
2584 * might be necessary to retry.
2586 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2587 xig.xig_sogen = so_gencnt;
2588 xig.xig_count = V_tcbinfo.ipi_count +
2589 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2590 error = SYSCTL_OUT(req, &xig, sizeof xig);
2596 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2597 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2598 NULL, 0, tcp_pcblist, "S,xtcpcb",
2599 "List of active TCP connections");
2603 tcp_getcred(SYSCTL_HANDLER_ARGS)
2606 struct sockaddr_in addrs[2];
2607 struct epoch_tracker et;
2611 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2614 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2617 NET_EPOCH_ENTER(et);
2618 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2619 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2623 error = cr_canseeinpcb(req->td->td_ucred, inp);
2625 cru2x(inp->inp_cred, &xuc);
2630 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2634 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2635 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2636 0, 0, tcp_getcred, "S,xucred",
2637 "Get the xucred of a TCP connection");
2642 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2644 struct epoch_tracker et;
2646 struct sockaddr_in6 addrs[2];
2653 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2656 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2659 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2660 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2663 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2665 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2672 NET_EPOCH_ENTER(et);
2675 inp = in_pcblookup(&V_tcbinfo,
2676 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2678 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2679 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2682 inp = in6_pcblookup(&V_tcbinfo,
2683 &addrs[1].sin6_addr, addrs[1].sin6_port,
2684 &addrs[0].sin6_addr, addrs[0].sin6_port,
2685 INPLOOKUP_RLOCKPCB, NULL);
2689 error = cr_canseeinpcb(req->td->td_ucred, inp);
2691 cru2x(inp->inp_cred, &xuc);
2696 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2700 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2701 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2702 0, 0, tcp6_getcred, "S,xucred",
2703 "Get the xucred of a TCP6 connection");
2707 /* Path MTU to try next when a fragmentation-needed message is received. */
2709 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2711 int mtu = ntohs(icp->icmp_nextmtu);
2713 /* If no alternative MTU was proposed, try the next smaller one. */
2715 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2716 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2717 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2723 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2729 struct inpcb *(*notify)(struct inpcb *, int);
2730 struct in_conninfo inc;
2731 tcp_seq icmp_tcp_seq;
2734 errno = icmp_errmap(icp);
2739 notify = tcp_mtudisc_notify;
2743 notify = tcp_drop_syn_sent;
2745 notify = tcp_notify;
2748 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2749 notify = tcp_drop_syn_sent;
2751 notify = tcp_notify;
2754 notify = tcp_notify;
2758 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2759 icmp_tcp_seq = th->th_seq;
2760 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2761 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2763 tp = intotcpcb(inp);
2765 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2767 * MTU discovery for offloaded connections. Let
2768 * the TOE driver verify seq# and process it.
2770 mtu = tcp_next_pmtu(icp, ip);
2771 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2775 if (tp->t_port != port)
2777 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2778 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2779 if (errno == EMSGSIZE) {
2781 * MTU discovery: we got a needfrag and
2782 * will potentially try a lower MTU.
2784 mtu = tcp_next_pmtu(icp, ip);
2787 * Only process the offered MTU if it
2788 * is smaller than the current one.
2790 if (mtu < tp->t_maxseg +
2791 sizeof(struct tcpiphdr)) {
2792 bzero(&inc, sizeof(inc));
2793 inc.inc_faddr = ip->ip_dst;
2795 inp->inp_inc.inc_fibnum;
2796 tcp_hc_updatemtu(&inc, mtu);
2797 inp = tcp_mtudisc(inp, mtu);
2800 inp = (*notify)(inp, errno);
2803 bzero(&inc, sizeof(inc));
2804 inc.inc_fport = th->th_dport;
2805 inc.inc_lport = th->th_sport;
2806 inc.inc_faddr = ip->ip_dst;
2807 inc.inc_laddr = ip->ip_src;
2808 syncache_unreach(&inc, icmp_tcp_seq, port);
2816 tcp_ctlinput(struct icmp *icmp)
2818 tcp_ctlinput_with_port(icmp, htons(0));
2822 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2824 /* Its a tunneled TCP over UDP icmp */
2825 struct icmp *icmp = param.icmp;
2826 struct ip *outer_ip, *inner_ip;
2828 struct tcphdr *th, ttemp;
2832 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2833 inner_ip = &icmp->icmp_ip;
2834 i_hlen = inner_ip->ip_hl << 2;
2835 o_len = ntohs(outer_ip->ip_len);
2837 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2838 /* Not enough data present */
2841 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2842 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2843 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2846 port = udp->uh_dport;
2847 th = (struct tcphdr *)(udp + 1);
2848 memcpy(&ttemp, th, sizeof(struct tcphdr));
2849 memcpy(udp, &ttemp, sizeof(struct tcphdr));
2850 /* Now adjust down the size of the outer IP header */
2851 o_len -= sizeof(struct udphdr);
2852 outer_ip->ip_len = htons(o_len);
2853 /* Now call in to the normal handling code */
2854 tcp_ctlinput_with_port(icmp, port);
2860 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2862 int mtu = ntohl(icmp6->icmp6_mtu);
2865 * If no alternative MTU was proposed, or the proposed MTU was too
2866 * small, set to the min.
2868 if (mtu < IPV6_MMTU)
2869 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */
2874 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2876 struct in6_addr *dst;
2877 struct inpcb *(*notify)(struct inpcb *, int);
2878 struct ip6_hdr *ip6;
2882 struct icmp6_hdr *icmp6;
2883 struct in_conninfo inc;
2888 tcp_seq icmp_tcp_seq;
2893 icmp6 = ip6cp->ip6c_icmp6;
2895 ip6 = ip6cp->ip6c_ip6;
2896 off = ip6cp->ip6c_off;
2897 dst = &ip6cp->ip6c_finaldst->sin6_addr;
2899 errno = icmp6_errmap(icmp6);
2904 notify = tcp_mtudisc_notify;
2908 notify = tcp_drop_syn_sent;
2910 notify = tcp_notify;
2914 * There are only four ICMPs that may reset connection:
2915 * - administratively prohibited
2916 * - port unreachable
2917 * - time exceeded in transit
2918 * - unknown next header
2920 if (V_icmp_may_rst &&
2921 ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
2922 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
2923 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
2924 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
2925 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
2926 (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
2927 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
2928 notify = tcp_drop_syn_sent;
2930 notify = tcp_notify;
2933 notify = tcp_notify;
2936 /* Check if we can safely get the ports from the tcp hdr */
2939 (int32_t) (off + sizeof(struct tcp_ports)))) {
2942 bzero(&t_ports, sizeof(struct tcp_ports));
2943 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2944 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2945 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2946 off += sizeof(struct tcp_ports);
2947 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2950 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2952 tp = intotcpcb(inp);
2954 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2955 /* MTU discovery for offloaded connections. */
2956 mtu = tcp6_next_pmtu(icmp6);
2957 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2961 if (tp->t_port != port)
2963 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2964 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2965 if (errno == EMSGSIZE) {
2968 * If we got a needfrag set the MTU
2969 * in the route to the suggested new
2970 * value (if given) and then notify.
2972 mtu = tcp6_next_pmtu(icmp6);
2974 bzero(&inc, sizeof(inc));
2975 inc.inc_fibnum = M_GETFIB(m);
2976 inc.inc_flags |= INC_ISIPV6;
2977 inc.inc6_faddr = *dst;
2978 if (in6_setscope(&inc.inc6_faddr,
2979 m->m_pkthdr.rcvif, NULL))
2982 * Only process the offered MTU if it
2983 * is smaller than the current one.
2985 if (mtu < tp->t_maxseg +
2986 sizeof (struct tcphdr) +
2987 sizeof (struct ip6_hdr)) {
2988 tcp_hc_updatemtu(&inc, mtu);
2989 tcp_mtudisc(inp, mtu);
2990 ICMP6STAT_INC(icp6s_pmtuchg);
2993 inp = (*notify)(inp, errno);
2996 bzero(&inc, sizeof(inc));
2997 inc.inc_fibnum = M_GETFIB(m);
2998 inc.inc_flags |= INC_ISIPV6;
2999 inc.inc_fport = t_ports.th_dport;
3000 inc.inc_lport = t_ports.th_sport;
3001 inc.inc6_faddr = *dst;
3002 inc.inc6_laddr = ip6->ip6_src;
3003 syncache_unreach(&inc, icmp_tcp_seq, port);
3011 tcp6_ctlinput(struct ip6ctlparam *ctl)
3013 tcp6_ctlinput_with_port(ctl, htons(0));
3017 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3019 struct ip6ctlparam *ip6cp = param.ip6cp;
3024 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3028 udp = mtod(m, struct udphdr *);
3029 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3032 port = udp->uh_dport;
3033 m_adj(m, sizeof(struct udphdr));
3034 if ((m->m_flags & M_PKTHDR) == 0) {
3035 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3037 /* Now call in to the normal handling code */
3038 tcp6_ctlinput_with_port(ip6cp, port);
3044 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3049 KASSERT(len >= SIPHASH_KEY_LENGTH,
3050 ("%s: keylen %u too short ", __func__, len));
3051 SipHash24_Init(&ctx);
3052 SipHash_SetKey(&ctx, (uint8_t *)key);
3053 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3054 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3055 switch (inc->inc_flags & INC_ISIPV6) {
3058 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3059 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3064 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3065 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3069 SipHash_Final((uint8_t *)hash, &ctx);
3071 return (hash[0] ^ hash[1]);
3075 tcp_new_ts_offset(struct in_conninfo *inc)
3077 struct in_conninfo inc_store, *local_inc;
3079 if (!V_tcp_ts_offset_per_conn) {
3080 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3081 inc_store.inc_lport = 0;
3082 inc_store.inc_fport = 0;
3083 local_inc = &inc_store;
3087 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3088 sizeof(V_ts_offset_secret)));
3092 * Following is where TCP initial sequence number generation occurs.
3094 * There are two places where we must use initial sequence numbers:
3095 * 1. In SYN-ACK packets.
3096 * 2. In SYN packets.
3098 * All ISNs for SYN-ACK packets are generated by the syncache. See
3099 * tcp_syncache.c for details.
3101 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3102 * depends on this property. In addition, these ISNs should be
3103 * unguessable so as to prevent connection hijacking. To satisfy
3104 * the requirements of this situation, the algorithm outlined in
3105 * RFC 1948 is used, with only small modifications.
3107 * Implementation details:
3109 * Time is based off the system timer, and is corrected so that it
3110 * increases by one megabyte per second. This allows for proper
3111 * recycling on high speed LANs while still leaving over an hour
3114 * As reading the *exact* system time is too expensive to be done
3115 * whenever setting up a TCP connection, we increment the time
3116 * offset in two ways. First, a small random positive increment
3117 * is added to isn_offset for each connection that is set up.
3118 * Second, the function tcp_isn_tick fires once per clock tick
3119 * and increments isn_offset as necessary so that sequence numbers
3120 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3121 * random positive increments serve only to ensure that the same
3122 * exact sequence number is never sent out twice (as could otherwise
3123 * happen when a port is recycled in less than the system tick
3126 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3127 * between seeding of isn_secret. This is normally set to zero,
3128 * as reseeding should not be necessary.
3130 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3131 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3132 * general, this means holding an exclusive (write) lock.
3135 #define ISN_BYTES_PER_SECOND 1048576
3136 #define ISN_STATIC_INCREMENT 4096
3137 #define ISN_RANDOM_INCREMENT (4096 - 1)
3138 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3140 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3141 VNET_DEFINE_STATIC(int, isn_last);
3142 VNET_DEFINE_STATIC(int, isn_last_reseed);
3143 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3144 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3146 #define V_isn_secret VNET(isn_secret)
3147 #define V_isn_last VNET(isn_last)
3148 #define V_isn_last_reseed VNET(isn_last_reseed)
3149 #define V_isn_offset VNET(isn_offset)
3150 #define V_isn_offset_old VNET(isn_offset_old)
3153 tcp_new_isn(struct in_conninfo *inc)
3156 u_int32_t projected_offset;
3159 /* Seed if this is the first use, reseed if requested. */
3160 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3161 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3163 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3164 V_isn_last_reseed = ticks;
3167 /* Compute the hash and return the ISN. */
3168 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3169 sizeof(V_isn_secret));
3170 V_isn_offset += ISN_STATIC_INCREMENT +
3171 (arc4random() & ISN_RANDOM_INCREMENT);
3172 if (ticks != V_isn_last) {
3173 projected_offset = V_isn_offset_old +
3174 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3175 if (SEQ_GT(projected_offset, V_isn_offset))
3176 V_isn_offset = projected_offset;
3177 V_isn_offset_old = V_isn_offset;
3180 new_isn += V_isn_offset;
3186 * When a specific ICMP unreachable message is received and the
3187 * connection state is SYN-SENT, drop the connection. This behavior
3188 * is controlled by the icmp_may_rst sysctl.
3190 static struct inpcb *
3191 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3196 INP_WLOCK_ASSERT(inp);
3198 tp = intotcpcb(inp);
3199 if (tp->t_state != TCPS_SYN_SENT)
3202 if (IS_FASTOPEN(tp->t_flags))
3203 tcp_fastopen_disable_path(tp);
3205 tp = tcp_drop(tp, errno);
3213 * When `need fragmentation' ICMP is received, update our idea of the MSS
3214 * based on the new value. Also nudge TCP to send something, since we
3215 * know the packet we just sent was dropped.
3216 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3218 static struct inpcb *
3219 tcp_mtudisc_notify(struct inpcb *inp, int error)
3222 return (tcp_mtudisc(inp, -1));
3225 static struct inpcb *
3226 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3231 INP_WLOCK_ASSERT(inp);
3233 tp = intotcpcb(inp);
3234 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3236 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3238 so = inp->inp_socket;
3239 SOCKBUF_LOCK(&so->so_snd);
3240 /* If the mss is larger than the socket buffer, decrease the mss. */
3241 if (so->so_snd.sb_hiwat < tp->t_maxseg)
3242 tp->t_maxseg = so->so_snd.sb_hiwat;
3243 SOCKBUF_UNLOCK(&so->so_snd);
3245 TCPSTAT_INC(tcps_mturesent);
3247 tp->snd_nxt = tp->snd_una;
3248 tcp_free_sackholes(tp);
3249 tp->snd_recover = tp->snd_max;
3250 if (tp->t_flags & TF_SACK_PERMIT)
3251 EXIT_FASTRECOVERY(tp->t_flags);
3252 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3254 * Conceptually the snd_nxt setting
3255 * and freeing sack holes should
3256 * be done by the default stacks
3257 * own tfb_tcp_mtu_chg().
3259 tp->t_fb->tfb_tcp_mtu_chg(tp);
3261 if (tcp_output(tp) < 0)
3269 * Look-up the routing entry to the peer of this inpcb. If no route
3270 * is found and it cannot be allocated, then return 0. This routine
3271 * is called by TCP routines that access the rmx structure and by
3272 * tcp_mss_update to get the peer/interface MTU.
3275 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3277 struct nhop_object *nh;
3279 uint32_t maxmtu = 0;
3281 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3283 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3284 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3289 maxmtu = nh->nh_mtu;
3291 /* Report additional interface capabilities. */
3293 if (ifp->if_capenable & IFCAP_TSO4 &&
3294 ifp->if_hwassist & CSUM_TSO) {
3295 cap->ifcap |= CSUM_TSO;
3296 cap->tsomax = ifp->if_hw_tsomax;
3297 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3298 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3308 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3310 struct nhop_object *nh;
3311 struct in6_addr dst6;
3314 uint32_t maxmtu = 0;
3316 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3318 if (inc->inc_flags & INC_IPV6MINMTU)
3321 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3322 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3323 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3328 maxmtu = nh->nh_mtu;
3330 /* Report additional interface capabilities. */
3332 if (ifp->if_capenable & IFCAP_TSO6 &&
3333 ifp->if_hwassist & CSUM_TSO) {
3334 cap->ifcap |= CSUM_TSO;
3335 cap->tsomax = ifp->if_hw_tsomax;
3336 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3337 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3346 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3348 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3349 * The right place to do that is ip6_setpktopt() that has just been
3350 * executed. By the way it just filled ip6po_minmtu for us.
3353 tcp6_use_min_mtu(struct tcpcb *tp)
3355 struct inpcb *inp = tptoinpcb(tp);
3357 INP_WLOCK_ASSERT(inp);
3359 * In case of the IPV6_USE_MIN_MTU socket
3360 * option, the INC_IPV6MINMTU flag to announce
3361 * a corresponding MSS during the initial
3362 * handshake. If the TCP connection is not in
3363 * the front states, just reduce the MSS being
3364 * used. This avoids the sending of TCP
3365 * segments which will be fragmented at the
3368 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3369 if ((tp->t_state >= TCPS_SYN_SENT) &&
3370 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3371 struct ip6_pktopts *opt;
3373 opt = inp->in6p_outputopts;
3374 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3375 tp->t_maxseg > TCP6_MSS)
3376 tp->t_maxseg = TCP6_MSS;
3382 * Calculate effective SMSS per RFC5681 definition for a given TCP
3383 * connection at its current state, taking into account SACK and etc.
3386 tcp_maxseg(const struct tcpcb *tp)
3390 if (tp->t_flags & TF_NOOPT)
3391 return (tp->t_maxseg);
3394 * Here we have a simplified code from tcp_addoptions(),
3395 * without a proper loop, and having most of paddings hardcoded.
3396 * We might make mistakes with padding here in some edge cases,
3397 * but this is harmless, since result of tcp_maxseg() is used
3398 * only in cwnd and ssthresh estimations.
3400 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3401 if (tp->t_flags & TF_RCVD_TSTMP)
3402 optlen = TCPOLEN_TSTAMP_APPA;
3405 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3406 if (tp->t_flags & TF_SIGNATURE)
3407 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3409 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3410 optlen += TCPOLEN_SACKHDR;
3411 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3412 optlen = PADTCPOLEN(optlen);
3415 if (tp->t_flags & TF_REQ_TSTMP)
3416 optlen = TCPOLEN_TSTAMP_APPA;
3418 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3419 if (tp->t_flags & TF_REQ_SCALE)
3420 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3421 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3422 if (tp->t_flags & TF_SIGNATURE)
3423 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3425 if (tp->t_flags & TF_SACK_PERMIT)
3426 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3429 optlen = min(optlen, TCP_MAXOLEN);
3430 return (tp->t_maxseg - optlen);
3435 tcp_fixed_maxseg(const struct tcpcb *tp)
3439 if (tp->t_flags & TF_NOOPT)
3440 return (tp->t_maxseg);
3443 * Here we have a simplified code from tcp_addoptions(),
3444 * without a proper loop, and having most of paddings hardcoded.
3445 * We only consider fixed options that we would send every
3446 * time I.e. SACK is not considered. This is important
3447 * for cc modules to figure out what the modulo of the
3450 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3451 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3452 if (tp->t_flags & TF_RCVD_TSTMP)
3453 optlen = TCPOLEN_TSTAMP_APPA;
3456 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3457 if (tp->t_flags & TF_SIGNATURE)
3458 optlen += PAD(TCPOLEN_SIGNATURE);
3461 if (tp->t_flags & TF_REQ_TSTMP)
3462 optlen = TCPOLEN_TSTAMP_APPA;
3464 optlen = PAD(TCPOLEN_MAXSEG);
3465 if (tp->t_flags & TF_REQ_SCALE)
3466 optlen += PAD(TCPOLEN_WINDOW);
3467 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3468 if (tp->t_flags & TF_SIGNATURE)
3469 optlen += PAD(TCPOLEN_SIGNATURE);
3471 if (tp->t_flags & TF_SACK_PERMIT)
3472 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3475 optlen = min(optlen, TCP_MAXOLEN);
3476 return (tp->t_maxseg - optlen);
3482 sysctl_drop(SYSCTL_HANDLER_ARGS)
3484 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3485 struct sockaddr_storage addrs[2];
3489 struct sockaddr_in *fin = NULL, *lin = NULL;
3491 struct epoch_tracker et;
3493 struct sockaddr_in6 *fin6, *lin6;
3503 if (req->oldptr != NULL || req->oldlen != 0)
3505 if (req->newptr == NULL)
3507 if (req->newlen < sizeof(addrs))
3509 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3513 switch (addrs[0].ss_family) {
3516 fin6 = (struct sockaddr_in6 *)&addrs[0];
3517 lin6 = (struct sockaddr_in6 *)&addrs[1];
3518 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3519 lin6->sin6_len != sizeof(struct sockaddr_in6))
3521 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3522 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3524 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3525 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3527 fin = (struct sockaddr_in *)&addrs[0];
3528 lin = (struct sockaddr_in *)&addrs[1];
3532 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3535 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3542 fin = (struct sockaddr_in *)&addrs[0];
3543 lin = (struct sockaddr_in *)&addrs[1];
3544 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3545 lin->sin_len != sizeof(struct sockaddr_in))
3552 NET_EPOCH_ENTER(et);
3553 switch (addrs[0].ss_family) {
3556 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3557 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3558 INPLOOKUP_WLOCKPCB, NULL);
3563 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3564 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3569 if (!SOLISTENING(inp->inp_socket)) {
3570 tp = intotcpcb(inp);
3571 tp = tcp_drop(tp, ECONNABORTED);
3582 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3583 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3584 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3585 "Drop TCP connection");
3588 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3590 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3591 &tcp_ctloutput_set));
3594 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3595 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3596 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3597 "Set socket option for TCP endpoint");
3601 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3603 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3604 struct sockaddr_storage addrs[2];
3607 struct sockaddr_in *fin = NULL, *lin = NULL;
3609 struct epoch_tracker et;
3611 struct sockaddr_in6 *fin6, *lin6;
3621 if (req->oldptr != NULL || req->oldlen != 0)
3623 if (req->newptr == NULL)
3625 if (req->newlen < sizeof(addrs))
3627 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3631 switch (addrs[0].ss_family) {
3634 fin6 = (struct sockaddr_in6 *)&addrs[0];
3635 lin6 = (struct sockaddr_in6 *)&addrs[1];
3636 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3637 lin6->sin6_len != sizeof(struct sockaddr_in6))
3639 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3640 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3642 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3643 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3645 fin = (struct sockaddr_in *)&addrs[0];
3646 lin = (struct sockaddr_in *)&addrs[1];
3650 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3653 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3660 fin = (struct sockaddr_in *)&addrs[0];
3661 lin = (struct sockaddr_in *)&addrs[1];
3662 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3663 lin->sin_len != sizeof(struct sockaddr_in))
3670 NET_EPOCH_ENTER(et);
3671 switch (addrs[0].ss_family) {
3674 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3675 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3676 INPLOOKUP_WLOCKPCB, NULL);
3681 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3682 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3690 so = inp->inp_socket;
3692 error = ktls_set_tx_mode(so,
3693 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3701 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3702 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3703 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3704 "Switch TCP connection to SW TLS");
3705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3706 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3707 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3708 "Switch TCP connection to ifnet TLS");
3712 * Generate a standardized TCP log line for use throughout the
3713 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3714 * allow use in the interrupt context.
3716 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3717 * NB: The function may return NULL if memory allocation failed.
3719 * Due to header inclusion and ordering limitations the struct ip
3720 * and ip6_hdr pointers have to be passed as void pointers.
3723 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3727 /* Is logging enabled? */
3728 if (V_tcp_log_in_vain == 0)
3731 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3735 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3739 /* Is logging enabled? */
3740 if (tcp_log_debug == 0)
3743 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3747 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3753 const struct ip *ip = (const struct ip *)ip4hdr;
3756 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3760 * The log line looks like this:
3761 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3763 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3764 sizeof(PRINT_TH_FLAGS) + 1 +
3766 2 * INET6_ADDRSTRLEN;
3768 2 * INET_ADDRSTRLEN;
3771 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3775 strcat(s, "TCP: [");
3778 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3779 inet_ntoa_r(inc->inc_faddr, sp);
3781 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3783 inet_ntoa_r(inc->inc_laddr, sp);
3785 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3788 ip6_sprintf(sp, &inc->inc6_faddr);
3790 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3792 ip6_sprintf(sp, &inc->inc6_laddr);
3794 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3795 } else if (ip6 && th) {
3796 ip6_sprintf(sp, &ip6->ip6_src);
3798 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3800 ip6_sprintf(sp, &ip6->ip6_dst);
3802 sprintf(sp, "]:%i", ntohs(th->th_dport));
3805 } else if (ip && th) {
3806 inet_ntoa_r(ip->ip_src, sp);
3808 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3810 inet_ntoa_r(ip->ip_dst, sp);
3812 sprintf(sp, "]:%i", ntohs(th->th_dport));
3820 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3821 if (*(s + size - 1) != '\0')
3822 panic("%s: string too long", __func__);
3827 * A subroutine which makes it easy to track TCP state changes with DTrace.
3828 * This function shouldn't be called for t_state initializations that don't
3829 * correspond to actual TCP state transitions.
3832 tcp_state_change(struct tcpcb *tp, int newstate)
3834 #if defined(KDTRACE_HOOKS)
3835 int pstate = tp->t_state;
3838 TCPSTATES_DEC(tp->t_state);
3839 TCPSTATES_INC(newstate);
3840 tp->t_state = newstate;
3841 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3845 * Create an external-format (``xtcpcb'') structure using the information in
3846 * the kernel-format tcpcb structure pointed to by tp. This is done to
3847 * reduce the spew of irrelevant information over this interface, to isolate
3848 * user code from changes in the kernel structure, and potentially to provide
3849 * information-hiding if we decide that some of this information should be
3850 * hidden from users.
3853 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3855 struct tcpcb *tp = intotcpcb(inp);
3858 bzero(xt, sizeof(*xt));
3859 xt->t_state = tp->t_state;
3860 xt->t_logstate = tcp_get_bblog_state(tp);
3861 xt->t_flags = tp->t_flags;
3862 xt->t_sndzerowin = tp->t_sndzerowin;
3863 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3864 xt->t_rcvoopack = tp->t_rcvoopack;
3865 xt->t_rcv_wnd = tp->rcv_wnd;
3866 xt->t_snd_wnd = tp->snd_wnd;
3867 xt->t_snd_cwnd = tp->snd_cwnd;
3868 xt->t_snd_ssthresh = tp->snd_ssthresh;
3869 xt->t_dsack_bytes = tp->t_dsack_bytes;
3870 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3871 xt->t_dsack_pack = tp->t_dsack_pack;
3872 xt->t_maxseg = tp->t_maxseg;
3873 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3874 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3876 now = getsbinuptime();
3877 #define COPYTIMER(which,where) do { \
3878 if (tp->t_timers[which] != SBT_MAX) \
3879 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \
3883 COPYTIMER(TT_DELACK, tt_delack);
3884 COPYTIMER(TT_REXMT, tt_rexmt);
3885 COPYTIMER(TT_PERSIST, tt_persist);
3886 COPYTIMER(TT_KEEP, tt_keep);
3887 COPYTIMER(TT_2MSL, tt_2msl);
3889 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3891 xt->xt_encaps_port = tp->t_port;
3892 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3893 TCP_FUNCTION_NAME_LEN_MAX);
3894 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3896 (void)tcp_log_get_id(tp, xt->xt_logid);
3899 xt->xt_len = sizeof(struct xtcpcb);
3900 in_pcbtoxinpcb(inp, &xt->xt_inp);
3904 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3909 (status > TCP_EI_STATUS_MAX_VALUE) ||
3914 if (status > (sizeof(uint32_t) * 8)) {
3915 /* Should this be a KASSERT? */
3918 bit = 1U << (status - 1);
3919 if (bit & tp->t_end_info_status) {
3920 /* already logged */
3923 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3924 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3925 tp->t_end_info_bytes[i] = status;
3926 tp->t_end_info_status |= bit;
3933 tcp_can_enable_pacing(void)
3936 if ((tcp_pacing_limit == -1) ||
3937 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
3938 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
3939 shadow_num_connections = number_of_tcp_connections_pacing;
3946 static uint8_t tcp_pacing_warning = 0;
3949 tcp_decrement_paced_conn(void)
3953 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
3954 shadow_num_connections = number_of_tcp_connections_pacing;
3955 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
3957 if (tcp_pacing_limit != -1) {
3958 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
3959 tcp_pacing_limit = 0;
3960 } else if (tcp_pacing_warning == 0) {
3961 printf("Warning pacing count is invalid, invalid decrement\n");
3962 tcp_pacing_warning = 1;
3967 #ifdef TCP_ACCOUNTING
3969 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss)
3971 if (SEQ_LT(th->th_ack, tp->snd_una)) {
3972 /* Do we have a SACK? */
3973 if (to->to_flags & TOF_SACK) {
3974 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
3975 tp->tcp_cnt_counters[ACK_SACK]++;
3979 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
3980 tp->tcp_cnt_counters[ACK_BEHIND]++;
3982 return (ACK_BEHIND);
3984 } else if (th->th_ack == tp->snd_una) {
3985 /* Do we have a SACK? */
3986 if (to->to_flags & TOF_SACK) {
3987 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
3988 tp->tcp_cnt_counters[ACK_SACK]++;
3991 } else if (tiwin != tp->snd_wnd) {
3992 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
3993 tp->tcp_cnt_counters[ACK_RWND]++;
3997 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
3998 tp->tcp_cnt_counters[ACK_DUPACK]++;
4000 return (ACK_DUPACK);
4003 if (!SEQ_GT(th->th_ack, tp->snd_max)) {
4004 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4005 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss);
4008 if (to->to_flags & TOF_SACK) {
4009 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4010 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++;
4012 return (ACK_CUMACK_SACK);
4014 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4015 tp->tcp_cnt_counters[ACK_CUMACK]++;
4017 return (ACK_CUMACK);