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"
41 #include "opt_tcpdebug.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
49 #include <sys/hhook.h>
51 #include <sys/kernel.h>
53 #include <sys/khelp.h>
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
66 #include <sys/domain.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
78 #include <net/route.h>
79 #include <net/route/nhop.h>
81 #include <net/if_var.h>
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
103 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_seq.h>
109 #include <netinet/tcp_timer.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet/tcp_log_buf.h>
112 #include <netinet/tcp_syncache.h>
113 #include <netinet/tcp_hpts.h>
114 #include <netinet/cc/cc.h>
116 #include <netinet6/tcp6_var.h>
118 #include <netinet/tcpip.h>
119 #include <netinet/tcp_fastopen.h>
121 #include <netinet/tcp_pcap.h>
124 #include <netinet/tcp_debug.h>
127 #include <netinet6/ip6protosw.h>
130 #include <netinet/tcp_offload.h>
132 #include <netinet/udp.h>
133 #include <netinet/udp_var.h>
135 #include <netipsec/ipsec_support.h>
137 #include <machine/in_cksum.h>
138 #include <crypto/siphash/siphash.h>
140 #include <security/mac/mac_framework.h>
142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
147 #ifdef NETFLIX_EXP_DETECTION
148 /* Sack attack detection thresholds and such */
149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
150 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
151 "Sack Attack detection thresholds");
152 int32_t tcp_force_detection = 0;
153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
155 &tcp_force_detection, 0,
156 "Do we force detection even if the INP has it off?");
157 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
160 &tcp_sack_to_ack_thresh, 700,
161 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
162 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
165 &tcp_sack_to_move_thresh, 600,
166 "Percentage of sack moves we must see above (10.1 percent is 101)");
167 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
170 &tcp_restoral_thresh, 550,
171 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
172 int32_t tcp_sad_decay_val = 800;
173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
175 &tcp_sad_decay_val, 800,
176 "The decay percentage (10.1 percent equals 101 )");
177 int32_t tcp_map_minimum = 500;
178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
180 &tcp_map_minimum, 500,
181 "Number of Map enteries before we start detection");
182 int32_t tcp_attack_on_turns_on_logging = 0;
183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
185 &tcp_attack_on_turns_on_logging, 0,
186 "When we have a positive hit on attack, do we turn on logging?");
187 int32_t tcp_sad_pacing_interval = 2000;
188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
190 &tcp_sad_pacing_interval, 2000,
191 "What is the minimum pacing interval for a classified attacker?");
193 int32_t tcp_sad_low_pps = 100;
194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
196 &tcp_sad_low_pps, 100,
197 "What is the input pps that below which we do not decay?");
199 uint32_t tcp_ack_war_time_window = 1000;
200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
202 &tcp_ack_war_time_window, 1000,
203 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
204 uint32_t tcp_ack_war_cnt = 5;
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
208 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
210 struct rwlock tcp_function_lock;
213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
218 error = sysctl_handle_int(oidp, &new, 0, req);
219 if (error == 0 && req->newptr) {
220 if (new < TCP_MINMSS)
228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
229 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
230 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
231 "Default TCP Maximum Segment Size");
235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
239 new = V_tcp_v6mssdflt;
240 error = sysctl_handle_int(oidp, &new, 0, req);
241 if (error == 0 && req->newptr) {
242 if (new < TCP_MINMSS)
245 V_tcp_v6mssdflt = new;
250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
251 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
252 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
253 "Default TCP Maximum Segment Size for IPv6");
257 * Minimum MSS we accept and use. This prevents DoS attacks where
258 * we are forced to a ridiculous low MSS like 20 and send hundreds
259 * of packets instead of one. The effect scales with the available
260 * bandwidth and quickly saturates the CPU and network interface
261 * with packet generation and sending. Set to zero to disable MINMSS
262 * checking. This setting prevents us from sending too small packets.
264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
266 &VNET_NAME(tcp_minmss), 0,
267 "Minimum TCP Maximum Segment Size");
269 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
271 &VNET_NAME(tcp_do_rfc1323), 0,
272 "Enable rfc1323 (high performance TCP) extensions");
275 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
276 * Some stacks negotiate TS, but never send them after connection setup. Some
277 * stacks negotiate TS, but don't send them when sending keep-alive segments.
278 * These include modern widely deployed TCP stacks.
279 * Therefore tolerating violations for now...
281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
283 &VNET_NAME(tcp_tolerate_missing_ts), 0,
284 "Tolerate missing TCP timestamps");
286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
288 &VNET_NAME(tcp_ts_offset_per_conn), 0,
289 "Initialize TCP timestamps per connection instead of per host pair");
291 /* How many connections are pacing */
292 static volatile uint32_t number_of_tcp_connections_pacing = 0;
293 static uint32_t shadow_num_connections = 0;
295 static int tcp_pacing_limit = 10000;
296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
297 &tcp_pacing_limit, 1000,
298 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
301 &shadow_num_connections, 0, "Number of TCP connections being paced");
303 static int tcp_log_debug = 0;
304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
305 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
307 static int tcp_tcbhashsize;
308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
309 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
311 static int do_tcpdrain = 1;
312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
313 "Enable tcp_drain routine for extra help when low on mbufs");
315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
316 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
319 #define V_icmp_may_rst VNET(icmp_may_rst)
320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
321 &VNET_NAME(icmp_may_rst), 0,
322 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
325 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
327 &VNET_NAME(tcp_isn_reseed_interval), 0,
328 "Seconds between reseeding of ISN secret");
330 static int tcp_soreceive_stream;
331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
332 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
334 VNET_DEFINE(uma_zone_t, sack_hole_zone);
335 #define V_sack_hole_zone VNET(sack_hole_zone)
336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
343 new = V_tcp_map_entries_limit;
344 error = sysctl_handle_int(oidp, &new, 0, req);
345 if (error == 0 && req->newptr) {
346 /* only allow "0" and value > minimum */
347 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
350 V_tcp_map_entries_limit = new;
354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
355 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
356 &VNET_NAME(tcp_map_entries_limit), 0,
357 &sysctl_net_inet_tcp_map_limit_check, "IU",
358 "Total sendmap entries limit");
360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
362 &VNET_NAME(tcp_map_split_limit), 0,
363 "Total sendmap split entries limit");
366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
371 #define V_ts_offset_secret VNET(ts_offset_secret)
373 static int tcp_default_fb_init(struct tcpcb *tp);
374 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
375 static int tcp_default_handoff_ok(struct tcpcb *tp);
376 static struct inpcb *tcp_notify(struct inpcb *, int);
377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
378 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
379 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
380 const void *ip4hdr, const void *ip6hdr);
382 static struct tcp_function_block tcp_def_funcblk = {
383 .tfb_tcp_block_name = "freebsd",
384 .tfb_tcp_output = tcp_default_output,
385 .tfb_tcp_do_segment = tcp_do_segment,
386 .tfb_tcp_ctloutput = tcp_default_ctloutput,
387 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
388 .tfb_tcp_fb_init = tcp_default_fb_init,
389 .tfb_tcp_fb_fini = tcp_default_fb_fini,
392 static int tcp_fb_cnt = 0;
393 struct tcp_funchead t_functions;
394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
399 TCPSTAT_INC(tcps_dsack_count);
402 if (SEQ_GT(end, start)) {
403 tp->t_dsack_bytes += (end - start);
404 TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
406 tp->t_dsack_tlp_bytes += (start - end);
407 TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
410 if (SEQ_GT(end, start)) {
411 tp->t_dsack_bytes += (end - start);
412 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
414 tp->t_dsack_tlp_bytes += (start - end);
415 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
420 static struct tcp_function_block *
421 find_tcp_functions_locked(struct tcp_function_set *fs)
423 struct tcp_function *f;
424 struct tcp_function_block *blk=NULL;
426 TAILQ_FOREACH(f, &t_functions, tf_next) {
427 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
435 static struct tcp_function_block *
436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
438 struct tcp_function_block *rblk=NULL;
439 struct tcp_function *f;
441 TAILQ_FOREACH(f, &t_functions, tf_next) {
442 if (f->tf_fb == blk) {
453 struct tcp_function_block *
454 find_and_ref_tcp_functions(struct tcp_function_set *fs)
456 struct tcp_function_block *blk;
458 rw_rlock(&tcp_function_lock);
459 blk = find_tcp_functions_locked(fs);
461 refcount_acquire(&blk->tfb_refcnt);
462 rw_runlock(&tcp_function_lock);
466 struct tcp_function_block *
467 find_and_ref_tcp_fb(struct tcp_function_block *blk)
469 struct tcp_function_block *rblk;
471 rw_rlock(&tcp_function_lock);
472 rblk = find_tcp_fb_locked(blk, NULL);
474 refcount_acquire(&rblk->tfb_refcnt);
475 rw_runlock(&tcp_function_lock);
479 /* Find a matching alias for the given tcp_function_block. */
481 find_tcp_function_alias(struct tcp_function_block *blk,
482 struct tcp_function_set *fs)
484 struct tcp_function *f;
488 rw_rlock(&tcp_function_lock);
489 TAILQ_FOREACH(f, &t_functions, tf_next) {
490 if ((f->tf_fb == blk) &&
491 (strncmp(f->tf_name, blk->tfb_tcp_block_name,
492 TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
493 /* Matching function block with different name. */
494 strncpy(fs->function_set_name, f->tf_name,
495 TCP_FUNCTION_NAME_LEN_MAX);
500 /* Null terminate the string appropriately. */
502 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
504 fs->function_set_name[0] = '\0';
506 rw_runlock(&tcp_function_lock);
510 static struct tcp_function_block *
511 find_and_ref_tcp_default_fb(void)
513 struct tcp_function_block *rblk;
515 rw_rlock(&tcp_function_lock);
516 rblk = tcp_func_set_ptr;
517 refcount_acquire(&rblk->tfb_refcnt);
518 rw_runlock(&tcp_function_lock);
523 tcp_switch_back_to_default(struct tcpcb *tp)
525 struct tcp_function_block *tfb;
527 KASSERT(tp->t_fb != &tcp_def_funcblk,
528 ("%s: called by the built-in default stack", __func__));
531 * Release the old stack. This function will either find a new one
534 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
535 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
536 refcount_release(&tp->t_fb->tfb_refcnt);
539 * Now, we'll find a new function block to use.
540 * Start by trying the current user-selected
541 * default, unless this stack is the user-selected
544 tfb = find_and_ref_tcp_default_fb();
545 if (tfb == tp->t_fb) {
546 refcount_release(&tfb->tfb_refcnt);
549 /* Does the stack accept this connection? */
550 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 refcount_release(&tfb->tfb_refcnt);
555 /* Try to use that stack. */
557 /* Initialize the new stack. If it succeeds, we are done. */
559 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
560 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
564 * Initialization failed. Release the reference count on
567 refcount_release(&tfb->tfb_refcnt);
571 * If that wasn't feasible, use the built-in default
572 * stack which is not allowed to reject anyone.
574 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
576 /* there always should be a default */
577 panic("Can't refer to tcp_def_funcblk");
579 if (tfb->tfb_tcp_handoff_ok != NULL) {
580 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
581 /* The default stack cannot say no */
582 panic("Default stack rejects a new session?");
586 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
587 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
588 /* The default stack cannot fail */
589 panic("Default stack initialization failed");
594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
595 const struct sockaddr *sa, void *ctx)
606 TCPSTAT_INC(tcps_tunneled_pkts);
607 if ((m->m_flags & M_PKTHDR) == 0) {
608 /* Can't handle one that is not a pkt hdr */
609 TCPSTAT_INC(tcps_tunneled_errs);
612 thlen = sizeof(struct tcphdr);
613 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
614 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
615 TCPSTAT_INC(tcps_tunneled_errs);
618 iph = mtod(m, struct ip *);
619 uh = (struct udphdr *)((caddr_t)iph + off);
620 th = (struct tcphdr *)(uh + 1);
621 thlen = th->th_off << 2;
622 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
623 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
625 TCPSTAT_INC(tcps_tunneled_errs);
628 iph = mtod(m, struct ip *);
629 uh = (struct udphdr *)((caddr_t)iph + off);
630 th = (struct tcphdr *)(uh + 1);
633 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
634 bcopy(th, uh, m->m_len - off);
635 m->m_len -= sizeof(struct udphdr);
636 m->m_pkthdr.len -= sizeof(struct udphdr);
638 * We use the same algorithm for
639 * both UDP and TCP for c-sum. So
640 * the code in tcp_input will skip
641 * the checksum. So we do nothing
642 * with the flag (m->m_pkthdr.csum_flags).
647 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
648 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
652 case IPV6_VERSION >> 4:
653 ip6 = mtod(m, struct ip6_hdr *);
654 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
655 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
670 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
673 struct tcp_function_set fs;
674 struct tcp_function_block *blk;
676 memset(&fs, 0, sizeof(fs));
677 rw_rlock(&tcp_function_lock);
678 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
681 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
682 fs.pcbcnt = blk->tfb_refcnt;
684 rw_runlock(&tcp_function_lock);
685 error = sysctl_handle_string(oidp, fs.function_set_name,
686 sizeof(fs.function_set_name), req);
688 /* Check for error or no change */
689 if (error != 0 || req->newptr == NULL)
692 rw_wlock(&tcp_function_lock);
693 blk = find_tcp_functions_locked(&fs);
695 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
699 tcp_func_set_ptr = blk;
701 rw_wunlock(&tcp_function_lock);
705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
706 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
707 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
708 "Set/get the default TCP functions");
711 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
713 int error, cnt, linesz;
714 struct tcp_function *f;
720 rw_rlock(&tcp_function_lock);
721 TAILQ_FOREACH(f, &t_functions, tf_next) {
724 rw_runlock(&tcp_function_lock);
726 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
727 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
732 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
733 "Alias", "PCB count");
738 rw_rlock(&tcp_function_lock);
739 TAILQ_FOREACH(f, &t_functions, tf_next) {
740 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
741 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
742 f->tf_fb->tfb_tcp_block_name,
743 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
744 alias ? f->tf_name : "-",
745 f->tf_fb->tfb_refcnt);
746 if (linesz >= bufsz) {
754 rw_runlock(&tcp_function_lock);
756 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
757 free(buffer, M_TEMP);
761 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
762 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
763 NULL, 0, sysctl_net_inet_list_available, "A",
764 "list available TCP Function sets");
766 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
769 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
770 #define V_udp4_tun_socket VNET(udp4_tun_socket)
773 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
774 #define V_udp6_tun_socket VNET(udp6_tun_socket)
778 tcp_over_udp_stop(void)
781 * This function assumes sysctl caller holds inp_rinfo_lock()
785 if (V_udp4_tun_socket != NULL) {
786 soclose(V_udp4_tun_socket);
787 V_udp4_tun_socket = NULL;
791 if (V_udp6_tun_socket != NULL) {
792 soclose(V_udp6_tun_socket);
793 V_udp6_tun_socket = NULL;
799 tcp_over_udp_start(void)
804 struct sockaddr_in sin;
807 struct sockaddr_in6 sin6;
810 * This function assumes sysctl caller holds inp_info_rlock()
813 port = V_tcp_udp_tunneling_port;
814 if (ntohs(port) == 0) {
815 /* Must have a port set */
819 if (V_udp4_tun_socket != NULL) {
820 /* Already running -- must stop first */
825 if (V_udp6_tun_socket != NULL) {
826 /* Already running -- must stop first */
831 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
832 SOCK_DGRAM, IPPROTO_UDP,
833 curthread->td_ucred, curthread))) {
837 /* Call the special UDP hook. */
838 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
839 tcp_recv_udp_tunneled_packet,
845 /* Ok, we have a socket, bind it to the port. */
846 memset(&sin, 0, sizeof(struct sockaddr_in));
847 sin.sin_len = sizeof(struct sockaddr_in);
848 sin.sin_family = AF_INET;
849 sin.sin_port = htons(port);
850 if ((ret = sobind(V_udp4_tun_socket,
851 (struct sockaddr *)&sin, curthread))) {
857 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
858 SOCK_DGRAM, IPPROTO_UDP,
859 curthread->td_ucred, curthread))) {
863 /* Call the special UDP hook. */
864 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
865 tcp_recv_udp_tunneled_packet,
866 tcp6_ctlinput_viaudp,
871 /* Ok, we have a socket, bind it to the port. */
872 memset(&sin6, 0, sizeof(struct sockaddr_in6));
873 sin6.sin6_len = sizeof(struct sockaddr_in6);
874 sin6.sin6_family = AF_INET6;
875 sin6.sin6_port = htons(port);
876 if ((ret = sobind(V_udp6_tun_socket,
877 (struct sockaddr *)&sin6, curthread))) {
886 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
891 old = V_tcp_udp_tunneling_port;
893 error = sysctl_handle_int(oidp, &new, 0, req);
895 (req->newptr != NULL)) {
896 if ((new < TCP_TUNNELING_PORT_MIN) ||
897 (new > TCP_TUNNELING_PORT_MAX)) {
900 V_tcp_udp_tunneling_port = new;
905 error = tcp_over_udp_start();
912 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
913 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
914 &VNET_NAME(tcp_udp_tunneling_port),
915 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
916 "Tunneling port for tcp over udp");
918 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
921 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
925 new = V_tcp_udp_tunneling_overhead;
926 error = sysctl_handle_int(oidp, &new, 0, req);
927 if (error == 0 && req->newptr) {
928 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
929 (new > TCP_TUNNELING_OVERHEAD_MAX))
932 V_tcp_udp_tunneling_overhead = new;
937 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
938 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
939 &VNET_NAME(tcp_udp_tunneling_overhead),
940 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
941 "MSS reduction when using tcp over udp");
944 * Exports one (struct tcp_function_info) for each alias/name.
947 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
950 struct tcp_function *f;
951 struct tcp_function_info tfi;
954 * We don't allow writes.
956 if (req->newptr != NULL)
960 * Wire the old buffer so we can directly copy the functions to
961 * user space without dropping the lock.
963 if (req->oldptr != NULL) {
964 error = sysctl_wire_old_buffer(req, 0);
970 * Walk the list and copy out matching entries. If INVARIANTS
971 * is compiled in, also walk the list to verify the length of
972 * the list matches what we have recorded.
974 rw_rlock(&tcp_function_lock);
978 if (req->oldptr == NULL) {
983 TAILQ_FOREACH(f, &t_functions, tf_next) {
987 if (req->oldptr != NULL) {
988 bzero(&tfi, sizeof(tfi));
989 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
990 tfi.tfi_id = f->tf_fb->tfb_id;
991 (void)strlcpy(tfi.tfi_alias, f->tf_name,
992 sizeof(tfi.tfi_alias));
993 (void)strlcpy(tfi.tfi_name,
994 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
995 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
997 * Don't stop on error, as that is the
998 * mechanism we use to accumulate length
999 * information if the buffer was too short.
1003 KASSERT(cnt == tcp_fb_cnt,
1004 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1008 rw_runlock(&tcp_function_lock);
1009 if (req->oldptr == NULL)
1010 error = SYSCTL_OUT(req, NULL,
1011 (cnt + 1) * sizeof(struct tcp_function_info));
1016 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1017 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1018 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1019 "List TCP function block name-to-ID mappings");
1022 * tfb_tcp_handoff_ok() function for the default stack.
1023 * Note that we'll basically try to take all comers.
1026 tcp_default_handoff_ok(struct tcpcb *tp)
1033 * tfb_tcp_fb_init() function for the default stack.
1035 * This handles making sure we have appropriate timers set if you are
1036 * transitioning a socket that has some amount of setup done.
1038 * The init() fuction from the default can *never* return non-zero i.e.
1039 * it is required to always succeed since it is the stack of last resort!
1042 tcp_default_fb_init(struct tcpcb *tp)
1047 INP_WLOCK_ASSERT(tp->t_inpcb);
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 so = tp->t_inpcb->inp_socket;
1065 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1066 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1067 tcp_timer_active(tp, TT_PERSIST))) {
1069 * If the session has established and it looks like it should
1070 * be in the persist state, set the persist timer. Otherwise,
1071 * set the retransmit timer.
1073 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1074 (int32_t)(tp->snd_nxt - tp->snd_una) <
1075 (int32_t)sbavail(&so->so_snd))
1078 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1081 /* All non-embryonic sessions get a keepalive timer. */
1082 if (!tcp_timer_active(tp, TT_KEEP))
1083 tcp_timer_activate(tp, TT_KEEP,
1084 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1088 * Make sure critical variables are initialized
1089 * if transitioning while in Recovery.
1091 if IN_FASTRECOVERY(tp->t_flags) {
1092 if (tp->sackhint.recover_fs == 0)
1093 tp->sackhint.recover_fs = max(1,
1094 tp->snd_nxt - tp->snd_una);
1101 * tfb_tcp_fb_fini() function for the default stack.
1103 * This changes state as necessary (or prudent) to prepare for another stack
1104 * to assume responsibility for the connection.
1107 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1110 INP_WLOCK_ASSERT(tp->t_inpcb);
1115 * Target size of TCP PCB hash tables. Must be a power of two.
1117 * Note that this can be overridden by the kernel environment
1118 * variable net.inet.tcp.tcbhashsize
1121 #define TCBHASHSIZE 0
1126 * Callouts should be moved into struct tcp directly. They are currently
1127 * separate because the tcpcb structure is exported to userland for sysctl
1128 * parsing purposes, which do not know about callouts.
1132 struct tcp_timer tt;
1139 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1140 #define V_tcpcb_zone VNET(tcpcb_zone)
1142 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1143 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1145 static struct mtx isn_mtx;
1147 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1148 #define ISN_LOCK() mtx_lock(&isn_mtx)
1149 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1151 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1154 * Take a value and get the next power of 2 that doesn't overflow.
1155 * Used to size the tcp_inpcb hash buckets.
1158 maketcp_hashsize(int size)
1164 * get the next power of 2 higher than maxsockets.
1166 hashsize = 1 << fls(size);
1167 /* catch overflow, and just go one power of 2 smaller */
1168 if (hashsize < size) {
1169 hashsize = 1 << (fls(size) - 1);
1174 static volatile int next_tcp_stack_id = 1;
1177 * Register a TCP function block with the name provided in the names
1178 * array. (Note that this function does NOT automatically register
1179 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1180 * explicitly include blk->tfb_tcp_block_name in the list of names if
1181 * you wish to register the stack with that name.)
1183 * Either all name registrations will succeed or all will fail. If
1184 * a name registration fails, the function will update the num_names
1185 * argument to point to the array index of the name that encountered
1188 * Returns 0 on success, or an error code on failure.
1191 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1192 const char *names[], int *num_names)
1194 struct tcp_function *n;
1195 struct tcp_function_set fs;
1198 KASSERT(names != NULL && *num_names > 0,
1199 ("%s: Called with 0-length name list", __func__));
1200 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1201 KASSERT(rw_initialized(&tcp_function_lock),
1202 ("%s: called too early", __func__));
1204 if ((blk->tfb_tcp_output == NULL) ||
1205 (blk->tfb_tcp_do_segment == NULL) ||
1206 (blk->tfb_tcp_ctloutput == NULL) ||
1207 (strlen(blk->tfb_tcp_block_name) == 0)) {
1209 * These functions are required and you
1215 if (blk->tfb_tcp_timer_stop_all ||
1216 blk->tfb_tcp_timer_activate ||
1217 blk->tfb_tcp_timer_active ||
1218 blk->tfb_tcp_timer_stop) {
1220 * If you define one timer function you
1221 * must have them all.
1223 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1224 (blk->tfb_tcp_timer_activate == NULL) ||
1225 (blk->tfb_tcp_timer_active == NULL) ||
1226 (blk->tfb_tcp_timer_stop == NULL)) {
1232 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1237 refcount_init(&blk->tfb_refcnt, 0);
1238 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1239 for (i = 0; i < *num_names; i++) {
1240 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1247 (void)strlcpy(fs.function_set_name, names[i],
1248 sizeof(fs.function_set_name));
1249 rw_wlock(&tcp_function_lock);
1250 if (find_tcp_functions_locked(&fs) != NULL) {
1251 /* Duplicate name space not allowed */
1252 rw_wunlock(&tcp_function_lock);
1253 free(n, M_TCPFUNCTIONS);
1257 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1258 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1260 rw_wunlock(&tcp_function_lock);
1266 * Deregister the names we just added. Because registration failed
1267 * for names[i], we don't need to deregister that name.
1270 rw_wlock(&tcp_function_lock);
1272 TAILQ_FOREACH(n, &t_functions, tf_next) {
1273 if (!strncmp(n->tf_name, names[i],
1274 TCP_FUNCTION_NAME_LEN_MAX)) {
1275 TAILQ_REMOVE(&t_functions, n, tf_next);
1278 free(n, M_TCPFUNCTIONS);
1283 rw_wunlock(&tcp_function_lock);
1288 * Register a TCP function block using the name provided in the name
1291 * Returns 0 on success, or an error code on failure.
1294 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1297 const char *name_list[1];
1302 name_list[0] = name;
1304 name_list[0] = blk->tfb_tcp_block_name;
1305 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1310 * Register a TCP function block using the name defined in
1311 * blk->tfb_tcp_block_name.
1313 * Returns 0 on success, or an error code on failure.
1316 register_tcp_functions(struct tcp_function_block *blk, int wait)
1319 return (register_tcp_functions_as_name(blk, NULL, wait));
1323 * Deregister all names associated with a function block. This
1324 * functionally removes the function block from use within the system.
1326 * When called with a true quiesce argument, mark the function block
1327 * as being removed so no more stacks will use it and determine
1328 * whether the removal would succeed.
1330 * When called with a false quiesce argument, actually attempt the
1333 * When called with a force argument, attempt to switch all TCBs to
1334 * use the default stack instead of returning EBUSY.
1336 * Returns 0 on success (or if the removal would succeed, or an error
1340 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1343 struct tcp_function *f;
1345 if (blk == &tcp_def_funcblk) {
1346 /* You can't un-register the default */
1349 rw_wlock(&tcp_function_lock);
1350 if (blk == tcp_func_set_ptr) {
1351 /* You can't free the current default */
1352 rw_wunlock(&tcp_function_lock);
1355 /* Mark the block so no more stacks can use it. */
1356 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1358 * If TCBs are still attached to the stack, attempt to switch them
1359 * to the default stack.
1361 if (force && blk->tfb_refcnt) {
1362 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1363 INPLOOKUP_WLOCKPCB);
1366 VNET_ITERATOR_DECL(vnet_iter);
1368 rw_wunlock(&tcp_function_lock);
1371 VNET_FOREACH(vnet_iter) {
1372 CURVNET_SET(vnet_iter);
1373 while ((inp = inp_next(&inpi)) != NULL) {
1374 if (inp->inp_flags & INP_TIMEWAIT)
1376 tp = intotcpcb(inp);
1377 if (tp == NULL || tp->t_fb != blk)
1379 tcp_switch_back_to_default(tp);
1383 VNET_LIST_RUNLOCK();
1385 rw_wlock(&tcp_function_lock);
1387 if (blk->tfb_refcnt) {
1388 /* TCBs still attached. */
1389 rw_wunlock(&tcp_function_lock);
1394 rw_wunlock(&tcp_function_lock);
1397 /* Remove any function names that map to this function block. */
1398 while (find_tcp_fb_locked(blk, &f) != NULL) {
1399 TAILQ_REMOVE(&t_functions, f, tf_next);
1402 free(f, M_TCPFUNCTIONS);
1404 rw_wunlock(&tcp_function_lock);
1409 tcp_vnet_init(void *arg __unused)
1413 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1414 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1415 printf("%s: WARNING: unable to register helper hook\n", __func__);
1416 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1417 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1418 printf("%s: WARNING: unable to register helper hook\n", __func__);
1421 if (tcp_stats_init())
1422 printf("%s: WARNING: unable to initialise TCP stats\n",
1425 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1429 * These have to be type stable for the benefit of the timers.
1431 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1432 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1433 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1434 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1440 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1441 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1442 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1444 tcp_fastopen_init();
1446 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1447 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1449 V_tcp_msl = TCPTV_MSL;
1451 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1452 tcp_vnet_init, NULL);
1455 tcp_init(void *arg __unused)
1457 const char *tcbhash_tuneable;
1460 tcp_reass_global_init();
1462 /* XXX virtualize those below? */
1463 tcp_delacktime = TCPTV_DELACK;
1464 tcp_keepinit = TCPTV_KEEP_INIT;
1465 tcp_keepidle = TCPTV_KEEP_IDLE;
1466 tcp_keepintvl = TCPTV_KEEPINTVL;
1467 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1468 tcp_rexmit_initial = TCPTV_RTOBASE;
1469 if (tcp_rexmit_initial < 1)
1470 tcp_rexmit_initial = 1;
1471 tcp_rexmit_min = TCPTV_MIN;
1472 if (tcp_rexmit_min < 1)
1474 tcp_persmin = TCPTV_PERSMIN;
1475 tcp_persmax = TCPTV_PERSMAX;
1476 tcp_rexmit_slop = TCPTV_CPU_VAR;
1477 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1479 /* Setup the tcp function block list */
1480 TAILQ_INIT(&t_functions);
1481 rw_init(&tcp_function_lock, "tcp_func_lock");
1482 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1484 /* Initialize the TCP logging data. */
1487 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1489 if (tcp_soreceive_stream) {
1491 tcp_usrreqs.pru_soreceive = soreceive_stream;
1494 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1499 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1501 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1503 if (max_protohdr < TCP_MINPROTOHDR)
1504 max_protohdr = TCP_MINPROTOHDR;
1505 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1507 #undef TCP_MINPROTOHDR
1510 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1511 SHUTDOWN_PRI_DEFAULT);
1513 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1514 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1515 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1516 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1517 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1518 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1519 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1520 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1521 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1526 hashsize = TCBHASHSIZE;
1527 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1528 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1529 if (hashsize == 0) {
1531 * Auto tune the hash size based on maxsockets.
1532 * A perfect hash would have a 1:1 mapping
1533 * (hashsize = maxsockets) however it's been
1534 * suggested that O(2) average is better.
1536 hashsize = maketcp_hashsize(maxsockets / 4);
1538 * Our historical default is 512,
1539 * do not autotune lower than this.
1544 printf("%s: %s auto tuned to %d\n", __func__,
1545 tcbhash_tuneable, hashsize);
1548 * We require a hashsize to be a power of two.
1549 * Previously if it was not a power of two we would just reset it
1550 * back to 512, which could be a nasty surprise if you did not notice
1551 * the error message.
1552 * Instead what we do is clip it to the closest power of two lower
1553 * than the specified hash value.
1555 if (!powerof2(hashsize)) {
1556 int oldhashsize = hashsize;
1558 hashsize = maketcp_hashsize(hashsize);
1559 /* prevent absurdly low value */
1562 printf("%s: WARNING: TCB hash size not a power of 2, "
1563 "clipped from %d to %d.\n", __func__, oldhashsize,
1566 tcp_tcbhashsize = hashsize;
1568 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1572 tcp_destroy(void *unused __unused)
1580 * All our processes are gone, all our sockets should be cleaned
1581 * up, which means, we should be past the tcp_discardcb() calls.
1582 * Sleep to let all tcpcb timers really disappear and cleanup.
1585 INP_INFO_WLOCK(&V_tcbinfo);
1586 n = V_tcbinfo.ipi_count;
1587 INP_INFO_WUNLOCK(&V_tcbinfo);
1590 pause("tcpdes", hz / 10);
1595 in_pcbinfo_destroy(&V_tcbinfo);
1596 /* tcp_discardcb() clears the sack_holes up. */
1597 uma_zdestroy(V_sack_hole_zone);
1598 uma_zdestroy(V_tcpcb_zone);
1601 * Cannot free the zone until all tcpcbs are released as we attach
1602 * the allocations to them.
1604 tcp_fastopen_destroy();
1606 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1607 VNET_PCPUSTAT_FREE(tcpstat);
1610 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1612 printf("%s: WARNING: unable to deregister helper hook "
1613 "type=%d, id=%d: error %d returned\n", __func__,
1614 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1616 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1618 printf("%s: WARNING: unable to deregister helper hook "
1619 "type=%d, id=%d: error %d returned\n", __func__,
1620 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1624 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1634 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1635 * tcp_template used to store this data in mbufs, but we now recopy it out
1636 * of the tcpcb each time to conserve mbufs.
1639 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1641 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1643 INP_WLOCK_ASSERT(inp);
1646 if ((inp->inp_vflag & INP_IPV6) != 0) {
1647 struct ip6_hdr *ip6;
1649 ip6 = (struct ip6_hdr *)ip_ptr;
1650 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1651 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1652 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1653 (IPV6_VERSION & IPV6_VERSION_MASK);
1655 ip6->ip6_nxt = IPPROTO_TCP;
1657 ip6->ip6_nxt = IPPROTO_UDP;
1658 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1659 ip6->ip6_src = inp->in6p_laddr;
1660 ip6->ip6_dst = inp->in6p_faddr;
1663 #if defined(INET6) && defined(INET)
1670 ip = (struct ip *)ip_ptr;
1671 ip->ip_v = IPVERSION;
1673 ip->ip_tos = inp->inp_ip_tos;
1677 ip->ip_ttl = inp->inp_ip_ttl;
1680 ip->ip_p = IPPROTO_TCP;
1682 ip->ip_p = IPPROTO_UDP;
1683 ip->ip_src = inp->inp_laddr;
1684 ip->ip_dst = inp->inp_faddr;
1687 th->th_sport = inp->inp_lport;
1688 th->th_dport = inp->inp_fport;
1692 tcp_set_flags(th, 0);
1695 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1699 * Create template to be used to send tcp packets on a connection.
1700 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1701 * use for this function is in keepalives, which use tcp_respond.
1704 tcpip_maketemplate(struct inpcb *inp)
1708 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1711 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1716 * Send a single message to the TCP at address specified by
1717 * the given TCP/IP header. If m == NULL, then we make a copy
1718 * of the tcpiphdr at th and send directly to the addressed host.
1719 * This is used to force keep alive messages out using the TCP
1720 * template for a connection. If flags are given then we send
1721 * a message back to the TCP which originated the segment th,
1722 * and discard the mbuf containing it and any other attached mbufs.
1724 * In any case the ack and sequence number of the transmitted
1725 * segment are as specified by the parameters.
1727 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1730 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1731 tcp_seq ack, tcp_seq seq, int flags)
1737 struct udphdr *uh = NULL;
1739 struct tcp_log_buffer *lgb;
1742 struct ip6_hdr *ip6;
1745 int optlen, tlen, win, ulen;
1750 int thflags = tcp_get_flags(th);
1753 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1757 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1764 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1765 INP_LOCK_ASSERT(inp);
1771 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1772 port = m->m_pkthdr.tcp_tun_port;
1775 if (ip && (ip->ip_p == IPPROTO_UDP))
1776 port = m->m_pkthdr.tcp_tun_port;
1785 if (!(flags & TH_RST)) {
1786 win = sbspace(&inp->inp_socket->so_rcv);
1787 if (win > TCP_MAXWIN << tp->rcv_scale)
1788 win = TCP_MAXWIN << tp->rcv_scale;
1790 if ((tp->t_flags & TF_NOOPT) == 0)
1794 m = m_gethdr(M_NOWAIT, MT_DATA);
1797 m->m_data += max_linkhdr;
1800 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1801 sizeof(struct ip6_hdr));
1802 ip6 = mtod(m, struct ip6_hdr *);
1803 nth = (struct tcphdr *)(ip6 + 1);
1805 /* Insert a UDP header */
1806 uh = (struct udphdr *)nth;
1807 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1808 uh->uh_dport = port;
1809 nth = (struct tcphdr *)(uh + 1);
1814 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1815 ip = mtod(m, struct ip *);
1816 nth = (struct tcphdr *)(ip + 1);
1818 /* Insert a UDP header */
1819 uh = (struct udphdr *)nth;
1820 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1821 uh->uh_dport = port;
1822 nth = (struct tcphdr *)(uh + 1);
1825 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1827 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1830 /* Can't reuse 'm', allocate a new mbuf. */
1831 n = m_gethdr(M_NOWAIT, MT_DATA);
1837 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1843 n->m_data += max_linkhdr;
1844 /* m_len is set later */
1845 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1848 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1849 sizeof(struct ip6_hdr));
1850 ip6 = mtod(n, struct ip6_hdr *);
1851 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1852 nth = (struct tcphdr *)(ip6 + 1);
1854 /* Insert a UDP header */
1855 uh = (struct udphdr *)nth;
1856 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1857 uh->uh_dport = port;
1858 nth = (struct tcphdr *)(uh + 1);
1863 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1864 ip = mtod(n, struct ip *);
1865 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1866 nth = (struct tcphdr *)(ip + 1);
1868 /* Insert a UDP header */
1869 uh = (struct udphdr *)nth;
1870 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1871 uh->uh_dport = port;
1872 nth = (struct tcphdr *)(uh + 1);
1875 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1876 xchg(nth->th_dport, nth->th_sport, uint16_t);
1883 * XXX MRT We inherit the FIB, which is lucky.
1887 m->m_data = (caddr_t)ipgen;
1888 /* m_len is set later */
1891 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1892 nth = (struct tcphdr *)(ip6 + 1);
1896 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1897 nth = (struct tcphdr *)(ip + 1);
1901 * this is usually a case when an extension header
1902 * exists between the IPv6 header and the
1905 nth->th_sport = th->th_sport;
1906 nth->th_dport = th->th_dport;
1908 xchg(nth->th_dport, nth->th_sport, uint16_t);
1914 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1916 #if defined(INET) && defined(INET6)
1920 tlen = sizeof (struct tcpiphdr);
1923 tlen += sizeof (struct udphdr);
1926 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1927 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1928 m, tlen, (long)M_TRAILINGSPACE(m)));
1933 /* Make sure we have room. */
1934 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1935 m->m_next = m_get(M_NOWAIT, MT_DATA);
1937 optp = mtod(m->m_next, u_char *);
1942 optp = (u_char *) (nth + 1);
1948 if (tp->t_flags & TF_RCVD_TSTMP) {
1949 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1950 to.to_tsecr = tp->ts_recent;
1951 to.to_flags |= TOF_TS;
1953 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1954 /* TCP-MD5 (RFC2385). */
1955 if (tp->t_flags & TF_SIGNATURE)
1956 to.to_flags |= TOF_SIGNATURE;
1958 /* Add the options. */
1959 tlen += optlen = tcp_addoptions(&to, optp);
1961 /* Update m_len in the correct mbuf. */
1962 optm->m_len += optlen;
1968 ulen = tlen - sizeof(struct ip6_hdr);
1969 uh->uh_ulen = htons(ulen);
1972 ip6->ip6_vfc = IPV6_VERSION;
1974 ip6->ip6_nxt = IPPROTO_UDP;
1976 ip6->ip6_nxt = IPPROTO_TCP;
1977 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1980 #if defined(INET) && defined(INET6)
1986 ulen = tlen - sizeof(struct ip);
1987 uh->uh_ulen = htons(ulen);
1989 ip->ip_len = htons(tlen);
1990 ip->ip_ttl = V_ip_defttl;
1992 ip->ip_p = IPPROTO_UDP;
1994 ip->ip_p = IPPROTO_TCP;
1996 if (V_path_mtu_discovery)
1997 ip->ip_off |= htons(IP_DF);
2000 m->m_pkthdr.len = tlen;
2001 m->m_pkthdr.rcvif = NULL;
2005 * Packet is associated with a socket, so allow the
2006 * label of the response to reflect the socket label.
2008 INP_LOCK_ASSERT(inp);
2009 mac_inpcb_create_mbuf(inp, m);
2012 * Packet is not associated with a socket, so possibly
2013 * update the label in place.
2015 mac_netinet_tcp_reply(m);
2018 nth->th_seq = htonl(seq);
2019 nth->th_ack = htonl(ack);
2020 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2021 tcp_set_flags(nth, flags);
2023 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2025 nth->th_win = htons((u_short)win);
2028 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2029 if (to.to_flags & TOF_SIGNATURE) {
2030 if (!TCPMD5_ENABLED() ||
2031 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2041 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2042 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2043 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2046 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2047 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2048 nth->th_sum = in6_cksum_pseudo(ip6,
2049 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2051 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2055 #if defined(INET6) && defined(INET)
2061 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2062 htons(ulen + IPPROTO_UDP));
2063 m->m_pkthdr.csum_flags = CSUM_UDP;
2064 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2067 m->m_pkthdr.csum_flags = CSUM_TCP;
2068 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2069 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2070 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2075 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2076 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2078 TCP_PROBE3(debug__output, tp, th, m);
2080 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2082 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2083 if (INP_WLOCKED(inp)) {
2084 union tcp_log_stackspecific log;
2087 memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2088 log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2089 log.u_bbr.flex8 = 4;
2090 log.u_bbr.pkts_out = tp->t_maxseg;
2091 log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2092 log.u_bbr.delivered = 0;
2093 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2094 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2097 * We can not log the packet, since we only own the
2098 * read lock, but a write lock is needed. The read lock
2099 * is not upgraded to a write lock, since only getting
2100 * the read lock was done intentionally to improve the
2101 * handling of SYN flooding attacks.
2102 * This happens only for pure SYN segments received in
2103 * the initial CLOSED state, or received in a more
2104 * advanced state than listen and the UDP encapsulation
2105 * port is unexpected.
2106 * The incoming SYN segments do not really belong to
2107 * the TCP connection and the handling does not change
2108 * the state of the TCP connection. Therefore, the
2109 * sending of the RST segments is not logged. Please
2110 * note that also the incoming SYN segments are not
2113 * The following code ensures that the above description
2114 * is and stays correct.
2116 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2117 (tp->t_state == TCPS_CLOSED ||
2118 (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2119 ("%s: Logging of TCP segment with flags 0x%b and "
2120 "UDP encapsulation port %u skipped in state %s",
2121 __func__, thflags, PRINT_TH_FLAGS,
2122 ntohs(port), tcpstates[tp->t_state]));
2128 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2129 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2132 #if defined(INET) && defined(INET6)
2137 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2138 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2142 lgb->tlb_errno = output_ret;
2146 * Create a new TCP control block, making an
2147 * empty reassembly queue and hooking it to the argument
2148 * protocol control block. The `inp' parameter must have
2149 * come from the zone allocator set up in tcp_init().
2152 tcp_newtcpcb(struct inpcb *inp)
2154 struct tcpcb_mem *tm;
2157 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2160 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2165 /* Initialise cc_var struct for this tcpcb. */
2167 tp->ccv->type = IPPROTO_TCP;
2168 tp->ccv->ccvc.tcp = tp;
2169 rw_rlock(&tcp_function_lock);
2170 tp->t_fb = tcp_func_set_ptr;
2171 refcount_acquire(&tp->t_fb->tfb_refcnt);
2172 rw_runlock(&tcp_function_lock);
2174 * Use the current system default CC algorithm.
2176 cc_attach(tp, CC_DEFAULT_ALGO());
2179 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2182 in_pcbref(inp); /* Reference for tcpcb */
2185 if (CC_ALGO(tp)->cb_init != NULL)
2186 if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2188 if (tp->t_fb->tfb_tcp_fb_fini)
2189 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2190 in_pcbrele_wlocked(inp);
2191 refcount_release(&tp->t_fb->tfb_refcnt);
2192 uma_zfree(V_tcpcb_zone, tm);
2198 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2199 if (tp->t_fb->tfb_tcp_fb_fini)
2200 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2201 in_pcbrele_wlocked(inp);
2202 refcount_release(&tp->t_fb->tfb_refcnt);
2203 uma_zfree(V_tcpcb_zone, tm);
2209 tp->t_vnet = inp->inp_vnet;
2211 tp->t_timers = &tm->tt;
2212 TAILQ_INIT(&tp->t_segq);
2215 isipv6 ? V_tcp_v6mssdflt :
2219 /* Set up our timeouts. */
2220 callout_init(&tp->t_timers->tt_rexmt, 1);
2221 callout_init(&tp->t_timers->tt_persist, 1);
2222 callout_init(&tp->t_timers->tt_keep, 1);
2223 callout_init(&tp->t_timers->tt_2msl, 1);
2224 callout_init(&tp->t_timers->tt_delack, 1);
2226 if (V_tcp_do_rfc1323)
2227 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2229 tp->t_flags |= TF_SACK_PERMIT;
2230 TAILQ_INIT(&tp->snd_holes);
2233 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2234 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2235 * reasonable initial retransmit time.
2237 tp->t_srtt = TCPTV_SRTTBASE;
2238 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2239 tp->t_rttmin = tcp_rexmit_min;
2240 tp->t_rxtcur = tcp_rexmit_initial;
2241 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2242 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2243 tp->t_rcvtime = ticks;
2245 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2246 * because the socket may be bound to an IPv6 wildcard address,
2247 * which may match an IPv4-mapped IPv6 address.
2249 inp->inp_ip_ttl = V_ip_defttl;
2253 * Init the TCP PCAP queues.
2255 tcp_pcap_tcpcb_init(tp);
2258 /* Initialize the per-TCPCB log data. */
2259 tcp_log_tcpcbinit(tp);
2261 tp->t_pacing_rate = -1;
2262 if (tp->t_fb->tfb_tcp_fb_init) {
2263 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2264 refcount_release(&tp->t_fb->tfb_refcnt);
2265 in_pcbrele_wlocked(inp);
2266 uma_zfree(V_tcpcb_zone, tm);
2271 if (V_tcp_perconn_stats_enable == 1)
2272 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2275 tp->t_flags |= TF_LRD;
2276 return (tp); /* XXX */
2280 * Drop a TCP connection, reporting
2281 * the specified error. If connection is synchronized,
2282 * then send a RST to peer.
2285 tcp_drop(struct tcpcb *tp, int errno)
2287 struct socket *so = tp->t_inpcb->inp_socket;
2290 INP_WLOCK_ASSERT(tp->t_inpcb);
2292 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2293 tcp_state_change(tp, TCPS_CLOSED);
2294 /* Don't use tcp_output() here due to possible recursion. */
2295 (void)tcp_output_nodrop(tp);
2296 TCPSTAT_INC(tcps_drops);
2298 TCPSTAT_INC(tcps_conndrops);
2299 if (errno == ETIMEDOUT && tp->t_softerror)
2300 errno = tp->t_softerror;
2301 so->so_error = errno;
2302 return (tcp_close(tp));
2306 tcp_discardcb(struct tcpcb *tp)
2308 struct inpcb *inp = tp->t_inpcb;
2310 INP_WLOCK_ASSERT(inp);
2313 * Make sure that all of our timers are stopped before we delete the
2316 * If stopping a timer fails, we schedule a discard function in same
2317 * callout, and the last discard function called will take care of
2318 * deleting the tcpcb.
2320 tp->t_timers->tt_draincnt = 0;
2321 tcp_timer_stop(tp, TT_REXMT);
2322 tcp_timer_stop(tp, TT_PERSIST);
2323 tcp_timer_stop(tp, TT_KEEP);
2324 tcp_timer_stop(tp, TT_2MSL);
2325 tcp_timer_stop(tp, TT_DELACK);
2326 if (tp->t_fb->tfb_tcp_timer_stop_all) {
2328 * Call the stop-all function of the methods,
2329 * this function should call the tcp_timer_stop()
2330 * method with each of the function specific timeouts.
2331 * That stop will be called via the tfb_tcp_timer_stop()
2332 * which should use the async drain function of the
2333 * callout system (see tcp_var.h).
2335 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2338 /* free the reassembly queue, if any */
2339 tcp_reass_flush(tp);
2342 /* Disconnect offload device, if any. */
2343 if (tp->t_flags & TF_TOE)
2344 tcp_offload_detach(tp);
2347 tcp_free_sackholes(tp);
2350 /* Free the TCP PCAP queues. */
2351 tcp_pcap_drain(&(tp->t_inpkts));
2352 tcp_pcap_drain(&(tp->t_outpkts));
2355 /* Allow the CC algorithm to clean up after itself. */
2356 if (CC_ALGO(tp)->cb_destroy != NULL)
2357 CC_ALGO(tp)->cb_destroy(tp->ccv);
2359 /* Detach from the CC algorithm */
2363 khelp_destroy_osd(tp->osd);
2366 stats_blob_destroy(tp->t_stats);
2370 inp->inp_ppcb = NULL;
2371 if (tp->t_timers->tt_draincnt == 0) {
2372 bool released __diagused;
2374 released = tcp_freecb(tp);
2375 KASSERT(!released, ("%s: inp %p should not have been released "
2376 "here", __func__, inp));
2381 tcp_freecb(struct tcpcb *tp)
2383 struct inpcb *inp = tp->t_inpcb;
2384 struct socket *so = inp->inp_socket;
2386 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2389 INP_WLOCK_ASSERT(inp);
2390 MPASS(tp->t_timers->tt_draincnt == 0);
2392 /* We own the last reference on tcpcb, let's free it. */
2394 tcp_log_tcpcbfini(tp);
2396 TCPSTATES_DEC(tp->t_state);
2397 if (tp->t_fb->tfb_tcp_fb_fini)
2398 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2401 * If we got enough samples through the srtt filter,
2402 * save the rtt and rttvar in the routing entry.
2403 * 'Enough' is arbitrarily defined as 4 rtt samples.
2404 * 4 samples is enough for the srtt filter to converge
2405 * to within enough % of the correct value; fewer samples
2406 * and we could save a bogus rtt. The danger is not high
2407 * as tcp quickly recovers from everything.
2408 * XXX: Works very well but needs some more statistics!
2410 * XXXRRS: Updating must be after the stack fini() since
2411 * that may be converting some internal representation of
2412 * say srtt etc into the general one used by other stacks.
2413 * Lets also at least protect against the so being NULL
2414 * as RW stated below.
2416 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2417 struct hc_metrics_lite metrics;
2420 bzero(&metrics, sizeof(metrics));
2422 * Update the ssthresh always when the conditions below
2423 * are satisfied. This gives us better new start value
2424 * for the congestion avoidance for new connections.
2425 * ssthresh is only set if packet loss occurred on a session.
2427 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2428 * being torn down. Ideally this code would not use 'so'.
2430 ssthresh = tp->snd_ssthresh;
2431 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2433 * convert the limit from user data bytes to
2434 * packets then to packet data bytes.
2436 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2439 ssthresh *= (tp->t_maxseg +
2441 (isipv6 ? sizeof (struct ip6_hdr) +
2442 sizeof (struct tcphdr) :
2444 sizeof (struct tcpiphdr)
2451 metrics.rmx_ssthresh = ssthresh;
2453 metrics.rmx_rtt = tp->t_srtt;
2454 metrics.rmx_rttvar = tp->t_rttvar;
2455 metrics.rmx_cwnd = tp->snd_cwnd;
2456 metrics.rmx_sendpipe = 0;
2457 metrics.rmx_recvpipe = 0;
2459 tcp_hc_update(&inp->inp_inc, &metrics);
2462 refcount_release(&tp->t_fb->tfb_refcnt);
2463 uma_zfree(V_tcpcb_zone, tp);
2465 return (in_pcbrele_wlocked(inp));
2469 * Attempt to close a TCP control block, marking it as dropped, and freeing
2470 * the socket if we hold the only reference.
2473 tcp_close(struct tcpcb *tp)
2475 struct inpcb *inp = tp->t_inpcb;
2478 INP_WLOCK_ASSERT(inp);
2481 if (tp->t_state == TCPS_LISTEN)
2482 tcp_offload_listen_stop(tp);
2485 * This releases the TFO pending counter resource for TFO listen
2486 * sockets as well as passively-created TFO sockets that transition
2487 * from SYN_RECEIVED to CLOSED.
2489 if (tp->t_tfo_pending) {
2490 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2491 tp->t_tfo_pending = NULL;
2494 tcp_hpts_remove(inp);
2497 TCPSTAT_INC(tcps_closed);
2498 if (tp->t_state != TCPS_CLOSED)
2499 tcp_state_change(tp, TCPS_CLOSED);
2500 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2501 so = inp->inp_socket;
2502 soisdisconnected(so);
2503 if (inp->inp_flags & INP_SOCKREF) {
2504 inp->inp_flags &= ~INP_SOCKREF;
2515 VNET_ITERATOR_DECL(vnet_iter);
2520 VNET_LIST_RLOCK_NOSLEEP();
2521 VNET_FOREACH(vnet_iter) {
2522 CURVNET_SET(vnet_iter);
2523 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2524 INPLOOKUP_WLOCKPCB);
2529 * Walk the tcpbs, if existing, and flush the reassembly queue,
2530 * if there is one...
2531 * XXX: The "Net/3" implementation doesn't imply that the TCP
2532 * reassembly queue should be flushed, but in a situation
2533 * where we're really low on mbufs, this is potentially
2536 while ((inpb = inp_next(&inpi)) != NULL) {
2537 if (inpb->inp_flags & INP_TIMEWAIT)
2539 if ((tcpb = intotcpcb(inpb)) != NULL) {
2540 tcp_reass_flush(tcpb);
2541 tcp_clean_sackreport(tcpb);
2543 tcp_log_drain(tcpb);
2546 if (tcp_pcap_aggressive_free) {
2547 /* Free the TCP PCAP queues. */
2548 tcp_pcap_drain(&(tcpb->t_inpkts));
2549 tcp_pcap_drain(&(tcpb->t_outpkts));
2556 VNET_LIST_RUNLOCK_NOSLEEP();
2560 * Notify a tcp user of an asynchronous error;
2561 * store error as soft error, but wake up user
2562 * (for now, won't do anything until can select for soft error).
2564 * Do not wake up user since there currently is no mechanism for
2565 * reporting soft errors (yet - a kqueue filter may be added).
2567 static struct inpcb *
2568 tcp_notify(struct inpcb *inp, int error)
2572 INP_WLOCK_ASSERT(inp);
2574 if ((inp->inp_flags & INP_TIMEWAIT) ||
2575 (inp->inp_flags & INP_DROPPED))
2578 tp = intotcpcb(inp);
2579 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2582 * Ignore some errors if we are hooked up.
2583 * If connection hasn't completed, has retransmitted several times,
2584 * and receives a second error, give up now. This is better
2585 * than waiting a long time to establish a connection that
2586 * can never complete.
2588 if (tp->t_state == TCPS_ESTABLISHED &&
2589 (error == EHOSTUNREACH || error == ENETUNREACH ||
2590 error == EHOSTDOWN)) {
2591 if (inp->inp_route.ro_nh) {
2592 NH_FREE(inp->inp_route.ro_nh);
2593 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2596 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2598 tp = tcp_drop(tp, error);
2604 tp->t_softerror = error;
2608 wakeup( &so->so_timeo);
2615 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2617 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2618 INPLOOKUP_RLOCKPCB);
2623 if (req->newptr != NULL)
2626 if (req->oldptr == NULL) {
2629 n = V_tcbinfo.ipi_count +
2630 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2631 n += imax(n / 8, 10);
2632 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2636 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2639 bzero(&xig, sizeof(xig));
2640 xig.xig_len = sizeof xig;
2641 xig.xig_count = V_tcbinfo.ipi_count +
2642 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2643 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2644 xig.xig_sogen = so_gencnt;
2645 error = SYSCTL_OUT(req, &xig, sizeof xig);
2649 error = syncache_pcblist(req);
2653 while ((inp = inp_next(&inpi)) != NULL) {
2654 if (inp->inp_gencnt <= xig.xig_gen) {
2658 * XXX: This use of cr_cansee(), introduced with
2659 * TCP state changes, is not quite right, but for
2660 * now, better than nothing.
2662 if (inp->inp_flags & INP_TIMEWAIT) {
2663 if (intotw(inp) != NULL)
2664 crerr = cr_cansee(req->td->td_ucred,
2665 intotw(inp)->tw_cred);
2667 crerr = EINVAL; /* Skip this inp. */
2669 crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2673 tcp_inptoxtp(inp, &xt);
2674 error = SYSCTL_OUT(req, &xt, sizeof xt);
2686 * Give the user an updated idea of our state.
2687 * If the generation differs from what we told
2688 * her before, she knows that something happened
2689 * while we were processing this request, and it
2690 * might be necessary to retry.
2692 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2693 xig.xig_sogen = so_gencnt;
2694 xig.xig_count = V_tcbinfo.ipi_count +
2695 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2696 error = SYSCTL_OUT(req, &xig, sizeof xig);
2702 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2703 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2704 NULL, 0, tcp_pcblist, "S,xtcpcb",
2705 "List of active TCP connections");
2709 tcp_getcred(SYSCTL_HANDLER_ARGS)
2712 struct sockaddr_in addrs[2];
2713 struct epoch_tracker et;
2717 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2720 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2723 NET_EPOCH_ENTER(et);
2724 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2725 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2728 if (inp->inp_socket == NULL)
2731 error = cr_canseeinpcb(req->td->td_ucred, inp);
2733 cru2x(inp->inp_cred, &xuc);
2738 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2742 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2743 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2744 0, 0, tcp_getcred, "S,xucred",
2745 "Get the xucred of a TCP connection");
2750 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2752 struct epoch_tracker et;
2754 struct sockaddr_in6 addrs[2];
2761 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2764 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2767 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2768 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2771 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2773 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2780 NET_EPOCH_ENTER(et);
2783 inp = in_pcblookup(&V_tcbinfo,
2784 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2786 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2787 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2790 inp = in6_pcblookup(&V_tcbinfo,
2791 &addrs[1].sin6_addr, addrs[1].sin6_port,
2792 &addrs[0].sin6_addr, addrs[0].sin6_port,
2793 INPLOOKUP_RLOCKPCB, NULL);
2796 if (inp->inp_socket == NULL)
2799 error = cr_canseeinpcb(req->td->td_ucred, inp);
2801 cru2x(inp->inp_cred, &xuc);
2806 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2810 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2811 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2812 0, 0, tcp6_getcred, "S,xucred",
2813 "Get the xucred of a TCP6 connection");
2817 /* Path MTU to try next when a fragmentation-needed message is received. */
2819 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2821 int mtu = ntohs(icp->icmp_nextmtu);
2823 /* If no alternative MTU was proposed, try the next smaller one. */
2825 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2826 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2827 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2833 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2835 struct ip *ip = vip;
2837 struct in_addr faddr;
2840 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2842 struct in_conninfo inc;
2843 tcp_seq icmp_tcp_seq;
2846 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2847 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2850 if (cmd == PRC_MSGSIZE)
2851 notify = tcp_mtudisc_notify;
2852 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2853 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2854 cmd == PRC_TIMXCEED_INTRANS) && ip)
2855 notify = tcp_drop_syn_sent;
2858 * Hostdead is ugly because it goes linearly through all PCBs.
2859 * XXX: We never get this from ICMP, otherwise it makes an
2860 * excellent DoS attack on machines with many connections.
2862 else if (cmd == PRC_HOSTDEAD)
2864 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2868 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2872 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2873 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2874 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2875 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2876 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2877 /* signal EHOSTDOWN, as it flushes the cached route */
2878 inp = (*notify)(inp, EHOSTDOWN);
2881 icmp_tcp_seq = th->th_seq;
2883 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2884 !(inp->inp_flags & INP_DROPPED) &&
2885 !(inp->inp_socket == NULL)) {
2886 tp = intotcpcb(inp);
2888 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2890 * MTU discovery for offloaded connections. Let
2891 * the TOE driver verify seq# and process it.
2893 mtu = tcp_next_pmtu(icp, ip);
2894 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2898 if (tp->t_port != port) {
2901 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2902 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2903 if (cmd == PRC_MSGSIZE) {
2905 * MTU discovery: we got a needfrag and
2906 * will potentially try a lower MTU.
2908 mtu = tcp_next_pmtu(icp, ip);
2911 * Only process the offered MTU if it
2912 * is smaller than the current one.
2914 if (mtu < tp->t_maxseg +
2915 sizeof(struct tcpiphdr)) {
2916 bzero(&inc, sizeof(inc));
2917 inc.inc_faddr = faddr;
2919 inp->inp_inc.inc_fibnum;
2920 tcp_hc_updatemtu(&inc, mtu);
2921 inp = tcp_mtudisc(inp, mtu);
2924 inp = (*notify)(inp,
2925 inetctlerrmap[cmd]);
2929 bzero(&inc, sizeof(inc));
2930 inc.inc_fport = th->th_dport;
2931 inc.inc_lport = th->th_sport;
2932 inc.inc_faddr = faddr;
2933 inc.inc_laddr = ip->ip_src;
2934 syncache_unreach(&inc, icmp_tcp_seq, port);
2942 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2944 tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2948 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2950 /* Its a tunneled TCP over UDP icmp */
2951 struct ip *outer_ip, *inner_ip;
2954 struct tcphdr *th, ttemp;
2958 inner_ip = (struct ip *)vip;
2959 icmp = (struct icmp *)((caddr_t)inner_ip -
2960 (sizeof(struct icmp) - sizeof(struct ip)));
2961 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2962 i_hlen = inner_ip->ip_hl << 2;
2963 o_len = ntohs(outer_ip->ip_len);
2965 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2966 /* Not enough data present */
2969 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2970 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2971 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2974 port = udp->uh_dport;
2975 th = (struct tcphdr *)(udp + 1);
2976 memcpy(&ttemp, th, sizeof(struct tcphdr));
2977 memcpy(udp, &ttemp, sizeof(struct tcphdr));
2978 /* Now adjust down the size of the outer IP header */
2979 o_len -= sizeof(struct udphdr);
2980 outer_ip->ip_len = htons(o_len);
2981 /* Now call in to the normal handling code */
2982 tcp_ctlinput_with_port(cmd, sa, vip, port);
2988 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2990 int mtu = ntohl(icmp6->icmp6_mtu);
2993 * If no alternative MTU was proposed, or the proposed MTU was too
2994 * small, set to the min.
2996 if (mtu < IPV6_MMTU)
2997 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */
3002 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3004 struct in6_addr *dst;
3005 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3006 struct ip6_hdr *ip6;
3010 struct icmp6_hdr *icmp6;
3011 struct ip6ctlparam *ip6cp = NULL;
3012 const struct sockaddr_in6 *sa6_src = NULL;
3013 struct in_conninfo inc;
3018 tcp_seq icmp_tcp_seq;
3022 if (sa->sa_family != AF_INET6 ||
3023 sa->sa_len != sizeof(struct sockaddr_in6))
3026 /* if the parameter is from icmp6, decode it. */
3028 ip6cp = (struct ip6ctlparam *)d;
3029 icmp6 = ip6cp->ip6c_icmp6;
3031 ip6 = ip6cp->ip6c_ip6;
3032 off = ip6cp->ip6c_off;
3033 sa6_src = ip6cp->ip6c_src;
3034 dst = ip6cp->ip6c_finaldst;
3038 off = 0; /* fool gcc */
3043 if (cmd == PRC_MSGSIZE)
3044 notify = tcp_mtudisc_notify;
3045 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3046 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3047 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3048 notify = tcp_drop_syn_sent;
3051 * Hostdead is ugly because it goes linearly through all PCBs.
3052 * XXX: We never get this from ICMP, otherwise it makes an
3053 * excellent DoS attack on machines with many connections.
3055 else if (cmd == PRC_HOSTDEAD)
3057 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3061 in6_pcbnotify(&V_tcbinfo, sa, 0,
3062 (const struct sockaddr *)sa6_src,
3063 0, cmd, NULL, notify);
3067 /* Check if we can safely get the ports from the tcp hdr */
3070 (int32_t) (off + sizeof(struct tcp_ports)))) {
3073 bzero(&t_ports, sizeof(struct tcp_ports));
3074 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3075 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3076 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3077 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3078 /* signal EHOSTDOWN, as it flushes the cached route */
3079 inp = (*notify)(inp, EHOSTDOWN);
3082 off += sizeof(struct tcp_ports);
3083 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3086 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3088 if (!(inp->inp_flags & INP_TIMEWAIT) &&
3089 !(inp->inp_flags & INP_DROPPED) &&
3090 !(inp->inp_socket == NULL)) {
3091 tp = intotcpcb(inp);
3093 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3094 /* MTU discovery for offloaded connections. */
3095 mtu = tcp6_next_pmtu(icmp6);
3096 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3100 if (tp->t_port != port) {
3103 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3104 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3105 if (cmd == PRC_MSGSIZE) {
3108 * If we got a needfrag set the MTU
3109 * in the route to the suggested new
3110 * value (if given) and then notify.
3112 mtu = tcp6_next_pmtu(icmp6);
3114 bzero(&inc, sizeof(inc));
3115 inc.inc_fibnum = M_GETFIB(m);
3116 inc.inc_flags |= INC_ISIPV6;
3117 inc.inc6_faddr = *dst;
3118 if (in6_setscope(&inc.inc6_faddr,
3119 m->m_pkthdr.rcvif, NULL))
3122 * Only process the offered MTU if it
3123 * is smaller than the current one.
3125 if (mtu < tp->t_maxseg +
3126 sizeof (struct tcphdr) +
3127 sizeof (struct ip6_hdr)) {
3128 tcp_hc_updatemtu(&inc, mtu);
3129 tcp_mtudisc(inp, mtu);
3130 ICMP6STAT_INC(icp6s_pmtuchg);
3133 inp = (*notify)(inp,
3134 inet6ctlerrmap[cmd]);
3138 bzero(&inc, sizeof(inc));
3139 inc.inc_fibnum = M_GETFIB(m);
3140 inc.inc_flags |= INC_ISIPV6;
3141 inc.inc_fport = t_ports.th_dport;
3142 inc.inc_lport = t_ports.th_sport;
3143 inc.inc6_faddr = *dst;
3144 inc.inc6_laddr = ip6->ip6_src;
3145 syncache_unreach(&inc, icmp_tcp_seq, port);
3153 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3155 tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3159 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3161 struct ip6ctlparam *ip6cp;
3166 ip6cp = (struct ip6ctlparam *)d;
3167 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3171 udp = mtod(m, struct udphdr *);
3172 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3175 port = udp->uh_dport;
3176 m_adj(m, sizeof(struct udphdr));
3177 if ((m->m_flags & M_PKTHDR) == 0) {
3178 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3180 /* Now call in to the normal handling code */
3181 tcp6_ctlinput_with_port(cmd, sa, d, port);
3187 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3192 KASSERT(len >= SIPHASH_KEY_LENGTH,
3193 ("%s: keylen %u too short ", __func__, len));
3194 SipHash24_Init(&ctx);
3195 SipHash_SetKey(&ctx, (uint8_t *)key);
3196 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3197 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3198 switch (inc->inc_flags & INC_ISIPV6) {
3201 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3202 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3207 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3208 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3212 SipHash_Final((uint8_t *)hash, &ctx);
3214 return (hash[0] ^ hash[1]);
3218 tcp_new_ts_offset(struct in_conninfo *inc)
3220 struct in_conninfo inc_store, *local_inc;
3222 if (!V_tcp_ts_offset_per_conn) {
3223 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3224 inc_store.inc_lport = 0;
3225 inc_store.inc_fport = 0;
3226 local_inc = &inc_store;
3230 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3231 sizeof(V_ts_offset_secret)));
3235 * Following is where TCP initial sequence number generation occurs.
3237 * There are two places where we must use initial sequence numbers:
3238 * 1. In SYN-ACK packets.
3239 * 2. In SYN packets.
3241 * All ISNs for SYN-ACK packets are generated by the syncache. See
3242 * tcp_syncache.c for details.
3244 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3245 * depends on this property. In addition, these ISNs should be
3246 * unguessable so as to prevent connection hijacking. To satisfy
3247 * the requirements of this situation, the algorithm outlined in
3248 * RFC 1948 is used, with only small modifications.
3250 * Implementation details:
3252 * Time is based off the system timer, and is corrected so that it
3253 * increases by one megabyte per second. This allows for proper
3254 * recycling on high speed LANs while still leaving over an hour
3257 * As reading the *exact* system time is too expensive to be done
3258 * whenever setting up a TCP connection, we increment the time
3259 * offset in two ways. First, a small random positive increment
3260 * is added to isn_offset for each connection that is set up.
3261 * Second, the function tcp_isn_tick fires once per clock tick
3262 * and increments isn_offset as necessary so that sequence numbers
3263 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3264 * random positive increments serve only to ensure that the same
3265 * exact sequence number is never sent out twice (as could otherwise
3266 * happen when a port is recycled in less than the system tick
3269 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3270 * between seeding of isn_secret. This is normally set to zero,
3271 * as reseeding should not be necessary.
3273 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3274 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3275 * general, this means holding an exclusive (write) lock.
3278 #define ISN_BYTES_PER_SECOND 1048576
3279 #define ISN_STATIC_INCREMENT 4096
3280 #define ISN_RANDOM_INCREMENT (4096 - 1)
3281 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3283 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3284 VNET_DEFINE_STATIC(int, isn_last);
3285 VNET_DEFINE_STATIC(int, isn_last_reseed);
3286 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3287 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3289 #define V_isn_secret VNET(isn_secret)
3290 #define V_isn_last VNET(isn_last)
3291 #define V_isn_last_reseed VNET(isn_last_reseed)
3292 #define V_isn_offset VNET(isn_offset)
3293 #define V_isn_offset_old VNET(isn_offset_old)
3296 tcp_new_isn(struct in_conninfo *inc)
3299 u_int32_t projected_offset;
3302 /* Seed if this is the first use, reseed if requested. */
3303 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3304 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3306 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3307 V_isn_last_reseed = ticks;
3310 /* Compute the hash and return the ISN. */
3311 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3312 sizeof(V_isn_secret));
3313 V_isn_offset += ISN_STATIC_INCREMENT +
3314 (arc4random() & ISN_RANDOM_INCREMENT);
3315 if (ticks != V_isn_last) {
3316 projected_offset = V_isn_offset_old +
3317 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3318 if (SEQ_GT(projected_offset, V_isn_offset))
3319 V_isn_offset = projected_offset;
3320 V_isn_offset_old = V_isn_offset;
3323 new_isn += V_isn_offset;
3329 * When a specific ICMP unreachable message is received and the
3330 * connection state is SYN-SENT, drop the connection. This behavior
3331 * is controlled by the icmp_may_rst sysctl.
3334 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3339 INP_WLOCK_ASSERT(inp);
3341 if ((inp->inp_flags & INP_TIMEWAIT) ||
3342 (inp->inp_flags & INP_DROPPED))
3345 tp = intotcpcb(inp);
3346 if (tp->t_state != TCPS_SYN_SENT)
3349 if (IS_FASTOPEN(tp->t_flags))
3350 tcp_fastopen_disable_path(tp);
3352 tp = tcp_drop(tp, errno);
3360 * When `need fragmentation' ICMP is received, update our idea of the MSS
3361 * based on the new value. Also nudge TCP to send something, since we
3362 * know the packet we just sent was dropped.
3363 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3365 static struct inpcb *
3366 tcp_mtudisc_notify(struct inpcb *inp, int error)
3369 return (tcp_mtudisc(inp, -1));
3372 static struct inpcb *
3373 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3378 INP_WLOCK_ASSERT(inp);
3379 if ((inp->inp_flags & INP_TIMEWAIT) ||
3380 (inp->inp_flags & INP_DROPPED))
3383 tp = intotcpcb(inp);
3384 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3386 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3388 so = inp->inp_socket;
3389 SOCKBUF_LOCK(&so->so_snd);
3390 /* If the mss is larger than the socket buffer, decrease the mss. */
3391 if (so->so_snd.sb_hiwat < tp->t_maxseg)
3392 tp->t_maxseg = so->so_snd.sb_hiwat;
3393 SOCKBUF_UNLOCK(&so->so_snd);
3395 TCPSTAT_INC(tcps_mturesent);
3397 tp->snd_nxt = tp->snd_una;
3398 tcp_free_sackholes(tp);
3399 tp->snd_recover = tp->snd_max;
3400 if (tp->t_flags & TF_SACK_PERMIT)
3401 EXIT_FASTRECOVERY(tp->t_flags);
3402 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3404 * Conceptually the snd_nxt setting
3405 * and freeing sack holes should
3406 * be done by the default stacks
3407 * own tfb_tcp_mtu_chg().
3409 tp->t_fb->tfb_tcp_mtu_chg(tp);
3411 if (tcp_output(tp) < 0)
3419 * Look-up the routing entry to the peer of this inpcb. If no route
3420 * is found and it cannot be allocated, then return 0. This routine
3421 * is called by TCP routines that access the rmx structure and by
3422 * tcp_mss_update to get the peer/interface MTU.
3425 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3427 struct nhop_object *nh;
3429 uint32_t maxmtu = 0;
3431 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3433 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3434 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3439 maxmtu = nh->nh_mtu;
3441 /* Report additional interface capabilities. */
3443 if (ifp->if_capenable & IFCAP_TSO4 &&
3444 ifp->if_hwassist & CSUM_TSO) {
3445 cap->ifcap |= CSUM_TSO;
3446 cap->tsomax = ifp->if_hw_tsomax;
3447 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3448 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3458 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3460 struct nhop_object *nh;
3461 struct in6_addr dst6;
3464 uint32_t maxmtu = 0;
3466 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3468 if (inc->inc_flags & INC_IPV6MINMTU)
3471 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3472 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3473 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3478 maxmtu = nh->nh_mtu;
3480 /* Report additional interface capabilities. */
3482 if (ifp->if_capenable & IFCAP_TSO6 &&
3483 ifp->if_hwassist & CSUM_TSO) {
3484 cap->ifcap |= CSUM_TSO;
3485 cap->tsomax = ifp->if_hw_tsomax;
3486 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3487 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3496 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3498 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3499 * The right place to do that is ip6_setpktopt() that has just been
3500 * executed. By the way it just filled ip6po_minmtu for us.
3503 tcp6_use_min_mtu(struct tcpcb *tp)
3505 struct inpcb *inp = tp->t_inpcb;
3507 INP_WLOCK_ASSERT(inp);
3509 * In case of the IPV6_USE_MIN_MTU socket
3510 * option, the INC_IPV6MINMTU flag to announce
3511 * a corresponding MSS during the initial
3512 * handshake. If the TCP connection is not in
3513 * the front states, just reduce the MSS being
3514 * used. This avoids the sending of TCP
3515 * segments which will be fragmented at the
3518 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3519 if ((tp->t_state >= TCPS_SYN_SENT) &&
3520 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3521 struct ip6_pktopts *opt;
3523 opt = inp->in6p_outputopts;
3524 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3525 tp->t_maxseg > TCP6_MSS)
3526 tp->t_maxseg = TCP6_MSS;
3532 * Calculate effective SMSS per RFC5681 definition for a given TCP
3533 * connection at its current state, taking into account SACK and etc.
3536 tcp_maxseg(const struct tcpcb *tp)
3540 if (tp->t_flags & TF_NOOPT)
3541 return (tp->t_maxseg);
3544 * Here we have a simplified code from tcp_addoptions(),
3545 * without a proper loop, and having most of paddings hardcoded.
3546 * We might make mistakes with padding here in some edge cases,
3547 * but this is harmless, since result of tcp_maxseg() is used
3548 * only in cwnd and ssthresh estimations.
3550 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3551 if (tp->t_flags & TF_RCVD_TSTMP)
3552 optlen = TCPOLEN_TSTAMP_APPA;
3555 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3556 if (tp->t_flags & TF_SIGNATURE)
3557 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3559 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3560 optlen += TCPOLEN_SACKHDR;
3561 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3562 optlen = PADTCPOLEN(optlen);
3565 if (tp->t_flags & TF_REQ_TSTMP)
3566 optlen = TCPOLEN_TSTAMP_APPA;
3568 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3569 if (tp->t_flags & TF_REQ_SCALE)
3570 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3571 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3572 if (tp->t_flags & TF_SIGNATURE)
3573 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3575 if (tp->t_flags & TF_SACK_PERMIT)
3576 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3579 optlen = min(optlen, TCP_MAXOLEN);
3580 return (tp->t_maxseg - optlen);
3585 tcp_fixed_maxseg(const struct tcpcb *tp)
3589 if (tp->t_flags & TF_NOOPT)
3590 return (tp->t_maxseg);
3593 * Here we have a simplified code from tcp_addoptions(),
3594 * without a proper loop, and having most of paddings hardcoded.
3595 * We only consider fixed options that we would send every
3596 * time I.e. SACK is not considered. This is important
3597 * for cc modules to figure out what the modulo of the
3600 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3601 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3602 if (tp->t_flags & TF_RCVD_TSTMP)
3603 optlen = TCPOLEN_TSTAMP_APPA;
3606 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3607 if (tp->t_flags & TF_SIGNATURE)
3608 optlen += PAD(TCPOLEN_SIGNATURE);
3611 if (tp->t_flags & TF_REQ_TSTMP)
3612 optlen = TCPOLEN_TSTAMP_APPA;
3614 optlen = PAD(TCPOLEN_MAXSEG);
3615 if (tp->t_flags & TF_REQ_SCALE)
3616 optlen += PAD(TCPOLEN_WINDOW);
3617 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3618 if (tp->t_flags & TF_SIGNATURE)
3619 optlen += PAD(TCPOLEN_SIGNATURE);
3621 if (tp->t_flags & TF_SACK_PERMIT)
3622 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3625 optlen = min(optlen, TCP_MAXOLEN);
3626 return (tp->t_maxseg - optlen);
3632 sysctl_drop(SYSCTL_HANDLER_ARGS)
3634 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3635 struct sockaddr_storage addrs[2];
3640 struct sockaddr_in *fin = NULL, *lin = NULL;
3642 struct epoch_tracker et;
3644 struct sockaddr_in6 *fin6, *lin6;
3654 if (req->oldptr != NULL || req->oldlen != 0)
3656 if (req->newptr == NULL)
3658 if (req->newlen < sizeof(addrs))
3660 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3664 switch (addrs[0].ss_family) {
3667 fin6 = (struct sockaddr_in6 *)&addrs[0];
3668 lin6 = (struct sockaddr_in6 *)&addrs[1];
3669 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3670 lin6->sin6_len != sizeof(struct sockaddr_in6))
3672 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3673 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3675 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3676 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3678 fin = (struct sockaddr_in *)&addrs[0];
3679 lin = (struct sockaddr_in *)&addrs[1];
3683 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3686 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3693 fin = (struct sockaddr_in *)&addrs[0];
3694 lin = (struct sockaddr_in *)&addrs[1];
3695 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3696 lin->sin_len != sizeof(struct sockaddr_in))
3703 NET_EPOCH_ENTER(et);
3704 switch (addrs[0].ss_family) {
3707 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3708 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3709 INPLOOKUP_WLOCKPCB, NULL);
3714 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3715 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3720 if (inp->inp_flags & INP_TIMEWAIT) {
3722 * XXXRW: There currently exists a state where an
3723 * inpcb is present, but its timewait state has been
3724 * discarded. For now, don't allow dropping of this
3732 } else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3733 !SOLISTENING(inp->inp_socket)) {
3734 tp = intotcpcb(inp);
3735 tp = tcp_drop(tp, ECONNABORTED);
3746 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3747 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3748 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3749 "Drop TCP connection");
3752 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3754 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3755 &tcp_ctloutput_set));
3758 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3759 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3760 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3761 "Set socket option for TCP endpoint");
3765 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3767 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3768 struct sockaddr_storage addrs[2];
3771 struct sockaddr_in *fin = NULL, *lin = NULL;
3773 struct epoch_tracker et;
3775 struct sockaddr_in6 *fin6, *lin6;
3785 if (req->oldptr != NULL || req->oldlen != 0)
3787 if (req->newptr == NULL)
3789 if (req->newlen < sizeof(addrs))
3791 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3795 switch (addrs[0].ss_family) {
3798 fin6 = (struct sockaddr_in6 *)&addrs[0];
3799 lin6 = (struct sockaddr_in6 *)&addrs[1];
3800 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3801 lin6->sin6_len != sizeof(struct sockaddr_in6))
3803 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3804 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3806 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3807 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3809 fin = (struct sockaddr_in *)&addrs[0];
3810 lin = (struct sockaddr_in *)&addrs[1];
3814 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3817 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3824 fin = (struct sockaddr_in *)&addrs[0];
3825 lin = (struct sockaddr_in *)&addrs[1];
3826 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3827 lin->sin_len != sizeof(struct sockaddr_in))
3834 NET_EPOCH_ENTER(et);
3835 switch (addrs[0].ss_family) {
3838 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3839 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3840 INPLOOKUP_WLOCKPCB, NULL);
3845 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3846 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3852 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3853 inp->inp_socket == NULL) {
3859 so = inp->inp_socket;
3861 error = ktls_set_tx_mode(so,
3862 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3871 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3872 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3873 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3874 "Switch TCP connection to SW TLS");
3875 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3876 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3877 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3878 "Switch TCP connection to ifnet TLS");
3882 * Generate a standardized TCP log line for use throughout the
3883 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3884 * allow use in the interrupt context.
3886 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3887 * NB: The function may return NULL if memory allocation failed.
3889 * Due to header inclusion and ordering limitations the struct ip
3890 * and ip6_hdr pointers have to be passed as void pointers.
3893 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3897 /* Is logging enabled? */
3898 if (V_tcp_log_in_vain == 0)
3901 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3905 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3909 /* Is logging enabled? */
3910 if (tcp_log_debug == 0)
3913 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3917 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3923 const struct ip *ip = (const struct ip *)ip4hdr;
3926 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3930 * The log line looks like this:
3931 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3933 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3934 sizeof(PRINT_TH_FLAGS) + 1 +
3936 2 * INET6_ADDRSTRLEN;
3938 2 * INET_ADDRSTRLEN;
3941 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3945 strcat(s, "TCP: [");
3948 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3949 inet_ntoa_r(inc->inc_faddr, sp);
3951 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3953 inet_ntoa_r(inc->inc_laddr, sp);
3955 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3958 ip6_sprintf(sp, &inc->inc6_faddr);
3960 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3962 ip6_sprintf(sp, &inc->inc6_laddr);
3964 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3965 } else if (ip6 && th) {
3966 ip6_sprintf(sp, &ip6->ip6_src);
3968 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3970 ip6_sprintf(sp, &ip6->ip6_dst);
3972 sprintf(sp, "]:%i", ntohs(th->th_dport));
3975 } else if (ip && th) {
3976 inet_ntoa_r(ip->ip_src, sp);
3978 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3980 inet_ntoa_r(ip->ip_dst, sp);
3982 sprintf(sp, "]:%i", ntohs(th->th_dport));
3990 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3991 if (*(s + size - 1) != '\0')
3992 panic("%s: string too long", __func__);
3997 * A subroutine which makes it easy to track TCP state changes with DTrace.
3998 * This function shouldn't be called for t_state initializations that don't
3999 * correspond to actual TCP state transitions.
4002 tcp_state_change(struct tcpcb *tp, int newstate)
4004 #if defined(KDTRACE_HOOKS)
4005 int pstate = tp->t_state;
4008 TCPSTATES_DEC(tp->t_state);
4009 TCPSTATES_INC(newstate);
4010 tp->t_state = newstate;
4011 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4015 * Create an external-format (``xtcpcb'') structure using the information in
4016 * the kernel-format tcpcb structure pointed to by tp. This is done to
4017 * reduce the spew of irrelevant information over this interface, to isolate
4018 * user code from changes in the kernel structure, and potentially to provide
4019 * information-hiding if we decide that some of this information should be
4020 * hidden from users.
4023 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4025 struct tcpcb *tp = intotcpcb(inp);
4026 struct tcptw *tw = intotw(inp);
4029 bzero(xt, sizeof(*xt));
4030 if (inp->inp_flags & INP_TIMEWAIT) {
4031 xt->t_state = TCPS_TIME_WAIT;
4032 xt->xt_encaps_port = tw->t_port;
4034 xt->t_state = tp->t_state;
4035 xt->t_logstate = tp->t_logstate;
4036 xt->t_flags = tp->t_flags;
4037 xt->t_sndzerowin = tp->t_sndzerowin;
4038 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4039 xt->t_rcvoopack = tp->t_rcvoopack;
4040 xt->t_rcv_wnd = tp->rcv_wnd;
4041 xt->t_snd_wnd = tp->snd_wnd;
4042 xt->t_snd_cwnd = tp->snd_cwnd;
4043 xt->t_snd_ssthresh = tp->snd_ssthresh;
4044 xt->t_dsack_bytes = tp->t_dsack_bytes;
4045 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4046 xt->t_dsack_pack = tp->t_dsack_pack;
4047 xt->t_maxseg = tp->t_maxseg;
4048 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4049 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4051 now = getsbinuptime();
4052 #define COPYTIMER(ttt) do { \
4053 if (callout_active(&tp->t_timers->ttt)) \
4054 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
4059 COPYTIMER(tt_delack);
4060 COPYTIMER(tt_rexmt);
4061 COPYTIMER(tt_persist);
4065 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4067 xt->xt_encaps_port = tp->t_port;
4068 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4069 TCP_FUNCTION_NAME_LEN_MAX);
4070 bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4073 (void)tcp_log_get_id(tp, xt->xt_logid);
4077 xt->xt_len = sizeof(struct xtcpcb);
4078 in_pcbtoxinpcb(inp, &xt->xt_inp);
4079 if (inp->inp_socket == NULL)
4080 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4084 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4089 (status > TCP_EI_STATUS_MAX_VALUE) ||
4094 if (status > (sizeof(uint32_t) * 8)) {
4095 /* Should this be a KASSERT? */
4098 bit = 1U << (status - 1);
4099 if (bit & tp->t_end_info_status) {
4100 /* already logged */
4103 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4104 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4105 tp->t_end_info_bytes[i] = status;
4106 tp->t_end_info_status |= bit;
4113 tcp_can_enable_pacing(void)
4116 if ((tcp_pacing_limit == -1) ||
4117 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4118 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4119 shadow_num_connections = number_of_tcp_connections_pacing;
4126 static uint8_t tcp_pacing_warning = 0;
4129 tcp_decrement_paced_conn(void)
4133 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4134 shadow_num_connections = number_of_tcp_connections_pacing;
4135 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4137 if (tcp_pacing_limit != -1) {
4138 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4139 tcp_pacing_limit = 0;
4140 } else if (tcp_pacing_warning == 0) {
4141 printf("Warning pacing count is invalid, invalid decrement\n");
4142 tcp_pacing_warning = 1;