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>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_syncache.h>
110 #include <netinet/tcp_hpts.h>
111 #include <netinet/cc/cc.h>
113 #include <netinet6/tcp6_var.h>
115 #include <netinet/tcpip.h>
116 #include <netinet/tcp_fastopen.h>
118 #include <netinet/tcp_pcap.h>
121 #include <netinet/tcp_debug.h>
124 #include <netinet6/ip6protosw.h>
127 #include <netinet/tcp_offload.h>
129 #include <netinet/udp.h>
130 #include <netinet/udp_var.h>
132 #include <netipsec/ipsec_support.h>
134 #include <machine/in_cksum.h>
135 #include <crypto/siphash/siphash.h>
137 #include <security/mac/mac_framework.h>
139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
144 #ifdef NETFLIX_EXP_DETECTION
145 /* Sack attack detection thresholds and such */
146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
147 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
148 "Sack Attack detection thresholds");
149 int32_t tcp_force_detection = 0;
150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
152 &tcp_force_detection, 0,
153 "Do we force detection even if the INP has it off?");
154 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
157 &tcp_sack_to_ack_thresh, 700,
158 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
159 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
162 &tcp_sack_to_move_thresh, 600,
163 "Percentage of sack moves we must see above (10.1 percent is 101)");
164 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
167 &tcp_restoral_thresh, 550,
168 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
169 int32_t tcp_sad_decay_val = 800;
170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
172 &tcp_sad_decay_val, 800,
173 "The decay percentage (10.1 percent equals 101 )");
174 int32_t tcp_map_minimum = 500;
175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
177 &tcp_map_minimum, 500,
178 "Number of Map enteries before we start detection");
179 int32_t tcp_attack_on_turns_on_logging = 0;
180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
182 &tcp_attack_on_turns_on_logging, 0,
183 "When we have a positive hit on attack, do we turn on logging?");
184 int32_t tcp_sad_pacing_interval = 2000;
185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
187 &tcp_sad_pacing_interval, 2000,
188 "What is the minimum pacing interval for a classified attacker?");
190 int32_t tcp_sad_low_pps = 100;
191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
193 &tcp_sad_low_pps, 100,
194 "What is the input pps that below which we do not decay?");
196 uint32_t tcp_ack_war_time_window = 1000;
197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
199 &tcp_ack_war_time_window, 1000,
200 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
201 uint32_t tcp_ack_war_cnt = 5;
202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
205 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
207 struct rwlock tcp_function_lock;
210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
215 error = sysctl_handle_int(oidp, &new, 0, req);
216 if (error == 0 && req->newptr) {
217 if (new < TCP_MINMSS)
225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
226 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
227 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
228 "Default TCP Maximum Segment Size");
232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
236 new = V_tcp_v6mssdflt;
237 error = sysctl_handle_int(oidp, &new, 0, req);
238 if (error == 0 && req->newptr) {
239 if (new < TCP_MINMSS)
242 V_tcp_v6mssdflt = new;
247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
248 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
250 "Default TCP Maximum Segment Size for IPv6");
254 * Minimum MSS we accept and use. This prevents DoS attacks where
255 * we are forced to a ridiculous low MSS like 20 and send hundreds
256 * of packets instead of one. The effect scales with the available
257 * bandwidth and quickly saturates the CPU and network interface
258 * with packet generation and sending. Set to zero to disable MINMSS
259 * checking. This setting prevents us from sending too small packets.
261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
263 &VNET_NAME(tcp_minmss), 0,
264 "Minimum TCP Maximum Segment Size");
266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
268 &VNET_NAME(tcp_do_rfc1323), 0,
269 "Enable rfc1323 (high performance TCP) extensions");
272 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
273 * Some stacks negotiate TS, but never send them after connection setup. Some
274 * stacks negotiate TS, but don't send them when sending keep-alive segments.
275 * These include modern widely deployed TCP stacks.
276 * Therefore tolerating violations for now...
278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
280 &VNET_NAME(tcp_tolerate_missing_ts), 0,
281 "Tolerate missing TCP timestamps");
283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
285 &VNET_NAME(tcp_ts_offset_per_conn), 0,
286 "Initialize TCP timestamps per connection instead of per host pair");
288 /* How many connections are pacing */
289 static volatile uint32_t number_of_tcp_connections_pacing = 0;
290 static uint32_t shadow_num_connections = 0;
292 static int tcp_pacing_limit = 10000;
293 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
294 &tcp_pacing_limit, 1000,
295 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
297 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
298 &shadow_num_connections, 0, "Number of TCP connections being paced");
300 static int tcp_log_debug = 0;
301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
302 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
304 static int tcp_tcbhashsize;
305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
306 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
308 static int do_tcpdrain = 1;
309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
310 "Enable tcp_drain routine for extra help when low on mbufs");
312 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
313 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
315 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
316 #define V_icmp_may_rst VNET(icmp_may_rst)
317 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
318 &VNET_NAME(icmp_may_rst), 0,
319 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
321 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
322 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
324 &VNET_NAME(tcp_isn_reseed_interval), 0,
325 "Seconds between reseeding of ISN secret");
327 static int tcp_soreceive_stream;
328 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
329 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
331 VNET_DEFINE(uma_zone_t, sack_hole_zone);
332 #define V_sack_hole_zone VNET(sack_hole_zone)
333 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
335 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
340 new = V_tcp_map_entries_limit;
341 error = sysctl_handle_int(oidp, &new, 0, req);
342 if (error == 0 && req->newptr) {
343 /* only allow "0" and value > minimum */
344 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
347 V_tcp_map_entries_limit = new;
351 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
352 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
353 &VNET_NAME(tcp_map_entries_limit), 0,
354 &sysctl_net_inet_tcp_map_limit_check, "IU",
355 "Total sendmap entries limit");
357 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
358 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
359 &VNET_NAME(tcp_map_split_limit), 0,
360 "Total sendmap split entries limit");
363 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
366 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
367 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
368 #define V_ts_offset_secret VNET(ts_offset_secret)
370 static int tcp_default_fb_init(struct tcpcb *tp);
371 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
372 static int tcp_default_handoff_ok(struct tcpcb *tp);
373 static struct inpcb *tcp_notify(struct inpcb *, int);
374 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
375 static void tcp_mtudisc(struct inpcb *, int);
376 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
377 void *ip4hdr, const void *ip6hdr);
379 static struct tcp_function_block tcp_def_funcblk = {
380 .tfb_tcp_block_name = "freebsd",
381 .tfb_tcp_output = tcp_output,
382 .tfb_tcp_do_segment = tcp_do_segment,
383 .tfb_tcp_ctloutput = tcp_default_ctloutput,
384 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
385 .tfb_tcp_fb_init = tcp_default_fb_init,
386 .tfb_tcp_fb_fini = tcp_default_fb_fini,
389 static int tcp_fb_cnt = 0;
390 struct tcp_funchead t_functions;
391 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
393 static struct tcp_function_block *
394 find_tcp_functions_locked(struct tcp_function_set *fs)
396 struct tcp_function *f;
397 struct tcp_function_block *blk=NULL;
399 TAILQ_FOREACH(f, &t_functions, tf_next) {
400 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
408 static struct tcp_function_block *
409 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
411 struct tcp_function_block *rblk=NULL;
412 struct tcp_function *f;
414 TAILQ_FOREACH(f, &t_functions, tf_next) {
415 if (f->tf_fb == blk) {
426 struct tcp_function_block *
427 find_and_ref_tcp_functions(struct tcp_function_set *fs)
429 struct tcp_function_block *blk;
431 rw_rlock(&tcp_function_lock);
432 blk = find_tcp_functions_locked(fs);
434 refcount_acquire(&blk->tfb_refcnt);
435 rw_runlock(&tcp_function_lock);
439 struct tcp_function_block *
440 find_and_ref_tcp_fb(struct tcp_function_block *blk)
442 struct tcp_function_block *rblk;
444 rw_rlock(&tcp_function_lock);
445 rblk = find_tcp_fb_locked(blk, NULL);
447 refcount_acquire(&rblk->tfb_refcnt);
448 rw_runlock(&tcp_function_lock);
452 static struct tcp_function_block *
453 find_and_ref_tcp_default_fb(void)
455 struct tcp_function_block *rblk;
457 rw_rlock(&tcp_function_lock);
458 rblk = tcp_func_set_ptr;
459 refcount_acquire(&rblk->tfb_refcnt);
460 rw_runlock(&tcp_function_lock);
465 tcp_switch_back_to_default(struct tcpcb *tp)
467 struct tcp_function_block *tfb;
469 KASSERT(tp->t_fb != &tcp_def_funcblk,
470 ("%s: called by the built-in default stack", __func__));
473 * Release the old stack. This function will either find a new one
476 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
477 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
478 refcount_release(&tp->t_fb->tfb_refcnt);
481 * Now, we'll find a new function block to use.
482 * Start by trying the current user-selected
483 * default, unless this stack is the user-selected
486 tfb = find_and_ref_tcp_default_fb();
487 if (tfb == tp->t_fb) {
488 refcount_release(&tfb->tfb_refcnt);
491 /* Does the stack accept this connection? */
492 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
493 (*tfb->tfb_tcp_handoff_ok)(tp)) {
494 refcount_release(&tfb->tfb_refcnt);
497 /* Try to use that stack. */
499 /* Initialize the new stack. If it succeeds, we are done. */
501 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
502 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
506 * Initialization failed. Release the reference count on
509 refcount_release(&tfb->tfb_refcnt);
513 * If that wasn't feasible, use the built-in default
514 * stack which is not allowed to reject anyone.
516 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
518 /* there always should be a default */
519 panic("Can't refer to tcp_def_funcblk");
521 if (tfb->tfb_tcp_handoff_ok != NULL) {
522 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
523 /* The default stack cannot say no */
524 panic("Default stack rejects a new session?");
528 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
529 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
530 /* The default stack cannot fail */
531 panic("Default stack initialization failed");
536 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
537 const struct sockaddr *sa, void *ctx)
548 TCPSTAT_INC(tcps_tunneled_pkts);
549 if ((m->m_flags & M_PKTHDR) == 0) {
550 /* Can't handle one that is not a pkt hdr */
551 TCPSTAT_INC(tcps_tunneled_errs);
554 thlen = sizeof(struct tcphdr);
555 if (m->m_len < off + sizeof(struct udphdr) + thlen &&
556 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
557 TCPSTAT_INC(tcps_tunneled_errs);
560 iph = mtod(m, struct ip *);
561 uh = (struct udphdr *)((caddr_t)iph + off);
562 th = (struct tcphdr *)(uh + 1);
563 thlen = th->th_off << 2;
564 if (m->m_len < off + sizeof(struct udphdr) + thlen) {
565 m = m_pullup(m, off + sizeof(struct udphdr) + thlen);
567 TCPSTAT_INC(tcps_tunneled_errs);
570 iph = mtod(m, struct ip *);
571 uh = (struct udphdr *)((caddr_t)iph + off);
572 th = (struct tcphdr *)(uh + 1);
575 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
576 bcopy(th, uh, m->m_len - off);
577 m->m_len -= sizeof(struct udphdr);
578 m->m_pkthdr.len -= sizeof(struct udphdr);
580 * We use the same algorithm for
581 * both UDP and TCP for c-sum. So
582 * the code in tcp_input will skip
583 * the checksum. So we do nothing
584 * with the flag (m->m_pkthdr.csum_flags).
589 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
590 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
594 case IPV6_VERSION >> 4:
595 ip6 = mtod(m, struct ip6_hdr *);
596 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
597 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
610 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
613 struct tcp_function_set fs;
614 struct tcp_function_block *blk;
616 memset(&fs, 0, sizeof(fs));
617 rw_rlock(&tcp_function_lock);
618 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
621 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
622 fs.pcbcnt = blk->tfb_refcnt;
624 rw_runlock(&tcp_function_lock);
625 error = sysctl_handle_string(oidp, fs.function_set_name,
626 sizeof(fs.function_set_name), req);
628 /* Check for error or no change */
629 if (error != 0 || req->newptr == NULL)
632 rw_wlock(&tcp_function_lock);
633 blk = find_tcp_functions_locked(&fs);
635 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
639 tcp_func_set_ptr = blk;
641 rw_wunlock(&tcp_function_lock);
645 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
646 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
647 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
648 "Set/get the default TCP functions");
651 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
653 int error, cnt, linesz;
654 struct tcp_function *f;
660 rw_rlock(&tcp_function_lock);
661 TAILQ_FOREACH(f, &t_functions, tf_next) {
664 rw_runlock(&tcp_function_lock);
666 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
667 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
672 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
673 "Alias", "PCB count");
678 rw_rlock(&tcp_function_lock);
679 TAILQ_FOREACH(f, &t_functions, tf_next) {
680 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
681 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
682 f->tf_fb->tfb_tcp_block_name,
683 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
684 alias ? f->tf_name : "-",
685 f->tf_fb->tfb_refcnt);
686 if (linesz >= bufsz) {
694 rw_runlock(&tcp_function_lock);
696 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
697 free(buffer, M_TEMP);
701 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
702 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
703 NULL, 0, sysctl_net_inet_list_available, "A",
704 "list available TCP Function sets");
706 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
709 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
710 #define V_udp4_tun_socket VNET(udp4_tun_socket)
713 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
714 #define V_udp6_tun_socket VNET(udp6_tun_socket)
718 tcp_over_udp_stop(void)
721 * This function assumes sysctl caller holds inp_rinfo_lock()
725 if (V_udp4_tun_socket != NULL) {
726 soclose(V_udp4_tun_socket);
727 V_udp4_tun_socket = NULL;
731 if (V_udp6_tun_socket != NULL) {
732 soclose(V_udp6_tun_socket);
733 V_udp6_tun_socket = NULL;
739 tcp_over_udp_start(void)
744 struct sockaddr_in sin;
747 struct sockaddr_in6 sin6;
750 * This function assumes sysctl caller holds inp_info_rlock()
753 port = V_tcp_udp_tunneling_port;
754 if (ntohs(port) == 0) {
755 /* Must have a port set */
759 if (V_udp4_tun_socket != NULL) {
760 /* Already running -- must stop first */
765 if (V_udp6_tun_socket != NULL) {
766 /* Already running -- must stop first */
771 if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
772 SOCK_DGRAM, IPPROTO_UDP,
773 curthread->td_ucred, curthread))) {
777 /* Call the special UDP hook. */
778 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
779 tcp_recv_udp_tunneled_packet,
785 /* Ok, we have a socket, bind it to the port. */
786 memset(&sin, 0, sizeof(struct sockaddr_in));
787 sin.sin_len = sizeof(struct sockaddr_in);
788 sin.sin_family = AF_INET;
789 sin.sin_port = htons(port);
790 if ((ret = sobind(V_udp4_tun_socket,
791 (struct sockaddr *)&sin, curthread))) {
797 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
798 SOCK_DGRAM, IPPROTO_UDP,
799 curthread->td_ucred, curthread))) {
803 /* Call the special UDP hook. */
804 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
805 tcp_recv_udp_tunneled_packet,
806 tcp6_ctlinput_viaudp,
811 /* Ok, we have a socket, bind it to the port. */
812 memset(&sin6, 0, sizeof(struct sockaddr_in6));
813 sin6.sin6_len = sizeof(struct sockaddr_in6);
814 sin6.sin6_family = AF_INET6;
815 sin6.sin6_port = htons(port);
816 if ((ret = sobind(V_udp6_tun_socket,
817 (struct sockaddr *)&sin6, curthread))) {
826 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
831 old = V_tcp_udp_tunneling_port;
833 error = sysctl_handle_int(oidp, &new, 0, req);
835 (req->newptr != NULL)) {
836 if ((new < TCP_TUNNELING_PORT_MIN) ||
837 (new > TCP_TUNNELING_PORT_MAX)) {
840 V_tcp_udp_tunneling_port = new;
845 error = tcp_over_udp_start();
852 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
853 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
854 &VNET_NAME(tcp_udp_tunneling_port),
855 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
856 "Tunneling port for tcp over udp");
858 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
861 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
865 new = V_tcp_udp_tunneling_overhead;
866 error = sysctl_handle_int(oidp, &new, 0, req);
867 if (error == 0 && req->newptr) {
868 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
869 (new > TCP_TUNNELING_OVERHEAD_MAX))
872 V_tcp_udp_tunneling_overhead = new;
877 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
878 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
879 &VNET_NAME(tcp_udp_tunneling_overhead),
880 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
881 "MSS reduction when using tcp over udp");
884 * Exports one (struct tcp_function_info) for each alias/name.
887 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
890 struct tcp_function *f;
891 struct tcp_function_info tfi;
894 * We don't allow writes.
896 if (req->newptr != NULL)
900 * Wire the old buffer so we can directly copy the functions to
901 * user space without dropping the lock.
903 if (req->oldptr != NULL) {
904 error = sysctl_wire_old_buffer(req, 0);
910 * Walk the list and copy out matching entries. If INVARIANTS
911 * is compiled in, also walk the list to verify the length of
912 * the list matches what we have recorded.
914 rw_rlock(&tcp_function_lock);
918 if (req->oldptr == NULL) {
923 TAILQ_FOREACH(f, &t_functions, tf_next) {
927 if (req->oldptr != NULL) {
928 bzero(&tfi, sizeof(tfi));
929 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
930 tfi.tfi_id = f->tf_fb->tfb_id;
931 (void)strlcpy(tfi.tfi_alias, f->tf_name,
932 sizeof(tfi.tfi_alias));
933 (void)strlcpy(tfi.tfi_name,
934 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
935 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
937 * Don't stop on error, as that is the
938 * mechanism we use to accumulate length
939 * information if the buffer was too short.
943 KASSERT(cnt == tcp_fb_cnt,
944 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
948 rw_runlock(&tcp_function_lock);
949 if (req->oldptr == NULL)
950 error = SYSCTL_OUT(req, NULL,
951 (cnt + 1) * sizeof(struct tcp_function_info));
956 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
957 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
958 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
959 "List TCP function block name-to-ID mappings");
962 * tfb_tcp_handoff_ok() function for the default stack.
963 * Note that we'll basically try to take all comers.
966 tcp_default_handoff_ok(struct tcpcb *tp)
973 * tfb_tcp_fb_init() function for the default stack.
975 * This handles making sure we have appropriate timers set if you are
976 * transitioning a socket that has some amount of setup done.
978 * The init() fuction from the default can *never* return non-zero i.e.
979 * it is required to always succeed since it is the stack of last resort!
982 tcp_default_fb_init(struct tcpcb *tp)
987 INP_WLOCK_ASSERT(tp->t_inpcb);
989 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
990 ("%s: connection %p in unexpected state %d", __func__, tp,
994 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
995 * know what to do for unexpected states (which includes TIME_WAIT).
997 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1001 * Make sure some kind of transmission timer is set if there is
1004 so = tp->t_inpcb->inp_socket;
1005 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1006 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1007 tcp_timer_active(tp, TT_PERSIST))) {
1009 * If the session has established and it looks like it should
1010 * be in the persist state, set the persist timer. Otherwise,
1011 * set the retransmit timer.
1013 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1014 (int32_t)(tp->snd_nxt - tp->snd_una) <
1015 (int32_t)sbavail(&so->so_snd))
1018 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1021 /* All non-embryonic sessions get a keepalive timer. */
1022 if (!tcp_timer_active(tp, TT_KEEP))
1023 tcp_timer_activate(tp, TT_KEEP,
1024 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1028 * Make sure critical variables are initialized
1029 * if transitioning while in Recovery.
1031 if IN_FASTRECOVERY(tp->t_flags) {
1032 if (tp->sackhint.recover_fs == 0)
1033 tp->sackhint.recover_fs = max(1,
1034 tp->snd_nxt - tp->snd_una);
1041 * tfb_tcp_fb_fini() function for the default stack.
1043 * This changes state as necessary (or prudent) to prepare for another stack
1044 * to assume responsibility for the connection.
1047 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1050 INP_WLOCK_ASSERT(tp->t_inpcb);
1055 * Target size of TCP PCB hash tables. Must be a power of two.
1057 * Note that this can be overridden by the kernel environment
1058 * variable net.inet.tcp.tcbhashsize
1061 #define TCBHASHSIZE 0
1066 * Callouts should be moved into struct tcp directly. They are currently
1067 * separate because the tcpcb structure is exported to userland for sysctl
1068 * parsing purposes, which do not know about callouts.
1072 struct tcp_timer tt;
1079 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1080 #define V_tcpcb_zone VNET(tcpcb_zone)
1082 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1083 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1085 static struct mtx isn_mtx;
1087 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1088 #define ISN_LOCK() mtx_lock(&isn_mtx)
1089 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
1092 * TCP initialization.
1095 tcp_zone_change(void *tag)
1098 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
1099 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1100 tcp_tw_zone_change();
1104 tcp_inpcb_init(void *mem, int size, int flags)
1106 struct inpcb *inp = mem;
1108 INP_LOCK_INIT(inp, "inp", "tcpinp");
1113 * Take a value and get the next power of 2 that doesn't overflow.
1114 * Used to size the tcp_inpcb hash buckets.
1117 maketcp_hashsize(int size)
1123 * get the next power of 2 higher than maxsockets.
1125 hashsize = 1 << fls(size);
1126 /* catch overflow, and just go one power of 2 smaller */
1127 if (hashsize < size) {
1128 hashsize = 1 << (fls(size) - 1);
1133 static volatile int next_tcp_stack_id = 1;
1136 * Register a TCP function block with the name provided in the names
1137 * array. (Note that this function does NOT automatically register
1138 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
1139 * explicitly include blk->tfb_tcp_block_name in the list of names if
1140 * you wish to register the stack with that name.)
1142 * Either all name registrations will succeed or all will fail. If
1143 * a name registration fails, the function will update the num_names
1144 * argument to point to the array index of the name that encountered
1147 * Returns 0 on success, or an error code on failure.
1150 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1151 const char *names[], int *num_names)
1153 struct tcp_function *n;
1154 struct tcp_function_set fs;
1157 KASSERT(names != NULL && *num_names > 0,
1158 ("%s: Called with 0-length name list", __func__));
1159 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1160 KASSERT(rw_initialized(&tcp_function_lock),
1161 ("%s: called too early", __func__));
1163 if ((blk->tfb_tcp_output == NULL) ||
1164 (blk->tfb_tcp_do_segment == NULL) ||
1165 (blk->tfb_tcp_ctloutput == NULL) ||
1166 (strlen(blk->tfb_tcp_block_name) == 0)) {
1168 * These functions are required and you
1174 if (blk->tfb_tcp_timer_stop_all ||
1175 blk->tfb_tcp_timer_activate ||
1176 blk->tfb_tcp_timer_active ||
1177 blk->tfb_tcp_timer_stop) {
1179 * If you define one timer function you
1180 * must have them all.
1182 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1183 (blk->tfb_tcp_timer_activate == NULL) ||
1184 (blk->tfb_tcp_timer_active == NULL) ||
1185 (blk->tfb_tcp_timer_stop == NULL)) {
1191 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1196 refcount_init(&blk->tfb_refcnt, 0);
1197 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1198 for (i = 0; i < *num_names; i++) {
1199 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1206 (void)strlcpy(fs.function_set_name, names[i],
1207 sizeof(fs.function_set_name));
1208 rw_wlock(&tcp_function_lock);
1209 if (find_tcp_functions_locked(&fs) != NULL) {
1210 /* Duplicate name space not allowed */
1211 rw_wunlock(&tcp_function_lock);
1212 free(n, M_TCPFUNCTIONS);
1216 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1217 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1219 rw_wunlock(&tcp_function_lock);
1225 * Deregister the names we just added. Because registration failed
1226 * for names[i], we don't need to deregister that name.
1229 rw_wlock(&tcp_function_lock);
1231 TAILQ_FOREACH(n, &t_functions, tf_next) {
1232 if (!strncmp(n->tf_name, names[i],
1233 TCP_FUNCTION_NAME_LEN_MAX)) {
1234 TAILQ_REMOVE(&t_functions, n, tf_next);
1237 free(n, M_TCPFUNCTIONS);
1242 rw_wunlock(&tcp_function_lock);
1247 * Register a TCP function block using the name provided in the name
1250 * Returns 0 on success, or an error code on failure.
1253 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1256 const char *name_list[1];
1261 name_list[0] = name;
1263 name_list[0] = blk->tfb_tcp_block_name;
1264 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1269 * Register a TCP function block using the name defined in
1270 * blk->tfb_tcp_block_name.
1272 * Returns 0 on success, or an error code on failure.
1275 register_tcp_functions(struct tcp_function_block *blk, int wait)
1278 return (register_tcp_functions_as_name(blk, NULL, wait));
1282 * Deregister all names associated with a function block. This
1283 * functionally removes the function block from use within the system.
1285 * When called with a true quiesce argument, mark the function block
1286 * as being removed so no more stacks will use it and determine
1287 * whether the removal would succeed.
1289 * When called with a false quiesce argument, actually attempt the
1292 * When called with a force argument, attempt to switch all TCBs to
1293 * use the default stack instead of returning EBUSY.
1295 * Returns 0 on success (or if the removal would succeed, or an error
1299 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1302 struct tcp_function *f;
1304 if (blk == &tcp_def_funcblk) {
1305 /* You can't un-register the default */
1308 rw_wlock(&tcp_function_lock);
1309 if (blk == tcp_func_set_ptr) {
1310 /* You can't free the current default */
1311 rw_wunlock(&tcp_function_lock);
1314 /* Mark the block so no more stacks can use it. */
1315 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1317 * If TCBs are still attached to the stack, attempt to switch them
1318 * to the default stack.
1320 if (force && blk->tfb_refcnt) {
1323 VNET_ITERATOR_DECL(vnet_iter);
1325 rw_wunlock(&tcp_function_lock);
1328 VNET_FOREACH(vnet_iter) {
1329 CURVNET_SET(vnet_iter);
1330 INP_INFO_WLOCK(&V_tcbinfo);
1331 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1333 if (inp->inp_flags & INP_TIMEWAIT) {
1337 tp = intotcpcb(inp);
1338 if (tp == NULL || tp->t_fb != blk) {
1342 tcp_switch_back_to_default(tp);
1345 INP_INFO_WUNLOCK(&V_tcbinfo);
1348 VNET_LIST_RUNLOCK();
1350 rw_wlock(&tcp_function_lock);
1352 if (blk->tfb_refcnt) {
1353 /* TCBs still attached. */
1354 rw_wunlock(&tcp_function_lock);
1359 rw_wunlock(&tcp_function_lock);
1362 /* Remove any function names that map to this function block. */
1363 while (find_tcp_fb_locked(blk, &f) != NULL) {
1364 TAILQ_REMOVE(&t_functions, f, tf_next);
1367 free(f, M_TCPFUNCTIONS);
1369 rw_wunlock(&tcp_function_lock);
1376 const char *tcbhash_tuneable;
1379 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1382 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1383 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1384 printf("%s: WARNING: unable to register helper hook\n", __func__);
1385 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1386 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1387 printf("%s: WARNING: unable to register helper hook\n", __func__);
1390 if (tcp_stats_init())
1391 printf("%s: WARNING: unable to initialise TCP stats\n",
1394 hashsize = TCBHASHSIZE;
1395 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1396 if (hashsize == 0) {
1398 * Auto tune the hash size based on maxsockets.
1399 * A perfect hash would have a 1:1 mapping
1400 * (hashsize = maxsockets) however it's been
1401 * suggested that O(2) average is better.
1403 hashsize = maketcp_hashsize(maxsockets / 4);
1405 * Our historical default is 512,
1406 * do not autotune lower than this.
1410 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1411 printf("%s: %s auto tuned to %d\n", __func__,
1412 tcbhash_tuneable, hashsize);
1415 * We require a hashsize to be a power of two.
1416 * Previously if it was not a power of two we would just reset it
1417 * back to 512, which could be a nasty surprise if you did not notice
1418 * the error message.
1419 * Instead what we do is clip it to the closest power of two lower
1420 * than the specified hash value.
1422 if (!powerof2(hashsize)) {
1423 int oldhashsize = hashsize;
1425 hashsize = maketcp_hashsize(hashsize);
1426 /* prevent absurdly low value */
1429 printf("%s: WARNING: TCB hash size not a power of 2, "
1430 "clipped from %d to %d.\n", __func__, oldhashsize,
1433 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1434 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1437 * These have to be type stable for the benefit of the timers.
1439 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1440 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1441 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1442 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1448 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1449 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1450 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1452 tcp_fastopen_init();
1454 /* Skip initialization of globals for non-default instances. */
1455 if (!IS_DEFAULT_VNET(curvnet))
1458 tcp_reass_global_init();
1460 /* XXX virtualize those bellow? */
1461 tcp_delacktime = TCPTV_DELACK;
1462 tcp_keepinit = TCPTV_KEEP_INIT;
1463 tcp_keepidle = TCPTV_KEEP_IDLE;
1464 tcp_keepintvl = TCPTV_KEEPINTVL;
1465 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1466 tcp_msl = TCPTV_MSL;
1467 tcp_rexmit_initial = TCPTV_RTOBASE;
1468 if (tcp_rexmit_initial < 1)
1469 tcp_rexmit_initial = 1;
1470 tcp_rexmit_min = TCPTV_MIN;
1471 if (tcp_rexmit_min < 1)
1473 tcp_persmin = TCPTV_PERSMIN;
1474 tcp_persmax = TCPTV_PERSMAX;
1475 tcp_rexmit_slop = TCPTV_CPU_VAR;
1476 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1477 tcp_tcbhashsize = hashsize;
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);
1512 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1513 EVENTHANDLER_PRI_ANY);
1515 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1516 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1517 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1518 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1519 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1520 tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1521 tcp_comp_total = counter_u64_alloc(M_WAITOK);
1522 tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1523 tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1531 tcp_destroy(void *unused __unused)
1539 * All our processes are gone, all our sockets should be cleaned
1540 * up, which means, we should be past the tcp_discardcb() calls.
1541 * Sleep to let all tcpcb timers really disappear and cleanup.
1544 INP_LIST_RLOCK(&V_tcbinfo);
1545 n = V_tcbinfo.ipi_count;
1546 INP_LIST_RUNLOCK(&V_tcbinfo);
1549 pause("tcpdes", hz / 10);
1554 in_pcbinfo_destroy(&V_tcbinfo);
1555 /* tcp_discardcb() clears the sack_holes up. */
1556 uma_zdestroy(V_sack_hole_zone);
1557 uma_zdestroy(V_tcpcb_zone);
1560 * Cannot free the zone until all tcpcbs are released as we attach
1561 * the allocations to them.
1563 tcp_fastopen_destroy();
1566 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1568 printf("%s: WARNING: unable to deregister helper hook "
1569 "type=%d, id=%d: error %d returned\n", __func__,
1570 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1572 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1574 printf("%s: WARNING: unable to deregister helper hook "
1575 "type=%d, id=%d: error %d returned\n", __func__,
1576 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1580 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1590 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1591 * tcp_template used to store this data in mbufs, but we now recopy it out
1592 * of the tcpcb each time to conserve mbufs.
1595 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1597 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1599 INP_WLOCK_ASSERT(inp);
1602 if ((inp->inp_vflag & INP_IPV6) != 0) {
1603 struct ip6_hdr *ip6;
1605 ip6 = (struct ip6_hdr *)ip_ptr;
1606 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1607 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1608 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1609 (IPV6_VERSION & IPV6_VERSION_MASK);
1611 ip6->ip6_nxt = IPPROTO_TCP;
1613 ip6->ip6_nxt = IPPROTO_UDP;
1614 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1615 ip6->ip6_src = inp->in6p_laddr;
1616 ip6->ip6_dst = inp->in6p_faddr;
1619 #if defined(INET6) && defined(INET)
1626 ip = (struct ip *)ip_ptr;
1627 ip->ip_v = IPVERSION;
1629 ip->ip_tos = inp->inp_ip_tos;
1633 ip->ip_ttl = inp->inp_ip_ttl;
1636 ip->ip_p = IPPROTO_TCP;
1638 ip->ip_p = IPPROTO_UDP;
1639 ip->ip_src = inp->inp_laddr;
1640 ip->ip_dst = inp->inp_faddr;
1643 th->th_sport = inp->inp_lport;
1644 th->th_dport = inp->inp_fport;
1652 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1656 * Create template to be used to send tcp packets on a connection.
1657 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1658 * use for this function is in keepalives, which use tcp_respond.
1661 tcpip_maketemplate(struct inpcb *inp)
1665 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1668 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1673 * Send a single message to the TCP at address specified by
1674 * the given TCP/IP header. If m == NULL, then we make a copy
1675 * of the tcpiphdr at th and send directly to the addressed host.
1676 * This is used to force keep alive messages out using the TCP
1677 * template for a connection. If flags are given then we send
1678 * a message back to the TCP which originated the segment th,
1679 * and discard the mbuf containing it and any other attached mbufs.
1681 * In any case the ack and sequence number of the transmitted
1682 * segment are as specified by the parameters.
1684 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1687 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1688 tcp_seq ack, tcp_seq seq, int flags)
1694 struct udphdr *uh = NULL;
1698 struct ip6_hdr *ip6;
1701 int optlen, tlen, win, ulen;
1705 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1709 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1716 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1717 INP_LOCK_ASSERT(inp);
1723 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1724 port = m->m_pkthdr.tcp_tun_port;
1727 if (ip && (ip->ip_p == IPPROTO_UDP))
1728 port = m->m_pkthdr.tcp_tun_port;
1737 if (!(flags & TH_RST)) {
1738 win = sbspace(&inp->inp_socket->so_rcv);
1739 if (win > TCP_MAXWIN << tp->rcv_scale)
1740 win = TCP_MAXWIN << tp->rcv_scale;
1742 if ((tp->t_flags & TF_NOOPT) == 0)
1746 m = m_gethdr(M_NOWAIT, MT_DATA);
1749 m->m_data += max_linkhdr;
1752 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1753 sizeof(struct ip6_hdr));
1754 ip6 = mtod(m, struct ip6_hdr *);
1755 nth = (struct tcphdr *)(ip6 + 1);
1757 /* Insert a UDP header */
1758 uh = (struct udphdr *)nth;
1759 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1760 uh->uh_dport = port;
1761 nth = (struct tcphdr *)(uh + 1);
1766 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1767 ip = mtod(m, struct ip *);
1768 nth = (struct tcphdr *)(ip + 1);
1770 /* Insert a UDP header */
1771 uh = (struct udphdr *)nth;
1772 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1773 uh->uh_dport = port;
1774 nth = (struct tcphdr *)(uh + 1);
1777 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1779 } else if ((!M_WRITABLE(m)) || (port != 0)) {
1782 /* Can't reuse 'm', allocate a new mbuf. */
1783 n = m_gethdr(M_NOWAIT, MT_DATA);
1789 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1795 n->m_data += max_linkhdr;
1796 /* m_len is set later */
1797 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1800 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1801 sizeof(struct ip6_hdr));
1802 ip6 = mtod(n, struct ip6_hdr *);
1803 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1804 nth = (struct tcphdr *)(ip6 + 1);
1806 /* Insert a UDP header */
1807 uh = (struct udphdr *)nth;
1808 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1809 uh->uh_dport = port;
1810 nth = (struct tcphdr *)(uh + 1);
1815 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1816 ip = mtod(n, struct ip *);
1817 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1818 nth = (struct tcphdr *)(ip + 1);
1820 /* Insert a UDP header */
1821 uh = (struct udphdr *)nth;
1822 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1823 uh->uh_dport = port;
1824 nth = (struct tcphdr *)(uh + 1);
1827 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1828 xchg(nth->th_dport, nth->th_sport, uint16_t);
1835 * XXX MRT We inherit the FIB, which is lucky.
1839 m->m_data = (caddr_t)ipgen;
1840 /* m_len is set later */
1843 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1844 nth = (struct tcphdr *)(ip6 + 1);
1848 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1849 nth = (struct tcphdr *)(ip + 1);
1853 * this is usually a case when an extension header
1854 * exists between the IPv6 header and the
1857 nth->th_sport = th->th_sport;
1858 nth->th_dport = th->th_dport;
1860 xchg(nth->th_dport, nth->th_sport, uint16_t);
1866 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1868 #if defined(INET) && defined(INET6)
1872 tlen = sizeof (struct tcpiphdr);
1875 tlen += sizeof (struct udphdr);
1878 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1879 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1880 m, tlen, (long)M_TRAILINGSPACE(m)));
1885 /* Make sure we have room. */
1886 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1887 m->m_next = m_get(M_NOWAIT, MT_DATA);
1889 optp = mtod(m->m_next, u_char *);
1894 optp = (u_char *) (nth + 1);
1900 if (tp->t_flags & TF_RCVD_TSTMP) {
1901 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1902 to.to_tsecr = tp->ts_recent;
1903 to.to_flags |= TOF_TS;
1905 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1906 /* TCP-MD5 (RFC2385). */
1907 if (tp->t_flags & TF_SIGNATURE)
1908 to.to_flags |= TOF_SIGNATURE;
1910 /* Add the options. */
1911 tlen += optlen = tcp_addoptions(&to, optp);
1913 /* Update m_len in the correct mbuf. */
1914 optm->m_len += optlen;
1920 ulen = tlen - sizeof(struct ip6_hdr);
1921 uh->uh_ulen = htons(ulen);
1924 ip6->ip6_vfc = IPV6_VERSION;
1926 ip6->ip6_nxt = IPPROTO_UDP;
1928 ip6->ip6_nxt = IPPROTO_TCP;
1929 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1932 #if defined(INET) && defined(INET6)
1938 ulen = tlen - sizeof(struct ip);
1939 uh->uh_ulen = htons(ulen);
1941 ip->ip_len = htons(tlen);
1942 ip->ip_ttl = V_ip_defttl;
1944 ip->ip_p = IPPROTO_UDP;
1946 ip->ip_p = IPPROTO_TCP;
1948 if (V_path_mtu_discovery)
1949 ip->ip_off |= htons(IP_DF);
1952 m->m_pkthdr.len = tlen;
1953 m->m_pkthdr.rcvif = NULL;
1957 * Packet is associated with a socket, so allow the
1958 * label of the response to reflect the socket label.
1960 INP_LOCK_ASSERT(inp);
1961 mac_inpcb_create_mbuf(inp, m);
1964 * Packet is not associated with a socket, so possibly
1965 * update the label in place.
1967 mac_netinet_tcp_reply(m);
1970 nth->th_seq = htonl(seq);
1971 nth->th_ack = htonl(ack);
1973 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1974 nth->th_flags = flags;
1976 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1978 nth->th_win = htons((u_short)win);
1981 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1982 if (to.to_flags & TOF_SIGNATURE) {
1983 if (!TCPMD5_ENABLED() ||
1984 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1994 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
1995 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1996 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
1999 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2000 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2001 nth->th_sum = in6_cksum_pseudo(ip6,
2002 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2004 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2008 #if defined(INET6) && defined(INET)
2014 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2015 htons(ulen + IPPROTO_UDP));
2016 m->m_pkthdr.csum_flags = CSUM_UDP;
2017 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2020 m->m_pkthdr.csum_flags = CSUM_TCP;
2021 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2022 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2023 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2028 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2029 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2031 TCP_PROBE3(debug__output, tp, th, m);
2033 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2037 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2038 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2041 #if defined(INET) && defined(INET6)
2046 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2047 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
2053 * Create a new TCP control block, making an
2054 * empty reassembly queue and hooking it to the argument
2055 * protocol control block. The `inp' parameter must have
2056 * come from the zone allocator set up in tcp_init().
2059 tcp_newtcpcb(struct inpcb *inp)
2061 struct tcpcb_mem *tm;
2064 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2067 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2072 /* Initialise cc_var struct for this tcpcb. */
2074 tp->ccv->type = IPPROTO_TCP;
2075 tp->ccv->ccvc.tcp = tp;
2076 rw_rlock(&tcp_function_lock);
2077 tp->t_fb = tcp_func_set_ptr;
2078 refcount_acquire(&tp->t_fb->tfb_refcnt);
2079 rw_runlock(&tcp_function_lock);
2081 * Use the current system default CC algorithm.
2084 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2085 CC_ALGO(tp) = CC_DEFAULT();
2088 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2091 in_pcbref(inp); /* Reference for tcpcb */
2094 if (CC_ALGO(tp)->cb_init != NULL)
2095 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
2096 if (tp->t_fb->tfb_tcp_fb_fini)
2097 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2098 in_pcbrele_wlocked(inp);
2099 refcount_release(&tp->t_fb->tfb_refcnt);
2100 uma_zfree(V_tcpcb_zone, tm);
2106 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2107 if (tp->t_fb->tfb_tcp_fb_fini)
2108 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2109 in_pcbrele_wlocked(inp);
2110 refcount_release(&tp->t_fb->tfb_refcnt);
2111 uma_zfree(V_tcpcb_zone, tm);
2117 tp->t_vnet = inp->inp_vnet;
2119 tp->t_timers = &tm->tt;
2120 TAILQ_INIT(&tp->t_segq);
2123 isipv6 ? V_tcp_v6mssdflt :
2127 /* Set up our timeouts. */
2128 callout_init(&tp->t_timers->tt_rexmt, 1);
2129 callout_init(&tp->t_timers->tt_persist, 1);
2130 callout_init(&tp->t_timers->tt_keep, 1);
2131 callout_init(&tp->t_timers->tt_2msl, 1);
2132 callout_init(&tp->t_timers->tt_delack, 1);
2134 if (V_tcp_do_rfc1323)
2135 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2137 tp->t_flags |= TF_SACK_PERMIT;
2138 TAILQ_INIT(&tp->snd_holes);
2141 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2142 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
2143 * reasonable initial retransmit time.
2145 tp->t_srtt = TCPTV_SRTTBASE;
2146 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2147 tp->t_rttmin = tcp_rexmit_min;
2148 tp->t_rxtcur = tcp_rexmit_initial;
2149 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2150 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2151 tp->t_rcvtime = ticks;
2153 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2154 * because the socket may be bound to an IPv6 wildcard address,
2155 * which may match an IPv4-mapped IPv6 address.
2157 inp->inp_ip_ttl = V_ip_defttl;
2161 * Init the TCP PCAP queues.
2163 tcp_pcap_tcpcb_init(tp);
2166 /* Initialize the per-TCPCB log data. */
2167 tcp_log_tcpcbinit(tp);
2169 tp->t_pacing_rate = -1;
2170 if (tp->t_fb->tfb_tcp_fb_init) {
2171 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2172 refcount_release(&tp->t_fb->tfb_refcnt);
2173 in_pcbrele_wlocked(inp);
2174 uma_zfree(V_tcpcb_zone, tm);
2179 if (V_tcp_perconn_stats_enable == 1)
2180 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2183 tp->t_flags |= TF_LRD;
2184 return (tp); /* XXX */
2188 * Switch the congestion control algorithm back to NewReno for any active
2189 * control blocks using an algorithm which is about to go away.
2190 * This ensures the CC framework can allow the unload to proceed without leaving
2191 * any dangling pointers which would trigger a panic.
2192 * Returning non-zero would inform the CC framework that something went wrong
2193 * and it would be unsafe to allow the unload to proceed. However, there is no
2194 * way for this to occur with this implementation so we always return zero.
2197 tcp_ccalgounload(struct cc_algo *unload_algo)
2199 struct cc_algo *tmpalgo;
2202 VNET_ITERATOR_DECL(vnet_iter);
2205 * Check all active control blocks across all network stacks and change
2206 * any that are using "unload_algo" back to NewReno. If "unload_algo"
2207 * requires cleanup code to be run, call it.
2210 VNET_FOREACH(vnet_iter) {
2211 CURVNET_SET(vnet_iter);
2212 INP_INFO_WLOCK(&V_tcbinfo);
2214 * New connections already part way through being initialised
2215 * with the CC algo we're removing will not race with this code
2216 * because the INP_INFO_WLOCK is held during initialisation. We
2217 * therefore don't enter the loop below until the connection
2218 * list has stabilised.
2220 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
2222 /* Important to skip tcptw structs. */
2223 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2224 (tp = intotcpcb(inp)) != NULL) {
2226 * By holding INP_WLOCK here, we are assured
2227 * that the connection is not currently
2228 * executing inside the CC module's functions
2229 * i.e. it is safe to make the switch back to
2232 if (CC_ALGO(tp) == unload_algo) {
2233 tmpalgo = CC_ALGO(tp);
2234 if (tmpalgo->cb_destroy != NULL)
2235 tmpalgo->cb_destroy(tp->ccv);
2238 * NewReno may allocate memory on
2239 * demand for certain stateful
2240 * configuration as needed, but is
2241 * coded to never fail on memory
2242 * allocation failure so it is a safe
2245 CC_ALGO(tp) = &newreno_cc_algo;
2250 INP_INFO_WUNLOCK(&V_tcbinfo);
2253 VNET_LIST_RUNLOCK();
2259 * Drop a TCP connection, reporting
2260 * the specified error. If connection is synchronized,
2261 * then send a RST to peer.
2264 tcp_drop(struct tcpcb *tp, int errno)
2266 struct socket *so = tp->t_inpcb->inp_socket;
2269 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2270 INP_WLOCK_ASSERT(tp->t_inpcb);
2272 if (TCPS_HAVERCVDSYN(tp->t_state)) {
2273 tcp_state_change(tp, TCPS_CLOSED);
2274 (void) tp->t_fb->tfb_tcp_output(tp);
2275 TCPSTAT_INC(tcps_drops);
2277 TCPSTAT_INC(tcps_conndrops);
2278 if (errno == ETIMEDOUT && tp->t_softerror)
2279 errno = tp->t_softerror;
2280 so->so_error = errno;
2281 return (tcp_close(tp));
2285 tcp_discardcb(struct tcpcb *tp)
2287 struct inpcb *inp = tp->t_inpcb;
2288 struct socket *so = inp->inp_socket;
2290 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2292 int released __unused;
2294 INP_WLOCK_ASSERT(inp);
2297 * Make sure that all of our timers are stopped before we delete the
2300 * If stopping a timer fails, we schedule a discard function in same
2301 * callout, and the last discard function called will take care of
2302 * deleting the tcpcb.
2304 tp->t_timers->tt_draincnt = 0;
2305 tcp_timer_stop(tp, TT_REXMT);
2306 tcp_timer_stop(tp, TT_PERSIST);
2307 tcp_timer_stop(tp, TT_KEEP);
2308 tcp_timer_stop(tp, TT_2MSL);
2309 tcp_timer_stop(tp, TT_DELACK);
2310 if (tp->t_fb->tfb_tcp_timer_stop_all) {
2312 * Call the stop-all function of the methods,
2313 * this function should call the tcp_timer_stop()
2314 * method with each of the function specific timeouts.
2315 * That stop will be called via the tfb_tcp_timer_stop()
2316 * which should use the async drain function of the
2317 * callout system (see tcp_var.h).
2319 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2322 /* free the reassembly queue, if any */
2323 tcp_reass_flush(tp);
2326 /* Disconnect offload device, if any. */
2327 if (tp->t_flags & TF_TOE)
2328 tcp_offload_detach(tp);
2331 tcp_free_sackholes(tp);
2334 /* Free the TCP PCAP queues. */
2335 tcp_pcap_drain(&(tp->t_inpkts));
2336 tcp_pcap_drain(&(tp->t_outpkts));
2339 /* Allow the CC algorithm to clean up after itself. */
2340 if (CC_ALGO(tp)->cb_destroy != NULL)
2341 CC_ALGO(tp)->cb_destroy(tp->ccv);
2345 khelp_destroy_osd(tp->osd);
2348 stats_blob_destroy(tp->t_stats);
2352 inp->inp_ppcb = NULL;
2353 if (tp->t_timers->tt_draincnt == 0) {
2354 /* We own the last reference on tcpcb, let's free it. */
2356 tcp_log_tcpcbfini(tp);
2358 TCPSTATES_DEC(tp->t_state);
2359 if (tp->t_fb->tfb_tcp_fb_fini)
2360 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2363 * If we got enough samples through the srtt filter,
2364 * save the rtt and rttvar in the routing entry.
2365 * 'Enough' is arbitrarily defined as 4 rtt samples.
2366 * 4 samples is enough for the srtt filter to converge
2367 * to within enough % of the correct value; fewer samples
2368 * and we could save a bogus rtt. The danger is not high
2369 * as tcp quickly recovers from everything.
2370 * XXX: Works very well but needs some more statistics!
2372 * XXXRRS: Updating must be after the stack fini() since
2373 * that may be converting some internal representation of
2374 * say srtt etc into the general one used by other stacks.
2375 * Lets also at least protect against the so being NULL
2376 * as RW stated below.
2378 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2379 struct hc_metrics_lite metrics;
2382 bzero(&metrics, sizeof(metrics));
2384 * Update the ssthresh always when the conditions below
2385 * are satisfied. This gives us better new start value
2386 * for the congestion avoidance for new connections.
2387 * ssthresh is only set if packet loss occurred on a session.
2389 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2390 * being torn down. Ideally this code would not use 'so'.
2392 ssthresh = tp->snd_ssthresh;
2393 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2395 * convert the limit from user data bytes to
2396 * packets then to packet data bytes.
2398 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2401 ssthresh *= (tp->t_maxseg +
2403 (isipv6 ? sizeof (struct ip6_hdr) +
2404 sizeof (struct tcphdr) :
2406 sizeof (struct tcpiphdr)
2413 metrics.rmx_ssthresh = ssthresh;
2415 metrics.rmx_rtt = tp->t_srtt;
2416 metrics.rmx_rttvar = tp->t_rttvar;
2417 metrics.rmx_cwnd = tp->snd_cwnd;
2418 metrics.rmx_sendpipe = 0;
2419 metrics.rmx_recvpipe = 0;
2421 tcp_hc_update(&inp->inp_inc, &metrics);
2423 refcount_release(&tp->t_fb->tfb_refcnt);
2425 uma_zfree(V_tcpcb_zone, tp);
2426 released = in_pcbrele_wlocked(inp);
2427 KASSERT(!released, ("%s: inp %p should not have been released "
2428 "here", __func__, inp));
2433 tcp_timer_discard(void *ptp)
2437 struct epoch_tracker et;
2439 tp = (struct tcpcb *)ptp;
2440 CURVNET_SET(tp->t_vnet);
2441 NET_EPOCH_ENTER(et);
2443 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2446 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2447 ("%s: tcpcb has to be stopped here", __func__));
2448 tp->t_timers->tt_draincnt--;
2449 if (tp->t_timers->tt_draincnt == 0) {
2450 /* We own the last reference on this tcpcb, let's free it. */
2452 tcp_log_tcpcbfini(tp);
2454 TCPSTATES_DEC(tp->t_state);
2455 if (tp->t_fb->tfb_tcp_fb_fini)
2456 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2457 refcount_release(&tp->t_fb->tfb_refcnt);
2459 uma_zfree(V_tcpcb_zone, tp);
2460 if (in_pcbrele_wlocked(inp)) {
2472 * Attempt to close a TCP control block, marking it as dropped, and freeing
2473 * the socket if we hold the only reference.
2476 tcp_close(struct tcpcb *tp)
2478 struct inpcb *inp = tp->t_inpcb;
2481 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2482 INP_WLOCK_ASSERT(inp);
2485 if (tp->t_state == TCPS_LISTEN)
2486 tcp_offload_listen_stop(tp);
2489 * This releases the TFO pending counter resource for TFO listen
2490 * sockets as well as passively-created TFO sockets that transition
2491 * from SYN_RECEIVED to CLOSED.
2493 if (tp->t_tfo_pending) {
2494 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2495 tp->t_tfo_pending = NULL;
2498 TCPSTAT_INC(tcps_closed);
2499 if (tp->t_state != TCPS_CLOSED)
2500 tcp_state_change(tp, TCPS_CLOSED);
2501 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2502 so = inp->inp_socket;
2503 soisdisconnected(so);
2504 if (inp->inp_flags & INP_SOCKREF) {
2505 KASSERT(so->so_state & SS_PROTOREF,
2506 ("tcp_close: !SS_PROTOREF"));
2507 inp->inp_flags &= ~INP_SOCKREF;
2510 so->so_state &= ~SS_PROTOREF;
2520 VNET_ITERATOR_DECL(vnet_iter);
2525 VNET_LIST_RLOCK_NOSLEEP();
2526 VNET_FOREACH(vnet_iter) {
2527 CURVNET_SET(vnet_iter);
2532 * Walk the tcpbs, if existing, and flush the reassembly queue,
2533 * if there is one...
2534 * XXX: The "Net/3" implementation doesn't imply that the TCP
2535 * reassembly queue should be flushed, but in a situation
2536 * where we're really low on mbufs, this is potentially
2539 INP_INFO_WLOCK(&V_tcbinfo);
2540 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2542 if (inpb->inp_flags & INP_TIMEWAIT) {
2546 if ((tcpb = intotcpcb(inpb)) != NULL) {
2547 tcp_reass_flush(tcpb);
2548 tcp_clean_sackreport(tcpb);
2550 tcp_log_drain(tcpb);
2553 if (tcp_pcap_aggressive_free) {
2554 /* Free the TCP PCAP queues. */
2555 tcp_pcap_drain(&(tcpb->t_inpkts));
2556 tcp_pcap_drain(&(tcpb->t_outpkts));
2562 INP_INFO_WUNLOCK(&V_tcbinfo);
2565 VNET_LIST_RUNLOCK_NOSLEEP();
2569 * Notify a tcp user of an asynchronous error;
2570 * store error as soft error, but wake up user
2571 * (for now, won't do anything until can select for soft error).
2573 * Do not wake up user since there currently is no mechanism for
2574 * reporting soft errors (yet - a kqueue filter may be added).
2576 static struct inpcb *
2577 tcp_notify(struct inpcb *inp, int error)
2581 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2582 INP_WLOCK_ASSERT(inp);
2584 if ((inp->inp_flags & INP_TIMEWAIT) ||
2585 (inp->inp_flags & INP_DROPPED))
2588 tp = intotcpcb(inp);
2589 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2592 * Ignore some errors if we are hooked up.
2593 * If connection hasn't completed, has retransmitted several times,
2594 * and receives a second error, give up now. This is better
2595 * than waiting a long time to establish a connection that
2596 * can never complete.
2598 if (tp->t_state == TCPS_ESTABLISHED &&
2599 (error == EHOSTUNREACH || error == ENETUNREACH ||
2600 error == EHOSTDOWN)) {
2601 if (inp->inp_route.ro_nh) {
2602 NH_FREE(inp->inp_route.ro_nh);
2603 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2606 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2608 tp = tcp_drop(tp, error);
2614 tp->t_softerror = error;
2618 wakeup( &so->so_timeo);
2625 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2627 struct epoch_tracker et;
2632 if (req->newptr != NULL)
2635 if (req->oldptr == NULL) {
2638 n = V_tcbinfo.ipi_count +
2639 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2640 n += imax(n / 8, 10);
2641 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2645 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2648 bzero(&xig, sizeof(xig));
2649 xig.xig_len = sizeof xig;
2650 xig.xig_count = V_tcbinfo.ipi_count +
2651 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2652 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2653 xig.xig_sogen = so_gencnt;
2654 error = SYSCTL_OUT(req, &xig, sizeof xig);
2658 error = syncache_pcblist(req);
2662 NET_EPOCH_ENTER(et);
2663 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2665 inp = CK_LIST_NEXT(inp, inp_list)) {
2667 if (inp->inp_gencnt <= xig.xig_gen) {
2671 * XXX: This use of cr_cansee(), introduced with
2672 * TCP state changes, is not quite right, but for
2673 * now, better than nothing.
2675 if (inp->inp_flags & INP_TIMEWAIT) {
2676 if (intotw(inp) != NULL)
2677 crerr = cr_cansee(req->td->td_ucred,
2678 intotw(inp)->tw_cred);
2680 crerr = EINVAL; /* Skip this inp. */
2682 crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2686 tcp_inptoxtp(inp, &xt);
2688 error = SYSCTL_OUT(req, &xt, sizeof xt);
2701 * Give the user an updated idea of our state.
2702 * If the generation differs from what we told
2703 * her before, she knows that something happened
2704 * while we were processing this request, and it
2705 * might be necessary to retry.
2707 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2708 xig.xig_sogen = so_gencnt;
2709 xig.xig_count = V_tcbinfo.ipi_count +
2710 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2711 error = SYSCTL_OUT(req, &xig, sizeof xig);
2717 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2718 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2719 NULL, 0, tcp_pcblist, "S,xtcpcb",
2720 "List of active TCP connections");
2724 tcp_getcred(SYSCTL_HANDLER_ARGS)
2727 struct sockaddr_in addrs[2];
2728 struct epoch_tracker et;
2732 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2735 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2738 NET_EPOCH_ENTER(et);
2739 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2740 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2743 if (inp->inp_socket == NULL)
2746 error = cr_canseeinpcb(req->td->td_ucred, inp);
2748 cru2x(inp->inp_cred, &xuc);
2753 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2757 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2758 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2759 0, 0, tcp_getcred, "S,xucred",
2760 "Get the xucred of a TCP connection");
2765 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2767 struct epoch_tracker et;
2769 struct sockaddr_in6 addrs[2];
2776 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2779 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2782 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2783 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2786 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2788 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2795 NET_EPOCH_ENTER(et);
2798 inp = in_pcblookup(&V_tcbinfo,
2799 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2801 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2802 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2805 inp = in6_pcblookup(&V_tcbinfo,
2806 &addrs[1].sin6_addr, addrs[1].sin6_port,
2807 &addrs[0].sin6_addr, addrs[0].sin6_port,
2808 INPLOOKUP_RLOCKPCB, NULL);
2811 if (inp->inp_socket == NULL)
2814 error = cr_canseeinpcb(req->td->td_ucred, inp);
2816 cru2x(inp->inp_cred, &xuc);
2821 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2825 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2826 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2827 0, 0, tcp6_getcred, "S,xucred",
2828 "Get the xucred of a TCP6 connection");
2832 /* Path MTU to try next when a fragmentation-needed message is received. */
2834 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2836 int mtu = ntohs(icp->icmp_nextmtu);
2838 /* If no alternative MTU was proposed, try the next smaller one. */
2840 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2841 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2842 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2848 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2850 struct ip *ip = vip;
2852 struct in_addr faddr;
2855 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2857 struct in_conninfo inc;
2858 tcp_seq icmp_tcp_seq;
2861 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2862 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2865 if (cmd == PRC_MSGSIZE)
2866 notify = tcp_mtudisc_notify;
2867 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2868 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2869 cmd == PRC_TIMXCEED_INTRANS) && ip)
2870 notify = tcp_drop_syn_sent;
2873 * Hostdead is ugly because it goes linearly through all PCBs.
2874 * XXX: We never get this from ICMP, otherwise it makes an
2875 * excellent DoS attack on machines with many connections.
2877 else if (cmd == PRC_HOSTDEAD)
2879 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2883 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2887 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2888 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2889 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2890 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2891 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2892 /* signal EHOSTDOWN, as it flushes the cached route */
2893 inp = (*notify)(inp, EHOSTDOWN);
2896 icmp_tcp_seq = th->th_seq;
2898 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2899 !(inp->inp_flags & INP_DROPPED) &&
2900 !(inp->inp_socket == NULL)) {
2901 tp = intotcpcb(inp);
2903 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2905 * MTU discovery for offloaded connections. Let
2906 * the TOE driver verify seq# and process it.
2908 mtu = tcp_next_pmtu(icp, ip);
2909 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2913 if (tp->t_port != port) {
2916 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2917 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2918 if (cmd == PRC_MSGSIZE) {
2920 * MTU discovery: we got a needfrag and
2921 * will potentially try a lower MTU.
2923 mtu = tcp_next_pmtu(icp, ip);
2926 * Only process the offered MTU if it
2927 * is smaller than the current one.
2929 if (mtu < tp->t_maxseg +
2930 sizeof(struct tcpiphdr)) {
2931 bzero(&inc, sizeof(inc));
2932 inc.inc_faddr = faddr;
2934 inp->inp_inc.inc_fibnum;
2935 tcp_hc_updatemtu(&inc, mtu);
2936 tcp_mtudisc(inp, mtu);
2939 inp = (*notify)(inp,
2940 inetctlerrmap[cmd]);
2944 bzero(&inc, sizeof(inc));
2945 inc.inc_fport = th->th_dport;
2946 inc.inc_lport = th->th_sport;
2947 inc.inc_faddr = faddr;
2948 inc.inc_laddr = ip->ip_src;
2949 syncache_unreach(&inc, icmp_tcp_seq, port);
2957 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2959 tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2963 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2965 /* Its a tunneled TCP over UDP icmp */
2966 struct ip *outer_ip, *inner_ip;
2969 struct tcphdr *th, ttemp;
2973 inner_ip = (struct ip *)vip;
2974 icmp = (struct icmp *)((caddr_t)inner_ip -
2975 (sizeof(struct icmp) - sizeof(struct ip)));
2976 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2977 i_hlen = inner_ip->ip_hl << 2;
2978 o_len = ntohs(outer_ip->ip_len);
2980 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2981 /* Not enough data present */
2984 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2985 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2986 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2989 port = udp->uh_dport;
2990 th = (struct tcphdr *)(udp + 1);
2991 memcpy(&ttemp, th, sizeof(struct tcphdr));
2992 memcpy(udp, &ttemp, sizeof(struct tcphdr));
2993 /* Now adjust down the size of the outer IP header */
2994 o_len -= sizeof(struct udphdr);
2995 outer_ip->ip_len = htons(o_len);
2996 /* Now call in to the normal handling code */
2997 tcp_ctlinput_with_port(cmd, sa, vip, port);
3003 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3005 int mtu = ntohl(icmp6->icmp6_mtu);
3008 * If no alternative MTU was proposed, or the proposed MTU was too
3009 * small, set to the min.
3011 if (mtu < IPV6_MMTU)
3012 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */
3017 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3019 struct in6_addr *dst;
3020 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3021 struct ip6_hdr *ip6;
3025 struct icmp6_hdr *icmp6;
3026 struct ip6ctlparam *ip6cp = NULL;
3027 const struct sockaddr_in6 *sa6_src = NULL;
3028 struct in_conninfo inc;
3033 tcp_seq icmp_tcp_seq;
3037 if (sa->sa_family != AF_INET6 ||
3038 sa->sa_len != sizeof(struct sockaddr_in6))
3041 /* if the parameter is from icmp6, decode it. */
3043 ip6cp = (struct ip6ctlparam *)d;
3044 icmp6 = ip6cp->ip6c_icmp6;
3046 ip6 = ip6cp->ip6c_ip6;
3047 off = ip6cp->ip6c_off;
3048 sa6_src = ip6cp->ip6c_src;
3049 dst = ip6cp->ip6c_finaldst;
3053 off = 0; /* fool gcc */
3058 if (cmd == PRC_MSGSIZE)
3059 notify = tcp_mtudisc_notify;
3060 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3061 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3062 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3063 notify = tcp_drop_syn_sent;
3066 * Hostdead is ugly because it goes linearly through all PCBs.
3067 * XXX: We never get this from ICMP, otherwise it makes an
3068 * excellent DoS attack on machines with many connections.
3070 else if (cmd == PRC_HOSTDEAD)
3072 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3076 in6_pcbnotify(&V_tcbinfo, sa, 0,
3077 (const struct sockaddr *)sa6_src,
3078 0, cmd, NULL, notify);
3082 /* Check if we can safely get the ports from the tcp hdr */
3085 (int32_t) (off + sizeof(struct tcp_ports)))) {
3088 bzero(&t_ports, sizeof(struct tcp_ports));
3089 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3090 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3091 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3092 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3093 /* signal EHOSTDOWN, as it flushes the cached route */
3094 inp = (*notify)(inp, EHOSTDOWN);
3097 off += sizeof(struct tcp_ports);
3098 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3101 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3103 if (!(inp->inp_flags & INP_TIMEWAIT) &&
3104 !(inp->inp_flags & INP_DROPPED) &&
3105 !(inp->inp_socket == NULL)) {
3106 tp = intotcpcb(inp);
3108 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3109 /* MTU discovery for offloaded connections. */
3110 mtu = tcp6_next_pmtu(icmp6);
3111 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3115 if (tp->t_port != port) {
3118 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3119 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3120 if (cmd == PRC_MSGSIZE) {
3123 * If we got a needfrag set the MTU
3124 * in the route to the suggested new
3125 * value (if given) and then notify.
3127 mtu = tcp6_next_pmtu(icmp6);
3129 bzero(&inc, sizeof(inc));
3130 inc.inc_fibnum = M_GETFIB(m);
3131 inc.inc_flags |= INC_ISIPV6;
3132 inc.inc6_faddr = *dst;
3133 if (in6_setscope(&inc.inc6_faddr,
3134 m->m_pkthdr.rcvif, NULL))
3137 * Only process the offered MTU if it
3138 * is smaller than the current one.
3140 if (mtu < tp->t_maxseg +
3141 sizeof (struct tcphdr) +
3142 sizeof (struct ip6_hdr)) {
3143 tcp_hc_updatemtu(&inc, mtu);
3144 tcp_mtudisc(inp, mtu);
3145 ICMP6STAT_INC(icp6s_pmtuchg);
3148 inp = (*notify)(inp,
3149 inet6ctlerrmap[cmd]);
3153 bzero(&inc, sizeof(inc));
3154 inc.inc_fibnum = M_GETFIB(m);
3155 inc.inc_flags |= INC_ISIPV6;
3156 inc.inc_fport = t_ports.th_dport;
3157 inc.inc_lport = t_ports.th_sport;
3158 inc.inc6_faddr = *dst;
3159 inc.inc6_laddr = ip6->ip6_src;
3160 syncache_unreach(&inc, icmp_tcp_seq, port);
3168 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3170 tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3174 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3176 struct ip6ctlparam *ip6cp;
3181 ip6cp = (struct ip6ctlparam *)d;
3182 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3186 udp = mtod(m, struct udphdr *);
3187 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3190 port = udp->uh_dport;
3191 m_adj(m, sizeof(struct udphdr));
3192 if ((m->m_flags & M_PKTHDR) == 0) {
3193 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3195 /* Now call in to the normal handling code */
3196 tcp6_ctlinput_with_port(cmd, sa, d, port);
3202 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3207 KASSERT(len >= SIPHASH_KEY_LENGTH,
3208 ("%s: keylen %u too short ", __func__, len));
3209 SipHash24_Init(&ctx);
3210 SipHash_SetKey(&ctx, (uint8_t *)key);
3211 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3212 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3213 switch (inc->inc_flags & INC_ISIPV6) {
3216 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3217 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3222 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3223 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3227 SipHash_Final((uint8_t *)hash, &ctx);
3229 return (hash[0] ^ hash[1]);
3233 tcp_new_ts_offset(struct in_conninfo *inc)
3235 struct in_conninfo inc_store, *local_inc;
3237 if (!V_tcp_ts_offset_per_conn) {
3238 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3239 inc_store.inc_lport = 0;
3240 inc_store.inc_fport = 0;
3241 local_inc = &inc_store;
3245 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3246 sizeof(V_ts_offset_secret)));
3250 * Following is where TCP initial sequence number generation occurs.
3252 * There are two places where we must use initial sequence numbers:
3253 * 1. In SYN-ACK packets.
3254 * 2. In SYN packets.
3256 * All ISNs for SYN-ACK packets are generated by the syncache. See
3257 * tcp_syncache.c for details.
3259 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3260 * depends on this property. In addition, these ISNs should be
3261 * unguessable so as to prevent connection hijacking. To satisfy
3262 * the requirements of this situation, the algorithm outlined in
3263 * RFC 1948 is used, with only small modifications.
3265 * Implementation details:
3267 * Time is based off the system timer, and is corrected so that it
3268 * increases by one megabyte per second. This allows for proper
3269 * recycling on high speed LANs while still leaving over an hour
3272 * As reading the *exact* system time is too expensive to be done
3273 * whenever setting up a TCP connection, we increment the time
3274 * offset in two ways. First, a small random positive increment
3275 * is added to isn_offset for each connection that is set up.
3276 * Second, the function tcp_isn_tick fires once per clock tick
3277 * and increments isn_offset as necessary so that sequence numbers
3278 * are incremented at approximately ISN_BYTES_PER_SECOND. The
3279 * random positive increments serve only to ensure that the same
3280 * exact sequence number is never sent out twice (as could otherwise
3281 * happen when a port is recycled in less than the system tick
3284 * net.inet.tcp.isn_reseed_interval controls the number of seconds
3285 * between seeding of isn_secret. This is normally set to zero,
3286 * as reseeding should not be necessary.
3288 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3289 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
3290 * general, this means holding an exclusive (write) lock.
3293 #define ISN_BYTES_PER_SECOND 1048576
3294 #define ISN_STATIC_INCREMENT 4096
3295 #define ISN_RANDOM_INCREMENT (4096 - 1)
3296 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
3298 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3299 VNET_DEFINE_STATIC(int, isn_last);
3300 VNET_DEFINE_STATIC(int, isn_last_reseed);
3301 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3302 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3304 #define V_isn_secret VNET(isn_secret)
3305 #define V_isn_last VNET(isn_last)
3306 #define V_isn_last_reseed VNET(isn_last_reseed)
3307 #define V_isn_offset VNET(isn_offset)
3308 #define V_isn_offset_old VNET(isn_offset_old)
3311 tcp_new_isn(struct in_conninfo *inc)
3314 u_int32_t projected_offset;
3317 /* Seed if this is the first use, reseed if requested. */
3318 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3319 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3321 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3322 V_isn_last_reseed = ticks;
3325 /* Compute the hash and return the ISN. */
3326 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3327 sizeof(V_isn_secret));
3328 V_isn_offset += ISN_STATIC_INCREMENT +
3329 (arc4random() & ISN_RANDOM_INCREMENT);
3330 if (ticks != V_isn_last) {
3331 projected_offset = V_isn_offset_old +
3332 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3333 if (SEQ_GT(projected_offset, V_isn_offset))
3334 V_isn_offset = projected_offset;
3335 V_isn_offset_old = V_isn_offset;
3338 new_isn += V_isn_offset;
3344 * When a specific ICMP unreachable message is received and the
3345 * connection state is SYN-SENT, drop the connection. This behavior
3346 * is controlled by the icmp_may_rst sysctl.
3349 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3354 INP_WLOCK_ASSERT(inp);
3356 if ((inp->inp_flags & INP_TIMEWAIT) ||
3357 (inp->inp_flags & INP_DROPPED))
3360 tp = intotcpcb(inp);
3361 if (tp->t_state != TCPS_SYN_SENT)
3364 if (IS_FASTOPEN(tp->t_flags))
3365 tcp_fastopen_disable_path(tp);
3367 tp = tcp_drop(tp, errno);
3375 * When `need fragmentation' ICMP is received, update our idea of the MSS
3376 * based on the new value. Also nudge TCP to send something, since we
3377 * know the packet we just sent was dropped.
3378 * This duplicates some code in the tcp_mss() function in tcp_input.c.
3380 static struct inpcb *
3381 tcp_mtudisc_notify(struct inpcb *inp, int error)
3384 tcp_mtudisc(inp, -1);
3389 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3394 INP_WLOCK_ASSERT(inp);
3395 if ((inp->inp_flags & INP_TIMEWAIT) ||
3396 (inp->inp_flags & INP_DROPPED))
3399 tp = intotcpcb(inp);
3400 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3402 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3404 so = inp->inp_socket;
3405 SOCKBUF_LOCK(&so->so_snd);
3406 /* If the mss is larger than the socket buffer, decrease the mss. */
3407 if (so->so_snd.sb_hiwat < tp->t_maxseg)
3408 tp->t_maxseg = so->so_snd.sb_hiwat;
3409 SOCKBUF_UNLOCK(&so->so_snd);
3411 TCPSTAT_INC(tcps_mturesent);
3413 tp->snd_nxt = tp->snd_una;
3414 tcp_free_sackholes(tp);
3415 tp->snd_recover = tp->snd_max;
3416 if (tp->t_flags & TF_SACK_PERMIT)
3417 EXIT_FASTRECOVERY(tp->t_flags);
3418 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3420 * Conceptually the snd_nxt setting
3421 * and freeing sack holes should
3422 * be done by the default stacks
3423 * own tfb_tcp_mtu_chg().
3425 tp->t_fb->tfb_tcp_mtu_chg(tp);
3427 tp->t_fb->tfb_tcp_output(tp);
3432 * Look-up the routing entry to the peer of this inpcb. If no route
3433 * is found and it cannot be allocated, then return 0. This routine
3434 * is called by TCP routines that access the rmx structure and by
3435 * tcp_mss_update to get the peer/interface MTU.
3438 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3440 struct nhop_object *nh;
3442 uint32_t maxmtu = 0;
3444 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3446 if (inc->inc_faddr.s_addr != INADDR_ANY) {
3447 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3452 maxmtu = nh->nh_mtu;
3454 /* Report additional interface capabilities. */
3456 if (ifp->if_capenable & IFCAP_TSO4 &&
3457 ifp->if_hwassist & CSUM_TSO) {
3458 cap->ifcap |= CSUM_TSO;
3459 cap->tsomax = ifp->if_hw_tsomax;
3460 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3461 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3471 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3473 struct nhop_object *nh;
3474 struct in6_addr dst6;
3477 uint32_t maxmtu = 0;
3479 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3481 if (inc->inc_flags & INC_IPV6MINMTU)
3484 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3485 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3486 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3491 maxmtu = nh->nh_mtu;
3493 /* Report additional interface capabilities. */
3495 if (ifp->if_capenable & IFCAP_TSO6 &&
3496 ifp->if_hwassist & CSUM_TSO) {
3497 cap->ifcap |= CSUM_TSO;
3498 cap->tsomax = ifp->if_hw_tsomax;
3499 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3500 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3510 * Calculate effective SMSS per RFC5681 definition for a given TCP
3511 * connection at its current state, taking into account SACK and etc.
3514 tcp_maxseg(const struct tcpcb *tp)
3518 if (tp->t_flags & TF_NOOPT)
3519 return (tp->t_maxseg);
3522 * Here we have a simplified code from tcp_addoptions(),
3523 * without a proper loop, and having most of paddings hardcoded.
3524 * We might make mistakes with padding here in some edge cases,
3525 * but this is harmless, since result of tcp_maxseg() is used
3526 * only in cwnd and ssthresh estimations.
3528 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3529 if (tp->t_flags & TF_RCVD_TSTMP)
3530 optlen = TCPOLEN_TSTAMP_APPA;
3533 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3534 if (tp->t_flags & TF_SIGNATURE)
3535 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3537 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3538 optlen += TCPOLEN_SACKHDR;
3539 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3540 optlen = PADTCPOLEN(optlen);
3543 if (tp->t_flags & TF_REQ_TSTMP)
3544 optlen = TCPOLEN_TSTAMP_APPA;
3546 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3547 if (tp->t_flags & TF_REQ_SCALE)
3548 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3549 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3550 if (tp->t_flags & TF_SIGNATURE)
3551 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3553 if (tp->t_flags & TF_SACK_PERMIT)
3554 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3557 optlen = min(optlen, TCP_MAXOLEN);
3558 return (tp->t_maxseg - optlen);
3563 tcp_fixed_maxseg(const struct tcpcb *tp)
3567 if (tp->t_flags & TF_NOOPT)
3568 return (tp->t_maxseg);
3571 * Here we have a simplified code from tcp_addoptions(),
3572 * without a proper loop, and having most of paddings hardcoded.
3573 * We only consider fixed options that we would send every
3574 * time I.e. SACK is not considered. This is important
3575 * for cc modules to figure out what the modulo of the
3578 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3579 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3580 if (tp->t_flags & TF_RCVD_TSTMP)
3581 optlen = TCPOLEN_TSTAMP_APPA;
3584 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3585 if (tp->t_flags & TF_SIGNATURE)
3586 optlen += PAD(TCPOLEN_SIGNATURE);
3589 if (tp->t_flags & TF_REQ_TSTMP)
3590 optlen = TCPOLEN_TSTAMP_APPA;
3592 optlen = PAD(TCPOLEN_MAXSEG);
3593 if (tp->t_flags & TF_REQ_SCALE)
3594 optlen += PAD(TCPOLEN_WINDOW);
3595 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3596 if (tp->t_flags & TF_SIGNATURE)
3597 optlen += PAD(TCPOLEN_SIGNATURE);
3599 if (tp->t_flags & TF_SACK_PERMIT)
3600 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3603 optlen = min(optlen, TCP_MAXOLEN);
3604 return (tp->t_maxseg - optlen);
3610 sysctl_drop(SYSCTL_HANDLER_ARGS)
3612 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3613 struct sockaddr_storage addrs[2];
3617 struct sockaddr_in *fin, *lin;
3618 struct epoch_tracker et;
3620 struct sockaddr_in6 *fin6, *lin6;
3631 if (req->oldptr != NULL || req->oldlen != 0)
3633 if (req->newptr == NULL)
3635 if (req->newlen < sizeof(addrs))
3637 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3641 switch (addrs[0].ss_family) {
3644 fin6 = (struct sockaddr_in6 *)&addrs[0];
3645 lin6 = (struct sockaddr_in6 *)&addrs[1];
3646 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3647 lin6->sin6_len != sizeof(struct sockaddr_in6))
3649 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3650 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3652 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3653 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3654 fin = (struct sockaddr_in *)&addrs[0];
3655 lin = (struct sockaddr_in *)&addrs[1];
3658 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3661 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3668 fin = (struct sockaddr_in *)&addrs[0];
3669 lin = (struct sockaddr_in *)&addrs[1];
3670 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3671 lin->sin_len != sizeof(struct sockaddr_in))
3678 NET_EPOCH_ENTER(et);
3679 switch (addrs[0].ss_family) {
3682 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3683 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3684 INPLOOKUP_WLOCKPCB, NULL);
3689 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3690 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3695 if (inp->inp_flags & INP_TIMEWAIT) {
3697 * XXXRW: There currently exists a state where an
3698 * inpcb is present, but its timewait state has been
3699 * discarded. For now, don't allow dropping of this
3707 } else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3708 !SOLISTENING(inp->inp_socket)) {
3709 tp = intotcpcb(inp);
3710 tp = tcp_drop(tp, ECONNABORTED);
3721 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3722 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3723 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3724 "Drop TCP connection");
3728 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3730 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3731 struct sockaddr_storage addrs[2];
3733 struct sockaddr_in *fin, *lin;
3734 struct epoch_tracker et;
3736 struct sockaddr_in6 *fin6, *lin6;
3747 if (req->oldptr != NULL || req->oldlen != 0)
3749 if (req->newptr == NULL)
3751 if (req->newlen < sizeof(addrs))
3753 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3757 switch (addrs[0].ss_family) {
3760 fin6 = (struct sockaddr_in6 *)&addrs[0];
3761 lin6 = (struct sockaddr_in6 *)&addrs[1];
3762 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3763 lin6->sin6_len != sizeof(struct sockaddr_in6))
3765 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3766 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3768 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3769 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3770 fin = (struct sockaddr_in *)&addrs[0];
3771 lin = (struct sockaddr_in *)&addrs[1];
3774 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3777 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3784 fin = (struct sockaddr_in *)&addrs[0];
3785 lin = (struct sockaddr_in *)&addrs[1];
3786 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3787 lin->sin_len != sizeof(struct sockaddr_in))
3794 NET_EPOCH_ENTER(et);
3795 switch (addrs[0].ss_family) {
3798 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3799 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3800 INPLOOKUP_WLOCKPCB, NULL);
3805 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3806 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3812 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3813 inp->inp_socket == NULL) {
3819 so = inp->inp_socket;
3821 error = ktls_set_tx_mode(so,
3822 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3832 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3833 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3834 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3835 "Switch TCP connection to SW TLS");
3836 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3837 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3838 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3839 "Switch TCP connection to ifnet TLS");
3843 * Generate a standardized TCP log line for use throughout the
3844 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3845 * allow use in the interrupt context.
3847 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3848 * NB: The function may return NULL if memory allocation failed.
3850 * Due to header inclusion and ordering limitations the struct ip
3851 * and ip6_hdr pointers have to be passed as void pointers.
3854 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3858 /* Is logging enabled? */
3859 if (V_tcp_log_in_vain == 0)
3862 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3866 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3870 /* Is logging enabled? */
3871 if (tcp_log_debug == 0)
3874 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3878 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3885 const struct ip6_hdr *ip6;
3887 ip6 = (const struct ip6_hdr *)ip6hdr;
3889 ip = (struct ip *)ip4hdr;
3892 * The log line looks like this:
3893 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3895 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3896 sizeof(PRINT_TH_FLAGS) + 1 +
3898 2 * INET6_ADDRSTRLEN;
3900 2 * INET_ADDRSTRLEN;
3903 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3907 strcat(s, "TCP: [");
3910 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3911 inet_ntoa_r(inc->inc_faddr, sp);
3913 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3915 inet_ntoa_r(inc->inc_laddr, sp);
3917 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3920 ip6_sprintf(sp, &inc->inc6_faddr);
3922 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3924 ip6_sprintf(sp, &inc->inc6_laddr);
3926 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3927 } else if (ip6 && th) {
3928 ip6_sprintf(sp, &ip6->ip6_src);
3930 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3932 ip6_sprintf(sp, &ip6->ip6_dst);
3934 sprintf(sp, "]:%i", ntohs(th->th_dport));
3937 } else if (ip && th) {
3938 inet_ntoa_r(ip->ip_src, sp);
3940 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3942 inet_ntoa_r(ip->ip_dst, sp);
3944 sprintf(sp, "]:%i", ntohs(th->th_dport));
3952 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3953 if (*(s + size - 1) != '\0')
3954 panic("%s: string too long", __func__);
3959 * A subroutine which makes it easy to track TCP state changes with DTrace.
3960 * This function shouldn't be called for t_state initializations that don't
3961 * correspond to actual TCP state transitions.
3964 tcp_state_change(struct tcpcb *tp, int newstate)
3966 #if defined(KDTRACE_HOOKS)
3967 int pstate = tp->t_state;
3970 TCPSTATES_DEC(tp->t_state);
3971 TCPSTATES_INC(newstate);
3972 tp->t_state = newstate;
3973 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3977 * Create an external-format (``xtcpcb'') structure using the information in
3978 * the kernel-format tcpcb structure pointed to by tp. This is done to
3979 * reduce the spew of irrelevant information over this interface, to isolate
3980 * user code from changes in the kernel structure, and potentially to provide
3981 * information-hiding if we decide that some of this information should be
3982 * hidden from users.
3985 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3987 struct tcpcb *tp = intotcpcb(inp);
3988 struct tcptw *tw = intotw(inp);
3991 bzero(xt, sizeof(*xt));
3992 if (inp->inp_flags & INP_TIMEWAIT) {
3993 xt->t_state = TCPS_TIME_WAIT;
3994 xt->xt_encaps_port = tw->t_port;
3996 xt->t_state = tp->t_state;
3997 xt->t_logstate = tp->t_logstate;
3998 xt->t_flags = tp->t_flags;
3999 xt->t_sndzerowin = tp->t_sndzerowin;
4000 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4001 xt->t_rcvoopack = tp->t_rcvoopack;
4002 xt->t_rcv_wnd = tp->rcv_wnd;
4003 xt->t_snd_wnd = tp->snd_wnd;
4004 xt->t_snd_cwnd = tp->snd_cwnd;
4005 xt->t_snd_ssthresh = tp->snd_ssthresh;
4006 xt->t_maxseg = tp->t_maxseg;
4007 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4008 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4010 now = getsbinuptime();
4011 #define COPYTIMER(ttt) do { \
4012 if (callout_active(&tp->t_timers->ttt)) \
4013 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
4018 COPYTIMER(tt_delack);
4019 COPYTIMER(tt_rexmt);
4020 COPYTIMER(tt_persist);
4024 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4026 xt->xt_encaps_port = tp->t_port;
4027 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4028 TCP_FUNCTION_NAME_LEN_MAX);
4029 bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4032 (void)tcp_log_get_id(tp, xt->xt_logid);
4036 xt->xt_len = sizeof(struct xtcpcb);
4037 in_pcbtoxinpcb(inp, &xt->xt_inp);
4038 if (inp->inp_socket == NULL)
4039 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4043 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4048 (status > TCP_EI_STATUS_MAX_VALUE) ||
4053 if (status > (sizeof(uint32_t) * 8)) {
4054 /* Should this be a KASSERT? */
4057 bit = 1U << (status - 1);
4058 if (bit & tp->t_end_info_status) {
4059 /* already logged */
4062 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4063 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4064 tp->t_end_info_bytes[i] = status;
4065 tp->t_end_info_status |= bit;
4072 tcp_can_enable_pacing(void)
4075 if ((tcp_pacing_limit == -1) ||
4076 (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4077 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4078 shadow_num_connections = number_of_tcp_connections_pacing;
4085 static uint8_t tcp_pacing_warning = 0;
4088 tcp_decrement_paced_conn(void)
4092 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4093 shadow_num_connections = number_of_tcp_connections_pacing;
4094 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4096 if (tcp_pacing_limit != -1) {
4097 printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4098 tcp_pacing_limit = 0;
4099 } else if (tcp_pacing_warning == 0) {
4100 printf("Warning pacing count is invalid, invalid decrement\n");
4101 tcp_pacing_warning = 1;