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>
80 #include <net/if_var.h>
83 #include <netinet/in.h>
84 #include <netinet/in_fib.h>
85 #include <netinet/in_kdtrace.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/ip_icmp.h>
91 #include <netinet/ip_var.h>
93 #include <netinet/icmp6.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/in6_fib.h>
96 #include <netinet6/in6_pcb.h>
97 #include <netinet6/ip6_var.h>
98 #include <netinet6/scope6_var.h>
99 #include <netinet6/nd6.h>
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcp_log_buf.h>
108 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_hpts.h>
110 #include <netinet/cc/cc.h>
112 #include <netinet6/tcp6_var.h>
114 #include <netinet/tcpip.h>
115 #include <netinet/tcp_fastopen.h>
117 #include <netinet/tcp_pcap.h>
120 #include <netinet/tcp_debug.h>
123 #include <netinet6/ip6protosw.h>
126 #include <netinet/tcp_offload.h>
129 #include <netipsec/ipsec_support.h>
131 #include <machine/in_cksum.h>
132 #include <crypto/siphash/siphash.h>
134 #include <security/mac/mac_framework.h>
136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
141 #ifdef NETFLIX_EXP_DETECTION
142 /* Sack attack detection thresholds and such */
143 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, CTLFLAG_RW, 0,
144 "Sack Attack detection thresholds");
145 int32_t tcp_force_detection = 0;
146 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
148 &tcp_force_detection, 0,
149 "Do we force detection even if the INP has it off?");
150 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
151 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
153 &tcp_sack_to_ack_thresh, 700,
154 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
155 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
156 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
158 &tcp_sack_to_move_thresh, 600,
159 "Percentage of sack moves we must see above (10.1 percent is 101)");
160 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
161 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
163 &tcp_restoral_thresh, 550,
164 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
165 int32_t tcp_sad_decay_val = 800;
166 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
168 &tcp_sad_decay_val, 800,
169 "The decay percentage (10.1 percent equals 101 )");
170 int32_t tcp_map_minimum = 500;
171 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
173 &tcp_map_minimum, 500,
174 "Number of Map enteries before we start detection");
175 int32_t tcp_attack_on_turns_on_logging = 0;
176 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
178 &tcp_attack_on_turns_on_logging, 0,
179 "When we have a positive hit on attack, do we turn on logging?");
180 int32_t tcp_sad_pacing_interval = 2000;
181 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
183 &tcp_sad_pacing_interval, 2000,
184 "What is the minimum pacing interval for a classified attacker?");
186 int32_t tcp_sad_low_pps = 100;
187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
189 &tcp_sad_low_pps, 100,
190 "What is the input pps that below which we do not decay?");
193 struct rwlock tcp_function_lock;
196 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
201 error = sysctl_handle_int(oidp, &new, 0, req);
202 if (error == 0 && req->newptr) {
203 if (new < TCP_MINMSS)
211 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
212 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
213 &sysctl_net_inet_tcp_mss_check, "I",
214 "Default TCP Maximum Segment Size");
218 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
222 new = V_tcp_v6mssdflt;
223 error = sysctl_handle_int(oidp, &new, 0, req);
224 if (error == 0 && req->newptr) {
225 if (new < TCP_MINMSS)
228 V_tcp_v6mssdflt = new;
233 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
234 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
235 &sysctl_net_inet_tcp_mss_v6_check, "I",
236 "Default TCP Maximum Segment Size for IPv6");
240 * Minimum MSS we accept and use. This prevents DoS attacks where
241 * we are forced to a ridiculous low MSS like 20 and send hundreds
242 * of packets instead of one. The effect scales with the available
243 * bandwidth and quickly saturates the CPU and network interface
244 * with packet generation and sending. Set to zero to disable MINMSS
245 * checking. This setting prevents us from sending too small packets.
247 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
248 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
249 &VNET_NAME(tcp_minmss), 0,
250 "Minimum TCP Maximum Segment Size");
252 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
253 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
254 &VNET_NAME(tcp_do_rfc1323), 0,
255 "Enable rfc1323 (high performance TCP) extensions");
257 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
258 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
259 &VNET_NAME(tcp_ts_offset_per_conn), 0,
260 "Initialize TCP timestamps per connection instead of per host pair");
262 static int tcp_log_debug = 0;
263 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
264 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
266 static int tcp_tcbhashsize;
267 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
268 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
270 static int do_tcpdrain = 1;
271 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
272 "Enable tcp_drain routine for extra help when low on mbufs");
274 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
275 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
277 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
278 #define V_icmp_may_rst VNET(icmp_may_rst)
279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
280 &VNET_NAME(icmp_may_rst), 0,
281 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
283 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
284 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
285 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
286 &VNET_NAME(tcp_isn_reseed_interval), 0,
287 "Seconds between reseeding of ISN secret");
289 static int tcp_soreceive_stream;
290 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
291 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
293 VNET_DEFINE(uma_zone_t, sack_hole_zone);
294 #define V_sack_hole_zone VNET(sack_hole_zone)
295 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
297 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
302 new = V_tcp_map_entries_limit;
303 error = sysctl_handle_int(oidp, &new, 0, req);
304 if (error == 0 && req->newptr) {
305 /* only allow "0" and value > minimum */
306 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
309 V_tcp_map_entries_limit = new;
313 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
314 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
315 &VNET_NAME(tcp_map_entries_limit), 0,
316 &sysctl_net_inet_tcp_map_limit_check, "IU",
317 "Total sendmap entries limit");
319 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
320 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
321 &VNET_NAME(tcp_map_split_limit), 0,
322 "Total sendmap split entries limit");
325 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
328 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
329 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
330 #define V_ts_offset_secret VNET(ts_offset_secret)
332 static int tcp_default_fb_init(struct tcpcb *tp);
333 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
334 static int tcp_default_handoff_ok(struct tcpcb *tp);
335 static struct inpcb *tcp_notify(struct inpcb *, int);
336 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
337 static void tcp_mtudisc(struct inpcb *, int);
338 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
339 void *ip4hdr, const void *ip6hdr);
342 static struct tcp_function_block tcp_def_funcblk = {
343 .tfb_tcp_block_name = "freebsd",
344 .tfb_tcp_output = tcp_output,
345 .tfb_tcp_do_segment = tcp_do_segment,
346 .tfb_tcp_ctloutput = tcp_default_ctloutput,
347 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
348 .tfb_tcp_fb_init = tcp_default_fb_init,
349 .tfb_tcp_fb_fini = tcp_default_fb_fini,
352 static int tcp_fb_cnt = 0;
353 struct tcp_funchead t_functions;
354 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
356 static struct tcp_function_block *
357 find_tcp_functions_locked(struct tcp_function_set *fs)
359 struct tcp_function *f;
360 struct tcp_function_block *blk=NULL;
362 TAILQ_FOREACH(f, &t_functions, tf_next) {
363 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
371 static struct tcp_function_block *
372 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
374 struct tcp_function_block *rblk=NULL;
375 struct tcp_function *f;
377 TAILQ_FOREACH(f, &t_functions, tf_next) {
378 if (f->tf_fb == blk) {
389 struct tcp_function_block *
390 find_and_ref_tcp_functions(struct tcp_function_set *fs)
392 struct tcp_function_block *blk;
394 rw_rlock(&tcp_function_lock);
395 blk = find_tcp_functions_locked(fs);
397 refcount_acquire(&blk->tfb_refcnt);
398 rw_runlock(&tcp_function_lock);
402 struct tcp_function_block *
403 find_and_ref_tcp_fb(struct tcp_function_block *blk)
405 struct tcp_function_block *rblk;
407 rw_rlock(&tcp_function_lock);
408 rblk = find_tcp_fb_locked(blk, NULL);
410 refcount_acquire(&rblk->tfb_refcnt);
411 rw_runlock(&tcp_function_lock);
415 static struct tcp_function_block *
416 find_and_ref_tcp_default_fb(void)
418 struct tcp_function_block *rblk;
420 rw_rlock(&tcp_function_lock);
421 rblk = tcp_func_set_ptr;
422 refcount_acquire(&rblk->tfb_refcnt);
423 rw_runlock(&tcp_function_lock);
428 tcp_switch_back_to_default(struct tcpcb *tp)
430 struct tcp_function_block *tfb;
432 KASSERT(tp->t_fb != &tcp_def_funcblk,
433 ("%s: called by the built-in default stack", __func__));
436 * Release the old stack. This function will either find a new one
439 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
440 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
441 refcount_release(&tp->t_fb->tfb_refcnt);
444 * Now, we'll find a new function block to use.
445 * Start by trying the current user-selected
446 * default, unless this stack is the user-selected
449 tfb = find_and_ref_tcp_default_fb();
450 if (tfb == tp->t_fb) {
451 refcount_release(&tfb->tfb_refcnt);
454 /* Does the stack accept this connection? */
455 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
456 (*tfb->tfb_tcp_handoff_ok)(tp)) {
457 refcount_release(&tfb->tfb_refcnt);
460 /* Try to use that stack. */
462 /* Initialize the new stack. If it succeeds, we are done. */
464 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
465 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
469 * Initialization failed. Release the reference count on
472 refcount_release(&tfb->tfb_refcnt);
476 * If that wasn't feasible, use the built-in default
477 * stack which is not allowed to reject anyone.
479 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
481 /* there always should be a default */
482 panic("Can't refer to tcp_def_funcblk");
484 if (tfb->tfb_tcp_handoff_ok != NULL) {
485 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
486 /* The default stack cannot say no */
487 panic("Default stack rejects a new session?");
491 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
492 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
493 /* The default stack cannot fail */
494 panic("Default stack initialization failed");
499 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
502 struct tcp_function_set fs;
503 struct tcp_function_block *blk;
505 memset(&fs, 0, sizeof(fs));
506 rw_rlock(&tcp_function_lock);
507 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
510 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
511 fs.pcbcnt = blk->tfb_refcnt;
513 rw_runlock(&tcp_function_lock);
514 error = sysctl_handle_string(oidp, fs.function_set_name,
515 sizeof(fs.function_set_name), req);
517 /* Check for error or no change */
518 if (error != 0 || req->newptr == NULL)
521 rw_wlock(&tcp_function_lock);
522 blk = find_tcp_functions_locked(&fs);
524 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
528 tcp_func_set_ptr = blk;
530 rw_wunlock(&tcp_function_lock);
534 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
535 CTLTYPE_STRING | CTLFLAG_RW,
536 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
537 "Set/get the default TCP functions");
540 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
542 int error, cnt, linesz;
543 struct tcp_function *f;
549 rw_rlock(&tcp_function_lock);
550 TAILQ_FOREACH(f, &t_functions, tf_next) {
553 rw_runlock(&tcp_function_lock);
555 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
556 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
561 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
562 "Alias", "PCB count");
567 rw_rlock(&tcp_function_lock);
568 TAILQ_FOREACH(f, &t_functions, tf_next) {
569 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
570 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
571 f->tf_fb->tfb_tcp_block_name,
572 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
573 alias ? f->tf_name : "-",
574 f->tf_fb->tfb_refcnt);
575 if (linesz >= bufsz) {
583 rw_runlock(&tcp_function_lock);
585 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
586 free(buffer, M_TEMP);
590 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
591 CTLTYPE_STRING|CTLFLAG_RD,
592 NULL, 0, sysctl_net_inet_list_available, "A",
593 "list available TCP Function sets");
596 * Exports one (struct tcp_function_info) for each alias/name.
599 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
602 struct tcp_function *f;
603 struct tcp_function_info tfi;
606 * We don't allow writes.
608 if (req->newptr != NULL)
612 * Wire the old buffer so we can directly copy the functions to
613 * user space without dropping the lock.
615 if (req->oldptr != NULL) {
616 error = sysctl_wire_old_buffer(req, 0);
622 * Walk the list and copy out matching entries. If INVARIANTS
623 * is compiled in, also walk the list to verify the length of
624 * the list matches what we have recorded.
626 rw_rlock(&tcp_function_lock);
630 if (req->oldptr == NULL) {
635 TAILQ_FOREACH(f, &t_functions, tf_next) {
639 if (req->oldptr != NULL) {
640 bzero(&tfi, sizeof(tfi));
641 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
642 tfi.tfi_id = f->tf_fb->tfb_id;
643 (void)strlcpy(tfi.tfi_alias, f->tf_name,
644 sizeof(tfi.tfi_alias));
645 (void)strlcpy(tfi.tfi_name,
646 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
647 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
649 * Don't stop on error, as that is the
650 * mechanism we use to accumulate length
651 * information if the buffer was too short.
655 KASSERT(cnt == tcp_fb_cnt,
656 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
660 rw_runlock(&tcp_function_lock);
661 if (req->oldptr == NULL)
662 error = SYSCTL_OUT(req, NULL,
663 (cnt + 1) * sizeof(struct tcp_function_info));
668 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
669 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
670 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
671 "List TCP function block name-to-ID mappings");
674 * tfb_tcp_handoff_ok() function for the default stack.
675 * Note that we'll basically try to take all comers.
678 tcp_default_handoff_ok(struct tcpcb *tp)
685 * tfb_tcp_fb_init() function for the default stack.
687 * This handles making sure we have appropriate timers set if you are
688 * transitioning a socket that has some amount of setup done.
690 * The init() fuction from the default can *never* return non-zero i.e.
691 * it is required to always succeed since it is the stack of last resort!
694 tcp_default_fb_init(struct tcpcb *tp)
699 INP_WLOCK_ASSERT(tp->t_inpcb);
701 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
702 ("%s: connection %p in unexpected state %d", __func__, tp,
706 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
707 * know what to do for unexpected states (which includes TIME_WAIT).
709 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
713 * Make sure some kind of transmission timer is set if there is
716 so = tp->t_inpcb->inp_socket;
717 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
718 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
719 tcp_timer_active(tp, TT_PERSIST))) {
721 * If the session has established and it looks like it should
722 * be in the persist state, set the persist timer. Otherwise,
723 * set the retransmit timer.
725 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
726 (int32_t)(tp->snd_nxt - tp->snd_una) <
727 (int32_t)sbavail(&so->so_snd))
730 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
733 /* All non-embryonic sessions get a keepalive timer. */
734 if (!tcp_timer_active(tp, TT_KEEP))
735 tcp_timer_activate(tp, TT_KEEP,
736 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
743 * tfb_tcp_fb_fini() function for the default stack.
745 * This changes state as necessary (or prudent) to prepare for another stack
746 * to assume responsibility for the connection.
749 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
752 INP_WLOCK_ASSERT(tp->t_inpcb);
757 * Target size of TCP PCB hash tables. Must be a power of two.
759 * Note that this can be overridden by the kernel environment
760 * variable net.inet.tcp.tcbhashsize
763 #define TCBHASHSIZE 0
768 * Callouts should be moved into struct tcp directly. They are currently
769 * separate because the tcpcb structure is exported to userland for sysctl
770 * parsing purposes, which do not know about callouts.
781 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
782 #define V_tcpcb_zone VNET(tcpcb_zone)
784 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
785 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
787 static struct mtx isn_mtx;
789 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
790 #define ISN_LOCK() mtx_lock(&isn_mtx)
791 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
794 * TCP initialization.
797 tcp_zone_change(void *tag)
800 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
801 uma_zone_set_max(V_tcpcb_zone, maxsockets);
802 tcp_tw_zone_change();
806 tcp_inpcb_init(void *mem, int size, int flags)
808 struct inpcb *inp = mem;
810 INP_LOCK_INIT(inp, "inp", "tcpinp");
815 * Take a value and get the next power of 2 that doesn't overflow.
816 * Used to size the tcp_inpcb hash buckets.
819 maketcp_hashsize(int size)
825 * get the next power of 2 higher than maxsockets.
827 hashsize = 1 << fls(size);
828 /* catch overflow, and just go one power of 2 smaller */
829 if (hashsize < size) {
830 hashsize = 1 << (fls(size) - 1);
835 static volatile int next_tcp_stack_id = 1;
838 * Register a TCP function block with the name provided in the names
839 * array. (Note that this function does NOT automatically register
840 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
841 * explicitly include blk->tfb_tcp_block_name in the list of names if
842 * you wish to register the stack with that name.)
844 * Either all name registrations will succeed or all will fail. If
845 * a name registration fails, the function will update the num_names
846 * argument to point to the array index of the name that encountered
849 * Returns 0 on success, or an error code on failure.
852 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
853 const char *names[], int *num_names)
855 struct tcp_function *n;
856 struct tcp_function_set fs;
859 KASSERT(names != NULL && *num_names > 0,
860 ("%s: Called with 0-length name list", __func__));
861 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
862 KASSERT(rw_initialized(&tcp_function_lock),
863 ("%s: called too early", __func__));
865 if ((blk->tfb_tcp_output == NULL) ||
866 (blk->tfb_tcp_do_segment == NULL) ||
867 (blk->tfb_tcp_ctloutput == NULL) ||
868 (strlen(blk->tfb_tcp_block_name) == 0)) {
870 * These functions are required and you
876 if (blk->tfb_tcp_timer_stop_all ||
877 blk->tfb_tcp_timer_activate ||
878 blk->tfb_tcp_timer_active ||
879 blk->tfb_tcp_timer_stop) {
881 * If you define one timer function you
882 * must have them all.
884 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
885 (blk->tfb_tcp_timer_activate == NULL) ||
886 (blk->tfb_tcp_timer_active == NULL) ||
887 (blk->tfb_tcp_timer_stop == NULL)) {
893 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
898 refcount_init(&blk->tfb_refcnt, 0);
899 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
900 for (i = 0; i < *num_names; i++) {
901 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
908 (void)strlcpy(fs.function_set_name, names[i],
909 sizeof(fs.function_set_name));
910 rw_wlock(&tcp_function_lock);
911 if (find_tcp_functions_locked(&fs) != NULL) {
912 /* Duplicate name space not allowed */
913 rw_wunlock(&tcp_function_lock);
914 free(n, M_TCPFUNCTIONS);
918 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
919 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
921 rw_wunlock(&tcp_function_lock);
927 * Deregister the names we just added. Because registration failed
928 * for names[i], we don't need to deregister that name.
931 rw_wlock(&tcp_function_lock);
933 TAILQ_FOREACH(n, &t_functions, tf_next) {
934 if (!strncmp(n->tf_name, names[i],
935 TCP_FUNCTION_NAME_LEN_MAX)) {
936 TAILQ_REMOVE(&t_functions, n, tf_next);
939 free(n, M_TCPFUNCTIONS);
944 rw_wunlock(&tcp_function_lock);
949 * Register a TCP function block using the name provided in the name
952 * Returns 0 on success, or an error code on failure.
955 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
958 const char *name_list[1];
965 name_list[0] = blk->tfb_tcp_block_name;
966 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
971 * Register a TCP function block using the name defined in
972 * blk->tfb_tcp_block_name.
974 * Returns 0 on success, or an error code on failure.
977 register_tcp_functions(struct tcp_function_block *blk, int wait)
980 return (register_tcp_functions_as_name(blk, NULL, wait));
984 * Deregister all names associated with a function block. This
985 * functionally removes the function block from use within the system.
987 * When called with a true quiesce argument, mark the function block
988 * as being removed so no more stacks will use it and determine
989 * whether the removal would succeed.
991 * When called with a false quiesce argument, actually attempt the
994 * When called with a force argument, attempt to switch all TCBs to
995 * use the default stack instead of returning EBUSY.
997 * Returns 0 on success (or if the removal would succeed, or an error
1001 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1004 struct tcp_function *f;
1006 if (blk == &tcp_def_funcblk) {
1007 /* You can't un-register the default */
1010 rw_wlock(&tcp_function_lock);
1011 if (blk == tcp_func_set_ptr) {
1012 /* You can't free the current default */
1013 rw_wunlock(&tcp_function_lock);
1016 /* Mark the block so no more stacks can use it. */
1017 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1019 * If TCBs are still attached to the stack, attempt to switch them
1020 * to the default stack.
1022 if (force && blk->tfb_refcnt) {
1025 VNET_ITERATOR_DECL(vnet_iter);
1027 rw_wunlock(&tcp_function_lock);
1030 VNET_FOREACH(vnet_iter) {
1031 CURVNET_SET(vnet_iter);
1032 INP_INFO_WLOCK(&V_tcbinfo);
1033 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1035 if (inp->inp_flags & INP_TIMEWAIT) {
1039 tp = intotcpcb(inp);
1040 if (tp == NULL || tp->t_fb != blk) {
1044 tcp_switch_back_to_default(tp);
1047 INP_INFO_WUNLOCK(&V_tcbinfo);
1050 VNET_LIST_RUNLOCK();
1052 rw_wlock(&tcp_function_lock);
1054 if (blk->tfb_refcnt) {
1055 /* TCBs still attached. */
1056 rw_wunlock(&tcp_function_lock);
1061 rw_wunlock(&tcp_function_lock);
1064 /* Remove any function names that map to this function block. */
1065 while (find_tcp_fb_locked(blk, &f) != NULL) {
1066 TAILQ_REMOVE(&t_functions, f, tf_next);
1069 free(f, M_TCPFUNCTIONS);
1071 rw_wunlock(&tcp_function_lock);
1078 const char *tcbhash_tuneable;
1081 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1084 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1085 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1086 printf("%s: WARNING: unable to register helper hook\n", __func__);
1087 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1088 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1089 printf("%s: WARNING: unable to register helper hook\n", __func__);
1092 if (tcp_stats_init())
1093 printf("%s: WARNING: unable to initialise TCP stats\n",
1096 hashsize = TCBHASHSIZE;
1097 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1098 if (hashsize == 0) {
1100 * Auto tune the hash size based on maxsockets.
1101 * A perfect hash would have a 1:1 mapping
1102 * (hashsize = maxsockets) however it's been
1103 * suggested that O(2) average is better.
1105 hashsize = maketcp_hashsize(maxsockets / 4);
1107 * Our historical default is 512,
1108 * do not autotune lower than this.
1112 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1113 printf("%s: %s auto tuned to %d\n", __func__,
1114 tcbhash_tuneable, hashsize);
1117 * We require a hashsize to be a power of two.
1118 * Previously if it was not a power of two we would just reset it
1119 * back to 512, which could be a nasty surprise if you did not notice
1120 * the error message.
1121 * Instead what we do is clip it to the closest power of two lower
1122 * than the specified hash value.
1124 if (!powerof2(hashsize)) {
1125 int oldhashsize = hashsize;
1127 hashsize = maketcp_hashsize(hashsize);
1128 /* prevent absurdly low value */
1131 printf("%s: WARNING: TCB hash size not a power of 2, "
1132 "clipped from %d to %d.\n", __func__, oldhashsize,
1135 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1136 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1139 * These have to be type stable for the benefit of the timers.
1141 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1142 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1143 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1144 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1150 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1151 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1152 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1154 tcp_fastopen_init();
1156 /* Skip initialization of globals for non-default instances. */
1157 if (!IS_DEFAULT_VNET(curvnet))
1160 tcp_reass_global_init();
1162 /* XXX virtualize those bellow? */
1163 tcp_delacktime = TCPTV_DELACK;
1164 tcp_keepinit = TCPTV_KEEP_INIT;
1165 tcp_keepidle = TCPTV_KEEP_IDLE;
1166 tcp_keepintvl = TCPTV_KEEPINTVL;
1167 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1168 tcp_msl = TCPTV_MSL;
1169 tcp_rexmit_initial = TCPTV_RTOBASE;
1170 if (tcp_rexmit_initial < 1)
1171 tcp_rexmit_initial = 1;
1172 tcp_rexmit_min = TCPTV_MIN;
1173 if (tcp_rexmit_min < 1)
1175 tcp_persmin = TCPTV_PERSMIN;
1176 tcp_persmax = TCPTV_PERSMAX;
1177 tcp_rexmit_slop = TCPTV_CPU_VAR;
1178 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1179 tcp_tcbhashsize = hashsize;
1181 /* Setup the tcp function block list */
1182 TAILQ_INIT(&t_functions);
1183 rw_init(&tcp_function_lock, "tcp_func_lock");
1184 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1186 /* Initialize the TCP logging data. */
1189 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1191 if (tcp_soreceive_stream) {
1193 tcp_usrreqs.pru_soreceive = soreceive_stream;
1196 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1201 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1203 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1205 if (max_protohdr < TCP_MINPROTOHDR)
1206 max_protohdr = TCP_MINPROTOHDR;
1207 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1209 #undef TCP_MINPROTOHDR
1212 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1213 SHUTDOWN_PRI_DEFAULT);
1214 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1215 EVENTHANDLER_PRI_ANY);
1217 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1218 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1219 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1220 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK);
1221 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1222 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK);
1230 tcp_destroy(void *unused __unused)
1238 * All our processes are gone, all our sockets should be cleaned
1239 * up, which means, we should be past the tcp_discardcb() calls.
1240 * Sleep to let all tcpcb timers really disappear and cleanup.
1243 INP_LIST_RLOCK(&V_tcbinfo);
1244 n = V_tcbinfo.ipi_count;
1245 INP_LIST_RUNLOCK(&V_tcbinfo);
1248 pause("tcpdes", hz / 10);
1253 in_pcbinfo_destroy(&V_tcbinfo);
1254 /* tcp_discardcb() clears the sack_holes up. */
1255 uma_zdestroy(V_sack_hole_zone);
1256 uma_zdestroy(V_tcpcb_zone);
1259 * Cannot free the zone until all tcpcbs are released as we attach
1260 * the allocations to them.
1262 tcp_fastopen_destroy();
1265 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1267 printf("%s: WARNING: unable to deregister helper hook "
1268 "type=%d, id=%d: error %d returned\n", __func__,
1269 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1271 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1273 printf("%s: WARNING: unable to deregister helper hook "
1274 "type=%d, id=%d: error %d returned\n", __func__,
1275 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1279 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1289 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1290 * tcp_template used to store this data in mbufs, but we now recopy it out
1291 * of the tcpcb each time to conserve mbufs.
1294 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1296 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1298 INP_WLOCK_ASSERT(inp);
1301 if ((inp->inp_vflag & INP_IPV6) != 0) {
1302 struct ip6_hdr *ip6;
1304 ip6 = (struct ip6_hdr *)ip_ptr;
1305 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1306 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1307 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1308 (IPV6_VERSION & IPV6_VERSION_MASK);
1309 ip6->ip6_nxt = IPPROTO_TCP;
1310 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1311 ip6->ip6_src = inp->in6p_laddr;
1312 ip6->ip6_dst = inp->in6p_faddr;
1315 #if defined(INET6) && defined(INET)
1322 ip = (struct ip *)ip_ptr;
1323 ip->ip_v = IPVERSION;
1325 ip->ip_tos = inp->inp_ip_tos;
1329 ip->ip_ttl = inp->inp_ip_ttl;
1331 ip->ip_p = IPPROTO_TCP;
1332 ip->ip_src = inp->inp_laddr;
1333 ip->ip_dst = inp->inp_faddr;
1336 th->th_sport = inp->inp_lport;
1337 th->th_dport = inp->inp_fport;
1345 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1349 * Create template to be used to send tcp packets on a connection.
1350 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1351 * use for this function is in keepalives, which use tcp_respond.
1354 tcpip_maketemplate(struct inpcb *inp)
1358 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1361 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1366 * Send a single message to the TCP at address specified by
1367 * the given TCP/IP header. If m == NULL, then we make a copy
1368 * of the tcpiphdr at th and send directly to the addressed host.
1369 * This is used to force keep alive messages out using the TCP
1370 * template for a connection. If flags are given then we send
1371 * a message back to the TCP which originated the segment th,
1372 * and discard the mbuf containing it and any other attached mbufs.
1374 * In any case the ack and sequence number of the transmitted
1375 * segment are as specified by the parameters.
1377 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1380 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1381 tcp_seq ack, tcp_seq seq, int flags)
1390 struct ip6_hdr *ip6;
1393 int optlen, tlen, win;
1396 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1399 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1406 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1407 INP_WLOCK_ASSERT(inp);
1414 if (!(flags & TH_RST)) {
1415 win = sbspace(&inp->inp_socket->so_rcv);
1416 if (win > TCP_MAXWIN << tp->rcv_scale)
1417 win = TCP_MAXWIN << tp->rcv_scale;
1419 if ((tp->t_flags & TF_NOOPT) == 0)
1423 m = m_gethdr(M_NOWAIT, MT_DATA);
1426 m->m_data += max_linkhdr;
1429 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1430 sizeof(struct ip6_hdr));
1431 ip6 = mtod(m, struct ip6_hdr *);
1432 nth = (struct tcphdr *)(ip6 + 1);
1436 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1437 ip = mtod(m, struct ip *);
1438 nth = (struct tcphdr *)(ip + 1);
1440 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1442 } else if (!M_WRITABLE(m)) {
1445 /* Can't reuse 'm', allocate a new mbuf. */
1446 n = m_gethdr(M_NOWAIT, MT_DATA);
1452 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1458 n->m_data += max_linkhdr;
1459 /* m_len is set later */
1460 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1463 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1464 sizeof(struct ip6_hdr));
1465 ip6 = mtod(n, struct ip6_hdr *);
1466 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1467 nth = (struct tcphdr *)(ip6 + 1);
1471 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1472 ip = mtod(n, struct ip *);
1473 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1474 nth = (struct tcphdr *)(ip + 1);
1476 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1477 xchg(nth->th_dport, nth->th_sport, uint16_t);
1484 * XXX MRT We inherit the FIB, which is lucky.
1488 m->m_data = (caddr_t)ipgen;
1489 /* m_len is set later */
1492 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1493 nth = (struct tcphdr *)(ip6 + 1);
1497 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1498 nth = (struct tcphdr *)(ip + 1);
1502 * this is usually a case when an extension header
1503 * exists between the IPv6 header and the
1506 nth->th_sport = th->th_sport;
1507 nth->th_dport = th->th_dport;
1509 xchg(nth->th_dport, nth->th_sport, uint16_t);
1515 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1517 #if defined(INET) && defined(INET6)
1521 tlen = sizeof (struct tcpiphdr);
1525 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1526 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1527 m, tlen, (long)M_TRAILINGSPACE(m)));
1532 /* Make sure we have room. */
1533 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1534 m->m_next = m_get(M_NOWAIT, MT_DATA);
1536 optp = mtod(m->m_next, u_char *);
1541 optp = (u_char *) (nth + 1);
1547 if (tp->t_flags & TF_RCVD_TSTMP) {
1548 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1549 to.to_tsecr = tp->ts_recent;
1550 to.to_flags |= TOF_TS;
1552 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1553 /* TCP-MD5 (RFC2385). */
1554 if (tp->t_flags & TF_SIGNATURE)
1555 to.to_flags |= TOF_SIGNATURE;
1557 /* Add the options. */
1558 tlen += optlen = tcp_addoptions(&to, optp);
1560 /* Update m_len in the correct mbuf. */
1561 optm->m_len += optlen;
1567 ip6->ip6_vfc = IPV6_VERSION;
1568 ip6->ip6_nxt = IPPROTO_TCP;
1569 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1572 #if defined(INET) && defined(INET6)
1577 ip->ip_len = htons(tlen);
1578 ip->ip_ttl = V_ip_defttl;
1579 if (V_path_mtu_discovery)
1580 ip->ip_off |= htons(IP_DF);
1583 m->m_pkthdr.len = tlen;
1584 m->m_pkthdr.rcvif = NULL;
1588 * Packet is associated with a socket, so allow the
1589 * label of the response to reflect the socket label.
1591 INP_WLOCK_ASSERT(inp);
1592 mac_inpcb_create_mbuf(inp, m);
1595 * Packet is not associated with a socket, so possibly
1596 * update the label in place.
1598 mac_netinet_tcp_reply(m);
1601 nth->th_seq = htonl(seq);
1602 nth->th_ack = htonl(ack);
1604 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1605 nth->th_flags = flags;
1607 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1609 nth->th_win = htons((u_short)win);
1612 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1613 if (to.to_flags & TOF_SIGNATURE) {
1614 if (!TCPMD5_ENABLED() ||
1615 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1622 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1625 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1626 nth->th_sum = in6_cksum_pseudo(ip6,
1627 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1628 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1632 #if defined(INET6) && defined(INET)
1637 m->m_pkthdr.csum_flags = CSUM_TCP;
1638 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1639 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1643 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1644 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1646 TCP_PROBE3(debug__output, tp, th, m);
1648 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1652 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1653 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1656 #if defined(INET) && defined(INET6)
1661 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1662 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1668 * Create a new TCP control block, making an
1669 * empty reassembly queue and hooking it to the argument
1670 * protocol control block. The `inp' parameter must have
1671 * come from the zone allocator set up in tcp_init().
1674 tcp_newtcpcb(struct inpcb *inp)
1676 struct tcpcb_mem *tm;
1679 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1682 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1687 /* Initialise cc_var struct for this tcpcb. */
1689 tp->ccv->type = IPPROTO_TCP;
1690 tp->ccv->ccvc.tcp = tp;
1691 rw_rlock(&tcp_function_lock);
1692 tp->t_fb = tcp_func_set_ptr;
1693 refcount_acquire(&tp->t_fb->tfb_refcnt);
1694 rw_runlock(&tcp_function_lock);
1696 * Use the current system default CC algorithm.
1699 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1700 CC_ALGO(tp) = CC_DEFAULT();
1703 if (CC_ALGO(tp)->cb_init != NULL)
1704 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1705 if (tp->t_fb->tfb_tcp_fb_fini)
1706 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1707 refcount_release(&tp->t_fb->tfb_refcnt);
1708 uma_zfree(V_tcpcb_zone, tm);
1714 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1715 if (tp->t_fb->tfb_tcp_fb_fini)
1716 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1717 refcount_release(&tp->t_fb->tfb_refcnt);
1718 uma_zfree(V_tcpcb_zone, tm);
1724 tp->t_vnet = inp->inp_vnet;
1726 tp->t_timers = &tm->tt;
1727 TAILQ_INIT(&tp->t_segq);
1730 isipv6 ? V_tcp_v6mssdflt :
1734 /* Set up our timeouts. */
1735 callout_init(&tp->t_timers->tt_rexmt, 1);
1736 callout_init(&tp->t_timers->tt_persist, 1);
1737 callout_init(&tp->t_timers->tt_keep, 1);
1738 callout_init(&tp->t_timers->tt_2msl, 1);
1739 callout_init(&tp->t_timers->tt_delack, 1);
1741 if (V_tcp_do_rfc1323)
1742 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1744 tp->t_flags |= TF_SACK_PERMIT;
1745 TAILQ_INIT(&tp->snd_holes);
1747 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1750 in_pcbref(inp); /* Reference for tcpcb */
1754 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1755 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1756 * reasonable initial retransmit time.
1758 tp->t_srtt = TCPTV_SRTTBASE;
1759 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1760 tp->t_rttmin = tcp_rexmit_min;
1761 tp->t_rxtcur = tcp_rexmit_initial;
1762 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1763 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1764 tp->t_rcvtime = ticks;
1766 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1767 * because the socket may be bound to an IPv6 wildcard address,
1768 * which may match an IPv4-mapped IPv6 address.
1770 inp->inp_ip_ttl = V_ip_defttl;
1774 * Init the TCP PCAP queues.
1776 tcp_pcap_tcpcb_init(tp);
1779 /* Initialize the per-TCPCB log data. */
1780 tcp_log_tcpcbinit(tp);
1782 if (tp->t_fb->tfb_tcp_fb_init) {
1783 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1786 if (V_tcp_perconn_stats_enable == 1)
1787 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
1789 return (tp); /* XXX */
1793 * Switch the congestion control algorithm back to NewReno for any active
1794 * control blocks using an algorithm which is about to go away.
1795 * This ensures the CC framework can allow the unload to proceed without leaving
1796 * any dangling pointers which would trigger a panic.
1797 * Returning non-zero would inform the CC framework that something went wrong
1798 * and it would be unsafe to allow the unload to proceed. However, there is no
1799 * way for this to occur with this implementation so we always return zero.
1802 tcp_ccalgounload(struct cc_algo *unload_algo)
1804 struct cc_algo *tmpalgo;
1807 VNET_ITERATOR_DECL(vnet_iter);
1810 * Check all active control blocks across all network stacks and change
1811 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1812 * requires cleanup code to be run, call it.
1815 VNET_FOREACH(vnet_iter) {
1816 CURVNET_SET(vnet_iter);
1817 INP_INFO_WLOCK(&V_tcbinfo);
1819 * New connections already part way through being initialised
1820 * with the CC algo we're removing will not race with this code
1821 * because the INP_INFO_WLOCK is held during initialisation. We
1822 * therefore don't enter the loop below until the connection
1823 * list has stabilised.
1825 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1827 /* Important to skip tcptw structs. */
1828 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1829 (tp = intotcpcb(inp)) != NULL) {
1831 * By holding INP_WLOCK here, we are assured
1832 * that the connection is not currently
1833 * executing inside the CC module's functions
1834 * i.e. it is safe to make the switch back to
1837 if (CC_ALGO(tp) == unload_algo) {
1838 tmpalgo = CC_ALGO(tp);
1839 if (tmpalgo->cb_destroy != NULL)
1840 tmpalgo->cb_destroy(tp->ccv);
1843 * NewReno may allocate memory on
1844 * demand for certain stateful
1845 * configuration as needed, but is
1846 * coded to never fail on memory
1847 * allocation failure so it is a safe
1850 CC_ALGO(tp) = &newreno_cc_algo;
1855 INP_INFO_WUNLOCK(&V_tcbinfo);
1858 VNET_LIST_RUNLOCK();
1864 * Drop a TCP connection, reporting
1865 * the specified error. If connection is synchronized,
1866 * then send a RST to peer.
1869 tcp_drop(struct tcpcb *tp, int errno)
1871 struct socket *so = tp->t_inpcb->inp_socket;
1873 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1874 INP_WLOCK_ASSERT(tp->t_inpcb);
1876 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1877 tcp_state_change(tp, TCPS_CLOSED);
1878 (void) tp->t_fb->tfb_tcp_output(tp);
1879 TCPSTAT_INC(tcps_drops);
1881 TCPSTAT_INC(tcps_conndrops);
1882 if (errno == ETIMEDOUT && tp->t_softerror)
1883 errno = tp->t_softerror;
1884 so->so_error = errno;
1885 return (tcp_close(tp));
1889 tcp_discardcb(struct tcpcb *tp)
1891 struct inpcb *inp = tp->t_inpcb;
1892 struct socket *so = inp->inp_socket;
1894 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1896 int released __unused;
1898 INP_WLOCK_ASSERT(inp);
1901 * Make sure that all of our timers are stopped before we delete the
1904 * If stopping a timer fails, we schedule a discard function in same
1905 * callout, and the last discard function called will take care of
1906 * deleting the tcpcb.
1908 tp->t_timers->tt_draincnt = 0;
1909 tcp_timer_stop(tp, TT_REXMT);
1910 tcp_timer_stop(tp, TT_PERSIST);
1911 tcp_timer_stop(tp, TT_KEEP);
1912 tcp_timer_stop(tp, TT_2MSL);
1913 tcp_timer_stop(tp, TT_DELACK);
1914 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1916 * Call the stop-all function of the methods,
1917 * this function should call the tcp_timer_stop()
1918 * method with each of the function specific timeouts.
1919 * That stop will be called via the tfb_tcp_timer_stop()
1920 * which should use the async drain function of the
1921 * callout system (see tcp_var.h).
1923 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1927 * If we got enough samples through the srtt filter,
1928 * save the rtt and rttvar in the routing entry.
1929 * 'Enough' is arbitrarily defined as 4 rtt samples.
1930 * 4 samples is enough for the srtt filter to converge
1931 * to within enough % of the correct value; fewer samples
1932 * and we could save a bogus rtt. The danger is not high
1933 * as tcp quickly recovers from everything.
1934 * XXX: Works very well but needs some more statistics!
1936 if (tp->t_rttupdated >= 4) {
1937 struct hc_metrics_lite metrics;
1940 bzero(&metrics, sizeof(metrics));
1942 * Update the ssthresh always when the conditions below
1943 * are satisfied. This gives us better new start value
1944 * for the congestion avoidance for new connections.
1945 * ssthresh is only set if packet loss occurred on a session.
1947 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1948 * being torn down. Ideally this code would not use 'so'.
1950 ssthresh = tp->snd_ssthresh;
1951 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1953 * convert the limit from user data bytes to
1954 * packets then to packet data bytes.
1956 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1959 ssthresh *= (tp->t_maxseg +
1961 (isipv6 ? sizeof (struct ip6_hdr) +
1962 sizeof (struct tcphdr) :
1964 sizeof (struct tcpiphdr)
1971 metrics.rmx_ssthresh = ssthresh;
1973 metrics.rmx_rtt = tp->t_srtt;
1974 metrics.rmx_rttvar = tp->t_rttvar;
1975 metrics.rmx_cwnd = tp->snd_cwnd;
1976 metrics.rmx_sendpipe = 0;
1977 metrics.rmx_recvpipe = 0;
1979 tcp_hc_update(&inp->inp_inc, &metrics);
1982 /* free the reassembly queue, if any */
1983 tcp_reass_flush(tp);
1986 /* Disconnect offload device, if any. */
1987 if (tp->t_flags & TF_TOE)
1988 tcp_offload_detach(tp);
1991 tcp_free_sackholes(tp);
1994 /* Free the TCP PCAP queues. */
1995 tcp_pcap_drain(&(tp->t_inpkts));
1996 tcp_pcap_drain(&(tp->t_outpkts));
1999 /* Allow the CC algorithm to clean up after itself. */
2000 if (CC_ALGO(tp)->cb_destroy != NULL)
2001 CC_ALGO(tp)->cb_destroy(tp->ccv);
2005 khelp_destroy_osd(tp->osd);
2008 stats_blob_destroy(tp->t_stats);
2012 inp->inp_ppcb = NULL;
2013 if (tp->t_timers->tt_draincnt == 0) {
2014 /* We own the last reference on tcpcb, let's free it. */
2016 tcp_log_tcpcbfini(tp);
2018 TCPSTATES_DEC(tp->t_state);
2019 if (tp->t_fb->tfb_tcp_fb_fini)
2020 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2021 refcount_release(&tp->t_fb->tfb_refcnt);
2023 uma_zfree(V_tcpcb_zone, tp);
2024 released = in_pcbrele_wlocked(inp);
2025 KASSERT(!released, ("%s: inp %p should not have been released "
2026 "here", __func__, inp));
2031 tcp_timer_discard(void *ptp)
2035 struct epoch_tracker et;
2037 tp = (struct tcpcb *)ptp;
2038 CURVNET_SET(tp->t_vnet);
2039 NET_EPOCH_ENTER(et);
2041 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2044 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2045 ("%s: tcpcb has to be stopped here", __func__));
2046 tp->t_timers->tt_draincnt--;
2047 if (tp->t_timers->tt_draincnt == 0) {
2048 /* We own the last reference on this tcpcb, let's free it. */
2050 tcp_log_tcpcbfini(tp);
2052 TCPSTATES_DEC(tp->t_state);
2053 if (tp->t_fb->tfb_tcp_fb_fini)
2054 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2055 refcount_release(&tp->t_fb->tfb_refcnt);
2057 uma_zfree(V_tcpcb_zone, tp);
2058 if (in_pcbrele_wlocked(inp)) {
2070 * Attempt to close a TCP control block, marking it as dropped, and freeing
2071 * the socket if we hold the only reference.
2074 tcp_close(struct tcpcb *tp)
2076 struct inpcb *inp = tp->t_inpcb;
2079 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2080 INP_WLOCK_ASSERT(inp);
2083 if (tp->t_state == TCPS_LISTEN)
2084 tcp_offload_listen_stop(tp);
2087 * This releases the TFO pending counter resource for TFO listen
2088 * sockets as well as passively-created TFO sockets that transition
2089 * from SYN_RECEIVED to CLOSED.
2091 if (tp->t_tfo_pending) {
2092 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2093 tp->t_tfo_pending = NULL;
2096 TCPSTAT_INC(tcps_closed);
2097 if (tp->t_state != TCPS_CLOSED)
2098 tcp_state_change(tp, TCPS_CLOSED);
2099 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2100 so = inp->inp_socket;
2101 soisdisconnected(so);
2102 if (inp->inp_flags & INP_SOCKREF) {
2103 KASSERT(so->so_state & SS_PROTOREF,
2104 ("tcp_close: !SS_PROTOREF"));
2105 inp->inp_flags &= ~INP_SOCKREF;
2108 so->so_state &= ~SS_PROTOREF;
2118 VNET_ITERATOR_DECL(vnet_iter);
2123 VNET_LIST_RLOCK_NOSLEEP();
2124 VNET_FOREACH(vnet_iter) {
2125 CURVNET_SET(vnet_iter);
2130 * Walk the tcpbs, if existing, and flush the reassembly queue,
2131 * if there is one...
2132 * XXX: The "Net/3" implementation doesn't imply that the TCP
2133 * reassembly queue should be flushed, but in a situation
2134 * where we're really low on mbufs, this is potentially
2137 INP_INFO_WLOCK(&V_tcbinfo);
2138 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2140 if (inpb->inp_flags & INP_TIMEWAIT) {
2144 if ((tcpb = intotcpcb(inpb)) != NULL) {
2145 tcp_reass_flush(tcpb);
2146 tcp_clean_sackreport(tcpb);
2148 tcp_log_drain(tcpb);
2151 if (tcp_pcap_aggressive_free) {
2152 /* Free the TCP PCAP queues. */
2153 tcp_pcap_drain(&(tcpb->t_inpkts));
2154 tcp_pcap_drain(&(tcpb->t_outpkts));
2160 INP_INFO_WUNLOCK(&V_tcbinfo);
2163 VNET_LIST_RUNLOCK_NOSLEEP();
2167 * Notify a tcp user of an asynchronous error;
2168 * store error as soft error, but wake up user
2169 * (for now, won't do anything until can select for soft error).
2171 * Do not wake up user since there currently is no mechanism for
2172 * reporting soft errors (yet - a kqueue filter may be added).
2174 static struct inpcb *
2175 tcp_notify(struct inpcb *inp, int error)
2179 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2180 INP_WLOCK_ASSERT(inp);
2182 if ((inp->inp_flags & INP_TIMEWAIT) ||
2183 (inp->inp_flags & INP_DROPPED))
2186 tp = intotcpcb(inp);
2187 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2190 * Ignore some errors if we are hooked up.
2191 * If connection hasn't completed, has retransmitted several times,
2192 * and receives a second error, give up now. This is better
2193 * than waiting a long time to establish a connection that
2194 * can never complete.
2196 if (tp->t_state == TCPS_ESTABLISHED &&
2197 (error == EHOSTUNREACH || error == ENETUNREACH ||
2198 error == EHOSTDOWN)) {
2199 if (inp->inp_route.ro_rt) {
2200 RTFREE(inp->inp_route.ro_rt);
2201 inp->inp_route.ro_rt = (struct rtentry *)NULL;
2204 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2206 tp = tcp_drop(tp, error);
2212 tp->t_softerror = error;
2216 wakeup( &so->so_timeo);
2223 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2225 struct epoch_tracker et;
2230 if (req->newptr != NULL)
2233 if (req->oldptr == NULL) {
2236 n = V_tcbinfo.ipi_count +
2237 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2238 n += imax(n / 8, 10);
2239 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2243 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2246 bzero(&xig, sizeof(xig));
2247 xig.xig_len = sizeof xig;
2248 xig.xig_count = V_tcbinfo.ipi_count +
2249 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2250 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2251 xig.xig_sogen = so_gencnt;
2252 error = SYSCTL_OUT(req, &xig, sizeof xig);
2256 error = syncache_pcblist(req);
2260 NET_EPOCH_ENTER(et);
2261 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2263 inp = CK_LIST_NEXT(inp, inp_list)) {
2265 if (inp->inp_gencnt <= xig.xig_gen) {
2267 * XXX: This use of cr_cansee(), introduced with
2268 * TCP state changes, is not quite right, but for
2269 * now, better than nothing.
2271 if (inp->inp_flags & INP_TIMEWAIT) {
2272 if (intotw(inp) != NULL)
2273 error = cr_cansee(req->td->td_ucred,
2274 intotw(inp)->tw_cred);
2276 error = EINVAL; /* Skip this inp. */
2278 error = cr_canseeinpcb(req->td->td_ucred, inp);
2282 tcp_inptoxtp(inp, &xt);
2284 error = SYSCTL_OUT(req, &xt, sizeof xt);
2297 * Give the user an updated idea of our state.
2298 * If the generation differs from what we told
2299 * her before, she knows that something happened
2300 * while we were processing this request, and it
2301 * might be necessary to retry.
2303 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2304 xig.xig_sogen = so_gencnt;
2305 xig.xig_count = V_tcbinfo.ipi_count +
2306 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2307 error = SYSCTL_OUT(req, &xig, sizeof xig);
2313 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2314 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2315 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2319 tcp_getcred(SYSCTL_HANDLER_ARGS)
2322 struct sockaddr_in addrs[2];
2323 struct epoch_tracker et;
2327 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2330 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2333 NET_EPOCH_ENTER(et);
2334 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2335 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2338 if (inp->inp_socket == NULL)
2341 error = cr_canseeinpcb(req->td->td_ucred, inp);
2343 cru2x(inp->inp_cred, &xuc);
2348 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2352 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2353 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2354 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2359 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2361 struct epoch_tracker et;
2363 struct sockaddr_in6 addrs[2];
2370 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2373 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2376 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2377 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2380 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2382 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2389 NET_EPOCH_ENTER(et);
2392 inp = in_pcblookup(&V_tcbinfo,
2393 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2395 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2396 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2399 inp = in6_pcblookup(&V_tcbinfo,
2400 &addrs[1].sin6_addr, addrs[1].sin6_port,
2401 &addrs[0].sin6_addr, addrs[0].sin6_port,
2402 INPLOOKUP_RLOCKPCB, NULL);
2405 if (inp->inp_socket == NULL)
2408 error = cr_canseeinpcb(req->td->td_ucred, inp);
2410 cru2x(inp->inp_cred, &xuc);
2415 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2419 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2420 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2421 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2427 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2429 struct ip *ip = vip;
2431 struct in_addr faddr;
2434 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2436 struct in_conninfo inc;
2437 tcp_seq icmp_tcp_seq;
2440 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2441 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2444 if (cmd == PRC_MSGSIZE)
2445 notify = tcp_mtudisc_notify;
2446 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2447 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2448 cmd == PRC_TIMXCEED_INTRANS) && ip)
2449 notify = tcp_drop_syn_sent;
2452 * Hostdead is ugly because it goes linearly through all PCBs.
2453 * XXX: We never get this from ICMP, otherwise it makes an
2454 * excellent DoS attack on machines with many connections.
2456 else if (cmd == PRC_HOSTDEAD)
2458 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2462 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2466 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2467 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2468 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2469 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2470 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2471 /* signal EHOSTDOWN, as it flushes the cached route */
2472 inp = (*notify)(inp, EHOSTDOWN);
2475 icmp_tcp_seq = th->th_seq;
2477 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2478 !(inp->inp_flags & INP_DROPPED) &&
2479 !(inp->inp_socket == NULL)) {
2480 tp = intotcpcb(inp);
2481 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2482 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2483 if (cmd == PRC_MSGSIZE) {
2486 * If we got a needfrag set the MTU
2487 * in the route to the suggested new
2488 * value (if given) and then notify.
2490 mtu = ntohs(icp->icmp_nextmtu);
2492 * If no alternative MTU was
2493 * proposed, try the next smaller
2498 ntohs(ip->ip_len), 1);
2499 if (mtu < V_tcp_minmss +
2500 sizeof(struct tcpiphdr))
2501 mtu = V_tcp_minmss +
2502 sizeof(struct tcpiphdr);
2504 * Only process the offered MTU if it
2505 * is smaller than the current one.
2507 if (mtu < tp->t_maxseg +
2508 sizeof(struct tcpiphdr)) {
2509 bzero(&inc, sizeof(inc));
2510 inc.inc_faddr = faddr;
2512 inp->inp_inc.inc_fibnum;
2513 tcp_hc_updatemtu(&inc, mtu);
2514 tcp_mtudisc(inp, mtu);
2517 inp = (*notify)(inp,
2518 inetctlerrmap[cmd]);
2522 bzero(&inc, sizeof(inc));
2523 inc.inc_fport = th->th_dport;
2524 inc.inc_lport = th->th_sport;
2525 inc.inc_faddr = faddr;
2526 inc.inc_laddr = ip->ip_src;
2527 syncache_unreach(&inc, icmp_tcp_seq);
2537 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2539 struct in6_addr *dst;
2540 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2541 struct ip6_hdr *ip6;
2545 struct icmp6_hdr *icmp6;
2546 struct ip6ctlparam *ip6cp = NULL;
2547 const struct sockaddr_in6 *sa6_src = NULL;
2548 struct in_conninfo inc;
2553 tcp_seq icmp_tcp_seq;
2557 if (sa->sa_family != AF_INET6 ||
2558 sa->sa_len != sizeof(struct sockaddr_in6))
2561 /* if the parameter is from icmp6, decode it. */
2563 ip6cp = (struct ip6ctlparam *)d;
2564 icmp6 = ip6cp->ip6c_icmp6;
2566 ip6 = ip6cp->ip6c_ip6;
2567 off = ip6cp->ip6c_off;
2568 sa6_src = ip6cp->ip6c_src;
2569 dst = ip6cp->ip6c_finaldst;
2573 off = 0; /* fool gcc */
2578 if (cmd == PRC_MSGSIZE)
2579 notify = tcp_mtudisc_notify;
2580 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2581 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2582 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2583 notify = tcp_drop_syn_sent;
2586 * Hostdead is ugly because it goes linearly through all PCBs.
2587 * XXX: We never get this from ICMP, otherwise it makes an
2588 * excellent DoS attack on machines with many connections.
2590 else if (cmd == PRC_HOSTDEAD)
2592 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2596 in6_pcbnotify(&V_tcbinfo, sa, 0,
2597 (const struct sockaddr *)sa6_src,
2598 0, cmd, NULL, notify);
2602 /* Check if we can safely get the ports from the tcp hdr */
2605 (int32_t) (off + sizeof(struct tcp_ports)))) {
2608 bzero(&t_ports, sizeof(struct tcp_ports));
2609 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2610 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2611 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2612 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2613 /* signal EHOSTDOWN, as it flushes the cached route */
2614 inp = (*notify)(inp, EHOSTDOWN);
2617 off += sizeof(struct tcp_ports);
2618 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2621 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2623 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2624 !(inp->inp_flags & INP_DROPPED) &&
2625 !(inp->inp_socket == NULL)) {
2626 tp = intotcpcb(inp);
2627 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2628 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2629 if (cmd == PRC_MSGSIZE) {
2632 * If we got a needfrag set the MTU
2633 * in the route to the suggested new
2634 * value (if given) and then notify.
2636 mtu = ntohl(icmp6->icmp6_mtu);
2638 * If no alternative MTU was
2639 * proposed, or the proposed
2640 * MTU was too small, set to
2643 if (mtu < IPV6_MMTU)
2644 mtu = IPV6_MMTU - 8;
2645 bzero(&inc, sizeof(inc));
2646 inc.inc_fibnum = M_GETFIB(m);
2647 inc.inc_flags |= INC_ISIPV6;
2648 inc.inc6_faddr = *dst;
2649 if (in6_setscope(&inc.inc6_faddr,
2650 m->m_pkthdr.rcvif, NULL))
2653 * Only process the offered MTU if it
2654 * is smaller than the current one.
2656 if (mtu < tp->t_maxseg +
2657 sizeof (struct tcphdr) +
2658 sizeof (struct ip6_hdr)) {
2659 tcp_hc_updatemtu(&inc, mtu);
2660 tcp_mtudisc(inp, mtu);
2661 ICMP6STAT_INC(icp6s_pmtuchg);
2664 inp = (*notify)(inp,
2665 inet6ctlerrmap[cmd]);
2669 bzero(&inc, sizeof(inc));
2670 inc.inc_fibnum = M_GETFIB(m);
2671 inc.inc_flags |= INC_ISIPV6;
2672 inc.inc_fport = t_ports.th_dport;
2673 inc.inc_lport = t_ports.th_sport;
2674 inc.inc6_faddr = *dst;
2675 inc.inc6_laddr = ip6->ip6_src;
2676 syncache_unreach(&inc, icmp_tcp_seq);
2685 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2690 KASSERT(len >= SIPHASH_KEY_LENGTH,
2691 ("%s: keylen %u too short ", __func__, len));
2692 SipHash24_Init(&ctx);
2693 SipHash_SetKey(&ctx, (uint8_t *)key);
2694 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2695 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2696 switch (inc->inc_flags & INC_ISIPV6) {
2699 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2700 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2705 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2706 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2710 SipHash_Final((uint8_t *)hash, &ctx);
2712 return (hash[0] ^ hash[1]);
2716 tcp_new_ts_offset(struct in_conninfo *inc)
2718 struct in_conninfo inc_store, *local_inc;
2720 if (!V_tcp_ts_offset_per_conn) {
2721 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2722 inc_store.inc_lport = 0;
2723 inc_store.inc_fport = 0;
2724 local_inc = &inc_store;
2728 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2729 sizeof(V_ts_offset_secret)));
2733 * Following is where TCP initial sequence number generation occurs.
2735 * There are two places where we must use initial sequence numbers:
2736 * 1. In SYN-ACK packets.
2737 * 2. In SYN packets.
2739 * All ISNs for SYN-ACK packets are generated by the syncache. See
2740 * tcp_syncache.c for details.
2742 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2743 * depends on this property. In addition, these ISNs should be
2744 * unguessable so as to prevent connection hijacking. To satisfy
2745 * the requirements of this situation, the algorithm outlined in
2746 * RFC 1948 is used, with only small modifications.
2748 * Implementation details:
2750 * Time is based off the system timer, and is corrected so that it
2751 * increases by one megabyte per second. This allows for proper
2752 * recycling on high speed LANs while still leaving over an hour
2755 * As reading the *exact* system time is too expensive to be done
2756 * whenever setting up a TCP connection, we increment the time
2757 * offset in two ways. First, a small random positive increment
2758 * is added to isn_offset for each connection that is set up.
2759 * Second, the function tcp_isn_tick fires once per clock tick
2760 * and increments isn_offset as necessary so that sequence numbers
2761 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2762 * random positive increments serve only to ensure that the same
2763 * exact sequence number is never sent out twice (as could otherwise
2764 * happen when a port is recycled in less than the system tick
2767 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2768 * between seeding of isn_secret. This is normally set to zero,
2769 * as reseeding should not be necessary.
2771 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2772 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2773 * general, this means holding an exclusive (write) lock.
2776 #define ISN_BYTES_PER_SECOND 1048576
2777 #define ISN_STATIC_INCREMENT 4096
2778 #define ISN_RANDOM_INCREMENT (4096 - 1)
2779 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
2781 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2782 VNET_DEFINE_STATIC(int, isn_last);
2783 VNET_DEFINE_STATIC(int, isn_last_reseed);
2784 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2785 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2787 #define V_isn_secret VNET(isn_secret)
2788 #define V_isn_last VNET(isn_last)
2789 #define V_isn_last_reseed VNET(isn_last_reseed)
2790 #define V_isn_offset VNET(isn_offset)
2791 #define V_isn_offset_old VNET(isn_offset_old)
2794 tcp_new_isn(struct in_conninfo *inc)
2797 u_int32_t projected_offset;
2800 /* Seed if this is the first use, reseed if requested. */
2801 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2802 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2804 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2805 V_isn_last_reseed = ticks;
2808 /* Compute the hash and return the ISN. */
2809 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2810 sizeof(V_isn_secret));
2811 V_isn_offset += ISN_STATIC_INCREMENT +
2812 (arc4random() & ISN_RANDOM_INCREMENT);
2813 if (ticks != V_isn_last) {
2814 projected_offset = V_isn_offset_old +
2815 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2816 if (SEQ_GT(projected_offset, V_isn_offset))
2817 V_isn_offset = projected_offset;
2818 V_isn_offset_old = V_isn_offset;
2821 new_isn += V_isn_offset;
2827 * When a specific ICMP unreachable message is received and the
2828 * connection state is SYN-SENT, drop the connection. This behavior
2829 * is controlled by the icmp_may_rst sysctl.
2832 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2837 INP_WLOCK_ASSERT(inp);
2839 if ((inp->inp_flags & INP_TIMEWAIT) ||
2840 (inp->inp_flags & INP_DROPPED))
2843 tp = intotcpcb(inp);
2844 if (tp->t_state != TCPS_SYN_SENT)
2847 if (IS_FASTOPEN(tp->t_flags))
2848 tcp_fastopen_disable_path(tp);
2850 tp = tcp_drop(tp, errno);
2858 * When `need fragmentation' ICMP is received, update our idea of the MSS
2859 * based on the new value. Also nudge TCP to send something, since we
2860 * know the packet we just sent was dropped.
2861 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2863 static struct inpcb *
2864 tcp_mtudisc_notify(struct inpcb *inp, int error)
2867 tcp_mtudisc(inp, -1);
2872 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2877 INP_WLOCK_ASSERT(inp);
2878 if ((inp->inp_flags & INP_TIMEWAIT) ||
2879 (inp->inp_flags & INP_DROPPED))
2882 tp = intotcpcb(inp);
2883 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2885 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2887 so = inp->inp_socket;
2888 SOCKBUF_LOCK(&so->so_snd);
2889 /* If the mss is larger than the socket buffer, decrease the mss. */
2890 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2891 tp->t_maxseg = so->so_snd.sb_hiwat;
2892 SOCKBUF_UNLOCK(&so->so_snd);
2894 TCPSTAT_INC(tcps_mturesent);
2896 tp->snd_nxt = tp->snd_una;
2897 tcp_free_sackholes(tp);
2898 tp->snd_recover = tp->snd_max;
2899 if (tp->t_flags & TF_SACK_PERMIT)
2900 EXIT_FASTRECOVERY(tp->t_flags);
2901 tp->t_fb->tfb_tcp_output(tp);
2906 * Look-up the routing entry to the peer of this inpcb. If no route
2907 * is found and it cannot be allocated, then return 0. This routine
2908 * is called by TCP routines that access the rmx structure and by
2909 * tcp_mss_update to get the peer/interface MTU.
2912 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2914 struct nhop4_extended nh4;
2916 uint32_t maxmtu = 0;
2918 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2920 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2922 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2923 NHR_REF, 0, &nh4) != 0)
2927 maxmtu = nh4.nh_mtu;
2929 /* Report additional interface capabilities. */
2931 if (ifp->if_capenable & IFCAP_TSO4 &&
2932 ifp->if_hwassist & CSUM_TSO) {
2933 cap->ifcap |= CSUM_TSO;
2934 cap->tsomax = ifp->if_hw_tsomax;
2935 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2936 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2939 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2947 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2949 struct nhop6_extended nh6;
2950 struct in6_addr dst6;
2953 uint32_t maxmtu = 0;
2955 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2957 if (inc->inc_flags & INC_IPV6MINMTU)
2960 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2961 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2962 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2967 maxmtu = nh6.nh_mtu;
2969 /* Report additional interface capabilities. */
2971 if (ifp->if_capenable & IFCAP_TSO6 &&
2972 ifp->if_hwassist & CSUM_TSO) {
2973 cap->ifcap |= CSUM_TSO;
2974 cap->tsomax = ifp->if_hw_tsomax;
2975 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2976 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2979 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2987 * Calculate effective SMSS per RFC5681 definition for a given TCP
2988 * connection at its current state, taking into account SACK and etc.
2991 tcp_maxseg(const struct tcpcb *tp)
2995 if (tp->t_flags & TF_NOOPT)
2996 return (tp->t_maxseg);
2999 * Here we have a simplified code from tcp_addoptions(),
3000 * without a proper loop, and having most of paddings hardcoded.
3001 * We might make mistakes with padding here in some edge cases,
3002 * but this is harmless, since result of tcp_maxseg() is used
3003 * only in cwnd and ssthresh estimations.
3005 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3006 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3007 if (tp->t_flags & TF_RCVD_TSTMP)
3008 optlen = TCPOLEN_TSTAMP_APPA;
3011 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3012 if (tp->t_flags & TF_SIGNATURE)
3013 optlen += PAD(TCPOLEN_SIGNATURE);
3015 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3016 optlen += TCPOLEN_SACKHDR;
3017 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3018 optlen = PAD(optlen);
3021 if (tp->t_flags & TF_REQ_TSTMP)
3022 optlen = TCPOLEN_TSTAMP_APPA;
3024 optlen = PAD(TCPOLEN_MAXSEG);
3025 if (tp->t_flags & TF_REQ_SCALE)
3026 optlen += PAD(TCPOLEN_WINDOW);
3027 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3028 if (tp->t_flags & TF_SIGNATURE)
3029 optlen += PAD(TCPOLEN_SIGNATURE);
3031 if (tp->t_flags & TF_SACK_PERMIT)
3032 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3035 optlen = min(optlen, TCP_MAXOLEN);
3036 return (tp->t_maxseg - optlen);
3040 sysctl_drop(SYSCTL_HANDLER_ARGS)
3042 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3043 struct sockaddr_storage addrs[2];
3047 struct sockaddr_in *fin, *lin;
3048 struct epoch_tracker et;
3050 struct sockaddr_in6 *fin6, *lin6;
3061 if (req->oldptr != NULL || req->oldlen != 0)
3063 if (req->newptr == NULL)
3065 if (req->newlen < sizeof(addrs))
3067 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3071 switch (addrs[0].ss_family) {
3074 fin6 = (struct sockaddr_in6 *)&addrs[0];
3075 lin6 = (struct sockaddr_in6 *)&addrs[1];
3076 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3077 lin6->sin6_len != sizeof(struct sockaddr_in6))
3079 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3080 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3082 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3083 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3084 fin = (struct sockaddr_in *)&addrs[0];
3085 lin = (struct sockaddr_in *)&addrs[1];
3088 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3091 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3098 fin = (struct sockaddr_in *)&addrs[0];
3099 lin = (struct sockaddr_in *)&addrs[1];
3100 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3101 lin->sin_len != sizeof(struct sockaddr_in))
3108 NET_EPOCH_ENTER(et);
3109 switch (addrs[0].ss_family) {
3112 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3113 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3114 INPLOOKUP_WLOCKPCB, NULL);
3119 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3120 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3125 if (inp->inp_flags & INP_TIMEWAIT) {
3127 * XXXRW: There currently exists a state where an
3128 * inpcb is present, but its timewait state has been
3129 * discarded. For now, don't allow dropping of this
3137 } else if (!(inp->inp_flags & INP_DROPPED) &&
3138 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3139 tp = intotcpcb(inp);
3140 tp = tcp_drop(tp, ECONNABORTED);
3151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3152 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3153 0, sysctl_drop, "", "Drop TCP connection");
3157 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3159 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3160 struct sockaddr_storage addrs[2];
3162 struct sockaddr_in *fin, *lin;
3163 struct epoch_tracker et;
3165 struct sockaddr_in6 *fin6, *lin6;
3176 if (req->oldptr != NULL || req->oldlen != 0)
3178 if (req->newptr == NULL)
3180 if (req->newlen < sizeof(addrs))
3182 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3186 switch (addrs[0].ss_family) {
3189 fin6 = (struct sockaddr_in6 *)&addrs[0];
3190 lin6 = (struct sockaddr_in6 *)&addrs[1];
3191 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3192 lin6->sin6_len != sizeof(struct sockaddr_in6))
3194 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3195 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3197 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3198 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3199 fin = (struct sockaddr_in *)&addrs[0];
3200 lin = (struct sockaddr_in *)&addrs[1];
3203 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3206 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3213 fin = (struct sockaddr_in *)&addrs[0];
3214 lin = (struct sockaddr_in *)&addrs[1];
3215 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3216 lin->sin_len != sizeof(struct sockaddr_in))
3223 NET_EPOCH_ENTER(et);
3224 switch (addrs[0].ss_family) {
3227 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3228 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3229 INPLOOKUP_WLOCKPCB, NULL);
3234 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3235 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3241 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3242 inp->inp_socket == NULL) {
3248 so = inp->inp_socket;
3250 error = ktls_set_tx_mode(so,
3251 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3261 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3262 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3263 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS");
3264 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3265 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3266 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS");
3270 * Generate a standardized TCP log line for use throughout the
3271 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3272 * allow use in the interrupt context.
3274 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3275 * NB: The function may return NULL if memory allocation failed.
3277 * Due to header inclusion and ordering limitations the struct ip
3278 * and ip6_hdr pointers have to be passed as void pointers.
3281 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3285 /* Is logging enabled? */
3286 if (tcp_log_in_vain == 0)
3289 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3293 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3297 /* Is logging enabled? */
3298 if (tcp_log_debug == 0)
3301 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3305 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3312 const struct ip6_hdr *ip6;
3314 ip6 = (const struct ip6_hdr *)ip6hdr;
3316 ip = (struct ip *)ip4hdr;
3319 * The log line looks like this:
3320 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3322 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3323 sizeof(PRINT_TH_FLAGS) + 1 +
3325 2 * INET6_ADDRSTRLEN;
3327 2 * INET_ADDRSTRLEN;
3330 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3334 strcat(s, "TCP: [");
3337 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3338 inet_ntoa_r(inc->inc_faddr, sp);
3340 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3342 inet_ntoa_r(inc->inc_laddr, sp);
3344 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3347 ip6_sprintf(sp, &inc->inc6_faddr);
3349 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3351 ip6_sprintf(sp, &inc->inc6_laddr);
3353 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3354 } else if (ip6 && th) {
3355 ip6_sprintf(sp, &ip6->ip6_src);
3357 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3359 ip6_sprintf(sp, &ip6->ip6_dst);
3361 sprintf(sp, "]:%i", ntohs(th->th_dport));
3364 } else if (ip && th) {
3365 inet_ntoa_r(ip->ip_src, sp);
3367 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3369 inet_ntoa_r(ip->ip_dst, sp);
3371 sprintf(sp, "]:%i", ntohs(th->th_dport));
3379 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3380 if (*(s + size - 1) != '\0')
3381 panic("%s: string too long", __func__);
3386 * A subroutine which makes it easy to track TCP state changes with DTrace.
3387 * This function shouldn't be called for t_state initializations that don't
3388 * correspond to actual TCP state transitions.
3391 tcp_state_change(struct tcpcb *tp, int newstate)
3393 #if defined(KDTRACE_HOOKS)
3394 int pstate = tp->t_state;
3397 TCPSTATES_DEC(tp->t_state);
3398 TCPSTATES_INC(newstate);
3399 tp->t_state = newstate;
3400 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3404 * Create an external-format (``xtcpcb'') structure using the information in
3405 * the kernel-format tcpcb structure pointed to by tp. This is done to
3406 * reduce the spew of irrelevant information over this interface, to isolate
3407 * user code from changes in the kernel structure, and potentially to provide
3408 * information-hiding if we decide that some of this information should be
3409 * hidden from users.
3412 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3414 struct tcpcb *tp = intotcpcb(inp);
3417 bzero(xt, sizeof(*xt));
3418 if (inp->inp_flags & INP_TIMEWAIT) {
3419 xt->t_state = TCPS_TIME_WAIT;
3421 xt->t_state = tp->t_state;
3422 xt->t_logstate = tp->t_logstate;
3423 xt->t_flags = tp->t_flags;
3424 xt->t_sndzerowin = tp->t_sndzerowin;
3425 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3426 xt->t_rcvoopack = tp->t_rcvoopack;
3428 now = getsbinuptime();
3429 #define COPYTIMER(ttt) do { \
3430 if (callout_active(&tp->t_timers->ttt)) \
3431 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3436 COPYTIMER(tt_delack);
3437 COPYTIMER(tt_rexmt);
3438 COPYTIMER(tt_persist);
3442 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3444 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3445 TCP_FUNCTION_NAME_LEN_MAX);
3447 (void)tcp_log_get_id(tp, xt->xt_logid);
3451 xt->xt_len = sizeof(struct xtcpcb);
3452 in_pcbtoxinpcb(inp, &xt->xt_inp);
3453 if (inp->inp_socket == NULL)
3454 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;