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>
130 #include <netipsec/ipsec_support.h>
132 #include <machine/in_cksum.h>
133 #include <crypto/siphash/siphash.h>
135 #include <security/mac/mac_framework.h>
137 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
139 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
142 #ifdef NETFLIX_EXP_DETECTION
143 /* Sack attack detection thresholds and such */
144 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
145 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
146 "Sack Attack detection thresholds");
147 int32_t tcp_force_detection = 0;
148 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
150 &tcp_force_detection, 0,
151 "Do we force detection even if the INP has it off?");
152 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */
153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
155 &tcp_sack_to_ack_thresh, 700,
156 "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
157 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */
158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
160 &tcp_sack_to_move_thresh, 600,
161 "Percentage of sack moves we must see above (10.1 percent is 101)");
162 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */
163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
165 &tcp_restoral_thresh, 550,
166 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
167 int32_t tcp_sad_decay_val = 800;
168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
170 &tcp_sad_decay_val, 800,
171 "The decay percentage (10.1 percent equals 101 )");
172 int32_t tcp_map_minimum = 500;
173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
175 &tcp_map_minimum, 500,
176 "Number of Map enteries before we start detection");
177 int32_t tcp_attack_on_turns_on_logging = 0;
178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
180 &tcp_attack_on_turns_on_logging, 0,
181 "When we have a positive hit on attack, do we turn on logging?");
182 int32_t tcp_sad_pacing_interval = 2000;
183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
185 &tcp_sad_pacing_interval, 2000,
186 "What is the minimum pacing interval for a classified attacker?");
188 int32_t tcp_sad_low_pps = 100;
189 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
191 &tcp_sad_low_pps, 100,
192 "What is the input pps that below which we do not decay?");
195 struct rwlock tcp_function_lock;
198 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
203 error = sysctl_handle_int(oidp, &new, 0, req);
204 if (error == 0 && req->newptr) {
205 if (new < TCP_MINMSS)
213 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
214 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
215 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
216 "Default TCP Maximum Segment Size");
220 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
224 new = V_tcp_v6mssdflt;
225 error = sysctl_handle_int(oidp, &new, 0, req);
226 if (error == 0 && req->newptr) {
227 if (new < TCP_MINMSS)
230 V_tcp_v6mssdflt = new;
235 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
236 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
237 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
238 "Default TCP Maximum Segment Size for IPv6");
242 * Minimum MSS we accept and use. This prevents DoS attacks where
243 * we are forced to a ridiculous low MSS like 20 and send hundreds
244 * of packets instead of one. The effect scales with the available
245 * bandwidth and quickly saturates the CPU and network interface
246 * with packet generation and sending. Set to zero to disable MINMSS
247 * checking. This setting prevents us from sending too small packets.
249 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
251 &VNET_NAME(tcp_minmss), 0,
252 "Minimum TCP Maximum Segment Size");
254 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
255 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
256 &VNET_NAME(tcp_do_rfc1323), 0,
257 "Enable rfc1323 (high performance TCP) extensions");
259 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 0;
260 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
261 &VNET_NAME(tcp_tolerate_missing_ts), 0,
262 "Tolerate missing TCP timestamps");
264 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
266 &VNET_NAME(tcp_ts_offset_per_conn), 0,
267 "Initialize TCP timestamps per connection instead of per host pair");
269 static int tcp_log_debug = 0;
270 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
271 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
273 static int tcp_tcbhashsize;
274 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
275 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
277 static int do_tcpdrain = 1;
278 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
279 "Enable tcp_drain routine for extra help when low on mbufs");
281 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
282 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
284 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
285 #define V_icmp_may_rst VNET(icmp_may_rst)
286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
287 &VNET_NAME(icmp_may_rst), 0,
288 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
290 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
291 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
293 &VNET_NAME(tcp_isn_reseed_interval), 0,
294 "Seconds between reseeding of ISN secret");
296 static int tcp_soreceive_stream;
297 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
298 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
300 VNET_DEFINE(uma_zone_t, sack_hole_zone);
301 #define V_sack_hole_zone VNET(sack_hole_zone)
302 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
304 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
309 new = V_tcp_map_entries_limit;
310 error = sysctl_handle_int(oidp, &new, 0, req);
311 if (error == 0 && req->newptr) {
312 /* only allow "0" and value > minimum */
313 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
316 V_tcp_map_entries_limit = new;
320 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
321 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
322 &VNET_NAME(tcp_map_entries_limit), 0,
323 &sysctl_net_inet_tcp_map_limit_check, "IU",
324 "Total sendmap entries limit");
326 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
327 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
328 &VNET_NAME(tcp_map_split_limit), 0,
329 "Total sendmap split entries limit");
332 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
335 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
336 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
337 #define V_ts_offset_secret VNET(ts_offset_secret)
339 static int tcp_default_fb_init(struct tcpcb *tp);
340 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
341 static int tcp_default_handoff_ok(struct tcpcb *tp);
342 static struct inpcb *tcp_notify(struct inpcb *, int);
343 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
344 static void tcp_mtudisc(struct inpcb *, int);
345 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
346 void *ip4hdr, const void *ip6hdr);
348 static struct tcp_function_block tcp_def_funcblk = {
349 .tfb_tcp_block_name = "freebsd",
350 .tfb_tcp_output = tcp_output,
351 .tfb_tcp_do_segment = tcp_do_segment,
352 .tfb_tcp_ctloutput = tcp_default_ctloutput,
353 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
354 .tfb_tcp_fb_init = tcp_default_fb_init,
355 .tfb_tcp_fb_fini = tcp_default_fb_fini,
358 static int tcp_fb_cnt = 0;
359 struct tcp_funchead t_functions;
360 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
362 static struct tcp_function_block *
363 find_tcp_functions_locked(struct tcp_function_set *fs)
365 struct tcp_function *f;
366 struct tcp_function_block *blk=NULL;
368 TAILQ_FOREACH(f, &t_functions, tf_next) {
369 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
377 static struct tcp_function_block *
378 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
380 struct tcp_function_block *rblk=NULL;
381 struct tcp_function *f;
383 TAILQ_FOREACH(f, &t_functions, tf_next) {
384 if (f->tf_fb == blk) {
395 struct tcp_function_block *
396 find_and_ref_tcp_functions(struct tcp_function_set *fs)
398 struct tcp_function_block *blk;
400 rw_rlock(&tcp_function_lock);
401 blk = find_tcp_functions_locked(fs);
403 refcount_acquire(&blk->tfb_refcnt);
404 rw_runlock(&tcp_function_lock);
408 struct tcp_function_block *
409 find_and_ref_tcp_fb(struct tcp_function_block *blk)
411 struct tcp_function_block *rblk;
413 rw_rlock(&tcp_function_lock);
414 rblk = find_tcp_fb_locked(blk, NULL);
416 refcount_acquire(&rblk->tfb_refcnt);
417 rw_runlock(&tcp_function_lock);
421 static struct tcp_function_block *
422 find_and_ref_tcp_default_fb(void)
424 struct tcp_function_block *rblk;
426 rw_rlock(&tcp_function_lock);
427 rblk = tcp_func_set_ptr;
428 refcount_acquire(&rblk->tfb_refcnt);
429 rw_runlock(&tcp_function_lock);
434 tcp_switch_back_to_default(struct tcpcb *tp)
436 struct tcp_function_block *tfb;
438 KASSERT(tp->t_fb != &tcp_def_funcblk,
439 ("%s: called by the built-in default stack", __func__));
442 * Release the old stack. This function will either find a new one
445 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
446 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
447 refcount_release(&tp->t_fb->tfb_refcnt);
450 * Now, we'll find a new function block to use.
451 * Start by trying the current user-selected
452 * default, unless this stack is the user-selected
455 tfb = find_and_ref_tcp_default_fb();
456 if (tfb == tp->t_fb) {
457 refcount_release(&tfb->tfb_refcnt);
460 /* Does the stack accept this connection? */
461 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
462 (*tfb->tfb_tcp_handoff_ok)(tp)) {
463 refcount_release(&tfb->tfb_refcnt);
466 /* Try to use that stack. */
468 /* Initialize the new stack. If it succeeds, we are done. */
470 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
471 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
475 * Initialization failed. Release the reference count on
478 refcount_release(&tfb->tfb_refcnt);
482 * If that wasn't feasible, use the built-in default
483 * stack which is not allowed to reject anyone.
485 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
487 /* there always should be a default */
488 panic("Can't refer to tcp_def_funcblk");
490 if (tfb->tfb_tcp_handoff_ok != NULL) {
491 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
492 /* The default stack cannot say no */
493 panic("Default stack rejects a new session?");
497 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
498 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
499 /* The default stack cannot fail */
500 panic("Default stack initialization failed");
505 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
508 struct tcp_function_set fs;
509 struct tcp_function_block *blk;
511 memset(&fs, 0, sizeof(fs));
512 rw_rlock(&tcp_function_lock);
513 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
516 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
517 fs.pcbcnt = blk->tfb_refcnt;
519 rw_runlock(&tcp_function_lock);
520 error = sysctl_handle_string(oidp, fs.function_set_name,
521 sizeof(fs.function_set_name), req);
523 /* Check for error or no change */
524 if (error != 0 || req->newptr == NULL)
527 rw_wlock(&tcp_function_lock);
528 blk = find_tcp_functions_locked(&fs);
530 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
534 tcp_func_set_ptr = blk;
536 rw_wunlock(&tcp_function_lock);
540 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
541 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
542 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
543 "Set/get the default TCP functions");
546 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
548 int error, cnt, linesz;
549 struct tcp_function *f;
555 rw_rlock(&tcp_function_lock);
556 TAILQ_FOREACH(f, &t_functions, tf_next) {
559 rw_runlock(&tcp_function_lock);
561 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
562 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
567 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
568 "Alias", "PCB count");
573 rw_rlock(&tcp_function_lock);
574 TAILQ_FOREACH(f, &t_functions, tf_next) {
575 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
576 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
577 f->tf_fb->tfb_tcp_block_name,
578 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
579 alias ? f->tf_name : "-",
580 f->tf_fb->tfb_refcnt);
581 if (linesz >= bufsz) {
589 rw_runlock(&tcp_function_lock);
591 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
592 free(buffer, M_TEMP);
596 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
597 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
598 NULL, 0, sysctl_net_inet_list_available, "A",
599 "list available TCP Function sets");
602 * Exports one (struct tcp_function_info) for each alias/name.
605 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
608 struct tcp_function *f;
609 struct tcp_function_info tfi;
612 * We don't allow writes.
614 if (req->newptr != NULL)
618 * Wire the old buffer so we can directly copy the functions to
619 * user space without dropping the lock.
621 if (req->oldptr != NULL) {
622 error = sysctl_wire_old_buffer(req, 0);
628 * Walk the list and copy out matching entries. If INVARIANTS
629 * is compiled in, also walk the list to verify the length of
630 * the list matches what we have recorded.
632 rw_rlock(&tcp_function_lock);
636 if (req->oldptr == NULL) {
641 TAILQ_FOREACH(f, &t_functions, tf_next) {
645 if (req->oldptr != NULL) {
646 bzero(&tfi, sizeof(tfi));
647 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
648 tfi.tfi_id = f->tf_fb->tfb_id;
649 (void)strlcpy(tfi.tfi_alias, f->tf_name,
650 sizeof(tfi.tfi_alias));
651 (void)strlcpy(tfi.tfi_name,
652 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
653 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
655 * Don't stop on error, as that is the
656 * mechanism we use to accumulate length
657 * information if the buffer was too short.
661 KASSERT(cnt == tcp_fb_cnt,
662 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
666 rw_runlock(&tcp_function_lock);
667 if (req->oldptr == NULL)
668 error = SYSCTL_OUT(req, NULL,
669 (cnt + 1) * sizeof(struct tcp_function_info));
674 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
675 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
676 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
677 "List TCP function block name-to-ID mappings");
680 * tfb_tcp_handoff_ok() function for the default stack.
681 * Note that we'll basically try to take all comers.
684 tcp_default_handoff_ok(struct tcpcb *tp)
691 * tfb_tcp_fb_init() function for the default stack.
693 * This handles making sure we have appropriate timers set if you are
694 * transitioning a socket that has some amount of setup done.
696 * The init() fuction from the default can *never* return non-zero i.e.
697 * it is required to always succeed since it is the stack of last resort!
700 tcp_default_fb_init(struct tcpcb *tp)
705 INP_WLOCK_ASSERT(tp->t_inpcb);
707 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
708 ("%s: connection %p in unexpected state %d", __func__, tp,
712 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
713 * know what to do for unexpected states (which includes TIME_WAIT).
715 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
719 * Make sure some kind of transmission timer is set if there is
722 so = tp->t_inpcb->inp_socket;
723 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
724 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
725 tcp_timer_active(tp, TT_PERSIST))) {
727 * If the session has established and it looks like it should
728 * be in the persist state, set the persist timer. Otherwise,
729 * set the retransmit timer.
731 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
732 (int32_t)(tp->snd_nxt - tp->snd_una) <
733 (int32_t)sbavail(&so->so_snd))
736 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
739 /* All non-embryonic sessions get a keepalive timer. */
740 if (!tcp_timer_active(tp, TT_KEEP))
741 tcp_timer_activate(tp, TT_KEEP,
742 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
746 * Make sure critical variables are initialized
747 * if transitioning while in Recovery.
749 if IN_FASTRECOVERY(tp->t_flags) {
750 if (tp->sackhint.recover_fs == 0)
751 tp->sackhint.recover_fs = max(1,
752 tp->snd_nxt - tp->snd_una);
759 * tfb_tcp_fb_fini() function for the default stack.
761 * This changes state as necessary (or prudent) to prepare for another stack
762 * to assume responsibility for the connection.
765 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
768 INP_WLOCK_ASSERT(tp->t_inpcb);
773 * Target size of TCP PCB hash tables. Must be a power of two.
775 * Note that this can be overridden by the kernel environment
776 * variable net.inet.tcp.tcbhashsize
779 #define TCBHASHSIZE 0
784 * Callouts should be moved into struct tcp directly. They are currently
785 * separate because the tcpcb structure is exported to userland for sysctl
786 * parsing purposes, which do not know about callouts.
797 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
798 #define V_tcpcb_zone VNET(tcpcb_zone)
800 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
801 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
803 static struct mtx isn_mtx;
805 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
806 #define ISN_LOCK() mtx_lock(&isn_mtx)
807 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
810 * TCP initialization.
813 tcp_zone_change(void *tag)
816 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
817 uma_zone_set_max(V_tcpcb_zone, maxsockets);
818 tcp_tw_zone_change();
822 tcp_inpcb_init(void *mem, int size, int flags)
824 struct inpcb *inp = mem;
826 INP_LOCK_INIT(inp, "inp", "tcpinp");
831 * Take a value and get the next power of 2 that doesn't overflow.
832 * Used to size the tcp_inpcb hash buckets.
835 maketcp_hashsize(int size)
841 * get the next power of 2 higher than maxsockets.
843 hashsize = 1 << fls(size);
844 /* catch overflow, and just go one power of 2 smaller */
845 if (hashsize < size) {
846 hashsize = 1 << (fls(size) - 1);
851 static volatile int next_tcp_stack_id = 1;
854 * Register a TCP function block with the name provided in the names
855 * array. (Note that this function does NOT automatically register
856 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
857 * explicitly include blk->tfb_tcp_block_name in the list of names if
858 * you wish to register the stack with that name.)
860 * Either all name registrations will succeed or all will fail. If
861 * a name registration fails, the function will update the num_names
862 * argument to point to the array index of the name that encountered
865 * Returns 0 on success, or an error code on failure.
868 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
869 const char *names[], int *num_names)
871 struct tcp_function *n;
872 struct tcp_function_set fs;
875 KASSERT(names != NULL && *num_names > 0,
876 ("%s: Called with 0-length name list", __func__));
877 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
878 KASSERT(rw_initialized(&tcp_function_lock),
879 ("%s: called too early", __func__));
881 if ((blk->tfb_tcp_output == NULL) ||
882 (blk->tfb_tcp_do_segment == NULL) ||
883 (blk->tfb_tcp_ctloutput == NULL) ||
884 (strlen(blk->tfb_tcp_block_name) == 0)) {
886 * These functions are required and you
892 if (blk->tfb_tcp_timer_stop_all ||
893 blk->tfb_tcp_timer_activate ||
894 blk->tfb_tcp_timer_active ||
895 blk->tfb_tcp_timer_stop) {
897 * If you define one timer function you
898 * must have them all.
900 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
901 (blk->tfb_tcp_timer_activate == NULL) ||
902 (blk->tfb_tcp_timer_active == NULL) ||
903 (blk->tfb_tcp_timer_stop == NULL)) {
909 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
914 refcount_init(&blk->tfb_refcnt, 0);
915 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
916 for (i = 0; i < *num_names; i++) {
917 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
924 (void)strlcpy(fs.function_set_name, names[i],
925 sizeof(fs.function_set_name));
926 rw_wlock(&tcp_function_lock);
927 if (find_tcp_functions_locked(&fs) != NULL) {
928 /* Duplicate name space not allowed */
929 rw_wunlock(&tcp_function_lock);
930 free(n, M_TCPFUNCTIONS);
934 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
935 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
937 rw_wunlock(&tcp_function_lock);
943 * Deregister the names we just added. Because registration failed
944 * for names[i], we don't need to deregister that name.
947 rw_wlock(&tcp_function_lock);
949 TAILQ_FOREACH(n, &t_functions, tf_next) {
950 if (!strncmp(n->tf_name, names[i],
951 TCP_FUNCTION_NAME_LEN_MAX)) {
952 TAILQ_REMOVE(&t_functions, n, tf_next);
955 free(n, M_TCPFUNCTIONS);
960 rw_wunlock(&tcp_function_lock);
965 * Register a TCP function block using the name provided in the name
968 * Returns 0 on success, or an error code on failure.
971 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
974 const char *name_list[1];
981 name_list[0] = blk->tfb_tcp_block_name;
982 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
987 * Register a TCP function block using the name defined in
988 * blk->tfb_tcp_block_name.
990 * Returns 0 on success, or an error code on failure.
993 register_tcp_functions(struct tcp_function_block *blk, int wait)
996 return (register_tcp_functions_as_name(blk, NULL, wait));
1000 * Deregister all names associated with a function block. This
1001 * functionally removes the function block from use within the system.
1003 * When called with a true quiesce argument, mark the function block
1004 * as being removed so no more stacks will use it and determine
1005 * whether the removal would succeed.
1007 * When called with a false quiesce argument, actually attempt the
1010 * When called with a force argument, attempt to switch all TCBs to
1011 * use the default stack instead of returning EBUSY.
1013 * Returns 0 on success (or if the removal would succeed, or an error
1017 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1020 struct tcp_function *f;
1022 if (blk == &tcp_def_funcblk) {
1023 /* You can't un-register the default */
1026 rw_wlock(&tcp_function_lock);
1027 if (blk == tcp_func_set_ptr) {
1028 /* You can't free the current default */
1029 rw_wunlock(&tcp_function_lock);
1032 /* Mark the block so no more stacks can use it. */
1033 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1035 * If TCBs are still attached to the stack, attempt to switch them
1036 * to the default stack.
1038 if (force && blk->tfb_refcnt) {
1041 VNET_ITERATOR_DECL(vnet_iter);
1043 rw_wunlock(&tcp_function_lock);
1046 VNET_FOREACH(vnet_iter) {
1047 CURVNET_SET(vnet_iter);
1048 INP_INFO_WLOCK(&V_tcbinfo);
1049 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1051 if (inp->inp_flags & INP_TIMEWAIT) {
1055 tp = intotcpcb(inp);
1056 if (tp == NULL || tp->t_fb != blk) {
1060 tcp_switch_back_to_default(tp);
1063 INP_INFO_WUNLOCK(&V_tcbinfo);
1066 VNET_LIST_RUNLOCK();
1068 rw_wlock(&tcp_function_lock);
1070 if (blk->tfb_refcnt) {
1071 /* TCBs still attached. */
1072 rw_wunlock(&tcp_function_lock);
1077 rw_wunlock(&tcp_function_lock);
1080 /* Remove any function names that map to this function block. */
1081 while (find_tcp_fb_locked(blk, &f) != NULL) {
1082 TAILQ_REMOVE(&t_functions, f, tf_next);
1085 free(f, M_TCPFUNCTIONS);
1087 rw_wunlock(&tcp_function_lock);
1094 const char *tcbhash_tuneable;
1097 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1100 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1101 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1102 printf("%s: WARNING: unable to register helper hook\n", __func__);
1103 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1104 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1105 printf("%s: WARNING: unable to register helper hook\n", __func__);
1108 if (tcp_stats_init())
1109 printf("%s: WARNING: unable to initialise TCP stats\n",
1112 hashsize = TCBHASHSIZE;
1113 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1114 if (hashsize == 0) {
1116 * Auto tune the hash size based on maxsockets.
1117 * A perfect hash would have a 1:1 mapping
1118 * (hashsize = maxsockets) however it's been
1119 * suggested that O(2) average is better.
1121 hashsize = maketcp_hashsize(maxsockets / 4);
1123 * Our historical default is 512,
1124 * do not autotune lower than this.
1128 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1129 printf("%s: %s auto tuned to %d\n", __func__,
1130 tcbhash_tuneable, hashsize);
1133 * We require a hashsize to be a power of two.
1134 * Previously if it was not a power of two we would just reset it
1135 * back to 512, which could be a nasty surprise if you did not notice
1136 * the error message.
1137 * Instead what we do is clip it to the closest power of two lower
1138 * than the specified hash value.
1140 if (!powerof2(hashsize)) {
1141 int oldhashsize = hashsize;
1143 hashsize = maketcp_hashsize(hashsize);
1144 /* prevent absurdly low value */
1147 printf("%s: WARNING: TCB hash size not a power of 2, "
1148 "clipped from %d to %d.\n", __func__, oldhashsize,
1151 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1152 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1155 * These have to be type stable for the benefit of the timers.
1157 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1158 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1159 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1160 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1166 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1167 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1168 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1170 tcp_fastopen_init();
1172 /* Skip initialization of globals for non-default instances. */
1173 if (!IS_DEFAULT_VNET(curvnet))
1176 tcp_reass_global_init();
1178 /* XXX virtualize those bellow? */
1179 tcp_delacktime = TCPTV_DELACK;
1180 tcp_keepinit = TCPTV_KEEP_INIT;
1181 tcp_keepidle = TCPTV_KEEP_IDLE;
1182 tcp_keepintvl = TCPTV_KEEPINTVL;
1183 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1184 tcp_msl = TCPTV_MSL;
1185 tcp_rexmit_initial = TCPTV_RTOBASE;
1186 if (tcp_rexmit_initial < 1)
1187 tcp_rexmit_initial = 1;
1188 tcp_rexmit_min = TCPTV_MIN;
1189 if (tcp_rexmit_min < 1)
1191 tcp_persmin = TCPTV_PERSMIN;
1192 tcp_persmax = TCPTV_PERSMAX;
1193 tcp_rexmit_slop = TCPTV_CPU_VAR;
1194 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1195 tcp_tcbhashsize = hashsize;
1197 /* Setup the tcp function block list */
1198 TAILQ_INIT(&t_functions);
1199 rw_init(&tcp_function_lock, "tcp_func_lock");
1200 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1202 /* Initialize the TCP logging data. */
1205 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1207 if (tcp_soreceive_stream) {
1209 tcp_usrreqs.pru_soreceive = soreceive_stream;
1212 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1217 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1219 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1221 if (max_protohdr < TCP_MINPROTOHDR)
1222 max_protohdr = TCP_MINPROTOHDR;
1223 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1225 #undef TCP_MINPROTOHDR
1228 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1229 SHUTDOWN_PRI_DEFAULT);
1230 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1231 EVENTHANDLER_PRI_ANY);
1233 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1234 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1235 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1236 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK);
1237 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1238 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK);
1246 tcp_destroy(void *unused __unused)
1254 * All our processes are gone, all our sockets should be cleaned
1255 * up, which means, we should be past the tcp_discardcb() calls.
1256 * Sleep to let all tcpcb timers really disappear and cleanup.
1259 INP_LIST_RLOCK(&V_tcbinfo);
1260 n = V_tcbinfo.ipi_count;
1261 INP_LIST_RUNLOCK(&V_tcbinfo);
1264 pause("tcpdes", hz / 10);
1269 in_pcbinfo_destroy(&V_tcbinfo);
1270 /* tcp_discardcb() clears the sack_holes up. */
1271 uma_zdestroy(V_sack_hole_zone);
1272 uma_zdestroy(V_tcpcb_zone);
1275 * Cannot free the zone until all tcpcbs are released as we attach
1276 * the allocations to them.
1278 tcp_fastopen_destroy();
1281 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1283 printf("%s: WARNING: unable to deregister helper hook "
1284 "type=%d, id=%d: error %d returned\n", __func__,
1285 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1287 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1289 printf("%s: WARNING: unable to deregister helper hook "
1290 "type=%d, id=%d: error %d returned\n", __func__,
1291 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1295 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1305 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1306 * tcp_template used to store this data in mbufs, but we now recopy it out
1307 * of the tcpcb each time to conserve mbufs.
1310 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1312 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1314 INP_WLOCK_ASSERT(inp);
1317 if ((inp->inp_vflag & INP_IPV6) != 0) {
1318 struct ip6_hdr *ip6;
1320 ip6 = (struct ip6_hdr *)ip_ptr;
1321 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1322 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1323 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1324 (IPV6_VERSION & IPV6_VERSION_MASK);
1325 ip6->ip6_nxt = IPPROTO_TCP;
1326 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1327 ip6->ip6_src = inp->in6p_laddr;
1328 ip6->ip6_dst = inp->in6p_faddr;
1331 #if defined(INET6) && defined(INET)
1338 ip = (struct ip *)ip_ptr;
1339 ip->ip_v = IPVERSION;
1341 ip->ip_tos = inp->inp_ip_tos;
1345 ip->ip_ttl = inp->inp_ip_ttl;
1347 ip->ip_p = IPPROTO_TCP;
1348 ip->ip_src = inp->inp_laddr;
1349 ip->ip_dst = inp->inp_faddr;
1352 th->th_sport = inp->inp_lport;
1353 th->th_dport = inp->inp_fport;
1361 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1365 * Create template to be used to send tcp packets on a connection.
1366 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1367 * use for this function is in keepalives, which use tcp_respond.
1370 tcpip_maketemplate(struct inpcb *inp)
1374 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1377 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1382 * Send a single message to the TCP at address specified by
1383 * the given TCP/IP header. If m == NULL, then we make a copy
1384 * of the tcpiphdr at th and send directly to the addressed host.
1385 * This is used to force keep alive messages out using the TCP
1386 * template for a connection. If flags are given then we send
1387 * a message back to the TCP which originated the segment th,
1388 * and discard the mbuf containing it and any other attached mbufs.
1390 * In any case the ack and sequence number of the transmitted
1391 * segment are as specified by the parameters.
1393 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1396 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1397 tcp_seq ack, tcp_seq seq, int flags)
1406 struct ip6_hdr *ip6;
1409 int optlen, tlen, win;
1412 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1416 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1423 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1424 INP_WLOCK_ASSERT(inp);
1431 if (!(flags & TH_RST)) {
1432 win = sbspace(&inp->inp_socket->so_rcv);
1433 if (win > TCP_MAXWIN << tp->rcv_scale)
1434 win = TCP_MAXWIN << tp->rcv_scale;
1436 if ((tp->t_flags & TF_NOOPT) == 0)
1440 m = m_gethdr(M_NOWAIT, MT_DATA);
1443 m->m_data += max_linkhdr;
1446 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1447 sizeof(struct ip6_hdr));
1448 ip6 = mtod(m, struct ip6_hdr *);
1449 nth = (struct tcphdr *)(ip6 + 1);
1453 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1454 ip = mtod(m, struct ip *);
1455 nth = (struct tcphdr *)(ip + 1);
1457 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1459 } else if (!M_WRITABLE(m)) {
1462 /* Can't reuse 'm', allocate a new mbuf. */
1463 n = m_gethdr(M_NOWAIT, MT_DATA);
1469 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1475 n->m_data += max_linkhdr;
1476 /* m_len is set later */
1477 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1480 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1481 sizeof(struct ip6_hdr));
1482 ip6 = mtod(n, struct ip6_hdr *);
1483 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1484 nth = (struct tcphdr *)(ip6 + 1);
1488 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1489 ip = mtod(n, struct ip *);
1490 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1491 nth = (struct tcphdr *)(ip + 1);
1493 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1494 xchg(nth->th_dport, nth->th_sport, uint16_t);
1501 * XXX MRT We inherit the FIB, which is lucky.
1505 m->m_data = (caddr_t)ipgen;
1506 /* m_len is set later */
1509 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1510 nth = (struct tcphdr *)(ip6 + 1);
1514 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1515 nth = (struct tcphdr *)(ip + 1);
1519 * this is usually a case when an extension header
1520 * exists between the IPv6 header and the
1523 nth->th_sport = th->th_sport;
1524 nth->th_dport = th->th_dport;
1526 xchg(nth->th_dport, nth->th_sport, uint16_t);
1532 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1534 #if defined(INET) && defined(INET6)
1538 tlen = sizeof (struct tcpiphdr);
1542 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1543 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1544 m, tlen, (long)M_TRAILINGSPACE(m)));
1549 /* Make sure we have room. */
1550 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1551 m->m_next = m_get(M_NOWAIT, MT_DATA);
1553 optp = mtod(m->m_next, u_char *);
1558 optp = (u_char *) (nth + 1);
1564 if (tp->t_flags & TF_RCVD_TSTMP) {
1565 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1566 to.to_tsecr = tp->ts_recent;
1567 to.to_flags |= TOF_TS;
1569 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1570 /* TCP-MD5 (RFC2385). */
1571 if (tp->t_flags & TF_SIGNATURE)
1572 to.to_flags |= TOF_SIGNATURE;
1574 /* Add the options. */
1575 tlen += optlen = tcp_addoptions(&to, optp);
1577 /* Update m_len in the correct mbuf. */
1578 optm->m_len += optlen;
1584 ip6->ip6_vfc = IPV6_VERSION;
1585 ip6->ip6_nxt = IPPROTO_TCP;
1586 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1589 #if defined(INET) && defined(INET6)
1594 ip->ip_len = htons(tlen);
1595 ip->ip_ttl = V_ip_defttl;
1596 if (V_path_mtu_discovery)
1597 ip->ip_off |= htons(IP_DF);
1600 m->m_pkthdr.len = tlen;
1601 m->m_pkthdr.rcvif = NULL;
1605 * Packet is associated with a socket, so allow the
1606 * label of the response to reflect the socket label.
1608 INP_WLOCK_ASSERT(inp);
1609 mac_inpcb_create_mbuf(inp, m);
1612 * Packet is not associated with a socket, so possibly
1613 * update the label in place.
1615 mac_netinet_tcp_reply(m);
1618 nth->th_seq = htonl(seq);
1619 nth->th_ack = htonl(ack);
1621 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1622 nth->th_flags = flags;
1624 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1626 nth->th_win = htons((u_short)win);
1629 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1630 if (to.to_flags & TOF_SIGNATURE) {
1631 if (!TCPMD5_ENABLED() ||
1632 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1639 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1642 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1643 nth->th_sum = in6_cksum_pseudo(ip6,
1644 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1645 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1649 #if defined(INET6) && defined(INET)
1654 m->m_pkthdr.csum_flags = CSUM_TCP;
1655 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1656 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1660 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1661 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1663 TCP_PROBE3(debug__output, tp, th, m);
1665 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1669 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1670 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1673 #if defined(INET) && defined(INET6)
1678 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1679 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1685 * Create a new TCP control block, making an
1686 * empty reassembly queue and hooking it to the argument
1687 * protocol control block. The `inp' parameter must have
1688 * come from the zone allocator set up in tcp_init().
1691 tcp_newtcpcb(struct inpcb *inp)
1693 struct tcpcb_mem *tm;
1696 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1699 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1704 /* Initialise cc_var struct for this tcpcb. */
1706 tp->ccv->type = IPPROTO_TCP;
1707 tp->ccv->ccvc.tcp = tp;
1708 rw_rlock(&tcp_function_lock);
1709 tp->t_fb = tcp_func_set_ptr;
1710 refcount_acquire(&tp->t_fb->tfb_refcnt);
1711 rw_runlock(&tcp_function_lock);
1713 * Use the current system default CC algorithm.
1716 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1717 CC_ALGO(tp) = CC_DEFAULT();
1720 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1723 in_pcbref(inp); /* Reference for tcpcb */
1726 if (CC_ALGO(tp)->cb_init != NULL)
1727 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1728 if (tp->t_fb->tfb_tcp_fb_fini)
1729 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1730 in_pcbrele_wlocked(inp);
1731 refcount_release(&tp->t_fb->tfb_refcnt);
1732 uma_zfree(V_tcpcb_zone, tm);
1738 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1739 if (tp->t_fb->tfb_tcp_fb_fini)
1740 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1741 in_pcbrele_wlocked(inp);
1742 refcount_release(&tp->t_fb->tfb_refcnt);
1743 uma_zfree(V_tcpcb_zone, tm);
1749 tp->t_vnet = inp->inp_vnet;
1751 tp->t_timers = &tm->tt;
1752 TAILQ_INIT(&tp->t_segq);
1755 isipv6 ? V_tcp_v6mssdflt :
1759 /* Set up our timeouts. */
1760 callout_init(&tp->t_timers->tt_rexmt, 1);
1761 callout_init(&tp->t_timers->tt_persist, 1);
1762 callout_init(&tp->t_timers->tt_keep, 1);
1763 callout_init(&tp->t_timers->tt_2msl, 1);
1764 callout_init(&tp->t_timers->tt_delack, 1);
1766 if (V_tcp_do_rfc1323)
1767 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1769 tp->t_flags |= TF_SACK_PERMIT;
1770 TAILQ_INIT(&tp->snd_holes);
1773 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1774 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1775 * reasonable initial retransmit time.
1777 tp->t_srtt = TCPTV_SRTTBASE;
1778 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1779 tp->t_rttmin = tcp_rexmit_min;
1780 tp->t_rxtcur = tcp_rexmit_initial;
1781 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1782 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1783 tp->t_rcvtime = ticks;
1785 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1786 * because the socket may be bound to an IPv6 wildcard address,
1787 * which may match an IPv4-mapped IPv6 address.
1789 inp->inp_ip_ttl = V_ip_defttl;
1793 * Init the TCP PCAP queues.
1795 tcp_pcap_tcpcb_init(tp);
1798 /* Initialize the per-TCPCB log data. */
1799 tcp_log_tcpcbinit(tp);
1801 tp->t_pacing_rate = -1;
1802 if (tp->t_fb->tfb_tcp_fb_init) {
1803 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
1804 refcount_release(&tp->t_fb->tfb_refcnt);
1805 in_pcbrele_wlocked(inp);
1806 uma_zfree(V_tcpcb_zone, tm);
1811 if (V_tcp_perconn_stats_enable == 1)
1812 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
1814 return (tp); /* XXX */
1818 * Switch the congestion control algorithm back to NewReno for any active
1819 * control blocks using an algorithm which is about to go away.
1820 * This ensures the CC framework can allow the unload to proceed without leaving
1821 * any dangling pointers which would trigger a panic.
1822 * Returning non-zero would inform the CC framework that something went wrong
1823 * and it would be unsafe to allow the unload to proceed. However, there is no
1824 * way for this to occur with this implementation so we always return zero.
1827 tcp_ccalgounload(struct cc_algo *unload_algo)
1829 struct cc_algo *tmpalgo;
1832 VNET_ITERATOR_DECL(vnet_iter);
1835 * Check all active control blocks across all network stacks and change
1836 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1837 * requires cleanup code to be run, call it.
1840 VNET_FOREACH(vnet_iter) {
1841 CURVNET_SET(vnet_iter);
1842 INP_INFO_WLOCK(&V_tcbinfo);
1844 * New connections already part way through being initialised
1845 * with the CC algo we're removing will not race with this code
1846 * because the INP_INFO_WLOCK is held during initialisation. We
1847 * therefore don't enter the loop below until the connection
1848 * list has stabilised.
1850 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1852 /* Important to skip tcptw structs. */
1853 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1854 (tp = intotcpcb(inp)) != NULL) {
1856 * By holding INP_WLOCK here, we are assured
1857 * that the connection is not currently
1858 * executing inside the CC module's functions
1859 * i.e. it is safe to make the switch back to
1862 if (CC_ALGO(tp) == unload_algo) {
1863 tmpalgo = CC_ALGO(tp);
1864 if (tmpalgo->cb_destroy != NULL)
1865 tmpalgo->cb_destroy(tp->ccv);
1868 * NewReno may allocate memory on
1869 * demand for certain stateful
1870 * configuration as needed, but is
1871 * coded to never fail on memory
1872 * allocation failure so it is a safe
1875 CC_ALGO(tp) = &newreno_cc_algo;
1880 INP_INFO_WUNLOCK(&V_tcbinfo);
1883 VNET_LIST_RUNLOCK();
1889 * Drop a TCP connection, reporting
1890 * the specified error. If connection is synchronized,
1891 * then send a RST to peer.
1894 tcp_drop(struct tcpcb *tp, int errno)
1896 struct socket *so = tp->t_inpcb->inp_socket;
1899 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1900 INP_WLOCK_ASSERT(tp->t_inpcb);
1902 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1903 tcp_state_change(tp, TCPS_CLOSED);
1904 (void) tp->t_fb->tfb_tcp_output(tp);
1905 TCPSTAT_INC(tcps_drops);
1907 TCPSTAT_INC(tcps_conndrops);
1908 if (errno == ETIMEDOUT && tp->t_softerror)
1909 errno = tp->t_softerror;
1910 so->so_error = errno;
1911 return (tcp_close(tp));
1915 tcp_discardcb(struct tcpcb *tp)
1917 struct inpcb *inp = tp->t_inpcb;
1918 struct socket *so = inp->inp_socket;
1920 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1922 int released __unused;
1924 INP_WLOCK_ASSERT(inp);
1927 * Make sure that all of our timers are stopped before we delete the
1930 * If stopping a timer fails, we schedule a discard function in same
1931 * callout, and the last discard function called will take care of
1932 * deleting the tcpcb.
1934 tp->t_timers->tt_draincnt = 0;
1935 tcp_timer_stop(tp, TT_REXMT);
1936 tcp_timer_stop(tp, TT_PERSIST);
1937 tcp_timer_stop(tp, TT_KEEP);
1938 tcp_timer_stop(tp, TT_2MSL);
1939 tcp_timer_stop(tp, TT_DELACK);
1940 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1942 * Call the stop-all function of the methods,
1943 * this function should call the tcp_timer_stop()
1944 * method with each of the function specific timeouts.
1945 * That stop will be called via the tfb_tcp_timer_stop()
1946 * which should use the async drain function of the
1947 * callout system (see tcp_var.h).
1949 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1953 * If we got enough samples through the srtt filter,
1954 * save the rtt and rttvar in the routing entry.
1955 * 'Enough' is arbitrarily defined as 4 rtt samples.
1956 * 4 samples is enough for the srtt filter to converge
1957 * to within enough % of the correct value; fewer samples
1958 * and we could save a bogus rtt. The danger is not high
1959 * as tcp quickly recovers from everything.
1960 * XXX: Works very well but needs some more statistics!
1962 if (tp->t_rttupdated >= 4) {
1963 struct hc_metrics_lite metrics;
1966 bzero(&metrics, sizeof(metrics));
1968 * Update the ssthresh always when the conditions below
1969 * are satisfied. This gives us better new start value
1970 * for the congestion avoidance for new connections.
1971 * ssthresh is only set if packet loss occurred on a session.
1973 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1974 * being torn down. Ideally this code would not use 'so'.
1976 ssthresh = tp->snd_ssthresh;
1977 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1979 * convert the limit from user data bytes to
1980 * packets then to packet data bytes.
1982 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1985 ssthresh *= (tp->t_maxseg +
1987 (isipv6 ? sizeof (struct ip6_hdr) +
1988 sizeof (struct tcphdr) :
1990 sizeof (struct tcpiphdr)
1997 metrics.rmx_ssthresh = ssthresh;
1999 metrics.rmx_rtt = tp->t_srtt;
2000 metrics.rmx_rttvar = tp->t_rttvar;
2001 metrics.rmx_cwnd = tp->snd_cwnd;
2002 metrics.rmx_sendpipe = 0;
2003 metrics.rmx_recvpipe = 0;
2005 tcp_hc_update(&inp->inp_inc, &metrics);
2008 /* free the reassembly queue, if any */
2009 tcp_reass_flush(tp);
2012 /* Disconnect offload device, if any. */
2013 if (tp->t_flags & TF_TOE)
2014 tcp_offload_detach(tp);
2017 tcp_free_sackholes(tp);
2020 /* Free the TCP PCAP queues. */
2021 tcp_pcap_drain(&(tp->t_inpkts));
2022 tcp_pcap_drain(&(tp->t_outpkts));
2025 /* Allow the CC algorithm to clean up after itself. */
2026 if (CC_ALGO(tp)->cb_destroy != NULL)
2027 CC_ALGO(tp)->cb_destroy(tp->ccv);
2031 khelp_destroy_osd(tp->osd);
2034 stats_blob_destroy(tp->t_stats);
2038 inp->inp_ppcb = NULL;
2039 if (tp->t_timers->tt_draincnt == 0) {
2040 /* We own the last reference on tcpcb, let's free it. */
2042 tcp_log_tcpcbfini(tp);
2044 TCPSTATES_DEC(tp->t_state);
2045 if (tp->t_fb->tfb_tcp_fb_fini)
2046 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2047 refcount_release(&tp->t_fb->tfb_refcnt);
2049 uma_zfree(V_tcpcb_zone, tp);
2050 released = in_pcbrele_wlocked(inp);
2051 KASSERT(!released, ("%s: inp %p should not have been released "
2052 "here", __func__, inp));
2057 tcp_timer_discard(void *ptp)
2061 struct epoch_tracker et;
2063 tp = (struct tcpcb *)ptp;
2064 CURVNET_SET(tp->t_vnet);
2065 NET_EPOCH_ENTER(et);
2067 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2070 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2071 ("%s: tcpcb has to be stopped here", __func__));
2072 tp->t_timers->tt_draincnt--;
2073 if (tp->t_timers->tt_draincnt == 0) {
2074 /* We own the last reference on this tcpcb, let's free it. */
2076 tcp_log_tcpcbfini(tp);
2078 TCPSTATES_DEC(tp->t_state);
2079 if (tp->t_fb->tfb_tcp_fb_fini)
2080 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2081 refcount_release(&tp->t_fb->tfb_refcnt);
2083 uma_zfree(V_tcpcb_zone, tp);
2084 if (in_pcbrele_wlocked(inp)) {
2096 * Attempt to close a TCP control block, marking it as dropped, and freeing
2097 * the socket if we hold the only reference.
2100 tcp_close(struct tcpcb *tp)
2102 struct inpcb *inp = tp->t_inpcb;
2105 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2106 INP_WLOCK_ASSERT(inp);
2109 if (tp->t_state == TCPS_LISTEN)
2110 tcp_offload_listen_stop(tp);
2113 * This releases the TFO pending counter resource for TFO listen
2114 * sockets as well as passively-created TFO sockets that transition
2115 * from SYN_RECEIVED to CLOSED.
2117 if (tp->t_tfo_pending) {
2118 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2119 tp->t_tfo_pending = NULL;
2122 TCPSTAT_INC(tcps_closed);
2123 if (tp->t_state != TCPS_CLOSED)
2124 tcp_state_change(tp, TCPS_CLOSED);
2125 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2126 so = inp->inp_socket;
2127 soisdisconnected(so);
2128 if (inp->inp_flags & INP_SOCKREF) {
2129 KASSERT(so->so_state & SS_PROTOREF,
2130 ("tcp_close: !SS_PROTOREF"));
2131 inp->inp_flags &= ~INP_SOCKREF;
2134 so->so_state &= ~SS_PROTOREF;
2144 VNET_ITERATOR_DECL(vnet_iter);
2149 VNET_LIST_RLOCK_NOSLEEP();
2150 VNET_FOREACH(vnet_iter) {
2151 CURVNET_SET(vnet_iter);
2156 * Walk the tcpbs, if existing, and flush the reassembly queue,
2157 * if there is one...
2158 * XXX: The "Net/3" implementation doesn't imply that the TCP
2159 * reassembly queue should be flushed, but in a situation
2160 * where we're really low on mbufs, this is potentially
2163 INP_INFO_WLOCK(&V_tcbinfo);
2164 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2166 if (inpb->inp_flags & INP_TIMEWAIT) {
2170 if ((tcpb = intotcpcb(inpb)) != NULL) {
2171 tcp_reass_flush(tcpb);
2172 tcp_clean_sackreport(tcpb);
2174 tcp_log_drain(tcpb);
2177 if (tcp_pcap_aggressive_free) {
2178 /* Free the TCP PCAP queues. */
2179 tcp_pcap_drain(&(tcpb->t_inpkts));
2180 tcp_pcap_drain(&(tcpb->t_outpkts));
2186 INP_INFO_WUNLOCK(&V_tcbinfo);
2189 VNET_LIST_RUNLOCK_NOSLEEP();
2193 * Notify a tcp user of an asynchronous error;
2194 * store error as soft error, but wake up user
2195 * (for now, won't do anything until can select for soft error).
2197 * Do not wake up user since there currently is no mechanism for
2198 * reporting soft errors (yet - a kqueue filter may be added).
2200 static struct inpcb *
2201 tcp_notify(struct inpcb *inp, int error)
2205 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2206 INP_WLOCK_ASSERT(inp);
2208 if ((inp->inp_flags & INP_TIMEWAIT) ||
2209 (inp->inp_flags & INP_DROPPED))
2212 tp = intotcpcb(inp);
2213 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2216 * Ignore some errors if we are hooked up.
2217 * If connection hasn't completed, has retransmitted several times,
2218 * and receives a second error, give up now. This is better
2219 * than waiting a long time to establish a connection that
2220 * can never complete.
2222 if (tp->t_state == TCPS_ESTABLISHED &&
2223 (error == EHOSTUNREACH || error == ENETUNREACH ||
2224 error == EHOSTDOWN)) {
2225 if (inp->inp_route.ro_nh) {
2226 NH_FREE(inp->inp_route.ro_nh);
2227 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2230 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2232 tp = tcp_drop(tp, error);
2238 tp->t_softerror = error;
2242 wakeup( &so->so_timeo);
2249 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2251 struct epoch_tracker et;
2256 if (req->newptr != NULL)
2259 if (req->oldptr == NULL) {
2262 n = V_tcbinfo.ipi_count +
2263 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2264 n += imax(n / 8, 10);
2265 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2269 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2272 bzero(&xig, sizeof(xig));
2273 xig.xig_len = sizeof xig;
2274 xig.xig_count = V_tcbinfo.ipi_count +
2275 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2276 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2277 xig.xig_sogen = so_gencnt;
2278 error = SYSCTL_OUT(req, &xig, sizeof xig);
2282 error = syncache_pcblist(req);
2286 NET_EPOCH_ENTER(et);
2287 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2289 inp = CK_LIST_NEXT(inp, inp_list)) {
2291 if (inp->inp_gencnt <= xig.xig_gen) {
2295 * XXX: This use of cr_cansee(), introduced with
2296 * TCP state changes, is not quite right, but for
2297 * now, better than nothing.
2299 if (inp->inp_flags & INP_TIMEWAIT) {
2300 if (intotw(inp) != NULL)
2301 crerr = cr_cansee(req->td->td_ucred,
2302 intotw(inp)->tw_cred);
2304 crerr = EINVAL; /* Skip this inp. */
2306 crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2310 tcp_inptoxtp(inp, &xt);
2312 error = SYSCTL_OUT(req, &xt, sizeof xt);
2325 * Give the user an updated idea of our state.
2326 * If the generation differs from what we told
2327 * her before, she knows that something happened
2328 * while we were processing this request, and it
2329 * might be necessary to retry.
2331 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2332 xig.xig_sogen = so_gencnt;
2333 xig.xig_count = V_tcbinfo.ipi_count +
2334 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2335 error = SYSCTL_OUT(req, &xig, sizeof xig);
2341 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2342 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2343 NULL, 0, tcp_pcblist, "S,xtcpcb",
2344 "List of active TCP connections");
2348 tcp_getcred(SYSCTL_HANDLER_ARGS)
2351 struct sockaddr_in addrs[2];
2352 struct epoch_tracker et;
2356 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2359 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2362 NET_EPOCH_ENTER(et);
2363 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2364 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2367 if (inp->inp_socket == NULL)
2370 error = cr_canseeinpcb(req->td->td_ucred, inp);
2372 cru2x(inp->inp_cred, &xuc);
2377 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2381 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2382 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2383 0, 0, tcp_getcred, "S,xucred",
2384 "Get the xucred of a TCP connection");
2389 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2391 struct epoch_tracker et;
2393 struct sockaddr_in6 addrs[2];
2400 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2403 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2406 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2407 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2410 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2412 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2419 NET_EPOCH_ENTER(et);
2422 inp = in_pcblookup(&V_tcbinfo,
2423 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2425 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2426 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2429 inp = in6_pcblookup(&V_tcbinfo,
2430 &addrs[1].sin6_addr, addrs[1].sin6_port,
2431 &addrs[0].sin6_addr, addrs[0].sin6_port,
2432 INPLOOKUP_RLOCKPCB, NULL);
2435 if (inp->inp_socket == NULL)
2438 error = cr_canseeinpcb(req->td->td_ucred, inp);
2440 cru2x(inp->inp_cred, &xuc);
2445 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2449 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2450 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2451 0, 0, tcp6_getcred, "S,xucred",
2452 "Get the xucred of a TCP6 connection");
2457 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2459 struct ip *ip = vip;
2461 struct in_addr faddr;
2464 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2466 struct in_conninfo inc;
2467 tcp_seq icmp_tcp_seq;
2470 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2471 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2474 if (cmd == PRC_MSGSIZE)
2475 notify = tcp_mtudisc_notify;
2476 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2477 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2478 cmd == PRC_TIMXCEED_INTRANS) && ip)
2479 notify = tcp_drop_syn_sent;
2482 * Hostdead is ugly because it goes linearly through all PCBs.
2483 * XXX: We never get this from ICMP, otherwise it makes an
2484 * excellent DoS attack on machines with many connections.
2486 else if (cmd == PRC_HOSTDEAD)
2488 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2492 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2496 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2497 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2498 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2499 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2500 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2501 /* signal EHOSTDOWN, as it flushes the cached route */
2502 inp = (*notify)(inp, EHOSTDOWN);
2505 icmp_tcp_seq = th->th_seq;
2507 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2508 !(inp->inp_flags & INP_DROPPED) &&
2509 !(inp->inp_socket == NULL)) {
2510 tp = intotcpcb(inp);
2511 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2512 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2513 if (cmd == PRC_MSGSIZE) {
2516 * If we got a needfrag set the MTU
2517 * in the route to the suggested new
2518 * value (if given) and then notify.
2520 mtu = ntohs(icp->icmp_nextmtu);
2522 * If no alternative MTU was
2523 * proposed, try the next smaller
2528 ntohs(ip->ip_len), 1);
2529 if (mtu < V_tcp_minmss +
2530 sizeof(struct tcpiphdr))
2531 mtu = V_tcp_minmss +
2532 sizeof(struct tcpiphdr);
2534 * Only process the offered MTU if it
2535 * is smaller than the current one.
2537 if (mtu < tp->t_maxseg +
2538 sizeof(struct tcpiphdr)) {
2539 bzero(&inc, sizeof(inc));
2540 inc.inc_faddr = faddr;
2542 inp->inp_inc.inc_fibnum;
2543 tcp_hc_updatemtu(&inc, mtu);
2544 tcp_mtudisc(inp, mtu);
2547 inp = (*notify)(inp,
2548 inetctlerrmap[cmd]);
2552 bzero(&inc, sizeof(inc));
2553 inc.inc_fport = th->th_dport;
2554 inc.inc_lport = th->th_sport;
2555 inc.inc_faddr = faddr;
2556 inc.inc_laddr = ip->ip_src;
2557 syncache_unreach(&inc, icmp_tcp_seq);
2567 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2569 struct in6_addr *dst;
2570 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2571 struct ip6_hdr *ip6;
2575 struct icmp6_hdr *icmp6;
2576 struct ip6ctlparam *ip6cp = NULL;
2577 const struct sockaddr_in6 *sa6_src = NULL;
2578 struct in_conninfo inc;
2583 tcp_seq icmp_tcp_seq;
2587 if (sa->sa_family != AF_INET6 ||
2588 sa->sa_len != sizeof(struct sockaddr_in6))
2591 /* if the parameter is from icmp6, decode it. */
2593 ip6cp = (struct ip6ctlparam *)d;
2594 icmp6 = ip6cp->ip6c_icmp6;
2596 ip6 = ip6cp->ip6c_ip6;
2597 off = ip6cp->ip6c_off;
2598 sa6_src = ip6cp->ip6c_src;
2599 dst = ip6cp->ip6c_finaldst;
2603 off = 0; /* fool gcc */
2608 if (cmd == PRC_MSGSIZE)
2609 notify = tcp_mtudisc_notify;
2610 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2611 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2612 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2613 notify = tcp_drop_syn_sent;
2616 * Hostdead is ugly because it goes linearly through all PCBs.
2617 * XXX: We never get this from ICMP, otherwise it makes an
2618 * excellent DoS attack on machines with many connections.
2620 else if (cmd == PRC_HOSTDEAD)
2622 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2626 in6_pcbnotify(&V_tcbinfo, sa, 0,
2627 (const struct sockaddr *)sa6_src,
2628 0, cmd, NULL, notify);
2632 /* Check if we can safely get the ports from the tcp hdr */
2635 (int32_t) (off + sizeof(struct tcp_ports)))) {
2638 bzero(&t_ports, sizeof(struct tcp_ports));
2639 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2640 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2641 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2642 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2643 /* signal EHOSTDOWN, as it flushes the cached route */
2644 inp = (*notify)(inp, EHOSTDOWN);
2647 off += sizeof(struct tcp_ports);
2648 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2651 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2653 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2654 !(inp->inp_flags & INP_DROPPED) &&
2655 !(inp->inp_socket == NULL)) {
2656 tp = intotcpcb(inp);
2657 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2658 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2659 if (cmd == PRC_MSGSIZE) {
2662 * If we got a needfrag set the MTU
2663 * in the route to the suggested new
2664 * value (if given) and then notify.
2666 mtu = ntohl(icmp6->icmp6_mtu);
2668 * If no alternative MTU was
2669 * proposed, or the proposed
2670 * MTU was too small, set to
2673 if (mtu < IPV6_MMTU)
2674 mtu = IPV6_MMTU - 8;
2675 bzero(&inc, sizeof(inc));
2676 inc.inc_fibnum = M_GETFIB(m);
2677 inc.inc_flags |= INC_ISIPV6;
2678 inc.inc6_faddr = *dst;
2679 if (in6_setscope(&inc.inc6_faddr,
2680 m->m_pkthdr.rcvif, NULL))
2683 * Only process the offered MTU if it
2684 * is smaller than the current one.
2686 if (mtu < tp->t_maxseg +
2687 sizeof (struct tcphdr) +
2688 sizeof (struct ip6_hdr)) {
2689 tcp_hc_updatemtu(&inc, mtu);
2690 tcp_mtudisc(inp, mtu);
2691 ICMP6STAT_INC(icp6s_pmtuchg);
2694 inp = (*notify)(inp,
2695 inet6ctlerrmap[cmd]);
2699 bzero(&inc, sizeof(inc));
2700 inc.inc_fibnum = M_GETFIB(m);
2701 inc.inc_flags |= INC_ISIPV6;
2702 inc.inc_fport = t_ports.th_dport;
2703 inc.inc_lport = t_ports.th_sport;
2704 inc.inc6_faddr = *dst;
2705 inc.inc6_laddr = ip6->ip6_src;
2706 syncache_unreach(&inc, icmp_tcp_seq);
2715 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2720 KASSERT(len >= SIPHASH_KEY_LENGTH,
2721 ("%s: keylen %u too short ", __func__, len));
2722 SipHash24_Init(&ctx);
2723 SipHash_SetKey(&ctx, (uint8_t *)key);
2724 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2725 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2726 switch (inc->inc_flags & INC_ISIPV6) {
2729 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2730 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2735 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2736 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2740 SipHash_Final((uint8_t *)hash, &ctx);
2742 return (hash[0] ^ hash[1]);
2746 tcp_new_ts_offset(struct in_conninfo *inc)
2748 struct in_conninfo inc_store, *local_inc;
2750 if (!V_tcp_ts_offset_per_conn) {
2751 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2752 inc_store.inc_lport = 0;
2753 inc_store.inc_fport = 0;
2754 local_inc = &inc_store;
2758 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2759 sizeof(V_ts_offset_secret)));
2763 * Following is where TCP initial sequence number generation occurs.
2765 * There are two places where we must use initial sequence numbers:
2766 * 1. In SYN-ACK packets.
2767 * 2. In SYN packets.
2769 * All ISNs for SYN-ACK packets are generated by the syncache. See
2770 * tcp_syncache.c for details.
2772 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2773 * depends on this property. In addition, these ISNs should be
2774 * unguessable so as to prevent connection hijacking. To satisfy
2775 * the requirements of this situation, the algorithm outlined in
2776 * RFC 1948 is used, with only small modifications.
2778 * Implementation details:
2780 * Time is based off the system timer, and is corrected so that it
2781 * increases by one megabyte per second. This allows for proper
2782 * recycling on high speed LANs while still leaving over an hour
2785 * As reading the *exact* system time is too expensive to be done
2786 * whenever setting up a TCP connection, we increment the time
2787 * offset in two ways. First, a small random positive increment
2788 * is added to isn_offset for each connection that is set up.
2789 * Second, the function tcp_isn_tick fires once per clock tick
2790 * and increments isn_offset as necessary so that sequence numbers
2791 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2792 * random positive increments serve only to ensure that the same
2793 * exact sequence number is never sent out twice (as could otherwise
2794 * happen when a port is recycled in less than the system tick
2797 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2798 * between seeding of isn_secret. This is normally set to zero,
2799 * as reseeding should not be necessary.
2801 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2802 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2803 * general, this means holding an exclusive (write) lock.
2806 #define ISN_BYTES_PER_SECOND 1048576
2807 #define ISN_STATIC_INCREMENT 4096
2808 #define ISN_RANDOM_INCREMENT (4096 - 1)
2809 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
2811 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2812 VNET_DEFINE_STATIC(int, isn_last);
2813 VNET_DEFINE_STATIC(int, isn_last_reseed);
2814 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2815 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2817 #define V_isn_secret VNET(isn_secret)
2818 #define V_isn_last VNET(isn_last)
2819 #define V_isn_last_reseed VNET(isn_last_reseed)
2820 #define V_isn_offset VNET(isn_offset)
2821 #define V_isn_offset_old VNET(isn_offset_old)
2824 tcp_new_isn(struct in_conninfo *inc)
2827 u_int32_t projected_offset;
2830 /* Seed if this is the first use, reseed if requested. */
2831 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2832 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2834 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2835 V_isn_last_reseed = ticks;
2838 /* Compute the hash and return the ISN. */
2839 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2840 sizeof(V_isn_secret));
2841 V_isn_offset += ISN_STATIC_INCREMENT +
2842 (arc4random() & ISN_RANDOM_INCREMENT);
2843 if (ticks != V_isn_last) {
2844 projected_offset = V_isn_offset_old +
2845 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2846 if (SEQ_GT(projected_offset, V_isn_offset))
2847 V_isn_offset = projected_offset;
2848 V_isn_offset_old = V_isn_offset;
2851 new_isn += V_isn_offset;
2857 * When a specific ICMP unreachable message is received and the
2858 * connection state is SYN-SENT, drop the connection. This behavior
2859 * is controlled by the icmp_may_rst sysctl.
2862 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2867 INP_WLOCK_ASSERT(inp);
2869 if ((inp->inp_flags & INP_TIMEWAIT) ||
2870 (inp->inp_flags & INP_DROPPED))
2873 tp = intotcpcb(inp);
2874 if (tp->t_state != TCPS_SYN_SENT)
2877 if (IS_FASTOPEN(tp->t_flags))
2878 tcp_fastopen_disable_path(tp);
2880 tp = tcp_drop(tp, errno);
2888 * When `need fragmentation' ICMP is received, update our idea of the MSS
2889 * based on the new value. Also nudge TCP to send something, since we
2890 * know the packet we just sent was dropped.
2891 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2893 static struct inpcb *
2894 tcp_mtudisc_notify(struct inpcb *inp, int error)
2897 tcp_mtudisc(inp, -1);
2902 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2907 INP_WLOCK_ASSERT(inp);
2908 if ((inp->inp_flags & INP_TIMEWAIT) ||
2909 (inp->inp_flags & INP_DROPPED))
2912 tp = intotcpcb(inp);
2913 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2915 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2917 so = inp->inp_socket;
2918 SOCKBUF_LOCK(&so->so_snd);
2919 /* If the mss is larger than the socket buffer, decrease the mss. */
2920 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2921 tp->t_maxseg = so->so_snd.sb_hiwat;
2922 SOCKBUF_UNLOCK(&so->so_snd);
2924 TCPSTAT_INC(tcps_mturesent);
2926 tp->snd_nxt = tp->snd_una;
2927 tcp_free_sackholes(tp);
2928 tp->snd_recover = tp->snd_max;
2929 if (tp->t_flags & TF_SACK_PERMIT)
2930 EXIT_FASTRECOVERY(tp->t_flags);
2931 tp->t_fb->tfb_tcp_output(tp);
2936 * Look-up the routing entry to the peer of this inpcb. If no route
2937 * is found and it cannot be allocated, then return 0. This routine
2938 * is called by TCP routines that access the rmx structure and by
2939 * tcp_mss_update to get the peer/interface MTU.
2942 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2944 struct nhop_object *nh;
2946 uint32_t maxmtu = 0;
2948 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2950 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2951 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
2956 maxmtu = nh->nh_mtu;
2958 /* Report additional interface capabilities. */
2960 if (ifp->if_capenable & IFCAP_TSO4 &&
2961 ifp->if_hwassist & CSUM_TSO) {
2962 cap->ifcap |= CSUM_TSO;
2963 cap->tsomax = ifp->if_hw_tsomax;
2964 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2965 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2975 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2977 struct nhop_object *nh;
2978 struct in6_addr dst6;
2981 uint32_t maxmtu = 0;
2983 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2985 if (inc->inc_flags & INC_IPV6MINMTU)
2988 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2989 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2990 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
2995 maxmtu = nh->nh_mtu;
2997 /* Report additional interface capabilities. */
2999 if (ifp->if_capenable & IFCAP_TSO6 &&
3000 ifp->if_hwassist & CSUM_TSO) {
3001 cap->ifcap |= CSUM_TSO;
3002 cap->tsomax = ifp->if_hw_tsomax;
3003 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3004 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3014 * Calculate effective SMSS per RFC5681 definition for a given TCP
3015 * connection at its current state, taking into account SACK and etc.
3018 tcp_maxseg(const struct tcpcb *tp)
3022 if (tp->t_flags & TF_NOOPT)
3023 return (tp->t_maxseg);
3026 * Here we have a simplified code from tcp_addoptions(),
3027 * without a proper loop, and having most of paddings hardcoded.
3028 * We might make mistakes with padding here in some edge cases,
3029 * but this is harmless, since result of tcp_maxseg() is used
3030 * only in cwnd and ssthresh estimations.
3032 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3033 if (tp->t_flags & TF_RCVD_TSTMP)
3034 optlen = TCPOLEN_TSTAMP_APPA;
3037 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3038 if (tp->t_flags & TF_SIGNATURE)
3039 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3041 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3042 optlen += TCPOLEN_SACKHDR;
3043 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3044 optlen = PADTCPOLEN(optlen);
3047 if (tp->t_flags & TF_REQ_TSTMP)
3048 optlen = TCPOLEN_TSTAMP_APPA;
3050 optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3051 if (tp->t_flags & TF_REQ_SCALE)
3052 optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3053 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3054 if (tp->t_flags & TF_SIGNATURE)
3055 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3057 if (tp->t_flags & TF_SACK_PERMIT)
3058 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3061 optlen = min(optlen, TCP_MAXOLEN);
3062 return (tp->t_maxseg - optlen);
3066 sysctl_drop(SYSCTL_HANDLER_ARGS)
3068 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3069 struct sockaddr_storage addrs[2];
3073 struct sockaddr_in *fin, *lin;
3074 struct epoch_tracker et;
3076 struct sockaddr_in6 *fin6, *lin6;
3087 if (req->oldptr != NULL || req->oldlen != 0)
3089 if (req->newptr == NULL)
3091 if (req->newlen < sizeof(addrs))
3093 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3097 switch (addrs[0].ss_family) {
3100 fin6 = (struct sockaddr_in6 *)&addrs[0];
3101 lin6 = (struct sockaddr_in6 *)&addrs[1];
3102 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3103 lin6->sin6_len != sizeof(struct sockaddr_in6))
3105 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3106 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3108 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3109 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3110 fin = (struct sockaddr_in *)&addrs[0];
3111 lin = (struct sockaddr_in *)&addrs[1];
3114 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3117 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3124 fin = (struct sockaddr_in *)&addrs[0];
3125 lin = (struct sockaddr_in *)&addrs[1];
3126 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3127 lin->sin_len != sizeof(struct sockaddr_in))
3134 NET_EPOCH_ENTER(et);
3135 switch (addrs[0].ss_family) {
3138 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3139 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3140 INPLOOKUP_WLOCKPCB, NULL);
3145 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3146 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3151 if (inp->inp_flags & INP_TIMEWAIT) {
3153 * XXXRW: There currently exists a state where an
3154 * inpcb is present, but its timewait state has been
3155 * discarded. For now, don't allow dropping of this
3163 } else if (!(inp->inp_flags & INP_DROPPED) &&
3164 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3165 tp = intotcpcb(inp);
3166 tp = tcp_drop(tp, ECONNABORTED);
3177 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3178 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3179 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3180 "Drop TCP connection");
3184 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3186 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3187 struct sockaddr_storage addrs[2];
3189 struct sockaddr_in *fin, *lin;
3190 struct epoch_tracker et;
3192 struct sockaddr_in6 *fin6, *lin6;
3203 if (req->oldptr != NULL || req->oldlen != 0)
3205 if (req->newptr == NULL)
3207 if (req->newlen < sizeof(addrs))
3209 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3213 switch (addrs[0].ss_family) {
3216 fin6 = (struct sockaddr_in6 *)&addrs[0];
3217 lin6 = (struct sockaddr_in6 *)&addrs[1];
3218 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3219 lin6->sin6_len != sizeof(struct sockaddr_in6))
3221 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3222 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3224 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3225 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3226 fin = (struct sockaddr_in *)&addrs[0];
3227 lin = (struct sockaddr_in *)&addrs[1];
3230 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3233 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3240 fin = (struct sockaddr_in *)&addrs[0];
3241 lin = (struct sockaddr_in *)&addrs[1];
3242 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3243 lin->sin_len != sizeof(struct sockaddr_in))
3250 NET_EPOCH_ENTER(et);
3251 switch (addrs[0].ss_family) {
3254 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3255 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3256 INPLOOKUP_WLOCKPCB, NULL);
3261 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3262 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3268 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3269 inp->inp_socket == NULL) {
3275 so = inp->inp_socket;
3277 error = ktls_set_tx_mode(so,
3278 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3288 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3289 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3290 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3291 "Switch TCP connection to SW TLS");
3292 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3293 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3294 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3295 "Switch TCP connection to ifnet TLS");
3299 * Generate a standardized TCP log line for use throughout the
3300 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3301 * allow use in the interrupt context.
3303 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3304 * NB: The function may return NULL if memory allocation failed.
3306 * Due to header inclusion and ordering limitations the struct ip
3307 * and ip6_hdr pointers have to be passed as void pointers.
3310 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3314 /* Is logging enabled? */
3315 if (V_tcp_log_in_vain == 0)
3318 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3322 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3326 /* Is logging enabled? */
3327 if (tcp_log_debug == 0)
3330 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3334 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3341 const struct ip6_hdr *ip6;
3343 ip6 = (const struct ip6_hdr *)ip6hdr;
3345 ip = (struct ip *)ip4hdr;
3348 * The log line looks like this:
3349 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3351 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3352 sizeof(PRINT_TH_FLAGS) + 1 +
3354 2 * INET6_ADDRSTRLEN;
3356 2 * INET_ADDRSTRLEN;
3359 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3363 strcat(s, "TCP: [");
3366 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3367 inet_ntoa_r(inc->inc_faddr, sp);
3369 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3371 inet_ntoa_r(inc->inc_laddr, sp);
3373 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3376 ip6_sprintf(sp, &inc->inc6_faddr);
3378 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3380 ip6_sprintf(sp, &inc->inc6_laddr);
3382 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3383 } else if (ip6 && th) {
3384 ip6_sprintf(sp, &ip6->ip6_src);
3386 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3388 ip6_sprintf(sp, &ip6->ip6_dst);
3390 sprintf(sp, "]:%i", ntohs(th->th_dport));
3393 } else if (ip && th) {
3394 inet_ntoa_r(ip->ip_src, sp);
3396 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3398 inet_ntoa_r(ip->ip_dst, sp);
3400 sprintf(sp, "]:%i", ntohs(th->th_dport));
3408 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3409 if (*(s + size - 1) != '\0')
3410 panic("%s: string too long", __func__);
3415 * A subroutine which makes it easy to track TCP state changes with DTrace.
3416 * This function shouldn't be called for t_state initializations that don't
3417 * correspond to actual TCP state transitions.
3420 tcp_state_change(struct tcpcb *tp, int newstate)
3422 #if defined(KDTRACE_HOOKS)
3423 int pstate = tp->t_state;
3426 TCPSTATES_DEC(tp->t_state);
3427 TCPSTATES_INC(newstate);
3428 tp->t_state = newstate;
3429 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3433 * Create an external-format (``xtcpcb'') structure using the information in
3434 * the kernel-format tcpcb structure pointed to by tp. This is done to
3435 * reduce the spew of irrelevant information over this interface, to isolate
3436 * user code from changes in the kernel structure, and potentially to provide
3437 * information-hiding if we decide that some of this information should be
3438 * hidden from users.
3441 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3443 struct tcpcb *tp = intotcpcb(inp);
3446 bzero(xt, sizeof(*xt));
3447 if (inp->inp_flags & INP_TIMEWAIT) {
3448 xt->t_state = TCPS_TIME_WAIT;
3450 xt->t_state = tp->t_state;
3451 xt->t_logstate = tp->t_logstate;
3452 xt->t_flags = tp->t_flags;
3453 xt->t_sndzerowin = tp->t_sndzerowin;
3454 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3455 xt->t_rcvoopack = tp->t_rcvoopack;
3456 xt->t_rcv_wnd = tp->rcv_wnd;
3457 xt->t_snd_wnd = tp->snd_wnd;
3458 xt->t_snd_cwnd = tp->snd_cwnd;
3459 xt->t_snd_ssthresh = tp->snd_ssthresh;
3460 xt->t_maxseg = tp->t_maxseg;
3461 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3462 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3464 now = getsbinuptime();
3465 #define COPYTIMER(ttt) do { \
3466 if (callout_active(&tp->t_timers->ttt)) \
3467 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3472 COPYTIMER(tt_delack);
3473 COPYTIMER(tt_rexmt);
3474 COPYTIMER(tt_persist);
3478 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3480 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3481 TCP_FUNCTION_NAME_LEN_MAX);
3482 bcopy(CC_ALGO(tp)->name, xt->xt_cc,
3485 (void)tcp_log_get_id(tp, xt->xt_logid);
3489 xt->xt_len = sizeof(struct xtcpcb);
3490 in_pcbtoxinpcb(inp, &xt->xt_inp);
3491 if (inp->inp_socket == NULL)
3492 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3496 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3501 (status > TCP_EI_STATUS_MAX_VALUE) ||
3506 if (status > (sizeof(uint32_t) * 8)) {
3507 /* Should this be a KASSERT? */
3510 bit = 1U << (status - 1);
3511 if (bit & tp->t_end_info_status) {
3512 /* already logged */
3515 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3516 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3517 tp->t_end_info_bytes[i] = status;
3518 tp->t_end_info_status |= bit;