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_ts_offset_per_conn) = 1;
260 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
261 &VNET_NAME(tcp_ts_offset_per_conn), 0,
262 "Initialize TCP timestamps per connection instead of per host pair");
264 static int tcp_log_debug = 0;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
266 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
268 static int tcp_tcbhashsize;
269 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
270 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
272 static int do_tcpdrain = 1;
273 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
274 "Enable tcp_drain routine for extra help when low on mbufs");
276 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
277 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
279 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
280 #define V_icmp_may_rst VNET(icmp_may_rst)
281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
282 &VNET_NAME(icmp_may_rst), 0,
283 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
285 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
286 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
288 &VNET_NAME(tcp_isn_reseed_interval), 0,
289 "Seconds between reseeding of ISN secret");
291 static int tcp_soreceive_stream;
292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
293 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
295 VNET_DEFINE(uma_zone_t, sack_hole_zone);
296 #define V_sack_hole_zone VNET(sack_hole_zone)
297 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */
299 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
304 new = V_tcp_map_entries_limit;
305 error = sysctl_handle_int(oidp, &new, 0, req);
306 if (error == 0 && req->newptr) {
307 /* only allow "0" and value > minimum */
308 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
311 V_tcp_map_entries_limit = new;
315 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
316 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
317 &VNET_NAME(tcp_map_entries_limit), 0,
318 &sysctl_net_inet_tcp_map_limit_check, "IU",
319 "Total sendmap entries limit");
321 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
322 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
323 &VNET_NAME(tcp_map_split_limit), 0,
324 "Total sendmap split entries limit");
327 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
330 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
331 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
332 #define V_ts_offset_secret VNET(ts_offset_secret)
334 static int tcp_default_fb_init(struct tcpcb *tp);
335 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
336 static int tcp_default_handoff_ok(struct tcpcb *tp);
337 static struct inpcb *tcp_notify(struct inpcb *, int);
338 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
339 static void tcp_mtudisc(struct inpcb *, int);
340 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
341 void *ip4hdr, const void *ip6hdr);
344 static struct tcp_function_block tcp_def_funcblk = {
345 .tfb_tcp_block_name = "freebsd",
346 .tfb_tcp_output = tcp_output,
347 .tfb_tcp_do_segment = tcp_do_segment,
348 .tfb_tcp_ctloutput = tcp_default_ctloutput,
349 .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
350 .tfb_tcp_fb_init = tcp_default_fb_init,
351 .tfb_tcp_fb_fini = tcp_default_fb_fini,
354 static int tcp_fb_cnt = 0;
355 struct tcp_funchead t_functions;
356 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
358 static struct tcp_function_block *
359 find_tcp_functions_locked(struct tcp_function_set *fs)
361 struct tcp_function *f;
362 struct tcp_function_block *blk=NULL;
364 TAILQ_FOREACH(f, &t_functions, tf_next) {
365 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
373 static struct tcp_function_block *
374 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
376 struct tcp_function_block *rblk=NULL;
377 struct tcp_function *f;
379 TAILQ_FOREACH(f, &t_functions, tf_next) {
380 if (f->tf_fb == blk) {
391 struct tcp_function_block *
392 find_and_ref_tcp_functions(struct tcp_function_set *fs)
394 struct tcp_function_block *blk;
396 rw_rlock(&tcp_function_lock);
397 blk = find_tcp_functions_locked(fs);
399 refcount_acquire(&blk->tfb_refcnt);
400 rw_runlock(&tcp_function_lock);
404 struct tcp_function_block *
405 find_and_ref_tcp_fb(struct tcp_function_block *blk)
407 struct tcp_function_block *rblk;
409 rw_rlock(&tcp_function_lock);
410 rblk = find_tcp_fb_locked(blk, NULL);
412 refcount_acquire(&rblk->tfb_refcnt);
413 rw_runlock(&tcp_function_lock);
417 static struct tcp_function_block *
418 find_and_ref_tcp_default_fb(void)
420 struct tcp_function_block *rblk;
422 rw_rlock(&tcp_function_lock);
423 rblk = tcp_func_set_ptr;
424 refcount_acquire(&rblk->tfb_refcnt);
425 rw_runlock(&tcp_function_lock);
430 tcp_switch_back_to_default(struct tcpcb *tp)
432 struct tcp_function_block *tfb;
434 KASSERT(tp->t_fb != &tcp_def_funcblk,
435 ("%s: called by the built-in default stack", __func__));
438 * Release the old stack. This function will either find a new one
441 if (tp->t_fb->tfb_tcp_fb_fini != NULL)
442 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
443 refcount_release(&tp->t_fb->tfb_refcnt);
446 * Now, we'll find a new function block to use.
447 * Start by trying the current user-selected
448 * default, unless this stack is the user-selected
451 tfb = find_and_ref_tcp_default_fb();
452 if (tfb == tp->t_fb) {
453 refcount_release(&tfb->tfb_refcnt);
456 /* Does the stack accept this connection? */
457 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
458 (*tfb->tfb_tcp_handoff_ok)(tp)) {
459 refcount_release(&tfb->tfb_refcnt);
462 /* Try to use that stack. */
464 /* Initialize the new stack. If it succeeds, we are done. */
466 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
467 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
471 * Initialization failed. Release the reference count on
474 refcount_release(&tfb->tfb_refcnt);
478 * If that wasn't feasible, use the built-in default
479 * stack which is not allowed to reject anyone.
481 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
483 /* there always should be a default */
484 panic("Can't refer to tcp_def_funcblk");
486 if (tfb->tfb_tcp_handoff_ok != NULL) {
487 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
488 /* The default stack cannot say no */
489 panic("Default stack rejects a new session?");
493 if (tp->t_fb->tfb_tcp_fb_init != NULL &&
494 (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
495 /* The default stack cannot fail */
496 panic("Default stack initialization failed");
501 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
504 struct tcp_function_set fs;
505 struct tcp_function_block *blk;
507 memset(&fs, 0, sizeof(fs));
508 rw_rlock(&tcp_function_lock);
509 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
512 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
513 fs.pcbcnt = blk->tfb_refcnt;
515 rw_runlock(&tcp_function_lock);
516 error = sysctl_handle_string(oidp, fs.function_set_name,
517 sizeof(fs.function_set_name), req);
519 /* Check for error or no change */
520 if (error != 0 || req->newptr == NULL)
523 rw_wlock(&tcp_function_lock);
524 blk = find_tcp_functions_locked(&fs);
526 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
530 tcp_func_set_ptr = blk;
532 rw_wunlock(&tcp_function_lock);
536 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
537 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
538 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
539 "Set/get the default TCP functions");
542 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
544 int error, cnt, linesz;
545 struct tcp_function *f;
551 rw_rlock(&tcp_function_lock);
552 TAILQ_FOREACH(f, &t_functions, tf_next) {
555 rw_runlock(&tcp_function_lock);
557 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
558 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
563 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
564 "Alias", "PCB count");
569 rw_rlock(&tcp_function_lock);
570 TAILQ_FOREACH(f, &t_functions, tf_next) {
571 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
572 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
573 f->tf_fb->tfb_tcp_block_name,
574 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
575 alias ? f->tf_name : "-",
576 f->tf_fb->tfb_refcnt);
577 if (linesz >= bufsz) {
585 rw_runlock(&tcp_function_lock);
587 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
588 free(buffer, M_TEMP);
592 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
593 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
594 NULL, 0, sysctl_net_inet_list_available, "A",
595 "list available TCP Function sets");
598 * Exports one (struct tcp_function_info) for each alias/name.
601 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
604 struct tcp_function *f;
605 struct tcp_function_info tfi;
608 * We don't allow writes.
610 if (req->newptr != NULL)
614 * Wire the old buffer so we can directly copy the functions to
615 * user space without dropping the lock.
617 if (req->oldptr != NULL) {
618 error = sysctl_wire_old_buffer(req, 0);
624 * Walk the list and copy out matching entries. If INVARIANTS
625 * is compiled in, also walk the list to verify the length of
626 * the list matches what we have recorded.
628 rw_rlock(&tcp_function_lock);
632 if (req->oldptr == NULL) {
637 TAILQ_FOREACH(f, &t_functions, tf_next) {
641 if (req->oldptr != NULL) {
642 bzero(&tfi, sizeof(tfi));
643 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
644 tfi.tfi_id = f->tf_fb->tfb_id;
645 (void)strlcpy(tfi.tfi_alias, f->tf_name,
646 sizeof(tfi.tfi_alias));
647 (void)strlcpy(tfi.tfi_name,
648 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
649 error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
651 * Don't stop on error, as that is the
652 * mechanism we use to accumulate length
653 * information if the buffer was too short.
657 KASSERT(cnt == tcp_fb_cnt,
658 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
662 rw_runlock(&tcp_function_lock);
663 if (req->oldptr == NULL)
664 error = SYSCTL_OUT(req, NULL,
665 (cnt + 1) * sizeof(struct tcp_function_info));
670 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
671 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
672 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
673 "List TCP function block name-to-ID mappings");
676 * tfb_tcp_handoff_ok() function for the default stack.
677 * Note that we'll basically try to take all comers.
680 tcp_default_handoff_ok(struct tcpcb *tp)
687 * tfb_tcp_fb_init() function for the default stack.
689 * This handles making sure we have appropriate timers set if you are
690 * transitioning a socket that has some amount of setup done.
692 * The init() fuction from the default can *never* return non-zero i.e.
693 * it is required to always succeed since it is the stack of last resort!
696 tcp_default_fb_init(struct tcpcb *tp)
701 INP_WLOCK_ASSERT(tp->t_inpcb);
703 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
704 ("%s: connection %p in unexpected state %d", __func__, tp,
708 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
709 * know what to do for unexpected states (which includes TIME_WAIT).
711 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
715 * Make sure some kind of transmission timer is set if there is
718 so = tp->t_inpcb->inp_socket;
719 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
720 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
721 tcp_timer_active(tp, TT_PERSIST))) {
723 * If the session has established and it looks like it should
724 * be in the persist state, set the persist timer. Otherwise,
725 * set the retransmit timer.
727 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
728 (int32_t)(tp->snd_nxt - tp->snd_una) <
729 (int32_t)sbavail(&so->so_snd))
732 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
735 /* All non-embryonic sessions get a keepalive timer. */
736 if (!tcp_timer_active(tp, TT_KEEP))
737 tcp_timer_activate(tp, TT_KEEP,
738 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
745 * tfb_tcp_fb_fini() function for the default stack.
747 * This changes state as necessary (or prudent) to prepare for another stack
748 * to assume responsibility for the connection.
751 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
754 INP_WLOCK_ASSERT(tp->t_inpcb);
759 * Target size of TCP PCB hash tables. Must be a power of two.
761 * Note that this can be overridden by the kernel environment
762 * variable net.inet.tcp.tcbhashsize
765 #define TCBHASHSIZE 0
770 * Callouts should be moved into struct tcp directly. They are currently
771 * separate because the tcpcb structure is exported to userland for sysctl
772 * parsing purposes, which do not know about callouts.
783 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
784 #define V_tcpcb_zone VNET(tcpcb_zone)
786 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
787 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
789 static struct mtx isn_mtx;
791 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
792 #define ISN_LOCK() mtx_lock(&isn_mtx)
793 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
796 * TCP initialization.
799 tcp_zone_change(void *tag)
802 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
803 uma_zone_set_max(V_tcpcb_zone, maxsockets);
804 tcp_tw_zone_change();
808 tcp_inpcb_init(void *mem, int size, int flags)
810 struct inpcb *inp = mem;
812 INP_LOCK_INIT(inp, "inp", "tcpinp");
817 * Take a value and get the next power of 2 that doesn't overflow.
818 * Used to size the tcp_inpcb hash buckets.
821 maketcp_hashsize(int size)
827 * get the next power of 2 higher than maxsockets.
829 hashsize = 1 << fls(size);
830 /* catch overflow, and just go one power of 2 smaller */
831 if (hashsize < size) {
832 hashsize = 1 << (fls(size) - 1);
837 static volatile int next_tcp_stack_id = 1;
840 * Register a TCP function block with the name provided in the names
841 * array. (Note that this function does NOT automatically register
842 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
843 * explicitly include blk->tfb_tcp_block_name in the list of names if
844 * you wish to register the stack with that name.)
846 * Either all name registrations will succeed or all will fail. If
847 * a name registration fails, the function will update the num_names
848 * argument to point to the array index of the name that encountered
851 * Returns 0 on success, or an error code on failure.
854 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
855 const char *names[], int *num_names)
857 struct tcp_function *n;
858 struct tcp_function_set fs;
861 KASSERT(names != NULL && *num_names > 0,
862 ("%s: Called with 0-length name list", __func__));
863 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
864 KASSERT(rw_initialized(&tcp_function_lock),
865 ("%s: called too early", __func__));
867 if ((blk->tfb_tcp_output == NULL) ||
868 (blk->tfb_tcp_do_segment == NULL) ||
869 (blk->tfb_tcp_ctloutput == NULL) ||
870 (strlen(blk->tfb_tcp_block_name) == 0)) {
872 * These functions are required and you
878 if (blk->tfb_tcp_timer_stop_all ||
879 blk->tfb_tcp_timer_activate ||
880 blk->tfb_tcp_timer_active ||
881 blk->tfb_tcp_timer_stop) {
883 * If you define one timer function you
884 * must have them all.
886 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
887 (blk->tfb_tcp_timer_activate == NULL) ||
888 (blk->tfb_tcp_timer_active == NULL) ||
889 (blk->tfb_tcp_timer_stop == NULL)) {
895 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
900 refcount_init(&blk->tfb_refcnt, 0);
901 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
902 for (i = 0; i < *num_names; i++) {
903 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
910 (void)strlcpy(fs.function_set_name, names[i],
911 sizeof(fs.function_set_name));
912 rw_wlock(&tcp_function_lock);
913 if (find_tcp_functions_locked(&fs) != NULL) {
914 /* Duplicate name space not allowed */
915 rw_wunlock(&tcp_function_lock);
916 free(n, M_TCPFUNCTIONS);
920 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
921 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
923 rw_wunlock(&tcp_function_lock);
929 * Deregister the names we just added. Because registration failed
930 * for names[i], we don't need to deregister that name.
933 rw_wlock(&tcp_function_lock);
935 TAILQ_FOREACH(n, &t_functions, tf_next) {
936 if (!strncmp(n->tf_name, names[i],
937 TCP_FUNCTION_NAME_LEN_MAX)) {
938 TAILQ_REMOVE(&t_functions, n, tf_next);
941 free(n, M_TCPFUNCTIONS);
946 rw_wunlock(&tcp_function_lock);
951 * Register a TCP function block using the name provided in the name
954 * Returns 0 on success, or an error code on failure.
957 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
960 const char *name_list[1];
967 name_list[0] = blk->tfb_tcp_block_name;
968 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
973 * Register a TCP function block using the name defined in
974 * blk->tfb_tcp_block_name.
976 * Returns 0 on success, or an error code on failure.
979 register_tcp_functions(struct tcp_function_block *blk, int wait)
982 return (register_tcp_functions_as_name(blk, NULL, wait));
986 * Deregister all names associated with a function block. This
987 * functionally removes the function block from use within the system.
989 * When called with a true quiesce argument, mark the function block
990 * as being removed so no more stacks will use it and determine
991 * whether the removal would succeed.
993 * When called with a false quiesce argument, actually attempt the
996 * When called with a force argument, attempt to switch all TCBs to
997 * use the default stack instead of returning EBUSY.
999 * Returns 0 on success (or if the removal would succeed, or an error
1003 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1006 struct tcp_function *f;
1008 if (blk == &tcp_def_funcblk) {
1009 /* You can't un-register the default */
1012 rw_wlock(&tcp_function_lock);
1013 if (blk == tcp_func_set_ptr) {
1014 /* You can't free the current default */
1015 rw_wunlock(&tcp_function_lock);
1018 /* Mark the block so no more stacks can use it. */
1019 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1021 * If TCBs are still attached to the stack, attempt to switch them
1022 * to the default stack.
1024 if (force && blk->tfb_refcnt) {
1027 VNET_ITERATOR_DECL(vnet_iter);
1029 rw_wunlock(&tcp_function_lock);
1032 VNET_FOREACH(vnet_iter) {
1033 CURVNET_SET(vnet_iter);
1034 INP_INFO_WLOCK(&V_tcbinfo);
1035 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1037 if (inp->inp_flags & INP_TIMEWAIT) {
1041 tp = intotcpcb(inp);
1042 if (tp == NULL || tp->t_fb != blk) {
1046 tcp_switch_back_to_default(tp);
1049 INP_INFO_WUNLOCK(&V_tcbinfo);
1052 VNET_LIST_RUNLOCK();
1054 rw_wlock(&tcp_function_lock);
1056 if (blk->tfb_refcnt) {
1057 /* TCBs still attached. */
1058 rw_wunlock(&tcp_function_lock);
1063 rw_wunlock(&tcp_function_lock);
1066 /* Remove any function names that map to this function block. */
1067 while (find_tcp_fb_locked(blk, &f) != NULL) {
1068 TAILQ_REMOVE(&t_functions, f, tf_next);
1071 free(f, M_TCPFUNCTIONS);
1073 rw_wunlock(&tcp_function_lock);
1080 const char *tcbhash_tuneable;
1083 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1086 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1087 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1088 printf("%s: WARNING: unable to register helper hook\n", __func__);
1089 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1090 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1091 printf("%s: WARNING: unable to register helper hook\n", __func__);
1094 if (tcp_stats_init())
1095 printf("%s: WARNING: unable to initialise TCP stats\n",
1098 hashsize = TCBHASHSIZE;
1099 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1100 if (hashsize == 0) {
1102 * Auto tune the hash size based on maxsockets.
1103 * A perfect hash would have a 1:1 mapping
1104 * (hashsize = maxsockets) however it's been
1105 * suggested that O(2) average is better.
1107 hashsize = maketcp_hashsize(maxsockets / 4);
1109 * Our historical default is 512,
1110 * do not autotune lower than this.
1114 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1115 printf("%s: %s auto tuned to %d\n", __func__,
1116 tcbhash_tuneable, hashsize);
1119 * We require a hashsize to be a power of two.
1120 * Previously if it was not a power of two we would just reset it
1121 * back to 512, which could be a nasty surprise if you did not notice
1122 * the error message.
1123 * Instead what we do is clip it to the closest power of two lower
1124 * than the specified hash value.
1126 if (!powerof2(hashsize)) {
1127 int oldhashsize = hashsize;
1129 hashsize = maketcp_hashsize(hashsize);
1130 /* prevent absurdly low value */
1133 printf("%s: WARNING: TCB hash size not a power of 2, "
1134 "clipped from %d to %d.\n", __func__, oldhashsize,
1137 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1138 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1141 * These have to be type stable for the benefit of the timers.
1143 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1144 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1145 uma_zone_set_max(V_tcpcb_zone, maxsockets);
1146 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1152 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1153 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1154 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1156 tcp_fastopen_init();
1158 /* Skip initialization of globals for non-default instances. */
1159 if (!IS_DEFAULT_VNET(curvnet))
1162 tcp_reass_global_init();
1164 /* XXX virtualize those bellow? */
1165 tcp_delacktime = TCPTV_DELACK;
1166 tcp_keepinit = TCPTV_KEEP_INIT;
1167 tcp_keepidle = TCPTV_KEEP_IDLE;
1168 tcp_keepintvl = TCPTV_KEEPINTVL;
1169 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1170 tcp_msl = TCPTV_MSL;
1171 tcp_rexmit_initial = TCPTV_RTOBASE;
1172 if (tcp_rexmit_initial < 1)
1173 tcp_rexmit_initial = 1;
1174 tcp_rexmit_min = TCPTV_MIN;
1175 if (tcp_rexmit_min < 1)
1177 tcp_persmin = TCPTV_PERSMIN;
1178 tcp_persmax = TCPTV_PERSMAX;
1179 tcp_rexmit_slop = TCPTV_CPU_VAR;
1180 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1181 tcp_tcbhashsize = hashsize;
1183 /* Setup the tcp function block list */
1184 TAILQ_INIT(&t_functions);
1185 rw_init(&tcp_function_lock, "tcp_func_lock");
1186 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1188 /* Initialize the TCP logging data. */
1191 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1193 if (tcp_soreceive_stream) {
1195 tcp_usrreqs.pru_soreceive = soreceive_stream;
1198 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1203 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1205 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1207 if (max_protohdr < TCP_MINPROTOHDR)
1208 max_protohdr = TCP_MINPROTOHDR;
1209 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1211 #undef TCP_MINPROTOHDR
1214 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1215 SHUTDOWN_PRI_DEFAULT);
1216 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1217 EVENTHANDLER_PRI_ANY);
1219 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1220 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1221 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1222 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK);
1223 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1224 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK);
1232 tcp_destroy(void *unused __unused)
1240 * All our processes are gone, all our sockets should be cleaned
1241 * up, which means, we should be past the tcp_discardcb() calls.
1242 * Sleep to let all tcpcb timers really disappear and cleanup.
1245 INP_LIST_RLOCK(&V_tcbinfo);
1246 n = V_tcbinfo.ipi_count;
1247 INP_LIST_RUNLOCK(&V_tcbinfo);
1250 pause("tcpdes", hz / 10);
1255 in_pcbinfo_destroy(&V_tcbinfo);
1256 /* tcp_discardcb() clears the sack_holes up. */
1257 uma_zdestroy(V_sack_hole_zone);
1258 uma_zdestroy(V_tcpcb_zone);
1261 * Cannot free the zone until all tcpcbs are released as we attach
1262 * the allocations to them.
1264 tcp_fastopen_destroy();
1267 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1269 printf("%s: WARNING: unable to deregister helper hook "
1270 "type=%d, id=%d: error %d returned\n", __func__,
1271 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1273 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1275 printf("%s: WARNING: unable to deregister helper hook "
1276 "type=%d, id=%d: error %d returned\n", __func__,
1277 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1281 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1291 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1292 * tcp_template used to store this data in mbufs, but we now recopy it out
1293 * of the tcpcb each time to conserve mbufs.
1296 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1298 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1300 INP_WLOCK_ASSERT(inp);
1303 if ((inp->inp_vflag & INP_IPV6) != 0) {
1304 struct ip6_hdr *ip6;
1306 ip6 = (struct ip6_hdr *)ip_ptr;
1307 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1308 (inp->inp_flow & IPV6_FLOWINFO_MASK);
1309 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1310 (IPV6_VERSION & IPV6_VERSION_MASK);
1311 ip6->ip6_nxt = IPPROTO_TCP;
1312 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1313 ip6->ip6_src = inp->in6p_laddr;
1314 ip6->ip6_dst = inp->in6p_faddr;
1317 #if defined(INET6) && defined(INET)
1324 ip = (struct ip *)ip_ptr;
1325 ip->ip_v = IPVERSION;
1327 ip->ip_tos = inp->inp_ip_tos;
1331 ip->ip_ttl = inp->inp_ip_ttl;
1333 ip->ip_p = IPPROTO_TCP;
1334 ip->ip_src = inp->inp_laddr;
1335 ip->ip_dst = inp->inp_faddr;
1338 th->th_sport = inp->inp_lport;
1339 th->th_dport = inp->inp_fport;
1347 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
1351 * Create template to be used to send tcp packets on a connection.
1352 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
1353 * use for this function is in keepalives, which use tcp_respond.
1356 tcpip_maketemplate(struct inpcb *inp)
1360 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1363 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1368 * Send a single message to the TCP at address specified by
1369 * the given TCP/IP header. If m == NULL, then we make a copy
1370 * of the tcpiphdr at th and send directly to the addressed host.
1371 * This is used to force keep alive messages out using the TCP
1372 * template for a connection. If flags are given then we send
1373 * a message back to the TCP which originated the segment th,
1374 * and discard the mbuf containing it and any other attached mbufs.
1376 * In any case the ack and sequence number of the transmitted
1377 * segment are as specified by the parameters.
1379 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1382 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1383 tcp_seq ack, tcp_seq seq, int flags)
1392 struct ip6_hdr *ip6;
1395 int optlen, tlen, win;
1398 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1402 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1409 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1410 INP_WLOCK_ASSERT(inp);
1417 if (!(flags & TH_RST)) {
1418 win = sbspace(&inp->inp_socket->so_rcv);
1419 if (win > TCP_MAXWIN << tp->rcv_scale)
1420 win = TCP_MAXWIN << tp->rcv_scale;
1422 if ((tp->t_flags & TF_NOOPT) == 0)
1426 m = m_gethdr(M_NOWAIT, MT_DATA);
1429 m->m_data += max_linkhdr;
1432 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1433 sizeof(struct ip6_hdr));
1434 ip6 = mtod(m, struct ip6_hdr *);
1435 nth = (struct tcphdr *)(ip6 + 1);
1439 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1440 ip = mtod(m, struct ip *);
1441 nth = (struct tcphdr *)(ip + 1);
1443 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1445 } else if (!M_WRITABLE(m)) {
1448 /* Can't reuse 'm', allocate a new mbuf. */
1449 n = m_gethdr(M_NOWAIT, MT_DATA);
1455 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1461 n->m_data += max_linkhdr;
1462 /* m_len is set later */
1463 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1466 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1467 sizeof(struct ip6_hdr));
1468 ip6 = mtod(n, struct ip6_hdr *);
1469 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1470 nth = (struct tcphdr *)(ip6 + 1);
1474 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1475 ip = mtod(n, struct ip *);
1476 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1477 nth = (struct tcphdr *)(ip + 1);
1479 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1480 xchg(nth->th_dport, nth->th_sport, uint16_t);
1487 * XXX MRT We inherit the FIB, which is lucky.
1491 m->m_data = (caddr_t)ipgen;
1492 /* m_len is set later */
1495 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1496 nth = (struct tcphdr *)(ip6 + 1);
1500 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1501 nth = (struct tcphdr *)(ip + 1);
1505 * this is usually a case when an extension header
1506 * exists between the IPv6 header and the
1509 nth->th_sport = th->th_sport;
1510 nth->th_dport = th->th_dport;
1512 xchg(nth->th_dport, nth->th_sport, uint16_t);
1518 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1520 #if defined(INET) && defined(INET6)
1524 tlen = sizeof (struct tcpiphdr);
1528 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1529 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1530 m, tlen, (long)M_TRAILINGSPACE(m)));
1535 /* Make sure we have room. */
1536 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1537 m->m_next = m_get(M_NOWAIT, MT_DATA);
1539 optp = mtod(m->m_next, u_char *);
1544 optp = (u_char *) (nth + 1);
1550 if (tp->t_flags & TF_RCVD_TSTMP) {
1551 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1552 to.to_tsecr = tp->ts_recent;
1553 to.to_flags |= TOF_TS;
1555 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1556 /* TCP-MD5 (RFC2385). */
1557 if (tp->t_flags & TF_SIGNATURE)
1558 to.to_flags |= TOF_SIGNATURE;
1560 /* Add the options. */
1561 tlen += optlen = tcp_addoptions(&to, optp);
1563 /* Update m_len in the correct mbuf. */
1564 optm->m_len += optlen;
1570 ip6->ip6_vfc = IPV6_VERSION;
1571 ip6->ip6_nxt = IPPROTO_TCP;
1572 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1575 #if defined(INET) && defined(INET6)
1580 ip->ip_len = htons(tlen);
1581 ip->ip_ttl = V_ip_defttl;
1582 if (V_path_mtu_discovery)
1583 ip->ip_off |= htons(IP_DF);
1586 m->m_pkthdr.len = tlen;
1587 m->m_pkthdr.rcvif = NULL;
1591 * Packet is associated with a socket, so allow the
1592 * label of the response to reflect the socket label.
1594 INP_WLOCK_ASSERT(inp);
1595 mac_inpcb_create_mbuf(inp, m);
1598 * Packet is not associated with a socket, so possibly
1599 * update the label in place.
1601 mac_netinet_tcp_reply(m);
1604 nth->th_seq = htonl(seq);
1605 nth->th_ack = htonl(ack);
1607 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1608 nth->th_flags = flags;
1610 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1612 nth->th_win = htons((u_short)win);
1615 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1616 if (to.to_flags & TOF_SIGNATURE) {
1617 if (!TCPMD5_ENABLED() ||
1618 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1625 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1628 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1629 nth->th_sum = in6_cksum_pseudo(ip6,
1630 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1631 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1635 #if defined(INET6) && defined(INET)
1640 m->m_pkthdr.csum_flags = CSUM_TCP;
1641 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1642 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1646 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1647 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1649 TCP_PROBE3(debug__output, tp, th, m);
1651 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1655 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1656 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1659 #if defined(INET) && defined(INET6)
1664 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1665 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1671 * Create a new TCP control block, making an
1672 * empty reassembly queue and hooking it to the argument
1673 * protocol control block. The `inp' parameter must have
1674 * come from the zone allocator set up in tcp_init().
1677 tcp_newtcpcb(struct inpcb *inp)
1679 struct tcpcb_mem *tm;
1682 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1685 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1690 /* Initialise cc_var struct for this tcpcb. */
1692 tp->ccv->type = IPPROTO_TCP;
1693 tp->ccv->ccvc.tcp = tp;
1694 rw_rlock(&tcp_function_lock);
1695 tp->t_fb = tcp_func_set_ptr;
1696 refcount_acquire(&tp->t_fb->tfb_refcnt);
1697 rw_runlock(&tcp_function_lock);
1699 * Use the current system default CC algorithm.
1702 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1703 CC_ALGO(tp) = CC_DEFAULT();
1706 if (CC_ALGO(tp)->cb_init != NULL)
1707 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1708 if (tp->t_fb->tfb_tcp_fb_fini)
1709 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1710 refcount_release(&tp->t_fb->tfb_refcnt);
1711 uma_zfree(V_tcpcb_zone, tm);
1717 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1718 if (tp->t_fb->tfb_tcp_fb_fini)
1719 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1720 refcount_release(&tp->t_fb->tfb_refcnt);
1721 uma_zfree(V_tcpcb_zone, tm);
1727 tp->t_vnet = inp->inp_vnet;
1729 tp->t_timers = &tm->tt;
1730 TAILQ_INIT(&tp->t_segq);
1733 isipv6 ? V_tcp_v6mssdflt :
1737 /* Set up our timeouts. */
1738 callout_init(&tp->t_timers->tt_rexmt, 1);
1739 callout_init(&tp->t_timers->tt_persist, 1);
1740 callout_init(&tp->t_timers->tt_keep, 1);
1741 callout_init(&tp->t_timers->tt_2msl, 1);
1742 callout_init(&tp->t_timers->tt_delack, 1);
1744 if (V_tcp_do_rfc1323)
1745 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1747 tp->t_flags |= TF_SACK_PERMIT;
1748 TAILQ_INIT(&tp->snd_holes);
1750 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1753 in_pcbref(inp); /* Reference for tcpcb */
1757 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1758 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1759 * reasonable initial retransmit time.
1761 tp->t_srtt = TCPTV_SRTTBASE;
1762 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1763 tp->t_rttmin = tcp_rexmit_min;
1764 tp->t_rxtcur = tcp_rexmit_initial;
1765 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1766 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1767 tp->t_rcvtime = ticks;
1769 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1770 * because the socket may be bound to an IPv6 wildcard address,
1771 * which may match an IPv4-mapped IPv6 address.
1773 inp->inp_ip_ttl = V_ip_defttl;
1777 * Init the TCP PCAP queues.
1779 tcp_pcap_tcpcb_init(tp);
1782 /* Initialize the per-TCPCB log data. */
1783 tcp_log_tcpcbinit(tp);
1785 if (tp->t_fb->tfb_tcp_fb_init) {
1786 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1789 if (V_tcp_perconn_stats_enable == 1)
1790 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
1792 return (tp); /* XXX */
1796 * Switch the congestion control algorithm back to NewReno for any active
1797 * control blocks using an algorithm which is about to go away.
1798 * This ensures the CC framework can allow the unload to proceed without leaving
1799 * any dangling pointers which would trigger a panic.
1800 * Returning non-zero would inform the CC framework that something went wrong
1801 * and it would be unsafe to allow the unload to proceed. However, there is no
1802 * way for this to occur with this implementation so we always return zero.
1805 tcp_ccalgounload(struct cc_algo *unload_algo)
1807 struct cc_algo *tmpalgo;
1810 VNET_ITERATOR_DECL(vnet_iter);
1813 * Check all active control blocks across all network stacks and change
1814 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1815 * requires cleanup code to be run, call it.
1818 VNET_FOREACH(vnet_iter) {
1819 CURVNET_SET(vnet_iter);
1820 INP_INFO_WLOCK(&V_tcbinfo);
1822 * New connections already part way through being initialised
1823 * with the CC algo we're removing will not race with this code
1824 * because the INP_INFO_WLOCK is held during initialisation. We
1825 * therefore don't enter the loop below until the connection
1826 * list has stabilised.
1828 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1830 /* Important to skip tcptw structs. */
1831 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1832 (tp = intotcpcb(inp)) != NULL) {
1834 * By holding INP_WLOCK here, we are assured
1835 * that the connection is not currently
1836 * executing inside the CC module's functions
1837 * i.e. it is safe to make the switch back to
1840 if (CC_ALGO(tp) == unload_algo) {
1841 tmpalgo = CC_ALGO(tp);
1842 if (tmpalgo->cb_destroy != NULL)
1843 tmpalgo->cb_destroy(tp->ccv);
1846 * NewReno may allocate memory on
1847 * demand for certain stateful
1848 * configuration as needed, but is
1849 * coded to never fail on memory
1850 * allocation failure so it is a safe
1853 CC_ALGO(tp) = &newreno_cc_algo;
1858 INP_INFO_WUNLOCK(&V_tcbinfo);
1861 VNET_LIST_RUNLOCK();
1867 * Drop a TCP connection, reporting
1868 * the specified error. If connection is synchronized,
1869 * then send a RST to peer.
1872 tcp_drop(struct tcpcb *tp, int errno)
1874 struct socket *so = tp->t_inpcb->inp_socket;
1877 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1878 INP_WLOCK_ASSERT(tp->t_inpcb);
1880 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1881 tcp_state_change(tp, TCPS_CLOSED);
1882 (void) tp->t_fb->tfb_tcp_output(tp);
1883 TCPSTAT_INC(tcps_drops);
1885 TCPSTAT_INC(tcps_conndrops);
1886 if (errno == ETIMEDOUT && tp->t_softerror)
1887 errno = tp->t_softerror;
1888 so->so_error = errno;
1889 return (tcp_close(tp));
1893 tcp_discardcb(struct tcpcb *tp)
1895 struct inpcb *inp = tp->t_inpcb;
1896 struct socket *so = inp->inp_socket;
1898 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1900 int released __unused;
1902 INP_WLOCK_ASSERT(inp);
1905 * Make sure that all of our timers are stopped before we delete the
1908 * If stopping a timer fails, we schedule a discard function in same
1909 * callout, and the last discard function called will take care of
1910 * deleting the tcpcb.
1912 tp->t_timers->tt_draincnt = 0;
1913 tcp_timer_stop(tp, TT_REXMT);
1914 tcp_timer_stop(tp, TT_PERSIST);
1915 tcp_timer_stop(tp, TT_KEEP);
1916 tcp_timer_stop(tp, TT_2MSL);
1917 tcp_timer_stop(tp, TT_DELACK);
1918 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1920 * Call the stop-all function of the methods,
1921 * this function should call the tcp_timer_stop()
1922 * method with each of the function specific timeouts.
1923 * That stop will be called via the tfb_tcp_timer_stop()
1924 * which should use the async drain function of the
1925 * callout system (see tcp_var.h).
1927 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1931 * If we got enough samples through the srtt filter,
1932 * save the rtt and rttvar in the routing entry.
1933 * 'Enough' is arbitrarily defined as 4 rtt samples.
1934 * 4 samples is enough for the srtt filter to converge
1935 * to within enough % of the correct value; fewer samples
1936 * and we could save a bogus rtt. The danger is not high
1937 * as tcp quickly recovers from everything.
1938 * XXX: Works very well but needs some more statistics!
1940 if (tp->t_rttupdated >= 4) {
1941 struct hc_metrics_lite metrics;
1944 bzero(&metrics, sizeof(metrics));
1946 * Update the ssthresh always when the conditions below
1947 * are satisfied. This gives us better new start value
1948 * for the congestion avoidance for new connections.
1949 * ssthresh is only set if packet loss occurred on a session.
1951 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1952 * being torn down. Ideally this code would not use 'so'.
1954 ssthresh = tp->snd_ssthresh;
1955 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1957 * convert the limit from user data bytes to
1958 * packets then to packet data bytes.
1960 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1963 ssthresh *= (tp->t_maxseg +
1965 (isipv6 ? sizeof (struct ip6_hdr) +
1966 sizeof (struct tcphdr) :
1968 sizeof (struct tcpiphdr)
1975 metrics.rmx_ssthresh = ssthresh;
1977 metrics.rmx_rtt = tp->t_srtt;
1978 metrics.rmx_rttvar = tp->t_rttvar;
1979 metrics.rmx_cwnd = tp->snd_cwnd;
1980 metrics.rmx_sendpipe = 0;
1981 metrics.rmx_recvpipe = 0;
1983 tcp_hc_update(&inp->inp_inc, &metrics);
1986 /* free the reassembly queue, if any */
1987 tcp_reass_flush(tp);
1990 /* Disconnect offload device, if any. */
1991 if (tp->t_flags & TF_TOE)
1992 tcp_offload_detach(tp);
1995 tcp_free_sackholes(tp);
1998 /* Free the TCP PCAP queues. */
1999 tcp_pcap_drain(&(tp->t_inpkts));
2000 tcp_pcap_drain(&(tp->t_outpkts));
2003 /* Allow the CC algorithm to clean up after itself. */
2004 if (CC_ALGO(tp)->cb_destroy != NULL)
2005 CC_ALGO(tp)->cb_destroy(tp->ccv);
2009 khelp_destroy_osd(tp->osd);
2012 stats_blob_destroy(tp->t_stats);
2016 inp->inp_ppcb = NULL;
2017 if (tp->t_timers->tt_draincnt == 0) {
2018 /* We own the last reference on tcpcb, let's free it. */
2020 tcp_log_tcpcbfini(tp);
2022 TCPSTATES_DEC(tp->t_state);
2023 if (tp->t_fb->tfb_tcp_fb_fini)
2024 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2025 refcount_release(&tp->t_fb->tfb_refcnt);
2027 uma_zfree(V_tcpcb_zone, tp);
2028 released = in_pcbrele_wlocked(inp);
2029 KASSERT(!released, ("%s: inp %p should not have been released "
2030 "here", __func__, inp));
2035 tcp_timer_discard(void *ptp)
2039 struct epoch_tracker et;
2041 tp = (struct tcpcb *)ptp;
2042 CURVNET_SET(tp->t_vnet);
2043 NET_EPOCH_ENTER(et);
2045 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2048 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2049 ("%s: tcpcb has to be stopped here", __func__));
2050 tp->t_timers->tt_draincnt--;
2051 if (tp->t_timers->tt_draincnt == 0) {
2052 /* We own the last reference on this tcpcb, let's free it. */
2054 tcp_log_tcpcbfini(tp);
2056 TCPSTATES_DEC(tp->t_state);
2057 if (tp->t_fb->tfb_tcp_fb_fini)
2058 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2059 refcount_release(&tp->t_fb->tfb_refcnt);
2061 uma_zfree(V_tcpcb_zone, tp);
2062 if (in_pcbrele_wlocked(inp)) {
2074 * Attempt to close a TCP control block, marking it as dropped, and freeing
2075 * the socket if we hold the only reference.
2078 tcp_close(struct tcpcb *tp)
2080 struct inpcb *inp = tp->t_inpcb;
2083 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2084 INP_WLOCK_ASSERT(inp);
2087 if (tp->t_state == TCPS_LISTEN)
2088 tcp_offload_listen_stop(tp);
2091 * This releases the TFO pending counter resource for TFO listen
2092 * sockets as well as passively-created TFO sockets that transition
2093 * from SYN_RECEIVED to CLOSED.
2095 if (tp->t_tfo_pending) {
2096 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2097 tp->t_tfo_pending = NULL;
2100 TCPSTAT_INC(tcps_closed);
2101 if (tp->t_state != TCPS_CLOSED)
2102 tcp_state_change(tp, TCPS_CLOSED);
2103 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2104 so = inp->inp_socket;
2105 soisdisconnected(so);
2106 if (inp->inp_flags & INP_SOCKREF) {
2107 KASSERT(so->so_state & SS_PROTOREF,
2108 ("tcp_close: !SS_PROTOREF"));
2109 inp->inp_flags &= ~INP_SOCKREF;
2112 so->so_state &= ~SS_PROTOREF;
2122 VNET_ITERATOR_DECL(vnet_iter);
2127 VNET_LIST_RLOCK_NOSLEEP();
2128 VNET_FOREACH(vnet_iter) {
2129 CURVNET_SET(vnet_iter);
2134 * Walk the tcpbs, if existing, and flush the reassembly queue,
2135 * if there is one...
2136 * XXX: The "Net/3" implementation doesn't imply that the TCP
2137 * reassembly queue should be flushed, but in a situation
2138 * where we're really low on mbufs, this is potentially
2141 INP_INFO_WLOCK(&V_tcbinfo);
2142 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2144 if (inpb->inp_flags & INP_TIMEWAIT) {
2148 if ((tcpb = intotcpcb(inpb)) != NULL) {
2149 tcp_reass_flush(tcpb);
2150 tcp_clean_sackreport(tcpb);
2152 tcp_log_drain(tcpb);
2155 if (tcp_pcap_aggressive_free) {
2156 /* Free the TCP PCAP queues. */
2157 tcp_pcap_drain(&(tcpb->t_inpkts));
2158 tcp_pcap_drain(&(tcpb->t_outpkts));
2164 INP_INFO_WUNLOCK(&V_tcbinfo);
2167 VNET_LIST_RUNLOCK_NOSLEEP();
2171 * Notify a tcp user of an asynchronous error;
2172 * store error as soft error, but wake up user
2173 * (for now, won't do anything until can select for soft error).
2175 * Do not wake up user since there currently is no mechanism for
2176 * reporting soft errors (yet - a kqueue filter may be added).
2178 static struct inpcb *
2179 tcp_notify(struct inpcb *inp, int error)
2183 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2184 INP_WLOCK_ASSERT(inp);
2186 if ((inp->inp_flags & INP_TIMEWAIT) ||
2187 (inp->inp_flags & INP_DROPPED))
2190 tp = intotcpcb(inp);
2191 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2194 * Ignore some errors if we are hooked up.
2195 * If connection hasn't completed, has retransmitted several times,
2196 * and receives a second error, give up now. This is better
2197 * than waiting a long time to establish a connection that
2198 * can never complete.
2200 if (tp->t_state == TCPS_ESTABLISHED &&
2201 (error == EHOSTUNREACH || error == ENETUNREACH ||
2202 error == EHOSTDOWN)) {
2203 if (inp->inp_route.ro_nh) {
2204 NH_FREE(inp->inp_route.ro_nh);
2205 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2208 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2210 tp = tcp_drop(tp, error);
2216 tp->t_softerror = error;
2220 wakeup( &so->so_timeo);
2227 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2229 struct epoch_tracker et;
2234 if (req->newptr != NULL)
2237 if (req->oldptr == NULL) {
2240 n = V_tcbinfo.ipi_count +
2241 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2242 n += imax(n / 8, 10);
2243 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2247 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2250 bzero(&xig, sizeof(xig));
2251 xig.xig_len = sizeof xig;
2252 xig.xig_count = V_tcbinfo.ipi_count +
2253 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2254 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2255 xig.xig_sogen = so_gencnt;
2256 error = SYSCTL_OUT(req, &xig, sizeof xig);
2260 error = syncache_pcblist(req);
2264 NET_EPOCH_ENTER(et);
2265 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2267 inp = CK_LIST_NEXT(inp, inp_list)) {
2269 if (inp->inp_gencnt <= xig.xig_gen) {
2273 * XXX: This use of cr_cansee(), introduced with
2274 * TCP state changes, is not quite right, but for
2275 * now, better than nothing.
2277 if (inp->inp_flags & INP_TIMEWAIT) {
2278 if (intotw(inp) != NULL)
2279 crerr = cr_cansee(req->td->td_ucred,
2280 intotw(inp)->tw_cred);
2282 crerr = EINVAL; /* Skip this inp. */
2284 crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2288 tcp_inptoxtp(inp, &xt);
2290 error = SYSCTL_OUT(req, &xt, sizeof xt);
2303 * Give the user an updated idea of our state.
2304 * If the generation differs from what we told
2305 * her before, she knows that something happened
2306 * while we were processing this request, and it
2307 * might be necessary to retry.
2309 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2310 xig.xig_sogen = so_gencnt;
2311 xig.xig_count = V_tcbinfo.ipi_count +
2312 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2313 error = SYSCTL_OUT(req, &xig, sizeof xig);
2319 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2320 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2321 NULL, 0, tcp_pcblist, "S,xtcpcb",
2322 "List of active TCP connections");
2326 tcp_getcred(SYSCTL_HANDLER_ARGS)
2329 struct sockaddr_in addrs[2];
2330 struct epoch_tracker et;
2334 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2337 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2340 NET_EPOCH_ENTER(et);
2341 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2342 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2345 if (inp->inp_socket == NULL)
2348 error = cr_canseeinpcb(req->td->td_ucred, inp);
2350 cru2x(inp->inp_cred, &xuc);
2355 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2359 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2360 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2361 0, 0, tcp_getcred, "S,xucred",
2362 "Get the xucred of a TCP connection");
2367 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2369 struct epoch_tracker et;
2371 struct sockaddr_in6 addrs[2];
2378 error = priv_check(req->td, PRIV_NETINET_GETCRED);
2381 error = SYSCTL_IN(req, addrs, sizeof(addrs));
2384 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2385 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2388 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2390 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2397 NET_EPOCH_ENTER(et);
2400 inp = in_pcblookup(&V_tcbinfo,
2401 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2403 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2404 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2407 inp = in6_pcblookup(&V_tcbinfo,
2408 &addrs[1].sin6_addr, addrs[1].sin6_port,
2409 &addrs[0].sin6_addr, addrs[0].sin6_port,
2410 INPLOOKUP_RLOCKPCB, NULL);
2413 if (inp->inp_socket == NULL)
2416 error = cr_canseeinpcb(req->td->td_ucred, inp);
2418 cru2x(inp->inp_cred, &xuc);
2423 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2427 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2428 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2429 0, 0, tcp6_getcred, "S,xucred",
2430 "Get the xucred of a TCP6 connection");
2436 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2438 struct ip *ip = vip;
2440 struct in_addr faddr;
2443 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2445 struct in_conninfo inc;
2446 tcp_seq icmp_tcp_seq;
2449 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2450 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2453 if (cmd == PRC_MSGSIZE)
2454 notify = tcp_mtudisc_notify;
2455 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2456 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2457 cmd == PRC_TIMXCEED_INTRANS) && ip)
2458 notify = tcp_drop_syn_sent;
2461 * Hostdead is ugly because it goes linearly through all PCBs.
2462 * XXX: We never get this from ICMP, otherwise it makes an
2463 * excellent DoS attack on machines with many connections.
2465 else if (cmd == PRC_HOSTDEAD)
2467 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2471 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2475 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2476 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2477 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2478 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2479 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2480 /* signal EHOSTDOWN, as it flushes the cached route */
2481 inp = (*notify)(inp, EHOSTDOWN);
2484 icmp_tcp_seq = th->th_seq;
2486 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2487 !(inp->inp_flags & INP_DROPPED) &&
2488 !(inp->inp_socket == NULL)) {
2489 tp = intotcpcb(inp);
2490 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2491 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2492 if (cmd == PRC_MSGSIZE) {
2495 * If we got a needfrag set the MTU
2496 * in the route to the suggested new
2497 * value (if given) and then notify.
2499 mtu = ntohs(icp->icmp_nextmtu);
2501 * If no alternative MTU was
2502 * proposed, try the next smaller
2507 ntohs(ip->ip_len), 1);
2508 if (mtu < V_tcp_minmss +
2509 sizeof(struct tcpiphdr))
2510 mtu = V_tcp_minmss +
2511 sizeof(struct tcpiphdr);
2513 * Only process the offered MTU if it
2514 * is smaller than the current one.
2516 if (mtu < tp->t_maxseg +
2517 sizeof(struct tcpiphdr)) {
2518 bzero(&inc, sizeof(inc));
2519 inc.inc_faddr = faddr;
2521 inp->inp_inc.inc_fibnum;
2522 tcp_hc_updatemtu(&inc, mtu);
2523 tcp_mtudisc(inp, mtu);
2526 inp = (*notify)(inp,
2527 inetctlerrmap[cmd]);
2531 bzero(&inc, sizeof(inc));
2532 inc.inc_fport = th->th_dport;
2533 inc.inc_lport = th->th_sport;
2534 inc.inc_faddr = faddr;
2535 inc.inc_laddr = ip->ip_src;
2536 syncache_unreach(&inc, icmp_tcp_seq);
2546 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2548 struct in6_addr *dst;
2549 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2550 struct ip6_hdr *ip6;
2554 struct icmp6_hdr *icmp6;
2555 struct ip6ctlparam *ip6cp = NULL;
2556 const struct sockaddr_in6 *sa6_src = NULL;
2557 struct in_conninfo inc;
2562 tcp_seq icmp_tcp_seq;
2566 if (sa->sa_family != AF_INET6 ||
2567 sa->sa_len != sizeof(struct sockaddr_in6))
2570 /* if the parameter is from icmp6, decode it. */
2572 ip6cp = (struct ip6ctlparam *)d;
2573 icmp6 = ip6cp->ip6c_icmp6;
2575 ip6 = ip6cp->ip6c_ip6;
2576 off = ip6cp->ip6c_off;
2577 sa6_src = ip6cp->ip6c_src;
2578 dst = ip6cp->ip6c_finaldst;
2582 off = 0; /* fool gcc */
2587 if (cmd == PRC_MSGSIZE)
2588 notify = tcp_mtudisc_notify;
2589 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2590 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2591 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2592 notify = tcp_drop_syn_sent;
2595 * Hostdead is ugly because it goes linearly through all PCBs.
2596 * XXX: We never get this from ICMP, otherwise it makes an
2597 * excellent DoS attack on machines with many connections.
2599 else if (cmd == PRC_HOSTDEAD)
2601 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2605 in6_pcbnotify(&V_tcbinfo, sa, 0,
2606 (const struct sockaddr *)sa6_src,
2607 0, cmd, NULL, notify);
2611 /* Check if we can safely get the ports from the tcp hdr */
2614 (int32_t) (off + sizeof(struct tcp_ports)))) {
2617 bzero(&t_ports, sizeof(struct tcp_ports));
2618 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2619 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2620 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2621 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2622 /* signal EHOSTDOWN, as it flushes the cached route */
2623 inp = (*notify)(inp, EHOSTDOWN);
2626 off += sizeof(struct tcp_ports);
2627 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2630 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2632 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2633 !(inp->inp_flags & INP_DROPPED) &&
2634 !(inp->inp_socket == NULL)) {
2635 tp = intotcpcb(inp);
2636 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2637 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2638 if (cmd == PRC_MSGSIZE) {
2641 * If we got a needfrag set the MTU
2642 * in the route to the suggested new
2643 * value (if given) and then notify.
2645 mtu = ntohl(icmp6->icmp6_mtu);
2647 * If no alternative MTU was
2648 * proposed, or the proposed
2649 * MTU was too small, set to
2652 if (mtu < IPV6_MMTU)
2653 mtu = IPV6_MMTU - 8;
2654 bzero(&inc, sizeof(inc));
2655 inc.inc_fibnum = M_GETFIB(m);
2656 inc.inc_flags |= INC_ISIPV6;
2657 inc.inc6_faddr = *dst;
2658 if (in6_setscope(&inc.inc6_faddr,
2659 m->m_pkthdr.rcvif, NULL))
2662 * Only process the offered MTU if it
2663 * is smaller than the current one.
2665 if (mtu < tp->t_maxseg +
2666 sizeof (struct tcphdr) +
2667 sizeof (struct ip6_hdr)) {
2668 tcp_hc_updatemtu(&inc, mtu);
2669 tcp_mtudisc(inp, mtu);
2670 ICMP6STAT_INC(icp6s_pmtuchg);
2673 inp = (*notify)(inp,
2674 inet6ctlerrmap[cmd]);
2678 bzero(&inc, sizeof(inc));
2679 inc.inc_fibnum = M_GETFIB(m);
2680 inc.inc_flags |= INC_ISIPV6;
2681 inc.inc_fport = t_ports.th_dport;
2682 inc.inc_lport = t_ports.th_sport;
2683 inc.inc6_faddr = *dst;
2684 inc.inc6_laddr = ip6->ip6_src;
2685 syncache_unreach(&inc, icmp_tcp_seq);
2694 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2699 KASSERT(len >= SIPHASH_KEY_LENGTH,
2700 ("%s: keylen %u too short ", __func__, len));
2701 SipHash24_Init(&ctx);
2702 SipHash_SetKey(&ctx, (uint8_t *)key);
2703 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2704 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2705 switch (inc->inc_flags & INC_ISIPV6) {
2708 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2709 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2714 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2715 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2719 SipHash_Final((uint8_t *)hash, &ctx);
2721 return (hash[0] ^ hash[1]);
2725 tcp_new_ts_offset(struct in_conninfo *inc)
2727 struct in_conninfo inc_store, *local_inc;
2729 if (!V_tcp_ts_offset_per_conn) {
2730 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2731 inc_store.inc_lport = 0;
2732 inc_store.inc_fport = 0;
2733 local_inc = &inc_store;
2737 return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2738 sizeof(V_ts_offset_secret)));
2742 * Following is where TCP initial sequence number generation occurs.
2744 * There are two places where we must use initial sequence numbers:
2745 * 1. In SYN-ACK packets.
2746 * 2. In SYN packets.
2748 * All ISNs for SYN-ACK packets are generated by the syncache. See
2749 * tcp_syncache.c for details.
2751 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2752 * depends on this property. In addition, these ISNs should be
2753 * unguessable so as to prevent connection hijacking. To satisfy
2754 * the requirements of this situation, the algorithm outlined in
2755 * RFC 1948 is used, with only small modifications.
2757 * Implementation details:
2759 * Time is based off the system timer, and is corrected so that it
2760 * increases by one megabyte per second. This allows for proper
2761 * recycling on high speed LANs while still leaving over an hour
2764 * As reading the *exact* system time is too expensive to be done
2765 * whenever setting up a TCP connection, we increment the time
2766 * offset in two ways. First, a small random positive increment
2767 * is added to isn_offset for each connection that is set up.
2768 * Second, the function tcp_isn_tick fires once per clock tick
2769 * and increments isn_offset as necessary so that sequence numbers
2770 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2771 * random positive increments serve only to ensure that the same
2772 * exact sequence number is never sent out twice (as could otherwise
2773 * happen when a port is recycled in less than the system tick
2776 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2777 * between seeding of isn_secret. This is normally set to zero,
2778 * as reseeding should not be necessary.
2780 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2781 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2782 * general, this means holding an exclusive (write) lock.
2785 #define ISN_BYTES_PER_SECOND 1048576
2786 #define ISN_STATIC_INCREMENT 4096
2787 #define ISN_RANDOM_INCREMENT (4096 - 1)
2788 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH
2790 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2791 VNET_DEFINE_STATIC(int, isn_last);
2792 VNET_DEFINE_STATIC(int, isn_last_reseed);
2793 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2794 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2796 #define V_isn_secret VNET(isn_secret)
2797 #define V_isn_last VNET(isn_last)
2798 #define V_isn_last_reseed VNET(isn_last_reseed)
2799 #define V_isn_offset VNET(isn_offset)
2800 #define V_isn_offset_old VNET(isn_offset_old)
2803 tcp_new_isn(struct in_conninfo *inc)
2806 u_int32_t projected_offset;
2809 /* Seed if this is the first use, reseed if requested. */
2810 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2811 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2813 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2814 V_isn_last_reseed = ticks;
2817 /* Compute the hash and return the ISN. */
2818 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2819 sizeof(V_isn_secret));
2820 V_isn_offset += ISN_STATIC_INCREMENT +
2821 (arc4random() & ISN_RANDOM_INCREMENT);
2822 if (ticks != V_isn_last) {
2823 projected_offset = V_isn_offset_old +
2824 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2825 if (SEQ_GT(projected_offset, V_isn_offset))
2826 V_isn_offset = projected_offset;
2827 V_isn_offset_old = V_isn_offset;
2830 new_isn += V_isn_offset;
2836 * When a specific ICMP unreachable message is received and the
2837 * connection state is SYN-SENT, drop the connection. This behavior
2838 * is controlled by the icmp_may_rst sysctl.
2841 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2846 INP_WLOCK_ASSERT(inp);
2848 if ((inp->inp_flags & INP_TIMEWAIT) ||
2849 (inp->inp_flags & INP_DROPPED))
2852 tp = intotcpcb(inp);
2853 if (tp->t_state != TCPS_SYN_SENT)
2856 if (IS_FASTOPEN(tp->t_flags))
2857 tcp_fastopen_disable_path(tp);
2859 tp = tcp_drop(tp, errno);
2867 * When `need fragmentation' ICMP is received, update our idea of the MSS
2868 * based on the new value. Also nudge TCP to send something, since we
2869 * know the packet we just sent was dropped.
2870 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2872 static struct inpcb *
2873 tcp_mtudisc_notify(struct inpcb *inp, int error)
2876 tcp_mtudisc(inp, -1);
2881 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2886 INP_WLOCK_ASSERT(inp);
2887 if ((inp->inp_flags & INP_TIMEWAIT) ||
2888 (inp->inp_flags & INP_DROPPED))
2891 tp = intotcpcb(inp);
2892 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2894 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2896 so = inp->inp_socket;
2897 SOCKBUF_LOCK(&so->so_snd);
2898 /* If the mss is larger than the socket buffer, decrease the mss. */
2899 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2900 tp->t_maxseg = so->so_snd.sb_hiwat;
2901 SOCKBUF_UNLOCK(&so->so_snd);
2903 TCPSTAT_INC(tcps_mturesent);
2905 tp->snd_nxt = tp->snd_una;
2906 tcp_free_sackholes(tp);
2907 tp->snd_recover = tp->snd_max;
2908 if (tp->t_flags & TF_SACK_PERMIT)
2909 EXIT_FASTRECOVERY(tp->t_flags);
2910 tp->t_fb->tfb_tcp_output(tp);
2915 * Look-up the routing entry to the peer of this inpcb. If no route
2916 * is found and it cannot be allocated, then return 0. This routine
2917 * is called by TCP routines that access the rmx structure and by
2918 * tcp_mss_update to get the peer/interface MTU.
2921 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2923 struct nhop_object *nh;
2925 uint32_t maxmtu = 0;
2927 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2929 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2931 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
2936 maxmtu = nh->nh_mtu;
2938 /* Report additional interface capabilities. */
2940 if (ifp->if_capenable & IFCAP_TSO4 &&
2941 ifp->if_hwassist & CSUM_TSO) {
2942 cap->ifcap |= CSUM_TSO;
2943 cap->tsomax = ifp->if_hw_tsomax;
2944 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2945 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2955 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2957 struct nhop_object *nh;
2958 struct in6_addr dst6;
2961 uint32_t maxmtu = 0;
2963 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2965 if (inc->inc_flags & INC_IPV6MINMTU)
2968 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2969 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2970 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
2975 maxmtu = nh->nh_mtu;
2977 /* Report additional interface capabilities. */
2979 if (ifp->if_capenable & IFCAP_TSO6 &&
2980 ifp->if_hwassist & CSUM_TSO) {
2981 cap->ifcap |= CSUM_TSO;
2982 cap->tsomax = ifp->if_hw_tsomax;
2983 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2984 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2994 * Calculate effective SMSS per RFC5681 definition for a given TCP
2995 * connection at its current state, taking into account SACK and etc.
2998 tcp_maxseg(const struct tcpcb *tp)
3002 if (tp->t_flags & TF_NOOPT)
3003 return (tp->t_maxseg);
3006 * Here we have a simplified code from tcp_addoptions(),
3007 * without a proper loop, and having most of paddings hardcoded.
3008 * We might make mistakes with padding here in some edge cases,
3009 * but this is harmless, since result of tcp_maxseg() is used
3010 * only in cwnd and ssthresh estimations.
3012 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
3013 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3014 if (tp->t_flags & TF_RCVD_TSTMP)
3015 optlen = TCPOLEN_TSTAMP_APPA;
3018 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3019 if (tp->t_flags & TF_SIGNATURE)
3020 optlen += PAD(TCPOLEN_SIGNATURE);
3022 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3023 optlen += TCPOLEN_SACKHDR;
3024 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3025 optlen = PAD(optlen);
3028 if (tp->t_flags & TF_REQ_TSTMP)
3029 optlen = TCPOLEN_TSTAMP_APPA;
3031 optlen = PAD(TCPOLEN_MAXSEG);
3032 if (tp->t_flags & TF_REQ_SCALE)
3033 optlen += PAD(TCPOLEN_WINDOW);
3034 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3035 if (tp->t_flags & TF_SIGNATURE)
3036 optlen += PAD(TCPOLEN_SIGNATURE);
3038 if (tp->t_flags & TF_SACK_PERMIT)
3039 optlen += PAD(TCPOLEN_SACK_PERMITTED);
3042 optlen = min(optlen, TCP_MAXOLEN);
3043 return (tp->t_maxseg - optlen);
3047 sysctl_drop(SYSCTL_HANDLER_ARGS)
3049 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3050 struct sockaddr_storage addrs[2];
3054 struct sockaddr_in *fin, *lin;
3055 struct epoch_tracker et;
3057 struct sockaddr_in6 *fin6, *lin6;
3068 if (req->oldptr != NULL || req->oldlen != 0)
3070 if (req->newptr == NULL)
3072 if (req->newlen < sizeof(addrs))
3074 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3078 switch (addrs[0].ss_family) {
3081 fin6 = (struct sockaddr_in6 *)&addrs[0];
3082 lin6 = (struct sockaddr_in6 *)&addrs[1];
3083 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3084 lin6->sin6_len != sizeof(struct sockaddr_in6))
3086 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3087 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3089 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3090 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3091 fin = (struct sockaddr_in *)&addrs[0];
3092 lin = (struct sockaddr_in *)&addrs[1];
3095 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3098 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3105 fin = (struct sockaddr_in *)&addrs[0];
3106 lin = (struct sockaddr_in *)&addrs[1];
3107 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3108 lin->sin_len != sizeof(struct sockaddr_in))
3115 NET_EPOCH_ENTER(et);
3116 switch (addrs[0].ss_family) {
3119 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3120 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3121 INPLOOKUP_WLOCKPCB, NULL);
3126 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3127 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3132 if (inp->inp_flags & INP_TIMEWAIT) {
3134 * XXXRW: There currently exists a state where an
3135 * inpcb is present, but its timewait state has been
3136 * discarded. For now, don't allow dropping of this
3144 } else if (!(inp->inp_flags & INP_DROPPED) &&
3145 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3146 tp = intotcpcb(inp);
3147 tp = tcp_drop(tp, ECONNABORTED);
3158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3159 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3160 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3161 "Drop TCP connection");
3165 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3167 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3168 struct sockaddr_storage addrs[2];
3170 struct sockaddr_in *fin, *lin;
3171 struct epoch_tracker et;
3173 struct sockaddr_in6 *fin6, *lin6;
3184 if (req->oldptr != NULL || req->oldlen != 0)
3186 if (req->newptr == NULL)
3188 if (req->newlen < sizeof(addrs))
3190 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3194 switch (addrs[0].ss_family) {
3197 fin6 = (struct sockaddr_in6 *)&addrs[0];
3198 lin6 = (struct sockaddr_in6 *)&addrs[1];
3199 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3200 lin6->sin6_len != sizeof(struct sockaddr_in6))
3202 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3203 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3205 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3206 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3207 fin = (struct sockaddr_in *)&addrs[0];
3208 lin = (struct sockaddr_in *)&addrs[1];
3211 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3214 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3221 fin = (struct sockaddr_in *)&addrs[0];
3222 lin = (struct sockaddr_in *)&addrs[1];
3223 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3224 lin->sin_len != sizeof(struct sockaddr_in))
3231 NET_EPOCH_ENTER(et);
3232 switch (addrs[0].ss_family) {
3235 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3236 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3237 INPLOOKUP_WLOCKPCB, NULL);
3242 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3243 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3249 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3250 inp->inp_socket == NULL) {
3256 so = inp->inp_socket;
3258 error = ktls_set_tx_mode(so,
3259 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3269 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3270 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3271 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3272 "Switch TCP connection to SW TLS");
3273 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3274 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3275 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3276 "Switch TCP connection to ifnet TLS");
3280 * Generate a standardized TCP log line for use throughout the
3281 * tcp subsystem. Memory allocation is done with M_NOWAIT to
3282 * allow use in the interrupt context.
3284 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3285 * NB: The function may return NULL if memory allocation failed.
3287 * Due to header inclusion and ordering limitations the struct ip
3288 * and ip6_hdr pointers have to be passed as void pointers.
3291 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3295 /* Is logging enabled? */
3296 if (V_tcp_log_in_vain == 0)
3299 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3303 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3307 /* Is logging enabled? */
3308 if (tcp_log_debug == 0)
3311 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3315 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3322 const struct ip6_hdr *ip6;
3324 ip6 = (const struct ip6_hdr *)ip6hdr;
3326 ip = (struct ip *)ip4hdr;
3329 * The log line looks like this:
3330 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3332 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3333 sizeof(PRINT_TH_FLAGS) + 1 +
3335 2 * INET6_ADDRSTRLEN;
3337 2 * INET_ADDRSTRLEN;
3340 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3344 strcat(s, "TCP: [");
3347 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3348 inet_ntoa_r(inc->inc_faddr, sp);
3350 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3352 inet_ntoa_r(inc->inc_laddr, sp);
3354 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3357 ip6_sprintf(sp, &inc->inc6_faddr);
3359 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3361 ip6_sprintf(sp, &inc->inc6_laddr);
3363 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3364 } else if (ip6 && th) {
3365 ip6_sprintf(sp, &ip6->ip6_src);
3367 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3369 ip6_sprintf(sp, &ip6->ip6_dst);
3371 sprintf(sp, "]:%i", ntohs(th->th_dport));
3374 } else if (ip && th) {
3375 inet_ntoa_r(ip->ip_src, sp);
3377 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3379 inet_ntoa_r(ip->ip_dst, sp);
3381 sprintf(sp, "]:%i", ntohs(th->th_dport));
3389 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3390 if (*(s + size - 1) != '\0')
3391 panic("%s: string too long", __func__);
3396 * A subroutine which makes it easy to track TCP state changes with DTrace.
3397 * This function shouldn't be called for t_state initializations that don't
3398 * correspond to actual TCP state transitions.
3401 tcp_state_change(struct tcpcb *tp, int newstate)
3403 #if defined(KDTRACE_HOOKS)
3404 int pstate = tp->t_state;
3407 TCPSTATES_DEC(tp->t_state);
3408 TCPSTATES_INC(newstate);
3409 tp->t_state = newstate;
3410 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3414 * Create an external-format (``xtcpcb'') structure using the information in
3415 * the kernel-format tcpcb structure pointed to by tp. This is done to
3416 * reduce the spew of irrelevant information over this interface, to isolate
3417 * user code from changes in the kernel structure, and potentially to provide
3418 * information-hiding if we decide that some of this information should be
3419 * hidden from users.
3422 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3424 struct tcpcb *tp = intotcpcb(inp);
3427 bzero(xt, sizeof(*xt));
3428 if (inp->inp_flags & INP_TIMEWAIT) {
3429 xt->t_state = TCPS_TIME_WAIT;
3431 xt->t_state = tp->t_state;
3432 xt->t_logstate = tp->t_logstate;
3433 xt->t_flags = tp->t_flags;
3434 xt->t_sndzerowin = tp->t_sndzerowin;
3435 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3436 xt->t_rcvoopack = tp->t_rcvoopack;
3438 now = getsbinuptime();
3439 #define COPYTIMER(ttt) do { \
3440 if (callout_active(&tp->t_timers->ttt)) \
3441 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
3446 COPYTIMER(tt_delack);
3447 COPYTIMER(tt_rexmt);
3448 COPYTIMER(tt_persist);
3452 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3454 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3455 TCP_FUNCTION_NAME_LEN_MAX);
3457 (void)tcp_log_get_id(tp, xt->xt_logid);
3461 xt->xt_len = sizeof(struct xtcpcb);
3462 in_pcbtoxinpcb(inp, &xt->xt_inp);
3463 if (inp->inp_socket == NULL)
3464 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3468 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3473 (status > TCP_EI_STATUS_MAX_VALUE) ||
3478 if (status > (sizeof(uint32_t) * 8)) {
3479 /* Should this be a KASSERT? */
3482 bit = 1U << (status - 1);
3483 if (bit & tp->t_end_info_status) {
3484 /* already logged */
3487 for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3488 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3489 tp->t_end_info_bytes[i] = status;
3490 tp->t_end_info_status |= bit;