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("$FreeBSD$");
37 #include "opt_compat.h"
39 #include "opt_inet6.h"
40 #include "opt_ipsec.h"
41 #include "opt_tcpdebug.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
48 #include <sys/hhook.h>
50 #include <sys/kernel.h>
52 #include <sys/khelp.h>
54 #include <sys/sysctl.h>
56 #include <sys/malloc.h>
57 #include <sys/refcount.h>
60 #include <sys/domain.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/protosw.h>
68 #include <sys/random.h>
72 #include <net/route.h>
74 #include <net/if_var.h>
77 #include <netinet/in.h>
78 #include <netinet/in_fib.h>
79 #include <netinet/in_kdtrace.h>
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_systm.h>
82 #include <netinet/in_var.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_icmp.h>
85 #include <netinet/ip_var.h>
87 #include <netinet/icmp6.h>
88 #include <netinet/ip6.h>
89 #include <netinet6/in6_fib.h>
90 #include <netinet6/in6_pcb.h>
91 #include <netinet6/ip6_var.h>
92 #include <netinet6/scope6_var.h>
93 #include <netinet6/nd6.h>
97 #include <netinet/tcp_fastopen.h>
99 #include <netinet/tcp.h>
100 #include <netinet/tcp_fsm.h>
101 #include <netinet/tcp_seq.h>
102 #include <netinet/tcp_timer.h>
103 #include <netinet/tcp_var.h>
104 #include <netinet/tcp_syncache.h>
105 #include <netinet/cc/cc.h>
107 #include <netinet6/tcp6_var.h>
109 #include <netinet/tcpip.h>
111 #include <netinet/tcp_pcap.h>
114 #include <netinet/tcp_debug.h>
117 #include <netinet6/ip6protosw.h>
120 #include <netinet/tcp_offload.h>
123 #include <netipsec/ipsec_support.h>
125 #include <machine/in_cksum.h>
128 #include <security/mac/mac_framework.h>
130 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
132 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
135 struct rwlock tcp_function_lock;
138 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
143 error = sysctl_handle_int(oidp, &new, 0, req);
144 if (error == 0 && req->newptr) {
145 if (new < TCP_MINMSS)
153 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
154 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
155 &sysctl_net_inet_tcp_mss_check, "I",
156 "Default TCP Maximum Segment Size");
160 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
164 new = V_tcp_v6mssdflt;
165 error = sysctl_handle_int(oidp, &new, 0, req);
166 if (error == 0 && req->newptr) {
167 if (new < TCP_MINMSS)
170 V_tcp_v6mssdflt = new;
175 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
177 &sysctl_net_inet_tcp_mss_v6_check, "I",
178 "Default TCP Maximum Segment Size for IPv6");
182 * Minimum MSS we accept and use. This prevents DoS attacks where
183 * we are forced to a ridiculous low MSS like 20 and send hundreds
184 * of packets instead of one. The effect scales with the available
185 * bandwidth and quickly saturates the CPU and network interface
186 * with packet generation and sending. Set to zero to disable MINMSS
187 * checking. This setting prevents us from sending too small packets.
189 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_minmss), 0,
192 "Minimum TCP Maximum Segment Size");
194 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
195 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
196 &VNET_NAME(tcp_do_rfc1323), 0,
197 "Enable rfc1323 (high performance TCP) extensions");
199 static int tcp_log_debug = 0;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
201 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
203 static int tcp_tcbhashsize;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
205 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
207 static int do_tcpdrain = 1;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
209 "Enable tcp_drain routine for extra help when low on mbufs");
211 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
212 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
214 static VNET_DEFINE(int, icmp_may_rst) = 1;
215 #define V_icmp_may_rst VNET(icmp_may_rst)
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(icmp_may_rst), 0,
218 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
220 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
221 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
223 &VNET_NAME(tcp_isn_reseed_interval), 0,
224 "Seconds between reseeding of ISN secret");
226 static int tcp_soreceive_stream;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
228 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
230 VNET_DEFINE(uma_zone_t, sack_hole_zone);
231 #define V_sack_hole_zone VNET(sack_hole_zone)
234 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
237 static struct inpcb *tcp_notify(struct inpcb *, int);
238 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
239 static void tcp_mtudisc(struct inpcb *, int);
240 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
241 void *ip4hdr, const void *ip6hdr);
244 static struct tcp_function_block tcp_def_funcblk = {
248 tcp_default_ctloutput,
259 int t_functions_inited = 0;
260 struct tcp_funchead t_functions;
261 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
264 init_tcp_functions(void)
266 if (t_functions_inited == 0) {
267 TAILQ_INIT(&t_functions);
268 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
269 t_functions_inited = 1;
273 static struct tcp_function_block *
274 find_tcp_functions_locked(struct tcp_function_set *fs)
276 struct tcp_function *f;
277 struct tcp_function_block *blk=NULL;
279 TAILQ_FOREACH(f, &t_functions, tf_next) {
280 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
288 static struct tcp_function_block *
289 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
291 struct tcp_function_block *rblk=NULL;
292 struct tcp_function *f;
294 TAILQ_FOREACH(f, &t_functions, tf_next) {
295 if (f->tf_fb == blk) {
306 struct tcp_function_block *
307 find_and_ref_tcp_functions(struct tcp_function_set *fs)
309 struct tcp_function_block *blk;
311 rw_rlock(&tcp_function_lock);
312 blk = find_tcp_functions_locked(fs);
314 refcount_acquire(&blk->tfb_refcnt);
315 rw_runlock(&tcp_function_lock);
319 struct tcp_function_block *
320 find_and_ref_tcp_fb(struct tcp_function_block *blk)
322 struct tcp_function_block *rblk;
324 rw_rlock(&tcp_function_lock);
325 rblk = find_tcp_fb_locked(blk, NULL);
327 refcount_acquire(&rblk->tfb_refcnt);
328 rw_runlock(&tcp_function_lock);
334 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
337 struct tcp_function_set fs;
338 struct tcp_function_block *blk;
340 memset(&fs, 0, sizeof(fs));
341 rw_rlock(&tcp_function_lock);
342 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
345 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
346 fs.pcbcnt = blk->tfb_refcnt;
348 rw_runlock(&tcp_function_lock);
349 error = sysctl_handle_string(oidp, fs.function_set_name,
350 sizeof(fs.function_set_name), req);
352 /* Check for error or no change */
353 if (error != 0 || req->newptr == NULL)
356 rw_wlock(&tcp_function_lock);
357 blk = find_tcp_functions_locked(&fs);
359 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
363 tcp_func_set_ptr = blk;
365 rw_wunlock(&tcp_function_lock);
369 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
370 CTLTYPE_STRING | CTLFLAG_RW,
371 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
372 "Set/get the default TCP functions");
375 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
377 int error, cnt, linesz;
378 struct tcp_function *f;
384 rw_rlock(&tcp_function_lock);
385 TAILQ_FOREACH(f, &t_functions, tf_next) {
388 rw_runlock(&tcp_function_lock);
390 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
391 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
396 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
397 "Alias", "PCB count");
402 rw_rlock(&tcp_function_lock);
403 TAILQ_FOREACH(f, &t_functions, tf_next) {
404 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
405 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
406 f->tf_fb->tfb_tcp_block_name,
407 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
408 alias ? f->tf_name : "-",
409 f->tf_fb->tfb_refcnt);
410 if (linesz >= bufsz) {
418 rw_runlock(&tcp_function_lock);
420 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
421 free(buffer, M_TEMP);
425 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
426 CTLTYPE_STRING|CTLFLAG_RD,
427 NULL, 0, sysctl_net_inet_list_available, "A",
428 "list available TCP Function sets");
431 * Target size of TCP PCB hash tables. Must be a power of two.
433 * Note that this can be overridden by the kernel environment
434 * variable net.inet.tcp.tcbhashsize
437 #define TCBHASHSIZE 0
442 * Callouts should be moved into struct tcp directly. They are currently
443 * separate because the tcpcb structure is exported to userland for sysctl
444 * parsing purposes, which do not know about callouts.
455 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
456 #define V_tcpcb_zone VNET(tcpcb_zone)
458 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
459 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
461 static struct mtx isn_mtx;
463 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
464 #define ISN_LOCK() mtx_lock(&isn_mtx)
465 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
468 * TCP initialization.
471 tcp_zone_change(void *tag)
474 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
475 uma_zone_set_max(V_tcpcb_zone, maxsockets);
476 tcp_tw_zone_change();
480 tcp_inpcb_init(void *mem, int size, int flags)
482 struct inpcb *inp = mem;
484 INP_LOCK_INIT(inp, "inp", "tcpinp");
489 * Take a value and get the next power of 2 that doesn't overflow.
490 * Used to size the tcp_inpcb hash buckets.
493 maketcp_hashsize(int size)
499 * get the next power of 2 higher than maxsockets.
501 hashsize = 1 << fls(size);
502 /* catch overflow, and just go one power of 2 smaller */
503 if (hashsize < size) {
504 hashsize = 1 << (fls(size) - 1);
510 * Register a TCP function block with the name provided in the names
511 * array. (Note that this function does NOT automatically register
512 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
513 * explicitly include blk->tfb_tcp_block_name in the list of names if
514 * you wish to register the stack with that name.)
516 * Either all name registrations will succeed or all will fail. If
517 * a name registration fails, the function will update the num_names
518 * argument to point to the array index of the name that encountered
521 * Returns 0 on success, or an error code on failure.
524 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
525 const char *names[], int *num_names)
527 struct tcp_function *n;
528 struct tcp_function_set fs;
531 KASSERT(names != NULL && *num_names > 0,
532 ("%s: Called with 0-length name list", __func__));
533 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
535 if (t_functions_inited == 0) {
536 init_tcp_functions();
538 if ((blk->tfb_tcp_output == NULL) ||
539 (blk->tfb_tcp_do_segment == NULL) ||
540 (blk->tfb_tcp_ctloutput == NULL) ||
541 (strlen(blk->tfb_tcp_block_name) == 0)) {
543 * These functions are required and you
549 if (blk->tfb_tcp_timer_stop_all ||
550 blk->tfb_tcp_timer_activate ||
551 blk->tfb_tcp_timer_active ||
552 blk->tfb_tcp_timer_stop) {
554 * If you define one timer function you
555 * must have them all.
557 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
558 (blk->tfb_tcp_timer_activate == NULL) ||
559 (blk->tfb_tcp_timer_active == NULL) ||
560 (blk->tfb_tcp_timer_stop == NULL)) {
566 refcount_init(&blk->tfb_refcnt, 0);
568 for (i = 0; i < *num_names; i++) {
569 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
576 (void)strncpy(fs.function_set_name, names[i],
577 TCP_FUNCTION_NAME_LEN_MAX);
578 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
579 rw_wlock(&tcp_function_lock);
580 if (find_tcp_functions_locked(&fs) != NULL) {
581 /* Duplicate name space not allowed */
582 rw_wunlock(&tcp_function_lock);
583 free(n, M_TCPFUNCTIONS);
587 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
588 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
589 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
590 rw_wunlock(&tcp_function_lock);
596 * Deregister the names we just added. Because registration failed
597 * for names[i], we don't need to deregister that name.
600 rw_wlock(&tcp_function_lock);
602 TAILQ_FOREACH(n, &t_functions, tf_next) {
603 if (!strncmp(n->tf_name, names[i],
604 TCP_FUNCTION_NAME_LEN_MAX)) {
605 TAILQ_REMOVE(&t_functions, n, tf_next);
607 free(n, M_TCPFUNCTIONS);
612 rw_wunlock(&tcp_function_lock);
617 * Register a TCP function block using the name provided in the name
620 * Returns 0 on success, or an error code on failure.
623 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
626 const char *name_list[1];
633 name_list[0] = blk->tfb_tcp_block_name;
634 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
639 * Register a TCP function block using the name defined in
640 * blk->tfb_tcp_block_name.
642 * Returns 0 on success, or an error code on failure.
645 register_tcp_functions(struct tcp_function_block *blk, int wait)
648 return (register_tcp_functions_as_name(blk, NULL, wait));
652 deregister_tcp_functions(struct tcp_function_block *blk)
654 struct tcp_function *f;
657 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
658 /* You can't un-register the default */
661 rw_wlock(&tcp_function_lock);
662 if (blk == tcp_func_set_ptr) {
663 /* You can't free the current default */
664 rw_wunlock(&tcp_function_lock);
667 if (blk->tfb_refcnt) {
668 /* Still tcb attached, mark it. */
669 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
670 rw_wunlock(&tcp_function_lock);
673 while (find_tcp_fb_locked(blk, &f) != NULL) {
675 TAILQ_REMOVE(&t_functions, f, tf_next);
677 free(f, M_TCPFUNCTIONS);
680 rw_wunlock(&tcp_function_lock);
687 const char *tcbhash_tuneable;
690 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
693 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
694 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
695 printf("%s: WARNING: unable to register helper hook\n", __func__);
696 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
697 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
698 printf("%s: WARNING: unable to register helper hook\n", __func__);
700 hashsize = TCBHASHSIZE;
701 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
704 * Auto tune the hash size based on maxsockets.
705 * A perfect hash would have a 1:1 mapping
706 * (hashsize = maxsockets) however it's been
707 * suggested that O(2) average is better.
709 hashsize = maketcp_hashsize(maxsockets / 4);
711 * Our historical default is 512,
712 * do not autotune lower than this.
716 if (bootverbose && IS_DEFAULT_VNET(curvnet))
717 printf("%s: %s auto tuned to %d\n", __func__,
718 tcbhash_tuneable, hashsize);
721 * We require a hashsize to be a power of two.
722 * Previously if it was not a power of two we would just reset it
723 * back to 512, which could be a nasty surprise if you did not notice
725 * Instead what we do is clip it to the closest power of two lower
726 * than the specified hash value.
728 if (!powerof2(hashsize)) {
729 int oldhashsize = hashsize;
731 hashsize = maketcp_hashsize(hashsize);
732 /* prevent absurdly low value */
735 printf("%s: WARNING: TCB hash size not a power of 2, "
736 "clipped from %d to %d.\n", __func__, oldhashsize,
739 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
740 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
743 * These have to be type stable for the benefit of the timers.
745 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
746 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
747 uma_zone_set_max(V_tcpcb_zone, maxsockets);
748 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
754 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
755 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
756 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
762 /* Skip initialization of globals for non-default instances. */
763 if (!IS_DEFAULT_VNET(curvnet))
766 tcp_reass_global_init();
768 /* XXX virtualize those bellow? */
769 tcp_delacktime = TCPTV_DELACK;
770 tcp_keepinit = TCPTV_KEEP_INIT;
771 tcp_keepidle = TCPTV_KEEP_IDLE;
772 tcp_keepintvl = TCPTV_KEEPINTVL;
773 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
775 tcp_rexmit_min = TCPTV_MIN;
776 if (tcp_rexmit_min < 1)
778 tcp_persmin = TCPTV_PERSMIN;
779 tcp_persmax = TCPTV_PERSMAX;
780 tcp_rexmit_slop = TCPTV_CPU_VAR;
781 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
782 tcp_tcbhashsize = hashsize;
783 /* Setup the tcp function block list */
784 init_tcp_functions();
785 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
787 if (tcp_soreceive_stream) {
789 tcp_usrreqs.pru_soreceive = soreceive_stream;
792 tcp6_usrreqs.pru_soreceive = soreceive_stream;
797 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
799 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
801 if (max_protohdr < TCP_MINPROTOHDR)
802 max_protohdr = TCP_MINPROTOHDR;
803 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
805 #undef TCP_MINPROTOHDR
808 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
809 SHUTDOWN_PRI_DEFAULT);
810 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
811 EVENTHANDLER_PRI_ANY);
819 tcp_destroy(void *unused __unused)
827 * All our processes are gone, all our sockets should be cleaned
828 * up, which means, we should be past the tcp_discardcb() calls.
829 * Sleep to let all tcpcb timers really disappear and cleanup.
832 INP_LIST_RLOCK(&V_tcbinfo);
833 n = V_tcbinfo.ipi_count;
834 INP_LIST_RUNLOCK(&V_tcbinfo);
837 pause("tcpdes", hz / 10);
842 in_pcbinfo_destroy(&V_tcbinfo);
843 /* tcp_discardcb() clears the sack_holes up. */
844 uma_zdestroy(V_sack_hole_zone);
845 uma_zdestroy(V_tcpcb_zone);
849 * Cannot free the zone until all tcpcbs are released as we attach
850 * the allocations to them.
852 tcp_fastopen_destroy();
856 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
858 printf("%s: WARNING: unable to deregister helper hook "
859 "type=%d, id=%d: error %d returned\n", __func__,
860 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
862 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
864 printf("%s: WARNING: unable to deregister helper hook "
865 "type=%d, id=%d: error %d returned\n", __func__,
866 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
870 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
880 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
881 * tcp_template used to store this data in mbufs, but we now recopy it out
882 * of the tcpcb each time to conserve mbufs.
885 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
887 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
889 INP_WLOCK_ASSERT(inp);
892 if ((inp->inp_vflag & INP_IPV6) != 0) {
895 ip6 = (struct ip6_hdr *)ip_ptr;
896 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
897 (inp->inp_flow & IPV6_FLOWINFO_MASK);
898 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
899 (IPV6_VERSION & IPV6_VERSION_MASK);
900 ip6->ip6_nxt = IPPROTO_TCP;
901 ip6->ip6_plen = htons(sizeof(struct tcphdr));
902 ip6->ip6_src = inp->in6p_laddr;
903 ip6->ip6_dst = inp->in6p_faddr;
906 #if defined(INET6) && defined(INET)
913 ip = (struct ip *)ip_ptr;
914 ip->ip_v = IPVERSION;
916 ip->ip_tos = inp->inp_ip_tos;
920 ip->ip_ttl = inp->inp_ip_ttl;
922 ip->ip_p = IPPROTO_TCP;
923 ip->ip_src = inp->inp_laddr;
924 ip->ip_dst = inp->inp_faddr;
927 th->th_sport = inp->inp_lport;
928 th->th_dport = inp->inp_fport;
936 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
940 * Create template to be used to send tcp packets on a connection.
941 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
942 * use for this function is in keepalives, which use tcp_respond.
945 tcpip_maketemplate(struct inpcb *inp)
949 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
952 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
957 * Send a single message to the TCP at address specified by
958 * the given TCP/IP header. If m == NULL, then we make a copy
959 * of the tcpiphdr at th and send directly to the addressed host.
960 * This is used to force keep alive messages out using the TCP
961 * template for a connection. If flags are given then we send
962 * a message back to the TCP which originated the segment th,
963 * and discard the mbuf containing it and any other attached mbufs.
965 * In any case the ack and sequence number of the transmitted
966 * segment are as specified by the parameters.
968 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
971 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
972 tcp_seq ack, tcp_seq seq, int flags)
984 int optlen, tlen, win;
987 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
990 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
997 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
998 INP_WLOCK_ASSERT(inp);
1005 if (!(flags & TH_RST)) {
1006 win = sbspace(&inp->inp_socket->so_rcv);
1007 if (win > TCP_MAXWIN << tp->rcv_scale)
1008 win = TCP_MAXWIN << tp->rcv_scale;
1010 if ((tp->t_flags & TF_NOOPT) == 0)
1014 m = m_gethdr(M_NOWAIT, MT_DATA);
1017 m->m_data += max_linkhdr;
1020 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1021 sizeof(struct ip6_hdr));
1022 ip6 = mtod(m, struct ip6_hdr *);
1023 nth = (struct tcphdr *)(ip6 + 1);
1027 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1028 ip = mtod(m, struct ip *);
1029 nth = (struct tcphdr *)(ip + 1);
1031 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1033 } else if (!M_WRITABLE(m)) {
1036 /* Can't reuse 'm', allocate a new mbuf. */
1037 n = m_gethdr(M_NOWAIT, MT_DATA);
1043 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1049 n->m_data += max_linkhdr;
1050 /* m_len is set later */
1051 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1054 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1055 sizeof(struct ip6_hdr));
1056 ip6 = mtod(n, struct ip6_hdr *);
1057 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1058 nth = (struct tcphdr *)(ip6 + 1);
1062 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1063 ip = mtod(n, struct ip *);
1064 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1065 nth = (struct tcphdr *)(ip + 1);
1067 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1068 xchg(nth->th_dport, nth->th_sport, uint16_t);
1075 * XXX MRT We inherit the FIB, which is lucky.
1079 m->m_data = (caddr_t)ipgen;
1080 /* m_len is set later */
1083 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1084 nth = (struct tcphdr *)(ip6 + 1);
1088 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1089 nth = (struct tcphdr *)(ip + 1);
1093 * this is usually a case when an extension header
1094 * exists between the IPv6 header and the
1097 nth->th_sport = th->th_sport;
1098 nth->th_dport = th->th_dport;
1100 xchg(nth->th_dport, nth->th_sport, uint16_t);
1106 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1108 #if defined(INET) && defined(INET6)
1112 tlen = sizeof (struct tcpiphdr);
1116 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1117 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1118 m, tlen, (long)M_TRAILINGSPACE(m)));
1123 /* Make sure we have room. */
1124 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1125 m->m_next = m_get(M_NOWAIT, MT_DATA);
1127 optp = mtod(m->m_next, u_char *);
1132 optp = (u_char *) (nth + 1);
1138 if (tp->t_flags & TF_RCVD_TSTMP) {
1139 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1140 to.to_tsecr = tp->ts_recent;
1141 to.to_flags |= TOF_TS;
1143 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1144 /* TCP-MD5 (RFC2385). */
1145 if (tp->t_flags & TF_SIGNATURE)
1146 to.to_flags |= TOF_SIGNATURE;
1148 /* Add the options. */
1149 tlen += optlen = tcp_addoptions(&to, optp);
1151 /* Update m_len in the correct mbuf. */
1152 optm->m_len += optlen;
1158 ip6->ip6_vfc = IPV6_VERSION;
1159 ip6->ip6_nxt = IPPROTO_TCP;
1160 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1163 #if defined(INET) && defined(INET6)
1168 ip->ip_len = htons(tlen);
1169 ip->ip_ttl = V_ip_defttl;
1170 if (V_path_mtu_discovery)
1171 ip->ip_off |= htons(IP_DF);
1174 m->m_pkthdr.len = tlen;
1175 m->m_pkthdr.rcvif = NULL;
1179 * Packet is associated with a socket, so allow the
1180 * label of the response to reflect the socket label.
1182 INP_WLOCK_ASSERT(inp);
1183 mac_inpcb_create_mbuf(inp, m);
1186 * Packet is not associated with a socket, so possibly
1187 * update the label in place.
1189 mac_netinet_tcp_reply(m);
1192 nth->th_seq = htonl(seq);
1193 nth->th_ack = htonl(ack);
1195 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1196 nth->th_flags = flags;
1198 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1200 nth->th_win = htons((u_short)win);
1203 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1204 if (to.to_flags & TOF_SIGNATURE) {
1205 if (!TCPMD5_ENABLED() ||
1206 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1213 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1216 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1217 nth->th_sum = in6_cksum_pseudo(ip6,
1218 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1219 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1223 #if defined(INET6) && defined(INET)
1228 m->m_pkthdr.csum_flags = CSUM_TCP;
1229 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1230 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1234 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1235 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1237 TCP_PROBE3(debug__output, tp, th, m);
1239 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1243 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1244 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1247 #if defined(INET) && defined(INET6)
1252 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1253 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1259 * Create a new TCP control block, making an
1260 * empty reassembly queue and hooking it to the argument
1261 * protocol control block. The `inp' parameter must have
1262 * come from the zone allocator set up in tcp_init().
1265 tcp_newtcpcb(struct inpcb *inp)
1267 struct tcpcb_mem *tm;
1270 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1273 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1278 /* Initialise cc_var struct for this tcpcb. */
1280 tp->ccv->type = IPPROTO_TCP;
1281 tp->ccv->ccvc.tcp = tp;
1282 rw_rlock(&tcp_function_lock);
1283 tp->t_fb = tcp_func_set_ptr;
1284 refcount_acquire(&tp->t_fb->tfb_refcnt);
1285 rw_runlock(&tcp_function_lock);
1287 * Use the current system default CC algorithm.
1290 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1291 CC_ALGO(tp) = CC_DEFAULT();
1294 if (CC_ALGO(tp)->cb_init != NULL)
1295 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1296 if (tp->t_fb->tfb_tcp_fb_fini)
1297 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1298 refcount_release(&tp->t_fb->tfb_refcnt);
1299 uma_zfree(V_tcpcb_zone, tm);
1305 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1306 if (tp->t_fb->tfb_tcp_fb_fini)
1307 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1308 refcount_release(&tp->t_fb->tfb_refcnt);
1309 uma_zfree(V_tcpcb_zone, tm);
1315 tp->t_vnet = inp->inp_vnet;
1317 tp->t_timers = &tm->tt;
1318 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1321 isipv6 ? V_tcp_v6mssdflt :
1325 /* Set up our timeouts. */
1326 callout_init(&tp->t_timers->tt_rexmt, 1);
1327 callout_init(&tp->t_timers->tt_persist, 1);
1328 callout_init(&tp->t_timers->tt_keep, 1);
1329 callout_init(&tp->t_timers->tt_2msl, 1);
1330 callout_init(&tp->t_timers->tt_delack, 1);
1332 if (V_tcp_do_rfc1323)
1333 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1335 tp->t_flags |= TF_SACK_PERMIT;
1336 TAILQ_INIT(&tp->snd_holes);
1338 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1341 in_pcbref(inp); /* Reference for tcpcb */
1345 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1346 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1347 * reasonable initial retransmit time.
1349 tp->t_srtt = TCPTV_SRTTBASE;
1350 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1351 tp->t_rttmin = tcp_rexmit_min;
1352 tp->t_rxtcur = TCPTV_RTOBASE;
1353 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1354 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1355 tp->t_rcvtime = ticks;
1357 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1358 * because the socket may be bound to an IPv6 wildcard address,
1359 * which may match an IPv4-mapped IPv6 address.
1361 inp->inp_ip_ttl = V_ip_defttl;
1365 * Init the TCP PCAP queues.
1367 tcp_pcap_tcpcb_init(tp);
1369 if (tp->t_fb->tfb_tcp_fb_init) {
1370 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1372 return (tp); /* XXX */
1376 * Switch the congestion control algorithm back to NewReno for any active
1377 * control blocks using an algorithm which is about to go away.
1378 * This ensures the CC framework can allow the unload to proceed without leaving
1379 * any dangling pointers which would trigger a panic.
1380 * Returning non-zero would inform the CC framework that something went wrong
1381 * and it would be unsafe to allow the unload to proceed. However, there is no
1382 * way for this to occur with this implementation so we always return zero.
1385 tcp_ccalgounload(struct cc_algo *unload_algo)
1387 struct cc_algo *tmpalgo;
1390 VNET_ITERATOR_DECL(vnet_iter);
1393 * Check all active control blocks across all network stacks and change
1394 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1395 * requires cleanup code to be run, call it.
1398 VNET_FOREACH(vnet_iter) {
1399 CURVNET_SET(vnet_iter);
1400 INP_INFO_WLOCK(&V_tcbinfo);
1402 * New connections already part way through being initialised
1403 * with the CC algo we're removing will not race with this code
1404 * because the INP_INFO_WLOCK is held during initialisation. We
1405 * therefore don't enter the loop below until the connection
1406 * list has stabilised.
1408 LIST_FOREACH(inp, &V_tcb, inp_list) {
1410 /* Important to skip tcptw structs. */
1411 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1412 (tp = intotcpcb(inp)) != NULL) {
1414 * By holding INP_WLOCK here, we are assured
1415 * that the connection is not currently
1416 * executing inside the CC module's functions
1417 * i.e. it is safe to make the switch back to
1420 if (CC_ALGO(tp) == unload_algo) {
1421 tmpalgo = CC_ALGO(tp);
1422 /* NewReno does not require any init. */
1423 CC_ALGO(tp) = &newreno_cc_algo;
1424 if (tmpalgo->cb_destroy != NULL)
1425 tmpalgo->cb_destroy(tp->ccv);
1430 INP_INFO_WUNLOCK(&V_tcbinfo);
1433 VNET_LIST_RUNLOCK();
1439 * Drop a TCP connection, reporting
1440 * the specified error. If connection is synchronized,
1441 * then send a RST to peer.
1444 tcp_drop(struct tcpcb *tp, int errno)
1446 struct socket *so = tp->t_inpcb->inp_socket;
1448 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1449 INP_WLOCK_ASSERT(tp->t_inpcb);
1451 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1452 tcp_state_change(tp, TCPS_CLOSED);
1453 (void) tp->t_fb->tfb_tcp_output(tp);
1454 TCPSTAT_INC(tcps_drops);
1456 TCPSTAT_INC(tcps_conndrops);
1457 if (errno == ETIMEDOUT && tp->t_softerror)
1458 errno = tp->t_softerror;
1459 so->so_error = errno;
1460 return (tcp_close(tp));
1464 tcp_discardcb(struct tcpcb *tp)
1466 struct inpcb *inp = tp->t_inpcb;
1467 struct socket *so = inp->inp_socket;
1469 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1473 INP_WLOCK_ASSERT(inp);
1476 * Make sure that all of our timers are stopped before we delete the
1479 * If stopping a timer fails, we schedule a discard function in same
1480 * callout, and the last discard function called will take care of
1481 * deleting the tcpcb.
1483 tp->t_timers->tt_draincnt = 0;
1484 tcp_timer_stop(tp, TT_REXMT);
1485 tcp_timer_stop(tp, TT_PERSIST);
1486 tcp_timer_stop(tp, TT_KEEP);
1487 tcp_timer_stop(tp, TT_2MSL);
1488 tcp_timer_stop(tp, TT_DELACK);
1489 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1491 * Call the stop-all function of the methods,
1492 * this function should call the tcp_timer_stop()
1493 * method with each of the function specific timeouts.
1494 * That stop will be called via the tfb_tcp_timer_stop()
1495 * which should use the async drain function of the
1496 * callout system (see tcp_var.h).
1498 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1502 * If we got enough samples through the srtt filter,
1503 * save the rtt and rttvar in the routing entry.
1504 * 'Enough' is arbitrarily defined as 4 rtt samples.
1505 * 4 samples is enough for the srtt filter to converge
1506 * to within enough % of the correct value; fewer samples
1507 * and we could save a bogus rtt. The danger is not high
1508 * as tcp quickly recovers from everything.
1509 * XXX: Works very well but needs some more statistics!
1511 if (tp->t_rttupdated >= 4) {
1512 struct hc_metrics_lite metrics;
1515 bzero(&metrics, sizeof(metrics));
1517 * Update the ssthresh always when the conditions below
1518 * are satisfied. This gives us better new start value
1519 * for the congestion avoidance for new connections.
1520 * ssthresh is only set if packet loss occurred on a session.
1522 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1523 * being torn down. Ideally this code would not use 'so'.
1525 ssthresh = tp->snd_ssthresh;
1526 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1528 * convert the limit from user data bytes to
1529 * packets then to packet data bytes.
1531 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1534 ssthresh *= (tp->t_maxseg +
1536 (isipv6 ? sizeof (struct ip6_hdr) +
1537 sizeof (struct tcphdr) :
1539 sizeof (struct tcpiphdr)
1546 metrics.rmx_ssthresh = ssthresh;
1548 metrics.rmx_rtt = tp->t_srtt;
1549 metrics.rmx_rttvar = tp->t_rttvar;
1550 metrics.rmx_cwnd = tp->snd_cwnd;
1551 metrics.rmx_sendpipe = 0;
1552 metrics.rmx_recvpipe = 0;
1554 tcp_hc_update(&inp->inp_inc, &metrics);
1557 /* free the reassembly queue, if any */
1558 tcp_reass_flush(tp);
1561 /* Disconnect offload device, if any. */
1562 if (tp->t_flags & TF_TOE)
1563 tcp_offload_detach(tp);
1566 tcp_free_sackholes(tp);
1569 /* Free the TCP PCAP queues. */
1570 tcp_pcap_drain(&(tp->t_inpkts));
1571 tcp_pcap_drain(&(tp->t_outpkts));
1574 /* Allow the CC algorithm to clean up after itself. */
1575 if (CC_ALGO(tp)->cb_destroy != NULL)
1576 CC_ALGO(tp)->cb_destroy(tp->ccv);
1579 khelp_destroy_osd(tp->osd);
1583 inp->inp_ppcb = NULL;
1584 if (tp->t_timers->tt_draincnt == 0) {
1585 /* We own the last reference on tcpcb, let's free it. */
1586 TCPSTATES_DEC(tp->t_state);
1587 if (tp->t_fb->tfb_tcp_fb_fini)
1588 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1589 refcount_release(&tp->t_fb->tfb_refcnt);
1591 uma_zfree(V_tcpcb_zone, tp);
1592 released = in_pcbrele_wlocked(inp);
1593 KASSERT(!released, ("%s: inp %p should not have been released "
1594 "here", __func__, inp));
1599 tcp_timer_discard(void *ptp)
1604 tp = (struct tcpcb *)ptp;
1605 CURVNET_SET(tp->t_vnet);
1606 INP_INFO_RLOCK(&V_tcbinfo);
1608 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1611 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1612 ("%s: tcpcb has to be stopped here", __func__));
1613 tp->t_timers->tt_draincnt--;
1614 if (tp->t_timers->tt_draincnt == 0) {
1615 /* We own the last reference on this tcpcb, let's free it. */
1616 TCPSTATES_DEC(tp->t_state);
1617 if (tp->t_fb->tfb_tcp_fb_fini)
1618 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1619 refcount_release(&tp->t_fb->tfb_refcnt);
1621 uma_zfree(V_tcpcb_zone, tp);
1622 if (in_pcbrele_wlocked(inp)) {
1623 INP_INFO_RUNLOCK(&V_tcbinfo);
1629 INP_INFO_RUNLOCK(&V_tcbinfo);
1634 * Attempt to close a TCP control block, marking it as dropped, and freeing
1635 * the socket if we hold the only reference.
1638 tcp_close(struct tcpcb *tp)
1640 struct inpcb *inp = tp->t_inpcb;
1643 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1644 INP_WLOCK_ASSERT(inp);
1647 if (tp->t_state == TCPS_LISTEN)
1648 tcp_offload_listen_stop(tp);
1652 * This releases the TFO pending counter resource for TFO listen
1653 * sockets as well as passively-created TFO sockets that transition
1654 * from SYN_RECEIVED to CLOSED.
1656 if (tp->t_tfo_pending) {
1657 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1658 tp->t_tfo_pending = NULL;
1662 TCPSTAT_INC(tcps_closed);
1663 if (tp->t_state != TCPS_CLOSED)
1664 tcp_state_change(tp, TCPS_CLOSED);
1665 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1666 so = inp->inp_socket;
1667 soisdisconnected(so);
1668 if (inp->inp_flags & INP_SOCKREF) {
1669 KASSERT(so->so_state & SS_PROTOREF,
1670 ("tcp_close: !SS_PROTOREF"));
1671 inp->inp_flags &= ~INP_SOCKREF;
1674 so->so_state &= ~SS_PROTOREF;
1684 VNET_ITERATOR_DECL(vnet_iter);
1689 VNET_LIST_RLOCK_NOSLEEP();
1690 VNET_FOREACH(vnet_iter) {
1691 CURVNET_SET(vnet_iter);
1696 * Walk the tcpbs, if existing, and flush the reassembly queue,
1697 * if there is one...
1698 * XXX: The "Net/3" implementation doesn't imply that the TCP
1699 * reassembly queue should be flushed, but in a situation
1700 * where we're really low on mbufs, this is potentially
1703 INP_INFO_WLOCK(&V_tcbinfo);
1704 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1705 if (inpb->inp_flags & INP_TIMEWAIT)
1708 if ((tcpb = intotcpcb(inpb)) != NULL) {
1709 tcp_reass_flush(tcpb);
1710 tcp_clean_sackreport(tcpb);
1712 if (tcp_pcap_aggressive_free) {
1713 /* Free the TCP PCAP queues. */
1714 tcp_pcap_drain(&(tcpb->t_inpkts));
1715 tcp_pcap_drain(&(tcpb->t_outpkts));
1721 INP_INFO_WUNLOCK(&V_tcbinfo);
1724 VNET_LIST_RUNLOCK_NOSLEEP();
1728 * Notify a tcp user of an asynchronous error;
1729 * store error as soft error, but wake up user
1730 * (for now, won't do anything until can select for soft error).
1732 * Do not wake up user since there currently is no mechanism for
1733 * reporting soft errors (yet - a kqueue filter may be added).
1735 static struct inpcb *
1736 tcp_notify(struct inpcb *inp, int error)
1740 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1741 INP_WLOCK_ASSERT(inp);
1743 if ((inp->inp_flags & INP_TIMEWAIT) ||
1744 (inp->inp_flags & INP_DROPPED))
1747 tp = intotcpcb(inp);
1748 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1751 * Ignore some errors if we are hooked up.
1752 * If connection hasn't completed, has retransmitted several times,
1753 * and receives a second error, give up now. This is better
1754 * than waiting a long time to establish a connection that
1755 * can never complete.
1757 if (tp->t_state == TCPS_ESTABLISHED &&
1758 (error == EHOSTUNREACH || error == ENETUNREACH ||
1759 error == EHOSTDOWN)) {
1760 if (inp->inp_route.ro_rt) {
1761 RTFREE(inp->inp_route.ro_rt);
1762 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1765 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1767 tp = tcp_drop(tp, error);
1773 tp->t_softerror = error;
1777 wakeup( &so->so_timeo);
1784 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1786 int error, i, m, n, pcb_count;
1787 struct inpcb *inp, **inp_list;
1792 * The process of preparing the TCB list is too time-consuming and
1793 * resource-intensive to repeat twice on every request.
1795 if (req->oldptr == NULL) {
1796 n = V_tcbinfo.ipi_count +
1797 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1798 n += imax(n / 8, 10);
1799 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1803 if (req->newptr != NULL)
1807 * OK, now we're committed to doing something.
1809 INP_LIST_RLOCK(&V_tcbinfo);
1810 gencnt = V_tcbinfo.ipi_gencnt;
1811 n = V_tcbinfo.ipi_count;
1812 INP_LIST_RUNLOCK(&V_tcbinfo);
1814 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1816 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1817 + (n + m) * sizeof(struct xtcpcb));
1821 xig.xig_len = sizeof xig;
1822 xig.xig_count = n + m;
1823 xig.xig_gen = gencnt;
1824 xig.xig_sogen = so_gencnt;
1825 error = SYSCTL_OUT(req, &xig, sizeof xig);
1829 error = syncache_pcblist(req, m, &pcb_count);
1833 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1835 INP_INFO_WLOCK(&V_tcbinfo);
1836 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1837 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1839 if (inp->inp_gencnt <= gencnt) {
1841 * XXX: This use of cr_cansee(), introduced with
1842 * TCP state changes, is not quite right, but for
1843 * now, better than nothing.
1845 if (inp->inp_flags & INP_TIMEWAIT) {
1846 if (intotw(inp) != NULL)
1847 error = cr_cansee(req->td->td_ucred,
1848 intotw(inp)->tw_cred);
1850 error = EINVAL; /* Skip this inp. */
1852 error = cr_canseeinpcb(req->td->td_ucred, inp);
1855 inp_list[i++] = inp;
1860 INP_INFO_WUNLOCK(&V_tcbinfo);
1864 for (i = 0; i < n; i++) {
1867 if (inp->inp_gencnt <= gencnt) {
1870 tcp_inptoxtp(inp, &xt);
1872 error = SYSCTL_OUT(req, &xt, sizeof xt);
1876 INP_INFO_RLOCK(&V_tcbinfo);
1877 for (i = 0; i < n; i++) {
1880 if (!in_pcbrele_rlocked(inp))
1883 INP_INFO_RUNLOCK(&V_tcbinfo);
1887 * Give the user an updated idea of our state.
1888 * If the generation differs from what we told
1889 * her before, she knows that something happened
1890 * while we were processing this request, and it
1891 * might be necessary to retry.
1893 INP_LIST_RLOCK(&V_tcbinfo);
1894 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1895 xig.xig_sogen = so_gencnt;
1896 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1897 INP_LIST_RUNLOCK(&V_tcbinfo);
1898 error = SYSCTL_OUT(req, &xig, sizeof xig);
1900 free(inp_list, M_TEMP);
1904 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1905 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1906 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1910 tcp_getcred(SYSCTL_HANDLER_ARGS)
1913 struct sockaddr_in addrs[2];
1917 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1920 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1923 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1924 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1926 if (inp->inp_socket == NULL)
1929 error = cr_canseeinpcb(req->td->td_ucred, inp);
1931 cru2x(inp->inp_cred, &xuc);
1936 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1940 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1941 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1942 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1947 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1950 struct sockaddr_in6 addrs[2];
1957 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1960 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1963 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1964 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1967 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1969 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1978 inp = in_pcblookup(&V_tcbinfo,
1979 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1981 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1982 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1985 inp = in6_pcblookup(&V_tcbinfo,
1986 &addrs[1].sin6_addr, addrs[1].sin6_port,
1987 &addrs[0].sin6_addr, addrs[0].sin6_port,
1988 INPLOOKUP_RLOCKPCB, NULL);
1990 if (inp->inp_socket == NULL)
1993 error = cr_canseeinpcb(req->td->td_ucred, inp);
1995 cru2x(inp->inp_cred, &xuc);
2000 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2004 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2005 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2006 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2012 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2014 struct ip *ip = vip;
2016 struct in_addr faddr;
2019 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2021 struct in_conninfo inc;
2022 tcp_seq icmp_tcp_seq;
2025 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2026 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2029 if (cmd == PRC_MSGSIZE)
2030 notify = tcp_mtudisc_notify;
2031 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2032 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2033 cmd == PRC_TIMXCEED_INTRANS) && ip)
2034 notify = tcp_drop_syn_sent;
2037 * Hostdead is ugly because it goes linearly through all PCBs.
2038 * XXX: We never get this from ICMP, otherwise it makes an
2039 * excellent DoS attack on machines with many connections.
2041 else if (cmd == PRC_HOSTDEAD)
2043 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2047 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2051 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2052 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2053 INP_INFO_RLOCK(&V_tcbinfo);
2054 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2055 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2056 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2057 /* signal EHOSTDOWN, as it flushes the cached route */
2058 inp = (*notify)(inp, EHOSTDOWN);
2061 icmp_tcp_seq = th->th_seq;
2063 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2064 !(inp->inp_flags & INP_DROPPED) &&
2065 !(inp->inp_socket == NULL)) {
2066 tp = intotcpcb(inp);
2067 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2068 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2069 if (cmd == PRC_MSGSIZE) {
2072 * If we got a needfrag set the MTU
2073 * in the route to the suggested new
2074 * value (if given) and then notify.
2076 mtu = ntohs(icp->icmp_nextmtu);
2078 * If no alternative MTU was
2079 * proposed, try the next smaller
2084 ntohs(ip->ip_len), 1);
2085 if (mtu < V_tcp_minmss +
2086 sizeof(struct tcpiphdr))
2087 mtu = V_tcp_minmss +
2088 sizeof(struct tcpiphdr);
2090 * Only process the offered MTU if it
2091 * is smaller than the current one.
2093 if (mtu < tp->t_maxseg +
2094 sizeof(struct tcpiphdr)) {
2095 bzero(&inc, sizeof(inc));
2096 inc.inc_faddr = faddr;
2098 inp->inp_inc.inc_fibnum;
2099 tcp_hc_updatemtu(&inc, mtu);
2100 tcp_mtudisc(inp, mtu);
2103 inp = (*notify)(inp,
2104 inetctlerrmap[cmd]);
2108 bzero(&inc, sizeof(inc));
2109 inc.inc_fport = th->th_dport;
2110 inc.inc_lport = th->th_sport;
2111 inc.inc_faddr = faddr;
2112 inc.inc_laddr = ip->ip_src;
2113 syncache_unreach(&inc, icmp_tcp_seq);
2118 INP_INFO_RUNLOCK(&V_tcbinfo);
2124 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2126 struct in6_addr *dst;
2127 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2128 struct ip6_hdr *ip6;
2132 struct icmp6_hdr *icmp6;
2133 struct ip6ctlparam *ip6cp = NULL;
2134 const struct sockaddr_in6 *sa6_src = NULL;
2135 struct in_conninfo inc;
2140 tcp_seq icmp_tcp_seq;
2144 if (sa->sa_family != AF_INET6 ||
2145 sa->sa_len != sizeof(struct sockaddr_in6))
2148 /* if the parameter is from icmp6, decode it. */
2150 ip6cp = (struct ip6ctlparam *)d;
2151 icmp6 = ip6cp->ip6c_icmp6;
2153 ip6 = ip6cp->ip6c_ip6;
2154 off = ip6cp->ip6c_off;
2155 sa6_src = ip6cp->ip6c_src;
2156 dst = ip6cp->ip6c_finaldst;
2160 off = 0; /* fool gcc */
2165 if (cmd == PRC_MSGSIZE)
2166 notify = tcp_mtudisc_notify;
2167 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2168 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2169 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2170 notify = tcp_drop_syn_sent;
2173 * Hostdead is ugly because it goes linearly through all PCBs.
2174 * XXX: We never get this from ICMP, otherwise it makes an
2175 * excellent DoS attack on machines with many connections.
2177 else if (cmd == PRC_HOSTDEAD)
2179 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2183 in6_pcbnotify(&V_tcbinfo, sa, 0,
2184 (const struct sockaddr *)sa6_src,
2185 0, cmd, NULL, notify);
2189 /* Check if we can safely get the ports from the tcp hdr */
2192 (int32_t) (off + sizeof(struct tcp_ports)))) {
2195 bzero(&t_ports, sizeof(struct tcp_ports));
2196 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2197 INP_INFO_RLOCK(&V_tcbinfo);
2198 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2199 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2200 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2201 /* signal EHOSTDOWN, as it flushes the cached route */
2202 inp = (*notify)(inp, EHOSTDOWN);
2205 off += sizeof(struct tcp_ports);
2206 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2209 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2211 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2212 !(inp->inp_flags & INP_DROPPED) &&
2213 !(inp->inp_socket == NULL)) {
2214 tp = intotcpcb(inp);
2215 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2216 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2217 if (cmd == PRC_MSGSIZE) {
2220 * If we got a needfrag set the MTU
2221 * in the route to the suggested new
2222 * value (if given) and then notify.
2224 mtu = ntohl(icmp6->icmp6_mtu);
2226 * If no alternative MTU was
2227 * proposed, or the proposed
2228 * MTU was too small, set to
2231 if (mtu < IPV6_MMTU)
2232 mtu = IPV6_MMTU - 8;
2233 bzero(&inc, sizeof(inc));
2234 inc.inc_fibnum = M_GETFIB(m);
2235 inc.inc_flags |= INC_ISIPV6;
2236 inc.inc6_faddr = *dst;
2237 if (in6_setscope(&inc.inc6_faddr,
2238 m->m_pkthdr.rcvif, NULL))
2241 * Only process the offered MTU if it
2242 * is smaller than the current one.
2244 if (mtu < tp->t_maxseg +
2245 sizeof (struct tcphdr) +
2246 sizeof (struct ip6_hdr)) {
2247 tcp_hc_updatemtu(&inc, mtu);
2248 tcp_mtudisc(inp, mtu);
2249 ICMP6STAT_INC(icp6s_pmtuchg);
2252 inp = (*notify)(inp,
2253 inet6ctlerrmap[cmd]);
2257 bzero(&inc, sizeof(inc));
2258 inc.inc_fibnum = M_GETFIB(m);
2259 inc.inc_flags |= INC_ISIPV6;
2260 inc.inc_fport = t_ports.th_dport;
2261 inc.inc_lport = t_ports.th_sport;
2262 inc.inc6_faddr = *dst;
2263 inc.inc6_laddr = ip6->ip6_src;
2264 syncache_unreach(&inc, icmp_tcp_seq);
2269 INP_INFO_RUNLOCK(&V_tcbinfo);
2275 * Following is where TCP initial sequence number generation occurs.
2277 * There are two places where we must use initial sequence numbers:
2278 * 1. In SYN-ACK packets.
2279 * 2. In SYN packets.
2281 * All ISNs for SYN-ACK packets are generated by the syncache. See
2282 * tcp_syncache.c for details.
2284 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2285 * depends on this property. In addition, these ISNs should be
2286 * unguessable so as to prevent connection hijacking. To satisfy
2287 * the requirements of this situation, the algorithm outlined in
2288 * RFC 1948 is used, with only small modifications.
2290 * Implementation details:
2292 * Time is based off the system timer, and is corrected so that it
2293 * increases by one megabyte per second. This allows for proper
2294 * recycling on high speed LANs while still leaving over an hour
2297 * As reading the *exact* system time is too expensive to be done
2298 * whenever setting up a TCP connection, we increment the time
2299 * offset in two ways. First, a small random positive increment
2300 * is added to isn_offset for each connection that is set up.
2301 * Second, the function tcp_isn_tick fires once per clock tick
2302 * and increments isn_offset as necessary so that sequence numbers
2303 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2304 * random positive increments serve only to ensure that the same
2305 * exact sequence number is never sent out twice (as could otherwise
2306 * happen when a port is recycled in less than the system tick
2309 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2310 * between seeding of isn_secret. This is normally set to zero,
2311 * as reseeding should not be necessary.
2313 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2314 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2315 * general, this means holding an exclusive (write) lock.
2318 #define ISN_BYTES_PER_SECOND 1048576
2319 #define ISN_STATIC_INCREMENT 4096
2320 #define ISN_RANDOM_INCREMENT (4096 - 1)
2322 static VNET_DEFINE(u_char, isn_secret[32]);
2323 static VNET_DEFINE(int, isn_last);
2324 static VNET_DEFINE(int, isn_last_reseed);
2325 static VNET_DEFINE(u_int32_t, isn_offset);
2326 static VNET_DEFINE(u_int32_t, isn_offset_old);
2328 #define V_isn_secret VNET(isn_secret)
2329 #define V_isn_last VNET(isn_last)
2330 #define V_isn_last_reseed VNET(isn_last_reseed)
2331 #define V_isn_offset VNET(isn_offset)
2332 #define V_isn_offset_old VNET(isn_offset_old)
2335 tcp_new_isn(struct tcpcb *tp)
2338 u_int32_t md5_buffer[4];
2340 u_int32_t projected_offset;
2342 INP_WLOCK_ASSERT(tp->t_inpcb);
2345 /* Seed if this is the first use, reseed if requested. */
2346 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2347 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2349 read_random(&V_isn_secret, sizeof(V_isn_secret));
2350 V_isn_last_reseed = ticks;
2353 /* Compute the md5 hash and return the ISN. */
2355 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2356 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2358 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2359 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2360 sizeof(struct in6_addr));
2361 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2362 sizeof(struct in6_addr));
2366 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2367 sizeof(struct in_addr));
2368 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2369 sizeof(struct in_addr));
2371 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2372 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2373 new_isn = (tcp_seq) md5_buffer[0];
2374 V_isn_offset += ISN_STATIC_INCREMENT +
2375 (arc4random() & ISN_RANDOM_INCREMENT);
2376 if (ticks != V_isn_last) {
2377 projected_offset = V_isn_offset_old +
2378 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2379 if (SEQ_GT(projected_offset, V_isn_offset))
2380 V_isn_offset = projected_offset;
2381 V_isn_offset_old = V_isn_offset;
2384 new_isn += V_isn_offset;
2390 * When a specific ICMP unreachable message is received and the
2391 * connection state is SYN-SENT, drop the connection. This behavior
2392 * is controlled by the icmp_may_rst sysctl.
2395 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2399 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2400 INP_WLOCK_ASSERT(inp);
2402 if ((inp->inp_flags & INP_TIMEWAIT) ||
2403 (inp->inp_flags & INP_DROPPED))
2406 tp = intotcpcb(inp);
2407 if (tp->t_state != TCPS_SYN_SENT)
2410 tp = tcp_drop(tp, errno);
2418 * When `need fragmentation' ICMP is received, update our idea of the MSS
2419 * based on the new value. Also nudge TCP to send something, since we
2420 * know the packet we just sent was dropped.
2421 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2423 static struct inpcb *
2424 tcp_mtudisc_notify(struct inpcb *inp, int error)
2427 tcp_mtudisc(inp, -1);
2432 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2437 INP_WLOCK_ASSERT(inp);
2438 if ((inp->inp_flags & INP_TIMEWAIT) ||
2439 (inp->inp_flags & INP_DROPPED))
2442 tp = intotcpcb(inp);
2443 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2445 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2447 so = inp->inp_socket;
2448 SOCKBUF_LOCK(&so->so_snd);
2449 /* If the mss is larger than the socket buffer, decrease the mss. */
2450 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2451 tp->t_maxseg = so->so_snd.sb_hiwat;
2452 SOCKBUF_UNLOCK(&so->so_snd);
2454 TCPSTAT_INC(tcps_mturesent);
2456 tp->snd_nxt = tp->snd_una;
2457 tcp_free_sackholes(tp);
2458 tp->snd_recover = tp->snd_max;
2459 if (tp->t_flags & TF_SACK_PERMIT)
2460 EXIT_FASTRECOVERY(tp->t_flags);
2461 tp->t_fb->tfb_tcp_output(tp);
2466 * Look-up the routing entry to the peer of this inpcb. If no route
2467 * is found and it cannot be allocated, then return 0. This routine
2468 * is called by TCP routines that access the rmx structure and by
2469 * tcp_mss_update to get the peer/interface MTU.
2472 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2474 struct nhop4_extended nh4;
2476 uint32_t maxmtu = 0;
2478 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2480 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2482 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2483 NHR_REF, 0, &nh4) != 0)
2487 maxmtu = nh4.nh_mtu;
2489 /* Report additional interface capabilities. */
2491 if (ifp->if_capenable & IFCAP_TSO4 &&
2492 ifp->if_hwassist & CSUM_TSO) {
2493 cap->ifcap |= CSUM_TSO;
2494 cap->tsomax = ifp->if_hw_tsomax;
2495 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2496 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2499 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2507 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2509 struct nhop6_extended nh6;
2510 struct in6_addr dst6;
2513 uint32_t maxmtu = 0;
2515 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2517 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2518 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2519 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2524 maxmtu = nh6.nh_mtu;
2526 /* Report additional interface capabilities. */
2528 if (ifp->if_capenable & IFCAP_TSO6 &&
2529 ifp->if_hwassist & CSUM_TSO) {
2530 cap->ifcap |= CSUM_TSO;
2531 cap->tsomax = ifp->if_hw_tsomax;
2532 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2533 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2536 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2544 * Calculate effective SMSS per RFC5681 definition for a given TCP
2545 * connection at its current state, taking into account SACK and etc.
2548 tcp_maxseg(const struct tcpcb *tp)
2552 if (tp->t_flags & TF_NOOPT)
2553 return (tp->t_maxseg);
2556 * Here we have a simplified code from tcp_addoptions(),
2557 * without a proper loop, and having most of paddings hardcoded.
2558 * We might make mistakes with padding here in some edge cases,
2559 * but this is harmless, since result of tcp_maxseg() is used
2560 * only in cwnd and ssthresh estimations.
2562 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2563 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2564 if (tp->t_flags & TF_RCVD_TSTMP)
2565 optlen = TCPOLEN_TSTAMP_APPA;
2568 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2569 if (tp->t_flags & TF_SIGNATURE)
2570 optlen += PAD(TCPOLEN_SIGNATURE);
2572 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2573 optlen += TCPOLEN_SACKHDR;
2574 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2575 optlen = PAD(optlen);
2578 if (tp->t_flags & TF_REQ_TSTMP)
2579 optlen = TCPOLEN_TSTAMP_APPA;
2581 optlen = PAD(TCPOLEN_MAXSEG);
2582 if (tp->t_flags & TF_REQ_SCALE)
2583 optlen += PAD(TCPOLEN_WINDOW);
2584 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2585 if (tp->t_flags & TF_SIGNATURE)
2586 optlen += PAD(TCPOLEN_SIGNATURE);
2588 if (tp->t_flags & TF_SACK_PERMIT)
2589 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2592 optlen = min(optlen, TCP_MAXOLEN);
2593 return (tp->t_maxseg - optlen);
2597 sysctl_drop(SYSCTL_HANDLER_ARGS)
2599 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2600 struct sockaddr_storage addrs[2];
2604 struct sockaddr_in *fin, *lin;
2606 struct sockaddr_in6 *fin6, *lin6;
2617 if (req->oldptr != NULL || req->oldlen != 0)
2619 if (req->newptr == NULL)
2621 if (req->newlen < sizeof(addrs))
2623 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2627 switch (addrs[0].ss_family) {
2630 fin6 = (struct sockaddr_in6 *)&addrs[0];
2631 lin6 = (struct sockaddr_in6 *)&addrs[1];
2632 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2633 lin6->sin6_len != sizeof(struct sockaddr_in6))
2635 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2636 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2638 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2639 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2640 fin = (struct sockaddr_in *)&addrs[0];
2641 lin = (struct sockaddr_in *)&addrs[1];
2644 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2647 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2654 fin = (struct sockaddr_in *)&addrs[0];
2655 lin = (struct sockaddr_in *)&addrs[1];
2656 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2657 lin->sin_len != sizeof(struct sockaddr_in))
2664 INP_INFO_RLOCK(&V_tcbinfo);
2665 switch (addrs[0].ss_family) {
2668 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2669 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2670 INPLOOKUP_WLOCKPCB, NULL);
2675 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2676 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2681 if (inp->inp_flags & INP_TIMEWAIT) {
2683 * XXXRW: There currently exists a state where an
2684 * inpcb is present, but its timewait state has been
2685 * discarded. For now, don't allow dropping of this
2693 } else if (!(inp->inp_flags & INP_DROPPED) &&
2694 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2695 tp = intotcpcb(inp);
2696 tp = tcp_drop(tp, ECONNABORTED);
2703 INP_INFO_RUNLOCK(&V_tcbinfo);
2707 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2708 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2709 0, sysctl_drop, "", "Drop TCP connection");
2712 * Generate a standardized TCP log line for use throughout the
2713 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2714 * allow use in the interrupt context.
2716 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2717 * NB: The function may return NULL if memory allocation failed.
2719 * Due to header inclusion and ordering limitations the struct ip
2720 * and ip6_hdr pointers have to be passed as void pointers.
2723 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2727 /* Is logging enabled? */
2728 if (tcp_log_in_vain == 0)
2731 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2735 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2739 /* Is logging enabled? */
2740 if (tcp_log_debug == 0)
2743 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2747 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2754 const struct ip6_hdr *ip6;
2756 ip6 = (const struct ip6_hdr *)ip6hdr;
2758 ip = (struct ip *)ip4hdr;
2761 * The log line looks like this:
2762 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2764 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2765 sizeof(PRINT_TH_FLAGS) + 1 +
2767 2 * INET6_ADDRSTRLEN;
2769 2 * INET_ADDRSTRLEN;
2772 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2776 strcat(s, "TCP: [");
2779 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2780 inet_ntoa_r(inc->inc_faddr, sp);
2782 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2784 inet_ntoa_r(inc->inc_laddr, sp);
2786 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2789 ip6_sprintf(sp, &inc->inc6_faddr);
2791 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2793 ip6_sprintf(sp, &inc->inc6_laddr);
2795 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2796 } else if (ip6 && th) {
2797 ip6_sprintf(sp, &ip6->ip6_src);
2799 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2801 ip6_sprintf(sp, &ip6->ip6_dst);
2803 sprintf(sp, "]:%i", ntohs(th->th_dport));
2806 } else if (ip && th) {
2807 inet_ntoa_r(ip->ip_src, sp);
2809 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2811 inet_ntoa_r(ip->ip_dst, sp);
2813 sprintf(sp, "]:%i", ntohs(th->th_dport));
2821 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2822 if (*(s + size - 1) != '\0')
2823 panic("%s: string too long", __func__);
2828 * A subroutine which makes it easy to track TCP state changes with DTrace.
2829 * This function shouldn't be called for t_state initializations that don't
2830 * correspond to actual TCP state transitions.
2833 tcp_state_change(struct tcpcb *tp, int newstate)
2835 #if defined(KDTRACE_HOOKS)
2836 int pstate = tp->t_state;
2839 TCPSTATES_DEC(tp->t_state);
2840 TCPSTATES_INC(newstate);
2841 tp->t_state = newstate;
2842 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2846 * Create an external-format (``xtcpcb'') structure using the information in
2847 * the kernel-format tcpcb structure pointed to by tp. This is done to
2848 * reduce the spew of irrelevant information over this interface, to isolate
2849 * user code from changes in the kernel structure, and potentially to provide
2850 * information-hiding if we decide that some of this information should be
2851 * hidden from users.
2854 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2856 struct tcpcb *tp = intotcpcb(inp);
2859 if (inp->inp_flags & INP_TIMEWAIT) {
2860 bzero(xt, sizeof(struct xtcpcb));
2861 xt->t_state = TCPS_TIME_WAIT;
2863 xt->t_state = tp->t_state;
2864 xt->t_flags = tp->t_flags;
2865 xt->t_sndzerowin = tp->t_sndzerowin;
2866 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2867 xt->t_rcvoopack = tp->t_rcvoopack;
2869 now = getsbinuptime();
2870 #define COPYTIMER(ttt) do { \
2871 if (callout_active(&tp->t_timers->ttt)) \
2872 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
2877 COPYTIMER(tt_delack);
2878 COPYTIMER(tt_rexmt);
2879 COPYTIMER(tt_persist);
2883 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2885 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2886 TCP_FUNCTION_NAME_LEN_MAX);
2889 xt->xt_len = sizeof(struct xtcpcb);
2890 in_pcbtoxinpcb(inp, &xt->xt_inp);
2891 if (inp->inp_socket == NULL)
2892 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;