2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
46 #include <sys/hhook.h>
48 #include <sys/kernel.h>
50 #include <sys/khelp.h>
52 #include <sys/sysctl.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
58 #include <sys/domain.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/protosw.h>
66 #include <sys/random.h>
70 #include <net/route.h>
72 #include <net/if_var.h>
75 #include <netinet/in.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_kdtrace.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_icmp.h>
83 #include <netinet/ip_var.h>
85 #include <netinet/icmp6.h>
86 #include <netinet/ip6.h>
87 #include <netinet6/in6_fib.h>
88 #include <netinet6/in6_pcb.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/scope6_var.h>
91 #include <netinet6/nd6.h>
95 #include <netinet/tcp_fastopen.h>
97 #include <netinet/tcp.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 #include <netinet/tcp_syncache.h>
103 #include <netinet/cc/cc.h>
105 #include <netinet6/tcp6_var.h>
107 #include <netinet/tcpip.h>
109 #include <netinet/tcp_pcap.h>
112 #include <netinet/tcp_debug.h>
115 #include <netinet6/ip6protosw.h>
118 #include <netinet/tcp_offload.h>
121 #include <netipsec/ipsec_support.h>
123 #include <machine/in_cksum.h>
126 #include <security/mac/mac_framework.h>
128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
133 struct rwlock tcp_function_lock;
136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
141 error = sysctl_handle_int(oidp, &new, 0, req);
142 if (error == 0 && req->newptr) {
143 if (new < TCP_MINMSS)
151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
152 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
153 &sysctl_net_inet_tcp_mss_check, "I",
154 "Default TCP Maximum Segment Size");
158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
162 new = V_tcp_v6mssdflt;
163 error = sysctl_handle_int(oidp, &new, 0, req);
164 if (error == 0 && req->newptr) {
165 if (new < TCP_MINMSS)
168 V_tcp_v6mssdflt = new;
173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
175 &sysctl_net_inet_tcp_mss_v6_check, "I",
176 "Default TCP Maximum Segment Size for IPv6");
180 * Minimum MSS we accept and use. This prevents DoS attacks where
181 * we are forced to a ridiculous low MSS like 20 and send hundreds
182 * of packets instead of one. The effect scales with the available
183 * bandwidth and quickly saturates the CPU and network interface
184 * with packet generation and sending. Set to zero to disable MINMSS
185 * checking. This setting prevents us from sending too small packets.
187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_minmss), 0,
190 "Minimum TCP Maximum Segment Size");
192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_do_rfc1323), 0,
195 "Enable rfc1323 (high performance TCP) extensions");
197 static int tcp_log_debug = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 static int tcp_tcbhashsize;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 static int do_tcpdrain = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
207 "Enable tcp_drain routine for extra help when low on mbufs");
209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 static VNET_DEFINE(int, icmp_may_rst) = 1;
213 #define V_icmp_may_rst VNET(icmp_may_rst)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(icmp_may_rst), 0,
216 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_isn_reseed_interval), 0,
222 "Seconds between reseeding of ISN secret");
224 static int tcp_soreceive_stream;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 VNET_DEFINE(uma_zone_t, sack_hole_zone);
229 #define V_sack_hole_zone VNET(sack_hole_zone)
232 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
235 static struct inpcb *tcp_notify(struct inpcb *, int);
236 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
237 static void tcp_mtudisc(struct inpcb *, int);
238 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
239 void *ip4hdr, const void *ip6hdr);
242 static struct tcp_function_block tcp_def_funcblk = {
246 tcp_default_ctloutput,
257 int t_functions_inited = 0;
258 struct tcp_funchead t_functions;
259 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
262 init_tcp_functions(void)
264 if (t_functions_inited == 0) {
265 TAILQ_INIT(&t_functions);
266 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
267 t_functions_inited = 1;
271 static struct tcp_function_block *
272 find_tcp_functions_locked(struct tcp_function_set *fs)
274 struct tcp_function *f;
275 struct tcp_function_block *blk=NULL;
277 TAILQ_FOREACH(f, &t_functions, tf_next) {
278 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
286 static struct tcp_function_block *
287 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
289 struct tcp_function_block *rblk=NULL;
290 struct tcp_function *f;
292 TAILQ_FOREACH(f, &t_functions, tf_next) {
293 if (f->tf_fb == blk) {
304 struct tcp_function_block *
305 find_and_ref_tcp_functions(struct tcp_function_set *fs)
307 struct tcp_function_block *blk;
309 rw_rlock(&tcp_function_lock);
310 blk = find_tcp_functions_locked(fs);
312 refcount_acquire(&blk->tfb_refcnt);
313 rw_runlock(&tcp_function_lock);
317 struct tcp_function_block *
318 find_and_ref_tcp_fb(struct tcp_function_block *blk)
320 struct tcp_function_block *rblk;
322 rw_rlock(&tcp_function_lock);
323 rblk = find_tcp_fb_locked(blk, NULL);
325 refcount_acquire(&rblk->tfb_refcnt);
326 rw_runlock(&tcp_function_lock);
332 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
335 struct tcp_function_set fs;
336 struct tcp_function_block *blk;
338 memset(&fs, 0, sizeof(fs));
339 rw_rlock(&tcp_function_lock);
340 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
343 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
344 fs.pcbcnt = blk->tfb_refcnt;
346 rw_runlock(&tcp_function_lock);
347 error = sysctl_handle_string(oidp, fs.function_set_name,
348 sizeof(fs.function_set_name), req);
350 /* Check for error or no change */
351 if (error != 0 || req->newptr == NULL)
354 rw_wlock(&tcp_function_lock);
355 blk = find_tcp_functions_locked(&fs);
357 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
361 tcp_func_set_ptr = blk;
363 rw_wunlock(&tcp_function_lock);
367 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
368 CTLTYPE_STRING | CTLFLAG_RW,
369 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
370 "Set/get the default TCP functions");
373 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
375 int error, cnt, linesz;
376 struct tcp_function *f;
382 rw_rlock(&tcp_function_lock);
383 TAILQ_FOREACH(f, &t_functions, tf_next) {
386 rw_runlock(&tcp_function_lock);
388 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
389 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
394 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
395 "Alias", "PCB count");
400 rw_rlock(&tcp_function_lock);
401 TAILQ_FOREACH(f, &t_functions, tf_next) {
402 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
403 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
404 f->tf_fb->tfb_tcp_block_name,
405 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
406 alias ? f->tf_name : "-",
407 f->tf_fb->tfb_refcnt);
408 if (linesz >= bufsz) {
416 rw_runlock(&tcp_function_lock);
418 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
419 free(buffer, M_TEMP);
423 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
424 CTLTYPE_STRING|CTLFLAG_RD,
425 NULL, 0, sysctl_net_inet_list_available, "A",
426 "list available TCP Function sets");
429 * Target size of TCP PCB hash tables. Must be a power of two.
431 * Note that this can be overridden by the kernel environment
432 * variable net.inet.tcp.tcbhashsize
435 #define TCBHASHSIZE 0
440 * Callouts should be moved into struct tcp directly. They are currently
441 * separate because the tcpcb structure is exported to userland for sysctl
442 * parsing purposes, which do not know about callouts.
453 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
454 #define V_tcpcb_zone VNET(tcpcb_zone)
456 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
457 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
459 static struct mtx isn_mtx;
461 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
462 #define ISN_LOCK() mtx_lock(&isn_mtx)
463 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
466 * TCP initialization.
469 tcp_zone_change(void *tag)
472 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
473 uma_zone_set_max(V_tcpcb_zone, maxsockets);
474 tcp_tw_zone_change();
478 tcp_inpcb_init(void *mem, int size, int flags)
480 struct inpcb *inp = mem;
482 INP_LOCK_INIT(inp, "inp", "tcpinp");
487 * Take a value and get the next power of 2 that doesn't overflow.
488 * Used to size the tcp_inpcb hash buckets.
491 maketcp_hashsize(int size)
497 * get the next power of 2 higher than maxsockets.
499 hashsize = 1 << fls(size);
500 /* catch overflow, and just go one power of 2 smaller */
501 if (hashsize < size) {
502 hashsize = 1 << (fls(size) - 1);
508 * Register a TCP function block with the name provided in the names
509 * array. (Note that this function does NOT automatically register
510 * blk->tfb_tcp_block_name as a stack name. Therefore, you should
511 * explicitly include blk->tfb_tcp_block_name in the list of names if
512 * you wish to register the stack with that name.)
514 * Either all name registrations will succeed or all will fail. If
515 * a name registration fails, the function will update the num_names
516 * argument to point to the array index of the name that encountered
519 * Returns 0 on success, or an error code on failure.
522 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
523 const char *names[], int *num_names)
525 struct tcp_function *n;
526 struct tcp_function_set fs;
529 KASSERT(names != NULL && *num_names > 0,
530 ("%s: Called with 0-length name list", __func__));
531 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
533 if (t_functions_inited == 0) {
534 init_tcp_functions();
536 if ((blk->tfb_tcp_output == NULL) ||
537 (blk->tfb_tcp_do_segment == NULL) ||
538 (blk->tfb_tcp_ctloutput == NULL) ||
539 (strlen(blk->tfb_tcp_block_name) == 0)) {
541 * These functions are required and you
547 if (blk->tfb_tcp_timer_stop_all ||
548 blk->tfb_tcp_timer_activate ||
549 blk->tfb_tcp_timer_active ||
550 blk->tfb_tcp_timer_stop) {
552 * If you define one timer function you
553 * must have them all.
555 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
556 (blk->tfb_tcp_timer_activate == NULL) ||
557 (blk->tfb_tcp_timer_active == NULL) ||
558 (blk->tfb_tcp_timer_stop == NULL)) {
564 refcount_init(&blk->tfb_refcnt, 0);
566 for (i = 0; i < *num_names; i++) {
567 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
574 (void)strncpy(fs.function_set_name, names[i],
575 TCP_FUNCTION_NAME_LEN_MAX);
576 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
577 rw_wlock(&tcp_function_lock);
578 if (find_tcp_functions_locked(&fs) != NULL) {
579 /* Duplicate name space not allowed */
580 rw_wunlock(&tcp_function_lock);
581 free(n, M_TCPFUNCTIONS);
585 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
586 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
587 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
588 rw_wunlock(&tcp_function_lock);
594 * Deregister the names we just added. Because registration failed
595 * for names[i], we don't need to deregister that name.
598 rw_wlock(&tcp_function_lock);
600 TAILQ_FOREACH(n, &t_functions, tf_next) {
601 if (!strncmp(n->tf_name, names[i],
602 TCP_FUNCTION_NAME_LEN_MAX)) {
603 TAILQ_REMOVE(&t_functions, n, tf_next);
605 free(n, M_TCPFUNCTIONS);
610 rw_wunlock(&tcp_function_lock);
615 * Register a TCP function block using the name provided in the name
618 * Returns 0 on success, or an error code on failure.
621 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
624 const char *name_list[1];
631 name_list[0] = blk->tfb_tcp_block_name;
632 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
637 * Register a TCP function block using the name defined in
638 * blk->tfb_tcp_block_name.
640 * Returns 0 on success, or an error code on failure.
643 register_tcp_functions(struct tcp_function_block *blk, int wait)
646 return (register_tcp_functions_as_name(blk, NULL, wait));
650 deregister_tcp_functions(struct tcp_function_block *blk)
652 struct tcp_function *f;
655 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
656 /* You can't un-register the default */
659 rw_wlock(&tcp_function_lock);
660 if (blk == tcp_func_set_ptr) {
661 /* You can't free the current default */
662 rw_wunlock(&tcp_function_lock);
665 if (blk->tfb_refcnt) {
666 /* Still tcb attached, mark it. */
667 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
668 rw_wunlock(&tcp_function_lock);
671 while (find_tcp_fb_locked(blk, &f) != NULL) {
673 TAILQ_REMOVE(&t_functions, f, tf_next);
675 free(f, M_TCPFUNCTIONS);
678 rw_wunlock(&tcp_function_lock);
685 const char *tcbhash_tuneable;
688 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
691 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
692 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
693 printf("%s: WARNING: unable to register helper hook\n", __func__);
694 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
695 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
696 printf("%s: WARNING: unable to register helper hook\n", __func__);
698 hashsize = TCBHASHSIZE;
699 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
702 * Auto tune the hash size based on maxsockets.
703 * A perfect hash would have a 1:1 mapping
704 * (hashsize = maxsockets) however it's been
705 * suggested that O(2) average is better.
707 hashsize = maketcp_hashsize(maxsockets / 4);
709 * Our historical default is 512,
710 * do not autotune lower than this.
714 if (bootverbose && IS_DEFAULT_VNET(curvnet))
715 printf("%s: %s auto tuned to %d\n", __func__,
716 tcbhash_tuneable, hashsize);
719 * We require a hashsize to be a power of two.
720 * Previously if it was not a power of two we would just reset it
721 * back to 512, which could be a nasty surprise if you did not notice
723 * Instead what we do is clip it to the closest power of two lower
724 * than the specified hash value.
726 if (!powerof2(hashsize)) {
727 int oldhashsize = hashsize;
729 hashsize = maketcp_hashsize(hashsize);
730 /* prevent absurdly low value */
733 printf("%s: WARNING: TCB hash size not a power of 2, "
734 "clipped from %d to %d.\n", __func__, oldhashsize,
737 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
738 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
741 * These have to be type stable for the benefit of the timers.
743 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
744 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
745 uma_zone_set_max(V_tcpcb_zone, maxsockets);
746 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
752 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
753 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
754 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
760 /* Skip initialization of globals for non-default instances. */
761 if (!IS_DEFAULT_VNET(curvnet))
764 tcp_reass_global_init();
766 /* XXX virtualize those bellow? */
767 tcp_delacktime = TCPTV_DELACK;
768 tcp_keepinit = TCPTV_KEEP_INIT;
769 tcp_keepidle = TCPTV_KEEP_IDLE;
770 tcp_keepintvl = TCPTV_KEEPINTVL;
771 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
773 tcp_rexmit_min = TCPTV_MIN;
774 if (tcp_rexmit_min < 1)
776 tcp_persmin = TCPTV_PERSMIN;
777 tcp_persmax = TCPTV_PERSMAX;
778 tcp_rexmit_slop = TCPTV_CPU_VAR;
779 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
780 tcp_tcbhashsize = hashsize;
781 /* Setup the tcp function block list */
782 init_tcp_functions();
783 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
785 if (tcp_soreceive_stream) {
787 tcp_usrreqs.pru_soreceive = soreceive_stream;
790 tcp6_usrreqs.pru_soreceive = soreceive_stream;
795 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
797 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
799 if (max_protohdr < TCP_MINPROTOHDR)
800 max_protohdr = TCP_MINPROTOHDR;
801 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
803 #undef TCP_MINPROTOHDR
806 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
807 SHUTDOWN_PRI_DEFAULT);
808 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
809 EVENTHANDLER_PRI_ANY);
817 tcp_destroy(void *unused __unused)
825 * All our processes are gone, all our sockets should be cleaned
826 * up, which means, we should be past the tcp_discardcb() calls.
827 * Sleep to let all tcpcb timers really disappear and cleanup.
830 INP_LIST_RLOCK(&V_tcbinfo);
831 n = V_tcbinfo.ipi_count;
832 INP_LIST_RUNLOCK(&V_tcbinfo);
835 pause("tcpdes", hz / 10);
840 in_pcbinfo_destroy(&V_tcbinfo);
841 /* tcp_discardcb() clears the sack_holes up. */
842 uma_zdestroy(V_sack_hole_zone);
843 uma_zdestroy(V_tcpcb_zone);
847 * Cannot free the zone until all tcpcbs are released as we attach
848 * the allocations to them.
850 tcp_fastopen_destroy();
854 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
856 printf("%s: WARNING: unable to deregister helper hook "
857 "type=%d, id=%d: error %d returned\n", __func__,
858 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
860 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
862 printf("%s: WARNING: unable to deregister helper hook "
863 "type=%d, id=%d: error %d returned\n", __func__,
864 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
868 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
878 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
879 * tcp_template used to store this data in mbufs, but we now recopy it out
880 * of the tcpcb each time to conserve mbufs.
883 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
885 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
887 INP_WLOCK_ASSERT(inp);
890 if ((inp->inp_vflag & INP_IPV6) != 0) {
893 ip6 = (struct ip6_hdr *)ip_ptr;
894 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
895 (inp->inp_flow & IPV6_FLOWINFO_MASK);
896 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
897 (IPV6_VERSION & IPV6_VERSION_MASK);
898 ip6->ip6_nxt = IPPROTO_TCP;
899 ip6->ip6_plen = htons(sizeof(struct tcphdr));
900 ip6->ip6_src = inp->in6p_laddr;
901 ip6->ip6_dst = inp->in6p_faddr;
904 #if defined(INET6) && defined(INET)
911 ip = (struct ip *)ip_ptr;
912 ip->ip_v = IPVERSION;
914 ip->ip_tos = inp->inp_ip_tos;
918 ip->ip_ttl = inp->inp_ip_ttl;
920 ip->ip_p = IPPROTO_TCP;
921 ip->ip_src = inp->inp_laddr;
922 ip->ip_dst = inp->inp_faddr;
925 th->th_sport = inp->inp_lport;
926 th->th_dport = inp->inp_fport;
934 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
938 * Create template to be used to send tcp packets on a connection.
939 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
940 * use for this function is in keepalives, which use tcp_respond.
943 tcpip_maketemplate(struct inpcb *inp)
947 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
950 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
955 * Send a single message to the TCP at address specified by
956 * the given TCP/IP header. If m == NULL, then we make a copy
957 * of the tcpiphdr at th and send directly to the addressed host.
958 * This is used to force keep alive messages out using the TCP
959 * template for a connection. If flags are given then we send
960 * a message back to the TCP which originated the segment th,
961 * and discard the mbuf containing it and any other attached mbufs.
963 * In any case the ack and sequence number of the transmitted
964 * segment are as specified by the parameters.
966 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
969 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
970 tcp_seq ack, tcp_seq seq, int flags)
982 int optlen, tlen, win;
985 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
988 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
995 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
996 INP_WLOCK_ASSERT(inp);
1003 if (!(flags & TH_RST)) {
1004 win = sbspace(&inp->inp_socket->so_rcv);
1005 if (win > TCP_MAXWIN << tp->rcv_scale)
1006 win = TCP_MAXWIN << tp->rcv_scale;
1008 if ((tp->t_flags & TF_NOOPT) == 0)
1012 m = m_gethdr(M_NOWAIT, MT_DATA);
1015 m->m_data += max_linkhdr;
1018 bcopy((caddr_t)ip6, mtod(m, caddr_t),
1019 sizeof(struct ip6_hdr));
1020 ip6 = mtod(m, struct ip6_hdr *);
1021 nth = (struct tcphdr *)(ip6 + 1);
1025 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1026 ip = mtod(m, struct ip *);
1027 nth = (struct tcphdr *)(ip + 1);
1029 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1031 } else if (!M_WRITABLE(m)) {
1034 /* Can't reuse 'm', allocate a new mbuf. */
1035 n = m_gethdr(M_NOWAIT, MT_DATA);
1041 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1047 n->m_data += max_linkhdr;
1048 /* m_len is set later */
1049 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1052 bcopy((caddr_t)ip6, mtod(n, caddr_t),
1053 sizeof(struct ip6_hdr));
1054 ip6 = mtod(n, struct ip6_hdr *);
1055 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1056 nth = (struct tcphdr *)(ip6 + 1);
1060 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1061 ip = mtod(n, struct ip *);
1062 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1063 nth = (struct tcphdr *)(ip + 1);
1065 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1066 xchg(nth->th_dport, nth->th_sport, uint16_t);
1073 * XXX MRT We inherit the FIB, which is lucky.
1077 m->m_data = (caddr_t)ipgen;
1078 /* m_len is set later */
1081 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1082 nth = (struct tcphdr *)(ip6 + 1);
1086 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1087 nth = (struct tcphdr *)(ip + 1);
1091 * this is usually a case when an extension header
1092 * exists between the IPv6 header and the
1095 nth->th_sport = th->th_sport;
1096 nth->th_dport = th->th_dport;
1098 xchg(nth->th_dport, nth->th_sport, uint16_t);
1104 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1106 #if defined(INET) && defined(INET6)
1110 tlen = sizeof (struct tcpiphdr);
1114 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1115 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1116 m, tlen, (long)M_TRAILINGSPACE(m)));
1121 /* Make sure we have room. */
1122 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1123 m->m_next = m_get(M_NOWAIT, MT_DATA);
1125 optp = mtod(m->m_next, u_char *);
1130 optp = (u_char *) (nth + 1);
1136 if (tp->t_flags & TF_RCVD_TSTMP) {
1137 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1138 to.to_tsecr = tp->ts_recent;
1139 to.to_flags |= TOF_TS;
1141 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1142 /* TCP-MD5 (RFC2385). */
1143 if (tp->t_flags & TF_SIGNATURE)
1144 to.to_flags |= TOF_SIGNATURE;
1146 /* Add the options. */
1147 tlen += optlen = tcp_addoptions(&to, optp);
1149 /* Update m_len in the correct mbuf. */
1150 optm->m_len += optlen;
1156 ip6->ip6_vfc = IPV6_VERSION;
1157 ip6->ip6_nxt = IPPROTO_TCP;
1158 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1161 #if defined(INET) && defined(INET6)
1166 ip->ip_len = htons(tlen);
1167 ip->ip_ttl = V_ip_defttl;
1168 if (V_path_mtu_discovery)
1169 ip->ip_off |= htons(IP_DF);
1172 m->m_pkthdr.len = tlen;
1173 m->m_pkthdr.rcvif = NULL;
1177 * Packet is associated with a socket, so allow the
1178 * label of the response to reflect the socket label.
1180 INP_WLOCK_ASSERT(inp);
1181 mac_inpcb_create_mbuf(inp, m);
1184 * Packet is not associated with a socket, so possibly
1185 * update the label in place.
1187 mac_netinet_tcp_reply(m);
1190 nth->th_seq = htonl(seq);
1191 nth->th_ack = htonl(ack);
1193 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1194 nth->th_flags = flags;
1196 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1198 nth->th_win = htons((u_short)win);
1201 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1202 if (to.to_flags & TOF_SIGNATURE) {
1203 if (!TCPMD5_ENABLED() ||
1204 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1211 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1214 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1215 nth->th_sum = in6_cksum_pseudo(ip6,
1216 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1217 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1221 #if defined(INET6) && defined(INET)
1226 m->m_pkthdr.csum_flags = CSUM_TCP;
1227 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1228 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1232 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1233 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1235 TCP_PROBE3(debug__output, tp, th, m);
1237 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1239 TCP_PROBE5(send, NULL, tp, m, tp, nth);
1242 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1244 #if defined(INET) && defined(INET6)
1248 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1253 * Create a new TCP control block, making an
1254 * empty reassembly queue and hooking it to the argument
1255 * protocol control block. The `inp' parameter must have
1256 * come from the zone allocator set up in tcp_init().
1259 tcp_newtcpcb(struct inpcb *inp)
1261 struct tcpcb_mem *tm;
1264 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1267 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1272 /* Initialise cc_var struct for this tcpcb. */
1274 tp->ccv->type = IPPROTO_TCP;
1275 tp->ccv->ccvc.tcp = tp;
1276 rw_rlock(&tcp_function_lock);
1277 tp->t_fb = tcp_func_set_ptr;
1278 refcount_acquire(&tp->t_fb->tfb_refcnt);
1279 rw_runlock(&tcp_function_lock);
1281 * Use the current system default CC algorithm.
1284 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1285 CC_ALGO(tp) = CC_DEFAULT();
1288 if (CC_ALGO(tp)->cb_init != NULL)
1289 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1290 if (tp->t_fb->tfb_tcp_fb_fini)
1291 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1292 refcount_release(&tp->t_fb->tfb_refcnt);
1293 uma_zfree(V_tcpcb_zone, tm);
1299 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1300 if (tp->t_fb->tfb_tcp_fb_fini)
1301 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1302 refcount_release(&tp->t_fb->tfb_refcnt);
1303 uma_zfree(V_tcpcb_zone, tm);
1309 tp->t_vnet = inp->inp_vnet;
1311 tp->t_timers = &tm->tt;
1312 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1315 isipv6 ? V_tcp_v6mssdflt :
1319 /* Set up our timeouts. */
1320 callout_init(&tp->t_timers->tt_rexmt, 1);
1321 callout_init(&tp->t_timers->tt_persist, 1);
1322 callout_init(&tp->t_timers->tt_keep, 1);
1323 callout_init(&tp->t_timers->tt_2msl, 1);
1324 callout_init(&tp->t_timers->tt_delack, 1);
1326 if (V_tcp_do_rfc1323)
1327 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1329 tp->t_flags |= TF_SACK_PERMIT;
1330 TAILQ_INIT(&tp->snd_holes);
1332 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1335 in_pcbref(inp); /* Reference for tcpcb */
1339 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1340 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1341 * reasonable initial retransmit time.
1343 tp->t_srtt = TCPTV_SRTTBASE;
1344 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1345 tp->t_rttmin = tcp_rexmit_min;
1346 tp->t_rxtcur = TCPTV_RTOBASE;
1347 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1348 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1349 tp->t_rcvtime = ticks;
1351 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1352 * because the socket may be bound to an IPv6 wildcard address,
1353 * which may match an IPv4-mapped IPv6 address.
1355 inp->inp_ip_ttl = V_ip_defttl;
1359 * Init the TCP PCAP queues.
1361 tcp_pcap_tcpcb_init(tp);
1363 if (tp->t_fb->tfb_tcp_fb_init) {
1364 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1366 return (tp); /* XXX */
1370 * Switch the congestion control algorithm back to NewReno for any active
1371 * control blocks using an algorithm which is about to go away.
1372 * This ensures the CC framework can allow the unload to proceed without leaving
1373 * any dangling pointers which would trigger a panic.
1374 * Returning non-zero would inform the CC framework that something went wrong
1375 * and it would be unsafe to allow the unload to proceed. However, there is no
1376 * way for this to occur with this implementation so we always return zero.
1379 tcp_ccalgounload(struct cc_algo *unload_algo)
1381 struct cc_algo *tmpalgo;
1384 VNET_ITERATOR_DECL(vnet_iter);
1387 * Check all active control blocks across all network stacks and change
1388 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1389 * requires cleanup code to be run, call it.
1392 VNET_FOREACH(vnet_iter) {
1393 CURVNET_SET(vnet_iter);
1394 INP_INFO_WLOCK(&V_tcbinfo);
1396 * New connections already part way through being initialised
1397 * with the CC algo we're removing will not race with this code
1398 * because the INP_INFO_WLOCK is held during initialisation. We
1399 * therefore don't enter the loop below until the connection
1400 * list has stabilised.
1402 LIST_FOREACH(inp, &V_tcb, inp_list) {
1404 /* Important to skip tcptw structs. */
1405 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1406 (tp = intotcpcb(inp)) != NULL) {
1408 * By holding INP_WLOCK here, we are assured
1409 * that the connection is not currently
1410 * executing inside the CC module's functions
1411 * i.e. it is safe to make the switch back to
1414 if (CC_ALGO(tp) == unload_algo) {
1415 tmpalgo = CC_ALGO(tp);
1416 /* NewReno does not require any init. */
1417 CC_ALGO(tp) = &newreno_cc_algo;
1418 if (tmpalgo->cb_destroy != NULL)
1419 tmpalgo->cb_destroy(tp->ccv);
1424 INP_INFO_WUNLOCK(&V_tcbinfo);
1427 VNET_LIST_RUNLOCK();
1433 * Drop a TCP connection, reporting
1434 * the specified error. If connection is synchronized,
1435 * then send a RST to peer.
1438 tcp_drop(struct tcpcb *tp, int errno)
1440 struct socket *so = tp->t_inpcb->inp_socket;
1442 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1443 INP_WLOCK_ASSERT(tp->t_inpcb);
1445 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1446 tcp_state_change(tp, TCPS_CLOSED);
1447 (void) tp->t_fb->tfb_tcp_output(tp);
1448 TCPSTAT_INC(tcps_drops);
1450 TCPSTAT_INC(tcps_conndrops);
1451 if (errno == ETIMEDOUT && tp->t_softerror)
1452 errno = tp->t_softerror;
1453 so->so_error = errno;
1454 return (tcp_close(tp));
1458 tcp_discardcb(struct tcpcb *tp)
1460 struct inpcb *inp = tp->t_inpcb;
1461 struct socket *so = inp->inp_socket;
1463 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1467 INP_WLOCK_ASSERT(inp);
1470 * Make sure that all of our timers are stopped before we delete the
1473 * If stopping a timer fails, we schedule a discard function in same
1474 * callout, and the last discard function called will take care of
1475 * deleting the tcpcb.
1477 tp->t_timers->tt_draincnt = 0;
1478 tcp_timer_stop(tp, TT_REXMT);
1479 tcp_timer_stop(tp, TT_PERSIST);
1480 tcp_timer_stop(tp, TT_KEEP);
1481 tcp_timer_stop(tp, TT_2MSL);
1482 tcp_timer_stop(tp, TT_DELACK);
1483 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1485 * Call the stop-all function of the methods,
1486 * this function should call the tcp_timer_stop()
1487 * method with each of the function specific timeouts.
1488 * That stop will be called via the tfb_tcp_timer_stop()
1489 * which should use the async drain function of the
1490 * callout system (see tcp_var.h).
1492 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1496 * If we got enough samples through the srtt filter,
1497 * save the rtt and rttvar in the routing entry.
1498 * 'Enough' is arbitrarily defined as 4 rtt samples.
1499 * 4 samples is enough for the srtt filter to converge
1500 * to within enough % of the correct value; fewer samples
1501 * and we could save a bogus rtt. The danger is not high
1502 * as tcp quickly recovers from everything.
1503 * XXX: Works very well but needs some more statistics!
1505 if (tp->t_rttupdated >= 4) {
1506 struct hc_metrics_lite metrics;
1509 bzero(&metrics, sizeof(metrics));
1511 * Update the ssthresh always when the conditions below
1512 * are satisfied. This gives us better new start value
1513 * for the congestion avoidance for new connections.
1514 * ssthresh is only set if packet loss occurred on a session.
1516 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1517 * being torn down. Ideally this code would not use 'so'.
1519 ssthresh = tp->snd_ssthresh;
1520 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1522 * convert the limit from user data bytes to
1523 * packets then to packet data bytes.
1525 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1528 ssthresh *= (tp->t_maxseg +
1530 (isipv6 ? sizeof (struct ip6_hdr) +
1531 sizeof (struct tcphdr) :
1533 sizeof (struct tcpiphdr)
1540 metrics.rmx_ssthresh = ssthresh;
1542 metrics.rmx_rtt = tp->t_srtt;
1543 metrics.rmx_rttvar = tp->t_rttvar;
1544 metrics.rmx_cwnd = tp->snd_cwnd;
1545 metrics.rmx_sendpipe = 0;
1546 metrics.rmx_recvpipe = 0;
1548 tcp_hc_update(&inp->inp_inc, &metrics);
1551 /* free the reassembly queue, if any */
1552 tcp_reass_flush(tp);
1555 /* Disconnect offload device, if any. */
1556 if (tp->t_flags & TF_TOE)
1557 tcp_offload_detach(tp);
1560 tcp_free_sackholes(tp);
1563 /* Free the TCP PCAP queues. */
1564 tcp_pcap_drain(&(tp->t_inpkts));
1565 tcp_pcap_drain(&(tp->t_outpkts));
1568 /* Allow the CC algorithm to clean up after itself. */
1569 if (CC_ALGO(tp)->cb_destroy != NULL)
1570 CC_ALGO(tp)->cb_destroy(tp->ccv);
1573 khelp_destroy_osd(tp->osd);
1577 inp->inp_ppcb = NULL;
1578 if (tp->t_timers->tt_draincnt == 0) {
1579 /* We own the last reference on tcpcb, let's free it. */
1580 TCPSTATES_DEC(tp->t_state);
1581 if (tp->t_fb->tfb_tcp_fb_fini)
1582 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1583 refcount_release(&tp->t_fb->tfb_refcnt);
1585 uma_zfree(V_tcpcb_zone, tp);
1586 released = in_pcbrele_wlocked(inp);
1587 KASSERT(!released, ("%s: inp %p should not have been released "
1588 "here", __func__, inp));
1593 tcp_timer_discard(void *ptp)
1598 tp = (struct tcpcb *)ptp;
1599 CURVNET_SET(tp->t_vnet);
1600 INP_INFO_RLOCK(&V_tcbinfo);
1602 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1605 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1606 ("%s: tcpcb has to be stopped here", __func__));
1607 tp->t_timers->tt_draincnt--;
1608 if (tp->t_timers->tt_draincnt == 0) {
1609 /* We own the last reference on this tcpcb, let's free it. */
1610 TCPSTATES_DEC(tp->t_state);
1611 if (tp->t_fb->tfb_tcp_fb_fini)
1612 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1613 refcount_release(&tp->t_fb->tfb_refcnt);
1615 uma_zfree(V_tcpcb_zone, tp);
1616 if (in_pcbrele_wlocked(inp)) {
1617 INP_INFO_RUNLOCK(&V_tcbinfo);
1623 INP_INFO_RUNLOCK(&V_tcbinfo);
1628 * Attempt to close a TCP control block, marking it as dropped, and freeing
1629 * the socket if we hold the only reference.
1632 tcp_close(struct tcpcb *tp)
1634 struct inpcb *inp = tp->t_inpcb;
1637 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1638 INP_WLOCK_ASSERT(inp);
1641 if (tp->t_state == TCPS_LISTEN)
1642 tcp_offload_listen_stop(tp);
1646 * This releases the TFO pending counter resource for TFO listen
1647 * sockets as well as passively-created TFO sockets that transition
1648 * from SYN_RECEIVED to CLOSED.
1650 if (tp->t_tfo_pending) {
1651 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1652 tp->t_tfo_pending = NULL;
1656 TCPSTAT_INC(tcps_closed);
1657 if (tp->t_state != TCPS_CLOSED)
1658 tcp_state_change(tp, TCPS_CLOSED);
1659 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1660 so = inp->inp_socket;
1661 soisdisconnected(so);
1662 if (inp->inp_flags & INP_SOCKREF) {
1663 KASSERT(so->so_state & SS_PROTOREF,
1664 ("tcp_close: !SS_PROTOREF"));
1665 inp->inp_flags &= ~INP_SOCKREF;
1668 so->so_state &= ~SS_PROTOREF;
1678 VNET_ITERATOR_DECL(vnet_iter);
1683 VNET_LIST_RLOCK_NOSLEEP();
1684 VNET_FOREACH(vnet_iter) {
1685 CURVNET_SET(vnet_iter);
1690 * Walk the tcpbs, if existing, and flush the reassembly queue,
1691 * if there is one...
1692 * XXX: The "Net/3" implementation doesn't imply that the TCP
1693 * reassembly queue should be flushed, but in a situation
1694 * where we're really low on mbufs, this is potentially
1697 INP_INFO_WLOCK(&V_tcbinfo);
1698 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1699 if (inpb->inp_flags & INP_TIMEWAIT)
1702 if ((tcpb = intotcpcb(inpb)) != NULL) {
1703 tcp_reass_flush(tcpb);
1704 tcp_clean_sackreport(tcpb);
1706 if (tcp_pcap_aggressive_free) {
1707 /* Free the TCP PCAP queues. */
1708 tcp_pcap_drain(&(tcpb->t_inpkts));
1709 tcp_pcap_drain(&(tcpb->t_outpkts));
1715 INP_INFO_WUNLOCK(&V_tcbinfo);
1718 VNET_LIST_RUNLOCK_NOSLEEP();
1722 * Notify a tcp user of an asynchronous error;
1723 * store error as soft error, but wake up user
1724 * (for now, won't do anything until can select for soft error).
1726 * Do not wake up user since there currently is no mechanism for
1727 * reporting soft errors (yet - a kqueue filter may be added).
1729 static struct inpcb *
1730 tcp_notify(struct inpcb *inp, int error)
1734 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1735 INP_WLOCK_ASSERT(inp);
1737 if ((inp->inp_flags & INP_TIMEWAIT) ||
1738 (inp->inp_flags & INP_DROPPED))
1741 tp = intotcpcb(inp);
1742 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1745 * Ignore some errors if we are hooked up.
1746 * If connection hasn't completed, has retransmitted several times,
1747 * and receives a second error, give up now. This is better
1748 * than waiting a long time to establish a connection that
1749 * can never complete.
1751 if (tp->t_state == TCPS_ESTABLISHED &&
1752 (error == EHOSTUNREACH || error == ENETUNREACH ||
1753 error == EHOSTDOWN)) {
1754 if (inp->inp_route.ro_rt) {
1755 RTFREE(inp->inp_route.ro_rt);
1756 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1759 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1761 tp = tcp_drop(tp, error);
1767 tp->t_softerror = error;
1771 wakeup( &so->so_timeo);
1778 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1780 int error, i, m, n, pcb_count;
1781 struct inpcb *inp, **inp_list;
1786 * The process of preparing the TCB list is too time-consuming and
1787 * resource-intensive to repeat twice on every request.
1789 if (req->oldptr == NULL) {
1790 n = V_tcbinfo.ipi_count +
1791 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1792 n += imax(n / 8, 10);
1793 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1797 if (req->newptr != NULL)
1801 * OK, now we're committed to doing something.
1803 INP_LIST_RLOCK(&V_tcbinfo);
1804 gencnt = V_tcbinfo.ipi_gencnt;
1805 n = V_tcbinfo.ipi_count;
1806 INP_LIST_RUNLOCK(&V_tcbinfo);
1808 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1810 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1811 + (n + m) * sizeof(struct xtcpcb));
1815 xig.xig_len = sizeof xig;
1816 xig.xig_count = n + m;
1817 xig.xig_gen = gencnt;
1818 xig.xig_sogen = so_gencnt;
1819 error = SYSCTL_OUT(req, &xig, sizeof xig);
1823 error = syncache_pcblist(req, m, &pcb_count);
1827 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1829 INP_INFO_WLOCK(&V_tcbinfo);
1830 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1831 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1833 if (inp->inp_gencnt <= gencnt) {
1835 * XXX: This use of cr_cansee(), introduced with
1836 * TCP state changes, is not quite right, but for
1837 * now, better than nothing.
1839 if (inp->inp_flags & INP_TIMEWAIT) {
1840 if (intotw(inp) != NULL)
1841 error = cr_cansee(req->td->td_ucred,
1842 intotw(inp)->tw_cred);
1844 error = EINVAL; /* Skip this inp. */
1846 error = cr_canseeinpcb(req->td->td_ucred, inp);
1849 inp_list[i++] = inp;
1854 INP_INFO_WUNLOCK(&V_tcbinfo);
1858 for (i = 0; i < n; i++) {
1861 if (inp->inp_gencnt <= gencnt) {
1864 tcp_inptoxtp(inp, &xt);
1866 error = SYSCTL_OUT(req, &xt, sizeof xt);
1870 INP_INFO_RLOCK(&V_tcbinfo);
1871 for (i = 0; i < n; i++) {
1874 if (!in_pcbrele_rlocked(inp))
1877 INP_INFO_RUNLOCK(&V_tcbinfo);
1881 * Give the user an updated idea of our state.
1882 * If the generation differs from what we told
1883 * her before, she knows that something happened
1884 * while we were processing this request, and it
1885 * might be necessary to retry.
1887 INP_LIST_RLOCK(&V_tcbinfo);
1888 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1889 xig.xig_sogen = so_gencnt;
1890 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1891 INP_LIST_RUNLOCK(&V_tcbinfo);
1892 error = SYSCTL_OUT(req, &xig, sizeof xig);
1894 free(inp_list, M_TEMP);
1898 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1899 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1900 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1904 tcp_getcred(SYSCTL_HANDLER_ARGS)
1907 struct sockaddr_in addrs[2];
1911 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1914 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1917 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1918 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1920 if (inp->inp_socket == NULL)
1923 error = cr_canseeinpcb(req->td->td_ucred, inp);
1925 cru2x(inp->inp_cred, &xuc);
1930 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1934 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1935 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1936 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1941 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1944 struct sockaddr_in6 addrs[2];
1951 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1954 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1957 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1958 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1961 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1963 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1972 inp = in_pcblookup(&V_tcbinfo,
1973 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1975 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1976 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1979 inp = in6_pcblookup(&V_tcbinfo,
1980 &addrs[1].sin6_addr, addrs[1].sin6_port,
1981 &addrs[0].sin6_addr, addrs[0].sin6_port,
1982 INPLOOKUP_RLOCKPCB, NULL);
1984 if (inp->inp_socket == NULL)
1987 error = cr_canseeinpcb(req->td->td_ucred, inp);
1989 cru2x(inp->inp_cred, &xuc);
1994 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1998 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1999 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2000 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2006 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2008 struct ip *ip = vip;
2010 struct in_addr faddr;
2013 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2015 struct in_conninfo inc;
2016 tcp_seq icmp_tcp_seq;
2019 faddr = ((struct sockaddr_in *)sa)->sin_addr;
2020 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2023 if (cmd == PRC_MSGSIZE)
2024 notify = tcp_mtudisc_notify;
2025 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2026 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2027 cmd == PRC_TIMXCEED_INTRANS) && ip)
2028 notify = tcp_drop_syn_sent;
2031 * Hostdead is ugly because it goes linearly through all PCBs.
2032 * XXX: We never get this from ICMP, otherwise it makes an
2033 * excellent DoS attack on machines with many connections.
2035 else if (cmd == PRC_HOSTDEAD)
2037 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2041 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2045 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2046 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2047 INP_INFO_RLOCK(&V_tcbinfo);
2048 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2049 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2050 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2051 /* signal EHOSTDOWN, as it flushes the cached route */
2052 inp = (*notify)(inp, EHOSTDOWN);
2055 icmp_tcp_seq = th->th_seq;
2057 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2058 !(inp->inp_flags & INP_DROPPED) &&
2059 !(inp->inp_socket == NULL)) {
2060 tp = intotcpcb(inp);
2061 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2062 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2063 if (cmd == PRC_MSGSIZE) {
2066 * If we got a needfrag set the MTU
2067 * in the route to the suggested new
2068 * value (if given) and then notify.
2070 mtu = ntohs(icp->icmp_nextmtu);
2072 * If no alternative MTU was
2073 * proposed, try the next smaller
2078 ntohs(ip->ip_len), 1);
2079 if (mtu < V_tcp_minmss +
2080 sizeof(struct tcpiphdr))
2081 mtu = V_tcp_minmss +
2082 sizeof(struct tcpiphdr);
2084 * Only process the offered MTU if it
2085 * is smaller than the current one.
2087 if (mtu < tp->t_maxseg +
2088 sizeof(struct tcpiphdr)) {
2089 bzero(&inc, sizeof(inc));
2090 inc.inc_faddr = faddr;
2092 inp->inp_inc.inc_fibnum;
2093 tcp_hc_updatemtu(&inc, mtu);
2094 tcp_mtudisc(inp, mtu);
2097 inp = (*notify)(inp,
2098 inetctlerrmap[cmd]);
2102 bzero(&inc, sizeof(inc));
2103 inc.inc_fport = th->th_dport;
2104 inc.inc_lport = th->th_sport;
2105 inc.inc_faddr = faddr;
2106 inc.inc_laddr = ip->ip_src;
2107 syncache_unreach(&inc, icmp_tcp_seq);
2112 INP_INFO_RUNLOCK(&V_tcbinfo);
2118 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2120 struct in6_addr *dst;
2121 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2122 struct ip6_hdr *ip6;
2126 struct icmp6_hdr *icmp6;
2127 struct ip6ctlparam *ip6cp = NULL;
2128 const struct sockaddr_in6 *sa6_src = NULL;
2129 struct in_conninfo inc;
2134 tcp_seq icmp_tcp_seq;
2138 if (sa->sa_family != AF_INET6 ||
2139 sa->sa_len != sizeof(struct sockaddr_in6))
2142 /* if the parameter is from icmp6, decode it. */
2144 ip6cp = (struct ip6ctlparam *)d;
2145 icmp6 = ip6cp->ip6c_icmp6;
2147 ip6 = ip6cp->ip6c_ip6;
2148 off = ip6cp->ip6c_off;
2149 sa6_src = ip6cp->ip6c_src;
2150 dst = ip6cp->ip6c_finaldst;
2154 off = 0; /* fool gcc */
2159 if (cmd == PRC_MSGSIZE)
2160 notify = tcp_mtudisc_notify;
2161 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2162 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2163 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2164 notify = tcp_drop_syn_sent;
2167 * Hostdead is ugly because it goes linearly through all PCBs.
2168 * XXX: We never get this from ICMP, otherwise it makes an
2169 * excellent DoS attack on machines with many connections.
2171 else if (cmd == PRC_HOSTDEAD)
2173 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2177 in6_pcbnotify(&V_tcbinfo, sa, 0,
2178 (const struct sockaddr *)sa6_src,
2179 0, cmd, NULL, notify);
2183 /* Check if we can safely get the ports from the tcp hdr */
2186 (int32_t) (off + sizeof(struct tcp_ports)))) {
2189 bzero(&t_ports, sizeof(struct tcp_ports));
2190 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2191 INP_INFO_RLOCK(&V_tcbinfo);
2192 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2193 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2194 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2195 /* signal EHOSTDOWN, as it flushes the cached route */
2196 inp = (*notify)(inp, EHOSTDOWN);
2199 off += sizeof(struct tcp_ports);
2200 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2203 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2205 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2206 !(inp->inp_flags & INP_DROPPED) &&
2207 !(inp->inp_socket == NULL)) {
2208 tp = intotcpcb(inp);
2209 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2210 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2211 if (cmd == PRC_MSGSIZE) {
2214 * If we got a needfrag set the MTU
2215 * in the route to the suggested new
2216 * value (if given) and then notify.
2218 mtu = ntohl(icmp6->icmp6_mtu);
2220 * If no alternative MTU was
2221 * proposed, or the proposed
2222 * MTU was too small, set to
2225 if (mtu < IPV6_MMTU)
2226 mtu = IPV6_MMTU - 8;
2227 bzero(&inc, sizeof(inc));
2228 inc.inc_fibnum = M_GETFIB(m);
2229 inc.inc_flags |= INC_ISIPV6;
2230 inc.inc6_faddr = *dst;
2231 if (in6_setscope(&inc.inc6_faddr,
2232 m->m_pkthdr.rcvif, NULL))
2235 * Only process the offered MTU if it
2236 * is smaller than the current one.
2238 if (mtu < tp->t_maxseg +
2239 sizeof (struct tcphdr) +
2240 sizeof (struct ip6_hdr)) {
2241 tcp_hc_updatemtu(&inc, mtu);
2242 tcp_mtudisc(inp, mtu);
2243 ICMP6STAT_INC(icp6s_pmtuchg);
2246 inp = (*notify)(inp,
2247 inet6ctlerrmap[cmd]);
2251 bzero(&inc, sizeof(inc));
2252 inc.inc_fibnum = M_GETFIB(m);
2253 inc.inc_flags |= INC_ISIPV6;
2254 inc.inc_fport = t_ports.th_dport;
2255 inc.inc_lport = t_ports.th_sport;
2256 inc.inc6_faddr = *dst;
2257 inc.inc6_laddr = ip6->ip6_src;
2258 syncache_unreach(&inc, icmp_tcp_seq);
2263 INP_INFO_RUNLOCK(&V_tcbinfo);
2269 * Following is where TCP initial sequence number generation occurs.
2271 * There are two places where we must use initial sequence numbers:
2272 * 1. In SYN-ACK packets.
2273 * 2. In SYN packets.
2275 * All ISNs for SYN-ACK packets are generated by the syncache. See
2276 * tcp_syncache.c for details.
2278 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2279 * depends on this property. In addition, these ISNs should be
2280 * unguessable so as to prevent connection hijacking. To satisfy
2281 * the requirements of this situation, the algorithm outlined in
2282 * RFC 1948 is used, with only small modifications.
2284 * Implementation details:
2286 * Time is based off the system timer, and is corrected so that it
2287 * increases by one megabyte per second. This allows for proper
2288 * recycling on high speed LANs while still leaving over an hour
2291 * As reading the *exact* system time is too expensive to be done
2292 * whenever setting up a TCP connection, we increment the time
2293 * offset in two ways. First, a small random positive increment
2294 * is added to isn_offset for each connection that is set up.
2295 * Second, the function tcp_isn_tick fires once per clock tick
2296 * and increments isn_offset as necessary so that sequence numbers
2297 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2298 * random positive increments serve only to ensure that the same
2299 * exact sequence number is never sent out twice (as could otherwise
2300 * happen when a port is recycled in less than the system tick
2303 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2304 * between seeding of isn_secret. This is normally set to zero,
2305 * as reseeding should not be necessary.
2307 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2308 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2309 * general, this means holding an exclusive (write) lock.
2312 #define ISN_BYTES_PER_SECOND 1048576
2313 #define ISN_STATIC_INCREMENT 4096
2314 #define ISN_RANDOM_INCREMENT (4096 - 1)
2316 static VNET_DEFINE(u_char, isn_secret[32]);
2317 static VNET_DEFINE(int, isn_last);
2318 static VNET_DEFINE(int, isn_last_reseed);
2319 static VNET_DEFINE(u_int32_t, isn_offset);
2320 static VNET_DEFINE(u_int32_t, isn_offset_old);
2322 #define V_isn_secret VNET(isn_secret)
2323 #define V_isn_last VNET(isn_last)
2324 #define V_isn_last_reseed VNET(isn_last_reseed)
2325 #define V_isn_offset VNET(isn_offset)
2326 #define V_isn_offset_old VNET(isn_offset_old)
2329 tcp_new_isn(struct tcpcb *tp)
2332 u_int32_t md5_buffer[4];
2334 u_int32_t projected_offset;
2336 INP_WLOCK_ASSERT(tp->t_inpcb);
2339 /* Seed if this is the first use, reseed if requested. */
2340 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2341 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2343 read_random(&V_isn_secret, sizeof(V_isn_secret));
2344 V_isn_last_reseed = ticks;
2347 /* Compute the md5 hash and return the ISN. */
2349 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2350 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2352 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2353 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2354 sizeof(struct in6_addr));
2355 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2356 sizeof(struct in6_addr));
2360 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2361 sizeof(struct in_addr));
2362 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2363 sizeof(struct in_addr));
2365 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2366 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2367 new_isn = (tcp_seq) md5_buffer[0];
2368 V_isn_offset += ISN_STATIC_INCREMENT +
2369 (arc4random() & ISN_RANDOM_INCREMENT);
2370 if (ticks != V_isn_last) {
2371 projected_offset = V_isn_offset_old +
2372 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2373 if (SEQ_GT(projected_offset, V_isn_offset))
2374 V_isn_offset = projected_offset;
2375 V_isn_offset_old = V_isn_offset;
2378 new_isn += V_isn_offset;
2384 * When a specific ICMP unreachable message is received and the
2385 * connection state is SYN-SENT, drop the connection. This behavior
2386 * is controlled by the icmp_may_rst sysctl.
2389 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2393 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2394 INP_WLOCK_ASSERT(inp);
2396 if ((inp->inp_flags & INP_TIMEWAIT) ||
2397 (inp->inp_flags & INP_DROPPED))
2400 tp = intotcpcb(inp);
2401 if (tp->t_state != TCPS_SYN_SENT)
2404 tp = tcp_drop(tp, errno);
2412 * When `need fragmentation' ICMP is received, update our idea of the MSS
2413 * based on the new value. Also nudge TCP to send something, since we
2414 * know the packet we just sent was dropped.
2415 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2417 static struct inpcb *
2418 tcp_mtudisc_notify(struct inpcb *inp, int error)
2421 tcp_mtudisc(inp, -1);
2426 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2431 INP_WLOCK_ASSERT(inp);
2432 if ((inp->inp_flags & INP_TIMEWAIT) ||
2433 (inp->inp_flags & INP_DROPPED))
2436 tp = intotcpcb(inp);
2437 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2439 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2441 so = inp->inp_socket;
2442 SOCKBUF_LOCK(&so->so_snd);
2443 /* If the mss is larger than the socket buffer, decrease the mss. */
2444 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2445 tp->t_maxseg = so->so_snd.sb_hiwat;
2446 SOCKBUF_UNLOCK(&so->so_snd);
2448 TCPSTAT_INC(tcps_mturesent);
2450 tp->snd_nxt = tp->snd_una;
2451 tcp_free_sackholes(tp);
2452 tp->snd_recover = tp->snd_max;
2453 if (tp->t_flags & TF_SACK_PERMIT)
2454 EXIT_FASTRECOVERY(tp->t_flags);
2455 tp->t_fb->tfb_tcp_output(tp);
2460 * Look-up the routing entry to the peer of this inpcb. If no route
2461 * is found and it cannot be allocated, then return 0. This routine
2462 * is called by TCP routines that access the rmx structure and by
2463 * tcp_mss_update to get the peer/interface MTU.
2466 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2468 struct nhop4_extended nh4;
2470 uint32_t maxmtu = 0;
2472 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2474 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2476 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2477 NHR_REF, 0, &nh4) != 0)
2481 maxmtu = nh4.nh_mtu;
2483 /* Report additional interface capabilities. */
2485 if (ifp->if_capenable & IFCAP_TSO4 &&
2486 ifp->if_hwassist & CSUM_TSO) {
2487 cap->ifcap |= CSUM_TSO;
2488 cap->tsomax = ifp->if_hw_tsomax;
2489 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2490 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2493 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2501 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2503 struct nhop6_extended nh6;
2504 struct in6_addr dst6;
2507 uint32_t maxmtu = 0;
2509 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2511 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2512 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2513 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2518 maxmtu = nh6.nh_mtu;
2520 /* Report additional interface capabilities. */
2522 if (ifp->if_capenable & IFCAP_TSO6 &&
2523 ifp->if_hwassist & CSUM_TSO) {
2524 cap->ifcap |= CSUM_TSO;
2525 cap->tsomax = ifp->if_hw_tsomax;
2526 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2527 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2530 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2538 * Calculate effective SMSS per RFC5681 definition for a given TCP
2539 * connection at its current state, taking into account SACK and etc.
2542 tcp_maxseg(const struct tcpcb *tp)
2546 if (tp->t_flags & TF_NOOPT)
2547 return (tp->t_maxseg);
2550 * Here we have a simplified code from tcp_addoptions(),
2551 * without a proper loop, and having most of paddings hardcoded.
2552 * We might make mistakes with padding here in some edge cases,
2553 * but this is harmless, since result of tcp_maxseg() is used
2554 * only in cwnd and ssthresh estimations.
2556 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2557 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2558 if (tp->t_flags & TF_RCVD_TSTMP)
2559 optlen = TCPOLEN_TSTAMP_APPA;
2562 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2563 if (tp->t_flags & TF_SIGNATURE)
2564 optlen += PAD(TCPOLEN_SIGNATURE);
2566 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2567 optlen += TCPOLEN_SACKHDR;
2568 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2569 optlen = PAD(optlen);
2572 if (tp->t_flags & TF_REQ_TSTMP)
2573 optlen = TCPOLEN_TSTAMP_APPA;
2575 optlen = PAD(TCPOLEN_MAXSEG);
2576 if (tp->t_flags & TF_REQ_SCALE)
2577 optlen += PAD(TCPOLEN_WINDOW);
2578 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2579 if (tp->t_flags & TF_SIGNATURE)
2580 optlen += PAD(TCPOLEN_SIGNATURE);
2582 if (tp->t_flags & TF_SACK_PERMIT)
2583 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2586 optlen = min(optlen, TCP_MAXOLEN);
2587 return (tp->t_maxseg - optlen);
2591 sysctl_drop(SYSCTL_HANDLER_ARGS)
2593 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2594 struct sockaddr_storage addrs[2];
2598 struct sockaddr_in *fin, *lin;
2600 struct sockaddr_in6 *fin6, *lin6;
2611 if (req->oldptr != NULL || req->oldlen != 0)
2613 if (req->newptr == NULL)
2615 if (req->newlen < sizeof(addrs))
2617 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2621 switch (addrs[0].ss_family) {
2624 fin6 = (struct sockaddr_in6 *)&addrs[0];
2625 lin6 = (struct sockaddr_in6 *)&addrs[1];
2626 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2627 lin6->sin6_len != sizeof(struct sockaddr_in6))
2629 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2630 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2632 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2633 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2634 fin = (struct sockaddr_in *)&addrs[0];
2635 lin = (struct sockaddr_in *)&addrs[1];
2638 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2641 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2648 fin = (struct sockaddr_in *)&addrs[0];
2649 lin = (struct sockaddr_in *)&addrs[1];
2650 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2651 lin->sin_len != sizeof(struct sockaddr_in))
2658 INP_INFO_RLOCK(&V_tcbinfo);
2659 switch (addrs[0].ss_family) {
2662 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2663 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2664 INPLOOKUP_WLOCKPCB, NULL);
2669 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2670 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2675 if (inp->inp_flags & INP_TIMEWAIT) {
2677 * XXXRW: There currently exists a state where an
2678 * inpcb is present, but its timewait state has been
2679 * discarded. For now, don't allow dropping of this
2687 } else if (!(inp->inp_flags & INP_DROPPED) &&
2688 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2689 tp = intotcpcb(inp);
2690 tp = tcp_drop(tp, ECONNABORTED);
2697 INP_INFO_RUNLOCK(&V_tcbinfo);
2701 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2702 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2703 0, sysctl_drop, "", "Drop TCP connection");
2706 * Generate a standardized TCP log line for use throughout the
2707 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2708 * allow use in the interrupt context.
2710 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2711 * NB: The function may return NULL if memory allocation failed.
2713 * Due to header inclusion and ordering limitations the struct ip
2714 * and ip6_hdr pointers have to be passed as void pointers.
2717 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2721 /* Is logging enabled? */
2722 if (tcp_log_in_vain == 0)
2725 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2729 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2733 /* Is logging enabled? */
2734 if (tcp_log_debug == 0)
2737 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2741 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2748 const struct ip6_hdr *ip6;
2750 ip6 = (const struct ip6_hdr *)ip6hdr;
2752 ip = (struct ip *)ip4hdr;
2755 * The log line looks like this:
2756 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2758 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2759 sizeof(PRINT_TH_FLAGS) + 1 +
2761 2 * INET6_ADDRSTRLEN;
2763 2 * INET_ADDRSTRLEN;
2766 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2770 strcat(s, "TCP: [");
2773 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2774 inet_ntoa_r(inc->inc_faddr, sp);
2776 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2778 inet_ntoa_r(inc->inc_laddr, sp);
2780 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2783 ip6_sprintf(sp, &inc->inc6_faddr);
2785 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2787 ip6_sprintf(sp, &inc->inc6_laddr);
2789 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2790 } else if (ip6 && th) {
2791 ip6_sprintf(sp, &ip6->ip6_src);
2793 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2795 ip6_sprintf(sp, &ip6->ip6_dst);
2797 sprintf(sp, "]:%i", ntohs(th->th_dport));
2800 } else if (ip && th) {
2801 inet_ntoa_r(ip->ip_src, sp);
2803 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2805 inet_ntoa_r(ip->ip_dst, sp);
2807 sprintf(sp, "]:%i", ntohs(th->th_dport));
2815 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2816 if (*(s + size - 1) != '\0')
2817 panic("%s: string too long", __func__);
2822 * A subroutine which makes it easy to track TCP state changes with DTrace.
2823 * This function shouldn't be called for t_state initializations that don't
2824 * correspond to actual TCP state transitions.
2827 tcp_state_change(struct tcpcb *tp, int newstate)
2829 #if defined(KDTRACE_HOOKS)
2830 int pstate = tp->t_state;
2833 TCPSTATES_DEC(tp->t_state);
2834 TCPSTATES_INC(newstate);
2835 tp->t_state = newstate;
2836 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2840 * Create an external-format (``xtcpcb'') structure using the information in
2841 * the kernel-format tcpcb structure pointed to by tp. This is done to
2842 * reduce the spew of irrelevant information over this interface, to isolate
2843 * user code from changes in the kernel structure, and potentially to provide
2844 * information-hiding if we decide that some of this information should be
2845 * hidden from users.
2848 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2850 struct tcpcb *tp = intotcpcb(inp);
2853 if (inp->inp_flags & INP_TIMEWAIT) {
2854 bzero(xt, sizeof(struct xtcpcb));
2855 xt->t_state = TCPS_TIME_WAIT;
2857 xt->t_state = tp->t_state;
2858 xt->t_flags = tp->t_flags;
2859 xt->t_sndzerowin = tp->t_sndzerowin;
2860 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2861 xt->t_rcvoopack = tp->t_rcvoopack;
2863 now = getsbinuptime();
2864 #define COPYTIMER(ttt) do { \
2865 if (callout_active(&tp->t_timers->ttt)) \
2866 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
2871 COPYTIMER(tt_delack);
2872 COPYTIMER(tt_rexmt);
2873 COPYTIMER(tt_persist);
2877 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2879 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2880 TCP_FUNCTION_NAME_LEN_MAX);
2883 xt->xt_len = sizeof(struct xtcpcb);
2884 in_pcbtoxinpcb(inp, &xt->xt_inp);
2885 if (inp->inp_socket == NULL)
2886 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;