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_fb->tfb_tcp_block_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;
381 rw_rlock(&tcp_function_lock);
382 TAILQ_FOREACH(f, &t_functions, tf_next) {
385 rw_runlock(&tcp_function_lock);
387 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
388 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
393 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
398 rw_rlock(&tcp_function_lock);
399 TAILQ_FOREACH(f, &t_functions, tf_next) {
400 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
401 f->tf_fb->tfb_tcp_block_name,
402 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
403 f->tf_fb->tfb_refcnt);
404 if (linesz >= bufsz) {
412 rw_runlock(&tcp_function_lock);
414 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
415 free(buffer, M_TEMP);
419 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
420 CTLTYPE_STRING|CTLFLAG_RD,
421 NULL, 0, sysctl_net_inet_list_available, "A",
422 "list available TCP Function sets");
425 * Target size of TCP PCB hash tables. Must be a power of two.
427 * Note that this can be overridden by the kernel environment
428 * variable net.inet.tcp.tcbhashsize
431 #define TCBHASHSIZE 0
436 * Callouts should be moved into struct tcp directly. They are currently
437 * separate because the tcpcb structure is exported to userland for sysctl
438 * parsing purposes, which do not know about callouts.
449 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
450 #define V_tcpcb_zone VNET(tcpcb_zone)
452 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
453 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
455 static struct mtx isn_mtx;
457 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
458 #define ISN_LOCK() mtx_lock(&isn_mtx)
459 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
462 * TCP initialization.
465 tcp_zone_change(void *tag)
468 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
469 uma_zone_set_max(V_tcpcb_zone, maxsockets);
470 tcp_tw_zone_change();
474 tcp_inpcb_init(void *mem, int size, int flags)
476 struct inpcb *inp = mem;
478 INP_LOCK_INIT(inp, "inp", "tcpinp");
483 * Take a value and get the next power of 2 that doesn't overflow.
484 * Used to size the tcp_inpcb hash buckets.
487 maketcp_hashsize(int size)
493 * get the next power of 2 higher than maxsockets.
495 hashsize = 1 << fls(size);
496 /* catch overflow, and just go one power of 2 smaller */
497 if (hashsize < size) {
498 hashsize = 1 << (fls(size) - 1);
504 register_tcp_functions(struct tcp_function_block *blk, int wait)
506 struct tcp_function_block *lblk;
507 struct tcp_function *n;
508 struct tcp_function_set fs;
510 if (t_functions_inited == 0) {
511 init_tcp_functions();
513 if ((blk->tfb_tcp_output == NULL) ||
514 (blk->tfb_tcp_do_segment == NULL) ||
515 (blk->tfb_tcp_ctloutput == NULL) ||
516 (strlen(blk->tfb_tcp_block_name) == 0)) {
518 * These functions are required and you
523 if (blk->tfb_tcp_timer_stop_all ||
524 blk->tfb_tcp_timer_activate ||
525 blk->tfb_tcp_timer_active ||
526 blk->tfb_tcp_timer_stop) {
528 * If you define one timer function you
529 * must have them all.
531 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
532 (blk->tfb_tcp_timer_activate == NULL) ||
533 (blk->tfb_tcp_timer_active == NULL) ||
534 (blk->tfb_tcp_timer_stop == NULL)) {
538 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
543 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
544 rw_wlock(&tcp_function_lock);
545 lblk = find_tcp_functions_locked(&fs);
547 /* Duplicate name space not allowed */
548 rw_wunlock(&tcp_function_lock);
549 free(n, M_TCPFUNCTIONS);
552 refcount_init(&blk->tfb_refcnt, 0);
554 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
555 rw_wunlock(&tcp_function_lock);
560 deregister_tcp_functions(struct tcp_function_block *blk)
562 struct tcp_function_block *lblk;
563 struct tcp_function *f;
566 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
567 /* You can't un-register the default */
570 rw_wlock(&tcp_function_lock);
571 if (blk == tcp_func_set_ptr) {
572 /* You can't free the current default */
573 rw_wunlock(&tcp_function_lock);
576 if (blk->tfb_refcnt) {
577 /* Still tcb attached, mark it. */
578 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
579 rw_wunlock(&tcp_function_lock);
582 lblk = find_tcp_fb_locked(blk, &f);
585 TAILQ_REMOVE(&t_functions, f, tf_next);
587 free(f, M_TCPFUNCTIONS);
590 rw_wunlock(&tcp_function_lock);
597 const char *tcbhash_tuneable;
600 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
603 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
604 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
605 printf("%s: WARNING: unable to register helper hook\n", __func__);
606 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
607 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
608 printf("%s: WARNING: unable to register helper hook\n", __func__);
610 hashsize = TCBHASHSIZE;
611 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
614 * Auto tune the hash size based on maxsockets.
615 * A perfect hash would have a 1:1 mapping
616 * (hashsize = maxsockets) however it's been
617 * suggested that O(2) average is better.
619 hashsize = maketcp_hashsize(maxsockets / 4);
621 * Our historical default is 512,
622 * do not autotune lower than this.
626 if (bootverbose && IS_DEFAULT_VNET(curvnet))
627 printf("%s: %s auto tuned to %d\n", __func__,
628 tcbhash_tuneable, hashsize);
631 * We require a hashsize to be a power of two.
632 * Previously if it was not a power of two we would just reset it
633 * back to 512, which could be a nasty surprise if you did not notice
635 * Instead what we do is clip it to the closest power of two lower
636 * than the specified hash value.
638 if (!powerof2(hashsize)) {
639 int oldhashsize = hashsize;
641 hashsize = maketcp_hashsize(hashsize);
642 /* prevent absurdly low value */
645 printf("%s: WARNING: TCB hash size not a power of 2, "
646 "clipped from %d to %d.\n", __func__, oldhashsize,
649 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
650 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
653 * These have to be type stable for the benefit of the timers.
655 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
656 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
657 uma_zone_set_max(V_tcpcb_zone, maxsockets);
658 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
664 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
665 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
666 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
672 /* Skip initialization of globals for non-default instances. */
673 if (!IS_DEFAULT_VNET(curvnet))
676 tcp_reass_global_init();
678 /* XXX virtualize those bellow? */
679 tcp_delacktime = TCPTV_DELACK;
680 tcp_keepinit = TCPTV_KEEP_INIT;
681 tcp_keepidle = TCPTV_KEEP_IDLE;
682 tcp_keepintvl = TCPTV_KEEPINTVL;
683 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
685 tcp_rexmit_min = TCPTV_MIN;
686 if (tcp_rexmit_min < 1)
688 tcp_persmin = TCPTV_PERSMIN;
689 tcp_persmax = TCPTV_PERSMAX;
690 tcp_rexmit_slop = TCPTV_CPU_VAR;
691 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
692 tcp_tcbhashsize = hashsize;
693 /* Setup the tcp function block list */
694 init_tcp_functions();
695 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
697 if (tcp_soreceive_stream) {
699 tcp_usrreqs.pru_soreceive = soreceive_stream;
702 tcp6_usrreqs.pru_soreceive = soreceive_stream;
707 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
709 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
711 if (max_protohdr < TCP_MINPROTOHDR)
712 max_protohdr = TCP_MINPROTOHDR;
713 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
715 #undef TCP_MINPROTOHDR
718 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
719 SHUTDOWN_PRI_DEFAULT);
720 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
721 EVENTHANDLER_PRI_ANY);
729 tcp_destroy(void *unused __unused)
737 * All our processes are gone, all our sockets should be cleaned
738 * up, which means, we should be past the tcp_discardcb() calls.
739 * Sleep to let all tcpcb timers really disappear and cleanup.
742 INP_LIST_RLOCK(&V_tcbinfo);
743 n = V_tcbinfo.ipi_count;
744 INP_LIST_RUNLOCK(&V_tcbinfo);
747 pause("tcpdes", hz / 10);
752 in_pcbinfo_destroy(&V_tcbinfo);
753 /* tcp_discardcb() clears the sack_holes up. */
754 uma_zdestroy(V_sack_hole_zone);
755 uma_zdestroy(V_tcpcb_zone);
759 * Cannot free the zone until all tcpcbs are released as we attach
760 * the allocations to them.
762 tcp_fastopen_destroy();
766 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
768 printf("%s: WARNING: unable to deregister helper hook "
769 "type=%d, id=%d: error %d returned\n", __func__,
770 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
772 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
774 printf("%s: WARNING: unable to deregister helper hook "
775 "type=%d, id=%d: error %d returned\n", __func__,
776 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
780 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
790 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
791 * tcp_template used to store this data in mbufs, but we now recopy it out
792 * of the tcpcb each time to conserve mbufs.
795 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
797 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
799 INP_WLOCK_ASSERT(inp);
802 if ((inp->inp_vflag & INP_IPV6) != 0) {
805 ip6 = (struct ip6_hdr *)ip_ptr;
806 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
807 (inp->inp_flow & IPV6_FLOWINFO_MASK);
808 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
809 (IPV6_VERSION & IPV6_VERSION_MASK);
810 ip6->ip6_nxt = IPPROTO_TCP;
811 ip6->ip6_plen = htons(sizeof(struct tcphdr));
812 ip6->ip6_src = inp->in6p_laddr;
813 ip6->ip6_dst = inp->in6p_faddr;
816 #if defined(INET6) && defined(INET)
823 ip = (struct ip *)ip_ptr;
824 ip->ip_v = IPVERSION;
826 ip->ip_tos = inp->inp_ip_tos;
830 ip->ip_ttl = inp->inp_ip_ttl;
832 ip->ip_p = IPPROTO_TCP;
833 ip->ip_src = inp->inp_laddr;
834 ip->ip_dst = inp->inp_faddr;
837 th->th_sport = inp->inp_lport;
838 th->th_dport = inp->inp_fport;
846 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
850 * Create template to be used to send tcp packets on a connection.
851 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
852 * use for this function is in keepalives, which use tcp_respond.
855 tcpip_maketemplate(struct inpcb *inp)
859 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
862 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
867 * Send a single message to the TCP at address specified by
868 * the given TCP/IP header. If m == NULL, then we make a copy
869 * of the tcpiphdr at th and send directly to the addressed host.
870 * This is used to force keep alive messages out using the TCP
871 * template for a connection. If flags are given then we send
872 * a message back to the TCP which originated the segment th,
873 * and discard the mbuf containing it and any other attached mbufs.
875 * In any case the ack and sequence number of the transmitted
876 * segment are as specified by the parameters.
878 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
881 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
882 tcp_seq ack, tcp_seq seq, int flags)
894 int optlen, tlen, win;
897 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
900 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
907 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
908 INP_WLOCK_ASSERT(inp);
915 if (!(flags & TH_RST)) {
916 win = sbspace(&inp->inp_socket->so_rcv);
917 if (win > TCP_MAXWIN << tp->rcv_scale)
918 win = TCP_MAXWIN << tp->rcv_scale;
920 if ((tp->t_flags & TF_NOOPT) == 0)
924 m = m_gethdr(M_NOWAIT, MT_DATA);
927 m->m_data += max_linkhdr;
930 bcopy((caddr_t)ip6, mtod(m, caddr_t),
931 sizeof(struct ip6_hdr));
932 ip6 = mtod(m, struct ip6_hdr *);
933 nth = (struct tcphdr *)(ip6 + 1);
937 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
938 ip = mtod(m, struct ip *);
939 nth = (struct tcphdr *)(ip + 1);
941 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
943 } else if (!M_WRITABLE(m)) {
946 /* Can't reuse 'm', allocate a new mbuf. */
947 n = m_gethdr(M_NOWAIT, MT_DATA);
953 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
959 n->m_data += max_linkhdr;
960 /* m_len is set later */
961 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
964 bcopy((caddr_t)ip6, mtod(n, caddr_t),
965 sizeof(struct ip6_hdr));
966 ip6 = mtod(n, struct ip6_hdr *);
967 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
968 nth = (struct tcphdr *)(ip6 + 1);
972 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
973 ip = mtod(n, struct ip *);
974 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
975 nth = (struct tcphdr *)(ip + 1);
977 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
978 xchg(nth->th_dport, nth->th_sport, uint16_t);
985 * XXX MRT We inherit the FIB, which is lucky.
989 m->m_data = (caddr_t)ipgen;
990 /* m_len is set later */
993 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
994 nth = (struct tcphdr *)(ip6 + 1);
998 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
999 nth = (struct tcphdr *)(ip + 1);
1003 * this is usually a case when an extension header
1004 * exists between the IPv6 header and the
1007 nth->th_sport = th->th_sport;
1008 nth->th_dport = th->th_dport;
1010 xchg(nth->th_dport, nth->th_sport, uint16_t);
1016 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1018 #if defined(INET) && defined(INET6)
1022 tlen = sizeof (struct tcpiphdr);
1026 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1027 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1028 m, tlen, (long)M_TRAILINGSPACE(m)));
1033 /* Make sure we have room. */
1034 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1035 m->m_next = m_get(M_NOWAIT, MT_DATA);
1037 optp = mtod(m->m_next, u_char *);
1042 optp = (u_char *) (nth + 1);
1048 if (tp->t_flags & TF_RCVD_TSTMP) {
1049 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1050 to.to_tsecr = tp->ts_recent;
1051 to.to_flags |= TOF_TS;
1053 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1054 /* TCP-MD5 (RFC2385). */
1055 if (tp->t_flags & TF_SIGNATURE)
1056 to.to_flags |= TOF_SIGNATURE;
1058 /* Add the options. */
1059 tlen += optlen = tcp_addoptions(&to, optp);
1061 /* Update m_len in the correct mbuf. */
1062 optm->m_len += optlen;
1068 ip6->ip6_vfc = IPV6_VERSION;
1069 ip6->ip6_nxt = IPPROTO_TCP;
1070 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1073 #if defined(INET) && defined(INET6)
1078 ip->ip_len = htons(tlen);
1079 ip->ip_ttl = V_ip_defttl;
1080 if (V_path_mtu_discovery)
1081 ip->ip_off |= htons(IP_DF);
1084 m->m_pkthdr.len = tlen;
1085 m->m_pkthdr.rcvif = NULL;
1089 * Packet is associated with a socket, so allow the
1090 * label of the response to reflect the socket label.
1092 INP_WLOCK_ASSERT(inp);
1093 mac_inpcb_create_mbuf(inp, m);
1096 * Packet is not associated with a socket, so possibly
1097 * update the label in place.
1099 mac_netinet_tcp_reply(m);
1102 nth->th_seq = htonl(seq);
1103 nth->th_ack = htonl(ack);
1105 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1106 nth->th_flags = flags;
1108 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1110 nth->th_win = htons((u_short)win);
1113 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1114 if (to.to_flags & TOF_SIGNATURE) {
1115 if (!TCPMD5_ENABLED() ||
1116 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1123 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1126 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1127 nth->th_sum = in6_cksum_pseudo(ip6,
1128 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1129 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1133 #if defined(INET6) && defined(INET)
1138 m->m_pkthdr.csum_flags = CSUM_TCP;
1139 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1140 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1144 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1145 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1147 TCP_PROBE3(debug__output, tp, th, m);
1149 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1151 TCP_PROBE5(send, NULL, tp, m, tp, nth);
1154 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1156 #if defined(INET) && defined(INET6)
1160 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1165 * Create a new TCP control block, making an
1166 * empty reassembly queue and hooking it to the argument
1167 * protocol control block. The `inp' parameter must have
1168 * come from the zone allocator set up in tcp_init().
1171 tcp_newtcpcb(struct inpcb *inp)
1173 struct tcpcb_mem *tm;
1176 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1179 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1184 /* Initialise cc_var struct for this tcpcb. */
1186 tp->ccv->type = IPPROTO_TCP;
1187 tp->ccv->ccvc.tcp = tp;
1188 rw_rlock(&tcp_function_lock);
1189 tp->t_fb = tcp_func_set_ptr;
1190 refcount_acquire(&tp->t_fb->tfb_refcnt);
1191 rw_runlock(&tcp_function_lock);
1193 * Use the current system default CC algorithm.
1196 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1197 CC_ALGO(tp) = CC_DEFAULT();
1200 if (CC_ALGO(tp)->cb_init != NULL)
1201 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1202 if (tp->t_fb->tfb_tcp_fb_fini)
1203 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1204 refcount_release(&tp->t_fb->tfb_refcnt);
1205 uma_zfree(V_tcpcb_zone, tm);
1211 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1212 if (tp->t_fb->tfb_tcp_fb_fini)
1213 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1214 refcount_release(&tp->t_fb->tfb_refcnt);
1215 uma_zfree(V_tcpcb_zone, tm);
1221 tp->t_vnet = inp->inp_vnet;
1223 tp->t_timers = &tm->tt;
1224 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1227 isipv6 ? V_tcp_v6mssdflt :
1231 /* Set up our timeouts. */
1232 callout_init(&tp->t_timers->tt_rexmt, 1);
1233 callout_init(&tp->t_timers->tt_persist, 1);
1234 callout_init(&tp->t_timers->tt_keep, 1);
1235 callout_init(&tp->t_timers->tt_2msl, 1);
1236 callout_init(&tp->t_timers->tt_delack, 1);
1238 if (V_tcp_do_rfc1323)
1239 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1241 tp->t_flags |= TF_SACK_PERMIT;
1242 TAILQ_INIT(&tp->snd_holes);
1244 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1247 in_pcbref(inp); /* Reference for tcpcb */
1251 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1252 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1253 * reasonable initial retransmit time.
1255 tp->t_srtt = TCPTV_SRTTBASE;
1256 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1257 tp->t_rttmin = tcp_rexmit_min;
1258 tp->t_rxtcur = TCPTV_RTOBASE;
1259 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1260 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1261 tp->t_rcvtime = ticks;
1263 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1264 * because the socket may be bound to an IPv6 wildcard address,
1265 * which may match an IPv4-mapped IPv6 address.
1267 inp->inp_ip_ttl = V_ip_defttl;
1271 * Init the TCP PCAP queues.
1273 tcp_pcap_tcpcb_init(tp);
1275 if (tp->t_fb->tfb_tcp_fb_init) {
1276 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1278 return (tp); /* XXX */
1282 * Switch the congestion control algorithm back to NewReno for any active
1283 * control blocks using an algorithm which is about to go away.
1284 * This ensures the CC framework can allow the unload to proceed without leaving
1285 * any dangling pointers which would trigger a panic.
1286 * Returning non-zero would inform the CC framework that something went wrong
1287 * and it would be unsafe to allow the unload to proceed. However, there is no
1288 * way for this to occur with this implementation so we always return zero.
1291 tcp_ccalgounload(struct cc_algo *unload_algo)
1293 struct cc_algo *tmpalgo;
1296 VNET_ITERATOR_DECL(vnet_iter);
1299 * Check all active control blocks across all network stacks and change
1300 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1301 * requires cleanup code to be run, call it.
1304 VNET_FOREACH(vnet_iter) {
1305 CURVNET_SET(vnet_iter);
1306 INP_INFO_WLOCK(&V_tcbinfo);
1308 * New connections already part way through being initialised
1309 * with the CC algo we're removing will not race with this code
1310 * because the INP_INFO_WLOCK is held during initialisation. We
1311 * therefore don't enter the loop below until the connection
1312 * list has stabilised.
1314 LIST_FOREACH(inp, &V_tcb, inp_list) {
1316 /* Important to skip tcptw structs. */
1317 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1318 (tp = intotcpcb(inp)) != NULL) {
1320 * By holding INP_WLOCK here, we are assured
1321 * that the connection is not currently
1322 * executing inside the CC module's functions
1323 * i.e. it is safe to make the switch back to
1326 if (CC_ALGO(tp) == unload_algo) {
1327 tmpalgo = CC_ALGO(tp);
1328 /* NewReno does not require any init. */
1329 CC_ALGO(tp) = &newreno_cc_algo;
1330 if (tmpalgo->cb_destroy != NULL)
1331 tmpalgo->cb_destroy(tp->ccv);
1336 INP_INFO_WUNLOCK(&V_tcbinfo);
1339 VNET_LIST_RUNLOCK();
1345 * Drop a TCP connection, reporting
1346 * the specified error. If connection is synchronized,
1347 * then send a RST to peer.
1350 tcp_drop(struct tcpcb *tp, int errno)
1352 struct socket *so = tp->t_inpcb->inp_socket;
1354 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1355 INP_WLOCK_ASSERT(tp->t_inpcb);
1357 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1358 tcp_state_change(tp, TCPS_CLOSED);
1359 (void) tp->t_fb->tfb_tcp_output(tp);
1360 TCPSTAT_INC(tcps_drops);
1362 TCPSTAT_INC(tcps_conndrops);
1363 if (errno == ETIMEDOUT && tp->t_softerror)
1364 errno = tp->t_softerror;
1365 so->so_error = errno;
1366 return (tcp_close(tp));
1370 tcp_discardcb(struct tcpcb *tp)
1372 struct inpcb *inp = tp->t_inpcb;
1373 struct socket *so = inp->inp_socket;
1375 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1379 INP_WLOCK_ASSERT(inp);
1382 * Make sure that all of our timers are stopped before we delete the
1385 * If stopping a timer fails, we schedule a discard function in same
1386 * callout, and the last discard function called will take care of
1387 * deleting the tcpcb.
1389 tp->t_timers->tt_draincnt = 0;
1390 tcp_timer_stop(tp, TT_REXMT);
1391 tcp_timer_stop(tp, TT_PERSIST);
1392 tcp_timer_stop(tp, TT_KEEP);
1393 tcp_timer_stop(tp, TT_2MSL);
1394 tcp_timer_stop(tp, TT_DELACK);
1395 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1397 * Call the stop-all function of the methods,
1398 * this function should call the tcp_timer_stop()
1399 * method with each of the function specific timeouts.
1400 * That stop will be called via the tfb_tcp_timer_stop()
1401 * which should use the async drain function of the
1402 * callout system (see tcp_var.h).
1404 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1408 * If we got enough samples through the srtt filter,
1409 * save the rtt and rttvar in the routing entry.
1410 * 'Enough' is arbitrarily defined as 4 rtt samples.
1411 * 4 samples is enough for the srtt filter to converge
1412 * to within enough % of the correct value; fewer samples
1413 * and we could save a bogus rtt. The danger is not high
1414 * as tcp quickly recovers from everything.
1415 * XXX: Works very well but needs some more statistics!
1417 if (tp->t_rttupdated >= 4) {
1418 struct hc_metrics_lite metrics;
1421 bzero(&metrics, sizeof(metrics));
1423 * Update the ssthresh always when the conditions below
1424 * are satisfied. This gives us better new start value
1425 * for the congestion avoidance for new connections.
1426 * ssthresh is only set if packet loss occurred on a session.
1428 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1429 * being torn down. Ideally this code would not use 'so'.
1431 ssthresh = tp->snd_ssthresh;
1432 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1434 * convert the limit from user data bytes to
1435 * packets then to packet data bytes.
1437 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1440 ssthresh *= (tp->t_maxseg +
1442 (isipv6 ? sizeof (struct ip6_hdr) +
1443 sizeof (struct tcphdr) :
1445 sizeof (struct tcpiphdr)
1452 metrics.rmx_ssthresh = ssthresh;
1454 metrics.rmx_rtt = tp->t_srtt;
1455 metrics.rmx_rttvar = tp->t_rttvar;
1456 metrics.rmx_cwnd = tp->snd_cwnd;
1457 metrics.rmx_sendpipe = 0;
1458 metrics.rmx_recvpipe = 0;
1460 tcp_hc_update(&inp->inp_inc, &metrics);
1463 /* free the reassembly queue, if any */
1464 tcp_reass_flush(tp);
1467 /* Disconnect offload device, if any. */
1468 if (tp->t_flags & TF_TOE)
1469 tcp_offload_detach(tp);
1472 tcp_free_sackholes(tp);
1475 /* Free the TCP PCAP queues. */
1476 tcp_pcap_drain(&(tp->t_inpkts));
1477 tcp_pcap_drain(&(tp->t_outpkts));
1480 /* Allow the CC algorithm to clean up after itself. */
1481 if (CC_ALGO(tp)->cb_destroy != NULL)
1482 CC_ALGO(tp)->cb_destroy(tp->ccv);
1485 khelp_destroy_osd(tp->osd);
1489 inp->inp_ppcb = NULL;
1490 if (tp->t_timers->tt_draincnt == 0) {
1491 /* We own the last reference on tcpcb, let's free it. */
1492 TCPSTATES_DEC(tp->t_state);
1493 if (tp->t_fb->tfb_tcp_fb_fini)
1494 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1495 refcount_release(&tp->t_fb->tfb_refcnt);
1497 uma_zfree(V_tcpcb_zone, tp);
1498 released = in_pcbrele_wlocked(inp);
1499 KASSERT(!released, ("%s: inp %p should not have been released "
1500 "here", __func__, inp));
1505 tcp_timer_discard(void *ptp)
1510 tp = (struct tcpcb *)ptp;
1511 CURVNET_SET(tp->t_vnet);
1512 INP_INFO_RLOCK(&V_tcbinfo);
1514 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1517 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1518 ("%s: tcpcb has to be stopped here", __func__));
1519 tp->t_timers->tt_draincnt--;
1520 if (tp->t_timers->tt_draincnt == 0) {
1521 /* We own the last reference on this tcpcb, let's free it. */
1522 TCPSTATES_DEC(tp->t_state);
1523 if (tp->t_fb->tfb_tcp_fb_fini)
1524 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1525 refcount_release(&tp->t_fb->tfb_refcnt);
1527 uma_zfree(V_tcpcb_zone, tp);
1528 if (in_pcbrele_wlocked(inp)) {
1529 INP_INFO_RUNLOCK(&V_tcbinfo);
1535 INP_INFO_RUNLOCK(&V_tcbinfo);
1540 * Attempt to close a TCP control block, marking it as dropped, and freeing
1541 * the socket if we hold the only reference.
1544 tcp_close(struct tcpcb *tp)
1546 struct inpcb *inp = tp->t_inpcb;
1549 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1550 INP_WLOCK_ASSERT(inp);
1553 if (tp->t_state == TCPS_LISTEN)
1554 tcp_offload_listen_stop(tp);
1558 * This releases the TFO pending counter resource for TFO listen
1559 * sockets as well as passively-created TFO sockets that transition
1560 * from SYN_RECEIVED to CLOSED.
1562 if (tp->t_tfo_pending) {
1563 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1564 tp->t_tfo_pending = NULL;
1568 TCPSTAT_INC(tcps_closed);
1569 if (tp->t_state != TCPS_CLOSED)
1570 tcp_state_change(tp, TCPS_CLOSED);
1571 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1572 so = inp->inp_socket;
1573 soisdisconnected(so);
1574 if (inp->inp_flags & INP_SOCKREF) {
1575 KASSERT(so->so_state & SS_PROTOREF,
1576 ("tcp_close: !SS_PROTOREF"));
1577 inp->inp_flags &= ~INP_SOCKREF;
1581 so->so_state &= ~SS_PROTOREF;
1591 VNET_ITERATOR_DECL(vnet_iter);
1596 VNET_LIST_RLOCK_NOSLEEP();
1597 VNET_FOREACH(vnet_iter) {
1598 CURVNET_SET(vnet_iter);
1603 * Walk the tcpbs, if existing, and flush the reassembly queue,
1604 * if there is one...
1605 * XXX: The "Net/3" implementation doesn't imply that the TCP
1606 * reassembly queue should be flushed, but in a situation
1607 * where we're really low on mbufs, this is potentially
1610 INP_INFO_WLOCK(&V_tcbinfo);
1611 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1612 if (inpb->inp_flags & INP_TIMEWAIT)
1615 if ((tcpb = intotcpcb(inpb)) != NULL) {
1616 tcp_reass_flush(tcpb);
1617 tcp_clean_sackreport(tcpb);
1619 if (tcp_pcap_aggressive_free) {
1620 /* Free the TCP PCAP queues. */
1621 tcp_pcap_drain(&(tcpb->t_inpkts));
1622 tcp_pcap_drain(&(tcpb->t_outpkts));
1628 INP_INFO_WUNLOCK(&V_tcbinfo);
1631 VNET_LIST_RUNLOCK_NOSLEEP();
1635 * Notify a tcp user of an asynchronous error;
1636 * store error as soft error, but wake up user
1637 * (for now, won't do anything until can select for soft error).
1639 * Do not wake up user since there currently is no mechanism for
1640 * reporting soft errors (yet - a kqueue filter may be added).
1642 static struct inpcb *
1643 tcp_notify(struct inpcb *inp, int error)
1647 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1648 INP_WLOCK_ASSERT(inp);
1650 if ((inp->inp_flags & INP_TIMEWAIT) ||
1651 (inp->inp_flags & INP_DROPPED))
1654 tp = intotcpcb(inp);
1655 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1658 * Ignore some errors if we are hooked up.
1659 * If connection hasn't completed, has retransmitted several times,
1660 * and receives a second error, give up now. This is better
1661 * than waiting a long time to establish a connection that
1662 * can never complete.
1664 if (tp->t_state == TCPS_ESTABLISHED &&
1665 (error == EHOSTUNREACH || error == ENETUNREACH ||
1666 error == EHOSTDOWN)) {
1667 if (inp->inp_route.ro_rt) {
1668 RTFREE(inp->inp_route.ro_rt);
1669 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1672 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1674 tp = tcp_drop(tp, error);
1680 tp->t_softerror = error;
1684 wakeup( &so->so_timeo);
1691 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1693 int error, i, m, n, pcb_count;
1694 struct inpcb *inp, **inp_list;
1699 * The process of preparing the TCB list is too time-consuming and
1700 * resource-intensive to repeat twice on every request.
1702 if (req->oldptr == NULL) {
1703 n = V_tcbinfo.ipi_count +
1704 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1705 n += imax(n / 8, 10);
1706 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1710 if (req->newptr != NULL)
1714 * OK, now we're committed to doing something.
1716 INP_LIST_RLOCK(&V_tcbinfo);
1717 gencnt = V_tcbinfo.ipi_gencnt;
1718 n = V_tcbinfo.ipi_count;
1719 INP_LIST_RUNLOCK(&V_tcbinfo);
1721 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1723 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1724 + (n + m) * sizeof(struct xtcpcb));
1728 xig.xig_len = sizeof xig;
1729 xig.xig_count = n + m;
1730 xig.xig_gen = gencnt;
1731 xig.xig_sogen = so_gencnt;
1732 error = SYSCTL_OUT(req, &xig, sizeof xig);
1736 error = syncache_pcblist(req, m, &pcb_count);
1740 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1742 INP_INFO_WLOCK(&V_tcbinfo);
1743 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1744 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1746 if (inp->inp_gencnt <= gencnt) {
1748 * XXX: This use of cr_cansee(), introduced with
1749 * TCP state changes, is not quite right, but for
1750 * now, better than nothing.
1752 if (inp->inp_flags & INP_TIMEWAIT) {
1753 if (intotw(inp) != NULL)
1754 error = cr_cansee(req->td->td_ucred,
1755 intotw(inp)->tw_cred);
1757 error = EINVAL; /* Skip this inp. */
1759 error = cr_canseeinpcb(req->td->td_ucred, inp);
1762 inp_list[i++] = inp;
1767 INP_INFO_WUNLOCK(&V_tcbinfo);
1771 for (i = 0; i < n; i++) {
1774 if (inp->inp_gencnt <= gencnt) {
1777 tcp_inptoxtp(inp, &xt);
1779 error = SYSCTL_OUT(req, &xt, sizeof xt);
1783 INP_INFO_RLOCK(&V_tcbinfo);
1784 for (i = 0; i < n; i++) {
1787 if (!in_pcbrele_rlocked(inp))
1790 INP_INFO_RUNLOCK(&V_tcbinfo);
1794 * Give the user an updated idea of our state.
1795 * If the generation differs from what we told
1796 * her before, she knows that something happened
1797 * while we were processing this request, and it
1798 * might be necessary to retry.
1800 INP_LIST_RLOCK(&V_tcbinfo);
1801 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1802 xig.xig_sogen = so_gencnt;
1803 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1804 INP_LIST_RUNLOCK(&V_tcbinfo);
1805 error = SYSCTL_OUT(req, &xig, sizeof xig);
1807 free(inp_list, M_TEMP);
1811 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1812 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1813 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1817 tcp_getcred(SYSCTL_HANDLER_ARGS)
1820 struct sockaddr_in addrs[2];
1824 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1827 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1830 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1831 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1833 if (inp->inp_socket == NULL)
1836 error = cr_canseeinpcb(req->td->td_ucred, inp);
1838 cru2x(inp->inp_cred, &xuc);
1843 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1847 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1848 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1849 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1854 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1857 struct sockaddr_in6 addrs[2];
1864 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1867 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1870 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1871 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1874 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1876 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1885 inp = in_pcblookup(&V_tcbinfo,
1886 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1888 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1889 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1892 inp = in6_pcblookup(&V_tcbinfo,
1893 &addrs[1].sin6_addr, addrs[1].sin6_port,
1894 &addrs[0].sin6_addr, addrs[0].sin6_port,
1895 INPLOOKUP_RLOCKPCB, NULL);
1897 if (inp->inp_socket == NULL)
1900 error = cr_canseeinpcb(req->td->td_ucred, inp);
1902 cru2x(inp->inp_cred, &xuc);
1907 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1911 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1912 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1913 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1919 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1921 struct ip *ip = vip;
1923 struct in_addr faddr;
1926 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1928 struct in_conninfo inc;
1929 tcp_seq icmp_tcp_seq;
1932 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1933 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1936 if (cmd == PRC_MSGSIZE)
1937 notify = tcp_mtudisc_notify;
1938 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1939 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1940 cmd == PRC_TIMXCEED_INTRANS) && ip)
1941 notify = tcp_drop_syn_sent;
1944 * Hostdead is ugly because it goes linearly through all PCBs.
1945 * XXX: We never get this from ICMP, otherwise it makes an
1946 * excellent DoS attack on machines with many connections.
1948 else if (cmd == PRC_HOSTDEAD)
1950 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1954 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1958 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1959 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1960 INP_INFO_RLOCK(&V_tcbinfo);
1961 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1962 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1963 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1964 /* signal EHOSTDOWN, as it flushes the cached route */
1965 inp = (*notify)(inp, EHOSTDOWN);
1968 } else if (inp != NULL) {
1969 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1970 !(inp->inp_flags & INP_DROPPED) &&
1971 !(inp->inp_socket == NULL)) {
1972 icmp_tcp_seq = ntohl(th->th_seq);
1973 tp = intotcpcb(inp);
1974 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1975 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1976 if (cmd == PRC_MSGSIZE) {
1979 * If we got a needfrag set the MTU
1980 * in the route to the suggested new
1981 * value (if given) and then notify.
1983 mtu = ntohs(icp->icmp_nextmtu);
1985 * If no alternative MTU was
1986 * proposed, try the next smaller
1991 ntohs(ip->ip_len), 1);
1992 if (mtu < V_tcp_minmss +
1993 sizeof(struct tcpiphdr))
1994 mtu = V_tcp_minmss +
1995 sizeof(struct tcpiphdr);
1997 * Only process the offered MTU if it
1998 * is smaller than the current one.
2000 if (mtu < tp->t_maxseg +
2001 sizeof(struct tcpiphdr)) {
2002 bzero(&inc, sizeof(inc));
2003 inc.inc_faddr = faddr;
2005 inp->inp_inc.inc_fibnum;
2006 tcp_hc_updatemtu(&inc, mtu);
2007 tcp_mtudisc(inp, mtu);
2010 inp = (*notify)(inp,
2011 inetctlerrmap[cmd]);
2017 bzero(&inc, sizeof(inc));
2018 inc.inc_fport = th->th_dport;
2019 inc.inc_lport = th->th_sport;
2020 inc.inc_faddr = faddr;
2021 inc.inc_laddr = ip->ip_src;
2022 syncache_unreach(&inc, th);
2024 INP_INFO_RUNLOCK(&V_tcbinfo);
2030 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2032 struct in6_addr *dst;
2034 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2035 struct ip6_hdr *ip6;
2039 struct icmp6_hdr *icmp6;
2040 struct ip6ctlparam *ip6cp = NULL;
2041 const struct sockaddr_in6 *sa6_src = NULL;
2042 struct in_conninfo inc;
2043 tcp_seq icmp_tcp_seq;
2048 if (sa->sa_family != AF_INET6 ||
2049 sa->sa_len != sizeof(struct sockaddr_in6))
2052 /* if the parameter is from icmp6, decode it. */
2054 ip6cp = (struct ip6ctlparam *)d;
2055 icmp6 = ip6cp->ip6c_icmp6;
2057 ip6 = ip6cp->ip6c_ip6;
2058 off = ip6cp->ip6c_off;
2059 sa6_src = ip6cp->ip6c_src;
2060 dst = ip6cp->ip6c_finaldst;
2064 off = 0; /* fool gcc */
2069 if (cmd == PRC_MSGSIZE)
2070 notify = tcp_mtudisc_notify;
2071 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2072 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2073 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2074 notify = tcp_drop_syn_sent;
2077 * Hostdead is ugly because it goes linearly through all PCBs.
2078 * XXX: We never get this from ICMP, otherwise it makes an
2079 * excellent DoS attack on machines with many connections.
2081 else if (cmd == PRC_HOSTDEAD)
2083 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2087 in6_pcbnotify(&V_tcbinfo, sa, 0,
2088 (const struct sockaddr *)sa6_src,
2089 0, cmd, NULL, notify);
2093 /* Check if we can safely get the ports from the tcp hdr */
2096 (int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2100 th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2101 INP_INFO_RLOCK(&V_tcbinfo);
2102 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2103 &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2104 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2105 /* signal EHOSTDOWN, as it flushes the cached route */
2106 inp = (*notify)(inp, EHOSTDOWN);
2109 } else if (inp != NULL) {
2110 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2111 !(inp->inp_flags & INP_DROPPED) &&
2112 !(inp->inp_socket == NULL)) {
2113 icmp_tcp_seq = ntohl(th->th_seq);
2114 tp = intotcpcb(inp);
2115 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2116 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2117 if (cmd == PRC_MSGSIZE) {
2120 * If we got a needfrag set the MTU
2121 * in the route to the suggested new
2122 * value (if given) and then notify.
2124 mtu = ntohl(icmp6->icmp6_mtu);
2126 * If no alternative MTU was
2127 * proposed, or the proposed
2128 * MTU was too small, set to
2131 if (mtu < IPV6_MMTU)
2132 mtu = IPV6_MMTU - 8;
2135 bzero(&inc, sizeof(inc));
2136 inc.inc_fibnum = M_GETFIB(m);
2137 inc.inc_flags |= INC_ISIPV6;
2138 inc.inc6_faddr = *dst;
2139 if (in6_setscope(&inc.inc6_faddr,
2140 m->m_pkthdr.rcvif, NULL))
2144 * Only process the offered MTU if it
2145 * is smaller than the current one.
2147 if (mtu < tp->t_maxseg +
2148 (sizeof (*th) + sizeof (*ip6))) {
2149 tcp_hc_updatemtu(&inc, mtu);
2150 tcp_mtudisc(inp, mtu);
2151 ICMP6STAT_INC(icp6s_pmtuchg);
2154 inp = (*notify)(inp,
2155 inet6ctlerrmap[cmd]);
2162 bzero(&inc, sizeof(inc));
2163 inc.inc_fibnum = M_GETFIB(m);
2164 inc.inc_flags |= INC_ISIPV6;
2165 inc.inc_fport = th->th_dport;
2166 inc.inc_lport = th->th_sport;
2167 inc.inc6_faddr = *dst;
2168 inc.inc6_laddr = ip6->ip6_src;
2169 syncache_unreach(&inc, th);
2171 INP_INFO_RUNLOCK(&V_tcbinfo);
2177 * Following is where TCP initial sequence number generation occurs.
2179 * There are two places where we must use initial sequence numbers:
2180 * 1. In SYN-ACK packets.
2181 * 2. In SYN packets.
2183 * All ISNs for SYN-ACK packets are generated by the syncache. See
2184 * tcp_syncache.c for details.
2186 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2187 * depends on this property. In addition, these ISNs should be
2188 * unguessable so as to prevent connection hijacking. To satisfy
2189 * the requirements of this situation, the algorithm outlined in
2190 * RFC 1948 is used, with only small modifications.
2192 * Implementation details:
2194 * Time is based off the system timer, and is corrected so that it
2195 * increases by one megabyte per second. This allows for proper
2196 * recycling on high speed LANs while still leaving over an hour
2199 * As reading the *exact* system time is too expensive to be done
2200 * whenever setting up a TCP connection, we increment the time
2201 * offset in two ways. First, a small random positive increment
2202 * is added to isn_offset for each connection that is set up.
2203 * Second, the function tcp_isn_tick fires once per clock tick
2204 * and increments isn_offset as necessary so that sequence numbers
2205 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2206 * random positive increments serve only to ensure that the same
2207 * exact sequence number is never sent out twice (as could otherwise
2208 * happen when a port is recycled in less than the system tick
2211 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2212 * between seeding of isn_secret. This is normally set to zero,
2213 * as reseeding should not be necessary.
2215 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2216 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2217 * general, this means holding an exclusive (write) lock.
2220 #define ISN_BYTES_PER_SECOND 1048576
2221 #define ISN_STATIC_INCREMENT 4096
2222 #define ISN_RANDOM_INCREMENT (4096 - 1)
2224 static VNET_DEFINE(u_char, isn_secret[32]);
2225 static VNET_DEFINE(int, isn_last);
2226 static VNET_DEFINE(int, isn_last_reseed);
2227 static VNET_DEFINE(u_int32_t, isn_offset);
2228 static VNET_DEFINE(u_int32_t, isn_offset_old);
2230 #define V_isn_secret VNET(isn_secret)
2231 #define V_isn_last VNET(isn_last)
2232 #define V_isn_last_reseed VNET(isn_last_reseed)
2233 #define V_isn_offset VNET(isn_offset)
2234 #define V_isn_offset_old VNET(isn_offset_old)
2237 tcp_new_isn(struct tcpcb *tp)
2240 u_int32_t md5_buffer[4];
2242 u_int32_t projected_offset;
2244 INP_WLOCK_ASSERT(tp->t_inpcb);
2247 /* Seed if this is the first use, reseed if requested. */
2248 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2249 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2251 read_random(&V_isn_secret, sizeof(V_isn_secret));
2252 V_isn_last_reseed = ticks;
2255 /* Compute the md5 hash and return the ISN. */
2257 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2258 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2260 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2261 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2262 sizeof(struct in6_addr));
2263 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2264 sizeof(struct in6_addr));
2268 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2269 sizeof(struct in_addr));
2270 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2271 sizeof(struct in_addr));
2273 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2274 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2275 new_isn = (tcp_seq) md5_buffer[0];
2276 V_isn_offset += ISN_STATIC_INCREMENT +
2277 (arc4random() & ISN_RANDOM_INCREMENT);
2278 if (ticks != V_isn_last) {
2279 projected_offset = V_isn_offset_old +
2280 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2281 if (SEQ_GT(projected_offset, V_isn_offset))
2282 V_isn_offset = projected_offset;
2283 V_isn_offset_old = V_isn_offset;
2286 new_isn += V_isn_offset;
2292 * When a specific ICMP unreachable message is received and the
2293 * connection state is SYN-SENT, drop the connection. This behavior
2294 * is controlled by the icmp_may_rst sysctl.
2297 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2301 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2302 INP_WLOCK_ASSERT(inp);
2304 if ((inp->inp_flags & INP_TIMEWAIT) ||
2305 (inp->inp_flags & INP_DROPPED))
2308 tp = intotcpcb(inp);
2309 if (tp->t_state != TCPS_SYN_SENT)
2312 tp = tcp_drop(tp, errno);
2320 * When `need fragmentation' ICMP is received, update our idea of the MSS
2321 * based on the new value. Also nudge TCP to send something, since we
2322 * know the packet we just sent was dropped.
2323 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2325 static struct inpcb *
2326 tcp_mtudisc_notify(struct inpcb *inp, int error)
2329 tcp_mtudisc(inp, -1);
2334 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2339 INP_WLOCK_ASSERT(inp);
2340 if ((inp->inp_flags & INP_TIMEWAIT) ||
2341 (inp->inp_flags & INP_DROPPED))
2344 tp = intotcpcb(inp);
2345 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2347 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2349 so = inp->inp_socket;
2350 SOCKBUF_LOCK(&so->so_snd);
2351 /* If the mss is larger than the socket buffer, decrease the mss. */
2352 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2353 tp->t_maxseg = so->so_snd.sb_hiwat;
2354 SOCKBUF_UNLOCK(&so->so_snd);
2356 TCPSTAT_INC(tcps_mturesent);
2358 tp->snd_nxt = tp->snd_una;
2359 tcp_free_sackholes(tp);
2360 tp->snd_recover = tp->snd_max;
2361 if (tp->t_flags & TF_SACK_PERMIT)
2362 EXIT_FASTRECOVERY(tp->t_flags);
2363 tp->t_fb->tfb_tcp_output(tp);
2368 * Look-up the routing entry to the peer of this inpcb. If no route
2369 * is found and it cannot be allocated, then return 0. This routine
2370 * is called by TCP routines that access the rmx structure and by
2371 * tcp_mss_update to get the peer/interface MTU.
2374 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2376 struct nhop4_extended nh4;
2378 uint32_t maxmtu = 0;
2380 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2382 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2384 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2385 NHR_REF, 0, &nh4) != 0)
2389 maxmtu = nh4.nh_mtu;
2391 /* Report additional interface capabilities. */
2393 if (ifp->if_capenable & IFCAP_TSO4 &&
2394 ifp->if_hwassist & CSUM_TSO) {
2395 cap->ifcap |= CSUM_TSO;
2396 cap->tsomax = ifp->if_hw_tsomax;
2397 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2398 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2401 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2409 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2411 struct nhop6_extended nh6;
2412 struct in6_addr dst6;
2415 uint32_t maxmtu = 0;
2417 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2419 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2420 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2421 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2426 maxmtu = nh6.nh_mtu;
2428 /* Report additional interface capabilities. */
2430 if (ifp->if_capenable & IFCAP_TSO6 &&
2431 ifp->if_hwassist & CSUM_TSO) {
2432 cap->ifcap |= CSUM_TSO;
2433 cap->tsomax = ifp->if_hw_tsomax;
2434 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2435 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2438 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2446 * Calculate effective SMSS per RFC5681 definition for a given TCP
2447 * connection at its current state, taking into account SACK and etc.
2450 tcp_maxseg(const struct tcpcb *tp)
2454 if (tp->t_flags & TF_NOOPT)
2455 return (tp->t_maxseg);
2458 * Here we have a simplified code from tcp_addoptions(),
2459 * without a proper loop, and having most of paddings hardcoded.
2460 * We might make mistakes with padding here in some edge cases,
2461 * but this is harmless, since result of tcp_maxseg() is used
2462 * only in cwnd and ssthresh estimations.
2464 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2465 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2466 if (tp->t_flags & TF_RCVD_TSTMP)
2467 optlen = TCPOLEN_TSTAMP_APPA;
2470 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2471 if (tp->t_flags & TF_SIGNATURE)
2472 optlen += PAD(TCPOLEN_SIGNATURE);
2474 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2475 optlen += TCPOLEN_SACKHDR;
2476 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2477 optlen = PAD(optlen);
2480 if (tp->t_flags & TF_REQ_TSTMP)
2481 optlen = TCPOLEN_TSTAMP_APPA;
2483 optlen = PAD(TCPOLEN_MAXSEG);
2484 if (tp->t_flags & TF_REQ_SCALE)
2485 optlen += PAD(TCPOLEN_WINDOW);
2486 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2487 if (tp->t_flags & TF_SIGNATURE)
2488 optlen += PAD(TCPOLEN_SIGNATURE);
2490 if (tp->t_flags & TF_SACK_PERMIT)
2491 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2494 optlen = min(optlen, TCP_MAXOLEN);
2495 return (tp->t_maxseg - optlen);
2499 sysctl_drop(SYSCTL_HANDLER_ARGS)
2501 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2502 struct sockaddr_storage addrs[2];
2506 struct sockaddr_in *fin, *lin;
2508 struct sockaddr_in6 *fin6, *lin6;
2519 if (req->oldptr != NULL || req->oldlen != 0)
2521 if (req->newptr == NULL)
2523 if (req->newlen < sizeof(addrs))
2525 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2529 switch (addrs[0].ss_family) {
2532 fin6 = (struct sockaddr_in6 *)&addrs[0];
2533 lin6 = (struct sockaddr_in6 *)&addrs[1];
2534 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2535 lin6->sin6_len != sizeof(struct sockaddr_in6))
2537 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2538 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2540 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2541 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2542 fin = (struct sockaddr_in *)&addrs[0];
2543 lin = (struct sockaddr_in *)&addrs[1];
2546 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2549 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2556 fin = (struct sockaddr_in *)&addrs[0];
2557 lin = (struct sockaddr_in *)&addrs[1];
2558 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2559 lin->sin_len != sizeof(struct sockaddr_in))
2566 INP_INFO_RLOCK(&V_tcbinfo);
2567 switch (addrs[0].ss_family) {
2570 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2571 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2572 INPLOOKUP_WLOCKPCB, NULL);
2577 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2578 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2583 if (inp->inp_flags & INP_TIMEWAIT) {
2585 * XXXRW: There currently exists a state where an
2586 * inpcb is present, but its timewait state has been
2587 * discarded. For now, don't allow dropping of this
2595 } else if (!(inp->inp_flags & INP_DROPPED) &&
2596 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2597 tp = intotcpcb(inp);
2598 tp = tcp_drop(tp, ECONNABORTED);
2605 INP_INFO_RUNLOCK(&V_tcbinfo);
2609 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2610 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2611 0, sysctl_drop, "", "Drop TCP connection");
2614 * Generate a standardized TCP log line for use throughout the
2615 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2616 * allow use in the interrupt context.
2618 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2619 * NB: The function may return NULL if memory allocation failed.
2621 * Due to header inclusion and ordering limitations the struct ip
2622 * and ip6_hdr pointers have to be passed as void pointers.
2625 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2629 /* Is logging enabled? */
2630 if (tcp_log_in_vain == 0)
2633 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2637 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2641 /* Is logging enabled? */
2642 if (tcp_log_debug == 0)
2645 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2649 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2656 const struct ip6_hdr *ip6;
2658 ip6 = (const struct ip6_hdr *)ip6hdr;
2660 ip = (struct ip *)ip4hdr;
2663 * The log line looks like this:
2664 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2666 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2667 sizeof(PRINT_TH_FLAGS) + 1 +
2669 2 * INET6_ADDRSTRLEN;
2671 2 * INET_ADDRSTRLEN;
2674 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2678 strcat(s, "TCP: [");
2681 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2682 inet_ntoa_r(inc->inc_faddr, sp);
2684 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2686 inet_ntoa_r(inc->inc_laddr, sp);
2688 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2691 ip6_sprintf(sp, &inc->inc6_faddr);
2693 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2695 ip6_sprintf(sp, &inc->inc6_laddr);
2697 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2698 } else if (ip6 && th) {
2699 ip6_sprintf(sp, &ip6->ip6_src);
2701 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2703 ip6_sprintf(sp, &ip6->ip6_dst);
2705 sprintf(sp, "]:%i", ntohs(th->th_dport));
2708 } else if (ip && th) {
2709 inet_ntoa_r(ip->ip_src, sp);
2711 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2713 inet_ntoa_r(ip->ip_dst, sp);
2715 sprintf(sp, "]:%i", ntohs(th->th_dport));
2723 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2724 if (*(s + size - 1) != '\0')
2725 panic("%s: string too long", __func__);
2730 * A subroutine which makes it easy to track TCP state changes with DTrace.
2731 * This function shouldn't be called for t_state initializations that don't
2732 * correspond to actual TCP state transitions.
2735 tcp_state_change(struct tcpcb *tp, int newstate)
2737 #if defined(KDTRACE_HOOKS)
2738 int pstate = tp->t_state;
2741 TCPSTATES_DEC(tp->t_state);
2742 TCPSTATES_INC(newstate);
2743 tp->t_state = newstate;
2744 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2748 * Create an external-format (``xtcpcb'') structure using the information in
2749 * the kernel-format tcpcb structure pointed to by tp. This is done to
2750 * reduce the spew of irrelevant information over this interface, to isolate
2751 * user code from changes in the kernel structure, and potentially to provide
2752 * information-hiding if we decide that some of this information should be
2753 * hidden from users.
2756 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2758 struct tcpcb *tp = intotcpcb(inp);
2761 if (inp->inp_flags & INP_TIMEWAIT) {
2762 bzero(xt, sizeof(struct xtcpcb));
2763 xt->t_state = TCPS_TIME_WAIT;
2765 xt->t_state = tp->t_state;
2766 xt->t_flags = tp->t_flags;
2767 xt->t_sndzerowin = tp->t_sndzerowin;
2768 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2769 xt->t_rcvoopack = tp->t_rcvoopack;
2771 now = getsbinuptime();
2772 #define COPYTIMER(ttt) do { \
2773 if (callout_active(&tp->t_timers->ttt)) \
2774 xt->ttt = (tp->t_timers->ttt.c_time - now) / \
2779 COPYTIMER(tt_delack);
2780 COPYTIMER(tt_rexmt);
2781 COPYTIMER(tt_persist);
2785 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2787 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2788 TCP_FUNCTION_NAME_LEN_MAX);
2791 xt->xt_len = sizeof(struct xtcpcb);
2792 in_pcbtoxinpcb(inp, &xt->xt_inp);
2793 if (inp->inp_socket == NULL)
2794 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;