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 * 4. 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>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/refcount.h>
54 #include <sys/domain.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/protosw.h>
62 #include <sys/random.h>
66 #include <net/route.h>
68 #include <net/if_var.h>
71 #include <netinet/in.h>
72 #include <netinet/in_fib.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/ip_var.h>
81 #include <netinet/icmp6.h>
82 #include <netinet/ip6.h>
83 #include <netinet6/in6_fib.h>
84 #include <netinet6/in6_pcb.h>
85 #include <netinet6/ip6_var.h>
86 #include <netinet6/scope6_var.h>
87 #include <netinet6/nd6.h>
91 #include <netinet/tcp_fastopen.h>
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_syncache.h>
99 #include <netinet/cc/cc.h>
101 #include <netinet6/tcp6_var.h>
103 #include <netinet/tcpip.h>
105 #include <netinet/tcp_pcap.h>
108 #include <netinet/tcp_debug.h>
111 #include <netinet6/ip6protosw.h>
114 #include <netinet/tcp_offload.h>
117 #include <netipsec/ipsec_support.h>
119 #include <machine/in_cksum.h>
122 #include <security/mac/mac_framework.h>
124 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
126 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
129 struct rwlock tcp_function_lock;
132 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
137 error = sysctl_handle_int(oidp, &new, 0, req);
138 if (error == 0 && req->newptr) {
139 if (new < TCP_MINMSS)
147 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
148 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
149 &sysctl_net_inet_tcp_mss_check, "I",
150 "Default TCP Maximum Segment Size");
154 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
158 new = V_tcp_v6mssdflt;
159 error = sysctl_handle_int(oidp, &new, 0, req);
160 if (error == 0 && req->newptr) {
161 if (new < TCP_MINMSS)
164 V_tcp_v6mssdflt = new;
169 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
171 &sysctl_net_inet_tcp_mss_v6_check, "I",
172 "Default TCP Maximum Segment Size for IPv6");
176 * Minimum MSS we accept and use. This prevents DoS attacks where
177 * we are forced to a ridiculous low MSS like 20 and send hundreds
178 * of packets instead of one. The effect scales with the available
179 * bandwidth and quickly saturates the CPU and network interface
180 * with packet generation and sending. Set to zero to disable MINMSS
181 * checking. This setting prevents us from sending too small packets.
183 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
185 &VNET_NAME(tcp_minmss), 0,
186 "Minimum TCP Maximum Segment Size");
188 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
189 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_do_rfc1323), 0,
191 "Enable rfc1323 (high performance TCP) extensions");
193 static int tcp_log_debug = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
195 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
197 static int tcp_tcbhashsize;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
199 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
201 static int do_tcpdrain = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
203 "Enable tcp_drain routine for extra help when low on mbufs");
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
206 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
208 static VNET_DEFINE(int, icmp_may_rst) = 1;
209 #define V_icmp_may_rst VNET(icmp_may_rst)
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
211 &VNET_NAME(icmp_may_rst), 0,
212 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
214 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
215 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_isn_reseed_interval), 0,
218 "Seconds between reseeding of ISN secret");
220 static int tcp_soreceive_stream;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
222 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define V_sack_hole_zone VNET(sack_hole_zone)
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
229 VNET_DEFINE(u_char, ts_offset_secret[32]);
230 #define V_ts_offset_secret VNET(ts_offset_secret)
232 static struct inpcb *tcp_notify(struct inpcb *, int);
233 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234 static void tcp_mtudisc(struct inpcb *, int);
235 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
236 void *ip4hdr, const void *ip6hdr);
239 static struct tcp_function_block tcp_def_funcblk = {
243 tcp_default_ctloutput,
254 int t_functions_inited = 0;
255 struct tcp_funchead t_functions;
256 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
259 init_tcp_functions(void)
261 if (t_functions_inited == 0) {
262 TAILQ_INIT(&t_functions);
263 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
264 t_functions_inited = 1;
268 static struct tcp_function_block *
269 find_tcp_functions_locked(struct tcp_function_set *fs)
271 struct tcp_function *f;
272 struct tcp_function_block *blk=NULL;
274 TAILQ_FOREACH(f, &t_functions, tf_next) {
275 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
283 static struct tcp_function_block *
284 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
286 struct tcp_function_block *rblk=NULL;
287 struct tcp_function *f;
289 TAILQ_FOREACH(f, &t_functions, tf_next) {
290 if (f->tf_fb == blk) {
301 struct tcp_function_block *
302 find_and_ref_tcp_functions(struct tcp_function_set *fs)
304 struct tcp_function_block *blk;
306 rw_rlock(&tcp_function_lock);
307 blk = find_tcp_functions_locked(fs);
309 refcount_acquire(&blk->tfb_refcnt);
310 rw_runlock(&tcp_function_lock);
314 struct tcp_function_block *
315 find_and_ref_tcp_fb(struct tcp_function_block *blk)
317 struct tcp_function_block *rblk;
319 rw_rlock(&tcp_function_lock);
320 rblk = find_tcp_fb_locked(blk, NULL);
322 refcount_acquire(&rblk->tfb_refcnt);
323 rw_runlock(&tcp_function_lock);
329 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
332 struct tcp_function_set fs;
333 struct tcp_function_block *blk;
335 memset(&fs, 0, sizeof(fs));
336 rw_rlock(&tcp_function_lock);
337 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
340 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
341 fs.pcbcnt = blk->tfb_refcnt;
343 rw_runlock(&tcp_function_lock);
344 error = sysctl_handle_string(oidp, fs.function_set_name,
345 sizeof(fs.function_set_name), req);
347 /* Check for error or no change */
348 if (error != 0 || req->newptr == NULL)
351 rw_wlock(&tcp_function_lock);
352 blk = find_tcp_functions_locked(&fs);
354 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
358 tcp_func_set_ptr = blk;
360 rw_wunlock(&tcp_function_lock);
364 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
365 CTLTYPE_STRING | CTLFLAG_RW,
366 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
367 "Set/get the default TCP functions");
370 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
372 int error, cnt, linesz;
373 struct tcp_function *f;
378 rw_rlock(&tcp_function_lock);
379 TAILQ_FOREACH(f, &t_functions, tf_next) {
382 rw_runlock(&tcp_function_lock);
384 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
385 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
390 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
395 rw_rlock(&tcp_function_lock);
396 TAILQ_FOREACH(f, &t_functions, tf_next) {
397 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
398 f->tf_fb->tfb_tcp_block_name,
399 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
400 f->tf_fb->tfb_refcnt);
401 if (linesz >= bufsz) {
409 rw_runlock(&tcp_function_lock);
411 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
412 free(buffer, M_TEMP);
416 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
417 CTLTYPE_STRING|CTLFLAG_RD,
418 NULL, 0, sysctl_net_inet_list_available, "A",
419 "list available TCP Function sets");
422 * Target size of TCP PCB hash tables. Must be a power of two.
424 * Note that this can be overridden by the kernel environment
425 * variable net.inet.tcp.tcbhashsize
428 #define TCBHASHSIZE 0
433 * Callouts should be moved into struct tcp directly. They are currently
434 * separate because the tcpcb structure is exported to userland for sysctl
435 * parsing purposes, which do not know about callouts.
444 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
445 #define V_tcpcb_zone VNET(tcpcb_zone)
447 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
448 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
450 static struct mtx isn_mtx;
452 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
453 #define ISN_LOCK() mtx_lock(&isn_mtx)
454 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
457 * TCP initialization.
460 tcp_zone_change(void *tag)
463 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
464 uma_zone_set_max(V_tcpcb_zone, maxsockets);
465 tcp_tw_zone_change();
469 tcp_inpcb_init(void *mem, int size, int flags)
471 struct inpcb *inp = mem;
473 INP_LOCK_INIT(inp, "inp", "tcpinp");
478 * Take a value and get the next power of 2 that doesn't overflow.
479 * Used to size the tcp_inpcb hash buckets.
482 maketcp_hashsize(int size)
488 * get the next power of 2 higher than maxsockets.
490 hashsize = 1 << fls(size);
491 /* catch overflow, and just go one power of 2 smaller */
492 if (hashsize < size) {
493 hashsize = 1 << (fls(size) - 1);
499 register_tcp_functions(struct tcp_function_block *blk, int wait)
501 struct tcp_function_block *lblk;
502 struct tcp_function *n;
503 struct tcp_function_set fs;
505 if (t_functions_inited == 0) {
506 init_tcp_functions();
508 if ((blk->tfb_tcp_output == NULL) ||
509 (blk->tfb_tcp_do_segment == NULL) ||
510 (blk->tfb_tcp_ctloutput == NULL) ||
511 (strlen(blk->tfb_tcp_block_name) == 0)) {
513 * These functions are required and you
518 if (blk->tfb_tcp_timer_stop_all ||
519 blk->tfb_tcp_timer_activate ||
520 blk->tfb_tcp_timer_active ||
521 blk->tfb_tcp_timer_stop) {
523 * If you define one timer function you
524 * must have them all.
526 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
527 (blk->tfb_tcp_timer_activate == NULL) ||
528 (blk->tfb_tcp_timer_active == NULL) ||
529 (blk->tfb_tcp_timer_stop == NULL)) {
533 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
538 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
539 rw_wlock(&tcp_function_lock);
540 lblk = find_tcp_functions_locked(&fs);
542 /* Duplicate name space not allowed */
543 rw_wunlock(&tcp_function_lock);
544 free(n, M_TCPFUNCTIONS);
547 refcount_init(&blk->tfb_refcnt, 0);
549 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
550 rw_wunlock(&tcp_function_lock);
555 deregister_tcp_functions(struct tcp_function_block *blk)
557 struct tcp_function_block *lblk;
558 struct tcp_function *f;
561 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
562 /* You can't un-register the default */
565 rw_wlock(&tcp_function_lock);
566 if (blk == tcp_func_set_ptr) {
567 /* You can't free the current default */
568 rw_wunlock(&tcp_function_lock);
571 if (blk->tfb_refcnt) {
572 /* Still tcb attached, mark it. */
573 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
574 rw_wunlock(&tcp_function_lock);
577 lblk = find_tcp_fb_locked(blk, &f);
580 TAILQ_REMOVE(&t_functions, f, tf_next);
582 free(f, M_TCPFUNCTIONS);
585 rw_wunlock(&tcp_function_lock);
592 const char *tcbhash_tuneable;
595 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
598 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
599 printf("%s: WARNING: unable to register helper hook\n", __func__);
600 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
601 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
602 printf("%s: WARNING: unable to register helper hook\n", __func__);
603 hashsize = TCBHASHSIZE;
604 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
607 * Auto tune the hash size based on maxsockets.
608 * A perfect hash would have a 1:1 mapping
609 * (hashsize = maxsockets) however it's been
610 * suggested that O(2) average is better.
612 hashsize = maketcp_hashsize(maxsockets / 4);
614 * Our historical default is 512,
615 * do not autotune lower than this.
619 if (bootverbose && IS_DEFAULT_VNET(curvnet))
620 printf("%s: %s auto tuned to %d\n", __func__,
621 tcbhash_tuneable, hashsize);
624 * We require a hashsize to be a power of two.
625 * Previously if it was not a power of two we would just reset it
626 * back to 512, which could be a nasty surprise if you did not notice
628 * Instead what we do is clip it to the closest power of two lower
629 * than the specified hash value.
631 if (!powerof2(hashsize)) {
632 int oldhashsize = hashsize;
634 hashsize = maketcp_hashsize(hashsize);
635 /* prevent absurdly low value */
638 printf("%s: WARNING: TCB hash size not a power of 2, "
639 "clipped from %d to %d.\n", __func__, oldhashsize,
642 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
643 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
646 * These have to be type stable for the benefit of the timers.
648 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
649 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
650 uma_zone_set_max(V_tcpcb_zone, maxsockets);
651 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
657 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
658 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
659 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
665 /* Skip initialization of globals for non-default instances. */
666 if (!IS_DEFAULT_VNET(curvnet))
669 tcp_reass_global_init();
671 /* XXX virtualize those bellow? */
672 tcp_delacktime = TCPTV_DELACK;
673 tcp_keepinit = TCPTV_KEEP_INIT;
674 tcp_keepidle = TCPTV_KEEP_IDLE;
675 tcp_keepintvl = TCPTV_KEEPINTVL;
676 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
678 tcp_rexmit_min = TCPTV_MIN;
679 if (tcp_rexmit_min < 1)
681 tcp_persmin = TCPTV_PERSMIN;
682 tcp_persmax = TCPTV_PERSMAX;
683 tcp_rexmit_slop = TCPTV_CPU_VAR;
684 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
685 tcp_tcbhashsize = hashsize;
686 /* Setup the tcp function block list */
687 init_tcp_functions();
688 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
689 read_random(&V_ts_offset_secret, sizeof(V_ts_offset_secret));
691 if (tcp_soreceive_stream) {
693 tcp_usrreqs.pru_soreceive = soreceive_stream;
696 tcp6_usrreqs.pru_soreceive = soreceive_stream;
701 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
703 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
705 if (max_protohdr < TCP_MINPROTOHDR)
706 max_protohdr = TCP_MINPROTOHDR;
707 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
709 #undef TCP_MINPROTOHDR
712 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
713 SHUTDOWN_PRI_DEFAULT);
714 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
715 EVENTHANDLER_PRI_ANY);
723 tcp_destroy(void *unused __unused)
728 * All our processes are gone, all our sockets should be cleaned
729 * up, which means, we should be past the tcp_discardcb() calls.
730 * Sleep to let all tcpcb timers really disappear and cleanup.
733 INP_LIST_RLOCK(&V_tcbinfo);
734 n = V_tcbinfo.ipi_count;
735 INP_LIST_RUNLOCK(&V_tcbinfo);
738 pause("tcpdes", hz / 10);
743 in_pcbinfo_destroy(&V_tcbinfo);
744 /* tcp_discardcb() clears the sack_holes up. */
745 uma_zdestroy(V_sack_hole_zone);
746 uma_zdestroy(V_tcpcb_zone);
750 * Cannot free the zone until all tcpcbs are released as we attach
751 * the allocations to them.
753 tcp_fastopen_destroy();
756 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
758 printf("%s: WARNING: unable to deregister helper hook "
759 "type=%d, id=%d: error %d returned\n", __func__,
760 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
762 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
764 printf("%s: WARNING: unable to deregister helper hook "
765 "type=%d, id=%d: error %d returned\n", __func__,
766 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
769 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
779 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
780 * tcp_template used to store this data in mbufs, but we now recopy it out
781 * of the tcpcb each time to conserve mbufs.
784 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
786 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
788 INP_WLOCK_ASSERT(inp);
791 if ((inp->inp_vflag & INP_IPV6) != 0) {
794 ip6 = (struct ip6_hdr *)ip_ptr;
795 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
796 (inp->inp_flow & IPV6_FLOWINFO_MASK);
797 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
798 (IPV6_VERSION & IPV6_VERSION_MASK);
799 ip6->ip6_nxt = IPPROTO_TCP;
800 ip6->ip6_plen = htons(sizeof(struct tcphdr));
801 ip6->ip6_src = inp->in6p_laddr;
802 ip6->ip6_dst = inp->in6p_faddr;
805 #if defined(INET6) && defined(INET)
812 ip = (struct ip *)ip_ptr;
813 ip->ip_v = IPVERSION;
815 ip->ip_tos = inp->inp_ip_tos;
819 ip->ip_ttl = inp->inp_ip_ttl;
821 ip->ip_p = IPPROTO_TCP;
822 ip->ip_src = inp->inp_laddr;
823 ip->ip_dst = inp->inp_faddr;
826 th->th_sport = inp->inp_lport;
827 th->th_dport = inp->inp_fport;
835 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
839 * Create template to be used to send tcp packets on a connection.
840 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
841 * use for this function is in keepalives, which use tcp_respond.
844 tcpip_maketemplate(struct inpcb *inp)
848 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
851 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
856 * Send a single message to the TCP at address specified by
857 * the given TCP/IP header. If m == NULL, then we make a copy
858 * of the tcpiphdr at th and send directly to the addressed host.
859 * This is used to force keep alive messages out using the TCP
860 * template for a connection. If flags are given then we send
861 * a message back to the TCP which originated the segment th,
862 * and discard the mbuf containing it and any other attached mbufs.
864 * In any case the ack and sequence number of the transmitted
865 * segment are as specified by the parameters.
867 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
870 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
871 tcp_seq ack, tcp_seq seq, int flags)
883 int optlen, tlen, win;
886 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
889 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
896 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
897 INP_WLOCK_ASSERT(inp);
904 if (!(flags & TH_RST)) {
905 win = sbspace(&inp->inp_socket->so_rcv);
906 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
907 win = (long)TCP_MAXWIN << tp->rcv_scale;
909 if ((tp->t_flags & TF_NOOPT) == 0)
913 m = m_gethdr(M_NOWAIT, MT_DATA);
916 m->m_data += max_linkhdr;
919 bcopy((caddr_t)ip6, mtod(m, caddr_t),
920 sizeof(struct ip6_hdr));
921 ip6 = mtod(m, struct ip6_hdr *);
922 nth = (struct tcphdr *)(ip6 + 1);
926 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
927 ip = mtod(m, struct ip *);
928 nth = (struct tcphdr *)(ip + 1);
930 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
932 } else if (!M_WRITABLE(m)) {
935 /* Can't reuse 'm', allocate a new mbuf. */
936 n = m_gethdr(M_NOWAIT, MT_DATA);
942 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
948 n->m_data += max_linkhdr;
949 /* m_len is set later */
950 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
953 bcopy((caddr_t)ip6, mtod(n, caddr_t),
954 sizeof(struct ip6_hdr));
955 ip6 = mtod(n, struct ip6_hdr *);
956 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
957 nth = (struct tcphdr *)(ip6 + 1);
961 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
962 ip = mtod(n, struct ip *);
963 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
964 nth = (struct tcphdr *)(ip + 1);
966 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
967 xchg(nth->th_dport, nth->th_sport, uint16_t);
974 * XXX MRT We inherit the FIB, which is lucky.
978 m->m_data = (caddr_t)ipgen;
979 /* m_len is set later */
982 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
983 nth = (struct tcphdr *)(ip6 + 1);
987 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
988 nth = (struct tcphdr *)(ip + 1);
992 * this is usually a case when an extension header
993 * exists between the IPv6 header and the
996 nth->th_sport = th->th_sport;
997 nth->th_dport = th->th_dport;
999 xchg(nth->th_dport, nth->th_sport, uint16_t);
1005 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1007 #if defined(INET) && defined(INET6)
1011 tlen = sizeof (struct tcpiphdr);
1015 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1016 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1017 m, tlen, (long)M_TRAILINGSPACE(m)));
1022 /* Make sure we have room. */
1023 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1024 m->m_next = m_get(M_NOWAIT, MT_DATA);
1026 optp = mtod(m->m_next, u_char *);
1031 optp = (u_char *) (nth + 1);
1037 if (tp->t_flags & TF_RCVD_TSTMP) {
1038 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1039 to.to_tsecr = tp->ts_recent;
1040 to.to_flags |= TOF_TS;
1042 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1043 /* TCP-MD5 (RFC2385). */
1044 if (tp->t_flags & TF_SIGNATURE)
1045 to.to_flags |= TOF_SIGNATURE;
1047 /* Add the options. */
1048 tlen += optlen = tcp_addoptions(&to, optp);
1050 /* Update m_len in the correct mbuf. */
1051 optm->m_len += optlen;
1057 ip6->ip6_vfc = IPV6_VERSION;
1058 ip6->ip6_nxt = IPPROTO_TCP;
1059 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1062 #if defined(INET) && defined(INET6)
1067 ip->ip_len = htons(tlen);
1068 ip->ip_ttl = V_ip_defttl;
1069 if (V_path_mtu_discovery)
1070 ip->ip_off |= htons(IP_DF);
1073 m->m_pkthdr.len = tlen;
1074 m->m_pkthdr.rcvif = NULL;
1078 * Packet is associated with a socket, so allow the
1079 * label of the response to reflect the socket label.
1081 INP_WLOCK_ASSERT(inp);
1082 mac_inpcb_create_mbuf(inp, m);
1085 * Packet is not associated with a socket, so possibly
1086 * update the label in place.
1088 mac_netinet_tcp_reply(m);
1091 nth->th_seq = htonl(seq);
1092 nth->th_ack = htonl(ack);
1094 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1095 nth->th_flags = flags;
1097 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1099 nth->th_win = htons((u_short)win);
1102 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1103 if (to.to_flags & TOF_SIGNATURE) {
1104 if (!TCPMD5_ENABLED() ||
1105 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1112 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1115 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1116 nth->th_sum = in6_cksum_pseudo(ip6,
1117 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1118 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1122 #if defined(INET6) && defined(INET)
1127 m->m_pkthdr.csum_flags = CSUM_TCP;
1128 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1129 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1133 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1134 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1136 TCP_PROBE3(debug__output, tp, th, m);
1138 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1142 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1143 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1146 #if defined(INET) && defined(INET6)
1151 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1152 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1158 * Create a new TCP control block, making an
1159 * empty reassembly queue and hooking it to the argument
1160 * protocol control block. The `inp' parameter must have
1161 * come from the zone allocator set up in tcp_init().
1164 tcp_newtcpcb(struct inpcb *inp)
1166 struct tcpcb_mem *tm;
1169 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1172 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1177 /* Initialise cc_var struct for this tcpcb. */
1179 tp->ccv->type = IPPROTO_TCP;
1180 tp->ccv->ccvc.tcp = tp;
1181 rw_rlock(&tcp_function_lock);
1182 tp->t_fb = tcp_func_set_ptr;
1183 refcount_acquire(&tp->t_fb->tfb_refcnt);
1184 rw_runlock(&tcp_function_lock);
1185 if (tp->t_fb->tfb_tcp_fb_init) {
1186 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1189 * Use the current system default CC algorithm.
1192 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1193 CC_ALGO(tp) = CC_DEFAULT();
1196 if (CC_ALGO(tp)->cb_init != NULL)
1197 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1198 if (tp->t_fb->tfb_tcp_fb_fini)
1199 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1200 refcount_release(&tp->t_fb->tfb_refcnt);
1201 uma_zfree(V_tcpcb_zone, tm);
1206 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1207 if (tp->t_fb->tfb_tcp_fb_fini)
1208 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1209 refcount_release(&tp->t_fb->tfb_refcnt);
1210 uma_zfree(V_tcpcb_zone, tm);
1215 tp->t_vnet = inp->inp_vnet;
1217 tp->t_timers = &tm->tt;
1218 TAILQ_INIT(&tp->t_segq);
1221 isipv6 ? V_tcp_v6mssdflt :
1225 /* Set up our timeouts. */
1226 callout_init(&tp->t_timers->tt_rexmt, 1);
1227 callout_init(&tp->t_timers->tt_persist, 1);
1228 callout_init(&tp->t_timers->tt_keep, 1);
1229 callout_init(&tp->t_timers->tt_2msl, 1);
1230 callout_init(&tp->t_timers->tt_delack, 1);
1232 if (V_tcp_do_rfc1323)
1233 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1235 tp->t_flags |= TF_SACK_PERMIT;
1236 TAILQ_INIT(&tp->snd_holes);
1238 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1241 in_pcbref(inp); /* Reference for tcpcb */
1245 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1246 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1247 * reasonable initial retransmit time.
1249 tp->t_srtt = TCPTV_SRTTBASE;
1250 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1251 tp->t_rttmin = tcp_rexmit_min;
1252 tp->t_rxtcur = TCPTV_RTOBASE;
1253 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1254 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1255 tp->t_rcvtime = ticks;
1257 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1258 * because the socket may be bound to an IPv6 wildcard address,
1259 * which may match an IPv4-mapped IPv6 address.
1261 inp->inp_ip_ttl = V_ip_defttl;
1265 * Init the TCP PCAP queues.
1267 tcp_pcap_tcpcb_init(tp);
1269 return (tp); /* XXX */
1273 * Switch the congestion control algorithm back to NewReno for any active
1274 * control blocks using an algorithm which is about to go away.
1275 * This ensures the CC framework can allow the unload to proceed without leaving
1276 * any dangling pointers which would trigger a panic.
1277 * Returning non-zero would inform the CC framework that something went wrong
1278 * and it would be unsafe to allow the unload to proceed. However, there is no
1279 * way for this to occur with this implementation so we always return zero.
1282 tcp_ccalgounload(struct cc_algo *unload_algo)
1284 struct cc_algo *tmpalgo;
1287 VNET_ITERATOR_DECL(vnet_iter);
1290 * Check all active control blocks across all network stacks and change
1291 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1292 * requires cleanup code to be run, call it.
1295 VNET_FOREACH(vnet_iter) {
1296 CURVNET_SET(vnet_iter);
1297 INP_INFO_WLOCK(&V_tcbinfo);
1299 * New connections already part way through being initialised
1300 * with the CC algo we're removing will not race with this code
1301 * because the INP_INFO_WLOCK is held during initialisation. We
1302 * therefore don't enter the loop below until the connection
1303 * list has stabilised.
1305 LIST_FOREACH(inp, &V_tcb, inp_list) {
1307 /* Important to skip tcptw structs. */
1308 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1309 (tp = intotcpcb(inp)) != NULL) {
1311 * By holding INP_WLOCK here, we are assured
1312 * that the connection is not currently
1313 * executing inside the CC module's functions
1314 * i.e. it is safe to make the switch back to
1317 if (CC_ALGO(tp) == unload_algo) {
1318 tmpalgo = CC_ALGO(tp);
1319 /* NewReno does not require any init. */
1320 CC_ALGO(tp) = &newreno_cc_algo;
1321 if (tmpalgo->cb_destroy != NULL)
1322 tmpalgo->cb_destroy(tp->ccv);
1327 INP_INFO_WUNLOCK(&V_tcbinfo);
1330 VNET_LIST_RUNLOCK();
1336 * Drop a TCP connection, reporting
1337 * the specified error. If connection is synchronized,
1338 * then send a RST to peer.
1341 tcp_drop(struct tcpcb *tp, int errno)
1343 struct socket *so = tp->t_inpcb->inp_socket;
1345 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1346 INP_WLOCK_ASSERT(tp->t_inpcb);
1348 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1349 tcp_state_change(tp, TCPS_CLOSED);
1350 (void) tp->t_fb->tfb_tcp_output(tp);
1351 TCPSTAT_INC(tcps_drops);
1353 TCPSTAT_INC(tcps_conndrops);
1354 if (errno == ETIMEDOUT && tp->t_softerror)
1355 errno = tp->t_softerror;
1356 so->so_error = errno;
1357 return (tcp_close(tp));
1361 tcp_discardcb(struct tcpcb *tp)
1363 struct inpcb *inp = tp->t_inpcb;
1364 struct socket *so = inp->inp_socket;
1366 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1370 INP_WLOCK_ASSERT(inp);
1373 * Make sure that all of our timers are stopped before we delete the
1376 * If stopping a timer fails, we schedule a discard function in same
1377 * callout, and the last discard function called will take care of
1378 * deleting the tcpcb.
1380 tp->t_timers->tt_draincnt = 0;
1381 tcp_timer_stop(tp, TT_REXMT);
1382 tcp_timer_stop(tp, TT_PERSIST);
1383 tcp_timer_stop(tp, TT_KEEP);
1384 tcp_timer_stop(tp, TT_2MSL);
1385 tcp_timer_stop(tp, TT_DELACK);
1386 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1388 * Call the stop-all function of the methods,
1389 * this function should call the tcp_timer_stop()
1390 * method with each of the function specific timeouts.
1391 * That stop will be called via the tfb_tcp_timer_stop()
1392 * which should use the async drain function of the
1393 * callout system (see tcp_var.h).
1395 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1399 * If we got enough samples through the srtt filter,
1400 * save the rtt and rttvar in the routing entry.
1401 * 'Enough' is arbitrarily defined as 4 rtt samples.
1402 * 4 samples is enough for the srtt filter to converge
1403 * to within enough % of the correct value; fewer samples
1404 * and we could save a bogus rtt. The danger is not high
1405 * as tcp quickly recovers from everything.
1406 * XXX: Works very well but needs some more statistics!
1408 if (tp->t_rttupdated >= 4) {
1409 struct hc_metrics_lite metrics;
1412 bzero(&metrics, sizeof(metrics));
1414 * Update the ssthresh always when the conditions below
1415 * are satisfied. This gives us better new start value
1416 * for the congestion avoidance for new connections.
1417 * ssthresh is only set if packet loss occurred on a session.
1419 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1420 * being torn down. Ideally this code would not use 'so'.
1422 ssthresh = tp->snd_ssthresh;
1423 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1425 * convert the limit from user data bytes to
1426 * packets then to packet data bytes.
1428 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1431 ssthresh *= (u_long)(tp->t_maxseg +
1433 (isipv6 ? sizeof (struct ip6_hdr) +
1434 sizeof (struct tcphdr) :
1436 sizeof (struct tcpiphdr)
1443 metrics.rmx_ssthresh = ssthresh;
1445 metrics.rmx_rtt = tp->t_srtt;
1446 metrics.rmx_rttvar = tp->t_rttvar;
1447 metrics.rmx_cwnd = tp->snd_cwnd;
1448 metrics.rmx_sendpipe = 0;
1449 metrics.rmx_recvpipe = 0;
1451 tcp_hc_update(&inp->inp_inc, &metrics);
1454 /* free the reassembly queue, if any */
1455 tcp_reass_flush(tp);
1458 /* Disconnect offload device, if any. */
1459 if (tp->t_flags & TF_TOE)
1460 tcp_offload_detach(tp);
1463 tcp_free_sackholes(tp);
1466 /* Free the TCP PCAP queues. */
1467 tcp_pcap_drain(&(tp->t_inpkts));
1468 tcp_pcap_drain(&(tp->t_outpkts));
1471 /* Allow the CC algorithm to clean up after itself. */
1472 if (CC_ALGO(tp)->cb_destroy != NULL)
1473 CC_ALGO(tp)->cb_destroy(tp->ccv);
1475 khelp_destroy_osd(tp->osd);
1478 inp->inp_ppcb = NULL;
1479 if (tp->t_timers->tt_draincnt == 0) {
1480 /* We own the last reference on tcpcb, let's free it. */
1481 TCPSTATES_DEC(tp->t_state);
1482 if (tp->t_fb->tfb_tcp_fb_fini)
1483 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1484 refcount_release(&tp->t_fb->tfb_refcnt);
1486 uma_zfree(V_tcpcb_zone, tp);
1487 released = in_pcbrele_wlocked(inp);
1488 KASSERT(!released, ("%s: inp %p should not have been released "
1489 "here", __func__, inp));
1494 tcp_timer_discard(void *ptp)
1499 tp = (struct tcpcb *)ptp;
1500 CURVNET_SET(tp->t_vnet);
1501 INP_INFO_RLOCK(&V_tcbinfo);
1503 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1506 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1507 ("%s: tcpcb has to be stopped here", __func__));
1508 tp->t_timers->tt_draincnt--;
1509 if (tp->t_timers->tt_draincnt == 0) {
1510 /* We own the last reference on this tcpcb, let's free it. */
1511 TCPSTATES_DEC(tp->t_state);
1512 if (tp->t_fb->tfb_tcp_fb_fini)
1513 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1514 refcount_release(&tp->t_fb->tfb_refcnt);
1516 uma_zfree(V_tcpcb_zone, tp);
1517 if (in_pcbrele_wlocked(inp)) {
1518 INP_INFO_RUNLOCK(&V_tcbinfo);
1524 INP_INFO_RUNLOCK(&V_tcbinfo);
1529 * Attempt to close a TCP control block, marking it as dropped, and freeing
1530 * the socket if we hold the only reference.
1533 tcp_close(struct tcpcb *tp)
1535 struct inpcb *inp = tp->t_inpcb;
1538 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1539 INP_WLOCK_ASSERT(inp);
1542 if (tp->t_state == TCPS_LISTEN)
1543 tcp_offload_listen_stop(tp);
1547 * This releases the TFO pending counter resource for TFO listen
1548 * sockets as well as passively-created TFO sockets that transition
1549 * from SYN_RECEIVED to CLOSED.
1551 if (tp->t_tfo_pending) {
1552 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1553 tp->t_tfo_pending = NULL;
1557 TCPSTAT_INC(tcps_closed);
1558 if (tp->t_state != TCPS_CLOSED)
1559 tcp_state_change(tp, TCPS_CLOSED);
1560 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1561 so = inp->inp_socket;
1562 soisdisconnected(so);
1563 if (inp->inp_flags & INP_SOCKREF) {
1564 KASSERT(so->so_state & SS_PROTOREF,
1565 ("tcp_close: !SS_PROTOREF"));
1566 inp->inp_flags &= ~INP_SOCKREF;
1570 so->so_state &= ~SS_PROTOREF;
1580 VNET_ITERATOR_DECL(vnet_iter);
1585 VNET_LIST_RLOCK_NOSLEEP();
1586 VNET_FOREACH(vnet_iter) {
1587 CURVNET_SET(vnet_iter);
1592 * Walk the tcpbs, if existing, and flush the reassembly queue,
1593 * if there is one...
1594 * XXX: The "Net/3" implementation doesn't imply that the TCP
1595 * reassembly queue should be flushed, but in a situation
1596 * where we're really low on mbufs, this is potentially
1599 INP_INFO_WLOCK(&V_tcbinfo);
1600 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1601 if (inpb->inp_flags & INP_TIMEWAIT)
1604 if ((tcpb = intotcpcb(inpb)) != NULL) {
1605 tcp_reass_flush(tcpb);
1606 tcp_clean_sackreport(tcpb);
1608 if (tcp_pcap_aggressive_free) {
1609 /* Free the TCP PCAP queues. */
1610 tcp_pcap_drain(&(tcpb->t_inpkts));
1611 tcp_pcap_drain(&(tcpb->t_outpkts));
1617 INP_INFO_WUNLOCK(&V_tcbinfo);
1620 VNET_LIST_RUNLOCK_NOSLEEP();
1624 * Notify a tcp user of an asynchronous error;
1625 * store error as soft error, but wake up user
1626 * (for now, won't do anything until can select for soft error).
1628 * Do not wake up user since there currently is no mechanism for
1629 * reporting soft errors (yet - a kqueue filter may be added).
1631 static struct inpcb *
1632 tcp_notify(struct inpcb *inp, int error)
1636 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1637 INP_WLOCK_ASSERT(inp);
1639 if ((inp->inp_flags & INP_TIMEWAIT) ||
1640 (inp->inp_flags & INP_DROPPED))
1643 tp = intotcpcb(inp);
1644 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1647 * Ignore some errors if we are hooked up.
1648 * If connection hasn't completed, has retransmitted several times,
1649 * and receives a second error, give up now. This is better
1650 * than waiting a long time to establish a connection that
1651 * can never complete.
1653 if (tp->t_state == TCPS_ESTABLISHED &&
1654 (error == EHOSTUNREACH || error == ENETUNREACH ||
1655 error == EHOSTDOWN)) {
1656 if (inp->inp_route.ro_rt) {
1657 RTFREE(inp->inp_route.ro_rt);
1658 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1661 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1663 tp = tcp_drop(tp, error);
1669 tp->t_softerror = error;
1673 wakeup( &so->so_timeo);
1680 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1682 int error, i, m, n, pcb_count;
1683 struct inpcb *inp, **inp_list;
1688 * The process of preparing the TCB list is too time-consuming and
1689 * resource-intensive to repeat twice on every request.
1691 if (req->oldptr == NULL) {
1692 n = V_tcbinfo.ipi_count +
1693 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1694 n += imax(n / 8, 10);
1695 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1699 if (req->newptr != NULL)
1703 * OK, now we're committed to doing something.
1705 INP_LIST_RLOCK(&V_tcbinfo);
1706 gencnt = V_tcbinfo.ipi_gencnt;
1707 n = V_tcbinfo.ipi_count;
1708 INP_LIST_RUNLOCK(&V_tcbinfo);
1710 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1712 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1713 + (n + m) * sizeof(struct xtcpcb));
1717 bzero(&xig, sizeof(xig));
1718 xig.xig_len = sizeof xig;
1719 xig.xig_count = n + m;
1720 xig.xig_gen = gencnt;
1721 xig.xig_sogen = so_gencnt;
1722 error = SYSCTL_OUT(req, &xig, sizeof xig);
1726 error = syncache_pcblist(req, m, &pcb_count);
1730 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1732 INP_INFO_WLOCK(&V_tcbinfo);
1733 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1734 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1736 if (inp->inp_gencnt <= gencnt) {
1738 * XXX: This use of cr_cansee(), introduced with
1739 * TCP state changes, is not quite right, but for
1740 * now, better than nothing.
1742 if (inp->inp_flags & INP_TIMEWAIT) {
1743 if (intotw(inp) != NULL)
1744 error = cr_cansee(req->td->td_ucred,
1745 intotw(inp)->tw_cred);
1747 error = EINVAL; /* Skip this inp. */
1749 error = cr_canseeinpcb(req->td->td_ucred, inp);
1752 inp_list[i++] = inp;
1757 INP_INFO_WUNLOCK(&V_tcbinfo);
1761 for (i = 0; i < n; i++) {
1764 if (inp->inp_gencnt <= gencnt) {
1768 bzero(&xt, sizeof(xt));
1769 xt.xt_len = sizeof xt;
1770 /* XXX should avoid extra copy */
1771 bcopy(inp, &xt.xt_inp, sizeof *inp);
1772 inp_ppcb = inp->inp_ppcb;
1773 if (inp_ppcb == NULL)
1774 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1775 else if (inp->inp_flags & INP_TIMEWAIT) {
1776 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1777 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1779 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1780 if (xt.xt_tp.t_timers)
1781 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1783 if (inp->inp_socket != NULL)
1784 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1786 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1787 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1789 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1791 error = SYSCTL_OUT(req, &xt, sizeof xt);
1795 INP_INFO_RLOCK(&V_tcbinfo);
1796 for (i = 0; i < n; i++) {
1799 if (!in_pcbrele_rlocked(inp))
1802 INP_INFO_RUNLOCK(&V_tcbinfo);
1806 * Give the user an updated idea of our state.
1807 * If the generation differs from what we told
1808 * her before, she knows that something happened
1809 * while we were processing this request, and it
1810 * might be necessary to retry.
1812 INP_LIST_RLOCK(&V_tcbinfo);
1813 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1814 xig.xig_sogen = so_gencnt;
1815 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1816 INP_LIST_RUNLOCK(&V_tcbinfo);
1817 error = SYSCTL_OUT(req, &xig, sizeof xig);
1819 free(inp_list, M_TEMP);
1823 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1824 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1825 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1829 tcp_getcred(SYSCTL_HANDLER_ARGS)
1832 struct sockaddr_in addrs[2];
1836 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1839 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1842 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1843 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1845 if (inp->inp_socket == NULL)
1848 error = cr_canseeinpcb(req->td->td_ucred, inp);
1850 cru2x(inp->inp_cred, &xuc);
1855 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1859 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1860 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1861 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1866 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1869 struct sockaddr_in6 addrs[2];
1876 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1879 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1882 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1883 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1886 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1888 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1897 inp = in_pcblookup(&V_tcbinfo,
1898 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1900 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1901 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1904 inp = in6_pcblookup(&V_tcbinfo,
1905 &addrs[1].sin6_addr, addrs[1].sin6_port,
1906 &addrs[0].sin6_addr, addrs[0].sin6_port,
1907 INPLOOKUP_RLOCKPCB, NULL);
1909 if (inp->inp_socket == NULL)
1912 error = cr_canseeinpcb(req->td->td_ucred, inp);
1914 cru2x(inp->inp_cred, &xuc);
1919 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1923 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1924 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1925 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1931 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1933 struct ip *ip = vip;
1935 struct in_addr faddr;
1938 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1940 struct in_conninfo inc;
1941 tcp_seq icmp_tcp_seq;
1944 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1945 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1948 if (cmd == PRC_MSGSIZE)
1949 notify = tcp_mtudisc_notify;
1950 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1951 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1952 cmd == PRC_TIMXCEED_INTRANS) && ip)
1953 notify = tcp_drop_syn_sent;
1956 * Hostdead is ugly because it goes linearly through all PCBs.
1957 * XXX: We never get this from ICMP, otherwise it makes an
1958 * excellent DoS attack on machines with many connections.
1960 else if (cmd == PRC_HOSTDEAD)
1962 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1966 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1970 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1971 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1972 INP_INFO_RLOCK(&V_tcbinfo);
1973 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1974 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1975 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1976 /* signal EHOSTDOWN, as it flushes the cached route */
1977 inp = (*notify)(inp, EHOSTDOWN);
1980 icmp_tcp_seq = th->th_seq;
1982 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1983 !(inp->inp_flags & INP_DROPPED) &&
1984 !(inp->inp_socket == NULL)) {
1985 tp = intotcpcb(inp);
1986 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
1987 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
1988 if (cmd == PRC_MSGSIZE) {
1991 * If we got a needfrag set the MTU
1992 * in the route to the suggested new
1993 * value (if given) and then notify.
1995 mtu = ntohs(icp->icmp_nextmtu);
1997 * If no alternative MTU was
1998 * proposed, try the next smaller
2003 ntohs(ip->ip_len), 1);
2004 if (mtu < V_tcp_minmss +
2005 sizeof(struct tcpiphdr))
2006 mtu = V_tcp_minmss +
2007 sizeof(struct tcpiphdr);
2009 * Only process the offered MTU if it
2010 * is smaller than the current one.
2012 if (mtu < tp->t_maxseg +
2013 sizeof(struct tcpiphdr)) {
2014 bzero(&inc, sizeof(inc));
2015 inc.inc_faddr = faddr;
2017 inp->inp_inc.inc_fibnum;
2018 tcp_hc_updatemtu(&inc, mtu);
2019 tcp_mtudisc(inp, mtu);
2022 inp = (*notify)(inp,
2023 inetctlerrmap[cmd]);
2027 bzero(&inc, sizeof(inc));
2028 inc.inc_fport = th->th_dport;
2029 inc.inc_lport = th->th_sport;
2030 inc.inc_faddr = faddr;
2031 inc.inc_laddr = ip->ip_src;
2032 syncache_unreach(&inc, icmp_tcp_seq);
2037 INP_INFO_RUNLOCK(&V_tcbinfo);
2043 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2045 struct in6_addr *dst;
2046 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2047 struct ip6_hdr *ip6;
2051 struct icmp6_hdr *icmp6;
2052 struct ip6ctlparam *ip6cp = NULL;
2053 const struct sockaddr_in6 *sa6_src = NULL;
2054 struct in_conninfo inc;
2059 tcp_seq icmp_tcp_seq;
2063 if (sa->sa_family != AF_INET6 ||
2064 sa->sa_len != sizeof(struct sockaddr_in6))
2067 /* if the parameter is from icmp6, decode it. */
2069 ip6cp = (struct ip6ctlparam *)d;
2070 icmp6 = ip6cp->ip6c_icmp6;
2072 ip6 = ip6cp->ip6c_ip6;
2073 off = ip6cp->ip6c_off;
2074 sa6_src = ip6cp->ip6c_src;
2075 dst = ip6cp->ip6c_finaldst;
2079 off = 0; /* fool gcc */
2084 if (cmd == PRC_MSGSIZE)
2085 notify = tcp_mtudisc_notify;
2086 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2087 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2088 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2089 notify = tcp_drop_syn_sent;
2092 * Hostdead is ugly because it goes linearly through all PCBs.
2093 * XXX: We never get this from ICMP, otherwise it makes an
2094 * excellent DoS attack on machines with many connections.
2096 else if (cmd == PRC_HOSTDEAD)
2098 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2102 in6_pcbnotify(&V_tcbinfo, sa, 0,
2103 (const struct sockaddr *)sa6_src,
2104 0, cmd, NULL, notify);
2108 /* Check if we can safely get the ports from the tcp hdr */
2111 (int32_t) (off + sizeof(struct tcp_ports)))) {
2114 bzero(&t_ports, sizeof(struct tcp_ports));
2115 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2116 INP_INFO_RLOCK(&V_tcbinfo);
2117 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2118 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2119 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2120 /* signal EHOSTDOWN, as it flushes the cached route */
2121 inp = (*notify)(inp, EHOSTDOWN);
2124 off += sizeof(struct tcp_ports);
2125 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2128 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2130 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2131 !(inp->inp_flags & INP_DROPPED) &&
2132 !(inp->inp_socket == NULL)) {
2133 tp = intotcpcb(inp);
2134 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2135 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2136 if (cmd == PRC_MSGSIZE) {
2139 * If we got a needfrag set the MTU
2140 * in the route to the suggested new
2141 * value (if given) and then notify.
2143 mtu = ntohl(icmp6->icmp6_mtu);
2145 * If no alternative MTU was
2146 * proposed, or the proposed
2147 * MTU was too small, set to
2150 if (mtu < IPV6_MMTU)
2151 mtu = IPV6_MMTU - 8;
2152 bzero(&inc, sizeof(inc));
2153 inc.inc_fibnum = M_GETFIB(m);
2154 inc.inc_flags |= INC_ISIPV6;
2155 inc.inc6_faddr = *dst;
2156 if (in6_setscope(&inc.inc6_faddr,
2157 m->m_pkthdr.rcvif, NULL))
2160 * Only process the offered MTU if it
2161 * is smaller than the current one.
2163 if (mtu < tp->t_maxseg +
2164 sizeof (struct tcphdr) +
2165 sizeof (struct ip6_hdr)) {
2166 tcp_hc_updatemtu(&inc, mtu);
2167 tcp_mtudisc(inp, mtu);
2168 ICMP6STAT_INC(icp6s_pmtuchg);
2171 inp = (*notify)(inp,
2172 inet6ctlerrmap[cmd]);
2176 bzero(&inc, sizeof(inc));
2177 inc.inc_fibnum = M_GETFIB(m);
2178 inc.inc_flags |= INC_ISIPV6;
2179 inc.inc_fport = t_ports.th_dport;
2180 inc.inc_lport = t_ports.th_sport;
2181 inc.inc6_faddr = *dst;
2182 inc.inc6_laddr = ip6->ip6_src;
2183 syncache_unreach(&inc, icmp_tcp_seq);
2188 INP_INFO_RUNLOCK(&V_tcbinfo);
2193 tcp_keyed_hash(struct in_conninfo *inc, u_char *key)
2199 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2200 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2201 switch (inc->inc_flags & INC_ISIPV6) {
2204 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2205 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2210 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2211 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2215 MD5Update(&ctx, key, 32);
2216 MD5Final((unsigned char *)hash, &ctx);
2222 tcp_new_ts_offset(struct in_conninfo *inc)
2224 return (tcp_keyed_hash(inc, V_ts_offset_secret));
2228 * Following is where TCP initial sequence number generation occurs.
2230 * There are two places where we must use initial sequence numbers:
2231 * 1. In SYN-ACK packets.
2232 * 2. In SYN packets.
2234 * All ISNs for SYN-ACK packets are generated by the syncache. See
2235 * tcp_syncache.c for details.
2237 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2238 * depends on this property. In addition, these ISNs should be
2239 * unguessable so as to prevent connection hijacking. To satisfy
2240 * the requirements of this situation, the algorithm outlined in
2241 * RFC 1948 is used, with only small modifications.
2243 * Implementation details:
2245 * Time is based off the system timer, and is corrected so that it
2246 * increases by one megabyte per second. This allows for proper
2247 * recycling on high speed LANs while still leaving over an hour
2250 * As reading the *exact* system time is too expensive to be done
2251 * whenever setting up a TCP connection, we increment the time
2252 * offset in two ways. First, a small random positive increment
2253 * is added to isn_offset for each connection that is set up.
2254 * Second, the function tcp_isn_tick fires once per clock tick
2255 * and increments isn_offset as necessary so that sequence numbers
2256 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2257 * random positive increments serve only to ensure that the same
2258 * exact sequence number is never sent out twice (as could otherwise
2259 * happen when a port is recycled in less than the system tick
2262 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2263 * between seeding of isn_secret. This is normally set to zero,
2264 * as reseeding should not be necessary.
2266 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2267 * isn_offset_old, and isn_ctx is performed using the ISN lock. In
2268 * general, this means holding an exclusive (write) lock.
2271 #define ISN_BYTES_PER_SECOND 1048576
2272 #define ISN_STATIC_INCREMENT 4096
2273 #define ISN_RANDOM_INCREMENT (4096 - 1)
2275 static VNET_DEFINE(u_char, isn_secret[32]);
2276 static VNET_DEFINE(int, isn_last);
2277 static VNET_DEFINE(int, isn_last_reseed);
2278 static VNET_DEFINE(u_int32_t, isn_offset);
2279 static VNET_DEFINE(u_int32_t, isn_offset_old);
2281 #define V_isn_secret VNET(isn_secret)
2282 #define V_isn_last VNET(isn_last)
2283 #define V_isn_last_reseed VNET(isn_last_reseed)
2284 #define V_isn_offset VNET(isn_offset)
2285 #define V_isn_offset_old VNET(isn_offset_old)
2288 tcp_new_isn(struct in_conninfo *inc)
2291 u_int32_t projected_offset;
2294 /* Seed if this is the first use, reseed if requested. */
2295 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2296 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2298 read_random(&V_isn_secret, sizeof(V_isn_secret));
2299 V_isn_last_reseed = ticks;
2302 /* Compute the md5 hash and return the ISN. */
2303 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret);
2304 V_isn_offset += ISN_STATIC_INCREMENT +
2305 (arc4random() & ISN_RANDOM_INCREMENT);
2306 if (ticks != V_isn_last) {
2307 projected_offset = V_isn_offset_old +
2308 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2309 if (SEQ_GT(projected_offset, V_isn_offset))
2310 V_isn_offset = projected_offset;
2311 V_isn_offset_old = V_isn_offset;
2314 new_isn += V_isn_offset;
2320 * When a specific ICMP unreachable message is received and the
2321 * connection state is SYN-SENT, drop the connection. This behavior
2322 * is controlled by the icmp_may_rst sysctl.
2325 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2329 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2330 INP_WLOCK_ASSERT(inp);
2332 if ((inp->inp_flags & INP_TIMEWAIT) ||
2333 (inp->inp_flags & INP_DROPPED))
2336 tp = intotcpcb(inp);
2337 if (tp->t_state != TCPS_SYN_SENT)
2340 tp = tcp_drop(tp, errno);
2348 * When `need fragmentation' ICMP is received, update our idea of the MSS
2349 * based on the new value. Also nudge TCP to send something, since we
2350 * know the packet we just sent was dropped.
2351 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2353 static struct inpcb *
2354 tcp_mtudisc_notify(struct inpcb *inp, int error)
2357 tcp_mtudisc(inp, -1);
2362 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2367 INP_WLOCK_ASSERT(inp);
2368 if ((inp->inp_flags & INP_TIMEWAIT) ||
2369 (inp->inp_flags & INP_DROPPED))
2372 tp = intotcpcb(inp);
2373 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2375 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2377 so = inp->inp_socket;
2378 SOCKBUF_LOCK(&so->so_snd);
2379 /* If the mss is larger than the socket buffer, decrease the mss. */
2380 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2381 tp->t_maxseg = so->so_snd.sb_hiwat;
2382 SOCKBUF_UNLOCK(&so->so_snd);
2384 TCPSTAT_INC(tcps_mturesent);
2386 tp->snd_nxt = tp->snd_una;
2387 tcp_free_sackholes(tp);
2388 tp->snd_recover = tp->snd_max;
2389 if (tp->t_flags & TF_SACK_PERMIT)
2390 EXIT_FASTRECOVERY(tp->t_flags);
2391 tp->t_fb->tfb_tcp_output(tp);
2396 * Look-up the routing entry to the peer of this inpcb. If no route
2397 * is found and it cannot be allocated, then return 0. This routine
2398 * is called by TCP routines that access the rmx structure and by
2399 * tcp_mss_update to get the peer/interface MTU.
2402 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2404 struct nhop4_extended nh4;
2408 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2410 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2412 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2413 NHR_REF, 0, &nh4) != 0)
2417 maxmtu = nh4.nh_mtu;
2419 /* Report additional interface capabilities. */
2421 if (ifp->if_capenable & IFCAP_TSO4 &&
2422 ifp->if_hwassist & CSUM_TSO) {
2423 cap->ifcap |= CSUM_TSO;
2424 cap->tsomax = ifp->if_hw_tsomax;
2425 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2426 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2429 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2437 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2439 struct nhop6_extended nh6;
2440 struct in6_addr dst6;
2445 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2447 if (inc->inc_flags & INC_IPV6MINMTU)
2450 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2451 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2452 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2457 maxmtu = nh6.nh_mtu;
2459 /* Report additional interface capabilities. */
2461 if (ifp->if_capenable & IFCAP_TSO6 &&
2462 ifp->if_hwassist & CSUM_TSO) {
2463 cap->ifcap |= CSUM_TSO;
2464 cap->tsomax = ifp->if_hw_tsomax;
2465 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2466 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2469 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2477 * Calculate effective SMSS per RFC5681 definition for a given TCP
2478 * connection at its current state, taking into account SACK and etc.
2481 tcp_maxseg(const struct tcpcb *tp)
2485 if (tp->t_flags & TF_NOOPT)
2486 return (tp->t_maxseg);
2489 * Here we have a simplified code from tcp_addoptions(),
2490 * without a proper loop, and having most of paddings hardcoded.
2491 * We might make mistakes with padding here in some edge cases,
2492 * but this is harmless, since result of tcp_maxseg() is used
2493 * only in cwnd and ssthresh estimations.
2495 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2496 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2497 if (tp->t_flags & TF_RCVD_TSTMP)
2498 optlen = TCPOLEN_TSTAMP_APPA;
2501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2502 if (tp->t_flags & TF_SIGNATURE)
2503 optlen += PAD(TCPOLEN_SIGNATURE);
2505 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2506 optlen += TCPOLEN_SACKHDR;
2507 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2508 optlen = PAD(optlen);
2511 if (tp->t_flags & TF_REQ_TSTMP)
2512 optlen = TCPOLEN_TSTAMP_APPA;
2514 optlen = PAD(TCPOLEN_MAXSEG);
2515 if (tp->t_flags & TF_REQ_SCALE)
2516 optlen += PAD(TCPOLEN_WINDOW);
2517 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2518 if (tp->t_flags & TF_SIGNATURE)
2519 optlen += PAD(TCPOLEN_SIGNATURE);
2521 if (tp->t_flags & TF_SACK_PERMIT)
2522 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2525 optlen = min(optlen, TCP_MAXOLEN);
2526 return (tp->t_maxseg - optlen);
2530 sysctl_drop(SYSCTL_HANDLER_ARGS)
2532 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2533 struct sockaddr_storage addrs[2];
2537 struct sockaddr_in *fin, *lin;
2539 struct sockaddr_in6 *fin6, *lin6;
2550 if (req->oldptr != NULL || req->oldlen != 0)
2552 if (req->newptr == NULL)
2554 if (req->newlen < sizeof(addrs))
2556 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2560 switch (addrs[0].ss_family) {
2563 fin6 = (struct sockaddr_in6 *)&addrs[0];
2564 lin6 = (struct sockaddr_in6 *)&addrs[1];
2565 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2566 lin6->sin6_len != sizeof(struct sockaddr_in6))
2568 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2569 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2571 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2572 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2573 fin = (struct sockaddr_in *)&addrs[0];
2574 lin = (struct sockaddr_in *)&addrs[1];
2577 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2580 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2587 fin = (struct sockaddr_in *)&addrs[0];
2588 lin = (struct sockaddr_in *)&addrs[1];
2589 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2590 lin->sin_len != sizeof(struct sockaddr_in))
2597 INP_INFO_RLOCK(&V_tcbinfo);
2598 switch (addrs[0].ss_family) {
2601 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2602 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2603 INPLOOKUP_WLOCKPCB, NULL);
2608 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2609 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2614 if (inp->inp_flags & INP_TIMEWAIT) {
2616 * XXXRW: There currently exists a state where an
2617 * inpcb is present, but its timewait state has been
2618 * discarded. For now, don't allow dropping of this
2626 } else if (!(inp->inp_flags & INP_DROPPED) &&
2627 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2628 tp = intotcpcb(inp);
2629 tp = tcp_drop(tp, ECONNABORTED);
2636 INP_INFO_RUNLOCK(&V_tcbinfo);
2640 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2641 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2642 0, sysctl_drop, "", "Drop TCP connection");
2645 * Generate a standardized TCP log line for use throughout the
2646 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2647 * allow use in the interrupt context.
2649 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2650 * NB: The function may return NULL if memory allocation failed.
2652 * Due to header inclusion and ordering limitations the struct ip
2653 * and ip6_hdr pointers have to be passed as void pointers.
2656 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2660 /* Is logging enabled? */
2661 if (tcp_log_in_vain == 0)
2664 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2668 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2672 /* Is logging enabled? */
2673 if (tcp_log_debug == 0)
2676 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2680 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2687 const struct ip6_hdr *ip6;
2689 ip6 = (const struct ip6_hdr *)ip6hdr;
2691 ip = (struct ip *)ip4hdr;
2694 * The log line looks like this:
2695 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2697 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2698 sizeof(PRINT_TH_FLAGS) + 1 +
2700 2 * INET6_ADDRSTRLEN;
2702 2 * INET_ADDRSTRLEN;
2705 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2709 strcat(s, "TCP: [");
2712 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2713 inet_ntoa_r(inc->inc_faddr, sp);
2715 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2717 inet_ntoa_r(inc->inc_laddr, sp);
2719 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2722 ip6_sprintf(sp, &inc->inc6_faddr);
2724 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2726 ip6_sprintf(sp, &inc->inc6_laddr);
2728 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2729 } else if (ip6 && th) {
2730 ip6_sprintf(sp, &ip6->ip6_src);
2732 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2734 ip6_sprintf(sp, &ip6->ip6_dst);
2736 sprintf(sp, "]:%i", ntohs(th->th_dport));
2739 } else if (ip && th) {
2740 inet_ntoa_r(ip->ip_src, sp);
2742 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2744 inet_ntoa_r(ip->ip_dst, sp);
2746 sprintf(sp, "]:%i", ntohs(th->th_dport));
2754 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2755 if (*(s + size - 1) != '\0')
2756 panic("%s: string too long", __func__);
2761 * A subroutine which makes it easy to track TCP state changes with DTrace.
2762 * This function shouldn't be called for t_state initializations that don't
2763 * correspond to actual TCP state transitions.
2766 tcp_state_change(struct tcpcb *tp, int newstate)
2768 #if defined(KDTRACE_HOOKS)
2769 int pstate = tp->t_state;
2772 TCPSTATES_DEC(tp->t_state);
2773 TCPSTATES_INC(newstate);
2774 tp->t_state = newstate;
2775 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);