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 static struct inpcb *tcp_notify(struct inpcb *, int);
230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231 static void tcp_mtudisc(struct inpcb *, int);
232 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
233 void *ip4hdr, const void *ip6hdr);
236 static struct tcp_function_block tcp_def_funcblk = {
240 tcp_default_ctloutput,
251 int t_functions_inited = 0;
252 struct tcp_funchead t_functions;
253 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
256 init_tcp_functions(void)
258 if (t_functions_inited == 0) {
259 TAILQ_INIT(&t_functions);
260 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
261 t_functions_inited = 1;
265 static struct tcp_function_block *
266 find_tcp_functions_locked(struct tcp_function_set *fs)
268 struct tcp_function *f;
269 struct tcp_function_block *blk=NULL;
271 TAILQ_FOREACH(f, &t_functions, tf_next) {
272 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
280 static struct tcp_function_block *
281 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
283 struct tcp_function_block *rblk=NULL;
284 struct tcp_function *f;
286 TAILQ_FOREACH(f, &t_functions, tf_next) {
287 if (f->tf_fb == blk) {
298 struct tcp_function_block *
299 find_and_ref_tcp_functions(struct tcp_function_set *fs)
301 struct tcp_function_block *blk;
303 rw_rlock(&tcp_function_lock);
304 blk = find_tcp_functions_locked(fs);
306 refcount_acquire(&blk->tfb_refcnt);
307 rw_runlock(&tcp_function_lock);
311 struct tcp_function_block *
312 find_and_ref_tcp_fb(struct tcp_function_block *blk)
314 struct tcp_function_block *rblk;
316 rw_rlock(&tcp_function_lock);
317 rblk = find_tcp_fb_locked(blk, NULL);
319 refcount_acquire(&rblk->tfb_refcnt);
320 rw_runlock(&tcp_function_lock);
326 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
329 struct tcp_function_set fs;
330 struct tcp_function_block *blk;
332 memset(&fs, 0, sizeof(fs));
333 rw_rlock(&tcp_function_lock);
334 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
337 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
338 fs.pcbcnt = blk->tfb_refcnt;
340 rw_runlock(&tcp_function_lock);
341 error = sysctl_handle_string(oidp, fs.function_set_name,
342 sizeof(fs.function_set_name), req);
344 /* Check for error or no change */
345 if (error != 0 || req->newptr == NULL)
348 rw_wlock(&tcp_function_lock);
349 blk = find_tcp_functions_locked(&fs);
351 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
355 tcp_func_set_ptr = blk;
357 rw_wunlock(&tcp_function_lock);
361 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
362 CTLTYPE_STRING | CTLFLAG_RW,
363 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
364 "Set/get the default TCP functions");
367 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
369 int error, cnt, linesz;
370 struct tcp_function *f;
375 rw_rlock(&tcp_function_lock);
376 TAILQ_FOREACH(f, &t_functions, tf_next) {
379 rw_runlock(&tcp_function_lock);
381 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
382 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
387 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
392 rw_rlock(&tcp_function_lock);
393 TAILQ_FOREACH(f, &t_functions, tf_next) {
394 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
395 f->tf_fb->tfb_tcp_block_name,
396 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
397 f->tf_fb->tfb_refcnt);
398 if (linesz >= bufsz) {
406 rw_runlock(&tcp_function_lock);
408 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
409 free(buffer, M_TEMP);
413 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
414 CTLTYPE_STRING|CTLFLAG_RD,
415 NULL, 0, sysctl_net_inet_list_available, "A",
416 "list available TCP Function sets");
419 * Target size of TCP PCB hash tables. Must be a power of two.
421 * Note that this can be overridden by the kernel environment
422 * variable net.inet.tcp.tcbhashsize
425 #define TCBHASHSIZE 0
430 * Callouts should be moved into struct tcp directly. They are currently
431 * separate because the tcpcb structure is exported to userland for sysctl
432 * parsing purposes, which do not know about callouts.
441 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
442 #define V_tcpcb_zone VNET(tcpcb_zone)
444 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
445 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
447 static struct mtx isn_mtx;
449 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
450 #define ISN_LOCK() mtx_lock(&isn_mtx)
451 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
454 * TCP initialization.
457 tcp_zone_change(void *tag)
460 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
461 uma_zone_set_max(V_tcpcb_zone, maxsockets);
462 tcp_tw_zone_change();
466 tcp_inpcb_init(void *mem, int size, int flags)
468 struct inpcb *inp = mem;
470 INP_LOCK_INIT(inp, "inp", "tcpinp");
475 * Take a value and get the next power of 2 that doesn't overflow.
476 * Used to size the tcp_inpcb hash buckets.
479 maketcp_hashsize(int size)
485 * get the next power of 2 higher than maxsockets.
487 hashsize = 1 << fls(size);
488 /* catch overflow, and just go one power of 2 smaller */
489 if (hashsize < size) {
490 hashsize = 1 << (fls(size) - 1);
496 register_tcp_functions(struct tcp_function_block *blk, int wait)
498 struct tcp_function_block *lblk;
499 struct tcp_function *n;
500 struct tcp_function_set fs;
502 if (t_functions_inited == 0) {
503 init_tcp_functions();
505 if ((blk->tfb_tcp_output == NULL) ||
506 (blk->tfb_tcp_do_segment == NULL) ||
507 (blk->tfb_tcp_ctloutput == NULL) ||
508 (strlen(blk->tfb_tcp_block_name) == 0)) {
510 * These functions are required and you
515 if (blk->tfb_tcp_timer_stop_all ||
516 blk->tfb_tcp_timer_activate ||
517 blk->tfb_tcp_timer_active ||
518 blk->tfb_tcp_timer_stop) {
520 * If you define one timer function you
521 * must have them all.
523 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
524 (blk->tfb_tcp_timer_activate == NULL) ||
525 (blk->tfb_tcp_timer_active == NULL) ||
526 (blk->tfb_tcp_timer_stop == NULL)) {
530 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
535 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
536 rw_wlock(&tcp_function_lock);
537 lblk = find_tcp_functions_locked(&fs);
539 /* Duplicate name space not allowed */
540 rw_wunlock(&tcp_function_lock);
541 free(n, M_TCPFUNCTIONS);
544 refcount_init(&blk->tfb_refcnt, 0);
546 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
547 rw_wunlock(&tcp_function_lock);
552 deregister_tcp_functions(struct tcp_function_block *blk)
554 struct tcp_function_block *lblk;
555 struct tcp_function *f;
558 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
559 /* You can't un-register the default */
562 rw_wlock(&tcp_function_lock);
563 if (blk == tcp_func_set_ptr) {
564 /* You can't free the current default */
565 rw_wunlock(&tcp_function_lock);
568 if (blk->tfb_refcnt) {
569 /* Still tcb attached, mark it. */
570 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
571 rw_wunlock(&tcp_function_lock);
574 lblk = find_tcp_fb_locked(blk, &f);
577 TAILQ_REMOVE(&t_functions, f, tf_next);
579 free(f, M_TCPFUNCTIONS);
582 rw_wunlock(&tcp_function_lock);
589 const char *tcbhash_tuneable;
592 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
594 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
595 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
596 printf("%s: WARNING: unable to register helper hook\n", __func__);
597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
598 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
599 printf("%s: WARNING: unable to register helper hook\n", __func__);
600 hashsize = TCBHASHSIZE;
601 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
604 * Auto tune the hash size based on maxsockets.
605 * A perfect hash would have a 1:1 mapping
606 * (hashsize = maxsockets) however it's been
607 * suggested that O(2) average is better.
609 hashsize = maketcp_hashsize(maxsockets / 4);
611 * Our historical default is 512,
612 * do not autotune lower than this.
616 if (bootverbose && IS_DEFAULT_VNET(curvnet))
617 printf("%s: %s auto tuned to %d\n", __func__,
618 tcbhash_tuneable, hashsize);
621 * We require a hashsize to be a power of two.
622 * Previously if it was not a power of two we would just reset it
623 * back to 512, which could be a nasty surprise if you did not notice
625 * Instead what we do is clip it to the closest power of two lower
626 * than the specified hash value.
628 if (!powerof2(hashsize)) {
629 int oldhashsize = hashsize;
631 hashsize = maketcp_hashsize(hashsize);
632 /* prevent absurdly low value */
635 printf("%s: WARNING: TCB hash size not a power of 2, "
636 "clipped from %d to %d.\n", __func__, oldhashsize,
639 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
640 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
643 * These have to be type stable for the benefit of the timers.
645 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
646 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
647 uma_zone_set_max(V_tcpcb_zone, maxsockets);
648 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
654 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
655 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
656 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
658 /* Skip initialization of globals for non-default instances. */
659 if (!IS_DEFAULT_VNET(curvnet))
662 tcp_reass_global_init();
664 /* XXX virtualize those bellow? */
665 tcp_delacktime = TCPTV_DELACK;
666 tcp_keepinit = TCPTV_KEEP_INIT;
667 tcp_keepidle = TCPTV_KEEP_IDLE;
668 tcp_keepintvl = TCPTV_KEEPINTVL;
669 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
671 tcp_rexmit_min = TCPTV_MIN;
672 if (tcp_rexmit_min < 1)
674 tcp_persmin = TCPTV_PERSMIN;
675 tcp_persmax = TCPTV_PERSMAX;
676 tcp_rexmit_slop = TCPTV_CPU_VAR;
677 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
678 tcp_tcbhashsize = hashsize;
679 /* Setup the tcp function block list */
680 init_tcp_functions();
681 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
683 if (tcp_soreceive_stream) {
685 tcp_usrreqs.pru_soreceive = soreceive_stream;
688 tcp6_usrreqs.pru_soreceive = soreceive_stream;
693 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
695 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
697 if (max_protohdr < TCP_MINPROTOHDR)
698 max_protohdr = TCP_MINPROTOHDR;
699 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
701 #undef TCP_MINPROTOHDR
704 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
705 SHUTDOWN_PRI_DEFAULT);
706 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
707 EVENTHANDLER_PRI_ANY);
719 tcp_destroy(void *unused __unused)
724 * All our processes are gone, all our sockets should be cleaned
725 * up, which means, we should be past the tcp_discardcb() calls.
726 * Sleep to let all tcpcb timers really disappear and cleanup.
729 INP_LIST_RLOCK(&V_tcbinfo);
730 n = V_tcbinfo.ipi_count;
731 INP_LIST_RUNLOCK(&V_tcbinfo);
734 pause("tcpdes", hz / 10);
739 in_pcbinfo_destroy(&V_tcbinfo);
740 /* tcp_discardcb() clears the sack_holes up. */
741 uma_zdestroy(V_sack_hole_zone);
742 uma_zdestroy(V_tcpcb_zone);
746 * Cannot free the zone until all tcpcbs are released as we attach
747 * the allocations to them.
749 tcp_fastopen_destroy();
752 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
754 printf("%s: WARNING: unable to deregister helper hook "
755 "type=%d, id=%d: error %d returned\n", __func__,
756 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
758 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
760 printf("%s: WARNING: unable to deregister helper hook "
761 "type=%d, id=%d: error %d returned\n", __func__,
762 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
765 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
775 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
776 * tcp_template used to store this data in mbufs, but we now recopy it out
777 * of the tcpcb each time to conserve mbufs.
780 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
782 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
784 INP_WLOCK_ASSERT(inp);
787 if ((inp->inp_vflag & INP_IPV6) != 0) {
790 ip6 = (struct ip6_hdr *)ip_ptr;
791 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
792 (inp->inp_flow & IPV6_FLOWINFO_MASK);
793 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
794 (IPV6_VERSION & IPV6_VERSION_MASK);
795 ip6->ip6_nxt = IPPROTO_TCP;
796 ip6->ip6_plen = htons(sizeof(struct tcphdr));
797 ip6->ip6_src = inp->in6p_laddr;
798 ip6->ip6_dst = inp->in6p_faddr;
801 #if defined(INET6) && defined(INET)
808 ip = (struct ip *)ip_ptr;
809 ip->ip_v = IPVERSION;
811 ip->ip_tos = inp->inp_ip_tos;
815 ip->ip_ttl = inp->inp_ip_ttl;
817 ip->ip_p = IPPROTO_TCP;
818 ip->ip_src = inp->inp_laddr;
819 ip->ip_dst = inp->inp_faddr;
822 th->th_sport = inp->inp_lport;
823 th->th_dport = inp->inp_fport;
831 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
835 * Create template to be used to send tcp packets on a connection.
836 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
837 * use for this function is in keepalives, which use tcp_respond.
840 tcpip_maketemplate(struct inpcb *inp)
844 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
847 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
852 * Send a single message to the TCP at address specified by
853 * the given TCP/IP header. If m == NULL, then we make a copy
854 * of the tcpiphdr at th and send directly to the addressed host.
855 * This is used to force keep alive messages out using the TCP
856 * template for a connection. If flags are given then we send
857 * a message back to the TCP which originated the segment th,
858 * and discard the mbuf containing it and any other attached mbufs.
860 * In any case the ack and sequence number of the transmitted
861 * segment are as specified by the parameters.
863 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
866 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
867 tcp_seq ack, tcp_seq seq, int flags)
879 int optlen, tlen, win;
882 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
885 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
892 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
893 INP_WLOCK_ASSERT(inp);
900 if (!(flags & TH_RST)) {
901 win = sbspace(&inp->inp_socket->so_rcv);
902 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
903 win = (long)TCP_MAXWIN << tp->rcv_scale;
905 if ((tp->t_flags & TF_NOOPT) == 0)
909 m = m_gethdr(M_NOWAIT, MT_DATA);
912 m->m_data += max_linkhdr;
915 bcopy((caddr_t)ip6, mtod(m, caddr_t),
916 sizeof(struct ip6_hdr));
917 ip6 = mtod(m, struct ip6_hdr *);
918 nth = (struct tcphdr *)(ip6 + 1);
922 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
923 ip = mtod(m, struct ip *);
924 nth = (struct tcphdr *)(ip + 1);
926 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
928 } else if (!M_WRITABLE(m)) {
931 /* Can't reuse 'm', allocate a new mbuf. */
932 n = m_gethdr(M_NOWAIT, MT_DATA);
938 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
944 n->m_data += max_linkhdr;
945 /* m_len is set later */
946 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
949 bcopy((caddr_t)ip6, mtod(n, caddr_t),
950 sizeof(struct ip6_hdr));
951 ip6 = mtod(n, struct ip6_hdr *);
952 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
953 nth = (struct tcphdr *)(ip6 + 1);
957 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
958 ip = mtod(n, struct ip *);
959 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
960 nth = (struct tcphdr *)(ip + 1);
962 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
963 xchg(nth->th_dport, nth->th_sport, uint16_t);
970 * XXX MRT We inherit the FIB, which is lucky.
974 m->m_data = (caddr_t)ipgen;
975 /* m_len is set later */
978 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
979 nth = (struct tcphdr *)(ip6 + 1);
983 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
984 nth = (struct tcphdr *)(ip + 1);
988 * this is usually a case when an extension header
989 * exists between the IPv6 header and the
992 nth->th_sport = th->th_sport;
993 nth->th_dport = th->th_dport;
995 xchg(nth->th_dport, nth->th_sport, uint16_t);
1001 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1003 #if defined(INET) && defined(INET6)
1007 tlen = sizeof (struct tcpiphdr);
1011 KASSERT(M_TRAILINGSPACE(m) >= tlen,
1012 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1013 m, tlen, (long)M_TRAILINGSPACE(m)));
1018 /* Make sure we have room. */
1019 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1020 m->m_next = m_get(M_NOWAIT, MT_DATA);
1022 optp = mtod(m->m_next, u_char *);
1027 optp = (u_char *) (nth + 1);
1033 if (tp->t_flags & TF_RCVD_TSTMP) {
1034 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1035 to.to_tsecr = tp->ts_recent;
1036 to.to_flags |= TOF_TS;
1038 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1039 /* TCP-MD5 (RFC2385). */
1040 if (tp->t_flags & TF_SIGNATURE)
1041 to.to_flags |= TOF_SIGNATURE;
1043 /* Add the options. */
1044 tlen += optlen = tcp_addoptions(&to, optp);
1046 /* Update m_len in the correct mbuf. */
1047 optm->m_len += optlen;
1053 ip6->ip6_vfc = IPV6_VERSION;
1054 ip6->ip6_nxt = IPPROTO_TCP;
1055 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1058 #if defined(INET) && defined(INET6)
1063 ip->ip_len = htons(tlen);
1064 ip->ip_ttl = V_ip_defttl;
1065 if (V_path_mtu_discovery)
1066 ip->ip_off |= htons(IP_DF);
1069 m->m_pkthdr.len = tlen;
1070 m->m_pkthdr.rcvif = NULL;
1074 * Packet is associated with a socket, so allow the
1075 * label of the response to reflect the socket label.
1077 INP_WLOCK_ASSERT(inp);
1078 mac_inpcb_create_mbuf(inp, m);
1081 * Packet is not associated with a socket, so possibly
1082 * update the label in place.
1084 mac_netinet_tcp_reply(m);
1087 nth->th_seq = htonl(seq);
1088 nth->th_ack = htonl(ack);
1090 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1091 nth->th_flags = flags;
1093 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1095 nth->th_win = htons((u_short)win);
1098 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1099 if (to.to_flags & TOF_SIGNATURE) {
1100 if (!TCPMD5_ENABLED() ||
1101 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1108 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1111 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1112 nth->th_sum = in6_cksum_pseudo(ip6,
1113 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1114 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1118 #if defined(INET6) && defined(INET)
1123 m->m_pkthdr.csum_flags = CSUM_TCP;
1124 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1125 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1129 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1130 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1132 TCP_PROBE3(debug__output, tp, th, m);
1134 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1136 TCP_PROBE5(send, NULL, tp, m, tp, nth);
1139 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1141 #if defined(INET) && defined(INET6)
1145 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1150 * Create a new TCP control block, making an
1151 * empty reassembly queue and hooking it to the argument
1152 * protocol control block. The `inp' parameter must have
1153 * come from the zone allocator set up in tcp_init().
1156 tcp_newtcpcb(struct inpcb *inp)
1158 struct tcpcb_mem *tm;
1161 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1164 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1169 /* Initialise cc_var struct for this tcpcb. */
1171 tp->ccv->type = IPPROTO_TCP;
1172 tp->ccv->ccvc.tcp = tp;
1173 rw_rlock(&tcp_function_lock);
1174 tp->t_fb = tcp_func_set_ptr;
1175 refcount_acquire(&tp->t_fb->tfb_refcnt);
1176 rw_runlock(&tcp_function_lock);
1177 if (tp->t_fb->tfb_tcp_fb_init) {
1178 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1181 * Use the current system default CC algorithm.
1184 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1185 CC_ALGO(tp) = CC_DEFAULT();
1188 if (CC_ALGO(tp)->cb_init != NULL)
1189 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1190 if (tp->t_fb->tfb_tcp_fb_fini)
1191 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1192 refcount_release(&tp->t_fb->tfb_refcnt);
1193 uma_zfree(V_tcpcb_zone, tm);
1198 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1199 if (tp->t_fb->tfb_tcp_fb_fini)
1200 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1201 refcount_release(&tp->t_fb->tfb_refcnt);
1202 uma_zfree(V_tcpcb_zone, tm);
1207 tp->t_vnet = inp->inp_vnet;
1209 tp->t_timers = &tm->tt;
1210 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1213 isipv6 ? V_tcp_v6mssdflt :
1217 /* Set up our timeouts. */
1218 callout_init(&tp->t_timers->tt_rexmt, 1);
1219 callout_init(&tp->t_timers->tt_persist, 1);
1220 callout_init(&tp->t_timers->tt_keep, 1);
1221 callout_init(&tp->t_timers->tt_2msl, 1);
1222 callout_init(&tp->t_timers->tt_delack, 1);
1224 if (V_tcp_do_rfc1323)
1225 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1227 tp->t_flags |= TF_SACK_PERMIT;
1228 TAILQ_INIT(&tp->snd_holes);
1230 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1233 in_pcbref(inp); /* Reference for tcpcb */
1237 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1238 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1239 * reasonable initial retransmit time.
1241 tp->t_srtt = TCPTV_SRTTBASE;
1242 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1243 tp->t_rttmin = tcp_rexmit_min;
1244 tp->t_rxtcur = TCPTV_RTOBASE;
1245 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1246 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1247 tp->t_rcvtime = ticks;
1249 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1250 * because the socket may be bound to an IPv6 wildcard address,
1251 * which may match an IPv4-mapped IPv6 address.
1253 inp->inp_ip_ttl = V_ip_defttl;
1257 * Init the TCP PCAP queues.
1259 tcp_pcap_tcpcb_init(tp);
1261 return (tp); /* XXX */
1265 * Switch the congestion control algorithm back to NewReno for any active
1266 * control blocks using an algorithm which is about to go away.
1267 * This ensures the CC framework can allow the unload to proceed without leaving
1268 * any dangling pointers which would trigger a panic.
1269 * Returning non-zero would inform the CC framework that something went wrong
1270 * and it would be unsafe to allow the unload to proceed. However, there is no
1271 * way for this to occur with this implementation so we always return zero.
1274 tcp_ccalgounload(struct cc_algo *unload_algo)
1276 struct cc_algo *tmpalgo;
1279 VNET_ITERATOR_DECL(vnet_iter);
1282 * Check all active control blocks across all network stacks and change
1283 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1284 * requires cleanup code to be run, call it.
1287 VNET_FOREACH(vnet_iter) {
1288 CURVNET_SET(vnet_iter);
1289 INP_INFO_WLOCK(&V_tcbinfo);
1291 * New connections already part way through being initialised
1292 * with the CC algo we're removing will not race with this code
1293 * because the INP_INFO_WLOCK is held during initialisation. We
1294 * therefore don't enter the loop below until the connection
1295 * list has stabilised.
1297 LIST_FOREACH(inp, &V_tcb, inp_list) {
1299 /* Important to skip tcptw structs. */
1300 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1301 (tp = intotcpcb(inp)) != NULL) {
1303 * By holding INP_WLOCK here, we are assured
1304 * that the connection is not currently
1305 * executing inside the CC module's functions
1306 * i.e. it is safe to make the switch back to
1309 if (CC_ALGO(tp) == unload_algo) {
1310 tmpalgo = CC_ALGO(tp);
1311 /* NewReno does not require any init. */
1312 CC_ALGO(tp) = &newreno_cc_algo;
1313 if (tmpalgo->cb_destroy != NULL)
1314 tmpalgo->cb_destroy(tp->ccv);
1319 INP_INFO_WUNLOCK(&V_tcbinfo);
1322 VNET_LIST_RUNLOCK();
1328 * Drop a TCP connection, reporting
1329 * the specified error. If connection is synchronized,
1330 * then send a RST to peer.
1333 tcp_drop(struct tcpcb *tp, int errno)
1335 struct socket *so = tp->t_inpcb->inp_socket;
1337 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1338 INP_WLOCK_ASSERT(tp->t_inpcb);
1340 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1341 tcp_state_change(tp, TCPS_CLOSED);
1342 (void) tp->t_fb->tfb_tcp_output(tp);
1343 TCPSTAT_INC(tcps_drops);
1345 TCPSTAT_INC(tcps_conndrops);
1346 if (errno == ETIMEDOUT && tp->t_softerror)
1347 errno = tp->t_softerror;
1348 so->so_error = errno;
1349 return (tcp_close(tp));
1353 tcp_discardcb(struct tcpcb *tp)
1355 struct inpcb *inp = tp->t_inpcb;
1356 struct socket *so = inp->inp_socket;
1358 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1362 INP_WLOCK_ASSERT(inp);
1365 * Make sure that all of our timers are stopped before we delete the
1368 * If stopping a timer fails, we schedule a discard function in same
1369 * callout, and the last discard function called will take care of
1370 * deleting the tcpcb.
1372 tp->t_timers->tt_draincnt = 0;
1373 tcp_timer_stop(tp, TT_REXMT);
1374 tcp_timer_stop(tp, TT_PERSIST);
1375 tcp_timer_stop(tp, TT_KEEP);
1376 tcp_timer_stop(tp, TT_2MSL);
1377 tcp_timer_stop(tp, TT_DELACK);
1378 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1380 * Call the stop-all function of the methods,
1381 * this function should call the tcp_timer_stop()
1382 * method with each of the function specific timeouts.
1383 * That stop will be called via the tfb_tcp_timer_stop()
1384 * which should use the async drain function of the
1385 * callout system (see tcp_var.h).
1387 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1391 * If we got enough samples through the srtt filter,
1392 * save the rtt and rttvar in the routing entry.
1393 * 'Enough' is arbitrarily defined as 4 rtt samples.
1394 * 4 samples is enough for the srtt filter to converge
1395 * to within enough % of the correct value; fewer samples
1396 * and we could save a bogus rtt. The danger is not high
1397 * as tcp quickly recovers from everything.
1398 * XXX: Works very well but needs some more statistics!
1400 if (tp->t_rttupdated >= 4) {
1401 struct hc_metrics_lite metrics;
1404 bzero(&metrics, sizeof(metrics));
1406 * Update the ssthresh always when the conditions below
1407 * are satisfied. This gives us better new start value
1408 * for the congestion avoidance for new connections.
1409 * ssthresh is only set if packet loss occurred on a session.
1411 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1412 * being torn down. Ideally this code would not use 'so'.
1414 ssthresh = tp->snd_ssthresh;
1415 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1417 * convert the limit from user data bytes to
1418 * packets then to packet data bytes.
1420 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1423 ssthresh *= (u_long)(tp->t_maxseg +
1425 (isipv6 ? sizeof (struct ip6_hdr) +
1426 sizeof (struct tcphdr) :
1428 sizeof (struct tcpiphdr)
1435 metrics.rmx_ssthresh = ssthresh;
1437 metrics.rmx_rtt = tp->t_srtt;
1438 metrics.rmx_rttvar = tp->t_rttvar;
1439 metrics.rmx_cwnd = tp->snd_cwnd;
1440 metrics.rmx_sendpipe = 0;
1441 metrics.rmx_recvpipe = 0;
1443 tcp_hc_update(&inp->inp_inc, &metrics);
1446 /* free the reassembly queue, if any */
1447 tcp_reass_flush(tp);
1450 /* Disconnect offload device, if any. */
1451 if (tp->t_flags & TF_TOE)
1452 tcp_offload_detach(tp);
1455 tcp_free_sackholes(tp);
1458 /* Free the TCP PCAP queues. */
1459 tcp_pcap_drain(&(tp->t_inpkts));
1460 tcp_pcap_drain(&(tp->t_outpkts));
1463 /* Allow the CC algorithm to clean up after itself. */
1464 if (CC_ALGO(tp)->cb_destroy != NULL)
1465 CC_ALGO(tp)->cb_destroy(tp->ccv);
1467 khelp_destroy_osd(tp->osd);
1470 inp->inp_ppcb = NULL;
1471 if (tp->t_timers->tt_draincnt == 0) {
1472 /* We own the last reference on tcpcb, let's free it. */
1473 TCPSTATES_DEC(tp->t_state);
1474 if (tp->t_fb->tfb_tcp_fb_fini)
1475 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1476 refcount_release(&tp->t_fb->tfb_refcnt);
1478 uma_zfree(V_tcpcb_zone, tp);
1479 released = in_pcbrele_wlocked(inp);
1480 KASSERT(!released, ("%s: inp %p should not have been released "
1481 "here", __func__, inp));
1486 tcp_timer_discard(void *ptp)
1491 tp = (struct tcpcb *)ptp;
1492 CURVNET_SET(tp->t_vnet);
1493 INP_INFO_RLOCK(&V_tcbinfo);
1495 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1498 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1499 ("%s: tcpcb has to be stopped here", __func__));
1500 tp->t_timers->tt_draincnt--;
1501 if (tp->t_timers->tt_draincnt == 0) {
1502 /* We own the last reference on this tcpcb, let's free it. */
1503 TCPSTATES_DEC(tp->t_state);
1504 if (tp->t_fb->tfb_tcp_fb_fini)
1505 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1506 refcount_release(&tp->t_fb->tfb_refcnt);
1508 uma_zfree(V_tcpcb_zone, tp);
1509 if (in_pcbrele_wlocked(inp)) {
1510 INP_INFO_RUNLOCK(&V_tcbinfo);
1516 INP_INFO_RUNLOCK(&V_tcbinfo);
1521 * Attempt to close a TCP control block, marking it as dropped, and freeing
1522 * the socket if we hold the only reference.
1525 tcp_close(struct tcpcb *tp)
1527 struct inpcb *inp = tp->t_inpcb;
1530 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1531 INP_WLOCK_ASSERT(inp);
1534 if (tp->t_state == TCPS_LISTEN)
1535 tcp_offload_listen_stop(tp);
1539 * This releases the TFO pending counter resource for TFO listen
1540 * sockets as well as passively-created TFO sockets that transition
1541 * from SYN_RECEIVED to CLOSED.
1543 if (tp->t_tfo_pending) {
1544 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1545 tp->t_tfo_pending = NULL;
1549 TCPSTAT_INC(tcps_closed);
1550 if (tp->t_state != TCPS_CLOSED)
1551 tcp_state_change(tp, TCPS_CLOSED);
1552 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1553 so = inp->inp_socket;
1554 soisdisconnected(so);
1555 if (inp->inp_flags & INP_SOCKREF) {
1556 KASSERT(so->so_state & SS_PROTOREF,
1557 ("tcp_close: !SS_PROTOREF"));
1558 inp->inp_flags &= ~INP_SOCKREF;
1562 so->so_state &= ~SS_PROTOREF;
1572 VNET_ITERATOR_DECL(vnet_iter);
1577 VNET_LIST_RLOCK_NOSLEEP();
1578 VNET_FOREACH(vnet_iter) {
1579 CURVNET_SET(vnet_iter);
1584 * Walk the tcpbs, if existing, and flush the reassembly queue,
1585 * if there is one...
1586 * XXX: The "Net/3" implementation doesn't imply that the TCP
1587 * reassembly queue should be flushed, but in a situation
1588 * where we're really low on mbufs, this is potentially
1591 INP_INFO_WLOCK(&V_tcbinfo);
1592 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1593 if (inpb->inp_flags & INP_TIMEWAIT)
1596 if ((tcpb = intotcpcb(inpb)) != NULL) {
1597 tcp_reass_flush(tcpb);
1598 tcp_clean_sackreport(tcpb);
1600 if (tcp_pcap_aggressive_free) {
1601 /* Free the TCP PCAP queues. */
1602 tcp_pcap_drain(&(tcpb->t_inpkts));
1603 tcp_pcap_drain(&(tcpb->t_outpkts));
1609 INP_INFO_WUNLOCK(&V_tcbinfo);
1612 VNET_LIST_RUNLOCK_NOSLEEP();
1616 * Notify a tcp user of an asynchronous error;
1617 * store error as soft error, but wake up user
1618 * (for now, won't do anything until can select for soft error).
1620 * Do not wake up user since there currently is no mechanism for
1621 * reporting soft errors (yet - a kqueue filter may be added).
1623 static struct inpcb *
1624 tcp_notify(struct inpcb *inp, int error)
1628 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1629 INP_WLOCK_ASSERT(inp);
1631 if ((inp->inp_flags & INP_TIMEWAIT) ||
1632 (inp->inp_flags & INP_DROPPED))
1635 tp = intotcpcb(inp);
1636 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1639 * Ignore some errors if we are hooked up.
1640 * If connection hasn't completed, has retransmitted several times,
1641 * and receives a second error, give up now. This is better
1642 * than waiting a long time to establish a connection that
1643 * can never complete.
1645 if (tp->t_state == TCPS_ESTABLISHED &&
1646 (error == EHOSTUNREACH || error == ENETUNREACH ||
1647 error == EHOSTDOWN)) {
1648 if (inp->inp_route.ro_rt) {
1649 RTFREE(inp->inp_route.ro_rt);
1650 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1653 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1655 tp = tcp_drop(tp, error);
1661 tp->t_softerror = error;
1665 wakeup( &so->so_timeo);
1672 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1674 int error, i, m, n, pcb_count;
1675 struct inpcb *inp, **inp_list;
1680 * The process of preparing the TCB list is too time-consuming and
1681 * resource-intensive to repeat twice on every request.
1683 if (req->oldptr == NULL) {
1684 n = V_tcbinfo.ipi_count +
1685 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1686 n += imax(n / 8, 10);
1687 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1691 if (req->newptr != NULL)
1695 * OK, now we're committed to doing something.
1697 INP_LIST_RLOCK(&V_tcbinfo);
1698 gencnt = V_tcbinfo.ipi_gencnt;
1699 n = V_tcbinfo.ipi_count;
1700 INP_LIST_RUNLOCK(&V_tcbinfo);
1702 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1704 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1705 + (n + m) * sizeof(struct xtcpcb));
1709 xig.xig_len = sizeof xig;
1710 xig.xig_count = n + m;
1711 xig.xig_gen = gencnt;
1712 xig.xig_sogen = so_gencnt;
1713 error = SYSCTL_OUT(req, &xig, sizeof xig);
1717 error = syncache_pcblist(req, m, &pcb_count);
1721 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1723 INP_INFO_WLOCK(&V_tcbinfo);
1724 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1725 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1727 if (inp->inp_gencnt <= gencnt) {
1729 * XXX: This use of cr_cansee(), introduced with
1730 * TCP state changes, is not quite right, but for
1731 * now, better than nothing.
1733 if (inp->inp_flags & INP_TIMEWAIT) {
1734 if (intotw(inp) != NULL)
1735 error = cr_cansee(req->td->td_ucred,
1736 intotw(inp)->tw_cred);
1738 error = EINVAL; /* Skip this inp. */
1740 error = cr_canseeinpcb(req->td->td_ucred, inp);
1743 inp_list[i++] = inp;
1748 INP_INFO_WUNLOCK(&V_tcbinfo);
1752 for (i = 0; i < n; i++) {
1755 if (inp->inp_gencnt <= gencnt) {
1759 bzero(&xt, sizeof(xt));
1760 xt.xt_len = sizeof xt;
1761 /* XXX should avoid extra copy */
1762 bcopy(inp, &xt.xt_inp, sizeof *inp);
1763 inp_ppcb = inp->inp_ppcb;
1764 if (inp_ppcb == NULL)
1765 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1766 else if (inp->inp_flags & INP_TIMEWAIT) {
1767 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1768 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1770 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1771 if (xt.xt_tp.t_timers)
1772 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1774 if (inp->inp_socket != NULL)
1775 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1777 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1778 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1780 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1782 error = SYSCTL_OUT(req, &xt, sizeof xt);
1786 INP_INFO_RLOCK(&V_tcbinfo);
1787 for (i = 0; i < n; i++) {
1790 if (!in_pcbrele_rlocked(inp))
1793 INP_INFO_RUNLOCK(&V_tcbinfo);
1797 * Give the user an updated idea of our state.
1798 * If the generation differs from what we told
1799 * her before, she knows that something happened
1800 * while we were processing this request, and it
1801 * might be necessary to retry.
1803 INP_LIST_RLOCK(&V_tcbinfo);
1804 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1805 xig.xig_sogen = so_gencnt;
1806 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1807 INP_LIST_RUNLOCK(&V_tcbinfo);
1808 error = SYSCTL_OUT(req, &xig, sizeof xig);
1810 free(inp_list, M_TEMP);
1814 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1815 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1816 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1820 tcp_getcred(SYSCTL_HANDLER_ARGS)
1823 struct sockaddr_in addrs[2];
1827 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1830 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1833 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1834 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1836 if (inp->inp_socket == NULL)
1839 error = cr_canseeinpcb(req->td->td_ucred, inp);
1841 cru2x(inp->inp_cred, &xuc);
1846 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1850 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1851 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1852 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1857 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1860 struct sockaddr_in6 addrs[2];
1867 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1870 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1873 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1874 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1877 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1879 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1888 inp = in_pcblookup(&V_tcbinfo,
1889 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1891 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1892 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1895 inp = in6_pcblookup(&V_tcbinfo,
1896 &addrs[1].sin6_addr, addrs[1].sin6_port,
1897 &addrs[0].sin6_addr, addrs[0].sin6_port,
1898 INPLOOKUP_RLOCKPCB, NULL);
1900 if (inp->inp_socket == NULL)
1903 error = cr_canseeinpcb(req->td->td_ucred, inp);
1905 cru2x(inp->inp_cred, &xuc);
1910 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1914 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1915 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1916 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1922 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1924 struct ip *ip = vip;
1926 struct in_addr faddr;
1929 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1931 struct in_conninfo inc;
1932 tcp_seq icmp_tcp_seq;
1935 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1936 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1939 if (cmd == PRC_MSGSIZE)
1940 notify = tcp_mtudisc_notify;
1941 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1942 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1943 cmd == PRC_TIMXCEED_INTRANS) && ip)
1944 notify = tcp_drop_syn_sent;
1947 * Hostdead is ugly because it goes linearly through all PCBs.
1948 * XXX: We never get this from ICMP, otherwise it makes an
1949 * excellent DoS attack on machines with many connections.
1951 else if (cmd == PRC_HOSTDEAD)
1953 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1957 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1961 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1962 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1963 INP_INFO_RLOCK(&V_tcbinfo);
1964 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1965 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1966 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1967 /* signal EHOSTDOWN, as it flushes the cached route */
1968 inp = (*notify)(inp, EHOSTDOWN);
1971 } else if (inp != NULL) {
1972 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1973 !(inp->inp_flags & INP_DROPPED) &&
1974 !(inp->inp_socket == NULL)) {
1975 icmp_tcp_seq = ntohl(th->th_seq);
1976 tp = intotcpcb(inp);
1977 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1978 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1979 if (cmd == PRC_MSGSIZE) {
1982 * If we got a needfrag set the MTU
1983 * in the route to the suggested new
1984 * value (if given) and then notify.
1986 mtu = ntohs(icp->icmp_nextmtu);
1988 * If no alternative MTU was
1989 * proposed, try the next smaller
1994 ntohs(ip->ip_len), 1);
1995 if (mtu < V_tcp_minmss +
1996 sizeof(struct tcpiphdr))
1997 mtu = V_tcp_minmss +
1998 sizeof(struct tcpiphdr);
2000 * Only process the offered MTU if it
2001 * is smaller than the current one.
2003 if (mtu < tp->t_maxseg +
2004 sizeof(struct tcpiphdr)) {
2005 bzero(&inc, sizeof(inc));
2006 inc.inc_faddr = faddr;
2008 inp->inp_inc.inc_fibnum;
2009 tcp_hc_updatemtu(&inc, mtu);
2010 tcp_mtudisc(inp, mtu);
2013 inp = (*notify)(inp,
2014 inetctlerrmap[cmd]);
2020 bzero(&inc, sizeof(inc));
2021 inc.inc_fport = th->th_dport;
2022 inc.inc_lport = th->th_sport;
2023 inc.inc_faddr = faddr;
2024 inc.inc_laddr = ip->ip_src;
2025 syncache_unreach(&inc, th);
2027 INP_INFO_RUNLOCK(&V_tcbinfo);
2033 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2035 struct in6_addr *dst;
2037 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2038 struct ip6_hdr *ip6;
2042 struct icmp6_hdr *icmp6;
2043 struct ip6ctlparam *ip6cp = NULL;
2044 const struct sockaddr_in6 *sa6_src = NULL;
2045 struct in_conninfo inc;
2046 tcp_seq icmp_tcp_seq;
2051 if (sa->sa_family != AF_INET6 ||
2052 sa->sa_len != sizeof(struct sockaddr_in6))
2055 /* if the parameter is from icmp6, decode it. */
2057 ip6cp = (struct ip6ctlparam *)d;
2058 icmp6 = ip6cp->ip6c_icmp6;
2060 ip6 = ip6cp->ip6c_ip6;
2061 off = ip6cp->ip6c_off;
2062 sa6_src = ip6cp->ip6c_src;
2063 dst = ip6cp->ip6c_finaldst;
2067 off = 0; /* fool gcc */
2072 if (cmd == PRC_MSGSIZE)
2073 notify = tcp_mtudisc_notify;
2074 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2075 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2076 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2077 notify = tcp_drop_syn_sent;
2080 * Hostdead is ugly because it goes linearly through all PCBs.
2081 * XXX: We never get this from ICMP, otherwise it makes an
2082 * excellent DoS attack on machines with many connections.
2084 else if (cmd == PRC_HOSTDEAD)
2086 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2090 in6_pcbnotify(&V_tcbinfo, sa, 0,
2091 (const struct sockaddr *)sa6_src,
2092 0, cmd, NULL, notify);
2096 /* Check if we can safely get the ports from the tcp hdr */
2099 (int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2103 th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2104 INP_INFO_RLOCK(&V_tcbinfo);
2105 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2106 &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2107 if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2108 /* signal EHOSTDOWN, as it flushes the cached route */
2109 inp = (*notify)(inp, EHOSTDOWN);
2112 } else if (inp != NULL) {
2113 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2114 !(inp->inp_flags & INP_DROPPED) &&
2115 !(inp->inp_socket == NULL)) {
2116 icmp_tcp_seq = ntohl(th->th_seq);
2117 tp = intotcpcb(inp);
2118 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2119 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2120 if (cmd == PRC_MSGSIZE) {
2123 * If we got a needfrag set the MTU
2124 * in the route to the suggested new
2125 * value (if given) and then notify.
2127 mtu = ntohl(icmp6->icmp6_mtu);
2129 * If no alternative MTU was
2130 * proposed, or the proposed
2131 * MTU was too small, set to
2134 if (mtu < IPV6_MMTU)
2135 mtu = IPV6_MMTU - 8;
2138 bzero(&inc, sizeof(inc));
2139 inc.inc_fibnum = M_GETFIB(m);
2140 inc.inc_flags |= INC_ISIPV6;
2141 inc.inc6_faddr = *dst;
2142 if (in6_setscope(&inc.inc6_faddr,
2143 m->m_pkthdr.rcvif, NULL))
2147 * Only process the offered MTU if it
2148 * is smaller than the current one.
2150 if (mtu < tp->t_maxseg +
2151 (sizeof (*th) + sizeof (*ip6))) {
2152 tcp_hc_updatemtu(&inc, mtu);
2153 tcp_mtudisc(inp, mtu);
2154 ICMP6STAT_INC(icp6s_pmtuchg);
2157 inp = (*notify)(inp,
2158 inet6ctlerrmap[cmd]);
2165 bzero(&inc, sizeof(inc));
2166 inc.inc_fibnum = M_GETFIB(m);
2167 inc.inc_flags |= INC_ISIPV6;
2168 inc.inc_fport = th->th_dport;
2169 inc.inc_lport = th->th_sport;
2170 inc.inc6_faddr = *dst;
2171 inc.inc6_laddr = ip6->ip6_src;
2172 syncache_unreach(&inc, th);
2174 INP_INFO_RUNLOCK(&V_tcbinfo);
2180 * Following is where TCP initial sequence number generation occurs.
2182 * There are two places where we must use initial sequence numbers:
2183 * 1. In SYN-ACK packets.
2184 * 2. In SYN packets.
2186 * All ISNs for SYN-ACK packets are generated by the syncache. See
2187 * tcp_syncache.c for details.
2189 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2190 * depends on this property. In addition, these ISNs should be
2191 * unguessable so as to prevent connection hijacking. To satisfy
2192 * the requirements of this situation, the algorithm outlined in
2193 * RFC 1948 is used, with only small modifications.
2195 * Implementation details:
2197 * Time is based off the system timer, and is corrected so that it
2198 * increases by one megabyte per second. This allows for proper
2199 * recycling on high speed LANs while still leaving over an hour
2202 * As reading the *exact* system time is too expensive to be done
2203 * whenever setting up a TCP connection, we increment the time
2204 * offset in two ways. First, a small random positive increment
2205 * is added to isn_offset for each connection that is set up.
2206 * Second, the function tcp_isn_tick fires once per clock tick
2207 * and increments isn_offset as necessary so that sequence numbers
2208 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2209 * random positive increments serve only to ensure that the same
2210 * exact sequence number is never sent out twice (as could otherwise
2211 * happen when a port is recycled in less than the system tick
2214 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2215 * between seeding of isn_secret. This is normally set to zero,
2216 * as reseeding should not be necessary.
2218 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2219 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2220 * general, this means holding an exclusive (write) lock.
2223 #define ISN_BYTES_PER_SECOND 1048576
2224 #define ISN_STATIC_INCREMENT 4096
2225 #define ISN_RANDOM_INCREMENT (4096 - 1)
2227 static VNET_DEFINE(u_char, isn_secret[32]);
2228 static VNET_DEFINE(int, isn_last);
2229 static VNET_DEFINE(int, isn_last_reseed);
2230 static VNET_DEFINE(u_int32_t, isn_offset);
2231 static VNET_DEFINE(u_int32_t, isn_offset_old);
2233 #define V_isn_secret VNET(isn_secret)
2234 #define V_isn_last VNET(isn_last)
2235 #define V_isn_last_reseed VNET(isn_last_reseed)
2236 #define V_isn_offset VNET(isn_offset)
2237 #define V_isn_offset_old VNET(isn_offset_old)
2240 tcp_new_isn(struct tcpcb *tp)
2243 u_int32_t md5_buffer[4];
2245 u_int32_t projected_offset;
2247 INP_WLOCK_ASSERT(tp->t_inpcb);
2250 /* Seed if this is the first use, reseed if requested. */
2251 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2252 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2254 read_random(&V_isn_secret, sizeof(V_isn_secret));
2255 V_isn_last_reseed = ticks;
2258 /* Compute the md5 hash and return the ISN. */
2260 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2261 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2263 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2264 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2265 sizeof(struct in6_addr));
2266 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2267 sizeof(struct in6_addr));
2271 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2272 sizeof(struct in_addr));
2273 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2274 sizeof(struct in_addr));
2276 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2277 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2278 new_isn = (tcp_seq) md5_buffer[0];
2279 V_isn_offset += ISN_STATIC_INCREMENT +
2280 (arc4random() & ISN_RANDOM_INCREMENT);
2281 if (ticks != V_isn_last) {
2282 projected_offset = V_isn_offset_old +
2283 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2284 if (SEQ_GT(projected_offset, V_isn_offset))
2285 V_isn_offset = projected_offset;
2286 V_isn_offset_old = V_isn_offset;
2289 new_isn += V_isn_offset;
2295 * When a specific ICMP unreachable message is received and the
2296 * connection state is SYN-SENT, drop the connection. This behavior
2297 * is controlled by the icmp_may_rst sysctl.
2300 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2304 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2305 INP_WLOCK_ASSERT(inp);
2307 if ((inp->inp_flags & INP_TIMEWAIT) ||
2308 (inp->inp_flags & INP_DROPPED))
2311 tp = intotcpcb(inp);
2312 if (tp->t_state != TCPS_SYN_SENT)
2315 tp = tcp_drop(tp, errno);
2323 * When `need fragmentation' ICMP is received, update our idea of the MSS
2324 * based on the new value. Also nudge TCP to send something, since we
2325 * know the packet we just sent was dropped.
2326 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2328 static struct inpcb *
2329 tcp_mtudisc_notify(struct inpcb *inp, int error)
2332 tcp_mtudisc(inp, -1);
2337 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2342 INP_WLOCK_ASSERT(inp);
2343 if ((inp->inp_flags & INP_TIMEWAIT) ||
2344 (inp->inp_flags & INP_DROPPED))
2347 tp = intotcpcb(inp);
2348 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2350 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2352 so = inp->inp_socket;
2353 SOCKBUF_LOCK(&so->so_snd);
2354 /* If the mss is larger than the socket buffer, decrease the mss. */
2355 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2356 tp->t_maxseg = so->so_snd.sb_hiwat;
2357 SOCKBUF_UNLOCK(&so->so_snd);
2359 TCPSTAT_INC(tcps_mturesent);
2361 tp->snd_nxt = tp->snd_una;
2362 tcp_free_sackholes(tp);
2363 tp->snd_recover = tp->snd_max;
2364 if (tp->t_flags & TF_SACK_PERMIT)
2365 EXIT_FASTRECOVERY(tp->t_flags);
2366 tp->t_fb->tfb_tcp_output(tp);
2371 * Look-up the routing entry to the peer of this inpcb. If no route
2372 * is found and it cannot be allocated, then return 0. This routine
2373 * is called by TCP routines that access the rmx structure and by
2374 * tcp_mss_update to get the peer/interface MTU.
2377 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2379 struct nhop4_extended nh4;
2383 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2385 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2387 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2388 NHR_REF, 0, &nh4) != 0)
2392 maxmtu = nh4.nh_mtu;
2394 /* Report additional interface capabilities. */
2396 if (ifp->if_capenable & IFCAP_TSO4 &&
2397 ifp->if_hwassist & CSUM_TSO) {
2398 cap->ifcap |= CSUM_TSO;
2399 cap->tsomax = ifp->if_hw_tsomax;
2400 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2401 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2404 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2412 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2414 struct nhop6_extended nh6;
2415 struct in6_addr dst6;
2420 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2422 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2423 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2424 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2429 maxmtu = nh6.nh_mtu;
2431 /* Report additional interface capabilities. */
2433 if (ifp->if_capenable & IFCAP_TSO6 &&
2434 ifp->if_hwassist & CSUM_TSO) {
2435 cap->ifcap |= CSUM_TSO;
2436 cap->tsomax = ifp->if_hw_tsomax;
2437 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2438 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2441 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2449 * Calculate effective SMSS per RFC5681 definition for a given TCP
2450 * connection at its current state, taking into account SACK and etc.
2453 tcp_maxseg(const struct tcpcb *tp)
2457 if (tp->t_flags & TF_NOOPT)
2458 return (tp->t_maxseg);
2461 * Here we have a simplified code from tcp_addoptions(),
2462 * without a proper loop, and having most of paddings hardcoded.
2463 * We might make mistakes with padding here in some edge cases,
2464 * but this is harmless, since result of tcp_maxseg() is used
2465 * only in cwnd and ssthresh estimations.
2467 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2468 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2469 if (tp->t_flags & TF_RCVD_TSTMP)
2470 optlen = TCPOLEN_TSTAMP_APPA;
2473 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2474 if (tp->t_flags & TF_SIGNATURE)
2475 optlen += PAD(TCPOLEN_SIGNATURE);
2477 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2478 optlen += TCPOLEN_SACKHDR;
2479 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2480 optlen = PAD(optlen);
2483 if (tp->t_flags & TF_REQ_TSTMP)
2484 optlen = TCPOLEN_TSTAMP_APPA;
2486 optlen = PAD(TCPOLEN_MAXSEG);
2487 if (tp->t_flags & TF_REQ_SCALE)
2488 optlen += PAD(TCPOLEN_WINDOW);
2489 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2490 if (tp->t_flags & TF_SIGNATURE)
2491 optlen += PAD(TCPOLEN_SIGNATURE);
2493 if (tp->t_flags & TF_SACK_PERMIT)
2494 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2497 optlen = min(optlen, TCP_MAXOLEN);
2498 return (tp->t_maxseg - optlen);
2502 sysctl_drop(SYSCTL_HANDLER_ARGS)
2504 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2505 struct sockaddr_storage addrs[2];
2509 struct sockaddr_in *fin, *lin;
2511 struct sockaddr_in6 *fin6, *lin6;
2522 if (req->oldptr != NULL || req->oldlen != 0)
2524 if (req->newptr == NULL)
2526 if (req->newlen < sizeof(addrs))
2528 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2532 switch (addrs[0].ss_family) {
2535 fin6 = (struct sockaddr_in6 *)&addrs[0];
2536 lin6 = (struct sockaddr_in6 *)&addrs[1];
2537 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2538 lin6->sin6_len != sizeof(struct sockaddr_in6))
2540 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2541 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2543 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2544 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2545 fin = (struct sockaddr_in *)&addrs[0];
2546 lin = (struct sockaddr_in *)&addrs[1];
2549 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2552 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2559 fin = (struct sockaddr_in *)&addrs[0];
2560 lin = (struct sockaddr_in *)&addrs[1];
2561 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2562 lin->sin_len != sizeof(struct sockaddr_in))
2569 INP_INFO_RLOCK(&V_tcbinfo);
2570 switch (addrs[0].ss_family) {
2573 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2574 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2575 INPLOOKUP_WLOCKPCB, NULL);
2580 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2581 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2586 if (inp->inp_flags & INP_TIMEWAIT) {
2588 * XXXRW: There currently exists a state where an
2589 * inpcb is present, but its timewait state has been
2590 * discarded. For now, don't allow dropping of this
2598 } else if (!(inp->inp_flags & INP_DROPPED) &&
2599 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2600 tp = intotcpcb(inp);
2601 tp = tcp_drop(tp, ECONNABORTED);
2608 INP_INFO_RUNLOCK(&V_tcbinfo);
2612 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2613 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2614 0, sysctl_drop, "", "Drop TCP connection");
2617 * Generate a standardized TCP log line for use throughout the
2618 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2619 * allow use in the interrupt context.
2621 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2622 * NB: The function may return NULL if memory allocation failed.
2624 * Due to header inclusion and ordering limitations the struct ip
2625 * and ip6_hdr pointers have to be passed as void pointers.
2628 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2632 /* Is logging enabled? */
2633 if (tcp_log_in_vain == 0)
2636 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2640 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2644 /* Is logging enabled? */
2645 if (tcp_log_debug == 0)
2648 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2652 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2659 const struct ip6_hdr *ip6;
2661 ip6 = (const struct ip6_hdr *)ip6hdr;
2663 ip = (struct ip *)ip4hdr;
2666 * The log line looks like this:
2667 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2669 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2670 sizeof(PRINT_TH_FLAGS) + 1 +
2672 2 * INET6_ADDRSTRLEN;
2674 2 * INET_ADDRSTRLEN;
2677 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2681 strcat(s, "TCP: [");
2684 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2685 inet_ntoa_r(inc->inc_faddr, sp);
2687 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2689 inet_ntoa_r(inc->inc_laddr, sp);
2691 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2694 ip6_sprintf(sp, &inc->inc6_faddr);
2696 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2698 ip6_sprintf(sp, &inc->inc6_laddr);
2700 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2701 } else if (ip6 && th) {
2702 ip6_sprintf(sp, &ip6->ip6_src);
2704 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2706 ip6_sprintf(sp, &ip6->ip6_dst);
2708 sprintf(sp, "]:%i", ntohs(th->th_dport));
2711 } else if (ip && th) {
2712 inet_ntoa_r(ip->ip_src, sp);
2714 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2716 inet_ntoa_r(ip->ip_dst, sp);
2718 sprintf(sp, "]:%i", ntohs(th->th_dport));
2726 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2727 if (*(s + size - 1) != '\0')
2728 panic("%s: string too long", __func__);
2733 * A subroutine which makes it easy to track TCP state changes with DTrace.
2734 * This function shouldn't be called for t_state initializations that don't
2735 * correspond to actual TCP state transitions.
2738 tcp_state_change(struct tcpcb *tp, int newstate)
2740 #if defined(KDTRACE_HOOKS)
2741 int pstate = tp->t_state;
2744 TCPSTATES_DEC(tp->t_state);
2745 TCPSTATES_INC(newstate);
2746 tp->t_state = newstate;
2747 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);