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/ip6.h>
82 #include <netinet6/in6_fib.h>
83 #include <netinet6/in6_pcb.h>
84 #include <netinet6/ip6_var.h>
85 #include <netinet6/scope6_var.h>
86 #include <netinet6/nd6.h>
90 #include <netinet/tcp_fastopen.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcp_syncache.h>
98 #include <netinet/cc/cc.h>
100 #include <netinet6/tcp6_var.h>
102 #include <netinet/tcpip.h>
104 #include <netinet/tcp_pcap.h>
107 #include <netinet/tcp_debug.h>
110 #include <netinet6/ip6protosw.h>
113 #include <netinet/tcp_offload.h>
117 #include <netipsec/ipsec.h>
118 #include <netipsec/xform.h>
120 #include <netipsec/ipsec6.h>
122 #include <netipsec/key.h>
123 #include <sys/syslog.h>
126 #include <machine/in_cksum.h>
129 #include <security/mac/mac_framework.h>
131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
136 struct rwlock tcp_function_lock;
139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
144 error = sysctl_handle_int(oidp, &new, 0, req);
145 if (error == 0 && req->newptr) {
146 if (new < TCP_MINMSS)
154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
155 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
156 &sysctl_net_inet_tcp_mss_check, "I",
157 "Default TCP Maximum Segment Size");
161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
165 new = V_tcp_v6mssdflt;
166 error = sysctl_handle_int(oidp, &new, 0, req);
167 if (error == 0 && req->newptr) {
168 if (new < TCP_MINMSS)
171 V_tcp_v6mssdflt = new;
176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
178 &sysctl_net_inet_tcp_mss_v6_check, "I",
179 "Default TCP Maximum Segment Size for IPv6");
183 * Minimum MSS we accept and use. This prevents DoS attacks where
184 * we are forced to a ridiculous low MSS like 20 and send hundreds
185 * of packets instead of one. The effect scales with the available
186 * bandwidth and quickly saturates the CPU and network interface
187 * with packet generation and sending. Set to zero to disable MINMSS
188 * checking. This setting prevents us from sending too small packets.
190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_minmss), 0,
193 "Minimum TCP Maximum Segment Size");
195 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_rfc1323), 0,
198 "Enable rfc1323 (high performance TCP) extensions");
200 static int tcp_log_debug = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
202 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
204 static int tcp_tcbhashsize;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
206 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
208 static int do_tcpdrain = 1;
209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
210 "Enable tcp_drain routine for extra help when low on mbufs");
212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
213 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
215 static VNET_DEFINE(int, icmp_may_rst) = 1;
216 #define V_icmp_may_rst VNET(icmp_may_rst)
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(icmp_may_rst), 0,
219 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
222 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_isn_reseed_interval), 0,
225 "Seconds between reseeding of ISN secret");
227 static int tcp_soreceive_stream;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
229 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
232 static int tcp_sig_checksigs = 1;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
234 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
237 VNET_DEFINE(uma_zone_t, sack_hole_zone);
238 #define V_sack_hole_zone VNET(sack_hole_zone)
240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
244 static void tcp_mtudisc(struct inpcb *, int);
245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
246 void *ip4hdr, const void *ip6hdr);
247 static void tcp_timer_discard(struct tcpcb *, uint32_t);
250 static struct tcp_function_block tcp_def_funcblk = {
254 tcp_default_ctloutput,
266 int t_functions_inited = 0;
267 struct tcp_funchead t_functions;
268 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
271 init_tcp_functions(void)
273 if (t_functions_inited == 0) {
274 TAILQ_INIT(&t_functions);
275 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
276 t_functions_inited = 1;
280 static struct tcp_function_block *
281 find_tcp_functions_locked(struct tcp_function_set *fs)
283 struct tcp_function *f;
284 struct tcp_function_block *blk=NULL;
286 TAILQ_FOREACH(f, &t_functions, tf_next) {
287 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
295 static struct tcp_function_block *
296 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
298 struct tcp_function_block *rblk=NULL;
299 struct tcp_function *f;
301 TAILQ_FOREACH(f, &t_functions, tf_next) {
302 if (f->tf_fb == blk) {
313 struct tcp_function_block *
314 find_and_ref_tcp_functions(struct tcp_function_set *fs)
316 struct tcp_function_block *blk;
318 rw_rlock(&tcp_function_lock);
319 blk = find_tcp_functions_locked(fs);
321 refcount_acquire(&blk->tfb_refcnt);
322 rw_runlock(&tcp_function_lock);
326 struct tcp_function_block *
327 find_and_ref_tcp_fb(struct tcp_function_block *blk)
329 struct tcp_function_block *rblk;
331 rw_rlock(&tcp_function_lock);
332 rblk = find_tcp_fb_locked(blk, NULL);
334 refcount_acquire(&rblk->tfb_refcnt);
335 rw_runlock(&tcp_function_lock);
341 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
344 struct tcp_function_set fs;
345 struct tcp_function_block *blk;
347 memset(&fs, 0, sizeof(fs));
348 rw_rlock(&tcp_function_lock);
349 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
352 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
353 fs.pcbcnt = blk->tfb_refcnt;
355 rw_runlock(&tcp_function_lock);
356 error = sysctl_handle_string(oidp, fs.function_set_name,
357 sizeof(fs.function_set_name), req);
359 /* Check for error or no change */
360 if (error != 0 || req->newptr == NULL)
363 rw_wlock(&tcp_function_lock);
364 blk = find_tcp_functions_locked(&fs);
366 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
370 tcp_func_set_ptr = blk;
372 rw_wunlock(&tcp_function_lock);
376 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
377 CTLTYPE_STRING | CTLFLAG_RW,
378 NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
379 "Set/get the default TCP functions");
382 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
384 int error, cnt, linesz;
385 struct tcp_function *f;
390 rw_rlock(&tcp_function_lock);
391 TAILQ_FOREACH(f, &t_functions, tf_next) {
394 rw_runlock(&tcp_function_lock);
396 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
397 buffer = malloc(bufsz, M_TEMP, M_WAITOK);
402 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
407 rw_rlock(&tcp_function_lock);
408 TAILQ_FOREACH(f, &t_functions, tf_next) {
409 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
410 f->tf_fb->tfb_tcp_block_name,
411 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
412 f->tf_fb->tfb_refcnt);
413 if (linesz >= bufsz) {
421 rw_runlock(&tcp_function_lock);
423 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
424 free(buffer, M_TEMP);
428 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
429 CTLTYPE_STRING|CTLFLAG_RD,
430 NULL, 0, sysctl_net_inet_list_available, "A",
431 "list available TCP Function sets");
434 * Target size of TCP PCB hash tables. Must be a power of two.
436 * Note that this can be overridden by the kernel environment
437 * variable net.inet.tcp.tcbhashsize
440 #define TCBHASHSIZE 0
445 * Callouts should be moved into struct tcp directly. They are currently
446 * separate because the tcpcb structure is exported to userland for sysctl
447 * parsing purposes, which do not know about callouts.
456 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
457 #define V_tcpcb_zone VNET(tcpcb_zone)
459 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
460 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
462 static struct mtx isn_mtx;
464 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
465 #define ISN_LOCK() mtx_lock(&isn_mtx)
466 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
469 * TCP initialization.
472 tcp_zone_change(void *tag)
475 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
476 uma_zone_set_max(V_tcpcb_zone, maxsockets);
477 tcp_tw_zone_change();
481 tcp_inpcb_init(void *mem, int size, int flags)
483 struct inpcb *inp = mem;
485 INP_LOCK_INIT(inp, "inp", "tcpinp");
490 * Take a value and get the next power of 2 that doesn't overflow.
491 * Used to size the tcp_inpcb hash buckets.
494 maketcp_hashsize(int size)
500 * get the next power of 2 higher than maxsockets.
502 hashsize = 1 << fls(size);
503 /* catch overflow, and just go one power of 2 smaller */
504 if (hashsize < size) {
505 hashsize = 1 << (fls(size) - 1);
511 register_tcp_functions(struct tcp_function_block *blk, int wait)
513 struct tcp_function_block *lblk;
514 struct tcp_function *n;
515 struct tcp_function_set fs;
517 if (t_functions_inited == 0) {
518 init_tcp_functions();
520 if ((blk->tfb_tcp_output == NULL) ||
521 (blk->tfb_tcp_do_segment == NULL) ||
522 (blk->tfb_tcp_ctloutput == NULL) ||
523 (strlen(blk->tfb_tcp_block_name) == 0)) {
525 * These functions are required and you
530 if (blk->tfb_tcp_timer_stop_all ||
531 blk->tfb_tcp_timers_left ||
532 blk->tfb_tcp_timer_activate ||
533 blk->tfb_tcp_timer_active ||
534 blk->tfb_tcp_timer_stop) {
536 * If you define one timer function you
537 * must have them all.
539 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
540 (blk->tfb_tcp_timers_left == NULL) ||
541 (blk->tfb_tcp_timer_activate == NULL) ||
542 (blk->tfb_tcp_timer_active == NULL) ||
543 (blk->tfb_tcp_timer_stop == NULL)) {
547 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
552 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
553 rw_wlock(&tcp_function_lock);
554 lblk = find_tcp_functions_locked(&fs);
556 /* Duplicate name space not allowed */
557 rw_wunlock(&tcp_function_lock);
558 free(n, M_TCPFUNCTIONS);
561 refcount_init(&blk->tfb_refcnt, 0);
563 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
564 rw_wunlock(&tcp_function_lock);
569 deregister_tcp_functions(struct tcp_function_block *blk)
571 struct tcp_function_block *lblk;
572 struct tcp_function *f;
575 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
576 /* You can't un-register the default */
579 rw_wlock(&tcp_function_lock);
580 if (blk == tcp_func_set_ptr) {
581 /* You can't free the current default */
582 rw_wunlock(&tcp_function_lock);
585 if (blk->tfb_refcnt) {
586 /* Still tcb attached, mark it. */
587 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
588 rw_wunlock(&tcp_function_lock);
591 lblk = find_tcp_fb_locked(blk, &f);
594 TAILQ_REMOVE(&t_functions, f, tf_next);
596 free(f, M_TCPFUNCTIONS);
599 rw_wunlock(&tcp_function_lock);
606 const char *tcbhash_tuneable;
609 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
611 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
612 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
613 printf("%s: WARNING: unable to register helper hook\n", __func__);
614 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
615 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
616 printf("%s: WARNING: unable to register helper hook\n", __func__);
617 hashsize = TCBHASHSIZE;
618 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
621 * Auto tune the hash size based on maxsockets.
622 * A perfect hash would have a 1:1 mapping
623 * (hashsize = maxsockets) however it's been
624 * suggested that O(2) average is better.
626 hashsize = maketcp_hashsize(maxsockets / 4);
628 * Our historical default is 512,
629 * do not autotune lower than this.
633 if (bootverbose && IS_DEFAULT_VNET(curvnet))
634 printf("%s: %s auto tuned to %d\n", __func__,
635 tcbhash_tuneable, hashsize);
638 * We require a hashsize to be a power of two.
639 * Previously if it was not a power of two we would just reset it
640 * back to 512, which could be a nasty surprise if you did not notice
642 * Instead what we do is clip it to the closest power of two lower
643 * than the specified hash value.
645 if (!powerof2(hashsize)) {
646 int oldhashsize = hashsize;
648 hashsize = maketcp_hashsize(hashsize);
649 /* prevent absurdly low value */
652 printf("%s: WARNING: TCB hash size not a power of 2, "
653 "clipped from %d to %d.\n", __func__, oldhashsize,
656 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
657 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
658 IPI_HASHFIELDS_4TUPLE);
661 * These have to be type stable for the benefit of the timers.
663 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
664 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
665 uma_zone_set_max(V_tcpcb_zone, maxsockets);
666 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
672 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
673 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
674 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
676 /* Skip initialization of globals for non-default instances. */
677 if (!IS_DEFAULT_VNET(curvnet))
680 tcp_reass_global_init();
682 /* XXX virtualize those bellow? */
683 tcp_delacktime = TCPTV_DELACK;
684 tcp_keepinit = TCPTV_KEEP_INIT;
685 tcp_keepidle = TCPTV_KEEP_IDLE;
686 tcp_keepintvl = TCPTV_KEEPINTVL;
687 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
689 tcp_rexmit_min = TCPTV_MIN;
690 if (tcp_rexmit_min < 1)
692 tcp_persmin = TCPTV_PERSMIN;
693 tcp_persmax = TCPTV_PERSMAX;
694 tcp_rexmit_slop = TCPTV_CPU_VAR;
695 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
696 tcp_tcbhashsize = hashsize;
697 /* Setup the tcp function block list */
698 init_tcp_functions();
699 register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
701 if (tcp_soreceive_stream) {
703 tcp_usrreqs.pru_soreceive = soreceive_stream;
706 tcp6_usrreqs.pru_soreceive = soreceive_stream;
711 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
713 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
715 if (max_protohdr < TCP_MINPROTOHDR)
716 max_protohdr = TCP_MINPROTOHDR;
717 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
719 #undef TCP_MINPROTOHDR
722 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
723 SHUTDOWN_PRI_DEFAULT);
724 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
725 EVENTHANDLER_PRI_ANY);
742 tcp_fastopen_destroy();
747 in_pcbinfo_destroy(&V_tcbinfo);
748 uma_zdestroy(V_sack_hole_zone);
749 uma_zdestroy(V_tcpcb_zone);
751 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
753 printf("%s: WARNING: unable to deregister helper hook "
754 "type=%d, id=%d: error %d returned\n", __func__,
755 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
757 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
759 printf("%s: WARNING: unable to deregister helper hook "
760 "type=%d, id=%d: error %d returned\n", __func__,
761 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
773 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
774 * tcp_template used to store this data in mbufs, but we now recopy it out
775 * of the tcpcb each time to conserve mbufs.
778 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
780 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
782 INP_WLOCK_ASSERT(inp);
785 if ((inp->inp_vflag & INP_IPV6) != 0) {
788 ip6 = (struct ip6_hdr *)ip_ptr;
789 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
790 (inp->inp_flow & IPV6_FLOWINFO_MASK);
791 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
792 (IPV6_VERSION & IPV6_VERSION_MASK);
793 ip6->ip6_nxt = IPPROTO_TCP;
794 ip6->ip6_plen = htons(sizeof(struct tcphdr));
795 ip6->ip6_src = inp->in6p_laddr;
796 ip6->ip6_dst = inp->in6p_faddr;
799 #if defined(INET6) && defined(INET)
806 ip = (struct ip *)ip_ptr;
807 ip->ip_v = IPVERSION;
809 ip->ip_tos = inp->inp_ip_tos;
813 ip->ip_ttl = inp->inp_ip_ttl;
815 ip->ip_p = IPPROTO_TCP;
816 ip->ip_src = inp->inp_laddr;
817 ip->ip_dst = inp->inp_faddr;
820 th->th_sport = inp->inp_lport;
821 th->th_dport = inp->inp_fport;
829 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
833 * Create template to be used to send tcp packets on a connection.
834 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
835 * use for this function is in keepalives, which use tcp_respond.
838 tcpip_maketemplate(struct inpcb *inp)
842 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
845 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
850 * Send a single message to the TCP at address specified by
851 * the given TCP/IP header. If m == NULL, then we make a copy
852 * of the tcpiphdr at th and send directly to the addressed host.
853 * This is used to force keep alive messages out using the TCP
854 * template for a connection. If flags are given then we send
855 * a message back to the TCP which originated the segment th,
856 * and discard the mbuf containing it and any other attached mbufs.
858 * In any case the ack and sequence number of the transmitted
859 * segment are as specified by the parameters.
861 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
864 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
865 tcp_seq ack, tcp_seq seq, int flags)
876 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
879 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
886 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
887 INP_WLOCK_ASSERT(inp);
893 if (!(flags & TH_RST)) {
894 win = sbspace(&inp->inp_socket->so_rcv);
895 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
896 win = (long)TCP_MAXWIN << tp->rcv_scale;
900 m = m_gethdr(M_NOWAIT, MT_DATA);
904 m->m_data += max_linkhdr;
907 bcopy((caddr_t)ip6, mtod(m, caddr_t),
908 sizeof(struct ip6_hdr));
909 ip6 = mtod(m, struct ip6_hdr *);
910 nth = (struct tcphdr *)(ip6 + 1);
914 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
915 ip = mtod(m, struct ip *);
916 nth = (struct tcphdr *)(ip + 1);
918 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
923 * XXX MRT We inherrit the FIB, which is lucky.
927 m->m_data = (caddr_t)ipgen;
928 /* m_len is set later */
930 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
933 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
934 nth = (struct tcphdr *)(ip6 + 1);
938 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
939 nth = (struct tcphdr *)(ip + 1);
943 * this is usually a case when an extension header
944 * exists between the IPv6 header and the
947 nth->th_sport = th->th_sport;
948 nth->th_dport = th->th_dport;
950 xchg(nth->th_dport, nth->th_sport, uint16_t);
956 ip6->ip6_vfc = IPV6_VERSION;
957 ip6->ip6_nxt = IPPROTO_TCP;
958 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
959 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
962 #if defined(INET) && defined(INET6)
967 tlen += sizeof (struct tcpiphdr);
968 ip->ip_len = htons(tlen);
969 ip->ip_ttl = V_ip_defttl;
970 if (V_path_mtu_discovery)
971 ip->ip_off |= htons(IP_DF);
975 m->m_pkthdr.len = tlen;
976 m->m_pkthdr.rcvif = NULL;
980 * Packet is associated with a socket, so allow the
981 * label of the response to reflect the socket label.
983 INP_WLOCK_ASSERT(inp);
984 mac_inpcb_create_mbuf(inp, m);
987 * Packet is not associated with a socket, so possibly
988 * update the label in place.
990 mac_netinet_tcp_reply(m);
993 nth->th_seq = htonl(seq);
994 nth->th_ack = htonl(ack);
996 nth->th_off = sizeof (struct tcphdr) >> 2;
997 nth->th_flags = flags;
999 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1001 nth->th_win = htons((u_short)win);
1004 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1007 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1008 nth->th_sum = in6_cksum_pseudo(ip6,
1009 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1010 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1014 #if defined(INET6) && defined(INET)
1019 m->m_pkthdr.csum_flags = CSUM_TCP;
1020 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1021 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1025 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1026 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1028 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1030 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1033 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1036 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1038 #if defined(INET) && defined(INET6)
1042 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1047 * Create a new TCP control block, making an
1048 * empty reassembly queue and hooking it to the argument
1049 * protocol control block. The `inp' parameter must have
1050 * come from the zone allocator set up in tcp_init().
1053 tcp_newtcpcb(struct inpcb *inp)
1055 struct tcpcb_mem *tm;
1058 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1061 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1066 /* Initialise cc_var struct for this tcpcb. */
1068 tp->ccv->type = IPPROTO_TCP;
1069 tp->ccv->ccvc.tcp = tp;
1070 rw_rlock(&tcp_function_lock);
1071 tp->t_fb = tcp_func_set_ptr;
1072 refcount_acquire(&tp->t_fb->tfb_refcnt);
1073 rw_runlock(&tcp_function_lock);
1074 if (tp->t_fb->tfb_tcp_fb_init) {
1075 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1078 * Use the current system default CC algorithm.
1081 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1082 CC_ALGO(tp) = CC_DEFAULT();
1085 if (CC_ALGO(tp)->cb_init != NULL)
1086 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1087 if (tp->t_fb->tfb_tcp_fb_fini)
1088 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1089 refcount_release(&tp->t_fb->tfb_refcnt);
1090 uma_zfree(V_tcpcb_zone, tm);
1095 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1096 if (tp->t_fb->tfb_tcp_fb_fini)
1097 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1098 refcount_release(&tp->t_fb->tfb_refcnt);
1099 uma_zfree(V_tcpcb_zone, tm);
1104 tp->t_vnet = inp->inp_vnet;
1106 tp->t_timers = &tm->tt;
1107 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
1110 isipv6 ? V_tcp_v6mssdflt :
1114 /* Set up our timeouts. */
1115 callout_init(&tp->t_timers->tt_rexmt, 1);
1116 callout_init(&tp->t_timers->tt_persist, 1);
1117 callout_init(&tp->t_timers->tt_keep, 1);
1118 callout_init(&tp->t_timers->tt_2msl, 1);
1119 callout_init(&tp->t_timers->tt_delack, 1);
1121 if (V_tcp_do_rfc1323)
1122 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1124 tp->t_flags |= TF_SACK_PERMIT;
1125 TAILQ_INIT(&tp->snd_holes);
1127 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1130 in_pcbref(inp); /* Reference for tcpcb */
1134 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1135 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1136 * reasonable initial retransmit time.
1138 tp->t_srtt = TCPTV_SRTTBASE;
1139 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1140 tp->t_rttmin = tcp_rexmit_min;
1141 tp->t_rxtcur = TCPTV_RTOBASE;
1142 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1143 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1144 tp->t_rcvtime = ticks;
1146 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1147 * because the socket may be bound to an IPv6 wildcard address,
1148 * which may match an IPv4-mapped IPv6 address.
1150 inp->inp_ip_ttl = V_ip_defttl;
1154 * Init the TCP PCAP queues.
1156 tcp_pcap_tcpcb_init(tp);
1158 return (tp); /* XXX */
1162 * Switch the congestion control algorithm back to NewReno for any active
1163 * control blocks using an algorithm which is about to go away.
1164 * This ensures the CC framework can allow the unload to proceed without leaving
1165 * any dangling pointers which would trigger a panic.
1166 * Returning non-zero would inform the CC framework that something went wrong
1167 * and it would be unsafe to allow the unload to proceed. However, there is no
1168 * way for this to occur with this implementation so we always return zero.
1171 tcp_ccalgounload(struct cc_algo *unload_algo)
1173 struct cc_algo *tmpalgo;
1176 VNET_ITERATOR_DECL(vnet_iter);
1179 * Check all active control blocks across all network stacks and change
1180 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1181 * requires cleanup code to be run, call it.
1184 VNET_FOREACH(vnet_iter) {
1185 CURVNET_SET(vnet_iter);
1186 INP_INFO_WLOCK(&V_tcbinfo);
1188 * New connections already part way through being initialised
1189 * with the CC algo we're removing will not race with this code
1190 * because the INP_INFO_WLOCK is held during initialisation. We
1191 * therefore don't enter the loop below until the connection
1192 * list has stabilised.
1194 LIST_FOREACH(inp, &V_tcb, inp_list) {
1196 /* Important to skip tcptw structs. */
1197 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1198 (tp = intotcpcb(inp)) != NULL) {
1200 * By holding INP_WLOCK here, we are assured
1201 * that the connection is not currently
1202 * executing inside the CC module's functions
1203 * i.e. it is safe to make the switch back to
1206 if (CC_ALGO(tp) == unload_algo) {
1207 tmpalgo = CC_ALGO(tp);
1208 /* NewReno does not require any init. */
1209 CC_ALGO(tp) = &newreno_cc_algo;
1210 if (tmpalgo->cb_destroy != NULL)
1211 tmpalgo->cb_destroy(tp->ccv);
1216 INP_INFO_WUNLOCK(&V_tcbinfo);
1219 VNET_LIST_RUNLOCK();
1225 * Drop a TCP connection, reporting
1226 * the specified error. If connection is synchronized,
1227 * then send a RST to peer.
1230 tcp_drop(struct tcpcb *tp, int errno)
1232 struct socket *so = tp->t_inpcb->inp_socket;
1234 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1235 INP_WLOCK_ASSERT(tp->t_inpcb);
1237 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1238 tcp_state_change(tp, TCPS_CLOSED);
1239 (void) tp->t_fb->tfb_tcp_output(tp);
1240 TCPSTAT_INC(tcps_drops);
1242 TCPSTAT_INC(tcps_conndrops);
1243 if (errno == ETIMEDOUT && tp->t_softerror)
1244 errno = tp->t_softerror;
1245 so->so_error = errno;
1246 return (tcp_close(tp));
1250 tcp_discardcb(struct tcpcb *tp)
1252 struct inpcb *inp = tp->t_inpcb;
1253 struct socket *so = inp->inp_socket;
1255 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1259 INP_WLOCK_ASSERT(inp);
1262 * Make sure that all of our timers are stopped before we delete the
1265 * If stopping a timer fails, we schedule a discard function in same
1266 * callout, and the last discard function called will take care of
1267 * deleting the tcpcb.
1269 tcp_timer_stop(tp, TT_REXMT);
1270 tcp_timer_stop(tp, TT_PERSIST);
1271 tcp_timer_stop(tp, TT_KEEP);
1272 tcp_timer_stop(tp, TT_2MSL);
1273 tcp_timer_stop(tp, TT_DELACK);
1274 if (tp->t_fb->tfb_tcp_timer_stop_all) {
1275 /* Call the stop-all function of the methods */
1276 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1280 * If we got enough samples through the srtt filter,
1281 * save the rtt and rttvar in the routing entry.
1282 * 'Enough' is arbitrarily defined as 4 rtt samples.
1283 * 4 samples is enough for the srtt filter to converge
1284 * to within enough % of the correct value; fewer samples
1285 * and we could save a bogus rtt. The danger is not high
1286 * as tcp quickly recovers from everything.
1287 * XXX: Works very well but needs some more statistics!
1289 if (tp->t_rttupdated >= 4) {
1290 struct hc_metrics_lite metrics;
1293 bzero(&metrics, sizeof(metrics));
1295 * Update the ssthresh always when the conditions below
1296 * are satisfied. This gives us better new start value
1297 * for the congestion avoidance for new connections.
1298 * ssthresh is only set if packet loss occured on a session.
1300 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1301 * being torn down. Ideally this code would not use 'so'.
1303 ssthresh = tp->snd_ssthresh;
1304 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1306 * convert the limit from user data bytes to
1307 * packets then to packet data bytes.
1309 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1312 ssthresh *= (u_long)(tp->t_maxseg +
1314 (isipv6 ? sizeof (struct ip6_hdr) +
1315 sizeof (struct tcphdr) :
1317 sizeof (struct tcpiphdr)
1324 metrics.rmx_ssthresh = ssthresh;
1326 metrics.rmx_rtt = tp->t_srtt;
1327 metrics.rmx_rttvar = tp->t_rttvar;
1328 metrics.rmx_cwnd = tp->snd_cwnd;
1329 metrics.rmx_sendpipe = 0;
1330 metrics.rmx_recvpipe = 0;
1332 tcp_hc_update(&inp->inp_inc, &metrics);
1335 /* free the reassembly queue, if any */
1336 tcp_reass_flush(tp);
1339 /* Disconnect offload device, if any. */
1340 if (tp->t_flags & TF_TOE)
1341 tcp_offload_detach(tp);
1344 tcp_free_sackholes(tp);
1347 /* Free the TCP PCAP queues. */
1348 tcp_pcap_drain(&(tp->t_inpkts));
1349 tcp_pcap_drain(&(tp->t_outpkts));
1352 /* Allow the CC algorithm to clean up after itself. */
1353 if (CC_ALGO(tp)->cb_destroy != NULL)
1354 CC_ALGO(tp)->cb_destroy(tp->ccv);
1356 khelp_destroy_osd(tp->osd);
1359 inp->inp_ppcb = NULL;
1360 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1361 /* We own the last reference on tcpcb, let's free it. */
1362 if ((tp->t_fb->tfb_tcp_timers_left) &&
1363 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1364 /* Some fb timers left running! */
1367 if (tp->t_fb->tfb_tcp_fb_fini)
1368 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1369 refcount_release(&tp->t_fb->tfb_refcnt);
1371 uma_zfree(V_tcpcb_zone, tp);
1372 released = in_pcbrele_wlocked(inp);
1373 KASSERT(!released, ("%s: inp %p should not have been released "
1374 "here", __func__, inp));
1379 tcp_timer_2msl_discard(void *xtp)
1382 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1386 tcp_timer_keep_discard(void *xtp)
1389 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1393 tcp_timer_persist_discard(void *xtp)
1396 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1400 tcp_timer_rexmt_discard(void *xtp)
1403 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1407 tcp_timer_delack_discard(void *xtp)
1410 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1414 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1418 CURVNET_SET(tp->t_vnet);
1419 INP_INFO_RLOCK(&V_tcbinfo);
1421 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1424 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1425 ("%s: tcpcb has to be stopped here", __func__));
1426 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1427 ("%s: discard callout should be running", __func__));
1428 tp->t_timers->tt_flags &= ~timer_type;
1429 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1430 /* We own the last reference on this tcpcb, let's free it. */
1431 if ((tp->t_fb->tfb_tcp_timers_left) &&
1432 (tp->t_fb->tfb_tcp_timers_left(tp))) {
1433 /* Some fb timers left running! */
1436 if (tp->t_fb->tfb_tcp_fb_fini)
1437 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1438 refcount_release(&tp->t_fb->tfb_refcnt);
1440 uma_zfree(V_tcpcb_zone, tp);
1441 if (in_pcbrele_wlocked(inp)) {
1442 INP_INFO_RUNLOCK(&V_tcbinfo);
1449 INP_INFO_RUNLOCK(&V_tcbinfo);
1454 * Attempt to close a TCP control block, marking it as dropped, and freeing
1455 * the socket if we hold the only reference.
1458 tcp_close(struct tcpcb *tp)
1460 struct inpcb *inp = tp->t_inpcb;
1463 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1464 INP_WLOCK_ASSERT(inp);
1467 if (tp->t_state == TCPS_LISTEN)
1468 tcp_offload_listen_stop(tp);
1472 * This releases the TFO pending counter resource for TFO listen
1473 * sockets as well as passively-created TFO sockets that transition
1474 * from SYN_RECEIVED to CLOSED.
1476 if (tp->t_tfo_pending) {
1477 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1478 tp->t_tfo_pending = NULL;
1482 TCPSTAT_INC(tcps_closed);
1483 TCPSTAT_DEC(tcps_states[tp->t_state]);
1484 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1485 so = inp->inp_socket;
1486 soisdisconnected(so);
1487 if (inp->inp_flags & INP_SOCKREF) {
1488 KASSERT(so->so_state & SS_PROTOREF,
1489 ("tcp_close: !SS_PROTOREF"));
1490 inp->inp_flags &= ~INP_SOCKREF;
1494 so->so_state &= ~SS_PROTOREF;
1504 VNET_ITERATOR_DECL(vnet_iter);
1509 VNET_LIST_RLOCK_NOSLEEP();
1510 VNET_FOREACH(vnet_iter) {
1511 CURVNET_SET(vnet_iter);
1516 * Walk the tcpbs, if existing, and flush the reassembly queue,
1517 * if there is one...
1518 * XXX: The "Net/3" implementation doesn't imply that the TCP
1519 * reassembly queue should be flushed, but in a situation
1520 * where we're really low on mbufs, this is potentially
1523 INP_INFO_WLOCK(&V_tcbinfo);
1524 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1525 if (inpb->inp_flags & INP_TIMEWAIT)
1528 if ((tcpb = intotcpcb(inpb)) != NULL) {
1529 tcp_reass_flush(tcpb);
1530 tcp_clean_sackreport(tcpb);
1534 INP_INFO_WUNLOCK(&V_tcbinfo);
1537 VNET_LIST_RUNLOCK_NOSLEEP();
1541 * Notify a tcp user of an asynchronous error;
1542 * store error as soft error, but wake up user
1543 * (for now, won't do anything until can select for soft error).
1545 * Do not wake up user since there currently is no mechanism for
1546 * reporting soft errors (yet - a kqueue filter may be added).
1548 static struct inpcb *
1549 tcp_notify(struct inpcb *inp, int error)
1553 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1554 INP_WLOCK_ASSERT(inp);
1556 if ((inp->inp_flags & INP_TIMEWAIT) ||
1557 (inp->inp_flags & INP_DROPPED))
1560 tp = intotcpcb(inp);
1561 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1564 * Ignore some errors if we are hooked up.
1565 * If connection hasn't completed, has retransmitted several times,
1566 * and receives a second error, give up now. This is better
1567 * than waiting a long time to establish a connection that
1568 * can never complete.
1570 if (tp->t_state == TCPS_ESTABLISHED &&
1571 (error == EHOSTUNREACH || error == ENETUNREACH ||
1572 error == EHOSTDOWN)) {
1574 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1576 tp = tcp_drop(tp, error);
1582 tp->t_softerror = error;
1586 wakeup( &so->so_timeo);
1593 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1595 int error, i, m, n, pcb_count;
1596 struct inpcb *inp, **inp_list;
1601 * The process of preparing the TCB list is too time-consuming and
1602 * resource-intensive to repeat twice on every request.
1604 if (req->oldptr == NULL) {
1605 n = V_tcbinfo.ipi_count +
1606 TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]);
1607 n += imax(n / 8, 10);
1608 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1612 if (req->newptr != NULL)
1616 * OK, now we're committed to doing something.
1618 INP_LIST_RLOCK(&V_tcbinfo);
1619 gencnt = V_tcbinfo.ipi_gencnt;
1620 n = V_tcbinfo.ipi_count;
1621 INP_LIST_RUNLOCK(&V_tcbinfo);
1623 m = TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]);
1625 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1626 + (n + m) * sizeof(struct xtcpcb));
1630 xig.xig_len = sizeof xig;
1631 xig.xig_count = n + m;
1632 xig.xig_gen = gencnt;
1633 xig.xig_sogen = so_gencnt;
1634 error = SYSCTL_OUT(req, &xig, sizeof xig);
1638 error = syncache_pcblist(req, m, &pcb_count);
1642 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1643 if (inp_list == NULL)
1646 INP_INFO_WLOCK(&V_tcbinfo);
1647 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1648 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1650 if (inp->inp_gencnt <= gencnt) {
1652 * XXX: This use of cr_cansee(), introduced with
1653 * TCP state changes, is not quite right, but for
1654 * now, better than nothing.
1656 if (inp->inp_flags & INP_TIMEWAIT) {
1657 if (intotw(inp) != NULL)
1658 error = cr_cansee(req->td->td_ucred,
1659 intotw(inp)->tw_cred);
1661 error = EINVAL; /* Skip this inp. */
1663 error = cr_canseeinpcb(req->td->td_ucred, inp);
1666 inp_list[i++] = inp;
1671 INP_INFO_WUNLOCK(&V_tcbinfo);
1675 for (i = 0; i < n; i++) {
1678 if (inp->inp_gencnt <= gencnt) {
1682 bzero(&xt, sizeof(xt));
1683 xt.xt_len = sizeof xt;
1684 /* XXX should avoid extra copy */
1685 bcopy(inp, &xt.xt_inp, sizeof *inp);
1686 inp_ppcb = inp->inp_ppcb;
1687 if (inp_ppcb == NULL)
1688 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1689 else if (inp->inp_flags & INP_TIMEWAIT) {
1690 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1691 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1693 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1694 if (xt.xt_tp.t_timers)
1695 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1697 if (inp->inp_socket != NULL)
1698 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1700 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1701 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1703 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1705 error = SYSCTL_OUT(req, &xt, sizeof xt);
1709 INP_INFO_RLOCK(&V_tcbinfo);
1710 for (i = 0; i < n; i++) {
1713 if (!in_pcbrele_rlocked(inp))
1716 INP_INFO_RUNLOCK(&V_tcbinfo);
1720 * Give the user an updated idea of our state.
1721 * If the generation differs from what we told
1722 * her before, she knows that something happened
1723 * while we were processing this request, and it
1724 * might be necessary to retry.
1726 INP_LIST_RLOCK(&V_tcbinfo);
1727 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1728 xig.xig_sogen = so_gencnt;
1729 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1730 INP_LIST_RUNLOCK(&V_tcbinfo);
1731 error = SYSCTL_OUT(req, &xig, sizeof xig);
1733 free(inp_list, M_TEMP);
1737 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1738 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1739 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1743 tcp_getcred(SYSCTL_HANDLER_ARGS)
1746 struct sockaddr_in addrs[2];
1750 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1753 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1756 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1757 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1759 if (inp->inp_socket == NULL)
1762 error = cr_canseeinpcb(req->td->td_ucred, inp);
1764 cru2x(inp->inp_cred, &xuc);
1769 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1773 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1774 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1775 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1780 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1783 struct sockaddr_in6 addrs[2];
1790 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1793 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1796 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1797 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1800 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1802 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1811 inp = in_pcblookup(&V_tcbinfo,
1812 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1814 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1815 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1818 inp = in6_pcblookup(&V_tcbinfo,
1819 &addrs[1].sin6_addr, addrs[1].sin6_port,
1820 &addrs[0].sin6_addr, addrs[0].sin6_port,
1821 INPLOOKUP_RLOCKPCB, NULL);
1823 if (inp->inp_socket == NULL)
1826 error = cr_canseeinpcb(req->td->td_ucred, inp);
1828 cru2x(inp->inp_cred, &xuc);
1833 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1837 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1838 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1839 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1845 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1847 struct ip *ip = vip;
1849 struct in_addr faddr;
1852 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1854 struct in_conninfo inc;
1855 tcp_seq icmp_tcp_seq;
1858 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1859 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1862 if (cmd == PRC_MSGSIZE)
1863 notify = tcp_mtudisc_notify;
1864 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1865 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1866 notify = tcp_drop_syn_sent;
1868 * Redirects don't need to be handled up here.
1870 else if (PRC_IS_REDIRECT(cmd))
1873 * Hostdead is ugly because it goes linearly through all PCBs.
1874 * XXX: We never get this from ICMP, otherwise it makes an
1875 * excellent DoS attack on machines with many connections.
1877 else if (cmd == PRC_HOSTDEAD)
1879 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1883 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1887 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1888 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1889 INP_INFO_RLOCK(&V_tcbinfo);
1890 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1891 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1893 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1894 !(inp->inp_flags & INP_DROPPED) &&
1895 !(inp->inp_socket == NULL)) {
1896 icmp_tcp_seq = ntohl(th->th_seq);
1897 tp = intotcpcb(inp);
1898 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1899 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1900 if (cmd == PRC_MSGSIZE) {
1903 * If we got a needfrag set the MTU
1904 * in the route to the suggested new
1905 * value (if given) and then notify.
1907 mtu = ntohs(icp->icmp_nextmtu);
1909 * If no alternative MTU was
1910 * proposed, try the next smaller
1915 ntohs(ip->ip_len), 1);
1916 if (mtu < V_tcp_minmss +
1917 sizeof(struct tcpiphdr))
1918 mtu = V_tcp_minmss +
1919 sizeof(struct tcpiphdr);
1921 * Only process the offered MTU if it
1922 * is smaller than the current one.
1924 if (mtu < tp->t_maxseg +
1925 sizeof(struct tcpiphdr)) {
1926 bzero(&inc, sizeof(inc));
1927 inc.inc_faddr = faddr;
1929 inp->inp_inc.inc_fibnum;
1930 tcp_hc_updatemtu(&inc, mtu);
1931 tcp_mtudisc(inp, mtu);
1934 inp = (*notify)(inp,
1935 inetctlerrmap[cmd]);
1941 bzero(&inc, sizeof(inc));
1942 inc.inc_fport = th->th_dport;
1943 inc.inc_lport = th->th_sport;
1944 inc.inc_faddr = faddr;
1945 inc.inc_laddr = ip->ip_src;
1946 syncache_unreach(&inc, th);
1948 INP_INFO_RUNLOCK(&V_tcbinfo);
1954 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1957 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1958 struct ip6_hdr *ip6;
1960 struct ip6ctlparam *ip6cp = NULL;
1961 const struct sockaddr_in6 *sa6_src = NULL;
1963 struct tcp_portonly {
1968 if (sa->sa_family != AF_INET6 ||
1969 sa->sa_len != sizeof(struct sockaddr_in6))
1972 if (cmd == PRC_MSGSIZE)
1973 notify = tcp_mtudisc_notify;
1974 else if (!PRC_IS_REDIRECT(cmd) &&
1975 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1978 /* if the parameter is from icmp6, decode it. */
1980 ip6cp = (struct ip6ctlparam *)d;
1982 ip6 = ip6cp->ip6c_ip6;
1983 off = ip6cp->ip6c_off;
1984 sa6_src = ip6cp->ip6c_src;
1988 off = 0; /* fool gcc */
1993 struct in_conninfo inc;
1995 * XXX: We assume that when IPV6 is non NULL,
1996 * M and OFF are valid.
1999 /* check if we can safely examine src and dst ports */
2000 if (m->m_pkthdr.len < off + sizeof(*thp))
2003 bzero(&th, sizeof(th));
2004 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2006 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
2007 (struct sockaddr *)ip6cp->ip6c_src,
2008 th.th_sport, cmd, NULL, notify);
2010 bzero(&inc, sizeof(inc));
2011 inc.inc_fport = th.th_dport;
2012 inc.inc_lport = th.th_sport;
2013 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
2014 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
2015 inc.inc_flags |= INC_ISIPV6;
2016 INP_INFO_RLOCK(&V_tcbinfo);
2017 syncache_unreach(&inc, &th);
2018 INP_INFO_RUNLOCK(&V_tcbinfo);
2020 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
2021 0, cmd, NULL, notify);
2027 * Following is where TCP initial sequence number generation occurs.
2029 * There are two places where we must use initial sequence numbers:
2030 * 1. In SYN-ACK packets.
2031 * 2. In SYN packets.
2033 * All ISNs for SYN-ACK packets are generated by the syncache. See
2034 * tcp_syncache.c for details.
2036 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2037 * depends on this property. In addition, these ISNs should be
2038 * unguessable so as to prevent connection hijacking. To satisfy
2039 * the requirements of this situation, the algorithm outlined in
2040 * RFC 1948 is used, with only small modifications.
2042 * Implementation details:
2044 * Time is based off the system timer, and is corrected so that it
2045 * increases by one megabyte per second. This allows for proper
2046 * recycling on high speed LANs while still leaving over an hour
2049 * As reading the *exact* system time is too expensive to be done
2050 * whenever setting up a TCP connection, we increment the time
2051 * offset in two ways. First, a small random positive increment
2052 * is added to isn_offset for each connection that is set up.
2053 * Second, the function tcp_isn_tick fires once per clock tick
2054 * and increments isn_offset as necessary so that sequence numbers
2055 * are incremented at approximately ISN_BYTES_PER_SECOND. The
2056 * random positive increments serve only to ensure that the same
2057 * exact sequence number is never sent out twice (as could otherwise
2058 * happen when a port is recycled in less than the system tick
2061 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2062 * between seeding of isn_secret. This is normally set to zero,
2063 * as reseeding should not be necessary.
2065 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2066 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
2067 * general, this means holding an exclusive (write) lock.
2070 #define ISN_BYTES_PER_SECOND 1048576
2071 #define ISN_STATIC_INCREMENT 4096
2072 #define ISN_RANDOM_INCREMENT (4096 - 1)
2074 static VNET_DEFINE(u_char, isn_secret[32]);
2075 static VNET_DEFINE(int, isn_last);
2076 static VNET_DEFINE(int, isn_last_reseed);
2077 static VNET_DEFINE(u_int32_t, isn_offset);
2078 static VNET_DEFINE(u_int32_t, isn_offset_old);
2080 #define V_isn_secret VNET(isn_secret)
2081 #define V_isn_last VNET(isn_last)
2082 #define V_isn_last_reseed VNET(isn_last_reseed)
2083 #define V_isn_offset VNET(isn_offset)
2084 #define V_isn_offset_old VNET(isn_offset_old)
2087 tcp_new_isn(struct tcpcb *tp)
2090 u_int32_t md5_buffer[4];
2092 u_int32_t projected_offset;
2094 INP_WLOCK_ASSERT(tp->t_inpcb);
2097 /* Seed if this is the first use, reseed if requested. */
2098 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2099 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2101 read_random(&V_isn_secret, sizeof(V_isn_secret));
2102 V_isn_last_reseed = ticks;
2105 /* Compute the md5 hash and return the ISN. */
2107 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2108 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2110 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2111 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2112 sizeof(struct in6_addr));
2113 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2114 sizeof(struct in6_addr));
2118 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2119 sizeof(struct in_addr));
2120 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2121 sizeof(struct in_addr));
2123 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2124 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2125 new_isn = (tcp_seq) md5_buffer[0];
2126 V_isn_offset += ISN_STATIC_INCREMENT +
2127 (arc4random() & ISN_RANDOM_INCREMENT);
2128 if (ticks != V_isn_last) {
2129 projected_offset = V_isn_offset_old +
2130 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2131 if (SEQ_GT(projected_offset, V_isn_offset))
2132 V_isn_offset = projected_offset;
2133 V_isn_offset_old = V_isn_offset;
2136 new_isn += V_isn_offset;
2142 * When a specific ICMP unreachable message is received and the
2143 * connection state is SYN-SENT, drop the connection. This behavior
2144 * is controlled by the icmp_may_rst sysctl.
2147 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2151 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2152 INP_WLOCK_ASSERT(inp);
2154 if ((inp->inp_flags & INP_TIMEWAIT) ||
2155 (inp->inp_flags & INP_DROPPED))
2158 tp = intotcpcb(inp);
2159 if (tp->t_state != TCPS_SYN_SENT)
2162 tp = tcp_drop(tp, errno);
2170 * When `need fragmentation' ICMP is received, update our idea of the MSS
2171 * based on the new value. Also nudge TCP to send something, since we
2172 * know the packet we just sent was dropped.
2173 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2175 static struct inpcb *
2176 tcp_mtudisc_notify(struct inpcb *inp, int error)
2179 tcp_mtudisc(inp, -1);
2184 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2189 INP_WLOCK_ASSERT(inp);
2190 if ((inp->inp_flags & INP_TIMEWAIT) ||
2191 (inp->inp_flags & INP_DROPPED))
2194 tp = intotcpcb(inp);
2195 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2197 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2199 so = inp->inp_socket;
2200 SOCKBUF_LOCK(&so->so_snd);
2201 /* If the mss is larger than the socket buffer, decrease the mss. */
2202 if (so->so_snd.sb_hiwat < tp->t_maxseg)
2203 tp->t_maxseg = so->so_snd.sb_hiwat;
2204 SOCKBUF_UNLOCK(&so->so_snd);
2206 TCPSTAT_INC(tcps_mturesent);
2208 tp->snd_nxt = tp->snd_una;
2209 tcp_free_sackholes(tp);
2210 tp->snd_recover = tp->snd_max;
2211 if (tp->t_flags & TF_SACK_PERMIT)
2212 EXIT_FASTRECOVERY(tp->t_flags);
2213 tp->t_fb->tfb_tcp_output(tp);
2218 * Look-up the routing entry to the peer of this inpcb. If no route
2219 * is found and it cannot be allocated, then return 0. This routine
2220 * is called by TCP routines that access the rmx structure and by
2221 * tcp_mss_update to get the peer/interface MTU.
2224 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2226 struct nhop4_extended nh4;
2230 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2232 if (inc->inc_faddr.s_addr != INADDR_ANY) {
2234 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2235 NHR_REF, 0, &nh4) != 0)
2239 maxmtu = nh4.nh_mtu;
2241 /* Report additional interface capabilities. */
2243 if (ifp->if_capenable & IFCAP_TSO4 &&
2244 ifp->if_hwassist & CSUM_TSO) {
2245 cap->ifcap |= CSUM_TSO;
2246 cap->tsomax = ifp->if_hw_tsomax;
2247 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2248 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2251 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2259 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2261 struct nhop6_extended nh6;
2262 struct in6_addr dst6;
2267 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2269 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2270 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2271 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2276 maxmtu = nh6.nh_mtu;
2278 /* Report additional interface capabilities. */
2280 if (ifp->if_capenable & IFCAP_TSO6 &&
2281 ifp->if_hwassist & CSUM_TSO) {
2282 cap->ifcap |= CSUM_TSO;
2283 cap->tsomax = ifp->if_hw_tsomax;
2284 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2285 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2288 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2296 * Calculate effective SMSS per RFC5681 definition for a given TCP
2297 * connection at its current state, taking into account SACK and etc.
2300 tcp_maxseg(const struct tcpcb *tp)
2304 if (tp->t_flags & TF_NOOPT)
2305 return (tp->t_maxseg);
2308 * Here we have a simplified code from tcp_addoptions(),
2309 * without a proper loop, and having most of paddings hardcoded.
2310 * We might make mistakes with padding here in some edge cases,
2311 * but this is harmless, since result of tcp_maxseg() is used
2312 * only in cwnd and ssthresh estimations.
2314 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4)
2315 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2316 if (tp->t_flags & TF_RCVD_TSTMP)
2317 optlen = TCPOLEN_TSTAMP_APPA;
2320 #ifdef TCP_SIGNATURE
2321 if (tp->t_flags & TF_SIGNATURE)
2322 optlen += PAD(TCPOLEN_SIGNATURE);
2324 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2325 optlen += TCPOLEN_SACKHDR;
2326 optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2327 optlen = PAD(optlen);
2330 if (tp->t_flags & TF_REQ_TSTMP)
2331 optlen = TCPOLEN_TSTAMP_APPA;
2333 optlen = PAD(TCPOLEN_MAXSEG);
2334 if (tp->t_flags & TF_REQ_SCALE)
2335 optlen += PAD(TCPOLEN_WINDOW);
2336 #ifdef TCP_SIGNATURE
2337 if (tp->t_flags & TF_SIGNATURE)
2338 optlen += PAD(TCPOLEN_SIGNATURE);
2340 if (tp->t_flags & TF_SACK_PERMIT)
2341 optlen += PAD(TCPOLEN_SACK_PERMITTED);
2344 optlen = min(optlen, TCP_MAXOLEN);
2345 return (tp->t_maxseg - optlen);
2349 /* compute ESP/AH header size for TCP, including outer IP header. */
2351 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2358 struct ip6_hdr *ip6;
2362 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2363 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2365 m = m_gethdr(M_NOWAIT, MT_DATA);
2370 if ((inp->inp_vflag & INP_IPV6) != 0) {
2371 ip6 = mtod(m, struct ip6_hdr *);
2372 th = (struct tcphdr *)(ip6 + 1);
2373 m->m_pkthdr.len = m->m_len =
2374 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2375 tcpip_fillheaders(inp, ip6, th);
2376 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2380 ip = mtod(m, struct ip *);
2381 th = (struct tcphdr *)(ip + 1);
2382 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2383 tcpip_fillheaders(inp, ip, th);
2384 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2392 #ifdef TCP_SIGNATURE
2394 * Callback function invoked by m_apply() to digest TCP segment data
2395 * contained within an mbuf chain.
2398 tcp_signature_apply(void *fstate, void *data, u_int len)
2401 MD5Update(fstate, (u_char *)data, len);
2406 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2407 * search with the destination IP address, and a 'magic SPI' to be
2408 * determined by the application. This is hardcoded elsewhere to 1179
2411 tcp_get_sav(struct mbuf *m, u_int direction)
2413 union sockaddr_union dst;
2414 struct secasvar *sav;
2417 struct ip6_hdr *ip6;
2418 char ip6buf[INET6_ADDRSTRLEN];
2421 /* Extract the destination from the IP header in the mbuf. */
2422 bzero(&dst, sizeof(union sockaddr_union));
2423 ip = mtod(m, struct ip *);
2425 ip6 = NULL; /* Make the compiler happy. */
2430 dst.sa.sa_len = sizeof(struct sockaddr_in);
2431 dst.sa.sa_family = AF_INET;
2432 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2433 ip->ip_src : ip->ip_dst;
2437 case (IPV6_VERSION >> 4):
2438 ip6 = mtod(m, struct ip6_hdr *);
2439 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2440 dst.sa.sa_family = AF_INET6;
2441 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2442 ip6->ip6_src : ip6->ip6_dst;
2451 /* Look up an SADB entry which matches the address of the peer. */
2452 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2454 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2455 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2457 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2458 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2467 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2470 * m pointer to head of mbuf chain
2471 * len length of TCP segment data, excluding options
2472 * optlen length of TCP segment options
2473 * buf pointer to storage for computed MD5 digest
2474 * sav pointer to security assosiation
2476 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2477 * When called from tcp_input(), we can be sure that th_sum has been
2478 * zeroed out and verified already.
2480 * Releases reference to SADB key before return.
2482 * Return 0 if successful, otherwise return -1.
2486 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2487 u_char *buf, struct secasvar *sav)
2490 struct ippseudo ippseudo;
2496 struct ipovly *ipovly;
2500 struct ip6_hdr *ip6;
2501 struct in6_addr in6;
2507 KASSERT(m != NULL, ("NULL mbuf chain"));
2508 KASSERT(buf != NULL, ("NULL signature pointer"));
2510 /* Extract the destination from the IP header in the mbuf. */
2511 ip = mtod(m, struct ip *);
2513 ip6 = NULL; /* Make the compiler happy. */
2518 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2520 * XXX The ippseudo header MUST be digested in network byte order,
2521 * or else we'll fail the regression test. Assume all fields we've
2522 * been doing arithmetic on have been in host byte order.
2523 * XXX One cannot depend on ipovly->ih_len here. When called from
2524 * tcp_output(), the underlying ip_len member has not yet been set.
2529 ipovly = (struct ipovly *)ip;
2530 ippseudo.ippseudo_src = ipovly->ih_src;
2531 ippseudo.ippseudo_dst = ipovly->ih_dst;
2532 ippseudo.ippseudo_pad = 0;
2533 ippseudo.ippseudo_p = IPPROTO_TCP;
2534 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2536 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2538 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2539 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2544 * RFC 2385, 2.0 Proposal
2545 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2546 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2547 * extended next header value (to form 32 bits), and 32-bit segment
2549 * Note: Upper-Layer Packet Length comes before Next Header.
2551 case (IPV6_VERSION >> 4):
2553 in6_clearscope(&in6);
2554 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2556 in6_clearscope(&in6);
2557 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2558 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2559 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2561 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2562 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2563 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2565 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2567 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2568 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2580 * Step 2: Update MD5 hash with TCP header, excluding options.
2581 * The TCP checksum must be set to zero.
2583 savecsum = th->th_sum;
2585 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2586 th->th_sum = savecsum;
2589 * Step 3: Update MD5 hash with TCP segment data.
2590 * Use m_apply() to avoid an early m_pullup().
2593 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2596 * Step 4: Update MD5 hash with shared secret.
2598 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2599 MD5Final(buf, &ctx);
2601 key_sa_recordxfer(sav, m);
2607 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2609 * Return 0 if successful, otherwise return -1.
2612 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2613 u_char *buf, u_int direction)
2615 struct secasvar *sav;
2617 if ((sav = tcp_get_sav(m, direction)) == NULL)
2620 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2624 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2627 * m pointer to head of mbuf chain
2628 * len length of TCP segment data, excluding options
2629 * optlen length of TCP segment options
2630 * buf pointer to storage for computed MD5 digest
2631 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2633 * Return 1 if successful, otherwise return 0.
2636 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2637 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2639 char tmpdigest[TCP_SIGLEN];
2641 if (tcp_sig_checksigs == 0)
2643 if ((tcpbflag & TF_SIGNATURE) == 0) {
2644 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2647 * If this socket is not expecting signature but
2648 * the segment contains signature just fail.
2650 TCPSTAT_INC(tcps_sig_err_sigopt);
2651 TCPSTAT_INC(tcps_sig_rcvbadsig);
2655 /* Signature is not expected, and not present in segment. */
2660 * If this socket is expecting signature but the segment does not
2661 * contain any just fail.
2663 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2664 TCPSTAT_INC(tcps_sig_err_nosigopt);
2665 TCPSTAT_INC(tcps_sig_rcvbadsig);
2668 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2669 IPSEC_DIR_INBOUND) == -1) {
2670 TCPSTAT_INC(tcps_sig_err_buildsig);
2671 TCPSTAT_INC(tcps_sig_rcvbadsig);
2675 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2676 TCPSTAT_INC(tcps_sig_rcvbadsig);
2679 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2682 #endif /* TCP_SIGNATURE */
2685 sysctl_drop(SYSCTL_HANDLER_ARGS)
2687 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2688 struct sockaddr_storage addrs[2];
2692 struct sockaddr_in *fin, *lin;
2694 struct sockaddr_in6 *fin6, *lin6;
2705 if (req->oldptr != NULL || req->oldlen != 0)
2707 if (req->newptr == NULL)
2709 if (req->newlen < sizeof(addrs))
2711 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2715 switch (addrs[0].ss_family) {
2718 fin6 = (struct sockaddr_in6 *)&addrs[0];
2719 lin6 = (struct sockaddr_in6 *)&addrs[1];
2720 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2721 lin6->sin6_len != sizeof(struct sockaddr_in6))
2723 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2724 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2726 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2727 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2728 fin = (struct sockaddr_in *)&addrs[0];
2729 lin = (struct sockaddr_in *)&addrs[1];
2732 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2735 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2742 fin = (struct sockaddr_in *)&addrs[0];
2743 lin = (struct sockaddr_in *)&addrs[1];
2744 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2745 lin->sin_len != sizeof(struct sockaddr_in))
2752 INP_INFO_RLOCK(&V_tcbinfo);
2753 switch (addrs[0].ss_family) {
2756 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2757 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2758 INPLOOKUP_WLOCKPCB, NULL);
2763 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2764 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2769 if (inp->inp_flags & INP_TIMEWAIT) {
2771 * XXXRW: There currently exists a state where an
2772 * inpcb is present, but its timewait state has been
2773 * discarded. For now, don't allow dropping of this
2781 } else if (!(inp->inp_flags & INP_DROPPED) &&
2782 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2783 tp = intotcpcb(inp);
2784 tp = tcp_drop(tp, ECONNABORTED);
2791 INP_INFO_RUNLOCK(&V_tcbinfo);
2795 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2796 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2797 0, sysctl_drop, "", "Drop TCP connection");
2800 * Generate a standardized TCP log line for use throughout the
2801 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2802 * allow use in the interrupt context.
2804 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2805 * NB: The function may return NULL if memory allocation failed.
2807 * Due to header inclusion and ordering limitations the struct ip
2808 * and ip6_hdr pointers have to be passed as void pointers.
2811 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2815 /* Is logging enabled? */
2816 if (tcp_log_in_vain == 0)
2819 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2823 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2827 /* Is logging enabled? */
2828 if (tcp_log_debug == 0)
2831 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2835 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2842 const struct ip6_hdr *ip6;
2844 ip6 = (const struct ip6_hdr *)ip6hdr;
2846 ip = (struct ip *)ip4hdr;
2849 * The log line looks like this:
2850 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2852 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2853 sizeof(PRINT_TH_FLAGS) + 1 +
2855 2 * INET6_ADDRSTRLEN;
2857 2 * INET_ADDRSTRLEN;
2860 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2864 strcat(s, "TCP: [");
2867 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2868 inet_ntoa_r(inc->inc_faddr, sp);
2870 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2872 inet_ntoa_r(inc->inc_laddr, sp);
2874 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2877 ip6_sprintf(sp, &inc->inc6_faddr);
2879 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2881 ip6_sprintf(sp, &inc->inc6_laddr);
2883 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2884 } else if (ip6 && th) {
2885 ip6_sprintf(sp, &ip6->ip6_src);
2887 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2889 ip6_sprintf(sp, &ip6->ip6_dst);
2891 sprintf(sp, "]:%i", ntohs(th->th_dport));
2894 } else if (ip && th) {
2895 inet_ntoa_r(ip->ip_src, sp);
2897 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2899 inet_ntoa_r(ip->ip_dst, sp);
2901 sprintf(sp, "]:%i", ntohs(th->th_dport));
2909 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2910 if (*(s + size - 1) != '\0')
2911 panic("%s: string too long", __func__);
2916 * A subroutine which makes it easy to track TCP state changes with DTrace.
2917 * This function shouldn't be called for t_state initializations that don't
2918 * correspond to actual TCP state transitions.
2921 tcp_state_change(struct tcpcb *tp, int newstate)
2923 #if defined(KDTRACE_HOOKS)
2924 int pstate = tp->t_state;
2927 TCPSTAT_DEC(tcps_states[tp->t_state]);
2928 TCPSTAT_INC(tcps_states[newstate]);
2929 tp->t_state = newstate;
2930 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);