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
33 #include "opt_compat.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_sack.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
50 #include <sys/domain.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
60 #include <net/route.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
67 #include <netinet/ip6.h>
69 #include <netinet/in_pcb.h>
71 #include <netinet6/in6_pcb.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
80 #include <netinet/ip_icmp.h>
81 #include <netinet/tcp.h>
82 #include <netinet/tcp_fsm.h>
83 #include <netinet/tcp_seq.h>
84 #include <netinet/tcp_timer.h>
85 #include <netinet/tcp_var.h>
87 #include <netinet6/tcp6_var.h>
89 #include <netinet/tcpip.h>
91 #include <netinet/tcp_debug.h>
93 #include <netinet6/ip6protosw.h>
96 #include <netinet6/ipsec.h>
98 #include <netinet6/ipsec6.h>
100 #include <netkey/key.h>
104 #include <netipsec/ipsec.h>
105 #include <netipsec/xform.h>
107 #include <netipsec/ipsec6.h>
109 #include <netipsec/key.h>
111 #endif /*FAST_IPSEC*/
113 #include <machine/in_cksum.h>
116 int tcp_mssdflt = TCP_MSS;
117 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
121 int tcp_v6mssdflt = TCP6_MSS;
122 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
123 CTLFLAG_RW, &tcp_v6mssdflt , 0,
124 "Default TCP Maximum Segment Size for IPv6");
128 * Minimum MSS we accept and use. This prevents DoS attacks where
129 * we are forced to a ridiculous low MSS like 20 and send hundreds
130 * of packets instead of one. The effect scales with the available
131 * bandwidth and quickly saturates the CPU and network interface
132 * with packet generation and sending. Set to zero to disable MINMSS
133 * checking. This setting prevents us from sending too small packets.
135 int tcp_minmss = TCP_MINMSS;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
139 * Number of TCP segments per second we accept from remote host
140 * before we start to calculate average segment size. If average
141 * segment size drops below the minimum TCP MSS we assume a DoS
142 * attack and reset+drop the connection. Care has to be taken not to
143 * set this value too small to not kill interactive type connections
144 * (telnet, SSH) which send many small packets.
146 int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
149 "be under the MINMSS Size");
152 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
153 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
157 int tcp_do_rfc1323 = 1;
158 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
161 static int tcp_tcbhashsize = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
165 static int do_tcpdrain = 1;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
167 "Enable tcp_drain routine for extra help when low on mbufs");
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
170 &tcbinfo.ipi_count, 0, "Number of active PCBs");
172 static int icmp_may_rst = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
174 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
176 static int tcp_isn_reseed_interval = 0;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, maxtcptw, CTLFLAG_RDTUN,
182 &maxtcptw, 0, "Maximum number of compressed TCP TIME_WAIT entries");
185 * TCP bandwidth limiting sysctls. Note that the default lower bound of
186 * 1024 exists only for debugging. A good production default would be
187 * something like 6100.
189 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
190 "TCP inflight data limiting");
192 static int tcp_inflight_enable = 1;
193 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
194 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
196 static int tcp_inflight_debug = 0;
197 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
198 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
200 static int tcp_inflight_rttthresh;
201 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
202 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
203 "RTT threshold below which inflight will deactivate itself");
205 static int tcp_inflight_min = 6144;
206 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
207 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
209 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
210 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
211 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
213 static int tcp_inflight_stab = 20;
214 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
215 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
217 uma_zone_t sack_hole_zone;
219 static struct inpcb *tcp_notify(struct inpcb *, int);
220 static void tcp_isn_tick(void *);
223 * Target size of TCP PCB hash tables. Must be a power of two.
225 * Note that this can be overridden by the kernel environment
226 * variable net.inet.tcp.tcbhashsize
229 #define TCBHASHSIZE 512
234 * Callouts should be moved into struct tcp directly. They are currently
235 * separate because the tcpcb structure is exported to userland for sysctl
236 * parsing purposes, which do not know about callouts.
240 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
241 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
244 static uma_zone_t tcpcb_zone;
245 static uma_zone_t tcptw_zone;
246 struct callout isn_callout;
247 static struct mtx isn_mtx;
249 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
250 #define ISN_LOCK() mtx_lock(&isn_mtx)
251 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
254 * TCP initialization.
257 tcp_zone_change(void *tag)
260 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
261 uma_zone_set_max(tcpcb_zone, maxsockets);
262 uma_zone_set_max(tcptw_zone, maxsockets / 5);
266 tcp_inpcb_init(void *mem, int size, int flags)
268 struct inpcb *inp = (struct inpcb *) mem;
269 INP_LOCK_INIT(inp, "inp", "tcpinp");
276 int hashsize = TCBHASHSIZE;
278 tcp_delacktime = TCPTV_DELACK;
279 tcp_keepinit = TCPTV_KEEP_INIT;
280 tcp_keepidle = TCPTV_KEEP_IDLE;
281 tcp_keepintvl = TCPTV_KEEPINTVL;
282 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
284 tcp_rexmit_min = TCPTV_MIN;
285 tcp_rexmit_slop = TCPTV_CPU_VAR;
286 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
288 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
290 tcbinfo.listhead = &tcb;
291 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
292 if (!powerof2(hashsize)) {
293 printf("WARNING: TCB hash size not a power of 2\n");
294 hashsize = 512; /* safe default */
296 tcp_tcbhashsize = hashsize;
297 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
298 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
299 &tcbinfo.porthashmask);
300 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
301 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
302 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
304 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
306 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
308 if (max_protohdr < TCP_MINPROTOHDR)
309 max_protohdr = TCP_MINPROTOHDR;
310 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
312 #undef TCP_MINPROTOHDR
314 * These have to be type stable for the benefit of the timers.
316 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
317 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
318 uma_zone_set_max(tcpcb_zone, maxsockets);
319 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
321 maxtcptw = maxsockets / 5;
322 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
323 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
324 uma_zone_set_max(tcptw_zone, maxtcptw);
330 callout_init(&isn_callout, CALLOUT_MPSAFE);
332 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
333 SHUTDOWN_PRI_DEFAULT);
334 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
335 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
336 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
337 EVENTHANDLER_PRI_ANY);
344 callout_stop(&isn_callout);
348 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
349 * tcp_template used to store this data in mbufs, but we now recopy it out
350 * of the tcpcb each time to conserve mbufs.
353 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
355 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
357 INP_LOCK_ASSERT(inp);
360 if ((inp->inp_vflag & INP_IPV6) != 0) {
363 ip6 = (struct ip6_hdr *)ip_ptr;
364 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
365 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
366 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
367 (IPV6_VERSION & IPV6_VERSION_MASK);
368 ip6->ip6_nxt = IPPROTO_TCP;
369 ip6->ip6_plen = sizeof(struct tcphdr);
370 ip6->ip6_src = inp->in6p_laddr;
371 ip6->ip6_dst = inp->in6p_faddr;
377 ip = (struct ip *)ip_ptr;
378 ip->ip_v = IPVERSION;
380 ip->ip_tos = inp->inp_ip_tos;
384 ip->ip_ttl = inp->inp_ip_ttl;
386 ip->ip_p = IPPROTO_TCP;
387 ip->ip_src = inp->inp_laddr;
388 ip->ip_dst = inp->inp_faddr;
390 th->th_sport = inp->inp_lport;
391 th->th_dport = inp->inp_fport;
399 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
403 * Create template to be used to send tcp packets on a connection.
404 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
405 * use for this function is in keepalives, which use tcp_respond.
408 tcpip_maketemplate(struct inpcb *inp)
413 m = m_get(M_DONTWAIT, MT_DATA);
416 m->m_len = sizeof(struct tcptemp);
417 n = mtod(m, struct tcptemp *);
419 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
424 * Send a single message to the TCP at address specified by
425 * the given TCP/IP header. If m == NULL, then we make a copy
426 * of the tcpiphdr at ti and send directly to the addressed host.
427 * This is used to force keep alive messages out using the TCP
428 * template for a connection. If flags are given then we send
429 * a message back to the TCP which originated the * segment ti,
430 * and discard the mbuf containing it and any other attached mbufs.
432 * In any case the ack and sequence number of the transmitted
433 * segment are as specified by the parameters.
435 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
438 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
439 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
452 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
455 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
462 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
463 INP_INFO_WLOCK_ASSERT(&tcbinfo);
464 INP_LOCK_ASSERT(inp);
469 if (!(flags & TH_RST)) {
470 win = sbspace(&inp->inp_socket->so_rcv);
471 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
472 win = (long)TCP_MAXWIN << tp->rcv_scale;
476 m = m_gethdr(M_DONTWAIT, MT_DATA);
480 m->m_data += max_linkhdr;
483 bcopy((caddr_t)ip6, mtod(m, caddr_t),
484 sizeof(struct ip6_hdr));
485 ip6 = mtod(m, struct ip6_hdr *);
486 nth = (struct tcphdr *)(ip6 + 1);
490 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
491 ip = mtod(m, struct ip *);
492 nth = (struct tcphdr *)(ip + 1);
494 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
499 m->m_data = (caddr_t)ipgen;
500 /* m_len is set later */
502 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
505 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
506 nth = (struct tcphdr *)(ip6 + 1);
510 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
511 nth = (struct tcphdr *)(ip + 1);
515 * this is usually a case when an extension header
516 * exists between the IPv6 header and the
519 nth->th_sport = th->th_sport;
520 nth->th_dport = th->th_dport;
522 xchg(nth->th_dport, nth->th_sport, n_short);
528 ip6->ip6_vfc = IPV6_VERSION;
529 ip6->ip6_nxt = IPPROTO_TCP;
530 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
532 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
536 tlen += sizeof (struct tcpiphdr);
538 ip->ip_ttl = ip_defttl;
539 if (path_mtu_discovery)
543 m->m_pkthdr.len = tlen;
544 m->m_pkthdr.rcvif = NULL;
548 * Packet is associated with a socket, so allow the
549 * label of the response to reflect the socket label.
551 INP_LOCK_ASSERT(inp);
552 mac_create_mbuf_from_inpcb(inp, m);
555 * Packet is not associated with a socket, so possibly
556 * update the label in place.
558 mac_reflect_mbuf_tcp(m);
561 nth->th_seq = htonl(seq);
562 nth->th_ack = htonl(ack);
564 nth->th_off = sizeof (struct tcphdr) >> 2;
565 nth->th_flags = flags;
567 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
569 nth->th_win = htons((u_short)win);
574 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
575 sizeof(struct ip6_hdr),
576 tlen - sizeof(struct ip6_hdr));
577 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
582 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
583 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
584 m->m_pkthdr.csum_flags = CSUM_TCP;
585 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
588 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
589 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
593 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
596 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
600 * Create a new TCP control block, making an
601 * empty reassembly queue and hooking it to the argument
602 * protocol control block. The `inp' parameter must have
603 * come from the zone allocator set up in tcp_init().
606 tcp_newtcpcb(struct inpcb *inp)
608 struct tcpcb_mem *tm;
611 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
614 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
618 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
619 tp->t_maxseg = tp->t_maxopd =
621 isipv6 ? tcp_v6mssdflt :
625 /* Set up our timeouts. */
626 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
627 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
628 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
629 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
630 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
633 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
634 tp->sack_enable = tcp_do_sack;
635 TAILQ_INIT(&tp->snd_holes);
636 tp->t_inpcb = inp; /* XXX */
638 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
639 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
640 * reasonable initial retransmit time.
642 tp->t_srtt = TCPTV_SRTTBASE;
643 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
644 tp->t_rttmin = tcp_rexmit_min;
645 tp->t_rxtcur = TCPTV_RTOBASE;
646 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
647 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
648 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
649 tp->t_rcvtime = ticks;
650 tp->t_bw_rtttime = ticks;
652 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
653 * because the socket may be bound to an IPv6 wildcard address,
654 * which may match an IPv4-mapped IPv6 address.
656 inp->inp_ip_ttl = ip_defttl;
658 return (tp); /* XXX */
662 * Drop a TCP connection, reporting
663 * the specified error. If connection is synchronized,
664 * then send a RST to peer.
667 tcp_drop(struct tcpcb *tp, int errno)
669 struct socket *so = tp->t_inpcb->inp_socket;
671 INP_INFO_WLOCK_ASSERT(&tcbinfo);
672 INP_LOCK_ASSERT(tp->t_inpcb);
674 if (TCPS_HAVERCVDSYN(tp->t_state)) {
675 tp->t_state = TCPS_CLOSED;
676 (void) tcp_output(tp);
677 tcpstat.tcps_drops++;
679 tcpstat.tcps_conndrops++;
680 if (errno == ETIMEDOUT && tp->t_softerror)
681 errno = tp->t_softerror;
682 so->so_error = errno;
683 return (tcp_close(tp));
687 tcp_discardcb(struct tcpcb *tp)
690 struct inpcb *inp = tp->t_inpcb;
691 struct socket *so = inp->inp_socket;
693 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
696 INP_LOCK_ASSERT(inp);
699 * Make sure that all of our timers are stopped before we
702 callout_stop(tp->tt_rexmt);
703 callout_stop(tp->tt_persist);
704 callout_stop(tp->tt_keep);
705 callout_stop(tp->tt_2msl);
706 callout_stop(tp->tt_delack);
709 * If we got enough samples through the srtt filter,
710 * save the rtt and rttvar in the routing entry.
711 * 'Enough' is arbitrarily defined as 4 rtt samples.
712 * 4 samples is enough for the srtt filter to converge
713 * to within enough % of the correct value; fewer samples
714 * and we could save a bogus rtt. The danger is not high
715 * as tcp quickly recovers from everything.
716 * XXX: Works very well but needs some more statistics!
718 if (tp->t_rttupdated >= 4) {
719 struct hc_metrics_lite metrics;
722 bzero(&metrics, sizeof(metrics));
724 * Update the ssthresh always when the conditions below
725 * are satisfied. This gives us better new start value
726 * for the congestion avoidance for new connections.
727 * ssthresh is only set if packet loss occured on a session.
729 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
730 * being torn down. Ideally this code would not use 'so'.
732 ssthresh = tp->snd_ssthresh;
733 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
735 * convert the limit from user data bytes to
736 * packets then to packet data bytes.
738 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
741 ssthresh *= (u_long)(tp->t_maxseg +
743 (isipv6 ? sizeof (struct ip6_hdr) +
744 sizeof (struct tcphdr) :
746 sizeof (struct tcpiphdr)
753 metrics.rmx_ssthresh = ssthresh;
755 metrics.rmx_rtt = tp->t_srtt;
756 metrics.rmx_rttvar = tp->t_rttvar;
757 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
758 metrics.rmx_bandwidth = tp->snd_bandwidth;
759 metrics.rmx_cwnd = tp->snd_cwnd;
760 metrics.rmx_sendpipe = 0;
761 metrics.rmx_recvpipe = 0;
763 tcp_hc_update(&inp->inp_inc, &metrics);
766 /* free the reassembly queue, if any */
767 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
768 LIST_REMOVE(q, tqe_q);
770 uma_zfree(tcp_reass_zone, q);
774 tcp_free_sackholes(tp);
775 inp->inp_ppcb = NULL;
777 uma_zfree(tcpcb_zone, tp);
781 * Attempt to close a TCP control block, marking it as dropped, and freeing
782 * the socket if we hold the only reference.
785 tcp_close(struct tcpcb *tp)
787 struct inpcb *inp = tp->t_inpcb;
790 INP_INFO_WLOCK_ASSERT(&tcbinfo);
791 INP_LOCK_ASSERT(inp);
794 tcpstat.tcps_closed++;
795 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
796 so = inp->inp_socket;
797 soisdisconnected(so);
798 if (inp->inp_vflag & INP_SOCKREF) {
799 KASSERT(so->so_state & SS_PROTOREF,
800 ("tcp_close: !SS_PROTOREF"));
801 inp->inp_vflag &= ~INP_SOCKREF;
805 so->so_state &= ~SS_PROTOREF;
819 struct tseg_qent *te;
822 * Walk the tcpbs, if existing, and flush the reassembly queue,
824 * XXX: The "Net/3" implementation doesn't imply that the TCP
825 * reassembly queue should be flushed, but in a situation
826 * where we're really low on mbufs, this is potentially
829 INP_INFO_RLOCK(&tcbinfo);
830 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
831 if (inpb->inp_vflag & INP_TIMEWAIT)
834 if ((tcpb = intotcpcb(inpb)) != NULL) {
835 while ((te = LIST_FIRST(&tcpb->t_segq))
837 LIST_REMOVE(te, tqe_q);
839 uma_zfree(tcp_reass_zone, te);
843 tcp_clean_sackreport(tcpb);
847 INP_INFO_RUNLOCK(&tcbinfo);
852 * Notify a tcp user of an asynchronous error;
853 * store error as soft error, but wake up user
854 * (for now, won't do anything until can select for soft error).
856 * Do not wake up user since there currently is no mechanism for
857 * reporting soft errors (yet - a kqueue filter may be added).
859 static struct inpcb *
860 tcp_notify(struct inpcb *inp, int error)
864 INP_INFO_WLOCK_ASSERT(&tcbinfo);
865 INP_LOCK_ASSERT(inp);
867 if ((inp->inp_vflag & INP_TIMEWAIT) ||
868 (inp->inp_vflag & INP_DROPPED))
872 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
875 * Ignore some errors if we are hooked up.
876 * If connection hasn't completed, has retransmitted several times,
877 * and receives a second error, give up now. This is better
878 * than waiting a long time to establish a connection that
879 * can never complete.
881 if (tp->t_state == TCPS_ESTABLISHED &&
882 (error == EHOSTUNREACH || error == ENETUNREACH ||
883 error == EHOSTDOWN)) {
885 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
887 tp = tcp_drop(tp, error);
893 tp->t_softerror = error;
897 wakeup( &so->so_timeo);
904 tcp_pcblist(SYSCTL_HANDLER_ARGS)
907 struct inpcb *inp, **inp_list;
912 * The process of preparing the TCB list is too time-consuming and
913 * resource-intensive to repeat twice on every request.
915 if (req->oldptr == NULL) {
916 n = tcbinfo.ipi_count;
917 req->oldidx = 2 * (sizeof xig)
918 + (n + n/8) * sizeof(struct xtcpcb);
922 if (req->newptr != NULL)
926 * OK, now we're committed to doing something.
928 INP_INFO_RLOCK(&tcbinfo);
929 gencnt = tcbinfo.ipi_gencnt;
930 n = tcbinfo.ipi_count;
931 INP_INFO_RUNLOCK(&tcbinfo);
933 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
934 + n * sizeof(struct xtcpcb));
938 xig.xig_len = sizeof xig;
940 xig.xig_gen = gencnt;
941 xig.xig_sogen = so_gencnt;
942 error = SYSCTL_OUT(req, &xig, sizeof xig);
946 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
947 if (inp_list == NULL)
950 INP_INFO_RLOCK(&tcbinfo);
951 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
952 inp = LIST_NEXT(inp, inp_list)) {
954 if (inp->inp_gencnt <= gencnt) {
956 * XXX: This use of cr_cansee(), introduced with
957 * TCP state changes, is not quite right, but for
958 * now, better than nothing.
960 if (inp->inp_vflag & INP_TIMEWAIT) {
961 if (intotw(inp) != NULL)
962 error = cr_cansee(req->td->td_ucred,
963 intotw(inp)->tw_cred);
965 error = EINVAL; /* Skip this inp. */
967 error = cr_canseesocket(req->td->td_ucred,
974 INP_INFO_RUNLOCK(&tcbinfo);
978 for (i = 0; i < n; i++) {
981 if (inp->inp_gencnt <= gencnt) {
985 bzero(&xt, sizeof(xt));
986 xt.xt_len = sizeof xt;
987 /* XXX should avoid extra copy */
988 bcopy(inp, &xt.xt_inp, sizeof *inp);
989 inp_ppcb = inp->inp_ppcb;
990 if (inp_ppcb == NULL)
991 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
992 else if (inp->inp_vflag & INP_TIMEWAIT) {
993 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
994 xt.xt_tp.t_state = TCPS_TIME_WAIT;
996 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
997 if (inp->inp_socket != NULL)
998 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1000 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1001 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1003 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1005 error = SYSCTL_OUT(req, &xt, sizeof xt);
1012 * Give the user an updated idea of our state.
1013 * If the generation differs from what we told
1014 * her before, she knows that something happened
1015 * while we were processing this request, and it
1016 * might be necessary to retry.
1018 INP_INFO_RLOCK(&tcbinfo);
1019 xig.xig_gen = tcbinfo.ipi_gencnt;
1020 xig.xig_sogen = so_gencnt;
1021 xig.xig_count = tcbinfo.ipi_count;
1022 INP_INFO_RUNLOCK(&tcbinfo);
1023 error = SYSCTL_OUT(req, &xig, sizeof xig);
1025 free(inp_list, M_TEMP);
1029 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1030 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1033 tcp_getcred(SYSCTL_HANDLER_ARGS)
1036 struct sockaddr_in addrs[2];
1040 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1043 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1046 INP_INFO_RLOCK(&tcbinfo);
1047 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1048 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1054 if (inp->inp_socket == NULL) {
1058 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1061 cru2x(inp->inp_socket->so_cred, &xuc);
1065 INP_INFO_RUNLOCK(&tcbinfo);
1067 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1071 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1072 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1073 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1077 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1080 struct sockaddr_in6 addrs[2];
1082 int error, mapped = 0;
1084 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1087 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1090 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1091 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1094 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1095 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1101 INP_INFO_RLOCK(&tcbinfo);
1103 inp = in_pcblookup_hash(&tcbinfo,
1104 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1106 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1110 inp = in6_pcblookup_hash(&tcbinfo,
1111 &addrs[1].sin6_addr, addrs[1].sin6_port,
1112 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1118 if (inp->inp_socket == NULL) {
1122 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1125 cru2x(inp->inp_socket->so_cred, &xuc);
1129 INP_INFO_RUNLOCK(&tcbinfo);
1131 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1135 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1136 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1137 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1142 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1144 struct ip *ip = vip;
1146 struct in_addr faddr;
1149 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1151 struct in_conninfo inc;
1152 tcp_seq icmp_tcp_seq;
1155 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1156 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1159 if (cmd == PRC_MSGSIZE)
1160 notify = tcp_mtudisc;
1161 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1162 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1163 notify = tcp_drop_syn_sent;
1165 * Redirects don't need to be handled up here.
1167 else if (PRC_IS_REDIRECT(cmd))
1170 * Source quench is depreciated.
1172 else if (cmd == PRC_QUENCH)
1175 * Hostdead is ugly because it goes linearly through all PCBs.
1176 * XXX: We never get this from ICMP, otherwise it makes an
1177 * excellent DoS attack on machines with many connections.
1179 else if (cmd == PRC_HOSTDEAD)
1181 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1184 icp = (struct icmp *)((caddr_t)ip
1185 - offsetof(struct icmp, icmp_ip));
1186 th = (struct tcphdr *)((caddr_t)ip
1187 + (ip->ip_hl << 2));
1188 INP_INFO_WLOCK(&tcbinfo);
1189 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1190 ip->ip_src, th->th_sport, 0, NULL);
1193 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1194 !(inp->inp_vflag & INP_DROPPED) &&
1195 !(inp->inp_socket == NULL)) {
1196 icmp_tcp_seq = htonl(th->th_seq);
1197 tp = intotcpcb(inp);
1198 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1199 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1200 if (cmd == PRC_MSGSIZE) {
1203 * If we got a needfrag set the MTU
1204 * in the route to the suggested new
1205 * value (if given) and then notify.
1207 bzero(&inc, sizeof(inc));
1208 inc.inc_flags = 0; /* IPv4 */
1209 inc.inc_faddr = faddr;
1211 mtu = ntohs(icp->icmp_nextmtu);
1213 * If no alternative MTU was
1214 * proposed, try the next smaller
1215 * one. ip->ip_len has already
1216 * been swapped in icmp_input().
1219 mtu = ip_next_mtu(ip->ip_len,
1221 if (mtu < max(296, (tcp_minmss)
1222 + sizeof(struct tcpiphdr)))
1226 + sizeof(struct tcpiphdr);
1228 * Only cache the the MTU if it
1229 * is smaller than the interface
1230 * or route MTU. tcp_mtudisc()
1231 * will do right thing by itself.
1233 if (mtu <= tcp_maxmtu(&inc))
1234 tcp_hc_updatemtu(&inc, mtu);
1237 inp = (*notify)(inp, inetctlerrmap[cmd]);
1243 inc.inc_fport = th->th_dport;
1244 inc.inc_lport = th->th_sport;
1245 inc.inc_faddr = faddr;
1246 inc.inc_laddr = ip->ip_src;
1250 syncache_unreach(&inc, th);
1252 INP_INFO_WUNLOCK(&tcbinfo);
1254 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1259 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1262 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1263 struct ip6_hdr *ip6;
1265 struct ip6ctlparam *ip6cp = NULL;
1266 const struct sockaddr_in6 *sa6_src = NULL;
1268 struct tcp_portonly {
1273 if (sa->sa_family != AF_INET6 ||
1274 sa->sa_len != sizeof(struct sockaddr_in6))
1277 if (cmd == PRC_MSGSIZE)
1278 notify = tcp_mtudisc;
1279 else if (!PRC_IS_REDIRECT(cmd) &&
1280 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1282 /* Source quench is depreciated. */
1283 else if (cmd == PRC_QUENCH)
1286 /* if the parameter is from icmp6, decode it. */
1288 ip6cp = (struct ip6ctlparam *)d;
1290 ip6 = ip6cp->ip6c_ip6;
1291 off = ip6cp->ip6c_off;
1292 sa6_src = ip6cp->ip6c_src;
1296 off = 0; /* fool gcc */
1301 struct in_conninfo inc;
1303 * XXX: We assume that when IPV6 is non NULL,
1304 * M and OFF are valid.
1307 /* check if we can safely examine src and dst ports */
1308 if (m->m_pkthdr.len < off + sizeof(*thp))
1311 bzero(&th, sizeof(th));
1312 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1314 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1315 (struct sockaddr *)ip6cp->ip6c_src,
1316 th.th_sport, cmd, NULL, notify);
1318 inc.inc_fport = th.th_dport;
1319 inc.inc_lport = th.th_sport;
1320 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1321 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1323 INP_INFO_WLOCK(&tcbinfo);
1324 syncache_unreach(&inc, &th);
1325 INP_INFO_WUNLOCK(&tcbinfo);
1327 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1328 0, cmd, NULL, notify);
1334 * Following is where TCP initial sequence number generation occurs.
1336 * There are two places where we must use initial sequence numbers:
1337 * 1. In SYN-ACK packets.
1338 * 2. In SYN packets.
1340 * All ISNs for SYN-ACK packets are generated by the syncache. See
1341 * tcp_syncache.c for details.
1343 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1344 * depends on this property. In addition, these ISNs should be
1345 * unguessable so as to prevent connection hijacking. To satisfy
1346 * the requirements of this situation, the algorithm outlined in
1347 * RFC 1948 is used, with only small modifications.
1349 * Implementation details:
1351 * Time is based off the system timer, and is corrected so that it
1352 * increases by one megabyte per second. This allows for proper
1353 * recycling on high speed LANs while still leaving over an hour
1356 * As reading the *exact* system time is too expensive to be done
1357 * whenever setting up a TCP connection, we increment the time
1358 * offset in two ways. First, a small random positive increment
1359 * is added to isn_offset for each connection that is set up.
1360 * Second, the function tcp_isn_tick fires once per clock tick
1361 * and increments isn_offset as necessary so that sequence numbers
1362 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1363 * random positive increments serve only to ensure that the same
1364 * exact sequence number is never sent out twice (as could otherwise
1365 * happen when a port is recycled in less than the system tick
1368 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1369 * between seeding of isn_secret. This is normally set to zero,
1370 * as reseeding should not be necessary.
1372 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1373 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1374 * general, this means holding an exclusive (write) lock.
1377 #define ISN_BYTES_PER_SECOND 1048576
1378 #define ISN_STATIC_INCREMENT 4096
1379 #define ISN_RANDOM_INCREMENT (4096 - 1)
1381 static u_char isn_secret[32];
1382 static int isn_last_reseed;
1383 static u_int32_t isn_offset, isn_offset_old;
1384 static MD5_CTX isn_ctx;
1387 tcp_new_isn(struct tcpcb *tp)
1389 u_int32_t md5_buffer[4];
1392 INP_LOCK_ASSERT(tp->t_inpcb);
1395 /* Seed if this is the first use, reseed if requested. */
1396 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1397 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1399 read_random(&isn_secret, sizeof(isn_secret));
1400 isn_last_reseed = ticks;
1403 /* Compute the md5 hash and return the ISN. */
1405 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1406 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1408 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1409 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1410 sizeof(struct in6_addr));
1411 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1412 sizeof(struct in6_addr));
1416 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1417 sizeof(struct in_addr));
1418 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1419 sizeof(struct in_addr));
1421 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1422 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1423 new_isn = (tcp_seq) md5_buffer[0];
1424 isn_offset += ISN_STATIC_INCREMENT +
1425 (arc4random() & ISN_RANDOM_INCREMENT);
1426 new_isn += isn_offset;
1432 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1433 * to keep time flowing at a relatively constant rate. If the random
1434 * increments have already pushed us past the projected offset, do nothing.
1437 tcp_isn_tick(void *xtp)
1439 u_int32_t projected_offset;
1442 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1444 if (projected_offset > isn_offset)
1445 isn_offset = projected_offset;
1447 isn_offset_old = isn_offset;
1448 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1453 * When a specific ICMP unreachable message is received and the
1454 * connection state is SYN-SENT, drop the connection. This behavior
1455 * is controlled by the icmp_may_rst sysctl.
1458 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1462 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1463 INP_LOCK_ASSERT(inp);
1465 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1466 (inp->inp_vflag & INP_DROPPED))
1469 tp = intotcpcb(inp);
1470 if (tp->t_state != TCPS_SYN_SENT)
1473 tp = tcp_drop(tp, errno);
1481 * When `need fragmentation' ICMP is received, update our idea of the MSS
1482 * based on the new value in the route. Also nudge TCP to send something,
1483 * since we know the packet we just sent was dropped.
1484 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1487 tcp_mtudisc(struct inpcb *inp, int errno)
1490 struct socket *so = inp->inp_socket;
1498 INP_LOCK_ASSERT(inp);
1499 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1500 (inp->inp_vflag & INP_DROPPED))
1503 tp = intotcpcb(inp);
1504 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1507 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1509 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1512 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1514 tcp_maxmtu(&inp->inp_inc);
1518 maxmtu = min(maxmtu, romtu);
1520 tp->t_maxopd = tp->t_maxseg =
1522 isipv6 ? tcp_v6mssdflt :
1529 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1531 sizeof(struct tcpiphdr)
1538 * XXX - The above conditional probably violates the TCP
1539 * spec. The problem is that, since we don't know the
1540 * other end's MSS, we are supposed to use a conservative
1541 * default. But, if we do that, then MTU discovery will
1542 * never actually take place, because the conservative
1543 * default is much less than the MTUs typically seen
1544 * on the Internet today. For the moment, we'll sweep
1545 * this under the carpet.
1547 * The conservative default might not actually be a problem
1548 * if the only case this occurs is when sending an initial
1549 * SYN with options and data to a host we've never talked
1550 * to before. Then, they will reply with an MSS value which
1551 * will get recorded and the new parameters should get
1552 * recomputed. For Further Study.
1554 if (tp->t_maxopd <= mss)
1558 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1559 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1560 mss -= TCPOLEN_TSTAMP_APPA;
1561 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1563 mss &= ~(MCLBYTES-1);
1566 mss = mss / MCLBYTES * MCLBYTES;
1568 if (so->so_snd.sb_hiwat < mss)
1569 mss = so->so_snd.sb_hiwat;
1573 tcpstat.tcps_mturesent++;
1575 tp->snd_nxt = tp->snd_una;
1576 tcp_free_sackholes(tp);
1577 tp->snd_recover = tp->snd_max;
1578 if (tp->sack_enable)
1579 EXIT_FASTRECOVERY(tp);
1585 * Look-up the routing entry to the peer of this inpcb. If no route
1586 * is found and it cannot be allocated, then return NULL. This routine
1587 * is called by TCP routines that access the rmx structure and by tcp_mss
1588 * to get the interface MTU.
1591 tcp_maxmtu(struct in_conninfo *inc)
1594 struct sockaddr_in *dst;
1598 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1600 bzero(&sro, sizeof(sro));
1601 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1602 dst = (struct sockaddr_in *)&sro.ro_dst;
1603 dst->sin_family = AF_INET;
1604 dst->sin_len = sizeof(*dst);
1605 dst->sin_addr = inc->inc_faddr;
1606 rtalloc_ign(&sro, RTF_CLONING);
1608 if (sro.ro_rt != NULL) {
1609 ifp = sro.ro_rt->rt_ifp;
1610 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1611 maxmtu = ifp->if_mtu;
1613 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1621 tcp_maxmtu6(struct in_conninfo *inc)
1623 struct route_in6 sro6;
1627 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1629 bzero(&sro6, sizeof(sro6));
1630 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1631 sro6.ro_dst.sin6_family = AF_INET6;
1632 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1633 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1634 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1636 if (sro6.ro_rt != NULL) {
1637 ifp = sro6.ro_rt->rt_ifp;
1638 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1639 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1641 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1642 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1651 /* compute ESP/AH header size for TCP, including outer IP header. */
1653 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1660 struct ip6_hdr *ip6;
1664 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1666 MGETHDR(m, M_DONTWAIT, MT_DATA);
1671 if ((inp->inp_vflag & INP_IPV6) != 0) {
1672 ip6 = mtod(m, struct ip6_hdr *);
1673 th = (struct tcphdr *)(ip6 + 1);
1674 m->m_pkthdr.len = m->m_len =
1675 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1676 tcpip_fillheaders(inp, ip6, th);
1677 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1681 ip = mtod(m, struct ip *);
1682 th = (struct tcphdr *)(ip + 1);
1683 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1684 tcpip_fillheaders(inp, ip, th);
1685 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1694 * Move a TCP connection into TIME_WAIT state.
1695 * tcbinfo is locked.
1696 * inp is locked, and is unlocked before returning.
1699 tcp_twstart(struct tcpcb *tp)
1703 int tw_time, acknow;
1706 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1707 INP_LOCK_ASSERT(tp->t_inpcb);
1709 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1711 tw = tcp_timer_2msl_tw(1);
1715 INP_UNLOCK(tp->t_inpcb);
1723 * Recover last window size sent.
1725 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1728 * Set t_recent if timestamps are used on the connection.
1730 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1731 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1732 tw->t_recent = tp->ts_recent;
1736 tw->snd_nxt = tp->snd_nxt;
1737 tw->rcv_nxt = tp->rcv_nxt;
1740 tw->t_starttime = tp->t_starttime;
1745 * be used for fin-wait-2 state also, then we may need
1746 * a ts_recent from the last segment.
1748 tw_time = 2 * tcp_msl;
1749 acknow = tp->t_flags & TF_ACKNOW;
1752 * First, discard tcpcb state, which includes stopping its timers and
1753 * freeing it. tcp_discardcb() used to also release the inpcb, but
1754 * that work is now done in the caller.
1756 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
1757 * and might not be needed here any longer.
1760 so = inp->inp_socket;
1761 soisdisconnected(so);
1763 tw->tw_cred = crhold(so->so_cred);
1764 tw->tw_so_options = so->so_options;
1767 tcp_twrespond(tw, TH_ACK);
1769 inp->inp_vflag |= INP_TIMEWAIT;
1770 tcp_timer_2msl_reset(tw, tw_time, 0);
1773 * If the inpcb owns the sole reference to the socket, then we can
1774 * detach and free the socket as it is not needed in time wait.
1776 if (inp->inp_vflag & INP_SOCKREF) {
1777 KASSERT(so->so_state & SS_PROTOREF,
1778 ("tcp_twstart: !SS_PROTOREF"));
1779 inp->inp_vflag &= ~INP_SOCKREF;
1783 so->so_state &= ~SS_PROTOREF;
1790 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1791 * the actual rate is slightly higher due to the addition of
1792 * random positive increments.
1794 * Most other new OSes use semi-randomized ISN values, so we
1795 * do not need to worry about them.
1797 #define MS_ISN_BYTES_PER_SECOND 250000
1800 * Determine if the ISN we will generate has advanced beyond the last
1801 * sequence number used by the previous connection. If so, indicate
1802 * that it is safe to recycle this tw socket by returning 1.
1804 * XXXRW: This function should assert the inpcb lock as it does multiple
1805 * non-atomic reads from the tcptw, but is currently called without it from
1806 * in_pcb.c:in_pcblookup_local().
1809 tcp_twrecycleable(struct tcptw *tw)
1811 tcp_seq new_iss = tw->iss;
1812 tcp_seq new_irs = tw->irs;
1814 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1815 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1817 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1824 tcp_twclose(struct tcptw *tw, int reuse)
1830 * At this point, we are in one of two situations:
1832 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
1835 * (2) We have a socket -- if we own a reference, release it and
1836 * notify the socket layer.
1839 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1840 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1841 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1842 INP_LOCK_ASSERT(inp);
1844 tw->tw_inpcb = NULL;
1845 tcp_timer_2msl_stop(tw);
1846 inp->inp_ppcb = NULL;
1849 so = inp->inp_socket;
1852 * If there's a socket, handle two cases: first, we own a
1853 * strong reference, which we will now release, or we don't
1854 * in which case another reference exists (XXXRW: think
1855 * about this more), and we don't need to take action.
1857 if (inp->inp_vflag & INP_SOCKREF) {
1858 inp->inp_vflag &= ~INP_SOCKREF;
1862 KASSERT(so->so_state & SS_PROTOREF,
1863 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1864 so->so_state &= ~SS_PROTOREF;
1868 * If we don't own the only reference, the socket and
1869 * inpcb need to be left around to be handled by
1870 * tcp_usr_detach() later.
1876 if (inp->inp_vflag & INP_IPV6PROTO)
1882 tcpstat.tcps_closed++;
1883 crfree(tw->tw_cred);
1887 uma_zfree(tcptw_zone, tw);
1891 tcp_twrespond(struct tcptw *tw, int flags)
1893 struct inpcb *inp = tw->tw_inpcb;
1896 struct ip *ip = NULL;
1898 u_int hdrlen, optlen;
1901 struct ip6_hdr *ip6 = NULL;
1902 int isipv6 = inp->inp_inc.inc_isipv6;
1905 INP_LOCK_ASSERT(inp);
1907 m = m_gethdr(M_DONTWAIT, MT_DATA);
1910 m->m_data += max_linkhdr;
1913 mac_create_mbuf_from_inpcb(inp, m);
1918 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1919 ip6 = mtod(m, struct ip6_hdr *);
1920 th = (struct tcphdr *)(ip6 + 1);
1921 tcpip_fillheaders(inp, ip6, th);
1925 hdrlen = sizeof(struct tcpiphdr);
1926 ip = mtod(m, struct ip *);
1927 th = (struct tcphdr *)(ip + 1);
1928 tcpip_fillheaders(inp, ip, th);
1930 optp = (u_int8_t *)(th + 1);
1933 * Send a timestamp and echo-reply if both our side and our peer
1934 * have sent timestamps in our SYN's and this is not a RST.
1936 if (tw->t_recent && flags == TH_ACK) {
1937 u_int32_t *lp = (u_int32_t *)optp;
1939 /* Form timestamp option as shown in appendix A of RFC 1323. */
1940 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1941 *lp++ = htonl(ticks);
1942 *lp = htonl(tw->t_recent);
1943 optp += TCPOLEN_TSTAMP_APPA;
1946 optlen = optp - (u_int8_t *)(th + 1);
1948 m->m_len = hdrlen + optlen;
1949 m->m_pkthdr.len = m->m_len;
1951 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1953 th->th_seq = htonl(tw->snd_nxt);
1954 th->th_ack = htonl(tw->rcv_nxt);
1955 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1956 th->th_flags = flags;
1957 th->th_win = htons(tw->last_win);
1961 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1962 sizeof(struct tcphdr) + optlen);
1963 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1964 error = ip6_output(m, inp->in6p_outputopts, NULL,
1965 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1969 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1970 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1971 m->m_pkthdr.csum_flags = CSUM_TCP;
1972 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1973 ip->ip_len = m->m_pkthdr.len;
1974 if (path_mtu_discovery)
1975 ip->ip_off |= IP_DF;
1976 error = ip_output(m, inp->inp_options, NULL,
1977 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1981 tcpstat.tcps_sndacks++;
1983 tcpstat.tcps_sndctrl++;
1984 tcpstat.tcps_sndtotal++;
1989 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1991 * This code attempts to calculate the bandwidth-delay product as a
1992 * means of determining the optimal window size to maximize bandwidth,
1993 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1994 * routers. This code also does a fairly good job keeping RTTs in check
1995 * across slow links like modems. We implement an algorithm which is very
1996 * similar (but not meant to be) TCP/Vegas. The code operates on the
1997 * transmitter side of a TCP connection and so only effects the transmit
1998 * side of the connection.
2000 * BACKGROUND: TCP makes no provision for the management of buffer space
2001 * at the end points or at the intermediate routers and switches. A TCP
2002 * stream, whether using NewReno or not, will eventually buffer as
2003 * many packets as it is able and the only reason this typically works is
2004 * due to the fairly small default buffers made available for a connection
2005 * (typicaly 16K or 32K). As machines use larger windows and/or window
2006 * scaling it is now fairly easy for even a single TCP connection to blow-out
2007 * all available buffer space not only on the local interface, but on
2008 * intermediate routers and switches as well. NewReno makes a misguided
2009 * attempt to 'solve' this problem by waiting for an actual failure to occur,
2010 * then backing off, then steadily increasing the window again until another
2011 * failure occurs, ad-infinitum. This results in terrible oscillation that
2012 * is only made worse as network loads increase and the idea of intentionally
2013 * blowing out network buffers is, frankly, a terrible way to manage network
2016 * It is far better to limit the transmit window prior to the failure
2017 * condition being achieved. There are two general ways to do this: First
2018 * you can 'scan' through different transmit window sizes and locate the
2019 * point where the RTT stops increasing, indicating that you have filled the
2020 * pipe, then scan backwards until you note that RTT stops decreasing, then
2021 * repeat ad-infinitum. This method works in principle but has severe
2022 * implementation issues due to RTT variances, timer granularity, and
2023 * instability in the algorithm which can lead to many false positives and
2024 * create oscillations as well as interact badly with other TCP streams
2025 * implementing the same algorithm.
2027 * The second method is to limit the window to the bandwidth delay product
2028 * of the link. This is the method we implement. RTT variances and our
2029 * own manipulation of the congestion window, bwnd, can potentially
2030 * destabilize the algorithm. For this reason we have to stabilize the
2031 * elements used to calculate the window. We do this by using the minimum
2032 * observed RTT, the long term average of the observed bandwidth, and
2033 * by adding two segments worth of slop. It isn't perfect but it is able
2034 * to react to changing conditions and gives us a very stable basis on
2035 * which to extend the algorithm.
2038 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2044 INP_LOCK_ASSERT(tp->t_inpcb);
2047 * If inflight_enable is disabled in the middle of a tcp connection,
2048 * make sure snd_bwnd is effectively disabled.
2050 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2051 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2052 tp->snd_bandwidth = 0;
2057 * Figure out the bandwidth. Due to the tick granularity this
2058 * is a very rough number and it MUST be averaged over a fairly
2059 * long period of time. XXX we need to take into account a link
2060 * that is not using all available bandwidth, but for now our
2061 * slop will ramp us up if this case occurs and the bandwidth later
2064 * Note: if ticks rollover 'bw' may wind up negative. We must
2065 * effectively reset t_bw_rtttime for this case.
2068 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2071 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2072 (save_ticks - tp->t_bw_rtttime);
2073 tp->t_bw_rtttime = save_ticks;
2074 tp->t_bw_rtseq = ack_seq;
2075 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2077 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2079 tp->snd_bandwidth = bw;
2082 * Calculate the semi-static bandwidth delay product, plus two maximal
2083 * segments. The additional slop puts us squarely in the sweet
2084 * spot and also handles the bandwidth run-up case and stabilization.
2085 * Without the slop we could be locking ourselves into a lower
2088 * Situations Handled:
2089 * (1) Prevents over-queueing of packets on LANs, especially on
2090 * high speed LANs, allowing larger TCP buffers to be
2091 * specified, and also does a good job preventing
2092 * over-queueing of packets over choke points like modems
2093 * (at least for the transmit side).
2095 * (2) Is able to handle changing network loads (bandwidth
2096 * drops so bwnd drops, bandwidth increases so bwnd
2099 * (3) Theoretically should stabilize in the face of multiple
2100 * connections implementing the same algorithm (this may need
2103 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2104 * be adjusted with a sysctl but typically only needs to be
2105 * on very slow connections. A value no smaller then 5
2106 * should be used, but only reduce this default if you have
2109 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2110 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2113 if (tcp_inflight_debug > 0) {
2115 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2117 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2126 if ((long)bwnd < tcp_inflight_min)
2127 bwnd = tcp_inflight_min;
2128 if (bwnd > tcp_inflight_max)
2129 bwnd = tcp_inflight_max;
2130 if ((long)bwnd < tp->t_maxseg * 2)
2131 bwnd = tp->t_maxseg * 2;
2132 tp->snd_bwnd = bwnd;
2135 #ifdef TCP_SIGNATURE
2137 * Callback function invoked by m_apply() to digest TCP segment data
2138 * contained within an mbuf chain.
2141 tcp_signature_apply(void *fstate, void *data, u_int len)
2144 MD5Update(fstate, (u_char *)data, len);
2149 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2152 * m pointer to head of mbuf chain
2153 * off0 offset to TCP header within the mbuf chain
2154 * len length of TCP segment data, excluding options
2155 * optlen length of TCP segment options
2156 * buf pointer to storage for computed MD5 digest
2157 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2159 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2160 * When called from tcp_input(), we can be sure that th_sum has been
2161 * zeroed out and verified already.
2163 * This function is for IPv4 use only. Calling this function with an
2164 * IPv6 packet in the mbuf chain will yield undefined results.
2166 * Return 0 if successful, otherwise return -1.
2168 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2169 * search with the destination IP address, and a 'magic SPI' to be
2170 * determined by the application. This is hardcoded elsewhere to 1179
2171 * right now. Another branch of this code exists which uses the SPD to
2172 * specify per-application flows but it is unstable.
2175 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2176 u_char *buf, u_int direction)
2178 union sockaddr_union dst;
2179 struct ippseudo ippseudo;
2183 struct ipovly *ipovly;
2184 struct secasvar *sav;
2188 KASSERT(m != NULL, ("NULL mbuf chain"));
2189 KASSERT(buf != NULL, ("NULL signature pointer"));
2191 /* Extract the destination from the IP header in the mbuf. */
2192 ip = mtod(m, struct ip *);
2193 bzero(&dst, sizeof(union sockaddr_union));
2194 dst.sa.sa_len = sizeof(struct sockaddr_in);
2195 dst.sa.sa_family = AF_INET;
2196 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2197 ip->ip_src : ip->ip_dst;
2199 /* Look up an SADB entry which matches the address of the peer. */
2200 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2202 printf("%s: SADB lookup failed for %s\n", __func__,
2203 inet_ntoa(dst.sin.sin_addr));
2208 ipovly = (struct ipovly *)ip;
2209 th = (struct tcphdr *)((u_char *)ip + off0);
2210 doff = off0 + sizeof(struct tcphdr) + optlen;
2213 * Step 1: Update MD5 hash with IP pseudo-header.
2215 * XXX The ippseudo header MUST be digested in network byte order,
2216 * or else we'll fail the regression test. Assume all fields we've
2217 * been doing arithmetic on have been in host byte order.
2218 * XXX One cannot depend on ipovly->ih_len here. When called from
2219 * tcp_output(), the underlying ip_len member has not yet been set.
2221 ippseudo.ippseudo_src = ipovly->ih_src;
2222 ippseudo.ippseudo_dst = ipovly->ih_dst;
2223 ippseudo.ippseudo_pad = 0;
2224 ippseudo.ippseudo_p = IPPROTO_TCP;
2225 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2226 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2229 * Step 2: Update MD5 hash with TCP header, excluding options.
2230 * The TCP checksum must be set to zero.
2232 savecsum = th->th_sum;
2234 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2235 th->th_sum = savecsum;
2238 * Step 3: Update MD5 hash with TCP segment data.
2239 * Use m_apply() to avoid an early m_pullup().
2242 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2245 * Step 4: Update MD5 hash with shared secret.
2247 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2248 MD5Final(buf, &ctx);
2250 key_sa_recordxfer(sav, m);
2254 #endif /* TCP_SIGNATURE */
2257 sysctl_drop(SYSCTL_HANDLER_ARGS)
2259 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2260 struct sockaddr_storage addrs[2];
2264 struct sockaddr_in *fin, *lin;
2266 struct sockaddr_in6 *fin6, *lin6;
2267 struct in6_addr f6, l6;
2278 if (req->oldptr != NULL || req->oldlen != 0)
2280 if (req->newptr == NULL)
2282 if (req->newlen < sizeof(addrs))
2284 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2288 switch (addrs[0].ss_family) {
2291 fin6 = (struct sockaddr_in6 *)&addrs[0];
2292 lin6 = (struct sockaddr_in6 *)&addrs[1];
2293 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2294 lin6->sin6_len != sizeof(struct sockaddr_in6))
2296 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2297 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2299 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2300 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2301 fin = (struct sockaddr_in *)&addrs[0];
2302 lin = (struct sockaddr_in *)&addrs[1];
2305 error = sa6_embedscope(fin6, ip6_use_defzone);
2308 error = sa6_embedscope(lin6, ip6_use_defzone);
2314 fin = (struct sockaddr_in *)&addrs[0];
2315 lin = (struct sockaddr_in *)&addrs[1];
2316 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2317 lin->sin_len != sizeof(struct sockaddr_in))
2323 INP_INFO_WLOCK(&tcbinfo);
2324 switch (addrs[0].ss_family) {
2327 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2328 &l6, lin6->sin6_port, 0, NULL);
2332 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2333 lin->sin_addr, lin->sin_port, 0, NULL);
2338 if (inp->inp_vflag & INP_TIMEWAIT) {
2340 * XXXRW: There currently exists a state where an
2341 * inpcb is present, but its timewait state has been
2342 * discarded. For now, don't allow dropping of this
2348 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2349 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2350 tp = intotcpcb(inp);
2351 tcp_drop(tp, ECONNABORTED);
2356 INP_INFO_WUNLOCK(&tcbinfo);
2360 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2361 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2362 0, sysctl_drop, "", "Drop TCP connection");