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"
40 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
50 #include <sys/domain.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 #include <sys/vimage.h>
62 #include <net/route.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
69 #include <netinet/ip6.h>
71 #include <netinet/in_pcb.h>
73 #include <netinet6/in6_pcb.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
91 #include <netinet6/tcp6_var.h>
93 #include <netinet/tcpip.h>
95 #include <netinet/tcp_debug.h>
97 #include <netinet6/ip6protosw.h>
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
103 #include <netipsec/ipsec6.h>
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
109 #include <machine/in_cksum.h>
112 #include <security/mac/mac_framework.h>
114 int tcp_mssdflt = TCP_MSS;
116 int tcp_v6mssdflt = TCP6_MSS;
120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
122 INIT_VNET_INET(TD_TO_VNET(curthread));
126 error = sysctl_handle_int(oidp, &new, 0, req);
127 if (error == 0 && req->newptr) {
128 if (new < TCP_MINMSS)
136 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLTYPE_INT|CTLFLAG_RW,
137 &tcp_mssdflt, 0, &sysctl_net_inet_tcp_mss_check, "I",
138 "Default TCP Maximum Segment Size");
142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
144 INIT_VNET_INET6(TD_TO_VNET(curthread));
147 new = V_tcp_v6mssdflt;
148 error = sysctl_handle_int(oidp, &new, 0, req);
149 if (error == 0 && req->newptr) {
150 if (new < TCP_MINMSS)
153 V_tcp_v6mssdflt = new;
158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLTYPE_INT|CTLFLAG_RW,
159 &tcp_v6mssdflt, 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
160 "Default TCP Maximum Segment Size for IPv6");
164 * Minimum MSS we accept and use. This prevents DoS attacks where
165 * we are forced to a ridiculous low MSS like 20 and send hundreds
166 * of packets instead of one. The effect scales with the available
167 * bandwidth and quickly saturates the CPU and network interface
168 * with packet generation and sending. Set to zero to disable MINMSS
169 * checking. This setting prevents us from sending too small packets.
171 int tcp_minmss = TCP_MINMSS;
172 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
173 CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
175 int tcp_do_rfc1323 = 1;
176 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
177 CTLFLAG_RW, tcp_do_rfc1323, 0,
178 "Enable rfc1323 (high performance TCP) extensions");
180 static int tcp_log_debug = 0;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
182 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
184 static int tcp_tcbhashsize = 0;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
186 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
188 static int do_tcpdrain = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
190 "Enable tcp_drain routine for extra help when low on mbufs");
192 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
193 CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs");
195 static int icmp_may_rst = 1;
196 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
197 CTLFLAG_RW, icmp_may_rst, 0,
198 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
200 static int tcp_isn_reseed_interval = 0;
201 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
202 CTLFLAG_RW, tcp_isn_reseed_interval, 0,
203 "Seconds between reseeding of ISN secret");
206 * TCP bandwidth limiting sysctls. Note that the default lower bound of
207 * 1024 exists only for debugging. A good production default would be
208 * something like 6100.
210 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
211 "TCP inflight data limiting");
213 static int tcp_inflight_enable = 1;
214 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
215 CTLFLAG_RW, tcp_inflight_enable, 0,
216 "Enable automatic TCP inflight data limiting");
218 static int tcp_inflight_debug = 0;
219 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
220 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
222 static int tcp_inflight_rttthresh;
223 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
224 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
225 "RTT threshold below which inflight will deactivate itself");
227 static int tcp_inflight_min = 6144;
228 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
229 CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
231 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
232 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
233 CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
235 static int tcp_inflight_stab = 20;
236 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
237 CTLFLAG_RW, tcp_inflight_stab, 0,
238 "Inflight Algorithm Stabilization 20 = 2 packets");
240 uma_zone_t sack_hole_zone;
242 static struct inpcb *tcp_notify(struct inpcb *, int);
243 static void tcp_isn_tick(void *);
246 * Target size of TCP PCB hash tables. Must be a power of two.
248 * Note that this can be overridden by the kernel environment
249 * variable net.inet.tcp.tcbhashsize
252 #define TCBHASHSIZE 512
257 * Callouts should be moved into struct tcp directly. They are currently
258 * separate because the tcpcb structure is exported to userland for sysctl
259 * parsing purposes, which do not know about callouts.
266 static uma_zone_t tcpcb_zone;
267 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
268 struct callout isn_callout;
269 static struct mtx isn_mtx;
271 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
272 #define ISN_LOCK() mtx_lock(&isn_mtx)
273 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
276 * TCP initialization.
279 tcp_zone_change(void *tag)
282 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
283 uma_zone_set_max(tcpcb_zone, maxsockets);
284 tcp_tw_zone_change();
288 tcp_inpcb_init(void *mem, int size, int flags)
290 struct inpcb *inp = mem;
292 INP_LOCK_INIT(inp, "inp", "tcpinp");
299 INIT_VNET_INET(curvnet);
301 int hashsize = TCBHASHSIZE;
302 tcp_delacktime = TCPTV_DELACK;
303 tcp_keepinit = TCPTV_KEEP_INIT;
304 tcp_keepidle = TCPTV_KEEP_IDLE;
305 tcp_keepintvl = TCPTV_KEEPINTVL;
306 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
308 tcp_rexmit_min = TCPTV_MIN;
309 if (tcp_rexmit_min < 1)
311 tcp_rexmit_slop = TCPTV_CPU_VAR;
312 V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
313 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
315 INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
317 V_tcbinfo.ipi_listhead = &V_tcb;
318 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
319 if (!powerof2(hashsize)) {
320 printf("WARNING: TCB hash size not a power of 2\n");
321 hashsize = 512; /* safe default */
323 tcp_tcbhashsize = hashsize;
324 V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
325 &V_tcbinfo.ipi_hashmask);
326 V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
327 &V_tcbinfo.ipi_porthashmask);
328 V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
329 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
330 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
332 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
334 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
336 if (max_protohdr < TCP_MINPROTOHDR)
337 max_protohdr = TCP_MINPROTOHDR;
338 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
340 #undef TCP_MINPROTOHDR
342 * These have to be type stable for the benefit of the timers.
344 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
345 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
346 uma_zone_set_max(tcpcb_zone, maxsockets);
352 callout_init(&isn_callout, CALLOUT_MPSAFE);
354 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
355 SHUTDOWN_PRI_DEFAULT);
356 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
357 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
358 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
359 EVENTHANDLER_PRI_ANY);
366 callout_stop(&isn_callout);
370 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
371 * tcp_template used to store this data in mbufs, but we now recopy it out
372 * of the tcpcb each time to conserve mbufs.
375 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
377 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
379 INP_WLOCK_ASSERT(inp);
382 if ((inp->inp_vflag & INP_IPV6) != 0) {
385 ip6 = (struct ip6_hdr *)ip_ptr;
386 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
387 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
388 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
389 (IPV6_VERSION & IPV6_VERSION_MASK);
390 ip6->ip6_nxt = IPPROTO_TCP;
391 ip6->ip6_plen = htons(sizeof(struct tcphdr));
392 ip6->ip6_src = inp->in6p_laddr;
393 ip6->ip6_dst = inp->in6p_faddr;
399 ip = (struct ip *)ip_ptr;
400 ip->ip_v = IPVERSION;
402 ip->ip_tos = inp->inp_ip_tos;
406 ip->ip_ttl = inp->inp_ip_ttl;
408 ip->ip_p = IPPROTO_TCP;
409 ip->ip_src = inp->inp_laddr;
410 ip->ip_dst = inp->inp_faddr;
412 th->th_sport = inp->inp_lport;
413 th->th_dport = inp->inp_fport;
421 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
425 * Create template to be used to send tcp packets on a connection.
426 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
427 * use for this function is in keepalives, which use tcp_respond.
430 tcpip_maketemplate(struct inpcb *inp)
434 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
437 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
442 * Send a single message to the TCP at address specified by
443 * the given TCP/IP header. If m == NULL, then we make a copy
444 * of the tcpiphdr at ti and send directly to the addressed host.
445 * This is used to force keep alive messages out using the TCP
446 * template for a connection. If flags are given then we send
447 * a message back to the TCP which originated the * segment ti,
448 * and discard the mbuf containing it and any other attached mbufs.
450 * In any case the ack and sequence number of the transmitted
451 * segment are as specified by the parameters.
453 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
456 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
457 tcp_seq ack, tcp_seq seq, int flags)
459 INIT_VNET_INET(curvnet);
471 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
474 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
481 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
482 INP_WLOCK_ASSERT(inp);
487 if (!(flags & TH_RST)) {
488 win = sbspace(&inp->inp_socket->so_rcv);
489 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
490 win = (long)TCP_MAXWIN << tp->rcv_scale;
494 m = m_gethdr(M_DONTWAIT, MT_DATA);
498 m->m_data += max_linkhdr;
501 bcopy((caddr_t)ip6, mtod(m, caddr_t),
502 sizeof(struct ip6_hdr));
503 ip6 = mtod(m, struct ip6_hdr *);
504 nth = (struct tcphdr *)(ip6 + 1);
508 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
509 ip = mtod(m, struct ip *);
510 nth = (struct tcphdr *)(ip + 1);
512 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
517 * XXX MRT We inherrit the FIB, which is lucky.
521 m->m_data = (caddr_t)ipgen;
522 /* m_len is set later */
524 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
527 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
528 nth = (struct tcphdr *)(ip6 + 1);
532 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
533 nth = (struct tcphdr *)(ip + 1);
537 * this is usually a case when an extension header
538 * exists between the IPv6 header and the
541 nth->th_sport = th->th_sport;
542 nth->th_dport = th->th_dport;
544 xchg(nth->th_dport, nth->th_sport, n_short);
550 ip6->ip6_vfc = IPV6_VERSION;
551 ip6->ip6_nxt = IPPROTO_TCP;
552 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
554 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
558 tlen += sizeof (struct tcpiphdr);
560 ip->ip_ttl = V_ip_defttl;
561 if (V_path_mtu_discovery)
565 m->m_pkthdr.len = tlen;
566 m->m_pkthdr.rcvif = NULL;
570 * Packet is associated with a socket, so allow the
571 * label of the response to reflect the socket label.
573 INP_WLOCK_ASSERT(inp);
574 mac_inpcb_create_mbuf(inp, m);
577 * Packet is not associated with a socket, so possibly
578 * update the label in place.
580 mac_netinet_tcp_reply(m);
583 nth->th_seq = htonl(seq);
584 nth->th_ack = htonl(ack);
586 nth->th_off = sizeof (struct tcphdr) >> 2;
587 nth->th_flags = flags;
589 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
591 nth->th_win = htons((u_short)win);
596 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
597 sizeof(struct ip6_hdr),
598 tlen - sizeof(struct ip6_hdr));
599 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
604 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
605 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
606 m->m_pkthdr.csum_flags = CSUM_TCP;
607 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
610 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
611 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
615 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
618 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
622 * Create a new TCP control block, making an
623 * empty reassembly queue and hooking it to the argument
624 * protocol control block. The `inp' parameter must have
625 * come from the zone allocator set up in tcp_init().
628 tcp_newtcpcb(struct inpcb *inp)
630 INIT_VNET_INET(inp->inp_vnet);
631 struct tcpcb_mem *tm;
634 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
637 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
641 tp->t_timers = &tm->tt;
642 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
643 tp->t_maxseg = tp->t_maxopd =
645 isipv6 ? V_tcp_v6mssdflt :
649 /* Set up our timeouts. */
650 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
651 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
652 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
653 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
654 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
656 if (V_tcp_do_rfc1323)
657 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
659 tp->t_flags |= TF_SACK_PERMIT;
660 TAILQ_INIT(&tp->snd_holes);
661 tp->t_inpcb = inp; /* XXX */
663 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
664 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
665 * reasonable initial retransmit time.
667 tp->t_srtt = TCPTV_SRTTBASE;
668 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
669 tp->t_rttmin = tcp_rexmit_min;
670 tp->t_rxtcur = TCPTV_RTOBASE;
671 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
672 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
673 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
674 tp->t_rcvtime = ticks;
675 tp->t_bw_rtttime = ticks;
677 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
678 * because the socket may be bound to an IPv6 wildcard address,
679 * which may match an IPv4-mapped IPv6 address.
681 inp->inp_ip_ttl = V_ip_defttl;
683 return (tp); /* XXX */
687 * Drop a TCP connection, reporting
688 * the specified error. If connection is synchronized,
689 * then send a RST to peer.
692 tcp_drop(struct tcpcb *tp, int errno)
694 INIT_VNET_INET(tp->t_inpcb->inp_vnet);
695 struct socket *so = tp->t_inpcb->inp_socket;
697 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
698 INP_WLOCK_ASSERT(tp->t_inpcb);
700 if (TCPS_HAVERCVDSYN(tp->t_state)) {
701 tp->t_state = TCPS_CLOSED;
702 (void) tcp_output_reset(tp);
703 V_tcpstat.tcps_drops++;
705 V_tcpstat.tcps_conndrops++;
706 if (errno == ETIMEDOUT && tp->t_softerror)
707 errno = tp->t_softerror;
708 so->so_error = errno;
709 return (tcp_close(tp));
713 tcp_discardcb(struct tcpcb *tp)
715 INIT_VNET_INET(tp->t_vnet);
717 struct inpcb *inp = tp->t_inpcb;
718 struct socket *so = inp->inp_socket;
720 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
723 INP_WLOCK_ASSERT(inp);
726 * Make sure that all of our timers are stopped before we
729 callout_stop(&tp->t_timers->tt_rexmt);
730 callout_stop(&tp->t_timers->tt_persist);
731 callout_stop(&tp->t_timers->tt_keep);
732 callout_stop(&tp->t_timers->tt_2msl);
733 callout_stop(&tp->t_timers->tt_delack);
736 * If we got enough samples through the srtt filter,
737 * save the rtt and rttvar in the routing entry.
738 * 'Enough' is arbitrarily defined as 4 rtt samples.
739 * 4 samples is enough for the srtt filter to converge
740 * to within enough % of the correct value; fewer samples
741 * and we could save a bogus rtt. The danger is not high
742 * as tcp quickly recovers from everything.
743 * XXX: Works very well but needs some more statistics!
745 if (tp->t_rttupdated >= 4) {
746 struct hc_metrics_lite metrics;
749 bzero(&metrics, sizeof(metrics));
751 * Update the ssthresh always when the conditions below
752 * are satisfied. This gives us better new start value
753 * for the congestion avoidance for new connections.
754 * ssthresh is only set if packet loss occured on a session.
756 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
757 * being torn down. Ideally this code would not use 'so'.
759 ssthresh = tp->snd_ssthresh;
760 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
762 * convert the limit from user data bytes to
763 * packets then to packet data bytes.
765 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
768 ssthresh *= (u_long)(tp->t_maxseg +
770 (isipv6 ? sizeof (struct ip6_hdr) +
771 sizeof (struct tcphdr) :
773 sizeof (struct tcpiphdr)
780 metrics.rmx_ssthresh = ssthresh;
782 metrics.rmx_rtt = tp->t_srtt;
783 metrics.rmx_rttvar = tp->t_rttvar;
784 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
785 metrics.rmx_bandwidth = tp->snd_bandwidth;
786 metrics.rmx_cwnd = tp->snd_cwnd;
787 metrics.rmx_sendpipe = 0;
788 metrics.rmx_recvpipe = 0;
790 tcp_hc_update(&inp->inp_inc, &metrics);
793 /* free the reassembly queue, if any */
794 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
795 LIST_REMOVE(q, tqe_q);
797 uma_zfree(tcp_reass_zone, q);
801 /* Disconnect offload device, if any. */
802 tcp_offload_detach(tp);
804 tcp_free_sackholes(tp);
805 inp->inp_ppcb = NULL;
807 uma_zfree(tcpcb_zone, tp);
811 * Attempt to close a TCP control block, marking it as dropped, and freeing
812 * the socket if we hold the only reference.
815 tcp_close(struct tcpcb *tp)
817 INIT_VNET_INET(tp->t_inpcb->inp_vnet);
818 struct inpcb *inp = tp->t_inpcb;
821 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
822 INP_WLOCK_ASSERT(inp);
824 /* Notify any offload devices of listener close */
825 if (tp->t_state == TCPS_LISTEN)
826 tcp_offload_listen_close(tp);
828 V_tcpstat.tcps_closed++;
829 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
830 so = inp->inp_socket;
831 soisdisconnected(so);
832 if (inp->inp_vflag & INP_SOCKREF) {
833 KASSERT(so->so_state & SS_PROTOREF,
834 ("tcp_close: !SS_PROTOREF"));
835 inp->inp_vflag &= ~INP_SOCKREF;
839 so->so_state &= ~SS_PROTOREF;
849 VNET_ITERATOR_DECL(vnet_iter);
855 VNET_FOREACH(vnet_iter) {
856 CURVNET_SET(vnet_iter);
857 INIT_VNET_INET(vnet_iter);
860 struct tseg_qent *te;
863 * Walk the tcpbs, if existing, and flush the reassembly queue,
865 * XXX: The "Net/3" implementation doesn't imply that the TCP
866 * reassembly queue should be flushed, but in a situation
867 * where we're really low on mbufs, this is potentially
870 INP_INFO_RLOCK(&V_tcbinfo);
871 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
872 if (inpb->inp_vflag & INP_TIMEWAIT)
875 if ((tcpb = intotcpcb(inpb)) != NULL) {
876 while ((te = LIST_FIRST(&tcpb->t_segq))
878 LIST_REMOVE(te, tqe_q);
880 uma_zfree(tcp_reass_zone, te);
884 tcp_clean_sackreport(tcpb);
888 INP_INFO_RUNLOCK(&V_tcbinfo);
895 * Notify a tcp user of an asynchronous error;
896 * store error as soft error, but wake up user
897 * (for now, won't do anything until can select for soft error).
899 * Do not wake up user since there currently is no mechanism for
900 * reporting soft errors (yet - a kqueue filter may be added).
902 static struct inpcb *
903 tcp_notify(struct inpcb *inp, int error)
907 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
908 INP_WLOCK_ASSERT(inp);
910 if ((inp->inp_vflag & INP_TIMEWAIT) ||
911 (inp->inp_vflag & INP_DROPPED))
915 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
918 * Ignore some errors if we are hooked up.
919 * If connection hasn't completed, has retransmitted several times,
920 * and receives a second error, give up now. This is better
921 * than waiting a long time to establish a connection that
922 * can never complete.
924 if (tp->t_state == TCPS_ESTABLISHED &&
925 (error == EHOSTUNREACH || error == ENETUNREACH ||
926 error == EHOSTDOWN)) {
928 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
930 tp = tcp_drop(tp, error);
936 tp->t_softerror = error;
940 wakeup( &so->so_timeo);
947 tcp_pcblist(SYSCTL_HANDLER_ARGS)
949 INIT_VNET_INET(curvnet);
950 int error, i, m, n, pcb_count;
951 struct inpcb *inp, **inp_list;
956 * The process of preparing the TCB list is too time-consuming and
957 * resource-intensive to repeat twice on every request.
959 if (req->oldptr == NULL) {
960 m = syncache_pcbcount();
961 n = V_tcbinfo.ipi_count;
962 req->oldidx = 2 * (sizeof xig)
963 + ((m + n) + n/8) * sizeof(struct xtcpcb);
967 if (req->newptr != NULL)
971 * OK, now we're committed to doing something.
973 INP_INFO_RLOCK(&V_tcbinfo);
974 gencnt = V_tcbinfo.ipi_gencnt;
975 n = V_tcbinfo.ipi_count;
976 INP_INFO_RUNLOCK(&V_tcbinfo);
978 m = syncache_pcbcount();
980 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
981 + (n + m) * sizeof(struct xtcpcb));
985 xig.xig_len = sizeof xig;
986 xig.xig_count = n + m;
987 xig.xig_gen = gencnt;
988 xig.xig_sogen = so_gencnt;
989 error = SYSCTL_OUT(req, &xig, sizeof xig);
993 error = syncache_pcblist(req, m, &pcb_count);
997 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
998 if (inp_list == NULL)
1001 INP_INFO_RLOCK(&V_tcbinfo);
1002 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1003 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1005 if (inp->inp_gencnt <= gencnt) {
1007 * XXX: This use of cr_cansee(), introduced with
1008 * TCP state changes, is not quite right, but for
1009 * now, better than nothing.
1011 if (inp->inp_vflag & INP_TIMEWAIT) {
1012 if (intotw(inp) != NULL)
1013 error = cr_cansee(req->td->td_ucred,
1014 intotw(inp)->tw_cred);
1016 error = EINVAL; /* Skip this inp. */
1018 error = cr_canseeinpcb(req->td->td_ucred, inp);
1020 inp_list[i++] = inp;
1024 INP_INFO_RUNLOCK(&V_tcbinfo);
1028 for (i = 0; i < n; i++) {
1031 if (inp->inp_gencnt <= gencnt) {
1035 bzero(&xt, sizeof(xt));
1036 xt.xt_len = sizeof xt;
1037 /* XXX should avoid extra copy */
1038 bcopy(inp, &xt.xt_inp, sizeof *inp);
1039 inp_ppcb = inp->inp_ppcb;
1040 if (inp_ppcb == NULL)
1041 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1042 else if (inp->inp_vflag & INP_TIMEWAIT) {
1043 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1044 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1046 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1047 if (inp->inp_socket != NULL)
1048 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1050 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1051 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1053 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1055 error = SYSCTL_OUT(req, &xt, sizeof xt);
1062 * Give the user an updated idea of our state.
1063 * If the generation differs from what we told
1064 * her before, she knows that something happened
1065 * while we were processing this request, and it
1066 * might be necessary to retry.
1068 INP_INFO_RLOCK(&V_tcbinfo);
1069 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1070 xig.xig_sogen = so_gencnt;
1071 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1072 INP_INFO_RUNLOCK(&V_tcbinfo);
1073 error = SYSCTL_OUT(req, &xig, sizeof xig);
1075 free(inp_list, M_TEMP);
1079 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1080 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1083 tcp_getcred(SYSCTL_HANDLER_ARGS)
1085 INIT_VNET_INET(curvnet);
1087 struct sockaddr_in addrs[2];
1091 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1094 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1097 INP_INFO_RLOCK(&V_tcbinfo);
1098 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1099 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1102 INP_INFO_RUNLOCK(&V_tcbinfo);
1103 if (inp->inp_socket == NULL)
1106 error = cr_canseeinpcb(req->td->td_ucred, inp);
1108 cru2x(inp->inp_cred, &xuc);
1111 INP_INFO_RUNLOCK(&V_tcbinfo);
1115 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1120 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1121 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1125 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1127 INIT_VNET_INET(curvnet);
1128 INIT_VNET_INET6(curvnet);
1130 struct sockaddr_in6 addrs[2];
1132 int error, mapped = 0;
1134 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1137 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1140 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1141 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1144 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1145 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1151 INP_INFO_RLOCK(&V_tcbinfo);
1153 inp = in_pcblookup_hash(&V_tcbinfo,
1154 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1156 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1160 inp = in6_pcblookup_hash(&V_tcbinfo,
1161 &addrs[1].sin6_addr, addrs[1].sin6_port,
1162 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1165 INP_INFO_RUNLOCK(&V_tcbinfo);
1166 if (inp->inp_socket == NULL)
1169 error = cr_canseeinpcb(req->td->td_ucred, inp);
1171 cru2x(inp->inp_cred, &xuc);
1174 INP_INFO_RUNLOCK(&V_tcbinfo);
1178 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1182 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1183 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1184 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1189 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1191 INIT_VNET_INET(curvnet);
1192 struct ip *ip = vip;
1194 struct in_addr faddr;
1197 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1199 struct in_conninfo inc;
1200 tcp_seq icmp_tcp_seq;
1203 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1204 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1207 if (cmd == PRC_MSGSIZE)
1208 notify = tcp_mtudisc;
1209 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1210 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1211 notify = tcp_drop_syn_sent;
1213 * Redirects don't need to be handled up here.
1215 else if (PRC_IS_REDIRECT(cmd))
1218 * Source quench is depreciated.
1220 else if (cmd == PRC_QUENCH)
1223 * Hostdead is ugly because it goes linearly through all PCBs.
1224 * XXX: We never get this from ICMP, otherwise it makes an
1225 * excellent DoS attack on machines with many connections.
1227 else if (cmd == PRC_HOSTDEAD)
1229 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1232 icp = (struct icmp *)((caddr_t)ip
1233 - offsetof(struct icmp, icmp_ip));
1234 th = (struct tcphdr *)((caddr_t)ip
1235 + (ip->ip_hl << 2));
1236 INP_INFO_WLOCK(&V_tcbinfo);
1237 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1238 ip->ip_src, th->th_sport, 0, NULL);
1241 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1242 !(inp->inp_vflag & INP_DROPPED) &&
1243 !(inp->inp_socket == NULL)) {
1244 icmp_tcp_seq = htonl(th->th_seq);
1245 tp = intotcpcb(inp);
1246 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1247 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1248 if (cmd == PRC_MSGSIZE) {
1251 * If we got a needfrag set the MTU
1252 * in the route to the suggested new
1253 * value (if given) and then notify.
1255 bzero(&inc, sizeof(inc));
1256 inc.inc_flags = 0; /* IPv4 */
1257 inc.inc_faddr = faddr;
1259 inp->inp_inc.inc_fibnum;
1261 mtu = ntohs(icp->icmp_nextmtu);
1263 * If no alternative MTU was
1264 * proposed, try the next smaller
1265 * one. ip->ip_len has already
1266 * been swapped in icmp_input().
1269 mtu = ip_next_mtu(ip->ip_len,
1271 if (mtu < max(296, V_tcp_minmss
1272 + sizeof(struct tcpiphdr)))
1276 + sizeof(struct tcpiphdr);
1278 * Only cache the the MTU if it
1279 * is smaller than the interface
1280 * or route MTU. tcp_mtudisc()
1281 * will do right thing by itself.
1283 if (mtu <= tcp_maxmtu(&inc, NULL))
1284 tcp_hc_updatemtu(&inc, mtu);
1287 inp = (*notify)(inp, inetctlerrmap[cmd]);
1293 inc.inc_fport = th->th_dport;
1294 inc.inc_lport = th->th_sport;
1295 inc.inc_faddr = faddr;
1296 inc.inc_laddr = ip->ip_src;
1300 syncache_unreach(&inc, th);
1302 INP_INFO_WUNLOCK(&V_tcbinfo);
1304 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1309 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1311 INIT_VNET_INET(curvnet);
1313 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1314 struct ip6_hdr *ip6;
1316 struct ip6ctlparam *ip6cp = NULL;
1317 const struct sockaddr_in6 *sa6_src = NULL;
1319 struct tcp_portonly {
1324 if (sa->sa_family != AF_INET6 ||
1325 sa->sa_len != sizeof(struct sockaddr_in6))
1328 if (cmd == PRC_MSGSIZE)
1329 notify = tcp_mtudisc;
1330 else if (!PRC_IS_REDIRECT(cmd) &&
1331 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1333 /* Source quench is depreciated. */
1334 else if (cmd == PRC_QUENCH)
1337 /* if the parameter is from icmp6, decode it. */
1339 ip6cp = (struct ip6ctlparam *)d;
1341 ip6 = ip6cp->ip6c_ip6;
1342 off = ip6cp->ip6c_off;
1343 sa6_src = ip6cp->ip6c_src;
1347 off = 0; /* fool gcc */
1352 struct in_conninfo inc;
1354 * XXX: We assume that when IPV6 is non NULL,
1355 * M and OFF are valid.
1358 /* check if we can safely examine src and dst ports */
1359 if (m->m_pkthdr.len < off + sizeof(*thp))
1362 bzero(&th, sizeof(th));
1363 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1365 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1366 (struct sockaddr *)ip6cp->ip6c_src,
1367 th.th_sport, cmd, NULL, notify);
1369 inc.inc_fport = th.th_dport;
1370 inc.inc_lport = th.th_sport;
1371 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1372 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1374 INP_INFO_WLOCK(&V_tcbinfo);
1375 syncache_unreach(&inc, &th);
1376 INP_INFO_WUNLOCK(&V_tcbinfo);
1378 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1379 0, cmd, NULL, notify);
1385 * Following is where TCP initial sequence number generation occurs.
1387 * There are two places where we must use initial sequence numbers:
1388 * 1. In SYN-ACK packets.
1389 * 2. In SYN packets.
1391 * All ISNs for SYN-ACK packets are generated by the syncache. See
1392 * tcp_syncache.c for details.
1394 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1395 * depends on this property. In addition, these ISNs should be
1396 * unguessable so as to prevent connection hijacking. To satisfy
1397 * the requirements of this situation, the algorithm outlined in
1398 * RFC 1948 is used, with only small modifications.
1400 * Implementation details:
1402 * Time is based off the system timer, and is corrected so that it
1403 * increases by one megabyte per second. This allows for proper
1404 * recycling on high speed LANs while still leaving over an hour
1407 * As reading the *exact* system time is too expensive to be done
1408 * whenever setting up a TCP connection, we increment the time
1409 * offset in two ways. First, a small random positive increment
1410 * is added to isn_offset for each connection that is set up.
1411 * Second, the function tcp_isn_tick fires once per clock tick
1412 * and increments isn_offset as necessary so that sequence numbers
1413 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1414 * random positive increments serve only to ensure that the same
1415 * exact sequence number is never sent out twice (as could otherwise
1416 * happen when a port is recycled in less than the system tick
1419 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1420 * between seeding of isn_secret. This is normally set to zero,
1421 * as reseeding should not be necessary.
1423 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1424 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1425 * general, this means holding an exclusive (write) lock.
1428 #define ISN_BYTES_PER_SECOND 1048576
1429 #define ISN_STATIC_INCREMENT 4096
1430 #define ISN_RANDOM_INCREMENT (4096 - 1)
1432 static u_char isn_secret[32];
1433 static int isn_last_reseed;
1434 static u_int32_t isn_offset, isn_offset_old;
1435 static MD5_CTX isn_ctx;
1438 tcp_new_isn(struct tcpcb *tp)
1440 INIT_VNET_INET(tp->t_vnet);
1441 u_int32_t md5_buffer[4];
1444 INP_WLOCK_ASSERT(tp->t_inpcb);
1447 /* Seed if this is the first use, reseed if requested. */
1448 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1449 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1451 read_random(&V_isn_secret, sizeof(V_isn_secret));
1452 V_isn_last_reseed = ticks;
1455 /* Compute the md5 hash and return the ISN. */
1456 MD5Init(&V_isn_ctx);
1457 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1458 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1460 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1461 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1462 sizeof(struct in6_addr));
1463 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1464 sizeof(struct in6_addr));
1468 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1469 sizeof(struct in_addr));
1470 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1471 sizeof(struct in_addr));
1473 MD5Update(&V_isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1474 MD5Final((u_char *) &md5_buffer, &V_isn_ctx);
1475 new_isn = (tcp_seq) md5_buffer[0];
1476 V_isn_offset += ISN_STATIC_INCREMENT +
1477 (arc4random() & ISN_RANDOM_INCREMENT);
1478 new_isn += V_isn_offset;
1484 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1485 * to keep time flowing at a relatively constant rate. If the random
1486 * increments have already pushed us past the projected offset, do nothing.
1489 tcp_isn_tick(void *xtp)
1491 VNET_ITERATOR_DECL(vnet_iter);
1492 u_int32_t projected_offset;
1496 VNET_FOREACH(vnet_iter) {
1497 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1498 INIT_VNET_INET(curvnet);
1500 V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1502 if (SEQ_GT(projected_offset, V_isn_offset))
1503 V_isn_offset = projected_offset;
1505 V_isn_offset_old = V_isn_offset;
1508 VNET_LIST_RUNLOCK();
1509 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1514 * When a specific ICMP unreachable message is received and the
1515 * connection state is SYN-SENT, drop the connection. This behavior
1516 * is controlled by the icmp_may_rst sysctl.
1519 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1522 INIT_VNET_INET(inp->inp_vnet);
1526 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1527 INP_WLOCK_ASSERT(inp);
1529 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1530 (inp->inp_vflag & INP_DROPPED))
1533 tp = intotcpcb(inp);
1534 if (tp->t_state != TCPS_SYN_SENT)
1537 tp = tcp_drop(tp, errno);
1545 * When `need fragmentation' ICMP is received, update our idea of the MSS
1546 * based on the new value in the route. Also nudge TCP to send something,
1547 * since we know the packet we just sent was dropped.
1548 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1551 tcp_mtudisc(struct inpcb *inp, int errno)
1553 INIT_VNET_INET(inp->inp_vnet);
1557 INP_WLOCK_ASSERT(inp);
1558 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1559 (inp->inp_vflag & INP_DROPPED))
1562 tp = intotcpcb(inp);
1563 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1565 tcp_mss_update(tp, -1, NULL);
1567 so = inp->inp_socket;
1568 SOCKBUF_LOCK(&so->so_snd);
1569 /* If the mss is larger than the socket buffer, decrease the mss. */
1570 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1571 tp->t_maxseg = so->so_snd.sb_hiwat;
1572 SOCKBUF_UNLOCK(&so->so_snd);
1574 V_tcpstat.tcps_mturesent++;
1576 tp->snd_nxt = tp->snd_una;
1577 tcp_free_sackholes(tp);
1578 tp->snd_recover = tp->snd_max;
1579 if (tp->t_flags & TF_SACK_PERMIT)
1580 EXIT_FASTRECOVERY(tp);
1581 tcp_output_send(tp);
1586 * Look-up the routing entry to the peer of this inpcb. If no route
1587 * is found and it cannot be allocated, then return NULL. This routine
1588 * is called by TCP routines that access the rmx structure and by tcp_mss
1589 * to get the interface MTU.
1592 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1595 struct sockaddr_in *dst;
1599 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1601 bzero(&sro, sizeof(sro));
1602 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1603 dst = (struct sockaddr_in *)&sro.ro_dst;
1604 dst->sin_family = AF_INET;
1605 dst->sin_len = sizeof(*dst);
1606 dst->sin_addr = inc->inc_faddr;
1607 in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1609 if (sro.ro_rt != NULL) {
1610 ifp = sro.ro_rt->rt_ifp;
1611 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1612 maxmtu = ifp->if_mtu;
1614 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1616 /* Report additional interface capabilities. */
1617 if (flags != NULL) {
1618 if (ifp->if_capenable & IFCAP_TSO4 &&
1619 ifp->if_hwassist & CSUM_TSO)
1629 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1631 struct route_in6 sro6;
1635 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1637 bzero(&sro6, sizeof(sro6));
1638 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1639 sro6.ro_dst.sin6_family = AF_INET6;
1640 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1641 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1642 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1644 if (sro6.ro_rt != NULL) {
1645 ifp = sro6.ro_rt->rt_ifp;
1646 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1647 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1649 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1650 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1652 /* Report additional interface capabilities. */
1653 if (flags != NULL) {
1654 if (ifp->if_capenable & IFCAP_TSO6 &&
1655 ifp->if_hwassist & CSUM_TSO)
1666 /* compute ESP/AH header size for TCP, including outer IP header. */
1668 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1675 struct ip6_hdr *ip6;
1679 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1681 MGETHDR(m, M_DONTWAIT, MT_DATA);
1686 if ((inp->inp_vflag & INP_IPV6) != 0) {
1687 ip6 = mtod(m, struct ip6_hdr *);
1688 th = (struct tcphdr *)(ip6 + 1);
1689 m->m_pkthdr.len = m->m_len =
1690 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1691 tcpip_fillheaders(inp, ip6, th);
1692 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1696 ip = mtod(m, struct ip *);
1697 th = (struct tcphdr *)(ip + 1);
1698 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1699 tcpip_fillheaders(inp, ip, th);
1700 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1709 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1711 * This code attempts to calculate the bandwidth-delay product as a
1712 * means of determining the optimal window size to maximize bandwidth,
1713 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1714 * routers. This code also does a fairly good job keeping RTTs in check
1715 * across slow links like modems. We implement an algorithm which is very
1716 * similar (but not meant to be) TCP/Vegas. The code operates on the
1717 * transmitter side of a TCP connection and so only effects the transmit
1718 * side of the connection.
1720 * BACKGROUND: TCP makes no provision for the management of buffer space
1721 * at the end points or at the intermediate routers and switches. A TCP
1722 * stream, whether using NewReno or not, will eventually buffer as
1723 * many packets as it is able and the only reason this typically works is
1724 * due to the fairly small default buffers made available for a connection
1725 * (typicaly 16K or 32K). As machines use larger windows and/or window
1726 * scaling it is now fairly easy for even a single TCP connection to blow-out
1727 * all available buffer space not only on the local interface, but on
1728 * intermediate routers and switches as well. NewReno makes a misguided
1729 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1730 * then backing off, then steadily increasing the window again until another
1731 * failure occurs, ad-infinitum. This results in terrible oscillation that
1732 * is only made worse as network loads increase and the idea of intentionally
1733 * blowing out network buffers is, frankly, a terrible way to manage network
1736 * It is far better to limit the transmit window prior to the failure
1737 * condition being achieved. There are two general ways to do this: First
1738 * you can 'scan' through different transmit window sizes and locate the
1739 * point where the RTT stops increasing, indicating that you have filled the
1740 * pipe, then scan backwards until you note that RTT stops decreasing, then
1741 * repeat ad-infinitum. This method works in principle but has severe
1742 * implementation issues due to RTT variances, timer granularity, and
1743 * instability in the algorithm which can lead to many false positives and
1744 * create oscillations as well as interact badly with other TCP streams
1745 * implementing the same algorithm.
1747 * The second method is to limit the window to the bandwidth delay product
1748 * of the link. This is the method we implement. RTT variances and our
1749 * own manipulation of the congestion window, bwnd, can potentially
1750 * destabilize the algorithm. For this reason we have to stabilize the
1751 * elements used to calculate the window. We do this by using the minimum
1752 * observed RTT, the long term average of the observed bandwidth, and
1753 * by adding two segments worth of slop. It isn't perfect but it is able
1754 * to react to changing conditions and gives us a very stable basis on
1755 * which to extend the algorithm.
1758 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1760 INIT_VNET_INET(tp->t_vnet);
1765 INP_WLOCK_ASSERT(tp->t_inpcb);
1768 * If inflight_enable is disabled in the middle of a tcp connection,
1769 * make sure snd_bwnd is effectively disabled.
1771 if (V_tcp_inflight_enable == 0 ||
1772 tp->t_rttlow < V_tcp_inflight_rttthresh) {
1773 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1774 tp->snd_bandwidth = 0;
1779 * Figure out the bandwidth. Due to the tick granularity this
1780 * is a very rough number and it MUST be averaged over a fairly
1781 * long period of time. XXX we need to take into account a link
1782 * that is not using all available bandwidth, but for now our
1783 * slop will ramp us up if this case occurs and the bandwidth later
1786 * Note: if ticks rollover 'bw' may wind up negative. We must
1787 * effectively reset t_bw_rtttime for this case.
1790 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1793 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1794 (save_ticks - tp->t_bw_rtttime);
1795 tp->t_bw_rtttime = save_ticks;
1796 tp->t_bw_rtseq = ack_seq;
1797 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1799 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1801 tp->snd_bandwidth = bw;
1804 * Calculate the semi-static bandwidth delay product, plus two maximal
1805 * segments. The additional slop puts us squarely in the sweet
1806 * spot and also handles the bandwidth run-up case and stabilization.
1807 * Without the slop we could be locking ourselves into a lower
1810 * Situations Handled:
1811 * (1) Prevents over-queueing of packets on LANs, especially on
1812 * high speed LANs, allowing larger TCP buffers to be
1813 * specified, and also does a good job preventing
1814 * over-queueing of packets over choke points like modems
1815 * (at least for the transmit side).
1817 * (2) Is able to handle changing network loads (bandwidth
1818 * drops so bwnd drops, bandwidth increases so bwnd
1821 * (3) Theoretically should stabilize in the face of multiple
1822 * connections implementing the same algorithm (this may need
1825 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1826 * be adjusted with a sysctl but typically only needs to be
1827 * on very slow connections. A value no smaller then 5
1828 * should be used, but only reduce this default if you have
1831 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1832 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1835 if (tcp_inflight_debug > 0) {
1837 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1839 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1848 if ((long)bwnd < V_tcp_inflight_min)
1849 bwnd = V_tcp_inflight_min;
1850 if (bwnd > V_tcp_inflight_max)
1851 bwnd = V_tcp_inflight_max;
1852 if ((long)bwnd < tp->t_maxseg * 2)
1853 bwnd = tp->t_maxseg * 2;
1854 tp->snd_bwnd = bwnd;
1857 #ifdef TCP_SIGNATURE
1859 * Callback function invoked by m_apply() to digest TCP segment data
1860 * contained within an mbuf chain.
1863 tcp_signature_apply(void *fstate, void *data, u_int len)
1866 MD5Update(fstate, (u_char *)data, len);
1871 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1874 * m pointer to head of mbuf chain
1876 * len length of TCP segment data, excluding options
1877 * optlen length of TCP segment options
1878 * buf pointer to storage for computed MD5 digest
1879 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1881 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1882 * When called from tcp_input(), we can be sure that th_sum has been
1883 * zeroed out and verified already.
1885 * Return 0 if successful, otherwise return -1.
1887 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1888 * search with the destination IP address, and a 'magic SPI' to be
1889 * determined by the application. This is hardcoded elsewhere to 1179
1890 * right now. Another branch of this code exists which uses the SPD to
1891 * specify per-application flows but it is unstable.
1894 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1895 u_char *buf, u_int direction)
1897 union sockaddr_union dst;
1898 struct ippseudo ippseudo;
1902 struct ipovly *ipovly;
1903 struct secasvar *sav;
1906 struct ip6_hdr *ip6;
1907 struct in6_addr in6;
1908 char ip6buf[INET6_ADDRSTRLEN];
1914 KASSERT(m != NULL, ("NULL mbuf chain"));
1915 KASSERT(buf != NULL, ("NULL signature pointer"));
1917 /* Extract the destination from the IP header in the mbuf. */
1918 bzero(&dst, sizeof(union sockaddr_union));
1919 ip = mtod(m, struct ip *);
1921 ip6 = NULL; /* Make the compiler happy. */
1925 dst.sa.sa_len = sizeof(struct sockaddr_in);
1926 dst.sa.sa_family = AF_INET;
1927 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1928 ip->ip_src : ip->ip_dst;
1931 case (IPV6_VERSION >> 4):
1932 ip6 = mtod(m, struct ip6_hdr *);
1933 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1934 dst.sa.sa_family = AF_INET6;
1935 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1936 ip6->ip6_src : ip6->ip6_dst;
1945 /* Look up an SADB entry which matches the address of the peer. */
1946 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1948 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1949 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1951 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1952 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1960 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1962 * XXX The ippseudo header MUST be digested in network byte order,
1963 * or else we'll fail the regression test. Assume all fields we've
1964 * been doing arithmetic on have been in host byte order.
1965 * XXX One cannot depend on ipovly->ih_len here. When called from
1966 * tcp_output(), the underlying ip_len member has not yet been set.
1970 ipovly = (struct ipovly *)ip;
1971 ippseudo.ippseudo_src = ipovly->ih_src;
1972 ippseudo.ippseudo_dst = ipovly->ih_dst;
1973 ippseudo.ippseudo_pad = 0;
1974 ippseudo.ippseudo_p = IPPROTO_TCP;
1975 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1977 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1979 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1980 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1984 * RFC 2385, 2.0 Proposal
1985 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1986 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1987 * extended next header value (to form 32 bits), and 32-bit segment
1989 * Note: Upper-Layer Packet Length comes before Next Header.
1991 case (IPV6_VERSION >> 4):
1993 in6_clearscope(&in6);
1994 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1996 in6_clearscope(&in6);
1997 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1998 plen = htonl(len + sizeof(struct tcphdr) + optlen);
1999 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2001 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2002 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2003 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2005 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2007 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2008 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2019 * Step 2: Update MD5 hash with TCP header, excluding options.
2020 * The TCP checksum must be set to zero.
2022 savecsum = th->th_sum;
2024 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2025 th->th_sum = savecsum;
2028 * Step 3: Update MD5 hash with TCP segment data.
2029 * Use m_apply() to avoid an early m_pullup().
2032 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2035 * Step 4: Update MD5 hash with shared secret.
2037 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2038 MD5Final(buf, &ctx);
2040 key_sa_recordxfer(sav, m);
2044 #endif /* TCP_SIGNATURE */
2047 sysctl_drop(SYSCTL_HANDLER_ARGS)
2049 INIT_VNET_INET(curvnet);
2051 INIT_VNET_INET6(curvnet);
2053 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2054 struct sockaddr_storage addrs[2];
2058 struct sockaddr_in *fin, *lin;
2060 struct sockaddr_in6 *fin6, *lin6;
2061 struct in6_addr f6, l6;
2072 if (req->oldptr != NULL || req->oldlen != 0)
2074 if (req->newptr == NULL)
2076 if (req->newlen < sizeof(addrs))
2078 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2082 switch (addrs[0].ss_family) {
2085 fin6 = (struct sockaddr_in6 *)&addrs[0];
2086 lin6 = (struct sockaddr_in6 *)&addrs[1];
2087 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2088 lin6->sin6_len != sizeof(struct sockaddr_in6))
2090 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2091 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2093 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2094 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2095 fin = (struct sockaddr_in *)&addrs[0];
2096 lin = (struct sockaddr_in *)&addrs[1];
2099 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2102 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2108 fin = (struct sockaddr_in *)&addrs[0];
2109 lin = (struct sockaddr_in *)&addrs[1];
2110 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2111 lin->sin_len != sizeof(struct sockaddr_in))
2117 INP_INFO_WLOCK(&V_tcbinfo);
2118 switch (addrs[0].ss_family) {
2121 inp = in6_pcblookup_hash(&V_tcbinfo, &f6, fin6->sin6_port,
2122 &l6, lin6->sin6_port, 0, NULL);
2126 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2127 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2132 if (inp->inp_vflag & INP_TIMEWAIT) {
2134 * XXXRW: There currently exists a state where an
2135 * inpcb is present, but its timewait state has been
2136 * discarded. For now, don't allow dropping of this
2144 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2145 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2146 tp = intotcpcb(inp);
2147 tp = tcp_drop(tp, ECONNABORTED);
2154 INP_INFO_WUNLOCK(&V_tcbinfo);
2158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2159 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2160 0, sysctl_drop, "", "Drop TCP connection");
2163 * Generate a standardized TCP log line for use throughout the
2164 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2165 * allow use in the interrupt context.
2167 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2168 * NB: The function may return NULL if memory allocation failed.
2170 * Due to header inclusion and ordering limitations the struct ip
2171 * and ip6_hdr pointers have to be passed as void pointers.
2174 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2181 const struct ip6_hdr *ip6;
2183 ip6 = (const struct ip6_hdr *)ip6hdr;
2185 ip = (struct ip *)ip4hdr;
2188 * The log line looks like this:
2189 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2191 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2192 sizeof(PRINT_TH_FLAGS) + 1 +
2194 2 * INET6_ADDRSTRLEN;
2196 2 * INET_ADDRSTRLEN;
2199 /* Is logging enabled? */
2200 if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2203 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2207 strcat(s, "TCP: [");
2210 if (inc && inc->inc_isipv6 == 0) {
2211 inet_ntoa_r(inc->inc_faddr, sp);
2213 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2215 inet_ntoa_r(inc->inc_laddr, sp);
2217 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2220 ip6_sprintf(sp, &inc->inc6_faddr);
2222 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2224 ip6_sprintf(sp, &inc->inc6_laddr);
2226 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2227 } else if (ip6 && th) {
2228 ip6_sprintf(sp, &ip6->ip6_src);
2230 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2232 ip6_sprintf(sp, &ip6->ip6_dst);
2234 sprintf(sp, "]:%i", ntohs(th->th_dport));
2236 } else if (ip && th) {
2237 inet_ntoa_r(ip->ip_src, sp);
2239 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2241 inet_ntoa_r(ip->ip_dst, sp);
2243 sprintf(sp, "]:%i", ntohs(th->th_dport));
2250 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2251 if (*(s + size - 1) != '\0')
2252 panic("%s: string too long", __func__);