2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #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>
61 #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 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
116 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
125 error = sysctl_handle_int(oidp, &new, 0, req);
126 if (error == 0 && req->newptr) {
127 if (new < TCP_MINMSS)
135 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
136 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
137 &sysctl_net_inet_tcp_mss_check, "I",
138 "Default TCP Maximum Segment Size");
142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
146 new = V_tcp_v6mssdflt;
147 error = sysctl_handle_int(oidp, &new, 0, req);
148 if (error == 0 && req->newptr) {
149 if (new < TCP_MINMSS)
152 V_tcp_v6mssdflt = new;
157 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
158 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
159 &sysctl_net_inet_tcp_mss_v6_check, "I",
160 "Default TCP Maximum Segment Size for IPv6");
164 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
167 VNET_SYSCTL_ARG(req, arg1);
168 return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
172 * Minimum MSS we accept and use. This prevents DoS attacks where
173 * we are forced to a ridiculous low MSS like 20 and send hundreds
174 * of packets instead of one. The effect scales with the available
175 * bandwidth and quickly saturates the CPU and network interface
176 * with packet generation and sending. Set to zero to disable MINMSS
177 * checking. This setting prevents us from sending too small packets.
179 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
180 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
181 &VNET_NAME(tcp_minmss), 0,
182 "Minmum TCP Maximum Segment Size");
184 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
186 &VNET_NAME(tcp_do_rfc1323), 0,
187 "Enable rfc1323 (high performance TCP) extensions");
189 static int tcp_log_debug = 0;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
191 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
193 static int tcp_tcbhashsize = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
195 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
197 static int do_tcpdrain = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
199 "Enable tcp_drain routine for extra help when low on mbufs");
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
202 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
204 static VNET_DEFINE(int, icmp_may_rst) = 1;
205 #define V_icmp_may_rst VNET(icmp_may_rst)
206 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
207 &VNET_NAME(icmp_may_rst), 0,
208 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
210 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
211 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
212 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
213 &VNET_NAME(tcp_isn_reseed_interval), 0,
214 "Seconds between reseeding of ISN secret");
217 * TCP bandwidth limiting sysctls. Note that the default lower bound of
218 * 1024 exists only for debugging. A good production default would be
219 * something like 6100.
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222 "TCP inflight data limiting");
224 static VNET_DEFINE(int, tcp_inflight_enable) = 0;
225 #define V_tcp_inflight_enable VNET(tcp_inflight_enable)
226 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
227 &VNET_NAME(tcp_inflight_enable), 0,
228 "Enable automatic TCP inflight data limiting");
230 static int tcp_inflight_debug = 0;
231 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
232 &tcp_inflight_debug, 0,
233 "Debug TCP inflight calculations");
235 static VNET_DEFINE(int, tcp_inflight_rttthresh);
236 #define V_tcp_inflight_rttthresh VNET(tcp_inflight_rttthresh)
237 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
238 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
239 vnet_sysctl_msec_to_ticks, "I",
240 "RTT threshold below which inflight will deactivate itself");
242 static VNET_DEFINE(int, tcp_inflight_min) = 6144;
243 #define V_tcp_inflight_min VNET(tcp_inflight_min)
244 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
245 &VNET_NAME(tcp_inflight_min), 0,
246 "Lower-bound for TCP inflight window");
248 static VNET_DEFINE(int, tcp_inflight_max) = TCP_MAXWIN << TCP_MAX_WINSHIFT;
249 #define V_tcp_inflight_max VNET(tcp_inflight_max)
250 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
251 &VNET_NAME(tcp_inflight_max), 0,
252 "Upper-bound for TCP inflight window");
254 static VNET_DEFINE(int, tcp_inflight_stab) = 20;
255 #define V_tcp_inflight_stab VNET(tcp_inflight_stab)
256 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
257 &VNET_NAME(tcp_inflight_stab), 0,
258 "Inflight Algorithm Stabilization 20 = 2 packets");
260 VNET_DEFINE(uma_zone_t, sack_hole_zone);
261 #define V_sack_hole_zone VNET(sack_hole_zone)
263 static struct inpcb *tcp_notify(struct inpcb *, int);
264 static void tcp_isn_tick(void *);
265 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
266 void *ip4hdr, const void *ip6hdr);
269 * Target size of TCP PCB hash tables. Must be a power of two.
271 * Note that this can be overridden by the kernel environment
272 * variable net.inet.tcp.tcbhashsize
275 #define TCBHASHSIZE 512
280 * Callouts should be moved into struct tcp directly. They are currently
281 * separate because the tcpcb structure is exported to userland for sysctl
282 * parsing purposes, which do not know about callouts.
289 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
290 #define V_tcpcb_zone VNET(tcpcb_zone)
292 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
293 struct callout isn_callout;
294 static struct mtx isn_mtx;
296 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
297 #define ISN_LOCK() mtx_lock(&isn_mtx)
298 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
301 * TCP initialization.
304 tcp_zone_change(void *tag)
307 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
308 uma_zone_set_max(V_tcpcb_zone, maxsockets);
309 tcp_tw_zone_change();
313 tcp_inpcb_init(void *mem, int size, int flags)
315 struct inpcb *inp = mem;
317 INP_LOCK_INIT(inp, "inp", "tcpinp");
326 INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
329 V_tcbinfo.ipi_vnet = curvnet;
331 V_tcbinfo.ipi_listhead = &V_tcb;
332 hashsize = TCBHASHSIZE;
333 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
334 if (!powerof2(hashsize)) {
335 printf("WARNING: TCB hash size not a power of 2\n");
336 hashsize = 512; /* safe default */
338 V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
339 &V_tcbinfo.ipi_hashmask);
340 V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
341 &V_tcbinfo.ipi_porthashmask);
342 V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
343 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
344 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
345 V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
348 * These have to be type stable for the benefit of the timers.
350 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
351 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
352 uma_zone_set_max(V_tcpcb_zone, maxsockets);
359 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
360 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
361 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
363 /* Skip initialization of globals for non-default instances. */
364 if (!IS_DEFAULT_VNET(curvnet))
367 /* XXX virtualize those bellow? */
368 tcp_delacktime = TCPTV_DELACK;
369 tcp_keepinit = TCPTV_KEEP_INIT;
370 tcp_keepidle = TCPTV_KEEP_IDLE;
371 tcp_keepintvl = TCPTV_KEEPINTVL;
372 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
374 tcp_rexmit_min = TCPTV_MIN;
375 if (tcp_rexmit_min < 1)
377 tcp_rexmit_slop = TCPTV_CPU_VAR;
378 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
379 tcp_tcbhashsize = hashsize;
382 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
384 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
386 if (max_protohdr < TCP_MINPROTOHDR)
387 max_protohdr = TCP_MINPROTOHDR;
388 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
390 #undef TCP_MINPROTOHDR
393 callout_init(&isn_callout, CALLOUT_MPSAFE);
394 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
395 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
396 SHUTDOWN_PRI_DEFAULT);
397 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
398 EVENTHANDLER_PRI_ANY);
411 /* XXX check that hashes are empty! */
412 hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
413 V_tcbinfo.ipi_hashmask);
414 hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
415 V_tcbinfo.ipi_porthashmask);
417 uma_zdestroy(V_sack_hole_zone);
418 uma_zdestroy(V_tcpcb_zone);
419 uma_zdestroy(V_tcbinfo.ipi_zone);
421 INP_INFO_LOCK_DESTROY(&V_tcbinfo);
429 callout_stop(&isn_callout);
433 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
434 * tcp_template used to store this data in mbufs, but we now recopy it out
435 * of the tcpcb each time to conserve mbufs.
438 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
440 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
442 INP_WLOCK_ASSERT(inp);
445 if ((inp->inp_vflag & INP_IPV6) != 0) {
448 ip6 = (struct ip6_hdr *)ip_ptr;
449 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
450 (inp->inp_flow & IPV6_FLOWINFO_MASK);
451 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
452 (IPV6_VERSION & IPV6_VERSION_MASK);
453 ip6->ip6_nxt = IPPROTO_TCP;
454 ip6->ip6_plen = htons(sizeof(struct tcphdr));
455 ip6->ip6_src = inp->in6p_laddr;
456 ip6->ip6_dst = inp->in6p_faddr;
462 ip = (struct ip *)ip_ptr;
463 ip->ip_v = IPVERSION;
465 ip->ip_tos = inp->inp_ip_tos;
469 ip->ip_ttl = inp->inp_ip_ttl;
471 ip->ip_p = IPPROTO_TCP;
472 ip->ip_src = inp->inp_laddr;
473 ip->ip_dst = inp->inp_faddr;
475 th->th_sport = inp->inp_lport;
476 th->th_dport = inp->inp_fport;
484 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
488 * Create template to be used to send tcp packets on a connection.
489 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
490 * use for this function is in keepalives, which use tcp_respond.
493 tcpip_maketemplate(struct inpcb *inp)
497 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
500 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
505 * Send a single message to the TCP at address specified by
506 * the given TCP/IP header. If m == NULL, then we make a copy
507 * of the tcpiphdr at ti and send directly to the addressed host.
508 * This is used to force keep alive messages out using the TCP
509 * template for a connection. If flags are given then we send
510 * a message back to the TCP which originated the * segment ti,
511 * and discard the mbuf containing it and any other attached mbufs.
513 * In any case the ack and sequence number of the transmitted
514 * segment are as specified by the parameters.
516 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
519 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
520 tcp_seq ack, tcp_seq seq, int flags)
533 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
536 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
543 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
544 INP_WLOCK_ASSERT(inp);
549 if (!(flags & TH_RST)) {
550 win = sbspace(&inp->inp_socket->so_rcv);
551 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
552 win = (long)TCP_MAXWIN << tp->rcv_scale;
556 m = m_gethdr(M_DONTWAIT, MT_DATA);
560 m->m_data += max_linkhdr;
563 bcopy((caddr_t)ip6, mtod(m, caddr_t),
564 sizeof(struct ip6_hdr));
565 ip6 = mtod(m, struct ip6_hdr *);
566 nth = (struct tcphdr *)(ip6 + 1);
570 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
571 ip = mtod(m, struct ip *);
572 nth = (struct tcphdr *)(ip + 1);
574 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
579 * XXX MRT We inherrit the FIB, which is lucky.
583 m->m_data = (caddr_t)ipgen;
584 /* m_len is set later */
586 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
589 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
590 nth = (struct tcphdr *)(ip6 + 1);
594 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
595 nth = (struct tcphdr *)(ip + 1);
599 * this is usually a case when an extension header
600 * exists between the IPv6 header and the
603 nth->th_sport = th->th_sport;
604 nth->th_dport = th->th_dport;
606 xchg(nth->th_dport, nth->th_sport, uint16_t);
612 ip6->ip6_vfc = IPV6_VERSION;
613 ip6->ip6_nxt = IPPROTO_TCP;
614 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
616 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
620 tlen += sizeof (struct tcpiphdr);
622 ip->ip_ttl = V_ip_defttl;
623 if (V_path_mtu_discovery)
627 m->m_pkthdr.len = tlen;
628 m->m_pkthdr.rcvif = NULL;
632 * Packet is associated with a socket, so allow the
633 * label of the response to reflect the socket label.
635 INP_WLOCK_ASSERT(inp);
636 mac_inpcb_create_mbuf(inp, m);
639 * Packet is not associated with a socket, so possibly
640 * update the label in place.
642 mac_netinet_tcp_reply(m);
645 nth->th_seq = htonl(seq);
646 nth->th_ack = htonl(ack);
648 nth->th_off = sizeof (struct tcphdr) >> 2;
649 nth->th_flags = flags;
651 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
653 nth->th_win = htons((u_short)win);
658 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
659 sizeof(struct ip6_hdr),
660 tlen - sizeof(struct ip6_hdr));
661 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
666 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
667 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
668 m->m_pkthdr.csum_flags = CSUM_TCP;
669 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
672 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
673 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
677 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
680 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
684 * Create a new TCP control block, making an
685 * empty reassembly queue and hooking it to the argument
686 * protocol control block. The `inp' parameter must have
687 * come from the zone allocator set up in tcp_init().
690 tcp_newtcpcb(struct inpcb *inp)
692 struct tcpcb_mem *tm;
695 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
698 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
703 tp->t_vnet = inp->inp_vnet;
705 tp->t_timers = &tm->tt;
706 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
707 tp->t_maxseg = tp->t_maxopd =
709 isipv6 ? V_tcp_v6mssdflt :
713 /* Set up our timeouts. */
714 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
715 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
716 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
717 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
718 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
720 if (V_tcp_do_rfc1323)
721 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
723 tp->t_flags |= TF_SACK_PERMIT;
724 TAILQ_INIT(&tp->snd_holes);
725 tp->t_inpcb = inp; /* XXX */
727 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
728 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
729 * reasonable initial retransmit time.
731 tp->t_srtt = TCPTV_SRTTBASE;
732 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
733 tp->t_rttmin = tcp_rexmit_min;
734 tp->t_rxtcur = TCPTV_RTOBASE;
735 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
736 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
737 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
738 tp->t_rcvtime = ticks;
739 tp->t_bw_rtttime = ticks;
741 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
742 * because the socket may be bound to an IPv6 wildcard address,
743 * which may match an IPv4-mapped IPv6 address.
745 inp->inp_ip_ttl = V_ip_defttl;
747 return (tp); /* XXX */
751 * Drop a TCP connection, reporting
752 * the specified error. If connection is synchronized,
753 * then send a RST to peer.
756 tcp_drop(struct tcpcb *tp, int errno)
758 struct socket *so = tp->t_inpcb->inp_socket;
760 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
761 INP_WLOCK_ASSERT(tp->t_inpcb);
763 if (TCPS_HAVERCVDSYN(tp->t_state)) {
764 tp->t_state = TCPS_CLOSED;
765 (void) tcp_output_reset(tp);
766 TCPSTAT_INC(tcps_drops);
768 TCPSTAT_INC(tcps_conndrops);
769 if (errno == ETIMEDOUT && tp->t_softerror)
770 errno = tp->t_softerror;
771 so->so_error = errno;
772 return (tcp_close(tp));
776 tcp_discardcb(struct tcpcb *tp)
778 struct inpcb *inp = tp->t_inpcb;
779 struct socket *so = inp->inp_socket;
781 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
784 INP_WLOCK_ASSERT(inp);
787 * Make sure that all of our timers are stopped before we
790 callout_stop(&tp->t_timers->tt_rexmt);
791 callout_stop(&tp->t_timers->tt_persist);
792 callout_stop(&tp->t_timers->tt_keep);
793 callout_stop(&tp->t_timers->tt_2msl);
794 callout_stop(&tp->t_timers->tt_delack);
797 * If we got enough samples through the srtt filter,
798 * save the rtt and rttvar in the routing entry.
799 * 'Enough' is arbitrarily defined as 4 rtt samples.
800 * 4 samples is enough for the srtt filter to converge
801 * to within enough % of the correct value; fewer samples
802 * and we could save a bogus rtt. The danger is not high
803 * as tcp quickly recovers from everything.
804 * XXX: Works very well but needs some more statistics!
806 if (tp->t_rttupdated >= 4) {
807 struct hc_metrics_lite metrics;
810 bzero(&metrics, sizeof(metrics));
812 * Update the ssthresh always when the conditions below
813 * are satisfied. This gives us better new start value
814 * for the congestion avoidance for new connections.
815 * ssthresh is only set if packet loss occured on a session.
817 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
818 * being torn down. Ideally this code would not use 'so'.
820 ssthresh = tp->snd_ssthresh;
821 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
823 * convert the limit from user data bytes to
824 * packets then to packet data bytes.
826 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
829 ssthresh *= (u_long)(tp->t_maxseg +
831 (isipv6 ? sizeof (struct ip6_hdr) +
832 sizeof (struct tcphdr) :
834 sizeof (struct tcpiphdr)
841 metrics.rmx_ssthresh = ssthresh;
843 metrics.rmx_rtt = tp->t_srtt;
844 metrics.rmx_rttvar = tp->t_rttvar;
845 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
846 metrics.rmx_bandwidth = tp->snd_bandwidth;
847 metrics.rmx_cwnd = tp->snd_cwnd;
848 metrics.rmx_sendpipe = 0;
849 metrics.rmx_recvpipe = 0;
851 tcp_hc_update(&inp->inp_inc, &metrics);
854 /* free the reassembly queue, if any */
856 /* Disconnect offload device, if any. */
857 tcp_offload_detach(tp);
859 tcp_free_sackholes(tp);
860 inp->inp_ppcb = NULL;
862 uma_zfree(V_tcpcb_zone, tp);
866 * Attempt to close a TCP control block, marking it as dropped, and freeing
867 * the socket if we hold the only reference.
870 tcp_close(struct tcpcb *tp)
872 struct inpcb *inp = tp->t_inpcb;
875 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
876 INP_WLOCK_ASSERT(inp);
878 /* Notify any offload devices of listener close */
879 if (tp->t_state == TCPS_LISTEN)
880 tcp_offload_listen_close(tp);
882 TCPSTAT_INC(tcps_closed);
883 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
884 so = inp->inp_socket;
885 soisdisconnected(so);
886 if (inp->inp_flags & INP_SOCKREF) {
887 KASSERT(so->so_state & SS_PROTOREF,
888 ("tcp_close: !SS_PROTOREF"));
889 inp->inp_flags &= ~INP_SOCKREF;
893 so->so_state &= ~SS_PROTOREF;
903 VNET_ITERATOR_DECL(vnet_iter);
908 VNET_LIST_RLOCK_NOSLEEP();
909 VNET_FOREACH(vnet_iter) {
910 CURVNET_SET(vnet_iter);
915 * Walk the tcpbs, if existing, and flush the reassembly queue,
917 * XXX: The "Net/3" implementation doesn't imply that the TCP
918 * reassembly queue should be flushed, but in a situation
919 * where we're really low on mbufs, this is potentially
922 INP_INFO_RLOCK(&V_tcbinfo);
923 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
924 if (inpb->inp_flags & INP_TIMEWAIT)
927 if ((tcpb = intotcpcb(inpb)) != NULL) {
928 tcp_reass_flush(tcpb);
929 tcp_clean_sackreport(tcpb);
933 INP_INFO_RUNLOCK(&V_tcbinfo);
936 VNET_LIST_RUNLOCK_NOSLEEP();
940 * Notify a tcp user of an asynchronous error;
941 * store error as soft error, but wake up user
942 * (for now, won't do anything until can select for soft error).
944 * Do not wake up user since there currently is no mechanism for
945 * reporting soft errors (yet - a kqueue filter may be added).
947 static struct inpcb *
948 tcp_notify(struct inpcb *inp, int error)
952 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
953 INP_WLOCK_ASSERT(inp);
955 if ((inp->inp_flags & INP_TIMEWAIT) ||
956 (inp->inp_flags & INP_DROPPED))
960 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
963 * Ignore some errors if we are hooked up.
964 * If connection hasn't completed, has retransmitted several times,
965 * and receives a second error, give up now. This is better
966 * than waiting a long time to establish a connection that
967 * can never complete.
969 if (tp->t_state == TCPS_ESTABLISHED &&
970 (error == EHOSTUNREACH || error == ENETUNREACH ||
971 error == EHOSTDOWN)) {
973 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
975 tp = tcp_drop(tp, error);
981 tp->t_softerror = error;
985 wakeup( &so->so_timeo);
992 tcp_pcblist(SYSCTL_HANDLER_ARGS)
994 int error, i, m, n, pcb_count;
995 struct inpcb *inp, **inp_list;
1000 * The process of preparing the TCB list is too time-consuming and
1001 * resource-intensive to repeat twice on every request.
1003 if (req->oldptr == NULL) {
1004 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1005 n += imax(n / 8, 10);
1006 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1010 if (req->newptr != NULL)
1014 * OK, now we're committed to doing something.
1016 INP_INFO_RLOCK(&V_tcbinfo);
1017 gencnt = V_tcbinfo.ipi_gencnt;
1018 n = V_tcbinfo.ipi_count;
1019 INP_INFO_RUNLOCK(&V_tcbinfo);
1021 m = syncache_pcbcount();
1023 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1024 + (n + m) * sizeof(struct xtcpcb));
1028 xig.xig_len = sizeof xig;
1029 xig.xig_count = n + m;
1030 xig.xig_gen = gencnt;
1031 xig.xig_sogen = so_gencnt;
1032 error = SYSCTL_OUT(req, &xig, sizeof xig);
1036 error = syncache_pcblist(req, m, &pcb_count);
1040 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1041 if (inp_list == NULL)
1044 INP_INFO_RLOCK(&V_tcbinfo);
1045 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1046 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1048 if (inp->inp_gencnt <= gencnt) {
1050 * XXX: This use of cr_cansee(), introduced with
1051 * TCP state changes, is not quite right, but for
1052 * now, better than nothing.
1054 if (inp->inp_flags & INP_TIMEWAIT) {
1055 if (intotw(inp) != NULL)
1056 error = cr_cansee(req->td->td_ucred,
1057 intotw(inp)->tw_cred);
1059 error = EINVAL; /* Skip this inp. */
1061 error = cr_canseeinpcb(req->td->td_ucred, inp);
1064 inp_list[i++] = inp;
1069 INP_INFO_RUNLOCK(&V_tcbinfo);
1073 for (i = 0; i < n; i++) {
1076 if (inp->inp_gencnt <= gencnt) {
1080 bzero(&xt, sizeof(xt));
1081 xt.xt_len = sizeof xt;
1082 /* XXX should avoid extra copy */
1083 bcopy(inp, &xt.xt_inp, sizeof *inp);
1084 inp_ppcb = inp->inp_ppcb;
1085 if (inp_ppcb == NULL)
1086 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1087 else if (inp->inp_flags & INP_TIMEWAIT) {
1088 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1089 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1091 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1092 if (inp->inp_socket != NULL)
1093 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1095 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1096 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1098 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1100 error = SYSCTL_OUT(req, &xt, sizeof xt);
1104 INP_INFO_WLOCK(&V_tcbinfo);
1105 for (i = 0; i < n; i++) {
1108 if (!in_pcbrele(inp))
1111 INP_INFO_WUNLOCK(&V_tcbinfo);
1115 * Give the user an updated idea of our state.
1116 * If the generation differs from what we told
1117 * her before, she knows that something happened
1118 * while we were processing this request, and it
1119 * might be necessary to retry.
1121 INP_INFO_RLOCK(&V_tcbinfo);
1122 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1123 xig.xig_sogen = so_gencnt;
1124 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1125 INP_INFO_RUNLOCK(&V_tcbinfo);
1126 error = SYSCTL_OUT(req, &xig, sizeof xig);
1128 free(inp_list, M_TEMP);
1132 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1133 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1136 tcp_getcred(SYSCTL_HANDLER_ARGS)
1139 struct sockaddr_in addrs[2];
1143 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1146 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1149 INP_INFO_RLOCK(&V_tcbinfo);
1150 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1151 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1154 INP_INFO_RUNLOCK(&V_tcbinfo);
1155 if (inp->inp_socket == NULL)
1158 error = cr_canseeinpcb(req->td->td_ucred, inp);
1160 cru2x(inp->inp_cred, &xuc);
1163 INP_INFO_RUNLOCK(&V_tcbinfo);
1167 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1171 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1172 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1173 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1177 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1180 struct sockaddr_in6 addrs[2];
1182 int error, mapped = 0;
1184 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1187 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1190 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1191 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1194 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1195 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1201 INP_INFO_RLOCK(&V_tcbinfo);
1203 inp = in_pcblookup_hash(&V_tcbinfo,
1204 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1206 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1210 inp = in6_pcblookup_hash(&V_tcbinfo,
1211 &addrs[1].sin6_addr, addrs[1].sin6_port,
1212 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1215 INP_INFO_RUNLOCK(&V_tcbinfo);
1216 if (inp->inp_socket == NULL)
1219 error = cr_canseeinpcb(req->td->td_ucred, inp);
1221 cru2x(inp->inp_cred, &xuc);
1224 INP_INFO_RUNLOCK(&V_tcbinfo);
1228 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1232 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1233 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1234 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1239 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1241 struct ip *ip = vip;
1243 struct in_addr faddr;
1246 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1248 struct in_conninfo inc;
1249 tcp_seq icmp_tcp_seq;
1252 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1253 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1256 if (cmd == PRC_MSGSIZE)
1257 notify = tcp_mtudisc;
1258 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1259 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1260 notify = tcp_drop_syn_sent;
1262 * Redirects don't need to be handled up here.
1264 else if (PRC_IS_REDIRECT(cmd))
1267 * Source quench is depreciated.
1269 else if (cmd == PRC_QUENCH)
1272 * Hostdead is ugly because it goes linearly through all PCBs.
1273 * XXX: We never get this from ICMP, otherwise it makes an
1274 * excellent DoS attack on machines with many connections.
1276 else if (cmd == PRC_HOSTDEAD)
1278 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1281 icp = (struct icmp *)((caddr_t)ip
1282 - offsetof(struct icmp, icmp_ip));
1283 th = (struct tcphdr *)((caddr_t)ip
1284 + (ip->ip_hl << 2));
1285 INP_INFO_WLOCK(&V_tcbinfo);
1286 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1287 ip->ip_src, th->th_sport, 0, NULL);
1290 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1291 !(inp->inp_flags & INP_DROPPED) &&
1292 !(inp->inp_socket == NULL)) {
1293 icmp_tcp_seq = htonl(th->th_seq);
1294 tp = intotcpcb(inp);
1295 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1296 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1297 if (cmd == PRC_MSGSIZE) {
1300 * If we got a needfrag set the MTU
1301 * in the route to the suggested new
1302 * value (if given) and then notify.
1304 bzero(&inc, sizeof(inc));
1305 inc.inc_faddr = faddr;
1307 inp->inp_inc.inc_fibnum;
1309 mtu = ntohs(icp->icmp_nextmtu);
1311 * If no alternative MTU was
1312 * proposed, try the next smaller
1313 * one. ip->ip_len has already
1314 * been swapped in icmp_input().
1317 mtu = ip_next_mtu(ip->ip_len,
1319 if (mtu < V_tcp_minmss
1320 + sizeof(struct tcpiphdr))
1322 + sizeof(struct tcpiphdr);
1324 * Only cache the the MTU if it
1325 * is smaller than the interface
1326 * or route MTU. tcp_mtudisc()
1327 * will do right thing by itself.
1329 if (mtu <= tcp_maxmtu(&inc, NULL))
1330 tcp_hc_updatemtu(&inc, mtu);
1333 inp = (*notify)(inp, inetctlerrmap[cmd]);
1339 bzero(&inc, sizeof(inc));
1340 inc.inc_fport = th->th_dport;
1341 inc.inc_lport = th->th_sport;
1342 inc.inc_faddr = faddr;
1343 inc.inc_laddr = ip->ip_src;
1344 syncache_unreach(&inc, th);
1346 INP_INFO_WUNLOCK(&V_tcbinfo);
1348 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1353 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1356 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1357 struct ip6_hdr *ip6;
1359 struct ip6ctlparam *ip6cp = NULL;
1360 const struct sockaddr_in6 *sa6_src = NULL;
1362 struct tcp_portonly {
1367 if (sa->sa_family != AF_INET6 ||
1368 sa->sa_len != sizeof(struct sockaddr_in6))
1371 if (cmd == PRC_MSGSIZE)
1372 notify = tcp_mtudisc;
1373 else if (!PRC_IS_REDIRECT(cmd) &&
1374 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1376 /* Source quench is depreciated. */
1377 else if (cmd == PRC_QUENCH)
1380 /* if the parameter is from icmp6, decode it. */
1382 ip6cp = (struct ip6ctlparam *)d;
1384 ip6 = ip6cp->ip6c_ip6;
1385 off = ip6cp->ip6c_off;
1386 sa6_src = ip6cp->ip6c_src;
1390 off = 0; /* fool gcc */
1395 struct in_conninfo inc;
1397 * XXX: We assume that when IPV6 is non NULL,
1398 * M and OFF are valid.
1401 /* check if we can safely examine src and dst ports */
1402 if (m->m_pkthdr.len < off + sizeof(*thp))
1405 bzero(&th, sizeof(th));
1406 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1408 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1409 (struct sockaddr *)ip6cp->ip6c_src,
1410 th.th_sport, cmd, NULL, notify);
1412 bzero(&inc, sizeof(inc));
1413 inc.inc_fport = th.th_dport;
1414 inc.inc_lport = th.th_sport;
1415 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1416 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1417 inc.inc_flags |= INC_ISIPV6;
1418 INP_INFO_WLOCK(&V_tcbinfo);
1419 syncache_unreach(&inc, &th);
1420 INP_INFO_WUNLOCK(&V_tcbinfo);
1422 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1423 0, cmd, NULL, notify);
1429 * Following is where TCP initial sequence number generation occurs.
1431 * There are two places where we must use initial sequence numbers:
1432 * 1. In SYN-ACK packets.
1433 * 2. In SYN packets.
1435 * All ISNs for SYN-ACK packets are generated by the syncache. See
1436 * tcp_syncache.c for details.
1438 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1439 * depends on this property. In addition, these ISNs should be
1440 * unguessable so as to prevent connection hijacking. To satisfy
1441 * the requirements of this situation, the algorithm outlined in
1442 * RFC 1948 is used, with only small modifications.
1444 * Implementation details:
1446 * Time is based off the system timer, and is corrected so that it
1447 * increases by one megabyte per second. This allows for proper
1448 * recycling on high speed LANs while still leaving over an hour
1451 * As reading the *exact* system time is too expensive to be done
1452 * whenever setting up a TCP connection, we increment the time
1453 * offset in two ways. First, a small random positive increment
1454 * is added to isn_offset for each connection that is set up.
1455 * Second, the function tcp_isn_tick fires once per clock tick
1456 * and increments isn_offset as necessary so that sequence numbers
1457 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1458 * random positive increments serve only to ensure that the same
1459 * exact sequence number is never sent out twice (as could otherwise
1460 * happen when a port is recycled in less than the system tick
1463 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1464 * between seeding of isn_secret. This is normally set to zero,
1465 * as reseeding should not be necessary.
1467 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1468 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1469 * general, this means holding an exclusive (write) lock.
1472 #define ISN_BYTES_PER_SECOND 1048576
1473 #define ISN_STATIC_INCREMENT 4096
1474 #define ISN_RANDOM_INCREMENT (4096 - 1)
1476 static VNET_DEFINE(u_char, isn_secret[32]);
1477 static VNET_DEFINE(int, isn_last_reseed);
1478 static VNET_DEFINE(u_int32_t, isn_offset);
1479 static VNET_DEFINE(u_int32_t, isn_offset_old);
1481 #define V_isn_secret VNET(isn_secret)
1482 #define V_isn_last_reseed VNET(isn_last_reseed)
1483 #define V_isn_offset VNET(isn_offset)
1484 #define V_isn_offset_old VNET(isn_offset_old)
1487 tcp_new_isn(struct tcpcb *tp)
1490 u_int32_t md5_buffer[4];
1493 INP_WLOCK_ASSERT(tp->t_inpcb);
1496 /* Seed if this is the first use, reseed if requested. */
1497 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1498 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1500 read_random(&V_isn_secret, sizeof(V_isn_secret));
1501 V_isn_last_reseed = ticks;
1504 /* Compute the md5 hash and return the ISN. */
1506 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1507 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1509 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1510 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1511 sizeof(struct in6_addr));
1512 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1513 sizeof(struct in6_addr));
1517 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1518 sizeof(struct in_addr));
1519 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1520 sizeof(struct in_addr));
1522 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1523 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1524 new_isn = (tcp_seq) md5_buffer[0];
1525 V_isn_offset += ISN_STATIC_INCREMENT +
1526 (arc4random() & ISN_RANDOM_INCREMENT);
1527 new_isn += V_isn_offset;
1533 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1534 * to keep time flowing at a relatively constant rate. If the random
1535 * increments have already pushed us past the projected offset, do nothing.
1538 tcp_isn_tick(void *xtp)
1540 VNET_ITERATOR_DECL(vnet_iter);
1541 u_int32_t projected_offset;
1543 VNET_LIST_RLOCK_NOSLEEP();
1545 VNET_FOREACH(vnet_iter) {
1546 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1548 V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1550 if (SEQ_GT(projected_offset, V_isn_offset))
1551 V_isn_offset = projected_offset;
1553 V_isn_offset_old = V_isn_offset;
1557 VNET_LIST_RUNLOCK_NOSLEEP();
1558 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1562 * When a specific ICMP unreachable message is received and the
1563 * connection state is SYN-SENT, drop the connection. This behavior
1564 * is controlled by the icmp_may_rst sysctl.
1567 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1571 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1572 INP_WLOCK_ASSERT(inp);
1574 if ((inp->inp_flags & INP_TIMEWAIT) ||
1575 (inp->inp_flags & INP_DROPPED))
1578 tp = intotcpcb(inp);
1579 if (tp->t_state != TCPS_SYN_SENT)
1582 tp = tcp_drop(tp, errno);
1590 * When `need fragmentation' ICMP is received, update our idea of the MSS
1591 * based on the new value in the route. Also nudge TCP to send something,
1592 * since we know the packet we just sent was dropped.
1593 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1596 tcp_mtudisc(struct inpcb *inp, int errno)
1601 INP_WLOCK_ASSERT(inp);
1602 if ((inp->inp_flags & INP_TIMEWAIT) ||
1603 (inp->inp_flags & INP_DROPPED))
1606 tp = intotcpcb(inp);
1607 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1609 tcp_mss_update(tp, -1, NULL, NULL);
1611 so = inp->inp_socket;
1612 SOCKBUF_LOCK(&so->so_snd);
1613 /* If the mss is larger than the socket buffer, decrease the mss. */
1614 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1615 tp->t_maxseg = so->so_snd.sb_hiwat;
1616 SOCKBUF_UNLOCK(&so->so_snd);
1618 TCPSTAT_INC(tcps_mturesent);
1620 tp->snd_nxt = tp->snd_una;
1621 tcp_free_sackholes(tp);
1622 tp->snd_recover = tp->snd_max;
1623 if (tp->t_flags & TF_SACK_PERMIT)
1624 EXIT_FASTRECOVERY(tp);
1625 tcp_output_send(tp);
1630 * Look-up the routing entry to the peer of this inpcb. If no route
1631 * is found and it cannot be allocated, then return 0. This routine
1632 * is called by TCP routines that access the rmx structure and by
1633 * tcp_mss_update to get the peer/interface MTU.
1636 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1639 struct sockaddr_in *dst;
1643 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1645 bzero(&sro, sizeof(sro));
1646 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1647 dst = (struct sockaddr_in *)&sro.ro_dst;
1648 dst->sin_family = AF_INET;
1649 dst->sin_len = sizeof(*dst);
1650 dst->sin_addr = inc->inc_faddr;
1651 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1653 if (sro.ro_rt != NULL) {
1654 ifp = sro.ro_rt->rt_ifp;
1655 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1656 maxmtu = ifp->if_mtu;
1658 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1660 /* Report additional interface capabilities. */
1661 if (flags != NULL) {
1662 if (ifp->if_capenable & IFCAP_TSO4 &&
1663 ifp->if_hwassist & CSUM_TSO)
1673 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1675 struct route_in6 sro6;
1679 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1681 bzero(&sro6, sizeof(sro6));
1682 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1683 sro6.ro_dst.sin6_family = AF_INET6;
1684 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1685 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1686 rtalloc_ign((struct route *)&sro6, 0);
1688 if (sro6.ro_rt != NULL) {
1689 ifp = sro6.ro_rt->rt_ifp;
1690 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1691 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1693 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1694 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1696 /* Report additional interface capabilities. */
1697 if (flags != NULL) {
1698 if (ifp->if_capenable & IFCAP_TSO6 &&
1699 ifp->if_hwassist & CSUM_TSO)
1710 /* compute ESP/AH header size for TCP, including outer IP header. */
1712 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1719 struct ip6_hdr *ip6;
1723 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1725 MGETHDR(m, M_DONTWAIT, MT_DATA);
1730 if ((inp->inp_vflag & INP_IPV6) != 0) {
1731 ip6 = mtod(m, struct ip6_hdr *);
1732 th = (struct tcphdr *)(ip6 + 1);
1733 m->m_pkthdr.len = m->m_len =
1734 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1735 tcpip_fillheaders(inp, ip6, th);
1736 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1740 ip = mtod(m, struct ip *);
1741 th = (struct tcphdr *)(ip + 1);
1742 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1743 tcpip_fillheaders(inp, ip, th);
1744 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1753 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1755 * This code attempts to calculate the bandwidth-delay product as a
1756 * means of determining the optimal window size to maximize bandwidth,
1757 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1758 * routers. This code also does a fairly good job keeping RTTs in check
1759 * across slow links like modems. We implement an algorithm which is very
1760 * similar (but not meant to be) TCP/Vegas. The code operates on the
1761 * transmitter side of a TCP connection and so only effects the transmit
1762 * side of the connection.
1764 * BACKGROUND: TCP makes no provision for the management of buffer space
1765 * at the end points or at the intermediate routers and switches. A TCP
1766 * stream, whether using NewReno or not, will eventually buffer as
1767 * many packets as it is able and the only reason this typically works is
1768 * due to the fairly small default buffers made available for a connection
1769 * (typicaly 16K or 32K). As machines use larger windows and/or window
1770 * scaling it is now fairly easy for even a single TCP connection to blow-out
1771 * all available buffer space not only on the local interface, but on
1772 * intermediate routers and switches as well. NewReno makes a misguided
1773 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1774 * then backing off, then steadily increasing the window again until another
1775 * failure occurs, ad-infinitum. This results in terrible oscillation that
1776 * is only made worse as network loads increase and the idea of intentionally
1777 * blowing out network buffers is, frankly, a terrible way to manage network
1780 * It is far better to limit the transmit window prior to the failure
1781 * condition being achieved. There are two general ways to do this: First
1782 * you can 'scan' through different transmit window sizes and locate the
1783 * point where the RTT stops increasing, indicating that you have filled the
1784 * pipe, then scan backwards until you note that RTT stops decreasing, then
1785 * repeat ad-infinitum. This method works in principle but has severe
1786 * implementation issues due to RTT variances, timer granularity, and
1787 * instability in the algorithm which can lead to many false positives and
1788 * create oscillations as well as interact badly with other TCP streams
1789 * implementing the same algorithm.
1791 * The second method is to limit the window to the bandwidth delay product
1792 * of the link. This is the method we implement. RTT variances and our
1793 * own manipulation of the congestion window, bwnd, can potentially
1794 * destabilize the algorithm. For this reason we have to stabilize the
1795 * elements used to calculate the window. We do this by using the minimum
1796 * observed RTT, the long term average of the observed bandwidth, and
1797 * by adding two segments worth of slop. It isn't perfect but it is able
1798 * to react to changing conditions and gives us a very stable basis on
1799 * which to extend the algorithm.
1802 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1808 INP_WLOCK_ASSERT(tp->t_inpcb);
1811 * If inflight_enable is disabled in the middle of a tcp connection,
1812 * make sure snd_bwnd is effectively disabled.
1814 if (V_tcp_inflight_enable == 0 ||
1815 tp->t_rttlow < V_tcp_inflight_rttthresh) {
1816 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1817 tp->snd_bandwidth = 0;
1822 * Figure out the bandwidth. Due to the tick granularity this
1823 * is a very rough number and it MUST be averaged over a fairly
1824 * long period of time. XXX we need to take into account a link
1825 * that is not using all available bandwidth, but for now our
1826 * slop will ramp us up if this case occurs and the bandwidth later
1829 * Note: if ticks rollover 'bw' may wind up negative. We must
1830 * effectively reset t_bw_rtttime for this case.
1833 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1836 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1837 (save_ticks - tp->t_bw_rtttime);
1838 tp->t_bw_rtttime = save_ticks;
1839 tp->t_bw_rtseq = ack_seq;
1840 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1842 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1844 tp->snd_bandwidth = bw;
1847 * Calculate the semi-static bandwidth delay product, plus two maximal
1848 * segments. The additional slop puts us squarely in the sweet
1849 * spot and also handles the bandwidth run-up case and stabilization.
1850 * Without the slop we could be locking ourselves into a lower
1853 * Situations Handled:
1854 * (1) Prevents over-queueing of packets on LANs, especially on
1855 * high speed LANs, allowing larger TCP buffers to be
1856 * specified, and also does a good job preventing
1857 * over-queueing of packets over choke points like modems
1858 * (at least for the transmit side).
1860 * (2) Is able to handle changing network loads (bandwidth
1861 * drops so bwnd drops, bandwidth increases so bwnd
1864 * (3) Theoretically should stabilize in the face of multiple
1865 * connections implementing the same algorithm (this may need
1868 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1869 * be adjusted with a sysctl but typically only needs to be
1870 * on very slow connections. A value no smaller then 5
1871 * should be used, but only reduce this default if you have
1874 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1875 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1878 if (tcp_inflight_debug > 0) {
1880 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1882 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1891 if ((long)bwnd < V_tcp_inflight_min)
1892 bwnd = V_tcp_inflight_min;
1893 if (bwnd > V_tcp_inflight_max)
1894 bwnd = V_tcp_inflight_max;
1895 if ((long)bwnd < tp->t_maxseg * 2)
1896 bwnd = tp->t_maxseg * 2;
1897 tp->snd_bwnd = bwnd;
1900 #ifdef TCP_SIGNATURE
1902 * Callback function invoked by m_apply() to digest TCP segment data
1903 * contained within an mbuf chain.
1906 tcp_signature_apply(void *fstate, void *data, u_int len)
1909 MD5Update(fstate, (u_char *)data, len);
1914 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1917 * m pointer to head of mbuf chain
1919 * len length of TCP segment data, excluding options
1920 * optlen length of TCP segment options
1921 * buf pointer to storage for computed MD5 digest
1922 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1924 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1925 * When called from tcp_input(), we can be sure that th_sum has been
1926 * zeroed out and verified already.
1928 * Return 0 if successful, otherwise return -1.
1930 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1931 * search with the destination IP address, and a 'magic SPI' to be
1932 * determined by the application. This is hardcoded elsewhere to 1179
1933 * right now. Another branch of this code exists which uses the SPD to
1934 * specify per-application flows but it is unstable.
1937 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1938 u_char *buf, u_int direction)
1940 union sockaddr_union dst;
1941 struct ippseudo ippseudo;
1945 struct ipovly *ipovly;
1946 struct secasvar *sav;
1949 struct ip6_hdr *ip6;
1950 struct in6_addr in6;
1951 char ip6buf[INET6_ADDRSTRLEN];
1957 KASSERT(m != NULL, ("NULL mbuf chain"));
1958 KASSERT(buf != NULL, ("NULL signature pointer"));
1960 /* Extract the destination from the IP header in the mbuf. */
1961 bzero(&dst, sizeof(union sockaddr_union));
1962 ip = mtod(m, struct ip *);
1964 ip6 = NULL; /* Make the compiler happy. */
1968 dst.sa.sa_len = sizeof(struct sockaddr_in);
1969 dst.sa.sa_family = AF_INET;
1970 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1971 ip->ip_src : ip->ip_dst;
1974 case (IPV6_VERSION >> 4):
1975 ip6 = mtod(m, struct ip6_hdr *);
1976 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1977 dst.sa.sa_family = AF_INET6;
1978 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1979 ip6->ip6_src : ip6->ip6_dst;
1988 /* Look up an SADB entry which matches the address of the peer. */
1989 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1991 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1992 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1994 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1995 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2003 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2005 * XXX The ippseudo header MUST be digested in network byte order,
2006 * or else we'll fail the regression test. Assume all fields we've
2007 * been doing arithmetic on have been in host byte order.
2008 * XXX One cannot depend on ipovly->ih_len here. When called from
2009 * tcp_output(), the underlying ip_len member has not yet been set.
2013 ipovly = (struct ipovly *)ip;
2014 ippseudo.ippseudo_src = ipovly->ih_src;
2015 ippseudo.ippseudo_dst = ipovly->ih_dst;
2016 ippseudo.ippseudo_pad = 0;
2017 ippseudo.ippseudo_p = IPPROTO_TCP;
2018 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2020 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2022 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2023 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2027 * RFC 2385, 2.0 Proposal
2028 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2029 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2030 * extended next header value (to form 32 bits), and 32-bit segment
2032 * Note: Upper-Layer Packet Length comes before Next Header.
2034 case (IPV6_VERSION >> 4):
2036 in6_clearscope(&in6);
2037 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2039 in6_clearscope(&in6);
2040 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2041 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2042 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2044 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2045 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2046 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2048 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2050 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2051 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2062 * Step 2: Update MD5 hash with TCP header, excluding options.
2063 * The TCP checksum must be set to zero.
2065 savecsum = th->th_sum;
2067 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2068 th->th_sum = savecsum;
2071 * Step 3: Update MD5 hash with TCP segment data.
2072 * Use m_apply() to avoid an early m_pullup().
2075 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2078 * Step 4: Update MD5 hash with shared secret.
2080 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2081 MD5Final(buf, &ctx);
2083 key_sa_recordxfer(sav, m);
2087 #endif /* TCP_SIGNATURE */
2090 sysctl_drop(SYSCTL_HANDLER_ARGS)
2092 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2093 struct sockaddr_storage addrs[2];
2097 struct sockaddr_in *fin, *lin;
2099 struct sockaddr_in6 *fin6, *lin6;
2110 if (req->oldptr != NULL || req->oldlen != 0)
2112 if (req->newptr == NULL)
2114 if (req->newlen < sizeof(addrs))
2116 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2120 switch (addrs[0].ss_family) {
2123 fin6 = (struct sockaddr_in6 *)&addrs[0];
2124 lin6 = (struct sockaddr_in6 *)&addrs[1];
2125 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2126 lin6->sin6_len != sizeof(struct sockaddr_in6))
2128 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2129 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2131 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2132 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2133 fin = (struct sockaddr_in *)&addrs[0];
2134 lin = (struct sockaddr_in *)&addrs[1];
2137 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2140 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2146 fin = (struct sockaddr_in *)&addrs[0];
2147 lin = (struct sockaddr_in *)&addrs[1];
2148 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2149 lin->sin_len != sizeof(struct sockaddr_in))
2155 INP_INFO_WLOCK(&V_tcbinfo);
2156 switch (addrs[0].ss_family) {
2159 inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2160 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2165 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2166 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2171 if (inp->inp_flags & INP_TIMEWAIT) {
2173 * XXXRW: There currently exists a state where an
2174 * inpcb is present, but its timewait state has been
2175 * discarded. For now, don't allow dropping of this
2183 } else if (!(inp->inp_flags & INP_DROPPED) &&
2184 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2185 tp = intotcpcb(inp);
2186 tp = tcp_drop(tp, ECONNABORTED);
2193 INP_INFO_WUNLOCK(&V_tcbinfo);
2197 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2198 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2199 0, sysctl_drop, "", "Drop TCP connection");
2202 * Generate a standardized TCP log line for use throughout the
2203 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2204 * allow use in the interrupt context.
2206 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2207 * NB: The function may return NULL if memory allocation failed.
2209 * Due to header inclusion and ordering limitations the struct ip
2210 * and ip6_hdr pointers have to be passed as void pointers.
2213 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2217 /* Is logging enabled? */
2218 if (tcp_log_in_vain == 0)
2221 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2225 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2229 /* Is logging enabled? */
2230 if (tcp_log_debug == 0)
2233 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2237 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2244 const struct ip6_hdr *ip6;
2246 ip6 = (const struct ip6_hdr *)ip6hdr;
2248 ip = (struct ip *)ip4hdr;
2251 * The log line looks like this:
2252 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2254 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2255 sizeof(PRINT_TH_FLAGS) + 1 +
2257 2 * INET6_ADDRSTRLEN;
2259 2 * INET_ADDRSTRLEN;
2262 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2266 strcat(s, "TCP: [");
2269 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2270 inet_ntoa_r(inc->inc_faddr, sp);
2272 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2274 inet_ntoa_r(inc->inc_laddr, sp);
2276 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2279 ip6_sprintf(sp, &inc->inc6_faddr);
2281 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2283 ip6_sprintf(sp, &inc->inc6_laddr);
2285 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2286 } else if (ip6 && th) {
2287 ip6_sprintf(sp, &ip6->ip6_src);
2289 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2291 ip6_sprintf(sp, &ip6->ip6_dst);
2293 sprintf(sp, "]:%i", ntohs(th->th_dport));
2295 } else if (ip && th) {
2296 inet_ntoa_r(ip->ip_src, sp);
2298 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2300 inet_ntoa_r(ip->ip_dst, sp);
2302 sprintf(sp, "]:%i", ntohs(th->th_dport));
2309 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2310 if (*(s + size - 1) != '\0')
2311 panic("%s: string too long", __func__);