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/hhook.h>
45 #include <sys/kernel.h>
46 #include <sys/khelp.h>
47 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
52 #include <sys/domain.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/protosw.h>
59 #include <sys/random.h>
63 #include <net/route.h>
67 #include <netinet/cc.h>
68 #include <netinet/in.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/ip.h>
72 #include <netinet/ip6.h>
74 #include <netinet/in_pcb.h>
76 #include <netinet6/in6_pcb.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip_var.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #include <netinet6/nd6.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/tcp_fsm.h>
87 #include <netinet/tcp_seq.h>
88 #include <netinet/tcp_timer.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/tcp_syncache.h>
91 #include <netinet/tcp_offload.h>
93 #include <netinet6/tcp6_var.h>
95 #include <netinet/tcpip.h>
97 #include <netinet/tcp_debug.h>
99 #include <netinet6/ip6protosw.h>
102 #include <netipsec/ipsec.h>
103 #include <netipsec/xform.h>
105 #include <netipsec/ipsec6.h>
107 #include <netipsec/key.h>
108 #include <sys/syslog.h>
111 #include <machine/in_cksum.h>
114 #include <security/mac/mac_framework.h>
116 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
118 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
122 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
127 error = sysctl_handle_int(oidp, &new, 0, req);
128 if (error == 0 && req->newptr) {
129 if (new < TCP_MINMSS)
137 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
138 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
139 &sysctl_net_inet_tcp_mss_check, "I",
140 "Default TCP Maximum Segment Size");
144 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
148 new = V_tcp_v6mssdflt;
149 error = sysctl_handle_int(oidp, &new, 0, req);
150 if (error == 0 && req->newptr) {
151 if (new < TCP_MINMSS)
154 V_tcp_v6mssdflt = new;
159 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
160 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
161 &sysctl_net_inet_tcp_mss_v6_check, "I",
162 "Default TCP Maximum Segment Size for IPv6");
166 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
169 VNET_SYSCTL_ARG(req, arg1);
170 return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
174 * Minimum MSS we accept and use. This prevents DoS attacks where
175 * we are forced to a ridiculous low MSS like 20 and send hundreds
176 * of packets instead of one. The effect scales with the available
177 * bandwidth and quickly saturates the CPU and network interface
178 * with packet generation and sending. Set to zero to disable MINMSS
179 * checking. This setting prevents us from sending too small packets.
181 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
182 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
183 &VNET_NAME(tcp_minmss), 0,
184 "Minimum TCP Maximum Segment Size");
186 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
187 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
188 &VNET_NAME(tcp_do_rfc1323), 0,
189 "Enable rfc1323 (high performance TCP) extensions");
191 static int tcp_log_debug = 0;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
193 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
195 static int tcp_tcbhashsize = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
197 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
199 static int do_tcpdrain = 1;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
201 "Enable tcp_drain routine for extra help when low on mbufs");
203 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
204 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
206 static VNET_DEFINE(int, icmp_may_rst) = 1;
207 #define V_icmp_may_rst VNET(icmp_may_rst)
208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
209 &VNET_NAME(icmp_may_rst), 0,
210 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
212 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
213 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
215 &VNET_NAME(tcp_isn_reseed_interval), 0,
216 "Seconds between reseeding of ISN secret");
219 * TCP bandwidth limiting sysctls. Note that the default lower bound of
220 * 1024 exists only for debugging. A good production default would be
221 * something like 6100.
223 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
224 "TCP inflight data limiting");
226 static VNET_DEFINE(int, tcp_inflight_enable) = 0;
227 #define V_tcp_inflight_enable VNET(tcp_inflight_enable)
228 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
229 &VNET_NAME(tcp_inflight_enable), 0,
230 "Enable automatic TCP inflight data limiting");
232 static int tcp_inflight_debug = 0;
233 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
234 &tcp_inflight_debug, 0,
235 "Debug TCP inflight calculations");
237 static VNET_DEFINE(int, tcp_inflight_rttthresh);
238 #define V_tcp_inflight_rttthresh VNET(tcp_inflight_rttthresh)
239 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
240 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
241 vnet_sysctl_msec_to_ticks, "I",
242 "RTT threshold below which inflight will deactivate itself");
244 static VNET_DEFINE(int, tcp_inflight_min) = 6144;
245 #define V_tcp_inflight_min VNET(tcp_inflight_min)
246 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
247 &VNET_NAME(tcp_inflight_min), 0,
248 "Lower-bound for TCP inflight window");
250 static VNET_DEFINE(int, tcp_inflight_max) = TCP_MAXWIN << TCP_MAX_WINSHIFT;
251 #define V_tcp_inflight_max VNET(tcp_inflight_max)
252 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
253 &VNET_NAME(tcp_inflight_max), 0,
254 "Upper-bound for TCP inflight window");
256 static VNET_DEFINE(int, tcp_inflight_stab) = 20;
257 #define V_tcp_inflight_stab VNET(tcp_inflight_stab)
258 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
259 &VNET_NAME(tcp_inflight_stab), 0,
260 "Inflight Algorithm Stabilization 20 = 2 packets");
263 static int tcp_sig_checksigs = 1;
264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
265 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
268 VNET_DEFINE(uma_zone_t, sack_hole_zone);
269 #define V_sack_hole_zone VNET(sack_hole_zone)
271 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
273 static struct inpcb *tcp_notify(struct inpcb *, int);
274 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
275 static void tcp_isn_tick(void *);
276 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
277 void *ip4hdr, const void *ip6hdr);
280 * Target size of TCP PCB hash tables. Must be a power of two.
282 * Note that this can be overridden by the kernel environment
283 * variable net.inet.tcp.tcbhashsize
286 #define TCBHASHSIZE 512
291 * Callouts should be moved into struct tcp directly. They are currently
292 * separate because the tcpcb structure is exported to userland for sysctl
293 * parsing purposes, which do not know about callouts.
302 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
303 #define V_tcpcb_zone VNET(tcpcb_zone)
305 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
306 struct callout isn_callout;
307 static struct mtx isn_mtx;
309 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
310 #define ISN_LOCK() mtx_lock(&isn_mtx)
311 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
314 * TCP initialization.
317 tcp_zone_change(void *tag)
320 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
321 uma_zone_set_max(V_tcpcb_zone, maxsockets);
322 tcp_tw_zone_change();
326 tcp_inpcb_init(void *mem, int size, int flags)
328 struct inpcb *inp = mem;
330 INP_LOCK_INIT(inp, "inp", "tcpinp");
339 INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
342 V_tcbinfo.ipi_vnet = curvnet;
344 V_tcbinfo.ipi_listhead = &V_tcb;
346 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
347 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
348 printf("%s: WARNING: unable to register helper hook\n", __func__);
349 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
350 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
351 printf("%s: WARNING: unable to register helper hook\n", __func__);
353 hashsize = TCBHASHSIZE;
354 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
355 if (!powerof2(hashsize)) {
356 printf("WARNING: TCB hash size not a power of 2\n");
357 hashsize = 512; /* safe default */
359 V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
360 &V_tcbinfo.ipi_hashmask);
361 V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
362 &V_tcbinfo.ipi_porthashmask);
363 V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
364 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
365 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
366 V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
369 * These have to be type stable for the benefit of the timers.
371 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
372 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
373 uma_zone_set_max(V_tcpcb_zone, maxsockets);
379 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
380 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
381 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
383 /* Skip initialization of globals for non-default instances. */
384 if (!IS_DEFAULT_VNET(curvnet))
387 tcp_reass_global_init();
389 /* XXX virtualize those bellow? */
390 tcp_delacktime = TCPTV_DELACK;
391 tcp_keepinit = TCPTV_KEEP_INIT;
392 tcp_keepidle = TCPTV_KEEP_IDLE;
393 tcp_keepintvl = TCPTV_KEEPINTVL;
394 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
396 tcp_rexmit_min = TCPTV_MIN;
397 if (tcp_rexmit_min < 1)
399 tcp_rexmit_slop = TCPTV_CPU_VAR;
400 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
401 tcp_tcbhashsize = hashsize;
404 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
406 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
408 if (max_protohdr < TCP_MINPROTOHDR)
409 max_protohdr = TCP_MINPROTOHDR;
410 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
412 #undef TCP_MINPROTOHDR
415 callout_init(&isn_callout, CALLOUT_MPSAFE);
416 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
417 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
418 SHUTDOWN_PRI_DEFAULT);
419 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
420 EVENTHANDLER_PRI_ANY);
432 /* XXX check that hashes are empty! */
433 hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
434 V_tcbinfo.ipi_hashmask);
435 hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
436 V_tcbinfo.ipi_porthashmask);
438 uma_zdestroy(V_sack_hole_zone);
439 uma_zdestroy(V_tcpcb_zone);
440 uma_zdestroy(V_tcbinfo.ipi_zone);
442 INP_INFO_LOCK_DESTROY(&V_tcbinfo);
450 callout_stop(&isn_callout);
454 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
455 * tcp_template used to store this data in mbufs, but we now recopy it out
456 * of the tcpcb each time to conserve mbufs.
459 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
461 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
463 INP_WLOCK_ASSERT(inp);
466 if ((inp->inp_vflag & INP_IPV6) != 0) {
469 ip6 = (struct ip6_hdr *)ip_ptr;
470 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
471 (inp->inp_flow & IPV6_FLOWINFO_MASK);
472 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
473 (IPV6_VERSION & IPV6_VERSION_MASK);
474 ip6->ip6_nxt = IPPROTO_TCP;
475 ip6->ip6_plen = htons(sizeof(struct tcphdr));
476 ip6->ip6_src = inp->in6p_laddr;
477 ip6->ip6_dst = inp->in6p_faddr;
483 ip = (struct ip *)ip_ptr;
484 ip->ip_v = IPVERSION;
486 ip->ip_tos = inp->inp_ip_tos;
490 ip->ip_ttl = inp->inp_ip_ttl;
492 ip->ip_p = IPPROTO_TCP;
493 ip->ip_src = inp->inp_laddr;
494 ip->ip_dst = inp->inp_faddr;
496 th->th_sport = inp->inp_lport;
497 th->th_dport = inp->inp_fport;
505 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
509 * Create template to be used to send tcp packets on a connection.
510 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
511 * use for this function is in keepalives, which use tcp_respond.
514 tcpip_maketemplate(struct inpcb *inp)
518 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
521 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
526 * Send a single message to the TCP at address specified by
527 * the given TCP/IP header. If m == NULL, then we make a copy
528 * of the tcpiphdr at ti and send directly to the addressed host.
529 * This is used to force keep alive messages out using the TCP
530 * template for a connection. If flags are given then we send
531 * a message back to the TCP which originated the * segment ti,
532 * and discard the mbuf containing it and any other attached mbufs.
534 * In any case the ack and sequence number of the transmitted
535 * segment are as specified by the parameters.
537 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
540 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
541 tcp_seq ack, tcp_seq seq, int flags)
554 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
557 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
564 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
565 INP_WLOCK_ASSERT(inp);
570 if (!(flags & TH_RST)) {
571 win = sbspace(&inp->inp_socket->so_rcv);
572 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
573 win = (long)TCP_MAXWIN << tp->rcv_scale;
577 m = m_gethdr(M_DONTWAIT, MT_DATA);
581 m->m_data += max_linkhdr;
584 bcopy((caddr_t)ip6, mtod(m, caddr_t),
585 sizeof(struct ip6_hdr));
586 ip6 = mtod(m, struct ip6_hdr *);
587 nth = (struct tcphdr *)(ip6 + 1);
591 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
592 ip = mtod(m, struct ip *);
593 nth = (struct tcphdr *)(ip + 1);
595 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
600 * XXX MRT We inherrit the FIB, which is lucky.
604 m->m_data = (caddr_t)ipgen;
605 /* m_len is set later */
607 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
610 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
611 nth = (struct tcphdr *)(ip6 + 1);
615 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
616 nth = (struct tcphdr *)(ip + 1);
620 * this is usually a case when an extension header
621 * exists between the IPv6 header and the
624 nth->th_sport = th->th_sport;
625 nth->th_dport = th->th_dport;
627 xchg(nth->th_dport, nth->th_sport, uint16_t);
633 ip6->ip6_vfc = IPV6_VERSION;
634 ip6->ip6_nxt = IPPROTO_TCP;
635 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
637 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
641 tlen += sizeof (struct tcpiphdr);
643 ip->ip_ttl = V_ip_defttl;
644 if (V_path_mtu_discovery)
648 m->m_pkthdr.len = tlen;
649 m->m_pkthdr.rcvif = NULL;
653 * Packet is associated with a socket, so allow the
654 * label of the response to reflect the socket label.
656 INP_WLOCK_ASSERT(inp);
657 mac_inpcb_create_mbuf(inp, m);
660 * Packet is not associated with a socket, so possibly
661 * update the label in place.
663 mac_netinet_tcp_reply(m);
666 nth->th_seq = htonl(seq);
667 nth->th_ack = htonl(ack);
669 nth->th_off = sizeof (struct tcphdr) >> 2;
670 nth->th_flags = flags;
672 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
674 nth->th_win = htons((u_short)win);
679 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
680 sizeof(struct ip6_hdr),
681 tlen - sizeof(struct ip6_hdr));
682 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
687 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
688 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
689 m->m_pkthdr.csum_flags = CSUM_TCP;
690 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
693 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
694 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
698 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
701 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
705 * Create a new TCP control block, making an
706 * empty reassembly queue and hooking it to the argument
707 * protocol control block. The `inp' parameter must have
708 * come from the zone allocator set up in tcp_init().
711 tcp_newtcpcb(struct inpcb *inp)
713 struct tcpcb_mem *tm;
716 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
719 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
724 /* Initialise cc_var struct for this tcpcb. */
726 tp->ccv->type = IPPROTO_TCP;
727 tp->ccv->ccvc.tcp = tp;
730 * Use the current system default CC algorithm.
733 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
734 CC_ALGO(tp) = CC_DEFAULT();
737 if (CC_ALGO(tp)->cb_init != NULL)
738 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
739 uma_zfree(V_tcpcb_zone, tm);
744 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
745 uma_zfree(V_tcpcb_zone, tm);
750 tp->t_vnet = inp->inp_vnet;
752 tp->t_timers = &tm->tt;
753 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
754 tp->t_maxseg = tp->t_maxopd =
756 isipv6 ? V_tcp_v6mssdflt :
760 /* Set up our timeouts. */
761 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
762 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
763 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
764 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
765 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
767 if (V_tcp_do_rfc1323)
768 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
770 tp->t_flags |= TF_SACK_PERMIT;
771 TAILQ_INIT(&tp->snd_holes);
772 tp->t_inpcb = inp; /* XXX */
774 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
775 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
776 * reasonable initial retransmit time.
778 tp->t_srtt = TCPTV_SRTTBASE;
779 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
780 tp->t_rttmin = tcp_rexmit_min;
781 tp->t_rxtcur = TCPTV_RTOBASE;
782 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
783 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
784 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
785 tp->t_rcvtime = ticks;
786 tp->t_bw_rtttime = ticks;
788 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
789 * because the socket may be bound to an IPv6 wildcard address,
790 * which may match an IPv4-mapped IPv6 address.
792 inp->inp_ip_ttl = V_ip_defttl;
794 return (tp); /* XXX */
798 * Switch the congestion control algorithm back to NewReno for any active
799 * control blocks using an algorithm which is about to go away.
800 * This ensures the CC framework can allow the unload to proceed without leaving
801 * any dangling pointers which would trigger a panic.
802 * Returning non-zero would inform the CC framework that something went wrong
803 * and it would be unsafe to allow the unload to proceed. However, there is no
804 * way for this to occur with this implementation so we always return zero.
807 tcp_ccalgounload(struct cc_algo *unload_algo)
809 struct cc_algo *tmpalgo;
812 VNET_ITERATOR_DECL(vnet_iter);
815 * Check all active control blocks across all network stacks and change
816 * any that are using "unload_algo" back to NewReno. If "unload_algo"
817 * requires cleanup code to be run, call it.
820 VNET_FOREACH(vnet_iter) {
821 CURVNET_SET(vnet_iter);
822 INP_INFO_RLOCK(&V_tcbinfo);
824 * New connections already part way through being initialised
825 * with the CC algo we're removing will not race with this code
826 * because the INP_INFO_WLOCK is held during initialisation. We
827 * therefore don't enter the loop below until the connection
828 * list has stabilised.
830 LIST_FOREACH(inp, &V_tcb, inp_list) {
832 /* Important to skip tcptw structs. */
833 if (!(inp->inp_flags & INP_TIMEWAIT) &&
834 (tp = intotcpcb(inp)) != NULL) {
836 * By holding INP_WLOCK here, we are assured
837 * that the connection is not currently
838 * executing inside the CC module's functions
839 * i.e. it is safe to make the switch back to
842 if (CC_ALGO(tp) == unload_algo) {
843 tmpalgo = CC_ALGO(tp);
844 /* NewReno does not require any init. */
845 CC_ALGO(tp) = &newreno_cc_algo;
846 if (tmpalgo->cb_destroy != NULL)
847 tmpalgo->cb_destroy(tp->ccv);
852 INP_INFO_RUNLOCK(&V_tcbinfo);
861 * Drop a TCP connection, reporting
862 * the specified error. If connection is synchronized,
863 * then send a RST to peer.
866 tcp_drop(struct tcpcb *tp, int errno)
868 struct socket *so = tp->t_inpcb->inp_socket;
870 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
871 INP_WLOCK_ASSERT(tp->t_inpcb);
873 if (TCPS_HAVERCVDSYN(tp->t_state)) {
874 tp->t_state = TCPS_CLOSED;
875 (void) tcp_output_reset(tp);
876 TCPSTAT_INC(tcps_drops);
878 TCPSTAT_INC(tcps_conndrops);
879 if (errno == ETIMEDOUT && tp->t_softerror)
880 errno = tp->t_softerror;
881 so->so_error = errno;
882 return (tcp_close(tp));
886 tcp_discardcb(struct tcpcb *tp)
888 struct inpcb *inp = tp->t_inpcb;
889 struct socket *so = inp->inp_socket;
891 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
894 INP_WLOCK_ASSERT(inp);
897 * Make sure that all of our timers are stopped before we
900 callout_stop(&tp->t_timers->tt_rexmt);
901 callout_stop(&tp->t_timers->tt_persist);
902 callout_stop(&tp->t_timers->tt_keep);
903 callout_stop(&tp->t_timers->tt_2msl);
904 callout_stop(&tp->t_timers->tt_delack);
907 * If we got enough samples through the srtt filter,
908 * save the rtt and rttvar in the routing entry.
909 * 'Enough' is arbitrarily defined as 4 rtt samples.
910 * 4 samples is enough for the srtt filter to converge
911 * to within enough % of the correct value; fewer samples
912 * and we could save a bogus rtt. The danger is not high
913 * as tcp quickly recovers from everything.
914 * XXX: Works very well but needs some more statistics!
916 if (tp->t_rttupdated >= 4) {
917 struct hc_metrics_lite metrics;
920 bzero(&metrics, sizeof(metrics));
922 * Update the ssthresh always when the conditions below
923 * are satisfied. This gives us better new start value
924 * for the congestion avoidance for new connections.
925 * ssthresh is only set if packet loss occured on a session.
927 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
928 * being torn down. Ideally this code would not use 'so'.
930 ssthresh = tp->snd_ssthresh;
931 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
933 * convert the limit from user data bytes to
934 * packets then to packet data bytes.
936 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
939 ssthresh *= (u_long)(tp->t_maxseg +
941 (isipv6 ? sizeof (struct ip6_hdr) +
942 sizeof (struct tcphdr) :
944 sizeof (struct tcpiphdr)
951 metrics.rmx_ssthresh = ssthresh;
953 metrics.rmx_rtt = tp->t_srtt;
954 metrics.rmx_rttvar = tp->t_rttvar;
955 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
956 metrics.rmx_bandwidth = tp->snd_bandwidth;
957 metrics.rmx_cwnd = tp->snd_cwnd;
958 metrics.rmx_sendpipe = 0;
959 metrics.rmx_recvpipe = 0;
961 tcp_hc_update(&inp->inp_inc, &metrics);
964 /* free the reassembly queue, if any */
966 /* Disconnect offload device, if any. */
967 tcp_offload_detach(tp);
969 tcp_free_sackholes(tp);
971 /* Allow the CC algorithm to clean up after itself. */
972 if (CC_ALGO(tp)->cb_destroy != NULL)
973 CC_ALGO(tp)->cb_destroy(tp->ccv);
975 khelp_destroy_osd(tp->osd);
978 inp->inp_ppcb = NULL;
980 uma_zfree(V_tcpcb_zone, tp);
984 * Attempt to close a TCP control block, marking it as dropped, and freeing
985 * the socket if we hold the only reference.
988 tcp_close(struct tcpcb *tp)
990 struct inpcb *inp = tp->t_inpcb;
993 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
994 INP_WLOCK_ASSERT(inp);
996 /* Notify any offload devices of listener close */
997 if (tp->t_state == TCPS_LISTEN)
998 tcp_offload_listen_close(tp);
1000 TCPSTAT_INC(tcps_closed);
1001 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1002 so = inp->inp_socket;
1003 soisdisconnected(so);
1004 if (inp->inp_flags & INP_SOCKREF) {
1005 KASSERT(so->so_state & SS_PROTOREF,
1006 ("tcp_close: !SS_PROTOREF"));
1007 inp->inp_flags &= ~INP_SOCKREF;
1011 so->so_state &= ~SS_PROTOREF;
1021 VNET_ITERATOR_DECL(vnet_iter);
1026 VNET_LIST_RLOCK_NOSLEEP();
1027 VNET_FOREACH(vnet_iter) {
1028 CURVNET_SET(vnet_iter);
1033 * Walk the tcpbs, if existing, and flush the reassembly queue,
1034 * if there is one...
1035 * XXX: The "Net/3" implementation doesn't imply that the TCP
1036 * reassembly queue should be flushed, but in a situation
1037 * where we're really low on mbufs, this is potentially
1040 INP_INFO_RLOCK(&V_tcbinfo);
1041 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1042 if (inpb->inp_flags & INP_TIMEWAIT)
1045 if ((tcpb = intotcpcb(inpb)) != NULL) {
1046 tcp_reass_flush(tcpb);
1047 tcp_clean_sackreport(tcpb);
1051 INP_INFO_RUNLOCK(&V_tcbinfo);
1054 VNET_LIST_RUNLOCK_NOSLEEP();
1058 * Notify a tcp user of an asynchronous error;
1059 * store error as soft error, but wake up user
1060 * (for now, won't do anything until can select for soft error).
1062 * Do not wake up user since there currently is no mechanism for
1063 * reporting soft errors (yet - a kqueue filter may be added).
1065 static struct inpcb *
1066 tcp_notify(struct inpcb *inp, int error)
1070 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1071 INP_WLOCK_ASSERT(inp);
1073 if ((inp->inp_flags & INP_TIMEWAIT) ||
1074 (inp->inp_flags & INP_DROPPED))
1077 tp = intotcpcb(inp);
1078 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1081 * Ignore some errors if we are hooked up.
1082 * If connection hasn't completed, has retransmitted several times,
1083 * and receives a second error, give up now. This is better
1084 * than waiting a long time to establish a connection that
1085 * can never complete.
1087 if (tp->t_state == TCPS_ESTABLISHED &&
1088 (error == EHOSTUNREACH || error == ENETUNREACH ||
1089 error == EHOSTDOWN)) {
1091 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1093 tp = tcp_drop(tp, error);
1099 tp->t_softerror = error;
1103 wakeup( &so->so_timeo);
1110 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1112 int error, i, m, n, pcb_count;
1113 struct inpcb *inp, **inp_list;
1118 * The process of preparing the TCB list is too time-consuming and
1119 * resource-intensive to repeat twice on every request.
1121 if (req->oldptr == NULL) {
1122 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1123 n += imax(n / 8, 10);
1124 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1128 if (req->newptr != NULL)
1132 * OK, now we're committed to doing something.
1134 INP_INFO_RLOCK(&V_tcbinfo);
1135 gencnt = V_tcbinfo.ipi_gencnt;
1136 n = V_tcbinfo.ipi_count;
1137 INP_INFO_RUNLOCK(&V_tcbinfo);
1139 m = syncache_pcbcount();
1141 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1142 + (n + m) * sizeof(struct xtcpcb));
1146 xig.xig_len = sizeof xig;
1147 xig.xig_count = n + m;
1148 xig.xig_gen = gencnt;
1149 xig.xig_sogen = so_gencnt;
1150 error = SYSCTL_OUT(req, &xig, sizeof xig);
1154 error = syncache_pcblist(req, m, &pcb_count);
1158 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1159 if (inp_list == NULL)
1162 INP_INFO_RLOCK(&V_tcbinfo);
1163 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1164 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1166 if (inp->inp_gencnt <= gencnt) {
1168 * XXX: This use of cr_cansee(), introduced with
1169 * TCP state changes, is not quite right, but for
1170 * now, better than nothing.
1172 if (inp->inp_flags & INP_TIMEWAIT) {
1173 if (intotw(inp) != NULL)
1174 error = cr_cansee(req->td->td_ucred,
1175 intotw(inp)->tw_cred);
1177 error = EINVAL; /* Skip this inp. */
1179 error = cr_canseeinpcb(req->td->td_ucred, inp);
1182 inp_list[i++] = inp;
1187 INP_INFO_RUNLOCK(&V_tcbinfo);
1191 for (i = 0; i < n; i++) {
1194 if (inp->inp_gencnt <= gencnt) {
1198 bzero(&xt, sizeof(xt));
1199 xt.xt_len = sizeof xt;
1200 /* XXX should avoid extra copy */
1201 bcopy(inp, &xt.xt_inp, sizeof *inp);
1202 inp_ppcb = inp->inp_ppcb;
1203 if (inp_ppcb == NULL)
1204 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1205 else if (inp->inp_flags & INP_TIMEWAIT) {
1206 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1207 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1209 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1210 if (inp->inp_socket != NULL)
1211 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1213 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1214 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1216 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1218 error = SYSCTL_OUT(req, &xt, sizeof xt);
1222 INP_INFO_WLOCK(&V_tcbinfo);
1223 for (i = 0; i < n; i++) {
1226 if (!in_pcbrele(inp))
1229 INP_INFO_WUNLOCK(&V_tcbinfo);
1233 * Give the user an updated idea of our state.
1234 * If the generation differs from what we told
1235 * her before, she knows that something happened
1236 * while we were processing this request, and it
1237 * might be necessary to retry.
1239 INP_INFO_RLOCK(&V_tcbinfo);
1240 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1241 xig.xig_sogen = so_gencnt;
1242 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1243 INP_INFO_RUNLOCK(&V_tcbinfo);
1244 error = SYSCTL_OUT(req, &xig, sizeof xig);
1246 free(inp_list, M_TEMP);
1250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1251 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1252 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1255 tcp_getcred(SYSCTL_HANDLER_ARGS)
1258 struct sockaddr_in addrs[2];
1262 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1265 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1268 INP_INFO_RLOCK(&V_tcbinfo);
1269 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1270 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1273 INP_INFO_RUNLOCK(&V_tcbinfo);
1274 if (inp->inp_socket == NULL)
1277 error = cr_canseeinpcb(req->td->td_ucred, inp);
1279 cru2x(inp->inp_cred, &xuc);
1282 INP_INFO_RUNLOCK(&V_tcbinfo);
1286 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1290 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1291 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1292 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1296 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1299 struct sockaddr_in6 addrs[2];
1301 int error, mapped = 0;
1303 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1306 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1309 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1310 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1313 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1314 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1320 INP_INFO_RLOCK(&V_tcbinfo);
1322 inp = in_pcblookup_hash(&V_tcbinfo,
1323 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1325 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1329 inp = in6_pcblookup_hash(&V_tcbinfo,
1330 &addrs[1].sin6_addr, addrs[1].sin6_port,
1331 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1334 INP_INFO_RUNLOCK(&V_tcbinfo);
1335 if (inp->inp_socket == NULL)
1338 error = cr_canseeinpcb(req->td->td_ucred, inp);
1340 cru2x(inp->inp_cred, &xuc);
1343 INP_INFO_RUNLOCK(&V_tcbinfo);
1347 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1351 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1352 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1353 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1358 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1360 struct ip *ip = vip;
1362 struct in_addr faddr;
1365 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1367 struct in_conninfo inc;
1368 tcp_seq icmp_tcp_seq;
1371 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1372 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1375 if (cmd == PRC_MSGSIZE)
1376 notify = tcp_mtudisc_notify;
1377 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1378 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1379 notify = tcp_drop_syn_sent;
1381 * Redirects don't need to be handled up here.
1383 else if (PRC_IS_REDIRECT(cmd))
1386 * Source quench is depreciated.
1388 else if (cmd == PRC_QUENCH)
1391 * Hostdead is ugly because it goes linearly through all PCBs.
1392 * XXX: We never get this from ICMP, otherwise it makes an
1393 * excellent DoS attack on machines with many connections.
1395 else if (cmd == PRC_HOSTDEAD)
1397 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1400 icp = (struct icmp *)((caddr_t)ip
1401 - offsetof(struct icmp, icmp_ip));
1402 th = (struct tcphdr *)((caddr_t)ip
1403 + (ip->ip_hl << 2));
1404 INP_INFO_WLOCK(&V_tcbinfo);
1405 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1406 ip->ip_src, th->th_sport, 0, NULL);
1409 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1410 !(inp->inp_flags & INP_DROPPED) &&
1411 !(inp->inp_socket == NULL)) {
1412 icmp_tcp_seq = htonl(th->th_seq);
1413 tp = intotcpcb(inp);
1414 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1415 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1416 if (cmd == PRC_MSGSIZE) {
1419 * If we got a needfrag set the MTU
1420 * in the route to the suggested new
1421 * value (if given) and then notify.
1423 bzero(&inc, sizeof(inc));
1424 inc.inc_faddr = faddr;
1426 inp->inp_inc.inc_fibnum;
1428 mtu = ntohs(icp->icmp_nextmtu);
1430 * If no alternative MTU was
1431 * proposed, try the next smaller
1432 * one. ip->ip_len has already
1433 * been swapped in icmp_input().
1436 mtu = ip_next_mtu(ip->ip_len,
1438 if (mtu < V_tcp_minmss
1439 + sizeof(struct tcpiphdr))
1441 + sizeof(struct tcpiphdr);
1443 * Only cache the MTU if it
1444 * is smaller than the interface
1445 * or route MTU. tcp_mtudisc()
1446 * will do right thing by itself.
1448 if (mtu <= tcp_maxmtu(&inc, NULL))
1449 tcp_hc_updatemtu(&inc, mtu);
1450 tcp_mtudisc(inp, mtu);
1452 inp = (*notify)(inp,
1453 inetctlerrmap[cmd]);
1459 bzero(&inc, sizeof(inc));
1460 inc.inc_fport = th->th_dport;
1461 inc.inc_lport = th->th_sport;
1462 inc.inc_faddr = faddr;
1463 inc.inc_laddr = ip->ip_src;
1464 syncache_unreach(&inc, th);
1466 INP_INFO_WUNLOCK(&V_tcbinfo);
1468 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1473 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1476 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1477 struct ip6_hdr *ip6;
1479 struct ip6ctlparam *ip6cp = NULL;
1480 const struct sockaddr_in6 *sa6_src = NULL;
1482 struct tcp_portonly {
1487 if (sa->sa_family != AF_INET6 ||
1488 sa->sa_len != sizeof(struct sockaddr_in6))
1491 if (cmd == PRC_MSGSIZE)
1492 notify = tcp_mtudisc_notify;
1493 else if (!PRC_IS_REDIRECT(cmd) &&
1494 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1496 /* Source quench is depreciated. */
1497 else if (cmd == PRC_QUENCH)
1500 /* if the parameter is from icmp6, decode it. */
1502 ip6cp = (struct ip6ctlparam *)d;
1504 ip6 = ip6cp->ip6c_ip6;
1505 off = ip6cp->ip6c_off;
1506 sa6_src = ip6cp->ip6c_src;
1510 off = 0; /* fool gcc */
1515 struct in_conninfo inc;
1517 * XXX: We assume that when IPV6 is non NULL,
1518 * M and OFF are valid.
1521 /* check if we can safely examine src and dst ports */
1522 if (m->m_pkthdr.len < off + sizeof(*thp))
1525 bzero(&th, sizeof(th));
1526 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1528 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1529 (struct sockaddr *)ip6cp->ip6c_src,
1530 th.th_sport, cmd, NULL, notify);
1532 bzero(&inc, sizeof(inc));
1533 inc.inc_fport = th.th_dport;
1534 inc.inc_lport = th.th_sport;
1535 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1536 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1537 inc.inc_flags |= INC_ISIPV6;
1538 INP_INFO_WLOCK(&V_tcbinfo);
1539 syncache_unreach(&inc, &th);
1540 INP_INFO_WUNLOCK(&V_tcbinfo);
1542 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1543 0, cmd, NULL, notify);
1549 * Following is where TCP initial sequence number generation occurs.
1551 * There are two places where we must use initial sequence numbers:
1552 * 1. In SYN-ACK packets.
1553 * 2. In SYN packets.
1555 * All ISNs for SYN-ACK packets are generated by the syncache. See
1556 * tcp_syncache.c for details.
1558 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1559 * depends on this property. In addition, these ISNs should be
1560 * unguessable so as to prevent connection hijacking. To satisfy
1561 * the requirements of this situation, the algorithm outlined in
1562 * RFC 1948 is used, with only small modifications.
1564 * Implementation details:
1566 * Time is based off the system timer, and is corrected so that it
1567 * increases by one megabyte per second. This allows for proper
1568 * recycling on high speed LANs while still leaving over an hour
1571 * As reading the *exact* system time is too expensive to be done
1572 * whenever setting up a TCP connection, we increment the time
1573 * offset in two ways. First, a small random positive increment
1574 * is added to isn_offset for each connection that is set up.
1575 * Second, the function tcp_isn_tick fires once per clock tick
1576 * and increments isn_offset as necessary so that sequence numbers
1577 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1578 * random positive increments serve only to ensure that the same
1579 * exact sequence number is never sent out twice (as could otherwise
1580 * happen when a port is recycled in less than the system tick
1583 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1584 * between seeding of isn_secret. This is normally set to zero,
1585 * as reseeding should not be necessary.
1587 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1588 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1589 * general, this means holding an exclusive (write) lock.
1592 #define ISN_BYTES_PER_SECOND 1048576
1593 #define ISN_STATIC_INCREMENT 4096
1594 #define ISN_RANDOM_INCREMENT (4096 - 1)
1596 static VNET_DEFINE(u_char, isn_secret[32]);
1597 static VNET_DEFINE(int, isn_last_reseed);
1598 static VNET_DEFINE(u_int32_t, isn_offset);
1599 static VNET_DEFINE(u_int32_t, isn_offset_old);
1601 #define V_isn_secret VNET(isn_secret)
1602 #define V_isn_last_reseed VNET(isn_last_reseed)
1603 #define V_isn_offset VNET(isn_offset)
1604 #define V_isn_offset_old VNET(isn_offset_old)
1607 tcp_new_isn(struct tcpcb *tp)
1610 u_int32_t md5_buffer[4];
1613 INP_WLOCK_ASSERT(tp->t_inpcb);
1616 /* Seed if this is the first use, reseed if requested. */
1617 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1618 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1620 read_random(&V_isn_secret, sizeof(V_isn_secret));
1621 V_isn_last_reseed = ticks;
1624 /* Compute the md5 hash and return the ISN. */
1626 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1627 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1629 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1630 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1631 sizeof(struct in6_addr));
1632 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1633 sizeof(struct in6_addr));
1637 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1638 sizeof(struct in_addr));
1639 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1640 sizeof(struct in_addr));
1642 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1643 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1644 new_isn = (tcp_seq) md5_buffer[0];
1645 V_isn_offset += ISN_STATIC_INCREMENT +
1646 (arc4random() & ISN_RANDOM_INCREMENT);
1647 new_isn += V_isn_offset;
1653 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1654 * to keep time flowing at a relatively constant rate. If the random
1655 * increments have already pushed us past the projected offset, do nothing.
1658 tcp_isn_tick(void *xtp)
1660 VNET_ITERATOR_DECL(vnet_iter);
1661 u_int32_t projected_offset;
1663 VNET_LIST_RLOCK_NOSLEEP();
1665 VNET_FOREACH(vnet_iter) {
1666 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1668 V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1670 if (SEQ_GT(projected_offset, V_isn_offset))
1671 V_isn_offset = projected_offset;
1673 V_isn_offset_old = V_isn_offset;
1677 VNET_LIST_RUNLOCK_NOSLEEP();
1678 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1682 * When a specific ICMP unreachable message is received and the
1683 * connection state is SYN-SENT, drop the connection. This behavior
1684 * is controlled by the icmp_may_rst sysctl.
1687 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1691 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1692 INP_WLOCK_ASSERT(inp);
1694 if ((inp->inp_flags & INP_TIMEWAIT) ||
1695 (inp->inp_flags & INP_DROPPED))
1698 tp = intotcpcb(inp);
1699 if (tp->t_state != TCPS_SYN_SENT)
1702 tp = tcp_drop(tp, errno);
1710 * When `need fragmentation' ICMP is received, update our idea of the MSS
1711 * based on the new value. Also nudge TCP to send something, since we
1712 * know the packet we just sent was dropped.
1713 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1715 static struct inpcb *
1716 tcp_mtudisc_notify(struct inpcb *inp, int error)
1719 return (tcp_mtudisc(inp, -1));
1723 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1728 INP_WLOCK_ASSERT(inp);
1729 if ((inp->inp_flags & INP_TIMEWAIT) ||
1730 (inp->inp_flags & INP_DROPPED))
1733 tp = intotcpcb(inp);
1734 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1736 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1738 so = inp->inp_socket;
1739 SOCKBUF_LOCK(&so->so_snd);
1740 /* If the mss is larger than the socket buffer, decrease the mss. */
1741 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1742 tp->t_maxseg = so->so_snd.sb_hiwat;
1743 SOCKBUF_UNLOCK(&so->so_snd);
1745 TCPSTAT_INC(tcps_mturesent);
1747 tp->snd_nxt = tp->snd_una;
1748 tcp_free_sackholes(tp);
1749 tp->snd_recover = tp->snd_max;
1750 if (tp->t_flags & TF_SACK_PERMIT)
1751 EXIT_FASTRECOVERY(tp->t_flags);
1752 tcp_output_send(tp);
1757 * Look-up the routing entry to the peer of this inpcb. If no route
1758 * is found and it cannot be allocated, then return 0. This routine
1759 * is called by TCP routines that access the rmx structure and by
1760 * tcp_mss_update to get the peer/interface MTU.
1763 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1766 struct sockaddr_in *dst;
1770 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1772 bzero(&sro, sizeof(sro));
1773 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1774 dst = (struct sockaddr_in *)&sro.ro_dst;
1775 dst->sin_family = AF_INET;
1776 dst->sin_len = sizeof(*dst);
1777 dst->sin_addr = inc->inc_faddr;
1778 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1780 if (sro.ro_rt != NULL) {
1781 ifp = sro.ro_rt->rt_ifp;
1782 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1783 maxmtu = ifp->if_mtu;
1785 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1787 /* Report additional interface capabilities. */
1788 if (flags != NULL) {
1789 if (ifp->if_capenable & IFCAP_TSO4 &&
1790 ifp->if_hwassist & CSUM_TSO)
1800 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1802 struct route_in6 sro6;
1806 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1808 bzero(&sro6, sizeof(sro6));
1809 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1810 sro6.ro_dst.sin6_family = AF_INET6;
1811 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1812 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1813 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1815 if (sro6.ro_rt != NULL) {
1816 ifp = sro6.ro_rt->rt_ifp;
1817 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1818 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1820 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1821 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1823 /* Report additional interface capabilities. */
1824 if (flags != NULL) {
1825 if (ifp->if_capenable & IFCAP_TSO6 &&
1826 ifp->if_hwassist & CSUM_TSO)
1837 /* compute ESP/AH header size for TCP, including outer IP header. */
1839 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1846 struct ip6_hdr *ip6;
1850 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1852 MGETHDR(m, M_DONTWAIT, MT_DATA);
1857 if ((inp->inp_vflag & INP_IPV6) != 0) {
1858 ip6 = mtod(m, struct ip6_hdr *);
1859 th = (struct tcphdr *)(ip6 + 1);
1860 m->m_pkthdr.len = m->m_len =
1861 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1862 tcpip_fillheaders(inp, ip6, th);
1863 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1867 ip = mtod(m, struct ip *);
1868 th = (struct tcphdr *)(ip + 1);
1869 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1870 tcpip_fillheaders(inp, ip, th);
1871 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1880 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1882 * This code attempts to calculate the bandwidth-delay product as a
1883 * means of determining the optimal window size to maximize bandwidth,
1884 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1885 * routers. This code also does a fairly good job keeping RTTs in check
1886 * across slow links like modems. We implement an algorithm which is very
1887 * similar (but not meant to be) TCP/Vegas. The code operates on the
1888 * transmitter side of a TCP connection and so only effects the transmit
1889 * side of the connection.
1891 * BACKGROUND: TCP makes no provision for the management of buffer space
1892 * at the end points or at the intermediate routers and switches. A TCP
1893 * stream, whether using NewReno or not, will eventually buffer as
1894 * many packets as it is able and the only reason this typically works is
1895 * due to the fairly small default buffers made available for a connection
1896 * (typicaly 16K or 32K). As machines use larger windows and/or window
1897 * scaling it is now fairly easy for even a single TCP connection to blow-out
1898 * all available buffer space not only on the local interface, but on
1899 * intermediate routers and switches as well. NewReno makes a misguided
1900 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1901 * then backing off, then steadily increasing the window again until another
1902 * failure occurs, ad-infinitum. This results in terrible oscillation that
1903 * is only made worse as network loads increase and the idea of intentionally
1904 * blowing out network buffers is, frankly, a terrible way to manage network
1907 * It is far better to limit the transmit window prior to the failure
1908 * condition being achieved. There are two general ways to do this: First
1909 * you can 'scan' through different transmit window sizes and locate the
1910 * point where the RTT stops increasing, indicating that you have filled the
1911 * pipe, then scan backwards until you note that RTT stops decreasing, then
1912 * repeat ad-infinitum. This method works in principle but has severe
1913 * implementation issues due to RTT variances, timer granularity, and
1914 * instability in the algorithm which can lead to many false positives and
1915 * create oscillations as well as interact badly with other TCP streams
1916 * implementing the same algorithm.
1918 * The second method is to limit the window to the bandwidth delay product
1919 * of the link. This is the method we implement. RTT variances and our
1920 * own manipulation of the congestion window, bwnd, can potentially
1921 * destabilize the algorithm. For this reason we have to stabilize the
1922 * elements used to calculate the window. We do this by using the minimum
1923 * observed RTT, the long term average of the observed bandwidth, and
1924 * by adding two segments worth of slop. It isn't perfect but it is able
1925 * to react to changing conditions and gives us a very stable basis on
1926 * which to extend the algorithm.
1929 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1935 INP_WLOCK_ASSERT(tp->t_inpcb);
1938 * If inflight_enable is disabled in the middle of a tcp connection,
1939 * make sure snd_bwnd is effectively disabled.
1941 if (V_tcp_inflight_enable == 0 ||
1942 tp->t_rttlow < V_tcp_inflight_rttthresh) {
1943 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1944 tp->snd_bandwidth = 0;
1949 * Figure out the bandwidth. Due to the tick granularity this
1950 * is a very rough number and it MUST be averaged over a fairly
1951 * long period of time. XXX we need to take into account a link
1952 * that is not using all available bandwidth, but for now our
1953 * slop will ramp us up if this case occurs and the bandwidth later
1956 * Note: if ticks rollover 'bw' may wind up negative. We must
1957 * effectively reset t_bw_rtttime for this case.
1960 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1963 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1964 (save_ticks - tp->t_bw_rtttime);
1965 tp->t_bw_rtttime = save_ticks;
1966 tp->t_bw_rtseq = ack_seq;
1967 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1969 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1971 tp->snd_bandwidth = bw;
1974 * Calculate the semi-static bandwidth delay product, plus two maximal
1975 * segments. The additional slop puts us squarely in the sweet
1976 * spot and also handles the bandwidth run-up case and stabilization.
1977 * Without the slop we could be locking ourselves into a lower
1980 * Situations Handled:
1981 * (1) Prevents over-queueing of packets on LANs, especially on
1982 * high speed LANs, allowing larger TCP buffers to be
1983 * specified, and also does a good job preventing
1984 * over-queueing of packets over choke points like modems
1985 * (at least for the transmit side).
1987 * (2) Is able to handle changing network loads (bandwidth
1988 * drops so bwnd drops, bandwidth increases so bwnd
1991 * (3) Theoretically should stabilize in the face of multiple
1992 * connections implementing the same algorithm (this may need
1995 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1996 * be adjusted with a sysctl but typically only needs to be
1997 * on very slow connections. A value no smaller then 5
1998 * should be used, but only reduce this default if you have
2001 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2002 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
2005 if (tcp_inflight_debug > 0) {
2007 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2009 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2018 if ((long)bwnd < V_tcp_inflight_min)
2019 bwnd = V_tcp_inflight_min;
2020 if (bwnd > V_tcp_inflight_max)
2021 bwnd = V_tcp_inflight_max;
2022 if ((long)bwnd < tp->t_maxseg * 2)
2023 bwnd = tp->t_maxseg * 2;
2024 tp->snd_bwnd = bwnd;
2027 #ifdef TCP_SIGNATURE
2029 * Callback function invoked by m_apply() to digest TCP segment data
2030 * contained within an mbuf chain.
2033 tcp_signature_apply(void *fstate, void *data, u_int len)
2036 MD5Update(fstate, (u_char *)data, len);
2041 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2044 * m pointer to head of mbuf chain
2046 * len length of TCP segment data, excluding options
2047 * optlen length of TCP segment options
2048 * buf pointer to storage for computed MD5 digest
2049 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2051 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2052 * When called from tcp_input(), we can be sure that th_sum has been
2053 * zeroed out and verified already.
2055 * Return 0 if successful, otherwise return -1.
2057 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2058 * search with the destination IP address, and a 'magic SPI' to be
2059 * determined by the application. This is hardcoded elsewhere to 1179
2060 * right now. Another branch of this code exists which uses the SPD to
2061 * specify per-application flows but it is unstable.
2064 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2065 u_char *buf, u_int direction)
2067 union sockaddr_union dst;
2068 struct ippseudo ippseudo;
2072 struct ipovly *ipovly;
2073 struct secasvar *sav;
2076 struct ip6_hdr *ip6;
2077 struct in6_addr in6;
2078 char ip6buf[INET6_ADDRSTRLEN];
2084 KASSERT(m != NULL, ("NULL mbuf chain"));
2085 KASSERT(buf != NULL, ("NULL signature pointer"));
2087 /* Extract the destination from the IP header in the mbuf. */
2088 bzero(&dst, sizeof(union sockaddr_union));
2089 ip = mtod(m, struct ip *);
2091 ip6 = NULL; /* Make the compiler happy. */
2095 dst.sa.sa_len = sizeof(struct sockaddr_in);
2096 dst.sa.sa_family = AF_INET;
2097 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2098 ip->ip_src : ip->ip_dst;
2101 case (IPV6_VERSION >> 4):
2102 ip6 = mtod(m, struct ip6_hdr *);
2103 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2104 dst.sa.sa_family = AF_INET6;
2105 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2106 ip6->ip6_src : ip6->ip6_dst;
2115 /* Look up an SADB entry which matches the address of the peer. */
2116 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2118 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2119 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2121 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2122 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2130 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2132 * XXX The ippseudo header MUST be digested in network byte order,
2133 * or else we'll fail the regression test. Assume all fields we've
2134 * been doing arithmetic on have been in host byte order.
2135 * XXX One cannot depend on ipovly->ih_len here. When called from
2136 * tcp_output(), the underlying ip_len member has not yet been set.
2140 ipovly = (struct ipovly *)ip;
2141 ippseudo.ippseudo_src = ipovly->ih_src;
2142 ippseudo.ippseudo_dst = ipovly->ih_dst;
2143 ippseudo.ippseudo_pad = 0;
2144 ippseudo.ippseudo_p = IPPROTO_TCP;
2145 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2147 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2149 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2150 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2154 * RFC 2385, 2.0 Proposal
2155 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2156 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2157 * extended next header value (to form 32 bits), and 32-bit segment
2159 * Note: Upper-Layer Packet Length comes before Next Header.
2161 case (IPV6_VERSION >> 4):
2163 in6_clearscope(&in6);
2164 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2166 in6_clearscope(&in6);
2167 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2168 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2169 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2171 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2172 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2173 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2175 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2177 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2178 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2189 * Step 2: Update MD5 hash with TCP header, excluding options.
2190 * The TCP checksum must be set to zero.
2192 savecsum = th->th_sum;
2194 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2195 th->th_sum = savecsum;
2198 * Step 3: Update MD5 hash with TCP segment data.
2199 * Use m_apply() to avoid an early m_pullup().
2202 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2205 * Step 4: Update MD5 hash with shared secret.
2207 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2208 MD5Final(buf, &ctx);
2210 key_sa_recordxfer(sav, m);
2216 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2219 * m pointer to head of mbuf chain
2220 * len length of TCP segment data, excluding options
2221 * optlen length of TCP segment options
2222 * buf pointer to storage for computed MD5 digest
2223 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2225 * Return 1 if successful, otherwise return 0.
2228 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2229 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2231 char tmpdigest[TCP_SIGLEN];
2233 if (tcp_sig_checksigs == 0)
2235 if ((tcpbflag & TF_SIGNATURE) == 0) {
2236 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2239 * If this socket is not expecting signature but
2240 * the segment contains signature just fail.
2242 TCPSTAT_INC(tcps_sig_err_sigopt);
2243 TCPSTAT_INC(tcps_sig_rcvbadsig);
2247 /* Signature is not expected, and not present in segment. */
2252 * If this socket is expecting signature but the segment does not
2253 * contain any just fail.
2255 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2256 TCPSTAT_INC(tcps_sig_err_nosigopt);
2257 TCPSTAT_INC(tcps_sig_rcvbadsig);
2260 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2261 IPSEC_DIR_INBOUND) == -1) {
2262 TCPSTAT_INC(tcps_sig_err_buildsig);
2263 TCPSTAT_INC(tcps_sig_rcvbadsig);
2267 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2268 TCPSTAT_INC(tcps_sig_rcvbadsig);
2271 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2274 #endif /* TCP_SIGNATURE */
2277 sysctl_drop(SYSCTL_HANDLER_ARGS)
2279 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2280 struct sockaddr_storage addrs[2];
2284 struct sockaddr_in *fin, *lin;
2286 struct sockaddr_in6 *fin6, *lin6;
2297 if (req->oldptr != NULL || req->oldlen != 0)
2299 if (req->newptr == NULL)
2301 if (req->newlen < sizeof(addrs))
2303 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2307 switch (addrs[0].ss_family) {
2310 fin6 = (struct sockaddr_in6 *)&addrs[0];
2311 lin6 = (struct sockaddr_in6 *)&addrs[1];
2312 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2313 lin6->sin6_len != sizeof(struct sockaddr_in6))
2315 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2316 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2318 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2319 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2320 fin = (struct sockaddr_in *)&addrs[0];
2321 lin = (struct sockaddr_in *)&addrs[1];
2324 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2327 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2333 fin = (struct sockaddr_in *)&addrs[0];
2334 lin = (struct sockaddr_in *)&addrs[1];
2335 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2336 lin->sin_len != sizeof(struct sockaddr_in))
2342 INP_INFO_WLOCK(&V_tcbinfo);
2343 switch (addrs[0].ss_family) {
2346 inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2347 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2352 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2353 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2358 if (inp->inp_flags & INP_TIMEWAIT) {
2360 * XXXRW: There currently exists a state where an
2361 * inpcb is present, but its timewait state has been
2362 * discarded. For now, don't allow dropping of this
2370 } else if (!(inp->inp_flags & INP_DROPPED) &&
2371 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2372 tp = intotcpcb(inp);
2373 tp = tcp_drop(tp, ECONNABORTED);
2380 INP_INFO_WUNLOCK(&V_tcbinfo);
2384 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2385 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2386 0, sysctl_drop, "", "Drop TCP connection");
2389 * Generate a standardized TCP log line for use throughout the
2390 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2391 * allow use in the interrupt context.
2393 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2394 * NB: The function may return NULL if memory allocation failed.
2396 * Due to header inclusion and ordering limitations the struct ip
2397 * and ip6_hdr pointers have to be passed as void pointers.
2400 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2404 /* Is logging enabled? */
2405 if (tcp_log_in_vain == 0)
2408 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2412 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2416 /* Is logging enabled? */
2417 if (tcp_log_debug == 0)
2420 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2424 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2431 const struct ip6_hdr *ip6;
2433 ip6 = (const struct ip6_hdr *)ip6hdr;
2435 ip = (struct ip *)ip4hdr;
2438 * The log line looks like this:
2439 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2441 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2442 sizeof(PRINT_TH_FLAGS) + 1 +
2444 2 * INET6_ADDRSTRLEN;
2446 2 * INET_ADDRSTRLEN;
2449 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2453 strcat(s, "TCP: [");
2456 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2457 inet_ntoa_r(inc->inc_faddr, sp);
2459 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2461 inet_ntoa_r(inc->inc_laddr, sp);
2463 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2466 ip6_sprintf(sp, &inc->inc6_faddr);
2468 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2470 ip6_sprintf(sp, &inc->inc6_laddr);
2472 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2473 } else if (ip6 && th) {
2474 ip6_sprintf(sp, &ip6->ip6_src);
2476 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2478 ip6_sprintf(sp, &ip6->ip6_dst);
2480 sprintf(sp, "]:%i", ntohs(th->th_dport));
2482 } else if (ip && th) {
2483 inet_ntoa_r(ip->ip_src, sp);
2485 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2487 inet_ntoa_r(ip->ip_dst, sp);
2489 sprintf(sp, "]:%i", ntohs(th->th_dport));
2496 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2497 if (*(s + size - 1) != '\0')
2498 panic("%s: string too long", __func__);