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_kdtrace.h"
40 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
53 #include <sys/domain.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/protosw.h>
61 #include <sys/random.h>
65 #include <net/route.h>
69 #include <netinet/cc.h>
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/in_var.h>
75 #include <netinet/ip.h>
76 #include <netinet/ip_icmp.h>
77 #include <netinet/ip_var.h>
79 #include <netinet/ip6.h>
80 #include <netinet6/in6_pcb.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #include <netinet6/nd6.h>
87 #include <netinet/tcp_fastopen.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/tcp_syncache.h>
95 #include <netinet6/tcp6_var.h>
97 #include <netinet/tcpip.h>
99 #include <netinet/tcp_debug.h>
102 #include <netinet6/ip6protosw.h>
105 #include <netinet/tcp_offload.h>
109 #include <netipsec/ipsec.h>
110 #include <netipsec/xform.h>
112 #include <netipsec/ipsec6.h>
114 #include <netipsec/key.h>
115 #include <sys/syslog.h>
118 #include <machine/in_cksum.h>
121 #include <security/mac/mac_framework.h>
123 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
125 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
129 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
134 error = sysctl_handle_int(oidp, &new, 0, req);
135 if (error == 0 && req->newptr) {
136 if (new < TCP_MINMSS)
144 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
145 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
146 &sysctl_net_inet_tcp_mss_check, "I",
147 "Default TCP Maximum Segment Size");
151 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
155 new = V_tcp_v6mssdflt;
156 error = sysctl_handle_int(oidp, &new, 0, req);
157 if (error == 0 && req->newptr) {
158 if (new < TCP_MINMSS)
161 V_tcp_v6mssdflt = new;
166 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
167 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
168 &sysctl_net_inet_tcp_mss_v6_check, "I",
169 "Default TCP Maximum Segment Size for IPv6");
173 * Minimum MSS we accept and use. This prevents DoS attacks where
174 * we are forced to a ridiculous low MSS like 20 and send hundreds
175 * of packets instead of one. The effect scales with the available
176 * bandwidth and quickly saturates the CPU and network interface
177 * with packet generation and sending. Set to zero to disable MINMSS
178 * checking. This setting prevents us from sending too small packets.
180 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
181 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
182 &VNET_NAME(tcp_minmss), 0,
183 "Minimum TCP Maximum Segment Size");
185 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
186 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc1323), 0,
188 "Enable rfc1323 (high performance TCP) extensions");
190 static int tcp_log_debug = 0;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
192 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
194 static int tcp_tcbhashsize = 0;
195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
196 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
198 static int do_tcpdrain = 1;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
200 "Enable tcp_drain routine for extra help when low on mbufs");
202 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
203 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
205 static VNET_DEFINE(int, icmp_may_rst) = 1;
206 #define V_icmp_may_rst VNET(icmp_may_rst)
207 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
208 &VNET_NAME(icmp_may_rst), 0,
209 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
211 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
212 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
213 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
214 &VNET_NAME(tcp_isn_reseed_interval), 0,
215 "Seconds between reseeding of ISN secret");
217 static int tcp_soreceive_stream = 0;
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
219 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
222 static int tcp_sig_checksigs = 1;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
224 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
227 VNET_DEFINE(uma_zone_t, sack_hole_zone);
228 #define V_sack_hole_zone VNET(sack_hole_zone)
230 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
232 static struct inpcb *tcp_notify(struct inpcb *, int);
233 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
235 void *ip4hdr, const void *ip6hdr);
236 static void tcp_timer_discard(struct tcpcb *, uint32_t);
239 * Target size of TCP PCB hash tables. Must be a power of two.
241 * Note that this can be overridden by the kernel environment
242 * variable net.inet.tcp.tcbhashsize
245 #define TCBHASHSIZE 0
250 * Callouts should be moved into struct tcp directly. They are currently
251 * separate because the tcpcb structure is exported to userland for sysctl
252 * parsing purposes, which do not know about callouts.
261 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
262 #define V_tcpcb_zone VNET(tcpcb_zone)
264 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
265 static struct mtx isn_mtx;
267 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
268 #define ISN_LOCK() mtx_lock(&isn_mtx)
269 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
272 * TCP initialization.
275 tcp_zone_change(void *tag)
278 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
279 uma_zone_set_max(V_tcpcb_zone, maxsockets);
280 tcp_tw_zone_change();
284 tcp_inpcb_init(void *mem, int size, int flags)
286 struct inpcb *inp = mem;
288 INP_LOCK_INIT(inp, "inp", "tcpinp");
293 * Take a value and get the next power of 2 that doesn't overflow.
294 * Used to size the tcp_inpcb hash buckets.
297 maketcp_hashsize(int size)
303 * get the next power of 2 higher than maxsockets.
305 hashsize = 1 << fls(size);
306 /* catch overflow, and just go one power of 2 smaller */
307 if (hashsize < size) {
308 hashsize = 1 << (fls(size) - 1);
316 const char *tcbhash_tuneable;
319 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
321 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
322 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 printf("%s: WARNING: unable to register helper hook\n", __func__);
324 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
325 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
326 printf("%s: WARNING: unable to register helper hook\n", __func__);
328 hashsize = TCBHASHSIZE;
329 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
332 * Auto tune the hash size based on maxsockets.
333 * A perfect hash would have a 1:1 mapping
334 * (hashsize = maxsockets) however it's been
335 * suggested that O(2) average is better.
337 hashsize = maketcp_hashsize(maxsockets / 4);
339 * Our historical default is 512,
340 * do not autotune lower than this.
345 printf("%s: %s auto tuned to %d\n", __func__,
346 tcbhash_tuneable, hashsize);
349 * We require a hashsize to be a power of two.
350 * Previously if it was not a power of two we would just reset it
351 * back to 512, which could be a nasty surprise if you did not notice
353 * Instead what we do is clip it to the closest power of two lower
354 * than the specified hash value.
356 if (!powerof2(hashsize)) {
357 int oldhashsize = hashsize;
359 hashsize = maketcp_hashsize(hashsize);
360 /* prevent absurdly low value */
363 printf("%s: WARNING: TCB hash size not a power of 2, "
364 "clipped from %d to %d.\n", __func__, oldhashsize,
367 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
368 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
369 IPI_HASHFIELDS_4TUPLE);
372 * These have to be type stable for the benefit of the timers.
374 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
375 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
376 uma_zone_set_max(V_tcpcb_zone, maxsockets);
377 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
383 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
384 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
385 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
387 /* Skip initialization of globals for non-default instances. */
388 if (!IS_DEFAULT_VNET(curvnet))
391 tcp_reass_global_init();
393 /* XXX virtualize those bellow? */
394 tcp_delacktime = TCPTV_DELACK;
395 tcp_keepinit = TCPTV_KEEP_INIT;
396 tcp_keepidle = TCPTV_KEEP_IDLE;
397 tcp_keepintvl = TCPTV_KEEPINTVL;
398 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
400 tcp_rexmit_min = TCPTV_MIN;
401 if (tcp_rexmit_min < 1)
403 tcp_rexmit_slop = TCPTV_CPU_VAR;
404 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
405 tcp_tcbhashsize = hashsize;
407 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
408 if (tcp_soreceive_stream) {
410 tcp_usrreqs.pru_soreceive = soreceive_stream;
413 tcp6_usrreqs.pru_soreceive = soreceive_stream;
418 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
420 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
422 if (max_protohdr < TCP_MINPROTOHDR)
423 max_protohdr = TCP_MINPROTOHDR;
424 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
426 #undef TCP_MINPROTOHDR
429 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
430 SHUTDOWN_PRI_DEFAULT);
431 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
432 EVENTHANDLER_PRI_ANY);
445 tcp_fastopen_destroy();
450 in_pcbinfo_destroy(&V_tcbinfo);
451 uma_zdestroy(V_sack_hole_zone);
452 uma_zdestroy(V_tcpcb_zone);
463 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
464 * tcp_template used to store this data in mbufs, but we now recopy it out
465 * of the tcpcb each time to conserve mbufs.
468 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
470 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
472 INP_WLOCK_ASSERT(inp);
475 if ((inp->inp_vflag & INP_IPV6) != 0) {
478 ip6 = (struct ip6_hdr *)ip_ptr;
479 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
480 (inp->inp_flow & IPV6_FLOWINFO_MASK);
481 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
482 (IPV6_VERSION & IPV6_VERSION_MASK);
483 ip6->ip6_nxt = IPPROTO_TCP;
484 ip6->ip6_plen = htons(sizeof(struct tcphdr));
485 ip6->ip6_src = inp->in6p_laddr;
486 ip6->ip6_dst = inp->in6p_faddr;
489 #if defined(INET6) && defined(INET)
496 ip = (struct ip *)ip_ptr;
497 ip->ip_v = IPVERSION;
499 ip->ip_tos = inp->inp_ip_tos;
503 ip->ip_ttl = inp->inp_ip_ttl;
505 ip->ip_p = IPPROTO_TCP;
506 ip->ip_src = inp->inp_laddr;
507 ip->ip_dst = inp->inp_faddr;
510 th->th_sport = inp->inp_lport;
511 th->th_dport = inp->inp_fport;
519 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
523 * Create template to be used to send tcp packets on a connection.
524 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
525 * use for this function is in keepalives, which use tcp_respond.
528 tcpip_maketemplate(struct inpcb *inp)
532 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
535 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
540 * Send a single message to the TCP at address specified by
541 * the given TCP/IP header. If m == NULL, then we make a copy
542 * of the tcpiphdr at ti and send directly to the addressed host.
543 * This is used to force keep alive messages out using the TCP
544 * template for a connection. If flags are given then we send
545 * a message back to the TCP which originated the * segment ti,
546 * and discard the mbuf containing it and any other attached mbufs.
548 * In any case the ack and sequence number of the transmitted
549 * segment are as specified by the parameters.
551 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
554 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
555 tcp_seq ack, tcp_seq seq, int flags)
568 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
571 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
578 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
579 INP_WLOCK_ASSERT(inp);
584 if (!(flags & TH_RST)) {
585 win = sbspace(&inp->inp_socket->so_rcv);
586 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
587 win = (long)TCP_MAXWIN << tp->rcv_scale;
591 m = m_gethdr(M_NOWAIT, MT_DATA);
595 m->m_data += max_linkhdr;
598 bcopy((caddr_t)ip6, mtod(m, caddr_t),
599 sizeof(struct ip6_hdr));
600 ip6 = mtod(m, struct ip6_hdr *);
601 nth = (struct tcphdr *)(ip6 + 1);
605 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
606 ip = mtod(m, struct ip *);
607 nth = (struct tcphdr *)(ip + 1);
609 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
614 * XXX MRT We inherrit the FIB, which is lucky.
618 m->m_data = (caddr_t)ipgen;
619 /* m_len is set later */
621 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
624 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
625 nth = (struct tcphdr *)(ip6 + 1);
629 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
630 nth = (struct tcphdr *)(ip + 1);
634 * this is usually a case when an extension header
635 * exists between the IPv6 header and the
638 nth->th_sport = th->th_sport;
639 nth->th_dport = th->th_dport;
641 xchg(nth->th_dport, nth->th_sport, uint16_t);
647 ip6->ip6_vfc = IPV6_VERSION;
648 ip6->ip6_nxt = IPPROTO_TCP;
649 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
650 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
653 #if defined(INET) && defined(INET6)
658 tlen += sizeof (struct tcpiphdr);
659 ip->ip_len = htons(tlen);
660 ip->ip_ttl = V_ip_defttl;
661 if (V_path_mtu_discovery)
662 ip->ip_off |= htons(IP_DF);
666 m->m_pkthdr.len = tlen;
667 m->m_pkthdr.rcvif = NULL;
671 * Packet is associated with a socket, so allow the
672 * label of the response to reflect the socket label.
674 INP_WLOCK_ASSERT(inp);
675 mac_inpcb_create_mbuf(inp, m);
678 * Packet is not associated with a socket, so possibly
679 * update the label in place.
681 mac_netinet_tcp_reply(m);
684 nth->th_seq = htonl(seq);
685 nth->th_ack = htonl(ack);
687 nth->th_off = sizeof (struct tcphdr) >> 2;
688 nth->th_flags = flags;
690 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
692 nth->th_win = htons((u_short)win);
695 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
698 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
699 nth->th_sum = in6_cksum_pseudo(ip6,
700 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
701 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
705 #if defined(INET6) && defined(INET)
710 m->m_pkthdr.csum_flags = CSUM_TCP;
711 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
712 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
716 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
717 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
720 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
723 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
726 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
728 #if defined(INET) && defined(INET6)
732 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
737 * Create a new TCP control block, making an
738 * empty reassembly queue and hooking it to the argument
739 * protocol control block. The `inp' parameter must have
740 * come from the zone allocator set up in tcp_init().
743 tcp_newtcpcb(struct inpcb *inp)
745 struct tcpcb_mem *tm;
748 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
751 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
756 /* Initialise cc_var struct for this tcpcb. */
758 tp->ccv->type = IPPROTO_TCP;
759 tp->ccv->ccvc.tcp = tp;
762 * Use the current system default CC algorithm.
765 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
766 CC_ALGO(tp) = CC_DEFAULT();
769 if (CC_ALGO(tp)->cb_init != NULL)
770 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
771 uma_zfree(V_tcpcb_zone, tm);
776 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
777 uma_zfree(V_tcpcb_zone, tm);
782 tp->t_vnet = inp->inp_vnet;
784 tp->t_timers = &tm->tt;
785 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
786 tp->t_maxseg = tp->t_maxopd =
788 isipv6 ? V_tcp_v6mssdflt :
792 /* Set up our timeouts. */
793 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
794 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
795 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
796 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
797 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
799 if (V_tcp_do_rfc1323)
800 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
802 tp->t_flags |= TF_SACK_PERMIT;
803 TAILQ_INIT(&tp->snd_holes);
805 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
808 in_pcbref(inp); /* Reference for tcpcb */
812 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
813 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
814 * reasonable initial retransmit time.
816 tp->t_srtt = TCPTV_SRTTBASE;
817 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
818 tp->t_rttmin = tcp_rexmit_min;
819 tp->t_rxtcur = TCPTV_RTOBASE;
820 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
821 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
822 tp->t_rcvtime = ticks;
824 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
825 * because the socket may be bound to an IPv6 wildcard address,
826 * which may match an IPv4-mapped IPv6 address.
828 inp->inp_ip_ttl = V_ip_defttl;
830 return (tp); /* XXX */
834 * Switch the congestion control algorithm back to NewReno for any active
835 * control blocks using an algorithm which is about to go away.
836 * This ensures the CC framework can allow the unload to proceed without leaving
837 * any dangling pointers which would trigger a panic.
838 * Returning non-zero would inform the CC framework that something went wrong
839 * and it would be unsafe to allow the unload to proceed. However, there is no
840 * way for this to occur with this implementation so we always return zero.
843 tcp_ccalgounload(struct cc_algo *unload_algo)
845 struct cc_algo *tmpalgo;
848 VNET_ITERATOR_DECL(vnet_iter);
851 * Check all active control blocks across all network stacks and change
852 * any that are using "unload_algo" back to NewReno. If "unload_algo"
853 * requires cleanup code to be run, call it.
856 VNET_FOREACH(vnet_iter) {
857 CURVNET_SET(vnet_iter);
858 INP_INFO_RLOCK(&V_tcbinfo);
860 * New connections already part way through being initialised
861 * with the CC algo we're removing will not race with this code
862 * because the INP_INFO_WLOCK is held during initialisation. We
863 * therefore don't enter the loop below until the connection
864 * list has stabilised.
866 LIST_FOREACH(inp, &V_tcb, inp_list) {
868 /* Important to skip tcptw structs. */
869 if (!(inp->inp_flags & INP_TIMEWAIT) &&
870 (tp = intotcpcb(inp)) != NULL) {
872 * By holding INP_WLOCK here, we are assured
873 * that the connection is not currently
874 * executing inside the CC module's functions
875 * i.e. it is safe to make the switch back to
878 if (CC_ALGO(tp) == unload_algo) {
879 tmpalgo = CC_ALGO(tp);
880 /* NewReno does not require any init. */
881 CC_ALGO(tp) = &newreno_cc_algo;
882 if (tmpalgo->cb_destroy != NULL)
883 tmpalgo->cb_destroy(tp->ccv);
888 INP_INFO_RUNLOCK(&V_tcbinfo);
897 * Drop a TCP connection, reporting
898 * the specified error. If connection is synchronized,
899 * then send a RST to peer.
902 tcp_drop(struct tcpcb *tp, int errno)
904 struct socket *so = tp->t_inpcb->inp_socket;
906 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
907 INP_WLOCK_ASSERT(tp->t_inpcb);
909 if (TCPS_HAVERCVDSYN(tp->t_state)) {
910 tcp_state_change(tp, TCPS_CLOSED);
911 (void) tcp_output(tp);
912 TCPSTAT_INC(tcps_drops);
914 TCPSTAT_INC(tcps_conndrops);
915 if (errno == ETIMEDOUT && tp->t_softerror)
916 errno = tp->t_softerror;
917 so->so_error = errno;
918 return (tcp_close(tp));
922 tcp_discardcb(struct tcpcb *tp)
924 struct inpcb *inp = tp->t_inpcb;
925 struct socket *so = inp->inp_socket;
927 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
931 INP_WLOCK_ASSERT(inp);
934 * Make sure that all of our timers are stopped before we delete the
937 * If stopping a timer fails, we schedule a discard function in same
938 * callout, and the last discard function called will take care of
939 * deleting the tcpcb.
941 tcp_timer_stop(tp, TT_REXMT);
942 tcp_timer_stop(tp, TT_PERSIST);
943 tcp_timer_stop(tp, TT_KEEP);
944 tcp_timer_stop(tp, TT_2MSL);
945 tcp_timer_stop(tp, TT_DELACK);
948 * If we got enough samples through the srtt filter,
949 * save the rtt and rttvar in the routing entry.
950 * 'Enough' is arbitrarily defined as 4 rtt samples.
951 * 4 samples is enough for the srtt filter to converge
952 * to within enough % of the correct value; fewer samples
953 * and we could save a bogus rtt. The danger is not high
954 * as tcp quickly recovers from everything.
955 * XXX: Works very well but needs some more statistics!
957 if (tp->t_rttupdated >= 4) {
958 struct hc_metrics_lite metrics;
961 bzero(&metrics, sizeof(metrics));
963 * Update the ssthresh always when the conditions below
964 * are satisfied. This gives us better new start value
965 * for the congestion avoidance for new connections.
966 * ssthresh is only set if packet loss occured on a session.
968 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
969 * being torn down. Ideally this code would not use 'so'.
971 ssthresh = tp->snd_ssthresh;
972 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
974 * convert the limit from user data bytes to
975 * packets then to packet data bytes.
977 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
980 ssthresh *= (u_long)(tp->t_maxseg +
982 (isipv6 ? sizeof (struct ip6_hdr) +
983 sizeof (struct tcphdr) :
985 sizeof (struct tcpiphdr)
992 metrics.rmx_ssthresh = ssthresh;
994 metrics.rmx_rtt = tp->t_srtt;
995 metrics.rmx_rttvar = tp->t_rttvar;
996 metrics.rmx_cwnd = tp->snd_cwnd;
997 metrics.rmx_sendpipe = 0;
998 metrics.rmx_recvpipe = 0;
1000 tcp_hc_update(&inp->inp_inc, &metrics);
1003 /* free the reassembly queue, if any */
1004 tcp_reass_flush(tp);
1007 /* Disconnect offload device, if any. */
1008 if (tp->t_flags & TF_TOE)
1009 tcp_offload_detach(tp);
1012 tcp_free_sackholes(tp);
1014 /* Allow the CC algorithm to clean up after itself. */
1015 if (CC_ALGO(tp)->cb_destroy != NULL)
1016 CC_ALGO(tp)->cb_destroy(tp->ccv);
1018 khelp_destroy_osd(tp->osd);
1021 inp->inp_ppcb = NULL;
1022 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1023 /* We own the last reference on tcpcb, let's free it. */
1025 uma_zfree(V_tcpcb_zone, tp);
1026 released = in_pcbrele_wlocked(inp);
1027 KASSERT(!released, ("%s: inp %p should not have been released "
1028 "here", __func__, inp));
1033 tcp_timer_2msl_discard(void *xtp)
1036 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1040 tcp_timer_keep_discard(void *xtp)
1043 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1047 tcp_timer_persist_discard(void *xtp)
1050 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1054 tcp_timer_rexmt_discard(void *xtp)
1057 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1061 tcp_timer_delack_discard(void *xtp)
1064 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1068 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1072 CURVNET_SET(tp->t_vnet);
1073 INP_INFO_WLOCK(&V_tcbinfo);
1075 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1078 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1079 ("%s: tcpcb has to be stopped here", __func__));
1080 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1081 ("%s: discard callout should be running", __func__));
1082 tp->t_timers->tt_flags &= ~timer_type;
1083 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1084 /* We own the last reference on this tcpcb, let's free it. */
1086 uma_zfree(V_tcpcb_zone, tp);
1087 if (in_pcbrele_wlocked(inp)) {
1088 INP_INFO_WUNLOCK(&V_tcbinfo);
1094 INP_INFO_WUNLOCK(&V_tcbinfo);
1099 * Attempt to close a TCP control block, marking it as dropped, and freeing
1100 * the socket if we hold the only reference.
1103 tcp_close(struct tcpcb *tp)
1105 struct inpcb *inp = tp->t_inpcb;
1108 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1109 INP_WLOCK_ASSERT(inp);
1112 if (tp->t_state == TCPS_LISTEN)
1113 tcp_offload_listen_stop(tp);
1117 * This releases the TFO pending counter resource for TFO listen
1118 * sockets as well as passively-created TFO sockets that transition
1119 * from SYN_RECEIVED to CLOSED.
1121 if (tp->t_tfo_pending) {
1122 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1123 tp->t_tfo_pending = NULL;
1127 TCPSTAT_INC(tcps_closed);
1128 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1129 so = inp->inp_socket;
1130 soisdisconnected(so);
1131 if (inp->inp_flags & INP_SOCKREF) {
1132 KASSERT(so->so_state & SS_PROTOREF,
1133 ("tcp_close: !SS_PROTOREF"));
1134 inp->inp_flags &= ~INP_SOCKREF;
1138 so->so_state &= ~SS_PROTOREF;
1148 VNET_ITERATOR_DECL(vnet_iter);
1153 VNET_LIST_RLOCK_NOSLEEP();
1154 VNET_FOREACH(vnet_iter) {
1155 CURVNET_SET(vnet_iter);
1160 * Walk the tcpbs, if existing, and flush the reassembly queue,
1161 * if there is one...
1162 * XXX: The "Net/3" implementation doesn't imply that the TCP
1163 * reassembly queue should be flushed, but in a situation
1164 * where we're really low on mbufs, this is potentially
1167 INP_INFO_RLOCK(&V_tcbinfo);
1168 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1169 if (inpb->inp_flags & INP_TIMEWAIT)
1172 if ((tcpb = intotcpcb(inpb)) != NULL) {
1173 tcp_reass_flush(tcpb);
1174 tcp_clean_sackreport(tcpb);
1178 INP_INFO_RUNLOCK(&V_tcbinfo);
1181 VNET_LIST_RUNLOCK_NOSLEEP();
1185 * Notify a tcp user of an asynchronous error;
1186 * store error as soft error, but wake up user
1187 * (for now, won't do anything until can select for soft error).
1189 * Do not wake up user since there currently is no mechanism for
1190 * reporting soft errors (yet - a kqueue filter may be added).
1192 static struct inpcb *
1193 tcp_notify(struct inpcb *inp, int error)
1197 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1198 INP_WLOCK_ASSERT(inp);
1200 if ((inp->inp_flags & INP_TIMEWAIT) ||
1201 (inp->inp_flags & INP_DROPPED))
1204 tp = intotcpcb(inp);
1205 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1208 * Ignore some errors if we are hooked up.
1209 * If connection hasn't completed, has retransmitted several times,
1210 * and receives a second error, give up now. This is better
1211 * than waiting a long time to establish a connection that
1212 * can never complete.
1214 if (tp->t_state == TCPS_ESTABLISHED &&
1215 (error == EHOSTUNREACH || error == ENETUNREACH ||
1216 error == EHOSTDOWN)) {
1218 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1220 tp = tcp_drop(tp, error);
1226 tp->t_softerror = error;
1230 wakeup( &so->so_timeo);
1237 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1239 int error, i, m, n, pcb_count;
1240 struct inpcb *inp, **inp_list;
1245 * The process of preparing the TCB list is too time-consuming and
1246 * resource-intensive to repeat twice on every request.
1248 if (req->oldptr == NULL) {
1249 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1250 n += imax(n / 8, 10);
1251 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1255 if (req->newptr != NULL)
1259 * OK, now we're committed to doing something.
1261 INP_INFO_RLOCK(&V_tcbinfo);
1262 gencnt = V_tcbinfo.ipi_gencnt;
1263 n = V_tcbinfo.ipi_count;
1264 INP_INFO_RUNLOCK(&V_tcbinfo);
1266 m = syncache_pcbcount();
1268 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1269 + (n + m) * sizeof(struct xtcpcb));
1273 xig.xig_len = sizeof xig;
1274 xig.xig_count = n + m;
1275 xig.xig_gen = gencnt;
1276 xig.xig_sogen = so_gencnt;
1277 error = SYSCTL_OUT(req, &xig, sizeof xig);
1281 error = syncache_pcblist(req, m, &pcb_count);
1285 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1286 if (inp_list == NULL)
1289 INP_INFO_RLOCK(&V_tcbinfo);
1290 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1291 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1293 if (inp->inp_gencnt <= gencnt) {
1295 * XXX: This use of cr_cansee(), introduced with
1296 * TCP state changes, is not quite right, but for
1297 * now, better than nothing.
1299 if (inp->inp_flags & INP_TIMEWAIT) {
1300 if (intotw(inp) != NULL)
1301 error = cr_cansee(req->td->td_ucred,
1302 intotw(inp)->tw_cred);
1304 error = EINVAL; /* Skip this inp. */
1306 error = cr_canseeinpcb(req->td->td_ucred, inp);
1309 inp_list[i++] = inp;
1314 INP_INFO_RUNLOCK(&V_tcbinfo);
1318 for (i = 0; i < n; i++) {
1321 if (inp->inp_gencnt <= gencnt) {
1325 bzero(&xt, sizeof(xt));
1326 xt.xt_len = sizeof xt;
1327 /* XXX should avoid extra copy */
1328 bcopy(inp, &xt.xt_inp, sizeof *inp);
1329 inp_ppcb = inp->inp_ppcb;
1330 if (inp_ppcb == NULL)
1331 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1332 else if (inp->inp_flags & INP_TIMEWAIT) {
1333 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1334 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1336 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1337 if (xt.xt_tp.t_timers)
1338 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1340 if (inp->inp_socket != NULL)
1341 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1343 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1344 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1346 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1348 error = SYSCTL_OUT(req, &xt, sizeof xt);
1352 INP_INFO_WLOCK(&V_tcbinfo);
1353 for (i = 0; i < n; i++) {
1356 if (!in_pcbrele_rlocked(inp))
1359 INP_INFO_WUNLOCK(&V_tcbinfo);
1363 * Give the user an updated idea of our state.
1364 * If the generation differs from what we told
1365 * her before, she knows that something happened
1366 * while we were processing this request, and it
1367 * might be necessary to retry.
1369 INP_INFO_RLOCK(&V_tcbinfo);
1370 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1371 xig.xig_sogen = so_gencnt;
1372 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1373 INP_INFO_RUNLOCK(&V_tcbinfo);
1374 error = SYSCTL_OUT(req, &xig, sizeof xig);
1376 free(inp_list, M_TEMP);
1380 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1381 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1382 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1386 tcp_getcred(SYSCTL_HANDLER_ARGS)
1389 struct sockaddr_in addrs[2];
1393 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1396 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1399 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1400 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1402 if (inp->inp_socket == NULL)
1405 error = cr_canseeinpcb(req->td->td_ucred, inp);
1407 cru2x(inp->inp_cred, &xuc);
1412 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1416 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1417 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1418 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1423 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1426 struct sockaddr_in6 addrs[2];
1433 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1436 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1439 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1440 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1443 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1445 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1454 inp = in_pcblookup(&V_tcbinfo,
1455 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1457 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1458 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1461 inp = in6_pcblookup(&V_tcbinfo,
1462 &addrs[1].sin6_addr, addrs[1].sin6_port,
1463 &addrs[0].sin6_addr, addrs[0].sin6_port,
1464 INPLOOKUP_RLOCKPCB, NULL);
1466 if (inp->inp_socket == NULL)
1469 error = cr_canseeinpcb(req->td->td_ucred, inp);
1471 cru2x(inp->inp_cred, &xuc);
1476 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1480 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1481 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1482 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1488 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1490 struct ip *ip = vip;
1492 struct in_addr faddr;
1495 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1497 struct in_conninfo inc;
1498 tcp_seq icmp_tcp_seq;
1501 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1502 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1505 if (cmd == PRC_MSGSIZE)
1506 notify = tcp_mtudisc_notify;
1507 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1508 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1509 notify = tcp_drop_syn_sent;
1511 * Redirects don't need to be handled up here.
1513 else if (PRC_IS_REDIRECT(cmd))
1516 * Source quench is depreciated.
1518 else if (cmd == PRC_QUENCH)
1521 * Hostdead is ugly because it goes linearly through all PCBs.
1522 * XXX: We never get this from ICMP, otherwise it makes an
1523 * excellent DoS attack on machines with many connections.
1525 else if (cmd == PRC_HOSTDEAD)
1527 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1530 icp = (struct icmp *)((caddr_t)ip
1531 - offsetof(struct icmp, icmp_ip));
1532 th = (struct tcphdr *)((caddr_t)ip
1533 + (ip->ip_hl << 2));
1534 INP_INFO_WLOCK(&V_tcbinfo);
1535 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1536 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1538 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1539 !(inp->inp_flags & INP_DROPPED) &&
1540 !(inp->inp_socket == NULL)) {
1541 icmp_tcp_seq = htonl(th->th_seq);
1542 tp = intotcpcb(inp);
1543 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1544 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1545 if (cmd == PRC_MSGSIZE) {
1548 * If we got a needfrag set the MTU
1549 * in the route to the suggested new
1550 * value (if given) and then notify.
1552 bzero(&inc, sizeof(inc));
1553 inc.inc_faddr = faddr;
1555 inp->inp_inc.inc_fibnum;
1557 mtu = ntohs(icp->icmp_nextmtu);
1559 * If no alternative MTU was
1560 * proposed, try the next smaller
1565 ntohs(ip->ip_len), 1);
1566 if (mtu < V_tcp_minmss
1567 + sizeof(struct tcpiphdr))
1569 + sizeof(struct tcpiphdr);
1571 * Only cache the MTU if it
1572 * is smaller than the interface
1573 * or route MTU. tcp_mtudisc()
1574 * will do right thing by itself.
1576 if (mtu <= tcp_maxmtu(&inc, NULL))
1577 tcp_hc_updatemtu(&inc, mtu);
1578 tcp_mtudisc(inp, mtu);
1580 inp = (*notify)(inp,
1581 inetctlerrmap[cmd]);
1587 bzero(&inc, sizeof(inc));
1588 inc.inc_fport = th->th_dport;
1589 inc.inc_lport = th->th_sport;
1590 inc.inc_faddr = faddr;
1591 inc.inc_laddr = ip->ip_src;
1592 syncache_unreach(&inc, th);
1594 INP_INFO_WUNLOCK(&V_tcbinfo);
1596 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1602 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1605 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1606 struct ip6_hdr *ip6;
1608 struct ip6ctlparam *ip6cp = NULL;
1609 const struct sockaddr_in6 *sa6_src = NULL;
1611 struct tcp_portonly {
1616 if (sa->sa_family != AF_INET6 ||
1617 sa->sa_len != sizeof(struct sockaddr_in6))
1620 if (cmd == PRC_MSGSIZE)
1621 notify = tcp_mtudisc_notify;
1622 else if (!PRC_IS_REDIRECT(cmd) &&
1623 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1625 /* Source quench is depreciated. */
1626 else if (cmd == PRC_QUENCH)
1629 /* if the parameter is from icmp6, decode it. */
1631 ip6cp = (struct ip6ctlparam *)d;
1633 ip6 = ip6cp->ip6c_ip6;
1634 off = ip6cp->ip6c_off;
1635 sa6_src = ip6cp->ip6c_src;
1639 off = 0; /* fool gcc */
1644 struct in_conninfo inc;
1646 * XXX: We assume that when IPV6 is non NULL,
1647 * M and OFF are valid.
1650 /* check if we can safely examine src and dst ports */
1651 if (m->m_pkthdr.len < off + sizeof(*thp))
1654 bzero(&th, sizeof(th));
1655 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1657 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1658 (struct sockaddr *)ip6cp->ip6c_src,
1659 th.th_sport, cmd, NULL, notify);
1661 bzero(&inc, sizeof(inc));
1662 inc.inc_fport = th.th_dport;
1663 inc.inc_lport = th.th_sport;
1664 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1665 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1666 inc.inc_flags |= INC_ISIPV6;
1667 INP_INFO_WLOCK(&V_tcbinfo);
1668 syncache_unreach(&inc, &th);
1669 INP_INFO_WUNLOCK(&V_tcbinfo);
1671 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1672 0, cmd, NULL, notify);
1678 * Following is where TCP initial sequence number generation occurs.
1680 * There are two places where we must use initial sequence numbers:
1681 * 1. In SYN-ACK packets.
1682 * 2. In SYN packets.
1684 * All ISNs for SYN-ACK packets are generated by the syncache. See
1685 * tcp_syncache.c for details.
1687 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1688 * depends on this property. In addition, these ISNs should be
1689 * unguessable so as to prevent connection hijacking. To satisfy
1690 * the requirements of this situation, the algorithm outlined in
1691 * RFC 1948 is used, with only small modifications.
1693 * Implementation details:
1695 * Time is based off the system timer, and is corrected so that it
1696 * increases by one megabyte per second. This allows for proper
1697 * recycling on high speed LANs while still leaving over an hour
1700 * As reading the *exact* system time is too expensive to be done
1701 * whenever setting up a TCP connection, we increment the time
1702 * offset in two ways. First, a small random positive increment
1703 * is added to isn_offset for each connection that is set up.
1704 * Second, the function tcp_isn_tick fires once per clock tick
1705 * and increments isn_offset as necessary so that sequence numbers
1706 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1707 * random positive increments serve only to ensure that the same
1708 * exact sequence number is never sent out twice (as could otherwise
1709 * happen when a port is recycled in less than the system tick
1712 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1713 * between seeding of isn_secret. This is normally set to zero,
1714 * as reseeding should not be necessary.
1716 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1717 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1718 * general, this means holding an exclusive (write) lock.
1721 #define ISN_BYTES_PER_SECOND 1048576
1722 #define ISN_STATIC_INCREMENT 4096
1723 #define ISN_RANDOM_INCREMENT (4096 - 1)
1725 static VNET_DEFINE(u_char, isn_secret[32]);
1726 static VNET_DEFINE(int, isn_last);
1727 static VNET_DEFINE(int, isn_last_reseed);
1728 static VNET_DEFINE(u_int32_t, isn_offset);
1729 static VNET_DEFINE(u_int32_t, isn_offset_old);
1731 #define V_isn_secret VNET(isn_secret)
1732 #define V_isn_last VNET(isn_last)
1733 #define V_isn_last_reseed VNET(isn_last_reseed)
1734 #define V_isn_offset VNET(isn_offset)
1735 #define V_isn_offset_old VNET(isn_offset_old)
1738 tcp_new_isn(struct tcpcb *tp)
1741 u_int32_t md5_buffer[4];
1743 u_int32_t projected_offset;
1745 INP_WLOCK_ASSERT(tp->t_inpcb);
1748 /* Seed if this is the first use, reseed if requested. */
1749 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1750 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1752 read_random(&V_isn_secret, sizeof(V_isn_secret));
1753 V_isn_last_reseed = ticks;
1756 /* Compute the md5 hash and return the ISN. */
1758 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1759 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1761 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1762 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1763 sizeof(struct in6_addr));
1764 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1765 sizeof(struct in6_addr));
1769 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1770 sizeof(struct in_addr));
1771 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1772 sizeof(struct in_addr));
1774 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1775 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1776 new_isn = (tcp_seq) md5_buffer[0];
1777 V_isn_offset += ISN_STATIC_INCREMENT +
1778 (arc4random() & ISN_RANDOM_INCREMENT);
1779 if (ticks != V_isn_last) {
1780 projected_offset = V_isn_offset_old +
1781 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1782 if (SEQ_GT(projected_offset, V_isn_offset))
1783 V_isn_offset = projected_offset;
1784 V_isn_offset_old = V_isn_offset;
1787 new_isn += V_isn_offset;
1793 * When a specific ICMP unreachable message is received and the
1794 * connection state is SYN-SENT, drop the connection. This behavior
1795 * is controlled by the icmp_may_rst sysctl.
1798 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1802 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1803 INP_WLOCK_ASSERT(inp);
1805 if ((inp->inp_flags & INP_TIMEWAIT) ||
1806 (inp->inp_flags & INP_DROPPED))
1809 tp = intotcpcb(inp);
1810 if (tp->t_state != TCPS_SYN_SENT)
1813 tp = tcp_drop(tp, errno);
1821 * When `need fragmentation' ICMP is received, update our idea of the MSS
1822 * based on the new value. Also nudge TCP to send something, since we
1823 * know the packet we just sent was dropped.
1824 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1826 static struct inpcb *
1827 tcp_mtudisc_notify(struct inpcb *inp, int error)
1830 return (tcp_mtudisc(inp, -1));
1834 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1839 INP_WLOCK_ASSERT(inp);
1840 if ((inp->inp_flags & INP_TIMEWAIT) ||
1841 (inp->inp_flags & INP_DROPPED))
1844 tp = intotcpcb(inp);
1845 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1847 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1849 so = inp->inp_socket;
1850 SOCKBUF_LOCK(&so->so_snd);
1851 /* If the mss is larger than the socket buffer, decrease the mss. */
1852 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1853 tp->t_maxseg = so->so_snd.sb_hiwat;
1854 SOCKBUF_UNLOCK(&so->so_snd);
1856 TCPSTAT_INC(tcps_mturesent);
1858 tp->snd_nxt = tp->snd_una;
1859 tcp_free_sackholes(tp);
1860 tp->snd_recover = tp->snd_max;
1861 if (tp->t_flags & TF_SACK_PERMIT)
1862 EXIT_FASTRECOVERY(tp->t_flags);
1869 * Look-up the routing entry to the peer of this inpcb. If no route
1870 * is found and it cannot be allocated, then return 0. This routine
1871 * is called by TCP routines that access the rmx structure and by
1872 * tcp_mss_update to get the peer/interface MTU.
1875 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1878 struct sockaddr_in *dst;
1882 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1884 bzero(&sro, sizeof(sro));
1885 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1886 dst = (struct sockaddr_in *)&sro.ro_dst;
1887 dst->sin_family = AF_INET;
1888 dst->sin_len = sizeof(*dst);
1889 dst->sin_addr = inc->inc_faddr;
1890 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1892 if (sro.ro_rt != NULL) {
1893 ifp = sro.ro_rt->rt_ifp;
1894 if (sro.ro_rt->rt_mtu == 0)
1895 maxmtu = ifp->if_mtu;
1897 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1899 /* Report additional interface capabilities. */
1901 if (ifp->if_capenable & IFCAP_TSO4 &&
1902 ifp->if_hwassist & CSUM_TSO) {
1903 cap->ifcap |= CSUM_TSO;
1904 cap->tsomax = ifp->if_hw_tsomax;
1905 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1906 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1917 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1919 struct route_in6 sro6;
1923 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1925 bzero(&sro6, sizeof(sro6));
1926 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1927 sro6.ro_dst.sin6_family = AF_INET6;
1928 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1929 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1930 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1932 if (sro6.ro_rt != NULL) {
1933 ifp = sro6.ro_rt->rt_ifp;
1934 if (sro6.ro_rt->rt_mtu == 0)
1935 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1937 maxmtu = min(sro6.ro_rt->rt_mtu,
1938 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1940 /* Report additional interface capabilities. */
1942 if (ifp->if_capenable & IFCAP_TSO6 &&
1943 ifp->if_hwassist & CSUM_TSO) {
1944 cap->ifcap |= CSUM_TSO;
1945 cap->tsomax = ifp->if_hw_tsomax;
1946 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1947 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1958 /* compute ESP/AH header size for TCP, including outer IP header. */
1960 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1967 struct ip6_hdr *ip6;
1971 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
1972 (!key_havesp(IPSEC_DIR_OUTBOUND)))
1974 m = m_gethdr(M_NOWAIT, MT_DATA);
1979 if ((inp->inp_vflag & INP_IPV6) != 0) {
1980 ip6 = mtod(m, struct ip6_hdr *);
1981 th = (struct tcphdr *)(ip6 + 1);
1982 m->m_pkthdr.len = m->m_len =
1983 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1984 tcpip_fillheaders(inp, ip6, th);
1985 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1989 ip = mtod(m, struct ip *);
1990 th = (struct tcphdr *)(ip + 1);
1991 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1992 tcpip_fillheaders(inp, ip, th);
1993 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2001 #ifdef TCP_SIGNATURE
2003 * Callback function invoked by m_apply() to digest TCP segment data
2004 * contained within an mbuf chain.
2007 tcp_signature_apply(void *fstate, void *data, u_int len)
2010 MD5Update(fstate, (u_char *)data, len);
2015 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2018 * m pointer to head of mbuf chain
2020 * len length of TCP segment data, excluding options
2021 * optlen length of TCP segment options
2022 * buf pointer to storage for computed MD5 digest
2023 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2025 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2026 * When called from tcp_input(), we can be sure that th_sum has been
2027 * zeroed out and verified already.
2029 * Return 0 if successful, otherwise return -1.
2031 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2032 * search with the destination IP address, and a 'magic SPI' to be
2033 * determined by the application. This is hardcoded elsewhere to 1179
2034 * right now. Another branch of this code exists which uses the SPD to
2035 * specify per-application flows but it is unstable.
2038 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2039 u_char *buf, u_int direction)
2041 union sockaddr_union dst;
2043 struct ippseudo ippseudo;
2049 struct ipovly *ipovly;
2051 struct secasvar *sav;
2054 struct ip6_hdr *ip6;
2055 struct in6_addr in6;
2056 char ip6buf[INET6_ADDRSTRLEN];
2062 KASSERT(m != NULL, ("NULL mbuf chain"));
2063 KASSERT(buf != NULL, ("NULL signature pointer"));
2065 /* Extract the destination from the IP header in the mbuf. */
2066 bzero(&dst, sizeof(union sockaddr_union));
2067 ip = mtod(m, struct ip *);
2069 ip6 = NULL; /* Make the compiler happy. */
2074 dst.sa.sa_len = sizeof(struct sockaddr_in);
2075 dst.sa.sa_family = AF_INET;
2076 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2077 ip->ip_src : ip->ip_dst;
2081 case (IPV6_VERSION >> 4):
2082 ip6 = mtod(m, struct ip6_hdr *);
2083 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2084 dst.sa.sa_family = AF_INET6;
2085 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2086 ip6->ip6_src : ip6->ip6_dst;
2095 /* Look up an SADB entry which matches the address of the peer. */
2096 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2098 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2099 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2101 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2102 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2110 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2112 * XXX The ippseudo header MUST be digested in network byte order,
2113 * or else we'll fail the regression test. Assume all fields we've
2114 * been doing arithmetic on have been in host byte order.
2115 * XXX One cannot depend on ipovly->ih_len here. When called from
2116 * tcp_output(), the underlying ip_len member has not yet been set.
2121 ipovly = (struct ipovly *)ip;
2122 ippseudo.ippseudo_src = ipovly->ih_src;
2123 ippseudo.ippseudo_dst = ipovly->ih_dst;
2124 ippseudo.ippseudo_pad = 0;
2125 ippseudo.ippseudo_p = IPPROTO_TCP;
2126 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2128 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2130 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2131 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2136 * RFC 2385, 2.0 Proposal
2137 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2138 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2139 * extended next header value (to form 32 bits), and 32-bit segment
2141 * Note: Upper-Layer Packet Length comes before Next Header.
2143 case (IPV6_VERSION >> 4):
2145 in6_clearscope(&in6);
2146 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2148 in6_clearscope(&in6);
2149 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2150 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2151 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2153 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2154 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2155 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2157 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2159 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2160 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2171 * Step 2: Update MD5 hash with TCP header, excluding options.
2172 * The TCP checksum must be set to zero.
2174 savecsum = th->th_sum;
2176 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2177 th->th_sum = savecsum;
2180 * Step 3: Update MD5 hash with TCP segment data.
2181 * Use m_apply() to avoid an early m_pullup().
2184 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2187 * Step 4: Update MD5 hash with shared secret.
2189 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2190 MD5Final(buf, &ctx);
2192 key_sa_recordxfer(sav, m);
2198 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2201 * m pointer to head of mbuf chain
2202 * len length of TCP segment data, excluding options
2203 * optlen length of TCP segment options
2204 * buf pointer to storage for computed MD5 digest
2205 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2207 * Return 1 if successful, otherwise return 0.
2210 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2211 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2213 char tmpdigest[TCP_SIGLEN];
2215 if (tcp_sig_checksigs == 0)
2217 if ((tcpbflag & TF_SIGNATURE) == 0) {
2218 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2221 * If this socket is not expecting signature but
2222 * the segment contains signature just fail.
2224 TCPSTAT_INC(tcps_sig_err_sigopt);
2225 TCPSTAT_INC(tcps_sig_rcvbadsig);
2229 /* Signature is not expected, and not present in segment. */
2234 * If this socket is expecting signature but the segment does not
2235 * contain any just fail.
2237 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2238 TCPSTAT_INC(tcps_sig_err_nosigopt);
2239 TCPSTAT_INC(tcps_sig_rcvbadsig);
2242 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2243 IPSEC_DIR_INBOUND) == -1) {
2244 TCPSTAT_INC(tcps_sig_err_buildsig);
2245 TCPSTAT_INC(tcps_sig_rcvbadsig);
2249 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2250 TCPSTAT_INC(tcps_sig_rcvbadsig);
2253 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2256 #endif /* TCP_SIGNATURE */
2259 sysctl_drop(SYSCTL_HANDLER_ARGS)
2261 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2262 struct sockaddr_storage addrs[2];
2266 struct sockaddr_in *fin, *lin;
2268 struct sockaddr_in6 *fin6, *lin6;
2279 if (req->oldptr != NULL || req->oldlen != 0)
2281 if (req->newptr == NULL)
2283 if (req->newlen < sizeof(addrs))
2285 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2289 switch (addrs[0].ss_family) {
2292 fin6 = (struct sockaddr_in6 *)&addrs[0];
2293 lin6 = (struct sockaddr_in6 *)&addrs[1];
2294 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2295 lin6->sin6_len != sizeof(struct sockaddr_in6))
2297 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2298 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2300 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2301 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2302 fin = (struct sockaddr_in *)&addrs[0];
2303 lin = (struct sockaddr_in *)&addrs[1];
2306 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2309 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2316 fin = (struct sockaddr_in *)&addrs[0];
2317 lin = (struct sockaddr_in *)&addrs[1];
2318 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2319 lin->sin_len != sizeof(struct sockaddr_in))
2326 INP_INFO_WLOCK(&V_tcbinfo);
2327 switch (addrs[0].ss_family) {
2330 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2331 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2332 INPLOOKUP_WLOCKPCB, NULL);
2337 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2338 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2343 if (inp->inp_flags & INP_TIMEWAIT) {
2345 * XXXRW: There currently exists a state where an
2346 * inpcb is present, but its timewait state has been
2347 * discarded. For now, don't allow dropping of this
2355 } else if (!(inp->inp_flags & INP_DROPPED) &&
2356 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2357 tp = intotcpcb(inp);
2358 tp = tcp_drop(tp, ECONNABORTED);
2365 INP_INFO_WUNLOCK(&V_tcbinfo);
2369 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2370 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2371 0, sysctl_drop, "", "Drop TCP connection");
2374 * Generate a standardized TCP log line for use throughout the
2375 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2376 * allow use in the interrupt context.
2378 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2379 * NB: The function may return NULL if memory allocation failed.
2381 * Due to header inclusion and ordering limitations the struct ip
2382 * and ip6_hdr pointers have to be passed as void pointers.
2385 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2389 /* Is logging enabled? */
2390 if (tcp_log_in_vain == 0)
2393 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2397 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2401 /* Is logging enabled? */
2402 if (tcp_log_debug == 0)
2405 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2409 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2416 const struct ip6_hdr *ip6;
2418 ip6 = (const struct ip6_hdr *)ip6hdr;
2420 ip = (struct ip *)ip4hdr;
2423 * The log line looks like this:
2424 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2426 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2427 sizeof(PRINT_TH_FLAGS) + 1 +
2429 2 * INET6_ADDRSTRLEN;
2431 2 * INET_ADDRSTRLEN;
2434 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2438 strcat(s, "TCP: [");
2441 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2442 inet_ntoa_r(inc->inc_faddr, sp);
2444 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2446 inet_ntoa_r(inc->inc_laddr, sp);
2448 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2451 ip6_sprintf(sp, &inc->inc6_faddr);
2453 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2455 ip6_sprintf(sp, &inc->inc6_laddr);
2457 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2458 } else if (ip6 && th) {
2459 ip6_sprintf(sp, &ip6->ip6_src);
2461 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2463 ip6_sprintf(sp, &ip6->ip6_dst);
2465 sprintf(sp, "]:%i", ntohs(th->th_dport));
2468 } else if (ip && th) {
2469 inet_ntoa_r(ip->ip_src, sp);
2471 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2473 inet_ntoa_r(ip->ip_dst, sp);
2475 sprintf(sp, "]:%i", ntohs(th->th_dport));
2483 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2484 if (*(s + size - 1) != '\0')
2485 panic("%s: string too long", __func__);
2490 * A subroutine which makes it easy to track TCP state changes with DTrace.
2491 * This function shouldn't be called for t_state initializations that don't
2492 * correspond to actual TCP state transitions.
2495 tcp_state_change(struct tcpcb *tp, int newstate)
2497 #if defined(KDTRACE_HOOKS)
2498 int pstate = tp->t_state;
2501 tp->t_state = newstate;
2502 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);