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_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_var.h>
76 #include <netinet/ip6.h>
77 #include <netinet6/in6_pcb.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
89 #include <netinet6/tcp6_var.h>
91 #include <netinet/tcpip.h>
93 #include <netinet/tcp_debug.h>
96 #include <netinet6/ip6protosw.h>
99 #include <netinet/tcp_offload.h>
103 #include <netipsec/ipsec.h>
104 #include <netipsec/xform.h>
106 #include <netipsec/ipsec6.h>
108 #include <netipsec/key.h>
109 #include <sys/syslog.h>
112 #include <machine/in_cksum.h>
115 #include <security/mac/mac_framework.h>
117 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
119 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
123 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
128 error = sysctl_handle_int(oidp, &new, 0, req);
129 if (error == 0 && req->newptr) {
130 if (new < TCP_MINMSS)
138 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
139 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
140 &sysctl_net_inet_tcp_mss_check, "I",
141 "Default TCP Maximum Segment Size");
145 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
149 new = V_tcp_v6mssdflt;
150 error = sysctl_handle_int(oidp, &new, 0, req);
151 if (error == 0 && req->newptr) {
152 if (new < TCP_MINMSS)
155 V_tcp_v6mssdflt = new;
160 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
161 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
162 &sysctl_net_inet_tcp_mss_v6_check, "I",
163 "Default TCP Maximum Segment Size for IPv6");
167 * Minimum MSS we accept and use. This prevents DoS attacks where
168 * we are forced to a ridiculous low MSS like 20 and send hundreds
169 * of packets instead of one. The effect scales with the available
170 * bandwidth and quickly saturates the CPU and network interface
171 * with packet generation and sending. Set to zero to disable MINMSS
172 * checking. This setting prevents us from sending too small packets.
174 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
176 &VNET_NAME(tcp_minmss), 0,
177 "Minmum TCP Maximum Segment Size");
179 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
180 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_rfc1323), 0,
182 "Enable rfc1323 (high performance TCP) extensions");
184 static int tcp_log_debug = 0;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
186 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
188 static int tcp_tcbhashsize = 0;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
190 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
192 static int do_tcpdrain = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
194 "Enable tcp_drain routine for extra help when low on mbufs");
196 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
197 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
199 static VNET_DEFINE(int, icmp_may_rst) = 1;
200 #define V_icmp_may_rst VNET(icmp_may_rst)
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
202 &VNET_NAME(icmp_may_rst), 0,
203 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
205 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
206 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
207 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
208 &VNET_NAME(tcp_isn_reseed_interval), 0,
209 "Seconds between reseeding of ISN secret");
211 static int tcp_soreceive_stream = 0;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
213 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
216 static int tcp_sig_checksigs = 1;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
218 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
221 VNET_DEFINE(uma_zone_t, sack_hole_zone);
222 #define V_sack_hole_zone VNET(sack_hole_zone)
224 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
226 static struct inpcb *tcp_notify(struct inpcb *, int);
227 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
228 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
229 void *ip4hdr, const void *ip6hdr);
232 * Target size of TCP PCB hash tables. Must be a power of two.
234 * Note that this can be overridden by the kernel environment
235 * variable net.inet.tcp.tcbhashsize
238 #define TCBHASHSIZE 0
243 * Callouts should be moved into struct tcp directly. They are currently
244 * separate because the tcpcb structure is exported to userland for sysctl
245 * parsing purposes, which do not know about callouts.
254 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
255 #define V_tcpcb_zone VNET(tcpcb_zone)
257 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
258 static struct mtx isn_mtx;
260 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
261 #define ISN_LOCK() mtx_lock(&isn_mtx)
262 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
265 * TCP initialization.
268 tcp_zone_change(void *tag)
271 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
272 uma_zone_set_max(V_tcpcb_zone, maxsockets);
273 tcp_tw_zone_change();
277 tcp_inpcb_init(void *mem, int size, int flags)
279 struct inpcb *inp = mem;
281 INP_LOCK_INIT(inp, "inp", "tcpinp");
286 * Take a value and get the next power of 2 that doesn't overflow.
287 * Used to size the tcp_inpcb hash buckets.
290 maketcp_hashsize(int size)
296 * get the next power of 2 higher than maxsockets.
298 hashsize = 1 << fls(size);
299 /* catch overflow, and just go one power of 2 smaller */
300 if (hashsize < size) {
301 hashsize = 1 << (fls(size) - 1);
309 const char *tcbhash_tuneable;
312 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
314 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
315 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
316 printf("%s: WARNING: unable to register helper hook\n", __func__);
317 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
318 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 printf("%s: WARNING: unable to register helper hook\n", __func__);
321 hashsize = TCBHASHSIZE;
322 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
325 * Auto tune the hash size based on maxsockets.
326 * A perfect hash would have a 1:1 mapping
327 * (hashsize = maxsockets) however it's been
328 * suggested that O(2) average is better.
330 hashsize = maketcp_hashsize(maxsockets / 4);
332 * Our historical default is 512,
333 * do not autotune lower than this.
338 printf("%s: %s auto tuned to %d\n", __func__,
339 tcbhash_tuneable, hashsize);
342 * We require a hashsize to be a power of two.
343 * Previously if it was not a power of two we would just reset it
344 * back to 512, which could be a nasty surprise if you did not notice
346 * Instead what we do is clip it to the closest power of two lower
347 * than the specified hash value.
349 if (!powerof2(hashsize)) {
350 int oldhashsize = hashsize;
352 hashsize = maketcp_hashsize(hashsize);
353 /* prevent absurdly low value */
356 printf("%s: WARNING: TCB hash size not a power of 2, "
357 "clipped from %d to %d.\n", __func__, oldhashsize,
360 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
361 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
362 IPI_HASHFIELDS_4TUPLE);
365 * These have to be type stable for the benefit of the timers.
367 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
368 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
369 uma_zone_set_max(V_tcpcb_zone, maxsockets);
370 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
377 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
378 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
379 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
381 /* Skip initialization of globals for non-default instances. */
382 if (!IS_DEFAULT_VNET(curvnet))
385 /* XXX virtualize those bellow? */
386 tcp_delacktime = TCPTV_DELACK;
387 tcp_keepinit = TCPTV_KEEP_INIT;
388 tcp_keepidle = TCPTV_KEEP_IDLE;
389 tcp_keepintvl = TCPTV_KEEPINTVL;
390 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
392 tcp_rexmit_min = TCPTV_MIN;
393 if (tcp_rexmit_min < 1)
395 tcp_rexmit_slop = TCPTV_CPU_VAR;
396 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
397 tcp_tcbhashsize = hashsize;
399 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
400 if (tcp_soreceive_stream) {
402 tcp_usrreqs.pru_soreceive = soreceive_stream;
405 tcp6_usrreqs.pru_soreceive = soreceive_stream;
410 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
412 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
414 if (max_protohdr < TCP_MINPROTOHDR)
415 max_protohdr = TCP_MINPROTOHDR;
416 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
418 #undef TCP_MINPROTOHDR
421 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
422 SHUTDOWN_PRI_DEFAULT);
423 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
424 EVENTHANDLER_PRI_ANY);
436 in_pcbinfo_destroy(&V_tcbinfo);
437 uma_zdestroy(V_sack_hole_zone);
438 uma_zdestroy(V_tcpcb_zone);
449 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
450 * tcp_template used to store this data in mbufs, but we now recopy it out
451 * of the tcpcb each time to conserve mbufs.
454 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
456 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
458 INP_WLOCK_ASSERT(inp);
461 if ((inp->inp_vflag & INP_IPV6) != 0) {
464 ip6 = (struct ip6_hdr *)ip_ptr;
465 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
466 (inp->inp_flow & IPV6_FLOWINFO_MASK);
467 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
468 (IPV6_VERSION & IPV6_VERSION_MASK);
469 ip6->ip6_nxt = IPPROTO_TCP;
470 ip6->ip6_plen = htons(sizeof(struct tcphdr));
471 ip6->ip6_src = inp->in6p_laddr;
472 ip6->ip6_dst = inp->in6p_faddr;
475 #if defined(INET6) && defined(INET)
482 ip = (struct ip *)ip_ptr;
483 ip->ip_v = IPVERSION;
485 ip->ip_tos = inp->inp_ip_tos;
489 ip->ip_ttl = inp->inp_ip_ttl;
491 ip->ip_p = IPPROTO_TCP;
492 ip->ip_src = inp->inp_laddr;
493 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 == (IPV6_VERSION >> 4);
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_NOWAIT, 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 = 0; /* Set in ip6_output(). */
636 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
639 #if defined(INET) && defined(INET6)
644 tlen += sizeof (struct tcpiphdr);
645 ip->ip_len = htons(tlen);
646 ip->ip_ttl = V_ip_defttl;
647 if (V_path_mtu_discovery)
648 ip->ip_off |= htons(IP_DF);
652 m->m_pkthdr.len = tlen;
653 m->m_pkthdr.rcvif = NULL;
657 * Packet is associated with a socket, so allow the
658 * label of the response to reflect the socket label.
660 INP_WLOCK_ASSERT(inp);
661 mac_inpcb_create_mbuf(inp, m);
664 * Packet is not associated with a socket, so possibly
665 * update the label in place.
667 mac_netinet_tcp_reply(m);
670 nth->th_seq = htonl(seq);
671 nth->th_ack = htonl(ack);
673 nth->th_off = sizeof (struct tcphdr) >> 2;
674 nth->th_flags = flags;
676 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
678 nth->th_win = htons((u_short)win);
681 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
684 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
685 nth->th_sum = in6_cksum_pseudo(ip6,
686 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
687 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
691 #if defined(INET6) && defined(INET)
696 m->m_pkthdr.csum_flags = CSUM_TCP;
697 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
698 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
702 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
703 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
707 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
709 #if defined(INET) && defined(INET6)
713 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
718 * Create a new TCP control block, making an
719 * empty reassembly queue and hooking it to the argument
720 * protocol control block. The `inp' parameter must have
721 * come from the zone allocator set up in tcp_init().
724 tcp_newtcpcb(struct inpcb *inp)
726 struct tcpcb_mem *tm;
729 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
732 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
737 /* Initialise cc_var struct for this tcpcb. */
739 tp->ccv->type = IPPROTO_TCP;
740 tp->ccv->ccvc.tcp = tp;
743 * Use the current system default CC algorithm.
746 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
747 CC_ALGO(tp) = CC_DEFAULT();
750 if (CC_ALGO(tp)->cb_init != NULL)
751 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
752 uma_zfree(V_tcpcb_zone, tm);
757 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
758 uma_zfree(V_tcpcb_zone, tm);
763 tp->t_vnet = inp->inp_vnet;
765 tp->t_timers = &tm->tt;
766 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
767 tp->t_maxseg = tp->t_maxopd =
769 isipv6 ? V_tcp_v6mssdflt :
773 /* Set up our timeouts. */
774 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
775 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
776 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
777 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
778 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
780 if (V_tcp_do_rfc1323)
781 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
783 tp->t_flags |= TF_SACK_PERMIT;
784 TAILQ_INIT(&tp->snd_holes);
785 tp->t_inpcb = inp; /* XXX */
787 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
788 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
789 * reasonable initial retransmit time.
791 tp->t_srtt = TCPTV_SRTTBASE;
792 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
793 tp->t_rttmin = tcp_rexmit_min;
794 tp->t_rxtcur = TCPTV_RTOBASE;
795 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
796 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
797 tp->t_rcvtime = ticks;
799 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
800 * because the socket may be bound to an IPv6 wildcard address,
801 * which may match an IPv4-mapped IPv6 address.
803 inp->inp_ip_ttl = V_ip_defttl;
805 return (tp); /* XXX */
809 * Switch the congestion control algorithm back to NewReno for any active
810 * control blocks using an algorithm which is about to go away.
811 * This ensures the CC framework can allow the unload to proceed without leaving
812 * any dangling pointers which would trigger a panic.
813 * Returning non-zero would inform the CC framework that something went wrong
814 * and it would be unsafe to allow the unload to proceed. However, there is no
815 * way for this to occur with this implementation so we always return zero.
818 tcp_ccalgounload(struct cc_algo *unload_algo)
820 struct cc_algo *tmpalgo;
823 VNET_ITERATOR_DECL(vnet_iter);
826 * Check all active control blocks across all network stacks and change
827 * any that are using "unload_algo" back to NewReno. If "unload_algo"
828 * requires cleanup code to be run, call it.
831 VNET_FOREACH(vnet_iter) {
832 CURVNET_SET(vnet_iter);
833 INP_INFO_RLOCK(&V_tcbinfo);
835 * New connections already part way through being initialised
836 * with the CC algo we're removing will not race with this code
837 * because the INP_INFO_WLOCK is held during initialisation. We
838 * therefore don't enter the loop below until the connection
839 * list has stabilised.
841 LIST_FOREACH(inp, &V_tcb, inp_list) {
843 /* Important to skip tcptw structs. */
844 if (!(inp->inp_flags & INP_TIMEWAIT) &&
845 (tp = intotcpcb(inp)) != NULL) {
847 * By holding INP_WLOCK here, we are assured
848 * that the connection is not currently
849 * executing inside the CC module's functions
850 * i.e. it is safe to make the switch back to
853 if (CC_ALGO(tp) == unload_algo) {
854 tmpalgo = CC_ALGO(tp);
855 /* NewReno does not require any init. */
856 CC_ALGO(tp) = &newreno_cc_algo;
857 if (tmpalgo->cb_destroy != NULL)
858 tmpalgo->cb_destroy(tp->ccv);
863 INP_INFO_RUNLOCK(&V_tcbinfo);
872 * Drop a TCP connection, reporting
873 * the specified error. If connection is synchronized,
874 * then send a RST to peer.
877 tcp_drop(struct tcpcb *tp, int errno)
879 struct socket *so = tp->t_inpcb->inp_socket;
881 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
882 INP_WLOCK_ASSERT(tp->t_inpcb);
884 if (TCPS_HAVERCVDSYN(tp->t_state)) {
885 tp->t_state = TCPS_CLOSED;
886 (void) tcp_output(tp);
887 TCPSTAT_INC(tcps_drops);
889 TCPSTAT_INC(tcps_conndrops);
890 if (errno == ETIMEDOUT && tp->t_softerror)
891 errno = tp->t_softerror;
892 so->so_error = errno;
893 return (tcp_close(tp));
897 tcp_discardcb(struct tcpcb *tp)
899 struct inpcb *inp = tp->t_inpcb;
900 struct socket *so = inp->inp_socket;
902 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
905 INP_WLOCK_ASSERT(inp);
908 * Make sure that all of our timers are stopped before we delete the
911 * XXXRW: Really, we would like to use callout_drain() here in order
912 * to avoid races experienced in tcp_timer.c where a timer is already
913 * executing at this point. However, we can't, both because we're
914 * running in a context where we can't sleep, and also because we
915 * hold locks required by the timers. What we instead need to do is
916 * test to see if callout_drain() is required, and if so, defer some
917 * portion of the remainder of tcp_discardcb() to an asynchronous
918 * context that can callout_drain() and then continue. Some care
919 * will be required to ensure that no further processing takes place
920 * on the tcpcb, even though it hasn't been freed (a flag?).
922 callout_stop(&tp->t_timers->tt_rexmt);
923 callout_stop(&tp->t_timers->tt_persist);
924 callout_stop(&tp->t_timers->tt_keep);
925 callout_stop(&tp->t_timers->tt_2msl);
926 callout_stop(&tp->t_timers->tt_delack);
929 * If we got enough samples through the srtt filter,
930 * save the rtt and rttvar in the routing entry.
931 * 'Enough' is arbitrarily defined as 4 rtt samples.
932 * 4 samples is enough for the srtt filter to converge
933 * to within enough % of the correct value; fewer samples
934 * and we could save a bogus rtt. The danger is not high
935 * as tcp quickly recovers from everything.
936 * XXX: Works very well but needs some more statistics!
938 if (tp->t_rttupdated >= 4) {
939 struct hc_metrics_lite metrics;
942 bzero(&metrics, sizeof(metrics));
944 * Update the ssthresh always when the conditions below
945 * are satisfied. This gives us better new start value
946 * for the congestion avoidance for new connections.
947 * ssthresh is only set if packet loss occured on a session.
949 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
950 * being torn down. Ideally this code would not use 'so'.
952 ssthresh = tp->snd_ssthresh;
953 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
955 * convert the limit from user data bytes to
956 * packets then to packet data bytes.
958 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
961 ssthresh *= (u_long)(tp->t_maxseg +
963 (isipv6 ? sizeof (struct ip6_hdr) +
964 sizeof (struct tcphdr) :
966 sizeof (struct tcpiphdr)
973 metrics.rmx_ssthresh = ssthresh;
975 metrics.rmx_rtt = tp->t_srtt;
976 metrics.rmx_rttvar = tp->t_rttvar;
977 metrics.rmx_cwnd = tp->snd_cwnd;
978 metrics.rmx_sendpipe = 0;
979 metrics.rmx_recvpipe = 0;
981 tcp_hc_update(&inp->inp_inc, &metrics);
984 /* free the reassembly queue, if any */
988 /* Disconnect offload device, if any. */
989 if (tp->t_flags & TF_TOE)
990 tcp_offload_detach(tp);
993 tcp_free_sackholes(tp);
995 /* Allow the CC algorithm to clean up after itself. */
996 if (CC_ALGO(tp)->cb_destroy != NULL)
997 CC_ALGO(tp)->cb_destroy(tp->ccv);
999 khelp_destroy_osd(tp->osd);
1002 inp->inp_ppcb = NULL;
1004 uma_zfree(V_tcpcb_zone, tp);
1008 * Attempt to close a TCP control block, marking it as dropped, and freeing
1009 * the socket if we hold the only reference.
1012 tcp_close(struct tcpcb *tp)
1014 struct inpcb *inp = tp->t_inpcb;
1017 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1018 INP_WLOCK_ASSERT(inp);
1021 if (tp->t_state == TCPS_LISTEN)
1022 tcp_offload_listen_stop(tp);
1025 TCPSTAT_INC(tcps_closed);
1026 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1027 so = inp->inp_socket;
1028 soisdisconnected(so);
1029 if (inp->inp_flags & INP_SOCKREF) {
1030 KASSERT(so->so_state & SS_PROTOREF,
1031 ("tcp_close: !SS_PROTOREF"));
1032 inp->inp_flags &= ~INP_SOCKREF;
1036 so->so_state &= ~SS_PROTOREF;
1046 VNET_ITERATOR_DECL(vnet_iter);
1051 VNET_LIST_RLOCK_NOSLEEP();
1052 VNET_FOREACH(vnet_iter) {
1053 CURVNET_SET(vnet_iter);
1058 * Walk the tcpbs, if existing, and flush the reassembly queue,
1059 * if there is one...
1060 * XXX: The "Net/3" implementation doesn't imply that the TCP
1061 * reassembly queue should be flushed, but in a situation
1062 * where we're really low on mbufs, this is potentially
1065 INP_INFO_RLOCK(&V_tcbinfo);
1066 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1067 if (inpb->inp_flags & INP_TIMEWAIT)
1070 if ((tcpb = intotcpcb(inpb)) != NULL) {
1071 tcp_reass_flush(tcpb);
1072 tcp_clean_sackreport(tcpb);
1076 INP_INFO_RUNLOCK(&V_tcbinfo);
1079 VNET_LIST_RUNLOCK_NOSLEEP();
1083 * Notify a tcp user of an asynchronous error;
1084 * store error as soft error, but wake up user
1085 * (for now, won't do anything until can select for soft error).
1087 * Do not wake up user since there currently is no mechanism for
1088 * reporting soft errors (yet - a kqueue filter may be added).
1090 static struct inpcb *
1091 tcp_notify(struct inpcb *inp, int error)
1095 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1096 INP_WLOCK_ASSERT(inp);
1098 if ((inp->inp_flags & INP_TIMEWAIT) ||
1099 (inp->inp_flags & INP_DROPPED))
1102 tp = intotcpcb(inp);
1103 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1106 * Ignore some errors if we are hooked up.
1107 * If connection hasn't completed, has retransmitted several times,
1108 * and receives a second error, give up now. This is better
1109 * than waiting a long time to establish a connection that
1110 * can never complete.
1112 if (tp->t_state == TCPS_ESTABLISHED &&
1113 (error == EHOSTUNREACH || error == ENETUNREACH ||
1114 error == EHOSTDOWN)) {
1116 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1118 tp = tcp_drop(tp, error);
1124 tp->t_softerror = error;
1128 wakeup( &so->so_timeo);
1135 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1137 int error, i, m, n, pcb_count;
1138 struct inpcb *inp, **inp_list;
1143 * The process of preparing the TCB list is too time-consuming and
1144 * resource-intensive to repeat twice on every request.
1146 if (req->oldptr == NULL) {
1147 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1148 n += imax(n / 8, 10);
1149 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1153 if (req->newptr != NULL)
1157 * OK, now we're committed to doing something.
1159 INP_INFO_RLOCK(&V_tcbinfo);
1160 gencnt = V_tcbinfo.ipi_gencnt;
1161 n = V_tcbinfo.ipi_count;
1162 INP_INFO_RUNLOCK(&V_tcbinfo);
1164 m = syncache_pcbcount();
1166 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1167 + (n + m) * sizeof(struct xtcpcb));
1171 xig.xig_len = sizeof xig;
1172 xig.xig_count = n + m;
1173 xig.xig_gen = gencnt;
1174 xig.xig_sogen = so_gencnt;
1175 error = SYSCTL_OUT(req, &xig, sizeof xig);
1179 error = syncache_pcblist(req, m, &pcb_count);
1183 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1184 if (inp_list == NULL)
1187 INP_INFO_RLOCK(&V_tcbinfo);
1188 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1189 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1191 if (inp->inp_gencnt <= gencnt) {
1193 * XXX: This use of cr_cansee(), introduced with
1194 * TCP state changes, is not quite right, but for
1195 * now, better than nothing.
1197 if (inp->inp_flags & INP_TIMEWAIT) {
1198 if (intotw(inp) != NULL)
1199 error = cr_cansee(req->td->td_ucred,
1200 intotw(inp)->tw_cred);
1202 error = EINVAL; /* Skip this inp. */
1204 error = cr_canseeinpcb(req->td->td_ucred, inp);
1207 inp_list[i++] = inp;
1212 INP_INFO_RUNLOCK(&V_tcbinfo);
1216 for (i = 0; i < n; i++) {
1219 if (inp->inp_gencnt <= gencnt) {
1223 bzero(&xt, sizeof(xt));
1224 xt.xt_len = sizeof xt;
1225 /* XXX should avoid extra copy */
1226 bcopy(inp, &xt.xt_inp, sizeof *inp);
1227 inp_ppcb = inp->inp_ppcb;
1228 if (inp_ppcb == NULL)
1229 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1230 else if (inp->inp_flags & INP_TIMEWAIT) {
1231 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1232 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1234 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1235 if (xt.xt_tp.t_timers)
1236 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1238 if (inp->inp_socket != NULL)
1239 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1241 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1242 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1244 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1246 error = SYSCTL_OUT(req, &xt, sizeof xt);
1250 INP_INFO_WLOCK(&V_tcbinfo);
1251 for (i = 0; i < n; i++) {
1254 if (!in_pcbrele_rlocked(inp))
1257 INP_INFO_WUNLOCK(&V_tcbinfo);
1261 * Give the user an updated idea of our state.
1262 * If the generation differs from what we told
1263 * her before, she knows that something happened
1264 * while we were processing this request, and it
1265 * might be necessary to retry.
1267 INP_INFO_RLOCK(&V_tcbinfo);
1268 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1269 xig.xig_sogen = so_gencnt;
1270 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1271 INP_INFO_RUNLOCK(&V_tcbinfo);
1272 error = SYSCTL_OUT(req, &xig, sizeof xig);
1274 free(inp_list, M_TEMP);
1278 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1279 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1280 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1284 tcp_getcred(SYSCTL_HANDLER_ARGS)
1287 struct sockaddr_in addrs[2];
1291 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1294 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1297 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1298 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1300 if (inp->inp_socket == NULL)
1303 error = cr_canseeinpcb(req->td->td_ucred, inp);
1305 cru2x(inp->inp_cred, &xuc);
1310 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1314 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1315 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1316 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1321 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1324 struct sockaddr_in6 addrs[2];
1331 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1334 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1337 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1338 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1341 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1343 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1352 inp = in_pcblookup(&V_tcbinfo,
1353 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1355 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1356 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1359 inp = in6_pcblookup(&V_tcbinfo,
1360 &addrs[1].sin6_addr, addrs[1].sin6_port,
1361 &addrs[0].sin6_addr, addrs[0].sin6_port,
1362 INPLOOKUP_RLOCKPCB, NULL);
1364 if (inp->inp_socket == NULL)
1367 error = cr_canseeinpcb(req->td->td_ucred, inp);
1369 cru2x(inp->inp_cred, &xuc);
1374 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1378 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1379 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1380 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1386 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1388 struct ip *ip = vip;
1390 struct in_addr faddr;
1393 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1395 struct in_conninfo inc;
1396 tcp_seq icmp_tcp_seq;
1399 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1400 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1403 if (cmd == PRC_MSGSIZE)
1404 notify = tcp_mtudisc_notify;
1405 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1406 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1407 notify = tcp_drop_syn_sent;
1409 * Redirects don't need to be handled up here.
1411 else if (PRC_IS_REDIRECT(cmd))
1414 * Source quench is depreciated.
1416 else if (cmd == PRC_QUENCH)
1419 * Hostdead is ugly because it goes linearly through all PCBs.
1420 * XXX: We never get this from ICMP, otherwise it makes an
1421 * excellent DoS attack on machines with many connections.
1423 else if (cmd == PRC_HOSTDEAD)
1425 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1428 icp = (struct icmp *)((caddr_t)ip
1429 - offsetof(struct icmp, icmp_ip));
1430 th = (struct tcphdr *)((caddr_t)ip
1431 + (ip->ip_hl << 2));
1432 INP_INFO_WLOCK(&V_tcbinfo);
1433 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1434 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1436 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1437 !(inp->inp_flags & INP_DROPPED) &&
1438 !(inp->inp_socket == NULL)) {
1439 icmp_tcp_seq = htonl(th->th_seq);
1440 tp = intotcpcb(inp);
1441 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1442 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1443 if (cmd == PRC_MSGSIZE) {
1446 * If we got a needfrag set the MTU
1447 * in the route to the suggested new
1448 * value (if given) and then notify.
1450 bzero(&inc, sizeof(inc));
1451 inc.inc_faddr = faddr;
1453 inp->inp_inc.inc_fibnum;
1455 mtu = ntohs(icp->icmp_nextmtu);
1457 * If no alternative MTU was
1458 * proposed, try the next smaller
1463 ntohs(ip->ip_len), 1);
1464 if (mtu < V_tcp_minmss
1465 + sizeof(struct tcpiphdr))
1467 + sizeof(struct tcpiphdr);
1469 * Only cache the MTU if it
1470 * is smaller than the interface
1471 * or route MTU. tcp_mtudisc()
1472 * will do right thing by itself.
1474 if (mtu <= tcp_maxmtu(&inc, NULL))
1475 tcp_hc_updatemtu(&inc, mtu);
1476 tcp_mtudisc(inp, mtu);
1478 inp = (*notify)(inp,
1479 inetctlerrmap[cmd]);
1485 bzero(&inc, sizeof(inc));
1486 inc.inc_fport = th->th_dport;
1487 inc.inc_lport = th->th_sport;
1488 inc.inc_faddr = faddr;
1489 inc.inc_laddr = ip->ip_src;
1490 syncache_unreach(&inc, th);
1492 INP_INFO_WUNLOCK(&V_tcbinfo);
1494 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1500 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1503 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1504 struct ip6_hdr *ip6;
1506 struct ip6ctlparam *ip6cp = NULL;
1507 const struct sockaddr_in6 *sa6_src = NULL;
1509 struct tcp_portonly {
1514 if (sa->sa_family != AF_INET6 ||
1515 sa->sa_len != sizeof(struct sockaddr_in6))
1518 if (cmd == PRC_MSGSIZE)
1519 notify = tcp_mtudisc_notify;
1520 else if (!PRC_IS_REDIRECT(cmd) &&
1521 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1523 /* Source quench is depreciated. */
1524 else if (cmd == PRC_QUENCH)
1527 /* if the parameter is from icmp6, decode it. */
1529 ip6cp = (struct ip6ctlparam *)d;
1531 ip6 = ip6cp->ip6c_ip6;
1532 off = ip6cp->ip6c_off;
1533 sa6_src = ip6cp->ip6c_src;
1537 off = 0; /* fool gcc */
1542 struct in_conninfo inc;
1544 * XXX: We assume that when IPV6 is non NULL,
1545 * M and OFF are valid.
1548 /* check if we can safely examine src and dst ports */
1549 if (m->m_pkthdr.len < off + sizeof(*thp))
1552 bzero(&th, sizeof(th));
1553 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1555 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1556 (struct sockaddr *)ip6cp->ip6c_src,
1557 th.th_sport, cmd, NULL, notify);
1559 bzero(&inc, sizeof(inc));
1560 inc.inc_fport = th.th_dport;
1561 inc.inc_lport = th.th_sport;
1562 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1563 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1564 inc.inc_flags |= INC_ISIPV6;
1565 INP_INFO_WLOCK(&V_tcbinfo);
1566 syncache_unreach(&inc, &th);
1567 INP_INFO_WUNLOCK(&V_tcbinfo);
1569 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1570 0, cmd, NULL, notify);
1576 * Following is where TCP initial sequence number generation occurs.
1578 * There are two places where we must use initial sequence numbers:
1579 * 1. In SYN-ACK packets.
1580 * 2. In SYN packets.
1582 * All ISNs for SYN-ACK packets are generated by the syncache. See
1583 * tcp_syncache.c for details.
1585 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1586 * depends on this property. In addition, these ISNs should be
1587 * unguessable so as to prevent connection hijacking. To satisfy
1588 * the requirements of this situation, the algorithm outlined in
1589 * RFC 1948 is used, with only small modifications.
1591 * Implementation details:
1593 * Time is based off the system timer, and is corrected so that it
1594 * increases by one megabyte per second. This allows for proper
1595 * recycling on high speed LANs while still leaving over an hour
1598 * As reading the *exact* system time is too expensive to be done
1599 * whenever setting up a TCP connection, we increment the time
1600 * offset in two ways. First, a small random positive increment
1601 * is added to isn_offset for each connection that is set up.
1602 * Second, the function tcp_isn_tick fires once per clock tick
1603 * and increments isn_offset as necessary so that sequence numbers
1604 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1605 * random positive increments serve only to ensure that the same
1606 * exact sequence number is never sent out twice (as could otherwise
1607 * happen when a port is recycled in less than the system tick
1610 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1611 * between seeding of isn_secret. This is normally set to zero,
1612 * as reseeding should not be necessary.
1614 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1615 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1616 * general, this means holding an exclusive (write) lock.
1619 #define ISN_BYTES_PER_SECOND 1048576
1620 #define ISN_STATIC_INCREMENT 4096
1621 #define ISN_RANDOM_INCREMENT (4096 - 1)
1623 static VNET_DEFINE(u_char, isn_secret[32]);
1624 static VNET_DEFINE(int, isn_last);
1625 static VNET_DEFINE(int, isn_last_reseed);
1626 static VNET_DEFINE(u_int32_t, isn_offset);
1627 static VNET_DEFINE(u_int32_t, isn_offset_old);
1629 #define V_isn_secret VNET(isn_secret)
1630 #define V_isn_last VNET(isn_last)
1631 #define V_isn_last_reseed VNET(isn_last_reseed)
1632 #define V_isn_offset VNET(isn_offset)
1633 #define V_isn_offset_old VNET(isn_offset_old)
1636 tcp_new_isn(struct tcpcb *tp)
1639 u_int32_t md5_buffer[4];
1641 u_int32_t projected_offset;
1643 INP_WLOCK_ASSERT(tp->t_inpcb);
1646 /* Seed if this is the first use, reseed if requested. */
1647 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1648 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1650 read_random(&V_isn_secret, sizeof(V_isn_secret));
1651 V_isn_last_reseed = ticks;
1654 /* Compute the md5 hash and return the ISN. */
1656 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1657 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1659 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1660 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1661 sizeof(struct in6_addr));
1662 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1663 sizeof(struct in6_addr));
1667 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1668 sizeof(struct in_addr));
1669 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1670 sizeof(struct in_addr));
1672 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1673 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1674 new_isn = (tcp_seq) md5_buffer[0];
1675 V_isn_offset += ISN_STATIC_INCREMENT +
1676 (arc4random() & ISN_RANDOM_INCREMENT);
1677 if (ticks != V_isn_last) {
1678 projected_offset = V_isn_offset_old +
1679 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1680 if (SEQ_GT(projected_offset, V_isn_offset))
1681 V_isn_offset = projected_offset;
1682 V_isn_offset_old = V_isn_offset;
1685 new_isn += V_isn_offset;
1691 * When a specific ICMP unreachable message is received and the
1692 * connection state is SYN-SENT, drop the connection. This behavior
1693 * is controlled by the icmp_may_rst sysctl.
1696 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1700 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1701 INP_WLOCK_ASSERT(inp);
1703 if ((inp->inp_flags & INP_TIMEWAIT) ||
1704 (inp->inp_flags & INP_DROPPED))
1707 tp = intotcpcb(inp);
1708 if (tp->t_state != TCPS_SYN_SENT)
1711 tp = tcp_drop(tp, errno);
1719 * When `need fragmentation' ICMP is received, update our idea of the MSS
1720 * based on the new value. Also nudge TCP to send something, since we
1721 * know the packet we just sent was dropped.
1722 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1724 static struct inpcb *
1725 tcp_mtudisc_notify(struct inpcb *inp, int error)
1728 return (tcp_mtudisc(inp, -1));
1732 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1737 INP_WLOCK_ASSERT(inp);
1738 if ((inp->inp_flags & INP_TIMEWAIT) ||
1739 (inp->inp_flags & INP_DROPPED))
1742 tp = intotcpcb(inp);
1743 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1745 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1747 so = inp->inp_socket;
1748 SOCKBUF_LOCK(&so->so_snd);
1749 /* If the mss is larger than the socket buffer, decrease the mss. */
1750 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1751 tp->t_maxseg = so->so_snd.sb_hiwat;
1752 SOCKBUF_UNLOCK(&so->so_snd);
1754 TCPSTAT_INC(tcps_mturesent);
1756 tp->snd_nxt = tp->snd_una;
1757 tcp_free_sackholes(tp);
1758 tp->snd_recover = tp->snd_max;
1759 if (tp->t_flags & TF_SACK_PERMIT)
1760 EXIT_FASTRECOVERY(tp->t_flags);
1767 * Look-up the routing entry to the peer of this inpcb. If no route
1768 * is found and it cannot be allocated, then return 0. This routine
1769 * is called by TCP routines that access the rmx structure and by
1770 * tcp_mss_update to get the peer/interface MTU.
1773 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1776 struct sockaddr_in *dst;
1780 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1782 bzero(&sro, sizeof(sro));
1783 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1784 dst = (struct sockaddr_in *)&sro.ro_dst;
1785 dst->sin_family = AF_INET;
1786 dst->sin_len = sizeof(*dst);
1787 dst->sin_addr = inc->inc_faddr;
1788 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1790 if (sro.ro_rt != NULL) {
1791 ifp = sro.ro_rt->rt_ifp;
1792 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1793 maxmtu = ifp->if_mtu;
1795 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1797 /* Report additional interface capabilities. */
1798 if (flags != NULL) {
1799 if (ifp->if_capenable & IFCAP_TSO4 &&
1800 ifp->if_hwassist & CSUM_TSO)
1811 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1813 struct route_in6 sro6;
1817 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1819 bzero(&sro6, sizeof(sro6));
1820 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1821 sro6.ro_dst.sin6_family = AF_INET6;
1822 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1823 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1824 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1826 if (sro6.ro_rt != NULL) {
1827 ifp = sro6.ro_rt->rt_ifp;
1828 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1829 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1831 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1832 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1834 /* Report additional interface capabilities. */
1835 if (flags != NULL) {
1836 if (ifp->if_capenable & IFCAP_TSO6 &&
1837 ifp->if_hwassist & CSUM_TSO)
1848 /* compute ESP/AH header size for TCP, including outer IP header. */
1850 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1857 struct ip6_hdr *ip6;
1861 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1863 MGETHDR(m, M_NOWAIT, MT_DATA);
1868 if ((inp->inp_vflag & INP_IPV6) != 0) {
1869 ip6 = mtod(m, struct ip6_hdr *);
1870 th = (struct tcphdr *)(ip6 + 1);
1871 m->m_pkthdr.len = m->m_len =
1872 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1873 tcpip_fillheaders(inp, ip6, th);
1874 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1878 ip = mtod(m, struct ip *);
1879 th = (struct tcphdr *)(ip + 1);
1880 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1881 tcpip_fillheaders(inp, ip, th);
1882 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1890 #ifdef TCP_SIGNATURE
1892 * Callback function invoked by m_apply() to digest TCP segment data
1893 * contained within an mbuf chain.
1896 tcp_signature_apply(void *fstate, void *data, u_int len)
1899 MD5Update(fstate, (u_char *)data, len);
1904 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1907 * m pointer to head of mbuf chain
1909 * len length of TCP segment data, excluding options
1910 * optlen length of TCP segment options
1911 * buf pointer to storage for computed MD5 digest
1912 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1914 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1915 * When called from tcp_input(), we can be sure that th_sum has been
1916 * zeroed out and verified already.
1918 * Return 0 if successful, otherwise return -1.
1920 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1921 * search with the destination IP address, and a 'magic SPI' to be
1922 * determined by the application. This is hardcoded elsewhere to 1179
1923 * right now. Another branch of this code exists which uses the SPD to
1924 * specify per-application flows but it is unstable.
1927 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1928 u_char *buf, u_int direction)
1930 union sockaddr_union dst;
1932 struct ippseudo ippseudo;
1938 struct ipovly *ipovly;
1940 struct secasvar *sav;
1943 struct ip6_hdr *ip6;
1944 struct in6_addr in6;
1945 char ip6buf[INET6_ADDRSTRLEN];
1951 KASSERT(m != NULL, ("NULL mbuf chain"));
1952 KASSERT(buf != NULL, ("NULL signature pointer"));
1954 /* Extract the destination from the IP header in the mbuf. */
1955 bzero(&dst, sizeof(union sockaddr_union));
1956 ip = mtod(m, struct ip *);
1958 ip6 = NULL; /* Make the compiler happy. */
1963 dst.sa.sa_len = sizeof(struct sockaddr_in);
1964 dst.sa.sa_family = AF_INET;
1965 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1966 ip->ip_src : ip->ip_dst;
1970 case (IPV6_VERSION >> 4):
1971 ip6 = mtod(m, struct ip6_hdr *);
1972 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1973 dst.sa.sa_family = AF_INET6;
1974 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1975 ip6->ip6_src : ip6->ip6_dst;
1984 /* Look up an SADB entry which matches the address of the peer. */
1985 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1987 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1988 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1990 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1991 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1999 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2001 * XXX The ippseudo header MUST be digested in network byte order,
2002 * or else we'll fail the regression test. Assume all fields we've
2003 * been doing arithmetic on have been in host byte order.
2004 * XXX One cannot depend on ipovly->ih_len here. When called from
2005 * tcp_output(), the underlying ip_len member has not yet been set.
2010 ipovly = (struct ipovly *)ip;
2011 ippseudo.ippseudo_src = ipovly->ih_src;
2012 ippseudo.ippseudo_dst = ipovly->ih_dst;
2013 ippseudo.ippseudo_pad = 0;
2014 ippseudo.ippseudo_p = IPPROTO_TCP;
2015 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2017 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2019 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2020 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2025 * RFC 2385, 2.0 Proposal
2026 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2027 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2028 * extended next header value (to form 32 bits), and 32-bit segment
2030 * Note: Upper-Layer Packet Length comes before Next Header.
2032 case (IPV6_VERSION >> 4):
2034 in6_clearscope(&in6);
2035 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2037 in6_clearscope(&in6);
2038 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2039 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2040 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2042 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2043 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2044 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2046 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2048 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2049 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2060 * Step 2: Update MD5 hash with TCP header, excluding options.
2061 * The TCP checksum must be set to zero.
2063 savecsum = th->th_sum;
2065 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2066 th->th_sum = savecsum;
2069 * Step 3: Update MD5 hash with TCP segment data.
2070 * Use m_apply() to avoid an early m_pullup().
2073 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2076 * Step 4: Update MD5 hash with shared secret.
2078 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2079 MD5Final(buf, &ctx);
2081 key_sa_recordxfer(sav, m);
2087 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2090 * m pointer to head of mbuf chain
2091 * len length of TCP segment data, excluding options
2092 * optlen length of TCP segment options
2093 * buf pointer to storage for computed MD5 digest
2094 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2096 * Return 1 if successful, otherwise return 0.
2099 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2100 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2102 char tmpdigest[TCP_SIGLEN];
2104 if (tcp_sig_checksigs == 0)
2106 if ((tcpbflag & TF_SIGNATURE) == 0) {
2107 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2110 * If this socket is not expecting signature but
2111 * the segment contains signature just fail.
2113 TCPSTAT_INC(tcps_sig_err_sigopt);
2114 TCPSTAT_INC(tcps_sig_rcvbadsig);
2118 /* Signature is not expected, and not present in segment. */
2123 * If this socket is expecting signature but the segment does not
2124 * contain any just fail.
2126 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2127 TCPSTAT_INC(tcps_sig_err_nosigopt);
2128 TCPSTAT_INC(tcps_sig_rcvbadsig);
2131 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2132 IPSEC_DIR_INBOUND) == -1) {
2133 TCPSTAT_INC(tcps_sig_err_buildsig);
2134 TCPSTAT_INC(tcps_sig_rcvbadsig);
2138 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2139 TCPSTAT_INC(tcps_sig_rcvbadsig);
2142 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2145 #endif /* TCP_SIGNATURE */
2148 sysctl_drop(SYSCTL_HANDLER_ARGS)
2150 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2151 struct sockaddr_storage addrs[2];
2155 struct sockaddr_in *fin, *lin;
2157 struct sockaddr_in6 *fin6, *lin6;
2168 if (req->oldptr != NULL || req->oldlen != 0)
2170 if (req->newptr == NULL)
2172 if (req->newlen < sizeof(addrs))
2174 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2178 switch (addrs[0].ss_family) {
2181 fin6 = (struct sockaddr_in6 *)&addrs[0];
2182 lin6 = (struct sockaddr_in6 *)&addrs[1];
2183 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2184 lin6->sin6_len != sizeof(struct sockaddr_in6))
2186 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2187 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2189 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2190 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2191 fin = (struct sockaddr_in *)&addrs[0];
2192 lin = (struct sockaddr_in *)&addrs[1];
2195 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2198 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2205 fin = (struct sockaddr_in *)&addrs[0];
2206 lin = (struct sockaddr_in *)&addrs[1];
2207 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2208 lin->sin_len != sizeof(struct sockaddr_in))
2215 INP_INFO_WLOCK(&V_tcbinfo);
2216 switch (addrs[0].ss_family) {
2219 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2220 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2221 INPLOOKUP_WLOCKPCB, NULL);
2226 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2227 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2232 if (inp->inp_flags & INP_TIMEWAIT) {
2234 * XXXRW: There currently exists a state where an
2235 * inpcb is present, but its timewait state has been
2236 * discarded. For now, don't allow dropping of this
2244 } else if (!(inp->inp_flags & INP_DROPPED) &&
2245 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2246 tp = intotcpcb(inp);
2247 tp = tcp_drop(tp, ECONNABORTED);
2254 INP_INFO_WUNLOCK(&V_tcbinfo);
2258 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2259 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2260 0, sysctl_drop, "", "Drop TCP connection");
2263 * Generate a standardized TCP log line for use throughout the
2264 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2265 * allow use in the interrupt context.
2267 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2268 * NB: The function may return NULL if memory allocation failed.
2270 * Due to header inclusion and ordering limitations the struct ip
2271 * and ip6_hdr pointers have to be passed as void pointers.
2274 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2278 /* Is logging enabled? */
2279 if (tcp_log_in_vain == 0)
2282 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2286 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2290 /* Is logging enabled? */
2291 if (tcp_log_debug == 0)
2294 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2298 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2305 const struct ip6_hdr *ip6;
2307 ip6 = (const struct ip6_hdr *)ip6hdr;
2309 ip = (struct ip *)ip4hdr;
2312 * The log line looks like this:
2313 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2315 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2316 sizeof(PRINT_TH_FLAGS) + 1 +
2318 2 * INET6_ADDRSTRLEN;
2320 2 * INET_ADDRSTRLEN;
2323 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2327 strcat(s, "TCP: [");
2330 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2331 inet_ntoa_r(inc->inc_faddr, sp);
2333 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2335 inet_ntoa_r(inc->inc_laddr, sp);
2337 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2340 ip6_sprintf(sp, &inc->inc6_faddr);
2342 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2344 ip6_sprintf(sp, &inc->inc6_laddr);
2346 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2347 } else if (ip6 && th) {
2348 ip6_sprintf(sp, &ip6->ip6_src);
2350 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2352 ip6_sprintf(sp, &ip6->ip6_dst);
2354 sprintf(sp, "]:%i", ntohs(th->th_dport));
2357 } else if (ip && th) {
2358 inet_ntoa_r(ip->ip_src, sp);
2360 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2362 inet_ntoa_r(ip->ip_dst, sp);
2364 sprintf(sp, "]:%i", ntohs(th->th_dport));
2372 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2373 if (*(s + size - 1) != '\0')
2374 panic("%s: string too long", __func__);