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>
86 #include <netinet/tcp_fsm.h>
87 #include <netinet/tcp_seq.h>
88 #include <netinet/tcp_timer.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/tcp_syncache.h>
92 #include <netinet6/tcp6_var.h>
94 #include <netinet/tcpip.h>
96 #include <netinet/tcp_debug.h>
99 #include <netinet6/ip6protosw.h>
102 #include <netinet/tcp_offload.h>
106 #include <netipsec/ipsec.h>
107 #include <netipsec/xform.h>
109 #include <netipsec/ipsec6.h>
111 #include <netipsec/key.h>
112 #include <sys/syslog.h>
115 #include <machine/in_cksum.h>
118 #include <security/mac/mac_framework.h>
120 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
122 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
126 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
131 error = sysctl_handle_int(oidp, &new, 0, req);
132 if (error == 0 && req->newptr) {
133 if (new < TCP_MINMSS)
141 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
142 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
143 &sysctl_net_inet_tcp_mss_check, "I",
144 "Default TCP Maximum Segment Size");
148 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
152 new = V_tcp_v6mssdflt;
153 error = sysctl_handle_int(oidp, &new, 0, req);
154 if (error == 0 && req->newptr) {
155 if (new < TCP_MINMSS)
158 V_tcp_v6mssdflt = new;
163 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
164 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
165 &sysctl_net_inet_tcp_mss_v6_check, "I",
166 "Default TCP Maximum Segment Size for IPv6");
170 * Minimum MSS we accept and use. This prevents DoS attacks where
171 * we are forced to a ridiculous low MSS like 20 and send hundreds
172 * of packets instead of one. The effect scales with the available
173 * bandwidth and quickly saturates the CPU and network interface
174 * with packet generation and sending. Set to zero to disable MINMSS
175 * checking. This setting prevents us from sending too small packets.
177 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
179 &VNET_NAME(tcp_minmss), 0,
180 "Minimum TCP Maximum Segment Size");
182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
184 &VNET_NAME(tcp_do_rfc1323), 0,
185 "Enable rfc1323 (high performance TCP) extensions");
187 static int tcp_log_debug = 0;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
189 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
191 static int tcp_tcbhashsize = 0;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
193 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
195 static int do_tcpdrain = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
197 "Enable tcp_drain routine for extra help when low on mbufs");
199 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
200 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
202 static VNET_DEFINE(int, icmp_may_rst) = 1;
203 #define V_icmp_may_rst VNET(icmp_may_rst)
204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
205 &VNET_NAME(icmp_may_rst), 0,
206 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
208 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
209 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
211 &VNET_NAME(tcp_isn_reseed_interval), 0,
212 "Seconds between reseeding of ISN secret");
214 static int tcp_soreceive_stream = 0;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
216 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
219 static int tcp_sig_checksigs = 1;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
221 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define V_sack_hole_zone VNET(sack_hole_zone)
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
229 static struct inpcb *tcp_notify(struct inpcb *, int);
230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
232 void *ip4hdr, const void *ip6hdr);
233 static void tcp_timer_discard(struct tcpcb *, uint32_t);
236 * Target size of TCP PCB hash tables. Must be a power of two.
238 * Note that this can be overridden by the kernel environment
239 * variable net.inet.tcp.tcbhashsize
242 #define TCBHASHSIZE 0
247 * Callouts should be moved into struct tcp directly. They are currently
248 * separate because the tcpcb structure is exported to userland for sysctl
249 * parsing purposes, which do not know about callouts.
258 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
259 #define V_tcpcb_zone VNET(tcpcb_zone)
261 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
262 static struct mtx isn_mtx;
264 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
265 #define ISN_LOCK() mtx_lock(&isn_mtx)
266 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
269 * TCP initialization.
272 tcp_zone_change(void *tag)
275 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
276 uma_zone_set_max(V_tcpcb_zone, maxsockets);
277 tcp_tw_zone_change();
281 tcp_inpcb_init(void *mem, int size, int flags)
283 struct inpcb *inp = mem;
285 INP_LOCK_INIT(inp, "inp", "tcpinp");
290 * Take a value and get the next power of 2 that doesn't overflow.
291 * Used to size the tcp_inpcb hash buckets.
294 maketcp_hashsize(int size)
300 * get the next power of 2 higher than maxsockets.
302 hashsize = 1 << fls(size);
303 /* catch overflow, and just go one power of 2 smaller */
304 if (hashsize < size) {
305 hashsize = 1 << (fls(size) - 1);
313 const char *tcbhash_tuneable;
316 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
318 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
319 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
320 printf("%s: WARNING: unable to register helper hook\n", __func__);
321 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
322 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 printf("%s: WARNING: unable to register helper hook\n", __func__);
325 hashsize = TCBHASHSIZE;
326 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
329 * Auto tune the hash size based on maxsockets.
330 * A perfect hash would have a 1:1 mapping
331 * (hashsize = maxsockets) however it's been
332 * suggested that O(2) average is better.
334 hashsize = maketcp_hashsize(maxsockets / 4);
336 * Our historical default is 512,
337 * do not autotune lower than this.
342 printf("%s: %s auto tuned to %d\n", __func__,
343 tcbhash_tuneable, hashsize);
346 * We require a hashsize to be a power of two.
347 * Previously if it was not a power of two we would just reset it
348 * back to 512, which could be a nasty surprise if you did not notice
350 * Instead what we do is clip it to the closest power of two lower
351 * than the specified hash value.
353 if (!powerof2(hashsize)) {
354 int oldhashsize = hashsize;
356 hashsize = maketcp_hashsize(hashsize);
357 /* prevent absurdly low value */
360 printf("%s: WARNING: TCB hash size not a power of 2, "
361 "clipped from %d to %d.\n", __func__, oldhashsize,
364 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
365 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
366 IPI_HASHFIELDS_4TUPLE);
369 * These have to be type stable for the benefit of the timers.
371 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
372 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
373 uma_zone_set_max(V_tcpcb_zone, maxsockets);
374 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
380 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
381 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
382 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384 /* Skip initialization of globals for non-default instances. */
385 if (!IS_DEFAULT_VNET(curvnet))
388 tcp_reass_global_init();
390 /* XXX virtualize those bellow? */
391 tcp_delacktime = TCPTV_DELACK;
392 tcp_keepinit = TCPTV_KEEP_INIT;
393 tcp_keepidle = TCPTV_KEEP_IDLE;
394 tcp_keepintvl = TCPTV_KEEPINTVL;
395 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
397 tcp_rexmit_min = TCPTV_MIN;
398 if (tcp_rexmit_min < 1)
400 tcp_rexmit_slop = TCPTV_CPU_VAR;
401 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
402 tcp_tcbhashsize = hashsize;
404 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
405 if (tcp_soreceive_stream) {
407 tcp_usrreqs.pru_soreceive = soreceive_stream;
410 tcp6_usrreqs.pru_soreceive = soreceive_stream;
415 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
417 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
419 if (max_protohdr < TCP_MINPROTOHDR)
420 max_protohdr = TCP_MINPROTOHDR;
421 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
423 #undef TCP_MINPROTOHDR
426 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
427 SHUTDOWN_PRI_DEFAULT);
428 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
429 EVENTHANDLER_PRI_ANY);
440 in_pcbinfo_destroy(&V_tcbinfo);
441 uma_zdestroy(V_sack_hole_zone);
442 uma_zdestroy(V_tcpcb_zone);
453 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
454 * tcp_template used to store this data in mbufs, but we now recopy it out
455 * of the tcpcb each time to conserve mbufs.
458 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
460 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
462 INP_WLOCK_ASSERT(inp);
465 if ((inp->inp_vflag & INP_IPV6) != 0) {
468 ip6 = (struct ip6_hdr *)ip_ptr;
469 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
470 (inp->inp_flow & IPV6_FLOWINFO_MASK);
471 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
472 (IPV6_VERSION & IPV6_VERSION_MASK);
473 ip6->ip6_nxt = IPPROTO_TCP;
474 ip6->ip6_plen = htons(sizeof(struct tcphdr));
475 ip6->ip6_src = inp->in6p_laddr;
476 ip6->ip6_dst = inp->in6p_faddr;
479 #if defined(INET6) && defined(INET)
486 ip = (struct ip *)ip_ptr;
487 ip->ip_v = IPVERSION;
489 ip->ip_tos = inp->inp_ip_tos;
493 ip->ip_ttl = inp->inp_ip_ttl;
495 ip->ip_p = IPPROTO_TCP;
496 ip->ip_src = inp->inp_laddr;
497 ip->ip_dst = inp->inp_faddr;
500 th->th_sport = inp->inp_lport;
501 th->th_dport = inp->inp_fport;
509 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
513 * Create template to be used to send tcp packets on a connection.
514 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
515 * use for this function is in keepalives, which use tcp_respond.
518 tcpip_maketemplate(struct inpcb *inp)
522 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
525 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
530 * Send a single message to the TCP at address specified by
531 * the given TCP/IP header. If m == NULL, then we make a copy
532 * of the tcpiphdr at ti and send directly to the addressed host.
533 * This is used to force keep alive messages out using the TCP
534 * template for a connection. If flags are given then we send
535 * a message back to the TCP which originated the * segment ti,
536 * and discard the mbuf containing it and any other attached mbufs.
538 * In any case the ack and sequence number of the transmitted
539 * segment are as specified by the parameters.
541 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
544 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
545 tcp_seq ack, tcp_seq seq, int flags)
558 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
561 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
568 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
569 INP_WLOCK_ASSERT(inp);
574 if (!(flags & TH_RST)) {
575 win = sbspace(&inp->inp_socket->so_rcv);
576 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
577 win = (long)TCP_MAXWIN << tp->rcv_scale;
581 m = m_gethdr(M_NOWAIT, MT_DATA);
585 m->m_data += max_linkhdr;
588 bcopy((caddr_t)ip6, mtod(m, caddr_t),
589 sizeof(struct ip6_hdr));
590 ip6 = mtod(m, struct ip6_hdr *);
591 nth = (struct tcphdr *)(ip6 + 1);
595 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
596 ip = mtod(m, struct ip *);
597 nth = (struct tcphdr *)(ip + 1);
599 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
604 * XXX MRT We inherrit the FIB, which is lucky.
608 m->m_data = (caddr_t)ipgen;
609 /* m_len is set later */
611 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
614 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
615 nth = (struct tcphdr *)(ip6 + 1);
619 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
620 nth = (struct tcphdr *)(ip + 1);
624 * this is usually a case when an extension header
625 * exists between the IPv6 header and the
628 nth->th_sport = th->th_sport;
629 nth->th_dport = th->th_dport;
631 xchg(nth->th_dport, nth->th_sport, uint16_t);
637 ip6->ip6_vfc = IPV6_VERSION;
638 ip6->ip6_nxt = IPPROTO_TCP;
639 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
640 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
643 #if defined(INET) && defined(INET6)
648 tlen += sizeof (struct tcpiphdr);
649 ip->ip_len = htons(tlen);
650 ip->ip_ttl = V_ip_defttl;
651 if (V_path_mtu_discovery)
652 ip->ip_off |= htons(IP_DF);
656 m->m_pkthdr.len = tlen;
657 m->m_pkthdr.rcvif = NULL;
661 * Packet is associated with a socket, so allow the
662 * label of the response to reflect the socket label.
664 INP_WLOCK_ASSERT(inp);
665 mac_inpcb_create_mbuf(inp, m);
668 * Packet is not associated with a socket, so possibly
669 * update the label in place.
671 mac_netinet_tcp_reply(m);
674 nth->th_seq = htonl(seq);
675 nth->th_ack = htonl(ack);
677 nth->th_off = sizeof (struct tcphdr) >> 2;
678 nth->th_flags = flags;
680 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
682 nth->th_win = htons((u_short)win);
685 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
688 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
689 nth->th_sum = in6_cksum_pseudo(ip6,
690 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
691 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
695 #if defined(INET6) && defined(INET)
700 m->m_pkthdr.csum_flags = CSUM_TCP;
701 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
702 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
706 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
707 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
710 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
713 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
716 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
718 #if defined(INET) && defined(INET6)
722 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
727 * Create a new TCP control block, making an
728 * empty reassembly queue and hooking it to the argument
729 * protocol control block. The `inp' parameter must have
730 * come from the zone allocator set up in tcp_init().
733 tcp_newtcpcb(struct inpcb *inp)
735 struct tcpcb_mem *tm;
738 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
741 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
746 /* Initialise cc_var struct for this tcpcb. */
748 tp->ccv->type = IPPROTO_TCP;
749 tp->ccv->ccvc.tcp = tp;
752 * Use the current system default CC algorithm.
755 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
756 CC_ALGO(tp) = CC_DEFAULT();
759 if (CC_ALGO(tp)->cb_init != NULL)
760 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
761 uma_zfree(V_tcpcb_zone, tm);
766 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
767 uma_zfree(V_tcpcb_zone, tm);
772 tp->t_vnet = inp->inp_vnet;
774 tp->t_timers = &tm->tt;
775 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
776 tp->t_maxseg = tp->t_maxopd =
778 isipv6 ? V_tcp_v6mssdflt :
782 /* Set up our timeouts. */
783 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
784 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
785 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
786 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
787 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
789 if (V_tcp_do_rfc1323)
790 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
792 tp->t_flags |= TF_SACK_PERMIT;
793 TAILQ_INIT(&tp->snd_holes);
795 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
798 in_pcbref(inp); /* Reference for tcpcb */
802 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
803 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
804 * reasonable initial retransmit time.
806 tp->t_srtt = TCPTV_SRTTBASE;
807 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
808 tp->t_rttmin = tcp_rexmit_min;
809 tp->t_rxtcur = TCPTV_RTOBASE;
810 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
811 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
812 tp->t_rcvtime = ticks;
814 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
815 * because the socket may be bound to an IPv6 wildcard address,
816 * which may match an IPv4-mapped IPv6 address.
818 inp->inp_ip_ttl = V_ip_defttl;
820 return (tp); /* XXX */
824 * Switch the congestion control algorithm back to NewReno for any active
825 * control blocks using an algorithm which is about to go away.
826 * This ensures the CC framework can allow the unload to proceed without leaving
827 * any dangling pointers which would trigger a panic.
828 * Returning non-zero would inform the CC framework that something went wrong
829 * and it would be unsafe to allow the unload to proceed. However, there is no
830 * way for this to occur with this implementation so we always return zero.
833 tcp_ccalgounload(struct cc_algo *unload_algo)
835 struct cc_algo *tmpalgo;
838 VNET_ITERATOR_DECL(vnet_iter);
841 * Check all active control blocks across all network stacks and change
842 * any that are using "unload_algo" back to NewReno. If "unload_algo"
843 * requires cleanup code to be run, call it.
846 VNET_FOREACH(vnet_iter) {
847 CURVNET_SET(vnet_iter);
848 INP_INFO_RLOCK(&V_tcbinfo);
850 * New connections already part way through being initialised
851 * with the CC algo we're removing will not race with this code
852 * because the INP_INFO_WLOCK is held during initialisation. We
853 * therefore don't enter the loop below until the connection
854 * list has stabilised.
856 LIST_FOREACH(inp, &V_tcb, inp_list) {
858 /* Important to skip tcptw structs. */
859 if (!(inp->inp_flags & INP_TIMEWAIT) &&
860 (tp = intotcpcb(inp)) != NULL) {
862 * By holding INP_WLOCK here, we are assured
863 * that the connection is not currently
864 * executing inside the CC module's functions
865 * i.e. it is safe to make the switch back to
868 if (CC_ALGO(tp) == unload_algo) {
869 tmpalgo = CC_ALGO(tp);
870 /* NewReno does not require any init. */
871 CC_ALGO(tp) = &newreno_cc_algo;
872 if (tmpalgo->cb_destroy != NULL)
873 tmpalgo->cb_destroy(tp->ccv);
878 INP_INFO_RUNLOCK(&V_tcbinfo);
887 * Drop a TCP connection, reporting
888 * the specified error. If connection is synchronized,
889 * then send a RST to peer.
892 tcp_drop(struct tcpcb *tp, int errno)
894 struct socket *so = tp->t_inpcb->inp_socket;
896 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
897 INP_WLOCK_ASSERT(tp->t_inpcb);
899 if (TCPS_HAVERCVDSYN(tp->t_state)) {
900 tcp_state_change(tp, TCPS_CLOSED);
901 (void) tcp_output(tp);
902 TCPSTAT_INC(tcps_drops);
904 TCPSTAT_INC(tcps_conndrops);
905 if (errno == ETIMEDOUT && tp->t_softerror)
906 errno = tp->t_softerror;
907 so->so_error = errno;
908 return (tcp_close(tp));
912 tcp_discardcb(struct tcpcb *tp)
914 struct inpcb *inp = tp->t_inpcb;
915 struct socket *so = inp->inp_socket;
917 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
921 INP_WLOCK_ASSERT(inp);
924 * Make sure that all of our timers are stopped before we delete the
927 * If stopping a timer fails, we schedule a discard function in same
928 * callout, and the last discard function called will take care of
929 * deleting the tcpcb.
931 tcp_timer_stop(tp, TT_REXMT);
932 tcp_timer_stop(tp, TT_PERSIST);
933 tcp_timer_stop(tp, TT_KEEP);
934 tcp_timer_stop(tp, TT_2MSL);
935 tcp_timer_stop(tp, TT_DELACK);
938 * If we got enough samples through the srtt filter,
939 * save the rtt and rttvar in the routing entry.
940 * 'Enough' is arbitrarily defined as 4 rtt samples.
941 * 4 samples is enough for the srtt filter to converge
942 * to within enough % of the correct value; fewer samples
943 * and we could save a bogus rtt. The danger is not high
944 * as tcp quickly recovers from everything.
945 * XXX: Works very well but needs some more statistics!
947 if (tp->t_rttupdated >= 4) {
948 struct hc_metrics_lite metrics;
951 bzero(&metrics, sizeof(metrics));
953 * Update the ssthresh always when the conditions below
954 * are satisfied. This gives us better new start value
955 * for the congestion avoidance for new connections.
956 * ssthresh is only set if packet loss occured on a session.
958 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
959 * being torn down. Ideally this code would not use 'so'.
961 ssthresh = tp->snd_ssthresh;
962 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
964 * convert the limit from user data bytes to
965 * packets then to packet data bytes.
967 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
970 ssthresh *= (u_long)(tp->t_maxseg +
972 (isipv6 ? sizeof (struct ip6_hdr) +
973 sizeof (struct tcphdr) :
975 sizeof (struct tcpiphdr)
982 metrics.rmx_ssthresh = ssthresh;
984 metrics.rmx_rtt = tp->t_srtt;
985 metrics.rmx_rttvar = tp->t_rttvar;
986 metrics.rmx_cwnd = tp->snd_cwnd;
987 metrics.rmx_sendpipe = 0;
988 metrics.rmx_recvpipe = 0;
990 tcp_hc_update(&inp->inp_inc, &metrics);
993 /* free the reassembly queue, if any */
997 /* Disconnect offload device, if any. */
998 if (tp->t_flags & TF_TOE)
999 tcp_offload_detach(tp);
1002 tcp_free_sackholes(tp);
1004 /* Allow the CC algorithm to clean up after itself. */
1005 if (CC_ALGO(tp)->cb_destroy != NULL)
1006 CC_ALGO(tp)->cb_destroy(tp->ccv);
1008 khelp_destroy_osd(tp->osd);
1011 inp->inp_ppcb = NULL;
1012 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1013 /* We own the last reference on tcpcb, let's free it. */
1015 uma_zfree(V_tcpcb_zone, tp);
1016 released = in_pcbrele_wlocked(inp);
1017 KASSERT(!released, ("%s: inp %p should not have been released "
1018 "here", __func__, inp));
1023 tcp_timer_2msl_discard(void *xtp)
1026 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1030 tcp_timer_keep_discard(void *xtp)
1033 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1037 tcp_timer_persist_discard(void *xtp)
1040 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1044 tcp_timer_rexmt_discard(void *xtp)
1047 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1051 tcp_timer_delack_discard(void *xtp)
1054 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1058 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1062 CURVNET_SET(tp->t_vnet);
1063 INP_INFO_WLOCK(&V_tcbinfo);
1065 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1068 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1069 ("%s: tcpcb has to be stopped here", __func__));
1070 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1071 ("%s: discard callout should be running", __func__));
1072 tp->t_timers->tt_flags &= ~timer_type;
1073 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1074 /* We own the last reference on this tcpcb, let's free it. */
1076 uma_zfree(V_tcpcb_zone, tp);
1077 if (in_pcbrele_wlocked(inp)) {
1078 INP_INFO_WUNLOCK(&V_tcbinfo);
1084 INP_INFO_WUNLOCK(&V_tcbinfo);
1089 * Attempt to close a TCP control block, marking it as dropped, and freeing
1090 * the socket if we hold the only reference.
1093 tcp_close(struct tcpcb *tp)
1095 struct inpcb *inp = tp->t_inpcb;
1098 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1099 INP_WLOCK_ASSERT(inp);
1102 if (tp->t_state == TCPS_LISTEN)
1103 tcp_offload_listen_stop(tp);
1106 TCPSTAT_INC(tcps_closed);
1107 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1108 so = inp->inp_socket;
1109 soisdisconnected(so);
1110 if (inp->inp_flags & INP_SOCKREF) {
1111 KASSERT(so->so_state & SS_PROTOREF,
1112 ("tcp_close: !SS_PROTOREF"));
1113 inp->inp_flags &= ~INP_SOCKREF;
1117 so->so_state &= ~SS_PROTOREF;
1127 VNET_ITERATOR_DECL(vnet_iter);
1132 VNET_LIST_RLOCK_NOSLEEP();
1133 VNET_FOREACH(vnet_iter) {
1134 CURVNET_SET(vnet_iter);
1139 * Walk the tcpbs, if existing, and flush the reassembly queue,
1140 * if there is one...
1141 * XXX: The "Net/3" implementation doesn't imply that the TCP
1142 * reassembly queue should be flushed, but in a situation
1143 * where we're really low on mbufs, this is potentially
1146 INP_INFO_RLOCK(&V_tcbinfo);
1147 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1148 if (inpb->inp_flags & INP_TIMEWAIT)
1151 if ((tcpb = intotcpcb(inpb)) != NULL) {
1152 tcp_reass_flush(tcpb);
1153 tcp_clean_sackreport(tcpb);
1157 INP_INFO_RUNLOCK(&V_tcbinfo);
1160 VNET_LIST_RUNLOCK_NOSLEEP();
1164 * Notify a tcp user of an asynchronous error;
1165 * store error as soft error, but wake up user
1166 * (for now, won't do anything until can select for soft error).
1168 * Do not wake up user since there currently is no mechanism for
1169 * reporting soft errors (yet - a kqueue filter may be added).
1171 static struct inpcb *
1172 tcp_notify(struct inpcb *inp, int error)
1176 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1177 INP_WLOCK_ASSERT(inp);
1179 if ((inp->inp_flags & INP_TIMEWAIT) ||
1180 (inp->inp_flags & INP_DROPPED))
1183 tp = intotcpcb(inp);
1184 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1187 * Ignore some errors if we are hooked up.
1188 * If connection hasn't completed, has retransmitted several times,
1189 * and receives a second error, give up now. This is better
1190 * than waiting a long time to establish a connection that
1191 * can never complete.
1193 if (tp->t_state == TCPS_ESTABLISHED &&
1194 (error == EHOSTUNREACH || error == ENETUNREACH ||
1195 error == EHOSTDOWN)) {
1197 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1199 tp = tcp_drop(tp, error);
1205 tp->t_softerror = error;
1209 wakeup( &so->so_timeo);
1216 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1218 int error, i, m, n, pcb_count;
1219 struct inpcb *inp, **inp_list;
1224 * The process of preparing the TCB list is too time-consuming and
1225 * resource-intensive to repeat twice on every request.
1227 if (req->oldptr == NULL) {
1228 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1229 n += imax(n / 8, 10);
1230 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1234 if (req->newptr != NULL)
1238 * OK, now we're committed to doing something.
1240 INP_INFO_RLOCK(&V_tcbinfo);
1241 gencnt = V_tcbinfo.ipi_gencnt;
1242 n = V_tcbinfo.ipi_count;
1243 INP_INFO_RUNLOCK(&V_tcbinfo);
1245 m = syncache_pcbcount();
1247 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1248 + (n + m) * sizeof(struct xtcpcb));
1252 xig.xig_len = sizeof xig;
1253 xig.xig_count = n + m;
1254 xig.xig_gen = gencnt;
1255 xig.xig_sogen = so_gencnt;
1256 error = SYSCTL_OUT(req, &xig, sizeof xig);
1260 error = syncache_pcblist(req, m, &pcb_count);
1264 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1265 if (inp_list == NULL)
1268 INP_INFO_RLOCK(&V_tcbinfo);
1269 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1270 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1272 if (inp->inp_gencnt <= gencnt) {
1274 * XXX: This use of cr_cansee(), introduced with
1275 * TCP state changes, is not quite right, but for
1276 * now, better than nothing.
1278 if (inp->inp_flags & INP_TIMEWAIT) {
1279 if (intotw(inp) != NULL)
1280 error = cr_cansee(req->td->td_ucred,
1281 intotw(inp)->tw_cred);
1283 error = EINVAL; /* Skip this inp. */
1285 error = cr_canseeinpcb(req->td->td_ucred, inp);
1288 inp_list[i++] = inp;
1293 INP_INFO_RUNLOCK(&V_tcbinfo);
1297 for (i = 0; i < n; i++) {
1300 if (inp->inp_gencnt <= gencnt) {
1304 bzero(&xt, sizeof(xt));
1305 xt.xt_len = sizeof xt;
1306 /* XXX should avoid extra copy */
1307 bcopy(inp, &xt.xt_inp, sizeof *inp);
1308 inp_ppcb = inp->inp_ppcb;
1309 if (inp_ppcb == NULL)
1310 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1311 else if (inp->inp_flags & INP_TIMEWAIT) {
1312 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1313 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1315 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1316 if (xt.xt_tp.t_timers)
1317 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1319 if (inp->inp_socket != NULL)
1320 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1322 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1323 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1325 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1327 error = SYSCTL_OUT(req, &xt, sizeof xt);
1331 INP_INFO_WLOCK(&V_tcbinfo);
1332 for (i = 0; i < n; i++) {
1335 if (!in_pcbrele_rlocked(inp))
1338 INP_INFO_WUNLOCK(&V_tcbinfo);
1342 * Give the user an updated idea of our state.
1343 * If the generation differs from what we told
1344 * her before, she knows that something happened
1345 * while we were processing this request, and it
1346 * might be necessary to retry.
1348 INP_INFO_RLOCK(&V_tcbinfo);
1349 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1350 xig.xig_sogen = so_gencnt;
1351 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1352 INP_INFO_RUNLOCK(&V_tcbinfo);
1353 error = SYSCTL_OUT(req, &xig, sizeof xig);
1355 free(inp_list, M_TEMP);
1359 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1360 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1361 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1365 tcp_getcred(SYSCTL_HANDLER_ARGS)
1368 struct sockaddr_in addrs[2];
1372 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1375 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1378 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1379 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1381 if (inp->inp_socket == NULL)
1384 error = cr_canseeinpcb(req->td->td_ucred, inp);
1386 cru2x(inp->inp_cred, &xuc);
1391 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1395 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1396 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1397 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1402 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1405 struct sockaddr_in6 addrs[2];
1412 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1415 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1418 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1419 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1422 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1424 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1433 inp = in_pcblookup(&V_tcbinfo,
1434 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1436 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1437 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1440 inp = in6_pcblookup(&V_tcbinfo,
1441 &addrs[1].sin6_addr, addrs[1].sin6_port,
1442 &addrs[0].sin6_addr, addrs[0].sin6_port,
1443 INPLOOKUP_RLOCKPCB, NULL);
1445 if (inp->inp_socket == NULL)
1448 error = cr_canseeinpcb(req->td->td_ucred, inp);
1450 cru2x(inp->inp_cred, &xuc);
1455 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1459 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1460 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1461 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1467 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1469 struct ip *ip = vip;
1471 struct in_addr faddr;
1474 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1476 struct in_conninfo inc;
1477 tcp_seq icmp_tcp_seq;
1480 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1481 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1484 if (cmd == PRC_MSGSIZE)
1485 notify = tcp_mtudisc_notify;
1486 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1487 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1488 notify = tcp_drop_syn_sent;
1490 * Redirects don't need to be handled up here.
1492 else if (PRC_IS_REDIRECT(cmd))
1495 * Source quench is depreciated.
1497 else if (cmd == PRC_QUENCH)
1500 * Hostdead is ugly because it goes linearly through all PCBs.
1501 * XXX: We never get this from ICMP, otherwise it makes an
1502 * excellent DoS attack on machines with many connections.
1504 else if (cmd == PRC_HOSTDEAD)
1506 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1509 icp = (struct icmp *)((caddr_t)ip
1510 - offsetof(struct icmp, icmp_ip));
1511 th = (struct tcphdr *)((caddr_t)ip
1512 + (ip->ip_hl << 2));
1513 INP_INFO_WLOCK(&V_tcbinfo);
1514 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1515 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1517 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1518 !(inp->inp_flags & INP_DROPPED) &&
1519 !(inp->inp_socket == NULL)) {
1520 icmp_tcp_seq = htonl(th->th_seq);
1521 tp = intotcpcb(inp);
1522 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1523 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1524 if (cmd == PRC_MSGSIZE) {
1527 * If we got a needfrag set the MTU
1528 * in the route to the suggested new
1529 * value (if given) and then notify.
1531 bzero(&inc, sizeof(inc));
1532 inc.inc_faddr = faddr;
1534 inp->inp_inc.inc_fibnum;
1536 mtu = ntohs(icp->icmp_nextmtu);
1538 * If no alternative MTU was
1539 * proposed, try the next smaller
1544 ntohs(ip->ip_len), 1);
1545 if (mtu < V_tcp_minmss
1546 + sizeof(struct tcpiphdr))
1548 + sizeof(struct tcpiphdr);
1550 * Only cache the MTU if it
1551 * is smaller than the interface
1552 * or route MTU. tcp_mtudisc()
1553 * will do right thing by itself.
1555 if (mtu <= tcp_maxmtu(&inc, NULL))
1556 tcp_hc_updatemtu(&inc, mtu);
1557 tcp_mtudisc(inp, mtu);
1559 inp = (*notify)(inp,
1560 inetctlerrmap[cmd]);
1566 bzero(&inc, sizeof(inc));
1567 inc.inc_fport = th->th_dport;
1568 inc.inc_lport = th->th_sport;
1569 inc.inc_faddr = faddr;
1570 inc.inc_laddr = ip->ip_src;
1571 syncache_unreach(&inc, th);
1573 INP_INFO_WUNLOCK(&V_tcbinfo);
1575 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1581 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1584 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1585 struct ip6_hdr *ip6;
1587 struct ip6ctlparam *ip6cp = NULL;
1588 const struct sockaddr_in6 *sa6_src = NULL;
1590 struct tcp_portonly {
1595 if (sa->sa_family != AF_INET6 ||
1596 sa->sa_len != sizeof(struct sockaddr_in6))
1599 if (cmd == PRC_MSGSIZE)
1600 notify = tcp_mtudisc_notify;
1601 else if (!PRC_IS_REDIRECT(cmd) &&
1602 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1604 /* Source quench is depreciated. */
1605 else if (cmd == PRC_QUENCH)
1608 /* if the parameter is from icmp6, decode it. */
1610 ip6cp = (struct ip6ctlparam *)d;
1612 ip6 = ip6cp->ip6c_ip6;
1613 off = ip6cp->ip6c_off;
1614 sa6_src = ip6cp->ip6c_src;
1618 off = 0; /* fool gcc */
1623 struct in_conninfo inc;
1625 * XXX: We assume that when IPV6 is non NULL,
1626 * M and OFF are valid.
1629 /* check if we can safely examine src and dst ports */
1630 if (m->m_pkthdr.len < off + sizeof(*thp))
1633 bzero(&th, sizeof(th));
1634 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1636 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1637 (struct sockaddr *)ip6cp->ip6c_src,
1638 th.th_sport, cmd, NULL, notify);
1640 bzero(&inc, sizeof(inc));
1641 inc.inc_fport = th.th_dport;
1642 inc.inc_lport = th.th_sport;
1643 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1644 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1645 inc.inc_flags |= INC_ISIPV6;
1646 INP_INFO_WLOCK(&V_tcbinfo);
1647 syncache_unreach(&inc, &th);
1648 INP_INFO_WUNLOCK(&V_tcbinfo);
1650 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1651 0, cmd, NULL, notify);
1657 * Following is where TCP initial sequence number generation occurs.
1659 * There are two places where we must use initial sequence numbers:
1660 * 1. In SYN-ACK packets.
1661 * 2. In SYN packets.
1663 * All ISNs for SYN-ACK packets are generated by the syncache. See
1664 * tcp_syncache.c for details.
1666 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1667 * depends on this property. In addition, these ISNs should be
1668 * unguessable so as to prevent connection hijacking. To satisfy
1669 * the requirements of this situation, the algorithm outlined in
1670 * RFC 1948 is used, with only small modifications.
1672 * Implementation details:
1674 * Time is based off the system timer, and is corrected so that it
1675 * increases by one megabyte per second. This allows for proper
1676 * recycling on high speed LANs while still leaving over an hour
1679 * As reading the *exact* system time is too expensive to be done
1680 * whenever setting up a TCP connection, we increment the time
1681 * offset in two ways. First, a small random positive increment
1682 * is added to isn_offset for each connection that is set up.
1683 * Second, the function tcp_isn_tick fires once per clock tick
1684 * and increments isn_offset as necessary so that sequence numbers
1685 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1686 * random positive increments serve only to ensure that the same
1687 * exact sequence number is never sent out twice (as could otherwise
1688 * happen when a port is recycled in less than the system tick
1691 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1692 * between seeding of isn_secret. This is normally set to zero,
1693 * as reseeding should not be necessary.
1695 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1696 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1697 * general, this means holding an exclusive (write) lock.
1700 #define ISN_BYTES_PER_SECOND 1048576
1701 #define ISN_STATIC_INCREMENT 4096
1702 #define ISN_RANDOM_INCREMENT (4096 - 1)
1704 static VNET_DEFINE(u_char, isn_secret[32]);
1705 static VNET_DEFINE(int, isn_last);
1706 static VNET_DEFINE(int, isn_last_reseed);
1707 static VNET_DEFINE(u_int32_t, isn_offset);
1708 static VNET_DEFINE(u_int32_t, isn_offset_old);
1710 #define V_isn_secret VNET(isn_secret)
1711 #define V_isn_last VNET(isn_last)
1712 #define V_isn_last_reseed VNET(isn_last_reseed)
1713 #define V_isn_offset VNET(isn_offset)
1714 #define V_isn_offset_old VNET(isn_offset_old)
1717 tcp_new_isn(struct tcpcb *tp)
1720 u_int32_t md5_buffer[4];
1722 u_int32_t projected_offset;
1724 INP_WLOCK_ASSERT(tp->t_inpcb);
1727 /* Seed if this is the first use, reseed if requested. */
1728 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1729 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1731 read_random(&V_isn_secret, sizeof(V_isn_secret));
1732 V_isn_last_reseed = ticks;
1735 /* Compute the md5 hash and return the ISN. */
1737 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1738 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1740 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1741 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1742 sizeof(struct in6_addr));
1743 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1744 sizeof(struct in6_addr));
1748 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1749 sizeof(struct in_addr));
1750 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1751 sizeof(struct in_addr));
1753 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1754 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1755 new_isn = (tcp_seq) md5_buffer[0];
1756 V_isn_offset += ISN_STATIC_INCREMENT +
1757 (arc4random() & ISN_RANDOM_INCREMENT);
1758 if (ticks != V_isn_last) {
1759 projected_offset = V_isn_offset_old +
1760 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1761 if (SEQ_GT(projected_offset, V_isn_offset))
1762 V_isn_offset = projected_offset;
1763 V_isn_offset_old = V_isn_offset;
1766 new_isn += V_isn_offset;
1772 * When a specific ICMP unreachable message is received and the
1773 * connection state is SYN-SENT, drop the connection. This behavior
1774 * is controlled by the icmp_may_rst sysctl.
1777 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1781 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1782 INP_WLOCK_ASSERT(inp);
1784 if ((inp->inp_flags & INP_TIMEWAIT) ||
1785 (inp->inp_flags & INP_DROPPED))
1788 tp = intotcpcb(inp);
1789 if (tp->t_state != TCPS_SYN_SENT)
1792 tp = tcp_drop(tp, errno);
1800 * When `need fragmentation' ICMP is received, update our idea of the MSS
1801 * based on the new value. Also nudge TCP to send something, since we
1802 * know the packet we just sent was dropped.
1803 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1805 static struct inpcb *
1806 tcp_mtudisc_notify(struct inpcb *inp, int error)
1809 return (tcp_mtudisc(inp, -1));
1813 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1818 INP_WLOCK_ASSERT(inp);
1819 if ((inp->inp_flags & INP_TIMEWAIT) ||
1820 (inp->inp_flags & INP_DROPPED))
1823 tp = intotcpcb(inp);
1824 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1826 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1828 so = inp->inp_socket;
1829 SOCKBUF_LOCK(&so->so_snd);
1830 /* If the mss is larger than the socket buffer, decrease the mss. */
1831 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1832 tp->t_maxseg = so->so_snd.sb_hiwat;
1833 SOCKBUF_UNLOCK(&so->so_snd);
1835 TCPSTAT_INC(tcps_mturesent);
1837 tp->snd_nxt = tp->snd_una;
1838 tcp_free_sackholes(tp);
1839 tp->snd_recover = tp->snd_max;
1840 if (tp->t_flags & TF_SACK_PERMIT)
1841 EXIT_FASTRECOVERY(tp->t_flags);
1848 * Look-up the routing entry to the peer of this inpcb. If no route
1849 * is found and it cannot be allocated, then return 0. This routine
1850 * is called by TCP routines that access the rmx structure and by
1851 * tcp_mss_update to get the peer/interface MTU.
1854 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1857 struct sockaddr_in *dst;
1861 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1863 bzero(&sro, sizeof(sro));
1864 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1865 dst = (struct sockaddr_in *)&sro.ro_dst;
1866 dst->sin_family = AF_INET;
1867 dst->sin_len = sizeof(*dst);
1868 dst->sin_addr = inc->inc_faddr;
1869 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1871 if (sro.ro_rt != NULL) {
1872 ifp = sro.ro_rt->rt_ifp;
1873 if (sro.ro_rt->rt_mtu == 0)
1874 maxmtu = ifp->if_mtu;
1876 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1878 /* Report additional interface capabilities. */
1880 if (ifp->if_capenable & IFCAP_TSO4 &&
1881 ifp->if_hwassist & CSUM_TSO) {
1882 cap->ifcap |= CSUM_TSO;
1883 cap->tsomax = ifp->if_hw_tsomax;
1884 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1885 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1896 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1898 struct route_in6 sro6;
1902 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1904 bzero(&sro6, sizeof(sro6));
1905 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1906 sro6.ro_dst.sin6_family = AF_INET6;
1907 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1908 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1909 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1911 if (sro6.ro_rt != NULL) {
1912 ifp = sro6.ro_rt->rt_ifp;
1913 if (sro6.ro_rt->rt_mtu == 0)
1914 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1916 maxmtu = min(sro6.ro_rt->rt_mtu,
1917 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1919 /* Report additional interface capabilities. */
1921 if (ifp->if_capenable & IFCAP_TSO6 &&
1922 ifp->if_hwassist & CSUM_TSO) {
1923 cap->ifcap |= CSUM_TSO;
1924 cap->tsomax = ifp->if_hw_tsomax;
1925 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1926 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1937 /* compute ESP/AH header size for TCP, including outer IP header. */
1939 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1946 struct ip6_hdr *ip6;
1950 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
1951 (!key_havesp(IPSEC_DIR_OUTBOUND)))
1953 m = m_gethdr(M_NOWAIT, MT_DATA);
1958 if ((inp->inp_vflag & INP_IPV6) != 0) {
1959 ip6 = mtod(m, struct ip6_hdr *);
1960 th = (struct tcphdr *)(ip6 + 1);
1961 m->m_pkthdr.len = m->m_len =
1962 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1963 tcpip_fillheaders(inp, ip6, th);
1964 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1968 ip = mtod(m, struct ip *);
1969 th = (struct tcphdr *)(ip + 1);
1970 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1971 tcpip_fillheaders(inp, ip, th);
1972 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1980 #ifdef TCP_SIGNATURE
1982 * Callback function invoked by m_apply() to digest TCP segment data
1983 * contained within an mbuf chain.
1986 tcp_signature_apply(void *fstate, void *data, u_int len)
1989 MD5Update(fstate, (u_char *)data, len);
1994 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1997 * m pointer to head of mbuf chain
1999 * len length of TCP segment data, excluding options
2000 * optlen length of TCP segment options
2001 * buf pointer to storage for computed MD5 digest
2002 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2004 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2005 * When called from tcp_input(), we can be sure that th_sum has been
2006 * zeroed out and verified already.
2008 * Return 0 if successful, otherwise return -1.
2010 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2011 * search with the destination IP address, and a 'magic SPI' to be
2012 * determined by the application. This is hardcoded elsewhere to 1179
2013 * right now. Another branch of this code exists which uses the SPD to
2014 * specify per-application flows but it is unstable.
2017 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2018 u_char *buf, u_int direction)
2020 union sockaddr_union dst;
2022 struct ippseudo ippseudo;
2028 struct ipovly *ipovly;
2030 struct secasvar *sav;
2033 struct ip6_hdr *ip6;
2034 struct in6_addr in6;
2035 char ip6buf[INET6_ADDRSTRLEN];
2041 KASSERT(m != NULL, ("NULL mbuf chain"));
2042 KASSERT(buf != NULL, ("NULL signature pointer"));
2044 /* Extract the destination from the IP header in the mbuf. */
2045 bzero(&dst, sizeof(union sockaddr_union));
2046 ip = mtod(m, struct ip *);
2048 ip6 = NULL; /* Make the compiler happy. */
2053 dst.sa.sa_len = sizeof(struct sockaddr_in);
2054 dst.sa.sa_family = AF_INET;
2055 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2056 ip->ip_src : ip->ip_dst;
2060 case (IPV6_VERSION >> 4):
2061 ip6 = mtod(m, struct ip6_hdr *);
2062 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2063 dst.sa.sa_family = AF_INET6;
2064 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2065 ip6->ip6_src : ip6->ip6_dst;
2074 /* Look up an SADB entry which matches the address of the peer. */
2075 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2077 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2078 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2080 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2081 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2089 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2091 * XXX The ippseudo header MUST be digested in network byte order,
2092 * or else we'll fail the regression test. Assume all fields we've
2093 * been doing arithmetic on have been in host byte order.
2094 * XXX One cannot depend on ipovly->ih_len here. When called from
2095 * tcp_output(), the underlying ip_len member has not yet been set.
2100 ipovly = (struct ipovly *)ip;
2101 ippseudo.ippseudo_src = ipovly->ih_src;
2102 ippseudo.ippseudo_dst = ipovly->ih_dst;
2103 ippseudo.ippseudo_pad = 0;
2104 ippseudo.ippseudo_p = IPPROTO_TCP;
2105 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2107 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2109 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2110 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2115 * RFC 2385, 2.0 Proposal
2116 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2117 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2118 * extended next header value (to form 32 bits), and 32-bit segment
2120 * Note: Upper-Layer Packet Length comes before Next Header.
2122 case (IPV6_VERSION >> 4):
2124 in6_clearscope(&in6);
2125 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2127 in6_clearscope(&in6);
2128 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2129 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2130 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2132 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2133 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2134 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2136 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2138 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2139 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2150 * Step 2: Update MD5 hash with TCP header, excluding options.
2151 * The TCP checksum must be set to zero.
2153 savecsum = th->th_sum;
2155 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2156 th->th_sum = savecsum;
2159 * Step 3: Update MD5 hash with TCP segment data.
2160 * Use m_apply() to avoid an early m_pullup().
2163 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2166 * Step 4: Update MD5 hash with shared secret.
2168 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2169 MD5Final(buf, &ctx);
2171 key_sa_recordxfer(sav, m);
2177 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2180 * m pointer to head of mbuf chain
2181 * len length of TCP segment data, excluding options
2182 * optlen length of TCP segment options
2183 * buf pointer to storage for computed MD5 digest
2184 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2186 * Return 1 if successful, otherwise return 0.
2189 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2190 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2192 char tmpdigest[TCP_SIGLEN];
2194 if (tcp_sig_checksigs == 0)
2196 if ((tcpbflag & TF_SIGNATURE) == 0) {
2197 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2200 * If this socket is not expecting signature but
2201 * the segment contains signature just fail.
2203 TCPSTAT_INC(tcps_sig_err_sigopt);
2204 TCPSTAT_INC(tcps_sig_rcvbadsig);
2208 /* Signature is not expected, and not present in segment. */
2213 * If this socket is expecting signature but the segment does not
2214 * contain any just fail.
2216 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2217 TCPSTAT_INC(tcps_sig_err_nosigopt);
2218 TCPSTAT_INC(tcps_sig_rcvbadsig);
2221 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2222 IPSEC_DIR_INBOUND) == -1) {
2223 TCPSTAT_INC(tcps_sig_err_buildsig);
2224 TCPSTAT_INC(tcps_sig_rcvbadsig);
2228 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2229 TCPSTAT_INC(tcps_sig_rcvbadsig);
2232 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2235 #endif /* TCP_SIGNATURE */
2238 sysctl_drop(SYSCTL_HANDLER_ARGS)
2240 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2241 struct sockaddr_storage addrs[2];
2245 struct sockaddr_in *fin, *lin;
2247 struct sockaddr_in6 *fin6, *lin6;
2258 if (req->oldptr != NULL || req->oldlen != 0)
2260 if (req->newptr == NULL)
2262 if (req->newlen < sizeof(addrs))
2264 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2268 switch (addrs[0].ss_family) {
2271 fin6 = (struct sockaddr_in6 *)&addrs[0];
2272 lin6 = (struct sockaddr_in6 *)&addrs[1];
2273 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2274 lin6->sin6_len != sizeof(struct sockaddr_in6))
2276 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2277 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2279 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2280 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2281 fin = (struct sockaddr_in *)&addrs[0];
2282 lin = (struct sockaddr_in *)&addrs[1];
2285 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2288 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2295 fin = (struct sockaddr_in *)&addrs[0];
2296 lin = (struct sockaddr_in *)&addrs[1];
2297 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2298 lin->sin_len != sizeof(struct sockaddr_in))
2305 INP_INFO_WLOCK(&V_tcbinfo);
2306 switch (addrs[0].ss_family) {
2309 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2310 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2311 INPLOOKUP_WLOCKPCB, NULL);
2316 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2317 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2322 if (inp->inp_flags & INP_TIMEWAIT) {
2324 * XXXRW: There currently exists a state where an
2325 * inpcb is present, but its timewait state has been
2326 * discarded. For now, don't allow dropping of this
2334 } else if (!(inp->inp_flags & INP_DROPPED) &&
2335 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2336 tp = intotcpcb(inp);
2337 tp = tcp_drop(tp, ECONNABORTED);
2344 INP_INFO_WUNLOCK(&V_tcbinfo);
2348 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2349 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2350 0, sysctl_drop, "", "Drop TCP connection");
2353 * Generate a standardized TCP log line for use throughout the
2354 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2355 * allow use in the interrupt context.
2357 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2358 * NB: The function may return NULL if memory allocation failed.
2360 * Due to header inclusion and ordering limitations the struct ip
2361 * and ip6_hdr pointers have to be passed as void pointers.
2364 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2368 /* Is logging enabled? */
2369 if (tcp_log_in_vain == 0)
2372 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2376 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2380 /* Is logging enabled? */
2381 if (tcp_log_debug == 0)
2384 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2388 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2395 const struct ip6_hdr *ip6;
2397 ip6 = (const struct ip6_hdr *)ip6hdr;
2399 ip = (struct ip *)ip4hdr;
2402 * The log line looks like this:
2403 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2405 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2406 sizeof(PRINT_TH_FLAGS) + 1 +
2408 2 * INET6_ADDRSTRLEN;
2410 2 * INET_ADDRSTRLEN;
2413 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2417 strcat(s, "TCP: [");
2420 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2421 inet_ntoa_r(inc->inc_faddr, sp);
2423 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2425 inet_ntoa_r(inc->inc_laddr, sp);
2427 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2430 ip6_sprintf(sp, &inc->inc6_faddr);
2432 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2434 ip6_sprintf(sp, &inc->inc6_laddr);
2436 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2437 } else if (ip6 && th) {
2438 ip6_sprintf(sp, &ip6->ip6_src);
2440 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2442 ip6_sprintf(sp, &ip6->ip6_dst);
2444 sprintf(sp, "]:%i", ntohs(th->th_dport));
2447 } else if (ip && th) {
2448 inet_ntoa_r(ip->ip_src, sp);
2450 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2452 inet_ntoa_r(ip->ip_dst, sp);
2454 sprintf(sp, "]:%i", ntohs(th->th_dport));
2462 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2463 if (*(s + size - 1) != '\0')
2464 panic("%s: string too long", __func__);
2469 * A subroutine which makes it easy to track TCP state changes with DTrace.
2470 * This function shouldn't be called for t_state initializations that don't
2471 * correspond to actual TCP state transitions.
2474 tcp_state_change(struct tcpcb *tp, int newstate)
2476 #if defined(KDTRACE_HOOKS)
2477 int pstate = tp->t_state;
2480 tp->t_state = newstate;
2481 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);