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>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/random.h>
64 #include <net/route.h>
66 #include <net/if_var.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_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
142 CTLFLAG_VNET | 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_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
164 CTLFLAG_VNET | 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_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_minmss), 0,
180 "Minimum TCP Maximum Segment Size");
182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | 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;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
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_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | 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_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | 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_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
211 &VNET_NAME(tcp_isn_reseed_interval), 0,
212 "Seconds between reseeding of ISN secret");
214 static int tcp_soreceive_stream;
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 /* XXX virtualize those bellow? */
389 tcp_delacktime = TCPTV_DELACK;
390 tcp_keepinit = TCPTV_KEEP_INIT;
391 tcp_keepidle = TCPTV_KEEP_IDLE;
392 tcp_keepintvl = TCPTV_KEEPINTVL;
393 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
395 tcp_rexmit_min = TCPTV_MIN;
396 if (tcp_rexmit_min < 1)
398 tcp_rexmit_slop = TCPTV_CPU_VAR;
399 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
400 tcp_tcbhashsize = hashsize;
402 if (tcp_soreceive_stream) {
404 tcp_usrreqs.pru_soreceive = soreceive_stream;
407 tcp6_usrreqs.pru_soreceive = soreceive_stream;
412 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
414 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
416 if (max_protohdr < TCP_MINPROTOHDR)
417 max_protohdr = TCP_MINPROTOHDR;
418 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
420 #undef TCP_MINPROTOHDR
423 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
424 SHUTDOWN_PRI_DEFAULT);
425 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
426 EVENTHANDLER_PRI_ANY);
438 in_pcbinfo_destroy(&V_tcbinfo);
439 uma_zdestroy(V_sack_hole_zone);
440 uma_zdestroy(V_tcpcb_zone);
442 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
444 printf("%s: WARNING: unable to deregister helper hook "
445 "type=%d, id=%d: error %d returned\n", __func__,
446 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
448 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
450 printf("%s: WARNING: unable to deregister helper hook "
451 "type=%d, id=%d: error %d returned\n", __func__,
452 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
464 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
465 * tcp_template used to store this data in mbufs, but we now recopy it out
466 * of the tcpcb each time to conserve mbufs.
469 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
471 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
473 INP_WLOCK_ASSERT(inp);
476 if ((inp->inp_vflag & INP_IPV6) != 0) {
479 ip6 = (struct ip6_hdr *)ip_ptr;
480 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
481 (inp->inp_flow & IPV6_FLOWINFO_MASK);
482 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
483 (IPV6_VERSION & IPV6_VERSION_MASK);
484 ip6->ip6_nxt = IPPROTO_TCP;
485 ip6->ip6_plen = htons(sizeof(struct tcphdr));
486 ip6->ip6_src = inp->in6p_laddr;
487 ip6->ip6_dst = inp->in6p_faddr;
490 #if defined(INET6) && defined(INET)
497 ip = (struct ip *)ip_ptr;
498 ip->ip_v = IPVERSION;
500 ip->ip_tos = inp->inp_ip_tos;
504 ip->ip_ttl = inp->inp_ip_ttl;
506 ip->ip_p = IPPROTO_TCP;
507 ip->ip_src = inp->inp_laddr;
508 ip->ip_dst = inp->inp_faddr;
511 th->th_sport = inp->inp_lport;
512 th->th_dport = inp->inp_fport;
520 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
524 * Create template to be used to send tcp packets on a connection.
525 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
526 * use for this function is in keepalives, which use tcp_respond.
529 tcpip_maketemplate(struct inpcb *inp)
533 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
536 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
541 * Send a single message to the TCP at address specified by
542 * the given TCP/IP header. If m == NULL, then we make a copy
543 * of the tcpiphdr at th and send directly to the addressed host.
544 * This is used to force keep alive messages out using the TCP
545 * template for a connection. If flags are given then we send
546 * a message back to the TCP which originated the segment th,
547 * and discard the mbuf containing it and any other attached mbufs.
549 * In any case the ack and sequence number of the transmitted
550 * segment are as specified by the parameters.
552 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
555 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
556 tcp_seq ack, tcp_seq seq, int flags)
569 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
572 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
579 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
580 INP_WLOCK_ASSERT(inp);
585 if (!(flags & TH_RST)) {
586 win = sbspace(&inp->inp_socket->so_rcv);
587 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
588 win = (long)TCP_MAXWIN << tp->rcv_scale;
592 m = m_gethdr(M_NOWAIT, MT_DATA);
596 m->m_data += max_linkhdr;
599 bcopy((caddr_t)ip6, mtod(m, caddr_t),
600 sizeof(struct ip6_hdr));
601 ip6 = mtod(m, struct ip6_hdr *);
602 nth = (struct tcphdr *)(ip6 + 1);
606 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
607 ip = mtod(m, struct ip *);
608 nth = (struct tcphdr *)(ip + 1);
610 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
615 * XXX MRT We inherrit the FIB, which is lucky.
619 m->m_data = (caddr_t)ipgen;
620 /* m_len is set later */
622 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
625 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
626 nth = (struct tcphdr *)(ip6 + 1);
630 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
631 nth = (struct tcphdr *)(ip + 1);
635 * this is usually a case when an extension header
636 * exists between the IPv6 header and the
639 nth->th_sport = th->th_sport;
640 nth->th_dport = th->th_dport;
642 xchg(nth->th_dport, nth->th_sport, uint16_t);
648 ip6->ip6_vfc = IPV6_VERSION;
649 ip6->ip6_nxt = IPPROTO_TCP;
650 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
651 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
654 #if defined(INET) && defined(INET6)
659 tlen += sizeof (struct tcpiphdr);
660 ip->ip_len = htons(tlen);
661 ip->ip_ttl = V_ip_defttl;
662 if (V_path_mtu_discovery)
663 ip->ip_off |= htons(IP_DF);
667 m->m_pkthdr.len = tlen;
668 m->m_pkthdr.rcvif = NULL;
672 * Packet is associated with a socket, so allow the
673 * label of the response to reflect the socket label.
675 INP_WLOCK_ASSERT(inp);
676 mac_inpcb_create_mbuf(inp, m);
679 * Packet is not associated with a socket, so possibly
680 * update the label in place.
682 mac_netinet_tcp_reply(m);
685 nth->th_seq = htonl(seq);
686 nth->th_ack = htonl(ack);
688 nth->th_off = sizeof (struct tcphdr) >> 2;
689 nth->th_flags = flags;
691 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
693 nth->th_win = htons((u_short)win);
696 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
699 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
700 nth->th_sum = in6_cksum_pseudo(ip6,
701 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
702 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
706 #if defined(INET6) && defined(INET)
711 m->m_pkthdr.csum_flags = CSUM_TCP;
712 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
713 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
717 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
718 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
721 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
724 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
727 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
729 #if defined(INET) && defined(INET6)
733 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
738 * Create a new TCP control block, making an
739 * empty reassembly queue and hooking it to the argument
740 * protocol control block. The `inp' parameter must have
741 * come from the zone allocator set up in tcp_init().
744 tcp_newtcpcb(struct inpcb *inp)
746 struct tcpcb_mem *tm;
749 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
752 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
757 /* Initialise cc_var struct for this tcpcb. */
759 tp->ccv->type = IPPROTO_TCP;
760 tp->ccv->ccvc.tcp = tp;
763 * Use the current system default CC algorithm.
766 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
767 CC_ALGO(tp) = CC_DEFAULT();
770 if (CC_ALGO(tp)->cb_init != NULL)
771 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
772 uma_zfree(V_tcpcb_zone, tm);
777 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
778 uma_zfree(V_tcpcb_zone, tm);
783 tp->t_vnet = inp->inp_vnet;
785 tp->t_timers = &tm->tt;
786 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
787 tp->t_maxseg = tp->t_maxopd =
789 isipv6 ? V_tcp_v6mssdflt :
793 /* Set up our timeouts. */
794 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
795 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
796 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
797 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
798 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
800 if (V_tcp_do_rfc1323)
801 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
803 tp->t_flags |= TF_SACK_PERMIT;
804 TAILQ_INIT(&tp->snd_holes);
806 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
809 in_pcbref(inp); /* Reference for tcpcb */
813 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
814 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
815 * reasonable initial retransmit time.
817 tp->t_srtt = TCPTV_SRTTBASE;
818 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
819 tp->t_rttmin = tcp_rexmit_min;
820 tp->t_rxtcur = TCPTV_RTOBASE;
821 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
822 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
823 tp->t_rcvtime = ticks;
825 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
826 * because the socket may be bound to an IPv6 wildcard address,
827 * which may match an IPv4-mapped IPv6 address.
829 inp->inp_ip_ttl = V_ip_defttl;
831 return (tp); /* XXX */
835 * Switch the congestion control algorithm back to NewReno for any active
836 * control blocks using an algorithm which is about to go away.
837 * This ensures the CC framework can allow the unload to proceed without leaving
838 * any dangling pointers which would trigger a panic.
839 * Returning non-zero would inform the CC framework that something went wrong
840 * and it would be unsafe to allow the unload to proceed. However, there is no
841 * way for this to occur with this implementation so we always return zero.
844 tcp_ccalgounload(struct cc_algo *unload_algo)
846 struct cc_algo *tmpalgo;
849 VNET_ITERATOR_DECL(vnet_iter);
852 * Check all active control blocks across all network stacks and change
853 * any that are using "unload_algo" back to NewReno. If "unload_algo"
854 * requires cleanup code to be run, call it.
857 VNET_FOREACH(vnet_iter) {
858 CURVNET_SET(vnet_iter);
859 INP_INFO_RLOCK(&V_tcbinfo);
861 * New connections already part way through being initialised
862 * with the CC algo we're removing will not race with this code
863 * because the INP_INFO_WLOCK is held during initialisation. We
864 * therefore don't enter the loop below until the connection
865 * list has stabilised.
867 LIST_FOREACH(inp, &V_tcb, inp_list) {
869 /* Important to skip tcptw structs. */
870 if (!(inp->inp_flags & INP_TIMEWAIT) &&
871 (tp = intotcpcb(inp)) != NULL) {
873 * By holding INP_WLOCK here, we are assured
874 * that the connection is not currently
875 * executing inside the CC module's functions
876 * i.e. it is safe to make the switch back to
879 if (CC_ALGO(tp) == unload_algo) {
880 tmpalgo = CC_ALGO(tp);
881 /* NewReno does not require any init. */
882 CC_ALGO(tp) = &newreno_cc_algo;
883 if (tmpalgo->cb_destroy != NULL)
884 tmpalgo->cb_destroy(tp->ccv);
889 INP_INFO_RUNLOCK(&V_tcbinfo);
898 * Drop a TCP connection, reporting
899 * the specified error. If connection is synchronized,
900 * then send a RST to peer.
903 tcp_drop(struct tcpcb *tp, int errno)
905 struct socket *so = tp->t_inpcb->inp_socket;
907 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
908 INP_WLOCK_ASSERT(tp->t_inpcb);
910 if (TCPS_HAVERCVDSYN(tp->t_state)) {
911 tcp_state_change(tp, TCPS_CLOSED);
912 (void) tcp_output(tp);
913 TCPSTAT_INC(tcps_drops);
915 TCPSTAT_INC(tcps_conndrops);
916 if (errno == ETIMEDOUT && tp->t_softerror)
917 errno = tp->t_softerror;
918 so->so_error = errno;
919 return (tcp_close(tp));
923 tcp_discardcb(struct tcpcb *tp)
925 struct inpcb *inp = tp->t_inpcb;
926 struct socket *so = inp->inp_socket;
928 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
932 INP_WLOCK_ASSERT(inp);
935 * Make sure that all of our timers are stopped before we delete the
938 * If stopping a timer fails, we schedule a discard function in same
939 * callout, and the last discard function called will take care of
940 * deleting the tcpcb.
942 tcp_timer_stop(tp, TT_REXMT);
943 tcp_timer_stop(tp, TT_PERSIST);
944 tcp_timer_stop(tp, TT_KEEP);
945 tcp_timer_stop(tp, TT_2MSL);
946 tcp_timer_stop(tp, TT_DELACK);
949 * If we got enough samples through the srtt filter,
950 * save the rtt and rttvar in the routing entry.
951 * 'Enough' is arbitrarily defined as 4 rtt samples.
952 * 4 samples is enough for the srtt filter to converge
953 * to within enough % of the correct value; fewer samples
954 * and we could save a bogus rtt. The danger is not high
955 * as tcp quickly recovers from everything.
956 * XXX: Works very well but needs some more statistics!
958 if (tp->t_rttupdated >= 4) {
959 struct hc_metrics_lite metrics;
962 bzero(&metrics, sizeof(metrics));
964 * Update the ssthresh always when the conditions below
965 * are satisfied. This gives us better new start value
966 * for the congestion avoidance for new connections.
967 * ssthresh is only set if packet loss occured on a session.
969 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
970 * being torn down. Ideally this code would not use 'so'.
972 ssthresh = tp->snd_ssthresh;
973 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
975 * convert the limit from user data bytes to
976 * packets then to packet data bytes.
978 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
981 ssthresh *= (u_long)(tp->t_maxseg +
983 (isipv6 ? sizeof (struct ip6_hdr) +
984 sizeof (struct tcphdr) :
986 sizeof (struct tcpiphdr)
993 metrics.rmx_ssthresh = ssthresh;
995 metrics.rmx_rtt = tp->t_srtt;
996 metrics.rmx_rttvar = tp->t_rttvar;
997 metrics.rmx_cwnd = tp->snd_cwnd;
998 metrics.rmx_sendpipe = 0;
999 metrics.rmx_recvpipe = 0;
1001 tcp_hc_update(&inp->inp_inc, &metrics);
1004 /* free the reassembly queue, if any */
1005 tcp_reass_flush(tp);
1008 /* Disconnect offload device, if any. */
1009 if (tp->t_flags & TF_TOE)
1010 tcp_offload_detach(tp);
1013 tcp_free_sackholes(tp);
1015 /* Allow the CC algorithm to clean up after itself. */
1016 if (CC_ALGO(tp)->cb_destroy != NULL)
1017 CC_ALGO(tp)->cb_destroy(tp->ccv);
1019 khelp_destroy_osd(tp->osd);
1022 inp->inp_ppcb = NULL;
1023 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1024 /* We own the last reference on tcpcb, let's free it. */
1026 uma_zfree(V_tcpcb_zone, tp);
1027 released = in_pcbrele_wlocked(inp);
1028 KASSERT(!released, ("%s: inp %p should not have been released "
1029 "here", __func__, inp));
1034 tcp_timer_2msl_discard(void *xtp)
1037 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1041 tcp_timer_keep_discard(void *xtp)
1044 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1048 tcp_timer_persist_discard(void *xtp)
1051 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1055 tcp_timer_rexmt_discard(void *xtp)
1058 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1062 tcp_timer_delack_discard(void *xtp)
1065 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1069 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1073 CURVNET_SET(tp->t_vnet);
1074 INP_INFO_WLOCK(&V_tcbinfo);
1076 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1079 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1080 ("%s: tcpcb has to be stopped here", __func__));
1081 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1082 ("%s: discard callout should be running", __func__));
1083 tp->t_timers->tt_flags &= ~timer_type;
1084 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1085 /* We own the last reference on this tcpcb, let's free it. */
1087 uma_zfree(V_tcpcb_zone, tp);
1088 if (in_pcbrele_wlocked(inp)) {
1089 INP_INFO_WUNLOCK(&V_tcbinfo);
1095 INP_INFO_WUNLOCK(&V_tcbinfo);
1100 * Attempt to close a TCP control block, marking it as dropped, and freeing
1101 * the socket if we hold the only reference.
1104 tcp_close(struct tcpcb *tp)
1106 struct inpcb *inp = tp->t_inpcb;
1109 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1110 INP_WLOCK_ASSERT(inp);
1113 if (tp->t_state == TCPS_LISTEN)
1114 tcp_offload_listen_stop(tp);
1117 TCPSTAT_INC(tcps_closed);
1118 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1119 so = inp->inp_socket;
1120 soisdisconnected(so);
1121 if (inp->inp_flags & INP_SOCKREF) {
1122 KASSERT(so->so_state & SS_PROTOREF,
1123 ("tcp_close: !SS_PROTOREF"));
1124 inp->inp_flags &= ~INP_SOCKREF;
1128 so->so_state &= ~SS_PROTOREF;
1138 VNET_ITERATOR_DECL(vnet_iter);
1143 VNET_LIST_RLOCK_NOSLEEP();
1144 VNET_FOREACH(vnet_iter) {
1145 CURVNET_SET(vnet_iter);
1150 * Walk the tcpbs, if existing, and flush the reassembly queue,
1151 * if there is one...
1152 * XXX: The "Net/3" implementation doesn't imply that the TCP
1153 * reassembly queue should be flushed, but in a situation
1154 * where we're really low on mbufs, this is potentially
1157 INP_INFO_RLOCK(&V_tcbinfo);
1158 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1159 if (inpb->inp_flags & INP_TIMEWAIT)
1162 if ((tcpb = intotcpcb(inpb)) != NULL) {
1163 tcp_reass_flush(tcpb);
1164 tcp_clean_sackreport(tcpb);
1168 INP_INFO_RUNLOCK(&V_tcbinfo);
1171 VNET_LIST_RUNLOCK_NOSLEEP();
1175 * Notify a tcp user of an asynchronous error;
1176 * store error as soft error, but wake up user
1177 * (for now, won't do anything until can select for soft error).
1179 * Do not wake up user since there currently is no mechanism for
1180 * reporting soft errors (yet - a kqueue filter may be added).
1182 static struct inpcb *
1183 tcp_notify(struct inpcb *inp, int error)
1187 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1188 INP_WLOCK_ASSERT(inp);
1190 if ((inp->inp_flags & INP_TIMEWAIT) ||
1191 (inp->inp_flags & INP_DROPPED))
1194 tp = intotcpcb(inp);
1195 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1198 * Ignore some errors if we are hooked up.
1199 * If connection hasn't completed, has retransmitted several times,
1200 * and receives a second error, give up now. This is better
1201 * than waiting a long time to establish a connection that
1202 * can never complete.
1204 if (tp->t_state == TCPS_ESTABLISHED &&
1205 (error == EHOSTUNREACH || error == ENETUNREACH ||
1206 error == EHOSTDOWN)) {
1208 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1210 tp = tcp_drop(tp, error);
1216 tp->t_softerror = error;
1220 wakeup( &so->so_timeo);
1227 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1229 int error, i, m, n, pcb_count;
1230 struct inpcb *inp, **inp_list;
1235 * The process of preparing the TCB list is too time-consuming and
1236 * resource-intensive to repeat twice on every request.
1238 if (req->oldptr == NULL) {
1239 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1240 n += imax(n / 8, 10);
1241 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1245 if (req->newptr != NULL)
1249 * OK, now we're committed to doing something.
1251 INP_INFO_RLOCK(&V_tcbinfo);
1252 gencnt = V_tcbinfo.ipi_gencnt;
1253 n = V_tcbinfo.ipi_count;
1254 INP_INFO_RUNLOCK(&V_tcbinfo);
1256 m = syncache_pcbcount();
1258 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1259 + (n + m) * sizeof(struct xtcpcb));
1263 xig.xig_len = sizeof xig;
1264 xig.xig_count = n + m;
1265 xig.xig_gen = gencnt;
1266 xig.xig_sogen = so_gencnt;
1267 error = SYSCTL_OUT(req, &xig, sizeof xig);
1271 error = syncache_pcblist(req, m, &pcb_count);
1275 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1276 if (inp_list == NULL)
1279 INP_INFO_RLOCK(&V_tcbinfo);
1280 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1281 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1283 if (inp->inp_gencnt <= gencnt) {
1285 * XXX: This use of cr_cansee(), introduced with
1286 * TCP state changes, is not quite right, but for
1287 * now, better than nothing.
1289 if (inp->inp_flags & INP_TIMEWAIT) {
1290 if (intotw(inp) != NULL)
1291 error = cr_cansee(req->td->td_ucred,
1292 intotw(inp)->tw_cred);
1294 error = EINVAL; /* Skip this inp. */
1296 error = cr_canseeinpcb(req->td->td_ucred, inp);
1299 inp_list[i++] = inp;
1304 INP_INFO_RUNLOCK(&V_tcbinfo);
1308 for (i = 0; i < n; i++) {
1311 if (inp->inp_gencnt <= gencnt) {
1315 bzero(&xt, sizeof(xt));
1316 xt.xt_len = sizeof xt;
1317 /* XXX should avoid extra copy */
1318 bcopy(inp, &xt.xt_inp, sizeof *inp);
1319 inp_ppcb = inp->inp_ppcb;
1320 if (inp_ppcb == NULL)
1321 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1322 else if (inp->inp_flags & INP_TIMEWAIT) {
1323 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1324 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1326 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1327 if (xt.xt_tp.t_timers)
1328 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1330 if (inp->inp_socket != NULL)
1331 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1333 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1334 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1336 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1338 error = SYSCTL_OUT(req, &xt, sizeof xt);
1342 INP_INFO_WLOCK(&V_tcbinfo);
1343 for (i = 0; i < n; i++) {
1346 if (!in_pcbrele_rlocked(inp))
1349 INP_INFO_WUNLOCK(&V_tcbinfo);
1353 * Give the user an updated idea of our state.
1354 * If the generation differs from what we told
1355 * her before, she knows that something happened
1356 * while we were processing this request, and it
1357 * might be necessary to retry.
1359 INP_INFO_RLOCK(&V_tcbinfo);
1360 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1361 xig.xig_sogen = so_gencnt;
1362 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1363 INP_INFO_RUNLOCK(&V_tcbinfo);
1364 error = SYSCTL_OUT(req, &xig, sizeof xig);
1366 free(inp_list, M_TEMP);
1370 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1371 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1372 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1376 tcp_getcred(SYSCTL_HANDLER_ARGS)
1379 struct sockaddr_in addrs[2];
1383 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1386 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1389 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1390 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1392 if (inp->inp_socket == NULL)
1395 error = cr_canseeinpcb(req->td->td_ucred, inp);
1397 cru2x(inp->inp_cred, &xuc);
1402 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1406 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1407 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1408 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1413 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1416 struct sockaddr_in6 addrs[2];
1423 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1426 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1429 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1430 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1433 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1435 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1444 inp = in_pcblookup(&V_tcbinfo,
1445 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1447 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1448 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1451 inp = in6_pcblookup(&V_tcbinfo,
1452 &addrs[1].sin6_addr, addrs[1].sin6_port,
1453 &addrs[0].sin6_addr, addrs[0].sin6_port,
1454 INPLOOKUP_RLOCKPCB, NULL);
1456 if (inp->inp_socket == NULL)
1459 error = cr_canseeinpcb(req->td->td_ucred, inp);
1461 cru2x(inp->inp_cred, &xuc);
1466 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1470 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1471 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1472 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1478 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1480 struct ip *ip = vip;
1482 struct in_addr faddr;
1485 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1487 struct in_conninfo inc;
1488 tcp_seq icmp_tcp_seq;
1491 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1492 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1495 if (cmd == PRC_MSGSIZE)
1496 notify = tcp_mtudisc_notify;
1497 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1498 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1499 notify = tcp_drop_syn_sent;
1501 * Redirects don't need to be handled up here.
1503 else if (PRC_IS_REDIRECT(cmd))
1506 * Hostdead is ugly because it goes linearly through all PCBs.
1507 * XXX: We never get this from ICMP, otherwise it makes an
1508 * excellent DoS attack on machines with many connections.
1510 else if (cmd == PRC_HOSTDEAD)
1512 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1515 icp = (struct icmp *)((caddr_t)ip
1516 - offsetof(struct icmp, icmp_ip));
1517 th = (struct tcphdr *)((caddr_t)ip
1518 + (ip->ip_hl << 2));
1519 INP_INFO_WLOCK(&V_tcbinfo);
1520 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1521 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1523 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1524 !(inp->inp_flags & INP_DROPPED) &&
1525 !(inp->inp_socket == NULL)) {
1526 icmp_tcp_seq = htonl(th->th_seq);
1527 tp = intotcpcb(inp);
1528 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1529 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1530 if (cmd == PRC_MSGSIZE) {
1533 * If we got a needfrag set the MTU
1534 * in the route to the suggested new
1535 * value (if given) and then notify.
1537 bzero(&inc, sizeof(inc));
1538 inc.inc_faddr = faddr;
1540 inp->inp_inc.inc_fibnum;
1542 mtu = ntohs(icp->icmp_nextmtu);
1544 * If no alternative MTU was
1545 * proposed, try the next smaller
1550 ntohs(ip->ip_len), 1);
1551 if (mtu < V_tcp_minmss
1552 + sizeof(struct tcpiphdr))
1554 + sizeof(struct tcpiphdr);
1556 * Only cache the MTU if it
1557 * is smaller than the interface
1558 * or route MTU. tcp_mtudisc()
1559 * will do right thing by itself.
1561 if (mtu <= tcp_maxmtu(&inc, NULL))
1562 tcp_hc_updatemtu(&inc, mtu);
1563 tcp_mtudisc(inp, mtu);
1565 inp = (*notify)(inp,
1566 inetctlerrmap[cmd]);
1572 bzero(&inc, sizeof(inc));
1573 inc.inc_fport = th->th_dport;
1574 inc.inc_lport = th->th_sport;
1575 inc.inc_faddr = faddr;
1576 inc.inc_laddr = ip->ip_src;
1577 syncache_unreach(&inc, th);
1579 INP_INFO_WUNLOCK(&V_tcbinfo);
1581 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1587 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1590 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1591 struct ip6_hdr *ip6;
1593 struct ip6ctlparam *ip6cp = NULL;
1594 const struct sockaddr_in6 *sa6_src = NULL;
1596 struct tcp_portonly {
1601 if (sa->sa_family != AF_INET6 ||
1602 sa->sa_len != sizeof(struct sockaddr_in6))
1605 if (cmd == PRC_MSGSIZE)
1606 notify = tcp_mtudisc_notify;
1607 else if (!PRC_IS_REDIRECT(cmd) &&
1608 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1611 /* if the parameter is from icmp6, decode it. */
1613 ip6cp = (struct ip6ctlparam *)d;
1615 ip6 = ip6cp->ip6c_ip6;
1616 off = ip6cp->ip6c_off;
1617 sa6_src = ip6cp->ip6c_src;
1621 off = 0; /* fool gcc */
1626 struct in_conninfo inc;
1628 * XXX: We assume that when IPV6 is non NULL,
1629 * M and OFF are valid.
1632 /* check if we can safely examine src and dst ports */
1633 if (m->m_pkthdr.len < off + sizeof(*thp))
1636 bzero(&th, sizeof(th));
1637 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1639 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1640 (struct sockaddr *)ip6cp->ip6c_src,
1641 th.th_sport, cmd, NULL, notify);
1643 bzero(&inc, sizeof(inc));
1644 inc.inc_fport = th.th_dport;
1645 inc.inc_lport = th.th_sport;
1646 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1647 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1648 inc.inc_flags |= INC_ISIPV6;
1649 INP_INFO_WLOCK(&V_tcbinfo);
1650 syncache_unreach(&inc, &th);
1651 INP_INFO_WUNLOCK(&V_tcbinfo);
1653 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1654 0, cmd, NULL, notify);
1660 * Following is where TCP initial sequence number generation occurs.
1662 * There are two places where we must use initial sequence numbers:
1663 * 1. In SYN-ACK packets.
1664 * 2. In SYN packets.
1666 * All ISNs for SYN-ACK packets are generated by the syncache. See
1667 * tcp_syncache.c for details.
1669 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1670 * depends on this property. In addition, these ISNs should be
1671 * unguessable so as to prevent connection hijacking. To satisfy
1672 * the requirements of this situation, the algorithm outlined in
1673 * RFC 1948 is used, with only small modifications.
1675 * Implementation details:
1677 * Time is based off the system timer, and is corrected so that it
1678 * increases by one megabyte per second. This allows for proper
1679 * recycling on high speed LANs while still leaving over an hour
1682 * As reading the *exact* system time is too expensive to be done
1683 * whenever setting up a TCP connection, we increment the time
1684 * offset in two ways. First, a small random positive increment
1685 * is added to isn_offset for each connection that is set up.
1686 * Second, the function tcp_isn_tick fires once per clock tick
1687 * and increments isn_offset as necessary so that sequence numbers
1688 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1689 * random positive increments serve only to ensure that the same
1690 * exact sequence number is never sent out twice (as could otherwise
1691 * happen when a port is recycled in less than the system tick
1694 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1695 * between seeding of isn_secret. This is normally set to zero,
1696 * as reseeding should not be necessary.
1698 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1699 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1700 * general, this means holding an exclusive (write) lock.
1703 #define ISN_BYTES_PER_SECOND 1048576
1704 #define ISN_STATIC_INCREMENT 4096
1705 #define ISN_RANDOM_INCREMENT (4096 - 1)
1707 static VNET_DEFINE(u_char, isn_secret[32]);
1708 static VNET_DEFINE(int, isn_last);
1709 static VNET_DEFINE(int, isn_last_reseed);
1710 static VNET_DEFINE(u_int32_t, isn_offset);
1711 static VNET_DEFINE(u_int32_t, isn_offset_old);
1713 #define V_isn_secret VNET(isn_secret)
1714 #define V_isn_last VNET(isn_last)
1715 #define V_isn_last_reseed VNET(isn_last_reseed)
1716 #define V_isn_offset VNET(isn_offset)
1717 #define V_isn_offset_old VNET(isn_offset_old)
1720 tcp_new_isn(struct tcpcb *tp)
1723 u_int32_t md5_buffer[4];
1725 u_int32_t projected_offset;
1727 INP_WLOCK_ASSERT(tp->t_inpcb);
1730 /* Seed if this is the first use, reseed if requested. */
1731 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1732 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1734 read_random(&V_isn_secret, sizeof(V_isn_secret));
1735 V_isn_last_reseed = ticks;
1738 /* Compute the md5 hash and return the ISN. */
1740 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1741 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1743 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1744 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1745 sizeof(struct in6_addr));
1746 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1747 sizeof(struct in6_addr));
1751 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1752 sizeof(struct in_addr));
1753 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1754 sizeof(struct in_addr));
1756 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1757 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1758 new_isn = (tcp_seq) md5_buffer[0];
1759 V_isn_offset += ISN_STATIC_INCREMENT +
1760 (arc4random() & ISN_RANDOM_INCREMENT);
1761 if (ticks != V_isn_last) {
1762 projected_offset = V_isn_offset_old +
1763 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1764 if (SEQ_GT(projected_offset, V_isn_offset))
1765 V_isn_offset = projected_offset;
1766 V_isn_offset_old = V_isn_offset;
1769 new_isn += V_isn_offset;
1775 * When a specific ICMP unreachable message is received and the
1776 * connection state is SYN-SENT, drop the connection. This behavior
1777 * is controlled by the icmp_may_rst sysctl.
1780 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1784 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1785 INP_WLOCK_ASSERT(inp);
1787 if ((inp->inp_flags & INP_TIMEWAIT) ||
1788 (inp->inp_flags & INP_DROPPED))
1791 tp = intotcpcb(inp);
1792 if (tp->t_state != TCPS_SYN_SENT)
1795 tp = tcp_drop(tp, errno);
1803 * When `need fragmentation' ICMP is received, update our idea of the MSS
1804 * based on the new value. Also nudge TCP to send something, since we
1805 * know the packet we just sent was dropped.
1806 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1808 static struct inpcb *
1809 tcp_mtudisc_notify(struct inpcb *inp, int error)
1812 return (tcp_mtudisc(inp, -1));
1816 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1821 INP_WLOCK_ASSERT(inp);
1822 if ((inp->inp_flags & INP_TIMEWAIT) ||
1823 (inp->inp_flags & INP_DROPPED))
1826 tp = intotcpcb(inp);
1827 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1829 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1831 so = inp->inp_socket;
1832 SOCKBUF_LOCK(&so->so_snd);
1833 /* If the mss is larger than the socket buffer, decrease the mss. */
1834 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1835 tp->t_maxseg = so->so_snd.sb_hiwat;
1836 SOCKBUF_UNLOCK(&so->so_snd);
1838 TCPSTAT_INC(tcps_mturesent);
1840 tp->snd_nxt = tp->snd_una;
1841 tcp_free_sackholes(tp);
1842 tp->snd_recover = tp->snd_max;
1843 if (tp->t_flags & TF_SACK_PERMIT)
1844 EXIT_FASTRECOVERY(tp->t_flags);
1851 * Look-up the routing entry to the peer of this inpcb. If no route
1852 * is found and it cannot be allocated, then return 0. This routine
1853 * is called by TCP routines that access the rmx structure and by
1854 * tcp_mss_update to get the peer/interface MTU.
1857 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1860 struct sockaddr_in *dst;
1864 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1866 bzero(&sro, sizeof(sro));
1867 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1868 dst = (struct sockaddr_in *)&sro.ro_dst;
1869 dst->sin_family = AF_INET;
1870 dst->sin_len = sizeof(*dst);
1871 dst->sin_addr = inc->inc_faddr;
1872 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1874 if (sro.ro_rt != NULL) {
1875 ifp = sro.ro_rt->rt_ifp;
1876 if (sro.ro_rt->rt_mtu == 0)
1877 maxmtu = ifp->if_mtu;
1879 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1881 /* Report additional interface capabilities. */
1883 if (ifp->if_capenable & IFCAP_TSO4 &&
1884 ifp->if_hwassist & CSUM_TSO) {
1885 cap->ifcap |= CSUM_TSO;
1886 cap->tsomax = ifp->if_hw_tsomax;
1887 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1888 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1899 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1901 struct route_in6 sro6;
1905 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1907 bzero(&sro6, sizeof(sro6));
1908 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1909 sro6.ro_dst.sin6_family = AF_INET6;
1910 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1911 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1912 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1914 if (sro6.ro_rt != NULL) {
1915 ifp = sro6.ro_rt->rt_ifp;
1916 if (sro6.ro_rt->rt_mtu == 0)
1917 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1919 maxmtu = min(sro6.ro_rt->rt_mtu,
1920 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1922 /* Report additional interface capabilities. */
1924 if (ifp->if_capenable & IFCAP_TSO6 &&
1925 ifp->if_hwassist & CSUM_TSO) {
1926 cap->ifcap |= CSUM_TSO;
1927 cap->tsomax = ifp->if_hw_tsomax;
1928 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1929 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1940 /* compute ESP/AH header size for TCP, including outer IP header. */
1942 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1949 struct ip6_hdr *ip6;
1953 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1955 m = m_gethdr(M_NOWAIT, MT_DATA);
1960 if ((inp->inp_vflag & INP_IPV6) != 0) {
1961 ip6 = mtod(m, struct ip6_hdr *);
1962 th = (struct tcphdr *)(ip6 + 1);
1963 m->m_pkthdr.len = m->m_len =
1964 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1965 tcpip_fillheaders(inp, ip6, th);
1966 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1970 ip = mtod(m, struct ip *);
1971 th = (struct tcphdr *)(ip + 1);
1972 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1973 tcpip_fillheaders(inp, ip, th);
1974 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1982 #ifdef TCP_SIGNATURE
1984 * Callback function invoked by m_apply() to digest TCP segment data
1985 * contained within an mbuf chain.
1988 tcp_signature_apply(void *fstate, void *data, u_int len)
1991 MD5Update(fstate, (u_char *)data, len);
1996 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1997 * search with the destination IP address, and a 'magic SPI' to be
1998 * determined by the application. This is hardcoded elsewhere to 1179
2001 tcp_get_sav(struct mbuf *m, u_int direction)
2003 union sockaddr_union dst;
2004 struct secasvar *sav;
2007 struct ip6_hdr *ip6;
2008 char ip6buf[INET6_ADDRSTRLEN];
2011 /* Extract the destination from the IP header in the mbuf. */
2012 bzero(&dst, sizeof(union sockaddr_union));
2013 ip = mtod(m, struct ip *);
2015 ip6 = NULL; /* Make the compiler happy. */
2020 dst.sa.sa_len = sizeof(struct sockaddr_in);
2021 dst.sa.sa_family = AF_INET;
2022 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2023 ip->ip_src : ip->ip_dst;
2027 case (IPV6_VERSION >> 4):
2028 ip6 = mtod(m, struct ip6_hdr *);
2029 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2030 dst.sa.sa_family = AF_INET6;
2031 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2032 ip6->ip6_src : ip6->ip6_dst;
2041 /* Look up an SADB entry which matches the address of the peer. */
2042 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2044 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2045 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2047 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2048 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2057 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2060 * m pointer to head of mbuf chain
2061 * len length of TCP segment data, excluding options
2062 * optlen length of TCP segment options
2063 * buf pointer to storage for computed MD5 digest
2064 * sav pointer to security assosiation
2066 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2067 * When called from tcp_input(), we can be sure that th_sum has been
2068 * zeroed out and verified already.
2070 * Releases reference to SADB key before return.
2072 * Return 0 if successful, otherwise return -1.
2076 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2077 u_char *buf, struct secasvar *sav)
2080 struct ippseudo ippseudo;
2086 struct ipovly *ipovly;
2090 struct ip6_hdr *ip6;
2091 struct in6_addr in6;
2097 KASSERT(m != NULL, ("NULL mbuf chain"));
2098 KASSERT(buf != NULL, ("NULL signature pointer"));
2100 /* Extract the destination from the IP header in the mbuf. */
2101 ip = mtod(m, struct ip *);
2103 ip6 = NULL; /* Make the compiler happy. */
2108 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2110 * XXX The ippseudo header MUST be digested in network byte order,
2111 * or else we'll fail the regression test. Assume all fields we've
2112 * been doing arithmetic on have been in host byte order.
2113 * XXX One cannot depend on ipovly->ih_len here. When called from
2114 * tcp_output(), the underlying ip_len member has not yet been set.
2119 ipovly = (struct ipovly *)ip;
2120 ippseudo.ippseudo_src = ipovly->ih_src;
2121 ippseudo.ippseudo_dst = ipovly->ih_dst;
2122 ippseudo.ippseudo_pad = 0;
2123 ippseudo.ippseudo_p = IPPROTO_TCP;
2124 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2126 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2128 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2129 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2134 * RFC 2385, 2.0 Proposal
2135 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2136 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2137 * extended next header value (to form 32 bits), and 32-bit segment
2139 * Note: Upper-Layer Packet Length comes before Next Header.
2141 case (IPV6_VERSION >> 4):
2143 in6_clearscope(&in6);
2144 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2146 in6_clearscope(&in6);
2147 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2148 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2149 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2151 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2152 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2153 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2155 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2157 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2158 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2169 * Step 2: Update MD5 hash with TCP header, excluding options.
2170 * The TCP checksum must be set to zero.
2172 savecsum = th->th_sum;
2174 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2175 th->th_sum = savecsum;
2178 * Step 3: Update MD5 hash with TCP segment data.
2179 * Use m_apply() to avoid an early m_pullup().
2182 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2185 * Step 4: Update MD5 hash with shared secret.
2187 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2188 MD5Final(buf, &ctx);
2190 key_sa_recordxfer(sav, m);
2196 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2198 * Return 0 if successful, otherwise return -1.
2201 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2202 u_char *buf, u_int direction)
2204 struct secasvar *sav;
2206 if ((sav = tcp_get_sav(m, direction)) == NULL)
2209 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2213 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2216 * m pointer to head of mbuf chain
2217 * len length of TCP segment data, excluding options
2218 * optlen length of TCP segment options
2219 * buf pointer to storage for computed MD5 digest
2220 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2222 * Return 1 if successful, otherwise return 0.
2225 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2226 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2228 char tmpdigest[TCP_SIGLEN];
2230 if (tcp_sig_checksigs == 0)
2232 if ((tcpbflag & TF_SIGNATURE) == 0) {
2233 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2236 * If this socket is not expecting signature but
2237 * the segment contains signature just fail.
2239 TCPSTAT_INC(tcps_sig_err_sigopt);
2240 TCPSTAT_INC(tcps_sig_rcvbadsig);
2244 /* Signature is not expected, and not present in segment. */
2249 * If this socket is expecting signature but the segment does not
2250 * contain any just fail.
2252 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2253 TCPSTAT_INC(tcps_sig_err_nosigopt);
2254 TCPSTAT_INC(tcps_sig_rcvbadsig);
2257 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2258 IPSEC_DIR_INBOUND) == -1) {
2259 TCPSTAT_INC(tcps_sig_err_buildsig);
2260 TCPSTAT_INC(tcps_sig_rcvbadsig);
2264 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2265 TCPSTAT_INC(tcps_sig_rcvbadsig);
2268 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2271 #endif /* TCP_SIGNATURE */
2274 sysctl_drop(SYSCTL_HANDLER_ARGS)
2276 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2277 struct sockaddr_storage addrs[2];
2281 struct sockaddr_in *fin, *lin;
2283 struct sockaddr_in6 *fin6, *lin6;
2294 if (req->oldptr != NULL || req->oldlen != 0)
2296 if (req->newptr == NULL)
2298 if (req->newlen < sizeof(addrs))
2300 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2304 switch (addrs[0].ss_family) {
2307 fin6 = (struct sockaddr_in6 *)&addrs[0];
2308 lin6 = (struct sockaddr_in6 *)&addrs[1];
2309 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2310 lin6->sin6_len != sizeof(struct sockaddr_in6))
2312 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2313 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2315 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2316 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2317 fin = (struct sockaddr_in *)&addrs[0];
2318 lin = (struct sockaddr_in *)&addrs[1];
2321 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2324 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2331 fin = (struct sockaddr_in *)&addrs[0];
2332 lin = (struct sockaddr_in *)&addrs[1];
2333 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2334 lin->sin_len != sizeof(struct sockaddr_in))
2341 INP_INFO_WLOCK(&V_tcbinfo);
2342 switch (addrs[0].ss_family) {
2345 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2346 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2347 INPLOOKUP_WLOCKPCB, NULL);
2352 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2353 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2358 if (inp->inp_flags & INP_TIMEWAIT) {
2360 * XXXRW: There currently exists a state where an
2361 * inpcb is present, but its timewait state has been
2362 * discarded. For now, don't allow dropping of this
2370 } else if (!(inp->inp_flags & INP_DROPPED) &&
2371 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2372 tp = intotcpcb(inp);
2373 tp = tcp_drop(tp, ECONNABORTED);
2380 INP_INFO_WUNLOCK(&V_tcbinfo);
2384 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2385 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2386 0, sysctl_drop, "", "Drop TCP connection");
2389 * Generate a standardized TCP log line for use throughout the
2390 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2391 * allow use in the interrupt context.
2393 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2394 * NB: The function may return NULL if memory allocation failed.
2396 * Due to header inclusion and ordering limitations the struct ip
2397 * and ip6_hdr pointers have to be passed as void pointers.
2400 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2404 /* Is logging enabled? */
2405 if (tcp_log_in_vain == 0)
2408 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2412 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2416 /* Is logging enabled? */
2417 if (tcp_log_debug == 0)
2420 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2424 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2431 const struct ip6_hdr *ip6;
2433 ip6 = (const struct ip6_hdr *)ip6hdr;
2435 ip = (struct ip *)ip4hdr;
2438 * The log line looks like this:
2439 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2441 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2442 sizeof(PRINT_TH_FLAGS) + 1 +
2444 2 * INET6_ADDRSTRLEN;
2446 2 * INET_ADDRSTRLEN;
2449 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2453 strcat(s, "TCP: [");
2456 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2457 inet_ntoa_r(inc->inc_faddr, sp);
2459 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2461 inet_ntoa_r(inc->inc_laddr, sp);
2463 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2466 ip6_sprintf(sp, &inc->inc6_faddr);
2468 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2470 ip6_sprintf(sp, &inc->inc6_laddr);
2472 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2473 } else if (ip6 && th) {
2474 ip6_sprintf(sp, &ip6->ip6_src);
2476 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2478 ip6_sprintf(sp, &ip6->ip6_dst);
2480 sprintf(sp, "]:%i", ntohs(th->th_dport));
2483 } else if (ip && th) {
2484 inet_ntoa_r(ip->ip_src, sp);
2486 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2488 inet_ntoa_r(ip->ip_dst, sp);
2490 sprintf(sp, "]:%i", ntohs(th->th_dport));
2498 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2499 if (*(s + size - 1) != '\0')
2500 panic("%s: string too long", __func__);
2505 * A subroutine which makes it easy to track TCP state changes with DTrace.
2506 * This function shouldn't be called for t_state initializations that don't
2507 * correspond to actual TCP state transitions.
2510 tcp_state_change(struct tcpcb *tp, int newstate)
2512 #if defined(KDTRACE_HOOKS)
2513 int pstate = tp->t_state;
2516 tp->t_state = newstate;
2517 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);