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);
235 * Target size of TCP PCB hash tables. Must be a power of two.
237 * Note that this can be overridden by the kernel environment
238 * variable net.inet.tcp.tcbhashsize
241 #define TCBHASHSIZE 0
246 * Callouts should be moved into struct tcp directly. They are currently
247 * separate because the tcpcb structure is exported to userland for sysctl
248 * parsing purposes, which do not know about callouts.
257 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
258 #define V_tcpcb_zone VNET(tcpcb_zone)
260 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
261 static struct mtx isn_mtx;
263 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
264 #define ISN_LOCK() mtx_lock(&isn_mtx)
265 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
268 * TCP initialization.
271 tcp_zone_change(void *tag)
274 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
275 uma_zone_set_max(V_tcpcb_zone, maxsockets);
276 tcp_tw_zone_change();
280 tcp_inpcb_init(void *mem, int size, int flags)
282 struct inpcb *inp = mem;
284 INP_LOCK_INIT(inp, "inp", "tcpinp");
289 * Take a value and get the next power of 2 that doesn't overflow.
290 * Used to size the tcp_inpcb hash buckets.
293 maketcp_hashsize(int size)
299 * get the next power of 2 higher than maxsockets.
301 hashsize = 1 << fls(size);
302 /* catch overflow, and just go one power of 2 smaller */
303 if (hashsize < size) {
304 hashsize = 1 << (fls(size) - 1);
312 const char *tcbhash_tuneable;
315 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
317 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
318 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 printf("%s: WARNING: unable to register helper hook\n", __func__);
320 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
321 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
322 printf("%s: WARNING: unable to register helper hook\n", __func__);
324 hashsize = TCBHASHSIZE;
325 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
328 * Auto tune the hash size based on maxsockets.
329 * A perfect hash would have a 1:1 mapping
330 * (hashsize = maxsockets) however it's been
331 * suggested that O(2) average is better.
333 hashsize = maketcp_hashsize(maxsockets / 4);
335 * Our historical default is 512,
336 * do not autotune lower than this.
341 printf("%s: %s auto tuned to %d\n", __func__,
342 tcbhash_tuneable, hashsize);
345 * We require a hashsize to be a power of two.
346 * Previously if it was not a power of two we would just reset it
347 * back to 512, which could be a nasty surprise if you did not notice
349 * Instead what we do is clip it to the closest power of two lower
350 * than the specified hash value.
352 if (!powerof2(hashsize)) {
353 int oldhashsize = hashsize;
355 hashsize = maketcp_hashsize(hashsize);
356 /* prevent absurdly low value */
359 printf("%s: WARNING: TCB hash size not a power of 2, "
360 "clipped from %d to %d.\n", __func__, oldhashsize,
363 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
364 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
365 IPI_HASHFIELDS_4TUPLE);
368 * These have to be type stable for the benefit of the timers.
370 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
371 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
372 uma_zone_set_max(V_tcpcb_zone, maxsockets);
373 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
379 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
380 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
381 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
383 /* Skip initialization of globals for non-default instances. */
384 if (!IS_DEFAULT_VNET(curvnet))
387 /* XXX virtualize those bellow? */
388 tcp_delacktime = TCPTV_DELACK;
389 tcp_keepinit = TCPTV_KEEP_INIT;
390 tcp_keepidle = TCPTV_KEEP_IDLE;
391 tcp_keepintvl = TCPTV_KEEPINTVL;
392 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
394 tcp_rexmit_min = TCPTV_MIN;
395 if (tcp_rexmit_min < 1)
397 tcp_rexmit_slop = TCPTV_CPU_VAR;
398 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
399 tcp_tcbhashsize = hashsize;
401 if (tcp_soreceive_stream) {
403 tcp_usrreqs.pru_soreceive = soreceive_stream;
406 tcp6_usrreqs.pru_soreceive = soreceive_stream;
411 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
413 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
415 if (max_protohdr < TCP_MINPROTOHDR)
416 max_protohdr = TCP_MINPROTOHDR;
417 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
419 #undef TCP_MINPROTOHDR
422 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
423 SHUTDOWN_PRI_DEFAULT);
424 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
425 EVENTHANDLER_PRI_ANY);
437 in_pcbinfo_destroy(&V_tcbinfo);
438 uma_zdestroy(V_sack_hole_zone);
439 uma_zdestroy(V_tcpcb_zone);
441 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
443 printf("%s: WARNING: unable to deregister helper hook "
444 "type=%d, id=%d: error %d returned\n", __func__,
445 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
447 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
449 printf("%s: WARNING: unable to deregister helper hook "
450 "type=%d, id=%d: error %d returned\n", __func__,
451 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
463 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
464 * tcp_template used to store this data in mbufs, but we now recopy it out
465 * of the tcpcb each time to conserve mbufs.
468 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
470 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
472 INP_WLOCK_ASSERT(inp);
475 if ((inp->inp_vflag & INP_IPV6) != 0) {
478 ip6 = (struct ip6_hdr *)ip_ptr;
479 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
480 (inp->inp_flow & IPV6_FLOWINFO_MASK);
481 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
482 (IPV6_VERSION & IPV6_VERSION_MASK);
483 ip6->ip6_nxt = IPPROTO_TCP;
484 ip6->ip6_plen = htons(sizeof(struct tcphdr));
485 ip6->ip6_src = inp->in6p_laddr;
486 ip6->ip6_dst = inp->in6p_faddr;
489 #if defined(INET6) && defined(INET)
496 ip = (struct ip *)ip_ptr;
497 ip->ip_v = IPVERSION;
499 ip->ip_tos = inp->inp_ip_tos;
503 ip->ip_ttl = inp->inp_ip_ttl;
505 ip->ip_p = IPPROTO_TCP;
506 ip->ip_src = inp->inp_laddr;
507 ip->ip_dst = inp->inp_faddr;
510 th->th_sport = inp->inp_lport;
511 th->th_dport = inp->inp_fport;
519 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
523 * Create template to be used to send tcp packets on a connection.
524 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
525 * use for this function is in keepalives, which use tcp_respond.
528 tcpip_maketemplate(struct inpcb *inp)
532 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
535 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
540 * Send a single message to the TCP at address specified by
541 * the given TCP/IP header. If m == NULL, then we make a copy
542 * of the tcpiphdr at th and send directly to the addressed host.
543 * This is used to force keep alive messages out using the TCP
544 * template for a connection. If flags are given then we send
545 * a message back to the TCP which originated the segment th,
546 * and discard the mbuf containing it and any other attached mbufs.
548 * In any case the ack and sequence number of the transmitted
549 * segment are as specified by the parameters.
551 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
554 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
555 tcp_seq ack, tcp_seq seq, int flags)
568 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
571 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
578 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
579 INP_WLOCK_ASSERT(inp);
584 if (!(flags & TH_RST)) {
585 win = sbspace(&inp->inp_socket->so_rcv);
586 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
587 win = (long)TCP_MAXWIN << tp->rcv_scale;
591 m = m_gethdr(M_NOWAIT, MT_DATA);
595 m->m_data += max_linkhdr;
598 bcopy((caddr_t)ip6, mtod(m, caddr_t),
599 sizeof(struct ip6_hdr));
600 ip6 = mtod(m, struct ip6_hdr *);
601 nth = (struct tcphdr *)(ip6 + 1);
605 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
606 ip = mtod(m, struct ip *);
607 nth = (struct tcphdr *)(ip + 1);
609 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
614 * XXX MRT We inherrit the FIB, which is lucky.
618 m->m_data = (caddr_t)ipgen;
619 /* m_len is set later */
621 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
624 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
625 nth = (struct tcphdr *)(ip6 + 1);
629 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
630 nth = (struct tcphdr *)(ip + 1);
634 * this is usually a case when an extension header
635 * exists between the IPv6 header and the
638 nth->th_sport = th->th_sport;
639 nth->th_dport = th->th_dport;
641 xchg(nth->th_dport, nth->th_sport, uint16_t);
647 ip6->ip6_vfc = IPV6_VERSION;
648 ip6->ip6_nxt = IPPROTO_TCP;
649 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
650 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
653 #if defined(INET) && defined(INET6)
658 tlen += sizeof (struct tcpiphdr);
659 ip->ip_len = htons(tlen);
660 ip->ip_ttl = V_ip_defttl;
661 if (V_path_mtu_discovery)
662 ip->ip_off |= htons(IP_DF);
666 m->m_pkthdr.len = tlen;
667 m->m_pkthdr.rcvif = NULL;
671 * Packet is associated with a socket, so allow the
672 * label of the response to reflect the socket label.
674 INP_WLOCK_ASSERT(inp);
675 mac_inpcb_create_mbuf(inp, m);
678 * Packet is not associated with a socket, so possibly
679 * update the label in place.
681 mac_netinet_tcp_reply(m);
684 nth->th_seq = htonl(seq);
685 nth->th_ack = htonl(ack);
687 nth->th_off = sizeof (struct tcphdr) >> 2;
688 nth->th_flags = flags;
690 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
692 nth->th_win = htons((u_short)win);
695 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
698 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
699 nth->th_sum = in6_cksum_pseudo(ip6,
700 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
701 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
705 #if defined(INET6) && defined(INET)
710 m->m_pkthdr.csum_flags = CSUM_TCP;
711 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
712 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
716 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
717 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
720 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
723 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
726 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
728 #if defined(INET) && defined(INET6)
732 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
737 * Create a new TCP control block, making an
738 * empty reassembly queue and hooking it to the argument
739 * protocol control block. The `inp' parameter must have
740 * come from the zone allocator set up in tcp_init().
743 tcp_newtcpcb(struct inpcb *inp)
745 struct tcpcb_mem *tm;
748 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
751 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
756 /* Initialise cc_var struct for this tcpcb. */
758 tp->ccv->type = IPPROTO_TCP;
759 tp->ccv->ccvc.tcp = tp;
762 * Use the current system default CC algorithm.
765 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
766 CC_ALGO(tp) = CC_DEFAULT();
769 if (CC_ALGO(tp)->cb_init != NULL)
770 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
771 uma_zfree(V_tcpcb_zone, tm);
776 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
777 uma_zfree(V_tcpcb_zone, tm);
782 tp->t_vnet = inp->inp_vnet;
784 tp->t_timers = &tm->tt;
785 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
786 tp->t_maxseg = tp->t_maxopd =
788 isipv6 ? V_tcp_v6mssdflt :
792 /* Set up our timeouts. */
793 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
794 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
795 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
796 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
797 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
799 if (V_tcp_do_rfc1323)
800 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
802 tp->t_flags |= TF_SACK_PERMIT;
803 TAILQ_INIT(&tp->snd_holes);
804 tp->t_inpcb = inp; /* XXX */
806 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
807 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
808 * reasonable initial retransmit time.
810 tp->t_srtt = TCPTV_SRTTBASE;
811 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
812 tp->t_rttmin = tcp_rexmit_min;
813 tp->t_rxtcur = TCPTV_RTOBASE;
814 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
815 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
816 tp->t_rcvtime = ticks;
818 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
819 * because the socket may be bound to an IPv6 wildcard address,
820 * which may match an IPv4-mapped IPv6 address.
822 inp->inp_ip_ttl = V_ip_defttl;
824 return (tp); /* XXX */
828 * Switch the congestion control algorithm back to NewReno for any active
829 * control blocks using an algorithm which is about to go away.
830 * This ensures the CC framework can allow the unload to proceed without leaving
831 * any dangling pointers which would trigger a panic.
832 * Returning non-zero would inform the CC framework that something went wrong
833 * and it would be unsafe to allow the unload to proceed. However, there is no
834 * way for this to occur with this implementation so we always return zero.
837 tcp_ccalgounload(struct cc_algo *unload_algo)
839 struct cc_algo *tmpalgo;
842 VNET_ITERATOR_DECL(vnet_iter);
845 * Check all active control blocks across all network stacks and change
846 * any that are using "unload_algo" back to NewReno. If "unload_algo"
847 * requires cleanup code to be run, call it.
850 VNET_FOREACH(vnet_iter) {
851 CURVNET_SET(vnet_iter);
852 INP_INFO_RLOCK(&V_tcbinfo);
854 * New connections already part way through being initialised
855 * with the CC algo we're removing will not race with this code
856 * because the INP_INFO_WLOCK is held during initialisation. We
857 * therefore don't enter the loop below until the connection
858 * list has stabilised.
860 LIST_FOREACH(inp, &V_tcb, inp_list) {
862 /* Important to skip tcptw structs. */
863 if (!(inp->inp_flags & INP_TIMEWAIT) &&
864 (tp = intotcpcb(inp)) != NULL) {
866 * By holding INP_WLOCK here, we are assured
867 * that the connection is not currently
868 * executing inside the CC module's functions
869 * i.e. it is safe to make the switch back to
872 if (CC_ALGO(tp) == unload_algo) {
873 tmpalgo = CC_ALGO(tp);
874 /* NewReno does not require any init. */
875 CC_ALGO(tp) = &newreno_cc_algo;
876 if (tmpalgo->cb_destroy != NULL)
877 tmpalgo->cb_destroy(tp->ccv);
882 INP_INFO_RUNLOCK(&V_tcbinfo);
891 * Drop a TCP connection, reporting
892 * the specified error. If connection is synchronized,
893 * then send a RST to peer.
896 tcp_drop(struct tcpcb *tp, int errno)
898 struct socket *so = tp->t_inpcb->inp_socket;
900 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
901 INP_WLOCK_ASSERT(tp->t_inpcb);
903 if (TCPS_HAVERCVDSYN(tp->t_state)) {
904 tcp_state_change(tp, TCPS_CLOSED);
905 (void) tcp_output(tp);
906 TCPSTAT_INC(tcps_drops);
908 TCPSTAT_INC(tcps_conndrops);
909 if (errno == ETIMEDOUT && tp->t_softerror)
910 errno = tp->t_softerror;
911 so->so_error = errno;
912 return (tcp_close(tp));
916 tcp_discardcb(struct tcpcb *tp)
918 struct inpcb *inp = tp->t_inpcb;
919 struct socket *so = inp->inp_socket;
921 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
924 INP_WLOCK_ASSERT(inp);
927 * Make sure that all of our timers are stopped before we delete the
930 * XXXRW: Really, we would like to use callout_drain() here in order
931 * to avoid races experienced in tcp_timer.c where a timer is already
932 * executing at this point. However, we can't, both because we're
933 * running in a context where we can't sleep, and also because we
934 * hold locks required by the timers. What we instead need to do is
935 * test to see if callout_drain() is required, and if so, defer some
936 * portion of the remainder of tcp_discardcb() to an asynchronous
937 * context that can callout_drain() and then continue. Some care
938 * will be required to ensure that no further processing takes place
939 * on the tcpcb, even though it hasn't been freed (a flag?).
941 callout_stop(&tp->t_timers->tt_rexmt);
942 callout_stop(&tp->t_timers->tt_persist);
943 callout_stop(&tp->t_timers->tt_keep);
944 callout_stop(&tp->t_timers->tt_2msl);
945 callout_stop(&tp->t_timers->tt_delack);
948 * If we got enough samples through the srtt filter,
949 * save the rtt and rttvar in the routing entry.
950 * 'Enough' is arbitrarily defined as 4 rtt samples.
951 * 4 samples is enough for the srtt filter to converge
952 * to within enough % of the correct value; fewer samples
953 * and we could save a bogus rtt. The danger is not high
954 * as tcp quickly recovers from everything.
955 * XXX: Works very well but needs some more statistics!
957 if (tp->t_rttupdated >= 4) {
958 struct hc_metrics_lite metrics;
961 bzero(&metrics, sizeof(metrics));
963 * Update the ssthresh always when the conditions below
964 * are satisfied. This gives us better new start value
965 * for the congestion avoidance for new connections.
966 * ssthresh is only set if packet loss occured on a session.
968 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
969 * being torn down. Ideally this code would not use 'so'.
971 ssthresh = tp->snd_ssthresh;
972 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
974 * convert the limit from user data bytes to
975 * packets then to packet data bytes.
977 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
980 ssthresh *= (u_long)(tp->t_maxseg +
982 (isipv6 ? sizeof (struct ip6_hdr) +
983 sizeof (struct tcphdr) :
985 sizeof (struct tcpiphdr)
992 metrics.rmx_ssthresh = ssthresh;
994 metrics.rmx_rtt = tp->t_srtt;
995 metrics.rmx_rttvar = tp->t_rttvar;
996 metrics.rmx_cwnd = tp->snd_cwnd;
997 metrics.rmx_sendpipe = 0;
998 metrics.rmx_recvpipe = 0;
1000 tcp_hc_update(&inp->inp_inc, &metrics);
1003 /* free the reassembly queue, if any */
1004 tcp_reass_flush(tp);
1007 /* Disconnect offload device, if any. */
1008 if (tp->t_flags & TF_TOE)
1009 tcp_offload_detach(tp);
1012 tcp_free_sackholes(tp);
1014 /* Allow the CC algorithm to clean up after itself. */
1015 if (CC_ALGO(tp)->cb_destroy != NULL)
1016 CC_ALGO(tp)->cb_destroy(tp->ccv);
1018 khelp_destroy_osd(tp->osd);
1021 inp->inp_ppcb = NULL;
1023 uma_zfree(V_tcpcb_zone, tp);
1027 * Attempt to close a TCP control block, marking it as dropped, and freeing
1028 * the socket if we hold the only reference.
1031 tcp_close(struct tcpcb *tp)
1033 struct inpcb *inp = tp->t_inpcb;
1036 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1037 INP_WLOCK_ASSERT(inp);
1040 if (tp->t_state == TCPS_LISTEN)
1041 tcp_offload_listen_stop(tp);
1044 TCPSTAT_INC(tcps_closed);
1045 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1046 so = inp->inp_socket;
1047 soisdisconnected(so);
1048 if (inp->inp_flags & INP_SOCKREF) {
1049 KASSERT(so->so_state & SS_PROTOREF,
1050 ("tcp_close: !SS_PROTOREF"));
1051 inp->inp_flags &= ~INP_SOCKREF;
1055 so->so_state &= ~SS_PROTOREF;
1065 VNET_ITERATOR_DECL(vnet_iter);
1070 VNET_LIST_RLOCK_NOSLEEP();
1071 VNET_FOREACH(vnet_iter) {
1072 CURVNET_SET(vnet_iter);
1077 * Walk the tcpbs, if existing, and flush the reassembly queue,
1078 * if there is one...
1079 * XXX: The "Net/3" implementation doesn't imply that the TCP
1080 * reassembly queue should be flushed, but in a situation
1081 * where we're really low on mbufs, this is potentially
1084 INP_INFO_RLOCK(&V_tcbinfo);
1085 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1086 if (inpb->inp_flags & INP_TIMEWAIT)
1089 if ((tcpb = intotcpcb(inpb)) != NULL) {
1090 tcp_reass_flush(tcpb);
1091 tcp_clean_sackreport(tcpb);
1095 INP_INFO_RUNLOCK(&V_tcbinfo);
1098 VNET_LIST_RUNLOCK_NOSLEEP();
1102 * Notify a tcp user of an asynchronous error;
1103 * store error as soft error, but wake up user
1104 * (for now, won't do anything until can select for soft error).
1106 * Do not wake up user since there currently is no mechanism for
1107 * reporting soft errors (yet - a kqueue filter may be added).
1109 static struct inpcb *
1110 tcp_notify(struct inpcb *inp, int error)
1114 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1115 INP_WLOCK_ASSERT(inp);
1117 if ((inp->inp_flags & INP_TIMEWAIT) ||
1118 (inp->inp_flags & INP_DROPPED))
1121 tp = intotcpcb(inp);
1122 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1125 * Ignore some errors if we are hooked up.
1126 * If connection hasn't completed, has retransmitted several times,
1127 * and receives a second error, give up now. This is better
1128 * than waiting a long time to establish a connection that
1129 * can never complete.
1131 if (tp->t_state == TCPS_ESTABLISHED &&
1132 (error == EHOSTUNREACH || error == ENETUNREACH ||
1133 error == EHOSTDOWN)) {
1135 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1137 tp = tcp_drop(tp, error);
1143 tp->t_softerror = error;
1147 wakeup( &so->so_timeo);
1154 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1156 int error, i, m, n, pcb_count;
1157 struct inpcb *inp, **inp_list;
1162 * The process of preparing the TCB list is too time-consuming and
1163 * resource-intensive to repeat twice on every request.
1165 if (req->oldptr == NULL) {
1166 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1167 n += imax(n / 8, 10);
1168 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1172 if (req->newptr != NULL)
1176 * OK, now we're committed to doing something.
1178 INP_INFO_RLOCK(&V_tcbinfo);
1179 gencnt = V_tcbinfo.ipi_gencnt;
1180 n = V_tcbinfo.ipi_count;
1181 INP_INFO_RUNLOCK(&V_tcbinfo);
1183 m = syncache_pcbcount();
1185 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1186 + (n + m) * sizeof(struct xtcpcb));
1190 xig.xig_len = sizeof xig;
1191 xig.xig_count = n + m;
1192 xig.xig_gen = gencnt;
1193 xig.xig_sogen = so_gencnt;
1194 error = SYSCTL_OUT(req, &xig, sizeof xig);
1198 error = syncache_pcblist(req, m, &pcb_count);
1202 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1203 if (inp_list == NULL)
1206 INP_INFO_RLOCK(&V_tcbinfo);
1207 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1208 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1210 if (inp->inp_gencnt <= gencnt) {
1212 * XXX: This use of cr_cansee(), introduced with
1213 * TCP state changes, is not quite right, but for
1214 * now, better than nothing.
1216 if (inp->inp_flags & INP_TIMEWAIT) {
1217 if (intotw(inp) != NULL)
1218 error = cr_cansee(req->td->td_ucred,
1219 intotw(inp)->tw_cred);
1221 error = EINVAL; /* Skip this inp. */
1223 error = cr_canseeinpcb(req->td->td_ucred, inp);
1226 inp_list[i++] = inp;
1231 INP_INFO_RUNLOCK(&V_tcbinfo);
1235 for (i = 0; i < n; i++) {
1238 if (inp->inp_gencnt <= gencnt) {
1242 bzero(&xt, sizeof(xt));
1243 xt.xt_len = sizeof xt;
1244 /* XXX should avoid extra copy */
1245 bcopy(inp, &xt.xt_inp, sizeof *inp);
1246 inp_ppcb = inp->inp_ppcb;
1247 if (inp_ppcb == NULL)
1248 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1249 else if (inp->inp_flags & INP_TIMEWAIT) {
1250 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1251 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1253 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1254 if (xt.xt_tp.t_timers)
1255 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1257 if (inp->inp_socket != NULL)
1258 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1260 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1261 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1263 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1265 error = SYSCTL_OUT(req, &xt, sizeof xt);
1269 INP_INFO_WLOCK(&V_tcbinfo);
1270 for (i = 0; i < n; i++) {
1273 if (!in_pcbrele_rlocked(inp))
1276 INP_INFO_WUNLOCK(&V_tcbinfo);
1280 * Give the user an updated idea of our state.
1281 * If the generation differs from what we told
1282 * her before, she knows that something happened
1283 * while we were processing this request, and it
1284 * might be necessary to retry.
1286 INP_INFO_RLOCK(&V_tcbinfo);
1287 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1288 xig.xig_sogen = so_gencnt;
1289 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1290 INP_INFO_RUNLOCK(&V_tcbinfo);
1291 error = SYSCTL_OUT(req, &xig, sizeof xig);
1293 free(inp_list, M_TEMP);
1297 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1298 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1299 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1303 tcp_getcred(SYSCTL_HANDLER_ARGS)
1306 struct sockaddr_in addrs[2];
1310 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1313 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1316 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1317 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1319 if (inp->inp_socket == NULL)
1322 error = cr_canseeinpcb(req->td->td_ucred, inp);
1324 cru2x(inp->inp_cred, &xuc);
1329 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1333 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1334 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1335 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1340 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1343 struct sockaddr_in6 addrs[2];
1350 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1353 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1356 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1357 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1360 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1362 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1371 inp = in_pcblookup(&V_tcbinfo,
1372 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1374 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1375 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1378 inp = in6_pcblookup(&V_tcbinfo,
1379 &addrs[1].sin6_addr, addrs[1].sin6_port,
1380 &addrs[0].sin6_addr, addrs[0].sin6_port,
1381 INPLOOKUP_RLOCKPCB, NULL);
1383 if (inp->inp_socket == NULL)
1386 error = cr_canseeinpcb(req->td->td_ucred, inp);
1388 cru2x(inp->inp_cred, &xuc);
1393 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1397 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1398 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1399 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1405 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1407 struct ip *ip = vip;
1409 struct in_addr faddr;
1412 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1414 struct in_conninfo inc;
1415 tcp_seq icmp_tcp_seq;
1418 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1419 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1422 if (cmd == PRC_MSGSIZE)
1423 notify = tcp_mtudisc_notify;
1424 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1425 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1426 notify = tcp_drop_syn_sent;
1428 * Redirects don't need to be handled up here.
1430 else if (PRC_IS_REDIRECT(cmd))
1433 * Hostdead is ugly because it goes linearly through all PCBs.
1434 * XXX: We never get this from ICMP, otherwise it makes an
1435 * excellent DoS attack on machines with many connections.
1437 else if (cmd == PRC_HOSTDEAD)
1439 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1442 icp = (struct icmp *)((caddr_t)ip
1443 - offsetof(struct icmp, icmp_ip));
1444 th = (struct tcphdr *)((caddr_t)ip
1445 + (ip->ip_hl << 2));
1446 INP_INFO_WLOCK(&V_tcbinfo);
1447 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1448 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1450 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1451 !(inp->inp_flags & INP_DROPPED) &&
1452 !(inp->inp_socket == NULL)) {
1453 icmp_tcp_seq = htonl(th->th_seq);
1454 tp = intotcpcb(inp);
1455 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1456 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1457 if (cmd == PRC_MSGSIZE) {
1460 * If we got a needfrag set the MTU
1461 * in the route to the suggested new
1462 * value (if given) and then notify.
1464 bzero(&inc, sizeof(inc));
1465 inc.inc_faddr = faddr;
1467 inp->inp_inc.inc_fibnum;
1469 mtu = ntohs(icp->icmp_nextmtu);
1471 * If no alternative MTU was
1472 * proposed, try the next smaller
1477 ntohs(ip->ip_len), 1);
1478 if (mtu < V_tcp_minmss
1479 + sizeof(struct tcpiphdr))
1481 + sizeof(struct tcpiphdr);
1483 * Only cache the MTU if it
1484 * is smaller than the interface
1485 * or route MTU. tcp_mtudisc()
1486 * will do right thing by itself.
1488 if (mtu <= tcp_maxmtu(&inc, NULL))
1489 tcp_hc_updatemtu(&inc, mtu);
1490 tcp_mtudisc(inp, mtu);
1492 inp = (*notify)(inp,
1493 inetctlerrmap[cmd]);
1499 bzero(&inc, sizeof(inc));
1500 inc.inc_fport = th->th_dport;
1501 inc.inc_lport = th->th_sport;
1502 inc.inc_faddr = faddr;
1503 inc.inc_laddr = ip->ip_src;
1504 syncache_unreach(&inc, th);
1506 INP_INFO_WUNLOCK(&V_tcbinfo);
1508 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1514 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1517 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1518 struct ip6_hdr *ip6;
1520 struct ip6ctlparam *ip6cp = NULL;
1521 const struct sockaddr_in6 *sa6_src = NULL;
1523 struct tcp_portonly {
1528 if (sa->sa_family != AF_INET6 ||
1529 sa->sa_len != sizeof(struct sockaddr_in6))
1532 if (cmd == PRC_MSGSIZE)
1533 notify = tcp_mtudisc_notify;
1534 else if (!PRC_IS_REDIRECT(cmd) &&
1535 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1538 /* if the parameter is from icmp6, decode it. */
1540 ip6cp = (struct ip6ctlparam *)d;
1542 ip6 = ip6cp->ip6c_ip6;
1543 off = ip6cp->ip6c_off;
1544 sa6_src = ip6cp->ip6c_src;
1548 off = 0; /* fool gcc */
1553 struct in_conninfo inc;
1555 * XXX: We assume that when IPV6 is non NULL,
1556 * M and OFF are valid.
1559 /* check if we can safely examine src and dst ports */
1560 if (m->m_pkthdr.len < off + sizeof(*thp))
1563 bzero(&th, sizeof(th));
1564 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1566 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1567 (struct sockaddr *)ip6cp->ip6c_src,
1568 th.th_sport, cmd, NULL, notify);
1570 bzero(&inc, sizeof(inc));
1571 inc.inc_fport = th.th_dport;
1572 inc.inc_lport = th.th_sport;
1573 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1574 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1575 inc.inc_flags |= INC_ISIPV6;
1576 INP_INFO_WLOCK(&V_tcbinfo);
1577 syncache_unreach(&inc, &th);
1578 INP_INFO_WUNLOCK(&V_tcbinfo);
1580 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1581 0, cmd, NULL, notify);
1587 * Following is where TCP initial sequence number generation occurs.
1589 * There are two places where we must use initial sequence numbers:
1590 * 1. In SYN-ACK packets.
1591 * 2. In SYN packets.
1593 * All ISNs for SYN-ACK packets are generated by the syncache. See
1594 * tcp_syncache.c for details.
1596 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1597 * depends on this property. In addition, these ISNs should be
1598 * unguessable so as to prevent connection hijacking. To satisfy
1599 * the requirements of this situation, the algorithm outlined in
1600 * RFC 1948 is used, with only small modifications.
1602 * Implementation details:
1604 * Time is based off the system timer, and is corrected so that it
1605 * increases by one megabyte per second. This allows for proper
1606 * recycling on high speed LANs while still leaving over an hour
1609 * As reading the *exact* system time is too expensive to be done
1610 * whenever setting up a TCP connection, we increment the time
1611 * offset in two ways. First, a small random positive increment
1612 * is added to isn_offset for each connection that is set up.
1613 * Second, the function tcp_isn_tick fires once per clock tick
1614 * and increments isn_offset as necessary so that sequence numbers
1615 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1616 * random positive increments serve only to ensure that the same
1617 * exact sequence number is never sent out twice (as could otherwise
1618 * happen when a port is recycled in less than the system tick
1621 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1622 * between seeding of isn_secret. This is normally set to zero,
1623 * as reseeding should not be necessary.
1625 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1626 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1627 * general, this means holding an exclusive (write) lock.
1630 #define ISN_BYTES_PER_SECOND 1048576
1631 #define ISN_STATIC_INCREMENT 4096
1632 #define ISN_RANDOM_INCREMENT (4096 - 1)
1634 static VNET_DEFINE(u_char, isn_secret[32]);
1635 static VNET_DEFINE(int, isn_last);
1636 static VNET_DEFINE(int, isn_last_reseed);
1637 static VNET_DEFINE(u_int32_t, isn_offset);
1638 static VNET_DEFINE(u_int32_t, isn_offset_old);
1640 #define V_isn_secret VNET(isn_secret)
1641 #define V_isn_last VNET(isn_last)
1642 #define V_isn_last_reseed VNET(isn_last_reseed)
1643 #define V_isn_offset VNET(isn_offset)
1644 #define V_isn_offset_old VNET(isn_offset_old)
1647 tcp_new_isn(struct tcpcb *tp)
1650 u_int32_t md5_buffer[4];
1652 u_int32_t projected_offset;
1654 INP_WLOCK_ASSERT(tp->t_inpcb);
1657 /* Seed if this is the first use, reseed if requested. */
1658 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1659 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1661 read_random(&V_isn_secret, sizeof(V_isn_secret));
1662 V_isn_last_reseed = ticks;
1665 /* Compute the md5 hash and return the ISN. */
1667 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1668 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1670 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1671 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1672 sizeof(struct in6_addr));
1673 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1674 sizeof(struct in6_addr));
1678 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1679 sizeof(struct in_addr));
1680 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1681 sizeof(struct in_addr));
1683 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1684 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1685 new_isn = (tcp_seq) md5_buffer[0];
1686 V_isn_offset += ISN_STATIC_INCREMENT +
1687 (arc4random() & ISN_RANDOM_INCREMENT);
1688 if (ticks != V_isn_last) {
1689 projected_offset = V_isn_offset_old +
1690 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1691 if (SEQ_GT(projected_offset, V_isn_offset))
1692 V_isn_offset = projected_offset;
1693 V_isn_offset_old = V_isn_offset;
1696 new_isn += V_isn_offset;
1702 * When a specific ICMP unreachable message is received and the
1703 * connection state is SYN-SENT, drop the connection. This behavior
1704 * is controlled by the icmp_may_rst sysctl.
1707 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1711 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1712 INP_WLOCK_ASSERT(inp);
1714 if ((inp->inp_flags & INP_TIMEWAIT) ||
1715 (inp->inp_flags & INP_DROPPED))
1718 tp = intotcpcb(inp);
1719 if (tp->t_state != TCPS_SYN_SENT)
1722 tp = tcp_drop(tp, errno);
1730 * When `need fragmentation' ICMP is received, update our idea of the MSS
1731 * based on the new value. Also nudge TCP to send something, since we
1732 * know the packet we just sent was dropped.
1733 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1735 static struct inpcb *
1736 tcp_mtudisc_notify(struct inpcb *inp, int error)
1739 return (tcp_mtudisc(inp, -1));
1743 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1748 INP_WLOCK_ASSERT(inp);
1749 if ((inp->inp_flags & INP_TIMEWAIT) ||
1750 (inp->inp_flags & INP_DROPPED))
1753 tp = intotcpcb(inp);
1754 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1756 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1758 so = inp->inp_socket;
1759 SOCKBUF_LOCK(&so->so_snd);
1760 /* If the mss is larger than the socket buffer, decrease the mss. */
1761 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1762 tp->t_maxseg = so->so_snd.sb_hiwat;
1763 SOCKBUF_UNLOCK(&so->so_snd);
1765 TCPSTAT_INC(tcps_mturesent);
1767 tp->snd_nxt = tp->snd_una;
1768 tcp_free_sackholes(tp);
1769 tp->snd_recover = tp->snd_max;
1770 if (tp->t_flags & TF_SACK_PERMIT)
1771 EXIT_FASTRECOVERY(tp->t_flags);
1778 * Look-up the routing entry to the peer of this inpcb. If no route
1779 * is found and it cannot be allocated, then return 0. This routine
1780 * is called by TCP routines that access the rmx structure and by
1781 * tcp_mss_update to get the peer/interface MTU.
1784 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1787 struct sockaddr_in *dst;
1791 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1793 bzero(&sro, sizeof(sro));
1794 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1795 dst = (struct sockaddr_in *)&sro.ro_dst;
1796 dst->sin_family = AF_INET;
1797 dst->sin_len = sizeof(*dst);
1798 dst->sin_addr = inc->inc_faddr;
1799 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1801 if (sro.ro_rt != NULL) {
1802 ifp = sro.ro_rt->rt_ifp;
1803 if (sro.ro_rt->rt_mtu == 0)
1804 maxmtu = ifp->if_mtu;
1806 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1808 /* Report additional interface capabilities. */
1810 if (ifp->if_capenable & IFCAP_TSO4 &&
1811 ifp->if_hwassist & CSUM_TSO) {
1812 cap->ifcap |= CSUM_TSO;
1813 cap->tsomax = ifp->if_hw_tsomax;
1814 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1815 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1826 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1828 struct route_in6 sro6;
1832 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1834 bzero(&sro6, sizeof(sro6));
1835 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1836 sro6.ro_dst.sin6_family = AF_INET6;
1837 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1838 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1839 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1841 if (sro6.ro_rt != NULL) {
1842 ifp = sro6.ro_rt->rt_ifp;
1843 if (sro6.ro_rt->rt_mtu == 0)
1844 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1846 maxmtu = min(sro6.ro_rt->rt_mtu,
1847 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1849 /* Report additional interface capabilities. */
1851 if (ifp->if_capenable & IFCAP_TSO6 &&
1852 ifp->if_hwassist & CSUM_TSO) {
1853 cap->ifcap |= CSUM_TSO;
1854 cap->tsomax = ifp->if_hw_tsomax;
1855 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1856 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1867 /* compute ESP/AH header size for TCP, including outer IP header. */
1869 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1876 struct ip6_hdr *ip6;
1880 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1882 m = m_gethdr(M_NOWAIT, MT_DATA);
1887 if ((inp->inp_vflag & INP_IPV6) != 0) {
1888 ip6 = mtod(m, struct ip6_hdr *);
1889 th = (struct tcphdr *)(ip6 + 1);
1890 m->m_pkthdr.len = m->m_len =
1891 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1892 tcpip_fillheaders(inp, ip6, th);
1893 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1897 ip = mtod(m, struct ip *);
1898 th = (struct tcphdr *)(ip + 1);
1899 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1900 tcpip_fillheaders(inp, ip, th);
1901 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1909 #ifdef TCP_SIGNATURE
1911 * Callback function invoked by m_apply() to digest TCP segment data
1912 * contained within an mbuf chain.
1915 tcp_signature_apply(void *fstate, void *data, u_int len)
1918 MD5Update(fstate, (u_char *)data, len);
1923 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1924 * search with the destination IP address, and a 'magic SPI' to be
1925 * determined by the application. This is hardcoded elsewhere to 1179
1928 tcp_get_sav(struct mbuf *m, u_int direction)
1930 union sockaddr_union dst;
1931 struct secasvar *sav;
1934 struct ip6_hdr *ip6;
1935 char ip6buf[INET6_ADDRSTRLEN];
1938 /* Extract the destination from the IP header in the mbuf. */
1939 bzero(&dst, sizeof(union sockaddr_union));
1940 ip = mtod(m, struct ip *);
1942 ip6 = NULL; /* Make the compiler happy. */
1947 dst.sa.sa_len = sizeof(struct sockaddr_in);
1948 dst.sa.sa_family = AF_INET;
1949 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1950 ip->ip_src : ip->ip_dst;
1954 case (IPV6_VERSION >> 4):
1955 ip6 = mtod(m, struct ip6_hdr *);
1956 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1957 dst.sa.sa_family = AF_INET6;
1958 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1959 ip6->ip6_src : ip6->ip6_dst;
1968 /* Look up an SADB entry which matches the address of the peer. */
1969 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1971 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1972 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1974 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1975 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1984 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1987 * m pointer to head of mbuf chain
1988 * len length of TCP segment data, excluding options
1989 * optlen length of TCP segment options
1990 * buf pointer to storage for computed MD5 digest
1991 * sav pointer to security assosiation
1993 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1994 * When called from tcp_input(), we can be sure that th_sum has been
1995 * zeroed out and verified already.
1997 * Releases reference to SADB key before return.
1999 * Return 0 if successful, otherwise return -1.
2003 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2004 u_char *buf, struct secasvar *sav)
2007 struct ippseudo ippseudo;
2013 struct ipovly *ipovly;
2017 struct ip6_hdr *ip6;
2018 struct in6_addr in6;
2024 KASSERT(m != NULL, ("NULL mbuf chain"));
2025 KASSERT(buf != NULL, ("NULL signature pointer"));
2027 /* Extract the destination from the IP header in the mbuf. */
2028 ip = mtod(m, struct ip *);
2030 ip6 = NULL; /* Make the compiler happy. */
2035 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2037 * XXX The ippseudo header MUST be digested in network byte order,
2038 * or else we'll fail the regression test. Assume all fields we've
2039 * been doing arithmetic on have been in host byte order.
2040 * XXX One cannot depend on ipovly->ih_len here. When called from
2041 * tcp_output(), the underlying ip_len member has not yet been set.
2046 ipovly = (struct ipovly *)ip;
2047 ippseudo.ippseudo_src = ipovly->ih_src;
2048 ippseudo.ippseudo_dst = ipovly->ih_dst;
2049 ippseudo.ippseudo_pad = 0;
2050 ippseudo.ippseudo_p = IPPROTO_TCP;
2051 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2053 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2055 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2056 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2061 * RFC 2385, 2.0 Proposal
2062 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2063 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2064 * extended next header value (to form 32 bits), and 32-bit segment
2066 * Note: Upper-Layer Packet Length comes before Next Header.
2068 case (IPV6_VERSION >> 4):
2070 in6_clearscope(&in6);
2071 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2073 in6_clearscope(&in6);
2074 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2075 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2076 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2078 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2079 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2080 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2082 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2084 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2085 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2096 * Step 2: Update MD5 hash with TCP header, excluding options.
2097 * The TCP checksum must be set to zero.
2099 savecsum = th->th_sum;
2101 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2102 th->th_sum = savecsum;
2105 * Step 3: Update MD5 hash with TCP segment data.
2106 * Use m_apply() to avoid an early m_pullup().
2109 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2112 * Step 4: Update MD5 hash with shared secret.
2114 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2115 MD5Final(buf, &ctx);
2117 key_sa_recordxfer(sav, m);
2123 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2125 * Return 0 if successful, otherwise return -1.
2128 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2129 u_char *buf, u_int direction)
2131 struct secasvar *sav;
2133 if ((sav = tcp_get_sav(m, direction)) == NULL)
2136 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2140 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2143 * m pointer to head of mbuf chain
2144 * len length of TCP segment data, excluding options
2145 * optlen length of TCP segment options
2146 * buf pointer to storage for computed MD5 digest
2147 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2149 * Return 1 if successful, otherwise return 0.
2152 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2153 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2155 char tmpdigest[TCP_SIGLEN];
2157 if (tcp_sig_checksigs == 0)
2159 if ((tcpbflag & TF_SIGNATURE) == 0) {
2160 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2163 * If this socket is not expecting signature but
2164 * the segment contains signature just fail.
2166 TCPSTAT_INC(tcps_sig_err_sigopt);
2167 TCPSTAT_INC(tcps_sig_rcvbadsig);
2171 /* Signature is not expected, and not present in segment. */
2176 * If this socket is expecting signature but the segment does not
2177 * contain any just fail.
2179 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2180 TCPSTAT_INC(tcps_sig_err_nosigopt);
2181 TCPSTAT_INC(tcps_sig_rcvbadsig);
2184 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2185 IPSEC_DIR_INBOUND) == -1) {
2186 TCPSTAT_INC(tcps_sig_err_buildsig);
2187 TCPSTAT_INC(tcps_sig_rcvbadsig);
2191 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2192 TCPSTAT_INC(tcps_sig_rcvbadsig);
2195 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2198 #endif /* TCP_SIGNATURE */
2201 sysctl_drop(SYSCTL_HANDLER_ARGS)
2203 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2204 struct sockaddr_storage addrs[2];
2208 struct sockaddr_in *fin, *lin;
2210 struct sockaddr_in6 *fin6, *lin6;
2221 if (req->oldptr != NULL || req->oldlen != 0)
2223 if (req->newptr == NULL)
2225 if (req->newlen < sizeof(addrs))
2227 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2231 switch (addrs[0].ss_family) {
2234 fin6 = (struct sockaddr_in6 *)&addrs[0];
2235 lin6 = (struct sockaddr_in6 *)&addrs[1];
2236 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2237 lin6->sin6_len != sizeof(struct sockaddr_in6))
2239 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2240 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2242 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2243 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2244 fin = (struct sockaddr_in *)&addrs[0];
2245 lin = (struct sockaddr_in *)&addrs[1];
2248 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2251 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2258 fin = (struct sockaddr_in *)&addrs[0];
2259 lin = (struct sockaddr_in *)&addrs[1];
2260 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2261 lin->sin_len != sizeof(struct sockaddr_in))
2268 INP_INFO_WLOCK(&V_tcbinfo);
2269 switch (addrs[0].ss_family) {
2272 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2273 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2274 INPLOOKUP_WLOCKPCB, NULL);
2279 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2280 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2285 if (inp->inp_flags & INP_TIMEWAIT) {
2287 * XXXRW: There currently exists a state where an
2288 * inpcb is present, but its timewait state has been
2289 * discarded. For now, don't allow dropping of this
2297 } else if (!(inp->inp_flags & INP_DROPPED) &&
2298 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2299 tp = intotcpcb(inp);
2300 tp = tcp_drop(tp, ECONNABORTED);
2307 INP_INFO_WUNLOCK(&V_tcbinfo);
2311 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2312 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2313 0, sysctl_drop, "", "Drop TCP connection");
2316 * Generate a standardized TCP log line for use throughout the
2317 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2318 * allow use in the interrupt context.
2320 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2321 * NB: The function may return NULL if memory allocation failed.
2323 * Due to header inclusion and ordering limitations the struct ip
2324 * and ip6_hdr pointers have to be passed as void pointers.
2327 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2331 /* Is logging enabled? */
2332 if (tcp_log_in_vain == 0)
2335 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2339 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2343 /* Is logging enabled? */
2344 if (tcp_log_debug == 0)
2347 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2351 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2358 const struct ip6_hdr *ip6;
2360 ip6 = (const struct ip6_hdr *)ip6hdr;
2362 ip = (struct ip *)ip4hdr;
2365 * The log line looks like this:
2366 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2368 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2369 sizeof(PRINT_TH_FLAGS) + 1 +
2371 2 * INET6_ADDRSTRLEN;
2373 2 * INET_ADDRSTRLEN;
2376 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2380 strcat(s, "TCP: [");
2383 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2384 inet_ntoa_r(inc->inc_faddr, sp);
2386 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2388 inet_ntoa_r(inc->inc_laddr, sp);
2390 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2393 ip6_sprintf(sp, &inc->inc6_faddr);
2395 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2397 ip6_sprintf(sp, &inc->inc6_laddr);
2399 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2400 } else if (ip6 && th) {
2401 ip6_sprintf(sp, &ip6->ip6_src);
2403 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2405 ip6_sprintf(sp, &ip6->ip6_dst);
2407 sprintf(sp, "]:%i", ntohs(th->th_dport));
2410 } else if (ip && th) {
2411 inet_ntoa_r(ip->ip_src, sp);
2413 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2415 inet_ntoa_r(ip->ip_dst, sp);
2417 sprintf(sp, "]:%i", ntohs(th->th_dport));
2425 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2426 if (*(s + size - 1) != '\0')
2427 panic("%s: string too long", __func__);
2432 * A subroutine which makes it easy to track TCP state changes with DTrace.
2433 * This function shouldn't be called for t_state initializations that don't
2434 * correspond to actual TCP state transitions.
2437 tcp_state_change(struct tcpcb *tp, int newstate)
2439 #if defined(KDTRACE_HOOKS)
2440 int pstate = tp->t_state;
2443 tp->t_state = newstate;
2444 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);