1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
16 *
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
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 * $FreeBSD: head/sys/netinet/tcp_timewait.c 157927 2006-04-21 09:25:40Z ps $
31 */
32
33#include "opt_compat.h"
34#include "opt_inet.h"
35#include "opt_inet6.h"
36#include "opt_ipsec.h"
37#include "opt_mac.h"
38#include "opt_tcpdebug.h"
39#include "opt_tcp_sack.h"
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/callout.h>
44#include <sys/kernel.h>
45#include <sys/sysctl.h>
46#include <sys/mac.h>
47#include <sys/malloc.h>
48#include <sys/mbuf.h>
49#ifdef INET6
50#include <sys/domain.h>
51#endif
52#include <sys/proc.h>
53#include <sys/socket.h>
54#include <sys/socketvar.h>
55#include <sys/protosw.h>
56#include <sys/random.h>
57
58#include <vm/uma.h>
59
60#include <net/route.h>
61#include <net/if.h>
62
63#include <netinet/in.h>
64#include <netinet/in_systm.h>
65#include <netinet/ip.h>
66#ifdef INET6
67#include <netinet/ip6.h>
68#endif
69#include <netinet/in_pcb.h>
70#ifdef INET6
71#include <netinet6/in6_pcb.h>
72#endif
73#include <netinet/in_var.h>
74#include <netinet/ip_var.h>
75#ifdef INET6
76#include <netinet6/ip6_var.h>
77#include <netinet6/scope6_var.h>
78#include <netinet6/nd6.h>
79#endif
80#include <netinet/ip_icmp.h>
81#include <netinet/tcp.h>
82#include <netinet/tcp_fsm.h>
83#include <netinet/tcp_seq.h>
84#include <netinet/tcp_timer.h>
85#include <netinet/tcp_var.h>
86#ifdef INET6
87#include <netinet6/tcp6_var.h>
88#endif
89#include <netinet/tcpip.h>
90#ifdef TCPDEBUG
91#include <netinet/tcp_debug.h>
92#endif
93#include <netinet6/ip6protosw.h>
94
95#ifdef IPSEC
96#include <netinet6/ipsec.h>
97#ifdef INET6
98#include <netinet6/ipsec6.h>
99#endif
100#include <netkey/key.h>
101#endif /*IPSEC*/
102
103#ifdef FAST_IPSEC
104#include <netipsec/ipsec.h>
105#include <netipsec/xform.h>
106#ifdef INET6
107#include <netipsec/ipsec6.h>
108#endif
109#include <netipsec/key.h>
110#define IPSEC
111#endif /*FAST_IPSEC*/
112
113#include <machine/in_cksum.h>
114#include <sys/md5.h>
115
116int tcp_mssdflt = TCP_MSS;
117SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
119
120#ifdef INET6
121int tcp_v6mssdflt = TCP6_MSS;
122SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
123 CTLFLAG_RW, &tcp_v6mssdflt , 0,
124 "Default TCP Maximum Segment Size for IPv6");
125#endif
126
127/*
128 * Minimum MSS we accept and use. This prevents DoS attacks where
129 * we are forced to a ridiculous low MSS like 20 and send hundreds
130 * of packets instead of one. The effect scales with the available
131 * bandwidth and quickly saturates the CPU and network interface
132 * with packet generation and sending. Set to zero to disable MINMSS
133 * checking. This setting prevents us from sending too small packets.
134 */
135int tcp_minmss = TCP_MINMSS;
136SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
138/*
139 * Number of TCP segments per second we accept from remote host
140 * before we start to calculate average segment size. If average
141 * segment size drops below the minimum TCP MSS we assume a DoS
142 * attack and reset+drop the connection. Care has to be taken not to
143 * set this value too small to not kill interactive type connections
144 * (telnet, SSH) which send many small packets.
145 */
146int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
147SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
149 "be under the MINMSS Size");
150
151#if 0
152static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
153SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
155#endif
156
157int tcp_do_rfc1323 = 1;
158SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
160
161static int tcp_tcbhashsize = 0;
162SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
164
165static int do_tcpdrain = 1;
166SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
167 "Enable tcp_drain routine for extra help when low on mbufs");
168
169SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
170 &tcbinfo.ipi_count, 0, "Number of active PCBs");
171
172static int icmp_may_rst = 1;
173SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
174 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
175
176static int tcp_isn_reseed_interval = 0;
177SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
179
180static int maxtcptw;
181SYSCTL_INT(_net_inet_tcp, OID_AUTO, maxtcptw, CTLFLAG_RDTUN,
182 &maxtcptw, 0, "Maximum number of compressed TCP TIME_WAIT entries");
183
184/*
185 * TCP bandwidth limiting sysctls. Note that the default lower bound of
186 * 1024 exists only for debugging. A good production default would be
187 * something like 6100.
188 */
189SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
190 "TCP inflight data limiting");
191
192static int tcp_inflight_enable = 1;
193SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
194 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
195
196static int tcp_inflight_debug = 0;
197SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
198 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
199
200static int tcp_inflight_rttthresh;
201SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
202 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
203 "RTT threshold below which inflight will deactivate itself");
204
205static int tcp_inflight_min = 6144;
206SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
207 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
208
209static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
210SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
211 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
212
213static int tcp_inflight_stab = 20;
214SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
215 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
216
217uma_zone_t sack_hole_zone;
218
219static struct inpcb *tcp_notify(struct inpcb *, int);
220static void tcp_isn_tick(void *);
221
222/*
223 * Target size of TCP PCB hash tables. Must be a power of two.
224 *
225 * Note that this can be overridden by the kernel environment
226 * variable net.inet.tcp.tcbhashsize
227 */
228#ifndef TCBHASHSIZE
229#define TCBHASHSIZE 512
230#endif
231
232/*
233 * XXX
234 * Callouts should be moved into struct tcp directly. They are currently
235 * separate because the tcpcb structure is exported to userland for sysctl
236 * parsing purposes, which do not know about callouts.
237 */
238struct tcpcb_mem {
239 struct tcpcb tcb;
240 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
241 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
242};
243
244static uma_zone_t tcpcb_zone;
245static uma_zone_t tcptw_zone;
246struct callout isn_callout;
247
248/*
249 * TCP initialization.
250 */
251static void
252tcp_zone_change(void *tag)
253{
254
255 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
256 uma_zone_set_max(tcpcb_zone, maxsockets);
257 uma_zone_set_max(tcptw_zone, maxsockets / 5);
258}
259
260void
261tcp_init(void)
262{
263 int hashsize = TCBHASHSIZE;
264
265 tcp_delacktime = TCPTV_DELACK;
266 tcp_keepinit = TCPTV_KEEP_INIT;
267 tcp_keepidle = TCPTV_KEEP_IDLE;
268 tcp_keepintvl = TCPTV_KEEPINTVL;
269 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
270 tcp_msl = TCPTV_MSL;
271 tcp_rexmit_min = TCPTV_MIN;
272 tcp_rexmit_slop = TCPTV_CPU_VAR;
273 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
274
275 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
276 LIST_INIT(&tcb);
277 tcbinfo.listhead = &tcb;
278 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
279 if (!powerof2(hashsize)) {
280 printf("WARNING: TCB hash size not a power of 2\n");
281 hashsize = 512; /* safe default */
282 }
283 tcp_tcbhashsize = hashsize;
284 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
285 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
286 &tcbinfo.porthashmask);
287 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
288 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
289 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
290#ifdef INET6
291#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
292#else /* INET6 */
293#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
294#endif /* INET6 */
295 if (max_protohdr < TCP_MINPROTOHDR)
296 max_protohdr = TCP_MINPROTOHDR;
297 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
298 panic("tcp_init");
299#undef TCP_MINPROTOHDR
300 /*
301 * These have to be type stable for the benefit of the timers.
302 */
303 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
304 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 uma_zone_set_max(tcpcb_zone, maxsockets);
306 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
307 if (maxtcptw == 0)
308 maxtcptw = maxsockets / 5;
309 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
310 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
311 uma_zone_set_max(tcptw_zone, maxtcptw);
312 tcp_timer_init();
313 syncache_init();
314 tcp_hc_init();
315 tcp_reass_init();
316 callout_init(&isn_callout, CALLOUT_MPSAFE);
317 tcp_isn_tick(NULL);
318 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
319 SHUTDOWN_PRI_DEFAULT);
320 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
321 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
322 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
323 EVENTHANDLER_PRI_ANY);
324}
325
326void
327tcp_fini(void *xtp)
328{
329
330 callout_stop(&isn_callout);
331}
332
333/*
334 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
335 * tcp_template used to store this data in mbufs, but we now recopy it out
336 * of the tcpcb each time to conserve mbufs.
337 */
338void
339tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
340{
341 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
342
343 INP_LOCK_ASSERT(inp);
344
345#ifdef INET6
346 if ((inp->inp_vflag & INP_IPV6) != 0) {
347 struct ip6_hdr *ip6;
348
349 ip6 = (struct ip6_hdr *)ip_ptr;
350 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
351 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
352 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
353 (IPV6_VERSION & IPV6_VERSION_MASK);
354 ip6->ip6_nxt = IPPROTO_TCP;
355 ip6->ip6_plen = sizeof(struct tcphdr);
356 ip6->ip6_src = inp->in6p_laddr;
357 ip6->ip6_dst = inp->in6p_faddr;
358 } else
359#endif
360 {
361 struct ip *ip;
362
363 ip = (struct ip *)ip_ptr;
364 ip->ip_v = IPVERSION;
365 ip->ip_hl = 5;
366 ip->ip_tos = inp->inp_ip_tos;
367 ip->ip_len = 0;
368 ip->ip_id = 0;
369 ip->ip_off = 0;
370 ip->ip_ttl = inp->inp_ip_ttl;
371 ip->ip_sum = 0;
372 ip->ip_p = IPPROTO_TCP;
373 ip->ip_src = inp->inp_laddr;
374 ip->ip_dst = inp->inp_faddr;
375 }
376 th->th_sport = inp->inp_lport;
377 th->th_dport = inp->inp_fport;
378 th->th_seq = 0;
379 th->th_ack = 0;
380 th->th_x2 = 0;
381 th->th_off = 5;
382 th->th_flags = 0;
383 th->th_win = 0;
384 th->th_urp = 0;
385 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
386}
387
388/*
389 * Create template to be used to send tcp packets on a connection.
390 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
391 * use for this function is in keepalives, which use tcp_respond.
392 */
393struct tcptemp *
394tcpip_maketemplate(struct inpcb *inp)
395{
396 struct mbuf *m;
397 struct tcptemp *n;
398
399 m = m_get(M_DONTWAIT, MT_DATA);
400 if (m == NULL)
401 return (0);
402 m->m_len = sizeof(struct tcptemp);
403 n = mtod(m, struct tcptemp *);
404
405 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
406 return (n);
407}
408
409/*
410 * Send a single message to the TCP at address specified by
411 * the given TCP/IP header. If m == NULL, then we make a copy
412 * of the tcpiphdr at ti and send directly to the addressed host.
413 * This is used to force keep alive messages out using the TCP
414 * template for a connection. If flags are given then we send
415 * a message back to the TCP which originated the * segment ti,
416 * and discard the mbuf containing it and any other attached mbufs.
417 *
418 * In any case the ack and sequence number of the transmitted
419 * segment are as specified by the parameters.
420 *
421 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
422 */
423void
424tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
425 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
426{
427 register int tlen;
428 int win = 0;
429 struct ip *ip;
430 struct tcphdr *nth;
431#ifdef INET6
432 struct ip6_hdr *ip6;
433 int isipv6;
434#endif /* INET6 */
435 int ipflags = 0;
436 struct inpcb *inp;
437
438 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
439
440#ifdef INET6
441 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
442 ip6 = ipgen;
443#endif /* INET6 */
444 ip = ipgen;
445
446 if (tp != NULL) {
447 inp = tp->t_inpcb;
448 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
449 INP_INFO_WLOCK_ASSERT(&tcbinfo);
450 INP_LOCK_ASSERT(inp);
451 } else
452 inp = NULL;
453
454 if (tp != NULL) {
455 if (!(flags & TH_RST)) {
456 win = sbspace(&inp->inp_socket->so_rcv);
457 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
458 win = (long)TCP_MAXWIN << tp->rcv_scale;
459 }
460 }
461 if (m == NULL) {
462 m = m_gethdr(M_DONTWAIT, MT_DATA);
463 if (m == NULL)
464 return;
465 tlen = 0;
466 m->m_data += max_linkhdr;
467#ifdef INET6
468 if (isipv6) {
469 bcopy((caddr_t)ip6, mtod(m, caddr_t),
470 sizeof(struct ip6_hdr));
471 ip6 = mtod(m, struct ip6_hdr *);
472 nth = (struct tcphdr *)(ip6 + 1);
473 } else
474#endif /* INET6 */
475 {
476 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
477 ip = mtod(m, struct ip *);
478 nth = (struct tcphdr *)(ip + 1);
479 }
480 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
481 flags = TH_ACK;
482 } else {
483 m_freem(m->m_next);
484 m->m_next = NULL;
485 m->m_data = (caddr_t)ipgen;
486 /* m_len is set later */
487 tlen = 0;
488#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
489#ifdef INET6
490 if (isipv6) {
491 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
492 nth = (struct tcphdr *)(ip6 + 1);
493 } else
494#endif /* INET6 */
495 {
496 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
497 nth = (struct tcphdr *)(ip + 1);
498 }
499 if (th != nth) {
500 /*
501 * this is usually a case when an extension header
502 * exists between the IPv6 header and the
503 * TCP header.
504 */
505 nth->th_sport = th->th_sport;
506 nth->th_dport = th->th_dport;
507 }
508 xchg(nth->th_dport, nth->th_sport, n_short);
509#undef xchg
510 }
511#ifdef INET6
512 if (isipv6) {
513 ip6->ip6_flow = 0;
514 ip6->ip6_vfc = IPV6_VERSION;
515 ip6->ip6_nxt = IPPROTO_TCP;
516 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
517 tlen));
518 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
519 } else
520#endif
521 {
522 tlen += sizeof (struct tcpiphdr);
523 ip->ip_len = tlen;
524 ip->ip_ttl = ip_defttl;
525 if (path_mtu_discovery)
526 ip->ip_off |= IP_DF;
527 }
528 m->m_len = tlen;
529 m->m_pkthdr.len = tlen;
530 m->m_pkthdr.rcvif = NULL;
531#ifdef MAC
532 if (inp != NULL) {
533 /*
534 * Packet is associated with a socket, so allow the
535 * label of the response to reflect the socket label.
536 */
537 INP_LOCK_ASSERT(inp);
538 mac_create_mbuf_from_inpcb(inp, m);
539 } else {
540 /*
541 * Packet is not associated with a socket, so possibly
542 * update the label in place.
543 */
544 mac_reflect_mbuf_tcp(m);
545 }
546#endif
547 nth->th_seq = htonl(seq);
548 nth->th_ack = htonl(ack);
549 nth->th_x2 = 0;
550 nth->th_off = sizeof (struct tcphdr) >> 2;
551 nth->th_flags = flags;
552 if (tp != NULL)
553 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
554 else
555 nth->th_win = htons((u_short)win);
556 nth->th_urp = 0;
557#ifdef INET6
558 if (isipv6) {
559 nth->th_sum = 0;
560 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
561 sizeof(struct ip6_hdr),
562 tlen - sizeof(struct ip6_hdr));
563 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
564 NULL, NULL);
565 } else
566#endif /* INET6 */
567 {
568 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
569 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
570 m->m_pkthdr.csum_flags = CSUM_TCP;
571 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
572 }
573#ifdef TCPDEBUG
574 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
575 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
576#endif
577#ifdef INET6
578 if (isipv6)
579 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
580 else
581#endif /* INET6 */
582 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
583}
584
585/*
586 * Create a new TCP control block, making an
587 * empty reassembly queue and hooking it to the argument
588 * protocol control block. The `inp' parameter must have
589 * come from the zone allocator set up in tcp_init().
590 */
591struct tcpcb *
592tcp_newtcpcb(struct inpcb *inp)
593{
594 struct tcpcb_mem *tm;
595 struct tcpcb *tp;
596#ifdef INET6
597 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
598#endif /* INET6 */
599
600 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
601 if (tm == NULL)
602 return (NULL);
603 tp = &tm->tcb;
604 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
605 tp->t_maxseg = tp->t_maxopd =
606#ifdef INET6
607 isipv6 ? tcp_v6mssdflt :
608#endif /* INET6 */
609 tcp_mssdflt;
610
611 /* Set up our timeouts. */
612 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
613 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
614 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
615 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
616 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
617
618 if (tcp_do_rfc1323)
619 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
620 tp->sack_enable = tcp_do_sack;
621 TAILQ_INIT(&tp->snd_holes);
622 tp->t_inpcb = inp; /* XXX */
623 /*
624 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
625 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
626 * reasonable initial retransmit time.
627 */
628 tp->t_srtt = TCPTV_SRTTBASE;
629 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
630 tp->t_rttmin = tcp_rexmit_min;
631 tp->t_rxtcur = TCPTV_RTOBASE;
632 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
633 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
634 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
635 tp->t_rcvtime = ticks;
636 tp->t_bw_rtttime = ticks;
637 /*
638 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
639 * because the socket may be bound to an IPv6 wildcard address,
640 * which may match an IPv4-mapped IPv6 address.
641 */
642 inp->inp_ip_ttl = ip_defttl;
643 inp->inp_ppcb = tp;
644 return (tp); /* XXX */
645}
646
647/*
648 * Drop a TCP connection, reporting
649 * the specified error. If connection is synchronized,
650 * then send a RST to peer.
651 */
652struct tcpcb *
653tcp_drop(struct tcpcb *tp, int errno)
654{
655 struct socket *so = tp->t_inpcb->inp_socket;
656
657 INP_INFO_WLOCK_ASSERT(&tcbinfo);
658 INP_LOCK_ASSERT(tp->t_inpcb);
659
660 if (TCPS_HAVERCVDSYN(tp->t_state)) {
661 tp->t_state = TCPS_CLOSED;
662 (void) tcp_output(tp);
663 tcpstat.tcps_drops++;
664 } else
665 tcpstat.tcps_conndrops++;
666 if (errno == ETIMEDOUT && tp->t_softerror)
667 errno = tp->t_softerror;
668 so->so_error = errno;
669 return (tcp_close(tp));
670}
671
672void
673tcp_discardcb(struct tcpcb *tp)
674{
675 struct tseg_qent *q;
676 struct inpcb *inp = tp->t_inpcb;
677 struct socket *so = inp->inp_socket;
678#ifdef INET6
679 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
680#endif /* INET6 */
681
682 /*
683 * XXXRW: This is all very well and good, but actually, we might be
684 * discarding the tcpcb after the socket is gone, so we can't do
685 * this:
686 KASSERT(so != NULL, ("tcp_discardcb: so == NULL"));
687 */
688 INP_LOCK_ASSERT(inp);
689
690 /*
691 * Make sure that all of our timers are stopped before we
692 * delete the PCB.
693 */
694 callout_stop(tp->tt_rexmt);
695 callout_stop(tp->tt_persist);
696 callout_stop(tp->tt_keep);
697 callout_stop(tp->tt_2msl);
698 callout_stop(tp->tt_delack);
699
700 /*
701 * If we got enough samples through the srtt filter,
702 * save the rtt and rttvar in the routing entry.
703 * 'Enough' is arbitrarily defined as 4 rtt samples.
704 * 4 samples is enough for the srtt filter to converge
705 * to within enough % of the correct value; fewer samples
706 * and we could save a bogus rtt. The danger is not high
707 * as tcp quickly recovers from everything.
708 * XXX: Works very well but needs some more statistics!
709 */
710 if (tp->t_rttupdated >= 4) {
711 struct hc_metrics_lite metrics;
712 u_long ssthresh;
713
714 bzero(&metrics, sizeof(metrics));
715 /*
716 * Update the ssthresh always when the conditions below
717 * are satisfied. This gives us better new start value
718 * for the congestion avoidance for new connections.
719 * ssthresh is only set if packet loss occured on a session.
720 */
721 ssthresh = tp->snd_ssthresh;
722 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
723 /*
724 * convert the limit from user data bytes to
725 * packets then to packet data bytes.
726 */
727 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
728 if (ssthresh < 2)
729 ssthresh = 2;
730 ssthresh *= (u_long)(tp->t_maxseg +
731#ifdef INET6
732 (isipv6 ? sizeof (struct ip6_hdr) +
733 sizeof (struct tcphdr) :
734#endif
735 sizeof (struct tcpiphdr)
736#ifdef INET6
737 )
738#endif
739 );
740 } else
741 ssthresh = 0;
742 metrics.rmx_ssthresh = ssthresh;
743
744 metrics.rmx_rtt = tp->t_srtt;
745 metrics.rmx_rttvar = tp->t_rttvar;
746 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
747 metrics.rmx_bandwidth = tp->snd_bandwidth;
748 metrics.rmx_cwnd = tp->snd_cwnd;
749 metrics.rmx_sendpipe = 0;
750 metrics.rmx_recvpipe = 0;
751
752 tcp_hc_update(&inp->inp_inc, &metrics);
753 }
754
755 /* free the reassembly queue, if any */
756 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
757 LIST_REMOVE(q, tqe_q);
758 m_freem(q->tqe_m);
759 uma_zfree(tcp_reass_zone, q);
760 tp->t_segqlen--;
761 tcp_reass_qsize--;
762 }
763 tcp_free_sackholes(tp);
764 inp->inp_ppcb = NULL;
765 tp->t_inpcb = NULL;
766 uma_zfree(tcpcb_zone, tp);
767
768 /*
769 * XXXRW: This seems a bit unclean.
770 */
771 if (so != NULL)
772 soisdisconnected(so);
773}
774
775/*
776 * Attempt to close a TCP control block, marking it as dropped, and freeing
777 * the socket if we hold the only reference.
778 */
779struct tcpcb *
780tcp_close(struct tcpcb *tp)
781{
782 struct inpcb *inp = tp->t_inpcb;
783 struct socket *so;
784
785 INP_INFO_WLOCK_ASSERT(&tcbinfo);
786 INP_LOCK_ASSERT(inp);
787
788 inp->inp_vflag |= INP_DROPPED;
789
790 tcpstat.tcps_closed++;
791 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
792 so = inp->inp_socket;
793 soisdisconnected(so);
794 if (inp->inp_vflag & INP_SOCKREF) {
795 KASSERT(so->so_state & SS_PROTOREF,
796 ("tcp_close: !SS_PROTOREF"));
797 inp->inp_vflag &= ~INP_SOCKREF;
798 tcp_discardcb(tp);
799#ifdef INET6
800 if (inp->inp_vflag & INP_IPV6PROTO) {
801 in6_pcbdetach(inp);
802 in6_pcbfree(inp);
803 } else {
804#endif
805 in_pcbdetach(inp);
806 in_pcbfree(inp);
807#ifdef INET6
808 }
809#endif
810 ACCEPT_LOCK();
811 SOCK_LOCK(so);
812 so->so_state &= ~SS_PROTOREF;
813 sofree(so);
814 return (NULL);
815 }
816 return (tp);
817}
818
819void
820tcp_drain(void)
821{
822
823 if (do_tcpdrain) {
824 struct inpcb *inpb;
825 struct tcpcb *tcpb;
826 struct tseg_qent *te;
827
828 /*
829 * Walk the tcpbs, if existing, and flush the reassembly queue,
830 * if there is one...
831 * XXX: The "Net/3" implementation doesn't imply that the TCP
832 * reassembly queue should be flushed, but in a situation
833 * where we're really low on mbufs, this is potentially
834 * usefull.
835 */
836 INP_INFO_RLOCK(&tcbinfo);
837 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
838 if (inpb->inp_vflag & INP_TIMEWAIT)
839 continue;
840 INP_LOCK(inpb);
841 if ((tcpb = intotcpcb(inpb)) != NULL) {
842 while ((te = LIST_FIRST(&tcpb->t_segq))
843 != NULL) {
844 LIST_REMOVE(te, tqe_q);
845 m_freem(te->tqe_m);
846 uma_zfree(tcp_reass_zone, te);
847 tcpb->t_segqlen--;
848 tcp_reass_qsize--;
849 }
850 tcp_clean_sackreport(tcpb);
851 }
852 INP_UNLOCK(inpb);
853 }
854 INP_INFO_RUNLOCK(&tcbinfo);
855 }
856}
857
858/*
859 * Notify a tcp user of an asynchronous error;
860 * store error as soft error, but wake up user
861 * (for now, won't do anything until can select for soft error).
862 *
863 * Do not wake up user since there currently is no mechanism for
864 * reporting soft errors (yet - a kqueue filter may be added).
865 */
866static struct inpcb *
867tcp_notify(struct inpcb *inp, int error)
868{
869 struct tcpcb *tp;
870
871 INP_INFO_WLOCK_ASSERT(&tcbinfo);
872 INP_LOCK_ASSERT(inp);
873
874 if ((inp->inp_vflag & INP_TIMEWAIT) ||
875 (inp->inp_vflag & INP_DROPPED))
876 return (inp);
877
878 tp = intotcpcb(inp);
879 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
880
881 /*
882 * Ignore some errors if we are hooked up.
883 * If connection hasn't completed, has retransmitted several times,
884 * and receives a second error, give up now. This is better
885 * than waiting a long time to establish a connection that
886 * can never complete.
887 */
888 if (tp->t_state == TCPS_ESTABLISHED &&
889 (error == EHOSTUNREACH || error == ENETUNREACH ||
890 error == EHOSTDOWN)) {
891 return (inp);
892 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
893 tp->t_softerror) {
894 tp = tcp_drop(tp, error);
895 if (tp != NULL)
896 return (inp);
897 else
898 return (NULL);
899 } else {
900 tp->t_softerror = error;
901 return (inp);
902 }
903#if 0
904 wakeup( &so->so_timeo);
905 sorwakeup(so);
906 sowwakeup(so);
907#endif
908}
909
910static int
911tcp_pcblist(SYSCTL_HANDLER_ARGS)
912{
913 int error, i, n;
914 struct inpcb *inp, **inp_list;
915 inp_gen_t gencnt;
916 struct xinpgen xig;
917
918 /*
919 * The process of preparing the TCB list is too time-consuming and
920 * resource-intensive to repeat twice on every request.
921 */
922 if (req->oldptr == NULL) {
923 n = tcbinfo.ipi_count;
924 req->oldidx = 2 * (sizeof xig)
925 + (n + n/8) * sizeof(struct xtcpcb);
926 return (0);
927 }
928
929 if (req->newptr != NULL)
930 return (EPERM);
931
932 /*
933 * OK, now we're committed to doing something.
934 */
935 INP_INFO_RLOCK(&tcbinfo);
936 gencnt = tcbinfo.ipi_gencnt;
937 n = tcbinfo.ipi_count;
938 INP_INFO_RUNLOCK(&tcbinfo);
939
940 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
941 + n * sizeof(struct xtcpcb));
942 if (error != 0)
943 return (error);
944
945 xig.xig_len = sizeof xig;
946 xig.xig_count = n;
947 xig.xig_gen = gencnt;
948 xig.xig_sogen = so_gencnt;
949 error = SYSCTL_OUT(req, &xig, sizeof xig);
950 if (error)
951 return (error);
952
953 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
954 if (inp_list == NULL)
955 return (ENOMEM);
956
957 INP_INFO_RLOCK(&tcbinfo);
958 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
959 inp = LIST_NEXT(inp, inp_list)) {
960 INP_LOCK(inp);
961 if (inp->inp_gencnt <= gencnt) {
962 /*
963 * XXX: This use of cr_cansee(), introduced with
964 * TCP state changes, is not quite right, but for
965 * now, better than nothing.
966 */
967 if (inp->inp_vflag & INP_TIMEWAIT) {
968 if (intotw(inp) != NULL)
969 error = cr_cansee(req->td->td_ucred,
970 intotw(inp)->tw_cred);
971 else
972 error = EINVAL; /* Skip this inp. */
973 } else
974 error = cr_canseesocket(req->td->td_ucred,
975 inp->inp_socket);
976 if (error == 0)
977 inp_list[i++] = inp;
978 }
979 INP_UNLOCK(inp);
980 }
981 INP_INFO_RUNLOCK(&tcbinfo);
982 n = i;
983
984 error = 0;
985 for (i = 0; i < n; i++) {
986 inp = inp_list[i];
987 if (inp->inp_gencnt <= gencnt) {
988 struct xtcpcb xt;
989 void *inp_ppcb;
990
991 bzero(&xt, sizeof(xt));
992 xt.xt_len = sizeof xt;
993 /* XXX should avoid extra copy */
994 bcopy(inp, &xt.xt_inp, sizeof *inp);
995 inp_ppcb = inp->inp_ppcb;
996 if (inp_ppcb == NULL)
997 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
998 else if (inp->inp_vflag & INP_TIMEWAIT) {
999 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1000 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1001 } else
1002 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1003 if (inp->inp_socket != NULL)
1004 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1005 else {
1006 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1007 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1008 }
1009 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1010 error = SYSCTL_OUT(req, &xt, sizeof xt);
1011 }
1012 }
1013 if (!error) {
1014 /*
1015 * Give the user an updated idea of our state.
1016 * If the generation differs from what we told
1017 * her before, she knows that something happened
1018 * while we were processing this request, and it
1019 * might be necessary to retry.
1020 */
1021 INP_INFO_RLOCK(&tcbinfo);
1022 xig.xig_gen = tcbinfo.ipi_gencnt;
1023 xig.xig_sogen = so_gencnt;
1024 xig.xig_count = tcbinfo.ipi_count;
1025 INP_INFO_RUNLOCK(&tcbinfo);
1026 error = SYSCTL_OUT(req, &xig, sizeof xig);
1027 }
1028 free(inp_list, M_TEMP);
1029 return (error);
1030}
1031
1032SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1033 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1034
1035static int
1036tcp_getcred(SYSCTL_HANDLER_ARGS)
1037{
1038 struct xucred xuc;
1039 struct sockaddr_in addrs[2];
1040 struct inpcb *inp;
1041 int error;
1042
1043 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1044 if (error)
1045 return (error);
1046 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1047 if (error)
1048 return (error);
1049 INP_INFO_RLOCK(&tcbinfo);
1050 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1051 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1052 if (inp == NULL) {
1053 error = ENOENT;
1054 goto outunlocked;
1055 }
1056 INP_LOCK(inp);
1057 if (inp->inp_socket == NULL) {
1058 error = ENOENT;
1059 goto out;
1060 }
1061 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1062 if (error)
1063 goto out;
1064 cru2x(inp->inp_socket->so_cred, &xuc);
1065out:
1066 INP_UNLOCK(inp);
1067outunlocked:
1068 INP_INFO_RUNLOCK(&tcbinfo);
1069 if (error == 0)
1070 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1071 return (error);
1072}
1073
1074SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1075 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1076 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1077
1078#ifdef INET6
1079static int
1080tcp6_getcred(SYSCTL_HANDLER_ARGS)
1081{
1082 struct xucred xuc;
1083 struct sockaddr_in6 addrs[2];
1084 struct inpcb *inp;
1085 int error, mapped = 0;
1086
1087 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1088 if (error)
1089 return (error);
1090 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1091 if (error)
1092 return (error);
1093 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1094 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1095 return (error);
1096 }
1097 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1098 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1099 mapped = 1;
1100 else
1101 return (EINVAL);
1102 }
1103
1104 INP_INFO_RLOCK(&tcbinfo);
1105 if (mapped == 1)
1106 inp = in_pcblookup_hash(&tcbinfo,
1107 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1108 addrs[1].sin6_port,
1109 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1110 addrs[0].sin6_port,
1111 0, NULL);
1112 else
1113 inp = in6_pcblookup_hash(&tcbinfo,
1114 &addrs[1].sin6_addr, addrs[1].sin6_port,
1115 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1116 if (inp == NULL) {
1117 error = ENOENT;
1118 goto outunlocked;
1119 }
1120 INP_LOCK(inp);
1121 if (inp->inp_socket == NULL) {
1122 error = ENOENT;
1123 goto out;
1124 }
1125 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1126 if (error)
1127 goto out;
1128 cru2x(inp->inp_socket->so_cred, &xuc);
1129out:
1130 INP_UNLOCK(inp);
1131outunlocked:
1132 INP_INFO_RUNLOCK(&tcbinfo);
1133 if (error == 0)
1134 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1135 return (error);
1136}
1137
1138SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1139 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1140 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1141#endif
1142
1143
1144void
1145tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1146{
1147 struct ip *ip = vip;
1148 struct tcphdr *th;
1149 struct in_addr faddr;
1150 struct inpcb *inp;
1151 struct tcpcb *tp;
1152 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1153 struct icmp *icp;
1154 struct in_conninfo inc;
1155 tcp_seq icmp_tcp_seq;
1156 int mtu;
1157
1158 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1159 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1160 return;
1161
1162 if (cmd == PRC_MSGSIZE)
1163 notify = tcp_mtudisc;
1164 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1165 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1166 notify = tcp_drop_syn_sent;
1167 /*
1168 * Redirects don't need to be handled up here.
1169 */
1170 else if (PRC_IS_REDIRECT(cmd))
1171 return;
1172 /*
1173 * Source quench is depreciated.
1174 */
1175 else if (cmd == PRC_QUENCH)
1176 return;
1177 /*
1178 * Hostdead is ugly because it goes linearly through all PCBs.
1179 * XXX: We never get this from ICMP, otherwise it makes an
1180 * excellent DoS attack on machines with many connections.
1181 */
1182 else if (cmd == PRC_HOSTDEAD)
1183 ip = NULL;
1184 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1185 return;
1186 if (ip != NULL) {
1187 icp = (struct icmp *)((caddr_t)ip
1188 - offsetof(struct icmp, icmp_ip));
1189 th = (struct tcphdr *)((caddr_t)ip
1190 + (ip->ip_hl << 2));
1191 INP_INFO_WLOCK(&tcbinfo);
1192 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1193 ip->ip_src, th->th_sport, 0, NULL);
1194 if (inp != NULL) {
1195 INP_LOCK(inp);
1196 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1197 !(inp->inp_vflag & INP_DROPPED) &&
1198 !(inp->inp_socket == NULL)) {
1199 icmp_tcp_seq = htonl(th->th_seq);
1200 tp = intotcpcb(inp);
1201 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1202 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1203 if (cmd == PRC_MSGSIZE) {
1204 /*
1205 * MTU discovery:
1206 * If we got a needfrag set the MTU
1207 * in the route to the suggested new
1208 * value (if given) and then notify.
1209 */
1210 bzero(&inc, sizeof(inc));
1211 inc.inc_flags = 0; /* IPv4 */
1212 inc.inc_faddr = faddr;
1213
1214 mtu = ntohs(icp->icmp_nextmtu);
1215 /*
1216 * If no alternative MTU was
1217 * proposed, try the next smaller
1218 * one. ip->ip_len has already
1219 * been swapped in icmp_input().
1220 */
1221 if (!mtu)
1222 mtu = ip_next_mtu(ip->ip_len,
1223 1);
1224 if (mtu < max(296, (tcp_minmss)
1225 + sizeof(struct tcpiphdr)))
1226 mtu = 0;
1227 if (!mtu)
1228 mtu = tcp_mssdflt
1229 + sizeof(struct tcpiphdr);
1230 /*
1231 * Only cache the the MTU if it
1232 * is smaller than the interface
1233 * or route MTU. tcp_mtudisc()
1234 * will do right thing by itself.
1235 */
1236 if (mtu <= tcp_maxmtu(&inc))
1237 tcp_hc_updatemtu(&inc, mtu);
1238 }
1239
1240 inp = (*notify)(inp, inetctlerrmap[cmd]);
1241 }
1242 }
1243 if (inp != NULL)
1244 INP_UNLOCK(inp);
1245 } else {
1246 inc.inc_fport = th->th_dport;
1247 inc.inc_lport = th->th_sport;
1248 inc.inc_faddr = faddr;
1249 inc.inc_laddr = ip->ip_src;
1250#ifdef INET6
1251 inc.inc_isipv6 = 0;
1252#endif
1253 syncache_unreach(&inc, th);
1254 }
1255 INP_INFO_WUNLOCK(&tcbinfo);
1256 } else
1257 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1258}
1259
1260#ifdef INET6
1261void
1262tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1263{
1264 struct tcphdr th;
1265 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1266 struct ip6_hdr *ip6;
1267 struct mbuf *m;
1268 struct ip6ctlparam *ip6cp = NULL;
1269 const struct sockaddr_in6 *sa6_src = NULL;
1270 int off;
1271 struct tcp_portonly {
1272 u_int16_t th_sport;
1273 u_int16_t th_dport;
1274 } *thp;
1275
1276 if (sa->sa_family != AF_INET6 ||
1277 sa->sa_len != sizeof(struct sockaddr_in6))
1278 return;
1279
1280 if (cmd == PRC_MSGSIZE)
1281 notify = tcp_mtudisc;
1282 else if (!PRC_IS_REDIRECT(cmd) &&
1283 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1284 return;
1285 /* Source quench is depreciated. */
1286 else if (cmd == PRC_QUENCH)
1287 return;
1288
1289 /* if the parameter is from icmp6, decode it. */
1290 if (d != NULL) {
1291 ip6cp = (struct ip6ctlparam *)d;
1292 m = ip6cp->ip6c_m;
1293 ip6 = ip6cp->ip6c_ip6;
1294 off = ip6cp->ip6c_off;
1295 sa6_src = ip6cp->ip6c_src;
1296 } else {
1297 m = NULL;
1298 ip6 = NULL;
1299 off = 0; /* fool gcc */
1300 sa6_src = &sa6_any;
1301 }
1302
1303 if (ip6 != NULL) {
1304 struct in_conninfo inc;
1305 /*
1306 * XXX: We assume that when IPV6 is non NULL,
1307 * M and OFF are valid.
1308 */
1309
1310 /* check if we can safely examine src and dst ports */
1311 if (m->m_pkthdr.len < off + sizeof(*thp))
1312 return;
1313
1314 bzero(&th, sizeof(th));
1315 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1316
1317 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1318 (struct sockaddr *)ip6cp->ip6c_src,
1319 th.th_sport, cmd, NULL, notify);
1320
1321 inc.inc_fport = th.th_dport;
1322 inc.inc_lport = th.th_sport;
1323 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1324 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1325 inc.inc_isipv6 = 1;
1326 INP_INFO_WLOCK(&tcbinfo);
1327 syncache_unreach(&inc, &th);
1328 INP_INFO_WUNLOCK(&tcbinfo);
1329 } else
1330 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1331 0, cmd, NULL, notify);
1332}
1333#endif /* INET6 */
1334
1335
1336/*
1337 * Following is where TCP initial sequence number generation occurs.
1338 *
1339 * There are two places where we must use initial sequence numbers:
1340 * 1. In SYN-ACK packets.
1341 * 2. In SYN packets.
1342 *
1343 * All ISNs for SYN-ACK packets are generated by the syncache. See
1344 * tcp_syncache.c for details.
1345 *
1346 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1347 * depends on this property. In addition, these ISNs should be
1348 * unguessable so as to prevent connection hijacking. To satisfy
1349 * the requirements of this situation, the algorithm outlined in
1350 * RFC 1948 is used, with only small modifications.
1351 *
1352 * Implementation details:
1353 *
1354 * Time is based off the system timer, and is corrected so that it
1355 * increases by one megabyte per second. This allows for proper
1356 * recycling on high speed LANs while still leaving over an hour
1357 * before rollover.
1358 *
1359 * As reading the *exact* system time is too expensive to be done
1360 * whenever setting up a TCP connection, we increment the time
1361 * offset in two ways. First, a small random positive increment
1362 * is added to isn_offset for each connection that is set up.
1363 * Second, the function tcp_isn_tick fires once per clock tick
1364 * and increments isn_offset as necessary so that sequence numbers
1365 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1366 * random positive increments serve only to ensure that the same
1367 * exact sequence number is never sent out twice (as could otherwise
1368 * happen when a port is recycled in less than the system tick
1369 * interval.)
1370 *
1371 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1372 * between seeding of isn_secret. This is normally set to zero,
1373 * as reseeding should not be necessary.
1374 *
1375 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1376 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1377 * general, this means holding an exclusive (write) lock.
1378 */
1379
1380#define ISN_BYTES_PER_SECOND 1048576
1381#define ISN_STATIC_INCREMENT 4096
1382#define ISN_RANDOM_INCREMENT (4096 - 1)
1383
1384static u_char isn_secret[32];
1385static int isn_last_reseed;
1386static u_int32_t isn_offset, isn_offset_old;
1387static MD5_CTX isn_ctx;
1388
1389tcp_seq
1390tcp_new_isn(struct tcpcb *tp)
1391{
1392 u_int32_t md5_buffer[4];
1393 tcp_seq new_isn;
1394
1395 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1396 INP_LOCK_ASSERT(tp->t_inpcb);
1397
1398 /* Seed if this is the first use, reseed if requested. */
1399 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1400 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1401 < (u_int)ticks))) {
1402 read_random(&isn_secret, sizeof(isn_secret));
1403 isn_last_reseed = ticks;
1404 }
1405
1406 /* Compute the md5 hash and return the ISN. */
1407 MD5Init(&isn_ctx);
1408 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1409 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1410#ifdef INET6
1411 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1412 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1413 sizeof(struct in6_addr));
1414 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1415 sizeof(struct in6_addr));
1416 } else
1417#endif
1418 {
1419 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1420 sizeof(struct in_addr));
1421 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1422 sizeof(struct in_addr));
1423 }
1424 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1425 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1426 new_isn = (tcp_seq) md5_buffer[0];
1427 isn_offset += ISN_STATIC_INCREMENT +
1428 (arc4random() & ISN_RANDOM_INCREMENT);
1429 new_isn += isn_offset;
1430 return (new_isn);
1431}
1432
1433/*
1434 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1435 * to keep time flowing at a relatively constant rate. If the random
1436 * increments have already pushed us past the projected offset, do nothing.
1437 */
1438static void
1439tcp_isn_tick(void *xtp)
1440{
1441 u_int32_t projected_offset;
1442
1443 INP_INFO_WLOCK(&tcbinfo);
1444 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1445
1446 if (projected_offset > isn_offset)
1447 isn_offset = projected_offset;
1448
1449 isn_offset_old = isn_offset;
1450 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1451 INP_INFO_WUNLOCK(&tcbinfo);
1452}
1453
1454/*
1455 * When a specific ICMP unreachable message is received and the
1456 * connection state is SYN-SENT, drop the connection. This behavior
1457 * is controlled by the icmp_may_rst sysctl.
1458 */
1459struct inpcb *
1460tcp_drop_syn_sent(struct inpcb *inp, int errno)
1461{
1462 struct tcpcb *tp;
1463
1464 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1465 INP_LOCK_ASSERT(inp);
1466
1467 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1468 (inp->inp_vflag & INP_DROPPED))
1469 return (inp);
1470
1471 tp = intotcpcb(inp);
1472 if (tp->t_state != TCPS_SYN_SENT)
1473 return (inp);
1474
1475 tp = tcp_drop(tp, errno);
1476 if (tp != NULL)
1477 return (inp);
1478 else
1479 return (NULL);
1480}
1481
1482/*
1483 * When `need fragmentation' ICMP is received, update our idea of the MSS
1484 * based on the new value in the route. Also nudge TCP to send something,
1485 * since we know the packet we just sent was dropped.
1486 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1487 */
1488struct inpcb *
1489tcp_mtudisc(struct inpcb *inp, int errno)
1490{
1491 struct tcpcb *tp;
1492 struct socket *so = inp->inp_socket;
1493 u_int maxmtu;
1494 u_int romtu;
1495 int mss;
1496#ifdef INET6
1497 int isipv6;
1498#endif /* INET6 */
1499
1500 INP_LOCK_ASSERT(inp);
1501 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1502 (inp->inp_vflag & INP_DROPPED))
1503 return (inp);
1504
1505 tp = intotcpcb(inp);
1506 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1507
1508#ifdef INET6
1509 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1510#endif
1511 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1512 romtu =
1513#ifdef INET6
1514 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1515#endif /* INET6 */
1516 tcp_maxmtu(&inp->inp_inc);
1517 if (!maxmtu)
1518 maxmtu = romtu;
1519 else
1520 maxmtu = min(maxmtu, romtu);
1521 if (!maxmtu) {
1522 tp->t_maxopd = tp->t_maxseg =
1523#ifdef INET6
1524 isipv6 ? tcp_v6mssdflt :
1525#endif /* INET6 */
1526 tcp_mssdflt;
1527 return (inp);
1528 }
1529 mss = maxmtu -
1530#ifdef INET6
1531 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1532#endif /* INET6 */
1533 sizeof(struct tcpiphdr)
1534#ifdef INET6
1535 )
1536#endif /* INET6 */
1537 ;
1538
1539 /*
1540 * XXX - The above conditional probably violates the TCP
1541 * spec. The problem is that, since we don't know the
1542 * other end's MSS, we are supposed to use a conservative
1543 * default. But, if we do that, then MTU discovery will
1544 * never actually take place, because the conservative
1545 * default is much less than the MTUs typically seen
1546 * on the Internet today. For the moment, we'll sweep
1547 * this under the carpet.
1548 *
1549 * The conservative default might not actually be a problem
1550 * if the only case this occurs is when sending an initial
1551 * SYN with options and data to a host we've never talked
1552 * to before. Then, they will reply with an MSS value which
1553 * will get recorded and the new parameters should get
1554 * recomputed. For Further Study.
1555 */
1556 if (tp->t_maxopd <= mss)
1557 return (inp);
1558 tp->t_maxopd = mss;
1559
1560 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1561 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1562 mss -= TCPOLEN_TSTAMP_APPA;
1563#if (MCLBYTES & (MCLBYTES - 1)) == 0
1564 if (mss > MCLBYTES)
1565 mss &= ~(MCLBYTES-1);
1566#else
1567 if (mss > MCLBYTES)
1568 mss = mss / MCLBYTES * MCLBYTES;
1569#endif
1570 if (so->so_snd.sb_hiwat < mss)
1571 mss = so->so_snd.sb_hiwat;
1572
1573 tp->t_maxseg = mss;
1574
1575 tcpstat.tcps_mturesent++;
1576 tp->t_rtttime = 0;
1577 tp->snd_nxt = tp->snd_una;
1578 tcp_output(tp);
1579 return (inp);
1580}
1581
1582/*
1583 * Look-up the routing entry to the peer of this inpcb. If no route
1584 * is found and it cannot be allocated, then return NULL. This routine
1585 * is called by TCP routines that access the rmx structure and by tcp_mss
1586 * to get the interface MTU.
1587 */
1588u_long
1589tcp_maxmtu(struct in_conninfo *inc)
1590{
1591 struct route sro;
1592 struct sockaddr_in *dst;
1593 struct ifnet *ifp;
1594 u_long maxmtu = 0;
1595
1596 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1597
1598 bzero(&sro, sizeof(sro));
1599 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1600 dst = (struct sockaddr_in *)&sro.ro_dst;
1601 dst->sin_family = AF_INET;
1602 dst->sin_len = sizeof(*dst);
1603 dst->sin_addr = inc->inc_faddr;
1604 rtalloc_ign(&sro, RTF_CLONING);
1605 }
1606 if (sro.ro_rt != NULL) {
1607 ifp = sro.ro_rt->rt_ifp;
1608 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1609 maxmtu = ifp->if_mtu;
1610 else
1611 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1612 RTFREE(sro.ro_rt);
1613 }
1614 return (maxmtu);
1615}
1616
1617#ifdef INET6
1618u_long
1619tcp_maxmtu6(struct in_conninfo *inc)
1620{
1621 struct route_in6 sro6;
1622 struct ifnet *ifp;
1623 u_long maxmtu = 0;
1624
1625 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1626
1627 bzero(&sro6, sizeof(sro6));
1628 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1629 sro6.ro_dst.sin6_family = AF_INET6;
1630 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1631 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1632 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1633 }
1634 if (sro6.ro_rt != NULL) {
1635 ifp = sro6.ro_rt->rt_ifp;
1636 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1637 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1638 else
1639 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1640 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1641 RTFREE(sro6.ro_rt);
1642 }
1643
1644 return (maxmtu);
1645}
1646#endif /* INET6 */
1647
1648#ifdef IPSEC
1649/* compute ESP/AH header size for TCP, including outer IP header. */
1650size_t
1651ipsec_hdrsiz_tcp(struct tcpcb *tp)
1652{
1653 struct inpcb *inp;
1654 struct mbuf *m;
1655 size_t hdrsiz;
1656 struct ip *ip;
1657#ifdef INET6
1658 struct ip6_hdr *ip6;
1659#endif
1660 struct tcphdr *th;
1661
1662 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1663 return (0);
1664 MGETHDR(m, M_DONTWAIT, MT_DATA);
1665 if (!m)
1666 return (0);
1667
1668#ifdef INET6
1669 if ((inp->inp_vflag & INP_IPV6) != 0) {
1670 ip6 = mtod(m, struct ip6_hdr *);
1671 th = (struct tcphdr *)(ip6 + 1);
1672 m->m_pkthdr.len = m->m_len =
1673 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1674 tcpip_fillheaders(inp, ip6, th);
1675 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1676 } else
1677#endif /* INET6 */
1678 {
1679 ip = mtod(m, struct ip *);
1680 th = (struct tcphdr *)(ip + 1);
1681 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1682 tcpip_fillheaders(inp, ip, th);
1683 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1684 }
1685
1686 m_free(m);
1687 return (hdrsiz);
1688}
1689#endif /*IPSEC*/
1690
1691/*
1692 * Move a TCP connection into TIME_WAIT state.
1693 * tcbinfo is locked.
1694 * inp is locked, and is unlocked before returning.
1695 */
1696void
1697tcp_twstart(struct tcpcb *tp)
1698{
1699 struct tcptw *tw;
1700 struct inpcb *inp;
1701 int tw_time, acknow;
1702 struct socket *so;
1703
1704 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1705 INP_LOCK_ASSERT(tp->t_inpcb);
1706
1707 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1708 if (tw == NULL) {
1709 tw = tcp_timer_2msl_tw(1);
1710 if (tw == NULL) {
1711 tp = tcp_close(tp);
1712 if (tp != NULL)
1713 INP_UNLOCK(tp->t_inpcb);
1714 return;
1715 }
1716 }
1717 inp = tp->t_inpcb;
1718 tw->tw_inpcb = inp;
1719
1720 /*
1721 * Recover last window size sent.
1722 */
1723 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1724
1725 /*
1726 * Set t_recent if timestamps are used on the connection.
1727 */
1728 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1729 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1730 tw->t_recent = tp->ts_recent;
1731 else
1732 tw->t_recent = 0;
1733
1734 tw->snd_nxt = tp->snd_nxt;
1735 tw->rcv_nxt = tp->rcv_nxt;
1736 tw->iss = tp->iss;
1737 tw->irs = tp->irs;
1738 tw->t_starttime = tp->t_starttime;
1739 tw->tw_time = 0;
1740
1741/* XXX
1742 * If this code will
1743 * be used for fin-wait-2 state also, then we may need
1744 * a ts_recent from the last segment.
1745 */
1746 tw_time = 2 * tcp_msl;
1747 acknow = tp->t_flags & TF_ACKNOW;
1748
1749 /*
1750 * First, discard tcpcb state, which includes stopping its timers and
1751 * freeing it. tcp_discardcb() used to also release the inpcb, but
1752 * that work is now done in the caller.
1753 */
1754 tcp_discardcb(tp);
1755 so = inp->inp_socket;
1756 SOCK_LOCK(so);
1757 tw->tw_cred = crhold(so->so_cred);
1758 tw->tw_so_options = so->so_options;
1759 SOCK_UNLOCK(so);
1760 if (acknow)
1761 tcp_twrespond(tw, TH_ACK);
1762 inp->inp_ppcb = tw;
1763 inp->inp_vflag |= INP_TIMEWAIT;
1764 tcp_timer_2msl_reset(tw, tw_time);
1765
1766 /*
1767 * If the inpcb owns the sole reference to the socket, then we can
1768 * detach and free the socket as it is not needed in time wait.
1769 */
1770 if (inp->inp_vflag & INP_SOCKREF) {
1771 KASSERT(so->so_state & SS_PROTOREF,
1772 ("tcp_twstart: !SS_PROTOREF"));
1773 inp->inp_vflag &= ~INP_SOCKREF;
1774#ifdef INET6
1775 if (inp->inp_vflag & INP_IPV6PROTO)
1776 in6_pcbdetach(inp);
1777 else
1778#endif
1779 in_pcbdetach(inp);
1780 INP_UNLOCK(inp);
1781 ACCEPT_LOCK();
1782 SOCK_LOCK(so);
1783 so->so_state &= ~SS_PROTOREF;
1784 sofree(so);
1785 } else
1786 INP_UNLOCK(inp);
1787}
1788
1789/*
1790 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1791 * the actual rate is slightly higher due to the addition of
1792 * random positive increments.
1793 *
1794 * Most other new OSes use semi-randomized ISN values, so we
1795 * do not need to worry about them.
1796 */
1797#define MS_ISN_BYTES_PER_SECOND 250000
1798
1799/*
1800 * Determine if the ISN we will generate has advanced beyond the last
1801 * sequence number used by the previous connection. If so, indicate
1802 * that it is safe to recycle this tw socket by returning 1.
1803 *
1804 * XXXRW: This function should assert the inpcb lock as it does multiple
1805 * non-atomic reads from the tcptw, but is currently called without it from
1806 * in_pcb.c:in_pcblookup_local().
1807 */
1808int
1809tcp_twrecycleable(struct tcptw *tw)
1810{
1811 tcp_seq new_iss = tw->iss;
1812 tcp_seq new_irs = tw->irs;
1813
1814 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1815 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1816
1817 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1818 return (1);
1819 else
1820 return (0);
1821}
1822
1823void
1824tcp_twclose(struct tcptw *tw, int reuse)
1825{
1826 struct socket *so;
1827 struct inpcb *inp;
1828
1829 /*
1830 * At this point, we are in one of two situations:
1831 *
1832 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all
1833 * after validating.
1834 *
1835 * (2) We have a socket, which we may or may now own the reference
1836 * for. If we own the reference, release all the state after
1837 * validating. If not, leave it for the socket close to clean up.
1838 */
1839 inp = tw->tw_inpcb;
1840 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1841 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1842 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1843 INP_LOCK_ASSERT(inp);
1844
1845 tw->tw_inpcb = NULL;
1846 tcp_timer_2msl_stop(tw);
1847 inp->inp_ppcb = NULL;
1848 inp->inp_vflag |= INP_DROPPED;
1849
1850 so = inp->inp_socket;
1851 if (so != NULL) {
1852 if (inp->inp_vflag & INP_SOCKREF) {
1853 /*
1854 * If a socket is present, and we own the only
1855 * reference, we need to tear down the socket and the
1856 * inpcb.
1857 */
1858 inp->inp_vflag &= ~INP_SOCKREF;
1859#ifdef INET6
1860 if (inp->inp_vflag & INP_IPV6PROTO) {
1861 in6_pcbdetach(inp);
1862 in6_pcbfree(inp);
1863 } else {
1864 in_pcbdetach(inp);
1865 in_pcbfree(inp);
1866 }
1867#endif
1868 ACCEPT_LOCK();
1869 SOCK_LOCK(so);
1870 KASSERT(so->so_state & SS_PROTOREF,
1871 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1872 so->so_state &= ~SS_PROTOREF;
1873 sofree(so);
1874 } else {
1875 /*
1876 * If we don't own the only reference, the socket and
1877 * inpcb need to be left around to be handled by
1878 * tcp_usr_detach() later.
1879 */
1880 INP_UNLOCK(inp);
1881 }
1882 } else {
1883#ifdef INET6
1884 if (inp->inp_vflag & INP_IPV6PROTO)
1885 in6_pcbfree(inp);
1886 else
1887#endif
1888 in_pcbfree(inp);
1889 }
1890 tcpstat.tcps_closed++;
1891 crfree(tw->tw_cred);
1892 tw->tw_cred = NULL;
1893 if (reuse)
1894 return;
1895 uma_zfree(tcptw_zone, tw);
1896}
1897
1898int
1899tcp_twrespond(struct tcptw *tw, int flags)
1900{
1901 struct inpcb *inp = tw->tw_inpcb;
1902 struct tcphdr *th;
1903 struct mbuf *m;
1904 struct ip *ip = NULL;
1905 u_int8_t *optp;
1906 u_int hdrlen, optlen;
1907 int error;
1908#ifdef INET6
1909 struct ip6_hdr *ip6 = NULL;
1910 int isipv6 = inp->inp_inc.inc_isipv6;
1911#endif
1912
1913 INP_LOCK_ASSERT(inp);
1914
1915 m = m_gethdr(M_DONTWAIT, MT_DATA);
1916 if (m == NULL)
1917 return (ENOBUFS);
1918 m->m_data += max_linkhdr;
1919
1920#ifdef MAC
1921 mac_create_mbuf_from_inpcb(inp, m);
1922#endif
1923
1924#ifdef INET6
1925 if (isipv6) {
1926 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1927 ip6 = mtod(m, struct ip6_hdr *);
1928 th = (struct tcphdr *)(ip6 + 1);
1929 tcpip_fillheaders(inp, ip6, th);
1930 } else
1931#endif
1932 {
1933 hdrlen = sizeof(struct tcpiphdr);
1934 ip = mtod(m, struct ip *);
1935 th = (struct tcphdr *)(ip + 1);
1936 tcpip_fillheaders(inp, ip, th);
1937 }
1938 optp = (u_int8_t *)(th + 1);
1939
1940 /*
1941 * Send a timestamp and echo-reply if both our side and our peer
1942 * have sent timestamps in our SYN's and this is not a RST.
1943 */
1944 if (tw->t_recent && flags == TH_ACK) {
1945 u_int32_t *lp = (u_int32_t *)optp;
1946
1947 /* Form timestamp option as shown in appendix A of RFC 1323. */
1948 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1949 *lp++ = htonl(ticks);
1950 *lp = htonl(tw->t_recent);
1951 optp += TCPOLEN_TSTAMP_APPA;
1952 }
1953
1954 optlen = optp - (u_int8_t *)(th + 1);
1955
1956 m->m_len = hdrlen + optlen;
1957 m->m_pkthdr.len = m->m_len;
1958
1959 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1960
1961 th->th_seq = htonl(tw->snd_nxt);
1962 th->th_ack = htonl(tw->rcv_nxt);
1963 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1964 th->th_flags = flags;
1965 th->th_win = htons(tw->last_win);
1966
1967#ifdef INET6
1968 if (isipv6) {
1969 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1970 sizeof(struct tcphdr) + optlen);
1971 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1972 error = ip6_output(m, inp->in6p_outputopts, NULL,
1973 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1974 } else
1975#endif
1976 {
1977 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1978 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1979 m->m_pkthdr.csum_flags = CSUM_TCP;
1980 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1981 ip->ip_len = m->m_pkthdr.len;
1982 if (path_mtu_discovery)
1983 ip->ip_off |= IP_DF;
1984 error = ip_output(m, inp->inp_options, NULL,
1985 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1986 NULL, inp);
1987 }
1988 if (flags & TH_ACK)
1989 tcpstat.tcps_sndacks++;
1990 else
1991 tcpstat.tcps_sndctrl++;
1992 tcpstat.tcps_sndtotal++;
1993 return (error);
1994}
1995
1996/*
1997 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1998 *
1999 * This code attempts to calculate the bandwidth-delay product as a
2000 * means of determining the optimal window size to maximize bandwidth,
2001 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
2002 * routers. This code also does a fairly good job keeping RTTs in check
2003 * across slow links like modems. We implement an algorithm which is very
2004 * similar (but not meant to be) TCP/Vegas. The code operates on the
2005 * transmitter side of a TCP connection and so only effects the transmit
2006 * side of the connection.
2007 *
2008 * BACKGROUND: TCP makes no provision for the management of buffer space
2009 * at the end points or at the intermediate routers and switches. A TCP
2010 * stream, whether using NewReno or not, will eventually buffer as
2011 * many packets as it is able and the only reason this typically works is
2012 * due to the fairly small default buffers made available for a connection
2013 * (typicaly 16K or 32K). As machines use larger windows and/or window
2014 * scaling it is now fairly easy for even a single TCP connection to blow-out
2015 * all available buffer space not only on the local interface, but on
2016 * intermediate routers and switches as well. NewReno makes a misguided
2017 * attempt to 'solve' this problem by waiting for an actual failure to occur,
2018 * then backing off, then steadily increasing the window again until another
2019 * failure occurs, ad-infinitum. This results in terrible oscillation that
2020 * is only made worse as network loads increase and the idea of intentionally
2021 * blowing out network buffers is, frankly, a terrible way to manage network
2022 * resources.
2023 *
2024 * It is far better to limit the transmit window prior to the failure
2025 * condition being achieved. There are two general ways to do this: First
2026 * you can 'scan' through different transmit window sizes and locate the
2027 * point where the RTT stops increasing, indicating that you have filled the
2028 * pipe, then scan backwards until you note that RTT stops decreasing, then
2029 * repeat ad-infinitum. This method works in principle but has severe
2030 * implementation issues due to RTT variances, timer granularity, and
2031 * instability in the algorithm which can lead to many false positives and
2032 * create oscillations as well as interact badly with other TCP streams
2033 * implementing the same algorithm.
2034 *
2035 * The second method is to limit the window to the bandwidth delay product
2036 * of the link. This is the method we implement. RTT variances and our
2037 * own manipulation of the congestion window, bwnd, can potentially
2038 * destabilize the algorithm. For this reason we have to stabilize the
2039 * elements used to calculate the window. We do this by using the minimum
2040 * observed RTT, the long term average of the observed bandwidth, and
2041 * by adding two segments worth of slop. It isn't perfect but it is able
2042 * to react to changing conditions and gives us a very stable basis on
2043 * which to extend the algorithm.
2044 */
2045void
2046tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2047{
2048 u_long bw;
2049 u_long bwnd;
2050 int save_ticks;
2051
2052 INP_LOCK_ASSERT(tp->t_inpcb);
2053
2054 /*
2055 * If inflight_enable is disabled in the middle of a tcp connection,
2056 * make sure snd_bwnd is effectively disabled.
2057 */
2058 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2059 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2060 tp->snd_bandwidth = 0;
2061 return;
2062 }
2063
2064 /*
2065 * Figure out the bandwidth. Due to the tick granularity this
2066 * is a very rough number and it MUST be averaged over a fairly
2067 * long period of time. XXX we need to take into account a link
2068 * that is not using all available bandwidth, but for now our
2069 * slop will ramp us up if this case occurs and the bandwidth later
2070 * increases.
2071 *
2072 * Note: if ticks rollover 'bw' may wind up negative. We must
2073 * effectively reset t_bw_rtttime for this case.
2074 */
2075 save_ticks = ticks;
2076 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2077 return;
2078
2079 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2080 (save_ticks - tp->t_bw_rtttime);
2081 tp->t_bw_rtttime = save_ticks;
2082 tp->t_bw_rtseq = ack_seq;
2083 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2084 return;
2085 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2086
2087 tp->snd_bandwidth = bw;
2088
2089 /*
2090 * Calculate the semi-static bandwidth delay product, plus two maximal
2091 * segments. The additional slop puts us squarely in the sweet
2092 * spot and also handles the bandwidth run-up case and stabilization.
2093 * Without the slop we could be locking ourselves into a lower
2094 * bandwidth.
2095 *
2096 * Situations Handled:
2097 * (1) Prevents over-queueing of packets on LANs, especially on
2098 * high speed LANs, allowing larger TCP buffers to be
2099 * specified, and also does a good job preventing
2100 * over-queueing of packets over choke points like modems
2101 * (at least for the transmit side).
2102 *
2103 * (2) Is able to handle changing network loads (bandwidth
2104 * drops so bwnd drops, bandwidth increases so bwnd
2105 * increases).
2106 *
2107 * (3) Theoretically should stabilize in the face of multiple
2108 * connections implementing the same algorithm (this may need
2109 * a little work).
2110 *
2111 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2112 * be adjusted with a sysctl but typically only needs to be
2113 * on very slow connections. A value no smaller then 5
2114 * should be used, but only reduce this default if you have
2115 * no other choice.
2116 */
2117#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2118 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2119#undef USERTT
2120
2121 if (tcp_inflight_debug > 0) {
2122 static int ltime;
2123 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2124 ltime = ticks;
2125 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2126 tp,
2127 bw,
2128 tp->t_rttbest,
2129 tp->t_srtt,
2130 bwnd
2131 );
2132 }
2133 }
2134 if ((long)bwnd < tcp_inflight_min)
2135 bwnd = tcp_inflight_min;
2136 if (bwnd > tcp_inflight_max)
2137 bwnd = tcp_inflight_max;
2138 if ((long)bwnd < tp->t_maxseg * 2)
2139 bwnd = tp->t_maxseg * 2;
2140 tp->snd_bwnd = bwnd;
2141}
2142
2143#ifdef TCP_SIGNATURE
2144/*
2145 * Callback function invoked by m_apply() to digest TCP segment data
2146 * contained within an mbuf chain.
2147 */
2148static int
2149tcp_signature_apply(void *fstate, void *data, u_int len)
2150{
2151
2152 MD5Update(fstate, (u_char *)data, len);
2153 return (0);
2154}
2155
2156/*
2157 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2158 *
2159 * Parameters:
2160 * m pointer to head of mbuf chain
2161 * off0 offset to TCP header within the mbuf chain
2162 * len length of TCP segment data, excluding options
2163 * optlen length of TCP segment options
2164 * buf pointer to storage for computed MD5 digest
2165 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2166 *
2167 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2168 * When called from tcp_input(), we can be sure that th_sum has been
2169 * zeroed out and verified already.
2170 *
2171 * This function is for IPv4 use only. Calling this function with an
2172 * IPv6 packet in the mbuf chain will yield undefined results.
2173 *
2174 * Return 0 if successful, otherwise return -1.
2175 *
2176 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2177 * search with the destination IP address, and a 'magic SPI' to be
2178 * determined by the application. This is hardcoded elsewhere to 1179
2179 * right now. Another branch of this code exists which uses the SPD to
2180 * specify per-application flows but it is unstable.
2181 */
2182int
2183tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2184 u_char *buf, u_int direction)
2185{
2186 union sockaddr_union dst;
2187 struct ippseudo ippseudo;
2188 MD5_CTX ctx;
2189 int doff;
2190 struct ip *ip;
2191 struct ipovly *ipovly;
2192 struct secasvar *sav;
2193 struct tcphdr *th;
2194 u_short savecsum;
2195
2196 KASSERT(m != NULL, ("NULL mbuf chain"));
2197 KASSERT(buf != NULL, ("NULL signature pointer"));
2198
2199 /* Extract the destination from the IP header in the mbuf. */
2200 ip = mtod(m, struct ip *);
2201 bzero(&dst, sizeof(union sockaddr_union));
2202 dst.sa.sa_len = sizeof(struct sockaddr_in);
2203 dst.sa.sa_family = AF_INET;
2204 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2205 ip->ip_src : ip->ip_dst;
2206
2207 /* Look up an SADB entry which matches the address of the peer. */
2208 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2209 if (sav == NULL) {
2210 printf("%s: SADB lookup failed for %s\n", __func__,
2211 inet_ntoa(dst.sin.sin_addr));
2212 return (EINVAL);
2213 }
2214
2215 MD5Init(&ctx);
2216 ipovly = (struct ipovly *)ip;
2217 th = (struct tcphdr *)((u_char *)ip + off0);
2218 doff = off0 + sizeof(struct tcphdr) + optlen;
2219
2220 /*
2221 * Step 1: Update MD5 hash with IP pseudo-header.
2222 *
2223 * XXX The ippseudo header MUST be digested in network byte order,
2224 * or else we'll fail the regression test. Assume all fields we've
2225 * been doing arithmetic on have been in host byte order.
2226 * XXX One cannot depend on ipovly->ih_len here. When called from
2227 * tcp_output(), the underlying ip_len member has not yet been set.
2228 */
2229 ippseudo.ippseudo_src = ipovly->ih_src;
2230 ippseudo.ippseudo_dst = ipovly->ih_dst;
2231 ippseudo.ippseudo_pad = 0;
2232 ippseudo.ippseudo_p = IPPROTO_TCP;
2233 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2234 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2235
2236 /*
2237 * Step 2: Update MD5 hash with TCP header, excluding options.
2238 * The TCP checksum must be set to zero.
2239 */
2240 savecsum = th->th_sum;
2241 th->th_sum = 0;
2242 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2243 th->th_sum = savecsum;
2244
2245 /*
2246 * Step 3: Update MD5 hash with TCP segment data.
2247 * Use m_apply() to avoid an early m_pullup().
2248 */
2249 if (len > 0)
2250 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2251
2252 /*
2253 * Step 4: Update MD5 hash with shared secret.
2254 */
2255 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2256 MD5Final(buf, &ctx);
2257
2258 key_sa_recordxfer(sav, m);
2259 KEY_FREESAV(&sav);
2260 return (0);
2261}
2262#endif /* TCP_SIGNATURE */
2263
2264static int
2265sysctl_drop(SYSCTL_HANDLER_ARGS)
2266{
2267 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2268 struct sockaddr_storage addrs[2];
2269 struct inpcb *inp;
2270 struct tcpcb *tp;
2271 struct tcptw *tw;
2272 struct sockaddr_in *fin, *lin;
2273#ifdef INET6
2274 struct sockaddr_in6 *fin6, *lin6;
2275 struct in6_addr f6, l6;
2276#endif
2277 int error;
2278
2279 inp = NULL;
2280 fin = lin = NULL;
2281#ifdef INET6
2282 fin6 = lin6 = NULL;
2283#endif
2284 error = 0;
2285
2286 if (req->oldptr != NULL || req->oldlen != 0)
2287 return (EINVAL);
2288 if (req->newptr == NULL)
2289 return (EPERM);
2290 if (req->newlen < sizeof(addrs))
2291 return (ENOMEM);
2292 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2293 if (error)
2294 return (error);
2295
2296 switch (addrs[0].ss_family) {
2297#ifdef INET6
2298 case AF_INET6:
2299 fin6 = (struct sockaddr_in6 *)&addrs[0];
2300 lin6 = (struct sockaddr_in6 *)&addrs[1];
2301 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2302 lin6->sin6_len != sizeof(struct sockaddr_in6))
2303 return (EINVAL);
2304 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2305 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2306 return (EINVAL);
2307 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2308 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2309 fin = (struct sockaddr_in *)&addrs[0];
2310 lin = (struct sockaddr_in *)&addrs[1];
2311 break;
2312 }
2313 error = sa6_embedscope(fin6, ip6_use_defzone);
2314 if (error)
2315 return (error);
2316 error = sa6_embedscope(lin6, ip6_use_defzone);
2317 if (error)
2318 return (error);
2319 break;
2320#endif
2321 case AF_INET:
2322 fin = (struct sockaddr_in *)&addrs[0];
2323 lin = (struct sockaddr_in *)&addrs[1];
2324 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2325 lin->sin_len != sizeof(struct sockaddr_in))
2326 return (EINVAL);
2327 break;
2328 default:
2329 return (EINVAL);
2330 }
2331 INP_INFO_WLOCK(&tcbinfo);
2332 switch (addrs[0].ss_family) {
2333#ifdef INET6
2334 case AF_INET6:
2335 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2336 &l6, lin6->sin6_port, 0, NULL);
2337 break;
2338#endif
2339 case AF_INET:
2340 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2341 lin->sin_addr, lin->sin_port, 0, NULL);
2342 break;
2343 }
2344 if (inp != NULL) {
2345 INP_LOCK(inp);
2346 if (inp->inp_vflag & INP_TIMEWAIT) {
2347 /*
2348 * XXXRW: There currently exists a state where an
2349 * inpcb is present, but its timewait state has been
2350 * discarded. For now, don't allow dropping of this
2351 * type of inpcb.
2352 */
2353 tw = intotw(inp);
2354 if (tw != NULL)
2355 tcp_twclose(tw, 0);
2356 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2357 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2358 tp = intotcpcb(inp);
2359 tcp_drop(tp, ECONNABORTED);
2360 }
2361 INP_UNLOCK(inp);
2362 } else
2363 error = ESRCH;
2364 INP_INFO_WUNLOCK(&tcbinfo);
2365 return (error);
2366}
2367
2368SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2369 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2370 0, sysctl_drop, "", "Drop TCP connection");