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 157478 2006-04-04 14:31:37Z glebius $
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 */
251void
252tcp_init(void)
253{
254 int hashsize = TCBHASHSIZE;
255
256 tcp_delacktime = TCPTV_DELACK;
257 tcp_keepinit = TCPTV_KEEP_INIT;
258 tcp_keepidle = TCPTV_KEEP_IDLE;
259 tcp_keepintvl = TCPTV_KEEPINTVL;
260 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
261 tcp_msl = TCPTV_MSL;
262 tcp_rexmit_min = TCPTV_MIN;
263 tcp_rexmit_slop = TCPTV_CPU_VAR;
264 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
265
266 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
267 LIST_INIT(&tcb);
268 tcbinfo.listhead = &tcb;
269 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
270 if (!powerof2(hashsize)) {
271 printf("WARNING: TCB hash size not a power of 2\n");
272 hashsize = 512; /* safe default */
273 }
274 tcp_tcbhashsize = hashsize;
275 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
276 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
277 &tcbinfo.porthashmask);
278 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
279 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
280 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
281#ifdef INET6
282#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
283#else /* INET6 */
284#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
285#endif /* INET6 */
286 if (max_protohdr < TCP_MINPROTOHDR)
287 max_protohdr = TCP_MINPROTOHDR;
288 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
289 panic("tcp_init");
290#undef TCP_MINPROTOHDR
291 /*
292 * These have to be type stable for the benefit of the timers.
293 */
294 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
295 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
296 uma_zone_set_max(tcpcb_zone, maxsockets);
297 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
298 if (maxtcptw == 0)
299 maxtcptw = maxsockets / 5;
300 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
301 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
302 uma_zone_set_max(tcptw_zone, maxtcptw);
303 tcp_timer_init();
304 syncache_init();
305 tcp_hc_init();
306 tcp_reass_init();
307 callout_init(&isn_callout, CALLOUT_MPSAFE);
308 tcp_isn_tick(NULL);
309 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
310 SHUTDOWN_PRI_DEFAULT);
311 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
312 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313}
314
315void
316tcp_fini(void *xtp)
317{
318
319 callout_stop(&isn_callout);
320}
321
322/*
323 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
324 * tcp_template used to store this data in mbufs, but we now recopy it out
325 * of the tcpcb each time to conserve mbufs.
326 */
327void
328tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
329{
330 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
331
332 INP_LOCK_ASSERT(inp);
333
334#ifdef INET6
335 if ((inp->inp_vflag & INP_IPV6) != 0) {
336 struct ip6_hdr *ip6;
337
338 ip6 = (struct ip6_hdr *)ip_ptr;
339 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
340 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
341 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
342 (IPV6_VERSION & IPV6_VERSION_MASK);
343 ip6->ip6_nxt = IPPROTO_TCP;
344 ip6->ip6_plen = sizeof(struct tcphdr);
345 ip6->ip6_src = inp->in6p_laddr;
346 ip6->ip6_dst = inp->in6p_faddr;
347 } else
348#endif
349 {
350 struct ip *ip;
351
352 ip = (struct ip *)ip_ptr;
353 ip->ip_v = IPVERSION;
354 ip->ip_hl = 5;
355 ip->ip_tos = inp->inp_ip_tos;
356 ip->ip_len = 0;
357 ip->ip_id = 0;
358 ip->ip_off = 0;
359 ip->ip_ttl = inp->inp_ip_ttl;
360 ip->ip_sum = 0;
361 ip->ip_p = IPPROTO_TCP;
362 ip->ip_src = inp->inp_laddr;
363 ip->ip_dst = inp->inp_faddr;
364 }
365 th->th_sport = inp->inp_lport;
366 th->th_dport = inp->inp_fport;
367 th->th_seq = 0;
368 th->th_ack = 0;
369 th->th_x2 = 0;
370 th->th_off = 5;
371 th->th_flags = 0;
372 th->th_win = 0;
373 th->th_urp = 0;
374 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
375}
376
377/*
378 * Create template to be used to send tcp packets on a connection.
379 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
380 * use for this function is in keepalives, which use tcp_respond.
381 */
382struct tcptemp *
383tcpip_maketemplate(struct inpcb *inp)
384{
385 struct mbuf *m;
386 struct tcptemp *n;
387
388 m = m_get(M_DONTWAIT, MT_DATA);
389 if (m == NULL)
390 return (0);
391 m->m_len = sizeof(struct tcptemp);
392 n = mtod(m, struct tcptemp *);
393
394 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
395 return (n);
396}
397
398/*
399 * Send a single message to the TCP at address specified by
400 * the given TCP/IP header. If m == NULL, then we make a copy
401 * of the tcpiphdr at ti and send directly to the addressed host.
402 * This is used to force keep alive messages out using the TCP
403 * template for a connection. If flags are given then we send
404 * a message back to the TCP which originated the * segment ti,
405 * and discard the mbuf containing it and any other attached mbufs.
406 *
407 * In any case the ack and sequence number of the transmitted
408 * segment are as specified by the parameters.
409 *
410 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
411 */
412void
413tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
414 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
415{
416 register int tlen;
417 int win = 0;
418 struct ip *ip;
419 struct tcphdr *nth;
420#ifdef INET6
421 struct ip6_hdr *ip6;
422 int isipv6;
423#endif /* INET6 */
424 int ipflags = 0;
425 struct inpcb *inp;
426
427 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
428
429#ifdef INET6
430 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
431 ip6 = ipgen;
432#endif /* INET6 */
433 ip = ipgen;
434
435 if (tp != NULL) {
436 inp = tp->t_inpcb;
437 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
438 INP_INFO_WLOCK_ASSERT(&tcbinfo);
439 INP_LOCK_ASSERT(inp);
440 } else
441 inp = NULL;
442
443 if (tp != NULL) {
444 if (!(flags & TH_RST)) {
445 win = sbspace(&inp->inp_socket->so_rcv);
446 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
447 win = (long)TCP_MAXWIN << tp->rcv_scale;
448 }
449 }
450 if (m == NULL) {
451 m = m_gethdr(M_DONTWAIT, MT_DATA);
452 if (m == NULL)
453 return;
454 tlen = 0;
455 m->m_data += max_linkhdr;
456#ifdef INET6
457 if (isipv6) {
458 bcopy((caddr_t)ip6, mtod(m, caddr_t),
459 sizeof(struct ip6_hdr));
460 ip6 = mtod(m, struct ip6_hdr *);
461 nth = (struct tcphdr *)(ip6 + 1);
462 } else
463#endif /* INET6 */
464 {
465 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
466 ip = mtod(m, struct ip *);
467 nth = (struct tcphdr *)(ip + 1);
468 }
469 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
470 flags = TH_ACK;
471 } else {
472 m_freem(m->m_next);
473 m->m_next = NULL;
474 m->m_data = (caddr_t)ipgen;
475 /* m_len is set later */
476 tlen = 0;
477#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
478#ifdef INET6
479 if (isipv6) {
480 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
481 nth = (struct tcphdr *)(ip6 + 1);
482 } else
483#endif /* INET6 */
484 {
485 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
486 nth = (struct tcphdr *)(ip + 1);
487 }
488 if (th != nth) {
489 /*
490 * this is usually a case when an extension header
491 * exists between the IPv6 header and the
492 * TCP header.
493 */
494 nth->th_sport = th->th_sport;
495 nth->th_dport = th->th_dport;
496 }
497 xchg(nth->th_dport, nth->th_sport, n_short);
498#undef xchg
499 }
500#ifdef INET6
501 if (isipv6) {
502 ip6->ip6_flow = 0;
503 ip6->ip6_vfc = IPV6_VERSION;
504 ip6->ip6_nxt = IPPROTO_TCP;
505 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
506 tlen));
507 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
508 } else
509#endif
510 {
511 tlen += sizeof (struct tcpiphdr);
512 ip->ip_len = tlen;
513 ip->ip_ttl = ip_defttl;
514 if (path_mtu_discovery)
515 ip->ip_off |= IP_DF;
516 }
517 m->m_len = tlen;
518 m->m_pkthdr.len = tlen;
519 m->m_pkthdr.rcvif = NULL;
520#ifdef MAC
521 if (inp != NULL) {
522 /*
523 * Packet is associated with a socket, so allow the
524 * label of the response to reflect the socket label.
525 */
526 INP_LOCK_ASSERT(inp);
527 mac_create_mbuf_from_inpcb(inp, m);
528 } else {
529 /*
530 * Packet is not associated with a socket, so possibly
531 * update the label in place.
532 */
533 mac_reflect_mbuf_tcp(m);
534 }
535#endif
536 nth->th_seq = htonl(seq);
537 nth->th_ack = htonl(ack);
538 nth->th_x2 = 0;
539 nth->th_off = sizeof (struct tcphdr) >> 2;
540 nth->th_flags = flags;
541 if (tp != NULL)
542 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
543 else
544 nth->th_win = htons((u_short)win);
545 nth->th_urp = 0;
546#ifdef INET6
547 if (isipv6) {
548 nth->th_sum = 0;
549 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
550 sizeof(struct ip6_hdr),
551 tlen - sizeof(struct ip6_hdr));
552 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
553 NULL, NULL);
554 } else
555#endif /* INET6 */
556 {
557 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
558 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
559 m->m_pkthdr.csum_flags = CSUM_TCP;
560 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
561 }
562#ifdef TCPDEBUG
563 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
564 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
565#endif
566#ifdef INET6
567 if (isipv6)
568 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
569 else
570#endif /* INET6 */
571 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
572}
573
574/*
575 * Create a new TCP control block, making an
576 * empty reassembly queue and hooking it to the argument
577 * protocol control block. The `inp' parameter must have
578 * come from the zone allocator set up in tcp_init().
579 */
580struct tcpcb *
581tcp_newtcpcb(struct inpcb *inp)
582{
583 struct tcpcb_mem *tm;
584 struct tcpcb *tp;
585#ifdef INET6
586 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
587#endif /* INET6 */
588
589 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
590 if (tm == NULL)
591 return (NULL);
592 tp = &tm->tcb;
593 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
594 tp->t_maxseg = tp->t_maxopd =
595#ifdef INET6
596 isipv6 ? tcp_v6mssdflt :
597#endif /* INET6 */
598 tcp_mssdflt;
599
600 /* Set up our timeouts. */
601 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
602 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
603 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
604 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
605 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
606
607 if (tcp_do_rfc1323)
608 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
609 tp->sack_enable = tcp_do_sack;
610 TAILQ_INIT(&tp->snd_holes);
611 tp->t_inpcb = inp; /* XXX */
612 /*
613 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
614 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
615 * reasonable initial retransmit time.
616 */
617 tp->t_srtt = TCPTV_SRTTBASE;
618 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
619 tp->t_rttmin = tcp_rexmit_min;
620 tp->t_rxtcur = TCPTV_RTOBASE;
621 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
622 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
623 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
624 tp->t_rcvtime = ticks;
625 tp->t_bw_rtttime = ticks;
626 /*
627 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
628 * because the socket may be bound to an IPv6 wildcard address,
629 * which may match an IPv4-mapped IPv6 address.
630 */
631 inp->inp_ip_ttl = ip_defttl;
632 inp->inp_ppcb = tp;
633 return (tp); /* XXX */
634}
635
636/*
637 * Drop a TCP connection, reporting
638 * the specified error. If connection is synchronized,
639 * then send a RST to peer.
640 */
641struct tcpcb *
642tcp_drop(struct tcpcb *tp, int errno)
643{
644 struct socket *so = tp->t_inpcb->inp_socket;
645
646 INP_INFO_WLOCK_ASSERT(&tcbinfo);
647 INP_LOCK_ASSERT(tp->t_inpcb);
648
649 if (TCPS_HAVERCVDSYN(tp->t_state)) {
650 tp->t_state = TCPS_CLOSED;
651 (void) tcp_output(tp);
652 tcpstat.tcps_drops++;
653 } else
654 tcpstat.tcps_conndrops++;
655 if (errno == ETIMEDOUT && tp->t_softerror)
656 errno = tp->t_softerror;
657 so->so_error = errno;
658 return (tcp_close(tp));
659}
660
661void
662tcp_discardcb(struct tcpcb *tp)
663{
664 struct tseg_qent *q;
665 struct inpcb *inp = tp->t_inpcb;
666 struct socket *so = inp->inp_socket;
667#ifdef INET6
668 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
669#endif /* INET6 */
670
671 /*
672 * XXXRW: This is all very well and good, but actually, we might be
673 * discarding the tcpcb after the socket is gone, so we can't do
674 * this:
675 KASSERT(so != NULL, ("tcp_discardcb: so == NULL"));
676 */
677 INP_LOCK_ASSERT(inp);
678
679 /*
680 * Make sure that all of our timers are stopped before we
681 * delete the PCB.
682 */
683 callout_stop(tp->tt_rexmt);
684 callout_stop(tp->tt_persist);
685 callout_stop(tp->tt_keep);
686 callout_stop(tp->tt_2msl);
687 callout_stop(tp->tt_delack);
688
689 /*
690 * If we got enough samples through the srtt filter,
691 * save the rtt and rttvar in the routing entry.
692 * 'Enough' is arbitrarily defined as 4 rtt samples.
693 * 4 samples is enough for the srtt filter to converge
694 * to within enough % of the correct value; fewer samples
695 * and we could save a bogus rtt. The danger is not high
696 * as tcp quickly recovers from everything.
697 * XXX: Works very well but needs some more statistics!
698 */
699 if (tp->t_rttupdated >= 4) {
700 struct hc_metrics_lite metrics;
701 u_long ssthresh;
702
703 bzero(&metrics, sizeof(metrics));
704 /*
705 * Update the ssthresh always when the conditions below
706 * are satisfied. This gives us better new start value
707 * for the congestion avoidance for new connections.
708 * ssthresh is only set if packet loss occured on a session.
709 */
710 ssthresh = tp->snd_ssthresh;
711 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
712 /*
713 * convert the limit from user data bytes to
714 * packets then to packet data bytes.
715 */
716 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
717 if (ssthresh < 2)
718 ssthresh = 2;
719 ssthresh *= (u_long)(tp->t_maxseg +
720#ifdef INET6
721 (isipv6 ? sizeof (struct ip6_hdr) +
722 sizeof (struct tcphdr) :
723#endif
724 sizeof (struct tcpiphdr)
725#ifdef INET6
726 )
727#endif
728 );
729 } else
730 ssthresh = 0;
731 metrics.rmx_ssthresh = ssthresh;
732
733 metrics.rmx_rtt = tp->t_srtt;
734 metrics.rmx_rttvar = tp->t_rttvar;
735 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
736 metrics.rmx_bandwidth = tp->snd_bandwidth;
737 metrics.rmx_cwnd = tp->snd_cwnd;
738 metrics.rmx_sendpipe = 0;
739 metrics.rmx_recvpipe = 0;
740
741 tcp_hc_update(&inp->inp_inc, &metrics);
742 }
743
744 /* free the reassembly queue, if any */
745 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
746 LIST_REMOVE(q, tqe_q);
747 m_freem(q->tqe_m);
748 uma_zfree(tcp_reass_zone, q);
749 tp->t_segqlen--;
750 tcp_reass_qsize--;
751 }
752 tcp_free_sackholes(tp);
753 inp->inp_ppcb = NULL;
754 tp->t_inpcb = NULL;
755 uma_zfree(tcpcb_zone, tp);
756
757 /*
758 * XXXRW: This seems a bit unclean.
759 */
760 if (so != NULL)
761 soisdisconnected(so);
762}
763
764/*
765 * Attempt to close a TCP control block, marking it as dropped, and freeing
766 * the socket if we hold the only reference.
767 */
768struct tcpcb *
769tcp_close(struct tcpcb *tp)
770{
771 struct inpcb *inp = tp->t_inpcb;
772 struct socket *so;
773
774 INP_INFO_WLOCK_ASSERT(&tcbinfo);
775 INP_LOCK_ASSERT(inp);
776
777 inp->inp_vflag |= INP_DROPPED;
778
779 tcpstat.tcps_closed++;
780 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
781 so = inp->inp_socket;
782 soisdisconnected(so);
783 if (inp->inp_vflag & INP_SOCKREF) {
784 KASSERT(so->so_state & SS_PROTOREF,
785 ("tcp_close: !SS_PROTOREF"));
786 inp->inp_vflag &= ~INP_SOCKREF;
787 tcp_discardcb(tp);
788#ifdef INET6
789 if (inp->inp_vflag & INP_IPV6PROTO) {
790 in6_pcbdetach(inp);
791 in6_pcbfree(inp);
792 } else {
793#endif
794 in_pcbdetach(inp);
795 in_pcbfree(inp);
796#ifdef INET6
797 }
798#endif
799 ACCEPT_LOCK();
800 SOCK_LOCK(so);
801 so->so_state &= ~SS_PROTOREF;
802 sofree(so);
803 return (NULL);
804 }
805 return (tp);
806}
807
808void
809tcp_drain(void)
810{
811
812 if (do_tcpdrain) {
813 struct inpcb *inpb;
814 struct tcpcb *tcpb;
815 struct tseg_qent *te;
816
817 /*
818 * Walk the tcpbs, if existing, and flush the reassembly queue,
819 * if there is one...
820 * XXX: The "Net/3" implementation doesn't imply that the TCP
821 * reassembly queue should be flushed, but in a situation
822 * where we're really low on mbufs, this is potentially
823 * usefull.
824 */
825 INP_INFO_RLOCK(&tcbinfo);
826 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
827 if (inpb->inp_vflag & INP_TIMEWAIT)
828 continue;
829 INP_LOCK(inpb);
830 if ((tcpb = intotcpcb(inpb)) != NULL) {
831 while ((te = LIST_FIRST(&tcpb->t_segq))
832 != NULL) {
833 LIST_REMOVE(te, tqe_q);
834 m_freem(te->tqe_m);
835 uma_zfree(tcp_reass_zone, te);
836 tcpb->t_segqlen--;
837 tcp_reass_qsize--;
838 }
839 tcp_clean_sackreport(tcpb);
840 }
841 INP_UNLOCK(inpb);
842 }
843 INP_INFO_RUNLOCK(&tcbinfo);
844 }
845}
846
847/*
848 * Notify a tcp user of an asynchronous error;
849 * store error as soft error, but wake up user
850 * (for now, won't do anything until can select for soft error).
851 *
852 * Do not wake up user since there currently is no mechanism for
853 * reporting soft errors (yet - a kqueue filter may be added).
854 */
855static struct inpcb *
856tcp_notify(struct inpcb *inp, int error)
857{
858 struct tcpcb *tp;
859
860 INP_INFO_WLOCK_ASSERT(&tcbinfo);
861 INP_LOCK_ASSERT(inp);
862
863 if ((inp->inp_vflag & INP_TIMEWAIT) ||
864 (inp->inp_vflag & INP_DROPPED))
865 return (inp);
866
867 tp = intotcpcb(inp);
868 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
869
870 /*
871 * Ignore some errors if we are hooked up.
872 * If connection hasn't completed, has retransmitted several times,
873 * and receives a second error, give up now. This is better
874 * than waiting a long time to establish a connection that
875 * can never complete.
876 */
877 if (tp->t_state == TCPS_ESTABLISHED &&
878 (error == EHOSTUNREACH || error == ENETUNREACH ||
879 error == EHOSTDOWN)) {
880 return (inp);
881 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
882 tp->t_softerror) {
883 tp = tcp_drop(tp, error);
884 if (tp != NULL)
885 return (inp);
886 else
887 return (NULL);
888 } else {
889 tp->t_softerror = error;
890 return (inp);
891 }
892#if 0
893 wakeup( &so->so_timeo);
894 sorwakeup(so);
895 sowwakeup(so);
896#endif
897}
898
899static int
900tcp_pcblist(SYSCTL_HANDLER_ARGS)
901{
902 int error, i, n;
903 struct inpcb *inp, **inp_list;
904 inp_gen_t gencnt;
905 struct xinpgen xig;
906
907 /*
908 * The process of preparing the TCB list is too time-consuming and
909 * resource-intensive to repeat twice on every request.
910 */
911 if (req->oldptr == NULL) {
912 n = tcbinfo.ipi_count;
913 req->oldidx = 2 * (sizeof xig)
914 + (n + n/8) * sizeof(struct xtcpcb);
915 return (0);
916 }
917
918 if (req->newptr != NULL)
919 return (EPERM);
920
921 /*
922 * OK, now we're committed to doing something.
923 */
924 INP_INFO_RLOCK(&tcbinfo);
925 gencnt = tcbinfo.ipi_gencnt;
926 n = tcbinfo.ipi_count;
927 INP_INFO_RUNLOCK(&tcbinfo);
928
929 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
930 + n * sizeof(struct xtcpcb));
931 if (error != 0)
932 return (error);
933
934 xig.xig_len = sizeof xig;
935 xig.xig_count = n;
936 xig.xig_gen = gencnt;
937 xig.xig_sogen = so_gencnt;
938 error = SYSCTL_OUT(req, &xig, sizeof xig);
939 if (error)
940 return (error);
941
942 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
943 if (inp_list == NULL)
944 return (ENOMEM);
945
946 INP_INFO_RLOCK(&tcbinfo);
947 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
948 inp = LIST_NEXT(inp, inp_list)) {
949 INP_LOCK(inp);
950 if (inp->inp_gencnt <= gencnt) {
951 /*
952 * XXX: This use of cr_cansee(), introduced with
953 * TCP state changes, is not quite right, but for
954 * now, better than nothing.
955 */
956 if (inp->inp_vflag & INP_TIMEWAIT) {
957 if (intotw(inp) != NULL)
958 error = cr_cansee(req->td->td_ucred,
959 intotw(inp)->tw_cred);
960 else
961 error = EINVAL; /* Skip this inp. */
962 } else
963 error = cr_canseesocket(req->td->td_ucred,
964 inp->inp_socket);
965 if (error == 0)
966 inp_list[i++] = inp;
967 }
968 INP_UNLOCK(inp);
969 }
970 INP_INFO_RUNLOCK(&tcbinfo);
971 n = i;
972
973 error = 0;
974 for (i = 0; i < n; i++) {
975 inp = inp_list[i];
976 if (inp->inp_gencnt <= gencnt) {
977 struct xtcpcb xt;
978 void *inp_ppcb;
979
980 bzero(&xt, sizeof(xt));
981 xt.xt_len = sizeof xt;
982 /* XXX should avoid extra copy */
983 bcopy(inp, &xt.xt_inp, sizeof *inp);
984 inp_ppcb = inp->inp_ppcb;
985 if (inp_ppcb == NULL)
986 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
987 else if (inp->inp_vflag & INP_TIMEWAIT) {
988 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
989 xt.xt_tp.t_state = TCPS_TIME_WAIT;
990 } else
991 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
992 if (inp->inp_socket != NULL)
993 sotoxsocket(inp->inp_socket, &xt.xt_socket);
994 else {
995 bzero(&xt.xt_socket, sizeof xt.xt_socket);
996 xt.xt_socket.xso_protocol = IPPROTO_TCP;
997 }
998 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
999 error = SYSCTL_OUT(req, &xt, sizeof xt);
1000 }
1001 }
1002 if (!error) {
1003 /*
1004 * Give the user an updated idea of our state.
1005 * If the generation differs from what we told
1006 * her before, she knows that something happened
1007 * while we were processing this request, and it
1008 * might be necessary to retry.
1009 */
1010 INP_INFO_RLOCK(&tcbinfo);
1011 xig.xig_gen = tcbinfo.ipi_gencnt;
1012 xig.xig_sogen = so_gencnt;
1013 xig.xig_count = tcbinfo.ipi_count;
1014 INP_INFO_RUNLOCK(&tcbinfo);
1015 error = SYSCTL_OUT(req, &xig, sizeof xig);
1016 }
1017 free(inp_list, M_TEMP);
1018 return (error);
1019}
1020
1021SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1022 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1023
1024static int
1025tcp_getcred(SYSCTL_HANDLER_ARGS)
1026{
1027 struct xucred xuc;
1028 struct sockaddr_in addrs[2];
1029 struct inpcb *inp;
1030 int error;
1031
1032 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1033 if (error)
1034 return (error);
1035 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1036 if (error)
1037 return (error);
1038 INP_INFO_RLOCK(&tcbinfo);
1039 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1040 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1041 if (inp == NULL) {
1042 error = ENOENT;
1043 goto outunlocked;
1044 }
1045 INP_LOCK(inp);
1046 if (inp->inp_socket == NULL) {
1047 error = ENOENT;
1048 goto out;
1049 }
1050 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1051 if (error)
1052 goto out;
1053 cru2x(inp->inp_socket->so_cred, &xuc);
1054out:
1055 INP_UNLOCK(inp);
1056outunlocked:
1057 INP_INFO_RUNLOCK(&tcbinfo);
1058 if (error == 0)
1059 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1060 return (error);
1061}
1062
1063SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1064 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1065 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1066
1067#ifdef INET6
1068static int
1069tcp6_getcred(SYSCTL_HANDLER_ARGS)
1070{
1071 struct xucred xuc;
1072 struct sockaddr_in6 addrs[2];
1073 struct inpcb *inp;
1074 int error, mapped = 0;
1075
1076 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1077 if (error)
1078 return (error);
1079 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1080 if (error)
1081 return (error);
1082 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1083 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1084 return (error);
1085 }
1086 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1087 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1088 mapped = 1;
1089 else
1090 return (EINVAL);
1091 }
1092
1093 INP_INFO_RLOCK(&tcbinfo);
1094 if (mapped == 1)
1095 inp = in_pcblookup_hash(&tcbinfo,
1096 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1097 addrs[1].sin6_port,
1098 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1099 addrs[0].sin6_port,
1100 0, NULL);
1101 else
1102 inp = in6_pcblookup_hash(&tcbinfo,
1103 &addrs[1].sin6_addr, addrs[1].sin6_port,
1104 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1105 if (inp == NULL) {
1106 error = ENOENT;
1107 goto outunlocked;
1108 }
1109 INP_LOCK(inp);
1110 if (inp->inp_socket == NULL) {
1111 error = ENOENT;
1112 goto out;
1113 }
1114 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1115 if (error)
1116 goto out;
1117 cru2x(inp->inp_socket->so_cred, &xuc);
1118out:
1119 INP_UNLOCK(inp);
1120outunlocked:
1121 INP_INFO_RUNLOCK(&tcbinfo);
1122 if (error == 0)
1123 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1124 return (error);
1125}
1126
1127SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1128 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1129 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1130#endif
1131
1132
1133void
1134tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1135{
1136 struct ip *ip = vip;
1137 struct tcphdr *th;
1138 struct in_addr faddr;
1139 struct inpcb *inp;
1140 struct tcpcb *tp;
1141 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1142 struct icmp *icp;
1143 struct in_conninfo inc;
1144 tcp_seq icmp_tcp_seq;
1145 int mtu;
1146
1147 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1148 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1149 return;
1150
1151 if (cmd == PRC_MSGSIZE)
1152 notify = tcp_mtudisc;
1153 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1154 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1155 notify = tcp_drop_syn_sent;
1156 /*
1157 * Redirects don't need to be handled up here.
1158 */
1159 else if (PRC_IS_REDIRECT(cmd))
1160 return;
1161 /*
1162 * Source quench is depreciated.
1163 */
1164 else if (cmd == PRC_QUENCH)
1165 return;
1166 /*
1167 * Hostdead is ugly because it goes linearly through all PCBs.
1168 * XXX: We never get this from ICMP, otherwise it makes an
1169 * excellent DoS attack on machines with many connections.
1170 */
1171 else if (cmd == PRC_HOSTDEAD)
1172 ip = NULL;
1173 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1174 return;
1175 if (ip != NULL) {
1176 icp = (struct icmp *)((caddr_t)ip
1177 - offsetof(struct icmp, icmp_ip));
1178 th = (struct tcphdr *)((caddr_t)ip
1179 + (ip->ip_hl << 2));
1180 INP_INFO_WLOCK(&tcbinfo);
1181 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1182 ip->ip_src, th->th_sport, 0, NULL);
1183 if (inp != NULL) {
1184 INP_LOCK(inp);
1185 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1186 !(inp->inp_vflag & INP_DROPPED) &&
1187 !(inp->inp_socket == NULL)) {
1188 icmp_tcp_seq = htonl(th->th_seq);
1189 tp = intotcpcb(inp);
1190 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1191 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1192 if (cmd == PRC_MSGSIZE) {
1193 /*
1194 * MTU discovery:
1195 * If we got a needfrag set the MTU
1196 * in the route to the suggested new
1197 * value (if given) and then notify.
1198 */
1199 bzero(&inc, sizeof(inc));
1200 inc.inc_flags = 0; /* IPv4 */
1201 inc.inc_faddr = faddr;
1202
1203 mtu = ntohs(icp->icmp_nextmtu);
1204 /*
1205 * If no alternative MTU was
1206 * proposed, try the next smaller
1207 * one. ip->ip_len has already
1208 * been swapped in icmp_input().
1209 */
1210 if (!mtu)
1211 mtu = ip_next_mtu(ip->ip_len,
1212 1);
1213 if (mtu < max(296, (tcp_minmss)
1214 + sizeof(struct tcpiphdr)))
1215 mtu = 0;
1216 if (!mtu)
1217 mtu = tcp_mssdflt
1218 + sizeof(struct tcpiphdr);
1219 /*
1220 * Only cache the the MTU if it
1221 * is smaller than the interface
1222 * or route MTU. tcp_mtudisc()
1223 * will do right thing by itself.
1224 */
1225 if (mtu <= tcp_maxmtu(&inc))
1226 tcp_hc_updatemtu(&inc, mtu);
1227 }
1228
1229 inp = (*notify)(inp, inetctlerrmap[cmd]);
1230 }
1231 }
1232 if (inp != NULL)
1233 INP_UNLOCK(inp);
1234 } else {
1235 inc.inc_fport = th->th_dport;
1236 inc.inc_lport = th->th_sport;
1237 inc.inc_faddr = faddr;
1238 inc.inc_laddr = ip->ip_src;
1239#ifdef INET6
1240 inc.inc_isipv6 = 0;
1241#endif
1242 syncache_unreach(&inc, th);
1243 }
1244 INP_INFO_WUNLOCK(&tcbinfo);
1245 } else
1246 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1247}
1248
1249#ifdef INET6
1250void
1251tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1252{
1253 struct tcphdr th;
1254 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1255 struct ip6_hdr *ip6;
1256 struct mbuf *m;
1257 struct ip6ctlparam *ip6cp = NULL;
1258 const struct sockaddr_in6 *sa6_src = NULL;
1259 int off;
1260 struct tcp_portonly {
1261 u_int16_t th_sport;
1262 u_int16_t th_dport;
1263 } *thp;
1264
1265 if (sa->sa_family != AF_INET6 ||
1266 sa->sa_len != sizeof(struct sockaddr_in6))
1267 return;
1268
1269 if (cmd == PRC_MSGSIZE)
1270 notify = tcp_mtudisc;
1271 else if (!PRC_IS_REDIRECT(cmd) &&
1272 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1273 return;
1274 /* Source quench is depreciated. */
1275 else if (cmd == PRC_QUENCH)
1276 return;
1277
1278 /* if the parameter is from icmp6, decode it. */
1279 if (d != NULL) {
1280 ip6cp = (struct ip6ctlparam *)d;
1281 m = ip6cp->ip6c_m;
1282 ip6 = ip6cp->ip6c_ip6;
1283 off = ip6cp->ip6c_off;
1284 sa6_src = ip6cp->ip6c_src;
1285 } else {
1286 m = NULL;
1287 ip6 = NULL;
1288 off = 0; /* fool gcc */
1289 sa6_src = &sa6_any;
1290 }
1291
1292 if (ip6 != NULL) {
1293 struct in_conninfo inc;
1294 /*
1295 * XXX: We assume that when IPV6 is non NULL,
1296 * M and OFF are valid.
1297 */
1298
1299 /* check if we can safely examine src and dst ports */
1300 if (m->m_pkthdr.len < off + sizeof(*thp))
1301 return;
1302
1303 bzero(&th, sizeof(th));
1304 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1305
1306 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1307 (struct sockaddr *)ip6cp->ip6c_src,
1308 th.th_sport, cmd, NULL, notify);
1309
1310 inc.inc_fport = th.th_dport;
1311 inc.inc_lport = th.th_sport;
1312 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1313 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1314 inc.inc_isipv6 = 1;
1315 INP_INFO_WLOCK(&tcbinfo);
1316 syncache_unreach(&inc, &th);
1317 INP_INFO_WUNLOCK(&tcbinfo);
1318 } else
1319 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1320 0, cmd, NULL, notify);
1321}
1322#endif /* INET6 */
1323
1324
1325/*
1326 * Following is where TCP initial sequence number generation occurs.
1327 *
1328 * There are two places where we must use initial sequence numbers:
1329 * 1. In SYN-ACK packets.
1330 * 2. In SYN packets.
1331 *
1332 * All ISNs for SYN-ACK packets are generated by the syncache. See
1333 * tcp_syncache.c for details.
1334 *
1335 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1336 * depends on this property. In addition, these ISNs should be
1337 * unguessable so as to prevent connection hijacking. To satisfy
1338 * the requirements of this situation, the algorithm outlined in
1339 * RFC 1948 is used, with only small modifications.
1340 *
1341 * Implementation details:
1342 *
1343 * Time is based off the system timer, and is corrected so that it
1344 * increases by one megabyte per second. This allows for proper
1345 * recycling on high speed LANs while still leaving over an hour
1346 * before rollover.
1347 *
1348 * As reading the *exact* system time is too expensive to be done
1349 * whenever setting up a TCP connection, we increment the time
1350 * offset in two ways. First, a small random positive increment
1351 * is added to isn_offset for each connection that is set up.
1352 * Second, the function tcp_isn_tick fires once per clock tick
1353 * and increments isn_offset as necessary so that sequence numbers
1354 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1355 * random positive increments serve only to ensure that the same
1356 * exact sequence number is never sent out twice (as could otherwise
1357 * happen when a port is recycled in less than the system tick
1358 * interval.)
1359 *
1360 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1361 * between seeding of isn_secret. This is normally set to zero,
1362 * as reseeding should not be necessary.
1363 *
1364 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1365 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1366 * general, this means holding an exclusive (write) lock.
1367 */
1368
1369#define ISN_BYTES_PER_SECOND 1048576
1370#define ISN_STATIC_INCREMENT 4096
1371#define ISN_RANDOM_INCREMENT (4096 - 1)
1372
1373static u_char isn_secret[32];
1374static int isn_last_reseed;
1375static u_int32_t isn_offset, isn_offset_old;
1376static MD5_CTX isn_ctx;
1377
1378tcp_seq
1379tcp_new_isn(struct tcpcb *tp)
1380{
1381 u_int32_t md5_buffer[4];
1382 tcp_seq new_isn;
1383
1384 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1385 INP_LOCK_ASSERT(tp->t_inpcb);
1386
1387 /* Seed if this is the first use, reseed if requested. */
1388 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1389 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1390 < (u_int)ticks))) {
1391 read_random(&isn_secret, sizeof(isn_secret));
1392 isn_last_reseed = ticks;
1393 }
1394
1395 /* Compute the md5 hash and return the ISN. */
1396 MD5Init(&isn_ctx);
1397 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1398 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1399#ifdef INET6
1400 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1401 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1402 sizeof(struct in6_addr));
1403 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1404 sizeof(struct in6_addr));
1405 } else
1406#endif
1407 {
1408 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1409 sizeof(struct in_addr));
1410 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1411 sizeof(struct in_addr));
1412 }
1413 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1414 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1415 new_isn = (tcp_seq) md5_buffer[0];
1416 isn_offset += ISN_STATIC_INCREMENT +
1417 (arc4random() & ISN_RANDOM_INCREMENT);
1418 new_isn += isn_offset;
1419 return (new_isn);
1420}
1421
1422/*
1423 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1424 * to keep time flowing at a relatively constant rate. If the random
1425 * increments have already pushed us past the projected offset, do nothing.
1426 */
1427static void
1428tcp_isn_tick(void *xtp)
1429{
1430 u_int32_t projected_offset;
1431
1432 INP_INFO_WLOCK(&tcbinfo);
1433 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1434
1435 if (projected_offset > isn_offset)
1436 isn_offset = projected_offset;
1437
1438 isn_offset_old = isn_offset;
1439 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1440 INP_INFO_WUNLOCK(&tcbinfo);
1441}
1442
1443/*
1444 * When a specific ICMP unreachable message is received and the
1445 * connection state is SYN-SENT, drop the connection. This behavior
1446 * is controlled by the icmp_may_rst sysctl.
1447 */
1448struct inpcb *
1449tcp_drop_syn_sent(struct inpcb *inp, int errno)
1450{
1451 struct tcpcb *tp;
1452
1453 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1454 INP_LOCK_ASSERT(inp);
1455
1456 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1457 (inp->inp_vflag & INP_DROPPED))
1458 return (inp);
1459
1460 tp = intotcpcb(inp);
1461 if (tp->t_state != TCPS_SYN_SENT)
1462 return (inp);
1463
1464 tp = tcp_drop(tp, errno);
1465 if (tp != NULL)
1466 return (inp);
1467 else
1468 return (NULL);
1469}
1470
1471/*
1472 * When `need fragmentation' ICMP is received, update our idea of the MSS
1473 * based on the new value in the route. Also nudge TCP to send something,
1474 * since we know the packet we just sent was dropped.
1475 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1476 */
1477struct inpcb *
1478tcp_mtudisc(struct inpcb *inp, int errno)
1479{
1480 struct tcpcb *tp;
1481 struct socket *so = inp->inp_socket;
1482 u_int maxmtu;
1483 u_int romtu;
1484 int mss;
1485#ifdef INET6
1486 int isipv6;
1487#endif /* INET6 */
1488
1489 INP_LOCK_ASSERT(inp);
1490 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1491 (inp->inp_vflag & INP_DROPPED))
1492 return (inp);
1493
1494 tp = intotcpcb(inp);
1495 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1496
1497#ifdef INET6
1498 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1499#endif
1500 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1501 romtu =
1502#ifdef INET6
1503 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1504#endif /* INET6 */
1505 tcp_maxmtu(&inp->inp_inc);
1506 if (!maxmtu)
1507 maxmtu = romtu;
1508 else
1509 maxmtu = min(maxmtu, romtu);
1510 if (!maxmtu) {
1511 tp->t_maxopd = tp->t_maxseg =
1512#ifdef INET6
1513 isipv6 ? tcp_v6mssdflt :
1514#endif /* INET6 */
1515 tcp_mssdflt;
1516 return (inp);
1517 }
1518 mss = maxmtu -
1519#ifdef INET6
1520 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1521#endif /* INET6 */
1522 sizeof(struct tcpiphdr)
1523#ifdef INET6
1524 )
1525#endif /* INET6 */
1526 ;
1527
1528 /*
1529 * XXX - The above conditional probably violates the TCP
1530 * spec. The problem is that, since we don't know the
1531 * other end's MSS, we are supposed to use a conservative
1532 * default. But, if we do that, then MTU discovery will
1533 * never actually take place, because the conservative
1534 * default is much less than the MTUs typically seen
1535 * on the Internet today. For the moment, we'll sweep
1536 * this under the carpet.
1537 *
1538 * The conservative default might not actually be a problem
1539 * if the only case this occurs is when sending an initial
1540 * SYN with options and data to a host we've never talked
1541 * to before. Then, they will reply with an MSS value which
1542 * will get recorded and the new parameters should get
1543 * recomputed. For Further Study.
1544 */
1545 if (tp->t_maxopd <= mss)
1546 return (inp);
1547 tp->t_maxopd = mss;
1548
1549 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1550 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1551 mss -= TCPOLEN_TSTAMP_APPA;
1552#if (MCLBYTES & (MCLBYTES - 1)) == 0
1553 if (mss > MCLBYTES)
1554 mss &= ~(MCLBYTES-1);
1555#else
1556 if (mss > MCLBYTES)
1557 mss = mss / MCLBYTES * MCLBYTES;
1558#endif
1559 if (so->so_snd.sb_hiwat < mss)
1560 mss = so->so_snd.sb_hiwat;
1561
1562 tp->t_maxseg = mss;
1563
1564 tcpstat.tcps_mturesent++;
1565 tp->t_rtttime = 0;
1566 tp->snd_nxt = tp->snd_una;
1567 tcp_output(tp);
1568 return (inp);
1569}
1570
1571/*
1572 * Look-up the routing entry to the peer of this inpcb. If no route
1573 * is found and it cannot be allocated, then return NULL. This routine
1574 * is called by TCP routines that access the rmx structure and by tcp_mss
1575 * to get the interface MTU.
1576 */
1577u_long
1578tcp_maxmtu(struct in_conninfo *inc)
1579{
1580 struct route sro;
1581 struct sockaddr_in *dst;
1582 struct ifnet *ifp;
1583 u_long maxmtu = 0;
1584
1585 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1586
1587 bzero(&sro, sizeof(sro));
1588 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1589 dst = (struct sockaddr_in *)&sro.ro_dst;
1590 dst->sin_family = AF_INET;
1591 dst->sin_len = sizeof(*dst);
1592 dst->sin_addr = inc->inc_faddr;
1593 rtalloc_ign(&sro, RTF_CLONING);
1594 }
1595 if (sro.ro_rt != NULL) {
1596 ifp = sro.ro_rt->rt_ifp;
1597 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1598 maxmtu = ifp->if_mtu;
1599 else
1600 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1601 RTFREE(sro.ro_rt);
1602 }
1603 return (maxmtu);
1604}
1605
1606#ifdef INET6
1607u_long
1608tcp_maxmtu6(struct in_conninfo *inc)
1609{
1610 struct route_in6 sro6;
1611 struct ifnet *ifp;
1612 u_long maxmtu = 0;
1613
1614 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1615
1616 bzero(&sro6, sizeof(sro6));
1617 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1618 sro6.ro_dst.sin6_family = AF_INET6;
1619 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1620 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1621 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1622 }
1623 if (sro6.ro_rt != NULL) {
1624 ifp = sro6.ro_rt->rt_ifp;
1625 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1626 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1627 else
1628 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1629 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1630 RTFREE(sro6.ro_rt);
1631 }
1632
1633 return (maxmtu);
1634}
1635#endif /* INET6 */
1636
1637#ifdef IPSEC
1638/* compute ESP/AH header size for TCP, including outer IP header. */
1639size_t
1640ipsec_hdrsiz_tcp(struct tcpcb *tp)
1641{
1642 struct inpcb *inp;
1643 struct mbuf *m;
1644 size_t hdrsiz;
1645 struct ip *ip;
1646#ifdef INET6
1647 struct ip6_hdr *ip6;
1648#endif
1649 struct tcphdr *th;
1650
1651 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1652 return (0);
1653 MGETHDR(m, M_DONTWAIT, MT_DATA);
1654 if (!m)
1655 return (0);
1656
1657#ifdef INET6
1658 if ((inp->inp_vflag & INP_IPV6) != 0) {
1659 ip6 = mtod(m, struct ip6_hdr *);
1660 th = (struct tcphdr *)(ip6 + 1);
1661 m->m_pkthdr.len = m->m_len =
1662 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1663 tcpip_fillheaders(inp, ip6, th);
1664 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1665 } else
1666#endif /* INET6 */
1667 {
1668 ip = mtod(m, struct ip *);
1669 th = (struct tcphdr *)(ip + 1);
1670 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1671 tcpip_fillheaders(inp, ip, th);
1672 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1673 }
1674
1675 m_free(m);
1676 return (hdrsiz);
1677}
1678#endif /*IPSEC*/
1679
1680/*
1681 * Move a TCP connection into TIME_WAIT state.
1682 * tcbinfo is locked.
1683 * inp is locked, and is unlocked before returning.
1684 */
1685void
1686tcp_twstart(struct tcpcb *tp)
1687{
1688 struct tcptw *tw;
1689 struct inpcb *inp;
1690 int tw_time, acknow;
1691 struct socket *so;
1692
1693 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1694 INP_LOCK_ASSERT(tp->t_inpcb);
1695
1696 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1697 if (tw == NULL) {
1698 tw = tcp_timer_2msl_tw(1);
1699 if (tw == NULL) {
1700 tp = tcp_close(tp);
1701 if (tp != NULL)
1702 INP_UNLOCK(tp->t_inpcb);
1703 return;
1704 }
1705 }
1706 inp = tp->t_inpcb;
1707 tw->tw_inpcb = inp;
1708
1709 /*
1710 * Recover last window size sent.
1711 */
1712 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1713
1714 /*
1715 * Set t_recent if timestamps are used on the connection.
1716 */
1717 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1718 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1719 tw->t_recent = tp->ts_recent;
1720 else
1721 tw->t_recent = 0;
1722
1723 tw->snd_nxt = tp->snd_nxt;
1724 tw->rcv_nxt = tp->rcv_nxt;
1725 tw->iss = tp->iss;
1726 tw->irs = tp->irs;
1727 tw->t_starttime = tp->t_starttime;
1728 tw->tw_time = 0;
1729
1730/* XXX
1731 * If this code will
1732 * be used for fin-wait-2 state also, then we may need
1733 * a ts_recent from the last segment.
1734 */
1735 tw_time = 2 * tcp_msl;
1736 acknow = tp->t_flags & TF_ACKNOW;
1737
1738 /*
1739 * First, discard tcpcb state, which includes stopping its timers and
1740 * freeing it. tcp_discardcb() used to also release the inpcb, but
1741 * that work is now done in the caller.
1742 */
1743 tcp_discardcb(tp);
1744 so = inp->inp_socket;
1745 SOCK_LOCK(so);
1746 tw->tw_cred = crhold(so->so_cred);
1747 tw->tw_so_options = so->so_options;
1748 SOCK_UNLOCK(so);
1749 if (acknow)
1750 tcp_twrespond(tw, TH_ACK);
1751 inp->inp_ppcb = tw;
1752 inp->inp_vflag |= INP_TIMEWAIT;
1753 tcp_timer_2msl_reset(tw, tw_time);
1754
1755 /*
1756 * If the inpcb owns the sole reference to the socket, then we can
1757 * detach and free the socket as it is not needed in time wait.
1758 */
1759 if (inp->inp_vflag & INP_SOCKREF) {
1760 KASSERT(so->so_state & SS_PROTOREF,
1761 ("tcp_twstart: !SS_PROTOREF"));
1762 inp->inp_vflag &= ~INP_SOCKREF;
1763#ifdef INET6
1764 if (inp->inp_vflag & INP_IPV6PROTO)
1765 in6_pcbdetach(inp);
1766 else
1767#endif
1768 in_pcbdetach(inp);
1769 INP_UNLOCK(inp);
1770 ACCEPT_LOCK();
1771 SOCK_LOCK(so);
1772 so->so_state &= ~SS_PROTOREF;
1773 sofree(so);
1774 } else
1775 INP_UNLOCK(inp);
1776}
1777
1778/*
1779 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1780 * the actual rate is slightly higher due to the addition of
1781 * random positive increments.
1782 *
1783 * Most other new OSes use semi-randomized ISN values, so we
1784 * do not need to worry about them.
1785 */
1786#define MS_ISN_BYTES_PER_SECOND 250000
1787
1788/*
1789 * Determine if the ISN we will generate has advanced beyond the last
1790 * sequence number used by the previous connection. If so, indicate
1791 * that it is safe to recycle this tw socket by returning 1.
1792 *
1793 * XXXRW: This function should assert the inpcb lock as it does multiple
1794 * non-atomic reads from the tcptw, but is currently called without it from
1795 * in_pcb.c:in_pcblookup_local().
1796 */
1797int
1798tcp_twrecycleable(struct tcptw *tw)
1799{
1800 tcp_seq new_iss = tw->iss;
1801 tcp_seq new_irs = tw->irs;
1802
1803 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1804 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1805
1806 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1807 return (1);
1808 else
1809 return (0);
1810}
1811
1812void
1813tcp_twclose(struct tcptw *tw, int reuse)
1814{
1815 struct socket *so;
1816 struct inpcb *inp;
1817
1818 /*
1819 * At this point, we are in one of two situations:
1820 *
1821 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all
1822 * after validating.
1823 *
1824 * (2) We have a socket, which we may or may now own the reference
1825 * for. If we own the reference, release all the state after
1826 * validating. If not, leave it for the socket close to clean up.
1827 */
1828 inp = tw->tw_inpcb;
1829 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1830 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1831 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1832 INP_LOCK_ASSERT(inp);
1833
1834 tw->tw_inpcb = NULL;
1835 tcp_timer_2msl_stop(tw);
1836 inp->inp_ppcb = NULL;
1837 inp->inp_vflag |= INP_DROPPED;
1838
1839 so = inp->inp_socket;
1840 if (so != NULL) {
1841 if (inp->inp_vflag & INP_SOCKREF) {
1842 /*
1843 * If a socket is present, and we own the only
1844 * reference, we need to tear down the socket and the
1845 * inpcb.
1846 */
1847 inp->inp_vflag &= ~INP_SOCKREF;
1848#ifdef INET6
1849 if (inp->inp_vflag & INP_IPV6PROTO) {
1850 in6_pcbdetach(inp);
1851 in6_pcbfree(inp);
1852 } else {
1853 in_pcbdetach(inp);
1854 in_pcbfree(inp);
1855 }
1856#endif
1857 ACCEPT_LOCK();
1858 SOCK_LOCK(so);
1859 KASSERT(so->so_state & SS_PROTOREF,
1860 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1861 so->so_state &= ~SS_PROTOREF;
1862 sofree(so);
1863 } else {
1864 /*
1865 * If we don't own the only reference, the socket and
1866 * inpcb need to be left around to be handled by
1867 * tcp_usr_detach() later.
1868 */
1869 INP_UNLOCK(inp);
1870 }
1871 } else {
1872#ifdef INET6
1873 if (inp->inp_vflag & INP_IPV6PROTO)
1874 in6_pcbfree(inp);
1875 else
1876#endif
1877 in_pcbfree(inp);
1878 }
1879 tcpstat.tcps_closed++;
1880 crfree(tw->tw_cred);
1881 tw->tw_cred = NULL;
1882 if (reuse)
1883 return;
1884 uma_zfree(tcptw_zone, tw);
1885}
1886
1887int
1888tcp_twrespond(struct tcptw *tw, int flags)
1889{
1890 struct inpcb *inp = tw->tw_inpcb;
1891 struct tcphdr *th;
1892 struct mbuf *m;
1893 struct ip *ip = NULL;
1894 u_int8_t *optp;
1895 u_int hdrlen, optlen;
1896 int error;
1897#ifdef INET6
1898 struct ip6_hdr *ip6 = NULL;
1899 int isipv6 = inp->inp_inc.inc_isipv6;
1900#endif
1901
1902 INP_LOCK_ASSERT(inp);
1903
1904 m = m_gethdr(M_DONTWAIT, MT_DATA);
1905 if (m == NULL)
1906 return (ENOBUFS);
1907 m->m_data += max_linkhdr;
1908
1909#ifdef MAC
1910 mac_create_mbuf_from_inpcb(inp, m);
1911#endif
1912
1913#ifdef INET6
1914 if (isipv6) {
1915 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1916 ip6 = mtod(m, struct ip6_hdr *);
1917 th = (struct tcphdr *)(ip6 + 1);
1918 tcpip_fillheaders(inp, ip6, th);
1919 } else
1920#endif
1921 {
1922 hdrlen = sizeof(struct tcpiphdr);
1923 ip = mtod(m, struct ip *);
1924 th = (struct tcphdr *)(ip + 1);
1925 tcpip_fillheaders(inp, ip, th);
1926 }
1927 optp = (u_int8_t *)(th + 1);
1928
1929 /*
1930 * Send a timestamp and echo-reply if both our side and our peer
1931 * have sent timestamps in our SYN's and this is not a RST.
1932 */
1933 if (tw->t_recent && flags == TH_ACK) {
1934 u_int32_t *lp = (u_int32_t *)optp;
1935
1936 /* Form timestamp option as shown in appendix A of RFC 1323. */
1937 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1938 *lp++ = htonl(ticks);
1939 *lp = htonl(tw->t_recent);
1940 optp += TCPOLEN_TSTAMP_APPA;
1941 }
1942
1943 optlen = optp - (u_int8_t *)(th + 1);
1944
1945 m->m_len = hdrlen + optlen;
1946 m->m_pkthdr.len = m->m_len;
1947
1948 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1949
1950 th->th_seq = htonl(tw->snd_nxt);
1951 th->th_ack = htonl(tw->rcv_nxt);
1952 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1953 th->th_flags = flags;
1954 th->th_win = htons(tw->last_win);
1955
1956#ifdef INET6
1957 if (isipv6) {
1958 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1959 sizeof(struct tcphdr) + optlen);
1960 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1961 error = ip6_output(m, inp->in6p_outputopts, NULL,
1962 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1963 } else
1964#endif
1965 {
1966 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1967 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1968 m->m_pkthdr.csum_flags = CSUM_TCP;
1969 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1970 ip->ip_len = m->m_pkthdr.len;
1971 if (path_mtu_discovery)
1972 ip->ip_off |= IP_DF;
1973 error = ip_output(m, inp->inp_options, NULL,
1974 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1975 NULL, inp);
1976 }
1977 if (flags & TH_ACK)
1978 tcpstat.tcps_sndacks++;
1979 else
1980 tcpstat.tcps_sndctrl++;
1981 tcpstat.tcps_sndtotal++;
1982 return (error);
1983}
1984
1985/*
1986 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1987 *
1988 * This code attempts to calculate the bandwidth-delay product as a
1989 * means of determining the optimal window size to maximize bandwidth,
1990 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1991 * routers. This code also does a fairly good job keeping RTTs in check
1992 * across slow links like modems. We implement an algorithm which is very
1993 * similar (but not meant to be) TCP/Vegas. The code operates on the
1994 * transmitter side of a TCP connection and so only effects the transmit
1995 * side of the connection.
1996 *
1997 * BACKGROUND: TCP makes no provision for the management of buffer space
1998 * at the end points or at the intermediate routers and switches. A TCP
1999 * stream, whether using NewReno or not, will eventually buffer as
2000 * many packets as it is able and the only reason this typically works is
2001 * due to the fairly small default buffers made available for a connection
2002 * (typicaly 16K or 32K). As machines use larger windows and/or window
2003 * scaling it is now fairly easy for even a single TCP connection to blow-out
2004 * all available buffer space not only on the local interface, but on
2005 * intermediate routers and switches as well. NewReno makes a misguided
2006 * attempt to 'solve' this problem by waiting for an actual failure to occur,
2007 * then backing off, then steadily increasing the window again until another
2008 * failure occurs, ad-infinitum. This results in terrible oscillation that
2009 * is only made worse as network loads increase and the idea of intentionally
2010 * blowing out network buffers is, frankly, a terrible way to manage network
2011 * resources.
2012 *
2013 * It is far better to limit the transmit window prior to the failure
2014 * condition being achieved. There are two general ways to do this: First
2015 * you can 'scan' through different transmit window sizes and locate the
2016 * point where the RTT stops increasing, indicating that you have filled the
2017 * pipe, then scan backwards until you note that RTT stops decreasing, then
2018 * repeat ad-infinitum. This method works in principle but has severe
2019 * implementation issues due to RTT variances, timer granularity, and
2020 * instability in the algorithm which can lead to many false positives and
2021 * create oscillations as well as interact badly with other TCP streams
2022 * implementing the same algorithm.
2023 *
2024 * The second method is to limit the window to the bandwidth delay product
2025 * of the link. This is the method we implement. RTT variances and our
2026 * own manipulation of the congestion window, bwnd, can potentially
2027 * destabilize the algorithm. For this reason we have to stabilize the
2028 * elements used to calculate the window. We do this by using the minimum
2029 * observed RTT, the long term average of the observed bandwidth, and
2030 * by adding two segments worth of slop. It isn't perfect but it is able
2031 * to react to changing conditions and gives us a very stable basis on
2032 * which to extend the algorithm.
2033 */
2034void
2035tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2036{
2037 u_long bw;
2038 u_long bwnd;
2039 int save_ticks;
2040
2041 INP_LOCK_ASSERT(tp->t_inpcb);
2042
2043 /*
2044 * If inflight_enable is disabled in the middle of a tcp connection,
2045 * make sure snd_bwnd is effectively disabled.
2046 */
2047 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2048 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2049 tp->snd_bandwidth = 0;
2050 return;
2051 }
2052
2053 /*
2054 * Figure out the bandwidth. Due to the tick granularity this
2055 * is a very rough number and it MUST be averaged over a fairly
2056 * long period of time. XXX we need to take into account a link
2057 * that is not using all available bandwidth, but for now our
2058 * slop will ramp us up if this case occurs and the bandwidth later
2059 * increases.
2060 *
2061 * Note: if ticks rollover 'bw' may wind up negative. We must
2062 * effectively reset t_bw_rtttime for this case.
2063 */
2064 save_ticks = ticks;
2065 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2066 return;
2067
2068 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2069 (save_ticks - tp->t_bw_rtttime);
2070 tp->t_bw_rtttime = save_ticks;
2071 tp->t_bw_rtseq = ack_seq;
2072 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2073 return;
2074 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2075
2076 tp->snd_bandwidth = bw;
2077
2078 /*
2079 * Calculate the semi-static bandwidth delay product, plus two maximal
2080 * segments. The additional slop puts us squarely in the sweet
2081 * spot and also handles the bandwidth run-up case and stabilization.
2082 * Without the slop we could be locking ourselves into a lower
2083 * bandwidth.
2084 *
2085 * Situations Handled:
2086 * (1) Prevents over-queueing of packets on LANs, especially on
2087 * high speed LANs, allowing larger TCP buffers to be
2088 * specified, and also does a good job preventing
2089 * over-queueing of packets over choke points like modems
2090 * (at least for the transmit side).
2091 *
2092 * (2) Is able to handle changing network loads (bandwidth
2093 * drops so bwnd drops, bandwidth increases so bwnd
2094 * increases).
2095 *
2096 * (3) Theoretically should stabilize in the face of multiple
2097 * connections implementing the same algorithm (this may need
2098 * a little work).
2099 *
2100 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2101 * be adjusted with a sysctl but typically only needs to be
2102 * on very slow connections. A value no smaller then 5
2103 * should be used, but only reduce this default if you have
2104 * no other choice.
2105 */
2106#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2107 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2108#undef USERTT
2109
2110 if (tcp_inflight_debug > 0) {
2111 static int ltime;
2112 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2113 ltime = ticks;
2114 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2115 tp,
2116 bw,
2117 tp->t_rttbest,
2118 tp->t_srtt,
2119 bwnd
2120 );
2121 }
2122 }
2123 if ((long)bwnd < tcp_inflight_min)
2124 bwnd = tcp_inflight_min;
2125 if (bwnd > tcp_inflight_max)
2126 bwnd = tcp_inflight_max;
2127 if ((long)bwnd < tp->t_maxseg * 2)
2128 bwnd = tp->t_maxseg * 2;
2129 tp->snd_bwnd = bwnd;
2130}
2131
2132#ifdef TCP_SIGNATURE
2133/*
2134 * Callback function invoked by m_apply() to digest TCP segment data
2135 * contained within an mbuf chain.
2136 */
2137static int
2138tcp_signature_apply(void *fstate, void *data, u_int len)
2139{
2140
2141 MD5Update(fstate, (u_char *)data, len);
2142 return (0);
2143}
2144
2145/*
2146 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2147 *
2148 * Parameters:
2149 * m pointer to head of mbuf chain
2150 * off0 offset to TCP header within the mbuf chain
2151 * len length of TCP segment data, excluding options
2152 * optlen length of TCP segment options
2153 * buf pointer to storage for computed MD5 digest
2154 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2155 *
2156 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2157 * When called from tcp_input(), we can be sure that th_sum has been
2158 * zeroed out and verified already.
2159 *
2160 * This function is for IPv4 use only. Calling this function with an
2161 * IPv6 packet in the mbuf chain will yield undefined results.
2162 *
2163 * Return 0 if successful, otherwise return -1.
2164 *
2165 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2166 * search with the destination IP address, and a 'magic SPI' to be
2167 * determined by the application. This is hardcoded elsewhere to 1179
2168 * right now. Another branch of this code exists which uses the SPD to
2169 * specify per-application flows but it is unstable.
2170 */
2171int
2172tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2173 u_char *buf, u_int direction)
2174{
2175 union sockaddr_union dst;
2176 struct ippseudo ippseudo;
2177 MD5_CTX ctx;
2178 int doff;
2179 struct ip *ip;
2180 struct ipovly *ipovly;
2181 struct secasvar *sav;
2182 struct tcphdr *th;
2183 u_short savecsum;
2184
2185 KASSERT(m != NULL, ("NULL mbuf chain"));
2186 KASSERT(buf != NULL, ("NULL signature pointer"));
2187
2188 /* Extract the destination from the IP header in the mbuf. */
2189 ip = mtod(m, struct ip *);
2190 bzero(&dst, sizeof(union sockaddr_union));
2191 dst.sa.sa_len = sizeof(struct sockaddr_in);
2192 dst.sa.sa_family = AF_INET;
2193 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2194 ip->ip_src : ip->ip_dst;
2195
2196 /* Look up an SADB entry which matches the address of the peer. */
2197 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2198 if (sav == NULL) {
2199 printf("%s: SADB lookup failed for %s\n", __func__,
2200 inet_ntoa(dst.sin.sin_addr));
2201 return (EINVAL);
2202 }
2203
2204 MD5Init(&ctx);
2205 ipovly = (struct ipovly *)ip;
2206 th = (struct tcphdr *)((u_char *)ip + off0);
2207 doff = off0 + sizeof(struct tcphdr) + optlen;
2208
2209 /*
2210 * Step 1: Update MD5 hash with IP pseudo-header.
2211 *
2212 * XXX The ippseudo header MUST be digested in network byte order,
2213 * or else we'll fail the regression test. Assume all fields we've
2214 * been doing arithmetic on have been in host byte order.
2215 * XXX One cannot depend on ipovly->ih_len here. When called from
2216 * tcp_output(), the underlying ip_len member has not yet been set.
2217 */
2218 ippseudo.ippseudo_src = ipovly->ih_src;
2219 ippseudo.ippseudo_dst = ipovly->ih_dst;
2220 ippseudo.ippseudo_pad = 0;
2221 ippseudo.ippseudo_p = IPPROTO_TCP;
2222 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2223 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2224
2225 /*
2226 * Step 2: Update MD5 hash with TCP header, excluding options.
2227 * The TCP checksum must be set to zero.
2228 */
2229 savecsum = th->th_sum;
2230 th->th_sum = 0;
2231 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2232 th->th_sum = savecsum;
2233
2234 /*
2235 * Step 3: Update MD5 hash with TCP segment data.
2236 * Use m_apply() to avoid an early m_pullup().
2237 */
2238 if (len > 0)
2239 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2240
2241 /*
2242 * Step 4: Update MD5 hash with shared secret.
2243 */
2244 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2245 MD5Final(buf, &ctx);
2246
2247 key_sa_recordxfer(sav, m);
2248 KEY_FREESAV(&sav);
2249 return (0);
2250}
2251#endif /* TCP_SIGNATURE */
2252
2253static int
2254sysctl_drop(SYSCTL_HANDLER_ARGS)
2255{
2256 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2257 struct sockaddr_storage addrs[2];
2258 struct inpcb *inp;
2259 struct tcpcb *tp;
2260 struct tcptw *tw;
2261 struct sockaddr_in *fin, *lin;
2262#ifdef INET6
2263 struct sockaddr_in6 *fin6, *lin6;
2264 struct in6_addr f6, l6;
2265#endif
2266 int error;
2267
2268 inp = NULL;
2269 fin = lin = NULL;
2270#ifdef INET6
2271 fin6 = lin6 = NULL;
2272#endif
2273 error = 0;
2274
2275 if (req->oldptr != NULL || req->oldlen != 0)
2276 return (EINVAL);
2277 if (req->newptr == NULL)
2278 return (EPERM);
2279 if (req->newlen < sizeof(addrs))
2280 return (ENOMEM);
2281 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2282 if (error)
2283 return (error);
2284
2285 switch (addrs[0].ss_family) {
2286#ifdef INET6
2287 case AF_INET6:
2288 fin6 = (struct sockaddr_in6 *)&addrs[0];
2289 lin6 = (struct sockaddr_in6 *)&addrs[1];
2290 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2291 lin6->sin6_len != sizeof(struct sockaddr_in6))
2292 return (EINVAL);
2293 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2294 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2295 return (EINVAL);
2296 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2297 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2298 fin = (struct sockaddr_in *)&addrs[0];
2299 lin = (struct sockaddr_in *)&addrs[1];
2300 break;
2301 }
2302 error = sa6_embedscope(fin6, ip6_use_defzone);
2303 if (error)
2304 return (error);
2305 error = sa6_embedscope(lin6, ip6_use_defzone);
2306 if (error)
2307 return (error);
2308 break;
2309#endif
2310 case AF_INET:
2311 fin = (struct sockaddr_in *)&addrs[0];
2312 lin = (struct sockaddr_in *)&addrs[1];
2313 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2314 lin->sin_len != sizeof(struct sockaddr_in))
2315 return (EINVAL);
2316 break;
2317 default:
2318 return (EINVAL);
2319 }
2320 INP_INFO_WLOCK(&tcbinfo);
2321 switch (addrs[0].ss_family) {
2322#ifdef INET6
2323 case AF_INET6:
2324 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2325 &l6, lin6->sin6_port, 0, NULL);
2326 break;
2327#endif
2328 case AF_INET:
2329 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2330 lin->sin_addr, lin->sin_port, 0, NULL);
2331 break;
2332 }
2333 if (inp != NULL) {
2334 INP_LOCK(inp);
2335 if (inp->inp_vflag & INP_TIMEWAIT) {
2336 /*
2337 * XXXRW: There currently exists a state where an
2338 * inpcb is present, but its timewait state has been
2339 * discarded. For now, don't allow dropping of this
2340 * type of inpcb.
2341 */
2342 tw = intotw(inp);
2343 if (tw != NULL)
2344 tcp_twclose(tw, 0);
2345 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2346 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2347 tp = intotcpcb(inp);
2348 tcp_drop(tp, ECONNABORTED);
2349 }
2350 INP_UNLOCK(inp);
2351 } else
2352 error = ESRCH;
2353 INP_INFO_WUNLOCK(&tcbinfo);
2354 return (error);
2355}
2356
2357SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2358 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2359 0, sysctl_drop, "", "Drop TCP connection");
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 157478 2006-04-04 14:31:37Z glebius $
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 */
251void
252tcp_init(void)
253{
254 int hashsize = TCBHASHSIZE;
255
256 tcp_delacktime = TCPTV_DELACK;
257 tcp_keepinit = TCPTV_KEEP_INIT;
258 tcp_keepidle = TCPTV_KEEP_IDLE;
259 tcp_keepintvl = TCPTV_KEEPINTVL;
260 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
261 tcp_msl = TCPTV_MSL;
262 tcp_rexmit_min = TCPTV_MIN;
263 tcp_rexmit_slop = TCPTV_CPU_VAR;
264 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
265
266 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
267 LIST_INIT(&tcb);
268 tcbinfo.listhead = &tcb;
269 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
270 if (!powerof2(hashsize)) {
271 printf("WARNING: TCB hash size not a power of 2\n");
272 hashsize = 512; /* safe default */
273 }
274 tcp_tcbhashsize = hashsize;
275 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
276 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
277 &tcbinfo.porthashmask);
278 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
279 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
280 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
281#ifdef INET6
282#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
283#else /* INET6 */
284#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
285#endif /* INET6 */
286 if (max_protohdr < TCP_MINPROTOHDR)
287 max_protohdr = TCP_MINPROTOHDR;
288 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
289 panic("tcp_init");
290#undef TCP_MINPROTOHDR
291 /*
292 * These have to be type stable for the benefit of the timers.
293 */
294 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
295 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
296 uma_zone_set_max(tcpcb_zone, maxsockets);
297 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
298 if (maxtcptw == 0)
299 maxtcptw = maxsockets / 5;
300 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
301 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
302 uma_zone_set_max(tcptw_zone, maxtcptw);
303 tcp_timer_init();
304 syncache_init();
305 tcp_hc_init();
306 tcp_reass_init();
307 callout_init(&isn_callout, CALLOUT_MPSAFE);
308 tcp_isn_tick(NULL);
309 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
310 SHUTDOWN_PRI_DEFAULT);
311 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
312 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313}
314
315void
316tcp_fini(void *xtp)
317{
318
319 callout_stop(&isn_callout);
320}
321
322/*
323 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
324 * tcp_template used to store this data in mbufs, but we now recopy it out
325 * of the tcpcb each time to conserve mbufs.
326 */
327void
328tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
329{
330 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
331
332 INP_LOCK_ASSERT(inp);
333
334#ifdef INET6
335 if ((inp->inp_vflag & INP_IPV6) != 0) {
336 struct ip6_hdr *ip6;
337
338 ip6 = (struct ip6_hdr *)ip_ptr;
339 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
340 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
341 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
342 (IPV6_VERSION & IPV6_VERSION_MASK);
343 ip6->ip6_nxt = IPPROTO_TCP;
344 ip6->ip6_plen = sizeof(struct tcphdr);
345 ip6->ip6_src = inp->in6p_laddr;
346 ip6->ip6_dst = inp->in6p_faddr;
347 } else
348#endif
349 {
350 struct ip *ip;
351
352 ip = (struct ip *)ip_ptr;
353 ip->ip_v = IPVERSION;
354 ip->ip_hl = 5;
355 ip->ip_tos = inp->inp_ip_tos;
356 ip->ip_len = 0;
357 ip->ip_id = 0;
358 ip->ip_off = 0;
359 ip->ip_ttl = inp->inp_ip_ttl;
360 ip->ip_sum = 0;
361 ip->ip_p = IPPROTO_TCP;
362 ip->ip_src = inp->inp_laddr;
363 ip->ip_dst = inp->inp_faddr;
364 }
365 th->th_sport = inp->inp_lport;
366 th->th_dport = inp->inp_fport;
367 th->th_seq = 0;
368 th->th_ack = 0;
369 th->th_x2 = 0;
370 th->th_off = 5;
371 th->th_flags = 0;
372 th->th_win = 0;
373 th->th_urp = 0;
374 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
375}
376
377/*
378 * Create template to be used to send tcp packets on a connection.
379 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
380 * use for this function is in keepalives, which use tcp_respond.
381 */
382struct tcptemp *
383tcpip_maketemplate(struct inpcb *inp)
384{
385 struct mbuf *m;
386 struct tcptemp *n;
387
388 m = m_get(M_DONTWAIT, MT_DATA);
389 if (m == NULL)
390 return (0);
391 m->m_len = sizeof(struct tcptemp);
392 n = mtod(m, struct tcptemp *);
393
394 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
395 return (n);
396}
397
398/*
399 * Send a single message to the TCP at address specified by
400 * the given TCP/IP header. If m == NULL, then we make a copy
401 * of the tcpiphdr at ti and send directly to the addressed host.
402 * This is used to force keep alive messages out using the TCP
403 * template for a connection. If flags are given then we send
404 * a message back to the TCP which originated the * segment ti,
405 * and discard the mbuf containing it and any other attached mbufs.
406 *
407 * In any case the ack and sequence number of the transmitted
408 * segment are as specified by the parameters.
409 *
410 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
411 */
412void
413tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
414 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
415{
416 register int tlen;
417 int win = 0;
418 struct ip *ip;
419 struct tcphdr *nth;
420#ifdef INET6
421 struct ip6_hdr *ip6;
422 int isipv6;
423#endif /* INET6 */
424 int ipflags = 0;
425 struct inpcb *inp;
426
427 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
428
429#ifdef INET6
430 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
431 ip6 = ipgen;
432#endif /* INET6 */
433 ip = ipgen;
434
435 if (tp != NULL) {
436 inp = tp->t_inpcb;
437 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
438 INP_INFO_WLOCK_ASSERT(&tcbinfo);
439 INP_LOCK_ASSERT(inp);
440 } else
441 inp = NULL;
442
443 if (tp != NULL) {
444 if (!(flags & TH_RST)) {
445 win = sbspace(&inp->inp_socket->so_rcv);
446 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
447 win = (long)TCP_MAXWIN << tp->rcv_scale;
448 }
449 }
450 if (m == NULL) {
451 m = m_gethdr(M_DONTWAIT, MT_DATA);
452 if (m == NULL)
453 return;
454 tlen = 0;
455 m->m_data += max_linkhdr;
456#ifdef INET6
457 if (isipv6) {
458 bcopy((caddr_t)ip6, mtod(m, caddr_t),
459 sizeof(struct ip6_hdr));
460 ip6 = mtod(m, struct ip6_hdr *);
461 nth = (struct tcphdr *)(ip6 + 1);
462 } else
463#endif /* INET6 */
464 {
465 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
466 ip = mtod(m, struct ip *);
467 nth = (struct tcphdr *)(ip + 1);
468 }
469 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
470 flags = TH_ACK;
471 } else {
472 m_freem(m->m_next);
473 m->m_next = NULL;
474 m->m_data = (caddr_t)ipgen;
475 /* m_len is set later */
476 tlen = 0;
477#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
478#ifdef INET6
479 if (isipv6) {
480 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
481 nth = (struct tcphdr *)(ip6 + 1);
482 } else
483#endif /* INET6 */
484 {
485 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
486 nth = (struct tcphdr *)(ip + 1);
487 }
488 if (th != nth) {
489 /*
490 * this is usually a case when an extension header
491 * exists between the IPv6 header and the
492 * TCP header.
493 */
494 nth->th_sport = th->th_sport;
495 nth->th_dport = th->th_dport;
496 }
497 xchg(nth->th_dport, nth->th_sport, n_short);
498#undef xchg
499 }
500#ifdef INET6
501 if (isipv6) {
502 ip6->ip6_flow = 0;
503 ip6->ip6_vfc = IPV6_VERSION;
504 ip6->ip6_nxt = IPPROTO_TCP;
505 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
506 tlen));
507 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
508 } else
509#endif
510 {
511 tlen += sizeof (struct tcpiphdr);
512 ip->ip_len = tlen;
513 ip->ip_ttl = ip_defttl;
514 if (path_mtu_discovery)
515 ip->ip_off |= IP_DF;
516 }
517 m->m_len = tlen;
518 m->m_pkthdr.len = tlen;
519 m->m_pkthdr.rcvif = NULL;
520#ifdef MAC
521 if (inp != NULL) {
522 /*
523 * Packet is associated with a socket, so allow the
524 * label of the response to reflect the socket label.
525 */
526 INP_LOCK_ASSERT(inp);
527 mac_create_mbuf_from_inpcb(inp, m);
528 } else {
529 /*
530 * Packet is not associated with a socket, so possibly
531 * update the label in place.
532 */
533 mac_reflect_mbuf_tcp(m);
534 }
535#endif
536 nth->th_seq = htonl(seq);
537 nth->th_ack = htonl(ack);
538 nth->th_x2 = 0;
539 nth->th_off = sizeof (struct tcphdr) >> 2;
540 nth->th_flags = flags;
541 if (tp != NULL)
542 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
543 else
544 nth->th_win = htons((u_short)win);
545 nth->th_urp = 0;
546#ifdef INET6
547 if (isipv6) {
548 nth->th_sum = 0;
549 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
550 sizeof(struct ip6_hdr),
551 tlen - sizeof(struct ip6_hdr));
552 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
553 NULL, NULL);
554 } else
555#endif /* INET6 */
556 {
557 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
558 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
559 m->m_pkthdr.csum_flags = CSUM_TCP;
560 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
561 }
562#ifdef TCPDEBUG
563 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
564 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
565#endif
566#ifdef INET6
567 if (isipv6)
568 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
569 else
570#endif /* INET6 */
571 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
572}
573
574/*
575 * Create a new TCP control block, making an
576 * empty reassembly queue and hooking it to the argument
577 * protocol control block. The `inp' parameter must have
578 * come from the zone allocator set up in tcp_init().
579 */
580struct tcpcb *
581tcp_newtcpcb(struct inpcb *inp)
582{
583 struct tcpcb_mem *tm;
584 struct tcpcb *tp;
585#ifdef INET6
586 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
587#endif /* INET6 */
588
589 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
590 if (tm == NULL)
591 return (NULL);
592 tp = &tm->tcb;
593 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
594 tp->t_maxseg = tp->t_maxopd =
595#ifdef INET6
596 isipv6 ? tcp_v6mssdflt :
597#endif /* INET6 */
598 tcp_mssdflt;
599
600 /* Set up our timeouts. */
601 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
602 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
603 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
604 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
605 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
606
607 if (tcp_do_rfc1323)
608 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
609 tp->sack_enable = tcp_do_sack;
610 TAILQ_INIT(&tp->snd_holes);
611 tp->t_inpcb = inp; /* XXX */
612 /*
613 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
614 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
615 * reasonable initial retransmit time.
616 */
617 tp->t_srtt = TCPTV_SRTTBASE;
618 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
619 tp->t_rttmin = tcp_rexmit_min;
620 tp->t_rxtcur = TCPTV_RTOBASE;
621 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
622 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
623 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
624 tp->t_rcvtime = ticks;
625 tp->t_bw_rtttime = ticks;
626 /*
627 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
628 * because the socket may be bound to an IPv6 wildcard address,
629 * which may match an IPv4-mapped IPv6 address.
630 */
631 inp->inp_ip_ttl = ip_defttl;
632 inp->inp_ppcb = tp;
633 return (tp); /* XXX */
634}
635
636/*
637 * Drop a TCP connection, reporting
638 * the specified error. If connection is synchronized,
639 * then send a RST to peer.
640 */
641struct tcpcb *
642tcp_drop(struct tcpcb *tp, int errno)
643{
644 struct socket *so = tp->t_inpcb->inp_socket;
645
646 INP_INFO_WLOCK_ASSERT(&tcbinfo);
647 INP_LOCK_ASSERT(tp->t_inpcb);
648
649 if (TCPS_HAVERCVDSYN(tp->t_state)) {
650 tp->t_state = TCPS_CLOSED;
651 (void) tcp_output(tp);
652 tcpstat.tcps_drops++;
653 } else
654 tcpstat.tcps_conndrops++;
655 if (errno == ETIMEDOUT && tp->t_softerror)
656 errno = tp->t_softerror;
657 so->so_error = errno;
658 return (tcp_close(tp));
659}
660
661void
662tcp_discardcb(struct tcpcb *tp)
663{
664 struct tseg_qent *q;
665 struct inpcb *inp = tp->t_inpcb;
666 struct socket *so = inp->inp_socket;
667#ifdef INET6
668 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
669#endif /* INET6 */
670
671 /*
672 * XXXRW: This is all very well and good, but actually, we might be
673 * discarding the tcpcb after the socket is gone, so we can't do
674 * this:
675 KASSERT(so != NULL, ("tcp_discardcb: so == NULL"));
676 */
677 INP_LOCK_ASSERT(inp);
678
679 /*
680 * Make sure that all of our timers are stopped before we
681 * delete the PCB.
682 */
683 callout_stop(tp->tt_rexmt);
684 callout_stop(tp->tt_persist);
685 callout_stop(tp->tt_keep);
686 callout_stop(tp->tt_2msl);
687 callout_stop(tp->tt_delack);
688
689 /*
690 * If we got enough samples through the srtt filter,
691 * save the rtt and rttvar in the routing entry.
692 * 'Enough' is arbitrarily defined as 4 rtt samples.
693 * 4 samples is enough for the srtt filter to converge
694 * to within enough % of the correct value; fewer samples
695 * and we could save a bogus rtt. The danger is not high
696 * as tcp quickly recovers from everything.
697 * XXX: Works very well but needs some more statistics!
698 */
699 if (tp->t_rttupdated >= 4) {
700 struct hc_metrics_lite metrics;
701 u_long ssthresh;
702
703 bzero(&metrics, sizeof(metrics));
704 /*
705 * Update the ssthresh always when the conditions below
706 * are satisfied. This gives us better new start value
707 * for the congestion avoidance for new connections.
708 * ssthresh is only set if packet loss occured on a session.
709 */
710 ssthresh = tp->snd_ssthresh;
711 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
712 /*
713 * convert the limit from user data bytes to
714 * packets then to packet data bytes.
715 */
716 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
717 if (ssthresh < 2)
718 ssthresh = 2;
719 ssthresh *= (u_long)(tp->t_maxseg +
720#ifdef INET6
721 (isipv6 ? sizeof (struct ip6_hdr) +
722 sizeof (struct tcphdr) :
723#endif
724 sizeof (struct tcpiphdr)
725#ifdef INET6
726 )
727#endif
728 );
729 } else
730 ssthresh = 0;
731 metrics.rmx_ssthresh = ssthresh;
732
733 metrics.rmx_rtt = tp->t_srtt;
734 metrics.rmx_rttvar = tp->t_rttvar;
735 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
736 metrics.rmx_bandwidth = tp->snd_bandwidth;
737 metrics.rmx_cwnd = tp->snd_cwnd;
738 metrics.rmx_sendpipe = 0;
739 metrics.rmx_recvpipe = 0;
740
741 tcp_hc_update(&inp->inp_inc, &metrics);
742 }
743
744 /* free the reassembly queue, if any */
745 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
746 LIST_REMOVE(q, tqe_q);
747 m_freem(q->tqe_m);
748 uma_zfree(tcp_reass_zone, q);
749 tp->t_segqlen--;
750 tcp_reass_qsize--;
751 }
752 tcp_free_sackholes(tp);
753 inp->inp_ppcb = NULL;
754 tp->t_inpcb = NULL;
755 uma_zfree(tcpcb_zone, tp);
756
757 /*
758 * XXXRW: This seems a bit unclean.
759 */
760 if (so != NULL)
761 soisdisconnected(so);
762}
763
764/*
765 * Attempt to close a TCP control block, marking it as dropped, and freeing
766 * the socket if we hold the only reference.
767 */
768struct tcpcb *
769tcp_close(struct tcpcb *tp)
770{
771 struct inpcb *inp = tp->t_inpcb;
772 struct socket *so;
773
774 INP_INFO_WLOCK_ASSERT(&tcbinfo);
775 INP_LOCK_ASSERT(inp);
776
777 inp->inp_vflag |= INP_DROPPED;
778
779 tcpstat.tcps_closed++;
780 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
781 so = inp->inp_socket;
782 soisdisconnected(so);
783 if (inp->inp_vflag & INP_SOCKREF) {
784 KASSERT(so->so_state & SS_PROTOREF,
785 ("tcp_close: !SS_PROTOREF"));
786 inp->inp_vflag &= ~INP_SOCKREF;
787 tcp_discardcb(tp);
788#ifdef INET6
789 if (inp->inp_vflag & INP_IPV6PROTO) {
790 in6_pcbdetach(inp);
791 in6_pcbfree(inp);
792 } else {
793#endif
794 in_pcbdetach(inp);
795 in_pcbfree(inp);
796#ifdef INET6
797 }
798#endif
799 ACCEPT_LOCK();
800 SOCK_LOCK(so);
801 so->so_state &= ~SS_PROTOREF;
802 sofree(so);
803 return (NULL);
804 }
805 return (tp);
806}
807
808void
809tcp_drain(void)
810{
811
812 if (do_tcpdrain) {
813 struct inpcb *inpb;
814 struct tcpcb *tcpb;
815 struct tseg_qent *te;
816
817 /*
818 * Walk the tcpbs, if existing, and flush the reassembly queue,
819 * if there is one...
820 * XXX: The "Net/3" implementation doesn't imply that the TCP
821 * reassembly queue should be flushed, but in a situation
822 * where we're really low on mbufs, this is potentially
823 * usefull.
824 */
825 INP_INFO_RLOCK(&tcbinfo);
826 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
827 if (inpb->inp_vflag & INP_TIMEWAIT)
828 continue;
829 INP_LOCK(inpb);
830 if ((tcpb = intotcpcb(inpb)) != NULL) {
831 while ((te = LIST_FIRST(&tcpb->t_segq))
832 != NULL) {
833 LIST_REMOVE(te, tqe_q);
834 m_freem(te->tqe_m);
835 uma_zfree(tcp_reass_zone, te);
836 tcpb->t_segqlen--;
837 tcp_reass_qsize--;
838 }
839 tcp_clean_sackreport(tcpb);
840 }
841 INP_UNLOCK(inpb);
842 }
843 INP_INFO_RUNLOCK(&tcbinfo);
844 }
845}
846
847/*
848 * Notify a tcp user of an asynchronous error;
849 * store error as soft error, but wake up user
850 * (for now, won't do anything until can select for soft error).
851 *
852 * Do not wake up user since there currently is no mechanism for
853 * reporting soft errors (yet - a kqueue filter may be added).
854 */
855static struct inpcb *
856tcp_notify(struct inpcb *inp, int error)
857{
858 struct tcpcb *tp;
859
860 INP_INFO_WLOCK_ASSERT(&tcbinfo);
861 INP_LOCK_ASSERT(inp);
862
863 if ((inp->inp_vflag & INP_TIMEWAIT) ||
864 (inp->inp_vflag & INP_DROPPED))
865 return (inp);
866
867 tp = intotcpcb(inp);
868 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
869
870 /*
871 * Ignore some errors if we are hooked up.
872 * If connection hasn't completed, has retransmitted several times,
873 * and receives a second error, give up now. This is better
874 * than waiting a long time to establish a connection that
875 * can never complete.
876 */
877 if (tp->t_state == TCPS_ESTABLISHED &&
878 (error == EHOSTUNREACH || error == ENETUNREACH ||
879 error == EHOSTDOWN)) {
880 return (inp);
881 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
882 tp->t_softerror) {
883 tp = tcp_drop(tp, error);
884 if (tp != NULL)
885 return (inp);
886 else
887 return (NULL);
888 } else {
889 tp->t_softerror = error;
890 return (inp);
891 }
892#if 0
893 wakeup( &so->so_timeo);
894 sorwakeup(so);
895 sowwakeup(so);
896#endif
897}
898
899static int
900tcp_pcblist(SYSCTL_HANDLER_ARGS)
901{
902 int error, i, n;
903 struct inpcb *inp, **inp_list;
904 inp_gen_t gencnt;
905 struct xinpgen xig;
906
907 /*
908 * The process of preparing the TCB list is too time-consuming and
909 * resource-intensive to repeat twice on every request.
910 */
911 if (req->oldptr == NULL) {
912 n = tcbinfo.ipi_count;
913 req->oldidx = 2 * (sizeof xig)
914 + (n + n/8) * sizeof(struct xtcpcb);
915 return (0);
916 }
917
918 if (req->newptr != NULL)
919 return (EPERM);
920
921 /*
922 * OK, now we're committed to doing something.
923 */
924 INP_INFO_RLOCK(&tcbinfo);
925 gencnt = tcbinfo.ipi_gencnt;
926 n = tcbinfo.ipi_count;
927 INP_INFO_RUNLOCK(&tcbinfo);
928
929 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
930 + n * sizeof(struct xtcpcb));
931 if (error != 0)
932 return (error);
933
934 xig.xig_len = sizeof xig;
935 xig.xig_count = n;
936 xig.xig_gen = gencnt;
937 xig.xig_sogen = so_gencnt;
938 error = SYSCTL_OUT(req, &xig, sizeof xig);
939 if (error)
940 return (error);
941
942 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
943 if (inp_list == NULL)
944 return (ENOMEM);
945
946 INP_INFO_RLOCK(&tcbinfo);
947 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
948 inp = LIST_NEXT(inp, inp_list)) {
949 INP_LOCK(inp);
950 if (inp->inp_gencnt <= gencnt) {
951 /*
952 * XXX: This use of cr_cansee(), introduced with
953 * TCP state changes, is not quite right, but for
954 * now, better than nothing.
955 */
956 if (inp->inp_vflag & INP_TIMEWAIT) {
957 if (intotw(inp) != NULL)
958 error = cr_cansee(req->td->td_ucred,
959 intotw(inp)->tw_cred);
960 else
961 error = EINVAL; /* Skip this inp. */
962 } else
963 error = cr_canseesocket(req->td->td_ucred,
964 inp->inp_socket);
965 if (error == 0)
966 inp_list[i++] = inp;
967 }
968 INP_UNLOCK(inp);
969 }
970 INP_INFO_RUNLOCK(&tcbinfo);
971 n = i;
972
973 error = 0;
974 for (i = 0; i < n; i++) {
975 inp = inp_list[i];
976 if (inp->inp_gencnt <= gencnt) {
977 struct xtcpcb xt;
978 void *inp_ppcb;
979
980 bzero(&xt, sizeof(xt));
981 xt.xt_len = sizeof xt;
982 /* XXX should avoid extra copy */
983 bcopy(inp, &xt.xt_inp, sizeof *inp);
984 inp_ppcb = inp->inp_ppcb;
985 if (inp_ppcb == NULL)
986 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
987 else if (inp->inp_vflag & INP_TIMEWAIT) {
988 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
989 xt.xt_tp.t_state = TCPS_TIME_WAIT;
990 } else
991 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
992 if (inp->inp_socket != NULL)
993 sotoxsocket(inp->inp_socket, &xt.xt_socket);
994 else {
995 bzero(&xt.xt_socket, sizeof xt.xt_socket);
996 xt.xt_socket.xso_protocol = IPPROTO_TCP;
997 }
998 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
999 error = SYSCTL_OUT(req, &xt, sizeof xt);
1000 }
1001 }
1002 if (!error) {
1003 /*
1004 * Give the user an updated idea of our state.
1005 * If the generation differs from what we told
1006 * her before, she knows that something happened
1007 * while we were processing this request, and it
1008 * might be necessary to retry.
1009 */
1010 INP_INFO_RLOCK(&tcbinfo);
1011 xig.xig_gen = tcbinfo.ipi_gencnt;
1012 xig.xig_sogen = so_gencnt;
1013 xig.xig_count = tcbinfo.ipi_count;
1014 INP_INFO_RUNLOCK(&tcbinfo);
1015 error = SYSCTL_OUT(req, &xig, sizeof xig);
1016 }
1017 free(inp_list, M_TEMP);
1018 return (error);
1019}
1020
1021SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1022 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1023
1024static int
1025tcp_getcred(SYSCTL_HANDLER_ARGS)
1026{
1027 struct xucred xuc;
1028 struct sockaddr_in addrs[2];
1029 struct inpcb *inp;
1030 int error;
1031
1032 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1033 if (error)
1034 return (error);
1035 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1036 if (error)
1037 return (error);
1038 INP_INFO_RLOCK(&tcbinfo);
1039 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1040 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1041 if (inp == NULL) {
1042 error = ENOENT;
1043 goto outunlocked;
1044 }
1045 INP_LOCK(inp);
1046 if (inp->inp_socket == NULL) {
1047 error = ENOENT;
1048 goto out;
1049 }
1050 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1051 if (error)
1052 goto out;
1053 cru2x(inp->inp_socket->so_cred, &xuc);
1054out:
1055 INP_UNLOCK(inp);
1056outunlocked:
1057 INP_INFO_RUNLOCK(&tcbinfo);
1058 if (error == 0)
1059 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1060 return (error);
1061}
1062
1063SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1064 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1065 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1066
1067#ifdef INET6
1068static int
1069tcp6_getcred(SYSCTL_HANDLER_ARGS)
1070{
1071 struct xucred xuc;
1072 struct sockaddr_in6 addrs[2];
1073 struct inpcb *inp;
1074 int error, mapped = 0;
1075
1076 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1077 if (error)
1078 return (error);
1079 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1080 if (error)
1081 return (error);
1082 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1083 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1084 return (error);
1085 }
1086 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1087 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1088 mapped = 1;
1089 else
1090 return (EINVAL);
1091 }
1092
1093 INP_INFO_RLOCK(&tcbinfo);
1094 if (mapped == 1)
1095 inp = in_pcblookup_hash(&tcbinfo,
1096 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1097 addrs[1].sin6_port,
1098 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1099 addrs[0].sin6_port,
1100 0, NULL);
1101 else
1102 inp = in6_pcblookup_hash(&tcbinfo,
1103 &addrs[1].sin6_addr, addrs[1].sin6_port,
1104 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1105 if (inp == NULL) {
1106 error = ENOENT;
1107 goto outunlocked;
1108 }
1109 INP_LOCK(inp);
1110 if (inp->inp_socket == NULL) {
1111 error = ENOENT;
1112 goto out;
1113 }
1114 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1115 if (error)
1116 goto out;
1117 cru2x(inp->inp_socket->so_cred, &xuc);
1118out:
1119 INP_UNLOCK(inp);
1120outunlocked:
1121 INP_INFO_RUNLOCK(&tcbinfo);
1122 if (error == 0)
1123 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1124 return (error);
1125}
1126
1127SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1128 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1129 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1130#endif
1131
1132
1133void
1134tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1135{
1136 struct ip *ip = vip;
1137 struct tcphdr *th;
1138 struct in_addr faddr;
1139 struct inpcb *inp;
1140 struct tcpcb *tp;
1141 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1142 struct icmp *icp;
1143 struct in_conninfo inc;
1144 tcp_seq icmp_tcp_seq;
1145 int mtu;
1146
1147 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1148 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1149 return;
1150
1151 if (cmd == PRC_MSGSIZE)
1152 notify = tcp_mtudisc;
1153 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1154 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1155 notify = tcp_drop_syn_sent;
1156 /*
1157 * Redirects don't need to be handled up here.
1158 */
1159 else if (PRC_IS_REDIRECT(cmd))
1160 return;
1161 /*
1162 * Source quench is depreciated.
1163 */
1164 else if (cmd == PRC_QUENCH)
1165 return;
1166 /*
1167 * Hostdead is ugly because it goes linearly through all PCBs.
1168 * XXX: We never get this from ICMP, otherwise it makes an
1169 * excellent DoS attack on machines with many connections.
1170 */
1171 else if (cmd == PRC_HOSTDEAD)
1172 ip = NULL;
1173 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1174 return;
1175 if (ip != NULL) {
1176 icp = (struct icmp *)((caddr_t)ip
1177 - offsetof(struct icmp, icmp_ip));
1178 th = (struct tcphdr *)((caddr_t)ip
1179 + (ip->ip_hl << 2));
1180 INP_INFO_WLOCK(&tcbinfo);
1181 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1182 ip->ip_src, th->th_sport, 0, NULL);
1183 if (inp != NULL) {
1184 INP_LOCK(inp);
1185 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1186 !(inp->inp_vflag & INP_DROPPED) &&
1187 !(inp->inp_socket == NULL)) {
1188 icmp_tcp_seq = htonl(th->th_seq);
1189 tp = intotcpcb(inp);
1190 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1191 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1192 if (cmd == PRC_MSGSIZE) {
1193 /*
1194 * MTU discovery:
1195 * If we got a needfrag set the MTU
1196 * in the route to the suggested new
1197 * value (if given) and then notify.
1198 */
1199 bzero(&inc, sizeof(inc));
1200 inc.inc_flags = 0; /* IPv4 */
1201 inc.inc_faddr = faddr;
1202
1203 mtu = ntohs(icp->icmp_nextmtu);
1204 /*
1205 * If no alternative MTU was
1206 * proposed, try the next smaller
1207 * one. ip->ip_len has already
1208 * been swapped in icmp_input().
1209 */
1210 if (!mtu)
1211 mtu = ip_next_mtu(ip->ip_len,
1212 1);
1213 if (mtu < max(296, (tcp_minmss)
1214 + sizeof(struct tcpiphdr)))
1215 mtu = 0;
1216 if (!mtu)
1217 mtu = tcp_mssdflt
1218 + sizeof(struct tcpiphdr);
1219 /*
1220 * Only cache the the MTU if it
1221 * is smaller than the interface
1222 * or route MTU. tcp_mtudisc()
1223 * will do right thing by itself.
1224 */
1225 if (mtu <= tcp_maxmtu(&inc))
1226 tcp_hc_updatemtu(&inc, mtu);
1227 }
1228
1229 inp = (*notify)(inp, inetctlerrmap[cmd]);
1230 }
1231 }
1232 if (inp != NULL)
1233 INP_UNLOCK(inp);
1234 } else {
1235 inc.inc_fport = th->th_dport;
1236 inc.inc_lport = th->th_sport;
1237 inc.inc_faddr = faddr;
1238 inc.inc_laddr = ip->ip_src;
1239#ifdef INET6
1240 inc.inc_isipv6 = 0;
1241#endif
1242 syncache_unreach(&inc, th);
1243 }
1244 INP_INFO_WUNLOCK(&tcbinfo);
1245 } else
1246 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1247}
1248
1249#ifdef INET6
1250void
1251tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1252{
1253 struct tcphdr th;
1254 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1255 struct ip6_hdr *ip6;
1256 struct mbuf *m;
1257 struct ip6ctlparam *ip6cp = NULL;
1258 const struct sockaddr_in6 *sa6_src = NULL;
1259 int off;
1260 struct tcp_portonly {
1261 u_int16_t th_sport;
1262 u_int16_t th_dport;
1263 } *thp;
1264
1265 if (sa->sa_family != AF_INET6 ||
1266 sa->sa_len != sizeof(struct sockaddr_in6))
1267 return;
1268
1269 if (cmd == PRC_MSGSIZE)
1270 notify = tcp_mtudisc;
1271 else if (!PRC_IS_REDIRECT(cmd) &&
1272 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1273 return;
1274 /* Source quench is depreciated. */
1275 else if (cmd == PRC_QUENCH)
1276 return;
1277
1278 /* if the parameter is from icmp6, decode it. */
1279 if (d != NULL) {
1280 ip6cp = (struct ip6ctlparam *)d;
1281 m = ip6cp->ip6c_m;
1282 ip6 = ip6cp->ip6c_ip6;
1283 off = ip6cp->ip6c_off;
1284 sa6_src = ip6cp->ip6c_src;
1285 } else {
1286 m = NULL;
1287 ip6 = NULL;
1288 off = 0; /* fool gcc */
1289 sa6_src = &sa6_any;
1290 }
1291
1292 if (ip6 != NULL) {
1293 struct in_conninfo inc;
1294 /*
1295 * XXX: We assume that when IPV6 is non NULL,
1296 * M and OFF are valid.
1297 */
1298
1299 /* check if we can safely examine src and dst ports */
1300 if (m->m_pkthdr.len < off + sizeof(*thp))
1301 return;
1302
1303 bzero(&th, sizeof(th));
1304 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1305
1306 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1307 (struct sockaddr *)ip6cp->ip6c_src,
1308 th.th_sport, cmd, NULL, notify);
1309
1310 inc.inc_fport = th.th_dport;
1311 inc.inc_lport = th.th_sport;
1312 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1313 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1314 inc.inc_isipv6 = 1;
1315 INP_INFO_WLOCK(&tcbinfo);
1316 syncache_unreach(&inc, &th);
1317 INP_INFO_WUNLOCK(&tcbinfo);
1318 } else
1319 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1320 0, cmd, NULL, notify);
1321}
1322#endif /* INET6 */
1323
1324
1325/*
1326 * Following is where TCP initial sequence number generation occurs.
1327 *
1328 * There are two places where we must use initial sequence numbers:
1329 * 1. In SYN-ACK packets.
1330 * 2. In SYN packets.
1331 *
1332 * All ISNs for SYN-ACK packets are generated by the syncache. See
1333 * tcp_syncache.c for details.
1334 *
1335 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1336 * depends on this property. In addition, these ISNs should be
1337 * unguessable so as to prevent connection hijacking. To satisfy
1338 * the requirements of this situation, the algorithm outlined in
1339 * RFC 1948 is used, with only small modifications.
1340 *
1341 * Implementation details:
1342 *
1343 * Time is based off the system timer, and is corrected so that it
1344 * increases by one megabyte per second. This allows for proper
1345 * recycling on high speed LANs while still leaving over an hour
1346 * before rollover.
1347 *
1348 * As reading the *exact* system time is too expensive to be done
1349 * whenever setting up a TCP connection, we increment the time
1350 * offset in two ways. First, a small random positive increment
1351 * is added to isn_offset for each connection that is set up.
1352 * Second, the function tcp_isn_tick fires once per clock tick
1353 * and increments isn_offset as necessary so that sequence numbers
1354 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1355 * random positive increments serve only to ensure that the same
1356 * exact sequence number is never sent out twice (as could otherwise
1357 * happen when a port is recycled in less than the system tick
1358 * interval.)
1359 *
1360 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1361 * between seeding of isn_secret. This is normally set to zero,
1362 * as reseeding should not be necessary.
1363 *
1364 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1365 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1366 * general, this means holding an exclusive (write) lock.
1367 */
1368
1369#define ISN_BYTES_PER_SECOND 1048576
1370#define ISN_STATIC_INCREMENT 4096
1371#define ISN_RANDOM_INCREMENT (4096 - 1)
1372
1373static u_char isn_secret[32];
1374static int isn_last_reseed;
1375static u_int32_t isn_offset, isn_offset_old;
1376static MD5_CTX isn_ctx;
1377
1378tcp_seq
1379tcp_new_isn(struct tcpcb *tp)
1380{
1381 u_int32_t md5_buffer[4];
1382 tcp_seq new_isn;
1383
1384 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1385 INP_LOCK_ASSERT(tp->t_inpcb);
1386
1387 /* Seed if this is the first use, reseed if requested. */
1388 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1389 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1390 < (u_int)ticks))) {
1391 read_random(&isn_secret, sizeof(isn_secret));
1392 isn_last_reseed = ticks;
1393 }
1394
1395 /* Compute the md5 hash and return the ISN. */
1396 MD5Init(&isn_ctx);
1397 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1398 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1399#ifdef INET6
1400 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1401 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1402 sizeof(struct in6_addr));
1403 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1404 sizeof(struct in6_addr));
1405 } else
1406#endif
1407 {
1408 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1409 sizeof(struct in_addr));
1410 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1411 sizeof(struct in_addr));
1412 }
1413 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1414 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1415 new_isn = (tcp_seq) md5_buffer[0];
1416 isn_offset += ISN_STATIC_INCREMENT +
1417 (arc4random() & ISN_RANDOM_INCREMENT);
1418 new_isn += isn_offset;
1419 return (new_isn);
1420}
1421
1422/*
1423 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1424 * to keep time flowing at a relatively constant rate. If the random
1425 * increments have already pushed us past the projected offset, do nothing.
1426 */
1427static void
1428tcp_isn_tick(void *xtp)
1429{
1430 u_int32_t projected_offset;
1431
1432 INP_INFO_WLOCK(&tcbinfo);
1433 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1434
1435 if (projected_offset > isn_offset)
1436 isn_offset = projected_offset;
1437
1438 isn_offset_old = isn_offset;
1439 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1440 INP_INFO_WUNLOCK(&tcbinfo);
1441}
1442
1443/*
1444 * When a specific ICMP unreachable message is received and the
1445 * connection state is SYN-SENT, drop the connection. This behavior
1446 * is controlled by the icmp_may_rst sysctl.
1447 */
1448struct inpcb *
1449tcp_drop_syn_sent(struct inpcb *inp, int errno)
1450{
1451 struct tcpcb *tp;
1452
1453 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1454 INP_LOCK_ASSERT(inp);
1455
1456 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1457 (inp->inp_vflag & INP_DROPPED))
1458 return (inp);
1459
1460 tp = intotcpcb(inp);
1461 if (tp->t_state != TCPS_SYN_SENT)
1462 return (inp);
1463
1464 tp = tcp_drop(tp, errno);
1465 if (tp != NULL)
1466 return (inp);
1467 else
1468 return (NULL);
1469}
1470
1471/*
1472 * When `need fragmentation' ICMP is received, update our idea of the MSS
1473 * based on the new value in the route. Also nudge TCP to send something,
1474 * since we know the packet we just sent was dropped.
1475 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1476 */
1477struct inpcb *
1478tcp_mtudisc(struct inpcb *inp, int errno)
1479{
1480 struct tcpcb *tp;
1481 struct socket *so = inp->inp_socket;
1482 u_int maxmtu;
1483 u_int romtu;
1484 int mss;
1485#ifdef INET6
1486 int isipv6;
1487#endif /* INET6 */
1488
1489 INP_LOCK_ASSERT(inp);
1490 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1491 (inp->inp_vflag & INP_DROPPED))
1492 return (inp);
1493
1494 tp = intotcpcb(inp);
1495 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1496
1497#ifdef INET6
1498 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1499#endif
1500 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1501 romtu =
1502#ifdef INET6
1503 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1504#endif /* INET6 */
1505 tcp_maxmtu(&inp->inp_inc);
1506 if (!maxmtu)
1507 maxmtu = romtu;
1508 else
1509 maxmtu = min(maxmtu, romtu);
1510 if (!maxmtu) {
1511 tp->t_maxopd = tp->t_maxseg =
1512#ifdef INET6
1513 isipv6 ? tcp_v6mssdflt :
1514#endif /* INET6 */
1515 tcp_mssdflt;
1516 return (inp);
1517 }
1518 mss = maxmtu -
1519#ifdef INET6
1520 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1521#endif /* INET6 */
1522 sizeof(struct tcpiphdr)
1523#ifdef INET6
1524 )
1525#endif /* INET6 */
1526 ;
1527
1528 /*
1529 * XXX - The above conditional probably violates the TCP
1530 * spec. The problem is that, since we don't know the
1531 * other end's MSS, we are supposed to use a conservative
1532 * default. But, if we do that, then MTU discovery will
1533 * never actually take place, because the conservative
1534 * default is much less than the MTUs typically seen
1535 * on the Internet today. For the moment, we'll sweep
1536 * this under the carpet.
1537 *
1538 * The conservative default might not actually be a problem
1539 * if the only case this occurs is when sending an initial
1540 * SYN with options and data to a host we've never talked
1541 * to before. Then, they will reply with an MSS value which
1542 * will get recorded and the new parameters should get
1543 * recomputed. For Further Study.
1544 */
1545 if (tp->t_maxopd <= mss)
1546 return (inp);
1547 tp->t_maxopd = mss;
1548
1549 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1550 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1551 mss -= TCPOLEN_TSTAMP_APPA;
1552#if (MCLBYTES & (MCLBYTES - 1)) == 0
1553 if (mss > MCLBYTES)
1554 mss &= ~(MCLBYTES-1);
1555#else
1556 if (mss > MCLBYTES)
1557 mss = mss / MCLBYTES * MCLBYTES;
1558#endif
1559 if (so->so_snd.sb_hiwat < mss)
1560 mss = so->so_snd.sb_hiwat;
1561
1562 tp->t_maxseg = mss;
1563
1564 tcpstat.tcps_mturesent++;
1565 tp->t_rtttime = 0;
1566 tp->snd_nxt = tp->snd_una;
1567 tcp_output(tp);
1568 return (inp);
1569}
1570
1571/*
1572 * Look-up the routing entry to the peer of this inpcb. If no route
1573 * is found and it cannot be allocated, then return NULL. This routine
1574 * is called by TCP routines that access the rmx structure and by tcp_mss
1575 * to get the interface MTU.
1576 */
1577u_long
1578tcp_maxmtu(struct in_conninfo *inc)
1579{
1580 struct route sro;
1581 struct sockaddr_in *dst;
1582 struct ifnet *ifp;
1583 u_long maxmtu = 0;
1584
1585 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1586
1587 bzero(&sro, sizeof(sro));
1588 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1589 dst = (struct sockaddr_in *)&sro.ro_dst;
1590 dst->sin_family = AF_INET;
1591 dst->sin_len = sizeof(*dst);
1592 dst->sin_addr = inc->inc_faddr;
1593 rtalloc_ign(&sro, RTF_CLONING);
1594 }
1595 if (sro.ro_rt != NULL) {
1596 ifp = sro.ro_rt->rt_ifp;
1597 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1598 maxmtu = ifp->if_mtu;
1599 else
1600 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1601 RTFREE(sro.ro_rt);
1602 }
1603 return (maxmtu);
1604}
1605
1606#ifdef INET6
1607u_long
1608tcp_maxmtu6(struct in_conninfo *inc)
1609{
1610 struct route_in6 sro6;
1611 struct ifnet *ifp;
1612 u_long maxmtu = 0;
1613
1614 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1615
1616 bzero(&sro6, sizeof(sro6));
1617 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1618 sro6.ro_dst.sin6_family = AF_INET6;
1619 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1620 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1621 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1622 }
1623 if (sro6.ro_rt != NULL) {
1624 ifp = sro6.ro_rt->rt_ifp;
1625 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1626 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1627 else
1628 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1629 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1630 RTFREE(sro6.ro_rt);
1631 }
1632
1633 return (maxmtu);
1634}
1635#endif /* INET6 */
1636
1637#ifdef IPSEC
1638/* compute ESP/AH header size for TCP, including outer IP header. */
1639size_t
1640ipsec_hdrsiz_tcp(struct tcpcb *tp)
1641{
1642 struct inpcb *inp;
1643 struct mbuf *m;
1644 size_t hdrsiz;
1645 struct ip *ip;
1646#ifdef INET6
1647 struct ip6_hdr *ip6;
1648#endif
1649 struct tcphdr *th;
1650
1651 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1652 return (0);
1653 MGETHDR(m, M_DONTWAIT, MT_DATA);
1654 if (!m)
1655 return (0);
1656
1657#ifdef INET6
1658 if ((inp->inp_vflag & INP_IPV6) != 0) {
1659 ip6 = mtod(m, struct ip6_hdr *);
1660 th = (struct tcphdr *)(ip6 + 1);
1661 m->m_pkthdr.len = m->m_len =
1662 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1663 tcpip_fillheaders(inp, ip6, th);
1664 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1665 } else
1666#endif /* INET6 */
1667 {
1668 ip = mtod(m, struct ip *);
1669 th = (struct tcphdr *)(ip + 1);
1670 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1671 tcpip_fillheaders(inp, ip, th);
1672 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1673 }
1674
1675 m_free(m);
1676 return (hdrsiz);
1677}
1678#endif /*IPSEC*/
1679
1680/*
1681 * Move a TCP connection into TIME_WAIT state.
1682 * tcbinfo is locked.
1683 * inp is locked, and is unlocked before returning.
1684 */
1685void
1686tcp_twstart(struct tcpcb *tp)
1687{
1688 struct tcptw *tw;
1689 struct inpcb *inp;
1690 int tw_time, acknow;
1691 struct socket *so;
1692
1693 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1694 INP_LOCK_ASSERT(tp->t_inpcb);
1695
1696 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1697 if (tw == NULL) {
1698 tw = tcp_timer_2msl_tw(1);
1699 if (tw == NULL) {
1700 tp = tcp_close(tp);
1701 if (tp != NULL)
1702 INP_UNLOCK(tp->t_inpcb);
1703 return;
1704 }
1705 }
1706 inp = tp->t_inpcb;
1707 tw->tw_inpcb = inp;
1708
1709 /*
1710 * Recover last window size sent.
1711 */
1712 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1713
1714 /*
1715 * Set t_recent if timestamps are used on the connection.
1716 */
1717 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1718 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1719 tw->t_recent = tp->ts_recent;
1720 else
1721 tw->t_recent = 0;
1722
1723 tw->snd_nxt = tp->snd_nxt;
1724 tw->rcv_nxt = tp->rcv_nxt;
1725 tw->iss = tp->iss;
1726 tw->irs = tp->irs;
1727 tw->t_starttime = tp->t_starttime;
1728 tw->tw_time = 0;
1729
1730/* XXX
1731 * If this code will
1732 * be used for fin-wait-2 state also, then we may need
1733 * a ts_recent from the last segment.
1734 */
1735 tw_time = 2 * tcp_msl;
1736 acknow = tp->t_flags & TF_ACKNOW;
1737
1738 /*
1739 * First, discard tcpcb state, which includes stopping its timers and
1740 * freeing it. tcp_discardcb() used to also release the inpcb, but
1741 * that work is now done in the caller.
1742 */
1743 tcp_discardcb(tp);
1744 so = inp->inp_socket;
1745 SOCK_LOCK(so);
1746 tw->tw_cred = crhold(so->so_cred);
1747 tw->tw_so_options = so->so_options;
1748 SOCK_UNLOCK(so);
1749 if (acknow)
1750 tcp_twrespond(tw, TH_ACK);
1751 inp->inp_ppcb = tw;
1752 inp->inp_vflag |= INP_TIMEWAIT;
1753 tcp_timer_2msl_reset(tw, tw_time);
1754
1755 /*
1756 * If the inpcb owns the sole reference to the socket, then we can
1757 * detach and free the socket as it is not needed in time wait.
1758 */
1759 if (inp->inp_vflag & INP_SOCKREF) {
1760 KASSERT(so->so_state & SS_PROTOREF,
1761 ("tcp_twstart: !SS_PROTOREF"));
1762 inp->inp_vflag &= ~INP_SOCKREF;
1763#ifdef INET6
1764 if (inp->inp_vflag & INP_IPV6PROTO)
1765 in6_pcbdetach(inp);
1766 else
1767#endif
1768 in_pcbdetach(inp);
1769 INP_UNLOCK(inp);
1770 ACCEPT_LOCK();
1771 SOCK_LOCK(so);
1772 so->so_state &= ~SS_PROTOREF;
1773 sofree(so);
1774 } else
1775 INP_UNLOCK(inp);
1776}
1777
1778/*
1779 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1780 * the actual rate is slightly higher due to the addition of
1781 * random positive increments.
1782 *
1783 * Most other new OSes use semi-randomized ISN values, so we
1784 * do not need to worry about them.
1785 */
1786#define MS_ISN_BYTES_PER_SECOND 250000
1787
1788/*
1789 * Determine if the ISN we will generate has advanced beyond the last
1790 * sequence number used by the previous connection. If so, indicate
1791 * that it is safe to recycle this tw socket by returning 1.
1792 *
1793 * XXXRW: This function should assert the inpcb lock as it does multiple
1794 * non-atomic reads from the tcptw, but is currently called without it from
1795 * in_pcb.c:in_pcblookup_local().
1796 */
1797int
1798tcp_twrecycleable(struct tcptw *tw)
1799{
1800 tcp_seq new_iss = tw->iss;
1801 tcp_seq new_irs = tw->irs;
1802
1803 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1804 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1805
1806 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1807 return (1);
1808 else
1809 return (0);
1810}
1811
1812void
1813tcp_twclose(struct tcptw *tw, int reuse)
1814{
1815 struct socket *so;
1816 struct inpcb *inp;
1817
1818 /*
1819 * At this point, we are in one of two situations:
1820 *
1821 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all
1822 * after validating.
1823 *
1824 * (2) We have a socket, which we may or may now own the reference
1825 * for. If we own the reference, release all the state after
1826 * validating. If not, leave it for the socket close to clean up.
1827 */
1828 inp = tw->tw_inpcb;
1829 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1830 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1831 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1832 INP_LOCK_ASSERT(inp);
1833
1834 tw->tw_inpcb = NULL;
1835 tcp_timer_2msl_stop(tw);
1836 inp->inp_ppcb = NULL;
1837 inp->inp_vflag |= INP_DROPPED;
1838
1839 so = inp->inp_socket;
1840 if (so != NULL) {
1841 if (inp->inp_vflag & INP_SOCKREF) {
1842 /*
1843 * If a socket is present, and we own the only
1844 * reference, we need to tear down the socket and the
1845 * inpcb.
1846 */
1847 inp->inp_vflag &= ~INP_SOCKREF;
1848#ifdef INET6
1849 if (inp->inp_vflag & INP_IPV6PROTO) {
1850 in6_pcbdetach(inp);
1851 in6_pcbfree(inp);
1852 } else {
1853 in_pcbdetach(inp);
1854 in_pcbfree(inp);
1855 }
1856#endif
1857 ACCEPT_LOCK();
1858 SOCK_LOCK(so);
1859 KASSERT(so->so_state & SS_PROTOREF,
1860 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1861 so->so_state &= ~SS_PROTOREF;
1862 sofree(so);
1863 } else {
1864 /*
1865 * If we don't own the only reference, the socket and
1866 * inpcb need to be left around to be handled by
1867 * tcp_usr_detach() later.
1868 */
1869 INP_UNLOCK(inp);
1870 }
1871 } else {
1872#ifdef INET6
1873 if (inp->inp_vflag & INP_IPV6PROTO)
1874 in6_pcbfree(inp);
1875 else
1876#endif
1877 in_pcbfree(inp);
1878 }
1879 tcpstat.tcps_closed++;
1880 crfree(tw->tw_cred);
1881 tw->tw_cred = NULL;
1882 if (reuse)
1883 return;
1884 uma_zfree(tcptw_zone, tw);
1885}
1886
1887int
1888tcp_twrespond(struct tcptw *tw, int flags)
1889{
1890 struct inpcb *inp = tw->tw_inpcb;
1891 struct tcphdr *th;
1892 struct mbuf *m;
1893 struct ip *ip = NULL;
1894 u_int8_t *optp;
1895 u_int hdrlen, optlen;
1896 int error;
1897#ifdef INET6
1898 struct ip6_hdr *ip6 = NULL;
1899 int isipv6 = inp->inp_inc.inc_isipv6;
1900#endif
1901
1902 INP_LOCK_ASSERT(inp);
1903
1904 m = m_gethdr(M_DONTWAIT, MT_DATA);
1905 if (m == NULL)
1906 return (ENOBUFS);
1907 m->m_data += max_linkhdr;
1908
1909#ifdef MAC
1910 mac_create_mbuf_from_inpcb(inp, m);
1911#endif
1912
1913#ifdef INET6
1914 if (isipv6) {
1915 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1916 ip6 = mtod(m, struct ip6_hdr *);
1917 th = (struct tcphdr *)(ip6 + 1);
1918 tcpip_fillheaders(inp, ip6, th);
1919 } else
1920#endif
1921 {
1922 hdrlen = sizeof(struct tcpiphdr);
1923 ip = mtod(m, struct ip *);
1924 th = (struct tcphdr *)(ip + 1);
1925 tcpip_fillheaders(inp, ip, th);
1926 }
1927 optp = (u_int8_t *)(th + 1);
1928
1929 /*
1930 * Send a timestamp and echo-reply if both our side and our peer
1931 * have sent timestamps in our SYN's and this is not a RST.
1932 */
1933 if (tw->t_recent && flags == TH_ACK) {
1934 u_int32_t *lp = (u_int32_t *)optp;
1935
1936 /* Form timestamp option as shown in appendix A of RFC 1323. */
1937 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1938 *lp++ = htonl(ticks);
1939 *lp = htonl(tw->t_recent);
1940 optp += TCPOLEN_TSTAMP_APPA;
1941 }
1942
1943 optlen = optp - (u_int8_t *)(th + 1);
1944
1945 m->m_len = hdrlen + optlen;
1946 m->m_pkthdr.len = m->m_len;
1947
1948 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1949
1950 th->th_seq = htonl(tw->snd_nxt);
1951 th->th_ack = htonl(tw->rcv_nxt);
1952 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1953 th->th_flags = flags;
1954 th->th_win = htons(tw->last_win);
1955
1956#ifdef INET6
1957 if (isipv6) {
1958 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1959 sizeof(struct tcphdr) + optlen);
1960 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1961 error = ip6_output(m, inp->in6p_outputopts, NULL,
1962 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1963 } else
1964#endif
1965 {
1966 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1967 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1968 m->m_pkthdr.csum_flags = CSUM_TCP;
1969 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1970 ip->ip_len = m->m_pkthdr.len;
1971 if (path_mtu_discovery)
1972 ip->ip_off |= IP_DF;
1973 error = ip_output(m, inp->inp_options, NULL,
1974 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1975 NULL, inp);
1976 }
1977 if (flags & TH_ACK)
1978 tcpstat.tcps_sndacks++;
1979 else
1980 tcpstat.tcps_sndctrl++;
1981 tcpstat.tcps_sndtotal++;
1982 return (error);
1983}
1984
1985/*
1986 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1987 *
1988 * This code attempts to calculate the bandwidth-delay product as a
1989 * means of determining the optimal window size to maximize bandwidth,
1990 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1991 * routers. This code also does a fairly good job keeping RTTs in check
1992 * across slow links like modems. We implement an algorithm which is very
1993 * similar (but not meant to be) TCP/Vegas. The code operates on the
1994 * transmitter side of a TCP connection and so only effects the transmit
1995 * side of the connection.
1996 *
1997 * BACKGROUND: TCP makes no provision for the management of buffer space
1998 * at the end points or at the intermediate routers and switches. A TCP
1999 * stream, whether using NewReno or not, will eventually buffer as
2000 * many packets as it is able and the only reason this typically works is
2001 * due to the fairly small default buffers made available for a connection
2002 * (typicaly 16K or 32K). As machines use larger windows and/or window
2003 * scaling it is now fairly easy for even a single TCP connection to blow-out
2004 * all available buffer space not only on the local interface, but on
2005 * intermediate routers and switches as well. NewReno makes a misguided
2006 * attempt to 'solve' this problem by waiting for an actual failure to occur,
2007 * then backing off, then steadily increasing the window again until another
2008 * failure occurs, ad-infinitum. This results in terrible oscillation that
2009 * is only made worse as network loads increase and the idea of intentionally
2010 * blowing out network buffers is, frankly, a terrible way to manage network
2011 * resources.
2012 *
2013 * It is far better to limit the transmit window prior to the failure
2014 * condition being achieved. There are two general ways to do this: First
2015 * you can 'scan' through different transmit window sizes and locate the
2016 * point where the RTT stops increasing, indicating that you have filled the
2017 * pipe, then scan backwards until you note that RTT stops decreasing, then
2018 * repeat ad-infinitum. This method works in principle but has severe
2019 * implementation issues due to RTT variances, timer granularity, and
2020 * instability in the algorithm which can lead to many false positives and
2021 * create oscillations as well as interact badly with other TCP streams
2022 * implementing the same algorithm.
2023 *
2024 * The second method is to limit the window to the bandwidth delay product
2025 * of the link. This is the method we implement. RTT variances and our
2026 * own manipulation of the congestion window, bwnd, can potentially
2027 * destabilize the algorithm. For this reason we have to stabilize the
2028 * elements used to calculate the window. We do this by using the minimum
2029 * observed RTT, the long term average of the observed bandwidth, and
2030 * by adding two segments worth of slop. It isn't perfect but it is able
2031 * to react to changing conditions and gives us a very stable basis on
2032 * which to extend the algorithm.
2033 */
2034void
2035tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2036{
2037 u_long bw;
2038 u_long bwnd;
2039 int save_ticks;
2040
2041 INP_LOCK_ASSERT(tp->t_inpcb);
2042
2043 /*
2044 * If inflight_enable is disabled in the middle of a tcp connection,
2045 * make sure snd_bwnd is effectively disabled.
2046 */
2047 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2048 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2049 tp->snd_bandwidth = 0;
2050 return;
2051 }
2052
2053 /*
2054 * Figure out the bandwidth. Due to the tick granularity this
2055 * is a very rough number and it MUST be averaged over a fairly
2056 * long period of time. XXX we need to take into account a link
2057 * that is not using all available bandwidth, but for now our
2058 * slop will ramp us up if this case occurs and the bandwidth later
2059 * increases.
2060 *
2061 * Note: if ticks rollover 'bw' may wind up negative. We must
2062 * effectively reset t_bw_rtttime for this case.
2063 */
2064 save_ticks = ticks;
2065 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2066 return;
2067
2068 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2069 (save_ticks - tp->t_bw_rtttime);
2070 tp->t_bw_rtttime = save_ticks;
2071 tp->t_bw_rtseq = ack_seq;
2072 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2073 return;
2074 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2075
2076 tp->snd_bandwidth = bw;
2077
2078 /*
2079 * Calculate the semi-static bandwidth delay product, plus two maximal
2080 * segments. The additional slop puts us squarely in the sweet
2081 * spot and also handles the bandwidth run-up case and stabilization.
2082 * Without the slop we could be locking ourselves into a lower
2083 * bandwidth.
2084 *
2085 * Situations Handled:
2086 * (1) Prevents over-queueing of packets on LANs, especially on
2087 * high speed LANs, allowing larger TCP buffers to be
2088 * specified, and also does a good job preventing
2089 * over-queueing of packets over choke points like modems
2090 * (at least for the transmit side).
2091 *
2092 * (2) Is able to handle changing network loads (bandwidth
2093 * drops so bwnd drops, bandwidth increases so bwnd
2094 * increases).
2095 *
2096 * (3) Theoretically should stabilize in the face of multiple
2097 * connections implementing the same algorithm (this may need
2098 * a little work).
2099 *
2100 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2101 * be adjusted with a sysctl but typically only needs to be
2102 * on very slow connections. A value no smaller then 5
2103 * should be used, but only reduce this default if you have
2104 * no other choice.
2105 */
2106#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2107 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2108#undef USERTT
2109
2110 if (tcp_inflight_debug > 0) {
2111 static int ltime;
2112 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2113 ltime = ticks;
2114 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2115 tp,
2116 bw,
2117 tp->t_rttbest,
2118 tp->t_srtt,
2119 bwnd
2120 );
2121 }
2122 }
2123 if ((long)bwnd < tcp_inflight_min)
2124 bwnd = tcp_inflight_min;
2125 if (bwnd > tcp_inflight_max)
2126 bwnd = tcp_inflight_max;
2127 if ((long)bwnd < tp->t_maxseg * 2)
2128 bwnd = tp->t_maxseg * 2;
2129 tp->snd_bwnd = bwnd;
2130}
2131
2132#ifdef TCP_SIGNATURE
2133/*
2134 * Callback function invoked by m_apply() to digest TCP segment data
2135 * contained within an mbuf chain.
2136 */
2137static int
2138tcp_signature_apply(void *fstate, void *data, u_int len)
2139{
2140
2141 MD5Update(fstate, (u_char *)data, len);
2142 return (0);
2143}
2144
2145/*
2146 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2147 *
2148 * Parameters:
2149 * m pointer to head of mbuf chain
2150 * off0 offset to TCP header within the mbuf chain
2151 * len length of TCP segment data, excluding options
2152 * optlen length of TCP segment options
2153 * buf pointer to storage for computed MD5 digest
2154 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2155 *
2156 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2157 * When called from tcp_input(), we can be sure that th_sum has been
2158 * zeroed out and verified already.
2159 *
2160 * This function is for IPv4 use only. Calling this function with an
2161 * IPv6 packet in the mbuf chain will yield undefined results.
2162 *
2163 * Return 0 if successful, otherwise return -1.
2164 *
2165 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2166 * search with the destination IP address, and a 'magic SPI' to be
2167 * determined by the application. This is hardcoded elsewhere to 1179
2168 * right now. Another branch of this code exists which uses the SPD to
2169 * specify per-application flows but it is unstable.
2170 */
2171int
2172tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2173 u_char *buf, u_int direction)
2174{
2175 union sockaddr_union dst;
2176 struct ippseudo ippseudo;
2177 MD5_CTX ctx;
2178 int doff;
2179 struct ip *ip;
2180 struct ipovly *ipovly;
2181 struct secasvar *sav;
2182 struct tcphdr *th;
2183 u_short savecsum;
2184
2185 KASSERT(m != NULL, ("NULL mbuf chain"));
2186 KASSERT(buf != NULL, ("NULL signature pointer"));
2187
2188 /* Extract the destination from the IP header in the mbuf. */
2189 ip = mtod(m, struct ip *);
2190 bzero(&dst, sizeof(union sockaddr_union));
2191 dst.sa.sa_len = sizeof(struct sockaddr_in);
2192 dst.sa.sa_family = AF_INET;
2193 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2194 ip->ip_src : ip->ip_dst;
2195
2196 /* Look up an SADB entry which matches the address of the peer. */
2197 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2198 if (sav == NULL) {
2199 printf("%s: SADB lookup failed for %s\n", __func__,
2200 inet_ntoa(dst.sin.sin_addr));
2201 return (EINVAL);
2202 }
2203
2204 MD5Init(&ctx);
2205 ipovly = (struct ipovly *)ip;
2206 th = (struct tcphdr *)((u_char *)ip + off0);
2207 doff = off0 + sizeof(struct tcphdr) + optlen;
2208
2209 /*
2210 * Step 1: Update MD5 hash with IP pseudo-header.
2211 *
2212 * XXX The ippseudo header MUST be digested in network byte order,
2213 * or else we'll fail the regression test. Assume all fields we've
2214 * been doing arithmetic on have been in host byte order.
2215 * XXX One cannot depend on ipovly->ih_len here. When called from
2216 * tcp_output(), the underlying ip_len member has not yet been set.
2217 */
2218 ippseudo.ippseudo_src = ipovly->ih_src;
2219 ippseudo.ippseudo_dst = ipovly->ih_dst;
2220 ippseudo.ippseudo_pad = 0;
2221 ippseudo.ippseudo_p = IPPROTO_TCP;
2222 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2223 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2224
2225 /*
2226 * Step 2: Update MD5 hash with TCP header, excluding options.
2227 * The TCP checksum must be set to zero.
2228 */
2229 savecsum = th->th_sum;
2230 th->th_sum = 0;
2231 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2232 th->th_sum = savecsum;
2233
2234 /*
2235 * Step 3: Update MD5 hash with TCP segment data.
2236 * Use m_apply() to avoid an early m_pullup().
2237 */
2238 if (len > 0)
2239 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2240
2241 /*
2242 * Step 4: Update MD5 hash with shared secret.
2243 */
2244 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2245 MD5Final(buf, &ctx);
2246
2247 key_sa_recordxfer(sav, m);
2248 KEY_FREESAV(&sav);
2249 return (0);
2250}
2251#endif /* TCP_SIGNATURE */
2252
2253static int
2254sysctl_drop(SYSCTL_HANDLER_ARGS)
2255{
2256 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2257 struct sockaddr_storage addrs[2];
2258 struct inpcb *inp;
2259 struct tcpcb *tp;
2260 struct tcptw *tw;
2261 struct sockaddr_in *fin, *lin;
2262#ifdef INET6
2263 struct sockaddr_in6 *fin6, *lin6;
2264 struct in6_addr f6, l6;
2265#endif
2266 int error;
2267
2268 inp = NULL;
2269 fin = lin = NULL;
2270#ifdef INET6
2271 fin6 = lin6 = NULL;
2272#endif
2273 error = 0;
2274
2275 if (req->oldptr != NULL || req->oldlen != 0)
2276 return (EINVAL);
2277 if (req->newptr == NULL)
2278 return (EPERM);
2279 if (req->newlen < sizeof(addrs))
2280 return (ENOMEM);
2281 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2282 if (error)
2283 return (error);
2284
2285 switch (addrs[0].ss_family) {
2286#ifdef INET6
2287 case AF_INET6:
2288 fin6 = (struct sockaddr_in6 *)&addrs[0];
2289 lin6 = (struct sockaddr_in6 *)&addrs[1];
2290 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2291 lin6->sin6_len != sizeof(struct sockaddr_in6))
2292 return (EINVAL);
2293 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2294 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2295 return (EINVAL);
2296 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2297 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2298 fin = (struct sockaddr_in *)&addrs[0];
2299 lin = (struct sockaddr_in *)&addrs[1];
2300 break;
2301 }
2302 error = sa6_embedscope(fin6, ip6_use_defzone);
2303 if (error)
2304 return (error);
2305 error = sa6_embedscope(lin6, ip6_use_defzone);
2306 if (error)
2307 return (error);
2308 break;
2309#endif
2310 case AF_INET:
2311 fin = (struct sockaddr_in *)&addrs[0];
2312 lin = (struct sockaddr_in *)&addrs[1];
2313 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2314 lin->sin_len != sizeof(struct sockaddr_in))
2315 return (EINVAL);
2316 break;
2317 default:
2318 return (EINVAL);
2319 }
2320 INP_INFO_WLOCK(&tcbinfo);
2321 switch (addrs[0].ss_family) {
2322#ifdef INET6
2323 case AF_INET6:
2324 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2325 &l6, lin6->sin6_port, 0, NULL);
2326 break;
2327#endif
2328 case AF_INET:
2329 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2330 lin->sin_addr, lin->sin_port, 0, NULL);
2331 break;
2332 }
2333 if (inp != NULL) {
2334 INP_LOCK(inp);
2335 if (inp->inp_vflag & INP_TIMEWAIT) {
2336 /*
2337 * XXXRW: There currently exists a state where an
2338 * inpcb is present, but its timewait state has been
2339 * discarded. For now, don't allow dropping of this
2340 * type of inpcb.
2341 */
2342 tw = intotw(inp);
2343 if (tw != NULL)
2344 tcp_twclose(tw, 0);
2345 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2346 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2347 tp = intotcpcb(inp);
2348 tcp_drop(tp, ECONNABORTED);
2349 }
2350 INP_UNLOCK(inp);
2351 } else
2352 error = ESRCH;
2353 INP_INFO_WUNLOCK(&tcbinfo);
2354 return (error);
2355}
2356
2357SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2358 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2359 0, sysctl_drop, "", "Drop TCP connection");