proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[dragonfly.git] / sys / netinet / tcp_subr.c
CommitLineData
984263bc
MD
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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
34 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $
dadab5e9 35 * $DragonFly: src/sys/netinet/tcp_subr.c,v 1.4 2003/06/25 03:56:04 dillon Exp $
984263bc
MD
36 */
37
38#include "opt_compat.h"
39#include "opt_inet6.h"
40#include "opt_ipsec.h"
41#include "opt_tcpdebug.h"
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/callout.h>
46#include <sys/kernel.h>
47#include <sys/sysctl.h>
48#include <sys/malloc.h>
49#include <sys/mbuf.h>
50#ifdef INET6
51#include <sys/domain.h>
52#endif
53#include <sys/proc.h>
54#include <sys/socket.h>
55#include <sys/socketvar.h>
56#include <sys/protosw.h>
57#include <sys/random.h>
58
59#include <vm/vm_zone.h>
60
61#include <net/route.h>
62#include <net/if.h>
63
64#define _IP_VHL
65#include <netinet/in.h>
66#include <netinet/in_systm.h>
67#include <netinet/ip.h>
68#ifdef INET6
69#include <netinet/ip6.h>
70#endif
71#include <netinet/in_pcb.h>
72#ifdef INET6
73#include <netinet6/in6_pcb.h>
74#endif
75#include <netinet/in_var.h>
76#include <netinet/ip_var.h>
77#ifdef INET6
78#include <netinet6/ip6_var.h>
79#endif
80#include <netinet/tcp.h>
81#include <netinet/tcp_fsm.h>
82#include <netinet/tcp_seq.h>
83#include <netinet/tcp_timer.h>
84#include <netinet/tcp_var.h>
85#ifdef INET6
86#include <netinet6/tcp6_var.h>
87#endif
88#include <netinet/tcpip.h>
89#ifdef TCPDEBUG
90#include <netinet/tcp_debug.h>
91#endif
92#include <netinet6/ip6protosw.h>
93
94#ifdef IPSEC
95#include <netinet6/ipsec.h>
96#ifdef INET6
97#include <netinet6/ipsec6.h>
98#endif
99#endif /*IPSEC*/
100
101#ifdef FAST_IPSEC
102#include <netipsec/ipsec.h>
103#ifdef INET6
104#include <netipsec/ipsec6.h>
105#endif
106#define IPSEC
107#endif /*FAST_IPSEC*/
108
109#include <machine/in_cksum.h>
110#include <sys/md5.h>
111
112int tcp_mssdflt = TCP_MSS;
113SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
114 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
115
116#ifdef INET6
117int tcp_v6mssdflt = TCP6_MSS;
118SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
119 CTLFLAG_RW, &tcp_v6mssdflt , 0,
120 "Default TCP Maximum Segment Size for IPv6");
121#endif
122
123#if 0
124static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
125SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
126 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
127#endif
128
129int tcp_do_rfc1323 = 1;
130SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
131 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
132
133int tcp_do_rfc1644 = 0;
134SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
135 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
136
137static int tcp_tcbhashsize = 0;
138SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
139 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
140
141static int do_tcpdrain = 1;
142SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
143 "Enable tcp_drain routine for extra help when low on mbufs");
144
145SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
146 &tcbinfo.ipi_count, 0, "Number of active PCBs");
147
148static int icmp_may_rst = 1;
149SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
150 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
151
152static int tcp_isn_reseed_interval = 0;
153SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
154 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
155
156/*
157 * TCP bandwidth limiting sysctls. Note that the default lower bound of
158 * 1024 exists only for debugging. A good production default would be
159 * something like 6100.
160 */
161static int tcp_inflight_enable = 0;
162SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW,
163 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
164
165static int tcp_inflight_debug = 0;
166SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW,
167 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
168
169static int tcp_inflight_min = 6144;
170SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW,
171 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
172
173static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
174SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW,
175 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
176
177static int tcp_inflight_stab = 20;
178SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW,
179 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 2 packets)");
180
181static void tcp_cleartaocache __P((void));
182static void tcp_notify __P((struct inpcb *, int));
183
184/*
185 * Target size of TCP PCB hash tables. Must be a power of two.
186 *
187 * Note that this can be overridden by the kernel environment
188 * variable net.inet.tcp.tcbhashsize
189 */
190#ifndef TCBHASHSIZE
191#define TCBHASHSIZE 512
192#endif
193
194/*
195 * This is the actual shape of what we allocate using the zone
196 * allocator. Doing it this way allows us to protect both structures
197 * using the same generation count, and also eliminates the overhead
198 * of allocating tcpcbs separately. By hiding the structure here,
199 * we avoid changing most of the rest of the code (although it needs
200 * to be changed, eventually, for greater efficiency).
201 */
202#define ALIGNMENT 32
203#define ALIGNM1 (ALIGNMENT - 1)
204struct inp_tp {
205 union {
206 struct inpcb inp;
207 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
208 } inp_tp_u;
209 struct tcpcb tcb;
210 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
211 struct callout inp_tp_delack;
212};
213#undef ALIGNMENT
214#undef ALIGNM1
215
216/*
217 * Tcp initialization
218 */
219void
220tcp_init()
221{
222 int hashsize = TCBHASHSIZE;
223
224 tcp_ccgen = 1;
225 tcp_cleartaocache();
226
227 tcp_delacktime = TCPTV_DELACK;
228 tcp_keepinit = TCPTV_KEEP_INIT;
229 tcp_keepidle = TCPTV_KEEP_IDLE;
230 tcp_keepintvl = TCPTV_KEEPINTVL;
231 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
232 tcp_msl = TCPTV_MSL;
233 tcp_rexmit_min = TCPTV_MIN;
234 tcp_rexmit_slop = TCPTV_CPU_VAR;
235
236 LIST_INIT(&tcb);
237 tcbinfo.listhead = &tcb;
238 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
239 if (!powerof2(hashsize)) {
240 printf("WARNING: TCB hash size not a power of 2\n");
241 hashsize = 512; /* safe default */
242 }
243 tcp_tcbhashsize = hashsize;
244 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
245 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
246 &tcbinfo.porthashmask);
247 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
248 ZONE_INTERRUPT, 0);
249#ifdef INET6
250#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
251#else /* INET6 */
252#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
253#endif /* INET6 */
254 if (max_protohdr < TCP_MINPROTOHDR)
255 max_protohdr = TCP_MINPROTOHDR;
256 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
257 panic("tcp_init");
258#undef TCP_MINPROTOHDR
259
260 syncache_init();
261}
262
263/*
264 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
265 * tcp_template used to store this data in mbufs, but we now recopy it out
266 * of the tcpcb each time to conserve mbufs.
267 */
268void
269tcp_fillheaders(tp, ip_ptr, tcp_ptr)
270 struct tcpcb *tp;
271 void *ip_ptr;
272 void *tcp_ptr;
273{
274 struct inpcb *inp = tp->t_inpcb;
275 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
276
277#ifdef INET6
278 if ((inp->inp_vflag & INP_IPV6) != 0) {
279 struct ip6_hdr *ip6;
280
281 ip6 = (struct ip6_hdr *)ip_ptr;
282 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
283 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
284 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
285 (IPV6_VERSION & IPV6_VERSION_MASK);
286 ip6->ip6_nxt = IPPROTO_TCP;
287 ip6->ip6_plen = sizeof(struct tcphdr);
288 ip6->ip6_src = inp->in6p_laddr;
289 ip6->ip6_dst = inp->in6p_faddr;
290 tcp_hdr->th_sum = 0;
291 } else
292#endif
293 {
294 struct ip *ip = (struct ip *) ip_ptr;
295
296 ip->ip_vhl = IP_VHL_BORING;
297 ip->ip_tos = 0;
298 ip->ip_len = 0;
299 ip->ip_id = 0;
300 ip->ip_off = 0;
301 ip->ip_ttl = 0;
302 ip->ip_sum = 0;
303 ip->ip_p = IPPROTO_TCP;
304 ip->ip_src = inp->inp_laddr;
305 ip->ip_dst = inp->inp_faddr;
306 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
307 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
308 }
309
310 tcp_hdr->th_sport = inp->inp_lport;
311 tcp_hdr->th_dport = inp->inp_fport;
312 tcp_hdr->th_seq = 0;
313 tcp_hdr->th_ack = 0;
314 tcp_hdr->th_x2 = 0;
315 tcp_hdr->th_off = 5;
316 tcp_hdr->th_flags = 0;
317 tcp_hdr->th_win = 0;
318 tcp_hdr->th_urp = 0;
319}
320
321/*
322 * Create template to be used to send tcp packets on a connection.
323 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
324 * use for this function is in keepalives, which use tcp_respond.
325 */
326struct tcptemp *
327tcp_maketemplate(tp)
328 struct tcpcb *tp;
329{
330 struct mbuf *m;
331 struct tcptemp *n;
332
333 m = m_get(M_DONTWAIT, MT_HEADER);
334 if (m == NULL)
335 return (0);
336 m->m_len = sizeof(struct tcptemp);
337 n = mtod(m, struct tcptemp *);
338
339 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
340 return (n);
341}
342
343/*
344 * Send a single message to the TCP at address specified by
345 * the given TCP/IP header. If m == 0, then we make a copy
346 * of the tcpiphdr at ti and send directly to the addressed host.
347 * This is used to force keep alive messages out using the TCP
348 * template for a connection. If flags are given then we send
349 * a message back to the TCP which originated the * segment ti,
350 * and discard the mbuf containing it and any other attached mbufs.
351 *
352 * In any case the ack and sequence number of the transmitted
353 * segment are as specified by the parameters.
354 *
355 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
356 */
357void
358tcp_respond(tp, ipgen, th, m, ack, seq, flags)
359 struct tcpcb *tp;
360 void *ipgen;
361 register struct tcphdr *th;
362 register struct mbuf *m;
363 tcp_seq ack, seq;
364 int flags;
365{
366 register int tlen;
367 int win = 0;
368 struct route *ro = 0;
369 struct route sro;
370 struct ip *ip;
371 struct tcphdr *nth;
372#ifdef INET6
373 struct route_in6 *ro6 = 0;
374 struct route_in6 sro6;
375 struct ip6_hdr *ip6;
376 int isipv6;
377#endif /* INET6 */
378 int ipflags = 0;
379
380#ifdef INET6
381 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
382 ip6 = ipgen;
383#endif /* INET6 */
384 ip = ipgen;
385
386 if (tp) {
387 if (!(flags & TH_RST)) {
388 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
389 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
390 win = (long)TCP_MAXWIN << tp->rcv_scale;
391 }
392#ifdef INET6
393 if (isipv6)
394 ro6 = &tp->t_inpcb->in6p_route;
395 else
396#endif /* INET6 */
397 ro = &tp->t_inpcb->inp_route;
398 } else {
399#ifdef INET6
400 if (isipv6) {
401 ro6 = &sro6;
402 bzero(ro6, sizeof *ro6);
403 } else
404#endif /* INET6 */
405 {
406 ro = &sro;
407 bzero(ro, sizeof *ro);
408 }
409 }
410 if (m == 0) {
411 m = m_gethdr(M_DONTWAIT, MT_HEADER);
412 if (m == NULL)
413 return;
414 tlen = 0;
415 m->m_data += max_linkhdr;
416#ifdef INET6
417 if (isipv6) {
418 bcopy((caddr_t)ip6, mtod(m, caddr_t),
419 sizeof(struct ip6_hdr));
420 ip6 = mtod(m, struct ip6_hdr *);
421 nth = (struct tcphdr *)(ip6 + 1);
422 } else
423#endif /* INET6 */
424 {
425 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
426 ip = mtod(m, struct ip *);
427 nth = (struct tcphdr *)(ip + 1);
428 }
429 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
430 flags = TH_ACK;
431 } else {
432 m_freem(m->m_next);
433 m->m_next = 0;
434 m->m_data = (caddr_t)ipgen;
435 /* m_len is set later */
436 tlen = 0;
437#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
438#ifdef INET6
439 if (isipv6) {
440 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
441 nth = (struct tcphdr *)(ip6 + 1);
442 } else
443#endif /* INET6 */
444 {
445 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
446 nth = (struct tcphdr *)(ip + 1);
447 }
448 if (th != nth) {
449 /*
450 * this is usually a case when an extension header
451 * exists between the IPv6 header and the
452 * TCP header.
453 */
454 nth->th_sport = th->th_sport;
455 nth->th_dport = th->th_dport;
456 }
457 xchg(nth->th_dport, nth->th_sport, n_short);
458#undef xchg
459 }
460#ifdef INET6
461 if (isipv6) {
462 ip6->ip6_flow = 0;
463 ip6->ip6_vfc = IPV6_VERSION;
464 ip6->ip6_nxt = IPPROTO_TCP;
465 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
466 tlen));
467 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
468 } else
469#endif
470 {
471 tlen += sizeof (struct tcpiphdr);
472 ip->ip_len = tlen;
473 ip->ip_ttl = ip_defttl;
474 }
475 m->m_len = tlen;
476 m->m_pkthdr.len = tlen;
477 m->m_pkthdr.rcvif = (struct ifnet *) 0;
478 nth->th_seq = htonl(seq);
479 nth->th_ack = htonl(ack);
480 nth->th_x2 = 0;
481 nth->th_off = sizeof (struct tcphdr) >> 2;
482 nth->th_flags = flags;
483 if (tp)
484 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
485 else
486 nth->th_win = htons((u_short)win);
487 nth->th_urp = 0;
488#ifdef INET6
489 if (isipv6) {
490 nth->th_sum = 0;
491 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
492 sizeof(struct ip6_hdr),
493 tlen - sizeof(struct ip6_hdr));
494 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
495 ro6 && ro6->ro_rt ?
496 ro6->ro_rt->rt_ifp :
497 NULL);
498 } else
499#endif /* INET6 */
500 {
501 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
502 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
503 m->m_pkthdr.csum_flags = CSUM_TCP;
504 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
505 }
506#ifdef TCPDEBUG
507 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
508 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
509#endif
510#ifdef INET6
511 if (isipv6) {
512 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL,
513 tp ? tp->t_inpcb : NULL);
514 if (ro6 == &sro6 && ro6->ro_rt) {
515 RTFREE(ro6->ro_rt);
516 ro6->ro_rt = NULL;
517 }
518 } else
519#endif /* INET6 */
520 {
521 (void) ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL);
522 if (ro == &sro && ro->ro_rt) {
523 RTFREE(ro->ro_rt);
524 ro->ro_rt = NULL;
525 }
526 }
527}
528
529/*
530 * Create a new TCP control block, making an
531 * empty reassembly queue and hooking it to the argument
532 * protocol control block. The `inp' parameter must have
533 * come from the zone allocator set up in tcp_init().
534 */
535struct tcpcb *
536tcp_newtcpcb(inp)
537 struct inpcb *inp;
538{
539 struct inp_tp *it;
540 register struct tcpcb *tp;
541#ifdef INET6
542 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
543#endif /* INET6 */
544
545 it = (struct inp_tp *)inp;
546 tp = &it->tcb;
547 bzero((char *) tp, sizeof(struct tcpcb));
548 LIST_INIT(&tp->t_segq);
549 tp->t_maxseg = tp->t_maxopd =
550#ifdef INET6
551 isipv6 ? tcp_v6mssdflt :
552#endif /* INET6 */
553 tcp_mssdflt;
554
555 /* Set up our timeouts. */
556 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
557 callout_init(tp->tt_persist = &it->inp_tp_persist);
558 callout_init(tp->tt_keep = &it->inp_tp_keep);
559 callout_init(tp->tt_2msl = &it->inp_tp_2msl);
560 callout_init(tp->tt_delack = &it->inp_tp_delack);
561
562 if (tcp_do_rfc1323)
563 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
564 if (tcp_do_rfc1644)
565 tp->t_flags |= TF_REQ_CC;
566 tp->t_inpcb = inp; /* XXX */
567 /*
568 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
569 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
570 * reasonable initial retransmit time.
571 */
572 tp->t_srtt = TCPTV_SRTTBASE;
573 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
574 tp->t_rttmin = tcp_rexmit_min;
575 tp->t_rxtcur = TCPTV_RTOBASE;
576 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
577 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
578 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
579 tp->t_rcvtime = ticks;
580 tp->t_bw_rtttime = ticks;
581 /*
582 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
583 * because the socket may be bound to an IPv6 wildcard address,
584 * which may match an IPv4-mapped IPv6 address.
585 */
586 inp->inp_ip_ttl = ip_defttl;
587 inp->inp_ppcb = (caddr_t)tp;
588 return (tp); /* XXX */
589}
590
591/*
592 * Drop a TCP connection, reporting
593 * the specified error. If connection is synchronized,
594 * then send a RST to peer.
595 */
596struct tcpcb *
597tcp_drop(tp, errno)
598 register struct tcpcb *tp;
599 int errno;
600{
601 struct socket *so = tp->t_inpcb->inp_socket;
602
603 if (TCPS_HAVERCVDSYN(tp->t_state)) {
604 tp->t_state = TCPS_CLOSED;
605 (void) tcp_output(tp);
606 tcpstat.tcps_drops++;
607 } else
608 tcpstat.tcps_conndrops++;
609 if (errno == ETIMEDOUT && tp->t_softerror)
610 errno = tp->t_softerror;
611 so->so_error = errno;
612 return (tcp_close(tp));
613}
614
615/*
616 * Close a TCP control block:
617 * discard all space held by the tcp
618 * discard internet protocol block
619 * wake up any sleepers
620 */
621struct tcpcb *
622tcp_close(tp)
623 register struct tcpcb *tp;
624{
625 register struct tseg_qent *q;
626 struct inpcb *inp = tp->t_inpcb;
627 struct socket *so = inp->inp_socket;
628#ifdef INET6
629 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
630#endif /* INET6 */
631 register struct rtentry *rt;
632 int dosavessthresh;
633
634 /*
635 * Make sure that all of our timers are stopped before we
636 * delete the PCB.
637 */
638 callout_stop(tp->tt_rexmt);
639 callout_stop(tp->tt_persist);
640 callout_stop(tp->tt_keep);
641 callout_stop(tp->tt_2msl);
642 callout_stop(tp->tt_delack);
643
644 /*
645 * If we got enough samples through the srtt filter,
646 * save the rtt and rttvar in the routing entry.
647 * 'Enough' is arbitrarily defined as the 16 samples.
648 * 16 samples is enough for the srtt filter to converge
649 * to within 5% of the correct value; fewer samples and
650 * we could save a very bogus rtt.
651 *
652 * Don't update the default route's characteristics and don't
653 * update anything that the user "locked".
654 */
655 if (tp->t_rttupdated >= 16) {
656 register u_long i = 0;
657#ifdef INET6
658 if (isipv6) {
659 struct sockaddr_in6 *sin6;
660
661 if ((rt = inp->in6p_route.ro_rt) == NULL)
662 goto no_valid_rt;
663 sin6 = (struct sockaddr_in6 *)rt_key(rt);
664 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
665 goto no_valid_rt;
666 }
667 else
668#endif /* INET6 */
669 if ((rt = inp->inp_route.ro_rt) == NULL ||
670 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
671 == INADDR_ANY)
672 goto no_valid_rt;
673
674 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
675 i = tp->t_srtt *
676 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
677 if (rt->rt_rmx.rmx_rtt && i)
678 /*
679 * filter this update to half the old & half
680 * the new values, converting scale.
681 * See route.h and tcp_var.h for a
682 * description of the scaling constants.
683 */
684 rt->rt_rmx.rmx_rtt =
685 (rt->rt_rmx.rmx_rtt + i) / 2;
686 else
687 rt->rt_rmx.rmx_rtt = i;
688 tcpstat.tcps_cachedrtt++;
689 }
690 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
691 i = tp->t_rttvar *
692 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
693 if (rt->rt_rmx.rmx_rttvar && i)
694 rt->rt_rmx.rmx_rttvar =
695 (rt->rt_rmx.rmx_rttvar + i) / 2;
696 else
697 rt->rt_rmx.rmx_rttvar = i;
698 tcpstat.tcps_cachedrttvar++;
699 }
700 /*
701 * The old comment here said:
702 * update the pipelimit (ssthresh) if it has been updated
703 * already or if a pipesize was specified & the threshhold
704 * got below half the pipesize. I.e., wait for bad news
705 * before we start updating, then update on both good
706 * and bad news.
707 *
708 * But we want to save the ssthresh even if no pipesize is
709 * specified explicitly in the route, because such
710 * connections still have an implicit pipesize specified
711 * by the global tcp_sendspace. In the absence of a reliable
712 * way to calculate the pipesize, it will have to do.
713 */
714 i = tp->snd_ssthresh;
715 if (rt->rt_rmx.rmx_sendpipe != 0)
716 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
717 else
718 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
719 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
720 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
721 || dosavessthresh) {
722 /*
723 * convert the limit from user data bytes to
724 * packets then to packet data bytes.
725 */
726 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
727 if (i < 2)
728 i = 2;
729 i *= (u_long)(tp->t_maxseg +
730#ifdef INET6
731 (isipv6 ? sizeof (struct ip6_hdr) +
732 sizeof (struct tcphdr) :
733#endif
734 sizeof (struct tcpiphdr)
735#ifdef INET6
736 )
737#endif
738 );
739 if (rt->rt_rmx.rmx_ssthresh)
740 rt->rt_rmx.rmx_ssthresh =
741 (rt->rt_rmx.rmx_ssthresh + i) / 2;
742 else
743 rt->rt_rmx.rmx_ssthresh = i;
744 tcpstat.tcps_cachedssthresh++;
745 }
746 }
747 no_valid_rt:
748 /* free the reassembly queue, if any */
749 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
750 LIST_REMOVE(q, tqe_q);
751 m_freem(q->tqe_m);
752 FREE(q, M_TSEGQ);
753 }
754 inp->inp_ppcb = NULL;
755 soisdisconnected(so);
756#ifdef INET6
757 if (INP_CHECK_SOCKAF(so, AF_INET6))
758 in6_pcbdetach(inp);
759 else
760#endif /* INET6 */
761 in_pcbdetach(inp);
762 tcpstat.tcps_closed++;
763 return ((struct tcpcb *)0);
764}
765
766void
767tcp_drain()
768{
769 if (do_tcpdrain)
770 {
771 struct inpcb *inpb;
772 struct tcpcb *tcpb;
773 struct tseg_qent *te;
774
775 /*
776 * Walk the tcpbs, if existing, and flush the reassembly queue,
777 * if there is one...
778 * XXX: The "Net/3" implementation doesn't imply that the TCP
779 * reassembly queue should be flushed, but in a situation
780 * where we're really low on mbufs, this is potentially
781 * usefull.
782 */
783 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
784 if ((tcpb = intotcpcb(inpb))) {
785 while ((te = LIST_FIRST(&tcpb->t_segq))
786 != NULL) {
787 LIST_REMOVE(te, tqe_q);
788 m_freem(te->tqe_m);
789 FREE(te, M_TSEGQ);
790 }
791 }
792 }
793
794 }
795}
796
797/*
798 * Notify a tcp user of an asynchronous error;
799 * store error as soft error, but wake up user
800 * (for now, won't do anything until can select for soft error).
801 *
802 * Do not wake up user since there currently is no mechanism for
803 * reporting soft errors (yet - a kqueue filter may be added).
804 */
805static void
806tcp_notify(inp, error)
807 struct inpcb *inp;
808 int error;
809{
810 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
811
812 /*
813 * Ignore some errors if we are hooked up.
814 * If connection hasn't completed, has retransmitted several times,
815 * and receives a second error, give up now. This is better
816 * than waiting a long time to establish a connection that
817 * can never complete.
818 */
819 if (tp->t_state == TCPS_ESTABLISHED &&
820 (error == EHOSTUNREACH || error == ENETUNREACH ||
821 error == EHOSTDOWN)) {
822 return;
823 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
824 tp->t_softerror)
825 tcp_drop(tp, error);
826 else
827 tp->t_softerror = error;
828#if 0
829 wakeup((caddr_t) &so->so_timeo);
830 sorwakeup(so);
831 sowwakeup(so);
832#endif
833}
834
835static int
836tcp_pcblist(SYSCTL_HANDLER_ARGS)
837{
838 int error, i, n, s;
839 struct inpcb *inp, **inp_list;
840 inp_gen_t gencnt;
841 struct xinpgen xig;
842
843 /*
844 * The process of preparing the TCB list is too time-consuming and
845 * resource-intensive to repeat twice on every request.
846 */
847 if (req->oldptr == 0) {
848 n = tcbinfo.ipi_count;
849 req->oldidx = 2 * (sizeof xig)
850 + (n + n/8) * sizeof(struct xtcpcb);
851 return 0;
852 }
853
854 if (req->newptr != 0)
855 return EPERM;
856
857 /*
858 * OK, now we're committed to doing something.
859 */
860 s = splnet();
861 gencnt = tcbinfo.ipi_gencnt;
862 n = tcbinfo.ipi_count;
863 splx(s);
864
865 xig.xig_len = sizeof xig;
866 xig.xig_count = n;
867 xig.xig_gen = gencnt;
868 xig.xig_sogen = so_gencnt;
869 error = SYSCTL_OUT(req, &xig, sizeof xig);
870 if (error)
871 return error;
872
873 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
874 if (inp_list == 0)
875 return ENOMEM;
876
877 s = splnet();
878 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
879 inp = LIST_NEXT(inp, inp_list)) {
dadab5e9 880 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp))
984263bc
MD
881 inp_list[i++] = inp;
882 }
883 splx(s);
884 n = i;
885
886 error = 0;
887 for (i = 0; i < n; i++) {
888 inp = inp_list[i];
889 if (inp->inp_gencnt <= gencnt) {
890 struct xtcpcb xt;
891 caddr_t inp_ppcb;
892 xt.xt_len = sizeof xt;
893 /* XXX should avoid extra copy */
894 bcopy(inp, &xt.xt_inp, sizeof *inp);
895 inp_ppcb = inp->inp_ppcb;
896 if (inp_ppcb != NULL)
897 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
898 else
899 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
900 if (inp->inp_socket)
901 sotoxsocket(inp->inp_socket, &xt.xt_socket);
902 error = SYSCTL_OUT(req, &xt, sizeof xt);
903 }
904 }
905 if (!error) {
906 /*
907 * Give the user an updated idea of our state.
908 * If the generation differs from what we told
909 * her before, she knows that something happened
910 * while we were processing this request, and it
911 * might be necessary to retry.
912 */
913 s = splnet();
914 xig.xig_gen = tcbinfo.ipi_gencnt;
915 xig.xig_sogen = so_gencnt;
916 xig.xig_count = tcbinfo.ipi_count;
917 splx(s);
918 error = SYSCTL_OUT(req, &xig, sizeof xig);
919 }
920 free(inp_list, M_TEMP);
921 return error;
922}
923
924SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
925 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
926
927static int
928tcp_getcred(SYSCTL_HANDLER_ARGS)
929{
930 struct sockaddr_in addrs[2];
931 struct inpcb *inp;
932 int error, s;
933
dadab5e9 934 error = suser(req->td);
984263bc
MD
935 if (error)
936 return (error);
937 error = SYSCTL_IN(req, addrs, sizeof(addrs));
938 if (error)
939 return (error);
940 s = splnet();
941 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
942 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
943 if (inp == NULL || inp->inp_socket == NULL) {
944 error = ENOENT;
945 goto out;
946 }
947 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
948out:
949 splx(s);
950 return (error);
951}
952
953SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
954 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
955
956#ifdef INET6
957static int
958tcp6_getcred(SYSCTL_HANDLER_ARGS)
959{
960 struct sockaddr_in6 addrs[2];
961 struct inpcb *inp;
962 int error, s, mapped = 0;
963
dadab5e9 964 error = suser_cred(req->p->p_ucred, 0);
984263bc
MD
965 if (error)
966 return (error);
967 error = SYSCTL_IN(req, addrs, sizeof(addrs));
968 if (error)
969 return (error);
970 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
971 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
972 mapped = 1;
973 else
974 return (EINVAL);
975 }
976 s = splnet();
977 if (mapped == 1)
978 inp = in_pcblookup_hash(&tcbinfo,
979 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
980 addrs[1].sin6_port,
981 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
982 addrs[0].sin6_port,
983 0, NULL);
984 else
985 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
986 addrs[1].sin6_port,
987 &addrs[0].sin6_addr, addrs[0].sin6_port,
988 0, NULL);
989 if (inp == NULL || inp->inp_socket == NULL) {
990 error = ENOENT;
991 goto out;
992 }
993 error = SYSCTL_OUT(req, inp->inp_socket->so_cred,
994 sizeof(struct ucred));
995out:
996 splx(s);
997 return (error);
998}
999
1000SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
1001 0, 0,
1002 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
1003#endif
1004
1005
1006void
1007tcp_ctlinput(cmd, sa, vip)
1008 int cmd;
1009 struct sockaddr *sa;
1010 void *vip;
1011{
1012 struct ip *ip = vip;
1013 struct tcphdr *th;
1014 struct in_addr faddr;
1015 struct inpcb *inp;
1016 struct tcpcb *tp;
1017 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1018 tcp_seq icmp_seq;
1019 int s;
1020
1021 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1022 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1023 return;
1024
1025 if (cmd == PRC_QUENCH)
1026 notify = tcp_quench;
1027 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1028 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1029 notify = tcp_drop_syn_sent;
1030 else if (cmd == PRC_MSGSIZE)
1031 notify = tcp_mtudisc;
1032 else if (PRC_IS_REDIRECT(cmd)) {
1033 ip = 0;
1034 notify = in_rtchange;
1035 } else if (cmd == PRC_HOSTDEAD)
1036 ip = 0;
1037 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
1038 return;
1039 if (ip) {
1040 s = splnet();
1041 th = (struct tcphdr *)((caddr_t)ip
1042 + (IP_VHL_HL(ip->ip_vhl) << 2));
1043 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1044 ip->ip_src, th->th_sport, 0, NULL);
1045 if (inp != NULL && inp->inp_socket != NULL) {
1046 icmp_seq = htonl(th->th_seq);
1047 tp = intotcpcb(inp);
1048 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1049 SEQ_LT(icmp_seq, tp->snd_max))
1050 (*notify)(inp, inetctlerrmap[cmd]);
1051 } else {
1052 struct in_conninfo inc;
1053
1054 inc.inc_fport = th->th_dport;
1055 inc.inc_lport = th->th_sport;
1056 inc.inc_faddr = faddr;
1057 inc.inc_laddr = ip->ip_src;
1058#ifdef INET6
1059 inc.inc_isipv6 = 0;
1060#endif
1061 syncache_unreach(&inc, th);
1062 }
1063 splx(s);
1064 } else
1065 in_pcbnotifyall(&tcb, faddr, inetctlerrmap[cmd], notify);
1066}
1067
1068#ifdef INET6
1069void
1070tcp6_ctlinput(cmd, sa, d)
1071 int cmd;
1072 struct sockaddr *sa;
1073 void *d;
1074{
1075 struct tcphdr th;
1076 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1077 struct ip6_hdr *ip6;
1078 struct mbuf *m;
1079 struct ip6ctlparam *ip6cp = NULL;
1080 const struct sockaddr_in6 *sa6_src = NULL;
1081 int off;
1082 struct tcp_portonly {
1083 u_int16_t th_sport;
1084 u_int16_t th_dport;
1085 } *thp;
1086
1087 if (sa->sa_family != AF_INET6 ||
1088 sa->sa_len != sizeof(struct sockaddr_in6))
1089 return;
1090
1091 if (cmd == PRC_QUENCH)
1092 notify = tcp_quench;
1093 else if (cmd == PRC_MSGSIZE)
1094 notify = tcp_mtudisc;
1095 else if (!PRC_IS_REDIRECT(cmd) &&
1096 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1097 return;
1098
1099 /* if the parameter is from icmp6, decode it. */
1100 if (d != NULL) {
1101 ip6cp = (struct ip6ctlparam *)d;
1102 m = ip6cp->ip6c_m;
1103 ip6 = ip6cp->ip6c_ip6;
1104 off = ip6cp->ip6c_off;
1105 sa6_src = ip6cp->ip6c_src;
1106 } else {
1107 m = NULL;
1108 ip6 = NULL;
1109 off = 0; /* fool gcc */
1110 sa6_src = &sa6_any;
1111 }
1112
1113 if (ip6) {
1114 struct in_conninfo inc;
1115 /*
1116 * XXX: We assume that when IPV6 is non NULL,
1117 * M and OFF are valid.
1118 */
1119
1120 /* check if we can safely examine src and dst ports */
1121 if (m->m_pkthdr.len < off + sizeof(*thp))
1122 return;
1123
1124 bzero(&th, sizeof(th));
1125 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1126
1127 in6_pcbnotify(&tcb, sa, th.th_dport,
1128 (struct sockaddr *)ip6cp->ip6c_src,
1129 th.th_sport, cmd, notify);
1130
1131 inc.inc_fport = th.th_dport;
1132 inc.inc_lport = th.th_sport;
1133 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1134 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1135 inc.inc_isipv6 = 1;
1136 syncache_unreach(&inc, &th);
1137 } else
1138 in6_pcbnotify(&tcb, sa, 0, (struct sockaddr *)sa6_src,
1139 0, cmd, notify);
1140}
1141#endif /* INET6 */
1142
1143
1144/*
1145 * Following is where TCP initial sequence number generation occurs.
1146 *
1147 * There are two places where we must use initial sequence numbers:
1148 * 1. In SYN-ACK packets.
1149 * 2. In SYN packets.
1150 *
1151 * All ISNs for SYN-ACK packets are generated by the syncache. See
1152 * tcp_syncache.c for details.
1153 *
1154 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1155 * depends on this property. In addition, these ISNs should be
1156 * unguessable so as to prevent connection hijacking. To satisfy
1157 * the requirements of this situation, the algorithm outlined in
1158 * RFC 1948 is used to generate sequence numbers.
1159 *
1160 * Implementation details:
1161 *
1162 * Time is based off the system timer, and is corrected so that it
1163 * increases by one megabyte per second. This allows for proper
1164 * recycling on high speed LANs while still leaving over an hour
1165 * before rollover.
1166 *
1167 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1168 * between seeding of isn_secret. This is normally set to zero,
1169 * as reseeding should not be necessary.
1170 *
1171 */
1172
1173#define ISN_BYTES_PER_SECOND 1048576
1174
1175u_char isn_secret[32];
1176int isn_last_reseed;
1177MD5_CTX isn_ctx;
1178
1179tcp_seq
1180tcp_new_isn(tp)
1181 struct tcpcb *tp;
1182{
1183 u_int32_t md5_buffer[4];
1184 tcp_seq new_isn;
1185
1186 /* Seed if this is the first use, reseed if requested. */
1187 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1188 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1189 < (u_int)ticks))) {
1190 read_random_unlimited(&isn_secret, sizeof(isn_secret));
1191 isn_last_reseed = ticks;
1192 }
1193
1194 /* Compute the md5 hash and return the ISN. */
1195 MD5Init(&isn_ctx);
1196 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1197 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1198#ifdef INET6
1199 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1200 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1201 sizeof(struct in6_addr));
1202 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1203 sizeof(struct in6_addr));
1204 } else
1205#endif
1206 {
1207 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1208 sizeof(struct in_addr));
1209 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1210 sizeof(struct in_addr));
1211 }
1212 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1213 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1214 new_isn = (tcp_seq) md5_buffer[0];
1215 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
1216 return new_isn;
1217}
1218
1219/*
1220 * When a source quench is received, close congestion window
1221 * to one segment. We will gradually open it again as we proceed.
1222 */
1223void
1224tcp_quench(inp, errno)
1225 struct inpcb *inp;
1226 int errno;
1227{
1228 struct tcpcb *tp = intotcpcb(inp);
1229
1230 if (tp)
1231 tp->snd_cwnd = tp->t_maxseg;
1232}
1233
1234/*
1235 * When a specific ICMP unreachable message is received and the
1236 * connection state is SYN-SENT, drop the connection. This behavior
1237 * is controlled by the icmp_may_rst sysctl.
1238 */
1239void
1240tcp_drop_syn_sent(inp, errno)
1241 struct inpcb *inp;
1242 int errno;
1243{
1244 struct tcpcb *tp = intotcpcb(inp);
1245
1246 if (tp && tp->t_state == TCPS_SYN_SENT)
1247 tcp_drop(tp, errno);
1248}
1249
1250/*
1251 * When `need fragmentation' ICMP is received, update our idea of the MSS
1252 * based on the new value in the route. Also nudge TCP to send something,
1253 * since we know the packet we just sent was dropped.
1254 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1255 */
1256void
1257tcp_mtudisc(inp, errno)
1258 struct inpcb *inp;
1259 int errno;
1260{
1261 struct tcpcb *tp = intotcpcb(inp);
1262 struct rtentry *rt;
1263 struct rmxp_tao *taop;
1264 struct socket *so = inp->inp_socket;
1265 int offered;
1266 int mss;
1267#ifdef INET6
1268 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1269#endif /* INET6 */
1270
1271 if (tp) {
1272#ifdef INET6
1273 if (isipv6)
1274 rt = tcp_rtlookup6(&inp->inp_inc);
1275 else
1276#endif /* INET6 */
1277 rt = tcp_rtlookup(&inp->inp_inc);
1278 if (!rt || !rt->rt_rmx.rmx_mtu) {
1279 tp->t_maxopd = tp->t_maxseg =
1280#ifdef INET6
1281 isipv6 ? tcp_v6mssdflt :
1282#endif /* INET6 */
1283 tcp_mssdflt;
1284 return;
1285 }
1286 taop = rmx_taop(rt->rt_rmx);
1287 offered = taop->tao_mssopt;
1288 mss = rt->rt_rmx.rmx_mtu -
1289#ifdef INET6
1290 (isipv6 ?
1291 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1292#endif /* INET6 */
1293 sizeof(struct tcpiphdr)
1294#ifdef INET6
1295 )
1296#endif /* INET6 */
1297 ;
1298
1299 if (offered)
1300 mss = min(mss, offered);
1301 /*
1302 * XXX - The above conditional probably violates the TCP
1303 * spec. The problem is that, since we don't know the
1304 * other end's MSS, we are supposed to use a conservative
1305 * default. But, if we do that, then MTU discovery will
1306 * never actually take place, because the conservative
1307 * default is much less than the MTUs typically seen
1308 * on the Internet today. For the moment, we'll sweep
1309 * this under the carpet.
1310 *
1311 * The conservative default might not actually be a problem
1312 * if the only case this occurs is when sending an initial
1313 * SYN with options and data to a host we've never talked
1314 * to before. Then, they will reply with an MSS value which
1315 * will get recorded and the new parameters should get
1316 * recomputed. For Further Study.
1317 */
1318 if (tp->t_maxopd <= mss)
1319 return;
1320 tp->t_maxopd = mss;
1321
1322 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1323 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1324 mss -= TCPOLEN_TSTAMP_APPA;
1325 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1326 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1327 mss -= TCPOLEN_CC_APPA;
1328#if (MCLBYTES & (MCLBYTES - 1)) == 0
1329 if (mss > MCLBYTES)
1330 mss &= ~(MCLBYTES-1);
1331#else
1332 if (mss > MCLBYTES)
1333 mss = mss / MCLBYTES * MCLBYTES;
1334#endif
1335 if (so->so_snd.sb_hiwat < mss)
1336 mss = so->so_snd.sb_hiwat;
1337
1338 tp->t_maxseg = mss;
1339
1340 tcpstat.tcps_mturesent++;
1341 tp->t_rtttime = 0;
1342 tp->snd_nxt = tp->snd_una;
1343 tcp_output(tp);
1344 }
1345}
1346
1347/*
1348 * Look-up the routing entry to the peer of this inpcb. If no route
1349 * is found and it cannot be allocated the return NULL. This routine
1350 * is called by TCP routines that access the rmx structure and by tcp_mss
1351 * to get the interface MTU.
1352 */
1353struct rtentry *
1354tcp_rtlookup(inc)
1355 struct in_conninfo *inc;
1356{
1357 struct route *ro;
1358 struct rtentry *rt;
1359
1360 ro = &inc->inc_route;
1361 rt = ro->ro_rt;
1362 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1363 /* No route yet, so try to acquire one */
1364 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1365 ro->ro_dst.sa_family = AF_INET;
1366 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1367 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1368 inc->inc_faddr;
1369 rtalloc(ro);
1370 rt = ro->ro_rt;
1371 }
1372 }
1373 return rt;
1374}
1375
1376#ifdef INET6
1377struct rtentry *
1378tcp_rtlookup6(inc)
1379 struct in_conninfo *inc;
1380{
1381 struct route_in6 *ro6;
1382 struct rtentry *rt;
1383
1384 ro6 = &inc->inc6_route;
1385 rt = ro6->ro_rt;
1386 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1387 /* No route yet, so try to acquire one */
1388 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1389 ro6->ro_dst.sin6_family = AF_INET6;
1390 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1391 ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1392 rtalloc((struct route *)ro6);
1393 rt = ro6->ro_rt;
1394 }
1395 }
1396 return rt;
1397}
1398#endif /* INET6 */
1399
1400#ifdef IPSEC
1401/* compute ESP/AH header size for TCP, including outer IP header. */
1402size_t
1403ipsec_hdrsiz_tcp(tp)
1404 struct tcpcb *tp;
1405{
1406 struct inpcb *inp;
1407 struct mbuf *m;
1408 size_t hdrsiz;
1409 struct ip *ip;
1410#ifdef INET6
1411 struct ip6_hdr *ip6;
1412#endif /* INET6 */
1413 struct tcphdr *th;
1414
1415 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1416 return 0;
1417 MGETHDR(m, M_DONTWAIT, MT_DATA);
1418 if (!m)
1419 return 0;
1420
1421#ifdef INET6
1422 if ((inp->inp_vflag & INP_IPV6) != 0) {
1423 ip6 = mtod(m, struct ip6_hdr *);
1424 th = (struct tcphdr *)(ip6 + 1);
1425 m->m_pkthdr.len = m->m_len =
1426 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1427 tcp_fillheaders(tp, ip6, th);
1428 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1429 } else
1430#endif /* INET6 */
1431 {
1432 ip = mtod(m, struct ip *);
1433 th = (struct tcphdr *)(ip + 1);
1434 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1435 tcp_fillheaders(tp, ip, th);
1436 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1437 }
1438
1439 m_free(m);
1440 return hdrsiz;
1441}
1442#endif /*IPSEC*/
1443
1444/*
1445 * Return a pointer to the cached information about the remote host.
1446 * The cached information is stored in the protocol specific part of
1447 * the route metrics.
1448 */
1449struct rmxp_tao *
1450tcp_gettaocache(inc)
1451 struct in_conninfo *inc;
1452{
1453 struct rtentry *rt;
1454
1455#ifdef INET6
1456 if (inc->inc_isipv6)
1457 rt = tcp_rtlookup6(inc);
1458 else
1459#endif /* INET6 */
1460 rt = tcp_rtlookup(inc);
1461
1462 /* Make sure this is a host route and is up. */
1463 if (rt == NULL ||
1464 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1465 return NULL;
1466
1467 return rmx_taop(rt->rt_rmx);
1468}
1469
1470/*
1471 * Clear all the TAO cache entries, called from tcp_init.
1472 *
1473 * XXX
1474 * This routine is just an empty one, because we assume that the routing
1475 * routing tables are initialized at the same time when TCP, so there is
1476 * nothing in the cache left over.
1477 */
1478static void
1479tcp_cleartaocache()
1480{
1481}
1482
1483/*
1484 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1485 *
1486 * This code attempts to calculate the bandwidth-delay product as a
1487 * means of determining the optimal window size to maximize bandwidth,
1488 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1489 * routers. This code also does a fairly good job keeping RTTs in check
1490 * across slow links like modems. We implement an algorithm which is very
1491 * similar (but not meant to be) TCP/Vegas. The code operates on the
1492 * transmitter side of a TCP connection and so only effects the transmit
1493 * side of the connection.
1494 *
1495 * BACKGROUND: TCP makes no provision for the management of buffer space
1496 * at the end points or at the intermediate routers and switches. A TCP
1497 * stream, whether using NewReno or not, will eventually buffer as
1498 * many packets as it is able and the only reason this typically works is
1499 * due to the fairly small default buffers made available for a connection
1500 * (typicaly 16K or 32K). As machines use larger windows and/or window
1501 * scaling it is now fairly easy for even a single TCP connection to blow-out
1502 * all available buffer space not only on the local interface, but on
1503 * intermediate routers and switches as well. NewReno makes a misguided
1504 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1505 * then backing off, then steadily increasing the window again until another
1506 * failure occurs, ad-infinitum. This results in terrible oscillation that
1507 * is only made worse as network loads increase and the idea of intentionally
1508 * blowing out network buffers is, frankly, a terrible way to manage network
1509 * resources.
1510 *
1511 * It is far better to limit the transmit window prior to the failure
1512 * condition being achieved. There are two general ways to do this: First
1513 * you can 'scan' through different transmit window sizes and locate the
1514 * point where the RTT stops increasing, indicating that you have filled the
1515 * pipe, then scan backwards until you note that RTT stops decreasing, then
1516 * repeat ad-infinitum. This method works in principle but has severe
1517 * implementation issues due to RTT variances, timer granularity, and
1518 * instability in the algorithm which can lead to many false positives and
1519 * create oscillations as well as interact badly with other TCP streams
1520 * implementing the same algorithm.
1521 *
1522 * The second method is to limit the window to the bandwidth delay product
1523 * of the link. This is the method we implement. RTT variances and our
1524 * own manipulation of the congestion window, bwnd, can potentially
1525 * destabilize the algorithm. For this reason we have to stabilize the
1526 * elements used to calculate the window. We do this by using the minimum
1527 * observed RTT, the long term average of the observed bandwidth, and
1528 * by adding two segments worth of slop. It isn't perfect but it is able
1529 * to react to changing conditions and gives us a very stable basis on
1530 * which to extend the algorithm.
1531 */
1532void
1533tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1534{
1535 u_long bw;
1536 u_long bwnd;
1537 int save_ticks;
1538
1539 /*
1540 * If inflight_enable is disabled in the middle of a tcp connection,
1541 * make sure snd_bwnd is effectively disabled.
1542 */
1543 if (tcp_inflight_enable == 0) {
1544 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1545 tp->snd_bandwidth = 0;
1546 return;
1547 }
1548
1549 /*
1550 * Figure out the bandwidth. Due to the tick granularity this
1551 * is a very rough number and it MUST be averaged over a fairly
1552 * long period of time. XXX we need to take into account a link
1553 * that is not using all available bandwidth, but for now our
1554 * slop will ramp us up if this case occurs and the bandwidth later
1555 * increases.
1556 *
1557 * Note: if ticks rollover 'bw' may wind up negative. We must
1558 * effectively reset t_bw_rtttime for this case.
1559 */
1560 save_ticks = ticks;
1561 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1562 return;
1563
1564 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1565 (save_ticks - tp->t_bw_rtttime);
1566 tp->t_bw_rtttime = save_ticks;
1567 tp->t_bw_rtseq = ack_seq;
1568 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1569 return;
1570 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1571
1572 tp->snd_bandwidth = bw;
1573
1574 /*
1575 * Calculate the semi-static bandwidth delay product, plus two maximal
1576 * segments. The additional slop puts us squarely in the sweet
1577 * spot and also handles the bandwidth run-up case. Without the
1578 * slop we could be locking ourselves into a lower bandwidth.
1579 *
1580 * Situations Handled:
1581 * (1) Prevents over-queueing of packets on LANs, especially on
1582 * high speed LANs, allowing larger TCP buffers to be
1583 * specified, and also does a good job preventing
1584 * over-queueing of packets over choke points like modems
1585 * (at least for the transmit side).
1586 *
1587 * (2) Is able to handle changing network loads (bandwidth
1588 * drops so bwnd drops, bandwidth increases so bwnd
1589 * increases).
1590 *
1591 * (3) Theoretically should stabilize in the face of multiple
1592 * connections implementing the same algorithm (this may need
1593 * a little work).
1594 *
1595 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1596 * be adjusted with a sysctl but typically only needs to be on
1597 * very slow connections. A value no smaller then 5 should
1598 * be used, but only reduce this default if you have no other
1599 * choice.
1600 */
1601#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1602 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * (int)tp->t_maxseg / 10;
1603#undef USERTT
1604
1605 if (tcp_inflight_debug > 0) {
1606 static int ltime;
1607 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1608 ltime = ticks;
1609 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1610 tp,
1611 bw,
1612 tp->t_rttbest,
1613 tp->t_srtt,
1614 bwnd
1615 );
1616 }
1617 }
1618 if ((long)bwnd < tcp_inflight_min)
1619 bwnd = tcp_inflight_min;
1620 if (bwnd > tcp_inflight_max)
1621 bwnd = tcp_inflight_max;
1622 if ((long)bwnd < tp->t_maxseg * 2)
1623 bwnd = tp->t_maxseg * 2;
1624 tp->snd_bwnd = bwnd;
1625}
1626