Revamp the initial lwkt_abortmsg() support to normalize the abstraction. Now
[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 $
b44419cb 35 * $DragonFly: src/sys/netinet/tcp_subr.c,v 1.27 2004/04/20 01:52:28 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>
3f9db7f8 58#include <sys/in_cksum.h>
984263bc
MD
59
60#include <vm/vm_zone.h>
61
62#include <net/route.h>
63#include <net/if.h>
0ddb6032 64#include <net/netisr.h>
984263bc 65
707ad4ed 66#define _IP_VHL
984263bc
MD
67#include <netinet/in.h>
68#include <netinet/in_systm.h>
69#include <netinet/ip.h>
984263bc 70#include <netinet/ip6.h>
984263bc 71#include <netinet/in_pcb.h>
984263bc 72#include <netinet6/in6_pcb.h>
984263bc
MD
73#include <netinet/in_var.h>
74#include <netinet/ip_var.h>
984263bc 75#include <netinet6/ip6_var.h>
984263bc
MD
76#include <netinet/tcp.h>
77#include <netinet/tcp_fsm.h>
78#include <netinet/tcp_seq.h>
79#include <netinet/tcp_timer.h>
80#include <netinet/tcp_var.h>
984263bc 81#include <netinet6/tcp6_var.h>
984263bc
MD
82#include <netinet/tcpip.h>
83#ifdef TCPDEBUG
84#include <netinet/tcp_debug.h>
85#endif
86#include <netinet6/ip6protosw.h>
87
88#ifdef IPSEC
89#include <netinet6/ipsec.h>
90#ifdef INET6
91#include <netinet6/ipsec6.h>
92#endif
707ad4ed 93#endif
984263bc
MD
94
95#ifdef FAST_IPSEC
96#include <netipsec/ipsec.h>
97#ifdef INET6
98#include <netipsec/ipsec6.h>
99#endif
100#define IPSEC
707ad4ed 101#endif
984263bc 102
984263bc
MD
103#include <sys/md5.h>
104
0ddb6032
JH
105#include <sys/msgport2.h>
106
707ad4ed 107int tcp_mssdflt = TCP_MSS;
984263bc 108SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
707ad4ed 109 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
984263bc
MD
110
111#ifdef INET6
707ad4ed
JH
112int tcp_v6mssdflt = TCP6_MSS;
113SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW,
114 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6");
984263bc
MD
115#endif
116
117#if 0
707ad4ed 118static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
984263bc 119SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
707ad4ed 120 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time");
984263bc
MD
121#endif
122
707ad4ed 123int tcp_do_rfc1323 = 1;
984263bc 124SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
707ad4ed 125 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
984263bc 126
707ad4ed 127int tcp_do_rfc1644 = 0;
984263bc 128SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
707ad4ed 129 &tcp_do_rfc1644, 0, "Enable rfc1644 (TTCP) extensions");
984263bc 130
707ad4ed 131static int tcp_tcbhashsize = 0;
984263bc 132SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
707ad4ed 133 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable");
984263bc 134
707ad4ed 135static int do_tcpdrain = 1;
984263bc
MD
136SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
137 "Enable tcp_drain routine for extra help when low on mbufs");
138
d371a63a 139/* XXX JH */
984263bc 140SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
d371a63a 141 &tcbinfo[0].ipi_count, 0, "Number of active PCBs");
984263bc 142
707ad4ed 143static int icmp_may_rst = 1;
984263bc
MD
144SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
145 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
146
707ad4ed 147static int tcp_isn_reseed_interval = 0;
984263bc
MD
148SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
149 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
150
151/*
152 * TCP bandwidth limiting sysctls. Note that the default lower bound of
153 * 1024 exists only for debugging. A good production default would be
154 * something like 6100.
155 */
707ad4ed 156static int tcp_inflight_enable = 0;
984263bc
MD
157SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW,
158 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
159
707ad4ed 160static int tcp_inflight_debug = 0;
984263bc
MD
161SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW,
162 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
163
707ad4ed 164static int tcp_inflight_min = 6144;
984263bc 165SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW,
707ad4ed 166 &tcp_inflight_min, 0, "Lower bound for TCP inflight window");
984263bc 167
707ad4ed 168static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
984263bc 169SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW,
707ad4ed 170 &tcp_inflight_max, 0, "Upper bound for TCP inflight window");
984263bc 171
707ad4ed 172static int tcp_inflight_stab = 20;
984263bc
MD
173SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW,
174 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 2 packets)");
175
707ad4ed
JH
176static void tcp_cleartaocache (void);
177static void tcp_notify (struct inpcb *, int);
984263bc 178
2b57d013
MD
179struct tcp_stats tcpstats_ary[MAXCPU];
180#ifdef SMP
181static int
182sysctl_tcpstats(SYSCTL_HANDLER_ARGS)
183{
707ad4ed 184 int cpu, error = 0;
2b57d013 185
707ad4ed 186 for (cpu = 0; cpu < ncpus; ++cpu) {
2b57d013 187 if ((error = SYSCTL_OUT(req, (void *)&tcpstats_ary[cpu],
707ad4ed 188 sizeof(struct tcp_stats))))
2b57d013
MD
189 break;
190 if ((error = SYSCTL_IN(req, (void *)&tcpstats_ary[cpu],
707ad4ed 191 sizeof(struct tcp_stats))))
2b57d013
MD
192 break;
193 }
194
195 return (error);
196}
707ad4ed
JH
197SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
198 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics");
199#else
2b57d013 200SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
707ad4ed 201 &tcpstat, tcp_stats, "TCP statistics");
2b57d013
MD
202#endif
203
984263bc
MD
204/*
205 * Target size of TCP PCB hash tables. Must be a power of two.
206 *
207 * Note that this can be overridden by the kernel environment
208 * variable net.inet.tcp.tcbhashsize
209 */
210#ifndef TCBHASHSIZE
707ad4ed 211#define TCBHASHSIZE 512
984263bc
MD
212#endif
213
214/*
215 * This is the actual shape of what we allocate using the zone
216 * allocator. Doing it this way allows us to protect both structures
217 * using the same generation count, and also eliminates the overhead
218 * of allocating tcpcbs separately. By hiding the structure here,
219 * we avoid changing most of the rest of the code (although it needs
220 * to be changed, eventually, for greater efficiency).
221 */
222#define ALIGNMENT 32
223#define ALIGNM1 (ALIGNMENT - 1)
224struct inp_tp {
225 union {
226 struct inpcb inp;
227 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
228 } inp_tp_u;
229 struct tcpcb tcb;
230 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
231 struct callout inp_tp_delack;
232};
233#undef ALIGNMENT
234#undef ALIGNM1
235
236/*
237 * Tcp initialization
238 */
239void
240tcp_init()
241{
d371a63a
JH
242 struct inpcbporthead *porthashbase;
243 u_long porthashmask;
ef962575 244 struct inpcontainerhead *wildcardhashbase;
2322f6ca 245 u_long wildcardhashmask;
d371a63a 246 struct vm_zone *ipi_zone;
984263bc 247 int hashsize = TCBHASHSIZE;
d371a63a 248 int cpu;
bf82f9b7 249
984263bc
MD
250 tcp_ccgen = 1;
251 tcp_cleartaocache();
252
253 tcp_delacktime = TCPTV_DELACK;
254 tcp_keepinit = TCPTV_KEEP_INIT;
255 tcp_keepidle = TCPTV_KEEP_IDLE;
256 tcp_keepintvl = TCPTV_KEEPINTVL;
257 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
258 tcp_msl = TCPTV_MSL;
259 tcp_rexmit_min = TCPTV_MIN;
260 tcp_rexmit_slop = TCPTV_CPU_VAR;
261
984263bc
MD
262 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
263 if (!powerof2(hashsize)) {
264 printf("WARNING: TCB hash size not a power of 2\n");
265 hashsize = 512; /* safe default */
266 }
267 tcp_tcbhashsize = hashsize;
d371a63a 268 porthashbase = hashinit(hashsize, M_PCB, &porthashmask);
2322f6ca 269 wildcardhashbase = hashinit(hashsize, M_PCB, &wildcardhashmask);
d371a63a
JH
270 ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
271 ZONE_INTERRUPT, 0);
272
273 for (cpu = 0; cpu < ncpus2; cpu++) {
274 LIST_INIT(&tcbinfo[cpu].listhead);
275 tcbinfo[cpu].hashbase = hashinit(hashsize, M_PCB,
276 &tcbinfo[cpu].hashmask);
277 tcbinfo[cpu].porthashbase = porthashbase;
278 tcbinfo[cpu].porthashmask = porthashmask;
2322f6ca
JH
279 tcbinfo[cpu].wildcardhashbase = wildcardhashbase;
280 tcbinfo[cpu].wildcardhashmask = wildcardhashmask;
d371a63a
JH
281 tcbinfo[cpu].ipi_zone = ipi_zone;
282 }
3edf7c37
RG
283
284 tcp_reass_maxseg = nmbclusters / 16;
707ad4ed 285 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg);
3edf7c37 286
984263bc 287#ifdef INET6
707ad4ed
JH
288#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
289#else
290#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
291#endif
984263bc
MD
292 if (max_protohdr < TCP_MINPROTOHDR)
293 max_protohdr = TCP_MINPROTOHDR;
294 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
295 panic("tcp_init");
296#undef TCP_MINPROTOHDR
297
2b57d013
MD
298 /*
299 * Initialize TCP statistics.
300 *
301 * It is layed out as an array which is has one element for UP,
302 * and SMP_MAXCPU elements for SMP. This allows us to retain
303 * the access mechanism from userland for both UP and SMP.
304 */
305#ifdef SMP
306 for (cpu = 0; cpu < ncpus; ++cpu) {
307 bzero(&tcpstats_ary[cpu], sizeof(struct tcp_stats));
308 }
309#else
310 bzero(&tcpstat, sizeof(struct tcp_stats));
311#endif
312
984263bc 313 syncache_init();
bf82f9b7 314 tcp_thread_init();
984263bc
MD
315}
316
317/*
318 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
319 * tcp_template used to store this data in mbufs, but we now recopy it out
320 * of the tcpcb each time to conserve mbufs.
321 */
322void
707ad4ed 323tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr)
984263bc
MD
324{
325 struct inpcb *inp = tp->t_inpcb;
326 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
327
328#ifdef INET6
707ad4ed 329 if (inp->inp_vflag & INP_IPV6) {
984263bc
MD
330 struct ip6_hdr *ip6;
331
332 ip6 = (struct ip6_hdr *)ip_ptr;
333 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
334 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
335 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
336 (IPV6_VERSION & IPV6_VERSION_MASK);
337 ip6->ip6_nxt = IPPROTO_TCP;
338 ip6->ip6_plen = sizeof(struct tcphdr);
339 ip6->ip6_src = inp->in6p_laddr;
340 ip6->ip6_dst = inp->in6p_faddr;
341 tcp_hdr->th_sum = 0;
342 } else
343#endif
344 {
707ad4ed
JH
345 struct ip *ip = (struct ip *) ip_ptr;
346
347 ip->ip_vhl = IP_VHL_BORING;
348 ip->ip_tos = 0;
349 ip->ip_len = 0;
350 ip->ip_id = 0;
351 ip->ip_off = 0;
352 ip->ip_ttl = 0;
353 ip->ip_sum = 0;
354 ip->ip_p = IPPROTO_TCP;
355 ip->ip_src = inp->inp_laddr;
356 ip->ip_dst = inp->inp_faddr;
357 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr,
358 ip->ip_dst.s_addr,
359 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
984263bc
MD
360 }
361
362 tcp_hdr->th_sport = inp->inp_lport;
363 tcp_hdr->th_dport = inp->inp_fport;
364 tcp_hdr->th_seq = 0;
365 tcp_hdr->th_ack = 0;
366 tcp_hdr->th_x2 = 0;
367 tcp_hdr->th_off = 5;
368 tcp_hdr->th_flags = 0;
369 tcp_hdr->th_win = 0;
370 tcp_hdr->th_urp = 0;
371}
372
373/*
374 * Create template to be used to send tcp packets on a connection.
375 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
376 * use for this function is in keepalives, which use tcp_respond.
377 */
378struct tcptemp *
707ad4ed 379tcp_maketemplate(struct tcpcb *tp)
984263bc
MD
380{
381 struct mbuf *m;
382 struct tcptemp *n;
383
384 m = m_get(M_DONTWAIT, MT_HEADER);
385 if (m == NULL)
707ad4ed 386 return (NULL);
984263bc
MD
387 m->m_len = sizeof(struct tcptemp);
388 n = mtod(m, struct tcptemp *);
389
390 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
391 return (n);
392}
393
394/*
395 * Send a single message to the TCP at address specified by
707ad4ed 396 * the given TCP/IP header. If m == NULL, then we make a copy
984263bc
MD
397 * of the tcpiphdr at ti and send directly to the addressed host.
398 * This is used to force keep alive messages out using the TCP
399 * template for a connection. If flags are given then we send
400 * a message back to the TCP which originated the * segment ti,
401 * and discard the mbuf containing it and any other attached mbufs.
402 *
403 * In any case the ack and sequence number of the transmitted
404 * segment are as specified by the parameters.
405 *
406 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
407 */
408void
707ad4ed
JH
409tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
410 tcp_seq ack, tcp_seq seq, int flags)
984263bc 411{
2256ba69 412 int tlen;
984263bc 413 int win = 0;
707ad4ed 414 struct route *ro = NULL;
984263bc 415 struct route sro;
707ad4ed 416 struct ip *ip = ipgen;
984263bc 417 struct tcphdr *nth;
984263bc 418 int ipflags = 0;
707ad4ed
JH
419 struct route_in6 *ro6 = NULL;
420 struct route_in6 sro6;
421 struct ip6_hdr *ip6 = ipgen;
984263bc 422#ifdef INET6
707ad4ed
JH
423 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6);
424#else
425 const boolean_t isipv6 = FALSE;
426#endif
984263bc 427
707ad4ed 428 if (tp != NULL) {
984263bc
MD
429 if (!(flags & TH_RST)) {
430 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
431 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
432 win = (long)TCP_MAXWIN << tp->rcv_scale;
433 }
984263bc
MD
434 if (isipv6)
435 ro6 = &tp->t_inpcb->in6p_route;
436 else
707ad4ed 437 ro = &tp->t_inpcb->inp_route;
984263bc 438 } else {
984263bc
MD
439 if (isipv6) {
440 ro6 = &sro6;
441 bzero(ro6, sizeof *ro6);
707ad4ed
JH
442 } else {
443 ro = &sro;
444 bzero(ro, sizeof *ro);
445 }
984263bc 446 }
707ad4ed 447 if (m == NULL) {
984263bc
MD
448 m = m_gethdr(M_DONTWAIT, MT_HEADER);
449 if (m == NULL)
450 return;
451 tlen = 0;
452 m->m_data += max_linkhdr;
984263bc 453 if (isipv6) {
707ad4ed 454 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr));
984263bc
MD
455 ip6 = mtod(m, struct ip6_hdr *);
456 nth = (struct tcphdr *)(ip6 + 1);
707ad4ed
JH
457 } else {
458 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
459 ip = mtod(m, struct ip *);
460 nth = (struct tcphdr *)(ip + 1);
461 }
462 bcopy(th, nth, sizeof(struct tcphdr));
984263bc
MD
463 flags = TH_ACK;
464 } else {
465 m_freem(m->m_next);
707ad4ed 466 m->m_next = NULL;
984263bc
MD
467 m->m_data = (caddr_t)ipgen;
468 /* m_len is set later */
469 tlen = 0;
707ad4ed 470#define xchg(a, b, type) { type t; t = a; a = b; b = t; }
984263bc
MD
471 if (isipv6) {
472 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
473 nth = (struct tcphdr *)(ip6 + 1);
707ad4ed
JH
474 } else {
475 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
476 nth = (struct tcphdr *)(ip + 1);
477 }
984263bc
MD
478 if (th != nth) {
479 /*
480 * this is usually a case when an extension header
481 * exists between the IPv6 header and the
482 * TCP header.
483 */
484 nth->th_sport = th->th_sport;
485 nth->th_dport = th->th_dport;
486 }
487 xchg(nth->th_dport, nth->th_sport, n_short);
488#undef xchg
489 }
984263bc
MD
490 if (isipv6) {
491 ip6->ip6_flow = 0;
492 ip6->ip6_vfc = IPV6_VERSION;
493 ip6->ip6_nxt = IPPROTO_TCP;
707ad4ed
JH
494 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen));
495 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
496 } else {
497 tlen += sizeof(struct tcpiphdr);
498 ip->ip_len = tlen;
499 ip->ip_ttl = ip_defttl;
500 }
984263bc
MD
501 m->m_len = tlen;
502 m->m_pkthdr.len = tlen;
707ad4ed 503 m->m_pkthdr.rcvif = (struct ifnet *) NULL;
984263bc
MD
504 nth->th_seq = htonl(seq);
505 nth->th_ack = htonl(ack);
506 nth->th_x2 = 0;
707ad4ed 507 nth->th_off = sizeof(struct tcphdr) >> 2;
984263bc 508 nth->th_flags = flags;
707ad4ed 509 if (tp != NULL)
984263bc
MD
510 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
511 else
512 nth->th_win = htons((u_short)win);
513 nth->th_urp = 0;
984263bc
MD
514 if (isipv6) {
515 nth->th_sum = 0;
516 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
517 sizeof(struct ip6_hdr),
518 tlen - sizeof(struct ip6_hdr));
519 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
707ad4ed
JH
520 (ro6 && ro6->ro_rt) ?
521 ro6->ro_rt->rt_ifp : NULL);
522 } else {
523 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
524 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
525 m->m_pkthdr.csum_flags = CSUM_TCP;
526 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
527 }
984263bc
MD
528#ifdef TCPDEBUG
529 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
530 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
531#endif
984263bc
MD
532 if (isipv6) {
533 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL,
707ad4ed
JH
534 tp ? tp->t_inpcb : NULL);
535 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) {
984263bc
MD
536 RTFREE(ro6->ro_rt);
537 ro6->ro_rt = NULL;
538 }
707ad4ed
JH
539 } else {
540 (void)ip_output(m, NULL, ro, ipflags, NULL,
541 tp ? tp->t_inpcb : NULL);
542 if ((ro == &sro) && (ro->ro_rt != NULL)) {
543 RTFREE(ro->ro_rt);
544 ro->ro_rt = NULL;
545 }
984263bc 546 }
984263bc
MD
547}
548
549/*
550 * Create a new TCP control block, making an
551 * empty reassembly queue and hooking it to the argument
552 * protocol control block. The `inp' parameter must have
553 * come from the zone allocator set up in tcp_init().
554 */
555struct tcpcb *
707ad4ed 556tcp_newtcpcb(struct inpcb *inp)
984263bc
MD
557{
558 struct inp_tp *it;
2256ba69 559 struct tcpcb *tp;
984263bc 560#ifdef INET6
707ad4ed
JH
561 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
562#else
563 const boolean_t isipv6 = FALSE;
564#endif
984263bc
MD
565
566 it = (struct inp_tp *)inp;
567 tp = &it->tcb;
707ad4ed 568 bzero(tp, sizeof(struct tcpcb));
984263bc 569 LIST_INIT(&tp->t_segq);
707ad4ed 570 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
984263bc
MD
571
572 /* Set up our timeouts. */
573 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
574 callout_init(tp->tt_persist = &it->inp_tp_persist);
575 callout_init(tp->tt_keep = &it->inp_tp_keep);
576 callout_init(tp->tt_2msl = &it->inp_tp_2msl);
577 callout_init(tp->tt_delack = &it->inp_tp_delack);
578
579 if (tcp_do_rfc1323)
707ad4ed 580 tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP);
984263bc
MD
581 if (tcp_do_rfc1644)
582 tp->t_flags |= TF_REQ_CC;
583 tp->t_inpcb = inp; /* XXX */
584 /*
585 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
586 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
587 * reasonable initial retransmit time.
588 */
589 tp->t_srtt = TCPTV_SRTTBASE;
707ad4ed
JH
590 tp->t_rttvar =
591 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
984263bc
MD
592 tp->t_rttmin = tcp_rexmit_min;
593 tp->t_rxtcur = TCPTV_RTOBASE;
594 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
595 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
596 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
597 tp->t_rcvtime = ticks;
598 tp->t_bw_rtttime = ticks;
707ad4ed 599 /*
984263bc
MD
600 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
601 * because the socket may be bound to an IPv6 wildcard address,
602 * which may match an IPv4-mapped IPv6 address.
603 */
604 inp->inp_ip_ttl = ip_defttl;
605 inp->inp_ppcb = (caddr_t)tp;
606 return (tp); /* XXX */
607}
608
609/*
707ad4ed
JH
610 * Drop a TCP connection, reporting the specified error.
611 * If connection is synchronized, then send a RST to peer.
984263bc
MD
612 */
613struct tcpcb *
707ad4ed 614tcp_drop(struct tcpcb *tp, int errno)
984263bc
MD
615{
616 struct socket *so = tp->t_inpcb->inp_socket;
617
618 if (TCPS_HAVERCVDSYN(tp->t_state)) {
619 tp->t_state = TCPS_CLOSED;
620 (void) tcp_output(tp);
621 tcpstat.tcps_drops++;
622 } else
623 tcpstat.tcps_conndrops++;
624 if (errno == ETIMEDOUT && tp->t_softerror)
625 errno = tp->t_softerror;
626 so->so_error = errno;
627 return (tcp_close(tp));
628}
629
630/*
631 * Close a TCP control block:
632 * discard all space held by the tcp
633 * discard internet protocol block
634 * wake up any sleepers
635 */
636struct tcpcb *
707ad4ed 637tcp_close(struct tcpcb *tp)
984263bc 638{
2256ba69 639 struct tseg_qent *q;
984263bc
MD
640 struct inpcb *inp = tp->t_inpcb;
641 struct socket *so = inp->inp_socket;
2256ba69 642 struct rtentry *rt;
707ad4ed
JH
643 boolean_t dosavessthresh;
644#ifdef INET6
645 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
646#else
647 const boolean_t isipv6 = FALSE;
648#endif
984263bc
MD
649
650 /*
651 * Make sure that all of our timers are stopped before we
652 * delete the PCB.
653 */
654 callout_stop(tp->tt_rexmt);
655 callout_stop(tp->tt_persist);
656 callout_stop(tp->tt_keep);
657 callout_stop(tp->tt_2msl);
658 callout_stop(tp->tt_delack);
659
660 /*
661 * If we got enough samples through the srtt filter,
662 * save the rtt and rttvar in the routing entry.
663 * 'Enough' is arbitrarily defined as the 16 samples.
664 * 16 samples is enough for the srtt filter to converge
665 * to within 5% of the correct value; fewer samples and
666 * we could save a very bogus rtt.
667 *
668 * Don't update the default route's characteristics and don't
669 * update anything that the user "locked".
670 */
671 if (tp->t_rttupdated >= 16) {
2256ba69 672 u_long i = 0;
707ad4ed 673
984263bc
MD
674 if (isipv6) {
675 struct sockaddr_in6 *sin6;
676
677 if ((rt = inp->in6p_route.ro_rt) == NULL)
678 goto no_valid_rt;
679 sin6 = (struct sockaddr_in6 *)rt_key(rt);
680 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
681 goto no_valid_rt;
707ad4ed
JH
682 } else
683 if ((rt = inp->inp_route.ro_rt) == NULL ||
684 ((struct sockaddr_in *)rt_key(rt))->
685 sin_addr.s_addr == INADDR_ANY)
686 goto no_valid_rt;
687
688 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) {
689 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
984263bc
MD
690 if (rt->rt_rmx.rmx_rtt && i)
691 /*
692 * filter this update to half the old & half
693 * the new values, converting scale.
694 * See route.h and tcp_var.h for a
695 * description of the scaling constants.
696 */
697 rt->rt_rmx.rmx_rtt =
698 (rt->rt_rmx.rmx_rtt + i) / 2;
699 else
700 rt->rt_rmx.rmx_rtt = i;
701 tcpstat.tcps_cachedrtt++;
702 }
707ad4ed 703 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) {
984263bc
MD
704 i = tp->t_rttvar *
705 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
706 if (rt->rt_rmx.rmx_rttvar && i)
707 rt->rt_rmx.rmx_rttvar =
708 (rt->rt_rmx.rmx_rttvar + i) / 2;
709 else
710 rt->rt_rmx.rmx_rttvar = i;
711 tcpstat.tcps_cachedrttvar++;
712 }
713 /*
714 * The old comment here said:
715 * update the pipelimit (ssthresh) if it has been updated
716 * already or if a pipesize was specified & the threshhold
717 * got below half the pipesize. I.e., wait for bad news
718 * before we start updating, then update on both good
719 * and bad news.
720 *
721 * But we want to save the ssthresh even if no pipesize is
722 * specified explicitly in the route, because such
723 * connections still have an implicit pipesize specified
724 * by the global tcp_sendspace. In the absence of a reliable
725 * way to calculate the pipesize, it will have to do.
726 */
727 i = tp->snd_ssthresh;
728 if (rt->rt_rmx.rmx_sendpipe != 0)
707ad4ed 729 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2);
984263bc 730 else
707ad4ed
JH
731 dosavessthresh = (i < so->so_snd.sb_hiwat/2);
732 if (dosavessthresh ||
733 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) &&
734 (rt->rt_rmx.rmx_ssthresh != 0))) {
984263bc
MD
735 /*
736 * convert the limit from user data bytes to
737 * packets then to packet data bytes.
738 */
739 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
740 if (i < 2)
741 i = 2;
707ad4ed
JH
742 i *= tp->t_maxseg +
743 (isipv6 ?
744 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
745 sizeof(struct tcpiphdr));
984263bc
MD
746 if (rt->rt_rmx.rmx_ssthresh)
747 rt->rt_rmx.rmx_ssthresh =
748 (rt->rt_rmx.rmx_ssthresh + i) / 2;
749 else
750 rt->rt_rmx.rmx_ssthresh = i;
751 tcpstat.tcps_cachedssthresh++;
752 }
753 }
707ad4ed
JH
754
755no_valid_rt:
984263bc
MD
756 /* free the reassembly queue, if any */
757 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
758 LIST_REMOVE(q, tqe_q);
759 m_freem(q->tqe_m);
760 FREE(q, M_TSEGQ);
3edf7c37 761 tcp_reass_qsize--;
984263bc
MD
762 }
763 inp->inp_ppcb = NULL;
764 soisdisconnected(so);
765#ifdef INET6
766 if (INP_CHECK_SOCKAF(so, AF_INET6))
767 in6_pcbdetach(inp);
768 else
707ad4ed
JH
769#endif
770 in_pcbdetach(inp);
984263bc 771 tcpstat.tcps_closed++;
707ad4ed 772 return (NULL);
984263bc
MD
773}
774
3f48f9c5
JH
775static __inline void
776tcp_drain_oncpu(struct inpcbhead *head)
984263bc 777{
d371a63a
JH
778 struct inpcb *inpb;
779 struct tcpcb *tcpb;
780 struct tseg_qent *te;
3f48f9c5
JH
781
782 LIST_FOREACH(inpb, head, inp_list) {
783 if ((tcpb = intotcpcb(inpb))) {
784 while ((te = LIST_FIRST(&tcpb->t_segq)) != NULL) {
785 LIST_REMOVE(te, tqe_q);
786 m_freem(te->tqe_m);
787 FREE(te, M_TSEGQ);
788 tcp_reass_qsize--;
789 }
790 }
791 }
792}
793
0ddb6032
JH
794#ifdef SMP
795struct netmsg_tcp_drain {
796 struct lwkt_msg nm_lmsg;
0ddb6032
JH
797 struct inpcbhead *nm_head;
798};
799
b44419cb
MD
800static int
801tcp_drain_handler(lwkt_msg_t lmsg)
0ddb6032 802{
b44419cb 803 struct netmsg_tcp_drain *nm = (void *)lmsg;
0ddb6032
JH
804
805 tcp_drain_oncpu(nm->nm_head);
b44419cb
MD
806 lwkt_replymsg(lmsg, 0);
807 return(EASYNC);
0ddb6032
JH
808}
809#endif
810
3f48f9c5
JH
811void
812tcp_drain()
813{
93c8e032 814#ifdef SMP
d371a63a 815 int cpu;
93c8e032 816#endif
d371a63a
JH
817
818 if (!do_tcpdrain)
819 return;
984263bc
MD
820
821 /*
822 * Walk the tcpbs, if existing, and flush the reassembly queue,
823 * if there is one...
824 * XXX: The "Net/3" implementation doesn't imply that the TCP
707ad4ed
JH
825 * reassembly queue should be flushed, but in a situation
826 * where we're really low on mbufs, this is potentially
827 * useful.
984263bc 828 */
3f48f9c5 829#ifdef SMP
d371a63a 830 for (cpu = 0; cpu < ncpus2; cpu++) {
3f48f9c5
JH
831 struct netmsg_tcp_drain *msg;
832
0ddb6032 833 if (cpu == mycpu->gd_cpuid) {
3f48f9c5
JH
834 tcp_drain_oncpu(&tcbinfo[cpu].listhead);
835 } else {
836 msg = malloc(sizeof(struct netmsg_tcp_drain),
837 M_LWKTMSG, M_NOWAIT);
838 if (!msg)
839 continue;
b44419cb
MD
840 lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0,
841 lwkt_cmd_func(tcp_drain_handler),
842 lwkt_cmd_op_none);
3f48f9c5
JH
843 msg->nm_head = &tcbinfo[cpu].listhead;
844 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_lmsg);
984263bc 845 }
984263bc 846 }
3f48f9c5
JH
847#else
848 tcp_drain_oncpu(&tcbinfo[0].listhead);
849#endif
984263bc
MD
850}
851
852/*
853 * Notify a tcp user of an asynchronous error;
854 * store error as soft error, but wake up user
855 * (for now, won't do anything until can select for soft error).
856 *
857 * Do not wake up user since there currently is no mechanism for
858 * reporting soft errors (yet - a kqueue filter may be added).
859 */
860static void
707ad4ed 861tcp_notify(struct inpcb *inp, int error)
984263bc 862{
707ad4ed 863 struct tcpcb *tp = intotcpcb(inp);
984263bc
MD
864
865 /*
866 * Ignore some errors if we are hooked up.
867 * If connection hasn't completed, has retransmitted several times,
868 * and receives a second error, give up now. This is better
869 * than waiting a long time to establish a connection that
870 * can never complete.
871 */
872 if (tp->t_state == TCPS_ESTABLISHED &&
873 (error == EHOSTUNREACH || error == ENETUNREACH ||
874 error == EHOSTDOWN)) {
875 return;
876 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
877 tp->t_softerror)
878 tcp_drop(tp, error);
879 else
880 tp->t_softerror = error;
881#if 0
882 wakeup((caddr_t) &so->so_timeo);
883 sorwakeup(so);
884 sowwakeup(so);
885#endif
886}
887
888static int
889tcp_pcblist(SYSCTL_HANDLER_ARGS)
890{
891 int error, i, n, s;
892 struct inpcb *inp, **inp_list;
893 inp_gen_t gencnt;
894 struct xinpgen xig;
895
896 /*
897 * The process of preparing the TCB list is too time-consuming and
898 * resource-intensive to repeat twice on every request.
899 */
707ad4ed 900 if (req->oldptr == NULL) {
d371a63a 901 n = tcbinfo[mycpu->gd_cpuid].ipi_count;
707ad4ed
JH
902 req->oldidx = 2 * (sizeof xig) +
903 (n + n/8) * sizeof(struct xtcpcb);
904 return (0);
984263bc
MD
905 }
906
707ad4ed
JH
907 if (req->newptr != NULL)
908 return (EPERM);
984263bc
MD
909
910 /*
911 * OK, now we're committed to doing something.
912 */
913 s = splnet();
d371a63a
JH
914 gencnt = tcbinfo[mycpu->gd_cpuid].ipi_gencnt;
915 n = tcbinfo[mycpu->gd_cpuid].ipi_count;
984263bc
MD
916 splx(s);
917
918 xig.xig_len = sizeof xig;
919 xig.xig_count = n;
920 xig.xig_gen = gencnt;
921 xig.xig_sogen = so_gencnt;
922 error = SYSCTL_OUT(req, &xig, sizeof xig);
707ad4ed
JH
923 if (error != 0)
924 return (error);
984263bc
MD
925
926 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
707ad4ed
JH
927 if (inp_list == NULL)
928 return (ENOMEM);
929
984263bc 930 s = splnet();
d371a63a
JH
931 for (inp = LIST_FIRST(&tcbinfo[mycpu->gd_cpuid].listhead), i = 0;
932 inp && i < n; inp = LIST_NEXT(inp, inp_list)) {
dadab5e9 933 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp))
984263bc
MD
934 inp_list[i++] = inp;
935 }
936 splx(s);
937 n = i;
938
939 error = 0;
940 for (i = 0; i < n; i++) {
941 inp = inp_list[i];
942 if (inp->inp_gencnt <= gencnt) {
943 struct xtcpcb xt;
944 caddr_t inp_ppcb;
945 xt.xt_len = sizeof xt;
946 /* XXX should avoid extra copy */
947 bcopy(inp, &xt.xt_inp, sizeof *inp);
948 inp_ppcb = inp->inp_ppcb;
949 if (inp_ppcb != NULL)
950 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
951 else
707ad4ed 952 bzero(&xt.xt_tp, sizeof xt.xt_tp);
984263bc
MD
953 if (inp->inp_socket)
954 sotoxsocket(inp->inp_socket, &xt.xt_socket);
955 error = SYSCTL_OUT(req, &xt, sizeof xt);
956 }
957 }
958 if (!error) {
959 /*
960 * Give the user an updated idea of our state.
961 * If the generation differs from what we told
962 * her before, she knows that something happened
963 * while we were processing this request, and it
964 * might be necessary to retry.
965 */
966 s = splnet();
d371a63a 967 xig.xig_gen = tcbinfo[mycpu->gd_cpuid].ipi_gencnt;
984263bc 968 xig.xig_sogen = so_gencnt;
d371a63a 969 xig.xig_count = tcbinfo[mycpu->gd_cpuid].ipi_count;
984263bc
MD
970 splx(s);
971 error = SYSCTL_OUT(req, &xig, sizeof xig);
972 }
973 free(inp_list, M_TEMP);
707ad4ed 974 return (error);
984263bc
MD
975}
976
977SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
978 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
979
980static int
981tcp_getcred(SYSCTL_HANDLER_ARGS)
982{
983 struct sockaddr_in addrs[2];
984 struct inpcb *inp;
d371a63a 985 int cpu;
984263bc
MD
986 int error, s;
987
dadab5e9 988 error = suser(req->td);
707ad4ed 989 if (error != 0)
984263bc 990 return (error);
707ad4ed
JH
991 error = SYSCTL_IN(req, addrs, sizeof addrs);
992 if (error != 0)
984263bc
MD
993 return (error);
994 s = splnet();
6ca1a1cd 995
d371a63a
JH
996 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
997 addrs[0].sin_addr.s_addr, addrs[0].sin_port);
998 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr,
999 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
984263bc
MD
1000 if (inp == NULL || inp->inp_socket == NULL) {
1001 error = ENOENT;
1002 goto out;
1003 }
1004 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
1005out:
1006 splx(s);
1007 return (error);
1008}
1009
707ad4ed 1010SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
984263bc
MD
1011 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
1012
1013#ifdef INET6
1014static int
1015tcp6_getcred(SYSCTL_HANDLER_ARGS)
1016{
1017 struct sockaddr_in6 addrs[2];
1018 struct inpcb *inp;
707ad4ed
JH
1019 int error, s;
1020 boolean_t mapped = FALSE;
984263bc 1021
ab1c1714 1022 error = suser(req->td);
707ad4ed 1023 if (error != 0)
984263bc 1024 return (error);
707ad4ed
JH
1025 error = SYSCTL_IN(req, addrs, sizeof addrs);
1026 if (error != 0)
984263bc
MD
1027 return (error);
1028 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1029 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
707ad4ed 1030 mapped = TRUE;
984263bc
MD
1031 else
1032 return (EINVAL);
1033 }
1034 s = splnet();
707ad4ed 1035 if (mapped) {
d371a63a
JH
1036 inp = in_pcblookup_hash(&tcbinfo[0],
1037 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1038 addrs[1].sin6_port,
1039 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1040 addrs[0].sin6_port,
1041 0, NULL);
1042 } else {
1043 inp = in6_pcblookup_hash(&tcbinfo[0],
1044 &addrs[1].sin6_addr, addrs[1].sin6_port,
1045 &addrs[0].sin6_addr, addrs[0].sin6_port,
1046 0, NULL);
1047 }
984263bc
MD
1048 if (inp == NULL || inp->inp_socket == NULL) {
1049 error = ENOENT;
1050 goto out;
1051 }
707ad4ed 1052 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
984263bc
MD
1053out:
1054 splx(s);
1055 return (error);
1056}
1057
707ad4ed 1058SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
984263bc
MD
1059 0, 0,
1060 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
1061#endif
1062
984263bc 1063void
707ad4ed 1064tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
984263bc
MD
1065{
1066 struct ip *ip = vip;
1067 struct tcphdr *th;
1068 struct in_addr faddr;
1069 struct inpcb *inp;
1070 struct tcpcb *tp;
707ad4ed 1071 void (*notify)(struct inpcb *, int) = tcp_notify;
984263bc 1072 tcp_seq icmp_seq;
d371a63a 1073 int cpu;
984263bc
MD
1074 int s;
1075
1076 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1077 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1078 return;
1079
1080 if (cmd == PRC_QUENCH)
1081 notify = tcp_quench;
707ad4ed
JH
1082 else if (icmp_may_rst &&
1083 (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT ||
1084 cmd == PRC_TIMXCEED_INTRANS) &&
1085 ip != NULL)
984263bc
MD
1086 notify = tcp_drop_syn_sent;
1087 else if (cmd == PRC_MSGSIZE)
1088 notify = tcp_mtudisc;
1089 else if (PRC_IS_REDIRECT(cmd)) {
707ad4ed 1090 ip = NULL;
984263bc
MD
1091 notify = in_rtchange;
1092 } else if (cmd == PRC_HOSTDEAD)
707ad4ed 1093 ip = NULL;
984263bc
MD
1094 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
1095 return;
707ad4ed 1096 if (ip != NULL) {
984263bc 1097 s = splnet();
707ad4ed
JH
1098 th = (struct tcphdr *)((caddr_t)ip +
1099 (IP_VHL_HL(ip->ip_vhl) << 2));
d371a63a 1100 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
707ad4ed 1101 ip->ip_src.s_addr, th->th_sport);
d371a63a 1102 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport,
707ad4ed
JH
1103 ip->ip_src, th->th_sport, 0, NULL);
1104 if ((inp != NULL) && (inp->inp_socket != NULL)) {
984263bc
MD
1105 icmp_seq = htonl(th->th_seq);
1106 tp = intotcpcb(inp);
1107 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1108 SEQ_LT(icmp_seq, tp->snd_max))
1109 (*notify)(inp, inetctlerrmap[cmd]);
1110 } else {
1111 struct in_conninfo inc;
1112
1113 inc.inc_fport = th->th_dport;
1114 inc.inc_lport = th->th_sport;
1115 inc.inc_faddr = faddr;
1116 inc.inc_laddr = ip->ip_src;
1117#ifdef INET6
1118 inc.inc_isipv6 = 0;
1119#endif
1120 syncache_unreach(&inc, th);
1121 }
1122 splx(s);
d371a63a 1123 } else {
6ca1a1cd 1124 for (cpu = 0; cpu < ncpus2; cpu++) {
d371a63a
JH
1125 in_pcbnotifyall(&tcbinfo[cpu].listhead, faddr,
1126 inetctlerrmap[cmd], notify);
6ca1a1cd 1127 }
d371a63a 1128 }
984263bc
MD
1129}
1130
1131#ifdef INET6
1132void
707ad4ed 1133tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
984263bc
MD
1134{
1135 struct tcphdr th;
42a7fc75 1136 void (*notify) (struct inpcb *, int) = tcp_notify;
984263bc
MD
1137 struct ip6_hdr *ip6;
1138 struct mbuf *m;
1139 struct ip6ctlparam *ip6cp = NULL;
1140 const struct sockaddr_in6 *sa6_src = NULL;
1141 int off;
1142 struct tcp_portonly {
1143 u_int16_t th_sport;
1144 u_int16_t th_dport;
1145 } *thp;
1146
1147 if (sa->sa_family != AF_INET6 ||
1148 sa->sa_len != sizeof(struct sockaddr_in6))
1149 return;
1150
1151 if (cmd == PRC_QUENCH)
1152 notify = tcp_quench;
1153 else if (cmd == PRC_MSGSIZE)
1154 notify = tcp_mtudisc;
1155 else if (!PRC_IS_REDIRECT(cmd) &&
1156 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1157 return;
1158
1159 /* if the parameter is from icmp6, decode it. */
1160 if (d != NULL) {
1161 ip6cp = (struct ip6ctlparam *)d;
1162 m = ip6cp->ip6c_m;
1163 ip6 = ip6cp->ip6c_ip6;
1164 off = ip6cp->ip6c_off;
1165 sa6_src = ip6cp->ip6c_src;
1166 } else {
1167 m = NULL;
1168 ip6 = NULL;
1169 off = 0; /* fool gcc */
1170 sa6_src = &sa6_any;
1171 }
1172
707ad4ed 1173 if (ip6 != NULL) {
984263bc
MD
1174 struct in_conninfo inc;
1175 /*
1176 * XXX: We assume that when IPV6 is non NULL,
1177 * M and OFF are valid.
1178 */
1179
1180 /* check if we can safely examine src and dst ports */
707ad4ed 1181 if (m->m_pkthdr.len < off + sizeof *thp)
984263bc
MD
1182 return;
1183
707ad4ed
JH
1184 bzero(&th, sizeof th);
1185 m_copydata(m, off, sizeof *thp, (caddr_t)&th);
984263bc 1186
d371a63a 1187 in6_pcbnotify(&tcbinfo[0].listhead, sa, th.th_dport,
984263bc
MD
1188 (struct sockaddr *)ip6cp->ip6c_src,
1189 th.th_sport, cmd, notify);
1190
1191 inc.inc_fport = th.th_dport;
1192 inc.inc_lport = th.th_sport;
1193 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1194 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1195 inc.inc_isipv6 = 1;
1196 syncache_unreach(&inc, &th);
1197 } else
d371a63a
JH
1198 in6_pcbnotify(&tcbinfo[0].listhead, sa, 0,
1199 (const struct sockaddr *)sa6_src, 0, cmd, notify);
984263bc 1200}
707ad4ed 1201#endif
984263bc
MD
1202
1203/*
1204 * Following is where TCP initial sequence number generation occurs.
1205 *
1206 * There are two places where we must use initial sequence numbers:
1207 * 1. In SYN-ACK packets.
1208 * 2. In SYN packets.
1209 *
1210 * All ISNs for SYN-ACK packets are generated by the syncache. See
1211 * tcp_syncache.c for details.
1212 *
1213 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1214 * depends on this property. In addition, these ISNs should be
1215 * unguessable so as to prevent connection hijacking. To satisfy
1216 * the requirements of this situation, the algorithm outlined in
1217 * RFC 1948 is used to generate sequence numbers.
1218 *
1219 * Implementation details:
1220 *
1221 * Time is based off the system timer, and is corrected so that it
1222 * increases by one megabyte per second. This allows for proper
1223 * recycling on high speed LANs while still leaving over an hour
1224 * before rollover.
1225 *
1226 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1227 * between seeding of isn_secret. This is normally set to zero,
1228 * as reseeding should not be necessary.
1229 *
1230 */
1231
707ad4ed 1232#define ISN_BYTES_PER_SECOND 1048576
984263bc
MD
1233
1234u_char isn_secret[32];
1235int isn_last_reseed;
1236MD5_CTX isn_ctx;
1237
1238tcp_seq
707ad4ed 1239tcp_new_isn(struct tcpcb *tp)
984263bc
MD
1240{
1241 u_int32_t md5_buffer[4];
1242 tcp_seq new_isn;
1243
1244 /* Seed if this is the first use, reseed if requested. */
1245 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1246 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1247 < (u_int)ticks))) {
707ad4ed 1248 read_random_unlimited(&isn_secret, sizeof isn_secret);
984263bc
MD
1249 isn_last_reseed = ticks;
1250 }
707ad4ed 1251
984263bc
MD
1252 /* Compute the md5 hash and return the ISN. */
1253 MD5Init(&isn_ctx);
707ad4ed
JH
1254 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short));
1255 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short));
984263bc 1256#ifdef INET6
707ad4ed 1257 if (tp->t_inpcb->inp_vflag & INP_IPV6) {
984263bc
MD
1258 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1259 sizeof(struct in6_addr));
1260 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1261 sizeof(struct in6_addr));
1262 } else
1263#endif
1264 {
1265 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1266 sizeof(struct in_addr));
1267 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1268 sizeof(struct in_addr));
1269 }
1270 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1271 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1272 new_isn = (tcp_seq) md5_buffer[0];
1273 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
707ad4ed 1274 return (new_isn);
984263bc
MD
1275}
1276
1277/*
1278 * When a source quench is received, close congestion window
1279 * to one segment. We will gradually open it again as we proceed.
1280 */
1281void
707ad4ed 1282tcp_quench(struct inpcb *inp, int errno)
984263bc
MD
1283{
1284 struct tcpcb *tp = intotcpcb(inp);
1285
707ad4ed 1286 if (tp != NULL)
984263bc
MD
1287 tp->snd_cwnd = tp->t_maxseg;
1288}
1289
1290/*
1291 * When a specific ICMP unreachable message is received and the
1292 * connection state is SYN-SENT, drop the connection. This behavior
1293 * is controlled by the icmp_may_rst sysctl.
1294 */
1295void
707ad4ed 1296tcp_drop_syn_sent(struct inpcb *inp, int errno)
984263bc
MD
1297{
1298 struct tcpcb *tp = intotcpcb(inp);
1299
707ad4ed 1300 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT))
984263bc
MD
1301 tcp_drop(tp, errno);
1302}
1303
1304/*
1305 * When `need fragmentation' ICMP is received, update our idea of the MSS
1306 * based on the new value in the route. Also nudge TCP to send something,
1307 * since we know the packet we just sent was dropped.
1308 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1309 */
1310void
707ad4ed 1311tcp_mtudisc(struct inpcb *inp, int errno)
984263bc
MD
1312{
1313 struct tcpcb *tp = intotcpcb(inp);
1314 struct rtentry *rt;
1315 struct rmxp_tao *taop;
1316 struct socket *so = inp->inp_socket;
1317 int offered;
1318 int mss;
1319#ifdef INET6
707ad4ed
JH
1320 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0);
1321#else
1322 const boolean_t isipv6 = FALSE;
1323#endif
984263bc 1324
707ad4ed 1325 if (tp != NULL) {
984263bc
MD
1326 if (isipv6)
1327 rt = tcp_rtlookup6(&inp->inp_inc);
1328 else
707ad4ed
JH
1329 rt = tcp_rtlookup(&inp->inp_inc);
1330 if (rt == NULL || rt->rt_rmx.rmx_mtu == 0) {
984263bc 1331 tp->t_maxopd = tp->t_maxseg =
707ad4ed 1332 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
984263bc
MD
1333 return;
1334 }
1335 taop = rmx_taop(rt->rt_rmx);
1336 offered = taop->tao_mssopt;
1337 mss = rt->rt_rmx.rmx_mtu -
984263bc
MD
1338 (isipv6 ?
1339 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
707ad4ed 1340 sizeof(struct tcpiphdr));
984263bc 1341
707ad4ed 1342 if (offered != 0)
984263bc
MD
1343 mss = min(mss, offered);
1344 /*
1345 * XXX - The above conditional probably violates the TCP
1346 * spec. The problem is that, since we don't know the
1347 * other end's MSS, we are supposed to use a conservative
1348 * default. But, if we do that, then MTU discovery will
1349 * never actually take place, because the conservative
1350 * default is much less than the MTUs typically seen
1351 * on the Internet today. For the moment, we'll sweep
1352 * this under the carpet.
1353 *
1354 * The conservative default might not actually be a problem
1355 * if the only case this occurs is when sending an initial
1356 * SYN with options and data to a host we've never talked
1357 * to before. Then, they will reply with an MSS value which
1358 * will get recorded and the new parameters should get
1359 * recomputed. For Further Study.
1360 */
1361 if (tp->t_maxopd <= mss)
1362 return;
1363 tp->t_maxopd = mss;
1364
707ad4ed 1365 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
984263bc
MD
1366 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1367 mss -= TCPOLEN_TSTAMP_APPA;
707ad4ed 1368 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
984263bc
MD
1369 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1370 mss -= TCPOLEN_CC_APPA;
1371#if (MCLBYTES & (MCLBYTES - 1)) == 0
1372 if (mss > MCLBYTES)
707ad4ed 1373 mss &= ~(MCLBYTES - 1);
984263bc
MD
1374#else
1375 if (mss > MCLBYTES)
1376 mss = mss / MCLBYTES * MCLBYTES;
1377#endif
1378 if (so->so_snd.sb_hiwat < mss)
1379 mss = so->so_snd.sb_hiwat;
1380
1381 tp->t_maxseg = mss;
1382
1383 tcpstat.tcps_mturesent++;
1384 tp->t_rtttime = 0;
1385 tp->snd_nxt = tp->snd_una;
1386 tcp_output(tp);
1387 }
1388}
1389
1390/*
1391 * Look-up the routing entry to the peer of this inpcb. If no route
1392 * is found and it cannot be allocated the return NULL. This routine
1393 * is called by TCP routines that access the rmx structure and by tcp_mss
1394 * to get the interface MTU.
1395 */
1396struct rtentry *
707ad4ed 1397tcp_rtlookup(struct in_conninfo *inc)
984263bc
MD
1398{
1399 struct route *ro;
1400 struct rtentry *rt;
1401
1402 ro = &inc->inc_route;
1403 rt = ro->ro_rt;
1404 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1405 /* No route yet, so try to acquire one */
1406 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1407 ro->ro_dst.sa_family = AF_INET;
1408 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1409 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1410 inc->inc_faddr;
1411 rtalloc(ro);
1412 rt = ro->ro_rt;
1413 }
1414 }
707ad4ed 1415 return (rt);
984263bc
MD
1416}
1417
1418#ifdef INET6
1419struct rtentry *
707ad4ed 1420tcp_rtlookup6(struct in_conninfo *inc)
984263bc
MD
1421{
1422 struct route_in6 *ro6;
1423 struct rtentry *rt;
1424
1425 ro6 = &inc->inc6_route;
1426 rt = ro6->ro_rt;
1427 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1428 /* No route yet, so try to acquire one */
1429 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1430 ro6->ro_dst.sin6_family = AF_INET6;
1431 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1432 ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1433 rtalloc((struct route *)ro6);
1434 rt = ro6->ro_rt;
1435 }
1436 }
707ad4ed 1437 return (rt);
984263bc 1438}
707ad4ed 1439#endif
984263bc
MD
1440
1441#ifdef IPSEC
1442/* compute ESP/AH header size for TCP, including outer IP header. */
1443size_t
707ad4ed 1444ipsec_hdrsiz_tcp(struct tcpcb *tp)
984263bc
MD
1445{
1446 struct inpcb *inp;
1447 struct mbuf *m;
1448 size_t hdrsiz;
1449 struct ip *ip;
984263bc
MD
1450 struct tcphdr *th;
1451
1452 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
707ad4ed 1453 return (0);
984263bc
MD
1454 MGETHDR(m, M_DONTWAIT, MT_DATA);
1455 if (!m)
707ad4ed 1456 return (0);
984263bc
MD
1457
1458#ifdef INET6
707ad4ed
JH
1459 if (inp->inp_vflag & INP_IPV6) {
1460 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1461
984263bc
MD
1462 th = (struct tcphdr *)(ip6 + 1);
1463 m->m_pkthdr.len = m->m_len =
707ad4ed 1464 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
984263bc
MD
1465 tcp_fillheaders(tp, ip6, th);
1466 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1467 } else
707ad4ed
JH
1468#endif
1469 {
1470 ip = mtod(m, struct ip *);
1471 th = (struct tcphdr *)(ip + 1);
1472 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1473 tcp_fillheaders(tp, ip, th);
1474 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1475 }
984263bc
MD
1476
1477 m_free(m);
707ad4ed 1478 return (hdrsiz);
984263bc 1479}
707ad4ed 1480#endif
984263bc
MD
1481
1482/*
1483 * Return a pointer to the cached information about the remote host.
1484 * The cached information is stored in the protocol specific part of
1485 * the route metrics.
1486 */
1487struct rmxp_tao *
707ad4ed 1488tcp_gettaocache(struct in_conninfo *inc)
984263bc
MD
1489{
1490 struct rtentry *rt;
1491
1492#ifdef INET6
1493 if (inc->inc_isipv6)
1494 rt = tcp_rtlookup6(inc);
1495 else
707ad4ed
JH
1496#endif
1497 rt = tcp_rtlookup(inc);
984263bc
MD
1498
1499 /* Make sure this is a host route and is up. */
1500 if (rt == NULL ||
707ad4ed
JH
1501 (rt->rt_flags & (RTF_UP | RTF_HOST)) != (RTF_UP | RTF_HOST))
1502 return (NULL);
984263bc 1503
707ad4ed 1504 return (rmx_taop(rt->rt_rmx));
984263bc
MD
1505}
1506
1507/*
1508 * Clear all the TAO cache entries, called from tcp_init.
1509 *
1510 * XXX
1511 * This routine is just an empty one, because we assume that the routing
1512 * routing tables are initialized at the same time when TCP, so there is
1513 * nothing in the cache left over.
1514 */
1515static void
1516tcp_cleartaocache()
1517{
1518}
1519
1520/*
1521 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1522 *
1523 * This code attempts to calculate the bandwidth-delay product as a
1524 * means of determining the optimal window size to maximize bandwidth,
1525 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1526 * routers. This code also does a fairly good job keeping RTTs in check
1527 * across slow links like modems. We implement an algorithm which is very
1528 * similar (but not meant to be) TCP/Vegas. The code operates on the
1529 * transmitter side of a TCP connection and so only effects the transmit
1530 * side of the connection.
1531 *
1532 * BACKGROUND: TCP makes no provision for the management of buffer space
1533 * at the end points or at the intermediate routers and switches. A TCP
1534 * stream, whether using NewReno or not, will eventually buffer as
1535 * many packets as it is able and the only reason this typically works is
1536 * due to the fairly small default buffers made available for a connection
1537 * (typicaly 16K or 32K). As machines use larger windows and/or window
1538 * scaling it is now fairly easy for even a single TCP connection to blow-out
1539 * all available buffer space not only on the local interface, but on
1540 * intermediate routers and switches as well. NewReno makes a misguided
1541 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1542 * then backing off, then steadily increasing the window again until another
1543 * failure occurs, ad-infinitum. This results in terrible oscillation that
1544 * is only made worse as network loads increase and the idea of intentionally
1545 * blowing out network buffers is, frankly, a terrible way to manage network
1546 * resources.
1547 *
1548 * It is far better to limit the transmit window prior to the failure
1549 * condition being achieved. There are two general ways to do this: First
1550 * you can 'scan' through different transmit window sizes and locate the
1551 * point where the RTT stops increasing, indicating that you have filled the
1552 * pipe, then scan backwards until you note that RTT stops decreasing, then
1553 * repeat ad-infinitum. This method works in principle but has severe
1554 * implementation issues due to RTT variances, timer granularity, and
1555 * instability in the algorithm which can lead to many false positives and
1556 * create oscillations as well as interact badly with other TCP streams
1557 * implementing the same algorithm.
1558 *
1559 * The second method is to limit the window to the bandwidth delay product
1560 * of the link. This is the method we implement. RTT variances and our
1561 * own manipulation of the congestion window, bwnd, can potentially
1562 * destabilize the algorithm. For this reason we have to stabilize the
1563 * elements used to calculate the window. We do this by using the minimum
1564 * observed RTT, the long term average of the observed bandwidth, and
1565 * by adding two segments worth of slop. It isn't perfect but it is able
1566 * to react to changing conditions and gives us a very stable basis on
1567 * which to extend the algorithm.
1568 */
1569void
1570tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1571{
1572 u_long bw;
1573 u_long bwnd;
1574 int save_ticks;
1575
1576 /*
1577 * If inflight_enable is disabled in the middle of a tcp connection,
1578 * make sure snd_bwnd is effectively disabled.
1579 */
707ad4ed 1580 if (!tcp_inflight_enable) {
984263bc
MD
1581 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1582 tp->snd_bandwidth = 0;
1583 return;
1584 }
1585
1586 /*
1587 * Figure out the bandwidth. Due to the tick granularity this
1588 * is a very rough number and it MUST be averaged over a fairly
1589 * long period of time. XXX we need to take into account a link
1590 * that is not using all available bandwidth, but for now our
1591 * slop will ramp us up if this case occurs and the bandwidth later
1592 * increases.
1593 *
1594 * Note: if ticks rollover 'bw' may wind up negative. We must
1595 * effectively reset t_bw_rtttime for this case.
1596 */
1597 save_ticks = ticks;
1598 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1599 return;
1600
1601 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1602 (save_ticks - tp->t_bw_rtttime);
1603 tp->t_bw_rtttime = save_ticks;
1604 tp->t_bw_rtseq = ack_seq;
1605 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1606 return;
1607 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1608
1609 tp->snd_bandwidth = bw;
1610
1611 /*
1612 * Calculate the semi-static bandwidth delay product, plus two maximal
1613 * segments. The additional slop puts us squarely in the sweet
1614 * spot and also handles the bandwidth run-up case. Without the
1615 * slop we could be locking ourselves into a lower bandwidth.
1616 *
1617 * Situations Handled:
1618 * (1) Prevents over-queueing of packets on LANs, especially on
1619 * high speed LANs, allowing larger TCP buffers to be
1620 * specified, and also does a good job preventing
1621 * over-queueing of packets over choke points like modems
1622 * (at least for the transmit side).
1623 *
1624 * (2) Is able to handle changing network loads (bandwidth
1625 * drops so bwnd drops, bandwidth increases so bwnd
1626 * increases).
1627 *
1628 * (3) Theoretically should stabilize in the face of multiple
1629 * connections implementing the same algorithm (this may need
1630 * a little work).
1631 *
1632 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1633 * be adjusted with a sysctl but typically only needs to be on
1634 * very slow connections. A value no smaller then 5 should
1635 * be used, but only reduce this default if you have no other
1636 * choice.
1637 */
707ad4ed
JH
1638
1639#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1640 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) +
1641 tcp_inflight_stab * (int)tp->t_maxseg / 10;
984263bc
MD
1642#undef USERTT
1643
1644 if (tcp_inflight_debug > 0) {
1645 static int ltime;
1646 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1647 ltime = ticks;
1648 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
707ad4ed 1649 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd);
984263bc
MD
1650 }
1651 }
1652 if ((long)bwnd < tcp_inflight_min)
1653 bwnd = tcp_inflight_min;
1654 if (bwnd > tcp_inflight_max)
1655 bwnd = tcp_inflight_max;
1656 if ((long)bwnd < tp->t_maxseg * 2)
1657 bwnd = tp->t_maxseg * 2;
1658 tp->snd_bwnd = bwnd;
1659}