kernel - Replace zalloc zones with kmalloc for PCBs
[dragonfly.git] / sys / netinet / tcp_subr.c
CommitLineData
984263bc 1/*
66d6c637
JH
2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
f23061d4 4 *
66d6c637
JH
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
f23061d4 7 *
66d6c637
JH
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
f23061d4 19 *
66d6c637
JH
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
66d6c637 34/*
984263bc
MD
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $
14572273 68 * $DragonFly: src/sys/netinet/tcp_subr.c,v 1.63 2008/11/11 10:46:58 sephe Exp $
984263bc
MD
69 */
70
71#include "opt_compat.h"
72#include "opt_inet6.h"
73#include "opt_ipsec.h"
74#include "opt_tcpdebug.h"
75
76#include <sys/param.h>
77#include <sys/systm.h>
78#include <sys/callout.h>
79#include <sys/kernel.h>
80#include <sys/sysctl.h>
81#include <sys/malloc.h>
dd2b0fb4 82#include <sys/mpipe.h>
984263bc
MD
83#include <sys/mbuf.h>
84#ifdef INET6
85#include <sys/domain.h>
86#endif
87#include <sys/proc.h>
895c1f85 88#include <sys/priv.h>
984263bc
MD
89#include <sys/socket.h>
90#include <sys/socketvar.h>
91#include <sys/protosw.h>
92#include <sys/random.h>
3f9db7f8 93#include <sys/in_cksum.h>
c7afbe76 94#include <sys/ktr.h>
984263bc 95
984263bc
MD
96#include <net/route.h>
97#include <net/if.h>
0ddb6032 98#include <net/netisr.h>
984263bc 99
707ad4ed 100#define _IP_VHL
984263bc
MD
101#include <netinet/in.h>
102#include <netinet/in_systm.h>
103#include <netinet/ip.h>
984263bc 104#include <netinet/ip6.h>
984263bc 105#include <netinet/in_pcb.h>
984263bc 106#include <netinet6/in6_pcb.h>
984263bc
MD
107#include <netinet/in_var.h>
108#include <netinet/ip_var.h>
984263bc 109#include <netinet6/ip6_var.h>
7d448528
JH
110#include <netinet/ip_icmp.h>
111#ifdef INET6
112#include <netinet/icmp6.h>
113#endif
984263bc
MD
114#include <netinet/tcp.h>
115#include <netinet/tcp_fsm.h>
116#include <netinet/tcp_seq.h>
117#include <netinet/tcp_timer.h>
a48c5dd5 118#include <netinet/tcp_timer2.h>
984263bc 119#include <netinet/tcp_var.h>
984263bc 120#include <netinet6/tcp6_var.h>
984263bc
MD
121#include <netinet/tcpip.h>
122#ifdef TCPDEBUG
123#include <netinet/tcp_debug.h>
124#endif
125#include <netinet6/ip6protosw.h>
126
127#ifdef IPSEC
128#include <netinet6/ipsec.h>
129#ifdef INET6
130#include <netinet6/ipsec6.h>
131#endif
707ad4ed 132#endif
984263bc
MD
133
134#ifdef FAST_IPSEC
bf844ffa 135#include <netproto/ipsec/ipsec.h>
984263bc 136#ifdef INET6
bf844ffa 137#include <netproto/ipsec/ipsec6.h>
984263bc
MD
138#endif
139#define IPSEC
707ad4ed 140#endif
984263bc 141
984263bc 142#include <sys/md5.h>
d2e9e54c
MD
143#include <machine/smp.h>
144
684a93c4
MD
145#include <sys/msgport2.h>
146#include <sys/mplock2.h>
4599cf19
MD
147#include <net/netmsg2.h>
148
c7afbe76
MD
149#if !defined(KTR_TCP)
150#define KTR_TCP KTR_ALL
151#endif
152KTR_INFO_MASTER(tcp);
153KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0);
154KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0);
155KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0);
156#define logtcp(name) KTR_LOG(tcp_ ## name)
157
2b1ce38a
MD
158struct inpcbinfo tcbinfo[MAXCPU];
159struct tcpcbackqhead tcpcbackq[MAXCPU];
160
92db3805
SZ
161int tcp_mpsafe_proto = 0;
162TUNABLE_INT("net.inet.tcp.mpsafe_proto", &tcp_mpsafe_proto);
163
730902da 164static int tcp_mpsafe_thread = NETMSG_SERVICE_ADAPTIVE;
92db3805
SZ
165TUNABLE_INT("net.inet.tcp.mpsafe_thread", &tcp_mpsafe_thread);
166SYSCTL_INT(_net_inet_tcp, OID_AUTO, mpsafe_thread, CTLFLAG_RW,
167 &tcp_mpsafe_thread, 0,
168 "0:BGL, 1:Adaptive BGL, 2:No BGL(experimental)");
169
707ad4ed 170int tcp_mssdflt = TCP_MSS;
f23061d4 171SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
707ad4ed 172 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
984263bc
MD
173
174#ifdef INET6
707ad4ed
JH
175int tcp_v6mssdflt = TCP6_MSS;
176SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW,
177 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6");
984263bc
MD
178#endif
179
5b0b9fa5
PA
180/*
181 * Minimum MSS we accept and use. This prevents DoS attacks where
182 * we are forced to a ridiculous low MSS like 20 and send hundreds
183 * of packets instead of one. The effect scales with the available
184 * bandwidth and quickly saturates the CPU and network interface
185 * with packet generation and sending. Set to zero to disable MINMSS
186 * checking. This setting prevents us from sending too small packets.
187 */
188int tcp_minmss = TCP_MINMSS;
189SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
190 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
5b0b9fa5 191
984263bc 192#if 0
707ad4ed 193static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
f23061d4 194SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
707ad4ed 195 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time");
984263bc
MD
196#endif
197
707ad4ed 198int tcp_do_rfc1323 = 1;
f23061d4 199SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
707ad4ed 200 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
984263bc 201
707ad4ed 202static int tcp_tcbhashsize = 0;
984263bc 203SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
707ad4ed 204 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable");
984263bc 205
707ad4ed 206static int do_tcpdrain = 1;
984263bc
MD
207SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208 "Enable tcp_drain routine for extra help when low on mbufs");
209
707ad4ed 210static int icmp_may_rst = 1;
f23061d4 211SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
984263bc
MD
212 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
213
707ad4ed 214static int tcp_isn_reseed_interval = 0;
984263bc
MD
215SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
216 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
217
218/*
d66b98eb
MD
219 * TCP bandwidth limiting sysctls. The inflight limiter is now turned on
220 * by default, but with generous values which should allow maximal
221 * bandwidth. In particular, the slop defaults to 50 (5 packets).
222 *
223 * The reason for doing this is that the limiter is the only mechanism we
224 * have which seems to do a really good job preventing receiver RX rings
225 * on network interfaces from getting blown out. Even though GigE/10GigE
226 * is supposed to flow control it looks like either it doesn't actually
227 * do it or Open Source drivers do not properly enable it.
228 *
229 * People using the limiter to reduce bottlenecks on slower WAN connections
230 * should set the slop to 20 (2 packets).
984263bc 231 */
d66b98eb 232static int tcp_inflight_enable = 1;
984263bc
MD
233SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW,
234 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
235
707ad4ed 236static int tcp_inflight_debug = 0;
984263bc
MD
237SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW,
238 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
239
707ad4ed 240static int tcp_inflight_min = 6144;
984263bc 241SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW,
707ad4ed 242 &tcp_inflight_min, 0, "Lower bound for TCP inflight window");
984263bc 243
707ad4ed 244static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
984263bc 245SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW,
707ad4ed 246 &tcp_inflight_max, 0, "Upper bound for TCP inflight window");
984263bc 247
d66b98eb 248static int tcp_inflight_stab = 50;
984263bc 249SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW,
d66b98eb 250 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 3 packets)");
984263bc 251
dd2b0fb4
MD
252static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives");
253static struct malloc_pipe tcptemp_mpipe;
254
92db3805 255static void tcp_willblock(int);
707ad4ed 256static void tcp_notify (struct inpcb *, int);
984263bc 257
5f7ab76b 258struct tcp_stats tcpstats_percpu[MAXCPU];
2b57d013
MD
259#ifdef SMP
260static int
261sysctl_tcpstats(SYSCTL_HANDLER_ARGS)
262{
707ad4ed 263 int cpu, error = 0;
2b57d013 264
707ad4ed 265 for (cpu = 0; cpu < ncpus; ++cpu) {
5f7ab76b 266 if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu],
707ad4ed 267 sizeof(struct tcp_stats))))
2b57d013 268 break;
5f7ab76b 269 if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu],
707ad4ed 270 sizeof(struct tcp_stats))))
2b57d013
MD
271 break;
272 }
273
274 return (error);
275}
707ad4ed
JH
276SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
277 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics");
278#else
2b57d013 279SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
707ad4ed 280 &tcpstat, tcp_stats, "TCP statistics");
2b57d013
MD
281#endif
282
984263bc
MD
283/*
284 * Target size of TCP PCB hash tables. Must be a power of two.
285 *
286 * Note that this can be overridden by the kernel environment
287 * variable net.inet.tcp.tcbhashsize
288 */
289#ifndef TCBHASHSIZE
707ad4ed 290#define TCBHASHSIZE 512
984263bc
MD
291#endif
292
293/*
294 * This is the actual shape of what we allocate using the zone
295 * allocator. Doing it this way allows us to protect both structures
296 * using the same generation count, and also eliminates the overhead
297 * of allocating tcpcbs separately. By hiding the structure here,
298 * we avoid changing most of the rest of the code (although it needs
299 * to be changed, eventually, for greater efficiency).
300 */
301#define ALIGNMENT 32
302#define ALIGNM1 (ALIGNMENT - 1)
303struct inp_tp {
304 union {
305 struct inpcb inp;
306 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
307 } inp_tp_u;
308 struct tcpcb tcb;
a48c5dd5
SZ
309 struct tcp_callout inp_tp_rexmt;
310 struct tcp_callout inp_tp_persist;
311 struct tcp_callout inp_tp_keep;
312 struct tcp_callout inp_tp_2msl;
313 struct tcp_callout inp_tp_delack;
0f758523 314 struct netmsg_tcp_timer inp_tp_timermsg;
984263bc
MD
315};
316#undef ALIGNMENT
317#undef ALIGNM1
318
319/*
320 * Tcp initialization
321 */
322void
f3f70f0d 323tcp_init(void)
984263bc 324{
d371a63a
JH
325 struct inpcbporthead *porthashbase;
326 u_long porthashmask;
984263bc 327 int hashsize = TCBHASHSIZE;
d371a63a 328 int cpu;
bf82f9b7 329
d2e9e54c
MD
330 /*
331 * note: tcptemp is used for keepalives, and it is ok for an
332 * allocation to fail so do not specify MPF_INT.
333 */
dd2b0fb4
MD
334 mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp),
335 25, -1, 0, NULL);
336
984263bc
MD
337 tcp_delacktime = TCPTV_DELACK;
338 tcp_keepinit = TCPTV_KEEP_INIT;
339 tcp_keepidle = TCPTV_KEEP_IDLE;
340 tcp_keepintvl = TCPTV_KEEPINTVL;
341 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
342 tcp_msl = TCPTV_MSL;
343 tcp_rexmit_min = TCPTV_MIN;
344 tcp_rexmit_slop = TCPTV_CPU_VAR;
345
984263bc
MD
346 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
347 if (!powerof2(hashsize)) {
a6ec04bc 348 kprintf("WARNING: TCB hash size not a power of 2\n");
984263bc
MD
349 hashsize = 512; /* safe default */
350 }
351 tcp_tcbhashsize = hashsize;
d371a63a 352 porthashbase = hashinit(hashsize, M_PCB, &porthashmask);
d371a63a
JH
353
354 for (cpu = 0; cpu < ncpus2; cpu++) {
d2e9e54c 355 in_pcbinfo_init(&tcbinfo[cpu]);
eb594563 356 tcbinfo[cpu].cpu = cpu;
d371a63a
JH
357 tcbinfo[cpu].hashbase = hashinit(hashsize, M_PCB,
358 &tcbinfo[cpu].hashmask);
359 tcbinfo[cpu].porthashbase = porthashbase;
360 tcbinfo[cpu].porthashmask = porthashmask;
8affadf8
JH
361 tcbinfo[cpu].wildcardhashbase = hashinit(hashsize, M_PCB,
362 &tcbinfo[cpu].wildcardhashmask);
9f42c129 363 tcbinfo[cpu].ipi_size = sizeof(struct inp_tp);
2b1ce38a 364 TAILQ_INIT(&tcpcbackq[cpu]);
d371a63a 365 }
3edf7c37
RG
366
367 tcp_reass_maxseg = nmbclusters / 16;
707ad4ed 368 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg);
3edf7c37 369
984263bc 370#ifdef INET6
707ad4ed
JH
371#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
372#else
373#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
374#endif
984263bc
MD
375 if (max_protohdr < TCP_MINPROTOHDR)
376 max_protohdr = TCP_MINPROTOHDR;
377 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
378 panic("tcp_init");
379#undef TCP_MINPROTOHDR
380
2b57d013 381 /*
5f7ab76b 382 * Initialize TCP statistics counters for each CPU.
2b57d013
MD
383 */
384#ifdef SMP
385 for (cpu = 0; cpu < ncpus; ++cpu) {
5f7ab76b 386 bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats));
2b57d013
MD
387 }
388#else
389 bzero(&tcpstat, sizeof(struct tcp_stats));
390#endif
391
984263bc 392 syncache_init();
bf82f9b7 393 tcp_thread_init();
984263bc
MD
394}
395
2b1ce38a
MD
396void
397tcpmsg_service_loop(void *dummy)
f23061d4 398{
2b1ce38a 399 struct netmsg *msg;
92db3805
SZ
400 int mplocked;
401
402 /*
403 * Thread was started with TDF_MPSAFE
404 */
405 mplocked = 0;
2b1ce38a 406
1e3f8217 407 while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2b1ce38a 408 do {
c7afbe76 409 logtcp(rxmsg);
92db3805
SZ
410 mplocked = netmsg_service(msg, tcp_mpsafe_thread,
411 mplocked);
2b1ce38a 412 } while ((msg = lwkt_getport(&curthread->td_msgport)) != NULL);
92db3805 413
c7afbe76 414 logtcp(delayed);
92db3805 415 tcp_willblock(mplocked);
c7afbe76 416 logtcp(wait);
2b1ce38a
MD
417 }
418}
419
420static void
92db3805 421tcp_willblock(int mplocked)
2b1ce38a
MD
422{
423 struct tcpcb *tp;
424 int cpu = mycpu->gd_cpuid;
92db3805
SZ
425 int unlock = 0;
426
427 if (!mplocked && !tcp_mpsafe_proto) {
428 if (TAILQ_EMPTY(&tcpcbackq[cpu]))
429 return;
430
431 get_mplock();
432 mplocked = 1;
433 unlock = 1;
434 }
2b1ce38a
MD
435
436 while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) {
437 KKASSERT(tp->t_flags & TF_ONOUTPUTQ);
438 tp->t_flags &= ~TF_ONOUTPUTQ;
439 TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq);
440 tcp_output(tp);
441 }
92db3805
SZ
442
443 if (unlock)
444 rel_mplock();
2b1ce38a
MD
445}
446
447
984263bc
MD
448/*
449 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
450 * tcp_template used to store this data in mbufs, but we now recopy it out
451 * of the tcpcb each time to conserve mbufs.
452 */
453void
707ad4ed 454tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr)
984263bc
MD
455{
456 struct inpcb *inp = tp->t_inpcb;
457 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
458
459#ifdef INET6
707ad4ed 460 if (inp->inp_vflag & INP_IPV6) {
984263bc
MD
461 struct ip6_hdr *ip6;
462
463 ip6 = (struct ip6_hdr *)ip_ptr;
464 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
465 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
466 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
467 (IPV6_VERSION & IPV6_VERSION_MASK);
468 ip6->ip6_nxt = IPPROTO_TCP;
469 ip6->ip6_plen = sizeof(struct tcphdr);
470 ip6->ip6_src = inp->in6p_laddr;
471 ip6->ip6_dst = inp->in6p_faddr;
472 tcp_hdr->th_sum = 0;
473 } else
474#endif
475 {
707ad4ed
JH
476 struct ip *ip = (struct ip *) ip_ptr;
477
478 ip->ip_vhl = IP_VHL_BORING;
479 ip->ip_tos = 0;
480 ip->ip_len = 0;
481 ip->ip_id = 0;
482 ip->ip_off = 0;
483 ip->ip_ttl = 0;
484 ip->ip_sum = 0;
485 ip->ip_p = IPPROTO_TCP;
486 ip->ip_src = inp->inp_laddr;
487 ip->ip_dst = inp->inp_faddr;
488 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr,
489 ip->ip_dst.s_addr,
490 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
984263bc
MD
491 }
492
493 tcp_hdr->th_sport = inp->inp_lport;
494 tcp_hdr->th_dport = inp->inp_fport;
495 tcp_hdr->th_seq = 0;
496 tcp_hdr->th_ack = 0;
497 tcp_hdr->th_x2 = 0;
498 tcp_hdr->th_off = 5;
499 tcp_hdr->th_flags = 0;
500 tcp_hdr->th_win = 0;
501 tcp_hdr->th_urp = 0;
502}
503
504/*
505 * Create template to be used to send tcp packets on a connection.
506 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
507 * use for this function is in keepalives, which use tcp_respond.
508 */
509struct tcptemp *
707ad4ed 510tcp_maketemplate(struct tcpcb *tp)
984263bc 511{
dd2b0fb4 512 struct tcptemp *tmp;
984263bc 513
dd2b0fb4 514 if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL)
707ad4ed 515 return (NULL);
f23061d4 516 tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t);
dd2b0fb4
MD
517 return (tmp);
518}
984263bc 519
dd2b0fb4
MD
520void
521tcp_freetemplate(struct tcptemp *tmp)
522{
523 mpipe_free(&tcptemp_mpipe, tmp);
984263bc
MD
524}
525
526/*
527 * Send a single message to the TCP at address specified by
707ad4ed 528 * the given TCP/IP header. If m == NULL, then we make a copy
984263bc
MD
529 * of the tcpiphdr at ti and send directly to the addressed host.
530 * This is used to force keep alive messages out using the TCP
531 * template for a connection. If flags are given then we send
532 * a message back to the TCP which originated the * segment ti,
533 * and discard the mbuf containing it and any other attached mbufs.
534 *
535 * In any case the ack and sequence number of the transmitted
536 * segment are as specified by the parameters.
537 *
538 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
539 */
540void
707ad4ed
JH
541tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
542 tcp_seq ack, tcp_seq seq, int flags)
984263bc 543{
2256ba69 544 int tlen;
984263bc 545 int win = 0;
707ad4ed 546 struct route *ro = NULL;
984263bc 547 struct route sro;
707ad4ed 548 struct ip *ip = ipgen;
984263bc 549 struct tcphdr *nth;
984263bc 550 int ipflags = 0;
707ad4ed
JH
551 struct route_in6 *ro6 = NULL;
552 struct route_in6 sro6;
553 struct ip6_hdr *ip6 = ipgen;
5a0e5b43 554 boolean_t use_tmpro = TRUE;
984263bc 555#ifdef INET6
707ad4ed
JH
556 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6);
557#else
558 const boolean_t isipv6 = FALSE;
559#endif
984263bc 560
707ad4ed 561 if (tp != NULL) {
984263bc 562 if (!(flags & TH_RST)) {
6d49aa6f 563 win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv);
46e92930
MD
564 if (win < 0)
565 win = 0;
984263bc
MD
566 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
567 win = (long)TCP_MAXWIN << tp->rcv_scale;
568 }
5a0e5b43
SZ
569 /*
570 * Don't use the route cache of a listen socket,
571 * it is not MPSAFE; use temporary route cache.
572 */
573 if (tp->t_state != TCPS_LISTEN) {
574 if (isipv6)
575 ro6 = &tp->t_inpcb->in6p_route;
576 else
577 ro = &tp->t_inpcb->inp_route;
578 use_tmpro = FALSE;
579 }
580 }
581 if (use_tmpro) {
984263bc
MD
582 if (isipv6) {
583 ro6 = &sro6;
584 bzero(ro6, sizeof *ro6);
707ad4ed
JH
585 } else {
586 ro = &sro;
587 bzero(ro, sizeof *ro);
588 }
984263bc 589 }
707ad4ed 590 if (m == NULL) {
74f1caca 591 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
984263bc
MD
592 if (m == NULL)
593 return;
594 tlen = 0;
595 m->m_data += max_linkhdr;
984263bc 596 if (isipv6) {
707ad4ed 597 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr));
984263bc
MD
598 ip6 = mtod(m, struct ip6_hdr *);
599 nth = (struct tcphdr *)(ip6 + 1);
707ad4ed
JH
600 } else {
601 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
602 ip = mtod(m, struct ip *);
603 nth = (struct tcphdr *)(ip + 1);
604 }
605 bcopy(th, nth, sizeof(struct tcphdr));
984263bc
MD
606 flags = TH_ACK;
607 } else {
608 m_freem(m->m_next);
707ad4ed 609 m->m_next = NULL;
984263bc
MD
610 m->m_data = (caddr_t)ipgen;
611 /* m_len is set later */
612 tlen = 0;
707ad4ed 613#define xchg(a, b, type) { type t; t = a; a = b; b = t; }
984263bc
MD
614 if (isipv6) {
615 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
616 nth = (struct tcphdr *)(ip6 + 1);
707ad4ed
JH
617 } else {
618 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
619 nth = (struct tcphdr *)(ip + 1);
620 }
984263bc
MD
621 if (th != nth) {
622 /*
623 * this is usually a case when an extension header
624 * exists between the IPv6 header and the
625 * TCP header.
626 */
627 nth->th_sport = th->th_sport;
628 nth->th_dport = th->th_dport;
629 }
630 xchg(nth->th_dport, nth->th_sport, n_short);
631#undef xchg
632 }
984263bc
MD
633 if (isipv6) {
634 ip6->ip6_flow = 0;
635 ip6->ip6_vfc = IPV6_VERSION;
636 ip6->ip6_nxt = IPPROTO_TCP;
707ad4ed
JH
637 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen));
638 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
639 } else {
640 tlen += sizeof(struct tcpiphdr);
641 ip->ip_len = tlen;
642 ip->ip_ttl = ip_defttl;
643 }
984263bc
MD
644 m->m_len = tlen;
645 m->m_pkthdr.len = tlen;
2038fb68 646 m->m_pkthdr.rcvif = NULL;
984263bc
MD
647 nth->th_seq = htonl(seq);
648 nth->th_ack = htonl(ack);
649 nth->th_x2 = 0;
707ad4ed 650 nth->th_off = sizeof(struct tcphdr) >> 2;
984263bc 651 nth->th_flags = flags;
707ad4ed 652 if (tp != NULL)
984263bc
MD
653 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
654 else
655 nth->th_win = htons((u_short)win);
656 nth->th_urp = 0;
984263bc
MD
657 if (isipv6) {
658 nth->th_sum = 0;
659 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
660 sizeof(struct ip6_hdr),
661 tlen - sizeof(struct ip6_hdr));
662 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
707ad4ed 663 (ro6 && ro6->ro_rt) ?
f23061d4 664 ro6->ro_rt->rt_ifp : NULL);
707ad4ed
JH
665 } else {
666 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
667 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
668 m->m_pkthdr.csum_flags = CSUM_TCP;
669 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
670 }
984263bc
MD
671#ifdef TCPDEBUG
672 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
673 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
674#endif
984263bc 675 if (isipv6) {
f23061d4
JH
676 ip6_output(m, NULL, ro6, ipflags, NULL, NULL,
677 tp ? tp->t_inpcb : NULL);
707ad4ed 678 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) {
984263bc
MD
679 RTFREE(ro6->ro_rt);
680 ro6->ro_rt = NULL;
681 }
707ad4ed 682 } else {
1dbb3516 683 ipflags |= IP_DEBUGROUTE;
f23061d4 684 ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL);
707ad4ed
JH
685 if ((ro == &sro) && (ro->ro_rt != NULL)) {
686 RTFREE(ro->ro_rt);
687 ro->ro_rt = NULL;
688 }
984263bc 689 }
984263bc
MD
690}
691
692/*
693 * Create a new TCP control block, making an
694 * empty reassembly queue and hooking it to the argument
695 * protocol control block. The `inp' parameter must have
696 * come from the zone allocator set up in tcp_init().
697 */
698struct tcpcb *
707ad4ed 699tcp_newtcpcb(struct inpcb *inp)
984263bc
MD
700{
701 struct inp_tp *it;
2256ba69 702 struct tcpcb *tp;
984263bc 703#ifdef INET6
707ad4ed
JH
704 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
705#else
706 const boolean_t isipv6 = FALSE;
707#endif
984263bc
MD
708
709 it = (struct inp_tp *)inp;
710 tp = &it->tcb;
707ad4ed 711 bzero(tp, sizeof(struct tcpcb));
984263bc 712 LIST_INIT(&tp->t_segq);
707ad4ed 713 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
984263bc
MD
714
715 /* Set up our timeouts. */
a48c5dd5
SZ
716 tp->tt_rexmt = &it->inp_tp_rexmt;
717 tp->tt_persist = &it->inp_tp_persist;
718 tp->tt_keep = &it->inp_tp_keep;
719 tp->tt_2msl = &it->inp_tp_2msl;
720 tp->tt_delack = &it->inp_tp_delack;
721 tcp_inittimers(tp);
984263bc 722
3db1c8a3
SZ
723 /*
724 * Zero out timer message. We don't create it here,
725 * since the current CPU may not be the owner of this
726 * inpcb.
727 */
0f758523 728 tp->tt_msg = &it->inp_tp_timermsg;
3db1c8a3 729 bzero(tp->tt_msg, sizeof(*tp->tt_msg));
0f758523 730
984263bc 731 if (tcp_do_rfc1323)
707ad4ed 732 tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP);
984263bc 733 tp->t_inpcb = inp; /* XXX */
eb594563 734 tp->t_state = TCPS_CLOSED;
984263bc
MD
735 /*
736 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
737 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
738 * reasonable initial retransmit time.
739 */
740 tp->t_srtt = TCPTV_SRTTBASE;
707ad4ed
JH
741 tp->t_rttvar =
742 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
984263bc
MD
743 tp->t_rttmin = tcp_rexmit_min;
744 tp->t_rxtcur = TCPTV_RTOBASE;
745 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
746 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
747 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
748 tp->t_rcvtime = ticks;
707ad4ed 749 /*
984263bc
MD
750 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
751 * because the socket may be bound to an IPv6 wildcard address,
752 * which may match an IPv4-mapped IPv6 address.
753 */
754 inp->inp_ip_ttl = ip_defttl;
f23061d4 755 inp->inp_ppcb = tp;
91489f6b 756 tcp_sack_tcpcb_init(tp);
984263bc
MD
757 return (tp); /* XXX */
758}
759
760/*
707ad4ed
JH
761 * Drop a TCP connection, reporting the specified error.
762 * If connection is synchronized, then send a RST to peer.
984263bc
MD
763 */
764struct tcpcb *
71f385dc 765tcp_drop(struct tcpcb *tp, int error)
984263bc
MD
766{
767 struct socket *so = tp->t_inpcb->inp_socket;
768
769 if (TCPS_HAVERCVDSYN(tp->t_state)) {
770 tp->t_state = TCPS_CLOSED;
f23061d4 771 tcp_output(tp);
984263bc
MD
772 tcpstat.tcps_drops++;
773 } else
774 tcpstat.tcps_conndrops++;
71f385dc
MD
775 if (error == ETIMEDOUT && tp->t_softerror)
776 error = tp->t_softerror;
777 so->so_error = error;
984263bc
MD
778 return (tcp_close(tp));
779}
780
8affadf8 781#ifdef SMP
eb594563 782
8affadf8 783struct netmsg_remwildcard {
4599cf19 784 struct netmsg nm_netmsg;
8affadf8
JH
785 struct inpcb *nm_inp;
786 struct inpcbinfo *nm_pcbinfo;
eb594563
MD
787#if defined(INET6)
788 int nm_isinet6;
789#else
790 int nm_unused01;
791#endif
8affadf8
JH
792};
793
eb594563
MD
794/*
795 * Wildcard inpcb's on SMP boxes must be removed from all cpus before the
796 * inp can be detached. We do this by cycling through the cpus, ending up
797 * on the cpu controlling the inp last and then doing the disconnect.
798 */
4599cf19
MD
799static void
800in_pcbremwildcardhash_handler(struct netmsg *msg0)
8affadf8
JH
801{
802 struct netmsg_remwildcard *msg = (struct netmsg_remwildcard *)msg0;
eb594563 803 int cpu;
8affadf8 804
eb594563
MD
805 cpu = msg->nm_pcbinfo->cpu;
806
807 if (cpu == msg->nm_inp->inp_pcbinfo->cpu) {
808 /* note: detach removes any wildcard hash entry */
809#ifdef INET6
810 if (msg->nm_isinet6)
811 in6_pcbdetach(msg->nm_inp);
812 else
813#endif
814 in_pcbdetach(msg->nm_inp);
4599cf19 815 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
eb594563
MD
816 } else {
817 in_pcbremwildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
818 cpu = (cpu + 1) % ncpus2;
819 msg->nm_pcbinfo = &tcbinfo[cpu];
4599cf19 820 lwkt_forwardmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
eb594563 821 }
8affadf8 822}
eb594563 823
8affadf8
JH
824#endif
825
984263bc
MD
826/*
827 * Close a TCP control block:
828 * discard all space held by the tcp
829 * discard internet protocol block
830 * wake up any sleepers
831 */
832struct tcpcb *
707ad4ed 833tcp_close(struct tcpcb *tp)
984263bc 834{
2256ba69 835 struct tseg_qent *q;
984263bc
MD
836 struct inpcb *inp = tp->t_inpcb;
837 struct socket *so = inp->inp_socket;
2256ba69 838 struct rtentry *rt;
707ad4ed 839 boolean_t dosavessthresh;
8affadf8
JH
840#ifdef SMP
841 int cpu;
842#endif
707ad4ed
JH
843#ifdef INET6
844 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
eb594563 845 boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0);
707ad4ed
JH
846#else
847 const boolean_t isipv6 = FALSE;
848#endif
984263bc 849
eb594563
MD
850 /*
851 * The tp is not instantly destroyed in the wildcard case. Setting
852 * the state to TCPS_TERMINATING will prevent the TCP stack from
853 * messing with it, though it should be noted that this change may
854 * not take effect on other cpus until we have chained the wildcard
855 * hash removal.
856 *
857 * XXX we currently depend on the BGL to synchronize the tp->t_state
858 * update and prevent other tcp protocol threads from accepting new
859 * connections on the listen socket we might be trying to close down.
860 */
861 KKASSERT(tp->t_state != TCPS_TERMINATING);
862 tp->t_state = TCPS_TERMINATING;
863
984263bc
MD
864 /*
865 * Make sure that all of our timers are stopped before we
2d42d2b0 866 * delete the PCB. For listen TCP socket (tp->tt_msg == NULL),
697aadcd
SZ
867 * timers are never used. If timer message is never created
868 * (tp->tt_msg->tt_tcb == NULL), timers are never used too.
984263bc 869 */
697aadcd 870 if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) {
2d42d2b0
SZ
871 tcp_callout_stop(tp, tp->tt_rexmt);
872 tcp_callout_stop(tp, tp->tt_persist);
873 tcp_callout_stop(tp, tp->tt_keep);
874 tcp_callout_stop(tp, tp->tt_2msl);
875 tcp_callout_stop(tp, tp->tt_delack);
876 }
984263bc 877
2b1ce38a
MD
878 if (tp->t_flags & TF_ONOUTPUTQ) {
879 KKASSERT(tp->tt_cpu == mycpu->gd_cpuid);
880 TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq);
881 tp->t_flags &= ~TF_ONOUTPUTQ;
882 }
883
984263bc
MD
884 /*
885 * If we got enough samples through the srtt filter,
886 * save the rtt and rttvar in the routing entry.
887 * 'Enough' is arbitrarily defined as the 16 samples.
888 * 16 samples is enough for the srtt filter to converge
889 * to within 5% of the correct value; fewer samples and
890 * we could save a very bogus rtt.
891 *
892 * Don't update the default route's characteristics and don't
893 * update anything that the user "locked".
894 */
895 if (tp->t_rttupdated >= 16) {
2256ba69 896 u_long i = 0;
707ad4ed 897
984263bc
MD
898 if (isipv6) {
899 struct sockaddr_in6 *sin6;
900
901 if ((rt = inp->in6p_route.ro_rt) == NULL)
902 goto no_valid_rt;
903 sin6 = (struct sockaddr_in6 *)rt_key(rt);
904 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
905 goto no_valid_rt;
707ad4ed
JH
906 } else
907 if ((rt = inp->inp_route.ro_rt) == NULL ||
908 ((struct sockaddr_in *)rt_key(rt))->
909 sin_addr.s_addr == INADDR_ANY)
910 goto no_valid_rt;
911
912 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) {
913 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
984263bc
MD
914 if (rt->rt_rmx.rmx_rtt && i)
915 /*
916 * filter this update to half the old & half
917 * the new values, converting scale.
918 * See route.h and tcp_var.h for a
919 * description of the scaling constants.
920 */
921 rt->rt_rmx.rmx_rtt =
922 (rt->rt_rmx.rmx_rtt + i) / 2;
923 else
924 rt->rt_rmx.rmx_rtt = i;
925 tcpstat.tcps_cachedrtt++;
926 }
707ad4ed 927 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) {
984263bc
MD
928 i = tp->t_rttvar *
929 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
930 if (rt->rt_rmx.rmx_rttvar && i)
931 rt->rt_rmx.rmx_rttvar =
932 (rt->rt_rmx.rmx_rttvar + i) / 2;
933 else
934 rt->rt_rmx.rmx_rttvar = i;
935 tcpstat.tcps_cachedrttvar++;
936 }
937 /*
938 * The old comment here said:
939 * update the pipelimit (ssthresh) if it has been updated
940 * already or if a pipesize was specified & the threshhold
941 * got below half the pipesize. I.e., wait for bad news
942 * before we start updating, then update on both good
943 * and bad news.
944 *
945 * But we want to save the ssthresh even if no pipesize is
946 * specified explicitly in the route, because such
947 * connections still have an implicit pipesize specified
948 * by the global tcp_sendspace. In the absence of a reliable
949 * way to calculate the pipesize, it will have to do.
950 */
951 i = tp->snd_ssthresh;
952 if (rt->rt_rmx.rmx_sendpipe != 0)
707ad4ed 953 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2);
984263bc 954 else
6d49aa6f 955 dosavessthresh = (i < so->so_snd.ssb_hiwat/2);
707ad4ed
JH
956 if (dosavessthresh ||
957 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) &&
958 (rt->rt_rmx.rmx_ssthresh != 0))) {
984263bc
MD
959 /*
960 * convert the limit from user data bytes to
961 * packets then to packet data bytes.
962 */
963 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
964 if (i < 2)
965 i = 2;
707ad4ed
JH
966 i *= tp->t_maxseg +
967 (isipv6 ?
968 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
969 sizeof(struct tcpiphdr));
984263bc
MD
970 if (rt->rt_rmx.rmx_ssthresh)
971 rt->rt_rmx.rmx_ssthresh =
972 (rt->rt_rmx.rmx_ssthresh + i) / 2;
973 else
974 rt->rt_rmx.rmx_ssthresh = i;
975 tcpstat.tcps_cachedssthresh++;
976 }
977 }
707ad4ed
JH
978
979no_valid_rt:
984263bc
MD
980 /* free the reassembly queue, if any */
981 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
982 LIST_REMOVE(q, tqe_q);
983 m_freem(q->tqe_m);
984 FREE(q, M_TSEGQ);
3edf7c37 985 tcp_reass_qsize--;
984263bc 986 }
91489f6b
JH
987 /* throw away SACK blocks in scoreboard*/
988 if (TCP_DO_SACK(tp))
989 tcp_sack_cleanup(&tp->scb);
eb594563 990
984263bc
MD
991 inp->inp_ppcb = NULL;
992 soisdisconnected(so);
0f758523
SZ
993
994 tcp_destroy_timermsg(tp);
995
eb594563
MD
996 /*
997 * Discard the inp. In the SMP case a wildcard inp's hash (created
998 * by a listen socket or an INADDR_ANY udp socket) is replicated
999 * for each protocol thread and must be removed in the context of
1000 * that thread. This is accomplished by chaining the message
1001 * through the cpus.
1002 *
1003 * If the inp is not wildcarded we simply detach, which will remove
1004 * the any hashes still present for this inp.
1005 */
8affadf8 1006#ifdef SMP
1e019813 1007 if (inp->inp_flags & INP_WILDCARD_MP) {
eb594563 1008 struct netmsg_remwildcard *msg;
1e019813 1009
eb594563 1010 cpu = (inp->inp_pcbinfo->cpu + 1) % ncpus2;
77652cad 1011 msg = kmalloc(sizeof(struct netmsg_remwildcard),
4599cf19 1012 M_LWKTMSG, M_INTWAIT);
48e7b118
MD
1013 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
1014 0, in_pcbremwildcardhash_handler);
eb594563
MD
1015#ifdef INET6
1016 msg->nm_isinet6 = isafinet6;
8affadf8 1017#endif
eb594563
MD
1018 msg->nm_inp = inp;
1019 msg->nm_pcbinfo = &tcbinfo[cpu];
4599cf19 1020 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
f23061d4 1021 } else
eb594563
MD
1022#endif
1023 {
1024 /* note: detach removes any wildcard hash entry */
984263bc 1025#ifdef INET6
eb594563
MD
1026 if (isafinet6)
1027 in6_pcbdetach(inp);
1028 else
707ad4ed 1029#endif
eb594563
MD
1030 in_pcbdetach(inp);
1031 }
984263bc 1032 tcpstat.tcps_closed++;
707ad4ed 1033 return (NULL);
984263bc
MD
1034}
1035
3f48f9c5
JH
1036static __inline void
1037tcp_drain_oncpu(struct inpcbhead *head)
984263bc 1038{
d371a63a
JH
1039 struct inpcb *inpb;
1040 struct tcpcb *tcpb;
1041 struct tseg_qent *te;
3f48f9c5
JH
1042
1043 LIST_FOREACH(inpb, head, inp_list) {
d2e9e54c
MD
1044 if (inpb->inp_flags & INP_PLACEMARKER)
1045 continue;
3f48f9c5
JH
1046 if ((tcpb = intotcpcb(inpb))) {
1047 while ((te = LIST_FIRST(&tcpb->t_segq)) != NULL) {
1048 LIST_REMOVE(te, tqe_q);
1049 m_freem(te->tqe_m);
1050 FREE(te, M_TSEGQ);
1051 tcp_reass_qsize--;
1052 }
1053 }
1054 }
1055}
1056
0ddb6032
JH
1057#ifdef SMP
1058struct netmsg_tcp_drain {
4599cf19 1059 struct netmsg nm_netmsg;
0ddb6032
JH
1060 struct inpcbhead *nm_head;
1061};
1062
4599cf19
MD
1063static void
1064tcp_drain_handler(netmsg_t netmsg)
0ddb6032 1065{
4599cf19 1066 struct netmsg_tcp_drain *nm = (void *)netmsg;
0ddb6032
JH
1067
1068 tcp_drain_oncpu(nm->nm_head);
4599cf19 1069 lwkt_replymsg(&nm->nm_netmsg.nm_lmsg, 0);
0ddb6032
JH
1070}
1071#endif
1072
3f48f9c5 1073void
f3f70f0d 1074tcp_drain(void)
3f48f9c5 1075{
93c8e032 1076#ifdef SMP
d371a63a 1077 int cpu;
93c8e032 1078#endif
d371a63a
JH
1079
1080 if (!do_tcpdrain)
1081 return;
984263bc
MD
1082
1083 /*
1084 * Walk the tcpbs, if existing, and flush the reassembly queue,
1085 * if there is one...
1086 * XXX: The "Net/3" implementation doesn't imply that the TCP
707ad4ed
JH
1087 * reassembly queue should be flushed, but in a situation
1088 * where we're really low on mbufs, this is potentially
1089 * useful.
984263bc 1090 */
3f48f9c5 1091#ifdef SMP
d371a63a 1092 for (cpu = 0; cpu < ncpus2; cpu++) {
3f48f9c5
JH
1093 struct netmsg_tcp_drain *msg;
1094
0ddb6032 1095 if (cpu == mycpu->gd_cpuid) {
d2e9e54c 1096 tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead);
3f48f9c5 1097 } else {
77652cad 1098 msg = kmalloc(sizeof(struct netmsg_tcp_drain),
b76bed62
MD
1099 M_LWKTMSG, M_NOWAIT);
1100 if (msg == NULL)
3f48f9c5 1101 continue;
48e7b118
MD
1102 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
1103 0, tcp_drain_handler);
d2e9e54c 1104 msg->nm_head = &tcbinfo[cpu].pcblisthead;
4599cf19 1105 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
984263bc 1106 }
984263bc 1107 }
3f48f9c5 1108#else
d2e9e54c 1109 tcp_drain_oncpu(&tcbinfo[0].pcblisthead);
3f48f9c5 1110#endif
984263bc
MD
1111}
1112
1113/*
1114 * Notify a tcp user of an asynchronous error;
1115 * store error as soft error, but wake up user
1116 * (for now, won't do anything until can select for soft error).
1117 *
1118 * Do not wake up user since there currently is no mechanism for
1119 * reporting soft errors (yet - a kqueue filter may be added).
1120 */
1121static void
707ad4ed 1122tcp_notify(struct inpcb *inp, int error)
984263bc 1123{
707ad4ed 1124 struct tcpcb *tp = intotcpcb(inp);
984263bc
MD
1125
1126 /*
1127 * Ignore some errors if we are hooked up.
1128 * If connection hasn't completed, has retransmitted several times,
1129 * and receives a second error, give up now. This is better
1130 * than waiting a long time to establish a connection that
1131 * can never complete.
1132 */
1133 if (tp->t_state == TCPS_ESTABLISHED &&
1134 (error == EHOSTUNREACH || error == ENETUNREACH ||
1135 error == EHOSTDOWN)) {
1136 return;
1137 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1138 tp->t_softerror)
1139 tcp_drop(tp, error);
1140 else
1141 tp->t_softerror = error;
1142#if 0
f23061d4 1143 wakeup(&so->so_timeo);
984263bc
MD
1144 sorwakeup(so);
1145 sowwakeup(so);
1146#endif
1147}
1148
1149static int
1150tcp_pcblist(SYSCTL_HANDLER_ARGS)
1151{
d2e9e54c
MD
1152 int error, i, n;
1153 struct inpcb *marker;
1154 struct inpcb *inp;
984263bc 1155 inp_gen_t gencnt;
d2e9e54c
MD
1156 globaldata_t gd;
1157 int origcpu, ccpu;
1158
1159 error = 0;
1160 n = 0;
984263bc
MD
1161
1162 /*
1163 * The process of preparing the TCB list is too time-consuming and
1164 * resource-intensive to repeat twice on every request.
1165 */
707ad4ed 1166 if (req->oldptr == NULL) {
d2e9e54c
MD
1167 for (ccpu = 0; ccpu < ncpus; ++ccpu) {
1168 gd = globaldata_find(ccpu);
1169 n += tcbinfo[gd->gd_cpuid].ipi_count;
1170 }
8d7c364e 1171 req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb);
707ad4ed 1172 return (0);
984263bc
MD
1173 }
1174
707ad4ed
JH
1175 if (req->newptr != NULL)
1176 return (EPERM);
984263bc 1177
efda3bd0 1178 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
d2e9e54c
MD
1179 marker->inp_flags |= INP_PLACEMARKER;
1180
984263bc 1181 /*
d2e9e54c 1182 * OK, now we're committed to doing something. Run the inpcb list
f23061d4 1183 * for each cpu in the system and construct the output. Use a
d2e9e54c
MD
1184 * list placemarker to deal with list changes occuring during
1185 * copyout blockages (but otherwise depend on being on the correct
1186 * cpu to avoid races).
984263bc 1187 */
d2e9e54c
MD
1188 origcpu = mycpu->gd_cpuid;
1189 for (ccpu = 1; ccpu <= ncpus && error == 0; ++ccpu) {
1190 globaldata_t rgd;
1191 caddr_t inp_ppcb;
1192 struct xtcpcb xt;
1193 int cpu_id;
1194
1195 cpu_id = (origcpu + ccpu) % ncpus;
1196 if ((smp_active_mask & (1 << cpu_id)) == 0)
1197 continue;
1198 rgd = globaldata_find(cpu_id);
1199 lwkt_setcpu_self(rgd);
1200
d2e9e54c
MD
1201 gencnt = tcbinfo[cpu_id].ipi_gencnt;
1202 n = tcbinfo[cpu_id].ipi_count;
1203
d2e9e54c
MD
1204 LIST_INSERT_HEAD(&tcbinfo[cpu_id].pcblisthead, marker, inp_list);
1205 i = 0;
1206 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
1207 /*
1208 * process a snapshot of pcbs, ignoring placemarkers
1209 * and using our own to allow SYSCTL_OUT to block.
1210 */
1211 LIST_REMOVE(marker, inp_list);
1212 LIST_INSERT_AFTER(inp, marker, inp_list);
707ad4ed 1213
d2e9e54c
MD
1214 if (inp->inp_flags & INP_PLACEMARKER)
1215 continue;
1216 if (inp->inp_gencnt > gencnt)
1217 continue;
1218 if (prison_xinpcb(req->td, inp))
1219 continue;
984263bc 1220
984263bc 1221 xt.xt_len = sizeof xt;
984263bc
MD
1222 bcopy(inp, &xt.xt_inp, sizeof *inp);
1223 inp_ppcb = inp->inp_ppcb;
1224 if (inp_ppcb != NULL)
1225 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1226 else
707ad4ed 1227 bzero(&xt.xt_tp, sizeof xt.xt_tp);
984263bc
MD
1228 if (inp->inp_socket)
1229 sotoxsocket(inp->inp_socket, &xt.xt_socket);
d2e9e54c
MD
1230 if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0)
1231 break;
1232 ++i;
1233 }
1234 LIST_REMOVE(marker, inp_list);
1235 if (error == 0 && i < n) {
0c3c561c
JH
1236 bzero(&xt, sizeof xt);
1237 xt.xt_len = sizeof xt;
d2e9e54c 1238 while (i < n) {
f23061d4 1239 error = SYSCTL_OUT(req, &xt, sizeof xt);
d2e9e54c
MD
1240 if (error)
1241 break;
1242 ++i;
1243 }
1244 }
984263bc 1245 }
d2e9e54c
MD
1246
1247 /*
1248 * Make sure we are on the same cpu we were on originally, since
1249 * higher level callers expect this. Also don't pollute caches with
1250 * migrated userland data by (eventually) returning to userland
1251 * on a different cpu.
1252 */
1253 lwkt_setcpu_self(globaldata_find(origcpu));
efda3bd0 1254 kfree(marker, M_TEMP);
707ad4ed 1255 return (error);
984263bc
MD
1256}
1257
1258SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1259 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1260
1261static int
1262tcp_getcred(SYSCTL_HANDLER_ARGS)
1263{
1264 struct sockaddr_in addrs[2];
1265 struct inpcb *inp;
d371a63a 1266 int cpu;
1cae611f 1267 int error;
984263bc 1268
895c1f85 1269 error = priv_check(req->td, PRIV_ROOT);
707ad4ed 1270 if (error != 0)
984263bc 1271 return (error);
707ad4ed
JH
1272 error = SYSCTL_IN(req, addrs, sizeof addrs);
1273 if (error != 0)
984263bc 1274 return (error);
1cae611f 1275 crit_enter();
d371a63a
JH
1276 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
1277 addrs[0].sin_addr.s_addr, addrs[0].sin_port);
1278 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr,
1279 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
984263bc
MD
1280 if (inp == NULL || inp->inp_socket == NULL) {
1281 error = ENOENT;
1282 goto out;
1283 }
1284 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
1285out:
1cae611f 1286 crit_exit();
984263bc
MD
1287 return (error);
1288}
1289
707ad4ed 1290SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
984263bc
MD
1291 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
1292
1293#ifdef INET6
1294static int
1295tcp6_getcred(SYSCTL_HANDLER_ARGS)
1296{
1297 struct sockaddr_in6 addrs[2];
1298 struct inpcb *inp;
1cae611f 1299 int error;
707ad4ed 1300 boolean_t mapped = FALSE;
984263bc 1301
895c1f85 1302 error = priv_check(req->td, PRIV_ROOT);
707ad4ed 1303 if (error != 0)
984263bc 1304 return (error);
707ad4ed
JH
1305 error = SYSCTL_IN(req, addrs, sizeof addrs);
1306 if (error != 0)
984263bc
MD
1307 return (error);
1308 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1309 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
707ad4ed 1310 mapped = TRUE;
984263bc
MD
1311 else
1312 return (EINVAL);
1313 }
1cae611f 1314 crit_enter();
707ad4ed 1315 if (mapped) {
d371a63a
JH
1316 inp = in_pcblookup_hash(&tcbinfo[0],
1317 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1318 addrs[1].sin6_port,
1319 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1320 addrs[0].sin6_port,
1321 0, NULL);
1322 } else {
1323 inp = in6_pcblookup_hash(&tcbinfo[0],
1324 &addrs[1].sin6_addr, addrs[1].sin6_port,
1325 &addrs[0].sin6_addr, addrs[0].sin6_port,
1326 0, NULL);
1327 }
984263bc
MD
1328 if (inp == NULL || inp->inp_socket == NULL) {
1329 error = ENOENT;
1330 goto out;
1331 }
707ad4ed 1332 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
984263bc 1333out:
1cae611f 1334 crit_exit();
984263bc
MD
1335 return (error);
1336}
1337
707ad4ed 1338SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
984263bc
MD
1339 0, 0,
1340 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
1341#endif
1342
14572273
SZ
1343struct netmsg_tcp_notify {
1344 struct netmsg nm_nmsg;
1345 void (*nm_notify)(struct inpcb *, int);
1346 struct in_addr nm_faddr;
1347 int nm_arg;
1348};
1349
1350static void
1351tcp_notifyall_oncpu(struct netmsg *netmsg)
1352{
1353 struct netmsg_tcp_notify *nmsg = (struct netmsg_tcp_notify *)netmsg;
1354 int nextcpu;
1355
1356 in_pcbnotifyall(&tcbinfo[mycpuid].pcblisthead, nmsg->nm_faddr,
1357 nmsg->nm_arg, nmsg->nm_notify);
1358
1359 nextcpu = mycpuid + 1;
1360 if (nextcpu < ncpus2)
1361 lwkt_forwardmsg(tcp_cport(nextcpu), &netmsg->nm_lmsg);
1362 else
1363 lwkt_replymsg(&netmsg->nm_lmsg, 0);
1364}
1365
984263bc 1366void
707ad4ed 1367tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
984263bc
MD
1368{
1369 struct ip *ip = vip;
1370 struct tcphdr *th;
1371 struct in_addr faddr;
1372 struct inpcb *inp;
1373 struct tcpcb *tp;
707ad4ed 1374 void (*notify)(struct inpcb *, int) = tcp_notify;
7d448528 1375 tcp_seq icmpseq;
1cae611f 1376 int arg, cpu;
7d448528
JH
1377
1378 if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
1379 return;
1380 }
984263bc
MD
1381
1382 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1383 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1384 return;
1385
7d448528
JH
1386 arg = inetctlerrmap[cmd];
1387 if (cmd == PRC_QUENCH) {
984263bc 1388 notify = tcp_quench;
7d448528
JH
1389 } else if (icmp_may_rst &&
1390 (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1391 cmd == PRC_UNREACH_PORT ||
1392 cmd == PRC_TIMXCEED_INTRANS) &&
1393 ip != NULL) {
984263bc 1394 notify = tcp_drop_syn_sent;
7d448528
JH
1395 } else if (cmd == PRC_MSGSIZE) {
1396 struct icmp *icmp = (struct icmp *)
1397 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1398
1399 arg = ntohs(icmp->icmp_nextmtu);
984263bc 1400 notify = tcp_mtudisc;
7d448528 1401 } else if (PRC_IS_REDIRECT(cmd)) {
707ad4ed 1402 ip = NULL;
984263bc 1403 notify = in_rtchange;
7d448528 1404 } else if (cmd == PRC_HOSTDEAD) {
707ad4ed 1405 ip = NULL;
7d448528
JH
1406 }
1407
707ad4ed 1408 if (ip != NULL) {
1cae611f 1409 crit_enter();
707ad4ed
JH
1410 th = (struct tcphdr *)((caddr_t)ip +
1411 (IP_VHL_HL(ip->ip_vhl) << 2));
d371a63a 1412 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
707ad4ed 1413 ip->ip_src.s_addr, th->th_sport);
d371a63a 1414 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport,
707ad4ed
JH
1415 ip->ip_src, th->th_sport, 0, NULL);
1416 if ((inp != NULL) && (inp->inp_socket != NULL)) {
7d448528 1417 icmpseq = htonl(th->th_seq);
984263bc 1418 tp = intotcpcb(inp);
7d448528
JH
1419 if (SEQ_GEQ(icmpseq, tp->snd_una) &&
1420 SEQ_LT(icmpseq, tp->snd_max))
1421 (*notify)(inp, arg);
984263bc
MD
1422 } else {
1423 struct in_conninfo inc;
1424
1425 inc.inc_fport = th->th_dport;
1426 inc.inc_lport = th->th_sport;
1427 inc.inc_faddr = faddr;
1428 inc.inc_laddr = ip->ip_src;
1429#ifdef INET6
1430 inc.inc_isipv6 = 0;
1431#endif
1432 syncache_unreach(&inc, th);
1433 }
1cae611f 1434 crit_exit();
d371a63a 1435 } else {
14572273
SZ
1436 struct netmsg_tcp_notify nmsg;
1437
1438 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
48e7b118
MD
1439 netmsg_init(&nmsg.nm_nmsg, NULL, &curthread->td_msgport,
1440 0, tcp_notifyall_oncpu);
14572273
SZ
1441 nmsg.nm_faddr = faddr;
1442 nmsg.nm_arg = arg;
1443 nmsg.nm_notify = notify;
1444
1445 lwkt_domsg(tcp_cport(0), &nmsg.nm_nmsg.nm_lmsg, 0);
d371a63a 1446 }
984263bc
MD
1447}
1448
1449#ifdef INET6
1450void
707ad4ed 1451tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
984263bc
MD
1452{
1453 struct tcphdr th;
42a7fc75 1454 void (*notify) (struct inpcb *, int) = tcp_notify;
984263bc
MD
1455 struct ip6_hdr *ip6;
1456 struct mbuf *m;
1457 struct ip6ctlparam *ip6cp = NULL;
1458 const struct sockaddr_in6 *sa6_src = NULL;
1459 int off;
1460 struct tcp_portonly {
1461 u_int16_t th_sport;
1462 u_int16_t th_dport;
1463 } *thp;
7d448528 1464 int arg;
984263bc
MD
1465
1466 if (sa->sa_family != AF_INET6 ||
1467 sa->sa_len != sizeof(struct sockaddr_in6))
1468 return;
1469
7d448528 1470 arg = 0;
984263bc
MD
1471 if (cmd == PRC_QUENCH)
1472 notify = tcp_quench;
7d448528
JH
1473 else if (cmd == PRC_MSGSIZE) {
1474 struct ip6ctlparam *ip6cp = d;
1475 struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6;
1476
1477 arg = ntohl(icmp6->icmp6_mtu);
984263bc 1478 notify = tcp_mtudisc;
7d448528
JH
1479 } else if (!PRC_IS_REDIRECT(cmd) &&
1480 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) {
984263bc 1481 return;
7d448528 1482 }
984263bc
MD
1483
1484 /* if the parameter is from icmp6, decode it. */
1485 if (d != NULL) {
1486 ip6cp = (struct ip6ctlparam *)d;
1487 m = ip6cp->ip6c_m;
1488 ip6 = ip6cp->ip6c_ip6;
1489 off = ip6cp->ip6c_off;
1490 sa6_src = ip6cp->ip6c_src;
1491 } else {
1492 m = NULL;
1493 ip6 = NULL;
1494 off = 0; /* fool gcc */
1495 sa6_src = &sa6_any;
1496 }
1497
707ad4ed 1498 if (ip6 != NULL) {
984263bc
MD
1499 struct in_conninfo inc;
1500 /*
1501 * XXX: We assume that when IPV6 is non NULL,
1502 * M and OFF are valid.
1503 */
1504
1505 /* check if we can safely examine src and dst ports */
707ad4ed 1506 if (m->m_pkthdr.len < off + sizeof *thp)
984263bc
MD
1507 return;
1508
707ad4ed
JH
1509 bzero(&th, sizeof th);
1510 m_copydata(m, off, sizeof *thp, (caddr_t)&th);
984263bc 1511
d2e9e54c 1512 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, th.th_dport,
984263bc 1513 (struct sockaddr *)ip6cp->ip6c_src,
7d448528 1514 th.th_sport, cmd, arg, notify);
984263bc
MD
1515
1516 inc.inc_fport = th.th_dport;
1517 inc.inc_lport = th.th_sport;
1518 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1519 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1520 inc.inc_isipv6 = 1;
1521 syncache_unreach(&inc, &th);
1522 } else
d2e9e54c 1523 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, 0,
7d448528 1524 (const struct sockaddr *)sa6_src, 0, cmd, arg, notify);
984263bc 1525}
707ad4ed 1526#endif
984263bc
MD
1527
1528/*
1529 * Following is where TCP initial sequence number generation occurs.
1530 *
1531 * There are two places where we must use initial sequence numbers:
1532 * 1. In SYN-ACK packets.
1533 * 2. In SYN packets.
1534 *
1535 * All ISNs for SYN-ACK packets are generated by the syncache. See
1536 * tcp_syncache.c for details.
1537 *
1538 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1539 * depends on this property. In addition, these ISNs should be
1540 * unguessable so as to prevent connection hijacking. To satisfy
1541 * the requirements of this situation, the algorithm outlined in
1542 * RFC 1948 is used to generate sequence numbers.
1543 *
1544 * Implementation details:
1545 *
1546 * Time is based off the system timer, and is corrected so that it
1547 * increases by one megabyte per second. This allows for proper
1548 * recycling on high speed LANs while still leaving over an hour
1549 * before rollover.
1550 *
1551 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1552 * between seeding of isn_secret. This is normally set to zero,
1553 * as reseeding should not be necessary.
1554 *
1555 */
1556
707ad4ed 1557#define ISN_BYTES_PER_SECOND 1048576
984263bc
MD
1558
1559u_char isn_secret[32];
1560int isn_last_reseed;
1561MD5_CTX isn_ctx;
1562
1563tcp_seq
707ad4ed 1564tcp_new_isn(struct tcpcb *tp)
984263bc
MD
1565{
1566 u_int32_t md5_buffer[4];
1567 tcp_seq new_isn;
1568
1569 /* Seed if this is the first use, reseed if requested. */
1570 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1571 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1572 < (u_int)ticks))) {
707ad4ed 1573 read_random_unlimited(&isn_secret, sizeof isn_secret);
984263bc
MD
1574 isn_last_reseed = ticks;
1575 }
707ad4ed 1576
984263bc
MD
1577 /* Compute the md5 hash and return the ISN. */
1578 MD5Init(&isn_ctx);
707ad4ed
JH
1579 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short));
1580 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short));
984263bc 1581#ifdef INET6
707ad4ed 1582 if (tp->t_inpcb->inp_vflag & INP_IPV6) {
984263bc
MD
1583 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1584 sizeof(struct in6_addr));
1585 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1586 sizeof(struct in6_addr));
1587 } else
1588#endif
1589 {
1590 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1591 sizeof(struct in_addr));
1592 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1593 sizeof(struct in_addr));
1594 }
1595 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1596 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1597 new_isn = (tcp_seq) md5_buffer[0];
1598 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
707ad4ed 1599 return (new_isn);
984263bc
MD
1600}
1601
1602/*
1603 * When a source quench is received, close congestion window
1604 * to one segment. We will gradually open it again as we proceed.
1605 */
1606void
71f385dc 1607tcp_quench(struct inpcb *inp, int error)
984263bc
MD
1608{
1609 struct tcpcb *tp = intotcpcb(inp);
1610
8acdb67c 1611 if (tp != NULL) {
984263bc 1612 tp->snd_cwnd = tp->t_maxseg;
8acdb67c
JH
1613 tp->snd_wacked = 0;
1614 }
984263bc
MD
1615}
1616
1617/*
1618 * When a specific ICMP unreachable message is received and the
1619 * connection state is SYN-SENT, drop the connection. This behavior
1620 * is controlled by the icmp_may_rst sysctl.
1621 */
1622void
71f385dc 1623tcp_drop_syn_sent(struct inpcb *inp, int error)
984263bc
MD
1624{
1625 struct tcpcb *tp = intotcpcb(inp);
1626
707ad4ed 1627 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT))
71f385dc 1628 tcp_drop(tp, error);
984263bc
MD
1629}
1630
1631/*
7d448528 1632 * When a `need fragmentation' ICMP is received, update our idea of the MSS
984263bc
MD
1633 * based on the new value in the route. Also nudge TCP to send something,
1634 * since we know the packet we just sent was dropped.
1635 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1636 */
1637void
7d448528 1638tcp_mtudisc(struct inpcb *inp, int mtu)
984263bc
MD
1639{
1640 struct tcpcb *tp = intotcpcb(inp);
1641 struct rtentry *rt;
984263bc 1642 struct socket *so = inp->inp_socket;
7d448528 1643 int maxopd, mss;
984263bc 1644#ifdef INET6
707ad4ed
JH
1645 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0);
1646#else
1647 const boolean_t isipv6 = FALSE;
1648#endif
984263bc 1649
7d448528
JH
1650 if (tp == NULL)
1651 return;
1652
1653 /*
1654 * If no MTU is provided in the ICMP message, use the
1655 * next lower likely value, as specified in RFC 1191.
1656 */
1657 if (mtu == 0) {
1658 int oldmtu;
1659
1660 oldmtu = tp->t_maxopd +
1661 (isipv6 ?
1662 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1663 sizeof(struct tcpiphdr));
1664 mtu = ip_next_mtu(oldmtu, 0);
1665 }
1666
1667 if (isipv6)
1668 rt = tcp_rtlookup6(&inp->inp_inc);
1669 else
1670 rt = tcp_rtlookup(&inp->inp_inc);
1671 if (rt != NULL) {
7d448528
JH
1672 if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu)
1673 mtu = rt->rt_rmx.rmx_mtu;
1674
1675 maxopd = mtu -
1676 (isipv6 ?
1677 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1678 sizeof(struct tcpiphdr));
1679
984263bc 1680 /*
7d448528 1681 * XXX - The following conditional probably violates the TCP
984263bc
MD
1682 * spec. The problem is that, since we don't know the
1683 * other end's MSS, we are supposed to use a conservative
1684 * default. But, if we do that, then MTU discovery will
1685 * never actually take place, because the conservative
1686 * default is much less than the MTUs typically seen
1687 * on the Internet today. For the moment, we'll sweep
1688 * this under the carpet.
1689 *
1690 * The conservative default might not actually be a problem
1691 * if the only case this occurs is when sending an initial
1692 * SYN with options and data to a host we've never talked
1693 * to before. Then, they will reply with an MSS value which
1694 * will get recorded and the new parameters should get
1695 * recomputed. For Further Study.
1696 */
27b8aee3
AE
1697 if (rt->rt_rmx.rmx_mssopt && rt->rt_rmx.rmx_mssopt < maxopd)
1698 maxopd = rt->rt_rmx.rmx_mssopt;
7d448528
JH
1699 } else
1700 maxopd = mtu -
1701 (isipv6 ?
1702 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1703 sizeof(struct tcpiphdr));
1704
1705 if (tp->t_maxopd <= maxopd)
1706 return;
1707 tp->t_maxopd = maxopd;
1708
1709 mss = maxopd;
1710 if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) ==
1711 (TF_REQ_TSTMP | TF_RCVD_TSTMP))
1712 mss -= TCPOLEN_TSTAMP_APPA;
1713
7d448528
JH
1714 /* round down to multiple of MCLBYTES */
1715#if (MCLBYTES & (MCLBYTES - 1)) == 0 /* test if MCLBYTES power of 2 */
1716 if (mss > MCLBYTES)
1717 mss &= ~(MCLBYTES - 1);
984263bc 1718#else
7d448528
JH
1719 if (mss > MCLBYTES)
1720 mss = (mss / MCLBYTES) * MCLBYTES;
984263bc 1721#endif
984263bc 1722
6d49aa6f
MD
1723 if (so->so_snd.ssb_hiwat < mss)
1724 mss = so->so_snd.ssb_hiwat;
984263bc 1725
7d448528
JH
1726 tp->t_maxseg = mss;
1727 tp->t_rtttime = 0;
1728 tp->snd_nxt = tp->snd_una;
1729 tcp_output(tp);
1730 tcpstat.tcps_mturesent++;
984263bc
MD
1731}
1732
1733/*
1734 * Look-up the routing entry to the peer of this inpcb. If no route
1735 * is found and it cannot be allocated the return NULL. This routine
1736 * is called by TCP routines that access the rmx structure and by tcp_mss
1737 * to get the interface MTU.
1738 */
1739struct rtentry *
707ad4ed 1740tcp_rtlookup(struct in_conninfo *inc)
984263bc 1741{
f23061d4 1742 struct route *ro = &inc->inc_route;
984263bc 1743
f23061d4 1744 if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) {
984263bc
MD
1745 /* No route yet, so try to acquire one */
1746 if (inc->inc_faddr.s_addr != INADDR_ANY) {
88fcebeb
MD
1747 /*
1748 * unused portions of the structure MUST be zero'd
1749 * out because rtalloc() treats it as opaque data
1750 */
1751 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
984263bc
MD
1752 ro->ro_dst.sa_family = AF_INET;
1753 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1754 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1755 inc->inc_faddr;
1756 rtalloc(ro);
984263bc
MD
1757 }
1758 }
f23061d4 1759 return (ro->ro_rt);
984263bc
MD
1760}
1761
1762#ifdef INET6
1763struct rtentry *
707ad4ed 1764tcp_rtlookup6(struct in_conninfo *inc)
984263bc 1765{
f23061d4 1766 struct route_in6 *ro6 = &inc->inc6_route;
984263bc 1767
f23061d4 1768 if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) {
984263bc
MD
1769 /* No route yet, so try to acquire one */
1770 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
88fcebeb
MD
1771 /*
1772 * unused portions of the structure MUST be zero'd
1773 * out because rtalloc() treats it as opaque data
1774 */
1775 bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6));
984263bc
MD
1776 ro6->ro_dst.sin6_family = AF_INET6;
1777 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1778 ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1779 rtalloc((struct route *)ro6);
984263bc
MD
1780 }
1781 }
f23061d4 1782 return (ro6->ro_rt);
984263bc 1783}
707ad4ed 1784#endif
984263bc
MD
1785
1786#ifdef IPSEC
1787/* compute ESP/AH header size for TCP, including outer IP header. */
1788size_t
707ad4ed 1789ipsec_hdrsiz_tcp(struct tcpcb *tp)
984263bc
MD
1790{
1791 struct inpcb *inp;
1792 struct mbuf *m;
1793 size_t hdrsiz;
1794 struct ip *ip;
984263bc
MD
1795 struct tcphdr *th;
1796
1797 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
707ad4ed 1798 return (0);
74f1caca 1799 MGETHDR(m, MB_DONTWAIT, MT_DATA);
984263bc 1800 if (!m)
707ad4ed 1801 return (0);
984263bc
MD
1802
1803#ifdef INET6
707ad4ed
JH
1804 if (inp->inp_vflag & INP_IPV6) {
1805 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1806
984263bc
MD
1807 th = (struct tcphdr *)(ip6 + 1);
1808 m->m_pkthdr.len = m->m_len =
707ad4ed 1809 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
984263bc
MD
1810 tcp_fillheaders(tp, ip6, th);
1811 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1812 } else
707ad4ed
JH
1813#endif
1814 {
1815 ip = mtod(m, struct ip *);
1816 th = (struct tcphdr *)(ip + 1);
1817 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1818 tcp_fillheaders(tp, ip, th);
1819 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1820 }
984263bc
MD
1821
1822 m_free(m);
707ad4ed 1823 return (hdrsiz);
984263bc 1824}
707ad4ed 1825#endif
984263bc 1826
984263bc
MD
1827/*
1828 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1829 *
1830 * This code attempts to calculate the bandwidth-delay product as a
1831 * means of determining the optimal window size to maximize bandwidth,
1832 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1833 * routers. This code also does a fairly good job keeping RTTs in check
1834 * across slow links like modems. We implement an algorithm which is very
1835 * similar (but not meant to be) TCP/Vegas. The code operates on the
1836 * transmitter side of a TCP connection and so only effects the transmit
1837 * side of the connection.
1838 *
1839 * BACKGROUND: TCP makes no provision for the management of buffer space
f23061d4 1840 * at the end points or at the intermediate routers and switches. A TCP
984263bc
MD
1841 * stream, whether using NewReno or not, will eventually buffer as
1842 * many packets as it is able and the only reason this typically works is
1843 * due to the fairly small default buffers made available for a connection
1844 * (typicaly 16K or 32K). As machines use larger windows and/or window
1845 * scaling it is now fairly easy for even a single TCP connection to blow-out
f23061d4 1846 * all available buffer space not only on the local interface, but on
984263bc
MD
1847 * intermediate routers and switches as well. NewReno makes a misguided
1848 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1849 * then backing off, then steadily increasing the window again until another
1850 * failure occurs, ad-infinitum. This results in terrible oscillation that
1851 * is only made worse as network loads increase and the idea of intentionally
1852 * blowing out network buffers is, frankly, a terrible way to manage network
1853 * resources.
1854 *
1855 * It is far better to limit the transmit window prior to the failure
1856 * condition being achieved. There are two general ways to do this: First
1857 * you can 'scan' through different transmit window sizes and locate the
1858 * point where the RTT stops increasing, indicating that you have filled the
1859 * pipe, then scan backwards until you note that RTT stops decreasing, then
1860 * repeat ad-infinitum. This method works in principle but has severe
1861 * implementation issues due to RTT variances, timer granularity, and
1862 * instability in the algorithm which can lead to many false positives and
1863 * create oscillations as well as interact badly with other TCP streams
1864 * implementing the same algorithm.
1865 *
1866 * The second method is to limit the window to the bandwidth delay product
1867 * of the link. This is the method we implement. RTT variances and our
f23061d4 1868 * own manipulation of the congestion window, bwnd, can potentially
984263bc
MD
1869 * destabilize the algorithm. For this reason we have to stabilize the
1870 * elements used to calculate the window. We do this by using the minimum
1871 * observed RTT, the long term average of the observed bandwidth, and
1872 * by adding two segments worth of slop. It isn't perfect but it is able
1873 * to react to changing conditions and gives us a very stable basis on
1874 * which to extend the algorithm.
1875 */
1876void
1877tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1878{
1879 u_long bw;
1880 u_long bwnd;
1881 int save_ticks;
421de19e 1882 int delta_ticks;
984263bc
MD
1883
1884 /*
1885 * If inflight_enable is disabled in the middle of a tcp connection,
1886 * make sure snd_bwnd is effectively disabled.
1887 */
707ad4ed 1888 if (!tcp_inflight_enable) {
984263bc
MD
1889 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1890 tp->snd_bandwidth = 0;
1891 return;
1892 }
1893
421de19e
MD
1894 /*
1895 * Validate the delta time. If a connection is new or has been idle
1896 * a long time we have to reset the bandwidth calculator.
1897 */
1898 save_ticks = ticks;
1899 delta_ticks = save_ticks - tp->t_bw_rtttime;
1900 if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) {
1901 tp->t_bw_rtttime = ticks;
1902 tp->t_bw_rtseq = ack_seq;
1903 if (tp->snd_bandwidth == 0)
1904 tp->snd_bandwidth = tcp_inflight_min;
1905 return;
1906 }
1907 if (delta_ticks == 0)
1908 return;
1909
1910 /*
1911 * Sanity check, plus ignore pure window update acks.
1912 */
1913 if ((int)(ack_seq - tp->t_bw_rtseq) <= 0)
1914 return;
1915
984263bc
MD
1916 /*
1917 * Figure out the bandwidth. Due to the tick granularity this
1918 * is a very rough number and it MUST be averaged over a fairly
1919 * long period of time. XXX we need to take into account a link
1920 * that is not using all available bandwidth, but for now our
1921 * slop will ramp us up if this case occurs and the bandwidth later
1922 * increases.
984263bc 1923 */
421de19e 1924 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks;
984263bc
MD
1925 tp->t_bw_rtttime = save_ticks;
1926 tp->t_bw_rtseq = ack_seq;
984263bc
MD
1927 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1928
1929 tp->snd_bandwidth = bw;
1930
1931 /*
1932 * Calculate the semi-static bandwidth delay product, plus two maximal
1933 * segments. The additional slop puts us squarely in the sweet
1934 * spot and also handles the bandwidth run-up case. Without the
1935 * slop we could be locking ourselves into a lower bandwidth.
1936 *
1937 * Situations Handled:
1938 * (1) Prevents over-queueing of packets on LANs, especially on
1939 * high speed LANs, allowing larger TCP buffers to be
f23061d4 1940 * specified, and also does a good job preventing
984263bc
MD
1941 * over-queueing of packets over choke points like modems
1942 * (at least for the transmit side).
1943 *
1944 * (2) Is able to handle changing network loads (bandwidth
1945 * drops so bwnd drops, bandwidth increases so bwnd
1946 * increases).
1947 *
1948 * (3) Theoretically should stabilize in the face of multiple
1949 * connections implementing the same algorithm (this may need
1950 * a little work).
1951 *
f23061d4 1952 * (4) Stability value (defaults to 20 = 2 maximal packets) can
984263bc
MD
1953 * be adjusted with a sysctl but typically only needs to be on
1954 * very slow connections. A value no smaller then 5 should
1955 * be used, but only reduce this default if you have no other
1956 * choice.
1957 */
707ad4ed
JH
1958
1959#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1960 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) +
1961 tcp_inflight_stab * (int)tp->t_maxseg / 10;
984263bc
MD
1962#undef USERTT
1963
1964 if (tcp_inflight_debug > 0) {
1965 static int ltime;
1966 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1967 ltime = ticks;
a6ec04bc 1968 kprintf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
707ad4ed 1969 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd);
984263bc
MD
1970 }
1971 }
1972 if ((long)bwnd < tcp_inflight_min)
1973 bwnd = tcp_inflight_min;
1974 if (bwnd > tcp_inflight_max)
1975 bwnd = tcp_inflight_max;
1976 if ((long)bwnd < tp->t_maxseg * 2)
1977 bwnd = tp->t_maxseg * 2;
1978 tp->snd_bwnd = bwnd;
1979}