1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 * $FreeBSD: releng/6.2/sys/netinet/tcp_subr.c 162908 2006-10-01 05:33:50Z silby $
31 */
32
33 #include "opt_compat.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_sack.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/mac.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/proc.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
57
58 #include <vm/uma.h>
59
60 #include <net/route.h>
61 #include <net/if.h>
62
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69 #include <netinet/in_pcb.h>
70 #ifdef INET6
71 #include <netinet6/in6_pcb.h>
72 #endif
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #endif
80 #include <netinet/ip_icmp.h>
81 #include <netinet/tcp.h>
82 #include <netinet/tcp_fsm.h>
83 #include <netinet/tcp_seq.h>
84 #include <netinet/tcp_timer.h>
85 #include <netinet/tcp_var.h>
86 #ifdef INET6
87 #include <netinet6/tcp6_var.h>
88 #endif
89 #include <netinet/tcpip.h>
90 #ifdef TCPDEBUG
91 #include <netinet/tcp_debug.h>
92 #endif
93 #include <netinet6/ip6protosw.h>
94
95 #ifdef IPSEC
96 #include <netinet6/ipsec.h>
97 #ifdef INET6
98 #include <netinet6/ipsec6.h>
99 #endif
100 #include <netkey/key.h>
101 #endif /*IPSEC*/
102
103 #ifdef FAST_IPSEC
104 #include <netipsec/ipsec.h>
105 #include <netipsec/xform.h>
106 #ifdef INET6
107 #include <netipsec/ipsec6.h>
108 #endif
109 #include <netipsec/key.h>
110 #define IPSEC
111 #endif /*FAST_IPSEC*/
112
113 #include <machine/in_cksum.h>
114 #include <sys/md5.h>
115
116 int tcp_mssdflt = TCP_MSS;
117 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
119
120 #ifdef INET6
121 int tcp_v6mssdflt = TCP6_MSS;
122 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
123 CTLFLAG_RW, &tcp_v6mssdflt , 0,
124 "Default TCP Maximum Segment Size for IPv6");
125 #endif
126
127 /*
128 * Minimum MSS we accept and use. This prevents DoS attacks where
129 * we are forced to a ridiculous low MSS like 20 and send hundreds
130 * of packets instead of one. The effect scales with the available
131 * bandwidth and quickly saturates the CPU and network interface
132 * with packet generation and sending. Set to zero to disable MINMSS
133 * checking. This setting prevents us from sending too small packets.
134 */
135 int tcp_minmss = TCP_MINMSS;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
138 /*
139 * Number of TCP segments per second we accept from remote host
140 * before we start to calculate average segment size. If average
141 * segment size drops below the minimum TCP MSS we assume a DoS
142 * attack and reset+drop the connection. Care has to be taken not to
143 * set this value too small to not kill interactive type connections
144 * (telnet, SSH) which send many small packets.
145 */
146 int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
149 "be under the MINMSS Size");
150
151 #if 0
152 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
153 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
155 #endif
156
157 int tcp_do_rfc1323 = 1;
158 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
160
161 static int tcp_tcbhashsize = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
164
165 static int do_tcpdrain = 1;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
167 "Enable tcp_drain routine for extra help when low on mbufs");
168
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
170 &tcbinfo.ipi_count, 0, "Number of active PCBs");
171
172 static int icmp_may_rst = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
174 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
175
176 static int tcp_isn_reseed_interval = 0;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
179
180 static uma_zone_t tcptw_zone;
181 static int maxtcptw;
182
183 static int
184 tcptw_auto_size(void)
185 {
186 int halfrange;
187
188 /*
189 * Max out at half the ephemeral port range so that TIME_WAIT
190 * sockets don't tie up too many ephemeral ports.
191 */
192 if (ipport_lastauto > ipport_firstauto)
193 halfrange = (ipport_lastauto - ipport_firstauto) / 2;
194 else
195 halfrange = (ipport_firstauto - ipport_lastauto) / 2;
196 /* Protect against goofy port ranges smaller than 32. */
197 return (imin(imax(halfrange, 32), maxsockets / 5));
198 }
199
200 static int
201 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
202 {
203 int error, new;
204
205 if (maxtcptw == 0)
206 new = tcptw_auto_size();
207 else
208 new = maxtcptw;
209 error = sysctl_handle_int(oidp, &new, sizeof(int), req);
210 if (error == 0 && req->newptr)
211 if (new >= 32) {
212 maxtcptw = new;
213 uma_zone_set_max(tcptw_zone, maxtcptw);
214 }
215 return (error);
216 }
217 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
218 &maxtcptw, 0, sysctl_maxtcptw, "IU",
219 "Maximum number of compressed TCP TIME_WAIT entries");
220
221 static int nolocaltimewait = 0;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
223 &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries"
224 "for local connections");
225
226 /*
227 * TCP bandwidth limiting sysctls. Note that the default lower bound of
228 * 1024 exists only for debugging. A good production default would be
229 * something like 6100.
230 */
231 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
232 "TCP inflight data limiting");
233
234 static int tcp_inflight_enable = 1;
235 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
236 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
237
238 static int tcp_inflight_debug = 0;
239 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
240 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
241
242 static int tcp_inflight_rttthresh;
243 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
244 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
245 "RTT threshold below which inflight will deactivate itself");
246
247 static int tcp_inflight_min = 6144;
248 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
249 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
250
251 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
252 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
253 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
254
255 static int tcp_inflight_stab = 20;
256 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
257 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
258
259 uma_zone_t sack_hole_zone;
260
261 static struct inpcb *tcp_notify(struct inpcb *, int);
262 static void tcp_discardcb(struct tcpcb *);
263 static void tcp_isn_tick(void *);
264
265 /*
266 * Target size of TCP PCB hash tables. Must be a power of two.
267 *
268 * Note that this can be overridden by the kernel environment
269 * variable net.inet.tcp.tcbhashsize
270 */
271 #ifndef TCBHASHSIZE
272 #define TCBHASHSIZE 512
273 #endif
274
275 /*
276 * XXX
277 * Callouts should be moved into struct tcp directly. They are currently
278 * separate because the tcpcb structure is exported to userland for sysctl
279 * parsing purposes, which do not know about callouts.
280 */
281 struct tcpcb_mem {
282 struct tcpcb tcb;
283 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
284 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
285 };
286
287 static uma_zone_t tcpcb_zone;
288 struct callout isn_callout;
289
290 /*
291 * Tcp initialization
292 */
293 static void
294 tcp_zone_change(void *tag)
295 {
296
297 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
298 uma_zone_set_max(tcpcb_zone, maxsockets);
299 if (maxtcptw == 0)
300 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
301 }
302
303 void
304 tcp_init()
305 {
306 int hashsize = TCBHASHSIZE;
307
308 tcp_delacktime = TCPTV_DELACK;
309 tcp_keepinit = TCPTV_KEEP_INIT;
310 tcp_keepidle = TCPTV_KEEP_IDLE;
311 tcp_keepintvl = TCPTV_KEEPINTVL;
312 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
313 tcp_msl = TCPTV_MSL;
314 tcp_rexmit_min = TCPTV_MIN;
315 tcp_rexmit_slop = TCPTV_CPU_VAR;
316 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
317
318 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
319 LIST_INIT(&tcb);
320 tcbinfo.listhead = &tcb;
321 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
322 if (!powerof2(hashsize)) {
323 printf("WARNING: TCB hash size not a power of 2\n");
324 hashsize = 512; /* safe default */
325 }
326 tcp_tcbhashsize = hashsize;
327 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
328 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
329 &tcbinfo.porthashmask);
330 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
331 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
332 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
333 #ifdef INET6
334 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
335 #else /* INET6 */
336 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
337 #endif /* INET6 */
338 if (max_protohdr < TCP_MINPROTOHDR)
339 max_protohdr = TCP_MINPROTOHDR;
340 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
341 panic("tcp_init");
342 #undef TCP_MINPROTOHDR
343 /*
344 * These have to be type stable for the benefit of the timers.
345 */
346 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
347 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
348 uma_zone_set_max(tcpcb_zone, maxsockets);
349 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
350 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
351 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
352 if (maxtcptw == 0)
353 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
354 else
355 uma_zone_set_max(tcptw_zone, maxtcptw);
356 tcp_timer_init();
357 syncache_init();
358 tcp_hc_init();
359 tcp_reass_init();
360 callout_init(&isn_callout, CALLOUT_MPSAFE);
361 tcp_isn_tick(NULL);
362 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
363 SHUTDOWN_PRI_DEFAULT);
364 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
365 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
366 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
367 EVENTHANDLER_PRI_ANY);
368 }
369
370 void
371 tcp_fini(xtp)
372 void *xtp;
373 {
374 callout_stop(&isn_callout);
375
376 }
377
378 /*
379 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
380 * tcp_template used to store this data in mbufs, but we now recopy it out
381 * of the tcpcb each time to conserve mbufs.
382 */
383 void
384 tcpip_fillheaders(inp, ip_ptr, tcp_ptr)
385 struct inpcb *inp;
386 void *ip_ptr;
387 void *tcp_ptr;
388 {
389 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
390
391 INP_LOCK_ASSERT(inp);
392
393 #ifdef INET6
394 if ((inp->inp_vflag & INP_IPV6) != 0) {
395 struct ip6_hdr *ip6;
396
397 ip6 = (struct ip6_hdr *)ip_ptr;
398 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
399 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
400 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
401 (IPV6_VERSION & IPV6_VERSION_MASK);
402 ip6->ip6_nxt = IPPROTO_TCP;
403 ip6->ip6_plen = sizeof(struct tcphdr);
404 ip6->ip6_src = inp->in6p_laddr;
405 ip6->ip6_dst = inp->in6p_faddr;
406 } else
407 #endif
408 {
409 struct ip *ip;
410
411 ip = (struct ip *)ip_ptr;
412 ip->ip_v = IPVERSION;
413 ip->ip_hl = 5;
414 ip->ip_tos = inp->inp_ip_tos;
415 ip->ip_len = 0;
416 ip->ip_id = 0;
417 ip->ip_off = 0;
418 ip->ip_ttl = inp->inp_ip_ttl;
419 ip->ip_sum = 0;
420 ip->ip_p = IPPROTO_TCP;
421 ip->ip_src = inp->inp_laddr;
422 ip->ip_dst = inp->inp_faddr;
423 }
424 th->th_sport = inp->inp_lport;
425 th->th_dport = inp->inp_fport;
426 th->th_seq = 0;
427 th->th_ack = 0;
428 th->th_x2 = 0;
429 th->th_off = 5;
430 th->th_flags = 0;
431 th->th_win = 0;
432 th->th_urp = 0;
433 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
434 }
435
436 /*
437 * Create template to be used to send tcp packets on a connection.
438 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
439 * use for this function is in keepalives, which use tcp_respond.
440 */
441 struct tcptemp *
442 tcpip_maketemplate(inp)
443 struct inpcb *inp;
444 {
445 struct mbuf *m;
446 struct tcptemp *n;
447
448 m = m_get(M_DONTWAIT, MT_DATA);
449 if (m == NULL)
450 return (0);
451 m->m_len = sizeof(struct tcptemp);
452 n = mtod(m, struct tcptemp *);
453
454 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
455 return (n);
456 }
457
458 /*
459 * Send a single message to the TCP at address specified by
460 * the given TCP/IP header. If m == NULL, then we make a copy
461 * of the tcpiphdr at ti and send directly to the addressed host.
462 * This is used to force keep alive messages out using the TCP
463 * template for a connection. If flags are given then we send
464 * a message back to the TCP which originated the * segment ti,
465 * and discard the mbuf containing it and any other attached mbufs.
466 *
467 * In any case the ack and sequence number of the transmitted
468 * segment are as specified by the parameters.
469 *
470 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
471 */
472 void
473 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
474 struct tcpcb *tp;
475 void *ipgen;
476 register struct tcphdr *th;
477 register struct mbuf *m;
478 tcp_seq ack, seq;
479 int flags;
480 {
481 register int tlen;
482 int win = 0;
483 struct ip *ip;
484 struct tcphdr *nth;
485 #ifdef INET6
486 struct ip6_hdr *ip6;
487 int isipv6;
488 #endif /* INET6 */
489 int ipflags = 0;
490 struct inpcb *inp;
491
492 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
493
494 #ifdef INET6
495 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
496 ip6 = ipgen;
497 #endif /* INET6 */
498 ip = ipgen;
499
500 if (tp != NULL) {
501 inp = tp->t_inpcb;
502 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
503 INP_INFO_WLOCK_ASSERT(&tcbinfo);
504 INP_LOCK_ASSERT(inp);
505 } else
506 inp = NULL;
507
508 if (tp != NULL) {
509 if (!(flags & TH_RST)) {
510 win = sbspace(&inp->inp_socket->so_rcv);
511 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
512 win = (long)TCP_MAXWIN << tp->rcv_scale;
513 }
514 }
515 if (m == NULL) {
516 m = m_gethdr(M_DONTWAIT, MT_HEADER);
517 if (m == NULL)
518 return;
519 tlen = 0;
520 m->m_data += max_linkhdr;
521 #ifdef INET6
522 if (isipv6) {
523 bcopy((caddr_t)ip6, mtod(m, caddr_t),
524 sizeof(struct ip6_hdr));
525 ip6 = mtod(m, struct ip6_hdr *);
526 nth = (struct tcphdr *)(ip6 + 1);
527 } else
528 #endif /* INET6 */
529 {
530 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
531 ip = mtod(m, struct ip *);
532 nth = (struct tcphdr *)(ip + 1);
533 }
534 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
535 flags = TH_ACK;
536 } else {
537 m_freem(m->m_next);
538 m->m_next = NULL;
539 m->m_data = (caddr_t)ipgen;
540 /* m_len is set later */
541 tlen = 0;
542 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
543 #ifdef INET6
544 if (isipv6) {
545 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
546 nth = (struct tcphdr *)(ip6 + 1);
547 } else
548 #endif /* INET6 */
549 {
550 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
551 nth = (struct tcphdr *)(ip + 1);
552 }
553 if (th != nth) {
554 /*
555 * this is usually a case when an extension header
556 * exists between the IPv6 header and the
557 * TCP header.
558 */
559 nth->th_sport = th->th_sport;
560 nth->th_dport = th->th_dport;
561 }
562 xchg(nth->th_dport, nth->th_sport, n_short);
563 #undef xchg
564 }
565 #ifdef INET6
566 if (isipv6) {
567 ip6->ip6_flow = 0;
568 ip6->ip6_vfc = IPV6_VERSION;
569 ip6->ip6_nxt = IPPROTO_TCP;
570 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
571 tlen));
572 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
573 } else
574 #endif
575 {
576 tlen += sizeof (struct tcpiphdr);
577 ip->ip_len = tlen;
578 ip->ip_ttl = ip_defttl;
579 if (path_mtu_discovery)
580 ip->ip_off |= IP_DF;
581 }
582 m->m_len = tlen;
583 m->m_pkthdr.len = tlen;
584 m->m_pkthdr.rcvif = NULL;
585 #ifdef MAC
586 if (inp != NULL) {
587 /*
588 * Packet is associated with a socket, so allow the
589 * label of the response to reflect the socket label.
590 */
591 INP_LOCK_ASSERT(inp);
592 mac_create_mbuf_from_inpcb(inp, m);
593 } else {
594 /*
595 * Packet is not associated with a socket, so possibly
596 * update the label in place.
597 */
598 mac_reflect_mbuf_tcp(m);
599 }
600 #endif
601 nth->th_seq = htonl(seq);
602 nth->th_ack = htonl(ack);
603 nth->th_x2 = 0;
604 nth->th_off = sizeof (struct tcphdr) >> 2;
605 nth->th_flags = flags;
606 if (tp != NULL)
607 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
608 else
609 nth->th_win = htons((u_short)win);
610 nth->th_urp = 0;
611 #ifdef INET6
612 if (isipv6) {
613 nth->th_sum = 0;
614 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
615 sizeof(struct ip6_hdr),
616 tlen - sizeof(struct ip6_hdr));
617 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
618 NULL, NULL);
619 } else
620 #endif /* INET6 */
621 {
622 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
623 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
624 m->m_pkthdr.csum_flags = CSUM_TCP;
625 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
626 }
627 #ifdef TCPDEBUG
628 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
629 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
630 #endif
631 #ifdef INET6
632 if (isipv6)
633 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
634 else
635 #endif /* INET6 */
636 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
637 }
638
639 /*
640 * Create a new TCP control block, making an
641 * empty reassembly queue and hooking it to the argument
642 * protocol control block. The `inp' parameter must have
643 * come from the zone allocator set up in tcp_init().
644 */
645 struct tcpcb *
646 tcp_newtcpcb(inp)
647 struct inpcb *inp;
648 {
649 struct tcpcb_mem *tm;
650 struct tcpcb *tp;
651 #ifdef INET6
652 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
653 #endif /* INET6 */
654
655 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
656 if (tm == NULL)
657 return (NULL);
658 tp = &tm->tcb;
659 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
660 tp->t_maxseg = tp->t_maxopd =
661 #ifdef INET6
662 isipv6 ? tcp_v6mssdflt :
663 #endif /* INET6 */
664 tcp_mssdflt;
665
666 /* Set up our timeouts. */
667 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
668 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
669 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
670 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
671 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
672
673 if (tcp_do_rfc1323)
674 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
675 tp->sack_enable = tcp_do_sack;
676 TAILQ_INIT(&tp->snd_holes);
677 tp->t_inpcb = inp; /* XXX */
678 /*
679 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
680 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
681 * reasonable initial retransmit time.
682 */
683 tp->t_srtt = TCPTV_SRTTBASE;
684 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
685 tp->t_rttmin = tcp_rexmit_min;
686 tp->t_rxtcur = TCPTV_RTOBASE;
687 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
688 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
689 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
690 tp->t_rcvtime = ticks;
691 tp->t_bw_rtttime = ticks;
692 /*
693 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
694 * because the socket may be bound to an IPv6 wildcard address,
695 * which may match an IPv4-mapped IPv6 address.
696 */
697 inp->inp_ip_ttl = ip_defttl;
698 inp->inp_ppcb = (caddr_t)tp;
699 return (tp); /* XXX */
700 }
701
702 /*
703 * Drop a TCP connection, reporting
704 * the specified error. If connection is synchronized,
705 * then send a RST to peer.
706 */
707 struct tcpcb *
708 tcp_drop(tp, errno)
709 register struct tcpcb *tp;
710 int errno;
711 {
712 struct socket *so = tp->t_inpcb->inp_socket;
713
714 INP_INFO_WLOCK_ASSERT(&tcbinfo);
715 INP_LOCK_ASSERT(tp->t_inpcb);
716
717 if (TCPS_HAVERCVDSYN(tp->t_state)) {
718 tp->t_state = TCPS_CLOSED;
719 (void) tcp_output(tp);
720 tcpstat.tcps_drops++;
721 } else
722 tcpstat.tcps_conndrops++;
723 if (errno == ETIMEDOUT && tp->t_softerror)
724 errno = tp->t_softerror;
725 so->so_error = errno;
726 return (tcp_close(tp));
727 }
728
729 static void
730 tcp_discardcb(tp)
731 struct tcpcb *tp;
732 {
733 struct tseg_qent *q;
734 struct inpcb *inp = tp->t_inpcb;
735 struct socket *so = inp->inp_socket;
736 #ifdef INET6
737 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
738 #endif /* INET6 */
739
740 INP_LOCK_ASSERT(inp);
741
742 /*
743 * Make sure that all of our timers are stopped before we
744 * delete the PCB.
745 */
746 callout_stop(tp->tt_rexmt);
747 callout_stop(tp->tt_persist);
748 callout_stop(tp->tt_keep);
749 callout_stop(tp->tt_2msl);
750 callout_stop(tp->tt_delack);
751
752 /*
753 * If we got enough samples through the srtt filter,
754 * save the rtt and rttvar in the routing entry.
755 * 'Enough' is arbitrarily defined as 4 rtt samples.
756 * 4 samples is enough for the srtt filter to converge
757 * to within enough % of the correct value; fewer samples
758 * and we could save a bogus rtt. The danger is not high
759 * as tcp quickly recovers from everything.
760 * XXX: Works very well but needs some more statistics!
761 */
762 if (tp->t_rttupdated >= 4) {
763 struct hc_metrics_lite metrics;
764 u_long ssthresh;
765
766 bzero(&metrics, sizeof(metrics));
767 /*
768 * Update the ssthresh always when the conditions below
769 * are satisfied. This gives us better new start value
770 * for the congestion avoidance for new connections.
771 * ssthresh is only set if packet loss occured on a session.
772 */
773 ssthresh = tp->snd_ssthresh;
774 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
775 /*
776 * convert the limit from user data bytes to
777 * packets then to packet data bytes.
778 */
779 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
780 if (ssthresh < 2)
781 ssthresh = 2;
782 ssthresh *= (u_long)(tp->t_maxseg +
783 #ifdef INET6
784 (isipv6 ? sizeof (struct ip6_hdr) +
785 sizeof (struct tcphdr) :
786 #endif
787 sizeof (struct tcpiphdr)
788 #ifdef INET6
789 )
790 #endif
791 );
792 } else
793 ssthresh = 0;
794 metrics.rmx_ssthresh = ssthresh;
795
796 metrics.rmx_rtt = tp->t_srtt;
797 metrics.rmx_rttvar = tp->t_rttvar;
798 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
799 metrics.rmx_bandwidth = tp->snd_bandwidth;
800 metrics.rmx_cwnd = tp->snd_cwnd;
801 metrics.rmx_sendpipe = 0;
802 metrics.rmx_recvpipe = 0;
803
804 tcp_hc_update(&inp->inp_inc, &metrics);
805 }
806
807 /* free the reassembly queue, if any */
808 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
809 LIST_REMOVE(q, tqe_q);
810 m_freem(q->tqe_m);
811 uma_zfree(tcp_reass_zone, q);
812 tp->t_segqlen--;
813 tcp_reass_qsize--;
814 }
815 tcp_free_sackholes(tp);
816 inp->inp_ppcb = NULL;
817 tp->t_inpcb = NULL;
818 uma_zfree(tcpcb_zone, tp);
819 soisdisconnected(so);
820 }
821
822 /*
823 * Close a TCP control block:
824 * discard all space held by the tcp
825 * discard internet protocol block
826 * wake up any sleepers
827 */
828 struct tcpcb *
829 tcp_close(tp)
830 struct tcpcb *tp;
831 {
832 struct inpcb *inp = tp->t_inpcb;
833 #ifdef INET6
834 struct socket *so = inp->inp_socket;
835 #endif
836
837 INP_INFO_WLOCK_ASSERT(&tcbinfo);
838 INP_LOCK_ASSERT(inp);
839
840 tcp_discardcb(tp);
841 #ifdef INET6
842 if (INP_CHECK_SOCKAF(so, AF_INET6))
843 in6_pcbdetach(inp);
844 else
845 #endif
846 in_pcbdetach(inp);
847 tcpstat.tcps_closed++;
848 return (NULL);
849 }
850
851 void
852 tcp_drain()
853 {
854 if (do_tcpdrain)
855 {
856 struct inpcb *inpb;
857 struct tcpcb *tcpb;
858 struct tseg_qent *te;
859
860 /*
861 * Walk the tcpbs, if existing, and flush the reassembly queue,
862 * if there is one...
863 * XXX: The "Net/3" implementation doesn't imply that the TCP
864 * reassembly queue should be flushed, but in a situation
865 * where we're really low on mbufs, this is potentially
866 * usefull.
867 */
868 INP_INFO_RLOCK(&tcbinfo);
869 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
870 if (inpb->inp_vflag & INP_TIMEWAIT)
871 continue;
872 INP_LOCK(inpb);
873 if ((tcpb = intotcpcb(inpb)) != NULL) {
874 while ((te = LIST_FIRST(&tcpb->t_segq))
875 != NULL) {
876 LIST_REMOVE(te, tqe_q);
877 m_freem(te->tqe_m);
878 uma_zfree(tcp_reass_zone, te);
879 tcpb->t_segqlen--;
880 tcp_reass_qsize--;
881 }
882 tcp_clean_sackreport(tcpb);
883 }
884 INP_UNLOCK(inpb);
885 }
886 INP_INFO_RUNLOCK(&tcbinfo);
887 }
888 }
889
890 /*
891 * Notify a tcp user of an asynchronous error;
892 * store error as soft error, but wake up user
893 * (for now, won't do anything until can select for soft error).
894 *
895 * Do not wake up user since there currently is no mechanism for
896 * reporting soft errors (yet - a kqueue filter may be added).
897 */
898 static struct inpcb *
899 tcp_notify(inp, error)
900 struct inpcb *inp;
901 int error;
902 {
903 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
904
905 INP_INFO_WLOCK_ASSERT(&tcbinfo);
906 INP_LOCK_ASSERT(inp);
907
908 /*
909 * Ignore some errors if we are hooked up.
910 * If connection hasn't completed, has retransmitted several times,
911 * and receives a second error, give up now. This is better
912 * than waiting a long time to establish a connection that
913 * can never complete.
914 */
915 if (tp->t_state == TCPS_ESTABLISHED &&
916 (error == EHOSTUNREACH || error == ENETUNREACH ||
917 error == EHOSTDOWN)) {
918 return (inp);
919 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
920 tp->t_softerror) {
921 tcp_drop(tp, error);
922 return (struct inpcb *)0;
923 } else {
924 tp->t_softerror = error;
925 return (inp);
926 }
927 #if 0
928 wakeup( &so->so_timeo);
929 sorwakeup(so);
930 sowwakeup(so);
931 #endif
932 }
933
934 static int
935 tcp_pcblist(SYSCTL_HANDLER_ARGS)
936 {
937 int error, i, n;
938 struct inpcb *inp, **inp_list;
939 inp_gen_t gencnt;
940 struct xinpgen xig;
941
942 /*
943 * The process of preparing the TCB list is too time-consuming and
944 * resource-intensive to repeat twice on every request.
945 */
946 if (req->oldptr == NULL) {
947 n = tcbinfo.ipi_count;
948 req->oldidx = 2 * (sizeof xig)
949 + (n + n/8) * sizeof(struct xtcpcb);
950 return (0);
951 }
952
953 if (req->newptr != NULL)
954 return (EPERM);
955
956 /*
957 * OK, now we're committed to doing something.
958 */
959 INP_INFO_RLOCK(&tcbinfo);
960 gencnt = tcbinfo.ipi_gencnt;
961 n = tcbinfo.ipi_count;
962 INP_INFO_RUNLOCK(&tcbinfo);
963
964 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
965 + n * sizeof(struct xtcpcb));
966 if (error != 0)
967 return (error);
968
969 xig.xig_len = sizeof xig;
970 xig.xig_count = n;
971 xig.xig_gen = gencnt;
972 xig.xig_sogen = so_gencnt;
973 error = SYSCTL_OUT(req, &xig, sizeof xig);
974 if (error)
975 return (error);
976
977 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
978 if (inp_list == NULL)
979 return (ENOMEM);
980
981 INP_INFO_RLOCK(&tcbinfo);
982 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
983 inp = LIST_NEXT(inp, inp_list)) {
984 INP_LOCK(inp);
985 if (inp->inp_gencnt <= gencnt) {
986 /*
987 * XXX: This use of cr_cansee(), introduced with
988 * TCP state changes, is not quite right, but for
989 * now, better than nothing.
990 */
991 if (inp->inp_vflag & INP_TIMEWAIT)
992 error = cr_cansee(req->td->td_ucred,
993 intotw(inp)->tw_cred);
994 else
995 error = cr_canseesocket(req->td->td_ucred,
996 inp->inp_socket);
997 if (error == 0)
998 inp_list[i++] = inp;
999 }
1000 INP_UNLOCK(inp);
1001 }
1002 INP_INFO_RUNLOCK(&tcbinfo);
1003 n = i;
1004
1005 error = 0;
1006 for (i = 0; i < n; i++) {
1007 inp = inp_list[i];
1008 if (inp->inp_gencnt <= gencnt) {
1009 struct xtcpcb xt;
1010 caddr_t inp_ppcb;
1011
1012 bzero(&xt, sizeof(xt));
1013 xt.xt_len = sizeof xt;
1014 /* XXX should avoid extra copy */
1015 bcopy(inp, &xt.xt_inp, sizeof *inp);
1016 inp_ppcb = inp->inp_ppcb;
1017 if (inp_ppcb == NULL)
1018 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1019 else if (inp->inp_vflag & INP_TIMEWAIT) {
1020 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1021 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1022 } else
1023 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1024 if (inp->inp_socket != NULL)
1025 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1026 else {
1027 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1028 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1029 }
1030 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1031 error = SYSCTL_OUT(req, &xt, sizeof xt);
1032 }
1033 }
1034 if (!error) {
1035 /*
1036 * Give the user an updated idea of our state.
1037 * If the generation differs from what we told
1038 * her before, she knows that something happened
1039 * while we were processing this request, and it
1040 * might be necessary to retry.
1041 */
1042 INP_INFO_RLOCK(&tcbinfo);
1043 xig.xig_gen = tcbinfo.ipi_gencnt;
1044 xig.xig_sogen = so_gencnt;
1045 xig.xig_count = tcbinfo.ipi_count;
1046 INP_INFO_RUNLOCK(&tcbinfo);
1047 error = SYSCTL_OUT(req, &xig, sizeof xig);
1048 }
1049 free(inp_list, M_TEMP);
1050 return (error);
1051 }
1052
1053 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1054 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1055
1056 static int
1057 tcp_getcred(SYSCTL_HANDLER_ARGS)
1058 {
1059 struct xucred xuc;
1060 struct sockaddr_in addrs[2];
1061 struct inpcb *inp;
1062 int error;
1063
1064 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1065 if (error)
1066 return (error);
1067 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1068 if (error)
1069 return (error);
1070 INP_INFO_RLOCK(&tcbinfo);
1071 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1072 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1073 if (inp == NULL) {
1074 error = ENOENT;
1075 goto outunlocked;
1076 }
1077 INP_LOCK(inp);
1078 if (inp->inp_socket == NULL) {
1079 error = ENOENT;
1080 goto out;
1081 }
1082 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1083 if (error)
1084 goto out;
1085 cru2x(inp->inp_socket->so_cred, &xuc);
1086 out:
1087 INP_UNLOCK(inp);
1088 outunlocked:
1089 INP_INFO_RUNLOCK(&tcbinfo);
1090 if (error == 0)
1091 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1092 return (error);
1093 }
1094
1095 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1096 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1097 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1098
1099 #ifdef INET6
1100 static int
1101 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1102 {
1103 struct xucred xuc;
1104 struct sockaddr_in6 addrs[2];
1105 struct inpcb *inp;
1106 int error, mapped = 0;
1107
1108 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1109 if (error)
1110 return (error);
1111 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1112 if (error)
1113 return (error);
1114 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1115 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1116 return (error);
1117 }
1118 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1119 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1120 mapped = 1;
1121 else
1122 return (EINVAL);
1123 }
1124
1125 INP_INFO_RLOCK(&tcbinfo);
1126 if (mapped == 1)
1127 inp = in_pcblookup_hash(&tcbinfo,
1128 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1129 addrs[1].sin6_port,
1130 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1131 addrs[0].sin6_port,
1132 0, NULL);
1133 else
1134 inp = in6_pcblookup_hash(&tcbinfo,
1135 &addrs[1].sin6_addr, addrs[1].sin6_port,
1136 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1137 if (inp == NULL) {
1138 error = ENOENT;
1139 goto outunlocked;
1140 }
1141 INP_LOCK(inp);
1142 if (inp->inp_socket == NULL) {
1143 error = ENOENT;
1144 goto out;
1145 }
1146 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1147 if (error)
1148 goto out;
1149 cru2x(inp->inp_socket->so_cred, &xuc);
1150 out:
1151 INP_UNLOCK(inp);
1152 outunlocked:
1153 INP_INFO_RUNLOCK(&tcbinfo);
1154 if (error == 0)
1155 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1156 return (error);
1157 }
1158
1159 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1160 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1161 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1162 #endif
1163
1164
1165 void
1166 tcp_ctlinput(cmd, sa, vip)
1167 int cmd;
1168 struct sockaddr *sa;
1169 void *vip;
1170 {
1171 struct ip *ip = vip;
1172 struct tcphdr *th;
1173 struct in_addr faddr;
1174 struct inpcb *inp;
1175 struct tcpcb *tp;
1176 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1177 struct icmp *icp;
1178 struct in_conninfo inc;
1179 tcp_seq icmp_tcp_seq;
1180 int mtu;
1181
1182 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1183 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1184 return;
1185
1186 if (cmd == PRC_MSGSIZE)
1187 notify = tcp_mtudisc;
1188 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1189 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1190 notify = tcp_drop_syn_sent;
1191 /*
1192 * Redirects don't need to be handled up here.
1193 */
1194 else if (PRC_IS_REDIRECT(cmd))
1195 return;
1196 /*
1197 * Source quench is depreciated.
1198 */
1199 else if (cmd == PRC_QUENCH)
1200 return;
1201 /*
1202 * Hostdead is ugly because it goes linearly through all PCBs.
1203 * XXX: We never get this from ICMP, otherwise it makes an
1204 * excellent DoS attack on machines with many connections.
1205 */
1206 else if (cmd == PRC_HOSTDEAD)
1207 ip = NULL;
1208 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1209 return;
1210 if (ip != NULL) {
1211 icp = (struct icmp *)((caddr_t)ip
1212 - offsetof(struct icmp, icmp_ip));
1213 th = (struct tcphdr *)((caddr_t)ip
1214 + (ip->ip_hl << 2));
1215 INP_INFO_WLOCK(&tcbinfo);
1216 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1217 ip->ip_src, th->th_sport, 0, NULL);
1218 if (inp != NULL) {
1219 INP_LOCK(inp);
1220 if (inp->inp_socket != NULL) {
1221 icmp_tcp_seq = htonl(th->th_seq);
1222 tp = intotcpcb(inp);
1223 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1224 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1225 if (cmd == PRC_MSGSIZE) {
1226 /*
1227 * MTU discovery:
1228 * If we got a needfrag set the MTU
1229 * in the route to the suggested new
1230 * value (if given) and then notify.
1231 */
1232 bzero(&inc, sizeof(inc));
1233 inc.inc_flags = 0; /* IPv4 */
1234 inc.inc_faddr = faddr;
1235
1236 mtu = ntohs(icp->icmp_nextmtu);
1237 /*
1238 * If no alternative MTU was
1239 * proposed, try the next smaller
1240 * one. ip->ip_len has already
1241 * been swapped in icmp_input().
1242 */
1243 if (!mtu)
1244 mtu = ip_next_mtu(ip->ip_len,
1245 1);
1246 if (mtu < max(296, (tcp_minmss)
1247 + sizeof(struct tcpiphdr)))
1248 mtu = 0;
1249 if (!mtu)
1250 mtu = tcp_mssdflt
1251 + sizeof(struct tcpiphdr);
1252 /*
1253 * Only cache the the MTU if it
1254 * is smaller than the interface
1255 * or route MTU. tcp_mtudisc()
1256 * will do right thing by itself.
1257 */
1258 if (mtu <= tcp_maxmtu(&inc))
1259 tcp_hc_updatemtu(&inc, mtu);
1260 }
1261
1262 inp = (*notify)(inp, inetctlerrmap[cmd]);
1263 }
1264 }
1265 if (inp != NULL)
1266 INP_UNLOCK(inp);
1267 } else {
1268 inc.inc_fport = th->th_dport;
1269 inc.inc_lport = th->th_sport;
1270 inc.inc_faddr = faddr;
1271 inc.inc_laddr = ip->ip_src;
1272 #ifdef INET6
1273 inc.inc_isipv6 = 0;
1274 #endif
1275 syncache_unreach(&inc, th);
1276 }
1277 INP_INFO_WUNLOCK(&tcbinfo);
1278 } else
1279 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1280 }
1281
1282 #ifdef INET6
1283 void
1284 tcp6_ctlinput(cmd, sa, d)
1285 int cmd;
1286 struct sockaddr *sa;
1287 void *d;
1288 {
1289 struct tcphdr th;
1290 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1291 struct ip6_hdr *ip6;
1292 struct mbuf *m;
1293 struct ip6ctlparam *ip6cp = NULL;
1294 const struct sockaddr_in6 *sa6_src = NULL;
1295 int off;
1296 struct tcp_portonly {
1297 u_int16_t th_sport;
1298 u_int16_t th_dport;
1299 } *thp;
1300
1301 if (sa->sa_family != AF_INET6 ||
1302 sa->sa_len != sizeof(struct sockaddr_in6))
1303 return;
1304
1305 if (cmd == PRC_MSGSIZE)
1306 notify = tcp_mtudisc;
1307 else if (!PRC_IS_REDIRECT(cmd) &&
1308 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1309 return;
1310 /* Source quench is depreciated. */
1311 else if (cmd == PRC_QUENCH)
1312 return;
1313
1314 /* if the parameter is from icmp6, decode it. */
1315 if (d != NULL) {
1316 ip6cp = (struct ip6ctlparam *)d;
1317 m = ip6cp->ip6c_m;
1318 ip6 = ip6cp->ip6c_ip6;
1319 off = ip6cp->ip6c_off;
1320 sa6_src = ip6cp->ip6c_src;
1321 } else {
1322 m = NULL;
1323 ip6 = NULL;
1324 off = 0; /* fool gcc */
1325 sa6_src = &sa6_any;
1326 }
1327
1328 if (ip6 != NULL) {
1329 struct in_conninfo inc;
1330 /*
1331 * XXX: We assume that when IPV6 is non NULL,
1332 * M and OFF are valid.
1333 */
1334
1335 /* check if we can safely examine src and dst ports */
1336 if (m->m_pkthdr.len < off + sizeof(*thp))
1337 return;
1338
1339 bzero(&th, sizeof(th));
1340 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1341
1342 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1343 (struct sockaddr *)ip6cp->ip6c_src,
1344 th.th_sport, cmd, NULL, notify);
1345
1346 inc.inc_fport = th.th_dport;
1347 inc.inc_lport = th.th_sport;
1348 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1349 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1350 inc.inc_isipv6 = 1;
1351 INP_INFO_WLOCK(&tcbinfo);
1352 syncache_unreach(&inc, &th);
1353 INP_INFO_WUNLOCK(&tcbinfo);
1354 } else
1355 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1356 0, cmd, NULL, notify);
1357 }
1358 #endif /* INET6 */
1359
1360
1361 /*
1362 * Following is where TCP initial sequence number generation occurs.
1363 *
1364 * There are two places where we must use initial sequence numbers:
1365 * 1. In SYN-ACK packets.
1366 * 2. In SYN packets.
1367 *
1368 * All ISNs for SYN-ACK packets are generated by the syncache. See
1369 * tcp_syncache.c for details.
1370 *
1371 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1372 * depends on this property. In addition, these ISNs should be
1373 * unguessable so as to prevent connection hijacking. To satisfy
1374 * the requirements of this situation, the algorithm outlined in
1375 * RFC 1948 is used, with only small modifications.
1376 *
1377 * Implementation details:
1378 *
1379 * Time is based off the system timer, and is corrected so that it
1380 * increases by one megabyte per second. This allows for proper
1381 * recycling on high speed LANs while still leaving over an hour
1382 * before rollover.
1383 *
1384 * As reading the *exact* system time is too expensive to be done
1385 * whenever setting up a TCP connection, we increment the time
1386 * offset in two ways. First, a small random positive increment
1387 * is added to isn_offset for each connection that is set up.
1388 * Second, the function tcp_isn_tick fires once per clock tick
1389 * and increments isn_offset as necessary so that sequence numbers
1390 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1391 * random positive increments serve only to ensure that the same
1392 * exact sequence number is never sent out twice (as could otherwise
1393 * happen when a port is recycled in less than the system tick
1394 * interval.)
1395 *
1396 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1397 * between seeding of isn_secret. This is normally set to zero,
1398 * as reseeding should not be necessary.
1399 *
1400 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1401 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1402 * general, this means holding an exclusive (write) lock.
1403 */
1404
1405 #define ISN_BYTES_PER_SECOND 1048576
1406 #define ISN_STATIC_INCREMENT 4096
1407 #define ISN_RANDOM_INCREMENT (4096 - 1)
1408
1409 static u_char isn_secret[32];
1410 static int isn_last_reseed;
1411 static u_int32_t isn_offset, isn_offset_old;
1412 static MD5_CTX isn_ctx;
1413
1414 tcp_seq
1415 tcp_new_isn(tp)
1416 struct tcpcb *tp;
1417 {
1418 u_int32_t md5_buffer[4];
1419 tcp_seq new_isn;
1420
1421 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1422 INP_LOCK_ASSERT(tp->t_inpcb);
1423
1424 /* Seed if this is the first use, reseed if requested. */
1425 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1426 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1427 < (u_int)ticks))) {
1428 read_random(&isn_secret, sizeof(isn_secret));
1429 isn_last_reseed = ticks;
1430 }
1431
1432 /* Compute the md5 hash and return the ISN. */
1433 MD5Init(&isn_ctx);
1434 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1435 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1436 #ifdef INET6
1437 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1438 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1439 sizeof(struct in6_addr));
1440 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1441 sizeof(struct in6_addr));
1442 } else
1443 #endif
1444 {
1445 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1446 sizeof(struct in_addr));
1447 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1448 sizeof(struct in_addr));
1449 }
1450 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1451 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1452 new_isn = (tcp_seq) md5_buffer[0];
1453 isn_offset += ISN_STATIC_INCREMENT +
1454 (arc4random() & ISN_RANDOM_INCREMENT);
1455 new_isn += isn_offset;
1456 return (new_isn);
1457 }
1458
1459 /*
1460 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1461 * to keep time flowing at a relatively constant rate. If the random
1462 * increments have already pushed us past the projected offset, do nothing.
1463 */
1464 static void
1465 tcp_isn_tick(xtp)
1466 void *xtp;
1467 {
1468 u_int32_t projected_offset;
1469
1470 INP_INFO_WLOCK(&tcbinfo);
1471 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1472
1473 if (projected_offset > isn_offset)
1474 isn_offset = projected_offset;
1475
1476 isn_offset_old = isn_offset;
1477 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1478 INP_INFO_WUNLOCK(&tcbinfo);
1479 }
1480
1481 /*
1482 * When a specific ICMP unreachable message is received and the
1483 * connection state is SYN-SENT, drop the connection. This behavior
1484 * is controlled by the icmp_may_rst sysctl.
1485 */
1486 struct inpcb *
1487 tcp_drop_syn_sent(inp, errno)
1488 struct inpcb *inp;
1489 int errno;
1490 {
1491 struct tcpcb *tp = intotcpcb(inp);
1492
1493 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1494 INP_LOCK_ASSERT(inp);
1495
1496 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) {
1497 tcp_drop(tp, errno);
1498 return (NULL);
1499 }
1500 return (inp);
1501 }
1502
1503 /*
1504 * When `need fragmentation' ICMP is received, update our idea of the MSS
1505 * based on the new value in the route. Also nudge TCP to send something,
1506 * since we know the packet we just sent was dropped.
1507 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1508 */
1509 struct inpcb *
1510 tcp_mtudisc(inp, errno)
1511 struct inpcb *inp;
1512 int errno;
1513 {
1514 struct tcpcb *tp = intotcpcb(inp);
1515 struct socket *so = inp->inp_socket;
1516 u_int maxmtu;
1517 u_int romtu;
1518 int mss;
1519 #ifdef INET6
1520 int isipv6;
1521 #endif /* INET6 */
1522
1523 INP_LOCK_ASSERT(inp);
1524 if (tp != NULL) {
1525 #ifdef INET6
1526 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1527 #endif
1528 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1529 romtu =
1530 #ifdef INET6
1531 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1532 #endif /* INET6 */
1533 tcp_maxmtu(&inp->inp_inc);
1534 if (!maxmtu)
1535 maxmtu = romtu;
1536 else
1537 maxmtu = min(maxmtu, romtu);
1538 if (!maxmtu) {
1539 tp->t_maxopd = tp->t_maxseg =
1540 #ifdef INET6
1541 isipv6 ? tcp_v6mssdflt :
1542 #endif /* INET6 */
1543 tcp_mssdflt;
1544 return (inp);
1545 }
1546 mss = maxmtu -
1547 #ifdef INET6
1548 (isipv6 ?
1549 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1550 #endif /* INET6 */
1551 sizeof(struct tcpiphdr)
1552 #ifdef INET6
1553 )
1554 #endif /* INET6 */
1555 ;
1556
1557 /*
1558 * XXX - The above conditional probably violates the TCP
1559 * spec. The problem is that, since we don't know the
1560 * other end's MSS, we are supposed to use a conservative
1561 * default. But, if we do that, then MTU discovery will
1562 * never actually take place, because the conservative
1563 * default is much less than the MTUs typically seen
1564 * on the Internet today. For the moment, we'll sweep
1565 * this under the carpet.
1566 *
1567 * The conservative default might not actually be a problem
1568 * if the only case this occurs is when sending an initial
1569 * SYN with options and data to a host we've never talked
1570 * to before. Then, they will reply with an MSS value which
1571 * will get recorded and the new parameters should get
1572 * recomputed. For Further Study.
1573 */
1574 if (tp->t_maxopd <= mss)
1575 return (inp);
1576 tp->t_maxopd = mss;
1577
1578 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1579 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1580 mss -= TCPOLEN_TSTAMP_APPA;
1581 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1582 if (mss > MCLBYTES)
1583 mss &= ~(MCLBYTES-1);
1584 #else
1585 if (mss > MCLBYTES)
1586 mss = mss / MCLBYTES * MCLBYTES;
1587 #endif
1588 if (so->so_snd.sb_hiwat < mss)
1589 mss = so->so_snd.sb_hiwat;
1590
1591 tp->t_maxseg = mss;
1592
1593 tcpstat.tcps_mturesent++;
1594 tp->t_rtttime = 0;
1595 tp->snd_nxt = tp->snd_una;
1596 tcp_free_sackholes(tp);
1597 tp->snd_recover = tp->snd_max;
1598 if (tp->sack_enable)
1599 EXIT_FASTRECOVERY(tp);
1600 tcp_output(tp);
1601 }
1602 return (inp);
1603 }
1604
1605 /*
1606 * Look-up the routing entry to the peer of this inpcb. If no route
1607 * is found and it cannot be allocated, then return NULL. This routine
1608 * is called by TCP routines that access the rmx structure and by tcp_mss
1609 * to get the interface MTU.
1610 */
1611 u_long
1612 tcp_maxmtu(inc)
1613 struct in_conninfo *inc;
1614 {
1615 struct route sro;
1616 struct sockaddr_in *dst;
1617 struct ifnet *ifp;
1618 u_long maxmtu = 0;
1619
1620 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1621
1622 bzero(&sro, sizeof(sro));
1623 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1624 dst = (struct sockaddr_in *)&sro.ro_dst;
1625 dst->sin_family = AF_INET;
1626 dst->sin_len = sizeof(*dst);
1627 dst->sin_addr = inc->inc_faddr;
1628 rtalloc_ign(&sro, RTF_CLONING);
1629 }
1630 if (sro.ro_rt != NULL) {
1631 ifp = sro.ro_rt->rt_ifp;
1632 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1633 maxmtu = ifp->if_mtu;
1634 else
1635 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1636 RTFREE(sro.ro_rt);
1637 }
1638 return (maxmtu);
1639 }
1640
1641 #ifdef INET6
1642 u_long
1643 tcp_maxmtu6(inc)
1644 struct in_conninfo *inc;
1645 {
1646 struct route_in6 sro6;
1647 struct ifnet *ifp;
1648 u_long maxmtu = 0;
1649
1650 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1651
1652 bzero(&sro6, sizeof(sro6));
1653 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1654 sro6.ro_dst.sin6_family = AF_INET6;
1655 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1656 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1657 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1658 }
1659 if (sro6.ro_rt != NULL) {
1660 ifp = sro6.ro_rt->rt_ifp;
1661 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1662 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1663 else
1664 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1665 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1666 RTFREE(sro6.ro_rt);
1667 }
1668
1669 return (maxmtu);
1670 }
1671 #endif /* INET6 */
1672
1673 #ifdef IPSEC
1674 /* compute ESP/AH header size for TCP, including outer IP header. */
1675 size_t
1676 ipsec_hdrsiz_tcp(tp)
1677 struct tcpcb *tp;
1678 {
1679 struct inpcb *inp;
1680 struct mbuf *m;
1681 size_t hdrsiz;
1682 struct ip *ip;
1683 #ifdef INET6
1684 struct ip6_hdr *ip6;
1685 #endif
1686 struct tcphdr *th;
1687
1688 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1689 return (0);
1690 MGETHDR(m, M_DONTWAIT, MT_DATA);
1691 if (!m)
1692 return (0);
1693
1694 #ifdef INET6
1695 if ((inp->inp_vflag & INP_IPV6) != 0) {
1696 ip6 = mtod(m, struct ip6_hdr *);
1697 th = (struct tcphdr *)(ip6 + 1);
1698 m->m_pkthdr.len = m->m_len =
1699 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1700 tcpip_fillheaders(inp, ip6, th);
1701 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1702 } else
1703 #endif /* INET6 */
1704 {
1705 ip = mtod(m, struct ip *);
1706 th = (struct tcphdr *)(ip + 1);
1707 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1708 tcpip_fillheaders(inp, ip, th);
1709 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1710 }
1711
1712 m_free(m);
1713 return (hdrsiz);
1714 }
1715 #endif /*IPSEC*/
1716
1717 /*
1718 * Move a TCP connection into TIME_WAIT state.
1719 * tcbinfo is locked.
1720 * inp is locked, and is unlocked before returning.
1721 */
1722 void
1723 tcp_twstart(tp)
1724 struct tcpcb *tp;
1725 {
1726 struct tcptw *tw;
1727 struct inpcb *inp = tp->t_inpcb;
1728 int acknow;
1729 struct socket *so;
1730
1731 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1732 INP_LOCK_ASSERT(inp);
1733
1734 if (nolocaltimewait && in_localip(inp->inp_faddr)) {
1735 tcp_close(tp);
1736 return;
1737 }
1738
1739 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1740 if (tw == NULL) {
1741 tw = tcp_timer_2msl_tw(1);
1742 if (tw == NULL) {
1743 tcp_close(tp);
1744 return;
1745 }
1746 }
1747 tw->tw_inpcb = inp;
1748
1749 /*
1750 * Recover last window size sent.
1751 */
1752 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1753
1754 /*
1755 * Set t_recent if timestamps are used on the connection.
1756 */
1757 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1758 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1759 tw->t_recent = tp->ts_recent;
1760 else
1761 tw->t_recent = 0;
1762
1763 tw->snd_nxt = tp->snd_nxt;
1764 tw->rcv_nxt = tp->rcv_nxt;
1765 tw->iss = tp->iss;
1766 tw->irs = tp->irs;
1767 tw->t_starttime = tp->t_starttime;
1768 tw->tw_time = 0;
1769
1770 /* XXX
1771 * If this code will
1772 * be used for fin-wait-2 state also, then we may need
1773 * a ts_recent from the last segment.
1774 */
1775 acknow = tp->t_flags & TF_ACKNOW;
1776 tcp_discardcb(tp);
1777 so = inp->inp_socket;
1778 ACCEPT_LOCK();
1779 SOCK_LOCK(so);
1780 so->so_pcb = NULL;
1781 tw->tw_cred = crhold(so->so_cred);
1782 tw->tw_so_options = so->so_options;
1783 sotryfree(so);
1784 inp->inp_socket = NULL;
1785 if (acknow)
1786 tcp_twrespond(tw, TH_ACK);
1787 inp->inp_ppcb = (caddr_t)tw;
1788 inp->inp_vflag |= INP_TIMEWAIT;
1789 tcp_timer_2msl_reset(tw, 0);
1790 INP_UNLOCK(inp);
1791 }
1792
1793 #if 0
1794 /*
1795 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1796 * the actual rate is slightly higher due to the addition of
1797 * random positive increments.
1798 *
1799 * Most other new OSes use semi-randomized ISN values, so we
1800 * do not need to worry about them.
1801 */
1802 #define MS_ISN_BYTES_PER_SECOND 250000
1803
1804 /*
1805 * Determine if the ISN we will generate has advanced beyond the last
1806 * sequence number used by the previous connection. If so, indicate
1807 * that it is safe to recycle this tw socket by returning 1.
1808 */
1809 int
1810 tcp_twrecycleable(struct tcptw *tw)
1811 {
1812 tcp_seq new_iss = tw->iss;
1813 tcp_seq new_irs = tw->irs;
1814
1815 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1816 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1817 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1818
1819 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1820 return (1);
1821 else
1822 return (0);
1823 }
1824 #endif
1825
1826 void
1827 tcp_twclose(struct tcptw *tw, int reuse)
1828 {
1829 struct inpcb *inp;
1830
1831 inp = tw->tw_inpcb;
1832 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1833 INP_LOCK_ASSERT(inp);
1834
1835 tw->tw_inpcb = NULL;
1836 tcp_timer_2msl_stop(tw);
1837 inp->inp_ppcb = NULL;
1838 #ifdef INET6
1839 if (inp->inp_vflag & INP_IPV6PROTO)
1840 in6_pcbdetach(inp);
1841 else
1842 #endif
1843 in_pcbdetach(inp);
1844 tcpstat.tcps_closed++;
1845 crfree(tw->tw_cred);
1846 tw->tw_cred = NULL;
1847 if (reuse)
1848 return;
1849 uma_zfree(tcptw_zone, tw);
1850 }
1851
1852 int
1853 tcp_twrespond(struct tcptw *tw, int flags)
1854 {
1855 struct inpcb *inp = tw->tw_inpcb;
1856 struct tcphdr *th;
1857 struct mbuf *m;
1858 struct ip *ip = NULL;
1859 u_int8_t *optp;
1860 u_int hdrlen, optlen;
1861 int error;
1862 #ifdef INET6
1863 struct ip6_hdr *ip6 = NULL;
1864 int isipv6 = inp->inp_inc.inc_isipv6;
1865 #endif
1866
1867 INP_LOCK_ASSERT(inp);
1868
1869 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1870 if (m == NULL)
1871 return (ENOBUFS);
1872 m->m_data += max_linkhdr;
1873
1874 #ifdef MAC
1875 mac_create_mbuf_from_inpcb(inp, m);
1876 #endif
1877
1878 #ifdef INET6
1879 if (isipv6) {
1880 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1881 ip6 = mtod(m, struct ip6_hdr *);
1882 th = (struct tcphdr *)(ip6 + 1);
1883 tcpip_fillheaders(inp, ip6, th);
1884 } else
1885 #endif
1886 {
1887 hdrlen = sizeof(struct tcpiphdr);
1888 ip = mtod(m, struct ip *);
1889 th = (struct tcphdr *)(ip + 1);
1890 tcpip_fillheaders(inp, ip, th);
1891 }
1892 optp = (u_int8_t *)(th + 1);
1893
1894 /*
1895 * Send a timestamp and echo-reply if both our side and our peer
1896 * have sent timestamps in our SYN's and this is not a RST.
1897 */
1898 if (tw->t_recent && flags == TH_ACK) {
1899 u_int32_t *lp = (u_int32_t *)optp;
1900
1901 /* Form timestamp option as shown in appendix A of RFC 1323. */
1902 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1903 *lp++ = htonl(ticks);
1904 *lp = htonl(tw->t_recent);
1905 optp += TCPOLEN_TSTAMP_APPA;
1906 }
1907
1908 optlen = optp - (u_int8_t *)(th + 1);
1909
1910 m->m_len = hdrlen + optlen;
1911 m->m_pkthdr.len = m->m_len;
1912
1913 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1914
1915 th->th_seq = htonl(tw->snd_nxt);
1916 th->th_ack = htonl(tw->rcv_nxt);
1917 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1918 th->th_flags = flags;
1919 th->th_win = htons(tw->last_win);
1920
1921 #ifdef INET6
1922 if (isipv6) {
1923 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1924 sizeof(struct tcphdr) + optlen);
1925 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1926 error = ip6_output(m, inp->in6p_outputopts, NULL,
1927 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1928 } else
1929 #endif
1930 {
1931 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1932 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1933 m->m_pkthdr.csum_flags = CSUM_TCP;
1934 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1935 ip->ip_len = m->m_pkthdr.len;
1936 if (path_mtu_discovery)
1937 ip->ip_off |= IP_DF;
1938 error = ip_output(m, inp->inp_options, NULL,
1939 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1940 NULL, inp);
1941 }
1942 if (flags & TH_ACK)
1943 tcpstat.tcps_sndacks++;
1944 else
1945 tcpstat.tcps_sndctrl++;
1946 tcpstat.tcps_sndtotal++;
1947 return (error);
1948 }
1949
1950 /*
1951 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1952 *
1953 * This code attempts to calculate the bandwidth-delay product as a
1954 * means of determining the optimal window size to maximize bandwidth,
1955 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1956 * routers. This code also does a fairly good job keeping RTTs in check
1957 * across slow links like modems. We implement an algorithm which is very
1958 * similar (but not meant to be) TCP/Vegas. The code operates on the
1959 * transmitter side of a TCP connection and so only effects the transmit
1960 * side of the connection.
1961 *
1962 * BACKGROUND: TCP makes no provision for the management of buffer space
1963 * at the end points or at the intermediate routers and switches. A TCP
1964 * stream, whether using NewReno or not, will eventually buffer as
1965 * many packets as it is able and the only reason this typically works is
1966 * due to the fairly small default buffers made available for a connection
1967 * (typicaly 16K or 32K). As machines use larger windows and/or window
1968 * scaling it is now fairly easy for even a single TCP connection to blow-out
1969 * all available buffer space not only on the local interface, but on
1970 * intermediate routers and switches as well. NewReno makes a misguided
1971 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1972 * then backing off, then steadily increasing the window again until another
1973 * failure occurs, ad-infinitum. This results in terrible oscillation that
1974 * is only made worse as network loads increase and the idea of intentionally
1975 * blowing out network buffers is, frankly, a terrible way to manage network
1976 * resources.
1977 *
1978 * It is far better to limit the transmit window prior to the failure
1979 * condition being achieved. There are two general ways to do this: First
1980 * you can 'scan' through different transmit window sizes and locate the
1981 * point where the RTT stops increasing, indicating that you have filled the
1982 * pipe, then scan backwards until you note that RTT stops decreasing, then
1983 * repeat ad-infinitum. This method works in principle but has severe
1984 * implementation issues due to RTT variances, timer granularity, and
1985 * instability in the algorithm which can lead to many false positives and
1986 * create oscillations as well as interact badly with other TCP streams
1987 * implementing the same algorithm.
1988 *
1989 * The second method is to limit the window to the bandwidth delay product
1990 * of the link. This is the method we implement. RTT variances and our
1991 * own manipulation of the congestion window, bwnd, can potentially
1992 * destabilize the algorithm. For this reason we have to stabilize the
1993 * elements used to calculate the window. We do this by using the minimum
1994 * observed RTT, the long term average of the observed bandwidth, and
1995 * by adding two segments worth of slop. It isn't perfect but it is able
1996 * to react to changing conditions and gives us a very stable basis on
1997 * which to extend the algorithm.
1998 */
1999 void
2000 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2001 {
2002 u_long bw;
2003 u_long bwnd;
2004 int save_ticks;
2005
2006 INP_LOCK_ASSERT(tp->t_inpcb);
2007
2008 /*
2009 * If inflight_enable is disabled in the middle of a tcp connection,
2010 * make sure snd_bwnd is effectively disabled.
2011 */
2012 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2013 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2014 tp->snd_bandwidth = 0;
2015 return;
2016 }
2017
2018 /*
2019 * Figure out the bandwidth. Due to the tick granularity this
2020 * is a very rough number and it MUST be averaged over a fairly
2021 * long period of time. XXX we need to take into account a link
2022 * that is not using all available bandwidth, but for now our
2023 * slop will ramp us up if this case occurs and the bandwidth later
2024 * increases.
2025 *
2026 * Note: if ticks rollover 'bw' may wind up negative. We must
2027 * effectively reset t_bw_rtttime for this case.
2028 */
2029 save_ticks = ticks;
2030 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2031 return;
2032
2033 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2034 (save_ticks - tp->t_bw_rtttime);
2035 tp->t_bw_rtttime = save_ticks;
2036 tp->t_bw_rtseq = ack_seq;
2037 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2038 return;
2039 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2040
2041 tp->snd_bandwidth = bw;
2042
2043 /*
2044 * Calculate the semi-static bandwidth delay product, plus two maximal
2045 * segments. The additional slop puts us squarely in the sweet
2046 * spot and also handles the bandwidth run-up case and stabilization.
2047 * Without the slop we could be locking ourselves into a lower
2048 * bandwidth.
2049 *
2050 * Situations Handled:
2051 * (1) Prevents over-queueing of packets on LANs, especially on
2052 * high speed LANs, allowing larger TCP buffers to be
2053 * specified, and also does a good job preventing
2054 * over-queueing of packets over choke points like modems
2055 * (at least for the transmit side).
2056 *
2057 * (2) Is able to handle changing network loads (bandwidth
2058 * drops so bwnd drops, bandwidth increases so bwnd
2059 * increases).
2060 *
2061 * (3) Theoretically should stabilize in the face of multiple
2062 * connections implementing the same algorithm (this may need
2063 * a little work).
2064 *
2065 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2066 * be adjusted with a sysctl but typically only needs to be
2067 * on very slow connections. A value no smaller then 5
2068 * should be used, but only reduce this default if you have
2069 * no other choice.
2070 */
2071 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2072 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2073 #undef USERTT
2074
2075 if (tcp_inflight_debug > 0) {
2076 static int ltime;
2077 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2078 ltime = ticks;
2079 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2080 tp,
2081 bw,
2082 tp->t_rttbest,
2083 tp->t_srtt,
2084 bwnd
2085 );
2086 }
2087 }
2088 if ((long)bwnd < tcp_inflight_min)
2089 bwnd = tcp_inflight_min;
2090 if (bwnd > tcp_inflight_max)
2091 bwnd = tcp_inflight_max;
2092 if ((long)bwnd < tp->t_maxseg * 2)
2093 bwnd = tp->t_maxseg * 2;
2094 tp->snd_bwnd = bwnd;
2095 }
2096
2097 #ifdef TCP_SIGNATURE
2098 /*
2099 * Callback function invoked by m_apply() to digest TCP segment data
2100 * contained within an mbuf chain.
2101 */
2102 static int
2103 tcp_signature_apply(void *fstate, void *data, u_int len)
2104 {
2105
2106 MD5Update(fstate, (u_char *)data, len);
2107 return (0);
2108 }
2109
2110 /*
2111 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2112 *
2113 * Parameters:
2114 * m pointer to head of mbuf chain
2115 * off0 offset to TCP header within the mbuf chain
2116 * len length of TCP segment data, excluding options
2117 * optlen length of TCP segment options
2118 * buf pointer to storage for computed MD5 digest
2119 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2120 *
2121 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2122 * When called from tcp_input(), we can be sure that th_sum has been
2123 * zeroed out and verified already.
2124 *
2125 * This function is for IPv4 use only. Calling this function with an
2126 * IPv6 packet in the mbuf chain will yield undefined results.
2127 *
2128 * Return 0 if successful, otherwise return -1.
2129 *
2130 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2131 * search with the destination IP address, and a 'magic SPI' to be
2132 * determined by the application. This is hardcoded elsewhere to 1179
2133 * right now. Another branch of this code exists which uses the SPD to
2134 * specify per-application flows but it is unstable.
2135 */
2136 int
2137 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2138 u_char *buf, u_int direction)
2139 {
2140 union sockaddr_union dst;
2141 struct ippseudo ippseudo;
2142 MD5_CTX ctx;
2143 int doff;
2144 struct ip *ip;
2145 struct ipovly *ipovly;
2146 struct secasvar *sav;
2147 struct tcphdr *th;
2148 u_short savecsum;
2149
2150 KASSERT(m != NULL, ("NULL mbuf chain"));
2151 KASSERT(buf != NULL, ("NULL signature pointer"));
2152
2153 /* Extract the destination from the IP header in the mbuf. */
2154 ip = mtod(m, struct ip *);
2155 bzero(&dst, sizeof(union sockaddr_union));
2156 dst.sa.sa_len = sizeof(struct sockaddr_in);
2157 dst.sa.sa_family = AF_INET;
2158 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2159 ip->ip_src : ip->ip_dst;
2160
2161 /* Look up an SADB entry which matches the address of the peer. */
2162 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2163 if (sav == NULL) {
2164 printf("%s: SADB lookup failed for %s\n", __func__,
2165 inet_ntoa(dst.sin.sin_addr));
2166 return (EINVAL);
2167 }
2168
2169 MD5Init(&ctx);
2170 ipovly = (struct ipovly *)ip;
2171 th = (struct tcphdr *)((u_char *)ip + off0);
2172 doff = off0 + sizeof(struct tcphdr) + optlen;
2173
2174 /*
2175 * Step 1: Update MD5 hash with IP pseudo-header.
2176 *
2177 * XXX The ippseudo header MUST be digested in network byte order,
2178 * or else we'll fail the regression test. Assume all fields we've
2179 * been doing arithmetic on have been in host byte order.
2180 * XXX One cannot depend on ipovly->ih_len here. When called from
2181 * tcp_output(), the underlying ip_len member has not yet been set.
2182 */
2183 ippseudo.ippseudo_src = ipovly->ih_src;
2184 ippseudo.ippseudo_dst = ipovly->ih_dst;
2185 ippseudo.ippseudo_pad = 0;
2186 ippseudo.ippseudo_p = IPPROTO_TCP;
2187 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2188 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2189
2190 /*
2191 * Step 2: Update MD5 hash with TCP header, excluding options.
2192 * The TCP checksum must be set to zero.
2193 */
2194 savecsum = th->th_sum;
2195 th->th_sum = 0;
2196 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2197 th->th_sum = savecsum;
2198
2199 /*
2200 * Step 3: Update MD5 hash with TCP segment data.
2201 * Use m_apply() to avoid an early m_pullup().
2202 */
2203 if (len > 0)
2204 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2205
2206 /*
2207 * Step 4: Update MD5 hash with shared secret.
2208 */
2209 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2210 MD5Final(buf, &ctx);
2211
2212 key_sa_recordxfer(sav, m);
2213 KEY_FREESAV(&sav);
2214 return (0);
2215 }
2216 #endif /* TCP_SIGNATURE */
2217
2218 static int
2219 sysctl_drop(SYSCTL_HANDLER_ARGS)
2220 {
2221 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2222 struct sockaddr_storage addrs[2];
2223 struct inpcb *inp;
2224 struct tcpcb *tp;
2225 struct tcptw *tw;
2226 struct sockaddr_in *fin, *lin;
2227 #ifdef INET6
2228 struct sockaddr_in6 *fin6, *lin6;
2229 struct in6_addr f6, l6;
2230 #endif
2231 int error;
2232
2233 inp = NULL;
2234 fin = lin = NULL;
2235 #ifdef INET6
2236 fin6 = lin6 = NULL;
2237 #endif
2238 error = 0;
2239
2240 if (req->oldptr != NULL || req->oldlen != 0)
2241 return (EINVAL);
2242 if (req->newptr == NULL)
2243 return (EPERM);
2244 if (req->newlen < sizeof(addrs))
2245 return (ENOMEM);
2246 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2247 if (error)
2248 return (error);
2249
2250 switch (addrs[0].ss_family) {
2251 #ifdef INET6
2252 case AF_INET6:
2253 fin6 = (struct sockaddr_in6 *)&addrs[0];
2254 lin6 = (struct sockaddr_in6 *)&addrs[1];
2255 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2256 lin6->sin6_len != sizeof(struct sockaddr_in6))
2257 return (EINVAL);
2258 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2259 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2260 return (EINVAL);
2261 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2262 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2263 fin = (struct sockaddr_in *)&addrs[0];
2264 lin = (struct sockaddr_in *)&addrs[1];
2265 break;
2266 }
2267 error = sa6_embedscope(fin6, ip6_use_defzone);
2268 if (error)
2269 return (error);
2270 error = sa6_embedscope(lin6, ip6_use_defzone);
2271 if (error)
2272 return (error);
2273 break;
2274 #endif
2275 case AF_INET:
2276 fin = (struct sockaddr_in *)&addrs[0];
2277 lin = (struct sockaddr_in *)&addrs[1];
2278 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2279 lin->sin_len != sizeof(struct sockaddr_in))
2280 return (EINVAL);
2281 break;
2282 default:
2283 return (EINVAL);
2284 }
2285 INP_INFO_WLOCK(&tcbinfo);
2286 switch (addrs[0].ss_family) {
2287 #ifdef INET6
2288 case AF_INET6:
2289 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2290 &l6, lin6->sin6_port, 0, NULL);
2291 break;
2292 #endif
2293 case AF_INET:
2294 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2295 lin->sin_addr, lin->sin_port, 0, NULL);
2296 break;
2297 }
2298 if (inp != NULL) {
2299 INP_LOCK(inp);
2300 if ((tw = intotw(inp)) &&
2301 (inp->inp_vflag & INP_TIMEWAIT) != 0) {
2302 (void) tcp_twclose(tw, 0);
2303 } else if ((tp = intotcpcb(inp)) &&
2304 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
2305 tp = tcp_drop(tp, ECONNABORTED);
2306 if (tp != NULL)
2307 INP_UNLOCK(inp);
2308 } else
2309 INP_UNLOCK(inp);
2310 } else
2311 error = ESRCH;
2312 INP_INFO_WUNLOCK(&tcbinfo);
2313 return (error);
2314 }
2315
2316 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2317 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2318 0, sysctl_drop, "", "Drop TCP connection");
Cache object: 26b4d90a75a29148d2d544e3238a2fb7
|