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/6.0/sys/netinet/tcp_subr.c 151169 2005-10-09 08:21:34Z maxim $
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/nd6.h>
78 #endif
79 #include <netinet/ip_icmp.h>
80 #include <netinet/tcp.h>
81 #include <netinet/tcp_fsm.h>
82 #include <netinet/tcp_seq.h>
83 #include <netinet/tcp_timer.h>
84 #include <netinet/tcp_var.h>
85 #ifdef INET6
86 #include <netinet6/tcp6_var.h>
87 #endif
88 #include <netinet/tcpip.h>
89 #ifdef TCPDEBUG
90 #include <netinet/tcp_debug.h>
91 #endif
92 #include <netinet6/ip6protosw.h>
93
94 #ifdef IPSEC
95 #include <netinet6/ipsec.h>
96 #ifdef INET6
97 #include <netinet6/ipsec6.h>
98 #endif
99 #include <netkey/key.h>
100 #endif /*IPSEC*/
101
102 #ifdef FAST_IPSEC
103 #include <netipsec/ipsec.h>
104 #include <netipsec/xform.h>
105 #ifdef INET6
106 #include <netipsec/ipsec6.h>
107 #endif
108 #include <netipsec/key.h>
109 #define IPSEC
110 #endif /*FAST_IPSEC*/
111
112 #include <machine/in_cksum.h>
113 #include <sys/md5.h>
114
115 int tcp_mssdflt = TCP_MSS;
116 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
117 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
118
119 #ifdef INET6
120 int tcp_v6mssdflt = TCP6_MSS;
121 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
122 CTLFLAG_RW, &tcp_v6mssdflt , 0,
123 "Default TCP Maximum Segment Size for IPv6");
124 #endif
125
126 /*
127 * Minimum MSS we accept and use. This prevents DoS attacks where
128 * we are forced to a ridiculous low MSS like 20 and send hundreds
129 * of packets instead of one. The effect scales with the available
130 * bandwidth and quickly saturates the CPU and network interface
131 * with packet generation and sending. Set to zero to disable MINMSS
132 * checking. This setting prevents us from sending too small packets.
133 */
134 int tcp_minmss = TCP_MINMSS;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
136 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
137 /*
138 * Number of TCP segments per second we accept from remote host
139 * before we start to calculate average segment size. If average
140 * segment size drops below the minimum TCP MSS we assume a DoS
141 * attack and reset+drop the connection. Care has to be taken not to
142 * set this value too small to not kill interactive type connections
143 * (telnet, SSH) which send many small packets.
144 */
145 int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
147 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
148 "be under the MINMSS Size");
149
150 #if 0
151 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
152 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
153 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
154 #endif
155
156 int tcp_do_rfc1323 = 1;
157 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
158 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
159
160 static int tcp_tcbhashsize = 0;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
162 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
163
164 static int do_tcpdrain = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
166 "Enable tcp_drain routine for extra help when low on mbufs");
167
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
169 &tcbinfo.ipi_count, 0, "Number of active PCBs");
170
171 static int icmp_may_rst = 1;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
173 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
174
175 static int tcp_isn_reseed_interval = 0;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
177 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
178
179 /*
180 * TCP bandwidth limiting sysctls. Note that the default lower bound of
181 * 1024 exists only for debugging. A good production default would be
182 * something like 6100.
183 */
184 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
185 "TCP inflight data limiting");
186
187 static int tcp_inflight_enable = 1;
188 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
189 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
190
191 static int tcp_inflight_debug = 0;
192 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
193 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
194
195 static int tcp_inflight_min = 6144;
196 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
197 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
198
199 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
200 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
201 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
202
203 static int tcp_inflight_stab = 20;
204 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
205 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
206
207 uma_zone_t sack_hole_zone;
208
209 static struct inpcb *tcp_notify(struct inpcb *, int);
210 static void tcp_discardcb(struct tcpcb *);
211 static void tcp_isn_tick(void *);
212
213 /*
214 * Target size of TCP PCB hash tables. Must be a power of two.
215 *
216 * Note that this can be overridden by the kernel environment
217 * variable net.inet.tcp.tcbhashsize
218 */
219 #ifndef TCBHASHSIZE
220 #define TCBHASHSIZE 512
221 #endif
222
223 /*
224 * XXX
225 * Callouts should be moved into struct tcp directly. They are currently
226 * separate because the tcpcb structure is exported to userland for sysctl
227 * parsing purposes, which do not know about callouts.
228 */
229 struct tcpcb_mem {
230 struct tcpcb tcb;
231 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
232 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
233 };
234
235 static uma_zone_t tcpcb_zone;
236 static uma_zone_t tcptw_zone;
237 struct callout isn_callout;
238
239 /*
240 * Tcp initialization
241 */
242 void
243 tcp_init()
244 {
245 int hashsize = TCBHASHSIZE;
246
247 tcp_delacktime = TCPTV_DELACK;
248 tcp_keepinit = TCPTV_KEEP_INIT;
249 tcp_keepidle = TCPTV_KEEP_IDLE;
250 tcp_keepintvl = TCPTV_KEEPINTVL;
251 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
252 tcp_msl = TCPTV_MSL;
253 tcp_rexmit_min = TCPTV_MIN;
254 tcp_rexmit_slop = TCPTV_CPU_VAR;
255
256 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
257 LIST_INIT(&tcb);
258 tcbinfo.listhead = &tcb;
259 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
260 if (!powerof2(hashsize)) {
261 printf("WARNING: TCB hash size not a power of 2\n");
262 hashsize = 512; /* safe default */
263 }
264 tcp_tcbhashsize = hashsize;
265 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
266 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
267 &tcbinfo.porthashmask);
268 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
269 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
270 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
271 #ifdef INET6
272 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
273 #else /* INET6 */
274 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
275 #endif /* INET6 */
276 if (max_protohdr < TCP_MINPROTOHDR)
277 max_protohdr = TCP_MINPROTOHDR;
278 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
279 panic("tcp_init");
280 #undef TCP_MINPROTOHDR
281 /*
282 * These have to be type stable for the benefit of the timers.
283 */
284 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
285 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
286 uma_zone_set_max(tcpcb_zone, maxsockets);
287 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
288 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
289 uma_zone_set_max(tcptw_zone, maxsockets / 5);
290 tcp_timer_init();
291 syncache_init();
292 tcp_hc_init();
293 tcp_reass_init();
294 callout_init(&isn_callout, CALLOUT_MPSAFE);
295 tcp_isn_tick(NULL);
296 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
297 SHUTDOWN_PRI_DEFAULT);
298 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
299 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
300 }
301
302 void
303 tcp_fini(xtp)
304 void *xtp;
305 {
306 callout_stop(&isn_callout);
307
308 }
309
310 /*
311 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
312 * tcp_template used to store this data in mbufs, but we now recopy it out
313 * of the tcpcb each time to conserve mbufs.
314 */
315 void
316 tcpip_fillheaders(inp, ip_ptr, tcp_ptr)
317 struct inpcb *inp;
318 void *ip_ptr;
319 void *tcp_ptr;
320 {
321 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
322
323 INP_LOCK_ASSERT(inp);
324
325 #ifdef INET6
326 if ((inp->inp_vflag & INP_IPV6) != 0) {
327 struct ip6_hdr *ip6;
328
329 ip6 = (struct ip6_hdr *)ip_ptr;
330 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
331 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
332 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
333 (IPV6_VERSION & IPV6_VERSION_MASK);
334 ip6->ip6_nxt = IPPROTO_TCP;
335 ip6->ip6_plen = sizeof(struct tcphdr);
336 ip6->ip6_src = inp->in6p_laddr;
337 ip6->ip6_dst = inp->in6p_faddr;
338 } else
339 #endif
340 {
341 struct ip *ip;
342
343 ip = (struct ip *)ip_ptr;
344 ip->ip_v = IPVERSION;
345 ip->ip_hl = 5;
346 ip->ip_tos = inp->inp_ip_tos;
347 ip->ip_len = 0;
348 ip->ip_id = 0;
349 ip->ip_off = 0;
350 ip->ip_ttl = inp->inp_ip_ttl;
351 ip->ip_sum = 0;
352 ip->ip_p = IPPROTO_TCP;
353 ip->ip_src = inp->inp_laddr;
354 ip->ip_dst = inp->inp_faddr;
355 }
356 th->th_sport = inp->inp_lport;
357 th->th_dport = inp->inp_fport;
358 th->th_seq = 0;
359 th->th_ack = 0;
360 th->th_x2 = 0;
361 th->th_off = 5;
362 th->th_flags = 0;
363 th->th_win = 0;
364 th->th_urp = 0;
365 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
366 }
367
368 /*
369 * Create template to be used to send tcp packets on a connection.
370 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
371 * use for this function is in keepalives, which use tcp_respond.
372 */
373 struct tcptemp *
374 tcpip_maketemplate(inp)
375 struct inpcb *inp;
376 {
377 struct mbuf *m;
378 struct tcptemp *n;
379
380 m = m_get(M_DONTWAIT, MT_DATA);
381 if (m == NULL)
382 return (0);
383 m->m_len = sizeof(struct tcptemp);
384 n = mtod(m, struct tcptemp *);
385
386 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
387 return (n);
388 }
389
390 /*
391 * Send a single message to the TCP at address specified by
392 * the given TCP/IP header. If m == NULL, then we make a copy
393 * of the tcpiphdr at ti and send directly to the addressed host.
394 * This is used to force keep alive messages out using the TCP
395 * template for a connection. If flags are given then we send
396 * a message back to the TCP which originated the * segment ti,
397 * and discard the mbuf containing it and any other attached mbufs.
398 *
399 * In any case the ack and sequence number of the transmitted
400 * segment are as specified by the parameters.
401 *
402 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
403 */
404 void
405 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
406 struct tcpcb *tp;
407 void *ipgen;
408 register struct tcphdr *th;
409 register struct mbuf *m;
410 tcp_seq ack, seq;
411 int flags;
412 {
413 register int tlen;
414 int win = 0;
415 struct ip *ip;
416 struct tcphdr *nth;
417 #ifdef INET6
418 struct ip6_hdr *ip6;
419 int isipv6;
420 #endif /* INET6 */
421 int ipflags = 0;
422 struct inpcb *inp;
423
424 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
425
426 #ifdef INET6
427 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
428 ip6 = ipgen;
429 #endif /* INET6 */
430 ip = ipgen;
431
432 if (tp != NULL) {
433 inp = tp->t_inpcb;
434 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
435 INP_INFO_WLOCK_ASSERT(&tcbinfo);
436 INP_LOCK_ASSERT(inp);
437 } else
438 inp = NULL;
439
440 if (tp != NULL) {
441 if (!(flags & TH_RST)) {
442 win = sbspace(&inp->inp_socket->so_rcv);
443 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
444 win = (long)TCP_MAXWIN << tp->rcv_scale;
445 }
446 }
447 if (m == NULL) {
448 m = m_gethdr(M_DONTWAIT, MT_HEADER);
449 if (m == NULL)
450 return;
451 tlen = 0;
452 m->m_data += max_linkhdr;
453 #ifdef INET6
454 if (isipv6) {
455 bcopy((caddr_t)ip6, mtod(m, caddr_t),
456 sizeof(struct ip6_hdr));
457 ip6 = mtod(m, struct ip6_hdr *);
458 nth = (struct tcphdr *)(ip6 + 1);
459 } else
460 #endif /* INET6 */
461 {
462 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
463 ip = mtod(m, struct ip *);
464 nth = (struct tcphdr *)(ip + 1);
465 }
466 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
467 flags = TH_ACK;
468 } else {
469 m_freem(m->m_next);
470 m->m_next = NULL;
471 m->m_data = (caddr_t)ipgen;
472 /* m_len is set later */
473 tlen = 0;
474 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
475 #ifdef INET6
476 if (isipv6) {
477 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
478 nth = (struct tcphdr *)(ip6 + 1);
479 } else
480 #endif /* INET6 */
481 {
482 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
483 nth = (struct tcphdr *)(ip + 1);
484 }
485 if (th != nth) {
486 /*
487 * this is usually a case when an extension header
488 * exists between the IPv6 header and the
489 * TCP header.
490 */
491 nth->th_sport = th->th_sport;
492 nth->th_dport = th->th_dport;
493 }
494 xchg(nth->th_dport, nth->th_sport, n_short);
495 #undef xchg
496 }
497 #ifdef INET6
498 if (isipv6) {
499 ip6->ip6_flow = 0;
500 ip6->ip6_vfc = IPV6_VERSION;
501 ip6->ip6_nxt = IPPROTO_TCP;
502 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
503 tlen));
504 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
505 } else
506 #endif
507 {
508 tlen += sizeof (struct tcpiphdr);
509 ip->ip_len = tlen;
510 ip->ip_ttl = ip_defttl;
511 if (path_mtu_discovery)
512 ip->ip_off |= IP_DF;
513 }
514 m->m_len = tlen;
515 m->m_pkthdr.len = tlen;
516 m->m_pkthdr.rcvif = NULL;
517 #ifdef MAC
518 if (inp != NULL) {
519 /*
520 * Packet is associated with a socket, so allow the
521 * label of the response to reflect the socket label.
522 */
523 INP_LOCK_ASSERT(inp);
524 mac_create_mbuf_from_inpcb(inp, m);
525 } else {
526 /*
527 * Packet is not associated with a socket, so possibly
528 * update the label in place.
529 */
530 mac_reflect_mbuf_tcp(m);
531 }
532 #endif
533 nth->th_seq = htonl(seq);
534 nth->th_ack = htonl(ack);
535 nth->th_x2 = 0;
536 nth->th_off = sizeof (struct tcphdr) >> 2;
537 nth->th_flags = flags;
538 if (tp != NULL)
539 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
540 else
541 nth->th_win = htons((u_short)win);
542 nth->th_urp = 0;
543 #ifdef INET6
544 if (isipv6) {
545 nth->th_sum = 0;
546 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
547 sizeof(struct ip6_hdr),
548 tlen - sizeof(struct ip6_hdr));
549 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
550 NULL, NULL);
551 } else
552 #endif /* INET6 */
553 {
554 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
555 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
556 m->m_pkthdr.csum_flags = CSUM_TCP;
557 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
558 }
559 #ifdef TCPDEBUG
560 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
561 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
562 #endif
563 #ifdef INET6
564 if (isipv6)
565 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
566 else
567 #endif /* INET6 */
568 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
569 }
570
571 /*
572 * Create a new TCP control block, making an
573 * empty reassembly queue and hooking it to the argument
574 * protocol control block. The `inp' parameter must have
575 * come from the zone allocator set up in tcp_init().
576 */
577 struct tcpcb *
578 tcp_newtcpcb(inp)
579 struct inpcb *inp;
580 {
581 struct tcpcb_mem *tm;
582 struct tcpcb *tp;
583 #ifdef INET6
584 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
585 #endif /* INET6 */
586
587 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
588 if (tm == NULL)
589 return (NULL);
590 tp = &tm->tcb;
591 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
592 tp->t_maxseg = tp->t_maxopd =
593 #ifdef INET6
594 isipv6 ? tcp_v6mssdflt :
595 #endif /* INET6 */
596 tcp_mssdflt;
597
598 /* Set up our timeouts. */
599 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
600 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
601 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
602 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
603 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
604
605 if (tcp_do_rfc1323)
606 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
607 tp->sack_enable = tcp_do_sack;
608 TAILQ_INIT(&tp->snd_holes);
609 tp->t_inpcb = inp; /* XXX */
610 /*
611 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
612 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
613 * reasonable initial retransmit time.
614 */
615 tp->t_srtt = TCPTV_SRTTBASE;
616 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
617 tp->t_rttmin = tcp_rexmit_min;
618 tp->t_rxtcur = TCPTV_RTOBASE;
619 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
620 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
621 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
622 tp->t_rcvtime = ticks;
623 tp->t_bw_rtttime = ticks;
624 /*
625 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
626 * because the socket may be bound to an IPv6 wildcard address,
627 * which may match an IPv4-mapped IPv6 address.
628 */
629 inp->inp_ip_ttl = ip_defttl;
630 inp->inp_ppcb = (caddr_t)tp;
631 return (tp); /* XXX */
632 }
633
634 /*
635 * Drop a TCP connection, reporting
636 * the specified error. If connection is synchronized,
637 * then send a RST to peer.
638 */
639 struct tcpcb *
640 tcp_drop(tp, errno)
641 register struct tcpcb *tp;
642 int errno;
643 {
644 struct socket *so = tp->t_inpcb->inp_socket;
645
646 INP_INFO_WLOCK_ASSERT(&tcbinfo);
647 INP_LOCK_ASSERT(tp->t_inpcb);
648
649 if (TCPS_HAVERCVDSYN(tp->t_state)) {
650 tp->t_state = TCPS_CLOSED;
651 (void) tcp_output(tp);
652 tcpstat.tcps_drops++;
653 } else
654 tcpstat.tcps_conndrops++;
655 if (errno == ETIMEDOUT && tp->t_softerror)
656 errno = tp->t_softerror;
657 so->so_error = errno;
658 return (tcp_close(tp));
659 }
660
661 static void
662 tcp_discardcb(tp)
663 struct tcpcb *tp;
664 {
665 struct tseg_qent *q;
666 struct inpcb *inp = tp->t_inpcb;
667 struct socket *so = inp->inp_socket;
668 #ifdef INET6
669 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
670 #endif /* INET6 */
671
672 INP_LOCK_ASSERT(inp);
673
674 /*
675 * Make sure that all of our timers are stopped before we
676 * delete the PCB.
677 */
678 callout_stop(tp->tt_rexmt);
679 callout_stop(tp->tt_persist);
680 callout_stop(tp->tt_keep);
681 callout_stop(tp->tt_2msl);
682 callout_stop(tp->tt_delack);
683
684 /*
685 * If we got enough samples through the srtt filter,
686 * save the rtt and rttvar in the routing entry.
687 * 'Enough' is arbitrarily defined as 4 rtt samples.
688 * 4 samples is enough for the srtt filter to converge
689 * to within enough % of the correct value; fewer samples
690 * and we could save a bogus rtt. The danger is not high
691 * as tcp quickly recovers from everything.
692 * XXX: Works very well but needs some more statistics!
693 */
694 if (tp->t_rttupdated >= 4) {
695 struct hc_metrics_lite metrics;
696 u_long ssthresh;
697
698 bzero(&metrics, sizeof(metrics));
699 /*
700 * Update the ssthresh always when the conditions below
701 * are satisfied. This gives us better new start value
702 * for the congestion avoidance for new connections.
703 * ssthresh is only set if packet loss occured on a session.
704 */
705 ssthresh = tp->snd_ssthresh;
706 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
707 /*
708 * convert the limit from user data bytes to
709 * packets then to packet data bytes.
710 */
711 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
712 if (ssthresh < 2)
713 ssthresh = 2;
714 ssthresh *= (u_long)(tp->t_maxseg +
715 #ifdef INET6
716 (isipv6 ? sizeof (struct ip6_hdr) +
717 sizeof (struct tcphdr) :
718 #endif
719 sizeof (struct tcpiphdr)
720 #ifdef INET6
721 )
722 #endif
723 );
724 } else
725 ssthresh = 0;
726 metrics.rmx_ssthresh = ssthresh;
727
728 metrics.rmx_rtt = tp->t_srtt;
729 metrics.rmx_rttvar = tp->t_rttvar;
730 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
731 metrics.rmx_bandwidth = tp->snd_bandwidth;
732 metrics.rmx_cwnd = tp->snd_cwnd;
733 metrics.rmx_sendpipe = 0;
734 metrics.rmx_recvpipe = 0;
735
736 tcp_hc_update(&inp->inp_inc, &metrics);
737 }
738
739 /* free the reassembly queue, if any */
740 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
741 LIST_REMOVE(q, tqe_q);
742 m_freem(q->tqe_m);
743 uma_zfree(tcp_reass_zone, q);
744 tp->t_segqlen--;
745 tcp_reass_qsize--;
746 }
747 tcp_free_sackholes(tp);
748 inp->inp_ppcb = NULL;
749 tp->t_inpcb = NULL;
750 uma_zfree(tcpcb_zone, tp);
751 soisdisconnected(so);
752 }
753
754 /*
755 * Close a TCP control block:
756 * discard all space held by the tcp
757 * discard internet protocol block
758 * wake up any sleepers
759 */
760 struct tcpcb *
761 tcp_close(tp)
762 struct tcpcb *tp;
763 {
764 struct inpcb *inp = tp->t_inpcb;
765 #ifdef INET6
766 struct socket *so = inp->inp_socket;
767 #endif
768
769 INP_INFO_WLOCK_ASSERT(&tcbinfo);
770 INP_LOCK_ASSERT(inp);
771
772 tcp_discardcb(tp);
773 #ifdef INET6
774 if (INP_CHECK_SOCKAF(so, AF_INET6))
775 in6_pcbdetach(inp);
776 else
777 #endif
778 in_pcbdetach(inp);
779 tcpstat.tcps_closed++;
780 return (NULL);
781 }
782
783 void
784 tcp_drain()
785 {
786 if (do_tcpdrain)
787 {
788 struct inpcb *inpb;
789 struct tcpcb *tcpb;
790 struct tseg_qent *te;
791
792 /*
793 * Walk the tcpbs, if existing, and flush the reassembly queue,
794 * if there is one...
795 * XXX: The "Net/3" implementation doesn't imply that the TCP
796 * reassembly queue should be flushed, but in a situation
797 * where we're really low on mbufs, this is potentially
798 * usefull.
799 */
800 INP_INFO_RLOCK(&tcbinfo);
801 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
802 if (inpb->inp_vflag & INP_TIMEWAIT)
803 continue;
804 INP_LOCK(inpb);
805 if ((tcpb = intotcpcb(inpb)) != NULL) {
806 while ((te = LIST_FIRST(&tcpb->t_segq))
807 != NULL) {
808 LIST_REMOVE(te, tqe_q);
809 m_freem(te->tqe_m);
810 uma_zfree(tcp_reass_zone, te);
811 tcpb->t_segqlen--;
812 tcp_reass_qsize--;
813 }
814 tcp_clean_sackreport(tcpb);
815 }
816 INP_UNLOCK(inpb);
817 }
818 INP_INFO_RUNLOCK(&tcbinfo);
819 }
820 }
821
822 /*
823 * Notify a tcp user of an asynchronous error;
824 * store error as soft error, but wake up user
825 * (for now, won't do anything until can select for soft error).
826 *
827 * Do not wake up user since there currently is no mechanism for
828 * reporting soft errors (yet - a kqueue filter may be added).
829 */
830 static struct inpcb *
831 tcp_notify(inp, error)
832 struct inpcb *inp;
833 int error;
834 {
835 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
836
837 INP_INFO_WLOCK_ASSERT(&tcbinfo);
838 INP_LOCK_ASSERT(inp);
839
840 /*
841 * Ignore some errors if we are hooked up.
842 * If connection hasn't completed, has retransmitted several times,
843 * and receives a second error, give up now. This is better
844 * than waiting a long time to establish a connection that
845 * can never complete.
846 */
847 if (tp->t_state == TCPS_ESTABLISHED &&
848 (error == EHOSTUNREACH || error == ENETUNREACH ||
849 error == EHOSTDOWN)) {
850 return (inp);
851 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
852 tp->t_softerror) {
853 tcp_drop(tp, error);
854 return (struct inpcb *)0;
855 } else {
856 tp->t_softerror = error;
857 return (inp);
858 }
859 #if 0
860 wakeup( &so->so_timeo);
861 sorwakeup(so);
862 sowwakeup(so);
863 #endif
864 }
865
866 static int
867 tcp_pcblist(SYSCTL_HANDLER_ARGS)
868 {
869 int error, i, n;
870 struct inpcb *inp, **inp_list;
871 inp_gen_t gencnt;
872 struct xinpgen xig;
873
874 /*
875 * The process of preparing the TCB list is too time-consuming and
876 * resource-intensive to repeat twice on every request.
877 */
878 if (req->oldptr == NULL) {
879 n = tcbinfo.ipi_count;
880 req->oldidx = 2 * (sizeof xig)
881 + (n + n/8) * sizeof(struct xtcpcb);
882 return (0);
883 }
884
885 if (req->newptr != NULL)
886 return (EPERM);
887
888 /*
889 * OK, now we're committed to doing something.
890 */
891 INP_INFO_RLOCK(&tcbinfo);
892 gencnt = tcbinfo.ipi_gencnt;
893 n = tcbinfo.ipi_count;
894 INP_INFO_RUNLOCK(&tcbinfo);
895
896 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
897 + n * sizeof(struct xtcpcb));
898 if (error != 0)
899 return (error);
900
901 xig.xig_len = sizeof xig;
902 xig.xig_count = n;
903 xig.xig_gen = gencnt;
904 xig.xig_sogen = so_gencnt;
905 error = SYSCTL_OUT(req, &xig, sizeof xig);
906 if (error)
907 return (error);
908
909 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
910 if (inp_list == NULL)
911 return (ENOMEM);
912
913 INP_INFO_RLOCK(&tcbinfo);
914 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
915 inp = LIST_NEXT(inp, inp_list)) {
916 INP_LOCK(inp);
917 if (inp->inp_gencnt <= gencnt) {
918 /*
919 * XXX: This use of cr_cansee(), introduced with
920 * TCP state changes, is not quite right, but for
921 * now, better than nothing.
922 */
923 if (inp->inp_vflag & INP_TIMEWAIT)
924 error = cr_cansee(req->td->td_ucred,
925 intotw(inp)->tw_cred);
926 else
927 error = cr_canseesocket(req->td->td_ucred,
928 inp->inp_socket);
929 if (error == 0)
930 inp_list[i++] = inp;
931 }
932 INP_UNLOCK(inp);
933 }
934 INP_INFO_RUNLOCK(&tcbinfo);
935 n = i;
936
937 error = 0;
938 for (i = 0; i < n; i++) {
939 inp = inp_list[i];
940 if (inp->inp_gencnt <= gencnt) {
941 struct xtcpcb xt;
942 caddr_t inp_ppcb;
943
944 bzero(&xt, sizeof(xt));
945 xt.xt_len = sizeof xt;
946 /* XXX should avoid extra copy */
947 bcopy(inp, &xt.xt_inp, sizeof *inp);
948 inp_ppcb = inp->inp_ppcb;
949 if (inp_ppcb == NULL)
950 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
951 else if (inp->inp_vflag & INP_TIMEWAIT) {
952 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
953 xt.xt_tp.t_state = TCPS_TIME_WAIT;
954 } else
955 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
956 if (inp->inp_socket != NULL)
957 sotoxsocket(inp->inp_socket, &xt.xt_socket);
958 else {
959 bzero(&xt.xt_socket, sizeof xt.xt_socket);
960 xt.xt_socket.xso_protocol = IPPROTO_TCP;
961 }
962 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
963 error = SYSCTL_OUT(req, &xt, sizeof xt);
964 }
965 }
966 if (!error) {
967 /*
968 * Give the user an updated idea of our state.
969 * If the generation differs from what we told
970 * her before, she knows that something happened
971 * while we were processing this request, and it
972 * might be necessary to retry.
973 */
974 INP_INFO_RLOCK(&tcbinfo);
975 xig.xig_gen = tcbinfo.ipi_gencnt;
976 xig.xig_sogen = so_gencnt;
977 xig.xig_count = tcbinfo.ipi_count;
978 INP_INFO_RUNLOCK(&tcbinfo);
979 error = SYSCTL_OUT(req, &xig, sizeof xig);
980 }
981 free(inp_list, M_TEMP);
982 return (error);
983 }
984
985 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
986 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
987
988 static int
989 tcp_getcred(SYSCTL_HANDLER_ARGS)
990 {
991 struct xucred xuc;
992 struct sockaddr_in addrs[2];
993 struct inpcb *inp;
994 int error;
995
996 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
997 if (error)
998 return (error);
999 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1000 if (error)
1001 return (error);
1002 INP_INFO_RLOCK(&tcbinfo);
1003 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1004 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1005 if (inp == NULL) {
1006 error = ENOENT;
1007 goto outunlocked;
1008 }
1009 INP_LOCK(inp);
1010 if (inp->inp_socket == NULL) {
1011 error = ENOENT;
1012 goto out;
1013 }
1014 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1015 if (error)
1016 goto out;
1017 cru2x(inp->inp_socket->so_cred, &xuc);
1018 out:
1019 INP_UNLOCK(inp);
1020 outunlocked:
1021 INP_INFO_RUNLOCK(&tcbinfo);
1022 if (error == 0)
1023 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1024 return (error);
1025 }
1026
1027 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1028 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1029 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1030
1031 #ifdef INET6
1032 static int
1033 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1034 {
1035 struct xucred xuc;
1036 struct sockaddr_in6 addrs[2];
1037 struct in6_addr a6[2];
1038 struct inpcb *inp;
1039 int error, mapped = 0;
1040
1041 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1042 if (error)
1043 return (error);
1044 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1045 if (error)
1046 return (error);
1047 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1048 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1049 mapped = 1;
1050 else
1051 return (EINVAL);
1052 } else {
1053 error = in6_embedscope(&a6[0], &addrs[0], NULL, NULL);
1054 if (error)
1055 return (EINVAL);
1056 error = in6_embedscope(&a6[1], &addrs[1], NULL, NULL);
1057 if (error)
1058 return (EINVAL);
1059 }
1060 INP_INFO_RLOCK(&tcbinfo);
1061 if (mapped == 1)
1062 inp = in_pcblookup_hash(&tcbinfo,
1063 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1064 addrs[1].sin6_port,
1065 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1066 addrs[0].sin6_port,
1067 0, NULL);
1068 else
1069 inp = in6_pcblookup_hash(&tcbinfo, &a6[1], addrs[1].sin6_port,
1070 &a6[0], addrs[0].sin6_port, 0, NULL);
1071 if (inp == NULL) {
1072 error = ENOENT;
1073 goto outunlocked;
1074 }
1075 INP_LOCK(inp);
1076 if (inp->inp_socket == NULL) {
1077 error = ENOENT;
1078 goto out;
1079 }
1080 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1081 if (error)
1082 goto out;
1083 cru2x(inp->inp_socket->so_cred, &xuc);
1084 out:
1085 INP_UNLOCK(inp);
1086 outunlocked:
1087 INP_INFO_RUNLOCK(&tcbinfo);
1088 if (error == 0)
1089 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1090 return (error);
1091 }
1092
1093 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1094 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1095 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1096 #endif
1097
1098
1099 void
1100 tcp_ctlinput(cmd, sa, vip)
1101 int cmd;
1102 struct sockaddr *sa;
1103 void *vip;
1104 {
1105 struct ip *ip = vip;
1106 struct tcphdr *th;
1107 struct in_addr faddr;
1108 struct inpcb *inp;
1109 struct tcpcb *tp;
1110 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1111 struct icmp *icp;
1112 struct in_conninfo inc;
1113 tcp_seq icmp_tcp_seq;
1114 int mtu;
1115
1116 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1117 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1118 return;
1119
1120 if (cmd == PRC_MSGSIZE)
1121 notify = tcp_mtudisc;
1122 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1123 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1124 notify = tcp_drop_syn_sent;
1125 /*
1126 * Redirects don't need to be handled up here.
1127 */
1128 else if (PRC_IS_REDIRECT(cmd))
1129 return;
1130 /*
1131 * Source quench is depreciated.
1132 */
1133 else if (cmd == PRC_QUENCH)
1134 return;
1135 /*
1136 * Hostdead is ugly because it goes linearly through all PCBs.
1137 * XXX: We never get this from ICMP, otherwise it makes an
1138 * excellent DoS attack on machines with many connections.
1139 */
1140 else if (cmd == PRC_HOSTDEAD)
1141 ip = NULL;
1142 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1143 return;
1144 if (ip != NULL) {
1145 icp = (struct icmp *)((caddr_t)ip
1146 - offsetof(struct icmp, icmp_ip));
1147 th = (struct tcphdr *)((caddr_t)ip
1148 + (ip->ip_hl << 2));
1149 INP_INFO_WLOCK(&tcbinfo);
1150 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1151 ip->ip_src, th->th_sport, 0, NULL);
1152 if (inp != NULL) {
1153 INP_LOCK(inp);
1154 if (inp->inp_socket != NULL) {
1155 icmp_tcp_seq = htonl(th->th_seq);
1156 tp = intotcpcb(inp);
1157 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1158 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1159 if (cmd == PRC_MSGSIZE) {
1160 /*
1161 * MTU discovery:
1162 * If we got a needfrag set the MTU
1163 * in the route to the suggested new
1164 * value (if given) and then notify.
1165 */
1166 bzero(&inc, sizeof(inc));
1167 inc.inc_flags = 0; /* IPv4 */
1168 inc.inc_faddr = faddr;
1169
1170 mtu = ntohs(icp->icmp_nextmtu);
1171 /*
1172 * If no alternative MTU was
1173 * proposed, try the next smaller
1174 * one. ip->ip_len has already
1175 * been swapped in icmp_input().
1176 */
1177 if (!mtu)
1178 mtu = ip_next_mtu(ip->ip_len,
1179 1);
1180 if (mtu < max(296, (tcp_minmss)
1181 + sizeof(struct tcpiphdr)))
1182 mtu = 0;
1183 if (!mtu)
1184 mtu = tcp_mssdflt
1185 + sizeof(struct tcpiphdr);
1186 /*
1187 * Only cache the the MTU if it
1188 * is smaller than the interface
1189 * or route MTU. tcp_mtudisc()
1190 * will do right thing by itself.
1191 */
1192 if (mtu <= tcp_maxmtu(&inc))
1193 tcp_hc_updatemtu(&inc, mtu);
1194 }
1195
1196 inp = (*notify)(inp, inetctlerrmap[cmd]);
1197 }
1198 }
1199 if (inp != NULL)
1200 INP_UNLOCK(inp);
1201 } else {
1202 inc.inc_fport = th->th_dport;
1203 inc.inc_lport = th->th_sport;
1204 inc.inc_faddr = faddr;
1205 inc.inc_laddr = ip->ip_src;
1206 #ifdef INET6
1207 inc.inc_isipv6 = 0;
1208 #endif
1209 syncache_unreach(&inc, th);
1210 }
1211 INP_INFO_WUNLOCK(&tcbinfo);
1212 } else
1213 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1214 }
1215
1216 #ifdef INET6
1217 void
1218 tcp6_ctlinput(cmd, sa, d)
1219 int cmd;
1220 struct sockaddr *sa;
1221 void *d;
1222 {
1223 struct tcphdr th;
1224 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1225 struct ip6_hdr *ip6;
1226 struct mbuf *m;
1227 struct ip6ctlparam *ip6cp = NULL;
1228 const struct sockaddr_in6 *sa6_src = NULL;
1229 int off;
1230 struct tcp_portonly {
1231 u_int16_t th_sport;
1232 u_int16_t th_dport;
1233 } *thp;
1234
1235 if (sa->sa_family != AF_INET6 ||
1236 sa->sa_len != sizeof(struct sockaddr_in6))
1237 return;
1238
1239 if (cmd == PRC_MSGSIZE)
1240 notify = tcp_mtudisc;
1241 else if (!PRC_IS_REDIRECT(cmd) &&
1242 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1243 return;
1244 /* Source quench is depreciated. */
1245 else if (cmd == PRC_QUENCH)
1246 return;
1247
1248 /* if the parameter is from icmp6, decode it. */
1249 if (d != NULL) {
1250 ip6cp = (struct ip6ctlparam *)d;
1251 m = ip6cp->ip6c_m;
1252 ip6 = ip6cp->ip6c_ip6;
1253 off = ip6cp->ip6c_off;
1254 sa6_src = ip6cp->ip6c_src;
1255 } else {
1256 m = NULL;
1257 ip6 = NULL;
1258 off = 0; /* fool gcc */
1259 sa6_src = &sa6_any;
1260 }
1261
1262 if (ip6 != NULL) {
1263 struct in_conninfo inc;
1264 /*
1265 * XXX: We assume that when IPV6 is non NULL,
1266 * M and OFF are valid.
1267 */
1268
1269 /* check if we can safely examine src and dst ports */
1270 if (m->m_pkthdr.len < off + sizeof(*thp))
1271 return;
1272
1273 bzero(&th, sizeof(th));
1274 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1275
1276 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1277 (struct sockaddr *)ip6cp->ip6c_src,
1278 th.th_sport, cmd, NULL, notify);
1279
1280 inc.inc_fport = th.th_dport;
1281 inc.inc_lport = th.th_sport;
1282 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1283 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1284 inc.inc_isipv6 = 1;
1285 INP_INFO_WLOCK(&tcbinfo);
1286 syncache_unreach(&inc, &th);
1287 INP_INFO_WUNLOCK(&tcbinfo);
1288 } else
1289 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1290 0, cmd, NULL, notify);
1291 }
1292 #endif /* INET6 */
1293
1294
1295 /*
1296 * Following is where TCP initial sequence number generation occurs.
1297 *
1298 * There are two places where we must use initial sequence numbers:
1299 * 1. In SYN-ACK packets.
1300 * 2. In SYN packets.
1301 *
1302 * All ISNs for SYN-ACK packets are generated by the syncache. See
1303 * tcp_syncache.c for details.
1304 *
1305 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1306 * depends on this property. In addition, these ISNs should be
1307 * unguessable so as to prevent connection hijacking. To satisfy
1308 * the requirements of this situation, the algorithm outlined in
1309 * RFC 1948 is used, with only small modifications.
1310 *
1311 * Implementation details:
1312 *
1313 * Time is based off the system timer, and is corrected so that it
1314 * increases by one megabyte per second. This allows for proper
1315 * recycling on high speed LANs while still leaving over an hour
1316 * before rollover.
1317 *
1318 * As reading the *exact* system time is too expensive to be done
1319 * whenever setting up a TCP connection, we increment the time
1320 * offset in two ways. First, a small random positive increment
1321 * is added to isn_offset for each connection that is set up.
1322 * Second, the function tcp_isn_tick fires once per clock tick
1323 * and increments isn_offset as necessary so that sequence numbers
1324 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1325 * random positive increments serve only to ensure that the same
1326 * exact sequence number is never sent out twice (as could otherwise
1327 * happen when a port is recycled in less than the system tick
1328 * interval.)
1329 *
1330 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1331 * between seeding of isn_secret. This is normally set to zero,
1332 * as reseeding should not be necessary.
1333 *
1334 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1335 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1336 * general, this means holding an exclusive (write) lock.
1337 */
1338
1339 #define ISN_BYTES_PER_SECOND 1048576
1340 #define ISN_STATIC_INCREMENT 4096
1341 #define ISN_RANDOM_INCREMENT (4096 - 1)
1342
1343 static u_char isn_secret[32];
1344 static int isn_last_reseed;
1345 static u_int32_t isn_offset, isn_offset_old;
1346 static MD5_CTX isn_ctx;
1347
1348 tcp_seq
1349 tcp_new_isn(tp)
1350 struct tcpcb *tp;
1351 {
1352 u_int32_t md5_buffer[4];
1353 tcp_seq new_isn;
1354
1355 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1356 INP_LOCK_ASSERT(tp->t_inpcb);
1357
1358 /* Seed if this is the first use, reseed if requested. */
1359 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1360 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1361 < (u_int)ticks))) {
1362 read_random(&isn_secret, sizeof(isn_secret));
1363 isn_last_reseed = ticks;
1364 }
1365
1366 /* Compute the md5 hash and return the ISN. */
1367 MD5Init(&isn_ctx);
1368 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1369 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1370 #ifdef INET6
1371 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1372 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1373 sizeof(struct in6_addr));
1374 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1375 sizeof(struct in6_addr));
1376 } else
1377 #endif
1378 {
1379 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1380 sizeof(struct in_addr));
1381 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1382 sizeof(struct in_addr));
1383 }
1384 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1385 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1386 new_isn = (tcp_seq) md5_buffer[0];
1387 isn_offset += ISN_STATIC_INCREMENT +
1388 (arc4random() & ISN_RANDOM_INCREMENT);
1389 new_isn += isn_offset;
1390 return (new_isn);
1391 }
1392
1393 /*
1394 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1395 * to keep time flowing at a relatively constant rate. If the random
1396 * increments have already pushed us past the projected offset, do nothing.
1397 */
1398 static void
1399 tcp_isn_tick(xtp)
1400 void *xtp;
1401 {
1402 u_int32_t projected_offset;
1403
1404 INP_INFO_WLOCK(&tcbinfo);
1405 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1406
1407 if (projected_offset > isn_offset)
1408 isn_offset = projected_offset;
1409
1410 isn_offset_old = isn_offset;
1411 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1412 INP_INFO_WUNLOCK(&tcbinfo);
1413 }
1414
1415 /*
1416 * When a specific ICMP unreachable message is received and the
1417 * connection state is SYN-SENT, drop the connection. This behavior
1418 * is controlled by the icmp_may_rst sysctl.
1419 */
1420 struct inpcb *
1421 tcp_drop_syn_sent(inp, errno)
1422 struct inpcb *inp;
1423 int errno;
1424 {
1425 struct tcpcb *tp = intotcpcb(inp);
1426
1427 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1428 INP_LOCK_ASSERT(inp);
1429
1430 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) {
1431 tcp_drop(tp, errno);
1432 return (NULL);
1433 }
1434 return (inp);
1435 }
1436
1437 /*
1438 * When `need fragmentation' ICMP is received, update our idea of the MSS
1439 * based on the new value in the route. Also nudge TCP to send something,
1440 * since we know the packet we just sent was dropped.
1441 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1442 */
1443 struct inpcb *
1444 tcp_mtudisc(inp, errno)
1445 struct inpcb *inp;
1446 int errno;
1447 {
1448 struct tcpcb *tp = intotcpcb(inp);
1449 struct socket *so = inp->inp_socket;
1450 u_int maxmtu;
1451 u_int romtu;
1452 int mss;
1453 #ifdef INET6
1454 int isipv6;
1455 #endif /* INET6 */
1456
1457 INP_LOCK_ASSERT(inp);
1458 if (tp != NULL) {
1459 #ifdef INET6
1460 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1461 #endif
1462 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1463 romtu =
1464 #ifdef INET6
1465 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1466 #endif /* INET6 */
1467 tcp_maxmtu(&inp->inp_inc);
1468 if (!maxmtu)
1469 maxmtu = romtu;
1470 else
1471 maxmtu = min(maxmtu, romtu);
1472 if (!maxmtu) {
1473 tp->t_maxopd = tp->t_maxseg =
1474 #ifdef INET6
1475 isipv6 ? tcp_v6mssdflt :
1476 #endif /* INET6 */
1477 tcp_mssdflt;
1478 return (inp);
1479 }
1480 mss = maxmtu -
1481 #ifdef INET6
1482 (isipv6 ?
1483 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1484 #endif /* INET6 */
1485 sizeof(struct tcpiphdr)
1486 #ifdef INET6
1487 )
1488 #endif /* INET6 */
1489 ;
1490
1491 /*
1492 * XXX - The above conditional probably violates the TCP
1493 * spec. The problem is that, since we don't know the
1494 * other end's MSS, we are supposed to use a conservative
1495 * default. But, if we do that, then MTU discovery will
1496 * never actually take place, because the conservative
1497 * default is much less than the MTUs typically seen
1498 * on the Internet today. For the moment, we'll sweep
1499 * this under the carpet.
1500 *
1501 * The conservative default might not actually be a problem
1502 * if the only case this occurs is when sending an initial
1503 * SYN with options and data to a host we've never talked
1504 * to before. Then, they will reply with an MSS value which
1505 * will get recorded and the new parameters should get
1506 * recomputed. For Further Study.
1507 */
1508 if (tp->t_maxopd <= mss)
1509 return (inp);
1510 tp->t_maxopd = mss;
1511
1512 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1513 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1514 mss -= TCPOLEN_TSTAMP_APPA;
1515 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1516 if (mss > MCLBYTES)
1517 mss &= ~(MCLBYTES-1);
1518 #else
1519 if (mss > MCLBYTES)
1520 mss = mss / MCLBYTES * MCLBYTES;
1521 #endif
1522 if (so->so_snd.sb_hiwat < mss)
1523 mss = so->so_snd.sb_hiwat;
1524
1525 tp->t_maxseg = mss;
1526
1527 tcpstat.tcps_mturesent++;
1528 tp->t_rtttime = 0;
1529 tp->snd_nxt = tp->snd_una;
1530 tcp_output(tp);
1531 }
1532 return (inp);
1533 }
1534
1535 /*
1536 * Look-up the routing entry to the peer of this inpcb. If no route
1537 * is found and it cannot be allocated, then return NULL. This routine
1538 * is called by TCP routines that access the rmx structure and by tcp_mss
1539 * to get the interface MTU.
1540 */
1541 u_long
1542 tcp_maxmtu(inc)
1543 struct in_conninfo *inc;
1544 {
1545 struct route sro;
1546 struct sockaddr_in *dst;
1547 struct ifnet *ifp;
1548 u_long maxmtu = 0;
1549
1550 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1551
1552 bzero(&sro, sizeof(sro));
1553 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1554 dst = (struct sockaddr_in *)&sro.ro_dst;
1555 dst->sin_family = AF_INET;
1556 dst->sin_len = sizeof(*dst);
1557 dst->sin_addr = inc->inc_faddr;
1558 rtalloc_ign(&sro, RTF_CLONING);
1559 }
1560 if (sro.ro_rt != NULL) {
1561 ifp = sro.ro_rt->rt_ifp;
1562 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1563 maxmtu = ifp->if_mtu;
1564 else
1565 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1566 RTFREE(sro.ro_rt);
1567 }
1568 return (maxmtu);
1569 }
1570
1571 #ifdef INET6
1572 u_long
1573 tcp_maxmtu6(inc)
1574 struct in_conninfo *inc;
1575 {
1576 struct route_in6 sro6;
1577 struct ifnet *ifp;
1578 u_long maxmtu = 0;
1579
1580 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1581
1582 bzero(&sro6, sizeof(sro6));
1583 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1584 sro6.ro_dst.sin6_family = AF_INET6;
1585 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1586 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1587 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1588 }
1589 if (sro6.ro_rt != NULL) {
1590 ifp = sro6.ro_rt->rt_ifp;
1591 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1592 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1593 else
1594 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1595 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1596 RTFREE(sro6.ro_rt);
1597 }
1598
1599 return (maxmtu);
1600 }
1601 #endif /* INET6 */
1602
1603 #ifdef IPSEC
1604 /* compute ESP/AH header size for TCP, including outer IP header. */
1605 size_t
1606 ipsec_hdrsiz_tcp(tp)
1607 struct tcpcb *tp;
1608 {
1609 struct inpcb *inp;
1610 struct mbuf *m;
1611 size_t hdrsiz;
1612 struct ip *ip;
1613 #ifdef INET6
1614 struct ip6_hdr *ip6;
1615 #endif
1616 struct tcphdr *th;
1617
1618 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1619 return (0);
1620 MGETHDR(m, M_DONTWAIT, MT_DATA);
1621 if (!m)
1622 return (0);
1623
1624 #ifdef INET6
1625 if ((inp->inp_vflag & INP_IPV6) != 0) {
1626 ip6 = mtod(m, struct ip6_hdr *);
1627 th = (struct tcphdr *)(ip6 + 1);
1628 m->m_pkthdr.len = m->m_len =
1629 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1630 tcpip_fillheaders(inp, ip6, th);
1631 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1632 } else
1633 #endif /* INET6 */
1634 {
1635 ip = mtod(m, struct ip *);
1636 th = (struct tcphdr *)(ip + 1);
1637 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1638 tcpip_fillheaders(inp, ip, th);
1639 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1640 }
1641
1642 m_free(m);
1643 return (hdrsiz);
1644 }
1645 #endif /*IPSEC*/
1646
1647 /*
1648 * Move a TCP connection into TIME_WAIT state.
1649 * tcbinfo is locked.
1650 * inp is locked, and is unlocked before returning.
1651 */
1652 void
1653 tcp_twstart(tp)
1654 struct tcpcb *tp;
1655 {
1656 struct tcptw *tw;
1657 struct inpcb *inp;
1658 int tw_time, acknow;
1659 struct socket *so;
1660
1661 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1662 INP_LOCK_ASSERT(tp->t_inpcb);
1663
1664 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1665 if (tw == NULL) {
1666 tw = tcp_timer_2msl_tw(1);
1667 if (tw == NULL) {
1668 tcp_close(tp);
1669 return;
1670 }
1671 }
1672 inp = tp->t_inpcb;
1673 tw->tw_inpcb = inp;
1674
1675 /*
1676 * Recover last window size sent.
1677 */
1678 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1679
1680 /*
1681 * Set t_recent if timestamps are used on the connection.
1682 */
1683 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1684 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1685 tw->t_recent = tp->ts_recent;
1686 else
1687 tw->t_recent = 0;
1688
1689 tw->snd_nxt = tp->snd_nxt;
1690 tw->rcv_nxt = tp->rcv_nxt;
1691 tw->iss = tp->iss;
1692 tw->irs = tp->irs;
1693 tw->t_starttime = tp->t_starttime;
1694 tw->tw_time = 0;
1695
1696 /* XXX
1697 * If this code will
1698 * be used for fin-wait-2 state also, then we may need
1699 * a ts_recent from the last segment.
1700 */
1701 tw_time = 2 * tcp_msl;
1702 acknow = tp->t_flags & TF_ACKNOW;
1703 tcp_discardcb(tp);
1704 so = inp->inp_socket;
1705 ACCEPT_LOCK();
1706 SOCK_LOCK(so);
1707 so->so_pcb = NULL;
1708 tw->tw_cred = crhold(so->so_cred);
1709 tw->tw_so_options = so->so_options;
1710 sotryfree(so);
1711 inp->inp_socket = NULL;
1712 if (acknow)
1713 tcp_twrespond(tw, TH_ACK);
1714 inp->inp_ppcb = (caddr_t)tw;
1715 inp->inp_vflag |= INP_TIMEWAIT;
1716 tcp_timer_2msl_reset(tw, tw_time);
1717 INP_UNLOCK(inp);
1718 }
1719
1720 /*
1721 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1722 * the actual rate is slightly higher due to the addition of
1723 * random positive increments.
1724 *
1725 * Most other new OSes use semi-randomized ISN values, so we
1726 * do not need to worry about them.
1727 */
1728 #define MS_ISN_BYTES_PER_SECOND 250000
1729
1730 /*
1731 * Determine if the ISN we will generate has advanced beyond the last
1732 * sequence number used by the previous connection. If so, indicate
1733 * that it is safe to recycle this tw socket by returning 1.
1734 *
1735 * XXXRW: This function should assert the inpcb lock as it does multiple
1736 * non-atomic reads from the tcptw, but is currently called without it from
1737 * in_pcb.c:in_pcblookup_local().
1738 */
1739 int
1740 tcp_twrecycleable(struct tcptw *tw)
1741 {
1742 tcp_seq new_iss = tw->iss;
1743 tcp_seq new_irs = tw->irs;
1744
1745 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1746 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1747
1748 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1749 return (1);
1750 else
1751 return (0);
1752 }
1753
1754 struct tcptw *
1755 tcp_twclose(struct tcptw *tw, int reuse)
1756 {
1757 struct inpcb *inp;
1758
1759 inp = tw->tw_inpcb;
1760 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1761 INP_LOCK_ASSERT(inp);
1762
1763 tw->tw_inpcb = NULL;
1764 tcp_timer_2msl_stop(tw);
1765 inp->inp_ppcb = NULL;
1766 #ifdef INET6
1767 if (inp->inp_vflag & INP_IPV6PROTO)
1768 in6_pcbdetach(inp);
1769 else
1770 #endif
1771 in_pcbdetach(inp);
1772 tcpstat.tcps_closed++;
1773 crfree(tw->tw_cred);
1774 tw->tw_cred = NULL;
1775 if (reuse)
1776 return (tw);
1777 uma_zfree(tcptw_zone, tw);
1778 return (NULL);
1779 }
1780
1781 int
1782 tcp_twrespond(struct tcptw *tw, int flags)
1783 {
1784 struct inpcb *inp = tw->tw_inpcb;
1785 struct tcphdr *th;
1786 struct mbuf *m;
1787 struct ip *ip = NULL;
1788 u_int8_t *optp;
1789 u_int hdrlen, optlen;
1790 int error;
1791 #ifdef INET6
1792 struct ip6_hdr *ip6 = NULL;
1793 int isipv6 = inp->inp_inc.inc_isipv6;
1794 #endif
1795
1796 INP_LOCK_ASSERT(inp);
1797
1798 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1799 if (m == NULL)
1800 return (ENOBUFS);
1801 m->m_data += max_linkhdr;
1802
1803 #ifdef MAC
1804 mac_create_mbuf_from_inpcb(inp, m);
1805 #endif
1806
1807 #ifdef INET6
1808 if (isipv6) {
1809 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1810 ip6 = mtod(m, struct ip6_hdr *);
1811 th = (struct tcphdr *)(ip6 + 1);
1812 tcpip_fillheaders(inp, ip6, th);
1813 } else
1814 #endif
1815 {
1816 hdrlen = sizeof(struct tcpiphdr);
1817 ip = mtod(m, struct ip *);
1818 th = (struct tcphdr *)(ip + 1);
1819 tcpip_fillheaders(inp, ip, th);
1820 }
1821 optp = (u_int8_t *)(th + 1);
1822
1823 /*
1824 * Send a timestamp and echo-reply if both our side and our peer
1825 * have sent timestamps in our SYN's and this is not a RST.
1826 */
1827 if (tw->t_recent && flags == TH_ACK) {
1828 u_int32_t *lp = (u_int32_t *)optp;
1829
1830 /* Form timestamp option as shown in appendix A of RFC 1323. */
1831 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1832 *lp++ = htonl(ticks);
1833 *lp = htonl(tw->t_recent);
1834 optp += TCPOLEN_TSTAMP_APPA;
1835 }
1836
1837 optlen = optp - (u_int8_t *)(th + 1);
1838
1839 m->m_len = hdrlen + optlen;
1840 m->m_pkthdr.len = m->m_len;
1841
1842 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1843
1844 th->th_seq = htonl(tw->snd_nxt);
1845 th->th_ack = htonl(tw->rcv_nxt);
1846 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1847 th->th_flags = flags;
1848 th->th_win = htons(tw->last_win);
1849
1850 #ifdef INET6
1851 if (isipv6) {
1852 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1853 sizeof(struct tcphdr) + optlen);
1854 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1855 error = ip6_output(m, inp->in6p_outputopts, NULL,
1856 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1857 } else
1858 #endif
1859 {
1860 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1861 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1862 m->m_pkthdr.csum_flags = CSUM_TCP;
1863 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1864 ip->ip_len = m->m_pkthdr.len;
1865 if (path_mtu_discovery)
1866 ip->ip_off |= IP_DF;
1867 error = ip_output(m, inp->inp_options, NULL,
1868 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1869 NULL, inp);
1870 }
1871 if (flags & TH_ACK)
1872 tcpstat.tcps_sndacks++;
1873 else
1874 tcpstat.tcps_sndctrl++;
1875 tcpstat.tcps_sndtotal++;
1876 return (error);
1877 }
1878
1879 /*
1880 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1881 *
1882 * This code attempts to calculate the bandwidth-delay product as a
1883 * means of determining the optimal window size to maximize bandwidth,
1884 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1885 * routers. This code also does a fairly good job keeping RTTs in check
1886 * across slow links like modems. We implement an algorithm which is very
1887 * similar (but not meant to be) TCP/Vegas. The code operates on the
1888 * transmitter side of a TCP connection and so only effects the transmit
1889 * side of the connection.
1890 *
1891 * BACKGROUND: TCP makes no provision for the management of buffer space
1892 * at the end points or at the intermediate routers and switches. A TCP
1893 * stream, whether using NewReno or not, will eventually buffer as
1894 * many packets as it is able and the only reason this typically works is
1895 * due to the fairly small default buffers made available for a connection
1896 * (typicaly 16K or 32K). As machines use larger windows and/or window
1897 * scaling it is now fairly easy for even a single TCP connection to blow-out
1898 * all available buffer space not only on the local interface, but on
1899 * intermediate routers and switches as well. NewReno makes a misguided
1900 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1901 * then backing off, then steadily increasing the window again until another
1902 * failure occurs, ad-infinitum. This results in terrible oscillation that
1903 * is only made worse as network loads increase and the idea of intentionally
1904 * blowing out network buffers is, frankly, a terrible way to manage network
1905 * resources.
1906 *
1907 * It is far better to limit the transmit window prior to the failure
1908 * condition being achieved. There are two general ways to do this: First
1909 * you can 'scan' through different transmit window sizes and locate the
1910 * point where the RTT stops increasing, indicating that you have filled the
1911 * pipe, then scan backwards until you note that RTT stops decreasing, then
1912 * repeat ad-infinitum. This method works in principle but has severe
1913 * implementation issues due to RTT variances, timer granularity, and
1914 * instability in the algorithm which can lead to many false positives and
1915 * create oscillations as well as interact badly with other TCP streams
1916 * implementing the same algorithm.
1917 *
1918 * The second method is to limit the window to the bandwidth delay product
1919 * of the link. This is the method we implement. RTT variances and our
1920 * own manipulation of the congestion window, bwnd, can potentially
1921 * destabilize the algorithm. For this reason we have to stabilize the
1922 * elements used to calculate the window. We do this by using the minimum
1923 * observed RTT, the long term average of the observed bandwidth, and
1924 * by adding two segments worth of slop. It isn't perfect but it is able
1925 * to react to changing conditions and gives us a very stable basis on
1926 * which to extend the algorithm.
1927 */
1928 void
1929 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1930 {
1931 u_long bw;
1932 u_long bwnd;
1933 int save_ticks;
1934
1935 INP_LOCK_ASSERT(tp->t_inpcb);
1936
1937 /*
1938 * If inflight_enable is disabled in the middle of a tcp connection,
1939 * make sure snd_bwnd is effectively disabled.
1940 */
1941 if (tcp_inflight_enable == 0) {
1942 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1943 tp->snd_bandwidth = 0;
1944 return;
1945 }
1946
1947 /*
1948 * Figure out the bandwidth. Due to the tick granularity this
1949 * is a very rough number and it MUST be averaged over a fairly
1950 * long period of time. XXX we need to take into account a link
1951 * that is not using all available bandwidth, but for now our
1952 * slop will ramp us up if this case occurs and the bandwidth later
1953 * increases.
1954 *
1955 * Note: if ticks rollover 'bw' may wind up negative. We must
1956 * effectively reset t_bw_rtttime for this case.
1957 */
1958 save_ticks = ticks;
1959 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1960 return;
1961
1962 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1963 (save_ticks - tp->t_bw_rtttime);
1964 tp->t_bw_rtttime = save_ticks;
1965 tp->t_bw_rtseq = ack_seq;
1966 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1967 return;
1968 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1969
1970 tp->snd_bandwidth = bw;
1971
1972 /*
1973 * Calculate the semi-static bandwidth delay product, plus two maximal
1974 * segments. The additional slop puts us squarely in the sweet
1975 * spot and also handles the bandwidth run-up case and stabilization.
1976 * Without the slop we could be locking ourselves into a lower
1977 * bandwidth.
1978 *
1979 * Situations Handled:
1980 * (1) Prevents over-queueing of packets on LANs, especially on
1981 * high speed LANs, allowing larger TCP buffers to be
1982 * specified, and also does a good job preventing
1983 * over-queueing of packets over choke points like modems
1984 * (at least for the transmit side).
1985 *
1986 * (2) Is able to handle changing network loads (bandwidth
1987 * drops so bwnd drops, bandwidth increases so bwnd
1988 * increases).
1989 *
1990 * (3) Theoretically should stabilize in the face of multiple
1991 * connections implementing the same algorithm (this may need
1992 * a little work).
1993 *
1994 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1995 * be adjusted with a sysctl but typically only needs to be
1996 * on very slow connections. A value no smaller then 5
1997 * should be used, but only reduce this default if you have
1998 * no other choice.
1999 */
2000 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2001 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2002 #undef USERTT
2003
2004 if (tcp_inflight_debug > 0) {
2005 static int ltime;
2006 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2007 ltime = ticks;
2008 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2009 tp,
2010 bw,
2011 tp->t_rttbest,
2012 tp->t_srtt,
2013 bwnd
2014 );
2015 }
2016 }
2017 if ((long)bwnd < tcp_inflight_min)
2018 bwnd = tcp_inflight_min;
2019 if (bwnd > tcp_inflight_max)
2020 bwnd = tcp_inflight_max;
2021 if ((long)bwnd < tp->t_maxseg * 2)
2022 bwnd = tp->t_maxseg * 2;
2023 tp->snd_bwnd = bwnd;
2024 }
2025
2026 #ifdef TCP_SIGNATURE
2027 /*
2028 * Callback function invoked by m_apply() to digest TCP segment data
2029 * contained within an mbuf chain.
2030 */
2031 static int
2032 tcp_signature_apply(void *fstate, void *data, u_int len)
2033 {
2034
2035 MD5Update(fstate, (u_char *)data, len);
2036 return (0);
2037 }
2038
2039 /*
2040 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2041 *
2042 * Parameters:
2043 * m pointer to head of mbuf chain
2044 * off0 offset to TCP header within the mbuf chain
2045 * len length of TCP segment data, excluding options
2046 * optlen length of TCP segment options
2047 * buf pointer to storage for computed MD5 digest
2048 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2049 *
2050 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2051 * When called from tcp_input(), we can be sure that th_sum has been
2052 * zeroed out and verified already.
2053 *
2054 * This function is for IPv4 use only. Calling this function with an
2055 * IPv6 packet in the mbuf chain will yield undefined results.
2056 *
2057 * Return 0 if successful, otherwise return -1.
2058 *
2059 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2060 * search with the destination IP address, and a 'magic SPI' to be
2061 * determined by the application. This is hardcoded elsewhere to 1179
2062 * right now. Another branch of this code exists which uses the SPD to
2063 * specify per-application flows but it is unstable.
2064 */
2065 int
2066 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2067 u_char *buf, u_int direction)
2068 {
2069 union sockaddr_union dst;
2070 struct ippseudo ippseudo;
2071 MD5_CTX ctx;
2072 int doff;
2073 struct ip *ip;
2074 struct ipovly *ipovly;
2075 struct secasvar *sav;
2076 struct tcphdr *th;
2077 u_short savecsum;
2078
2079 KASSERT(m != NULL, ("NULL mbuf chain"));
2080 KASSERT(buf != NULL, ("NULL signature pointer"));
2081
2082 /* Extract the destination from the IP header in the mbuf. */
2083 ip = mtod(m, struct ip *);
2084 bzero(&dst, sizeof(union sockaddr_union));
2085 dst.sa.sa_len = sizeof(struct sockaddr_in);
2086 dst.sa.sa_family = AF_INET;
2087 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2088 ip->ip_src : ip->ip_dst;
2089
2090 /* Look up an SADB entry which matches the address of the peer. */
2091 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2092 if (sav == NULL) {
2093 printf("%s: SADB lookup failed for %s\n", __func__,
2094 inet_ntoa(dst.sin.sin_addr));
2095 return (EINVAL);
2096 }
2097
2098 MD5Init(&ctx);
2099 ipovly = (struct ipovly *)ip;
2100 th = (struct tcphdr *)((u_char *)ip + off0);
2101 doff = off0 + sizeof(struct tcphdr) + optlen;
2102
2103 /*
2104 * Step 1: Update MD5 hash with IP pseudo-header.
2105 *
2106 * XXX The ippseudo header MUST be digested in network byte order,
2107 * or else we'll fail the regression test. Assume all fields we've
2108 * been doing arithmetic on have been in host byte order.
2109 * XXX One cannot depend on ipovly->ih_len here. When called from
2110 * tcp_output(), the underlying ip_len member has not yet been set.
2111 */
2112 ippseudo.ippseudo_src = ipovly->ih_src;
2113 ippseudo.ippseudo_dst = ipovly->ih_dst;
2114 ippseudo.ippseudo_pad = 0;
2115 ippseudo.ippseudo_p = IPPROTO_TCP;
2116 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2117 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2118
2119 /*
2120 * Step 2: Update MD5 hash with TCP header, excluding options.
2121 * The TCP checksum must be set to zero.
2122 */
2123 savecsum = th->th_sum;
2124 th->th_sum = 0;
2125 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2126 th->th_sum = savecsum;
2127
2128 /*
2129 * Step 3: Update MD5 hash with TCP segment data.
2130 * Use m_apply() to avoid an early m_pullup().
2131 */
2132 if (len > 0)
2133 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2134
2135 /*
2136 * Step 4: Update MD5 hash with shared secret.
2137 */
2138 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2139 MD5Final(buf, &ctx);
2140
2141 key_sa_recordxfer(sav, m);
2142 KEY_FREESAV(&sav);
2143 return (0);
2144 }
2145 #endif /* TCP_SIGNATURE */
2146
2147 static int
2148 sysctl_drop(SYSCTL_HANDLER_ARGS)
2149 {
2150 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2151 struct sockaddr_storage addrs[2];
2152 struct inpcb *inp;
2153 struct tcpcb *tp;
2154 struct tcptw *tw;
2155 struct sockaddr_in *fin, *lin;
2156 #ifdef INET6
2157 struct sockaddr_in6 *fin6, *lin6;
2158 struct in6_addr f6, l6;
2159 #endif
2160 int error;
2161
2162 inp = NULL;
2163 fin = lin = NULL;
2164 #ifdef INET6
2165 fin6 = lin6 = NULL;
2166 #endif
2167 error = 0;
2168
2169 if (req->oldptr != NULL || req->oldlen != 0)
2170 return (EINVAL);
2171 if (req->newptr == NULL)
2172 return (EPERM);
2173 if (req->newlen < sizeof(addrs))
2174 return (ENOMEM);
2175 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2176 if (error)
2177 return (error);
2178
2179 switch (addrs[0].ss_family) {
2180 #ifdef INET6
2181 case AF_INET6:
2182 fin6 = (struct sockaddr_in6 *)&addrs[0];
2183 lin6 = (struct sockaddr_in6 *)&addrs[1];
2184 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2185 lin6->sin6_len != sizeof(struct sockaddr_in6))
2186 return (EINVAL);
2187 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2188 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2189 return (EINVAL);
2190 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2191 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2192 fin = (struct sockaddr_in *)&addrs[0];
2193 lin = (struct sockaddr_in *)&addrs[1];
2194 break;
2195 }
2196 error = in6_embedscope(&f6, fin6, NULL, NULL);
2197 if (error)
2198 return (EINVAL);
2199 error = in6_embedscope(&l6, lin6, NULL, NULL);
2200 if (error)
2201 return (EINVAL);
2202 break;
2203 #endif
2204 case AF_INET:
2205 fin = (struct sockaddr_in *)&addrs[0];
2206 lin = (struct sockaddr_in *)&addrs[1];
2207 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2208 lin->sin_len != sizeof(struct sockaddr_in))
2209 return (EINVAL);
2210 break;
2211 default:
2212 return (EINVAL);
2213 }
2214 INP_INFO_WLOCK(&tcbinfo);
2215 switch (addrs[0].ss_family) {
2216 #ifdef INET6
2217 case AF_INET6:
2218 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2219 &l6, lin6->sin6_port, 0, NULL);
2220 break;
2221 #endif
2222 case AF_INET:
2223 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2224 lin->sin_addr, lin->sin_port, 0, NULL);
2225 break;
2226 }
2227 if (inp != NULL) {
2228 INP_LOCK(inp);
2229 if ((tw = intotw(inp)) &&
2230 (inp->inp_vflag & INP_TIMEWAIT) != 0) {
2231 (void) tcp_twclose(tw, 0);
2232 } else if ((tp = intotcpcb(inp)) &&
2233 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
2234 tp = tcp_drop(tp, ECONNABORTED);
2235 if (tp != NULL)
2236 INP_UNLOCK(inp);
2237 } else
2238 INP_UNLOCK(inp);
2239 } else
2240 error = ESRCH;
2241 INP_INFO_WUNLOCK(&tcbinfo);
2242 return (error);
2243 }
2244
2245 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2246 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2247 0, sysctl_drop, "", "Drop TCP connection");
Cache object: 1d7433c45b2cb528c2bd615f1b26b979
|